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
|
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
// RUN: %clang_cc1 -triple armv8-arm-none-eabi \
// RUN: -target-feature +neon -target-feature +bf16 -mfloat-abi soft \
// RUN: -disable-O0-optnone -S -emit-llvm -fno-legacy-pass-manager -o - %s \
// RUN: | opt -S -mem2reg | FileCheck %s
// RUN: %clang_cc1 -triple armv8-arm-none-eabi \
// RUN: -target-feature +neon -target-feature +bf16 -mfloat-abi hard \
// RUN: -disable-O0-optnone -S -emit-llvm -fno-legacy-pass-manager -o - %s \
// RUN: | opt -S -mem2reg | FileCheck %s
#include <arm_neon.h>
// CHECK-LABEL: @test_vbfdot_f32(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <2 x float> [[R:%.*]] to <8 x i8>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <4 x bfloat> [[A:%.*]] to <8 x i8>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <4 x bfloat> [[B:%.*]] to <8 x i8>
// CHECK-NEXT: [[VBFDOT3_I:%.*]] = call <2 x float> @llvm.arm.neon.bfdot.v2f32.v4bf16(<2 x float> [[R]], <4 x bfloat> [[A]], <4 x bfloat> [[B]]) #[[ATTR3:[0-9]+]]
// 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 <16 x i8>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[B:%.*]] to <16 x i8>
// CHECK-NEXT: [[VBFDOT3_I:%.*]] = call <4 x float> @llvm.arm.neon.bfdot.v4f32.v8bf16(<4 x float> [[R]], <8 x bfloat> [[A]], <8 x bfloat> [[B]]) #[[ATTR3]]
// 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: [[__REINT_144:%.*]] = alloca <4 x bfloat>, align 8
// CHECK-NEXT: [[__REINT1_144:%.*]] = alloca <2 x float>, align 8
// CHECK-NEXT: store <4 x bfloat> [[B:%.*]], <4 x bfloat>* [[__REINT_144]], align 8
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x bfloat>* [[__REINT_144]] to <2 x float>*
// CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, <2 x float>* [[TMP0]], align 8
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <2 x float> [[TMP1]] to <8 x i8>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP2]] to <2 x float>
// CHECK-NEXT: [[LANE:%.*]] = shufflevector <2 x float> [[TMP3]], <2 x float> [[TMP3]], <2 x i32> zeroinitializer
// CHECK-NEXT: store <2 x float> [[LANE]], <2 x float>* [[__REINT1_144]], align 8
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <2 x float>* [[__REINT1_144]] to <4 x bfloat>*
// CHECK-NEXT: [[TMP5:%.*]] = load <4 x bfloat>, <4 x bfloat>* [[TMP4]], align 8
// CHECK-NEXT: [[TMP6:%.*]] = bitcast <2 x float> [[R:%.*]] to <8 x i8>
// CHECK-NEXT: [[TMP7:%.*]] = bitcast <4 x bfloat> [[A:%.*]] to <8 x i8>
// CHECK-NEXT: [[TMP8:%.*]] = bitcast <4 x bfloat> [[TMP5]] to <8 x i8>
// CHECK-NEXT: [[VBFDOT3_I:%.*]] = call <2 x float> @llvm.arm.neon.bfdot.v2f32.v4bf16(<2 x float> [[R]], <4 x bfloat> [[A]], <4 x bfloat> [[TMP5]]) #[[ATTR3]]
// 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: [[__REINT_146:%.*]] = alloca <8 x bfloat>, align 8
// CHECK-NEXT: [[__REINT1_146:%.*]] = alloca <4 x float>, align 8
// CHECK-NEXT: store <8 x bfloat> [[B:%.*]], <8 x bfloat>* [[__REINT_146]], align 8
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <8 x bfloat>* [[__REINT_146]] to <4 x float>*
// CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, <4 x float>* [[TMP0]], align 8
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <4 x float> [[TMP1]] to <16 x i8>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <16 x i8> [[TMP2]] to <4 x float>
// CHECK-NEXT: [[LANE:%.*]] = shufflevector <4 x float> [[TMP3]], <4 x float> [[TMP3]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
// CHECK-NEXT: store <4 x float> [[LANE]], <4 x float>* [[__REINT1_146]], align 8
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <4 x float>* [[__REINT1_146]] to <8 x bfloat>*
// CHECK-NEXT: [[TMP5:%.*]] = load <8 x bfloat>, <8 x bfloat>* [[TMP4]], align 8
