[llvm] f541a3a - [AMDGPU] SIInstrInfo: Fix resultDependsOnExec for VOPC instructions (#134629)

[via llvm-commits]

Author: Frederik Harwath
Date: 2025-04-22T10:17:35+02:00
New Revision: f541a3aad8e70b613ea411bf2d3e64a220f98e82

URL: https://github.com/llvm/llvm-project/commit/f541a3aad8e70b613ea411bf2d3e64a220f98e82
DIFF: https://github.com/llvm/llvm-project/commit/f541a3aad8e70b613ea411bf2d3e64a220f98e82.diff

LOG: [AMDGPU] SIInstrInfo: Fix resultDependsOnExec for VOPC instructions (#134629)

SIInstrInfo::resultDependsOnExec assumes that operand 0 of a comparison
is always the destination of the instruction. This is not true for
instructions in VOPC form where it is "src0". This led to a crash in
machine-cse.

---------

Co-authored-by: Matt Arsenault <arsenm2 at gmail.com>

Added: 
    llvm/test/CodeGen/AMDGPU/si-instr-info-vopc-exec.mir

Modified: 
    llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
    llvm/lib/Target/AMDGPU/SIInstrInfo.h

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
index ecfffa5067a39..e6d54860df221 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -147,14 +147,19 @@ bool SIInstrInfo::isReallyTriviallyReMaterializable(
 }
 
 // Returns true if the scalar result of a VALU instruction depends on exec.
-static bool resultDependsOnExec(const MachineInstr &MI) {
+bool SIInstrInfo::resultDependsOnExec(const MachineInstr &MI) const {
   // Ignore comparisons which are only used masked with exec.
   // This allows some hoisting/sinking of VALU comparisons.
   if (MI.isCompare()) {
-    const MachineRegisterInfo &MRI = MI.getParent()->getParent()->getRegInfo();
-    Register DstReg = MI.getOperand(0).getReg();
+    const MachineOperand *Dst = getNamedOperand(MI, AMDGPU::OpName::sdst);
+    if (!Dst)
+      return true;
+
+    Register DstReg = Dst->getReg();
     if (!DstReg.isVirtual())
       return true;
+
+    const MachineRegisterInfo &MRI = MI.getParent()->getParent()->getRegInfo();
     for (MachineInstr &Use : MRI.use_nodbg_instructions(DstReg)) {
       switch (Use.getOpcode()) {
       case AMDGPU::S_AND_SAVEEXEC_B32:

diff  --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.h b/llvm/lib/Target/AMDGPU/SIInstrInfo.h
index 9eb9444d0fe96..a3a54659d299a 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.h
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.h
@@ -183,6 +183,8 @@ class SIInstrInfo final : public AMDGPUGenInstrInfo {
   bool verifyCopy(const MachineInstr &MI, const MachineRegisterInfo &MRI,
                   StringRef &ErrInfo) const;
 
+  bool resultDependsOnExec(const MachineInstr &MI) const;
+
 protected:
   /// If the specific machine instruction is a instruction that moves/copies
   /// value from one register to another register return destination and source

diff  --git a/llvm/test/CodeGen/AMDGPU/si-instr-info-vopc-exec.mir b/llvm/test/CodeGen/AMDGPU/si-instr-info-vopc-exec.mir
new file mode 100644
index 0000000000000..52a11ee3cdedf
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/si-instr-info-vopc-exec.mir
@@ -0,0 +1,25 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+# RUN: llc -mtriple=amdgcn -mcpu=gfx1010 -run-pass=machine-cse %s -o - | FileCheck %s
+
+# Check only that V_CMP_EQ_U32_e32 does not lead to a crash.
+
+---
+name: depends_on_exec_check_can_handle_vopc
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    ; CHECK-LABEL: name: depends_on_exec_check_can_handle_vopc
+    ; CHECK: [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+    ; CHECK-NEXT: V_CMP_EQ_U32_e32 1, killed undef [[DEF]], implicit-def $vcc_lo, implicit $exec
+    ; CHECK-NEXT: [[V_CNDMASK_B32_e32_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e32 16256, undef [[DEF]], implicit $vcc_lo, implicit $exec
+    ; CHECK-NEXT: [[V_LSHLREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 16, undef [[V_CNDMASK_B32_e32_]], implicit $exec
+    ; CHECK-NEXT: SI_RETURN
+    %0:vgpr_32 = IMPLICIT_DEF
+    V_CMP_EQ_U32_e32 1, killed undef %0, implicit-def $vcc, implicit $exec
+    %1:vgpr_32 = V_CNDMASK_B32_e32 16256, undef %0, implicit $vcc, implicit $exec
+    V_CMP_EQ_U32_e32 1, killed undef %0, implicit-def $vcc, implicit $exec
+    %2:vgpr_32 = V_CNDMASK_B32_e32 16256, undef %0, implicit $vcc, implicit $exec
+    %3:vgpr_32 = V_LSHLREV_B32_e64 16, killed undef %2, implicit $exec
+    %4:vgpr_32 = V_LSHLREV_B32_e64 16, killed undef %1, implicit $exec
+    SI_RETURN
+...