[VPlan] Merge fcmp uno feeding AnyOf.
#166823
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Member
|
@llvm/pr-subscribers-vectorizers @llvm/pr-subscribers-llvm-transforms Author: Florian Hahn (fhahn) ChangesFold This pattern is generated to check if any vector lane is NaN, and combining multiple compares is beneficial on architectures that have dedicated instructions. Alive2 Proof: https://alive2.llvm.org/ce/z/vA_aoM Combine suggested as part of #161735 Full diff: https://github.com/llvm/llvm-project/pull/166823.diff 5 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h b/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h
index b57c44872c1b6..8b2931637113f 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h
@@ -417,6 +417,10 @@ m_BranchOnCount(const Op0_t &Op0, const Op1_t &Op1) {
return m_VPInstruction<VPInstruction::BranchOnCount>(Op0, Op1);
}
+inline VPInstruction_match<VPInstruction::AnyOf> m_AnyOf() {
+ return m_VPInstruction<VPInstruction::AnyOf>();
+}
+
template <typename Op0_t>
inline VPInstruction_match<VPInstruction::AnyOf, Op0_t>
m_AnyOf(const Op0_t &Op0) {
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 48bd697397f41..937c420dcd6ee 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -1221,6 +1221,29 @@ static void simplifyRecipe(VPSingleDefRecipe *Def, VPTypeAnalysis &TypeInfo) {
}
}
+ // Fold any-of (fcmp uno %A, %A), (fcmp uno %B, %B), ... ->
+ // any-of (fcmp uno %A, %B), ...
+ if (match(Def, m_AnyOf()) && Def->getNumOperands() % 2 == 0) {
+ SmallVector<VPValue *, 4> NewOps;
+ unsigned NumOps = Def->getNumOperands();
+ for (unsigned I = 0; I < NumOps; I += 2) {
+ VPValue *A, *B;
+ if (!match(
+ Def->getOperand(I),
+ m_SpecificCmp(CmpInst::FCMP_UNO, m_VPValue(A), m_Deferred(A))) ||
+ !match(Def->getOperand(I + 1),
+ m_SpecificCmp(CmpInst::FCMP_UNO, m_VPValue(B), m_Deferred(B))))
+ break;
+
+ NewOps.push_back(Builder.createFCmp(CmpInst::FCMP_UNO, A, B));
+ }
+
+ if (NewOps.size() == NumOps / 2) {
+ VPValue *NewAnyOf = Builder.createNaryOp(VPInstruction::AnyOf, NewOps);
+ return Def->replaceAllUsesWith(NewAnyOf);
+ }
+ }
+
// Remove redundant DerviedIVs, that is 0 + A * 1 -> A and 0 + 0 * x -> 0.
if ((match(Def, m_DerivedIV(m_ZeroInt(), m_VPValue(A), m_One())) ||
match(Def, m_DerivedIV(m_ZeroInt(), m_ZeroInt(), m_VPValue()))) &&
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/fmax-without-fast-math-flags.ll b/llvm/test/Transforms/LoopVectorize/AArch64/fmax-without-fast-math-flags.ll
index 3c83c01929aae..f4e5085c604f6 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/fmax-without-fast-math-flags.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/fmax-without-fast-math-flags.ll
@@ -60,12 +60,9 @@ define float @fmaxnum(ptr %src, i64 %n) {
; CHECK-NEXT: [[TMP8]] = call <4 x float> @llvm.maxnum.v4f32(<4 x float> [[VEC_PHI1]], <4 x float> [[WIDE_LOAD2]])
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[IV]], 8
; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: [[TMP3:%.*]] = fcmp uno <4 x float> [[WIDE_LOAD]], [[WIDE_LOAD]]
-; CHECK-NEXT: [[TMP4:%.*]] = fcmp uno <4 x float> [[WIDE_LOAD2]], [[WIDE_LOAD2]]
-; CHECK-NEXT: [[TMP18:%.*]] = freeze <4 x i1> [[TMP3]]
-; CHECK-NEXT: [[TMP15:%.*]] = freeze <4 x i1> [[TMP4]]
-; CHECK-NEXT: [[TMP5:%.*]] = or <4 x i1> [[TMP18]], [[TMP15]]
-; CHECK-NEXT: [[TMP6:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP5]])
+; CHECK-NEXT: [[TMP18:%.*]] = fcmp uno <4 x float> [[WIDE_LOAD]], [[WIDE_LOAD2]]
+; CHECK-NEXT: [[TMP19:%.*]] = freeze <4 x i1> [[TMP18]]
+; CHECK-NEXT: [[TMP6:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP19]])
