LoopVectorize: guard marking iv as scalar; fix bug #88730
Merged
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
When collecting loop scalars, LoopVectorize over-eagerly marks the induction variable and its update as scalars after vectorization, even if the induction variable update is a first-order recurrence. Guard the process with this check, fixing a crash. Fixes llvm#72969.
|
@llvm/pr-subscribers-llvm-transforms Author: Ramkumar Ramachandra (artagnon) ChangesWhen collecting loop scalars, LoopVectorize over-eagerly marks the induction variable and its update as scalars after vectorization, even if the induction variable update is a first-order recurrence. Guard the process with this check, fixing a crash. Fixes #72969. Full diff: https://github.com/llvm/llvm-project/pull/88730.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 5535cc55e93216..fa11127e496888 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -3875,6 +3875,13 @@ void LoopVectorizationCostModel::collectLoopScalars(ElementCount VF) {
if (!ScalarInd)
continue;
+ // If the induction variable update is a fixed-order recurrence, neither the
+ // induction variable or its update should be marked scalar after
+ // vectorization.
+ auto *IndUpdatePhi = dyn_cast<PHINode>(IndUpdate);
+ if (IndUpdatePhi && Legal->isFixedOrderRecurrence(IndUpdatePhi))
+ continue;
+
// Determine if all users of the induction variable update instruction are
// scalar after vectorization.
auto ScalarIndUpdate =
diff --git a/llvm/test/Transforms/LoopVectorize/X86/pr72969.ll b/llvm/test/Transforms/LoopVectorize/X86/pr72969.ll
index f982695983330f..45af544f0304d4 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/pr72969.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/pr72969.ll
@@ -1,10 +1,111 @@
-; REQUIRES: asserts
-; RUN: not --crash opt -mtriple=x86_64 -mattr=-avx,-avx2,-avx512f,+sse,-sse2,-sse3,-sse4.2 -passes=loop-vectorize -S < %s
-; RUN: not --crash opt -mtriple=x86_64 -mattr=-avx,-avx2,-avx512f,+sse,-sse2,-sse3,-sse4.2 -passes=loop-vectorize -force-vector-width=4 -S < %s
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt -mtriple=x86_64 -mattr=-avx,-avx2,-avx512f,+sse,-sse2,-sse3,-sse4.2 -passes=loop-vectorize -S %s | FileCheck --check-prefix=NOVEC %s
+; RUN: opt -mtriple=x86_64 -mattr=-avx,-avx2,-avx512f,+sse,-sse2,-sse3,-sse4.2 -passes=loop-vectorize -force-vector-width=4 -S %s | FileCheck --check-prefix=VEC %s
@h = global i64 0
define void @test(ptr %p) {
+; NOVEC-LABEL: define void @test(
+; NOVEC-SAME: ptr [[P:%.*]]) #[[ATTR0:[0-9]+]] {
+; NOVEC-NEXT: entry:
+; NOVEC-NEXT: br label [[FOR_BODY:%.*]]
+; NOVEC: for.body:
+; NOVEC-NEXT: [[IDX_EXT_MERGE:%.*]] = phi i64 [ 1, [[ENTRY:%.*]] ], [ [[IDX:%.*]], [[FOR_BODY]] ]
+; NOVEC-NEXT: [[INC_MERGE:%.*]] = phi i16 [ 1, [[ENTRY]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
+; NOVEC-NEXT: [[IDX_MERGE:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[IDX_EXT_MERGE]], [[FOR_BODY]] ]
+; NOVEC-NEXT: [[ADD:%.*]] = shl i64 [[IDX_MERGE]], 1
+; NOVEC-NEXT: [[ARRAYIDX:%.*]] = getelementptr i64, ptr [[P]], i64 [[ADD]]
