Skip to content

AMDGPU: Fix assert when multi operands to update after folding imm #148205

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.

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

Merged
14 changes: 10 additions & 4 deletions llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -1761,6 +1761,7 @@ bool SIFoldOperandsImpl::foldInstOperand(MachineInstr &MI,
for (MachineInstr *Copy : CopiesToReplace)
Copy->addImplicitDefUseOperands(*MF);

SetVector<MachineInstr *> ConstantFoldCandidates;
for (FoldCandidate &Fold : FoldList) {
assert(!Fold.isReg() || Fold.Def.OpToFold);
if (Fold.isReg() && Fold.getReg().isVirtual()) {
Expand All @@ -1783,16 +1784,21 @@ bool SIFoldOperandsImpl::foldInstOperand(MachineInstr &MI,
<< static_cast<int>(Fold.UseOpNo) << " of "
<< *Fold.UseMI);

if (Fold.isImm() && tryConstantFoldOp(Fold.UseMI)) {
LLVM_DEBUG(dbgs() << "Constant folded " << *Fold.UseMI);
Changed = true;
}
if (Fold.isImm())
ConstantFoldCandidates.insert(Fold.UseMI);

} else if (Fold.Commuted) {
// Restoring instruction's original operand order if fold has failed.
TII->commuteInstruction(*Fold.UseMI, false);
}
}

for (MachineInstr *MI : ConstantFoldCandidates) {
if (tryConstantFoldOp(MI)) {
LLVM_DEBUG(dbgs() << "Constant folded " << *MI);
Changed = true;
}
}
return true;
}

Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1,15 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
# RUN: llc -mtriple=amdgcn-amd-hsa -mcpu=gfx1031 -run-pass=si-fold-operands -o - %s | FileCheck %s
---
name: snork
body: |
bb.0:
; CHECK-LABEL: name: snork
; CHECK: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_]], %subreg.sub1, [[S_MOV_B32_]], %subreg.sub2, [[S_MOV_B32_]], %subreg.sub3
; CHECK-NEXT: SI_RETURN
%0:sreg_32 = S_MOV_B32 0
%1:sgpr_128 = REG_SEQUENCE %0, %subreg.sub0, %0, %subreg.sub1, %0, %subreg.sub2, %0, %subreg.sub3
%2:sreg_32 = S_OR_B32 %1.sub0, %1.sub3, implicit-def dead $scc
SI_RETURN
...