Index: llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp =================================================================== --- llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp +++ llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp @@ -24,6 +24,7 @@ #include "AMDGPUGlobalISelUtils.h" #include "AMDGPUTargetMachine.h" #include "SIMachineFunctionInfo.h" +#include "llvm/ADT/ScopeExit.h" #include "llvm/CodeGen/GlobalISel/LegalizerHelper.h" #include "llvm/CodeGen/GlobalISel/MachineIRBuilder.h" #include "llvm/CodeGen/GlobalISel/MIPatternMatch.h" @@ -3006,7 +3007,24 @@ } } -// Return number of address operands in an image intrinsic. +/// Convert from separate vaddr components to a single vector address register, +/// and replace the remaining operands with $noreg. +static void convertImageAddrToPacked(MachineIRBuilder &B, MachineInstr &MI, + int DimIdx, int NumVAddrs) { + const LLT S32 = LLT::scalar(32); + + SmallVector AddrRegs(NumVAddrs); + for (int I = 0; I != NumVAddrs; ++I) { + AddrRegs[I] = MI.getOperand(DimIdx + I).getReg(); + assert(B.getMRI()->getType(AddrRegs[I]) == S32); + } + + auto VAddr = B.buildBuildVector(LLT::vector(NumVAddrs, 32), AddrRegs); + MI.getOperand(DimIdx).setReg(VAddr.getReg(0)); + for (int I = 1; I != NumVAddrs; ++I) + MI.getOperand(DimIdx + I).setReg(AMDGPU::NoRegister); +} + static int getImageNumVAddr(const AMDGPU::ImageDimIntrinsicInfo *ImageDimIntr, const AMDGPU::MIMGBaseOpcodeInfo *BaseOpcode) { const AMDGPU::MIMGDimInfo *DimInfo @@ -3054,6 +3072,16 @@ const AMDGPU::MIMGBaseOpcodeInfo *BaseOpcode = AMDGPU::getMIMGBaseOpcodeInfo(ImageDimIntr->BaseOpcode); + Observer.changingInstr(MI); + auto ChangedInstr = make_scope_exit([&] { Observer.changedInstr(MI); }); + + + unsigned NewOpcode = NumDefs == 0 ? + AMDGPU::G_AMDGPU_INTRIN_IMAGE_STORE : AMDGPU::G_AMDGPU_INTRIN_IMAGE_LOAD; + + // Track that we legalized this + MI.setDesc(B.getTII().get(NewOpcode)); + B.setInstr(MI); MachineRegisterInfo *MRI = B.getMRI(); @@ -3068,28 +3096,30 @@ LLT AddrTy = MRI->getType(MI.getOperand(DimIdx).getReg()); const bool IsA16 = AddrTy == S16; - // TODO: Handle NSA vs. non-NSA for non-a16 case. + const int NumVAddrs = getImageNumVAddr(ImageDimIntr, BaseOpcode); + + // If the register allocator cannot place the address registers contiguously + // without introducing moves, then using the non-sequential address encoding + // is always preferable, since it saves VALU instructions and is usually a + // wash in terms of code size or even better. + // + // However, we currently have no way of hinting to the register allocator + // that MIMG addresses should be placed contiguously when it is possible to + // do so, so force non-NSA for the common 2-address case as a heuristic. + // + // SIShrinkInstructions will convert NSA encodings to non-NSA after register + // allocation when possible. + const bool UseNSA = NumVAddrs >= 3 && + ST.hasFeature(AMDGPU::FeatureNSAEncoding); // Rewrite the addressing register layout before doing anything else. if (IsA16) { +#if 0 + // FIXME: this feature is missing from gfx10. When that is fixed, this check + // should be introduced. if (!ST.hasFeature(AMDGPU::FeatureR128A16)) return false; - - const int NumVAddrs = getImageNumVAddr(ImageDimIntr, BaseOpcode); - - // If the register allocator cannot place the address registers contiguously - // without introducing moves, then using the non-sequential address encoding - // is always preferable, since it saves VALU instructions and is usually a - // wash in terms of code size or even better. - // - // However, we currently have no way of hinting to the register allocator - // that MIMG addresses should be placed contiguously when it is possible to - // do so, so force non-NSA for the common 2-address case as a heuristic. - // - // SIShrinkInstructions will convert NSA encodings to non-NSA after register - // allocation when possible. - const bool UseNSA = NumVAddrs >= 3 && - ST.hasFeature(AMDGPU::FeatureNSAEncoding); +#endif if (NumVAddrs > 1) { SmallVector PackedRegs; @@ -3102,10 +3132,6 @@ PackedRegs.resize(1); } - // FIXME: We'll notify the observer multiple times if there are further - // modifications later. - Observer.changingInstr(MI); - const int NumPacked = PackedRegs.size(); for (int I = 0; I != NumVAddrs; ++I) { assert(MI.getOperand(DimIdx + I).getReg() != AMDGPU::NoRegister); @@ -3115,9 +3141,9 @@ else MI.getOperand(DimIdx + I).setReg(AMDGPU::NoRegister); } - - Observer.changedInstr(MI); } + } else if (!UseNSA && NumVAddrs > 1) { + convertImageAddrToPacked(B, MI, DimIdx, NumVAddrs); } if (BaseOpcode->Atomic) // No d16 atomics, or TFE. @@ -3133,9 +3159,7 @@ Register RepackedReg = handleD16VData(B, *MRI, VData); if (RepackedReg != VData) { - Observer.changingInstr(MI); MI.getOperand(1).setReg(RepackedReg); - Observer.changedInstr(MI); } return true; @@ -3176,13 +3200,10 @@ } Register TFEReg = MRI->createGenericVirtualRegister(TFETy); - Observer.changingInstr(MI); MI.getOperand(0).setReg(TFEReg); MI.RemoveOperand(1); - Observer.changedInstr(MI); - // Insert after the instruction. B.setInsertPt(*MI.getParent(), ++MI.getIterator()); @@ -3230,9 +3251,7 @@ LLT WidenedTy = Ty.changeElementType(S32); Register WideDstReg = MRI->createGenericVirtualRegister(WidenedTy); - Observer.changingInstr(MI); MI.getOperand(0).setReg(WideDstReg); - Observer.changedInstr(MI); repackUnpackedD16Load(B, DstReg, WideDstReg); return true; Index: llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp =================================================================== --- llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp +++ llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp @@ -2347,6 +2347,17 @@ } break; } + case AMDGPU::G_AMDGPU_INTRIN_IMAGE_LOAD: + case AMDGPU::G_AMDGPU_INTRIN_IMAGE_STORE: { + const AMDGPU::RsrcIntrinsic *RSrcIntrin + = AMDGPU::lookupRsrcIntrinsic(MI.getIntrinsicID()); + assert(RSrcIntrin && RSrcIntrin->IsImage); + // Non-images can have complications from operands that allow both SGPR + // and VGPR. For now it's too complicated to figure out the final opcode + // to derive the register bank from the MCInstrDesc. + applyMappingImage(MI, OpdMapper, MRI, RSrcIntrin->RsrcArg); + return; + } case AMDGPU::G_INTRINSIC_W_SIDE_EFFECTS: { auto IntrID = MI.getIntrinsicID(); switch (IntrID) { @@ -3424,6 +3435,17 @@ } break; } + case AMDGPU::G_AMDGPU_INTRIN_IMAGE_LOAD: + case AMDGPU::G_AMDGPU_INTRIN_IMAGE_STORE: { + auto IntrID = MI.getIntrinsicID(); + const AMDGPU::RsrcIntrinsic *RSrcIntrin = AMDGPU::lookupRsrcIntrinsic(IntrID); + assert(RSrcIntrin && "missing RsrcIntrinsic for image intrinsic"); + // Non-images can have complications from operands that allow both SGPR + // and VGPR. For now it's too complicated to figure out the final opcode + // to derive the register bank from the MCInstrDesc. + assert(RSrcIntrin->IsImage); + return getImageMapping(MRI, MI, RSrcIntrin->RsrcArg); + } case AMDGPU::G_INTRINSIC_W_SIDE_EFFECTS: { auto IntrID = MI.getIntrinsicID(); switch (IntrID) { @@ -3556,15 +3578,6 @@ break; } default: - if (const AMDGPU::RsrcIntrinsic *RSrcIntrin = - AMDGPU::lookupRsrcIntrinsic(IntrID)) { - // Non-images can have complications from operands that allow both SGPR - // and VGPR. For now it's too complicated to figure out the final opcode - // to derive the register bank from the MCInstrDesc. - if (RSrcIntrin->IsImage) - return getImageMapping(MRI, MI, RSrcIntrin->RsrcArg); - } - return getInvalidInstructionMapping(); } break; Index: llvm/lib/Target/AMDGPU/SIInstructions.td =================================================================== --- llvm/lib/Target/AMDGPU/SIInstructions.td +++ llvm/lib/Target/AMDGPU/SIInstructions.td @@ -2265,3 +2265,21 @@ let mayLoad = 1; let mayStore = 1; } + +// This is equivalent to the G_INTRINSIC*, but the operands may have +// been legalized depending on the subtarget requirements. +def G_AMDGPU_INTRIN_IMAGE_LOAD : AMDGPUGenericInstruction { + let OutOperandList = (outs type0:$dst); + let InOperandList = (ins unknown:$intrin, variable_ops); + let hasSideEffects = 0; + let mayLoad = 1; +} + +// This is equivalent to the G_INTRINSIC*, but the operands may have +// been legalized depending on the subtarget requirements. +def G_AMDGPU_INTRIN_IMAGE_STORE : AMDGPUGenericInstruction { + let OutOperandList = (outs); + let InOperandList = (ins unknown:$intrin, variable_ops); + let hasSideEffects = 0; + let mayStore = 1; +} Index: llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.atomic.dim.a16.ll =================================================================== --- llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.atomic.dim.a16.ll +++ llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.atomic.dim.a16.ll @@ -18,8 +18,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.swap.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.swap.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_swap_1d ; GFX10NSA: bb.1.main_body: @@ -36,8 +36,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.swap.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.swap.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.swap.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0) @@ -61,8 +61,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_add_1d ; GFX10NSA: bb.1.main_body: @@ -79,8 +79,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.add.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0) @@ -104,8 +104,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.sub.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.sub.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_sub_1d ; GFX10NSA: bb.1.main_body: @@ -122,8 +122,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.sub.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.sub.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.sub.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0) @@ -147,8 +147,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.smin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.smin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_smin_1d ; GFX10NSA: bb.1.main_body: @@ -165,8 +165,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.smin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.smin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.smin.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0) @@ -191,8 +191,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.umin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.umin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_umin_1d ; GFX10NSA: bb.1.main_body: @@ -209,8 +209,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.umin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.umin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.umin.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0) @@ -234,8 +234,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.smax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.smax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_smax_1d ; GFX10NSA: bb.1.main_body: @@ -252,8 +252,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.smax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.smax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.smax.