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
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
|
/*
* Copyright (c) 2017 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#pragma once
#ifdef __aarch64__
#include <arm_neon.h>
inline void a64_gemm_u8_4x4(const uint8_t *Apanel, const uint8_t *Bpanel, uint32_t *Cpanel, int ablocks, int bblocks, int K) {
const uint8_t *a_ptr = Apanel;
uint32_t *c_ptr = Cpanel;
K /= 16;
int oddk = (K & 1);
for (int yb=0; yb<ablocks; yb++) {
const uint8_t *a_ptr0 = a_ptr;
const uint8_t *b_ptr = Bpanel;
for (int xb=0; xb<bblocks; xb++) {
a_ptr = a_ptr0;
int k = ((K+1)/2)-1;
register uint8x16_t b0 asm("v4");
register uint8x16_t b1 asm("v5");
register uint8x16_t b2 asm("v6");
register uint8x16_t b3 asm("v7");
register uint8x16_t b0a asm("v8");
register uint8x16_t b1a asm("v9");
register uint8x16_t b2a asm("v10");
register uint8x16_t b3a asm("v11");
__asm __volatile (
"movi v16.4s, #0x0\n"
"ldr q0, [%[a_ptr]]\n"
"movi v17.4s, #0x0\n"
"ldr %q[b0], [%[b_ptr]]\n"
"movi v18.4s, #0x0\n"
"ldr %q[b1], [%[b_ptr], #16]\n"
"movi v19.4s, #0x0\n"
"ldr %q[b2], [%[b_ptr], #32]\n"
"movi v20.4s, #0x0\n"
"ldr %q[b3], [%[b_ptr], #48]\n"
"movi v21.4s, #0x0\n"
"ldr q1, [%[a_ptr], #16]\n"
"movi v22.4s, #0x0\n"
"ldr q2, [%[a_ptr], #32]\n"
"movi v23.4s, #0x0\n"
"ldr q3, [%[a_ptr], #48]\n"
"movi v24.4s, #0x0\n"
ASM_PREFETCH("[%[b_ptr], #64]")
"movi v25.4s, #0x0\n"
ASM_PREFETCH("[%[a_ptr], #64]")
"movi v26.4s, #0x0\n"
ASM_PREFETCH("[%[b_ptr], #128]")
"movi v27.4s, #0x0\n"
ASM_PREFETCH("[%[a_ptr], #128]")
"movi v28.4s, #0x0\n"
ASM_PREFETCH("[%[b_ptr], #192]")
"movi v29.4s, #0x0\n"
ASM_PREFETCH("[%[a_ptr], #192]")
"movi v30.4s, #0x0\n"
ASM_PREFETCH("[%[b_ptr], #256]")
"movi v31.4s, #0x0\n"
ASM_PREFETCH("[%[a_ptr], #256]")
// Loop structure optimized for A57 (after r0).
// Unavoidably, the multiply will "dribble" if
// dual issued with an add.
// Minimize the effect of this by making sure
// there are 2 adds to run under the dribbled
// multiply.
// Pipeline in blocks of 8 multiplies - combine
// this iteration's multiplies with adds from
// the previous iteration.
// So the first block doesn't have any adds to
// do - but because all the adds are at the
// start of the block it's only the first couple
// of multiplies that need to be pulled out.
// Start of unroll 0 (first iteration)
"umull v12.8h, v0.8b, %[b0].8b\n"
"umull v13.8h, v0.8b, %[b1].8b\n"
// Skip loop if we are doing zero iterations of it.
