[AMDGPU] Fix SGPR checks in S_MOV_B64_IMM_PSEUDO generation.

[abinavpp]

Summary

The function to generate S_MOV_B64_IMM_PSEUDO was recently modified to
optimize AGPR to AGPR copy but it missed checking for the SGPR
clobbering for the S_MOV_B64_IMM_PSEUDO generation.

[abinavpp]

This fixes the codegen of test_rocrand_kernel_xorwow.cpp of rocRAND (tracked by SWDEV-306338).

[arsenm]

llvm/lib/Target/AMDGPU/GCNPreRAOptimizations.cpp

96–97

Can you just change the range loop to preincrement the iterator?

102–116

Checking for specific opcodes with specific subreg indices is definitely not the right way to check for register interference.

[vangthao]

llvm/lib/Target/AMDGPU/GCNPreRAOptimizations.cpp

102

Why do we check for COPY here? If Reg is not AGPR then we will not touch any COPY instructions.

119

Is there a reason why we need a new loop to check this?

This should do the same as the removed lines of code below:

if (Def0)
  return false;

if (Def1)
  return false;

As in if there are more than one def of a subreg then bail.

[rampitec]

llvm/lib/Target/AMDGPU/GCNPreRAOptimizations.cpp

178

If you do a separate loop I do not understand why this handling is still in the second loop.

[abinavpp]

We don't need the new loop. Sorry for the noise. We just need to make sure that
we don't do the S_MOV_B64_IMM_PSEUDO generation when we have an SGPR COPY like
how D104874 did.

[vangthao]

LGTM.

llvm/lib/Target/AMDGPU/GCNPreRAOptimizations.cpp

 
 FunctionPass *llvm::createGCNPreRAOptimizationsPass() {
   return new GCNPreRAOptimizations();
 }
 
 bool GCNPreRAOptimizations::processReg(Register Reg) {
   MachineInstr *Def0 = nullptr;
   MachineInstr *Def1 = nullptr;
-  uint64_t Init = 0;
+  uint64_t Init = 0, NumDef0 = 0, NumDef1 = 0;
   bool Changed = false;
   SmallSet<Register, 32> ModifiedRegs;
   bool IsAGPRDst = TRI->isAGPRClass(MRI->getRegClass(Reg));
 
+  // Bail out if emitting AMDGPU::S_MOV_B64_IMM_PSEUDO will clobber the SGPR
+  // (sub)reg.
+  //
+  // Note: It's tricky to do this in the next def_instructions() loop since the
+  // AMDGPU::COPY handling there can potentially modify the IR. TODO: Can we
+  // move the AGPR copy optimization somewhere else?
+  if (!IsAGPRDst) {
+    for (MachineInstr &I : MRI->def_instructions(Reg)) {
+      unsigned Opc = I.getOpcode();
+      if (Opc == AMDGPU::COPY || Opc == AMDGPU::S_MOV_B32) {
+        switch (I.getOperand(0).getSubReg()) {
+        default:
+          return false;
+
+        case AMDGPU::sub0:
+          ++NumDef0;
+          break;
+
+        case AMDGPU::sub1:
+          ++NumDef1;
+          break;
+        }
+      }
+    }
+
+    if (NumDef0 > 1 || NumDef1 > 1)
+      return false;
+  }
+
   for (MachineInstr &I : MRI->def_instructions(Reg)) {
     switch (I.getOpcode()) {
     default:
       return false;
     case AMDGPU::V_ACCVGPR_WRITE_B32_e64:
       break;
     case AMDGPU::COPY: {
       // Some subtargets cannot do an AGPR to AGPR copy directly, and need an
case AMDGPU::COPY: {
           }
 
           // Found the defining accvgpr_write, stop looking any further.
           break;
         }
       }
       break;
     }
     case AMDGPU::S_MOV_B32:
       if (I.getOperand(0).getReg() != Reg || !I.getOperand(1).isImm() ||
           I.getNumOperands() != 2)
         return false;
 
       switch (I.getOperand(0).getSubReg()) {
       default:
         return false;
       case AMDGPU::sub0:
-        if (Def0)
-          return false;
         Def0 = &I;
         Init |= I.getOperand(1).getImm() & 0xffffffff;
         break;
       case AMDGPU::sub1:
-        if (Def1)
-          return false;
         Def1 = &I;
         Init |= static_cast<uint64_t>(I.getOperand(1).getImm()) << 32;
         break;
       }
       break;
     }
   }
 

llvm/test/CodeGen/AMDGPU/agpr-to-agpr-copy.mir

 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 # RUN: llc -march=amdgcn -mcpu=gfx908 -run-pass=liveintervals,amdgpu-pre-ra-optimizations -verify-machineinstrs %s -o - | FileCheck -check-prefix=GFX908 %s
 
