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
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
|
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
// RUN: %clang_cc1 -triple aarch64 -target-feature +neon -target-feature +bf16 \
// RUN: -disable-O0-optnone -emit-llvm %s -o - | opt -S -passes=mem2reg,sroa | FileCheck %s
// REQUIRES: aarch64-registered-target || arm-registered-target
#include <arm_neon.h>
// CHECK-LABEL: @test_vbfdot_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <2 x float> [[R:%.*]] to <2 x i32>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <4 x bfloat> [[A:%.*]] to <4 x i16>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <4 x bfloat> [[B:%.*]] to <4 x i16>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <2 x i32> [[TMP0]] to <8 x i8>
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <4 x i16> [[TMP1]] to <8 x i8>
// CHECK-NEXT: [[TMP5:%.*]] = bitcast <4 x i16> [[TMP2]] to <8 x i8>
// CHECK-NEXT: [[VBFDOT_I:%.*]] = bitcast <8 x i8> [[TMP3]] to <2 x float>
// CHECK-NEXT: [[VBFDOT1_I:%.*]] = bitcast <8 x i8> [[TMP4]] to <4 x bfloat>
// CHECK-NEXT: [[VBFDOT2_I:%.*]] = bitcast <8 x i8> [[TMP5]] to <4 x bfloat>
// CHECK-NEXT: [[VBFDOT3_I:%.*]] = call <2 x float> @llvm.aarch64.neon.bfdot.v2f32.v4bf16(<2 x float> [[VBFDOT_I]], <4 x bfloat> [[VBFDOT1_I]], <4 x bfloat> [[VBFDOT2_I]])
// CHECK-NEXT: ret <2 x float> [[VBFDOT3_I]]
//
float32x2_t test_vbfdot_f32(float32x2_t r, bfloat16x4_t a, bfloat16x4_t b) {
return vbfdot_f32(r, a, b);
}
// CHECK-LABEL: @test_vbfdotq_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[R:%.*]] to <4 x i32>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <8 x i16>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[B:%.*]] to <8 x i16>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x i32> [[TMP0]] to <16 x i8>
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <8 x i16> [[TMP1]] to <16 x i8>
// CHECK-NEXT: [[TMP5:%.*]] = bitcast <8 x i16> [[TMP2]] to <16 x i8>
// CHECK-NEXT: [[VBFDOT_I:%.*]] = bitcast <16 x i8> [[TMP3]] to <4 x float>
// CHECK-NEXT: [[VBFDOT1_I:%.*]] = bitcast <16 x i8> [[TMP4]] to <8 x bfloat>
// CHECK-NEXT: [[VBFDOT2_I:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x bfloat>
// CHECK-NEXT: [[VBFDOT3_I:%.*]] = call <4 x float> @llvm.aarch64.neon.bfdot.v4f32.v8bf16(<4 x float> [[VBFDOT_I]], <8 x bfloat> [[VBFDOT1_I]], <8 x bfloat> [[VBFDOT2_I]])
// CHECK-NEXT: ret <4 x float> [[VBFDOT3_I]]
//
float32x4_t test_vbfdotq_f32(float32x4_t r, bfloat16x8_t a, bfloat16x8_t b){
return vbfdotq_f32(r, a, b);
}
// CHECK-LABEL: @test_vbfdot_lane_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x bfloat> [[B:%.*]] to <2 x i32>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i32> [[TMP0]] to <2 x float>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <2 x float> [[TMP1]] to <2 x i32>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <2 x i32> [[TMP2]] to <8 x i8>
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <8 x i8> [[TMP3]] to <2 x float>
