[AMDGPU] Switch to backwards scavenging in eliminateFrameIndex

[foad]

Summary

Frame index elimination runs backwards so we must use backwards
scavenging. Otherwise, when a scavenged register is spilled, the
scavenger will remember that the register is in use until the restore
point, but it will never reach that restore point. The result is that in
some cases it will keep scavenging different registers instead of
reusing the same one.

[foad]

llvm/test/CodeGen/AMDGPU/GlobalISel/call-outgoing-stack-args.ll

163–172

Good example of resuing s0 instead of lots of different SGPRs.

llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

struct SGPRSpillBuilder {
   void prepare() {
     // Scavenged temporary VGPR to use. It must be scavenged once for any number
     // of spilled subregs.
     // FIXME: The liveness analysis is limited and does not tell if a register
     // is in use in lanes that are currently inactive. We can never be sure if
     // a register as actually in use in another lane, so we need to save all
     // used lanes of the chosen VGPR.
     assert(RS && "Cannot spill SGPR to memory without RegScavenger");
-    TmpVGPR = RS->scavengeRegister(&AMDGPU::VGPR_32RegClass, MI, 0, false);
+    TmpVGPR = RS->scavengeRegisterBackwards(AMDGPU::VGPR_32RegClass, MI, false,
+                                            0, false);
 
     // Reserve temporary stack slot
     TmpVGPRIndex = MFI.getScavengeFI(MF.getFrameInfo(), TRI);
     if (TmpVGPR) {
       // Found a register that is dead in the currently active lanes, we only
       // need to spill inactive lanes.
       TmpVGPRLive = false;
     } else {
void prepare() {
     // the same register.
     RS->setRegUsed(TmpVGPR);
 
     // Try to scavenge SGPRs to save exec
     assert(!SavedExecReg && "Exec is already saved, refuse to save again");
     const TargetRegisterClass &RC =
         IsWave32 ? AMDGPU::SGPR_32RegClass : AMDGPU::SGPR_64RegClass;
     RS->setRegUsed(SuperReg);
-    SavedExecReg = RS->scavengeRegister(&RC, MI, 0, false);
+    SavedExecReg = RS->scavengeRegisterBackwards(RC, MI, false, 0, false);
 
     int64_t VGPRLanes = getPerVGPRData().VGPRLanes;
 
     if (SavedExecReg) {
       RS->setRegUsed(SavedExecReg);
       // Set exec to needed lanes
       BuildMI(*MBB, MI, DL, TII.get(MovOpc), SavedExecReg).addReg(ExecReg);
       auto I =
if (IsAGPR) {
         if (NeedSuperRegDef)
           AccRead.addReg(ValueReg, RegState::ImplicitDefine);
         AccRead->setAsmPrinterFlag(MachineInstr::ReloadReuse);
       }
       SubReg = TmpIntermediateVGPR;
     } else if (UseVGPROffset) {
       // FIXME: change to scavengeRegisterBackwards()
       if (!TmpOffsetVGPR) {
-        TmpOffsetVGPR = RS->scavengeRegister(&AMDGPU::VGPR_32RegClass, MI, 0);
+        TmpOffsetVGPR = RS->scavengeRegisterBackwards(AMDGPU::VGPR_32RegClass,
+                                                      MI, false, 0);
         RS->setRegUsed(TmpOffsetVGPR);
       }
     }
 
     MachinePointerInfo PInfo = BasePtrInfo.getWithOffset(RegOffset);
     MachineMemOperand *NewMMO =
         MF->getMachineMemOperand(PInfo, MMO->getFlags(), RemEltSize,
                                  commonAlignment(Alignment, RegOffset));
default: {
         if (!Offset && FrameReg && UseSGPR) {
           FIOp.setReg(FrameReg);
           return false;
         }
 
         const TargetRegisterClass *RC = UseSGPR ? &AMDGPU::SReg_32_XM0RegClass
                                                 : &AMDGPU::VGPR_32RegClass;
 
-        Register TmpReg = RS->scavengeRegister(RC, MI, 0, !UseSGPR);
+        Register TmpReg =
+            RS->scavengeRegisterBackwards(*RC, MI, false, 0, !UseSGPR);
         FIOp.setReg(TmpReg);
         FIOp.setIsKill();
 
         if ((!FrameReg || !Offset) && TmpReg) {
           unsigned Opc = UseSGPR ? AMDGPU::S_MOV_B32 : AMDGPU::V_MOV_B32_e32;
           auto MIB = BuildMI(*MBB, MI, DL, TII->get(Opc), TmpReg);
           if (FrameReg)
             MIB.addReg(FrameReg);
           else
             MIB.addImm(Offset);
 
           return false;
         }
 
         bool NeedSaveSCC =
             RS->isRegUsed(AMDGPU::SCC) && !MI->definesRegister(AMDGPU::SCC);
 
         Register TmpSReg =
             UseSGPR ? TmpReg
-                    : RS->scavengeRegister(&AMDGPU::SReg_32_XM0RegClass, MI, 0,
-                                           !UseSGPR);
+                    : RS->scavengeRegisterBackwards(AMDGPU::SReg_32_XM0RegClass,
+                                                    MI, false, 0, !UseSGPR);
 
         // TODO: for flat scratch another attempt can be made with a VGPR index
         //       if no SGPRs can be scavenged.
         if ((!TmpSReg && !FrameReg) || (!TmpReg && !UseSGPR))
           report_fatal_error("Cannot scavenge register in FI elimination!");
 
         if (!TmpSReg) {
           // Use frame register and restore it after.
default: {
         bool IsSALU = isSGPRClass(TII->getOpRegClass(*MI, FIOperandNum));
         bool LiveSCC =
             RS->isRegUsed(AMDGPU::SCC) && !MI->definesRegister(AMDGPU::SCC);
         const TargetRegisterClass *RC = IsSALU && !LiveSCC
                                             ? &AMDGPU::SReg_32RegClass
                                             : &AMDGPU::VGPR_32RegClass;
         bool IsCopy = MI->getOpcode() == AMDGPU::V_MOV_B32_e32 ||
                       MI->getOpcode() == AMDGPU::V_MOV_B32_e64;
-        Register ResultReg = IsCopy ? MI->getOperand(0).getReg()
-                                    : RS->scavengeRegister(RC, MI, 0);
+        Register ResultReg =
+            IsCopy ? MI->getOperand(0).getReg()
+                   : RS->scavengeRegisterBackwards(*RC, MI, false, 0);
 
         int64_t Offset = FrameInfo.getObjectOffset(Index);
         if (Offset == 0) {
           unsigned OpCode = IsSALU && !LiveSCC ? AMDGPU::S_LSHR_B32
                                                : AMDGPU::V_LSHRREV_B32_e64;
           // XXX - This never happens because of emergency scavenging slot at 0?
           auto Shift = BuildMI(*MBB, MI, DL, TII->get(OpCode), ResultReg)
                            .addImm(ST.getWavefrontSizeLog2())
                            .addReg(FrameReg);
           if (IsSALU && !LiveSCC)
             Shift.getInstr()->getOperand(3).setIsDead(); // Mark SCC as dead.
           if (IsSALU && LiveSCC) {
-            Register NewDest =
-                RS->scavengeRegister(&AMDGPU::SReg_32RegClass, Shift, 0);
+            Register NewDest = RS->scavengeRegisterBackwards(
+                AMDGPU::SReg_32RegClass, Shift, false, 0);
             BuildMI(*MBB, MI, DL, TII->get(AMDGPU::V_READFIRSTLANE_B32),
                     NewDest)
                 .addReg(ResultReg);
             ResultReg = NewDest;
           }
         } else {
           MachineInstrBuilder MIB;
           if (!IsSALU) {
default: {
           }
           if (!MIB || IsSALU) {
             // We have to produce a carry out, and there isn't a free SGPR pair
             // for it. We can keep the whole computation on the SALU to avoid
             // clobbering an additional register at the cost of an extra mov.
 
             // We may have 1 free scratch SGPR even though a carry out is
             // unavailable. Only one additional mov is needed.
-            Register TmpScaledReg =
-              RS->scavengeRegister(&AMDGPU::SReg_32_XM0RegClass, MI, 0, false);
+            Register TmpScaledReg = RS->scavengeRegisterBackwards(
+                AMDGPU::SReg_32_XM0RegClass, MI, false, 0, false);
             Register ScaledReg = TmpScaledReg.isValid() ? TmpScaledReg : FrameReg;
 
             BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_LSHR_B32), ScaledReg)
               .addReg(FrameReg)
               .addImm(ST.getWavefrontSizeLog2());
             BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_ADD_I32), ScaledReg)
                 .addReg(ScaledReg, RegState::Kill)
                 .addImm(Offset);
default: {
         }
       }
 
       // If the offset is simply too big, don't convert to a scratch wave offset
       // relative index.
 
       FIOp.ChangeToImmediate(Offset);
       if (!TII->isImmOperandLegal(*MI, FIOperandNum, FIOp)) {
-        Register TmpReg = RS->scavengeRegister(&AMDGPU::VGPR_32RegClass, MI, 0);
+        Register TmpReg = RS->scavengeRegisterBackwards(AMDGPU::VGPR_32RegClass,
+                                                        MI, false, 0);
         BuildMI(*MBB, MI, DL, TII->get(AMDGPU::V_MOV_B32_e32), TmpReg)
           .addImm(Offset);
         FIOp.ChangeToRegister(TmpReg, false, false, true);
       }
     }
   }
   return false;
 }

llvm/test/CodeGen/AMDGPU/GlobalISel/call-outgoing-stack-args.ll

 ; MUBUF-NEXT:    s_endpgm
 ;
 ; FLATSCR-LABEL: kernel_caller_byval:
 ; FLATSCR:       ; %bb.0:
 ; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s0, s3
 ; FLATSCR-NEXT:    v_mov_b32_e32 v0, 0
 ; FLATSCR-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
 ; FLATSCR-NEXT:    v_mov_b32_e32 v1, 0
-; FLATSCR-NEXT:    s_mov_b32 s8, 0
-; FLATSCR-NEXT:    s_mov_b32 s12, 0
-; FLATSCR-NEXT:    s_mov_b32 s11, 0
-; FLATSCR-NEXT:    s_mov_b32 s10, 0
-; FLATSCR-NEXT:    s_mov_b32 s9, 0
-; FLATSCR-NEXT:    s_mov_b32 s13, 0
-; FLATSCR-NEXT:    s_mov_b32 s7, 0
-; FLATSCR-NEXT:    s_mov_b32 s5, 0
-; FLATSCR-NEXT:    s_mov_b32 s3, 0
-; FLATSCR-NEXT:    s_mov_b32 s1, 0
 ; FLATSCR-NEXT:    s_mov_b32 s0, 0
-; FLATSCR-NEXT:    s_mov_b32 s2, 0
-; FLATSCR-NEXT:    s_mov_b32 s4, 0
-; FLATSCR-NEXT:    s_mov_b32 s6, 0
-; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s8 offset:8
-; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s12 offset:16
-; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s11 offset:24
-; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s10 offset:32
-; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s9 offset:40
-; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s13 offset:48
-; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s7 offset:56
-; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s5 offset:64
-; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s3 offset:72
-; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s1 offset:80
+; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:8
+; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:16
+; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:24
+; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:32
+; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:40
+; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:48
+; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:56
+; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:64
+; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:72
+; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:80
 ; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:88
-; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s2 offset:96
-; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s4 offset:104
-; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s6 offset:112
-; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s8 offset:120
-; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s12 offset:128
-; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s11 offset:8
+; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:96
+; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:104
+; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:112
+; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:120
+; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 offset:128
+; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s0 offset:8
 ; FLATSCR-NEXT:    s_nop 0
-; FLATSCR-NEXT:    scratch_load_dwordx2 v[2:3], off, s10 offset:16
-; FLATSCR-NEXT:    scratch_load_dwordx2 v[4:5], off, s9 offset:24
-; FLATSCR-NEXT:    s_mov_b32 s37, 0
-; FLATSCR-NEXT:    scratch_load_dwordx2 v[6:7], off, s37 offset:32
-; FLATSCR-NEXT:    s_mov_b32 s36, 0
-; FLATSCR-NEXT:    scratch_load_dwordx2 v[8:9], off, s36 offset:40
-; FLATSCR-NEXT:    s_mov_b32 s35, 0
-; FLATSCR-NEXT:    scratch_load_dwordx2 v[10:11], off, s35 offset:48
-; FLATSCR-NEXT:    s_mov_b32 s34, 0
-; FLATSCR-NEXT:    scratch_load_dwordx2 v[12:13], off, s34 offset:56
-; FLATSCR-NEXT:    s_mov_b32 s33, 0
-; FLATSCR-NEXT:    scratch_load_dwordx2 v[14:15], off, s33 offset:64
+; FLATSCR-NEXT:    scratch_load_dwordx2 v[2:3], off, s0 offset:16
+; FLATSCR-NEXT:    scratch_load_dwordx2 v[4:5], off, s0 offset:24
+; FLATSCR-NEXT:    scratch_load_dwordx2 v[6:7], off, s0 offset:32
+; FLATSCR-NEXT:    scratch_load_dwordx2 v[8:9], off, s0 offset:40
+; FLATSCR-NEXT:    scratch_load_dwordx2 v[10:11], off, s0 offset:48
+; FLATSCR-NEXT:    scratch_load_dwordx2 v[12:13], off, s0 offset:56
+; FLATSCR-NEXT:    scratch_load_dwordx2 v[14:15], off, s0 offset:64
 ; FLATSCR-NEXT:    s_movk_i32 s32, 0x50
 ; FLATSCR-NEXT:    s_getpc_b64 s[0:1]
 ; FLATSCR-NEXT:    s_add_u32 s0, s0, external_void_func_byval@rel32@lo+4
 ; FLATSCR-NEXT:    s_addc_u32 s1, s1, external_void_func_byval@rel32@hi+12
 ; FLATSCR-NEXT:    s_add_u32 s2, s32, 8
 ; FLATSCR-NEXT:    s_add_u32 s3, s32, 16
 ; FLATSCR-NEXT:    s_add_u32 s4, s32, 24
 ; FLATSCR-NEXT:    s_add_u32 s5, s32, 32

llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll

 ; GCN-NEXT:    buffer_store_dword v13, off, s[0:3], s33 offset:552 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v14, off, s[0:3], s33 offset:556 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v15, off, s[0:3], s33 offset:560 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v16, off, s[0:3], s33 offset:564 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v17, off, s[0:3], s33 offset:568 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v18, off, s[0:3], s33 offset:572 ; 4-byte Folded Spill
 ; GCN-NEXT:    global_load_dwordx4 v[8:11], v[0:1], off offset:224
 ; GCN-NEXT:    global_load_dwordx4 v[12:15], v[0:1], off offset:240
+; GCN-NEXT:    v_lshrrev_b32_e64 v1, 6, s33
+; GCN-NEXT:    v_add_u32_e32 v1, 0x100, v1
 ; GCN-NEXT:    buffer_store_dword v2, off, s[0:3], s33 offset:256
 ; GCN-NEXT:    buffer_store_dword v3, off, s[0:3], s33 offset:260
 ; GCN-NEXT:    buffer_store_dword v4, off, s[0:3], s33 offset:264
 ; GCN-NEXT:    buffer_store_dword v5, off, s[0:3], s33 offset:268
 ; GCN-NEXT:    buffer_store_dword v16, off, s[0:3], s33 offset:272
 ; GCN-NEXT:    buffer_store_dword v17, off, s[0:3], s33 offset:276
 ; GCN-NEXT:    buffer_store_dword v18, off, s[0:3], s33 offset:280
 ; GCN-NEXT:    buffer_store_dword v19, off, s[0:3], s33 offset:284

 ; GCN-NEXT:    buffer_load_dword v25, off, s[0:3], s33 offset:548 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v26, off, s[0:3], s33 offset:552 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v27, off, s[0:3], s33 offset:556 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v28, off, s[0:3], s33 offset:560 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v29, off, s[0:3], s33 offset:564 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v30, off, s[0:3], s33 offset:568 ; 4-byte Folded Reload
 ; GCN-NEXT:    buffer_load_dword v31, off, s[0:3], s33 offset:572 ; 4-byte Folded Reload
 ; GCN-NEXT:    v_bfe_u32 v0, v6, 1, 6
-; GCN-NEXT:    v_lshrrev_b32_e64 v2, 6, s33
 ; GCN-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GCN-NEXT:    v_add_u32_e32 v2, 0x100, v2
-; GCN-NEXT:    v_add_u32_e32 v0, v2, v0
+; GCN-NEXT:    v_add_u32_e32 v0, v1, v0
 ; GCN-NEXT:    v_and_b32_e32 v1, 1, v6
 ; GCN-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_mov_b32_e32 v16, v20
 ; GCN-NEXT:    v_mov_b32_e32 v17, v21
 ; GCN-NEXT:    v_mov_b32_e32 v18, v22
 ; GCN-NEXT:    v_mov_b32_e32 v19, v23
 ; GCN-NEXT:    buffer_store_dword v16, off, s[0:3], s33 offset:464

llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll

 }
 
 define amdgpu_kernel void @store_load_sindex_small_offset_kernel(i32 %idx) {
 ; GFX9-LABEL: store_load_sindex_small_offset_kernel:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_load_dword s0, s[0:1], 0x24
 ; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s2, s5
 ; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
-; GFX9-NEXT:    s_mov_b32 s2, 0
-; GFX9-NEXT:    scratch_load_dword v0, off, s2 offset:4 glc
+; GFX9-NEXT:    s_mov_b32 s1, 0
+; GFX9-NEXT:    scratch_load_dword v0, off, s1 offset:4 glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    s_lshl_b32 s1, s0, 2
 ; GFX9-NEXT:    s_and_b32 s0, s0, 15
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX9-NEXT:    s_addk_i32 s1, 0x104
 ; GFX9-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX9-NEXT:    scratch_store_dword off, v0, s1
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)

 
 define void @store_load_vindex_small_offset_foo(i32 %idx) {
 ; GFX9-LABEL: store_load_vindex_small_offset_foo:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    scratch_load_dword v1, off, s32 glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
-; GFX9-NEXT:    s_add_i32 s1, s32, 0x100
+; GFX9-NEXT:    s_add_i32 s0, s32, 0x100
 ; GFX9-NEXT:    v_and_b32_e32 v0, 15, v0
-; GFX9-NEXT:    v_add_u32_e32 v1, s1, v1
+; GFX9-NEXT:    v_add_u32_e32 v1, s0, v1
 ; GFX9-NEXT:    v_mov_b32_e32 v2, 15
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    s_add_i32 s0, s32, 0x100
 ; GFX9-NEXT:    scratch_store_dword v1, v2, off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_add_u32_e32 v0, s0, v0
 ; GFX9-NEXT:    scratch_load_dword v0, v0, off glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: store_load_vindex_small_offset_foo:
 ; GFX10:       ; %bb.0: ; %bb
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_and_b32_e32 v1, 15, v0
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_add_i32 s1, s32, 0x100
 ; GFX10-NEXT:    s_add_i32 s0, s32, 0x100
 ; GFX10-NEXT:    v_mov_b32_e32 v2, 15
-; GFX10-NEXT:    v_lshlrev_b32_e32 v1, 2, v1
-; GFX10-NEXT:    v_add_nc_u32_e32 v0, s1, v0
 ; GFX10-NEXT:    scratch_load_dword v3, off, s32 glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    v_lshlrev_b32_e32 v1, 2, v1
+; GFX10-NEXT:    v_add_nc_u32_e32 v0, s0, v0
+; GFX10-NEXT:    s_add_i32 s0, s32, 0x100
 ; GFX10-NEXT:    v_add_nc_u32_e32 v1, s0, v1
 ; GFX10-NEXT:    scratch_store_dword v0, v2, off
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    scratch_load_dword v0, v1, off glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX940-LABEL: store_load_vindex_small_offset_foo:

 }
 
 define amdgpu_kernel void @store_load_sindex_large_offset_kernel(i32 %idx) {
 ; GFX9-LABEL: store_load_sindex_large_offset_kernel:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_load_dword s0, s[0:1], 0x24
 ; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s2, s5
 ; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
-; GFX9-NEXT:    s_mov_b32 s2, 0
-; GFX9-NEXT:    scratch_load_dword v0, off, s2 offset:4 glc
+; GFX9-NEXT:    s_mov_b32 s1, 0
+; GFX9-NEXT:    scratch_load_dword v0, off, s1 offset:4 glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    s_lshl_b32 s1, s0, 2
 ; GFX9-NEXT:    s_and_b32 s0, s0, 15
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX9-NEXT:    s_addk_i32 s1, 0x4004
 ; GFX9-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX9-NEXT:    scratch_store_dword off, v0, s1
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)

 
 define void @store_load_vindex_large_offset_foo(i32 %idx) {
 ; GFX9-LABEL: store_load_vindex_large_offset_foo:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    scratch_load_dword v1, off, s32 offset:4 glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
-; GFX9-NEXT:    s_add_i32 s1, s32, 0x4004
+; GFX9-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX9-NEXT:    v_and_b32_e32 v0, 15, v0
-; GFX9-NEXT:    v_add_u32_e32 v1, s1, v1
+; GFX9-NEXT:    v_add_u32_e32 v1, s0, v1
 ; GFX9-NEXT:    v_mov_b32_e32 v2, 15
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX9-NEXT:    scratch_store_dword v1, v2, off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_add_u32_e32 v0, s0, v0
 ; GFX9-NEXT:    scratch_load_dword v0, v0, off glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: store_load_vindex_large_offset_foo:
 ; GFX10:       ; %bb.0: ; %bb
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_and_b32_e32 v1, 15, v0
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_add_i32 s1, s32, 0x4004
 ; GFX10-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX10-NEXT:    v_mov_b32_e32 v2, 15
-; GFX10-NEXT:    v_lshlrev_b32_e32 v1, 2, v1
-; GFX10-NEXT:    v_add_nc_u32_e32 v0, s1, v0
 ; GFX10-NEXT:    scratch_load_dword v3, off, s32 offset:4 glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    v_lshlrev_b32_e32 v1, 2, v1
+; GFX10-NEXT:    v_add_nc_u32_e32 v0, s0, v0
+; GFX10-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX10-NEXT:    v_add_nc_u32_e32 v1, s0, v1
 ; GFX10-NEXT:    scratch_store_dword v0, v2, off
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    scratch_load_dword v0, v1, off glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX940-LABEL: store_load_vindex_large_offset_foo:
 ; GFX940:       ; %bb.0: ; %bb
 ; GFX940-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX940-NEXT:    scratch_load_dword v1, off, s32 offset:4 sc0 sc1
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
 ; GFX940-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
-; GFX940-NEXT:    s_add_i32 s1, s32, 0x4004
-; GFX940-NEXT:    v_add_u32_e32 v1, s1, v1
+; GFX940-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX940-NEXT:    v_add_u32_e32 v1, s0, v1
 ; GFX940-NEXT:    v_mov_b32_e32 v2, 15
 ; GFX940-NEXT:    v_and_b32_e32 v0, 15, v0
 ; GFX940-NEXT:    scratch_store_dword v1, v2, off sc0 sc1
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
 ; GFX940-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX940-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX940-NEXT:    scratch_load_dword v0, v0, s0 sc0 sc1
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
 ; GFX940-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: store_load_vindex_large_offset_foo:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    v_dual_mov_b32 v2, 15 :: v_dual_lshlrev_b32 v1, 2, v0
 ; GFX11-NEXT:    v_and_b32_e32 v0, 15, v0
-; GFX11-NEXT:    s_add_i32 s1, s32, 0x4004
 ; GFX11-NEXT:    s_add_i32 s0, s32, 0x4004
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT:    v_add_nc_u32_e32 v1, s1, v1
 ; GFX11-NEXT:    scratch_load_b32 v3, off, s32 offset:4 glc dlc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_add_nc_u32_e32 v1, s0, v1
+; GFX11-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX11-NEXT:    scratch_store_b32 v1, v2, off dlc
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    scratch_load_b32 v0, v0, s0 glc dlc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 bb:
   %padding = alloca [4096 x i32], align 4, addrspace(5)

 }
 
 define amdgpu_kernel void @store_load_large_imm_offset_kernel() {
 ; GFX9-LABEL: store_load_large_imm_offset_kernel:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s0, s3
 ; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 13
-; GFX9-NEXT:    s_mov_b32 s1, 0
-; GFX9-NEXT:    s_movk_i32 s0, 0x3e80
-; GFX9-NEXT:    scratch_store_dword off, v0, s1 offset:4
+; GFX9-NEXT:    s_mov_b32 s0, 0
+; GFX9-NEXT:    scratch_store_dword off, v0, s0 offset:4
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_movk_i32 s0, 0x3e80
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX9-NEXT:    s_add_i32 s0, s0, 4
 ; GFX9-NEXT:    scratch_store_dword off, v0, s0
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    scratch_load_dword v0, off, s0 glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_endpgm
 ;

llvm/test/CodeGen/AMDGPU/chain-hi-to-lo.ll

 ; GFX900-NEXT:    s_endpgm
 ;
 ; FLATSCR-LABEL: vload2_private:
 ; FLATSCR:       ; %bb.0: ; %entry
 ; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s2, s5
 ; FLATSCR-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
 ; FLATSCR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; FLATSCR-NEXT:    v_mov_b32_e32 v2, 0
-; FLATSCR-NEXT:    s_mov_b32 s6, 0
-; FLATSCR-NEXT:    s_mov_b32 s5, 0
 ; FLATSCR-NEXT:    s_mov_b32 s4, 0
 ; FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
 ; FLATSCR-NEXT:    global_load_ushort v0, v2, s[0:1]
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
-; FLATSCR-NEXT:    scratch_store_short off, v0, s6 offset:4
+; FLATSCR-NEXT:    scratch_store_short off, v0, s4 offset:4
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    global_load_ushort v0, v2, s[0:1] offset:2
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
-; FLATSCR-NEXT:    scratch_store_short off, v0, s5 offset:6
+; FLATSCR-NEXT:    scratch_store_short off, v0, s4 offset:6
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    global_load_ushort v0, v2, s[0:1] offset:4
-; FLATSCR-NEXT:    s_mov_b32 s1, 0
 ; FLATSCR-NEXT:    s_mov_b32 s0, 0
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
-; FLATSCR-NEXT:    scratch_store_short off, v0, s4 offset:8
+; FLATSCR-NEXT:    scratch_store_short off, v0, s0 offset:8
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
-; FLATSCR-NEXT:    scratch_load_dword v0, off, s1 offset:4
+; FLATSCR-NEXT:    scratch_load_dword v0, off, s0 offset:4
 ; FLATSCR-NEXT:    scratch_load_dword v1, off, s0 offset:6
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
 ; FLATSCR-NEXT:    s_endpgm
 ;
 ; GFX10_DEFAULT-LABEL: vload2_private:
 ; GFX10_DEFAULT:       ; %bb.0: ; %entry
 ; GFX10_DEFAULT-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0

 ; FLATSCR_GFX10-LABEL: vload2_private:
 ; FLATSCR_GFX10:       ; %bb.0: ; %entry
 ; FLATSCR_GFX10-NEXT:    s_add_u32 s2, s2, s5
 ; FLATSCR_GFX10-NEXT:    s_addc_u32 s3, s3, 0
 ; FLATSCR_GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2
 ; FLATSCR_GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3
 ; FLATSCR_GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; FLATSCR_GFX10-NEXT:    v_mov_b32_e32 v2, 0
-; FLATSCR_GFX10-NEXT:    s_mov_b32 s6, 0
-; FLATSCR_GFX10-NEXT:    s_mov_b32 s5, 0
 ; FLATSCR_GFX10-NEXT:    s_mov_b32 s4, 0
 ; FLATSCR_GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; FLATSCR_GFX10-NEXT:    global_load_ushort v0, v2, s[0:1]
 ; FLATSCR_GFX10-NEXT:    s_waitcnt vmcnt(0)
-; FLATSCR_GFX10-NEXT:    scratch_store_short off, v0, s6 offset:4
+; FLATSCR_GFX10-NEXT:    scratch_store_short off, v0, s4 offset:4
 ; FLATSCR_GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; FLATSCR_GFX10-NEXT:    global_load_ushort v0, v2, s[0:1] offset:2
 ; FLATSCR_GFX10-NEXT:    s_waitcnt vmcnt(0)
-; FLATSCR_GFX10-NEXT:    scratch_store_short off, v0, s5 offset:6
+; FLATSCR_GFX10-NEXT:    scratch_store_short off, v0, s4 offset:6
 ; FLATSCR_GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; FLATSCR_GFX10-NEXT:    global_load_ushort v0, v2, s[0:1] offset:4
 ; FLATSCR_GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; FLATSCR_GFX10-NEXT:    s_mov_b32 s1, 0
 ; FLATSCR_GFX10-NEXT:    s_mov_b32 s0, 0
 ; FLATSCR_GFX10-NEXT:    s_waitcnt vmcnt(0)
-; FLATSCR_GFX10-NEXT:    scratch_store_short off, v0, s4 offset:8
+; FLATSCR_GFX10-NEXT:    scratch_store_short off, v0, s0 offset:8
 ; FLATSCR_GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; FLATSCR_GFX10-NEXT:    s_clause 0x1
-; FLATSCR_GFX10-NEXT:    scratch_load_dword v0, off, s1 offset:4
+; FLATSCR_GFX10-NEXT:    scratch_load_dword v0, off, s0 offset:4
 ; FLATSCR_GFX10-NEXT:    scratch_load_dword v1, off, s0 offset:6
 ; FLATSCR_GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR_GFX10-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
 ; FLATSCR_GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: vload2_private:
 ; GFX11:       ; %bb.0: ; %entry
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0

llvm/test/CodeGen/AMDGPU/extract-load-i1.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa < %s | FileCheck %s
 
 ; FIXME: Inefficient codegen which skips an optimization of load +
 ; extractelement when the vector element type is not byte-sized.
 define i1 @extractloadi1(ptr %ptr, i32 %idx) {
 ; CHECK-LABEL: extractloadi1:
 ; CHECK:       ; %bb.0:
 ; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; CHECK-NEXT:    flat_load_ubyte v0, v[0:1]
 ; CHECK-NEXT:    v_and_b32_e32 v1, 7, v2
-; CHECK-NEXT:    v_lshr_b32_e64 v9, s32, 6
-; CHECK-NEXT:    v_or_b32_e32 v1, v9, v1
+; CHECK-NEXT:    v_lshr_b32_e64 v2, s32, 6
+; CHECK-NEXT:    v_or_b32_e32 v1, v2, v1
 ; CHECK-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; CHECK-NEXT:    v_bfe_u32 v2, v0, 1, 1
 ; CHECK-NEXT:    v_bfe_u32 v3, v0, 2, 2
 ; CHECK-NEXT:    v_bfe_u32 v4, v0, 3, 1
 ; CHECK-NEXT:    v_lshrrev_b32_e32 v5, 4, v0
 ; CHECK-NEXT:    v_bfe_u32 v6, v0, 5, 1
 ; CHECK-NEXT:    v_lshrrev_b32_e32 v7, 6, v0
 ; CHECK-NEXT:    v_lshrrev_b32_e32 v8, 7, v0

llvm/test/CodeGen/AMDGPU/flat-scratch-init.ll

 define amdgpu_kernel void @test(ptr addrspace(1) %out, i32 %in) {
 ; FLAT_SCR_OPT-LABEL: test:
 ; FLAT_SCR_OPT:       ; %bb.0:
 ; FLAT_SCR_OPT-NEXT:    s_add_u32 s2, s2, s5
 ; FLAT_SCR_OPT-NEXT:    s_addc_u32 s3, s3, 0
 ; FLAT_SCR_OPT-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2
 ; FLAT_SCR_OPT-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3
 ; FLAT_SCR_OPT-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
-; FLAT_SCR_OPT-NEXT:    s_mov_b32 s104, exec_lo
+; FLAT_SCR_OPT-NEXT:    s_mov_b32 s4, exec_lo
 ; FLAT_SCR_OPT-NEXT:    s_mov_b32 exec_lo, 3
-; FLAT_SCR_OPT-NEXT:    s_mov_b32 s4, 0
-; FLAT_SCR_OPT-NEXT:    scratch_store_dword off, v72, s4
+; FLAT_SCR_OPT-NEXT:    s_mov_b32 s5, 0
+; FLAT_SCR_OPT-NEXT:    scratch_store_dword off, v0, s5
 ; FLAT_SCR_OPT-NEXT:    s_waitcnt lgkmcnt(0)
-; FLAT_SCR_OPT-NEXT:    v_writelane_b32 v72, s2, 0
-; FLAT_SCR_OPT-NEXT:    s_mov_b32 s4, 4
-; FLAT_SCR_OPT-NEXT:    v_writelane_b32 v72, s3, 1
-; FLAT_SCR_OPT-NEXT:    scratch_store_dword off, v72, s4 ; 4-byte Folded Spill
+; FLAT_SCR_OPT-NEXT:    v_writelane_b32 v0, s2, 0
+; FLAT_SCR_OPT-NEXT:    s_mov_b32 s5, 4
+; FLAT_SCR_OPT-NEXT:    v_writelane_b32 v0, s3, 1
+; FLAT_SCR_OPT-NEXT:    scratch_store_dword off, v0, s5 ; 4-byte Folded Spill
 ; FLAT_SCR_OPT-NEXT:    s_waitcnt_depctr 0xffe3
-; FLAT_SCR_OPT-NEXT:    s_mov_b32 s4, 0
-; FLAT_SCR_OPT-NEXT:    scratch_load_dword v72, off, s4
+; FLAT_SCR_OPT-NEXT:    s_mov_b32 s5, 0
+; FLAT_SCR_OPT-NEXT:    scratch_load_dword v0, off, s5
 ; FLAT_SCR_OPT-NEXT:    s_waitcnt vmcnt(0)
 ; FLAT_SCR_OPT-NEXT:    s_waitcnt_depctr 0xffe3
-; FLAT_SCR_OPT-NEXT:    s_mov_b32 exec_lo, s104
+; FLAT_SCR_OPT-NEXT:    s_mov_b32 exec_lo, s4
 ; FLAT_SCR_OPT-NEXT:    s_load_dword vcc_lo, s[0:1], 0x8
 ; FLAT_SCR_OPT-NEXT:    ; kill: killed $sgpr0_sgpr1
 ; FLAT_SCR_OPT-NEXT:    ;;#ASMSTART
 ; FLAT_SCR_OPT-NEXT:    ;;#ASMEND
 ; FLAT_SCR_OPT-NEXT:    ;;#ASMSTART
 ; FLAT_SCR_OPT-NEXT:    ;;#ASMEND
 ; FLAT_SCR_OPT-NEXT:    ;;#ASMSTART
 ; FLAT_SCR_OPT-NEXT:    ;;#ASMEND