 ---
 name: test_mfma_f32_4x4x1f32_propagate_vgpr
 tracksRegLiveness: true
 
 body: |
   bb.0:
     liveins: $sgpr0_sgpr1
     ; GFX908-LABEL: name: test_mfma_f32_4x4x1f32_propagate_vgpr
     ; GFX908: liveins: $sgpr0_sgpr1
-    ; GFX908: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr0_sgpr1
-    ; GFX908: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[COPY]](p4), 36, 0 :: (dereferenceable invariant load (s64), align 4, addrspace 4)
-    ; GFX908: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
-    ; GFX908: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 1123418112, implicit $exec
-    ; GFX908: undef %4.sub0:areg_128 = V_ACCVGPR_WRITE_B32_e64 [[V_MOV_B32_e32_1]], implicit $exec
-    ; GFX908: %4.sub1:areg_128 = COPY [[V_MOV_B32_e32_1]]
-    ; GFX908: %4.sub2:areg_128 = COPY [[V_MOV_B32_e32_1]]
-    ; GFX908: %4.sub3:areg_128 = COPY [[V_MOV_B32_e32_1]]
-    ; GFX908: [[V_MOV_B32_e32_2:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 1073741824, implicit $exec
-    ; GFX908: [[V_MOV_B32_e32_3:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 1065353216, implicit $exec
-    ; GFX908: [[V_MFMA_F32_4X4X1F32_e64_:%[0-9]+]]:areg_128 = V_MFMA_F32_4X4X1F32_e64 [[V_MOV_B32_e32_3]], [[V_MOV_B32_e32_2]], %4, 0, 0, 0, implicit $mode, implicit $exec
-    ; GFX908: [[COPY1:%[0-9]+]]:vreg_128 = COPY [[V_MFMA_F32_4X4X1F32_e64_]]
-    ; GFX908: GLOBAL_STORE_DWORDX4_SADDR [[V_MOV_B32_e32_]], [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s128), addrspace 1)
-    ; GFX908: S_ENDPGM 0
+    ; GFX908-NEXT: {{  $}}
+    ; GFX908-NEXT: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr0_sgpr1
+    ; GFX908-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[COPY]](p4), 36, 0 :: (dereferenceable invariant load (s64), align 4, addrspace 4)
+    ; GFX908-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
+    ; GFX908-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 1123418112, implicit $exec
+    ; GFX908-NEXT: undef %4.sub0:areg_128 = V_ACCVGPR_WRITE_B32_e64 [[V_MOV_B32_e32_1]], implicit $exec
+    ; GFX908-NEXT: %4.sub1:areg_128 = COPY [[V_MOV_B32_e32_1]]
+    ; GFX908-NEXT: %4.sub2:areg_128 = COPY [[V_MOV_B32_e32_1]]
+    ; GFX908-NEXT: %4.sub3:areg_128 = COPY [[V_MOV_B32_e32_1]]
+    ; GFX908-NEXT: [[V_MOV_B32_e32_2:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 1073741824, implicit $exec
+    ; GFX908-NEXT: [[V_MOV_B32_e32_3:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 1065353216, implicit $exec
+    ; GFX908-NEXT: [[V_MFMA_F32_4X4X1F32_e64_:%[0-9]+]]:areg_128 = V_MFMA_F32_4X4X1F32_e64 [[V_MOV_B32_e32_3]], [[V_MOV_B32_e32_2]], %4, 0, 0, 0, implicit $mode, implicit $exec
+    ; GFX908-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY [[V_MFMA_F32_4X4X1F32_e64_]]
+    ; GFX908-NEXT: GLOBAL_STORE_DWORDX4_SADDR [[V_MOV_B32_e32_]], [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s128), addrspace 1)
+    ; GFX908-NEXT: S_ENDPGM 0
     %1:sgpr_64(p4) = COPY $sgpr0_sgpr1
     %4:sreg_64_xexec = S_LOAD_DWORDX2_IMM %1:sgpr_64(p4), 36, 0 :: (dereferenceable invariant load (s64), align 4, addrspace 4)
     %5:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
     %13:vgpr_32 = V_MOV_B32_e32 1123418112, implicit $exec
     undef %11.sub0:areg_128 = V_ACCVGPR_WRITE_B32_e64 %13:vgpr_32, implicit $exec
     %11.sub1:areg_128 = COPY %11.sub0:areg_128
     %11.sub2:areg_128 = COPY %11.sub0:areg_128
     %11.sub3:areg_128 = COPY %11.sub0:areg_128
     %8:vgpr_32 = V_MOV_B32_e32 1073741824, implicit $exec
     %9:vgpr_32 = V_MOV_B32_e32 1065353216, implicit $exec
     %10:areg_128 = V_MFMA_F32_4X4X1F32_e64 %9:vgpr_32, %8:vgpr_32, %11:areg_128, 0, 0, 0, implicit $mode, implicit $exec
     %12:vreg_128 = COPY %10:areg_128
     GLOBAL_STORE_DWORDX4_SADDR %5:vgpr_32, %12:vreg_128, %4:sreg_64_xexec, 0, 0, implicit $exec :: (store (s128), addrspace 1)
     S_ENDPGM 0
 ...
 ---
 name: test_mfma_f32_4x4x1f32_no_propagate_imm
 tracksRegLiveness: true
 
