diff --git a/llvm/lib/Target/AMDGPU/SIFrameLowering.h b/llvm/lib/Target/AMDGPU/SIFrameLowering.h --- a/llvm/lib/Target/AMDGPU/SIFrameLowering.h +++ b/llvm/lib/Target/AMDGPU/SIFrameLowering.h @@ -56,6 +56,11 @@ MachineBasicBlock &MBB, MachineBasicBlock::iterator MI) const override; + bool spillCalleeSavedRegisters(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MBBI, + const ArrayRef CSI, + const TargetRegisterInfo *TRI) const override; + private: void saveSGPRToMemory(MachineBasicBlock &MBB, MachineBasicBlock::iterator MI, Register Reg, const int FI, LivePhysRegs &LiveRegs, @@ -97,6 +102,12 @@ buildCFI(MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI, const DebugLoc &DL, const MCCFIInstruction &CFIInst, MachineInstr::MIFlag flag = MachineInstr::FrameSetup) const; + /// Create a CFI index describing a spill of the register \p Reg to another + /// register \p RegCopy and build a MachineInstr around it. + MachineInstr *buildCFIForRegToRegSpill(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MBBI, + const DebugLoc &DL, const Register Reg, + const Register RegCopy) const; /// Create a CFI index describing a spill of an SGPR to a single lane of /// a VGPR and build a MachineInstr around it. MachineInstr *buildCFIForSGPRToVGPRSpill(MachineBasicBlock &MBB, @@ -115,6 +126,12 @@ MachineBasicBlock::iterator MBBI, const DebugLoc &DL, unsigned SGPR, int64_t Offset) const; + /// Create a CFI index describing a spill of a VGPR to VMEM and + /// build a MachineInstr around it. + MachineInstr *buildCFIForVGPRToVMEMSpill(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MBBI, + const DebugLoc &DL, unsigned VGPR, + int64_t Offset) const; MachineInstr *buildCFIForRegToSGPRPairSpill(MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI, const DebugLoc &DL, Register Reg, diff --git a/llvm/lib/Target/AMDGPU/SIFrameLowering.cpp b/llvm/lib/Target/AMDGPU/SIFrameLowering.cpp --- a/llvm/lib/Target/AMDGPU/SIFrameLowering.cpp +++ b/llvm/lib/Target/AMDGPU/SIFrameLowering.cpp @@ -1691,6 +1691,46 @@ return frameTriviallyRequiresSP(MFI); } +bool SIFrameLowering::spillCalleeSavedRegisters( + MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI, + const ArrayRef CSI, const TargetRegisterInfo *TRI) const { + MachineFunction &MF = *MBB.getParent(); + const GCNSubtarget &ST = MF.getSubtarget(); + const SIRegisterInfo *RI = ST.getRegisterInfo(); + const SIInstrInfo *TII = ST.getInstrInfo(); + + // FIXME: Switch to emitting frame moves conditionally. + const bool NeedsFrameMoves = true; + + for (const CalleeSavedInfo &CS : CSI) { + // Insert the spill to the stack frame. + unsigned Reg = CS.getReg(); + + if (CS.isSpilledToReg()) { + BuildMI(MBB, MBBI, DebugLoc(), TII->get(TargetOpcode::COPY), + CS.getDstReg()) + .addReg(Reg, getKillRegState(true)); + } else { + const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass( + Reg, Reg == RI->getReturnAddressReg(MF) ? MVT::i64 : MVT::i32); + const MachineRegisterInfo &MRI = MF.getRegInfo(); + // If this value was already livein, we probably have a direct use of the + // incoming register value, so don't kill at the spill point. This happens + // since we pass some special inputs (workgroup IDs) in the callee saved + // range. + const bool IsLiveIn = MRI.isLiveIn(Reg); + if (NeedsFrameMoves) + TII->storeRegToStackSlotCFI(MBB, MBBI, Reg, !IsLiveIn, CS.getFrameIdx(), + RC, TRI); + else + TII->storeRegToStackSlot(MBB, MBBI, Reg, !IsLiveIn, CS.getFrameIdx(), + RC, TRI); + } + } + + return true; +} + // Build a CFI MachineInstr for the \p CFIInst CFI rule MachineInstr *SIFrameLowering::buildCFI(MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI, @@ -1704,6 +1744,17 @@ .setMIFlag(Flag); } +MachineInstr *SIFrameLowering::buildCFIForRegToRegSpill( + MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI, + const DebugLoc &DL, const Register Reg, const Register RegCopy) const { + MachineFunction &MF = *MBB.getParent(); + const MCRegisterInfo &MCRI = *MF.getMMI().getContext().getRegisterInfo(); + return buildCFI( + MBB, MBBI, DL, + MCCFIInstruction::createRegister(nullptr, MCRI.getDwarfRegNum(Reg, false), + MCRI.getDwarfRegNum(RegCopy, false))); +} + static void encodeDwarfRegisterLocation(int DwarfReg, raw_ostream &OS) { if (DwarfReg < 32) { OS << uint8_t(dwarf::DW_OP_reg0 + DwarfReg); @@ -1820,6 +1871,42 @@ nullptr, MCRI.getDwarfRegNum(SGPR, false), Offset)); } +MachineInstr *SIFrameLowering::buildCFIForVGPRToVMEMSpill( + MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI, + const DebugLoc &DL, unsigned VGPR, int64_t Offset) const { + const MachineFunction &MF = *MBB.getParent(); + const MCRegisterInfo &MCRI = *MF.getMMI().getContext().getRegisterInfo(); + const GCNSubtarget &ST = MF.getSubtarget(); + + int DwarfVGPR = MCRI.getDwarfRegNum(VGPR, false); + + SmallString<20> Block; + raw_svector_ostream OSBlock(Block); + encodeDwarfRegisterLocation(DwarfVGPR, OSBlock); + OSBlock << uint8_t(dwarf::DW_OP_swap); + OSBlock << uint8_t(dwarf::DW_OP_LLVM_offset_uconst); + encodeULEB128(Offset, OSBlock); + OSBlock << uint8_t(dwarf::DW_OP_LLVM_call_frame_entry_reg); + encodeULEB128(MCRI.getDwarfRegNum( + ST.isWave32() ? AMDGPU::EXEC_LO : AMDGPU::EXEC, false), + OSBlock); + OSBlock << uint8_t(dwarf::DW_OP_deref_size); + OSBlock << uint8_t(ST.getWavefrontSize() / 8); + OSBlock << uint8_t(dwarf::DW_OP_LLVM_select_bit_piece); + encodeULEB128(VGPRLaneBitSize, OSBlock); + encodeULEB128(ST.getWavefrontSize(), OSBlock); + + SmallString<20> CFIInst; + raw_svector_ostream OSCFIInst(CFIInst); + OSCFIInst << uint8_t(dwarf::DW_CFA_expression); + encodeULEB128(DwarfVGPR, OSCFIInst); + encodeULEB128(Block.size(), OSCFIInst); + OSCFIInst << Block; + + return buildCFI(MBB, MBBI, DL, + MCCFIInstruction::createEscape(nullptr, OSCFIInst.str())); +} + MachineInstr *SIFrameLowering::buildCFIForRegToSGPRPairSpill( MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI, const DebugLoc &DL, const Register Reg, const Register SGPRPair) const { diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.h b/llvm/lib/Target/AMDGPU/SIInstrInfo.h --- a/llvm/lib/Target/AMDGPU/SIInstrInfo.h +++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.h @@ -226,12 +226,27 @@ MachineBasicBlock::iterator I, const DebugLoc &DL, Register SrcReg, int Value) const; +private: + void storeRegToStackSlotImpl(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MI, Register SrcReg, + bool isKill, int FrameIndex, + const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI, + bool NeedsCFI) const; + +public: void storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator MI, Register SrcReg, bool isKill, int FrameIndex, const TargetRegisterClass *RC, const TargetRegisterInfo *TRI) const override; + void storeRegToStackSlotCFI(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MI, Register SrcReg, + bool isKill, int FrameIndex, + const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI) const; + void loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator MI, Register DestReg, int FrameIndex, const TargetRegisterClass *RC, diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp --- a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp +++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp @@ -1403,120 +1403,149 @@ return get(getIndirectVGPRWriteMovRelPseudoOpc(VecSize)); } -static unsigned getSGPRSpillSaveOpcode(unsigned Size) { +static unsigned getSGPRSpillSaveOpcode(unsigned Size, bool NeedsCFI) { switch (Size) { case 4: - return AMDGPU::SI_SPILL_S32_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_S32_CFI_SAVE : AMDGPU::SI_SPILL_S32_SAVE; case 8: - return AMDGPU::SI_SPILL_S64_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_S64_CFI_SAVE : AMDGPU::SI_SPILL_S64_SAVE; case 12: - return AMDGPU::SI_SPILL_S96_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_S96_CFI_SAVE : AMDGPU::SI_SPILL_S96_SAVE; case 16: - return AMDGPU::SI_SPILL_S128_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_S128_CFI_SAVE + : AMDGPU::SI_SPILL_S128_SAVE; case 20: - return AMDGPU::SI_SPILL_S160_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_S160_CFI_SAVE + : AMDGPU::SI_SPILL_S160_SAVE; case 24: - return AMDGPU::SI_SPILL_S192_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_S192_CFI_SAVE + : AMDGPU::SI_SPILL_S192_SAVE; case 28: - return AMDGPU::SI_SPILL_S224_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_S224_CFI_SAVE + : AMDGPU::SI_SPILL_S224_SAVE; case 32: - return AMDGPU::SI_SPILL_S256_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_S256_CFI_SAVE + : AMDGPU::SI_SPILL_S256_SAVE; case 64: - return AMDGPU::SI_SPILL_S512_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_S512_CFI_SAVE + : AMDGPU::SI_SPILL_S512_SAVE; case 128: - return AMDGPU::SI_SPILL_S1024_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_S1024_CFI_SAVE + : AMDGPU::SI_SPILL_S1024_SAVE; default: llvm_unreachable("unknown register size"); } } -static unsigned getVGPRSpillSaveOpcode(unsigned Size) { +static unsigned getVGPRSpillSaveOpcode(unsigned Size, bool NeedsCFI) { switch (Size) { case 4: - return AMDGPU::SI_SPILL_V32_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_V32_CFI_SAVE : AMDGPU::SI_SPILL_V32_SAVE; case 8: - return AMDGPU::SI_SPILL_V64_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_V64_CFI_SAVE : AMDGPU::SI_SPILL_V64_SAVE; case 12: - return AMDGPU::SI_SPILL_V96_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_V96_CFI_SAVE : AMDGPU::SI_SPILL_V96_SAVE; case 16: - return AMDGPU::SI_SPILL_V128_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_V128_CFI_SAVE + : AMDGPU::SI_SPILL_V128_SAVE; case 20: - return AMDGPU::SI_SPILL_V160_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_V160_CFI_SAVE + : AMDGPU::SI_SPILL_V160_SAVE; case 24: - return AMDGPU::SI_SPILL_V192_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_V192_CFI_SAVE + : AMDGPU::SI_SPILL_V192_SAVE; case 28: - return AMDGPU::SI_SPILL_V224_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_V224_CFI_SAVE + : AMDGPU::SI_SPILL_V224_SAVE; case 32: - return AMDGPU::SI_SPILL_V256_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_V256_CFI_SAVE + : AMDGPU::SI_SPILL_V256_SAVE; case 64: - return AMDGPU::SI_SPILL_V512_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_V512_CFI_SAVE + : AMDGPU::SI_SPILL_V512_SAVE; case 128: - return AMDGPU::SI_SPILL_V1024_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_V1024_CFI_SAVE + : AMDGPU::SI_SPILL_V1024_SAVE; default: llvm_unreachable("unknown register size"); } } -static unsigned getAGPRSpillSaveOpcode(unsigned Size) { +static unsigned getAGPRSpillSaveOpcode(unsigned Size, bool NeedsCFI) { switch (Size) { case 4: - return AMDGPU::SI_SPILL_A32_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_A32_CFI_SAVE : AMDGPU::SI_SPILL_A32_SAVE; case 8: - return AMDGPU::SI_SPILL_A64_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_A64_CFI_SAVE : AMDGPU::SI_SPILL_A64_SAVE; case 12: - return AMDGPU::SI_SPILL_A96_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_A96_CFI_SAVE : AMDGPU::SI_SPILL_A96_SAVE; case 16: - return AMDGPU::SI_SPILL_A128_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_A128_CFI_SAVE + : AMDGPU::SI_SPILL_A128_SAVE; case 20: - return AMDGPU::SI_SPILL_A160_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_A160_CFI_SAVE + : AMDGPU::SI_SPILL_A160_SAVE; case 24: - return AMDGPU::SI_SPILL_A192_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_A192_CFI_SAVE + : AMDGPU::SI_SPILL_A192_SAVE; case 28: - return AMDGPU::SI_SPILL_A224_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_A224_CFI_SAVE + : AMDGPU::SI_SPILL_A224_SAVE; case 32: - return AMDGPU::SI_SPILL_A256_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_A256_CFI_SAVE + : AMDGPU::SI_SPILL_A256_SAVE; case 64: - return AMDGPU::SI_SPILL_A512_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_A512_CFI_SAVE + : AMDGPU::SI_SPILL_A512_SAVE; case 128: - return AMDGPU::SI_SPILL_A1024_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_A1024_CFI_SAVE + : AMDGPU::SI_SPILL_A1024_SAVE; default: llvm_unreachable("unknown register size"); } } -static unsigned getAVSpillSaveOpcode(unsigned Size) { +static unsigned getAVSpillSaveOpcode(unsigned Size, bool NeedsCFI) { switch (Size) { case 4: - return AMDGPU::SI_SPILL_AV32_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_AV32_CFI_SAVE + : AMDGPU::SI_SPILL_AV32_SAVE; case 8: - return AMDGPU::SI_SPILL_AV64_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_AV64_CFI_SAVE + : AMDGPU::SI_SPILL_AV64_SAVE; case 12: - return AMDGPU::SI_SPILL_AV96_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_AV96_CFI_SAVE + : AMDGPU::SI_SPILL_AV96_SAVE; case 16: - return AMDGPU::SI_SPILL_AV128_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_AV128_CFI_SAVE + : AMDGPU::SI_SPILL_AV128_SAVE; case 20: - return AMDGPU::SI_SPILL_AV160_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_AV160_CFI_SAVE + : AMDGPU::SI_SPILL_AV160_SAVE; case 24: - return AMDGPU::SI_SPILL_AV192_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_AV192_CFI_SAVE + : AMDGPU::SI_SPILL_AV192_SAVE; case 28: - return AMDGPU::SI_SPILL_AV224_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_AV224_CFI_SAVE + : AMDGPU::SI_SPILL_AV224_SAVE; case 32: - return AMDGPU::SI_SPILL_AV256_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_AV256_CFI_SAVE + : AMDGPU::SI_SPILL_AV256_SAVE; case 64: - return AMDGPU::SI_SPILL_AV512_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_AV512_CFI_SAVE + : AMDGPU::SI_SPILL_AV512_SAVE; case 128: - return AMDGPU::SI_SPILL_AV1024_SAVE; + return NeedsCFI ? AMDGPU::SI_SPILL_AV1024_CFI_SAVE + : AMDGPU::SI_SPILL_AV1024_SAVE; default: llvm_unreachable("unknown register size"); } } -void SIInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB, - MachineBasicBlock::iterator MI, - Register SrcReg, bool isKill, - int FrameIndex, - const TargetRegisterClass *RC, - const TargetRegisterInfo *TRI) const { +void SIInstrInfo::storeRegToStackSlotImpl( + MachineBasicBlock &MBB, MachineBasicBlock::iterator MI, Register SrcReg, + bool isKill, int FrameIndex, const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI, bool NeedsCFI) const { MachineFunction *MF = MBB.getParent(); SIMachineFunctionInfo *MFI = MF->getInfo(); MachineFrameInfo &FrameInfo = MF->getFrameInfo(); @@ -1538,7 +1567,8 @@ // We are only allowed to create one new instruction when spilling // registers, so we need to use pseudo instruction for spilling SGPRs. - const MCInstrDesc &OpDesc = get(getSGPRSpillSaveOpcode(SpillSize)); + const MCInstrDesc &OpDesc = + get(getSGPRSpillSaveOpcode(SpillSize, NeedsCFI)); // The SGPR spill/restore instructions only work on number sgprs, so we need // to make sure we are using the correct register class. @@ -1557,9 +1587,11 @@ return; } - unsigned Opcode = RI.isVectorSuperClass(RC) ? getAVSpillSaveOpcode(SpillSize) - : RI.isAGPRClass(RC) ? getAGPRSpillSaveOpcode(SpillSize) - : getVGPRSpillSaveOpcode(SpillSize); + unsigned Opcode = RI.isVectorSuperClass(RC) + ? getAVSpillSaveOpcode(SpillSize, NeedsCFI) + : RI.isAGPRClass(RC) + ? getAGPRSpillSaveOpcode(SpillSize, NeedsCFI) + : getVGPRSpillSaveOpcode(SpillSize, NeedsCFI); MFI->setHasSpilledVGPRs(); BuildMI(MBB, MI, DL, get(Opcode)) @@ -1570,6 +1602,26 @@ .addMemOperand(MMO); } +void SIInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MI, + Register SrcReg, bool isKill, + int FrameIndex, + const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI) const { + storeRegToStackSlotImpl(MBB, MI, SrcReg, isKill, FrameIndex, RC, TRI, + /*NeedsCFI*/ false); +} + +void SIInstrInfo::storeRegToStackSlotCFI(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MI, + Register SrcReg, bool isKill, + int FrameIndex, + const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI) const { + storeRegToStackSlotImpl(MBB, MI, SrcReg, isKill, FrameIndex, RC, TRI, + /*NeedsCFI*/ true); +} + static unsigned getSGPRSpillRestoreOpcode(unsigned Size) { switch (Size) { case 4: diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td --- a/llvm/lib/Target/AMDGPU/SIInstructions.td +++ b/llvm/lib/Target/AMDGPU/SIInstructions.td @@ -748,6 +748,13 @@ let mayLoad = 0; } + def _CFI_SAVE : PseudoInstSI < + (outs), + (ins sgpr_class:$data, i32imm:$addr)> { + let mayStore = 1; + let mayLoad = 0; + } + def _RESTORE : PseudoInstSI < (outs sgpr_class:$data), (ins i32imm:$addr)> { @@ -790,6 +797,18 @@ let Size = !if(!le(MaxSize, 256), MaxSize, 252); } + def _CFI_SAVE : VPseudoInstSI < + (outs), + (ins vgpr_class:$vdata, i32imm:$vaddr, + SReg_32:$soffset, i32imm:$offset)> { + let mayStore = 1; + let mayLoad = 0; + // (2 * 4) + (8 * num_subregs) bytes maximum + int MaxSize = !add(!shl(!srl(vgpr_class.Size, 5), !add(UsesTmp, 3)), 8); + // Size field is unsigned char and cannot fit more. + let Size = !if(!le(MaxSize, 256), MaxSize, 252); + } + def _RESTORE : VPseudoInstSI < (outs vgpr_class:$vdata), (ins i32imm:$vaddr, diff --git a/llvm/lib/Target/AMDGPU/SILowerSGPRSpills.cpp b/llvm/lib/Target/AMDGPU/SILowerSGPRSpills.cpp --- a/llvm/lib/Target/AMDGPU/SILowerSGPRSpills.cpp +++ b/llvm/lib/Target/AMDGPU/SILowerSGPRSpills.cpp @@ -73,44 +73,19 @@ char &llvm::SILowerSGPRSpillsID = SILowerSGPRSpills::ID; /// Insert spill code for the callee-saved registers used in the function. -static void insertCSRSaves(MachineBasicBlock &SaveBlock, - ArrayRef CSI, - LiveIntervals *LIS) { - MachineFunction &MF = *SaveBlock.getParent(); - const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo(); - const TargetFrameLowering *TFI = MF.getSubtarget().getFrameLowering(); - const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo(); - const GCNSubtarget &ST = MF.getSubtarget(); - const SIRegisterInfo *RI = ST.getRegisterInfo(); - +static void insertCSRSaves(const GCNSubtarget &ST, MachineBasicBlock &SaveBlock, + ArrayRef CSI, LiveIntervals *LIS) { + const TargetFrameLowering *TFI = ST.getFrameLowering(); + const TargetRegisterInfo *TRI = ST.getRegisterInfo(); MachineBasicBlock::iterator I = SaveBlock.begin(); - if (!TFI->spillCalleeSavedRegisters(SaveBlock, I, CSI, TRI)) { - const MachineRegisterInfo &MRI = MF.getRegInfo(); - - for (const CalleeSavedInfo &CS : CSI) { - // Insert the spill to the stack frame. - MCRegister Reg = CS.getReg(); - - MachineInstrSpan MIS(I, &SaveBlock); - const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass( - Reg, Reg == RI->getReturnAddressReg(MF) ? MVT::i64 : MVT::i32); - - // If this value was already livein, we probably have a direct use of the - // incoming register value, so don't kill at the spill point. This happens - // since we pass some special inputs (workgroup IDs) in the callee saved - // range. - const bool IsLiveIn = MRI.isLiveIn(Reg); - TII.storeRegToStackSlot(SaveBlock, I, Reg, !IsLiveIn, CS.getFrameIdx(), - RC, TRI); - - if (LIS) { - assert(std::distance(MIS.begin(), I) == 1); - MachineInstr &Inst = *std::prev(I); - - LIS->InsertMachineInstrInMaps(Inst); - LIS->removeAllRegUnitsForPhysReg(Reg); - } - } + MachineInstrSpan MIS(I, &SaveBlock); + bool Success = TFI->spillCalleeSavedRegisters(SaveBlock, I, CSI, TRI); + assert(Success && "spillCalleeSavedRegisters should always succeed"); + (void)Success; + + if (LIS) { + for (MachineInstr &Inst : make_range(MIS.begin(), I)) + LIS->InsertMachineInstrInMaps(Inst); } } @@ -212,11 +187,19 @@ std::vector CSI; const MCPhysReg *CSRegs = MRI.getCalleeSavedRegs(); + Register RetAddrReg = TRI->getReturnAddressReg(MF); + bool SpillRetAddrReg = false; for (unsigned I = 0; CSRegs[I]; ++I) { MCRegister Reg = CSRegs[I]; if (SavedRegs.test(Reg)) { + if (Reg == TRI->getSubReg(RetAddrReg, AMDGPU::sub0) || + Reg == TRI->getSubReg(RetAddrReg, AMDGPU::sub1)) { + SpillRetAddrReg = true; + continue; + } + const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg, MVT::i32); int JunkFI = MFI.CreateStackObject(TRI->getSpillSize(*RC), @@ -226,9 +209,20 @@ } } + // Return address uses a register pair. Add the super register to the + // CSI list so that it's easier to identify the entire spill and CFI + // can be emitted appropriately. + if (SpillRetAddrReg) { + const TargetRegisterClass *RC = + TRI->getMinimalPhysRegClass(RetAddrReg, MVT::i64); + int JunkFI = MFI.CreateStackObject(TRI->getSpillSize(*RC), + TRI->getSpillAlign(*RC), true); + CSI.push_back(CalleeSavedInfo(RetAddrReg, JunkFI)); + } + if (!CSI.empty()) { for (MachineBasicBlock *SaveBlock : SaveBlocks) - insertCSRSaves(*SaveBlock, CSI, LIS); + insertCSRSaves(ST, *SaveBlock, CSI, LIS); // Add live ins to save blocks. assert(SaveBlocks.size() == 1 && "shrink wrapping not fully implemented"); diff --git a/llvm/lib/Target/AMDGPU/SIRegisterInfo.h b/llvm/lib/Target/AMDGPU/SIRegisterInfo.h --- a/llvm/lib/Target/AMDGPU/SIRegisterInfo.h +++ b/llvm/lib/Target/AMDGPU/SIRegisterInfo.h @@ -137,10 +137,9 @@ bool IsLoad, bool IsKill = true) const; /// If \p OnlyToVGPR is true, this will only succeed if this - bool spillSGPR(MachineBasicBlock::iterator MI, - int FI, RegScavenger *RS, - LiveIntervals *LIS = nullptr, - bool OnlyToVGPR = false) const; + bool spillSGPR(MachineBasicBlock::iterator MI, int FI, RegScavenger *RS, + LiveIntervals *LIS = nullptr, bool OnlyToVGPR = false, + bool NeedsCFI = false) const; bool restoreSGPR(MachineBasicBlock::iterator MI, int FI, RegScavenger *RS, @@ -412,8 +411,8 @@ unsigned LoadStoreOp, int Index, Register ValueReg, bool ValueIsKill, MCRegister ScratchOffsetReg, int64_t InstrOffset, MachineMemOperand *MMO, - RegScavenger *RS, - LivePhysRegs *LiveRegs = nullptr) const; + RegScavenger *RS, LivePhysRegs *LiveRegs = nullptr, + bool NeedsCFI = false) const; }; } // End namespace llvm diff --git a/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp b/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp --- a/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp +++ b/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp @@ -939,93 +939,133 @@ switch (Op) { case AMDGPU::SI_SPILL_S1024_SAVE: + case AMDGPU::SI_SPILL_S1024_CFI_SAVE: case AMDGPU::SI_SPILL_S1024_RESTORE: case AMDGPU::SI_SPILL_V1024_SAVE: + case AMDGPU::SI_SPILL_V1024_CFI_SAVE: case AMDGPU::SI_SPILL_V1024_RESTORE: case AMDGPU::SI_SPILL_A1024_SAVE: + case AMDGPU::SI_SPILL_A1024_CFI_SAVE: case AMDGPU::SI_SPILL_A1024_RESTORE: case AMDGPU::SI_SPILL_AV1024_SAVE: + case AMDGPU::SI_SPILL_AV1024_CFI_SAVE: case AMDGPU::SI_SPILL_AV1024_RESTORE: return 32; case AMDGPU::SI_SPILL_S512_SAVE: + case AMDGPU::SI_SPILL_S512_CFI_SAVE: case AMDGPU::SI_SPILL_S512_RESTORE: case AMDGPU::SI_SPILL_V512_SAVE: + case AMDGPU::SI_SPILL_V512_CFI_SAVE: case AMDGPU::SI_SPILL_V512_RESTORE: case AMDGPU::SI_SPILL_A512_SAVE: + case AMDGPU::SI_SPILL_A512_CFI_SAVE: case AMDGPU::SI_SPILL_A512_RESTORE: case AMDGPU::SI_SPILL_AV512_SAVE: + case AMDGPU::SI_SPILL_AV512_CFI_SAVE: case AMDGPU::SI_SPILL_AV512_RESTORE: return 16; case AMDGPU::SI_SPILL_S256_SAVE: + case AMDGPU::SI_SPILL_S256_CFI_SAVE: case AMDGPU::SI_SPILL_S256_RESTORE: case AMDGPU::SI_SPILL_V256_SAVE: + case AMDGPU::SI_SPILL_V256_CFI_SAVE: case AMDGPU::SI_SPILL_V256_RESTORE: case AMDGPU::SI_SPILL_A256_SAVE: + case AMDGPU::SI_SPILL_A256_CFI_SAVE: case AMDGPU::SI_SPILL_A256_RESTORE: case AMDGPU::SI_SPILL_AV256_SAVE: + case AMDGPU::SI_SPILL_AV256_CFI_SAVE: case AMDGPU::SI_SPILL_AV256_RESTORE: return 8; case AMDGPU::SI_SPILL_S224_SAVE: + case AMDGPU::SI_SPILL_S224_CFI_SAVE: case AMDGPU::SI_SPILL_S224_RESTORE: case AMDGPU::SI_SPILL_V224_SAVE: + case AMDGPU::SI_SPILL_V224_CFI_SAVE: case AMDGPU::SI_SPILL_V224_RESTORE: case AMDGPU::SI_SPILL_A224_SAVE: + case AMDGPU::SI_SPILL_A224_CFI_SAVE: case AMDGPU::SI_SPILL_A224_RESTORE: case AMDGPU::SI_SPILL_AV224_SAVE: + case AMDGPU::SI_SPILL_AV224_CFI_SAVE: case AMDGPU::SI_SPILL_AV224_RESTORE: return 7; case AMDGPU::SI_SPILL_S192_SAVE: + case AMDGPU::SI_SPILL_S192_CFI_SAVE: case AMDGPU::SI_SPILL_S192_RESTORE: case AMDGPU::SI_SPILL_V192_SAVE: + case AMDGPU::SI_SPILL_V192_CFI_SAVE: case AMDGPU::SI_SPILL_V192_RESTORE: case AMDGPU::SI_SPILL_A192_SAVE: + case AMDGPU::SI_SPILL_A192_CFI_SAVE: case AMDGPU::SI_SPILL_A192_RESTORE: case AMDGPU::SI_SPILL_AV192_SAVE: + case AMDGPU::SI_SPILL_AV192_CFI_SAVE: case AMDGPU::SI_SPILL_AV192_RESTORE: return 6; case AMDGPU::SI_SPILL_S160_SAVE: + case AMDGPU::SI_SPILL_S160_CFI_SAVE: case AMDGPU::SI_SPILL_S160_RESTORE: case AMDGPU::SI_SPILL_V160_SAVE: + case AMDGPU::SI_SPILL_V160_CFI_SAVE: case AMDGPU::SI_SPILL_V160_RESTORE: case AMDGPU::SI_SPILL_A160_SAVE: + case AMDGPU::SI_SPILL_A160_CFI_SAVE: case AMDGPU::SI_SPILL_A160_RESTORE: case AMDGPU::SI_SPILL_AV160_SAVE: + case AMDGPU::SI_SPILL_AV160_CFI_SAVE: case AMDGPU::SI_SPILL_AV160_RESTORE: return 5; case AMDGPU::SI_SPILL_S128_SAVE: + case AMDGPU::SI_SPILL_S128_CFI_SAVE: case AMDGPU::SI_SPILL_S128_RESTORE: case AMDGPU::SI_SPILL_V128_SAVE: + case AMDGPU::SI_SPILL_V128_CFI_SAVE: case AMDGPU::SI_SPILL_V128_RESTORE: case AMDGPU::SI_SPILL_A128_SAVE: + case AMDGPU::SI_SPILL_A128_CFI_SAVE: case AMDGPU::SI_SPILL_A128_RESTORE: case AMDGPU::SI_SPILL_AV128_SAVE: + case AMDGPU::SI_SPILL_AV128_CFI_SAVE: case AMDGPU::SI_SPILL_AV128_RESTORE: return 4; case AMDGPU::SI_SPILL_S96_SAVE: + case AMDGPU::SI_SPILL_S96_CFI_SAVE: case AMDGPU::SI_SPILL_S96_RESTORE: case AMDGPU::SI_SPILL_V96_SAVE: + case AMDGPU::SI_SPILL_V96_CFI_SAVE: case AMDGPU::SI_SPILL_V96_RESTORE: case AMDGPU::SI_SPILL_A96_SAVE: + case AMDGPU::SI_SPILL_A96_CFI_SAVE: case AMDGPU::SI_SPILL_A96_RESTORE: case AMDGPU::SI_SPILL_AV96_SAVE: + case AMDGPU::SI_SPILL_AV96_CFI_SAVE: case AMDGPU::SI_SPILL_AV96_RESTORE: return 3; case AMDGPU::SI_SPILL_S64_SAVE: + case AMDGPU::SI_SPILL_S64_CFI_SAVE: case AMDGPU::SI_SPILL_S64_RESTORE: case AMDGPU::SI_SPILL_V64_SAVE: + case AMDGPU::SI_SPILL_V64_CFI_SAVE: case AMDGPU::SI_SPILL_V64_RESTORE: case AMDGPU::SI_SPILL_A64_SAVE: + case AMDGPU::SI_SPILL_A64_CFI_SAVE: case AMDGPU::SI_SPILL_A64_RESTORE: case AMDGPU::SI_SPILL_AV64_SAVE: + case AMDGPU::SI_SPILL_AV64_CFI_SAVE: case AMDGPU::SI_SPILL_AV64_RESTORE: return 2; case AMDGPU::SI_SPILL_S32_SAVE: + case AMDGPU::SI_SPILL_S32_CFI_SAVE: case AMDGPU::SI_SPILL_S32_RESTORE: case AMDGPU::SI_SPILL_V32_SAVE: + case AMDGPU::SI_SPILL_V32_CFI_SAVE: case AMDGPU::SI_SPILL_V32_RESTORE: case AMDGPU::SI_SPILL_A32_SAVE: + case AMDGPU::SI_SPILL_A32_CFI_SAVE: case AMDGPU::SI_SPILL_A32_RESTORE: case AMDGPU::SI_SPILL_AV32_SAVE: + case AMDGPU::SI_SPILL_AV32_CFI_SAVE: case AMDGPU::SI_SPILL_AV32_RESTORE: return 1; default: llvm_unreachable("Invalid spill opcode"); @@ -1148,14 +1188,14 @@ } } -static MachineInstrBuilder spillVGPRtoAGPR(const GCNSubtarget &ST, - MachineBasicBlock &MBB, - MachineBasicBlock::iterator MI, - int Index, unsigned Lane, - unsigned ValueReg, bool IsKill) { +static MachineInstrBuilder +spillVGPRtoAGPR(const GCNSubtarget &ST, MachineBasicBlock &MBB, + MachineBasicBlock::iterator MI, int Index, unsigned Lane, + unsigned ValueReg, bool IsKill, bool NeedsCFI) { MachineFunction *MF = MBB.getParent(); SIMachineFunctionInfo *MFI = MF->getInfo(); const SIInstrInfo *TII = ST.getInstrInfo(); + const SIFrameLowering *TFL = ST.getFrameLowering(); MCPhysReg Reg = MFI->getVGPRToAGPRSpill(Index, Lane); @@ -1178,6 +1218,8 @@ auto CopyMIB = BuildMI(MBB, MI, DL, TII->get(AMDGPU::COPY), Dst) .addReg(Src, getKillRegState(IsKill)); CopyMIB->setAsmPrinterFlag(MachineInstr::ReloadReuse); + if (NeedsCFI) + TFL->buildCFIForRegToRegSpill(MBB, MI, DL, Src, Dst); return CopyMIB; } unsigned Opc = (IsStore ^ IsVGPR) ? AMDGPU::V_ACCVGPR_WRITE_B32_e64 @@ -1186,6 +1228,8 @@ auto MIB = BuildMI(MBB, MI, DL, TII->get(Opc), Dst) .addReg(Src, getKillRegState(IsKill)); MIB->setAsmPrinterFlag(MachineInstr::ReloadReuse); + if (NeedsCFI) + TFL->buildCFIForRegToRegSpill(MBB, MI, DL, Src, Dst); return MIB; } @@ -1208,7 +1252,8 @@ return false; const MachineOperand *Reg = TII->getNamedOperand(*MI, AMDGPU::OpName::vdata); - if (spillVGPRtoAGPR(ST, *MBB, MI, Index, 0, Reg->getReg(), false).getInstr()) + if (spillVGPRtoAGPR(ST, *MBB, MI, Index, 0, Reg->getReg(), false, false) + .getInstr()) return true; MachineInstrBuilder NewMI = @@ -1271,12 +1316,13 @@ MachineBasicBlock &MBB, MachineBasicBlock::iterator MI, const DebugLoc &DL, unsigned LoadStoreOp, int Index, Register ValueReg, bool IsKill, MCRegister ScratchOffsetReg, int64_t InstOffset, MachineMemOperand *MMO, - RegScavenger *RS, LivePhysRegs *LiveRegs) const { + RegScavenger *RS, LivePhysRegs *LiveRegs, bool NeedsCFI) const { assert((!RS || !LiveRegs) && "Only RS or LiveRegs can be set but not both"); MachineFunction *MF = MBB.getParent(); const SIInstrInfo *TII = ST.getInstrInfo(); const MachineFrameInfo &MFI = MF->getFrameInfo(); + const SIFrameLowering *TFL = ST.getFrameLowering(); const SIMachineFunctionInfo *FuncInfo = MF->getInfo(); const MCInstrDesc *Desc = &TII->get(LoadStoreOp); @@ -1304,6 +1350,7 @@ int64_t MaxOffset = Offset + Size + RemSize - EltSize; int64_t ScratchOffsetRegDelta = 0; + int64_t AdditionalCFIOffset = 0; if (IsFlat && EltSize > 4) { LoadStoreOp = getFlatScratchSpillOpcode(TII, LoadStoreOp, EltSize); @@ -1416,6 +1463,7 @@ Scavenged = true; } + AdditionalCFIOffset = Offset; // We currently only support spilling VGPRs to EltSize boundaries, meaning // we can simplify the adjustment of Offset here to just scale with // WavefrontSize. @@ -1516,7 +1564,8 @@ Register Sub = IsSubReg ? Register(getSubReg(ValueReg, getSubRegFromChannel(Lane))) : ValueReg; - auto MIB = spillVGPRtoAGPR(ST, MBB, MI, Index, Lane, Sub, IsKill); + auto MIB = + spillVGPRtoAGPR(ST, MBB, MI, Index, Lane, Sub, IsKill, NeedsCFI); if (!MIB.getInstr()) break; if (NeedSuperRegDef || (IsSubReg && IsStore && Lane == LaneS && !i)) { @@ -1611,6 +1660,11 @@ .addImm(0); // swz MIB.addMemOperand(NewMMO); + if (IsStore && NeedsCFI) + TFL->buildCFIForVGPRToVMEMSpill( + MBB, MI, DebugLoc(), SubReg, + (Offset + RegOffset) * ST.getWavefrontSize() + AdditionalCFIOffset); + if (!IsAGPR && NeedSuperRegDef) MIB.addReg(ValueReg, RegState::ImplicitDefine); @@ -1666,11 +1720,9 @@ } } -bool SIRegisterInfo::spillSGPR(MachineBasicBlock::iterator MI, - int Index, - RegScavenger *RS, - LiveIntervals *LIS, - bool OnlyToVGPR) const { +bool SIRegisterInfo::spillSGPR(MachineBasicBlock::iterator MI, int Index, + RegScavenger *RS, LiveIntervals *LIS, + bool OnlyToVGPR, bool NeedsCFI) const { SGPRSpillBuilder SB(*this, *ST.getInstrInfo(), isWave32, MI, Index, RS); ArrayRef VGPRSpills = SB.MFI.getSGPRToVGPRSpills(Index); @@ -1678,6 +1730,8 @@ if (OnlyToVGPR && !SpillToVGPR) return false; + const SIFrameLowering *TFL = ST.getFrameLowering(); + assert(SpillToVGPR || (SB.SuperReg != SB.MFI.getStackPtrOffsetReg() && SB.SuperReg != SB.MFI.getFrameOffsetReg())); @@ -1702,11 +1756,27 @@ .addReg(SubReg, getKillRegState(UseKill)) .addImm(Spill.Lane) .addReg(Spill.VGPR); + + MachineInstr *CFI = nullptr; + if (NeedsCFI) { + if (SB.SuperReg == SB.TRI.getReturnAddressReg(SB.MF)) { + if (i == e - 1) + CFI = TFL->buildCFIForSGPRToVGPRSpill(*SB.MBB, MI, DebugLoc(), + AMDGPU::PC_REG, VGPRSpills); + } else { + CFI = TFL->buildCFIForSGPRToVGPRSpill(*SB.MBB, MI, DebugLoc(), SubReg, + Spill.VGPR, Spill.Lane); + } + } + if (LIS) { if (i == 0) LIS->ReplaceMachineInstrInMaps(*MI, *MIB); else LIS->InsertMachineInstrInMaps(*MIB); + + if (CFI) + LIS->InsertMachineInstrInMaps(*CFI); } if (i == 0 && SB.NumSubRegs > 1) { @@ -1771,6 +1841,18 @@ // Write out VGPR SB.readWriteTmpVGPR(Offset, /*IsLoad*/ false); + + // TODO: Implement CFI for SpillToVMEM for all scenarios. + MachineInstr *CFI = nullptr; + if (NeedsCFI && SB.SuperReg == SB.TRI.getReturnAddressReg(SB.MF)) { + int64_t CFIOffset = (Offset * SB.EltSize + + SB.MF.getFrameInfo().getObjectOffset(Index)) * + ST.getWavefrontSize(); + CFI = TFL->buildCFIForSGPRToVMEMSpill(*SB.MBB, MI, DebugLoc(), + AMDGPU::PC_REG, CFIOffset); + } + if (LIS && CFI) + LIS->InsertMachineInstrInMaps(*CFI); } SB.restore(); @@ -1942,7 +2024,20 @@ int FI, RegScavenger *RS, LiveIntervals *LIS) const { + bool NeedsCFI = false; switch (MI->getOpcode()) { + case AMDGPU::SI_SPILL_S1024_CFI_SAVE: + case AMDGPU::SI_SPILL_S512_CFI_SAVE: + case AMDGPU::SI_SPILL_S256_CFI_SAVE: + case AMDGPU::SI_SPILL_S224_CFI_SAVE: + case AMDGPU::SI_SPILL_S192_CFI_SAVE: + case AMDGPU::SI_SPILL_S160_CFI_SAVE: + case AMDGPU::SI_SPILL_S128_CFI_SAVE: + case AMDGPU::SI_SPILL_S96_CFI_SAVE: + case AMDGPU::SI_SPILL_S64_CFI_SAVE: + case AMDGPU::SI_SPILL_S32_CFI_SAVE: + NeedsCFI = true; + LLVM_FALLTHROUGH; case AMDGPU::SI_SPILL_S1024_SAVE: case AMDGPU::SI_SPILL_S512_SAVE: case AMDGPU::SI_SPILL_S256_SAVE: @@ -1953,7 +2048,7 @@ case AMDGPU::SI_SPILL_S96_SAVE: case AMDGPU::SI_SPILL_S64_SAVE: case AMDGPU::SI_SPILL_S32_SAVE: - return spillSGPR(MI, FI, RS, LIS, true); + return spillSGPR(MI, FI, RS, LIS, true, NeedsCFI); case AMDGPU::SI_SPILL_S1024_RESTORE: case AMDGPU::SI_SPILL_S512_RESTORE: case AMDGPU::SI_SPILL_S256_RESTORE: @@ -1989,8 +2084,23 @@ ? getBaseRegister() : getFrameRegister(*MF); + bool NeedsCFI = false; + switch (MI->getOpcode()) { // SGPR register spill + case AMDGPU::SI_SPILL_S1024_CFI_SAVE: + case AMDGPU::SI_SPILL_S512_CFI_SAVE: + case AMDGPU::SI_SPILL_S256_CFI_SAVE: + case AMDGPU::SI_SPILL_S224_CFI_SAVE: + case AMDGPU::SI_SPILL_S192_CFI_SAVE: + case AMDGPU::SI_SPILL_S160_CFI_SAVE: + case AMDGPU::SI_SPILL_S128_CFI_SAVE: + case AMDGPU::SI_SPILL_S96_CFI_SAVE: + case AMDGPU::SI_SPILL_S64_CFI_SAVE: + case AMDGPU::SI_SPILL_S32_CFI_SAVE: { + NeedsCFI = true; + LLVM_FALLTHROUGH; + } case AMDGPU::SI_SPILL_S1024_SAVE: case AMDGPU::SI_SPILL_S512_SAVE: case AMDGPU::SI_SPILL_S256_SAVE: @@ -2001,7 +2111,7 @@ case AMDGPU::SI_SPILL_S96_SAVE: case AMDGPU::SI_SPILL_S64_SAVE: case AMDGPU::SI_SPILL_S32_SAVE: { - spillSGPR(MI, Index, RS); + spillSGPR(MI, Index, RS, nullptr, false, NeedsCFI); break; } @@ -2021,6 +2131,38 @@ } // VGPR register spill + case AMDGPU::SI_SPILL_V1024_CFI_SAVE: + case AMDGPU::SI_SPILL_V512_CFI_SAVE: + case AMDGPU::SI_SPILL_V256_CFI_SAVE: + case AMDGPU::SI_SPILL_V224_CFI_SAVE: + case AMDGPU::SI_SPILL_V192_CFI_SAVE: + case AMDGPU::SI_SPILL_V160_CFI_SAVE: + case AMDGPU::SI_SPILL_V128_CFI_SAVE: + case AMDGPU::SI_SPILL_V96_CFI_SAVE: + case AMDGPU::SI_SPILL_V64_CFI_SAVE: + case AMDGPU::SI_SPILL_V32_CFI_SAVE: + case AMDGPU::SI_SPILL_A1024_CFI_SAVE: + case AMDGPU::SI_SPILL_A512_CFI_SAVE: + case AMDGPU::SI_SPILL_A256_CFI_SAVE: + case AMDGPU::SI_SPILL_A224_CFI_SAVE: + case AMDGPU::SI_SPILL_A192_CFI_SAVE: + case AMDGPU::SI_SPILL_A160_CFI_SAVE: + case AMDGPU::SI_SPILL_A128_CFI_SAVE: + case AMDGPU::SI_SPILL_A96_CFI_SAVE: + case AMDGPU::SI_SPILL_A64_CFI_SAVE: + case AMDGPU::SI_SPILL_A32_CFI_SAVE: + case AMDGPU::SI_SPILL_AV1024_CFI_SAVE: + case AMDGPU::SI_SPILL_AV512_CFI_SAVE: + case AMDGPU::SI_SPILL_AV256_CFI_SAVE: + case AMDGPU::SI_SPILL_AV224_CFI_SAVE: + case AMDGPU::SI_SPILL_AV192_CFI_SAVE: + case AMDGPU::SI_SPILL_AV160_CFI_SAVE: + case AMDGPU::SI_SPILL_AV128_CFI_SAVE: + case AMDGPU::SI_SPILL_AV96_CFI_SAVE: + case AMDGPU::SI_SPILL_AV64_CFI_SAVE: + case AMDGPU::SI_SPILL_AV32_CFI_SAVE: + NeedsCFI = true; + LLVM_FALLTHROUGH; case AMDGPU::SI_SPILL_V1024_SAVE: case AMDGPU::SI_SPILL_V512_SAVE: case AMDGPU::SI_SPILL_V256_SAVE: @@ -2058,11 +2200,11 @@ unsigned Opc = ST.enableFlatScratch() ? AMDGPU::SCRATCH_STORE_DWORD_SADDR : AMDGPU::BUFFER_STORE_DWORD_OFFSET; - auto *MBB = MI->getParent(); + auto MBB = MI->getParent(); buildSpillLoadStore( *MBB, MI, DL, Opc, Index, VData->getReg(), VData->isKill(), FrameReg, TII->getNamedOperand(*MI, AMDGPU::OpName::offset)->getImm(), - *MI->memoperands_begin(), RS); + *MI->memoperands_begin(), RS, nullptr, NeedsCFI); MFI->addToSpilledVGPRs(getNumSubRegsForSpillOp(MI->getOpcode())); MI->eraseFromParent(); break; @@ -2104,7 +2246,7 @@ unsigned Opc = ST.enableFlatScratch() ? AMDGPU::SCRATCH_LOAD_DWORD_SADDR : AMDGPU::BUFFER_LOAD_DWORD_OFFSET; - auto *MBB = MI->getParent(); + auto MBB = MI->getParent(); buildSpillLoadStore( *MBB, MI, DL, Opc, Index, VData->getReg(), VData->isKill(), FrameReg, TII->getNamedOperand(*MI, AMDGPU::OpName::offset)->getImm(), diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/assert-align.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/assert-align.ll --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/assert-align.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/assert-align.ll @@ -12,20 +12,20 @@ ; CHECK-NEXT: s_mov_b64 exec, s[16:17] ; CHECK-NEXT: v_writelane_b32 v40, s33, 2 ; CHECK-NEXT: s_mov_b32 s33, s32 -; CHECK-NEXT: s_addk_i32 s32, 0x400 ; CHECK-NEXT: v_writelane_b32 v40, s30, 0 +; CHECK-NEXT: s_addk_i32 s32, 0x400 +; CHECK-NEXT: v_writelane_b32 v40, s31, 1 ; CHECK-NEXT: v_mov_b32_e32 v0, 0 ; CHECK-NEXT: v_mov_b32_e32 v1, 0 -; CHECK-NEXT: v_writelane_b32 v40, s31, 1 ; CHECK-NEXT: s_getpc_b64 s[16:17] ; CHECK-NEXT: s_add_u32 s16, s16, ext@rel32@lo+4 ; CHECK-NEXT: s_addc_u32 s17, s17, ext@rel32@hi+12 ; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17] ; CHECK-NEXT: v_mov_b32_e32 v2, 0 +; CHECK-NEXT: v_readlane_b32 s30, v40, 0 ; CHECK-NEXT: global_store_dword v[0:1], v2, off ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: v_readlane_b32 s31, v40, 1 -; CHECK-NEXT: v_readlane_b32 s30, v40, 0 ; CHECK-NEXT: s_addk_i32 s32, 0xfc00 ; CHECK-NEXT: v_readlane_b32 s33, v40, 2 ; CHECK-NEXT: s_or_saveexec_b64 s[4:5], -1 diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/call-outgoing-stack-args.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/call-outgoing-stack-args.ll --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/call-outgoing-stack-args.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/call-outgoing-stack-args.ll @@ -241,23 +241,23 @@ ; MUBUF-NEXT: s_mov_b64 exec, s[4:5] ; MUBUF-NEXT: v_writelane_b32 v40, s33, 2 ; MUBUF-NEXT: s_mov_b32 s33, s32 +; MUBUF-NEXT: v_writelane_b32 v40, s30, 0 ; MUBUF-NEXT: s_addk_i32 s32, 0x400 +; MUBUF-NEXT: v_writelane_b32 v40, s31, 1 ; MUBUF-NEXT: v_mov_b32_e32 v0, 9 ; MUBUF-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:4 ; MUBUF-NEXT: v_mov_b32_e32 v0, 10 ; MUBUF-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:8 ; MUBUF-NEXT: v_mov_b32_e32 v0, 11 -; MUBUF-NEXT: v_writelane_b32 v40, s30, 0 ; MUBUF-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:12 ; MUBUF-NEXT: v_mov_b32_e32 v0, 12 -; MUBUF-NEXT: v_writelane_b32 v40, s31, 1 ; MUBUF-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:16 ; MUBUF-NEXT: s_getpc_b64 s[4:5] ; MUBUF-NEXT: s_add_u32 s4, s4, external_void_func_v16i32_v16i32_v4i32@rel32@lo+4 ; MUBUF-NEXT: s_addc_u32 s5, s5, external_void_func_v16i32_v16i32_v4i32@rel32@hi+12 ; MUBUF-NEXT: s_swappc_b64 s[30:31], s[4:5] -; MUBUF-NEXT: v_readlane_b32 s31, v40, 1 ; MUBUF-NEXT: v_readlane_b32 s30, v40, 0 +; MUBUF-NEXT: v_readlane_b32 s31, v40, 1 ; MUBUF-NEXT: s_addk_i32 s32, 0xfc00 ; MUBUF-NEXT: v_readlane_b32 s33, v40, 2 ; MUBUF-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -274,23 +274,23 @@ ; FLATSCR-NEXT: s_mov_b64 exec, s[0:1] ; FLATSCR-NEXT: v_writelane_b32 v40, s33, 2 ; FLATSCR-NEXT: s_mov_b32 s33, s32 +; FLATSCR-NEXT: v_writelane_b32 v40, s30, 0 ; FLATSCR-NEXT: s_add_i32 s32, s32, 16 +; FLATSCR-NEXT: v_writelane_b32 v40, s31, 1 ; FLATSCR-NEXT: v_mov_b32_e32 v0, 9 ; FLATSCR-NEXT: scratch_store_dword off, v0, s32 offset:4 ; FLATSCR-NEXT: v_mov_b32_e32 v0, 10 ; FLATSCR-NEXT: scratch_store_dword off, v0, s32 offset:8 ; FLATSCR-NEXT: v_mov_b32_e32 v0, 11 -; FLATSCR-NEXT: v_writelane_b32 v40, s30, 0 ; FLATSCR-NEXT: scratch_store_dword off, v0, s32 offset:12 ; FLATSCR-NEXT: v_mov_b32_e32 v0, 12 -; FLATSCR-NEXT: v_writelane_b32 v40, s31, 1 ; FLATSCR-NEXT: scratch_store_dword off, v0, s32 offset:16 ; FLATSCR-NEXT: s_getpc_b64 s[0:1] ; FLATSCR-NEXT: s_add_u32 s0, s0, external_void_func_v16i32_v16i32_v4i32@rel32@lo+4 ; FLATSCR-NEXT: s_addc_u32 s1, s1, external_void_func_v16i32_v16i32_v4i32@rel32@hi+12 ; FLATSCR-NEXT: s_swappc_b64 s[30:31], s[0:1] -; FLATSCR-NEXT: v_readlane_b32 s31, v40, 1 ; FLATSCR-NEXT: v_readlane_b32 s30, v40, 0 +; FLATSCR-NEXT: v_readlane_b32 s31, v40, 1 ; FLATSCR-NEXT: s_add_i32 s32, s32, -16 ; FLATSCR-NEXT: v_readlane_b32 s33, v40, 2 ; FLATSCR-NEXT: s_or_saveexec_b64 s[0:1], -1 @@ -311,11 +311,11 @@ ; MUBUF-NEXT: s_mov_b64 exec, s[4:5] ; MUBUF-NEXT: v_writelane_b32 v40, s33, 2 ; MUBUF-NEXT: s_mov_b32 s33, s32 -; MUBUF-NEXT: buffer_load_dword v1, v0, s[0:3], 0 offen -; MUBUF-NEXT: buffer_load_dword v2, v0, s[0:3], 0 offen offset:4 -; MUBUF-NEXT: s_addk_i32 s32, 0x400 ; MUBUF-NEXT: v_writelane_b32 v40, s30, 0 +; MUBUF-NEXT: s_addk_i32 s32, 0x400 ; MUBUF-NEXT: v_writelane_b32 v40, s31, 1 +; MUBUF-NEXT: buffer_load_dword v1, v0, s[0:3], 0 offen +; MUBUF-NEXT: buffer_load_dword v2, v0, s[0:3], 0 offen offset:4 ; MUBUF-NEXT: s_getpc_b64 s[4:5] ; MUBUF-NEXT: s_add_u32 s4, s4, external_void_func_byval@rel32@lo+4 ; MUBUF-NEXT: s_addc_u32 s5, s5, external_void_func_byval@rel32@hi+12 @@ -373,8 +373,8 @@ ; MUBUF-NEXT: s_waitcnt vmcnt(1) ; MUBUF-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:60 ; MUBUF-NEXT: s_swappc_b64 s[30:31], s[4:5] -; MUBUF-NEXT: v_readlane_b32 s31, v40, 1 ; MUBUF-NEXT: v_readlane_b32 s30, v40, 0 +; MUBUF-NEXT: v_readlane_b32 s31, v40, 1 ; MUBUF-NEXT: s_addk_i32 s32, 0xfc00 ; MUBUF-NEXT: v_readlane_b32 s33, v40, 2 ; MUBUF-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -391,10 +391,10 @@ ; FLATSCR-NEXT: s_mov_b64 exec, s[0:1] ; FLATSCR-NEXT: v_writelane_b32 v40, s33, 2 ; FLATSCR-NEXT: s_mov_b32 s33, s32 -; FLATSCR-NEXT: scratch_load_dwordx2 v[1:2], v0, off -; FLATSCR-NEXT: s_add_i32 s32, s32, 16 ; FLATSCR-NEXT: v_writelane_b32 v40, s30, 0 +; FLATSCR-NEXT: s_add_i32 s32, s32, 16 ; FLATSCR-NEXT: v_writelane_b32 v40, s31, 1 +; FLATSCR-NEXT: scratch_load_dwordx2 v[1:2], v0, off ; FLATSCR-NEXT: s_getpc_b64 s[0:1] ; FLATSCR-NEXT: s_add_u32 s0, s0, external_void_func_byval@rel32@lo+4 ; FLATSCR-NEXT: s_addc_u32 s1, s1, external_void_func_byval@rel32@hi+12 @@ -422,8 +422,8 @@ ; FLATSCR-NEXT: s_waitcnt vmcnt(0) ; FLATSCR-NEXT: scratch_store_dwordx2 off, v[0:1], s32 offset:56 ; FLATSCR-NEXT: s_swappc_b64 s[30:31], s[0:1] -; FLATSCR-NEXT: v_readlane_b32 s31, v40, 1 ; FLATSCR-NEXT: v_readlane_b32 s30, v40, 0 +; FLATSCR-NEXT: v_readlane_b32 s31, v40, 1 ; FLATSCR-NEXT: s_add_i32 s32, s32, -16 ; FLATSCR-NEXT: v_readlane_b32 s33, v40, 2 ; FLATSCR-NEXT: s_or_saveexec_b64 s[0:1], -1 diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll @@ -11,6 +11,7 @@ ; GCN-NEXT: s_mov_b32 s4, s33 ; GCN-NEXT: s_add_i32 s33, s32, 0x3fc0 ; GCN-NEXT: s_and_b32 s33, s33, 0xffffc000 +; GCN-NEXT: s_add_i32 s32, s32, 0x10000 ; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:56 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:52 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:48 ; 4-byte Folded Spill @@ -38,7 +39,6 @@ ; GCN-NEXT: global_load_dwordx4 v[39:42], v[0:1], off offset:144 ; GCN-NEXT: global_load_dwordx4 v[43:46], v[0:1], off offset:160 ; GCN-NEXT: global_load_dwordx4 v[47:50], v[0:1], off offset:176 -; GCN-NEXT: s_add_i32 s32, s32, 0x10000 ; GCN-NEXT: s_add_i32 s32, s32, 0xffff0000 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_store_dword v35, off, s[0:3], s33 offset:640 ; 4-byte Folded Spill @@ -257,6 +257,7 @@ ; GCN-NEXT: s_mov_b32 s4, s33 ; GCN-NEXT: s_add_i32 s33, s32, 0x3fc0 ; GCN-NEXT: s_and_b32 s33, s33, 0xffffc000 +; GCN-NEXT: s_add_i32 s32, s32, 0x10000 ; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:56 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:52 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:48 ; 4-byte Folded Spill @@ -284,7 +285,6 @@ ; GCN-NEXT: global_load_dwordx4 v[39:42], v[0:1], off offset:144 ; GCN-NEXT: global_load_dwordx4 v[43:46], v[0:1], off offset:160 ; GCN-NEXT: global_load_dwordx4 v[47:50], v[0:1], off offset:176 -; GCN-NEXT: s_add_i32 s32, s32, 0x10000 ; GCN-NEXT: s_add_i32 s32, s32, 0xffff0000 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_store_dword v35, off, s[0:3], s33 offset:640 ; 4-byte Folded Spill @@ -507,6 +507,7 @@ ; GCN-NEXT: s_mov_b32 s4, s33 ; GCN-NEXT: s_add_i32 s33, s32, 0x3fc0 ; GCN-NEXT: s_and_b32 s33, s33, 0xffffc000 +; GCN-NEXT: s_add_i32 s32, s32, 0x10000 ; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:44 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:40 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:36 ; 4-byte Folded Spill @@ -532,7 +533,6 @@ ; GCN-NEXT: global_load_dwordx4 v[48:51], v[0:1], off offset:160 ; GCN-NEXT: global_load_dwordx4 v[52:55], v[0:1], off offset:176 ; GCN-NEXT: global_load_dwordx4 v[11:14], v[0:1], off offset:192 -; GCN-NEXT: s_add_i32 s32, s32, 0x10000 ; GCN-NEXT: s_add_i32 s32, s32, 0xffff0000 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_store_dword v11, off, s[0:3], s33 offset:576 ; 4-byte Folded Spill diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/localizer.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/localizer.ll --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/localizer.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/localizer.ll @@ -237,16 +237,16 @@ ; GFX9-NEXT: s_mov_b64 exec, s[16:17] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 0 ; GFX9-NEXT: v_mov_b32_e32 v1, 0 ; GFX9-NEXT: global_load_dword v0, v[0:1], off glc ; GFX9-NEXT: s_waitcnt vmcnt(0) -; GFX9-NEXT: v_writelane_b32 v40, s30, 0 -; GFX9-NEXT: s_addk_i32 s32, 0x400 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_swappc_b64 s[30:31], 0 -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[4:5], -1 diff --git a/llvm/test/CodeGen/AMDGPU/abi-attribute-hints-undefined-behavior.ll b/llvm/test/CodeGen/AMDGPU/abi-attribute-hints-undefined-behavior.ll --- a/llvm/test/CodeGen/AMDGPU/abi-attribute-hints-undefined-behavior.ll +++ b/llvm/test/CodeGen/AMDGPU/abi-attribute-hints-undefined-behavior.ll @@ -22,15 +22,15 @@ ; FIXEDABI-NEXT: s_mov_b64 exec, s[16:17] ; FIXEDABI-NEXT: v_writelane_b32 v40, s33, 2 ; FIXEDABI-NEXT: s_mov_b32 s33, s32 -; FIXEDABI-NEXT: s_addk_i32 s32, 0x400 ; FIXEDABI-NEXT: v_writelane_b32 v40, s30, 0 +; FIXEDABI-NEXT: s_addk_i32 s32, 0x400 ; FIXEDABI-NEXT: v_writelane_b32 v40, s31, 1 ; FIXEDABI-NEXT: s_getpc_b64 s[16:17] ; FIXEDABI-NEXT: s_add_u32 s16, s16, requires_all_inputs@rel32@lo+4 ; FIXEDABI-NEXT: s_addc_u32 s17, s17, requires_all_inputs@rel32@hi+12 ; FIXEDABI-NEXT: s_swappc_b64 s[30:31], s[16:17] -; FIXEDABI-NEXT: v_readlane_b32 s31, v40, 1 ; FIXEDABI-NEXT: v_readlane_b32 s30, v40, 0 +; FIXEDABI-NEXT: v_readlane_b32 s31, v40, 1 ; FIXEDABI-NEXT: s_addk_i32 s32, 0xfc00 ; FIXEDABI-NEXT: v_readlane_b32 s33, v40, 2 ; FIXEDABI-NEXT: s_or_saveexec_b64 s[4:5], -1 diff --git a/llvm/test/CodeGen/AMDGPU/accvgpr-spill-scc-clobber.mir b/llvm/test/CodeGen/AMDGPU/accvgpr-spill-scc-clobber.mir --- a/llvm/test/CodeGen/AMDGPU/accvgpr-spill-scc-clobber.mir +++ b/llvm/test/CodeGen/AMDGPU/accvgpr-spill-scc-clobber.mir @@ -443,229 +443,453 @@ ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr28_lo16 ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr29_lo16 ; GFX90A-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr32, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr32_lo16, $vgpr0_lo16 ; GFX90A-NEXT: $vgpr1 = V_ACCVGPR_READ_B32_e64 killed $agpr33, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr33_lo16, $vgpr1_lo16 ; GFX90A-NEXT: $vgpr2 = V_ACCVGPR_READ_B32_e64 killed $agpr34, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr34_lo16, $vgpr2_lo16 ; GFX90A-NEXT: $vgpr3 = V_ACCVGPR_READ_B32_e64 killed $agpr35, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr35_lo16, $vgpr3_lo16 ; GFX90A-NEXT: $vgpr4 = V_ACCVGPR_READ_B32_e64 killed $agpr36, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr36_lo16, $vgpr4_lo16 ; GFX90A-NEXT: $vgpr5 = V_ACCVGPR_READ_B32_e64 killed $agpr37, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr37_lo16, $vgpr5_lo16 ; GFX90A-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 killed $agpr38, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr38_lo16, $vgpr6_lo16 ; GFX90A-NEXT: $vgpr7 = V_ACCVGPR_READ_B32_e64 killed $agpr39, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr39_lo16, $vgpr7_lo16 ; GFX90A-NEXT: $vgpr8 = V_ACCVGPR_READ_B32_e64 killed $agpr40, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr40_lo16, $vgpr8_lo16 ; GFX90A-NEXT: $vgpr9 = V_ACCVGPR_READ_B32_e64 killed $agpr41, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr41_lo16, $vgpr9_lo16 ; GFX90A-NEXT: $vgpr10 = V_ACCVGPR_READ_B32_e64 killed $agpr42, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr42_lo16, $vgpr10_lo16 ; GFX90A-NEXT: $vgpr11 = V_ACCVGPR_READ_B32_e64 killed $agpr43, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr43_lo16, $vgpr11_lo16 ; GFX90A-NEXT: $vgpr12 = V_ACCVGPR_READ_B32_e64 killed $agpr44, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr44_lo16, $vgpr12_lo16 ; GFX90A-NEXT: $vgpr13 = V_ACCVGPR_READ_B32_e64 killed $agpr45, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr45_lo16, $vgpr13_lo16 ; GFX90A-NEXT: $vgpr14 = V_ACCVGPR_READ_B32_e64 killed $agpr46, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr46_lo16, $vgpr14_lo16 ; GFX90A-NEXT: $vgpr15 = V_ACCVGPR_READ_B32_e64 killed $agpr47, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr47_lo16, $vgpr15_lo16 ; GFX90A-NEXT: $vgpr16 = V_ACCVGPR_READ_B32_e64 killed $agpr48, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr48_lo16, $vgpr16_lo16 ; GFX90A-NEXT: $vgpr17 = V_ACCVGPR_READ_B32_e64 killed $agpr49, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr49_lo16, $vgpr17_lo16 ; GFX90A-NEXT: $vgpr18 = V_ACCVGPR_READ_B32_e64 killed $agpr50, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr50_lo16, $vgpr18_lo16 ; GFX90A-NEXT: $vgpr19 = V_ACCVGPR_READ_B32_e64 killed $agpr51, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr51_lo16, $vgpr19_lo16 ; GFX90A-NEXT: $vgpr20 = V_ACCVGPR_READ_B32_e64 killed $agpr52, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr52_lo16, $vgpr20_lo16 ; GFX90A-NEXT: $vgpr21 = V_ACCVGPR_READ_B32_e64 killed $agpr53, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr53_lo16, $vgpr21_lo16 ; GFX90A-NEXT: $vgpr22 = V_ACCVGPR_READ_B32_e64 killed $agpr54, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr54_lo16, $vgpr22_lo16 ; GFX90A-NEXT: $vgpr23 = V_ACCVGPR_READ_B32_e64 killed $agpr55, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr55_lo16, $vgpr23_lo16 ; GFX90A-NEXT: $vgpr24 = V_ACCVGPR_READ_B32_e64 killed $agpr56, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr56_lo16, $vgpr24_lo16 ; GFX90A-NEXT: $vgpr25 = V_ACCVGPR_READ_B32_e64 killed $agpr57, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr57_lo16, $vgpr25_lo16 ; GFX90A-NEXT: $vgpr26 = V_ACCVGPR_READ_B32_e64 killed $agpr58, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr58_lo16, $vgpr26_lo16 ; GFX90A-NEXT: $vgpr27 = V_ACCVGPR_READ_B32_e64 killed $agpr59, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr59_lo16, $vgpr27_lo16 ; GFX90A-NEXT: $vgpr28 = V_ACCVGPR_READ_B32_e64 killed $agpr60, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr60_lo16, $vgpr28_lo16 ; GFX90A-NEXT: $vgpr29 = V_ACCVGPR_READ_B32_e64 killed $agpr61, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr61_lo16, $vgpr29_lo16 ; GFX90A-NEXT: $vgpr30 = V_ACCVGPR_READ_B32_e64 killed $agpr62, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr62_lo16, $vgpr30_lo16 ; GFX90A-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 killed $agpr63, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr63_lo16, $vgpr31_lo16 ; GFX90A-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr64, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr64_lo16, $vgpr32_lo16 ; GFX90A-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 killed $agpr65, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr65_lo16, $vgpr33_lo16 ; GFX90A-NEXT: $vgpr34 = V_ACCVGPR_READ_B32_e64 killed $agpr66, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr66_lo16, $vgpr34_lo16 ; GFX90A-NEXT: $vgpr35 = V_ACCVGPR_READ_B32_e64 killed $agpr67, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr67_lo16, $vgpr35_lo16 ; GFX90A-NEXT: $vgpr36 = V_ACCVGPR_READ_B32_e64 killed $agpr68, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr68_lo16, $vgpr36_lo16 ; GFX90A-NEXT: $vgpr37 = V_ACCVGPR_READ_B32_e64 killed $agpr69, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr69_lo16, $vgpr37_lo16 ; GFX90A-NEXT: $vgpr38 = V_ACCVGPR_READ_B32_e64 killed $agpr70, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr70_lo16, $vgpr38_lo16 ; GFX90A-NEXT: $vgpr39 = V_ACCVGPR_READ_B32_e64 killed $agpr71, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr71_lo16, $vgpr39_lo16 ; GFX90A-NEXT: $vgpr48 = V_ACCVGPR_READ_B32_e64 killed $agpr72, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr72_lo16, $vgpr48_lo16 ; GFX90A-NEXT: $vgpr49 = V_ACCVGPR_READ_B32_e64 killed $agpr73, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr73_lo16, $vgpr49_lo16 ; GFX90A-NEXT: $vgpr50 = V_ACCVGPR_READ_B32_e64 killed $agpr74, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr74_lo16, $vgpr50_lo16 ; GFX90A-NEXT: $vgpr51 = V_ACCVGPR_READ_B32_e64 killed $agpr75, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr75_lo16, $vgpr51_lo16 ; GFX90A-NEXT: $vgpr52 = V_ACCVGPR_READ_B32_e64 killed $agpr76, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr76_lo16, $vgpr52_lo16 ; GFX90A-NEXT: $vgpr53 = V_ACCVGPR_READ_B32_e64 killed $agpr77, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr77_lo16, $vgpr53_lo16 ; GFX90A-NEXT: $vgpr54 = V_ACCVGPR_READ_B32_e64 killed $agpr78, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr78_lo16, $vgpr54_lo16 ; GFX90A-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 killed $agpr79, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr79_lo16, $vgpr55_lo16 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr80, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 700, 0, 0, 0, implicit $exec :: (store (s32) into %stack.50, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x18, 0x0f, 0x90, 0xd0, 0x18, 0x16, 0xec, 0x80, 0xde, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr81, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 696, 0, 0, 0, implicit $exec :: (store (s32) into %stack.51, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x18, 0x0f, 0x90, 0xd1, 0x18, 0x16, 0xec, 0x80, 0xdc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr82, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 692, 0, 0, 0, implicit $exec :: (store (s32) into %stack.52, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x18, 0x0f, 0x90, 0xd2, 0x18, 0x16, 0xec, 0x80, 0xda, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr83, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 688, 0, 0, 0, implicit $exec :: (store (s32) into %stack.53, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x18, 0x0f, 0x90, 0xd3, 0x18, 0x16, 0xec, 0x80, 0xd8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr84, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 684, 0, 0, 0, implicit $exec :: (store (s32) into %stack.54, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x18, 0x0f, 0x90, 0xd4, 0x18, 0x16, 0xec, 0x80, 0xd6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr85, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 680, 0, 0, 0, implicit $exec :: (store (s32) into %stack.55, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x18, 0x0f, 0x90, 0xd5, 0x18, 0x16, 0xec, 0x80, 0xd4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr86, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 676, 0, 0, 0, implicit $exec :: (store (s32) into %stack.56, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x18, 0x0f, 0x90, 0xd6, 0x18, 0x16, 0xec, 0x80, 0xd2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr87, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 672, 0, 0, 0, implicit $exec :: (store (s32) into %stack.57, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x18, 0x0f, 0x90, 0xd7, 0x18, 0x16, 0xec, 0x80, 0xd0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr88, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 668, 0, 0, 0, implicit $exec :: (store (s32) into %stack.58, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x18, 0x0f, 0x90, 0xd8, 0x18, 0x16, 0xec, 0x80, 0xce, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr89, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 664, 0, 0, 0, implicit $exec :: (store (s32) into %stack.59, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x18, 0x0f, 0x90, 0xd9, 0x18, 0x16, 0xec, 0x80, 0xcc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr90, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 660, 0, 0, 0, implicit $exec :: (store (s32) into %stack.60, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x18, 0x0f, 0x90, 0xda, 0x18, 0x16, 0xec, 0x80, 0xca, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr91, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 656, 0, 0, 0, implicit $exec :: (store (s32) into %stack.61, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x18, 0x0f, 0x90, 0xdb, 0x18, 0x16, 0xec, 0x80, 0xc8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr92, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 652, 0, 0, 0, implicit $exec :: (store (s32) into %stack.62, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x18, 0x0f, 0x90, 0xdc, 0x18, 0x16, 0xec, 0x80, 0xc6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr93, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 648, 0, 0, 0, implicit $exec :: (store (s32) into %stack.63, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x18, 0x0f, 0x90, 0xdd, 0x18, 0x16, 0xec, 0x80, 0xc4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr94, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 644, 0, 0, 0, implicit $exec :: (store (s32) into %stack.64, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x18, 0x0f, 0x90, 0xde, 0x18, 0x16, 0xec, 0x80, 0xc2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr95, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 640, 0, 0, 0, implicit $exec :: (store (s32) into %stack.65, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x18, 0x0f, 0x90, 0xdf, 0x18, 0x16, 0xec, 0x80, 0xc0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr96, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 636, 0, 0, 0, implicit $exec :: (store (s32) into %stack.66, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x18, 0x0f, 0x90, 0xe0, 0x18, 0x16, 0xec, 0x80, 0xbe, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr97, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 632, 0, 0, 0, implicit $exec :: (store (s32) into %stack.67, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x18, 0x0f, 0x90, 0xe1, 0x18, 0x16, 0xec, 0x80, 0xbc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr98, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 628, 0, 0, 0, implicit $exec :: (store (s32) into %stack.68, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x18, 0x0f, 0x90, 0xe2, 0x18, 0x16, 0xec, 0x80, 0xba, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr99, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 624, 0, 0, 0, implicit $exec :: (store (s32) into %stack.69, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x18, 0x0f, 0x90, 0xe3, 0x18, 0x16, 0xec, 0x80, 0xb8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr100, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 620, 0, 0, 0, implicit $exec :: (store (s32) into %stack.70, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x18, 0x0f, 0x90, 0xe4, 0x18, 0x16, 0xec, 0x80, 0xb6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr101, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 616, 0, 0, 0, implicit $exec :: (store (s32) into %stack.71, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x18, 0x0f, 0x90, 0xe5, 0x18, 0x16, 0xec, 0x80, 0xb4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr102, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 612, 0, 0, 0, implicit $exec :: (store (s32) into %stack.72, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x18, 0x0f, 0x90, 0xe6, 0x18, 0x16, 0xec, 0x80, 0xb2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr103, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 608, 0, 0, 0, implicit $exec :: (store (s32) into %stack.73, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x18, 0x0f, 0x90, 0xe7, 0x18, 0x16, 0xec, 0x80, 0xb0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr104, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 604, 0, 0, 0, implicit $exec :: (store (s32) into %stack.74, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x18, 0x0f, 0x90, 0xe8, 0x18, 0x16, 0xec, 0x80, 0xae, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr105, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 600, 0, 0, 0, implicit $exec :: (store (s32) into %stack.75, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x18, 0x0f, 0x90, 0xe9, 0x18, 0x16, 0xec, 0x80, 0xac, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr106, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 596, 0, 0, 0, implicit $exec :: (store (s32) into %stack.76, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x18, 0x0f, 0x90, 0xea, 0x18, 0x16, 0xec, 0x80, 0xaa, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr107, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 592, 0, 0, 0, implicit $exec :: (store (s32) into %stack.77, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x18, 0x0f, 0x90, 0xeb, 0x18, 0x16, 0xec, 0x80, 0xa8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr108, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 588, 0, 0, 0, implicit $exec :: (store (s32) into %stack.78, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x18, 0x0f, 0x90, 0xec, 0x18, 0x16, 0xec, 0x80, 0xa6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr109, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 584, 0, 0, 0, implicit $exec :: (store (s32) into %stack.79, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x18, 0x0f, 0x90, 0xed, 0x18, 0x16, 0xec, 0x80, 0xa4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr110, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 580, 0, 0, 0, implicit $exec :: (store (s32) into %stack.80, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x18, 0x0f, 0x90, 0xee, 0x18, 0x16, 0xec, 0x80, 0xa2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr111, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 576, 0, 0, 0, implicit $exec :: (store (s32) into %stack.81, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x18, 0x0f, 0x90, 0xef, 0x18, 0x16, 0xec, 0x80, 0xa0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr112, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 572, 0, 0, 0, implicit $exec :: (store (s32) into %stack.82, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x18, 0x0f, 0x90, 0xf0, 0x18, 0x16, 0xec, 0x80, 0x9e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr113, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 568, 0, 0, 0, implicit $exec :: (store (s32) into %stack.83, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x18, 0x0f, 0x90, 0xf1, 0x18, 0x16, 0xec, 0x80, 0x9c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr114, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 564, 0, 0, 0, implicit $exec :: (store (s32) into %stack.84, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x18, 0x0f, 0x90, 0xf2, 0x18, 0x16, 0xec, 0x80, 0x9a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr115, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 560, 0, 0, 0, implicit $exec :: (store (s32) into %stack.85, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x18, 0x0f, 0x90, 0xf3, 0x18, 0x16, 0xec, 0x80, 0x98, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr116, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 556, 0, 0, 0, implicit $exec :: (store (s32) into %stack.86, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x18, 0x0f, 0x90, 0xf4, 0x18, 0x16, 0xec, 0x80, 0x96, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr117, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 552, 0, 0, 0, implicit $exec :: (store (s32) into %stack.87, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x18, 0x0f, 0x90, 0xf5, 0x18, 0x16, 0xec, 0x80, 0x94, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr118, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 548, 0, 0, 0, implicit $exec :: (store (s32) into %stack.88, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x18, 0x0f, 0x90, 0xf6, 0x18, 0x16, 0xec, 0x80, 0x92, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr119, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 544, 0, 0, 0, implicit $exec :: (store (s32) into %stack.89, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x18, 0x0f, 0x90, 0xf7, 0x18, 0x16, 0xec, 0x80, 0x90, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr120, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 540, 0, 0, 0, implicit $exec :: (store (s32) into %stack.90, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x18, 0x0f, 0x90, 0xf8, 0x18, 0x16, 0xec, 0x80, 0x8e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr121, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 536, 0, 0, 0, implicit $exec :: (store (s32) into %stack.91, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x18, 0x0f, 0x90, 0xf9, 0x18, 0x16, 0xec, 0x80, 0x8c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr122, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 532, 0, 0, 0, implicit $exec :: (store (s32) into %stack.92, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x18, 0x0f, 0x90, 0xfa, 0x18, 0x16, 0xec, 0x80, 0x8a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr123, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 528, 0, 0, 0, implicit $exec :: (store (s32) into %stack.93, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x18, 0x0f, 0x90, 0xfb, 0x18, 0x16, 0xec, 0x80, 0x88, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr124, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 524, 0, 0, 0, implicit $exec :: (store (s32) into %stack.94, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x18, 0x0f, 0x90, 0xfc, 0x18, 0x16, 0xec, 0x80, 0x86, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr125, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 520, 0, 0, 0, implicit $exec :: (store (s32) into %stack.95, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x18, 0x0f, 0x90, 0xfd, 0x18, 0x16, 0xec, 0x80, 0x84, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr126, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 516, 0, 0, 0, implicit $exec :: (store (s32) into %stack.96, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x18, 0x0f, 0x90, 0xfe, 0x18, 0x16, 0xec, 0x80, 0x82, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr127, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 512, 0, 0, 0, implicit $exec :: (store (s32) into %stack.97, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x18, 0x0f, 0x90, 0xff, 0x18, 0x16, 0xec, 0x80, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr128, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 508, 0, 0, 0, implicit $exec :: (store (s32) into %stack.98, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x80, 0x19, 0x0f, 0x90, 0x80, 0x19, 0x16, 0xec, 0x80, 0xfe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr129, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 504, 0, 0, 0, implicit $exec :: (store (s32) into %stack.99, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x81, 0x19, 0x0f, 0x90, 0x81, 0x19, 0x16, 0xec, 0x80, 0xfc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr130, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 500, 0, 0, 0, implicit $exec :: (store (s32) into %stack.100, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x82, 0x19, 0x0f, 0x90, 0x82, 0x19, 0x16, 0xec, 0x80, 0xfa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr131, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 496, 0, 0, 0, implicit $exec :: (store (s32) into %stack.101, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x83, 0x19, 0x0f, 0x90, 0x83, 0x19, 0x16, 0xec, 0x80, 0xf8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr132, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 492, 0, 0, 0, implicit $exec :: (store (s32) into %stack.102, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x84, 0x19, 0x0f, 0x90, 0x84, 0x19, 0x16, 0xec, 0x80, 0xf6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr133, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 488, 0, 0, 0, implicit $exec :: (store (s32) into %stack.103, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x85, 0x19, 0x0f, 0x90, 0x85, 0x19, 0x16, 0xec, 0x80, 0xf4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr134, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 484, 0, 0, 0, implicit $exec :: (store (s32) into %stack.104, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x86, 0x19, 0x0f, 0x90, 0x86, 0x19, 0x16, 0xec, 0x80, 0xf2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr135, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 480, 0, 0, 0, implicit $exec :: (store (s32) into %stack.105, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x87, 0x19, 0x0f, 0x90, 0x87, 0x19, 0x16, 0xec, 0x80, 0xf0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr136, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 476, 0, 0, 0, implicit $exec :: (store (s32) into %stack.106, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x88, 0x19, 0x0f, 0x90, 0x88, 0x19, 0x16, 0xec, 0x80, 0xee, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr137, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 472, 0, 0, 0, implicit $exec :: (store (s32) into %stack.107, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x89, 0x19, 0x0f, 0x90, 0x89, 0x19, 0x16, 0xec, 0x80, 0xec, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr138, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 468, 0, 0, 0, implicit $exec :: (store (s32) into %stack.108, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8a, 0x19, 0x0f, 0x90, 0x8a, 0x19, 0x16, 0xec, 0x80, 0xea, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr139, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 464, 0, 0, 0, implicit $exec :: (store (s32) into %stack.109, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8b, 0x19, 0x0f, 0x90, 0x8b, 0x19, 0x16, 0xec, 0x80, 0xe8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr140, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 460, 0, 0, 0, implicit $exec :: (store (s32) into %stack.110, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8c, 0x19, 0x0f, 0x90, 0x8c, 0x19, 0x16, 0xec, 0x80, 0xe6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr141, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 456, 0, 0, 0, implicit $exec :: (store (s32) into %stack.111, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8d, 0x19, 0x0f, 0x90, 0x8d, 0x19, 0x16, 0xec, 0x80, 0xe4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr142, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 452, 0, 0, 0, implicit $exec :: (store (s32) into %stack.112, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8e, 0x19, 0x0f, 0x90, 0x8e, 0x19, 0x16, 0xec, 0x80, 0xe2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr143, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 448, 0, 0, 0, implicit $exec :: (store (s32) into %stack.113, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8f, 0x19, 0x0f, 0x90, 0x8f, 0x19, 0x16, 0xec, 0x80, 0xe0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr144, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 444, 0, 0, 0, implicit $exec :: (store (s32) into %stack.114, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x90, 0x19, 0x0f, 0x90, 0x90, 0x19, 0x16, 0xec, 0x80, 0xde, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr145, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 440, 0, 0, 0, implicit $exec :: (store (s32) into %stack.115, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x91, 0x19, 0x0f, 0x90, 0x91, 0x19, 0x16, 0xec, 0x80, 0xdc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr146, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 436, 0, 0, 0, implicit $exec :: (store (s32) into %stack.116, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x92, 0x19, 0x0f, 0x90, 0x92, 0x19, 0x16, 0xec, 0x80, 0xda, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr147, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 432, 0, 0, 0, implicit $exec :: (store (s32) into %stack.117, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x93, 0x19, 0x0f, 0x90, 0x93, 0x19, 0x16, 0xec, 0x80, 0xd8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr148, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 428, 0, 0, 0, implicit $exec :: (store (s32) into %stack.118, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x94, 0x19, 0x0f, 0x90, 0x94, 0x19, 0x16, 0xec, 0x80, 0xd6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr149, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 424, 0, 0, 0, implicit $exec :: (store (s32) into %stack.119, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x95, 0x19, 0x0f, 0x90, 0x95, 0x19, 0x16, 0xec, 0x80, 0xd4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr150, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 420, 0, 0, 0, implicit $exec :: (store (s32) into %stack.120, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x96, 0x19, 0x0f, 0x90, 0x96, 0x19, 0x16, 0xec, 0x80, 0xd2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr151, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 416, 0, 0, 0, implicit $exec :: (store (s32) into %stack.121, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x97, 0x19, 0x0f, 0x90, 0x97, 0x19, 0x16, 0xec, 0x80, 0xd0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr152, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 412, 0, 0, 0, implicit $exec :: (store (s32) into %stack.122, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x98, 0x19, 0x0f, 0x90, 0x98, 0x19, 0x16, 0xec, 0x80, 0xce, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr153, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 408, 0, 0, 0, implicit $exec :: (store (s32) into %stack.123, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x99, 0x19, 0x0f, 0x90, 0x99, 0x19, 0x16, 0xec, 0x80, 0xcc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr154, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 404, 0, 0, 0, implicit $exec :: (store (s32) into %stack.124, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9a, 0x19, 0x0f, 0x90, 0x9a, 0x19, 0x16, 0xec, 0x80, 0xca, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr155, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 400, 0, 0, 0, implicit $exec :: (store (s32) into %stack.125, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9b, 0x19, 0x0f, 0x90, 0x9b, 0x19, 0x16, 0xec, 0x80, 0xc8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr156, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 396, 0, 0, 0, implicit $exec :: (store (s32) into %stack.126, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9c, 0x19, 0x0f, 0x90, 0x9c, 0x19, 0x16, 0xec, 0x80, 0xc6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr157, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 392, 0, 0, 0, implicit $exec :: (store (s32) into %stack.127, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9d, 0x19, 0x0f, 0x90, 0x9d, 0x19, 0x16, 0xec, 0x80, 0xc4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr158, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 388, 0, 0, 0, implicit $exec :: (store (s32) into %stack.128, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9e, 0x19, 0x0f, 0x90, 0x9e, 0x19, 0x16, 0xec, 0x80, 0xc2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr159, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 384, 0, 0, 0, implicit $exec :: (store (s32) into %stack.129, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9f, 0x19, 0x0f, 0x90, 0x9f, 0x19, 0x16, 0xec, 0x80, 0xc0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr160, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 380, 0, 0, 0, implicit $exec :: (store (s32) into %stack.130, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa0, 0x19, 0x0f, 0x90, 0xa0, 0x19, 0x16, 0xec, 0x80, 0xbe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr161, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 376, 0, 0, 0, implicit $exec :: (store (s32) into %stack.131, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa1, 0x19, 0x0f, 0x90, 0xa1, 0x19, 0x16, 0xec, 0x80, 0xbc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr162, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 372, 0, 0, 0, implicit $exec :: (store (s32) into %stack.132, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa2, 0x19, 0x0f, 0x90, 0xa2, 0x19, 0x16, 0xec, 0x80, 0xba, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr163, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 368, 0, 0, 0, implicit $exec :: (store (s32) into %stack.133, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa3, 0x19, 0x0f, 0x90, 0xa3, 0x19, 0x16, 0xec, 0x80, 0xb8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr164, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 364, 0, 0, 0, implicit $exec :: (store (s32) into %stack.134, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa4, 0x19, 0x0f, 0x90, 0xa4, 0x19, 0x16, 0xec, 0x80, 0xb6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr165, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 360, 0, 0, 0, implicit $exec :: (store (s32) into %stack.135, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa5, 0x19, 0x0f, 0x90, 0xa5, 0x19, 0x16, 0xec, 0x80, 0xb4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr166, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 356, 0, 0, 0, implicit $exec :: (store (s32) into %stack.136, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa6, 0x19, 0x0f, 0x90, 0xa6, 0x19, 0x16, 0xec, 0x80, 0xb2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr167, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 352, 0, 0, 0, implicit $exec :: (store (s32) into %stack.137, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa7, 0x19, 0x0f, 0x90, 0xa7, 0x19, 0x16, 0xec, 0x80, 0xb0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr168, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 348, 0, 0, 0, implicit $exec :: (store (s32) into %stack.138, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa8, 0x19, 0x0f, 0x90, 0xa8, 0x19, 0x16, 0xec, 0x80, 0xae, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr169, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 344, 0, 0, 0, implicit $exec :: (store (s32) into %stack.139, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa9, 0x19, 0x0f, 0x90, 0xa9, 0x19, 0x16, 0xec, 0x80, 0xac, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr170, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 340, 0, 0, 0, implicit $exec :: (store (s32) into %stack.140, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaa, 0x19, 0x0f, 0x90, 0xaa, 0x19, 0x16, 0xec, 0x80, 0xaa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr171, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 336, 0, 0, 0, implicit $exec :: (store (s32) into %stack.141, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xab, 0x19, 0x0f, 0x90, 0xab, 0x19, 0x16, 0xec, 0x80, 0xa8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr172, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 332, 0, 0, 0, implicit $exec :: (store (s32) into %stack.142, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xac, 0x19, 0x0f, 0x90, 0xac, 0x19, 0x16, 0xec, 0x80, 0xa6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr173, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 328, 0, 0, 0, implicit $exec :: (store (s32) into %stack.143, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xad, 0x19, 0x0f, 0x90, 0xad, 0x19, 0x16, 0xec, 0x80, 0xa4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr174, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 324, 0, 0, 0, implicit $exec :: (store (s32) into %stack.144, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xae, 0x19, 0x0f, 0x90, 0xae, 0x19, 0x16, 0xec, 0x80, 0xa2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr175, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 320, 0, 0, 0, implicit $exec :: (store (s32) into %stack.145, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaf, 0x19, 0x0f, 0x90, 0xaf, 0x19, 0x16, 0xec, 0x80, 0xa0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr176, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 316, 0, 0, 0, implicit $exec :: (store (s32) into %stack.146, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb0, 0x19, 0x0f, 0x90, 0xb0, 0x19, 0x16, 0xec, 0x80, 0x9e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr177, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 312, 0, 0, 0, implicit $exec :: (store (s32) into %stack.147, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb1, 0x19, 0x0f, 0x90, 0xb1, 0x19, 0x16, 0xec, 0x80, 0x9c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr178, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 308, 0, 0, 0, implicit $exec :: (store (s32) into %stack.148, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb2, 0x19, 0x0f, 0x90, 0xb2, 0x19, 0x16, 0xec, 0x80, 0x9a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr179, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 304, 0, 0, 0, implicit $exec :: (store (s32) into %stack.149, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb3, 0x19, 0x0f, 0x90, 0xb3, 0x19, 0x16, 0xec, 0x80, 0x98, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr180, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 300, 0, 0, 0, implicit $exec :: (store (s32) into %stack.150, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb4, 0x19, 0x0f, 0x90, 0xb4, 0x19, 0x16, 0xec, 0x80, 0x96, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr181, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 296, 0, 0, 0, implicit $exec :: (store (s32) into %stack.151, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb5, 0x19, 0x0f, 0x90, 0xb5, 0x19, 0x16, 0xec, 0x80, 0x94, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr182, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 292, 0, 0, 0, implicit $exec :: (store (s32) into %stack.152, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb6, 0x19, 0x0f, 0x90, 0xb6, 0x19, 0x16, 0xec, 0x80, 0x92, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr183, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 288, 0, 0, 0, implicit $exec :: (store (s32) into %stack.153, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb7, 0x19, 0x0f, 0x90, 0xb7, 0x19, 0x16, 0xec, 0x80, 0x90, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr184, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 284, 0, 0, 0, implicit $exec :: (store (s32) into %stack.154, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb8, 0x19, 0x0f, 0x90, 0xb8, 0x19, 0x16, 0xec, 0x80, 0x8e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr185, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 280, 0, 0, 0, implicit $exec :: (store (s32) into %stack.155, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb9, 0x19, 0x0f, 0x90, 0xb9, 0x19, 0x16, 0xec, 0x80, 0x8c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr186, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 276, 0, 0, 0, implicit $exec :: (store (s32) into %stack.156, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xba, 0x19, 0x0f, 0x90, 0xba, 0x19, 0x16, 0xec, 0x80, 0x8a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr187, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 272, 0, 0, 0, implicit $exec :: (store (s32) into %stack.157, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbb, 0x19, 0x0f, 0x90, 0xbb, 0x19, 0x16, 0xec, 0x80, 0x88, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr188, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 268, 0, 0, 0, implicit $exec :: (store (s32) into %stack.158, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbc, 0x19, 0x0f, 0x90, 0xbc, 0x19, 0x16, 0xec, 0x80, 0x86, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr189, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 264, 0, 0, 0, implicit $exec :: (store (s32) into %stack.159, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbd, 0x19, 0x0f, 0x90, 0xbd, 0x19, 0x16, 0xec, 0x80, 0x84, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr190, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 260, 0, 0, 0, implicit $exec :: (store (s32) into %stack.160, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbe, 0x19, 0x0f, 0x90, 0xbe, 0x19, 0x16, 0xec, 0x80, 0x82, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr191, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 256, 0, 0, 0, implicit $exec :: (store (s32) into %stack.161, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbf, 0x19, 0x0f, 0x90, 0xbf, 0x19, 0x16, 0xec, 0x80, 0x80, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr192, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 252, 0, 0, 0, implicit $exec :: (store (s32) into %stack.162, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc0, 0x19, 0x0e, 0x90, 0xc0, 0x19, 0x16, 0xec, 0x80, 0x7e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr193, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 248, 0, 0, 0, implicit $exec :: (store (s32) into %stack.163, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc1, 0x19, 0x0e, 0x90, 0xc1, 0x19, 0x16, 0xec, 0x80, 0x7c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr194, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 244, 0, 0, 0, implicit $exec :: (store (s32) into %stack.164, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc2, 0x19, 0x0e, 0x90, 0xc2, 0x19, 0x16, 0xec, 0x80, 0x7a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr195, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 240, 0, 0, 0, implicit $exec :: (store (s32) into %stack.165, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc3, 0x19, 0x0e, 0x90, 0xc3, 0x19, 0x16, 0xec, 0x80, 0x78, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr196, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 236, 0, 0, 0, implicit $exec :: (store (s32) into %stack.166, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc4, 0x19, 0x0e, 0x90, 0xc4, 0x19, 0x16, 0xec, 0x80, 0x76, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr197, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 232, 0, 0, 0, implicit $exec :: (store (s32) into %stack.167, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc5, 0x19, 0x0e, 0x90, 0xc5, 0x19, 0x16, 0xec, 0x80, 0x74, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr198, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 228, 0, 0, 0, implicit $exec :: (store (s32) into %stack.168, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc6, 0x19, 0x0e, 0x90, 0xc6, 0x19, 0x16, 0xec, 0x80, 0x72, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr199, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 224, 0, 0, 0, implicit $exec :: (store (s32) into %stack.169, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc7, 0x19, 0x0e, 0x90, 0xc7, 0x19, 0x16, 0xec, 0x80, 0x70, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr200, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 220, 0, 0, 0, implicit $exec :: (store (s32) into %stack.170, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc8, 0x19, 0x0e, 0x90, 0xc8, 0x19, 0x16, 0xec, 0x80, 0x6e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr201, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 216, 0, 0, 0, implicit $exec :: (store (s32) into %stack.171, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc9, 0x19, 0x0e, 0x90, 0xc9, 0x19, 0x16, 0xec, 0x80, 0x6c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr202, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 212, 0, 0, 0, implicit $exec :: (store (s32) into %stack.172, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xca, 0x19, 0x0e, 0x90, 0xca, 0x19, 0x16, 0xec, 0x80, 0x6a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr203, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 208, 0, 0, 0, implicit $exec :: (store (s32) into %stack.173, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcb, 0x19, 0x0e, 0x90, 0xcb, 0x19, 0x16, 0xec, 0x80, 0x68, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr204, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 204, 0, 0, 0, implicit $exec :: (store (s32) into %stack.174, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcc, 0x19, 0x0e, 0x90, 0xcc, 0x19, 0x16, 0xec, 0x80, 0x66, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr205, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 200, 0, 0, 0, implicit $exec :: (store (s32) into %stack.175, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcd, 0x19, 0x0e, 0x90, 0xcd, 0x19, 0x16, 0xec, 0x80, 0x64, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr206, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 196, 0, 0, 0, implicit $exec :: (store (s32) into %stack.176, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xce, 0x19, 0x0e, 0x90, 0xce, 0x19, 0x16, 0xec, 0x80, 0x62, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr207, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 192, 0, 0, 0, implicit $exec :: (store (s32) into %stack.177, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcf, 0x19, 0x0e, 0x90, 0xcf, 0x19, 0x16, 0xec, 0x80, 0x60, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr208, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 188, 0, 0, 0, implicit $exec :: (store (s32) into %stack.178, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x19, 0x0e, 0x90, 0xd0, 0x19, 0x16, 0xec, 0x80, 0x5e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr209, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 184, 0, 0, 0, implicit $exec :: (store (s32) into %stack.179, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x19, 0x0e, 0x90, 0xd1, 0x19, 0x16, 0xec, 0x80, 0x5c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr210, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 180, 0, 0, 0, implicit $exec :: (store (s32) into %stack.180, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x19, 0x0e, 0x90, 0xd2, 0x19, 0x16, 0xec, 0x80, 0x5a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr211, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 176, 0, 0, 0, implicit $exec :: (store (s32) into %stack.181, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x19, 0x0e, 0x90, 0xd3, 0x19, 0x16, 0xec, 0x80, 0x58, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr212, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 172, 0, 0, 0, implicit $exec :: (store (s32) into %stack.182, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x19, 0x0e, 0x90, 0xd4, 0x19, 0x16, 0xec, 0x80, 0x56, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr213, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 168, 0, 0, 0, implicit $exec :: (store (s32) into %stack.183, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x19, 0x0e, 0x90, 0xd5, 0x19, 0x16, 0xec, 0x80, 0x54, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr214, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 164, 0, 0, 0, implicit $exec :: (store (s32) into %stack.184, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x19, 0x0e, 0x90, 0xd6, 0x19, 0x16, 0xec, 0x80, 0x52, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr215, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 160, 0, 0, 0, implicit $exec :: (store (s32) into %stack.185, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x19, 0x0e, 0x90, 0xd7, 0x19, 0x16, 0xec, 0x80, 0x50, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr216, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 156, 0, 0, 0, implicit $exec :: (store (s32) into %stack.186, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x19, 0x0e, 0x90, 0xd8, 0x19, 0x16, 0xec, 0x80, 0x4e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr217, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 152, 0, 0, 0, implicit $exec :: (store (s32) into %stack.187, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x19, 0x0e, 0x90, 0xd9, 0x19, 0x16, 0xec, 0x80, 0x4c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr218, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 148, 0, 0, 0, implicit $exec :: (store (s32) into %stack.188, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x19, 0x0e, 0x90, 0xda, 0x19, 0x16, 0xec, 0x80, 0x4a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr219, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 144, 0, 0, 0, implicit $exec :: (store (s32) into %stack.189, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x19, 0x0e, 0x90, 0xdb, 0x19, 0x16, 0xec, 0x80, 0x48, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr220, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 140, 0, 0, 0, implicit $exec :: (store (s32) into %stack.190, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x19, 0x0e, 0x90, 0xdc, 0x19, 0x16, 0xec, 0x80, 0x46, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr221, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 136, 0, 0, 0, implicit $exec :: (store (s32) into %stack.191, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x19, 0x0e, 0x90, 0xdd, 0x19, 0x16, 0xec, 0x80, 0x44, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr222, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 132, 0, 0, 0, implicit $exec :: (store (s32) into %stack.192, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x19, 0x0e, 0x90, 0xde, 0x19, 0x16, 0xec, 0x80, 0x42, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr223, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 128, 0, 0, 0, implicit $exec :: (store (s32) into %stack.193, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x19, 0x0e, 0x90, 0xdf, 0x19, 0x16, 0xec, 0x80, 0x40, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr224, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 124, 0, 0, 0, implicit $exec :: (store (s32) into %stack.194, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x19, 0x0e, 0x90, 0xe0, 0x19, 0x16, 0xec, 0x80, 0x3e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr225, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 120, 0, 0, 0, implicit $exec :: (store (s32) into %stack.195, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x19, 0x0e, 0x90, 0xe1, 0x19, 0x16, 0xec, 0x80, 0x3c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr226, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 116, 0, 0, 0, implicit $exec :: (store (s32) into %stack.196, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x19, 0x0e, 0x90, 0xe2, 0x19, 0x16, 0xec, 0x80, 0x3a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr227, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 112, 0, 0, 0, implicit $exec :: (store (s32) into %stack.197, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x19, 0x0e, 0x90, 0xe3, 0x19, 0x16, 0xec, 0x80, 0x38, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr228, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 108, 0, 0, 0, implicit $exec :: (store (s32) into %stack.198, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x19, 0x0e, 0x90, 0xe4, 0x19, 0x16, 0xec, 0x80, 0x36, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr229, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 104, 0, 0, 0, implicit $exec :: (store (s32) into %stack.199, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x19, 0x0e, 0x90, 0xe5, 0x19, 0x16, 0xec, 0x80, 0x34, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr230, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 100, 0, 0, 0, implicit $exec :: (store (s32) into %stack.200, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x19, 0x0e, 0x90, 0xe6, 0x19, 0x16, 0xec, 0x80, 0x32, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr231, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 96, 0, 0, 0, implicit $exec :: (store (s32) into %stack.201, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x19, 0x0e, 0x90, 0xe7, 0x19, 0x16, 0xec, 0x80, 0x30, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr232, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 92, 0, 0, 0, implicit $exec :: (store (s32) into %stack.202, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x19, 0x0e, 0x90, 0xe8, 0x19, 0x16, 0xec, 0x80, 0x2e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr233, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 88, 0, 0, 0, implicit $exec :: (store (s32) into %stack.203, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x19, 0x0e, 0x90, 0xe9, 0x19, 0x16, 0xec, 0x80, 0x2c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr234, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 84, 0, 0, 0, implicit $exec :: (store (s32) into %stack.204, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x19, 0x0e, 0x90, 0xea, 0x19, 0x16, 0xec, 0x80, 0x2a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr235, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 80, 0, 0, 0, implicit $exec :: (store (s32) into %stack.205, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x19, 0x0e, 0x90, 0xeb, 0x19, 0x16, 0xec, 0x80, 0x28, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr236, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 76, 0, 0, 0, implicit $exec :: (store (s32) into %stack.206, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x19, 0x0e, 0x90, 0xec, 0x19, 0x16, 0xec, 0x80, 0x26, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr237, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 72, 0, 0, 0, implicit $exec :: (store (s32) into %stack.207, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x19, 0x0e, 0x90, 0xed, 0x19, 0x16, 0xec, 0x80, 0x24, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr238, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 68, 0, 0, 0, implicit $exec :: (store (s32) into %stack.208, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x19, 0x0e, 0x90, 0xee, 0x19, 0x16, 0xec, 0x80, 0x22, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr239, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 64, 0, 0, 0, implicit $exec :: (store (s32) into %stack.209, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x19, 0x0e, 0x90, 0xef, 0x19, 0x16, 0xec, 0x80, 0x20, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr240, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec :: (store (s32) into %stack.210, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x19, 0x0e, 0x90, 0xf0, 0x19, 0x16, 0xec, 0x80, 0x1e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr241, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec :: (store (s32) into %stack.211, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x19, 0x0e, 0x90, 0xf1, 0x19, 0x16, 0xec, 0x80, 0x1c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr242, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec :: (store (s32) into %stack.212, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x19, 0x0e, 0x90, 0xf2, 0x19, 0x16, 0xec, 0x80, 0x1a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr243, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec :: (store (s32) into %stack.213, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x19, 0x0e, 0x90, 0xf3, 0x19, 0x16, 0xec, 0x80, 0x18, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr244, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec :: (store (s32) into %stack.214, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x19, 0x0e, 0x90, 0xf4, 0x19, 0x16, 0xec, 0x80, 0x16, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr245, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec :: (store (s32) into %stack.215, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x19, 0x0e, 0x90, 0xf5, 0x19, 0x16, 0xec, 0x80, 0x14, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr246, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec :: (store (s32) into %stack.216, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x19, 0x0e, 0x90, 0xf6, 0x19, 0x16, 0xec, 0x80, 0x12, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr247, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.217, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x19, 0x0e, 0x90, 0xf7, 0x19, 0x16, 0xec, 0x80, 0x10, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr248, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (store (s32) into %stack.218, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x19, 0x0e, 0x90, 0xf8, 0x19, 0x16, 0xec, 0x80, 0x0e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr249, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (store (s32) into %stack.219, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x19, 0x0e, 0x90, 0xf9, 0x19, 0x16, 0xec, 0x80, 0x0c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr250, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (store (s32) into %stack.220, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x19, 0x0e, 0x90, 0xfa, 0x19, 0x16, 0xec, 0x80, 0x0a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr251, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (store (s32) into %stack.221, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x19, 0x0e, 0x90, 0xfb, 0x19, 0x16, 0xec, 0x80, 0x08, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr252, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (store (s32) into %stack.222, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x19, 0x0e, 0x90, 0xfc, 0x19, 0x16, 0xec, 0x80, 0x06, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr253, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (store (s32) into %stack.223, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x19, 0x0e, 0x90, 0xfd, 0x19, 0x16, 0xec, 0x80, 0x04, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr254, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.224, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x19, 0x0e, 0x90, 0xfe, 0x19, 0x16, 0xec, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr255, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.225, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x19, 0x0d, 0x90, 0xff, 0x19, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: S_CMP_EQ_U32 0, 0, implicit-def $scc ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 704, 0, 0, 0, implicit $exec :: (store (s32) into %stack.226, addrspace 5) ; GFX90A-NEXT: $vgpr0 = V_MOV_B32_e32 8904, implicit $exec @@ -1334,229 +1558,453 @@ ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr28_lo16 ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr29_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr32, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr32_lo16, $vgpr0_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr1 = V_ACCVGPR_READ_B32_e64 killed $agpr33, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr33_lo16, $vgpr1_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr2 = V_ACCVGPR_READ_B32_e64 killed $agpr34, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr34_lo16, $vgpr2_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr3 = V_ACCVGPR_READ_B32_e64 killed $agpr35, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr35_lo16, $vgpr3_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr4 = V_ACCVGPR_READ_B32_e64 killed $agpr36, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr36_lo16, $vgpr4_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr5 = V_ACCVGPR_READ_B32_e64 killed $agpr37, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr37_lo16, $vgpr5_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 killed $agpr38, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr38_lo16, $vgpr6_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr7 = V_ACCVGPR_READ_B32_e64 killed $agpr39, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr39_lo16, $vgpr7_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr8 = V_ACCVGPR_READ_B32_e64 killed $agpr40, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr40_lo16, $vgpr8_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr9 = V_ACCVGPR_READ_B32_e64 killed $agpr41, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr41_lo16, $vgpr9_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr10 = V_ACCVGPR_READ_B32_e64 killed $agpr42, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr42_lo16, $vgpr10_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr11 = V_ACCVGPR_READ_B32_e64 killed $agpr43, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr43_lo16, $vgpr11_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr12 = V_ACCVGPR_READ_B32_e64 killed $agpr44, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr44_lo16, $vgpr12_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr13 = V_ACCVGPR_READ_B32_e64 killed $agpr45, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr45_lo16, $vgpr13_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr14 = V_ACCVGPR_READ_B32_e64 killed $agpr46, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr46_lo16, $vgpr14_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr15 = V_ACCVGPR_READ_B32_e64 killed $agpr47, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr47_lo16, $vgpr15_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr16 = V_ACCVGPR_READ_B32_e64 killed $agpr48, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr48_lo16, $vgpr16_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr17 = V_ACCVGPR_READ_B32_e64 killed $agpr49, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr49_lo16, $vgpr17_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr18 = V_ACCVGPR_READ_B32_e64 killed $agpr50, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr50_lo16, $vgpr18_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr19 = V_ACCVGPR_READ_B32_e64 killed $agpr51, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr51_lo16, $vgpr19_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr20 = V_ACCVGPR_READ_B32_e64 killed $agpr52, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr52_lo16, $vgpr20_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr21 = V_ACCVGPR_READ_B32_e64 killed $agpr53, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr53_lo16, $vgpr21_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr22 = V_ACCVGPR_READ_B32_e64 killed $agpr54, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr54_lo16, $vgpr22_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr23 = V_ACCVGPR_READ_B32_e64 killed $agpr55, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr55_lo16, $vgpr23_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr24 = V_ACCVGPR_READ_B32_e64 killed $agpr56, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr56_lo16, $vgpr24_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr25 = V_ACCVGPR_READ_B32_e64 killed $agpr57, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr57_lo16, $vgpr25_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr26 = V_ACCVGPR_READ_B32_e64 killed $agpr58, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr58_lo16, $vgpr26_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr27 = V_ACCVGPR_READ_B32_e64 killed $agpr59, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr59_lo16, $vgpr27_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr28 = V_ACCVGPR_READ_B32_e64 killed $agpr60, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr60_lo16, $vgpr28_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr29 = V_ACCVGPR_READ_B32_e64 killed $agpr61, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr61_lo16, $vgpr29_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr30 = V_ACCVGPR_READ_B32_e64 killed $agpr62, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr62_lo16, $vgpr30_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 killed $agpr63, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr63_lo16, $vgpr31_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr64, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr64_lo16, $vgpr32_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 killed $agpr65, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr65_lo16, $vgpr33_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr34 = V_ACCVGPR_READ_B32_e64 killed $agpr66, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr66_lo16, $vgpr34_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr35 = V_ACCVGPR_READ_B32_e64 killed $agpr67, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr67_lo16, $vgpr35_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr36 = V_ACCVGPR_READ_B32_e64 killed $agpr68, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr68_lo16, $vgpr36_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr37 = V_ACCVGPR_READ_B32_e64 killed $agpr69, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr69_lo16, $vgpr37_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr38 = V_ACCVGPR_READ_B32_e64 killed $agpr70, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr70_lo16, $vgpr38_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr39 = V_ACCVGPR_READ_B32_e64 killed $agpr71, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr71_lo16, $vgpr39_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr48 = V_ACCVGPR_READ_B32_e64 killed $agpr72, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr72_lo16, $vgpr48_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr49 = V_ACCVGPR_READ_B32_e64 killed $agpr73, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr73_lo16, $vgpr49_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr50 = V_ACCVGPR_READ_B32_e64 killed $agpr74, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr74_lo16, $vgpr50_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr51 = V_ACCVGPR_READ_B32_e64 killed $agpr75, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr75_lo16, $vgpr51_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr52 = V_ACCVGPR_READ_B32_e64 killed $agpr76, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr76_lo16, $vgpr52_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr53 = V_ACCVGPR_READ_B32_e64 killed $agpr77, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr77_lo16, $vgpr53_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr54 = V_ACCVGPR_READ_B32_e64 killed $agpr78, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr78_lo16, $vgpr54_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 killed $agpr79, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr79_lo16, $vgpr55_lo16 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr80, $sgpr32, 700, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.50, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x18, 0x0f, 0x90, 0xd0, 0x18, 0x16, 0xec, 0x80, 0xde, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr81, $sgpr32, 696, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.51, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x18, 0x0f, 0x90, 0xd1, 0x18, 0x16, 0xec, 0x80, 0xdc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr82, $sgpr32, 692, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.52, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x18, 0x0f, 0x90, 0xd2, 0x18, 0x16, 0xec, 0x80, 0xda, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr83, $sgpr32, 688, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.53, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x18, 0x0f, 0x90, 0xd3, 0x18, 0x16, 0xec, 0x80, 0xd8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr84, $sgpr32, 684, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.54, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x18, 0x0f, 0x90, 0xd4, 0x18, 0x16, 0xec, 0x80, 0xd6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr85, $sgpr32, 680, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.55, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x18, 0x0f, 0x90, 0xd5, 0x18, 0x16, 0xec, 0x80, 0xd4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr86, $sgpr32, 676, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.56, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x18, 0x0f, 0x90, 0xd6, 0x18, 0x16, 0xec, 0x80, 0xd2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr87, $sgpr32, 672, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.57, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x18, 0x0f, 0x90, 0xd7, 0x18, 0x16, 0xec, 0x80, 0xd0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr88, $sgpr32, 668, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.58, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x18, 0x0f, 0x90, 0xd8, 0x18, 0x16, 0xec, 0x80, 0xce, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr89, $sgpr32, 664, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.59, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x18, 0x0f, 0x90, 0xd9, 0x18, 0x16, 0xec, 0x80, 0xcc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr90, $sgpr32, 660, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.60, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x18, 0x0f, 0x90, 0xda, 0x18, 0x16, 0xec, 0x80, 0xca, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr91, $sgpr32, 656, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.61, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x18, 0x0f, 0x90, 0xdb, 0x18, 0x16, 0xec, 0x80, 0xc8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr92, $sgpr32, 652, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.62, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x18, 0x0f, 0x90, 0xdc, 0x18, 0x16, 0xec, 0x80, 0xc6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr93, $sgpr32, 648, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.63, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x18, 0x0f, 0x90, 0xdd, 0x18, 0x16, 0xec, 0x80, 0xc4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr94, $sgpr32, 644, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.64, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x18, 0x0f, 0x90, 0xde, 0x18, 0x16, 0xec, 0x80, 0xc2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr95, $sgpr32, 640, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.65, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x18, 0x0f, 0x90, 0xdf, 0x18, 0x16, 0xec, 0x80, 0xc0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr96, $sgpr32, 636, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.66, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x18, 0x0f, 0x90, 0xe0, 0x18, 0x16, 0xec, 0x80, 0xbe, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr97, $sgpr32, 632, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.67, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x18, 0x0f, 0x90, 0xe1, 0x18, 0x16, 0xec, 0x80, 0xbc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr98, $sgpr32, 628, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.68, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x18, 0x0f, 0x90, 0xe2, 0x18, 0x16, 0xec, 0x80, 0xba, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr99, $sgpr32, 624, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.69, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x18, 0x0f, 0x90, 0xe3, 0x18, 0x16, 0xec, 0x80, 0xb8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr100, $sgpr32, 620, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.70, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x18, 0x0f, 0x90, 0xe4, 0x18, 0x16, 0xec, 0x80, 0xb6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr101, $sgpr32, 616, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.71, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x18, 0x0f, 0x90, 0xe5, 0x18, 0x16, 0xec, 0x80, 0xb4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr102, $sgpr32, 612, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.72, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x18, 0x0f, 0x90, 0xe6, 0x18, 0x16, 0xec, 0x80, 0xb2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr103, $sgpr32, 608, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.73, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x18, 0x0f, 0x90, 0xe7, 0x18, 0x16, 0xec, 0x80, 0xb0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr104, $sgpr32, 604, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.74, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x18, 0x0f, 0x90, 0xe8, 0x18, 0x16, 0xec, 0x80, 0xae, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr105, $sgpr32, 600, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.75, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x18, 0x0f, 0x90, 0xe9, 0x18, 0x16, 0xec, 0x80, 0xac, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr106, $sgpr32, 596, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.76, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x18, 0x0f, 0x90, 0xea, 0x18, 0x16, 0xec, 0x80, 0xaa, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr107, $sgpr32, 592, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.77, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x18, 0x0f, 0x90, 0xeb, 0x18, 0x16, 0xec, 0x80, 0xa8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr108, $sgpr32, 588, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.78, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x18, 0x0f, 0x90, 0xec, 0x18, 0x16, 0xec, 0x80, 0xa6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr109, $sgpr32, 584, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.79, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x18, 0x0f, 0x90, 0xed, 0x18, 0x16, 0xec, 0x80, 0xa4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr110, $sgpr32, 580, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.80, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x18, 0x0f, 0x90, 0xee, 0x18, 0x16, 0xec, 0x80, 0xa2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr111, $sgpr32, 576, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.81, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x18, 0x0f, 0x90, 0xef, 0x18, 0x16, 0xec, 0x80, 0xa0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr112, $sgpr32, 572, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.82, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x18, 0x0f, 0x90, 0xf0, 0x18, 0x16, 0xec, 0x80, 0x9e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr113, $sgpr32, 568, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.83, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x18, 0x0f, 0x90, 0xf1, 0x18, 0x16, 0xec, 0x80, 0x9c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr114, $sgpr32, 564, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.84, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x18, 0x0f, 0x90, 0xf2, 0x18, 0x16, 0xec, 0x80, 0x9a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr115, $sgpr32, 560, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.85, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x18, 0x0f, 0x90, 0xf3, 0x18, 0x16, 0xec, 0x80, 0x98, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr116, $sgpr32, 556, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.86, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x18, 0x0f, 0x90, 0xf4, 0x18, 0x16, 0xec, 0x80, 0x96, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr117, $sgpr32, 552, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.87, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x18, 0x0f, 0x90, 0xf5, 0x18, 0x16, 0xec, 0x80, 0x94, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr118, $sgpr32, 548, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.88, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x18, 0x0f, 0x90, 0xf6, 0x18, 0x16, 0xec, 0x80, 0x92, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr119, $sgpr32, 544, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.89, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x18, 0x0f, 0x90, 0xf7, 0x18, 0x16, 0xec, 0x80, 0x90, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr120, $sgpr32, 540, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.90, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x18, 0x0f, 0x90, 0xf8, 0x18, 0x16, 0xec, 0x80, 0x8e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr121, $sgpr32, 536, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.91, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x18, 0x0f, 0x90, 0xf9, 0x18, 0x16, 0xec, 0x80, 0x8c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr122, $sgpr32, 532, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.92, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x18, 0x0f, 0x90, 0xfa, 0x18, 0x16, 0xec, 0x80, 0x8a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr123, $sgpr32, 528, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.93, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x18, 0x0f, 0x90, 0xfb, 0x18, 0x16, 0xec, 0x80, 0x88, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr124, $sgpr32, 524, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.94, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x18, 0x0f, 0x90, 0xfc, 0x18, 0x16, 0xec, 0x80, 0x86, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr125, $sgpr32, 520, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.95, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x18, 0x0f, 0x90, 0xfd, 0x18, 0x16, 0xec, 0x80, 0x84, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr126, $sgpr32, 516, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.96, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x18, 0x0f, 0x90, 0xfe, 0x18, 0x16, 0xec, 0x80, 0x82, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr127, $sgpr32, 512, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.97, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x18, 0x0f, 0x90, 0xff, 0x18, 0x16, 0xec, 0x80, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr128, $sgpr32, 508, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.98, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x80, 0x19, 0x0f, 0x90, 0x80, 0x19, 0x16, 0xec, 0x80, 0xfe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr129, $sgpr32, 504, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.99, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x81, 0x19, 0x0f, 0x90, 0x81, 0x19, 0x16, 0xec, 0x80, 0xfc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr130, $sgpr32, 500, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.100, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x82, 0x19, 0x0f, 0x90, 0x82, 0x19, 0x16, 0xec, 0x80, 0xfa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr131, $sgpr32, 496, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.101, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x83, 0x19, 0x0f, 0x90, 0x83, 0x19, 0x16, 0xec, 0x80, 0xf8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr132, $sgpr32, 492, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.102, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x84, 0x19, 0x0f, 0x90, 0x84, 0x19, 0x16, 0xec, 0x80, 0xf6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr133, $sgpr32, 488, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.103, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x85, 0x19, 0x0f, 0x90, 0x85, 0x19, 0x16, 0xec, 0x80, 0xf4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr134, $sgpr32, 484, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.104, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x86, 0x19, 0x0f, 0x90, 0x86, 0x19, 0x16, 0xec, 0x80, 0xf2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr135, $sgpr32, 480, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.105, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x87, 0x19, 0x0f, 0x90, 0x87, 0x19, 0x16, 0xec, 0x80, 0xf0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr136, $sgpr32, 476, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.106, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x88, 0x19, 0x0f, 0x90, 0x88, 0x19, 0x16, 0xec, 0x80, 0xee, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr137, $sgpr32, 472, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.107, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x89, 0x19, 0x0f, 0x90, 0x89, 0x19, 0x16, 0xec, 0x80, 0xec, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr138, $sgpr32, 468, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.108, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8a, 0x19, 0x0f, 0x90, 0x8a, 0x19, 0x16, 0xec, 0x80, 0xea, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr139, $sgpr32, 464, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.109, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8b, 0x19, 0x0f, 0x90, 0x8b, 0x19, 0x16, 0xec, 0x80, 0xe8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr140, $sgpr32, 460, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.110, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8c, 0x19, 0x0f, 0x90, 0x8c, 0x19, 0x16, 0xec, 0x80, 0xe6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr141, $sgpr32, 456, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.111, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8d, 0x19, 0x0f, 0x90, 0x8d, 0x19, 0x16, 0xec, 0x80, 0xe4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr142, $sgpr32, 452, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.112, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8e, 0x19, 0x0f, 0x90, 0x8e, 0x19, 0x16, 0xec, 0x80, 0xe2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr143, $sgpr32, 448, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.113, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8f, 0x19, 0x0f, 0x90, 0x8f, 0x19, 0x16, 0xec, 0x80, 0xe0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr144, $sgpr32, 444, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.114, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x90, 0x19, 0x0f, 0x90, 0x90, 0x19, 0x16, 0xec, 0x80, 0xde, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr145, $sgpr32, 440, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.115, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x91, 0x19, 0x0f, 0x90, 0x91, 0x19, 0x16, 0xec, 0x80, 0xdc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr146, $sgpr32, 436, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.116, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x92, 0x19, 0x0f, 0x90, 0x92, 0x19, 0x16, 0xec, 0x80, 0xda, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr147, $sgpr32, 432, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.117, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x93, 0x19, 0x0f, 0x90, 0x93, 0x19, 0x16, 0xec, 0x80, 0xd8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr148, $sgpr32, 428, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.118, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x94, 0x19, 0x0f, 0x90, 0x94, 0x19, 0x16, 0xec, 0x80, 0xd6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr149, $sgpr32, 424, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.119, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x95, 0x19, 0x0f, 0x90, 0x95, 0x19, 0x16, 0xec, 0x80, 0xd4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr150, $sgpr32, 420, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.120, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x96, 0x19, 0x0f, 0x90, 0x96, 0x19, 0x16, 0xec, 0x80, 0xd2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr151, $sgpr32, 416, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.121, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x97, 0x19, 0x0f, 0x90, 0x97, 0x19, 0x16, 0xec, 0x80, 0xd0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr152, $sgpr32, 412, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.122, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x98, 0x19, 0x0f, 0x90, 0x98, 0x19, 0x16, 0xec, 0x80, 0xce, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr153, $sgpr32, 408, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.123, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x99, 0x19, 0x0f, 0x90, 0x99, 0x19, 0x16, 0xec, 0x80, 0xcc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr154, $sgpr32, 404, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.124, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9a, 0x19, 0x0f, 0x90, 0x9a, 0x19, 0x16, 0xec, 0x80, 0xca, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr155, $sgpr32, 400, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.125, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9b, 0x19, 0x0f, 0x90, 0x9b, 0x19, 0x16, 0xec, 0x80, 0xc8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr156, $sgpr32, 396, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.126, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9c, 0x19, 0x0f, 0x90, 0x9c, 0x19, 0x16, 0xec, 0x80, 0xc6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr157, $sgpr32, 392, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.127, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9d, 0x19, 0x0f, 0x90, 0x9d, 0x19, 0x16, 0xec, 0x80, 0xc4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr158, $sgpr32, 388, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.128, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9e, 0x19, 0x0f, 0x90, 0x9e, 0x19, 0x16, 0xec, 0x80, 0xc2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr159, $sgpr32, 384, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.129, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9f, 0x19, 0x0f, 0x90, 0x9f, 0x19, 0x16, 0xec, 0x80, 0xc0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr160, $sgpr32, 380, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.130, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa0, 0x19, 0x0f, 0x90, 0xa0, 0x19, 0x16, 0xec, 0x80, 0xbe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr161, $sgpr32, 376, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.131, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa1, 0x19, 0x0f, 0x90, 0xa1, 0x19, 0x16, 0xec, 0x80, 0xbc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr162, $sgpr32, 372, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.132, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa2, 0x19, 0x0f, 0x90, 0xa2, 0x19, 0x16, 0xec, 0x80, 0xba, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr163, $sgpr32, 368, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.133, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa3, 0x19, 0x0f, 0x90, 0xa3, 0x19, 0x16, 0xec, 0x80, 0xb8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr164, $sgpr32, 364, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.134, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa4, 0x19, 0x0f, 0x90, 0xa4, 0x19, 0x16, 0xec, 0x80, 0xb6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr165, $sgpr32, 360, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.135, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa5, 0x19, 0x0f, 0x90, 0xa5, 0x19, 0x16, 0xec, 0x80, 0xb4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr166, $sgpr32, 356, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.136, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa6, 0x19, 0x0f, 0x90, 0xa6, 0x19, 0x16, 0xec, 0x80, 0xb2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr167, $sgpr32, 352, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.137, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa7, 0x19, 0x0f, 0x90, 0xa7, 0x19, 0x16, 0xec, 0x80, 0xb0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr168, $sgpr32, 348, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.138, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa8, 0x19, 0x0f, 0x90, 0xa8, 0x19, 0x16, 0xec, 0x80, 0xae, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr169, $sgpr32, 344, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.139, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa9, 0x19, 0x0f, 0x90, 0xa9, 0x19, 0x16, 0xec, 0x80, 0xac, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr170, $sgpr32, 340, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.140, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaa, 0x19, 0x0f, 0x90, 0xaa, 0x19, 0x16, 0xec, 0x80, 0xaa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr171, $sgpr32, 336, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.141, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xab, 0x19, 0x0f, 0x90, 0xab, 0x19, 0x16, 0xec, 0x80, 0xa8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr172, $sgpr32, 332, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.142, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xac, 0x19, 0x0f, 0x90, 0xac, 0x19, 0x16, 0xec, 0x80, 0xa6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr173, $sgpr32, 328, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.143, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xad, 0x19, 0x0f, 0x90, 0xad, 0x19, 0x16, 0xec, 0x80, 0xa4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr174, $sgpr32, 324, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.144, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xae, 0x19, 0x0f, 0x90, 0xae, 0x19, 0x16, 0xec, 0x80, 0xa2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr175, $sgpr32, 320, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.145, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaf, 0x19, 0x0f, 0x90, 0xaf, 0x19, 0x16, 0xec, 0x80, 0xa0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr176, $sgpr32, 316, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.146, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb0, 0x19, 0x0f, 0x90, 0xb0, 0x19, 0x16, 0xec, 0x80, 0x9e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr177, $sgpr32, 312, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.147, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb1, 0x19, 0x0f, 0x90, 0xb1, 0x19, 0x16, 0xec, 0x80, 0x9c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr178, $sgpr32, 308, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.148, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb2, 0x19, 0x0f, 0x90, 0xb2, 0x19, 0x16, 0xec, 0x80, 0x9a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr179, $sgpr32, 304, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.149, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb3, 0x19, 0x0f, 0x90, 0xb3, 0x19, 0x16, 0xec, 0x80, 0x98, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr180, $sgpr32, 300, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.150, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb4, 0x19, 0x0f, 0x90, 0xb4, 0x19, 0x16, 0xec, 0x80, 0x96, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr181, $sgpr32, 296, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.151, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb5, 0x19, 0x0f, 0x90, 0xb5, 0x19, 0x16, 0xec, 0x80, 0x94, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr182, $sgpr32, 292, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.152, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb6, 0x19, 0x0f, 0x90, 0xb6, 0x19, 0x16, 0xec, 0x80, 0x92, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr183, $sgpr32, 288, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.153, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb7, 0x19, 0x0f, 0x90, 0xb7, 0x19, 0x16, 0xec, 0x80, 0x90, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr184, $sgpr32, 284, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.154, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb8, 0x19, 0x0f, 0x90, 0xb8, 0x19, 0x16, 0xec, 0x80, 0x8e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr185, $sgpr32, 280, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.155, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb9, 0x19, 0x0f, 0x90, 0xb9, 0x19, 0x16, 0xec, 0x80, 0x8c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr186, $sgpr32, 276, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.156, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xba, 0x19, 0x0f, 0x90, 0xba, 0x19, 0x16, 0xec, 0x80, 0x8a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr187, $sgpr32, 272, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.157, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbb, 0x19, 0x0f, 0x90, 0xbb, 0x19, 0x16, 0xec, 0x80, 0x88, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr188, $sgpr32, 268, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.158, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbc, 0x19, 0x0f, 0x90, 0xbc, 0x19, 0x16, 0xec, 0x80, 0x86, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr189, $sgpr32, 264, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.159, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbd, 0x19, 0x0f, 0x90, 0xbd, 0x19, 0x16, 0xec, 0x80, 0x84, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr190, $sgpr32, 260, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.160, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbe, 0x19, 0x0f, 0x90, 0xbe, 0x19, 0x16, 0xec, 0x80, 0x82, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr191, $sgpr32, 256, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.161, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbf, 0x19, 0x0f, 0x90, 0xbf, 0x19, 0x16, 0xec, 0x80, 0x80, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr192, $sgpr32, 252, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.162, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc0, 0x19, 0x0e, 0x90, 0xc0, 0x19, 0x16, 0xec, 0x80, 0x7e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr193, $sgpr32, 248, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.163, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc1, 0x19, 0x0e, 0x90, 0xc1, 0x19, 0x16, 0xec, 0x80, 0x7c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr194, $sgpr32, 244, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.164, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc2, 0x19, 0x0e, 0x90, 0xc2, 0x19, 0x16, 0xec, 0x80, 0x7a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr195, $sgpr32, 240, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.165, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc3, 0x19, 0x0e, 0x90, 0xc3, 0x19, 0x16, 0xec, 0x80, 0x78, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr196, $sgpr32, 236, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.166, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc4, 0x19, 0x0e, 0x90, 0xc4, 0x19, 0x16, 0xec, 0x80, 0x76, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr197, $sgpr32, 232, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.167, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc5, 0x19, 0x0e, 0x90, 0xc5, 0x19, 0x16, 0xec, 0x80, 0x74, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr198, $sgpr32, 228, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.168, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc6, 0x19, 0x0e, 0x90, 0xc6, 0x19, 0x16, 0xec, 0x80, 0x72, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr199, $sgpr32, 224, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.169, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc7, 0x19, 0x0e, 0x90, 0xc7, 0x19, 0x16, 0xec, 0x80, 0x70, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr200, $sgpr32, 220, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.170, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc8, 0x19, 0x0e, 0x90, 0xc8, 0x19, 0x16, 0xec, 0x80, 0x6e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr201, $sgpr32, 216, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.171, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc9, 0x19, 0x0e, 0x90, 0xc9, 0x19, 0x16, 0xec, 0x80, 0x6c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr202, $sgpr32, 212, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.172, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xca, 0x19, 0x0e, 0x90, 0xca, 0x19, 0x16, 0xec, 0x80, 0x6a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr203, $sgpr32, 208, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.173, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcb, 0x19, 0x0e, 0x90, 0xcb, 0x19, 0x16, 0xec, 0x80, 0x68, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr204, $sgpr32, 204, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.174, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcc, 0x19, 0x0e, 0x90, 0xcc, 0x19, 0x16, 0xec, 0x80, 0x66, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr205, $sgpr32, 200, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.175, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcd, 0x19, 0x0e, 0x90, 0xcd, 0x19, 0x16, 0xec, 0x80, 0x64, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr206, $sgpr32, 196, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.176, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xce, 0x19, 0x0e, 0x90, 0xce, 0x19, 0x16, 0xec, 0x80, 0x62, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr207, $sgpr32, 192, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.177, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcf, 0x19, 0x0e, 0x90, 0xcf, 0x19, 0x16, 0xec, 0x80, 0x60, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr208, $sgpr32, 188, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.178, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x19, 0x0e, 0x90, 0xd0, 0x19, 0x16, 0xec, 0x80, 0x5e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr209, $sgpr32, 184, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.179, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x19, 0x0e, 0x90, 0xd1, 0x19, 0x16, 0xec, 0x80, 0x5c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr210, $sgpr32, 180, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.180, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x19, 0x0e, 0x90, 0xd2, 0x19, 0x16, 0xec, 0x80, 0x5a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr211, $sgpr32, 176, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.181, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x19, 0x0e, 0x90, 0xd3, 0x19, 0x16, 0xec, 0x80, 0x58, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr212, $sgpr32, 172, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.182, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x19, 0x0e, 0x90, 0xd4, 0x19, 0x16, 0xec, 0x80, 0x56, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr213, $sgpr32, 168, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.183, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x19, 0x0e, 0x90, 0xd5, 0x19, 0x16, 0xec, 0x80, 0x54, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr214, $sgpr32, 164, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.184, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x19, 0x0e, 0x90, 0xd6, 0x19, 0x16, 0xec, 0x80, 0x52, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr215, $sgpr32, 160, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.185, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x19, 0x0e, 0x90, 0xd7, 0x19, 0x16, 0xec, 0x80, 0x50, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr216, $sgpr32, 156, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.186, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x19, 0x0e, 0x90, 0xd8, 0x19, 0x16, 0xec, 0x80, 0x4e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr217, $sgpr32, 152, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.187, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x19, 0x0e, 0x90, 0xd9, 0x19, 0x16, 0xec, 0x80, 0x4c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr218, $sgpr32, 148, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.188, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x19, 0x0e, 0x90, 0xda, 0x19, 0x16, 0xec, 0x80, 0x4a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr219, $sgpr32, 144, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.189, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x19, 0x0e, 0x90, 0xdb, 0x19, 0x16, 0xec, 0x80, 0x48, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr220, $sgpr32, 140, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.190, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x19, 0x0e, 0x90, 0xdc, 0x19, 0x16, 0xec, 0x80, 