 ; FLAT_SCR_OPT-NEXT:    ;;#ASMEND
 ; FLAT_SCR_OPT-NEXT:    ;;#ASMSTART
 ; FLAT_SCR_OPT-NEXT:    ;;#ASMEND
 ; FLAT_SCR_OPT-NEXT:    ;;#ASMSTART
 ; FLAT_SCR_OPT-NEXT:    ;;#ASMEND
 ; FLAT_SCR_OPT-NEXT:    s_mov_b32 s2, exec_lo
 ; FLAT_SCR_OPT-NEXT:    s_mov_b32 exec_lo, 3
 ; FLAT_SCR_OPT-NEXT:    s_mov_b32 s3, 0
-; FLAT_SCR_OPT-NEXT:    scratch_store_dword off, v2, s3
+; FLAT_SCR_OPT-NEXT:    scratch_store_dword off, v1, s3
 ; FLAT_SCR_OPT-NEXT:    s_waitcnt_depctr 0xffe3
 ; FLAT_SCR_OPT-NEXT:    s_mov_b32 s3, 4
-; FLAT_SCR_OPT-NEXT:    scratch_load_dword v2, off, s3 ; 4-byte Folded Reload
+; FLAT_SCR_OPT-NEXT:    scratch_load_dword v1, off, s3 ; 4-byte Folded Reload
 ; FLAT_SCR_OPT-NEXT:    s_waitcnt_depctr 0xffe3
 ; FLAT_SCR_OPT-NEXT:    s_mov_b32 s3, 0
 ; FLAT_SCR_OPT-NEXT:    s_waitcnt vmcnt(0)
-; FLAT_SCR_OPT-NEXT:    v_readlane_b32 s0, v2, 0
-; FLAT_SCR_OPT-NEXT:    v_readlane_b32 s1, v2, 1
-; FLAT_SCR_OPT-NEXT:    scratch_load_dword v2, off, s3
+; FLAT_SCR_OPT-NEXT:    v_readlane_b32 s0, v1, 0
+; FLAT_SCR_OPT-NEXT:    v_readlane_b32 s1, v1, 1
+; FLAT_SCR_OPT-NEXT:    scratch_load_dword v1, off, s3
 ; FLAT_SCR_OPT-NEXT:    s_waitcnt vmcnt(0)
 ; FLAT_SCR_OPT-NEXT:    s_waitcnt_depctr 0xffe3
 ; FLAT_SCR_OPT-NEXT:    s_mov_b32 exec_lo, s2
 ; FLAT_SCR_OPT-NEXT:    v_mov_b32_e32 v1, 0
 ; FLAT_SCR_OPT-NEXT:    global_store_dword v1, v0, s[0:1]
 ; FLAT_SCR_OPT-NEXT:    s_endpgm
 ;
 ; FLAT_SCR_ARCH-LABEL: test:
 ; FLAT_SCR_ARCH:       ; %bb.0:
 ; FLAT_SCR_ARCH-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
-; FLAT_SCR_ARCH-NEXT:    s_mov_b32 s104, exec_lo
+; FLAT_SCR_ARCH-NEXT:    s_mov_b32 s4, exec_lo
 ; FLAT_SCR_ARCH-NEXT:    s_mov_b32 exec_lo, 3
-; FLAT_SCR_ARCH-NEXT:    s_mov_b32 s4, 0
-; FLAT_SCR_ARCH-NEXT:    scratch_store_dword off, v72, s4
+; FLAT_SCR_ARCH-NEXT:    s_mov_b32 s5, 0
+; FLAT_SCR_ARCH-NEXT:    scratch_store_dword off, v0, s5
 ; FLAT_SCR_ARCH-NEXT:    s_waitcnt lgkmcnt(0)
-; FLAT_SCR_ARCH-NEXT:    v_writelane_b32 v72, s2, 0
-; FLAT_SCR_ARCH-NEXT:    s_mov_b32 s4, 4
-; FLAT_SCR_ARCH-NEXT:    v_writelane_b32 v72, s3, 1
-; FLAT_SCR_ARCH-NEXT:    scratch_store_dword off, v72, s4 ; 4-byte Folded Spill
+; FLAT_SCR_ARCH-NEXT:    v_writelane_b32 v0, s2, 0
+; FLAT_SCR_ARCH-NEXT:    s_mov_b32 s5, 4
+; FLAT_SCR_ARCH-NEXT:    v_writelane_b32 v0, s3, 1
+; FLAT_SCR_ARCH-NEXT:    scratch_store_dword off, v0, s5 ; 4-byte Folded Spill
 ; FLAT_SCR_ARCH-NEXT:    s_waitcnt_depctr 0xffe3
-; FLAT_SCR_ARCH-NEXT:    s_mov_b32 s4, 0
-; FLAT_SCR_ARCH-NEXT:    scratch_load_dword v72, off, s4
+; FLAT_SCR_ARCH-NEXT:    s_mov_b32 s5, 0
+; FLAT_SCR_ARCH-NEXT:    scratch_load_dword v0, off, s5
 ; FLAT_SCR_ARCH-NEXT:    s_waitcnt vmcnt(0)
 ; FLAT_SCR_ARCH-NEXT:    s_waitcnt_depctr 0xffe3
-; FLAT_SCR_ARCH-NEXT:    s_mov_b32 exec_lo, s104
+; FLAT_SCR_ARCH-NEXT:    s_mov_b32 exec_lo, s4
 ; FLAT_SCR_ARCH-NEXT:    s_load_dword vcc_lo, s[0:1], 0x8
 ; FLAT_SCR_ARCH-NEXT:    ; kill: killed $sgpr0_sgpr1
 ; FLAT_SCR_ARCH-NEXT:    ;;#ASMSTART
 ; FLAT_SCR_ARCH-NEXT:    ;;#ASMEND
 ; FLAT_SCR_ARCH-NEXT:    ;;#ASMSTART
 ; FLAT_SCR_ARCH-NEXT:    ;;#ASMEND
 ; FLAT_SCR_ARCH-NEXT:    ;;#ASMSTART
 ; FLAT_SCR_ARCH-NEXT:    ;;#ASMEND

 ; FLAT_SCR_ARCH-NEXT:    ;;#ASMEND
 ; FLAT_SCR_ARCH-NEXT:    ;;#ASMSTART
 ; FLAT_SCR_ARCH-NEXT:    ;;#ASMEND
 ; FLAT_SCR_ARCH-NEXT:    ;;#ASMSTART
 ; FLAT_SCR_ARCH-NEXT:    ;;#ASMEND
 ; FLAT_SCR_ARCH-NEXT:    s_mov_b32 s2, exec_lo
 ; FLAT_SCR_ARCH-NEXT:    s_mov_b32 exec_lo, 3
 ; FLAT_SCR_ARCH-NEXT:    s_mov_b32 s3, 0
-; FLAT_SCR_ARCH-NEXT:    scratch_store_dword off, v2, s3
+; FLAT_SCR_ARCH-NEXT:    scratch_store_dword off, v1, s3
 ; FLAT_SCR_ARCH-NEXT:    s_waitcnt_depctr 0xffe3
 ; FLAT_SCR_ARCH-NEXT:    s_mov_b32 s3, 4
-; FLAT_SCR_ARCH-NEXT:    scratch_load_dword v2, off, s3 ; 4-byte Folded Reload
+; FLAT_SCR_ARCH-NEXT:    scratch_load_dword v1, off, s3 ; 4-byte Folded Reload
 ; FLAT_SCR_ARCH-NEXT:    s_waitcnt_depctr 0xffe3
 ; FLAT_SCR_ARCH-NEXT:    s_mov_b32 s3, 0
 ; FLAT_SCR_ARCH-NEXT:    s_waitcnt vmcnt(0)
-; FLAT_SCR_ARCH-NEXT:    v_readlane_b32 s0, v2, 0
-; FLAT_SCR_ARCH-NEXT:    v_readlane_b32 s1, v2, 1
-; FLAT_SCR_ARCH-NEXT:    scratch_load_dword v2, off, s3
+; FLAT_SCR_ARCH-NEXT:    v_readlane_b32 s0, v1, 0
+; FLAT_SCR_ARCH-NEXT:    v_readlane_b32 s1, v1, 1
+; FLAT_SCR_ARCH-NEXT:    scratch_load_dword v1, off, s3
 ; FLAT_SCR_ARCH-NEXT:    s_waitcnt vmcnt(0)
 ; FLAT_SCR_ARCH-NEXT:    s_waitcnt_depctr 0xffe3
 ; FLAT_SCR_ARCH-NEXT:    s_mov_b32 exec_lo, s2
 ; FLAT_SCR_ARCH-NEXT:    v_mov_b32_e32 v1, 0
 ; FLAT_SCR_ARCH-NEXT:    global_store_dword v1, v0, s[0:1]
 ; FLAT_SCR_ARCH-NEXT:    s_endpgm
   call void asm sideeffect "", "~{s[0:7]}" ()
   call void asm sideeffect "", "~{s[8:15]}" ()

llvm/test/CodeGen/AMDGPU/flat-scratch.ll

 ; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
 ; GFX9-NEXT:    s_mov_b32 s1, s0
 ; GFX9-NEXT:    s_mov_b32 s2, s0
 ; GFX9-NEXT:    s_mov_b32 s3, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s3
-; GFX9-NEXT:    s_mov_b32 s3, 0
-; GFX9-NEXT:    s_mov_b32 s2, 0
-; GFX9-NEXT:    s_mov_b32 s1, 0
-; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s3 offset:52
-; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:36
-; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:20
+; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:52
+; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:36
+; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:20
 ; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:4
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: zero_init_kernel:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_add_u32 s0, s0, s3
 ; GFX10-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0

 ; GFX9-PAL-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
 ; GFX9-PAL-NEXT:    s_mov_b32 s1, s0
 ; GFX9-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX9-PAL-NEXT:    s_mov_b32 s3, s0
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v3, s3
-; GFX9-PAL-NEXT:    s_mov_b32 s3, 0
-; GFX9-PAL-NEXT:    s_mov_b32 s2, 0
-; GFX9-PAL-NEXT:    s_mov_b32 s1, 0
-; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s3 offset:52
-; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:36
-; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:20
+; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:52
+; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:36
+; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:20
 ; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:4
 ; GFX9-PAL-NEXT:    s_endpgm
 ;
 ; GFX940-LABEL: zero_init_kernel:
 ; GFX940:       ; %bb.0:
 ; GFX940-NEXT:    s_mov_b32 s0, 0
 ; GFX940-NEXT:    s_mov_b32 s1, s0
 ; GFX940-NEXT:    s_mov_b32 s2, s0

 ; GFX1010-PAL-NEXT:    s_mov_b32 s0, 0
 ; GFX1010-PAL-NEXT:    s_mov_b32 s1, s0
 ; GFX1010-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX1010-PAL-NEXT:    s_mov_b32 s3, s0
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v3, s3
-; GFX1010-PAL-NEXT:    s_mov_b32 s3, 0
-; GFX1010-PAL-NEXT:    s_mov_b32 s2, 0
-; GFX1010-PAL-NEXT:    s_mov_b32 s1, 0
-; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s3 offset:52
-; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:36
-; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:20
+; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:52
+; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:36
+; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:20
 ; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:4
 ; GFX1010-PAL-NEXT:    s_endpgm
 ;
 ; GFX1030-PAL-LABEL: zero_init_kernel:
 ; GFX1030-PAL:       ; %bb.0:
 ; GFX1030-PAL-NEXT:    s_getpc_b64 s[2:3]
 ; GFX1030-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX1030-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
; GFX11-PAL-NEXT:    s_setpc_b64 s[30:31]
   ret void
 }
 
 define amdgpu_kernel void @zero_init_small_offset_kernel() {
 ; GFX9-LABEL: zero_init_small_offset_kernel:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s0, s3
 ; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
-; GFX9-NEXT:    s_mov_b32 s4, 0
-; GFX9-NEXT:    scratch_load_dword v0, off, s4 offset:4 glc
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_mov_b32 s0, 0
+; GFX9-NEXT:    scratch_load_dword v0, off, s0 offset:4 glc
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_mov_b32 s1, s0
 ; GFX9-NEXT:    s_mov_b32 s2, s0
 ; GFX9-NEXT:    s_mov_b32 s3, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s3
-; GFX9-NEXT:    s_mov_b32 s3, 0
-; GFX9-NEXT:    s_mov_b32 s2, 0
-; GFX9-NEXT:    s_mov_b32 s1, 0
-; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s3 offset:260
-; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:276
-; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:292
+; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:260
+; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:276
+; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:292
 ; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:308
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: zero_init_small_offset_kernel:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_add_u32 s0, s0, s3
 ; GFX10-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0

 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 ;
 ; GFX9-PAL-LABEL: zero_init_small_offset_kernel:
 ; GFX9-PAL:       ; %bb.0:
 ; GFX9-PAL-NEXT:    s_getpc_b64 s[2:3]
 ; GFX9-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX9-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
-; GFX9-PAL-NEXT:    s_mov_b32 s4, 0
 ; GFX9-PAL-NEXT:    s_mov_b32 s0, 0
 ; GFX9-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-PAL-NEXT:    s_and_b32 s3, s3, 0xffff
 ; GFX9-PAL-NEXT:    s_add_u32 flat_scratch_lo, s2, s1
 ; GFX9-PAL-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
-; GFX9-PAL-NEXT:    scratch_load_dword v0, off, s4 offset:4 glc
+; GFX9-PAL-NEXT:    scratch_load_dword v0, off, s0 offset:4 glc
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    s_mov_b32 s1, s0
 ; GFX9-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX9-PAL-NEXT:    s_mov_b32 s3, s0
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v3, s3
-; GFX9-PAL-NEXT:    s_mov_b32 s3, 0
-; GFX9-PAL-NEXT:    s_mov_b32 s2, 0
-; GFX9-PAL-NEXT:    s_mov_b32 s1, 0
-; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s3 offset:260
-; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:276
-; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:292
+; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:260
+; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:276
+; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:292
 ; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:308
 ; GFX9-PAL-NEXT:    s_endpgm
 ;
 ; GFX940-LABEL: zero_init_small_offset_kernel:
 ; GFX940:       ; %bb.0:
 ; GFX940-NEXT:    scratch_load_dword v0, off, off offset:4 sc0 sc1
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
 ; GFX940-NEXT:    s_mov_b32 s0, 0

 ; GFX1010-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX1010-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
 ; GFX1010-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1010-PAL-NEXT:    s_and_b32 s3, s3, 0xffff
 ; GFX1010-PAL-NEXT:    s_add_u32 s2, s2, s1
 ; GFX1010-PAL-NEXT:    s_addc_u32 s3, s3, 0
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3
-; GFX1010-PAL-NEXT:    s_mov_b32 s4, 0
 ; GFX1010-PAL-NEXT:    s_mov_b32 s0, 0
-; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, s4 offset:4 glc dlc
+; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, s0 offset:4 glc dlc
 ; GFX1010-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1010-PAL-NEXT:    s_mov_b32 s1, s0
 ; GFX1010-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX1010-PAL-NEXT:    s_mov_b32 s3, s0
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v3, s3
-; GFX1010-PAL-NEXT:    s_mov_b32 s3, 0
-; GFX1010-PAL-NEXT:    s_mov_b32 s2, 0
-; GFX1010-PAL-NEXT:    s_mov_b32 s1, 0
-; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s3 offset:260
-; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:276
-; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:292
+; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:260
+; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:276
+; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:292
 ; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:308
 ; GFX1010-PAL-NEXT:    s_endpgm
 ;
 ; GFX1030-PAL-LABEL: zero_init_small_offset_kernel:
 ; GFX1030-PAL:       ; %bb.0:
 ; GFX1030-PAL-NEXT:    s_getpc_b64 s[2:3]
 ; GFX1030-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX1030-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0

 }
 
 define amdgpu_kernel void @store_load_sindex_small_offset_kernel(i32 %idx) {
 ; GFX9-LABEL: store_load_sindex_small_offset_kernel:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_load_dword s0, s[0:1], 0x24
 ; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s2, s5
 ; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
-; GFX9-NEXT:    s_mov_b32 s2, 0
-; GFX9-NEXT:    scratch_load_dword v0, off, s2 offset:4 glc
+; GFX9-NEXT:    s_mov_b32 s1, 0
+; GFX9-NEXT:    scratch_load_dword v0, off, s1 offset:4 glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    s_lshl_b32 s1, s0, 2
 ; GFX9-NEXT:    s_and_b32 s0, s0, 15
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX9-NEXT:    s_addk_i32 s1, 0x104
 ; GFX9-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX9-NEXT:    scratch_store_dword off, v0, s1
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)

 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_endpgm
 ;
 ; GFX9-PAL-LABEL: store_load_sindex_small_offset_kernel:
 ; GFX9-PAL:       ; %bb.0: ; %bb
 ; GFX9-PAL-NEXT:    s_getpc_b64 s[4:5]
 ; GFX9-PAL-NEXT:    s_mov_b32 s4, s0
 ; GFX9-PAL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
-; GFX9-PAL-NEXT:    s_mov_b32 s2, 0
-; GFX9-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX9-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-PAL-NEXT:    s_and_b32 s5, s5, 0xffff
+; GFX9-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX9-PAL-NEXT:    s_add_u32 flat_scratch_lo, s4, s3
 ; GFX9-PAL-NEXT:    s_addc_u32 flat_scratch_hi, s5, 0
-; GFX9-PAL-NEXT:    scratch_load_dword v0, off, s2 offset:4 glc
-; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-PAL-NEXT:    s_mov_b32 s1, 0
+; GFX9-PAL-NEXT:    scratch_load_dword v0, off, s1 offset:4 glc
+; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX9-PAL-NEXT:    s_lshl_b32 s1, s0, 2
 ; GFX9-PAL-NEXT:    s_and_b32 s0, s0, 15
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX9-PAL-NEXT:    s_addk_i32 s1, 0x104
 ; GFX9-PAL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX9-PAL-NEXT:    scratch_store_dword off, v0, s1
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    s_addk_i32 s0, 0x104

 ; GFX1010-PAL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX1010-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1010-PAL-NEXT:    s_and_b32 s5, s5, 0xffff
 ; GFX1010-PAL-NEXT:    s_add_u32 s4, s4, s3
 ; GFX1010-PAL-NEXT:    s_addc_u32 s5, s5, 0
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s4
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s5
 ; GFX1010-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
-; GFX1010-PAL-NEXT:    s_mov_b32 s2, 0
-; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, s2 offset:4 glc dlc
+; GFX1010-PAL-NEXT:    s_mov_b32 s1, 0
+; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, s1 offset:4 glc dlc
 ; GFX1010-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX1010-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1010-PAL-NEXT:    s_and_b32 s1, s0, 15
 ; GFX1010-PAL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX1010-PAL-NEXT:    s_lshl_b32 s1, s1, 2
 ; GFX1010-PAL-NEXT:    s_addk_i32 s0, 0x104
 ; GFX1010-PAL-NEXT:    s_addk_i32 s1, 0x104
bb:
   ret void
 }
 
 define amdgpu_ps void @store_load_sindex_small_offset_foo(i32 inreg %idx) {
 ; GFX9-LABEL: store_load_sindex_small_offset_foo:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s0, s3
 ; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
-; GFX9-NEXT:    s_mov_b32 s1, 0
-; GFX9-NEXT:    scratch_load_dword v0, off, s1 offset:4 glc
+; GFX9-NEXT:    s_mov_b32 s0, 0
+; GFX9-NEXT:    scratch_load_dword v0, off, s0 offset:4 glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_lshl_b32 s0, s2, 2
 ; GFX9-NEXT:    s_addk_i32 s0, 0x104
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX9-NEXT:    scratch_store_dword off, v0, s0
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_and_b32 s0, s2, 15
 ; GFX9-NEXT:    s_lshl_b32 s0, s0, 2

 ; GFX9-PAL:       ; %bb.0: ; %bb
 ; GFX9-PAL-NEXT:    s_getpc_b64 s[2:3]
 ; GFX9-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX9-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
 ; GFX9-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-PAL-NEXT:    s_and_b32 s3, s3, 0xffff
 ; GFX9-PAL-NEXT:    s_add_u32 flat_scratch_lo, s2, s1
 ; GFX9-PAL-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
-; GFX9-PAL-NEXT:    s_mov_b32 s2, 0
-; GFX9-PAL-NEXT:    scratch_load_dword v0, off, s2 offset:4 glc
+; GFX9-PAL-NEXT:    s_mov_b32 s1, 0
+; GFX9-PAL-NEXT:    scratch_load_dword v0, off, s1 offset:4 glc
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    s_lshl_b32 s1, s0, 2
 ; GFX9-PAL-NEXT:    s_and_b32 s0, s0, 15
 ; GFX9-PAL-NEXT:    s_addk_i32 s1, 0x104
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX9-PAL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX9-PAL-NEXT:    scratch_store_dword off, v0, s1
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)

 ; GFX1010-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX1010-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
 ; GFX1010-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1010-PAL-NEXT:    s_and_b32 s3, s3, 0xffff
 ; GFX1010-PAL-NEXT:    s_add_u32 s2, s2, s1
 ; GFX1010-PAL-NEXT:    s_addc_u32 s3, s3, 0
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3
-; GFX1010-PAL-NEXT:    s_mov_b32 s2, 0
-; GFX1010-PAL-NEXT:    s_and_b32 s1, s0, 15
-; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, s2 offset:4 glc dlc
+; GFX1010-PAL-NEXT:    s_mov_b32 s1, 0
+; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, s1 offset:4 glc dlc
 ; GFX1010-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v0, 15
+; GFX1010-PAL-NEXT:    s_and_b32 s1, s0, 15
 ; GFX1010-PAL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX1010-PAL-NEXT:    s_lshl_b32 s1, s1, 2
 ; GFX1010-PAL-NEXT:    s_addk_i32 s0, 0x104
 ; GFX1010-PAL-NEXT:    s_addk_i32 s1, 0x104
 ; GFX1010-PAL-NEXT:    scratch_store_dword off, v0, s0
 ; GFX1010-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, s1 glc dlc
 ; GFX1010-PAL-NEXT:    s_waitcnt vmcnt(0)

 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: store_load_vindex_small_offset_foo:
 ; GFX10:       ; %bb.0: ; %bb
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_and_b32_e32 v1, 15, v0
-; GFX10-NEXT:    s_add_i32 s1, s32, 0x100
 ; GFX10-NEXT:    s_add_i32 s0, s32, 0x100
-; GFX10-NEXT:    v_lshl_add_u32 v0, v0, 2, s1
 ; GFX10-NEXT:    v_mov_b32_e32 v2, 15
-; GFX10-NEXT:    v_lshl_add_u32 v1, v1, 2, s0
+; GFX10-NEXT:    v_lshl_add_u32 v0, v0, 2, s0
+; GFX10-NEXT:    s_add_i32 s0, s32, 0x100
 ; GFX10-NEXT:    scratch_load_dword v3, off, s32 glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    v_lshl_add_u32 v1, v1, 2, s0
 ; GFX10-NEXT:    scratch_store_dword v0, v2, off
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    scratch_load_dword v0, v1, off glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: store_load_vindex_small_offset_foo:
 ; GFX11:       ; %bb.0: ; %bb

 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
 ; GFX940-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-PAL-LABEL: store_load_vindex_small_offset_foo:
 ; GFX10-PAL:       ; %bb.0: ; %bb
 ; GFX10-PAL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-PAL-NEXT:    v_and_b32_e32 v1, 15, v0
-; GFX10-PAL-NEXT:    s_add_i32 s1, s32, 0x100
 ; GFX10-PAL-NEXT:    s_add_i32 s0, s32, 0x100
-; GFX10-PAL-NEXT:    v_lshl_add_u32 v0, v0, 2, s1
 ; GFX10-PAL-NEXT:    v_mov_b32_e32 v2, 15
-; GFX10-PAL-NEXT:    v_lshl_add_u32 v1, v1, 2, s0
+; GFX10-PAL-NEXT:    v_lshl_add_u32 v0, v0, 2, s0
+; GFX10-PAL-NEXT:    s_add_i32 s0, s32, 0x100
 ; GFX10-PAL-NEXT:    scratch_load_dword v3, off, s32 glc dlc
 ; GFX10-PAL-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-PAL-NEXT:    v_lshl_add_u32 v1, v1, 2, s0
 ; GFX10-PAL-NEXT:    scratch_store_dword v0, v2, off
 ; GFX10-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-PAL-NEXT:    scratch_load_dword v0, v1, off glc dlc
 ; GFX10-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-PAL-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-PAL-LABEL: store_load_vindex_small_offset_foo:
 ; GFX11-PAL:       ; %bb.0: ; %bb
bb:
   ret void
 }
 
 define amdgpu_kernel void @zero_init_large_offset_kernel() {
 ; GFX9-LABEL: zero_init_large_offset_kernel:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s0, s3
 ; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
-; GFX9-NEXT:    s_mov_b32 s4, 0
-; GFX9-NEXT:    scratch_load_dword v0, off, s4 offset:4 glc
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_mov_b32 s0, 0
+; GFX9-NEXT:    scratch_load_dword v0, off, s0 offset:4 glc
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_mov_b32 s1, s0
 ; GFX9-NEXT:    s_mov_b32 s2, s0
 ; GFX9-NEXT:    s_mov_b32 s3, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s3
-; GFX9-NEXT:    s_movk_i32 s3, 0x4004
-; GFX9-NEXT:    s_movk_i32 s2, 0x4004
-; GFX9-NEXT:    s_movk_i32 s1, 0x4004
 ; GFX9-NEXT:    s_movk_i32 s0, 0x4004
-; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s3
-; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:16
-; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:32
+; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0
+; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:16
+; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:32
 ; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:48
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: zero_init_large_offset_kernel:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_add_u32 s0, s0, s3
 ; GFX10-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
 ; GFX10-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
 ; GFX10-NEXT:    scratch_load_dword v0, off, off offset:4 glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_mov_b32 s0, 0
 ; GFX10-NEXT:    s_mov_b32 s1, s0
 ; GFX10-NEXT:    s_mov_b32 s2, s0
 ; GFX10-NEXT:    s_mov_b32 s3, s0
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s3
-; GFX10-NEXT:    s_movk_i32 s3, 0x4004
-; GFX10-NEXT:    s_movk_i32 s2, 0x4004
-; GFX10-NEXT:    s_movk_i32 s1, 0x4004
 ; GFX10-NEXT:    s_movk_i32 s0, 0x4004
-; GFX10-NEXT:    scratch_store_dwordx4 off, v[0:3], s3
-; GFX10-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:16
-; GFX10-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:32
+; GFX10-NEXT:    scratch_store_dwordx4 off, v[0:3], s0
+; GFX10-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:16
+; GFX10-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:32
 ; GFX10-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:48
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: zero_init_large_offset_kernel:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    scratch_load_b32 v0, off, off offset:4 glc dlc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_mov_b32 s0, 0
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_mov_b32 s1, s0
 ; GFX11-NEXT:    s_mov_b32 s2, s0
 ; GFX11-NEXT:    s_mov_b32 s3, s0
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
-; GFX11-NEXT:    s_movk_i32 s3, 0x4004
-; GFX11-NEXT:    s_movk_i32 s2, 0x4004
-; GFX11-NEXT:    s_movk_i32 s1, 0x4004
 ; GFX11-NEXT:    s_movk_i32 s0, 0x4004
 ; GFX11-NEXT:    s_clause 0x3
-; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s3
-; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s2 offset:16
-; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s1 offset:32
+; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s0
+; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:16
+; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:32
 ; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:48
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 ;
 ; GFX9-PAL-LABEL: zero_init_large_offset_kernel:
 ; GFX9-PAL:       ; %bb.0:
 ; GFX9-PAL-NEXT:    s_getpc_b64 s[2:3]
 ; GFX9-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX9-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
-; GFX9-PAL-NEXT:    s_mov_b32 s4, 0
 ; GFX9-PAL-NEXT:    s_mov_b32 s0, 0
 ; GFX9-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-PAL-NEXT:    s_and_b32 s3, s3, 0xffff
 ; GFX9-PAL-NEXT:    s_add_u32 flat_scratch_lo, s2, s1
 ; GFX9-PAL-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
-; GFX9-PAL-NEXT:    scratch_load_dword v0, off, s4 offset:4 glc
+; GFX9-PAL-NEXT:    scratch_load_dword v0, off, s0 offset:4 glc
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    s_mov_b32 s1, s0
 ; GFX9-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX9-PAL-NEXT:    s_mov_b32 s3, s0
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v3, s3
-; GFX9-PAL-NEXT:    s_movk_i32 s3, 0x4004
-; GFX9-PAL-NEXT:    s_movk_i32 s2, 0x4004
-; GFX9-PAL-NEXT:    s_movk_i32 s1, 0x4004
 ; GFX9-PAL-NEXT:    s_movk_i32 s0, 0x4004
-; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s3
-; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:16
-; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:32
+; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0
+; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:16
+; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:32
 ; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:48
 ; GFX9-PAL-NEXT:    s_endpgm
 ;
 ; GFX940-LABEL: zero_init_large_offset_kernel:
 ; GFX940:       ; %bb.0:
 ; GFX940-NEXT:    scratch_load_dword v0, off, off offset:4 sc0 sc1
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
 ; GFX940-NEXT:    s_mov_b32 s0, 0
 ; GFX940-NEXT:    s_mov_b32 s1, s0
 ; GFX940-NEXT:    s_mov_b32 s2, s0
 ; GFX940-NEXT:    s_mov_b32 s3, s0
 ; GFX940-NEXT:    v_mov_b64_e32 v[0:1], s[0:1]
 ; GFX940-NEXT:    v_mov_b64_e32 v[2:3], s[2:3]
-; GFX940-NEXT:    s_movk_i32 s3, 0x4004
-; GFX940-NEXT:    s_movk_i32 s2, 0x4004
-; GFX940-NEXT:    s_movk_i32 s1, 0x4004
 ; GFX940-NEXT:    s_movk_i32 s0, 0x4004
-; GFX940-NEXT:    scratch_store_dwordx4 off, v[0:3], s3 sc0 sc1
-; GFX940-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:16 sc0 sc1
-; GFX940-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:32 sc0 sc1
+; GFX940-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 sc0 sc1
+; GFX940-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:16 sc0 sc1
+; GFX940-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:32 sc0 sc1
 ; GFX940-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:48 sc0 sc1
 ; GFX940-NEXT:    s_endpgm
 ;
 ; GFX1010-PAL-LABEL: zero_init_large_offset_kernel:
 ; GFX1010-PAL:       ; %bb.0:
 ; GFX1010-PAL-NEXT:    s_getpc_b64 s[2:3]
 ; GFX1010-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX1010-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
 ; GFX1010-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1010-PAL-NEXT:    s_and_b32 s3, s3, 0xffff
 ; GFX1010-PAL-NEXT:    s_add_u32 s2, s2, s1
 ; GFX1010-PAL-NEXT:    s_addc_u32 s3, s3, 0
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3
-; GFX1010-PAL-NEXT:    s_mov_b32 s4, 0
 ; GFX1010-PAL-NEXT:    s_mov_b32 s0, 0
-; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, s4 offset:4 glc dlc
+; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, s0 offset:4 glc dlc
 ; GFX1010-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1010-PAL-NEXT:    s_mov_b32 s1, s0
 ; GFX1010-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX1010-PAL-NEXT:    s_mov_b32 s3, s0
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v3, s3
-; GFX1010-PAL-NEXT:    s_movk_i32 s3, 0x4004
-; GFX1010-PAL-NEXT:    s_movk_i32 s2, 0x4004
-; GFX1010-PAL-NEXT:    s_movk_i32 s1, 0x4004
 ; GFX1010-PAL-NEXT:    s_movk_i32 s0, 0x4004
-; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s3
-; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:16
-; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:32
+; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0
+; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:16
+; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:32
 ; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:48
 ; GFX1010-PAL-NEXT:    s_endpgm
 ;
 ; GFX1030-PAL-LABEL: zero_init_large_offset_kernel:
 ; GFX1030-PAL:       ; %bb.0:
 ; GFX1030-PAL-NEXT:    s_getpc_b64 s[2:3]
 ; GFX1030-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX1030-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
 ; GFX1030-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1030-PAL-NEXT:    s_and_b32 s3, s3, 0xffff
 ; GFX1030-PAL-NEXT:    s_add_u32 s2, s2, s1
 ; GFX1030-PAL-NEXT:    s_addc_u32 s3, s3, 0
 ; GFX1030-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2
 ; GFX1030-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3
 ; GFX1030-PAL-NEXT:    scratch_load_dword v0, off, off offset:4 glc dlc
 ; GFX1030-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1030-PAL-NEXT:    s_mov_b32 s0, 0
 ; GFX1030-PAL-NEXT:    s_mov_b32 s1, s0
 ; GFX1030-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX1030-PAL-NEXT:    s_mov_b32 s3, s0
 ; GFX1030-PAL-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX1030-PAL-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX1030-PAL-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX1030-PAL-NEXT:    v_mov_b32_e32 v3, s3
-; GFX1030-PAL-NEXT:    s_movk_i32 s3, 0x4004
-; GFX1030-PAL-NEXT:    s_movk_i32 s2, 0x4004
-; GFX1030-PAL-NEXT:    s_movk_i32 s1, 0x4004
 ; GFX1030-PAL-NEXT:    s_movk_i32 s0, 0x4004
-; GFX1030-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s3
-; GFX1030-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:16
-; GFX1030-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:32
+; GFX1030-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0
+; GFX1030-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:16
+; GFX1030-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:32
 ; GFX1030-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:48
 ; GFX1030-PAL-NEXT:    s_endpgm
 ;
 ; GFX11-PAL-LABEL: zero_init_large_offset_kernel:
 ; GFX11-PAL:       ; %bb.0:
 ; GFX11-PAL-NEXT:    scratch_load_b32 v0, off, off offset:4 glc dlc
 ; GFX11-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-PAL-NEXT:    s_mov_b32 s0, 0
 ; GFX11-PAL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-PAL-NEXT:    s_mov_b32 s1, s0
 ; GFX11-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX11-PAL-NEXT:    s_mov_b32 s3, s0
 ; GFX11-PAL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-PAL-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
-; GFX11-PAL-NEXT:    s_movk_i32 s3, 0x4004
-; GFX11-PAL-NEXT:    s_movk_i32 s2, 0x4004
-; GFX11-PAL-NEXT:    s_movk_i32 s1, 0x4004
 ; GFX11-PAL-NEXT:    s_movk_i32 s0, 0x4004
 ; GFX11-PAL-NEXT:    s_clause 0x3
-; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s3
-; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s2 offset:16
-; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s1 offset:32
+; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s0
+; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:16
+; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:32
 ; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:48
 ; GFX11-PAL-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-PAL-NEXT:    s_endpgm
   %padding = alloca [4096 x i32], align 4, addrspace(5)
   %alloca = alloca [32 x i16], align 2, addrspace(5)
   %pad_gep = getelementptr inbounds [4096 x i32], ptr addrspace(5) %padding, i32 0, i32 undef
   %pad_load = load volatile i32, ptr addrspace(5) %pad_gep, align 4
   call void @llvm.memset.p5.i64(ptr addrspace(5) align 2 dereferenceable(64) %alloca, i8 0, i64 64, i1 false)

