[VectorCombine] foldShuffleToIdentity - peek through bitcasts to see if they come from the same value to form identity sequence #98334
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
|
@llvm/pr-subscribers-llvm-transforms Author: Simon Pilgrim (RKSimon) ChangesWorkaround until I can get #96884 fixed properly - when trying to find identity sequences, peek through any bitcasts to see if the values all came from the same source. We don't run CSE frequently enough to merge all the bitcasts that we end up with. Full diff: https://github.com/llvm/llvm-project/pull/98334.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
index 4896b8ed2595b..3a49f95d3f117 100644
--- a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
+++ b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
@@ -1859,13 +1859,19 @@ bool VectorCombine::foldShuffleToIdentity(Instruction &I) {
if (!FrontU)
return false;
+ // Helper to peek through bitcasts to the same value.
+ auto IsEquiv = [&](Value *X, Value *Y) {
+ return X->getType() == Y->getType() &&
+ peekThroughBitcasts(X) == peekThroughBitcasts(Y);
+ };
+
// Look for an identity value.
if (FrontLane == 0 &&
cast<FixedVectorType>(FrontU->get()->getType())->getNumElements() ==
Ty->getNumElements() &&
- all_of(drop_begin(enumerate(Item)), [Item](const auto &E) {
+ all_of(drop_begin(enumerate(Item)), [IsEquiv, Item](const auto &E) {
Value *FrontV = Item.front().first->get();
- return !E.value().first || (E.value().first->get() == FrontV &&
+ return !E.value().first || (IsEquiv(E.value().first->get(), FrontV) &&
E.value().second == (int)E.index());
})) {
IdentityLeafs.insert(FrontU);
diff --git a/llvm/test/Transforms/PhaseOrdering/X86/blendv-select.ll b/llvm/test/Transforms/PhaseOrdering/X86/blendv-select.ll
index 011d51600b51f..c2ed7b9c84523 100644
--- a/llvm/test/Transforms/PhaseOrdering/X86/blendv-select.ll
+++ b/llvm/test/Transforms/PhaseOrdering/X86/blendv-select.ll
@@ -88,20 +88,10 @@ define <4 x i64> @x86_pblendvb_v8i32_v4i32(<4 x i64> %a, <4 x i64> %b, <4 x i64>
; CHECK-NEXT: [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <8 x i32>
; CHECK-NEXT: [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <8 x i32>
; CHECK-NEXT: [[CMP:%.*]] = icmp slt <8 x i32> [[C_BC]], [[D_BC]]
-; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <8 x i1> [[CMP]], <8 x i1> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <4 x i64> [[A:%.*]] to <8 x i32>
-; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <8 x i32> [[TMP2]], <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT: [[TMP4:%.*]] = bitcast <4 x i64> [[B:%.*]] to <8 x i32>
-; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <8 x i32> [[TMP4]], <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT: [[TMP6:%.*]] = select <4 x i1> [[TMP1]], <4 x i32> [[TMP5]], <4 x i32> [[TMP3]]
-; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <8 x i1> [[CMP]], <8 x i1> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: [[TMP8:%.*]] = bitcast <4 x i64> [[A]] to <8 x i32>
-; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <8 x i32> [[TMP8]], <8 x i32> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: [[TMP10:%.*]] = bitcast <4 x i64> [[B]] to <8 x i32>
-; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <8 x i32> [[TMP10]], <8 x i32> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: [[TMP12:%.*]] = select <4 x i1> [[TMP7]], <4 x i32> [[TMP11]], <4 x i32> [[TMP9]]
-; CHECK-NEXT: [[TMP13:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> [[TMP12]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: [[RES:%.*]] = bitcast <8 x i32> [[TMP13]] to <4 x i64>
+; CHECK-NEXT: [[TMP1:%.*]] = bitcast <4 x i64> [[A:%.*]] to <8 x i32>
+; CHECK-NEXT: [[TMP2:%.*]] = bitcast <4 x i64> [[B:%.*]] to <8 x i32>
+; CHECK-NEXT: [[TMP3:%.*]] = select <8 x i1> [[CMP]], <8 x i32> [[TMP2]], <8 x i32> [[TMP1]]
+; CHECK-NEXT: [[RES:%.*]] = bitcast <8 x i32> [[TMP3]] to <4 x i64>
; CHECK-NEXT: ret <4 x i64> [[RES]]
;
%a.bc = bitcast <4 x i64> %a to <32 x i8>
@@ -129,20 +119,10 @@ define <4 x i64> @x86_pblendvb_v16i16_v8i16(<4 x i64> %a, <4 x i64> %b, <4 x i64
