Skip to content

Commit 23ad559

Browse files
committed
[SLP][REVEC] Ignore UserTreeIndex if it is empty.
Previously, the all_of check did not consider the case where the TreeEntry is empty (i.e., when it is the first entry).
1 parent e92f2ee commit 23ad559

File tree

2 files changed

+18
-3
lines changed

2 files changed

+18
-3
lines changed

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

Lines changed: 4 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -6449,9 +6449,10 @@ void BoUpSLP::reorderTopToBottom() {
64496449
assert(SLPReVec && "Only supported by REVEC.");
64506450
// ShuffleVectorInst does not do reorderOperands (and it should not
64516451
// because ShuffleVectorInst supports only a limited set of
6452-
// patterns). Only do reorderNodeWithReuses if all of the users are
6453-
// not ShuffleVectorInst.
6454-
if (isa<ShuffleVectorInst>(TE->UserTreeIndex.UserTE->getMainOp()))
6452+
// patterns). Only do reorderNodeWithReuses if the user is not
6453+
// ShuffleVectorInst.
6454+
if (TE->UserTreeIndex &&
6455+
isa<ShuffleVectorInst>(TE->UserTreeIndex.UserTE->getMainOp()))
64556456
continue;
64566457
assert((!TE->UserTreeIndex ||
64576458
!isa<ShuffleVectorInst>(

llvm/test/Transforms/SLPVectorizer/X86/revec-reduced-value-replace-extractelement.ll

Lines changed: 14 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,20 @@
22
; RUN: opt -S --passes=slp-vectorizer -slp-threshold=-99999 < %s -mtriple=x86_64-unknown-linux-gnu -slp-revec | FileCheck %s
33

44
define void @test() {
5+
; CHECK-LABEL: define void @test() {
6+
; CHECK-NEXT: [[BB:.*]]:
7+
; CHECK-NEXT: [[TRUNC:%.*]] = trunc i64 0 to i32
8+
; CHECK-NEXT: br label %[[BB1:.*]]
9+
; CHECK: [[BB1]]:
10+
; CHECK-NEXT: [[TMP0:%.*]] = phi <2 x i32> [ zeroinitializer, %[[BB]] ], [ [[TMP4:%.*]], %[[BB1]] ]
11+
; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x i32> [[TMP0]], i32 1
12+
; CHECK-NEXT: [[TMP2:%.*]] = call i8 @llvm.vector.reduce.mul.v4i8(<4 x i8> zeroinitializer)
13+
; CHECK-NEXT: [[TMP3:%.*]] = zext i8 [[TMP2]] to i32
14+
; CHECK-NEXT: [[OP_RDX:%.*]] = mul i32 [[TMP3]], [[TMP1]]
15+
; CHECK-NEXT: [[OP_RDX1:%.*]] = mul i32 [[OP_RDX]], [[TRUNC]]
16+
; CHECK-NEXT: [[TMP4]] = insertelement <2 x i32> <i32 0, i32 poison>, i32 [[OP_RDX1]], i32 1
17+
; CHECK-NEXT: br label %[[BB1]]
18+
;
519
bb:
620
br label %bb1
721

0 commit comments

Comments
 (0)