This is an archive of the discontinued LLVM Phabricator instance.

Differential D147157

[AMDGPU] Do not fix implicit vcc operand on INLINEASM
ClosedPublic

Authored by foad on Mar 29 2023, 7:22 AM.

Details

Reviewers
rampitec
Group Reviewers
Restricted Project
Commits
rG3e3594c771e6: [AMDGPU] Do not fix implicit vcc operand on INLINEASM
Summary

An INLINEASM can have an implicit def of vcc. It is not appropriate for
fixImplicitOperands to change this to vcc_lo on wave32.

Diff Detail

Event Timeline

foad created this revision.Mar 29 2023, 7:22 AM
Herald added a project: Restricted Project. · View Herald TranscriptMar 29 2023, 7:22 AM
Herald added subscribers: kosarev, StephenFan, kerbowa and 7 others. · View Herald Transcript
foad requested review of this revision.Mar 29 2023, 7:22 AM
Herald added a project: Restricted Project. · View Herald TranscriptMar 29 2023, 7:22 AM
Herald added subscribers: llvm-commits, wdng. · View Herald Transcript
foad added a child revision: D147158: [AMDGPU] Do not reserve 16-bit registers.Mar 29 2023, 7:23 AM
foad added inline comments.
llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
7880

Instead of blacklisting inlineasm I could whitelist VALU and SALU instructions. I'm not sure which is better.

foad mentioned this in D147158: [AMDGPU] Do not reserve 16-bit registers.Mar 29 2023, 7:27 AM
rampitec accepted this revision.Mar 29 2023, 11:09 AM
This revision is now accepted and ready to land.Mar 29 2023, 11:09 AM
Harbormaster completed remote builds in B222516: Diff 509357.Mar 29 2023, 11:21 AM
This revision was landed with ongoing or failed builds.Mar 29 2023, 12:33 PM
Closed by commit rG3e3594c771e6: [AMDGPU] Do not fix implicit vcc operand on INLINEASM (authored by foad). · Explain Why
This revision was automatically updated to reflect the committed changes.
foad added a commit: rG3e3594c771e6: [AMDGPU] Do not fix implicit vcc operand on INLINEASM.

Revision Contents

PathSize
llvm/
lib/
Target/
AMDGPU/
3 lines

Diff 509444

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 7,871 Lines • ▼ Show 20 Lines
} }
unsigned SIInstrInfo::getMaxMUBUFImmOffset() { return (1 << 12) - 1; } unsigned SIInstrInfo::getMaxMUBUFImmOffset() { return (1 << 12) - 1; }
void SIInstrInfo::fixImplicitOperands(MachineInstr &MI) const { void SIInstrInfo::fixImplicitOperands(MachineInstr &MI) const {
if (!ST.isWave32()) if (!ST.isWave32())
return; return;
if (MI.isInlineAsm())
foadAuthorUnsubmitted
Done
ReplyInline Actions

Instead of blacklisting inlineasm I could whitelist VALU and SALU instructions. I'm not sure which is better.

foad: Instead of blacklisting inlineasm I could whitelist VALU and SALU instructions. I'm not sure…
return;
for (auto &Op : MI.implicit_operands()) { for (auto &Op : MI.implicit_operands()) {
if (Op.isReg() && Op.getReg() == AMDGPU::VCC) if (Op.isReg() && Op.getReg() == AMDGPU::VCC)
Op.setReg(AMDGPU::VCC_LO); Op.setReg(AMDGPU::VCC_LO);
} }
} }
bool SIInstrInfo::isBufferSMRD(const MachineInstr &MI) const { bool SIInstrInfo::isBufferSMRD(const MachineInstr &MI) const {
if (!isSMRD(MI)) if (!isSMRD(MI))
▲ Show 20 Lines • Show All 916 Lines • Show Last 20 Lines