Skip to content

[AMDGPU] Do not remat instructions with PhysReg uses #124366

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 1 commit into from
Jan 27, 2025
Merged

[AMDGPU] Do not remat instructions with PhysReg uses #124366

merged 1 commit into from
Jan 27, 2025

Conversation

jrbyrnes
Copy link
Contributor

This blocks rematerialization during scheduling if the instruction has a non accepted PhysReg use.

Currently, there aren't any checks like this in place, and we may create invalid code: https://godbolt.org/z/xjPjdcorf

@jrbyrnes
[AMDGPU] Do not allow remat instructions with PhysReg uses
267f469 
Change-Id: Ibb1f27bf559d641e26f5a4ae80d3561516c7a5e0
@jrbyrnes jrbyrnes requested review from arsenm and rampitec January 24, 2025 23:29
@llvmbot
Copy link
Member

llvmbot commented Jan 24, 2025

@llvm/pr-subscribers-backend-amdgpu

Author: Jeffrey Byrnes (jrbyrnes)

Changes

This blocks rematerialization during scheduling if the instruction has a non accepted PhysReg use.

Currently, there aren't any checks like this in place, and we may create invalid code: https://godbolt.org/z/xjPjdcorf

Patch is 39.31 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/124366.diff

2 Files Affected:

  • (modified) llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp (+10-2)
  • (modified) llvm/test/CodeGen/AMDGPU/machine-scheduler-sink-trivial-remats.mir (+408-48)
diff --git a/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp b/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp
index b00105ae9bd528..6e693066de10b6 100644
--- a/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp
+++ b/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp
@@ -1842,15 +1842,23 @@ bool PreRARematStage::sinkTriviallyRematInsts(const GCNSubtarget &ST,
   return true;
 }
 
-// Copied from MachineLICM
 bool PreRARematStage::isTriviallyReMaterializable(const MachineInstr &MI) {
   if (!DAG.TII->isTriviallyReMaterializable(MI))
     return false;
 
-  for (const MachineOperand &MO : MI.all_uses())
+  for (const MachineOperand &MO : MI.all_uses()) {
     if (MO.getReg().isVirtual())
       return false;
 
+    // We can't remat physreg uses, unless it is a constant or an ignorable
+    // use (e.g. implicit exec use on VALU instructions)
+    if (MO.getReg().isPhysical()) {
+      if (DAG.MRI.isConstantPhysReg(MO.getReg()) || DAG.TII->isIgnorableUse(MO))
+        continue;
+      return false;
+    }
+  }
+
   return true;
 }
 
diff --git a/llvm/test/CodeGen/AMDGPU/machine-scheduler-sink-trivial-remats.mir b/llvm/test/CodeGen/AMDGPU/machine-scheduler-sink-trivial-remats.mir
index 9f264de531950b..7662abc0aaf858 100644
--- a/llvm/test/CodeGen/AMDGPU/machine-scheduler-sink-trivial-remats.mir
+++ b/llvm/test/CodeGen/AMDGPU/machine-scheduler-sink-trivial-remats.mir
@@ -84,13 +84,11 @@ body:             |
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     S_NOP 0
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %0, implicit %1
     S_NOP 0, implicit %2, implicit %3
@@ -191,14 +189,12 @@ body:             |
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %24:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
     S_NOP 0, implicit %24
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %23
     S_NOP 0, implicit %0, implicit %1
@@ -300,7 +296,6 @@ body:             |
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %24:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
@@ -308,7 +303,6 @@ body:             |
     S_NOP 0, implicit %23
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %0, implicit %1
     S_NOP 0, implicit %2, implicit %3
@@ -408,7 +402,6 @@ body:             |
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %24:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
@@ -416,7 +409,6 @@ body:             |
     S_NOP 0, implicit %22, implicit %23
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %0, implicit %1
     S_NOP 0, implicit %2, implicit %3
@@ -529,7 +521,6 @@ body:             |
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %24:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
@@ -537,14 +528,12 @@ body:             |
     S_NOP 0, implicit %23
 
   bb.2:
-  ; predcessors: %bb.1
     successors: %bb.3
 
     %25:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 25, implicit $exec, implicit $mode
     S_NOP 0
 