 body: |
   bb.0:
     liveins: $sgpr0_sgpr1
     ; GFX908-LABEL: name: test_mfma_f32_4x4x1f32_no_propagate_imm
     ; GFX908: liveins: $sgpr0_sgpr1
-    ; GFX908: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr0_sgpr1
-    ; GFX908: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[COPY]](p4), 36, 0 :: (dereferenceable invariant load (s64), align 4, addrspace 4)
-    ; GFX908: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
-    ; GFX908: undef %3.sub0:areg_128 = V_ACCVGPR_WRITE_B32_e64 1073741824, implicit $exec
-    ; GFX908: %3.sub1:areg_128 = COPY %3.sub0
-    ; GFX908: %3.sub2:areg_128 = COPY %3.sub0
-    ; GFX908: %3.sub3:areg_128 = COPY %3.sub0
-    ; GFX908: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 1073741824, implicit $exec
-    ; GFX908: [[V_MOV_B32_e32_2:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 1065353216, implicit $exec
-    ; GFX908: [[V_MFMA_F32_4X4X1F32_e64_:%[0-9]+]]:areg_128 = V_MFMA_F32_4X4X1F32_e64 [[V_MOV_B32_e32_2]], [[V_MOV_B32_e32_1]], %3, 0, 0, 0, implicit $mode, implicit $exec
-    ; GFX908: [[COPY1:%[0-9]+]]:vreg_128 = COPY [[V_MFMA_F32_4X4X1F32_e64_]]
-    ; GFX908: GLOBAL_STORE_DWORDX4_SADDR [[V_MOV_B32_e32_]], [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s128), addrspace 1)
-    ; GFX908: S_ENDPGM 0
+    ; GFX908-NEXT: {{  $}}
+    ; GFX908-NEXT: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr0_sgpr1
+    ; GFX908-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[COPY]](p4), 36, 0 :: (dereferenceable invariant load (s64), align 4, addrspace 4)
+    ; GFX908-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
+    ; GFX908-NEXT: undef %3.sub0:areg_128 = V_ACCVGPR_WRITE_B32_e64 1073741824, implicit $exec
+    ; GFX908-NEXT: %3.sub1:areg_128 = COPY %3.sub0
+    ; GFX908-NEXT: %3.sub2:areg_128 = COPY %3.sub0
+    ; GFX908-NEXT: %3.sub3:areg_128 = COPY %3.sub0
+    ; GFX908-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 1073741824, implicit $exec
+    ; GFX908-NEXT: [[V_MOV_B32_e32_2:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 1065353216, implicit $exec
+    ; GFX908-NEXT: [[V_MFMA_F32_4X4X1F32_e64_:%[0-9]+]]:areg_128 = V_MFMA_F32_4X4X1F32_e64 [[V_MOV_B32_e32_2]], [[V_MOV_B32_e32_1]], %3, 0, 0, 0, implicit $mode, implicit $exec
+    ; GFX908-NEXT: [[COPY1:%[0-9]+]]:vreg_128 = COPY [[V_MFMA_F32_4X4X1F32_e64_]]
+    ; GFX908-NEXT: GLOBAL_STORE_DWORDX4_SADDR [[V_MOV_B32_e32_]], [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s128), addrspace 1)
+    ; GFX908-NEXT: S_ENDPGM 0
     %1:sgpr_64(p4) = COPY $sgpr0_sgpr1
     %4:sreg_64_xexec = S_LOAD_DWORDX2_IMM %1:sgpr_64(p4), 36, 0 :: (dereferenceable invariant load (s64), align 4, addrspace 4)
     %5:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
     undef %11.sub0:areg_128 = V_ACCVGPR_WRITE_B32_e64 1073741824, implicit $exec
     %11.sub1:areg_128 = COPY %11.sub0:areg_128
     %11.sub2:areg_128 = COPY %11.sub0:areg_128
     %11.sub3:areg_128 = COPY %11.sub0:areg_128
     %8:vgpr_32 = V_MOV_B32_e32 1073741824, implicit $exec
     %9:vgpr_32 = V_MOV_B32_e32 1065353216, implicit $exec
     %10:areg_128 = V_MFMA_F32_4X4X1F32_e64 %9:vgpr_32, %8:vgpr_32, %11:areg_128, 0, 0, 0, implicit $mode, implicit $exec
     %12:vreg_128 = COPY %10:areg_128
     GLOBAL_STORE_DWORDX4_SADDR %5:vgpr_32, %12:vreg_128, %4:sreg_64_xexec, 0, 0, implicit $exec :: (store (s128), addrspace 1)
     S_ENDPGM 0
 ...
 ---
 name: test_vgpr_subreg_propagate
 tracksRegLiveness: true
 