 ; GFX9-NEXT:    s_mov_b32 s0, 0
 ; GFX9-NEXT:    s_mov_b32 s1, s0
 ; GFX9-NEXT:    s_mov_b32 s2, s0
 ; GFX9-NEXT:    s_mov_b32 s3, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s3
-; GFX9-NEXT:    s_add_i32 s3, s32, 0x4004
-; GFX9-NEXT:    s_add_i32 s2, s32, 0x4004
-; GFX9-NEXT:    s_add_i32 s1, s32, 0x4004
 ; GFX9-NEXT:    s_add_i32 s0, s32, 0x4004
-; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s3
-; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:16
-; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:32
+; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0
+; GFX9-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:16
+; GFX9-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:32
+; GFX9-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:48
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: zero_init_large_offset_foo:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    scratch_load_dword v0, off, s32 offset:4 glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_mov_b32 s0, 0
 ; GFX10-NEXT:    s_mov_b32 s1, s0
 ; GFX10-NEXT:    s_mov_b32 s2, s0
 ; GFX10-NEXT:    s_mov_b32 s3, s0
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s3
-; GFX10-NEXT:    s_add_i32 s3, s32, 0x4004
-; GFX10-NEXT:    s_add_i32 s2, s32, 0x4004
-; GFX10-NEXT:    s_add_i32 s1, s32, 0x4004
 ; GFX10-NEXT:    s_add_i32 s0, s32, 0x4004
-; GFX10-NEXT:    scratch_store_dwordx4 off, v[0:3], s3
-; GFX10-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:16
-; GFX10-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:32
+; GFX10-NEXT:    scratch_store_dwordx4 off, v[0:3], s0
+; GFX10-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX10-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:16
+; GFX10-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX10-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:32
+; GFX10-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX10-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:48
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: zero_init_large_offset_foo:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    scratch_load_b32 v0, off, s32 offset:4 glc dlc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_mov_b32 s0, 0
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_mov_b32 s1, s0
 ; GFX11-NEXT:    s_mov_b32 s2, s0
 ; GFX11-NEXT:    s_mov_b32 s3, s0
 ; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
-; GFX11-NEXT:    s_add_i32 s3, s32, 0x4004
-; GFX11-NEXT:    s_add_i32 s2, s32, 0x4004
-; GFX11-NEXT:    s_add_i32 s1, s32, 0x4004
 ; GFX11-NEXT:    s_add_i32 s0, s32, 0x4004
-; GFX11-NEXT:    s_clause 0x3
-; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s3
-; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s2 offset:16
-; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s1 offset:32
+; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s0
+; GFX11-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:16
+; GFX11-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:32
+; GFX11-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX11-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:48
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-PAL-LABEL: zero_init_large_offset_foo:
 ; GFX9-PAL:       ; %bb.0:
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-PAL-NEXT:    scratch_load_dword v0, off, s32 offset:4 glc
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    s_mov_b32 s0, 0
 ; GFX9-PAL-NEXT:    s_mov_b32 s1, s0
 ; GFX9-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX9-PAL-NEXT:    s_mov_b32 s3, s0
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v3, s3
-; GFX9-PAL-NEXT:    s_add_i32 s3, s32, 0x4004
-; GFX9-PAL-NEXT:    s_add_i32 s2, s32, 0x4004
-; GFX9-PAL-NEXT:    s_add_i32 s1, s32, 0x4004
 ; GFX9-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
-; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s3
-; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:16
-; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:32
+; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0
+; GFX9-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:16
+; GFX9-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:32
+; GFX9-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:48
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX940-LABEL: zero_init_large_offset_foo:
 ; GFX940:       ; %bb.0:
 ; GFX940-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX940-NEXT:    scratch_load_dword v0, off, s32 offset:4 sc0 sc1
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
 ; GFX940-NEXT:    s_mov_b32 s0, 0
 ; GFX940-NEXT:    s_mov_b32 s1, s0
 ; GFX940-NEXT:    s_mov_b32 s2, s0
 ; GFX940-NEXT:    s_mov_b32 s3, s0
 ; GFX940-NEXT:    v_mov_b64_e32 v[0:1], s[0:1]
 ; GFX940-NEXT:    v_mov_b64_e32 v[2:3], s[2:3]
-; GFX940-NEXT:    s_add_i32 s3, s32, 0x4004
-; GFX940-NEXT:    s_add_i32 s2, s32, 0x4004
-; GFX940-NEXT:    s_add_i32 s1, s32, 0x4004
 ; GFX940-NEXT:    s_add_i32 s0, s32, 0x4004
-; GFX940-NEXT:    scratch_store_dwordx4 off, v[0:3], s3 sc0 sc1
-; GFX940-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:16 sc0 sc1
-; GFX940-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:32 sc0 sc1
+; GFX940-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 sc0 sc1
+; GFX940-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX940-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:16 sc0 sc1
+; GFX940-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX940-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:32 sc0 sc1
+; GFX940-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX940-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:48 sc0 sc1
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
 ; GFX940-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX10-PAL-LABEL: zero_init_large_offset_foo:
-; GFX10-PAL:       ; %bb.0:
-; GFX10-PAL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-PAL-NEXT:    scratch_load_dword v0, off, s32 offset:4 glc dlc
-; GFX10-PAL-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-PAL-NEXT:    s_mov_b32 s0, 0
-; GFX10-PAL-NEXT:    s_mov_b32 s1, s0
-; GFX10-PAL-NEXT:    s_mov_b32 s2, s0
-; GFX10-PAL-NEXT:    s_mov_b32 s3, s0
-; GFX10-PAL-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-PAL-NEXT:    v_mov_b32_e32 v1, s1
-; GFX10-PAL-NEXT:    v_mov_b32_e32 v2, s2
-; GFX10-PAL-NEXT:    v_mov_b32_e32 v3, s3
-; GFX10-PAL-NEXT:    s_add_i32 s3, s32, 0x4004
-; GFX10-PAL-NEXT:    s_add_i32 s2, s32, 0x4004
-; GFX10-PAL-NEXT:    s_add_i32 s1, s32, 0x4004
-; GFX10-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
-; GFX10-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s3
-; GFX10-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s2 offset:16
-; GFX10-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:32
-; GFX10-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:48
-; GFX10-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX10-PAL-NEXT:    s_setpc_b64 s[30:31]
+; GFX1010-PAL-LABEL: zero_init_large_offset_foo:
+; GFX1010-PAL:       ; %bb.0:
+; GFX1010-PAL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1010-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, s32 offset:4 glc dlc
+; GFX1010-PAL-NEXT:    s_waitcnt vmcnt(0)
+; GFX1010-PAL-NEXT:    s_mov_b32 s0, 0
+; GFX1010-PAL-NEXT:    s_mov_b32 s1, s0
+; GFX1010-PAL-NEXT:    s_mov_b32 s2, s0
+; GFX1010-PAL-NEXT:    s_mov_b32 s3, s0
+; GFX1010-PAL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX1010-PAL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX1010-PAL-NEXT:    v_mov_b32_e32 v2, s2
+; GFX1010-PAL-NEXT:    v_mov_b32_e32 v3, s3
+; GFX1010-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0
+; GFX1010-PAL-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX1010-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:16
+; GFX1010-PAL-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX1010-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:32
+; GFX1010-PAL-NEXT:    s_waitcnt_depctr 0xffe3
+; GFX1010-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX1010-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:48
+; GFX1010-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX1010-PAL-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1030-PAL-LABEL: zero_init_large_offset_foo:
+; GFX1030-PAL:       ; %bb.0:
+; GFX1030-PAL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1030-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX1030-PAL-NEXT:    scratch_load_dword v0, off, s32 offset:4 glc dlc
+; GFX1030-PAL-NEXT:    s_waitcnt vmcnt(0)
+; GFX1030-PAL-NEXT:    s_mov_b32 s0, 0
+; GFX1030-PAL-NEXT:    s_mov_b32 s1, s0
+; GFX1030-PAL-NEXT:    s_mov_b32 s2, s0
+; GFX1030-PAL-NEXT:    s_mov_b32 s3, s0
+; GFX1030-PAL-NEXT:    v_mov_b32_e32 v0, s0
+; GFX1030-PAL-NEXT:    v_mov_b32_e32 v1, s1
+; GFX1030-PAL-NEXT:    v_mov_b32_e32 v2, s2
+; GFX1030-PAL-NEXT:    v_mov_b32_e32 v3, s3
+; GFX1030-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX1030-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0
+; GFX1030-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX1030-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:16
+; GFX1030-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX1030-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:32
+; GFX1030-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX1030-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:48
+; GFX1030-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX1030-PAL-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-PAL-LABEL: zero_init_large_offset_foo:
 ; GFX11-PAL:       ; %bb.0:
 ; GFX11-PAL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-PAL-NEXT:    scratch_load_b32 v0, off, s32 offset:4 glc dlc
 ; GFX11-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-PAL-NEXT:    s_mov_b32 s0, 0
 ; GFX11-PAL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX11-PAL-NEXT:    s_mov_b32 s1, s0
 ; GFX11-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX11-PAL-NEXT:    s_mov_b32 s3, s0
 ; GFX11-PAL-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-PAL-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
-; GFX11-PAL-NEXT:    s_add_i32 s3, s32, 0x4004
-; GFX11-PAL-NEXT:    s_add_i32 s2, s32, 0x4004
-; GFX11-PAL-NEXT:    s_add_i32 s1, s32, 0x4004
 ; GFX11-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
-; GFX11-PAL-NEXT:    s_clause 0x3
-; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s3
-; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s2 offset:16
-; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s1 offset:32
+; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s0
+; GFX11-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:16
+; GFX11-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:32
+; GFX11-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX11-PAL-NEXT:    scratch_store_b128 off, v[0:3], s0 offset:48
 ; GFX11-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-PAL-NEXT:    s_setpc_b64 s[30:31]
   %padding = alloca [4096 x i32], align 4, addrspace(5)
   %alloca = alloca [32 x i16], align 2, addrspace(5)
   %pad_gep = getelementptr inbounds [4096 x i32], ptr addrspace(5) %padding, i32 0, i32 undef
   %pad_load = load volatile i32, ptr addrspace(5) %pad_gep, align 4
   call void @llvm.memset.p5.i64(ptr addrspace(5) align 2 dereferenceable(64) %alloca, i8 0, i64 64, i1 false)
   ret void
 }
 
 define amdgpu_kernel void @store_load_sindex_large_offset_kernel(i32 %idx) {
 ; GFX9-LABEL: store_load_sindex_large_offset_kernel:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_load_dword s0, s[0:1], 0x24
 ; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s2, s5
 ; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
-; GFX9-NEXT:    s_mov_b32 s2, 0
-; GFX9-NEXT:    scratch_load_dword v0, off, s2 offset:4 glc
+; GFX9-NEXT:    s_mov_b32 s1, 0
+; GFX9-NEXT:    scratch_load_dword v0, off, s1 offset:4 glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    s_lshl_b32 s1, s0, 2
 ; GFX9-NEXT:    s_and_b32 s0, s0, 15
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX9-NEXT:    s_addk_i32 s1, 0x4004
 ; GFX9-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX9-NEXT:    scratch_store_dword off, v0, s1
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)

 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_endpgm
 ;
 ; GFX9-PAL-LABEL: store_load_sindex_large_offset_kernel:
 ; GFX9-PAL:       ; %bb.0: ; %bb
 ; GFX9-PAL-NEXT:    s_getpc_b64 s[4:5]
 ; GFX9-PAL-NEXT:    s_mov_b32 s4, s0
 ; GFX9-PAL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
-; GFX9-PAL-NEXT:    s_mov_b32 s2, 0
-; GFX9-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX9-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-PAL-NEXT:    s_and_b32 s5, s5, 0xffff
+; GFX9-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX9-PAL-NEXT:    s_add_u32 flat_scratch_lo, s4, s3
 ; GFX9-PAL-NEXT:    s_addc_u32 flat_scratch_hi, s5, 0
-; GFX9-PAL-NEXT:    scratch_load_dword v0, off, s2 offset:4 glc
-; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-PAL-NEXT:    s_mov_b32 s1, 0
+; GFX9-PAL-NEXT:    scratch_load_dword v0, off, s1 offset:4 glc
+; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX9-PAL-NEXT:    s_lshl_b32 s1, s0, 2
 ; GFX9-PAL-NEXT:    s_and_b32 s0, s0, 15
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX9-PAL-NEXT:    s_addk_i32 s1, 0x4004
 ; GFX9-PAL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX9-PAL-NEXT:    scratch_store_dword off, v0, s1
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    s_addk_i32 s0, 0x4004

 ; GFX1010-PAL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX1010-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1010-PAL-NEXT:    s_and_b32 s5, s5, 0xffff
 ; GFX1010-PAL-NEXT:    s_add_u32 s4, s4, s3
 ; GFX1010-PAL-NEXT:    s_addc_u32 s5, s5, 0
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s4
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s5
 ; GFX1010-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
-; GFX1010-PAL-NEXT:    s_mov_b32 s2, 0
-; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, s2 offset:4 glc dlc
+; GFX1010-PAL-NEXT:    s_mov_b32 s1, 0
+; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, s1 offset:4 glc dlc
 ; GFX1010-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX1010-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1010-PAL-NEXT:    s_and_b32 s1, s0, 15
 ; GFX1010-PAL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX1010-PAL-NEXT:    s_lshl_b32 s1, s1, 2
 ; GFX1010-PAL-NEXT:    s_addk_i32 s0, 0x4004
 ; GFX1010-PAL-NEXT:    s_addk_i32 s1, 0x4004
bb:
   ret void
 }
 
 define amdgpu_ps void @store_load_sindex_large_offset_foo(i32 inreg %idx) {
 ; GFX9-LABEL: store_load_sindex_large_offset_foo:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s0, s3
 ; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
-; GFX9-NEXT:    s_mov_b32 s1, 0
-; GFX9-NEXT:    scratch_load_dword v0, off, s1 offset:4 glc
+; GFX9-NEXT:    s_mov_b32 s0, 0
+; GFX9-NEXT:    scratch_load_dword v0, off, s0 offset:4 glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_lshl_b32 s0, s2, 2
 ; GFX9-NEXT:    s_addk_i32 s0, 0x4004
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX9-NEXT:    scratch_store_dword off, v0, s0
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_and_b32 s0, s2, 15
 ; GFX9-NEXT:    s_lshl_b32 s0, s0, 2

 ; GFX9-PAL:       ; %bb.0: ; %bb
 ; GFX9-PAL-NEXT:    s_getpc_b64 s[2:3]
 ; GFX9-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX9-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
 ; GFX9-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-PAL-NEXT:    s_and_b32 s3, s3, 0xffff
 ; GFX9-PAL-NEXT:    s_add_u32 flat_scratch_lo, s2, s1
 ; GFX9-PAL-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
-; GFX9-PAL-NEXT:    s_mov_b32 s2, 0
-; GFX9-PAL-NEXT:    scratch_load_dword v0, off, s2 offset:4 glc
+; GFX9-PAL-NEXT:    s_mov_b32 s1, 0
+; GFX9-PAL-NEXT:    scratch_load_dword v0, off, s1 offset:4 glc
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    s_lshl_b32 s1, s0, 2
 ; GFX9-PAL-NEXT:    s_and_b32 s0, s0, 15
 ; GFX9-PAL-NEXT:    s_addk_i32 s1, 0x4004
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX9-PAL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX9-PAL-NEXT:    scratch_store_dword off, v0, s1
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)

 ; GFX1010-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX1010-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
 ; GFX1010-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1010-PAL-NEXT:    s_and_b32 s3, s3, 0xffff
 ; GFX1010-PAL-NEXT:    s_add_u32 s2, s2, s1
 ; GFX1010-PAL-NEXT:    s_addc_u32 s3, s3, 0
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3
-; GFX1010-PAL-NEXT:    s_mov_b32 s2, 0
-; GFX1010-PAL-NEXT:    s_and_b32 s1, s0, 15
-; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, s2 offset:4 glc dlc
+; GFX1010-PAL-NEXT:    s_mov_b32 s1, 0
+; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, s1 offset:4 glc dlc
 ; GFX1010-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v0, 15
+; GFX1010-PAL-NEXT:    s_and_b32 s1, s0, 15
 ; GFX1010-PAL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX1010-PAL-NEXT:    s_lshl_b32 s1, s1, 2
 ; GFX1010-PAL-NEXT:    s_addk_i32 s0, 0x4004
 ; GFX1010-PAL-NEXT:    s_addk_i32 s1, 0x4004
 ; GFX1010-PAL-NEXT:    scratch_store_dword off, v0, s0
 ; GFX1010-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, s1 glc dlc
 ; GFX1010-PAL-NEXT:    s_waitcnt vmcnt(0)

 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: store_load_vindex_large_offset_foo:
 ; GFX10:       ; %bb.0: ; %bb
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_and_b32_e32 v1, 15, v0
-; GFX10-NEXT:    s_add_i32 s1, s32, 0x4004
 ; GFX10-NEXT:    s_add_i32 s0, s32, 0x4004
-; GFX10-NEXT:    v_lshl_add_u32 v0, v0, 2, s1
 ; GFX10-NEXT:    v_mov_b32_e32 v2, 15
-; GFX10-NEXT:    v_lshl_add_u32 v1, v1, 2, s0
+; GFX10-NEXT:    v_lshl_add_u32 v0, v0, 2, s0
+; GFX10-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX10-NEXT:    scratch_load_dword v3, off, s32 offset:4 glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    v_lshl_add_u32 v1, v1, 2, s0
 ; GFX10-NEXT:    scratch_store_dword v0, v2, off
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    scratch_load_dword v0, v1, off glc dlc
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: store_load_vindex_large_offset_foo:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    v_dual_mov_b32 v2, 15 :: v_dual_and_b32 v1, 15, v0
-; GFX11-NEXT:    s_add_i32 s1, s32, 0x4004
 ; GFX11-NEXT:    s_add_i32 s0, s32, 0x4004
-; GFX11-NEXT:    v_lshl_add_u32 v0, v0, 2, s1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT:    v_lshlrev_b32_e32 v1, 2, v1
 ; GFX11-NEXT:    scratch_load_b32 v3, off, s32 offset:4 glc dlc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_lshl_add_u32 v0, v0, 2, s0
+; GFX11-NEXT:    v_lshlrev_b32_e32 v1, 2, v1
+; GFX11-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX11-NEXT:    scratch_store_b32 v0, v2, off dlc
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    scratch_load_b32 v0, v1, s0 glc dlc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-PAL-LABEL: store_load_vindex_large_offset_foo:
 ; GFX9-PAL:       ; %bb.0: ; %bb

 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX940-LABEL: store_load_vindex_large_offset_foo:
 ; GFX940:       ; %bb.0: ; %bb
 ; GFX940-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX940-NEXT:    scratch_load_dword v1, off, s32 offset:4 sc0 sc1
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
-; GFX940-NEXT:    s_add_i32 s1, s32, 0x4004
-; GFX940-NEXT:    v_mov_b32_e32 v1, s1
+; GFX940-NEXT:    s_add_i32 s0, s32, 0x4004
+; GFX940-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX940-NEXT:    v_lshl_add_u32 v1, v0, 2, v1
 ; GFX940-NEXT:    v_mov_b32_e32 v2, 15
 ; GFX940-NEXT:    v_and_b32_e32 v0, 15, v0
 ; GFX940-NEXT:    scratch_store_dword v1, v2, off sc0 sc1
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
 ; GFX940-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX940-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX940-NEXT:    scratch_load_dword v0, v0, s0 sc0 sc1
 ; GFX940-NEXT:    s_waitcnt vmcnt(0)
 ; GFX940-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-PAL-LABEL: store_load_vindex_large_offset_foo:
 ; GFX10-PAL:       ; %bb.0: ; %bb
 ; GFX10-PAL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-PAL-NEXT:    v_and_b32_e32 v1, 15, v0
-; GFX10-PAL-NEXT:    s_add_i32 s1, s32, 0x4004
 ; GFX10-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
-; GFX10-PAL-NEXT:    v_lshl_add_u32 v0, v0, 2, s1
 ; GFX10-PAL-NEXT:    v_mov_b32_e32 v2, 15
-; GFX10-PAL-NEXT:    v_lshl_add_u32 v1, v1, 2, s0
+; GFX10-PAL-NEXT:    v_lshl_add_u32 v0, v0, 2, s0
+; GFX10-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX10-PAL-NEXT:    scratch_load_dword v3, off, s32 offset:4 glc dlc
 ; GFX10-PAL-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-PAL-NEXT:    v_lshl_add_u32 v1, v1, 2, s0
 ; GFX10-PAL-NEXT:    scratch_store_dword v0, v2, off
 ; GFX10-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-PAL-NEXT:    scratch_load_dword v0, v1, off glc dlc
 ; GFX10-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-PAL-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-PAL-LABEL: store_load_vindex_large_offset_foo:
 ; GFX11-PAL:       ; %bb.0: ; %bb
 ; GFX11-PAL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-PAL-NEXT:    v_dual_mov_b32 v2, 15 :: v_dual_and_b32 v1, 15, v0
-; GFX11-PAL-NEXT:    s_add_i32 s1, s32, 0x4004
 ; GFX11-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
-; GFX11-PAL-NEXT:    v_lshl_add_u32 v0, v0, 2, s1
-; GFX11-PAL-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-PAL-NEXT:    v_lshlrev_b32_e32 v1, 2, v1
 ; GFX11-PAL-NEXT:    scratch_load_b32 v3, off, s32 offset:4 glc dlc
 ; GFX11-PAL-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-PAL-NEXT:    v_lshl_add_u32 v0, v0, 2, s0
+; GFX11-PAL-NEXT:    v_lshlrev_b32_e32 v1, 2, v1
+; GFX11-PAL-NEXT:    s_add_i32 s0, s32, 0x4004
 ; GFX11-PAL-NEXT:    scratch_store_b32 v0, v2, off dlc
 ; GFX11-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-PAL-NEXT:    scratch_load_b32 v0, v1, s0 glc dlc
 ; GFX11-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-PAL-NEXT:    s_setpc_b64 s[30:31]
 bb:
   %padding = alloca [4096 x i32], align 4, addrspace(5)
   %i = alloca [32 x float], align 4, addrspace(5)
   %pad_gep = getelementptr inbounds [4096 x i32], ptr addrspace(5) %padding, i32 0, i32 undef
   %pad_load = load volatile i32, ptr addrspace(5) %pad_gep, align 4
   %i7 = getelementptr inbounds [32 x float], ptr addrspace(5) %i, i32 0, i32 %idx
   store volatile i32 15, ptr addrspace(5) %i7, align 4
   %i9 = and i32 %idx, 15
   %i10 = getelementptr inbounds [32 x float], ptr addrspace(5) %i, i32 0, i32 %i9
   %i12 = load volatile i32, ptr addrspace(5) %i10, align 4
   ret void
 }
 
 define amdgpu_kernel void @store_load_large_imm_offset_kernel() {
 ; GFX9-LABEL: store_load_large_imm_offset_kernel:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s0, s3
 ; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 13
-; GFX9-NEXT:    s_mov_b32 s1, 0
-; GFX9-NEXT:    s_movk_i32 s0, 0x3000
-; GFX9-NEXT:    scratch_store_dword off, v0, s1 offset:4
+; GFX9-NEXT:    s_mov_b32 s0, 0
+; GFX9-NEXT:    scratch_store_dword off, v0, s0 offset:4
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_movk_i32 s0, 0x3000
 ; GFX9-NEXT:    s_add_i32 s0, s0, 4
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX9-NEXT:    scratch_store_dword off, v0, s0 offset:3712
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    scratch_load_dword v0, off, s0 offset:3712 glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_endpgm
 ;

 ; GFX11-NEXT:    s_endpgm
 ;
 ; GFX9-PAL-LABEL: store_load_large_imm_offset_kernel:
 ; GFX9-PAL:       ; %bb.0: ; %bb
 ; GFX9-PAL-NEXT:    s_getpc_b64 s[2:3]
 ; GFX9-PAL-NEXT:    s_mov_b32 s2, s0
 ; GFX9-PAL-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v0, 13
-; GFX9-PAL-NEXT:    s_movk_i32 s0, 0x3000
+; GFX9-PAL-NEXT:    s_mov_b32 s0, 0
 ; GFX9-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-PAL-NEXT:    s_and_b32 s3, s3, 0xffff
 ; GFX9-PAL-NEXT:    s_add_u32 flat_scratch_lo, s2, s1
 ; GFX9-PAL-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
-; GFX9-PAL-NEXT:    s_mov_b32 s1, 0
-; GFX9-PAL-NEXT:    scratch_store_dword off, v0, s1 offset:4
+; GFX9-PAL-NEXT:    scratch_store_dword off, v0, s0 offset:4
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-PAL-NEXT:    s_movk_i32 s0, 0x3000
 ; GFX9-PAL-NEXT:    s_add_i32 s0, s0, 4
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX9-PAL-NEXT:    scratch_store_dword off, v0, s0 offset:3712
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    scratch_load_dword v0, off, s0 offset:3712 glc
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    s_endpgm
 ;

 ; GFX9-LABEL: large_offset:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s0, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v2, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v3, v0
-; GFX9-NEXT:    s_mov_b32 s1, 0
-; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:3024
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_mov_b32 s0, 0
+; GFX9-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:3024
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    scratch_load_dwordx4 v[0:3], off, s0 offset:3024 glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 16
 ; GFX9-NEXT:    ;;#ASMSTART
 ; GFX9-NEXT:    ; use v0
 ; GFX9-NEXT:    ;;#ASMEND
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0x810
 ; GFX9-NEXT:    ;;#ASMSTART

 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v2, v0
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v3, v0
 ; GFX9-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-PAL-NEXT:    s_and_b32 s3, s3, 0xffff
 ; GFX9-PAL-NEXT:    s_add_u32 flat_scratch_lo, s2, s0
 ; GFX9-PAL-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
-; GFX9-PAL-NEXT:    s_mov_b32 s1, 0
-; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s1 offset:3024
-; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    s_mov_b32 s0, 0
+; GFX9-PAL-NEXT:    scratch_store_dwordx4 off, v[0:3], s0 offset:3024
+; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    scratch_load_dwordx4 v[0:3], off, s0 offset:3024 glc
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v0, 16
 ; GFX9-PAL-NEXT:    ;;#ASMSTART
 ; GFX9-PAL-NEXT:    ; use v0
 ; GFX9-PAL-NEXT:    ;;#ASMEND
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v0, 0x810
 ; GFX9-PAL-NEXT:    ;;#ASMSTART

llvm/test/CodeGen/AMDGPU/frame-index.mir

machineFunctionInfo:
   stackPtrOffsetReg: '$sgpr32'
 body:             |
   bb.0:
     liveins: $sgpr30_sgpr31
 