; CHECK-NEXT: [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <16 x i16>
; CHECK-NEXT: [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <16 x i16>
; CHECK-NEXT: [[CMP:%.*]] = icmp slt <16 x i16> [[C_BC]], [[D_BC]]
-; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <16 x i1> [[CMP]], <16 x i1> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <4 x i64> [[A:%.*]] to <16 x i16>
-; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <16 x i16> [[TMP2]], <16 x i16> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: [[TMP4:%.*]] = bitcast <4 x i64> [[B:%.*]] to <16 x i16>
-; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <16 x i16> [[TMP4]], <16 x i16> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: [[TMP6:%.*]] = select <8 x i1> [[TMP1]], <8 x i16> [[TMP5]], <8 x i16> [[TMP3]]
-; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <16 x i1> [[CMP]], <16 x i1> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
-; CHECK-NEXT: [[TMP8:%.*]] = bitcast <4 x i64> [[A]] to <16 x i16>
-; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <16 x i16> [[TMP8]], <16 x i16> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
-; CHECK-NEXT: [[TMP10:%.*]] = bitcast <4 x i64> [[B]] to <16 x i16>
-; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <16 x i16> [[TMP10]], <16 x i16> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
-; CHECK-NEXT: [[TMP12:%.*]] = select <8 x i1> [[TMP7]], <8 x i16> [[TMP11]], <8 x i16> [[TMP9]]
-; CHECK-NEXT: [[TMP13:%.*]] = shufflevector <8 x i16> [[TMP6]], <8 x i16> [[TMP12]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
-; CHECK-NEXT: [[RES:%.*]] = bitcast <16 x i16> [[TMP13]] to <4 x i64>
+; CHECK-NEXT: [[TMP1:%.*]] = bitcast <4 x i64> [[A:%.*]] to <16 x i16>
+; CHECK-NEXT: [[TMP2:%.*]] = bitcast <4 x i64> [[B:%.*]] to <16 x i16>
+; CHECK-NEXT: [[TMP3:%.*]] = select <16 x i1> [[CMP]], <16 x i16> [[TMP2]], <16 x i16> [[TMP1]]
+; CHECK-NEXT: [[RES:%.*]] = bitcast <16 x i16> [[TMP3]] to <4 x i64>
; CHECK-NEXT: ret <4 x i64> [[RES]]
;
%a.bc = bitcast <4 x i64> %a to <32 x i8>
@@ -276,20 +256,10 @@ define <8 x i64> @x86_pblendvb_v16i32_v8i32(<8 x i64> %a, <8 x i64> %b, <8 x i64
; CHECK-NEXT: [[C_BC:%.*]] = bitcast <8 x i64> [[C:%.*]] to <16 x i32>
; CHECK-NEXT: [[D_BC:%.*]] = bitcast <8 x i64> [[D:%.*]] to <16 x i32>
; CHECK-NEXT: [[CMP:%.*]] = icmp slt <16 x i32> [[C_BC]], [[D_BC]]
-; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <16 x i1> [[CMP]], <16 x i1> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x i64> [[A:%.*]] to <16 x i32>
-; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <16 x i32> [[TMP2]], <16 x i32> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: [[TMP4:%.*]] = bitcast <8 x i64> [[B:%.*]] to <16 x i32>
-; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <16 x i32> [[TMP4]], <16 x i32> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: [[TMP6:%.*]] = select <8 x i1> [[TMP1]], <8 x i32> [[TMP5]], <8 x i32> [[TMP3]]
-; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <16 x i1> [[CMP]], <16 x i1> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
-; CHECK-NEXT: [[TMP8:%.*]] = bitcast <8 x i64> [[A]] to <16 x i32>
-; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <16 x i32> [[TMP8]], <16 x i32> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
-; CHECK-NEXT: [[TMP10:%.*]] = bitcast <8 x i64> [[B]] to <16 x i32>
-; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <16 x i32> [[TMP10]], <16 x i32> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
-; CHECK-NEXT: [[TMP12:%.*]] = select <8 x i1> [[TMP7]], <8 x i32> [[TMP11]], <8 x i32> [[TMP9]]
-; CHECK-NEXT: [[TMP13:%.*]] = shufflevector <8 x i32> [[TMP6]], <8 x i32> [[TMP12]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
-; CHECK-NEXT: [[RES:%.*]] = bitcast <16 x i32> [[TMP13]] to <8 x i64>
+; CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x i64> [[A:%.*]] to <16 x i32>
+; CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x i64> [[B:%.*]] to <16 x i32>
+; CHECK-NEXT: [[TMP3:%.*]] = select <16 x i1> [[CMP]], <16 x i32> [[TMP2]], <16 x i32> [[TMP1]]
