Skip to content

[SLP] Create groups before sorting Phis #111174

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

Closed
wants to merge 3 commits into from
Closed

[SLP] Create groups before sorting Phis #111174

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
36ac0d8
[SLP] Create groups before sorting Phis
omern1 Oct 4, 2024
f969b73
Merge remote-tracking branch 'upstream/main' into slp-phix
omern1 Oct 4, 2024
4f9eb2e
Fix formatting
omern1 Oct 4, 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
154 changes: 125 additions & 29 deletions llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5448,40 +5448,136 @@ BoUpSLP::getReorderingData(const TreeEntry &TE, bool TopToBottom) {
if (!TE.ReorderIndices.empty())
return TE.ReorderIndices;

auto PHICompare = [&](unsigned I1, unsigned I2) {
Value *V1 = TE.Scalars[I1];
Value *V2 = TE.Scalars[I2];
if (V1 == V2 || (V1->getNumUses() == 0 && V2->getNumUses() == 0))
return false;
if (V1->getNumUses() < V2->getNumUses())
return true;
if (V1->getNumUses() > V2->getNumUses())
return false;
auto *FirstUserOfPhi1 = cast<Instruction>(*V1->user_begin());
auto *FirstUserOfPhi2 = cast<Instruction>(*V2->user_begin());
if (auto *IE1 = dyn_cast<InsertElementInst>(FirstUserOfPhi1))
if (auto *IE2 = dyn_cast<InsertElementInst>(FirstUserOfPhi2)) {
if (!areTwoInsertFromSameBuildVector(
IE1, IE2,
[](InsertElementInst *II) { return II->getOperand(0); }))
return I1 < I2;
return getElementIndex(IE1) < getElementIndex(IE2);
}
if (auto *EE1 = dyn_cast<ExtractElementInst>(FirstUserOfPhi1))
if (auto *EE2 = dyn_cast<ExtractElementInst>(FirstUserOfPhi2)) {
if (EE1->getOperand(0) != EE2->getOperand(0))
return I1 < I2;
return getElementIndex(EE1) < getElementIndex(EE2);
}
return I1 < I2;
};
SmallDenseMap<unsigned, unsigned, 16> PhiToId;
DenseMap<unsigned, unsigned> PhiToId;
SmallVector<unsigned> Phis(TE.Scalars.size());
std::iota(Phis.begin(), Phis.end(), 0);

BitVector Seen(Phis.size());
SmallVector<SmallVector<unsigned>> Groups;
Groups.resize(Phis.size(), {});

for (auto const Phidx : Phis) {
// We've already found a group for this Phidx
if (Seen.test(Phidx))
continue;
Comment on lines +5459 to +5462
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hmm, all this code is meaningless. Is not it just this:

for (unsigned Phidx : seq<unsigned>(TE.Scalars.size())) {

?
Phis filled using std::iota, so Seen is to needed at all.


Groups[Phidx].push_back(Phidx);

auto UserIterPhi1 = TE.Scalars[Phidx]->user_begin();
if (UserIterPhi1.atEnd())
continue;

auto *FirstUserOfPhi1 = cast<Instruction>(*UserIterPhi1);

unsigned Count = 0;
if (auto *IE1 = dyn_cast<InsertElementInst>(FirstUserOfPhi1)) {
auto Width = IE1->getType()->getElementCount();
assert(!Width.isScalable());
Count = Width.getFixedValue() - 1;
} else if (auto *EE1 = dyn_cast<ExtractElementInst>(FirstUserOfPhi1)) {
// Count is unused in the case of extract element instructions.
Count = -1;
}

for (auto const PhidxB : ArrayRef(Phis).drop_front(Phidx + 1)) {

Comment on lines +5482 to +5483
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Again, O(N^2). I don't like it, it increases compile time. Alternative approach is here

#110529

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, would have been nice to have a heads up on the discourse thread because I didn't realize you had an alternative approach that you were working on.

Anyway. Thanks for fixing the issue.