0x46, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr221, $sgpr32, 136, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.191, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x19, 0x0e, 0x90, 0xdd, 0x19, 0x16, 0xec, 0x80, 0x44, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr222, $sgpr32, 132, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.192, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x19, 0x0e, 0x90, 0xde, 0x19, 0x16, 0xec, 0x80, 0x42, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr223, $sgpr32, 128, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.193, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x19, 0x0e, 0x90, 0xdf, 0x19, 0x16, 0xec, 0x80, 0x40, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr224, $sgpr32, 124, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.194, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x19, 0x0e, 0x90, 0xe0, 0x19, 0x16, 0xec, 0x80, 0x3e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr225, $sgpr32, 120, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.195, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x19, 0x0e, 0x90, 0xe1, 0x19, 0x16, 0xec, 0x80, 0x3c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr226, $sgpr32, 116, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.196, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x19, 0x0e, 0x90, 0xe2, 0x19, 0x16, 0xec, 0x80, 0x3a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr227, $sgpr32, 112, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.197, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x19, 0x0e, 0x90, 0xe3, 0x19, 0x16, 0xec, 0x80, 0x38, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr228, $sgpr32, 108, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.198, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x19, 0x0e, 0x90, 0xe4, 0x19, 0x16, 0xec, 0x80, 0x36, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr229, $sgpr32, 104, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.199, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x19, 0x0e, 0x90, 0xe5, 0x19, 0x16, 0xec, 0x80, 0x34, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr230, $sgpr32, 100, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.200, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x19, 0x0e, 0x90, 0xe6, 0x19, 0x16, 0xec, 0x80, 0x32, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr231, $sgpr32, 96, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.201, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x19, 0x0e, 0x90, 0xe7, 0x19, 0x16, 0xec, 0x80, 0x30, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr232, $sgpr32, 92, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.202, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x19, 0x0e, 0x90, 0xe8, 0x19, 0x16, 0xec, 0x80, 0x2e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr233, $sgpr32, 88, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.203, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x19, 0x0e, 0x90, 0xe9, 0x19, 0x16, 0xec, 0x80, 0x2c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr234, $sgpr32, 84, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.204, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x19, 0x0e, 0x90, 0xea, 0x19, 0x16, 0xec, 0x80, 0x2a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr235, $sgpr32, 80, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.205, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x19, 0x0e, 0x90, 0xeb, 0x19, 0x16, 0xec, 0x80, 0x28, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr236, $sgpr32, 76, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.206, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x19, 0x0e, 0x90, 0xec, 0x19, 0x16, 0xec, 0x80, 0x26, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr237, $sgpr32, 72, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.207, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x19, 0x0e, 0x90, 0xed, 0x19, 0x16, 0xec, 0x80, 0x24, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr238, $sgpr32, 68, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.208, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x19, 0x0e, 0x90, 0xee, 0x19, 0x16, 0xec, 0x80, 0x22, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr239, $sgpr32, 64, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.209, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x19, 0x0e, 0x90, 0xef, 0x19, 0x16, 0xec, 0x80, 0x20, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr240, $sgpr32, 60, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.210, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x19, 0x0e, 0x90, 0xf0, 0x19, 0x16, 0xec, 0x80, 0x1e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr241, $sgpr32, 56, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.211, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x19, 0x0e, 0x90, 0xf1, 0x19, 0x16, 0xec, 0x80, 0x1c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr242, $sgpr32, 52, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.212, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x19, 0x0e, 0x90, 0xf2, 0x19, 0x16, 0xec, 0x80, 0x1a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr243, $sgpr32, 48, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.213, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x19, 0x0e, 0x90, 0xf3, 0x19, 0x16, 0xec, 0x80, 0x18, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr244, $sgpr32, 44, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.214, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x19, 0x0e, 0x90, 0xf4, 0x19, 0x16, 0xec, 0x80, 0x16, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr245, $sgpr32, 40, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.215, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x19, 0x0e, 0x90, 0xf5, 0x19, 0x16, 0xec, 0x80, 0x14, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr246, $sgpr32, 36, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.216, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x19, 0x0e, 0x90, 0xf6, 0x19, 0x16, 0xec, 0x80, 0x12, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr247, $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.217, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x19, 0x0e, 0x90, 0xf7, 0x19, 0x16, 0xec, 0x80, 0x10, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr248, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.218, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x19, 0x0e, 0x90, 0xf8, 0x19, 0x16, 0xec, 0x80, 0x0e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr249, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.219, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x19, 0x0e, 0x90, 0xf9, 0x19, 0x16, 0xec, 0x80, 0x0c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr250, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.220, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x19, 0x0e, 0x90, 0xfa, 0x19, 0x16, 0xec, 0x80, 0x0a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr251, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.221, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x19, 0x0e, 0x90, 0xfb, 0x19, 0x16, 0xec, 0x80, 0x08, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr252, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.222, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x19, 0x0e, 0x90, 0xfc, 0x19, 0x16, 0xec, 0x80, 0x06, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr253, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.223, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x19, 0x0e, 0x90, 0xfd, 0x19, 0x16, 0xec, 0x80, 0x04, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr254, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.224, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x19, 0x0e, 0x90, 0xfe, 0x19, 0x16, 0xec, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr255, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.225, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x19, 0x0d, 0x90, 0xff, 0x19, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: S_CMP_EQ_U32 0, 0, implicit-def $scc ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 704, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.226, addrspace 5) ; GFX90A-FLATSCR-NEXT: $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec @@ -2255,229 +2703,453 @@ ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr28_lo16 ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr29_lo16 ; GFX90A-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr32, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr32_lo16, $vgpr0_lo16 ; GFX90A-NEXT: $vgpr1 = V_ACCVGPR_READ_B32_e64 killed $agpr33, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr33_lo16, $vgpr1_lo16 ; GFX90A-NEXT: $vgpr2 = V_ACCVGPR_READ_B32_e64 killed $agpr34, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr34_lo16, $vgpr2_lo16 ; GFX90A-NEXT: $vgpr3 = V_ACCVGPR_READ_B32_e64 killed $agpr35, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr35_lo16, $vgpr3_lo16 ; GFX90A-NEXT: $vgpr4 = V_ACCVGPR_READ_B32_e64 killed $agpr36, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr36_lo16, $vgpr4_lo16 ; GFX90A-NEXT: $vgpr5 = V_ACCVGPR_READ_B32_e64 killed $agpr37, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr37_lo16, $vgpr5_lo16 ; GFX90A-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 killed $agpr38, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr38_lo16, $vgpr6_lo16 ; GFX90A-NEXT: $vgpr7 = V_ACCVGPR_READ_B32_e64 killed $agpr39, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr39_lo16, $vgpr7_lo16 ; GFX90A-NEXT: $vgpr8 = V_ACCVGPR_READ_B32_e64 killed $agpr40, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr40_lo16, $vgpr8_lo16 ; GFX90A-NEXT: $vgpr9 = V_ACCVGPR_READ_B32_e64 killed $agpr41, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr41_lo16, $vgpr9_lo16 ; GFX90A-NEXT: $vgpr10 = V_ACCVGPR_READ_B32_e64 killed $agpr42, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr42_lo16, $vgpr10_lo16 ; GFX90A-NEXT: $vgpr11 = V_ACCVGPR_READ_B32_e64 killed $agpr43, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr43_lo16, $vgpr11_lo16 ; GFX90A-NEXT: $vgpr12 = V_ACCVGPR_READ_B32_e64 killed $agpr44, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr44_lo16, $vgpr12_lo16 ; GFX90A-NEXT: $vgpr13 = V_ACCVGPR_READ_B32_e64 killed $agpr45, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr45_lo16, $vgpr13_lo16 ; GFX90A-NEXT: $vgpr14 = V_ACCVGPR_READ_B32_e64 killed $agpr46, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr46_lo16, $vgpr14_lo16 ; GFX90A-NEXT: $vgpr15 = V_ACCVGPR_READ_B32_e64 killed $agpr47, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr47_lo16, $vgpr15_lo16 ; GFX90A-NEXT: $vgpr16 = V_ACCVGPR_READ_B32_e64 killed $agpr48, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr48_lo16, $vgpr16_lo16 ; GFX90A-NEXT: $vgpr17 = V_ACCVGPR_READ_B32_e64 killed $agpr49, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr49_lo16, $vgpr17_lo16 ; GFX90A-NEXT: $vgpr18 = V_ACCVGPR_READ_B32_e64 killed $agpr50, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr50_lo16, $vgpr18_lo16 ; GFX90A-NEXT: $vgpr19 = V_ACCVGPR_READ_B32_e64 killed $agpr51, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr51_lo16, $vgpr19_lo16 ; GFX90A-NEXT: $vgpr20 = V_ACCVGPR_READ_B32_e64 killed $agpr52, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr52_lo16, $vgpr20_lo16 ; GFX90A-NEXT: $vgpr21 = V_ACCVGPR_READ_B32_e64 killed $agpr53, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr53_lo16, $vgpr21_lo16 ; GFX90A-NEXT: $vgpr22 = V_ACCVGPR_READ_B32_e64 killed $agpr54, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr54_lo16, $vgpr22_lo16 ; GFX90A-NEXT: $vgpr23 = V_ACCVGPR_READ_B32_e64 killed $agpr55, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr55_lo16, $vgpr23_lo16 ; GFX90A-NEXT: $vgpr24 = V_ACCVGPR_READ_B32_e64 killed $agpr56, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr56_lo16, $vgpr24_lo16 ; GFX90A-NEXT: $vgpr25 = V_ACCVGPR_READ_B32_e64 killed $agpr57, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr57_lo16, $vgpr25_lo16 ; GFX90A-NEXT: $vgpr26 = V_ACCVGPR_READ_B32_e64 killed $agpr58, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr58_lo16, $vgpr26_lo16 ; GFX90A-NEXT: $vgpr27 = V_ACCVGPR_READ_B32_e64 killed $agpr59, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr59_lo16, $vgpr27_lo16 ; GFX90A-NEXT: $vgpr28 = V_ACCVGPR_READ_B32_e64 killed $agpr60, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr60_lo16, $vgpr28_lo16 ; GFX90A-NEXT: $vgpr29 = V_ACCVGPR_READ_B32_e64 killed $agpr61, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr61_lo16, $vgpr29_lo16 ; GFX90A-NEXT: $vgpr30 = V_ACCVGPR_READ_B32_e64 killed $agpr62, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr62_lo16, $vgpr30_lo16 ; GFX90A-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 killed $agpr63, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr63_lo16, $vgpr31_lo16 ; GFX90A-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr64, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr64_lo16, $vgpr32_lo16 ; GFX90A-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 killed $agpr65, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr65_lo16, $vgpr33_lo16 ; GFX90A-NEXT: $vgpr34 = V_ACCVGPR_READ_B32_e64 killed $agpr66, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr66_lo16, $vgpr34_lo16 ; GFX90A-NEXT: $vgpr35 = V_ACCVGPR_READ_B32_e64 killed $agpr67, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr67_lo16, $vgpr35_lo16 ; GFX90A-NEXT: $vgpr36 = V_ACCVGPR_READ_B32_e64 killed $agpr68, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr68_lo16, $vgpr36_lo16 ; GFX90A-NEXT: $vgpr37 = V_ACCVGPR_READ_B32_e64 killed $agpr69, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr69_lo16, $vgpr37_lo16 ; GFX90A-NEXT: $vgpr38 = V_ACCVGPR_READ_B32_e64 killed $agpr70, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr70_lo16, $vgpr38_lo16 ; GFX90A-NEXT: $vgpr39 = V_ACCVGPR_READ_B32_e64 killed $agpr71, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr71_lo16, $vgpr39_lo16 ; GFX90A-NEXT: $vgpr48 = V_ACCVGPR_READ_B32_e64 killed $agpr72, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr72_lo16, $vgpr48_lo16 ; GFX90A-NEXT: $vgpr49 = V_ACCVGPR_READ_B32_e64 killed $agpr73, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr73_lo16, $vgpr49_lo16 ; GFX90A-NEXT: $vgpr50 = V_ACCVGPR_READ_B32_e64 killed $agpr74, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr74_lo16, $vgpr50_lo16 ; GFX90A-NEXT: $vgpr51 = V_ACCVGPR_READ_B32_e64 killed $agpr75, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr75_lo16, $vgpr51_lo16 ; GFX90A-NEXT: $vgpr52 = V_ACCVGPR_READ_B32_e64 killed $agpr76, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr76_lo16, $vgpr52_lo16 ; GFX90A-NEXT: $vgpr53 = V_ACCVGPR_READ_B32_e64 killed $agpr77, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr77_lo16, $vgpr53_lo16 ; GFX90A-NEXT: $vgpr54 = V_ACCVGPR_READ_B32_e64 killed $agpr78, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr78_lo16, $vgpr54_lo16 ; GFX90A-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 killed $agpr79, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr79_lo16, $vgpr55_lo16 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr80, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 700, 0, 0, 0, implicit $exec :: (store (s32) into %stack.50, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x18, 0x0f, 0x90, 0xd0, 0x18, 0x16, 0xec, 0x80, 0xde, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr81, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 696, 0, 0, 0, implicit $exec :: (store (s32) into %stack.51, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x18, 0x0f, 0x90, 0xd1, 0x18, 0x16, 0xec, 0x80, 0xdc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr82, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 692, 0, 0, 0, implicit $exec :: (store (s32) into %stack.52, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x18, 0x0f, 0x90, 0xd2, 0x18, 0x16, 0xec, 0x80, 0xda, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr83, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 688, 0, 0, 0, implicit $exec :: (store (s32) into %stack.53, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x18, 0x0f, 0x90, 0xd3, 0x18, 0x16, 0xec, 0x80, 0xd8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr84, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 684, 0, 0, 0, implicit $exec :: (store (s32) into %stack.54, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x18, 0x0f, 0x90, 0xd4, 0x18, 0x16, 0xec, 0x80, 0xd6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr85, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 680, 0, 0, 0, implicit $exec :: (store (s32) into %stack.55, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x18, 0x0f, 0x90, 0xd5, 0x18, 0x16, 0xec, 0x80, 0xd4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr86, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 676, 0, 0, 0, implicit $exec :: (store (s32) into %stack.56, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x18, 0x0f, 0x90, 0xd6, 0x18, 0x16, 0xec, 0x80, 0xd2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr87, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 672, 0, 0, 0, implicit $exec :: (store (s32) into %stack.57, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x18, 0x0f, 0x90, 0xd7, 0x18, 0x16, 0xec, 0x80, 0xd0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr88, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 668, 0, 0, 0, implicit $exec :: (store (s32) into %stack.58, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x18, 0x0f, 0x90, 0xd8, 0x18, 0x16, 0xec, 0x80, 0xce, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr89, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 664, 0, 0, 0, implicit $exec :: (store (s32) into %stack.59, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x18, 0x0f, 0x90, 0xd9, 0x18, 0x16, 0xec, 0x80, 0xcc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr90, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 660, 0, 0, 0, implicit $exec :: (store (s32) into %stack.60, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x18, 0x0f, 0x90, 0xda, 0x18, 0x16, 0xec, 0x80, 0xca, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr91, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 656, 0, 0, 0, implicit $exec :: (store (s32) into %stack.61, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x18, 0x0f, 0x90, 0xdb, 0x18, 0x16, 0xec, 0x80, 0xc8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr92, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 652, 0, 0, 0, implicit $exec :: (store (s32) into %stack.62, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x18, 0x0f, 0x90, 0xdc, 0x18, 0x16, 0xec, 0x80, 0xc6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr93, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 648, 0, 0, 0, implicit $exec :: (store (s32) into %stack.63, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x18, 0x0f, 0x90, 0xdd, 0x18, 0x16, 0xec, 0x80, 0xc4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr94, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 644, 0, 0, 0, implicit $exec :: (store (s32) into %stack.64, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x18, 0x0f, 0x90, 0xde, 0x18, 0x16, 0xec, 0x80, 0xc2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr95, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 640, 0, 0, 0, implicit $exec :: (store (s32) into %stack.65, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x18, 0x0f, 0x90, 0xdf, 0x18, 0x16, 0xec, 0x80, 0xc0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr96, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 636, 0, 0, 0, implicit $exec :: (store (s32) into %stack.66, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x18, 0x0f, 0x90, 0xe0, 0x18, 0x16, 0xec, 0x80, 0xbe, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr97, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 632, 0, 0, 0, implicit $exec :: (store (s32) into %stack.67, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x18, 0x0f, 0x90, 0xe1, 0x18, 0x16, 0xec, 0x80, 0xbc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr98, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 628, 0, 0, 0, implicit $exec :: (store (s32) into %stack.68, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x18, 0x0f, 0x90, 0xe2, 0x18, 0x16, 0xec, 0x80, 0xba, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr99, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 624, 0, 0, 0, implicit $exec :: (store (s32) into %stack.69, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x18, 0x0f, 0x90, 0xe3, 0x18, 0x16, 0xec, 0x80, 0xb8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr100, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 620, 0, 0, 0, implicit $exec :: (store (s32) into %stack.70, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x18, 0x0f, 0x90, 0xe4, 0x18, 0x16, 0xec, 0x80, 0xb6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr101, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 616, 0, 0, 0, implicit $exec :: (store (s32) into %stack.71, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x18, 0x0f, 0x90, 0xe5, 0x18, 0x16, 0xec, 0x80, 0xb4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr102, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 612, 0, 0, 0, implicit $exec :: (store (s32) into %stack.72, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x18, 0x0f, 0x90, 0xe6, 0x18, 0x16, 0xec, 0x80, 0xb2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr103, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 608, 0, 0, 0, implicit $exec :: (store (s32) into %stack.73, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x18, 0x0f, 0x90, 0xe7, 0x18, 0x16, 0xec, 0x80, 0xb0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr104, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 604, 0, 0, 0, implicit $exec :: (store (s32) into %stack.74, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x18, 0x0f, 0x90, 0xe8, 0x18, 0x16, 0xec, 0x80, 0xae, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr105, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 600, 0, 0, 0, implicit $exec :: (store (s32) into %stack.75, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x18, 0x0f, 0x90, 0xe9, 0x18, 0x16, 0xec, 0x80, 0xac, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr106, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 596, 0, 0, 0, implicit $exec :: (store (s32) into %stack.76, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x18, 0x0f, 0x90, 0xea, 0x18, 0x16, 0xec, 0x80, 0xaa, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr107, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 592, 0, 0, 0, implicit $exec :: (store (s32) into %stack.77, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x18, 0x0f, 0x90, 0xeb, 0x18, 0x16, 0xec, 0x80, 0xa8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr108, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 588, 0, 0, 0, implicit $exec :: (store (s32) into %stack.78, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x18, 0x0f, 0x90, 0xec, 0x18, 0x16, 0xec, 0x80, 0xa6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr109, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 584, 0, 0, 0, implicit $exec :: (store (s32) into %stack.79, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x18, 0x0f, 0x90, 0xed, 0x18, 0x16, 0xec, 0x80, 0xa4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr110, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 580, 0, 0, 0, implicit $exec :: (store (s32) into %stack.80, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x18, 0x0f, 0x90, 0xee, 0x18, 0x16, 0xec, 0x80, 0xa2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr111, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 576, 0, 0, 0, implicit $exec :: (store (s32) into %stack.81, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x18, 0x0f, 0x90, 0xef, 0x18, 0x16, 0xec, 0x80, 0xa0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr112, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 572, 0, 0, 0, implicit $exec :: (store (s32) into %stack.82, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x18, 0x0f, 0x90, 0xf0, 0x18, 0x16, 0xec, 0x80, 0x9e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr113, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 568, 0, 0, 0, implicit $exec :: (store (s32) into %stack.83, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x18, 0x0f, 0x90, 0xf1, 0x18, 0x16, 0xec, 0x80, 0x9c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr114, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 564, 0, 0, 0, implicit $exec :: (store (s32) into %stack.84, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x18, 0x0f, 0x90, 0xf2, 0x18, 0x16, 0xec, 0x80, 0x9a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr115, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 560, 0, 0, 0, implicit $exec :: (store (s32) into %stack.85, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x18, 0x0f, 0x90, 0xf3, 0x18, 0x16, 0xec, 0x80, 0x98, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr116, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 556, 0, 0, 0, implicit $exec :: (store (s32) into %stack.86, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x18, 0x0f, 0x90, 0xf4, 0x18, 0x16, 0xec, 0x80, 0x96, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr117, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 552, 0, 0, 0, implicit $exec :: (store (s32) into %stack.87, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x18, 0x0f, 0x90, 0xf5, 0x18, 0x16, 0xec, 0x80, 0x94, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr118, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 548, 0, 0, 0, implicit $exec :: (store (s32) into %stack.88, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x18, 0x0f, 0x90, 0xf6, 0x18, 0x16, 0xec, 0x80, 0x92, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr119, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 544, 0, 0, 0, implicit $exec :: (store (s32) into %stack.89, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x18, 0x0f, 0x90, 0xf7, 0x18, 0x16, 0xec, 0x80, 0x90, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr120, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 540, 0, 0, 0, implicit $exec :: (store (s32) into %stack.90, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x18, 0x0f, 0x90, 0xf8, 0x18, 0x16, 0xec, 0x80, 0x8e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr121, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 536, 0, 0, 0, implicit $exec :: (store (s32) into %stack.91, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x18, 0x0f, 0x90, 0xf9, 0x18, 0x16, 0xec, 0x80, 0x8c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr122, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 532, 0, 0, 0, implicit $exec :: (store (s32) into %stack.92, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x18, 0x0f, 0x90, 0xfa, 0x18, 0x16, 0xec, 0x80, 0x8a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr123, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 528, 0, 0, 0, implicit $exec :: (store (s32) into %stack.93, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x18, 0x0f, 0x90, 0xfb, 0x18, 0x16, 0xec, 0x80, 0x88, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr124, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 524, 0, 0, 0, implicit $exec :: (store (s32) into %stack.94, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x18, 0x0f, 0x90, 0xfc, 0x18, 0x16, 0xec, 0x80, 0x86, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr125, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 520, 0, 0, 0, implicit $exec :: (store (s32) into %stack.95, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x18, 0x0f, 0x90, 0xfd, 0x18, 0x16, 0xec, 0x80, 0x84, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr126, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 516, 0, 0, 0, implicit $exec :: (store (s32) into %stack.96, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x18, 0x0f, 0x90, 0xfe, 0x18, 0x16, 0xec, 0x80, 0x82, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr127, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 512, 0, 0, 0, implicit $exec :: (store (s32) into %stack.97, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x18, 0x0f, 0x90, 0xff, 0x18, 0x16, 0xec, 0x80, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr128, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 508, 0, 0, 0, implicit $exec :: (store (s32) into %stack.98, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x80, 0x19, 0x0f, 0x90, 0x80, 0x19, 0x16, 0xec, 0x80, 0xfe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr129, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 504, 0, 0, 0, implicit $exec :: (store (s32) into %stack.99, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x81, 0x19, 0x0f, 0x90, 0x81, 0x19, 0x16, 0xec, 0x80, 0xfc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr130, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 500, 0, 0, 0, implicit $exec :: (store (s32) into %stack.100, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x82, 0x19, 0x0f, 0x90, 0x82, 0x19, 0x16, 0xec, 0x80, 0xfa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr131, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 496, 0, 0, 0, implicit $exec :: (store (s32) into %stack.101, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x83, 0x19, 0x0f, 0x90, 0x83, 0x19, 0x16, 0xec, 0x80, 0xf8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr132, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 492, 0, 0, 0, implicit $exec :: (store (s32) into %stack.102, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x84, 0x19, 0x0f, 0x90, 0x84, 0x19, 0x16, 0xec, 0x80, 0xf6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr133, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 488, 0, 0, 0, implicit $exec :: (store (s32) into %stack.103, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x85, 0x19, 0x0f, 0x90, 0x85, 0x19, 0x16, 0xec, 0x80, 0xf4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr134, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 484, 0, 0, 0, implicit $exec :: (store (s32) into %stack.104, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x86, 0x19, 0x0f, 0x90, 0x86, 0x19, 0x16, 0xec, 0x80, 0xf2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr135, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 480, 0, 0, 0, implicit $exec :: (store (s32) into %stack.105, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x87, 0x19, 0x0f, 0x90, 0x87, 0x19, 0x16, 0xec, 0x80, 0xf0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr136, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 476, 0, 0, 0, implicit $exec :: (store (s32) into %stack.106, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x88, 0x19, 0x0f, 0x90, 0x88, 0x19, 0x16, 0xec, 0x80, 0xee, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr137, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 472, 0, 0, 0, implicit $exec :: (store (s32) into %stack.107, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x89, 0x19, 0x0f, 0x90, 0x89, 0x19, 0x16, 0xec, 0x80, 0xec, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr138, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 468, 0, 0, 0, implicit $exec :: (store (s32) into %stack.108, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8a, 0x19, 0x0f, 0x90, 0x8a, 0x19, 0x16, 0xec, 0x80, 0xea, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr139, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 464, 0, 0, 0, implicit $exec :: (store (s32) into %stack.109, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8b, 0x19, 0x0f, 0x90, 0x8b, 0x19, 0x16, 0xec, 0x80, 0xe8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr140, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 460, 0, 0, 0, implicit $exec :: (store (s32) into %stack.110, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8c, 0x19, 0x0f, 0x90, 0x8c, 0x19, 0x16, 0xec, 0x80, 0xe6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr141, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 456, 0, 0, 0, implicit $exec :: (store (s32) into %stack.111, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8d, 0x19, 0x0f, 0x90, 0x8d, 0x19, 0x16, 0xec, 0x80, 0xe4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr142, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 452, 0, 0, 0, implicit $exec :: (store (s32) into %stack.112, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8e, 0x19, 0x0f, 0x90, 0x8e, 0x19, 0x16, 0xec, 0x80, 0xe2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr143, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 448, 0, 0, 0, implicit $exec :: (store (s32) into %stack.113, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8f, 0x19, 0x0f, 0x90, 0x8f, 0x19, 0x16, 0xec, 0x80, 0xe0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr144, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 444, 0, 0, 0, implicit $exec :: (store (s32) into %stack.114, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x90, 0x19, 0x0f, 0x90, 0x90, 0x19, 0x16, 0xec, 0x80, 0xde, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr145, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 440, 0, 0, 0, implicit $exec :: (store (s32) into %stack.115, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x91, 0x19, 0x0f, 0x90, 0x91, 0x19, 0x16, 0xec, 0x80, 0xdc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr146, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 436, 0, 0, 0, implicit $exec :: (store (s32) into %stack.116, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x92, 0x19, 0x0f, 0x90, 0x92, 0x19, 0x16, 0xec, 0x80, 0xda, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr147, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 432, 0, 0, 0, implicit $exec :: (store (s32) into %stack.117, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x93, 0x19, 0x0f, 0x90, 0x93, 0x19, 0x16, 0xec, 0x80, 0xd8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr148, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 428, 0, 0, 0, implicit $exec :: (store (s32) into %stack.118, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x94, 0x19, 0x0f, 0x90, 0x94, 0x19, 0x16, 0xec, 0x80, 0xd6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr149, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 424, 0, 0, 0, implicit $exec :: (store (s32) into %stack.119, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x95, 0x19, 0x0f, 0x90, 0x95, 0x19, 0x16, 0xec, 0x80, 0xd4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr150, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 420, 0, 0, 0, implicit $exec :: (store (s32) into %stack.120, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x96, 0x19, 0x0f, 0x90, 0x96, 0x19, 0x16, 0xec, 0x80, 0xd2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr151, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 416, 0, 0, 0, implicit $exec :: (store (s32) into %stack.121, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x97, 0x19, 0x0f, 0x90, 0x97, 0x19, 0x16, 0xec, 0x80, 0xd0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr152, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 412, 0, 0, 0, implicit $exec :: (store (s32) into %stack.122, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x98, 0x19, 0x0f, 0x90, 0x98, 0x19, 0x16, 0xec, 0x80, 0xce, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr153, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 408, 0, 0, 0, implicit $exec :: (store (s32) into %stack.123, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x99, 0x19, 0x0f, 0x90, 0x99, 0x19, 0x16, 0xec, 0x80, 0xcc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr154, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 404, 0, 0, 0, implicit $exec :: (store (s32) into %stack.124, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9a, 0x19, 0x0f, 0x90, 0x9a, 0x19, 0x16, 0xec, 0x80, 0xca, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr155, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 400, 0, 0, 0, implicit $exec :: (store (s32) into %stack.125, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9b, 0x19, 0x0f, 0x90, 0x9b, 0x19, 0x16, 0xec, 0x80, 0xc8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr156, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 396, 0, 0, 0, implicit $exec :: (store (s32) into %stack.126, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9c, 0x19, 0x0f, 0x90, 0x9c, 0x19, 0x16, 0xec, 0x80, 0xc6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr157, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 392, 0, 0, 0, implicit $exec :: (store (s32) into %stack.127, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9d, 0x19, 0x0f, 0x90, 0x9d, 0x19, 0x16, 0xec, 0x80, 0xc4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr158, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 388, 0, 0, 0, implicit $exec :: (store (s32) into %stack.128, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9e, 0x19, 0x0f, 0x90, 0x9e, 0x19, 0x16, 0xec, 0x80, 0xc2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr159, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 384, 0, 0, 0, implicit $exec :: (store (s32) into %stack.129, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9f, 0x19, 0x0f, 0x90, 0x9f, 0x19, 0x16, 0xec, 0x80, 0xc0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr160, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 380, 0, 0, 0, implicit $exec :: (store (s32) into %stack.130, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa0, 0x19, 0x0f, 0x90, 0xa0, 0x19, 0x16, 0xec, 0x80, 0xbe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr161, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 376, 0, 0, 0, implicit $exec :: (store (s32) into %stack.131, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa1, 0x19, 0x0f, 0x90, 0xa1, 0x19, 0x16, 0xec, 0x80, 0xbc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr162, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 372, 0, 0, 0, implicit $exec :: (store (s32) into %stack.132, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa2, 0x19, 0x0f, 0x90, 0xa2, 0x19, 0x16, 0xec, 0x80, 0xba, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr163, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 368, 0, 0, 0, implicit $exec :: (store (s32) into %stack.133, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa3, 0x19, 0x0f, 0x90, 0xa3, 0x19, 0x16, 0xec, 0x80, 0xb8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr164, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 364, 0, 0, 0, implicit $exec :: (store (s32) into %stack.134, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa4, 0x19, 0x0f, 0x90, 0xa4, 0x19, 0x16, 0xec, 0x80, 0xb6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr165, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 360, 0, 0, 0, implicit $exec :: (store (s32) into %stack.135, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa5, 0x19, 0x0f, 0x90, 0xa5, 0x19, 0x16, 0xec, 0x80, 0xb4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr166, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 356, 0, 0, 0, implicit $exec :: (store (s32) into %stack.136, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa6, 0x19, 0x0f, 0x90, 0xa6, 0x19, 0x16, 0xec, 0x80, 0xb2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr167, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 352, 0, 0, 0, implicit $exec :: (store (s32) into %stack.137, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa7, 0x19, 0x0f, 0x90, 0xa7, 0x19, 0x16, 0xec, 0x80, 0xb0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr168, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 348, 0, 0, 0, implicit $exec :: (store (s32) into %stack.138, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa8, 0x19, 0x0f, 0x90, 0xa8, 0x19, 0x16, 0xec, 0x80, 0xae, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr169, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 344, 0, 0, 0, implicit $exec :: (store (s32) into %stack.139, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa9, 0x19, 0x0f, 0x90, 0xa9, 0x19, 0x16, 0xec, 0x80, 0xac, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr170, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 340, 0, 0, 0, implicit $exec :: (store (s32) into %stack.140, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaa, 0x19, 0x0f, 0x90, 0xaa, 0x19, 0x16, 0xec, 0x80, 0xaa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr171, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 336, 0, 0, 0, implicit $exec :: (store (s32) into %stack.141, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xab, 0x19, 0x0f, 0x90, 0xab, 0x19, 0x16, 0xec, 0x80, 0xa8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr172, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 332, 0, 0, 0, implicit $exec :: (store (s32) into %stack.142, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xac, 0x19, 0x0f, 0x90, 0xac, 0x19, 0x16, 0xec, 0x80, 0xa6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr173, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 328, 0, 0, 0, implicit $exec :: (store (s32) into %stack.143, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xad, 0x19, 0x0f, 0x90, 0xad, 0x19, 0x16, 0xec, 0x80, 0xa4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr174, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 324, 0, 0, 0, implicit $exec :: (store (s32) into %stack.144, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xae, 0x19, 0x0f, 0x90, 0xae, 0x19, 0x16, 0xec, 0x80, 0xa2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr175, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 320, 0, 0, 0, implicit $exec :: (store (s32) into %stack.145, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaf, 0x19, 0x0f, 0x90, 0xaf, 0x19, 0x16, 0xec, 0x80, 0xa0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr176, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 316, 0, 0, 0, implicit $exec :: (store (s32) into %stack.146, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb0, 0x19, 0x0f, 0x90, 0xb0, 0x19, 0x16, 0xec, 0x80, 0x9e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr177, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 312, 0, 0, 0, implicit $exec :: (store (s32) into %stack.147, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb1, 0x19, 0x0f, 0x90, 0xb1, 0x19, 0x16, 0xec, 0x80, 0x9c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr178, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 308, 0, 0, 0, implicit $exec :: (store (s32) into %stack.148, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb2, 0x19, 0x0f, 0x90, 0xb2, 0x19, 0x16, 0xec, 0x80, 0x9a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr179, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 304, 0, 0, 0, implicit $exec :: (store (s32) into %stack.149, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb3, 0x19, 0x0f, 0x90, 0xb3, 0x19, 0x16, 0xec, 0x80, 0x98, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr180, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 300, 0, 0, 0, implicit $exec :: (store (s32) into %stack.150, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb4, 0x19, 0x0f, 0x90, 0xb4, 0x19, 0x16, 0xec, 0x80, 0x96, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr181, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 296, 0, 0, 0, implicit $exec :: (store (s32) into %stack.151, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb5, 0x19, 0x0f, 0x90, 0xb5, 0x19, 0x16, 0xec, 0x80, 0x94, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr182, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 292, 0, 0, 0, implicit $exec :: (store (s32) into %stack.152, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb6, 0x19, 0x0f, 0x90, 0xb6, 0x19, 0x16, 0xec, 0x80, 0x92, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr183, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 288, 0, 0, 0, implicit $exec :: (store (s32) into %stack.153, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb7, 0x19, 0x0f, 0x90, 0xb7, 0x19, 0x16, 0xec, 0x80, 0x90, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr184, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 284, 0, 0, 0, implicit $exec :: (store (s32) into %stack.154, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb8, 0x19, 0x0f, 0x90, 0xb8, 0x19, 0x16, 0xec, 0x80, 0x8e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr185, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 280, 0, 0, 0, implicit $exec :: (store (s32) into %stack.155, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb9, 0x19, 0x0f, 0x90, 0xb9, 0x19, 0x16, 0xec, 0x80, 0x8c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr186, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 276, 0, 0, 0, implicit $exec :: (store (s32) into %stack.156, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xba, 0x19, 0x0f, 0x90, 0xba, 0x19, 0x16, 0xec, 0x80, 0x8a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr187, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 272, 0, 0, 0, implicit $exec :: (store (s32) into %stack.157, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbb, 0x19, 0x0f, 0x90, 0xbb, 0x19, 0x16, 0xec, 0x80, 0x88, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr188, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 268, 0, 0, 0, implicit $exec :: (store (s32) into %stack.158, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbc, 0x19, 0x0f, 0x90, 0xbc, 0x19, 0x16, 0xec, 0x80, 0x86, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr189, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 264, 0, 0, 0, implicit $exec :: (store (s32) into %stack.159, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbd, 0x19, 0x0f, 0x90, 0xbd, 0x19, 0x16, 0xec, 0x80, 0x84, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr190, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 260, 0, 0, 0, implicit $exec :: (store (s32) into %stack.160, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbe, 0x19, 0x0f, 0x90, 0xbe, 0x19, 0x16, 0xec, 0x80, 0x82, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr191, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 256, 0, 0, 0, implicit $exec :: (store (s32) into %stack.161, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbf, 0x19, 0x0f, 0x90, 0xbf, 0x19, 0x16, 0xec, 0x80, 0x80, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr192, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 252, 0, 0, 0, implicit $exec :: (store (s32) into %stack.162, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc0, 0x19, 0x0e, 0x90, 0xc0, 0x19, 0x16, 0xec, 0x80, 0x7e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr193, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 248, 0, 0, 0, implicit $exec :: (store (s32) into %stack.163, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc1, 0x19, 0x0e, 0x90, 0xc1, 0x19, 0x16, 0xec, 0x80, 0x7c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr194, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 244, 0, 0, 0, implicit $exec :: (store (s32) into %stack.164, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc2, 0x19, 0x0e, 0x90, 0xc2, 0x19, 0x16, 0xec, 0x80, 0x7a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr195, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 240, 0, 0, 0, implicit $exec :: (store (s32) into %stack.165, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc3, 0x19, 0x0e, 0x90, 0xc3, 0x19, 0x16, 0xec, 0x80, 0x78, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr196, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 236, 0, 0, 0, implicit $exec :: (store (s32) into %stack.166, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc4, 0x19, 0x0e, 0x90, 0xc4, 0x19, 0x16, 0xec, 0x80, 0x76, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr197, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 232, 0, 0, 0, implicit $exec :: (store (s32) into %stack.167, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc5, 0x19, 0x0e, 0x90, 0xc5, 0x19, 0x16, 0xec, 0x80, 0x74, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr198, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 228, 0, 0, 0, implicit $exec :: (store (s32) into %stack.168, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc6, 0x19, 0x0e, 0x90, 0xc6, 0x19, 0x16, 0xec, 0x80, 0x72, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr199, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 224, 0, 0, 0, implicit $exec :: (store (s32) into %stack.169, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc7, 0x19, 0x0e, 0x90, 0xc7, 0x19, 0x16, 0xec, 0x80, 0x70, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr200, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 220, 0, 0, 0, implicit $exec :: (store (s32) into %stack.170, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc8, 0x19, 0x0e, 0x90, 0xc8, 0x19, 0x16, 0xec, 0x80, 0x6e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr201, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 216, 0, 0, 0, implicit $exec :: (store (s32) into %stack.171, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc9, 0x19, 0x0e, 0x90, 0xc9, 0x19, 0x16, 0xec, 0x80, 0x6c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr202, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 212, 0, 0, 0, implicit $exec :: (store (s32) into %stack.172, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xca, 0x19, 0x0e, 0x90, 0xca, 0x19, 0x16, 0xec, 0x80, 0x6a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr203, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 208, 0, 0, 0, implicit $exec :: (store (s32) into %stack.173, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcb, 0x19, 0x0e, 0x90, 0xcb, 0x19, 0x16, 0xec, 0x80, 0x68, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr204, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 204, 0, 0, 0, implicit $exec :: (store (s32) into %stack.174, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcc, 0x19, 0x0e, 0x90, 0xcc, 0x19, 0x16, 0xec, 0x80, 0x66, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr205, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 200, 0, 0, 0, implicit $exec :: (store (s32) into %stack.175, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcd, 0x19, 0x0e, 0x90, 0xcd, 0x19, 0x16, 0xec, 0x80, 0x64, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr206, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 196, 0, 0, 0, implicit $exec :: (store (s32) into %stack.176, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xce, 0x19, 0x0e, 0x90, 0xce, 0x19, 0x16, 0xec, 0x80, 0x62, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr207, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 192, 0, 0, 0, implicit $exec :: (store (s32) into %stack.177, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcf, 0x19, 0x0e, 0x90, 0xcf, 0x19, 0x16, 0xec, 0x80, 0x60, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr208, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 188, 0, 0, 0, implicit $exec :: (store (s32) into %stack.178, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x19, 0x0e, 0x90, 0xd0, 0x19, 0x16, 0xec, 0x80, 0x5e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr209, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 184, 0, 0, 0, implicit $exec :: (store (s32) into %stack.179, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x19, 0x0e, 0x90, 0xd1, 0x19, 0x16, 0xec, 0x80, 0x5c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr210, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 180, 0, 0, 0, implicit $exec :: (store (s32) into %stack.180, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x19, 0x0e, 0x90, 0xd2, 0x19, 0x16, 0xec, 0x80, 0x5a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr211, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 176, 0, 0, 0, implicit $exec :: (store (s32) into %stack.181, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x19, 0x0e, 0x90, 0xd3, 0x19, 0x16, 0xec, 0x80, 0x58, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr212, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 172, 0, 0, 0, implicit $exec :: (store (s32) into %stack.182, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x19, 0x0e, 0x90, 0xd4, 0x19, 0x16, 0xec, 0x80, 0x56, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr213, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 168, 0, 0, 0, implicit $exec :: (store (s32) into %stack.183, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x19, 0x0e, 0x90, 0xd5, 0x19, 0x16, 0xec, 0x80, 0x54, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr214, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 164, 0, 0, 0, implicit $exec :: (store (s32) into %stack.184, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x19, 0x0e, 0x90, 0xd6, 0x19, 0x16, 0xec, 0x80, 0x52, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr215, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 160, 0, 0, 0, implicit $exec :: (store (s32) into %stack.185, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x19, 0x0e, 0x90, 0xd7, 0x19, 0x16, 0xec, 0x80, 0x50, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr216, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 156, 0, 0, 0, implicit $exec :: (store (s32) into %stack.186, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x19, 0x0e, 0x90, 0xd8, 0x19, 0x16, 0xec, 0x80, 0x4e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr217, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 152, 0, 0, 0, implicit $exec :: (store (s32) into %stack.187, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x19, 0x0e, 0x90, 0xd9, 0x19, 0x16, 0xec, 0x80, 0x4c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr218, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 148, 0, 0, 0, implicit $exec :: (store (s32) into %stack.188, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x19, 0x0e, 0x90, 0xda, 0x19, 0x16, 0xec, 0x80, 0x4a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr219, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 144, 0, 0, 0, implicit $exec :: (store (s32) into %stack.189, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x19, 0x0e, 0x90, 0xdb, 0x19, 0x16, 0xec, 0x80, 0x48, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr220, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 140, 0, 0, 0, implicit $exec :: (store (s32) into %stack.190, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x19, 0x0e, 0x90, 0xdc, 0x19, 0x16, 0xec, 0x80, 0x46, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr221, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 136, 0, 0, 0, implicit $exec :: (store (s32) into %stack.191, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x19, 0x0e, 0x90, 0xdd, 0x19, 0x16, 0xec, 0x80, 0x44, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr222, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 132, 0, 0, 0, implicit $exec :: (store (s32) into %stack.192, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x19, 0x0e, 0x90, 0xde, 0x19, 0x16, 0xec, 0x80, 0x42, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr223, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 128, 0, 0, 0, implicit $exec :: (store (s32) into %stack.193, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x19, 0x0e, 0x90, 0xdf, 0x19, 0x16, 0xec, 0x80, 0x40, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr224, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 124, 0, 0, 0, implicit $exec :: (store (s32) into %stack.194, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x19, 0x0e, 0x90, 0xe0, 0x19, 0x16, 0xec, 0x80, 0x3e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr225, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 120, 0, 0, 0, implicit $exec :: (store (s32) into %stack.195, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x19, 0x0e, 0x90, 0xe1, 0x19, 0x16, 0xec, 0x80, 0x3c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr226, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 116, 0, 0, 0, implicit $exec :: (store (s32) into %stack.196, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x19, 0x0e, 0x90, 0xe2, 0x19, 0x16, 0xec, 0x80, 0x3a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr227, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 112, 0, 0, 0, implicit $exec :: (store (s32) into %stack.197, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x19, 0x0e, 0x90, 0xe3, 0x19, 0x16, 0xec, 0x80, 0x38, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr228, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 108, 0, 0, 0, implicit $exec :: (store (s32) into %stack.198, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x19, 0x0e, 0x90, 0xe4, 0x19, 0x16, 0xec, 0x80, 0x36, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr229, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 104, 0, 0, 0, implicit $exec :: (store (s32) into %stack.199, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x19, 0x0e, 0x90, 0xe5, 0x19, 0x16, 0xec, 0x80, 0x34, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr230, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 100, 0, 0, 0, implicit $exec :: (store (s32) into %stack.200, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x19, 0x0e, 0x90, 0xe6, 0x19, 0x16, 0xec, 0x80, 0x32, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr231, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 96, 0, 0, 0, implicit $exec :: (store (s32) into %stack.201, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x19, 0x0e, 0x90, 0xe7, 0x19, 0x16, 0xec, 0x80, 0x30, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr232, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 92, 0, 0, 0, implicit $exec :: (store (s32) into %stack.202, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x19, 0x0e, 0x90, 0xe8, 0x19, 0x16, 0xec, 0x80, 0x2e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr233, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 88, 0, 0, 0, implicit $exec :: (store (s32) into %stack.203, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x19, 0x0e, 0x90, 0xe9, 0x19, 0x16, 0xec, 0x80, 0x2c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr234, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 84, 0, 0, 0, implicit $exec :: (store (s32) into %stack.204, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x19, 0x0e, 0x90, 0xea, 0x19, 0x16, 0xec, 0x80, 0x2a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr235, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 80, 0, 0, 0, implicit $exec :: (store (s32) into %stack.205, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x19, 0x0e, 0x90, 0xeb, 0x19, 0x16, 0xec, 0x80, 0x28, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr236, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 76, 0, 0, 0, implicit $exec :: (store (s32) into %stack.206, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x19, 0x0e, 0x90, 0xec, 0x19, 0x16, 0xec, 0x80, 0x26, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr237, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 72, 0, 0, 0, implicit $exec :: (store (s32) into %stack.207, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x19, 0x0e, 0x90, 0xed, 0x19, 0x16, 0xec, 0x80, 0x24, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr238, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 68, 0, 0, 0, implicit $exec :: (store (s32) into %stack.208, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x19, 0x0e, 0x90, 0xee, 0x19, 0x16, 0xec, 0x80, 0x22, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr239, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 64, 0, 0, 0, implicit $exec :: (store (s32) into %stack.209, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x19, 0x0e, 0x90, 0xef, 0x19, 0x16, 0xec, 0x80, 0x20, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr240, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec :: (store (s32) into %stack.210, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x19, 0x0e, 0x90, 0xf0, 0x19, 0x16, 0xec, 0x80, 0x1e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr241, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec :: (store (s32) into %stack.211, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x19, 0x0e, 0x90, 0xf1, 0x19, 0x16, 0xec, 0x80, 0x1c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr242, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec :: (store (s32) into %stack.212, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x19, 0x0e, 0x90, 0xf2, 0x19, 0x16, 0xec, 0x80, 0x1a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr243, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec :: (store (s32) into %stack.213, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x19, 0x0e, 0x90, 0xf3, 0x19, 0x16, 0xec, 0x80, 0x18, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr244, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec :: (store (s32) into %stack.214, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x19, 0x0e, 0x90, 0xf4, 0x19, 0x16, 0xec, 0x80, 0x16, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr245, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec :: (store (s32) into %stack.215, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x19, 0x0e, 0x90, 0xf5, 0x19, 0x16, 0xec, 0x80, 0x14, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr246, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec :: (store (s32) into %stack.216, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x19, 0x0e, 0x90, 0xf6, 0x19, 0x16, 0xec, 0x80, 0x12, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr247, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.217, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x19, 0x0e, 0x90, 0xf7, 0x19, 0x16, 0xec, 0x80, 0x10, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr248, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (store (s32) into %stack.218, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x19, 0x0e, 0x90, 0xf8, 0x19, 0x16, 0xec, 0x80, 0x0e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr249, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (store (s32) into %stack.219, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x19, 0x0e, 0x90, 0xf9, 0x19, 0x16, 0xec, 0x80, 0x0c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr250, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (store (s32) into %stack.220, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x19, 0x0e, 0x90, 0xfa, 0x19, 0x16, 0xec, 0x80, 0x0a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr251, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (store (s32) into %stack.221, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x19, 0x0e, 0x90, 0xfb, 0x19, 0x16, 0xec, 0x80, 0x08, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr252, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (store (s32) into %stack.222, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x19, 0x0e, 0x90, 0xfc, 0x19, 0x16, 0xec, 0x80, 0x06, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr253, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (store (s32) into %stack.223, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x19, 0x0e, 0x90, 0xfd, 0x19, 0x16, 0xec, 0x80, 0x04, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr254, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.224, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x19, 0x0e, 0x90, 0xfe, 0x19, 0x16, 0xec, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr255, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.225, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x19, 0x0d, 0x90, 0xff, 0x19, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: S_CMP_EQ_U32 0, 0, implicit-def $scc ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 704, 0, 0, 0, implicit $exec :: (store (s32) into %stack.226, addrspace 5) ; GFX90A-NEXT: $vgpr0 = V_MOV_B32_e32 8904, implicit $exec @@ -3149,229 +3821,453 @@ ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr28_lo16 ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr29_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr32, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr32_lo16, $vgpr0_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr1 = V_ACCVGPR_READ_B32_e64 killed $agpr33, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr33_lo16, $vgpr1_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr2 = V_ACCVGPR_READ_B32_e64 killed $agpr34, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr34_lo16, $vgpr2_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr3 = V_ACCVGPR_READ_B32_e64 killed $agpr35, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr35_lo16, $vgpr3_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr4 = V_ACCVGPR_READ_B32_e64 killed $agpr36, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr36_lo16, $vgpr4_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr5 = V_ACCVGPR_READ_B32_e64 killed $agpr37, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr37_lo16, $vgpr5_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 killed $agpr38, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr38_lo16, $vgpr6_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr7 = V_ACCVGPR_READ_B32_e64 killed $agpr39, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr39_lo16, $vgpr7_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr8 = V_ACCVGPR_READ_B32_e64 killed $agpr40, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr40_lo16, $vgpr8_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr9 = V_ACCVGPR_READ_B32_e64 killed $agpr41, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr41_lo16, $vgpr9_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr10 = V_ACCVGPR_READ_B32_e64 killed $agpr42, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr42_lo16, $vgpr10_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr11 = V_ACCVGPR_READ_B32_e64 killed $agpr43, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr43_lo16, $vgpr11_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr12 = V_ACCVGPR_READ_B32_e64 killed $agpr44, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr44_lo16, $vgpr12_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr13 = V_ACCVGPR_READ_B32_e64 killed $agpr45, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr45_lo16, $vgpr13_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr14 = V_ACCVGPR_READ_B32_e64 killed $agpr46, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr46_lo16, $vgpr14_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr15 = V_ACCVGPR_READ_B32_e64 killed $agpr47, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr47_lo16, $vgpr15_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr16 = V_ACCVGPR_READ_B32_e64 killed $agpr48, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr48_lo16, $vgpr16_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr17 = V_ACCVGPR_READ_B32_e64 killed $agpr49, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr49_lo16, $vgpr17_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr18 = V_ACCVGPR_READ_B32_e64 killed $agpr50, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr50_lo16, $vgpr18_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr19 = V_ACCVGPR_READ_B32_e64 killed $agpr51, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr51_lo16, $vgpr19_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr20 = V_ACCVGPR_READ_B32_e64 killed $agpr52, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr52_lo16, $vgpr20_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr21 = V_ACCVGPR_READ_B32_e64 killed $agpr53, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr53_lo16, $vgpr21_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr22 = V_ACCVGPR_READ_B32_e64 killed $agpr54, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr54_lo16, $vgpr22_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr23 = V_ACCVGPR_READ_B32_e64 killed $agpr55, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr55_lo16, $vgpr23_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr24 = V_ACCVGPR_READ_B32_e64 killed $agpr56, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr56_lo16, $vgpr24_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr25 = V_ACCVGPR_READ_B32_e64 killed $agpr57, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr57_lo16, $vgpr25_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr26 = V_ACCVGPR_READ_B32_e64 killed $agpr58, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr58_lo16, $vgpr26_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr27 = V_ACCVGPR_READ_B32_e64 killed $agpr59, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr59_lo16, $vgpr27_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr28 = V_ACCVGPR_READ_B32_e64 killed $agpr60, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr60_lo16, $vgpr28_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr29 = V_ACCVGPR_READ_B32_e64 killed $agpr61, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr61_lo16, $vgpr29_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr30 = V_ACCVGPR_READ_B32_e64 killed $agpr62, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr62_lo16, $vgpr30_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 killed $agpr63, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr63_lo16, $vgpr31_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr64, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr64_lo16, $vgpr32_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 killed $agpr65, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr65_lo16, $vgpr33_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr34 = V_ACCVGPR_READ_B32_e64 killed $agpr66, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr66_lo16, $vgpr34_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr35 = V_ACCVGPR_READ_B32_e64 killed $agpr67, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr67_lo16, $vgpr35_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr36 = V_ACCVGPR_READ_B32_e64 killed $agpr68, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr68_lo16, $vgpr36_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr37 = V_ACCVGPR_READ_B32_e64 killed $agpr69, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr69_lo16, $vgpr37_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr38 = V_ACCVGPR_READ_B32_e64 killed $agpr70, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr70_lo16, $vgpr38_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr39 = V_ACCVGPR_READ_B32_e64 killed $agpr71, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr71_lo16, $vgpr39_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr48 = V_ACCVGPR_READ_B32_e64 killed $agpr72, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr72_lo16, $vgpr48_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr49 = V_ACCVGPR_READ_B32_e64 killed $agpr73, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr73_lo16, $vgpr49_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr50 = V_ACCVGPR_READ_B32_e64 killed $agpr74, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr74_lo16, $vgpr50_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr51 = V_ACCVGPR_READ_B32_e64 killed $agpr75, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr75_lo16, $vgpr51_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr52 = V_ACCVGPR_READ_B32_e64 killed $agpr76, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr76_lo16, $vgpr52_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr53 = V_ACCVGPR_READ_B32_e64 killed $agpr77, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr77_lo16, $vgpr53_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr54 = V_ACCVGPR_READ_B32_e64 killed $agpr78, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr78_lo16, $vgpr54_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 killed $agpr79, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr79_lo16, $vgpr55_lo16 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr80, $sgpr32, 700, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.50, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x18, 0x0f, 0x90, 0xd0, 0x18, 0x16, 0xec, 0x80, 0xde, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr81, $sgpr32, 696, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.51, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x18, 0x0f, 0x90, 0xd1, 0x18, 0x16, 0xec, 0x80, 0xdc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr82, $sgpr32, 692, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.52, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x18, 0x0f, 0x90, 0xd2, 0x18, 0x16, 0xec, 0x80, 0xda, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr83, $sgpr32, 688, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.53, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x18, 0x0f, 0x90, 0xd3, 0x18, 0x16, 0xec, 0x80, 0xd8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr84, $sgpr32, 684, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.54, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x18, 0x0f, 0x90, 0xd4, 0x18, 0x16, 0xec, 0x80, 0xd6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr85, $sgpr32, 680, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.55, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x18, 0x0f, 0x90, 0xd5, 0x18, 0x16, 0xec, 0x80, 0xd4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr86, $sgpr32, 676, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.56, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x18, 0x0f, 0x90, 0xd6, 0x18, 0x16, 0xec, 0x80, 0xd2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr87, $sgpr32, 672, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.57, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x18, 0x0f, 0x90, 0xd7, 0x18, 0x16, 0xec, 0x80, 0xd0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr88, $sgpr32, 668, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.58, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x18, 0x0f, 0x90, 0xd8, 0x18, 0x16, 0xec, 0x80, 0xce, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr89, $sgpr32, 664, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.59, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x18, 0x0f, 0x90, 0xd9, 0x18, 0x16, 0xec, 0x80, 0xcc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr90, $sgpr32, 660, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.60, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x18, 0x0f, 0x90, 0xda, 0x18, 0x16, 0xec, 0x80, 0xca, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr91, $sgpr32, 656, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.61, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x18, 0x0f, 0x90, 0xdb, 0x18, 0x16, 0xec, 0x80, 0xc8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr92, $sgpr32, 652, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.62, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x18, 0x0f, 0x90, 0xdc, 0x18, 0x16, 0xec, 0x80, 0xc6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr93, $sgpr32, 648, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.63, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x18, 0x0f, 0x90, 0xdd, 0x18, 0x16, 0xec, 0x80, 0xc4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr94, $sgpr32, 644, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.64, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x18, 0x0f, 0x90, 0xde, 0x18, 0x16, 0xec, 0x80, 0xc2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr95, $sgpr32, 640, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.65, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x18, 0x0f, 0x90, 0xdf, 0x18, 0x16, 0xec, 0x80, 0xc0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr96, $sgpr32, 636, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.66, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x18, 0x0f, 0x90, 0xe0, 0x18, 0x16, 0xec, 0x80, 0xbe, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr97, $sgpr32, 632, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.67, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x18, 0x0f, 0x90, 0xe1, 0x18, 0x16, 0xec, 0x80, 0xbc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr98, $sgpr32, 628, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.68, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x18, 0x0f, 0x90, 0xe2, 0x18, 0x16, 0xec, 0x80, 0xba, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr99, $sgpr32, 624, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.69, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x18, 0x0f, 0x90, 0xe3, 0x18, 0x16, 0xec, 0x80, 0xb8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr100, $sgpr32, 620, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.70, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x18, 0x0f, 0x90, 0xe4, 0x18, 0x16, 0xec, 0x80, 0xb6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr101, $sgpr32, 616, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.71, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x18, 0x0f, 0x90, 0xe5, 0x18, 0x16, 0xec, 0x80, 0xb4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr102, $sgpr32, 612, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.72, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x18, 0x0f, 0x90, 0xe6, 0x18, 0x16, 0xec, 0x80, 0xb2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr103, $sgpr32, 608, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.73, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x18, 0x0f, 0x90, 0xe7, 0x18, 0x16, 0xec, 0x80, 0xb0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr104, $sgpr32, 604, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.74, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x18, 0x0f, 0x90, 0xe8, 0x18, 0x16, 0xec, 0x80, 0xae, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr105, $sgpr32, 600, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.75, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x18, 0x0f, 0x90, 0xe9, 0x18, 0x16, 0xec, 0x80, 0xac, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr106, $sgpr32, 596, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.76, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x18, 0x0f, 0x90, 0xea, 0x18, 0x16, 0xec, 0x80, 0xaa, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr107, $sgpr32, 592, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.77, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x18, 0x0f, 0x90, 0xeb, 0x18, 0x16, 0xec, 0x80, 0xa8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr108, $sgpr32, 588, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.78, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x18, 0x0f, 0x90, 0xec, 0x18, 0x16, 0xec, 0x80, 0xa6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr109, $sgpr32, 584, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.79, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x18, 0x0f, 0x90, 0xed, 0x18, 0x16, 0xec, 0x80, 0xa4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr110, $sgpr32, 580, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.80, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x18, 0x0f, 0x90, 0xee, 0x18, 0x16, 0xec, 0x80, 0xa2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr111, $sgpr32, 576, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.81, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x18, 0x0f, 0x90, 0xef, 0x18, 0x16, 0xec, 0x80, 0xa0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr112, $sgpr32, 572, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.82, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x18, 0x0f, 0x90, 0xf0, 0x18, 0x16, 0xec, 0x80, 0x9e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr113, $sgpr32, 568, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.83, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x18, 0x0f, 0x90, 0xf1, 0x18, 0x16, 0xec, 0x80, 0x9c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr114, $sgpr32, 564, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.84, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x18, 0x0f, 0x90, 0xf2, 0x18, 0x16, 0xec, 0x80, 0x9a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr115, $sgpr32, 560, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.85, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x18, 0x0f, 0x90, 0xf3, 0x18, 0x16, 0xec, 0x80, 0x98, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr116, $sgpr32, 556, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.86, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x18, 0x0f, 0x90, 0xf4, 0x18, 0x16, 0xec, 0x80, 0x96, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr117, $sgpr32, 552, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.87, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x18, 0x0f, 0x90, 0xf5, 0x18, 0x16, 0xec, 0x80, 0x94, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr118, $sgpr32, 548, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.88, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x18, 0x0f, 0x90, 0xf6, 0x18, 0x16, 0xec, 0x80, 0x92, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr119, $sgpr32, 544, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.89, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x18, 0x0f, 0x90, 0xf7, 0x18, 0x16, 0xec, 0x80, 0x90, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr120, $sgpr32, 540, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.90, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x18, 0x0f, 0x90, 0xf8, 0x18, 0x16, 0xec, 0x80, 0x8e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr121, $sgpr32, 536, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.91, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x18, 0x0f, 0x90, 0xf9, 0x18, 0x16, 0xec, 0x80, 0x8c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr122, $sgpr32, 532, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.92, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x18, 0x0f, 0x90, 0xfa, 0x18, 0x16, 0xec, 0x80, 0x8a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr123, $sgpr32, 528, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.93, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x18, 0x0f, 0x90, 0xfb, 0x18, 0x16, 0xec, 0x80, 0x88, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr124, $sgpr32, 524, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.94, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x18, 0x0f, 0x90, 0xfc, 0x18, 0x16, 0xec, 0x80, 0x86, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr125, $sgpr32, 520, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.95, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x18, 0x0f, 0x90, 0xfd, 0x18, 0x16, 0xec, 0x80, 0x84, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr126, $sgpr32, 516, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.96, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x18, 0x0f, 0x90, 0xfe, 0x18, 0x16, 0xec, 0x80, 0x82, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr127, $sgpr32, 512, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.97, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x18, 0x0f, 0x90, 0xff, 0x18, 0x16, 0xec, 0x80, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr128, $sgpr32, 508, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.98, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x80, 0x19, 0x0f, 0x90, 0x80, 0x19, 0x16, 0xec, 0x80, 0xfe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr129, $sgpr32, 504, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.99, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x81, 0x19, 0x0f, 0x90, 0x81, 0x19, 0x16, 0xec, 0x80, 0xfc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr130, $sgpr32, 500, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.100, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x82, 0x19, 0x0f, 0x90, 0x82, 0x19, 0x16, 0xec, 0x80, 0xfa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr131, $sgpr32, 496, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.101, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x83, 0x19, 0x0f, 0x90, 0x83, 0x19, 0x16, 0xec, 0x80, 0xf8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr132, $sgpr32, 492, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.102, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x84, 0x19, 0x0f, 0x90, 0x84, 0x19, 0x16, 0xec, 0x80, 0xf6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr133, $sgpr32, 488, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.103, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x85, 0x19, 0x0f, 0x90, 0x85, 0x19, 0x16, 0xec, 0x80, 0xf4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr134, $sgpr32, 484, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.104, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x86, 0x19, 0x0f, 0x90, 0x86, 0x19, 0x16, 0xec, 0x80, 0xf2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr135, $sgpr32, 480, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.105, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x87, 0x19, 0x0f, 0x90, 0x87, 0x19, 0x16, 0xec, 0x80, 0xf0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr136, $sgpr32, 476, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.106, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x88, 0x19, 0x0f, 0x90, 0x88, 0x19, 0x16, 0xec, 0x80, 0xee, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr137, $sgpr32, 472, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.107, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x89, 0x19, 0x0f, 0x90, 0x89, 0x19, 0x16, 0xec, 0x80, 0xec, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr138, $sgpr32, 468, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.108, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8a, 0x19, 0x0f, 0x90, 0x8a, 0x19, 0x16, 0xec, 0x80, 0xea, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr139, $sgpr32, 464, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.109, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8b, 0x19, 0x0f, 0x90, 0x8b, 0x19, 0x16, 0xec, 0x80, 0xe8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr140, $sgpr32, 460, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.110, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8c, 0x19, 0x0f, 0x90, 0x8c, 0x19, 0x16, 0xec, 0x80, 0xe6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr141, $sgpr32, 456, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.111, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8d, 0x19, 0x0f, 0x90, 0x8d, 0x19, 0x16, 0xec, 0x80, 0xe4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr142, $sgpr32, 452, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.112, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8e, 0x19, 0x0f, 0x90, 0x8e, 0x19, 0x16, 0xec, 0x80, 0xe2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr143, $sgpr32, 448, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.113, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8f, 0x19, 0x0f, 0x90, 0x8f, 0x19, 0x16, 0xec, 0x80, 0xe0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr144, $sgpr32, 444, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.114, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x90, 0x19, 0x0f, 0x90, 0x90, 0x19, 0x16, 0xec, 0x80, 0xde, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr145, $sgpr32, 440, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.115, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x91, 0x19, 0x0f, 0x90, 0x91, 0x19, 0x16, 0xec, 0x80, 0xdc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr146, $sgpr32, 436, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.116, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x92, 0x19, 0x0f, 0x90, 0x92, 0x19, 0x16, 0xec, 0x80, 0xda, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr147, $sgpr32, 432, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.117, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x93, 0x19, 0x0f, 0x90, 0x93, 0x19, 0x16, 0xec, 0x80, 0xd8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr148, $sgpr32, 428, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.118, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x94, 0x19, 0x0f, 0x90, 0x94, 0x19, 0x16, 0xec, 0x80, 0xd6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr149, $sgpr32, 424, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.119, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x95, 0x19, 0x0f, 0x90, 0x95, 0x19, 0x16, 0xec, 0x80, 0xd4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr150, $sgpr32, 420, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.120, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x96, 0x19, 0x0f, 0x90, 0x96, 0x19, 0x16, 0xec, 0x80, 0xd2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr151, $sgpr32, 416, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.121, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x97, 0x19, 0x0f, 0x90, 0x97, 0x19, 0x16, 0xec, 0x80, 0xd0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr152, $sgpr32, 412, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.122, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x98, 0x19, 0x0f, 0x90, 0x98, 0x19, 0x16, 0xec, 0x80, 0xce, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr153, $sgpr32, 408, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.123, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x99, 0x19, 0x0f, 0x90, 0x99, 0x19, 0x16, 0xec, 0x80, 0xcc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr154, $sgpr32, 404, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.124, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9a, 0x19, 0x0f, 0x90, 0x9a, 0x19, 0x16, 0xec, 0x80, 0xca, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr155, $sgpr32, 400, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.125, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9b, 0x19, 0x0f, 0x90, 0x9b, 0x19, 0x16, 0xec, 0x80, 0xc8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr156, $sgpr32, 396, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.126, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9c, 0x19, 0x0f, 0x90, 0x9c, 0x19, 0x16, 0xec, 0x80, 0xc6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr157, $sgpr32, 392, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.127, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9d, 0x19, 0x0f, 0x90, 0x9d, 0x19, 0x16, 0xec, 0x80, 0xc4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr158, $sgpr32, 388, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.128, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9e, 0x19, 0x0f, 0x90, 0x9e, 0x19, 0x16, 0xec, 0x80, 0xc2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr159, $sgpr32, 384, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.129, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9f, 0x19, 0x0f, 0x90, 0x9f, 0x19, 0x16, 0xec, 0x80, 0xc0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr160, $sgpr32, 380, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.130, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa0, 0x19, 0x0f, 0x90, 0xa0, 0x19, 0x16, 0xec, 0x80, 0xbe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr161, $sgpr32, 376, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.131, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa1, 0x19, 0x0f, 0x90, 0xa1, 0x19, 0x16, 0xec, 0x80, 0xbc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr162, $sgpr32, 372, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.132, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa2, 0x19, 0x0f, 0x90, 0xa2, 0x19, 0x16, 0xec, 0x80, 0xba, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr163, $sgpr32, 368, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.133, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa3, 0x19, 0x0f, 0x90, 0xa3, 0x19, 0x16, 0xec, 0x80, 0xb8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr164, $sgpr32, 364, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.134, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa4, 0x19, 0x0f, 0x90, 0xa4, 0x19, 0x16, 0xec, 0x80, 0xb6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr165, $sgpr32, 360, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.135, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa5, 0x19, 0x0f, 0x90, 0xa5, 0x19, 0x16, 0xec, 0x80, 0xb4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr166, $sgpr32, 356, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.136, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa6, 0x19, 0x0f, 0x90, 0xa6, 0x19, 0x16, 0xec, 0x80, 0xb2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr167, $sgpr32, 352, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.137, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa7, 0x19, 0x0f, 0x90, 0xa7, 0x19, 0x16, 0xec, 0x80, 0xb0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr168, $sgpr32, 348, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.138, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa8, 0x19, 0x0f, 0x90, 0xa8, 0x19, 0x16, 0xec, 0x80, 0xae, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr169, $sgpr32, 344, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.139, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa9, 0x19, 0x0f, 0x90, 0xa9, 0x19, 0x16, 0xec, 0x80, 0xac, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr170, $sgpr32, 340, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.140, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaa, 0x19, 0x0f, 0x90, 0xaa, 0x19, 0x16, 0xec, 0x80, 0xaa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr171, $sgpr32, 336, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.141, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xab, 0x19, 0x0f, 0x90, 0xab, 0x19, 0x16, 0xec, 0x80, 0xa8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr172, $sgpr32, 332, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.142, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xac, 0x19, 0x0f, 0x90, 0xac, 0x19, 0x16, 0xec, 0x80, 0xa6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr173, $sgpr32, 328, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.143, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xad, 0x19, 0x0f, 0x90, 0xad, 0x19, 0x16, 0xec, 0x80, 0xa4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr174, $sgpr32, 324, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.144, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xae, 0x19, 0x0f, 0x90, 0xae, 0x19, 0x16, 0xec, 0x80, 0xa2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr175, $sgpr32, 320, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.145, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaf, 0x19, 0x0f, 0x90, 0xaf, 0x19, 0x16, 0xec, 0x80, 0xa0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr176, $sgpr32, 316, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.146, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb0, 0x19, 0x0f, 0x90, 0xb0, 0x19, 0x16, 0xec, 0x80, 0x9e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr177, $sgpr32, 312, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.147, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb1, 0x19, 0x0f, 0x90, 0xb1, 0x19, 0x16, 0xec, 0x80, 0x9c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr178, $sgpr32, 308, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.148, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb2, 0x19, 0x0f, 0x90, 0xb2, 0x19, 0x16, 0xec, 0x80, 0x9a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr179, $sgpr32, 304, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.149, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb3, 0x19, 0x0f, 0x90, 0xb3, 0x19, 0x16, 0xec, 0x80, 0x98, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr180, $sgpr32, 300, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.150, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb4, 0x19, 0x0f, 0x90, 0xb4, 0x19, 0x16, 0xec, 0x80, 0x96, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr181, $sgpr32, 296, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.151, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb5, 0x19, 0x0f, 0x90, 0xb5, 0x19, 0x16, 0xec, 0x80, 0x94, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr182, $sgpr32, 292, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.152, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb6, 0x19, 0x0f, 0x90, 0xb6, 0x19, 0x16, 0xec, 0x80, 0x92, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr183, $sgpr32, 288, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.153, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb7, 0x19, 0x0f, 0x90, 0xb7, 0x19, 0x16, 0xec, 0x80, 0x90, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr184, $sgpr32, 284, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.154, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb8, 0x19, 0x0f, 0x90, 0xb8, 0x19, 0x16, 0xec, 0x80, 0x8e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr185, $sgpr32, 280, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.155, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb9, 0x19, 0x0f, 0x90, 0xb9, 0x19, 0x16, 0xec, 0x80, 0x8c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr186, $sgpr32, 276, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.156, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xba, 0x19, 0x0f, 0x90, 0xba, 0x19, 0x16, 0xec, 0x80, 0x8a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr187, $sgpr32, 272, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.157, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbb, 0x19, 0x0f, 0x90, 0xbb, 0x19, 0x16, 0xec, 0x80, 0x88, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr188, $sgpr32, 268, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.158, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbc, 0x19, 0x0f, 0x90, 0xbc, 0x19, 0x16, 0xec, 0x80, 0x86, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr189, $sgpr32, 264, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.159, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbd, 0x19, 0x0f, 0x90, 0xbd, 0x19, 0x16, 0xec, 0x80, 0x84, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr190, $sgpr32, 260, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.160, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbe, 0x19, 0x0f, 0x90, 0xbe, 0x19, 0x16, 0xec, 0x80, 0x82, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr191, $sgpr32, 256, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.161, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbf, 0x19, 0x0f, 0x90, 0xbf, 0x19, 0x16, 0xec, 0x80, 0x80, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr192, $sgpr32, 252, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.162, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc0, 0x19, 0x0e, 0x90, 0xc0, 0x19, 0x16, 0xec, 0x80, 0x7e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr193, $sgpr32, 248, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.163, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc1, 0x19, 0x0e, 0x90, 0xc1, 0x19, 0x16, 0xec, 0x80, 0x7c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr194, $sgpr32, 244, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.164, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc2, 0x19, 0x0e, 0x90, 0xc2, 0x19, 0x16, 0xec, 0x80, 0x7a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr195, $sgpr32, 240, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.165, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc3, 0x19, 0x0e, 0x90, 0xc3, 0x19, 0x16, 0xec, 0x80, 0x78, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr196, $sgpr32, 236, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.166, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc4, 0x19, 0x0e, 0x90, 0xc4, 0x19, 0x16, 0xec, 0x80, 0x76, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr197, $sgpr32, 232, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.167, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc5, 0x19, 0x0e, 0x90, 0xc5, 0x19, 0x16, 0xec, 0x80, 0x74, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr198, $sgpr32, 228, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.168, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc6, 0x19, 0x0e, 0x90, 0xc6, 0x19, 0x16, 0xec, 0x80, 0x72, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr199, $sgpr32, 224, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.169, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc7, 0x19, 0x0e, 0x90, 0xc7, 0x19, 0x16, 0xec, 0x80, 0x70, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr200, $sgpr32, 220, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.170, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc8, 0x19, 0x0e, 0x90, 0xc8, 0x19, 0x16, 0xec, 0x80, 0x6e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr201, $sgpr32, 216, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.171, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc9, 0x19, 0x0e, 0x90, 0xc9, 0x19, 0x16, 0xec, 0x80, 0x6c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr202, $sgpr32, 212, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.172, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xca, 0x19, 0x0e, 0x90, 0xca, 0x19, 0x16, 0xec, 0x80, 0x6a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr203, $sgpr32, 208, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.173, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcb, 0x19, 0x0e, 0x90, 0xcb, 0x19, 0x16, 0xec, 0x80, 0x68, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr204, $sgpr32, 204, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.174, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcc, 0x19, 0x0e, 0x90, 0xcc, 0x19, 0x16, 0xec, 0x80, 0x66, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr205, $sgpr32, 200, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.175, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcd, 0x19, 0x0e, 0x90, 0xcd, 0x19, 0x16, 0xec, 0x80, 0x64, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr206, $sgpr32, 196, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.176, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xce, 0x19, 0x0e, 0x90, 0xce, 0x19, 0x16, 0xec, 0x80, 0x62, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr207, $sgpr32, 192, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.177, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcf, 0x19, 0x0e, 0x90, 0xcf, 0x19, 0x16, 0xec, 0x80, 0x60, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr208, $sgpr32, 188, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.178, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x19, 0x0e, 0x90, 0xd0, 0x19, 0x16, 0xec, 0x80, 0x5e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr209, $sgpr32, 184, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.179, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x19, 0x0e, 0x90, 0xd1, 0x19, 0x16, 0xec, 0x80, 0x5c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr210, $sgpr32, 180, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.180, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x19, 0x0e, 0x90, 0xd2, 0x19, 0x16, 0xec, 0x80, 0x5a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr211, $sgpr32, 176, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.181, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x19, 0x0e, 0x90, 0xd3, 0x19, 0x16, 0xec, 0x80, 0x58, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr212, $sgpr32, 172, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.182, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x19, 0x0e, 0x90, 0xd4, 0x19, 0x16, 0xec, 0x80, 0x56, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr213, $sgpr32, 168, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.183, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x19, 0x0e, 0x90, 0xd5, 0x19, 0x16, 0xec, 0x80, 0x54, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr214, $sgpr32, 164, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.184, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x19, 0x0e, 0x90, 0xd6, 0x19, 0x16, 0xec, 0x80, 0x52, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr215, $sgpr32, 160, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.185, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x19, 0x0e, 0x90, 0xd7, 0x19, 0x16, 0xec, 0x80, 0x50, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr216, $sgpr32, 156, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.186, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x19, 0x0e, 0x90, 0xd8, 0x19, 0x16, 0xec, 0x80, 0x4e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr217, $sgpr32, 152, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.187, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x19, 0x0e, 0x90, 0xd9, 0x19, 0x16, 0xec, 0x80, 0x4c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr218, $sgpr32, 148, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.188, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x19, 0x0e, 0x90, 0xda, 0x19, 0x16, 0xec, 0x80, 0x4a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr219, $sgpr32, 144, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.189, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x19, 0x0e, 0x90, 0xdb, 0x19, 0x16, 0xec, 0x80, 0x48, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr220, $sgpr32, 140, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.190, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x19, 0x0e, 0x90, 0xdc, 0x19, 0x16, 0xec, 0x80, 0x46, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr221, $sgpr32, 136, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.191, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x19, 0x0e, 0x90, 0xdd, 0x19, 0x16, 0xec, 0x80, 0x44, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr222, $sgpr32, 132, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.192, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x19, 0x0e, 0x90, 0xde, 0x19, 0x16, 0xec, 0x80, 0x42, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr223, $sgpr32, 128, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.193, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x19, 0x0e, 0x90, 0xdf, 0x19, 0x16, 0xec, 0x80, 0x40, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr224, $sgpr32, 124, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.194, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x19, 0x0e, 0x90, 0xe0, 0x19, 0x16, 0xec, 0x80, 0x3e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr225, $sgpr32, 120, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.195, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x19, 0x0e, 0x90, 0xe1, 0x19, 0x16, 0xec, 0x80, 0x3c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr226, $sgpr32, 116, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.196, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x19, 0x0e, 0x90, 0xe2, 0x19, 0x16, 0xec, 0x80, 0x3a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr227, $sgpr32, 112, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.197, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x19, 0x0e, 0x90, 0xe3, 0x19, 0x16, 0xec, 0x80, 0x38, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr228, $sgpr32, 108, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.198, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x19, 0x0e, 0x90, 0xe4, 0x19, 0x16, 0xec, 0x80, 0x36, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr229, $sgpr32, 104, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.199, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x19, 0x0e, 0x90, 0xe5, 0x19, 0x16, 0xec, 0x80, 0x34, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr230, $sgpr32, 100, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.200, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x19, 0x0e, 0x90, 0xe6, 0x19, 0x16, 0xec, 0x80, 0x32, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr231, $sgpr32, 96, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.201, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x19, 0x0e, 0x90, 0xe7, 0x19, 0x16, 0xec, 0x80, 0x30, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr232, $sgpr32, 92, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.202, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x19, 0x0e, 0x90, 0xe8, 0x19, 0x16, 0xec, 0x80, 0x2e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr233, $sgpr32, 88, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.203, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x19, 0x0e, 0x90, 0xe9, 0x19, 0x16, 0xec, 0x80, 0x2c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr234, $sgpr32, 84, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.204, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x19, 0x0e, 0x90, 0xea, 0x19, 0x16, 0xec, 0x80, 0x2a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr235, $sgpr32, 80, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.205, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x19, 0x0e, 0x90, 0xeb, 0x19, 0x16, 0xec, 0x80, 0x28, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr236, $sgpr32, 76, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.206, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x19, 0x0e, 0x90, 0xec, 0x19, 0x16, 0xec, 0x80, 0x26, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr237, $sgpr32, 72, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.207, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x19, 0x0e, 0x90, 0xed, 0x19, 0x16, 0xec, 0x80, 0x24, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr238, $sgpr32, 68, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.208, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x19, 0x0e, 0x90, 0xee, 0x19, 0x16, 0xec, 0x80, 0x22, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr239, $sgpr32, 64, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.209, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x19, 0x0e, 0x90, 0xef, 0x19, 0x16, 0xec, 0x80, 0x20, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr240, $sgpr32, 60, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.210, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x19, 0x0e, 0x90, 0xf0, 0x19, 0x16, 0xec, 0x80, 0x1e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr241, $sgpr32, 56, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.211, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x19, 0x0e, 0x90, 0xf1, 0x19, 0x16, 0xec, 0x80, 0x1c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr242, $sgpr32, 52, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.212, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x19, 0x0e, 0x90, 0xf2, 0x19, 0x16, 0xec, 0x80, 0x1a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr243, $sgpr32, 48, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.213, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x19, 0x0e, 0x90, 0xf3, 0x19, 0x16, 0xec, 0x80, 0x18, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr244, $sgpr32, 44, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.214, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x19, 0x0e, 0x90, 0xf4, 0x19, 0x16, 0xec, 0x80, 0x16, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr245, $sgpr32, 40, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.215, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x19, 0x0e, 0x90, 0xf5, 0x19, 0x16, 0xec, 0x80, 0x14, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr246, $sgpr32, 36, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.216, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x19, 0x0e, 0x90, 0xf6, 0x19, 0x16, 0xec, 0x80, 0x12, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr247, $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.217, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x19, 0x0e, 0x90, 0xf7, 0x19, 0x16, 0xec, 0x80, 0x10, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr248, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.218, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x19, 0x0e, 0x90, 0xf8, 0x19, 0x16, 0xec, 0x80, 0x0e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr249, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.219, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x19, 0x0e, 0x90, 0xf9, 0x19, 0x16, 0xec, 0x80, 0x0c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr250, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.220, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x19, 0x0e, 0x90, 0xfa, 0x19, 0x16, 0xec, 0x80, 0x0a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr251, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.221, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x19, 0x0e, 0x90, 0xfb, 0x19, 0x16, 0xec, 0x80, 0x08, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr252, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.222, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x19, 0x0e, 0x90, 0xfc, 0x19, 0x16, 0xec, 0x80, 0x06, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr253, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.223, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x19, 0x0e, 0x90, 0xfd, 0x19, 0x16, 0xec, 0x80, 0x04, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr254, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.224, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x19, 0x0e, 0x90, 0xfe, 0x19, 0x16, 0xec, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr255, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.225, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x19, 0x0d, 0x90, 0xff, 0x19, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: S_CMP_EQ_U32 0, 0, implicit-def $scc ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 704, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.226, addrspace 5) ; GFX90A-FLATSCR-NEXT: $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec @@ -4072,229 +4968,453 @@ ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr28_lo16 ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr29_lo16 ; GFX90A-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr32, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr32_lo16, $vgpr0_lo16 ; GFX90A-NEXT: $vgpr1 = V_ACCVGPR_READ_B32_e64 killed $agpr33, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr33_lo16, $vgpr1_lo16 ; GFX90A-NEXT: $vgpr2 = V_ACCVGPR_READ_B32_e64 killed $agpr34, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr34_lo16, $vgpr2_lo16 ; GFX90A-NEXT: $vgpr3 = V_ACCVGPR_READ_B32_e64 killed $agpr35, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr35_lo16, $vgpr3_lo16 ; GFX90A-NEXT: $vgpr4 = V_ACCVGPR_READ_B32_e64 killed $agpr36, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr36_lo16, $vgpr4_lo16 ; GFX90A-NEXT: $vgpr5 = V_ACCVGPR_READ_B32_e64 killed $agpr37, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr37_lo16, $vgpr5_lo16 ; GFX90A-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 killed $agpr38, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr38_lo16, $vgpr6_lo16 ; GFX90A-NEXT: $vgpr7 = V_ACCVGPR_READ_B32_e64 killed $agpr39, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr39_lo16, $vgpr7_lo16 ; GFX90A-NEXT: $vgpr8 = V_ACCVGPR_READ_B32_e64 killed $agpr40, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr40_lo16, $vgpr8_lo16 ; GFX90A-NEXT: $vgpr9 = V_ACCVGPR_READ_B32_e64 killed $agpr41, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr41_lo16, $vgpr9_lo16 ; GFX90A-NEXT: $vgpr10 = V_ACCVGPR_READ_B32_e64 killed $agpr42, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr42_lo16, $vgpr10_lo16 ; GFX90A-NEXT: $vgpr11 = V_ACCVGPR_READ_B32_e64 killed $agpr43, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr43_lo16, $vgpr11_lo16 ; GFX90A-NEXT: $vgpr12 = V_ACCVGPR_READ_B32_e64 killed $agpr44, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr44_lo16, $vgpr12_lo16 ; GFX90A-NEXT: $vgpr13 = V_ACCVGPR_READ_B32_e64 killed $agpr45, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr45_lo16, $vgpr13_lo16 ; GFX90A-NEXT: $vgpr14 = V_ACCVGPR_READ_B32_e64 killed $agpr46, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr46_lo16, $vgpr14_lo16 ; GFX90A-NEXT: $vgpr15 = V_ACCVGPR_READ_B32_e64 killed $agpr47, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr47_lo16, $vgpr15_lo16 ; GFX90A-NEXT: $vgpr16 = V_ACCVGPR_READ_B32_e64 killed $agpr48, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr48_lo16, $vgpr16_lo16 ; GFX90A-NEXT: $vgpr17 = V_ACCVGPR_READ_B32_e64 killed $agpr49, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr49_lo16, $vgpr17_lo16 ; GFX90A-NEXT: $vgpr18 = V_ACCVGPR_READ_B32_e64 killed $agpr50, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr50_lo16, $vgpr18_lo16 ; GFX90A-NEXT: $vgpr19 = V_ACCVGPR_READ_B32_e64 killed $agpr51, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr51_lo16, $vgpr19_lo16 ; GFX90A-NEXT: $vgpr20 = V_ACCVGPR_READ_B32_e64 killed $agpr52, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr52_lo16, $vgpr20_lo16 ; GFX90A-NEXT: $vgpr21 = V_ACCVGPR_READ_B32_e64 killed $agpr53, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr53_lo16, $vgpr21_lo16 ; GFX90A-NEXT: $vgpr22 = V_ACCVGPR_READ_B32_e64 killed $agpr54, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr54_lo16, $vgpr22_lo16 ; GFX90A-NEXT: $vgpr23 = V_ACCVGPR_READ_B32_e64 killed $agpr55, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr55_lo16, $vgpr23_lo16 ; GFX90A-NEXT: $vgpr24 = V_ACCVGPR_READ_B32_e64 killed $agpr56, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr56_lo16, $vgpr24_lo16 ; GFX90A-NEXT: $vgpr25 = V_ACCVGPR_READ_B32_e64 killed $agpr57, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr57_lo16, $vgpr25_lo16 ; GFX90A-NEXT: $vgpr26 = V_ACCVGPR_READ_B32_e64 killed $agpr58, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr58_lo16, $vgpr26_lo16 ; GFX90A-NEXT: $vgpr27 = V_ACCVGPR_READ_B32_e64 killed $agpr59, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr59_lo16, $vgpr27_lo16 ; GFX90A-NEXT: $vgpr28 = V_ACCVGPR_READ_B32_e64 killed $agpr60, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr60_lo16, $vgpr28_lo16 ; GFX90A-NEXT: $vgpr29 = V_ACCVGPR_READ_B32_e64 killed $agpr61, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr61_lo16, $vgpr29_lo16 ; GFX90A-NEXT: $vgpr30 = V_ACCVGPR_READ_B32_e64 killed $agpr62, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr62_lo16, $vgpr30_lo16 ; GFX90A-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 killed $agpr63, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr63_lo16, $vgpr31_lo16 ; GFX90A-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr64, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr64_lo16, $vgpr32_lo16 ; GFX90A-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 killed $agpr65, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr65_lo16, $vgpr33_lo16 ; GFX90A-NEXT: $vgpr34 = V_ACCVGPR_READ_B32_e64 killed $agpr66, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr66_lo16, $vgpr34_lo16 ; GFX90A-NEXT: $vgpr35 = V_ACCVGPR_READ_B32_e64 killed $agpr67, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr67_lo16, $vgpr35_lo16 ; GFX90A-NEXT: $vgpr36 = V_ACCVGPR_READ_B32_e64 killed $agpr68, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr68_lo16, $vgpr36_lo16 ; GFX90A-NEXT: $vgpr37 = V_ACCVGPR_READ_B32_e64 killed $agpr69, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr69_lo16, $vgpr37_lo16 ; GFX90A-NEXT: $vgpr38 = V_ACCVGPR_READ_B32_e64 killed $agpr70, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr70_lo16, $vgpr38_lo16 ; GFX90A-NEXT: $vgpr39 = V_ACCVGPR_READ_B32_e64 killed $agpr71, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr71_lo16, $vgpr39_lo16 ; GFX90A-NEXT: $vgpr48 = V_ACCVGPR_READ_B32_e64 killed $agpr72, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr72_lo16, $vgpr48_lo16 ; GFX90A-NEXT: $vgpr49 = V_ACCVGPR_READ_B32_e64 killed $agpr73, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr73_lo16, $vgpr49_lo16 ; GFX90A-NEXT: $vgpr50 = V_ACCVGPR_READ_B32_e64 killed $agpr74, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr74_lo16, $vgpr50_lo16 ; GFX90A-NEXT: $vgpr51 = V_ACCVGPR_READ_B32_e64 killed $agpr75, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr75_lo16, $vgpr51_lo16 ; GFX90A-NEXT: $vgpr52 = V_ACCVGPR_READ_B32_e64 killed $agpr76, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr76_lo16, $vgpr52_lo16 ; GFX90A-NEXT: $vgpr53 = V_ACCVGPR_READ_B32_e64 killed $agpr77, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr77_lo16, $vgpr53_lo16 ; GFX90A-NEXT: $vgpr54 = V_ACCVGPR_READ_B32_e64 killed $agpr78, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr78_lo16, $vgpr54_lo16 ; GFX90A-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 killed $agpr79, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr79_lo16, $vgpr55_lo16 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr80, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 700, 0, 0, 0, implicit $exec :: (store (s32) into %stack.50, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x18, 0x0f, 0x90, 0xd0, 0x18, 0x16, 0xec, 0x80, 0xde, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr81, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 696, 0, 0, 0, implicit $exec :: (store (s32) into %stack.51, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x18, 0x0f, 0x90, 0xd1, 0x18, 0x16, 0xec, 0x80, 0xdc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr82, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 692, 0, 0, 0, implicit $exec :: (store (s32) into %stack.52, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x18, 0x0f, 0x90, 0xd2, 0x18, 0x16, 0xec, 0x80, 0xda, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr83, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 688, 0, 0, 0, implicit $exec :: (store (s32) into %stack.53, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x18, 0x0f, 0x90, 0xd3, 0x18, 0x16, 0xec, 0x80, 0xd8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr84, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 684, 0, 0, 0, implicit $exec :: (store (s32) into %stack.54, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x18, 0x0f, 0x90, 0xd4, 0x18, 0x16, 0xec, 0x80, 0xd6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr85, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 680, 0, 0, 0, implicit $exec :: (store (s32) into %stack.55, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x18, 0x0f, 0x90, 0xd5, 0x18, 0x16, 0xec, 0x80, 0xd4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr86, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 676, 0, 0, 0, implicit $exec :: (store (s32) into %stack.56, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x18, 0x0f, 0x90, 0xd6, 0x18, 0x16, 0xec, 0x80, 0xd2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr87, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 672, 0, 0, 0, implicit $exec :: (store (s32) into %stack.57, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x18, 0x0f, 0x90, 0xd7, 0x18, 0x16, 0xec, 0x80, 0xd0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr88, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 668, 0, 0, 0, implicit $exec :: (store (s32) into %stack.58, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x18, 0x0f, 0x90, 0xd8, 0x18, 0x16, 0xec, 0x80, 0xce, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr89, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 664, 0, 0, 0, implicit $exec :: (store (s32) into %stack.59, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x18, 0x0f, 0x90, 0xd9, 0x18, 0x16, 0xec, 0x80, 0xcc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr90, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 660, 0, 0, 0, implicit $exec :: (store (s32) into %stack.60, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x18, 0x0f, 0x90, 0xda, 0x18, 0x16, 0xec, 0x80, 0xca, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr91, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 656, 0, 0, 0, implicit $exec :: (store (s32) into %stack.61, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x18, 0x0f, 0x90, 0xdb, 0x18, 0x16, 0xec, 0x80, 0xc8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr92, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 652, 0, 0, 0, implicit $exec :: (store (s32) into %stack.62, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x18, 0x0f, 0x90, 0xdc, 0x18, 0x16, 0xec, 0x80, 0xc6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr93, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 648, 0, 0, 0, implicit $exec :: (store (s32) into %stack.63, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x18, 0x0f, 0x90, 0xdd, 0x18, 0x16, 0xec, 0x80, 0xc4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr94, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 644, 0, 0, 0, implicit $exec :: (store (s32) into %stack.64, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x18, 0x0f, 0x90, 0xde, 0x18, 0x16, 0xec, 0x80, 0xc2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr95, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 640, 0, 0, 0, implicit $exec :: (store (s32) into %stack.65, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x18, 0x0f, 0x90, 0xdf, 0x18, 0x16, 0xec, 0x80, 0xc0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr96, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 636, 0, 0, 0, implicit $exec :: (store (s32) into %stack.66, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x18, 0x0f, 0x90, 0xe0, 0x18, 0x16, 0xec, 0x80, 0xbe, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr97, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 632, 0, 0, 0, implicit $exec :: (store (s32) into %stack.67, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x18, 0x0f, 0x90, 0xe1, 0x18, 0x16, 0xec, 0x80, 0xbc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr98, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 628, 0, 0, 0, implicit $exec :: (store (s32) into %stack.68, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x18, 0x0f, 0x90, 0xe2, 0x18, 0x16, 0xec, 0x80, 0xba, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr99, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 624, 0, 0, 0, implicit $exec :: (store (s32) into %stack.69, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x18, 0x0f, 0x90, 0xe3, 0x18, 0x16, 0xec, 0x80, 0xb8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr100, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 620, 0, 0, 0, implicit $exec :: (store (s32) into %stack.70, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x18, 0x0f, 0x90, 0xe4, 0x18, 0x16, 0xec, 0x80, 0xb6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr101, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 616, 0, 0, 0, implicit $exec :: (store (s32) into %stack.71, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x18, 0x0f, 0x90, 0xe5, 0x18, 0x16, 0xec, 0x80, 0xb4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr102, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 612, 0, 0, 0, implicit $exec :: (store (s32) into %stack.72, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x18, 0x0f, 0x90, 0xe6, 0x18, 0x16, 0xec, 0x80, 0xb2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr103, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 608, 0, 0, 0, implicit $exec :: (store (s32) into %stack.73, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x18, 0x0f, 0x90, 0xe7, 0x18, 0x16, 0xec, 0x80, 0xb0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr104, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 604, 0, 0, 0, implicit $exec :: (store (s32) into %stack.74, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x18, 0x0f, 0x90, 0xe8, 0x18, 0x16, 0xec, 0x80, 0xae, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr105, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 600, 0, 0, 0, implicit $exec :: (store (s32) into %stack.75, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x18, 0x0f, 0x90, 0xe9, 0x18, 0x16, 0xec, 0x80, 0xac, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr106, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 596, 0, 0, 0, implicit $exec :: (store (s32) into %stack.76, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x18, 0x0f, 0x90, 0xea, 0x18, 0x16, 0xec, 0x80, 0xaa, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr107, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 592, 0, 0, 0, implicit $exec :: (store (s32) into %stack.77, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x18, 0x0f, 0x90, 0xeb, 0x18, 0x16, 0xec, 0x80, 0xa8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr108, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 588, 0, 0, 0, implicit $exec :: (store (s32) into %stack.78, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x18, 0x0f, 0x90, 0xec, 0x18, 0x16, 0xec, 0x80, 0xa6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr109, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 584, 0, 0, 0, implicit $exec :: (store (s32) into %stack.79, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x18, 0x0f, 0x90, 0xed, 0x18, 0x16, 0xec, 0x80, 0xa4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr110, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 580, 0, 0, 0, implicit $exec :: (store (s32) into %stack.80, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x18, 0x0f, 0x90, 0xee, 0x18, 0x16, 0xec, 0x80, 0xa2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr111, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 576, 0, 0, 0, implicit $exec :: (store (s32) into %stack.81, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x18, 0x0f, 0x90, 0xef, 0x18, 0x16, 0xec, 0x80, 0xa0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr112, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 572, 0, 0, 0, implicit $exec :: (store (s32) into %stack.82, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x18, 0x0f, 0x90, 0xf0, 0x18, 0x16, 0xec, 0x80, 0x9e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr113, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 568, 0, 0, 0, implicit $exec :: (store (s32) into %stack.83, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x18, 0x0f, 0x90, 0xf1, 0x18, 0x16, 0xec, 0x80, 0x9c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr114, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 564, 0, 0, 0, implicit $exec :: (store (s32) into %stack.84, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x18, 0x0f, 0x90, 0xf2, 0x18, 0x16, 0xec, 0x80, 0x9a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr115, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 560, 0, 0, 0, implicit $exec :: (store (s32) into %stack.85, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x18, 0x0f, 0x90, 0xf3, 0x18, 0x16, 0xec, 0x80, 0x98, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr116, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 556, 0, 0, 0, implicit $exec :: (store (s32) into %stack.86, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x18, 0x0f, 0x90, 0xf4, 0x18, 0x16, 0xec, 0x80, 0x96, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr117, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 552, 0, 0, 0, implicit $exec :: (store (s32) into %stack.87, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x18, 0x0f, 0x90, 0xf5, 0x18, 0x16, 0xec, 0x80, 0x94, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr118, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 548, 0, 0, 0, implicit $exec :: (store (s32) into %stack.88, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x18, 0x0f, 0x90, 0xf6, 0x18, 0x16, 0xec, 0x80, 0x92, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr119, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 544, 0, 0, 0, implicit $exec :: (store (s32) into %stack.89, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x18, 0x0f, 0x90, 0xf7, 0x18, 0x16, 0xec, 0x80, 0x90, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr120, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 540, 0, 0, 0, implicit $exec :: (store (s32) into %stack.90, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x18, 0x0f, 0x90, 0xf8, 0x18, 0x16, 0xec, 0x80, 0x8e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr121, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 536, 0, 0, 0, implicit $exec :: (store (s32) into %stack.91, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x18, 0x0f, 0x90, 0xf9, 0x18, 0x16, 0xec, 0x80, 0x8c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr122, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 532, 0, 0, 0, implicit $exec :: (store (s32) into %stack.92, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x18, 0x0f, 0x90, 0xfa, 0x18, 0x16, 0xec, 0x80, 0x8a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr123, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 528, 0, 0, 0, implicit $exec :: (store (s32) into %stack.93, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x18, 0x0f, 0x90, 0xfb, 0x18, 0x16, 0xec, 0x80, 0x88, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr124, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 524, 0, 0, 0, implicit $exec :: (store (s32) into %stack.94, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x18, 0x0f, 0x90, 0xfc, 0x18, 0x16, 0xec, 0x80, 0x86, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr125, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 520, 0, 0, 0, implicit $exec :: (store (s32) into %stack.95, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x18, 0x0f, 0x90, 0xfd, 0x18, 0x16, 0xec, 0x80, 0x84, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr126, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 516, 0, 0, 0, implicit $exec :: (store (s32) into %stack.96, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x18, 0x0f, 0x90, 0xfe, 0x18, 0x16, 0xec, 0x80, 0x82, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr127, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 512, 0, 0, 0, implicit $exec :: (store (s32) into %stack.97, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x18, 0x0f, 0x90, 0xff, 0x18, 0x16, 0xec, 0x80, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr128, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 508, 0, 0, 0, implicit $exec :: (store (s32) into %stack.98, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x80, 0x19, 0x0f, 0x90, 0x80, 0x19, 0x16, 0xec, 0x80, 0xfe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr129, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 504, 0, 0, 0, implicit $exec :: (store (s32) into %stack.99, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x81, 0x19, 0x0f, 0x90, 0x81, 0x19, 0x16, 0xec, 0x80, 0xfc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr130, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 500, 0, 0, 0, implicit $exec :: (store (s32) into %stack.100, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x82, 0x19, 0x0f, 0x90, 0x82, 0x19, 0x16, 0xec, 0x80, 0xfa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr131, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 496, 0, 0, 0, implicit $exec :: (store (s32) into %stack.101, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x83, 0x19, 0x0f, 0x90, 0x83, 0x19, 0x16, 0xec, 0x80, 0xf8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr132, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 492, 0, 0, 0, implicit $exec :: (store (s32) into %stack.102, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x84, 0x19, 0x0f, 0x90, 0x84, 0x19, 0x16, 0xec, 0x80, 0xf6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr133, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 488, 0, 0, 0, implicit $exec :: (store (s32) into %stack.103, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x85, 0x19, 0x0f, 0x90, 0x85, 0x19, 0x16, 0xec, 0x80, 0xf4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr134, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 484, 0, 0, 0, implicit $exec :: (store (s32) into %stack.104, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x86, 0x19, 0x0f, 0x90, 0x86, 0x19, 0x16, 0xec, 0x80, 0xf2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr135, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 480, 0, 0, 0, implicit $exec :: (store (s32) into %stack.105, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x87, 0x19, 0x0f, 0x90, 0x87, 0x19, 0x16, 0xec, 0x80, 0xf0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr136, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 476, 0, 0, 0, implicit $exec :: (store (s32) into %stack.106, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x88, 0x19, 0x0f, 0x90, 0x88, 0x19, 0x16, 0xec, 0x80, 0xee, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr137, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 472, 0, 0, 0, implicit $exec :: (store (s32) into %stack.107, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x89, 0x19, 0x0f, 0x90, 0x89, 0x19, 0x16, 0xec, 0x80, 0xec, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr138, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 468, 0, 0, 0, implicit $exec :: (store (s32) into %stack.108, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8a, 0x19, 0x0f, 0x90, 0x8a, 0x19, 0x16, 0xec, 0x80, 0xea, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr139, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 464, 0, 0, 0, implicit $exec :: (store (s32) into %stack.109, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8b, 0x19, 0x0f, 0x90, 0x8b, 0x19, 0x16, 0xec, 0x80, 0xe8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr140, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 460, 0, 0, 0, implicit $exec :: (store (s32) into %stack.110, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8c, 0x19, 0x0f, 0x90, 0x8c, 0x19, 0x16, 0xec, 0x80, 0xe6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr141, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 456, 0, 0, 0, implicit $exec :: (store (s32) into %stack.111, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8d, 0x19, 0x0f, 0x90, 0x8d, 0x19, 0x16, 0xec, 0x80, 0xe4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr142, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 452, 0, 0, 0, implicit $exec :: (store (s32) into %stack.112, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8e, 0x19, 0x0f, 0x90, 0x8e, 0x19, 0x16, 0xec, 0x80, 0xe2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr143, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 448, 0, 0, 0, implicit $exec :: (store (s32) into %stack.113, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8f, 0x19, 0x0f, 0x90, 0x8f, 0x19, 0x16, 0xec, 0x80, 0xe0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr144, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 444, 0, 0, 0, implicit $exec :: (store (s32) into %stack.114, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x90, 0x19, 0x0f, 0x90, 0x90, 0x19, 0x16, 0xec, 0x80, 0xde, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr145, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 440, 0, 0, 0, implicit $exec :: (store (s32) into %stack.115, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x91, 0x19, 0x0f, 0x90, 0x91, 0x19, 0x16, 0xec, 0x80, 0xdc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr146, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 436, 0, 0, 0, implicit $exec :: (store (s32) into %stack.116, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x92, 0x19, 0x0f, 0x90, 0x92, 0x19, 0x16, 0xec, 0x80, 0xda, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr147, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 432, 0, 0, 0, implicit $exec :: (store (s32) into %stack.117, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x93, 0x19, 0x0f, 0x90, 0x93, 0x19, 0x16, 0xec, 0x80, 0xd8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr148, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 428, 0, 0, 0, implicit $exec :: (store (s32) into %stack.118, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x94, 0x19, 0x0f, 0x90, 0x94, 0x19, 0x16, 0xec, 0x80, 0xd6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr149, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 424, 0, 0, 0, implicit $exec :: (store (s32) into %stack.119, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x95, 0x19, 0x0f, 0x90, 0x95, 0x19, 0x16, 0xec, 0x80, 0xd4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr150, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 420, 0, 0, 0, implicit $exec :: (store (s32) into %stack.120, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x96, 0x19, 0x0f, 0x90, 0x96, 0x19, 0x16, 0xec, 0x80, 0xd2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr151, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 416, 0, 0, 0, implicit $exec :: (store (s32) into %stack.121, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x97, 0x19, 0x0f, 0x90, 0x97, 0x19, 0x16, 0xec, 0x80, 0xd0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr152, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 412, 0, 0, 0, implicit $exec :: (store (s32) into %stack.122, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x98, 0x19, 0x0f, 0x90, 0x98, 0x19, 0x16, 0xec, 0x80, 0xce, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr153, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 408, 0, 0, 0, implicit $exec :: (store (s32) into %stack.123, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x99, 0x19, 0x0f, 0x90, 0x99, 0x19, 0x16, 0xec, 0x80, 0xcc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr154, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 404, 0, 0, 0, implicit $exec :: (store (s32) into %stack.124, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9a, 0x19, 0x0f, 0x90, 0x9a, 0x19, 0x16, 0xec, 0x80, 0xca, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr155, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 400, 0, 0, 0, implicit $exec :: (store (s32) into %stack.125, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9b, 0x19, 0x0f, 0x90, 0x9b, 0x19, 0x16, 0xec, 0x80, 0xc8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr156, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 396, 0, 0, 0, implicit $exec :: (store (s32) into %stack.126, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9c, 0x19, 0x0f, 0x90, 0x9c, 0x19, 0x16, 0xec, 0x80, 0xc6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr157, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 392, 0, 0, 0, implicit $exec :: (store (s32) into %stack.127, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9d, 0x19, 0x0f, 0x90, 0x9d, 0x19, 0x16, 0xec, 0x80, 0xc4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr158, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 388, 0, 0, 0, implicit $exec :: (store (s32) into %stack.128, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9e, 0x19, 0x0f, 0x90, 0x9e, 0x19, 0x16, 0xec, 0x80, 0xc2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr159, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 384, 0, 0, 0, implicit $exec :: (store (s32) into %stack.129, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9f, 0x19, 0x0f, 0x90, 0x9f, 0x19, 0x16, 0xec, 0x80, 0xc0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr160, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 380, 0, 0, 0, implicit $exec :: (store (s32) into %stack.130, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa0, 0x19, 0x0f, 0x90, 0xa0, 0x19, 0x16, 0xec, 0x80, 0xbe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr161, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 376, 0, 0, 0, implicit $exec :: (store (s32) into %stack.131, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa1, 0x19, 0x0f, 0x90, 0xa1, 0x19, 0x16, 0xec, 0x80, 0xbc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr162, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 