// CHECK-NEXT: [[TMP6:%.*]] = bitcast <4 x float> [[R:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP7:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP8:%.*]] = bitcast <8 x bfloat> [[TMP5]] to <16 x i8>
// CHECK-NEXT: [[VBFDOT3_I:%.*]] = call <4 x float> @llvm.arm.neon.bfdot.v4f32.v8bf16(<4 x float> [[R]], <8 x bfloat> [[A]], <8 x bfloat> [[TMP5]]) #[[ATTR3]]
// 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: [[__REINT_148:%.*]] = alloca <8 x bfloat>, align 8
// CHECK-NEXT: [[__REINT1_148:%.*]] = alloca <2 x float>, align 8
// CHECK-NEXT: store <8 x bfloat> [[B:%.*]], <8 x bfloat>* [[__REINT_148]], align 8
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <8 x bfloat>* [[__REINT_148]] to <4 x float>*
// CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, <4 x float>* [[TMP0]], align 8
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <4 x float> [[TMP1]] to <16 x i8>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <16 x i8> [[TMP2]] to <4 x float>
// CHECK-NEXT: [[LANE:%.*]] = shufflevector <4 x float> [[TMP3]], <4 x float> [[TMP3]], <2 x i32> <i32 3, i32 3>
// CHECK-NEXT: store <2 x float> [[LANE]], <2 x float>* [[__REINT1_148]], align 8
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <2 x float>* [[__REINT1_148]] to <4 x bfloat>*
// CHECK-NEXT: [[TMP5:%.*]] = load <4 x bfloat>, <4 x bfloat>* [[TMP4]], align 8
// CHECK-NEXT: [[TMP6:%.*]] = bitcast <2 x float> [[R:%.*]] to <8 x i8>
// CHECK-NEXT: [[TMP7:%.*]] = bitcast <4 x bfloat> [[A:%.*]] to <8 x i8>
// CHECK-NEXT: [[TMP8:%.*]] = bitcast <4 x bfloat> [[TMP5]] to <8 x i8>
// CHECK-NEXT: [[VBFDOT3_I:%.*]] = call <2 x float> @llvm.arm.neon.bfdot.v2f32.v4bf16(<2 x float> [[R]], <4 x bfloat> [[A]], <4 x bfloat> [[TMP5]]) #[[ATTR3]]
// 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: [[__REINT_142:%.*]] = alloca <4 x bfloat>, align 8
// CHECK-NEXT: [[__REINT1_142:%.*]] = alloca <4 x float>, align 8
// CHECK-NEXT: store <4 x bfloat> [[B:%.*]], <4 x bfloat>* [[__REINT_142]], align 8
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x bfloat>* [[__REINT_142]] to <2 x float>*
// CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, <2 x float>* [[TMP0]], align 8
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <2 x float> [[TMP1]] to <8 x i8>
// CHECK-NEXT: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP2]] to <2 x float>
// CHECK-NEXT: [[LANE:%.*]] = shufflevector <2 x float> [[TMP3]], <2 x float> [[TMP3]], <4 x i32> zeroinitializer
// CHECK-NEXT: store <4 x float> [[LANE]], <4 x float>* [[__REINT1_142]], align 8
// CHECK-NEXT: [[TMP4:%.*]] = bitcast <4 x float>* [[__REINT1_142]] to <8 x bfloat>*
// CHECK-NEXT: [[TMP5:%.*]] = load <8 x bfloat>, <8 x bfloat>* [[TMP4]], align 8
// CHECK-NEXT: [[TMP6:%.*]] = bitcast <4 x float> [[R:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP7:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP8:%.*]] = bitcast <8 x bfloat> [[TMP5]] to <16 x i8>
// CHECK-NEXT: [[VBFDOT3_I:%.*]] = call <4 x float> @llvm.arm.neon.bfdot.v4f32.v8bf16(<4 x float> [[R]], <8 x bfloat> [[A]], <8 x bfloat> [[TMP5]]) #[[ATTR3]]
// 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 <16 x i8>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[B:%.*]] to <16 x i8>
// CHECK-NEXT: [[VBFMMLAQ_V3_I:%.*]] = call <4 x float> @llvm.arm.neon.bfmmla(<4 x float> [[R]], <8 x bfloat> [[A]], <8 x bfloat> [[B]]) #[[ATTR3]]
// CHECK-NEXT: [[VBFMMLAQ_V4_I:%.*]] = bitcast <4 x float> [[VBFMMLAQ_V3_I]] to <16 x i8>
// CHECK-NEXT: ret <4 x float> [[VBFMMLAQ_V3_I]]
//
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 <16 x i8>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[B:%.*]] to <16 x i8>
// CHECK-NEXT: [[VBFMLALBQ_V3_I:%.*]] = call <4 x float> @llvm.arm.neon.bfmlalb(<4 x float> [[R]], <8 x bfloat> [[A]], <8 x bfloat> [[B]]) #[[ATTR3]]