     ; GCN-LABEL: name: func_add_constant_to_fi_uniform_SCC_clobber_i32
     ; GCN: liveins: $sgpr30_sgpr31
     ; GCN-NEXT: {{  $}}
-    ; GCN-NEXT: $sgpr1 = S_LSHR_B32 6, $sgpr32, implicit-def dead $scc
-    ; GCN-NEXT: renamable $sgpr4 = nuw S_ADD_U32 killed $sgpr1, 4, implicit-def $scc
+    ; GCN-NEXT: $sgpr0 = S_LSHR_B32 6, $sgpr32, implicit-def dead $scc
+    ; GCN-NEXT: renamable $sgpr4 = nuw S_ADD_U32 killed $sgpr0, 4, implicit-def $scc
     ; GCN-NEXT: renamable $sgpr5 = S_ADDC_U32 $sgpr4, 1234567, implicit-def $scc, implicit $scc
     ; GCN-NEXT: $sgpr0 = S_LSHR_B32 $sgpr32, 6, implicit-def $scc
     ; GCN-NEXT: $sgpr0 = S_ADD_I32 killed $sgpr0, 8, implicit-def $scc
     ; GCN-NEXT: renamable $sgpr6 = S_MUL_I32 killed $sgpr0, $sgpr5
     ; GCN-NEXT: renamable $vgpr0 = COPY killed renamable $sgpr6, implicit $exec
     ; GCN-NEXT: $m0 = S_MOV_B32 -1
     ; GCN-NEXT: DS_WRITE_B32 undef renamable $vgpr0, killed renamable $vgpr0, 0, 0, implicit $m0, implicit $exec
     ; GCN-NEXT: S_SETPC_B64_return killed renamable $sgpr30_sgpr31
machineFunctionInfo:
   stackPtrOffsetReg: '$sgpr32'
 body:             |
   bb.0:
     liveins: $sgpr30_sgpr31, $sgpr10
     ; GCN-LABEL: name: func_add_constant_to_fi_uniform_live_SCC_i32
     ; GCN: liveins: $sgpr10, $sgpr30_sgpr31
     ; GCN-NEXT: {{  $}}
     ; GCN-NEXT: renamable $sgpr4 = nuw S_ADD_U32 $sgpr10, 4, implicit-def $scc
-    ; GCN-NEXT: $vgpr1 = V_LSHRREV_B32_e64 6, $sgpr32, implicit $exec
-    ; GCN-NEXT: $sgpr0 = V_READFIRSTLANE_B32 $vgpr1, implicit $exec
+    ; GCN-NEXT: $vgpr0 = V_LSHRREV_B32_e64 6, $sgpr32, implicit $exec
+    ; GCN-NEXT: $sgpr0 = V_READFIRSTLANE_B32 $vgpr0, implicit $exec
     ; GCN-NEXT: renamable $sgpr5 = S_MUL_I32 killed $sgpr0, $sgpr4
     ; GCN-NEXT: renamable $sgpr6 = S_ADDC_U32 $sgpr5, 1234567, implicit-def dead $scc, implicit $scc
     ; GCN-NEXT: renamable $vgpr0 = COPY killed renamable $sgpr6, implicit $exec
     ; GCN-NEXT: $m0 = S_MOV_B32 -1
     ; GCN-NEXT: DS_WRITE_B32 undef renamable $vgpr0, killed renamable $vgpr0, 0, 0, implicit $m0, implicit $exec
     ; GCN-NEXT: S_SETPC_B64_return killed renamable $sgpr30_sgpr31
     renamable $sgpr4 = nuw S_ADD_U32 $sgpr10, 4, implicit-def $scc
     renamable $sgpr5 = S_MUL_I32 %stack.0, $sgpr4

llvm/test/CodeGen/AMDGPU/frame-setup-without-sgpr-to-vgpr-spills.ll

 ; NO-SPILL-TO-VGPR-LABEL: callee_with_stack_and_call:
 ; NO-SPILL-TO-VGPR:       ; %bb.0:
 ; NO-SPILL-TO-VGPR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; NO-SPILL-TO-VGPR-NEXT:    s_mov_b32 s4, s33
 ; NO-SPILL-TO-VGPR-NEXT:    s_mov_b32 s33, s32
 ; NO-SPILL-TO-VGPR-NEXT:    v_mov_b32_e32 v0, s4
 ; NO-SPILL-TO-VGPR-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
 ; NO-SPILL-TO-VGPR-NEXT:    s_addk_i32 s32, 0x800
-; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 s[10:11], exec
+; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 s[4:5], exec
 ; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 exec, 1
-; NO-SPILL-TO-VGPR-NEXT:    buffer_store_dword v2, off, s[0:3], s33 offset:16
-; NO-SPILL-TO-VGPR-NEXT:    v_writelane_b32 v2, s30, 0
-; NO-SPILL-TO-VGPR-NEXT:    buffer_store_dword v2, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
-; NO-SPILL-TO-VGPR-NEXT:    buffer_load_dword v2, off, s[0:3], s33 offset:16
+; NO-SPILL-TO-VGPR-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:16
+; NO-SPILL-TO-VGPR-NEXT:    v_writelane_b32 v0, s30, 0
+; NO-SPILL-TO-VGPR-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
+; NO-SPILL-TO-VGPR-NEXT:    buffer_load_dword v0, off, s[0:3], s33 offset:16
 ; NO-SPILL-TO-VGPR-NEXT:    s_waitcnt vmcnt(0)
-; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 exec, s[10:11]
-; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 s[8:9], exec
+; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 exec, s[4:5]
+; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 s[4:5], exec
 ; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 exec, 1
-; NO-SPILL-TO-VGPR-NEXT:    buffer_store_dword v1, off, s[0:3], s33 offset:16
-; NO-SPILL-TO-VGPR-NEXT:    v_writelane_b32 v1, s31, 0
-; NO-SPILL-TO-VGPR-NEXT:    buffer_store_dword v1, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
-; NO-SPILL-TO-VGPR-NEXT:    buffer_load_dword v1, off, s[0:3], s33 offset:16
+; NO-SPILL-TO-VGPR-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:16
+; NO-SPILL-TO-VGPR-NEXT:    v_writelane_b32 v0, s31, 0
+; NO-SPILL-TO-VGPR-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
+; NO-SPILL-TO-VGPR-NEXT:    buffer_load_dword v0, off, s[0:3], s33 offset:16
 ; NO-SPILL-TO-VGPR-NEXT:    s_waitcnt vmcnt(0)
-; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 exec, s[8:9]
+; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 exec, s[4:5]
 ; NO-SPILL-TO-VGPR-NEXT:    v_mov_b32_e32 v0, 0
 ; NO-SPILL-TO-VGPR-NEXT:    buffer_store_dword v0, off, s[0:3], s33
 ; NO-SPILL-TO-VGPR-NEXT:    s_waitcnt vmcnt(0)
 ; NO-SPILL-TO-VGPR-NEXT:    s_getpc_b64 s[4:5]
 ; NO-SPILL-TO-VGPR-NEXT:    s_add_u32 s4, s4, external_void_func_void@rel32@lo+4
 ; NO-SPILL-TO-VGPR-NEXT:    s_addc_u32 s5, s5, external_void_func_void@rel32@hi+12
 ; NO-SPILL-TO-VGPR-NEXT:    s_swappc_b64 s[30:31], s[4:5]
-; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 s[8:9], exec
+; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 s[4:5], exec
 ; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 exec, 1
-; NO-SPILL-TO-VGPR-NEXT:    buffer_store_dword v2, off, s[0:3], s33 offset:16
-; NO-SPILL-TO-VGPR-NEXT:    buffer_load_dword v2, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
+; NO-SPILL-TO-VGPR-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:16
+; NO-SPILL-TO-VGPR-NEXT:    buffer_load_dword v0, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
 ; NO-SPILL-TO-VGPR-NEXT:    s_waitcnt vmcnt(0)
-; NO-SPILL-TO-VGPR-NEXT:    v_readlane_b32 s31, v2, 0
-; NO-SPILL-TO-VGPR-NEXT:    buffer_load_dword v2, off, s[0:3], s33 offset:16
+; NO-SPILL-TO-VGPR-NEXT:    v_readlane_b32 s31, v0, 0
+; NO-SPILL-TO-VGPR-NEXT:    buffer_load_dword v0, off, s[0:3], s33 offset:16
 ; NO-SPILL-TO-VGPR-NEXT:    s_waitcnt vmcnt(0)
-; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 exec, s[8:9]
-; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 s[6:7], exec
+; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 exec, s[4:5]
+; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 s[4:5], exec
 ; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 exec, 1
-; NO-SPILL-TO-VGPR-NEXT:    buffer_store_dword v1, off, s[0:3], s33 offset:16
-; NO-SPILL-TO-VGPR-NEXT:    buffer_load_dword v1, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
+; NO-SPILL-TO-VGPR-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:16
+; NO-SPILL-TO-VGPR-NEXT:    buffer_load_dword v0, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
 ; NO-SPILL-TO-VGPR-NEXT:    s_waitcnt vmcnt(0)
-; NO-SPILL-TO-VGPR-NEXT:    v_readlane_b32 s30, v1, 0
-; NO-SPILL-TO-VGPR-NEXT:    buffer_load_dword v1, off, s[0:3], s33 offset:16
+; NO-SPILL-TO-VGPR-NEXT:    v_readlane_b32 s30, v0, 0
+; NO-SPILL-TO-VGPR-NEXT:    buffer_load_dword v0, off, s[0:3], s33 offset:16
 ; NO-SPILL-TO-VGPR-NEXT:    s_waitcnt vmcnt(0)
-; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 exec, s[6:7]
+; NO-SPILL-TO-VGPR-NEXT:    s_mov_b64 exec, s[4:5]
 ; NO-SPILL-TO-VGPR-NEXT:    buffer_load_dword v0, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload
 ; NO-SPILL-TO-VGPR-NEXT:    s_addk_i32 s32, 0xf800
 ; NO-SPILL-TO-VGPR-NEXT:    s_waitcnt vmcnt(0)
 ; NO-SPILL-TO-VGPR-NEXT:    v_readfirstlane_b32 s4, v0
 ; NO-SPILL-TO-VGPR-NEXT:    s_mov_b32 s33, s4
 ; NO-SPILL-TO-VGPR-NEXT:    s_setpc_b64 s[30:31]
   %alloca = alloca i32, addrspace(5)
   store volatile i32 0, ptr addrspace(5) %alloca
   call void @external_void_func_void()
   ret void
 }
 
 attributes #0 = { nounwind "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }

llvm/test/CodeGen/AMDGPU/local-stack-alloc-block-sp-reference.ll

 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; MUBUF-NEXT:    s_endpgm
 ;
 ; FLATSCR-LABEL: local_stack_offset_uses_sp:
 ; FLATSCR:       ; %bb.0: ; %entry
 ; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s2, s5
 ; FLATSCR-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
 ; FLATSCR-NEXT:    v_mov_b32_e32 v0, 0
-; FLATSCR-NEXT:    s_movk_i32 s3, 0x2000
-; FLATSCR-NEXT:    s_mov_b32 s2, 0
-; FLATSCR-NEXT:    scratch_store_dword off, v0, s3
+; FLATSCR-NEXT:    s_movk_i32 s2, 0x2000
+; FLATSCR-NEXT:    scratch_store_dword off, v0, s2
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
+; FLATSCR-NEXT:    s_mov_b32 s2, 0
 ; FLATSCR-NEXT:  .LBB0_1: ; %loadstoreloop
 ; FLATSCR-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; FLATSCR-NEXT:    s_add_i32 s3, s2, 0x3000
 ; FLATSCR-NEXT:    s_add_i32 s2, s2, 1
 ; FLATSCR-NEXT:    s_cmpk_lt_u32 s2, 0x2120
 ; FLATSCR-NEXT:    scratch_store_byte off, v0, s3
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    s_cbranch_scc1 .LBB0_1

 ; FLATSCR-LABEL: func_local_stack_offset_uses_sp:
 ; FLATSCR:       ; %bb.0: ; %entry
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; FLATSCR-NEXT:    s_mov_b32 s2, s33
 ; FLATSCR-NEXT:    s_add_i32 s33, s32, 0x1fff
 ; FLATSCR-NEXT:    s_and_b32 s33, s33, 0xffffe000
 ; FLATSCR-NEXT:    s_add_i32 s32, s32, 0x8000
 ; FLATSCR-NEXT:    v_mov_b32_e32 v2, 0
-; FLATSCR-NEXT:    s_add_i32 s1, s33, 0x2000
-; FLATSCR-NEXT:    s_mov_b32 s0, 0
-; FLATSCR-NEXT:    scratch_store_dword off, v2, s1
+; FLATSCR-NEXT:    s_add_i32 s0, s33, 0x2000
+; FLATSCR-NEXT:    scratch_store_dword off, v2, s0
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
+; FLATSCR-NEXT:    s_mov_b32 s0, 0
 ; FLATSCR-NEXT:  .LBB1_1: ; %loadstoreloop
 ; FLATSCR-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; FLATSCR-NEXT:    s_add_i32 s3, s33, 0x3000
 ; FLATSCR-NEXT:    s_add_i32 s1, s0, s3
 ; FLATSCR-NEXT:    s_add_i32 s0, s0, 1
 ; FLATSCR-NEXT:    s_cmpk_lt_u32 s0, 0x2120
 ; FLATSCR-NEXT:    scratch_store_byte off, v2, s1
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)

 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; MUBUF-NEXT:    v_or_b32_e32 v2, 0x12c4, v0
 ; MUBUF-NEXT:    buffer_load_dword v6, v2, s[0:3], 0 offen glc
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; MUBUF-NEXT:    buffer_load_dword v7, v1, s[0:3], 0 offen glc
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; MUBUF-NEXT:    v_or_b32_e32 v1, 0x12cc, v0
 ; MUBUF-NEXT:    v_or_b32_e32 v0, 0x12c8, v0
-; MUBUF-NEXT:    v_mov_b32_e32 v18, 0x4000
-; MUBUF-NEXT:    v_mov_b32_e32 v17, 0x4000
-; MUBUF-NEXT:    v_mov_b32_e32 v16, 0x4000
+; MUBUF-NEXT:    v_mov_b32_e32 v2, 0x4000
 ; MUBUF-NEXT:    buffer_load_dword v1, v1, s[0:3], 0 offen glc
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
-; MUBUF-NEXT:    v_mov_b32_e32 v15, 0x4000
+; MUBUF-NEXT:    v_mov_b32_e32 v3, 0x4000
 ; MUBUF-NEXT:    buffer_load_dword v0, v0, s[0:3], 0 offen glc
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
-; MUBUF-NEXT:    v_mov_b32_e32 v14, 0x4000
-; MUBUF-NEXT:    buffer_load_dword v8, v18, s[0:3], 0 offen glc
-; MUBUF-NEXT:    s_waitcnt vmcnt(0)
-; MUBUF-NEXT:    buffer_load_dword v9, v17, s[0:3], 0 offen offset:4 glc
+; MUBUF-NEXT:    v_mov_b32_e32 v10, 0x4000
+; MUBUF-NEXT:    buffer_load_dword v8, v2, s[0:3], 0 offen glc
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
-; MUBUF-NEXT:    buffer_load_dword v2, v16, s[0:3], 0 offen offset:8 glc
+; MUBUF-NEXT:    v_mov_b32_e32 v2, 0x4000
+; MUBUF-NEXT:    buffer_load_dword v9, v2, s[0:3], 0 offen offset:4 glc
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
-; MUBUF-NEXT:    buffer_load_dword v3, v15, s[0:3], 0 offen offset:12 glc
+; MUBUF-NEXT:    v_mov_b32_e32 v11, 0x4000
+; MUBUF-NEXT:    buffer_load_dword v2, v3, s[0:3], 0 offen offset:8 glc
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
-; MUBUF-NEXT:    buffer_load_dword v10, v14, s[0:3], 0 offen offset:16 glc
-; MUBUF-NEXT:    s_waitcnt vmcnt(0)
-; MUBUF-NEXT:    v_mov_b32_e32 v13, 0x4000
-; MUBUF-NEXT:    buffer_load_dword v11, v13, s[0:3], 0 offen offset:20 glc
+; MUBUF-NEXT:    v_mov_b32_e32 v12, 0x4000
+; MUBUF-NEXT:    buffer_load_dword v3, v10, s[0:3], 0 offen offset:12 glc
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; MUBUF-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
-; MUBUF-NEXT:    v_mov_b32_e32 v12, 0
+; MUBUF-NEXT:    buffer_load_dword v10, v11, s[0:3], 0 offen offset:16 glc
+; MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; MUBUF-NEXT:    v_add_co_u32_e32 v2, vcc, v0, v2
+; MUBUF-NEXT:    buffer_load_dword v11, v12, s[0:3], 0 offen offset:20 glc
+; MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; MUBUF-NEXT:    v_addc_co_u32_e32 v3, vcc, v1, v3, vcc
 ; MUBUF-NEXT:    v_add_co_u32_e32 v0, vcc, v7, v8
 ; MUBUF-NEXT:    v_addc_co_u32_e32 v1, vcc, v6, v9, vcc
 ; MUBUF-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v10
+; MUBUF-NEXT:    v_mov_b32_e32 v12, 0
 ; MUBUF-NEXT:    v_addc_co_u32_e32 v5, vcc, v5, v11, vcc
 ; MUBUF-NEXT:    s_waitcnt lgkmcnt(0)
 ; MUBUF-NEXT:    global_store_dwordx2 v12, v[4:5], s[4:5] offset:16
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; MUBUF-NEXT:    global_store_dwordx4 v12, v[0:3], s[4:5]
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; MUBUF-NEXT:    s_endpgm
 ;
 ; FLATSCR-LABEL: local_stack_offset_uses_sp_flat:
 ; FLATSCR:       ; %bb.0: ; %entry
 ; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s2, s5
 ; FLATSCR-NEXT:    s_addc_u32 flat_scratch_hi, s3, 0
 ; FLATSCR-NEXT:    v_mov_b32_e32 v0, 0
-; FLATSCR-NEXT:    s_mov_b32 s3, 0
 ; FLATSCR-NEXT:    s_mov_b32 s2, 0
-; FLATSCR-NEXT:    scratch_store_dword off, v0, s3 offset:1024
+; FLATSCR-NEXT:    scratch_store_dword off, v0, s2 offset:1024
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:  .LBB2_1: ; %loadstoreloop
 ; FLATSCR-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; FLATSCR-NEXT:    s_add_i32 s3, s2, 0x2000
 ; FLATSCR-NEXT:    s_add_i32 s2, s2, 1
 ; FLATSCR-NEXT:    s_cmpk_lt_u32 s2, 0x2120
 ; FLATSCR-NEXT:    scratch_store_byte off, v0, s3
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    s_cbranch_scc1 .LBB2_1
 ; FLATSCR-NEXT:  ; %bb.2: ; %split
 ; FLATSCR-NEXT:    s_movk_i32 s2, 0x1000
 ; FLATSCR-NEXT:    s_addk_i32 s2, 0x2000
 ; FLATSCR-NEXT:    scratch_load_dwordx2 v[8:9], off, s2 offset:720 glc
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    scratch_load_dwordx4 v[0:3], off, s2 offset:704 glc
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
-; FLATSCR-NEXT:    s_movk_i32 s3, 0x2000
 ; FLATSCR-NEXT:    s_movk_i32 s2, 0x2000
-; FLATSCR-NEXT:    scratch_load_dwordx2 v[10:11], off, s3 offset:16 glc
+; FLATSCR-NEXT:    scratch_load_dwordx2 v[10:11], off, s2 offset:16 glc
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    scratch_load_dwordx4 v[4:7], off, s2 glc
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; FLATSCR-NEXT:    v_mov_b32_e32 v12, 0
 ; FLATSCR-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v6
 ; FLATSCR-NEXT:    v_addc_co_u32_e32 v3, vcc, v3, v7, vcc
 ; FLATSCR-NEXT:    v_add_co_u32_e32 v0, vcc, v0, v4

llvm/test/CodeGen/AMDGPU/pei-scavenge-vgpr-spill.mir

bb.0:
     ; GFX9-FLATSCR-NEXT: $sgpr33 = frame-setup S_ADD_I32 $sgpr32, 8191, implicit-def $scc
     ; GFX9-FLATSCR-NEXT: $sgpr33 = frame-setup S_AND_B32 killed $sgpr33, 4294959104, implicit-def dead $scc
     ; GFX9-FLATSCR-NEXT: $sgpr6_sgpr7 = S_XOR_SAVEEXEC_B64 -1, implicit-def $exec, implicit-def dead $scc, implicit $exec
     ; GFX9-FLATSCR-NEXT: $sgpr5 = S_ADD_I32 $sgpr33, 16388, implicit-def dead $scc
     ; GFX9-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR $vgpr2, killed $sgpr5, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.3, addrspace 5)
     ; GFX9-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr6_sgpr7
     ; GFX9-FLATSCR-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr4, 0, undef $vgpr2
     ; GFX9-FLATSCR-NEXT: $sgpr32 = frame-setup S_ADD_I32 $sgpr32, 32768, implicit-def dead $scc
-    ; GFX9-FLATSCR-NEXT: $sgpr9 = S_ADD_I32 $sgpr33, 8192, implicit-def $scc
-    ; GFX9-FLATSCR-NEXT: $vgpr0 = V_MOV_B32_e32 killed $sgpr9, implicit $exec
-    ; GFX9-FLATSCR-NEXT: $sgpr8 = S_ADD_I32 $sgpr33, 16384, implicit-def $scc
-    ; GFX9-FLATSCR-NEXT: $vgpr0 = V_OR_B32_e32 killed $sgpr8, $vgpr1, implicit $exec
+    ; GFX9-FLATSCR-NEXT: $sgpr4 = S_ADD_I32 $sgpr33, 8192, implicit-def $scc
+    ; GFX9-FLATSCR-NEXT: $vgpr0 = V_MOV_B32_e32 killed $sgpr4, implicit $exec
+    ; GFX9-FLATSCR-NEXT: $sgpr4 = S_ADD_I32 $sgpr33, 16384, implicit-def $scc
+    ; GFX9-FLATSCR-NEXT: $vgpr0 = V_OR_B32_e32 killed $sgpr4, $vgpr1, implicit $exec
     ; GFX9-FLATSCR-NEXT: $sgpr4 = V_READLANE_B32 $vgpr2, 0
     ; GFX9-FLATSCR-NEXT: $sgpr6_sgpr7 = S_XOR_SAVEEXEC_B64 -1, implicit-def $exec, implicit-def dead $scc, implicit $exec
     ; GFX9-FLATSCR-NEXT: $sgpr5 = S_ADD_I32 $sgpr33, 16388, implicit-def dead $scc
     ; GFX9-FLATSCR-NEXT: $vgpr2 = SCRATCH_LOAD_DWORD_SADDR killed $sgpr5, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.3, addrspace 5)
     ; GFX9-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr6_sgpr7
     ; GFX9-FLATSCR-NEXT: $sgpr32 = frame-destroy S_ADD_I32 $sgpr32, -32768, implicit-def dead $scc
     ; GFX9-FLATSCR-NEXT: $sgpr33 = COPY $sgpr4
     ; GFX9-FLATSCR-NEXT: S_ENDPGM 0, amdgpu_allvgprs
     $vgpr0 = V_MOV_B32_e32 %stack.0, implicit $exec
     $vgpr0 = V_OR_B32_e32 %stack.1, $vgpr1, implicit $exec
     S_ENDPGM 0, amdgpu_allvgprs
 ...

llvm/test/CodeGen/AMDGPU/scratch-simple.ll

 ; GFX10-FLATSCR: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
 ; GFX10-FLATSCR: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
 
 ; GFX9-FLATSCR-PAL-DAG: s_getpc_b64 s[2:3]
 ; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s2, s0
 ; GFX9-FLATSCR-PAL-DAG: s_load_dwordx2 s[2:3], s[2:3], 0x0
 ; GFX9-FLATSCR-PAL-DAG: v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-FLATSCR-PAL-DAG: v_mov_b32_e32 v0, 0xbf20e7f4
-; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s1, 0
+; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s0, 0
 ; GFX9-FLATSCR-PAL-DAG: s_waitcnt lgkmcnt(0)
 ; GFX9-FLATSCR-PAL-DAG: s_and_b32 s3, s3, 0xffff
 ; GFX9-FLATSCR-PAL-DAG: s_add_u32 flat_scratch_lo, s2, s0
 ; GFX9-FLATSCR-PAL-DAG: s_addc_u32 flat_scratch_hi, s3, 0
 ; GFX9-FLATSCR-PAL-DAG: v_and_b32_e32 [[CLAMP_IDX:v[0-9]+]], 0x1fc, v0
 
 ; GFX10-FLATSCR-PAL: s_getpc_b64 s[2:3]
 ; GFX10-FLATSCR-PAL: s_mov_b32 s2, s0

 ; GFX10-FLATSCR: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
 ; GFX10-FLATSCR: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
 
 ; GFX9-FLATSCR-PAL-DAG: s_getpc_b64 s[2:3]
 ; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s2, s0
 ; GFX9-FLATSCR-PAL-DAG: s_load_dwordx2 s[2:3], s[2:3], 0x0
 ; GFX9-FLATSCR-PAL-DAG: v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-FLATSCR-PAL-DAG: v_mov_b32_e32 v0, 0xbf20e7f4
-; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s1, 0
+; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s0, 0
 ; GFX9-FLATSCR-PAL-DAG: s_waitcnt lgkmcnt(0)
 ; GFX9-FLATSCR-PAL-DAG: s_and_b32 s3, s3, 0xffff
 ; GFX9-FLATSCR-PAL-DAG: s_add_u32 flat_scratch_lo, s2, s0
 ; GFX9-FLATSCR-PAL-DAG: s_addc_u32 flat_scratch_hi, s3, 0
 
 ; GFX10-FLATSCR-PAL: s_getpc_b64 s[2:3]
 ; GFX10-FLATSCR-PAL: s_mov_b32 s2, s0
 ; GFX10-FLATSCR-PAL: s_load_dwordx2 s[2:3], s[2:3], 0x0

 ; GFX10-FLATSCR: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
 ; GFX10-FLATSCR: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
 
 ; GFX9-FLATSCR-PAL-DAG: s_getpc_b64 s[2:3]
 ; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s2, s0
 ; GFX9-FLATSCR-PAL-DAG: s_load_dwordx2 s[2:3], s[2:3], 0x10
 ; GFX9-FLATSCR-PAL-DAG: v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-FLATSCR-PAL-DAG: v_mov_b32_e32 v0, 0xbf20e7f4
-; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s1, 0
+; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s0, 0
 ; GFX9-FLATSCR-PAL-DAG: s_waitcnt lgkmcnt(0)
 ; GFX9-FLATSCR-PAL-DAG: s_and_b32 s3, s3, 0xffff
 ; GFX9-FLATSCR-PAL-DAG: s_add_u32 flat_scratch_lo, s2, s0
 ; GFX9-FLATSCR-PAL-DAG: s_addc_u32 flat_scratch_hi, s3, 0
 
 ; GFX10-FLATSCR-PAL: s_getpc_b64 s[2:3]
 ; GFX10-FLATSCR-PAL: s_mov_b32 s2, s0
 ; GFX10-FLATSCR-PAL: s_load_dwordx2 s[2:3], s[2:3], 0x10

 ; GFX10-FLATSCR: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
 ; GFX10-FLATSCR: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
 
 ; GFX9-FLATSCR-PAL-DAG: s_getpc_b64 s[0:1]
 ; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s0, s8
 ; GFX9-FLATSCR-PAL-DAG: s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-FLATSCR-PAL-DAG: v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-FLATSCR-PAL-DAG: v_mov_b32_e32 v0, 0xbf20e7f4
-; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s1, 0
+; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s0, 0
 ; GFX9-FLATSCR-PAL-DAG: s_waitcnt lgkmcnt(0)
 ; GFX9-FLATSCR-PAL-DAG: s_and_b32 s1, s1, 0xffff
 ; GFX9-FLATSCR-PAL-DAG: s_add_u32 flat_scratch_lo, s0, s5
 ; GFX9-FLATSCR-PAL-DAG: s_addc_u32 flat_scratch_hi, s1, 0
 
 ; GFX10-FLATSCR-PAL: s_getpc_b64 s[0:1]
 ; GFX10-FLATSCR-PAL: s_mov_b32 s0, s8
 ; GFX10-FLATSCR-PAL: s_load_dwordx2 s[0:1], s[0:1], 0x0

 ; GFX10-FLATSCR: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
 ; GFX10-FLATSCR: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
 
 ; GFX9-FLATSCR-PAL-DAG: s_getpc_b64 s[0:1]
 ; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s0, s8
 ; GFX9-FLATSCR-PAL-DAG: s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-FLATSCR-PAL-DAG: v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-FLATSCR-PAL-DAG: v_mov_b32_e32 v0, 0xbf20e7f4
-; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s1, 0
+; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s0, 0
 ; GFX9-FLATSCR-PAL-DAG: s_waitcnt lgkmcnt(0)
 ; GFX9-FLATSCR-PAL-DAG: s_and_b32 s1, s1, 0xffff
 ; GFX9-FLATSCR-PAL-DAG: s_add_u32 flat_scratch_lo, s0, s5
 ; GFX9-FLATSCR-PAL-DAG: s_addc_u32 flat_scratch_hi, s1, 0
 
 ; GFX10-FLATSCR-PAL: s_getpc_b64 s[0:1]
 ; GFX10-FLATSCR-PAL: s_mov_b32 s0, s8
 ; GFX10-FLATSCR-PAL: s_load_dwordx2 s[0:1], s[0:1], 0x0

 ; GFX10-FLATSCR: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
 ; GFX10-FLATSCR: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
 
 ; GFX9-FLATSCR-PAL-DAG: s_getpc_b64 s[0:1]
 ; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s0, s8
 ; GFX9-FLATSCR-PAL-DAG: s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-FLATSCR-PAL-DAG: v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-FLATSCR-PAL-DAG: v_mov_b32_e32 v0, 0xbf20e7f4
-; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s1, 0
+; GFX9-FLATSCR-PAL-DAG: s_mov_b32 s0, 0
 ; GFX9-FLATSCR-PAL-DAG: s_waitcnt lgkmcnt(0)
 ; GFX9-FLATSCR-PAL-DAG: s_and_b32 s1, s1, 0xffff
 ; GFX9-FLATSCR-PAL-DAG: s_add_u32 flat_scratch_lo, s0, s5
 ; GFX9-FLATSCR-PAL-DAG: s_addc_u32 flat_scratch_hi, s1, 0
 