// CHECK-NEXT: [[LANE:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> [[TMP4]], <2 x i32> zeroinitializer
// CHECK-NEXT: [[TMP5:%.*]] = bitcast <2 x float> [[LANE]] to <2 x i32>
// CHECK-NEXT: [[TMP6:%.*]] = bitcast <2 x i32> [[TMP5]] to <4 x bfloat>
// CHECK-NEXT: [[TMP7:%.*]] = bitcast <2 x float> [[R:%.*]] to <2 x i32>
// CHECK-NEXT: [[TMP8:%.*]] = bitcast <4 x bfloat> [[A:%.*]] to <4 x i16>
// CHECK-NEXT: [[TMP9:%.*]] = bitcast <4 x bfloat> [[TMP6]] to <4 x i16>
// CHECK-NEXT: [[TMP10:%.*]] = bitcast <2 x i32> [[TMP7]] to <8 x i8>
// CHECK-NEXT: [[TMP11:%.*]] = bitcast <4 x i16> [[TMP8]] to <8 x i8>
// CHECK-NEXT: [[TMP12:%.*]] = bitcast <4 x i16> [[TMP9]] to <8 x i8>
// CHECK-NEXT: [[VBFDOT_I:%.*]] = bitcast <8 x i8> [[TMP10]] to <2 x float>
// CHECK-NEXT: [[VBFDOT1_I:%.*]] = bitcast <8 x i8> [[TMP11]] to <4 x bfloat>
// CHECK-NEXT: [[VBFDOT2_I:%.*]] = bitcast <8 x i8> [[TMP12]] to <4 x bfloat>
// CHECK-NEXT: [[VBFDOT3_I:%.*]] = call <2 x float> @llvm.aarch64.neon.bfdot.v2f32.v4bf16(<2 x float> [[VBFDOT_I]], <4 x bfloat> [[VBFDOT1_I]], <4 x bfloat> [[VBFDOT2_I]])
// CHECK-NEXT: ret <2 x float> [[VBFDOT3_I]]
//
float32x2_t test_vbfdot_lane_f32(float32x2_t r, bfloat16x4_t a, bfloat16x4_t b){
return vbfdot_lane_f32(r, a, b, 0);
}
// CHECK-LABEL: @test_vbfdotq_laneq_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <8 x bfloat> [[B:%.*]] to <4 x i32>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <4 x i32> [[TMP0]] to <4 x float>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <4 x float> [[TMP1]] to <4 x i32>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x i32> [[TMP2]] to <16 x i8>
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <16 x i8> [[TMP3]] to <4 x float>
// CHECK-NEXT: [[LANE:%.*]] = shufflevector <4 x float> [[TMP4]], <4 x float> [[TMP4]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
// CHECK-NEXT: [[TMP5:%.*]] = bitcast <4 x float> [[LANE]] to <4 x i32>
// CHECK-NEXT: [[TMP6:%.*]] = bitcast <4 x i32> [[TMP5]] to <8 x bfloat>
// CHECK-NEXT: [[TMP7:%.*]] = bitcast <4 x float> [[R:%.*]] to <4 x i32>
// CHECK-NEXT: [[TMP8:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <8 x i16>
// CHECK-NEXT: [[TMP9:%.*]] = bitcast <8 x bfloat> [[TMP6]] to <8 x i16>
// CHECK-NEXT: [[TMP10:%.*]] = bitcast <4 x i32> [[TMP7]] to <16 x i8>
// CHECK-NEXT: [[TMP11:%.*]] = bitcast <8 x i16> [[TMP8]] to <16 x i8>
// CHECK-NEXT: [[TMP12:%.*]] = bitcast <8 x i16> [[TMP9]] to <16 x i8>
// CHECK-NEXT: [[VBFDOT_I:%.*]] = bitcast <16 x i8> [[TMP10]] to <4 x float>
// CHECK-NEXT: [[VBFDOT1_I:%.*]] = bitcast <16 x i8> [[TMP11]] to <8 x bfloat>
// CHECK-NEXT: [[VBFDOT2_I:%.*]] = bitcast <16 x i8> [[TMP12]] to <8 x bfloat>
// CHECK-NEXT: [[VBFDOT3_I:%.*]] = call <4 x float> @llvm.aarch64.neon.bfdot.v4f32.v8bf16(<4 x float> [[VBFDOT_I]], <8 x bfloat> [[VBFDOT1_I]], <8 x bfloat> [[VBFDOT2_I]])
// CHECK-NEXT: ret <4 x float> [[VBFDOT3_I]]
//
float32x4_t test_vbfdotq_laneq_f32(float32x4_t r, bfloat16x8_t a, bfloat16x8_t b) {
return vbfdotq_laneq_f32(r, a, b, 3);
}