 body: |
   bb.0:
     liveins: $vgpr0_vgpr1_vgpr2_vgpr3
     ; GFX908-LABEL: name: test_vgpr_subreg_propagate
     ; GFX908: liveins: $vgpr0_vgpr1_vgpr2_vgpr3
-    ; GFX908: [[COPY:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3, implicit $exec
-    ; GFX908: undef %1.sub0:areg_128 = V_ACCVGPR_WRITE_B32_e64 [[COPY]].sub0, implicit $exec
-    ; GFX908: %1.sub1:areg_128 = COPY [[COPY]].sub0
-    ; GFX908: %1.sub2:areg_128 = COPY [[COPY]].sub0
-    ; GFX908: %1.sub3:areg_128 = COPY [[COPY]].sub0
-    ; GFX908: S_ENDPGM 0, implicit [[COPY]], implicit %1
+    ; GFX908-NEXT: {{  $}}
+    ; GFX908-NEXT: [[COPY:%[0-9]+]]:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3, implicit $exec
+    ; GFX908-NEXT: undef %1.sub0:areg_128 = V_ACCVGPR_WRITE_B32_e64 [[COPY]].sub0, implicit $exec
+    ; GFX908-NEXT: %1.sub1:areg_128 = COPY [[COPY]].sub0
+    ; GFX908-NEXT: %1.sub2:areg_128 = COPY [[COPY]].sub0
+    ; GFX908-NEXT: %1.sub3:areg_128 = COPY [[COPY]].sub0
+    ; GFX908-NEXT: S_ENDPGM 0, implicit [[COPY]], implicit %1
     %0:vreg_128 = COPY $vgpr0_vgpr1_vgpr2_vgpr3, implicit $exec
     undef %1.sub0:areg_128 = V_ACCVGPR_WRITE_B32_e64 %0.sub0, implicit $exec
     %1.sub1:areg_128 = COPY %1.sub0:areg_128
     %1.sub2:areg_128 = COPY %1.sub0:areg_128
     %1.sub3:areg_128 = COPY %1.sub0:areg_128
     S_ENDPGM 0, implicit %0, implicit %1
 ...
 ---
 name: test_nonmatching_agpr_subreg_no_propagate
 tracksRegLiveness: true
 