 ; GFX10-FLATSCR-PAL: s_getpc_b64 s[0:1]
 ; GFX10-FLATSCR-PAL: s_mov_b32 s0, s8
 ; GFX10-FLATSCR-PAL: s_load_dwordx2 s[0:1], s[0:1], 0x0

llvm/test/CodeGen/AMDGPU/sgpr-spill-to-vmem-scc-clobber.mir

body:             |
   ; VMEM-GFX8: bb.0:
   ; VMEM-GFX8-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
   ; VMEM-GFX8-NEXT:   liveins: $sgpr8_sgpr9
   ; VMEM-GFX8-NEXT: {{  $}}
   ; VMEM-GFX8-NEXT:   S_CMP_EQ_U32 0, 0, implicit-def $scc
   ; VMEM-GFX8-NEXT:   $sgpr4_sgpr5 = S_MOV_B64 $exec
   ; VMEM-GFX8-NEXT:   $exec = S_MOV_B64 3, implicit-def $vgpr0
   ; VMEM-GFX8-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
-  ; VMEM-GFX8-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr8, 0, undef $vgpr0
+  ; VMEM-GFX8-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr8, 0, undef $vgpr0, implicit $sgpr8_sgpr9
   ; VMEM-GFX8-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr9, 1, $vgpr0, implicit $sgpr8_sgpr9
   ; VMEM-GFX8-NEXT:   $vgpr1 = V_MOV_B32_e32 8200, implicit $exec
   ; VMEM-GFX8-NEXT:   BUFFER_STORE_DWORD_OFFEN killed $vgpr0, killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
   ; VMEM-GFX8-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
   ; VMEM-GFX8-NEXT:   $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr0
   ; VMEM-GFX8-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit $scc
   ; VMEM-GFX8-NEXT: {{  $}}
   ; VMEM-GFX8-NEXT: bb.1:
body:             |
   ; VMEM-GFX8: bb.0:
   ; VMEM-GFX8-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
   ; VMEM-GFX8-NEXT:   liveins: $sgpr8_sgpr9, $vgpr248_vgpr249_vgpr250_vgpr251, $vgpr252_vgpr253_vgpr254_vgpr255, $vgpr240_vgpr241_vgpr242_vgpr243_vgpr244_vgpr245_vgpr246_vgpr247, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47, $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71_vgpr72_vgpr73_vgpr74_vgpr75_vgpr76_vgpr77_vgpr78_vgpr79, $vgpr80_vgpr81_vgpr82_vgpr83_vgpr84_vgpr85_vgpr86_vgpr87_vgpr88_vgpr89_vgpr90_vgpr91_vgpr92_vgpr93_vgpr94_vgpr95, $vgpr96_vgpr97_vgpr98_vgpr99_vgpr100_vgpr101_vgpr102_vgpr103_vgpr104_vgpr105_vgpr106_vgpr107_vgpr108_vgpr109_vgpr110_vgpr111, $vgpr112_vgpr113_vgpr114_vgpr115_vgpr116_vgpr117_vgpr118_vgpr119_vgpr120_vgpr121_vgpr122_vgpr123_vgpr124_vgpr125_vgpr126_vgpr127, $vgpr128_vgpr129_vgpr130_vgpr131_vgpr132_vgpr133_vgpr134_vgpr135_vgpr136_vgpr137_vgpr138_vgpr139_vgpr140_vgpr141_vgpr142_vgpr143, $vgpr144_vgpr145_vgpr146_vgpr147_vgpr148_vgpr149_vgpr150_vgpr151_vgpr152_vgpr153_vgpr154_vgpr155_vgpr156_vgpr157_vgpr158_vgpr159, $vgpr160_vgpr161_vgpr162_vgpr163_vgpr164_vgpr165_vgpr166_vgpr167_vgpr168_vgpr169_vgpr170_vgpr171_vgpr172_vgpr173_vgpr174_vgpr175, $vgpr176_vgpr177_vgpr178_vgpr179_vgpr180_vgpr181_vgpr182_vgpr183_vgpr184_vgpr185_vgpr186_vgpr187_vgpr188_vgpr189_vgpr190_vgpr191, $vgpr192_vgpr193_vgpr194_vgpr195_vgpr196_vgpr197_vgpr198_vgpr199_vgpr200_vgpr201_vgpr202_vgpr203_vgpr204_vgpr205_vgpr206_vgpr207, $vgpr208_vgpr209_vgpr210_vgpr211_vgpr212_vgpr213_vgpr214_vgpr215_vgpr216_vgpr217_vgpr218_vgpr219_vgpr220_vgpr221_vgpr222_vgpr223, $vgpr224_vgpr225_vgpr226_vgpr227_vgpr228_vgpr229_vgpr230_vgpr231_vgpr232_vgpr233_vgpr234_vgpr235_vgpr236_vgpr237_vgpr238_vgpr239
   ; VMEM-GFX8-NEXT: {{  $}}
   ; VMEM-GFX8-NEXT:   S_CMP_EQ_U32 0, 0, implicit-def $scc
   ; VMEM-GFX8-NEXT:   $sgpr4_sgpr5 = S_MOV_B64 $exec
   ; VMEM-GFX8-NEXT:   $exec = S_MOV_B64 3
   ; VMEM-GFX8-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
-  ; VMEM-GFX8-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr8, 0, undef $vgpr0
+  ; VMEM-GFX8-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr8, 0, undef $vgpr0, implicit $sgpr8_sgpr9
   ; VMEM-GFX8-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr9, 1, $vgpr0, implicit $sgpr8_sgpr9
   ; VMEM-GFX8-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (store (s32) into %stack.3, addrspace 5)
   ; VMEM-GFX8-NEXT:   $vgpr1 = V_MOV_B32_e32 8200, implicit $exec
   ; VMEM-GFX8-NEXT:   BUFFER_STORE_DWORD_OFFEN killed $vgpr0, killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
   ; VMEM-GFX8-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
   ; VMEM-GFX8-NEXT:   $exec = S_MOV_B64 killed $sgpr4_sgpr5
   ; VMEM-GFX8-NEXT:   $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (load (s32) from %stack.3, addrspace 5)
   ; VMEM-GFX8-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit $scc

 
 body:             |
   ; VMEM-GFX8-LABEL: name: sgpr32_save_clobber_scc_emergency_stack_slot_x2
   ; VMEM-GFX8: bb.0:
   ; VMEM-GFX8-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
   ; VMEM-GFX8-NEXT:   liveins: $sgpr8, $sgpr9, $vgpr248_vgpr249_vgpr250_vgpr251, $vgpr252_vgpr253_vgpr254_vgpr255, $vgpr240_vgpr241_vgpr242_vgpr243_vgpr244_vgpr245_vgpr246_vgpr247, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47, $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71_vgpr72_vgpr73_vgpr74_vgpr75_vgpr76_vgpr77_vgpr78_vgpr79, $vgpr80_vgpr81_vgpr82_vgpr83_vgpr84_vgpr85_vgpr86_vgpr87_vgpr88_vgpr89_vgpr90_vgpr91_vgpr92_vgpr93_vgpr94_vgpr95, $vgpr96_vgpr97_vgpr98_vgpr99_vgpr100_vgpr101_vgpr102_vgpr103_vgpr104_vgpr105_vgpr106_vgpr107_vgpr108_vgpr109_vgpr110_vgpr111, $vgpr112_vgpr113_vgpr114_vgpr115_vgpr116_vgpr117_vgpr118_vgpr119_vgpr120_vgpr121_vgpr122_vgpr123_vgpr124_vgpr125_vgpr126_vgpr127, $vgpr128_vgpr129_vgpr130_vgpr131_vgpr132_vgpr133_vgpr134_vgpr135_vgpr136_vgpr137_vgpr138_vgpr139_vgpr140_vgpr141_vgpr142_vgpr143, $vgpr144_vgpr145_vgpr146_vgpr147_vgpr148_vgpr149_vgpr150_vgpr151_vgpr152_vgpr153_vgpr154_vgpr155_vgpr156_vgpr157_vgpr158_vgpr159, $vgpr160_vgpr161_vgpr162_vgpr163_vgpr164_vgpr165_vgpr166_vgpr167_vgpr168_vgpr169_vgpr170_vgpr171_vgpr172_vgpr173_vgpr174_vgpr175, $vgpr176_vgpr177_vgpr178_vgpr179_vgpr180_vgpr181_vgpr182_vgpr183_vgpr184_vgpr185_vgpr186_vgpr187_vgpr188_vgpr189_vgpr190_vgpr191, $vgpr192_vgpr193_vgpr194_vgpr195_vgpr196_vgpr197_vgpr198_vgpr199_vgpr200_vgpr201_vgpr202_vgpr203_vgpr204_vgpr205_vgpr206_vgpr207, $vgpr208_vgpr209_vgpr210_vgpr211_vgpr212_vgpr213_vgpr214_vgpr215_vgpr216_vgpr217_vgpr218_vgpr219_vgpr220_vgpr221_vgpr222_vgpr223, $vgpr224_vgpr225_vgpr226_vgpr227_vgpr228_vgpr229_vgpr230_vgpr231_vgpr232_vgpr233_vgpr234_vgpr235_vgpr236_vgpr237_vgpr238_vgpr239
   ; VMEM-GFX8-NEXT: {{  $}}
   ; VMEM-GFX8-NEXT:   S_CMP_EQ_U32 0, 0, implicit-def $scc
-  ; VMEM-GFX8-NEXT:   $sgpr6_sgpr7 = S_MOV_B64 $exec
+  ; VMEM-GFX8-NEXT:   $sgpr4_sgpr5 = S_MOV_B64 $exec
   ; VMEM-GFX8-NEXT:   $exec = S_MOV_B64 1
   ; VMEM-GFX8-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.3, addrspace 5)
   ; VMEM-GFX8-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr8, 0, undef $vgpr0
   ; VMEM-GFX8-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (store (s32) into %stack.4, addrspace 5)
   ; VMEM-GFX8-NEXT:   $vgpr1 = V_MOV_B32_e32 8200, implicit $exec
   ; VMEM-GFX8-NEXT:   BUFFER_STORE_DWORD_OFFEN killed $vgpr0, killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
   ; VMEM-GFX8-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.3, addrspace 5)
-  ; VMEM-GFX8-NEXT:   $exec = S_MOV_B64 killed $sgpr6_sgpr7
+  ; VMEM-GFX8-NEXT:   $exec = S_MOV_B64 killed $sgpr4_sgpr5
   ; VMEM-GFX8-NEXT:   $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (load (s32) from %stack.4, addrspace 5)
   ; VMEM-GFX8-NEXT:   $sgpr4_sgpr5 = S_MOV_B64 $exec
   ; VMEM-GFX8-NEXT:   $exec = S_MOV_B64 1
   ; VMEM-GFX8-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.3, addrspace 5)
   ; VMEM-GFX8-NEXT:   $vgpr0 = V_WRITELANE_B32 $sgpr9, 0, undef $vgpr0
   ; VMEM-GFX8-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (store (s32) into %stack.4, addrspace 5)
   ; VMEM-GFX8-NEXT:   $vgpr1 = V_MOV_B32_e32 8200, implicit $exec
   ; VMEM-GFX8-NEXT:   BUFFER_STORE_DWORD_OFFEN killed $vgpr0, killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)

 
 body:             |
   ; VMEM-GFX8-LABEL: name: sgpr32_restore_clobber_scc_emergency_stack_slot_x2
   ; VMEM-GFX8: bb.0:
   ; VMEM-GFX8-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
   ; VMEM-GFX8-NEXT:   liveins: $vgpr248_vgpr249_vgpr250_vgpr251, $vgpr252_vgpr253_vgpr254_vgpr255, $vgpr240_vgpr241_vgpr242_vgpr243_vgpr244_vgpr245_vgpr246_vgpr247, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47, $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71_vgpr72_vgpr73_vgpr74_vgpr75_vgpr76_vgpr77_vgpr78_vgpr79, $vgpr80_vgpr81_vgpr82_vgpr83_vgpr84_vgpr85_vgpr86_vgpr87_vgpr88_vgpr89_vgpr90_vgpr91_vgpr92_vgpr93_vgpr94_vgpr95, $vgpr96_vgpr97_vgpr98_vgpr99_vgpr100_vgpr101_vgpr102_vgpr103_vgpr104_vgpr105_vgpr106_vgpr107_vgpr108_vgpr109_vgpr110_vgpr111, $vgpr112_vgpr113_vgpr114_vgpr115_vgpr116_vgpr117_vgpr118_vgpr119_vgpr120_vgpr121_vgpr122_vgpr123_vgpr124_vgpr125_vgpr126_vgpr127, $vgpr128_vgpr129_vgpr130_vgpr131_vgpr132_vgpr133_vgpr134_vgpr135_vgpr136_vgpr137_vgpr138_vgpr139_vgpr140_vgpr141_vgpr142_vgpr143, $vgpr144_vgpr145_vgpr146_vgpr147_vgpr148_vgpr149_vgpr150_vgpr151_vgpr152_vgpr153_vgpr154_vgpr155_vgpr156_vgpr157_vgpr158_vgpr159, $vgpr160_vgpr161_vgpr162_vgpr163_vgpr164_vgpr165_vgpr166_vgpr167_vgpr168_vgpr169_vgpr170_vgpr171_vgpr172_vgpr173_vgpr174_vgpr175, $vgpr176_vgpr177_vgpr178_vgpr179_vgpr180_vgpr181_vgpr182_vgpr183_vgpr184_vgpr185_vgpr186_vgpr187_vgpr188_vgpr189_vgpr190_vgpr191, $vgpr192_vgpr193_vgpr194_vgpr195_vgpr196_vgpr197_vgpr198_vgpr199_vgpr200_vgpr201_vgpr202_vgpr203_vgpr204_vgpr205_vgpr206_vgpr207, $vgpr208_vgpr209_vgpr210_vgpr211_vgpr212_vgpr213_vgpr214_vgpr215_vgpr216_vgpr217_vgpr218_vgpr219_vgpr220_vgpr221_vgpr222_vgpr223, $vgpr224_vgpr225_vgpr226_vgpr227_vgpr228_vgpr229_vgpr230_vgpr231_vgpr232_vgpr233_vgpr234_vgpr235_vgpr236_vgpr237_vgpr238_vgpr239
   ; VMEM-GFX8-NEXT: {{  $}}
   ; VMEM-GFX8-NEXT:   S_CMP_EQ_U32 0, 0, implicit-def $scc
-  ; VMEM-GFX8-NEXT:   $sgpr6_sgpr7 = S_MOV_B64 $exec
+  ; VMEM-GFX8-NEXT:   $sgpr4_sgpr5 = S_MOV_B64 $exec
   ; VMEM-GFX8-NEXT:   $exec = S_MOV_B64 1
   ; VMEM-GFX8-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.3, addrspace 5)
   ; VMEM-GFX8-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (store (s32) into %stack.4, addrspace 5)
   ; VMEM-GFX8-NEXT:   $vgpr1 = V_MOV_B32_e32 8200, implicit $exec
   ; VMEM-GFX8-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFEN killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.1, addrspace 5)
   ; VMEM-GFX8-NEXT:   $sgpr8 = V_READLANE_B32 killed $vgpr0, 0
   ; VMEM-GFX8-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.3, addrspace 5)
-  ; VMEM-GFX8-NEXT:   $exec = S_MOV_B64 killed $sgpr6_sgpr7
+  ; VMEM-GFX8-NEXT:   $exec = S_MOV_B64 killed $sgpr4_sgpr5
   ; VMEM-GFX8-NEXT:   $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (load (s32) from %stack.4, addrspace 5)
   ; VMEM-GFX8-NEXT:   $sgpr4_sgpr5 = S_MOV_B64 $exec
   ; VMEM-GFX8-NEXT:   $exec = S_MOV_B64 1
   ; VMEM-GFX8-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.3, addrspace 5)
   ; VMEM-GFX8-NEXT:   BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (store (s32) into %stack.4, addrspace 5)
   ; VMEM-GFX8-NEXT:   $vgpr1 = V_MOV_B32_e32 16392, implicit $exec
   ; VMEM-GFX8-NEXT:   $vgpr0 = BUFFER_LOAD_DWORD_OFFEN killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
   ; VMEM-GFX8-NEXT:   $sgpr9 = V_READLANE_B32 killed $vgpr0, 0

llvm/test/CodeGen/AMDGPU/sgpr-spill-vmem-large-frame.mir

machineFunctionInfo:
   hasSpilledSGPRs: true
 body:             |
   bb.0:
     liveins: $sgpr30_sgpr31, $sgpr10, $sgpr11
     ; CHECK-LABEL: name: test
     ; CHECK: liveins: $sgpr10, $sgpr11, $sgpr30_sgpr31
     ; CHECK-NEXT: {{  $}}
     ; CHECK-NEXT: S_CMP_EQ_U32 0, 0, implicit-def $scc
-    ; CHECK-NEXT: $sgpr6_sgpr7 = S_MOV_B64 $exec
-    ; CHECK-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr2
-    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
-    ; CHECK-NEXT: $vgpr2 = V_WRITELANE_B32 killed $sgpr10, 0, undef $vgpr2
-    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
-    ; CHECK-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
-    ; CHECK-NEXT: $exec = S_MOV_B64 killed $sgpr6_sgpr7, implicit killed $vgpr2
+    ; CHECK-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
+    ; CHECK-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr1
+    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
+    ; CHECK-NEXT: $vgpr1 = V_WRITELANE_B32 killed $sgpr10, 0, undef $vgpr1
+    ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
+    ; CHECK-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
+    ; CHECK-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr1
     ; CHECK-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
     ; CHECK-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr1
     ; CHECK-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
     ; CHECK-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
     ; CHECK-NEXT: $sgpr10 = V_READLANE_B32 killed $vgpr1, 0
     ; CHECK-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
     ; CHECK-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr1
     ; CHECK-NEXT: S_SETPC_B64 $sgpr30_sgpr31, implicit $scc
     S_CMP_EQ_U32 0, 0, implicit-def $scc
     SI_SPILL_S32_SAVE killed $sgpr10, %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
     renamable $sgpr10 = SI_SPILL_S32_RESTORE %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
     S_SETPC_B64 $sgpr30_sgpr31, implicit $scc
 ...

