Skip to content

Commit 48a4b14

Browse files
alexey-bataevalexey-bataev
committed
[SLP]Fix whole vector registers calculations for compares
Need to check that the calculated number of the elements is not larger than the original number of scalars to prevent a compiler crash. Fixes llvm#134013
1 parent 65ed353 commit 48a4b14

File tree

2 files changed

+63
-0
lines changed

2 files changed

+63
-0
lines changed

‎llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -8873,6 +8873,7 @@ void BoUpSLP::buildTree_rec(ArrayRef<Value *> VL, unsigned Depth,
88738873
// Find the number of elements, which forms full vectors.
88748874
unsigned PWSz = getFullVectorNumberOfElements(
88758875
*TTI, UniqueValues.front()->getType(), UniqueValues.size());
8876+
PWSz = std::min<unsigned>(PWSz, VL.size());
88768877
if (PWSz == VL.size()) {
88778878
ReuseShuffleIndices.clear();
88788879
} else {

‎llvm/test/Transforms/SLPVectorizer/X86/whole-registers-compare.ll

Lines changed: 62 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,62 @@
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 -slp-threshold=-99999 < %s | FileCheck %s
3+
4+
define void @test() {
5+
; CHECK-LABEL: define void @test() {
6+
; CHECK-NEXT: [[ENTRY:.*]]:
7+
; CHECK-NEXT: br label %[[DO_BODY:.*]]
8+
; CHECK: [[DO_BODY]]:
9+
; CHECK-NEXT: [[TMP0:%.*]] = phi <12 x float> [ zeroinitializer, %[[ENTRY]] ], [ [[TMP7:%.*]], %[[DO_BODY]] ]
10+
; CHECK-NEXT: [[CMP119:%.*]] = fcmp uge float 0.000000e+00, 0.000000e+00
11+
; CHECK-NEXT: [[VAL_SROA_6_1:%.*]] = select i1 [[CMP119]], float 0.000000e+00, float 0.000000e+00
12+
; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x float> <float 0.000000e+00, float poison>, float [[VAL_SROA_6_1]], i32 1
13+
; CHECK-NEXT: [[TMP2:%.*]] = fcmp uge <2 x float> [[TMP1]], zeroinitializer
14+
; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <2 x i1> [[TMP2]], <2 x i1> poison, <12 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
15+
; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <12 x i1> <i1 true, i1 undef, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 undef, i1 undef, i1 undef>, <12 x i1> [[TMP3]], <12 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 12, i32 13, i32 11>
16+
; CHECK-NEXT: [[TMP5:%.*]] = insertelement <12 x i1> [[TMP4]], i1 [[CMP119]], i32 11
17+
; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <12 x i1> [[TMP5]], <12 x i1> poison, <12 x i32> <i32 0, i32 0, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>
18+
; CHECK-NEXT: [[TMP7]] = select <12 x i1> [[TMP6]], <12 x float> zeroinitializer, <12 x float> zeroinitializer
19+
; CHECK-NEXT: br label %[[DO_BODY]]
20+
;
21+
entry:
22+
br label %do.body
23+
24+
do.body:
25+
%val.sroa.66.0 = phi float [ 0.000000e+00, %entry ], [ %val.sroa.66.1, %do.body ]
26+
%val.sroa.60.0 = phi float [ 0.000000e+00, %entry ], [ %val.sroa.60.2, %do.body ]
27+
%val.sroa.54.0 = phi float [ 0.000000e+00, %entry ], [ %val.sroa.54.2, %do.body ]
28+
%val.sroa.48.0 = phi float [ 0.000000e+00, %entry ], [ %val.sroa.48.2, %do.body ]
29+
%val.sroa.42.0 = phi float [ 0.000000e+00, %entry ], [ %val.sroa.42.2, %do.body ]
30+
%val.sroa.36.0 = phi float [ 0.000000e+00, %entry ], [ %val.sroa.36.2, %do.body ]
31+
%val.sroa.30.0 = phi float [ 0.000000e+00, %entry ], [ %val.sroa.30.2, %do.body ]
32+
%val.sroa.24.0 = phi float [ 0.000000e+00, %entry ], [ %val.sroa.24.2, %do.body ]
33+
%val.sroa.18.0 = phi float [ 0.000000e+00, %entry ], [ %val.sroa.18.2, %do.body ]
34+
%val.sroa.12.0 = phi float [ 0.000000e+00, %entry ], [ %val.sroa.12.2, %do.body ]
35+
%val.sroa.6.0 = phi float [ 0.000000e+00, %entry ], [ %val.sroa.6.2, %do.body ]
36+
%val.sroa.0.0 = phi float [ 0.000000e+00, %entry ], [ %val.sroa.0.1, %do.body ]
37+
%cmp119 = fcmp uge float 0.000000e+00, 0.000000e+00
38+
%val.sroa.6.1 = select i1 %cmp119, float 0.000000e+00, float 0.000000e+00
39+
%val.sroa.0.1 = select i1 %cmp119, float 0.000000e+00, float 0.000000e+00
40+
%cmp119.1 = fcmp uge float %val.sroa.6.1, 0.000000e+00
41+
%val.sroa.6.2 = select i1 %cmp119.1, float 0.000000e+00, float 0.000000e+00
42+
%cmp119.2 = fcmp uge float 0.000000e+00, 0.000000e+00
43+
%val.sroa.12.2 = select i1 %cmp119.2, float 0.000000e+00, float 0.000000e+00
44+
%cmp119.3 = fcmp uge float 0.000000e+00, 0.000000e+00
45+
%val.sroa.18.2 = select i1 %cmp119.3, float 0.000000e+00, float 0.000000e+00
46+
%cmp119.4 = fcmp uge float 0.000000e+00, 0.000000e+00
47+
%val.sroa.24.2 = select i1 %cmp119.4, float 0.000000e+00, float 0.000000e+00
48+
%cmp119.5 = fcmp uge float 0.000000e+00, 0.000000e+00
49+
%val.sroa.30.2 = select i1 %cmp119.5, float 0.000000e+00, float 0.000000e+00
50+
%cmp119.6 = fcmp uge float 0.000000e+00, 0.000000e+00
51+
%val.sroa.36.2 = select i1 %cmp119.6, float 0.000000e+00, float 0.000000e+00
52+
%cmp119.7 = fcmp uge float 0.000000e+00, 0.000000e+00
53+
%val.sroa.42.2 = select i1 %cmp119.7, float 0.000000e+00, float 0.000000e+00
54+
%cmp119.8 = fcmp uge float 0.000000e+00, 0.000000e+00
55+
%val.sroa.48.2 = select i1 %cmp119.8, float 0.000000e+00, float 0.000000e+00
56+
%cmp119.9 = fcmp uge float 0.000000e+00, 0.000000e+00
57+
%val.sroa.54.2 = select i1 %cmp119.9, float 0.000000e+00, float 0.000000e+00
58+
%cmp119.10 = fcmp uge float 0.000000e+00, 0.000000e+00
59+
%val.sroa.66.1 = select i1 %cmp119.10, float 0.000000e+00, float 0.000000e+00
60+
%val.sroa.60.2 = select i1 %cmp119.10, float 0.000000e+00, float 0.000000e+00
61+
br label %do.body
62+
}

0 commit comments

Comments
 (0)