   bb.3:
-  ; predecessors: %bb.2
     successors: %bb.4
 
     %26:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 26, implicit $exec, implicit $mode, implicit-def $m0
@@ -552,7 +541,6 @@ body:             |
     S_NOP 0, implicit %25
 
   bb.4:
-  ; predcessors: %bb.3
 
     S_NOP 0, implicit %0, implicit %1
     S_NOP 0, implicit %2, implicit %3
@@ -666,7 +654,6 @@ body:             |
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %24:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
@@ -674,7 +661,6 @@ body:             |
     S_NOP 0, implicit %23, implicit %22
 
   bb.2:
-  ; predcessors: %bb.1
     successors: %bb.3
 
     %25:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 25, implicit $exec, implicit $mode
@@ -682,7 +668,6 @@ body:             |
     S_NOP 0
 
   bb.3:
-  ; predecessors: %bb.2
     successors: %bb.4
 
     %27:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 27, implicit $exec, implicit $mode, implicit-def $m0
@@ -690,7 +675,6 @@ body:             |
     S_NOP 0, implicit %25, implicit %26
 
   bb.4:
-  ; predcessors: %bb.3
 
     S_NOP 0, implicit %0, implicit %1
     S_NOP 0, implicit %2, implicit %3
@@ -949,14 +933,12 @@ body:             |
     undef %23.sub0:vreg_64 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %23.sub1:vreg_64 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
     S_NOP 0, implicit %23
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %0, implicit %1
     S_NOP 0, implicit %2, implicit %3
@@ -1053,7 +1035,6 @@ body:             |
     undef %21.sub0:vreg_128 = nofpexcept V_CVT_I32_F64_e32 21, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %21.sub1:vreg_128 = nofpexcept V_CVT_I32_F64_e32 22, implicit $exec, implicit $mode, implicit-def $m0
@@ -1062,7 +1043,6 @@ body:             |
     S_NOP 0, implicit %21
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %0, implicit %1
     S_NOP 0, implicit %2, implicit %3
@@ -1581,7 +1561,6 @@ body:             |
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %24:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
@@ -1589,7 +1568,6 @@ body:             |
     S_NOP 0, implicit %24, implicit %25
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %23
     S_NOP 0, implicit %0, implicit %1
@@ -2528,14 +2506,12 @@ body:             |
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %24:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
     S_NOP 0, implicit %24
 
   bb.2:
-  ; predcessors: %bb.1
     successors: %bb.3
 
     S_NOP 0, implicit %23
@@ -2543,7 +2519,6 @@ body:             |
     S_NOP 0
 
   bb.3:
-  ; predecessors: %bb.2
     successors: %bb.4
 
     %26:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 26, implicit $exec, implicit $mode, implicit-def $m0
@@ -2551,7 +2526,6 @@ body:             |
     S_NOP 0, implicit %26, implicit %27
 