+; NOVEC-NEXT: store i64 0, ptr [[ARRAYIDX]], align 8
+; NOVEC-NEXT: [[INC]] = add i16 [[INC_MERGE]], 1
+; NOVEC-NEXT: [[IDX]] = zext i16 [[INC]] to i64
+; NOVEC-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[P]], i64 [[IDX]]
+; NOVEC-NEXT: [[CMP:%.*]] = icmp ugt ptr [[GEP]], @h
+; NOVEC-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[FOR_BODY]]
+; NOVEC: exit:
+; NOVEC-NEXT: ret void
+;
+; VEC-LABEL: define void @test(
+; VEC-SAME: ptr [[P:%.*]]) #[[ATTR0:[0-9]+]] {
+; VEC-NEXT: entry:
+; VEC-NEXT: [[P1:%.*]] = ptrtoint ptr [[P]] to i64
+; VEC-NEXT: [[TMP0:%.*]] = add i64 [[P1]], 16
+; VEC-NEXT: [[UMAX2:%.*]] = call i64 @llvm.umax.i64(i64 [[TMP0]], i64 add (i64 ptrtoint (ptr @h to i64), i64 1))
+; VEC-NEXT: [[TMP1:%.*]] = add i64 [[UMAX2]], -9
+; VEC-NEXT: [[TMP2:%.*]] = sub i64 [[TMP1]], [[P1]]
+; VEC-NEXT: [[TMP3:%.*]] = lshr i64 [[TMP2]], 3
+; VEC-NEXT: [[TMP4:%.*]] = add nuw nsw i64 [[TMP3]], 1
+; VEC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP4]], 4
+; VEC-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
+; VEC: vector.scevcheck:
+; VEC-NEXT: [[TMP5:%.*]] = add i64 [[P1]], 16
+; VEC-NEXT: [[UMAX:%.*]] = call i64 @llvm.umax.i64(i64 [[TMP5]], i64 add (i64 ptrtoint (ptr @h to i64), i64 1))
+; VEC-NEXT: [[TMP6:%.*]] = add i64 [[UMAX]], -9
+; VEC-NEXT: [[TMP7:%.*]] = sub i64 [[TMP6]], [[P1]]
+; VEC-NEXT: [[TMP8:%.*]] = lshr i64 [[TMP7]], 3
+; VEC-NEXT: [[TMP9:%.*]] = icmp ugt i64 [[TMP8]], 65535
+; VEC-NEXT: [[TMP10:%.*]] = trunc i64 [[TMP8]] to i16
+; VEC-NEXT: [[TMP11:%.*]] = add i16 2, [[TMP10]]
+; VEC-NEXT: [[TMP12:%.*]] = icmp ult i16 [[TMP11]], 2
+; VEC-NEXT: [[TMP13:%.*]] = icmp ugt i64 [[TMP8]], 65535
+; VEC-NEXT: [[TMP14:%.*]] = or i1 [[TMP12]], [[TMP13]]
+; VEC-NEXT: br i1 [[TMP14]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
+; VEC: vector.ph:
+; VEC-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP4]], 4
+; VEC-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP4]], [[N_MOD_VF]]
+; VEC-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i16
+; VEC-NEXT: [[IND_END:%.*]] = add i16 1, [[DOTCAST]]
+; VEC-NEXT: br label [[VECTOR_BODY:%.*]]
+; VEC: vector.body:
+; VEC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; VEC-NEXT: [[VECTOR_RECUR:%.*]] = phi <4 x i64> [ <i64 poison, i64 poison, i64 poison, i64 1>, [[VECTOR_PH]] ], [ [[TMP28:%.*]], [[VECTOR_BODY]] ]
+; VEC-NEXT: [[VEC_IND:%.*]] = phi <4 x i16> [ <i16 1, i16 2, i16 3, i16 4>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
+; VEC-NEXT: [[TMP15:%.*]] = add i64 [[INDEX]], 0
+; VEC-NEXT: [[TMP16:%.*]] = add i64 [[INDEX]], 1
+; VEC-NEXT: [[TMP17:%.*]] = add i64 [[INDEX]], 2
+; VEC-NEXT: [[TMP18:%.*]] = add i64 [[INDEX]], 3
+; VEC-NEXT: [[TMP19:%.*]] = shl i64 [[TMP15]], 1
+; VEC-NEXT: [[TMP20:%.*]] = shl i64 [[TMP16]], 1
+; VEC-NEXT: [[TMP21:%.*]] = shl i64 [[TMP17]], 1
+; VEC-NEXT: [[TMP22:%.*]] = shl i64 [[TMP18]], 1
+; VEC-NEXT: [[TMP23:%.*]] = getelementptr i64, ptr [[P]], i64 [[TMP19]]
+; VEC-NEXT: [[TMP24:%.*]] = getelementptr i64, ptr [[P]], i64 [[TMP20]]
+; VEC-NEXT: [[TMP25:%.*]] = getelementptr i64, ptr [[P]], i64 [[TMP21]]