; CHECK-NEXT: [[TMP10:%.*]] = or i1 [[TMP6]], [[TMP9]]
; CHECK-NEXT: br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/fmin-without-fast-math-flags.ll b/llvm/test/Transforms/LoopVectorize/AArch64/fmin-without-fast-math-flags.ll
index 711a9cd03ac15..89b5d1395a4a0 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/fmin-without-fast-math-flags.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/fmin-without-fast-math-flags.ll
@@ -60,12 +60,9 @@ define float @fminnum(ptr %src, i64 %n) {
; CHECK-NEXT: [[TMP8]] = call <4 x float> @llvm.minnum.v4f32(<4 x float> [[VEC_PHI1]], <4 x float> [[WIDE_LOAD2]])
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[IV]], 8
; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: [[TMP3:%.*]] = fcmp uno <4 x float> [[WIDE_LOAD]], [[WIDE_LOAD]]
-; CHECK-NEXT: [[TMP4:%.*]] = fcmp uno <4 x float> [[WIDE_LOAD2]], [[WIDE_LOAD2]]
-; CHECK-NEXT: [[TMP15:%.*]] = freeze <4 x i1> [[TMP3]]
-; CHECK-NEXT: [[TMP18:%.*]] = freeze <4 x i1> [[TMP4]]
-; CHECK-NEXT: [[TMP5:%.*]] = or <4 x i1> [[TMP15]], [[TMP18]]
-; CHECK-NEXT: [[TMP6:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP5]])
+; CHECK-NEXT: [[TMP15:%.*]] = fcmp uno <4 x float> [[WIDE_LOAD]], [[WIDE_LOAD2]]
+; CHECK-NEXT: [[TMP19:%.*]] = freeze <4 x i1> [[TMP15]]
+; CHECK-NEXT: [[TMP6:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP19]])
; CHECK-NEXT: [[TMP10:%.*]] = or i1 [[TMP6]], [[TMP9]]
; CHECK-NEXT: br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
diff --git a/llvm/test/Transforms/LoopVectorize/fmax-without-fast-math-flags-interleave.ll b/llvm/test/Transforms/LoopVectorize/fmax-without-fast-math-flags-interleave.ll
index af648df9fc5c7..eab3492ebe2fa 100644
--- a/llvm/test/Transforms/LoopVectorize/fmax-without-fast-math-flags-interleave.ll
+++ b/llvm/test/Transforms/LoopVectorize/fmax-without-fast-math-flags-interleave.ll
@@ -60,12 +60,9 @@ define float @fmaxnum(ptr %src, i64 %n) {
; CHECK-NEXT: [[TMP8]] = call <4 x float> @llvm.maxnum.v4f32(<4 x float> [[VEC_PHI1]], <4 x float> [[WIDE_LOAD2]])
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[IV]], 8
; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: [[TMP3:%.*]] = fcmp uno <4 x float> [[WIDE_LOAD]], [[WIDE_LOAD]]
-; CHECK-NEXT: [[TMP4:%.*]] = fcmp uno <4 x float> [[WIDE_LOAD2]], [[WIDE_LOAD2]]
-; CHECK-NEXT: [[TMP15:%.*]] = freeze <4 x i1> [[TMP3]]
-; CHECK-NEXT: [[TMP18:%.*]] = freeze <4 x i1> [[TMP4]]
-; CHECK-NEXT: [[TMP5:%.*]] = or <4 x i1> [[TMP15]], [[TMP18]]
-; CHECK-NEXT: [[TMP6:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP5]])
+; CHECK-NEXT: [[TMP15:%.*]] = fcmp uno <4 x float> [[WIDE_LOAD]], [[WIDE_LOAD2]]
+; CHECK-NEXT: [[TMP19:%.*]] = freeze <4 x i1> [[TMP15]]
+; CHECK-NEXT: [[TMP6:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP19]])
; CHECK-NEXT: [[TMP10:%.*]] = or i1 [[TMP6]], [[TMP9]]
; CHECK-NEXT: br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
|
Fold any-of (fcmp uno %A, %A), (fcmp uno %B, %B), ... -> any-of (fcmp uno %A, %B), ... This pattern is generated to check if any vector lane is NaN, and combining multiple compares is beneficial on architectures that have dedicated instructions. Alive2 Proof: https://alive2.llvm.org/ce/z/vA_aoM
fhahn
force-pushed
the
vplan-merge-fcmp-uno
branch
from
c707451 to
c83adad
Compare
Mel-Chen
reviewed
Nov 14, 2025
Comment on lines
1241
to
1244
| if (NewOps.size() == NumOps / 2) { | ||
| VPValue *NewAnyOf = Builder.createNaryOp(VPInstruction::AnyOf, NewOps); | ||
| return Def->replaceAllUsesWith(NewAnyOf); | ||
| } |
Contributor
There was a problem hiding this comment.