// CHECK-NEXT: [[VBFMLALBQ_V4_I:%.*]] = bitcast <4 x float> [[VBFMLALBQ_V3_I]] to <16 x i8>
// CHECK-NEXT: ret <4 x float> [[VBFMLALBQ_V3_I]]
//
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 <16 x i8>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[B:%.*]] to <16 x i8>
// CHECK-NEXT: [[VBFMLALTQ_V3_I:%.*]] = call <4 x float> @llvm.arm.neon.bfmlalt(<4 x float> [[R]], <8 x bfloat> [[A]], <8 x bfloat> [[B]]) #[[ATTR3]]
// CHECK-NEXT: [[VBFMLALTQ_V4_I:%.*]] = bitcast <4 x float> [[VBFMLALTQ_V3_I]] to <16 x i8>
// CHECK-NEXT: ret <4 x float> [[VBFMLALTQ_V3_I]]
//
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> undef, bfloat [[VGET_LANE]], i32 0
// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <8 x bfloat> [[VECINIT]], bfloat [[VGET_LANE3]], i32 1
// CHECK-NEXT: [[VGET_LANE8:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT10:%.*]] = insertelement <8 x bfloat> [[VECINIT5]], bfloat [[VGET_LANE8]], i32 2
// CHECK-NEXT: [[VGET_LANE13:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT15:%.*]] = insertelement <8 x bfloat> [[VECINIT10]], bfloat [[VGET_LANE13]], i32 3
// CHECK-NEXT: [[VGET_LANE18:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT20:%.*]] = insertelement <8 x bfloat> [[VECINIT15]], bfloat [[VGET_LANE18]], i32 4
// CHECK-NEXT: [[VGET_LANE23:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT25:%.*]] = insertelement <8 x bfloat> [[VECINIT20]], bfloat [[VGET_LANE23]], i32 5
// CHECK-NEXT: [[VGET_LANE28:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT30:%.*]] = insertelement <8 x bfloat> [[VECINIT25]], bfloat [[VGET_LANE28]], i32 6
// CHECK-NEXT: [[VGET_LANE33:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT35:%.*]] = insertelement <8 x bfloat> [[VECINIT30]], bfloat [[VGET_LANE33]], i32 7
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[R:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[VECINIT35]] to <16 x i8>
// CHECK-NEXT: [[VBFMLALBQ_V3_I:%.*]] = call <4 x float> @llvm.arm.neon.bfmlalb(<4 x float> [[R]], <8 x bfloat> [[A]], <8 x bfloat> [[VECINIT35]]) #[[ATTR3]]
// CHECK-NEXT: [[VBFMLALBQ_V4_I:%.*]] = bitcast <4 x float> [[VBFMLALBQ_V3_I]] to <16 x i8>
// CHECK-NEXT: ret <4 x float> [[VBFMLALBQ_V3_I]]
//
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: [[VGET_LANE:%.*]] = extractelement <8 x bfloat> [[B:%.*]], i32 3
// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <8 x bfloat> undef, bfloat [[VGET_LANE]], i32 0
// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <8 x bfloat> [[VECINIT]], bfloat [[VGET_LANE3]], i32 1
// CHECK-NEXT: [[VGET_LANE8:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT10:%.*]] = insertelement <8 x bfloat> [[VECINIT5]], bfloat [[VGET_LANE8]], i32 2
// CHECK-NEXT: [[VGET_LANE13:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT15:%.*]] = insertelement <8 x bfloat> [[VECINIT10]], bfloat [[VGET_LANE13]], i32 3
// CHECK-NEXT: [[VGET_LANE18:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT20:%.*]] = insertelement <8 x bfloat> [[VECINIT15]], bfloat [[VGET_LANE18]], i32 4
// CHECK-NEXT: [[VGET_LANE23:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT25:%.*]] = insertelement <8 x bfloat> [[VECINIT20]], bfloat [[VGET_LANE23]], i32 5
// CHECK-NEXT: [[VGET_LANE28:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT30:%.*]] = insertelement <8 x bfloat> [[VECINIT25]], bfloat [[VGET_LANE28]], i32 6
// CHECK-NEXT: [[VGET_LANE33:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT35:%.*]] = insertelement <8 x bfloat> [[VECINIT30]], bfloat [[VGET_LANE33]], i32 7
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[R:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[VECINIT35]] to <16 x i8>
// CHECK-NEXT: [[VBFMLALBQ_V3_I:%.*]] = call <4 x float> @llvm.arm.neon.bfmlalb(<4 x float> [[R]], <8 x bfloat> [[A]], <8 x bfloat> [[VECINIT35]]) #[[ATTR3]]