+; CHECK-NEXT: [[RES:%.*]] = bitcast <16 x i32> [[TMP3]] to <8 x i64>
; CHECK-NEXT: ret <8 x i64> [[RES]]
;
%a.bc = bitcast <8 x i64> %a to <64 x i8>
@@ -317,20 +287,10 @@ define <8 x i64> @x86_pblendvb_v32i16_v16i16(<8 x i64> %a, <8 x i64> %b, <8 x i6
; CHECK-NEXT: [[C_BC:%.*]] = bitcast <8 x i64> [[C:%.*]] to <32 x i16>
; CHECK-NEXT: [[D_BC:%.*]] = bitcast <8 x i64> [[D:%.*]] to <32 x i16>
; CHECK-NEXT: [[CMP:%.*]] = icmp slt <32 x i16> [[C_BC]], [[D_BC]]
-; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <32 x i1> [[CMP]], <32 x i1> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
-; CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x i64> [[A:%.*]] to <32 x i16>
-; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <32 x i16> [[TMP2]], <32 x i16> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
-; CHECK-NEXT: [[TMP4:%.*]] = bitcast <8 x i64> [[B:%.*]] to <32 x i16>
-; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <32 x i16> [[TMP4]], <32 x i16> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
-; CHECK-NEXT: [[TMP6:%.*]] = select <16 x i1> [[TMP1]], <16 x i16> [[TMP5]], <16 x i16> [[TMP3]]
-; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <32 x i1> [[CMP]], <32 x i1> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, 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: [[TMP8:%.*]] = bitcast <8 x i64> [[A]] to <32 x i16>
-; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <32 x i16> [[TMP8]], <32 x i16> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, 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: [[TMP10:%.*]] = bitcast <8 x i64> [[B]] to <32 x i16>
-; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <32 x i16> [[TMP10]], <32 x i16> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, 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: [[TMP12:%.*]] = select <16 x i1> [[TMP7]], <16 x i16> [[TMP11]], <16 x i16> [[TMP9]]
-; CHECK-NEXT: [[TMP13:%.*]] = shufflevector <16 x i16> [[TMP6]], <16 x i16> [[TMP12]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, 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: [[RES:%.*]] = bitcast <32 x i16> [[TMP13]] to <8 x i64>
+; CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x i64> [[A:%.*]] to <32 x i16>
+; CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x i64> [[B:%.*]] to <32 x i16>
+; CHECK-NEXT: [[TMP3:%.*]] = select <32 x i1> [[CMP]], <32 x i16> [[TMP2]], <32 x i16> [[TMP1]]
+; CHECK-NEXT: [[RES:%.*]] = bitcast <32 x i16> [[TMP3]] to <8 x i64>
; CHECK-NEXT: ret <8 x i64> [[RES]]
;
%a.bc = bitcast <8 x i64> %a to <64 x i8>
|
davemgreen
reviewed
Jul 14, 2024
RKSimon
added a commit
that referenced
this pull request
Jul 15, 2024
Don't rely on phaseordering tests alone
…if they come from the same value to form identity sequence Workaround until I can get llvm#96884 fixed properly - when trying to find identity sequences, peek through any bitcasts to see if the values all came from the same source. We don't run CSE frequently enough to merge all the bitcasts that we end up with.
RKSimon
force-pushed
the
shuffle-identity-equivalent
branch
from
829ea31 to
50bcca9
Compare
|
Any more comments? |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/17/builds/858 Here is the relevant piece of the build log for the reference: |
yuxuanchen1997
pushed a commit
that referenced
this pull request
Jul 25, 2024
…if they come from the same value to form identity sequence (#98334) Summary: Workaround until I can get #96884 fixed properly - when trying to find identity sequences, peek through any bitcasts to see if the values all came from the same source. We don't run CSE frequently enough to merge all the bitcasts that we end up with. Test Plan: Reviewers: Subscribers: Tasks: Tags: Differential Revision: https://phabricator.intern.facebook.com/D60251606
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.
Workaround until I can get #96884 fixed properly - when trying to find identity sequences, peek through any bitcasts to see if the values all came from the same source. We don't run CSE frequently enough to merge all the bitcasts that we end up with.