Should we clean up the recipes in NewOps if NewOps.size() != NumOps / 2?
Contributor
Author
There was a problem hiding this comment.
The latest update should always use the newly created recipes, thanks
ayalz
reviewed
Nov 15, 2025
Comment on lines
+1224
to
+1225
| // Fold any-of (fcmp uno %A, %A), (fcmp uno %B, %B), ... -> | ||
| // any-of (fcmp uno %A, %B), ... |
Collaborator
There was a problem hiding this comment.
Perhaps better (than all-adjacent-pairs-or-nothing is) to search for a single pair of any two operands one doing (fcmp uno %A, %A) and the other (fcmp uno %B, %B) and replace them by one operand doing (fcmp uno %A, %B), to be repeated while found?
Tests below seem to cover the numOperands == 2 case only?
Contributor
Author
There was a problem hiding this comment.
Updated to keep track of unpaired compares, and merge the next matching pair. Should be tested now with tests with IC = 3 and IC =5, thanks
fhahn
added a commit
that referenced
this pull request
Nov 15, 2025
This adds additional test coverage for folding FCMP uno (#166823)
llvm-sync bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Nov 15, 2025
…r fmaxnum. This adds additional test coverage for folding FCMP uno (llvm/llvm-project#166823)
ayalz
approved these changes
Nov 16, 2025
Comment on lines
1234
to
1241
| continue; | ||
| } | ||
| if (UnpairedCmp) { | ||
| NewOps.push_back(Builder.createFCmp(CmpInst::FCMP_UNO, | ||
| UnpairedCmp->getOperand(0), X)); | ||
| UnpairedCmp = nullptr; | ||
| } else { | ||
| UnpairedCmp = Op->getDefiningRecipe(); |
Collaborator
There was a problem hiding this comment.
nit: clearer for first alternative to be w/o early continue and second alternative to be the simpler !UnpairedCmp case which at runtime precedes the (third) UnpairedCmp case?
Suggested change
| continue; | |
| } | |
| if (UnpairedCmp) { | |
| NewOps.push_back(Builder.createFCmp(CmpInst::FCMP_UNO, | |
| UnpairedCmp->getOperand(0), X)); | |
| UnpairedCmp = nullptr; | |
| } else { | |
| UnpairedCmp = Op->getDefiningRecipe(); | |
| } else if (!UnpairedCmp) { | |
| UnpairedCmp = Op->getDefiningRecipe(); | |
| } else { | |
| NewOps.push_back(Builder.createFCmp(CmpInst::FCMP_UNO, | |
| UnpairedCmp->getOperand(0), X)); | |
| UnpairedCmp = nullptr; | |
| } |
🐧 Linux x64 Test Results
|
llvm-sync bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Nov 23, 2025
Fold any-of (fcmp uno %A, %A), (fcmp uno %B, %B), ... -> any-of (fcmp uno %A, %B), ... This pattern is generated to check if any vector lane is NaN, and combining multiple compares is beneficial on architectures that have dedicated instructions. Alive2 Proof: https://alive2.llvm.org/ce/z/vA_aoM Combine suggested as part of llvm/llvm-project#161735 PR: llvm/llvm-project#166823
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/42/builds/7095 Here is the relevant piece of the build log for the reference |
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/110/builds/6461 Here is the relevant piece of the build log for the reference |
nekoshirro
pushed a commit
to nekoshirro/Alchemist-LLVM
that referenced
this pull request
Nov 24, 2025
This adds additional test coverage for folding FCMP uno (llvm/llvm-project#166823) Signed-off-by: Hafidz Muzakky <[email protected]>
nekoshirro
pushed a commit
to nekoshirro/Alchemist-LLVM
that referenced
this pull request
Nov 24, 2025
Fold any-of (fcmp uno %A, %A), (fcmp uno %B, %B), ... -> any-of (fcmp uno %A, %B), ... This pattern is generated to check if any vector lane is NaN, and combining multiple compares is beneficial on architectures that have dedicated instructions. Alive2 Proof: https://alive2.llvm.org/ce/z/vA_aoM Combine suggested as part of llvm/llvm-project#161735 PR: llvm/llvm-project#166823 Signed-off-by: Hafidz Muzakky <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fold
any-of (fcmp uno %A, %A), (fcmp uno %B, %B), ... ->
any-of (fcmp uno %A, %B), ...
This pattern is generated to check if any vector lane is NaN, and combining multiple compares is beneficial on architectures that have dedicated instructions.
Alive2 Proof: https://alive2.llvm.org/ce/z/vA_aoM
Combine suggested as part of #161735