Skip to content

Commit ead9d6a

Browse files
committed
[SLP]Check VectorizableTree is not empty before accessing elements
Need to check VectorizableTree is not empty before accessing elements. Fixes #131635
1 parent cfa07cc commit ead9d6a

File tree

2 files changed

+49
-1
lines changed

2 files changed

+49
-1
lines changed

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -8893,7 +8893,7 @@ void BoUpSLP::buildTree_rec(ArrayRef<Value *> VL, unsigned Depth,
88938893
TTI.getArithmeticInstrCost(Opcode1, Op2VecTy, Kind);
88948894
InstructionCost NewCost =
88958895
NewVecOpsCost + InsertCost +
8896-
(VectorizableTree.front()->hasState() &&
8896+
(!VectorizableTree.empty() && VectorizableTree.front()->hasState() &&
88978897
VectorizableTree.front()->getOpcode() == Instruction::Store
88988898
? NewShuffleCost
88998899
: 0);
Lines changed: 48 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,48 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
2+
; RUN: opt -S --passes=slp-vectorizer -mtriple=x86_64-unknown-linux-gnu -mcpu=core-avx2 < %s | FileCheck %s
3+
4+
%struct.1 = type { [258 x [1 x i32]], [56 x i8], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [2 x [16 x [1 x i32]]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [10 x [16 x [1 x i32]]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [16 x [1 x i32]], [2 x [1 x i32]], [56 x i8], [2 x [1 x i32]], [56 x i8], [2 x [1 x i32]], [56 x i8] }
5+
6+
define <8 x float> @foo(i32 %0, i32 %1, i32 %2) {
7+
; CHECK-LABEL: define <8 x float> @foo(
8+
; CHECK-SAME: i32 [[TMP0:%.*]], i32 [[TMP1:%.*]], i32 [[TMP2:%.*]]) #[[ATTR0:[0-9]+]] {
9+
; CHECK-NEXT: [[ENTRY:.*:]]
10+
; CHECK-NEXT: [[VECEXT_I615:%.*]] = extractelement <4 x i32> zeroinitializer, i64 0
11+
; CHECK-NEXT: [[IDX_EXT:%.*]] = sext i32 [[VECEXT_I615]] to i64
12+
; CHECK-NEXT: [[__U_I613:%.*]] = getelementptr i8, ptr getelementptr inbounds ([[STRUCT_1:%.*]], ptr null, i64 -35, i32 37, i64 0, i64 0), i64 [[IDX_EXT]]
13+
; CHECK-NEXT: [[IDX_EXT320:%.*]] = sext i32 [[TMP0]] to i64
14+
; CHECK-NEXT: [[__U_I611:%.*]] = getelementptr i8, ptr getelementptr inbounds ([[STRUCT_1]], ptr null, i64 -35, i32 37, i64 0, i64 0), i64 [[IDX_EXT320]]
15+
; CHECK-NEXT: [[IDX_EXT326:%.*]] = sext i32 [[TMP1]] to i64
16+
; CHECK-NEXT: [[__U_I609:%.*]] = getelementptr i8, ptr getelementptr inbounds ([[STRUCT_1]], ptr null, i64 -35, i32 37, i64 0, i64 0), i64 [[IDX_EXT326]]
17+
; CHECK-NEXT: [[IDX_EXT332:%.*]] = sext i32 [[TMP2]] to i64
18+
; CHECK-NEXT: [[__U_I607:%.*]] = getelementptr i8, ptr getelementptr inbounds ([[STRUCT_1]], ptr null, i64 -35, i32 37, i64 0, i64 0), i64 [[IDX_EXT332]]
19+
; CHECK-NEXT: [[IDX_EXT338:%.*]] = zext i32 [[TMP2]] to i64
20+
; CHECK-NEXT: [[__U_I605:%.*]] = getelementptr i8, ptr getelementptr inbounds ([[STRUCT_1]], ptr null, i64 -35, i32 37, i64 0, i64 0), i64 [[IDX_EXT338]]
21+
; CHECK-NEXT: [[IDX_EXT344:%.*]] = zext i32 [[TMP1]] to i64
22+
; CHECK-NEXT: [[__U_I603:%.*]] = getelementptr i8, ptr getelementptr inbounds ([[STRUCT_1]], ptr null, i64 -35, i32 37, i64 0, i64 0), i64 [[IDX_EXT344]]
23+
; CHECK-NEXT: [[IDX_EXT350:%.*]] = zext i32 0 to i64
24+
; CHECK-NEXT: [[__U_I601:%.*]] = getelementptr i8, ptr getelementptr inbounds ([[STRUCT_1]], ptr null, i64 -35, i32 37, i64 0, i64 0), i64 [[IDX_EXT350]]
25+
; CHECK-NEXT: [[IDX_EXT356:%.*]] = zext i32 [[TMP0]] to i64
26+
; CHECK-NEXT: [[__U_I:%.*]] = getelementptr i8, ptr getelementptr inbounds ([[STRUCT_1]], ptr null, i64 -35, i32 37, i64 0, i64 0), i64 [[IDX_EXT356]]
27+
; CHECK-NEXT: ret <8 x float> zeroinitializer
28+
;
29+
entry:
30+
%vecext.i615 = extractelement <4 x i32> zeroinitializer, i64 0
31+
%idx.ext = sext i32 %vecext.i615 to i64
32+
%__u.i613 = getelementptr i8, ptr getelementptr inbounds (%struct.1, ptr null, i64 -35, i32 37, i64 0, i64 0), i64 %idx.ext
33+
%idx.ext320 = sext i32 %0 to i64
34+
%__u.i611 = getelementptr i8, ptr getelementptr inbounds (%struct.1, ptr null, i64 -35, i32 37, i64 0, i64 0), i64 %idx.ext320
35+
%idx.ext326 = sext i32 %1 to i64
36+
%__u.i609 = getelementptr i8, ptr getelementptr inbounds (%struct.1, ptr null, i64 -35, i32 37, i64 0, i64 0), i64 %idx.ext326
37+
%idx.ext332 = sext i32 %2 to i64
38+
%__u.i607 = getelementptr i8, ptr getelementptr inbounds (%struct.1, ptr null, i64 -35, i32 37, i64 0, i64 0), i64 %idx.ext332
39+
%idx.ext338 = zext i32 %2 to i64
40+
%__u.i605 = getelementptr i8, ptr getelementptr inbounds (%struct.1, ptr null, i64 -35, i32 37, i64 0, i64 0), i64 %idx.ext338
41+
%idx.ext344 = zext i32 %1 to i64
42+
%__u.i603 = getelementptr i8, ptr getelementptr inbounds (%struct.1, ptr null, i64 -35, i32 37, i64 0, i64 0), i64 %idx.ext344
43+
%idx.ext350 = zext i32 0 to i64
44+
%__u.i601 = getelementptr i8, ptr getelementptr inbounds (%struct.1, ptr null, i64 -35, i32 37, i64 0, i64 0), i64 %idx.ext350
45+
%idx.ext356 = zext i32 %0 to i64
46+
%__u.i = getelementptr i8, ptr getelementptr inbounds (%struct.1, ptr null, i64 -35, i32 37, i64 0, i64 0), i64 %idx.ext356
47+
ret <8 x float> zeroinitializer
48+
}

0 commit comments

Comments
 (0)