Skip to content

[AMDGCN] Use ZExt when handling indices in insertment element #85718

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.

Sign up for GitHub

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
merged 4 commits into from
Mar 20, 2024
Merged

[AMDGCN] Use ZExt when handling indices in insertment element #85718

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
fe2d2aa
[AMDGCN] Use ZExt when handling indices in insertment element instruc…
DataCorrupted Mar 18, 2024
282d796
[AMDGPU] move test to codegenprepare-break-large-phis.ll
DataCorrupted Mar 19, 2024
e3e34bf
change function naming
DataCorrupted Mar 19, 2024
9aa773f
[AMDGPU] change test name and update groundtruth
DataCorrupted Mar 19, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1749,7 +1749,7 @@ static bool isInterestingPHIIncomingValue(const Value *V) {
// Non constant index/out of bounds index -> folding is unlikely.
// The latter is more of a sanity check because canonical IR should just
// have replaced those with poison.
if (!Idx || Idx->getSExtValue() >= FVT->getNumElements())
if (!Idx || Idx->getZExtValue() >= FVT->getNumElements())
return false;

const auto *VecSrc = IE->getOperand(0);
Expand All @@ -1761,7 +1761,7 @@ static bool isInterestingPHIIncomingValue(const Value *V) {
return false;

CurVal = VecSrc;
EltsCovered.set(Idx->getSExtValue());
EltsCovered.set(Idx->getZExtValue());

// All elements covered.
if (EltsCovered.all())
Expand Down
51 changes: 51 additions & 0 deletions llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-break-large-phis.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1197,3 +1197,54 @@ reallyfinally:
store <5 x double> %val, ptr %out, align 1
ret void
}

define amdgpu_kernel void @pr85718(i1 %Bool, ptr %Ptr, <4 x float> %Vec1, <4 x float> %Vec2) {
; OPT-LABEL: @pr85718(
; OPT-NEXT: BB0:
; OPT-NEXT: [[I:%.*]] = insertelement <4 x float> [[VEC1:%.*]], float 4.200000e+01, i1 true
; OPT-NEXT: br label [[BB1:%.*]]
; OPT: BB1:
; OPT-NEXT: [[TMP0:%.*]] = phi float [ [[LARGEPHI_EXTRACTSLICE0:%.*]], [[BB2:%.*]] ], [ [[LARGEPHI_EXTRACTSLICE1:%.*]], [[BB1]] ], [ 0.000000e+00, [[BB0:%.*]] ]
; OPT-NEXT: [[TMP1:%.*]] = phi float [ [[LARGEPHI_EXTRACTSLICE3:%.*]], [[BB2]] ], [ [[LARGEPHI_EXTRACTSLICE4:%.*]], [[BB1]] ], [ 0.000000e+00, [[BB0]] ]
; OPT-NEXT: [[TMP2:%.*]] = phi float [ [[LARGEPHI_EXTRACTSLICE6:%.*]], [[BB2]] ], [ [[LARGEPHI_EXTRACTSLICE7:%.*]], [[BB1]] ], [ 0.000000e+00, [[BB0]] ]
; OPT-NEXT: [[TMP3:%.*]] = phi float [ [[LARGEPHI_EXTRACTSLICE9:%.*]], [[BB2]] ], [ [[LARGEPHI_EXTRACTSLICE10:%.*]], [[BB1]] ], [ 0.000000e+00, [[BB0]] ]
; OPT-NEXT: [[LARGEPHI_INSERTSLICE0:%.*]] = insertelement <4 x float> poison, float [[TMP0]], i64 0
; OPT-NEXT: [[LARGEPHI_INSERTSLICE1:%.*]] = insertelement <4 x float> [[LARGEPHI_INSERTSLICE0]], float [[TMP1]], i64 1
; OPT-NEXT: [[LARGEPHI_INSERTSLICE2:%.*]] = insertelement <4 x float> [[LARGEPHI_INSERTSLICE1]], float [[TMP2]], i64 2
; OPT-NEXT: [[LARGEPHI_INSERTSLICE3:%.*]] = insertelement <4 x float> [[LARGEPHI_INSERTSLICE2]], float [[TMP3]], i64 3
; OPT-NEXT: store <4 x float> [[LARGEPHI_INSERTSLICE3]], ptr [[PTR:%.*]], align 128
; OPT-NEXT: [[LARGEPHI_EXTRACTSLICE1]] = extractelement <4 x float> [[VEC2:%.*]], i64 0
; OPT-NEXT: [[LARGEPHI_EXTRACTSLICE4]] = extractelement <4 x float> [[VEC2]], i64 1
; OPT-NEXT: [[LARGEPHI_EXTRACTSLICE7]] = extractelement <4 x float> [[VEC2]], i64 2
; OPT-NEXT: [[LARGEPHI_EXTRACTSLICE10]] = extractelement <4 x float> [[VEC2]], i64 3
; OPT-NEXT: br i1 [[BOOL:%.*]], label [[BB1]], label [[BB2]]
; OPT: BB2:
; OPT-NEXT: [[LARGEPHI_EXTRACTSLICE0]] = extractelement <4 x float> [[I]], i64 0
; OPT-NEXT: [[LARGEPHI_EXTRACTSLICE3]] = extractelement <4 x float> [[I]], i64 1
; OPT-NEXT: [[LARGEPHI_EXTRACTSLICE6]] = extractelement <4 x float> [[I]], i64 2
; OPT-NEXT: [[LARGEPHI_EXTRACTSLICE9]] = extractelement <4 x float> [[I]], i64 3
; OPT-NEXT: br label [[BB1]]
;
; NOOPT-LABEL: @pr85718(
; NOOPT-NEXT: BB0:
; NOOPT-NEXT: [[I:%.*]] = insertelement <4 x float> [[VEC1:%.*]], float 4.200000e+01, i1 true
; NOOPT-NEXT: br label [[BB1:%.*]]
; NOOPT: BB1:
; NOOPT-NEXT: [[PHI:%.*]] = phi <4 x float> [ [[I]], [[BB2:%.*]] ], [ [[VEC2:%.*]], [[BB1]] ], [ zeroinitializer, [[BB0:%.*]] ]
; NOOPT-NEXT: store <4 x float> [[PHI]], ptr [[PTR:%.*]], align 128
; NOOPT-NEXT: br i1 [[BOOL:%.*]], label [[BB1]], label [[BB2]]
; NOOPT: BB2:
; NOOPT-NEXT: br label [[BB1]]
;
BB0:
%I = insertelement <4 x float> %Vec1, float 4.200000e+01, i1 true
br label %BB1

BB1: ; preds = %BB0, %BB1, %BB2
%PHI = phi <4 x float> [ %I, %BB2 ], [ %Vec2, %BB1 ], [ zeroinitializer, %BB0 ]
store <4 x float> %PHI, ptr %Ptr, align 128
br i1 %Bool, label %BB1, label %BB2

BB2: ; preds = %BB1
br label %BB1
}