"cbz %w[k], 4f\n"
// Unroll 0 continuation (branch target)
"1:\n"
"umull v14.8h, v0.8b, %[b2].8b\n"
"subs %w[k], %w[k], #1\n"
"umull v15.8h, v0.8b, %[b3].8b\n"
"ldr %q[b0a], [%[b_ptr], #64]\n"
"umlal2 v12.8h, v0.16b, %[b0].16b\n"
"umlal2 v13.8h, v0.16b, %[b1].16b\n"
"ldr %q[b1a], [%[b_ptr], #80]\n"
"umlal2 v14.8h, v0.16b, %[b2].16b\n"
"umlal2 v15.8h, v0.16b, %[b3].16b\n"
"ldr q0, [%[a_ptr], #64]\n"
"uadalp v16.4s, v12.8h\n"
"umull v12.8h, v1.8b, %[b0].8b\n"
"uadalp v17.4s, v13.8h\n"
"uadalp v18.4s, v14.8h\n"
"umull v13.8h, v1.8b, %[b1].8b\n"
"uadalp v19.4s, v15.8h\n"
"umull v14.8h, v1.8b, %[b2].8b\n"
"ldr %q[b2a], [%[b_ptr], #96]\n"
"umull v15.8h, v1.8b, %[b3].8b\n"
"umlal2 v12.8h, v1.16b, %[b0].16b\n"
"ldr %q[b3a], [%[b_ptr], #112]\n"
"umlal2 v13.8h, v1.16b, %[b1].16b\n"
"add %[b_ptr], %[b_ptr], #128\n"
"umlal2 v14.8h, v1.16b, %[b2].16b\n"
"umlal2 v15.8h, v1.16b, %[b3].16b\n"
"ldr q1, [%[a_ptr], #80]\n"
"uadalp v20.4s, v12.8h\n"
"umull v12.8h, v2.8b, %[b0].8b\n"
"uadalp v21.4s, v13.8h\n"
"uadalp v22.4s, v14.8h\n"
"umull v13.8h, v2.8b, %[b1].8b\n"
"uadalp v23.4s, v15.8h\n"
"umull v14.8h, v2.8b, %[b2].8b\n"
"umull v15.8h, v2.8b, %[b3].8b\n"
"umlal2 v12.8h, v2.16b, %[b0].16b\n"
ASM_PREFETCH("[%[b_ptr], #192]")
"umlal2 v13.8h, v2.16b, %[b1].16b\n"
"umlal2 v14.8h, v2.16b, %[b2].16b\n"
ASM_PREFETCH("[%[a_ptr], #320]")
"umlal2 v15.8h, v2.16b, %[b3].16b\n"
"ldr q2, [%[a_ptr], #96]\n"
"uadalp v24.4s, v12.8h\n"
"umull v12.8h, v3.8b, %[b0].8b\n"
"uadalp v25.4s, v13.8h\n"
"uadalp v26.4s, v14.8h\n"
"umull v13.8h, v3.8b, %[b1].8b\n"
"uadalp v27.4s, v15.8h\n"
"umull v14.8h, v3.8b, %[b2].8b\n"
"umull v15.8h, v3.8b, %[b3].8b\n"
"umlal2 v12.8h, v3.16b, %[b0].16b\n"
"ldr %q[b0], [%[b_ptr], #0]\n"
"umlal2 v13.8h, v3.16b, %[b1].16b\n"
"umlal2 v14.8h, v3.16b, %[b2].16b\n"
"umlal2 v15.8h, v3.16b, %[b3].16b\n"
"ldr q3, [%[a_ptr], #112]\n"
// Unroll 1
"uadalp v28.4s, v12.8h\n"