372, 0, 0, 0, implicit $exec :: (store (s32) into %stack.132, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa2, 0x19, 0x0f, 0x90, 0xa2, 0x19, 0x16, 0xec, 0x80, 0xba, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr163, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 368, 0, 0, 0, implicit $exec :: (store (s32) into %stack.133, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa3, 0x19, 0x0f, 0x90, 0xa3, 0x19, 0x16, 0xec, 0x80, 0xb8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr164, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 364, 0, 0, 0, implicit $exec :: (store (s32) into %stack.134, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa4, 0x19, 0x0f, 0x90, 0xa4, 0x19, 0x16, 0xec, 0x80, 0xb6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr165, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 360, 0, 0, 0, implicit $exec :: (store (s32) into %stack.135, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa5, 0x19, 0x0f, 0x90, 0xa5, 0x19, 0x16, 0xec, 0x80, 0xb4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr166, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 356, 0, 0, 0, implicit $exec :: (store (s32) into %stack.136, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa6, 0x19, 0x0f, 0x90, 0xa6, 0x19, 0x16, 0xec, 0x80, 0xb2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr167, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 352, 0, 0, 0, implicit $exec :: (store (s32) into %stack.137, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa7, 0x19, 0x0f, 0x90, 0xa7, 0x19, 0x16, 0xec, 0x80, 0xb0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr168, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 348, 0, 0, 0, implicit $exec :: (store (s32) into %stack.138, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa8, 0x19, 0x0f, 0x90, 0xa8, 0x19, 0x16, 0xec, 0x80, 0xae, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr169, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 344, 0, 0, 0, implicit $exec :: (store (s32) into %stack.139, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa9, 0x19, 0x0f, 0x90, 0xa9, 0x19, 0x16, 0xec, 0x80, 0xac, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr170, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 340, 0, 0, 0, implicit $exec :: (store (s32) into %stack.140, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaa, 0x19, 0x0f, 0x90, 0xaa, 0x19, 0x16, 0xec, 0x80, 0xaa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr171, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 336, 0, 0, 0, implicit $exec :: (store (s32) into %stack.141, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xab, 0x19, 0x0f, 0x90, 0xab, 0x19, 0x16, 0xec, 0x80, 0xa8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr172, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 332, 0, 0, 0, implicit $exec :: (store (s32) into %stack.142, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xac, 0x19, 0x0f, 0x90, 0xac, 0x19, 0x16, 0xec, 0x80, 0xa6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr173, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 328, 0, 0, 0, implicit $exec :: (store (s32) into %stack.143, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xad, 0x19, 0x0f, 0x90, 0xad, 0x19, 0x16, 0xec, 0x80, 0xa4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr174, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 324, 0, 0, 0, implicit $exec :: (store (s32) into %stack.144, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xae, 0x19, 0x0f, 0x90, 0xae, 0x19, 0x16, 0xec, 0x80, 0xa2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr175, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 320, 0, 0, 0, implicit $exec :: (store (s32) into %stack.145, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaf, 0x19, 0x0f, 0x90, 0xaf, 0x19, 0x16, 0xec, 0x80, 0xa0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr176, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 316, 0, 0, 0, implicit $exec :: (store (s32) into %stack.146, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb0, 0x19, 0x0f, 0x90, 0xb0, 0x19, 0x16, 0xec, 0x80, 0x9e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr177, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 312, 0, 0, 0, implicit $exec :: (store (s32) into %stack.147, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb1, 0x19, 0x0f, 0x90, 0xb1, 0x19, 0x16, 0xec, 0x80, 0x9c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr178, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 308, 0, 0, 0, implicit $exec :: (store (s32) into %stack.148, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb2, 0x19, 0x0f, 0x90, 0xb2, 0x19, 0x16, 0xec, 0x80, 0x9a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr179, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 304, 0, 0, 0, implicit $exec :: (store (s32) into %stack.149, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb3, 0x19, 0x0f, 0x90, 0xb3, 0x19, 0x16, 0xec, 0x80, 0x98, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr180, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 300, 0, 0, 0, implicit $exec :: (store (s32) into %stack.150, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb4, 0x19, 0x0f, 0x90, 0xb4, 0x19, 0x16, 0xec, 0x80, 0x96, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr181, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 296, 0, 0, 0, implicit $exec :: (store (s32) into %stack.151, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb5, 0x19, 0x0f, 0x90, 0xb5, 0x19, 0x16, 0xec, 0x80, 0x94, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr182, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 292, 0, 0, 0, implicit $exec :: (store (s32) into %stack.152, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb6, 0x19, 0x0f, 0x90, 0xb6, 0x19, 0x16, 0xec, 0x80, 0x92, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr183, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 288, 0, 0, 0, implicit $exec :: (store (s32) into %stack.153, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb7, 0x19, 0x0f, 0x90, 0xb7, 0x19, 0x16, 0xec, 0x80, 0x90, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr184, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 284, 0, 0, 0, implicit $exec :: (store (s32) into %stack.154, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb8, 0x19, 0x0f, 0x90, 0xb8, 0x19, 0x16, 0xec, 0x80, 0x8e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr185, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 280, 0, 0, 0, implicit $exec :: (store (s32) into %stack.155, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb9, 0x19, 0x0f, 0x90, 0xb9, 0x19, 0x16, 0xec, 0x80, 0x8c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr186, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 276, 0, 0, 0, implicit $exec :: (store (s32) into %stack.156, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xba, 0x19, 0x0f, 0x90, 0xba, 0x19, 0x16, 0xec, 0x80, 0x8a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr187, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 272, 0, 0, 0, implicit $exec :: (store (s32) into %stack.157, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbb, 0x19, 0x0f, 0x90, 0xbb, 0x19, 0x16, 0xec, 0x80, 0x88, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr188, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 268, 0, 0, 0, implicit $exec :: (store (s32) into %stack.158, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbc, 0x19, 0x0f, 0x90, 0xbc, 0x19, 0x16, 0xec, 0x80, 0x86, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr189, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 264, 0, 0, 0, implicit $exec :: (store (s32) into %stack.159, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbd, 0x19, 0x0f, 0x90, 0xbd, 0x19, 0x16, 0xec, 0x80, 0x84, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr190, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 260, 0, 0, 0, implicit $exec :: (store (s32) into %stack.160, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbe, 0x19, 0x0f, 0x90, 0xbe, 0x19, 0x16, 0xec, 0x80, 0x82, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr191, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 256, 0, 0, 0, implicit $exec :: (store (s32) into %stack.161, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbf, 0x19, 0x0f, 0x90, 0xbf, 0x19, 0x16, 0xec, 0x80, 0x80, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr192, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 252, 0, 0, 0, implicit $exec :: (store (s32) into %stack.162, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc0, 0x19, 0x0e, 0x90, 0xc0, 0x19, 0x16, 0xec, 0x80, 0x7e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr193, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 248, 0, 0, 0, implicit $exec :: (store (s32) into %stack.163, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc1, 0x19, 0x0e, 0x90, 0xc1, 0x19, 0x16, 0xec, 0x80, 0x7c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr194, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 244, 0, 0, 0, implicit $exec :: (store (s32) into %stack.164, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc2, 0x19, 0x0e, 0x90, 0xc2, 0x19, 0x16, 0xec, 0x80, 0x7a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr195, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 240, 0, 0, 0, implicit $exec :: (store (s32) into %stack.165, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc3, 0x19, 0x0e, 0x90, 0xc3, 0x19, 0x16, 0xec, 0x80, 0x78, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr196, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 236, 0, 0, 0, implicit $exec :: (store (s32) into %stack.166, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc4, 0x19, 0x0e, 0x90, 0xc4, 0x19, 0x16, 0xec, 0x80, 0x76, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr197, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 232, 0, 0, 0, implicit $exec :: (store (s32) into %stack.167, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc5, 0x19, 0x0e, 0x90, 0xc5, 0x19, 0x16, 0xec, 0x80, 0x74, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr198, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 228, 0, 0, 0, implicit $exec :: (store (s32) into %stack.168, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc6, 0x19, 0x0e, 0x90, 0xc6, 0x19, 0x16, 0xec, 0x80, 0x72, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr199, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 224, 0, 0, 0, implicit $exec :: (store (s32) into %stack.169, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc7, 0x19, 0x0e, 0x90, 0xc7, 0x19, 0x16, 0xec, 0x80, 0x70, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr200, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 220, 0, 0, 0, implicit $exec :: (store (s32) into %stack.170, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc8, 0x19, 0x0e, 0x90, 0xc8, 0x19, 0x16, 0xec, 0x80, 0x6e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr201, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 216, 0, 0, 0, implicit $exec :: (store (s32) into %stack.171, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc9, 0x19, 0x0e, 0x90, 0xc9, 0x19, 0x16, 0xec, 0x80, 0x6c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr202, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 212, 0, 0, 0, implicit $exec :: (store (s32) into %stack.172, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xca, 0x19, 0x0e, 0x90, 0xca, 0x19, 0x16, 0xec, 0x80, 0x6a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr203, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 208, 0, 0, 0, implicit $exec :: (store (s32) into %stack.173, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcb, 0x19, 0x0e, 0x90, 0xcb, 0x19, 0x16, 0xec, 0x80, 0x68, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr204, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 204, 0, 0, 0, implicit $exec :: (store (s32) into %stack.174, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcc, 0x19, 0x0e, 0x90, 0xcc, 0x19, 0x16, 0xec, 0x80, 0x66, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr205, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 200, 0, 0, 0, implicit $exec :: (store (s32) into %stack.175, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcd, 0x19, 0x0e, 0x90, 0xcd, 0x19, 0x16, 0xec, 0x80, 0x64, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr206, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 196, 0, 0, 0, implicit $exec :: (store (s32) into %stack.176, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xce, 0x19, 0x0e, 0x90, 0xce, 0x19, 0x16, 0xec, 0x80, 0x62, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr207, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 192, 0, 0, 0, implicit $exec :: (store (s32) into %stack.177, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcf, 0x19, 0x0e, 0x90, 0xcf, 0x19, 0x16, 0xec, 0x80, 0x60, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr208, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 188, 0, 0, 0, implicit $exec :: (store (s32) into %stack.178, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x19, 0x0e, 0x90, 0xd0, 0x19, 0x16, 0xec, 0x80, 0x5e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr209, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 184, 0, 0, 0, implicit $exec :: (store (s32) into %stack.179, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x19, 0x0e, 0x90, 0xd1, 0x19, 0x16, 0xec, 0x80, 0x5c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr210, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 180, 0, 0, 0, implicit $exec :: (store (s32) into %stack.180, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x19, 0x0e, 0x90, 0xd2, 0x19, 0x16, 0xec, 0x80, 0x5a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr211, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 176, 0, 0, 0, implicit $exec :: (store (s32) into %stack.181, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x19, 0x0e, 0x90, 0xd3, 0x19, 0x16, 0xec, 0x80, 0x58, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr212, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 172, 0, 0, 0, implicit $exec :: (store (s32) into %stack.182, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x19, 0x0e, 0x90, 0xd4, 0x19, 0x16, 0xec, 0x80, 0x56, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr213, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 168, 0, 0, 0, implicit $exec :: (store (s32) into %stack.183, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x19, 0x0e, 0x90, 0xd5, 0x19, 0x16, 0xec, 0x80, 0x54, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr214, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 164, 0, 0, 0, implicit $exec :: (store (s32) into %stack.184, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x19, 0x0e, 0x90, 0xd6, 0x19, 0x16, 0xec, 0x80, 0x52, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr215, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 160, 0, 0, 0, implicit $exec :: (store (s32) into %stack.185, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x19, 0x0e, 0x90, 0xd7, 0x19, 0x16, 0xec, 0x80, 0x50, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr216, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 156, 0, 0, 0, implicit $exec :: (store (s32) into %stack.186, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x19, 0x0e, 0x90, 0xd8, 0x19, 0x16, 0xec, 0x80, 0x4e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr217, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 152, 0, 0, 0, implicit $exec :: (store (s32) into %stack.187, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x19, 0x0e, 0x90, 0xd9, 0x19, 0x16, 0xec, 0x80, 0x4c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr218, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 148, 0, 0, 0, implicit $exec :: (store (s32) into %stack.188, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x19, 0x0e, 0x90, 0xda, 0x19, 0x16, 0xec, 0x80, 0x4a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr219, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 144, 0, 0, 0, implicit $exec :: (store (s32) into %stack.189, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x19, 0x0e, 0x90, 0xdb, 0x19, 0x16, 0xec, 0x80, 0x48, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr220, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 140, 0, 0, 0, implicit $exec :: (store (s32) into %stack.190, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x19, 0x0e, 0x90, 0xdc, 0x19, 0x16, 0xec, 0x80, 0x46, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr221, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 136, 0, 0, 0, implicit $exec :: (store (s32) into %stack.191, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x19, 0x0e, 0x90, 0xdd, 0x19, 0x16, 0xec, 0x80, 0x44, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr222, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 132, 0, 0, 0, implicit $exec :: (store (s32) into %stack.192, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x19, 0x0e, 0x90, 0xde, 0x19, 0x16, 0xec, 0x80, 0x42, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr223, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 128, 0, 0, 0, implicit $exec :: (store (s32) into %stack.193, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x19, 0x0e, 0x90, 0xdf, 0x19, 0x16, 0xec, 0x80, 0x40, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr224, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 124, 0, 0, 0, implicit $exec :: (store (s32) into %stack.194, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x19, 0x0e, 0x90, 0xe0, 0x19, 0x16, 0xec, 0x80, 0x3e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr225, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 120, 0, 0, 0, implicit $exec :: (store (s32) into %stack.195, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x19, 0x0e, 0x90, 0xe1, 0x19, 0x16, 0xec, 0x80, 0x3c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr226, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 116, 0, 0, 0, implicit $exec :: (store (s32) into %stack.196, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x19, 0x0e, 0x90, 0xe2, 0x19, 0x16, 0xec, 0x80, 0x3a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr227, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 112, 0, 0, 0, implicit $exec :: (store (s32) into %stack.197, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x19, 0x0e, 0x90, 0xe3, 0x19, 0x16, 0xec, 0x80, 0x38, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr228, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 108, 0, 0, 0, implicit $exec :: (store (s32) into %stack.198, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x19, 0x0e, 0x90, 0xe4, 0x19, 0x16, 0xec, 0x80, 0x36, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr229, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 104, 0, 0, 0, implicit $exec :: (store (s32) into %stack.199, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x19, 0x0e, 0x90, 0xe5, 0x19, 0x16, 0xec, 0x80, 0x34, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr230, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 100, 0, 0, 0, implicit $exec :: (store (s32) into %stack.200, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x19, 0x0e, 0x90, 0xe6, 0x19, 0x16, 0xec, 0x80, 0x32, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr231, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 96, 0, 0, 0, implicit $exec :: (store (s32) into %stack.201, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x19, 0x0e, 0x90, 0xe7, 0x19, 0x16, 0xec, 0x80, 0x30, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr232, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 92, 0, 0, 0, implicit $exec :: (store (s32) into %stack.202, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x19, 0x0e, 0x90, 0xe8, 0x19, 0x16, 0xec, 0x80, 0x2e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr233, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 88, 0, 0, 0, implicit $exec :: (store (s32) into %stack.203, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x19, 0x0e, 0x90, 0xe9, 0x19, 0x16, 0xec, 0x80, 0x2c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr234, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 84, 0, 0, 0, implicit $exec :: (store (s32) into %stack.204, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x19, 0x0e, 0x90, 0xea, 0x19, 0x16, 0xec, 0x80, 0x2a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr235, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 80, 0, 0, 0, implicit $exec :: (store (s32) into %stack.205, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x19, 0x0e, 0x90, 0xeb, 0x19, 0x16, 0xec, 0x80, 0x28, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr236, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 76, 0, 0, 0, implicit $exec :: (store (s32) into %stack.206, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x19, 0x0e, 0x90, 0xec, 0x19, 0x16, 0xec, 0x80, 0x26, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr237, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 72, 0, 0, 0, implicit $exec :: (store (s32) into %stack.207, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x19, 0x0e, 0x90, 0xed, 0x19, 0x16, 0xec, 0x80, 0x24, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr238, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 68, 0, 0, 0, implicit $exec :: (store (s32) into %stack.208, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x19, 0x0e, 0x90, 0xee, 0x19, 0x16, 0xec, 0x80, 0x22, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr239, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 64, 0, 0, 0, implicit $exec :: (store (s32) into %stack.209, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x19, 0x0e, 0x90, 0xef, 0x19, 0x16, 0xec, 0x80, 0x20, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr240, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec :: (store (s32) into %stack.210, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x19, 0x0e, 0x90, 0xf0, 0x19, 0x16, 0xec, 0x80, 0x1e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr241, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec :: (store (s32) into %stack.211, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x19, 0x0e, 0x90, 0xf1, 0x19, 0x16, 0xec, 0x80, 0x1c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr242, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec :: (store (s32) into %stack.212, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x19, 0x0e, 0x90, 0xf2, 0x19, 0x16, 0xec, 0x80, 0x1a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr243, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec :: (store (s32) into %stack.213, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x19, 0x0e, 0x90, 0xf3, 0x19, 0x16, 0xec, 0x80, 0x18, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr244, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec :: (store (s32) into %stack.214, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x19, 0x0e, 0x90, 0xf4, 0x19, 0x16, 0xec, 0x80, 0x16, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr245, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec :: (store (s32) into %stack.215, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x19, 0x0e, 0x90, 0xf5, 0x19, 0x16, 0xec, 0x80, 0x14, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr246, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec :: (store (s32) into %stack.216, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x19, 0x0e, 0x90, 0xf6, 0x19, 0x16, 0xec, 0x80, 0x12, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr247, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.217, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x19, 0x0e, 0x90, 0xf7, 0x19, 0x16, 0xec, 0x80, 0x10, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr248, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (store (s32) into %stack.218, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x19, 0x0e, 0x90, 0xf8, 0x19, 0x16, 0xec, 0x80, 0x0e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr249, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (store (s32) into %stack.219, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x19, 0x0e, 0x90, 0xf9, 0x19, 0x16, 0xec, 0x80, 0x0c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr250, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (store (s32) into %stack.220, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x19, 0x0e, 0x90, 0xfa, 0x19, 0x16, 0xec, 0x80, 0x0a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr251, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (store (s32) into %stack.221, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x19, 0x0e, 0x90, 0xfb, 0x19, 0x16, 0xec, 0x80, 0x08, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr252, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (store (s32) into %stack.222, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x19, 0x0e, 0x90, 0xfc, 0x19, 0x16, 0xec, 0x80, 0x06, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr253, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (store (s32) into %stack.223, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x19, 0x0e, 0x90, 0xfd, 0x19, 0x16, 0xec, 0x80, 0x04, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr254, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.224, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x19, 0x0e, 0x90, 0xfe, 0x19, 0x16, 0xec, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr255, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.225, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x19, 0x0d, 0x90, 0xff, 0x19, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: S_CMP_EQ_U32 0, 0, implicit-def $scc ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 704, 0, 0, 0, implicit $exec :: (store (s32) into %stack.226, addrspace 5) ; GFX90A-NEXT: $vgpr0 = V_MOV_B32_e32 8904, implicit $exec @@ -4969,229 +6089,453 @@ ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr28_lo16 ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr29_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr32, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr32_lo16, $vgpr0_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr1 = V_ACCVGPR_READ_B32_e64 killed $agpr33, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr33_lo16, $vgpr1_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr2 = V_ACCVGPR_READ_B32_e64 killed $agpr34, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr34_lo16, $vgpr2_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr3 = V_ACCVGPR_READ_B32_e64 killed $agpr35, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr35_lo16, $vgpr3_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr4 = V_ACCVGPR_READ_B32_e64 killed $agpr36, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr36_lo16, $vgpr4_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr5 = V_ACCVGPR_READ_B32_e64 killed $agpr37, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr37_lo16, $vgpr5_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 killed $agpr38, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr38_lo16, $vgpr6_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr7 = V_ACCVGPR_READ_B32_e64 killed $agpr39, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr39_lo16, $vgpr7_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr8 = V_ACCVGPR_READ_B32_e64 killed $agpr40, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr40_lo16, $vgpr8_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr9 = V_ACCVGPR_READ_B32_e64 killed $agpr41, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr41_lo16, $vgpr9_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr10 = V_ACCVGPR_READ_B32_e64 killed $agpr42, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr42_lo16, $vgpr10_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr11 = V_ACCVGPR_READ_B32_e64 killed $agpr43, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr43_lo16, $vgpr11_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr12 = V_ACCVGPR_READ_B32_e64 killed $agpr44, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr44_lo16, $vgpr12_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr13 = V_ACCVGPR_READ_B32_e64 killed $agpr45, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr45_lo16, $vgpr13_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr14 = V_ACCVGPR_READ_B32_e64 killed $agpr46, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr46_lo16, $vgpr14_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr15 = V_ACCVGPR_READ_B32_e64 killed $agpr47, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr47_lo16, $vgpr15_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr16 = V_ACCVGPR_READ_B32_e64 killed $agpr48, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr48_lo16, $vgpr16_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr17 = V_ACCVGPR_READ_B32_e64 killed $agpr49, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr49_lo16, $vgpr17_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr18 = V_ACCVGPR_READ_B32_e64 killed $agpr50, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr50_lo16, $vgpr18_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr19 = V_ACCVGPR_READ_B32_e64 killed $agpr51, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr51_lo16, $vgpr19_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr20 = V_ACCVGPR_READ_B32_e64 killed $agpr52, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr52_lo16, $vgpr20_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr21 = V_ACCVGPR_READ_B32_e64 killed $agpr53, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr53_lo16, $vgpr21_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr22 = V_ACCVGPR_READ_B32_e64 killed $agpr54, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr54_lo16, $vgpr22_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr23 = V_ACCVGPR_READ_B32_e64 killed $agpr55, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr55_lo16, $vgpr23_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr24 = V_ACCVGPR_READ_B32_e64 killed $agpr56, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr56_lo16, $vgpr24_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr25 = V_ACCVGPR_READ_B32_e64 killed $agpr57, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr57_lo16, $vgpr25_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr26 = V_ACCVGPR_READ_B32_e64 killed $agpr58, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr58_lo16, $vgpr26_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr27 = V_ACCVGPR_READ_B32_e64 killed $agpr59, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr59_lo16, $vgpr27_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr28 = V_ACCVGPR_READ_B32_e64 killed $agpr60, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr60_lo16, $vgpr28_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr29 = V_ACCVGPR_READ_B32_e64 killed $agpr61, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr61_lo16, $vgpr29_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr30 = V_ACCVGPR_READ_B32_e64 killed $agpr62, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr62_lo16, $vgpr30_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 killed $agpr63, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr63_lo16, $vgpr31_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr64, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr64_lo16, $vgpr32_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 killed $agpr65, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr65_lo16, $vgpr33_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr34 = V_ACCVGPR_READ_B32_e64 killed $agpr66, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr66_lo16, $vgpr34_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr35 = V_ACCVGPR_READ_B32_e64 killed $agpr67, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr67_lo16, $vgpr35_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr36 = V_ACCVGPR_READ_B32_e64 killed $agpr68, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr68_lo16, $vgpr36_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr37 = V_ACCVGPR_READ_B32_e64 killed $agpr69, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr69_lo16, $vgpr37_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr38 = V_ACCVGPR_READ_B32_e64 killed $agpr70, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr70_lo16, $vgpr38_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr39 = V_ACCVGPR_READ_B32_e64 killed $agpr71, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr71_lo16, $vgpr39_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr48 = V_ACCVGPR_READ_B32_e64 killed $agpr72, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr72_lo16, $vgpr48_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr49 = V_ACCVGPR_READ_B32_e64 killed $agpr73, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr73_lo16, $vgpr49_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr50 = V_ACCVGPR_READ_B32_e64 killed $agpr74, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr74_lo16, $vgpr50_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr51 = V_ACCVGPR_READ_B32_e64 killed $agpr75, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr75_lo16, $vgpr51_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr52 = V_ACCVGPR_READ_B32_e64 killed $agpr76, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr76_lo16, $vgpr52_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr53 = V_ACCVGPR_READ_B32_e64 killed $agpr77, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr77_lo16, $vgpr53_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr54 = V_ACCVGPR_READ_B32_e64 killed $agpr78, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr78_lo16, $vgpr54_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 killed $agpr79, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr79_lo16, $vgpr55_lo16 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr80, $sgpr32, 700, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.50, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x18, 0x0f, 0x90, 0xd0, 0x18, 0x16, 0xec, 0x80, 0xde, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr81, $sgpr32, 696, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.51, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x18, 0x0f, 0x90, 0xd1, 0x18, 0x16, 0xec, 0x80, 0xdc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr82, $sgpr32, 692, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.52, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x18, 0x0f, 0x90, 0xd2, 0x18, 0x16, 0xec, 0x80, 0xda, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr83, $sgpr32, 688, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.53, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x18, 0x0f, 0x90, 0xd3, 0x18, 0x16, 0xec, 0x80, 0xd8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr84, $sgpr32, 684, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.54, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x18, 0x0f, 0x90, 0xd4, 0x18, 0x16, 0xec, 0x80, 0xd6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr85, $sgpr32, 680, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.55, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x18, 0x0f, 0x90, 0xd5, 0x18, 0x16, 0xec, 0x80, 0xd4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr86, $sgpr32, 676, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.56, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x18, 0x0f, 0x90, 0xd6, 0x18, 0x16, 0xec, 0x80, 0xd2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr87, $sgpr32, 672, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.57, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x18, 0x0f, 0x90, 0xd7, 0x18, 0x16, 0xec, 0x80, 0xd0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr88, $sgpr32, 668, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.58, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x18, 0x0f, 0x90, 0xd8, 0x18, 0x16, 0xec, 0x80, 0xce, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr89, $sgpr32, 664, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.59, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x18, 0x0f, 0x90, 0xd9, 0x18, 0x16, 0xec, 0x80, 0xcc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr90, $sgpr32, 660, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.60, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x18, 0x0f, 0x90, 0xda, 0x18, 0x16, 0xec, 0x80, 0xca, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr91, $sgpr32, 656, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.61, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x18, 0x0f, 0x90, 0xdb, 0x18, 0x16, 0xec, 0x80, 0xc8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr92, $sgpr32, 652, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.62, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x18, 0x0f, 0x90, 0xdc, 0x18, 0x16, 0xec, 0x80, 0xc6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr93, $sgpr32, 648, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.63, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x18, 0x0f, 0x90, 0xdd, 0x18, 0x16, 0xec, 0x80, 0xc4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr94, $sgpr32, 644, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.64, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x18, 0x0f, 0x90, 0xde, 0x18, 0x16, 0xec, 0x80, 0xc2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr95, $sgpr32, 640, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.65, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x18, 0x0f, 0x90, 0xdf, 0x18, 0x16, 0xec, 0x80, 0xc0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr96, $sgpr32, 636, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.66, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x18, 0x0f, 0x90, 0xe0, 0x18, 0x16, 0xec, 0x80, 0xbe, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr97, $sgpr32, 632, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.67, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x18, 0x0f, 0x90, 0xe1, 0x18, 0x16, 0xec, 0x80, 0xbc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr98, $sgpr32, 628, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.68, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x18, 0x0f, 0x90, 0xe2, 0x18, 0x16, 0xec, 0x80, 0xba, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr99, $sgpr32, 624, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.69, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x18, 0x0f, 0x90, 0xe3, 0x18, 0x16, 0xec, 0x80, 0xb8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr100, $sgpr32, 620, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.70, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x18, 0x0f, 0x90, 0xe4, 0x18, 0x16, 0xec, 0x80, 0xb6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr101, $sgpr32, 616, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.71, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x18, 0x0f, 0x90, 0xe5, 0x18, 0x16, 0xec, 0x80, 0xb4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr102, $sgpr32, 612, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.72, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x18, 0x0f, 0x90, 0xe6, 0x18, 0x16, 0xec, 0x80, 0xb2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr103, $sgpr32, 608, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.73, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x18, 0x0f, 0x90, 0xe7, 0x18, 0x16, 0xec, 0x80, 0xb0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr104, $sgpr32, 604, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.74, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x18, 0x0f, 0x90, 0xe8, 0x18, 0x16, 0xec, 0x80, 0xae, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr105, $sgpr32, 600, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.75, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x18, 0x0f, 0x90, 0xe9, 0x18, 0x16, 0xec, 0x80, 0xac, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr106, $sgpr32, 596, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.76, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x18, 0x0f, 0x90, 0xea, 0x18, 0x16, 0xec, 0x80, 0xaa, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr107, $sgpr32, 592, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.77, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x18, 0x0f, 0x90, 0xeb, 0x18, 0x16, 0xec, 0x80, 0xa8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr108, $sgpr32, 588, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.78, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x18, 0x0f, 0x90, 0xec, 0x18, 0x16, 0xec, 0x80, 0xa6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr109, $sgpr32, 584, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.79, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x18, 0x0f, 0x90, 0xed, 0x18, 0x16, 0xec, 0x80, 0xa4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr110, $sgpr32, 580, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.80, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x18, 0x0f, 0x90, 0xee, 0x18, 0x16, 0xec, 0x80, 0xa2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr111, $sgpr32, 576, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.81, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x18, 0x0f, 0x90, 0xef, 0x18, 0x16, 0xec, 0x80, 0xa0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr112, $sgpr32, 572, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.82, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x18, 0x0f, 0x90, 0xf0, 0x18, 0x16, 0xec, 0x80, 0x9e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr113, $sgpr32, 568, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.83, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x18, 0x0f, 0x90, 0xf1, 0x18, 0x16, 0xec, 0x80, 0x9c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr114, $sgpr32, 564, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.84, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x18, 0x0f, 0x90, 0xf2, 0x18, 0x16, 0xec, 0x80, 0x9a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr115, $sgpr32, 560, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.85, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x18, 0x0f, 0x90, 0xf3, 0x18, 0x16, 0xec, 0x80, 0x98, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr116, $sgpr32, 556, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.86, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x18, 0x0f, 0x90, 0xf4, 0x18, 0x16, 0xec, 0x80, 0x96, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr117, $sgpr32, 552, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.87, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x18, 0x0f, 0x90, 0xf5, 0x18, 0x16, 0xec, 0x80, 0x94, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr118, $sgpr32, 548, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.88, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x18, 0x0f, 0x90, 0xf6, 0x18, 0x16, 0xec, 0x80, 0x92, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr119, $sgpr32, 544, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.89, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x18, 0x0f, 0x90, 0xf7, 0x18, 0x16, 0xec, 0x80, 0x90, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr120, $sgpr32, 540, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.90, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x18, 0x0f, 0x90, 0xf8, 0x18, 0x16, 0xec, 0x80, 0x8e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr121, $sgpr32, 536, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.91, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x18, 0x0f, 0x90, 0xf9, 0x18, 0x16, 0xec, 0x80, 0x8c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr122, $sgpr32, 532, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.92, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x18, 0x0f, 0x90, 0xfa, 0x18, 0x16, 0xec, 0x80, 0x8a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr123, $sgpr32, 528, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.93, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x18, 0x0f, 0x90, 0xfb, 0x18, 0x16, 0xec, 0x80, 0x88, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr124, $sgpr32, 524, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.94, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x18, 0x0f, 0x90, 0xfc, 0x18, 0x16, 0xec, 0x80, 0x86, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr125, $sgpr32, 520, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.95, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x18, 0x0f, 0x90, 0xfd, 0x18, 0x16, 0xec, 0x80, 0x84, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr126, $sgpr32, 516, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.96, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x18, 0x0f, 0x90, 0xfe, 0x18, 0x16, 0xec, 0x80, 0x82, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr127, $sgpr32, 512, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.97, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x18, 0x0f, 0x90, 0xff, 0x18, 0x16, 0xec, 0x80, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr128, $sgpr32, 508, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.98, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x80, 0x19, 0x0f, 0x90, 0x80, 0x19, 0x16, 0xec, 0x80, 0xfe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr129, $sgpr32, 504, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.99, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x81, 0x19, 0x0f, 0x90, 0x81, 0x19, 0x16, 0xec, 0x80, 0xfc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr130, $sgpr32, 500, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.100, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x82, 0x19, 0x0f, 0x90, 0x82, 0x19, 0x16, 0xec, 0x80, 0xfa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr131, $sgpr32, 496, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.101, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x83, 0x19, 0x0f, 0x90, 0x83, 0x19, 0x16, 0xec, 0x80, 0xf8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr132, $sgpr32, 492, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.102, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x84, 0x19, 0x0f, 0x90, 0x84, 0x19, 0x16, 0xec, 0x80, 0xf6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr133, $sgpr32, 488, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.103, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x85, 0x19, 0x0f, 0x90, 0x85, 0x19, 0x16, 0xec, 0x80, 0xf4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr134, $sgpr32, 484, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.104, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x86, 0x19, 0x0f, 0x90, 0x86, 0x19, 0x16, 0xec, 0x80, 0xf2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr135, $sgpr32, 480, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.105, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x87, 0x19, 0x0f, 0x90, 0x87, 0x19, 0x16, 0xec, 0x80, 0xf0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr136, $sgpr32, 476, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.106, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x88, 0x19, 0x0f, 0x90, 0x88, 0x19, 0x16, 0xec, 0x80, 0xee, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr137, $sgpr32, 472, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.107, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x89, 0x19, 0x0f, 0x90, 0x89, 0x19, 0x16, 0xec, 0x80, 0xec, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr138, $sgpr32, 468, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.108, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8a, 0x19, 0x0f, 0x90, 0x8a, 0x19, 0x16, 0xec, 0x80, 0xea, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr139, $sgpr32, 464, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.109, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8b, 0x19, 0x0f, 0x90, 0x8b, 0x19, 0x16, 0xec, 0x80, 0xe8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr140, $sgpr32, 460, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.110, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8c, 0x19, 0x0f, 0x90, 0x8c, 0x19, 0x16, 0xec, 0x80, 0xe6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr141, $sgpr32, 456, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.111, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8d, 0x19, 0x0f, 0x90, 0x8d, 0x19, 0x16, 0xec, 0x80, 0xe4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr142, $sgpr32, 452, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.112, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8e, 0x19, 0x0f, 0x90, 0x8e, 0x19, 0x16, 0xec, 0x80, 0xe2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr143, $sgpr32, 448, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.113, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8f, 0x19, 0x0f, 0x90, 0x8f, 0x19, 0x16, 0xec, 0x80, 0xe0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr144, $sgpr32, 444, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.114, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x90, 0x19, 0x0f, 0x90, 0x90, 0x19, 0x16, 0xec, 0x80, 0xde, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr145, $sgpr32, 440, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.115, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x91, 0x19, 0x0f, 0x90, 0x91, 0x19, 0x16, 0xec, 0x80, 0xdc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr146, $sgpr32, 436, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.116, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x92, 0x19, 0x0f, 0x90, 0x92, 0x19, 0x16, 0xec, 0x80, 0xda, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr147, $sgpr32, 432, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.117, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x93, 0x19, 0x0f, 0x90, 0x93, 0x19, 0x16, 0xec, 0x80, 0xd8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr148, $sgpr32, 428, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.118, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x94, 0x19, 0x0f, 0x90, 0x94, 0x19, 0x16, 0xec, 0x80, 0xd6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr149, $sgpr32, 424, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.119, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x95, 0x19, 0x0f, 0x90, 0x95, 0x19, 0x16, 0xec, 0x80, 0xd4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr150, $sgpr32, 420, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.120, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x96, 0x19, 0x0f, 0x90, 0x96, 0x19, 0x16, 0xec, 0x80, 0xd2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr151, $sgpr32, 416, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.121, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x97, 0x19, 0x0f, 0x90, 0x97, 0x19, 0x16, 0xec, 0x80, 0xd0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr152, $sgpr32, 412, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.122, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x98, 0x19, 0x0f, 0x90, 0x98, 0x19, 0x16, 0xec, 0x80, 0xce, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr153, $sgpr32, 408, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.123, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x99, 0x19, 0x0f, 0x90, 0x99, 0x19, 0x16, 0xec, 0x80, 0xcc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr154, $sgpr32, 404, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.124, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9a, 0x19, 0x0f, 0x90, 0x9a, 0x19, 0x16, 0xec, 0x80, 0xca, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr155, $sgpr32, 400, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.125, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9b, 0x19, 0x0f, 0x90, 0x9b, 0x19, 0x16, 0xec, 0x80, 0xc8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr156, $sgpr32, 396, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.126, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9c, 0x19, 0x0f, 0x90, 0x9c, 0x19, 0x16, 0xec, 0x80, 0xc6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr157, $sgpr32, 392, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.127, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9d, 0x19, 0x0f, 0x90, 0x9d, 0x19, 0x16, 0xec, 0x80, 0xc4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr158, $sgpr32, 388, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.128, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9e, 0x19, 0x0f, 0x90, 0x9e, 0x19, 0x16, 0xec, 0x80, 0xc2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr159, $sgpr32, 384, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.129, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9f, 0x19, 0x0f, 0x90, 0x9f, 0x19, 0x16, 0xec, 0x80, 0xc0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr160, $sgpr32, 380, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.130, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa0, 0x19, 0x0f, 0x90, 0xa0, 0x19, 0x16, 0xec, 0x80, 0xbe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr161, $sgpr32, 376, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.131, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa1, 0x19, 0x0f, 0x90, 0xa1, 0x19, 0x16, 0xec, 0x80, 0xbc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr162, $sgpr32, 372, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.132, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa2, 0x19, 0x0f, 0x90, 0xa2, 0x19, 0x16, 0xec, 0x80, 0xba, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr163, $sgpr32, 368, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.133, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa3, 0x19, 0x0f, 0x90, 0xa3, 0x19, 0x16, 0xec, 0x80, 0xb8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr164, $sgpr32, 364, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.134, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa4, 0x19, 0x0f, 0x90, 0xa4, 0x19, 0x16, 0xec, 0x80, 0xb6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr165, $sgpr32, 360, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.135, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa5, 0x19, 0x0f, 0x90, 0xa5, 0x19, 0x16, 0xec, 0x80, 0xb4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr166, $sgpr32, 356, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.136, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa6, 0x19, 0x0f, 0x90, 0xa6, 0x19, 0x16, 0xec, 0x80, 0xb2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr167, $sgpr32, 352, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.137, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa7, 0x19, 0x0f, 0x90, 0xa7, 0x19, 0x16, 0xec, 0x80, 0xb0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr168, $sgpr32, 348, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.138, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa8, 0x19, 0x0f, 0x90, 0xa8, 0x19, 0x16, 0xec, 0x80, 0xae, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr169, $sgpr32, 344, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.139, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa9, 0x19, 0x0f, 0x90, 0xa9, 0x19, 0x16, 0xec, 0x80, 0xac, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr170, $sgpr32, 340, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.140, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaa, 0x19, 0x0f, 0x90, 0xaa, 0x19, 0x16, 0xec, 0x80, 0xaa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr171, $sgpr32, 336, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.141, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xab, 0x19, 0x0f, 0x90, 0xab, 0x19, 0x16, 0xec, 0x80, 0xa8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr172, $sgpr32, 332, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.142, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xac, 0x19, 0x0f, 0x90, 0xac, 0x19, 0x16, 0xec, 0x80, 0xa6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr173, $sgpr32, 328, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.143, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xad, 0x19, 0x0f, 0x90, 0xad, 0x19, 0x16, 0xec, 0x80, 0xa4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr174, $sgpr32, 324, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.144, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xae, 0x19, 0x0f, 0x90, 0xae, 0x19, 0x16, 0xec, 0x80, 0xa2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr175, $sgpr32, 320, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.145, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaf, 0x19, 0x0f, 0x90, 0xaf, 0x19, 0x16, 0xec, 0x80, 0xa0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr176, $sgpr32, 316, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.146, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb0, 0x19, 0x0f, 0x90, 0xb0, 0x19, 0x16, 0xec, 0x80, 0x9e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr177, $sgpr32, 312, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.147, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb1, 0x19, 0x0f, 0x90, 0xb1, 0x19, 0x16, 0xec, 0x80, 0x9c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr178, $sgpr32, 308, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.148, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb2, 0x19, 0x0f, 0x90, 0xb2, 0x19, 0x16, 0xec, 0x80, 0x9a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr179, $sgpr32, 304, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.149, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb3, 0x19, 0x0f, 0x90, 0xb3, 0x19, 0x16, 0xec, 0x80, 0x98, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr180, $sgpr32, 300, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.150, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb4, 0x19, 0x0f, 0x90, 0xb4, 0x19, 0x16, 0xec, 0x80, 0x96, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr181, $sgpr32, 296, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.151, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb5, 0x19, 0x0f, 0x90, 0xb5, 0x19, 0x16, 0xec, 0x80, 0x94, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr182, $sgpr32, 292, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.152, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb6, 0x19, 0x0f, 0x90, 0xb6, 0x19, 0x16, 0xec, 0x80, 0x92, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr183, $sgpr32, 288, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.153, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb7, 0x19, 0x0f, 0x90, 0xb7, 0x19, 0x16, 0xec, 0x80, 0x90, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr184, $sgpr32, 284, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.154, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb8, 0x19, 0x0f, 0x90, 0xb8, 0x19, 0x16, 0xec, 0x80, 0x8e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr185, $sgpr32, 280, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.155, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb9, 0x19, 0x0f, 0x90, 0xb9, 0x19, 0x16, 0xec, 0x80, 0x8c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr186, $sgpr32, 276, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.156, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xba, 0x19, 0x0f, 0x90, 0xba, 0x19, 0x16, 0xec, 0x80, 0x8a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr187, $sgpr32, 272, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.157, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbb, 0x19, 0x0f, 0x90, 0xbb, 0x19, 0x16, 0xec, 0x80, 0x88, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr188, $sgpr32, 268, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.158, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbc, 0x19, 0x0f, 0x90, 0xbc, 0x19, 0x16, 0xec, 0x80, 0x86, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr189, $sgpr32, 264, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.159, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbd, 0x19, 0x0f, 0x90, 0xbd, 0x19, 0x16, 0xec, 0x80, 0x84, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr190, $sgpr32, 260, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.160, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbe, 0x19, 0x0f, 0x90, 0xbe, 0x19, 0x16, 0xec, 0x80, 0x82, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr191, $sgpr32, 256, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.161, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbf, 0x19, 0x0f, 0x90, 0xbf, 0x19, 0x16, 0xec, 0x80, 0x80, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr192, $sgpr32, 252, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.162, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc0, 0x19, 0x0e, 0x90, 0xc0, 0x19, 0x16, 0xec, 0x80, 0x7e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr193, $sgpr32, 248, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.163, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc1, 0x19, 0x0e, 0x90, 0xc1, 0x19, 0x16, 0xec, 0x80, 0x7c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr194, $sgpr32, 244, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.164, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc2, 0x19, 0x0e, 0x90, 0xc2, 0x19, 0x16, 0xec, 0x80, 0x7a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr195, $sgpr32, 240, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.165, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc3, 0x19, 0x0e, 0x90, 0xc3, 0x19, 0x16, 0xec, 0x80, 0x78, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr196, $sgpr32, 236, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.166, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc4, 0x19, 0x0e, 0x90, 0xc4, 0x19, 0x16, 0xec, 0x80, 0x76, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr197, $sgpr32, 232, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.167, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc5, 0x19, 0x0e, 0x90, 0xc5, 0x19, 0x16, 0xec, 0x80, 0x74, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr198, $sgpr32, 228, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.168, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc6, 0x19, 0x0e, 0x90, 0xc6, 0x19, 0x16, 0xec, 0x80, 0x72, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr199, $sgpr32, 224, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.169, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc7, 0x19, 0x0e, 0x90, 0xc7, 0x19, 0x16, 0xec, 0x80, 0x70, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr200, $sgpr32, 220, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.170, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc8, 0x19, 0x0e, 0x90, 0xc8, 0x19, 0x16, 0xec, 0x80, 0x6e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr201, $sgpr32, 216, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.171, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc9, 0x19, 0x0e, 0x90, 0xc9, 0x19, 0x16, 0xec, 0x80, 0x6c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr202, $sgpr32, 212, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.172, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xca, 0x19, 0x0e, 0x90, 0xca, 0x19, 0x16, 0xec, 0x80, 0x6a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr203, $sgpr32, 208, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.173, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcb, 0x19, 0x0e, 0x90, 0xcb, 0x19, 0x16, 0xec, 0x80, 0x68, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr204, $sgpr32, 204, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.174, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcc, 0x19, 0x0e, 0x90, 0xcc, 0x19, 0x16, 0xec, 0x80, 0x66, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr205, $sgpr32, 200, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.175, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcd, 0x19, 0x0e, 0x90, 0xcd, 0x19, 0x16, 0xec, 0x80, 0x64, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr206, $sgpr32, 196, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.176, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xce, 0x19, 0x0e, 0x90, 0xce, 0x19, 0x16, 0xec, 0x80, 0x62, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr207, $sgpr32, 192, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.177, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcf, 0x19, 0x0e, 0x90, 0xcf, 0x19, 0x16, 0xec, 0x80, 0x60, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr208, $sgpr32, 188, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.178, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x19, 0x0e, 0x90, 0xd0, 0x19, 0x16, 0xec, 0x80, 0x5e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr209, $sgpr32, 184, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.179, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x19, 0x0e, 0x90, 0xd1, 0x19, 0x16, 0xec, 0x80, 0x5c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr210, $sgpr32, 180, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.180, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x19, 0x0e, 0x90, 0xd2, 0x19, 0x16, 0xec, 0x80, 0x5a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr211, $sgpr32, 176, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.181, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x19, 0x0e, 0x90, 0xd3, 0x19, 0x16, 0xec, 0x80, 0x58, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr212, $sgpr32, 172, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.182, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x19, 0x0e, 0x90, 0xd4, 0x19, 0x16, 0xec, 0x80, 0x56, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr213, $sgpr32, 168, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.183, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x19, 0x0e, 0x90, 0xd5, 0x19, 0x16, 0xec, 0x80, 0x54, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr214, $sgpr32, 164, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.184, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x19, 0x0e, 0x90, 0xd6, 0x19, 0x16, 0xec, 0x80, 0x52, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr215, $sgpr32, 160, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.185, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x19, 0x0e, 0x90, 0xd7, 0x19, 0x16, 0xec, 0x80, 0x50, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr216, $sgpr32, 156, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.186, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x19, 0x0e, 0x90, 0xd8, 0x19, 0x16, 0xec, 0x80, 0x4e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr217, $sgpr32, 152, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.187, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x19, 0x0e, 0x90, 0xd9, 0x19, 0x16, 0xec, 0x80, 0x4c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr218, $sgpr32, 148, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.188, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x19, 0x0e, 0x90, 0xda, 0x19, 0x16, 0xec, 0x80, 0x4a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr219, $sgpr32, 144, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.189, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x19, 0x0e, 0x90, 0xdb, 0x19, 0x16, 0xec, 0x80, 0x48, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr220, $sgpr32, 140, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.190, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x19, 0x0e, 0x90, 0xdc, 0x19, 0x16, 0xec, 0x80, 0x46, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr221, $sgpr32, 136, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.191, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x19, 0x0e, 0x90, 0xdd, 0x19, 0x16, 0xec, 0x80, 0x44, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr222, $sgpr32, 132, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.192, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x19, 0x0e, 0x90, 0xde, 0x19, 0x16, 0xec, 0x80, 0x42, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr223, $sgpr32, 128, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.193, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x19, 0x0e, 0x90, 0xdf, 0x19, 0x16, 0xec, 0x80, 0x40, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr224, $sgpr32, 124, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.194, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x19, 0x0e, 0x90, 0xe0, 0x19, 0x16, 0xec, 0x80, 0x3e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr225, $sgpr32, 120, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.195, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x19, 0x0e, 0x90, 0xe1, 0x19, 0x16, 0xec, 0x80, 0x3c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr226, $sgpr32, 116, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.196, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x19, 0x0e, 0x90, 0xe2, 0x19, 0x16, 0xec, 0x80, 0x3a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr227, $sgpr32, 112, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.197, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x19, 0x0e, 0x90, 0xe3, 0x19, 0x16, 0xec, 0x80, 0x38, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr228, $sgpr32, 108, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.198, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x19, 0x0e, 0x90, 0xe4, 0x19, 0x16, 0xec, 0x80, 0x36, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr229, $sgpr32, 104, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.199, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x19, 0x0e, 0x90, 0xe5, 0x19, 0x16, 0xec, 0x80, 0x34, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr230, $sgpr32, 100, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.200, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x19, 0x0e, 0x90, 0xe6, 0x19, 0x16, 0xec, 0x80, 0x32, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr231, $sgpr32, 96, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.201, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x19, 0x0e, 0x90, 0xe7, 0x19, 0x16, 0xec, 0x80, 0x30, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr232, $sgpr32, 92, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.202, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x19, 0x0e, 0x90, 0xe8, 0x19, 0x16, 0xec, 0x80, 0x2e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr233, $sgpr32, 88, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.203, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x19, 0x0e, 0x90, 0xe9, 0x19, 0x16, 0xec, 0x80, 0x2c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr234, $sgpr32, 84, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.204, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x19, 0x0e, 0x90, 0xea, 0x19, 0x16, 0xec, 0x80, 0x2a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr235, $sgpr32, 80, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.205, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x19, 0x0e, 0x90, 0xeb, 0x19, 0x16, 0xec, 0x80, 0x28, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr236, $sgpr32, 76, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.206, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x19, 0x0e, 0x90, 0xec, 0x19, 0x16, 0xec, 0x80, 0x26, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr237, $sgpr32, 72, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.207, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x19, 0x0e, 0x90, 0xed, 0x19, 0x16, 0xec, 0x80, 0x24, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr238, $sgpr32, 68, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.208, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x19, 0x0e, 0x90, 0xee, 0x19, 0x16, 0xec, 0x80, 0x22, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr239, $sgpr32, 64, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.209, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x19, 0x0e, 0x90, 0xef, 0x19, 0x16, 0xec, 0x80, 0x20, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr240, $sgpr32, 60, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.210, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x19, 0x0e, 0x90, 0xf0, 0x19, 0x16, 0xec, 0x80, 0x1e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr241, $sgpr32, 56, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.211, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x19, 0x0e, 0x90, 0xf1, 0x19, 0x16, 0xec, 0x80, 0x1c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr242, $sgpr32, 52, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.212, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x19, 0x0e, 0x90, 0xf2, 0x19, 0x16, 0xec, 0x80, 0x1a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr243, $sgpr32, 48, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.213, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x19, 0x0e, 0x90, 0xf3, 0x19, 0x16, 0xec, 0x80, 0x18, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr244, $sgpr32, 44, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.214, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x19, 0x0e, 0x90, 0xf4, 0x19, 0x16, 0xec, 0x80, 0x16, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr245, $sgpr32, 40, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.215, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x19, 0x0e, 0x90, 0xf5, 0x19, 0x16, 0xec, 0x80, 0x14, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr246, $sgpr32, 36, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.216, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x19, 0x0e, 0x90, 0xf6, 0x19, 0x16, 0xec, 0x80, 0x12, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr247, $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.217, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x19, 0x0e, 0x90, 0xf7, 0x19, 0x16, 0xec, 0x80, 0x10, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr248, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.218, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x19, 0x0e, 0x90, 0xf8, 0x19, 0x16, 0xec, 0x80, 0x0e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr249, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.219, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x19, 0x0e, 0x90, 0xf9, 0x19, 0x16, 0xec, 0x80, 0x0c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr250, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.220, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x19, 0x0e, 0x90, 0xfa, 0x19, 0x16, 0xec, 0x80, 0x0a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr251, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.221, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x19, 0x0e, 0x90, 0xfb, 0x19, 0x16, 0xec, 0x80, 0x08, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr252, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.222, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x19, 0x0e, 0x90, 0xfc, 0x19, 0x16, 0xec, 0x80, 0x06, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr253, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.223, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x19, 0x0e, 0x90, 0xfd, 0x19, 0x16, 0xec, 0x80, 0x04, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr254, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.224, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x19, 0x0e, 0x90, 0xfe, 0x19, 0x16, 0xec, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr255, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.225, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x19, 0x0d, 0x90, 0xff, 0x19, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: S_CMP_EQ_U32 0, 0, implicit-def $scc ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 704, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.226, addrspace 5) ; GFX90A-FLATSCR-NEXT: $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec @@ -5888,229 +7232,453 @@ ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr28_lo16 ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr29_lo16 ; GFX90A-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr32, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr32_lo16, $vgpr0_lo16 ; GFX90A-NEXT: $vgpr1 = V_ACCVGPR_READ_B32_e64 killed $agpr33, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr33_lo16, $vgpr1_lo16 ; GFX90A-NEXT: $vgpr2 = V_ACCVGPR_READ_B32_e64 killed $agpr34, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr34_lo16, $vgpr2_lo16 ; GFX90A-NEXT: $vgpr3 = V_ACCVGPR_READ_B32_e64 killed $agpr35, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr35_lo16, $vgpr3_lo16 ; GFX90A-NEXT: $vgpr4 = V_ACCVGPR_READ_B32_e64 killed $agpr36, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr36_lo16, $vgpr4_lo16 ; GFX90A-NEXT: $vgpr5 = V_ACCVGPR_READ_B32_e64 killed $agpr37, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr37_lo16, $vgpr5_lo16 ; GFX90A-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 killed $agpr38, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr38_lo16, $vgpr6_lo16 ; GFX90A-NEXT: $vgpr7 = V_ACCVGPR_READ_B32_e64 killed $agpr39, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr39_lo16, $vgpr7_lo16 ; GFX90A-NEXT: $vgpr8 = V_ACCVGPR_READ_B32_e64 killed $agpr40, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr40_lo16, $vgpr8_lo16 ; GFX90A-NEXT: $vgpr9 = V_ACCVGPR_READ_B32_e64 killed $agpr41, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr41_lo16, $vgpr9_lo16 ; GFX90A-NEXT: $vgpr10 = V_ACCVGPR_READ_B32_e64 killed $agpr42, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr42_lo16, $vgpr10_lo16 ; GFX90A-NEXT: $vgpr11 = V_ACCVGPR_READ_B32_e64 killed $agpr43, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr43_lo16, $vgpr11_lo16 ; GFX90A-NEXT: $vgpr12 = V_ACCVGPR_READ_B32_e64 killed $agpr44, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr44_lo16, $vgpr12_lo16 ; GFX90A-NEXT: $vgpr13 = V_ACCVGPR_READ_B32_e64 killed $agpr45, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr45_lo16, $vgpr13_lo16 ; GFX90A-NEXT: $vgpr14 = V_ACCVGPR_READ_B32_e64 killed $agpr46, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr46_lo16, $vgpr14_lo16 ; GFX90A-NEXT: $vgpr15 = V_ACCVGPR_READ_B32_e64 killed $agpr47, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr47_lo16, $vgpr15_lo16 ; GFX90A-NEXT: $vgpr16 = V_ACCVGPR_READ_B32_e64 killed $agpr48, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr48_lo16, $vgpr16_lo16 ; GFX90A-NEXT: $vgpr17 = V_ACCVGPR_READ_B32_e64 killed $agpr49, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr49_lo16, $vgpr17_lo16 ; GFX90A-NEXT: $vgpr18 = V_ACCVGPR_READ_B32_e64 killed $agpr50, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr50_lo16, $vgpr18_lo16 ; GFX90A-NEXT: $vgpr19 = V_ACCVGPR_READ_B32_e64 killed $agpr51, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr51_lo16, $vgpr19_lo16 ; GFX90A-NEXT: $vgpr20 = V_ACCVGPR_READ_B32_e64 killed $agpr52, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr52_lo16, $vgpr20_lo16 ; GFX90A-NEXT: $vgpr21 = V_ACCVGPR_READ_B32_e64 killed $agpr53, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr53_lo16, $vgpr21_lo16 ; GFX90A-NEXT: $vgpr22 = V_ACCVGPR_READ_B32_e64 killed $agpr54, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr54_lo16, $vgpr22_lo16 ; GFX90A-NEXT: $vgpr23 = V_ACCVGPR_READ_B32_e64 killed $agpr55, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr55_lo16, $vgpr23_lo16 ; GFX90A-NEXT: $vgpr24 = V_ACCVGPR_READ_B32_e64 killed $agpr56, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr56_lo16, $vgpr24_lo16 ; GFX90A-NEXT: $vgpr25 = V_ACCVGPR_READ_B32_e64 killed $agpr57, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr57_lo16, $vgpr25_lo16 ; GFX90A-NEXT: $vgpr26 = V_ACCVGPR_READ_B32_e64 killed $agpr58, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr58_lo16, $vgpr26_lo16 ; GFX90A-NEXT: $vgpr27 = V_ACCVGPR_READ_B32_e64 killed $agpr59, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr59_lo16, $vgpr27_lo16 ; GFX90A-NEXT: $vgpr28 = V_ACCVGPR_READ_B32_e64 killed $agpr60, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr60_lo16, $vgpr28_lo16 ; GFX90A-NEXT: $vgpr29 = V_ACCVGPR_READ_B32_e64 killed $agpr61, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr61_lo16, $vgpr29_lo16 ; GFX90A-NEXT: $vgpr30 = V_ACCVGPR_READ_B32_e64 killed $agpr62, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr62_lo16, $vgpr30_lo16 ; GFX90A-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 killed $agpr63, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr63_lo16, $vgpr31_lo16 ; GFX90A-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr64, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr64_lo16, $vgpr32_lo16 ; GFX90A-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 killed $agpr65, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr65_lo16, $vgpr33_lo16 ; GFX90A-NEXT: $vgpr34 = V_ACCVGPR_READ_B32_e64 killed $agpr66, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr66_lo16, $vgpr34_lo16 ; GFX90A-NEXT: $vgpr35 = V_ACCVGPR_READ_B32_e64 killed $agpr67, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr67_lo16, $vgpr35_lo16 ; GFX90A-NEXT: $vgpr36 = V_ACCVGPR_READ_B32_e64 killed $agpr68, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr68_lo16, $vgpr36_lo16 ; GFX90A-NEXT: $vgpr37 = V_ACCVGPR_READ_B32_e64 killed $agpr69, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr69_lo16, $vgpr37_lo16 ; GFX90A-NEXT: $vgpr38 = V_ACCVGPR_READ_B32_e64 killed $agpr70, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr70_lo16, $vgpr38_lo16 ; GFX90A-NEXT: $vgpr39 = V_ACCVGPR_READ_B32_e64 killed $agpr71, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr71_lo16, $vgpr39_lo16 ; GFX90A-NEXT: $vgpr48 = V_ACCVGPR_READ_B32_e64 killed $agpr72, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr72_lo16, $vgpr48_lo16 ; GFX90A-NEXT: $vgpr49 = V_ACCVGPR_READ_B32_e64 killed $agpr73, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr73_lo16, $vgpr49_lo16 ; GFX90A-NEXT: $vgpr50 = V_ACCVGPR_READ_B32_e64 killed $agpr74, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr74_lo16, $vgpr50_lo16 ; GFX90A-NEXT: $vgpr51 = V_ACCVGPR_READ_B32_e64 killed $agpr75, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr75_lo16, $vgpr51_lo16 ; GFX90A-NEXT: $vgpr52 = V_ACCVGPR_READ_B32_e64 killed $agpr76, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr76_lo16, $vgpr52_lo16 ; GFX90A-NEXT: $vgpr53 = V_ACCVGPR_READ_B32_e64 killed $agpr77, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr77_lo16, $vgpr53_lo16 ; GFX90A-NEXT: $vgpr54 = V_ACCVGPR_READ_B32_e64 killed $agpr78, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr78_lo16, $vgpr54_lo16 ; GFX90A-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 killed $agpr79, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr79_lo16, $vgpr55_lo16 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr80, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 700, 0, 0, 0, implicit $exec :: (store (s32) into %stack.50, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x18, 0x0f, 0x90, 0xd0, 0x18, 0x16, 0xec, 0x80, 0xde, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr81, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 696, 0, 0, 0, implicit $exec :: (store (s32) into %stack.51, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x18, 0x0f, 0x90, 0xd1, 0x18, 0x16, 0xec, 0x80, 0xdc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr82, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 692, 0, 0, 0, implicit $exec :: (store (s32) into %stack.52, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x18, 0x0f, 0x90, 0xd2, 0x18, 0x16, 0xec, 0x80, 0xda, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr83, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 688, 0, 0, 0, implicit $exec :: (store (s32) into %stack.53, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x18, 0x0f, 0x90, 0xd3, 0x18, 0x16, 0xec, 0x80, 0xd8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr84, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 684, 0, 0, 0, implicit $exec :: (store (s32) into %stack.54, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x18, 0x0f, 0x90, 0xd4, 0x18, 0x16, 0xec, 0x80, 0xd6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr85, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 680, 0, 0, 0, implicit $exec :: (store (s32) into %stack.55, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x18, 0x0f, 0x90, 0xd5, 0x18, 0x16, 0xec, 0x80, 0xd4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr86, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 676, 0, 0, 0, implicit $exec :: (store (s32) into %stack.56, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x18, 0x0f, 0x90, 0xd6, 0x18, 0x16, 0xec, 0x80, 0xd2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr87, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 672, 0, 0, 0, implicit $exec :: (store (s32) into %stack.57, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x18, 0x0f, 0x90, 0xd7, 0x18, 0x16, 0xec, 0x80, 0xd0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr88, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 668, 0, 0, 0, implicit $exec :: (store (s32) into %stack.58, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x18, 0x0f, 0x90, 0xd8, 0x18, 0x16, 0xec, 0x80, 0xce, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr89, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 664, 0, 0, 0, implicit $exec :: (store (s32) into %stack.59, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x18, 0x0f, 0x90, 0xd9, 0x18, 0x16, 0xec, 0x80, 0xcc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr90, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 660, 0, 0, 0, implicit $exec :: (store (s32) into %stack.60, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x18, 0x0f, 0x90, 0xda, 0x18, 0x16, 0xec, 0x80, 0xca, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr91, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 656, 0, 0, 0, implicit $exec :: (store (s32) into %stack.61, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x18, 0x0f, 0x90, 0xdb, 0x18, 0x16, 0xec, 0x80, 0xc8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr92, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 652, 0, 0, 0, implicit $exec :: (store (s32) into %stack.62, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x18, 0x0f, 0x90, 0xdc, 0x18, 0x16, 0xec, 0x80, 0xc6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr93, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 648, 0, 0, 0, implicit $exec :: (store (s32) into %stack.63, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x18, 0x0f, 0x90, 0xdd, 0x18, 0x16, 0xec, 0x80, 0xc4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr94, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 644, 0, 0, 0, implicit $exec :: (store (s32) into %stack.64, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x18, 0x0f, 0x90, 0xde, 0x18, 0x16, 0xec, 0x80, 0xc2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr95, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 640, 0, 0, 0, implicit $exec :: (store (s32) into %stack.65, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x18, 0x0f, 0x90, 0xdf, 0x18, 0x16, 0xec, 0x80, 0xc0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr96, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 636, 0, 0, 0, implicit $exec :: (store (s32) into %stack.66, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x18, 0x0f, 0x90, 0xe0, 0x18, 0x16, 0xec, 0x80, 0xbe, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr97, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 632, 0, 0, 0, implicit $exec :: (store (s32) into %stack.67, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x18, 0x0f, 0x90, 0xe1, 0x18, 0x16, 0xec, 0x80, 0xbc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr98, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 628, 0, 0, 0, implicit $exec :: (store (s32) into %stack.68, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x18, 0x0f, 0x90, 0xe2, 0x18, 0x16, 0xec, 0x80, 0xba, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr99, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 624, 0, 0, 0, implicit $exec :: (store (s32) into %stack.69, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x18, 0x0f, 0x90, 0xe3, 0x18, 0x16, 0xec, 0x80, 0xb8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr100, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 620, 0, 0, 0, implicit $exec :: (store (s32) into %stack.70, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x18, 0x0f, 0x90, 0xe4, 0x18, 0x16, 0xec, 0x80, 0xb6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr101, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 616, 0, 0, 0, implicit $exec :: (store (s32) into %stack.71, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x18, 0x0f, 0x90, 0xe5, 0x18, 0x16, 0xec, 0x80, 0xb4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr102, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 612, 0, 0, 0, implicit $exec :: (store (s32) into %stack.72, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x18, 0x0f, 0x90, 0xe6, 0x18, 0x16, 0xec, 0x80, 0xb2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr103, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 608, 0, 0, 0, implicit $exec :: (store (s32) into %stack.73, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x18, 0x0f, 0x90, 0xe7, 0x18, 0x16, 0xec, 0x80, 0xb0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr104, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 604, 0, 0, 0, implicit $exec :: (store (s32) into %stack.74, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x18, 0x0f, 0x90, 0xe8, 0x18, 0x16, 0xec, 0x80, 0xae, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr105, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 600, 0, 0, 0, implicit $exec :: (store (s32) into %stack.75, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x18, 0x0f, 0x90, 0xe9, 0x18, 0x16, 0xec, 0x80, 0xac, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr106, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 596, 0, 0, 0, implicit $exec :: (store (s32) into %stack.76, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x18, 0x0f, 0x90, 0xea, 0x18, 0x16, 0xec, 0x80, 0xaa, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr107, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 592, 0, 0, 0, implicit $exec :: (store (s32) into %stack.77, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x18, 0x0f, 0x90, 0xeb, 0x18, 0x16, 0xec, 0x80, 0xa8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr108, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 588, 0, 0, 0, implicit $exec :: (store (s32) into %stack.78, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x18, 0x0f, 0x90, 0xec, 0x18, 0x16, 0xec, 0x80, 0xa6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr109, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 584, 0, 0, 0, implicit $exec :: (store (s32) into %stack.79, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x18, 0x0f, 0x90, 0xed, 0x18, 0x16, 0xec, 0x80, 0xa4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr110, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 580, 0, 0, 0, implicit $exec :: (store (s32) into %stack.80, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x18, 0x0f, 0x90, 0xee, 0x18, 0x16, 0xec, 0x80, 0xa2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr111, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 576, 0, 0, 0, implicit $exec :: (store (s32) into %stack.81, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x18, 0x0f, 0x90, 0xef, 0x18, 0x16, 0xec, 0x80, 0xa0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr112, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 572, 0, 0, 0, implicit $exec :: (store (s32) into %stack.82, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x18, 0x0f, 0x90, 0xf0, 0x18, 0x16, 0xec, 0x80, 0x9e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr113, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 568, 0, 0, 0, implicit $exec :: (store (s32) into %stack.83, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x18, 0x0f, 0x90, 0xf1, 0x18, 0x16, 0xec, 0x80, 0x9c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr114, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 564, 0, 0, 0, implicit $exec :: (store (s32) into %stack.84, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x18, 0x0f, 0x90, 0xf2, 0x18, 0x16, 0xec, 0x80, 0x9a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr115, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 560, 0, 0, 0, implicit $exec :: (store (s32) into %stack.85, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x18, 0x0f, 0x90, 0xf3, 0x18, 0x16, 0xec, 0x80, 0x98, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr116, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 556, 0, 0, 0, implicit $exec :: (store (s32) into %stack.86, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x18, 0x0f, 0x90, 0xf4, 0x18, 0x16, 0xec, 0x80, 0x96, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr117, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 552, 0, 0, 0, implicit $exec :: (store (s32) into %stack.87, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x18, 0x0f, 0x90, 0xf5, 0x18, 0x16, 0xec, 0x80, 0x94, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr118, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 548, 0, 0, 0, implicit $exec :: (store (s32) into %stack.88, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x18, 0x0f, 0x90, 0xf6, 0x18, 0x16, 0xec, 0x80, 0x92, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr119, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 544, 0, 0, 0, implicit $exec :: (store (s32) into %stack.89, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x18, 0x0f, 0x90, 0xf7, 0x18, 0x16, 0xec, 0x80, 0x90, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr120, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 540, 0, 0, 0, implicit $exec :: (store (s32) into %stack.90, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x18, 0x0f, 0x90, 0xf8, 0x18, 0x16, 0xec, 0x80, 0x8e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr121, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 536, 0, 0, 0, implicit $exec :: (store (s32) into %stack.91, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x18, 0x0f, 0x90, 0xf9, 0x18, 0x16, 0xec, 0x80, 0x8c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr122, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 532, 0, 0, 0, implicit $exec :: (store (s32) into %stack.92, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x18, 0x0f, 0x90, 0xfa, 0x18, 0x16, 0xec, 0x80, 0x8a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr123, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 528, 0, 0, 0, implicit $exec :: (store (s32) into %stack.93, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x18, 0x0f, 0x90, 0xfb, 0x18, 0x16, 0xec, 0x80, 0x88, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr124, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 524, 0, 0, 0, implicit $exec :: (store (s32) into %stack.94, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x18, 0x0f, 0x90, 0xfc, 0x18, 0x16, 0xec, 0x80, 0x86, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr125, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 520, 0, 0, 0, implicit $exec :: (store (s32) into %stack.95, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x18, 0x0f, 0x90, 0xfd, 0x18, 0x16, 0xec, 0x80, 0x84, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr126, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 516, 0, 0, 0, implicit $exec :: (store (s32) into %stack.96, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x18, 0x0f, 0x90, 0xfe, 0x18, 0x16, 0xec, 0x80, 0x82, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr127, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 512, 0, 0, 0, implicit $exec :: (store (s32) into %stack.97, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x18, 0x0f, 0x90, 0xff, 0x18, 0x16, 0xec, 0x80, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr128, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 508, 0, 0, 0, implicit $exec :: (store (s32) into %stack.98, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x80, 0x19, 0x0f, 0x90, 0x80, 0x19, 0x16, 0xec, 0x80, 0xfe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr129, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 504, 0, 0, 0, implicit $exec :: (store (s32) into %stack.99, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x81, 0x19, 0x0f, 0x90, 0x81, 0x19, 0x16, 0xec, 0x80, 0xfc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr130, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 500, 0, 0, 0, implicit $exec :: (store (s32) into %stack.100, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x82, 0x19, 0x0f, 0x90, 0x82, 0x19, 0x16, 0xec, 0x80, 0xfa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr131, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 496, 0, 0, 0, implicit $exec :: (store (s32) into %stack.101, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x83, 0x19, 0x0f, 0x90, 0x83, 0x19, 0x16, 0xec, 0x80, 0xf8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr132, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 492, 0, 0, 0, implicit $exec :: (store (s32) into %stack.102, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x84, 0x19, 0x0f, 0x90, 0x84, 0x19, 0x16, 0xec, 0x80, 0xf6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr133, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 488, 0, 0, 0, implicit $exec :: (store (s32) into %stack.103, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x85, 0x19, 0x0f, 0x90, 0x85, 0x19, 0x16, 0xec, 0x80, 0xf4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr134, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 484, 0, 0, 0, implicit $exec :: (store (s32) into %stack.104, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x86, 0x19, 0x0f, 0x90, 0x86, 0x19, 0x16, 0xec, 0x80, 0xf2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr135, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 480, 0, 0, 0, implicit $exec :: (store (s32) into %stack.105, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x87, 0x19, 0x0f, 0x90, 0x87, 0x19, 0x16, 0xec, 0x80, 0xf0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr136, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 476, 0, 0, 0, implicit $exec :: (store (s32) into %stack.106, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x88, 0x19, 0x0f, 0x90, 0x88, 0x19, 0x16, 0xec, 0x80, 0xee, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr137, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 472, 0, 0, 0, implicit $exec :: (store (s32) into %stack.107, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x89, 0x19, 0x0f, 0x90, 0x89, 0x19, 0x16, 0xec, 0x80, 0xec, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr138, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 468, 0, 0, 0, implicit $exec :: (store (s32) into %stack.108, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8a, 0x19, 0x0f, 0x90, 0x8a, 0x19, 0x16, 0xec, 0x80, 0xea, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr139, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 464, 0, 0, 0, implicit $exec :: (store (s32) into %stack.109, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8b, 0x19, 0x0f, 0x90, 0x8b, 0x19, 0x16, 0xec, 0x80, 0xe8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr140, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 460, 0, 0, 0, implicit $exec :: (store (s32) into %stack.110, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8c, 0x19, 0x0f, 0x90, 0x8c, 0x19, 0x16, 0xec, 0x80, 0xe6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr141, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 456, 0, 0, 0, implicit $exec :: (store (s32) into %stack.111, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8d, 0x19, 0x0f, 0x90, 0x8d, 0x19, 0x16, 0xec, 0x80, 0xe4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr142, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 452, 0, 0, 0, implicit $exec :: (store (s32) into %stack.112, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8e, 0x19, 0x0f, 0x90, 0x8e, 0x19, 0x16, 0xec, 0x80, 0xe2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr143, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 448, 0, 0, 0, implicit $exec :: (store (s32) into %stack.113, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8f, 0x19, 0x0f, 0x90, 0x8f, 0x19, 0x16, 0xec, 0x80, 0xe0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr144, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 444, 0, 0, 0, implicit $exec :: (store (s32) into %stack.114, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x90, 0x19, 0x0f, 0x90, 0x90, 0x19, 0x16, 0xec, 0x80, 0xde, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr145, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 440, 0, 0, 0, implicit $exec :: (store (s32) into %stack.115, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x91, 0x19, 0x0f, 0x90, 0x91, 0x19, 0x16, 0xec, 0x80, 0xdc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr146, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 436, 0, 0, 0, implicit $exec :: (store (s32) into %stack.116, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x92, 0x19, 0x0f, 0x90, 0x92, 0x19, 0x16, 0xec, 0x80, 0xda, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr147, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 432, 0, 0, 0, implicit $exec :: (store (s32) into %stack.117, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x93, 0x19, 0x0f, 0x90, 0x93, 0x19, 0x16, 0xec, 0x80, 0xd8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr148, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 428, 0, 0, 0, implicit $exec :: (store (s32) into %stack.118, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x94, 0x19, 0x0f, 0x90, 0x94, 0x19, 0x16, 0xec, 0x80, 0xd6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr149, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 424, 0, 0, 0, implicit $exec :: (store (s32) into %stack.119, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x95, 0x19, 0x0f, 0x90, 0x95, 0x19, 0x16, 0xec, 0x80, 0xd4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr150, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 420, 0, 0, 0, implicit $exec :: (store (s32) into %stack.120, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x96, 0x19, 0x0f, 0x90, 0x96, 0x19, 0x16, 0xec, 0x80, 0xd2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr151, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 416, 0, 0, 0, implicit $exec :: (store (s32) into %stack.121, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x97, 0x19, 0x0f, 0x90, 0x97, 0x19, 0x16, 0xec, 0x80, 0xd0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr152, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 412, 0, 0, 0, implicit $exec :: (store (s32) into %stack.122, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x98, 0x19, 0x0f, 0x90, 0x98, 0x19, 0x16, 0xec, 0x80, 0xce, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr153, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 408, 0, 0, 0, implicit $exec :: (store (s32) into %stack.123, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x99, 0x19, 0x0f, 0x90, 0x99, 0x19, 0x16, 0xec, 0x80, 0xcc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr154, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 404, 0, 0, 0, implicit $exec :: (store (s32) into %stack.124, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9a, 0x19, 0x0f, 0x90, 0x9a, 0x19, 0x16, 0xec, 0x80, 0xca, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr155, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 400, 0, 0, 0, implicit $exec :: (store (s32) into %stack.125, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9b, 0x19, 0x0f, 0x90, 0x9b, 0x19, 0x16, 0xec, 0x80, 0xc8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr156, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 396, 0, 0, 0, implicit $exec :: (store (s32) into %stack.126, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9c, 0x19, 0x0f, 0x90, 0x9c, 0x19, 0x16, 0xec, 0x80, 0xc6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr157, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 392, 0, 0, 0, implicit $exec :: (store (s32) into %stack.127, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9d, 0x19, 0x0f, 0x90, 0x9d, 0x19, 0x16, 0xec, 0x80, 0xc4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr158, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 388, 0, 0, 0, implicit $exec :: (store (s32) into %stack.128, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9e, 0x19, 0x0f, 0x90, 0x9e, 0x19, 0x16, 0xec, 0x80, 0xc2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr159, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 384, 0, 0, 0, implicit $exec :: (store (s32) into %stack.129, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9f, 0x19, 0x0f, 0x90, 0x9f, 0x19, 0x16, 0xec, 0x80, 0xc0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr160, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 380, 0, 0, 0, implicit $exec :: (store (s32) into %stack.130, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa0, 0x19, 0x0f, 0x90, 0xa0, 0x19, 0x16, 0xec, 0x80, 0xbe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr161, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 376, 0, 0, 0, implicit $exec :: (store (s32) into %stack.131, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa1, 0x19, 0x0f, 0x90, 0xa1, 0x19, 0x16, 0xec, 0x80, 0xbc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr162, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 372, 0, 0, 0, implicit $exec :: (store (s32) into %stack.132, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa2, 0x19, 0x0f, 0x90, 0xa2, 0x19, 0x16, 0xec, 0x80, 0xba, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr163, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 368, 0, 0, 0, implicit $exec :: (store (s32) into %stack.133, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa3, 0x19, 0x0f, 0x90, 0xa3, 0x19, 0x16, 0xec, 0x80, 0xb8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr164, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 364, 0, 0, 0, implicit $exec :: (store (s32) into %stack.134, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa4, 0x19, 0x0f, 0x90, 0xa4, 0x19, 0x16, 0xec, 0x80, 0xb6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr165, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 360, 0, 0, 0, implicit $exec :: (store (s32) into %stack.135, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa5, 0x19, 0x0f, 0x90, 0xa5, 0x19, 0x16, 0xec, 0x80, 0xb4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr166, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 356, 0, 0, 0, implicit $exec :: (store (s32) into %stack.136, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa6, 0x19, 0x0f, 0x90, 0xa6, 0x19, 0x16, 0xec, 0x80, 0xb2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr167, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 352, 0, 0, 0, implicit $exec :: (store (s32) into %stack.137, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa7, 0x19, 0x0f, 0x90, 0xa7, 0x19, 0x16, 0xec, 0x80, 0xb0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr168, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 348, 0, 0, 0, implicit $exec :: (store (s32) into %stack.138, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa8, 0x19, 0x0f, 0x90, 0xa8, 0x19, 0x16, 0xec, 0x80, 0xae, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr169, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 344, 0, 0, 0, implicit $exec :: (store (s32) into %stack.139, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa9, 0x19, 0x0f, 0x90, 0xa9, 0x19, 0x16, 0xec, 0x80, 0xac, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr170, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 340, 0, 0, 0, implicit $exec :: (store (s32) into %stack.140, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaa, 0x19, 0x0f, 0x90, 0xaa, 0x19, 0x16, 0xec, 0x80, 0xaa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr171, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 336, 0, 0, 0, implicit $exec :: (store (s32) into %stack.141, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xab, 0x19, 0x0f, 0x90, 0xab, 0x19, 0x16, 0xec, 0x80, 0xa8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr172, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 332, 0, 0, 0, implicit $exec :: (store (s32) into %stack.142, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xac, 0x19, 0x0f, 0x90, 0xac, 0x19, 0x16, 0xec, 0x80, 0xa6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr173, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 328, 0, 0, 0, implicit $exec :: (store (s32) into %stack.143, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xad, 0x19, 0x0f, 0x90, 0xad, 0x19, 0x16, 0xec, 0x80, 0xa4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr174, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 324, 0, 0, 0, implicit $exec :: (store (s32) into %stack.144, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xae, 0x19, 0x0f, 0x90, 0xae, 0x19, 0x16, 0xec, 0x80, 0xa2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr175, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 320, 0, 0, 0, implicit $exec :: (store (s32) into %stack.145, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaf, 0x19, 0x0f, 0x90, 0xaf, 0x19, 0x16, 0xec, 0x80, 0xa0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr176, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 316, 0, 0, 0, implicit $exec :: (store (s32) into %stack.146, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb0, 0x19, 0x0f, 0x90, 0xb0, 0x19, 0x16, 0xec, 0x80, 0x9e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr177, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 312, 0, 0, 0, implicit $exec :: (store (s32) into %stack.147, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb1, 0x19, 0x0f, 0x90, 0xb1, 0x19, 0x16, 0xec, 0x80, 0x9c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr178, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 308, 0, 0, 0, implicit $exec :: (store (s32) into %stack.148, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb2, 0x19, 0x0f, 0x90, 0xb2, 0x19, 0x16, 0xec, 0x80, 0x9a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr179, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 304, 0, 0, 0, implicit $exec :: (store (s32) into %stack.149, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb3, 0x19, 0x0f, 0x90, 0xb3, 0x19, 0x16, 0xec, 0x80, 0x98, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr180, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 300, 0, 0, 0, implicit $exec :: (store (s32) into %stack.150, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb4, 0x19, 0x0f, 0x90, 0xb4, 0x19, 0x16, 0xec, 0x80, 0x96, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr181, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 296, 0, 0, 0, implicit $exec :: (store (s32) into %stack.151, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb5, 0x19, 0x0f, 0x90, 0xb5, 0x19, 0x16, 0xec, 0x80, 0x94, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr182, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 292, 0, 0, 0, implicit $exec :: (store (s32) into %stack.152, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb6, 0x19, 0x0f, 0x90, 0xb6, 0x19, 0x16, 0xec, 0x80, 0x92, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr183, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 288, 0, 0, 0, implicit $exec :: (store (s32) into %stack.153, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb7, 0x19, 0x0f, 0x90, 0xb7, 0x19, 0x16, 0xec, 0x80, 0x90, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr184, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 284, 0, 0, 0, implicit $exec :: (store (s32) into %stack.154, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb8, 0x19, 0x0f, 0x90, 0xb8, 0x19, 0x16, 0xec, 0x80, 0x8e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr185, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 280, 0, 0, 0, implicit $exec :: (store (s32) into %stack.155, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb9, 0x19, 0x0f, 0x90, 0xb9, 0x19, 0x16, 0xec, 0x80, 0x8c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr186, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 276, 0, 0, 0, implicit $exec :: (store (s32) into %stack.156, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xba, 0x19, 0x0f, 0x90, 0xba, 0x19, 0x16, 0xec, 0x80, 0x8a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr187, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 272, 0, 0, 0, implicit $exec :: (store (s32) into %stack.157, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbb, 0x19, 0x0f, 0x90, 0xbb, 0x19, 0x16, 0xec, 0x80, 0x88, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr188, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 268, 0, 0, 0, implicit $exec :: (store (s32) into %stack.158, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbc, 0x19, 0x0f, 0x90, 0xbc, 0x19, 0x16, 0xec, 0x80, 0x86, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr189, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 264, 0, 0, 0, implicit $exec :: (store (s32) into %stack.159, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbd, 0x19, 0x0f, 0x90, 0xbd, 0x19, 0x16, 0xec, 0x80, 0x84, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr190, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 260, 0, 0, 0, implicit $exec :: (store (s32) into %stack.160, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbe, 0x19, 0x0f, 0x90, 0xbe, 0x19, 0x16, 0xec, 0x80, 0x82, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr191, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 256, 0, 0, 0, implicit $exec :: (store (s32) into %stack.161, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbf, 0x19, 0x0f, 0x90, 0xbf, 0x19, 0x16, 0xec, 0x80, 0x80, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr192, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 252, 0, 0, 0, implicit $exec :: (store (s32) into %stack.162, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc0, 0x19, 0x0e, 0x90, 0xc0, 0x19, 0x16, 0xec, 0x80, 0x7e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr193, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 248, 0, 0, 0, implicit $exec :: (store (s32) into %stack.163, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc1, 0x19, 0x0e, 0x90, 0xc1, 0x19, 0x16, 0xec, 0x80, 0x7c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr194, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 244, 0, 0, 0, implicit $exec :: (store (s32) into %stack.164, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc2, 0x19, 0x0e, 0x90, 0xc2, 0x19, 0x16, 0xec, 0x80, 0x7a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr195, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 240, 0, 0, 0, implicit $exec :: (store (s32) into %stack.165, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc3, 0x19, 0x0e, 0x90, 0xc3, 0x19, 0x16, 0xec, 0x80, 0x78, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr196, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 236, 0, 0, 0, implicit $exec :: (store (s32) into %stack.166, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc4, 0x19, 0x0e, 0x90, 0xc4, 0x19, 0x16, 0xec, 0x80, 0x76, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr197, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 232, 0, 0, 0, implicit $exec :: (store (s32) into %stack.167, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc5, 0x19, 0x0e, 0x90, 0xc5, 0x19, 0x16, 0xec, 0x80, 0x74, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr198, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 228, 0, 0, 0, implicit $exec :: (store (s32) into %stack.168, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc6, 0x19, 0x0e, 0x90, 0xc6, 0x19, 0x16, 0xec, 0x80, 0x72, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr199, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 224, 0, 0, 0, implicit $exec :: (store (s32) into %stack.169, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc7, 0x19, 0x0e, 0x90, 0xc7, 0x19, 0x16, 0xec, 0x80, 0x70, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr200, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 220, 0, 0, 0, implicit $exec :: (store (s32) into %stack.170, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc8, 0x19, 0x0e, 0x90, 0xc8, 0x19, 0x16, 0xec, 0x80, 0x6e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr201, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 216, 0, 0, 0, implicit $exec :: (store (s32) into %stack.171, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc9, 0x19, 0x0e, 0x90, 0xc9, 0x19, 0x16, 0xec, 0x80, 0x6c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr202, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 212, 0, 0, 0, implicit $exec :: (store (s32) into %stack.172, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xca, 0x19, 0x0e, 0x90, 0xca, 0x19, 0x16, 0xec, 0x80, 0x6a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr203, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 208, 0, 0, 0, implicit $exec :: (store (s32) into %stack.173, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcb, 0x19, 0x0e, 0x90, 0xcb, 0x19, 0x16, 0xec, 0x80, 0x68, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr204, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 204, 0, 0, 0, implicit $exec :: (store (s32) into %stack.174, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcc, 0x19, 0x0e, 0x90, 0xcc, 0x19, 0x16, 0xec, 0x80, 0x66, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr205, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 200, 0, 0, 0, implicit $exec :: (store (s32) into %stack.175, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcd, 0x19, 0x0e, 0x90, 0xcd, 0x19, 0x16, 0xec, 0x80, 0x64, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr206, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 196, 0, 0, 0, implicit $exec :: (store (s32) into %stack.176, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xce, 0x19, 0x0e, 0x90, 0xce, 0x19, 0x16, 0xec, 0x80, 0x62, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr207, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 192, 0, 0, 0, implicit $exec :: (store (s32) into %stack.177, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcf, 0x19, 0x0e, 0x90, 0xcf, 0x19, 0x16, 0xec, 0x80, 0x60, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr208, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 188, 0, 0, 0, implicit $exec :: (store (s32) into %stack.178, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x19, 0x0e, 0x90, 0xd0, 0x19, 0x16, 0xec, 0x80, 0x5e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr209, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 184, 0, 0, 0, implicit $exec :: (store (s32) into %stack.179, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x19, 0x0e, 0x90, 0xd1, 0x19, 0x16, 0xec, 0x80, 0x5c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr210, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 180, 0, 0, 0, implicit $exec :: (store (s32) into %stack.180, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x19, 0x0e, 0x90, 0xd2, 0x19, 0x16, 0xec, 0x80, 0x5a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr211, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 176, 0, 0, 0, implicit $exec :: (store (s32) into %stack.181, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x19, 0x0e, 0x90, 0xd3, 0x19, 0x16, 0xec, 0x80, 0x58, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr212, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 172, 0, 0, 0, implicit $exec :: (store (s32) into %stack.182, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x19, 0x0e, 0x90, 0xd4, 0x19, 0x16, 0xec, 0x80, 0x56, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr213, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 168, 0, 0, 0, implicit $exec :: (store (s32) into %stack.183, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x19, 0x0e, 0x90, 0xd5, 0x19, 0x16, 0xec, 0x80, 0x54, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr214, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 164, 0, 0, 0, implicit $exec :: (store (s32) into %stack.184, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x19, 0x0e, 0x90, 0xd6, 0x19, 0x16, 0xec, 0x80, 0x52, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr215, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 160, 0, 0, 0, implicit $exec :: (store (s32) into %stack.185, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x19, 0x0e, 0x90, 0xd7, 0x19, 0x16, 0xec, 0x80, 0x50, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr216, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 156, 0, 0, 0, implicit $exec :: (store (s32) into %stack.186, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x19, 0x0e, 0x90, 0xd8, 0x19, 0x16, 0xec, 0x80, 0x4e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr217, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 152, 0, 0, 0, implicit $exec :: (store (s32) into %stack.187, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x19, 0x0e, 0x90, 0xd9, 0x19, 0x16, 0xec, 0x80, 0x4c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr218, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 148, 0, 0, 0, implicit $exec :: (store (s32) into %stack.188, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x19, 0x0e, 0x90, 0xda, 0x19, 0x16, 0xec, 0x80, 0x4a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr219, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 144, 0, 0, 0, implicit $exec :: (store (s32) into %stack.189, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x19, 0x0e, 0x90, 0xdb, 0x19, 0x16, 0xec, 0x80, 0x48, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr220, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 140, 0, 0, 0, implicit $exec :: (store (s32) into %stack.190, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x19, 0x0e, 0x90, 0xdc, 0x19, 0x16, 0xec, 0x80, 0x46, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr221, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 136, 0, 0, 0, implicit $exec :: (store (s32) into %stack.191, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x19, 0x0e, 0x90, 0xdd, 0x19, 0x16, 0xec, 0x80, 0x44, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr222, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 132, 0, 0, 0, implicit $exec :: (store (s32) into %stack.192, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x19, 0x0e, 0x90, 0xde, 0x19, 0x16, 0xec, 0x80, 0x42, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr223, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 128, 0, 0, 0, implicit $exec :: (store (s32) into %stack.193, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x19, 0x0e, 0x90, 0xdf, 0x19, 0x16, 0xec, 0x80, 0x40, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr224, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 124, 0, 0, 0, implicit $exec :: (store (s32) into %stack.194, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x19, 0x0e, 0x90, 0xe0, 0x19, 0x16, 0xec, 0x80, 0x3e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr225, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 120, 0, 0, 0, implicit $exec :: (store (s32) into %stack.195, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x19, 0x0e, 0x90, 0xe1, 0x19, 0x16, 0xec, 0x80, 0x3c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr226, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 116, 0, 0, 0, implicit $exec :: (store (s32) into %stack.196, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x19, 0x0e, 0x90, 0xe2, 0x19, 0x16, 0xec, 0x80, 0x3a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr227, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 112, 0, 0, 0, implicit $exec :: (store (s32) into %stack.197, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x19, 0x0e, 0x90, 0xe3, 0x19, 0x16, 0xec, 0x80, 0x38, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr228, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 108, 0, 0, 0, implicit $exec :: (store (s32) into %stack.198, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x19, 0x0e, 0x90, 0xe4, 0x19, 0x16, 0xec, 0x80, 0x36, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr229, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 104, 0, 0, 0, implicit $exec :: (store (s32) into %stack.199, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x19, 0x0e, 0x90, 0xe5, 0x19, 0x16, 0xec, 0x80, 0x34, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr230, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 100, 0, 0, 0, implicit $exec :: (store (s32) into %stack.200, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x19, 0x0e, 0x90, 0xe6, 0x19, 0x16, 0xec, 0x80, 0x32, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr231, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 96, 0, 0, 0, implicit $exec :: (store (s32) into %stack.201, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x19, 0x0e, 0x90, 0xe7, 0x19, 0x16, 0xec, 0x80, 0x30, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr232, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 92, 0, 0, 0, implicit $exec :: (store (s32) into %stack.202, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x19, 0x0e, 0x90, 0xe8, 0x19, 0x16, 0xec, 0x80, 0x2e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr233, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 88, 0, 0, 0, implicit $exec :: (store (s32) into %stack.203, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x19, 0x0e, 0x90, 0xe9, 0x19, 0x16, 0xec, 0x80, 0x2c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr234, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 84, 0, 0, 0, implicit $exec :: (store (s32) into %stack.204, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x19, 0x0e, 0x90, 0xea, 0x19, 0x16, 0xec, 0x80, 0x2a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr235, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 80, 0, 0, 0, implicit $exec :: (store (s32) into %stack.205, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x19, 0x0e, 0x90, 0xeb, 0x19, 0x16, 0xec, 0x80, 0x28, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr236, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 76, 0, 0, 0, implicit $exec :: (store (s32) into %stack.206, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x19, 0x0e, 0x90, 0xec, 0x19, 0x16, 0xec, 0x80, 0x26, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr237, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 72, 0, 0, 0, implicit $exec :: (store (s32) into %stack.207, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x19, 0x0e, 0x90, 0xed, 0x19, 0x16, 0xec, 0x80, 0x24, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr238, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 68, 0, 0, 0, implicit $exec :: (store (s32) into %stack.208, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x19, 0x0e, 0x90, 0xee, 0x19, 0x16, 0xec, 0x80, 0x22, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr239, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 64, 0, 0, 0, implicit $exec :: (store (s32) into %stack.209, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x19, 0x0e, 0x90, 0xef, 0x19, 0x16, 0xec, 0x80, 0x20, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr240, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec :: (store (s32) into %stack.210, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x19, 0x0e, 0x90, 0xf0, 0x19, 0x16, 0xec, 0x80, 0x1e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr241, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec :: (store (s32) into %stack.211, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x19, 0x0e, 0x90, 0xf1, 0x19, 0x16, 0xec, 0x80, 0x1c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr242, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec :: (store (s32) into %stack.212, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x19, 0x0e, 0x90, 0xf2, 0x19, 0x16, 0xec, 0x80, 0x1a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr243, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec :: (store (s32) into %stack.213, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x19, 0x0e, 0x90, 0xf3, 0x19, 0x16, 0xec, 0x80, 0x18, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr244, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec :: (store (s32) into %stack.214, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x19, 0x0e, 0x90, 0xf4, 0x19, 0x16, 0xec, 0x80, 0x16, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr245, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec :: (store (s32) into %stack.215, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x19, 0x0e, 0x90, 0xf5, 0x19, 0x16, 0xec, 0x80, 0x14, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr246, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec :: (store (s32) into %stack.216, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x19, 0x0e, 0x90, 0xf6, 0x19, 0x16, 0xec, 0x80, 0x12, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr247, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.217, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x19, 0x0e, 0x90, 0xf7, 0x19, 0x16, 0xec, 0x80, 0x10, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr248, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (store (s32) into %stack.218, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x19, 0x0e, 0x90, 0xf8, 0x19, 0x16, 0xec, 0x80, 0x0e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr249, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (store (s32) into %stack.219, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x19, 0x0e, 0x90, 0xf9, 0x19, 0x16, 0xec, 0x80, 0x0c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr250, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (store (s32) into %stack.220, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x19, 0x0e, 0x90, 0xfa, 0x19, 0x16, 0xec, 0x80, 0x0a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr251, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (store (s32) into %stack.221, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x19, 0x0e, 0x90, 0xfb, 0x19, 0x16, 0xec, 0x80, 0x08, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr252, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (store (s32) into %stack.222, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x19, 0x0e, 0x90, 0xfc, 0x19, 0x16, 0xec, 0x80, 0x06, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr253, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (store (s32) into %stack.223, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x19, 0x0e, 0x90, 0xfd, 0x19, 0x16, 0xec, 0x80, 0x04, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr254, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.224, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x19, 0x0e, 0x90, 0xfe, 0x19, 0x16, 0xec, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr255, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.225, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x19, 0x0d, 0x90, 0xff, 0x19, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: S_CMP_EQ_U32 0, 0, implicit-def $scc ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 704, 0, 0, 0, implicit $exec :: (store (s32) into %stack.226, addrspace 5) ; GFX90A-NEXT: $vgpr0 = V_MOV_B32_e32 8904, implicit $exec @@ -6779,229 +8347,453 @@ ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr28_lo16 ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr29_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr32, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr32_lo16, $vgpr0_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr1 = V_ACCVGPR_READ_B32_e64 killed $agpr33, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr33_lo16, $vgpr1_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr2 = V_ACCVGPR_READ_B32_e64 killed $agpr34, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr34_lo16, $vgpr2_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr3 = V_ACCVGPR_READ_B32_e64 killed $agpr35, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr35_lo16, $vgpr3_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr4 = V_ACCVGPR_READ_B32_e64 killed $agpr36, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr36_lo16, $vgpr4_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr5 = V_ACCVGPR_READ_B32_e64 killed $agpr37, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr37_lo16, $vgpr5_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 killed $agpr38, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr38_lo16, $vgpr6_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr7 = V_ACCVGPR_READ_B32_e64 killed $agpr39, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr39_lo16, $vgpr7_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr8 = V_ACCVGPR_READ_B32_e64 killed $agpr40, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr40_lo16, $vgpr8_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr9 = V_ACCVGPR_READ_B32_e64 killed $agpr41, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr41_lo16, $vgpr9_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr10 = V_ACCVGPR_READ_B32_e64 killed $agpr42, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr42_lo16, $vgpr10_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr11 = V_ACCVGPR_READ_B32_e64 killed $agpr43, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr43_lo16, $vgpr11_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr12 = V_ACCVGPR_READ_B32_e64 killed $agpr44, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr44_lo16, $vgpr12_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr13 = V_ACCVGPR_READ_B32_e64 killed $agpr45, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr45_lo16, $vgpr13_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr14 = V_ACCVGPR_READ_B32_e64 killed $agpr46, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr46_lo16, $vgpr14_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr15 = V_ACCVGPR_READ_B32_e64 killed $agpr47, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr47_lo16, $vgpr15_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr16 = V_ACCVGPR_READ_B32_e64 killed $agpr48, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr48_lo16, $vgpr16_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr17 = V_ACCVGPR_READ_B32_e64 killed $agpr49, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr49_lo16, $vgpr17_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr18 = V_ACCVGPR_READ_B32_e64 killed $agpr50, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr50_lo16, $vgpr18_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr19 = V_ACCVGPR_READ_B32_e64 killed $agpr51, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr51_lo16, $vgpr19_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr20 = V_ACCVGPR_READ_B32_e64 killed $agpr52, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr52_lo16, $vgpr20_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr21 = V_ACCVGPR_READ_B32_e64 killed $agpr53, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr53_lo16, $vgpr21_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr22 = V_ACCVGPR_READ_B32_e64 killed $agpr54, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr54_lo16, $vgpr22_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr23 = V_ACCVGPR_READ_B32_e64 killed $agpr55, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr55_lo16, $vgpr23_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr24 = V_ACCVGPR_READ_B32_e64 killed $agpr56, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr56_lo16, $vgpr24_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr25 = V_ACCVGPR_READ_B32_e64 killed $agpr57, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr57_lo16, $vgpr25_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr26 = V_ACCVGPR_READ_B32_e64 killed $agpr58, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr58_lo16, $vgpr26_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr27 = V_ACCVGPR_READ_B32_e64 killed $agpr59, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr59_lo16, $vgpr27_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr28 = V_ACCVGPR_READ_B32_e64 killed $agpr60, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr60_lo16, $vgpr28_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr29 = V_ACCVGPR_READ_B32_e64 killed $agpr61, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr61_lo16, $vgpr29_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr30 = V_ACCVGPR_READ_B32_e64 killed $agpr62, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr62_lo16, $vgpr30_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 killed $agpr63, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr63_lo16, $vgpr31_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr64, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr64_lo16, $vgpr32_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 killed $agpr65, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr65_lo16, $vgpr33_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr34 = V_ACCVGPR_READ_B32_e64 killed $agpr66, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr66_lo16, $vgpr34_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr35 = V_ACCVGPR_READ_B32_e64 killed $agpr67, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr67_lo16, $vgpr35_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr36 = V_ACCVGPR_READ_B32_e64 killed $agpr68, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr68_lo16, $vgpr36_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr37 = V_ACCVGPR_READ_B32_e64 killed $agpr69, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr69_lo16, $vgpr37_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr38 = V_ACCVGPR_READ_B32_e64 killed $agpr70, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr70_lo16, $vgpr38_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr39 = V_ACCVGPR_READ_B32_e64 killed $agpr71, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr71_lo16, $vgpr39_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr48 = V_ACCVGPR_READ_B32_e64 killed $agpr72, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr72_lo16, $vgpr48_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr49 = V_ACCVGPR_READ_B32_e64 killed $agpr73, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr73_lo16, $vgpr49_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr50 = V_ACCVGPR_READ_B32_e64 killed $agpr74, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr74_lo16, $vgpr50_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr51 = V_ACCVGPR_READ_B32_e64 killed $agpr75, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr75_lo16, $vgpr51_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr52 = V_ACCVGPR_READ_B32_e64 killed $agpr76, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr76_lo16, $vgpr52_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr53 = V_ACCVGPR_READ_B32_e64 killed $agpr77, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr77_lo16, $vgpr53_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr54 = V_ACCVGPR_READ_B32_e64 killed $agpr78, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr78_lo16, $vgpr54_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 killed $agpr79, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr79_lo16, $vgpr55_lo16 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr80, $sgpr32, 700, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.50, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x18, 0x0f, 0x90, 0xd0, 0x18, 0x16, 0xec, 0x80, 0xde, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr81, $sgpr32, 696, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.51, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x18, 0x0f, 0x90, 0xd1, 0x18, 0x16, 0xec, 0x80, 0xdc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr82, $sgpr32, 692, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.52, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x18, 0x0f, 0x90, 0xd2, 0x18, 0x16, 0xec, 0x80, 0xda, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr83, $sgpr32, 688, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.53, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x18, 0x0f, 0x90, 0xd3, 0x18, 0x16, 0xec, 0x80, 0xd8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr84, $sgpr32, 684, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.54, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x18, 0x0f, 0x90, 0xd4, 0x18, 0x16, 0xec, 0x80, 0xd6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr85, $sgpr32, 680, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.55, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x18, 0x0f, 0x90, 0xd5, 0x18, 0x16, 0xec, 0x80, 0xd4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr86, $sgpr32, 676, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.56, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x18, 0x0f, 0x90, 0xd6, 0x18, 0x16, 0xec, 0x80, 0xd2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr87, $sgpr32, 672, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.57, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x18, 0x0f, 0x90, 0xd7, 0x18, 0x16, 0xec, 0x80, 0xd0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr88, $sgpr32, 668, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.58, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x18, 0x0f, 0x90, 0xd8, 0x18, 0x16, 0xec, 0x80, 0xce, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr89, $sgpr32, 664, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.59, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x18, 0x0f, 0x90, 0xd9, 0x18, 0x16, 0xec, 0x80, 0xcc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr90, $sgpr32, 660, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.60, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x18, 0x0f, 0x90, 0xda, 0x18, 0x16, 0xec, 0x80, 0xca, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr91, $sgpr32, 656, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.61, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x18, 0x0f, 0x90, 0xdb, 0x18, 0x16, 0xec, 0x80, 0xc8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr92, $sgpr32, 652, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.62, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x18, 0x0f, 0x90, 0xdc, 0x18, 0x16, 0xec, 0x80, 0xc6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr93, $sgpr32, 648, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.63, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x18, 0x0f, 0x90, 0xdd, 0x18, 0x16, 0xec, 0x80, 0xc4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr94, $sgpr32, 644, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.64, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x18, 0x0f, 0x90, 0xde, 0x18, 0x16, 0xec, 0x80, 0xc2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr95, $sgpr32, 640, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.65, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x18, 0x0f, 0x90, 0xdf, 0x18, 0x16, 0xec, 0x80, 0xc0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr96, $sgpr32, 636, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.66, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x18, 0x0f, 0x90, 0xe0, 0x18, 0x16, 0xec, 0x80, 0xbe, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr97, $sgpr32, 632, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.67, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x18, 0x0f, 0x90, 0xe1, 0x18, 0x16, 0xec, 0x80, 