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0) @@ -277,8 +277,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.umax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.umax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_umax_1d ; GFX10NSA: bb.1.main_body: @@ -295,8 +295,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.umax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.umax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.umax.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0) @@ -320,8 +320,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.and.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.and.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_and_1d ; GFX10NSA: bb.1.main_body: @@ -338,8 +338,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.and.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.and.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.and.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0) @@ -363,8 +363,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.or.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.or.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_or_1d ; GFX10NSA: bb.1.main_body: @@ -381,8 +381,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.or.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.or.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.or.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0) @@ -406,8 +406,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.xor.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.xor.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_xor_1d ; GFX10NSA: bb.1.main_body: @@ -424,8 +424,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.xor.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.xor.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.xor.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0) @@ -449,8 +449,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.inc.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.inc.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_inc_1d ; GFX10NSA: bb.1.main_body: @@ -467,8 +467,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.inc.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.inc.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.inc.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0) @@ -492,8 +492,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.dec.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.dec.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_dec_1d ; GFX10NSA: bb.1.main_body: @@ -510,8 +510,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.dec.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.dec.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.dec.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0) @@ -536,8 +536,8 @@ ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2 ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY10]](s32) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.cmpswap.1d), [[COPY8]](s32), [[COPY9]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.1d), [[COPY8]](s32), [[COPY9]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_cmpswap_1d ; GFX10NSA: bb.1.main_body: @@ -555,8 +555,8 @@ ; GFX10NSA: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2 ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY10]](s32) ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.cmpswap.1d), [[COPY8]](s32), [[COPY9]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.1d), [[COPY8]](s32), [[COPY9]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.cmpswap.1d.i32.i16(i32 %cmp, i32 %swap, i16 %s, <8 x i32> %rsrc, i32 0, i32 0) @@ -583,8 +583,8 @@ ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32) ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.2d), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2d), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_add_2d ; GFX10NSA: bb.1.main_body: @@ -604,8 +604,8 @@ ; GFX10NSA: [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32) ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.2d), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2d), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.add.2d.i32.i16(i32 %data, i16 %s, i16 %t, <8 x i32> %rsrc, i32 0, i32 0) @@ -637,8 +637,8 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.3d), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.3d), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_add_3d ; GFX10NSA: bb.1.main_body: @@ -662,8 +662,8 @@ ; GFX10NSA: [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.3d), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.3d), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.add.3d.i32.i16(i32 %data, i16 %s, i16 %t, i16 %r, <8 x i32> %rsrc, i32 0, i32 0) @@ -695,8 +695,8 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.cube), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.cube), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_add_cube ; GFX10NSA: bb.1.main_body: @@ -720,8 +720,8 @@ ; GFX10NSA: [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.cube), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.cube), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.add.cube.i32.i16(i32 %data, i16 %s, i16 %t, i16 %face, <8 x i32> %rsrc, i32 0, i32 0) @@ -748,8 +748,8 @@ ; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32) ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.1darray), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1darray), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_add_1darray ; GFX10NSA: bb.1.main_body: @@ -769,8 +769,8 @@ ; GFX10NSA: [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32) ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.1darray), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1darray), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.add.1darray.i32.i16(i32 %data, i16 %s, i16 %slice, <8 x i32> %rsrc, i32 0, i32 0) @@ -802,8 +802,8 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.2darray), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2darray), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_add_2darray ; GFX10NSA: bb.1.main_body: @@ -827,8 +827,8 @@ ; GFX10NSA: [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.2darray), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2darray), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.add.2darray.i32.i16(i32 %data, i16 %s, i16 %t, i16 %slice, <8 x i32> %rsrc, i32 0, i32 0) @@ -860,8 +860,8 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.2dmsaa), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2dmsaa), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_add_2dmsaa ; GFX10NSA: bb.1.main_body: @@ -885,8 +885,8 @@ ; GFX10NSA: [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.2dmsaa), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2dmsaa), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.add.2dmsaa.i32.i16(i32 %data, i16 %s, i16 %t, i16 %fragid, <8 x i32> %rsrc, i32 0, i32 0) @@ -919,8 +919,8 @@ ; GFX9: [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.2darraymsaa), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2darraymsaa), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_add_2darraymsaa ; GFX10NSA: bb.1.main_body: @@ -945,8 +945,8 @@ ; GFX10NSA: [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32) ; GFX10NSA: [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.2darraymsaa), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2darraymsaa), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.add.2darraymsaa.i32.i16(i32 %data, i16 %s, i16 %t, i16 %slice, i16 %fragid, <8 x i32> %rsrc, i32 0, i32 0) @@ -970,8 +970,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_add_1d_slc ; GFX10NSA: bb.1.main_body: @@ -988,8 +988,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32) ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.add.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 2) @@ -1017,8 +1017,8 @@ ; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32) ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2d), [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2d), [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_cmpswap_2d ; GFX10NSA: bb.1.main_body: @@ -1039,8 +1039,8 @@ ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32) ; GFX10NSA: [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2d), [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2d), [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.cmpswap.2d.i32.i16(i32 %cmp, i32 %swap, i16 %s, i16 %t, <8 x i32> %rsrc, i32 0, i32 0) @@ -1073,8 +1073,8 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.cmpswap.3d), [[COPY8]](s32), [[COPY9]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.3d), [[COPY8]](s32), [[COPY9]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_cmpswap_3d ; GFX10NSA: bb.1.main_body: @@ -1099,8 +1099,8 @@ ; GFX10NSA: [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.cmpswap.3d), [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.3d), [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.cmpswap.3d.i32.i16(i32 %cmp, i32 %swap, i16 %s, i16 %t, i16 %r, <8 x i32> %rsrc, i32 0, i32 0) @@ -1134,8 +1134,8 @@ ; GFX9: [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2darraymsaa), [[COPY8]](s32), [[COPY9]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2darraymsaa), [[COPY8]](s32), [[COPY9]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: atomic_cmpswap_2darraymsaa ; GFX10NSA: bb.1.main_body: @@ -1161,8 +1161,8 @@ ; GFX10NSA: [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32) ; GFX10NSA: [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2darraymsaa), [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2darraymsaa), [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %v = call i32 @llvm.amdgcn.image.atomic.cmpswap.2darraymsaa.i32.i16(i32 %cmp, i32 %swap, i16 %s, i16 %t, i16 %slice, i16 %fragid, <8 x i32> %rsrc, i32 0, i32 0) Index: llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.dim.a16.ll =================================================================== --- llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.dim.a16.ll +++ llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.dim.a16.ll @@ -18,8 +18,8 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -40,8 +40,8 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -74,8 +74,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -101,8 +101,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; GFX10NSA: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -142,8 +142,8 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -174,8 +174,8 @@ ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.3d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -216,8 +216,8 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.cube), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.cube), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -248,8 +248,8 @@ ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.cube), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.cube), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -284,8 +284,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -311,8 +311,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; GFX10NSA: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -352,8 +352,8 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -384,8 +384,8 @@ ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -426,8 +426,8 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2dmsaa), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2dmsaa), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -458,8 +458,8 @@ ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2dmsaa), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2dmsaa), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -502,8 +502,8 @@ ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -536,8 +536,8 @@ ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -573,8 +573,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.mip.