"umull v12.8h, v0.8b, %[b0a].8b\n"
"uadalp v29.4s, v13.8h\n"
"uadalp v30.4s, v14.8h\n"
"umull v13.8h, v0.8b, %[b1a].8b\n"
"uadalp v31.4s, v15.8h\n"
"umull v14.8h, v0.8b, %[b2a].8b\n"
"umull v15.8h, v0.8b, %[b3a].8b\n"
"ldr %q[b1], [%[b_ptr], #16]\n"
"umlal2 v12.8h, v0.16b, %[b0a].16b\n"
"umlal2 v13.8h, v0.16b, %[b1a].16b\n"
"ldr %q[b2], [%[b_ptr], #32]\n"
"umlal2 v14.8h, v0.16b, %[b2a].16b\n"
"umlal2 v15.8h, v0.16b, %[b3a].16b\n"
"ldr q0, [%[a_ptr], #128]\n"
"uadalp v16.4s, v12.8h\n"
"umull v12.8h, v1.8b, %[b0a].8b\n"
"uadalp v17.4s, v13.8h\n"
"uadalp v18.4s, v14.8h\n"
"umull v13.8h, v1.8b, %[b1a].8b\n"
"uadalp v19.4s, v15.8h\n"
"add %[a_ptr], %[a_ptr], #128\n"
"umull v14.8h, v1.8b, %[b2a].8b\n"
"umull v15.8h, v1.8b, %[b3a].8b\n"
"ldr %q[b3], [%[b_ptr], #48]\n"
"umlal2 v12.8h, v1.16b, %[b0a].16b\n"
"umlal2 v13.8h, v1.16b, %[b1a].16b\n"
"umlal2 v14.8h, v1.16b, %[b2a].16b\n"
"umlal2 v15.8h, v1.16b, %[b3a].16b\n"
"ldr q1, [%[a_ptr], #16]\n"
"uadalp v20.4s, v12.8h\n"
"umull v12.8h, v2.8b, %[b0a].8b\n"
"uadalp v21.4s, v13.8h\n"
"uadalp v22.4s, v14.8h\n"
"umull v13.8h, v2.8b, %[b1a].8b\n"
"uadalp v23.4s, v15.8h\n"
"umull v14.8h, v2.8b, %[b2a].8b\n"
"umull v15.8h, v2.8b, %[b3a].8b\n"
"umlal2 v12.8h, v2.16b, %[b0a].16b\n"
ASM_PREFETCH("[%[b_ptr], #256]")
"umlal2 v13.8h, v2.16b, %[b1a].16b\n"
"umlal2 v14.8h, v2.16b, %[b2a].16b\n"
ASM_PREFETCH("[%[a_ptr], #256]")
"umlal2 v15.8h, v2.16b, %[b3a].16b\n"
"ldr q2, [%[a_ptr], #32]\n"
"uadalp v24.4s, v12.8h\n"
"umull v12.8h, v3.8b, %[b0a].8b\n"
"uadalp v25.4s, v13.8h\n"
"uadalp v26.4s, v14.8h\n"
"umull v13.8h, v3.8b, %[b1a].8b\n"
"uadalp v27.4s, v15.8h\n"
"umull v14.8h, v3.8b, %[b2a].8b\n"
"umull v15.8h, v3.8b, %[b3a].8b\n"
"umlal2 v12.8h, v3.16b, %[b0a].16b\n"
"umlal2 v13.8h, v3.16b, %[b1a].16b\n"
"umlal2 v14.8h, v3.16b, %[b2a].16b\n"
"umlal2 v15.8h, v3.16b, %[b3a].16b\n"
"ldr q3, [%[a_ptr], #48]\n"
// Start of unroll 0 for next iteration.