0xbc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr98, $sgpr32, 628, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.68, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x18, 0x0f, 0x90, 0xe2, 0x18, 0x16, 0xec, 0x80, 0xba, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr99, $sgpr32, 624, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.69, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x18, 0x0f, 0x90, 0xe3, 0x18, 0x16, 0xec, 0x80, 0xb8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr100, $sgpr32, 620, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.70, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x18, 0x0f, 0x90, 0xe4, 0x18, 0x16, 0xec, 0x80, 0xb6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr101, $sgpr32, 616, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.71, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x18, 0x0f, 0x90, 0xe5, 0x18, 0x16, 0xec, 0x80, 0xb4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr102, $sgpr32, 612, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.72, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x18, 0x0f, 0x90, 0xe6, 0x18, 0x16, 0xec, 0x80, 0xb2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr103, $sgpr32, 608, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.73, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x18, 0x0f, 0x90, 0xe7, 0x18, 0x16, 0xec, 0x80, 0xb0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr104, $sgpr32, 604, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.74, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x18, 0x0f, 0x90, 0xe8, 0x18, 0x16, 0xec, 0x80, 0xae, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr105, $sgpr32, 600, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.75, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x18, 0x0f, 0x90, 0xe9, 0x18, 0x16, 0xec, 0x80, 0xac, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr106, $sgpr32, 596, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.76, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x18, 0x0f, 0x90, 0xea, 0x18, 0x16, 0xec, 0x80, 0xaa, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr107, $sgpr32, 592, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.77, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x18, 0x0f, 0x90, 0xeb, 0x18, 0x16, 0xec, 0x80, 0xa8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr108, $sgpr32, 588, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.78, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x18, 0x0f, 0x90, 0xec, 0x18, 0x16, 0xec, 0x80, 0xa6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr109, $sgpr32, 584, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.79, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x18, 0x0f, 0x90, 0xed, 0x18, 0x16, 0xec, 0x80, 0xa4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr110, $sgpr32, 580, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.80, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x18, 0x0f, 0x90, 0xee, 0x18, 0x16, 0xec, 0x80, 0xa2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr111, $sgpr32, 576, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.81, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x18, 0x0f, 0x90, 0xef, 0x18, 0x16, 0xec, 0x80, 0xa0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr112, $sgpr32, 572, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.82, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x18, 0x0f, 0x90, 0xf0, 0x18, 0x16, 0xec, 0x80, 0x9e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr113, $sgpr32, 568, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.83, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x18, 0x0f, 0x90, 0xf1, 0x18, 0x16, 0xec, 0x80, 0x9c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr114, $sgpr32, 564, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.84, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x18, 0x0f, 0x90, 0xf2, 0x18, 0x16, 0xec, 0x80, 0x9a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr115, $sgpr32, 560, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.85, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x18, 0x0f, 0x90, 0xf3, 0x18, 0x16, 0xec, 0x80, 0x98, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr116, $sgpr32, 556, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.86, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x18, 0x0f, 0x90, 0xf4, 0x18, 0x16, 0xec, 0x80, 0x96, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr117, $sgpr32, 552, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.87, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x18, 0x0f, 0x90, 0xf5, 0x18, 0x16, 0xec, 0x80, 0x94, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr118, $sgpr32, 548, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.88, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x18, 0x0f, 0x90, 0xf6, 0x18, 0x16, 0xec, 0x80, 0x92, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr119, $sgpr32, 544, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.89, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x18, 0x0f, 0x90, 0xf7, 0x18, 0x16, 0xec, 0x80, 0x90, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr120, $sgpr32, 540, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.90, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x18, 0x0f, 0x90, 0xf8, 0x18, 0x16, 0xec, 0x80, 0x8e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr121, $sgpr32, 536, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.91, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x18, 0x0f, 0x90, 0xf9, 0x18, 0x16, 0xec, 0x80, 0x8c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr122, $sgpr32, 532, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.92, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x18, 0x0f, 0x90, 0xfa, 0x18, 0x16, 0xec, 0x80, 0x8a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr123, $sgpr32, 528, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.93, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x18, 0x0f, 0x90, 0xfb, 0x18, 0x16, 0xec, 0x80, 0x88, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr124, $sgpr32, 524, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.94, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x18, 0x0f, 0x90, 0xfc, 0x18, 0x16, 0xec, 0x80, 0x86, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr125, $sgpr32, 520, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.95, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x18, 0x0f, 0x90, 0xfd, 0x18, 0x16, 0xec, 0x80, 0x84, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr126, $sgpr32, 516, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.96, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x18, 0x0f, 0x90, 0xfe, 0x18, 0x16, 0xec, 0x80, 0x82, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr127, $sgpr32, 512, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.97, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x18, 0x0f, 0x90, 0xff, 0x18, 0x16, 0xec, 0x80, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr128, $sgpr32, 508, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.98, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x80, 0x19, 0x0f, 0x90, 0x80, 0x19, 0x16, 0xec, 0x80, 0xfe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr129, $sgpr32, 504, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.99, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x81, 0x19, 0x0f, 0x90, 0x81, 0x19, 0x16, 0xec, 0x80, 0xfc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr130, $sgpr32, 500, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.100, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x82, 0x19, 0x0f, 0x90, 0x82, 0x19, 0x16, 0xec, 0x80, 0xfa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr131, $sgpr32, 496, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.101, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x83, 0x19, 0x0f, 0x90, 0x83, 0x19, 0x16, 0xec, 0x80, 0xf8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr132, $sgpr32, 492, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.102, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x84, 0x19, 0x0f, 0x90, 0x84, 0x19, 0x16, 0xec, 0x80, 0xf6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr133, $sgpr32, 488, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.103, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x85, 0x19, 0x0f, 0x90, 0x85, 0x19, 0x16, 0xec, 0x80, 0xf4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr134, $sgpr32, 484, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.104, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x86, 0x19, 0x0f, 0x90, 0x86, 0x19, 0x16, 0xec, 0x80, 0xf2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr135, $sgpr32, 480, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.105, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x87, 0x19, 0x0f, 0x90, 0x87, 0x19, 0x16, 0xec, 0x80, 0xf0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr136, $sgpr32, 476, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.106, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x88, 0x19, 0x0f, 0x90, 0x88, 0x19, 0x16, 0xec, 0x80, 0xee, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr137, $sgpr32, 472, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.107, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x89, 0x19, 0x0f, 0x90, 0x89, 0x19, 0x16, 0xec, 0x80, 0xec, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr138, $sgpr32, 468, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.108, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8a, 0x19, 0x0f, 0x90, 0x8a, 0x19, 0x16, 0xec, 0x80, 0xea, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr139, $sgpr32, 464, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.109, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8b, 0x19, 0x0f, 0x90, 0x8b, 0x19, 0x16, 0xec, 0x80, 0xe8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr140, $sgpr32, 460, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.110, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8c, 0x19, 0x0f, 0x90, 0x8c, 0x19, 0x16, 0xec, 0x80, 0xe6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr141, $sgpr32, 456, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.111, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8d, 0x19, 0x0f, 0x90, 0x8d, 0x19, 0x16, 0xec, 0x80, 0xe4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr142, $sgpr32, 452, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.112, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8e, 0x19, 0x0f, 0x90, 0x8e, 0x19, 0x16, 0xec, 0x80, 0xe2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr143, $sgpr32, 448, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.113, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8f, 0x19, 0x0f, 0x90, 0x8f, 0x19, 0x16, 0xec, 0x80, 0xe0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr144, $sgpr32, 444, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.114, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x90, 0x19, 0x0f, 0x90, 0x90, 0x19, 0x16, 0xec, 0x80, 0xde, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr145, $sgpr32, 440, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.115, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x91, 0x19, 0x0f, 0x90, 0x91, 0x19, 0x16, 0xec, 0x80, 0xdc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr146, $sgpr32, 436, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.116, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x92, 0x19, 0x0f, 0x90, 0x92, 0x19, 0x16, 0xec, 0x80, 0xda, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr147, $sgpr32, 432, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.117, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x93, 0x19, 0x0f, 0x90, 0x93, 0x19, 0x16, 0xec, 0x80, 0xd8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr148, $sgpr32, 428, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.118, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x94, 0x19, 0x0f, 0x90, 0x94, 0x19, 0x16, 0xec, 0x80, 0xd6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr149, $sgpr32, 424, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.119, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x95, 0x19, 0x0f, 0x90, 0x95, 0x19, 0x16, 0xec, 0x80, 0xd4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr150, $sgpr32, 420, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.120, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x96, 0x19, 0x0f, 0x90, 0x96, 0x19, 0x16, 0xec, 0x80, 0xd2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr151, $sgpr32, 416, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.121, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x97, 0x19, 0x0f, 0x90, 0x97, 0x19, 0x16, 0xec, 0x80, 0xd0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr152, $sgpr32, 412, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.122, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x98, 0x19, 0x0f, 0x90, 0x98, 0x19, 0x16, 0xec, 0x80, 0xce, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr153, $sgpr32, 408, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.123, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x99, 0x19, 0x0f, 0x90, 0x99, 0x19, 0x16, 0xec, 0x80, 0xcc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr154, $sgpr32, 404, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.124, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9a, 0x19, 0x0f, 0x90, 0x9a, 0x19, 0x16, 0xec, 0x80, 0xca, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr155, $sgpr32, 400, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.125, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9b, 0x19, 0x0f, 0x90, 0x9b, 0x19, 0x16, 0xec, 0x80, 0xc8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr156, $sgpr32, 396, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.126, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9c, 0x19, 0x0f, 0x90, 0x9c, 0x19, 0x16, 0xec, 0x80, 0xc6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr157, $sgpr32, 392, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.127, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9d, 0x19, 0x0f, 0x90, 0x9d, 0x19, 0x16, 0xec, 0x80, 0xc4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr158, $sgpr32, 388, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.128, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9e, 0x19, 0x0f, 0x90, 0x9e, 0x19, 0x16, 0xec, 0x80, 0xc2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr159, $sgpr32, 384, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.129, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9f, 0x19, 0x0f, 0x90, 0x9f, 0x19, 0x16, 0xec, 0x80, 0xc0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr160, $sgpr32, 380, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.130, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa0, 0x19, 0x0f, 0x90, 0xa0, 0x19, 0x16, 0xec, 0x80, 0xbe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr161, $sgpr32, 376, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.131, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa1, 0x19, 0x0f, 0x90, 0xa1, 0x19, 0x16, 0xec, 0x80, 0xbc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr162, $sgpr32, 372, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.132, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa2, 0x19, 0x0f, 0x90, 0xa2, 0x19, 0x16, 0xec, 0x80, 0xba, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr163, $sgpr32, 368, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.133, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa3, 0x19, 0x0f, 0x90, 0xa3, 0x19, 0x16, 0xec, 0x80, 0xb8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr164, $sgpr32, 364, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.134, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa4, 0x19, 0x0f, 0x90, 0xa4, 0x19, 0x16, 0xec, 0x80, 0xb6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr165, $sgpr32, 360, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.135, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa5, 0x19, 0x0f, 0x90, 0xa5, 0x19, 0x16, 0xec, 0x80, 0xb4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr166, $sgpr32, 356, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.136, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa6, 0x19, 0x0f, 0x90, 0xa6, 0x19, 0x16, 0xec, 0x80, 0xb2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr167, $sgpr32, 352, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.137, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa7, 0x19, 0x0f, 0x90, 0xa7, 0x19, 0x16, 0xec, 0x80, 0xb0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr168, $sgpr32, 348, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.138, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa8, 0x19, 0x0f, 0x90, 0xa8, 0x19, 0x16, 0xec, 0x80, 0xae, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr169, $sgpr32, 344, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.139, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa9, 0x19, 0x0f, 0x90, 0xa9, 0x19, 0x16, 0xec, 0x80, 0xac, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr170, $sgpr32, 340, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.140, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaa, 0x19, 0x0f, 0x90, 0xaa, 0x19, 0x16, 0xec, 0x80, 0xaa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr171, $sgpr32, 336, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.141, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xab, 0x19, 0x0f, 0x90, 0xab, 0x19, 0x16, 0xec, 0x80, 0xa8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr172, $sgpr32, 332, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.142, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xac, 0x19, 0x0f, 0x90, 0xac, 0x19, 0x16, 0xec, 0x80, 0xa6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr173, $sgpr32, 328, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.143, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xad, 0x19, 0x0f, 0x90, 0xad, 0x19, 0x16, 0xec, 0x80, 0xa4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr174, $sgpr32, 324, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.144, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xae, 0x19, 0x0f, 0x90, 0xae, 0x19, 0x16, 0xec, 0x80, 0xa2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr175, $sgpr32, 320, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.145, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaf, 0x19, 0x0f, 0x90, 0xaf, 0x19, 0x16, 0xec, 0x80, 0xa0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr176, $sgpr32, 316, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.146, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb0, 0x19, 0x0f, 0x90, 0xb0, 0x19, 0x16, 0xec, 0x80, 0x9e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr177, $sgpr32, 312, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.147, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb1, 0x19, 0x0f, 0x90, 0xb1, 0x19, 0x16, 0xec, 0x80, 0x9c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr178, $sgpr32, 308, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.148, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb2, 0x19, 0x0f, 0x90, 0xb2, 0x19, 0x16, 0xec, 0x80, 0x9a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr179, $sgpr32, 304, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.149, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb3, 0x19, 0x0f, 0x90, 0xb3, 0x19, 0x16, 0xec, 0x80, 0x98, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr180, $sgpr32, 300, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.150, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb4, 0x19, 0x0f, 0x90, 0xb4, 0x19, 0x16, 0xec, 0x80, 0x96, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr181, $sgpr32, 296, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.151, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb5, 0x19, 0x0f, 0x90, 0xb5, 0x19, 0x16, 0xec, 0x80, 0x94, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr182, $sgpr32, 292, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.152, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb6, 0x19, 0x0f, 0x90, 0xb6, 0x19, 0x16, 0xec, 0x80, 0x92, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr183, $sgpr32, 288, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.153, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb7, 0x19, 0x0f, 0x90, 0xb7, 0x19, 0x16, 0xec, 0x80, 0x90, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr184, $sgpr32, 284, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.154, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb8, 0x19, 0x0f, 0x90, 0xb8, 0x19, 0x16, 0xec, 0x80, 0x8e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr185, $sgpr32, 280, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.155, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb9, 0x19, 0x0f, 0x90, 0xb9, 0x19, 0x16, 0xec, 0x80, 0x8c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr186, $sgpr32, 276, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.156, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xba, 0x19, 0x0f, 0x90, 0xba, 0x19, 0x16, 0xec, 0x80, 0x8a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr187, $sgpr32, 272, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.157, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbb, 0x19, 0x0f, 0x90, 0xbb, 0x19, 0x16, 0xec, 0x80, 0x88, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr188, $sgpr32, 268, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.158, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbc, 0x19, 0x0f, 0x90, 0xbc, 0x19, 0x16, 0xec, 0x80, 0x86, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr189, $sgpr32, 264, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.159, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbd, 0x19, 0x0f, 0x90, 0xbd, 0x19, 0x16, 0xec, 0x80, 0x84, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr190, $sgpr32, 260, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.160, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbe, 0x19, 0x0f, 0x90, 0xbe, 0x19, 0x16, 0xec, 0x80, 0x82, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr191, $sgpr32, 256, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.161, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbf, 0x19, 0x0f, 0x90, 0xbf, 0x19, 0x16, 0xec, 0x80, 0x80, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr192, $sgpr32, 252, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.162, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc0, 0x19, 0x0e, 0x90, 0xc0, 0x19, 0x16, 0xec, 0x80, 0x7e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr193, $sgpr32, 248, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.163, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc1, 0x19, 0x0e, 0x90, 0xc1, 0x19, 0x16, 0xec, 0x80, 0x7c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr194, $sgpr32, 244, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.164, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc2, 0x19, 0x0e, 0x90, 0xc2, 0x19, 0x16, 0xec, 0x80, 0x7a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr195, $sgpr32, 240, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.165, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc3, 0x19, 0x0e, 0x90, 0xc3, 0x19, 0x16, 0xec, 0x80, 0x78, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr196, $sgpr32, 236, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.166, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc4, 0x19, 0x0e, 0x90, 0xc4, 0x19, 0x16, 0xec, 0x80, 0x76, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr197, $sgpr32, 232, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.167, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc5, 0x19, 0x0e, 0x90, 0xc5, 0x19, 0x16, 0xec, 0x80, 0x74, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr198, $sgpr32, 228, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.168, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc6, 0x19, 0x0e, 0x90, 0xc6, 0x19, 0x16, 0xec, 0x80, 0x72, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr199, $sgpr32, 224, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.169, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc7, 0x19, 0x0e, 0x90, 0xc7, 0x19, 0x16, 0xec, 0x80, 0x70, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr200, $sgpr32, 220, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.170, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc8, 0x19, 0x0e, 0x90, 0xc8, 0x19, 0x16, 0xec, 0x80, 0x6e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr201, $sgpr32, 216, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.171, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc9, 0x19, 0x0e, 0x90, 0xc9, 0x19, 0x16, 0xec, 0x80, 0x6c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr202, $sgpr32, 212, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.172, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xca, 0x19, 0x0e, 0x90, 0xca, 0x19, 0x16, 0xec, 0x80, 0x6a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr203, $sgpr32, 208, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.173, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcb, 0x19, 0x0e, 0x90, 0xcb, 0x19, 0x16, 0xec, 0x80, 0x68, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr204, $sgpr32, 204, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.174, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcc, 0x19, 0x0e, 0x90, 0xcc, 0x19, 0x16, 0xec, 0x80, 0x66, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr205, $sgpr32, 200, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.175, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcd, 0x19, 0x0e, 0x90, 0xcd, 0x19, 0x16, 0xec, 0x80, 0x64, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr206, $sgpr32, 196, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.176, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xce, 0x19, 0x0e, 0x90, 0xce, 0x19, 0x16, 0xec, 0x80, 0x62, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr207, $sgpr32, 192, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.177, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcf, 0x19, 0x0e, 0x90, 0xcf, 0x19, 0x16, 0xec, 0x80, 0x60, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr208, $sgpr32, 188, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.178, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x19, 0x0e, 0x90, 0xd0, 0x19, 0x16, 0xec, 0x80, 0x5e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr209, $sgpr32, 184, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.179, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x19, 0x0e, 0x90, 0xd1, 0x19, 0x16, 0xec, 0x80, 0x5c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr210, $sgpr32, 180, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.180, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x19, 0x0e, 0x90, 0xd2, 0x19, 0x16, 0xec, 0x80, 0x5a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr211, $sgpr32, 176, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.181, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x19, 0x0e, 0x90, 0xd3, 0x19, 0x16, 0xec, 0x80, 0x58, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr212, $sgpr32, 172, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.182, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x19, 0x0e, 0x90, 0xd4, 0x19, 0x16, 0xec, 0x80, 0x56, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr213, $sgpr32, 168, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.183, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x19, 0x0e, 0x90, 0xd5, 0x19, 0x16, 0xec, 0x80, 0x54, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr214, $sgpr32, 164, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.184, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x19, 0x0e, 0x90, 0xd6, 0x19, 0x16, 0xec, 0x80, 0x52, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr215, $sgpr32, 160, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.185, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x19, 0x0e, 0x90, 0xd7, 0x19, 0x16, 0xec, 0x80, 0x50, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr216, $sgpr32, 156, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.186, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x19, 0x0e, 0x90, 0xd8, 0x19, 0x16, 0xec, 0x80, 0x4e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr217, $sgpr32, 152, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.187, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x19, 0x0e, 0x90, 0xd9, 0x19, 0x16, 0xec, 0x80, 0x4c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr218, $sgpr32, 148, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.188, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x19, 0x0e, 0x90, 0xda, 0x19, 0x16, 0xec, 0x80, 0x4a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr219, $sgpr32, 144, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.189, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x19, 0x0e, 0x90, 0xdb, 0x19, 0x16, 0xec, 0x80, 0x48, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr220, $sgpr32, 140, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.190, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x19, 0x0e, 0x90, 0xdc, 0x19, 0x16, 0xec, 0x80, 0x46, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr221, $sgpr32, 136, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.191, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x19, 0x0e, 0x90, 0xdd, 0x19, 0x16, 0xec, 0x80, 0x44, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr222, $sgpr32, 132, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.192, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x19, 0x0e, 0x90, 0xde, 0x19, 0x16, 0xec, 0x80, 0x42, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr223, $sgpr32, 128, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.193, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x19, 0x0e, 0x90, 0xdf, 0x19, 0x16, 0xec, 0x80, 0x40, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr224, $sgpr32, 124, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.194, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x19, 0x0e, 0x90, 0xe0, 0x19, 0x16, 0xec, 0x80, 0x3e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr225, $sgpr32, 120, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.195, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x19, 0x0e, 0x90, 0xe1, 0x19, 0x16, 0xec, 0x80, 0x3c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr226, $sgpr32, 116, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.196, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x19, 0x0e, 0x90, 0xe2, 0x19, 0x16, 0xec, 0x80, 0x3a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr227, $sgpr32, 112, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.197, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x19, 0x0e, 0x90, 0xe3, 0x19, 0x16, 0xec, 0x80, 0x38, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr228, $sgpr32, 108, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.198, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x19, 0x0e, 0x90, 0xe4, 0x19, 0x16, 0xec, 0x80, 0x36, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr229, $sgpr32, 104, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.199, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x19, 0x0e, 0x90, 0xe5, 0x19, 0x16, 0xec, 0x80, 0x34, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr230, $sgpr32, 100, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.200, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x19, 0x0e, 0x90, 0xe6, 0x19, 0x16, 0xec, 0x80, 0x32, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr231, $sgpr32, 96, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.201, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x19, 0x0e, 0x90, 0xe7, 0x19, 0x16, 0xec, 0x80, 0x30, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr232, $sgpr32, 92, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.202, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x19, 0x0e, 0x90, 0xe8, 0x19, 0x16, 0xec, 0x80, 0x2e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr233, $sgpr32, 88, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.203, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x19, 0x0e, 0x90, 0xe9, 0x19, 0x16, 0xec, 0x80, 0x2c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr234, $sgpr32, 84, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.204, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x19, 0x0e, 0x90, 0xea, 0x19, 0x16, 0xec, 0x80, 0x2a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr235, $sgpr32, 80, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.205, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x19, 0x0e, 0x90, 0xeb, 0x19, 0x16, 0xec, 0x80, 0x28, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr236, $sgpr32, 76, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.206, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x19, 0x0e, 0x90, 0xec, 0x19, 0x16, 0xec, 0x80, 0x26, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr237, $sgpr32, 72, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.207, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x19, 0x0e, 0x90, 0xed, 0x19, 0x16, 0xec, 0x80, 0x24, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr238, $sgpr32, 68, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.208, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x19, 0x0e, 0x90, 0xee, 0x19, 0x16, 0xec, 0x80, 0x22, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr239, $sgpr32, 64, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.209, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x19, 0x0e, 0x90, 0xef, 0x19, 0x16, 0xec, 0x80, 0x20, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr240, $sgpr32, 60, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.210, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x19, 0x0e, 0x90, 0xf0, 0x19, 0x16, 0xec, 0x80, 0x1e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr241, $sgpr32, 56, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.211, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x19, 0x0e, 0x90, 0xf1, 0x19, 0x16, 0xec, 0x80, 0x1c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr242, $sgpr32, 52, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.212, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x19, 0x0e, 0x90, 0xf2, 0x19, 0x16, 0xec, 0x80, 0x1a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr243, $sgpr32, 48, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.213, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x19, 0x0e, 0x90, 0xf3, 0x19, 0x16, 0xec, 0x80, 0x18, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr244, $sgpr32, 44, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.214, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x19, 0x0e, 0x90, 0xf4, 0x19, 0x16, 0xec, 0x80, 0x16, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr245, $sgpr32, 40, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.215, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x19, 0x0e, 0x90, 0xf5, 0x19, 0x16, 0xec, 0x80, 0x14, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr246, $sgpr32, 36, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.216, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x19, 0x0e, 0x90, 0xf6, 0x19, 0x16, 0xec, 0x80, 0x12, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr247, $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.217, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x19, 0x0e, 0x90, 0xf7, 0x19, 0x16, 0xec, 0x80, 0x10, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr248, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.218, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x19, 0x0e, 0x90, 0xf8, 0x19, 0x16, 0xec, 0x80, 0x0e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr249, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.219, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x19, 0x0e, 0x90, 0xf9, 0x19, 0x16, 0xec, 0x80, 0x0c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr250, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.220, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x19, 0x0e, 0x90, 0xfa, 0x19, 0x16, 0xec, 0x80, 0x0a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr251, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.221, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x19, 0x0e, 0x90, 0xfb, 0x19, 0x16, 0xec, 0x80, 0x08, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr252, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.222, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x19, 0x0e, 0x90, 0xfc, 0x19, 0x16, 0xec, 0x80, 0x06, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr253, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.223, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x19, 0x0e, 0x90, 0xfd, 0x19, 0x16, 0xec, 0x80, 0x04, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr254, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.224, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x19, 0x0e, 0x90, 0xfe, 0x19, 0x16, 0xec, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr255, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.225, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x19, 0x0d, 0x90, 0xff, 0x19, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: S_CMP_EQ_U32 0, 0, implicit-def $scc ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 704, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.226, addrspace 5) ; GFX90A-FLATSCR-NEXT: $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec @@ -7699,229 +9491,453 @@ ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr28_lo16 ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr29_lo16 ; GFX90A-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr32, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr32_lo16, $vgpr0_lo16 ; GFX90A-NEXT: $vgpr1 = V_ACCVGPR_READ_B32_e64 killed $agpr33, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr33_lo16, $vgpr1_lo16 ; GFX90A-NEXT: $vgpr2 = V_ACCVGPR_READ_B32_e64 killed $agpr34, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr34_lo16, $vgpr2_lo16 ; GFX90A-NEXT: $vgpr3 = V_ACCVGPR_READ_B32_e64 killed $agpr35, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr35_lo16, $vgpr3_lo16 ; GFX90A-NEXT: $vgpr4 = V_ACCVGPR_READ_B32_e64 killed $agpr36, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr36_lo16, $vgpr4_lo16 ; GFX90A-NEXT: $vgpr5 = V_ACCVGPR_READ_B32_e64 killed $agpr37, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr37_lo16, $vgpr5_lo16 ; GFX90A-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 killed $agpr38, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr38_lo16, $vgpr6_lo16 ; GFX90A-NEXT: $vgpr7 = V_ACCVGPR_READ_B32_e64 killed $agpr39, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr39_lo16, $vgpr7_lo16 ; GFX90A-NEXT: $vgpr8 = V_ACCVGPR_READ_B32_e64 killed $agpr40, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr40_lo16, $vgpr8_lo16 ; GFX90A-NEXT: $vgpr9 = V_ACCVGPR_READ_B32_e64 killed $agpr41, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr41_lo16, $vgpr9_lo16 ; GFX90A-NEXT: $vgpr10 = V_ACCVGPR_READ_B32_e64 killed $agpr42, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr42_lo16, $vgpr10_lo16 ; GFX90A-NEXT: $vgpr11 = V_ACCVGPR_READ_B32_e64 killed $agpr43, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr43_lo16, $vgpr11_lo16 ; GFX90A-NEXT: $vgpr12 = V_ACCVGPR_READ_B32_e64 killed $agpr44, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr44_lo16, $vgpr12_lo16 ; GFX90A-NEXT: $vgpr13 = V_ACCVGPR_READ_B32_e64 killed $agpr45, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr45_lo16, $vgpr13_lo16 ; GFX90A-NEXT: $vgpr14 = V_ACCVGPR_READ_B32_e64 killed $agpr46, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr46_lo16, $vgpr14_lo16 ; GFX90A-NEXT: $vgpr15 = V_ACCVGPR_READ_B32_e64 killed $agpr47, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr47_lo16, $vgpr15_lo16 ; GFX90A-NEXT: $vgpr16 = V_ACCVGPR_READ_B32_e64 killed $agpr48, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr48_lo16, $vgpr16_lo16 ; GFX90A-NEXT: $vgpr17 = V_ACCVGPR_READ_B32_e64 killed $agpr49, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr49_lo16, $vgpr17_lo16 ; GFX90A-NEXT: $vgpr18 = V_ACCVGPR_READ_B32_e64 killed $agpr50, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr50_lo16, $vgpr18_lo16 ; GFX90A-NEXT: $vgpr19 = V_ACCVGPR_READ_B32_e64 killed $agpr51, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr51_lo16, $vgpr19_lo16 ; GFX90A-NEXT: $vgpr20 = V_ACCVGPR_READ_B32_e64 killed $agpr52, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr52_lo16, $vgpr20_lo16 ; GFX90A-NEXT: $vgpr21 = V_ACCVGPR_READ_B32_e64 killed $agpr53, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr53_lo16, $vgpr21_lo16 ; GFX90A-NEXT: $vgpr22 = V_ACCVGPR_READ_B32_e64 killed $agpr54, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr54_lo16, $vgpr22_lo16 ; GFX90A-NEXT: $vgpr23 = V_ACCVGPR_READ_B32_e64 killed $agpr55, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr55_lo16, $vgpr23_lo16 ; GFX90A-NEXT: $vgpr24 = V_ACCVGPR_READ_B32_e64 killed $agpr56, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr56_lo16, $vgpr24_lo16 ; GFX90A-NEXT: $vgpr25 = V_ACCVGPR_READ_B32_e64 killed $agpr57, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr57_lo16, $vgpr25_lo16 ; GFX90A-NEXT: $vgpr26 = V_ACCVGPR_READ_B32_e64 killed $agpr58, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr58_lo16, $vgpr26_lo16 ; GFX90A-NEXT: $vgpr27 = V_ACCVGPR_READ_B32_e64 killed $agpr59, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr59_lo16, $vgpr27_lo16 ; GFX90A-NEXT: $vgpr28 = V_ACCVGPR_READ_B32_e64 killed $agpr60, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr60_lo16, $vgpr28_lo16 ; GFX90A-NEXT: $vgpr29 = V_ACCVGPR_READ_B32_e64 killed $agpr61, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr61_lo16, $vgpr29_lo16 ; GFX90A-NEXT: $vgpr30 = V_ACCVGPR_READ_B32_e64 killed $agpr62, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr62_lo16, $vgpr30_lo16 ; GFX90A-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 killed $agpr63, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr63_lo16, $vgpr31_lo16 ; GFX90A-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr64, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr64_lo16, $vgpr32_lo16 ; GFX90A-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 killed $agpr65, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr65_lo16, $vgpr33_lo16 ; GFX90A-NEXT: $vgpr34 = V_ACCVGPR_READ_B32_e64 killed $agpr66, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr66_lo16, $vgpr34_lo16 ; GFX90A-NEXT: $vgpr35 = V_ACCVGPR_READ_B32_e64 killed $agpr67, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr67_lo16, $vgpr35_lo16 ; GFX90A-NEXT: $vgpr36 = V_ACCVGPR_READ_B32_e64 killed $agpr68, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr68_lo16, $vgpr36_lo16 ; GFX90A-NEXT: $vgpr37 = V_ACCVGPR_READ_B32_e64 killed $agpr69, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr69_lo16, $vgpr37_lo16 ; GFX90A-NEXT: $vgpr38 = V_ACCVGPR_READ_B32_e64 killed $agpr70, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr70_lo16, $vgpr38_lo16 ; GFX90A-NEXT: $vgpr39 = V_ACCVGPR_READ_B32_e64 killed $agpr71, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr71_lo16, $vgpr39_lo16 ; GFX90A-NEXT: $vgpr48 = V_ACCVGPR_READ_B32_e64 killed $agpr72, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr72_lo16, $vgpr48_lo16 ; GFX90A-NEXT: $vgpr49 = V_ACCVGPR_READ_B32_e64 killed $agpr73, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr73_lo16, $vgpr49_lo16 ; GFX90A-NEXT: $vgpr50 = V_ACCVGPR_READ_B32_e64 killed $agpr74, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr74_lo16, $vgpr50_lo16 ; GFX90A-NEXT: $vgpr51 = V_ACCVGPR_READ_B32_e64 killed $agpr75, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr75_lo16, $vgpr51_lo16 ; GFX90A-NEXT: $vgpr52 = V_ACCVGPR_READ_B32_e64 killed $agpr76, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr76_lo16, $vgpr52_lo16 ; GFX90A-NEXT: $vgpr53 = V_ACCVGPR_READ_B32_e64 killed $agpr77, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr77_lo16, $vgpr53_lo16 ; GFX90A-NEXT: $vgpr54 = V_ACCVGPR_READ_B32_e64 killed $agpr78, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr78_lo16, $vgpr54_lo16 ; GFX90A-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 killed $agpr79, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr79_lo16, $vgpr55_lo16 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr80, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 700, 0, 0, 0, implicit $exec :: (store (s32) into %stack.50, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x18, 0x0f, 0x90, 0xd0, 0x18, 0x16, 0xec, 0x80, 0xde, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr81, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 696, 0, 0, 0, implicit $exec :: (store (s32) into %stack.51, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x18, 0x0f, 0x90, 0xd1, 0x18, 0x16, 0xec, 0x80, 0xdc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr82, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 692, 0, 0, 0, implicit $exec :: (store (s32) into %stack.52, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x18, 0x0f, 0x90, 0xd2, 0x18, 0x16, 0xec, 0x80, 0xda, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr83, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 688, 0, 0, 0, implicit $exec :: (store (s32) into %stack.53, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x18, 0x0f, 0x90, 0xd3, 0x18, 0x16, 0xec, 0x80, 0xd8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr84, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 684, 0, 0, 0, implicit $exec :: (store (s32) into %stack.54, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x18, 0x0f, 0x90, 0xd4, 0x18, 0x16, 0xec, 0x80, 0xd6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr85, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 680, 0, 0, 0, implicit $exec :: (store (s32) into %stack.55, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x18, 0x0f, 0x90, 0xd5, 0x18, 0x16, 0xec, 0x80, 0xd4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr86, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 676, 0, 0, 0, implicit $exec :: (store (s32) into %stack.56, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x18, 0x0f, 0x90, 0xd6, 0x18, 0x16, 0xec, 0x80, 0xd2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr87, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 672, 0, 0, 0, implicit $exec :: (store (s32) into %stack.57, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x18, 0x0f, 0x90, 0xd7, 0x18, 0x16, 0xec, 0x80, 0xd0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr88, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 668, 0, 0, 0, implicit $exec :: (store (s32) into %stack.58, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x18, 0x0f, 0x90, 0xd8, 0x18, 0x16, 0xec, 0x80, 0xce, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr89, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 664, 0, 0, 0, implicit $exec :: (store (s32) into %stack.59, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x18, 0x0f, 0x90, 0xd9, 0x18, 0x16, 0xec, 0x80, 0xcc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr90, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 660, 0, 0, 0, implicit $exec :: (store (s32) into %stack.60, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x18, 0x0f, 0x90, 0xda, 0x18, 0x16, 0xec, 0x80, 0xca, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr91, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 656, 0, 0, 0, implicit $exec :: (store (s32) into %stack.61, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x18, 0x0f, 0x90, 0xdb, 0x18, 0x16, 0xec, 0x80, 0xc8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr92, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 652, 0, 0, 0, implicit $exec :: (store (s32) into %stack.62, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x18, 0x0f, 0x90, 0xdc, 0x18, 0x16, 0xec, 0x80, 0xc6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr93, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 648, 0, 0, 0, implicit $exec :: (store (s32) into %stack.63, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x18, 0x0f, 0x90, 0xdd, 0x18, 0x16, 0xec, 0x80, 0xc4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr94, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 644, 0, 0, 0, implicit $exec :: (store (s32) into %stack.64, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x18, 0x0f, 0x90, 0xde, 0x18, 0x16, 0xec, 0x80, 0xc2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr95, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 640, 0, 0, 0, implicit $exec :: (store (s32) into %stack.65, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x18, 0x0f, 0x90, 0xdf, 0x18, 0x16, 0xec, 0x80, 0xc0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr96, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 636, 0, 0, 0, implicit $exec :: (store (s32) into %stack.66, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x18, 0x0f, 0x90, 0xe0, 0x18, 0x16, 0xec, 0x80, 0xbe, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr97, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 632, 0, 0, 0, implicit $exec :: (store (s32) into %stack.67, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x18, 0x0f, 0x90, 0xe1, 0x18, 0x16, 0xec, 0x80, 0xbc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr98, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 628, 0, 0, 0, implicit $exec :: (store (s32) into %stack.68, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x18, 0x0f, 0x90, 0xe2, 0x18, 0x16, 0xec, 0x80, 0xba, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr99, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 624, 0, 0, 0, implicit $exec :: (store (s32) into %stack.69, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x18, 0x0f, 0x90, 0xe3, 0x18, 0x16, 0xec, 0x80, 0xb8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr100, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 620, 0, 0, 0, implicit $exec :: (store (s32) into %stack.70, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x18, 0x0f, 0x90, 0xe4, 0x18, 0x16, 0xec, 0x80, 0xb6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr101, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 616, 0, 0, 0, implicit $exec :: (store (s32) into %stack.71, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x18, 0x0f, 0x90, 0xe5, 0x18, 0x16, 0xec, 0x80, 0xb4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr102, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 612, 0, 0, 0, implicit $exec :: (store (s32) into %stack.72, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x18, 0x0f, 0x90, 0xe6, 0x18, 0x16, 0xec, 0x80, 0xb2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr103, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 608, 0, 0, 0, implicit $exec :: (store (s32) into %stack.73, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x18, 0x0f, 0x90, 0xe7, 0x18, 0x16, 0xec, 0x80, 0xb0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr104, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 604, 0, 0, 0, implicit $exec :: (store (s32) into %stack.74, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x18, 0x0f, 0x90, 0xe8, 0x18, 0x16, 0xec, 0x80, 0xae, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr105, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 600, 0, 0, 0, implicit $exec :: (store (s32) into %stack.75, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x18, 0x0f, 0x90, 0xe9, 0x18, 0x16, 0xec, 0x80, 0xac, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr106, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 596, 0, 0, 0, implicit $exec :: (store (s32) into %stack.76, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x18, 0x0f, 0x90, 0xea, 0x18, 0x16, 0xec, 0x80, 0xaa, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr107, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 592, 0, 0, 0, implicit $exec :: (store (s32) into %stack.77, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x18, 0x0f, 0x90, 0xeb, 0x18, 0x16, 0xec, 0x80, 0xa8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr108, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 588, 0, 0, 0, implicit $exec :: (store (s32) into %stack.78, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x18, 0x0f, 0x90, 0xec, 0x18, 0x16, 0xec, 0x80, 0xa6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr109, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 584, 0, 0, 0, implicit $exec :: (store (s32) into %stack.79, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x18, 0x0f, 0x90, 0xed, 0x18, 0x16, 0xec, 0x80, 0xa4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr110, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 580, 0, 0, 0, implicit $exec :: (store (s32) into %stack.80, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x18, 0x0f, 0x90, 0xee, 0x18, 0x16, 0xec, 0x80, 0xa2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr111, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 576, 0, 0, 0, implicit $exec :: (store (s32) into %stack.81, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x18, 0x0f, 0x90, 0xef, 0x18, 0x16, 0xec, 0x80, 0xa0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr112, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 572, 0, 0, 0, implicit $exec :: (store (s32) into %stack.82, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x18, 0x0f, 0x90, 0xf0, 0x18, 0x16, 0xec, 0x80, 0x9e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr113, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 568, 0, 0, 0, implicit $exec :: (store (s32) into %stack.83, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x18, 0x0f, 0x90, 0xf1, 0x18, 0x16, 0xec, 0x80, 0x9c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr114, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 564, 0, 0, 0, implicit $exec :: (store (s32) into %stack.84, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x18, 0x0f, 0x90, 0xf2, 0x18, 0x16, 0xec, 0x80, 0x9a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr115, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 560, 0, 0, 0, implicit $exec :: (store (s32) into %stack.85, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x18, 0x0f, 0x90, 0xf3, 0x18, 0x16, 0xec, 0x80, 0x98, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr116, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 556, 0, 0, 0, implicit $exec :: (store (s32) into %stack.86, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x18, 0x0f, 0x90, 0xf4, 0x18, 0x16, 0xec, 0x80, 0x96, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr117, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 552, 0, 0, 0, implicit $exec :: (store (s32) into %stack.87, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x18, 0x0f, 0x90, 0xf5, 0x18, 0x16, 0xec, 0x80, 0x94, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr118, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 548, 0, 0, 0, implicit $exec :: (store (s32) into %stack.88, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x18, 0x0f, 0x90, 0xf6, 0x18, 0x16, 0xec, 0x80, 0x92, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr119, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 544, 0, 0, 0, implicit $exec :: (store (s32) into %stack.89, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x18, 0x0f, 0x90, 0xf7, 0x18, 0x16, 0xec, 0x80, 0x90, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr120, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 540, 0, 0, 0, implicit $exec :: (store (s32) into %stack.90, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x18, 0x0f, 0x90, 0xf8, 0x18, 0x16, 0xec, 0x80, 0x8e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr121, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 536, 0, 0, 0, implicit $exec :: (store (s32) into %stack.91, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x18, 0x0f, 0x90, 0xf9, 0x18, 0x16, 0xec, 0x80, 0x8c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr122, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 532, 0, 0, 0, implicit $exec :: (store (s32) into %stack.92, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x18, 0x0f, 0x90, 0xfa, 0x18, 0x16, 0xec, 0x80, 0x8a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr123, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 528, 0, 0, 0, implicit $exec :: (store (s32) into %stack.93, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x18, 0x0f, 0x90, 0xfb, 0x18, 0x16, 0xec, 0x80, 0x88, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr124, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 524, 0, 0, 0, implicit $exec :: (store (s32) into %stack.94, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x18, 0x0f, 0x90, 0xfc, 0x18, 0x16, 0xec, 0x80, 0x86, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr125, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 520, 0, 0, 0, implicit $exec :: (store (s32) into %stack.95, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x18, 0x0f, 0x90, 0xfd, 0x18, 0x16, 0xec, 0x80, 0x84, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr126, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 516, 0, 0, 0, implicit $exec :: (store (s32) into %stack.96, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x18, 0x0f, 0x90, 0xfe, 0x18, 0x16, 0xec, 0x80, 0x82, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr127, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 512, 0, 0, 0, implicit $exec :: (store (s32) into %stack.97, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x18, 0x0f, 0x90, 0xff, 0x18, 0x16, 0xec, 0x80, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr128, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 508, 0, 0, 0, implicit $exec :: (store (s32) into %stack.98, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x80, 0x19, 0x0f, 0x90, 0x80, 0x19, 0x16, 0xec, 0x80, 0xfe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr129, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 504, 0, 0, 0, implicit $exec :: (store (s32) into %stack.99, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x81, 0x19, 0x0f, 0x90, 0x81, 0x19, 0x16, 0xec, 0x80, 0xfc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr130, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 500, 0, 0, 0, implicit $exec :: (store (s32) into %stack.100, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x82, 0x19, 0x0f, 0x90, 0x82, 0x19, 0x16, 0xec, 0x80, 0xfa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr131, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 496, 0, 0, 0, implicit $exec :: (store (s32) into %stack.101, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x83, 0x19, 0x0f, 0x90, 0x83, 0x19, 0x16, 0xec, 0x80, 0xf8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr132, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 492, 0, 0, 0, implicit $exec :: (store (s32) into %stack.102, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x84, 0x19, 0x0f, 0x90, 0x84, 0x19, 0x16, 0xec, 0x80, 0xf6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr133, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 488, 0, 0, 0, implicit $exec :: (store (s32) into %stack.103, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x85, 0x19, 0x0f, 0x90, 0x85, 0x19, 0x16, 0xec, 0x80, 0xf4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr134, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 484, 0, 0, 0, implicit $exec :: (store (s32) into %stack.104, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x86, 0x19, 0x0f, 0x90, 0x86, 0x19, 0x16, 0xec, 0x80, 0xf2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr135, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 480, 0, 0, 0, implicit $exec :: (store (s32) into %stack.105, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x87, 0x19, 0x0f, 0x90, 0x87, 0x19, 0x16, 0xec, 0x80, 0xf0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr136, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 476, 0, 0, 0, implicit $exec :: (store (s32) into %stack.106, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x88, 0x19, 0x0f, 0x90, 0x88, 0x19, 0x16, 0xec, 0x80, 0xee, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr137, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 472, 0, 0, 0, implicit $exec :: (store (s32) into %stack.107, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x89, 0x19, 0x0f, 0x90, 0x89, 0x19, 0x16, 0xec, 0x80, 0xec, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr138, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 468, 0, 0, 0, implicit $exec :: (store (s32) into %stack.108, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8a, 0x19, 0x0f, 0x90, 0x8a, 0x19, 0x16, 0xec, 0x80, 0xea, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr139, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 464, 0, 0, 0, implicit $exec :: (store (s32) into %stack.109, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8b, 0x19, 0x0f, 0x90, 0x8b, 0x19, 0x16, 0xec, 0x80, 0xe8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr140, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 460, 0, 0, 0, implicit $exec :: (store (s32) into %stack.110, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8c, 0x19, 0x0f, 0x90, 0x8c, 0x19, 0x16, 0xec, 0x80, 0xe6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr141, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 456, 0, 0, 0, implicit $exec :: (store (s32) into %stack.111, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8d, 0x19, 0x0f, 0x90, 0x8d, 0x19, 0x16, 0xec, 0x80, 0xe4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr142, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 452, 0, 0, 0, implicit $exec :: (store (s32) into %stack.112, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8e, 0x19, 0x0f, 0x90, 0x8e, 0x19, 0x16, 0xec, 0x80, 0xe2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr143, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 448, 0, 0, 0, implicit $exec :: (store (s32) into %stack.113, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8f, 0x19, 0x0f, 0x90, 0x8f, 0x19, 0x16, 0xec, 0x80, 0xe0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr144, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 444, 0, 0, 0, implicit $exec :: (store (s32) into %stack.114, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x90, 0x19, 0x0f, 0x90, 0x90, 0x19, 0x16, 0xec, 0x80, 0xde, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr145, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 440, 0, 0, 0, implicit $exec :: (store (s32) into %stack.115, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x91, 0x19, 0x0f, 0x90, 0x91, 0x19, 0x16, 0xec, 0x80, 0xdc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr146, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 436, 0, 0, 0, implicit $exec :: (store (s32) into %stack.116, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x92, 0x19, 0x0f, 0x90, 0x92, 0x19, 0x16, 0xec, 0x80, 0xda, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr147, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 432, 0, 0, 0, implicit $exec :: (store (s32) into %stack.117, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x93, 0x19, 0x0f, 0x90, 0x93, 0x19, 0x16, 0xec, 0x80, 0xd8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr148, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 428, 0, 0, 0, implicit $exec :: (store (s32) into %stack.118, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x94, 0x19, 0x0f, 0x90, 0x94, 0x19, 0x16, 0xec, 0x80, 0xd6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr149, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 424, 0, 0, 0, implicit $exec :: (store (s32) into %stack.119, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x95, 0x19, 0x0f, 0x90, 0x95, 0x19, 0x16, 0xec, 0x80, 0xd4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr150, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 420, 0, 0, 0, implicit $exec :: (store (s32) into %stack.120, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x96, 0x19, 0x0f, 0x90, 0x96, 0x19, 0x16, 0xec, 0x80, 0xd2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr151, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 416, 0, 0, 0, implicit $exec :: (store (s32) into %stack.121, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x97, 0x19, 0x0f, 0x90, 0x97, 0x19, 0x16, 0xec, 0x80, 0xd0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr152, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 412, 0, 0, 0, implicit $exec :: (store (s32) into %stack.122, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x98, 0x19, 0x0f, 0x90, 0x98, 0x19, 0x16, 0xec, 0x80, 0xce, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr153, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 408, 0, 0, 0, implicit $exec :: (store (s32) into %stack.123, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x99, 0x19, 0x0f, 0x90, 0x99, 0x19, 0x16, 0xec, 0x80, 0xcc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr154, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 404, 0, 0, 0, implicit $exec :: (store (s32) into %stack.124, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9a, 0x19, 0x0f, 0x90, 0x9a, 0x19, 0x16, 0xec, 0x80, 0xca, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr155, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 400, 0, 0, 0, implicit $exec :: (store (s32) into %stack.125, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9b, 0x19, 0x0f, 0x90, 0x9b, 0x19, 0x16, 0xec, 0x80, 0xc8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr156, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 396, 0, 0, 0, implicit $exec :: (store (s32) into %stack.126, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9c, 0x19, 0x0f, 0x90, 0x9c, 0x19, 0x16, 0xec, 0x80, 0xc6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr157, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 392, 0, 0, 0, implicit $exec :: (store (s32) into %stack.127, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9d, 0x19, 0x0f, 0x90, 0x9d, 0x19, 0x16, 0xec, 0x80, 0xc4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr158, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 388, 0, 0, 0, implicit $exec :: (store (s32) into %stack.128, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9e, 0x19, 0x0f, 0x90, 0x9e, 0x19, 0x16, 0xec, 0x80, 0xc2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr159, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 384, 0, 0, 0, implicit $exec :: (store (s32) into %stack.129, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9f, 0x19, 0x0f, 0x90, 0x9f, 0x19, 0x16, 0xec, 0x80, 0xc0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr160, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 380, 0, 0, 0, implicit $exec :: (store (s32) into %stack.130, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa0, 0x19, 0x0f, 0x90, 0xa0, 0x19, 0x16, 0xec, 0x80, 0xbe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr161, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 376, 0, 0, 0, implicit $exec :: (store (s32) into %stack.131, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa1, 0x19, 0x0f, 0x90, 0xa1, 0x19, 0x16, 0xec, 0x80, 0xbc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr162, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 372, 0, 0, 0, implicit $exec :: (store (s32) into %stack.132, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa2, 0x19, 0x0f, 0x90, 0xa2, 0x19, 0x16, 0xec, 0x80, 0xba, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr163, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 368, 0, 0, 0, implicit $exec :: (store (s32) into %stack.133, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa3, 0x19, 0x0f, 0x90, 0xa3, 0x19, 0x16, 0xec, 0x80, 0xb8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr164, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 364, 0, 0, 0, implicit $exec :: (store (s32) into %stack.134, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa4, 0x19, 0x0f, 0x90, 0xa4, 0x19, 0x16, 0xec, 0x80, 0xb6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr165, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 360, 0, 0, 0, implicit $exec :: (store (s32) into %stack.135, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa5, 0x19, 0x0f, 0x90, 0xa5, 0x19, 0x16, 0xec, 0x80, 0xb4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr166, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 356, 0, 0, 0, implicit $exec :: (store (s32) into %stack.136, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa6, 0x19, 0x0f, 0x90, 0xa6, 0x19, 0x16, 0xec, 0x80, 0xb2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr167, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 352, 0, 0, 0, implicit $exec :: (store (s32) into %stack.137, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa7, 0x19, 0x0f, 0x90, 0xa7, 0x19, 0x16, 0xec, 0x80, 0xb0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr168, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 348, 0, 0, 0, implicit $exec :: (store (s32) into %stack.138, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa8, 0x19, 0x0f, 0x90, 0xa8, 0x19, 0x16, 0xec, 0x80, 0xae, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr169, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 344, 0, 0, 0, implicit $exec :: (store (s32) into %stack.139, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa9, 0x19, 0x0f, 0x90, 0xa9, 0x19, 0x16, 0xec, 0x80, 0xac, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr170, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 340, 0, 0, 0, implicit $exec :: (store (s32) into %stack.140, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaa, 0x19, 0x0f, 0x90, 0xaa, 0x19, 0x16, 0xec, 0x80, 0xaa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr171, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 336, 0, 0, 0, implicit $exec :: (store (s32) into %stack.141, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xab, 0x19, 0x0f, 0x90, 0xab, 0x19, 0x16, 0xec, 0x80, 0xa8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr172, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 332, 0, 0, 0, implicit $exec :: (store (s32) into %stack.142, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xac, 0x19, 0x0f, 0x90, 0xac, 0x19, 0x16, 0xec, 0x80, 0xa6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr173, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 328, 0, 0, 0, implicit $exec :: (store (s32) into %stack.143, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xad, 0x19, 0x0f, 0x90, 0xad, 0x19, 0x16, 0xec, 0x80, 0xa4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr174, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 324, 0, 0, 0, implicit $exec :: (store (s32) into %stack.144, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xae, 0x19, 0x0f, 0x90, 0xae, 0x19, 0x16, 0xec, 0x80, 0xa2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr175, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 320, 0, 0, 0, implicit $exec :: (store (s32) into %stack.145, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaf, 0x19, 0x0f, 0x90, 0xaf, 0x19, 0x16, 0xec, 0x80, 0xa0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr176, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 316, 0, 0, 0, implicit $exec :: (store (s32) into %stack.146, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb0, 0x19, 0x0f, 0x90, 0xb0, 0x19, 0x16, 0xec, 0x80, 0x9e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr177, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 312, 0, 0, 0, implicit $exec :: (store (s32) into %stack.147, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb1, 0x19, 0x0f, 0x90, 0xb1, 0x19, 0x16, 0xec, 0x80, 0x9c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr178, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 308, 0, 0, 0, implicit $exec :: (store (s32) into %stack.148, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb2, 0x19, 0x0f, 0x90, 0xb2, 0x19, 0x16, 0xec, 0x80, 0x9a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr179, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 304, 0, 0, 0, implicit $exec :: (store (s32) into %stack.149, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb3, 0x19, 0x0f, 0x90, 0xb3, 0x19, 0x16, 0xec, 0x80, 0x98, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr180, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 300, 0, 0, 0, implicit $exec :: (store (s32) into %stack.150, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb4, 0x19, 0x0f, 0x90, 0xb4, 0x19, 0x16, 0xec, 0x80, 0x96, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr181, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 296, 0, 0, 0, implicit $exec :: (store (s32) into %stack.151, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb5, 0x19, 0x0f, 0x90, 0xb5, 0x19, 0x16, 0xec, 0x80, 0x94, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr182, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 292, 0, 0, 0, implicit $exec :: (store (s32) into %stack.152, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb6, 0x19, 0x0f, 0x90, 0xb6, 0x19, 0x16, 0xec, 0x80, 0x92, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr183, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 288, 0, 0, 0, implicit $exec :: (store (s32) into %stack.153, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb7, 0x19, 0x0f, 0x90, 0xb7, 0x19, 0x16, 0xec, 0x80, 0x90, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr184, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 284, 0, 0, 0, implicit $exec :: (store (s32) into %stack.154, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb8, 0x19, 0x0f, 0x90, 0xb8, 0x19, 0x16, 0xec, 0x80, 0x8e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr185, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 280, 0, 0, 0, implicit $exec :: (store (s32) into %stack.155, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb9, 0x19, 0x0f, 0x90, 0xb9, 0x19, 0x16, 0xec, 0x80, 0x8c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr186, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 276, 0, 0, 0, implicit $exec :: (store (s32) into %stack.156, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xba, 0x19, 0x0f, 0x90, 0xba, 0x19, 0x16, 0xec, 0x80, 0x8a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr187, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 272, 0, 0, 0, implicit $exec :: (store (s32) into %stack.157, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbb, 0x19, 0x0f, 0x90, 0xbb, 0x19, 0x16, 0xec, 0x80, 0x88, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr188, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 268, 0, 0, 0, implicit $exec :: (store (s32) into %stack.158, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbc, 0x19, 0x0f, 0x90, 0xbc, 0x19, 0x16, 0xec, 0x80, 0x86, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr189, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 264, 0, 0, 0, implicit $exec :: (store (s32) into %stack.159, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbd, 0x19, 0x0f, 0x90, 0xbd, 0x19, 0x16, 0xec, 0x80, 0x84, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr190, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 260, 0, 0, 0, implicit $exec :: (store (s32) into %stack.160, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbe, 0x19, 0x0f, 0x90, 0xbe, 0x19, 0x16, 0xec, 0x80, 0x82, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr191, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 256, 0, 0, 0, implicit $exec :: (store (s32) into %stack.161, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbf, 0x19, 0x0f, 0x90, 0xbf, 0x19, 0x16, 0xec, 0x80, 0x80, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr192, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 252, 0, 0, 0, implicit $exec :: (store (s32) into %stack.162, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc0, 0x19, 0x0e, 0x90, 0xc0, 0x19, 0x16, 0xec, 0x80, 0x7e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr193, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 248, 0, 0, 0, implicit $exec :: (store (s32) into %stack.163, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc1, 0x19, 0x0e, 0x90, 0xc1, 0x19, 0x16, 0xec, 0x80, 0x7c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr194, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 244, 0, 0, 0, implicit $exec :: (store (s32) into %stack.164, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc2, 0x19, 0x0e, 0x90, 0xc2, 0x19, 0x16, 0xec, 0x80, 0x7a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr195, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 240, 0, 0, 0, implicit $exec :: (store (s32) into %stack.165, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc3, 0x19, 0x0e, 0x90, 0xc3, 0x19, 0x16, 0xec, 0x80, 0x78, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr196, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 236, 0, 0, 0, implicit $exec :: (store (s32) into %stack.166, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc4, 0x19, 0x0e, 0x90, 0xc4, 0x19, 0x16, 0xec, 0x80, 0x76, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr197, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 232, 0, 0, 0, implicit $exec :: (store (s32) into %stack.167, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc5, 0x19, 0x0e, 0x90, 0xc5, 0x19, 0x16, 0xec, 0x80, 0x74, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr198, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 228, 0, 0, 0, implicit $exec :: (store (s32) into %stack.168, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc6, 0x19, 0x0e, 0x90, 0xc6, 0x19, 0x16, 0xec, 0x80, 0x72, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr199, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 224, 0, 0, 0, implicit $exec :: (store (s32) into %stack.169, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc7, 0x19, 0x0e, 0x90, 0xc7, 0x19, 0x16, 0xec, 0x80, 0x70, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr200, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 220, 0, 0, 0, implicit $exec :: (store (s32) into %stack.170, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc8, 0x19, 0x0e, 0x90, 0xc8, 0x19, 0x16, 0xec, 0x80, 0x6e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr201, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 216, 0, 0, 0, implicit $exec :: (store (s32) into %stack.171, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc9, 0x19, 0x0e, 0x90, 0xc9, 0x19, 0x16, 0xec, 0x80, 0x6c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr202, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 212, 0, 0, 0, implicit $exec :: (store (s32) into %stack.172, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xca, 0x19, 0x0e, 0x90, 0xca, 0x19, 0x16, 0xec, 0x80, 0x6a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr203, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 208, 0, 0, 0, implicit $exec :: (store (s32) into %stack.173, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcb, 0x19, 0x0e, 0x90, 0xcb, 0x19, 0x16, 0xec, 0x80, 0x68, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr204, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 204, 0, 0, 0, implicit $exec :: (store (s32) into %stack.174, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcc, 0x19, 0x0e, 0x90, 0xcc, 0x19, 0x16, 0xec, 0x80, 0x66, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr205, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 200, 0, 0, 0, implicit $exec :: (store (s32) into %stack.175, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcd, 0x19, 0x0e, 0x90, 0xcd, 0x19, 0x16, 0xec, 0x80, 0x64, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr206, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 196, 0, 0, 0, implicit $exec :: (store (s32) into %stack.176, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xce, 0x19, 0x0e, 0x90, 0xce, 0x19, 0x16, 0xec, 0x80, 0x62, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr207, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 192, 0, 0, 0, implicit $exec :: (store (s32) into %stack.177, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcf, 0x19, 0x0e, 0x90, 0xcf, 0x19, 0x16, 0xec, 0x80, 0x60, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr208, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 188, 0, 0, 0, implicit $exec :: (store (s32) into %stack.178, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x19, 0x0e, 0x90, 0xd0, 0x19, 0x16, 0xec, 0x80, 0x5e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr209, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 184, 0, 0, 0, implicit $exec :: (store (s32) into %stack.179, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x19, 0x0e, 0x90, 0xd1, 0x19, 0x16, 0xec, 0x80, 0x5c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr210, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 180, 0, 0, 0, implicit $exec :: (store (s32) into %stack.180, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x19, 0x0e, 0x90, 0xd2, 0x19, 0x16, 0xec, 0x80, 0x5a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr211, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 176, 0, 0, 0, implicit $exec :: (store (s32) into %stack.181, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x19, 0x0e, 0x90, 0xd3, 0x19, 0x16, 0xec, 0x80, 0x58, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr212, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 172, 0, 0, 0, implicit $exec :: (store (s32) into %stack.182, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x19, 0x0e, 0x90, 0xd4, 0x19, 0x16, 0xec, 0x80, 0x56, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr213, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 168, 0, 0, 0, implicit $exec :: (store (s32) into %stack.183, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x19, 0x0e, 0x90, 0xd5, 0x19, 0x16, 0xec, 0x80, 0x54, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr214, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 164, 0, 0, 0, implicit $exec :: (store (s32) into %stack.184, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x19, 0x0e, 0x90, 0xd6, 0x19, 0x16, 0xec, 0x80, 0x52, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr215, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 160, 0, 0, 0, implicit $exec :: (store (s32) into %stack.185, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x19, 0x0e, 0x90, 0xd7, 0x19, 0x16, 0xec, 0x80, 0x50, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr216, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 156, 0, 0, 0, implicit $exec :: (store (s32) into %stack.186, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x19, 0x0e, 0x90, 0xd8, 0x19, 0x16, 0xec, 0x80, 0x4e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr217, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 152, 0, 0, 0, implicit $exec :: (store (s32) into %stack.187, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x19, 0x0e, 0x90, 0xd9, 0x19, 0x16, 0xec, 0x80, 0x4c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr218, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 148, 0, 0, 0, implicit $exec :: (store (s32) into %stack.188, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x19, 0x0e, 0x90, 0xda, 0x19, 0x16, 0xec, 0x80, 0x4a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr219, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 144, 0, 0, 0, implicit $exec :: (store (s32) into %stack.189, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x19, 0x0e, 0x90, 0xdb, 0x19, 0x16, 0xec, 0x80, 0x48, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr220, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 140, 0, 0, 0, implicit $exec :: (store (s32) into %stack.190, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x19, 0x0e, 0x90, 0xdc, 0x19, 0x16, 0xec, 0x80, 0x46, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr221, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 136, 0, 0, 0, implicit $exec :: (store (s32) into %stack.191, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x19, 0x0e, 0x90, 0xdd, 0x19, 0x16, 0xec, 0x80, 0x44, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr222, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 132, 0, 0, 0, implicit $exec :: (store (s32) into %stack.192, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x19, 0x0e, 0x90, 0xde, 0x19, 0x16, 0xec, 0x80, 0x42, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr223, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 128, 0, 0, 0, implicit $exec :: (store (s32) into %stack.193, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x19, 0x0e, 0x90, 0xdf, 0x19, 0x16, 0xec, 0x80, 0x40, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr224, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 124, 0, 0, 0, implicit $exec :: (store (s32) into %stack.194, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x19, 0x0e, 0x90, 0xe0, 0x19, 0x16, 0xec, 0x80, 0x3e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr225, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 120, 0, 0, 0, implicit $exec :: (store (s32) into %stack.195, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x19, 0x0e, 0x90, 0xe1, 0x19, 0x16, 0xec, 0x80, 0x3c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr226, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 116, 0, 0, 0, implicit $exec :: (store (s32) into %stack.196, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x19, 0x0e, 0x90, 0xe2, 0x19, 0x16, 0xec, 0x80, 0x3a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr227, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 112, 0, 0, 0, implicit $exec :: (store (s32) into %stack.197, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x19, 0x0e, 0x90, 0xe3, 0x19, 0x16, 0xec, 0x80, 0x38, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr228, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 108, 0, 0, 0, implicit $exec :: (store (s32) into %stack.198, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x19, 0x0e, 0x90, 0xe4, 0x19, 0x16, 0xec, 0x80, 0x36, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr229, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 104, 0, 0, 0, implicit $exec :: (store (s32) into %stack.199, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x19, 0x0e, 0x90, 0xe5, 0x19, 0x16, 0xec, 0x80, 0x34, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr230, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 100, 0, 0, 0, implicit $exec :: (store (s32) into %stack.200, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x19, 0x0e, 0x90, 0xe6, 0x19, 0x16, 0xec, 0x80, 0x32, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr231, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 96, 0, 0, 0, implicit $exec :: (store (s32) into %stack.201, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x19, 0x0e, 0x90, 0xe7, 0x19, 0x16, 0xec, 0x80, 0x30, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr232, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 92, 0, 0, 0, implicit $exec :: (store (s32) into %stack.202, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x19, 0x0e, 0x90, 0xe8, 0x19, 0x16, 0xec, 0x80, 0x2e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr233, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 88, 0, 0, 0, implicit $exec :: (store (s32) into %stack.203, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x19, 0x0e, 0x90, 0xe9, 0x19, 0x16, 0xec, 0x80, 0x2c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr234, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 84, 0, 0, 0, implicit $exec :: (store (s32) into %stack.204, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x19, 0x0e, 0x90, 0xea, 0x19, 0x16, 0xec, 0x80, 0x2a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr235, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 80, 0, 0, 0, implicit $exec :: (store (s32) into %stack.205, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x19, 0x0e, 0x90, 0xeb, 0x19, 0x16, 0xec, 0x80, 0x28, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr236, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 76, 0, 0, 0, implicit $exec :: (store (s32) into %stack.206, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x19, 0x0e, 0x90, 0xec, 0x19, 0x16, 0xec, 0x80, 0x26, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr237, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 72, 0, 0, 0, implicit $exec :: (store (s32) into %stack.207, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x19, 0x0e, 0x90, 0xed, 0x19, 0x16, 0xec, 0x80, 0x24, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr238, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 68, 0, 0, 0, implicit $exec :: (store (s32) into %stack.208, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x19, 0x0e, 0x90, 0xee, 0x19, 0x16, 0xec, 0x80, 0x22, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr239, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 64, 0, 0, 0, implicit $exec :: (store (s32) into %stack.209, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x19, 0x0e, 0x90, 0xef, 0x19, 0x16, 0xec, 0x80, 0x20, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr240, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec :: (store (s32) into %stack.210, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x19, 0x0e, 0x90, 0xf0, 0x19, 0x16, 0xec, 0x80, 0x1e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr241, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec :: (store (s32) into %stack.211, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x19, 0x0e, 0x90, 0xf1, 0x19, 0x16, 0xec, 0x80, 0x1c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr242, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec :: (store (s32) into %stack.212, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x19, 0x0e, 0x90, 0xf2, 0x19, 0x16, 0xec, 0x80, 0x1a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr243, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec :: (store (s32) into %stack.213, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x19, 0x0e, 0x90, 0xf3, 0x19, 0x16, 0xec, 0x80, 0x18, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr244, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec :: (store (s32) into %stack.214, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x19, 0x0e, 0x90, 0xf4, 0x19, 0x16, 0xec, 0x80, 0x16, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr245, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec :: (store (s32) into %stack.215, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x19, 0x0e, 0x90, 0xf5, 0x19, 0x16, 0xec, 0x80, 0x14, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr246, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec :: (store (s32) into %stack.216, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x19, 0x0e, 0x90, 0xf6, 0x19, 0x16, 0xec, 0x80, 0x12, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr247, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.217, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x19, 0x0e, 0x90, 0xf7, 0x19, 0x16, 0xec, 0x80, 0x10, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr248, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (store (s32) into %stack.218, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x19, 0x0e, 0x90, 0xf8, 0x19, 0x16, 0xec, 0x80, 0x0e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr249, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (store (s32) into %stack.219, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x19, 0x0e, 0x90, 0xf9, 0x19, 0x16, 0xec, 0x80, 0x0c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr250, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (store (s32) into %stack.220, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x19, 0x0e, 0x90, 0xfa, 0x19, 0x16, 0xec, 0x80, 0x0a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr251, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (store (s32) into %stack.221, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x19, 0x0e, 0x90, 0xfb, 0x19, 0x16, 0xec, 0x80, 0x08, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr252, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (store (s32) into %stack.222, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x19, 0x0e, 0x90, 0xfc, 0x19, 0x16, 0xec, 0x80, 0x06, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr253, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (store (s32) into %stack.223, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x19, 0x0e, 0x90, 0xfd, 0x19, 0x16, 0xec, 0x80, 0x04, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr254, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.224, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x19, 0x0e, 0x90, 0xfe, 0x19, 0x16, 0xec, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr255, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.225, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x19, 0x0d, 0x90, 0xff, 0x19, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: S_CMP_EQ_U32 0, 0, implicit-def $scc ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 704, 0, 0, 0, implicit $exec :: (store (s32) into %stack.226, addrspace 5) ; GFX90A-NEXT: $vgpr0 = V_MOV_B32_e32 8904, implicit $exec @@ -8593,229 +10609,453 @@ ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr28_lo16 ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr29_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr32, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr32_lo16, $vgpr0_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr1 = V_ACCVGPR_READ_B32_e64 killed $agpr33, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr33_lo16, $vgpr1_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr2 = V_ACCVGPR_READ_B32_e64 killed $agpr34, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr34_lo16, $vgpr2_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr3 = V_ACCVGPR_READ_B32_e64 killed $agpr35, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr35_lo16, $vgpr3_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr4 = V_ACCVGPR_READ_B32_e64 killed $agpr36, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr36_lo16, $vgpr4_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr5 = V_ACCVGPR_READ_B32_e64 killed $agpr37, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr37_lo16, $vgpr5_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 killed $agpr38, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr38_lo16, $vgpr6_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr7 = V_ACCVGPR_READ_B32_e64 killed $agpr39, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr39_lo16, $vgpr7_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr8 = V_ACCVGPR_READ_B32_e64 killed $agpr40, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr40_lo16, $vgpr8_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr9 = V_ACCVGPR_READ_B32_e64 killed $agpr41, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr41_lo16, $vgpr9_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr10 = V_ACCVGPR_READ_B32_e64 killed $agpr42, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr42_lo16, $vgpr10_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr11 = V_ACCVGPR_READ_B32_e64 killed $agpr43, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr43_lo16, $vgpr11_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr12 = V_ACCVGPR_READ_B32_e64 killed $agpr44, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr44_lo16, $vgpr12_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr13 = V_ACCVGPR_READ_B32_e64 killed $agpr45, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr45_lo16, $vgpr13_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr14 = V_ACCVGPR_READ_B32_e64 killed $agpr46, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr46_lo16, $vgpr14_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr15 = V_ACCVGPR_READ_B32_e64 killed $agpr47, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr47_lo16, $vgpr15_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr16 = V_ACCVGPR_READ_B32_e64 killed $agpr48, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr48_lo16, $vgpr16_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr17 = V_ACCVGPR_READ_B32_e64 killed $agpr49, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr49_lo16, $vgpr17_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr18 = V_ACCVGPR_READ_B32_e64 killed $agpr50, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr50_lo16, $vgpr18_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr19 = V_ACCVGPR_READ_B32_e64 killed $agpr51, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr51_lo16, $vgpr19_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr20 = V_ACCVGPR_READ_B32_e64 killed $agpr52, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr52_lo16, $vgpr20_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr21 = V_ACCVGPR_READ_B32_e64 killed $agpr53, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr53_lo16, $vgpr21_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr22 = V_ACCVGPR_READ_B32_e64 killed $agpr54, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr54_lo16, $vgpr22_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr23 = V_ACCVGPR_READ_B32_e64 killed $agpr55, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr55_lo16, $vgpr23_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr24 = V_ACCVGPR_READ_B32_e64 killed $agpr56, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr56_lo16, $vgpr24_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr25 = V_ACCVGPR_READ_B32_e64 killed $agpr57, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr57_lo16, $vgpr25_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr26 = V_ACCVGPR_READ_B32_e64 killed $agpr58, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr58_lo16, $vgpr26_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr27 = V_ACCVGPR_READ_B32_e64 killed $agpr59, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr59_lo16, $vgpr27_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr28 = V_ACCVGPR_READ_B32_e64 killed $agpr60, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr60_lo16, $vgpr28_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr29 = V_ACCVGPR_READ_B32_e64 killed $agpr61, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr61_lo16, $vgpr29_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr30 = V_ACCVGPR_READ_B32_e64 killed $agpr62, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr62_lo16, $vgpr30_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 killed $agpr63, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr63_lo16, $vgpr31_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr64, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr64_lo16, $vgpr32_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 killed $agpr65, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr65_lo16, $vgpr33_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr34 = V_ACCVGPR_READ_B32_e64 killed $agpr66, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr66_lo16, $vgpr34_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr35 = V_ACCVGPR_READ_B32_e64 killed $agpr67, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr67_lo16, $vgpr35_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr36 = V_ACCVGPR_READ_B32_e64 killed $agpr68, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr68_lo16, $vgpr36_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr37 = V_ACCVGPR_READ_B32_e64 killed $agpr69, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr69_lo16, $vgpr37_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr38 = V_ACCVGPR_READ_B32_e64 killed $agpr70, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr70_lo16, $vgpr38_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr39 = V_ACCVGPR_READ_B32_e64 killed $agpr71, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr71_lo16, $vgpr39_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr48 = V_ACCVGPR_READ_B32_e64 killed $agpr72, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr72_lo16, $vgpr48_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr49 = V_ACCVGPR_READ_B32_e64 killed $agpr73, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr73_lo16, $vgpr49_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr50 = V_ACCVGPR_READ_B32_e64 killed $agpr74, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr74_lo16, $vgpr50_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr51 = V_ACCVGPR_READ_B32_e64 killed $agpr75, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr75_lo16, $vgpr51_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr52 = V_ACCVGPR_READ_B32_e64 killed $agpr76, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr76_lo16, $vgpr52_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr53 = V_ACCVGPR_READ_B32_e64 killed $agpr77, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr77_lo16, $vgpr53_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr54 = V_ACCVGPR_READ_B32_e64 killed $agpr78, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr78_lo16, $vgpr54_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 killed $agpr79, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr79_lo16, $vgpr55_lo16 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr80, $sgpr32, 700, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.50, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x18, 0x0f, 0x90, 0xd0, 0x18, 0x16, 0xec, 0x80, 0xde, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr81, $sgpr32, 696, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.51, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x18, 0x0f, 0x90, 0xd1, 0x18, 0x16, 0xec, 0x80, 0xdc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr82, $sgpr32, 692, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.52, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x18, 0x0f, 0x90, 0xd2, 0x18, 0x16, 0xec, 0x80, 0xda, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr83, $sgpr32, 688, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.53, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x18, 0x0f, 0x90, 0xd3, 0x18, 0x16, 0xec, 0x80, 0xd8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr84, $sgpr32, 684, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.54, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x18, 0x0f, 0x90, 0xd4, 0x18, 0x16, 0xec, 0x80, 0xd6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr85, $sgpr32, 680, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.55, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x18, 0x0f, 0x90, 0xd5, 0x18, 0x16, 0xec, 0x80, 0xd4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr86, $sgpr32, 676, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.56, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x18, 0x0f, 0x90, 0xd6, 0x18, 0x16, 0xec, 0x80, 0xd2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr87, $sgpr32, 672, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.57, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x18, 0x0f, 0x90, 0xd7, 0x18, 0x16, 0xec, 0x80, 0xd0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr88, $sgpr32, 668, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.58, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x18, 0x0f, 0x90, 0xd8, 0x18, 0x16, 0xec, 0x80, 0xce, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr89, $sgpr32, 664, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.59, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x18, 0x0f, 0x90, 0xd9, 0x18, 0x16, 0xec, 0x80, 0xcc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr90, $sgpr32, 660, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.60, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x18, 0x0f, 0x90, 0xda, 0x18, 0x16, 0xec, 0x80, 0xca, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr91, $sgpr32, 656, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.61, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x18, 0x0f, 0x90, 0xdb, 0x18, 0x16, 0xec, 0x80, 0xc8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr92, $sgpr32, 652, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.62, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x18, 0x0f, 0x90, 0xdc, 0x18, 0x16, 0xec, 0x80, 0xc6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr93, $sgpr32, 648, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.63, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x18, 0x0f, 0x90, 0xdd, 0x18, 0x16, 0xec, 0x80, 0xc4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr94, $sgpr32, 644, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.64, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x18, 0x0f, 0x90, 0xde, 0x18, 0x16, 0xec, 0x80, 0xc2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr95, $sgpr32, 640, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.65, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x18, 0x0f, 0x90, 0xdf, 0x18, 0x16, 0xec, 0x80, 0xc0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr96, $sgpr32, 636, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.66, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x18, 0x0f, 0x90, 0xe0, 0x18, 0x16, 0xec, 0x80, 0xbe, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr97, $sgpr32, 632, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.67, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x18, 0x0f, 0x90, 0xe1, 0x18, 0x16, 0xec, 0x80, 0xbc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr98, $sgpr32, 628, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.68, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x18, 0x0f, 0x90, 0xe2, 0x18, 0x16, 0xec, 0x80, 0xba, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr99, $sgpr32, 624, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.69, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x18, 0x0f, 0x90, 0xe3, 0x18, 0x16, 0xec, 0x80, 0xb8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr100, $sgpr32, 620, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.70, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x18, 0x0f, 0x90, 0xe4, 0x18, 0x16, 0xec, 0x80, 0xb6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr101, $sgpr32, 616, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.71, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x18, 0x0f, 0x90, 0xe5, 0x18, 0x16, 0xec, 0x80, 0xb4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr102, $sgpr32, 612, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.72, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x18, 0x0f, 0x90, 0xe6, 0x18, 0x16, 0xec, 0x80, 0xb2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr103, $sgpr32, 608, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.73, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x18, 0x0f, 0x90, 0xe7, 0x18, 0x16, 0xec, 0x80, 0xb0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr104, $sgpr32, 604, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.74, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x18, 0x0f, 0x90, 0xe8, 0x18, 0x16, 0xec, 0x80, 0xae, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr105, $sgpr32, 600, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.75, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x18, 0x0f, 0x90, 0xe9, 0x18, 0x16, 0xec, 0x80, 0xac, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr106, $sgpr32, 596, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.76, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x18, 0x0f, 0x90, 0xea, 0x18, 0x16, 0xec, 0x80, 0xaa, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr107, $sgpr32, 592, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.77, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x18, 0x0f, 0x90, 0xeb, 0x18, 0x16, 0xec, 0x80, 0xa8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr108, $sgpr32, 588, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.78, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x18, 0x0f, 0x90, 0xec, 0x18, 0x16, 0xec, 0x80, 0xa6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr109, $sgpr32, 584, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.79, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x18, 0x0f, 0x90, 0xed, 0x18, 0x16, 0xec, 0x80, 0xa4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr110, $sgpr32, 580, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.80, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x18, 0x0f, 0x90, 0xee, 0x18, 0x16, 0xec, 0x80, 0xa2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr111, $sgpr32, 576, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.81, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x18, 0x0f, 0x90, 0xef, 0x18, 0x16, 0xec, 0x80, 0xa0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr112, $sgpr32, 572, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.82, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x18, 0x0f, 0x90, 0xf0, 0x18, 0x16, 0xec, 0x80, 0x9e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr113, $sgpr32, 568, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.83, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x18, 0x0f, 0x90, 0xf1, 0x18, 0x16, 0xec, 0x80, 0x9c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr114, $sgpr32, 564, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.84, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x18, 0x0f, 0x90, 0xf2, 0x18, 0x16, 0xec, 0x80, 0x9a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr115, $sgpr32, 560, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.85, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x18, 0x0f, 0x90, 0xf3, 0x18, 0x16, 0xec, 0x80, 0x98, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr116, $sgpr32, 556, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.86, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x18, 0x0f, 0x90, 0xf4, 0x18, 0x16, 0xec, 0x80, 0x96, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr117, $sgpr32, 552, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.87, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x18, 0x0f, 0x90, 0xf5, 0x18, 0x16, 0xec, 0x80, 0x94, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr118, $sgpr32, 548, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.88, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x18, 0x0f, 0x90, 0xf6, 0x18, 0x16, 0xec, 0x80, 0x92, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr119, $sgpr32, 544, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.89, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x18, 0x0f, 0x90, 0xf7, 0x18, 0x16, 0xec, 0x80, 0x90, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr120, $sgpr32, 540, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.90, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x18, 0x0f, 0x90, 0xf8, 0x18, 0x16, 0xec, 0x80, 0x8e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr121, $sgpr32, 536, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.91, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x18, 0x0f, 0x90, 0xf9, 0x18, 0x16, 0xec, 0x80, 0x8c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr122, $sgpr32, 532, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.92, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x18, 0x0f, 0x90, 0xfa, 0x18, 0x16, 0xec, 0x80, 0x8a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr123, $sgpr32, 528, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.93, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x18, 0x0f, 0x90, 0xfb, 0x18, 0x16, 0xec, 0x80, 0x88, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr124, $sgpr32, 524, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.94, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x18, 0x0f, 0x90, 0xfc, 0x18, 0x16, 0xec, 0x80, 0x86, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr125, $sgpr32, 520, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.95, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x18, 0x0f, 0x90, 0xfd, 0x18, 0x16, 0xec, 0x80, 0x84, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr126, $sgpr32, 516, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.96, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x18, 0x0f, 0x90, 0xfe, 0x18, 0x16, 0xec, 0x80, 0x82, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr127, $sgpr32, 512, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.97, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x18, 0x0f, 0x90, 0xff, 0x18, 0x16, 0xec, 0x80, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr128, $sgpr32, 508, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.98, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x80, 0x19, 0x0f, 0x90, 0x80, 0x19, 0x16, 0xec, 0x80, 0xfe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr129, $sgpr32, 504, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.99, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x81, 0x19, 0x0f, 0x90, 0x81, 0x19, 0x16, 0xec, 0x80, 0xfc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr130, $sgpr32, 500, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.100, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x82, 0x19, 0x0f, 0x90, 0x82, 0x19, 0x16, 0xec, 0x80, 0xfa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr131, $sgpr32, 496, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.101, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x83, 0x19, 0x0f, 0x90, 0x83, 0x19, 0x16, 0xec, 0x80, 0xf8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr132, $sgpr32, 492, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.102, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x84, 0x19, 0x0f, 0x90, 0x84, 0x19, 0x16, 0xec, 0x80, 0xf6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr133, $sgpr32, 488, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.103, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x85, 0x19, 0x0f, 0x90, 0x85, 0x19, 0x16, 0xec, 0x80, 0xf4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr134, $sgpr32, 484, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.104, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x86, 0x19, 0x0f, 0x90, 0x86, 0x19, 0x16, 0xec, 0x80, 0xf2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr135, $sgpr32, 480, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.105, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x87, 0x19, 0x0f, 0x90, 0x87, 0x19, 0x16, 0xec, 0x80, 0xf0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr136, $sgpr32, 476, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.106, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x88, 0x19, 0x0f, 0x90, 0x88, 0x19, 0x16, 0xec, 0x80, 0xee, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr137, $sgpr32, 472, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.107, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x89, 0x19, 0x0f, 0x90, 0x89, 0x19, 0x16, 0xec, 0x80, 0xec, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr138, $sgpr32, 468, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.108, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8a, 0x19, 0x0f, 0x90, 0x8a, 0x19, 0x16, 0xec, 0x80, 0xea, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr139, $sgpr32, 464, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.109, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8b, 0x19, 0x0f, 0x90, 0x8b, 0x19, 0x16, 0xec, 0x80, 0xe8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr140, $sgpr32, 460, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.110, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8c, 0x19, 0x0f, 0x90, 0x8c, 0x19, 0x16, 0xec, 0x80, 0xe6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr141, $sgpr32, 456, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.111, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8d, 0x19, 0x0f, 0x90, 0x8d, 0x19, 0x16, 0xec, 0x80, 0xe4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr142, $sgpr32, 452, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.112, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8e, 0x19, 0x0f, 0x90, 0x8e, 0x19, 0x16, 0xec, 0x80, 0xe2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr143, $sgpr32, 448, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.113, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8f, 0x19, 0x0f, 0x90, 0x8f, 0x19, 0x16, 0xec, 0x80, 0xe0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr144, $sgpr32, 444, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.114, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x90, 0x19, 0x0f, 0x90, 0x90, 0x19, 0x16, 0xec, 0x80, 0xde, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr145, $sgpr32, 440, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.115, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x91, 0x19, 0x0f, 0x90, 0x91, 0x19, 0x16, 0xec, 0x80, 0xdc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr146, $sgpr32, 436, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.116, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x92, 0x19, 0x0f, 0x90, 0x92, 0x19, 0x16, 0xec, 0x80, 0xda, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr147, $sgpr32, 432, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.117, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x93, 0x19, 0x0f, 0x90, 0x93, 0x19, 0x16, 0xec, 0x80, 0xd8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr148, $sgpr32, 428, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.118, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x94, 0x19, 0x0f, 0x90, 0x94, 0x19, 0x16, 0xec, 0x80, 0xd6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr149, $sgpr32, 424, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.119, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x95, 0x19, 0x0f, 0x90, 0x95, 0x19, 0x16, 0xec, 0x80, 0xd4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr150, $sgpr32, 420, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.120, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x96, 0x19, 0x0f, 0x90, 0x96, 0x19, 0x16, 0xec, 0x80, 0xd2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr151, $sgpr32, 416, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.121, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x97, 0x19, 0x0f, 0x90, 0x97, 0x19, 0x16, 0xec, 0x80, 0xd0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr152, $sgpr32, 412, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.122, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x98, 0x19, 0x0f, 0x90, 0x98, 0x19, 0x16, 0xec, 0x80, 0xce, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr153, $sgpr32, 408, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.123, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x99, 0x19, 0x0f, 0x90, 0x99, 0x19, 0x16, 0xec, 0x80, 0xcc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr154, $sgpr32, 404, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.124, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9a, 0x19, 0x0f, 0x90, 0x9a, 0x19, 0x16, 0xec, 0x80, 0xca, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr155, $sgpr32, 400, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.125, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9b, 0x19, 0x0f, 0x90, 0x9b, 0x19, 0x16, 0xec, 0x80, 0xc8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr156, $sgpr32, 396, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.126, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9c, 0x19, 0x0f, 0x90, 0x9c, 0x19, 0x16, 0xec, 0x80, 0xc6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr157, $sgpr32, 392, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.127, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9d, 0x19, 0x0f, 0x90, 0x9d, 0x19, 0x16, 0xec, 0x80, 0xc4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr158, $sgpr32, 388, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.128, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9e, 0x19, 0x0f, 0x90, 0x9e, 0x19, 0x16, 0xec, 0x80, 0xc2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr159, $sgpr32, 384, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.129, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9f, 0x19, 0x0f, 0x90, 0x9f, 0x19, 0x16, 0xec, 0x80, 0xc0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr160, $sgpr32, 380, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.130, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa0, 0x19, 0x0f, 0x90, 0xa0, 0x19, 0x16, 0xec, 0x80, 0xbe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr161, $sgpr32, 376, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.131, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa1, 0x19, 0x0f, 0x90, 0xa1, 0x19, 0x16, 0xec, 0x80, 0xbc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr162, $sgpr32, 372, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.132, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa2, 0x19, 0x0f, 0x90, 0xa2, 0x19, 0x16, 0xec, 0x80, 0xba, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr163, $sgpr32, 368, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.133, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa3, 0x19, 0x0f, 0x90, 0xa3, 0x19, 0x16, 0xec, 0x80, 0xb8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr164, $sgpr32, 364, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.134, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa4, 0x19, 0x0f, 0x90, 0xa4, 0x19, 0x16, 0xec, 0x80, 0xb6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr165, $sgpr32, 360, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.135, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa5, 0x19, 0x0f, 0x90, 0xa5, 0x19, 0x16, 0xec, 0x80, 0xb4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr166, $sgpr32, 356, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.136, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa6, 0x19, 0x0f, 0x90, 0xa6, 0x19, 0x16, 0xec, 0x80, 0xb2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr167, $sgpr32, 352, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.137, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa7, 0x19, 0x0f, 0x90, 0xa7, 0x19, 0x16, 0xec, 0x80, 0xb0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr168, $sgpr32, 348, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.138, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa8, 0x19, 0x0f, 0x90, 0xa8, 0x19, 0x16, 0xec, 0x80, 0xae, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr169, $sgpr32, 344, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.139, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa9, 0x19, 0x0f, 0x90, 0xa9, 0x19, 0x16, 0xec, 0x80, 0xac, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr170, $sgpr32, 340, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.140, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaa, 0x19, 0x0f, 0x90, 0xaa, 0x19, 0x16, 0xec, 0x80, 0xaa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr171, $sgpr32, 336, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.141, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xab, 0x19, 0x0f, 0x90, 0xab, 0x19, 0x16, 0xec, 0x80, 0xa8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr172, $sgpr32, 332, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.142, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xac, 0x19, 0x0f, 0x90, 0xac, 0x19, 0x16, 0xec, 0x80, 0xa6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr173, $sgpr32, 328, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.143, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xad, 0x19, 0x0f, 0x90, 0xad, 0x19, 0x16, 0xec, 0x80, 0xa4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr174, $sgpr32, 324, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.144, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xae, 0x19, 0x0f, 0x90, 0xae, 0x19, 0x16, 0xec, 0x80, 0xa2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr175, $sgpr32, 320, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.145, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaf, 0x19, 0x0f, 0x90, 0xaf, 0x19, 0x16, 0xec, 0x80, 0xa0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr176, $sgpr32, 316, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.146, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb0, 0x19, 0x0f, 0x90, 0xb0, 0x19, 0x16, 0xec, 0x80, 0x9e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr177, $sgpr32, 312, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.147, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb1, 0x19, 0x0f, 0x90, 0xb1, 0x19, 0x16, 0xec, 0x80, 0x9c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr178, $sgpr32, 308, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.148, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb2, 0x19, 0x0f, 0x90, 0xb2, 0x19, 0x16, 0xec, 0x80, 0x9a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr179, $sgpr32, 304, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.149, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb3, 0x19, 0x0f, 0x90, 0xb3, 0x19, 0x16, 0xec, 0x80, 0x98, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr180, $sgpr32, 300, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.150, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb4, 0x19, 0x0f, 0x90, 0xb4, 0x19, 0x16, 0xec, 0x80, 0x96, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr181, $sgpr32, 296, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.151, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb5, 0x19, 0x0f, 0x90, 0xb5, 0x19, 0x16, 0xec, 0x80, 0x94, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr182, $sgpr32, 292, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.152, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb6, 0x19, 0x0f, 0x90, 0xb6, 0x19, 0x16, 0xec, 0x80, 0x92, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr183, $sgpr32, 288, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.153, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb7, 0x19, 0x0f, 0x90, 0xb7, 0x19, 0x16, 0xec, 0x80, 0x90, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr184, $sgpr32, 284, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.154, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb8, 0x19, 0x0f, 0x90, 0xb8, 0x19, 0x16, 0xec, 0x80, 0x8e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr185, $sgpr32, 280, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.155, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb9, 0x19, 0x0f, 0x90, 0xb9, 0x19, 0x16, 0xec, 0x80, 0x8c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr186, $sgpr32, 276, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.156, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xba, 0x19, 0x0f, 0x90, 0xba, 0x19, 0x16, 0xec, 0x80, 0x8a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr187, $sgpr32, 272, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.157, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbb, 0x19, 0x0f, 0x90, 0xbb, 0x19, 0x16, 0xec, 0x80, 0x88, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr188, $sgpr32, 268, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.158, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbc, 0x19, 0x0f, 0x90, 0xbc, 0x19, 0x16, 0xec, 0x80, 0x86, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr189, $sgpr32, 264, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.159, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbd, 0x19, 0x0f, 0x90, 0xbd, 0x19, 0x16, 0xec, 0x80, 0x84, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr190, $sgpr32, 260, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.160, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbe, 0x19, 0x0f, 0x90, 0xbe, 0x19, 0x16, 0xec, 0x80, 0x82, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr191, $sgpr32, 256, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.161, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbf, 0x19, 0x0f, 0x90, 0xbf, 0x19, 0x16, 0xec, 0x80, 0x80, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr192, $sgpr32, 252, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.162, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc0, 0x19, 0x0e, 0x90, 0xc0, 0x19, 0x16, 0xec, 0x80, 0x7e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr193, $sgpr32, 248, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.163, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc1, 0x19, 0x0e, 0x90, 0xc1, 0x19, 0x16, 0xec, 0x80, 0x7c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr194, $sgpr32, 244, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.164, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc2, 0x19, 0x0e, 0x90, 0xc2, 0x19, 0x16, 0xec, 0x80, 0x7a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr195, $sgpr32, 240, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.165, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc3, 0x19, 0x0e, 0x90, 0xc3, 0x19, 0x16, 0xec, 0x80, 0x78, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr196, $sgpr32, 236, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.166, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc4, 0x19, 0x0e, 0x90, 0xc4, 0x19, 0x16, 0xec, 0x80, 0x76, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr197, $sgpr32, 232, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.167, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc5, 0x19, 0x0e, 0x90, 0xc5, 0x19, 0x16, 0xec, 0x80, 0x74, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr198, $sgpr32, 228, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.168, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc6, 0x19, 0x0e, 0x90, 0xc6, 0x19, 0x16, 0xec, 0x80, 0x72, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr199, $sgpr32, 224, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.169, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc7, 0x19, 0x0e, 0x90, 0xc7, 0x19, 0x16, 0xec, 0x80, 0x70, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr200, $sgpr32, 220, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.170, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc8, 0x19, 0x0e, 0x90, 0xc8, 0x19, 0x16, 0xec, 0x80, 0x6e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr201, $sgpr32, 216, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.171, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc9, 0x19, 0x0e, 0x90, 0xc9, 0x19, 0x16, 0xec, 0x80, 0x6c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr202, $sgpr32, 212, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.172, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xca, 0x19, 0x0e, 0x90, 0xca, 0x19, 0x16, 0xec, 0x80, 0x6a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr203, $sgpr32, 208, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.173, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcb, 0x19, 0x0e, 0x90, 0xcb, 0x19, 0x16, 0xec, 0x80, 0x68, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr204, $sgpr32, 204, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.174, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcc, 0x19, 0x0e, 0x90, 0xcc, 0x19, 0x16, 0xec, 0x80, 0x66, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr205, $sgpr32, 200, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.175, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcd, 0x19, 0x0e, 0x90, 0xcd, 0x19, 0x16, 0xec, 0x80, 0x64, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr206, $sgpr32, 196, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.176, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xce, 0x19, 0x0e, 0x90, 0xce, 0x19, 0x16, 0xec, 0x80, 0x62, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr207, $sgpr32, 192, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.177, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcf, 0x19, 0x0e, 0x90, 0xcf, 0x19, 0x16, 0xec, 0x80, 0x60, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr208, $sgpr32, 188, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.178, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x19, 0x0e, 0x90, 0xd0, 0x19, 0x16, 0xec, 0x80, 0x5e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr209, $sgpr32, 184, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.179, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x19, 0x0e, 0x90, 0xd1, 0x19, 0x16, 0xec, 0x80, 0x5c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr210, $sgpr32, 180, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.180, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x19, 0x0e, 0x90, 0xd2, 0x19, 0x16, 0xec, 0x80, 0x5a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr211, $sgpr32, 176, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.181, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x19, 0x0e, 0x90, 0xd3, 0x19, 0x16, 0xec, 0x80, 0x58, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr212, $sgpr32, 172, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.182, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x19, 0x0e, 0x90, 0xd4, 0x19, 0x16, 0xec, 0x80, 0x56, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr213, $sgpr32, 168, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.183, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x19, 0x0e, 0x90, 0xd5, 0x19, 0x16, 0xec, 0x80, 0x54, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr214, $sgpr32, 164, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.184, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x19, 0x0e, 0x90, 0xd6, 0x19, 0x16, 0xec, 0x80, 0x52, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr215, $sgpr32, 160, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.185, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x19, 0x0e, 0x90, 0xd7, 0x19, 0x16, 0xec, 0x80, 0x50, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr216, $sgpr32, 156, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.186, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x19, 0x0e, 0x90, 0xd8, 0x19, 0x16, 0xec, 0x80, 0x4e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr217, $sgpr32, 152, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.187, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x19, 0x0e, 0x90, 0xd9, 0x19, 0x16, 0xec, 0x80, 0x4c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr218, $sgpr32, 148, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.188, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x19, 0x0e, 0x90, 0xda, 0x19, 0x16, 0xec, 0x80, 0x4a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr219, $sgpr32, 144, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.189, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x19, 0x0e, 0x90, 0xdb, 0x19, 0x16, 0xec, 0x80, 0x48, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr220, $sgpr32, 140, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.190, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x19, 0x0e, 0x90, 0xdc, 0x19, 0x16, 0xec, 0x80, 0x46, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr221, $sgpr32, 136, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.191, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x19, 0x0e, 0x90, 0xdd, 0x19, 0x16, 0xec, 0x80, 0x44, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr222, $sgpr32, 132, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.192, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x19, 0x0e, 0x90, 0xde, 0x19, 0x16, 0xec, 0x80, 0x42, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr223, $sgpr32, 128, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.193, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x19, 0x0e, 0x90, 0xdf, 0x19, 0x16, 0xec, 0x80, 0x40, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr224, $sgpr32, 124, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.194, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x19, 0x0e, 0x90, 0xe0, 0x19, 0x16, 0xec, 0x80, 0x3e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr225, $sgpr32, 120, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.195, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x19, 0x0e, 0x90, 0xe1, 0x19, 0x16, 0xec, 0x80, 0x3c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr226, $sgpr32, 116, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.196, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x19, 0x0e, 0x90, 0xe2, 0x19, 0x16, 0xec, 0x80, 0x3a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr227, $sgpr32, 112, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.197, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x19, 0x0e, 0x90, 0xe3, 0x19, 0x16, 0xec, 0x80, 0x38, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr228, $sgpr32, 108, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.198, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x19, 0x0e, 0x90, 0xe4, 0x19, 0x16, 0xec, 0x80, 0x36, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr229, $sgpr32, 104, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.199, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x19, 0x0e, 0x90, 0xe5, 0x19, 0x16, 0xec, 0x80, 0x34, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr230, $sgpr32, 100, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.200, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x19, 0x0e, 0x90, 0xe6, 0x19, 0x16, 0xec, 0x80, 0x32, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr231, $sgpr32, 96, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.201, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x19, 0x0e, 0x90, 0xe7, 0x19, 0x16, 0xec, 0x80, 0x30, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr232, $sgpr32, 92, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.202, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x19, 0x0e, 0x90, 0xe8, 0x19, 0x16, 0xec, 0x80, 0x2e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr233, $sgpr32, 88, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.203, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x19, 0x0e, 0x90, 0xe9, 0x19, 0x16, 0xec, 0x80, 0x2c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr234, $sgpr32, 84, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.204, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x19, 0x0e, 0x90, 0xea, 0x19, 0x16, 0xec, 0x80, 0x2a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr235, $sgpr32, 80, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.205, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x19, 0x0e, 0x90, 0xeb, 0x19, 0x16, 0xec, 0x80, 0x28, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr236, $sgpr32, 76, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.206, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x19, 0x0e, 0x90, 0xec, 0x19, 0x16, 0xec, 0x80, 0x26, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr237, $sgpr32, 72, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.207, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x19, 0x0e, 0x90, 0xed, 0x19, 0x16, 0xec, 0x80, 0x24, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr238, $sgpr32, 68, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.208, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x19, 0x0e, 0x90, 0xee, 0x19, 0x16, 0xec, 0x80, 0x22, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr239, $sgpr32, 64, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.209, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x19, 0x0e, 0x90, 0xef, 0x19, 0x16, 0xec, 0x80, 0x20, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr240, $sgpr32, 60, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.210, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x19, 0x0e, 0x90, 0xf0, 0x19, 0x16, 0xec, 0x80, 0x1e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr241, $sgpr32, 56, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.211, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x19, 0x0e, 0x90, 0xf1, 0x19, 0x16, 0xec, 0x80, 0x1c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr242, $sgpr32, 52, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.212, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x19, 0x0e, 0x90, 0xf2, 0x19, 0x16, 0xec, 0x80, 0x1a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr243, $sgpr32, 48, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.213, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x19, 0x0e, 0x90, 0xf3, 0x19, 0x16, 0xec, 0x80, 0x18, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr244, $sgpr32, 44, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.214, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x19, 0x0e, 0x90, 0xf4, 0x19, 0x16, 0xec, 0x80, 0x16, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr245, $sgpr32, 40, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.215, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x19, 0x0e, 0x90, 0xf5, 0x19, 0x16, 0xec, 0x80, 0x14, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr246, $sgpr32, 36, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.216, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x19, 0x0e, 0x90, 0xf6, 0x19, 0x16, 0xec, 0x80, 0x12, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr247, $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.217, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x19, 0x0e, 0x90, 0xf7, 0x19, 0x16, 0xec, 0x80, 0x10, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr248, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.218, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x19, 0x0e, 0x90, 0xf8, 0x19, 0x16, 0xec, 0x80, 0x0e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr249, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.219, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x19, 0x0e, 0x90, 0xf9, 0x19, 0x16, 0xec, 0x80, 0x0c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr250, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.220, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x19, 0x0e, 0x90, 0xfa, 0x19, 0x16, 0xec, 0x80, 0x0a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr251, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.221, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x19, 0x0e, 0x90, 0xfb, 0x19, 0x16, 0xec, 0x80, 0x08, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr252, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.222, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x19, 0x0e, 0x90, 0xfc, 0x19, 0x16, 0xec, 0x80, 0x06, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr253, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.223, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x19, 0x0e, 0x90, 0xfd, 0x19, 0x16, 0xec, 0x80, 0x04, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr254, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.224, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x19, 0x0e, 0x90, 0xfe, 0x19, 0x16, 0xec, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr255, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.225, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x19, 0x0d, 0x90, 0xff, 0x19, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: S_CMP_EQ_U32 0, 0, implicit-def $scc ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 704, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.226, addrspace 5) ; GFX90A-FLATSCR-NEXT: $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec @@ -9514,229 +11754,453 @@ ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr28_lo16 ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr29_lo16 ; GFX90A-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr32, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr32_lo16, $vgpr0_lo16 ; GFX90A-NEXT: $vgpr1 = V_ACCVGPR_READ_B32_e64 killed $agpr33, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr33_lo16, $vgpr1_lo16 ; GFX90A-NEXT: $vgpr2 = V_ACCVGPR_READ_B32_e64 killed $agpr34, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr34_lo16, $vgpr2_lo16 ; GFX90A-NEXT: $vgpr3 = V_ACCVGPR_READ_B32_e64 killed $agpr35, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr35_lo16, $vgpr3_lo16 ; GFX90A-NEXT: $vgpr4 = V_ACCVGPR_READ_B32_e64 killed $agpr36, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr36_lo16, $vgpr4_lo16 ; GFX90A-NEXT: $vgpr5 = V_ACCVGPR_READ_B32_e64 killed $agpr37, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr37_lo16, $vgpr5_lo16 ; GFX90A-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 killed $agpr38, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr38_lo16, $vgpr6_lo16 ; GFX90A-NEXT: $vgpr7 = V_ACCVGPR_READ_B32_e64 killed $agpr39, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr39_lo16, $vgpr7_lo16 ; GFX90A-NEXT: $vgpr8 = V_ACCVGPR_READ_B32_e64 killed $agpr40, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr40_lo16, $vgpr8_lo16 ; GFX90A-NEXT: $vgpr9 = V_ACCVGPR_READ_B32_e64 killed $agpr41, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr41_lo16, $vgpr9_lo16 ; GFX90A-NEXT: $vgpr10 = V_ACCVGPR_READ_B32_e64 killed $agpr42, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr42_lo16, $vgpr10_lo16 ; GFX90A-NEXT: $vgpr11 = V_ACCVGPR_READ_B32_e64 killed $agpr43, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr43_lo16, $vgpr11_lo16 ; GFX90A-NEXT: $vgpr12 = V_ACCVGPR_READ_B32_e64 killed $agpr44, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr44_lo16, $vgpr12_lo16 ; GFX90A-NEXT: $vgpr13 = V_ACCVGPR_READ_B32_e64 killed $agpr45, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr45_lo16, $vgpr13_lo16 ; GFX90A-NEXT: $vgpr14 = V_ACCVGPR_READ_B32_e64 killed $agpr46, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr46_lo16, $vgpr14_lo16 ; GFX90A-NEXT: $vgpr15 = V_ACCVGPR_READ_B32_e64 killed $agpr47, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr47_lo16, $vgpr15_lo16 ; GFX90A-NEXT: $vgpr16 = V_ACCVGPR_READ_B32_e64 killed $agpr48, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr48_lo16, $vgpr16_lo16 ; GFX90A-NEXT: $vgpr17 = V_ACCVGPR_READ_B32_e64 killed $agpr49, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr49_lo16, $vgpr17_lo16 ; GFX90A-NEXT: $vgpr18 = V_ACCVGPR_READ_B32_e64 killed $agpr50, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr50_lo16, $vgpr18_lo16 ; GFX90A-NEXT: $vgpr19 = V_ACCVGPR_READ_B32_e64 killed $agpr51, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr51_lo16, $vgpr19_lo16 ; GFX90A-NEXT: $vgpr20 = V_ACCVGPR_READ_B32_e64 killed $agpr52, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr52_lo16, $vgpr20_lo16 ; GFX90A-NEXT: $vgpr21 = V_ACCVGPR_READ_B32_e64 killed $agpr53, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr53_lo16, $vgpr21_lo16 ; GFX90A-NEXT: $vgpr22 = V_ACCVGPR_READ_B32_e64 killed $agpr54, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr54_lo16, $vgpr22_lo16 ; GFX90A-NEXT: $vgpr23 = V_ACCVGPR_READ_B32_e64 killed $agpr55, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr55_lo16, $vgpr23_lo16 ; GFX90A-NEXT: $vgpr24 = V_ACCVGPR_READ_B32_e64 killed $agpr56, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr56_lo16, $vgpr24_lo16 ; GFX90A-NEXT: $vgpr25 = V_ACCVGPR_READ_B32_e64 killed $agpr57, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr57_lo16, $vgpr25_lo16 ; GFX90A-NEXT: $vgpr26 = V_ACCVGPR_READ_B32_e64 killed $agpr58, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr58_lo16, $vgpr26_lo16 ; GFX90A-NEXT: $vgpr27 = V_ACCVGPR_READ_B32_e64 killed $agpr59, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr59_lo16, $vgpr27_lo16 ; GFX90A-NEXT: $vgpr28 = V_ACCVGPR_READ_B32_e64 killed $agpr60, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr60_lo16, $vgpr28_lo16 ; GFX90A-NEXT: $vgpr29 = V_ACCVGPR_READ_B32_e64 killed $agpr61, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr61_lo16, $vgpr29_lo16 ; GFX90A-NEXT: $vgpr30 = V_ACCVGPR_READ_B32_e64 killed $agpr62, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr62_lo16, $vgpr30_lo16 ; GFX90A-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 killed $agpr63, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr63_lo16, $vgpr31_lo16 ; GFX90A-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr64, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr64_lo16, $vgpr32_lo16 ; GFX90A-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 killed $agpr65, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr65_lo16, $vgpr33_lo16 ; GFX90A-NEXT: $vgpr34 = V_ACCVGPR_READ_B32_e64 killed $agpr66, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr66_lo16, $vgpr34_lo16 ; GFX90A-NEXT: $vgpr35 = V_ACCVGPR_READ_B32_e64 killed $agpr67, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr67_lo16, $vgpr35_lo16 ; GFX90A-NEXT: $vgpr36 = V_ACCVGPR_READ_B32_e64 killed $agpr68, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr68_lo16, $vgpr36_lo16 ; GFX90A-NEXT: $vgpr37 = V_ACCVGPR_READ_B32_e64 killed $agpr69, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr69_lo16, $vgpr37_lo16 ; GFX90A-NEXT: $vgpr38 = V_ACCVGPR_READ_B32_e64 killed $agpr70, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr70_lo16, $vgpr38_lo16 ; GFX90A-NEXT: $vgpr39 = V_ACCVGPR_READ_B32_e64 killed $agpr71, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr71_lo16, $vgpr39_lo16 ; GFX90A-NEXT: $vgpr48 = V_ACCVGPR_READ_B32_e64 killed $agpr72, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr72_lo16, $vgpr48_lo16 ; GFX90A-NEXT: $vgpr49 = V_ACCVGPR_READ_B32_e64 killed $agpr73, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr73_lo16, $vgpr49_lo16 ; GFX90A-NEXT: $vgpr50 = V_ACCVGPR_READ_B32_e64 killed $agpr74, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr74_lo16, $vgpr50_lo16 ; GFX90A-NEXT: $vgpr51 = V_ACCVGPR_READ_B32_e64 killed $agpr75, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr75_lo16, $vgpr51_lo16 ; GFX90A-NEXT: $vgpr52 = V_ACCVGPR_READ_B32_e64 killed $agpr76, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr76_lo16, $vgpr52_lo16 ; GFX90A-NEXT: $vgpr53 = V_ACCVGPR_READ_B32_e64 killed $agpr77, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr77_lo16, $vgpr53_lo16 ; GFX90A-NEXT: $vgpr54 = V_ACCVGPR_READ_B32_e64 killed $agpr78, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr78_lo16, $vgpr54_lo16 ; GFX90A-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 killed $agpr79, implicit $exec + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION register $agpr79_lo16, $vgpr55_lo16 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr80, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 700, 0, 0, 0, implicit $exec :: (store (s32) into %stack.50, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x18, 0x0f, 0x90, 0xd0, 0x18, 0x16, 0xec, 0x80, 0xde, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr81, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 696, 0, 0, 0, implicit $exec :: (store (s32) into %stack.51, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x18, 0x0f, 0x90, 0xd1, 0x18, 0x16, 0xec, 0x80, 0xdc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr82, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 692, 0, 0, 0, implicit $exec :: (store (s32) into %stack.52, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x18, 0x0f, 0x90, 0xd2, 0x18, 0x16, 0xec, 0x80, 0xda, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr83, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 688, 0, 0, 0, implicit $exec :: (store (s32) into %stack.53, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x18, 0x0f, 0x90, 0xd3, 0x18, 0x16, 0xec, 0x80, 0xd8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr84, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 684, 0, 0, 0, implicit $exec :: (store (s32) into %stack.54, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x18, 0x0f, 0x90, 0xd4, 0x18, 0x16, 0xec, 0x80, 0xd6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr85, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 680, 0, 0, 0, implicit $exec :: (store (s32) into %stack.55, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x18, 0x0f, 0x90, 0xd5, 0x18, 0x16, 0xec, 0x80, 0xd4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr86, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 676, 0, 0, 0, implicit $exec :: (store (s32) into %stack.56, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x18, 0x0f, 0x90, 0xd6, 0x18, 0x16, 0xec, 0x80, 0xd2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr87, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 672, 0, 0, 0, implicit $exec :: (store (s32) into %stack.57, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x18, 0x0f, 0x90, 0xd7, 0x18, 0x16, 0xec, 0x80, 0xd0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr88, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 668, 0, 0, 0, implicit $exec :: (store (s32) into %stack.58, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x18, 0x0f, 0x90, 0xd8, 0x18, 0x16, 0xec, 0x80, 0xce, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr89, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 664, 0, 0, 0, implicit $exec :: (store (s32) into %stack.59, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x18, 0x0f, 0x90, 0xd9, 0x18, 0x16, 0xec, 0x80, 0xcc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr90, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 660, 0, 0, 0, implicit $exec :: (store (s32) into %stack.60, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x18, 0x0f, 0x90, 0xda, 0x18, 0x16, 0xec, 0x80, 0xca, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr91, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 656, 0, 0, 0, implicit $exec :: (store (s32) into %stack.61, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x18, 0x0f, 0x90, 0xdb, 0x18, 0x16, 0xec, 0x80, 0xc8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr92, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 652, 0, 0, 0, implicit $exec :: (store (s32) into %stack.62, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x18, 0x0f, 0x90, 0xdc, 0x18, 0x16, 0xec, 0x80, 0xc6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr93, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 648, 0, 0, 0, implicit $exec :: (store (s32) into %stack.63, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x18, 0x0f, 0x90, 0xdd, 0x18, 0x16, 0xec, 0x80, 0xc4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr94, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 644, 0, 0, 0, implicit $exec :: (store (s32) into %stack.64, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x18, 0x0f, 0x90, 0xde, 0x18, 0x16, 0xec, 0x80, 0xc2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr95, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 640, 0, 0, 0, implicit $exec :: (store (s32) into %stack.65, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x18, 0x0f, 0x90, 0xdf, 0x18, 0x16, 0xec, 0x80, 0xc0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr96, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 636, 0, 0, 0, implicit $exec :: (store (s32) into %stack.66, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x18, 0x0f, 0x90, 0xe0, 0x18, 0x16, 0xec, 0x80, 0xbe, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr97, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 632, 0, 0, 0, implicit $exec :: (store (s32) into %stack.67, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x18, 0x0f, 0x90, 0xe1, 0x18, 0x16, 0xec, 0x80, 0xbc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr98, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 628, 0, 0, 0, implicit $exec :: (store (s32) into %stack.68, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x18, 0x0f, 0x90, 0xe2, 0x18, 0x16, 0xec, 0x80, 0xba, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr99, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 624, 0, 0, 0, implicit $exec :: (store (s32) into %stack.69, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x18, 0x0f, 0x90, 0xe3, 0x18, 0x16, 0xec, 0x80, 0xb8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr100, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 620, 0, 0, 0, implicit $exec :: (store (s32) into %stack.70, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x18, 0x0f, 0x90, 0xe4, 0x18, 0x16, 0xec, 0x80, 0xb6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr101, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 616, 0, 0, 0, implicit $exec :: (store (s32) into %stack.71, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x18, 0x0f, 0x90, 0xe5, 0x18, 0x16, 0xec, 0x80, 0xb4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr102, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 612, 0, 0, 0, implicit $exec :: (store (s32) into %stack.72, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x18, 0x0f, 0x90, 0xe6, 0x18, 0x16, 0xec, 0x80, 0xb2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr103, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 608, 0, 0, 0, implicit $exec :: (store (s32) into %stack.73, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x18, 0x0f, 0x90, 0xe7, 0x18, 0x16, 0xec, 0x80, 0xb0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr104, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 604, 0, 0, 0, implicit $exec :: (store (s32) into %stack.74, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x18, 0x0f, 0x90, 0xe8, 0x18, 0x16, 0xec, 0x80, 0xae, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr105, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 600, 0, 0, 0, implicit $exec :: (store (s32) into %stack.75, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x18, 0x0f, 0x90, 0xe9, 0x18, 0x16, 0xec, 0x80, 0xac, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr106, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 596, 0, 0, 0, implicit $exec :: (store (s32) into %stack.76, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x18, 0x0f, 0x90, 0xea, 0x18, 0x16, 0xec, 0x80, 0xaa, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr107, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 592, 0, 0, 0, implicit $exec :: (store (s32) into %stack.77, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x18, 0x0f, 0x90, 0xeb, 0x18, 0x16, 0xec, 0x80, 0xa8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr108, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 588, 0, 0, 0, implicit $exec :: (store (s32) into %stack.78, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x18, 0x0f, 0x90, 0xec, 0x18, 0x16, 0xec, 0x80, 0xa6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr109, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 584, 0, 0, 0, implicit $exec :: (store (s32) into %stack.79, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x18, 0x0f, 0x90, 0xed, 0x18, 0x16, 0xec, 0x80, 0xa4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr110, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 580, 0, 0, 0, implicit $exec :: (store (s32) into %stack.80, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x18, 0x0f, 0x90, 0xee, 0x18, 0x16, 0xec, 0x80, 0xa2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr111, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 576, 0, 0, 0, implicit $exec :: (store (s32) into %stack.81, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x18, 0x0f, 0x90, 0xef, 0x18, 0x16, 0xec, 0x80, 0xa0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr112, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 572, 0, 0, 0, implicit $exec :: (store (s32) into %stack.82, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x18, 0x0f, 0x90, 0xf0, 0x18, 0x16, 0xec, 0x80, 0x9e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr113, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 568, 0, 0, 0, implicit $exec :: (store (s32) into %stack.83, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x18, 0x0f, 0x90, 0xf1, 0x18, 0x16, 0xec, 0x80, 0x9c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr114, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 564, 0, 0, 0, implicit $exec :: (store (s32) into %stack.84, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x18, 0x0f, 0x90, 0xf2, 0x18, 0x16, 0xec, 0x80, 0x9a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr115, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 560, 0, 0, 0, implicit $exec :: (store (s32) into %stack.85, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x18, 0x0f, 0x90, 0xf3, 0x18, 0x16, 0xec, 0x80, 0x98, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr116, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 556, 0, 0, 0, implicit $exec :: (store (s32) into %stack.86, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x18, 0x0f, 0x90, 0xf4, 0x18, 0x16, 0xec, 0x80, 0x96, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr117, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 552, 0, 0, 0, implicit $exec :: (store (s32) into %stack.87, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x18, 0x0f, 0x90, 0xf5, 0x18, 0x16, 0xec, 0x80, 0x94, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr118, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 548, 0, 0, 0, implicit $exec :: (store (s32) into %stack.88, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x18, 0x0f, 0x90, 0xf6, 0x18, 0x16, 0xec, 0x80, 0x92, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr119, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 544, 0, 0, 0, implicit $exec :: (store (s32) into %stack.89, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x18, 0x0f, 0x90, 0xf7, 0x18, 0x16, 0xec, 0x80, 0x90, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr120, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 540, 0, 0, 0, implicit $exec :: (store (s32) into %stack.90, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x18, 0x0f, 0x90, 0xf8, 0x18, 0x16, 0xec, 0x80, 0x8e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr121, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 536, 0, 0, 0, implicit $exec :: (store (s32) into %stack.91, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x18, 0x0f, 0x90, 0xf9, 0x18, 0x16, 0xec, 0x80, 0x8c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr122, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 532, 0, 0, 0, implicit $exec :: (store (s32) into %stack.92, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x18, 0x0f, 0x90, 0xfa, 0x18, 0x16, 0xec, 0x80, 0x8a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr123, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 528, 0, 0, 0, implicit $exec :: (store (s32) into %stack.93, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x18, 0x0f, 0x90, 0xfb, 0x18, 0x16, 0xec, 0x80, 0x88, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr124, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 524, 0, 0, 0, implicit $exec :: (store (s32) into %stack.94, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x18, 0x0f, 0x90, 0xfc, 0x18, 0x16, 0xec, 0x80, 0x86, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr125, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 520, 0, 0, 0, implicit $exec :: (store (s32) into %stack.95, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x18, 0x0f, 0x90, 0xfd, 0x18, 0x16, 0xec, 0x80, 0x84, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr126, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 516, 0, 0, 0, implicit $exec :: (store (s32) into %stack.96, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x18, 0x0f, 0x90, 0xfe, 0x18, 0x16, 0xec, 0x80, 0x82, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr127, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 512, 0, 0, 0, implicit $exec :: (store (s32) into %stack.97, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x18, 0x0f, 0x90, 0xff, 0x18, 0x16, 0xec, 0x80, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr128, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 508, 0, 0, 0, implicit $exec :: (store (s32) into %stack.98, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x80, 0x19, 0x0f, 0x90, 0x80, 0x19, 0x16, 0xec, 0x80, 0xfe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr129, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 504, 0, 0, 0, implicit $exec :: (store (s32) into %stack.99, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x81, 0x19, 0x0f, 0x90, 0x81, 0x19, 0x16, 0xec, 0x80, 0xfc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr130, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 500, 0, 0, 0, implicit $exec :: (store (s32) into %stack.100, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x82, 0x19, 0x0f, 0x90, 0x82, 0x19, 0x16, 0xec, 0x80, 0xfa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr131, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 496, 0, 0, 0, implicit $exec :: (store (s32) into %stack.101, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x83, 0x19, 0x0f, 0x90, 0x83, 0x19, 0x16, 0xec, 0x80, 0xf8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr132, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 492, 0, 0, 0, implicit $exec :: (store (s32) into %stack.102, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x84, 0x19, 0x0f, 0x90, 0x84, 0x19, 0x16, 0xec, 0x80, 0xf6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr133, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 488, 0, 0, 0, implicit $exec :: (store (s32) into %stack.103, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x85, 0x19, 0x0f, 0x90, 0x85, 0x19, 0x16, 0xec, 0x80, 0xf4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr134, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 484, 0, 0, 0, implicit $exec :: (store (s32) into %stack.104, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x86, 0x19, 0x0f, 0x90, 0x86, 0x19, 0x16, 0xec, 0x80, 0xf2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr135, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 480, 0, 0, 0, implicit $exec :: (store (s32) into %stack.105, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x87, 0x19, 0x0f, 0x90, 0x87, 0x19, 0x16, 0xec, 0x80, 0xf0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr136, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 476, 0, 0, 0, implicit $exec :: (store (s32) into %stack.106, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x88, 0x19, 0x0f, 0x90, 0x88, 0x19, 0x16, 0xec, 0x80, 0xee, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr137, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 472, 0, 0, 0, implicit $exec :: (store (s32) into %stack.107, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x89, 0x19, 0x0f, 0x90, 0x89, 0x19, 0x16, 0xec, 0x80, 0xec, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr138, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 468, 0, 0, 0, implicit $exec :: (store (s32) into %stack.108, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8a, 0x19, 0x0f, 0x90, 0x8a, 0x19, 0x16, 0xec, 0x80, 0xea, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr139, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 464, 0, 0, 0, implicit $exec :: (store (s32) into %stack.109, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8b, 0x19, 0x0f, 0x90, 0x8b, 0x19, 0x16, 0xec, 0x80, 0xe8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr140, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 460, 0, 0, 0, implicit $exec :: (store (s32) into %stack.110, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8c, 0x19, 0x0f, 0x90, 0x8c, 0x19, 0x16, 0xec, 0x80, 0xe6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr141, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 456, 0, 0, 0, implicit $exec :: (store (s32) into %stack.111, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8d, 0x19, 0x0f, 0x90, 0x8d, 0x19, 0x16, 0xec, 0x80, 0xe4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr142, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 452, 0, 0, 0, implicit $exec :: (store (s32) into %stack.112, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8e, 0x19, 0x0f, 0x90, 0x8e, 0x19, 0x16, 0xec, 0x80, 0xe2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr143, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 448, 0, 0, 0, implicit $exec :: (store (s32) into %stack.113, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8f, 0x19, 0x0f, 0x90, 0x8f, 0x19, 0x16, 0xec, 0x80, 0xe0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr144, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 444, 0, 0, 0, implicit $exec :: (store (s32) into %stack.114, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x90, 0x19, 0x0f, 0x90, 0x90, 0x19, 0x16, 0xec, 0x80, 0xde, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr145, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 440, 0, 0, 0, implicit $exec :: (store (s32) into %stack.115, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x91, 0x19, 0x0f, 0x90, 0x91, 0x19, 0x16, 0xec, 0x80, 0xdc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr146, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 436, 0, 0, 0, implicit $exec :: (store (s32) into %stack.116, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x92, 0x19, 0x0f, 0x90, 0x92, 0x19, 0x16, 0xec, 0x80, 0xda, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr147, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 432, 0, 0, 0, implicit $exec :: (store (s32) into %stack.117, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x93, 0x19, 0x0f, 0x90, 0x93, 0x19, 0x16, 0xec, 0x80, 0xd8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr148, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 428, 0, 0, 0, implicit $exec :: (store (s32) into %stack.118, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x94, 0x19, 0x0f, 0x90, 0x94, 0x19, 0x16, 0xec, 0x80, 0xd6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr149, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 424, 0, 0, 0, implicit $exec :: (store (s32) into %stack.119, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x95, 0x19, 0x0f, 0x90, 0x95, 0x19, 0x16, 0xec, 0x80, 0xd4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr150, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 420, 0, 0, 0, implicit $exec :: (store (s32) into %stack.120, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x96, 0x19, 0x0f, 0x90, 0x96, 0x19, 0x16, 0xec, 0x80, 0xd2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr151, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 416, 0, 0, 0, implicit $exec :: (store (s32) into %stack.121, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x97, 0x19, 0x0f, 0x90, 0x97, 0x19, 0x16, 0xec, 0x80, 0xd0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr152, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 412, 0, 0, 0, implicit $exec :: (store (s32) into %stack.122, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x98, 0x19, 0x0f, 0x90, 0x98, 0x19, 0x16, 0xec, 0x80, 0xce, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr153, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 408, 0, 0, 0, implicit $exec :: (store (s32) into %stack.123, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x99, 0x19, 0x0f, 0x90, 0x99, 0x19, 0x16, 0xec, 0x80, 0xcc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr154, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 404, 0, 0, 0, implicit $exec :: (store (s32) into %stack.124, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9a, 0x19, 0x0f, 0x90, 0x9a, 0x19, 0x16, 0xec, 0x80, 0xca, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr155, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 400, 0, 0, 0, implicit $exec :: (store (s32) into %stack.125, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9b, 0x19, 0x0f, 0x90, 0x9b, 0x19, 0x16, 0xec, 0x80, 0xc8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr156, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 396, 0, 0, 0, implicit $exec :: (store (s32) into %stack.126, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9c, 0x19, 0x0f, 0x90, 0x9c, 0x19, 0x16, 0xec, 0x80, 0xc6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr157, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 392, 0, 0, 0, implicit $exec :: (store (s32) into %stack.127, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9d, 0x19, 0x0f, 0x90, 0x9d, 0x19, 0x16, 0xec, 0x80, 0xc4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr158, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 388, 0, 0, 0, implicit $exec :: (store (s32) into %stack.128, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9e, 0x19, 0x0f, 0x90, 0x9e, 0x19, 0x16, 0xec, 0x80, 0xc2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr159, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 384, 0, 0, 0, implicit $exec :: (store (s32) into %stack.129, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9f, 0x19, 0x0f, 0x90, 0x9f, 0x19, 0x16, 0xec, 0x80, 0xc0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr160, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 380, 0, 0, 0, implicit $exec :: (store (s32) into %stack.130, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa0, 0x19, 0x0f, 0x90, 0xa0, 0x19, 0x16, 0xec, 0x80, 0xbe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr161, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 376, 0, 0, 0, implicit $exec :: (store (s32) into %stack.131, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa1, 0x19, 0x0f, 0x90, 0xa1, 0x19, 0x16, 0xec, 0x80, 0xbc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr162, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 372, 0, 0, 0, implicit $exec :: (store (s32) into %stack.132, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa2, 0x19, 0x0f, 0x90, 0xa2, 0x19, 0x16, 0xec, 0x80, 0xba, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr163, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 368, 0, 0, 0, implicit $exec :: (store (s32) into %stack.133, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa3, 0x19, 0x0f, 0x90, 0xa3, 0x19, 0x16, 0xec, 0x80, 0xb8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr164, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 364, 0, 0, 0, implicit $exec :: (store (s32) into %stack.134, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa4, 0x19, 0x0f, 0x90, 0xa4, 0x19, 0x16, 0xec, 0x80, 0xb6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr165, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 360, 0, 0, 0, implicit $exec :: (store (s32) into %stack.135, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa5, 0x19, 0x0f, 0x90, 0xa5, 0x19, 0x16, 0xec, 0x80, 0xb4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr166, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 356, 0, 0, 0, implicit $exec :: (store (s32) into %stack.136, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa6, 0x19, 0x0f, 0x90, 0xa6, 0x19, 0x16, 0xec, 0x80, 0xb2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr167, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 352, 0, 0, 0, implicit $exec :: (store (s32) into %stack.137, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa7, 0x19, 0x0f, 0x90, 0xa7, 0x19, 0x16, 0xec, 0x80, 0xb0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr168, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 348, 0, 0, 0, implicit $exec :: (store (s32) into %stack.138, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa8, 0x19, 0x0f, 0x90, 0xa8, 0x19, 0x16, 0xec, 0x80, 0xae, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr169, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 344, 0, 0, 0, implicit $exec :: (store (s32) into %stack.139, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa9, 0x19, 0x0f, 0x90, 0xa9, 0x19, 0x16, 0xec, 0x80, 0xac, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr170, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 340, 0, 0, 0, implicit $exec :: (store (s32) into %stack.140, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaa, 0x19, 0x0f, 0x90, 0xaa, 0x19, 0x16, 0xec, 0x80, 0xaa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr171, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 336, 0, 0, 0, implicit $exec :: (store (s32) into %stack.141, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xab, 0x19, 0x0f, 0x90, 0xab, 0x19, 0x16, 0xec, 0x80, 0xa8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr172, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 332, 0, 0, 0, implicit $exec :: (store (s32) into %stack.142, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xac, 0x19, 0x0f, 0x90, 0xac, 0x19, 0x16, 0xec, 0x80, 0xa6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr173, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 328, 0, 0, 0, implicit $exec :: (store (s32) into %stack.143, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xad, 0x19, 0x0f, 0x90, 0xad, 0x19, 0x16, 0xec, 0x80, 0xa4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr174, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 324, 0, 0, 0, implicit $exec :: (store (s32) into %stack.144, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xae, 0x19, 0x0f, 0x90, 0xae, 0x19, 0x16, 0xec, 0x80, 0xa2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr175, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 320, 0, 0, 0, implicit $exec :: (store (s32) into %stack.145, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaf, 0x19, 0x0f, 0x90, 0xaf, 0x19, 0x16, 0xec, 0x80, 0xa0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr176, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 316, 0, 0, 0, implicit $exec :: (store (s32) into %stack.146, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb0, 0x19, 0x0f, 0x90, 0xb0, 0x19, 0x16, 0xec, 0x80, 0x9e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr177, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 312, 0, 0, 0, implicit $exec :: (store (s32) into %stack.147, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb1, 0x19, 0x0f, 0x90, 0xb1, 0x19, 0x16, 0xec, 0x80, 0x9c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr178, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 308, 0, 0, 0, implicit $exec :: (store (s32) into %stack.148, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb2, 0x19, 0x0f, 0x90, 0xb2, 0x19, 0x16, 0xec, 0x80, 0x9a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr179, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 304, 0, 0, 0, implicit $exec :: (store (s32) into %stack.149, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb3, 0x19, 0x0f, 0x90, 0xb3, 0x19, 0x16, 0xec, 0x80, 0x98, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr180, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 300, 0, 0, 0, implicit $exec :: (store (s32) into %stack.150, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb4, 0x19, 0x0f, 0x90, 0xb4, 0x19, 0x16, 0xec, 0x80, 0x96, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr181, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 296, 0, 0, 0, implicit $exec :: (store (s32) into %stack.151, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb5, 0x19, 0x0f, 0x90, 0xb5, 0x19, 0x16, 0xec, 0x80, 0x94, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr182, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 292, 0, 0, 0, implicit $exec :: (store (s32) into %stack.152, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb6, 0x19, 0x0f, 0x90, 0xb6, 0x19, 0x16, 0xec, 0x80, 0x92, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr183, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 288, 0, 0, 0, implicit $exec :: (store (s32) into %stack.153, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb7, 0x19, 0x0f, 0x90, 0xb7, 0x19, 0x16, 0xec, 0x80, 0x90, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr184, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 284, 0, 0, 0, implicit $exec :: (store (s32) into %stack.154, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb8, 0x19, 0x0f, 0x90, 0xb8, 0x19, 0x16, 0xec, 0x80, 0x8e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr185, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 280, 0, 0, 0, implicit $exec :: (store (s32) into %stack.155, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb9, 0x19, 0x0f, 0x90, 0xb9, 0x19, 0x16, 0xec, 0x80, 0x8c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr186, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 276, 0, 0, 0, implicit $exec :: (store (s32) into %stack.156, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xba, 0x19, 0x0f, 0x90, 0xba, 0x19, 0x16, 0xec, 0x80, 0x8a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr187, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 272, 0, 0, 0, implicit $exec :: (store (s32) into %stack.157, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbb, 0x19, 0x0f, 0x90, 0xbb, 0x19, 0x16, 0xec, 0x80, 0x88, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr188, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 268, 0, 0, 0, implicit $exec :: (store (s32) into %stack.158, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbc, 0x19, 0x0f, 0x90, 0xbc, 0x19, 0x16, 0xec, 0x80, 0x86, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr189, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 264, 0, 0, 0, implicit $exec :: (store (s32) into %stack.159, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbd, 0x19, 0x0f, 0x90, 0xbd, 0x19, 0x16, 0xec, 0x80, 0x84, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr190, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 260, 0, 0, 0, implicit $exec :: (store (s32) into %stack.160, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbe, 0x19, 0x0f, 0x90, 0xbe, 0x19, 0x16, 0xec, 0x80, 0x82, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr191, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 256, 0, 0, 0, implicit $exec :: (store (s32) into %stack.161, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbf, 0x19, 0x0f, 0x90, 0xbf, 0x19, 0x16, 0xec, 0x80, 0x80, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr192, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 252, 0, 0, 0, implicit $exec :: (store (s32) into %stack.162, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc0, 0x19, 0x0e, 0x90, 0xc0, 0x19, 0x16, 0xec, 0x80, 0x7e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr193, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 248, 0, 0, 0, implicit $exec :: (store (s32) into %stack.163, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc1, 0x19, 0x0e, 0x90, 0xc1, 0x19, 0x16, 0xec, 0x80, 0x7c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr194, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 244, 0, 0, 0, implicit $exec :: (store (s32) into %stack.164, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc2, 0x19, 0x0e, 0x90, 0xc2, 0x19, 0x16, 0xec, 0x80, 0x7a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr195, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 240, 0, 0, 0, implicit $exec :: (store (s32) into %stack.165, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc3, 0x19, 0x0e, 0x90, 0xc3, 0x19, 0x16, 0xec, 0x80, 0x78, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr196, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 236, 0, 0, 0, implicit $exec :: (store (s32) into %stack.166, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc4, 0x19, 0x0e, 0x90, 0xc4, 0x19, 0x16, 0xec, 0x80, 0x76, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr197, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 232, 0, 0, 0, implicit $exec :: (store (s32) into %stack.167, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc5, 0x19, 0x0e, 0x90, 0xc5, 0x19, 0x16, 0xec, 0x80, 0x74, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr198, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 228, 0, 0, 0, implicit $exec :: (store (s32) into %stack.168, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc6, 0x19, 0x0e, 0x90, 0xc6, 0x19, 0x16, 0xec, 0x80, 0x72, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr199, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 224, 0, 0, 0, implicit $exec :: (store (s32) into %stack.169, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc7, 0x19, 0x0e, 0x90, 0xc7, 0x19, 0x16, 0xec, 0x80, 0x70, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr200, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 220, 0, 0, 0, implicit $exec :: (store (s32) into %stack.170, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc8, 0x19, 0x0e, 0x90, 0xc8, 0x19, 0x16, 0xec, 0x80, 0x6e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr201, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 216, 0, 0, 0, implicit $exec :: (store (s32) into %stack.171, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc9, 0x19, 0x0e, 0x90, 0xc9, 0x19, 0x16, 0xec, 0x80, 0x6c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr202, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 212, 0, 0, 0, implicit $exec :: (store (s32) into %stack.172, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xca, 0x19, 0x0e, 0x90, 0xca, 0x19, 0x16, 0xec, 0x80, 0x6a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr203, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 208, 0, 0, 0, implicit $exec :: (store (s32) into %stack.173, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcb, 0x19, 0x0e, 0x90, 0xcb, 0x19, 0x16, 0xec, 0x80, 0x68, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr204, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 204, 0, 0, 0, implicit $exec :: (store (s32) into %stack.174, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcc, 0x19, 0x0e, 0x90, 0xcc, 0x19, 0x16, 0xec, 0x80, 0x66, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr205, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 200, 0, 0, 0, implicit $exec :: (store (s32) into %stack.175, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcd, 0x19, 0x0e, 0x90, 0xcd, 0x19, 0x16, 0xec, 0x80, 0x64, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr206, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 196, 0, 0, 0, implicit $exec :: (store (s32) into %stack.176, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xce, 0x19, 0x0e, 0x90, 0xce, 0x19, 0x16, 0xec, 0x80, 0x62, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr207, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 192, 0, 0, 0, implicit $exec :: (store (s32) into %stack.177, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcf, 0x19, 0x0e, 0x90, 0xcf, 0x19, 0x16, 0xec, 0x80, 0x60, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr208, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 188, 0, 0, 0, implicit $exec :: (store (s32) into %stack.178, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x19, 0x0e, 0x90, 0xd0, 0x19, 0x16, 0xec, 0x80, 0x5e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr209, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 184, 0, 0, 0, implicit $exec :: (store (s32) into %stack.179, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x19, 0x0e, 0x90, 0xd1, 0x19, 0x16, 0xec, 0x80, 0x5c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr210, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 180, 0, 0, 0, implicit $exec :: (store (s32) into %stack.180, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x19, 0x0e, 0x90, 0xd2, 0x19, 0x16, 0xec, 0x80, 0x5a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr211, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 176, 0, 0, 0, implicit $exec :: (store (s32) into %stack.181, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x19, 0x0e, 0x90, 0xd3, 0x19, 0x16, 0xec, 0x80, 0x58, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr212, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 172, 0, 0, 0, implicit $exec :: (store (s32) into %stack.182, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x19, 0x0e, 0x90, 0xd4, 0x19, 0x16, 0xec, 0x80, 0x56, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr213, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 168, 0, 0, 0, implicit $exec :: (store (s32) into %stack.183, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x19, 0x0e, 0x90, 0xd5, 0x19, 0x16, 0xec, 0x80, 0x54, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr214, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 164, 0, 0, 0, implicit $exec :: (store (s32) into %stack.184, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x19, 0x0e, 0x90, 0xd6, 0x19, 0x16, 0xec, 0x80, 0x52, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr215, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 160, 0, 0, 0, implicit $exec :: (store (s32) into %stack.185, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x19, 0x0e, 0x90, 0xd7, 0x19, 0x16, 0xec, 0x80, 0x50, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr216, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 156, 0, 0, 0, implicit $exec :: (store (s32) into %stack.186, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x19, 0x0e, 0x90, 0xd8, 0x19, 0x16, 0xec, 0x80, 0x4e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr217, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 152, 0, 0, 0, implicit $exec :: (store (s32) into %stack.187, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x19, 0x0e, 0x90, 0xd9, 0x19, 0x16, 0xec, 0x80, 0x4c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr218, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 148, 0, 0, 0, implicit $exec :: (store (s32) into %stack.188, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x19, 0x0e, 0x90, 0xda, 0x19, 0x16, 0xec, 0x80, 0x4a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr219, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 144, 0, 0, 0, implicit $exec :: (store (s32) into %stack.189, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x19, 0x0e, 0x90, 0xdb, 0x19, 0x16, 0xec, 0x80, 0x48, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr220, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 140, 0, 0, 0, implicit $exec :: (store (s32) into %stack.190, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x19, 0x0e, 0x90, 0xdc, 0x19, 0x16, 0xec, 0x80, 0x46, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr221, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 136, 0, 0, 0, implicit $exec :: (store (s32) into %stack.191, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x19, 0x0e, 0x90, 0xdd, 0x19, 0x16, 0xec, 0x80, 0x44, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr222, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 132, 0, 0, 0, implicit $exec :: (store (s32) into %stack.192, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x19, 0x0e, 0x90, 0xde, 0x19, 0x16, 0xec, 0x80, 0x42, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr223, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 128, 0, 0, 0, implicit $exec :: (store (s32) into %stack.193, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x19, 0x0e, 0x90, 0xdf, 0x19, 0x16, 0xec, 0x80, 0x40, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr224, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 124, 0, 0, 0, implicit $exec :: (store (s32) into %stack.194, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x19, 0x0e, 0x90, 0xe0, 0x19, 0x16, 0xec, 0x80, 0x3e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr225, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 120, 0, 0, 0, implicit $exec :: (store (s32) into %stack.195, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x19, 0x0e, 0x90, 0xe1, 0x19, 0x16, 0xec, 0x80, 0x3c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr226, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 116, 0, 0, 0, implicit $exec :: (store (s32) into %stack.196, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x19, 0x0e, 0x90, 0xe2, 0x19, 0x16, 0xec, 0x80, 0x3a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr227, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 112, 0, 0, 0, implicit $exec :: (store (s32) into %stack.197, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x19, 0x0e, 0x90, 0xe3, 0x19, 0x16, 0xec, 0x80, 0x38, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr228, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 108, 0, 0, 0, implicit $exec :: (store (s32) into %stack.198, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x19, 0x0e, 0x90, 0xe4, 0x19, 0x16, 0xec, 0x80, 0x36, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr229, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 104, 0, 0, 0, implicit $exec :: (store (s32) into %stack.199, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x19, 0x0e, 0x90, 0xe5, 0x19, 0x16, 0xec, 0x80, 0x34, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr230, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 100, 0, 0, 0, implicit $exec :: (store (s32) into %stack.200, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x19, 0x0e, 0x90, 0xe6, 0x19, 0x16, 0xec, 0x80, 0x32, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr231, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 96, 0, 0, 0, implicit $exec :: (store (s32) into %stack.201, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x19, 0x0e, 0x90, 0xe7, 0x19, 0x16, 0xec, 0x80, 0x30, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr232, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 92, 0, 0, 0, implicit $exec :: (store (s32) into %stack.202, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x19, 0x0e, 0x90, 0xe8, 0x19, 0x16, 0xec, 0x80, 0x2e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr233, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 88, 0, 0, 0, implicit $exec :: (store (s32) into %stack.203, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x19, 0x0e, 0x90, 0xe9, 0x19, 0x16, 0xec, 0x80, 0x2c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr234, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 84, 0, 0, 0, implicit $exec :: (store (s32) into %stack.204, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x19, 0x0e, 0x90, 0xea, 0x19, 0x16, 0xec, 0x80, 0x2a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr235, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 80, 0, 0, 0, implicit $exec :: (store (s32) into %stack.205, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x19, 0x0e, 0x90, 0xeb, 0x19, 0x16, 0xec, 0x80, 0x28, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr236, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 76, 0, 0, 0, implicit $exec :: (store (s32) into %stack.206, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x19, 0x0e, 0x90, 0xec, 0x19, 0x16, 0xec, 0x80, 0x26, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr237, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 72, 0, 0, 0, implicit $exec :: (store (s32) into %stack.207, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x19, 0x0e, 0x90, 0xed, 0x19, 0x16, 0xec, 0x80, 0x24, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr238, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 68, 0, 0, 0, implicit $exec :: (store (s32) into %stack.208, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x19, 0x0e, 0x90, 0xee, 0x19, 0x16, 0xec, 0x80, 0x22, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr239, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 64, 0, 0, 0, implicit $exec :: (store (s32) into %stack.209, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x19, 0x0e, 0x90, 0xef, 0x19, 0x16, 0xec, 0x80, 0x20, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr240, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 60, 0, 0, 0, implicit $exec :: (store (s32) into %stack.210, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x19, 0x0e, 0x90, 0xf0, 0x19, 0x16, 0xec, 0x80, 0x1e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr241, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 56, 0, 0, 0, implicit $exec :: (store (s32) into %stack.211, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x19, 0x0e, 0x90, 0xf1, 0x19, 0x16, 0xec, 0x80, 0x1c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr242, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 52, 0, 0, 0, implicit $exec :: (store (s32) into %stack.212, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x19, 0x0e, 0x90, 0xf2, 0x19, 0x16, 0xec, 0x80, 0x1a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr243, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec :: (store (s32) into %stack.213, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x19, 0x0e, 0x90, 0xf3, 0x19, 0x16, 0xec, 0x80, 0x18, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr244, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec :: (store (s32) into %stack.214, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x19, 0x0e, 0x90, 0xf4, 0x19, 0x16, 0xec, 0x80, 0x16, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr245, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec :: (store (s32) into %stack.215, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x19, 0x0e, 0x90, 0xf5, 0x19, 0x16, 0xec, 0x80, 0x14, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr246, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec :: (store (s32) into %stack.216, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x19, 0x0e, 0x90, 0xf6, 0x19, 0x16, 0xec, 0x80, 0x12, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr247, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.217, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x19, 0x0e, 0x90, 0xf7, 0x19, 0x16, 0xec, 0x80, 0x10, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr248, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (store (s32) into %stack.218, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x19, 0x0e, 0x90, 0xf8, 0x19, 0x16, 0xec, 0x80, 0x0e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr249, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (store (s32) into %stack.219, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x19, 0x0e, 0x90, 0xf9, 0x19, 0x16, 0xec, 0x80, 0x0c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr250, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (store (s32) into %stack.220, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x19, 0x0e, 0x90, 0xfa, 0x19, 0x16, 0xec, 0x80, 0x0a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr251, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (store (s32) into %stack.221, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x19, 0x0e, 0x90, 0xfb, 0x19, 0x16, 0xec, 0x80, 0x08, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr252, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (store (s32) into %stack.222, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x19, 0x0e, 0x90, 0xfc, 0x19, 0x16, 0xec, 0x80, 0x06, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr253, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (store (s32) into %stack.223, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x19, 0x0e, 0x90, 0xfd, 0x19, 0x16, 0xec, 0x80, 0x04, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr254, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.224, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x19, 0x0e, 0x90, 0xfe, 0x19, 0x16, 0xec, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $agpr255, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.225, addrspace 5) + ; GFX90A-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x19, 0x0d, 0x90, 0xff, 0x19, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-NEXT: S_CMP_EQ_U32 0, 0, implicit-def $scc ; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 704, 0, 0, 0, implicit $exec :: (store (s32) into %stack.226, addrspace 5) ; GFX90A-NEXT: $vgpr0 = V_MOV_B32_e32 8904, implicit $exec @@ -10411,229 +12875,453 @@ ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr28_lo16 ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr29_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 killed $agpr32, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr32_lo16, $vgpr0_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr1 = V_ACCVGPR_READ_B32_e64 killed $agpr33, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr33_lo16, $vgpr1_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr2 = V_ACCVGPR_READ_B32_e64 killed $agpr34, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr34_lo16, $vgpr2_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr3 = V_ACCVGPR_READ_B32_e64 killed $agpr35, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr35_lo16, $vgpr3_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr4 = V_ACCVGPR_READ_B32_e64 killed $agpr36, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr36_lo16, $vgpr4_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr5 = V_ACCVGPR_READ_B32_e64 killed $agpr37, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr37_lo16, $vgpr5_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr6 = V_ACCVGPR_READ_B32_e64 killed $agpr38, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr38_lo16, $vgpr6_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr7 = V_ACCVGPR_READ_B32_e64 killed $agpr39, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr39_lo16, $vgpr7_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr8 = V_ACCVGPR_READ_B32_e64 killed $agpr40, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr40_lo16, $vgpr8_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr9 = V_ACCVGPR_READ_B32_e64 killed $agpr41, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr41_lo16, $vgpr9_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr10 = V_ACCVGPR_READ_B32_e64 killed $agpr42, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr42_lo16, $vgpr10_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr11 = V_ACCVGPR_READ_B32_e64 killed $agpr43, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr43_lo16, $vgpr11_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr12 = V_ACCVGPR_READ_B32_e64 killed $agpr44, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr44_lo16, $vgpr12_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr13 = V_ACCVGPR_READ_B32_e64 killed $agpr45, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr45_lo16, $vgpr13_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr14 = V_ACCVGPR_READ_B32_e64 killed $agpr46, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr46_lo16, $vgpr14_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr15 = V_ACCVGPR_READ_B32_e64 killed $agpr47, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr47_lo16, $vgpr15_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr16 = V_ACCVGPR_READ_B32_e64 killed $agpr48, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr48_lo16, $vgpr16_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr17 = V_ACCVGPR_READ_B32_e64 killed $agpr49, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr49_lo16, $vgpr17_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr18 = V_ACCVGPR_READ_B32_e64 killed $agpr50, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr50_lo16, $vgpr18_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr19 = V_ACCVGPR_READ_B32_e64 killed $agpr51, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr51_lo16, $vgpr19_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr20 = V_ACCVGPR_READ_B32_e64 killed $agpr52, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr52_lo16, $vgpr20_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr21 = V_ACCVGPR_READ_B32_e64 killed $agpr53, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr53_lo16, $vgpr21_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr22 = V_ACCVGPR_READ_B32_e64 killed $agpr54, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr54_lo16, $vgpr22_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr23 = V_ACCVGPR_READ_B32_e64 killed $agpr55, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr55_lo16, $vgpr23_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr24 = V_ACCVGPR_READ_B32_e64 killed $agpr56, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr56_lo16, $vgpr24_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr25 = V_ACCVGPR_READ_B32_e64 killed $agpr57, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr57_lo16, $vgpr25_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr26 = V_ACCVGPR_READ_B32_e64 killed $agpr58, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr58_lo16, $vgpr26_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr27 = V_ACCVGPR_READ_B32_e64 killed $agpr59, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr59_lo16, $vgpr27_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr28 = V_ACCVGPR_READ_B32_e64 killed $agpr60, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr60_lo16, $vgpr28_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr29 = V_ACCVGPR_READ_B32_e64 killed $agpr61, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr61_lo16, $vgpr29_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr30 = V_ACCVGPR_READ_B32_e64 killed $agpr62, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr62_lo16, $vgpr30_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 killed $agpr63, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr63_lo16, $vgpr31_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 killed $agpr64, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr64_lo16, $vgpr32_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 killed $agpr65, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr65_lo16, $vgpr33_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr34 = V_ACCVGPR_READ_B32_e64 killed $agpr66, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr66_lo16, $vgpr34_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr35 = V_ACCVGPR_READ_B32_e64 killed $agpr67, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr67_lo16, $vgpr35_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr36 = V_ACCVGPR_READ_B32_e64 killed $agpr68, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr68_lo16, $vgpr36_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr37 = V_ACCVGPR_READ_B32_e64 killed $agpr69, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr69_lo16, $vgpr37_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr38 = V_ACCVGPR_READ_B32_e64 killed $agpr70, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr70_lo16, $vgpr38_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr39 = V_ACCVGPR_READ_B32_e64 killed $agpr71, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr71_lo16, $vgpr39_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr48 = V_ACCVGPR_READ_B32_e64 killed $agpr72, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr72_lo16, $vgpr48_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr49 = V_ACCVGPR_READ_B32_e64 killed $agpr73, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr73_lo16, $vgpr49_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr50 = V_ACCVGPR_READ_B32_e64 killed $agpr74, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr74_lo16, $vgpr50_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr51 = V_ACCVGPR_READ_B32_e64 killed $agpr75, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr75_lo16, $vgpr51_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr52 = V_ACCVGPR_READ_B32_e64 killed $agpr76, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr76_lo16, $vgpr52_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr53 = V_ACCVGPR_READ_B32_e64 killed $agpr77, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr77_lo16, $vgpr53_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr54 = V_ACCVGPR_READ_B32_e64 killed $agpr78, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr78_lo16, $vgpr54_lo16 ; GFX90A-FLATSCR-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 killed $agpr79, implicit $exec + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION register $agpr79_lo16, $vgpr55_lo16 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr80, $sgpr32, 700, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.50, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x18, 0x0f, 0x90, 0xd0, 0x18, 0x16, 0xec, 0x80, 0xde, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr81, $sgpr32, 696, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.51, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x18, 0x0f, 0x90, 0xd1, 0x18, 0x16, 0xec, 0x80, 0xdc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr82, $sgpr32, 692, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.52, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x18, 0x0f, 0x90, 0xd2, 0x18, 0x16, 0xec, 0x80, 0xda, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr83, $sgpr32, 688, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.53, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x18, 0x0f, 0x90, 0xd3, 0x18, 0x16, 0xec, 0x80, 0xd8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr84, $sgpr32, 684, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.54, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x18, 0x0f, 0x90, 0xd4, 0x18, 0x16, 0xec, 0x80, 0xd6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr85, $sgpr32, 680, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.55, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x18, 0x0f, 0x90, 0xd5, 0x18, 0x16, 0xec, 0x80, 0xd4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr86, $sgpr32, 676, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.56, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x18, 0x0f, 0x90, 0xd6, 0x18, 0x16, 0xec, 0x80, 0xd2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr87, $sgpr32, 672, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.57, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x18, 0x0f, 0x90, 0xd7, 0x18, 0x16, 0xec, 0x80, 0xd0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr88, $sgpr32, 668, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.58, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x18, 0x0f, 0x90, 0xd8, 0x18, 0x16, 0xec, 0x80, 0xce, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr89, $sgpr32, 664, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.59, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x18, 0x0f, 0x90, 0xd9, 0x18, 0x16, 0xec, 0x80, 0xcc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr90, $sgpr32, 660, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.60, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x18, 0x0f, 0x90, 0xda, 0x18, 0x16, 0xec, 0x80, 0xca, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr91, $sgpr32, 656, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.61, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x18, 0x0f, 0x90, 0xdb, 0x18, 0x16, 0xec, 0x80, 0xc8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr92, $sgpr32, 652, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.62, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x18, 0x0f, 0x90, 0xdc, 0x18, 0x16, 0xec, 0x80, 0xc6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr93, $sgpr32, 648, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.63, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x18, 0x0f, 0x90, 0xdd, 0x18, 0x16, 0xec, 0x80, 0xc4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr94, $sgpr32, 644, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.64, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x18, 0x0f, 0x90, 0xde, 0x18, 0x16, 0xec, 0x80, 0xc2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr95, $sgpr32, 640, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.65, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x18, 0x0f, 0x90, 0xdf, 0x18, 0x16, 0xec, 0x80, 0xc0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr96, $sgpr32, 636, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.66, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x18, 0x0f, 0x90, 0xe0, 0x18, 0x16, 0xec, 0x80, 0xbe, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr97, $sgpr32, 632, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.67, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x18, 0x0f, 0x90, 0xe1, 0x18, 0x16, 0xec, 0x80, 0xbc, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr98, $sgpr32, 628, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.68, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x18, 0x0f, 0x90, 0xe2, 0x18, 0x16, 0xec, 0x80, 0xba, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr99, $sgpr32, 624, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.69, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x18, 0x0f, 0x90, 0xe3, 0x18, 0x16, 0xec, 0x80, 0xb8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr100, $sgpr32, 620, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.70, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x18, 0x0f, 0x90, 0xe4, 0x18, 0x16, 0xec, 0x80, 0xb6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr101, $sgpr32, 616, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.71, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x18, 0x0f, 0x90, 0xe5, 0x18, 0x16, 0xec, 0x80, 0xb4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr102, $sgpr32, 612, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.72, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x18, 0x0f, 0x90, 0xe6, 0x18, 0x16, 0xec, 0x80, 0xb2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr103, $sgpr32, 608, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.73, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x18, 0x0f, 0x90, 0xe7, 0x18, 0x16, 0xec, 0x80, 0xb0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr104, $sgpr32, 604, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.74, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x18, 0x0f, 0x90, 0xe8, 0x18, 0x16, 0xec, 0x80, 0xae, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr105, $sgpr32, 600, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.75, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x18, 0x0f, 0x90, 0xe9, 0x18, 0x16, 0xec, 0x80, 0xac, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr106, $sgpr32, 596, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.76, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x18, 0x0f, 0x90, 0xea, 0x18, 0x16, 0xec, 0x80, 0xaa, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr107, $sgpr32, 592, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.77, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x18, 0x0f, 0x90, 0xeb, 0x18, 0x16, 0xec, 0x80, 0xa8, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr108, $sgpr32, 588, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.78, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x18, 0x0f, 0x90, 0xec, 0x18, 0x16, 0xec, 0x80, 0xa6, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr109, $sgpr32, 584, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.79, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x18, 0x0f, 0x90, 0xed, 0x18, 0x16, 0xec, 0x80, 0xa4, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr110, $sgpr32, 580, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.80, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x18, 0x0f, 0x90, 0xee, 0x18, 0x16, 0xec, 0x80, 0xa2, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr111, $sgpr32, 576, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.81, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x18, 0x0f, 0x90, 0xef, 0x18, 0x16, 0xec, 0x80, 0xa0, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr112, $sgpr32, 572, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.82, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x18, 0x0f, 0x90, 0xf0, 0x18, 0x16, 0xec, 0x80, 0x9e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr113, $sgpr32, 568, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.83, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x18, 0x0f, 0x90, 0xf1, 0x18, 0x16, 0xec, 0x80, 0x9c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr114, $sgpr32, 564, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.84, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x18, 0x0f, 0x90, 0xf2, 0x18, 0x16, 0xec, 0x80, 0x9a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr115, $sgpr32, 560, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.85, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x18, 0x0f, 0x90, 0xf3, 0x18, 0x16, 0xec, 0x80, 0x98, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr116, $sgpr32, 556, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.86, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x18, 0x0f, 0x90, 0xf4, 0x18, 0x16, 0xec, 0x80, 0x96, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr117, $sgpr32, 552, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.87, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x18, 0x0f, 0x90, 0xf5, 0x18, 0x16, 0xec, 0x80, 0x94, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr118, $sgpr32, 548, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.88, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x18, 0x0f, 0x90, 0xf6, 0x18, 0x16, 0xec, 0x80, 0x92, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr119, $sgpr32, 544, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.89, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x18, 0x0f, 0x90, 0xf7, 0x18, 0x16, 0xec, 0x80, 0x90, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr120, $sgpr32, 540, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.90, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x18, 0x0f, 0x90, 0xf8, 0x18, 0x16, 0xec, 0x80, 0x8e, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr121, $sgpr32, 536, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.91, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x18, 0x0f, 0x90, 0xf9, 0x18, 0x16, 0xec, 0x80, 0x8c, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr122, $sgpr32, 532, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.92, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x18, 0x0f, 0x90, 0xfa, 0x18, 0x16, 0xec, 0x80, 0x8a, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr123, $sgpr32, 528, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.93, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x18, 0x0f, 0x90, 0xfb, 0x18, 0x16, 0xec, 0x80, 0x88, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr124, $sgpr32, 524, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.94, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x18, 0x0f, 0x90, 0xfc, 0x18, 0x16, 0xec, 0x80, 0x86, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr125, $sgpr32, 520, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.95, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x18, 0x0f, 0x90, 0xfd, 0x18, 0x16, 0xec, 0x80, 0x84, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr126, $sgpr32, 516, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.96, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x18, 0x0f, 0x90, 0xfe, 0x18, 0x16, 0xec, 0x80, 0x82, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr127, $sgpr32, 512, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.97, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x18, 0x0f, 0x90, 0xff, 0x18, 0x16, 0xec, 0x80, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr128, $sgpr32, 508, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.98, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x80, 0x19, 0x0f, 0x90, 0x80, 0x19, 0x16, 0xec, 0x80, 0xfe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr129, $sgpr32, 504, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.99, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x81, 0x19, 0x0f, 0x90, 0x81, 0x19, 0x16, 0xec, 0x80, 0xfc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr130, $sgpr32, 500, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.100, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x82, 0x19, 0x0f, 0x90, 0x82, 0x19, 0x16, 0xec, 0x80, 0xfa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr131, $sgpr32, 496, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.101, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x83, 0x19, 0x0f, 0x90, 0x83, 0x19, 0x16, 0xec, 0x80, 0xf8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr132, $sgpr32, 492, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.102, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x84, 0x19, 0x0f, 0x90, 0x84, 0x19, 0x16, 0xec, 0x80, 0xf6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr133, $sgpr32, 488, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.103, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x85, 0x19, 0x0f, 0x90, 0x85, 0x19, 0x16, 0xec, 0x80, 0xf4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr134, $sgpr32, 484, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.104, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x86, 0x19, 0x0f, 0x90, 0x86, 0x19, 0x16, 0xec, 0x80, 0xf2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr135, $sgpr32, 480, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.105, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x87, 0x19, 0x0f, 0x90, 0x87, 0x19, 0x16, 0xec, 0x80, 0xf0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr136, $sgpr32, 476, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.106, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x88, 0x19, 0x0f, 0x90, 0x88, 0x19, 0x16, 0xec, 0x80, 0xee, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr137, $sgpr32, 472, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.107, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x89, 0x19, 0x0f, 0x90, 0x89, 0x19, 0x16, 0xec, 0x80, 0xec, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr138, $sgpr32, 468, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.108, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8a, 0x19, 0x0f, 0x90, 0x8a, 0x19, 0x16, 0xec, 0x80, 0xea, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr139, $sgpr32, 464, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.109, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8b, 0x19, 0x0f, 0x90, 0x8b, 0x19, 0x16, 0xec, 0x80, 0xe8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr140, $sgpr32, 460, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.110, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8c, 0x19, 0x0f, 0x90, 0x8c, 0x19, 0x16, 0xec, 0x80, 0xe6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr141, $sgpr32, 456, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.111, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8d, 0x19, 0x0f, 0x90, 0x8d, 0x19, 0x16, 0xec, 0x80, 0xe4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr142, $sgpr32, 452, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.112, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8e, 0x19, 0x0f, 0x90, 0x8e, 0x19, 0x16, 0xec, 0x80, 0xe2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr143, $sgpr32, 448, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.113, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x8f, 0x19, 0x0f, 0x90, 0x8f, 0x19, 0x16, 0xec, 0x80, 0xe0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr144, $sgpr32, 444, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.114, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x90, 0x19, 0x0f, 0x90, 0x90, 0x19, 0x16, 0xec, 0x80, 0xde, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr145, $sgpr32, 440, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.115, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x91, 0x19, 0x0f, 0x90, 0x91, 0x19, 0x16, 0xec, 0x80, 0xdc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr146, $sgpr32, 436, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.116, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x92, 0x19, 0x0f, 0x90, 0x92, 0x19, 0x16, 0xec, 0x80, 0xda, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr147, $sgpr32, 432, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.117, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x93, 0x19, 0x0f, 0x90, 0x93, 0x19, 0x16, 0xec, 0x80, 0xd8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr148, $sgpr32, 428, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.118, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x94, 0x19, 0x0f, 0x90, 0x94, 0x19, 0x16, 0xec, 0x80, 0xd6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr149, $sgpr32, 424, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.119, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x95, 0x19, 0x0f, 0x90, 0x95, 0x19, 0x16, 0xec, 0x80, 0xd4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr150, $sgpr32, 420, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.120, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x96, 0x19, 0x0f, 0x90, 0x96, 0x19, 0x16, 0xec, 0x80, 0xd2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr151, $sgpr32, 416, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.121, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x97, 0x19, 0x0f, 0x90, 0x97, 0x19, 0x16, 0xec, 0x80, 0xd0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr152, $sgpr32, 412, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.122, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x98, 0x19, 0x0f, 0x90, 0x98, 0x19, 0x16, 0xec, 0x80, 0xce, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr153, $sgpr32, 408, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.123, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x99, 0x19, 0x0f, 0x90, 0x99, 0x19, 0x16, 0xec, 0x80, 0xcc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr154, $sgpr32, 404, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.124, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9a, 0x19, 0x0f, 0x90, 0x9a, 0x19, 0x16, 0xec, 0x80, 0xca, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr155, $sgpr32, 400, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.125, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9b, 0x19, 0x0f, 0x90, 0x9b, 0x19, 0x16, 0xec, 0x80, 0xc8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr156, $sgpr32, 396, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.126, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9c, 0x19, 0x0f, 0x90, 0x9c, 0x19, 0x16, 0xec, 0x80, 0xc6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr157, $sgpr32, 392, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.127, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9d, 0x19, 0x0f, 0x90, 0x9d, 0x19, 0x16, 0xec, 0x80, 0xc4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr158, $sgpr32, 388, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.128, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9e, 0x19, 0x0f, 0x90, 0x9e, 0x19, 0x16, 0xec, 0x80, 0xc2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr159, $sgpr32, 384, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.129, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x9f, 0x19, 0x0f, 0x90, 0x9f, 0x19, 0x16, 0xec, 0x80, 0xc0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr160, $sgpr32, 380, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.130, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa0, 0x19, 0x0f, 0x90, 0xa0, 0x19, 0x16, 0xec, 0x80, 0xbe, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr161, $sgpr32, 376, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.131, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa1, 0x19, 0x0f, 0x90, 0xa1, 0x19, 0x16, 0xec, 0x80, 0xbc, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr162, $sgpr32, 372, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.132, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa2, 0x19, 0x0f, 0x90, 0xa2, 0x19, 0x16, 0xec, 0x80, 0xba, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr163, $sgpr32, 368, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.133, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa3, 0x19, 0x0f, 0x90, 0xa3, 0x19, 0x16, 0xec, 0x80, 0xb8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr164, $sgpr32, 364, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.134, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa4, 0x19, 0x0f, 0x90, 0xa4, 0x19, 0x16, 0xec, 0x80, 0xb6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr165, $sgpr32, 360, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.135, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa5, 0x19, 0x0f, 0x90, 0xa5, 0x19, 0x16, 0xec, 0x80, 0xb4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr166, $sgpr32, 356, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.136, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa6, 0x19, 0x0f, 0x90, 0xa6, 0x19, 0x16, 0xec, 0x80, 0xb2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr167, $sgpr32, 352, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.137, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa7, 0x19, 0x0f, 0x90, 0xa7, 0x19, 0x16, 0xec, 0x80, 0xb0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr168, $sgpr32, 348, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.138, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa8, 0x19, 0x0f, 0x90, 0xa8, 0x19, 0x16, 0xec, 0x80, 0xae, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr169, $sgpr32, 344, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.139, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa9, 0x19, 0x0f, 0x90, 0xa9, 0x19, 0x16, 0xec, 0x80, 0xac, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr170, $sgpr32, 340, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.140, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaa, 0x19, 0x0f, 0x90, 0xaa, 0x19, 0x16, 0xec, 0x80, 0xaa, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr171, $sgpr32, 336, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.141, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xab, 0x19, 0x0f, 0x90, 0xab, 0x19, 0x16, 0xec, 0x80, 0xa8, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr172, $sgpr32, 332, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.142, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xac, 0x19, 0x0f, 0x90, 0xac, 0x19, 0x16, 0xec, 0x80, 0xa6, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr173, $sgpr32, 328, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.143, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xad, 0x19, 0x0f, 0x90, 0xad, 0x19, 0x16, 0xec, 0x80, 0xa4, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr174, $sgpr32, 324, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.144, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xae, 0x19, 0x0f, 0x90, 0xae, 0x19, 0x16, 0xec, 0x80, 0xa2, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr175, $sgpr32, 320, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.145, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaf, 0x19, 0x0f, 0x90, 0xaf, 0x19, 0x16, 0xec, 0x80, 0xa0, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr176, $sgpr32, 316, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.146, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb0, 0x19, 0x0f, 0x90, 0xb0, 0x19, 0x16, 0xec, 0x80, 0x9e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr177, $sgpr32, 312, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.147, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb1, 0x19, 0x0f, 0x90, 0xb1, 0x19, 0x16, 0xec, 0x80, 0x9c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr178, $sgpr32, 308, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.148, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb2, 0x19, 0x0f, 0x90, 0xb2, 0x19, 0x16, 0xec, 0x80, 0x9a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr179, $sgpr32, 304, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.149, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb3, 0x19, 0x0f, 0x90, 0xb3, 0x19, 0x16, 0xec, 0x80, 0x98, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr180, $sgpr32, 300, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.150, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb4, 0x19, 0x0f, 0x90, 0xb4, 0x19, 0x16, 0xec, 0x80, 0x96, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr181, $sgpr32, 296, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.151, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb5, 0x19, 0x0f, 0x90, 0xb5, 0x19, 0x16, 0xec, 0x80, 0x94, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr182, $sgpr32, 292, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.152, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb6, 0x19, 0x0f, 0x90, 0xb6, 0x19, 0x16, 0xec, 0x80, 0x92, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr183, $sgpr32, 288, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.153, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb7, 0x19, 0x0f, 0x90, 0xb7, 0x19, 0x16, 0xec, 0x80, 0x90, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr184, $sgpr32, 284, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.154, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb8, 0x19, 0x0f, 0x90, 0xb8, 0x19, 0x16, 0xec, 0x80, 0x8e, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr185, $sgpr32, 280, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.155, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb9, 0x19, 0x0f, 0x90, 0xb9, 0x19, 0x16, 0xec, 0x80, 0x8c, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr186, $sgpr32, 276, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.156, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xba, 0x19, 0x0f, 0x90, 0xba, 0x19, 0x16, 0xec, 0x80, 0x8a, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr187, $sgpr32, 272, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.157, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbb, 0x19, 0x0f, 0x90, 0xbb, 0x19, 0x16, 0xec, 0x80, 0x88, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr188, $sgpr32, 268, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.158, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbc, 0x19, 0x0f, 0x90, 0xbc, 0x19, 0x16, 0xec, 0x80, 0x86, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr189, $sgpr32, 264, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.159, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbd, 0x19, 0x0f, 0x90, 0xbd, 0x19, 0x16, 0xec, 0x80, 0x84, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr190, $sgpr32, 260, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.160, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbe, 0x19, 0x0f, 0x90, 0xbe, 0x19, 0x16, 0xec, 0x80, 0x82, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr191, $sgpr32, 256, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.161, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbf, 0x19, 0x0f, 0x90, 0xbf, 0x19, 0x16, 0xec, 0x80, 0x80, 0x01, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr192, $sgpr32, 252, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.162, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc0, 0x19, 0x0e, 0x90, 0xc0, 0x19, 0x16, 0xec, 0x80, 0x7e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr193, $sgpr32, 248, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.163, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc1, 0x19, 0x0e, 0x90, 0xc1, 0x19, 0x16, 0xec, 0x80, 0x7c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr194, $sgpr32, 244, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.164, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc2, 0x19, 0x0e, 0x90, 0xc2, 0x19, 0x16, 0xec, 0x80, 0x7a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr195, $sgpr32, 240, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.165, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc3, 0x19, 0x0e, 0x90, 0xc3, 0x19, 0x16, 0xec, 0x80, 0x78, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr196, $sgpr32, 236, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.166, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc4, 0x19, 0x0e, 0x90, 0xc4, 0x19, 0x16, 0xec, 0x80, 0x76, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr197, $sgpr32, 232, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.167, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc5, 0x19, 0x0e, 0x90, 0xc5, 0x19, 0x16, 0xec, 0x80, 0x74, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr198, $sgpr32, 228, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.168, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc6, 0x19, 0x0e, 0x90, 0xc6, 0x19, 0x16, 0xec, 0x80, 0x72, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr199, $sgpr32, 224, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.169, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc7, 0x19, 0x0e, 0x90, 0xc7, 0x19, 0x16, 0xec, 0x80, 0x70, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr200, $sgpr32, 220, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.170, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc8, 0x19, 0x0e, 0x90, 0xc8, 0x19, 0x16, 0xec, 0x80, 0x6e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr201, $sgpr32, 216, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.171, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xc9, 0x19, 0x0e, 0x90, 0xc9, 0x19, 0x16, 0xec, 0x80, 0x6c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr202, $sgpr32, 212, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.172, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xca, 0x19, 0x0e, 0x90, 0xca, 0x19, 0x16, 0xec, 0x80, 0x6a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr203, $sgpr32, 208, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.173, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcb, 0x19, 0x0e, 0x90, 0xcb, 0x19, 0x16, 0xec, 0x80, 0x68, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr204, $sgpr32, 204, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.174, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcc, 0x19, 0x0e, 0x90, 0xcc, 0x19, 0x16, 0xec, 0x80, 0x66, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr205, $sgpr32, 200, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.175, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcd, 0x19, 0x0e, 0x90, 0xcd, 0x19, 0x16, 0xec, 0x80, 0x64, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr206, $sgpr32, 196, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.176, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xce, 0x19, 0x0e, 0x90, 0xce, 0x19, 0x16, 0xec, 0x80, 0x62, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr207, $sgpr32, 192, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.177, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xcf, 0x19, 0x0e, 0x90, 0xcf, 0x19, 0x16, 0xec, 0x80, 0x60, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr208, $sgpr32, 188, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.178, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd0, 0x19, 0x0e, 0x90, 0xd0, 0x19, 0x16, 0xec, 0x80, 0x5e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr209, $sgpr32, 184, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.179, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd1, 0x19, 0x0e, 0x90, 0xd1, 0x19, 0x16, 0xec, 0x80, 0x5c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr210, $sgpr32, 180, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.180, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd2, 0x19, 0x0e, 0x90, 0xd2, 0x19, 0x16, 0xec, 0x80, 0x5a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr211, $sgpr32, 176, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.181, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd3, 0x19, 0x0e, 0x90, 0xd3, 0x19, 0x16, 0xec, 0x80, 0x58, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr212, $sgpr32, 172, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.182, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd4, 0x19, 0x0e, 0x90, 0xd4, 0x19, 0x16, 0xec, 0x80, 0x56, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr213, $sgpr32, 168, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.183, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd5, 0x19, 0x0e, 0x90, 0xd5, 0x19, 0x16, 0xec, 0x80, 0x54, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr214, $sgpr32, 164, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.184, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd6, 0x19, 0x0e, 0x90, 0xd6, 0x19, 0x16, 0xec, 0x80, 0x52, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr215, $sgpr32, 160, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.185, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd7, 0x19, 0x0e, 0x90, 0xd7, 0x19, 0x16, 0xec, 0x80, 0x50, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr216, $sgpr32, 156, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.186, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd8, 0x19, 0x0e, 0x90, 0xd8, 0x19, 0x16, 0xec, 0x80, 0x4e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr217, $sgpr32, 152, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.187, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xd9, 0x19, 0x0e, 0x90, 0xd9, 0x19, 0x16, 0xec, 0x80, 0x4c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr218, $sgpr32, 148, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.188, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xda, 0x19, 0x0e, 0x90, 0xda, 0x19, 0x16, 0xec, 0x80, 0x4a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr219, $sgpr32, 144, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.189, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdb, 0x19, 0x0e, 0x90, 0xdb, 0x19, 0x16, 0xec, 0x80, 0x48, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr220, $sgpr32, 140, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.190, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdc, 0x19, 0x0e, 0x90, 0xdc, 0x19, 0x16, 0xec, 0x80, 0x46, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr221, $sgpr32, 136, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.191, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdd, 0x19, 0x0e, 0x90, 0xdd, 0x19, 0x16, 0xec, 0x80, 0x44, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr222, $sgpr32, 132, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.192, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xde, 0x19, 0x0e, 0x90, 0xde, 0x19, 0x16, 0xec, 0x80, 0x42, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr223, $sgpr32, 128, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.193, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xdf, 0x19, 0x0e, 0x90, 0xdf, 0x19, 0x16, 0xec, 0x80, 0x40, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr224, $sgpr32, 124, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.194, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe0, 0x19, 0x0e, 0x90, 0xe0, 0x19, 0x16, 0xec, 0x80, 0x3e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr225, $sgpr32, 120, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.195, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe1, 0x19, 0x0e, 0x90, 0xe1, 0x19, 0x16, 0xec, 0x80, 0x3c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr226, $sgpr32, 116, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.196, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe2, 0x19, 0x0e, 0x90, 0xe2, 0x19, 0x16, 0xec, 0x80, 0x3a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr227, $sgpr32, 112, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.197, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe3, 0x19, 0x0e, 0x90, 0xe3, 0x19, 0x16, 0xec, 0x80, 0x38, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr228, $sgpr32, 108, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.198, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe4, 0x19, 0x0e, 0x90, 0xe4, 0x19, 0x16, 0xec, 0x80, 0x36, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr229, $sgpr32, 104, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.199, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe5, 0x19, 0x0e, 0x90, 0xe5, 0x19, 0x16, 0xec, 0x80, 0x34, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr230, $sgpr32, 100, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.200, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe6, 0x19, 0x0e, 0x90, 0xe6, 0x19, 0x16, 0xec, 0x80, 0x32, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr231, $sgpr32, 96, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.201, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe7, 0x19, 0x0e, 0x90, 0xe7, 0x19, 0x16, 0xec, 0x80, 0x30, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr232, $sgpr32, 92, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.202, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe8, 0x19, 0x0e, 0x90, 0xe8, 0x19, 0x16, 0xec, 0x80, 0x2e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr233, $sgpr32, 88, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.203, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xe9, 0x19, 0x0e, 0x90, 0xe9, 0x19, 0x16, 0xec, 0x80, 0x2c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr234, $sgpr32, 84, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.204, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xea, 0x19, 0x0e, 0x90, 0xea, 0x19, 0x16, 0xec, 0x80, 0x2a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr235, $sgpr32, 80, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.205, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xeb, 0x19, 0x0e, 0x90, 0xeb, 0x19, 0x16, 0xec, 0x80, 0x28, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr236, $sgpr32, 76, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.206, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xec, 0x19, 0x0e, 0x90, 0xec, 0x19, 0x16, 0xec, 0x80, 0x26, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr237, $sgpr32, 72, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.207, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xed, 0x19, 0x0e, 0x90, 0xed, 0x19, 0x16, 0xec, 0x80, 0x24, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr238, $sgpr32, 68, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.208, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xee, 0x19, 0x0e, 0x90, 0xee, 0x19, 0x16, 0xec, 0x80, 0x22, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr239, $sgpr32, 64, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.209, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xef, 0x19, 0x0e, 0x90, 0xef, 0x19, 0x16, 0xec, 0x80, 0x20, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr240, $sgpr32, 60, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.210, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf0, 0x19, 0x0e, 0x90, 0xf0, 0x19, 0x16, 0xec, 0x80, 0x1e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr241, $sgpr32, 56, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.211, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf1, 0x19, 0x0e, 0x90, 0xf1, 0x19, 0x16, 0xec, 0x80, 0x1c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr242, $sgpr32, 52, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.212, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf2, 0x19, 0x0e, 0x90, 0xf2, 0x19, 0x16, 0xec, 0x80, 0x1a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr243, $sgpr32, 48, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.213, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf3, 0x19, 0x0e, 0x90, 0xf3, 0x19, 0x16, 0xec, 0x80, 0x18, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr244, $sgpr32, 44, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.214, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf4, 0x19, 0x0e, 0x90, 0xf4, 0x19, 0x16, 0xec, 0x80, 0x16, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr245, $sgpr32, 40, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.215, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf5, 0x19, 0x0e, 0x90, 0xf5, 0x19, 0x16, 0xec, 0x80, 0x14, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr246, $sgpr32, 36, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.216, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf6, 0x19, 0x0e, 0x90, 0xf6, 0x19, 0x16, 0xec, 0x80, 0x12, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr247, $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.217, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf7, 0x19, 0x0e, 0x90, 0xf7, 0x19, 0x16, 0xec, 0x80, 0x10, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr248, $sgpr32, 28, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.218, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf8, 0x19, 0x0e, 0x90, 0xf8, 0x19, 0x16, 0xec, 0x80, 0x0e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr249, $sgpr32, 24, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.219, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xf9, 0x19, 0x0e, 0x90, 0xf9, 0x19, 0x16, 0xec, 0x80, 0x0c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr250, $sgpr32, 20, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.220, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfa, 0x19, 0x0e, 0x90, 0xfa, 0x19, 0x16, 0xec, 0x80, 0x0a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr251, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.221, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfb, 0x19, 0x0e, 0x90, 0xfb, 0x19, 0x16, 0xec, 0x80, 0x08, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr252, $sgpr32, 12, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.222, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfc, 0x19, 0x0e, 0x90, 0xfc, 0x19, 0x16, 0xec, 0x80, 0x06, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr253, $sgpr32, 8, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.223, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfd, 0x19, 0x0e, 0x90, 0xfd, 0x19, 0x16, 0xec, 0x80, 0x04, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr254, $sgpr32, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.224, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xfe, 0x19, 0x0e, 0x90, 0xfe, 0x19, 0x16, 0xec, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $agpr255, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.225, addrspace 5) + ; GFX90A-FLATSCR-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xff, 0x19, 0x0d, 0x90, 0xff, 0x19, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GFX90A-FLATSCR-NEXT: S_CMP_EQ_U32 0, 0, implicit-def $scc ; GFX90A-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr32, 704, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.226, addrspace 5) ; GFX90A-FLATSCR-NEXT: $vgpr0 = V_MOV_B32_e32 $sgpr32, implicit $exec diff --git a/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll b/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll --- a/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll +++ b/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll @@ -462,6 +462,7 @@ ; GFX90A-LABEL: v32_asm_def_use: ; GFX90A: ; %bb.0: ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX90A-NEXT: v_accvgpr_read_b32 v35, a32 ; Reload Reuse ; GFX90A-NEXT: v_mov_b32_e32 v34, v0 ; GFX90A-NEXT: v_mov_b32_e32 v33, v1 ; GFX90A-NEXT: ;;#ASMSTART @@ -489,8 +490,8 @@ ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; copy ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_accvgpr_read_b32 v35, a32 ; Reload Reuse ; GFX90A-NEXT: v_accvgpr_mov_b32 a32, a1 +; GFX90A-NEXT: s_nop 0 ; GFX90A-NEXT: v_mfma_f32_16x16x1f32 a[0:15], v34, v33, a[16:31] ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; copy @@ -1017,6 +1018,7 @@ ; GFX90A-LABEL: no_free_vgprs_at_sgpr_to_agpr_copy: ; GFX90A: ; %bb.0: ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX90A-NEXT: v_accvgpr_read_b32 v34, a32 ; Reload Reuse ; GFX90A-NEXT: v_mov_b32_e32 v33, v0 ; GFX90A-NEXT: v_mov_b32_e32 v32, v1 ; GFX90A-NEXT: ;;#ASMSTART @@ -1038,8 +1040,7 @@ ; GFX90A-NEXT: v_accvgpr_write_b32 a18, s2 ; GFX90A-NEXT: v_accvgpr_write_b32 a17, s1 ; GFX90A-NEXT: v_accvgpr_write_b32 a16, s0 -; GFX90A-NEXT: v_accvgpr_read_b32 v34, a32 ; Reload Reuse -; GFX90A-NEXT: s_nop 0 +; GFX90A-NEXT: s_nop 1 ; GFX90A-NEXT: v_mfma_f32_16x16x1f32 a[0:15], v33, v32, a[16:31] ; GFX90A-NEXT: s_nop 7 ; GFX90A-NEXT: s_nop 2 diff --git a/llvm/test/CodeGen/AMDGPU/av_spill_cross_bb_usage.mir b/llvm/test/CodeGen/AMDGPU/av_spill_cross_bb_usage.mir --- a/llvm/test/CodeGen/AMDGPU/av_spill_cross_bb_usage.mir +++ b/llvm/test/CodeGen/AMDGPU/av_spill_cross_bb_usage.mir @@ -458,18 +458,31 @@ ; GCN-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr28_lo16 ; GCN-NEXT: frame-setup CFI_INSTRUCTION undefined $sgpr29_lo16 ; GCN-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 48, 0, 0, 0, implicit $exec :: (store (s32) into %stack.3, addrspace 5) + ; GCN-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa8, 0x14, 0x0e, 0x90, 0xa8, 0x14, 0x16, 0xec, 0x80, 0x18, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GCN-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr41, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 44, 0, 0, 0, implicit $exec :: (store (s32) into %stack.4, addrspace 5) + ; GCN-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xa9, 0x14, 0x0e, 0x90, 0xa9, 0x14, 0x16, 0xec, 0x80, 0x16, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GCN-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr42, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 40, 0, 0, 0, implicit $exec :: (store (s32) into %stack.5, addrspace 5) + ; GCN-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xaa, 0x14, 0x0e, 0x90, 0xaa, 0x14, 0x16, 0xec, 0x80, 0x14, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GCN-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr43, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 36, 0, 0, 0, implicit $exec :: (store (s32) into %stack.6, addrspace 5) + ; GCN-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xab, 0x14, 0x0e, 0x90, 0xab, 0x14, 0x16, 0xec, 0x80, 0x12, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GCN-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr44, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.7, addrspace 5) + ; GCN-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xac, 0x14, 0x0e, 0x90, 0xac, 0x14, 0x16, 0xec, 0x80, 0x10, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GCN-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr45, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 28, 0, 0, 0, implicit $exec :: (store (s32) into %stack.8, addrspace 5) + ; GCN-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xad, 0x14, 0x0e, 0x90, 0xad, 0x14, 0x16, 0xec, 0x80, 0x0e, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GCN-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr46, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 24, 0, 0, 0, implicit $exec :: (store (s32) into %stack.9, addrspace 5) + ; GCN-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xae, 0x14, 0x0e, 0x90, 0xae, 0x14, 0x16, 0xec, 0x80, 0x0c, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GCN-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr56, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 20, 0, 0, 0, implicit $exec :: (store (s32) into %stack.10, addrspace 5) + ; GCN-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb8, 0x14, 0x0e, 0x90, 0xb8, 0x14, 0x16, 0xec, 0x80, 0x0a, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GCN-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr57, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 16, 0, 0, 0, implicit $exec :: (store (s32) into %stack.11, addrspace 5) + ; GCN-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xb9, 0x14, 0x0e, 0x90, 0xb9, 0x14, 0x16, 0xec, 0x80, 0x08, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GCN-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr58, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 12, 0, 0, 0, implicit $exec :: (store (s32) into %stack.12, addrspace 5) + ; GCN-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xba, 0x14, 0x0e, 0x90, 0xba, 0x14, 0x16, 0xec, 0x80, 0x06, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GCN-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr59, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, 0, implicit $exec :: (store (s32) into %stack.13, addrspace 5) + ; GCN-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbb, 0x14, 0x0e, 0x90, 0xbb, 0x14, 0x16, 0xec, 0x80, 0x04, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GCN-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr60, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, 0, implicit $exec :: (store (s32) into %stack.14, addrspace 5) + ; GCN-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbc, 0x14, 0x0e, 0x90, 0xbc, 0x14, 0x16, 0xec, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GCN-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr61, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, 0, implicit $exec :: (store (s32) into %stack.15, addrspace 5) + ; GCN-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0xbd, 0x14, 0x0d, 0x90, 0xbd, 0x14, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; GCN-NEXT: renamable $vgpr44 = COPY $vgpr13, implicit $exec ; GCN-NEXT: renamable $vgpr43 = COPY $vgpr12, implicit $exec ; GCN-NEXT: S_CBRANCH_SCC1 %bb.2, implicit undef $scc diff --git a/llvm/test/CodeGen/AMDGPU/branch-relax-spill.ll b/llvm/test/CodeGen/AMDGPU/branch-relax-spill.ll --- a/llvm/test/CodeGen/AMDGPU/branch-relax-spill.ll +++ b/llvm/test/CodeGen/AMDGPU/branch-relax-spill.ll @@ -903,79 +903,79 @@ ; CHECK-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill ; CHECK-NEXT: s_mov_b64 exec, s[4:5] ; CHECK-NEXT: s_waitcnt expcnt(1) -; CHECK-NEXT: v_writelane_b32 v0, s30, 0 -; CHECK-NEXT: v_writelane_b32 v0, s31, 1 -; CHECK-NEXT: v_writelane_b32 v0, s33, 2 -; CHECK-NEXT: v_writelane_b32 v0, s34, 3 -; CHECK-NEXT: v_writelane_b32 v0, s35, 4 -; CHECK-NEXT: v_writelane_b32 v0, s36, 5 -; CHECK-NEXT: v_writelane_b32 v0, s37, 6 -; CHECK-NEXT: v_writelane_b32 v0, s38, 7 -; CHECK-NEXT: v_writelane_b32 v0, s39, 8 -; CHECK-NEXT: v_writelane_b32 v0, s40, 9 -; CHECK-NEXT: v_writelane_b32 v0, s41, 10 -; CHECK-NEXT: v_writelane_b32 v0, s42, 11 -; CHECK-NEXT: v_writelane_b32 v0, s43, 12 -; CHECK-NEXT: v_writelane_b32 v0, s44, 13 -; CHECK-NEXT: v_writelane_b32 v0, s45, 14 -; CHECK-NEXT: v_writelane_b32 v0, s46, 15 -; CHECK-NEXT: v_writelane_b32 v0, s47, 16 -; CHECK-NEXT: v_writelane_b32 v0, s48, 17 -; CHECK-NEXT: v_writelane_b32 v0, s49, 18 -; CHECK-NEXT: v_writelane_b32 v0, s50, 19 -; CHECK-NEXT: v_writelane_b32 v0, s51, 20 -; CHECK-NEXT: v_writelane_b32 v0, s52, 21 -; CHECK-NEXT: v_writelane_b32 v0, s53, 22 -; CHECK-NEXT: v_writelane_b32 v0, s54, 23 -; CHECK-NEXT: v_writelane_b32 v0, s55, 24 -; CHECK-NEXT: v_writelane_b32 v0, s56, 25 -; CHECK-NEXT: v_writelane_b32 v0, s57, 26 -; CHECK-NEXT: v_writelane_b32 v0, s58, 27 -; CHECK-NEXT: v_writelane_b32 v0, s59, 28 -; CHECK-NEXT: v_writelane_b32 v0, s60, 29 -; CHECK-NEXT: v_writelane_b32 v0, s61, 30 -; CHECK-NEXT: v_writelane_b32 v0, s62, 31 -; CHECK-NEXT: v_writelane_b32 v0, s63, 32 -; CHECK-NEXT: v_writelane_b32 v0, s64, 33 -; CHECK-NEXT: v_writelane_b32 v0, s65, 34 -; CHECK-NEXT: v_writelane_b32 v0, s66, 35 -; CHECK-NEXT: v_writelane_b32 v0, s67, 36 -; CHECK-NEXT: v_writelane_b32 v0, s68, 37 -; CHECK-NEXT: v_writelane_b32 v0, s69, 38 -; CHECK-NEXT: v_writelane_b32 v0, s70, 39 -; CHECK-NEXT: v_writelane_b32 v0, s71, 40 -; CHECK-NEXT: v_writelane_b32 v0, s72, 41 -; CHECK-NEXT: v_writelane_b32 v0, s73, 42 -; CHECK-NEXT: v_writelane_b32 v0, s74, 43 -; CHECK-NEXT: v_writelane_b32 v0, s75, 44 -; CHECK-NEXT: v_writelane_b32 v0, s76, 45 -; CHECK-NEXT: v_writelane_b32 v0, s77, 46 -; CHECK-NEXT: v_writelane_b32 v0, s78, 47 -; CHECK-NEXT: v_writelane_b32 v0, s79, 48 -; CHECK-NEXT: v_writelane_b32 v0, s80, 49 -; CHECK-NEXT: v_writelane_b32 v0, s81, 50 -; CHECK-NEXT: v_writelane_b32 v0, s82, 51 -; CHECK-NEXT: v_writelane_b32 v0, s83, 52 -; CHECK-NEXT: v_writelane_b32 v0, s84, 53 -; CHECK-NEXT: v_writelane_b32 v0, s85, 54 -; CHECK-NEXT: v_writelane_b32 v0, s86, 55 -; CHECK-NEXT: v_writelane_b32 v0, s87, 56 -; CHECK-NEXT: v_writelane_b32 v0, s88, 57 +; CHECK-NEXT: v_writelane_b32 v0, s33, 0 +; CHECK-NEXT: v_writelane_b32 v0, s34, 1 +; CHECK-NEXT: v_writelane_b32 v0, s35, 2 +; CHECK-NEXT: v_writelane_b32 v0, s36, 3 +; CHECK-NEXT: v_writelane_b32 v0, s37, 4 +; CHECK-NEXT: v_writelane_b32 v0, s38, 5 +; CHECK-NEXT: v_writelane_b32 v0, s39, 6 +; CHECK-NEXT: v_writelane_b32 v0, s40, 7 +; CHECK-NEXT: v_writelane_b32 v0, s41, 8 +; CHECK-NEXT: v_writelane_b32 v0, s42, 9 +; CHECK-NEXT: v_writelane_b32 v0, s43, 10 +; CHECK-NEXT: v_writelane_b32 v0, s44, 11 +; CHECK-NEXT: v_writelane_b32 v0, s45, 12 +; CHECK-NEXT: v_writelane_b32 v0, s46, 13 +; CHECK-NEXT: v_writelane_b32 v0, s47, 14 +; CHECK-NEXT: v_writelane_b32 v0, s48, 15 +; CHECK-NEXT: v_writelane_b32 v0, s49, 16 +; CHECK-NEXT: v_writelane_b32 v0, s50, 17 +; CHECK-NEXT: v_writelane_b32 v0, s51, 18 +; CHECK-NEXT: v_writelane_b32 v0, s52, 19 +; CHECK-NEXT: v_writelane_b32 v0, s53, 20 +; CHECK-NEXT: v_writelane_b32 v0, s54, 21 +; CHECK-NEXT: v_writelane_b32 v0, s55, 22 +; CHECK-NEXT: v_writelane_b32 v0, s56, 23 +; CHECK-NEXT: v_writelane_b32 v0, s57, 24 +; CHECK-NEXT: v_writelane_b32 v0, s58, 25 +; CHECK-NEXT: v_writelane_b32 v0, s59, 26 +; CHECK-NEXT: v_writelane_b32 v0, s60, 27 +; CHECK-NEXT: v_writelane_b32 v0, s61, 28 +; CHECK-NEXT: v_writelane_b32 v0, s62, 29 +; CHECK-NEXT: v_writelane_b32 v0, s63, 30 +; CHECK-NEXT: v_writelane_b32 v0, s64, 31 +; CHECK-NEXT: v_writelane_b32 v0, s65, 32 +; CHECK-NEXT: v_writelane_b32 v0, s66, 33 +; CHECK-NEXT: v_writelane_b32 v0, s67, 34 +; CHECK-NEXT: v_writelane_b32 v0, s68, 35 +; CHECK-NEXT: v_writelane_b32 v0, s69, 36 +; CHECK-NEXT: v_writelane_b32 v0, s70, 37 +; CHECK-NEXT: v_writelane_b32 v0, s71, 38 +; CHECK-NEXT: v_writelane_b32 v0, s72, 39 +; CHECK-NEXT: v_writelane_b32 v0, s73, 40 +; CHECK-NEXT: v_writelane_b32 v0, s74, 41 +; CHECK-NEXT: v_writelane_b32 v0, s75, 42 +; CHECK-NEXT: v_writelane_b32 v0, s76, 43 +; CHECK-NEXT: v_writelane_b32 v0, s77, 44 +; CHECK-NEXT: v_writelane_b32 v0, s78, 45 +; CHECK-NEXT: v_writelane_b32 v0, s79, 46 +; CHECK-NEXT: v_writelane_b32 v0, s80, 47 +; CHECK-NEXT: v_writelane_b32 v0, s81, 48 +; CHECK-NEXT: v_writelane_b32 v0, s82, 49 +; CHECK-NEXT: v_writelane_b32 v0, s83, 50 +; CHECK-NEXT: v_writelane_b32 v0, s84, 51 +; CHECK-NEXT: v_writelane_b32 v0, s85, 52 +; CHECK-NEXT: v_writelane_b32 v0, s86, 53 +; CHECK-NEXT: v_writelane_b32 v0, s87, 54 +; CHECK-NEXT: v_writelane_b32 v0, s88, 55 +; CHECK-NEXT: v_writelane_b32 v0, s89, 56 +; CHECK-NEXT: v_writelane_b32 v0, s90, 57 +; CHECK-NEXT: v_writelane_b32 v0, s91, 58 +; CHECK-NEXT: v_writelane_b32 v0, s92, 59 +; CHECK-NEXT: v_writelane_b32 v0, s93, 60 +; CHECK-NEXT: v_writelane_b32 v0, s94, 61 +; CHECK-NEXT: v_writelane_b32 v0, s95, 62 +; CHECK-NEXT: v_writelane_b32 v0, s96, 63 ; CHECK-NEXT: s_waitcnt expcnt(0) -; CHECK-NEXT: v_writelane_b32 v1, s95, 0 -; CHECK-NEXT: v_writelane_b32 v0, s89, 58 -; CHECK-NEXT: v_writelane_b32 v1, s96, 1 -; CHECK-NEXT: v_writelane_b32 v0, s90, 59 -; CHECK-NEXT: v_writelane_b32 v1, s97, 2 -; CHECK-NEXT: v_writelane_b32 v0, s91, 60 -; CHECK-NEXT: v_writelane_b32 v1, s98, 3 -; CHECK-NEXT: v_writelane_b32 v0, s92, 61 -; CHECK-NEXT: v_writelane_b32 v1, s99, 4 -; CHECK-NEXT: v_writelane_b32 v0, s93, 62 -; CHECK-NEXT: v_writelane_b32 v1, s100, 5 +; CHECK-NEXT: v_writelane_b32 v1, s97, 0 +; CHECK-NEXT: v_writelane_b32 v1, s98, 1 +; CHECK-NEXT: v_writelane_b32 v1, s99, 2 +; CHECK-NEXT: v_writelane_b32 v1, s100, 3 +; CHECK-NEXT: v_writelane_b32 v1, s101, 4 +; CHECK-NEXT: v_writelane_b32 v1, s30, 5 +; CHECK-NEXT: v_writelane_b32 v1, s31, 6 ; CHECK-NEXT: s_mov_b32 s31, s12 -; CHECK-NEXT: v_writelane_b32 v0, s94, 63 -; CHECK-NEXT: v_writelane_b32 v1, s101, 6 ; CHECK-NEXT: s_cmp_eq_u32 s31, 0 ; CHECK-NEXT: ;;#ASMSTART ; CHECK-NEXT: s_mov_b32 s0, 0 @@ -1414,6 +1414,7 @@ ; CHECK-NEXT: ;;#ASMSTART ; CHECK-NEXT: ; reg use s31 ; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: v_readlane_b32 s30, v1, 5 ; CHECK-NEXT: ;;#ASMSTART ; CHECK-NEXT: ; reg use s32 ; CHECK-NEXT: ;;#ASMEND @@ -1630,77 +1631,76 @@ ; CHECK-NEXT: ;;#ASMSTART ; CHECK-NEXT: ; reg use vcc_hi ; CHECK-NEXT: ;;#ASMEND -; CHECK-NEXT: v_readlane_b32 s101, v1, 6 -; CHECK-NEXT: v_readlane_b32 s100, v1, 5 -; CHECK-NEXT: v_readlane_b32 s99, v1, 4 -; CHECK-NEXT: v_readlane_b32 s98, v1, 3 -; CHECK-NEXT: v_readlane_b32 s97, v1, 2 -; CHECK-NEXT: v_readlane_b32 s96, v1, 1 -; CHECK-NEXT: v_readlane_b32 s95, v1, 0 -; CHECK-NEXT: v_readlane_b32 s94, v0, 63 -; CHECK-NEXT: v_readlane_b32 s93, v0, 62 -; CHECK-NEXT: v_readlane_b32 s92, v0, 61 -; CHECK-NEXT: v_readlane_b32 s91, v0, 60 -; CHECK-NEXT: v_readlane_b32 s90, v0, 59 -; CHECK-NEXT: v_readlane_b32 s89, v0, 58 -; CHECK-NEXT: v_readlane_b32 s88, v0, 57 -; CHECK-NEXT: v_readlane_b32 s87, v0, 56 -; CHECK-NEXT: v_readlane_b32 s86, v0, 55 -; CHECK-NEXT: v_readlane_b32 s85, v0, 54 -; CHECK-NEXT: v_readlane_b32 s84, v0, 53 -; CHECK-NEXT: v_readlane_b32 s83, v0, 52 -; CHECK-NEXT: v_readlane_b32 s82, v0, 51 -; CHECK-NEXT: v_readlane_b32 s81, v0, 50 -; CHECK-NEXT: v_readlane_b32 s80, v0, 49 -; CHECK-NEXT: v_readlane_b32 s79, v0, 48 -; CHECK-NEXT: v_readlane_b32 s78, v0, 47 -; CHECK-NEXT: v_readlane_b32 s77, v0, 46 -; CHECK-NEXT: v_readlane_b32 s76, v0, 45 -; CHECK-NEXT: v_readlane_b32 s75, v0, 44 -; CHECK-NEXT: v_readlane_b32 s74, v0, 43 -; CHECK-NEXT: v_readlane_b32 s73, v0, 42 -; CHECK-NEXT: v_readlane_b32 s72, v0, 41 -; CHECK-NEXT: v_readlane_b32 s71, v0, 40 -; CHECK-NEXT: v_readlane_b32 s70, v0, 39 -; CHECK-NEXT: v_readlane_b32 s69, v0, 38 -; CHECK-NEXT: v_readlane_b32 s68, v0, 37 -; CHECK-NEXT: v_readlane_b32 s67, v0, 36 -; CHECK-NEXT: v_readlane_b32 s66, v0, 35 -; CHECK-NEXT: v_readlane_b32 s65, v0, 34 -; CHECK-NEXT: v_readlane_b32 s64, v0, 33 -; CHECK-NEXT: v_readlane_b32 s63, v0, 32 -; CHECK-NEXT: v_readlane_b32 s62, v0, 31 -; CHECK-NEXT: v_readlane_b32 s61, v0, 30 -; CHECK-NEXT: v_readlane_b32 s60, v0, 29 -; CHECK-NEXT: v_readlane_b32 s59, v0, 28 -; CHECK-NEXT: v_readlane_b32 s58, v0, 27 -; CHECK-NEXT: v_readlane_b32 s57, v0, 26 -; CHECK-NEXT: v_readlane_b32 s56, v0, 25 -; CHECK-NEXT: v_readlane_b32 s55, v0, 24 -; CHECK-NEXT: v_readlane_b32 s54, v0, 23 -; CHECK-NEXT: v_readlane_b32 s53, v0, 22 -; CHECK-NEXT: v_readlane_b32 s52, v0, 21 -; CHECK-NEXT: v_readlane_b32 s51, v0, 20 -; CHECK-NEXT: v_readlane_b32 s50, v0, 19 -; CHECK-NEXT: v_readlane_b32 s49, v0, 18 -; CHECK-NEXT: v_readlane_b32 s48, v0, 17 -; CHECK-NEXT: v_readlane_b32 s47, v0, 16 -; CHECK-NEXT: v_readlane_b32 s46, v0, 15 -; CHECK-NEXT: v_readlane_b32 s45, v0, 14 -; CHECK-NEXT: v_readlane_b32 s44, v0, 13 -; CHECK-NEXT: v_readlane_b32 s43, v0, 12 -; CHECK-NEXT: v_readlane_b32 s42, v0, 11 -; CHECK-NEXT: v_readlane_b32 s41, v0, 10 -; CHECK-NEXT: v_readlane_b32 s40, v0, 9 -; CHECK-NEXT: v_readlane_b32 s39, v0, 8 -; CHECK-NEXT: v_readlane_b32 s38, v0, 7 -; CHECK-NEXT: v_readlane_b32 s37, v0, 6 -; CHECK-NEXT: v_readlane_b32 s36, v0, 5 -; CHECK-NEXT: v_readlane_b32 s35, v0, 4 -; CHECK-NEXT: v_readlane_b32 s34, v0, 3 -; CHECK-NEXT: v_readlane_b32 s33, v0, 2 -; CHECK-NEXT: v_readlane_b32 s31, v0, 1 -; CHECK-NEXT: v_readlane_b32 s30, v0, 0 +; CHECK-NEXT: v_readlane_b32 s31, v1, 6 +; CHECK-NEXT: v_readlane_b32 s101, v1, 4 +; CHECK-NEXT: v_readlane_b32 s100, v1, 3 +; CHECK-NEXT: v_readlane_b32 s99, v1, 2 +; CHECK-NEXT: v_readlane_b32 s98, v1, 1 +; CHECK-NEXT: v_readlane_b32 s97, v1, 0 +; CHECK-NEXT: v_readlane_b32 s96, v0, 63 +; CHECK-NEXT: v_readlane_b32 s95, v0, 62 +; CHECK-NEXT: v_readlane_b32 s94, v0, 61 +; CHECK-NEXT: v_readlane_b32 s93, v0, 60 +; CHECK-NEXT: v_readlane_b32 s92, v0, 59 +; CHECK-NEXT: v_readlane_b32 s91, v0, 58 +; CHECK-NEXT: v_readlane_b32 s90, v0, 57 +; CHECK-NEXT: v_readlane_b32 s89, v0, 56 +; CHECK-NEXT: v_readlane_b32 s88, v0, 55 +; CHECK-NEXT: v_readlane_b32 s87, v0, 54 +; CHECK-NEXT: v_readlane_b32 s86, v0, 53 +; CHECK-NEXT: v_readlane_b32 s85, v0, 52 +; CHECK-NEXT: v_readlane_b32 s84, v0, 51 +; CHECK-NEXT: v_readlane_b32 s83, v0, 50 +; CHECK-NEXT: v_readlane_b32 s82, v0, 49 +; CHECK-NEXT: v_readlane_b32 s81, v0, 48 +; CHECK-NEXT: v_readlane_b32 s80, v0, 47 +; CHECK-NEXT: v_readlane_b32 s79, v0, 46 +; CHECK-NEXT: v_readlane_b32 s78, v0, 45 +; CHECK-NEXT: v_readlane_b32 s77, v0, 44 +; CHECK-NEXT: v_readlane_b32 s76, v0, 43 +; CHECK-NEXT: v_readlane_b32 s75, v0, 42 +; CHECK-NEXT: v_readlane_b32 s74, v0, 41 +; CHECK-NEXT: v_readlane_b32 s73, v0, 40 +; CHECK-NEXT: v_readlane_b32 s72, v0, 39 +; CHECK-NEXT: v_readlane_b32 s71, v0, 38 +; CHECK-NEXT: v_readlane_b32 s70, v0, 37 +; CHECK-NEXT: v_readlane_b32 s69, v0, 36 +; CHECK-NEXT: v_readlane_b32 s68, v0, 35 +; CHECK-NEXT: v_readlane_b32 s67, v0, 34 +; CHECK-NEXT: v_readlane_b32 s66, v0, 33 +; CHECK-NEXT: v_readlane_b32 s65, v0, 32 +; CHECK-NEXT: v_readlane_b32 s64, v0, 31 +; CHECK-NEXT: v_readlane_b32 s63, v0, 30 +; CHECK-NEXT: v_readlane_b32 s62, v0, 29 +; CHECK-NEXT: v_readlane_b32 s61, v0, 28 +; CHECK-NEXT: v_readlane_b32 s60, v0, 27 +; CHECK-NEXT: v_readlane_b32 s59, v0, 26 +; CHECK-NEXT: v_readlane_b32 s58, v0, 25 +; CHECK-NEXT: v_readlane_b32 s57, v0, 24 +; CHECK-NEXT: v_readlane_b32 s56, v0, 23 +; CHECK-NEXT: v_readlane_b32 s55, v0, 22 +; CHECK-NEXT: v_readlane_b32 s54, v0, 21 +; CHECK-NEXT: v_readlane_b32 s53, v0, 20 +; CHECK-NEXT: v_readlane_b32 s52, v0, 19 +; CHECK-NEXT: v_readlane_b32 s51, v0, 18 +; CHECK-NEXT: v_readlane_b32 s50, v0, 17 +; CHECK-NEXT: v_readlane_b32 s49, v0, 16 +; CHECK-NEXT: v_readlane_b32 s48, v0, 15 +; CHECK-NEXT: v_readlane_b32 s47, v0, 14 +; CHECK-NEXT: v_readlane_b32 s46, v0, 13 +; CHECK-NEXT: v_readlane_b32 s45, v0, 12 +; CHECK-NEXT: v_readlane_b32 s44, v0, 11 +; CHECK-NEXT: v_readlane_b32 s43, v0, 10 +; CHECK-NEXT: v_readlane_b32 s42, v0, 9 +; CHECK-NEXT: v_readlane_b32 s41, v0, 8 +; CHECK-NEXT: v_readlane_b32 s40, v0, 7 +; CHECK-NEXT: v_readlane_b32 s39, v0, 6 +; CHECK-NEXT: v_readlane_b32 s38, v0, 5 +; CHECK-NEXT: v_readlane_b32 s37, v0, 4 +; CHECK-NEXT: v_readlane_b32 s36, v0, 3 +; CHECK-NEXT: v_readlane_b32 s35, v0, 2 +; CHECK-NEXT: v_readlane_b32 s34, v0, 1 +; CHECK-NEXT: v_readlane_b32 s33, v0, 0 ; CHECK-NEXT: s_or_saveexec_b64 s[4:5], -1 ; CHECK-NEXT: buffer_load_dword v0, off, s[0:3], s32 ; 4-byte Folded Reload ; CHECK-NEXT: buffer_load_dword v1, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload diff --git a/llvm/test/CodeGen/AMDGPU/call-graph-register-usage.ll b/llvm/test/CodeGen/AMDGPU/call-graph-register-usage.ll --- a/llvm/test/CodeGen/AMDGPU/call-graph-register-usage.ll +++ b/llvm/test/CodeGen/AMDGPU/call-graph-register-usage.ll @@ -17,8 +17,8 @@ ; GCN: v_writelane_b32 v40, s30, 0 ; GCN: v_writelane_b32 v40, s31, 1 ; GCN: s_swappc_b64 -; GCN: v_readlane_b32 s31, v40, 1 -; GCN: v_readlane_b32 s30, v40, 0 +; GCN-DAG: v_readlane_b32 s31, v40, 1 +; GCN-DAG: v_readlane_b32 s30, v40, 0 ; GCN: v_readlane_b32 s33, v40, 2 ; GCN: s_setpc_b64 s[30:31] ; GCN: ; NumSgprs: 36 diff --git a/llvm/test/CodeGen/AMDGPU/call-preserved-registers.ll b/llvm/test/CodeGen/AMDGPU/call-preserved-registers.ll --- a/llvm/test/CodeGen/AMDGPU/call-preserved-registers.ll +++ b/llvm/test/CodeGen/AMDGPU/call-preserved-registers.ll @@ -25,21 +25,21 @@ ; MUBUF: buffer_store_dword ; FLATSCR: scratch_store_dword ; GCN: v_writelane_b32 v40, s33, 4 -; GCN: v_writelane_b32 v40, s30, 0 -; GCN: v_writelane_b32 v40, s31, 1 -; GCN: v_writelane_b32 v40, s34, 2 -; GCN: v_writelane_b32 v40, s35, 3 +; GCN: v_writelane_b32 v40, s34, 0 +; GCN: v_writelane_b32 v40, s35, 1 +; GCN: v_writelane_b32 v40, s30, 2 +; GCN: v_writelane_b32 v40, s31, 3 ; GCN: s_swappc_b64 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: ;;#ASMEND ; GCN-NEXT: s_swappc_b64 -; GCN: v_readlane_b32 s35, v40, 3 -; GCN: v_readlane_b32 s34, v40, 2 -; MUBUF-DAG: v_readlane_b32 s31, v40, 1 -; MUBUF-DAG: v_readlane_b32 s30, v40, 0 -; FLATSCR-DAG: v_readlane_b32 s31, v40, 1 -; FLATSCR-DAG: v_readlane_b32 s30, v40, 0 +; MUBUF-DAG: v_readlane_b32 s31, v40, 3 +; MUBUF-DAG: v_readlane_b32 s30, v40, 2 +; FLATSCR-DAG: v_readlane_b32 s31, v40, 3 +; FLATSCR-DAG: v_readlane_b32 s30, v40, 2 +; GCN: v_readlane_b32 s35, v40, 1 +; GCN: v_readlane_b32 s34, v40, 0 ; GCN: v_readlane_b32 s33, v40, 4 ; MUBUF: buffer_load_dword @@ -79,8 +79,8 @@ ; GCN-NEXT: #ASMSTART ; GCN: ; clobber ; GCN-NEXT: #ASMEND -; GCN: v_readlane_b32 s31, v0, 1 -; GCN: v_readlane_b32 s30, v0, 0 +; GCN-DAG: v_readlane_b32 s30, v0, 0 +; GCN-DAG: v_readlane_b32 s31, v0, 1 ; GCN: s_setpc_b64 s[30:31] define void @void_func_void_clobber_s30_s31() #2 { call void asm sideeffect "; clobber", "~{s[30:31]}"() #0 diff --git a/llvm/test/CodeGen/AMDGPU/callee-frame-setup.ll b/llvm/test/CodeGen/AMDGPU/callee-frame-setup.ll --- a/llvm/test/CodeGen/AMDGPU/callee-frame-setup.ll +++ b/llvm/test/CodeGen/AMDGPU/callee-frame-setup.ll @@ -235,9 +235,11 @@ ; GCN: s_waitcnt ; GCN-NEXT:s_mov_b32 [[FP_COPY:s[0-9]+]], s33 ; GCN-NEXT: s_mov_b32 s33, s32 -; GCN: v_mov_b32_e32 [[ZERO:v[0-9]+]], 0 +; MUBUF: s_addk_i32 s32, 0x300 +; FLATSCR: s_add_i32 s32, s32, 12 ; MUBUF-DAG: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill ; FLATSCR-DAG: scratch_store_dword off, v41, s33 ; 4-byte Folded Spill +; GCN: v_mov_b32_e32 [[ZERO:v[0-9]+]], 0 ; MUBUF-DAG: buffer_store_dword [[ZERO]], off, s[0:3], s33 offset:4 ; FLATSCR-DAG: scratch_store_dword off, [[ZERO]], s33 offset:4 @@ -247,10 +249,8 @@ ; MUBUF: buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload ; FLATSCR: scratch_load_dword v41, off, s33 ; 4-byte Folded Reload -; MUBUF: s_addk_i32 s32, 0x300 ; MUBUF-NEXT: s_addk_i32 s32, 0xfd00 ; MUBUF-NEXT: s_mov_b32 s33, s4 -; FLATSCR: s_add_i32 s32, s32, 12 ; FLATSCR-NEXT: s_add_i32 s32, s32, -12 ; FLATSCR-NEXT: s_mov_b32 s33, s0 ; GCN-NEXT: s_waitcnt vmcnt(0) @@ -271,19 +271,16 @@ ; GCN-NEXT: s_mov_b64 exec, [[COPY_EXEC0]] ; GCN-NEXT: v_writelane_b32 v0, s33, 63 ; GCN: s_mov_b32 s33, s32 -; GCN-COUNT-60: v_writelane_b32 v0 -; GCN: v_writelane_b32 v0 +; MUBUF: s_addk_i32 s32, 0x400 +; FLATSCR: s_add_i32 s32, s32, 16 ; MUBUF: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill ; FLATSCR: scratch_store_dword off, v41, s33 ; 4-byte Folded Spill -; GCN: v_writelane_b32 v0 +; GCN-COUNT-63: v_writelane_b32 v0 ; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[0:3], s33 offset:4 ; FLATSCR: scratch_store_dword off, v{{[0-9]+}}, s33 offset:4 ; GCN: ;;#ASMSTART -; GCN: v_writelane_b32 v0 -; MUBUF: s_addk_i32 s32, 0x400 ; MUBUF: s_addk_i32 s32, 0xfc00 -; FLATSCR: s_add_i32 s32, s32, 16 ; FLATSCR: s_add_i32 s32, s32, -16 ; GCN-NEXT: v_readlane_b32 s33, v0, 63 ; GCN-NEXT: s_or_saveexec_b64 [[COPY_EXEC1:s\[[0-9]+:[0-9]+\]]], -1{{$}} @@ -319,18 +316,16 @@ ; FLATSCR-NEXT: s_mov_b32 s33, s32 ; MUBUF: s_mov_b32 [[FP_COPY:s[0-9]+]], s33 ; MUBUF-NEXT: s_mov_b32 s33, s32 -; GCN-COUNT-61: v_writelane_b32 v0, +; MUBUF: s_addk_i32 s32, 0x400 +; FLATSCR: s_add_i32 s32, s32, 16 ; MUBUF: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill ; FLATSCR: scratch_store_dword off, v41, s33 ; 4-byte Folded Spill -; GCN: v_writelane_b32 v0, +; GCN-COUNT-63: v_writelane_b32 v0, ; MUBUF: buffer_store_dword ; FLATSCR: scratch_store_dword ; GCN: ;;#ASMSTART -; GCN: v_writelane_b32 v0, ; MUBUF: buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload ; FLATSCR: scratch_load_dword v41, off, s33 ; 4-byte Folded Reload -; MUBUF: s_addk_i32 s32, 0x400 -; FLATSCR: s_add_i32 s32, s32, 16 ; GCN-COUNT-64: v_readlane_b32 s{{[0-9]+}}, v0 ; MUBUF-NEXT: s_addk_i32 s32, 0xfc00 ; FLATSCR-NEXT: s_add_i32 s32, s32, -16 @@ -392,16 +387,16 @@ ; GCN-NEXT: v_writelane_b32 [[CSR_VGPR]], s33, 2 ; GCN-NEXT: s_mov_b32 s33, s32 ; GCN: v_writelane_b32 [[CSR_VGPR]], s30, 0 -; GCN: v_mov_b32_e32 [[ZERO:v[0-9]+]], 0 ; MUBUF: s_addk_i32 s32, 0x300 ; FLATSCR: s_add_i32 s32, s32, 12 ; GCN: v_writelane_b32 [[CSR_VGPR]], s31, 1 +; GCN: v_mov_b32_e32 [[ZERO:v[0-9]+]], 0 ; MUBUF: buffer_store_dword [[ZERO]], off, s[0:3], s33{{$}} ; FLATSCR: scratch_store_dword off, [[ZERO]], s33{{$}} ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN: ;;#ASMSTART -; GCN: v_readlane_b32 s31, [[CSR_VGPR]], 1 -; GCN: v_readlane_b32 s30, [[CSR_VGPR]], 0 +; GCN-DAG: v_readlane_b32 s31, [[CSR_VGPR]], 1 +; GCN-DAG: v_readlane_b32 s30, [[CSR_VGPR]], 0 ; MUBUF: s_addk_i32 s32, 0xfd00 ; FLATSCR: s_add_i32 s32, s32, -12 ; GCN-NEXT: v_readlane_b32 s33, [[CSR_VGPR]], 2 diff --git a/llvm/test/CodeGen/AMDGPU/cross-block-use-is-not-abi-copy.ll b/llvm/test/CodeGen/AMDGPU/cross-block-use-is-not-abi-copy.ll --- a/llvm/test/CodeGen/AMDGPU/cross-block-use-is-not-abi-copy.ll +++ b/llvm/test/CodeGen/AMDGPU/cross-block-use-is-not-abi-copy.ll @@ -32,15 +32,15 @@ ; GCN-NEXT: s_mov_b64 exec, s[16:17] ; GCN-NEXT: v_writelane_b32 v40, s33, 2 ; GCN-NEXT: s_mov_b32 s33, s32 -; GCN-NEXT: s_addk_i32 s32, 0x400 ; GCN-NEXT: v_writelane_b32 v40, s30, 0 +; GCN-NEXT: s_addk_i32 s32, 0x400 ; GCN-NEXT: v_writelane_b32 v40, s31, 1 ; GCN-NEXT: s_getpc_b64 s[16:17] ; GCN-NEXT: s_add_u32 s16, s16, func_v2f32@rel32@lo+4 ; GCN-NEXT: s_addc_u32 s17, s17, func_v2f32@rel32@hi+12 ; GCN-NEXT: s_swappc_b64 s[30:31], s[16:17] -; GCN-NEXT: v_readlane_b32 s31, v40, 1 ; GCN-NEXT: v_readlane_b32 s30, v40, 0 +; GCN-NEXT: v_readlane_b32 s31, v40, 1 ; GCN-NEXT: s_addk_i32 s32, 0xfc00 ; GCN-NEXT: v_readlane_b32 s33, v40, 2 ; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -66,15 +66,15 @@ ; GCN-NEXT: s_mov_b64 exec, s[16:17] ; GCN-NEXT: v_writelane_b32 v40, s33, 2 ; GCN-NEXT: s_mov_b32 s33, s32 -; GCN-NEXT: s_addk_i32 s32, 0x400 ; GCN-NEXT: v_writelane_b32 v40, s30, 0 +; GCN-NEXT: s_addk_i32 s32, 0x400 ; GCN-NEXT: v_writelane_b32 v40, s31, 1 ; GCN-NEXT: s_getpc_b64 s[16:17] ; GCN-NEXT: s_add_u32 s16, s16, func_v3f32@rel32@lo+4 ; GCN-NEXT: s_addc_u32 s17, s17, func_v3f32@rel32@hi+12 ; GCN-NEXT: s_swappc_b64 s[30:31], s[16:17] -; GCN-NEXT: v_readlane_b32 s31, v40, 1 ; GCN-NEXT: v_readlane_b32 s30, v40, 0 +; GCN-NEXT: v_readlane_b32 s31, v40, 1 ; GCN-NEXT: s_addk_i32 s32, 0xfc00 ; GCN-NEXT: v_readlane_b32 s33, v40, 2 ; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -100,15 +100,15 @@ ; GCN-NEXT: s_mov_b64 exec, s[16:17] ; GCN-NEXT: v_writelane_b32 v40, s33, 2 ; GCN-NEXT: s_mov_b32 s33, s32 -; GCN-NEXT: s_addk_i32 s32, 0x400 ; GCN-NEXT: v_writelane_b32 v40, s30, 0 +; GCN-NEXT: s_addk_i32 s32, 0x400 ; GCN-NEXT: v_writelane_b32 v40, s31, 1 ; GCN-NEXT: s_getpc_b64 s[16:17] ; GCN-NEXT: s_add_u32 s16, s16, func_v4f16@rel32@lo+4 ; GCN-NEXT: s_addc_u32 s17, s17, func_v4f16@rel32@hi+12 ; GCN-NEXT: s_swappc_b64 s[30:31], s[16:17] -; GCN-NEXT: v_readlane_b32 s31, v40, 1 ; GCN-NEXT: v_readlane_b32 s30, v40, 0 +; GCN-NEXT: v_readlane_b32 s31, v40, 1 ; GCN-NEXT: s_addk_i32 s32, 0xfc00 ; GCN-NEXT: v_readlane_b32 s33, v40, 2 ; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -134,16 +134,16 @@ ; GCN-NEXT: s_mov_b64 exec, s[16:17] ; GCN-NEXT: v_writelane_b32 v40, s33, 2 ; GCN-NEXT: s_mov_b32 s33, s32 -; GCN-NEXT: s_addk_i32 s32, 0x400 ; GCN-NEXT: v_writelane_b32 v40, s30, 0 +; GCN-NEXT: s_addk_i32 s32, 0x400 ; GCN-NEXT: v_writelane_b32 v40, s31, 1 ; GCN-NEXT: s_getpc_b64 s[16:17] ; GCN-NEXT: s_add_u32 s16, s16, func_struct@rel32@lo+4 ; GCN-NEXT: s_addc_u32 s17, s17, func_struct@rel32@hi+12 ; GCN-NEXT: s_swappc_b64 s[30:31], s[16:17] +; GCN-NEXT: v_readlane_b32 s30, v40, 0 ; GCN-NEXT: v_mov_b32_e32 v1, v4 ; GCN-NEXT: v_readlane_b32 s31, v40, 1 -; GCN-NEXT: v_readlane_b32 s30, v40, 0 ; GCN-NEXT: s_addk_i32 s32, 0xfc00 ; GCN-NEXT: v_readlane_b32 s33, v40, 2 ; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1 diff --git a/llvm/test/CodeGen/AMDGPU/csr-sgpr-spill-live-ins.mir b/llvm/test/CodeGen/AMDGPU/csr-sgpr-spill-live-ins.mir --- a/llvm/test/CodeGen/AMDGPU/csr-sgpr-spill-live-ins.mir +++ b/llvm/test/CodeGen/AMDGPU/csr-sgpr-spill-live-ins.mir @@ -24,9 +24,13 @@ ; CHECK-NEXT: frame-setup CFI_INSTRUCTION offset $vgpr0_lo16, 0 ; CHECK-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5 ; CHECK-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr42, 0, $vgpr0 + ; CHECK-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x4a, 0x05, 0x90, 0x80, 0x14, 0xec, 0x00 ; CHECK-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr43, 1, $vgpr0 + ; CHECK-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x4b, 0x05, 0x90, 0x80, 0x14, 0xec, 0x04 ; CHECK-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr46, 2, $vgpr0 + ; CHECK-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x4e, 0x05, 0x90, 0x80, 0x14, 0xec, 0x08 ; CHECK-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr47, 3, $vgpr0 + ; CHECK-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x4f, 0x05, 0x90, 0x80, 0x14, 0xec, 0x0c ; CHECK-NEXT: S_NOP 0 ; CHECK-NEXT: {{ $}} ; CHECK-NEXT: bb.1: diff --git a/llvm/test/CodeGen/AMDGPU/debug-frame.ll b/llvm/test/CodeGen/AMDGPU/debug-frame.ll --- a/llvm/test/CodeGen/AMDGPU/debug-frame.ll +++ b/llvm/test/CodeGen/AMDGPU/debug-frame.ll @@ -1,4 +1,7 @@ -; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -filetype=asm -o - %s | FileCheck %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -filetype=asm -o - %s | FileCheck --check-prefixes=CHECK,WAVE64,GFX900 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a -amdgpu-spill-vgpr-to-agpr=0 -filetype=asm -o - %s | FileCheck --check-prefixes=CHECK,WAVE64,GFX90A-V2A-DIS %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a -amdgpu-spill-vgpr-to-agpr=1 -filetype=asm -o - %s | FileCheck --check-prefixes=CHECK,WAVE64,GFX90A-V2A-EN %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -mattr=+wavefrontsize32,-wavefrontsize64 -filetype=asm -o - %s | FileCheck --check-prefixes=CHECK,WAVE32 %s ; CHECK-LABEL: kern1: ; CHECK: .cfi_startproc @@ -23,7 +26,779 @@ ret void } -attributes #0 = { nounwind } +; CHECK-LABEL: func_no_clobber: +; CHECK: .cfi_startproc + +; CHECK-NOT: .cfi_{{.*}} + +; CHECK: %bb.0: +; SGPR32 = 64 +; CHECK-NEXT: .cfi_llvm_def_aspace_cfa 64, 0, 6 +; DW_CFA_expression [0x10] +; PC_64 ULEB128(17)=[0x10] +; BLOCK_LENGTH ULEB128(8)=[0x08] +; DW_OP_regx [0x90] +; SGPR30 ULEB128(62)=[0x3e] +; DW_OP_piece [0x93] +; PIECE_SIZE [0x04] +; DW_OP_regx [0x90] +; SGPR31 ULEB128(63)=[0x3f] +; DW_OP_piece [0x93] +; PIECE_SIZE [0x04] +; CHECK-NEXT: .cfi_escape 0x10, 0x10, 0x08, 0x90, 0x3e, 0x93, 0x04, 0x90, 0x3f, 0x93, 0x04 + +; CHECK-NOT: .cfi_{{.*}} + +; CHECK: .cfi_endproc +define hidden void @func_no_clobber() #0 { +entry: + ret void +} + +; CHECK-LABEL: {{^}}callee_need_to_spill_fp_to_memory: +; CHECK: .cfi_startproc + +; SGPR33 = 65 +; CHECK: v_mov_b32_e32 [[TMP_VGPR:v[0-9]+]], s33 +; GFX900: buffer_store_dword [[TMP_VGPR]], off, s[0:3], s32 offset:448 ; 4-byte Folded Spill +; GFX90A-V2A-DIS: buffer_store_dword [[TMP_VGPR]], off, s[0:3], s32 offset:448 ; 4-byte Folded Spill +; GFX90A-V2A-EN: buffer_store_dword [[TMP_VGPR]], off, s[0:3], s32 offset:320 ; 4-byte Folded Spill + +; GFX900: .cfi_offset 65, 28672 +; GFX90A-V2A-DIS: .cfi_offset 65, 28672 +; GFX90A-V2A-EN: .cfi_offset 65, 20480 +; WAVE32: .cfi_offset 65, 14336 + +; CHECK: .cfi_endproc +define void @callee_need_to_spill_fp_to_memory() #1 { + call void asm sideeffect "; clobber nonpreserved SGPRs", + "~{s4},~{s5},~{s6},~{s7},~{s8},~{s9} + ,~{s10},~{s11},~{s12},~{s13},~{s14},~{s15},~{s16},~{s17},~{s18},~{s19} + ,~{s20},~{s21},~{s22},~{s23},~{s24},~{s25},~{s26},~{s27},~{s28},~{s29} + ,~{vcc}"() + + call void asm sideeffect "; clobber all VGPRs", + "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7},~{v8},~{v9} + ,~{v10},~{v11},~{v12},~{v13},~{v14},~{v15},~{v16},~{v17},~{v18},~{v19} + ,~{v20},~{v21},~{v22},~{v23},~{v24},~{v25},~{v26},~{v27},~{v28},~{v29} + ,~{v30},~{v31},~{v32},~{v33},~{v34},~{v35},~{v36},~{v37},~{v38},~{v39} + ,~{v40},~{v41},~{v42},~{v43},~{v44},~{v45},~{v46},~{v47},~{v48},~{v49} + ,~{v50},~{v51},~{v52},~{v53},~{v54},~{v55},~{v56},~{v57},~{v58},~{v59} + ,~{v60},~{v61},~{v62},~{v63},~{v64},~{v65},~{v66},~{v67},~{v68},~{v69} + ,~{v70},~{v71},~{v72},~{v73},~{v74},~{v75},~{v76},~{v77},~{v78},~{v79} + ,~{v80},~{v81},~{v82},~{v83},~{v84},~{v85},~{v86},~{v87},~{v88},~{v89} + ,~{v90},~{v91},~{v92},~{v93},~{v94},~{v95},~{v96},~{v97},~{v98},~{v99} + ,~{v100},~{v101},~{v102},~{v103},~{v104},~{v105},~{v106},~{v107},~{v108},~{v109} + ,~{v110},~{v111},~{v112},~{v113},~{v114},~{v115},~{v116},~{v117},~{v118},~{v119} + ,~{v120},~{v121},~{v122},~{v123},~{v124},~{v125},~{v126},~{v127},~{v128},~{v129} + ,~{v130},~{v131},~{v132},~{v133},~{v134},~{v135},~{v136},~{v137},~{v138},~{v139} + ,~{v140},~{v141},~{v142},~{v143},~{v144},~{v145},~{v146},~{v147},~{v148},~{v149} + ,~{v150},~{v151},~{v152},~{v153},~{v154},~{v155},~{v156},~{v157},~{v158},~{v159} + ,~{v160},~{v161},~{v162},~{v163},~{v164},~{v165},~{v166},~{v167},~{v168},~{v169} + ,~{v170},~{v171},~{v172},~{v173},~{v174},~{v175},~{v176},~{v177},~{v178},~{v179} + ,~{v180},~{v181},~{v182},~{v183},~{v184},~{v185},~{v186},~{v187},~{v188},~{v189} + ,~{v190},~{v191},~{v192},~{v193},~{v194},~{v195},~{v196},~{v197},~{v198},~{v199} + ,~{v200},~{v201},~{v202},~{v203},~{v204},~{v205},~{v206},~{v207},~{v208},~{v209} + ,~{v210},~{v211},~{v212},~{v213},~{v214},~{v215},~{v216},~{v217},~{v218},~{v219} + ,~{v220},~{v221},~{v222},~{v223},~{v224},~{v225},~{v226},~{v227},~{v228},~{v229} + ,~{v230},~{v231},~{v232},~{v233},~{v234},~{v235},~{v236},~{v237},~{v238},~{v239} + ,~{v240},~{v241},~{v242},~{v243},~{v244},~{v245},~{v246},~{v247},~{v248},~{v249} + ,~{v250},~{v251},~{v252},~{v253},~{v254},~{v255}"() + ret void +} + +declare hidden void @ex() #0 + +; CHECK-LABEL: func_call_clobber: +; CHECK: .cfi_startproc + +; CHECK-NOT: .cfi_{{.