 body: |
   bb.0:
     liveins: $vgpr0_vgpr1
     ; GFX908-LABEL: name: test_nonmatching_agpr_subreg_no_propagate
     ; GFX908: liveins: $vgpr0_vgpr1
-    ; GFX908: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1, implicit $exec
-    ; GFX908: undef %1.sub0:areg_64 = V_ACCVGPR_WRITE_B32_e64 [[COPY]].sub0, implicit $exec
-    ; GFX908: %1.sub1:areg_64 = V_ACCVGPR_WRITE_B32_e64 [[COPY]].sub1, implicit $exec
-    ; GFX908: [[COPY1:%[0-9]+]]:areg_64 = COPY %1
-    ; GFX908: S_ENDPGM 0, implicit [[COPY]], implicit %1, implicit [[COPY1]]
+    ; GFX908-NEXT: {{  $}}
+    ; GFX908-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1, implicit $exec
+    ; GFX908-NEXT: undef %1.sub0:areg_64 = V_ACCVGPR_WRITE_B32_e64 [[COPY]].sub0, implicit $exec
+    ; GFX908-NEXT: %1.sub1:areg_64 = V_ACCVGPR_WRITE_B32_e64 [[COPY]].sub1, implicit $exec
+    ; GFX908-NEXT: [[COPY1:%[0-9]+]]:areg_64 = COPY %1
+    ; GFX908-NEXT: S_ENDPGM 0, implicit [[COPY]], implicit %1, implicit [[COPY1]]
     %0:vreg_64 = COPY $vgpr0_vgpr1, implicit $exec
     undef %1.sub0:areg_64 = V_ACCVGPR_WRITE_B32_e64 %0.sub0, implicit $exec
     %1.sub1:areg_64 = V_ACCVGPR_WRITE_B32_e64 %0.sub1, implicit $exec
     %2:areg_64 = COPY %1:areg_64
     S_ENDPGM 0, implicit %0, implicit %1, implicit %2
 ...
 ---
 name: test_subreg_to_single_agpr_reg_propagate
 tracksRegLiveness: true
 
 body: |
   bb.0:
     liveins: $vgpr0_vgpr1
     ; GFX908-LABEL: name: test_subreg_to_single_agpr_reg_propagate
     ; GFX908: liveins: $vgpr0_vgpr1
-    ; GFX908: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1, implicit $exec
-    ; GFX908: undef %1.sub0:areg_64 = V_ACCVGPR_WRITE_B32_e64 [[COPY]].sub0, implicit $exec
-    ; GFX908: %1.sub1:areg_64 = V_ACCVGPR_WRITE_B32_e64 [[COPY]].sub1, implicit $exec
-    ; GFX908: [[COPY1:%[0-9]+]]:agpr_32 = COPY [[COPY]].sub1
-    ; GFX908: S_ENDPGM 0, implicit [[COPY]], implicit %1, implicit [[COPY1]]
+    ; GFX908-NEXT: {{  $}}
+    ; GFX908-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1, implicit $exec
+    ; GFX908-NEXT: undef %1.sub0:areg_64 = V_ACCVGPR_WRITE_B32_e64 [[COPY]].sub0, implicit $exec
+    ; GFX908-NEXT: %1.sub1:areg_64 = V_ACCVGPR_WRITE_B32_e64 [[COPY]].sub1, implicit $exec
+    ; GFX908-NEXT: [[COPY1:%[0-9]+]]:agpr_32 = COPY [[COPY]].sub1
+    ; GFX908-NEXT: S_ENDPGM 0, implicit [[COPY]], implicit %1, implicit [[COPY1]]
     %0:vreg_64 = COPY $vgpr0_vgpr1, implicit $exec
     undef %1.sub0:areg_64 = V_ACCVGPR_WRITE_B32_e64 %0.sub0, implicit $exec
     %1.sub1:areg_64 = V_ACCVGPR_WRITE_B32_e64 %0.sub1, implicit $exec
     %2:agpr_32 = COPY %1.sub1:areg_64
     S_ENDPGM 0, implicit %0, implicit %1, implicit %2
 ...

llvm/test/CodeGen/AMDGPU/combine-sreg64-inits.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 # RUN: llc -march=amdgcn -verify-machineinstrs -run-pass=liveintervals,amdgpu-pre-ra-optimizations %s -o - | FileCheck -check-prefix=GCN %s
 
