This is an archive of the discontinued LLVM Phabricator instance.

Differential D43049

AMDGPU: Don't crash when trying to fold implicit operands
ClosedPublic

Authored by arsenm on Feb 7 2018, 3:36 PM.

Details

Reviewers
rampitec
kzhuravl
cfang

Diff Detail

Event Timeline

arsenm created this revision.Feb 7 2018, 3:36 PM
Herald added subscribers: t-tye, tpr, dstuttard and 3 others. · View Herald TranscriptFeb 7 2018, 3:36 PM
cfang accepted this revision.Feb 7 2018, 4:03 PM

LGTM

This revision is now accepted and ready to land.Feb 7 2018, 4:03 PM
arsenm closed this revision.Feb 7 2018, 5:14 PM

r324550

Revision Contents

PathSize
lib/
Target/
AMDGPU/
1 line
test/
CodeGen/
AMDGPU/
14 lines

Diff 133326

lib/Target/AMDGPU/SIFoldOperands.cpp

Show First 20 Lines • Show All 339 Lines • ▼ Show 20 Lines if (FoldingImm && UseMI->isCopy()) {
UseMI->setDesc(TII->get(MovOp)); UseMI->setDesc(TII->get(MovOp));
CopiesToReplace.push_back(UseMI); CopiesToReplace.push_back(UseMI);
} else { } else {
const MCInstrDesc &UseDesc = UseMI->getDesc(); const MCInstrDesc &UseDesc = UseMI->getDesc();
// Don't fold into target independent nodes. Target independent opcodes // Don't fold into target independent nodes. Target independent opcodes
// don't have defined register classes. // don't have defined register classes.
if (UseDesc.isVariadic() || if (UseDesc.isVariadic() ||
UseOp.isImplicit() ||
UseDesc.OpInfo[UseOpIdx].RegClass == -1) UseDesc.OpInfo[UseOpIdx].RegClass == -1)
return; return;
} }
if (!FoldingImm) { if (!FoldingImm) {
tryAddToFoldList(FoldList, UseMI, UseOpIdx, &OpToFold, TII); tryAddToFoldList(FoldList, UseMI, UseOpIdx, &OpToFold, TII);
// FIXME: We could try to change the instruction from 64-bit to 32-bit // FIXME: We could try to change the instruction from 64-bit to 32-bit
▲ Show 20 Lines • Show All 631 Lines • Show Last 20 Lines

test/CodeGen/AMDGPU/fold-implicit-operand.mir

  • This file was added.
# RUN: llc -march=amdgcn -run-pass si-fold-operands -verify-machineinstrs -o - %s | FileCheck %s
---
# Make sure there is no crash when trying to fold an immediate into an
# implicit use
# CHECK: %0:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
# CHECK-NEXT: S_ENDPGM implicit %0
name: fold_imm_implicit_operand
body: |
bb.0:
%0:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
S_ENDPGM implicit %0
...