// CHECK-LABEL: @test_vbfdot_laneq_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <8 x bfloat> [[B:%.*]] to <4 x i32>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <4 x i32> [[TMP0]] to <4 x float>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <4 x float> [[TMP1]] to <4 x i32>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x i32> [[TMP2]] to <16 x i8>
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <16 x i8> [[TMP3]] to <4 x float>
// CHECK-NEXT: [[LANE:%.*]] = shufflevector <4 x float> [[TMP4]], <4 x float> [[TMP4]], <2 x i32> <i32 3, i32 3>
// CHECK-NEXT: [[TMP5:%.*]] = bitcast <2 x float> [[LANE]] to <2 x i32>
// CHECK-NEXT: [[TMP6:%.*]] = bitcast <2 x i32> [[TMP5]] to <4 x bfloat>
// CHECK-NEXT: [[TMP7:%.*]] = bitcast <2 x float> [[R:%.*]] to <2 x i32>
// CHECK-NEXT: [[TMP8:%.*]] = bitcast <4 x bfloat> [[A:%.*]] to <4 x i16>
// CHECK-NEXT: [[TMP9:%.*]] = bitcast <4 x bfloat> [[TMP6]] to <4 x i16>
// CHECK-NEXT: [[TMP10:%.*]] = bitcast <2 x i32> [[TMP7]] to <8 x i8>
// CHECK-NEXT: [[TMP11:%.*]] = bitcast <4 x i16> [[TMP8]] to <8 x i8>
// CHECK-NEXT: [[TMP12:%.*]] = bitcast <4 x i16> [[TMP9]] to <8 x i8>
// CHECK-NEXT: [[VBFDOT_I:%.*]] = bitcast <8 x i8> [[TMP10]] to <2 x float>
// CHECK-NEXT: [[VBFDOT1_I:%.*]] = bitcast <8 x i8> [[TMP11]] to <4 x bfloat>
// CHECK-NEXT: [[VBFDOT2_I:%.*]] = bitcast <8 x i8> [[TMP12]] to <4 x bfloat>
// CHECK-NEXT: [[VBFDOT3_I:%.*]] = call <2 x float> @llvm.aarch64.neon.bfdot.v2f32.v4bf16(<2 x float> [[VBFDOT_I]], <4 x bfloat> [[VBFDOT1_I]], <4 x bfloat> [[VBFDOT2_I]])
// CHECK-NEXT: ret <2 x float> [[VBFDOT3_I]]
//
float32x2_t test_vbfdot_laneq_f32(float32x2_t r, bfloat16x4_t a, bfloat16x8_t b) {
return vbfdot_laneq_f32(r, a, b, 3);
}
// CHECK-LABEL: @test_vbfdotq_lane_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x bfloat> [[B:%.*]] to <2 x i32>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i32> [[TMP0]] to <2 x float>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <2 x float> [[TMP1]] to <2 x i32>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <2 x i32> [[TMP2]] to <8 x i8>
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <8 x i8> [[TMP3]] to <2 x float>
// CHECK-NEXT: [[LANE:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> [[TMP4]], <4 x i32> zeroinitializer
// CHECK-NEXT: [[TMP5:%.*]] = bitcast <4 x float> [[LANE]] to <4 x i32>
// CHECK-NEXT: [[TMP6:%.*]] = bitcast <4 x i32> [[TMP5]] to <8 x bfloat>
// CHECK-NEXT: [[TMP7:%.*]] = bitcast <4 x float> [[R:%.*]] to <4 x i32>
// CHECK-NEXT: [[TMP8:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <8 x i16>
// CHECK-NEXT: [[TMP9:%.*]] = bitcast <8 x bfloat> [[TMP6]] to <8 x i16>
// CHECK-NEXT: [[TMP10:%.*]] = bitcast <4 x i32> [[TMP7]] to <16 x i8>
// CHECK-NEXT: [[TMP11:%.*]] = bitcast <8 x i16> [[TMP8]] to <16 x i8>
// CHECK-NEXT: [[TMP12:%.*]] = bitcast <8 x i16> [[TMP9]] to <16 x i8>
// CHECK-NEXT: [[VBFDOT_I:%.*]] = bitcast <16 x i8> [[TMP10]] to <4 x float>
// CHECK-NEXT: [[VBFDOT1_I:%.*]] = bitcast <16 x i8> [[TMP11]] to <8 x bfloat>
// CHECK-NEXT: [[VBFDOT2_I:%.*]] = bitcast <16 x i8> [[TMP12]] to <8 x bfloat>
