Skip to content

Commit 133c122

Browse files
alexey-bataevalexey-bataev
committed
[SLP]Fix a crash on accessing element with index -1 for reused mask with PoisonMaskElem
Need to check if the index from the ReuseShuffleIndices mask is not equal to PoisonMaskElem before trying to access the element by index.
1 parent 667815c commit 133c122

File tree

2 files changed

+188
-0
lines changed

2 files changed

+188
-0
lines changed

‎llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4578,6 +4578,8 @@ BoUpSLP::findReusedOrderedScalars(const BoUpSLP::TreeEntry &TE) {
45784578
continue;
45794579
if (!TE.ReuseShuffleIndices.empty())
45804580
K = TE.ReuseShuffleIndices[K];
4581+
if (K == PoisonMaskElem)
4582+
continue;
45814583
if (!TE.ReorderIndices.empty())
45824584
K = std::distance(TE.ReorderIndices.begin(),
45834585
find(TE.ReorderIndices, K));

‎llvm/test/Transforms/SLPVectorizer/X86/reused-mask-with-poison-index.ll

Lines changed: 186 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,186 @@
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 < %s | FileCheck %s
3+
4+
define fastcc i32 @test(ptr %0, <2 x float> %1, i1 %2, float %3, float %4) {
5+
; CHECK-LABEL: define fastcc i32 @test(
6+
; CHECK-SAME: ptr [[TMP0:%.*]], <2 x float> [[TMP1:%.*]], i1 [[TMP2:%.*]], float [[TMP3:%.*]], float [[TMP4:%.*]]) {
7+
; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x float> [[TMP1]], i64 0
8+
; CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x float> [[TMP1]], i64 1
9+
; CHECK-NEXT: br label %[[BB8:.*]]
10+
; CHECK: [[BB8]]:
11+
; CHECK-NEXT: [[TMP9:%.*]] = phi float [ 0.000000e+00, [[TMP5:%.*]] ], [ [[TMP58:%.*]], %[[TMP56:.*]] ]
12+
; CHECK-NEXT: [[TMP10:%.*]] = phi float [ 0.000000e+00, [[TMP5]] ], [ [[TMP59:%.*]], %[[TMP56]] ]
13+
; CHECK-NEXT: [[TMP11:%.*]] = phi float [ [[TMP4]], [[TMP5]] ], [ [[TMP60:%.*]], %[[TMP56]] ]
14+
; CHECK-NEXT: [[TMP12:%.*]] = phi float [ [[TMP7]], [[TMP5]] ], [ [[TMP61:%.*]], %[[TMP56]] ]
15+
; CHECK-NEXT: [[TMP13:%.*]] = phi float [ [[TMP6]], [[TMP5]] ], [ [[TMP62:%.*]], %[[TMP56]] ]
16+
; CHECK-NEXT: [[TMP14:%.*]] = phi float [ 0.000000e+00, [[TMP5]] ], [ [[TMP63:%.*]], %[[TMP56]] ]
17+
; CHECK-NEXT: [[TMP15:%.*]] = phi float [ 0.000000e+00, [[TMP5]] ], [ [[TMP64:%.*]], %[[TMP56]] ]
18+
; CHECK-NEXT: [[TMP16:%.*]] = phi float [ 0.000000e+00, [[TMP5]] ], [ [[TMP65:%.*]], %[[TMP56]] ]
19+
; CHECK-NEXT: [[TMP17:%.*]] = phi float [ undef, [[TMP5]] ], [ [[TMP66:%.*]], %[[TMP56]] ]
20+
; CHECK-NEXT: [[TMP18:%.*]] = phi float [ 0.000000e+00, [[TMP5]] ], [ [[TMP67:%.*]], %[[TMP56]] ]
21+
; CHECK-NEXT: [[TMP19:%.*]] = phi float [ [[TMP4]], [[TMP5]] ], [ [[TMP68:%.*]], %[[TMP56]] ]
22+
; CHECK-NEXT: [[TMP20:%.*]] = phi float [ [[TMP4]], [[TMP5]] ], [ [[TMP69:%.*]], %[[TMP56]] ]
23+
; CHECK-NEXT: [[TMP21:%.*]] = phi float [ [[TMP4]], [[TMP5]] ], [ [[TMP70:%.*]], %[[TMP56]] ]
24+