 ---
-# GCN-LABEL: name: combine_sreg64_inits
-# GCN: %0:sgpr_64 = S_MOV_B64_IMM_PSEUDO 8589934593
-# GCN: S_NOP 0
 name:            combine_sreg64_inits
 tracksRegLiveness: true
 body:             |
   bb.0:
+    ; GCN-LABEL: name: combine_sreg64_inits
+    ; GCN: dead %0:sgpr_64 = S_MOV_B64_IMM_PSEUDO 8589934593
+    ; GCN-NEXT: S_NOP 0
     undef %0.sub0:sgpr_64 = S_MOV_B32 1
     S_NOP 0
     %0.sub1:sgpr_64 = S_MOV_B32 2
 ...
 ---
-# GCN-LABEL: name: combine_sreg64_inits_swap
-# GCN: %0:sgpr_64 = S_MOV_B64_IMM_PSEUDO 8589934593
-# GCN: S_NOP 0
 name:            combine_sreg64_inits_swap
 tracksRegLiveness: true
 body:             |
   bb.0:
+    ; GCN-LABEL: name: combine_sreg64_inits_swap
+    ; GCN: dead %0:sgpr_64 = S_MOV_B64_IMM_PSEUDO 8589934593
+    ; GCN-NEXT: S_NOP 0
     undef %0.sub1:sgpr_64 = S_MOV_B32 2
     S_NOP 0
     %0.sub0:sgpr_64 = S_MOV_B32 1
 ...
 ---
-# GCN-LABEL: name: sreg64_inits_different_blocks
-# GCN: undef %0.sub0:sgpr_64 = S_MOV_B32 1
-# GCN: %0.sub1:sgpr_64 = S_MOV_B32 2
+name: sreg64_subreg_copy_0
+tracksRegLiveness: true
+body: |
+  bb.0:
+    ; GCN-LABEL: name: sreg64_subreg_copy_0
+    ; GCN: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+    ; GCN-NEXT: undef %1.sub0:sgpr_64 = COPY [[DEF]]
+    ; GCN-NEXT: %1.sub0:sgpr_64 = S_MOV_B32 1
+    ; GCN-NEXT: dead %1.sub1:sgpr_64 = S_MOV_B32 2
+    %0:sgpr_32 = IMPLICIT_DEF
+    undef %1.sub0:sgpr_64 = COPY %0:sgpr_32
+    %1.sub0:sgpr_64 = S_MOV_B32 1
+    %1.sub1:sgpr_64 = S_MOV_B32 2
+...
+---
+name: sreg64_subreg_copy_1
+tracksRegLiveness: true
+body: |
+  bb.0:
+    ; GCN-LABEL: name: sreg64_subreg_copy_1
+    ; GCN: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+    ; GCN-NEXT: undef %1.sub0:sgpr_64 = S_MOV_B32 1
+    ; GCN-NEXT: %1.sub1:sgpr_64 = COPY [[DEF]]
+    ; GCN-NEXT: dead %1.sub1:sgpr_64 = S_MOV_B32 2
+    %0:sgpr_32 = IMPLICIT_DEF
+    undef %1.sub0:sgpr_64 = S_MOV_B32 1
+    %1.sub1:sgpr_64 = COPY %0:sgpr_32
+    %1.sub1:sgpr_64 = S_MOV_B32 2
+...
+---
+name: sreg64_subreg_copy_2
+tracksRegLiveness: true
+body: |
+  bb.0:
+    ; GCN-LABEL: name: sreg64_subreg_copy_2
+    ; GCN: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
+    ; GCN-NEXT: undef %1.sub0:sgpr_64 = S_MOV_B32 1
+    ; GCN-NEXT: %1.sub1:sgpr_64 = S_MOV_B32 2
+    ; GCN-NEXT: dead %1.sub0:sgpr_64 = COPY [[DEF]]
+    %0:sgpr_32 = IMPLICIT_DEF
+    undef %1.sub0:sgpr_64 = S_MOV_B32 1
+    %1.sub1:sgpr_64 = S_MOV_B32 2
+    %1.sub0:sgpr_64 = COPY %0:sgpr_32
+...
+---
 name:            sreg64_inits_different_blocks
 tracksRegLiveness: true
 body:             |
+  ; GCN-LABEL: name: sreg64_inits_different_blocks
+  ; GCN: bb.0:
+  ; GCN-NEXT:   successors: %bb.1(0x80000000)
+  ; GCN-NEXT: {{  $}}
+  ; GCN-NEXT:   undef %0.sub0:sgpr_64 = S_MOV_B32 1
+  ; GCN-NEXT: {{  $}}
+  ; GCN-NEXT: bb.1:
+  ; GCN-NEXT:   dead %0.sub1:sgpr_64 = S_MOV_B32 2
   bb.0:
     undef %0.sub0:sgpr_64 = S_MOV_B32 1
 