// CHECK-NEXT: [[VBFDOT3_I:%.*]] = call <4 x float> @llvm.aarch64.neon.bfdot.v4f32.v8bf16(<4 x float> [[VBFDOT_I]], <8 x bfloat> [[VBFDOT1_I]], <8 x bfloat> [[VBFDOT2_I]])
// CHECK-NEXT: ret <4 x float> [[VBFDOT3_I]]
//
float32x4_t test_vbfdotq_lane_f32(float32x4_t r, bfloat16x8_t a, bfloat16x4_t b) {
return vbfdotq_lane_f32(r, a, b, 0);
}
// CHECK-LABEL: @test_vbfmmlaq_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[R:%.*]] to <4 x i32>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <8 x i16>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[B:%.*]] to <8 x i16>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x i32> [[TMP0]] to <16 x i8>
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <8 x i16> [[TMP1]] to <16 x i8>
// CHECK-NEXT: [[TMP5:%.*]] = bitcast <8 x i16> [[TMP2]] to <16 x i8>
// CHECK-NEXT: [[VBFMMLAQ_F32_I:%.*]] = bitcast <16 x i8> [[TMP3]] to <4 x float>
// CHECK-NEXT: [[VBFMMLAQ_F321_I:%.*]] = bitcast <16 x i8> [[TMP4]] to <8 x bfloat>
// CHECK-NEXT: [[VBFMMLAQ_F322_I:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x bfloat>
// CHECK-NEXT: [[VBFMMLAQ_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.bfmmla(<4 x float> [[VBFMMLAQ_F32_I]], <8 x bfloat> [[VBFMMLAQ_F321_I]], <8 x bfloat> [[VBFMMLAQ_F322_I]])
// CHECK-NEXT: [[VBFMMLAQ_F324_I:%.*]] = bitcast <4 x float> [[VBFMMLAQ_F323_I]] to <16 x i8>
// CHECK-NEXT: [[TMP6:%.*]] = bitcast <16 x i8> [[VBFMMLAQ_F324_I]] to <4 x i32>
// CHECK-NEXT: [[TMP7:%.*]] = bitcast <4 x i32> [[TMP6]] to <4 x float>
// CHECK-NEXT: ret <4 x float> [[TMP7]]
//
float32x4_t test_vbfmmlaq_f32(float32x4_t r, bfloat16x8_t a, bfloat16x8_t b) {
return vbfmmlaq_f32(r, a, b);
}
// CHECK-LABEL: @test_vbfmlalbq_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[R:%.*]] to <4 x i32>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <8 x i16>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[B:%.*]] to <8 x i16>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x i32> [[TMP0]] to <16 x i8>
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <8 x i16> [[TMP1]] to <16 x i8>
// CHECK-NEXT: [[TMP5:%.*]] = bitcast <8 x i16> [[TMP2]] to <16 x i8>
// CHECK-NEXT: [[VBFMLALBQ_F32_I:%.*]] = bitcast <16 x i8> [[TMP3]] to <4 x float>
// CHECK-NEXT: [[VBFMLALBQ_F321_I:%.*]] = bitcast <16 x i8> [[TMP4]] to <8 x bfloat>
// CHECK-NEXT: [[VBFMLALBQ_F322_I:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x bfloat>
// CHECK-NEXT: [[VBFMLALBQ_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.bfmlalb(<4 x float> [[VBFMLALBQ_F32_I]], <8 x bfloat> [[VBFMLALBQ_F321_I]], <8 x bfloat> [[VBFMLALBQ_F322_I]])
// CHECK-NEXT: [[VBFMLALBQ_F324_I:%.*]] = bitcast <4 x float> [[VBFMLALBQ_F323_I]] to <16 x i8>
// CHECK-NEXT: [[TMP6:%.*]] = bitcast <16 x i8> [[VBFMLALBQ_F324_I]] to <4 x i32>
// CHECK-NEXT: [[TMP7:%.*]] = bitcast <4 x i32> [[TMP6]] to <4 x float>
// CHECK-NEXT: ret <4 x float> [[TMP7]]
//
float32x4_t test_vbfmlalbq_f32(float32x4_t r, bfloat16x8_t a, bfloat16x8_t b) {
return vbfmlalbq_f32(r, a, b);
}
