maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[AMDGPU][GlobalISel] Fix insert point in FoldableFneg combine 

Newly created fneg was built after some of it's uses in some cases.
Now it will be built immediately after instruction whose dst it negates.

Differential Revision: https://reviews.llvm.org/D119459
This commit is contained in:
Mirko Brkusanin 2022-02-11 11:44:50 +01:00
parent fd16eeea9d
commit 5ff35ba8ae
2 changed files with 30 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
llvm/lib/Target/AMDGPU/AMDGPUCombinerHelper.cpp
View File

@ -373,7 +373,8 @@ void AMDGPUCombinerHelper::applyFoldableFneg(MachineInstr &MI,
replaceRegWith(MRI, Dst, NegatedMatchInfo);
// Recreate non negated value for other uses of old MatchInfoDst
Builder.setInstrAndDebugLoc(MI);
auto NextInst = ++MatchInfo->getIterator();
Builder.setInstrAndDebugLoc(*NextInst);
Builder.buildFNeg(MatchInfoDst, NegatedMatchInfo, MI.getFlags());
}

28
llvm/test/CodeGen/AMDGPU/GlobalISel/combine-foldable-fneg.mir
View File

@ -777,3 +777,31 @@ body: |
$vgpr2 = COPY %7:_(s32)
...
# Check if new fneg is inserted at the appropriate place
---
name: fneg_src_has_multiple_uses
body: |
bb.0:
liveins: $vgpr0, $vgpr1, $vgpr2
; CHECK-LABEL: name: fneg_src_has_multiple_uses
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; CHECK-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; CHECK-NEXT: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[COPY1]]
; CHECK-NEXT: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[COPY]], [[FNEG]]
; CHECK-NEXT: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FMUL]]
; CHECK-NEXT: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[FNEG1]], [[COPY2]]
; CHECK-NEXT: $vgpr0 = COPY [[FMUL1]](s32)
; CHECK-NEXT: $vgpr1 = COPY [[FMUL]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s32) = COPY $vgpr1
%2:_(s32) = COPY $vgpr2
%3:_(s32) = G_FMUL %0:_, %1:_
%4:_(s32) = G_FMUL %3:_, %2:_
%5:_(s32) = G_FNEG %3:_
$vgpr0 = COPY %4:_(s32)
$vgpr1 = COPY %5:_(s32)
...