"uadalp v28.4s, v12.8h\n"
"umull v12.8h, v0.8b, %[b0].8b\n"
"uadalp v29.4s, v13.8h\n"
"uadalp v30.4s, v14.8h\n"
"umull v13.8h, v0.8b, %[b1].8b\n"
"uadalp v31.4s, v15.8h\n"
"bne 1b\n"
// Target to use when K=1 or 2 (i.e. zero iterations of main loop)
"4:\n"
// Branch to alternative tail for odd K
"cbnz %w[oddk], 2f\n"
// Detached final iteration (even K)
"umull v14.8h, v0.8b, %[b2].8b\n"
"umull v15.8h, v0.8b, %[b3].8b\n"
"ldr %q[b0a], [%[b_ptr], #64]\n"
"umlal2 v12.8h, v0.16b, %[b0].16b\n"
"umlal2 v13.8h, v0.16b, %[b1].16b\n"
"ldr %q[b1a], [%[b_ptr], #80]\n"
"umlal2 v14.8h, v0.16b, %[b2].16b\n"
"umlal2 v15.8h, v0.16b, %[b3].16b\n"
"ldr q0, [%[a_ptr], #64]\n"
"uadalp v16.4s, v12.8h\n"
"umull v12.8h, v1.8b, %[b0].8b\n"
"uadalp v17.4s, v13.8h\n"
"uadalp v18.4s, v14.8h\n"
"umull v13.8h, v1.8b, %[b1].8b\n"
"uadalp v19.4s, v15.8h\n"
"umull v14.8h, v1.8b, %[b2].8b\n"
"ldr %q[b2a], [%[b_ptr], #96]\n"
"umull v15.8h, v1.8b, %[b3].8b\n"
"umlal2 v12.8h, v1.16b, %[b0].16b\n"
"ldr %q[b3a], [%[b_ptr], #112]\n"
"umlal2 v13.8h, v1.16b, %[b1].16b\n"
"add %[b_ptr], %[b_ptr], #128\n"
"umlal2 v14.8h, v1.16b, %[b2].16b\n"
"umlal2 v15.8h, v1.16b, %[b3].16b\n"
"ldr q1, [%[a_ptr], #80]\n"
"uadalp v20.4s, v12.8h\n"
"umull v12.8h, v2.8b, %[b0].8b\n"
"uadalp v21.4s, v13.8h\n"
"uadalp v22.4s, v14.8h\n"
"umull v13.8h, v2.8b, %[b1].8b\n"
"uadalp v23.4s, v15.8h\n"
"umull v14.8h, v2.8b, %[b2].8b\n"
"umull v15.8h, v2.8b, %[b3].8b\n"
"umlal2 v12.8h, v2.16b, %[b0].16b\n"
"umlal2 v13.8h, v2.16b, %[b1].16b\n"
"umlal2 v14.8h, v2.16b, %[b2].16b\n"
"umlal2 v15.8h, v2.16b, %[b3].16b\n"
"ldr q2, [%[a_ptr], #96]\n"
"uadalp v24.4s, v12.8h\n"
"umull v12.8h, v3.8b, %[b0].8b\n"
"uadalp v25.4s, v13.8h\n"
"uadalp v26.4s, v14.8h\n"
"umull v13.8h, v3.8b, %[b1].8b\n"
"uadalp v27.4s, v15.8h\n"
"umull v14.8h, v3.8b, %[b2].8b\n"
"umull v15.8h, v3.8b, %[b3].8b\n"
"umlal2 v12.8h, v3.16b, %[b0].16b\n"
"umlal2 v13.8h, v3.16b, %[b1].16b\n"
"umlal2 v14.8h, v3.16b, %[b2].16b\n"
"umlal2 v15.8h, v3.16b, %[b3].16b\n"
"ldr q3, [%[a_ptr], #112]\n"
// Unroll 1
"uadalp v28.4s, v12.8h\n"
"umull v12.8h, v0.8b, %[b0a].8b\n"
"uadalp v29.4s, v13.8h\n"
"uadalp v30.4s, v14.8h\n"
"umull v13.8h, v0.8b, %[b1a].8b\n"
"uadalp v31.4s, v15.8h\n"