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -600,8 +600,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; GFX10NSA: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.mip.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -641,8 +641,8 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.mip.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -673,8 +673,8 @@ ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.mip.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -717,8 +717,8 @@ ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.mip.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -751,8 +751,8 @@ ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.mip.3d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.3d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -796,8 +796,8 @@ ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.mip.cube), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.cube), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -830,8 +830,8 @@ ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.mip.cube), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.cube), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -873,8 +873,8 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.mip.1darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -905,8 +905,8 @@ ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.mip.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -949,8 +949,8 @@ ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.mip.2darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -983,8 +983,8 @@ ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.mip.2darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -1020,7 +1020,7 @@ ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_1d ; GFX10NSA: bb.1.main_body: @@ -1042,7 +1042,7 @@ ; GFX10NSA: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords, i32 0 @@ -1076,7 +1076,7 @@ ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; GFX9: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_2d ; GFX10NSA: bb.1.main_body: @@ -1103,7 +1103,7 @@ ; GFX10NSA: [[COPY13:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; GFX10NSA: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords, i32 0 @@ -1144,7 +1144,7 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_3d ; GFX10NSA: bb.1.main_body: @@ -1176,7 +1176,7 @@ ; GFX10NSA: [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords_lo, i32 0 @@ -1218,7 +1218,7 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_cube ; GFX10NSA: bb.1.main_body: @@ -1250,7 +1250,7 @@ ; GFX10NSA: [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords_lo, i32 0 @@ -1286,7 +1286,7 @@ ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; GFX9: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_1darray ; GFX10NSA: bb.1.main_body: @@ -1313,7 +1313,7 @@ ; GFX10NSA: [[COPY13:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; GFX10NSA: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords, i32 0 @@ -1354,7 +1354,7 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_2darray ; GFX10NSA: bb.1.main_body: @@ -1386,7 +1386,7 @@ ; GFX10NSA: [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords_lo, i32 0 @@ -1428,7 +1428,7 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2dmsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2dmsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_2dmsaa ; GFX10NSA: bb.1.main_body: @@ -1460,7 +1460,7 @@ ; GFX10NSA: [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2dmsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2dmsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords_lo, i32 0 @@ -1504,7 +1504,7 @@ ; GFX9: [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2darraymsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2darraymsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_2darraymsaa ; GFX10NSA: bb.1.main_body: @@ -1538,7 +1538,7 @@ ; GFX10NSA: [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2darraymsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2darraymsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords_lo, i32 0 @@ -1575,7 +1575,7 @@ ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; GFX9: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.mip.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_mip_1d ; GFX10NSA: bb.1.main_body: @@ -1602,7 +1602,7 @@ ; GFX10NSA: [[COPY13:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; GFX10NSA: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.mip.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords, i32 0 @@ -1643,7 +1643,7 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.mip.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_mip_2d ; GFX10NSA: bb.1.main_body: @@ -1675,7 +1675,7 @@ ; GFX10NSA: [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.mip.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords_lo, i32 0 @@ -1719,7 +1719,7 @@ ; GFX9: [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.mip.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_mip_3d ; GFX10NSA: bb.1.main_body: @@ -1753,7 +1753,7 @@ ; GFX10NSA: [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.mip.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords_lo, i32 0 @@ -1798,7 +1798,7 @@ ; GFX9: [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.mip.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_mip_cube ; GFX10NSA: bb.1.main_body: @@ -1832,7 +1832,7 @@ ; GFX10NSA: [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.mip.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords_lo, i32 0 @@ -1875,7 +1875,7 @@ ; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.mip.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_mip_1darray ; GFX10NSA: bb.1.main_body: @@ -1907,7 +1907,7 @@ ; GFX10NSA: [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.mip.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords_lo, i32 0 @@ -1951,7 +1951,7 @@ ; GFX9: [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.mip.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_mip_2darray ; GFX10NSA: bb.1.main_body: @@ -1985,7 +1985,7 @@ ; GFX10NSA: [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.mip.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords_lo, i32 0 @@ -2012,8 +2012,8 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -2034,8 +2034,8 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -2063,8 +2063,8 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.2d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -2085,8 +2085,8 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.2d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -2114,8 +2114,8 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.3d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.3d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -2136,8 +2136,8 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.3d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.3d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -2165,8 +2165,8 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.cube), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.cube), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -2187,8 +2187,8 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.cube), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.cube), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -2216,8 +2216,8 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.1darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -2238,8 +2238,8 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.1darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -2267,8 +2267,8 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.2darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -2289,8 +2289,8 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.2darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -2318,8 +2318,8 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.2dmsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2dmsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -2340,8 +2340,8 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.2dmsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2dmsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -2369,8 +2369,8 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.2darraymsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2darraymsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -2391,8 +2391,8 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.2darraymsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2darraymsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -2420,8 +2420,8 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 8, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8") - ; GFX9: $vgpr0 = COPY [[INT]](s32) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 8, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8") + ; GFX9: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0 ; GFX10NSA-LABEL: name: load_1d_V1 ; GFX10NSA: bb.1.main_body: @@ -2438,8 +2438,8 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 8, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8") - ; GFX10NSA: $vgpr0 = COPY [[INT]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 8, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 main_body: %s = extractelement <2 x i16> %coords, i32 0 @@ -2463,8 +2463,8 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<2 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 9, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<2 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 9, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1 @@ -2483,8 +2483,8 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<2 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 9, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<2 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 9, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1 @@ -2511,7 +2511,7 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.1d), [[COPY8]](s32), 2, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 4 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[COPY8]](s32), 2, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 4 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_1d_V1 ; GFX10NSA: bb.1.main_body: @@ -2529,7 +2529,7 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.1d), [[COPY8]](s32), 2, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 4 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[COPY8]](s32), 2, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 4 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords, i32 0 @@ -2556,7 +2556,7 @@ ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY10]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<2 x s32>), 12, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 8 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<2 x s32>), 12, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 8 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_1d_V2 ; GFX10NSA: bb.1.main_body: @@ -2576,7 +2576,7 @@ ; GFX10NSA: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY10]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<2 x s32>), 12, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 8 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<2 x s32>), 12, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 8 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords, i32 0 @@ -2600,8 +2600,8 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -2622,8 +2622,8 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -2651,8 +2651,8 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -2673,8 +2673,8 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -2702,8 +2702,8 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -2724,8 +2724,8 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -2758,7 +2758,7 @@ ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_1d_glc ; GFX10NSA: bb.1.main_body: @@ -2780,7 +2780,7 @@ ; GFX10NSA: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords, i32 0 @@ -2809,7 +2809,7 @@ ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_1d_slc ; GFX10NSA: bb.1.main_body: @@ -2831,7 +2831,7 @@ ; GFX10NSA: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords, i32 0 @@ -2860,7 +2860,7 @@ ; GFX9: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX9: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX9: S_ENDPGM 0 ; GFX10NSA-LABEL: name: store_1d_glc_slc ; GFX10NSA: bb.1.main_body: @@ -2882,7 +2882,7 @@ ; GFX10NSA: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8") + ; GFX10NSA: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8") ; GFX10NSA: S_ENDPGM 0 main_body: %s = extractelement <2 x i16> %coords, i32 0 @@ -2906,8 +2906,8 @@ ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.1d), 0, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1d), 0, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) ; GFX9: $vgpr2 = COPY [[UV2]](s32) @@ -2928,8 +2928,8 @@ ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.image.getresinfo.1d), 0, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1d), 0, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) @@ -2958,8 +2958,8 @@ ; GFX9: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<5 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<5 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>) ; GFX9: G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) @@ -2982,8 +2982,8 @@ ; GFX10NSA: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF ; GFX10NSA: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>) ; GFX10NSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<5 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<5 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>) ; GFX10NSA: G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) @@ -3021,8 +3021,8 @@ ; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32) - ; GFX9: [[INT:%[0-9]+]]:_(<5 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<5 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>) ; GFX9: G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) @@ -3050,8 +3050,8 @@ ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; GFX10NSA: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<5 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<5 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>) ; GFX10NSA: G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) @@ -3096,8 +3096,8 @@ ; GFX9: [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF1]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(<5 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<5 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>) ; GFX9: G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) @@ -3130,8 +3130,8 @@ ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF1]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<5 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.3d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<5 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>) ; GFX10NSA: G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) @@ -3179,8 +3179,8 @@ ; GFX9: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>) - ; GFX9: [[INT:%[0-9]+]]:_(<5 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<5 x s32>) + ; GFX9: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>) ; GFX9: G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; GFX9: $vgpr0 = COPY [[UV]](s32) ; GFX9: $vgpr1 = COPY [[UV1]](s32) @@ -3215,8 +3215,8 @@ ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) ; GFX10NSA: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; GFX10NSA: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32) - ; GFX10NSA: [[INT:%[0-9]+]]:_(<5 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<5 x s32>) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>) ; GFX10NSA: G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) Index: llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.2d.d16.ll =================================================================== --- llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.2d.d16.ll +++ llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.2d.d16.ll @@ -17,8 +17,9 @@ ; UNPACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 ; UNPACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; UNPACKED: [[INT:%[0-9]+]]:_(s16) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 1, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8") - ; UNPACKED: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[INT]](s16) + ; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s16) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8") + ; UNPACKED: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[AMDGPU_INTRIN_IMAGE_LOAD]](s16) ; UNPACKED: $vgpr0 = COPY [[ANYEXT]](s32) ; UNPACKED: SI_RETURN_TO_EPILOG implicit $vgpr0 ; PACKED-LABEL: name: image_load_f16 @@ -35,8 +36,9 @@ ; PACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 ; PACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; PACKED: [[INT:%[0-9]+]]:_(s16) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 1, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8") - ; PACKED: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[INT]](s16) + ; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s16) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8") + ; PACKED: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[AMDGPU_INTRIN_IMAGE_LOAD]](s16) ; PACKED: $vgpr0 = COPY [[ANYEXT]](s32) ; PACKED: SI_RETURN_TO_EPILOG implicit $vgpr0 %tex = call half @llvm.amdgcn.image.load.2d.f16.i32(i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0) @@ -58,8 +60,9 @@ ; UNPACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 ; UNPACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; UNPACKED: [[INT:%[0-9]+]]:_(<2 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 3, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8") - ; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<2 x s32>) + ; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8") + ; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>) ; UNPACKED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535 ; UNPACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32) ; UNPACKED: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]] @@ -85,8 +88,9 @@ ; PACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 ; PACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; PACKED: [[INT:%[0-9]+]]:_(<2 x s16>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 3, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8") - ; PACKED: $vgpr0 = COPY [[INT]](<2 x s16>) + ; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8") + ; PACKED: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s16>) ; PACKED: SI_RETURN_TO_EPILOG implicit $vgpr0 %tex = call <2 x half> @llvm.amdgcn.image.load.2d.v2f16.i32(i32 3, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0) ret <2 x half> %tex @@ -107,8 +111,9 @@ ; UNPACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 ; UNPACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; UNPACKED: [[INT:%[0-9]+]]:_(<3 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 7, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8) - ; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<3 x s32>) + ; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8) + ; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>) ; UNPACKED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535 ; UNPACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32) ; UNPACKED: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]] @@ -147,9 +152,10 @@ ; PACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 ; PACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; PACKED: [[INT:%[0-9]+]]:_(<3 x s16>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 7, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8) + ; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8) ; PACKED: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF - ; PACKED: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[INT]](<3 x s16>), 0 + ; PACKED: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s16>), 0 ; PACKED: [[EXTRACT:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 0 ; PACKED: [[EXTRACT1:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 32 ; PACKED: $vgpr0 = COPY [[EXTRACT]](<2 x s16>) @@ -174,8 +180,9 @@ ; UNPACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 ; UNPACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; UNPACKED: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 15, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8") - ; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8") + ; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; UNPACKED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535 ; UNPACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32) ; UNPACKED: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]] @@ -209,8 +216,9 @@ ; PACKED: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 ; PACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; PACKED: [[INT:%[0-9]+]]:_(<4 x s16>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 15, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8") - ; PACKED: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INT]](<4 x s16>) + ; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8") + ; PACKED: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s16>) ; PACKED: $vgpr0 = COPY [[UV]](<2 x s16>) ; PACKED: $vgpr1 = COPY [[UV1]](<2 x s16>) ; PACKED: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1 @@ -234,8 +242,9 @@ ; UNPACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; UNPACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF - ; UNPACKED: [[INT:%[0-9]+]]:_(<2 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 1, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8") - ; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<2 x s32>) + ; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8") + ; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>) ; UNPACKED: G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; UNPACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32) ; UNPACKED: $vgpr0 = COPY [[COPY10]](s32) @@ -255,8 +264,9 @@ ; PACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; PACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF - ; PACKED: [[INT:%[0-9]+]]:_(<2 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 1, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8") - ; PACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<2 x s32>) + ; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8") + ; PACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>) ; PACKED: G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; PACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32) ; PACKED: $vgpr0 = COPY [[COPY10]](s32) @@ -284,8 +294,9 @@ ; UNPACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; UNPACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF - ; UNPACKED: [[INT:%[0-9]+]]:_(<3 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 3, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8") - ; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<3 x s32>) + ; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8") + ; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>) ; UNPACKED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535 ; UNPACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32) ; UNPACKED: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]] @@ -313,8 +324,9 @@ ; PACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; PACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF - ; PACKED: [[INT:%[0-9]+]]:_(<2 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 3, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8") - ; PACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<2 x s32>) + ; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8") + ; PACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>) ; PACKED: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32) ; PACKED: G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; PACKED: $vgpr0 = COPY [[BITCAST]](<2 x s16>) @@ -342,8 +354,9 @@ ; UNPACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; UNPACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF - ; UNPACKED: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 7, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8) - ; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8) + ; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; UNPACKED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535 ; UNPACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32) ; UNPACKED: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]] @@ -384,8 +397,9 @@ ; PACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; PACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF - ; PACKED: [[INT:%[0-9]+]]:_(<3 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 7, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8) - ; PACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<3 x s32>) + ; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8) + ; PACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>) ; PACKED: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32) ; PACKED: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV1]](s32) ; PACKED: [[DEF1:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF @@ -422,8 +436,9 @@ ; UNPACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; UNPACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF - ; UNPACKED: [[INT:%[0-9]+]]:_(<5 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 15, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8") - ; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<5 x s32>) + ; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; UNPACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8") + ; UNPACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>) ; UNPACKED: G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; UNPACKED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535 ; UNPACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32) @@ -459,8 +474,9 @@ ; PACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; PACKED: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF - ; PACKED: [[INT:%[0-9]+]]:_(<3 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 15, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8") - ; PACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<3 x s32>) + ; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; PACKED: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8") + ; PACKED: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>) ; PACKED: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32) ; PACKED: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV1]](s32) ; PACKED: G_STORE [[UV2]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) Index: llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.2d.ll =================================================================== --- llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.2d.ll +++ llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.2d.ll @@ -16,8 +16,9 @@ ; GCN: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 ; GCN: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GCN: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 1, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8") - ; GCN: $vgpr0 = COPY [[INT]](s32) + ; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8") + ; GCN: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) ; GCN: SI_RETURN_TO_EPILOG implicit $vgpr0 %tex = call float @llvm.amdgcn.image.load.2d.f32.i32(i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0) ret float %tex @@ -38,8 +39,9 @@ ; GCN: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 ; GCN: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GCN: [[INT:%[0-9]+]]:_(<2 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 3, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8") - ; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<2 x s32>) + ; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8") + ; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>) ; GCN: $vgpr0 = COPY [[UV]](s32) ; GCN: $vgpr1 = COPY [[UV1]](s32) ; GCN: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1 @@ -62,8 +64,9 @@ ; GCN: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 ; GCN: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GCN: [[INT:%[0-9]+]]:_(<3 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 7, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 12 from custom "TargetCustom8", align 16) - ; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<3 x s32>) + ; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 12 from custom "TargetCustom8", align 16) + ; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>) ; GCN: $vgpr0 = COPY [[UV]](s32) ; GCN: $vgpr1 = COPY [[UV1]](s32) ; GCN: $vgpr2 = COPY [[UV2]](s32) @@ -87,8 +90,9 @@ ; GCN: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 ; GCN: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; GCN: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 15, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GCN: $vgpr0 = COPY [[UV]](s32) ; GCN: $vgpr1 = COPY [[UV1]](s32) ; GCN: $vgpr2 = COPY [[UV2]](s32) @@ -114,8 +118,9 @@ ; GCN: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GCN: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF - ; GCN: [[INT:%[0-9]+]]:_(<2 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 1, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8") - ; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<2 x s32>) + ; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8") + ; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>) ; GCN: G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; GCN: $vgpr0 = COPY [[UV]](s32) ; GCN: SI_RETURN_TO_EPILOG implicit $vgpr0 @@ -142,8 +147,9 @@ ; GCN: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GCN: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF - ; GCN: [[INT:%[0-9]+]]:_(<3 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 3, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8") - ; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<3 x s32>) + ; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8") + ; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>) ; GCN: G_STORE [[UV2]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; GCN: $vgpr0 = COPY [[UV]](s32) ; GCN: $vgpr1 = COPY [[UV1]](s32) @@ -171,8 +177,9 @@ ; GCN: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GCN: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF - ; GCN: [[INT:%[0-9]+]]:_(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 7, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 12 from custom "TargetCustom8", align 16) - ; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<4 x s32>) + ; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 12 from custom "TargetCustom8", align 16) + ; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) ; GCN: G_STORE [[UV3]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; GCN: $vgpr0 = COPY [[UV]](s32) ; GCN: $vgpr1 = COPY [[UV1]](s32) @@ -201,8 +208,9 @@ ; GCN: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GCN: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; GCN: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF - ; GCN: [[INT:%[0-9]+]]:_(<5 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.2d), 15, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](<5 x s32>) + ; GCN: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; GCN: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GCN: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>) ; GCN: G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) ; GCN: $vgpr0 = COPY [[UV]](s32) ; GCN: $vgpr1 = COPY [[UV1]](s32) Index: llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.2darraymsaa.ll =================================================================== --- /dev/null +++ llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.2darraymsaa.ll @@ -0,0 +1,123 @@ +; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py +; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -stop-after=legalizer -o - %s | FileCheck -check-prefix=GFX6 %s +; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -stop-after=legalizer -o - %s | FileCheck -check-prefix=GFX10NSA %s + +define amdgpu_ps <4 x float> @load_2darraymsaa(<8 x i32> inreg %rsrc, i32 %s, i32 %t, i32 %slice, i32 %fragid) { + ; GFX6-LABEL: name: load_2darraymsaa + ; GFX6: bb.1 (%ir-block.0): + ; GFX6: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX6: [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2 + ; GFX6: [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3 + ; GFX6: [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4 + ; GFX6: [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5 + ; GFX6: [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6 + ; GFX6: [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7 + ; GFX6: [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8 + ; GFX6: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9 + ; GFX6: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX6: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX6: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2 + ; GFX6: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3 + ; GFX6: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) + ; GFX6: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32) + ; GFX6: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[BUILD_VECTOR1]](<4 x s32>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX6: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) + ; GFX6: $vgpr0 = COPY [[UV]](s32) + ; GFX6: $vgpr1 = COPY [[UV1]](s32) + ; GFX6: $vgpr2 = COPY [[UV2]](s32) + ; GFX6: $vgpr3 = COPY [[UV3]](s32) + ; GFX6: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3 + ; GFX10NSA-LABEL: name: load_2darraymsaa + ; GFX10NSA: bb.1 (%ir-block.0): + ; GFX10NSA: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10NSA: [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2 + ; GFX10NSA: [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3 + ; GFX10NSA: [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4 + ; GFX10NSA: [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5 + ; GFX10NSA: [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6 + ; GFX10NSA: [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7 + ; GFX10NSA: [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8 + ; GFX10NSA: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9 + ; GFX10NSA: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX10NSA: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2 + ; GFX10NSA: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3 + ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>) + ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) + ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) + ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) + ; GFX10NSA: $vgpr3 = COPY [[UV3]](s32) + ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3 + %v = call <4 x float> @llvm.amdgcn.image.load.2darraymsaa.v4f32.i32(i32 15, i32 %s, i32 %t, i32 %slice, i32 %fragid, <8 x i32> %rsrc, i32 0, i32 0) + ret <4 x float> %v +} + +define amdgpu_ps <4 x float> @load_2darraymsaa_tfe(<8 x i32> inreg %rsrc, i32 addrspace(1)* inreg %out, i32 %s, i32 %t, i32 %slice, i32 %fragid) { + ; GFX6-LABEL: name: load_2darraymsaa_tfe + ; GFX6: bb.1 (%ir-block.0): + ; GFX6: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX6: [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2 + ; GFX6: [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3 + ; GFX6: [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4 + ; GFX6: [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5 + ; GFX6: [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6 + ; GFX6: [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7 + ; GFX6: [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8 + ; GFX6: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9 + ; GFX6: [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10 + ; GFX6: [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11 + ; GFX6: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX6: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX6: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr2 + ; GFX6: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr3 + ; GFX6: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) + ; GFX6: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32) + ; GFX6: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY10]](s32), [[COPY11]](s32), [[COPY12]](s32), [[COPY13]](s32) + ; GFX6: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[BUILD_VECTOR1]](<4 x s32>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX6: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>) + ; GFX6: G_STORE [[UV4]](s32), [[MV]](p1) :: (store 4 into %ir.out, addrspace 1) + ; GFX6: $vgpr0 = COPY [[UV]](s32) + ; GFX6: $vgpr1 = COPY [[UV1]](s32) + ; GFX6: $vgpr2 = COPY [[UV2]](s32) + ; GFX6: $vgpr3 = COPY [[UV3]](s32) + ; GFX6: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3 + ; GFX10NSA-LABEL: name: load_2darraymsaa_tfe + ; GFX10NSA: bb.1 (%ir-block.0): + ; GFX10NSA: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $vgpr0, $vgpr1, $vgpr2, $vgpr3 + ; GFX10NSA: [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2 + ; GFX10NSA: [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3 + ; GFX10NSA: [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4 + ; GFX10NSA: [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5 + ; GFX10NSA: [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6 + ; GFX10NSA: [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7 + ; GFX10NSA: [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8 + ; GFX10NSA: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9 + ; GFX10NSA: [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10 + ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11 + ; GFX10NSA: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX10NSA: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX10NSA: [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr2 + ; GFX10NSA: [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr3 + ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) + ; GFX10NSA: [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[COPY10]](s32), [[COPY11]](s32), [[COPY12]](s32), [[COPY13]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>) + ; GFX10NSA: G_STORE [[UV4]](s32), [[MV]](p1) :: (store 4 into %ir.out, addrspace 1) + ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) + ; GFX10NSA: $vgpr1 = COPY [[UV1]](s32) + ; GFX10NSA: $vgpr2 = COPY [[UV2]](s32) + ; GFX10NSA: $vgpr3 = COPY [[UV3]](s32) + ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3 + %v = call { <4 x float>, i32 } @llvm.amdgcn.image.load.2darraymsaa.sl_v4f32i32s.i32(i32 15, i32 %s, i32 %t, i32 %slice, i32 %fragid, <8 x i32> %rsrc, i32 1, i32 0) + %v.vec = extractvalue { <4 x float>, i32 } %v, 0 + %v.err = extractvalue { <4 x float>, i32 } %v, 1 + store i32 %v.err, i32 addrspace(1)* %out, align 4 + ret <4 x float> %v.vec +} + +declare <4 x float> @llvm.amdgcn.image.load.2darraymsaa.v4f32.i32(i32 immarg, i32, i32, i32, i32, <8 x i32>, i32 immarg, i32 immarg) #0 +declare { <4 x float>, i32 } @llvm.amdgcn.image.load.2darraymsaa.sl_v4f32i32s.i32(i32 immarg, i32, i32, i32, i32, <8 x i32>, i32 immarg, i32 immarg) #0 + +attributes #0 = { nounwind readonly } Index: llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.3d.ll =================================================================== --- /dev/null +++ llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.3d.ll @@ -0,0 +1,101 @@ +; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py +; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -stop-after=legalizer -o - %s | FileCheck -check-prefix=GFX6 %s +; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -stop-after=legalizer -o - %s | FileCheck -check-prefix=GFX10NSA %s + +define amdgpu_ps float @image_load_3d_f32(<8 x i32> inreg %rsrc, i32 %s, i32 %t, i32 %r) { + ; GFX6-LABEL: name: image_load_3d_f32 + ; GFX6: bb.1 (%ir-block.0): + ; GFX6: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2 + ; GFX6: [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2 + ; GFX6: [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3 + ; GFX6: [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4 + ; GFX6: [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5 + ; GFX6: [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6 + ; GFX6: [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7 + ; GFX6: [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8 + ; GFX6: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9 + ; GFX6: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX6: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX6: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2 + ; GFX6: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) + ; GFX6: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32) + ; GFX6: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 1, [[BUILD_VECTOR1]](<3 x s32>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8") + ; GFX6: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) + ; GFX6: SI_RETURN_TO_EPILOG implicit $vgpr0 + ; GFX10NSA-LABEL: name: image_load_3d_f32 + ; GFX10NSA: bb.1 (%ir-block.0): + ; GFX10NSA: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2 + ; GFX10NSA: [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2 + ; GFX10NSA: [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3 + ; GFX10NSA: [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4 + ; GFX10NSA: [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5 + ; GFX10NSA: [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6 + ; GFX10NSA: [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7 + ; GFX10NSA: [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8 + ; GFX10NSA: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9 + ; GFX10NSA: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX10NSA: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2 + ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 1, [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8") + ; GFX10NSA: $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32) + ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 + %tex = call float @llvm.amdgcn.image.load.3d.f32.i32(i32 1, i32 %s, i32 %t, i32 %r, <8 x i32> %rsrc, i32 0, i32 0) + ret float %tex +} + +define amdgpu_ps float @image_load_3d_tfe_f32(<8 x i32> inreg %rsrc, i32 %s, i32 %t, i32 %r) { + ; GFX6-LABEL: name: image_load_3d_tfe_f32 + ; GFX6: bb.1 (%ir-block.0): + ; GFX6: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2 + ; GFX6: [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2 + ; GFX6: [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3 + ; GFX6: [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4 + ; GFX6: [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5 + ; GFX6: [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6 + ; GFX6: [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7 + ; GFX6: [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8 + ; GFX6: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9 + ; GFX6: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX6: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX6: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2 + ; GFX6: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) + ; GFX6: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF + ; GFX6: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32) + ; GFX6: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 1, [[BUILD_VECTOR1]](<3 x s32>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8") + ; GFX6: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>) + ; GFX6: G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) + ; GFX6: $vgpr0 = COPY [[UV]](s32) + ; GFX6: SI_RETURN_TO_EPILOG implicit $vgpr0 + ; GFX10NSA-LABEL: name: image_load_3d_tfe_f32 + ; GFX10NSA: bb.1 (%ir-block.0): + ; GFX10NSA: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2 + ; GFX10NSA: [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2 + ; GFX10NSA: [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3 + ; GFX10NSA: [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4 + ; GFX10NSA: [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5 + ; GFX10NSA: [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6 + ; GFX10NSA: [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7 + ; GFX10NSA: [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8 + ; GFX10NSA: [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9 + ; GFX10NSA: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0 + ; GFX10NSA: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 + ; GFX10NSA: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2 + ; GFX10NSA: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) + ; GFX10NSA: [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF + ; GFX10NSA: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 1, [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8") + ; GFX10NSA: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>) + ; GFX10NSA: G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1) + ; GFX10NSA: $vgpr0 = COPY [[UV]](s32) + ; GFX10NSA: SI_RETURN_TO_EPILOG implicit $vgpr0 + %val = call { float, i32 } @llvm.amdgcn.image.load.3d.sl_f32i32s.i32(i32 1, i32 %s, i32 %t, i32 %r, <8 x i32> %rsrc, i32 1, i32 0) + %tex = extractvalue { float, i32 } %val, 0 + %tfe = extractvalue { float, i32 } %val, 1 + store i32 %tfe, i32 addrspace(1)* undef + ret float %tex +} + +declare float @llvm.amdgcn.image.load.3d.f32.i32(i32 immarg, i32, i32, i32, <8 x i32>, i32 immarg, i32 immarg) #0 +declare { float, i32 } @llvm.amdgcn.image.load.3d.sl_f32i32s.i32(i32 immarg, i32, i32, i32, <8 x i32>, i32 immarg, i32 immarg) #0 + +attributes #0 = { nounwind readonly } Index: llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.store.2d.d16.ll =================================================================== --- llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.store.2d.d16.ll +++ llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.store.2d.d16.ll @@ -19,7 +19,8 @@ ; UNPACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2 ; UNPACKED: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY10]](s32) ; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; UNPACKED: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2d), [[TRUNC]](s16), 1, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 2 into custom "TargetCustom8") + ; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; UNPACKED: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[TRUNC]](s16), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 2 into custom "TargetCustom8") ; UNPACKED: S_ENDPGM 0 ; PACKED-LABEL: name: image_store_f16 ; PACKED: bb.1 (%ir-block.0): @@ -37,7 +38,8 @@ ; PACKED: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2 ; PACKED: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY10]](s32) ; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; PACKED: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2d), [[TRUNC]](s16), 1, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 2 into custom "TargetCustom8") + ; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; PACKED: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[TRUNC]](s16), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 2 into custom "TargetCustom8") ; PACKED: S_ENDPGM 0 call void @llvm.amdgcn.image.store.2d.f16.i32(half %data, i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0) ret void @@ -59,13 +61,14 @@ ; UNPACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; UNPACKED: [[COPY10:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2 ; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) + ; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) ; UNPACKED: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY10]](<2 x s16>) ; UNPACKED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 ; UNPACKED: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) ; UNPACKED: [[COPY11:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; UNPACKED: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) - ; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY11]](s32), [[COPY12]](s32) - ; UNPACKED: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR1]](<2 x s32>), 3, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 4 into custom "TargetCustom8") + ; UNPACKED: [[BUILD_VECTOR2:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY11]](s32), [[COPY12]](s32) + ; UNPACKED: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR2]](<2 x s32>), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 4 into custom "TargetCustom8") ; UNPACKED: S_ENDPGM 0 ; PACKED-LABEL: name: image_store_v2f16 ; PACKED: bb.1 (%ir-block.0): @@ -82,7 +85,8 @@ ; PACKED: [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1 ; PACKED: [[COPY10:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2 ; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) - ; PACKED: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2d), [[COPY10]](<2 x s16>), 3, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 4 into custom "TargetCustom8") + ; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; PACKED: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[COPY10]](<2 x s16>), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 4 into custom "TargetCustom8") ; PACKED: S_ENDPGM 0 call void @llvm.amdgcn.image.store.2d.v2f16.i32(<2 x half> %in, i32 3, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0) ret void @@ -107,6 +111,7 @@ ; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; UNPACKED: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY10]](<2 x s16>), [[COPY11]](<2 x s16>) ; UNPACKED: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[CONCAT_VECTORS]](<4 x s16>), 0 + ; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) ; UNPACKED: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; UNPACKED: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT]](<3 x s16>), 0 ; UNPACKED: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) @@ -118,8 +123,8 @@ ; UNPACKED: [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) ; UNPACKED: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; UNPACKED: [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32) - ; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY12]](s32), [[COPY13]](s32), [[COPY14]](s32) - ; UNPACKED: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR1]](<3 x s32>), 7, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 6 into custom "TargetCustom8", align 8) + ; UNPACKED: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY12]](s32), [[COPY13]](s32), [[COPY14]](s32) + ; UNPACKED: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR2]](<3 x s32>), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 6 into custom "TargetCustom8", align 8) ; UNPACKED: S_ENDPGM 0 ; PACKED-LABEL: name: image_store_v3f16 ; PACKED: bb.1 (%ir-block.0): @@ -139,7 +144,8 @@ ; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; PACKED: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY10]](<2 x s16>), [[COPY11]](<2 x s16>) ; PACKED: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[CONCAT_VECTORS]](<4 x s16>), 0 - ; PACKED: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2d), [[EXTRACT]](<3 x s16>), 7, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 6 into custom "TargetCustom8", align 8) + ; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; PACKED: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[EXTRACT]](<3 x s16>), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 6 into custom "TargetCustom8", align 8) ; PACKED: S_ENDPGM 0 call void @llvm.amdgcn.image.store.2d.v3f16.i32(<3 x half> %in, i32 7, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0) ret void @@ -162,6 +168,7 @@ ; UNPACKED: [[COPY10:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2 ; UNPACKED: [[COPY11:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3 ; UNPACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) + ; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) ; UNPACKED: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY10]](<2 x s16>) ; UNPACKED: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 ; UNPACKED: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) @@ -171,8 +178,8 @@ ; UNPACKED: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; UNPACKED: [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32) ; UNPACKED: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) - ; UNPACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY12]](s32), [[COPY13]](s32), [[COPY14]](s32), [[COPY15]](s32) - ; UNPACKED: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 8 into custom "TargetCustom8") + ; UNPACKED: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY12]](s32), [[COPY13]](s32), [[COPY14]](s32), [[COPY15]](s32) + ; UNPACKED: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR2]](<4 x s32>), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 8 into custom "TargetCustom8") ; UNPACKED: S_ENDPGM 0 ; PACKED-LABEL: name: image_store_v4f16 ; PACKED: bb.1 (%ir-block.0): @@ -191,7 +198,8 @@ ; PACKED: [[COPY11:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3 ; PACKED: [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; PACKED: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY10]](<2 x s16>), [[COPY11]](<2 x s16>) - ; PACKED: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.store.2d), [[CONCAT_VECTORS]](<4 x s16>), 15, [[COPY8]](s32), [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 8 into custom "TargetCustom8") + ; PACKED: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32) + ; PACKED: G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[CONCAT_VECTORS]](<4 