llvm/test/CodeGen/AMDGPU/sgpr-spill.mir

bb.0:
     ; GCN64-MUBUF-NEXT: $sgpr33 = S_MOV_B32 0
     ; GCN64-MUBUF-NEXT: $sgpr28 = S_MOV_B32 &SCRATCH_RSRC_DWORD0, implicit-def $sgpr28_sgpr29_sgpr30_sgpr31
     ; GCN64-MUBUF-NEXT: $sgpr29 = S_MOV_B32 &SCRATCH_RSRC_DWORD1, implicit-def $sgpr28_sgpr29_sgpr30_sgpr31
     ; GCN64-MUBUF-NEXT: $sgpr30 = S_MOV_B32 4294967295, implicit-def $sgpr28_sgpr29_sgpr30_sgpr31
     ; GCN64-MUBUF-NEXT: $sgpr31 = S_MOV_B32 14680064, implicit-def $sgpr28_sgpr29_sgpr30_sgpr31
     ; GCN64-MUBUF-NEXT: $sgpr28 = S_ADD_U32 $sgpr28, $sgpr11, implicit-def $scc, implicit-def $sgpr28_sgpr29_sgpr30_sgpr31
     ; GCN64-MUBUF-NEXT: $sgpr29 = S_ADDC_U32 $sgpr29, 0, implicit-def dead $scc, implicit $scc, implicit-def $sgpr28_sgpr29_sgpr30_sgpr31
     ; GCN64-MUBUF-NEXT: renamable $sgpr12 = IMPLICIT_DEF
-    ; GCN64-MUBUF-NEXT: $sgpr6_sgpr7 = S_MOV_B64 $exec
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr1
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 killed $sgpr12, 0, undef $vgpr1
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr6_sgpr7, implicit killed $vgpr1
+    ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr0
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 killed $sgpr12, 0, undef $vgpr0
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-MUBUF-NEXT: renamable $sgpr12 = IMPLICIT_DEF
     ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr0
     ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0
     ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
     ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-MUBUF-NEXT: renamable $sgpr12_sgpr13 = IMPLICIT_DEF
-    ; GCN64-MUBUF-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr2
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr2, implicit $sgpr12_sgpr13
-    ; GCN64-MUBUF-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr13, 1, $vgpr2, implicit killed $sgpr12_sgpr13
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 8, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr2
+    ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit killed $sgpr12_sgpr13
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 8, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-MUBUF-NEXT: renamable $sgpr12_sgpr13 = IMPLICIT_DEF
-    ; GCN64-MUBUF-NEXT: $sgpr2_sgpr3 = S_MOV_B64 $exec
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr3
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr3, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr3 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr3, implicit $sgpr12_sgpr13
-    ; GCN64-MUBUF-NEXT: $vgpr3 = V_WRITELANE_B32 $sgpr13, 1, $vgpr3, implicit $sgpr12_sgpr13
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr3, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 8, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr3 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr2_sgpr3, implicit killed $vgpr3
+    ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 8, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-MUBUF-NEXT: renamable $sgpr12_sgpr13_sgpr14 = IMPLICIT_DEF
-    ; GCN64-MUBUF-NEXT: $sgpr6_sgpr7 = S_MOV_B64 $exec
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 7, implicit-def $vgpr1
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr1, implicit $sgpr12_sgpr13_sgpr14
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr13, 1, $vgpr1, implicit $sgpr12_sgpr13_sgpr14
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr14, 2, $vgpr1, implicit killed $sgpr12_sgpr13_sgpr14
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 16, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr6_sgpr7, implicit killed $vgpr1
+    ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 7, implicit-def $vgpr0
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13_sgpr14
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13_sgpr14
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr14, 2, $vgpr0, implicit killed $sgpr12_sgpr13_sgpr14
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 16, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-MUBUF-NEXT: renamable $sgpr12_sgpr13_sgpr14_sgpr15 = IMPLICIT_DEF
     ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 15, implicit-def $vgpr0
     ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr14, 2, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr15, 3, $vgpr0, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15
     ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 28, 0, 0, implicit $exec :: (store (s32) into %stack.3, addrspace 5)
     ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-MUBUF-NEXT: renamable $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16 = IMPLICIT_DEF
-    ; GCN64-MUBUF-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 31, implicit-def $vgpr2
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr2, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN64-MUBUF-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr13, 1, $vgpr2, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN64-MUBUF-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr14, 2, $vgpr2, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN64-MUBUF-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr15, 3, $vgpr2, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN64-MUBUF-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr16, 4, $vgpr2, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 44, 0, 0, implicit $exec :: (store (s32) into %stack.4, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr2
+    ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 31, implicit-def $vgpr0
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr14, 2, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr15, 3, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr16, 4, $vgpr0, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 44, 0, 0, implicit $exec :: (store (s32) into %stack.4, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-MUBUF-NEXT: renamable $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = IMPLICIT_DEF
-    ; GCN64-MUBUF-NEXT: $sgpr2_sgpr3 = S_MOV_B64 $exec
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 255, implicit-def $vgpr1
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr13, 1, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr14, 2, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr15, 3, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr16, 4, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr17, 5, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr18, 6, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr19, 7, $vgpr1, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 64, 0, 0, implicit $exec :: (store (s32) into %stack.5, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr2_sgpr3, implicit killed $vgpr1
+    ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 255, implicit-def $vgpr0
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr14, 2, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr15, 3, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr16, 4, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr17, 5, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr18, 6, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr19, 7, $vgpr0, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 64, 0, 0, implicit $exec :: (store (s32) into %stack.5, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-MUBUF-NEXT: renamable $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27 = IMPLICIT_DEF
     ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 65535, implicit-def $vgpr0
     ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr14, 2, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr15, 3, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr16, 4, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr17, 5, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr18, 6, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr19, 7, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr20, 8, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr21, 9, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr22, 10, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr23, 11, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr24, 12, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr25, 13, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr26, 14, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr27, 15, $vgpr0, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 96, 0, 0, implicit $exec :: (store (s32) into %stack.6, addrspace 5)
     ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-MUBUF-NEXT: renamable $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95 = IMPLICIT_DEF
-    ; GCN64-MUBUF-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 4294967295, implicit-def $vgpr1
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr64, 0, undef $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr65, 1, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr66, 2, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr67, 3, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr68, 4, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr69, 5, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr70, 6, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr71, 7, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr72, 8, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr73, 9, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr74, 10, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr75, 11, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr76, 12, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr77, 13, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr78, 14, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr79, 15, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr80, 16, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr81, 17, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr82, 18, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr83, 19, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr84, 20, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr85, 21, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr86, 22, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr87, 23, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr88, 24, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr89, 25, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr90, 26, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr91, 27, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr92, 28, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr93, 29, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr94, 30, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr95, 31, $vgpr1, implicit killed $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 160, 0, 0, implicit $exec :: (store (s32) into %stack.7, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr1
+    ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 4294967295, implicit-def $vgpr0
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr64, 0, undef $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr65, 1, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr66, 2, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr67, 3, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr68, 4, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr69, 5, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr70, 6, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr71, 7, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr72, 8, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr73, 9, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr74, 10, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr75, 11, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr76, 12, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr77, 13, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr78, 14, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr79, 15, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr80, 16, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr81, 17, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr82, 18, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr83, 19, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr84, 20, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr85, 21, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr86, 22, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr87, 23, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr88, 24, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr89, 25, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr90, 26, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr91, 27, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr92, 28, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr93, 29, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr94, 30, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr95, 31, $vgpr0, implicit killed $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 160, 0, 0, implicit $exec :: (store (s32) into %stack.7, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-MUBUF-NEXT: renamable $sgpr12 = IMPLICIT_DEF
     ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr0
     ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0
     ; GCN64-MUBUF-NEXT: $sgpr2 = S_ADD_I32 $sgpr33, 262144, implicit-def dead $scc
     ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, killed $sgpr2, 0, 0, 0, implicit $exec :: (store (s32) into %stack.8, align 4096, addrspace 5)
     ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN32-MUBUF-LABEL: name: check_spill
     ; GCN32-MUBUF: liveins: $sgpr8, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr11
     ; GCN32-MUBUF-NEXT: {{  $}}
     ; GCN32-MUBUF-NEXT: $sgpr33 = S_MOV_B32 0
     ; GCN32-MUBUF-NEXT: $sgpr96 = S_MOV_B32 &SCRATCH_RSRC_DWORD0, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GCN32-MUBUF-NEXT: $sgpr97 = S_MOV_B32 &SCRATCH_RSRC_DWORD1, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GCN32-MUBUF-NEXT: $sgpr98 = S_MOV_B32 4294967295, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GCN32-MUBUF-NEXT: $sgpr99 = S_MOV_B32 834756608, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GCN32-MUBUF-NEXT: $sgpr96 = S_ADD_U32 $sgpr96, $sgpr11, implicit-def $scc, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GCN32-MUBUF-NEXT: $sgpr97 = S_ADDC_U32 $sgpr97, 0, implicit-def dead $scc, implicit $scc, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GCN32-MUBUF-NEXT: renamable $sgpr12 = IMPLICIT_DEF
-    ; GCN32-MUBUF-NEXT: $sgpr3 = S_MOV_B32 $exec_lo
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 1, implicit-def $vgpr1
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 killed $sgpr12, 0, undef $vgpr1
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr3, implicit killed $vgpr1
+    ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 1, implicit-def $vgpr0
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 killed $sgpr12, 0, undef $vgpr0
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
     ; GCN32-MUBUF-NEXT: renamable $sgpr12 = IMPLICIT_DEF
     ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
     ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 1, implicit-def $vgpr0
     ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0
     ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
     ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
     ; GCN32-MUBUF-NEXT: renamable $sgpr12_sgpr13 = IMPLICIT_DEF
-    ; GCN32-MUBUF-NEXT: $sgpr2 = S_MOV_B32 $exec_lo
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 3, implicit-def $vgpr2
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr2, implicit $sgpr12_sgpr13
-    ; GCN32-MUBUF-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr13, 1, $vgpr2, implicit killed $sgpr12_sgpr13
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 8, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr2, implicit killed $vgpr2
+    ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 3, implicit-def $vgpr0
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit killed $sgpr12_sgpr13
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 8, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
     ; GCN32-MUBUF-NEXT: renamable $sgpr12_sgpr13 = IMPLICIT_DEF
-    ; GCN32-MUBUF-NEXT: $sgpr1 = S_MOV_B32 $exec_lo
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 3, implicit-def $vgpr3
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr3, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr3 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr3, implicit $sgpr12_sgpr13
-    ; GCN32-MUBUF-NEXT: $vgpr3 = V_WRITELANE_B32 $sgpr13, 1, $vgpr3, implicit $sgpr12_sgpr13
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr3, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 8, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr3 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr1, implicit killed $vgpr3
+    ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 3, implicit-def $vgpr0
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 8, 0, 0, implicit $exec :: (store (s32) into %stack.1, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
     ; GCN32-MUBUF-NEXT: renamable $sgpr12_sgpr13_sgpr14 = IMPLICIT_DEF
-    ; GCN32-MUBUF-NEXT: $sgpr3 = S_MOV_B32 $exec_lo
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 7, implicit-def $vgpr1
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr1, implicit $sgpr12_sgpr13_sgpr14
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr13, 1, $vgpr1, implicit $sgpr12_sgpr13_sgpr14
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr14, 2, $vgpr1, implicit killed $sgpr12_sgpr13_sgpr14
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 16, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr3, implicit killed $vgpr1
+    ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 7, implicit-def $vgpr0
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13_sgpr14
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13_sgpr14
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr14, 2, $vgpr0, implicit killed $sgpr12_sgpr13_sgpr14
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 16, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
     ; GCN32-MUBUF-NEXT: renamable $sgpr12_sgpr13_sgpr14_sgpr15 = IMPLICIT_DEF
     ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
     ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 15, implicit-def $vgpr0
     ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr14, 2, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr15, 3, $vgpr0, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15
     ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 28, 0, 0, implicit $exec :: (store (s32) into %stack.3, addrspace 5)
     ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
     ; GCN32-MUBUF-NEXT: renamable $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16 = IMPLICIT_DEF
-    ; GCN32-MUBUF-NEXT: $sgpr2 = S_MOV_B32 $exec_lo
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 31, implicit-def $vgpr2
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr2, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN32-MUBUF-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr13, 1, $vgpr2, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN32-MUBUF-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr14, 2, $vgpr2, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN32-MUBUF-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr15, 3, $vgpr2, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN32-MUBUF-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr16, 4, $vgpr2, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 44, 0, 0, implicit $exec :: (store (s32) into %stack.4, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr2, implicit killed $vgpr2
+    ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 31, implicit-def $vgpr0
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr14, 2, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr15, 3, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr16, 4, $vgpr0, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 44, 0, 0, implicit $exec :: (store (s32) into %stack.4, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
     ; GCN32-MUBUF-NEXT: renamable $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = IMPLICIT_DEF
-    ; GCN32-MUBUF-NEXT: $sgpr1 = S_MOV_B32 $exec_lo
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 255, implicit-def $vgpr1
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr13, 1, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr14, 2, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr15, 3, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr16, 4, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr17, 5, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr18, 6, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr19, 7, $vgpr1, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 64, 0, 0, implicit $exec :: (store (s32) into %stack.5, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr1, implicit killed $vgpr1
+    ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 255, implicit-def $vgpr0
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr14, 2, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr15, 3, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr16, 4, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr17, 5, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr18, 6, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr19, 7, $vgpr0, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 64, 0, 0, implicit $exec :: (store (s32) into %stack.5, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
     ; GCN32-MUBUF-NEXT: renamable $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27 = IMPLICIT_DEF
     ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
     ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 65535, implicit-def $vgpr0
     ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr14, 2, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr15, 3, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr16, 4, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr17, 5, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr18, 6, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr19, 7, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr20, 8, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr21, 9, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr22, 10, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr23, 11, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr24, 12, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr25, 13, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr26, 14, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr27, 15, $vgpr0, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 96, 0, 0, implicit $exec :: (store (s32) into %stack.6, addrspace 5)
     ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
     ; GCN32-MUBUF-NEXT: renamable $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95 = IMPLICIT_DEF
-    ; GCN32-MUBUF-NEXT: $sgpr2 = S_MOV_B32 $exec_lo
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 4294967295, implicit-def $vgpr1
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr64, 0, undef $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr65, 1, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr66, 2, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr67, 3, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr68, 4, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr69, 5, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr70, 6, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr71, 7, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr72, 8, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr73, 9, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr74, 10, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr75, 11, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr76, 12, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr77, 13, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr78, 14, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr79, 15, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr80, 16, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr81, 17, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr82, 18, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr83, 19, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr84, 20, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr85, 21, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr86, 22, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr87, 23, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr88, 24, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr89, 25, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr90, 26, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr91, 27, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr92, 28, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr93, 29, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr94, 30, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr95, 31, $vgpr1, implicit killed $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 160, 0, 0, implicit $exec :: (store (s32) into %stack.7, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr2, implicit killed $vgpr1
+    ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 4294967295, implicit-def $vgpr0
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr64, 0, undef $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr65, 1, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr66, 2, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr67, 3, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr68, 4, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr69, 5, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr70, 6, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr71, 7, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr72, 8, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr73, 9, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr74, 10, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr75, 11, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr76, 12, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr77, 13, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr78, 14, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr79, 15, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr80, 16, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr81, 17, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr82, 18, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr83, 19, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr84, 20, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr85, 21, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr86, 22, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr87, 23, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr88, 24, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr89, 25, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr90, 26, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr91, 27, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr92, 28, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr93, 29, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr94, 30, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr95, 31, $vgpr0, implicit killed $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 160, 0, 0, implicit $exec :: (store (s32) into %stack.7, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
     ; GCN32-MUBUF-NEXT: renamable $sgpr12 = IMPLICIT_DEF
     ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
     ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 1, implicit-def $vgpr0
     ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN32-MUBUF-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0
     ; GCN32-MUBUF-NEXT: $sgpr1 = S_ADD_I32 $sgpr33, 131072, implicit-def dead $scc
     ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, killed $sgpr1, 0, 0, 0, implicit $exec :: (store (s32) into %stack.8, align 4096, addrspace 5)
     ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
     ; GCN64-FLATSCR-LABEL: name: check_spill
     ; GCN64-FLATSCR: liveins: $sgpr8, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr11, $sgpr0_sgpr1
     ; GCN64-FLATSCR-NEXT: {{  $}}
     ; GCN64-FLATSCR-NEXT: $sgpr33 = S_MOV_B32 0
     ; GCN64-FLATSCR-NEXT: $flat_scr_lo = S_ADD_U32 $sgpr0, $sgpr11, implicit-def $scc
     ; GCN64-FLATSCR-NEXT: $flat_scr_hi = S_ADDC_U32 $sgpr1, 0, implicit-def dead $scc, implicit $scc
     ; GCN64-FLATSCR-NEXT: renamable $sgpr12 = IMPLICIT_DEF
-    ; GCN64-FLATSCR-NEXT: $sgpr6_sgpr7 = S_MOV_B64 $exec
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr1
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr1, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 killed $sgpr12, 0, undef $vgpr1
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr1, $sgpr33, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.0, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr6_sgpr7, implicit killed $vgpr1
+    ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr0
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 killed $sgpr12, 0, undef $vgpr0
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.0, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-FLATSCR-NEXT: renamable $sgpr12 = IMPLICIT_DEF
     ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr0
     ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0
     ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.0, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-FLATSCR-NEXT: renamable $sgpr12_sgpr13 = IMPLICIT_DEF
-    ; GCN64-FLATSCR-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr2
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr2, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr2, implicit $sgpr12_sgpr13
-    ; GCN64-FLATSCR-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr13, 1, $vgpr2, implicit killed $sgpr12_sgpr13
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr2, $sgpr33, 8, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.1, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr2 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr2
+    ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit killed $sgpr12_sgpr13
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 8, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.1, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-FLATSCR-NEXT: renamable $sgpr12_sgpr13 = IMPLICIT_DEF
-    ; GCN64-FLATSCR-NEXT: $sgpr2_sgpr3 = S_MOV_B64 $exec
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr3
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr3, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr3 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr3, implicit $sgpr12_sgpr13
-    ; GCN64-FLATSCR-NEXT: $vgpr3 = V_WRITELANE_B32 $sgpr13, 1, $vgpr3, implicit $sgpr12_sgpr13
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr3, $sgpr33, 8, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.1, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr3 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr2_sgpr3, implicit killed $vgpr3
+    ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 8, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.1, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-FLATSCR-NEXT: renamable $sgpr12_sgpr13_sgpr14 = IMPLICIT_DEF
-    ; GCN64-FLATSCR-NEXT: $sgpr6_sgpr7 = S_MOV_B64 $exec
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 7, implicit-def $vgpr1
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr1, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr1, implicit $sgpr12_sgpr13_sgpr14
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr13, 1, $vgpr1, implicit $sgpr12_sgpr13_sgpr14
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr14, 2, $vgpr1, implicit killed $sgpr12_sgpr13_sgpr14
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr1, $sgpr33, 16, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.2, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr6_sgpr7, implicit killed $vgpr1
+    ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 7, implicit-def $vgpr0
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13_sgpr14
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13_sgpr14
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr14, 2, $vgpr0, implicit killed $sgpr12_sgpr13_sgpr14
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 16, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.2, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-FLATSCR-NEXT: renamable $sgpr12_sgpr13_sgpr14_sgpr15 = IMPLICIT_DEF
     ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 15, implicit-def $vgpr0
     ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr14, 2, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr15, 3, $vgpr0, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15
     ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 28, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.3, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-FLATSCR-NEXT: renamable $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16 = IMPLICIT_DEF
-    ; GCN64-FLATSCR-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 31, implicit-def $vgpr2
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr2, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr2, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN64-FLATSCR-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr13, 1, $vgpr2, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN64-FLATSCR-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr14, 2, $vgpr2, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN64-FLATSCR-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr15, 3, $vgpr2, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN64-FLATSCR-NEXT: $vgpr2 = V_WRITELANE_B32 $sgpr16, 4, $vgpr2, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr2, $sgpr33, 44, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.4, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr2 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr2
+    ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 31, implicit-def $vgpr0
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr14, 2, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr15, 3, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr16, 4, $vgpr0, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 44, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.4, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-FLATSCR-NEXT: renamable $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = IMPLICIT_DEF
-    ; GCN64-FLATSCR-NEXT: $sgpr2_sgpr3 = S_MOV_B64 $exec
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 255, implicit-def $vgpr1
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr1, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr13, 1, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr14, 2, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr15, 3, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr16, 4, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr17, 5, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr18, 6, $vgpr1, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr19, 7, $vgpr1, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr1, $sgpr33, 64, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.5, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr2_sgpr3, implicit killed $vgpr1
+    ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 255, implicit-def $vgpr0
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr14, 2, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr15, 3, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr16, 4, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr17, 5, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr18, 6, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr19, 7, $vgpr0, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 64, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.5, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-FLATSCR-NEXT: renamable $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27 = IMPLICIT_DEF
     ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 65535, implicit-def $vgpr0
     ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr13, 1, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr14, 2, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr15, 3, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr16, 4, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr17, 5, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr18, 6, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr19, 7, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr20, 8, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr21, 9, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr22, 10, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr23, 11, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr24, 12, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr25, 13, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr26, 14, $vgpr0, implicit $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr27, 15, $vgpr0, implicit killed $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 96, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.6, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-FLATSCR-NEXT: renamable $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95 = IMPLICIT_DEF
-    ; GCN64-FLATSCR-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 4294967295, implicit-def $vgpr1
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr1, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr64, 0, undef $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr65, 1, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr66, 2, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr67, 3, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr68, 4, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr69, 5, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr70, 6, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr71, 7, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr72, 8, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr73, 9, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr74, 10, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr75, 11, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr76, 12, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr77, 13, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr78, 14, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr79, 15, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr80, 16, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr81, 17, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr82, 18, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr83, 19, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr84, 20, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr85, 21, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr86, 22, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr87, 23, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr88, 24, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr89, 25, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr90, 26, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr91, 27, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr92, 28, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr93, 29, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr94, 30, $vgpr1, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = V_WRITELANE_B32 $sgpr95, 31, $vgpr1, implicit killed $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr1, $sgpr33, 160, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.7, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr1
+    ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 4294967295, implicit-def $vgpr0
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr64, 0, undef $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr65, 1, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr66, 2, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr67, 3, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr68, 4, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr69, 5, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr70, 6, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr71, 7, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr72, 8, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr73, 9, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr74, 10, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr75, 11, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr76, 12, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr77, 13, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr78, 14, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr79, 15, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr80, 16, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr81, 17, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr82, 18, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr83, 19, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr84, 20, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr85, 21, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr86, 22, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr87, 23, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr88, 24, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr89, 25, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr90, 26, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr91, 27, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr92, 28, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr93, 29, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr94, 30, $vgpr0, implicit $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr95, 31, $vgpr0, implicit killed $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 160, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.7, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-FLATSCR-NEXT: renamable $sgpr12 = IMPLICIT_DEF
     ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr0
     ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 0, undef $vgpr0
     ; GCN64-FLATSCR-NEXT: $sgpr2 = S_ADD_I32 $sgpr33, 4096, implicit-def dead $scc
     ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, killed $sgpr2, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.8, align 4096, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
bb.0:
     ; GCN64-MUBUF-NEXT: {{  $}}
     ; GCN64-MUBUF-NEXT: $sgpr33 = S_MOV_B32 0
     ; GCN64-MUBUF-NEXT: $sgpr28 = S_MOV_B32 &SCRATCH_RSRC_DWORD0, implicit-def $sgpr28_sgpr29_sgpr30_sgpr31
     ; GCN64-MUBUF-NEXT: $sgpr29 = S_MOV_B32 &SCRATCH_RSRC_DWORD1, implicit-def $sgpr28_sgpr29_sgpr30_sgpr31
     ; GCN64-MUBUF-NEXT: $sgpr30 = S_MOV_B32 4294967295, implicit-def $sgpr28_sgpr29_sgpr30_sgpr31
     ; GCN64-MUBUF-NEXT: $sgpr31 = S_MOV_B32 14680064, implicit-def $sgpr28_sgpr29_sgpr30_sgpr31
     ; GCN64-MUBUF-NEXT: $sgpr28 = S_ADD_U32 $sgpr28, $sgpr11, implicit-def $scc, implicit-def $sgpr28_sgpr29_sgpr30_sgpr31
     ; GCN64-MUBUF-NEXT: $sgpr29 = S_ADDC_U32 $sgpr29, 0, implicit-def dead $scc, implicit $scc, implicit-def $sgpr28_sgpr29_sgpr30_sgpr31
-    ; GCN64-MUBUF-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr1
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 4, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 killed $vgpr1, 0
-    ; GCN64-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr1
-    ; GCN64-MUBUF-NEXT: $sgpr2_sgpr3 = S_MOV_B64 $exec
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr2
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 8, 0, 0, implicit $exec :: (load (s32) from %stack.1, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr2, 0, implicit-def $sgpr12_sgpr13
-    ; GCN64-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 killed $vgpr2, 1
-    ; GCN64-MUBUF-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr2_sgpr3, implicit killed $vgpr2
-    ; GCN64-MUBUF-NEXT: $sgpr6_sgpr7 = S_MOV_B64 $exec
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 7, implicit-def $vgpr3
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr3, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr3 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 16, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr3, 0, implicit-def $sgpr12_sgpr13_sgpr14
-    ; GCN64-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr3, 1
-    ; GCN64-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 killed $vgpr3, 2
-    ; GCN64-MUBUF-NEXT: $vgpr3 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr6_sgpr7, implicit killed $vgpr3
+    ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr0
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 4, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 killed $vgpr0, 0
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
+    ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 8, 0, 0, implicit $exec :: (load (s32) from %stack.1, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13
+    ; GCN64-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 killed $vgpr0, 1
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
+    ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 7, implicit-def $vgpr0
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 16, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13_sgpr14
+    ; GCN64-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 1
+    ; GCN64-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 killed $vgpr0, 2
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 15, implicit-def $vgpr0
     ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 28, 0, 0, implicit $exec :: (load (s32) from %stack.3, addrspace 5)
     ; GCN64-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
     ; GCN64-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 1
     ; GCN64-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 $vgpr0, 2
     ; GCN64-MUBUF-NEXT: $sgpr15 = V_READLANE_B32 killed $vgpr0, 3
     ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
-    ; GCN64-MUBUF-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 31, implicit-def $vgpr1
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 44, 0, 0, implicit $exec :: (load (s32) from %stack.4, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr1, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN64-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr1, 1
-    ; GCN64-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 $vgpr1, 2
-    ; GCN64-MUBUF-NEXT: $sgpr15 = V_READLANE_B32 $vgpr1, 3
-    ; GCN64-MUBUF-NEXT: $sgpr16 = V_READLANE_B32 killed $vgpr1, 4
-    ; GCN64-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr1
-    ; GCN64-MUBUF-NEXT: $sgpr2_sgpr3 = S_MOV_B64 $exec
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 255, implicit-def $vgpr2
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 64, 0, 0, implicit $exec :: (load (s32) from %stack.5, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr2, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr2, 1
-    ; GCN64-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 $vgpr2, 2
-    ; GCN64-MUBUF-NEXT: $sgpr15 = V_READLANE_B32 $vgpr2, 3
-    ; GCN64-MUBUF-NEXT: $sgpr16 = V_READLANE_B32 $vgpr2, 4
-    ; GCN64-MUBUF-NEXT: $sgpr17 = V_READLANE_B32 $vgpr2, 5
-    ; GCN64-MUBUF-NEXT: $sgpr18 = V_READLANE_B32 $vgpr2, 6
-    ; GCN64-MUBUF-NEXT: $sgpr19 = V_READLANE_B32 killed $vgpr2, 7
-    ; GCN64-MUBUF-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr2_sgpr3, implicit killed $vgpr2
+    ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 31, implicit-def $vgpr0
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 44, 0, 0, implicit $exec :: (load (s32) from %stack.4, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN64-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 1
+    ; GCN64-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 $vgpr0, 2
+    ; GCN64-MUBUF-NEXT: $sgpr15 = V_READLANE_B32 $vgpr0, 3
+    ; GCN64-MUBUF-NEXT: $sgpr16 = V_READLANE_B32 killed $vgpr0, 4
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
+    ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 255, implicit-def $vgpr0
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 64, 0, 0, implicit $exec :: (load (s32) from %stack.5, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 1
+    ; GCN64-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 $vgpr0, 2
+    ; GCN64-MUBUF-NEXT: $sgpr15 = V_READLANE_B32 $vgpr0, 3
+    ; GCN64-MUBUF-NEXT: $sgpr16 = V_READLANE_B32 $vgpr0, 4
+    ; GCN64-MUBUF-NEXT: $sgpr17 = V_READLANE_B32 $vgpr0, 5
+    ; GCN64-MUBUF-NEXT: $sgpr18 = V_READLANE_B32 $vgpr0, 6
+    ; GCN64-MUBUF-NEXT: $sgpr19 = V_READLANE_B32 killed $vgpr0, 7
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 65535, implicit-def $vgpr0
     ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 96, 0, 0, implicit $exec :: (load (s32) from %stack.6, addrspace 5)
     ; GCN64-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 1
     ; GCN64-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 $vgpr0, 2
     ; GCN64-MUBUF-NEXT: $sgpr15 = V_READLANE_B32 $vgpr0, 3
     ; GCN64-MUBUF-NEXT: $sgpr16 = V_READLANE_B32 $vgpr0, 4
     ; GCN64-MUBUF-NEXT: $sgpr17 = V_READLANE_B32 $vgpr0, 5
     ; GCN64-MUBUF-NEXT: $sgpr18 = V_READLANE_B32 $vgpr0, 6
     ; GCN64-MUBUF-NEXT: $sgpr19 = V_READLANE_B32 $vgpr0, 7
     ; GCN64-MUBUF-NEXT: $sgpr20 = V_READLANE_B32 $vgpr0, 8
     ; GCN64-MUBUF-NEXT: $sgpr21 = V_READLANE_B32 $vgpr0, 9
     ; GCN64-MUBUF-NEXT: $sgpr22 = V_READLANE_B32 $vgpr0, 10
     ; GCN64-MUBUF-NEXT: $sgpr23 = V_READLANE_B32 $vgpr0, 11
     ; GCN64-MUBUF-NEXT: $sgpr24 = V_READLANE_B32 $vgpr0, 12
     ; GCN64-MUBUF-NEXT: $sgpr25 = V_READLANE_B32 $vgpr0, 13
     ; GCN64-MUBUF-NEXT: $sgpr26 = V_READLANE_B32 $vgpr0, 14
     ; GCN64-MUBUF-NEXT: $sgpr27 = V_READLANE_B32 killed $vgpr0, 15
     ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
-    ; GCN64-MUBUF-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 4294967295, implicit-def $vgpr1
-    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 160, 0, 0, implicit $exec :: (load (s32) from %stack.7, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $sgpr64 = V_READLANE_B32 $vgpr1, 0, implicit-def $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-MUBUF-NEXT: $sgpr65 = V_READLANE_B32 $vgpr1, 1
-    ; GCN64-MUBUF-NEXT: $sgpr66 = V_READLANE_B32 $vgpr1, 2
-    ; GCN64-MUBUF-NEXT: $sgpr67 = V_READLANE_B32 $vgpr1, 3
-    ; GCN64-MUBUF-NEXT: $sgpr68 = V_READLANE_B32 $vgpr1, 4
-    ; GCN64-MUBUF-NEXT: $sgpr69 = V_READLANE_B32 $vgpr1, 5
-    ; GCN64-MUBUF-NEXT: $sgpr70 = V_READLANE_B32 $vgpr1, 6
-    ; GCN64-MUBUF-NEXT: $sgpr71 = V_READLANE_B32 $vgpr1, 7
-    ; GCN64-MUBUF-NEXT: $sgpr72 = V_READLANE_B32 $vgpr1, 8
-    ; GCN64-MUBUF-NEXT: $sgpr73 = V_READLANE_B32 $vgpr1, 9
-    ; GCN64-MUBUF-NEXT: $sgpr74 = V_READLANE_B32 $vgpr1, 10
-    ; GCN64-MUBUF-NEXT: $sgpr75 = V_READLANE_B32 $vgpr1, 11
-    ; GCN64-MUBUF-NEXT: $sgpr76 = V_READLANE_B32 $vgpr1, 12
-    ; GCN64-MUBUF-NEXT: $sgpr77 = V_READLANE_B32 $vgpr1, 13
-    ; GCN64-MUBUF-NEXT: $sgpr78 = V_READLANE_B32 $vgpr1, 14
-    ; GCN64-MUBUF-NEXT: $sgpr79 = V_READLANE_B32 $vgpr1, 15
-    ; GCN64-MUBUF-NEXT: $sgpr80 = V_READLANE_B32 $vgpr1, 16
-    ; GCN64-MUBUF-NEXT: $sgpr81 = V_READLANE_B32 $vgpr1, 17
-    ; GCN64-MUBUF-NEXT: $sgpr82 = V_READLANE_B32 $vgpr1, 18
-    ; GCN64-MUBUF-NEXT: $sgpr83 = V_READLANE_B32 $vgpr1, 19
-    ; GCN64-MUBUF-NEXT: $sgpr84 = V_READLANE_B32 $vgpr1, 20
-    ; GCN64-MUBUF-NEXT: $sgpr85 = V_READLANE_B32 $vgpr1, 21
-    ; GCN64-MUBUF-NEXT: $sgpr86 = V_READLANE_B32 $vgpr1, 22
-    ; GCN64-MUBUF-NEXT: $sgpr87 = V_READLANE_B32 $vgpr1, 23
-    ; GCN64-MUBUF-NEXT: $sgpr88 = V_READLANE_B32 $vgpr1, 24
-    ; GCN64-MUBUF-NEXT: $sgpr89 = V_READLANE_B32 $vgpr1, 25
-    ; GCN64-MUBUF-NEXT: $sgpr90 = V_READLANE_B32 $vgpr1, 26
-    ; GCN64-MUBUF-NEXT: $sgpr91 = V_READLANE_B32 $vgpr1, 27
-    ; GCN64-MUBUF-NEXT: $sgpr92 = V_READLANE_B32 $vgpr1, 28
-    ; GCN64-MUBUF-NEXT: $sgpr93 = V_READLANE_B32 $vgpr1, 29
-    ; GCN64-MUBUF-NEXT: $sgpr94 = V_READLANE_B32 $vgpr1, 30
-    ; GCN64-MUBUF-NEXT: $sgpr95 = V_READLANE_B32 killed $vgpr1, 31
-    ; GCN64-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr1
+    ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 4294967295, implicit-def $vgpr0
+    ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 160, 0, 0, implicit $exec :: (load (s32) from %stack.7, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $sgpr64 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-MUBUF-NEXT: $sgpr65 = V_READLANE_B32 $vgpr0, 1
+    ; GCN64-MUBUF-NEXT: $sgpr66 = V_READLANE_B32 $vgpr0, 2
+    ; GCN64-MUBUF-NEXT: $sgpr67 = V_READLANE_B32 $vgpr0, 3
+    ; GCN64-MUBUF-NEXT: $sgpr68 = V_READLANE_B32 $vgpr0, 4
+    ; GCN64-MUBUF-NEXT: $sgpr69 = V_READLANE_B32 $vgpr0, 5
+    ; GCN64-MUBUF-NEXT: $sgpr70 = V_READLANE_B32 $vgpr0, 6
+    ; GCN64-MUBUF-NEXT: $sgpr71 = V_READLANE_B32 $vgpr0, 7
+    ; GCN64-MUBUF-NEXT: $sgpr72 = V_READLANE_B32 $vgpr0, 8
+    ; GCN64-MUBUF-NEXT: $sgpr73 = V_READLANE_B32 $vgpr0, 9
+    ; GCN64-MUBUF-NEXT: $sgpr74 = V_READLANE_B32 $vgpr0, 10
+    ; GCN64-MUBUF-NEXT: $sgpr75 = V_READLANE_B32 $vgpr0, 11
+    ; GCN64-MUBUF-NEXT: $sgpr76 = V_READLANE_B32 $vgpr0, 12
+    ; GCN64-MUBUF-NEXT: $sgpr77 = V_READLANE_B32 $vgpr0, 13
+    ; GCN64-MUBUF-NEXT: $sgpr78 = V_READLANE_B32 $vgpr0, 14
+    ; GCN64-MUBUF-NEXT: $sgpr79 = V_READLANE_B32 $vgpr0, 15
+    ; GCN64-MUBUF-NEXT: $sgpr80 = V_READLANE_B32 $vgpr0, 16
+    ; GCN64-MUBUF-NEXT: $sgpr81 = V_READLANE_B32 $vgpr0, 17
+    ; GCN64-MUBUF-NEXT: $sgpr82 = V_READLANE_B32 $vgpr0, 18
+    ; GCN64-MUBUF-NEXT: $sgpr83 = V_READLANE_B32 $vgpr0, 19
+    ; GCN64-MUBUF-NEXT: $sgpr84 = V_READLANE_B32 $vgpr0, 20
+    ; GCN64-MUBUF-NEXT: $sgpr85 = V_READLANE_B32 $vgpr0, 21
+    ; GCN64-MUBUF-NEXT: $sgpr86 = V_READLANE_B32 $vgpr0, 22
+    ; GCN64-MUBUF-NEXT: $sgpr87 = V_READLANE_B32 $vgpr0, 23
+    ; GCN64-MUBUF-NEXT: $sgpr88 = V_READLANE_B32 $vgpr0, 24
+    ; GCN64-MUBUF-NEXT: $sgpr89 = V_READLANE_B32 $vgpr0, 25
+    ; GCN64-MUBUF-NEXT: $sgpr90 = V_READLANE_B32 $vgpr0, 26
+    ; GCN64-MUBUF-NEXT: $sgpr91 = V_READLANE_B32 $vgpr0, 27
+    ; GCN64-MUBUF-NEXT: $sgpr92 = V_READLANE_B32 $vgpr0, 28
+    ; GCN64-MUBUF-NEXT: $sgpr93 = V_READLANE_B32 $vgpr0, 29
+    ; GCN64-MUBUF-NEXT: $sgpr94 = V_READLANE_B32 $vgpr0, 30
+    ; GCN64-MUBUF-NEXT: $sgpr95 = V_READLANE_B32 killed $vgpr0, 31
+    ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-MUBUF-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr0
     ; GCN64-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-MUBUF-NEXT: $sgpr2 = S_ADD_I32 $sgpr33, 262144, implicit-def dead $scc
     ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, killed $sgpr2, 0, 0, 0, implicit $exec :: (load (s32) from %stack.8, align 4096, addrspace 5)
     ; GCN64-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 killed $vgpr0, 0
     ; GCN64-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr28_sgpr29_sgpr30_sgpr31, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-MUBUF-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN32-MUBUF-LABEL: name: check_reload
     ; GCN32-MUBUF: liveins: $sgpr8, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr11
     ; GCN32-MUBUF-NEXT: {{  $}}
     ; GCN32-MUBUF-NEXT: $sgpr33 = S_MOV_B32 0
     ; GCN32-MUBUF-NEXT: $sgpr96 = S_MOV_B32 &SCRATCH_RSRC_DWORD0, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GCN32-MUBUF-NEXT: $sgpr97 = S_MOV_B32 &SCRATCH_RSRC_DWORD1, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GCN32-MUBUF-NEXT: $sgpr98 = S_MOV_B32 4294967295, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GCN32-MUBUF-NEXT: $sgpr99 = S_MOV_B32 834756608, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GCN32-MUBUF-NEXT: $sgpr96 = S_ADD_U32 $sgpr96, $sgpr11, implicit-def $scc, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GCN32-MUBUF-NEXT: $sgpr97 = S_ADDC_U32 $sgpr97, 0, implicit-def dead $scc, implicit $scc, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
-    ; GCN32-MUBUF-NEXT: $sgpr2 = S_MOV_B32 $exec_lo
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 1, implicit-def $vgpr1
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 killed $vgpr1, 0
-    ; GCN32-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr2, implicit killed $vgpr1
-    ; GCN32-MUBUF-NEXT: $sgpr1 = S_MOV_B32 $exec_lo
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 3, implicit-def $vgpr2
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 8, 0, 0, implicit $exec :: (load (s32) from %stack.1, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr2, 0, implicit-def $sgpr12_sgpr13
-    ; GCN32-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 killed $vgpr2, 1
-    ; GCN32-MUBUF-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr1, implicit killed $vgpr2
-    ; GCN32-MUBUF-NEXT: $sgpr3 = S_MOV_B32 $exec_lo
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 7, implicit-def $vgpr3
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr3, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr3 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 16, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr3, 0, implicit-def $sgpr12_sgpr13_sgpr14
-    ; GCN32-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr3, 1
-    ; GCN32-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 killed $vgpr3, 2
-    ; GCN32-MUBUF-NEXT: $vgpr3 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr3, implicit killed $vgpr3
+    ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 1, implicit-def $vgpr0
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 killed $vgpr0, 0
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
+    ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 3, implicit-def $vgpr0
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 8, 0, 0, implicit $exec :: (load (s32) from %stack.1, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13
+    ; GCN32-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 killed $vgpr0, 1
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
+    ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 7, implicit-def $vgpr0
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 16, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13_sgpr14
+    ; GCN32-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 1
+    ; GCN32-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 killed $vgpr0, 2
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
     ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
     ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 15, implicit-def $vgpr0
     ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 28, 0, 0, implicit $exec :: (load (s32) from %stack.3, addrspace 5)
     ; GCN32-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
     ; GCN32-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 1
     ; GCN32-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 $vgpr0, 2
     ; GCN32-MUBUF-NEXT: $sgpr15 = V_READLANE_B32 killed $vgpr0, 3
     ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
-    ; GCN32-MUBUF-NEXT: $sgpr2 = S_MOV_B32 $exec_lo
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 31, implicit-def $vgpr1
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 44, 0, 0, implicit $exec :: (load (s32) from %stack.4, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr1, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN32-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr1, 1
-    ; GCN32-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 $vgpr1, 2
-    ; GCN32-MUBUF-NEXT: $sgpr15 = V_READLANE_B32 $vgpr1, 3
-    ; GCN32-MUBUF-NEXT: $sgpr16 = V_READLANE_B32 killed $vgpr1, 4
-    ; GCN32-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr2, implicit killed $vgpr1
-    ; GCN32-MUBUF-NEXT: $sgpr1 = S_MOV_B32 $exec_lo
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 255, implicit-def $vgpr2
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 64, 0, 0, implicit $exec :: (load (s32) from %stack.5, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr2, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN32-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr2, 1
-    ; GCN32-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 $vgpr2, 2
-    ; GCN32-MUBUF-NEXT: $sgpr15 = V_READLANE_B32 $vgpr2, 3
-    ; GCN32-MUBUF-NEXT: $sgpr16 = V_READLANE_B32 $vgpr2, 4
-    ; GCN32-MUBUF-NEXT: $sgpr17 = V_READLANE_B32 $vgpr2, 5
-    ; GCN32-MUBUF-NEXT: $sgpr18 = V_READLANE_B32 $vgpr2, 6
-    ; GCN32-MUBUF-NEXT: $sgpr19 = V_READLANE_B32 killed $vgpr2, 7
-    ; GCN32-MUBUF-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr1, implicit killed $vgpr2
+    ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 31, implicit-def $vgpr0
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 44, 0, 0, implicit $exec :: (load (s32) from %stack.4, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN32-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 1
+    ; GCN32-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 $vgpr0, 2
+    ; GCN32-MUBUF-NEXT: $sgpr15 = V_READLANE_B32 $vgpr0, 3
+    ; GCN32-MUBUF-NEXT: $sgpr16 = V_READLANE_B32 killed $vgpr0, 4
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
+    ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 255, implicit-def $vgpr0
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 64, 0, 0, implicit $exec :: (load (s32) from %stack.5, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN32-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 1
+    ; GCN32-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 $vgpr0, 2
+    ; GCN32-MUBUF-NEXT: $sgpr15 = V_READLANE_B32 $vgpr0, 3
+    ; GCN32-MUBUF-NEXT: $sgpr16 = V_READLANE_B32 $vgpr0, 4
+    ; GCN32-MUBUF-NEXT: $sgpr17 = V_READLANE_B32 $vgpr0, 5
+    ; GCN32-MUBUF-NEXT: $sgpr18 = V_READLANE_B32 $vgpr0, 6
+    ; GCN32-MUBUF-NEXT: $sgpr19 = V_READLANE_B32 killed $vgpr0, 7
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
     ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
     ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 65535, implicit-def $vgpr0
     ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 96, 0, 0, implicit $exec :: (load (s32) from %stack.6, addrspace 5)
     ; GCN32-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN32-MUBUF-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 1
     ; GCN32-MUBUF-NEXT: $sgpr14 = V_READLANE_B32 $vgpr0, 2
     ; GCN32-MUBUF-NEXT: $sgpr15 = V_READLANE_B32 $vgpr0, 3
     ; GCN32-MUBUF-NEXT: $sgpr16 = V_READLANE_B32 $vgpr0, 4
     ; GCN32-MUBUF-NEXT: $sgpr17 = V_READLANE_B32 $vgpr0, 5
     ; GCN32-MUBUF-NEXT: $sgpr18 = V_READLANE_B32 $vgpr0, 6
     ; GCN32-MUBUF-NEXT: $sgpr19 = V_READLANE_B32 $vgpr0, 7
     ; GCN32-MUBUF-NEXT: $sgpr20 = V_READLANE_B32 $vgpr0, 8
     ; GCN32-MUBUF-NEXT: $sgpr21 = V_READLANE_B32 $vgpr0, 9
     ; GCN32-MUBUF-NEXT: $sgpr22 = V_READLANE_B32 $vgpr0, 10
     ; GCN32-MUBUF-NEXT: $sgpr23 = V_READLANE_B32 $vgpr0, 11
     ; GCN32-MUBUF-NEXT: $sgpr24 = V_READLANE_B32 $vgpr0, 12
     ; GCN32-MUBUF-NEXT: $sgpr25 = V_READLANE_B32 $vgpr0, 13
     ; GCN32-MUBUF-NEXT: $sgpr26 = V_READLANE_B32 $vgpr0, 14
     ; GCN32-MUBUF-NEXT: $sgpr27 = V_READLANE_B32 killed $vgpr0, 15
     ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
-    ; GCN32-MUBUF-NEXT: $sgpr2 = S_MOV_B32 $exec_lo
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 4294967295, implicit-def $vgpr1
-    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 160, 0, 0, implicit $exec :: (load (s32) from %stack.7, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $sgpr64 = V_READLANE_B32 $vgpr1, 0, implicit-def $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN32-MUBUF-NEXT: $sgpr65 = V_READLANE_B32 $vgpr1, 1
-    ; GCN32-MUBUF-NEXT: $sgpr66 = V_READLANE_B32 $vgpr1, 2
-    ; GCN32-MUBUF-NEXT: $sgpr67 = V_READLANE_B32 $vgpr1, 3
-    ; GCN32-MUBUF-NEXT: $sgpr68 = V_READLANE_B32 $vgpr1, 4
-    ; GCN32-MUBUF-NEXT: $sgpr69 = V_READLANE_B32 $vgpr1, 5
-    ; GCN32-MUBUF-NEXT: $sgpr70 = V_READLANE_B32 $vgpr1, 6
-    ; GCN32-MUBUF-NEXT: $sgpr71 = V_READLANE_B32 $vgpr1, 7
-    ; GCN32-MUBUF-NEXT: $sgpr72 = V_READLANE_B32 $vgpr1, 8
-    ; GCN32-MUBUF-NEXT: $sgpr73 = V_READLANE_B32 $vgpr1, 9
-    ; GCN32-MUBUF-NEXT: $sgpr74 = V_READLANE_B32 $vgpr1, 10
-    ; GCN32-MUBUF-NEXT: $sgpr75 = V_READLANE_B32 $vgpr1, 11
-    ; GCN32-MUBUF-NEXT: $sgpr76 = V_READLANE_B32 $vgpr1, 12
-    ; GCN32-MUBUF-NEXT: $sgpr77 = V_READLANE_B32 $vgpr1, 13
-    ; GCN32-MUBUF-NEXT: $sgpr78 = V_READLANE_B32 $vgpr1, 14
-    ; GCN32-MUBUF-NEXT: $sgpr79 = V_READLANE_B32 $vgpr1, 15
-    ; GCN32-MUBUF-NEXT: $sgpr80 = V_READLANE_B32 $vgpr1, 16
-    ; GCN32-MUBUF-NEXT: $sgpr81 = V_READLANE_B32 $vgpr1, 17
-    ; GCN32-MUBUF-NEXT: $sgpr82 = V_READLANE_B32 $vgpr1, 18
-    ; GCN32-MUBUF-NEXT: $sgpr83 = V_READLANE_B32 $vgpr1, 19
-    ; GCN32-MUBUF-NEXT: $sgpr84 = V_READLANE_B32 $vgpr1, 20
-    ; GCN32-MUBUF-NEXT: $sgpr85 = V_READLANE_B32 $vgpr1, 21
-    ; GCN32-MUBUF-NEXT: $sgpr86 = V_READLANE_B32 $vgpr1, 22
-    ; GCN32-MUBUF-NEXT: $sgpr87 = V_READLANE_B32 $vgpr1, 23
-    ; GCN32-MUBUF-NEXT: $sgpr88 = V_READLANE_B32 $vgpr1, 24
-    ; GCN32-MUBUF-NEXT: $sgpr89 = V_READLANE_B32 $vgpr1, 25
-    ; GCN32-MUBUF-NEXT: $sgpr90 = V_READLANE_B32 $vgpr1, 26
-    ; GCN32-MUBUF-NEXT: $sgpr91 = V_READLANE_B32 $vgpr1, 27
-    ; GCN32-MUBUF-NEXT: $sgpr92 = V_READLANE_B32 $vgpr1, 28
-    ; GCN32-MUBUF-NEXT: $sgpr93 = V_READLANE_B32 $vgpr1, 29
-    ; GCN32-MUBUF-NEXT: $sgpr94 = V_READLANE_B32 $vgpr1, 30
-    ; GCN32-MUBUF-NEXT: $sgpr95 = V_READLANE_B32 killed $vgpr1, 31
-    ; GCN32-MUBUF-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr2, implicit killed $vgpr1
+    ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 4294967295, implicit-def $vgpr0
+    ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 160, 0, 0, implicit $exec :: (load (s32) from %stack.7, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $sgpr64 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN32-MUBUF-NEXT: $sgpr65 = V_READLANE_B32 $vgpr0, 1
+    ; GCN32-MUBUF-NEXT: $sgpr66 = V_READLANE_B32 $vgpr0, 2
+    ; GCN32-MUBUF-NEXT: $sgpr67 = V_READLANE_B32 $vgpr0, 3
+    ; GCN32-MUBUF-NEXT: $sgpr68 = V_READLANE_B32 $vgpr0, 4
+    ; GCN32-MUBUF-NEXT: $sgpr69 = V_READLANE_B32 $vgpr0, 5
+    ; GCN32-MUBUF-NEXT: $sgpr70 = V_READLANE_B32 $vgpr0, 6
+    ; GCN32-MUBUF-NEXT: $sgpr71 = V_READLANE_B32 $vgpr0, 7
+    ; GCN32-MUBUF-NEXT: $sgpr72 = V_READLANE_B32 $vgpr0, 8
+    ; GCN32-MUBUF-NEXT: $sgpr73 = V_READLANE_B32 $vgpr0, 9
+    ; GCN32-MUBUF-NEXT: $sgpr74 = V_READLANE_B32 $vgpr0, 10
+    ; GCN32-MUBUF-NEXT: $sgpr75 = V_READLANE_B32 $vgpr0, 11
+    ; GCN32-MUBUF-NEXT: $sgpr76 = V_READLANE_B32 $vgpr0, 12
+    ; GCN32-MUBUF-NEXT: $sgpr77 = V_READLANE_B32 $vgpr0, 13
+    ; GCN32-MUBUF-NEXT: $sgpr78 = V_READLANE_B32 $vgpr0, 14
+    ; GCN32-MUBUF-NEXT: $sgpr79 = V_READLANE_B32 $vgpr0, 15
+    ; GCN32-MUBUF-NEXT: $sgpr80 = V_READLANE_B32 $vgpr0, 16
+    ; GCN32-MUBUF-NEXT: $sgpr81 = V_READLANE_B32 $vgpr0, 17
+    ; GCN32-MUBUF-NEXT: $sgpr82 = V_READLANE_B32 $vgpr0, 18
+    ; GCN32-MUBUF-NEXT: $sgpr83 = V_READLANE_B32 $vgpr0, 19
+    ; GCN32-MUBUF-NEXT: $sgpr84 = V_READLANE_B32 $vgpr0, 20
+    ; GCN32-MUBUF-NEXT: $sgpr85 = V_READLANE_B32 $vgpr0, 21
+    ; GCN32-MUBUF-NEXT: $sgpr86 = V_READLANE_B32 $vgpr0, 22
+    ; GCN32-MUBUF-NEXT: $sgpr87 = V_READLANE_B32 $vgpr0, 23
+    ; GCN32-MUBUF-NEXT: $sgpr88 = V_READLANE_B32 $vgpr0, 24
+    ; GCN32-MUBUF-NEXT: $sgpr89 = V_READLANE_B32 $vgpr0, 25
+    ; GCN32-MUBUF-NEXT: $sgpr90 = V_READLANE_B32 $vgpr0, 26
+    ; GCN32-MUBUF-NEXT: $sgpr91 = V_READLANE_B32 $vgpr0, 27
+    ; GCN32-MUBUF-NEXT: $sgpr92 = V_READLANE_B32 $vgpr0, 28
+    ; GCN32-MUBUF-NEXT: $sgpr93 = V_READLANE_B32 $vgpr0, 29
+    ; GCN32-MUBUF-NEXT: $sgpr94 = V_READLANE_B32 $vgpr0, 30
+    ; GCN32-MUBUF-NEXT: $sgpr95 = V_READLANE_B32 killed $vgpr0, 31
+    ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
     ; GCN32-MUBUF-NEXT: $sgpr0 = S_MOV_B32 $exec_lo
     ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 1, implicit-def $vgpr0
     ; GCN32-MUBUF-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN32-MUBUF-NEXT: $sgpr1 = S_ADD_I32 $sgpr33, 131072, implicit-def dead $scc
     ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, killed $sgpr1, 0, 0, 0, implicit $exec :: (load (s32) from %stack.8, align 4096, addrspace 5)
     ; GCN32-MUBUF-NEXT: $sgpr12 = V_READLANE_B32 killed $vgpr0, 0
     ; GCN32-MUBUF-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN32-MUBUF-NEXT: $exec_lo = S_MOV_B32 killed $sgpr0, implicit killed $vgpr0
     ; GCN64-FLATSCR-LABEL: name: check_reload
     ; GCN64-FLATSCR: liveins: $sgpr8, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr11, $sgpr0_sgpr1
     ; GCN64-FLATSCR-NEXT: {{  $}}
     ; GCN64-FLATSCR-NEXT: $sgpr33 = S_MOV_B32 0
     ; GCN64-FLATSCR-NEXT: $flat_scr_lo = S_ADD_U32 $sgpr0, $sgpr11, implicit-def $scc
     ; GCN64-FLATSCR-NEXT: $flat_scr_hi = S_ADDC_U32 $sgpr1, 0, implicit-def dead $scc, implicit $scc
-    ; GCN64-FLATSCR-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr1
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr1, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $sgpr12 = V_READLANE_B32 killed $vgpr1, 0
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr1
-    ; GCN64-FLATSCR-NEXT: $sgpr2_sgpr3 = S_MOV_B64 $exec
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr2
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr2, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr2 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $sgpr12 = V_READLANE_B32 $vgpr2, 0, implicit-def $sgpr12_sgpr13
-    ; GCN64-FLATSCR-NEXT: $sgpr13 = V_READLANE_B32 killed $vgpr2, 1
-    ; GCN64-FLATSCR-NEXT: $vgpr2 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr2_sgpr3, implicit killed $vgpr2
-    ; GCN64-FLATSCR-NEXT: $sgpr6_sgpr7 = S_MOV_B64 $exec
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 7, implicit-def $vgpr3
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr3, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr3 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.2, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $sgpr12 = V_READLANE_B32 $vgpr3, 0, implicit-def $sgpr12_sgpr13_sgpr14
-    ; GCN64-FLATSCR-NEXT: $sgpr13 = V_READLANE_B32 $vgpr3, 1
-    ; GCN64-FLATSCR-NEXT: $sgpr14 = V_READLANE_B32 killed $vgpr3, 2
-    ; GCN64-FLATSCR-NEXT: $vgpr3 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr6_sgpr7, implicit killed $vgpr3
+    ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr0
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $sgpr12 = V_READLANE_B32 killed $vgpr0, 0
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
+    ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13
+    ; GCN64-FLATSCR-NEXT: $sgpr13 = V_READLANE_B32 killed $vgpr0, 1
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
+    ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 7, implicit-def $vgpr0
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 16, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.2, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13_sgpr14
+    ; GCN64-FLATSCR-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 1
+    ; GCN64-FLATSCR-NEXT: $sgpr14 = V_READLANE_B32 killed $vgpr0, 2
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 15, implicit-def $vgpr0
     ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 28, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.3, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
     ; GCN64-FLATSCR-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 1
     ; GCN64-FLATSCR-NEXT: $sgpr14 = V_READLANE_B32 $vgpr0, 2
     ; GCN64-FLATSCR-NEXT: $sgpr15 = V_READLANE_B32 killed $vgpr0, 3
     ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
-    ; GCN64-FLATSCR-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 31, implicit-def $vgpr1
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr1, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 44, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.4, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $sgpr12 = V_READLANE_B32 $vgpr1, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
-    ; GCN64-FLATSCR-NEXT: $sgpr13 = V_READLANE_B32 $vgpr1, 1
-    ; GCN64-FLATSCR-NEXT: $sgpr14 = V_READLANE_B32 $vgpr1, 2
-    ; GCN64-FLATSCR-NEXT: $sgpr15 = V_READLANE_B32 $vgpr1, 3
-    ; GCN64-FLATSCR-NEXT: $sgpr16 = V_READLANE_B32 killed $vgpr1, 4
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr1
-    ; GCN64-FLATSCR-NEXT: $sgpr2_sgpr3 = S_MOV_B64 $exec
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 255, implicit-def $vgpr2
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr2, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr2 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 64, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.5, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $sgpr12 = V_READLANE_B32 $vgpr2, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
-    ; GCN64-FLATSCR-NEXT: $sgpr13 = V_READLANE_B32 $vgpr2, 1
-    ; GCN64-FLATSCR-NEXT: $sgpr14 = V_READLANE_B32 $vgpr2, 2
-    ; GCN64-FLATSCR-NEXT: $sgpr15 = V_READLANE_B32 $vgpr2, 3
-    ; GCN64-FLATSCR-NEXT: $sgpr16 = V_READLANE_B32 $vgpr2, 4
-    ; GCN64-FLATSCR-NEXT: $sgpr17 = V_READLANE_B32 $vgpr2, 5
-    ; GCN64-FLATSCR-NEXT: $sgpr18 = V_READLANE_B32 $vgpr2, 6
-    ; GCN64-FLATSCR-NEXT: $sgpr19 = V_READLANE_B32 killed $vgpr2, 7
-    ; GCN64-FLATSCR-NEXT: $vgpr2 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr2_sgpr3, implicit killed $vgpr2
+    ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 31, implicit-def $vgpr0
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 44, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.4, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16
+    ; GCN64-FLATSCR-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 1
+    ; GCN64-FLATSCR-NEXT: $sgpr14 = V_READLANE_B32 $vgpr0, 2
+    ; GCN64-FLATSCR-NEXT: $sgpr15 = V_READLANE_B32 $vgpr0, 3
+    ; GCN64-FLATSCR-NEXT: $sgpr16 = V_READLANE_B32 killed $vgpr0, 4
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
+    ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 255, implicit-def $vgpr0
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 64, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.5, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
+    ; GCN64-FLATSCR-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 1
+    ; GCN64-FLATSCR-NEXT: $sgpr14 = V_READLANE_B32 $vgpr0, 2
+    ; GCN64-FLATSCR-NEXT: $sgpr15 = V_READLANE_B32 $vgpr0, 3
+    ; GCN64-FLATSCR-NEXT: $sgpr16 = V_READLANE_B32 $vgpr0, 4
+    ; GCN64-FLATSCR-NEXT: $sgpr17 = V_READLANE_B32 $vgpr0, 5
+    ; GCN64-FLATSCR-NEXT: $sgpr18 = V_READLANE_B32 $vgpr0, 6
+    ; GCN64-FLATSCR-NEXT: $sgpr19 = V_READLANE_B32 killed $vgpr0, 7
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 65535, implicit-def $vgpr0
     ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 96, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.6, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27
     ; GCN64-FLATSCR-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 1
     ; GCN64-FLATSCR-NEXT: $sgpr14 = V_READLANE_B32 $vgpr0, 2
     ; GCN64-FLATSCR-NEXT: $sgpr15 = V_READLANE_B32 $vgpr0, 3
     ; GCN64-FLATSCR-NEXT: $sgpr16 = V_READLANE_B32 $vgpr0, 4
     ; GCN64-FLATSCR-NEXT: $sgpr17 = V_READLANE_B32 $vgpr0, 5
     ; GCN64-FLATSCR-NEXT: $sgpr18 = V_READLANE_B32 $vgpr0, 6
     ; GCN64-FLATSCR-NEXT: $sgpr19 = V_READLANE_B32 $vgpr0, 7
     ; GCN64-FLATSCR-NEXT: $sgpr20 = V_READLANE_B32 $vgpr0, 8
     ; GCN64-FLATSCR-NEXT: $sgpr21 = V_READLANE_B32 $vgpr0, 9
     ; GCN64-FLATSCR-NEXT: $sgpr22 = V_READLANE_B32 $vgpr0, 10
     ; GCN64-FLATSCR-NEXT: $sgpr23 = V_READLANE_B32 $vgpr0, 11
     ; GCN64-FLATSCR-NEXT: $sgpr24 = V_READLANE_B32 $vgpr0, 12
     ; GCN64-FLATSCR-NEXT: $sgpr25 = V_READLANE_B32 $vgpr0, 13
     ; GCN64-FLATSCR-NEXT: $sgpr26 = V_READLANE_B32 $vgpr0, 14
     ; GCN64-FLATSCR-NEXT: $sgpr27 = V_READLANE_B32 killed $vgpr0, 15
     ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
-    ; GCN64-FLATSCR-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 4294967295, implicit-def $vgpr1
-    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr1, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 160, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.7, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $sgpr64 = V_READLANE_B32 $vgpr1, 0, implicit-def $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
-    ; GCN64-FLATSCR-NEXT: $sgpr65 = V_READLANE_B32 $vgpr1, 1
-    ; GCN64-FLATSCR-NEXT: $sgpr66 = V_READLANE_B32 $vgpr1, 2
-    ; GCN64-FLATSCR-NEXT: $sgpr67 = V_READLANE_B32 $vgpr1, 3
-    ; GCN64-FLATSCR-NEXT: $sgpr68 = V_READLANE_B32 $vgpr1, 4
-    ; GCN64-FLATSCR-NEXT: $sgpr69 = V_READLANE_B32 $vgpr1, 5
-    ; GCN64-FLATSCR-NEXT: $sgpr70 = V_READLANE_B32 $vgpr1, 6
-    ; GCN64-FLATSCR-NEXT: $sgpr71 = V_READLANE_B32 $vgpr1, 7
-    ; GCN64-FLATSCR-NEXT: $sgpr72 = V_READLANE_B32 $vgpr1, 8
-    ; GCN64-FLATSCR-NEXT: $sgpr73 = V_READLANE_B32 $vgpr1, 9
-    ; GCN64-FLATSCR-NEXT: $sgpr74 = V_READLANE_B32 $vgpr1, 10
-    ; GCN64-FLATSCR-NEXT: $sgpr75 = V_READLANE_B32 $vgpr1, 11
-    ; GCN64-FLATSCR-NEXT: $sgpr76 = V_READLANE_B32 $vgpr1, 12
-    ; GCN64-FLATSCR-NEXT: $sgpr77 = V_READLANE_B32 $vgpr1, 13
-    ; GCN64-FLATSCR-NEXT: $sgpr78 = V_READLANE_B32 $vgpr1, 14
-    ; GCN64-FLATSCR-NEXT: $sgpr79 = V_READLANE_B32 $vgpr1, 15
-    ; GCN64-FLATSCR-NEXT: $sgpr80 = V_READLANE_B32 $vgpr1, 16
-    ; GCN64-FLATSCR-NEXT: $sgpr81 = V_READLANE_B32 $vgpr1, 17
-    ; GCN64-FLATSCR-NEXT: $sgpr82 = V_READLANE_B32 $vgpr1, 18
-    ; GCN64-FLATSCR-NEXT: $sgpr83 = V_READLANE_B32 $vgpr1, 19
-    ; GCN64-FLATSCR-NEXT: $sgpr84 = V_READLANE_B32 $vgpr1, 20
-    ; GCN64-FLATSCR-NEXT: $sgpr85 = V_READLANE_B32 $vgpr1, 21
-    ; GCN64-FLATSCR-NEXT: $sgpr86 = V_READLANE_B32 $vgpr1, 22
-    ; GCN64-FLATSCR-NEXT: $sgpr87 = V_READLANE_B32 $vgpr1, 23
-    ; GCN64-FLATSCR-NEXT: $sgpr88 = V_READLANE_B32 $vgpr1, 24
-    ; GCN64-FLATSCR-NEXT: $sgpr89 = V_READLANE_B32 $vgpr1, 25
-    ; GCN64-FLATSCR-NEXT: $sgpr90 = V_READLANE_B32 $vgpr1, 26
-    ; GCN64-FLATSCR-NEXT: $sgpr91 = V_READLANE_B32 $vgpr1, 27
-    ; GCN64-FLATSCR-NEXT: $sgpr92 = V_READLANE_B32 $vgpr1, 28
-    ; GCN64-FLATSCR-NEXT: $sgpr93 = V_READLANE_B32 $vgpr1, 29
-    ; GCN64-FLATSCR-NEXT: $sgpr94 = V_READLANE_B32 $vgpr1, 30
-    ; GCN64-FLATSCR-NEXT: $sgpr95 = V_READLANE_B32 killed $vgpr1, 31
-    ; GCN64-FLATSCR-NEXT: $vgpr1 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr1
+    ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 4294967295, implicit-def $vgpr0
+    ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 160, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.7, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $sgpr64 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr64_sgpr65_sgpr66_sgpr67_sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75_sgpr76_sgpr77_sgpr78_sgpr79_sgpr80_sgpr81_sgpr82_sgpr83_sgpr84_sgpr85_sgpr86_sgpr87_sgpr88_sgpr89_sgpr90_sgpr91_sgpr92_sgpr93_sgpr94_sgpr95
+    ; GCN64-FLATSCR-NEXT: $sgpr65 = V_READLANE_B32 $vgpr0, 1
+    ; GCN64-FLATSCR-NEXT: $sgpr66 = V_READLANE_B32 $vgpr0, 2
+    ; GCN64-FLATSCR-NEXT: $sgpr67 = V_READLANE_B32 $vgpr0, 3
+    ; GCN64-FLATSCR-NEXT: $sgpr68 = V_READLANE_B32 $vgpr0, 4
+    ; GCN64-FLATSCR-NEXT: $sgpr69 = V_READLANE_B32 $vgpr0, 5
+    ; GCN64-FLATSCR-NEXT: $sgpr70 = V_READLANE_B32 $vgpr0, 6
+    ; GCN64-FLATSCR-NEXT: $sgpr71 = V_READLANE_B32 $vgpr0, 7
+    ; GCN64-FLATSCR-NEXT: $sgpr72 = V_READLANE_B32 $vgpr0, 8
+    ; GCN64-FLATSCR-NEXT: $sgpr73 = V_READLANE_B32 $vgpr0, 9
+    ; GCN64-FLATSCR-NEXT: $sgpr74 = V_READLANE_B32 $vgpr0, 10
+    ; GCN64-FLATSCR-NEXT: $sgpr75 = V_READLANE_B32 $vgpr0, 11
+    ; GCN64-FLATSCR-NEXT: $sgpr76 = V_READLANE_B32 $vgpr0, 12
+    ; GCN64-FLATSCR-NEXT: $sgpr77 = V_READLANE_B32 $vgpr0, 13
+    ; GCN64-FLATSCR-NEXT: $sgpr78 = V_READLANE_B32 $vgpr0, 14
+    ; GCN64-FLATSCR-NEXT: $sgpr79 = V_READLANE_B32 $vgpr0, 15
+    ; GCN64-FLATSCR-NEXT: $sgpr80 = V_READLANE_B32 $vgpr0, 16
+    ; GCN64-FLATSCR-NEXT: $sgpr81 = V_READLANE_B32 $vgpr0, 17
+    ; GCN64-FLATSCR-NEXT: $sgpr82 = V_READLANE_B32 $vgpr0, 18
+    ; GCN64-FLATSCR-NEXT: $sgpr83 = V_READLANE_B32 $vgpr0, 19
+    ; GCN64-FLATSCR-NEXT: $sgpr84 = V_READLANE_B32 $vgpr0, 20
+    ; GCN64-FLATSCR-NEXT: $sgpr85 = V_READLANE_B32 $vgpr0, 21
+    ; GCN64-FLATSCR-NEXT: $sgpr86 = V_READLANE_B32 $vgpr0, 22
+    ; GCN64-FLATSCR-NEXT: $sgpr87 = V_READLANE_B32 $vgpr0, 23
+    ; GCN64-FLATSCR-NEXT: $sgpr88 = V_READLANE_B32 $vgpr0, 24
+    ; GCN64-FLATSCR-NEXT: $sgpr89 = V_READLANE_B32 $vgpr0, 25
+    ; GCN64-FLATSCR-NEXT: $sgpr90 = V_READLANE_B32 $vgpr0, 26
+    ; GCN64-FLATSCR-NEXT: $sgpr91 = V_READLANE_B32 $vgpr0, 27
+    ; GCN64-FLATSCR-NEXT: $sgpr92 = V_READLANE_B32 $vgpr0, 28
+    ; GCN64-FLATSCR-NEXT: $sgpr93 = V_READLANE_B32 $vgpr0, 29
+    ; GCN64-FLATSCR-NEXT: $sgpr94 = V_READLANE_B32 $vgpr0, 30
+    ; GCN64-FLATSCR-NEXT: $sgpr95 = V_READLANE_B32 killed $vgpr0, 31
+    ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GCN64-FLATSCR-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 1, implicit-def $vgpr0
     ; GCN64-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $sgpr2 = S_ADD_I32 $sgpr33, 4096, implicit-def dead $scc
     ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR killed $sgpr2, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.8, align 4096, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $sgpr12 = V_READLANE_B32 killed $vgpr0, 0
     ; GCN64-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GCN64-FLATSCR-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0