"umull v14.8h, v0.8b, %[b2a].8b\n"
"add %[a_ptr], %[a_ptr], #128\n"
"umull v15.8h, v0.8b, %[b3a].8b\n"
"umlal2 v12.8h, v0.16b, %[b0a].16b\n"
"umlal2 v13.8h, v0.16b, %[b1a].16b\n"
"umlal2 v14.8h, v0.16b, %[b2a].16b\n"
"umlal2 v15.8h, v0.16b, %[b3a].16b\n"
"uadalp v16.4s, v12.8h\n"
"umull v12.8h, v1.8b, %[b0a].8b\n"
"uadalp v17.4s, v13.8h\n"
"uadalp v18.4s, v14.8h\n"
"umull v13.8h, v1.8b, %[b1a].8b\n"
"uadalp v19.4s, v15.8h\n"
"umull v14.8h, v1.8b, %[b2a].8b\n"
"umull v15.8h, v1.8b, %[b3a].8b\n"
"umlal2 v12.8h, v1.16b, %[b0a].16b\n"
"addp v16.4s, v16.4s, v17.4s\n"
"umlal2 v13.8h, v1.16b, %[b1a].16b\n"
"addp v17.4s, v18.4s, v19.4s\n"
"umlal2 v14.8h, v1.16b, %[b2a].16b\n"
"umlal2 v15.8h, v1.16b, %[b3a].16b\n"
"uadalp v20.4s, v12.8h\n"
"umull v12.8h, v2.8b, %[b0a].8b\n"
"uadalp v21.4s, v13.8h\n"
"uadalp v22.4s, v14.8h\n"
"umull v13.8h, v2.8b, %[b1a].8b\n"
"uadalp v23.4s, v15.8h\n"
"addp v16.4s, v16.4s, v17.4s\n"
"umull v14.8h, v2.8b, %[b2a].8b\n"
"addp v18.4s, v20.4s, v21.4s\n"
"addp v19.4s, v22.4s, v23.4s\n"
"umull v15.8h, v2.8b, %[b3a].8b\n"
"umlal2 v12.8h, v2.16b, %[b0a].16b\n"
"str q16, [%[c_ptr]]\n"
"umlal2 v13.8h, v2.16b, %[b1a].16b\n"
"umlal2 v14.8h, v2.16b, %[b2a].16b\n"
"umlal2 v15.8h, v2.16b, %[b3a].16b\n"
"uadalp v24.4s, v12.8h\n"
"umull v12.8h, v3.8b, %[b0a].8b\n"
"uadalp v25.4s, v13.8h\n"
"uadalp v26.4s, v14.8h\n"
"umull v13.8h, v3.8b, %[b1a].8b\n"
"uadalp v27.4s, v15.8h\n"
"addp v17.4s, v18.4s, v19.4s\n"
"umull v14.8h, v3.8b, %[b2a].8b\n"
"addp v20.4s, v24.4s, v25.4s\n"
"addp v21.4s, v26.4s, v27.4s\n"
"umull v15.8h, v3.8b, %[b3a].8b\n"
"umlal2 v12.8h, v3.16b, %[b0a].16b\n"
"str q17, [%[c_ptr], #16]\n"
"umlal2 v13.8h, v3.16b, %[b1a].16b\n"
"umlal2 v14.8h, v3.16b, %[b2a].16b\n"
"addp v18.4s, v20.4s, v21.4s\n"
"umlal2 v15.8h, v3.16b, %[b3a].16b\n"
"b 3f\n"
// Detached final iteration (odd K)
"2:\n"
"umull v14.8h, v0.8b, %[b2].8b\n"
"add %[a_ptr], %[a_ptr], #64\n"
"umull v15.8h, v0.8b, %[b3].8b\n"
"add %[b_ptr], %[b_ptr], #64\n"
"umlal2 v12.8h, v0.16b, %[b0].16b\n"
"umlal2 v13.8h, v0.16b, %[b1].16b\n"
"umlal2 v14.8h, v0.16b, %[b2].16b\n"
"umlal2 v15.8h, v0.16b, %[b3].16b\n"
"uadalp v16.4s, v12.8h\n"
"umull v12.8h, v1.8b, %[b0].8b\n"
"uadalp v17.4s, v13.8h\n"
"uadalp v18.4s, v14.8h\n"