// CHECK-LABEL: @test_vbfmlaltq_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[R:%.*]] to <4 x i32>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <8 x i16>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[B:%.*]] to <8 x i16>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x i32> [[TMP0]] to <16 x i8>
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <8 x i16> [[TMP1]] to <16 x i8>
// CHECK-NEXT: [[TMP5:%.*]] = bitcast <8 x i16> [[TMP2]] to <16 x i8>
// CHECK-NEXT: [[VBFMLALTQ_F32_I:%.*]] = bitcast <16 x i8> [[TMP3]] to <4 x float>
// CHECK-NEXT: [[VBFMLALTQ_F321_I:%.*]] = bitcast <16 x i8> [[TMP4]] to <8 x bfloat>
// CHECK-NEXT: [[VBFMLALTQ_F322_I:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x bfloat>
// CHECK-NEXT: [[VBFMLALTQ_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.bfmlalt(<4 x float> [[VBFMLALTQ_F32_I]], <8 x bfloat> [[VBFMLALTQ_F321_I]], <8 x bfloat> [[VBFMLALTQ_F322_I]])
// CHECK-NEXT: [[VBFMLALTQ_F324_I:%.*]] = bitcast <4 x float> [[VBFMLALTQ_F323_I]] to <16 x i8>
// CHECK-NEXT: [[TMP6:%.*]] = bitcast <16 x i8> [[VBFMLALTQ_F324_I]] to <4 x i32>
// CHECK-NEXT: [[TMP7:%.*]] = bitcast <4 x i32> [[TMP6]] to <4 x float>
// CHECK-NEXT: ret <4 x float> [[TMP7]]
//
float32x4_t test_vbfmlaltq_f32(float32x4_t r, bfloat16x8_t a, bfloat16x8_t b) {
return vbfmlaltq_f32(r, a, b);
}
// CHECK-LABEL: @test_vbfmlalbq_lane_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <4 x bfloat> [[B:%.*]], i32 0
// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <8 x bfloat> poison, bfloat [[VGET_LANE]], i32 0
// CHECK-NEXT: [[VGET_LANE4:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT6:%.*]] = insertelement <8 x bfloat> [[VECINIT]], bfloat [[VGET_LANE4]], i32 1
// CHECK-NEXT: [[VGET_LANE10:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT12:%.*]] = insertelement <8 x bfloat> [[VECINIT6]], bfloat [[VGET_LANE10]], i32 2
// CHECK-NEXT: [[VGET_LANE16:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT18:%.*]] = insertelement <8 x bfloat> [[VECINIT12]], bfloat [[VGET_LANE16]], i32 3
// CHECK-NEXT: [[VGET_LANE22:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT24:%.*]] = insertelement <8 x bfloat> [[VECINIT18]], bfloat [[VGET_LANE22]], i32 4
// CHECK-NEXT: [[VGET_LANE28:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT30:%.*]] = insertelement <8 x bfloat> [[VECINIT24]], bfloat [[VGET_LANE28]], i32 5
// CHECK-NEXT: [[VGET_LANE34:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT36:%.*]] = insertelement <8 x bfloat> [[VECINIT30]], bfloat [[VGET_LANE34]], i32 6
// CHECK-NEXT: [[VGET_LANE40:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT42:%.*]] = insertelement <8 x bfloat> [[VECINIT36]], bfloat [[VGET_LANE40]], i32 7
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[R:%.*]] to <4 x i32>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <8 x i16>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[VECINIT42]] to <8 x i16>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x i32> [[TMP0]] to <16 x i8>