; CHECK-NEXT: [[TMP22:%.*]] = phi float [ [[TMP4]], [[TMP5]] ], [ [[TMP71:%.*]], %[[TMP56]] ]
25+
; CHECK-NEXT: [[TMP23:%.*]] = phi float [ 0.000000e+00, [[TMP5]] ], [ [[TMP73:%.*]], %[[TMP56]] ]
26+
; CHECK-NEXT: [[TMP24:%.*]] = phi float [ 0.000000e+00, [[TMP5]] ], [ [[TMP72:%.*]], %[[TMP56]] ]
27+
; CHECK-NEXT: [[TMP25:%.*]] = phi <4 x float> [ zeroinitializer, [[TMP5]] ], [ poison, %[[TMP56]] ]
28+
; CHECK-NEXT: [[TMP26:%.*]] = phi <2 x float> [ zeroinitializer, [[TMP5]] ], [ poison, %[[TMP56]] ]
29+
; CHECK-NEXT: br i1 false, label %[[BB57:.*]], label %[[BB27:.*]]
30+
; CHECK: [[BB27]]:
31+
; CHECK-NEXT: [[TMP28:%.*]] = fcmp olt float [[TMP22]], 0.000000e+00
32+
; CHECK-NEXT: br i1 [[TMP28]], label %[[BB57]], label %[[BB29:.*]]
33+
; CHECK: [[BB29]]:
34+
; CHECK-NEXT: [[TMP30:%.*]] = fcmp olt float [[TMP21]], 0.000000e+00
35+
; CHECK-NEXT: br i1 [[TMP30]], label %[[BB57]], label %[[BB31:.*]]
36+
; CHECK: [[BB31]]:
37+
; CHECK-NEXT: [[TMP32:%.*]] = insertelement <4 x float> poison, float [[TMP3]], i32 0
38+
; CHECK-NEXT: [[TMP33:%.*]] = shufflevector <4 x float> [[TMP32]], <4 x float> poison, <4 x i32> zeroinitializer
39+
; CHECK-NEXT: [[TMP34:%.*]] = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> [[TMP33]], <4 x float> zeroinitializer, <4 x float> zeroinitializer)
40+
; CHECK-NEXT: [[TMP35:%.*]] = extractelement <4 x float> [[TMP25]], i32 0
41+
; CHECK-NEXT: [[TMP36:%.*]] = fsub float [[TMP17]], [[TMP35]]
42+
; CHECK-NEXT: [[TMP37:%.*]] = extractelement <4 x float> [[TMP25]], i32 1
43+
; CHECK-NEXT: [[TMP38:%.*]] = fsub float [[TMP15]], [[TMP37]]
44+
; CHECK-NEXT: [[TMP39:%.*]] = insertelement <2 x float> zeroinitializer, float [[TMP14]], i64 0
45+
; CHECK-NEXT: [[TMP40:%.*]] = shufflevector <4 x float> [[TMP25]], <4 x float> poison, <2 x i32> <i32 poison, i32 2>
46+
; CHECK-NEXT: [[TMP41:%.*]] = shufflevector <2 x float> [[TMP39]], <2 x float> [[TMP40]], <2 x i32> <i32 0, i32 3>
47+
; CHECK-NEXT: [[TMP42:%.*]] = insertelement <2 x float> zeroinitializer, float [[TMP38]], i64 0
48+
; CHECK-NEXT: [[TMP43:%.*]] = insertelement <2 x float> zeroinitializer, float [[TMP36]], i64 0
49+
; CHECK-NEXT: [[TMP44:%.*]] = fmul <2 x float> [[TMP42]], [[TMP43]]
50+
; CHECK-NEXT: [[TMP45:%.*]] = tail call <2 x float> @llvm.fmuladd.v2f32(<2 x float> [[TMP41]], <2 x float> [[TMP26]], <2 x float> [[TMP44]])
51+
; CHECK-NEXT: [[TMP46:%.*]] = extractelement <2 x float> [[TMP45]], i64 0
52+
; CHECK-NEXT: [[TMP47:%.*]] = fcmp ogt float [[TMP46]], 0.000000e+00
53+
; CHECK-NEXT: br i1 [[TMP47]], label %[[BB48:.*]], label %[[BB50:.*]]
54+
; CHECK: [[BB48]]:
55+
; CHECK-NEXT: br label %[[BB50]]
56+
; CHECK: [[BB49:.*]]:
57+
; CHECK-NEXT: br label %[[BB50]]
58+
; CHECK: [[BB50]]:
59+
; CHECK-NEXT: [[TMP51:%.*]] = phi float [ [[TMP12]], %[[BB49]] ], [ [[TMP19]], %[[BB48]] ], [ 0.000000e+00, %[[BB31]] ]
60+
; CHECK-NEXT: [[TMP52:%.*]] = phi float [ [[TMP13]], %[[BB49]] ], [ [[TMP20]], %[[BB48]] ], [ 0.000000e+00, %[[BB31]] ]
