This is an archive of the discontinued LLVM Phabricator instance.

Differential D71866

[amdgpu] Fix scoreboard updating on `s_waitcnt_vscnt`.
ClosedPublic

Authored by hliao on Dec 24 2019, 9:49 PM.

Details

Reviewers
msearles
arsenm
rampitec
Commits
rG79d401905fcf: [amdgpu] Fix scoreboard updating on `s_waitcnt_vscnt`.
Summary
  • Other counters are accidentally cleared.

Diff Detail

Event Timeline

hliao created this revision.Dec 24 2019, 9:49 PM
Herald added a project: Restricted Project. · View Herald TranscriptDec 24 2019, 9:49 PM
Herald added subscribers: llvm-commits, hiraditya, t-tye and 8 others. · View Herald Transcript
Harbormaster completed remote builds in B42932: Diff 235255.Dec 24 2019, 9:49 PM
msearles added a subscriber: msearles.Dec 25 2019, 12:01 AM
hliao updated this revision to Diff 235686.Dec 31 2019, 12:05 AM

Add test case.

Herald added a subscriber: jfb. · View Herald TranscriptDec 31 2019, 12:05 AM
hliao retitled this revision from [amdgpu] Fix scoreboard updating after `s_waitcnt_vscnt`. to [amdgpu] Fix scoreboard updating on `s_waitcnt_vscnt`..Dec 31 2019, 12:06 AM
hliao added reviewers: msearles, arsenm, rampitec.
Harbormaster completed remote builds in B43085: Diff 235686.Dec 31 2019, 12:07 AM
hliao marked an inline comment as done.Dec 31 2019, 12:07 AM

Add clarification on the test case.

llvm/test/CodeGen/AMDGPU/waitcnt-vscnt.mir
6

This S_WAITCNT is missing without this patch.

rampitec accepted this revision.Dec 31 2019, 10:44 AM

LGTM

This revision is now accepted and ready to land.Dec 31 2019, 10:44 AM
Closed by commit rG79d401905fcf: [amdgpu] Fix scoreboard updating on `s_waitcnt_vscnt`. (authored by hliao). · Explain WhyDec 31 2019, 11:27 AM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
llvm/
lib/
Target/
AMDGPU/
2 lines
test/
CodeGen/
AMDGPU/
17 lines

Diff 235723

llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp

Show First 20 Lines • Show All 1,075 Lines • ▼ Show 20 Lines if (OldWaitcntInstr) {
Modified = true; Modified = true;
} else if (II->getOpcode() == AMDGPU::S_WAITCNT) { } else if (II->getOpcode() == AMDGPU::S_WAITCNT) {
int64_t Imm = II->getOperand(0).getImm(); int64_t Imm = II->getOperand(0).getImm();
ScoreBrackets.applyWaitcnt(AMDGPU::decodeWaitcnt(IV, Imm)); ScoreBrackets.applyWaitcnt(AMDGPU::decodeWaitcnt(IV, Imm));
} else { } else {
assert(II->getOpcode() == AMDGPU::S_WAITCNT_VSCNT); assert(II->getOpcode() == AMDGPU::S_WAITCNT_VSCNT);
assert(II->getOperand(0).getReg() == AMDGPU::SGPR_NULL); assert(II->getOperand(0).getReg() == AMDGPU::SGPR_NULL);
ScoreBrackets.applyWaitcnt( ScoreBrackets.applyWaitcnt(
AMDGPU::Waitcnt(0, 0, 0, II->getOperand(1).getImm())); AMDGPU::Waitcnt(~0u, ~0u, ~0u, II->getOperand(1).getImm()));
} }
} }
} }
return Modified; return Modified;
} }
if (ForceEmitZeroWaitcnts) if (ForceEmitZeroWaitcnts)
Wait = AMDGPU::Waitcnt::allZero(IV); Wait = AMDGPU::Waitcnt::allZero(IV);
▲ Show 20 Lines • Show All 514 Lines • Show Last 20 Lines

llvm/test/CodeGen/AMDGPU/waitcnt-vscnt.mir

  • This file was added.
# RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs -run-pass si-insert-waitcnts -o - %s | FileCheck -check-prefix=GFX10 %s
# GFX10-LABEL: waitcnt-vscnt
# GFX10: GLOBAL_ATOMIC_ADD_RTN
# GFX10-NEXT: S_WAITCNT 49279
---
hliaoAuthorUnsubmitted
Done
ReplyInline Actions

This S_WAITCNT is missing without this patch.

hliao: This S_WAITCNT is missing without this patch.
name: waitcnt-vscnt
machineFunctionInfo:
isEntryFunction: true
body: |
bb.0:
liveins: $sgpr0_sgpr1
$sgpr4 = S_LOAD_DWORD_IMM $sgpr0_sgpr1, 4, 0, 0 :: (dereferenceable invariant load 4 from `i32 addrspace(4)* undef`)
S_WAITCNT_VSCNT undef $sgpr_null, 0
$vgpr0 = GLOBAL_ATOMIC_ADD_RTN $vgpr0_vgpr1, $vgpr2, 0, 0, implicit $exec :: (load store syncscope("agent") seq_cst 4, addrspace 1)
S_CMP_LG_U32 killed $sgpr4, 0, implicit-def $scc
...