   bb.4:
-  ; predcessors: %bb.3
 
     S_NOP 0, implicit %25
     S_NOP 0, implicit %0, implicit %1
@@ -2650,7 +2624,6 @@ body:             |
     %21.sub1:vreg_128 = nofpexcept V_CVT_I32_F64_e32 22, implicit $exec, implicit $mode, implicit-def $m0
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %21.sub2:vreg_128 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode, implicit-def $m0
@@ -2658,7 +2631,6 @@ body:             |
     S_NOP 0, implicit %21
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %0, implicit %1
     S_NOP 0, implicit %2, implicit %3
@@ -2759,7 +2731,6 @@ body:             |
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %24:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
@@ -2767,7 +2738,6 @@ body:             |
     S_NOP 0, implicit %23
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %23
     S_NOP 0, implicit %0, implicit %1
@@ -5030,7 +5000,6 @@ body:             |
     %21.sub1:vreg_128 = nofpexcept V_CVT_I32_F64_e32 22, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %21.sub2:vreg_128 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode, implicit-def $m0
@@ -5038,7 +5007,6 @@ body:             |
     S_NOP 0, implicit %21
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %0, implicit %1
     S_NOP 0, implicit %2, implicit %3
@@ -5137,14 +5105,12 @@ body:             |
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %24:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
     S_NOP 0, implicit %23, implicit %24
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %0, implicit %1
     S_NOP 0, implicit %2, implicit %3
@@ -5242,7 +5208,6 @@ body:             |
     %22:vreg_64 = nofpexcept V_CVT_F64_I32_e32 22, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode, implicit-def $m0
@@ -5250,7 +5215,6 @@ body:             |
     S_NOP 0, implicit %22
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %0, implicit %1
     S_NOP 0, implicit %2, implicit %3
@@ -5348,7 +5312,6 @@ body:             |
     %22:vreg_64 = nofpexcept V_CVT_F64_I32_e32 22, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode, implicit-def $m0
@@ -5357,7 +5320,6 @@ body:             |
     S_NOP 0, implicit %22
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %0, implicit %1
     S_NOP 0, implicit %2, implicit %3
@@ -5456,7 +5418,6 @@ body:             |
     %22:vreg_64 = nofpexcept V_CVT_F64_I32_e32 22, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode, implicit-def $m0
@@ -5466,7 +5427,6 @@ body:             |
     S_NOP 0, implicit %22
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %0, implicit %1
     S_NOP 0, implicit %2, implicit %3
@@ -5562,14 +5522,12 @@ body:             |
     %22:vreg_64 = nofpexcept V_CVT_F64_I32_e32 22, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode, implicit-def $m0
     S_NOP 0, implicit %22, implicit %23
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %0, implicit %1
     S_NOP 0, implicit %2, implicit %3
@@ -5669,14 +5627,12 @@ body:             |
     undef %23.sub1:vreg_64_align2 = V_MOV_B32_e32 23, implicit $exec
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %24:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
     S_NOP 0, implicit %24
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %23.sub1
     S_NOP 0, implicit %0, implicit %1
@@ -5779,14 +5735,12 @@ body:             |
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %24:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
     S_NOP 0, implicit %24
 
   bb.2:
-  ; predcessors: %bb.1
 
     DBG_VALUE %23, 0, 0
     S_NOP 0, implicit %23
@@ -5889,14 +5843,12 @@ body:             |
     %23:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 23, implicit $exec, implicit $mode
 
   bb.1:
-  ; predecessors: %bb.0
     successors: %bb.2
 
     %24:vgpr_32 = nofpexcept V_CVT_I32_F64_e32 24, implicit $exec, implicit $mode, implicit-def $m0
     S_NOP 0, implicit %24
 
   bb.2:
-  ; predcessors: %bb.1
 
     S_NOP 0, implicit %23
     S_NOP 0, implicit %0, implicit %1
@@ -5914,3 +5866,411 @@ body:             |
     S_ENDPGM 0
 ...
 