llvm/test/CodeGen/AMDGPU/sgpr-spills-split-regalloc.ll

 ; GCN-NEXT:    buffer_store_dword v248, off, s[0:3], s33 offset:28 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v249, off, s[0:3], s33 offset:24 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v250, off, s[0:3], s33 offset:20 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v251, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v252, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v253, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v254, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_store_dword v255, off, s[0:3], s33 ; 4-byte Folded Spill
-; GCN-NEXT:    s_mov_b64 s[14:15], exec
+; GCN-NEXT:    s_mov_b64 s[4:5], exec
 ; GCN-NEXT:    s_mov_b64 exec, 1
-; GCN-NEXT:    buffer_store_dword v1, off, s[0:3], s33 offset:456
-; GCN-NEXT:    v_writelane_b32 v1, s30, 0
-; GCN-NEXT:    buffer_store_dword v1, off, s[0:3], s33 offset:448 ; 4-byte Folded Spill
-; GCN-NEXT:    buffer_load_dword v1, off, s[0:3], s33 offset:456
+; GCN-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:456
+; GCN-NEXT:    v_writelane_b32 v0, s30, 0
+; GCN-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:448 ; 4-byte Folded Spill
+; GCN-NEXT:    buffer_load_dword v0, off, s[0:3], s33 offset:456
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    s_mov_b64 exec, s[14:15]
-; GCN-NEXT:    s_mov_b64 s[12:13], exec
+; GCN-NEXT:    s_mov_b64 exec, s[4:5]
+; GCN-NEXT:    s_mov_b64 s[4:5], exec
 ; GCN-NEXT:    s_mov_b64 exec, 1
 ; GCN-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:456
 ; GCN-NEXT:    v_writelane_b32 v0, s31, 0
 ; GCN-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:452 ; 4-byte Folded Spill
 ; GCN-NEXT:    buffer_load_dword v0, off, s[0:3], s33 offset:456
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    s_mov_b64 exec, s[12:13]
+; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_getpc_b64 s[4:5]
 ; GCN-NEXT:    s_add_u32 s4, s4, child_function_ipra@rel32@lo+4
 ; GCN-NEXT:    s_addc_u32 s5, s5, child_function_ipra@rel32@hi+12
 ; GCN-NEXT:    s_mov_b64 s[10:11], s[2:3]
 ; GCN-NEXT:    s_mov_b64 s[8:9], s[0:1]
 ; GCN-NEXT:    s_mov_b64 s[0:1], s[8:9]
 ; GCN-NEXT:    s_mov_b64 s[2:3], s[10:11]
 ; GCN-NEXT:    s_swappc_b64 s[30:31], s[4:5]
-; GCN-NEXT:    s_mov_b64 s[8:9], exec
+; GCN-NEXT:    s_mov_b64 s[4:5], exec
 ; GCN-NEXT:    s_mov_b64 exec, 1
-; GCN-NEXT:    buffer_store_dword v1, off, s[0:3], s33 offset:456
-; GCN-NEXT:    buffer_load_dword v1, off, s[0:3], s33 offset:452 ; 4-byte Folded Reload
+; GCN-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:456
+; GCN-NEXT:    buffer_load_dword v0, off, s[0:3], s33 offset:452 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    v_readlane_b32 s31, v1, 0
-; GCN-NEXT:    buffer_load_dword v1, off, s[0:3], s33 offset:456
+; GCN-NEXT:    v_readlane_b32 s31, v0, 0
+; GCN-NEXT:    buffer_load_dword v0, off, s[0:3], s33 offset:456
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    s_mov_b64 exec, s[8:9]
+; GCN-NEXT:    s_mov_b64 exec, s[4:5]
 ; GCN-NEXT:    s_mov_b64 s[4:5], exec
 ; GCN-NEXT:    s_mov_b64 exec, 1
 ; GCN-NEXT:    buffer_store_dword v0, off, s[0:3], s33 offset:456
 ; GCN-NEXT:    buffer_load_dword v0, off, s[0:3], s33 offset:448 ; 4-byte Folded Reload
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_readlane_b32 s30, v0, 0
 ; GCN-NEXT:    buffer_load_dword v0, off, s[0:3], s33 offset:456
 ; GCN-NEXT:    s_waitcnt vmcnt(0)

llvm/test/CodeGen/AMDGPU/spill-offset-calculation.ll

 ; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], 0 offset:8
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; MUBUF-NEXT:    s_endpgm
 ;
 ; FLATSCR-LABEL: test_inst_offset_kernel:
 ; FLATSCR:       ; %bb.0: ; %entry
 ; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s0, s3
 ; FLATSCR-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
-; FLATSCR-NEXT:    s_mov_b32 s1, 0
-; FLATSCR-NEXT:    scratch_load_dword v0, off, s1 offset:8 glc
+; FLATSCR-NEXT:    s_mov_b32 s0, 0
+; FLATSCR-NEXT:    scratch_load_dword v0, off, s0 offset:8 glc
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    s_movk_i32 s0, 0xffc
 ; FLATSCR-NEXT:    scratch_store_dword off, v0, s0 ; 4-byte Folded Spill
 ; FLATSCR-NEXT:    ;;#ASMSTART
 ; FLATSCR-NEXT:    ;;#ASMEND
 ; FLATSCR-NEXT:    scratch_load_dword v0, off, s0 ; 4-byte Folded Reload
 ; FLATSCR-NEXT:    s_mov_b32 s0, 0
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)

 ; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], 0 offset:8
 ; MUBUF-NEXT:    s_waitcnt vmcnt(0)
 ; MUBUF-NEXT:    s_endpgm
 ;
 ; FLATSCR-LABEL: test_sgpr_offset_kernel:
 ; FLATSCR:       ; %bb.0: ; %entry
 ; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s0, s3
 ; FLATSCR-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
-; FLATSCR-NEXT:    s_mov_b32 s1, 0
-; FLATSCR-NEXT:    scratch_load_dword v0, off, s1 offset:8 glc
+; FLATSCR-NEXT:    s_mov_b32 s0, 0
+; FLATSCR-NEXT:    scratch_load_dword v0, off, s0 offset:8 glc
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    s_movk_i32 s0, 0x1000
 ; FLATSCR-NEXT:    scratch_store_dword off, v0, s0 ; 4-byte Folded Spill
 ; FLATSCR-NEXT:    ;;#ASMSTART
 ; FLATSCR-NEXT:    ;;#ASMEND
 ; FLATSCR-NEXT:    scratch_load_dword v0, off, s0 ; 4-byte Folded Reload
 ; FLATSCR-NEXT:    s_mov_b32 s0, 0
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)

 ; MUBUF-NEXT:    ;;#ASMSTART
 ; MUBUF-NEXT:    ;;#ASMEND
 ; MUBUF-NEXT:    s_endpgm
 ;
 ; FLATSCR-LABEL: test_sgpr_offset_function_scavenge_fail_kernel:
 ; FLATSCR:       ; %bb.0: ; %entry
 ; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s0, s3
 ; FLATSCR-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
-; FLATSCR-NEXT:    s_mov_b32 s9, 0
+; FLATSCR-NEXT:    s_mov_b32 s8, 0
 ; FLATSCR-NEXT:    ;;#ASMSTART
 ; FLATSCR-NEXT:    ;;#ASMEND
-; FLATSCR-NEXT:    scratch_load_dword v0, off, s9 offset:8 glc
+; FLATSCR-NEXT:    scratch_load_dword v0, off, s8 offset:8 glc
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    s_movk_i32 s8, 0x1004
 ; FLATSCR-NEXT:    scratch_store_dword off, v0, s8 ; 4-byte Folded Spill
 ; FLATSCR-NEXT:    ;;#ASMSTART
 ; FLATSCR-NEXT:    ;;#ASMEND
 ; FLATSCR-NEXT:    ;;#ASMSTART
 ; FLATSCR-NEXT:    ;;#ASMEND
 ; FLATSCR-NEXT:    ;;#ASMSTART

 ; MUBUF-NEXT:    ; v[0:1]
 ; MUBUF-NEXT:    ;;#ASMEND
 ; MUBUF-NEXT:    s_endpgm
 ;
 ; FLATSCR-LABEL: test_sgpr_offset_subregs_kernel:
 ; FLATSCR:       ; %bb.0: ; %entry
 ; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s0, s3
 ; FLATSCR-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
-; FLATSCR-NEXT:    s_mov_b32 s2, 0
-; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s2 offset:12 glc
+; FLATSCR-NEXT:    s_mov_b32 s0, 0
+; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s0 offset:12 glc
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    s_movk_i32 s0, 0xff8
-; FLATSCR-NEXT:    s_mov_b32 s1, 0
 ; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 ; 8-byte Folded Spill
+; FLATSCR-NEXT:    s_mov_b32 s0, 0
 ; FLATSCR-NEXT:    ;;#ASMSTART
 ; FLATSCR-NEXT:    ;;#ASMEND
-; FLATSCR-NEXT:    scratch_load_dword v0, off, s1 offset:8 glc
+; FLATSCR-NEXT:    scratch_load_dword v0, off, s0 offset:8 glc
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
+; FLATSCR-NEXT:    s_movk_i32 s0, 0xff8
 ; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s0 ; 8-byte Folded Reload
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    ;;#ASMSTART
 ; FLATSCR-NEXT:    ; v[0:1]
 ; FLATSCR-NEXT:    ;;#ASMEND
 ; FLATSCR-NEXT:    s_endpgm
 entry:
   ; Occupy 4088 bytes of scratch, so that the spill of the last subreg of %a

 ; MUBUF-NEXT:    ; v[0:1]
 ; MUBUF-NEXT:    ;;#ASMEND
 ; MUBUF-NEXT:    s_endpgm
 ;
 ; FLATSCR-LABEL: test_inst_offset_subregs_kernel:
 ; FLATSCR:       ; %bb.0: ; %entry
 ; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s0, s3
 ; FLATSCR-NEXT:    s_addc_u32 flat_scratch_hi, s1, 0
-; FLATSCR-NEXT:    s_mov_b32 s2, 0
-; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s2 offset:12 glc
+; FLATSCR-NEXT:    s_mov_b32 s0, 0
+; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s0 offset:12 glc
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    s_movk_i32 s0, 0xffc
-; FLATSCR-NEXT:    s_mov_b32 s1, 0
 ; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 ; 8-byte Folded Spill
+; FLATSCR-NEXT:    s_mov_b32 s0, 0
 ; FLATSCR-NEXT:    ;;#ASMSTART
 ; FLATSCR-NEXT:    ;;#ASMEND
-; FLATSCR-NEXT:    scratch_load_dword v0, off, s1 offset:8 glc
+; FLATSCR-NEXT:    scratch_load_dword v0, off, s0 offset:8 glc
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
+; FLATSCR-NEXT:    s_movk_i32 s0, 0xffc
 ; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s0 ; 8-byte Folded Reload
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    ;;#ASMSTART
 ; FLATSCR-NEXT:    ; v[0:1]
 ; FLATSCR-NEXT:    ;;#ASMEND
 ; FLATSCR-NEXT:    s_endpgm
 entry:
   ; Occupy 4092 bytes of scratch, so that the spill of the last subreg of %a

llvm/test/CodeGen/AMDGPU/spill-scavenge-offset.ll

 ; GFX6-NEXT:    s_mov_b32 s39, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 s[36:37], s[2:3]
 ; GFX6-NEXT:    v_lshlrev_b32_e32 v7, 8, v5
 ; GFX6-NEXT:    v_mov_b32_e32 v8, v6
 ; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:240
 ; GFX6-NEXT:    s_addc_u32 s41, s41, 0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x83400
-; GFX6-NEXT:    s_mov_b64 s[44:45], exec
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:224

 ; GFX6-NEXT:    buffer_store_dword v10, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v11, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v12, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_load_dwordx4 v[13:16], v[7:8], s[36:39], 0 addr64 offset:32
 ; GFX6-NEXT:    buffer_load_dwordx4 v[17:20], v[7:8], s[36:39], 0 addr64 offset:48
 ; GFX6-NEXT:    v_lshlrev_b32_e32 v4, 13, v0
 ; GFX6-NEXT:    v_add_i32_e32 v4, vcc, 16, v4
 ; GFX6-NEXT:    v_mov_b32_e32 v7, 1
+; GFX6-NEXT:    s_mov_b64 s[2:3], exec
 ; GFX6-NEXT:    buffer_store_dword v7, v4, s[40:43], 0 offen
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ; def s[4:11]
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    v_writelane_b32 v4, s4, 0
 ; GFX6-NEXT:    v_writelane_b32 v4, s5, 1
 ; GFX6-NEXT:    v_writelane_b32 v4, s6, 2
 ; GFX6-NEXT:    v_writelane_b32 v4, s7, 3
 ; GFX6-NEXT:    v_writelane_b32 v4, s8, 4
 ; GFX6-NEXT:    v_writelane_b32 v4, s9, 5
 ; GFX6-NEXT:    v_writelane_b32 v4, s10, 6
 ; GFX6-NEXT:    v_writelane_b32 v4, s11, 7
-; GFX6-NEXT:    s_mov_b32 s2, 0x83800
-; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT:    s_mov_b32 s12, 0x83800
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], s12 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
+; GFX6-NEXT:    s_mov_b64 exec, s[2:3]
 ; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ; def s[8:15]
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ; def s[16:23]
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    ;;#ASMSTART