// CHECK-NEXT: [[VBFMLALBQ_V4_I:%.*]] = bitcast <4 x float> [[VBFMLALBQ_V3_I]] to <16 x i8>
// CHECK-NEXT: ret <4 x float> [[VBFMLALBQ_V3_I]]
//
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> undef, bfloat [[VGET_LANE]], i32 0
// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <8 x bfloat> [[VECINIT]], bfloat [[VGET_LANE3]], i32 1
// CHECK-NEXT: [[VGET_LANE8:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT10:%.*]] = insertelement <8 x bfloat> [[VECINIT5]], bfloat [[VGET_LANE8]], i32 2
// CHECK-NEXT: [[VGET_LANE13:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT15:%.*]] = insertelement <8 x bfloat> [[VECINIT10]], bfloat [[VGET_LANE13]], i32 3
// CHECK-NEXT: [[VGET_LANE18:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT20:%.*]] = insertelement <8 x bfloat> [[VECINIT15]], bfloat [[VGET_LANE18]], i32 4
// CHECK-NEXT: [[VGET_LANE23:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT25:%.*]] = insertelement <8 x bfloat> [[VECINIT20]], bfloat [[VGET_LANE23]], i32 5
// CHECK-NEXT: [[VGET_LANE28:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT30:%.*]] = insertelement <8 x bfloat> [[VECINIT25]], bfloat [[VGET_LANE28]], i32 6
// CHECK-NEXT: [[VGET_LANE33:%.*]] = extractelement <4 x bfloat> [[B]], i32 0
// CHECK-NEXT: [[VECINIT35:%.*]] = insertelement <8 x bfloat> [[VECINIT30]], bfloat [[VGET_LANE33]], i32 7
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[R:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[VECINIT35]] to <16 x i8>
// CHECK-NEXT: [[VBFMLALTQ_V3_I:%.*]] = call <4 x float> @llvm.arm.neon.bfmlalt(<4 x float> [[R]], <8 x bfloat> [[A]], <8 x bfloat> [[VECINIT35]]) #[[ATTR3]]
// CHECK-NEXT: [[VBFMLALTQ_V4_I:%.*]] = bitcast <4 x float> [[VBFMLALTQ_V3_I]] to <16 x i8>
// CHECK-NEXT: ret <4 x float> [[VBFMLALTQ_V3_I]]
//
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: [[VGET_LANE:%.*]] = extractelement <8 x bfloat> [[B:%.*]], i32 3
// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <8 x bfloat> undef, bfloat [[VGET_LANE]], i32 0
// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <8 x bfloat> [[VECINIT]], bfloat [[VGET_LANE3]], i32 1
// CHECK-NEXT: [[VGET_LANE8:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT10:%.*]] = insertelement <8 x bfloat> [[VECINIT5]], bfloat [[VGET_LANE8]], i32 2
// CHECK-NEXT: [[VGET_LANE13:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT15:%.*]] = insertelement <8 x bfloat> [[VECINIT10]], bfloat [[VGET_LANE13]], i32 3
// CHECK-NEXT: [[VGET_LANE18:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT20:%.*]] = insertelement <8 x bfloat> [[VECINIT15]], bfloat [[VGET_LANE18]], i32 4
// CHECK-NEXT: [[VGET_LANE23:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT25:%.*]] = insertelement <8 x bfloat> [[VECINIT20]], bfloat [[VGET_LANE23]], i32 5
// CHECK-NEXT: [[VGET_LANE28:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT30:%.*]] = insertelement <8 x bfloat> [[VECINIT25]], bfloat [[VGET_LANE28]], i32 6
// CHECK-NEXT: [[VGET_LANE33:%.*]] = extractelement <8 x bfloat> [[B]], i32 3
// CHECK-NEXT: [[VECINIT35:%.*]] = insertelement <8 x bfloat> [[VECINIT30]], bfloat [[VGET_LANE33]], i32 7
// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[R:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x bfloat> [[A:%.*]] to <16 x i8>
// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x bfloat> [[VECINIT35]] to <16 x i8>
// CHECK-NEXT: [[VBFMLALTQ_V3_I:%.*]] = call <4 x float> @llvm.arm.neon.bfmlalt(<4 x float> [[R]], <8 x bfloat> [[A]], <8 x bfloat> [[VECINIT35]]) #[[ATTR3]]
// CHECK-NEXT: [[VBFMLALTQ_V4_I:%.*]] = bitcast <4 x float> [[VBFMLALTQ_V3_I]] to <16 x i8>
// CHECK-NEXT: ret <4 x float> [[VBFMLALTQ_V3_I]]
//
float32x4_t test_vbfmlaltq_laneq_f32(float32x4_t r, bfloat16x8_t a, bfloat16x8_t b) {
return vbfmlaltq_laneq_f32(r, a, b, 3);
}
|