+---
+name:            test_occ_8_physreg_use
+tracksRegLiveness: true
+machineFunctionInfo:
+  isEntryFunction: true
+body:             |
+  ; GFX908-LABEL: name: test_occ_8_physreg_use
+  ; GFX908: bb.0:
+  ; GFX908-NEXT:   successors: %bb.1(0x80000000)
+  ; GFX908-NEXT:   liveins: $vgpr0, $sgpr0_sgpr1
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT:   [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr0_sgpr1
+  ; GFX908-NEXT:   [[COPY1:%[0-9]+]]:vgpr_32(s32) = COPY $vgpr0
+  ; GFX908-NEXT:   [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[COPY]](p4), 52, 0 :: (dereferenceable invariant load (s64), align 4, addrspace 4)
+  ; GFX908-NEXT:   undef [[S_MOV_B32_:%[0-9]+]].sub1:sreg_64 = S_MOV_B32 0
+  ; GFX908-NEXT:   $vgpr8 = IMPLICIT_DEF
+  ; GFX908-NEXT:   $vgpr9 = IMPLICIT_DEF
+  ; GFX908-NEXT:   dead [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF1:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF2:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF3:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF4:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF5:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF6:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF7:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF8:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF9:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF10:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF11:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF12:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF13:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF14:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF15:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF16:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF17:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   dead [[V_CVT_I32_F32_e32_:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF17]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[DEF18:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF19:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   dead [[V_CMP_GT_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_GT_U32_e64 [[S_LOAD_DWORDX2_IMM]].sub0, [[COPY1]](s32), implicit $exec
+  ; GFX908-NEXT:   dead undef [[S_MOV_B32_:%[0-9]+]].sub0:sreg_64 = COPY [[S_LOAD_DWORDX2_IMM]].sub1
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_1:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 $vgpr8, implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_2:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 $vgpr9, implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_3:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF1]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_4:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF2]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_5:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF3]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_6:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF4]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_7:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF5]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_8:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF6]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_9:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF7]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_10:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF8]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_11:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF9]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_12:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF10]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_13:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF11]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_14:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF12]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_15:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF13]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_16:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF14]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_17:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF15]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_18:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF16]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_19:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF18]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_20:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF19]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[DEF20:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_21:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF20]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[DEF21:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_22:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF21]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[DEF22:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_23:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF22]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[DEF23:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_24:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF23]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[DEF24:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_25:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF24]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[DEF25:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_26:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF25]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[DEF26:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_27:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF26]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[DEF27:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_28:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF27]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[DEF28:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_29:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF28]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[DEF29:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[DEF30:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_30:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF29]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   [[V_CVT_I32_F32_e32_31:%[0-9]+]]:vgpr_32 = nofpexcept V_CVT_I32_F32_e32 [[DEF30]], implicit $exec, implicit $mode
+  ; GFX908-NEXT:   S_BRANCH %bb.1
+  ; GFX908-NEXT: {{  $}}
+  ; GFX908-NEXT: bb.1:
+  ; GFX908-NEXT:   S_NOP 0, implicit [[V_CVT_I32_F32_e32_1]], implicit [[V_CVT_I32_F32_e32_30]], implicit [[DEF29]]
+  ; GFX908-NEXT:   S_NOP 0, implicit [[V_CVT_I32_F32_e32_2]], implicit [[V_CVT_I32_F32_e32_31]], implicit [[DEF30]]
+  ; GFX908-NEXT:   S_NOP 0, implicit [[V_CVT_I32_F32_e32_3]], implicit [[V_CVT_I32_F32_e32_11]]
+  ; GFX908-NEXT:   S_NOP 0, implicit [[V_CVT_I32_F32_e32_4]], implicit [[V_CVT_I32_F32_e32_12]]
+  ; GFX908-NEXT:   S_NOP 0, implicit [[V_CVT_I32_F32_e32_5]], implicit [[V_CVT_I32_F32_e32_13]]
+  ; GFX908-NEXT:   S_NOP 0, implicit [[V_CVT_I32_F32_e32_6]], implicit [[V_CVT_I32_F32_e32_14]]
+  ; GFX908-NEXT:   S_NOP 0, implicit [[V_CVT_I32_F32_e32_7]], implicit [[V_CVT_I32_F32_e32_15]]
+  ; GFX908-NEXT:   S_NOP 0, implicit [[V_CVT_I32_F32_e32_8]], implicit [[V_CVT_I32_F32_e32_16]]
+  ; GFX908-NEXT:   S_NOP 0, implicit [[V_CVT_I32_F32_e32_9]], implicit [[V_CVT_I32_F32_e32_17]]
+  ; GFX908-NEXT:   S_NOP 0, implicit [[V_CVT_I32_F32_e32_10]], implicit [[V_CVT_I32_F32_e32_18]]
+  ; GFX908-NEXT:   S_NOP 0, implicit [[V_CVT_I32_F32_e32_19]], implicit [[V_CVT_I32_F32_e32_20]]
+  ; GFX908-NEXT:...
[truncated]

@jrbyrnes jrbyrnes merged commit e77d428 into llvm:main Jan 27, 2025
10 checks passed
@jrbyrnes jrbyrnes mentioned this pull request Jan 27, 2025
Merged
jayfoad
jayfoad reviewed Jan 28, 2025
View reviewed changes
llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp
if (MO.getReg().isVirtual())
return false;

// We can't remat physreg uses, unless it is a constant or an ignorable
// use (e.g. implicit exec use on VALU instructions)
if (MO.getReg().isPhysical()) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You already know it's not virtual because of the "if" above.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed by #124327

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@jayfoad jayfoad jayfoad left review comments

@rampitec rampitec rampitec approved these changes

@arsenm arsenm Awaiting requested review from arsenm

Assignees
No one assigned
Labels
backend:AMDGPU
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@jrbyrnes @llvmbot @jayfoad @rampitec