"umull v13.8h, v1.8b, %[b1].8b\n"
"uadalp v19.4s, v15.8h\n"
"umull v14.8h, v1.8b, %[b2].8b\n"
"umull v15.8h, v1.8b, %[b3].8b\n"
"umlal2 v12.8h, v1.16b, %[b0].16b\n"
"addp v16.4s, v16.4s, v17.4s\n"
"umlal2 v13.8h, v1.16b, %[b1].16b\n"
"addp v17.4s, v18.4s, v19.4s\n"
"umlal2 v14.8h, v1.16b, %[b2].16b\n"
"umlal2 v15.8h, v1.16b, %[b3].16b\n"
"uadalp v20.4s, v12.8h\n"
"umull v12.8h, v2.8b, %[b0].8b\n"
"uadalp v21.4s, v13.8h\n"
"uadalp v22.4s, v14.8h\n"
"umull v13.8h, v2.8b, %[b1].8b\n"
"uadalp v23.4s, v15.8h\n"
"addp v16.4s, v16.4s, v17.4s\n"
"umull v14.8h, v2.8b, %[b2].8b\n"
"addp v18.4s, v20.4s, v21.4s\n"
"addp v19.4s, v22.4s, v23.4s\n"
"umull v15.8h, v2.8b, %[b3].8b\n"
"umlal2 v12.8h, v2.16b, %[b0].16b\n"
"str q16, [%[c_ptr]]\n"
"umlal2 v13.8h, v2.16b, %[b1].16b\n"
"umlal2 v14.8h, v2.16b, %[b2].16b\n"
"umlal2 v15.8h, v2.16b, %[b3].16b\n"
"uadalp v24.4s, v12.8h\n"
"umull v12.8h, v3.8b, %[b0].8b\n"
"uadalp v25.4s, v13.8h\n"
"uadalp v26.4s, v14.8h\n"
"umull v13.8h, v3.8b, %[b1].8b\n"
"uadalp v27.4s, v15.8h\n"
"addp v17.4s, v18.4s, v19.4s\n"
"umull v14.8h, v3.8b, %[b2].8b\n"
"addp v20.4s, v24.4s, v25.4s\n"
"addp v21.4s, v26.4s, v27.4s\n"
"umull v15.8h, v3.8b, %[b3].8b\n"
"umlal2 v12.8h, v3.16b, %[b0].16b\n"
"str q17, [%[c_ptr], #16]\n"
"umlal2 v13.8h, v3.16b, %[b1].16b\n"
"umlal2 v14.8h, v3.16b, %[b2].16b\n"
"addp v18.4s, v20.4s, v21.4s\n"
"umlal2 v15.8h, v3.16b, %[b3].16b\n"
"3:\n"
// Final additions
"uadalp v28.4s, v12.8h\n"
"str q18, [%[c_ptr], #32]\n"
"uadalp v29.4s, v13.8h\n"
"uadalp v30.4s, v14.8h\n"
"uadalp v31.4s, v15.8h\n"
// Horizontal reduction, phase 1
"addp v22.4s, v28.4s, v29.4s\n"
"addp v23.4s, v30.4s, v31.4s\n"
// Horizontal reduction, phase 2
"addp v19.4s, v22.4s, v23.4s\n"
"str q19, [%[c_ptr], #48]\n"
"add %[c_ptr], %[c_ptr], #64\n"
:
[a_ptr] "+r" (a_ptr), [b_ptr] "+r" (b_ptr), [c_ptr] "+r" (c_ptr),
[b0] "+w" (b0), [b1] "+w" (b1), [b2] "+w" (b2), [b3] "+w" (b3),
[b0a] "+w" (b0a), [b1a] "+w" (b1a), [b2a] "+w" (b2a), [b3a] "+w" (b3a),
[k] "+r" (k)
: [oddk] "r" (oddk)
: "x20", "x21", "v0","v1","v2","v3","v12","v13","v14","v15","v16","v17","v18","v19",
"v20","v21","v22","v23","v24","v25","v26","v27","v28","v29","v30","v31", "cc");
}
}
}
#endif // __aarch64__
|