1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
| ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -basicaa -slp-vectorizer -S | FileCheck %s -check-prefixes=CHECK,SSE
; RUN: opt < %s -basicaa -slp-vectorizer -S -mattr=+avx | FileCheck %s -check-prefixes=CHECK,AVX
target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-apple-macosx10.10.0"
define void @testfunc(float* nocapture %dest, float* nocapture readonly %src) {
; SSE-LABEL: @testfunc(
; SSE-NEXT: entry:
; SSE-NEXT: br label [[FOR_BODY:%.*]]
; SSE: for.body:
; SSE-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
; SSE-NEXT: [[ACC1_056:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[ADD13:%.*]], [[FOR_BODY]] ]
; SSE-NEXT: [[S1_055:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[COND_I40:%.*]], [[FOR_BODY]] ]
; SSE-NEXT: [[S0_054:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[COND_I44:%.*]], [[FOR_BODY]] ]
; SSE-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[SRC:%.*]], i64 [[INDVARS_IV]]
; SSE-NEXT: [[TMP0:%.*]] = load float, float* [[ARRAYIDX]], align 4
; SSE-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
; SSE-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[DEST:%.*]], i64 [[INDVARS_IV]]
; SSE-NEXT: store float [[ACC1_056]], float* [[ARRAYIDX2]], align 4
; SSE-NEXT: [[ADD:%.*]] = fadd float [[S0_054]], [[TMP0]]
; SSE-NEXT: [[ADD3:%.*]] = fadd float [[S1_055]], [[TMP0]]
; SSE-NEXT: [[MUL:%.*]] = fmul float [[S0_054]], 0.000000e+00
; SSE-NEXT: [[ADD4:%.*]] = fadd float [[MUL]], [[ADD3]]
; SSE-NEXT: [[MUL5:%.*]] = fmul float [[S1_055]], 0.000000e+00
; SSE-NEXT: [[ADD6:%.*]] = fadd float [[MUL5]], [[ADD]]
; SSE-NEXT: [[CMP_I:%.*]] = fcmp olt float [[ADD6]], 1.000000e+00
; SSE-NEXT: [[COND_I:%.*]] = select i1 [[CMP_I]], float [[ADD6]], float 1.000000e+00
; SSE-NEXT: [[CMP_I51:%.*]] = fcmp olt float [[COND_I]], -1.000000e+00
; SSE-NEXT: [[CMP_I49:%.*]] = fcmp olt float [[ADD4]], 1.000000e+00
; SSE-NEXT: [[COND_I50:%.*]] = select i1 [[CMP_I49]], float [[ADD4]], float 1.000000e+00
; SSE-NEXT: [[CMP_I47:%.*]] = fcmp olt float [[COND_I50]], -1.000000e+00
; SSE-NEXT: [[COND_I_OP:%.*]] = fmul float [[COND_I]], 0.000000e+00
; SSE-NEXT: [[MUL10:%.*]] = select i1 [[CMP_I51]], float -0.000000e+00, float [[COND_I_OP]]
; SSE-NEXT: [[COND_I50_OP:%.*]] = fmul float [[COND_I50]], 0.000000e+00
; SSE-NEXT: [[MUL11:%.*]] = select i1 [[CMP_I47]], float -0.000000e+00, float [[COND_I50_OP]]
; SSE-NEXT: [[ADD13]] = fadd float [[MUL10]], [[MUL11]]
; SSE-NEXT: [[CMP_I45:%.*]] = fcmp olt float [[ADD13]], 1.000000e+00
; SSE-NEXT: [[COND_I46:%.*]] = select i1 [[CMP_I45]], float [[ADD13]], float 1.000000e+00
; SSE-NEXT: [[CMP_I43:%.*]] = fcmp olt float [[COND_I46]], -1.000000e+00
; SSE-NEXT: [[COND_I44]] = select i1 [[CMP_I43]], float -1.000000e+00, float [[COND_I46]]
; SSE-NEXT: [[CMP_I41:%.*]] = fcmp olt float [[MUL11]], 1.000000e+00
; SSE-NEXT: [[COND_I42:%.*]] = select i1 [[CMP_I41]], float [[MUL11]], float 1.000000e+00
; SSE-NEXT: [[CMP_I39:%.*]] = fcmp olt float [[COND_I42]], -1.000000e+00
; SSE-NEXT: [[COND_I40]] = select i1 [[CMP_I39]], float -1.000000e+00, float [[COND_I42]]
; SSE-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 32
; SSE-NEXT: br i1 [[EXITCOND]], label [[FOR_END:%.*]], label [[FOR_BODY]]
; SSE: for.end:
; SSE-NEXT: ret void
;
; AVX-LABEL: @testfunc(
; AVX-NEXT: entry:
; AVX-NEXT: br label [[FOR_BODY:%.*]]
; AVX: for.body:
; AVX-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
; AVX-NEXT: [[ACC1_056:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[ADD13:%.*]], [[FOR_BODY]] ]
; AVX-NEXT: [[TMP0:%.*]] = phi <2 x float> [ zeroinitializer, [[ENTRY]] ], [ [[TMP23:%.*]], [[FOR_BODY]] ]
; AVX-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[SRC:%.*]], i64 [[INDVARS_IV]]
; AVX-NEXT: [[TMP1:%.*]] = load float, float* [[ARRAYIDX]], align 4
; AVX-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
; AVX-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[DEST:%.*]], i64 [[INDVARS_IV]]
; AVX-NEXT: store float [[ACC1_056]], float* [[ARRAYIDX2]], align 4
; AVX-NEXT: [[TMP2:%.*]] = extractelement <2 x float> [[TMP0]], i32 1
; AVX-NEXT: [[TMP3:%.*]] = insertelement <2 x float> undef, float [[TMP2]], i32 0
; AVX-NEXT: [[TMP4:%.*]] = extractelement <2 x float> [[TMP0]], i32 0
; AVX-NEXT: [[TMP5:%.*]] = insertelement <2 x float> [[TMP3]], float [[TMP4]], i32 1
; AVX-NEXT: [[TMP6:%.*]] = insertelement <2 x float> undef, float [[TMP1]], i32 0
; AVX-NEXT: [[TMP7:%.*]] = insertelement <2 x float> [[TMP6]], float [[TMP1]], i32 1
; AVX-NEXT: [[TMP8:%.*]] = fadd <2 x float> [[TMP5]], [[TMP7]]
; AVX-NEXT: [[TMP9:%.*]] = fmul <2 x float> [[TMP0]], zeroinitializer
; AVX-NEXT: [[TMP10:%.*]] = fadd <2 x float> [[TMP9]], [[TMP8]]
; AVX-NEXT: [[TMP11:%.*]] = fcmp olt <2 x float> [[TMP10]], <float 1.000000e+00, float 1.000000e+00>
; AVX-NEXT: [[TMP12:%.*]] = select <2 x i1> [[TMP11]], <2 x float> [[TMP10]], <2 x float> <float 1.000000e+00, float 1.000000e+00>
; AVX-NEXT: [[TMP13:%.*]] = fcmp olt <2 x float> [[TMP12]], <float -1.000000e+00, float -1.000000e+00>
; AVX-NEXT: [[TMP14:%.*]] = fmul <2 x float> [[TMP12]], zeroinitializer
; AVX-NEXT: [[TMP15:%.*]] = select <2 x i1> [[TMP13]], <2 x float> <float -0.000000e+00, float -0.000000e+00>, <2 x float> [[TMP14]]
; AVX-NEXT: [[TMP16:%.*]] = extractelement <2 x float> [[TMP15]], i32 0
; AVX-NEXT: [[TMP17:%.*]] = extractelement <2 x float> [[TMP15]], i32 1
; AVX-NEXT: [[ADD13]] = fadd float [[TMP16]], [[TMP17]]
; AVX-NEXT: [[TMP18:%.*]] = insertelement <2 x float> undef, float [[TMP17]], i32 0
; AVX-NEXT: [[TMP19:%.*]] = insertelement <2 x float> [[TMP18]], float [[ADD13]], i32 1