// At this point we know that we have found all the elements that fit in
// this group so we'll stop.
if (Count == 0)
break;

// B is already in a group so we don't want to look at it again.
if (Seen.test(PhidxB))
continue;

auto UserIterPhi2 = TE.Scalars[PhidxB]->user_begin();
if (UserIterPhi2.atEnd()) {
Seen.set(PhidxB);
continue;
}

auto *FirstUserOfPhi2 = cast<Instruction>(*UserIterPhi2);

if (auto *IE2 = dyn_cast<InsertElementInst>(FirstUserOfPhi2)) {
if (auto *IE1 = dyn_cast<InsertElementInst>(FirstUserOfPhi1)) {
if (areTwoInsertFromSameBuildVector(
IE1, IE2,
[](InsertElementInst *II) { return II->getOperand(0); })) {
Groups[Phidx].push_back(PhidxB);
Seen.set(PhidxB);
--Count;
continue;
}
}
} else if (auto *EE1 = dyn_cast<ExtractElementInst>(FirstUserOfPhi1)) {
if (auto *EE2 = dyn_cast<ExtractElementInst>(FirstUserOfPhi2)) {
if (EE1->getVectorOperand() == EE2->getVectorOperand()) {
Groups[Phidx].push_back(PhidxB);
Seen.set(PhidxB);
continue;
}
}
}
}

std::sort(
Groups[Phidx].begin(), Groups[Phidx].end(),
[TE](unsigned I1, unsigned I2) {
auto U1 = TE.Scalars[I1]->user_begin();
if (U1.atEnd())
return false;
auto U2 = TE.Scalars[I2]->user_begin();
if (U2.atEnd())
return true;

if (auto *FirstUserOfPhi1 = dyn_cast<InsertElementInst>(*(U1))) {
if (auto *FirstUserOfPhi2 = dyn_cast<InsertElementInst>(*(U2))) {
return getElementIndex(FirstUserOfPhi1) <
getElementIndex(FirstUserOfPhi2);
}
}

if (auto *FirstUserOfPhi1 = dyn_cast<ExtractElementInst>(*(U1))) {
if (auto *FirstUserOfPhi2 = dyn_cast<ExtractElementInst>(*(U2))) {
return FirstUserOfPhi1->getIndexOperand() <
FirstUserOfPhi2->getIndexOperand();
}
}

llvm_unreachable(
"Found something other than InsertElement or ExtractElement");
});
}

// Sort the groups on the basis of their ordering in the block.
std::sort(Groups.begin(), Groups.end(),
[TE](SmallVector<unsigned> I1, SmallVector<unsigned> I2) {
if (I1.empty())
return false;
if (I2.empty())
return true;

auto PhidxA = I1.front();
auto *InstA = TE.Scalars[PhidxA];
auto *Phi1 = cast<Instruction>(InstA);

auto PhidxB = I2.front();
auto *InstB = TE.Scalars[PhidxB];
auto *Phi2 = cast<Instruction>(InstB);

return Phi1->comesBefore(Phi2);
});

OrdersType ResOrder(TE.Scalars.size());
for (unsigned Id = 0, Sz = TE.Scalars.size(); Id < Sz; ++Id)
PhiToId[Id] = Id;
stable_sort(Phis, PHICompare);

Phis.clear();
for (auto const &group : Groups) {
for (auto const element : group)
Phis.push_back(element);
}

for (unsigned Id = 0, Sz = Phis.size(); Id < Sz; ++Id)
ResOrder[Id] = PhiToId[Phis[Id]];
if (isIdentityOrder(ResOrder))
Expand Down
21 changes: 11 additions & 10 deletions llvm/test/Transforms/SLPVectorizer/AArch64/landing_pad.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,7 +8,7 @@
; YAML-NEXT: Function: foo
; YAML-NEXT: Args:
; YAML-NEXT: - String: 'SLP vectorized with cost '
; YAML-NEXT: - Cost: '2'
; YAML-NEXT: - Cost: '0'
; YAML-NEXT: - String: ' and with tree size '
; YAML-NEXT: - TreeSize: '2'

Expand All @@ -28,7 +28,7 @@
; YAML-NEXT: Function: foo
; YAML-NEXT: Args:
; YAML-NEXT: - String: 'SLP vectorized with cost '
; YAML-NEXT: - Cost: '2'
; YAML-NEXT: - Cost: '3'
; YAML-NEXT: - String: ' and with tree size '
; YAML-NEXT: - TreeSize: '9'