   bb.1:
     %0.sub1:sgpr_64 = S_MOV_B32 2
 ...
 ---
-# GCN-LABEL: name: sreg64_inits_two_defs_sub1
-# GCN: undef %0.sub0:sgpr_64 = S_MOV_B32 1
-# GCN: %0.sub1:sgpr_64 = S_MOV_B32 2
-# GCN: %0.sub1:sgpr_64 = S_MOV_B32 3
 name:            sreg64_inits_two_defs_sub1
 tracksRegLiveness: true
 body:             |
   bb.0:
+    ; GCN-LABEL: name: sreg64_inits_two_defs_sub1
+    ; GCN: undef %0.sub0:sgpr_64 = S_MOV_B32 1
+    ; GCN-NEXT: %0.sub1:sgpr_64 = S_MOV_B32 2
+    ; GCN-NEXT: dead %0.sub1:sgpr_64 = S_MOV_B32 3
     undef %0.sub0:sgpr_64 = S_MOV_B32 1
     %0.sub1:sgpr_64 = S_MOV_B32 2
     %0.sub1:sgpr_64 = S_MOV_B32 3
 ...
 ---
-# GCN-LABEL: name: sreg64_inits_two_defs_sub0
-# GCN: undef %0.sub0:sgpr_64 = S_MOV_B32 1
-# GCN: %0.sub1:sgpr_64 = S_MOV_B32 2
-# GCN: %0.sub0:sgpr_64 = S_MOV_B32 3
 name:            sreg64_inits_two_defs_sub0
 tracksRegLiveness: true
 body:             |
   bb.0:
+    ; GCN-LABEL: name: sreg64_inits_two_defs_sub0
+    ; GCN: undef %0.sub0:sgpr_64 = S_MOV_B32 1
+    ; GCN-NEXT: %0.sub1:sgpr_64 = S_MOV_B32 2
+    ; GCN-NEXT: dead %0.sub0:sgpr_64 = S_MOV_B32 3
     undef %0.sub0:sgpr_64 = S_MOV_B32 1
     %0.sub1:sgpr_64 = S_MOV_B32 2
     %0.sub0:sgpr_64 = S_MOV_B32 3
 ...
 ---
-# GCN-LABEL: name: sreg64_inits_full_def
-# GCN: undef %1.sub0:sgpr_64 = S_MOV_B32 1
-# GCN: %0:sgpr_64 = S_MOV_B64 3
 name:            sreg64_inits_full_def
 tracksRegLiveness: true
 body:             |
   bb.0:
+    ; GCN-LABEL: name: sreg64_inits_full_def
+    ; GCN: dead undef %1.sub0:sgpr_64 = S_MOV_B32 1
+    ; GCN-NEXT: dead %0:sgpr_64 = S_MOV_B64 3
     undef %0.sub0:sgpr_64 = S_MOV_B32 1
     %0:sgpr_64 = S_MOV_B64 3
 ...
 ---
-# GCN-LABEL: name: sreg64_inits_imp_use
-# GCN: %0.sub0:sgpr_64 = S_MOV_B32 1, implicit $m0
-# GCN: %0.sub1:sgpr_64 = S_MOV_B32 2
 name:            sreg64_inits_imp_use
 tracksRegLiveness: true
 body:             |
   bb.0:
+    ; GCN-LABEL: name: sreg64_inits_imp_use
+    ; GCN: undef %0.sub0:sgpr_64 = S_MOV_B32 1, implicit $m0
+    ; GCN-NEXT: dead %0.sub1:sgpr_64 = S_MOV_B32 2
     undef %0.sub0:sgpr_64 = S_MOV_B32 1, implicit $m0
     %0.sub1:sgpr_64 = S_MOV_B32 2
 ...
 ---
-# GCN-LABEL: name: sreg64_inits_imp_def
-# GCN: %0.sub0:sgpr_64 = S_MOV_B32 1, implicit-def $scc
-# GCN: %0.sub1:sgpr_64 = S_MOV_B32 2
 name:            sreg64_inits_imp_def
 tracksRegLiveness: true
 body:             |
   bb.0:
+    ; GCN-LABEL: name: sreg64_inits_imp_def
+    ; GCN: undef %0.sub0:sgpr_64 = S_MOV_B32 1, implicit-def $scc
+    ; GCN-NEXT: dead %0.sub1:sgpr_64 = S_MOV_B32 2
     undef %0.sub0:sgpr_64 = S_MOV_B32 1, implicit-def $scc
     %0.sub1:sgpr_64 = S_MOV_B32 2
 ...