*}} + +; CHECK: %bb.0: +; CHECK-NEXT: .cfi_llvm_def_aspace_cfa 64, 0, 6 +; CHECK-NEXT: .cfi_escape 0x10, 0x10, 0x08, 0x90, 0x3e, 0x93, 0x04, 0x90, 0x3f, 0x93, 0x04 + +; VGPR0_wave64 = 2560 +; WAVE64-NEXT: .cfi_undefined 2560 +; WAVE64-NEXT: .cfi_undefined 2561 +; WAVE64-NEXT: .cfi_undefined 2562 +; WAVE64-NEXT: .cfi_undefined 2563 +; WAVE64-NEXT: .cfi_undefined 2564 +; WAVE64-NEXT: .cfi_undefined 2565 +; WAVE64-NEXT: .cfi_undefined 2566 +; WAVE64-NEXT: .cfi_undefined 2567 +; WAVE64-NEXT: .cfi_undefined 2568 +; WAVE64-NEXT: .cfi_undefined 2569 +; WAVE64-NEXT: .cfi_undefined 2570 +; WAVE64-NEXT: .cfi_undefined 2571 +; WAVE64-NEXT: .cfi_undefined 2572 +; WAVE64-NEXT: .cfi_undefined 2573 +; WAVE64-NEXT: .cfi_undefined 2574 +; WAVE64-NEXT: .cfi_undefined 2575 +; WAVE64-NEXT: .cfi_undefined 2576 +; WAVE64-NEXT: .cfi_undefined 2577 +; WAVE64-NEXT: .cfi_undefined 2578 +; WAVE64-NEXT: .cfi_undefined 2579 +; WAVE64-NEXT: .cfi_undefined 2580 +; WAVE64-NEXT: .cfi_undefined 2581 +; WAVE64-NEXT: .cfi_undefined 2582 +; WAVE64-NEXT: .cfi_undefined 2583 +; WAVE64-NEXT: .cfi_undefined 2584 +; WAVE64-NEXT: .cfi_undefined 2585 +; WAVE64-NEXT: .cfi_undefined 2586 +; WAVE64-NEXT: .cfi_undefined 2587 +; WAVE64-NEXT: .cfi_undefined 2588 +; WAVE64-NEXT: .cfi_undefined 2589 +; WAVE64-NEXT: .cfi_undefined 2590 +; WAVE64-NEXT: .cfi_undefined 2591 +; WAVE64-NEXT: .cfi_undefined 2592 +; WAVE64-NEXT: .cfi_undefined 2593 +; WAVE64-NEXT: .cfi_undefined 2594 +; WAVE64-NEXT: .cfi_undefined 2595 +; WAVE64-NEXT: .cfi_undefined 2596 +; WAVE64-NEXT: .cfi_undefined 2597 +; WAVE64-NEXT: .cfi_undefined 2598 +; WAVE64-NEXT: .cfi_undefined 2599 + +; VPGR48_wave64 = 2608 +; WAVE64-NEXT: .cfi_undefined 2608 +; WAVE64-NEXT: .cfi_undefined 2609 +; WAVE64-NEXT: .cfi_undefined 2610 +; WAVE64-NEXT: .cfi_undefined 2611 +; WAVE64-NEXT: .cfi_undefined 2612 +; WAVE64-NEXT: .cfi_undefined 2613 +; WAVE64-NEXT: .cfi_undefined 2614 +; WAVE64-NEXT: .cfi_undefined 2615 + +; WAVE64-NEXT: .cfi_undefined 2624 +; WAVE64-NEXT: .cfi_undefined 2625 +; WAVE64-NEXT: .cfi_undefined 2626 +; WAVE64-NEXT: .cfi_undefined 2627 +; WAVE64-NEXT: .cfi_undefined 2628 +; WAVE64-NEXT: .cfi_undefined 2629 +; WAVE64-NEXT: .cfi_undefined 2630 +; WAVE64-NEXT: .cfi_undefined 2631 + +; WAVE64-NEXT: .cfi_undefined 2640 +; WAVE64-NEXT: .cfi_undefined 2641 +; WAVE64-NEXT: .cfi_undefined 2642 +; WAVE64-NEXT: .cfi_undefined 2643 +; WAVE64-NEXT: .cfi_undefined 2644 +; WAVE64-NEXT: .cfi_undefined 2645 +; WAVE64-NEXT: .cfi_undefined 2646 +; WAVE64-NEXT: .cfi_undefined 2647 + +; WAVE64-NEXT: .cfi_undefined 2656 +; WAVE64-NEXT: .cfi_undefined 2657 +; WAVE64-NEXT: .cfi_undefined 2658 +; WAVE64-NEXT: .cfi_undefined 2659 +; WAVE64-NEXT: .cfi_undefined 2660 +; WAVE64-NEXT: .cfi_undefined 2661 +; WAVE64-NEXT: .cfi_undefined 2662 +; WAVE64-NEXT: .cfi_undefined 2663 + +; WAVE64-NEXT: .cfi_undefined 2672 +; WAVE64-NEXT: .cfi_undefined 2673 +; WAVE64-NEXT: .cfi_undefined 2674 +; WAVE64-NEXT: .cfi_undefined 2675 +; WAVE64-NEXT: .cfi_undefined 2676 +; WAVE64-NEXT: .cfi_undefined 2677 +; WAVE64-NEXT: .cfi_undefined 2678 +; WAVE64-NEXT: .cfi_undefined 2679 + +; WAVE64-NEXT: .cfi_undefined 2688 +; WAVE64-NEXT: .cfi_undefined 2689 +; WAVE64-NEXT: .cfi_undefined 2690 +; WAVE64-NEXT: .cfi_undefined 2691 +; WAVE64-NEXT: .cfi_undefined 2692 +; WAVE64-NEXT: .cfi_undefined 2693 +; WAVE64-NEXT: .cfi_undefined 2694 +; WAVE64-NEXT: .cfi_undefined 2695 + +; WAVE64-NEXT: .cfi_undefined 2704 +; WAVE64-NEXT: .cfi_undefined 2705 +; WAVE64-NEXT: .cfi_undefined 2706 +; WAVE64-NEXT: .cfi_undefined 2707 +; WAVE64-NEXT: .cfi_undefined 2708 +; WAVE64-NEXT: .cfi_undefined 2709 +; WAVE64-NEXT: .cfi_undefined 2710 +; WAVE64-NEXT: .cfi_undefined 2711 + +; WAVE64-NEXT: .cfi_undefined 2720 +; WAVE64-NEXT: .cfi_undefined 2721 +; WAVE64-NEXT: .cfi_undefined 2722 +; WAVE64-NEXT: .cfi_undefined 2723 +; WAVE64-NEXT: .cfi_undefined 2724 +; WAVE64-NEXT: .cfi_undefined 2725 +; WAVE64-NEXT: .cfi_undefined 2726 +; WAVE64-NEXT: .cfi_undefined 2727 + +; WAVE64-NEXT: .cfi_undefined 2736 +; WAVE64-NEXT: .cfi_undefined 2737 +; WAVE64-NEXT: .cfi_undefined 2738 +; WAVE64-NEXT: .cfi_undefined 2739 +; WAVE64-NEXT: .cfi_undefined 2740 +; WAVE64-NEXT: .cfi_undefined 2741 +; WAVE64-NEXT: .cfi_undefined 2742 +; WAVE64-NEXT: .cfi_undefined 2743 + +; WAVE64-NEXT: .cfi_undefined 2752 +; WAVE64-NEXT: .cfi_undefined 2753 +; WAVE64-NEXT: .cfi_undefined 2754 +; WAVE64-NEXT: .cfi_undefined 2755 +; WAVE64-NEXT: .cfi_undefined 2756 +; WAVE64-NEXT: .cfi_undefined 2757 +; WAVE64-NEXT: .cfi_undefined 2758 +; WAVE64-NEXT: .cfi_undefined 2759 + +; WAVE64-NEXT: .cfi_undefined 2768 +; WAVE64-NEXT: .cfi_undefined 2769 +; WAVE64-NEXT: .cfi_undefined 2770 +; WAVE64-NEXT: .cfi_undefined 2771 +; WAVE64-NEXT: .cfi_undefined 2772 +; WAVE64-NEXT: .cfi_undefined 2773 +; WAVE64-NEXT: .cfi_undefined 2774 +; WAVE64-NEXT: .cfi_undefined 2775 + +; WAVE64-NEXT: .cfi_undefined 2784 +; WAVE64-NEXT: .cfi_undefined 2785 +; WAVE64-NEXT: .cfi_undefined 2786 +; WAVE64-NEXT: .cfi_undefined 2787 +; WAVE64-NEXT: .cfi_undefined 2788 +; WAVE64-NEXT: .cfi_undefined 2789 +; WAVE64-NEXT: .cfi_undefined 2790 +; WAVE64-NEXT: .cfi_undefined 2791 + +; WAVE64-NEXT: .cfi_undefined 2800 +; WAVE64-NEXT: .cfi_undefined 2801 +; WAVE64-NEXT: .cfi_undefined 2802 +; WAVE64-NEXT: .cfi_undefined 2803 +; WAVE64-NEXT: .cfi_undefined 2804 +; WAVE64-NEXT: .cfi_undefined 2805 +; WAVE64-NEXT: .cfi_undefined 2806 +; WAVE64-NEXT: .cfi_undefined 2807 + +; AGPR0_wave64 = 3072 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3072 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3073 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3074 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3075 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3076 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3077 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3078 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3079 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3080 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3081 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3082 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3083 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3084 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3085 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3086 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3087 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3088 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3089 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3090 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3091 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3092 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3093 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3094 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3095 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3096 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3097 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3098 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3099 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3100 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3101 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3102 +; GFX90A-V2A-DIS-NEXT: .cfi_undefined 3103 + +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3072 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3073 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3074 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3075 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3076 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3077 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3078 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3079 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3080 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3081 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3082 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3083 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3084 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3085 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3086 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3087 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3088 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3089 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3090 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3091 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3092 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3093 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3094 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3095 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3096 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3097 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3098 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3099 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3100 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3101 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3102 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3103 + +; VGPR0_wave32 = 1536 +; WAVE32-NEXT: .cfi_undefined 1536 +; WAVE32-NEXT: .cfi_undefined 1537 +; WAVE32-NEXT: .cfi_undefined 1538 +; WAVE32-NEXT: .cfi_undefined 1539 +; WAVE32-NEXT: .cfi_undefined 1540 +; WAVE32-NEXT: .cfi_undefined 1541 +; WAVE32-NEXT: .cfi_undefined 1542 +; WAVE32-NEXT: .cfi_undefined 1543 +; WAVE32-NEXT: .cfi_undefined 1544 +; WAVE32-NEXT: .cfi_undefined 1545 +; WAVE32-NEXT: .cfi_undefined 1546 +; WAVE32-NEXT: .cfi_undefined 1547 +; WAVE32-NEXT: .cfi_undefined 1548 +; WAVE32-NEXT: .cfi_undefined 1549 +; WAVE32-NEXT: .cfi_undefined 1550 +; WAVE32-NEXT: .cfi_undefined 1551 +; WAVE32-NEXT: .cfi_undefined 1552 +; WAVE32-NEXT: .cfi_undefined 1553 +; WAVE32-NEXT: .cfi_undefined 1554 +; WAVE32-NEXT: .cfi_undefined 1555 +; WAVE32-NEXT: .cfi_undefined 1556 +; WAVE32-NEXT: .cfi_undefined 1557 +; WAVE32-NEXT: .cfi_undefined 1558 +; WAVE32-NEXT: .cfi_undefined 1559 +; WAVE32-NEXT: .cfi_undefined 1560 +; WAVE32-NEXT: .cfi_undefined 1561 +; WAVE32-NEXT: .cfi_undefined 1562 +; WAVE32-NEXT: .cfi_undefined 1563 +; WAVE32-NEXT: .cfi_undefined 1564 +; WAVE32-NEXT: .cfi_undefined 1565 +; WAVE32-NEXT: .cfi_undefined 1566 +; WAVE32-NEXT: .cfi_undefined 1567 +; WAVE32-NEXT: .cfi_undefined 1568 +; WAVE32-NEXT: .cfi_undefined 1569 +; WAVE32-NEXT: .cfi_undefined 1570 +; WAVE32-NEXT: .cfi_undefined 1571 +; WAVE32-NEXT: .cfi_undefined 1572 +; WAVE32-NEXT: .cfi_undefined 1573 +; WAVE32-NEXT: .cfi_undefined 1574 +; WAVE32-NEXT: .cfi_undefined 1575 + +; VPGR48_wave64 = 1584 +; WAVE32-NEXT: .cfi_undefined 1584 +; WAVE32-NEXT: .cfi_undefined 1585 +; WAVE32-NEXT: .cfi_undefined 1586 +; WAVE32-NEXT: .cfi_undefined 1587 +; WAVE32-NEXT: .cfi_undefined 1588 +; WAVE32-NEXT: .cfi_undefined 1589 +; WAVE32-NEXT: .cfi_undefined 1590 +; WAVE32-NEXT: .cfi_undefined 1591 + +; WAVE32-NEXT: .cfi_undefined 1600 +; WAVE32-NEXT: .cfi_undefined 1601 +; WAVE32-NEXT: .cfi_undefined 1602 +; WAVE32-NEXT: .cfi_undefined 1603 +; WAVE32-NEXT: .cfi_undefined 1604 +; WAVE32-NEXT: .cfi_undefined 1605 +; WAVE32-NEXT: .cfi_undefined 1606 +; WAVE32-NEXT: .cfi_undefined 1607 + +; WAVE32-NEXT: .cfi_undefined 1616 +; WAVE32-NEXT: .cfi_undefined 1617 +; WAVE32-NEXT: .cfi_undefined 1618 +; WAVE32-NEXT: .cfi_undefined 1619 +; WAVE32-NEXT: .cfi_undefined 1620 +; WAVE32-NEXT: .cfi_undefined 1621 +; WAVE32-NEXT: .cfi_undefined 1622 +; WAVE32-NEXT: .cfi_undefined 1623 + +; WAVE32-NEXT: .cfi_undefined 1632 +; WAVE32-NEXT: .cfi_undefined 1633 +; WAVE32-NEXT: .cfi_undefined 1634 +; WAVE32-NEXT: .cfi_undefined 1635 +; WAVE32-NEXT: .cfi_undefined 1636 +; WAVE32-NEXT: .cfi_undefined 1637 +; WAVE32-NEXT: .cfi_undefined 1638 +; WAVE32-NEXT: .cfi_undefined 1639 + +; WAVE32-NEXT: .cfi_undefined 1648 +; WAVE32-NEXT: .cfi_undefined 1649 +; WAVE32-NEXT: .cfi_undefined 1650 +; WAVE32-NEXT: .cfi_undefined 1651 +; WAVE32-NEXT: .cfi_undefined 1652 +; WAVE32-NEXT: .cfi_undefined 1653 +; WAVE32-NEXT: .cfi_undefined 1654 +; WAVE32-NEXT: .cfi_undefined 1655 + +; WAVE32-NEXT: .cfi_undefined 1664 +; WAVE32-NEXT: .cfi_undefined 1665 +; WAVE32-NEXT: .cfi_undefined 1666 +; WAVE32-NEXT: .cfi_undefined 1667 +; WAVE32-NEXT: .cfi_undefined 1668 +; WAVE32-NEXT: .cfi_undefined 1669 +; WAVE32-NEXT: .cfi_undefined 1670 +; WAVE32-NEXT: .cfi_undefined 1671 + +; WAVE32-NEXT: .cfi_undefined 1680 +; WAVE32-NEXT: .cfi_undefined 1681 +; WAVE32-NEXT: .cfi_undefined 1682 +; WAVE32-NEXT: .cfi_undefined 1683 +; WAVE32-NEXT: .cfi_undefined 1684 +; WAVE32-NEXT: .cfi_undefined 1685 +; WAVE32-NEXT: .cfi_undefined 1686 +; WAVE32-NEXT: .cfi_undefined 1687 + +; WAVE32-NEXT: .cfi_undefined 1696 +; WAVE32-NEXT: .cfi_undefined 1697 +; WAVE32-NEXT: .cfi_undefined 1698 +; WAVE32-NEXT: .cfi_undefined 1699 +; WAVE32-NEXT: .cfi_undefined 1700 +; WAVE32-NEXT: .cfi_undefined 1701 +; WAVE32-NEXT: .cfi_undefined 1702 +; WAVE32-NEXT: .cfi_undefined 1703 + +; WAVE32-NEXT: .cfi_undefined 1712 +; WAVE32-NEXT: .cfi_undefined 1713 +; WAVE32-NEXT: .cfi_undefined 1714 +; WAVE32-NEXT: .cfi_undefined 1715 +; WAVE32-NEXT: .cfi_undefined 1716 +; WAVE32-NEXT: .cfi_undefined 1717 +; WAVE32-NEXT: .cfi_undefined 1718 +; WAVE32-NEXT: .cfi_undefined 1719 + +; WAVE32-NEXT: .cfi_undefined 1728 +; WAVE32-NEXT: .cfi_undefined 1729 +; WAVE32-NEXT: .cfi_undefined 1730 +; WAVE32-NEXT: .cfi_undefined 1731 +; WAVE32-NEXT: .cfi_undefined 1732 +; WAVE32-NEXT: .cfi_undefined 1733 +; WAVE32-NEXT: .cfi_undefined 1734 +; WAVE32-NEXT: .cfi_undefined 1735 + +; WAVE32-NEXT: .cfi_undefined 1744 +; WAVE32-NEXT: .cfi_undefined 1745 +; WAVE32-NEXT: .cfi_undefined 1746 +; WAVE32-NEXT: .cfi_undefined 1747 +; WAVE32-NEXT: .cfi_undefined 1748 +; WAVE32-NEXT: .cfi_undefined 1749 +; WAVE32-NEXT: .cfi_undefined 1750 +; WAVE32-NEXT: .cfi_undefined 1751 + +; WAVE32-NEXT: .cfi_undefined 1760 +; WAVE32-NEXT: .cfi_undefined 1761 +; WAVE32-NEXT: .cfi_undefined 1762 +; WAVE32-NEXT: .cfi_undefined 1763 +; WAVE32-NEXT: .cfi_undefined 1764 +; WAVE32-NEXT: .cfi_undefined 1765 +; WAVE32-NEXT: .cfi_undefined 1766 +; WAVE32-NEXT: .cfi_undefined 1767 + +; WAVE32-NEXT: .cfi_undefined 1776 +; WAVE32-NEXT: .cfi_undefined 1777 +; WAVE32-NEXT: .cfi_undefined 1778 +; WAVE32-NEXT: .cfi_undefined 1779 +; WAVE32-NEXT: .cfi_undefined 1780 +; WAVE32-NEXT: .cfi_undefined 1781 +; WAVE32-NEXT: .cfi_undefined 1782 +; WAVE32-NEXT: .cfi_undefined 1783 + + +; SGPR0 = 32 +; CHECK-NEXT: .cfi_undefined 32 +; CHECK-NEXT: .cfi_undefined 33 +; CHECK-NEXT: .cfi_undefined 34 +; CHECK-NEXT: .cfi_undefined 35 +; CHECK-NEXT: .cfi_undefined 36 +; CHECK-NEXT: .cfi_undefined 37 +; CHECK-NEXT: .cfi_undefined 38 +; CHECK-NEXT: .cfi_undefined 39 +; CHECK-NEXT: .cfi_undefined 40 +; CHECK-NEXT: .cfi_undefined 41 +; CHECK-NEXT: .cfi_undefined 42 +; CHECK-NEXT: .cfi_undefined 43 +; CHECK-NEXT: .cfi_undefined 44 +; CHECK-NEXT: .cfi_undefined 45 +; CHECK-NEXT: .cfi_undefined 46 +; CHECK-NEXT: .cfi_undefined 47 +; CHECK-NEXT: .cfi_undefined 48 +; CHECK-NEXT: .cfi_undefined 49 +; CHECK-NEXT: .cfi_undefined 50 +; CHECK-NEXT: .cfi_undefined 51 +; CHECK-NEXT: .cfi_undefined 52 +; CHECK-NEXT: .cfi_undefined 53 +; CHECK-NEXT: .cfi_undefined 54 +; CHECK-NEXT: .cfi_undefined 55 +; CHECK-NEXT: .cfi_undefined 56 +; CHECK-NEXT: .cfi_undefined 57 +; CHECK-NEXT: .cfi_undefined 58 +; CHECK-NEXT: .cfi_undefined 59 +; CHECK-NEXT: .cfi_undefined 60 +; CHECK-NEXT: .cfi_undefined 61 + +; CHECK-NOT: .cfi_{{.*}} + +; WAVE64: s_or_saveexec_b64 [[EXEC_MASK:s\[[0-9]+:[0-9]+\]]], -1 +; WAVE32: s_or_saveexec_b32 [[EXEC_MASK:s[0-9]+]], -1 +; CHECK-NEXT: buffer_store_dword v40, off, s[0:3], s32 ; 4-byte Folded Spill +; VGPR40_wave64 = 2600 +; WAVE64-NEXT: .cfi_offset 2600, 0 +; VGPR40_wave32 = 1576 +; WAVE32-NEXT: .cfi_offset 1576, 0 +; CHECK-NOT: .cfi_{{.*}} +; WAVE64: s_mov_b64 exec, [[EXEC_MASK]] +; WAVE32: s_mov_b32 exec_lo, [[EXEC_MASK]] + +; CHECK-NOT: .cfi_{{.*}} + +; CHECK: v_writelane_b32 v40, s33, 2 + +; DW_CFA_expression [0x10] SGPR33 ULEB128(65)=[0x41] +; BLOCK_LENGTH ULEB128(5)=[0x05] +; DW_OP_regx [0x90] +; VGPR40_wave64 ULEB128(2600)=[0xa8, 0x14] +; DW_OP_LLVM_offset_uconst [0xec] +; OFFSET ULEB128(0x08) [0x08] +; WAVE64-NEXT: .cfi_escape 0x10, 0x41, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x08 + +; DW_CFA_expression [0x10] SGPR33 ULEB128(65)=[0x41] +; BLOCK_LENGTH ULEB128(5)=[0x05] +; DW_OP_regx [0x90] +; VGPR40_wave32 ULEB128(1576)=[0xa8, 0x0c] +; DW_OP_LLVM_offset_uconst [0xe4] +; OFFSET ULEB128(0x08) [0x08] +; WAVE32-NEXT: .cfi_escape 0x10, 0x41, 0x05, 0x90, 0xa8, 0x0c, 0xec, 0x08 + +; CHECK-NOT: .cfi_{{.*}} + +; CHECK: s_mov_b32 s33, s32 +; SGPR33 = 65 +; CHECK-NEXT: .cfi_def_cfa_register 65 + +; CHECK-NOT: .cfi_{{.*}} + +; CHECK: s_addk_i32 s32, +; CHECK: v_readlane_b32 s33, v40, 2 +; SGPR32 = 64 +; CHECK: .cfi_def_cfa_register 64 + +; CHECK-NOT: .cfi_{{.*}} + +; CHECK: .cfi_endproc +define hidden void @func_call_clobber() #0 { +entry: + call void @ex() #0 + ret void +} + +; CHECK-LABEL: func_spill_vgpr_to_vmem: +; CHECK: .cfi_startproc + +; CHECK-NOT: .cfi_{{.*}} + +; CHECK: %bb.0: +; SGPR32 = 64 +; CHECK-NEXT: .cfi_llvm_def_aspace_cfa 64, 0, 6 +; CHECK-NEXT: .cfi_escape 0x10, 0x10, 0x08, 0x90, 0x3e, 0x93, 0x04, 0x90, 0x3f, 0x93, 0x04 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 2560 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 2561 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3072 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3073 + +; CHECK-NOT: .cfi_{{.*}} + +; WAVE32: buffer_store_dword v40, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill +; GFX900: buffer_store_dword v40, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill +; GFX90A-V2A-DIS: buffer_store_dword v40, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill +; GFX90A-V2A-EN: v_accvgpr_write_b32 a[[#TMP_AGPR1:]], v[[#VGPR1:]] + +; DW_CFA_expression [0x10] +; VGPR40_wave64 ULEB128(1576)=[0xa8, 0x14] +; BLOCK_LENGTH ULEB128(14)=[0x0e] +; DW_OP_regx [0x90] +; VGPR40_wave64 ULEB128(1576)=[0xa8, 0x14] +; DW_OP_swap [0x16] +; DW_OP_LLVM_offset_uconst [0xec] +; OFFSET ULEB128(256)=[0x80, 0x02] / OFFSET ULEB128(256)=[0x80, 0x06] +; DW_OP_LLVM_call_frame_entry_reg [0xf5] +; EXEC_MASK_wave64 ULEB128(17)=[0x11] +; DW_OP_deref_size [0x94] +; SIZE [0x08] +; DW_OP_LLVM_select_bit_piece [0xfe] +; ELEMENT_SIZE [0x20] +; ELEMENT_COUNT [0x40] +; GFX900-NEXT: .cfi_escape 0x10, 0xa8, 0x14, 0x0e, 0x90, 0xa8, 0x14, 0x16, 0xec, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 +; GFX90A-V2A-DIS-NEXT: .cfi_escape 0x10, 0xa8, 0x14, 0x0e, 0x90, 0xa8, 0x14, 0x16, 0xec, 0x80, 0x06, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 +; GFX90A-V2A-EN-NEXT: .cfi_register [[#VGPR1+2560]], [[#TMP_AGPR1+3072]] + +; DW_CFA_expression [0x10] +; VGPR40_wave32 ULEB128(1576)=[0xa8, 0x0c] +; BLOCK_LENGTH ULEB128(14)=[0x0e] +; DW_OP_regx [0x90] +; VGPR40_wave32 ULEB128(1576)=[0xa8, 0x0c] +; DW_OP_swap [0x16] +; DW_OP_LLVM_offset_uconst [0xec] +; OFFSET ULEB128(128)=[0x80, 0x01] +; DW_OP_LLVM_call_frame_entry_reg [0xf5] +; EXEC_MASK_wave32 ULEB128(1)=[0x01] +; DW_OP_deref_size [0x94] +; SIZE [0x04] +; DW_OP_LLVM_select_bit_piece [0xfe] +; ELEMENT_SIZE [0x20] +; ELEMENT_COUNT [0x20] +; WAVE32-NEXT: .cfi_escape 0x10, 0xa8, 0x0c, 0x0e, 0x90, 0xa8, 0x0c, 0x16, 0xec, 0x80, 0x01, 0xf5, 0x01, 0x94, 0x04, 0xfe, 0x20, 0x20 + +; CHECK-NOT: .cfi_{{.*}} + +; WAVE32: buffer_store_dword v41, off, s[0:3], s32 ; 4-byte Folded Spill +; GFX900: buffer_store_dword v41, off, s[0:3], s32 ; 4-byte Folded Spill +; GFX90A-V2A-DIS: buffer_store_dword v41, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill +; GFX90A-V2A-EN: v_accvgpr_write_b32 a[[#TMP_AGPR2:]], v[[#VGPR2:]] + +; DW_CFA_expression [0x10] +; VGPR41_wave64 ULEB128(2601)=[0xa9, 0x14] +; BLOCK_LENGTH ULEB128(13)=[0x0d] +; DW_OP_regx [0x90] +; VGPR41_wave64 ULEB128(2601)=[0xa9, 0x14] +; DW_OP_swap [0x16] +; DW_OP_LLVM_offset_uconst [0xec] +; OFFSET ULEB128(0)=[0x00] / OFFSET ULEB128(128)=[0x80, 0x04] +; DW_OP_LLVM_call_frame_entry_reg [0xe6] +; EXEC_MASK_wave64 ULEB128(17)=[0x11] +; DW_OP_deref_size [0x94] +; SIZE [0x08] +; DW_OP_LLVM_select_bit_piece [0xfe] +; ELEMENT_SIZE [0x20] +; ELEMENT_COUNT [0x40] +; GFX900-NEXT: .cfi_escape 0x10, 0xa9, 0x14, 0x0d, 0x90, 0xa9, 0x14, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 +; GFX90A-V2A-DIS-NEXT: .cfi_escape 0x10, 0xa9, 0x14, 0x0e, 0x90, 0xa9, 0x14, 0x16, 0xec, 0x80, 0x04, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 +; GFX90A-V2A-EN-NEXT: .cfi_register [[#VGPR2+2560]], [[#TMP_AGPR2+3072]] + +; DW_CFA_expression [0x10] +; VGPR41_wave32 ULEB128(1577)=[0xa9, 0x0c] +; BLOCK_LENGTH ULEB128(13)=[0x0d] +; DW_OP_regx [0x90] +; VGPR41_wave32 ULEB128(1577)=[0xa9, 0x0c] +; DW_OP_swap [0x16] +; DW_OP_LLVM_offset_uconst [0xec] +; OFFSET ULEB128(0)=[0x00] +; DW_OP_LLVM_call_frame_entry_reg [0xf5] +; EXEC_MASK_wave32 ULEB128(1)=[0x01] +; DW_OP_deref_size [0x94] +; SIZE [0x04] +; DW_OP_LLVM_select_bit_piece [0xfe] +; ELEMENT_SIZE [0x20] +; ELEMENT_COUNT [0x20] +; WAVE32: .cfi_escape 0x10, 0xa9, 0x0c, 0x0d, 0x90, 0xa9, 0x0c, 0x16, 0xec, 0x00, 0xf5, 0x01, 0x94, 0x04, 0xfe, 0x20, 0x20 + +; GFX90A-V2A-DIS: buffer_store_dword a32, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill +; GFX90A-V2A-EN: v_accvgpr_read_b32 v[[#TMP_VGPR1:]], a[[#AGPR1:]] + +; DW_CFA_expression [0x10] +; AGPR32_wave64 ULEB128(3104)=[0xa0, 0x18] +; BLOCK_LENGTH ULEB128(14)=[0x0e] +; DW_OP_regx [0x90] +; AGPR32_wave64 ULEB128(3104)=[0xa0, 0x18] +; DW_OP_swap [0x16] +; DW_OP_LLVM_offset_uconst [0xec] +; OFFSET ULEB128(256)=[0x80, 0x02] +; DW_OP_LLVM_call_frame_entry_reg [0xe6] +; EXEC_MASK_wave64 ULEB128(17)=[0x11] +; DW_OP_deref_size [0x94] +; SIZE [0x08] +; DW_OP_LLVM_select_bit_piece [0xfe] +; ELEMENT_SIZE [0x20] +; ELEMENT_COUNT [0x40] +; GFX90A-V2A-DIS-NEXT: .cfi_escape 0x10, 0xa0, 0x18, 0x0e, 0x90, 0xa0, 0x18, 0x16, 0xec, 0x80, 0x02, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 +; GFX90A-V2A-EN-NEXT: .cfi_register [[#AGPR1+3072]], [[#TMP_VGPR1+2560]] + +; CHECK-NOT: .cfi_{{.*}} + +; GFX90A-V2A-DIS: buffer_store_dword a33, off, s[0:3], s32 ; 4-byte Folded Spill +; GFX90A-V2A-EN: v_accvgpr_read_b32 v[[#TMP_VGPR2:]], a[[#AGPR2:]] + +; DW_CFA_expression [0x10] +; AGPR32_wave64 ULEB128(3105)=[0xa1, 0x18] +; BLOCK_LENGTH ULEB128(14)=[0x0d] +; DW_OP_regx [0x90] +; AGPR32_wave64 ULEB128(3105)=[0xa1, 0x18] +; DW_OP_swap [0x16] +; DW_OP_LLVM_offset_uconst [0xec] +; OFFSET ULEB128(0)=[0x00] +; DW_OP_LLVM_call_frame_entry_reg [0xe6] +; EXEC_MASK_wave64 ULEB128(17)=[0x11] +; DW_OP_deref_size [0x94] +; SIZE [0x08] +; DW_OP_LLVM_select_bit_piece [0xfe] +; ELEMENT_SIZE [0x20] +; ELEMENT_COUNT [0x40] +; GFX90A-V2A-DIS-NEXT: .cfi_escape 0x10, 0xa1, 0x18, 0x0d, 0x90, 0xa1, 0x18, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 +; GFX90A-V2A-EN-NEXT: .cfi_register [[#AGPR2+3072]], [[#TMP_VGPR2+2560]] + +; CHECK-NOT: .cfi_{{.*}} + +; CHECK: .cfi_endproc +define hidden void @func_spill_vgpr_to_vmem() #0 { +entry: + call void asm sideeffect "; clobber", "~{v40}"() #0 + call void asm sideeffect "; clobber", "~{v41}"() #0 + call void asm sideeffect "; clobber", "~{a32}"() #0 + call void asm sideeffect "; clobber", "~{a33}"() #0 + ret void +} + +; CHECK-LABEL: func_spill_vgpr_to_agpr: +; CHECK: .cfi_startproc + +; CHECK-NOT: .cfi_{{.*}} + +; CHECK: %bb.0: +; CHECK-NEXT: .cfi_llvm_def_aspace_cfa 64, 0, 6 +; CHECK-NEXT: .cfi_escape 0x10, 0x10, 0x08, 0x90, 0x3e, 0x93, 0x04, 0x90, 0x3f, 0x93, 0x04 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 2560 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 2561 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3072 +; GFX90A-V2A-EN-NEXT: .cfi_undefined 3073 + +; CHECK-NOT: .cfi_{{.*}} + +; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX90A-V2A-EN-NEXT: v_accvgpr_write_b32 a[[#TMP_AGPR1:]], v[[#VGPR1:]] +; GFX90A-V2A-EN-NEXT: .cfi_register [[#VGPR1+2560]], [[#TMP_AGPR1+3072]] +; GFX90A-V2A-EN-NEXT: v_accvgpr_write_b32 a[[#TMP_AGPR2]], v[[#VGPR2]] +; GFX90A-V2A-EN-NEXT: .cfi_register [[#VGPR2+2560]], [[#TMP_AGPR2+3072]] +; GFX90A-V2A-EN-NEXT: v_accvgpr_read_b32 v[[#TMP_VGPR1:]], a[[#AGPR1:]] +; GFX90A-V2A-EN-NEXT: .cfi_register [[#AGPR1+3072]], [[#TMP_VGPR1+2560]] +; GFX90A-V2A-EN-NEXT: v_accvgpr_read_b32 v[[#TMP_VGPR2:]], a[[#AGPR2:]] +; GFX90A-V2A-EN-NEXT: .cfi_register [[#AGPR2+3072]], [[#TMP_VGPR2+2560]] +; GFX90A-V2A-EN: v_accvgpr_write_b32 a33, v1 +; GFX90A-V2A-EN-NEXT: v_accvgpr_write_b32 a32, v0 +; GFX90A-V2A-EN-NEXT: v_accvgpr_read_b32 v41, a1 +; GFX90A-V2A-EN-NEXT: v_accvgpr_read_b32 v40, a0 + +; CHECK: s_setpc_b64 s[30:31] + +; CHECK-NOT: .cfi_{{.*}} +; CHECK: .cfi_endproc + +define hidden void @func_spill_vgpr_to_agpr() #2 { + call void asm sideeffect "; clobber", "~{v40}"() + call void asm sideeffect "; clobber", "~{v41}"() + call void asm sideeffect "; clobber", "~{a32}"() + call void asm sideeffect "; clobber", "~{a33}"() + ret void +} + + +; NOTE: Number of VGPRs available to kernel, and in turn number of corresponding CFIs generated, +; is dependent on waves/WG size. Since the intent here is to check whether we generate the correct +; CFIs, doing it for any one set of details is sufficient which also makes the test insensitive to +; changes in those details. +attributes #0 = { nounwind "amdgpu-waves-per-eu"="1,1" "amdgpu-flat-work-group-size"="128,128" } +attributes #1 = { nounwind "amdgpu-waves-per-eu"="1,1" "amdgpu-flat-work-group-size"="128,128" "frame-pointer"="all" } +attributes #2 = { nounwind } !llvm.dbg.cu = !{!0} !llvm.module.flags = !{!2, !3} diff --git a/llvm/test/CodeGen/AMDGPU/dwarf-multi-register-use-crash.ll b/llvm/test/CodeGen/AMDGPU/dwarf-multi-register-use-crash.ll --- a/llvm/test/CodeGen/AMDGPU/dwarf-multi-register-use-crash.ll +++ b/llvm/test/CodeGen/AMDGPU/dwarf-multi-register-use-crash.ll @@ -456,22 +456,40 @@ ; CHECK-NEXT: .cfi_escape 0x10, 0x41, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x40 ; ; CHECK-NEXT: s_mov_b32 s33, s32 ; CHECK-NEXT: .cfi_def_cfa_register 65 -; CHECK-NEXT: v_writelane_b32 v40, s30, 0 -; CHECK-NEXT: v_writelane_b32 v40, s31, 1 -; CHECK-NEXT: v_writelane_b32 v40, s34, 2 -; CHECK-NEXT: v_writelane_b32 v40, s35, 3 -; CHECK-NEXT: v_writelane_b32 v40, s36, 4 -; CHECK-NEXT: v_writelane_b32 v40, s37, 5 -; CHECK-NEXT: v_writelane_b32 v40, s38, 6 -; CHECK-NEXT: v_writelane_b32 v40, s39, 7 -; CHECK-NEXT: v_writelane_b32 v40, s40, 8 -; CHECK-NEXT: v_writelane_b32 v40, s41, 9 -; CHECK-NEXT: v_writelane_b32 v40, s42, 10 -; CHECK-NEXT: v_writelane_b32 v40, s43, 11 -; CHECK-NEXT: v_writelane_b32 v40, s44, 12 ; CHECK-NEXT: s_addk_i32 s32, 0x400 -; CHECK-NEXT: v_writelane_b32 v40, s45, 13 -; CHECK-NEXT: v_writelane_b32 v40, s46, 14 +; CHECK-NEXT: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill +; CHECK-NEXT: .cfi_escape 0x10, 0xa9, 0x14, 0x0d, 0x90, 0xa9, 0x14, 0x16, 0xec, 0x00, 0xf5, 0x11, 0x94, 0x08, 0xfe, 0x20, 0x40 ; +; CHECK-NEXT: v_writelane_b32 v40, s34, 0 +; CHECK-NEXT: .cfi_escape 0x10, 0x42, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x00 ; +; CHECK-NEXT: v_writelane_b32 v40, s35, 1 +; CHECK-NEXT: .cfi_escape 0x10, 0x43, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x04 ; +; CHECK-NEXT: v_writelane_b32 v40, s36, 2 +; CHECK-NEXT: .cfi_escape 0x10, 0x44, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x08 ; +; CHECK-NEXT: v_writelane_b32 v40, s37, 3 +; CHECK-NEXT: .cfi_escape 0x10, 0x45, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x0c ; +; CHECK-NEXT: v_writelane_b32 v40, s38, 4 +; CHECK-NEXT: .cfi_escape 0x10, 0x46, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x10 ; +; CHECK-NEXT: v_writelane_b32 v40, s39, 5 +; CHECK-NEXT: .cfi_escape 0x10, 0x47, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x14 ; +; CHECK-NEXT: v_writelane_b32 v40, s40, 6 +; CHECK-NEXT: .cfi_escape 0x10, 0x48, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x18 ; +; CHECK-NEXT: v_writelane_b32 v40, s41, 7 +; CHECK-NEXT: .cfi_escape 0x10, 0x49, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x1c ; +; CHECK-NEXT: v_writelane_b32 v40, s42, 8 +; CHECK-NEXT: .cfi_escape 0x10, 0x4a, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x20 ; +; CHECK-NEXT: v_writelane_b32 v40, s43, 9 +; CHECK-NEXT: .cfi_escape 0x10, 0x4b, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x24 ; +; CHECK-NEXT: v_writelane_b32 v40, s44, 10 +; CHECK-NEXT: .cfi_escape 0x10, 0x4c, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x28 ; +; CHECK-NEXT: v_writelane_b32 v40, s45, 11 +; CHECK-NEXT: .cfi_escape 0x10, 0x4d, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x2c ; +; CHECK-NEXT: v_writelane_b32 v40, s46, 12 +; CHECK-NEXT: .cfi_escape 0x10, 0x4e, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x30 ; +; CHECK-NEXT: v_writelane_b32 v40, s47, 13 +; CHECK-NEXT: .cfi_escape 0x10, 0x4f, 0x05, 0x90, 0xa8, 0x14, 0xec, 0x34 ; +; CHECK-NEXT: v_writelane_b32 v40, s30, 14 +; CHECK-NEXT: v_writelane_b32 v40, s31, 15 +; CHECK-NEXT: .cfi_escape 0x10, 0x10, 0x0e, 0x90, 0xa8, 0x14, 0x9d, 0x20, 0xc0, 0x03, 0x90, 0xa8, 0x14, 0x9d, 0x20, 0xe0, 0x03 ; ; CHECK-NEXT: s_mov_b64 s[40:41], s[4:5] ; CHECK-NEXT: ;DEBUG_VALUE: dummy:dummy <- undef ; CHECK-NEXT: .Ltmp0: @@ -479,10 +497,8 @@ ; CHECK-NEXT: s_getpc_b64 s[4:5] ; CHECK-NEXT: s_add_u32 s4, s4, __kmpc_alloc_shared@gotpcrel32@lo+4 ; CHECK-NEXT: s_addc_u32 s5, s5, __kmpc_alloc_shared@gotpcrel32@hi+12 -; CHECK-NEXT: v_writelane_b32 v40, s47, 15 ; CHECK-NEXT: s_load_dwordx2 s[46:47], s[4:5], 0x0 ; CHECK-NEXT: s_mov_b64 s[4:5], s[40:41] -; CHECK-NEXT: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill ; CHECK-NEXT: v_mov_b32_e32 v41, v31 ; CHECK-NEXT: s_mov_b32 s42, s15 ; CHECK-NEXT: s_mov_b32 s43, s14 @@ -508,23 +524,23 @@ ; CHECK-NEXT: .loc 1 0 9 is_stmt 0 ; dummy:0:9 ; CHECK-NEXT: buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload ; CHECK-NEXT: v_mov_b32_e32 v2, 0 +; CHECK-NEXT: v_readlane_b32 s30, v40, 14 ; CHECK-NEXT: flat_store_dword v[0:1], v2 -; CHECK-NEXT: v_readlane_b32 s47, v40, 15 -; CHECK-NEXT: v_readlane_b32 s46, v40, 14 -; CHECK-NEXT: v_readlane_b32 s45, v40, 13 -; CHECK-NEXT: v_readlane_b32 s44, v40, 12 -; CHECK-NEXT: v_readlane_b32 s43, v40, 11 -; CHECK-NEXT: v_readlane_b32 s42, v40, 10 -; CHECK-NEXT: v_readlane_b32 s41, v40, 9 -; CHECK-NEXT: v_readlane_b32 s40, v40, 8 -; CHECK-NEXT: v_readlane_b32 s39, v40, 7 -; CHECK-NEXT: v_readlane_b32 s38, v40, 6 -; CHECK-NEXT: v_readlane_b32 s37, v40, 5 -; CHECK-NEXT: v_readlane_b32 s36, v40, 4 -; CHECK-NEXT: v_readlane_b32 s35, v40, 3 -; CHECK-NEXT: v_readlane_b32 s34, v40, 2 -; CHECK-NEXT: v_readlane_b32 s31, v40, 1 -; CHECK-NEXT: v_readlane_b32 s30, v40, 0 +; CHECK-NEXT: v_readlane_b32 s31, v40, 15 +; CHECK-NEXT: v_readlane_b32 s47, v40, 13 +; CHECK-NEXT: v_readlane_b32 s46, v40, 12 +; CHECK-NEXT: v_readlane_b32 s45, v40, 11 +; CHECK-NEXT: v_readlane_b32 s44, v40, 10 +; CHECK-NEXT: v_readlane_b32 s43, v40, 9 +; CHECK-NEXT: v_readlane_b32 s42, v40, 8 +; CHECK-NEXT: v_readlane_b32 s41, v40, 7 +; CHECK-NEXT: v_readlane_b32 s40, v40, 6 +; CHECK-NEXT: v_readlane_b32 s39, v40, 5 +; CHECK-NEXT: v_readlane_b32 s38, v40, 4 +; CHECK-NEXT: v_readlane_b32 s37, v40, 3 +; CHECK-NEXT: v_readlane_b32 s36, v40, 2 +; CHECK-NEXT: v_readlane_b32 s35, v40, 1 +; CHECK-NEXT: v_readlane_b32 s34, v40, 0 ; CHECK-NEXT: s_addk_i32 s32, 0xfc00 ; CHECK-NEXT: v_readlane_b32 s33, v40, 16 ; CHECK-NEXT: .cfi_def_cfa_register 64 diff --git a/llvm/test/CodeGen/AMDGPU/frame-setup-without-sgpr-to-vgpr-spills.ll b/llvm/test/CodeGen/AMDGPU/frame-setup-without-sgpr-to-vgpr-spills.ll --- a/llvm/test/CodeGen/AMDGPU/frame-setup-without-sgpr-to-vgpr-spills.ll +++ b/llvm/test/CodeGen/AMDGPU/frame-setup-without-sgpr-to-vgpr-spills.ll @@ -15,18 +15,18 @@ ; SPILL-TO-VGPR-NEXT: s_mov_b64 exec, s[4:5] ; SPILL-TO-VGPR-NEXT: v_writelane_b32 v40, s33, 2 ; SPILL-TO-VGPR-NEXT: s_mov_b32 s33, s32 -; SPILL-TO-VGPR-NEXT: s_addk_i32 s32, 0x400 ; SPILL-TO-VGPR-NEXT: v_writelane_b32 v40, s30, 0 -; SPILL-TO-VGPR-NEXT: v_mov_b32_e32 v0, 0 +; SPILL-TO-VGPR-NEXT: s_addk_i32 s32, 0x400 ; SPILL-TO-VGPR-NEXT: v_writelane_b32 v40, s31, 1 +; SPILL-TO-VGPR-NEXT: v_mov_b32_e32 v0, 0 ; SPILL-TO-VGPR-NEXT: buffer_store_dword v0, off, s[0:3], s33 ; SPILL-TO-VGPR-NEXT: s_waitcnt vmcnt(0) ; SPILL-TO-VGPR-NEXT: s_getpc_b64 s[4:5] ; SPILL-TO-VGPR-NEXT: s_add_u32 s4, s4, external_void_func_void@rel32@lo+4 ; SPILL-TO-VGPR-NEXT: s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12 ; SPILL-TO-VGPR-NEXT: s_swappc_b64 s[30:31], s[4:5] -; SPILL-TO-VGPR-NEXT: v_readlane_b32 s31, v40, 1 ; SPILL-TO-VGPR-NEXT: v_readlane_b32 s30, v40, 0 +; SPILL-TO-VGPR-NEXT: v_readlane_b32 s31, v40, 1 ; SPILL-TO-VGPR-NEXT: s_addk_i32 s32, 0xfc00 ; SPILL-TO-VGPR-NEXT: v_readlane_b32 s33, v40, 2 ; SPILL-TO-VGPR-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -43,21 +43,14 @@ ; NO-SPILL-TO-VGPR-NEXT: s_mov_b32 s33, s32 ; NO-SPILL-TO-VGPR-NEXT: s_addk_i32 s32, 0x800 ; NO-SPILL-TO-VGPR-NEXT: s_mov_b64 s[8:9], exec -; NO-SPILL-TO-VGPR-NEXT: s_mov_b64 exec, 1 +; NO-SPILL-TO-VGPR-NEXT: s_mov_b64 exec, 3 ; NO-SPILL-TO-VGPR-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:16 ; NO-SPILL-TO-VGPR-NEXT: v_writelane_b32 v1, s30, 0 +; NO-SPILL-TO-VGPR-NEXT: v_writelane_b32 v1, s31, 1 ; NO-SPILL-TO-VGPR-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill ; NO-SPILL-TO-VGPR-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:16 ; NO-SPILL-TO-VGPR-NEXT: s_waitcnt vmcnt(0) ; NO-SPILL-TO-VGPR-NEXT: s_mov_b64 exec, s[8:9] -; NO-SPILL-TO-VGPR-NEXT: s_mov_b64 s[8:9], exec -; NO-SPILL-TO-VGPR-NEXT: s_mov_b64 exec, 1 -; NO-SPILL-TO-VGPR-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:16 -; NO-SPILL-TO-VGPR-NEXT: v_writelane_b32 v1, s31, 0 -; NO-SPILL-TO-VGPR-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill -; NO-SPILL-TO-VGPR-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:16 -; NO-SPILL-TO-VGPR-NEXT: s_waitcnt vmcnt(0) -; NO-SPILL-TO-VGPR-NEXT: s_mov_b64 exec, s[8:9] ; NO-SPILL-TO-VGPR-NEXT: v_mov_b32_e32 v0, 0 ; NO-SPILL-TO-VGPR-NEXT: buffer_store_dword v0, off, s[0:3], s33 ; NO-SPILL-TO-VGPR-NEXT: s_waitcnt vmcnt(0) @@ -66,20 +59,12 @@ ; NO-SPILL-TO-VGPR-NEXT: s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12 ; NO-SPILL-TO-VGPR-NEXT: s_swappc_b64 s[30:31], s[4:5] ; NO-SPILL-TO-VGPR-NEXT: s_mov_b64 s[4:5], exec -; NO-SPILL-TO-VGPR-NEXT: s_mov_b64 exec, 1 -; NO-SPILL-TO-VGPR-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:16 -; NO-SPILL-TO-VGPR-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload -; NO-SPILL-TO-VGPR-NEXT: s_waitcnt vmcnt(0) -; NO-SPILL-TO-VGPR-NEXT: v_readlane_b32 s31, v1, 0 -; NO-SPILL-TO-VGPR-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:16 -; NO-SPILL-TO-VGPR-NEXT: s_waitcnt vmcnt(0) -; NO-SPILL-TO-VGPR-NEXT: s_mov_b64 exec, s[4:5] -; NO-SPILL-TO-VGPR-NEXT: s_mov_b64 s[4:5], exec -; NO-SPILL-TO-VGPR-NEXT: s_mov_b64 exec, 1 +; NO-SPILL-TO-VGPR-NEXT: s_mov_b64 exec, 3 ; NO-SPILL-TO-VGPR-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:16 ; NO-SPILL-TO-VGPR-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload ; NO-SPILL-TO-VGPR-NEXT: s_waitcnt vmcnt(0) ; NO-SPILL-TO-VGPR-NEXT: v_readlane_b32 s30, v1, 0 +; NO-SPILL-TO-VGPR-NEXT: v_readlane_b32 s31, v1, 1 ; NO-SPILL-TO-VGPR-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:16 ; NO-SPILL-TO-VGPR-NEXT: s_waitcnt vmcnt(0) ; NO-SPILL-TO-VGPR-NEXT: s_mov_b64 exec, s[4:5] diff --git a/llvm/test/CodeGen/AMDGPU/gfx-call-non-gfx-func.ll b/llvm/test/CodeGen/AMDGPU/gfx-call-non-gfx-func.ll --- a/llvm/test/CodeGen/AMDGPU/gfx-call-non-gfx-func.ll +++ b/llvm/test/CodeGen/AMDGPU/gfx-call-non-gfx-func.ll @@ -13,6 +13,7 @@ ; SDAG-NEXT: s_mov_b64 exec, s[34:35] ; SDAG-NEXT: v_writelane_b32 v40, s33, 28 ; SDAG-NEXT: s_mov_b32 s33, s32 +; SDAG-NEXT: s_addk_i32 s32, 0x400 ; SDAG-NEXT: v_writelane_b32 v40, s4, 0 ; SDAG-NEXT: v_writelane_b32 v40, s5, 1 ; SDAG-NEXT: v_writelane_b32 v40, s6, 2 @@ -33,24 +34,23 @@ ; SDAG-NEXT: v_writelane_b32 v40, s21, 17 ; SDAG-NEXT: v_writelane_b32 v40, s22, 18 ; SDAG-NEXT: v_writelane_b32 v40, s23, 19 -; SDAG-NEXT: s_addk_i32 s32, 0x400 ; SDAG-NEXT: v_writelane_b32 v40, s24, 20 ; SDAG-NEXT: v_writelane_b32 v40, s25, 21 -; SDAG-NEXT: s_getpc_b64 s[34:35] -; SDAG-NEXT: s_add_u32 s34, s34, extern_c_func@gotpcrel32@lo+4 -; SDAG-NEXT: s_addc_u32 s35, s35, extern_c_func@gotpcrel32@hi+12 ; SDAG-NEXT: v_writelane_b32 v40, s26, 22 -; SDAG-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; SDAG-NEXT: v_writelane_b32 v40, s27, 23 ; SDAG-NEXT: v_writelane_b32 v40, s28, 24 ; SDAG-NEXT: v_writelane_b32 v40, s29, 25 ; SDAG-NEXT: v_writelane_b32 v40, s30, 26 -; SDAG-NEXT: s_mov_b64 s[8:9], 0 ; SDAG-NEXT: v_writelane_b32 v40, s31, 27 +; SDAG-NEXT: s_getpc_b64 s[34:35] +; SDAG-NEXT: s_add_u32 s34, s34, extern_c_func@gotpcrel32@lo+4 +; SDAG-NEXT: s_addc_u32 s35, s35, extern_c_func@gotpcrel32@hi+12 +; SDAG-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; SDAG-NEXT: s_mov_b64 s[8:9], 0 ; SDAG-NEXT: s_waitcnt lgkmcnt(0) ; SDAG-NEXT: s_swappc_b64 s[30:31], s[34:35] -; SDAG-NEXT: v_readlane_b32 s31, v40, 27 ; SDAG-NEXT: v_readlane_b32 s30, v40, 26 +; SDAG-NEXT: v_readlane_b32 s31, v40, 27 ; SDAG-NEXT: v_readlane_b32 s29, v40, 25 ; SDAG-NEXT: v_readlane_b32 s28, v40, 24 ; SDAG-NEXT: v_readlane_b32 s27, v40, 23 @@ -93,6 +93,7 @@ ; GISEL-NEXT: s_mov_b64 exec, s[34:35] ; GISEL-NEXT: v_writelane_b32 v40, s33, 28 ; GISEL-NEXT: s_mov_b32 s33, s32 +; GISEL-NEXT: s_addk_i32 s32, 0x400 ; GISEL-NEXT: v_writelane_b32 v40, s4, 0 ; GISEL-NEXT: v_writelane_b32 v40, s5, 1 ; GISEL-NEXT: v_writelane_b32 v40, s6, 2 @@ -113,24 +114,23 @@ ; GISEL-NEXT: v_writelane_b32 v40, s21, 17 ; GISEL-NEXT: v_writelane_b32 v40, s22, 18 ; GISEL-NEXT: v_writelane_b32 v40, s23, 19 -; GISEL-NEXT: s_addk_i32 s32, 0x400 ; GISEL-NEXT: v_writelane_b32 v40, s24, 20 ; GISEL-NEXT: v_writelane_b32 v40, s25, 21 -; GISEL-NEXT: s_getpc_b64 s[34:35] -; GISEL-NEXT: s_add_u32 s34, s34, extern_c_func@gotpcrel32@lo+4 -; GISEL-NEXT: s_addc_u32 s35, s35, extern_c_func@gotpcrel32@hi+12 ; GISEL-NEXT: v_writelane_b32 v40, s26, 22 -; GISEL-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GISEL-NEXT: v_writelane_b32 v40, s27, 23 ; GISEL-NEXT: v_writelane_b32 v40, s28, 24 ; GISEL-NEXT: v_writelane_b32 v40, s29, 25 ; GISEL-NEXT: v_writelane_b32 v40, s30, 26 -; GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] ; GISEL-NEXT: v_writelane_b32 v40, s31, 27 +; GISEL-NEXT: s_getpc_b64 s[34:35] +; GISEL-NEXT: s_add_u32 s34, s34, extern_c_func@gotpcrel32@lo+4 +; GISEL-NEXT: s_addc_u32 s35, s35, extern_c_func@gotpcrel32@hi+12 +; GISEL-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] ; GISEL-NEXT: s_waitcnt lgkmcnt(0) ; GISEL-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GISEL-NEXT: v_readlane_b32 s31, v40, 27 ; GISEL-NEXT: v_readlane_b32 s30, v40, 26 +; GISEL-NEXT: v_readlane_b32 s31, v40, 27 ; GISEL-NEXT: v_readlane_b32 s29, v40, 25 ; GISEL-NEXT: v_readlane_b32 s28, v40, 24 ; GISEL-NEXT: v_readlane_b32 s27, v40, 23 diff --git a/llvm/test/CodeGen/AMDGPU/gfx-callable-argument-types.ll b/llvm/test/CodeGen/AMDGPU/gfx-callable-argument-types.ll --- a/llvm/test/CodeGen/AMDGPU/gfx-callable-argument-types.ll +++ b/llvm/test/CodeGen/AMDGPU/gfx-callable-argument-types.ll @@ -102,17 +102,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 -; GFX9-NEXT: v_mov_b32_e32 v0, 1 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: v_mov_b32_e32 v0, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i1@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i1@rel32@hi+12 ; GFX9-NEXT: buffer_store_byte v0, off, s[0:3], s32 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -132,16 +132,16 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i1@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i1@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: buffer_store_byte v0, off, s[0:3], s32 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -161,16 +161,16 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_mov_b32_e32 v0, 1 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_mov_b32_e32 v0, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i1@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i1@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: scratch_store_b8 off, v0, s32 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -190,16 +190,16 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i1@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i1@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: scratch_store_byte off, v0, s32 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -221,19 +221,19 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: global_load_ubyte v0, v[0:1], off glc -; GFX9-NEXT: s_waitcnt vmcnt(0) -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: global_load_ubyte v0, v[0:1], off glc +; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i1_signext@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i1_signext@rel32@hi+12 ; GFX9-NEXT: v_and_b32_e32 v0, 1, v0 ; GFX9-NEXT: buffer_store_byte v0, off, s[0:3], s32 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -252,19 +252,19 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: global_load_ubyte v0, v[0:1], off glc dlc -; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: global_load_ubyte v0, v[0:1], off glc dlc +; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i1_signext@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i1_signext@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_and_b32_e32 v0, 1, v0 ; GFX10-NEXT: buffer_store_byte v0, off, s[0:3], s32 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -283,19 +283,19 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: global_load_u8 v0, v[0:1], off glc dlc -; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: global_load_u8 v0, v[0:1], off glc dlc +; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i1_signext@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i1_signext@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: v_and_b32_e32 v0, 1, v0 ; GFX11-NEXT: scratch_store_b8 off, v0, s32 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -314,19 +314,19 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: global_load_ubyte v0, v[0:1], off glc dlc -; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(0) ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: global_load_ubyte v0, v[0:1], off glc dlc +; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(0) ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i1_signext@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i1_signext@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_and_b32_e32 v0, 1, v0 ; GFX10-SCRATCH-NEXT: scratch_store_byte off, v0, s32 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -349,19 +349,19 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: global_load_ubyte v0, v[0:1], off glc -; GFX9-NEXT: s_waitcnt vmcnt(0) -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: global_load_ubyte v0, v[0:1], off glc +; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i1_zeroext@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i1_zeroext@rel32@hi+12 ; GFX9-NEXT: v_and_b32_e32 v0, 1, v0 ; GFX9-NEXT: buffer_store_byte v0, off, s[0:3], s32 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -380,19 +380,19 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: global_load_ubyte v0, v[0:1], off glc dlc -; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: global_load_ubyte v0, v[0:1], off glc dlc +; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i1_zeroext@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i1_zeroext@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_and_b32_e32 v0, 1, v0 ; GFX10-NEXT: buffer_store_byte v0, off, s[0:3], s32 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -411,19 +411,19 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: global_load_u8 v0, v[0:1], off glc dlc -; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: global_load_u8 v0, v[0:1], off glc dlc +; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i1_zeroext@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i1_zeroext@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: v_and_b32_e32 v0, 1, v0 ; GFX11-NEXT: scratch_store_b8 off, v0, s32 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -442,19 +442,19 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: global_load_ubyte v0, v[0:1], off glc dlc -; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(0) ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: global_load_ubyte v0, v[0:1], off glc dlc +; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(0) ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i1_zeroext@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i1_zeroext@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_and_b32_e32 v0, 1, v0 ; GFX10-SCRATCH-NEXT: scratch_store_byte off, v0, s32 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -477,16 +477,16 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x7b +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i8@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i8@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -506,15 +506,15 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i8@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i8@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -534,16 +534,16 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i8@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i8@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -563,15 +563,15 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i8@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i8@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -593,17 +593,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: global_load_sbyte v0, v[0:1], off glc -; GFX9-NEXT: s_waitcnt vmcnt(0) -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: global_load_sbyte v0, v[0:1], off glc +; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i8_signext@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i8_signext@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -622,17 +622,17 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: global_load_sbyte v0, v[0:1], off glc dlc -; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: global_load_sbyte v0, v[0:1], off glc dlc +; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i8_signext@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i8_signext@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -651,18 +651,18 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: global_load_i8 v0, v[0:1], off glc dlc -; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: global_load_i8 v0, v[0:1], off glc dlc +; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i8_signext@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i8_signext@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -681,17 +681,17 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: global_load_sbyte v0, v[0:1], off glc dlc -; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(0) ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: global_load_sbyte v0, v[0:1], off glc dlc +; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(0) ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i8_signext@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i8_signext@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -714,17 +714,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: global_load_ubyte v0, v[0:1], off glc -; GFX9-NEXT: s_waitcnt vmcnt(0) -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: global_load_ubyte v0, v[0:1], off glc +; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i8_zeroext@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i8_zeroext@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -743,17 +743,17 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: global_load_ubyte v0, v[0:1], off glc dlc -; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: global_load_ubyte v0, v[0:1], off glc dlc +; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i8_zeroext@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i8_zeroext@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -772,18 +772,18 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: global_load_u8 v0, v[0:1], off glc dlc -; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: global_load_u8 v0, v[0:1], off glc dlc +; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i8_zeroext@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i8_zeroext@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -802,17 +802,17 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: global_load_ubyte v0, v[0:1], off glc dlc -; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(0) ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: global_load_ubyte v0, v[0:1], off glc dlc +; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(0) ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i8_zeroext@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i8_zeroext@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -835,16 +835,16 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x7b +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i16@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i16@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -864,15 +864,15 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i16@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i16@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -892,16 +892,16 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i16@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i16@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -921,15 +921,15 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i16@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i16@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -951,17 +951,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: global_load_ushort v0, v[0:1], off glc -; GFX9-NEXT: s_waitcnt vmcnt(0) -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: global_load_ushort v0, v[0:1], off glc +; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i16_signext@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i16_signext@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -980,17 +980,17 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: global_load_ushort v0, v[0:1], off glc dlc -; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: global_load_ushort v0, v[0:1], off glc dlc +; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i16_signext@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i16_signext@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -1009,18 +1009,18 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: global_load_u16 v0, v[0:1], off glc dlc -; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: global_load_u16 v0, v[0:1], off glc dlc +; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i16_signext@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i16_signext@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -1039,17 +1039,17 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: global_load_ushort v0, v[0:1], off glc dlc -; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(0) ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: global_load_ushort v0, v[0:1], off glc dlc +; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(0) ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i16_signext@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i16_signext@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -1072,17 +1072,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: global_load_ushort v0, v[0:1], off glc -; GFX9-NEXT: s_waitcnt vmcnt(0) -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: global_load_ushort v0, v[0:1], off glc +; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i16_zeroext@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i16_zeroext@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -1101,17 +1101,17 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: global_load_ushort v0, v[0:1], off glc dlc -; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: global_load_ushort v0, v[0:1], off glc dlc +; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i16_zeroext@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i16_zeroext@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -1130,18 +1130,18 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: global_load_u16 v0, v[0:1], off glc dlc -; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: s_getpc_b64 s[0:1] +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: global_load_u16 v0, v[0:1], off glc dlc +; GFX11-NEXT: s_waitcnt vmcnt(0) +; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i16_zeroext@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i16_zeroext@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -1160,17 +1160,17 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: global_load_ushort v0, v[0:1], off glc dlc -; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(0) ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: global_load_ushort v0, v[0:1], off glc dlc +; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(0) ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i16_zeroext@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i16_zeroext@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -1193,16 +1193,16 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 -; GFX9-NEXT: v_mov_b32_e32 v0, 42 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: v_mov_b32_e32 v0, 42 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -1222,15 +1222,15 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-NEXT: v_mov_b32_e32 v0, 42 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: v_mov_b32_e32 v0, 42 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i32@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -1250,16 +1250,16 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_mov_b32_e32 v0, 42 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_mov_b32_e32 v0, 42 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i32@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -1279,15 +1279,15 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 42 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 42 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i32@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -1309,17 +1309,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX9-NEXT: v_mov_b32_e32 v1, 0 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i64@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i64@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -1339,16 +1339,16 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX10-NEXT: v_mov_b32_e32 v1, 0 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i64@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i64@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -1368,16 +1368,16 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_dual_mov_b32 v0, 0x7b :: v_dual_mov_b32 v1, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v0, 0x7b :: v_dual_mov_b32 v1, 0 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i64@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i64@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -1397,16 +1397,16 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0x7b ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 0 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i64@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i64@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -1428,18 +1428,18 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 0 ; GFX9-NEXT: v_mov_b32_e32 v1, 0 ; GFX9-NEXT: global_load_dwordx4 v[0:3], v[0:1], off -; GFX9-NEXT: s_addk_i32 s32, 0x400 -; GFX9-NEXT: v_writelane_b32 v40, s30, 0 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2i64@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2i64@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -1458,18 +1458,18 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_mov_b32_e32 v0, 0 -; GFX10-NEXT: v_mov_b32_e32 v1, 0 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: v_mov_b32_e32 v0, 0 +; GFX10-NEXT: v_mov_b32_e32 v1, 0 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2i64@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2i64@rel32@hi+12 ; GFX10-NEXT: global_load_dwordx4 v[0:3], v[0:1], off -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -1488,19 +1488,18 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_mov_b32_e32 v0, 0 -; GFX11-NEXT: v_mov_b32_e32 v1, 0 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_mov_b32_e32 v0, 0 +; GFX11-NEXT: v_mov_b32_e32 v1, 0 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2i64@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2i64@rel32@hi+12 ; GFX11-NEXT: global_load_b128 v[0:3], v[0:1], off -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -1519,18 +1518,18 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 0 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2i64@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2i64@rel32@hi+12 ; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[0:3], v[0:1], off -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -1553,19 +1552,19 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 1 ; GFX9-NEXT: v_mov_b32_e32 v1, 2 ; GFX9-NEXT: v_mov_b32_e32 v2, 3 ; GFX9-NEXT: v_mov_b32_e32 v3, 4 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2i64@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2i64@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -1585,18 +1584,18 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-NEXT: v_mov_b32_e32 v1, 2 ; GFX10-NEXT: v_mov_b32_e32 v2, 3 ; GFX10-NEXT: v_mov_b32_e32 v3, 4 -; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2i64@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2i64@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -1616,17 +1615,17 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_dual_mov_b32 v0, 1 :: v_dual_mov_b32 v1, 2 -; GFX11-NEXT: v_dual_mov_b32 v2, 3 :: v_dual_mov_b32 v3, 4 ; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v0, 1 :: v_dual_mov_b32 v1, 2 +; GFX11-NEXT: v_dual_mov_b32 v2, 3 :: v_dual_mov_b32 v3, 4 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2i64@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2i64@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -1646,18 +1645,18 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 2 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 3 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v3, 4 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2i64@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2i64@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -1679,20 +1678,20 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 0 ; GFX9-NEXT: v_mov_b32_e32 v1, 0 ; GFX9-NEXT: global_load_dwordx4 v[0:3], v[0:1], off -; GFX9-NEXT: s_addk_i32 s32, 0x400 -; GFX9-NEXT: v_writelane_b32 v40, s30, 0 ; GFX9-NEXT: v_mov_b32_e32 v4, 1 ; GFX9-NEXT: v_mov_b32_e32 v5, 2 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3i64@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3i64@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -1711,20 +1710,20 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 0 ; GFX10-NEXT: v_mov_b32_e32 v1, 0 -; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: v_mov_b32_e32 v4, 1 ; GFX10-NEXT: v_mov_b32_e32 v5, 2 -; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: global_load_dwordx4 v[0:3], v[0:1], off -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3i64@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3i64@rel32@hi+12 +; GFX10-NEXT: global_load_dwordx4 v[0:3], v[0:1], off ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -1743,19 +1742,18 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v5, 2 -; GFX11-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v4, 1 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v5, 2 +; GFX11-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v4, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3i64@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3i64@rel32@hi+12 ; GFX11-NEXT: global_load_b128 v[0:3], v[0:1], off -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -1774,20 +1772,20 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 0 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v4, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v5, 2 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[0:3], v[0:1], off -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3i64@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3i64@rel32@hi+12 +; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[0:3], v[0:1], off ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -1812,22 +1810,22 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 0 ; GFX9-NEXT: v_mov_b32_e32 v1, 0 ; GFX9-NEXT: global_load_dwordx4 v[0:3], v[0:1], off -; GFX9-NEXT: s_addk_i32 s32, 0x400 -; GFX9-NEXT: v_writelane_b32 v40, s30, 0 ; GFX9-NEXT: v_mov_b32_e32 v4, 1 ; GFX9-NEXT: v_mov_b32_e32 v5, 2 ; GFX9-NEXT: v_mov_b32_e32 v6, 3 ; GFX9-NEXT: v_mov_b32_e32 v7, 4 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v4i64@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v4i64@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -1846,22 +1844,22 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 0 ; GFX10-NEXT: v_mov_b32_e32 v1, 0 -; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: v_mov_b32_e32 v4, 1 ; GFX10-NEXT: v_mov_b32_e32 v5, 2 ; GFX10-NEXT: v_mov_b32_e32 v6, 3 -; GFX10-NEXT: global_load_dwordx4 v[0:3], v[0:1], off ; GFX10-NEXT: v_mov_b32_e32 v7, 4 -; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: global_load_dwordx4 v[0:3], v[0:1], off ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v4i64@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v4i64@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -1880,20 +1878,19 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v5, 2 -; GFX11-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v4, 1 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_dual_mov_b32 v6, 3 :: v_dual_mov_b32 v7, 4 -; GFX11-NEXT: global_load_b128 v[0:3], v[0:1], off ; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v5, 2 +; GFX11-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v4, 1 +; GFX11-NEXT: v_dual_mov_b32 v6, 3 :: v_dual_mov_b32 v7, 4 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v4i64@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v4i64@rel32@hi+12 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX11-NEXT: global_load_b128 v[0:3], v[0:1], off ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -1912,22 +1909,22 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 0 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v4, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v5, 2 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v6, 3 -; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[0:3], v[0:1], off ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v7, 4 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[0:3], v[0:1], off ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v4i64@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v4i64@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -1951,16 +1948,16 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x4400 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: v_mov_b32_e32 v0, 0x4400 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_f16@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_f16@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -1980,15 +1977,15 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-NEXT: v_mov_b32_e32 v0, 0x4400 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: v_mov_b32_e32 v0, 0x4400 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_f16@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_f16@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -2008,16 +2005,16 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_mov_b32_e32 v0, 0x4400 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_mov_b32_e32 v0, 0x4400 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_f16@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_f16@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -2037,15 +2034,15 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0x4400 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0x4400 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_f16@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_f16@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -2067,16 +2064,16 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 -; GFX9-NEXT: v_mov_b32_e32 v0, 4.0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: v_mov_b32_e32 v0, 4.0 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_f32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_f32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -2096,15 +2093,15 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-NEXT: v_mov_b32_e32 v0, 4.0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: v_mov_b32_e32 v0, 4.0 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_f32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_f32@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -2124,16 +2121,16 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_mov_b32_e32 v0, 4.0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_mov_b32_e32 v0, 4.0 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_f32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_f32@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -2153,15 +2150,15 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 4.0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 4.0 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_f32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_f32@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -2183,17 +2180,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 1.0 ; GFX9-NEXT: v_mov_b32_e32 v1, 2.0 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2f32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2f32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -2213,16 +2210,16 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 1.0 ; GFX10-NEXT: v_mov_b32_e32 v1, 2.0 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2f32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2f32@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -2242,16 +2239,16 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_dual_mov_b32 v0, 1.0 :: v_dual_mov_b32 v1, 2.0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v0, 1.0 :: v_dual_mov_b32 v1, 2.0 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2f32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2f32@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -2271,16 +2268,16 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 1.0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 2.0 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2f32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2f32@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -2302,18 +2299,18 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 1.0 ; GFX9-NEXT: v_mov_b32_e32 v1, 2.0 ; GFX9-NEXT: v_mov_b32_e32 v2, 4.0 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3f32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3f32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -2333,17 +2330,17 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 1.0 ; GFX10-NEXT: v_mov_b32_e32 v1, 2.0 ; GFX10-NEXT: v_mov_b32_e32 v2, 4.0 -; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3f32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3f32@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -2363,17 +2360,17 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_dual_mov_b32 v0, 1.0 :: v_dual_mov_b32 v1, 2.0 -; GFX11-NEXT: v_mov_b32_e32 v2, 4.0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v0, 1.0 :: v_dual_mov_b32 v1, 2.0 +; GFX11-NEXT: v_mov_b32_e32 v2, 4.0 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3f32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3f32@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -2393,17 +2390,17 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 1.0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 2.0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 4.0 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3f32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3f32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -2425,20 +2422,20 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 1.0 ; GFX9-NEXT: v_mov_b32_e32 v1, 2.0 ; GFX9-NEXT: v_mov_b32_e32 v2, 4.0 ; GFX9-NEXT: v_mov_b32_e32 v3, -1.0 ; GFX9-NEXT: v_mov_b32_e32 v4, 0.5 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v5f32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v5f32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -2458,19 +2455,19 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 1.0 ; GFX10-NEXT: v_mov_b32_e32 v1, 2.0 ; GFX10-NEXT: v_mov_b32_e32 v2, 4.0 ; GFX10-NEXT: v_mov_b32_e32 v3, -1.0 ; GFX10-NEXT: v_mov_b32_e32 v4, 0.5 -; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v5f32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v5f32@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -2490,18 +2487,18 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 +; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: v_dual_mov_b32 v0, 1.0 :: v_dual_mov_b32 v1, 2.0 ; GFX11-NEXT: v_dual_mov_b32 v2, 4.0 :: v_dual_mov_b32 v3, -1.0 ; GFX11-NEXT: v_mov_b32_e32 v4, 0.5 -; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v5f32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v5f32@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -2521,19 +2518,19 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 1.0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 2.0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 4.0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v3, -1.0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v4, 0.5 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v5f32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v5f32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -2555,17 +2552,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 0 ; GFX9-NEXT: v_mov_b32_e32 v1, 0x40100000 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_f64@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_f64@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -2585,16 +2582,16 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 0 ; GFX10-NEXT: v_mov_b32_e32 v1, 0x40100000 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_f64@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_f64@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -2614,16 +2611,16 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x40100000 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0x40100000 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_f64@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_f64@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -2643,16 +2640,16 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 0x40100000 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_f64@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_f64@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -2674,19 +2671,19 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 0 ; GFX9-NEXT: v_mov_b32_e32 v1, 2.0 ; GFX9-NEXT: v_mov_b32_e32 v2, 0 ; GFX9-NEXT: v_mov_b32_e32 v3, 0x40100000 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2f64@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2f64@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -2706,18 +2703,18 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 0 ; GFX10-NEXT: v_mov_b32_e32 v1, 2.0 ; GFX10-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-NEXT: v_mov_b32_e32 v3, 0x40100000 -; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2f64@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2f64@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -2737,17 +2734,17 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 2.0 -; GFX11-NEXT: v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v3, 0x40100000 ; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 2.0 +; GFX11-NEXT: v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v3, 0x40100000 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2f64@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2f64@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -2767,18 +2764,18 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 2.0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v3, 0x40100000 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2f64@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2f64@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -2800,21 +2797,21 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 0 ; GFX9-NEXT: v_mov_b32_e32 v1, 2.0 ; GFX9-NEXT: v_mov_b32_e32 v2, 0 ; GFX9-NEXT: v_mov_b32_e32 v3, 0x40100000 ; GFX9-NEXT: v_mov_b32_e32 v4, 0 ; GFX9-NEXT: v_mov_b32_e32 v5, 0x40200000 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3f64@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3f64@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -2834,20 +2831,20 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 0 ; GFX10-NEXT: v_mov_b32_e32 v1, 2.0 ; GFX10-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-NEXT: v_mov_b32_e32 v3, 0x40100000 ; GFX10-NEXT: v_mov_b32_e32 v4, 0 ; GFX10-NEXT: v_mov_b32_e32 v5, 0x40200000 -; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3f64@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3f64@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -2867,18 +2864,18 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 +; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 2.0 ; GFX11-NEXT: v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v3, 0x40100000 ; GFX11-NEXT: v_dual_mov_b32 v4, 0 :: v_dual_mov_b32 v5, 0x40200000 -; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3f64@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3f64@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -2898,20 +2895,20 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 2.0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v3, 0x40100000 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v4, 0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v5, 0x40200000 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3f64@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3f64@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -2933,16 +2930,16 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: global_load_dword v0, v[0:1], off -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: global_load_dword v0, v[0:1], off ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2i16@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2i16@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -2961,16 +2958,16 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: global_load_dword v0, v[0:1], off ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: global_load_dword v0, v[0:1], off ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2i16@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2i16@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -2989,17 +2986,17 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: global_load_b32 v0, v[0:1], off ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: global_load_b32 v0, v[0:1], off ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2i16@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2i16@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -3018,16 +3015,16 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: global_load_dword v0, v[0:1], off ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: global_load_dword v0, v[0:1], off ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2i16@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2i16@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -3050,16 +3047,16 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3i16@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3i16@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -3078,16 +3075,16 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3i16@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3i16@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -3106,17 +3103,17 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: global_load_b64 v[0:1], v[0:1], off ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: global_load_b64 v[0:1], v[0:1], off ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3i16@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3i16@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -3135,16 +3132,16 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3i16@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3i16@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -3167,16 +3164,16 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3f16@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3f16@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -3195,16 +3192,16 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3f16@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3f16@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -3223,17 +3220,17 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: global_load_b64 v[0:1], v[0:1], off ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: global_load_b64 v[0:1], v[0:1], off ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3f16@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3f16@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -3252,16 +3249,16 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3f16@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3f16@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -3284,17 +3281,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 0x20001 ; GFX9-NEXT: v_mov_b32_e32 v1, 3 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3i16@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3i16@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -3314,16 +3311,16 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 0x20001 ; GFX10-NEXT: v_mov_b32_e32 v1, 3 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3i16@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3i16@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -3343,16 +3340,16 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_dual_mov_b32 v0, 0x20001 :: v_dual_mov_b32 v1, 3 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v0, 0x20001 :: v_dual_mov_b32 v1, 3 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3i16@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3i16@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -3372,16 +3369,16 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0x20001 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 3 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3i16@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3i16@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -3403,17 +3400,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 0x40003c00 ; GFX9-NEXT: v_mov_b32_e32 v1, 0x4400 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3f16@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3f16@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -3433,16 +3430,16 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 0x40003c00 ; GFX10-NEXT: v_mov_b32_e32 v1, 0x4400 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3f16@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3f16@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -3462,17 +3459,17 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 +; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: v_mov_b32_e32 v0, 0x40003c00 ; GFX11-NEXT: v_mov_b32_e32 v1, 0x4400 -; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3f16@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3f16@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -3492,16 +3489,16 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0x40003c00 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 0x4400 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3f16@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3f16@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -3523,16 +3520,16 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v4i16@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v4i16@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -3551,16 +3548,16 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v4i16@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v4i16@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -3579,17 +3576,17 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: global_load_b64 v[0:1], v[0:1], off ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: global_load_b64 v[0:1], v[0:1], off ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v4i16@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v4i16@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -3608,16 +3605,16 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v4i16@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v4i16@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -3640,17 +3637,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 0x20001 ; GFX9-NEXT: v_mov_b32_e32 v1, 0x40003 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v4i16@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v4i16@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -3670,16 +3667,16 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 0x20001 ; GFX10-NEXT: v_mov_b32_e32 v1, 0x40003 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v4i16@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v4i16@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -3699,17 +3696,17 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 +; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: v_mov_b32_e32 v0, 0x20001 ; GFX11-NEXT: v_mov_b32_e32 v1, 0x40003 -; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v4i16@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v4i16@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -3729,16 +3726,16 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0x20001 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 0x40003 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v4i16@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v4i16@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -3760,16 +3757,16 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: global_load_dword v0, v[0:1], off -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: global_load_dword v0, v[0:1], off ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2f16@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2f16@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -3788,16 +3785,16 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: global_load_dword v0, v[0:1], off ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: global_load_dword v0, v[0:1], off ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2f16@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2f16@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -3816,17 +3813,17 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: global_load_b32 v0, v[0:1], off ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: global_load_b32 v0, v[0:1], off ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2f16@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2f16@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -3845,16 +3842,16 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: global_load_dword v0, v[0:1], off ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: global_load_dword v0, v[0:1], off ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2f16@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2f16@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -3877,16 +3874,16 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -3905,16 +3902,16 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2i32@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -3933,17 +3930,17 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: global_load_b64 v[0:1], v[0:1], off ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: global_load_b64 v[0:1], v[0:1], off ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2i32@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -3962,16 +3959,16 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: global_load_dwordx2 v[0:1], v[0:1], off ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2i32@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -3994,17 +3991,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 1 ; GFX9-NEXT: v_mov_b32_e32 v1, 2 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -4024,16 +4021,16 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-NEXT: v_mov_b32_e32 v1, 2 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2i32@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -4053,16 +4050,16 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_dual_mov_b32 v0, 1 :: v_dual_mov_b32 v1, 2 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v0, 1 :: v_dual_mov_b32 v1, 2 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2i32@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -4082,16 +4079,16 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 2 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2i32@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -4113,18 +4110,18 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 3 ; GFX9-NEXT: v_mov_b32_e32 v1, 4 ; GFX9-NEXT: v_mov_b32_e32 v2, 5 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -4144,17 +4141,17 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 3 ; GFX10-NEXT: v_mov_b32_e32 v1, 4 ; GFX10-NEXT: v_mov_b32_e32 v2, 5 -; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3i32@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -4174,17 +4171,17 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_dual_mov_b32 v0, 3 :: v_dual_mov_b32 v1, 4 -; GFX11-NEXT: v_mov_b32_e32 v2, 5 ; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v0, 3 :: v_dual_mov_b32 v1, 4 +; GFX11-NEXT: v_mov_b32_e32 v2, 5 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3i32@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -4204,17 +4201,17 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 3 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 4 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 5 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3i32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -4236,19 +4233,19 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 3 ; GFX9-NEXT: v_mov_b32_e32 v1, 4 ; GFX9-NEXT: v_mov_b32_e32 v2, 5 ; GFX9-NEXT: v_mov_b32_e32 v3, 6 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3i32_i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3i32_i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -4268,18 +4265,18 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 3 ; GFX10-NEXT: v_mov_b32_e32 v1, 4 ; GFX10-NEXT: v_mov_b32_e32 v2, 5 ; GFX10-NEXT: v_mov_b32_e32 v3, 6 -; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3i32_i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3i32_i32@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -4299,17 +4296,17 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_dual_mov_b32 v0, 3 :: v_dual_mov_b32 v1, 4 -; GFX11-NEXT: v_dual_mov_b32 v2, 5 :: v_dual_mov_b32 v3, 6 ; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v0, 3 :: v_dual_mov_b32 v1, 4 +; GFX11-NEXT: v_dual_mov_b32 v2, 5 :: v_dual_mov_b32 v3, 6 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3i32_i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3i32_i32@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -4329,18 +4326,18 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 3 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 4 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 5 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v3, 6 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3i32_i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3i32_i32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -4362,16 +4359,16 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: global_load_dwordx4 v[0:3], v[0:1], off -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: global_load_dwordx4 v[0:3], v[0:1], off ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v4i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v4i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -4390,16 +4387,16 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: global_load_dwordx4 v[0:3], v[0:1], off ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: global_load_dwordx4 v[0:3], v[0:1], off ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v4i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v4i32@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -4418,17 +4415,17 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: global_load_b128 v[0:3], v[0:1], off ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: global_load_b128 v[0:3], v[0:1], off ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v4i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v4i32@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -4447,16 +4444,16 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[0:3], v[0:1], off ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[0:3], v[0:1], off ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v4i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v4i32@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -4479,19 +4476,19 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 1 ; GFX9-NEXT: v_mov_b32_e32 v1, 2 ; GFX9-NEXT: v_mov_b32_e32 v2, 3 ; GFX9-NEXT: v_mov_b32_e32 v3, 4 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v4i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v4i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -4511,18 +4508,18 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-NEXT: v_mov_b32_e32 v1, 2 ; GFX10-NEXT: v_mov_b32_e32 v2, 3 ; GFX10-NEXT: v_mov_b32_e32 v3, 4 -; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v4i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v4i32@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -4542,17 +4539,17 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_dual_mov_b32 v0, 1 :: v_dual_mov_b32 v1, 2 -; GFX11-NEXT: v_dual_mov_b32 v2, 3 :: v_dual_mov_b32 v3, 4 ; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v0, 1 :: v_dual_mov_b32 v1, 2 +; GFX11-NEXT: v_dual_mov_b32 v2, 3 :: v_dual_mov_b32 v3, 4 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v4i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v4i32@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -4572,18 +4569,18 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 2 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 3 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v3, 4 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v4i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v4i32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -4605,20 +4602,20 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 1 ; GFX9-NEXT: v_mov_b32_e32 v1, 2 ; GFX9-NEXT: v_mov_b32_e32 v2, 3 ; GFX9-NEXT: v_mov_b32_e32 v3, 4 ; GFX9-NEXT: v_mov_b32_e32 v4, 5 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v5i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v5i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -4638,19 +4635,19 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-NEXT: v_mov_b32_e32 v1, 2 ; GFX10-NEXT: v_mov_b32_e32 v2, 3 ; GFX10-NEXT: v_mov_b32_e32 v3, 4 ; GFX10-NEXT: v_mov_b32_e32 v4, 5 -; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v5i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v5i32@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -4670,18 +4667,18 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 +; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: v_dual_mov_b32 v0, 1 :: v_dual_mov_b32 v1, 2 ; GFX11-NEXT: v_dual_mov_b32 v2, 3 :: v_dual_mov_b32 v3, 4 ; GFX11-NEXT: v_mov_b32_e32 v4, 5 -; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v5i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v5i32@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -4701,19 +4698,19 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 2 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 3 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v3, 4 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v4, 5 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v5i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v5i32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -4735,11 +4732,11 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX9-NEXT: v_mov_b32_e32 v8, 0 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; GFX9-NEXT: v_mov_b32_e32 v8, 0 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: global_load_dwordx4 v[0:3], v8, s[34:35] ; GFX9-NEXT: global_load_dwordx4 v[4:7], v8, s[34:35] offset:16 @@ -4747,8 +4744,8 @@ ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v8i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v8i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -4767,21 +4764,22 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX10-NEXT: v_mov_b32_e32 v8, 0 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; GFX10-NEXT: v_mov_b32_e32 v8, 0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_clause 0x1 ; GFX10-NEXT: global_load_dwordx4 v[0:3], v8, s[34:35] ; GFX10-NEXT: global_load_dwordx4 v[4:7], v8, s[34:35] offset:16 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v8i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v8i32@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -4800,22 +4798,22 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 -; GFX11-NEXT: v_mov_b32_e32 v4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 +; GFX11-NEXT: v_mov_b32_e32 v4, 0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: global_load_b128 v[0:3], v4, s[0:1] ; GFX11-NEXT: global_load_b128 v[4:7], v4, s[0:1] offset:16 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v8i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v8i32@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -4834,21 +4832,22 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v8, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v8, 0 ; GFX10-SCRATCH-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-SCRATCH-NEXT: s_clause 0x1 ; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[0:3], v8, s[0:1] ; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[4:7], v8, s[0:1] offset:16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v8i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v8i32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -4872,8 +4871,9 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 1 ; GFX9-NEXT: v_mov_b32_e32 v1, 2 ; GFX9-NEXT: v_mov_b32_e32 v2, 3 @@ -4882,13 +4882,12 @@ ; GFX9-NEXT: v_mov_b32_e32 v5, 6 ; GFX9-NEXT: v_mov_b32_e32 v6, 7 ; GFX9-NEXT: v_mov_b32_e32 v7, 8 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v8i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v8i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -4908,6 +4907,8 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-NEXT: v_mov_b32_e32 v1, 2 ; GFX10-NEXT: v_mov_b32_e32 v2, 3 @@ -4916,14 +4917,12 @@ ; GFX10-NEXT: v_mov_b32_e32 v5, 6 ; GFX10-NEXT: v_mov_b32_e32 v6, 7 ; GFX10-NEXT: v_mov_b32_e32 v7, 8 -; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v8i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v8i32@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -4943,19 +4942,19 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 +; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: v_dual_mov_b32 v0, 1 :: v_dual_mov_b32 v1, 2 ; GFX11-NEXT: v_dual_mov_b32 v2, 3 :: v_dual_mov_b32 v3, 4 ; GFX11-NEXT: v_dual_mov_b32 v4, 5 :: v_dual_mov_b32 v5, 6 ; GFX11-NEXT: v_dual_mov_b32 v6, 7 :: v_dual_mov_b32 v7, 8 -; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v8i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v8i32@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -4975,6 +4974,8 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 2 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 3 @@ -4983,14 +4984,12 @@ ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v5, 6 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v6, 7 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v7, 8 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v8i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v8i32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -5012,11 +5011,11 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX9-NEXT: v_mov_b32_e32 v16, 0 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; GFX9-NEXT: v_mov_b32_e32 v16, 0 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: global_load_dwordx4 v[0:3], v16, s[34:35] ; GFX9-NEXT: global_load_dwordx4 v[4:7], v16, s[34:35] offset:16 @@ -5026,8 +5025,8 @@ ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v16i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v16i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -5046,23 +5045,24 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX10-NEXT: v_mov_b32_e32 v16, 0 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; GFX10-NEXT: v_mov_b32_e32 v16, 0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_clause 0x3 ; GFX10-NEXT: global_load_dwordx4 v[0:3], v16, s[34:35] ; GFX10-NEXT: global_load_dwordx4 v[4:7], v16, s[34:35] offset:16 ; GFX10-NEXT: global_load_dwordx4 v[8:11], v16, s[34:35] offset:32 ; GFX10-NEXT: global_load_dwordx4 v[12:15], v16, s[34:35] offset:48 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v16i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v16i32@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -5081,24 +5081,24 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 -; GFX11-NEXT: v_mov_b32_e32 v12, 0 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 +; GFX11-NEXT: v_mov_b32_e32 v12, 0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_clause 0x3 ; GFX11-NEXT: global_load_b128 v[0:3], v12, s[0:1] ; GFX11-NEXT: global_load_b128 v[4:7], v12, s[0:1] offset:16 ; GFX11-NEXT: global_load_b128 v[8:11], v12, s[0:1] offset:32 ; GFX11-NEXT: global_load_b128 v[12:15], v12, s[0:1] offset:48 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v16i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v16i32@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -5117,23 +5117,24 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v16, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v16, 0 ; GFX10-SCRATCH-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-SCRATCH-NEXT: s_clause 0x3 ; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[0:3], v16, s[0:1] ; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[4:7], v16, s[0:1] offset:16 ; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[8:11], v16, s[0:1] offset:32 ; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[12:15], v16, s[0:1] offset:48 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v16i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v16i32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -5157,11 +5158,11 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX9-NEXT: v_mov_b32_e32 v28, 0 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; GFX9-NEXT: v_mov_b32_e32 v28, 0 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: global_load_dwordx4 v[0:3], v28, s[34:35] ; GFX9-NEXT: global_load_dwordx4 v[4:7], v28, s[34:35] offset:16 @@ -5176,8 +5177,8 @@ ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v32i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v32i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -5196,10 +5197,11 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX10-NEXT: v_mov_b32_e32 v32, 0 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; GFX10-NEXT: v_mov_b32_e32 v32, 0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_clause 0x7 ; GFX10-NEXT: global_load_dwordx4 v[0:3], v32, s[34:35] @@ -5210,13 +5212,13 @@ ; GFX10-NEXT: global_load_dwordx4 v[20:23], v32, s[34:35] offset:80 ; GFX10-NEXT: global_load_dwordx4 v[24:27], v32, s[34:35] offset:96 ; GFX10-NEXT: global_load_dwordx4 v[28:31], v32, s[34:35] offset:112 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v32i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v32i32@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -5235,10 +5237,11 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 -; GFX11-NEXT: v_mov_b32_e32 v28, 0 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 +; GFX11-NEXT: v_mov_b32_e32 v28, 0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_clause 0x7 ; GFX11-NEXT: global_load_b128 v[0:3], v28, s[0:1] @@ -5249,14 +5252,13 @@ ; GFX11-NEXT: global_load_b128 v[20:23], v28, s[0:1] offset:80 ; GFX11-NEXT: global_load_b128 v[24:27], v28, s[0:1] offset:96 ; GFX11-NEXT: global_load_b128 v[28:31], v28, s[0:1] offset:112 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v32i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v32i32@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -5275,10 +5277,11 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v32, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v32, 0 ; GFX10-SCRATCH-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-SCRATCH-NEXT: s_clause 0x7 ; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[0:3], v32, s[0:1] @@ -5289,13 +5292,13 @@ ; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[20:23], v32, s[0:1] offset:80 ; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[24:27], v32, s[0:1] offset:96 ; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[28:31], v32, s[0:1] offset:112 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v32i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v32i32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -5319,11 +5322,12 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX9-NEXT: v_mov_b32_e32 v28, 0 ; GFX9-NEXT: global_load_dword v32, v[0:1], off -; GFX9-NEXT: s_addk_i32 s32, 0x400 -; GFX9-NEXT: v_writelane_b32 v40, s30, 0 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: global_load_dwordx4 v[0:3], v28, s[34:35] ; GFX9-NEXT: global_load_dwordx4 v[4:7], v28, s[34:35] offset:16 @@ -5334,15 +5338,14 @@ ; GFX9-NEXT: global_load_dwordx4 v[24:27], v28, s[34:35] offset:96 ; GFX9-NEXT: s_nop 0 ; GFX9-NEXT: global_load_dwordx4 v[28:31], v28, s[34:35] offset:112 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v32i32_i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v32i32_i32@rel32@hi+12 ; GFX9-NEXT: s_waitcnt vmcnt(8) ; GFX9-NEXT: buffer_store_dword v32, off, s[0:3], s32 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -5361,10 +5364,11 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX10-NEXT: v_mov_b32_e32 v32, 0 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; GFX10-NEXT: v_mov_b32_e32 v32, 0 ; GFX10-NEXT: global_load_dword v33, v[0:1], off ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_clause 0x7 @@ -5376,15 +5380,15 @@ ; GFX10-NEXT: global_load_dwordx4 v[20:23], v32, s[34:35] offset:80 ; GFX10-NEXT: global_load_dwordx4 v[24:27], v32, s[34:35] offset:96 ; GFX10-NEXT: global_load_dwordx4 v[28:31], v32, s[34:35] offset:112 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v32i32_i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v32i32_i32@rel32@hi+12 ; GFX10-NEXT: s_waitcnt vmcnt(8) ; GFX10-NEXT: buffer_store_dword v33, off, s[0:3], s32 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -5403,10 +5407,11 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 -; GFX11-NEXT: v_mov_b32_e32 v28, 0 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 +; GFX11-NEXT: v_mov_b32_e32 v28, 0 ; GFX11-NEXT: global_load_b32 v32, v[0:1], off ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_clause 0x7 @@ -5418,15 +5423,14 @@ ; GFX11-NEXT: global_load_b128 v[20:23], v28, s[0:1] offset:80 ; GFX11-NEXT: global_load_b128 v[24:27], v28, s[0:1] offset:96 ; GFX11-NEXT: global_load_b128 v[28:31], v28, s[0:1] offset:112 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v32i32_i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v32i32_i32@rel32@hi+12 ; GFX11-NEXT: s_waitcnt vmcnt(8) ; GFX11-NEXT: scratch_store_b32 off, v32, s32 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -5445,10 +5449,11 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v32, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v32, 0 ; GFX10-SCRATCH-NEXT: global_load_dword v33, v[0:1], off ; GFX10-SCRATCH-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-SCRATCH-NEXT: s_clause 0x7 @@ -5460,15 +5465,15 @@ ; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[20:23], v32, s[0:1] offset:80 ; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[24:27], v32, s[0:1] offset:96 ; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[28:31], v32, s[0:1] offset:112 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v32i32_i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v32i32_i32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(8) ; GFX10-SCRATCH-NEXT: scratch_store_dword off, v33, s32 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -5497,9 +5502,9 @@ ; GFX9-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill ; GFX9-NEXT: buffer_store_dword v42, off, s[0:3], s33 ; 4-byte Folded Spill ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v41, v0 ; GFX9-NEXT: v_mov_b32_e32 v0, 42 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v42, v1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_i32_func_i32@rel32@lo+4 @@ -5509,8 +5514,8 @@ ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: buffer_load_dword v42, off, s[0:3], s33 ; 4-byte Folded Reload ; GFX9-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -5529,14 +5534,14 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill ; GFX10-NEXT: buffer_store_dword v42, off, s[0:3], s33 ; 4-byte Folded Spill ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v41, v0 ; GFX10-NEXT: v_mov_b32_e32 v0, 42 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_mov_b32_e32 v42, v1 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_i32_func_i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_i32_func_i32@rel32@hi+12 @@ -5546,8 +5551,8 @@ ; GFX10-NEXT: s_clause 0x1 ; GFX10-NEXT: buffer_load_dword v42, off, s[0:3], s33 ; GFX10-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:4 -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -5566,14 +5571,15 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: scratch_store_b32 off, v41, s33 offset:4 +; GFX11-NEXT: ; meta instruction ; GFX11-NEXT: scratch_store_b32 off, v42, s33 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: v_dual_mov_b32 v42, v1 :: v_dual_mov_b32 v41, v0 ; GFX11-NEXT: v_mov_b32_e32 v0, 42 -; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_i32_func_i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_i32_func_i32@rel32@hi+12 @@ -5584,8 +5590,8 @@ ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: scratch_load_b32 v42, off, s33 ; GFX11-NEXT: scratch_load_b32 v41, off, s33 offset:4 -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -5604,14 +5610,14 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: scratch_store_dword off, v41, s33 offset:4 ; 4-byte Folded Spill ; GFX10-SCRATCH-NEXT: scratch_store_dword off, v42, s33 ; 4-byte Folded Spill ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v41, v0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 42 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v42, v1 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_i32_func_i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_i32_func_i32@rel32@hi+12 @@ -5621,8 +5627,8 @@ ; GFX10-SCRATCH-NEXT: s_clause 0x1 ; GFX10-SCRATCH-NEXT: scratch_load_dword v42, off, s33 ; GFX10-SCRATCH-NEXT: scratch_load_dword v41, off, s33 offset:4 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -5645,11 +5651,11 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX9-NEXT: v_mov_b32_e32 v2, 0 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: global_load_ubyte v0, v2, s[34:35] ; GFX9-NEXT: global_load_dword v1, v2, s[34:35] offset:4 @@ -5657,8 +5663,8 @@ ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_struct_i8_i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_struct_i8_i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -5677,21 +5683,22 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX10-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; GFX10-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_clause 0x1 ; GFX10-NEXT: global_load_ubyte v0, v2, s[34:35] ; GFX10-NEXT: global_load_dword v1, v2, s[34:35] offset:4 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_struct_i8_i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_struct_i8_i32@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -5710,22 +5717,22 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 -; GFX11-NEXT: v_mov_b32_e32 v1, 0 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 +; GFX11-NEXT: v_mov_b32_e32 v1, 0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: global_load_u8 v0, v1, s[0:1] ; GFX11-NEXT: global_load_b32 v1, v1, s[0:1] offset:4 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_struct_i8_i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_struct_i8_i32@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -5744,21 +5751,22 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-SCRATCH-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-SCRATCH-NEXT: s_clause 0x1 ; GFX10-SCRATCH-NEXT: global_load_ubyte v0, v2, s[0:1] ; GFX10-SCRATCH-NEXT: global_load_dword v1, v2, s[0:1] offset:4 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_struct_i8_i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_struct_i8_i32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -5782,20 +5790,20 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 3 ; GFX9-NEXT: buffer_store_byte v0, off, s[0:3], s33 ; GFX9-NEXT: v_mov_b32_e32 v0, 8 -; GFX9-NEXT: s_addk_i32 s32, 0x400 -; GFX9-NEXT: v_writelane_b32 v40, s30, 0 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:4 ; GFX9-NEXT: v_lshrrev_b32_e64 v0, 6, s33 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_byval_struct_i8_i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_byval_struct_i8_i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -5814,20 +5822,20 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_mov_b32_e32 v0, 3 -; GFX10-NEXT: v_mov_b32_e32 v1, 8 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: v_mov_b32_e32 v0, 3 +; GFX10-NEXT: v_mov_b32_e32 v1, 8 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_byval_struct_i8_i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_byval_struct_i8_i32@rel32@hi+12 ; GFX10-NEXT: buffer_store_byte v0, off, s[0:3], s33 ; GFX10-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:4 ; GFX10-NEXT: v_lshrrev_b32_e64 v0, 5, s33 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -5846,9 +5854,10 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_dual_mov_b32 v0, 3 :: v_dual_mov_b32 v1, 8 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v0, 3 :: v_dual_mov_b32 v1, 8 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_byval_struct_i8_i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_byval_struct_i8_i32@rel32@hi+12 @@ -5856,11 +5865,9 @@ ; GFX11-NEXT: scratch_store_b8 off, v0, s33 ; GFX11-NEXT: scratch_store_b32 off, v1, s33 offset:4 ; GFX11-NEXT: v_mov_b32_e32 v0, s33 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -5879,20 +5886,20 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 3 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 8 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 3 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 8 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_byval_struct_i8_i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_byval_struct_i8_i32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: scratch_store_byte off, v0, s33 ; GFX10-SCRATCH-NEXT: scratch_store_dword off, v1, s33 offset:4 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, s33 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -5919,24 +5926,24 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x800 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 3 ; GFX9-NEXT: buffer_store_byte v0, off, s[0:3], s33 ; GFX9-NEXT: v_mov_b32_e32 v0, 8 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:4 ; GFX9-NEXT: v_lshrrev_b32_e64 v0, 6, s33 -; GFX9-NEXT: s_addk_i32 s32, 0x800 -; GFX9-NEXT: v_writelane_b32 v40, s30, 0 ; GFX9-NEXT: v_add_u32_e32 v0, 8, v0 ; GFX9-NEXT: v_lshrrev_b32_e64 v1, 6, s33 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_sret_struct_i8_i32_byval_struct_i8_i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_sret_struct_i8_i32_byval_struct_i8_i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] ; GFX9-NEXT: buffer_load_ubyte v0, off, s[0:3], s33 offset:8 ; GFX9-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:12 -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xf800 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_waitcnt vmcnt(0) @@ -5960,10 +5967,11 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_mov_b32_e32 v0, 3 -; GFX10-NEXT: v_mov_b32_e32 v1, 8 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x400 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: v_mov_b32_e32 v0, 3 +; GFX10-NEXT: v_mov_b32_e32 v1, 8 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_sret_struct_i8_i32_byval_struct_i8_i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_sret_struct_i8_i32_byval_struct_i8_i32@rel32@hi+12 @@ -5971,14 +5979,13 @@ ; GFX10-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:4 ; GFX10-NEXT: v_lshrrev_b32_e64 v0, 5, s33 ; GFX10-NEXT: v_lshrrev_b32_e64 v1, 5, s33 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_add_nc_u32_e32 v0, 8, v0 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] ; GFX10-NEXT: s_clause 0x1 ; GFX10-NEXT: buffer_load_ubyte v0, off, s[0:3], s33 offset:8 ; GFX10-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:12 -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfc00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_waitcnt vmcnt(0) @@ -6003,24 +6010,24 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_dual_mov_b32 v0, 3 :: v_dual_mov_b32 v1, 8 +; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 32 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v0, 3 :: v_dual_mov_b32 v1, 8 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_sret_struct_i8_i32_byval_struct_i8_i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_sret_struct_i8_i32_byval_struct_i8_i32@rel32@hi+12 ; GFX11-NEXT: s_add_i32 vcc_lo, s33, 8 -; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: scratch_store_b8 off, v0, s33 ; GFX11-NEXT: scratch_store_b32 off, v1, s33 offset:4 ; GFX11-NEXT: v_dual_mov_b32 v0, vcc_lo :: v_dual_mov_b32 v1, s33 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: scratch_load_u8 v0, off, s33 offset:8 ; GFX11-NEXT: scratch_load_b32 v1, off, s33 offset:12 -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_addk_i32 s32, 0xffe0 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_waitcnt vmcnt(0) @@ -6045,25 +6052,25 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 3 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 32 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 3 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 8 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_sret_struct_i8_i32_byval_struct_i8_i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_sret_struct_i8_i32_byval_struct_i8_i32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_add_i32 vcc_lo, s33, 8 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: scratch_store_byte off, v0, s33 ; GFX10-SCRATCH-NEXT: scratch_store_dword off, v1, s33 offset:4 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, vcc_lo ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, s33 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] ; GFX10-SCRATCH-NEXT: s_clause 0x1 ; GFX10-SCRATCH-NEXT: scratch_load_ubyte v0, off, s33 offset:8 ; GFX10-SCRATCH-NEXT: scratch_load_dword v1, off, s33 offset:12 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_addk_i32 s32, 0xffe0 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(0) @@ -6104,11 +6111,11 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX9-NEXT: v_mov_b32_e32 v0, 0 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; GFX9-NEXT: v_mov_b32_e32 v0, 0 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: global_load_dwordx4 v[0:3], v0, s[34:35] ; GFX9-NEXT: s_getpc_b64 s[34:35] @@ -6134,8 +6141,8 @@ ; GFX9-NEXT: v_mov_b32_e32 v2, v17 ; GFX9-NEXT: v_mov_b32_e32 v3, v18 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -6154,11 +6161,11 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX10-NEXT: v_mov_b32_e32 v0, 0 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; GFX10-NEXT: v_mov_b32_e32 v0, 0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: global_load_dwordx4 v[0:3], v0, s[34:35] ; GFX10-NEXT: s_waitcnt_depctr 0xffe3 @@ -6185,8 +6192,8 @@ ; GFX10-NEXT: v_mov_b32_e32 v2, v17 ; GFX10-NEXT: v_mov_b32_e32 v3, v18 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -6205,11 +6212,11 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 -; GFX11-NEXT: v_mov_b32_e32 v0, 0 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 +; GFX11-NEXT: v_mov_b32_e32 v0, 0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: global_load_b128 v[0:3], v0, s[0:1] ; GFX11-NEXT: s_getpc_b64 s[0:1] @@ -6233,8 +6240,8 @@ ; GFX11-NEXT: v_dual_mov_b32 v12, v3 :: v_dual_mov_b32 v3, v18 ; GFX11-NEXT: v_dual_mov_b32 v1, v16 :: v_dual_mov_b32 v2, v17 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -6253,11 +6260,11 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0 ; GFX10-SCRATCH-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-SCRATCH-NEXT: global_load_dwordx4 v[0:3], v0, s[0:1] ; GFX10-SCRATCH-NEXT: s_waitcnt_depctr 0xffe3 @@ -6284,8 +6291,8 @@ ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, v17 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v3, v18 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -6309,42 +6316,42 @@ ; GFX9-NEXT: s_mov_b64 exec, s[4:5] ; GFX9-NEXT: v_writelane_b32 v40, s33, 32 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x800 +; GFX9-NEXT: v_writelane_b32 v40, s34, 0 +; GFX9-NEXT: v_writelane_b32 v40, s35, 1 +; GFX9-NEXT: v_writelane_b32 v40, s36, 2 +; GFX9-NEXT: v_writelane_b32 v40, s37, 3 +; GFX9-NEXT: v_writelane_b32 v40, s38, 4 +; GFX9-NEXT: v_writelane_b32 v40, s39, 5 +; GFX9-NEXT: v_writelane_b32 v40, s40, 6 +; GFX9-NEXT: v_writelane_b32 v40, s41, 7 +; GFX9-NEXT: v_writelane_b32 v40, s42, 8 +; GFX9-NEXT: v_writelane_b32 v40, s43, 9 +; GFX9-NEXT: v_writelane_b32 v40, s44, 10 +; GFX9-NEXT: v_writelane_b32 v40, s45, 11 +; GFX9-NEXT: v_writelane_b32 v40, s46, 12 +; GFX9-NEXT: v_writelane_b32 v40, s47, 13 +; GFX9-NEXT: v_writelane_b32 v40, s48, 14 +; GFX9-NEXT: v_writelane_b32 v40, s49, 15 +; GFX9-NEXT: v_writelane_b32 v40, s50, 16 +; GFX9-NEXT: v_writelane_b32 v40, s51, 17 +; GFX9-NEXT: v_writelane_b32 v40, s52, 18 +; GFX9-NEXT: v_writelane_b32 v40, s53, 19 +; GFX9-NEXT: v_writelane_b32 v40, s54, 20 +; GFX9-NEXT: v_writelane_b32 v40, s55, 21 +; GFX9-NEXT: v_writelane_b32 v40, s56, 22 +; GFX9-NEXT: v_writelane_b32 v40, s57, 23 +; GFX9-NEXT: v_writelane_b32 v40, s58, 24 +; GFX9-NEXT: v_writelane_b32 v40, s59, 25 +; GFX9-NEXT: v_writelane_b32 v40, s60, 26 +; GFX9-NEXT: v_writelane_b32 v40, s61, 27 +; GFX9-NEXT: v_writelane_b32 v40, s62, 28 +; GFX9-NEXT: v_writelane_b32 v40, s63, 29 +; GFX9-NEXT: v_writelane_b32 v40, s30, 30 +; GFX9-NEXT: v_writelane_b32 v40, s31, 31 ; GFX9-NEXT: buffer_load_dword v32, off, s[0:3], s33 offset:20 ; GFX9-NEXT: buffer_load_dword v33, off, s[0:3], s33 offset:16 ; GFX9-NEXT: buffer_load_dword v31, off, s[0:3], s33 -; GFX9-NEXT: v_writelane_b32 v40, s30, 0 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 -; GFX9-NEXT: v_writelane_b32 v40, s34, 2 -; GFX9-NEXT: v_writelane_b32 v40, s35, 3 -; GFX9-NEXT: v_writelane_b32 v40, s36, 4 -; GFX9-NEXT: v_writelane_b32 v40, s37, 5 -; GFX9-NEXT: v_writelane_b32 v40, s38, 6 -; GFX9-NEXT: v_writelane_b32 v40, s39, 7 -; GFX9-NEXT: v_writelane_b32 v40, s40, 8 -; GFX9-NEXT: v_writelane_b32 v40, s41, 9 -; GFX9-NEXT: v_writelane_b32 v40, s42, 10 -; GFX9-NEXT: v_writelane_b32 v40, s43, 11 -; GFX9-NEXT: v_writelane_b32 v40, s44, 12 -; GFX9-NEXT: v_writelane_b32 v40, s45, 13 -; GFX9-NEXT: v_writelane_b32 v40, s46, 14 -; GFX9-NEXT: v_writelane_b32 v40, s47, 15 -; GFX9-NEXT: v_writelane_b32 v40, s48, 16 -; GFX9-NEXT: v_writelane_b32 v40, s49, 17 -; GFX9-NEXT: v_writelane_b32 v40, s50, 18 -; GFX9-NEXT: v_writelane_b32 v40, s51, 19 -; GFX9-NEXT: v_writelane_b32 v40, s52, 20 -; GFX9-NEXT: v_writelane_b32 v40, s53, 21 -; GFX9-NEXT: v_writelane_b32 v40, s54, 22 -; GFX9-NEXT: v_writelane_b32 v40, s55, 23 -; GFX9-NEXT: v_writelane_b32 v40, s56, 24 -; GFX9-NEXT: v_writelane_b32 v40, s57, 25 -; GFX9-NEXT: v_writelane_b32 v40, s58, 26 -; GFX9-NEXT: v_writelane_b32 v40, s59, 27 -; GFX9-NEXT: v_writelane_b32 v40, s60, 28 -; GFX9-NEXT: v_writelane_b32 v40, s61, 29 -; GFX9-NEXT: s_addk_i32 s32, 0x800 -; GFX9-NEXT: v_writelane_b32 v40, s62, 30 -; GFX9-NEXT: v_writelane_b32 v40, s63, 31 ; GFX9-NEXT: s_getpc_b64 s[4:5] ; GFX9-NEXT: s_add_u32 s4, s4, byval_align16_f64_arg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s5, s5, byval_align16_f64_arg@rel32@hi+12 @@ -6353,38 +6360,38 @@ ; GFX9-NEXT: s_waitcnt vmcnt(2) ; GFX9-NEXT: buffer_store_dword v33, off, s[0:3], s32 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[4:5] -; GFX9-NEXT: v_readlane_b32 s63, v40, 31 -; GFX9-NEXT: v_readlane_b32 s62, v40, 30 -; GFX9-NEXT: v_readlane_b32 s61, v40, 29 -; GFX9-NEXT: v_readlane_b32 s60, v40, 28 -; GFX9-NEXT: v_readlane_b32 s59, v40, 27 -; GFX9-NEXT: v_readlane_b32 s58, v40, 26 -; GFX9-NEXT: v_readlane_b32 s57, v40, 25 -; GFX9-NEXT: v_readlane_b32 s56, v40, 24 -; GFX9-NEXT: v_readlane_b32 s55, v40, 23 -; GFX9-NEXT: v_readlane_b32 s54, v40, 22 -; GFX9-NEXT: v_readlane_b32 s53, v40, 21 -; GFX9-NEXT: v_readlane_b32 s52, v40, 20 -; GFX9-NEXT: v_readlane_b32 s51, v40, 19 -; GFX9-NEXT: v_readlane_b32 s50, v40, 18 -; GFX9-NEXT: v_readlane_b32 s49, v40, 17 -; GFX9-NEXT: v_readlane_b32 s48, v40, 16 -; GFX9-NEXT: v_readlane_b32 s47, v40, 15 -; GFX9-NEXT: v_readlane_b32 s46, v40, 14 -; GFX9-NEXT: v_readlane_b32 s45, v40, 13 -; GFX9-NEXT: v_readlane_b32 s44, v40, 12 -; GFX9-NEXT: v_readlane_b32 s43, v40, 11 -; GFX9-NEXT: v_readlane_b32 s42, v40, 10 -; GFX9-NEXT: v_readlane_b32 s41, v40, 9 -; GFX9-NEXT: v_readlane_b32 s40, v40, 8 -; GFX9-NEXT: v_readlane_b32 s39, v40, 7 -; GFX9-NEXT: v_readlane_b32 s38, v40, 6 -; GFX9-NEXT: v_readlane_b32 s37, v40, 5 -; GFX9-NEXT: v_readlane_b32 s36, v40, 4 -; GFX9-NEXT: v_readlane_b32 s35, v40, 3 -; GFX9-NEXT: v_readlane_b32 s34, v40, 2 -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 -; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s30, v40, 30 +; GFX9-NEXT: v_readlane_b32 s31, v40, 31 +; GFX9-NEXT: v_readlane_b32 s63, v40, 29 +; GFX9-NEXT: v_readlane_b32 s62, v40, 28 +; GFX9-NEXT: v_readlane_b32 s61, v40, 27 +; GFX9-NEXT: v_readlane_b32 s60, v40, 26 +; GFX9-NEXT: v_readlane_b32 s59, v40, 25 +; GFX9-NEXT: v_readlane_b32 s58, v40, 24 +; GFX9-NEXT: v_readlane_b32 s57, v40, 23 +; GFX9-NEXT: v_readlane_b32 s56, v40, 22 +; GFX9-NEXT: v_readlane_b32 s55, v40, 21 +; GFX9-NEXT: v_readlane_b32 s54, v40, 20 +; GFX9-NEXT: v_readlane_b32 s53, v40, 19 +; GFX9-NEXT: v_readlane_b32 s52, v40, 18 +; GFX9-NEXT: v_readlane_b32 s51, v40, 17 +; GFX9-NEXT: v_readlane_b32 s50, v40, 16 +; GFX9-NEXT: v_readlane_b32 s49, v40, 15 +; GFX9-NEXT: v_readlane_b32 s48, v40, 14 +; GFX9-NEXT: v_readlane_b32 s47, v40, 13 +; GFX9-NEXT: v_readlane_b32 s46, v40, 12 +; GFX9-NEXT: v_readlane_b32 s45, v40, 11 +; GFX9-NEXT: v_readlane_b32 s44, v40, 10 +; GFX9-NEXT: v_readlane_b32 s43, v40, 9 +; GFX9-NEXT: v_readlane_b32 s42, v40, 8 +; GFX9-NEXT: v_readlane_b32 s41, v40, 7 +; GFX9-NEXT: v_readlane_b32 s40, v40, 6 +; GFX9-NEXT: v_readlane_b32 s39, v40, 5 +; GFX9-NEXT: v_readlane_b32 s38, v40, 4 +; GFX9-NEXT: v_readlane_b32 s37, v40, 3 +; GFX9-NEXT: v_readlane_b32 s36, v40, 2 +; GFX9-NEXT: v_readlane_b32 s35, v40, 1 +; GFX9-NEXT: v_readlane_b32 s34, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xf800 ; GFX9-NEXT: v_readlane_b32 s33, v40, 32 ; GFX9-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -6403,12 +6410,43 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s4 ; GFX10-NEXT: s_mov_b32 s6, s33 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x400 +; GFX10-NEXT: v_writelane_b32 v40, s34, 0 +; GFX10-NEXT: v_writelane_b32 v40, s35, 1 +; GFX10-NEXT: v_writelane_b32 v40, s36, 2 +; GFX10-NEXT: v_writelane_b32 v40, s37, 3 +; GFX10-NEXT: v_writelane_b32 v40, s38, 4 +; GFX10-NEXT: v_writelane_b32 v40, s39, 5 +; GFX10-NEXT: v_writelane_b32 v40, s40, 6 +; GFX10-NEXT: v_writelane_b32 v40, s41, 7 +; GFX10-NEXT: v_writelane_b32 v40, s42, 8 +; GFX10-NEXT: v_writelane_b32 v40, s43, 9 +; GFX10-NEXT: v_writelane_b32 v40, s44, 10 +; GFX10-NEXT: v_writelane_b32 v40, s45, 11 +; GFX10-NEXT: v_writelane_b32 v40, s46, 12 +; GFX10-NEXT: v_writelane_b32 v40, s47, 13 +; GFX10-NEXT: v_writelane_b32 v40, s48, 14 +; GFX10-NEXT: v_writelane_b32 v40, s49, 15 +; GFX10-NEXT: v_writelane_b32 v40, s50, 16 +; GFX10-NEXT: v_writelane_b32 v40, s51, 17 +; GFX10-NEXT: v_writelane_b32 v40, s52, 18 +; GFX10-NEXT: v_writelane_b32 v40, s53, 19 +; GFX10-NEXT: v_writelane_b32 v40, s54, 20 +; GFX10-NEXT: v_writelane_b32 v40, s55, 21 +; GFX10-NEXT: v_writelane_b32 v40, s56, 22 +; GFX10-NEXT: v_writelane_b32 v40, s57, 23 +; GFX10-NEXT: v_writelane_b32 v40, s58, 24 +; GFX10-NEXT: v_writelane_b32 v40, s59, 25 +; GFX10-NEXT: v_writelane_b32 v40, s60, 26 +; GFX10-NEXT: v_writelane_b32 v40, s61, 27 +; GFX10-NEXT: v_writelane_b32 v40, s62, 28 +; GFX10-NEXT: v_writelane_b32 v40, s63, 29 +; GFX10-NEXT: v_writelane_b32 v40, s30, 30 +; GFX10-NEXT: v_writelane_b32 v40, s31, 31 ; GFX10-NEXT: s_clause 0x2 ; GFX10-NEXT: buffer_load_dword v32, off, s[0:3], s33 offset:20 ; GFX10-NEXT: buffer_load_dword v33, off, s[0:3], s33 offset:16 ; GFX10-NEXT: buffer_load_dword v31, off, s[0:3], s33 -; GFX10-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-NEXT: s_addk_i32 s32, 0x400 ; GFX10-NEXT: s_getpc_b64 s[4:5] ; GFX10-NEXT: s_add_u32 s4, s4, byval_align16_f64_arg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s5, s5, byval_align16_f64_arg@rel32@hi+12 @@ -6416,70 +6454,39 @@ ; GFX10-NEXT: buffer_store_dword v32, off, s[0:3], s32 offset:4 ; GFX10-NEXT: s_waitcnt vmcnt(1) ; GFX10-NEXT: buffer_store_dword v33, off, s[0:3], s32 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 -; GFX10-NEXT: v_writelane_b32 v40, s34, 2 -; GFX10-NEXT: v_writelane_b32 v40, s35, 3 -; GFX10-NEXT: v_writelane_b32 v40, s36, 4 -; GFX10-NEXT: v_writelane_b32 v40, s37, 5 -; GFX10-NEXT: v_writelane_b32 v40, s38, 6 -; GFX10-NEXT: v_writelane_b32 v40, s39, 7 -; GFX10-NEXT: v_writelane_b32 v40, s40, 8 -; GFX10-NEXT: v_writelane_b32 v40, s41, 9 -; GFX10-NEXT: v_writelane_b32 v40, s42, 10 -; GFX10-NEXT: v_writelane_b32 v40, s43, 11 -; GFX10-NEXT: v_writelane_b32 v40, s44, 12 -; GFX10-NEXT: v_writelane_b32 v40, s45, 13 -; GFX10-NEXT: v_writelane_b32 v40, s46, 14 -; GFX10-NEXT: v_writelane_b32 v40, s47, 15 -; GFX10-NEXT: v_writelane_b32 v40, s48, 16 -; GFX10-NEXT: v_writelane_b32 v40, s49, 17 -; GFX10-NEXT: v_writelane_b32 v40, s50, 18 -; GFX10-NEXT: v_writelane_b32 v40, s51, 19 -; GFX10-NEXT: v_writelane_b32 v40, s52, 20 -; GFX10-NEXT: v_writelane_b32 v40, s53, 21 -; GFX10-NEXT: v_writelane_b32 v40, s54, 22 -; GFX10-NEXT: v_writelane_b32 v40, s55, 23 -; GFX10-NEXT: v_writelane_b32 v40, s56, 24 -; GFX10-NEXT: v_writelane_b32 v40, s57, 25 -; GFX10-NEXT: v_writelane_b32 v40, s58, 26 -; GFX10-NEXT: v_writelane_b32 v40, s59, 27 -; GFX10-NEXT: v_writelane_b32 v40, s60, 28 -; GFX10-NEXT: v_writelane_b32 v40, s61, 29 -; GFX10-NEXT: v_writelane_b32 v40, s62, 30 -; GFX10-NEXT: v_writelane_b32 v40, s63, 31 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[4:5] -; GFX10-NEXT: v_readlane_b32 s63, v40, 31 -; GFX10-NEXT: v_readlane_b32 s62, v40, 30 -; GFX10-NEXT: v_readlane_b32 s61, v40, 29 -; GFX10-NEXT: v_readlane_b32 s60, v40, 28 -; GFX10-NEXT: v_readlane_b32 s59, v40, 27 -; GFX10-NEXT: v_readlane_b32 s58, v40, 26 -; GFX10-NEXT: v_readlane_b32 s57, v40, 25 -; GFX10-NEXT: v_readlane_b32 s56, v40, 24 -; GFX10-NEXT: v_readlane_b32 s55, v40, 23 -; GFX10-NEXT: v_readlane_b32 s54, v40, 22 -; GFX10-NEXT: v_readlane_b32 s53, v40, 21 -; GFX10-NEXT: v_readlane_b32 s52, v40, 20 -; GFX10-NEXT: v_readlane_b32 s51, v40, 19 -; GFX10-NEXT: v_readlane_b32 s50, v40, 18 -; GFX10-NEXT: v_readlane_b32 s49, v40, 17 -; GFX10-NEXT: v_readlane_b32 s48, v40, 16 -; GFX10-NEXT: v_readlane_b32 s47, v40, 15 -; GFX10-NEXT: v_readlane_b32 s46, v40, 14 -; GFX10-NEXT: v_readlane_b32 s45, v40, 13 -; GFX10-NEXT: v_readlane_b32 s44, v40, 12 -; GFX10-NEXT: v_readlane_b32 s43, v40, 11 -; GFX10-NEXT: v_readlane_b32 s42, v40, 10 -; GFX10-NEXT: v_readlane_b32 s41, v40, 9 -; GFX10-NEXT: v_readlane_b32 s40, v40, 8 -; GFX10-NEXT: v_readlane_b32 s39, v40, 7 -; GFX10-NEXT: v_readlane_b32 s38, v40, 6 -; GFX10-NEXT: v_readlane_b32 s37, v40, 5 -; GFX10-NEXT: v_readlane_b32 s36, v40, 4 -; GFX10-NEXT: v_readlane_b32 s35, v40, 3 -; GFX10-NEXT: v_readlane_b32 s34, v40, 2 -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 -; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s30, v40, 30 +; GFX10-NEXT: v_readlane_b32 s31, v40, 31 +; GFX10-NEXT: v_readlane_b32 s63, v40, 29 +; GFX10-NEXT: v_readlane_b32 s62, v40, 28 +; GFX10-NEXT: v_readlane_b32 s61, v40, 27 +; GFX10-NEXT: v_readlane_b32 s60, v40, 26 +; GFX10-NEXT: v_readlane_b32 s59, v40, 25 +; GFX10-NEXT: v_readlane_b32 s58, v40, 24 +; GFX10-NEXT: v_readlane_b32 s57, v40, 23 +; GFX10-NEXT: v_readlane_b32 s56, v40, 22 +; GFX10-NEXT: v_readlane_b32 s55, v40, 21 +; GFX10-NEXT: v_readlane_b32 s54, v40, 20 +; GFX10-NEXT: v_readlane_b32 s53, v40, 19 +; GFX10-NEXT: v_readlane_b32 s52, v40, 18 +; GFX10-NEXT: v_readlane_b32 s51, v40, 17 +; GFX10-NEXT: v_readlane_b32 s50, v40, 16 +; GFX10-NEXT: v_readlane_b32 s49, v40, 15 +; GFX10-NEXT: v_readlane_b32 s48, v40, 14 +; GFX10-NEXT: v_readlane_b32 s47, v40, 13 +; GFX10-NEXT: v_readlane_b32 s46, v40, 12 +; GFX10-NEXT: v_readlane_b32 s45, v40, 11 +; GFX10-NEXT: v_readlane_b32 s44, v40, 10 +; GFX10-NEXT: v_readlane_b32 s43, v40, 9 +; GFX10-NEXT: v_readlane_b32 s42, v40, 8 +; GFX10-NEXT: v_readlane_b32 s41, v40, 7 +; GFX10-NEXT: v_readlane_b32 s40, v40, 6 +; GFX10-NEXT: v_readlane_b32 s39, v40, 5 +; GFX10-NEXT: v_readlane_b32 s38, v40, 4 +; GFX10-NEXT: v_readlane_b32 s37, v40, 3 +; GFX10-NEXT: v_readlane_b32 s36, v40, 2 +; GFX10-NEXT: v_readlane_b32 s35, v40, 1 +; GFX10-NEXT: v_readlane_b32 s34, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfc00 ; GFX10-NEXT: s_mov_b32 s33, s6 ; GFX10-NEXT: s_or_saveexec_b32 s4, -1 @@ -6498,80 +6505,80 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: s_mov_b32 s4, s33 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 32 +; GFX11-NEXT: v_writelane_b32 v40, s34, 0 +; GFX11-NEXT: v_writelane_b32 v40, s35, 1 +; GFX11-NEXT: v_writelane_b32 v40, s36, 2 +; GFX11-NEXT: v_writelane_b32 v40, s37, 3 +; GFX11-NEXT: v_writelane_b32 v40, s38, 4 +; GFX11-NEXT: v_writelane_b32 v40, s39, 5 +; GFX11-NEXT: v_writelane_b32 v40, s40, 6 +; GFX11-NEXT: v_writelane_b32 v40, s41, 7 +; GFX11-NEXT: v_writelane_b32 v40, s42, 8 +; GFX11-NEXT: v_writelane_b32 v40, s43, 9 +; GFX11-NEXT: v_writelane_b32 v40, s44, 10 +; GFX11-NEXT: v_writelane_b32 v40, s45, 11 +; GFX11-NEXT: v_writelane_b32 v40, s46, 12 +; GFX11-NEXT: v_writelane_b32 v40, s47, 13 +; GFX11-NEXT: v_writelane_b32 v40, s48, 14 +; GFX11-NEXT: v_writelane_b32 v40, s49, 15 +; GFX11-NEXT: v_writelane_b32 v40, s50, 16 +; GFX11-NEXT: v_writelane_b32 v40, s51, 17 +; GFX11-NEXT: v_writelane_b32 v40, s52, 18 +; GFX11-NEXT: v_writelane_b32 v40, s53, 19 +; GFX11-NEXT: v_writelane_b32 v40, s54, 20 +; GFX11-NEXT: v_writelane_b32 v40, s55, 21 +; GFX11-NEXT: v_writelane_b32 v40, s56, 22 +; GFX11-NEXT: v_writelane_b32 v40, s57, 23 +; GFX11-NEXT: v_writelane_b32 v40, s58, 24 +; GFX11-NEXT: v_writelane_b32 v40, s59, 25 +; GFX11-NEXT: v_writelane_b32 v40, s60, 26 +; GFX11-NEXT: v_writelane_b32 v40, s61, 27 +; GFX11-NEXT: v_writelane_b32 v40, s62, 28 +; GFX11-NEXT: v_writelane_b32 v40, s63, 29 +; GFX11-NEXT: v_writelane_b32 v40, s30, 30 +; GFX11-NEXT: v_writelane_b32 v40, s31, 31 ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: scratch_load_b64 v[32:33], off, s33 offset:16 ; GFX11-NEXT: scratch_load_b32 v31, off, s33 -; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: s_add_i32 s32, s32, 32 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, byval_align16_f64_arg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, byval_align16_f64_arg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 -; GFX11-NEXT: v_writelane_b32 v40, s34, 2 -; GFX11-NEXT: v_writelane_b32 v40, s35, 3 -; GFX11-NEXT: v_writelane_b32 v40, s36, 4 -; GFX11-NEXT: v_writelane_b32 v40, s37, 5 -; GFX11-NEXT: v_writelane_b32 v40, s38, 6 -; GFX11-NEXT: v_writelane_b32 v40, s39, 7 -; GFX11-NEXT: v_writelane_b32 v40, s40, 8 -; GFX11-NEXT: v_writelane_b32 v40, s41, 9 -; GFX11-NEXT: v_writelane_b32 v40, s42, 10 -; GFX11-NEXT: v_writelane_b32 v40, s43, 11 -; GFX11-NEXT: v_writelane_b32 v40, s44, 12 -; GFX11-NEXT: v_writelane_b32 v40, s45, 13 -; GFX11-NEXT: v_writelane_b32 v40, s46, 14 -; GFX11-NEXT: v_writelane_b32 v40, s47, 15 -; GFX11-NEXT: v_writelane_b32 v40, s48, 16 -; GFX11-NEXT: v_writelane_b32 v40, s49, 17 -; GFX11-NEXT: v_writelane_b32 v40, s50, 18 -; GFX11-NEXT: v_writelane_b32 v40, s51, 19 -; GFX11-NEXT: v_writelane_b32 v40, s52, 20 -; GFX11-NEXT: v_writelane_b32 v40, s53, 21 -; GFX11-NEXT: v_writelane_b32 v40, s54, 22 -; GFX11-NEXT: v_writelane_b32 v40, s55, 23 -; GFX11-NEXT: v_writelane_b32 v40, s56, 24 -; GFX11-NEXT: v_writelane_b32 v40, s57, 25 -; GFX11-NEXT: v_writelane_b32 v40, s58, 26 -; GFX11-NEXT: v_writelane_b32 v40, s59, 27 -; GFX11-NEXT: v_writelane_b32 v40, s60, 28 -; GFX11-NEXT: v_writelane_b32 v40, s61, 29 -; GFX11-NEXT: v_writelane_b32 v40, s62, 30 -; GFX11-NEXT: v_writelane_b32 v40, s63, 31 ; GFX11-NEXT: s_waitcnt vmcnt(1) ; GFX11-NEXT: scratch_store_b64 off, v[32:33], s32 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s63, v40, 31 -; GFX11-NEXT: v_readlane_b32 s62, v40, 30 -; GFX11-NEXT: v_readlane_b32 s61, v40, 29 -; GFX11-NEXT: v_readlane_b32 s60, v40, 28 -; GFX11-NEXT: v_readlane_b32 s59, v40, 27 -; GFX11-NEXT: v_readlane_b32 s58, v40, 26 -; GFX11-NEXT: v_readlane_b32 s57, v40, 25 -; GFX11-NEXT: v_readlane_b32 s56, v40, 24 -; GFX11-NEXT: v_readlane_b32 s55, v40, 23 -; GFX11-NEXT: v_readlane_b32 s54, v40, 22 -; GFX11-NEXT: v_readlane_b32 s53, v40, 21 -; GFX11-NEXT: v_readlane_b32 s52, v40, 20 -; GFX11-NEXT: v_readlane_b32 s51, v40, 19 -; GFX11-NEXT: v_readlane_b32 s50, v40, 18 -; GFX11-NEXT: v_readlane_b32 s49, v40, 17 -; GFX11-NEXT: v_readlane_b32 s48, v40, 16 -; GFX11-NEXT: v_readlane_b32 s47, v40, 15 -; GFX11-NEXT: v_readlane_b32 s46, v40, 14 -; GFX11-NEXT: v_readlane_b32 s45, v40, 13 -; GFX11-NEXT: v_readlane_b32 s44, v40, 12 -; GFX11-NEXT: v_readlane_b32 s43, v40, 11 -; GFX11-NEXT: v_readlane_b32 s42, v40, 10 -; GFX11-NEXT: v_readlane_b32 s41, v40, 9 -; GFX11-NEXT: v_readlane_b32 s40, v40, 8 -; GFX11-NEXT: v_readlane_b32 s39, v40, 7 -; GFX11-NEXT: v_readlane_b32 s38, v40, 6 -; GFX11-NEXT: v_readlane_b32 s37, v40, 5 -; GFX11-NEXT: v_readlane_b32 s36, v40, 4 -; GFX11-NEXT: v_readlane_b32 s35, v40, 3 -; GFX11-NEXT: v_readlane_b32 s34, v40, 2 -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 -; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s30, v40, 30 +; GFX11-NEXT: v_readlane_b32 s31, v40, 31 +; GFX11-NEXT: v_readlane_b32 s63, v40, 29 +; GFX11-NEXT: v_readlane_b32 s62, v40, 28 +; GFX11-NEXT: v_readlane_b32 s61, v40, 27 +; GFX11-NEXT: v_readlane_b32 s60, v40, 26 +; GFX11-NEXT: v_readlane_b32 s59, v40, 25 +; GFX11-NEXT: v_readlane_b32 s58, v40, 24 +; GFX11-NEXT: v_readlane_b32 s57, v40, 23 +; GFX11-NEXT: v_readlane_b32 s56, v40, 22 +; GFX11-NEXT: v_readlane_b32 s55, v40, 21 +; GFX11-NEXT: v_readlane_b32 s54, v40, 20 +; GFX11-NEXT: v_readlane_b32 s53, v40, 19 +; GFX11-NEXT: v_readlane_b32 s52, v40, 18 +; GFX11-NEXT: v_readlane_b32 s51, v40, 17 +; GFX11-NEXT: v_readlane_b32 s50, v40, 16 +; GFX11-NEXT: v_readlane_b32 s49, v40, 15 +; GFX11-NEXT: v_readlane_b32 s48, v40, 14 +; GFX11-NEXT: v_readlane_b32 s47, v40, 13 +; GFX11-NEXT: v_readlane_b32 s46, v40, 12 +; GFX11-NEXT: v_readlane_b32 s45, v40, 11 +; GFX11-NEXT: v_readlane_b32 s44, v40, 10 +; GFX11-NEXT: v_readlane_b32 s43, v40, 9 +; GFX11-NEXT: v_readlane_b32 s42, v40, 8 +; GFX11-NEXT: v_readlane_b32 s41, v40, 7 +; GFX11-NEXT: v_readlane_b32 s40, v40, 6 +; GFX11-NEXT: v_readlane_b32 s39, v40, 5 +; GFX11-NEXT: v_readlane_b32 s38, v40, 4 +; GFX11-NEXT: v_readlane_b32 s37, v40, 3 +; GFX11-NEXT: v_readlane_b32 s36, v40, 2 +; GFX11-NEXT: v_readlane_b32 s35, v40, 1 +; GFX11-NEXT: v_readlane_b32 s34, v40, 0 ; GFX11-NEXT: s_addk_i32 s32, 0xffe0 ; GFX11-NEXT: s_mov_b32 s33, s4 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -6590,80 +6597,80 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: s_mov_b32 s4, s33 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 32 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s34, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s35, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s36, 2 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s37, 3 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s38, 4 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s39, 5 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s40, 6 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s41, 7 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s42, 8 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s43, 9 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s44, 10 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s45, 11 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s46, 12 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s47, 13 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s48, 14 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s49, 15 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s50, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s51, 17 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s52, 18 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s53, 19 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s54, 20 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s55, 21 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s56, 22 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s57, 23 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s58, 24 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s59, 25 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s60, 26 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s61, 27 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s62, 28 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s63, 29 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 30 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 31 ; GFX10-SCRATCH-NEXT: s_clause 0x1 ; GFX10-SCRATCH-NEXT: scratch_load_dwordx2 v[32:33], off, s33 offset:16 ; GFX10-SCRATCH-NEXT: scratch_load_dword v31, off, s33 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 32 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, byval_align16_f64_arg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, byval_align16_f64_arg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s34, 2 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s35, 3 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s36, 4 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s37, 5 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s38, 6 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s39, 7 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s40, 8 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s41, 9 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s42, 10 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s43, 11 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s44, 12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s45, 13 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s46, 14 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s47, 15 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s48, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s49, 17 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s50, 18 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s51, 19 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s52, 20 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s53, 21 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s54, 22 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s55, 23 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s56, 24 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s57, 25 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s58, 26 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s59, 27 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s60, 28 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s61, 29 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s62, 30 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s63, 31 ; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(1) ; GFX10-SCRATCH-NEXT: scratch_store_dwordx2 off, v[32:33], s32 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s63, v40, 31 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s62, v40, 30 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s61, v40, 29 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s60, v40, 28 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s59, v40, 27 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s58, v40, 26 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s57, v40, 25 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s56, v40, 24 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s55, v40, 23 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s54, v40, 22 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s53, v40, 21 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s52, v40, 20 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s51, v40, 19 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s50, v40, 18 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s49, v40, 17 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s48, v40, 16 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s47, v40, 15 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s46, v40, 14 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s45, v40, 13 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s44, v40, 12 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s43, v40, 11 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s42, v40, 10 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s41, v40, 9 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s40, v40, 8 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s39, v40, 7 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s38, v40, 6 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s37, v40, 5 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s36, v40, 4 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s35, v40, 3 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s34, v40, 2 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 -; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 30 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 31 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s63, v40, 29 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s62, v40, 28 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s61, v40, 27 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s60, v40, 26 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s59, v40, 25 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s58, v40, 24 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s57, v40, 23 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s56, v40, 22 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s55, v40, 21 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s54, v40, 20 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s53, v40, 19 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s52, v40, 18 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s51, v40, 17 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s50, v40, 16 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s49, v40, 15 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s48, v40, 14 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s47, v40, 13 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s46, v40, 12 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s45, v40, 11 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s44, v40, 10 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s43, v40, 9 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s42, v40, 8 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s41, v40, 7 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s40, v40, 6 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s39, v40, 5 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s38, v40, 4 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s37, v40, 3 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s36, v40, 2 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s35, v40, 1 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s34, v40, 0 ; GFX10-SCRATCH-NEXT: s_addk_i32 s32, 0xffe0 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s4 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -6688,17 +6695,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 -; GFX9-NEXT: v_mov_b32_e32 v0, 1 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: v_mov_b32_e32 v0, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i1_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i1_inreg@rel32@hi+12 ; GFX9-NEXT: buffer_store_byte v0, off, s[0:3], s32 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -6718,16 +6725,16 @@ ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i1_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i1_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: buffer_store_byte v0, off, s[0:3], s32 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -6747,16 +6754,16 @@ ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 -; GFX11-NEXT: v_mov_b32_e32 v0, 1 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_mov_b32_e32 v0, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i1_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i1_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: scratch_store_b8 off, v0, s32 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -6776,16 +6783,16 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i1_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i1_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: scratch_store_byte off, v0, s32 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -6807,17 +6814,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 3 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s30, 1 -; GFX9-NEXT: s_movk_i32 s4, 0x7b ; GFX9-NEXT: v_writelane_b32 v40, s31, 2 +; GFX9-NEXT: s_movk_i32 s4, 0x7b ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i8_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i8_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s30, v40, 1 +; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 3 @@ -6837,17 +6844,17 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_movk_i32 s4, 0x7b -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i8_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i8_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 3 @@ -6867,18 +6874,18 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 3 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s30, 1 +; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: s_movk_i32 s4, 0x7b -; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i8_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i8_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 1 -; GFX11-NEXT: v_writelane_b32 v40, s31, 2 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s30, v40, 1 +; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 3 @@ -6898,17 +6905,17 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-SCRATCH-NEXT: s_movk_i32 s4, 0x7b -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i8_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i8_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 3 @@ -6931,17 +6938,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 3 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s30, 1 -; GFX9-NEXT: s_movk_i32 s4, 0x7b ; GFX9-NEXT: v_writelane_b32 v40, s31, 2 +; GFX9-NEXT: s_movk_i32 s4, 0x7b ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i16_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i16_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s30, v40, 1 +; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 3 @@ -6961,17 +6968,17 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_movk_i32 s4, 0x7b -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i16_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i16_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 3 @@ -6991,18 +6998,18 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 3 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s30, 1 +; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: s_movk_i32 s4, 0x7b -; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i16_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i16_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 1 -; GFX11-NEXT: v_writelane_b32 v40, s31, 2 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s30, v40, 1 +; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 3 @@ -7022,17 +7029,17 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-SCRATCH-NEXT: s_movk_i32 s4, 0x7b -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i16_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i16_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 3 @@ -7055,17 +7062,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 3 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s30, 1 -; GFX9-NEXT: s_mov_b32 s4, 42 ; GFX9-NEXT: v_writelane_b32 v40, s31, 2 +; GFX9-NEXT: s_mov_b32 s4, 42 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i32_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s30, v40, 1 +; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 3 @@ -7085,17 +7092,17 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_mov_b32 s4, 42 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i32_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i32_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 3 @@ -7115,18 +7122,18 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 3 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s30, 1 +; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: s_mov_b32 s4, 42 -; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i32_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i32_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 1 -; GFX11-NEXT: v_writelane_b32 v40, s31, 2 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s30, v40, 1 +; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 3 @@ -7146,17 +7153,17 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 42 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i32_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i32_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 3 @@ -7179,19 +7186,19 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 4 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 2 +; GFX9-NEXT: v_writelane_b32 v40, s31, 3 ; GFX9-NEXT: s_movk_i32 s4, 0x7b ; GFX9-NEXT: s_mov_b32 s5, 0 -; GFX9-NEXT: v_writelane_b32 v40, s31, 3 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_i64_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_i64_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s30, v40, 2 +; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 @@ -7212,19 +7219,19 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_movk_i32 s4, 0x7b -; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: s_mov_b32 s5, 0 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_i64_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_i64_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 0 -; GFX10-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 @@ -7245,20 +7252,20 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 4 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s5, 1 +; GFX11-NEXT: v_writelane_b32 v40, s30, 2 +; GFX11-NEXT: v_writelane_b32 v40, s31, 3 ; GFX11-NEXT: s_movk_i32 s4, 0x7b -; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: s_mov_b32 s5, 0 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_i64_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_i64_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 0 -; GFX11-NEXT: v_writelane_b32 v40, s30, 2 -; GFX11-NEXT: v_writelane_b32 v40, s31, 3 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s30, v40, 2 +; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 @@ -7279,19 +7286,19 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_movk_i32 s4, 0x7b -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 0 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_i64_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_i64_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 0 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 @@ -7315,21 +7322,21 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 6 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 -; GFX9-NEXT: s_mov_b64 s[34:35], 0 ; GFX9-NEXT: v_writelane_b32 v40, s7, 3 -; GFX9-NEXT: s_load_dwordx4 s[4:7], s[34:35], 0x0 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 4 ; GFX9-NEXT: v_writelane_b32 v40, s31, 5 +; GFX9-NEXT: s_mov_b64 s[34:35], 0 +; GFX9-NEXT: s_load_dwordx4 s[4:7], s[34:35], 0x0 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2i64_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2i64_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 5 ; GFX9-NEXT: v_readlane_b32 s30, v40, 4 +; GFX9-NEXT: v_readlane_b32 s31, v40, 5 ; GFX9-NEXT: v_readlane_b32 s7, v40, 3 ; GFX9-NEXT: v_readlane_b32 s6, v40, 2 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 @@ -7352,21 +7359,21 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 6 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_mov_b64 s[34:35], 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 +; GFX10-NEXT: v_writelane_b32 v40, s30, 4 +; GFX10-NEXT: v_writelane_b32 v40, s31, 5 +; GFX10-NEXT: s_mov_b64 s[34:35], 0 ; GFX10-NEXT: s_load_dwordx4 s[4:7], s[34:35], 0x0 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2i64_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2i64_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 4 -; GFX10-NEXT: v_writelane_b32 v40, s31, 5 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-NEXT: v_readlane_b32 s30, v40, 4 +; GFX10-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 @@ -7389,22 +7396,22 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 6 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_mov_b64 s[0:1], 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 +; GFX11-NEXT: v_writelane_b32 v40, s30, 4 +; GFX11-NEXT: v_writelane_b32 v40, s31, 5 +; GFX11-NEXT: s_mov_b64 s[0:1], 0 ; GFX11-NEXT: s_load_b128 s[4:7], s[0:1], 0x0 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2i64_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2i64_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 4 -; GFX11-NEXT: v_writelane_b32 v40, s31, 5 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 5 ; GFX11-NEXT: v_readlane_b32 s30, v40, 4 +; GFX11-NEXT: v_readlane_b32 s31, v40, 5 ; GFX11-NEXT: v_readlane_b32 s7, v40, 3 ; GFX11-NEXT: v_readlane_b32 s6, v40, 2 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 @@ -7427,21 +7434,21 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 6 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_mov_b64 s[0:1], 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 4 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 5 +; GFX10-SCRATCH-NEXT: s_mov_b64 s[0:1], 0 ; GFX10-SCRATCH-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2i64_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2i64_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 4 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 5 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 4 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 @@ -7468,23 +7475,23 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 6 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 ; GFX9-NEXT: v_writelane_b32 v40, s7, 3 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 4 +; GFX9-NEXT: v_writelane_b32 v40, s31, 5 ; GFX9-NEXT: s_mov_b32 s4, 1 ; GFX9-NEXT: s_mov_b32 s5, 2 ; GFX9-NEXT: s_mov_b32 s6, 3 ; GFX9-NEXT: s_mov_b32 s7, 4 -; GFX9-NEXT: v_writelane_b32 v40, s31, 5 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2i64_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2i64_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 5 ; GFX9-NEXT: v_readlane_b32 s30, v40, 4 +; GFX9-NEXT: v_readlane_b32 s31, v40, 5 ; GFX9-NEXT: v_readlane_b32 s7, v40, 3 ; GFX9-NEXT: v_readlane_b32 s6, v40, 2 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 @@ -7507,23 +7514,23 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 6 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_mov_b32 s4, 1 ; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: s_getpc_b64 s[34:35] -; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2i64_inreg@rel32@lo+4 -; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2i64_inreg@rel32@hi+12 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 2 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-NEXT: s_mov_b32 s6, 3 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-NEXT: s_mov_b32 s7, 4 ; GFX10-NEXT: v_writelane_b32 v40, s30, 4 ; GFX10-NEXT: v_writelane_b32 v40, s31, 5 +; GFX10-NEXT: s_mov_b32 s4, 1 +; GFX10-NEXT: s_mov_b32 s5, 2 +; GFX10-NEXT: s_mov_b32 s6, 3 +; GFX10-NEXT: s_mov_b32 s7, 4 +; GFX10-NEXT: s_getpc_b64 s[34:35] +; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2i64_inreg@rel32@lo+4 +; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2i64_inreg@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-NEXT: v_readlane_b32 s30, v40, 4 +; GFX10-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 @@ -7546,24 +7553,24 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 6 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_mov_b32 s4, 1 ; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: s_getpc_b64 s[0:1] -; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2i64_inreg@rel32@lo+4 -; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2i64_inreg@rel32@hi+12 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 2 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 -; GFX11-NEXT: s_mov_b32 s6, 3 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 -; GFX11-NEXT: s_mov_b32 s7, 4 ; GFX11-NEXT: v_writelane_b32 v40, s30, 4 ; GFX11-NEXT: v_writelane_b32 v40, s31, 5 +; GFX11-NEXT: s_mov_b32 s4, 1 +; GFX11-NEXT: s_mov_b32 s5, 2 +; GFX11-NEXT: s_mov_b32 s6, 3 +; GFX11-NEXT: s_mov_b32 s7, 4 +; GFX11-NEXT: s_getpc_b64 s[0:1] +; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2i64_inreg@rel32@lo+4 +; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2i64_inreg@rel32@hi+12 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 5 ; GFX11-NEXT: v_readlane_b32 s30, v40, 4 +; GFX11-NEXT: v_readlane_b32 s31, v40, 5 ; GFX11-NEXT: v_readlane_b32 s7, v40, 3 ; GFX11-NEXT: v_readlane_b32 s6, v40, 2 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 @@ -7586,23 +7593,23 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 6 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] -; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2i64_inreg@rel32@lo+4 -; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2i64_inreg@rel32@hi+12 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 3 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-SCRATCH-NEXT: s_mov_b32 s7, 4 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 4 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 5 +; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 1 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2 +; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 3 +; GFX10-SCRATCH-NEXT: s_mov_b32 s7, 4 +; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] +; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2i64_inreg@rel32@lo+4 +; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2i64_inreg@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 4 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 @@ -7628,25 +7635,25 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 8 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 -; GFX9-NEXT: s_mov_b64 s[34:35], 0 ; GFX9-NEXT: v_writelane_b32 v40, s7, 3 -; GFX9-NEXT: s_load_dwordx4 s[4:7], s[34:35], 0x0 ; GFX9-NEXT: v_writelane_b32 v40, s8, 4 ; GFX9-NEXT: v_writelane_b32 v40, s9, 5 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 6 +; GFX9-NEXT: v_writelane_b32 v40, s31, 7 +; GFX9-NEXT: s_mov_b64 s[34:35], 0 +; GFX9-NEXT: s_load_dwordx4 s[4:7], s[34:35], 0x0 ; GFX9-NEXT: s_mov_b32 s8, 1 ; GFX9-NEXT: s_mov_b32 s9, 2 -; GFX9-NEXT: v_writelane_b32 v40, s31, 7 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3i64_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3i64_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 7 ; GFX9-NEXT: v_readlane_b32 s30, v40, 6 +; GFX9-NEXT: v_readlane_b32 s31, v40, 7 ; GFX9-NEXT: v_readlane_b32 s9, v40, 5 ; GFX9-NEXT: v_readlane_b32 s8, v40, 4 ; GFX9-NEXT: v_readlane_b32 s7, v40, 3 @@ -7671,25 +7678,25 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 8 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_mov_b64 s[34:35], 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-NEXT: s_load_dwordx4 s[4:7], s[34:35], 0x0 -; GFX10-NEXT: s_getpc_b64 s[34:35] -; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3i64_inreg@rel32@lo+4 -; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3i64_inreg@rel32@hi+12 ; GFX10-NEXT: v_writelane_b32 v40, s8, 4 -; GFX10-NEXT: s_mov_b32 s8, 1 ; GFX10-NEXT: v_writelane_b32 v40, s9, 5 -; GFX10-NEXT: s_mov_b32 s9, 2 ; GFX10-NEXT: v_writelane_b32 v40, s30, 6 ; GFX10-NEXT: v_writelane_b32 v40, s31, 7 +; GFX10-NEXT: s_mov_b64 s[34:35], 0 +; GFX10-NEXT: s_mov_b32 s8, 1 +; GFX10-NEXT: s_load_dwordx4 s[4:7], s[34:35], 0x0 +; GFX10-NEXT: s_mov_b32 s9, 2 +; GFX10-NEXT: s_getpc_b64 s[34:35] +; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3i64_inreg@rel32@lo+4 +; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3i64_inreg@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 7 ; GFX10-NEXT: v_readlane_b32 s30, v40, 6 +; GFX10-NEXT: v_readlane_b32 s31, v40, 7 ; GFX10-NEXT: v_readlane_b32 s9, v40, 5 ; GFX10-NEXT: v_readlane_b32 s8, v40, 4 ; GFX10-NEXT: v_readlane_b32 s7, v40, 3 @@ -7714,26 +7721,26 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 8 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_mov_b64 s[0:1], 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 -; GFX11-NEXT: s_load_b128 s[4:7], s[0:1], 0x0 -; GFX11-NEXT: s_getpc_b64 s[0:1] -; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3i64_inreg@rel32@lo+4 -; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3i64_inreg@rel32@hi+12 ; GFX11-NEXT: v_writelane_b32 v40, s8, 4 -; GFX11-NEXT: s_mov_b32 s8, 1 ; GFX11-NEXT: v_writelane_b32 v40, s9, 5 -; GFX11-NEXT: s_mov_b32 s9, 2 ; GFX11-NEXT: v_writelane_b32 v40, s30, 6 ; GFX11-NEXT: v_writelane_b32 v40, s31, 7 +; GFX11-NEXT: s_mov_b64 s[0:1], 0 +; GFX11-NEXT: s_mov_b32 s8, 1 +; GFX11-NEXT: s_load_b128 s[4:7], s[0:1], 0x0 +; GFX11-NEXT: s_mov_b32 s9, 2 +; GFX11-NEXT: s_getpc_b64 s[0:1] +; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3i64_inreg@rel32@lo+4 +; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3i64_inreg@rel32@hi+12 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 7 ; GFX11-NEXT: v_readlane_b32 s30, v40, 6 +; GFX11-NEXT: v_readlane_b32 s31, v40, 7 ; GFX11-NEXT: v_readlane_b32 s9, v40, 5 ; GFX11-NEXT: v_readlane_b32 s8, v40, 4 ; GFX11-NEXT: v_readlane_b32 s7, v40, 3 @@ -7758,25 +7765,25 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 8 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_mov_b64 s[0:1], 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-SCRATCH-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x0 -; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] -; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3i64_inreg@rel32@lo+4 -; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3i64_inreg@rel32@hi+12 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s8, 4 -; GFX10-SCRATCH-NEXT: s_mov_b32 s8, 1 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s9, 5 -; GFX10-SCRATCH-NEXT: s_mov_b32 s9, 2 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 6 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 7 +; GFX10-SCRATCH-NEXT: s_mov_b64 s[0:1], 0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s8, 1 +; GFX10-SCRATCH-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s9, 2 +; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] +; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3i64_inreg@rel32@lo+4 +; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3i64_inreg@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 7 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 6 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 7 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s9, v40, 5 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s8, v40, 4 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s7, v40, 3 @@ -7807,29 +7814,29 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 10 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 ; GFX9-NEXT: v_writelane_b32 v40, s7, 3 -; GFX9-NEXT: s_mov_b64 s[34:35], 0 ; GFX9-NEXT: v_writelane_b32 v40, s8, 4 -; GFX9-NEXT: s_load_dwordx4 s[4:7], s[34:35], 0x0 ; GFX9-NEXT: v_writelane_b32 v40, s9, 5 ; GFX9-NEXT: v_writelane_b32 v40, s10, 6 ; GFX9-NEXT: v_writelane_b32 v40, s11, 7 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 8 +; GFX9-NEXT: v_writelane_b32 v40, s31, 9 +; GFX9-NEXT: s_mov_b64 s[34:35], 0 +; GFX9-NEXT: s_load_dwordx4 s[4:7], s[34:35], 0x0 ; GFX9-NEXT: s_mov_b32 s8, 1 ; GFX9-NEXT: s_mov_b32 s9, 2 ; GFX9-NEXT: s_mov_b32 s10, 3 ; GFX9-NEXT: s_mov_b32 s11, 4 -; GFX9-NEXT: v_writelane_b32 v40, s31, 9 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v4i64_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v4i64_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 9 ; GFX9-NEXT: v_readlane_b32 s30, v40, 8 +; GFX9-NEXT: v_readlane_b32 s31, v40, 9 ; GFX9-NEXT: v_readlane_b32 s11, v40, 7 ; GFX9-NEXT: v_readlane_b32 s10, v40, 6 ; GFX9-NEXT: v_readlane_b32 s9, v40, 5 @@ -7856,29 +7863,29 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 10 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_mov_b64 s[34:35], 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-NEXT: s_load_dwordx4 s[4:7], s[34:35], 0x0 -; GFX10-NEXT: s_getpc_b64 s[34:35] -; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v4i64_inreg@rel32@lo+4 -; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v4i64_inreg@rel32@hi+12 ; GFX10-NEXT: v_writelane_b32 v40, s8, 4 -; GFX10-NEXT: s_mov_b32 s8, 1 ; GFX10-NEXT: v_writelane_b32 v40, s9, 5 -; GFX10-NEXT: s_mov_b32 s9, 2 ; GFX10-NEXT: v_writelane_b32 v40, s10, 6 -; GFX10-NEXT: s_mov_b32 s10, 3 ; GFX10-NEXT: v_writelane_b32 v40, s11, 7 -; GFX10-NEXT: s_mov_b32 s11, 4 ; GFX10-NEXT: v_writelane_b32 v40, s30, 8 ; GFX10-NEXT: v_writelane_b32 v40, s31, 9 +; GFX10-NEXT: s_mov_b64 s[34:35], 0 +; GFX10-NEXT: s_mov_b32 s8, 1 +; GFX10-NEXT: s_load_dwordx4 s[4:7], s[34:35], 0x0 +; GFX10-NEXT: s_mov_b32 s9, 2 +; GFX10-NEXT: s_mov_b32 s10, 3 +; GFX10-NEXT: s_mov_b32 s11, 4 +; GFX10-NEXT: s_getpc_b64 s[34:35] +; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v4i64_inreg@rel32@lo+4 +; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v4i64_inreg@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 9 ; GFX10-NEXT: v_readlane_b32 s30, v40, 8 +; GFX10-NEXT: v_readlane_b32 s31, v40, 9 ; GFX10-NEXT: v_readlane_b32 s11, v40, 7 ; GFX10-NEXT: v_readlane_b32 s10, v40, 6 ; GFX10-NEXT: v_readlane_b32 s9, v40, 5 @@ -7905,30 +7912,30 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 10 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_mov_b64 s[0:1], 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 -; GFX11-NEXT: s_load_b128 s[4:7], s[0:1], 0x0 -; GFX11-NEXT: s_getpc_b64 s[0:1] -; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v4i64_inreg@rel32@lo+4 -; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v4i64_inreg@rel32@hi+12 ; GFX11-NEXT: v_writelane_b32 v40, s8, 4 -; GFX11-NEXT: s_mov_b32 s8, 1 ; GFX11-NEXT: v_writelane_b32 v40, s9, 5 -; GFX11-NEXT: s_mov_b32 s9, 2 ; GFX11-NEXT: v_writelane_b32 v40, s10, 6 -; GFX11-NEXT: s_mov_b32 s10, 3 ; GFX11-NEXT: v_writelane_b32 v40, s11, 7 -; GFX11-NEXT: s_mov_b32 s11, 4 ; GFX11-NEXT: v_writelane_b32 v40, s30, 8 ; GFX11-NEXT: v_writelane_b32 v40, s31, 9 +; GFX11-NEXT: s_mov_b64 s[0:1], 0 +; GFX11-NEXT: s_mov_b32 s8, 1 +; GFX11-NEXT: s_load_b128 s[4:7], s[0:1], 0x0 +; GFX11-NEXT: s_mov_b32 s9, 2 +; GFX11-NEXT: s_mov_b32 s10, 3 +; GFX11-NEXT: s_mov_b32 s11, 4 +; GFX11-NEXT: s_getpc_b64 s[0:1] +; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v4i64_inreg@rel32@lo+4 +; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v4i64_inreg@rel32@hi+12 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 9 ; GFX11-NEXT: v_readlane_b32 s30, v40, 8 +; GFX11-NEXT: v_readlane_b32 s31, v40, 9 ; GFX11-NEXT: v_readlane_b32 s11, v40, 7 ; GFX11-NEXT: v_readlane_b32 s10, v40, 6 ; GFX11-NEXT: v_readlane_b32 s9, v40, 5 @@ -7955,29 +7962,29 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 10 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_mov_b64 s[0:1], 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-SCRATCH-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x0 -; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] -; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v4i64_inreg@rel32@lo+4 -; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v4i64_inreg@rel32@hi+12 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s8, 4 -; GFX10-SCRATCH-NEXT: s_mov_b32 s8, 1 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s9, 5 -; GFX10-SCRATCH-NEXT: s_mov_b32 s9, 2 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s10, 6 -; GFX10-SCRATCH-NEXT: s_mov_b32 s10, 3 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s11, 7 -; GFX10-SCRATCH-NEXT: s_mov_b32 s11, 4 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 8 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 9 +; GFX10-SCRATCH-NEXT: s_mov_b64 s[0:1], 0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s8, 1 +; GFX10-SCRATCH-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s9, 2 +; GFX10-SCRATCH-NEXT: s_mov_b32 s10, 3 +; GFX10-SCRATCH-NEXT: s_mov_b32 s11, 4 +; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] +; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v4i64_inreg@rel32@lo+4 +; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v4i64_inreg@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 9 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 8 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 9 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s11, v40, 7 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s10, v40, 6 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s9, v40, 5 @@ -8009,17 +8016,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 3 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s30, 1 -; GFX9-NEXT: s_movk_i32 s4, 0x4400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 2 +; GFX9-NEXT: s_movk_i32 s4, 0x4400 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_f16_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_f16_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s30, v40, 1 +; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 3 @@ -8039,17 +8046,17 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_movk_i32 s4, 0x4400 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_f16_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_f16_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 3 @@ -8069,18 +8076,18 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 3 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s30, 1 +; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: s_movk_i32 s4, 0x4400 -; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_f16_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_f16_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 1 -; GFX11-NEXT: v_writelane_b32 v40, s31, 2 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s30, v40, 1 +; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 3 @@ -8100,17 +8107,17 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-SCRATCH-NEXT: s_movk_i32 s4, 0x4400 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_f16_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_f16_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 3 @@ -8133,17 +8140,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 3 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s30, 1 -; GFX9-NEXT: s_mov_b32 s4, 4.0 ; GFX9-NEXT: v_writelane_b32 v40, s31, 2 +; GFX9-NEXT: s_mov_b32 s4, 4.0 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_f32_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_f32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s30, v40, 1 +; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 3 @@ -8163,17 +8170,17 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_mov_b32 s4, 4.0 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_f32_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_f32_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 3 @@ -8193,18 +8200,18 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 3 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s30, 1 +; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: s_mov_b32 s4, 4.0 -; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_f32_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_f32_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 1 -; GFX11-NEXT: v_writelane_b32 v40, s31, 2 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s30, v40, 1 +; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 3 @@ -8224,17 +8231,17 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 4.0 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_f32_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_f32_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 3 @@ -8257,19 +8264,19 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 4 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 2 +; GFX9-NEXT: v_writelane_b32 v40, s31, 3 ; GFX9-NEXT: s_mov_b32 s4, 1.0 ; GFX9-NEXT: s_mov_b32 s5, 2.0 -; GFX9-NEXT: v_writelane_b32 v40, s31, 3 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2f32_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2f32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s30, v40, 2 +; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 @@ -8290,19 +8297,19 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_mov_b32 s4, 1.0 -; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: s_mov_b32 s5, 2.0 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2f32_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2f32_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 2.0 -; GFX10-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 @@ -8323,20 +8330,20 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 4 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s5, 1 +; GFX11-NEXT: v_writelane_b32 v40, s30, 2 +; GFX11-NEXT: v_writelane_b32 v40, s31, 3 ; GFX11-NEXT: s_mov_b32 s4, 1.0 -; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: s_mov_b32 s5, 2.0 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2f32_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2f32_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 2.0 -; GFX11-NEXT: v_writelane_b32 v40, s30, 2 -; GFX11-NEXT: v_writelane_b32 v40, s31, 3 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s30, v40, 2 +; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 @@ -8357,19 +8364,19 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 1.0 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2.0 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2f32_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2f32_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2.0 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 @@ -8393,21 +8400,21 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 5 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 3 +; GFX9-NEXT: v_writelane_b32 v40, s31, 4 ; GFX9-NEXT: s_mov_b32 s4, 1.0 ; GFX9-NEXT: s_mov_b32 s5, 2.0 ; GFX9-NEXT: s_mov_b32 s6, 4.0 -; GFX9-NEXT: v_writelane_b32 v40, s31, 4 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3f32_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3f32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 4 ; GFX9-NEXT: v_readlane_b32 s30, v40, 3 +; GFX9-NEXT: v_readlane_b32 s31, v40, 4 ; GFX9-NEXT: v_readlane_b32 s6, v40, 2 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 @@ -8429,21 +8436,21 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 5 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_mov_b32 s4, 1.0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: s_getpc_b64 s[34:35] -; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3f32_inreg@rel32@lo+4 -; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3f32_inreg@rel32@hi+12 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 2.0 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-NEXT: s_mov_b32 s6, 4.0 ; GFX10-NEXT: v_writelane_b32 v40, s30, 3 ; GFX10-NEXT: v_writelane_b32 v40, s31, 4 +; GFX10-NEXT: s_mov_b32 s4, 1.0 +; GFX10-NEXT: s_mov_b32 s5, 2.0 +; GFX10-NEXT: s_mov_b32 s6, 4.0 +; GFX10-NEXT: s_getpc_b64 s[34:35] +; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3f32_inreg@rel32@lo+4 +; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3f32_inreg@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 4 ; GFX10-NEXT: v_readlane_b32 s30, v40, 3 +; GFX10-NEXT: v_readlane_b32 s31, v40, 4 ; GFX10-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 @@ -8465,22 +8472,22 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 5 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_mov_b32 s4, 1.0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: s_getpc_b64 s[0:1] -; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3f32_inreg@rel32@lo+4 -; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3f32_inreg@rel32@hi+12 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 2.0 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 -; GFX11-NEXT: s_mov_b32 s6, 4.0 ; GFX11-NEXT: v_writelane_b32 v40, s30, 3 ; GFX11-NEXT: v_writelane_b32 v40, s31, 4 +; GFX11-NEXT: s_mov_b32 s4, 1.0 +; GFX11-NEXT: s_mov_b32 s5, 2.0 +; GFX11-NEXT: s_mov_b32 s6, 4.0 +; GFX11-NEXT: s_getpc_b64 s[0:1] +; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3f32_inreg@rel32@lo+4 +; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3f32_inreg@rel32@hi+12 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 4 ; GFX11-NEXT: v_readlane_b32 s30, v40, 3 +; GFX11-NEXT: v_readlane_b32 s31, v40, 4 ; GFX11-NEXT: v_readlane_b32 s6, v40, 2 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 @@ -8502,21 +8509,21 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 5 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 1.0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] -; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3f32_inreg@rel32@lo+4 -; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3f32_inreg@rel32@hi+12 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2.0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 4.0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 3 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 4 +; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 1.0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2.0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 4.0 +; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] +; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3f32_inreg@rel32@lo+4 +; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3f32_inreg@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 4 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 3 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 4 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 @@ -8541,25 +8548,25 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 7 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 ; GFX9-NEXT: v_writelane_b32 v40, s7, 3 ; GFX9-NEXT: v_writelane_b32 v40, s8, 4 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 5 +; GFX9-NEXT: v_writelane_b32 v40, s31, 6 ; GFX9-NEXT: s_mov_b32 s4, 1.0 ; GFX9-NEXT: s_mov_b32 s5, 2.0 ; GFX9-NEXT: s_mov_b32 s6, 4.0 ; GFX9-NEXT: s_mov_b32 s7, -1.0 ; GFX9-NEXT: s_mov_b32 s8, 0.5 -; GFX9-NEXT: v_writelane_b32 v40, s31, 6 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v5f32_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v5f32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 6 ; GFX9-NEXT: v_readlane_b32 s30, v40, 5 +; GFX9-NEXT: v_readlane_b32 s31, v40, 6 ; GFX9-NEXT: v_readlane_b32 s8, v40, 4 ; GFX9-NEXT: v_readlane_b32 s7, v40, 3 ; GFX9-NEXT: v_readlane_b32 s6, v40, 2 @@ -8583,25 +8590,25 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 7 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_mov_b32 s4, 1.0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: s_getpc_b64 s[34:35] -; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v5f32_inreg@rel32@lo+4 -; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v5f32_inreg@rel32@hi+12 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 2.0 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-NEXT: s_mov_b32 s6, 4.0 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-NEXT: s_mov_b32 s7, -1.0 ; GFX10-NEXT: v_writelane_b32 v40, s8, 4 -; GFX10-NEXT: s_mov_b32 s8, 0.5 ; GFX10-NEXT: v_writelane_b32 v40, s30, 5 ; GFX10-NEXT: v_writelane_b32 v40, s31, 6 +; GFX10-NEXT: s_mov_b32 s4, 1.0 +; GFX10-NEXT: s_mov_b32 s5, 2.0 +; GFX10-NEXT: s_mov_b32 s6, 4.0 +; GFX10-NEXT: s_mov_b32 s7, -1.0 +; GFX10-NEXT: s_mov_b32 s8, 0.5 +; GFX10-NEXT: s_getpc_b64 s[34:35] +; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v5f32_inreg@rel32@lo+4 +; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v5f32_inreg@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 6 ; GFX10-NEXT: v_readlane_b32 s30, v40, 5 +; GFX10-NEXT: v_readlane_b32 s31, v40, 6 ; GFX10-NEXT: v_readlane_b32 s8, v40, 4 ; GFX10-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-NEXT: v_readlane_b32 s6, v40, 2 @@ -8625,26 +8632,26 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 7 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_mov_b32 s4, 1.0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: s_getpc_b64 s[0:1] -; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v5f32_inreg@rel32@lo+4 -; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v5f32_inreg@rel32@hi+12 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 2.0 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 -; GFX11-NEXT: s_mov_b32 s6, 4.0 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 -; GFX11-NEXT: s_mov_b32 s7, -1.0 ; GFX11-NEXT: v_writelane_b32 v40, s8, 4 -; GFX11-NEXT: s_mov_b32 s8, 0.5 ; GFX11-NEXT: v_writelane_b32 v40, s30, 5 ; GFX11-NEXT: v_writelane_b32 v40, s31, 6 +; GFX11-NEXT: s_mov_b32 s4, 1.0 +; GFX11-NEXT: s_mov_b32 s5, 2.0 +; GFX11-NEXT: s_mov_b32 s6, 4.0 +; GFX11-NEXT: s_mov_b32 s7, -1.0 +; GFX11-NEXT: s_mov_b32 s8, 0.5 +; GFX11-NEXT: s_getpc_b64 s[0:1] +; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v5f32_inreg@rel32@lo+4 +; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v5f32_inreg@rel32@hi+12 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 6 ; GFX11-NEXT: v_readlane_b32 s30, v40, 5 +; GFX11-NEXT: v_readlane_b32 s31, v40, 6 ; GFX11-NEXT: v_readlane_b32 s8, v40, 4 ; GFX11-NEXT: v_readlane_b32 s7, v40, 3 ; GFX11-NEXT: v_readlane_b32 s6, v40, 2 @@ -8668,25 +8675,25 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 7 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 1.0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] -; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v5f32_inreg@rel32@lo+4 -; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v5f32_inreg@rel32@hi+12 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2.0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 4.0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-SCRATCH-NEXT: s_mov_b32 s7, -1.0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s8, 4 -; GFX10-SCRATCH-NEXT: s_mov_b32 s8, 0.5 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 5 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 6 +; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 1.0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2.0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 4.0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s7, -1.0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s8, 0.5 +; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] +; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v5f32_inreg@rel32@lo+4 +; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v5f32_inreg@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 6 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 5 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 6 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s8, v40, 4 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s6, v40, 2 @@ -8713,19 +8720,19 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 4 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 2 +; GFX9-NEXT: v_writelane_b32 v40, s31, 3 ; GFX9-NEXT: s_mov_b32 s4, 0 ; GFX9-NEXT: s_mov_b32 s5, 0x40100000 -; GFX9-NEXT: v_writelane_b32 v40, s31, 3 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_f64_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_f64_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s30, v40, 2 +; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 @@ -8746,19 +8753,19 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: s_mov_b32 s5, 0x40100000 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_f64_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_f64_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 0x40100000 -; GFX10-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 @@ -8779,20 +8786,20 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 4 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s5, 1 +; GFX11-NEXT: v_writelane_b32 v40, s30, 2 +; GFX11-NEXT: v_writelane_b32 v40, s31, 3 ; GFX11-NEXT: s_mov_b32 s4, 0 -; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: s_mov_b32 s5, 0x40100000 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_f64_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_f64_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 0x40100000 -; GFX11-NEXT: v_writelane_b32 v40, s30, 2 -; GFX11-NEXT: v_writelane_b32 v40, s31, 3 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s30, v40, 2 +; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 @@ -8813,19 +8820,19 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 0 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 0x40100000 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_f64_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_f64_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 0x40100000 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 @@ -8849,23 +8856,23 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 6 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 ; GFX9-NEXT: v_writelane_b32 v40, s7, 3 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 4 +; GFX9-NEXT: v_writelane_b32 v40, s31, 5 ; GFX9-NEXT: s_mov_b32 s4, 0 ; GFX9-NEXT: s_mov_b32 s5, 2.0 ; GFX9-NEXT: s_mov_b32 s6, 0 ; GFX9-NEXT: s_mov_b32 s7, 0x40100000 -; GFX9-NEXT: v_writelane_b32 v40, s31, 5 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2f64_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2f64_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 5 ; GFX9-NEXT: v_readlane_b32 s30, v40, 4 +; GFX9-NEXT: v_readlane_b32 s31, v40, 5 ; GFX9-NEXT: v_readlane_b32 s7, v40, 3 ; GFX9-NEXT: v_readlane_b32 s6, v40, 2 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 @@ -8888,23 +8895,23 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 6 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: s_getpc_b64 s[34:35] -; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2f64_inreg@rel32@lo+4 -; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2f64_inreg@rel32@hi+12 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 2.0 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-NEXT: s_mov_b32 s6, 0 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-NEXT: s_mov_b32 s7, 0x40100000 ; GFX10-NEXT: v_writelane_b32 v40, s30, 4 ; GFX10-NEXT: v_writelane_b32 v40, s31, 5 +; GFX10-NEXT: s_mov_b32 s4, 0 +; GFX10-NEXT: s_mov_b32 s5, 2.0 +; GFX10-NEXT: s_mov_b32 s6, 0 +; GFX10-NEXT: s_mov_b32 s7, 0x40100000 +; GFX10-NEXT: s_getpc_b64 s[34:35] +; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2f64_inreg@rel32@lo+4 +; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2f64_inreg@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-NEXT: v_readlane_b32 s30, v40, 4 +; GFX10-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 @@ -8927,24 +8934,24 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 6 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_mov_b32 s4, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: s_getpc_b64 s[0:1] -; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2f64_inreg@rel32@lo+4 -; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2f64_inreg@rel32@hi+12 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 2.0 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 -; GFX11-NEXT: s_mov_b32 s6, 0 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 -; GFX11-NEXT: s_mov_b32 s7, 0x40100000 ; GFX11-NEXT: v_writelane_b32 v40, s30, 4 ; GFX11-NEXT: v_writelane_b32 v40, s31, 5 +; GFX11-NEXT: s_mov_b32 s4, 0 +; GFX11-NEXT: s_mov_b32 s5, 2.0 +; GFX11-NEXT: s_mov_b32 s6, 0 +; GFX11-NEXT: s_mov_b32 s7, 0x40100000 +; GFX11-NEXT: s_getpc_b64 s[0:1] +; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2f64_inreg@rel32@lo+4 +; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2f64_inreg@rel32@hi+12 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 5 ; GFX11-NEXT: v_readlane_b32 s30, v40, 4 +; GFX11-NEXT: v_readlane_b32 s31, v40, 5 ; GFX11-NEXT: v_readlane_b32 s7, v40, 3 ; GFX11-NEXT: v_readlane_b32 s6, v40, 2 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 @@ -8967,23 +8974,23 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 6 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] -; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2f64_inreg@rel32@lo+4 -; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2f64_inreg@rel32@hi+12 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2.0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-SCRATCH-NEXT: s_mov_b32 s7, 0x40100000 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 4 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 5 +; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2.0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s7, 0x40100000 +; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] +; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2f64_inreg@rel32@lo+4 +; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2f64_inreg@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 4 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 @@ -9009,27 +9016,27 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 8 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 ; GFX9-NEXT: v_writelane_b32 v40, s7, 3 ; GFX9-NEXT: v_writelane_b32 v40, s8, 4 ; GFX9-NEXT: v_writelane_b32 v40, s9, 5 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 6 +; GFX9-NEXT: v_writelane_b32 v40, s31, 7 ; GFX9-NEXT: s_mov_b32 s4, 0 ; GFX9-NEXT: s_mov_b32 s5, 2.0 ; GFX9-NEXT: s_mov_b32 s6, 0 ; GFX9-NEXT: s_mov_b32 s7, 0x40100000 ; GFX9-NEXT: s_mov_b32 s8, 0 ; GFX9-NEXT: s_mov_b32 s9, 0x40200000 -; GFX9-NEXT: v_writelane_b32 v40, s31, 7 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3f64_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3f64_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 7 ; GFX9-NEXT: v_readlane_b32 s30, v40, 6 +; GFX9-NEXT: v_readlane_b32 s31, v40, 7 ; GFX9-NEXT: v_readlane_b32 s9, v40, 5 ; GFX9-NEXT: v_readlane_b32 s8, v40, 4 ; GFX9-NEXT: v_readlane_b32 s7, v40, 3 @@ -9054,27 +9061,27 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 8 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: s_getpc_b64 s[34:35] -; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3f64_inreg@rel32@lo+4 -; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3f64_inreg@rel32@hi+12 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 2.0 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-NEXT: s_mov_b32 s6, 0 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-NEXT: s_mov_b32 s7, 0x40100000 ; GFX10-NEXT: v_writelane_b32 v40, s8, 4 -; GFX10-NEXT: s_mov_b32 s8, 0 ; GFX10-NEXT: v_writelane_b32 v40, s9, 5 -; GFX10-NEXT: s_mov_b32 s9, 0x40200000 ; GFX10-NEXT: v_writelane_b32 v40, s30, 6 ; GFX10-NEXT: v_writelane_b32 v40, s31, 7 +; GFX10-NEXT: s_mov_b32 s4, 0 +; GFX10-NEXT: s_mov_b32 s5, 2.0 +; GFX10-NEXT: s_mov_b32 s6, 0 +; GFX10-NEXT: s_mov_b32 s7, 0x40100000 +; GFX10-NEXT: s_mov_b32 s8, 0 +; GFX10-NEXT: s_mov_b32 s9, 0x40200000 +; GFX10-NEXT: s_getpc_b64 s[34:35] +; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3f64_inreg@rel32@lo+4 +; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3f64_inreg@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 7 ; GFX10-NEXT: v_readlane_b32 s30, v40, 6 +; GFX10-NEXT: v_readlane_b32 s31, v40, 7 ; GFX10-NEXT: v_readlane_b32 s9, v40, 5 ; GFX10-NEXT: v_readlane_b32 s8, v40, 4 ; GFX10-NEXT: v_readlane_b32 s7, v40, 3 @@ -9099,28 +9106,28 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 8 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_mov_b32 s4, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: s_getpc_b64 s[0:1] -; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3f64_inreg@rel32@lo+4 -; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3f64_inreg@rel32@hi+12 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 2.0 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 -; GFX11-NEXT: s_mov_b32 s6, 0 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 -; GFX11-NEXT: s_mov_b32 s7, 0x40100000 ; GFX11-NEXT: v_writelane_b32 v40, s8, 4 -; GFX11-NEXT: s_mov_b32 s8, 0 ; GFX11-NEXT: v_writelane_b32 v40, s9, 5 -; GFX11-NEXT: s_mov_b32 s9, 0x40200000 ; GFX11-NEXT: v_writelane_b32 v40, s30, 6 ; GFX11-NEXT: v_writelane_b32 v40, s31, 7 +; GFX11-NEXT: s_mov_b32 s4, 0 +; GFX11-NEXT: s_mov_b32 s5, 2.0 +; GFX11-NEXT: s_mov_b32 s6, 0 +; GFX11-NEXT: s_mov_b32 s7, 0x40100000 +; GFX11-NEXT: s_mov_b32 s8, 0 +; GFX11-NEXT: s_mov_b32 s9, 0x40200000 +; GFX11-NEXT: s_getpc_b64 s[0:1] +; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3f64_inreg@rel32@lo+4 +; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3f64_inreg@rel32@hi+12 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 7 ; GFX11-NEXT: v_readlane_b32 s30, v40, 6 +; GFX11-NEXT: v_readlane_b32 s31, v40, 7 ; GFX11-NEXT: v_readlane_b32 s9, v40, 5 ; GFX11-NEXT: v_readlane_b32 s8, v40, 4 ; GFX11-NEXT: v_readlane_b32 s7, v40, 3 @@ -9145,27 +9152,27 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 8 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] -; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3f64_inreg@rel32@lo+4 -; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3f64_inreg@rel32@hi+12 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2.0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-SCRATCH-NEXT: s_mov_b32 s7, 0x40100000 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s8, 4 -; GFX10-SCRATCH-NEXT: s_mov_b32 s8, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s9, 5 -; GFX10-SCRATCH-NEXT: s_mov_b32 s9, 0x40200000 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 6 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 7 +; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2.0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s7, 0x40100000 +; GFX10-SCRATCH-NEXT: s_mov_b32 s8, 0 +; GFX10-SCRATCH-NEXT: s_mov_b32 s9, 0x40200000 +; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] +; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3f64_inreg@rel32@lo+4 +; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3f64_inreg@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 7 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 6 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 7 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s9, v40, 5 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s8, v40, 4 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s7, v40, 3 @@ -9193,17 +9200,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 3 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: v_writelane_b32 v40, s4, 0 -; GFX9-NEXT: s_load_dword s4, s[34:35], 0x0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s30, 1 ; GFX9-NEXT: v_writelane_b32 v40, s31, 2 +; GFX9-NEXT: s_load_dword s4, s[34:35], 0x0 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2i16_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2i16_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s30, v40, 1 +; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 3 @@ -9223,17 +9230,17 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_load_dword s4, s[34:35], 0x0 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2i16_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2i16_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 3 @@ -9253,18 +9260,18 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 3 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s30, 1 +; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: s_load_b32 s4, s[0:1], 0x0 -; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2i16_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2i16_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 1 -; GFX11-NEXT: v_writelane_b32 v40, s31, 2 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s30, v40, 1 +; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 3 @@ -9284,17 +9291,17 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-SCRATCH-NEXT: s_load_dword s4, s[0:1], 0x0 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2i16_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2i16_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 3 @@ -9318,18 +9325,18 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 4 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 -; GFX9-NEXT: s_load_dwordx2 s[4:5], s[34:35], 0x0 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 2 ; GFX9-NEXT: v_writelane_b32 v40, s31, 3 +; GFX9-NEXT: s_load_dwordx2 s[4:5], s[34:35], 0x0 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3i16_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3i16_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s30, v40, 2 +; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 @@ -9350,18 +9357,18 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_load_dwordx2 s[4:5], s[34:35], 0x0 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3i16_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3i16_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 @@ -9382,19 +9389,19 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 4 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 +; GFX11-NEXT: v_writelane_b32 v40, s30, 2 +; GFX11-NEXT: v_writelane_b32 v40, s31, 3 ; GFX11-NEXT: s_load_b64 s[4:5], s[0:1], 0x0 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3i16_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3i16_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 2 -; GFX11-NEXT: v_writelane_b32 v40, s31, 3 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s30, v40, 2 +; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 @@ -9415,18 +9422,18 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3i16_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3i16_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 @@ -9451,18 +9458,18 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 4 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 -; GFX9-NEXT: s_load_dwordx2 s[4:5], s[34:35], 0x0 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 2 ; GFX9-NEXT: v_writelane_b32 v40, s31, 3 +; GFX9-NEXT: s_load_dwordx2 s[4:5], s[34:35], 0x0 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3f16_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3f16_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s30, v40, 2 +; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 @@ -9483,18 +9490,18 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_load_dwordx2 s[4:5], s[34:35], 0x0 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3f16_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3f16_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 @@ -9515,19 +9522,19 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 4 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 +; GFX11-NEXT: v_writelane_b32 v40, s30, 2 +; GFX11-NEXT: v_writelane_b32 v40, s31, 3 ; GFX11-NEXT: s_load_b64 s[4:5], s[0:1], 0x0 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3f16_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3f16_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 2 -; GFX11-NEXT: v_writelane_b32 v40, s31, 3 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s30, v40, 2 +; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 @@ -9548,18 +9555,18 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3f16_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3f16_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 @@ -9584,19 +9591,19 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 4 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 2 +; GFX9-NEXT: v_writelane_b32 v40, s31, 3 ; GFX9-NEXT: s_mov_b32 s4, 0x20001 ; GFX9-NEXT: s_mov_b32 s5, 3 -; GFX9-NEXT: v_writelane_b32 v40, s31, 3 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3i16_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3i16_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s30, v40, 2 +; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 @@ -9617,19 +9624,19 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_mov_b32 s4, 0x20001 -; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: s_mov_b32 s5, 3 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3i16_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3i16_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 3 -; GFX10-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 @@ -9650,20 +9657,20 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 4 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s5, 1 +; GFX11-NEXT: v_writelane_b32 v40, s30, 2 +; GFX11-NEXT: v_writelane_b32 v40, s31, 3 ; GFX11-NEXT: s_mov_b32 s4, 0x20001 -; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: s_mov_b32 s5, 3 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3i16_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3i16_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 3 -; GFX11-NEXT: v_writelane_b32 v40, s30, 2 -; GFX11-NEXT: v_writelane_b32 v40, s31, 3 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s30, v40, 2 +; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 @@ -9684,19 +9691,19 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 0x20001 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 3 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3i16_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3i16_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 3 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 @@ -9720,19 +9727,19 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 4 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 2 +; GFX9-NEXT: v_writelane_b32 v40, s31, 3 ; GFX9-NEXT: s_mov_b32 s4, 0x40003c00 ; GFX9-NEXT: s_movk_i32 s5, 0x4400 -; GFX9-NEXT: v_writelane_b32 v40, s31, 3 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3f16_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3f16_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s30, v40, 2 +; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 @@ -9753,19 +9760,19 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_mov_b32 s4, 0x40003c00 -; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: s_movk_i32 s5, 0x4400 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3f16_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3f16_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_movk_i32 s5, 0x4400 -; GFX10-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 @@ -9786,20 +9793,20 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 4 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s5, 1 +; GFX11-NEXT: v_writelane_b32 v40, s30, 2 +; GFX11-NEXT: v_writelane_b32 v40, s31, 3 ; GFX11-NEXT: s_mov_b32 s4, 0x40003c00 -; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: s_movk_i32 s5, 0x4400 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3f16_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3f16_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_movk_i32 s5, 0x4400 -; GFX11-NEXT: v_writelane_b32 v40, s30, 2 -; GFX11-NEXT: v_writelane_b32 v40, s31, 3 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s30, v40, 2 +; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 @@ -9820,19 +9827,19 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 0x40003c00 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: s_movk_i32 s5, 0x4400 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3f16_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3f16_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_movk_i32 s5, 0x4400 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 @@ -9856,18 +9863,18 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 4 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 -; GFX9-NEXT: s_load_dwordx2 s[4:5], s[34:35], 0x0 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 2 ; GFX9-NEXT: v_writelane_b32 v40, s31, 3 +; GFX9-NEXT: s_load_dwordx2 s[4:5], s[34:35], 0x0 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v4i16_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v4i16_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s30, v40, 2 +; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 @@ -9888,18 +9895,18 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_load_dwordx2 s[4:5], s[34:35], 0x0 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v4i16_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v4i16_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 @@ -9920,19 +9927,19 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 4 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 +; GFX11-NEXT: v_writelane_b32 v40, s30, 2 +; GFX11-NEXT: v_writelane_b32 v40, s31, 3 ; GFX11-NEXT: s_load_b64 s[4:5], s[0:1], 0x0 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v4i16_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v4i16_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 2 -; GFX11-NEXT: v_writelane_b32 v40, s31, 3 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s30, v40, 2 +; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 @@ -9953,18 +9960,18 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v4i16_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v4i16_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 @@ -9989,19 +9996,19 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 4 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 2 +; GFX9-NEXT: v_writelane_b32 v40, s31, 3 ; GFX9-NEXT: s_mov_b32 s4, 0x20001 ; GFX9-NEXT: s_mov_b32 s5, 0x40003 -; GFX9-NEXT: v_writelane_b32 v40, s31, 3 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v4i16_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v4i16_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s30, v40, 2 +; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 @@ -10022,19 +10029,19 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_mov_b32 s4, 0x20001 -; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: s_mov_b32 s5, 0x40003 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v4i16_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v4i16_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 0x40003 -; GFX10-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 @@ -10055,20 +10062,20 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 4 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s5, 1 +; GFX11-NEXT: v_writelane_b32 v40, s30, 2 +; GFX11-NEXT: v_writelane_b32 v40, s31, 3 ; GFX11-NEXT: s_mov_b32 s4, 0x20001 -; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: s_mov_b32 s5, 0x40003 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v4i16_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v4i16_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 0x40003 -; GFX11-NEXT: v_writelane_b32 v40, s30, 2 -; GFX11-NEXT: v_writelane_b32 v40, s31, 3 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s30, v40, 2 +; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 @@ -10089,19 +10096,19 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 0x20001 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 0x40003 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v4i16_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v4i16_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 0x40003 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 @@ -10125,17 +10132,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 3 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: v_writelane_b32 v40, s4, 0 -; GFX9-NEXT: s_load_dword s4, s[34:35], 0x0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s30, 1 ; GFX9-NEXT: v_writelane_b32 v40, s31, 2 +; GFX9-NEXT: s_load_dword s4, s[34:35], 0x0 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2f16_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2f16_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s30, v40, 1 +; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 3 @@ -10155,17 +10162,17 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_load_dword s4, s[34:35], 0x0 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2f16_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2f16_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 3 @@ -10185,18 +10192,18 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 3 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s30, 1 +; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: s_load_b32 s4, s[0:1], 0x0 -; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2f16_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2f16_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 1 -; GFX11-NEXT: v_writelane_b32 v40, s31, 2 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s30, v40, 1 +; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 3 @@ -10216,17 +10223,17 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-SCRATCH-NEXT: s_load_dword s4, s[0:1], 0x0 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2f16_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2f16_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 3 @@ -10250,18 +10257,18 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 4 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 -; GFX9-NEXT: s_load_dwordx2 s[4:5], s[34:35], 0x0 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 2 ; GFX9-NEXT: v_writelane_b32 v40, s31, 3 +; GFX9-NEXT: s_load_dwordx2 s[4:5], s[34:35], 0x0 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2i32_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2i32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s30, v40, 2 +; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 @@ -10282,18 +10289,18 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_load_dwordx2 s[4:5], s[34:35], 0x0 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2i32_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2i32_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 @@ -10314,19 +10321,19 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 4 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 +; GFX11-NEXT: v_writelane_b32 v40, s30, 2 +; GFX11-NEXT: v_writelane_b32 v40, s31, 3 ; GFX11-NEXT: s_load_b64 s[4:5], s[0:1], 0x0 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2i32_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2i32_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 2 -; GFX11-NEXT: v_writelane_b32 v40, s31, 3 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s30, v40, 2 +; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 @@ -10347,18 +10354,18 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2i32_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2i32_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 @@ -10383,19 +10390,19 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 4 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 2 +; GFX9-NEXT: v_writelane_b32 v40, s31, 3 ; GFX9-NEXT: s_mov_b32 s4, 1 ; GFX9-NEXT: s_mov_b32 s5, 2 -; GFX9-NEXT: v_writelane_b32 v40, s31, 3 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v2i32_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v2i32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s30, v40, 2 +; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 @@ -10416,19 +10423,19 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_mov_b32 s4, 1 -; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: s_mov_b32 s5, 2 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v2i32_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v2i32_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 2 -; GFX10-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 @@ -10449,20 +10456,20 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 4 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s5, 1 +; GFX11-NEXT: v_writelane_b32 v40, s30, 2 +; GFX11-NEXT: v_writelane_b32 v40, s31, 3 ; GFX11-NEXT: s_mov_b32 s4, 1 -; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: s_mov_b32 s5, 2 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v2i32_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v2i32_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 2 -; GFX11-NEXT: v_writelane_b32 v40, s30, 2 -; GFX11-NEXT: v_writelane_b32 v40, s31, 3 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s30, v40, 2 +; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 @@ -10483,19 +10490,19 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 1 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v2i32_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v2i32_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 @@ -10519,21 +10526,21 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 5 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 3 +; GFX9-NEXT: v_writelane_b32 v40, s31, 4 ; GFX9-NEXT: s_mov_b32 s4, 3 ; GFX9-NEXT: s_mov_b32 s5, 4 ; GFX9-NEXT: s_mov_b32 s6, 5 -; GFX9-NEXT: v_writelane_b32 v40, s31, 4 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3i32_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3i32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 4 ; GFX9-NEXT: v_readlane_b32 s30, v40, 3 +; GFX9-NEXT: v_readlane_b32 s31, v40, 4 ; GFX9-NEXT: v_readlane_b32 s6, v40, 2 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 @@ -10555,21 +10562,21 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 5 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_mov_b32 s4, 3 ; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: s_getpc_b64 s[34:35] -; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3i32_inreg@rel32@lo+4 -; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3i32_inreg@rel32@hi+12 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 4 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-NEXT: s_mov_b32 s6, 5 ; GFX10-NEXT: v_writelane_b32 v40, s30, 3 ; GFX10-NEXT: v_writelane_b32 v40, s31, 4 +; GFX10-NEXT: s_mov_b32 s4, 3 +; GFX10-NEXT: s_mov_b32 s5, 4 +; GFX10-NEXT: s_mov_b32 s6, 5 +; GFX10-NEXT: s_getpc_b64 s[34:35] +; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3i32_inreg@rel32@lo+4 +; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3i32_inreg@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 4 ; GFX10-NEXT: v_readlane_b32 s30, v40, 3 +; GFX10-NEXT: v_readlane_b32 s31, v40, 4 ; GFX10-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 @@ -10591,22 +10598,22 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 5 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_mov_b32 s4, 3 ; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: s_getpc_b64 s[0:1] -; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3i32_inreg@rel32@lo+4 -; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3i32_inreg@rel32@hi+12 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 4 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 -; GFX11-NEXT: s_mov_b32 s6, 5 ; GFX11-NEXT: v_writelane_b32 v40, s30, 3 ; GFX11-NEXT: v_writelane_b32 v40, s31, 4 +; GFX11-NEXT: s_mov_b32 s4, 3 +; GFX11-NEXT: s_mov_b32 s5, 4 +; GFX11-NEXT: s_mov_b32 s6, 5 +; GFX11-NEXT: s_getpc_b64 s[0:1] +; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3i32_inreg@rel32@lo+4 +; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3i32_inreg@rel32@hi+12 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 4 ; GFX11-NEXT: v_readlane_b32 s30, v40, 3 +; GFX11-NEXT: v_readlane_b32 s31, v40, 4 ; GFX11-NEXT: v_readlane_b32 s6, v40, 2 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 @@ -10628,21 +10635,21 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 5 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 3 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] -; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3i32_inreg@rel32@lo+4 -; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3i32_inreg@rel32@hi+12 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 4 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 5 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 3 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 4 +; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 3 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 4 +; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 5 +; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] +; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3i32_inreg@rel32@lo+4 +; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3i32_inreg@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 4 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 3 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 4 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s4, v40, 0 @@ -10667,23 +10674,23 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 6 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 ; GFX9-NEXT: v_writelane_b32 v40, s7, 3 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 4 +; GFX9-NEXT: v_writelane_b32 v40, s31, 5 ; GFX9-NEXT: s_mov_b32 s4, 3 ; GFX9-NEXT: s_mov_b32 s5, 4 ; GFX9-NEXT: s_mov_b32 s6, 5 ; GFX9-NEXT: s_mov_b32 s7, 6 -; GFX9-NEXT: v_writelane_b32 v40, s31, 5 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v3i32_i32_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v3i32_i32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 5 ; GFX9-NEXT: v_readlane_b32 s30, v40, 4 +; GFX9-NEXT: v_readlane_b32 s31, v40, 5 ; GFX9-NEXT: v_readlane_b32 s7, v40, 3 ; GFX9-NEXT: v_readlane_b32 s6, v40, 2 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 @@ -10706,23 +10713,23 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 6 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_mov_b32 s4, 3 ; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: s_getpc_b64 s[34:35] -; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3i32_i32_inreg@rel32@lo+4 -; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3i32_i32_inreg@rel32@hi+12 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 4 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-NEXT: s_mov_b32 s6, 5 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-NEXT: s_mov_b32 s7, 6 ; GFX10-NEXT: v_writelane_b32 v40, s30, 4 ; GFX10-NEXT: v_writelane_b32 v40, s31, 5 +; GFX10-NEXT: s_mov_b32 s4, 3 +; GFX10-NEXT: s_mov_b32 s5, 4 +; GFX10-NEXT: s_mov_b32 s6, 5 +; GFX10-NEXT: s_mov_b32 s7, 6 +; GFX10-NEXT: s_getpc_b64 s[34:35] +; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v3i32_i32_inreg@rel32@lo+4 +; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v3i32_i32_inreg@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-NEXT: v_readlane_b32 s30, v40, 4 +; GFX10-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 @@ -10745,24 +10752,24 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 6 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_mov_b32 s4, 3 ; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: s_getpc_b64 s[0:1] -; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3i32_i32_inreg@rel32@lo+4 -; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3i32_i32_inreg@rel32@hi+12 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 4 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 -; GFX11-NEXT: s_mov_b32 s6, 5 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 -; GFX11-NEXT: s_mov_b32 s7, 6 ; GFX11-NEXT: v_writelane_b32 v40, s30, 4 ; GFX11-NEXT: v_writelane_b32 v40, s31, 5 +; GFX11-NEXT: s_mov_b32 s4, 3 +; GFX11-NEXT: s_mov_b32 s5, 4 +; GFX11-NEXT: s_mov_b32 s6, 5 +; GFX11-NEXT: s_mov_b32 s7, 6 +; GFX11-NEXT: s_getpc_b64 s[0:1] +; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v3i32_i32_inreg@rel32@lo+4 +; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v3i32_i32_inreg@rel32@hi+12 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 5 ; GFX11-NEXT: v_readlane_b32 s30, v40, 4 +; GFX11-NEXT: v_readlane_b32 s31, v40, 5 ; GFX11-NEXT: v_readlane_b32 s7, v40, 3 ; GFX11-NEXT: v_readlane_b32 s6, v40, 2 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 @@ -10785,23 +10792,23 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 6 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 3 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] -; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3i32_i32_inreg@rel32@lo+4 -; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3i32_i32_inreg@rel32@hi+12 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 4 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 5 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-SCRATCH-NEXT: s_mov_b32 s7, 6 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 4 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 5 +; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 3 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 4 +; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 5 +; GFX10-SCRATCH-NEXT: s_mov_b32 s7, 6 +; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] +; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v3i32_i32_inreg@rel32@lo+4 +; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v3i32_i32_inreg@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 4 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 @@ -10827,20 +10834,20 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 6 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 ; GFX9-NEXT: v_writelane_b32 v40, s7, 3 -; GFX9-NEXT: s_load_dwordx4 s[4:7], s[34:35], 0x0 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 4 ; GFX9-NEXT: v_writelane_b32 v40, s31, 5 +; GFX9-NEXT: s_load_dwordx4 s[4:7], s[34:35], 0x0 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v4i32_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v4i32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 5 ; GFX9-NEXT: v_readlane_b32 s30, v40, 4 +; GFX9-NEXT: v_readlane_b32 s31, v40, 5 ; GFX9-NEXT: v_readlane_b32 s7, v40, 3 ; GFX9-NEXT: v_readlane_b32 s6, v40, 2 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 @@ -10863,20 +10870,20 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 6 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 +; GFX10-NEXT: v_writelane_b32 v40, s30, 4 +; GFX10-NEXT: v_writelane_b32 v40, s31, 5 ; GFX10-NEXT: s_load_dwordx4 s[4:7], s[34:35], 0x0 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v4i32_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v4i32_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 4 -; GFX10-NEXT: v_writelane_b32 v40, s31, 5 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-NEXT: v_readlane_b32 s30, v40, 4 +; GFX10-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 @@ -10899,21 +10906,21 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 6 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 +; GFX11-NEXT: v_writelane_b32 v40, s30, 4 +; GFX11-NEXT: v_writelane_b32 v40, s31, 5 ; GFX11-NEXT: s_load_b128 s[4:7], s[0:1], 0x0 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v4i32_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v4i32_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 4 -; GFX11-NEXT: v_writelane_b32 v40, s31, 5 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 5 ; GFX11-NEXT: v_readlane_b32 s30, v40, 4 +; GFX11-NEXT: v_readlane_b32 s31, v40, 5 ; GFX11-NEXT: v_readlane_b32 s7, v40, 3 ; GFX11-NEXT: v_readlane_b32 s6, v40, 2 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 @@ -10936,20 +10943,20 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 6 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 4 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 5 ; GFX10-SCRATCH-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v4i32_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v4i32_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 4 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 5 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 4 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 @@ -10976,23 +10983,23 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 6 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 ; GFX9-NEXT: v_writelane_b32 v40, s7, 3 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 4 +; GFX9-NEXT: v_writelane_b32 v40, s31, 5 ; GFX9-NEXT: s_mov_b32 s4, 1 ; GFX9-NEXT: s_mov_b32 s5, 2 ; GFX9-NEXT: s_mov_b32 s6, 3 ; GFX9-NEXT: s_mov_b32 s7, 4 -; GFX9-NEXT: v_writelane_b32 v40, s31, 5 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v4i32_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v4i32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 5 ; GFX9-NEXT: v_readlane_b32 s30, v40, 4 +; GFX9-NEXT: v_readlane_b32 s31, v40, 5 ; GFX9-NEXT: v_readlane_b32 s7, v40, 3 ; GFX9-NEXT: v_readlane_b32 s6, v40, 2 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 @@ -11015,23 +11022,23 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 6 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_mov_b32 s4, 1 ; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: s_getpc_b64 s[34:35] -; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v4i32_inreg@rel32@lo+4 -; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v4i32_inreg@rel32@hi+12 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 2 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-NEXT: s_mov_b32 s6, 3 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-NEXT: s_mov_b32 s7, 4 ; GFX10-NEXT: v_writelane_b32 v40, s30, 4 ; GFX10-NEXT: v_writelane_b32 v40, s31, 5 +; GFX10-NEXT: s_mov_b32 s4, 1 +; GFX10-NEXT: s_mov_b32 s5, 2 +; GFX10-NEXT: s_mov_b32 s6, 3 +; GFX10-NEXT: s_mov_b32 s7, 4 +; GFX10-NEXT: s_getpc_b64 s[34:35] +; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v4i32_inreg@rel32@lo+4 +; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v4i32_inreg@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-NEXT: v_readlane_b32 s30, v40, 4 +; GFX10-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 @@ -11054,24 +11061,24 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 6 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_mov_b32 s4, 1 ; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: s_getpc_b64 s[0:1] -; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v4i32_inreg@rel32@lo+4 -; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v4i32_inreg@rel32@hi+12 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 2 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 -; GFX11-NEXT: s_mov_b32 s6, 3 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 -; GFX11-NEXT: s_mov_b32 s7, 4 ; GFX11-NEXT: v_writelane_b32 v40, s30, 4 ; GFX11-NEXT: v_writelane_b32 v40, s31, 5 +; GFX11-NEXT: s_mov_b32 s4, 1 +; GFX11-NEXT: s_mov_b32 s5, 2 +; GFX11-NEXT: s_mov_b32 s6, 3 +; GFX11-NEXT: s_mov_b32 s7, 4 +; GFX11-NEXT: s_getpc_b64 s[0:1] +; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v4i32_inreg@rel32@lo+4 +; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v4i32_inreg@rel32@hi+12 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 5 ; GFX11-NEXT: v_readlane_b32 s30, v40, 4 +; GFX11-NEXT: v_readlane_b32 s31, v40, 5 ; GFX11-NEXT: v_readlane_b32 s7, v40, 3 ; GFX11-NEXT: v_readlane_b32 s6, v40, 2 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 @@ -11094,23 +11101,23 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 6 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] -; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v4i32_inreg@rel32@lo+4 -; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v4i32_inreg@rel32@hi+12 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 3 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-SCRATCH-NEXT: s_mov_b32 s7, 4 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 4 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 5 +; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 1 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2 +; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 3 +; GFX10-SCRATCH-NEXT: s_mov_b32 s7, 4 +; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] +; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v4i32_inreg@rel32@lo+4 +; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v4i32_inreg@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 4 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 5 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s6, v40, 2 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s5, v40, 1 @@ -11136,25 +11143,25 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 7 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 ; GFX9-NEXT: v_writelane_b32 v40, s7, 3 ; GFX9-NEXT: v_writelane_b32 v40, s8, 4 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 5 +; GFX9-NEXT: v_writelane_b32 v40, s31, 6 ; GFX9-NEXT: s_mov_b32 s4, 1 ; GFX9-NEXT: s_mov_b32 s5, 2 ; GFX9-NEXT: s_mov_b32 s6, 3 ; GFX9-NEXT: s_mov_b32 s7, 4 ; GFX9-NEXT: s_mov_b32 s8, 5 -; GFX9-NEXT: v_writelane_b32 v40, s31, 6 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v5i32_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v5i32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 6 ; GFX9-NEXT: v_readlane_b32 s30, v40, 5 +; GFX9-NEXT: v_readlane_b32 s31, v40, 6 ; GFX9-NEXT: v_readlane_b32 s8, v40, 4 ; GFX9-NEXT: v_readlane_b32 s7, v40, 3 ; GFX9-NEXT: v_readlane_b32 s6, v40, 2 @@ -11178,25 +11185,25 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 7 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_mov_b32 s4, 1 ; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: s_getpc_b64 s[34:35] -; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v5i32_inreg@rel32@lo+4 -; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v5i32_inreg@rel32@hi+12 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 2 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-NEXT: s_mov_b32 s6, 3 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-NEXT: s_mov_b32 s7, 4 ; GFX10-NEXT: v_writelane_b32 v40, s8, 4 -; GFX10-NEXT: s_mov_b32 s8, 5 ; GFX10-NEXT: v_writelane_b32 v40, s30, 5 ; GFX10-NEXT: v_writelane_b32 v40, s31, 6 +; GFX10-NEXT: s_mov_b32 s4, 1 +; GFX10-NEXT: s_mov_b32 s5, 2 +; GFX10-NEXT: s_mov_b32 s6, 3 +; GFX10-NEXT: s_mov_b32 s7, 4 +; GFX10-NEXT: s_mov_b32 s8, 5 +; GFX10-NEXT: s_getpc_b64 s[34:35] +; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v5i32_inreg@rel32@lo+4 +; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v5i32_inreg@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 6 ; GFX10-NEXT: v_readlane_b32 s30, v40, 5 +; GFX10-NEXT: v_readlane_b32 s31, v40, 6 ; GFX10-NEXT: v_readlane_b32 s8, v40, 4 ; GFX10-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-NEXT: v_readlane_b32 s6, v40, 2 @@ -11220,26 +11227,26 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 7 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_mov_b32 s4, 1 ; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: s_getpc_b64 s[0:1] -; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v5i32_inreg@rel32@lo+4 -; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v5i32_inreg@rel32@hi+12 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 2 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 -; GFX11-NEXT: s_mov_b32 s6, 3 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 -; GFX11-NEXT: s_mov_b32 s7, 4 ; GFX11-NEXT: v_writelane_b32 v40, s8, 4 -; GFX11-NEXT: s_mov_b32 s8, 5 ; GFX11-NEXT: v_writelane_b32 v40, s30, 5 ; GFX11-NEXT: v_writelane_b32 v40, s31, 6 +; GFX11-NEXT: s_mov_b32 s4, 1 +; GFX11-NEXT: s_mov_b32 s5, 2 +; GFX11-NEXT: s_mov_b32 s6, 3 +; GFX11-NEXT: s_mov_b32 s7, 4 +; GFX11-NEXT: s_mov_b32 s8, 5 +; GFX11-NEXT: s_getpc_b64 s[0:1] +; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v5i32_inreg@rel32@lo+4 +; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v5i32_inreg@rel32@hi+12 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 6 ; GFX11-NEXT: v_readlane_b32 s30, v40, 5 +; GFX11-NEXT: v_readlane_b32 s31, v40, 6 ; GFX11-NEXT: v_readlane_b32 s8, v40, 4 ; GFX11-NEXT: v_readlane_b32 s7, v40, 3 ; GFX11-NEXT: v_readlane_b32 s6, v40, 2 @@ -11263,25 +11270,25 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 7 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] -; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v5i32_inreg@rel32@lo+4 -; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v5i32_inreg@rel32@hi+12 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 3 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-SCRATCH-NEXT: s_mov_b32 s7, 4 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s8, 4 -; GFX10-SCRATCH-NEXT: s_mov_b32 s8, 5 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 5 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 6 +; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 1 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2 +; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 3 +; GFX10-SCRATCH-NEXT: s_mov_b32 s7, 4 +; GFX10-SCRATCH-NEXT: s_mov_b32 s8, 5 +; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] +; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v5i32_inreg@rel32@lo+4 +; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v5i32_inreg@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 6 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 5 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 6 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s8, v40, 4 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s7, v40, 3 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s6, v40, 2 @@ -11308,26 +11315,26 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 10 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 -; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX9-NEXT: v_writelane_b32 v40, s7, 3 ; GFX9-NEXT: v_writelane_b32 v40, s8, 4 ; GFX9-NEXT: v_writelane_b32 v40, s9, 5 ; GFX9-NEXT: v_writelane_b32 v40, s10, 6 ; GFX9-NEXT: v_writelane_b32 v40, s11, 7 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: s_load_dwordx8 s[4:11], s[34:35], 0x0 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 8 ; GFX9-NEXT: v_writelane_b32 v40, s31, 9 +; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_load_dwordx8 s[4:11], s[34:35], 0x0 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v8i32_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v8i32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 9 ; GFX9-NEXT: v_readlane_b32 s30, v40, 8 +; GFX9-NEXT: v_readlane_b32 s31, v40, 9 ; GFX9-NEXT: v_readlane_b32 s11, v40, 7 ; GFX9-NEXT: v_readlane_b32 s10, v40, 6 ; GFX9-NEXT: v_readlane_b32 s9, v40, 5 @@ -11354,9 +11361,8 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 10 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 @@ -11364,16 +11370,17 @@ ; GFX10-NEXT: v_writelane_b32 v40, s9, 5 ; GFX10-NEXT: v_writelane_b32 v40, s10, 6 ; GFX10-NEXT: v_writelane_b32 v40, s11, 7 +; GFX10-NEXT: v_writelane_b32 v40, s30, 8 +; GFX10-NEXT: v_writelane_b32 v40, s31, 9 +; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_load_dwordx8 s[4:11], s[34:35], 0x0 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v8i32_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v8i32_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 8 -; GFX10-NEXT: v_writelane_b32 v40, s31, 9 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 9 ; GFX10-NEXT: v_readlane_b32 s30, v40, 8 +; GFX10-NEXT: v_readlane_b32 s31, v40, 9 ; GFX10-NEXT: v_readlane_b32 s11, v40, 7 ; GFX10-NEXT: v_readlane_b32 s10, v40, 6 ; GFX10-NEXT: v_readlane_b32 s9, v40, 5 @@ -11400,9 +11407,8 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 10 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 @@ -11410,17 +11416,18 @@ ; GFX11-NEXT: v_writelane_b32 v40, s9, 5 ; GFX11-NEXT: v_writelane_b32 v40, s10, 6 ; GFX11-NEXT: v_writelane_b32 v40, s11, 7 +; GFX11-NEXT: v_writelane_b32 v40, s30, 8 +; GFX11-NEXT: v_writelane_b32 v40, s31, 9 +; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_load_b256 s[4:11], s[0:1], 0x0 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v8i32_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v8i32_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 8 -; GFX11-NEXT: v_writelane_b32 v40, s31, 9 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 9 ; GFX11-NEXT: v_readlane_b32 s30, v40, 8 +; GFX11-NEXT: v_readlane_b32 s31, v40, 9 ; GFX11-NEXT: v_readlane_b32 s11, v40, 7 ; GFX11-NEXT: v_readlane_b32 s10, v40, 6 ; GFX11-NEXT: v_readlane_b32 s9, v40, 5 @@ -11447,9 +11454,8 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 10 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 @@ -11457,16 +11463,17 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s9, 5 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s10, 6 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s11, 7 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 8 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 9 +; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-SCRATCH-NEXT: s_load_dwordx8 s[4:11], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v8i32_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v8i32_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 8 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 9 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 9 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 8 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 9 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s11, v40, 7 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s10, v40, 6 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s9, v40, 5 @@ -11498,6 +11505,7 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 10 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 @@ -11506,8 +11514,8 @@ ; GFX9-NEXT: v_writelane_b32 v40, s9, 5 ; GFX9-NEXT: v_writelane_b32 v40, s10, 6 ; GFX9-NEXT: v_writelane_b32 v40, s11, 7 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 8 +; GFX9-NEXT: v_writelane_b32 v40, s31, 9 ; GFX9-NEXT: s_mov_b32 s4, 1 ; GFX9-NEXT: s_mov_b32 s5, 2 ; GFX9-NEXT: s_mov_b32 s6, 3 @@ -11516,13 +11524,12 @@ ; GFX9-NEXT: s_mov_b32 s9, 6 ; GFX9-NEXT: s_mov_b32 s10, 7 ; GFX9-NEXT: s_mov_b32 s11, 8 -; GFX9-NEXT: v_writelane_b32 v40, s31, 9 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v8i32_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v8i32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 9 ; GFX9-NEXT: v_readlane_b32 s30, v40, 8 +; GFX9-NEXT: v_readlane_b32 s31, v40, 9 ; GFX9-NEXT: v_readlane_b32 s11, v40, 7 ; GFX9-NEXT: v_readlane_b32 s10, v40, 6 ; GFX9-NEXT: v_readlane_b32 s9, v40, 5 @@ -11549,31 +11556,31 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 10 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_mov_b32 s4, 1 ; GFX10-NEXT: s_addk_i32 s32, 0x200 -; GFX10-NEXT: s_getpc_b64 s[34:35] -; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v8i32_inreg@rel32@lo+4 -; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v8i32_inreg@rel32@hi+12 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-NEXT: s_mov_b32 s5, 2 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-NEXT: s_mov_b32 s6, 3 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-NEXT: s_mov_b32 s7, 4 ; GFX10-NEXT: v_writelane_b32 v40, s8, 4 +; GFX10-NEXT: v_writelane_b32 v40, s9, 5 +; GFX10-NEXT: v_writelane_b32 v40, s10, 6 +; GFX10-NEXT: v_writelane_b32 v40, s11, 7 +; GFX10-NEXT: v_writelane_b32 v40, s30, 8 +; GFX10-NEXT: v_writelane_b32 v40, s31, 9 +; GFX10-NEXT: s_mov_b32 s4, 1 +; GFX10-NEXT: s_mov_b32 s5, 2 +; GFX10-NEXT: s_mov_b32 s6, 3 +; GFX10-NEXT: s_mov_b32 s7, 4 ; GFX10-NEXT: s_mov_b32 s8, 5 -; GFX10-NEXT: v_writelane_b32 v40, s9, 5 ; GFX10-NEXT: s_mov_b32 s9, 6 -; GFX10-NEXT: v_writelane_b32 v40, s10, 6 ; GFX10-NEXT: s_mov_b32 s10, 7 -; GFX10-NEXT: v_writelane_b32 v40, s11, 7 ; GFX10-NEXT: s_mov_b32 s11, 8 -; GFX10-NEXT: v_writelane_b32 v40, s30, 8 -; GFX10-NEXT: v_writelane_b32 v40, s31, 9 +; GFX10-NEXT: s_getpc_b64 s[34:35] +; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v8i32_inreg@rel32@lo+4 +; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v8i32_inreg@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 9 ; GFX10-NEXT: v_readlane_b32 s30, v40, 8 +; GFX10-NEXT: v_readlane_b32 s31, v40, 9 ; GFX10-NEXT: v_readlane_b32 s11, v40, 7 ; GFX10-NEXT: v_readlane_b32 s10, v40, 6 ; GFX10-NEXT: v_readlane_b32 s9, v40, 5 @@ -11600,32 +11607,32 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 10 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_mov_b32 s4, 1 ; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: s_getpc_b64 s[0:1] -; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v8i32_inreg@rel32@lo+4 -; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v8i32_inreg@rel32@hi+12 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 -; GFX11-NEXT: s_mov_b32 s5, 2 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 -; GFX11-NEXT: s_mov_b32 s6, 3 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 -; GFX11-NEXT: s_mov_b32 s7, 4 ; GFX11-NEXT: v_writelane_b32 v40, s8, 4 -; GFX11-NEXT: s_mov_b32 s8, 5 ; GFX11-NEXT: v_writelane_b32 v40, s9, 5 -; GFX11-NEXT: s_mov_b32 s9, 6 ; GFX11-NEXT: v_writelane_b32 v40, s10, 6 -; GFX11-NEXT: s_mov_b32 s10, 7 ; GFX11-NEXT: v_writelane_b32 v40, s11, 7 -; GFX11-NEXT: s_mov_b32 s11, 8 ; GFX11-NEXT: v_writelane_b32 v40, s30, 8 ; GFX11-NEXT: v_writelane_b32 v40, s31, 9 +; GFX11-NEXT: s_mov_b32 s4, 1 +; GFX11-NEXT: s_mov_b32 s5, 2 +; GFX11-NEXT: s_mov_b32 s6, 3 +; GFX11-NEXT: s_mov_b32 s7, 4 +; GFX11-NEXT: s_mov_b32 s8, 5 +; GFX11-NEXT: s_mov_b32 s9, 6 +; GFX11-NEXT: s_mov_b32 s10, 7 +; GFX11-NEXT: s_mov_b32 s11, 8 +; GFX11-NEXT: s_getpc_b64 s[0:1] +; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v8i32_inreg@rel32@lo+4 +; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v8i32_inreg@rel32@hi+12 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 9 ; GFX11-NEXT: v_readlane_b32 s30, v40, 8 +; GFX11-NEXT: v_readlane_b32 s31, v40, 9 ; GFX11-NEXT: v_readlane_b32 s11, v40, 7 ; GFX11-NEXT: v_readlane_b32 s10, v40, 6 ; GFX11-NEXT: v_readlane_b32 s9, v40, 5 @@ -11652,31 +11659,31 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 10 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] -; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v8i32_inreg@rel32@lo+4 -; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v8i32_inreg@rel32@hi+12 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 -; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 -; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 3 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 -; GFX10-SCRATCH-NEXT: s_mov_b32 s7, 4 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s8, 4 -; GFX10-SCRATCH-NEXT: s_mov_b32 s8, 5 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s9, 5 -; GFX10-SCRATCH-NEXT: s_mov_b32 s9, 6 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s10, 6 -; GFX10-SCRATCH-NEXT: s_mov_b32 s10, 7 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s11, 7 -; GFX10-SCRATCH-NEXT: s_mov_b32 s11, 8 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 8 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 9 +; GFX10-SCRATCH-NEXT: s_mov_b32 s4, 1 +; GFX10-SCRATCH-NEXT: s_mov_b32 s5, 2 +; GFX10-SCRATCH-NEXT: s_mov_b32 s6, 3 +; GFX10-SCRATCH-NEXT: s_mov_b32 s7, 4 +; GFX10-SCRATCH-NEXT: s_mov_b32 s8, 5 +; GFX10-SCRATCH-NEXT: s_mov_b32 s9, 6 +; GFX10-SCRATCH-NEXT: s_mov_b32 s10, 7 +; GFX10-SCRATCH-NEXT: s_mov_b32 s11, 8 +; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] +; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v8i32_inreg@rel32@lo+4 +; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v8i32_inreg@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 9 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 8 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 9 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s11, v40, 7 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s10, v40, 6 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s9, v40, 5 @@ -11706,6 +11713,7 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 18 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 @@ -11717,23 +11725,22 @@ ; GFX9-NEXT: v_writelane_b32 v40, s12, 8 ; GFX9-NEXT: v_writelane_b32 v40, s13, 9 ; GFX9-NEXT: v_writelane_b32 v40, s14, 10 -; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX9-NEXT: v_writelane_b32 v40, s15, 11 ; GFX9-NEXT: v_writelane_b32 v40, s16, 12 ; GFX9-NEXT: v_writelane_b32 v40, s17, 13 ; GFX9-NEXT: v_writelane_b32 v40, s18, 14 ; GFX9-NEXT: v_writelane_b32 v40, s19, 15 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: s_load_dwordx16 s[4:19], s[34:35], 0x0 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 16 ; GFX9-NEXT: v_writelane_b32 v40, s31, 17 +; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_load_dwordx16 s[4:19], s[34:35], 0x0 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v16i32_inreg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v16i32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 17 ; GFX9-NEXT: v_readlane_b32 s30, v40, 16 +; GFX9-NEXT: v_readlane_b32 s31, v40, 17 ; GFX9-NEXT: v_readlane_b32 s19, v40, 15 ; GFX9-NEXT: v_readlane_b32 s18, v40, 14 ; GFX9-NEXT: v_readlane_b32 s17, v40, 13 @@ -11768,9 +11775,8 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 18 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 @@ -11786,16 +11792,17 @@ ; GFX10-NEXT: v_writelane_b32 v40, s17, 13 ; GFX10-NEXT: v_writelane_b32 v40, s18, 14 ; GFX10-NEXT: v_writelane_b32 v40, s19, 15 +; GFX10-NEXT: v_writelane_b32 v40, s30, 16 +; GFX10-NEXT: v_writelane_b32 v40, s31, 17 +; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_load_dwordx16 s[4:19], s[34:35], 0x0 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v16i32_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v16i32_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 16 -; GFX10-NEXT: v_writelane_b32 v40, s31, 17 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 17 ; GFX10-NEXT: v_readlane_b32 s30, v40, 16 +; GFX10-NEXT: v_readlane_b32 s31, v40, 17 ; GFX10-NEXT: v_readlane_b32 s19, v40, 15 ; GFX10-NEXT: v_readlane_b32 s18, v40, 14 ; GFX10-NEXT: v_readlane_b32 s17, v40, 13 @@ -11830,9 +11837,8 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 18 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 @@ -11848,17 +11854,18 @@ ; GFX11-NEXT: v_writelane_b32 v40, s17, 13 ; GFX11-NEXT: v_writelane_b32 v40, s18, 14 ; GFX11-NEXT: v_writelane_b32 v40, s19, 15 +; GFX11-NEXT: v_writelane_b32 v40, s30, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 17 +; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_load_b512 s[4:19], s[0:1], 0x0 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v16i32_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v16i32_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 16 -; GFX11-NEXT: v_writelane_b32 v40, s31, 17 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 17 ; GFX11-NEXT: v_readlane_b32 s30, v40, 16 +; GFX11-NEXT: v_readlane_b32 s31, v40, 17 ; GFX11-NEXT: v_readlane_b32 s19, v40, 15 ; GFX11-NEXT: v_readlane_b32 s18, v40, 14 ; GFX11-NEXT: v_readlane_b32 s17, v40, 13 @@ -11893,9 +11900,8 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 18 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 @@ -11911,16 +11917,17 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s17, 13 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s18, 14 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s19, 15 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 17 +; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-SCRATCH-NEXT: s_load_dwordx16 s[4:19], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v16i32_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v16i32_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 17 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 17 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 16 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 17 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s19, v40, 15 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s18, v40, 14 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s17, v40, 13 @@ -11960,6 +11967,7 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 28 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 @@ -11976,23 +11984,27 @@ ; GFX9-NEXT: v_writelane_b32 v40, s17, 13 ; GFX9-NEXT: v_writelane_b32 v40, s18, 14 ; GFX9-NEXT: v_writelane_b32 v40, s19, 15 -; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX9-NEXT: v_writelane_b32 v40, s20, 16 ; GFX9-NEXT: v_writelane_b32 v40, s21, 17 ; GFX9-NEXT: v_writelane_b32 v40, s22, 18 ; GFX9-NEXT: v_writelane_b32 v40, s23, 19 ; GFX9-NEXT: v_writelane_b32 v40, s24, 20 -; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: s_load_dwordx16 s[36:51], s[34:35], 0x40 -; GFX9-NEXT: s_load_dwordx16 s[4:19], s[34:35], 0x0 ; GFX9-NEXT: v_writelane_b32 v40, s25, 21 ; GFX9-NEXT: v_writelane_b32 v40, s26, 22 ; GFX9-NEXT: v_writelane_b32 v40, s27, 23 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s28, 24 +; GFX9-NEXT: v_writelane_b32 v40, s29, 25 +; GFX9-NEXT: v_writelane_b32 v40, s30, 26 +; GFX9-NEXT: v_writelane_b32 v40, s31, 27 +; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_load_dwordx16 s[36:51], s[34:35], 0x40 +; GFX9-NEXT: s_load_dwordx16 s[4:19], s[34:35], 0x0 +; GFX9-NEXT: s_getpc_b64 s[34:35] +; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v32i32_inreg@rel32@lo+4 +; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v32i32_inreg@rel32@hi+12 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: v_mov_b32_e32 v0, s46 -; GFX9-NEXT: v_writelane_b32 v40, s29, 25 ; GFX9-NEXT: v_mov_b32_e32 v1, s47 ; GFX9-NEXT: v_mov_b32_e32 v2, s48 ; GFX9-NEXT: v_mov_b32_e32 v3, s49 @@ -12001,7 +12013,6 @@ ; GFX9-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:8 ; GFX9-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:12 ; GFX9-NEXT: v_mov_b32_e32 v0, s50 -; GFX9-NEXT: v_writelane_b32 v40, s30, 26 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:16 ; GFX9-NEXT: v_mov_b32_e32 v0, s51 ; GFX9-NEXT: s_mov_b32 s20, s36 @@ -12014,14 +12025,10 @@ ; GFX9-NEXT: s_mov_b32 s27, s43 ; GFX9-NEXT: s_mov_b32 s28, s44 ; GFX9-NEXT: s_mov_b32 s29, s45 -; GFX9-NEXT: v_writelane_b32 v40, s31, 27 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:20 -; GFX9-NEXT: s_getpc_b64 s[34:35] -; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v32i32_inreg@rel32@lo+4 -; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v32i32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 27 ; GFX9-NEXT: v_readlane_b32 s30, v40, 26 +; GFX9-NEXT: v_readlane_b32 s31, v40, 27 ; GFX9-NEXT: v_readlane_b32 s29, v40, 25 ; GFX9-NEXT: v_readlane_b32 s28, v40, 24 ; GFX9-NEXT: v_readlane_b32 s27, v40, 23 @@ -12066,9 +12073,8 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 28 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 @@ -12084,6 +12090,19 @@ ; GFX10-NEXT: v_writelane_b32 v40, s17, 13 ; GFX10-NEXT: v_writelane_b32 v40, s18, 14 ; GFX10-NEXT: v_writelane_b32 v40, s19, 15 +; GFX10-NEXT: v_writelane_b32 v40, s20, 16 +; GFX10-NEXT: v_writelane_b32 v40, s21, 17 +; GFX10-NEXT: v_writelane_b32 v40, s22, 18 +; GFX10-NEXT: v_writelane_b32 v40, s23, 19 +; GFX10-NEXT: v_writelane_b32 v40, s24, 20 +; GFX10-NEXT: v_writelane_b32 v40, s25, 21 +; GFX10-NEXT: v_writelane_b32 v40, s26, 22 +; GFX10-NEXT: v_writelane_b32 v40, s27, 23 +; GFX10-NEXT: v_writelane_b32 v40, s28, 24 +; GFX10-NEXT: v_writelane_b32 v40, s29, 25 +; GFX10-NEXT: v_writelane_b32 v40, s30, 26 +; GFX10-NEXT: v_writelane_b32 v40, s31, 27 +; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_clause 0x1 ; GFX10-NEXT: s_load_dwordx16 s[36:51], s[34:35], 0x40 @@ -12091,23 +12110,21 @@ ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v32i32_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v32i32_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s20, 16 -; GFX10-NEXT: v_writelane_b32 v40, s21, 17 -; GFX10-NEXT: v_writelane_b32 v40, s22, 18 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: v_mov_b32_e32 v0, s46 -; GFX10-NEXT: v_writelane_b32 v40, s23, 19 ; GFX10-NEXT: v_mov_b32_e32 v1, s47 ; GFX10-NEXT: v_mov_b32_e32 v2, s48 ; GFX10-NEXT: v_mov_b32_e32 v3, s49 ; GFX10-NEXT: s_mov_b32 s20, s36 -; GFX10-NEXT: v_writelane_b32 v40, s24, 20 ; GFX10-NEXT: s_mov_b32 s21, s37 ; GFX10-NEXT: s_mov_b32 s22, s38 ; GFX10-NEXT: s_mov_b32 s23, s39 ; GFX10-NEXT: s_mov_b32 s24, s40 -; GFX10-NEXT: v_writelane_b32 v40, s25, 21 ; GFX10-NEXT: s_mov_b32 s25, s41 +; GFX10-NEXT: s_mov_b32 s26, s42 +; GFX10-NEXT: s_mov_b32 s27, s43 +; GFX10-NEXT: s_mov_b32 s28, s44 +; GFX10-NEXT: s_mov_b32 s29, s45 ; GFX10-NEXT: v_mov_b32_e32 v4, s50 ; GFX10-NEXT: v_mov_b32_e32 v5, s51 ; GFX10-NEXT: buffer_store_dword v0, off, s[0:3], s32 @@ -12116,19 +12133,9 @@ ; GFX10-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:12 ; GFX10-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:16 ; GFX10-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:20 -; GFX10-NEXT: v_writelane_b32 v40, s26, 22 -; GFX10-NEXT: s_mov_b32 s26, s42 -; GFX10-NEXT: v_writelane_b32 v40, s27, 23 -; GFX10-NEXT: s_mov_b32 s27, s43 -; GFX10-NEXT: v_writelane_b32 v40, s28, 24 -; GFX10-NEXT: s_mov_b32 s28, s44 -; GFX10-NEXT: v_writelane_b32 v40, s29, 25 -; GFX10-NEXT: s_mov_b32 s29, s45 -; GFX10-NEXT: v_writelane_b32 v40, s30, 26 -; GFX10-NEXT: v_writelane_b32 v40, s31, 27 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 27 ; GFX10-NEXT: v_readlane_b32 s30, v40, 26 +; GFX10-NEXT: v_readlane_b32 s31, v40, 27 ; GFX10-NEXT: v_readlane_b32 s29, v40, 25 ; GFX10-NEXT: v_readlane_b32 s28, v40, 24 ; GFX10-NEXT: v_readlane_b32 s27, v40, 23 @@ -12173,9 +12180,8 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 28 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 @@ -12191,6 +12197,19 @@ ; GFX11-NEXT: v_writelane_b32 v40, s17, 13 ; GFX11-NEXT: v_writelane_b32 v40, s18, 14 ; GFX11-NEXT: v_writelane_b32 v40, s19, 15 +; GFX11-NEXT: v_writelane_b32 v40, s20, 16 +; GFX11-NEXT: v_writelane_b32 v40, s21, 17 +; GFX11-NEXT: v_writelane_b32 v40, s22, 18 +; GFX11-NEXT: v_writelane_b32 v40, s23, 19 +; GFX11-NEXT: v_writelane_b32 v40, s24, 20 +; GFX11-NEXT: v_writelane_b32 v40, s25, 21 +; GFX11-NEXT: v_writelane_b32 v40, s26, 22 +; GFX11-NEXT: v_writelane_b32 v40, s27, 23 +; GFX11-NEXT: v_writelane_b32 v40, s28, 24 +; GFX11-NEXT: v_writelane_b32 v40, s29, 25 +; GFX11-NEXT: v_writelane_b32 v40, s30, 26 +; GFX11-NEXT: v_writelane_b32 v40, s31, 27 +; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: s_load_b512 s[36:51], s[0:1], 0x40 @@ -12198,39 +12217,26 @@ ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v32i32_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v32i32_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s20, 16 -; GFX11-NEXT: v_writelane_b32 v40, s21, 17 -; GFX11-NEXT: v_writelane_b32 v40, s22, 18 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: v_dual_mov_b32 v4, s50 :: v_dual_mov_b32 v5, s51 -; GFX11-NEXT: v_writelane_b32 v40, s23, 19 ; GFX11-NEXT: v_dual_mov_b32 v0, s46 :: v_dual_mov_b32 v1, s47 ; GFX11-NEXT: v_dual_mov_b32 v2, s48 :: v_dual_mov_b32 v3, s49 -; GFX11-NEXT: v_writelane_b32 v40, s24, 20 ; GFX11-NEXT: s_mov_b32 s20, s36 ; GFX11-NEXT: s_mov_b32 s21, s37 ; GFX11-NEXT: s_mov_b32 s22, s38 ; GFX11-NEXT: s_mov_b32 s23, s39 -; GFX11-NEXT: v_writelane_b32 v40, s25, 21 ; GFX11-NEXT: s_mov_b32 s24, s40 ; GFX11-NEXT: s_mov_b32 s25, s41 -; GFX11-NEXT: s_clause 0x1 -; GFX11-NEXT: scratch_store_b64 off, v[4:5], s32 offset:16 -; GFX11-NEXT: scratch_store_b128 off, v[0:3], s32 -; GFX11-NEXT: v_writelane_b32 v40, s26, 22 ; GFX11-NEXT: s_mov_b32 s26, s42 -; GFX11-NEXT: v_writelane_b32 v40, s27, 23 ; GFX11-NEXT: s_mov_b32 s27, s43 -; GFX11-NEXT: v_writelane_b32 v40, s28, 24 ; GFX11-NEXT: s_mov_b32 s28, s44 -; GFX11-NEXT: v_writelane_b32 v40, s29, 25 ; GFX11-NEXT: s_mov_b32 s29, s45 -; GFX11-NEXT: v_writelane_b32 v40, s30, 26 -; GFX11-NEXT: v_writelane_b32 v40, s31, 27 +; GFX11-NEXT: s_clause 0x1 +; GFX11-NEXT: scratch_store_b64 off, v[4:5], s32 offset:16 +; GFX11-NEXT: scratch_store_b128 off, v[0:3], s32 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 27 ; GFX11-NEXT: v_readlane_b32 s30, v40, 26 +; GFX11-NEXT: v_readlane_b32 s31, v40, 27 ; GFX11-NEXT: v_readlane_b32 s29, v40, 25 ; GFX11-NEXT: v_readlane_b32 s28, v40, 24 ; GFX11-NEXT: v_readlane_b32 s27, v40, 23 @@ -12275,9 +12281,8 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 28 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 @@ -12293,6 +12298,19 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s17, 13 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s18, 14 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s19, 15 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s20, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s21, 17 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s22, 18 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s23, 19 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s24, 20 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s25, 21 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s26, 22 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s27, 23 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s28, 24 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s29, 25 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 26 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 27 +; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-SCRATCH-NEXT: s_clause 0x1 ; GFX10-SCRATCH-NEXT: s_load_dwordx16 s[36:51], s[0:1], 0x40 @@ -12300,40 +12318,28 @@ ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v32i32_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v32i32_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s20, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s21, 17 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s22, 18 ; GFX10-SCRATCH-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v4, s50 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s23, 19 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v5, s51 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, s46 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, s47 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, s48 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s24, 20 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v3, s49 ; GFX10-SCRATCH-NEXT: s_mov_b32 s20, s36 ; GFX10-SCRATCH-NEXT: s_mov_b32 s21, s37 ; GFX10-SCRATCH-NEXT: s_mov_b32 s22, s38 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s25, 21 ; GFX10-SCRATCH-NEXT: s_mov_b32 s23, s39 ; GFX10-SCRATCH-NEXT: s_mov_b32 s24, s40 ; GFX10-SCRATCH-NEXT: s_mov_b32 s25, s41 -; GFX10-SCRATCH-NEXT: scratch_store_dwordx2 off, v[4:5], s32 offset:16 -; GFX10-SCRATCH-NEXT: scratch_store_dwordx4 off, v[0:3], s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s26, 22 ; GFX10-SCRATCH-NEXT: s_mov_b32 s26, s42 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s27, 23 ; GFX10-SCRATCH-NEXT: s_mov_b32 s27, s43 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s28, 24 ; GFX10-SCRATCH-NEXT: s_mov_b32 s28, s44 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s29, 25 ; GFX10-SCRATCH-NEXT: s_mov_b32 s29, s45 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 26 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 27 +; GFX10-SCRATCH-NEXT: scratch_store_dwordx2 off, v[4:5], s32 offset:16 +; GFX10-SCRATCH-NEXT: scratch_store_dwordx4 off, v[0:3], s32 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 27 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 26 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 27 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s29, v40, 25 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s28, v40, 24 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s27, v40, 23 @@ -12383,6 +12389,7 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 28 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 ; GFX9-NEXT: v_writelane_b32 v40, s6, 2 @@ -12398,38 +12405,41 @@ ; GFX9-NEXT: v_writelane_b32 v40, s16, 12 ; GFX9-NEXT: v_writelane_b32 v40, s17, 13 ; GFX9-NEXT: v_writelane_b32 v40, s18, 14 -; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX9-NEXT: v_writelane_b32 v40, s19, 15 ; GFX9-NEXT: v_writelane_b32 v40, s20, 16 ; GFX9-NEXT: v_writelane_b32 v40, s21, 17 ; GFX9-NEXT: v_writelane_b32 v40, s22, 18 ; GFX9-NEXT: v_writelane_b32 v40, s23, 19 +; GFX9-NEXT: v_writelane_b32 v40, s24, 20 +; GFX9-NEXT: v_writelane_b32 v40, s25, 21 +; GFX9-NEXT: v_writelane_b32 v40, s26, 22 +; GFX9-NEXT: v_writelane_b32 v40, s27, 23 +; GFX9-NEXT: v_writelane_b32 v40, s28, 24 +; GFX9-NEXT: v_writelane_b32 v40, s29, 25 +; GFX9-NEXT: v_writelane_b32 v40, s30, 26 +; GFX9-NEXT: v_writelane_b32 v40, s31, 27 +; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: s_load_dword s52, s[34:35], 0x0 ; GFX9-NEXT: ; kill: killed $sgpr34_sgpr35 ; GFX9-NEXT: ; kill: killed $sgpr34_sgpr35 ; GFX9-NEXT: s_load_dwordx16 s[36:51], s[34:35], 0x40 ; GFX9-NEXT: s_load_dwordx16 s[4:19], s[34:35], 0x0 -; GFX9-NEXT: v_writelane_b32 v40, s24, 20 -; GFX9-NEXT: v_writelane_b32 v40, s25, 21 -; GFX9-NEXT: s_addk_i32 s32, 0x400 -; GFX9-NEXT: v_writelane_b32 v40, s26, 22 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: v_mov_b32_e32 v0, s52 -; GFX9-NEXT: v_writelane_b32 v40, s27, 23 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:24 +; GFX9-NEXT: s_getpc_b64 s[34:35] +; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v32i32_i32_inreg@rel32@lo+4 +; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v32i32_i32_inreg@rel32@hi+12 ; GFX9-NEXT: v_mov_b32_e32 v0, s46 -; GFX9-NEXT: v_writelane_b32 v40, s28, 24 ; GFX9-NEXT: v_mov_b32_e32 v1, s47 ; GFX9-NEXT: v_mov_b32_e32 v2, s48 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 ; GFX9-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:4 ; GFX9-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:8 ; GFX9-NEXT: v_mov_b32_e32 v0, s49 -; GFX9-NEXT: v_writelane_b32 v40, s29, 25 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:12 ; GFX9-NEXT: v_mov_b32_e32 v0, s50 -; GFX9-NEXT: v_writelane_b32 v40, s30, 26 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:16 ; GFX9-NEXT: v_mov_b32_e32 v0, s51 ; GFX9-NEXT: s_mov_b32 s20, s36 @@ -12442,14 +12452,10 @@ ; GFX9-NEXT: s_mov_b32 s27, s43 ; GFX9-NEXT: s_mov_b32 s28, s44 ; GFX9-NEXT: s_mov_b32 s29, s45 -; GFX9-NEXT: v_writelane_b32 v40, s31, 27 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:20 -; GFX9-NEXT: s_getpc_b64 s[34:35] -; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_v32i32_i32_inreg@rel32@lo+4 -; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_v32i32_i32_inreg@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 27 ; GFX9-NEXT: v_readlane_b32 s30, v40, 26 +; GFX9-NEXT: v_readlane_b32 s31, v40, 27 ; GFX9-NEXT: v_readlane_b32 s29, v40, 25 ; GFX9-NEXT: v_readlane_b32 s28, v40, 24 ; GFX9-NEXT: v_readlane_b32 s27, v40, 23 @@ -12494,9 +12500,8 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 28 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-NEXT: v_writelane_b32 v40, s7, 3 @@ -12512,6 +12517,19 @@ ; GFX10-NEXT: v_writelane_b32 v40, s17, 13 ; GFX10-NEXT: v_writelane_b32 v40, s18, 14 ; GFX10-NEXT: v_writelane_b32 v40, s19, 15 +; GFX10-NEXT: v_writelane_b32 v40, s20, 16 +; GFX10-NEXT: v_writelane_b32 v40, s21, 17 +; GFX10-NEXT: v_writelane_b32 v40, s22, 18 +; GFX10-NEXT: v_writelane_b32 v40, s23, 19 +; GFX10-NEXT: v_writelane_b32 v40, s24, 20 +; GFX10-NEXT: v_writelane_b32 v40, s25, 21 +; GFX10-NEXT: v_writelane_b32 v40, s26, 22 +; GFX10-NEXT: v_writelane_b32 v40, s27, 23 +; GFX10-NEXT: v_writelane_b32 v40, s28, 24 +; GFX10-NEXT: v_writelane_b32 v40, s29, 25 +; GFX10-NEXT: v_writelane_b32 v40, s30, 26 +; GFX10-NEXT: v_writelane_b32 v40, s31, 27 +; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_clause 0x2 ; GFX10-NEXT: s_load_dword s52, s[34:35], 0x0 @@ -12522,46 +12540,34 @@ ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_v32i32_i32_inreg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_v32i32_i32_inreg@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s20, 16 -; GFX10-NEXT: v_writelane_b32 v40, s21, 17 -; GFX10-NEXT: v_writelane_b32 v40, s22, 18 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: v_mov_b32_e32 v0, s52 ; GFX10-NEXT: v_mov_b32_e32 v1, s47 -; GFX10-NEXT: v_writelane_b32 v40, s23, 19 +; GFX10-NEXT: v_mov_b32_e32 v2, s48 ; GFX10-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:24 ; GFX10-NEXT: v_mov_b32_e32 v0, s46 -; GFX10-NEXT: v_mov_b32_e32 v2, s48 ; GFX10-NEXT: v_mov_b32_e32 v3, s49 -; GFX10-NEXT: v_writelane_b32 v40, s24, 20 ; GFX10-NEXT: s_mov_b32 s20, s36 ; GFX10-NEXT: s_mov_b32 s21, s37 ; GFX10-NEXT: s_mov_b32 s22, s38 ; GFX10-NEXT: s_mov_b32 s23, s39 -; GFX10-NEXT: v_writelane_b32 v40, s25, 21 ; GFX10-NEXT: s_mov_b32 s24, s40 ; GFX10-NEXT: s_mov_b32 s25, s41 +; GFX10-NEXT: s_mov_b32 s26, s42 +; GFX10-NEXT: s_mov_b32 s27, s43 +; GFX10-NEXT: s_mov_b32 s28, s44 +; GFX10-NEXT: s_mov_b32 s29, s45 ; GFX10-NEXT: v_mov_b32_e32 v4, s50 ; GFX10-NEXT: v_mov_b32_e32 v5, s51 -; GFX10-NEXT: v_writelane_b32 v40, s26, 22 -; GFX10-NEXT: s_mov_b32 s26, s42 ; GFX10-NEXT: buffer_store_dword v0, off, s[0:3], s32 ; GFX10-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:4 ; GFX10-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:8 ; GFX10-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:12 ; GFX10-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:16 ; GFX10-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:20 -; GFX10-NEXT: v_writelane_b32 v40, s27, 23 -; GFX10-NEXT: s_mov_b32 s27, s43 -; GFX10-NEXT: v_writelane_b32 v40, s28, 24 -; GFX10-NEXT: s_mov_b32 s28, s44 -; GFX10-NEXT: v_writelane_b32 v40, s29, 25 -; GFX10-NEXT: s_mov_b32 s29, s45 -; GFX10-NEXT: v_writelane_b32 v40, s30, 26 -; GFX10-NEXT: v_writelane_b32 v40, s31, 27 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 27 ; GFX10-NEXT: v_readlane_b32 s30, v40, 26 +; GFX10-NEXT: v_readlane_b32 s31, v40, 27 ; GFX10-NEXT: v_readlane_b32 s29, v40, 25 ; GFX10-NEXT: v_readlane_b32 s28, v40, 24 ; GFX10-NEXT: v_readlane_b32 s27, v40, 23 @@ -12606,9 +12612,8 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 28 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 -; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 ; GFX11-NEXT: v_writelane_b32 v40, s6, 2 ; GFX11-NEXT: v_writelane_b32 v40, s7, 3 @@ -12624,6 +12629,19 @@ ; GFX11-NEXT: v_writelane_b32 v40, s17, 13 ; GFX11-NEXT: v_writelane_b32 v40, s18, 14 ; GFX11-NEXT: v_writelane_b32 v40, s19, 15 +; GFX11-NEXT: v_writelane_b32 v40, s20, 16 +; GFX11-NEXT: v_writelane_b32 v40, s21, 17 +; GFX11-NEXT: v_writelane_b32 v40, s22, 18 +; GFX11-NEXT: v_writelane_b32 v40, s23, 19 +; GFX11-NEXT: v_writelane_b32 v40, s24, 20 +; GFX11-NEXT: v_writelane_b32 v40, s25, 21 +; GFX11-NEXT: v_writelane_b32 v40, s26, 22 +; GFX11-NEXT: v_writelane_b32 v40, s27, 23 +; GFX11-NEXT: v_writelane_b32 v40, s28, 24 +; GFX11-NEXT: v_writelane_b32 v40, s29, 25 +; GFX11-NEXT: v_writelane_b32 v40, s30, 26 +; GFX11-NEXT: v_writelane_b32 v40, s31, 27 +; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_clause 0x2 ; GFX11-NEXT: s_load_b32 s2, s[0:1], 0x0 @@ -12632,41 +12650,28 @@ ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_v32i32_i32_inreg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_v32i32_i32_inreg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s20, 16 -; GFX11-NEXT: v_writelane_b32 v40, s21, 17 -; GFX11-NEXT: v_writelane_b32 v40, s22, 18 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: v_dual_mov_b32 v6, s2 :: v_dual_mov_b32 v5, s51 -; GFX11-NEXT: v_writelane_b32 v40, s23, 19 ; GFX11-NEXT: v_dual_mov_b32 v4, s50 :: v_dual_mov_b32 v1, s47 ; GFX11-NEXT: v_dual_mov_b32 v0, s46 :: v_dual_mov_b32 v3, s49 -; GFX11-NEXT: v_writelane_b32 v40, s24, 20 ; GFX11-NEXT: v_mov_b32_e32 v2, s48 ; GFX11-NEXT: s_mov_b32 s20, s36 ; GFX11-NEXT: s_mov_b32 s21, s37 ; GFX11-NEXT: s_mov_b32 s22, s38 -; GFX11-NEXT: v_writelane_b32 v40, s25, 21 ; GFX11-NEXT: s_mov_b32 s23, s39 ; GFX11-NEXT: s_mov_b32 s24, s40 ; GFX11-NEXT: s_mov_b32 s25, s41 -; GFX11-NEXT: s_clause 0x2 -; GFX11-NEXT: scratch_store_b32 off, v6, s32 offset:24 -; GFX11-NEXT: scratch_store_b64 off, v[4:5], s32 offset:16 -; GFX11-NEXT: scratch_store_b128 off, v[0:3], s32 -; GFX11-NEXT: v_writelane_b32 v40, s26, 22 ; GFX11-NEXT: s_mov_b32 s26, s42 -; GFX11-NEXT: v_writelane_b32 v40, s27, 23 ; GFX11-NEXT: s_mov_b32 s27, s43 -; GFX11-NEXT: v_writelane_b32 v40, s28, 24 ; GFX11-NEXT: s_mov_b32 s28, s44 -; GFX11-NEXT: v_writelane_b32 v40, s29, 25 ; GFX11-NEXT: s_mov_b32 s29, s45 -; GFX11-NEXT: v_writelane_b32 v40, s30, 26 -; GFX11-NEXT: v_writelane_b32 v40, s31, 27 +; GFX11-NEXT: s_clause 0x2 +; GFX11-NEXT: scratch_store_b32 off, v6, s32 offset:24 +; GFX11-NEXT: scratch_store_b64 off, v[4:5], s32 offset:16 +; GFX11-NEXT: scratch_store_b128 off, v[0:3], s32 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 27 ; GFX11-NEXT: v_readlane_b32 s30, v40, 26 +; GFX11-NEXT: v_readlane_b32 s31, v40, 27 ; GFX11-NEXT: v_readlane_b32 s29, v40, 25 ; GFX11-NEXT: v_readlane_b32 s28, v40, 24 ; GFX11-NEXT: v_readlane_b32 s27, v40, 23 @@ -12711,9 +12716,8 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 28 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 -; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s5, 1 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s6, 2 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s7, 3 @@ -12729,6 +12733,19 @@ ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s17, 13 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s18, 14 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s19, 15 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s20, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s21, 17 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s22, 18 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s23, 19 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s24, 20 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s25, 21 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s26, 22 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s27, 23 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s28, 24 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s29, 25 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 26 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 27 +; GFX10-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0 ; GFX10-SCRATCH-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-SCRATCH-NEXT: s_clause 0x2 ; GFX10-SCRATCH-NEXT: s_load_dword s2, s[0:1], 0x0 @@ -12739,42 +12756,30 @@ ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_v32i32_i32_inreg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_v32i32_i32_inreg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s20, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s21, 17 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s22, 18 ; GFX10-SCRATCH-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v6, s2 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v4, s50 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s23, 19 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v5, s51 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, s46 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, s47 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, s48 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s24, 20 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v3, s49 ; GFX10-SCRATCH-NEXT: s_mov_b32 s20, s36 ; GFX10-SCRATCH-NEXT: s_mov_b32 s21, s37 ; GFX10-SCRATCH-NEXT: s_mov_b32 s22, s38 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s25, 21 ; GFX10-SCRATCH-NEXT: s_mov_b32 s23, s39 ; GFX10-SCRATCH-NEXT: s_mov_b32 s24, s40 ; GFX10-SCRATCH-NEXT: s_mov_b32 s25, s41 -; GFX10-SCRATCH-NEXT: scratch_store_dword off, v6, s32 offset:24 -; GFX10-SCRATCH-NEXT: scratch_store_dwordx2 off, v[4:5], s32 offset:16 -; GFX10-SCRATCH-NEXT: scratch_store_dwordx4 off, v[0:3], s32 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s26, 22 ; GFX10-SCRATCH-NEXT: s_mov_b32 s26, s42 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s27, 23 ; GFX10-SCRATCH-NEXT: s_mov_b32 s27, s43 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s28, 24 ; GFX10-SCRATCH-NEXT: s_mov_b32 s28, s44 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s29, 25 ; GFX10-SCRATCH-NEXT: s_mov_b32 s29, s45 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 26 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 27 +; GFX10-SCRATCH-NEXT: scratch_store_dword off, v6, s32 offset:24 +; GFX10-SCRATCH-NEXT: scratch_store_dwordx2 off, v[4:5], s32 offset:16 +; GFX10-SCRATCH-NEXT: scratch_store_dwordx4 off, v[0:3], s32 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 27 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 26 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 27 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s29, v40, 25 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s28, v40, 24 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s27, v40, 23 @@ -12825,11 +12830,11 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: buffer_load_dword v32, off, s[0:3], s33 -; GFX9-NEXT: buffer_load_dword v33, off, s[0:3], s33 offset:4 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 +; GFX9-NEXT: buffer_load_dword v32, off, s[0:3], s33 +; GFX9-NEXT: buffer_load_dword v33, off, s[0:3], s33 offset:4 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, stack_passed_f64_arg@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, stack_passed_f64_arg@rel32@hi+12 @@ -12838,8 +12843,8 @@ ; GFX9-NEXT: s_waitcnt vmcnt(1) ; GFX9-NEXT: buffer_store_dword v33, off, s[0:3], s32 offset:4 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -12858,11 +12863,12 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_clause 0x1 ; GFX10-NEXT: buffer_load_dword v32, off, s[0:3], s33 ; GFX10-NEXT: buffer_load_dword v33, off, s[0:3], s33 offset:4 -; GFX10-NEXT: v_writelane_b32 v40, s30, 0 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, stack_passed_f64_arg@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, stack_passed_f64_arg@rel32@hi+12 @@ -12870,10 +12876,9 @@ ; GFX10-NEXT: buffer_store_dword v32, off, s[0:3], s32 ; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: buffer_store_dword v33, off, s[0:3], s32 offset:4 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -12892,18 +12897,18 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: scratch_load_b64 v[32:33], off, s33 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: scratch_load_b64 v[32:33], off, s33 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, stack_passed_f64_arg@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, stack_passed_f64_arg@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: scratch_store_b64 off, v[32:33], s32 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -12922,18 +12927,18 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 -; GFX10-SCRATCH-NEXT: scratch_load_dwordx2 v[32:33], off, s33 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 +; GFX10-SCRATCH-NEXT: scratch_load_dwordx2 v[32:33], off, s33 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, stack_passed_f64_arg@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, stack_passed_f64_arg@rel32@hi+12 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_waitcnt vmcnt(0) ; GFX10-SCRATCH-NEXT: scratch_store_dwordx2 off, v[32:33], s32 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -12956,7 +12961,9 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: v_writelane_b32 v40, s30, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 12 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 ; GFX9-NEXT: v_mov_b32_e32 v0, 13 @@ -12964,7 +12971,6 @@ ; GFX9-NEXT: v_mov_b32_e32 v0, 14 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:8 ; GFX9-NEXT: v_mov_b32_e32 v0, 15 -; GFX9-NEXT: v_writelane_b32 v40, s30, 0 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:12 ; GFX9-NEXT: v_mov_b32_e32 v0, 0 ; GFX9-NEXT: v_mov_b32_e32 v1, 0 @@ -12998,13 +13004,12 @@ ; GFX9-NEXT: v_mov_b32_e32 v29, 9 ; GFX9-NEXT: v_mov_b32_e32 v30, 10 ; GFX9-NEXT: v_mov_b32_e32 v31, 11 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_12xv3i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_12xv3i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -13023,12 +13028,14 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 12 ; GFX10-NEXT: v_mov_b32_e32 v1, 13 ; GFX10-NEXT: v_mov_b32_e32 v2, 14 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_mov_b32_e32 v3, 15 -; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: v_mov_b32_e32 v4, 1 ; GFX10-NEXT: buffer_store_dword v0, off, s[0:3], s32 ; GFX10-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:4 ; GFX10-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:8 @@ -13037,7 +13044,6 @@ ; GFX10-NEXT: v_mov_b32_e32 v1, 0 ; GFX10-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-NEXT: v_mov_b32_e32 v3, 1 -; GFX10-NEXT: v_mov_b32_e32 v4, 1 ; GFX10-NEXT: v_mov_b32_e32 v5, 1 ; GFX10-NEXT: v_mov_b32_e32 v6, 2 ; GFX10-NEXT: v_mov_b32_e32 v7, 2 @@ -13065,13 +13071,12 @@ ; GFX10-NEXT: v_mov_b32_e32 v29, 9 ; GFX10-NEXT: v_mov_b32_e32 v30, 10 ; GFX10-NEXT: v_mov_b32_e32 v31, 11 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_12xv3i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_12xv3i32@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -13090,15 +13095,16 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: v_writelane_b32 v40, s30, 0 +; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: v_dual_mov_b32 v0, 12 :: v_dual_mov_b32 v1, 13 ; GFX11-NEXT: v_dual_mov_b32 v2, 14 :: v_dual_mov_b32 v3, 15 -; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: v_dual_mov_b32 v4, 1 :: v_dual_mov_b32 v5, 1 +; GFX11-NEXT: v_dual_mov_b32 v6, 2 :: v_dual_mov_b32 v7, 2 ; GFX11-NEXT: scratch_store_b128 off, v[0:3], s32 ; GFX11-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, 0 ; GFX11-NEXT: v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v3, 1 -; GFX11-NEXT: v_dual_mov_b32 v6, 2 :: v_dual_mov_b32 v7, 2 ; GFX11-NEXT: v_dual_mov_b32 v8, 2 :: v_dual_mov_b32 v9, 3 ; GFX11-NEXT: v_dual_mov_b32 v10, 3 :: v_dual_mov_b32 v11, 3 ; GFX11-NEXT: v_dual_mov_b32 v12, 4 :: v_dual_mov_b32 v13, 4 @@ -13111,14 +13117,13 @@ ; GFX11-NEXT: v_dual_mov_b32 v26, 8 :: v_dual_mov_b32 v27, 9 ; GFX11-NEXT: v_dual_mov_b32 v28, 9 :: v_dual_mov_b32 v29, 9 ; GFX11-NEXT: v_dual_mov_b32 v30, 10 :: v_dual_mov_b32 v31, 11 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_12xv3i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_12xv3i32@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -13137,21 +13142,22 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 12 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 13 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 14 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v3, 15 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v4, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v5, 1 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v6, 2 +; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v7, 2 ; GFX10-SCRATCH-NEXT: scratch_store_dwordx4 off, v[0:3], s32 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v3, 1 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v6, 2 -; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v7, 2 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v8, 2 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v9, 3 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v10, 3 @@ -13176,13 +13182,12 @@ ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v29, 9 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v30, 10 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v31, 11 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_12xv3i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_12xv3i32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -13217,7 +13222,9 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: v_writelane_b32 v40, s30, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 8 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 ; GFX9-NEXT: v_mov_b32_e32 v0, 9 @@ -13233,7 +13240,6 @@ ; GFX9-NEXT: v_mov_b32_e32 v0, 14 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:24 ; GFX9-NEXT: v_mov_b32_e32 v0, 15 -; GFX9-NEXT: v_writelane_b32 v40, s30, 0 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:28 ; GFX9-NEXT: v_mov_b32_e32 v0, 0 ; GFX9-NEXT: v_mov_b32_e32 v1, 0 @@ -13267,13 +13273,12 @@ ; GFX9-NEXT: v_mov_b32_e32 v29, 5 ; GFX9-NEXT: v_mov_b32_e32 v30, 6 ; GFX9-NEXT: v_mov_b32_e32 v31, 7 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_8xv5i32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_8xv5i32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -13292,19 +13297,21 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 8 ; GFX10-NEXT: v_mov_b32_e32 v1, 9 ; GFX10-NEXT: v_mov_b32_e32 v2, 10 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_mov_b32_e32 v3, 14 +; GFX10-NEXT: v_mov_b32_e32 v4, 15 ; GFX10-NEXT: buffer_store_dword v0, off, s[0:3], s32 ; GFX10-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:4 ; GFX10-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:8 ; GFX10-NEXT: v_mov_b32_e32 v0, 11 ; GFX10-NEXT: v_mov_b32_e32 v1, 12 ; GFX10-NEXT: v_mov_b32_e32 v2, 13 -; GFX10-NEXT: v_mov_b32_e32 v4, 15 -; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: v_mov_b32_e32 v5, 1 ; GFX10-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:12 ; GFX10-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:16 ; GFX10-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:20 @@ -13315,7 +13322,6 @@ ; GFX10-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-NEXT: v_mov_b32_e32 v3, 0 ; GFX10-NEXT: v_mov_b32_e32 v4, 0 -; GFX10-NEXT: v_mov_b32_e32 v5, 1 ; GFX10-NEXT: v_mov_b32_e32 v6, 1 ; GFX10-NEXT: v_mov_b32_e32 v7, 1 ; GFX10-NEXT: v_mov_b32_e32 v8, 1 @@ -13342,13 +13348,12 @@ ; GFX10-NEXT: v_mov_b32_e32 v29, 5 ; GFX10-NEXT: v_mov_b32_e32 v30, 6 ; GFX10-NEXT: v_mov_b32_e32 v31, 7 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_8xv5i32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_8xv5i32@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -13367,12 +13372,13 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: v_writelane_b32 v40, s30, 0 +; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: v_dual_mov_b32 v0, 12 :: v_dual_mov_b32 v1, 13 ; GFX11-NEXT: v_dual_mov_b32 v2, 14 :: v_dual_mov_b32 v3, 15 ; GFX11-NEXT: v_dual_mov_b32 v4, 8 :: v_dual_mov_b32 v5, 9 ; GFX11-NEXT: v_dual_mov_b32 v6, 10 :: v_dual_mov_b32 v7, 11 -; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: scratch_store_b128 off, v[0:3], s32 offset:16 ; GFX11-NEXT: scratch_store_b128 off, v[4:7], s32 @@ -13392,14 +13398,13 @@ ; GFX11-NEXT: v_dual_mov_b32 v26, 5 :: v_dual_mov_b32 v27, 5 ; GFX11-NEXT: v_dual_mov_b32 v28, 5 :: v_dual_mov_b32 v29, 5 ; GFX11-NEXT: v_dual_mov_b32 v30, 6 :: v_dual_mov_b32 v31, 7 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_8xv5i32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_8xv5i32@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -13418,6 +13423,9 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 12 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 13 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 14 @@ -13426,8 +13434,6 @@ ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v5, 9 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v6, 10 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v7, 11 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: scratch_store_dwordx4 off, v[0:3], s32 offset:16 ; GFX10-SCRATCH-NEXT: scratch_store_dwordx4 off, v[4:7], s32 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0 @@ -13462,13 +13468,12 @@ ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v29, 5 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v30, 6 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v31, 7 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_8xv5i32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_8xv5i32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 @@ -13499,7 +13504,9 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: v_writelane_b32 v40, s30, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: v_mov_b32_e32 v0, 0x41000000 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 ; GFX9-NEXT: v_mov_b32_e32 v0, 0x41100000 @@ -13515,7 +13522,6 @@ ; GFX9-NEXT: v_mov_b32_e32 v0, 0x41600000 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:24 ; GFX9-NEXT: v_mov_b32_e32 v0, 0x41700000 -; GFX9-NEXT: v_writelane_b32 v40, s30, 0 ; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:28 ; GFX9-NEXT: v_mov_b32_e32 v0, 0 ; GFX9-NEXT: v_mov_b32_e32 v1, 0 @@ -13549,13 +13555,12 @@ ; GFX9-NEXT: v_mov_b32_e32 v29, 0x40a00000 ; GFX9-NEXT: v_mov_b32_e32 v30, 0x40c00000 ; GFX9-NEXT: v_mov_b32_e32 v31, 0x40e00000 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_8xv5f32@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_8xv5f32@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -13574,19 +13579,21 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v0, 0x41000000 ; GFX10-NEXT: v_mov_b32_e32 v1, 0x41100000 ; GFX10-NEXT: v_mov_b32_e32 v2, 0x41200000 -; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: v_mov_b32_e32 v3, 0x41600000 +; GFX10-NEXT: v_mov_b32_e32 v4, 0x41700000 ; GFX10-NEXT: buffer_store_dword v0, off, s[0:3], s32 ; GFX10-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:4 ; GFX10-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:8 ; GFX10-NEXT: v_mov_b32_e32 v0, 0x41300000 ; GFX10-NEXT: v_mov_b32_e32 v1, 0x41400000 ; GFX10-NEXT: v_mov_b32_e32 v2, 0x41500000 -; GFX10-NEXT: v_mov_b32_e32 v4, 0x41700000 -; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: v_mov_b32_e32 v5, 1.0 ; GFX10-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:12 ; GFX10-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:16 ; GFX10-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:20 @@ -13597,7 +13604,6 @@ ; GFX10-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-NEXT: v_mov_b32_e32 v3, 0 ; GFX10-NEXT: v_mov_b32_e32 v4, 0 -; GFX10-NEXT: v_mov_b32_e32 v5, 1.0 ; GFX10-NEXT: v_mov_b32_e32 v6, 1.0 ; GFX10-NEXT: v_mov_b32_e32 v7, 1.0 ; GFX10-NEXT: v_mov_b32_e32 v8, 1.0 @@ -13624,13 +13630,12 @@ ; GFX10-NEXT: v_mov_b32_e32 v29, 0x40a00000 ; GFX10-NEXT: v_mov_b32_e32 v30, 0x40c00000 ; GFX10-NEXT: v_mov_b32_e32 v31, 0x40e00000 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_8xv5f32@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_8xv5f32@rel32@hi+12 