Expand All @@ -45,36 +45,37 @@ define void @foo() personality ptr @bar {
; CHECK: bb3:
; CHECK-NEXT: [[TMP2:%.*]] = phi <2 x i64> [ [[TMP4:%.*]], [[BB6:%.*]] ], [ poison, [[BB1:%.*]] ]
; CHECK-NEXT: [[TMP3:%.*]] = invoke i64 poison(ptr addrspace(1) nonnull poison, i64 0, i64 0, i64 poison) [ "deopt"() ]
; CHECK-NEXT: to label [[BB4:%.*]] unwind label [[BB10:%.*]]
; CHECK-NEXT: to label [[BB4:%.*]] unwind label [[BB10:%.*]]
; CHECK: bb4:
; CHECK-NEXT: br i1 poison, label [[BB11:%.*]], label [[BB5:%.*]]
; CHECK: bb5:
; CHECK-NEXT: br label [[BB7:%.*]]
; CHECK: bb6:
; CHECK-NEXT: [[TMP4]] = phi <2 x i64> [ <i64 0, i64 poison>, [[BB8:%.*]] ]
; CHECK-NEXT: [[TMP4]] = phi <2 x i64> [ <i64 poison, i64 0>, [[BB8:%.*]] ]
; CHECK-NEXT: br label [[BB3]]
; CHECK: bb7:
; CHECK-NEXT: [[LOCAL_5_84111:%.*]] = phi i64 [ poison, [[BB8]] ], [ poison, [[BB5]] ]
; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x i64> poison, i64 [[LOCAL_5_84111]], i32 0
; CHECK-NEXT: [[TMP6:%.*]] = invoke i64 poison(ptr addrspace(1) nonnull poison, i64 poison, i64 poison, i64 poison) [ "deopt"() ]
; CHECK-NEXT: to label [[BB8]] unwind label [[BB12:%.*]]
; CHECK-NEXT: to label [[BB8]] unwind label [[BB12:%.*]]
; CHECK: bb8:
; CHECK-NEXT: br i1 poison, label [[BB7]], label [[BB6]]
; CHECK: bb9:
; CHECK-NEXT: [[INDVARS_IV528799:%.*]] = phi i64 [ poison, [[BB10]] ], [ poison, [[BB12]] ]
; CHECK-NEXT: [[TMP7]] = phi <2 x i64> [ [[TMP8:%.*]], [[BB10]] ], [ [[TMP9:%.*]], [[BB12]] ]
; CHECK-NEXT: [[TMP7]] = phi <2 x i64> [ [[TMP9:%.*]], [[BB10]] ], [ [[TMP10:%.*]], [[BB12]] ]
; CHECK-NEXT: br label [[BB2]]
; CHECK: bb10:
; CHECK-NEXT: [[TMP8]] = phi <2 x i64> [ [[TMP2]], [[BB3]] ]
; CHECK-NEXT: [[TMP8:%.*]] = phi <2 x i64> [ [[TMP2]], [[BB3]] ]
; CHECK-NEXT: [[LANDING_PAD68:%.*]] = landingpad { ptr, i64 }
; CHECK-NEXT: cleanup
; CHECK-NEXT: cleanup
; CHECK-NEXT: [[TMP9]] = shufflevector <2 x i64> [[TMP8]], <2 x i64> poison, <2 x i32> <i32 1, i32 0>
; CHECK-NEXT: br label [[BB9]]
; CHECK: bb11:
; CHECK-NEXT: ret void
; CHECK: bb12:
; CHECK-NEXT: [[TMP9]] = phi <2 x i64> [ [[TMP5]], [[BB7]] ]
; CHECK-NEXT: [[TMP10]] = phi <2 x i64> [ [[TMP5]], [[BB7]] ]
; CHECK-NEXT: [[LANDING_PAD149:%.*]] = landingpad { ptr, i64 }
; CHECK-NEXT: cleanup
; CHECK-NEXT: cleanup
; CHECK-NEXT: br label [[BB9]]
;
bb1:
Expand Down
54 changes: 26 additions & 28 deletions llvm/test/Transforms/SLPVectorizer/AArch64/reused-scalar-repeated-in-node.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,7 @@ define void @test() {
; CHECK-NEXT: br label %[[BB64]]
; CHECK: [[BB64]]:
; CHECK-NEXT: [[TMP0:%.*]] = phi <2 x float> [ poison, %[[BB61]] ], [ poison, %[[BB63]] ], [ poison, %[[BB62]] ]
; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <2 x float> [[TMP0]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
; CHECK-NEXT: [[I66:%.*]] = load float, ptr poison, align 16
; CHECK-NEXT: [[I67:%.*]] = load float, ptr poison, align 4
; CHECK-NEXT: [[I68:%.*]] = load float, ptr poison, align 8
Expand All @@ -25,48 +26,45 @@ define void @test() {