+; VEC-NEXT: [[TMP26:%.*]] = getelementptr i64, ptr [[P]], i64 [[TMP22]]
+; VEC-NEXT: store i64 0, ptr [[TMP23]], align 8
+; VEC-NEXT: store i64 0, ptr [[TMP24]], align 8
+; VEC-NEXT: store i64 0, ptr [[TMP25]], align 8
+; VEC-NEXT: store i64 0, ptr [[TMP26]], align 8
+; VEC-NEXT: [[TMP27:%.*]] = add <4 x i16> [[VEC_IND]], <i16 1, i16 1, i16 1, i16 1>
+; VEC-NEXT: [[TMP28]] = zext <4 x i16> [[TMP27]] to <4 x i64>
+; VEC-NEXT: [[TMP29:%.*]] = shufflevector <4 x i64> [[VECTOR_RECUR]], <4 x i64> [[TMP28]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
+; VEC-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
+; VEC-NEXT: [[VEC_IND_NEXT]] = add <4 x i16> [[VEC_IND]], <i16 4, i16 4, i16 4, i16 4>
+; VEC-NEXT: [[TMP30:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; VEC-NEXT: br i1 [[TMP30]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; VEC: middle.block:
+; VEC-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP4]], [[N_VEC]]
+; VEC-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <4 x i64> [[TMP28]], i32 3
+; VEC-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
+; VEC: scalar.ph:
+; VEC-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i64 [ 1, [[VECTOR_SCEVCHECK]] ], [ 1, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
+; VEC-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 1, [[ENTRY]] ], [ 1, [[VECTOR_SCEVCHECK]] ]
+; VEC-NEXT: [[BC_RESUME_VAL3:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
+; VEC-NEXT: br label [[FOR_BODY:%.*]]
+; VEC: for.body:
+; VEC-NEXT: [[SCALAR_RECUR:%.*]] = phi i64 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[IDX:%.*]], [[FOR_BODY]] ]
+; VEC-NEXT: [[INC_MERGE:%.*]] = phi i16 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
+; VEC-NEXT: [[IDX_MERGE:%.*]] = phi i64 [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ], [ [[SCALAR_RECUR]], [[FOR_BODY]] ]
+; VEC-NEXT: [[ADD:%.*]] = shl i64 [[IDX_MERGE]], 1
+; VEC-NEXT: [[ARRAYIDX:%.*]] = getelementptr i64, ptr [[P]], i64 [[ADD]]
+; VEC-NEXT: store i64 0, ptr [[ARRAYIDX]], align 8
+; VEC-NEXT: [[INC]] = add i16 [[INC_MERGE]], 1
+; VEC-NEXT: [[IDX]] = zext i16 [[INC]] to i64
+; VEC-NEXT: [[GEP:%.*]] = getelementptr i64, ptr [[P]], i64 [[IDX]]
+; VEC-NEXT: [[CMP:%.*]] = icmp ugt ptr [[GEP]], @h
+; VEC-NEXT: br i1 [[CMP]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
+; VEC: exit:
+; VEC-NEXT: ret void
+;
entry:
br label %for.body
@@ -24,3 +125,9 @@ for.body:
exit:
ret void
}
+;.
+; VEC: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
+; VEC: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
+; VEC: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
+; VEC: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]]}
+;.
|
david-arm
reviewed
Apr 16, 2024
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.
When collecting loop scalars, LoopVectorize over-eagerly marks the induction variable and its update as scalars after vectorization, even if the induction variable update is a first-order recurrence. Guard the process with this check, fixing a crash.
Fixes #72969.