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -13649,6 +13654,9 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: v_writelane_b32 v40, s30, 0 +; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: v_mov_b32_e32 v0, 0x41400000 ; GFX11-NEXT: v_mov_b32_e32 v1, 0x41500000 ; GFX11-NEXT: v_mov_b32_e32 v2, 0x41600000 @@ -13657,8 +13665,6 @@ ; GFX11-NEXT: v_mov_b32_e32 v5, 0x41100000 ; GFX11-NEXT: v_mov_b32_e32 v6, 0x41200000 ; GFX11-NEXT: v_mov_b32_e32 v7, 0x41300000 -; GFX11-NEXT: s_add_i32 s32, s32, 16 -; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: scratch_store_b128 off, v[0:3], s32 offset:16 ; GFX11-NEXT: scratch_store_b128 off, v[4:7], s32 @@ -13680,14 +13686,13 @@ ; GFX11-NEXT: v_mov_b32_e32 v29, 0x40a00000 ; GFX11-NEXT: v_mov_b32_e32 v30, 0x40c00000 ; GFX11-NEXT: v_mov_b32_e32 v31, 0x40e00000 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_8xv5f32@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_8xv5f32@rel32@hi+12 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -13706,6 +13711,9 @@ ; GFX10-SCRATCH-NEXT: s_mov_b32 exec_lo, s0 ; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-SCRATCH-NEXT: s_mov_b32 s33, s32 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 +; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0x41400000 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v1, 0x41500000 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v2, 0x41600000 @@ -13714,8 +13722,6 @@ ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v5, 0x41100000 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v6, 0x41200000 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v7, 0x41300000 -; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, 16 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-SCRATCH-NEXT: scratch_store_dwordx4 off, v[0:3], s32 offset:16 ; GFX10-SCRATCH-NEXT: scratch_store_dwordx4 off, v[4:7], s32 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v0, 0 @@ -13750,13 +13756,12 @@ ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v29, 0x40a00000 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v30, 0x40c00000 ; GFX10-SCRATCH-NEXT: v_mov_b32_e32 v31, 0x40e00000 -; GFX10-SCRATCH-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-SCRATCH-NEXT: s_getpc_b64 s[0:1] ; GFX10-SCRATCH-NEXT: s_add_u32 s0, s0, external_void_func_8xv5f32@rel32@lo+4 ; GFX10-SCRATCH-NEXT: s_addc_u32 s1, s1, external_void_func_8xv5f32@rel32@hi+12 ; GFX10-SCRATCH-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-SCRATCH-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-SCRATCH-NEXT: s_add_i32 s32, s32, -16 ; GFX10-SCRATCH-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-SCRATCH-NEXT: s_or_saveexec_b32 s0, -1 diff --git a/llvm/test/CodeGen/AMDGPU/gfx-callable-preserved-registers.ll b/llvm/test/CodeGen/AMDGPU/gfx-callable-preserved-registers.ll --- a/llvm/test/CodeGen/AMDGPU/gfx-callable-preserved-registers.ll +++ b/llvm/test/CodeGen/AMDGPU/gfx-callable-preserved-registers.ll @@ -14,9 +14,9 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 4 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s5, 1 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 2 ; GFX9-NEXT: v_writelane_b32 v40, s31, 3 ; GFX9-NEXT: s_getpc_b64 s[4:5] @@ -26,8 +26,8 @@ ; GFX9-NEXT: ;;#ASMSTART ; GFX9-NEXT: ;;#ASMEND ; GFX9-NEXT: s_swappc_b64 s[30:31], s[4:5] -; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s30, v40, 2 +; GFX9-NEXT: v_readlane_b32 s31, v40, 3 ; GFX9-NEXT: v_readlane_b32 s5, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 @@ -48,20 +48,20 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 4 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: v_writelane_b32 v40, s5, 1 +; GFX10-NEXT: v_writelane_b32 v40, s30, 2 +; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_getpc_b64 s[4:5] ; GFX10-NEXT: s_add_u32 s4, s4, external_void_func_void@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 2 -; GFX10-NEXT: v_writelane_b32 v40, s31, 3 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[4:5] ; GFX10-NEXT: ;;#ASMSTART ; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: s_swappc_b64 s[30:31], s[4:5] -; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s30, v40, 2 +; GFX10-NEXT: v_readlane_b32 s31, v40, 3 ; GFX10-NEXT: v_readlane_b32 s5, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 @@ -82,21 +82,21 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 4 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: v_writelane_b32 v40, s5, 1 +; GFX11-NEXT: v_writelane_b32 v40, s30, 2 +; GFX11-NEXT: v_writelane_b32 v40, s31, 3 ; GFX11-NEXT: s_getpc_b64 s[4:5] ; GFX11-NEXT: s_add_u32 s4, s4, external_void_func_void@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 2 -; GFX11-NEXT: v_writelane_b32 v40, s31, 3 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[4:5] ; GFX11-NEXT: ;;#ASMSTART ; GFX11-NEXT: ;;#ASMEND ; GFX11-NEXT: s_swappc_b64 s[30:31], s[4:5] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s30, v40, 2 +; GFX11-NEXT: v_readlane_b32 s31, v40, 3 ; GFX11-NEXT: v_readlane_b32 s5, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 @@ -129,8 +129,8 @@ ; GFX9-NEXT: ;;#ASMSTART ; GFX9-NEXT: ; clobber ; GFX9-NEXT: ;;#ASMEND -; GFX9-NEXT: v_readlane_b32 s31, v0, 3 ; GFX9-NEXT: v_readlane_b32 s30, v0, 2 +; GFX9-NEXT: v_readlane_b32 s31, v0, 3 ; GFX9-NEXT: v_readlane_b32 s29, v0, 1 ; GFX9-NEXT: v_readlane_b32 s28, v0, 0 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -157,8 +157,8 @@ ; GFX10-NEXT: ;;#ASMSTART ; GFX10-NEXT: ; clobber ; GFX10-NEXT: ;;#ASMEND -; GFX10-NEXT: v_readlane_b32 s31, v0, 3 ; GFX10-NEXT: v_readlane_b32 s30, v0, 2 +; GFX10-NEXT: v_readlane_b32 s31, v0, 3 ; GFX10-NEXT: v_readlane_b32 s29, v0, 1 ; GFX10-NEXT: v_readlane_b32 s28, v0, 0 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -187,8 +187,8 @@ ; GFX11-NEXT: ; clobber ; GFX11-NEXT: ;;#ASMEND ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v0, 3 ; GFX11-NEXT: v_readlane_b32 s30, v0, 2 +; GFX11-NEXT: v_readlane_b32 s31, v0, 3 ; GFX11-NEXT: v_readlane_b32 s29, v0, 1 ; GFX11-NEXT: v_readlane_b32 s28, v0, 0 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -216,8 +216,8 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 3 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s30, 1 ; GFX9-NEXT: v_writelane_b32 v40, s31, 2 ; GFX9-NEXT: ;;#ASMSTART @@ -232,8 +232,8 @@ ; GFX9-NEXT: ;;#ASMSTART ; GFX9-NEXT: ; use s31 ; GFX9-NEXT: ;;#ASMEND -; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s30, v40, 1 +; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 3 @@ -253,13 +253,13 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_void@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_void@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: ;;#ASMSTART ; GFX10-NEXT: ; def s31 ; GFX10-NEXT: ;;#ASMEND @@ -269,8 +269,8 @@ ; GFX10-NEXT: ;;#ASMSTART ; GFX10-NEXT: ; use s31 ; GFX10-NEXT: ;;#ASMEND -; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 3 @@ -290,13 +290,13 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 3 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s30, 1 +; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_void@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_void@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 1 -; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: ;;#ASMSTART ; GFX11-NEXT: ; def s31 ; GFX11-NEXT: ;;#ASMEND @@ -306,8 +306,8 @@ ; GFX11-NEXT: ;;#ASMSTART ; GFX11-NEXT: ; use s31 ; GFX11-NEXT: ;;#ASMEND -; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s30, v40, 1 +; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 3 @@ -332,8 +332,8 @@ ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 ; GFX9-NEXT: s_addk_i32 s32, 0x400 -; GFX9-NEXT: v_writelane_b32 v40, s30, 0 ; GFX9-NEXT: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill +; GFX9-NEXT: v_writelane_b32 v40, s30, 0 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: ;;#ASMSTART ; GFX9-NEXT: ; def v31 @@ -348,8 +348,8 @@ ; GFX9-NEXT: ; use v31 ; GFX9-NEXT: ;;#ASMEND ; GFX9-NEXT: buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -368,14 +368,14 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill +; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: ;;#ASMSTART ; GFX10-NEXT: ; def v31 ; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: v_mov_b32_e32 v41, v31 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_void@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_void@rel32@hi+12 @@ -385,8 +385,8 @@ ; GFX10-NEXT: ; use v31 ; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -405,14 +405,14 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: scratch_store_b32 off, v41, s33 ; 4-byte Folded Spill +; GFX11-NEXT: v_writelane_b32 v40, s30, 0 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: ;;#ASMSTART ; GFX11-NEXT: ; def v31 ; GFX11-NEXT: ;;#ASMEND ; GFX11-NEXT: v_mov_b32_e32 v41, v31 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_void@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_void@rel32@hi+12 @@ -423,8 +423,8 @@ ; GFX11-NEXT: ; use v31 ; GFX11-NEXT: ;;#ASMEND ; GFX11-NEXT: scratch_load_b32 v41, off, s33 ; 4-byte Folded Reload -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -448,8 +448,8 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 3 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s30, 1 ; GFX9-NEXT: v_writelane_b32 v40, s31, 2 ; GFX9-NEXT: ;;#ASMSTART @@ -461,11 +461,11 @@ ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_void@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] ; GFX9-NEXT: s_mov_b32 s33, s4 +; GFX9-NEXT: v_readlane_b32 s30, v40, 1 ; GFX9-NEXT: ;;#ASMSTART ; GFX9-NEXT: ; use s33 ; GFX9-NEXT: ;;#ASMEND ; GFX9-NEXT: v_readlane_b32 s31, v40, 2 -; GFX9-NEXT: v_readlane_b32 s30, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 3 @@ -485,8 +485,10 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_void@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_void@rel32@hi+12 @@ -494,15 +496,13 @@ ; GFX10-NEXT: ; def s33 ; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: s_mov_b32 s4, s33 -; GFX10-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] ; GFX10-NEXT: s_mov_b32 s33, s4 +; GFX10-NEXT: v_readlane_b32 s30, v40, 1 ; GFX10-NEXT: ;;#ASMSTART ; GFX10-NEXT: ; use s33 ; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: v_readlane_b32 s31, v40, 2 -; GFX10-NEXT: v_readlane_b32 s30, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 3 @@ -522,8 +522,10 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 3 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s30, 1 +; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_void@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_void@rel32@hi+12 @@ -531,16 +533,13 @@ ; GFX11-NEXT: ; def s33 ; GFX11-NEXT: ;;#ASMEND ; GFX11-NEXT: s_mov_b32 s4, s33 -; GFX11-NEXT: v_writelane_b32 v40, s30, 1 -; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] ; GFX11-NEXT: s_mov_b32 s33, s4 +; GFX11-NEXT: v_readlane_b32 s30, v40, 1 ; GFX11-NEXT: ;;#ASMSTART ; GFX11-NEXT: ; use s33 ; GFX11-NEXT: ;;#ASMEND -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-NEXT: v_readlane_b32 s31, v40, 2 -; GFX11-NEXT: v_readlane_b32 s30, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 3 @@ -564,24 +563,24 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 3 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s30, 1 +; GFX9-NEXT: v_writelane_b32 v40, s31, 2 ; GFX9-NEXT: ;;#ASMSTART ; GFX9-NEXT: ; def s34 ; GFX9-NEXT: ;;#ASMEND -; GFX9-NEXT: v_writelane_b32 v40, s31, 2 ; GFX9-NEXT: s_mov_b32 s4, s34 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_void@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_void@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] +; GFX9-NEXT: v_readlane_b32 s30, v40, 1 ; GFX9-NEXT: s_mov_b32 s34, s4 ; GFX9-NEXT: ;;#ASMSTART ; GFX9-NEXT: ; use s34 ; GFX9-NEXT: ;;#ASMEND ; GFX9-NEXT: v_readlane_b32 s31, v40, 2 -; GFX9-NEXT: v_readlane_b32 s30, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 3 @@ -601,8 +600,10 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_getpc_b64 s[36:37] ; GFX10-NEXT: s_add_u32 s36, s36, external_void_func_void@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s37, s37, external_void_func_void@rel32@hi+12 @@ -610,15 +611,13 @@ ; GFX10-NEXT: ; def s34 ; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: s_mov_b32 s4, s34 -; GFX10-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[36:37] +; GFX10-NEXT: v_readlane_b32 s30, v40, 1 ; GFX10-NEXT: s_mov_b32 s34, s4 ; GFX10-NEXT: ;;#ASMSTART ; GFX10-NEXT: ; use s34 ; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: v_readlane_b32 s31, v40, 2 -; GFX10-NEXT: v_readlane_b32 s30, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 3 @@ -638,8 +637,10 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 3 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s30, 1 +; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_void@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_void@rel32@hi+12 @@ -647,16 +648,13 @@ ; GFX11-NEXT: ; def s34 ; GFX11-NEXT: ;;#ASMEND ; GFX11-NEXT: s_mov_b32 s4, s34 -; GFX11-NEXT: v_writelane_b32 v40, s30, 1 -; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] +; GFX11-NEXT: v_readlane_b32 s30, v40, 1 ; GFX11-NEXT: s_mov_b32 s34, s4 ; GFX11-NEXT: ;;#ASMSTART ; GFX11-NEXT: ; use s34 ; GFX11-NEXT: ;;#ASMEND -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-NEXT: v_readlane_b32 s31, v40, 2 -; GFX11-NEXT: v_readlane_b32 s30, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 3 @@ -681,8 +679,8 @@ ; GFX9-NEXT: v_writelane_b32 v41, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 ; GFX9-NEXT: s_addk_i32 s32, 0x400 -; GFX9-NEXT: v_writelane_b32 v41, s30, 0 ; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill +; GFX9-NEXT: v_writelane_b32 v41, s30, 0 ; GFX9-NEXT: v_writelane_b32 v41, s31, 1 ; GFX9-NEXT: ;;#ASMSTART ; GFX9-NEXT: ; def v40 @@ -695,8 +693,8 @@ ; GFX9-NEXT: ; use v40 ; GFX9-NEXT: ;;#ASMEND ; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload -; GFX9-NEXT: v_readlane_b32 s31, v41, 1 ; GFX9-NEXT: v_readlane_b32 s30, v41, 0 +; GFX9-NEXT: v_readlane_b32 s31, v41, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v41, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -715,23 +713,23 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v41, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v41, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill +; GFX10-NEXT: v_writelane_b32 v41, s30, 0 +; GFX10-NEXT: v_writelane_b32 v41, s31, 1 ; GFX10-NEXT: ;;#ASMSTART ; GFX10-NEXT: ; def v40 ; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_void@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_void@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v41, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] ; GFX10-NEXT: ;;#ASMSTART ; GFX10-NEXT: ; use v40 ; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload -; GFX10-NEXT: v_readlane_b32 s31, v41, 1 ; GFX10-NEXT: v_readlane_b32 s30, v41, 0 +; GFX10-NEXT: v_readlane_b32 s31, v41, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v41, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -750,23 +748,24 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v41, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v41, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill +; GFX11-NEXT: v_writelane_b32 v41, s30, 0 +; GFX11-NEXT: v_writelane_b32 v41, s31, 1 ; GFX11-NEXT: ;;#ASMSTART ; GFX11-NEXT: ; def v40 ; GFX11-NEXT: ;;#ASMEND ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_void@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_void@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v41, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] ; GFX11-NEXT: ;;#ASMSTART ; GFX11-NEXT: ; use v40 ; GFX11-NEXT: ;;#ASMEND ; GFX11-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload -; GFX11-NEXT: v_readlane_b32 s31, v41, 1 ; GFX11-NEXT: v_readlane_b32 s30, v41, 0 +; GFX11-NEXT: v_readlane_b32 s31, v41, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v41, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -913,15 +912,15 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, void_func_void_clobber_s33@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, void_func_void_clobber_s33@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -942,13 +941,13 @@ ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, void_func_void_clobber_s33@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, void_func_void_clobber_s33@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -969,14 +968,14 @@ ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, void_func_void_clobber_s33@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, void_func_void_clobber_s33@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -997,15 +996,15 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x400 ; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, void_func_void_clobber_s34@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, void_func_void_clobber_s34@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -1026,13 +1025,13 @@ ; GFX10-NEXT: s_mov_b32 s33, s32 ; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, void_func_void_clobber_s34@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, void_func_void_clobber_s34@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -1053,14 +1052,14 @@ ; GFX11-NEXT: s_mov_b32 s33, s32 ; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, void_func_void_clobber_s34@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, void_func_void_clobber_s34@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -1081,8 +1080,8 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 3 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: v_writelane_b32 v40, s30, 1 ; GFX9-NEXT: v_writelane_b32 v40, s31, 2 ; GFX9-NEXT: ;;#ASMSTART @@ -1093,11 +1092,11 @@ ; GFX9-NEXT: s_add_u32 s34, s34, external_void_func_void@rel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, external_void_func_void@rel32@hi+12 ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] +; GFX9-NEXT: v_readlane_b32 s30, v40, 1 ; GFX9-NEXT: ;;#ASMSTART ; GFX9-NEXT: ; use s4 ; GFX9-NEXT: ;;#ASMEND ; GFX9-NEXT: v_readlane_b32 s31, v40, 2 -; GFX9-NEXT: v_readlane_b32 s30, v40, 1 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 3 @@ -1117,8 +1116,10 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_void@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_void@rel32@hi+12 @@ -1126,14 +1127,12 @@ ; GFX10-NEXT: ; def s40 ; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: s_mov_b32 s4, s40 -; GFX10-NEXT: v_writelane_b32 v40, s30, 1 -; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] +; GFX10-NEXT: v_readlane_b32 s30, v40, 1 ; GFX10-NEXT: ;;#ASMSTART ; GFX10-NEXT: ; use s4 ; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: v_readlane_b32 s31, v40, 2 -; GFX10-NEXT: v_readlane_b32 s30, v40, 1 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 3 @@ -1153,8 +1152,10 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 3 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s30, 1 +; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_void@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_void@rel32@hi+12 @@ -1162,15 +1163,12 @@ ; GFX11-NEXT: ; def s40 ; GFX11-NEXT: ;;#ASMEND ; GFX11-NEXT: s_mov_b32 s4, s40 -; GFX11-NEXT: v_writelane_b32 v40, s30, 1 -; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] +; GFX11-NEXT: v_readlane_b32 s30, v40, 1 ; GFX11-NEXT: ;;#ASMSTART ; GFX11-NEXT: ; use s4 ; GFX11-NEXT: ;;#ASMEND -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-NEXT: v_readlane_b32 s31, v40, 2 -; GFX11-NEXT: v_readlane_b32 s30, v40, 1 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 3 @@ -1194,10 +1192,10 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v40, s33, 3 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: v_writelane_b32 v40, s4, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 -; GFX9-NEXT: v_writelane_b32 v40, s30, 1 ; GFX9-NEXT: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill +; GFX9-NEXT: v_writelane_b32 v40, s4, 0 +; GFX9-NEXT: v_writelane_b32 v40, s30, 1 ; GFX9-NEXT: v_writelane_b32 v40, s31, 2 ; GFX9-NEXT: ;;#ASMSTART ; GFX9-NEXT: ; def s40 @@ -1218,8 +1216,8 @@ ; GFX9-NEXT: ; use v41 ; GFX9-NEXT: ;;#ASMEND ; GFX9-NEXT: buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload -; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s30, v40, 1 +; GFX9-NEXT: v_readlane_b32 s31, v40, 2 ; GFX9-NEXT: v_readlane_b32 s4, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v40, 3 @@ -1239,14 +1237,15 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v40, s33, 3 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s4, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 ; GFX10-NEXT: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill +; GFX10-NEXT: v_writelane_b32 v40, s4, 0 +; GFX10-NEXT: v_writelane_b32 v40, s30, 1 +; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: ;;#ASMSTART ; GFX10-NEXT: ; def s40 ; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: s_mov_b32 s4, s40 -; GFX10-NEXT: v_writelane_b32 v40, s30, 1 ; GFX10-NEXT: ;;#ASMSTART ; GFX10-NEXT: ; def v32 ; GFX10-NEXT: ;;#ASMEND @@ -1254,7 +1253,6 @@ ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, external_void_func_void@rel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, external_void_func_void@rel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s31, 2 ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] ; GFX10-NEXT: ;;#ASMSTART ; GFX10-NEXT: ; use s4 @@ -1263,8 +1261,8 @@ ; GFX10-NEXT: ; use v41 ; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload -; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s30, v40, 1 +; GFX10-NEXT: v_readlane_b32 s31, v40, 2 ; GFX10-NEXT: v_readlane_b32 s4, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 3 @@ -1284,14 +1282,15 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 3 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s4, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 ; GFX11-NEXT: scratch_store_b32 off, v41, s33 ; 4-byte Folded Spill +; GFX11-NEXT: v_writelane_b32 v40, s4, 0 +; GFX11-NEXT: v_writelane_b32 v40, s30, 1 +; GFX11-NEXT: v_writelane_b32 v40, s31, 2 ; GFX11-NEXT: ;;#ASMSTART ; GFX11-NEXT: ; def s40 ; GFX11-NEXT: ;;#ASMEND ; GFX11-NEXT: s_mov_b32 s4, s40 -; GFX11-NEXT: v_writelane_b32 v40, s30, 1 ; GFX11-NEXT: ;;#ASMSTART ; GFX11-NEXT: ; def v32 ; GFX11-NEXT: ;;#ASMEND @@ -1299,7 +1298,7 @@ ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, external_void_func_void@rel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, external_void_func_void@rel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s31, 2 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] ; GFX11-NEXT: ;;#ASMSTART ; GFX11-NEXT: ; use s4 @@ -1308,8 +1307,8 @@ ; GFX11-NEXT: ; use v41 ; GFX11-NEXT: ;;#ASMEND ; GFX11-NEXT: scratch_load_b32 v41, off, s33 ; 4-byte Folded Reload -; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s30, v40, 1 +; GFX11-NEXT: v_readlane_b32 s31, v40, 2 ; GFX11-NEXT: v_readlane_b32 s4, v40, 0 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v40, 3 diff --git a/llvm/test/CodeGen/AMDGPU/gfx-callable-return-types.ll b/llvm/test/CodeGen/AMDGPU/gfx-callable-return-types.ll --- a/llvm/test/CodeGen/AMDGPU/gfx-callable-return-types.ll +++ b/llvm/test/CodeGen/AMDGPU/gfx-callable-return-types.ll @@ -29,17 +29,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v1, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: v_writelane_b32 v1, s30, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v1, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, return_i1@gotpcrel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, return_i1@gotpcrel32@hi+12 ; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX9-NEXT: v_writelane_b32 v1, s30, 0 -; GFX9-NEXT: v_writelane_b32 v1, s31, 1 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v1, 1 ; GFX9-NEXT: v_readlane_b32 s30, v1, 0 +; GFX9-NEXT: v_readlane_b32 s31, v1, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v1, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -58,17 +58,17 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v1, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: v_writelane_b32 v1, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v1, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, return_i1@gotpcrel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, return_i1@gotpcrel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v1, s30, 0 ; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX10-NEXT: v_writelane_b32 v1, s31, 1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v1, 1 ; GFX10-NEXT: v_readlane_b32 s30, v1, 0 +; GFX10-NEXT: v_readlane_b32 s31, v1, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v1, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -87,18 +87,17 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v1, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: v_writelane_b32 v1, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v1, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, return_i1@gotpcrel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, return_i1@gotpcrel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v1, s30, 0 ; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 -; GFX11-NEXT: v_writelane_b32 v1, s31, 1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v1, 1 ; GFX11-NEXT: v_readlane_b32 s30, v1, 0 +; GFX11-NEXT: v_readlane_b32 s31, v1, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v1, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -137,17 +136,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v1, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: v_writelane_b32 v1, s30, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v1, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, return_i16@gotpcrel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, return_i16@gotpcrel32@hi+12 ; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX9-NEXT: v_writelane_b32 v1, s30, 0 -; GFX9-NEXT: v_writelane_b32 v1, s31, 1 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v1, 1 ; GFX9-NEXT: v_readlane_b32 s30, v1, 0 +; GFX9-NEXT: v_readlane_b32 s31, v1, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v1, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -166,17 +165,17 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v1, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: v_writelane_b32 v1, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v1, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, return_i16@gotpcrel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, return_i16@gotpcrel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v1, s30, 0 ; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX10-NEXT: v_writelane_b32 v1, s31, 1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v1, 1 ; GFX10-NEXT: v_readlane_b32 s30, v1, 0 +; GFX10-NEXT: v_readlane_b32 s31, v1, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v1, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -195,18 +194,17 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v1, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: v_writelane_b32 v1, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v1, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, return_i16@gotpcrel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, return_i16@gotpcrel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v1, s30, 0 ; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 -; GFX11-NEXT: v_writelane_b32 v1, s31, 1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v1, 1 ; GFX11-NEXT: v_readlane_b32 s30, v1, 0 +; GFX11-NEXT: v_readlane_b32 s31, v1, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v1, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -245,17 +243,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v1, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: v_writelane_b32 v1, s30, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v1, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, return_2xi16@gotpcrel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, return_2xi16@gotpcrel32@hi+12 ; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX9-NEXT: v_writelane_b32 v1, s30, 0 -; GFX9-NEXT: v_writelane_b32 v1, s31, 1 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v1, 1 ; GFX9-NEXT: v_readlane_b32 s30, v1, 0 +; GFX9-NEXT: v_readlane_b32 s31, v1, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v1, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -274,17 +272,17 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v1, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: v_writelane_b32 v1, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v1, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, return_2xi16@gotpcrel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, return_2xi16@gotpcrel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v1, s30, 0 ; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX10-NEXT: v_writelane_b32 v1, s31, 1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v1, 1 ; GFX10-NEXT: v_readlane_b32 s30, v1, 0 +; GFX10-NEXT: v_readlane_b32 s31, v1, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v1, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -303,18 +301,17 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v1, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: v_writelane_b32 v1, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v1, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, return_2xi16@gotpcrel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, return_2xi16@gotpcrel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v1, s30, 0 ; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 -; GFX11-NEXT: v_writelane_b32 v1, s31, 1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v1, 1 ; GFX11-NEXT: v_readlane_b32 s30, v1, 0 +; GFX11-NEXT: v_readlane_b32 s31, v1, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v1, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -362,17 +359,17 @@ ; GFX9-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-NEXT: v_writelane_b32 v2, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: v_writelane_b32 v2, s30, 0 ; GFX9-NEXT: s_addk_i32 s32, 0x400 +; GFX9-NEXT: v_writelane_b32 v2, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, return_3xi16@gotpcrel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, return_3xi16@gotpcrel32@hi+12 ; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX9-NEXT: v_writelane_b32 v2, s30, 0 -; GFX9-NEXT: v_writelane_b32 v2, s31, 1 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v2, 1 ; GFX9-NEXT: v_readlane_b32 s30, v2, 0 +; GFX9-NEXT: v_readlane_b32 s31, v2, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-NEXT: v_readlane_b32 s33, v2, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -391,17 +388,17 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s34 ; GFX10-NEXT: v_writelane_b32 v2, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: v_writelane_b32 v2, s30, 0 ; GFX10-NEXT: s_addk_i32 s32, 0x200 +; GFX10-NEXT: v_writelane_b32 v2, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, return_3xi16@gotpcrel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, return_3xi16@gotpcrel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v2, s30, 0 ; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX10-NEXT: v_writelane_b32 v2, s31, 1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v2, 1 ; GFX10-NEXT: v_readlane_b32 s30, v2, 0 +; GFX10-NEXT: v_readlane_b32 s31, v2, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfe00 ; GFX10-NEXT: v_readlane_b32 s33, v2, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -420,18 +417,17 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v2, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 +; GFX11-NEXT: v_writelane_b32 v2, s30, 0 ; GFX11-NEXT: s_add_i32 s32, s32, 16 +; GFX11-NEXT: v_writelane_b32 v2, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, return_3xi16@gotpcrel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, return_3xi16@gotpcrel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v2, s30, 0 ; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 -; GFX11-NEXT: v_writelane_b32 v2, s31, 1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v2, 1 ; GFX11-NEXT: v_readlane_b32 s30, v2, 0 +; GFX11-NEXT: v_readlane_b32 s31, v2, 1 ; GFX11-NEXT: s_add_i32 s32, s32, -16 ; GFX11-NEXT: v_readlane_b32 s33, v2, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -1644,18 +1640,18 @@ ; GFX9-NEXT: v_writelane_b32 v2, s33, 2 ; GFX9-NEXT: s_add_i32 s33, s32, 0x1ffc0 ; GFX9-NEXT: s_and_b32 s33, s33, 0xfffe0000 +; GFX9-NEXT: v_writelane_b32 v2, s30, 0 ; GFX9-NEXT: s_add_i32 s32, s32, 0x60000 +; GFX9-NEXT: v_writelane_b32 v2, s31, 1 ; GFX9-NEXT: s_getpc_b64 s[34:35] ; GFX9-NEXT: s_add_u32 s34, s34, return_512xi32@gotpcrel32@lo+4 ; GFX9-NEXT: s_addc_u32 s35, s35, return_512xi32@gotpcrel32@hi+12 ; GFX9-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 -; GFX9-NEXT: v_writelane_b32 v2, s30, 0 ; GFX9-NEXT: v_lshrrev_b32_e64 v0, 6, s33 -; GFX9-NEXT: v_writelane_b32 v2, s31, 1 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX9-NEXT: v_readlane_b32 s31, v2, 1 ; GFX9-NEXT: v_readlane_b32 s30, v2, 0 +; GFX9-NEXT: v_readlane_b32 s31, v2, 1 ; GFX9-NEXT: s_add_i32 s32, s32, 0xfffa0000 ; GFX9-NEXT: v_readlane_b32 s33, v2, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -1675,18 +1671,18 @@ ; GFX10-NEXT: v_writelane_b32 v2, s33, 2 ; GFX10-NEXT: s_add_i32 s33, s32, 0xffe0 ; GFX10-NEXT: s_and_b32 s33, s33, 0xffff0000 +; GFX10-NEXT: v_writelane_b32 v2, s30, 0 ; GFX10-NEXT: s_add_i32 s32, s32, 0x30000 +; GFX10-NEXT: v_writelane_b32 v2, s31, 1 ; GFX10-NEXT: s_getpc_b64 s[34:35] ; GFX10-NEXT: s_add_u32 s34, s34, return_512xi32@gotpcrel32@lo+4 ; GFX10-NEXT: s_addc_u32 s35, s35, return_512xi32@gotpcrel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v2, s30, 0 -; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX10-NEXT: v_lshrrev_b32_e64 v0, 5, s33 -; GFX10-NEXT: v_writelane_b32 v2, s31, 1 +; GFX10-NEXT: s_load_dwordx2 s[34:35], s[34:35], 0x0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GFX10-NEXT: v_readlane_b32 s31, v2, 1 ; GFX10-NEXT: v_readlane_b32 s30, v2, 0 +; GFX10-NEXT: v_readlane_b32 s31, v2, 1 ; GFX10-NEXT: s_add_i32 s32, s32, 0xfffd0000 ; GFX10-NEXT: v_readlane_b32 s33, v2, 2 ; GFX10-NEXT: s_or_saveexec_b32 s34, -1 @@ -1707,19 +1703,18 @@ ; GFX11-NEXT: s_add_i32 s33, s32, 0x7ff ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_b32 s33, s33, 0xfffff800 +; GFX11-NEXT: v_writelane_b32 v5, s30, 0 ; GFX11-NEXT: s_addk_i32 s32, 0x1800 +; GFX11-NEXT: v_writelane_b32 v5, s31, 1 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, return_512xi32@gotpcrel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, return_512xi32@gotpcrel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v5, s30, 0 -; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 ; GFX11-NEXT: v_mov_b32_e32 v0, s33 -; GFX11-NEXT: v_writelane_b32 v5, s31, 1 +; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_readlane_b32 s31, v5, 1 ; GFX11-NEXT: v_readlane_b32 s30, v5, 0 +; GFX11-NEXT: v_readlane_b32 s31, v5, 1 ; GFX11-NEXT: s_addk_i32 s32, 0xe800 ; GFX11-NEXT: v_readlane_b32 s33, v5, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 diff --git a/llvm/test/CodeGen/AMDGPU/indirect-call.ll b/llvm/test/CodeGen/AMDGPU/indirect-call.ll --- a/llvm/test/CodeGen/AMDGPU/indirect-call.ll +++ b/llvm/test/CodeGen/AMDGPU/indirect-call.ll @@ -399,24 +399,24 @@ ; GCN-NEXT: v_writelane_b32 v40, s33, 18 ; GCN-NEXT: s_mov_b32 s33, s32 ; GCN-NEXT: s_addk_i32 s32, 0x400 -; GCN-NEXT: v_writelane_b32 v40, s30, 0 -; GCN-NEXT: v_writelane_b32 v40, s31, 1 -; GCN-NEXT: v_writelane_b32 v40, s34, 2 -; GCN-NEXT: v_writelane_b32 v40, s35, 3 -; GCN-NEXT: v_writelane_b32 v40, s36, 4 -; GCN-NEXT: v_writelane_b32 v40, s37, 5 -; GCN-NEXT: v_writelane_b32 v40, s38, 6 -; GCN-NEXT: v_writelane_b32 v40, s39, 7 -; GCN-NEXT: v_writelane_b32 v40, s40, 8 -; GCN-NEXT: v_writelane_b32 v40, s41, 9 -; GCN-NEXT: v_writelane_b32 v40, s42, 10 -; GCN-NEXT: v_writelane_b32 v40, s43, 11 -; GCN-NEXT: v_writelane_b32 v40, s44, 12 -; GCN-NEXT: v_writelane_b32 v40, s45, 13 -; GCN-NEXT: v_writelane_b32 v40, s46, 14 -; GCN-NEXT: v_writelane_b32 v40, s47, 15 -; GCN-NEXT: v_writelane_b32 v40, s48, 16 -; GCN-NEXT: v_writelane_b32 v40, s49, 17 +; GCN-NEXT: v_writelane_b32 v40, s34, 0 +; GCN-NEXT: v_writelane_b32 v40, s35, 1 +; GCN-NEXT: v_writelane_b32 v40, s36, 2 +; GCN-NEXT: v_writelane_b32 v40, s37, 3 +; GCN-NEXT: v_writelane_b32 v40, s38, 4 +; GCN-NEXT: v_writelane_b32 v40, s39, 5 +; GCN-NEXT: v_writelane_b32 v40, s40, 6 +; GCN-NEXT: v_writelane_b32 v40, s41, 7 +; GCN-NEXT: v_writelane_b32 v40, s42, 8 +; GCN-NEXT: v_writelane_b32 v40, s43, 9 +; GCN-NEXT: v_writelane_b32 v40, s44, 10 +; GCN-NEXT: v_writelane_b32 v40, s45, 11 +; GCN-NEXT: v_writelane_b32 v40, s46, 12 +; GCN-NEXT: v_writelane_b32 v40, s47, 13 +; GCN-NEXT: v_writelane_b32 v40, s48, 14 +; GCN-NEXT: v_writelane_b32 v40, s49, 15 +; GCN-NEXT: v_writelane_b32 v40, s30, 16 +; GCN-NEXT: v_writelane_b32 v40, s31, 17 ; GCN-NEXT: s_mov_b32 s42, s15 ; GCN-NEXT: s_mov_b32 s43, s14 ; GCN-NEXT: s_mov_b32 s44, s13 @@ -446,24 +446,24 @@ ; GCN-NEXT: s_cbranch_execnz .LBB2_1 ; GCN-NEXT: ; %bb.2: ; GCN-NEXT: s_mov_b64 exec, s[46:47] -; GCN-NEXT: v_readlane_b32 s49, v40, 17 -; GCN-NEXT: v_readlane_b32 s48, v40, 16 -; GCN-NEXT: v_readlane_b32 s47, v40, 15 -; GCN-NEXT: v_readlane_b32 s46, v40, 14 -; GCN-NEXT: v_readlane_b32 s45, v40, 13 -; GCN-NEXT: v_readlane_b32 s44, v40, 12 -; GCN-NEXT: v_readlane_b32 s43, v40, 11 -; GCN-NEXT: v_readlane_b32 s42, v40, 10 -; GCN-NEXT: v_readlane_b32 s41, v40, 9 -; GCN-NEXT: v_readlane_b32 s40, v40, 8 -; GCN-NEXT: v_readlane_b32 s39, v40, 7 -; GCN-NEXT: v_readlane_b32 s38, v40, 6 -; GCN-NEXT: v_readlane_b32 s37, v40, 5 -; GCN-NEXT: v_readlane_b32 s36, v40, 4 -; GCN-NEXT: v_readlane_b32 s35, v40, 3 -; GCN-NEXT: v_readlane_b32 s34, v40, 2 -; GCN-NEXT: v_readlane_b32 s31, v40, 1 -; GCN-NEXT: v_readlane_b32 s30, v40, 0 +; GCN-NEXT: v_readlane_b32 s30, v40, 16 +; GCN-NEXT: v_readlane_b32 s31, v40, 17 +; GCN-NEXT: v_readlane_b32 s49, v40, 15 +; GCN-NEXT: v_readlane_b32 s48, v40, 14 +; GCN-NEXT: v_readlane_b32 s47, v40, 13 +; GCN-NEXT: v_readlane_b32 s46, v40, 12 +; GCN-NEXT: v_readlane_b32 s45, v40, 11 +; GCN-NEXT: v_readlane_b32 s44, v40, 10 +; GCN-NEXT: v_readlane_b32 s43, v40, 9 +; GCN-NEXT: v_readlane_b32 s42, v40, 8 +; GCN-NEXT: v_readlane_b32 s41, v40, 7 +; GCN-NEXT: v_readlane_b32 s40, v40, 6 +; GCN-NEXT: v_readlane_b32 s39, v40, 5 +; GCN-NEXT: v_readlane_b32 s38, v40, 4 +; GCN-NEXT: v_readlane_b32 s37, v40, 3 +; GCN-NEXT: v_readlane_b32 s36, v40, 2 +; GCN-NEXT: v_readlane_b32 s35, v40, 1 +; GCN-NEXT: v_readlane_b32 s34, v40, 0 ; GCN-NEXT: s_addk_i32 s32, 0xfc00 ; GCN-NEXT: v_readlane_b32 s33, v40, 18 ; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -481,24 +481,24 @@ ; GISEL-NEXT: v_writelane_b32 v40, s33, 18 ; GISEL-NEXT: s_mov_b32 s33, s32 ; GISEL-NEXT: s_addk_i32 s32, 0x400 -; GISEL-NEXT: v_writelane_b32 v40, s30, 0 -; GISEL-NEXT: v_writelane_b32 v40, s31, 1 -; GISEL-NEXT: v_writelane_b32 v40, s34, 2 -; GISEL-NEXT: v_writelane_b32 v40, s35, 3 -; GISEL-NEXT: v_writelane_b32 v40, s36, 4 -; GISEL-NEXT: v_writelane_b32 v40, s37, 5 -; GISEL-NEXT: v_writelane_b32 v40, s38, 6 -; GISEL-NEXT: v_writelane_b32 v40, s39, 7 -; GISEL-NEXT: v_writelane_b32 v40, s40, 8 -; GISEL-NEXT: v_writelane_b32 v40, s41, 9 -; GISEL-NEXT: v_writelane_b32 v40, s42, 10 -; GISEL-NEXT: v_writelane_b32 v40, s43, 11 -; GISEL-NEXT: v_writelane_b32 v40, s44, 12 -; GISEL-NEXT: v_writelane_b32 v40, s45, 13 -; GISEL-NEXT: v_writelane_b32 v40, s46, 14 -; GISEL-NEXT: v_writelane_b32 v40, s47, 15 -; GISEL-NEXT: v_writelane_b32 v40, s48, 16 -; GISEL-NEXT: v_writelane_b32 v40, s49, 17 +; GISEL-NEXT: v_writelane_b32 v40, s34, 0 +; GISEL-NEXT: v_writelane_b32 v40, s35, 1 +; GISEL-NEXT: v_writelane_b32 v40, s36, 2 +; GISEL-NEXT: v_writelane_b32 v40, s37, 3 +; GISEL-NEXT: v_writelane_b32 v40, s38, 4 +; GISEL-NEXT: v_writelane_b32 v40, s39, 5 +; GISEL-NEXT: v_writelane_b32 v40, s40, 6 +; GISEL-NEXT: v_writelane_b32 v40, s41, 7 +; GISEL-NEXT: v_writelane_b32 v40, s42, 8 +; GISEL-NEXT: v_writelane_b32 v40, s43, 9 +; GISEL-NEXT: v_writelane_b32 v40, s44, 10 +; GISEL-NEXT: v_writelane_b32 v40, s45, 11 +; GISEL-NEXT: v_writelane_b32 v40, s46, 12 +; GISEL-NEXT: v_writelane_b32 v40, s47, 13 +; GISEL-NEXT: v_writelane_b32 v40, s48, 14 +; GISEL-NEXT: v_writelane_b32 v40, s49, 15 +; GISEL-NEXT: v_writelane_b32 v40, s30, 16 +; GISEL-NEXT: v_writelane_b32 v40, s31, 17 ; GISEL-NEXT: s_mov_b32 s42, s15 ; GISEL-NEXT: s_mov_b32 s43, s14 ; GISEL-NEXT: s_mov_b32 s44, s13 @@ -528,24 +528,24 @@ ; GISEL-NEXT: s_cbranch_execnz .LBB2_1 ; GISEL-NEXT: ; %bb.2: ; GISEL-NEXT: s_mov_b64 exec, s[46:47] -; GISEL-NEXT: v_readlane_b32 s49, v40, 17 -; GISEL-NEXT: v_readlane_b32 s48, v40, 16 -; GISEL-NEXT: v_readlane_b32 s47, v40, 15 -; GISEL-NEXT: v_readlane_b32 s46, v40, 14 -; GISEL-NEXT: v_readlane_b32 s45, v40, 13 -; GISEL-NEXT: v_readlane_b32 s44, v40, 12 -; GISEL-NEXT: v_readlane_b32 s43, v40, 11 -; GISEL-NEXT: v_readlane_b32 s42, v40, 10 -; GISEL-NEXT: v_readlane_b32 s41, v40, 9 -; GISEL-NEXT: v_readlane_b32 s40, v40, 8 -; GISEL-NEXT: v_readlane_b32 s39, v40, 7 -; GISEL-NEXT: v_readlane_b32 s38, v40, 6 -; GISEL-NEXT: v_readlane_b32 s37, v40, 5 -; GISEL-NEXT: v_readlane_b32 s36, v40, 4 -; GISEL-NEXT: v_readlane_b32 s35, v40, 3 -; GISEL-NEXT: v_readlane_b32 s34, v40, 2 -; GISEL-NEXT: v_readlane_b32 s31, v40, 1 -; GISEL-NEXT: v_readlane_b32 s30, v40, 0 +; GISEL-NEXT: v_readlane_b32 s30, v40, 16 +; GISEL-NEXT: v_readlane_b32 s31, v40, 17 +; GISEL-NEXT: v_readlane_b32 s49, v40, 15 +; GISEL-NEXT: v_readlane_b32 s48, v40, 14 +; GISEL-NEXT: v_readlane_b32 s47, v40, 13 +; GISEL-NEXT: v_readlane_b32 s46, v40, 12 +; GISEL-NEXT: v_readlane_b32 s45, v40, 11 +; GISEL-NEXT: v_readlane_b32 s44, v40, 10 +; GISEL-NEXT: v_readlane_b32 s43, v40, 9 +; GISEL-NEXT: v_readlane_b32 s42, v40, 8 +; GISEL-NEXT: v_readlane_b32 s41, v40, 7 +; GISEL-NEXT: v_readlane_b32 s40, v40, 6 +; GISEL-NEXT: v_readlane_b32 s39, v40, 5 +; GISEL-NEXT: v_readlane_b32 s38, v40, 4 +; GISEL-NEXT: v_readlane_b32 s37, v40, 3 +; GISEL-NEXT: v_readlane_b32 s36, v40, 2 +; GISEL-NEXT: v_readlane_b32 s35, v40, 1 +; GISEL-NEXT: v_readlane_b32 s34, v40, 0 ; GISEL-NEXT: s_addk_i32 s32, 0xfc00 ; GISEL-NEXT: v_readlane_b32 s33, v40, 18 ; GISEL-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -567,24 +567,24 @@ ; GCN-NEXT: v_writelane_b32 v40, s33, 18 ; GCN-NEXT: s_mov_b32 s33, s32 ; GCN-NEXT: s_addk_i32 s32, 0x400 -; GCN-NEXT: v_writelane_b32 v40, s30, 0 -; GCN-NEXT: v_writelane_b32 v40, s31, 1 -; GCN-NEXT: v_writelane_b32 v40, s34, 2 -; GCN-NEXT: v_writelane_b32 v40, s35, 3 -; GCN-NEXT: v_writelane_b32 v40, s36, 4 -; GCN-NEXT: v_writelane_b32 v40, s37, 5 -; GCN-NEXT: v_writelane_b32 v40, s38, 6 -; GCN-NEXT: v_writelane_b32 v40, s39, 7 -; GCN-NEXT: v_writelane_b32 v40, s40, 8 -; GCN-NEXT: v_writelane_b32 v40, s41, 9 -; GCN-NEXT: v_writelane_b32 v40, s42, 10 -; GCN-NEXT: v_writelane_b32 v40, s43, 11 -; GCN-NEXT: v_writelane_b32 v40, s44, 12 -; GCN-NEXT: v_writelane_b32 v40, s45, 13 -; GCN-NEXT: v_writelane_b32 v40, s46, 14 -; GCN-NEXT: v_writelane_b32 v40, s47, 15 -; GCN-NEXT: v_writelane_b32 v40, s48, 16 -; GCN-NEXT: v_writelane_b32 v40, s49, 17 +; GCN-NEXT: v_writelane_b32 v40, s34, 0 +; GCN-NEXT: v_writelane_b32 v40, s35, 1 +; GCN-NEXT: v_writelane_b32 v40, s36, 2 +; GCN-NEXT: v_writelane_b32 v40, s37, 3 +; GCN-NEXT: v_writelane_b32 v40, s38, 4 +; GCN-NEXT: v_writelane_b32 v40, s39, 5 +; GCN-NEXT: v_writelane_b32 v40, s40, 6 +; GCN-NEXT: v_writelane_b32 v40, s41, 7 +; GCN-NEXT: v_writelane_b32 v40, s42, 8 +; GCN-NEXT: v_writelane_b32 v40, s43, 9 +; GCN-NEXT: v_writelane_b32 v40, s44, 10 +; GCN-NEXT: v_writelane_b32 v40, s45, 11 +; GCN-NEXT: v_writelane_b32 v40, s46, 12 +; GCN-NEXT: v_writelane_b32 v40, s47, 13 +; GCN-NEXT: v_writelane_b32 v40, s48, 14 +; GCN-NEXT: v_writelane_b32 v40, s49, 15 +; GCN-NEXT: v_writelane_b32 v40, s30, 16 +; GCN-NEXT: v_writelane_b32 v40, s31, 17 ; GCN-NEXT: s_mov_b32 s42, s15 ; GCN-NEXT: s_mov_b32 s43, s14 ; GCN-NEXT: s_mov_b32 s44, s13 @@ -617,24 +617,24 @@ ; GCN-NEXT: s_cbranch_execnz .LBB3_1 ; GCN-NEXT: ; %bb.2: ; GCN-NEXT: s_mov_b64 exec, s[46:47] -; GCN-NEXT: v_readlane_b32 s49, v40, 17 -; GCN-NEXT: v_readlane_b32 s48, v40, 16 -; GCN-NEXT: v_readlane_b32 s47, v40, 15 -; GCN-NEXT: v_readlane_b32 s46, v40, 14 -; GCN-NEXT: v_readlane_b32 s45, v40, 13 -; GCN-NEXT: v_readlane_b32 s44, v40, 12 -; GCN-NEXT: v_readlane_b32 s43, v40, 11 -; GCN-NEXT: v_readlane_b32 s42, v40, 10 -; GCN-NEXT: v_readlane_b32 s41, v40, 9 -; GCN-NEXT: v_readlane_b32 s40, v40, 8 -; GCN-NEXT: v_readlane_b32 s39, v40, 7 -; GCN-NEXT: v_readlane_b32 s38, v40, 6 -; GCN-NEXT: v_readlane_b32 s37, v40, 5 -; GCN-NEXT: v_readlane_b32 s36, v40, 4 -; GCN-NEXT: v_readlane_b32 s35, v40, 3 -; GCN-NEXT: v_readlane_b32 s34, v40, 2 -; GCN-NEXT: v_readlane_b32 s31, v40, 1 -; GCN-NEXT: v_readlane_b32 s30, v40, 0 +; GCN-NEXT: v_readlane_b32 s30, v40, 16 +; GCN-NEXT: v_readlane_b32 s31, v40, 17 +; GCN-NEXT: v_readlane_b32 s49, v40, 15 +; GCN-NEXT: v_readlane_b32 s48, v40, 14 +; GCN-NEXT: v_readlane_b32 s47, v40, 13 +; GCN-NEXT: v_readlane_b32 s46, v40, 12 +; GCN-NEXT: v_readlane_b32 s45, v40, 11 +; GCN-NEXT: v_readlane_b32 s44, v40, 10 +; GCN-NEXT: v_readlane_b32 s43, v40, 9 +; GCN-NEXT: v_readlane_b32 s42, v40, 8 +; GCN-NEXT: v_readlane_b32 s41, v40, 7 +; GCN-NEXT: v_readlane_b32 s40, v40, 6 +; GCN-NEXT: v_readlane_b32 s39, v40, 5 +; GCN-NEXT: v_readlane_b32 s38, v40, 4 +; GCN-NEXT: v_readlane_b32 s37, v40, 3 +; GCN-NEXT: v_readlane_b32 s36, v40, 2 +; GCN-NEXT: v_readlane_b32 s35, v40, 1 +; GCN-NEXT: v_readlane_b32 s34, v40, 0 ; GCN-NEXT: s_addk_i32 s32, 0xfc00 ; GCN-NEXT: v_readlane_b32 s33, v40, 18 ; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -652,24 +652,24 @@ ; GISEL-NEXT: v_writelane_b32 v40, s33, 18 ; GISEL-NEXT: s_mov_b32 s33, s32 ; GISEL-NEXT: s_addk_i32 s32, 0x400 -; GISEL-NEXT: v_writelane_b32 v40, s30, 0 -; GISEL-NEXT: v_writelane_b32 v40, s31, 1 -; GISEL-NEXT: v_writelane_b32 v40, s34, 2 -; GISEL-NEXT: v_writelane_b32 v40, s35, 3 -; GISEL-NEXT: v_writelane_b32 v40, s36, 4 -; GISEL-NEXT: v_writelane_b32 v40, s37, 5 -; GISEL-NEXT: v_writelane_b32 v40, s38, 6 -; GISEL-NEXT: v_writelane_b32 v40, s39, 7 -; GISEL-NEXT: v_writelane_b32 v40, s40, 8 -; GISEL-NEXT: v_writelane_b32 v40, s41, 9 -; GISEL-NEXT: v_writelane_b32 v40, s42, 10 -; GISEL-NEXT: v_writelane_b32 v40, s43, 11 -; GISEL-NEXT: v_writelane_b32 v40, s44, 12 -; GISEL-NEXT: v_writelane_b32 v40, s45, 13 -; GISEL-NEXT: v_writelane_b32 v40, s46, 14 -; GISEL-NEXT: v_writelane_b32 v40, s47, 15 -; GISEL-NEXT: v_writelane_b32 v40, s48, 16 -; GISEL-NEXT: v_writelane_b32 v40, s49, 17 +; GISEL-NEXT: v_writelane_b32 v40, s34, 0 +; GISEL-NEXT: v_writelane_b32 v40, s35, 1 +; GISEL-NEXT: v_writelane_b32 v40, s36, 2 +; GISEL-NEXT: v_writelane_b32 v40, s37, 3 +; GISEL-NEXT: v_writelane_b32 v40, s38, 4 +; GISEL-NEXT: v_writelane_b32 v40, s39, 5 +; GISEL-NEXT: v_writelane_b32 v40, s40, 6 +; GISEL-NEXT: v_writelane_b32 v40, s41, 7 +; GISEL-NEXT: v_writelane_b32 v40, s42, 8 +; GISEL-NEXT: v_writelane_b32 v40, s43, 9 +; GISEL-NEXT: v_writelane_b32 v40, s44, 10 +; GISEL-NEXT: v_writelane_b32 v40, s45, 11 +; GISEL-NEXT: v_writelane_b32 v40, s46, 12 +; GISEL-NEXT: v_writelane_b32 v40, s47, 13 +; GISEL-NEXT: v_writelane_b32 v40, s48, 14 +; GISEL-NEXT: v_writelane_b32 v40, s49, 15 +; GISEL-NEXT: v_writelane_b32 v40, s30, 16 +; GISEL-NEXT: v_writelane_b32 v40, s31, 17 ; GISEL-NEXT: s_mov_b32 s42, s15 ; GISEL-NEXT: s_mov_b32 s43, s14 ; GISEL-NEXT: s_mov_b32 s44, s13 @@ -700,24 +700,24 @@ ; GISEL-NEXT: s_cbranch_execnz .LBB3_1 ; GISEL-NEXT: ; %bb.2: ; GISEL-NEXT: s_mov_b64 exec, s[46:47] -; GISEL-NEXT: v_readlane_b32 s49, v40, 17 -; GISEL-NEXT: v_readlane_b32 s48, v40, 16 -; GISEL-NEXT: v_readlane_b32 s47, v40, 15 -; GISEL-NEXT: v_readlane_b32 s46, v40, 14 -; GISEL-NEXT: v_readlane_b32 s45, v40, 13 -; GISEL-NEXT: v_readlane_b32 s44, v40, 12 -; GISEL-NEXT: v_readlane_b32 s43, v40, 11 -; GISEL-NEXT: v_readlane_b32 s42, v40, 10 -; GISEL-NEXT: v_readlane_b32 s41, v40, 9 -; GISEL-NEXT: v_readlane_b32 s40, v40, 8 -; GISEL-NEXT: v_readlane_b32 s39, v40, 7 -; GISEL-NEXT: v_readlane_b32 s38, v40, 6 -; GISEL-NEXT: v_readlane_b32 s37, v40, 5 -; GISEL-NEXT: v_readlane_b32 s36, v40, 4 -; GISEL-NEXT: v_readlane_b32 s35, v40, 3 -; GISEL-NEXT: v_readlane_b32 s34, v40, 2 -; GISEL-NEXT: v_readlane_b32 s31, v40, 1 -; GISEL-NEXT: v_readlane_b32 s30, v40, 0 +; GISEL-NEXT: v_readlane_b32 s30, v40, 16 +; GISEL-NEXT: v_readlane_b32 s31, v40, 17 +; GISEL-NEXT: v_readlane_b32 s49, v40, 15 +; GISEL-NEXT: v_readlane_b32 s48, v40, 14 +; GISEL-NEXT: v_readlane_b32 s47, v40, 13 +; GISEL-NEXT: v_readlane_b32 s46, v40, 12 +; GISEL-NEXT: v_readlane_b32 s45, v40, 11 +; GISEL-NEXT: v_readlane_b32 s44, v40, 10 +; GISEL-NEXT: v_readlane_b32 s43, v40, 9 +; GISEL-NEXT: v_readlane_b32 s42, v40, 8 +; GISEL-NEXT: v_readlane_b32 s41, v40, 7 +; GISEL-NEXT: v_readlane_b32 s40, v40, 6 +; GISEL-NEXT: v_readlane_b32 s39, v40, 5 +; GISEL-NEXT: v_readlane_b32 s38, v40, 4 +; GISEL-NEXT: v_readlane_b32 s37, v40, 3 +; GISEL-NEXT: v_readlane_b32 s36, v40, 2 +; GISEL-NEXT: v_readlane_b32 s35, v40, 1 +; GISEL-NEXT: v_readlane_b32 s34, v40, 0 ; GISEL-NEXT: s_addk_i32 s32, 0xfc00 ; GISEL-NEXT: v_readlane_b32 s33, v40, 18 ; GISEL-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -739,24 +739,24 @@ ; GCN-NEXT: v_writelane_b32 v40, s33, 18 ; GCN-NEXT: s_mov_b32 s33, s32 ; GCN-NEXT: s_addk_i32 s32, 0x400 -; GCN-NEXT: v_writelane_b32 v40, s30, 0 -; GCN-NEXT: v_writelane_b32 v40, s31, 1 -; GCN-NEXT: v_writelane_b32 v40, s34, 2 -; GCN-NEXT: v_writelane_b32 v40, s35, 3 -; GCN-NEXT: v_writelane_b32 v40, s36, 4 -; GCN-NEXT: v_writelane_b32 v40, s37, 5 -; GCN-NEXT: v_writelane_b32 v40, s38, 6 -; GCN-NEXT: v_writelane_b32 v40, s39, 7 -; GCN-NEXT: v_writelane_b32 v40, s40, 8 -; GCN-NEXT: v_writelane_b32 v40, s41, 9 -; GCN-NEXT: v_writelane_b32 v40, s42, 10 -; GCN-NEXT: v_writelane_b32 v40, s43, 11 -; GCN-NEXT: v_writelane_b32 v40, s44, 12 -; GCN-NEXT: v_writelane_b32 v40, s45, 13 -; GCN-NEXT: v_writelane_b32 v40, s46, 14 -; GCN-NEXT: v_writelane_b32 v40, s47, 15 -; GCN-NEXT: v_writelane_b32 v40, s48, 16 -; GCN-NEXT: v_writelane_b32 v40, s49, 17 +; GCN-NEXT: v_writelane_b32 v40, s34, 0 +; GCN-NEXT: v_writelane_b32 v40, s35, 1 +; GCN-NEXT: v_writelane_b32 v40, s36, 2 +; GCN-NEXT: v_writelane_b32 v40, s37, 3 +; GCN-NEXT: v_writelane_b32 v40, s38, 4 +; GCN-NEXT: v_writelane_b32 v40, s39, 5 +; GCN-NEXT: v_writelane_b32 v40, s40, 6 +; GCN-NEXT: v_writelane_b32 v40, s41, 7 +; GCN-NEXT: v_writelane_b32 v40, s42, 8 +; GCN-NEXT: v_writelane_b32 v40, s43, 9 +; GCN-NEXT: v_writelane_b32 v40, s44, 10 +; GCN-NEXT: v_writelane_b32 v40, s45, 11 +; GCN-NEXT: v_writelane_b32 v40, s46, 12 +; GCN-NEXT: v_writelane_b32 v40, s47, 13 +; GCN-NEXT: v_writelane_b32 v40, s48, 14 +; GCN-NEXT: v_writelane_b32 v40, s49, 15 +; GCN-NEXT: v_writelane_b32 v40, s30, 16 +; GCN-NEXT: v_writelane_b32 v40, s31, 17 ; GCN-NEXT: s_mov_b32 s42, s15 ; GCN-NEXT: s_mov_b32 s43, s14 ; GCN-NEXT: s_mov_b32 s44, s13 @@ -788,24 +788,24 @@ ; GCN-NEXT: ; %bb.2: ; GCN-NEXT: s_mov_b64 exec, s[46:47] ; GCN-NEXT: v_add_i32_e32 v0, vcc, 1, v2 -; GCN-NEXT: v_readlane_b32 s49, v40, 17 -; GCN-NEXT: v_readlane_b32 s48, v40, 16 -; GCN-NEXT: v_readlane_b32 s47, v40, 15 -; GCN-NEXT: v_readlane_b32 s46, v40, 14 -; GCN-NEXT: v_readlane_b32 s45, v40, 13 -; GCN-NEXT: v_readlane_b32 s44, v40, 12 -; GCN-NEXT: v_readlane_b32 s43, v40, 11 -; GCN-NEXT: v_readlane_b32 s42, v40, 10 -; GCN-NEXT: v_readlane_b32 s41, v40, 9 -; GCN-NEXT: v_readlane_b32 s40, v40, 8 -; GCN-NEXT: v_readlane_b32 s39, v40, 7 -; GCN-NEXT: v_readlane_b32 s38, v40, 6 -; GCN-NEXT: v_readlane_b32 s37, v40, 5 -; GCN-NEXT: v_readlane_b32 s36, v40, 4 -; GCN-NEXT: v_readlane_b32 s35, v40, 3 -; GCN-NEXT: v_readlane_b32 s34, v40, 2 -; GCN-NEXT: v_readlane_b32 s31, v40, 1 -; GCN-NEXT: v_readlane_b32 s30, v40, 0 +; GCN-NEXT: v_readlane_b32 s30, v40, 16 +; GCN-NEXT: v_readlane_b32 s31, v40, 17 +; GCN-NEXT: v_readlane_b32 s49, v40, 15 +; GCN-NEXT: v_readlane_b32 s48, v40, 14 +; GCN-NEXT: v_readlane_b32 s47, v40, 13 +; GCN-NEXT: v_readlane_b32 s46, v40, 12 +; GCN-NEXT: v_readlane_b32 s45, v40, 11 +; GCN-NEXT: v_readlane_b32 s44, v40, 10 +; GCN-NEXT: v_readlane_b32 s43, v40, 9 +; GCN-NEXT: v_readlane_b32 s42, v40, 8 +; GCN-NEXT: v_readlane_b32 s41, v40, 7 +; GCN-NEXT: v_readlane_b32 s40, v40, 6 +; GCN-NEXT: v_readlane_b32 s39, v40, 5 +; GCN-NEXT: v_readlane_b32 s38, v40, 4 +; GCN-NEXT: v_readlane_b32 s37, v40, 3 +; GCN-NEXT: v_readlane_b32 s36, v40, 2 +; GCN-NEXT: v_readlane_b32 s35, v40, 1 +; GCN-NEXT: v_readlane_b32 s34, v40, 0 ; GCN-NEXT: s_addk_i32 s32, 0xfc00 ; GCN-NEXT: v_readlane_b32 s33, v40, 18 ; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -823,24 +823,24 @@ ; GISEL-NEXT: v_writelane_b32 v40, s33, 18 ; GISEL-NEXT: s_mov_b32 s33, s32 ; GISEL-NEXT: s_addk_i32 s32, 0x400 -; GISEL-NEXT: v_writelane_b32 v40, s30, 0 -; GISEL-NEXT: v_writelane_b32 v40, s31, 1 -; GISEL-NEXT: v_writelane_b32 v40, s34, 2 -; GISEL-NEXT: v_writelane_b32 v40, s35, 3 -; GISEL-NEXT: v_writelane_b32 v40, s36, 4 -; GISEL-NEXT: v_writelane_b32 v40, s37, 5 -; GISEL-NEXT: v_writelane_b32 v40, s38, 6 -; GISEL-NEXT: v_writelane_b32 v40, s39, 7 -; GISEL-NEXT: v_writelane_b32 v40, s40, 8 -; GISEL-NEXT: v_writelane_b32 v40, s41, 9 -; GISEL-NEXT: v_writelane_b32 v40, s42, 10 -; GISEL-NEXT: v_writelane_b32 v40, s43, 11 -; GISEL-NEXT: v_writelane_b32 v40, s44, 12 -; GISEL-NEXT: v_writelane_b32 v40, s45, 13 -; GISEL-NEXT: v_writelane_b32 v40, s46, 14 -; GISEL-NEXT: v_writelane_b32 v40, s47, 15 -; GISEL-NEXT: v_writelane_b32 v40, s48, 16 -; GISEL-NEXT: v_writelane_b32 v40, s49, 17 +; GISEL-NEXT: v_writelane_b32 v40, s34, 0 +; GISEL-NEXT: v_writelane_b32 v40, s35, 1 +; GISEL-NEXT: v_writelane_b32 v40, s36, 2 +; GISEL-NEXT: v_writelane_b32 v40, s37, 3 +; GISEL-NEXT: v_writelane_b32 v40, s38, 4 +; GISEL-NEXT: v_writelane_b32 v40, s39, 5 +; GISEL-NEXT: v_writelane_b32 v40, s40, 6 +; GISEL-NEXT: v_writelane_b32 v40, s41, 7 +; GISEL-NEXT: v_writelane_b32 v40, s42, 8 +; GISEL-NEXT: v_writelane_b32 v40, s43, 9 +; GISEL-NEXT: v_writelane_b32 v40, s44, 10 +; GISEL-NEXT: v_writelane_b32 v40, s45, 11 +; GISEL-NEXT: v_writelane_b32 v40, s46, 12 +; GISEL-NEXT: v_writelane_b32 v40, s47, 13 +; GISEL-NEXT: v_writelane_b32 v40, s48, 14 +; GISEL-NEXT: v_writelane_b32 v40, s49, 15 +; GISEL-NEXT: v_writelane_b32 v40, s30, 16 +; GISEL-NEXT: v_writelane_b32 v40, s31, 17 ; GISEL-NEXT: s_mov_b32 s42, s15 ; GISEL-NEXT: s_mov_b32 s43, s14 ; GISEL-NEXT: s_mov_b32 s44, s13 @@ -872,24 +872,24 @@ ; GISEL-NEXT: ; %bb.2: ; GISEL-NEXT: s_mov_b64 exec, s[46:47] ; GISEL-NEXT: v_add_i32_e32 v0, vcc, 1, v1 -; GISEL-NEXT: v_readlane_b32 s49, v40, 17 -; GISEL-NEXT: v_readlane_b32 s48, v40, 16 -; GISEL-NEXT: v_readlane_b32 s47, v40, 15 -; GISEL-NEXT: v_readlane_b32 s46, v40, 14 -; GISEL-NEXT: v_readlane_b32 s45, v40, 13 -; GISEL-NEXT: v_readlane_b32 s44, v40, 12 -; GISEL-NEXT: v_readlane_b32 s43, v40, 11 -; GISEL-NEXT: v_readlane_b32 s42, v40, 10 -; GISEL-NEXT: v_readlane_b32 s41, v40, 9 -; GISEL-NEXT: v_readlane_b32 s40, v40, 8 -; GISEL-NEXT: v_readlane_b32 s39, v40, 7 -; GISEL-NEXT: v_readlane_b32 s38, v40, 6 -; GISEL-NEXT: v_readlane_b32 s37, v40, 5 -; GISEL-NEXT: v_readlane_b32 s36, v40, 4 -; GISEL-NEXT: v_readlane_b32 s35, v40, 3 -; GISEL-NEXT: v_readlane_b32 s34, v40, 2 -; GISEL-NEXT: v_readlane_b32 s31, v40, 1 -; GISEL-NEXT: v_readlane_b32 s30, v40, 0 +; GISEL-NEXT: v_readlane_b32 s30, v40, 16 +; GISEL-NEXT: v_readlane_b32 s31, v40, 17 +; GISEL-NEXT: v_readlane_b32 s49, v40, 15 +; GISEL-NEXT: v_readlane_b32 s48, v40, 14 +; GISEL-NEXT: v_readlane_b32 s47, v40, 13 +; GISEL-NEXT: v_readlane_b32 s46, v40, 12 +; GISEL-NEXT: v_readlane_b32 s45, v40, 11 +; GISEL-NEXT: v_readlane_b32 s44, v40, 10 +; GISEL-NEXT: v_readlane_b32 s43, v40, 9 +; GISEL-NEXT: v_readlane_b32 s42, v40, 8 +; GISEL-NEXT: v_readlane_b32 s41, v40, 7 +; GISEL-NEXT: v_readlane_b32 s40, v40, 6 +; GISEL-NEXT: v_readlane_b32 s39, v40, 5 +; GISEL-NEXT: v_readlane_b32 s38, v40, 4 +; GISEL-NEXT: v_readlane_b32 s37, v40, 3 +; GISEL-NEXT: v_readlane_b32 s36, v40, 2 +; GISEL-NEXT: v_readlane_b32 s35, v40, 1 +; GISEL-NEXT: v_readlane_b32 s34, v40, 0 ; GISEL-NEXT: s_addk_i32 s32, 0xfc00 ; GISEL-NEXT: v_readlane_b32 s33, v40, 18 ; GISEL-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -912,26 +912,26 @@ ; GCN-NEXT: v_writelane_b32 v40, s33, 20 ; GCN-NEXT: s_mov_b32 s33, s32 ; GCN-NEXT: s_addk_i32 s32, 0x400 -; GCN-NEXT: v_writelane_b32 v40, s30, 0 -; GCN-NEXT: v_writelane_b32 v40, s31, 1 -; GCN-NEXT: v_writelane_b32 v40, s34, 2 -; GCN-NEXT: v_writelane_b32 v40, s35, 3 -; GCN-NEXT: v_writelane_b32 v40, s36, 4 -; GCN-NEXT: v_writelane_b32 v40, s37, 5 -; GCN-NEXT: v_writelane_b32 v40, s38, 6 -; GCN-NEXT: v_writelane_b32 v40, s39, 7 -; GCN-NEXT: v_writelane_b32 v40, s40, 8 -; GCN-NEXT: v_writelane_b32 v40, s41, 9 -; GCN-NEXT: v_writelane_b32 v40, s42, 10 -; GCN-NEXT: v_writelane_b32 v40, s43, 11 -; GCN-NEXT: v_writelane_b32 v40, s44, 12 -; GCN-NEXT: v_writelane_b32 v40, s45, 13 -; GCN-NEXT: v_writelane_b32 v40, s46, 14 -; GCN-NEXT: v_writelane_b32 v40, s47, 15 -; GCN-NEXT: v_writelane_b32 v40, s48, 16 -; GCN-NEXT: v_writelane_b32 v40, s49, 17 -; GCN-NEXT: v_writelane_b32 v40, s50, 18 -; GCN-NEXT: v_writelane_b32 v40, s51, 19 +; GCN-NEXT: v_writelane_b32 v40, s34, 0 +; GCN-NEXT: v_writelane_b32 v40, s35, 1 +; GCN-NEXT: v_writelane_b32 v40, s36, 2 +; GCN-NEXT: v_writelane_b32 v40, s37, 3 +; GCN-NEXT: v_writelane_b32 v40, s38, 4 +; GCN-NEXT: v_writelane_b32 v40, s39, 5 +; GCN-NEXT: v_writelane_b32 v40, s40, 6 +; GCN-NEXT: v_writelane_b32 v40, s41, 7 +; GCN-NEXT: v_writelane_b32 v40, s42, 8 +; GCN-NEXT: v_writelane_b32 v40, s43, 9 +; GCN-NEXT: v_writelane_b32 v40, s44, 10 +; GCN-NEXT: v_writelane_b32 v40, s45, 11 +; GCN-NEXT: v_writelane_b32 v40, s46, 12 +; GCN-NEXT: v_writelane_b32 v40, s47, 13 +; GCN-NEXT: v_writelane_b32 v40, s48, 14 +; GCN-NEXT: v_writelane_b32 v40, s49, 15 +; GCN-NEXT: v_writelane_b32 v40, s50, 16 +; GCN-NEXT: v_writelane_b32 v40, s51, 17 +; GCN-NEXT: v_writelane_b32 v40, s30, 18 +; GCN-NEXT: v_writelane_b32 v40, s31, 19 ; GCN-NEXT: s_mov_b32 s42, s15 ; GCN-NEXT: s_mov_b32 s43, s14 ; GCN-NEXT: s_mov_b32 s44, s13 @@ -968,26 +968,26 @@ ; GCN-NEXT: s_mov_b64 exec, s[48:49] ; GCN-NEXT: .LBB5_4: ; %bb2 ; GCN-NEXT: s_or_b64 exec, exec, s[46:47] -; GCN-NEXT: v_readlane_b32 s51, v40, 19 -; GCN-NEXT: v_readlane_b32 s50, v40, 18 -; GCN-NEXT: v_readlane_b32 s49, v40, 17 -; GCN-NEXT: v_readlane_b32 s48, v40, 16 -; GCN-NEXT: v_readlane_b32 s47, v40, 15 -; GCN-NEXT: v_readlane_b32 s46, v40, 14 -; GCN-NEXT: v_readlane_b32 s45, v40, 13 -; GCN-NEXT: v_readlane_b32 s44, v40, 12 -; GCN-NEXT: v_readlane_b32 s43, v40, 11 -; GCN-NEXT: v_readlane_b32 s42, v40, 10 -; GCN-NEXT: v_readlane_b32 s41, v40, 9 -; GCN-NEXT: v_readlane_b32 s40, v40, 8 -; GCN-NEXT: v_readlane_b32 s39, v40, 7 -; GCN-NEXT: v_readlane_b32 s38, v40, 6 -; GCN-NEXT: v_readlane_b32 s37, v40, 5 -; GCN-NEXT: v_readlane_b32 s36, v40, 4 -; GCN-NEXT: v_readlane_b32 s35, v40, 3 -; GCN-NEXT: v_readlane_b32 s34, v40, 2 -; GCN-NEXT: v_readlane_b32 s31, v40, 1 -; GCN-NEXT: v_readlane_b32 s30, v40, 0 +; GCN-NEXT: v_readlane_b32 s30, v40, 18 +; GCN-NEXT: v_readlane_b32 s31, v40, 19 +; GCN-NEXT: v_readlane_b32 s51, v40, 17 +; GCN-NEXT: v_readlane_b32 s50, v40, 16 +; GCN-NEXT: v_readlane_b32 s49, v40, 15 +; GCN-NEXT: v_readlane_b32 s48, v40, 14 +; GCN-NEXT: v_readlane_b32 s47, v40, 13 +; GCN-NEXT: v_readlane_b32 s46, v40, 12 +; GCN-NEXT: v_readlane_b32 s45, v40, 11 +; GCN-NEXT: v_readlane_b32 s44, v40, 10 +; GCN-NEXT: v_readlane_b32 s43, v40, 9 +; GCN-NEXT: v_readlane_b32 s42, v40, 8 +; GCN-NEXT: v_readlane_b32 s41, v40, 7 +; GCN-NEXT: v_readlane_b32 s40, v40, 6 +; GCN-NEXT: v_readlane_b32 s39, v40, 5 +; GCN-NEXT: v_readlane_b32 s38, v40, 4 +; GCN-NEXT: v_readlane_b32 s37, v40, 3 +; GCN-NEXT: v_readlane_b32 s36, v40, 2 +; GCN-NEXT: v_readlane_b32 s35, v40, 1 +; GCN-NEXT: v_readlane_b32 s34, v40, 0 ; GCN-NEXT: s_addk_i32 s32, 0xfc00 ; GCN-NEXT: v_readlane_b32 s33, v40, 20 ; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -1005,26 +1005,26 @@ ; GISEL-NEXT: v_writelane_b32 v40, s33, 20 ; GISEL-NEXT: s_mov_b32 s33, s32 ; GISEL-NEXT: s_addk_i32 s32, 0x400 -; GISEL-NEXT: v_writelane_b32 v40, s30, 0 -; GISEL-NEXT: v_writelane_b32 v40, s31, 1 -; GISEL-NEXT: v_writelane_b32 v40, s34, 2 -; GISEL-NEXT: v_writelane_b32 v40, s35, 3 -; GISEL-NEXT: v_writelane_b32 v40, s36, 4 -; GISEL-NEXT: v_writelane_b32 v40, s37, 5 -; GISEL-NEXT: v_writelane_b32 v40, s38, 6 -; GISEL-NEXT: v_writelane_b32 v40, s39, 7 -; GISEL-NEXT: v_writelane_b32 v40, s40, 8 -; GISEL-NEXT: v_writelane_b32 v40, s41, 9 -; GISEL-NEXT: v_writelane_b32 v40, s42, 10 -; GISEL-NEXT: v_writelane_b32 v40, s43, 11 -; GISEL-NEXT: v_writelane_b32 v40, s44, 12 -; GISEL-NEXT: v_writelane_b32 v40, s45, 13 -; GISEL-NEXT: v_writelane_b32 v40, s46, 14 -; GISEL-NEXT: v_writelane_b32 v40, s47, 15 -; GISEL-NEXT: v_writelane_b32 v40, s48, 16 -; GISEL-NEXT: v_writelane_b32 v40, s49, 17 -; GISEL-NEXT: v_writelane_b32 v40, s50, 18 -; GISEL-NEXT: v_writelane_b32 v40, s51, 19 +; GISEL-NEXT: v_writelane_b32 v40, s34, 0 +; GISEL-NEXT: v_writelane_b32 v40, s35, 1 +; GISEL-NEXT: v_writelane_b32 v40, s36, 2 +; GISEL-NEXT: v_writelane_b32 v40, s37, 3 +; GISEL-NEXT: v_writelane_b32 v40, s38, 4 +; GISEL-NEXT: v_writelane_b32 v40, s39, 5 +; GISEL-NEXT: v_writelane_b32 v40, s40, 6 +; GISEL-NEXT: v_writelane_b32 v40, s41, 7 +; GISEL-NEXT: v_writelane_b32 v40, s42, 8 +; GISEL-NEXT: v_writelane_b32 v40, s43, 9 +; GISEL-NEXT: v_writelane_b32 v40, s44, 10 +; GISEL-NEXT: v_writelane_b32 v40, s45, 11 +; GISEL-NEXT: v_writelane_b32 v40, s46, 12 +; GISEL-NEXT: v_writelane_b32 v40, s47, 13 +; GISEL-NEXT: v_writelane_b32 v40, s48, 14 +; GISEL-NEXT: v_writelane_b32 v40, s49, 15 +; GISEL-NEXT: v_writelane_b32 v40, s50, 16 +; GISEL-NEXT: v_writelane_b32 v40, s51, 17 +; GISEL-NEXT: v_writelane_b32 v40, s30, 18 +; GISEL-NEXT: v_writelane_b32 v40, s31, 19 ; GISEL-NEXT: s_mov_b32 s42, s15 ; GISEL-NEXT: s_mov_b32 s43, s14 ; GISEL-NEXT: s_mov_b32 s44, s13 @@ -1061,26 +1061,26 @@ ; GISEL-NEXT: s_mov_b64 exec, s[48:49] ; GISEL-NEXT: .LBB5_4: ; %bb2 ; GISEL-NEXT: s_or_b64 exec, exec, s[46:47] -; GISEL-NEXT: v_readlane_b32 s51, v40, 19 -; GISEL-NEXT: v_readlane_b32 s50, v40, 18 -; GISEL-NEXT: v_readlane_b32 s49, v40, 17 -; GISEL-NEXT: v_readlane_b32 s48, v40, 16 -; GISEL-NEXT: v_readlane_b32 s47, v40, 15 -; GISEL-NEXT: v_readlane_b32 s46, v40, 14 -; GISEL-NEXT: v_readlane_b32 s45, v40, 13 -; GISEL-NEXT: v_readlane_b32 s44, v40, 12 -; GISEL-NEXT: v_readlane_b32 s43, v40, 11 -; GISEL-NEXT: v_readlane_b32 s42, v40, 10 -; GISEL-NEXT: v_readlane_b32 s41, v40, 9 -; GISEL-NEXT: v_readlane_b32 s40, v40, 8 -; GISEL-NEXT: v_readlane_b32 s39, v40, 7 -; GISEL-NEXT: v_readlane_b32 s38, v40, 6 -; GISEL-NEXT: v_readlane_b32 s37, v40, 5 -; GISEL-NEXT: v_readlane_b32 s36, v40, 4 -; GISEL-NEXT: v_readlane_b32 s35, v40, 3 -; GISEL-NEXT: v_readlane_b32 s34, v40, 2 -; GISEL-NEXT: v_readlane_b32 s31, v40, 1 -; GISEL-NEXT: v_readlane_b32 s30, v40, 0 +; GISEL-NEXT: v_readlane_b32 s30, v40, 18 +; GISEL-NEXT: v_readlane_b32 s31, v40, 19 +; GISEL-NEXT: v_readlane_b32 s51, v40, 17 +; GISEL-NEXT: v_readlane_b32 s50, v40, 16 +; GISEL-NEXT: v_readlane_b32 s49, v40, 15 +; GISEL-NEXT: v_readlane_b32 s48, v40, 14 +; GISEL-NEXT: v_readlane_b32 s47, v40, 13 +; GISEL-NEXT: v_readlane_b32 s46, v40, 12 +; GISEL-NEXT: v_readlane_b32 s45, v40, 11 +; GISEL-NEXT: v_readlane_b32 s44, v40, 10 +; GISEL-NEXT: v_readlane_b32 s43, v40, 9 +; GISEL-NEXT: v_readlane_b32 s42, v40, 8 +; GISEL-NEXT: v_readlane_b32 s41, v40, 7 +; GISEL-NEXT: v_readlane_b32 s40, v40, 6 +; GISEL-NEXT: v_readlane_b32 s39, v40, 5 +; GISEL-NEXT: v_readlane_b32 s38, v40, 4 +; GISEL-NEXT: v_readlane_b32 s37, v40, 3 +; GISEL-NEXT: v_readlane_b32 s36, v40, 2 +; GISEL-NEXT: v_readlane_b32 s35, v40, 1 +; GISEL-NEXT: v_readlane_b32 s34, v40, 0 ; GISEL-NEXT: s_addk_i32 s32, 0xfc00 ; GISEL-NEXT: v_readlane_b32 s33, v40, 20 ; GISEL-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -1109,38 +1109,38 @@ ; GCN-NEXT: v_writelane_b32 v40, s33, 32 ; GCN-NEXT: s_mov_b32 s33, s32 ; GCN-NEXT: s_addk_i32 s32, 0x400 -; GCN-NEXT: v_writelane_b32 v40, s30, 0 -; GCN-NEXT: v_writelane_b32 v40, s31, 1 -; GCN-NEXT: v_writelane_b32 v40, s34, 2 -; GCN-NEXT: v_writelane_b32 v40, s35, 3 -; GCN-NEXT: v_writelane_b32 v40, s36, 4 -; GCN-NEXT: v_writelane_b32 v40, s37, 5 -; GCN-NEXT: v_writelane_b32 v40, s38, 6 -; GCN-NEXT: v_writelane_b32 v40, s39, 7 -; GCN-NEXT: v_writelane_b32 v40, s40, 8 -; GCN-NEXT: v_writelane_b32 v40, s41, 9 -; GCN-NEXT: v_writelane_b32 v40, s42, 10 -; GCN-NEXT: v_writelane_b32 v40, s43, 11 -; GCN-NEXT: v_writelane_b32 v40, s44, 12 -; GCN-NEXT: v_writelane_b32 v40, s45, 13 -; GCN-NEXT: v_writelane_b32 v40, s46, 14 -; GCN-NEXT: v_writelane_b32 v40, s47, 15 -; GCN-NEXT: v_writelane_b32 v40, s48, 16 -; GCN-NEXT: v_writelane_b32 v40, s49, 17 -; GCN-NEXT: v_writelane_b32 v40, s50, 18 -; GCN-NEXT: v_writelane_b32 v40, s51, 19 -; GCN-NEXT: v_writelane_b32 v40, s52, 20 -; GCN-NEXT: v_writelane_b32 v40, s53, 21 -; GCN-NEXT: v_writelane_b32 v40, s54, 22 -; GCN-NEXT: v_writelane_b32 v40, s55, 23 -; GCN-NEXT: v_writelane_b32 v40, s56, 24 -; GCN-NEXT: v_writelane_b32 v40, s57, 25 -; GCN-NEXT: v_writelane_b32 v40, s58, 26 -; GCN-NEXT: v_writelane_b32 v40, s59, 27 -; GCN-NEXT: v_writelane_b32 v40, s60, 28 -; GCN-NEXT: v_writelane_b32 v40, s61, 29 -; GCN-NEXT: v_writelane_b32 v40, s62, 30 -; GCN-NEXT: v_writelane_b32 v40, s63, 31 +; GCN-NEXT: v_writelane_b32 v40, s34, 0 +; GCN-NEXT: v_writelane_b32 v40, s35, 1 +; GCN-NEXT: v_writelane_b32 v40, s36, 2 +; GCN-NEXT: v_writelane_b32 v40, s37, 3 +; GCN-NEXT: v_writelane_b32 v40, s38, 4 +; GCN-NEXT: v_writelane_b32 v40, s39, 5 +; GCN-NEXT: v_writelane_b32 v40, s40, 6 +; GCN-NEXT: v_writelane_b32 v40, s41, 7 +; GCN-NEXT: v_writelane_b32 v40, s42, 8 +; GCN-NEXT: v_writelane_b32 v40, s43, 9 +; GCN-NEXT: v_writelane_b32 v40, s44, 10 +; GCN-NEXT: v_writelane_b32 v40, s45, 11 +; GCN-NEXT: v_writelane_b32 v40, s46, 12 +; GCN-NEXT: v_writelane_b32 v40, s47, 13 +; GCN-NEXT: v_writelane_b32 v40, s48, 14 +; GCN-NEXT: v_writelane_b32 v40, s49, 15 +; GCN-NEXT: v_writelane_b32 v40, s50, 16 +; GCN-NEXT: v_writelane_b32 v40, s51, 17 +; GCN-NEXT: v_writelane_b32 v40, s52, 18 +; GCN-NEXT: v_writelane_b32 v40, s53, 19 +; GCN-NEXT: v_writelane_b32 v40, s54, 20 +; GCN-NEXT: v_writelane_b32 v40, s55, 21 +; GCN-NEXT: v_writelane_b32 v40, s56, 22 +; GCN-NEXT: v_writelane_b32 v40, s57, 23 +; GCN-NEXT: v_writelane_b32 v40, s58, 24 +; GCN-NEXT: v_writelane_b32 v40, s59, 25 +; GCN-NEXT: v_writelane_b32 v40, s60, 26 +; GCN-NEXT: v_writelane_b32 v40, s61, 27 +; GCN-NEXT: v_writelane_b32 v40, s62, 28 +; GCN-NEXT: v_writelane_b32 v40, s63, 29 +; GCN-NEXT: v_writelane_b32 v40, s30, 30 +; GCN-NEXT: v_writelane_b32 v40, s31, 31 ; GCN-NEXT: s_mov_b64 s[6:7], exec ; GCN-NEXT: s_movk_i32 s4, 0x7b ; GCN-NEXT: .LBB6_1: ; =>This Inner Loop Header: Depth=1 @@ -1154,38 +1154,38 @@ ; GCN-NEXT: s_cbranch_execnz .LBB6_1 ; GCN-NEXT: ; %bb.2: ; GCN-NEXT: s_mov_b64 exec, s[6:7] -; GCN-NEXT: v_readlane_b32 s63, v40, 31 -; GCN-NEXT: v_readlane_b32 s62, v40, 30 -; GCN-NEXT: v_readlane_b32 s61, v40, 29 -; GCN-NEXT: v_readlane_b32 s60, v40, 28 -; GCN-NEXT: v_readlane_b32 s59, v40, 27 -; GCN-NEXT: v_readlane_b32 s58, v40, 26 -; GCN-NEXT: v_readlane_b32 s57, v40, 25 -; GCN-NEXT: v_readlane_b32 s56, v40, 24 -; GCN-NEXT: v_readlane_b32 s55, v40, 23 -; GCN-NEXT: v_readlane_b32 s54, v40, 22 -; GCN-NEXT: v_readlane_b32 s53, v40, 21 -; GCN-NEXT: v_readlane_b32 s52, v40, 20 -; GCN-NEXT: v_readlane_b32 s51, v40, 19 -; GCN-NEXT: v_readlane_b32 s50, v40, 18 -; GCN-NEXT: v_readlane_b32 s49, v40, 17 -; GCN-NEXT: v_readlane_b32 s48, v40, 16 -; GCN-NEXT: v_readlane_b32 s47, v40, 15 -; GCN-NEXT: v_readlane_b32 s46, v40, 14 -; GCN-NEXT: v_readlane_b32 s45, v40, 13 -; GCN-NEXT: v_readlane_b32 s44, v40, 12 -; GCN-NEXT: v_readlane_b32 s43, v40, 11 -; GCN-NEXT: v_readlane_b32 s42, v40, 10 -; GCN-NEXT: v_readlane_b32 s41, v40, 9 -; GCN-NEXT: v_readlane_b32 s40, v40, 8 -; GCN-NEXT: v_readlane_b32 s39, v40, 7 -; GCN-NEXT: v_readlane_b32 s38, v40, 6 -; GCN-NEXT: v_readlane_b32 s37, v40, 5 -; GCN-NEXT: v_readlane_b32 s36, v40, 4 -; GCN-NEXT: v_readlane_b32 s35, v40, 3 -; GCN-NEXT: v_readlane_b32 s34, v40, 2 -; GCN-NEXT: v_readlane_b32 s31, v40, 1 -; GCN-NEXT: v_readlane_b32 s30, v40, 0 +; GCN-NEXT: v_readlane_b32 s30, v40, 30 +; GCN-NEXT: v_readlane_b32 s31, v40, 31 +; GCN-NEXT: v_readlane_b32 s63, v40, 29 +; GCN-NEXT: v_readlane_b32 s62, v40, 28 +; GCN-NEXT: v_readlane_b32 s61, v40, 27 +; GCN-NEXT: v_readlane_b32 s60, v40, 26 +; GCN-NEXT: v_readlane_b32 s59, v40, 25 +; GCN-NEXT: v_readlane_b32 s58, v40, 24 +; GCN-NEXT: v_readlane_b32 s57, v40, 23 +; GCN-NEXT: v_readlane_b32 s56, v40, 22 +; GCN-NEXT: v_readlane_b32 s55, v40, 21 +; GCN-NEXT: v_readlane_b32 s54, v40, 20 +; GCN-NEXT: v_readlane_b32 s53, v40, 19 +; GCN-NEXT: v_readlane_b32 s52, v40, 18 +; GCN-NEXT: v_readlane_b32 s51, v40, 17 +; GCN-NEXT: v_readlane_b32 s50, v40, 16 +; GCN-NEXT: v_readlane_b32 s49, v40, 15 +; GCN-NEXT: v_readlane_b32 s48, v40, 14 +; GCN-NEXT: v_readlane_b32 s47, v40, 13 +; GCN-NEXT: v_readlane_b32 s46, v40, 12 +; GCN-NEXT: v_readlane_b32 s45, v40, 11 +; GCN-NEXT: v_readlane_b32 s44, v40, 10 +; GCN-NEXT: v_readlane_b32 s43, v40, 9 +; GCN-NEXT: v_readlane_b32 s42, v40, 8 +; GCN-NEXT: v_readlane_b32 s41, v40, 7 +; GCN-NEXT: v_readlane_b32 s40, v40, 6 +; GCN-NEXT: v_readlane_b32 s39, v40, 5 +; GCN-NEXT: v_readlane_b32 s38, v40, 4 +; GCN-NEXT: v_readlane_b32 s37, v40, 3 +; GCN-NEXT: v_readlane_b32 s36, v40, 2 +; GCN-NEXT: v_readlane_b32 s35, v40, 1 +; GCN-NEXT: v_readlane_b32 s34, v40, 0 ; GCN-NEXT: s_addk_i32 s32, 0xfc00 ; GCN-NEXT: v_readlane_b32 s33, v40, 32 ; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -1203,38 +1203,38 @@ ; GISEL-NEXT: v_writelane_b32 v40, s33, 32 ; GISEL-NEXT: s_mov_b32 s33, s32 ; GISEL-NEXT: s_addk_i32 s32, 0x400 -; GISEL-NEXT: v_writelane_b32 v40, s30, 0 -; GISEL-NEXT: v_writelane_b32 v40, s31, 1 -; GISEL-NEXT: v_writelane_b32 v40, s34, 2 -; GISEL-NEXT: v_writelane_b32 v40, s35, 3 -; GISEL-NEXT: v_writelane_b32 v40, s36, 4 -; GISEL-NEXT: v_writelane_b32 v40, s37, 5 -; GISEL-NEXT: v_writelane_b32 v40, s38, 6 -; GISEL-NEXT: v_writelane_b32 v40, s39, 7 -; GISEL-NEXT: v_writelane_b32 v40, s40, 8 -; GISEL-NEXT: v_writelane_b32 v40, s41, 9 -; GISEL-NEXT: v_writelane_b32 v40, s42, 10 -; GISEL-NEXT: v_writelane_b32 v40, s43, 11 -; GISEL-NEXT: v_writelane_b32 v40, s44, 12 -; GISEL-NEXT: v_writelane_b32 v40, s45, 13 -; GISEL-NEXT: v_writelane_b32 v40, s46, 14 -; GISEL-NEXT: v_writelane_b32 v40, s47, 15 -; GISEL-NEXT: v_writelane_b32 v40, s48, 16 -; GISEL-NEXT: v_writelane_b32 v40, s49, 17 -; GISEL-NEXT: v_writelane_b32 v40, s50, 18 -; GISEL-NEXT: v_writelane_b32 v40, s51, 19 -; GISEL-NEXT: v_writelane_b32 v40, s52, 20 -; GISEL-NEXT: v_writelane_b32 v40, s53, 21 -; GISEL-NEXT: v_writelane_b32 v40, s54, 22 -; GISEL-NEXT: v_writelane_b32 v40, s55, 23 -; GISEL-NEXT: v_writelane_b32 v40, s56, 24 -; GISEL-NEXT: v_writelane_b32 v40, s57, 25 -; GISEL-NEXT: v_writelane_b32 v40, s58, 26 -; GISEL-NEXT: v_writelane_b32 v40, s59, 27 -; GISEL-NEXT: v_writelane_b32 v40, s60, 28 -; GISEL-NEXT: v_writelane_b32 v40, s61, 29 -; GISEL-NEXT: v_writelane_b32 v40, s62, 30 -; GISEL-NEXT: v_writelane_b32 v40, s63, 31 +; GISEL-NEXT: v_writelane_b32 v40, s34, 0 +; GISEL-NEXT: v_writelane_b32 v40, s35, 1 +; GISEL-NEXT: v_writelane_b32 v40, s36, 2 +; GISEL-NEXT: v_writelane_b32 v40, s37, 3 +; GISEL-NEXT: v_writelane_b32 v40, s38, 4 +; GISEL-NEXT: v_writelane_b32 v40, s39, 5 +; GISEL-NEXT: v_writelane_b32 v40, s40, 6 +; GISEL-NEXT: v_writelane_b32 v40, s41, 7 +; GISEL-NEXT: v_writelane_b32 v40, s42, 8 +; GISEL-NEXT: v_writelane_b32 v40, s43, 9 +; GISEL-NEXT: v_writelane_b32 v40, s44, 10 +; GISEL-NEXT: v_writelane_b32 v40, s45, 11 +; GISEL-NEXT: v_writelane_b32 v40, s46, 12 +; GISEL-NEXT: v_writelane_b32 v40, s47, 13 +; GISEL-NEXT: v_writelane_b32 v40, s48, 14 +; GISEL-NEXT: v_writelane_b32 v40, s49, 15 +; GISEL-NEXT: v_writelane_b32 v40, s50, 16 +; GISEL-NEXT: v_writelane_b32 v40, s51, 17 +; GISEL-NEXT: v_writelane_b32 v40, s52, 18 +; GISEL-NEXT: v_writelane_b32 v40, s53, 19 +; GISEL-NEXT: v_writelane_b32 v40, s54, 20 +; GISEL-NEXT: v_writelane_b32 v40, s55, 21 +; GISEL-NEXT: v_writelane_b32 v40, s56, 22 +; GISEL-NEXT: v_writelane_b32 v40, s57, 23 +; GISEL-NEXT: v_writelane_b32 v40, s58, 24 +; GISEL-NEXT: v_writelane_b32 v40, s59, 25 +; GISEL-NEXT: v_writelane_b32 v40, s60, 26 +; GISEL-NEXT: v_writelane_b32 v40, s61, 27 +; GISEL-NEXT: v_writelane_b32 v40, s62, 28 +; GISEL-NEXT: v_writelane_b32 v40, s63, 29 +; GISEL-NEXT: v_writelane_b32 v40, s30, 30 +; GISEL-NEXT: v_writelane_b32 v40, s31, 31 ; GISEL-NEXT: s_mov_b64 s[6:7], exec ; GISEL-NEXT: s_movk_i32 s4, 0x7b ; GISEL-NEXT: .LBB6_1: ; =>This Inner Loop Header: Depth=1 @@ -1248,38 +1248,38 @@ ; GISEL-NEXT: s_cbranch_execnz .LBB6_1 ; GISEL-NEXT: ; %bb.2: ; GISEL-NEXT: s_mov_b64 exec, s[6:7] -; GISEL-NEXT: v_readlane_b32 s63, v40, 31 -; GISEL-NEXT: v_readlane_b32 s62, v40, 30 -; GISEL-NEXT: v_readlane_b32 s61, v40, 29 -; GISEL-NEXT: v_readlane_b32 s60, v40, 28 -; GISEL-NEXT: v_readlane_b32 s59, v40, 27 -; GISEL-NEXT: v_readlane_b32 s58, v40, 26 -; GISEL-NEXT: v_readlane_b32 s57, v40, 25 -; GISEL-NEXT: v_readlane_b32 s56, v40, 24 -; GISEL-NEXT: v_readlane_b32 s55, v40, 23 -; GISEL-NEXT: v_readlane_b32 s54, v40, 22 -; GISEL-NEXT: v_readlane_b32 s53, v40, 21 -; GISEL-NEXT: v_readlane_b32 s52, v40, 20 -; GISEL-NEXT: v_readlane_b32 s51, v40, 19 -; GISEL-NEXT: v_readlane_b32 s50, v40, 18 -; GISEL-NEXT: v_readlane_b32 s49, v40, 17 -; GISEL-NEXT: v_readlane_b32 s48, v40, 16 -; GISEL-NEXT: v_readlane_b32 s47, v40, 15 -; GISEL-NEXT: v_readlane_b32 s46, v40, 14 -; GISEL-NEXT: v_readlane_b32 s45, v40, 13 -; GISEL-NEXT: v_readlane_b32 s44, v40, 12 -; GISEL-NEXT: v_readlane_b32 s43, v40, 11 -; GISEL-NEXT: v_readlane_b32 s42, v40, 10 -; GISEL-NEXT: v_readlane_b32 s41, v40, 9 -; GISEL-NEXT: v_readlane_b32 s40, v40, 8 -; GISEL-NEXT: v_readlane_b32 s39, v40, 7 -; GISEL-NEXT: v_readlane_b32 s38, v40, 6 -; GISEL-NEXT: v_readlane_b32 s37, v40, 5 -; GISEL-NEXT: v_readlane_b32 s36, v40, 4 -; GISEL-NEXT: v_readlane_b32 s35, v40, 3 -; GISEL-NEXT: v_readlane_b32 s34, v40, 2 -; GISEL-NEXT: v_readlane_b32 s31, v40, 1 -; GISEL-NEXT: v_readlane_b32 s30, v40, 0 +; GISEL-NEXT: v_readlane_b32 s30, v40, 30 +; GISEL-NEXT: v_readlane_b32 s31, v40, 31 +; GISEL-NEXT: v_readlane_b32 s63, v40, 29 +; GISEL-NEXT: v_readlane_b32 s62, v40, 28 +; GISEL-NEXT: v_readlane_b32 s61, v40, 27 +; GISEL-NEXT: v_readlane_b32 s60, v40, 26 +; GISEL-NEXT: v_readlane_b32 s59, v40, 25 +; GISEL-NEXT: v_readlane_b32 s58, v40, 24 +; GISEL-NEXT: v_readlane_b32 s57, v40, 23 +; GISEL-NEXT: v_readlane_b32 s56, v40, 22 +; GISEL-NEXT: v_readlane_b32 s55, v40, 21 +; GISEL-NEXT: v_readlane_b32 s54, v40, 20 +; GISEL-NEXT: v_readlane_b32 s53, v40, 19 +; GISEL-NEXT: v_readlane_b32 s52, v40, 18 +; GISEL-NEXT: v_readlane_b32 s51, v40, 17 +; GISEL-NEXT: v_readlane_b32 s50, v40, 16 +; GISEL-NEXT: v_readlane_b32 s49, v40, 15 +; GISEL-NEXT: v_readlane_b32 s48, v40, 14 +; GISEL-NEXT: v_readlane_b32 s47, v40, 13 +; GISEL-NEXT: v_readlane_b32 s46, v40, 12 +; GISEL-NEXT: v_readlane_b32 s45, v40, 11 +; GISEL-NEXT: v_readlane_b32 s44, v40, 10 +; GISEL-NEXT: v_readlane_b32 s43, v40, 9 +; GISEL-NEXT: v_readlane_b32 s42, v40, 8 +; GISEL-NEXT: v_readlane_b32 s41, v40, 7 +; GISEL-NEXT: v_readlane_b32 s40, v40, 6 +; GISEL-NEXT: v_readlane_b32 s39, v40, 5 +; GISEL-NEXT: v_readlane_b32 s38, v40, 4 +; GISEL-NEXT: v_readlane_b32 s37, v40, 3 +; GISEL-NEXT: v_readlane_b32 s36, v40, 2 +; GISEL-NEXT: v_readlane_b32 s35, v40, 1 +; GISEL-NEXT: v_readlane_b32 s34, v40, 0 ; GISEL-NEXT: s_addk_i32 s32, 0xfc00 ; GISEL-NEXT: v_readlane_b32 s33, v40, 32 ; GISEL-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -1302,38 +1302,38 @@ ; GCN-NEXT: s_mov_b32 s33, s32 ; GCN-NEXT: s_addk_i32 s32, 0x400 ; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill -; GCN-NEXT: v_writelane_b32 v40, s30, 0 -; GCN-NEXT: v_writelane_b32 v40, s31, 1 -; GCN-NEXT: v_writelane_b32 v40, s34, 2 -; GCN-NEXT: v_writelane_b32 v40, s35, 3 -; GCN-NEXT: v_writelane_b32 v40, s36, 4 -; GCN-NEXT: v_writelane_b32 v40, s37, 5 -; GCN-NEXT: v_writelane_b32 v40, s38, 6 -; GCN-NEXT: v_writelane_b32 v40, s39, 7 -; GCN-NEXT: v_writelane_b32 v40, s40, 8 -; GCN-NEXT: v_writelane_b32 v40, s41, 9 -; GCN-NEXT: v_writelane_b32 v40, s42, 10 -; GCN-NEXT: v_writelane_b32 v40, s43, 11 -; GCN-NEXT: v_writelane_b32 v40, s44, 12 -; GCN-NEXT: v_writelane_b32 v40, s45, 13 -; GCN-NEXT: v_writelane_b32 v40, s46, 14 -; GCN-NEXT: v_writelane_b32 v40, s47, 15 -; GCN-NEXT: v_writelane_b32 v40, s48, 16 -; GCN-NEXT: v_writelane_b32 v40, s49, 17 -; GCN-NEXT: v_writelane_b32 v40, s50, 18 -; GCN-NEXT: v_writelane_b32 v40, s51, 19 -; GCN-NEXT: v_writelane_b32 v40, s52, 20 -; GCN-NEXT: v_writelane_b32 v40, s53, 21 -; GCN-NEXT: v_writelane_b32 v40, s54, 22 -; GCN-NEXT: v_writelane_b32 v40, s55, 23 -; GCN-NEXT: v_writelane_b32 v40, s56, 24 -; GCN-NEXT: v_writelane_b32 v40, s57, 25 -; GCN-NEXT: v_writelane_b32 v40, s58, 26 -; GCN-NEXT: v_writelane_b32 v40, s59, 27 -; GCN-NEXT: v_writelane_b32 v40, s60, 28 -; GCN-NEXT: v_writelane_b32 v40, s61, 29 -; GCN-NEXT: v_writelane_b32 v40, s62, 30 -; GCN-NEXT: v_writelane_b32 v40, s63, 31 +; GCN-NEXT: v_writelane_b32 v40, s34, 0 +; GCN-NEXT: v_writelane_b32 v40, s35, 1 +; GCN-NEXT: v_writelane_b32 v40, s36, 2 +; GCN-NEXT: v_writelane_b32 v40, s37, 3 +; GCN-NEXT: v_writelane_b32 v40, s38, 4 +; GCN-NEXT: v_writelane_b32 v40, s39, 5 +; GCN-NEXT: v_writelane_b32 v40, s40, 6 +; GCN-NEXT: v_writelane_b32 v40, s41, 7 +; GCN-NEXT: v_writelane_b32 v40, s42, 8 +; GCN-NEXT: v_writelane_b32 v40, s43, 9 +; GCN-NEXT: v_writelane_b32 v40, s44, 10 +; GCN-NEXT: v_writelane_b32 v40, s45, 11 +; GCN-NEXT: v_writelane_b32 v40, s46, 12 +; GCN-NEXT: v_writelane_b32 v40, s47, 13 +; GCN-NEXT: v_writelane_b32 v40, s48, 14 +; GCN-NEXT: v_writelane_b32 v40, s49, 15 +; GCN-NEXT: v_writelane_b32 v40, s50, 16 +; GCN-NEXT: v_writelane_b32 v40, s51, 17 +; GCN-NEXT: v_writelane_b32 v40, s52, 18 +; GCN-NEXT: v_writelane_b32 v40, s53, 19 +; GCN-NEXT: v_writelane_b32 v40, s54, 20 +; GCN-NEXT: v_writelane_b32 v40, s55, 21 +; GCN-NEXT: v_writelane_b32 v40, s56, 22 +; GCN-NEXT: v_writelane_b32 v40, s57, 23 +; GCN-NEXT: v_writelane_b32 v40, s58, 24 +; GCN-NEXT: v_writelane_b32 v40, s59, 25 +; GCN-NEXT: v_writelane_b32 v40, s60, 26 +; GCN-NEXT: v_writelane_b32 v40, s61, 27 +; GCN-NEXT: v_writelane_b32 v40, s62, 28 +; GCN-NEXT: v_writelane_b32 v40, s63, 29 +; GCN-NEXT: v_writelane_b32 v40, s30, 30 +; GCN-NEXT: v_writelane_b32 v40, s31, 31 ; GCN-NEXT: v_mov_b32_e32 v41, v0 ; GCN-NEXT: s_mov_b64 s[4:5], exec ; GCN-NEXT: .LBB7_1: ; =>This Inner Loop Header: Depth=1 @@ -1349,38 +1349,38 @@ ; GCN-NEXT: ; %bb.2: ; GCN-NEXT: s_mov_b64 exec, s[4:5] ; GCN-NEXT: v_mov_b32_e32 v0, v41 -; GCN-NEXT: v_readlane_b32 s63, v40, 31 -; GCN-NEXT: v_readlane_b32 s62, v40, 30 -; GCN-NEXT: v_readlane_b32 s61, v40, 29 -; GCN-NEXT: v_readlane_b32 s60, v40, 28 -; GCN-NEXT: v_readlane_b32 s59, v40, 27 -; GCN-NEXT: v_readlane_b32 s58, v40, 26 -; GCN-NEXT: v_readlane_b32 s57, v40, 25 -; GCN-NEXT: v_readlane_b32 s56, v40, 24 -; GCN-NEXT: v_readlane_b32 s55, v40, 23 -; GCN-NEXT: v_readlane_b32 s54, v40, 22 -; GCN-NEXT: v_readlane_b32 s53, v40, 21 -; GCN-NEXT: v_readlane_b32 s52, v40, 20 -; GCN-NEXT: v_readlane_b32 s51, v40, 19 -; GCN-NEXT: v_readlane_b32 s50, v40, 18 -; GCN-NEXT: v_readlane_b32 s49, v40, 17 -; GCN-NEXT: v_readlane_b32 s48, v40, 16 -; GCN-NEXT: v_readlane_b32 s47, v40, 15 -; GCN-NEXT: v_readlane_b32 s46, v40, 14 -; GCN-NEXT: v_readlane_b32 s45, v40, 13 -; GCN-NEXT: v_readlane_b32 s44, v40, 12 -; GCN-NEXT: v_readlane_b32 s43, v40, 11 -; GCN-NEXT: v_readlane_b32 s42, v40, 10 -; GCN-NEXT: v_readlane_b32 s41, v40, 9 -; GCN-NEXT: v_readlane_b32 s40, v40, 8 -; GCN-NEXT: v_readlane_b32 s39, v40, 7 -; GCN-NEXT: v_readlane_b32 s38, v40, 6 -; GCN-NEXT: v_readlane_b32 s37, v40, 5 -; GCN-NEXT: v_readlane_b32 s36, v40, 4 -; GCN-NEXT: v_readlane_b32 s35, v40, 3 -; GCN-NEXT: v_readlane_b32 s34, v40, 2 -; GCN-NEXT: v_readlane_b32 s31, v40, 1 -; GCN-NEXT: v_readlane_b32 s30, v40, 0 +; GCN-NEXT: v_readlane_b32 s30, v40, 30 +; GCN-NEXT: v_readlane_b32 s31, v40, 31 +; GCN-NEXT: v_readlane_b32 s63, v40, 29 +; GCN-NEXT: v_readlane_b32 s62, v40, 28 +; GCN-NEXT: v_readlane_b32 s61, v40, 27 +; GCN-NEXT: v_readlane_b32 s60, v40, 26 +; GCN-NEXT: v_readlane_b32 s59, v40, 25 +; GCN-NEXT: v_readlane_b32 s58, v40, 24 +; GCN-NEXT: v_readlane_b32 s57, v40, 23 +; GCN-NEXT: v_readlane_b32 s56, v40, 22 +; GCN-NEXT: v_readlane_b32 s55, v40, 21 +; GCN-NEXT: v_readlane_b32 s54, v40, 20 +; GCN-NEXT: v_readlane_b32 s53, v40, 19 +; GCN-NEXT: v_readlane_b32 s52, v40, 18 +; GCN-NEXT: v_readlane_b32 s51, v40, 17 +; GCN-NEXT: v_readlane_b32 s50, v40, 16 +; GCN-NEXT: v_readlane_b32 s49, v40, 15 +; GCN-NEXT: v_readlane_b32 s48, v40, 14 +; GCN-NEXT: v_readlane_b32 s47, v40, 13 +; GCN-NEXT: v_readlane_b32 s46, v40, 12 +; GCN-NEXT: v_readlane_b32 s45, v40, 11 +; GCN-NEXT: v_readlane_b32 s44, v40, 10 +; GCN-NEXT: v_readlane_b32 s43, v40, 9 +; GCN-NEXT: v_readlane_b32 s42, v40, 8 +; GCN-NEXT: v_readlane_b32 s41, v40, 7 +; GCN-NEXT: v_readlane_b32 s40, v40, 6 +; GCN-NEXT: v_readlane_b32 s39, v40, 5 +; GCN-NEXT: v_readlane_b32 s38, v40, 4 +; GCN-NEXT: v_readlane_b32 s37, v40, 3 +; GCN-NEXT: v_readlane_b32 s36, v40, 2 +; GCN-NEXT: v_readlane_b32 s35, v40, 1 +; GCN-NEXT: v_readlane_b32 s34, v40, 0 ; GCN-NEXT: buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload ; GCN-NEXT: s_addk_i32 s32, 0xfc00 ; GCN-NEXT: v_readlane_b32 s33, v40, 32 @@ -1400,38 +1400,38 @@ ; GISEL-NEXT: s_mov_b32 s33, s32 ; GISEL-NEXT: s_addk_i32 s32, 0x400 ; GISEL-NEXT: buffer_store_dword v41, off, s[0:3], s33 ; 4-byte Folded Spill -; GISEL-NEXT: v_writelane_b32 v40, s30, 0 -; GISEL-NEXT: v_writelane_b32 v40, s31, 1 -; GISEL-NEXT: v_writelane_b32 v40, s34, 2 -; GISEL-NEXT: v_writelane_b32 v40, s35, 3 -; GISEL-NEXT: v_writelane_b32 v40, s36, 4 -; GISEL-NEXT: v_writelane_b32 v40, s37, 5 -; GISEL-NEXT: v_writelane_b32 v40, s38, 6 -; GISEL-NEXT: v_writelane_b32 v40, s39, 7 -; GISEL-NEXT: v_writelane_b32 v40, s40, 8 -; GISEL-NEXT: v_writelane_b32 v40, s41, 9 -; GISEL-NEXT: v_writelane_b32 v40, s42, 10 -; GISEL-NEXT: v_writelane_b32 v40, s43, 11 -; GISEL-NEXT: v_writelane_b32 v40, s44, 12 -; GISEL-NEXT: v_writelane_b32 v40, s45, 13 -; GISEL-NEXT: v_writelane_b32 v40, s46, 14 -; GISEL-NEXT: v_writelane_b32 v40, s47, 15 -; GISEL-NEXT: v_writelane_b32 v40, s48, 16 -; GISEL-NEXT: v_writelane_b32 v40, s49, 17 -; GISEL-NEXT: v_writelane_b32 v40, s50, 18 -; GISEL-NEXT: v_writelane_b32 v40, s51, 19 -; GISEL-NEXT: v_writelane_b32 v40, s52, 20 -; GISEL-NEXT: v_writelane_b32 v40, s53, 21 -; GISEL-NEXT: v_writelane_b32 v40, s54, 22 -; GISEL-NEXT: v_writelane_b32 v40, s55, 23 -; GISEL-NEXT: v_writelane_b32 v40, s56, 24 -; GISEL-NEXT: v_writelane_b32 v40, s57, 25 -; GISEL-NEXT: v_writelane_b32 v40, s58, 26 -; GISEL-NEXT: v_writelane_b32 v40, s59, 27 -; GISEL-NEXT: v_writelane_b32 v40, s60, 28 -; GISEL-NEXT: v_writelane_b32 v40, s61, 29 -; GISEL-NEXT: v_writelane_b32 v40, s62, 30 -; GISEL-NEXT: v_writelane_b32 v40, s63, 31 +; GISEL-NEXT: v_writelane_b32 v40, s34, 0 +; GISEL-NEXT: v_writelane_b32 v40, s35, 1 +; GISEL-NEXT: v_writelane_b32 v40, s36, 2 +; GISEL-NEXT: v_writelane_b32 v40, s37, 3 +; GISEL-NEXT: v_writelane_b32 v40, s38, 4 +; GISEL-NEXT: v_writelane_b32 v40, s39, 5 +; GISEL-NEXT: v_writelane_b32 v40, s40, 6 +; GISEL-NEXT: v_writelane_b32 v40, s41, 7 +; GISEL-NEXT: v_writelane_b32 v40, s42, 8 +; GISEL-NEXT: v_writelane_b32 v40, s43, 9 +; GISEL-NEXT: v_writelane_b32 v40, s44, 10 +; GISEL-NEXT: v_writelane_b32 v40, s45, 11 +; GISEL-NEXT: v_writelane_b32 v40, s46, 12 +; GISEL-NEXT: v_writelane_b32 v40, s47, 13 +; GISEL-NEXT: v_writelane_b32 v40, s48, 14 +; GISEL-NEXT: v_writelane_b32 v40, s49, 15 +; GISEL-NEXT: v_writelane_b32 v40, s50, 16 +; GISEL-NEXT: v_writelane_b32 v40, s51, 17 +; GISEL-NEXT: v_writelane_b32 v40, s52, 18 +; GISEL-NEXT: v_writelane_b32 v40, s53, 19 +; GISEL-NEXT: v_writelane_b32 v40, s54, 20 +; GISEL-NEXT: v_writelane_b32 v40, s55, 21 +; GISEL-NEXT: v_writelane_b32 v40, s56, 22 +; GISEL-NEXT: v_writelane_b32 v40, s57, 23 +; GISEL-NEXT: v_writelane_b32 v40, s58, 24 +; GISEL-NEXT: v_writelane_b32 v40, s59, 25 +; GISEL-NEXT: v_writelane_b32 v40, s60, 26 +; GISEL-NEXT: v_writelane_b32 v40, s61, 27 +; GISEL-NEXT: v_writelane_b32 v40, s62, 28 +; GISEL-NEXT: v_writelane_b32 v40, s63, 29 +; GISEL-NEXT: v_writelane_b32 v40, s30, 30 +; GISEL-NEXT: v_writelane_b32 v40, s31, 31 ; GISEL-NEXT: v_mov_b32_e32 v41, v0 ; GISEL-NEXT: s_mov_b64 s[4:5], exec ; GISEL-NEXT: .LBB7_1: ; =>This Inner Loop Header: Depth=1 @@ -1447,38 +1447,38 @@ ; GISEL-NEXT: ; %bb.2: ; GISEL-NEXT: s_mov_b64 exec, s[4:5] ; GISEL-NEXT: v_mov_b32_e32 v0, v41 -; GISEL-NEXT: v_readlane_b32 s63, v40, 31 -; GISEL-NEXT: v_readlane_b32 s62, v40, 30 -; GISEL-NEXT: v_readlane_b32 s61, v40, 29 -; GISEL-NEXT: v_readlane_b32 s60, v40, 28 -; GISEL-NEXT: v_readlane_b32 s59, v40, 27 -; GISEL-NEXT: v_readlane_b32 s58, v40, 26 -; GISEL-NEXT: v_readlane_b32 s57, v40, 25 -; GISEL-NEXT: v_readlane_b32 s56, v40, 24 -; GISEL-NEXT: v_readlane_b32 s55, v40, 23 -; GISEL-NEXT: v_readlane_b32 s54, v40, 22 -; GISEL-NEXT: v_readlane_b32 s53, v40, 21 -; GISEL-NEXT: v_readlane_b32 s52, v40, 20 -; GISEL-NEXT: v_readlane_b32 s51, v40, 19 -; GISEL-NEXT: v_readlane_b32 s50, v40, 18 -; GISEL-NEXT: v_readlane_b32 s49, v40, 17 -; GISEL-NEXT: v_readlane_b32 s48, v40, 16 -; GISEL-NEXT: v_readlane_b32 s47, v40, 15 -; GISEL-NEXT: v_readlane_b32 s46, v40, 14 -; GISEL-NEXT: v_readlane_b32 s45, v40, 13 -; GISEL-NEXT: v_readlane_b32 s44, v40, 12 -; GISEL-NEXT: v_readlane_b32 s43, v40, 11 -; GISEL-NEXT: v_readlane_b32 s42, v40, 10 -; GISEL-NEXT: v_readlane_b32 s41, v40, 9 -; GISEL-NEXT: v_readlane_b32 s40, v40, 8 -; GISEL-NEXT: v_readlane_b32 s39, v40, 7 -; GISEL-NEXT: v_readlane_b32 s38, v40, 6 -; GISEL-NEXT: v_readlane_b32 s37, v40, 5 -; GISEL-NEXT: v_readlane_b32 s36, v40, 4 -; GISEL-NEXT: v_readlane_b32 s35, v40, 3 -; GISEL-NEXT: v_readlane_b32 s34, v40, 2 -; GISEL-NEXT: v_readlane_b32 s31, v40, 1 -; GISEL-NEXT: v_readlane_b32 s30, v40, 0 +; GISEL-NEXT: v_readlane_b32 s30, v40, 30 +; GISEL-NEXT: v_readlane_b32 s31, v40, 31 +; GISEL-NEXT: v_readlane_b32 s63, v40, 29 +; GISEL-NEXT: v_readlane_b32 s62, v40, 28 +; GISEL-NEXT: v_readlane_b32 s61, v40, 27 +; GISEL-NEXT: v_readlane_b32 s60, v40, 26 +; GISEL-NEXT: v_readlane_b32 s59, v40, 25 +; GISEL-NEXT: v_readlane_b32 s58, v40, 24 +; GISEL-NEXT: v_readlane_b32 s57, v40, 23 +; GISEL-NEXT: v_readlane_b32 s56, v40, 22 +; GISEL-NEXT: v_readlane_b32 s55, v40, 21 +; GISEL-NEXT: v_readlane_b32 s54, v40, 20 +; GISEL-NEXT: v_readlane_b32 s53, v40, 19 +; GISEL-NEXT: v_readlane_b32 s52, v40, 18 +; GISEL-NEXT: v_readlane_b32 s51, v40, 17 +; GISEL-NEXT: v_readlane_b32 s50, v40, 16 +; GISEL-NEXT: v_readlane_b32 s49, v40, 15 +; GISEL-NEXT: v_readlane_b32 s48, v40, 14 +; GISEL-NEXT: v_readlane_b32 s47, v40, 13 +; GISEL-NEXT: v_readlane_b32 s46, v40, 12 +; GISEL-NEXT: v_readlane_b32 s45, v40, 11 +; GISEL-NEXT: v_readlane_b32 s44, v40, 10 +; GISEL-NEXT: v_readlane_b32 s43, v40, 9 +; GISEL-NEXT: v_readlane_b32 s42, v40, 8 +; GISEL-NEXT: v_readlane_b32 s41, v40, 7 +; GISEL-NEXT: v_readlane_b32 s40, v40, 6 +; GISEL-NEXT: v_readlane_b32 s39, v40, 5 +; GISEL-NEXT: v_readlane_b32 s38, v40, 4 +; GISEL-NEXT: v_readlane_b32 s37, v40, 3 +; GISEL-NEXT: v_readlane_b32 s36, v40, 2 +; GISEL-NEXT: v_readlane_b32 s35, v40, 1 +; GISEL-NEXT: v_readlane_b32 s34, v40, 0 ; GISEL-NEXT: buffer_load_dword v41, off, s[0:3], s33 ; 4-byte Folded Reload ; GISEL-NEXT: s_addk_i32 s32, 0xfc00 ; GISEL-NEXT: v_readlane_b32 s33, v40, 32 @@ -1505,38 +1505,38 @@ ; GCN-NEXT: v_writelane_b32 v40, s33, 32 ; GCN-NEXT: s_mov_b32 s33, s32 ; GCN-NEXT: s_addk_i32 s32, 0x400 -; GCN-NEXT: v_writelane_b32 v40, s30, 0 -; GCN-NEXT: v_writelane_b32 v40, s31, 1 -; GCN-NEXT: v_writelane_b32 v40, s34, 2 -; GCN-NEXT: v_writelane_b32 v40, s35, 3 -; GCN-NEXT: v_writelane_b32 v40, s36, 4 -; GCN-NEXT: v_writelane_b32 v40, s37, 5 -; GCN-NEXT: v_writelane_b32 v40, s38, 6 -; GCN-NEXT: v_writelane_b32 v40, s39, 7 -; GCN-NEXT: v_writelane_b32 v40, s40, 8 -; GCN-NEXT: v_writelane_b32 v40, s41, 9 -; GCN-NEXT: v_writelane_b32 v40, s42, 10 -; GCN-NEXT: v_writelane_b32 v40, s43, 11 -; GCN-NEXT: v_writelane_b32 v40, s44, 12 -; GCN-NEXT: v_writelane_b32 v40, s45, 13 -; GCN-NEXT: v_writelane_b32 v40, s46, 14 -; GCN-NEXT: v_writelane_b32 v40, s47, 15 -; GCN-NEXT: v_writelane_b32 v40, s48, 16 -; GCN-NEXT: v_writelane_b32 v40, s49, 17 -; GCN-NEXT: v_writelane_b32 v40, s50, 18 -; GCN-NEXT: v_writelane_b32 v40, s51, 19 -; GCN-NEXT: v_writelane_b32 v40, s52, 20 -; GCN-NEXT: v_writelane_b32 v40, s53, 21 -; GCN-NEXT: v_writelane_b32 v40, s54, 22 -; GCN-NEXT: v_writelane_b32 v40, s55, 23 -; GCN-NEXT: v_writelane_b32 v40, s56, 24 -; GCN-NEXT: v_writelane_b32 v40, s57, 25 -; GCN-NEXT: v_writelane_b32 v40, s58, 26 -; GCN-NEXT: v_writelane_b32 v40, s59, 27 -; GCN-NEXT: v_writelane_b32 v40, s60, 28 -; GCN-NEXT: v_writelane_b32 v40, s61, 29 -; GCN-NEXT: v_writelane_b32 v40, s62, 30 -; GCN-NEXT: v_writelane_b32 v40, s63, 31 +; GCN-NEXT: v_writelane_b32 v40, s34, 0 +; GCN-NEXT: v_writelane_b32 v40, s35, 1 +; GCN-NEXT: v_writelane_b32 v40, s36, 2 +; GCN-NEXT: v_writelane_b32 v40, s37, 3 +; GCN-NEXT: v_writelane_b32 v40, s38, 4 +; GCN-NEXT: v_writelane_b32 v40, s39, 5 +; GCN-NEXT: v_writelane_b32 v40, s40, 6 +; GCN-NEXT: v_writelane_b32 v40, s41, 7 +; GCN-NEXT: v_writelane_b32 v40, s42, 8 +; GCN-NEXT: v_writelane_b32 v40, s43, 9 +; GCN-NEXT: v_writelane_b32 v40, s44, 10 +; GCN-NEXT: v_writelane_b32 v40, s45, 11 +; GCN-NEXT: v_writelane_b32 v40, s46, 12 +; GCN-NEXT: v_writelane_b32 v40, s47, 13 +; GCN-NEXT: v_writelane_b32 v40, s48, 14 +; GCN-NEXT: v_writelane_b32 v40, s49, 15 +; GCN-NEXT: v_writelane_b32 v40, s50, 16 +; GCN-NEXT: v_writelane_b32 v40, s51, 17 +; GCN-NEXT: v_writelane_b32 v40, s52, 18 +; GCN-NEXT: v_writelane_b32 v40, s53, 19 +; GCN-NEXT: v_writelane_b32 v40, s54, 20 +; GCN-NEXT: v_writelane_b32 v40, s55, 21 +; GCN-NEXT: v_writelane_b32 v40, s56, 22 +; GCN-NEXT: v_writelane_b32 v40, s57, 23 +; GCN-NEXT: v_writelane_b32 v40, s58, 24 +; GCN-NEXT: v_writelane_b32 v40, s59, 25 +; GCN-NEXT: v_writelane_b32 v40, s60, 26 +; GCN-NEXT: v_writelane_b32 v40, s61, 27 +; GCN-NEXT: v_writelane_b32 v40, s62, 28 +; GCN-NEXT: v_writelane_b32 v40, s63, 29 +; GCN-NEXT: v_writelane_b32 v40, s30, 30 +; GCN-NEXT: v_writelane_b32 v40, s31, 31 ; GCN-NEXT: s_mov_b64 s[4:5], exec ; GCN-NEXT: .LBB8_1: ; =>This Inner Loop Header: Depth=1 ; GCN-NEXT: v_readfirstlane_b32 s6, v1 @@ -1552,38 +1552,38 @@ ; GCN-NEXT: ; %bb.2: ; GCN-NEXT: s_mov_b64 exec, s[4:5] ; GCN-NEXT: v_mov_b32_e32 v0, v3 -; GCN-NEXT: v_readlane_b32 s63, v40, 31 -; GCN-NEXT: v_readlane_b32 s62, v40, 30 -; GCN-NEXT: v_readlane_b32 s61, v40, 29 -; GCN-NEXT: v_readlane_b32 s60, v40, 28 -; GCN-NEXT: v_readlane_b32 s59, v40, 27 -; GCN-NEXT: v_readlane_b32 s58, v40, 26 -; GCN-NEXT: v_readlane_b32 s57, v40, 25 -; GCN-NEXT: v_readlane_b32 s56, v40, 24 -; GCN-NEXT: v_readlane_b32 s55, v40, 23 -; GCN-NEXT: v_readlane_b32 s54, v40, 22 -; GCN-NEXT: v_readlane_b32 s53, v40, 21 -; GCN-NEXT: v_readlane_b32 s52, v40, 20 -; GCN-NEXT: v_readlane_b32 s51, v40, 19 -; GCN-NEXT: v_readlane_b32 s50, v40, 18 -; GCN-NEXT: v_readlane_b32 s49, v40, 17 -; GCN-NEXT: v_readlane_b32 s48, v40, 16 -; GCN-NEXT: v_readlane_b32 s47, v40, 15 -; GCN-NEXT: v_readlane_b32 s46, v40, 14 -; GCN-NEXT: v_readlane_b32 s45, v40, 13 -; GCN-NEXT: v_readlane_b32 s44, v40, 12 -; GCN-NEXT: v_readlane_b32 s43, v40, 11 -; GCN-NEXT: v_readlane_b32 s42, v40, 10 -; GCN-NEXT: v_readlane_b32 s41, v40, 9 -; GCN-NEXT: v_readlane_b32 s40, v40, 8 -; GCN-NEXT: v_readlane_b32 s39, v40, 7 -; GCN-NEXT: v_readlane_b32 s38, v40, 6 -; GCN-NEXT: v_readlane_b32 s37, v40, 5 -; GCN-NEXT: v_readlane_b32 s36, v40, 4 -; GCN-NEXT: v_readlane_b32 s35, v40, 3 -; GCN-NEXT: v_readlane_b32 s34, v40, 2 -; GCN-NEXT: v_readlane_b32 s31, v40, 1 -; GCN-NEXT: v_readlane_b32 s30, v40, 0 +; GCN-NEXT: v_readlane_b32 s30, v40, 30 +; GCN-NEXT: v_readlane_b32 s31, v40, 31 +; GCN-NEXT: v_readlane_b32 s63, v40, 29 +; GCN-NEXT: v_readlane_b32 s62, v40, 28 +; GCN-NEXT: v_readlane_b32 s61, v40, 27 +; GCN-NEXT: v_readlane_b32 s60, v40, 26 +; GCN-NEXT: v_readlane_b32 s59, v40, 25 +; GCN-NEXT: v_readlane_b32 s58, v40, 24 +; GCN-NEXT: v_readlane_b32 s57, v40, 23 +; GCN-NEXT: v_readlane_b32 s56, v40, 22 +; GCN-NEXT: v_readlane_b32 s55, v40, 21 +; GCN-NEXT: v_readlane_b32 s54, v40, 20 +; GCN-NEXT: v_readlane_b32 s53, v40, 19 +; GCN-NEXT: v_readlane_b32 s52, v40, 18 +; GCN-NEXT: v_readlane_b32 s51, v40, 17 +; GCN-NEXT: v_readlane_b32 s50, v40, 16 +; GCN-NEXT: v_readlane_b32 s49, v40, 15 +; GCN-NEXT: v_readlane_b32 s48, v40, 14 +; GCN-NEXT: v_readlane_b32 s47, v40, 13 +; GCN-NEXT: v_readlane_b32 s46, v40, 12 +; GCN-NEXT: v_readlane_b32 s45, v40, 11 +; GCN-NEXT: v_readlane_b32 s44, v40, 10 +; GCN-NEXT: v_readlane_b32 s43, v40, 9 +; GCN-NEXT: v_readlane_b32 s42, v40, 8 +; GCN-NEXT: v_readlane_b32 s41, v40, 7 +; GCN-NEXT: v_readlane_b32 s40, v40, 6 +; GCN-NEXT: v_readlane_b32 s39, v40, 5 +; GCN-NEXT: v_readlane_b32 s38, v40, 4 +; GCN-NEXT: v_readlane_b32 s37, v40, 3 +; GCN-NEXT: v_readlane_b32 s36, v40, 2 +; GCN-NEXT: v_readlane_b32 s35, v40, 1 +; GCN-NEXT: v_readlane_b32 s34, v40, 0 ; GCN-NEXT: s_addk_i32 s32, 0xfc00 ; GCN-NEXT: v_readlane_b32 s33, v40, 32 ; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -1601,38 +1601,38 @@ ; GISEL-NEXT: v_writelane_b32 v40, s33, 32 ; GISEL-NEXT: s_mov_b32 s33, s32 ; GISEL-NEXT: s_addk_i32 s32, 0x400 -; GISEL-NEXT: v_writelane_b32 v40, s30, 0 -; GISEL-NEXT: v_writelane_b32 v40, s31, 1 -; GISEL-NEXT: v_writelane_b32 v40, s34, 2 -; GISEL-NEXT: v_writelane_b32 v40, s35, 3 -; GISEL-NEXT: v_writelane_b32 v40, s36, 4 -; GISEL-NEXT: v_writelane_b32 v40, s37, 5 -; GISEL-NEXT: v_writelane_b32 v40, s38, 6 -; GISEL-NEXT: v_writelane_b32 v40, s39, 7 -; GISEL-NEXT: v_writelane_b32 v40, s40, 8 -; GISEL-NEXT: v_writelane_b32 v40, s41, 9 -; GISEL-NEXT: v_writelane_b32 v40, s42, 10 -; GISEL-NEXT: v_writelane_b32 v40, s43, 11 -; GISEL-NEXT: v_writelane_b32 v40, s44, 12 -; GISEL-NEXT: v_writelane_b32 v40, s45, 13 -; GISEL-NEXT: v_writelane_b32 v40, s46, 14 -; GISEL-NEXT: v_writelane_b32 v40, s47, 15 -; GISEL-NEXT: v_writelane_b32 v40, s48, 16 -; GISEL-NEXT: v_writelane_b32 v40, s49, 17 -; GISEL-NEXT: v_writelane_b32 v40, s50, 18 -; GISEL-NEXT: v_writelane_b32 v40, s51, 19 -; GISEL-NEXT: v_writelane_b32 v40, s52, 20 -; GISEL-NEXT: v_writelane_b32 v40, s53, 21 -; GISEL-NEXT: v_writelane_b32 v40, s54, 22 -; GISEL-NEXT: v_writelane_b32 v40, s55, 23 -; GISEL-NEXT: v_writelane_b32 v40, s56, 24 -; GISEL-NEXT: v_writelane_b32 v40, s57, 25 -; GISEL-NEXT: v_writelane_b32 v40, s58, 26 -; GISEL-NEXT: v_writelane_b32 v40, s59, 27 -; GISEL-NEXT: v_writelane_b32 v40, s60, 28 -; GISEL-NEXT: v_writelane_b32 v40, s61, 29 -; GISEL-NEXT: v_writelane_b32 v40, s62, 30 -; GISEL-NEXT: v_writelane_b32 v40, s63, 31 +; GISEL-NEXT: v_writelane_b32 v40, s34, 0 +; GISEL-NEXT: v_writelane_b32 v40, s35, 1 +; GISEL-NEXT: v_writelane_b32 v40, s36, 2 +; GISEL-NEXT: v_writelane_b32 v40, s37, 3 +; GISEL-NEXT: v_writelane_b32 v40, s38, 4 +; GISEL-NEXT: v_writelane_b32 v40, s39, 5 +; GISEL-NEXT: v_writelane_b32 v40, s40, 6 +; GISEL-NEXT: v_writelane_b32 v40, s41, 7 +; GISEL-NEXT: v_writelane_b32 v40, s42, 8 +; GISEL-NEXT: v_writelane_b32 v40, s43, 9 +; GISEL-NEXT: v_writelane_b32 v40, s44, 10 +; GISEL-NEXT: v_writelane_b32 v40, s45, 11 +; GISEL-NEXT: v_writelane_b32 v40, s46, 12 +; GISEL-NEXT: v_writelane_b32 v40, s47, 13 +; GISEL-NEXT: v_writelane_b32 v40, s48, 14 +; GISEL-NEXT: v_writelane_b32 v40, s49, 15 +; GISEL-NEXT: v_writelane_b32 v40, s50, 16 +; GISEL-NEXT: v_writelane_b32 v40, s51, 17 +; GISEL-NEXT: v_writelane_b32 v40, s52, 18 +; GISEL-NEXT: v_writelane_b32 v40, s53, 19 +; GISEL-NEXT: v_writelane_b32 v40, s54, 20 +; GISEL-NEXT: v_writelane_b32 v40, s55, 21 +; GISEL-NEXT: v_writelane_b32 v40, s56, 22 +; GISEL-NEXT: v_writelane_b32 v40, s57, 23 +; GISEL-NEXT: v_writelane_b32 v40, s58, 24 +; GISEL-NEXT: v_writelane_b32 v40, s59, 25 +; GISEL-NEXT: v_writelane_b32 v40, s60, 26 +; GISEL-NEXT: v_writelane_b32 v40, s61, 27 +; GISEL-NEXT: v_writelane_b32 v40, s62, 28 +; GISEL-NEXT: v_writelane_b32 v40, s63, 29 +; GISEL-NEXT: v_writelane_b32 v40, s30, 30 +; GISEL-NEXT: v_writelane_b32 v40, s31, 31 ; GISEL-NEXT: s_mov_b64 s[4:5], exec ; GISEL-NEXT: .LBB8_1: ; =>This Inner Loop Header: Depth=1 ; GISEL-NEXT: v_readfirstlane_b32 s8, v1 @@ -1648,38 +1648,38 @@ ; GISEL-NEXT: ; %bb.2: ; GISEL-NEXT: s_mov_b64 exec, s[4:5] ; GISEL-NEXT: v_mov_b32_e32 v0, v2 -; GISEL-NEXT: v_readlane_b32 s63, v40, 31 -; GISEL-NEXT: v_readlane_b32 s62, v40, 30 -; GISEL-NEXT: v_readlane_b32 s61, v40, 29 -; GISEL-NEXT: v_readlane_b32 s60, v40, 28 -; GISEL-NEXT: v_readlane_b32 s59, v40, 27 -; GISEL-NEXT: v_readlane_b32 s58, v40, 26 -; GISEL-NEXT: v_readlane_b32 s57, v40, 25 -; GISEL-NEXT: v_readlane_b32 s56, v40, 24 -; GISEL-NEXT: v_readlane_b32 s55, v40, 23 -; GISEL-NEXT: v_readlane_b32 s54, v40, 22 -; GISEL-NEXT: v_readlane_b32 s53, v40, 21 -; GISEL-NEXT: v_readlane_b32 s52, v40, 20 -; GISEL-NEXT: v_readlane_b32 s51, v40, 19 -; GISEL-NEXT: v_readlane_b32 s50, v40, 18 -; GISEL-NEXT: v_readlane_b32 s49, v40, 17 -; GISEL-NEXT: v_readlane_b32 s48, v40, 16 -; GISEL-NEXT: v_readlane_b32 s47, v40, 15 -; GISEL-NEXT: v_readlane_b32 s46, v40, 14 -; GISEL-NEXT: v_readlane_b32 s45, v40, 13 -; GISEL-NEXT: v_readlane_b32 s44, v40, 12 -; GISEL-NEXT: v_readlane_b32 s43, v40, 11 -; GISEL-NEXT: v_readlane_b32 s42, v40, 10 -; GISEL-NEXT: v_readlane_b32 s41, v40, 9 -; GISEL-NEXT: v_readlane_b32 s40, v40, 8 -; GISEL-NEXT: v_readlane_b32 s39, v40, 7 -; GISEL-NEXT: v_readlane_b32 s38, v40, 6 -; GISEL-NEXT: v_readlane_b32 s37, v40, 5 -; GISEL-NEXT: v_readlane_b32 s36, v40, 4 -; GISEL-NEXT: v_readlane_b32 s35, v40, 3 -; GISEL-NEXT: v_readlane_b32 s34, v40, 2 -; GISEL-NEXT: v_readlane_b32 s31, v40, 1 -; GISEL-NEXT: v_readlane_b32 s30, v40, 0 +; GISEL-NEXT: v_readlane_b32 s30, v40, 30 +; GISEL-NEXT: v_readlane_b32 s31, v40, 31 +; GISEL-NEXT: v_readlane_b32 s63, v40, 29 +; GISEL-NEXT: v_readlane_b32 s62, v40, 28 +; GISEL-NEXT: v_readlane_b32 s61, v40, 27 +; GISEL-NEXT: v_readlane_b32 s60, v40, 26 +; GISEL-NEXT: v_readlane_b32 s59, v40, 25 +; GISEL-NEXT: v_readlane_b32 s58, v40, 24 +; GISEL-NEXT: v_readlane_b32 s57, v40, 23 +; GISEL-NEXT: v_readlane_b32 s56, v40, 22 +; GISEL-NEXT: v_readlane_b32 s55, v40, 21 +; GISEL-NEXT: v_readlane_b32 s54, v40, 20 +; GISEL-NEXT: v_readlane_b32 s53, v40, 19 +; GISEL-NEXT: v_readlane_b32 s52, v40, 18 +; GISEL-NEXT: v_readlane_b32 s51, v40, 17 +; GISEL-NEXT: v_readlane_b32 s50, v40, 16 +; GISEL-NEXT: v_readlane_b32 s49, v40, 15 +; GISEL-NEXT: v_readlane_b32 s48, v40, 14 +; GISEL-NEXT: v_readlane_b32 s47, v40, 13 +; GISEL-NEXT: v_readlane_b32 s46, v40, 12 +; GISEL-NEXT: v_readlane_b32 s45, v40, 11 +; GISEL-NEXT: v_readlane_b32 s44, v40, 10 +; GISEL-NEXT: v_readlane_b32 s43, v40, 9 +; GISEL-NEXT: v_readlane_b32 s42, v40, 8 +; GISEL-NEXT: v_readlane_b32 s41, v40, 7 +; GISEL-NEXT: v_readlane_b32 s40, v40, 6 +; GISEL-NEXT: v_readlane_b32 s39, v40, 5 +; GISEL-NEXT: v_readlane_b32 s38, v40, 4 +; GISEL-NEXT: v_readlane_b32 s37, v40, 3 +; GISEL-NEXT: v_readlane_b32 s36, v40, 2 +; GISEL-NEXT: v_readlane_b32 s35, v40, 1 +; GISEL-NEXT: v_readlane_b32 s34, v40, 0 ; GISEL-NEXT: s_addk_i32 s32, 0xfc00 ; GISEL-NEXT: v_readlane_b32 s33, v40, 32 ; GISEL-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -1702,38 +1702,38 @@ ; GCN-NEXT: v_writelane_b32 v40, s33, 32 ; GCN-NEXT: s_mov_b32 s33, s32 ; GCN-NEXT: s_addk_i32 s32, 0x400 -; GCN-NEXT: v_writelane_b32 v40, s30, 0 -; GCN-NEXT: v_writelane_b32 v40, s31, 1 -; GCN-NEXT: v_writelane_b32 v40, s34, 2 -; GCN-NEXT: v_writelane_b32 v40, s35, 3 -; GCN-NEXT: v_writelane_b32 v40, s36, 4 -; GCN-NEXT: v_writelane_b32 v40, s37, 5 -; GCN-NEXT: v_writelane_b32 v40, s38, 6 -; GCN-NEXT: v_writelane_b32 v40, s39, 7 -; GCN-NEXT: v_writelane_b32 v40, s40, 8 -; GCN-NEXT: v_writelane_b32 v40, s41, 9 -; GCN-NEXT: v_writelane_b32 v40, s42, 10 -; GCN-NEXT: v_writelane_b32 v40, s43, 11 -; GCN-NEXT: v_writelane_b32 v40, s44, 12 -; GCN-NEXT: v_writelane_b32 v40, s45, 13 -; GCN-NEXT: v_writelane_b32 v40, s46, 14 -; GCN-NEXT: v_writelane_b32 v40, s47, 15 -; GCN-NEXT: v_writelane_b32 v40, s48, 16 -; GCN-NEXT: v_writelane_b32 v40, s49, 17 -; GCN-NEXT: v_writelane_b32 v40, s50, 18 -; GCN-NEXT: v_writelane_b32 v40, s51, 19 -; GCN-NEXT: v_writelane_b32 v40, s52, 20 -; GCN-NEXT: v_writelane_b32 v40, s53, 21 -; GCN-NEXT: v_writelane_b32 v40, s54, 22 -; GCN-NEXT: v_writelane_b32 v40, s55, 23 -; GCN-NEXT: v_writelane_b32 v40, s56, 24 -; GCN-NEXT: v_writelane_b32 v40, s57, 25 -; GCN-NEXT: v_writelane_b32 v40, s58, 26 -; GCN-NEXT: v_writelane_b32 v40, s59, 27 -; GCN-NEXT: v_writelane_b32 v40, s60, 28 -; GCN-NEXT: v_writelane_b32 v40, s61, 29 -; GCN-NEXT: v_writelane_b32 v40, s62, 30 -; GCN-NEXT: v_writelane_b32 v40, s63, 31 +; GCN-NEXT: v_writelane_b32 v40, s34, 0 +; GCN-NEXT: v_writelane_b32 v40, s35, 1 +; GCN-NEXT: v_writelane_b32 v40, s36, 2 +; GCN-NEXT: v_writelane_b32 v40, s37, 3 +; GCN-NEXT: v_writelane_b32 v40, s38, 4 +; GCN-NEXT: v_writelane_b32 v40, s39, 5 +; GCN-NEXT: v_writelane_b32 v40, s40, 6 +; GCN-NEXT: v_writelane_b32 v40, s41, 7 +; GCN-NEXT: v_writelane_b32 v40, s42, 8 +; GCN-NEXT: v_writelane_b32 v40, s43, 9 +; GCN-NEXT: v_writelane_b32 v40, s44, 10 +; GCN-NEXT: v_writelane_b32 v40, s45, 11 +; GCN-NEXT: v_writelane_b32 v40, s46, 12 +; GCN-NEXT: v_writelane_b32 v40, s47, 13 +; GCN-NEXT: v_writelane_b32 v40, s48, 14 +; GCN-NEXT: v_writelane_b32 v40, s49, 15 +; GCN-NEXT: v_writelane_b32 v40, s50, 16 +; GCN-NEXT: v_writelane_b32 v40, s51, 17 +; GCN-NEXT: v_writelane_b32 v40, s52, 18 +; GCN-NEXT: v_writelane_b32 v40, s53, 19 +; GCN-NEXT: v_writelane_b32 v40, s54, 20 +; GCN-NEXT: v_writelane_b32 v40, s55, 21 +; GCN-NEXT: v_writelane_b32 v40, s56, 22 +; GCN-NEXT: v_writelane_b32 v40, s57, 23 +; GCN-NEXT: v_writelane_b32 v40, s58, 24 +; GCN-NEXT: v_writelane_b32 v40, s59, 25 +; GCN-NEXT: v_writelane_b32 v40, s60, 26 +; GCN-NEXT: v_writelane_b32 v40, s61, 27 +; GCN-NEXT: v_writelane_b32 v40, s62, 28 +; GCN-NEXT: v_writelane_b32 v40, s63, 29 +; GCN-NEXT: v_writelane_b32 v40, s30, 30 +; GCN-NEXT: v_writelane_b32 v40, s31, 31 ; GCN-NEXT: s_mov_b64 s[4:5], exec ; GCN-NEXT: .LBB9_1: ; =>This Inner Loop Header: Depth=1 ; GCN-NEXT: v_readfirstlane_b32 s6, v0 @@ -1746,38 +1746,38 @@ ; GCN-NEXT: s_cbranch_execnz .LBB9_1 ; GCN-NEXT: ; %bb.2: ; GCN-NEXT: s_mov_b64 exec, s[4:5] -; GCN-NEXT: v_readlane_b32 s63, v40, 31 -; GCN-NEXT: v_readlane_b32 s62, v40, 30 -; GCN-NEXT: v_readlane_b32 s61, v40, 29 -; GCN-NEXT: v_readlane_b32 s60, v40, 28 -; GCN-NEXT: v_readlane_b32 s59, v40, 27 -; GCN-NEXT: v_readlane_b32 s58, v40, 26 -; GCN-NEXT: v_readlane_b32 s57, v40, 25 -; GCN-NEXT: v_readlane_b32 s56, v40, 24 -; GCN-NEXT: v_readlane_b32 s55, v40, 23 -; GCN-NEXT: v_readlane_b32 s54, v40, 22 -; GCN-NEXT: v_readlane_b32 s53, v40, 21 -; GCN-NEXT: v_readlane_b32 s52, v40, 20 -; GCN-NEXT: v_readlane_b32 s51, v40, 19 -; GCN-NEXT: v_readlane_b32 s50, v40, 18 -; GCN-NEXT: v_readlane_b32 s49, v40, 17 -; GCN-NEXT: v_readlane_b32 s48, v40, 16 -; GCN-NEXT: v_readlane_b32 s47, v40, 15 -; GCN-NEXT: v_readlane_b32 s46, v40, 14 -; GCN-NEXT: v_readlane_b32 s45, v40, 13 -; GCN-NEXT: v_readlane_b32 s44, v40, 12 -; GCN-NEXT: v_readlane_b32 s43, v40, 11 -; GCN-NEXT: v_readlane_b32 s42, v40, 10 -; GCN-NEXT: v_readlane_b32 s41, v40, 9 -; GCN-NEXT: v_readlane_b32 s40, v40, 8 -; GCN-NEXT: v_readlane_b32 s39, v40, 7 -; GCN-NEXT: v_readlane_b32 s38, v40, 6 -; GCN-NEXT: v_readlane_b32 s37, v40, 5 -; GCN-NEXT: v_readlane_b32 s36, v40, 4 -; GCN-NEXT: v_readlane_b32 s35, v40, 3 -; GCN-NEXT: v_readlane_b32 s34, v40, 2 -; GCN-NEXT: v_readlane_b32 s31, v40, 1 -; GCN-NEXT: v_readlane_b32 s30, v40, 0 +; GCN-NEXT: v_readlane_b32 s30, v40, 30 +; GCN-NEXT: v_readlane_b32 s31, v40, 31 +; GCN-NEXT: v_readlane_b32 s63, v40, 29 +; GCN-NEXT: v_readlane_b32 s62, v40, 28 +; GCN-NEXT: v_readlane_b32 s61, v40, 27 +; GCN-NEXT: v_readlane_b32 s60, v40, 26 +; GCN-NEXT: v_readlane_b32 s59, v40, 25 +; GCN-NEXT: v_readlane_b32 s58, v40, 24 +; GCN-NEXT: v_readlane_b32 s57, v40, 23 +; GCN-NEXT: v_readlane_b32 s56, v40, 22 +; GCN-NEXT: v_readlane_b32 s55, v40, 21 +; GCN-NEXT: v_readlane_b32 s54, v40, 20 +; GCN-NEXT: v_readlane_b32 s53, v40, 19 +; GCN-NEXT: v_readlane_b32 s52, v40, 18 +; GCN-NEXT: v_readlane_b32 s51, v40, 17 +; GCN-NEXT: v_readlane_b32 s50, v40, 16 +; GCN-NEXT: v_readlane_b32 s49, v40, 15 +; GCN-NEXT: v_readlane_b32 s48, v40, 14 +; GCN-NEXT: v_readlane_b32 s47, v40, 13 +; GCN-NEXT: v_readlane_b32 s46, v40, 12 +; GCN-NEXT: v_readlane_b32 s45, v40, 11 +; GCN-NEXT: v_readlane_b32 s44, v40, 10 +; GCN-NEXT: v_readlane_b32 s43, v40, 9 +; GCN-NEXT: v_readlane_b32 s42, v40, 8 +; GCN-NEXT: v_readlane_b32 s41, v40, 7 +; GCN-NEXT: v_readlane_b32 s40, v40, 6 +; GCN-NEXT: v_readlane_b32 s39, v40, 5 +; GCN-NEXT: v_readlane_b32 s38, v40, 4 +; GCN-NEXT: v_readlane_b32 s37, v40, 3 +; GCN-NEXT: v_readlane_b32 s36, v40, 2 +; GCN-NEXT: v_readlane_b32 s35, v40, 1 +; GCN-NEXT: v_readlane_b32 s34, v40, 0 ; GCN-NEXT: s_addk_i32 s32, 0xfc00 ; GCN-NEXT: v_readlane_b32 s33, v40, 32 ; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -1795,38 +1795,38 @@ ; GISEL-NEXT: v_writelane_b32 v40, s33, 32 ; GISEL-NEXT: s_mov_b32 s33, s32 ; GISEL-NEXT: s_addk_i32 s32, 0x400 -; GISEL-NEXT: v_writelane_b32 v40, s30, 0 -; GISEL-NEXT: v_writelane_b32 v40, s31, 1 -; GISEL-NEXT: v_writelane_b32 v40, s34, 2 -; GISEL-NEXT: v_writelane_b32 v40, s35, 3 -; GISEL-NEXT: v_writelane_b32 v40, s36, 4 -; GISEL-NEXT: v_writelane_b32 v40, s37, 5 -; GISEL-NEXT: v_writelane_b32 v40, s38, 6 -; GISEL-NEXT: v_writelane_b32 v40, s39, 7 -; GISEL-NEXT: v_writelane_b32 v40, s40, 8 -; GISEL-NEXT: v_writelane_b32 v40, s41, 9 -; GISEL-NEXT: v_writelane_b32 v40, s42, 10 -; GISEL-NEXT: v_writelane_b32 v40, s43, 11 -; GISEL-NEXT: v_writelane_b32 v40, s44, 12 -; GISEL-NEXT: v_writelane_b32 v40, s45, 13 -; GISEL-NEXT: v_writelane_b32 v40, s46, 14 -; GISEL-NEXT: v_writelane_b32 v40, s47, 15 -; GISEL-NEXT: v_writelane_b32 v40, s48, 16 -; GISEL-NEXT: v_writelane_b32 v40, s49, 17 -; GISEL-NEXT: v_writelane_b32 v40, s50, 18 -; GISEL-NEXT: v_writelane_b32 v40, s51, 19 -; GISEL-NEXT: v_writelane_b32 v40, s52, 20 -; GISEL-NEXT: v_writelane_b32 v40, s53, 21 -; GISEL-NEXT: v_writelane_b32 v40, s54, 22 -; GISEL-NEXT: v_writelane_b32 v40, s55, 23 -; GISEL-NEXT: v_writelane_b32 v40, s56, 24 -; GISEL-NEXT: v_writelane_b32 v40, s57, 25 -; GISEL-NEXT: v_writelane_b32 v40, s58, 26 -; GISEL-NEXT: v_writelane_b32 v40, s59, 27 -; GISEL-NEXT: v_writelane_b32 v40, s60, 28 -; GISEL-NEXT: v_writelane_b32 v40, s61, 29 -; GISEL-NEXT: v_writelane_b32 v40, s62, 30 -; GISEL-NEXT: v_writelane_b32 v40, s63, 31 +; GISEL-NEXT: v_writelane_b32 v40, s34, 0 +; GISEL-NEXT: v_writelane_b32 v40, s35, 1 +; GISEL-NEXT: v_writelane_b32 v40, s36, 2 +; GISEL-NEXT: v_writelane_b32 v40, s37, 3 +; GISEL-NEXT: v_writelane_b32 v40, s38, 4 +; GISEL-NEXT: v_writelane_b32 v40, s39, 5 +; GISEL-NEXT: v_writelane_b32 v40, s40, 6 +; GISEL-NEXT: v_writelane_b32 v40, s41, 7 +; GISEL-NEXT: v_writelane_b32 v40, s42, 8 +; GISEL-NEXT: v_writelane_b32 v40, s43, 9 +; GISEL-NEXT: v_writelane_b32 v40, s44, 10 +; GISEL-NEXT: v_writelane_b32 v40, s45, 11 +; GISEL-NEXT: v_writelane_b32 v40, s46, 12 +; GISEL-NEXT: v_writelane_b32 v40, s47, 13 +; GISEL-NEXT: v_writelane_b32 v40, s48, 14 +; GISEL-NEXT: v_writelane_b32 v40, s49, 15 +; GISEL-NEXT: v_writelane_b32 v40, s50, 16 +; GISEL-NEXT: v_writelane_b32 v40, s51, 17 +; GISEL-NEXT: v_writelane_b32 v40, s52, 18 +; GISEL-NEXT: v_writelane_b32 v40, s53, 19 +; GISEL-NEXT: v_writelane_b32 v40, s54, 20 +; GISEL-NEXT: v_writelane_b32 v40, s55, 21 +; GISEL-NEXT: v_writelane_b32 v40, s56, 22 +; GISEL-NEXT: v_writelane_b32 v40, s57, 23 +; GISEL-NEXT: v_writelane_b32 v40, s58, 24 +; GISEL-NEXT: v_writelane_b32 v40, s59, 25 +; GISEL-NEXT: v_writelane_b32 v40, s60, 26 +; GISEL-NEXT: v_writelane_b32 v40, s61, 27 +; GISEL-NEXT: v_writelane_b32 v40, s62, 28 +; GISEL-NEXT: v_writelane_b32 v40, s63, 29 +; GISEL-NEXT: v_writelane_b32 v40, s30, 30 +; GISEL-NEXT: v_writelane_b32 v40, s31, 31 ; GISEL-NEXT: s_mov_b64 s[4:5], exec ; GISEL-NEXT: .LBB9_1: ; =>This Inner Loop Header: Depth=1 ; GISEL-NEXT: v_readfirstlane_b32 s6, v0 @@ -1839,38 +1839,38 @@ ; GISEL-NEXT: s_cbranch_execnz .LBB9_1 ; GISEL-NEXT: ; %bb.2: ; GISEL-NEXT: s_mov_b64 exec, s[4:5] -; GISEL-NEXT: v_readlane_b32 s63, v40, 31 -; GISEL-NEXT: v_readlane_b32 s62, v40, 30 -; GISEL-NEXT: v_readlane_b32 s61, v40, 29 -; GISEL-NEXT: v_readlane_b32 s60, v40, 28 -; GISEL-NEXT: v_readlane_b32 s59, v40, 27 -; GISEL-NEXT: v_readlane_b32 s58, v40, 26 -; GISEL-NEXT: v_readlane_b32 s57, v40, 25 -; GISEL-NEXT: v_readlane_b32 s56, v40, 24 -; GISEL-NEXT: v_readlane_b32 s55, v40, 23 -; GISEL-NEXT: v_readlane_b32 s54, v40, 22 -; GISEL-NEXT: v_readlane_b32 s53, v40, 21 -; GISEL-NEXT: v_readlane_b32 s52, v40, 20 -; GISEL-NEXT: v_readlane_b32 s51, v40, 19 -; GISEL-NEXT: v_readlane_b32 s50, v40, 18 -; GISEL-NEXT: v_readlane_b32 s49, v40, 17 -; GISEL-NEXT: v_readlane_b32 s48, v40, 16 -; GISEL-NEXT: v_readlane_b32 s47, v40, 15 -; GISEL-NEXT: v_readlane_b32 s46, v40, 14 -; GISEL-NEXT: v_readlane_b32 s45, v40, 13 -; GISEL-NEXT: v_readlane_b32 s44, v40, 12 -; GISEL-NEXT: v_readlane_b32 s43, v40, 11 -; GISEL-NEXT: v_readlane_b32 s42, v40, 10 -; GISEL-NEXT: v_readlane_b32 s41, v40, 9 -; GISEL-NEXT: v_readlane_b32 s40, v40, 8 -; GISEL-NEXT: v_readlane_b32 s39, v40, 7 -; GISEL-NEXT: v_readlane_b32 s38, v40, 6 -; GISEL-NEXT: v_readlane_b32 s37, v40, 5 -; GISEL-NEXT: v_readlane_b32 s36, v40, 4 -; GISEL-NEXT: v_readlane_b32 s35, v40, 3 -; GISEL-NEXT: v_readlane_b32 s34, v40, 2 -; GISEL-NEXT: v_readlane_b32 s31, v40, 1 -; GISEL-NEXT: v_readlane_b32 s30, v40, 0 +; GISEL-NEXT: v_readlane_b32 s30, v40, 30 +; GISEL-NEXT: v_readlane_b32 s31, v40, 31 +; GISEL-NEXT: v_readlane_b32 s63, v40, 29 +; GISEL-NEXT: v_readlane_b32 s62, v40, 28 +; GISEL-NEXT: v_readlane_b32 s61, v40, 27 +; GISEL-NEXT: v_readlane_b32 s60, v40, 26 +; GISEL-NEXT: v_readlane_b32 s59, v40, 25 +; GISEL-NEXT: v_readlane_b32 s58, v40, 24 +; GISEL-NEXT: v_readlane_b32 s57, v40, 23 +; GISEL-NEXT: v_readlane_b32 s56, v40, 22 +; GISEL-NEXT: v_readlane_b32 s55, v40, 21 +; GISEL-NEXT: v_readlane_b32 s54, v40, 20 +; GISEL-NEXT: v_readlane_b32 s53, v40, 19 +; GISEL-NEXT: v_readlane_b32 s52, v40, 18 +; GISEL-NEXT: v_readlane_b32 s51, v40, 17 +; GISEL-NEXT: v_readlane_b32 s50, v40, 16 +; GISEL-NEXT: v_readlane_b32 s49, v40, 15 +; GISEL-NEXT: v_readlane_b32 s48, v40, 14 +; GISEL-NEXT: v_readlane_b32 s47, v40, 13 +; GISEL-NEXT: v_readlane_b32 s46, v40, 12 +; GISEL-NEXT: v_readlane_b32 s45, v40, 11 +; GISEL-NEXT: v_readlane_b32 s44, v40, 10 +; GISEL-NEXT: v_readlane_b32 s43, v40, 9 +; GISEL-NEXT: v_readlane_b32 s42, v40, 8 +; GISEL-NEXT: v_readlane_b32 s41, v40, 7 +; GISEL-NEXT: v_readlane_b32 s40, v40, 6 +; GISEL-NEXT: v_readlane_b32 s39, v40, 5 +; GISEL-NEXT: v_readlane_b32 s38, v40, 4 +; GISEL-NEXT: v_readlane_b32 s37, v40, 3 +; GISEL-NEXT: v_readlane_b32 s36, v40, 2 +; GISEL-NEXT: v_readlane_b32 s35, v40, 1 +; GISEL-NEXT: v_readlane_b32 s34, v40, 0 ; GISEL-NEXT: s_addk_i32 s32, 0xfc00 ; GISEL-NEXT: v_readlane_b32 s33, v40, 32 ; GISEL-NEXT: s_or_saveexec_b64 s[4:5], -1 diff --git a/llvm/test/CodeGen/AMDGPU/mul24-pass-ordering.ll b/llvm/test/CodeGen/AMDGPU/mul24-pass-ordering.ll --- a/llvm/test/CodeGen/AMDGPU/mul24-pass-ordering.ll +++ b/llvm/test/CodeGen/AMDGPU/mul24-pass-ordering.ll @@ -193,19 +193,19 @@ ; GFX9-NEXT: s_mov_b64 exec, s[4:5] ; GFX9-NEXT: v_writelane_b32 v40, s33, 5 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: v_writelane_b32 v40, s30, 0 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_addk_i32 s32, 0x800 -; GFX9-NEXT: v_writelane_b32 v40, s34, 2 -; GFX9-NEXT: v_writelane_b32 v40, s36, 3 +; GFX9-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill +; GFX9-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill +; GFX9-NEXT: buffer_store_dword v43, off, s[0:3], s33 ; 4-byte Folded Spill +; GFX9-NEXT: v_writelane_b32 v40, s34, 0 +; GFX9-NEXT: v_writelane_b32 v40, s36, 1 +; GFX9-NEXT: v_writelane_b32 v40, s37, 2 +; GFX9-NEXT: v_writelane_b32 v40, s30, 3 +; GFX9-NEXT: v_writelane_b32 v40, s31, 4 ; GFX9-NEXT: s_getpc_b64 s[4:5] ; GFX9-NEXT: s_add_u32 s4, s4, foo@gotpcrel32@lo+4 ; GFX9-NEXT: s_addc_u32 s5, s5, foo@gotpcrel32@hi+12 -; GFX9-NEXT: v_writelane_b32 v40, s37, 4 ; GFX9-NEXT: s_load_dwordx2 s[36:37], s[4:5], 0x0 -; GFX9-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill -; GFX9-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill -; GFX9-NEXT: buffer_store_dword v43, off, s[0:3], s33 ; 4-byte Folded Spill ; GFX9-NEXT: v_mov_b32_e32 v41, v1 ; GFX9-NEXT: v_mov_b32_e32 v42, v0 ; GFX9-NEXT: v_mul_u32_u24_e32 v0, v42, v41 @@ -223,11 +223,11 @@ ; GFX9-NEXT: buffer_load_dword v43, off, s[0:3], s33 ; 4-byte Folded Reload ; GFX9-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload ; GFX9-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload -; GFX9-NEXT: v_readlane_b32 s37, v40, 4 -; GFX9-NEXT: v_readlane_b32 s36, v40, 3 -; GFX9-NEXT: v_readlane_b32 s34, v40, 2 -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 -; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s30, v40, 3 +; GFX9-NEXT: v_readlane_b32 s31, v40, 4 +; GFX9-NEXT: v_readlane_b32 s37, v40, 2 +; GFX9-NEXT: v_readlane_b32 s36, v40, 1 +; GFX9-NEXT: v_readlane_b32 s34, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xf800 ; GFX9-NEXT: v_readlane_b32 s33, v40, 5 ; GFX9-NEXT: s_or_saveexec_b64 s[4:5], -1 diff --git a/llvm/test/CodeGen/AMDGPU/need-fp-from-vgpr-spills.ll b/llvm/test/CodeGen/AMDGPU/need-fp-from-vgpr-spills.ll --- a/llvm/test/CodeGen/AMDGPU/need-fp-from-vgpr-spills.ll +++ b/llvm/test/CodeGen/AMDGPU/need-fp-from-vgpr-spills.ll @@ -47,8 +47,8 @@ ; CHECK-NEXT: ;;#ASMSTART ; CHECK-NEXT: ; clobber csr v40 ; CHECK-NEXT: ;;#ASMEND -; CHECK-NEXT: v_readlane_b32 s31, v1, 1 ; CHECK-NEXT: v_readlane_b32 s30, v1, 0 +; CHECK-NEXT: v_readlane_b32 s31, v1, 1 ; CHECK-NEXT: buffer_load_dword v40, off, s[0:3], s33 ; 4-byte Folded Reload ; CHECK-NEXT: s_add_i32 s32, s32, 0xfffffc00 ; CHECK-NEXT: v_readlane_b32 s33, v1, 2 @@ -168,8 +168,8 @@ ; CHECK-NEXT: s_mov_b64 s[0:1], s[8:9] ; CHECK-NEXT: s_mov_b64 s[2:3], s[10:11] ; CHECK-NEXT: s_swappc_b64 s[30:31], s[4:5] -; CHECK-NEXT: v_readlane_b32 s31, v1, 1 ; CHECK-NEXT: v_readlane_b32 s30, v1, 0 +; CHECK-NEXT: v_readlane_b32 s31, v1, 1 ; CHECK-NEXT: s_add_i32 s32, s32, 0xfffffc00 ; CHECK-NEXT: v_readlane_b32 s33, v1, 2 ; CHECK-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -202,8 +202,8 @@ ; CHECK-NEXT: s_mov_b64 s[0:1], s[8:9] ; CHECK-NEXT: s_mov_b64 s[2:3], s[10:11] ; CHECK-NEXT: s_swappc_b64 s[30:31], s[4:5] -; CHECK-NEXT: v_readlane_b32 s31, v2, 1 ; CHECK-NEXT: v_readlane_b32 s30, v2, 0 +; CHECK-NEXT: v_readlane_b32 s31, v2, 1 ; CHECK-NEXT: s_add_i32 s32, s32, 0xfffffc00 ; CHECK-NEXT: v_readlane_b32 s33, v2, 2 ; CHECK-NEXT: s_or_saveexec_b64 s[4:5], -1 diff --git a/llvm/test/CodeGen/AMDGPU/nested-calls.ll b/llvm/test/CodeGen/AMDGPU/nested-calls.ll --- a/llvm/test/CodeGen/AMDGPU/nested-calls.ll +++ b/llvm/test/CodeGen/AMDGPU/nested-calls.ll @@ -22,8 +22,8 @@ ; GCN: s_swappc_b64 -; GCN: v_readlane_b32 s31, v40, 1 -; GCN: v_readlane_b32 s30, v40, 0 +; GCN-DAG: v_readlane_b32 s30, v40, 0 +; GCN-DAG: v_readlane_b32 s31, v40, 1 ; GCN-NEXT: s_addk_i32 s32, 0xfc00 ; GCN-NEXT: v_readlane_b32 s33, v40, 2 diff --git a/llvm/test/CodeGen/AMDGPU/no-source-locations-in-prologue.ll b/llvm/test/CodeGen/AMDGPU/no-source-locations-in-prologue.ll --- a/llvm/test/CodeGen/AMDGPU/no-source-locations-in-prologue.ll +++ b/llvm/test/CodeGen/AMDGPU/no-source-locations-in-prologue.ll @@ -200,6 +200,7 @@ ; CHECK-NEXT: s_add_i32 s32, s32, 0x400 ; CHECK-NEXT: v_writelane_b32 v40, s30, 0 ; CHECK-NEXT: v_writelane_b32 v40, s31, 1 +; CHECK-NEXT: .cfi_escape 0x10, 0x10, 0x0c, 0x90, 0xa8, 0x14, 0x9d, 0x20, 0x00, 0x90, 0xa8, 0x14, 0x9d, 0x20, 0x20 ; ; CHECK-NEXT: .Ltmp0: ; CHECK-NEXT: .loc 0 31 3 prologue_end ; lane-info.cpp:31:3 ; CHECK-NEXT: s_getpc_b64 s[16:17] @@ -214,8 +215,8 @@ ; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17] ; CHECK-NEXT: .Ltmp1: ; CHECK-NEXT: .loc 0 32 1 ; lane-info.cpp:32:1 -; CHECK-NEXT: v_readlane_b32 s31, v40, 1 ; CHECK-NEXT: v_readlane_b32 s30, v40, 0 +; CHECK-NEXT: v_readlane_b32 s31, v40, 1 ; CHECK-NEXT: s_add_i32 s32, s32, 0xfffffc00 ; CHECK-NEXT: v_readlane_b32 s33, v40, 2 ; CHECK-NEXT: .cfi_def_cfa_register 64 diff --git a/llvm/test/CodeGen/AMDGPU/sgpr-spills-split-regalloc.ll b/llvm/test/CodeGen/AMDGPU/sgpr-spills-split-regalloc.ll --- a/llvm/test/CodeGen/AMDGPU/sgpr-spills-split-regalloc.ll +++ b/llvm/test/CodeGen/AMDGPU/sgpr-spills-split-regalloc.ll @@ -11,8 +11,8 @@ ; GCN: v_writelane_b32 v255, s30, 0 ; GCN: v_writelane_b32 v255, s31, 1 ; GCN: s_swappc_b64 s[30:31], s[4:5] -; GCN: v_readlane_b32 s31, v255, 1 ; GCN: v_readlane_b32 s30, v255, 0 +; GCN: v_readlane_b32 s31, v255, 1 ; GCN: v_readlane_b32 s33, v255, 2 ; GCN: ; NumVgprs: 256 @@ -57,8 +57,8 @@ ; GCN: v_writelane_b32 v254, s30, 0 ; GCN: v_writelane_b32 v254, s31, 1 ; GCN: s_swappc_b64 s[30:31], s[4:5] -; GCN: v_readlane_b32 s31, v254, 1 ; GCN: v_readlane_b32 s30, v254, 0 +; GCN: v_readlane_b32 s31, v254, 1 ; GCN: v_readlane_b32 s33, v254, 2 define void @spill_to_lowest_available_vgpr() #0 { @@ -282,15 +282,15 @@ } ; GCN-LABEL: {{^}}spill_sgpr_no_free_vgpr_ipra: -; GCN: v_writelane_b32 v0, s30, 0 ; GCN: buffer_store_dword v0, off -; GCN: v_writelane_b32 v0, s31, 0 +; GCN: v_writelane_b32 v0, s30, 0 +; GCN: v_writelane_b32 v0, s31, 1 ; GCN: buffer_store_dword v0, off ; GCN: swappc ; GCN: buffer_load_dword v0, off -; GCN: v_readlane_b32 s31, v0, 0 -; GCN: buffer_load_dword v0, off ; GCN: v_readlane_b32 s30, v0, 0 +; GCN: v_readlane_b32 s31, v0, 1 +; GCN: buffer_load_dword v0, off define void @spill_sgpr_no_free_vgpr_ipra() #0 { call void @child_function_ipra() ret void diff --git a/llvm/test/CodeGen/AMDGPU/si-lower-sgpr-spills.mir b/llvm/test/CodeGen/AMDGPU/si-lower-sgpr-spills.mir --- a/llvm/test/CodeGen/AMDGPU/si-lower-sgpr-spills.mir +++ b/llvm/test/CodeGen/AMDGPU/si-lower-sgpr-spills.mir @@ -2,9 +2,13 @@ # CHECK-LABEL: name: empty_entry_block # CHECK: V_WRITELANE +# CHECK: frame-setup CFI_INSTRUCTION # CHECK-NEXT: V_WRITELANE +# CHECK: frame-setup CFI_INSTRUCTION # CHECK-NEXT: V_WRITELANE +# CHECK: frame-setup CFI_INSTRUCTION # CHECK-NEXT: V_WRITELANE +# CHECK: frame-setup CFI_INSTRUCTION # CHECK: V_READLANE # CHECK-NEXT: V_READLANE # CHECK-NEXT: V_READLANE diff --git a/llvm/test/CodeGen/AMDGPU/spill-sgpr-csr-live-ins.mir b/llvm/test/CodeGen/AMDGPU/spill-sgpr-csr-live-ins.mir --- a/llvm/test/CodeGen/AMDGPU/spill-sgpr-csr-live-ins.mir +++ b/llvm/test/CodeGen/AMDGPU/spill-sgpr-csr-live-ins.mir @@ -13,6 +13,7 @@ ; CHECK: liveins: $sgpr50, $vgpr0 ; CHECK-NEXT: {{ $}} ; CHECK-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr50, 0, $vgpr0 + ; CHECK-NEXT: frame-setup CFI_INSTRUCTION escape 0x10, 0x52, 0x05, 0x90, 0x80, 0x14, 0xec, 0x00 ; CHECK-NEXT: S_NOP 0, implicit $sgpr50 ; CHECK-NEXT: $sgpr50 = S_MOV_B32 0 S_NOP 0, implicit $sgpr50 diff --git a/llvm/test/CodeGen/AMDGPU/stack-realign.ll b/llvm/test/CodeGen/AMDGPU/stack-realign.ll --- a/llvm/test/CodeGen/AMDGPU/stack-realign.ll +++ b/llvm/test/CodeGen/AMDGPU/stack-realign.ll @@ -163,16 +163,16 @@ ; GCN-DAG: s_add_i32 [[SCRATCH_REG:s[0-9]+]], s32, 0xffc0 ; GCN: s_and_b32 s33, [[SCRATCH_REG]], 0xffff0000 ; GCN: s_mov_b32 s34, s32 +; GCN: s_add_i32 s32, s32, 0x30000 ; GCN: v_mov_b32_e32 v32, 0 ; GCN: buffer_store_dword v32, off, s[0:3], s33 offset:1024 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_load_dword v{{[0-9]+}}, off, s[0:3], s34 -; GCN-DAG: s_add_i32 s32, s32, 0x30000 ; GCN: buffer_store_dword v{{[0-9]+}}, off, s[0:3], s32 ; GCN: s_swappc_b64 s[30:31], -; GCN: v_readlane_b32 s31, [[VGPR_REG]], 1 -; GCN: v_readlane_b32 s30, [[VGPR_REG]], 0 +; GCN-DAG: v_readlane_b32 s31, [[VGPR_REG]], 1 +; GCN-DAG: v_readlane_b32 s30, [[VGPR_REG]], 0 ; GCN: s_add_i32 s32, s32, 0xfffd0000 ; GCN-NEXT: v_readlane_b32 s33, [[VGPR_REG]], 2 ; GCN-NEXT: v_readlane_b32 s34, [[VGPR_REG]], 3 diff --git a/llvm/test/CodeGen/AMDGPU/tail-call-amdgpu-gfx.ll b/llvm/test/CodeGen/AMDGPU/tail-call-amdgpu-gfx.ll --- a/llvm/test/CodeGen/AMDGPU/tail-call-amdgpu-gfx.ll +++ b/llvm/test/CodeGen/AMDGPU/tail-call-amdgpu-gfx.ll @@ -22,18 +22,18 @@ ; GCN-NEXT: s_mov_b64 exec, s[34:35] ; GCN-NEXT: v_writelane_b32 v1, s33, 3 ; GCN-NEXT: s_mov_b32 s33, s32 -; GCN-NEXT: v_writelane_b32 v1, s4, 0 ; GCN-NEXT: s_addk_i32 s32, 0x400 +; GCN-NEXT: v_writelane_b32 v1, s4, 0 ; GCN-NEXT: v_writelane_b32 v1, s30, 1 +; GCN-NEXT: v_writelane_b32 v1, s31, 2 ; GCN-NEXT: v_add_f32_e32 v0, 1.0, v0 ; GCN-NEXT: s_mov_b32 s4, 2.0 -; GCN-NEXT: v_writelane_b32 v1, s31, 2 ; GCN-NEXT: s_getpc_b64 s[34:35] ; GCN-NEXT: s_add_u32 s34, s34, callee@rel32@lo+4 ; GCN-NEXT: s_addc_u32 s35, s35, callee@rel32@hi+12 ; GCN-NEXT: s_swappc_b64 s[30:31], s[34:35] -; GCN-NEXT: v_readlane_b32 s31, v1, 2 ; GCN-NEXT: v_readlane_b32 s30, v1, 1 +; GCN-NEXT: v_readlane_b32 s31, v1, 2 ; GCN-NEXT: v_readlane_b32 s4, v1, 0 ; GCN-NEXT: s_addk_i32 s32, 0xfc00 ; GCN-NEXT: v_readlane_b32 s33, v1, 3 diff --git a/llvm/test/CodeGen/AMDGPU/unstructured-cfg-def-use-issue.ll b/llvm/test/CodeGen/AMDGPU/unstructured-cfg-def-use-issue.ll --- a/llvm/test/CodeGen/AMDGPU/unstructured-cfg-def-use-issue.ll +++ b/llvm/test/CodeGen/AMDGPU/unstructured-cfg-def-use-issue.ll @@ -49,8 +49,8 @@ ; GCN-NEXT: v_mov_b32_e32 v1, 0 ; GCN-NEXT: flat_store_dword v[0:1], v2 ; GCN-NEXT: .LBB0_7: ; %UnifiedReturnBlock -; GCN-NEXT: v_readlane_b32 s31, v40, 1 ; GCN-NEXT: v_readlane_b32 s30, v40, 0 +; GCN-NEXT: v_readlane_b32 s31, v40, 1 ; GCN-NEXT: s_addk_i32 s32, 0xfc00 ; GCN-NEXT: v_readlane_b32 s33, v40, 2 ; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -194,24 +194,24 @@ ; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 ; 4-byte Folded Spill -; GCN-NEXT: v_writelane_b32 v40, s30, 0 -; GCN-NEXT: v_writelane_b32 v40, s31, 1 -; GCN-NEXT: v_writelane_b32 v40, s34, 2 -; GCN-NEXT: v_writelane_b32 v40, s35, 3 -; GCN-NEXT: v_writelane_b32 v40, s36, 4 -; GCN-NEXT: v_writelane_b32 v40, s37, 5 -; GCN-NEXT: v_writelane_b32 v40, s38, 6 -; GCN-NEXT: v_writelane_b32 v40, s39, 7 -; GCN-NEXT: v_writelane_b32 v40, s40, 8 -; GCN-NEXT: v_writelane_b32 v40, s41, 9 -; GCN-NEXT: v_writelane_b32 v40, s42, 10 -; GCN-NEXT: v_writelane_b32 v40, s43, 11 -; GCN-NEXT: v_writelane_b32 v40, s44, 12 -; GCN-NEXT: v_writelane_b32 v40, s45, 13 -; GCN-NEXT: v_writelane_b32 v40, s46, 14 -; GCN-NEXT: v_writelane_b32 v40, s47, 15 -; GCN-NEXT: v_writelane_b32 v40, s48, 16 -; GCN-NEXT: v_writelane_b32 v40, s49, 17 +; GCN-NEXT: v_writelane_b32 v40, s34, 0 +; GCN-NEXT: v_writelane_b32 v40, s35, 1 +; GCN-NEXT: v_writelane_b32 v40, s36, 2 +; GCN-NEXT: v_writelane_b32 v40, s37, 3 +; GCN-NEXT: v_writelane_b32 v40, s38, 4 +; GCN-NEXT: v_writelane_b32 v40, s39, 5 +; GCN-NEXT: v_writelane_b32 v40, s40, 6 +; GCN-NEXT: v_writelane_b32 v40, s41, 7 +; GCN-NEXT: v_writelane_b32 v40, s42, 8 +; GCN-NEXT: v_writelane_b32 v40, s43, 9 +; GCN-NEXT: v_writelane_b32 v40, s44, 10 +; GCN-NEXT: v_writelane_b32 v40, s45, 11 +; GCN-NEXT: v_writelane_b32 v40, s46, 12 +; GCN-NEXT: v_writelane_b32 v40, s47, 13 +; GCN-NEXT: v_writelane_b32 v40, s48, 14 +; GCN-NEXT: v_writelane_b32 v40, s49, 15 +; GCN-NEXT: v_writelane_b32 v40, s30, 16 +; GCN-NEXT: v_writelane_b32 v40, s31, 17 ; GCN-NEXT: v_mov_b32_e32 v41, v31 ; GCN-NEXT: s_mov_b32 s44, s15 ; GCN-NEXT: s_mov_b32 s45, s14 diff --git a/llvm/test/CodeGen/AMDGPU/vgpr-tuple-allocation.ll b/llvm/test/CodeGen/AMDGPU/vgpr-tuple-allocation.ll --- a/llvm/test/CodeGen/AMDGPU/vgpr-tuple-allocation.ll +++ b/llvm/test/CodeGen/AMDGPU/vgpr-tuple-allocation.ll @@ -16,6 +16,13 @@ ; GFX9-NEXT: s_mov_b64 exec, s[4:5] ; GFX9-NEXT: v_writelane_b32 v40, s33, 2 ; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_addk_i32 s32, 0x800 +; GFX9-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill +; GFX9-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill +; GFX9-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill +; GFX9-NEXT: buffer_store_dword v44, off, s[0:3], s33 ; 4-byte Folded Spill +; GFX9-NEXT: v_writelane_b32 v40, s30, 0 +; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_mov_b32 s4, 0 ; GFX9-NEXT: v_mov_b32_e32 v36, v16 ; GFX9-NEXT: v_mov_b32_e32 v35, v15 @@ -29,10 +36,6 @@ ; GFX9-NEXT: s_mov_b32 s9, s4 ; GFX9-NEXT: s_mov_b32 s10, s4 ; GFX9-NEXT: s_mov_b32 s11, s4 -; GFX9-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill -; GFX9-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill -; GFX9-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill -; GFX9-NEXT: buffer_store_dword v44, off, s[0:3], s33 ; 4-byte Folded Spill ; GFX9-NEXT: ;;#ASMSTART ; GFX9-NEXT: ;;#ASMEND ; GFX9-NEXT: ;;#ASMSTART @@ -42,13 +45,10 @@ ; GFX9-NEXT: ;;#ASMSTART ; GFX9-NEXT: ;;#ASMEND ; GFX9-NEXT: image_gather4_c_b_cl v[41:44], v[32:36], s[4:11], s[4:7] dmask:0x1 -; GFX9-NEXT: s_addk_i32 s32, 0x800 ; GFX9-NEXT: s_getpc_b64 s[4:5] ; GFX9-NEXT: s_add_u32 s4, s4, extern_func@gotpcrel32@lo+4 ; GFX9-NEXT: s_addc_u32 s5, s5, extern_func@gotpcrel32@hi+12 ; GFX9-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x0 -; GFX9-NEXT: v_writelane_b32 v40, s30, 0 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: s_swappc_b64 s[30:31], s[4:5] ; GFX9-NEXT: v_mov_b32_e32 v0, v41 @@ -59,8 +59,8 @@ ; GFX9-NEXT: buffer_load_dword v43, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload ; GFX9-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload ; GFX9-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s31, v40, 1 ; GFX9-NEXT: s_addk_i32 s32, 0xf800 ; GFX9-NEXT: v_readlane_b32 s33, v40, 2 ; GFX9-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -79,16 +79,21 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s4 ; GFX10-NEXT: v_writelane_b32 v40, s33, 2 ; GFX10-NEXT: s_mov_b32 s33, s32 +; GFX10-NEXT: s_addk_i32 s32, 0x400 +; GFX10-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill +; GFX10-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill +; GFX10-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill +; GFX10-NEXT: buffer_store_dword v44, off, s[0:3], s33 ; 4-byte Folded Spill +; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: v_mov_b32_e32 v36, v16 ; GFX10-NEXT: v_mov_b32_e32 v35, v15 ; GFX10-NEXT: v_mov_b32_e32 v34, v14 ; GFX10-NEXT: v_mov_b32_e32 v33, v13 ; GFX10-NEXT: v_mov_b32_e32 v32, v12 ; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill -; GFX10-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill -; GFX10-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill -; GFX10-NEXT: buffer_store_dword v44, off, s[0:3], s33 ; 4-byte Folded Spill +; GFX10-NEXT: ;;#ASMSTART +; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: s_mov_b32 s5, s4 ; GFX10-NEXT: s_mov_b32 s6, s4 ; GFX10-NEXT: s_mov_b32 s7, s4 @@ -102,17 +107,12 @@ ; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: ;;#ASMSTART ; GFX10-NEXT: ;;#ASMEND -; GFX10-NEXT: ;;#ASMSTART -; GFX10-NEXT: ;;#ASMEND ; GFX10-NEXT: image_gather4_c_b_cl v[41:44], v[32:36], s[4:11], s[4:7] dmask:0x1 dim:SQ_RSRC_IMG_2D -; GFX10-NEXT: s_addk_i32 s32, 0x400 ; GFX10-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-NEXT: s_getpc_b64 s[4:5] ; GFX10-NEXT: s_add_u32 s4, s4, extern_func@gotpcrel32@lo+4 ; GFX10-NEXT: s_addc_u32 s5, s5, extern_func@gotpcrel32@hi+12 -; GFX10-NEXT: v_writelane_b32 v40, s30, 0 ; GFX10-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x0 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_swappc_b64 s[30:31], s[4:5] ; GFX10-NEXT: v_mov_b32_e32 v0, v41 @@ -124,8 +124,8 @@ ; GFX10-NEXT: buffer_load_dword v43, off, s[0:3], s33 offset:4 ; GFX10-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:8 ; GFX10-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:12 -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s31, v40, 1 ; GFX10-NEXT: s_addk_i32 s32, 0xfc00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 2 ; GFX10-NEXT: s_or_saveexec_b32 s4, -1 @@ -144,15 +144,23 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 2 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_dual_mov_b32 v36, v16 :: v_dual_mov_b32 v35, v15 -; GFX11-NEXT: v_dual_mov_b32 v34, v14 :: v_dual_mov_b32 v33, v13 -; GFX11-NEXT: v_mov_b32_e32 v32, v12 -; GFX11-NEXT: s_mov_b32 s0, 0 +; GFX11-NEXT: s_add_i32 s32, s32, 32 ; GFX11-NEXT: s_clause 0x3 ; GFX11-NEXT: scratch_store_b32 off, v41, s33 offset:12 +; GFX11-NEXT: ; meta instruction ; GFX11-NEXT: scratch_store_b32 off, v42, s33 offset:8 +; GFX11-NEXT: ; meta instruction ; GFX11-NEXT: scratch_store_b32 off, v43, s33 offset:4 +; GFX11-NEXT: ; meta instruction ; GFX11-NEXT: scratch_store_b32 off, v44, s33 +; GFX11-NEXT: v_writelane_b32 v40, s30, 0 +; GFX11-NEXT: v_writelane_b32 v40, s31, 1 +; GFX11-NEXT: v_dual_mov_b32 v36, v16 :: v_dual_mov_b32 v35, v15 +; GFX11-NEXT: v_dual_mov_b32 v34, v14 :: v_dual_mov_b32 v33, v13 +; GFX11-NEXT: v_mov_b32_e32 v32, v12 +; GFX11-NEXT: s_mov_b32 s0, 0 +; GFX11-NEXT: ;;#ASMSTART +; GFX11-NEXT: ;;#ASMEND ; GFX11-NEXT: s_mov_b32 s1, s0 ; GFX11-NEXT: s_mov_b32 s2, s0 ; GFX11-NEXT: s_mov_b32 s3, s0 @@ -166,16 +174,11 @@ ; GFX11-NEXT: ;;#ASMEND ; GFX11-NEXT: ;;#ASMSTART ; GFX11-NEXT: ;;#ASMEND -; GFX11-NEXT: ;;#ASMSTART -; GFX11-NEXT: ;;#ASMEND ; GFX11-NEXT: image_gather4_c_b_cl v[41:44], v[32:36], s[0:7], s[0:3] dmask:0x1 dim:SQ_RSRC_IMG_2D -; GFX11-NEXT: s_add_i32 s32, s32, 32 ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, extern_func@gotpcrel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, extern_func@gotpcrel32@hi+12 -; GFX11-NEXT: v_writelane_b32 v40, s30, 0 ; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_swappc_b64 s[30:31], s[0:1] ; GFX11-NEXT: v_dual_mov_b32 v0, v41 :: v_dual_mov_b32 v1, v42 @@ -185,8 +188,8 @@ ; GFX11-NEXT: scratch_load_b32 v43, off, s33 offset:4 ; GFX11-NEXT: scratch_load_b32 v42, off, s33 offset:8 ; GFX11-NEXT: scratch_load_b32 v41, off, s33 offset:12 -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s31, v40, 1 ; GFX11-NEXT: s_addk_i32 s32, 0xffe0 ; GFX11-NEXT: v_readlane_b32 s33, v40, 2 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 @@ -227,22 +230,23 @@ ; GFX9-NEXT: s_mov_b64 exec, s[4:5] ; GFX9-NEXT: v_writelane_b32 v40, s33, 10 ; GFX9-NEXT: s_mov_b32 s33, s32 -; GFX9-NEXT: v_writelane_b32 v40, s30, 0 -; GFX9-NEXT: v_writelane_b32 v40, s31, 1 -; GFX9-NEXT: v_writelane_b32 v40, s36, 2 -; GFX9-NEXT: v_writelane_b32 v40, s37, 3 -; GFX9-NEXT: v_writelane_b32 v40, s38, 4 -; GFX9-NEXT: v_writelane_b32 v40, s39, 5 -; GFX9-NEXT: v_writelane_b32 v40, s40, 6 -; GFX9-NEXT: v_writelane_b32 v40, s41, 7 -; GFX9-NEXT: v_writelane_b32 v40, s42, 8 -; GFX9-NEXT: s_mov_b32 s36, 0 +; GFX9-NEXT: s_addk_i32 s32, 0x800 ; GFX9-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill ; GFX9-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill ; GFX9-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill ; GFX9-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill ; GFX9-NEXT: buffer_store_dword v45, off, s[0:3], s33 ; 4-byte Folded Spill -; GFX9-NEXT: v_writelane_b32 v40, s43, 9 +; GFX9-NEXT: v_writelane_b32 v40, s36, 0 +; GFX9-NEXT: v_writelane_b32 v40, s37, 1 +; GFX9-NEXT: v_writelane_b32 v40, s38, 2 +; GFX9-NEXT: v_writelane_b32 v40, s39, 3 +; GFX9-NEXT: v_writelane_b32 v40, s40, 4 +; GFX9-NEXT: v_writelane_b32 v40, s41, 5 +; GFX9-NEXT: v_writelane_b32 v40, s42, 6 +; GFX9-NEXT: v_writelane_b32 v40, s43, 7 +; GFX9-NEXT: v_writelane_b32 v40, s30, 8 +; GFX9-NEXT: v_writelane_b32 v40, s31, 9 +; GFX9-NEXT: s_mov_b32 s36, 0 ; GFX9-NEXT: v_mov_b32_e32 v45, v16 ; GFX9-NEXT: v_mov_b32_e32 v44, v15 ; GFX9-NEXT: v_mov_b32_e32 v43, v14 @@ -256,7 +260,6 @@ ; GFX9-NEXT: s_mov_b32 s42, s36 ; GFX9-NEXT: s_mov_b32 s43, s36 ; GFX9-NEXT: image_gather4_c_b_cl v[0:3], v[41:45], s[36:43], s[4:7] dmask:0x1 -; GFX9-NEXT: s_addk_i32 s32, 0x800 ; GFX9-NEXT: s_getpc_b64 s[4:5] ; GFX9-NEXT: s_add_u32 s4, s4, extern_func@gotpcrel32@lo+4 ; GFX9-NEXT: s_addc_u32 s5, s5, extern_func@gotpcrel32@hi+12 @@ -272,16 +275,16 @@ ; GFX9-NEXT: buffer_load_dword v43, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload ; GFX9-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload ; GFX9-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:16 ; 4-byte Folded Reload -; GFX9-NEXT: v_readlane_b32 s43, v40, 9 -; GFX9-NEXT: v_readlane_b32 s42, v40, 8 -; GFX9-NEXT: v_readlane_b32 s41, v40, 7 -; GFX9-NEXT: v_readlane_b32 s40, v40, 6 -; GFX9-NEXT: v_readlane_b32 s39, v40, 5 -; GFX9-NEXT: v_readlane_b32 s38, v40, 4 -; GFX9-NEXT: v_readlane_b32 s37, v40, 3 -; GFX9-NEXT: v_readlane_b32 s36, v40, 2 -; GFX9-NEXT: v_readlane_b32 s31, v40, 1 -; GFX9-NEXT: v_readlane_b32 s30, v40, 0 +; GFX9-NEXT: v_readlane_b32 s30, v40, 8 +; GFX9-NEXT: v_readlane_b32 s31, v40, 9 +; GFX9-NEXT: v_readlane_b32 s43, v40, 7 +; GFX9-NEXT: v_readlane_b32 s42, v40, 6 +; GFX9-NEXT: v_readlane_b32 s41, v40, 5 +; GFX9-NEXT: v_readlane_b32 s40, v40, 4 +; GFX9-NEXT: v_readlane_b32 s39, v40, 3 +; GFX9-NEXT: v_readlane_b32 s38, v40, 2 +; GFX9-NEXT: v_readlane_b32 s37, v40, 1 +; GFX9-NEXT: v_readlane_b32 s36, v40, 0 ; GFX9-NEXT: s_addk_i32 s32, 0xf800 ; GFX9-NEXT: v_readlane_b32 s33, v40, 10 ; GFX9-NEXT: s_or_saveexec_b64 s[4:5], -1 @@ -300,41 +303,41 @@ ; GFX10-NEXT: s_mov_b32 exec_lo, s4 ; GFX10-NEXT: v_writelane_b32 v40, s33, 10 ; GFX10-NEXT: s_mov_b32 s33, s32 -; GFX10-NEXT: v_writelane_b32 v40, s30, 0 +; GFX10-NEXT: s_addk_i32 s32, 0x400 ; GFX10-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill ; GFX10-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill ; GFX10-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill ; GFX10-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill ; GFX10-NEXT: buffer_store_dword v45, off, s[0:3], s33 ; 4-byte Folded Spill -; GFX10-NEXT: s_addk_i32 s32, 0x400 -; GFX10-NEXT: v_mov_b32_e32 v41, v16 -; GFX10-NEXT: v_mov_b32_e32 v42, v15 -; GFX10-NEXT: v_writelane_b32 v40, s31, 1 -; GFX10-NEXT: v_mov_b32_e32 v43, v14 -; GFX10-NEXT: v_mov_b32_e32 v44, v13 -; GFX10-NEXT: v_mov_b32_e32 v45, v12 -; GFX10-NEXT: v_writelane_b32 v40, s36, 2 +; GFX10-NEXT: v_writelane_b32 v40, s36, 0 +; GFX10-NEXT: v_writelane_b32 v40, s37, 1 +; GFX10-NEXT: v_writelane_b32 v40, s38, 2 +; GFX10-NEXT: v_writelane_b32 v40, s39, 3 +; GFX10-NEXT: v_writelane_b32 v40, s40, 4 +; GFX10-NEXT: v_writelane_b32 v40, s41, 5 +; GFX10-NEXT: v_writelane_b32 v40, s42, 6 +; GFX10-NEXT: v_writelane_b32 v40, s43, 7 +; GFX10-NEXT: v_writelane_b32 v40, s30, 8 +; GFX10-NEXT: v_writelane_b32 v40, s31, 9 ; GFX10-NEXT: s_mov_b32 s36, 0 -; GFX10-NEXT: v_writelane_b32 v40, s37, 3 +; GFX10-NEXT: v_mov_b32_e32 v41, v16 ; GFX10-NEXT: s_mov_b32 s37, s36 -; GFX10-NEXT: v_writelane_b32 v40, s38, 4 ; GFX10-NEXT: s_mov_b32 s38, s36 -; GFX10-NEXT: v_writelane_b32 v40, s39, 5 ; GFX10-NEXT: s_mov_b32 s39, s36 -; GFX10-NEXT: v_writelane_b32 v40, s40, 6 ; GFX10-NEXT: s_mov_b32 s40, s36 -; GFX10-NEXT: v_writelane_b32 v40, s41, 7 ; GFX10-NEXT: s_mov_b32 s41, s36 -; GFX10-NEXT: v_writelane_b32 v40, s42, 8 ; GFX10-NEXT: s_mov_b32 s42, s36 -; GFX10-NEXT: v_writelane_b32 v40, s43, 9 ; GFX10-NEXT: s_mov_b32 s43, s36 +; GFX10-NEXT: v_mov_b32_e32 v42, v15 ; GFX10-NEXT: image_gather4_c_b_cl v[0:3], v[12:16], s[36:43], s[4:7] dmask:0x1 dim:SQ_RSRC_IMG_2D ; GFX10-NEXT: s_waitcnt_depctr 0xffe3 ; GFX10-NEXT: s_getpc_b64 s[4:5] ; GFX10-NEXT: s_add_u32 s4, s4, extern_func@gotpcrel32@lo+4 ; GFX10-NEXT: s_addc_u32 s5, s5, extern_func@gotpcrel32@hi+12 +; GFX10-NEXT: v_mov_b32_e32 v43, v14 ; GFX10-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x0 +; GFX10-NEXT: v_mov_b32_e32 v44, v13 +; GFX10-NEXT: v_mov_b32_e32 v45, v12 ; GFX10-NEXT: s_waitcnt vmcnt(0) ; GFX10-NEXT: global_store_dwordx4 v[0:1], v[0:3], off ; GFX10-NEXT: s_waitcnt lgkmcnt(0) @@ -346,16 +349,16 @@ ; GFX10-NEXT: buffer_load_dword v43, off, s[0:3], s33 offset:8 ; GFX10-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:12 ; GFX10-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:16 -; GFX10-NEXT: v_readlane_b32 s43, v40, 9 -; GFX10-NEXT: v_readlane_b32 s42, v40, 8 -; GFX10-NEXT: v_readlane_b32 s41, v40, 7 -; GFX10-NEXT: v_readlane_b32 s40, v40, 6 -; GFX10-NEXT: v_readlane_b32 s39, v40, 5 -; GFX10-NEXT: v_readlane_b32 s38, v40, 4 -; GFX10-NEXT: v_readlane_b32 s37, v40, 3 -; GFX10-NEXT: v_readlane_b32 s36, v40, 2 -; GFX10-NEXT: v_readlane_b32 s31, v40, 1 -; GFX10-NEXT: v_readlane_b32 s30, v40, 0 +; GFX10-NEXT: v_readlane_b32 s30, v40, 8 +; GFX10-NEXT: v_readlane_b32 s31, v40, 9 +; GFX10-NEXT: v_readlane_b32 s43, v40, 7 +; GFX10-NEXT: v_readlane_b32 s42, v40, 6 +; GFX10-NEXT: v_readlane_b32 s41, v40, 5 +; GFX10-NEXT: v_readlane_b32 s40, v40, 4 +; GFX10-NEXT: v_readlane_b32 s39, v40, 3 +; GFX10-NEXT: v_readlane_b32 s38, v40, 2 +; GFX10-NEXT: v_readlane_b32 s37, v40, 1 +; GFX10-NEXT: v_readlane_b32 s36, v40, 0 ; GFX10-NEXT: s_addk_i32 s32, 0xfc00 ; GFX10-NEXT: v_readlane_b32 s33, v40, 10 ; GFX10-NEXT: s_or_saveexec_b32 s4, -1 @@ -374,38 +377,42 @@ ; GFX11-NEXT: s_mov_b32 exec_lo, s0 ; GFX11-NEXT: v_writelane_b32 v40, s33, 10 ; GFX11-NEXT: s_mov_b32 s33, s32 -; GFX11-NEXT: v_writelane_b32 v40, s30, 0 +; GFX11-NEXT: s_add_i32 s32, s32, 32 ; GFX11-NEXT: s_clause 0x4 ; GFX11-NEXT: scratch_store_b32 off, v41, s33 offset:16 +; GFX11-NEXT: ; meta instruction ; GFX11-NEXT: scratch_store_b32 off, v42, s33 offset:12 +; GFX11-NEXT: ; meta instruction ; GFX11-NEXT: scratch_store_b32 off, v43, s33 offset:8 +; GFX11-NEXT: ; meta instruction ; GFX11-NEXT: scratch_store_b32 off, v44, s33 offset:4 +; GFX11-NEXT: ; meta instruction ; GFX11-NEXT: scratch_store_b32 off, v45, s33 -; GFX11-NEXT: s_add_i32 s32, s32, 32 -; GFX11-NEXT: v_dual_mov_b32 v41, v16 :: v_dual_mov_b32 v42, v15 -; GFX11-NEXT: v_writelane_b32 v40, s31, 1 -; GFX11-NEXT: v_dual_mov_b32 v43, v14 :: v_dual_mov_b32 v44, v13 -; GFX11-NEXT: v_mov_b32_e32 v45, v12 -; GFX11-NEXT: v_writelane_b32 v40, s36, 2 +; GFX11-NEXT: v_writelane_b32 v40, s36, 0 +; GFX11-NEXT: v_writelane_b32 v40, s37, 1 +; GFX11-NEXT: v_writelane_b32 v40, s38, 2 +; GFX11-NEXT: v_writelane_b32 v40, s39, 3 +; GFX11-NEXT: v_writelane_b32 v40, s40, 4 +; GFX11-NEXT: v_writelane_b32 v40, s41, 5 +; GFX11-NEXT: v_writelane_b32 v40, s42, 6 +; GFX11-NEXT: v_writelane_b32 v40, s43, 7 +; GFX11-NEXT: v_writelane_b32 v40, s30, 8 +; GFX11-NEXT: v_writelane_b32 v40, s31, 9 ; GFX11-NEXT: s_mov_b32 s36, 0 -; GFX11-NEXT: v_writelane_b32 v40, s37, 3 +; GFX11-NEXT: v_dual_mov_b32 v41, v16 :: v_dual_mov_b32 v42, v15 ; GFX11-NEXT: s_mov_b32 s37, s36 -; GFX11-NEXT: v_writelane_b32 v40, s38, 4 ; GFX11-NEXT: s_mov_b32 s38, s36 -; GFX11-NEXT: v_writelane_b32 v40, s39, 5 ; GFX11-NEXT: s_mov_b32 s39, s36 -; GFX11-NEXT: v_writelane_b32 v40, s40, 6 ; GFX11-NEXT: s_mov_b32 s40, s36 -; GFX11-NEXT: v_writelane_b32 v40, s41, 7 ; GFX11-NEXT: s_mov_b32 s41, s36 -; GFX11-NEXT: v_writelane_b32 v40, s42, 8 ; GFX11-NEXT: s_mov_b32 s42, s36 -; GFX11-NEXT: v_writelane_b32 v40, s43, 9 ; GFX11-NEXT: s_mov_b32 s43, s36 +; GFX11-NEXT: v_dual_mov_b32 v43, v14 :: v_dual_mov_b32 v44, v13 ; GFX11-NEXT: image_gather4_c_b_cl v[0:3], v[12:16], s[36:43], s[0:3] dmask:0x1 dim:SQ_RSRC_IMG_2D ; GFX11-NEXT: s_getpc_b64 s[0:1] ; GFX11-NEXT: s_add_u32 s0, s0, extern_func@gotpcrel32@lo+4 ; GFX11-NEXT: s_addc_u32 s1, s1, extern_func@gotpcrel32@hi+12 +; GFX11-NEXT: v_mov_b32_e32 v45, v12 ; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0 ; GFX11-NEXT: s_waitcnt vmcnt(0) ; GFX11-NEXT: global_store_b128 v[0:1], v[0:3], off @@ -418,16 +425,16 @@ ; GFX11-NEXT: scratch_load_b32 v43, off, s33 offset:8 ; GFX11-NEXT: scratch_load_b32 v42, off, s33 offset:12 ; GFX11-NEXT: scratch_load_b32 v41, off, s33 offset:16 -; GFX11-NEXT: v_readlane_b32 s43, v40, 9 -; GFX11-NEXT: v_readlane_b32 s42, v40, 8 -; GFX11-NEXT: v_readlane_b32 s41, v40, 7 -; GFX11-NEXT: v_readlane_b32 s40, v40, 6 -; GFX11-NEXT: v_readlane_b32 s39, v40, 5 -; GFX11-NEXT: v_readlane_b32 s38, v40, 4 -; GFX11-NEXT: v_readlane_b32 s37, v40, 3 -; GFX11-NEXT: v_readlane_b32 s36, v40, 2 -; GFX11-NEXT: v_readlane_b32 s31, v40, 1 -; GFX11-NEXT: v_readlane_b32 s30, v40, 0 +; GFX11-NEXT: v_readlane_b32 s30, v40, 8 +; GFX11-NEXT: v_readlane_b32 s31, v40, 9 +; GFX11-NEXT: v_readlane_b32 s43, v40, 7 +; GFX11-NEXT: v_readlane_b32 s42, v40, 6 +; GFX11-NEXT: v_readlane_b32 s41, v40, 5 +; GFX11-NEXT: v_readlane_b32 s40, v40, 4 +; GFX11-NEXT: v_readlane_b32 s39, v40, 3 +; GFX11-NEXT: v_readlane_b32 s38, v40, 2 +; GFX11-NEXT: v_readlane_b32 s37, v40, 1 +; GFX11-NEXT: v_readlane_b32 s36, v40, 0 ; GFX11-NEXT: s_addk_i32 s32, 0xffe0 ; GFX11-NEXT: v_readlane_b32 s33, v40, 10 ; GFX11-NEXT: s_or_saveexec_b32 s0, -1 diff --git a/llvm/test/CodeGen/AMDGPU/wave32.ll b/llvm/test/CodeGen/AMDGPU/wave32.ll --- a/llvm/test/CodeGen/AMDGPU/wave32.ll +++ b/llvm/test/CodeGen/AMDGPU/wave32.ll @@ -1118,12 +1118,12 @@ ; GCN-NEXT: v_writelane_b32 v40, s33, 2 ; GCN: s_mov_b32 s33, s32 +; GCN: v_writelane_b32 v40, s30, 0 ; GFX1064: s_addk_i32 s32, 0x400 ; GFX1032: s_addk_i32 s32, 0x200 -; GCN-DAG: v_writelane_b32 v40, s30, 0 -; GCN-DAG: v_writelane_b32 v40, s31, 1 +; GCN: v_writelane_b32 v40, s31, 1 ; GCN: s_swappc_b64 ; GCN-DAG: v_readlane_b32 s30, v40, 0 ; GCN-DAG: v_readlane_b32 s31, v40, 1 diff --git a/llvm/test/CodeGen/AMDGPU/wwm-reserved-spill.ll b/llvm/test/CodeGen/AMDGPU/wwm-reserved-spill.ll --- a/llvm/test/CodeGen/AMDGPU/wwm-reserved-spill.ll +++ b/llvm/test/CodeGen/AMDGPU/wwm-reserved-spill.ll @@ -368,8 +368,8 @@ ; GFX9-O0-NEXT: s_mov_b64 exec, s[40:41] ; GFX9-O0-NEXT: v_mov_b32_e32 v0, v1 ; GFX9-O0-NEXT: buffer_store_dword v0, off, s[36:39], s34 offset:4 -; GFX9-O0-NEXT: v_readlane_b32 s31, v3, 1 ; GFX9-O0-NEXT: v_readlane_b32 s30, v3, 0 +; GFX9-O0-NEXT: v_readlane_b32 s31, v3, 1 ; GFX9-O0-NEXT: s_add_i32 s32, s32, 0xfffffc00 ; GFX9-O0-NEXT: v_readlane_b32 s33, v3, 2 ; GFX9-O0-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -407,9 +407,9 @@ ; GFX9-O3-NEXT: v_add_u32_e32 v1, v1, v2 ; GFX9-O3-NEXT: s_mov_b64 exec, s[34:35] ; GFX9-O3-NEXT: v_mov_b32_e32 v0, v1 +; GFX9-O3-NEXT: v_readlane_b32 s30, v3, 0 ; GFX9-O3-NEXT: buffer_store_dword v0, off, s[4:7], 0 offset:4 ; GFX9-O3-NEXT: v_readlane_b32 s31, v3, 1 -; GFX9-O3-NEXT: v_readlane_b32 s30, v3, 0 ; GFX9-O3-NEXT: s_addk_i32 s32, 0xfc00 ; GFX9-O3-NEXT: v_readlane_b32 s33, v3, 2 ; GFX9-O3-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -596,8 +596,8 @@ ; GFX9-O0-NEXT: v_mov_b32_e32 v1, v3 ; GFX9-O0-NEXT: s_mov_b32 s34, 0 ; GFX9-O0-NEXT: buffer_store_dwordx2 v[0:1], off, s[36:39], s34 offset:4 -; GFX9-O0-NEXT: v_readlane_b32 s31, v10, 1 ; GFX9-O0-NEXT: v_readlane_b32 s30, v10, 0 +; GFX9-O0-NEXT: v_readlane_b32 s31, v10, 1 ; GFX9-O0-NEXT: s_add_i32 s32, s32, 0xfffff400 ; GFX9-O0-NEXT: v_readlane_b32 s33, v10, 8 ; GFX9-O0-NEXT: s_or_saveexec_b64 s[34:35], -1 @@ -668,9 +668,9 @@ ; GFX9-O3-NEXT: s_mov_b64 exec, s[38:39] ; GFX9-O3-NEXT: v_mov_b32_e32 v0, v2 ; GFX9-O3-NEXT: v_mov_b32_e32 v1, v3 +; GFX9-O3-NEXT: v_readlane_b32 s30, v8, 0 ; GFX9-O3-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 offset:4 ; GFX9-O3-NEXT: v_readlane_b32 s31, v8, 1 -; GFX9-O3-NEXT: v_readlane_b32 s30, v8, 0 ; GFX9-O3-NEXT: s_addk_i32 s32, 0xf800 ; GFX9-O3-NEXT: v_readlane_b32 s33, v8, 2 ; GFX9-O3-NEXT: s_or_saveexec_b64 s[34:35], -1