; CHECK-NEXT: [[I75:%.*]] = load float, ptr poison, align 16
; CHECK-NEXT: [[I76:%.*]] = load float, ptr poison, align 4
; CHECK-NEXT: [[TMP1:%.*]] = insertelement <16 x float> poison, float [[I76]], i32 0
; CHECK-NEXT: [[TMP2:%.*]] = insertelement <16 x float> [[TMP1]], float [[I75]], i32 1
; CHECK-NEXT: [[TMP3:%.*]] = insertelement <16 x float> [[TMP2]], float [[I74]], i32 2
; CHECK-NEXT: [[TMP4:%.*]] = insertelement <16 x float> [[TMP3]], float [[I73]], i32 3
; CHECK-NEXT: [[TMP5:%.*]] = insertelement <16 x float> [[TMP4]], float [[I71]], i32 4
; CHECK-NEXT: [[TMP6:%.*]] = insertelement <16 x float> [[TMP5]], float [[I70]], i32 5
; CHECK-NEXT: [[TMP7:%.*]] = insertelement <16 x float> [[TMP6]], float [[I68]], i32 6
; CHECK-NEXT: [[TMP8:%.*]] = insertelement <16 x float> [[TMP7]], float [[I66]], i32 7
; CHECK-NEXT: [[TMP9:%.*]] = insertelement <16 x float> [[TMP8]], float [[I72]], i32 13
; CHECK-NEXT: [[TMP10:%.*]] = insertelement <16 x float> [[TMP9]], float [[I69]], i32 14
; CHECK-NEXT: [[TMP11:%.*]] = insertelement <16 x float> [[TMP10]], float [[I67]], i32 15
; CHECK-NEXT: [[TMP3:%.*]] = insertelement <16 x float> [[TMP1]], float [[I75]], i32 3
; CHECK-NEXT: [[TMP4:%.*]] = insertelement <16 x float> [[TMP3]], float [[I74]], i32 4
; CHECK-NEXT: [[TMP5:%.*]] = insertelement <16 x float> [[TMP4]], float [[I73]], i32 5
; CHECK-NEXT: [[TMP6:%.*]] = insertelement <16 x float> [[TMP5]], float [[I72]], i32 6
; CHECK-NEXT: [[TMP7:%.*]] = insertelement <16 x float> [[TMP6]], float [[I71]], i32 7
; CHECK-NEXT: [[TMP8:%.*]] = insertelement <16 x float> [[TMP7]], float [[I70]], i32 8
; CHECK-NEXT: [[TMP9:%.*]] = insertelement <16 x float> [[TMP8]], float [[I69]], i32 10
; CHECK-NEXT: [[TMP10:%.*]] = insertelement <16 x float> [[TMP9]], float [[I68]], i32 13
; CHECK-NEXT: [[TMP11:%.*]] = insertelement <16 x float> [[TMP10]], float [[I67]], i32 14
; CHECK-NEXT: [[TMP12:%.*]] = insertelement <16 x float> [[TMP11]], float [[I66]], i32 15
; CHECK-NEXT: br i1 poison, label %[[BB167:.*]], label %[[BB77:.*]]
; CHECK: [[BB77]]:
; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <16 x float> [[TMP11]], <16 x float> poison, <8 x i32> <i32 poison, i32 5, i32 6, i32 7, i32 15, i32 15, i32 14, i32 15>
; CHECK-NEXT: [[TMP13:%.*]] = shufflevector <2 x float> [[TMP0]], <2 x float> poison, <16 x i32> <i32 poison, i32 poison, i32 1, i32 0, i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 1, i32 poison, i32 poison>
; CHECK-NEXT: [[TMP14:%.*]] = shufflevector <2 x float> [[TMP0]], <2 x float> poison, <16 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 0, i32 poison, i32 poison>
; CHECK-NEXT: [[TMP14:%.*]] = shufflevector <16 x float> [[TMP12]], <16 x float> poison, <8 x i32> <i32 poison, i32 15, i32 8, i32 13, i32 10, i32 14, i32 14, i32 14>
; CHECK-NEXT: [[TMP13:%.*]] = shufflevector <2 x float> [[TMP0]], <2 x float> poison, <16 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 1, i32 0, i32 1, i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; CHECK-NEXT: br label %[[BB78:.*]]
; CHECK: [[BB78]]:
; CHECK-NEXT: [[TMP15:%.*]] = phi <8 x float> [ [[TMP12]], %[[BB77]] ], [ [[TMP30:%.*]], %[[BB78]] ]
; CHECK-NEXT: [[TMP16:%.*]] = phi <2 x float> [ poison, %[[BB77]] ], [ [[TMP31:%.*]], %[[BB78]] ]
; CHECK-NEXT: [[TMP17:%.*]] = shufflevector <8 x float> [[TMP15]], <8 x float> poison, <16 x i32> <i32 0, i32 3, i32 1, i32 2, i32 3, i32 0, i32 2, i32 3, i32 2, i32 7, i32 2, i32 3, i32 0, i32 6, i32 7, i32 7>
; CHECK-NEXT: [[TMP15:%.*]] = phi <8 x float> [ [[TMP14]], %[[BB77]] ], [ [[TMP31:%.*]], %[[BB78]] ]
; CHECK-NEXT: [[TMP16:%.*]] = phi <2 x float> [ poison, %[[BB77]] ], [ [[TMP32:%.*]], %[[BB78]] ]
; CHECK-NEXT: [[TMP17:%.*]] = shufflevector <8 x float> [[TMP15]], <8 x float> poison, <16 x i32> <i32 0, i32 3, i32 5, i32 1, i32 2, i32 3, i32 4, i32 1, i32 0, i32 3, i32 5, i32 1, i32 0, i32 3, i32 5, i32 1>
; CHECK-NEXT: [[TMP18:%.*]] = fmul fast <16 x float> [[TMP17]], [[TMP13]]
; CHECK-NEXT: [[TMP19:%.*]] = shufflevector <8 x float> [[TMP15]], <8 x float> poison, <16 x i32> <i32 1, i32 poison, i32 0, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 1, i32 7, i32 6, i32 6>
; CHECK-NEXT: [[TMP19:%.*]] = shufflevector <8 x float> [[TMP15]], <8 x float> poison, <16 x i32> <i32 2, i32 poison, i32 4, i32 poison, i32 0, i32 poison, i32 5, i32 poison, i32 2, i32 poison, i32 4, i32 poison, i32 2, i32 poison, i32 4, i32 poison>
; CHECK-NEXT: [[TMP20:%.*]] = shufflevector <2 x float> [[TMP16]], <2 x float> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; CHECK-NEXT: [[TMP21:%.*]] = shufflevector <16 x float> [[TMP19]], <16 x float> [[TMP20]], <16 x i32> <i32 0, i32 17, i32 2, i32 16, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 12, i32 13, i32 14, i32 15>
; CHECK-NEXT: [[TMP22:%.*]] = shufflevector <8 x float> [[TMP15]], <8 x float> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; CHECK-NEXT: [[TMP23:%.*]] = shufflevector <16 x float> [[TMP21]], <16 x float> [[TMP22]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 17, i32 6, i32 7, i32 8, i32 22, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; CHECK-NEXT: [[TMP24:%.*]] = shufflevector <16 x float> [[TMP23]], <16 x float> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 1, i32 5, i32 3, i32 1, i32 3, i32 9, i32 3, i32 1, i32 12, i32 13, i32 14, i32 15>
; CHECK-NEXT: [[TMP25:%.*]] = call <16 x float> @llvm.vector.insert.v16f32.v2f32(<16 x float> [[TMP14]], <2 x float> [[TMP0]], i64 2)
; CHECK-NEXT: [[TMP21:%.*]] = shufflevector <16 x float> [[TMP19]], <16 x float> [[TMP20]], <16 x i32> <i32 0, i32 16, i32 2, i32 17, i32 4, i32 poison, i32 6, i32 poison, i32 8, i32 poison, i32 10, i32 poison, i32 12, i32 poison, i32 14, i32 poison>
; CHECK-NEXT: [[TMP24:%.*]] = shufflevector <16 x float> [[TMP21]], <16 x float> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 1, i32 6, i32 3, i32 8, i32 1, i32 10, i32 3, i32 12, i32 1, i32 14, i32 3>
; CHECK-NEXT: [[TMP25:%.*]] = call <16 x float> @llvm.vector.insert.v16f32.v4f32(<16 x float> poison, <4 x float> [[TMP2]], i64 4)