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <8 x i16> [[TMP1]] to <16 x i8>
// CHECK-NEXT: [[TMP5:%.*]] = bitcast <8 x i16> [[TMP2]] to <16 x i8>
// CHECK-NEXT: [[VBFMLALBQ_F32_I:%.*]] = bitcast <16 x i8> [[TMP3]] to <4 x float>
// CHECK-NEXT: [[VBFMLALBQ_F321_I:%.*]] = bitcast <16 x i8> [[TMP4]] to <8 x bfloat>
// CHECK-NEXT: [[VBFMLALBQ_F322_I:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x bfloat>
// CHECK-NEXT: [[VBFMLALBQ_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.bfmlalb(<4 x float> [[VBFMLALBQ_F32_I]], <8 x bfloat> [[VBFMLALBQ_F321_I]], <8 x bfloat> [[VBFMLALBQ_F322_I]])
// CHECK-NEXT: [[VBFMLALBQ_F324_I:%.*]] = bitcast <4 x float> [[VBFMLALBQ_F323_I]] to <16 x i8>
// CHECK-NEXT: [[TMP6:%.*]] = bitcast <16 x i8> [[VBFMLALBQ_F324_I]] to <4 x i32>
// CHECK-NEXT: [[TMP7:%.*]] = bitcast <4 x i32> [[TMP6]] to <4 x float>
// CHECK-NEXT: ret <4 x float> [[TMP7]]
//
float32x4_t test_vbfmlalbq_lane_f32(float32x4_t r, bfloat16x8_t a, bfloat16x4_t b) {
return vbfmlalbq_lane_f32(r, a, b, 0);
}
// CHECK-LABEL: @test_vbfmlalbq_laneq_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <8 x bfloat> [[B:%.*]], i32 3
// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <8 x bfloat> poison, bfloat [[VGETQ_LANE]], i32 0
// CHECK-NEXT: [[VGETQ_LANE4:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT6:%.*]] = insertelement <8 x bfloat> [[VECINIT]], bfloat [[VGETQ_LANE4]], i32 1
// CHECK-NEXT: [[VGETQ_LANE10:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT12:%.*]] = insertelement <8 x bfloat> [[VECINIT6]], bfloat [[VGETQ_LANE10]], i32 2
// CHECK-NEXT: [[VGETQ_LANE16:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT18:%.*]] = insertelement <8 x bfloat> [[VECINIT12]], bfloat [[VGETQ_LANE16]], i32 3
// CHECK-NEXT: [[VGETQ_LANE22:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT24:%.*]] = insertelement <8 x bfloat> [[VECINIT18]], bfloat [[VGETQ_LANE22]], i32 4
// CHECK-NEXT: [[VGETQ_LANE28:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT30:%.*]] = insertelement <8 x bfloat> [[VECINIT24]], bfloat [[VGETQ_LANE28]], i32 5
// CHECK-NEXT: [[VGETQ_LANE34:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT36:%.*]] = insertelement <8 x bfloat> [[VECINIT30]], bfloat [[VGETQ_LANE34]], i32 6
// CHECK-NEXT: [[VGETQ_LANE40:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT42:%.*]] = insertelement <8 x bfloat> [[VECINIT36]], bfloat [[VGETQ_LANE40]], i32 7
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[R:%.*]] to <4 x i32>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <8 x i16>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[VECINIT42]] to <8 x i16>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x i32> [[TMP0]] to <16 x i8>
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <8 x i16> [[TMP1]] to <16 x i8>
// CHECK-NEXT: [[TMP5:%.*]] = bitcast <8 x i16> [[TMP2]] to <16 x i8>
// CHECK-NEXT: [[VBFMLALBQ_F32_I:%.*]] = bitcast <16 x i8> [[TMP3]] to <4 x float>
// CHECK-NEXT: [[VBFMLALBQ_F321_I:%.*]] = bitcast <16 x i8> [[TMP4]] to <8 x bfloat>
