release/19.x: [VectorCombine] Do not try to operate on OperandBundles. (#111635) #111796
+79
−28
Conversation
This file contains 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
|
@RKSimon What do you think about merging this PR to the release branch? |
|
@llvm/pr-subscribers-llvm-transforms Author: None (llvmbot) ChangesBackport c136d32 Requested by: @davemgreen Full diff: https://github.com/llvm/llvm-project/pull/111796.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
index 444598520c981a..679934d07e36d0 100644
--- a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
+++ b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
@@ -1900,33 +1900,35 @@ bool VectorCombine::foldShuffleToIdentity(Instruction &I) {
// We need each element to be the same type of value, and check that each
// element has a single use.
- if (all_of(drop_begin(Item), [Item](InstLane IL) {
- Value *FrontV = Item.front().first->get();
- if (!IL.first)
- return true;
- Value *V = IL.first->get();
- if (auto *I = dyn_cast<Instruction>(V); I && !I->hasOneUse())
- return false;
- if (V->getValueID() != FrontV->getValueID())
- return false;
- if (auto *CI = dyn_cast<CmpInst>(V))
- if (CI->getPredicate() != cast<CmpInst>(FrontV)->getPredicate())
- return false;
- if (auto *CI = dyn_cast<CastInst>(V))
- if (CI->getSrcTy() != cast<CastInst>(FrontV)->getSrcTy())
- return false;
- if (auto *SI = dyn_cast<SelectInst>(V))
- if (!isa<VectorType>(SI->getOperand(0)->getType()) ||
- SI->getOperand(0)->getType() !=
- cast<SelectInst>(FrontV)->getOperand(0)->getType())
- return false;
- if (isa<CallInst>(V) && !isa<IntrinsicInst>(V))
- return false;
- auto *II = dyn_cast<IntrinsicInst>(V);
- return !II || (isa<IntrinsicInst>(FrontV) &&
- II->getIntrinsicID() ==
- cast<IntrinsicInst>(FrontV)->getIntrinsicID());
- })) {
+ auto CheckLaneIsEquivalentToFirst = [Item](InstLane IL) {
+ Value *FrontV = Item.front().first->get();
+ if (!IL.first)
+ return true;
+ Value *V = IL.first->get();
+ if (auto *I = dyn_cast<Instruction>(V); I && !I->hasOneUse())
+ return false;
+ if (V->getValueID() != FrontV->getValueID())
+ return false;
+ if (auto *CI = dyn_cast<CmpInst>(V))
+ if (CI->getPredicate() != cast<CmpInst>(FrontV)->getPredicate())
+ return false;
+ if (auto *CI = dyn_cast<CastInst>(V))
+ if (CI->getSrcTy() != cast<CastInst>(FrontV)->getSrcTy())
+ return false;
+ if (auto *SI = dyn_cast<SelectInst>(V))
+ if (!isa<VectorType>(SI->getOperand(0)->getType()) ||
+ SI->getOperand(0)->getType() !=
+ cast<SelectInst>(FrontV)->getOperand(0)->getType())
+ return false;
+ if (isa<CallInst>(V) && !isa<IntrinsicInst>(V))
+ return false;
+ auto *II = dyn_cast<IntrinsicInst>(V);
+ return !II || (isa<IntrinsicInst>(FrontV) &&
+ II->getIntrinsicID() ==
+ cast<IntrinsicInst>(FrontV)->getIntrinsicID() &&
+ !II->hasOperandBundles());
+ };
+ if (all_of(drop_begin(Item), CheckLaneIsEquivalentToFirst)) {
// Check the operator is one that we support.
if (isa<BinaryOperator, CmpInst>(FrontU)) {
// We exclude div/rem in case they hit UB from poison lanes.