; CHECK-NEXT: [[TMP26:%.*]] = fmul fast <16 x float> [[TMP24]], [[TMP25]]
; CHECK-NEXT: [[TMP27:%.*]] = fadd fast <16 x float> [[TMP26]], [[TMP18]]
; CHECK-NEXT: [[TMP28:%.*]] = fadd fast <16 x float> [[TMP27]], poison
; CHECK-NEXT: [[TMP29:%.*]] = fadd fast <16 x float> [[TMP28]], poison
; CHECK-NEXT: [[TMP30]] = shufflevector <16 x float> [[TMP29]], <16 x float> poison, <8 x i32> <i32 12, i32 5, i32 6, i32 7, i32 15, i32 15, i32 14, i32 15>
; CHECK-NEXT: [[TMP31]] = shufflevector <16 x float> [[TMP29]], <16 x float> poison, <2 x i32> <i32 10, i32 11>
; CHECK-NEXT: [[TMP31]] = shufflevector <16 x float> [[TMP29]], <16 x float> poison, <8 x i32> <i32 12, i32 15, i32 8, i32 13, i32 10, i32 14, i32 14, i32 14>
; CHECK-NEXT: [[TMP32]] = shufflevector <16 x float> [[TMP29]], <16 x float> poison, <2 x i32> <i32 9, i32 11>
; CHECK-NEXT: br i1 poison, label %[[BB78]], label %[[BB167]]
; CHECK: [[BB167]]:
; CHECK-NEXT: [[TMP32:%.*]] = phi <16 x float> [ [[TMP11]], %[[BB64]] ], [ [[TMP29]], %[[BB78]] ]
; CHECK-NEXT: [[TMP33:%.*]] = extractelement <16 x float> [[TMP32]], i32 15
; CHECK-NEXT: [[TMP30:%.*]] = phi <16 x float> [ [[TMP12]], %[[BB64]] ], [ [[TMP29]], %[[BB78]] ]
; CHECK-NEXT: [[TMP33:%.*]] = extractelement <16 x float> [[TMP30]], i32 14
; CHECK-NEXT: store float [[TMP33]], ptr poison, align 1
; CHECK-NEXT: [[TMP34:%.*]] = extractelement <16 x float> [[TMP32]], i32 13
; CHECK-NEXT: [[TMP34:%.*]] = extractelement <16 x float> [[TMP30]], i32 6
; CHECK-NEXT: store float [[TMP34]], ptr poison, align 1
; CHECK-NEXT: [[TMP35:%.*]] = extractelement <16 x float> [[TMP32]], i32 14
; CHECK-NEXT: [[TMP35:%.*]] = extractelement <16 x float> [[TMP30]], i32 10
; CHECK-NEXT: br i1 poison, label %[[BB186:.*]], label %[[BB184:.*]]
; CHECK: [[BB184]]:
; CHECK-NEXT: br label %[[BB185:.*]]
Expand Down
4 changes: 2 additions & 2 deletions llvm/test/Transforms/SLPVectorizer/X86/buildvector-postpone-for-dependency.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,11 +13,11 @@ define void @test() {
; CHECK: [[BB6]]:
; CHECK-NEXT: [[TMP1:%.*]] = phi <2 x i32> [ zeroinitializer, %[[BB]] ], [ [[TMP8:%.*]], %[[BB6]] ]
; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> <i32 0, i32 0, i32 poison, i32 poison>, <4 x i32> [[TMP6]], <4 x i32> <i32 0, i32 1, i32 5, i32 4>
; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> <i32 0, i32 0, i32 poison, i32 poison>, <4 x i32> [[TMP6]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
; CHECK-NEXT: [[TMP3:%.*]] = ashr <4 x i32> zeroinitializer, [[TMP2]]
; CHECK-NEXT: [[TMP4:%.*]] = mul <4 x i32> zeroinitializer, [[TMP2]]
; CHECK-NEXT: [[TMP5]] = shufflevector <4 x i32> [[TMP3]], <4 x i32> [[TMP4]], <4 x i32> <i32 0, i32 5, i32 6, i32 7>
; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> <i32 0, i32 poison>, <2 x i32> <i32 2, i32 1>
; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> <i32 poison, i32 0>, <2 x i32> <i32 0, i32 3>
; CHECK-NEXT: [[TMP8]] = mul <2 x i32> zeroinitializer, [[TMP7]]
; CHECK-NEXT: br i1 false, label %[[BB2]], label %[[BB6]]
;
Expand Down
Loading
Loading