// CHECK-NEXT: [[VBFMLALBQ_F322_I:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x bfloat>
// CHECK-NEXT: [[VBFMLALBQ_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.bfmlalb(<4 x float> [[VBFMLALBQ_F32_I]], <8 x bfloat> [[VBFMLALBQ_F321_I]], <8 x bfloat> [[VBFMLALBQ_F322_I]])
// CHECK-NEXT: [[VBFMLALBQ_F324_I:%.*]] = bitcast <4 x float> [[VBFMLALBQ_F323_I]] to <16 x i8>
// CHECK-NEXT: [[TMP6:%.*]] = bitcast <16 x i8> [[VBFMLALBQ_F324_I]] to <4 x i32>
// CHECK-NEXT: [[TMP7:%.*]] = bitcast <4 x i32> [[TMP6]] to <4 x float>
// CHECK-NEXT: ret <4 x float> [[TMP7]]
//
float32x4_t test_vbfmlalbq_laneq_f32(float32x4_t r, bfloat16x8_t a, bfloat16x8_t b) {
return vbfmlalbq_laneq_f32(r, a, b, 3);
}
// CHECK-LABEL: @test_vbfmlaltq_lane_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <4 x bfloat> [[B:%.*]], i32 0
// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <8 x bfloat> poison, bfloat [[VGET_LANE]], i32 0
// CHECK-NEXT: [[VGET_LANE4:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT6:%.*]] = insertelement <8 x bfloat> [[VECINIT]], bfloat [[VGET_LANE4]], i32 1
// CHECK-NEXT: [[VGET_LANE10:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT12:%.*]] = insertelement <8 x bfloat> [[VECINIT6]], bfloat [[VGET_LANE10]], i32 2
// CHECK-NEXT: [[VGET_LANE16:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT18:%.*]] = insertelement <8 x bfloat> [[VECINIT12]], bfloat [[VGET_LANE16]], i32 3
// CHECK-NEXT: [[VGET_LANE22:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT24:%.*]] = insertelement <8 x bfloat> [[VECINIT18]], bfloat [[VGET_LANE22]], i32 4
// CHECK-NEXT: [[VGET_LANE28:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT30:%.*]] = insertelement <8 x bfloat> [[VECINIT24]], bfloat [[VGET_LANE28]], i32 5
// CHECK-NEXT: [[VGET_LANE34:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT36:%.*]] = insertelement <8 x bfloat> [[VECINIT30]], bfloat [[VGET_LANE34]], i32 6
// CHECK-NEXT: [[VGET_LANE40:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT42:%.*]] = insertelement <8 x bfloat> [[VECINIT36]], bfloat [[VGET_LANE40]], i32 7
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[R:%.*]] to <4 x i32>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <8 x i16>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[VECINIT42]] to <8 x i16>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x i32> [[TMP0]] to <16 x i8>
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <8 x i16> [[TMP1]] to <16 x i8>
// CHECK-NEXT: [[TMP5:%.*]] = bitcast <8 x i16> [[TMP2]] to <16 x i8>
// CHECK-NEXT: [[VBFMLALTQ_F32_I:%.*]] = bitcast <16 x i8> [[TMP3]] to <4 x float>
// CHECK-NEXT: [[VBFMLALTQ_F321_I:%.*]] = bitcast <16 x i8> [[TMP4]] to <8 x bfloat>
// CHECK-NEXT: [[VBFMLALTQ_F322_I:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x bfloat>
// CHECK-NEXT: [[VBFMLALTQ_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.bfmlalt(<4 x float> [[VBFMLALTQ_F32_I]], <8 x bfloat> [[VBFMLALTQ_F321_I]], <8 x bfloat> [[VBFMLALTQ_F322_I]])
// CHECK-NEXT: [[VBFMLALTQ_F324_I:%.*]] = bitcast <4 x float> [[VBFMLALTQ_F323_I]] to <16 x i8>