@@ -1954,7 +1956,8 @@ bool VectorCombine::foldShuffleToIdentity(Instruction &I) {
Worklist.push_back(generateInstLaneVectorFromOperand(Item, 2));
continue;
} else if (auto *II = dyn_cast<IntrinsicInst>(FrontU);
- II && isTriviallyVectorizable(II->getIntrinsicID())) {
+ II && isTriviallyVectorizable(II->getIntrinsicID()) &&
+ !II->hasOperandBundles()) {
for (unsigned Op = 0, E = II->getNumOperands() - 1; Op < E; Op++) {
if (isVectorIntrinsicWithScalarOpAtArg(II->getIntrinsicID(), Op)) {
if (!all_of(drop_begin(Item), [Item, Op](InstLane &IL) {
diff --git a/llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll b/llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll
index af04fb0ab4621b..66fe11369d88be 100644
--- a/llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll
+++ b/llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll
@@ -1066,4 +1066,52 @@ entry:
ret <2 x float> %4
}
+define <16 x i64> @operandbundles(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c) {
+; CHECK-LABEL: @operandbundles(
+; CHECK-NEXT: [[CALL:%.*]] = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> [[A:%.*]], <4 x i64> [[B:%.*]], <4 x i64> [[C:%.*]]) [ "jl_roots"(ptr addrspace(10) null, ptr addrspace(10) null) ]
+; CHECK-NEXT: [[SHUFFLEVECTOR:%.*]] = shufflevector <4 x i64> [[CALL]], <4 x i64> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, 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: [[SHUFFLEVECTOR1:%.*]] = shufflevector <16 x i64> [[SHUFFLEVECTOR]], <16 x i64> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT: ret <16 x i64> [[SHUFFLEVECTOR1]]
+;
+ %call = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c) [ "jl_roots"(ptr addrspace(10) null, ptr addrspace(10) null) ]
+ %shufflevector = shufflevector <4 x i64> %call, <4 x i64> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+ %shufflevector1 = shufflevector <16 x i64> %shufflevector, <16 x i64> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+ ret <16 x i64> %shufflevector1
+}
+
+define <8 x i8> @operandbundles_first(<8 x i8> %a) {
+; CHECK-LABEL: @operandbundles_first(
+; CHECK-NEXT: [[AB:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT: [[AT:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
+; CHECK-NEXT: [[ABT:%.*]] = call <4 x i8> @llvm.abs.v4i8(<4 x i8> [[AT]], i1 false) [ "jl_roots"(ptr addrspace(10) null, ptr addrspace(10) null) ]
+; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x i8> [[AT]], <4 x i8> [[AB]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT: [[R:%.*]] = call <8 x i8> @llvm.abs.v8i8(<8 x i8> [[TMP1]], i1 false)
+; CHECK-NEXT: ret <8 x i8> [[R]]
+;
+ %ab = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+ %at = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
+ %abt = call <4 x i8> @llvm.abs.v4i8(<4 x i8> %at, i1 false) [ "jl_roots"(ptr addrspace(10) null, ptr addrspace(10) null) ]
+ %abb = call <4 x i8> @llvm.abs.v4i8(<4 x i8> %ab, i1 false)
+ %r = shufflevector <4 x i8> %abt, <4 x i8> %abb, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+ ret <8 x i8> %r
+}
+
+define <8 x i8> @operandbundles_second(<8 x i8> %a) {
+; CHECK-LABEL: @operandbundles_second(
+; CHECK-NEXT: [[AB:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT: [[AT:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
+; CHECK-NEXT: [[ABB:%.*]] = call <4 x i8> @llvm.abs.v4i8(<4 x i8> [[AB]], i1 false) [ "jl_roots"(ptr addrspace(10) null, ptr addrspace(10) null) ]
+; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x i8> [[AT]], <4 x i8> [[AB]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT: [[R:%.*]] = call <8 x i8> @llvm.abs.v8i8(<8 x i8> [[TMP1]], i1 false)
+; CHECK-NEXT: ret <8 x i8> [[R]]
+;
+ %ab = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+ %at = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
+ %abt = call <4 x i8> @llvm.abs.v4i8(<4 x i8> %at, i1 false)
+ %abb = call <4 x i8> @llvm.abs.v4i8(<4 x i8> %ab, i1 false) [ "jl_roots"(ptr addrspace(10) null, ptr addrspace(10) null) ]
+ %r = shufflevector <4 x i8> %abt, <4 x i8> %abb, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+ ret <8 x i8> %r
+}
+
+declare <4 x i64> @llvm.fshl.v4i64(<4 x i64>, <4 x i64>, <4 x i64>)
declare void @use(<4 x i8>)
|
|
@davemgreen The CI build failure is confusing - do OperandBundles occur in 19.x? |
|
I must admit OperandBundles are a new one for me. It has apparently been around for a while, and I had assumed from JuliaPackaging/Yggdrasil#9204 that this was related to pulling clang-19 into Julia. I will have to have a look at what is behaving differently in the test case. I'm hoping they just need to be regenerated due to the other changes in vectorcombine since clang-19. |
davemgreen
force-pushed
the
issue110382
branch
from
5092575 to
c6bcc80
Compare
|
It did look like the abs/shuffle was being optimized differently in the rest of vector combine. |
|
@RKSimon @phoebewang ok to backport? |
LLVM, rather than Clang specifically, but yes, we're updating it to v19 for Julia. Maybe @Zentrik can comment about how they got into #110382, if necessary. |
This bails out if we see an intrinsic with an operand bundle on it, to make sure we don't process the bundles incorrectly. Fixes llvm#110382. (cherry picked from commit c136d32)
|
@davemgreen (or anyone else). If you would like to add a note about this fix in the release notes (completely optional). Please reply to this comment with a one or two sentence description of the fix. When you are done, please add the release:note label to this PR. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Backport c136d32
Requested by: @davemgreen