x s16>), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 8 into custom "TargetCustom8") ; PACKED: S_ENDPGM 0 call void @llvm.amdgcn.image.store.2d.v4f16.i32(<4 x half> %in, i32 15, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0) ret void Index: llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.image.load.1d.ll =================================================================== --- llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.image.load.1d.ll +++ llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.image.load.1d.ll @@ -18,9 +18,9 @@ ; CHECK: [[COPY8:%[0-9]+]]:vgpr(s32) = COPY $vgpr0 ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; CHECK: [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF - ; CHECK: [[INT:%[0-9]+]]:vgpr(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY8]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; CHECK: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY8]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") ; CHECK: [[COPY9:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1) - ; CHECK: G_STORE [[INT]](<4 x s32>), [[COPY9]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) + ; CHECK: G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY9]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) ; CHECK: S_ENDPGM 0 %v = call <4 x float> @llvm.amdgcn.image.load.1d.v4f32.i32(i32 15, i32 %s, <8 x i32> %rsrc, i32 0, i32 0) store <4 x float> %v, <4 x float> addrspace(1)* undef @@ -44,9 +44,9 @@ ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; CHECK: [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF ; CHECK: [[COPY9:%[0-9]+]]:vgpr(s32) = COPY [[COPY8]](s32) - ; CHECK: [[INT:%[0-9]+]]:vgpr(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; CHECK: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") ; CHECK: [[COPY10:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1) - ; CHECK: G_STORE [[INT]](<4 x s32>), [[COPY10]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) + ; CHECK: G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY10]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) ; CHECK: S_ENDPGM 0 %v = call <4 x float> @llvm.amdgcn.image.load.1d.v4f32.i32(i32 15, i32 %s, <8 x i32> %rsrc, i32 0, i32 0) store <4 x float> %v, <4 x float> addrspace(1)* undef @@ -98,7 +98,7 @@ ; CHECK: [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV3]](s64), [[UV3]](s64), implicit $exec ; CHECK: [[S_AND_B64_2:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_3]], [[S_AND_B64_1]], implicit-def $scc ; CHECK: [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32) - ; CHECK: [[INT:%[0-9]+]]:vgpr(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY8]](s32), [[BUILD_VECTOR1]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; CHECK: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY8]](s32), [[BUILD_VECTOR1]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") ; CHECK: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_2]], implicit-def $exec, implicit-def $scc, implicit $exec ; CHECK: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc ; CHECK: S_CBRANCH_EXECNZ %bb.2, implicit $exec @@ -107,7 +107,7 @@ ; CHECK: $exec = S_MOV_B64_term [[S_MOV_B64_term]] ; CHECK: bb.4: ; CHECK: [[COPY9:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1) - ; CHECK: G_STORE [[INT]](<4 x s32>), [[COPY9]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) + ; CHECK: G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY9]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) ; CHECK: S_ENDPGM 0 %v = call <4 x float> @llvm.amdgcn.image.load.1d.v4f32.i32(i32 15, i32 %s, <8 x i32> %rsrc, i32 0, i32 0) store <4 x float> %v, <4 x float> addrspace(1)* undef @@ -160,7 +160,7 @@ ; CHECK: [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV3]](s64), [[UV3]](s64), implicit $exec ; CHECK: [[S_AND_B64_2:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_3]], [[S_AND_B64_1]], implicit-def $scc ; CHECK: [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32) - ; CHECK: [[INT:%[0-9]+]]:vgpr(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY9]](s32), [[BUILD_VECTOR1]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; CHECK: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY9]](s32), [[BUILD_VECTOR1]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") ; CHECK: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_2]], implicit-def $exec, implicit-def $scc, implicit $exec ; CHECK: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc ; CHECK: S_CBRANCH_EXECNZ %bb.2, implicit $exec @@ -169,7 +169,7 @@ ; CHECK: $exec = S_MOV_B64_term [[S_MOV_B64_term]] ; CHECK: bb.4: ; CHECK: [[COPY10:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1) - ; CHECK: G_STORE [[INT]](<4 x s32>), [[COPY10]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) + ; CHECK: G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY10]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) ; CHECK: S_ENDPGM 0 %v = call <4 x float> @llvm.amdgcn.image.load.1d.v4f32.i32(i32 15, i32 %s, <8 x i32> %rsrc, i32 0, i32 0) store <4 x float> %v, <4 x float> addrspace(1)* undef Index: llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.image.sample.1d.ll =================================================================== --- llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.image.sample.1d.ll +++ llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.image.sample.1d.ll @@ -1,6 +1,6 @@ ; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -stop-after=regbankselect -regbankselect-fast -o - %s | FileCheck %s -; XUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -stop-after=regbankselect -regbankselect-greedy -o - %s | FileCheck %s +; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -stop-after=regbankselect -regbankselect-greedy -o - %s | FileCheck %s ; Natural mapping define amdgpu_ps void @sample_1d_vgpr_vaddr__sgpr_rsrc__sgpr_samp(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) { @@ -23,8 +23,8 @@ ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32) ; CHECK: [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32) ; CHECK: [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF - ; CHECK: [[INT:%[0-9]+]]:vgpr(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; CHECK: G_STORE [[INT]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) + ; CHECK: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; CHECK: G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) ; CHECK: S_ENDPGM 0 %v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0) store <4 x float> %v, <4 x float> addrspace(1)* undef @@ -53,8 +53,8 @@ ; CHECK: [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32) ; CHECK: [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF ; CHECK: [[COPY13:%[0-9]+]]:vgpr(s32) = COPY [[COPY12]](s32) - ; CHECK: [[INT:%[0-9]+]]:vgpr(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY13]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") - ; CHECK: G_STORE [[INT]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) + ; CHECK: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY13]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; CHECK: G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) ; CHECK: S_ENDPGM 0 %v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0) store <4 x float> %v, <4 x float> addrspace(1)* undef @@ -111,7 +111,7 @@ ; CHECK: [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV3]](s64), [[UV3]](s64), implicit $exec ; CHECK: [[S_AND_B64_2:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_3]], [[S_AND_B64_1]], implicit-def $scc ; CHECK: [[BUILD_VECTOR2:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32) - ; CHECK: [[INT:%[0-9]+]]:vgpr(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR2]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; CHECK: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR2]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") ; CHECK: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_2]], implicit-def $exec, implicit-def $scc, implicit $exec ; CHECK: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc ; CHECK: S_CBRANCH_EXECNZ %bb.2, implicit $exec @@ -119,7 +119,7 @@ ; CHECK: successors: %bb.4(0x80000000) ; CHECK: $exec = S_MOV_B64_term [[S_MOV_B64_term]] ; CHECK: bb.4: - ; CHECK: G_STORE [[INT]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) + ; CHECK: G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) ; CHECK: S_ENDPGM 0 %v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0) store <4 x float> %v, <4 x float> addrspace(1)* undef @@ -166,7 +166,7 @@ ; CHECK: [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec ; CHECK: [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc ; CHECK: [[BUILD_VECTOR2:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32) - ; CHECK: [[INT:%[0-9]+]]:vgpr(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR2]](<4 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; CHECK: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR2]](<4 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") ; CHECK: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec ; CHECK: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc ; CHECK: S_CBRANCH_EXECNZ %bb.2, implicit $exec @@ -174,7 +174,7 @@ ; CHECK: successors: %bb.4(0x80000000) ; CHECK: $exec = S_MOV_B64_term [[S_MOV_B64_term]] ; CHECK: bb.4: - ; CHECK: G_STORE [[INT]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) + ; CHECK: G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) ; CHECK: S_ENDPGM 0 %v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0) store <4 x float> %v, <4 x float> addrspace(1)* undef @@ -243,7 +243,7 @@ ; CHECK: [[V_CMP_EQ_U64_e64_5:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV5]](s64), [[UV5]](s64), implicit $exec ; CHECK: [[S_AND_B64_4:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_5]], [[S_AND_B64_3]], implicit-def $scc ; CHECK: [[BUILD_VECTOR3:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_8]](s32), [[V_READFIRSTLANE_B32_9]](s32), [[V_READFIRSTLANE_B32_10]](s32), [[V_READFIRSTLANE_B32_11]](s32) - ; CHECK: [[INT:%[0-9]+]]:vgpr(<4 x s32>) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR2]](<8 x s32>), [[BUILD_VECTOR3]](<4 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") + ; CHECK: [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR2]](<8 x s32>), [[BUILD_VECTOR3]](<4 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8") ; CHECK: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_4]], implicit-def $exec, implicit-def $scc, implicit $exec ; CHECK: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc ; CHECK: S_CBRANCH_EXECNZ %bb.2, implicit $exec @@ -251,7 +251,7 @@ ; CHECK: successors: %bb.4(0x80000000) ; CHECK: $exec = S_MOV_B64_term [[S_MOV_B64_term]] ; CHECK: bb.4: - ; CHECK: G_STORE [[INT]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) + ; CHECK: G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1) ; CHECK: S_ENDPGM 0 %v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0) store <4 x float> %v, <4 x float> addrspace(1)* undef