; AVX-NEXT: [[TMP20:%.*]] = fcmp olt <2 x float> [[TMP19]], <float 1.000000e+00, float 1.000000e+00>
; AVX-NEXT: [[TMP21:%.*]] = select <2 x i1> [[TMP20]], <2 x float> [[TMP19]], <2 x float> <float 1.000000e+00, float 1.000000e+00>
; AVX-NEXT: [[TMP22:%.*]] = fcmp olt <2 x float> [[TMP21]], <float -1.000000e+00, float -1.000000e+00>
; AVX-NEXT: [[TMP23]] = select <2 x i1> [[TMP22]], <2 x float> <float -1.000000e+00, float -1.000000e+00>, <2 x float> [[TMP21]]
; AVX-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 32
; AVX-NEXT: br i1 [[EXITCOND]], label [[FOR_END:%.*]], label [[FOR_BODY]]
; AVX: for.end:
; AVX-NEXT: ret void
;
entry:
br label %for.body
for.body:
%indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
%acc1.056 = phi float [ 0.000000e+00, %entry ], [ %add13, %for.body ]
%s1.055 = phi float [ 0.000000e+00, %entry ], [ %cond.i40, %for.body ]
%s0.054 = phi float [ 0.000000e+00, %entry ], [ %cond.i44, %for.body ]
%arrayidx = getelementptr inbounds float, float* %src, i64 %indvars.iv
%0 = load float, float* %arrayidx, align 4
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
%arrayidx2 = getelementptr inbounds float, float* %dest, i64 %indvars.iv
store float %acc1.056, float* %arrayidx2, align 4
%add = fadd float %s0.054, %0
%add3 = fadd float %s1.055, %0
%mul = fmul float %s0.054, 0.000000e+00
%add4 = fadd float %mul, %add3
%mul5 = fmul float %s1.055, 0.000000e+00
%add6 = fadd float %mul5, %add
%cmp.i = fcmp olt float %add6, 1.000000e+00
%cond.i = select i1 %cmp.i, float %add6, float 1.000000e+00
%cmp.i51 = fcmp olt float %cond.i, -1.000000e+00
%cmp.i49 = fcmp olt float %add4, 1.000000e+00
%cond.i50 = select i1 %cmp.i49, float %add4, float 1.000000e+00
%cmp.i47 = fcmp olt float %cond.i50, -1.000000e+00
%cond.i.op = fmul float %cond.i, 0.000000e+00
%mul10 = select i1 %cmp.i51, float -0.000000e+00, float %cond.i.op
%cond.i50.op = fmul float %cond.i50, 0.000000e+00
%mul11 = select i1 %cmp.i47, float -0.000000e+00, float %cond.i50.op
%add13 = fadd float %mul10, %mul11
; The SLPVectorizer crashed in vectorizeChainsInBlock() because it tried
; to access the second operand of the following cmp after the cmp itself
; was already vectorized and deleted.
%cmp.i45 = fcmp olt float %add13, 1.000000e+00
%cond.i46 = select i1 %cmp.i45, float %add13, float 1.000000e+00
%cmp.i43 = fcmp olt float %cond.i46, -1.000000e+00
%cond.i44 = select i1 %cmp.i43, float -1.000000e+00, float %cond.i46
%cmp.i41 = fcmp olt float %mul11, 1.000000e+00
%cond.i42 = select i1 %cmp.i41, float %mul11, float 1.000000e+00
%cmp.i39 = fcmp olt float %cond.i42, -1.000000e+00
%cond.i40 = select i1 %cmp.i39, float -1.000000e+00, float %cond.i42
%exitcond = icmp eq i64 %indvars.iv.next, 32
br i1 %exitcond, label %for.end, label %for.body
for.end:
ret void
}
|