// CHECK-NEXT: [[TMP6:%.*]] = bitcast <16 x i8> [[VBFMLALTQ_F324_I]] to <4 x i32>
// CHECK-NEXT: [[TMP7:%.*]] = bitcast <4 x i32> [[TMP6]] to <4 x float>
// CHECK-NEXT: ret <4 x float> [[TMP7]]
//
float32x4_t test_vbfmlaltq_lane_f32(float32x4_t r, bfloat16x8_t a, bfloat16x4_t b) {
return vbfmlaltq_lane_f32(r, a, b, 0);
}
// CHECK-LABEL: @test_vbfmlaltq_laneq_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <8 x bfloat> [[B:%.*]], i32 3
// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <8 x bfloat> poison, bfloat [[VGETQ_LANE]], i32 0
// CHECK-NEXT: [[VGETQ_LANE4:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT6:%.*]] = insertelement <8 x bfloat> [[VECINIT]], bfloat [[VGETQ_LANE4]], i32 1
// CHECK-NEXT: [[VGETQ_LANE10:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT12:%.*]] = insertelement <8 x bfloat> [[VECINIT6]], bfloat [[VGETQ_LANE10]], i32 2
// CHECK-NEXT: [[VGETQ_LANE16:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT18:%.*]] = insertelement <8 x bfloat> [[VECINIT12]], bfloat [[VGETQ_LANE16]], i32 3
// CHECK-NEXT: [[VGETQ_LANE22:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT24:%.*]] = insertelement <8 x bfloat> [[VECINIT18]], bfloat [[VGETQ_LANE22]], i32 4
// CHECK-NEXT: [[VGETQ_LANE28:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT30:%.*]] = insertelement <8 x bfloat> [[VECINIT24]], bfloat [[VGETQ_LANE28]], i32 5
// CHECK-NEXT: [[VGETQ_LANE34:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT36:%.*]] = insertelement <8 x bfloat> [[VECINIT30]], bfloat [[VGETQ_LANE34]], i32 6
// CHECK-NEXT: [[VGETQ_LANE40:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT42:%.*]] = insertelement <8 x bfloat> [[VECINIT36]], bfloat [[VGETQ_LANE40]], i32 7
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[R:%.*]] to <4 x i32>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <8 x i16>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[VECINIT42]] to <8 x i16>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x i32> [[TMP0]] to <16 x i8>
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <8 x i16> [[TMP1]] to <16 x i8>
// CHECK-NEXT: [[TMP5:%.*]] = bitcast <8 x i16> [[TMP2]] to <16 x i8>
// CHECK-NEXT: [[VBFMLALTQ_F32_I:%.*]] = bitcast <16 x i8> [[TMP3]] to <4 x float>
// CHECK-NEXT: [[VBFMLALTQ_F321_I:%.*]] = bitcast <16 x i8> [[TMP4]] to <8 x bfloat>
// CHECK-NEXT: [[VBFMLALTQ_F322_I:%.*]] = bitcast <16 x i8> [[TMP5]] to <8 x bfloat>
// CHECK-NEXT: [[VBFMLALTQ_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.bfmlalt(<4 x float> [[VBFMLALTQ_F32_I]], <8 x bfloat> [[VBFMLALTQ_F321_I]], <8 x bfloat> [[VBFMLALTQ_F322_I]])
// CHECK-NEXT: [[VBFMLALTQ_F324_I:%.*]] = bitcast <4 x float> [[VBFMLALTQ_F323_I]] to <16 x i8>
// CHECK-NEXT: [[TMP6:%.*]] = bitcast <16 x i8> [[VBFMLALTQ_F324_I]] to <4 x i32>
// CHECK-NEXT: [[TMP7:%.*]] = bitcast <4 x i32> [[TMP6]] to <4 x float>
// CHECK-NEXT: ret <4 x float> [[TMP7]]
//
float32x4_t test_vbfmlaltq_laneq_f32(float32x4_t r, bfloat16x8_t a, bfloat16x8_t b) {
return vbfmlaltq_laneq_f32(r, a, b, 3);
}
|
