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
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
|
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -S --passes=slp-vectorizer -mtriple=x86_64-unknown-linux < %s | FileCheck %s
define <16 x half> @test(i32 %0, float %1, i32 %2) {
; CHECK-LABEL: define <16 x half> @test(
; CHECK-SAME: i32 [[TMP0:%.*]], float [[TMP1:%.*]], i32 [[TMP2:%.*]]) {
; CHECK-NEXT: [[TMP4:%.*]] = insertelement <16 x float> <float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float poison, float 0.000000e+00, float 0.000000e+00>, float [[TMP1]], i32 13
; CHECK-NEXT: [[TMP5:%.*]] = bitcast <16 x float> [[TMP4]] to <16 x i32>
; CHECK-NEXT: [[TMP6:%.*]] = call <2 x float> @llvm.fmuladd.v2f32(<2 x float> zeroinitializer, <2 x float> zeroinitializer, <2 x float> zeroinitializer)
; CHECK-NEXT: [[TMP7:%.*]] = icmp ugt i32 [[TMP2]], 0
; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i32 [[TMP0]], [[TMP0]]
; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x float> [[TMP6]], i32 0
; CHECK-NEXT: [[TMP10:%.*]] = fcmp ogt float [[TMP9]], 0.000000e+00
; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i32 [[TMP0]], 0
; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i32 [[TMP0]], 0
; CHECK-NEXT: [[TMP13:%.*]] = extractelement <16 x i32> [[TMP5]], i32 10
; CHECK-NEXT: [[TMP14:%.*]] = icmp eq i32 [[TMP13]], 0
; CHECK-NEXT: [[TMP15:%.*]] = bitcast float 0.000000e+00 to i32
; CHECK-NEXT: [[TMP16:%.*]] = icmp eq i32 [[TMP15]], 0
; CHECK-NEXT: [[TMP17:%.*]] = icmp eq i32 [[TMP0]], 0
; CHECK-NEXT: [[TMP18:%.*]] = icmp ult i32 0, 0
; CHECK-NEXT: [[TMP19:%.*]] = icmp ult i32 0, 0
; CHECK-NEXT: [[TMP20:%.*]] = icmp ult i32 0, 0
; CHECK-NEXT: [[TMP21:%.*]] = icmp eq i32 0, 0
; CHECK-NEXT: [[TMP22:%.*]] = extractelement <16 x i32> [[TMP5]], i32 4
; CHECK-NEXT: [[TMP23:%.*]] = icmp eq i32 [[TMP22]], 0
; CHECK-NEXT: [[TMP24:%.*]] = icmp eq i32 [[TMP0]], [[TMP0]]
; CHECK-NEXT: [[TMP25:%.*]] = bitcast float 0.000000e+00 to i32
; CHECK-NEXT: [[TMP26:%.*]] = icmp eq i32 [[TMP25]], 0
; CHECK-NEXT: [[TMP27:%.*]] = icmp ult <16 x i32> [[TMP5]], zeroinitializer
; CHECK-NEXT: [[TMP28:%.*]] = select <16 x i1> [[TMP27]], <16 x i32> zeroinitializer, <16 x i32> zeroinitializer
; CHECK-NEXT: [[TMP29:%.*]] = sitofp <16 x i32> [[TMP28]] to <16 x float>
; CHECK-NEXT: [[TMP30:%.*]] = call <16 x float> @llvm.fmuladd.v16f32(<16 x float> [[TMP29]], <16 x float> zeroinitializer, <16 x float> zeroinitializer)
; CHECK-NEXT: [[TMP31:%.*]] = fadd <16 x float> [[TMP30]], zeroinitializer
; CHECK-NEXT: [[TMP32:%.*]] = call <12 x i1> @llvm.vector.insert.v12i1.v2i1(<12 x i1> poison, <2 x i1> zeroinitializer, i64 0)
; CHECK-NEXT: [[TMP33:%.*]] = shufflevector <12 x i1> [[TMP32]], <12 x i1> <i1 poison, i1 false, i1 false, i1 poison, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false>, <12 x i32> <i32 0, i32 13, i32 14, i32 1, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
; CHECK-NEXT: [[TMP34:%.*]] = shufflevector <12 x i1> [[TMP33]], <12 x i1> poison, <16 x i32> <i32 0, i32 1, i32 0, i32 0, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 0, i32 10, i32 11, i32 0>
; CHECK-NEXT: [[TMP35:%.*]] = select <16 x i1> [[TMP34]], <16 x float> zeroinitializer, <16 x float> [[TMP31]]
; CHECK-NEXT: [[TMP36:%.*]] = bitcast <16 x float> [[TMP35]] to <16 x i32>
; CHECK-NEXT: [[TMP37:%.*]] = and <16 x i32> [[TMP36]], zeroinitializer
; CHECK-NEXT: [[TMP38:%.*]] = bitcast <16 x i32> [[TMP37]] to <16 x float>
; CHECK-NEXT: [[TMP39:%.*]] = call <16 x float> @llvm.vector.insert.v16f32.v2f32(<16 x float> <float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float poison, float poison>, <2 x float> [[TMP6]], i64 14)
; CHECK-NEXT: [[TMP40:%.*]] = call <16 x float> @llvm.fmuladd.v16f32(<16 x float> zeroinitializer, <16 x float> [[TMP38]], <16 x float> [[TMP39]])
; CHECK-NEXT: [[TMP41:%.*]] = extractelement <16 x float> [[TMP29]], i32 0
; CHECK-NEXT: [[TMP42:%.*]] = fcmp olt float [[TMP41]], 0.000000e+00
; CHECK-NEXT: [[TMP43:%.*]] = extractelement <16 x float> [[TMP29]], i32 14
; CHECK-NEXT: [[TMP44:%.*]] = fcmp ogt float [[TMP43]], 0.000000e+00
; CHECK-NEXT: [[TMP45:%.*]] = fcmp olt float [[TMP43]], 0.000000e+00
; CHECK-NEXT: [[TMP46:%.*]] = extractelement <16 x float> [[TMP29]], i32 13
; CHECK-NEXT: [[TMP47:%.*]] = fcmp ogt float [[TMP46]], 0.000000e+00
; CHECK-NEXT: [[TMP48:%.*]] = fcmp olt float [[TMP46]], 0.000000e+00
; CHECK-NEXT: [[TMP49:%.*]] = fcmp olt float [[TMP41]], 0.000000e+00
; CHECK-NEXT: [[TMP50:%.*]] = extractelement <16 x float> [[TMP29]], i32 1
; CHECK-NEXT: [[TMP51:%.*]] = fcmp ogt float [[TMP50]], 0.000000e+00
; CHECK-NEXT: [[TMP52:%.*]] = fcmp oeq <16 x float> [[TMP40]], zeroinitializer
; CHECK-NEXT: ret <16 x half> zeroinitializer
;
%4 = bitcast float 0.000000e+00 to i32
%5 = fcmp olt float 0.000000e+00, 0.000000e+00
%6 = icmp ult i32 %4, 0
%7 = select i1 %6, i32 0, i32 0
%8 = sitofp i32 %7 to float
%9 = tail call float @llvm.fmuladd.f32(float %8, float 0.000000e+00, float 0.000000e+00)
%10 = fadd float %9, 0.000000e+00
%11 = select i1 %5, float 0.000000e+00, float %10
%12 = bitcast float %11 to i32
%13 = and i32 %12, 0
%14 = bitcast i32 %13 to float
%15 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00)
%16 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %14, float %15)
%17 = fcmp oeq float %16, 0.000000e+00
%18 = fcmp olt float %8, 0.000000e+00
%19 = icmp ugt i32 %2, 0
%20 = bitcast float 0.000000e+00 to i32
%21 = icmp eq i32 %0, %0
%22 = icmp ult i32 %20, 0
%23 = select i1 %22, i32 0, i32 0
%24 = sitofp i32 %23 to float
%25 = tail call float @llvm.fmuladd.f32(float %24, float 0.000000e+00, float 0.000000e+00)
%26 = fadd float %25, 0.000000e+00
%27 = select i1 false, float 0.000000e+00, float %26
%28 = bitcast float %27 to i32
%29 = and i32 %28, 0
%30 = bitcast i32 %29 to float
%31 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00)
%32 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %30, float %31)
%33 = fcmp ogt float %24, 0.000000e+00
%34 = fcmp oeq float %32, 0.000000e+00
%35 = fcmp ogt float %31, 0.000000e+00
%36 = fcmp olt float %24, 0.000000e+00
%37 = bitcast float %1 to i32
%38 = icmp ult i32 %37, 0
%39 = select i1 %38, i32 0, i32 0
%40 = sitofp i32 %39 to float
%41 = tail call float @llvm.fmuladd.f32(float %40, float 0.000000e+00, float 0.000000e+00)
%42 = fadd float %41, 0.000000e+00
%43 = select i1 false, float 0.000000e+00, float %42
%44 = bitcast float %43 to i32
%45 = and i32 %44, 0
%46 = bitcast i32 %45 to float
%47 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %46, float 0.000000e+00)
%48 = fadd float 0.000000e+00, 0.000000e+00
%49 = fcmp ogt float %40, 0.000000e+00
%50 = fcmp oeq float %47, 0.000000e+00
%51 = fcmp ogt float %48, 0.000000e+00
%52 = fcmp olt float %40, 0.000000e+00
%53 = icmp eq i32 0, 0
%54 = bitcast float 0.000000e+00 to i32
%55 = icmp eq i32 0, 0
%56 = icmp ult i32 %54, 0
%57 = select i1 %56, i32 0, i32 0
%58 = sitofp i32 %57 to float
%59 = tail call float @llvm.fmuladd.f32(float %58, float 0.000000e+00, float 0.000000e+00)
%60 = fadd float %59, 0.000000e+00
%61 = select i1 %5, float 0.000000e+00, float %60
%62 = bitcast float %61 to i32
%63 = and i32 %62, 0
%64 = bitcast i32 %63 to float
%65 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %64, float 0.000000e+00)
%66 = fcmp oeq float %65, 0.000000e+00
%67 = bitcast float 0.000000e+00 to i32
%68 = icmp eq i32 %0, 0
%69 = icmp ult i32 %67, 0
%70 = select i1 %69, i32 0, i32 0
%71 = sitofp i32 %70 to float
%72 = tail call float @llvm.fmuladd.f32(float %71, float 0.000000e+00, float 0.000000e+00)
%73 = fadd float %72, 0.000000e+00
%74 = select i1 false, float 0.000000e+00, float %73
%75 = bitcast float %74 to i32
%76 = and i32 %75, 0
%77 = bitcast i32 %76 to float
%78 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %77, float 0.000000e+00)
%79 = fadd float 0.000000e+00, 0.000000e+00
%80 = fcmp oeq float %78, 0.000000e+00
%81 = fcmp ogt float %79, 0.000000e+00
%82 = icmp eq i32 %0, 0
%83 = bitcast float 0.000000e+00 to i32
%84 = icmp eq i32 %83, 0
%85 = icmp ult i32 %83, 0
%86 = select i1 %85, i32 0, i32 0
%87 = sitofp i32 %86 to float
%88 = tail call float @llvm.fmuladd.f32(float %87, float 0.000000e+00, float 0.000000e+00)
%89 = fadd float %88, 0.000000e+00
%90 = select i1 false, float 0.000000e+00, float %89
%91 = bitcast float %90 to i32
%92 = and i32 %91, 0
%93 = bitcast i32 %92 to float
%94 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %93, float 0.000000e+00)
%95 = fcmp oeq float %94, 0.000000e+00
%96 = bitcast float 0.000000e+00 to i32
%97 = bitcast float 0.000000e+00 to i32
%98 = icmp ult i32 %97, 0
%99 = select i1 %98, i32 0, i32 0
%100 = sitofp i32 %99 to float
%101 = tail call float @llvm.fmuladd.f32(float %100, float 0.000000e+00, float 0.000000e+00)
%102 = fadd float %101, 0.000000e+00
%103 = select i1 false, float 0.000000e+00, float %102
%104 = bitcast float %103 to i32
%105 = and i32 %104, 0
%106 = bitcast i32 %105 to float
%107 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %106, float 0.000000e+00)
%108 = fcmp oeq float %107, 0.000000e+00
%109 = icmp eq i32 %96, 0
%110 = icmp eq i32 %0, 0
%111 = icmp ult i32 0, 0
%112 = bitcast float 0.000000e+00 to i32
%113 = icmp ult i32 %112, 0
%114 = select i1 %113, i32 0, i32 0
%115 = sitofp i32 %114 to float
%116 = tail call float @llvm.fmuladd.f32(float %115, float 0.000000e+00, float 0.000000e+00)
%117 = fadd float %116, 0.000000e+00
%118 = select i1 false, float 0.000000e+00, float %117
%119 = bitcast float %118 to i32
%120 = and i32 %119, 0
%121 = bitcast i32 %120 to float
%122 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %121, float 0.000000e+00)
%123 = fadd float 0.000000e+00, 0.000000e+00
%124 = fcmp oeq float %122, 0.000000e+00
%125 = fcmp ogt float %123, 0.000000e+00
%126 = icmp ult i32 0, 0
%127 = bitcast float 0.000000e+00 to i32
%128 = icmp ult i32 %127, 0
%129 = select i1 %128, i32 0, i32 0
%130 = sitofp i32 %129 to float
%131 = tail call float @llvm.fmuladd.f32(float %130, float 0.000000e+00, float 0.000000e+00)
%132 = fadd float %131, 0.000000e+00
%133 = select i1 false, float 0.000000e+00, float %132
%134 = bitcast float %133 to i32
%135 = and i32 %134, 0
%136 = bitcast i32 %135 to float
%137 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %136, float 0.000000e+00)
%138 = fcmp oeq float %137, 0.000000e+00
%139 = icmp ult i32 0, 0
%140 = bitcast float 0.000000e+00 to i32
%141 = icmp eq i32 0, 0
%142 = icmp ult i32 %140, 0
%143 = select i1 %142, i32 0, i32 0
%144 = sitofp i32 %143 to float
%145 = tail call float @llvm.fmuladd.f32(float %144, float 0.000000e+00, float 0.000000e+00)
%146 = fadd float %145, 0.000000e+00
%147 = select i1 false, float 0.000000e+00, float %146
%148 = bitcast float %147 to i32
%149 = and i32 %148, 0
%150 = bitcast i32 %149 to float
%151 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %150, float 0.000000e+00)
%152 = fcmp oeq float %151, 0.000000e+00
%153 = fcmp olt float 0.000000e+00, 0.000000e+00
%154 = select i1 %153, float 0.000000e+00, float %10
%155 = bitcast float %154 to i32
%156 = and i32 %155, 0
%157 = bitcast i32 %156 to float
%158 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %157, float 0.000000e+00)
%159 = fcmp oeq float %158, 0.000000e+00
%160 = bitcast float 0.000000e+00 to i32
%161 = icmp eq i32 %160, 0
%162 = icmp ult i32 %160, 0
%163 = select i1 %162, i32 0, i32 0
%164 = sitofp i32 %163 to float
%165 = tail call float @llvm.fmuladd.f32(float %164, float 0.000000e+00, float 0.000000e+00)
%166 = fadd float %165, 0.000000e+00
%167 = select i1 false, float 0.000000e+00, float %166
%168 = bitcast float %167 to i32
%169 = and i32 %168, 0
%170 = bitcast i32 %169 to float
%171 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %170, float 0.000000e+00)
%172 = fcmp oeq float %171, 0.000000e+00
%173 = tail call float @llvm.fmuladd.f32(float %8, float 0.000000e+00, float 0.000000e+00)
%174 = fadd float %173, 0.000000e+00
%175 = select i1 %5, float 0.000000e+00, float %174
%176 = bitcast float %175 to i32
%177 = and i32 %176, 0
%178 = bitcast i32 %177 to float
%179 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %178, float 0.000000e+00)
%180 = fadd float 0.000000e+00, 0.000000e+00
%181 = fcmp oeq float %179, 0.000000e+00
%182 = fcmp ogt float %180, 0.000000e+00
%183 = fcmp olt float %8, 0.000000e+00
%184 = bitcast float 0.000000e+00 to i32
%185 = icmp eq i32 %0, %0
%186 = icmp ult i32 %184, 0
%187 = select i1 %186, i32 0, i32 0
%188 = sitofp i32 %187 to float
%189 = tail call float @llvm.fmuladd.f32(float %188, float 0.000000e+00, float 0.000000e+00)
%190 = fadd float %189, 0.000000e+00
%191 = select i1 %5, float 0.000000e+00, float %190
%192 = bitcast float %191 to i32
%193 = and i32 %192, 0
%194 = bitcast i32 %193 to float
%195 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %194, float 0.000000e+00)
%196 = fcmp oeq float %195, 0.000000e+00
%197 = bitcast float 0.000000e+00 to i32
%198 = icmp eq i32 %197, 0
%199 = icmp ult i32 %197, 0
%200 = select i1 %199, i32 0, i32 0
%201 = sitofp i32 %200 to float
%202 = tail call float @llvm.fmuladd.f32(float %201, float 0.000000e+00, float 0.000000e+00)
%203 = fadd float %202, 0.000000e+00
%204 = select i1 false, float 0.000000e+00, float %203
%205 = bitcast float %204 to i32
%206 = and i32 %205, 0
%207 = bitcast i32 %206 to float
%208 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %207, float 0.000000e+00)
%209 = fcmp oeq float %208, 0.000000e+00
%210 = fcmp ogt float %201, 0.000000e+00
%211 = tail call float @llvm.fmuladd.f32(float 0.000000e+00, float %14, float 0.000000e+00)
%212 = fcmp oeq float %211, 0.000000e+00
ret <16 x half> zeroinitializer
}
|