61+
; CHECK-NEXT: br i1 [[TMP2]], label %[[BB57]], label %[[BB53:.*]]
62+
; CHECK: [[BB53]]:
63+
; CHECK-NEXT: [[TMP54:%.*]] = extractelement <2 x float> [[TMP1]], i64 0
64+
; CHECK-NEXT: [[TMP55:%.*]] = extractelement <4 x float> [[TMP25]], i32 3
65+
; CHECK-NEXT: br label %[[BB57]]
66+
; CHECK: [[TMP56]]:
67+
; CHECK-NEXT: br label %[[BB8]]
68+
; CHECK: [[BB57]]:
69+
; CHECK-NEXT: [[TMP58]] = phi float [ [[TMP9]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ 0.000000e+00, %[[BB53]] ], [ [[TMP3]], %[[BB50]] ]
70+
; CHECK-NEXT: [[TMP59]] = phi float [ [[TMP10]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ [[TMP55]], %[[BB53]] ], [ 0.000000e+00, %[[BB50]] ]
71+
; CHECK-NEXT: [[TMP60]] = phi float [ [[TMP11]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ 0.000000e+00, %[[BB53]] ], [ [[TMP3]], %[[BB50]] ]
72+
; CHECK-NEXT: [[TMP61]] = phi float [ [[TMP12]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ 0.000000e+00, %[[BB53]] ], [ 0.000000e+00, %[[BB50]] ]
73+
; CHECK-NEXT: [[TMP62]] = phi float [ [[TMP13]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ [[TMP54]], %[[BB53]] ], [ 0.000000e+00, %[[BB50]] ]
74+
; CHECK-NEXT: [[TMP63]] = phi float [ [[TMP14]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ 0.000000e+00, %[[BB53]] ], [ [[TMP9]], %[[BB50]] ]
75+
; CHECK-NEXT: [[TMP64]] = phi float [ [[TMP15]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ [[TMP55]], %[[BB53]] ], [ [[TMP10]], %[[BB50]] ]
76+
; CHECK-NEXT: [[TMP65]] = phi float [ [[TMP16]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ 0.000000e+00, %[[BB53]] ], [ [[TMP11]], %[[BB50]] ]
77+
; CHECK-NEXT: [[TMP66]] = phi float [ [[TMP17]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ 0.000000e+00, %[[BB53]] ], [ 0.000000e+00, %[[BB50]] ]
78+
; CHECK-NEXT: [[TMP67]] = phi float [ [[TMP18]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ 0.000000e+00, %[[BB53]] ], [ 0.000000e+00, %[[BB50]] ]
79+
; CHECK-NEXT: [[TMP68]] = phi float [ [[TMP19]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ 0.000000e+00, %[[BB53]] ], [ [[TMP3]], %[[BB50]] ]
80+
; CHECK-NEXT: [[TMP69]] = phi float [ [[TMP20]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ [[TMP54]], %[[BB53]] ], [ 0.000000e+00, %[[BB50]] ]
81+
; CHECK-NEXT: [[TMP70]] = phi float [ [[TMP21]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ 0.000000e+00, %[[BB53]] ], [ [[TMP51]], %[[BB50]] ]
82+
; CHECK-NEXT: [[TMP71]] = phi float [ [[TMP22]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ [[TMP54]], %[[BB53]] ], [ [[TMP52]], %[[BB50]] ]
83+
; CHECK-NEXT: [[TMP72]] = phi float [ [[TMP24]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ 0.000000e+00, %[[BB53]] ], [ [[TMP24]], %[[BB50]] ]
84+
; CHECK-NEXT: [[TMP73]] = phi float [ [[TMP23]], %[[BB29]] ], [ 0.000000e+00, %[[BB27]] ], [ 0.000000e+00, %[[BB8]] ], [ 0.000000e+00, %[[BB53]] ], [ [[TMP23]], %[[BB50]] ]
85+
; CHECK-NEXT: [[TMP74:%.*]] = phi <4 x float> [ [[TMP25]], %[[BB29]] ], [ [[TMP25]], %[[BB27]] ], [ zeroinitializer, %[[BB8]] ], [ [[TMP34]], %[[BB53]] ], [ [[TMP34]], %[[BB50]] ]
86+
; CHECK-NEXT: [[TMP75:%.*]] = tail call float @llvm.fmuladd.f32(float [[TMP73]], float [[TMP72]], float 0.000000e+00)
87+
; CHECK-NEXT: store float [[TMP75]], ptr [[TMP0]], align 4
88+
; CHECK-NEXT: ret i32 0
89+
;
90+
%6 = extractelement <2 x float> %1, i64 0
91+
%7 = extractelement <2 x float> %1, i64 1
92+
br label %8
93+
94+
8:
95+
%9 = phi float [ 0.000000e+00, %5 ], [ %59, %57 ]
96+
%10 = phi float [ 0.000000e+00, %5 ], [ %60, %57 ]
97+
%11 = phi float [ 0.000000e+00, %5 ], [ %61, %57 ]
98+
%12 = phi float [ %4, %5 ], [ %62, %57 ]
99+
%13 = phi float [ %7, %5 ], [ %63, %57 ]
100+
%14 = phi float [ %6, %5 ], [ %64, %57 ]
101+
%15 = phi float [ 0.000000e+00, %5 ], [ %65, %57 ]
102+
%16 = phi float [ 0.000000e+00, %5 ], [ %66, %57 ]
103+
%17 = phi float [ 0.000000e+00, %5 ], [ %67, %57 ]
104+
%18 = phi float [ 0.000000e+00, %5 ], [ %68, %57 ]
105+
%19 = phi float [ undef, %5 ], [ %69, %57 ]
106+
%20 = phi float [ 0.000000e+00, %5 ], [ %70, %57 ]
107+
%21 = phi float [ 0.000000e+00, %5 ], [ %71, %57 ]
108+
%22 = phi float [ %4, %5 ], [ %72, %57 ]
109+
%23 = phi float [ %4, %5 ], [ %73, %57 ]
110+
%24 = phi float [ %4, %5 ], [ %74, %57 ]
111+
%25 = phi float [ %4, %5 ], [ %75, %57 ]
112+
%26 = phi float [ 0.000000e+00, %5 ], [ %76, %57 ]
113+
%27 = phi float [ 0.000000e+00, %5 ], [ %78, %57 ]
114+
%28 = phi float [ 0.000000e+00, %5 ], [ %77, %57 ]
115+
br i1 false, label %58, label %29
116+
117+
29:
118+
%30 = fcmp olt float %25, 0.000000e+00
119+
br i1 %30, label %58, label %31
120+
121+
31:
122+
%32 = fcmp olt float %24, 0.000000e+00
123+
br i1 %32, label %58, label %33
124+
125+
33:
126+
%34 = tail call float @llvm.fmuladd.f32(float %3, float 0.000000e+00, float 0.000000e+00)
127+
%35 = tail call float @llvm.fmuladd.f32(float %3, float 0.000000e+00, float 0.000000e+00)
128+
%36 = tail call float @llvm.fmuladd.f32(float %3, float 0.000000e+00, float 0.000000e+00)
129+
%37 = tail call float @llvm.fmuladd.f32(float %3, float 0.000000e+00, float 0.000000e+00)
130+
%38 = insertelement <2 x float> zeroinitializer, float %17, i64 0
131+
%39 = insertelement <2 x float> %38, float %21, i64 1
132+
%40 = fsub float %19, %18
133+
%41 = fsub float %16, %20
134+
%42 = insertelement <2 x float> zeroinitializer, float %15, i64 0
135+
%43 = insertelement <2 x float> %42, float %26, i64 1
136+
%44 = insertelement <2 x float> zeroinitializer, float %41, i64 0
137+
%45 = insertelement <2 x float> zeroinitializer, float %40, i64 0
138+
%46 = fmul <2 x float> %44, %45
139+
%47 = tail call <2 x float> @llvm.fmuladd.v2f32(<2 x float> %43, <2 x float> %39, <2 x float> %46)
140+
%48 = extractelement <2 x float> %47, i64 0
141+
%49 = fcmp ogt float %48, 0.000000e+00
142+
br i1 %49, label %50, label %52
143+
144+
50:
145+
br label %52
146+
147+
51:
148+
br label %52
149+
150+
52:
151+
%53 = phi float [ %13, %51 ], [ %22, %50 ], [ 0.000000e+00, %33 ]
152+
%54 = phi float [ %14, %51 ], [ %23, %50 ], [ 0.000000e+00, %33 ]
153+
br i1 %2, label %58, label %55
154+
155+
55:
156+
%56 = extractelement <2 x float> %1, i64 0
157+
br label %58
158+
159+
57:
160+
br label %8
161+
162+
58:
163+
%59 = phi float [ %9, %31 ], [ %9, %29 ], [ 0.000000e+00, %8 ], [ %36, %55 ], [ %36, %52 ]
164+
%60 = phi float [ %10, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ 0.000000e+00, %55 ], [ %3, %52 ]
165+
%61 = phi float [ %11, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ %9, %55 ], [ 0.000000e+00, %52 ]
166+
%62 = phi float [ %12, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ 0.000000e+00, %55 ], [ %3, %52 ]
167+
%63 = phi float [ %13, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ 0.000000e+00, %55 ], [ 0.000000e+00, %52 ]
168+
%64 = phi float [ %14, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ %56, %55 ], [ 0.000000e+00, %52 ]
169+
%65 = phi float [ %15, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ 0.000000e+00, %55 ], [ %10, %52 ]
170+
%66 = phi float [ %16, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ %9, %55 ], [ %11, %52 ]
171+
%67 = phi float [ %17, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ 0.000000e+00, %55 ], [ %12, %52 ]
172+
%68 = phi float [ %18, %31 ], [ %18, %29 ], [ 0.000000e+00, %8 ], [ %35, %55 ], [ %35, %52 ]
173+
%69 = phi float [ %19, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ 0.000000e+00, %55 ], [ 0.000000e+00, %52 ]
174+
%70 = phi float [ %20, %31 ], [ %20, %29 ], [ 0.000000e+00, %8 ], [ %34, %55 ], [ %34, %52 ]
175+
%71 = phi float [ %21, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ 0.000000e+00, %55 ], [ 0.000000e+00, %52 ]
176+
%72 = phi float [ %22, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ 0.000000e+00, %55 ], [ %3, %52 ]
177+
%73 = phi float [ %23, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ %56, %55 ], [ 0.000000e+00, %52 ]
178+
%74 = phi float [ %24, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ 0.000000e+00, %55 ], [ %53, %52 ]
179+
%75 = phi float [ %25, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ %56, %55 ], [ %54, %52 ]
180+
%76 = phi float [ %26, %31 ], [ %26, %29 ], [ 0.000000e+00, %8 ], [ %37, %55 ], [ %37, %52 ]
181+
%77 = phi float [ %28, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ 0.000000e+00, %55 ], [ %28, %52 ]
182+
%78 = phi float [ %27, %31 ], [ 0.000000e+00, %29 ], [ 0.000000e+00, %8 ], [ 0.000000e+00, %55 ], [ %27, %52 ]
183+
%79 = tail call float @llvm.fmuladd.f32(float %78, float %77, float 0.000000e+00)
184+
store float %79, ptr %0, align 4
185+
ret i32 0
186+
}

0 commit comments

Comments
 (0)