 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ; def s33
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    s_and_saveexec_b64 s[34:35], vcc
 ; GFX6-NEXT:    s_cbranch_execz .LBB1_2
 ; GFX6-NEXT:  ; %bb.1: ; %bb0
 ; GFX6-NEXT:    s_mov_b64 s[44:45], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
-; GFX6-NEXT:    buffer_store_dword v9, off, s[40:43], 0
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_writelane_b32 v9, s8, 0
-; GFX6-NEXT:    v_writelane_b32 v9, s9, 1
-; GFX6-NEXT:    v_writelane_b32 v9, s10, 2
-; GFX6-NEXT:    v_writelane_b32 v9, s11, 3
-; GFX6-NEXT:    v_writelane_b32 v9, s12, 4
-; GFX6-NEXT:    v_writelane_b32 v9, s13, 5
-; GFX6-NEXT:    v_writelane_b32 v9, s14, 6
-; GFX6-NEXT:    v_writelane_b32 v9, s15, 7
-; GFX6-NEXT:    v_mov_b32_e32 v4, 0x2100
-; GFX6-NEXT:    buffer_store_dword v9, v4, s[40:43], 0 offen ; 4-byte Folded Spill
+; GFX6-NEXT:    v_writelane_b32 v4, s8, 0
+; GFX6-NEXT:    v_writelane_b32 v4, s9, 1
+; GFX6-NEXT:    v_writelane_b32 v4, s10, 2
+; GFX6-NEXT:    v_writelane_b32 v4, s11, 3
+; GFX6-NEXT:    v_writelane_b32 v4, s12, 4
+; GFX6-NEXT:    v_writelane_b32 v4, s13, 5
+; GFX6-NEXT:    v_writelane_b32 v4, s14, 6
+; GFX6-NEXT:    v_writelane_b32 v4, s15, 7
+; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2100
+; GFX6-NEXT:    buffer_store_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], 0
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
 ; GFX6-NEXT:    s_mov_b64 s[44:45], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
-; GFX6-NEXT:    v_mov_b32_e32 v4, 0x20e0
-; GFX6-NEXT:    buffer_store_dword v8, off, s[40:43], 0
+; GFX6-NEXT:    v_mov_b32_e32 v7, 0x20e0
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v8, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_readlane_b32 s8, v8, 0
-; GFX6-NEXT:    v_readlane_b32 s9, v8, 1
-; GFX6-NEXT:    v_readlane_b32 s10, v8, 2
-; GFX6-NEXT:    v_readlane_b32 s11, v8, 3
-; GFX6-NEXT:    v_readlane_b32 s12, v8, 4
-; GFX6-NEXT:    v_readlane_b32 s13, v8, 5
-; GFX6-NEXT:    v_readlane_b32 s14, v8, 6
-; GFX6-NEXT:    v_readlane_b32 s15, v8, 7
-; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], 0
+; GFX6-NEXT:    v_readlane_b32 s8, v4, 0
+; GFX6-NEXT:    v_readlane_b32 s9, v4, 1
+; GFX6-NEXT:    v_readlane_b32 s10, v4, 2
+; GFX6-NEXT:    v_readlane_b32 s11, v4, 3
+; GFX6-NEXT:    v_readlane_b32 s12, v4, 4
+; GFX6-NEXT:    v_readlane_b32 s13, v4, 5
+; GFX6-NEXT:    v_readlane_b32 s14, v4, 6
+; GFX6-NEXT:    v_readlane_b32 s15, v4, 7
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
 ; GFX6-NEXT:    s_mov_b64 s[44:45], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
-; GFX6-NEXT:    buffer_store_dword v7, off, s[40:43], 0
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_writelane_b32 v7, s16, 0
-; GFX6-NEXT:    v_writelane_b32 v7, s17, 1
-; GFX6-NEXT:    v_writelane_b32 v7, s18, 2
-; GFX6-NEXT:    v_writelane_b32 v7, s19, 3
-; GFX6-NEXT:    v_writelane_b32 v7, s20, 4
-; GFX6-NEXT:    v_writelane_b32 v7, s21, 5
-; GFX6-NEXT:    v_writelane_b32 v7, s22, 6
-; GFX6-NEXT:    v_writelane_b32 v7, s23, 7
-; GFX6-NEXT:    v_mov_b32_e32 v4, 0x2120
-; GFX6-NEXT:    buffer_store_dword v7, v4, s[40:43], 0 offen ; 4-byte Folded Spill
+; GFX6-NEXT:    v_writelane_b32 v4, s16, 0
+; GFX6-NEXT:    v_writelane_b32 v4, s17, 1
+; GFX6-NEXT:    v_writelane_b32 v4, s18, 2
+; GFX6-NEXT:    v_writelane_b32 v4, s19, 3
+; GFX6-NEXT:    v_writelane_b32 v4, s20, 4
+; GFX6-NEXT:    v_writelane_b32 v4, s21, 5
+; GFX6-NEXT:    v_writelane_b32 v4, s22, 6
+; GFX6-NEXT:    v_writelane_b32 v4, s23, 7
+; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2120
+; GFX6-NEXT:    buffer_store_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], 0
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
 ; GFX6-NEXT:    s_mov_b64 s[44:45], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
-; GFX6-NEXT:    v_mov_b32_e32 v4, 0x2100
-; GFX6-NEXT:    buffer_store_dword v9, off, s[40:43], 0
+; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2100
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v9, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_readlane_b32 s16, v9, 0
-; GFX6-NEXT:    v_readlane_b32 s17, v9, 1
-; GFX6-NEXT:    v_readlane_b32 s18, v9, 2
-; GFX6-NEXT:    v_readlane_b32 s19, v9, 3
-; GFX6-NEXT:    v_readlane_b32 s20, v9, 4
-; GFX6-NEXT:    v_readlane_b32 s21, v9, 5
-; GFX6-NEXT:    v_readlane_b32 s22, v9, 6
-; GFX6-NEXT:    v_readlane_b32 s23, v9, 7
-; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], 0
+; GFX6-NEXT:    v_readlane_b32 s16, v4, 0
+; GFX6-NEXT:    v_readlane_b32 s17, v4, 1
+; GFX6-NEXT:    v_readlane_b32 s18, v4, 2
+; GFX6-NEXT:    v_readlane_b32 s19, v4, 3
+; GFX6-NEXT:    v_readlane_b32 s20, v4, 4
+; GFX6-NEXT:    v_readlane_b32 s21, v4, 5
+; GFX6-NEXT:    v_readlane_b32 s22, v4, 6
+; GFX6-NEXT:    v_readlane_b32 s23, v4, 7
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
 ; GFX6-NEXT:    s_mov_b64 s[44:45], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
-; GFX6-NEXT:    buffer_store_dword v8, off, s[40:43], 0
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_writelane_b32 v8, s24, 0
-; GFX6-NEXT:    v_writelane_b32 v8, s25, 1
-; GFX6-NEXT:    v_writelane_b32 v8, s26, 2
-; GFX6-NEXT:    v_writelane_b32 v8, s27, 3
-; GFX6-NEXT:    v_writelane_b32 v8, s28, 4
-; GFX6-NEXT:    v_writelane_b32 v8, s29, 5
-; GFX6-NEXT:    v_writelane_b32 v8, s30, 6
-; GFX6-NEXT:    v_writelane_b32 v8, s31, 7
-; GFX6-NEXT:    v_mov_b32_e32 v4, 0x2140
-; GFX6-NEXT:    buffer_store_dword v8, v4, s[40:43], 0 offen ; 4-byte Folded Spill
+; GFX6-NEXT:    v_writelane_b32 v4, s24, 0
+; GFX6-NEXT:    v_writelane_b32 v4, s25, 1
+; GFX6-NEXT:    v_writelane_b32 v4, s26, 2
+; GFX6-NEXT:    v_writelane_b32 v4, s27, 3
+; GFX6-NEXT:    v_writelane_b32 v4, s28, 4
+; GFX6-NEXT:    v_writelane_b32 v4, s29, 5
+; GFX6-NEXT:    v_writelane_b32 v4, s30, 6
+; GFX6-NEXT:    v_writelane_b32 v4, s31, 7
+; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2140
+; GFX6-NEXT:    buffer_store_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], 0
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
 ; GFX6-NEXT:    s_mov_b64 s[44:45], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
-; GFX6-NEXT:    v_mov_b32_e32 v4, 0x2120
-; GFX6-NEXT:    buffer_store_dword v7, off, s[40:43], 0
+; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2120
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v7, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_readlane_b32 s24, v7, 0
-; GFX6-NEXT:    v_readlane_b32 s25, v7, 1
-; GFX6-NEXT:    v_readlane_b32 s26, v7, 2
-; GFX6-NEXT:    v_readlane_b32 s27, v7, 3
-; GFX6-NEXT:    v_readlane_b32 s28, v7, 4
-; GFX6-NEXT:    v_readlane_b32 s29, v7, 5
-; GFX6-NEXT:    v_readlane_b32 s30, v7, 6
-; GFX6-NEXT:    v_readlane_b32 s31, v7, 7
-; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], 0
+; GFX6-NEXT:    v_readlane_b32 s24, v4, 0
+; GFX6-NEXT:    v_readlane_b32 s25, v4, 1
+; GFX6-NEXT:    v_readlane_b32 s26, v4, 2
+; GFX6-NEXT:    v_readlane_b32 s27, v4, 3
+; GFX6-NEXT:    v_readlane_b32 s28, v4, 4
+; GFX6-NEXT:    v_readlane_b32 s29, v4, 5
+; GFX6-NEXT:    v_readlane_b32 s30, v4, 6
+; GFX6-NEXT:    v_readlane_b32 s31, v4, 7
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
 ; GFX6-NEXT:    s_mov_b64 s[44:45], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 15
-; GFX6-NEXT:    buffer_store_dword v10, off, s[40:43], 0
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_writelane_b32 v10, s0, 0
-; GFX6-NEXT:    v_writelane_b32 v10, s1, 1
-; GFX6-NEXT:    v_writelane_b32 v10, s2, 2
-; GFX6-NEXT:    v_writelane_b32 v10, s3, 3
-; GFX6-NEXT:    v_mov_b32_e32 v4, 0x2160
-; GFX6-NEXT:    buffer_store_dword v10, v4, s[40:43], 0 offen ; 4-byte Folded Spill
+; GFX6-NEXT:    v_writelane_b32 v4, s0, 0
+; GFX6-NEXT:    v_writelane_b32 v4, s1, 1
+; GFX6-NEXT:    v_writelane_b32 v4, s2, 2
+; GFX6-NEXT:    v_writelane_b32 v4, s3, 3
+; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2160
+; GFX6-NEXT:    buffer_store_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v10, off, s[40:43], 0
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
-; GFX6-NEXT:    s_mov_b64 s[44:45], exec
+; GFX6-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 15
-; GFX6-NEXT:    buffer_store_dword v8, off, s[40:43], 0
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_writelane_b32 v8, s4, 0
-; GFX6-NEXT:    v_writelane_b32 v8, s5, 1
-; GFX6-NEXT:    v_writelane_b32 v8, s6, 2
-; GFX6-NEXT:    v_writelane_b32 v8, s7, 3
-; GFX6-NEXT:    s_mov_b32 s0, 0x85c00
-; GFX6-NEXT:    buffer_store_dword v8, off, s[40:43], s0 ; 4-byte Folded Spill
+; GFX6-NEXT:    v_writelane_b32 v4, s4, 0
+; GFX6-NEXT:    v_writelane_b32 v4, s5, 1
+; GFX6-NEXT:    v_writelane_b32 v4, s6, 2
+; GFX6-NEXT:    v_writelane_b32 v4, s7, 3
+; GFX6-NEXT:    s_mov_b32 s44, 0x85c00
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], s44 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], 0
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
+; GFX6-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX6-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 3
-; GFX6-NEXT:    buffer_store_dword v9, off, s[40:43], 0
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_writelane_b32 v9, s2, 0
-; GFX6-NEXT:    v_writelane_b32 v9, s3, 1
+; GFX6-NEXT:    v_writelane_b32 v4, s2, 0
+; GFX6-NEXT:    v_writelane_b32 v4, s3, 1
 ; GFX6-NEXT:    s_mov_b32 s4, 0x86600
-; GFX6-NEXT:    buffer_store_dword v9, off, s[40:43], s4 ; 4-byte Folded Spill
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], s4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], 0
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX6-NEXT:    s_mov_b64 s[44:45], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
-; GFX6-NEXT:    v_mov_b32_e32 v4, 0x2140
-; GFX6-NEXT:    buffer_store_dword v7, off, s[40:43], 0
+; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2140
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v7, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_readlane_b32 s0, v7, 0
-; GFX6-NEXT:    v_readlane_b32 s1, v7, 1
-; GFX6-NEXT:    v_readlane_b32 s2, v7, 2
-; GFX6-NEXT:    v_readlane_b32 s3, v7, 3
-; GFX6-NEXT:    v_readlane_b32 s4, v7, 4
-; GFX6-NEXT:    v_readlane_b32 s5, v7, 5
-; GFX6-NEXT:    v_readlane_b32 s6, v7, 6
-; GFX6-NEXT:    v_readlane_b32 s7, v7, 7
-; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], 0
+; GFX6-NEXT:    v_readlane_b32 s0, v4, 0
+; GFX6-NEXT:    v_readlane_b32 s1, v4, 1
+; GFX6-NEXT:    v_readlane_b32 s2, v4, 2
+; GFX6-NEXT:    v_readlane_b32 s3, v4, 3
+; GFX6-NEXT:    v_readlane_b32 s4, v4, 4
+; GFX6-NEXT:    v_readlane_b32 s5, v4, 5
+; GFX6-NEXT:    v_readlane_b32 s6, v4, 6
+; GFX6-NEXT:    v_readlane_b32 s7, v4, 7
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
 ; GFX6-NEXT:    s_mov_b64 s[44:45], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 15
-; GFX6-NEXT:    buffer_store_dword v8, off, s[40:43], 0
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_writelane_b32 v8, s36, 0
-; GFX6-NEXT:    v_writelane_b32 v8, s37, 1
-; GFX6-NEXT:    v_writelane_b32 v8, s38, 2
-; GFX6-NEXT:    v_writelane_b32 v8, s39, 3
-; GFX6-NEXT:    v_mov_b32_e32 v4, 0x2180
-; GFX6-NEXT:    buffer_store_dword v8, v4, s[40:43], 0 offen ; 4-byte Folded Spill
+; GFX6-NEXT:    v_writelane_b32 v4, s36, 0
+; GFX6-NEXT:    v_writelane_b32 v4, s37, 1
+; GFX6-NEXT:    v_writelane_b32 v4, s38, 2
+; GFX6-NEXT:    v_writelane_b32 v4, s39, 3
+; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2180
+; GFX6-NEXT:    buffer_store_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], 0
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
 ; GFX6-NEXT:    s_mov_b64 s[38:39], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 3
-; GFX6-NEXT:    buffer_store_dword v10, off, s[40:43], 0
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_writelane_b32 v10, s36, 0
-; GFX6-NEXT:    v_writelane_b32 v10, s37, 1
+; GFX6-NEXT:    v_writelane_b32 v4, s36, 0
+; GFX6-NEXT:    v_writelane_b32 v4, s37, 1
 ; GFX6-NEXT:    s_mov_b32 s44, 0x86400
-; GFX6-NEXT:    buffer_store_dword v10, off, s[40:43], s44 ; 4-byte Folded Spill
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], s44 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v10, off, s[40:43], 0
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[38:39]
 ; GFX6-NEXT:    s_mov_b64 s[44:45], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 15
-; GFX6-NEXT:    v_mov_b32_e32 v4, 0x2170
-; GFX6-NEXT:    buffer_store_dword v9, off, s[40:43], 0
+; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2170
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v9, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_readlane_b32 s36, v9, 0
-; GFX6-NEXT:    v_readlane_b32 s37, v9, 1
-; GFX6-NEXT:    v_readlane_b32 s38, v9, 2
-; GFX6-NEXT:    v_readlane_b32 s39, v9, 3
-; GFX6-NEXT:    buffer_load_dword v9, off, s[40:43], 0
+; GFX6-NEXT:    v_readlane_b32 s36, v4, 0
+; GFX6-NEXT:    v_readlane_b32 s37, v4, 1
+; GFX6-NEXT:    v_readlane_b32 s38, v4, 2
+; GFX6-NEXT:    v_readlane_b32 s39, v4, 3
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[44:45]
 ; GFX6-NEXT:    s_not_b64 exec, exec
-; GFX6-NEXT:    v_mov_b32_e32 v4, 0x2190
-; GFX6-NEXT:    buffer_store_dword v7, off, s[40:43], 0
+; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2190
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v7, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_not_b64 exec, exec
-; GFX6-NEXT:    v_mov_b32_e32 v4, 0x2190
-; GFX6-NEXT:    buffer_load_dword v7, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2190
+; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_not_b64 exec, exec
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_readlane_b32 s44, v7, 0
-; GFX6-NEXT:    v_readlane_b32 s45, v7, 1
-; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], 0
+; GFX6-NEXT:    v_readlane_b32 s44, v4, 0
+; GFX6-NEXT:    v_readlane_b32 s45, v4, 1
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_not_b64 exec, exec
 ; GFX6-NEXT:    s_mov_b64 vcc, s[34:35]
 ; GFX6-NEXT:    s_not_b64 exec, exec
-; GFX6-NEXT:    v_mov_b32_e32 v4, 0x2198
-; GFX6-NEXT:    buffer_store_dword v8, off, s[40:43], 0
+; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2198
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v8, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_not_b64 exec, exec
-; GFX6-NEXT:    v_mov_b32_e32 v4, 0x2198
-; GFX6-NEXT:    buffer_load_dword v8, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT:    v_mov_b32_e32 v7, 0x2198
+; GFX6-NEXT:    buffer_load_dword v4, v7, s[40:43], 0 offen ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_not_b64 exec, exec
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_readlane_b32 s34, v8, 0
-; GFX6-NEXT:    v_readlane_b32 s35, v8, 1
-; GFX6-NEXT:    buffer_load_dword v8, off, s[40:43], 0
+; GFX6-NEXT:    v_readlane_b32 s34, v4, 0
+; GFX6-NEXT:    v_readlane_b32 s35, v4, 1
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_not_b64 exec, exec
 ; GFX6-NEXT:    ;;#ASMSTART
 ; GFX6-NEXT:    ; use s[8:15],s[16:23],s[24:31],s[0:7],s[36:39],s[34:35],s[44:45]
 ; GFX6-NEXT:    ;;#ASMEND
 ; GFX6-NEXT:    s_mov_b64 s[34:35], vcc
-; GFX6-NEXT:    s_mov_b64 s[8:9], exec
+; GFX6-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 15
-; GFX6-NEXT:    s_mov_b32 s0, 0x86000
+; GFX6-NEXT:    s_mov_b32 s2, 0x86000
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], s0 ; 4-byte Folded Reload
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], s2 ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    v_readlane_b32 s36, v4, 0
 ; GFX6-NEXT:    v_readlane_b32 s37, v4, 1
 ; GFX6-NEXT:    v_readlane_b32 s38, v4, 2
 ; GFX6-NEXT:    v_readlane_b32 s39, v4, 3
 ; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    s_mov_b64 exec, s[8:9]
+; GFX6-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX6-NEXT:    s_mov_b64 s[4:5], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 15
 ; GFX6-NEXT:    s_mov_b32 s6, 0x85800
-; GFX6-NEXT:    buffer_store_dword v7, off, s[40:43], 0
+; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], s6 ; 4-byte Folded Reload
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], s6 ; 4-byte Folded Reload
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
-; GFX6-NEXT:    v_readlane_b32 s0, v7, 0
-; GFX6-NEXT:    v_readlane_b32 s1, v7, 1
-; GFX6-NEXT:    v_readlane_b32 s2, v7, 2
-; GFX6-NEXT:    v_readlane_b32 s3, v7, 3
-; GFX6-NEXT:    buffer_load_dword v7, off, s[40:43], 0
+; GFX6-NEXT:    v_readlane_b32 s0, v4, 0
+; GFX6-NEXT:    v_readlane_b32 s1, v4, 1
+; GFX6-NEXT:    v_readlane_b32 s2, v4, 2
+; GFX6-NEXT:    v_readlane_b32 s3, v4, 3
+; GFX6-NEXT:    buffer_load_dword v4, off, s[40:43], 0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX6-NEXT:    s_mov_b32 s2, 0x83800
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
 ; GFX6-NEXT:    buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill

llvm/test/CodeGen/AMDGPU/spill-special-sgpr.mir

bb.0:
     ; GFX9-NEXT: $sgpr33 = S_MOV_B32 0
     ; GFX9-NEXT: $sgpr12 = S_MOV_B32 &SCRATCH_RSRC_DWORD0, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
     ; GFX9-NEXT: $sgpr13 = S_MOV_B32 &SCRATCH_RSRC_DWORD1, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
     ; GFX9-NEXT: $sgpr14 = S_MOV_B32 4294967295, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
     ; GFX9-NEXT: $sgpr15 = S_MOV_B32 14680064, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
     ; GFX9-NEXT: $sgpr12 = S_ADD_U32 $sgpr12, $sgpr9, implicit-def $scc, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
     ; GFX9-NEXT: $sgpr13 = S_ADDC_U32 $sgpr13, 0, implicit-def dead $scc, implicit $scc, implicit-def $sgpr12_sgpr13_sgpr14_sgpr15
     ; GFX9-NEXT: $vcc = IMPLICIT_DEF
-    ; GFX9-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
-    ; GFX9-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr2
-    ; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GFX9-NEXT: $vgpr2 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr2, implicit $vcc
-    ; GFX9-NEXT: $vgpr2 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr2, implicit $vcc
-    ; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
-    ; GFX9-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GFX9-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr2
+    ; GFX9-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GFX9-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
+    ; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GFX9-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
+    ; GFX9-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit $vcc
+    ; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
+    ; GFX9-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GFX9-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GFX9-NEXT: $vcc = IMPLICIT_DEF
-    ; GFX9-NEXT: $sgpr2_sgpr3 = S_MOV_B64 $exec
-    ; GFX9-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr1
-    ; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GFX9-NEXT: $vgpr1 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr1, implicit $vcc
-    ; GFX9-NEXT: $vgpr1 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr1, implicit killed $vcc
-    ; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
-    ; GFX9-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GFX9-NEXT: $exec = S_MOV_B64 killed $sgpr2_sgpr3, implicit killed $vgpr1
+    ; GFX9-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GFX9-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
+    ; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GFX9-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
+    ; GFX9-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit killed $vcc
+    ; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
+    ; GFX9-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GFX9-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GFX9-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GFX9-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
     ; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GFX9-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 4, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
     ; GFX9-NEXT: $vcc_lo = V_READLANE_B32 $vgpr0, 0, implicit-def $vcc
     ; GFX9-NEXT: $vcc_hi = V_READLANE_B32 killed $vgpr0, 1
     ; GFX9-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr12_sgpr13_sgpr14_sgpr15, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GFX9-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GFX10-LABEL: name: check_vcc
     ; GFX10: liveins: $sgpr8, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr9
     ; GFX10-NEXT: {{  $}}
     ; GFX10-NEXT: $sgpr33 = S_MOV_B32 0
     ; GFX10-NEXT: $sgpr96 = S_MOV_B32 &SCRATCH_RSRC_DWORD0, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GFX10-NEXT: $sgpr97 = S_MOV_B32 &SCRATCH_RSRC_DWORD1, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GFX10-NEXT: $sgpr98 = S_MOV_B32 4294967295, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GFX10-NEXT: $sgpr99 = S_MOV_B32 836853760, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GFX10-NEXT: $sgpr96 = S_ADD_U32 $sgpr96, $sgpr9, implicit-def $scc, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GFX10-NEXT: $sgpr97 = S_ADDC_U32 $sgpr97, 0, implicit-def dead $scc, implicit $scc, implicit-def $sgpr96_sgpr97_sgpr98_sgpr99
     ; GFX10-NEXT: $vcc = IMPLICIT_DEF
-    ; GFX10-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
-    ; GFX10-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr2
-    ; GFX10-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GFX10-NEXT: $vgpr2 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr2, implicit $vcc
-    ; GFX10-NEXT: $vgpr2 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr2, implicit $vcc
-    ; GFX10-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr2, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
-    ; GFX10-NEXT: $vgpr2 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GFX10-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr2
+    ; GFX10-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GFX10-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
+    ; GFX10-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GFX10-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
+    ; GFX10-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit $vcc
+    ; GFX10-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
+    ; GFX10-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GFX10-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GFX10-NEXT: $vcc = IMPLICIT_DEF
-    ; GFX10-NEXT: $sgpr2_sgpr3 = S_MOV_B64 $exec
-    ; GFX10-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr1
-    ; GFX10-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GFX10-NEXT: $vgpr1 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr1, implicit $vcc
-    ; GFX10-NEXT: $vgpr1 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr1, implicit killed $vcc
-    ; GFX10-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
-    ; GFX10-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GFX10-NEXT: $exec = S_MOV_B64 killed $sgpr2_sgpr3, implicit killed $vgpr1
+    ; GFX10-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GFX10-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
+    ; GFX10-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GFX10-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
+    ; GFX10-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit killed $vcc
+    ; GFX10-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0, addrspace 5)
+    ; GFX10-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GFX10-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GFX10-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GFX10-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
     ; GFX10-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr0, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GFX10-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 4, 0, 0, implicit $exec :: (load (s32) from %stack.0, addrspace 5)
     ; GFX10-NEXT: $vcc_lo = V_READLANE_B32 $vgpr0, 0, implicit-def $vcc
     ; GFX10-NEXT: $vcc_hi = V_READLANE_B32 killed $vgpr0, 1
     ; GFX10-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFSET $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr33, 0, 0, 0, implicit $exec :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GFX10-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GFX11-LABEL: name: check_vcc
     ; GFX11: liveins: $sgpr8, $sgpr4_sgpr5, $sgpr6_sgpr7
     ; GFX11-NEXT: {{  $}}
     ; GFX11-NEXT: $sgpr33 = S_MOV_B32 0
     ; GFX11-NEXT: $vcc = IMPLICIT_DEF
-    ; GFX11-NEXT: $sgpr4_sgpr5 = S_MOV_B64 $exec
-    ; GFX11-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr2
-    ; GFX11-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr2, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GFX11-NEXT: $vgpr2 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr2, implicit $vcc
-    ; GFX11-NEXT: $vgpr2 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr2, implicit $vcc
-    ; GFX11-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr2, $sgpr33, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.0, addrspace 5)
-    ; GFX11-NEXT: $vgpr2 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GFX11-NEXT: $exec = S_MOV_B64 killed $sgpr4_sgpr5, implicit killed $vgpr2
+    ; GFX11-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GFX11-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
+    ; GFX11-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GFX11-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
+    ; GFX11-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit $vcc
+    ; GFX11-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.0, addrspace 5)
+    ; GFX11-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GFX11-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GFX11-NEXT: $vcc = IMPLICIT_DEF
-    ; GFX11-NEXT: $sgpr2_sgpr3 = S_MOV_B64 $exec
-    ; GFX11-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr1
-    ; GFX11-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr1, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
-    ; GFX11-NEXT: $vgpr1 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr1, implicit $vcc
-    ; GFX11-NEXT: $vgpr1 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr1, implicit killed $vcc
-    ; GFX11-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr1, $sgpr33, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.0, addrspace 5)
-    ; GFX11-NEXT: $vgpr1 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
-    ; GFX11-NEXT: $exec = S_MOV_B64 killed $sgpr2_sgpr3, implicit killed $vgpr1
+    ; GFX11-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
+    ; GFX11-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
+    ; GFX11-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
+    ; GFX11-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_lo, 0, undef $vgpr0, implicit $vcc
+    ; GFX11-NEXT: $vgpr0 = V_WRITELANE_B32 $vcc_hi, 1, $vgpr0, implicit killed $vcc
+    ; GFX11-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.0, addrspace 5)
+    ; GFX11-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
+    ; GFX11-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     ; GFX11-NEXT: $sgpr0_sgpr1 = S_MOV_B64 $exec
     ; GFX11-NEXT: $exec = S_MOV_B64 3, implicit-def $vgpr0
     ; GFX11-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr0, $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %fixed-stack.0, align 16, addrspace 5)
     ; GFX11-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.0, addrspace 5)
     ; GFX11-NEXT: $vcc_lo = V_READLANE_B32 $vgpr0, 0, implicit-def $vcc
     ; GFX11-NEXT: $vcc_hi = V_READLANE_B32 killed $vgpr0, 1
     ; GFX11-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD_SADDR $sgpr33, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %fixed-stack.0, align 16, addrspace 5)
     ; GFX11-NEXT: $exec = S_MOV_B64 killed $sgpr0_sgpr1, implicit killed $vgpr0
     $vcc = IMPLICIT_DEF
     SI_SPILL_S64_SAVE $vcc, %stack.0, implicit $exec, implicit $sgpr96_sgpr97_sgpr98_sgpr99, implicit $sgpr32
 
     $vcc = IMPLICIT_DEF
     SI_SPILL_S64_SAVE killed $vcc, %stack.0, implicit $exec, implicit $sgpr96_sgpr97_sgpr98_sgpr99, implicit $sgpr32
 
     $vcc = SI_SPILL_S64_RESTORE %stack.0, implicit $exec, implicit $sgpr96_sgpr97_sgpr98_sgpr99, implicit $sgpr32
 ...

llvm/test/CodeGen/AMDGPU/stack-pointer-offset-relative-frameindex.ll

 ; FLATSCR-NEXT:    s_addc_u32 s1, s1, svm_eval_nodes@rel32@hi+12
 ; FLATSCR-NEXT:    s_waitcnt lgkmcnt(0)
 ; FLATSCR-NEXT:    v_mov_b32_e32 v0, s2
 ; FLATSCR-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; FLATSCR-NEXT:    v_cmp_ne_u32_e32 vcc_lo, 0, v0
 ; FLATSCR-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; FLATSCR-NEXT:    s_cbranch_execz .LBB0_2
 ; FLATSCR-NEXT:  ; %bb.1: ; %if.then4.i
-; FLATSCR-NEXT:    s_movk_i32 s1, 0x4000
+; FLATSCR-NEXT:    s_movk_i32 s0, 0x4000
+; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s0 offset:4
+; FLATSCR-NEXT:    s_waitcnt_depctr 0xffe3
 ; FLATSCR-NEXT:    s_mov_b32 s0, 0x41c64e6d
-; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s1 offset:4
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR-NEXT:    v_add_nc_u32_e32 v0, v1, v0
 ; FLATSCR-NEXT:    v_mad_u64_u32 v[0:1], s0, v0, s0, 0x3039
 ; FLATSCR-NEXT:    scratch_store_dword off, v0, s0
 ; FLATSCR-NEXT:  .LBB0_2: ; %shader_eval_surface.exit
 ; FLATSCR-NEXT:    s_endpgm
 ;
 ; MUBUF11-LABEL: kernel_background_evaluate:
 ; MUBUF11:       ; %bb.0: ; %entry
 ; MUBUF11-NEXT:    s_load_b32 s2, s[0:1], 0x24
 ; MUBUF11-NEXT:    v_mov_b32_e32 v1, 0x2000
 ; MUBUF11-NEXT:    v_dual_mov_b32 v2, 0x4000 :: v_dual_mov_b32 v3, 0
 ; MUBUF11-NEXT:    v_mov_b32_e32 v4, 0x400000
 ; MUBUF11-NEXT:    s_movk_i32 s32, 0x6000
 ; MUBUF11-NEXT:    s_getpc_b64 s[0:1]
 ; MUBUF11-NEXT:    s_add_u32 s0, s0, svm_eval_nodes@rel32@lo+4
 ; MUBUF11-NEXT:    s_addc_u32 s1, s1, svm_eval_nodes@rel32@hi+12
 ; MUBUF11-NEXT:    s_waitcnt lgkmcnt(0)
 ; MUBUF11-NEXT:    v_mov_b32_e32 v0, s2
 ; MUBUF11-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; MUBUF11-NEXT:    s_mov_b32 s0, exec_lo
 ; MUBUF11-NEXT:    v_cmpx_ne_u32_e32 0, v0
 ; MUBUF11-NEXT:    s_cbranch_execz .LBB0_2
 ; MUBUF11-NEXT:  ; %bb.1: ; %if.then4.i
-; MUBUF11-NEXT:    s_movk_i32 s1, 0x4000
+; MUBUF11-NEXT:    s_movk_i32 s0, 0x4000
+; MUBUF11-NEXT:    scratch_load_b64 v[0:1], off, s0 offset:4
 ; MUBUF11-NEXT:    s_mov_b32 s0, 0x41c64e6d
-; MUBUF11-NEXT:    scratch_load_b64 v[0:1], off, s1 offset:4
 ; MUBUF11-NEXT:    s_waitcnt vmcnt(0)
 ; MUBUF11-NEXT:    v_add_nc_u32_e32 v2, v1, v0
 ; MUBUF11-NEXT:    v_mad_u64_u32 v[0:1], null, v2, s0, 0x3039
 ; MUBUF11-NEXT:    scratch_store_b32 off, v0, s0
 ; MUBUF11-NEXT:  .LBB0_2: ; %shader_eval_surface.exit
 ; MUBUF11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; MUBUF11-NEXT:    s_endpgm
 ;

 ; FLATSCR11-NEXT:    s_addc_u32 s1, s1, svm_eval_nodes@rel32@hi+12
 ; FLATSCR11-NEXT:    s_waitcnt lgkmcnt(0)
 ; FLATSCR11-NEXT:    v_mov_b32_e32 v0, s2
 ; FLATSCR11-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; FLATSCR11-NEXT:    s_mov_b32 s0, exec_lo
 ; FLATSCR11-NEXT:    v_cmpx_ne_u32_e32 0, v0
 ; FLATSCR11-NEXT:    s_cbranch_execz .LBB0_2
 ; FLATSCR11-NEXT:  ; %bb.1: ; %if.then4.i
-; FLATSCR11-NEXT:    s_movk_i32 s1, 0x4000
+; FLATSCR11-NEXT:    s_movk_i32 s0, 0x4000
+; FLATSCR11-NEXT:    scratch_load_b64 v[0:1], off, s0 offset:4
 ; FLATSCR11-NEXT:    s_mov_b32 s0, 0x41c64e6d
-; FLATSCR11-NEXT:    scratch_load_b64 v[0:1], off, s1 offset:4
 ; FLATSCR11-NEXT:    s_waitcnt vmcnt(0)
 ; FLATSCR11-NEXT:    v_add_nc_u32_e32 v2, v1, v0
 ; FLATSCR11-NEXT:    v_mad_u64_u32 v[0:1], null, v2, s0, 0x3039
 ; FLATSCR11-NEXT:    scratch_store_b32 off, v0, s0
 ; FLATSCR11-NEXT:  .LBB0_2: ; %shader_eval_surface.exit
 ; FLATSCR11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; FLATSCR11-NEXT:    s_endpgm
 entry: