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
|
/*
* Copyright (c) 2022 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.
*/
#ifdef __ARM_FEATURE_SVE
#ifdef ARM_COMPUTE_ENABLE_SME2
#include "arm_gemm.hpp"
#include <cstdint>
#include "../../asmlib.hpp"
#include "../../utils.hpp"
namespace arm_gemm {
void sme2_interleaved_nomerge_s8q_mopa_2VLx2VL(const int8_t *const A, const int8_t *const B, int8_t *const C, int ldc, const int M, const int N, const int K, const int32_t *const bias, const Requantize32 &rq, const int n_0, bool accumulate, int32_t *const accumulator_buffer)
{
struct KernelArgs
{
KernelArgs(
const int8_t *const A,
const int8_t *const B,
int8_t *const C, const int ldc,
const int M, const int N, const int K,
const int32_t *const bias,
const Requantize32 &rq,
const int n_0,
bool accumulate,
int32_t *const accumulator_buffer
) : A(A),
B(B), kstride_bytes(roundup(K, 4) * sizeof(int8_t)),
C(C), ldcb(ldc * sizeof(int8_t)),
M(M), N(N), K(K),
n_loops(((K / 4) - 1) / 2), n_tail_iters(((K / 4) - 1) % 2),
bias(bias), n_0(n_0),
accumulator_buffer(accumulator_buffer),
flags(0x0)
{
if (accumulate)
{
flags |= 1 << 0; // FILL_ACCUMULATORS_FROM_BUFFER
}
if (C == nullptr)
{
flags |= 1 << 1; // STORE_ACCUMULATORS_TO_BUFFER
}
if (rq.per_channel_requant)
{
flags |= 1 << 2; // PER_CHANNEL_QUANTISATION
}
}
const int8_t *const A;
const int8_t *const B;
const long kstride_bytes;
int8_t *const C;
const long ldcb;
const long M, N, K, n_loops, n_tail_iters;
int32_t min = std::numeric_limits<int8_t>::min();
int32_t max = std::numeric_limits<int8_t>::max();
const int32_t *const bias;
const int n_0;
int32_t *const accumulator_buffer;
uint64_t flags;
};
// Construct arguments for this kernel
KernelArgs args(A, B, C, ldc, M, N, K, bias, rq, n_0, accumulate, accumulator_buffer);
__asm__ __volatile__(
"ldr x15, [%x[args], %[offsetof_flags]]\n"
".inst 0xd503477f // SMSTART ZA\n"
"ptrue p1.b\n"
".inst 0x25207811 // ptrue pn9.b\n"
"ldr x14, [%x[args], %[offsetof_accumulator_buffer]]\n"
"ldr x13, [%x[args], %[offsetof_accumulator_buffer]]\n"
"tbz x15, #0, 2f\n"
"mov x12, #0x0\n"
"cntw x19\n"
"1:" // Initial accumulator load from buffer: Loop
".inst 0xa040c5c0 // ld1w { z0.s-z3.s }, pn9.b/Z, [x14]\n"
".inst 0xc0840400 // mova za0h.s[x12], { z0.s-z3.s }\n"
".inst 0xa041c5cc // ld1w { z12.s-z15.s }, pn9.b/Z, [x14, #0x4, MUL VL]\n"
".inst 0xc0840581 // mova za1h.s[x12], { z12.s-z15.s }\n"
".inst 0xa042c5c0 // ld1w { z0.s-z3.s }, pn9.b/Z, [x14, #0x8, MUL VL]\n"
".inst 0xc0840402 // mova za2h.s[x12], { z0.s-z3.s }\n"
".inst 0xa043c5dc // ld1w { z28.s-z31.s }, pn9.b/Z, [x14, #0xc, MUL VL]\n"
".inst 0xc0840783 // mova za3h.s[x12], { z28.s-z31.s }\n"
"add x12, x12, #0x4\n"
"cmp x12, x19\n"
"addvl x14, x14, #16\n"
"blt 1b\n"
"2:" // Initial accumulator load from buffer: End
"ldr w11, [%x[args], %[offsetof_M]]\n"
"mov x10, #0x0\n"
"mov x9, #0x0\n"
"ldr w28, [%x[args], %[offsetof_N]]\n"
"ldr x27, [%x[args], %[offsetof_A]]\n"
"3:" // M and N loop
"mov x26, x27\n"
".inst 0x25bc4530 // whilelt pn8.s, x9, x28, VLx2\n"
"tbnz x15, #0, 4f\n"
"ldr x19, [%x[args], %[offsetof_bias]]\n"
".inst 0xc00800ff // zero { zad0, zad1, zad2, zad3, zad4, zad5, zad6, zad7 }\n"
"cbz x19, 5f\n"
".inst 0xa0094275 // ldnt1w { z20.s-z21.s }, p8/Z, [x19, x9, LSL #2]\n"
".inst 0xc0902680 // addha za0.s, p1/M, p1/M, z20.s\n"
".inst 0xc09026a1 // addha za1.s, p1/M, p1/M, z21.s\n"
".inst 0xc0902682 // addha za2.s, p1/M, p1/M, z20.s\n"
".inst 0xc09026a3 // addha za3.s, p1/M, p1/M, z21.s\n"
"4:" // Prepare accumulators: Test for last block
"mov x19, x9\n"
"mov x20, x10\n"
"incw x19, ALL, MUL #2\n"
"incw x20, ALL, MUL #2\n"
"cmp x19, x28\n"
"csel x20, x10, x20, LT\n"
"mov x19, x15\n"
"bfm x15, XZR, #0x0, #0x0 // bfc x15, #0x0, #0x1\n"
"cmp x20, x11\n"
"csel x15, x19, x15, LT\n"
"5:" // Prepare accumulators: End
"ldr x19, [%x[args], %[offsetof_K]]\n"
"add x19, x19, #0x3\n"
"lsr x19, x19, #0x2\n"
"ldr x22, [%x[args], %[offsetof_B]]\n"
"lsr x21, x19, #0x2\n"
"and x20, x19, #0x3\n"
"ldr x19, [%x[args], %[offsetof_kstride_bytes]]\n"
"madd x22, x9, x19, x22\n" // bptr = B + n * kstride_bytes
"cbz x21, 8f\n"
"subs x21, x21, #0x1\n"
".inst 0xa040075e // ld1b { z30.b-z31.b }, pn9.b/Z, [x26]\n"
".inst 0xa04006d1 // ldnt1b { z16.b-z17.b }, pn9.b/Z, [x22]\n"
".inst 0xa041074e // ld1b { z14.b-z15.b }, pn9.b/Z, [x26, #0x2, MUL VL]\n"
".inst 0xa04106c9 // ldnt1b { z8.b-z9.b }, pn9.b/Z, [x22, #0x2, MUL VL]\n"
".inst 0xa0420740 // ld1b { z0.b-z1.b }, pn9.b/Z, [x26, #0x4, MUL VL]\n"
".inst 0xa14206dc // ldnt1b { z20.b, z28.b }, pn9.b/Z, [x22, #0x4, MUL VL]\n"
".inst 0xa0430744 // ld1b { z4.b-z5.b }, pn9.b/Z, [x26, #0x6, MUL VL]\n"
"addvl x26, x26, #8\n"
".inst 0xa14306ca // ldnt1b { z2.b, z10.b }, pn9.b/Z, [x22, #0x6, MUL VL]\n"
"addvl x22, x22, #8\n"
"ble 7f\n"
"6:" // K loop
".inst 0xa09027c0 // smopa za0.s, p1/M, p1/M, z30.b, z16.b\n"
"subs x21, x21, #0x1\n"
".inst 0xa09127c1 // smopa za1.s, p1/M, p1/M, z30.b, z17.b\n"
".inst 0xa09027e2 // smopa za2.s, p1/M, p1/M, z31.b, z16.b\n"
".inst 0xa09127e3 // smopa za3.s, p1/M, p1/M, z31.b, z17.b\n"
".inst 0xa040075e // ld1b { z30.b-z31.b }, pn9.b/Z, [x26]\n"
".inst 0xa08825c0 // smopa za0.s, p1/M, p1/M, z14.b, z8.b\n"
".inst 0xa04006d1 // ldnt1b { z16.b-z17.b }, pn9.b/Z, [x22]\n"
".inst 0xa08925c1 // smopa za1.s, p1/M, p1/M, z14.b, z9.b\n"
".inst 0xa08825e2 // smopa za2.s, p1/M, p1/M, z15.b, z8.b\n"
".inst 0xa08925e3 // smopa za3.s, p1/M, p1/M, z15.b, z9.b\n"
".inst 0xa041074e // ld1b { z14.b-z15.b }, pn9.b/Z, [x26, #0x2, MUL VL]\n"
".inst 0xa0942400 // smopa za0.s, p1/M, p1/M, z0.b, z20.b\n"
".inst 0xa04106c9 // ldnt1b { z8.b-z9.b }, pn9.b/Z, [x22, #0x2, MUL VL]\n"
".inst 0xa09c2401 // smopa za1.s, p1/M, p1/M, z0.b, z28.b\n"
".inst 0xa0942422 // smopa za2.s, p1/M, p1/M, z1.b, z20.b\n"
".inst 0xa09c2423 // smopa za3.s, p1/M, p1/M, z1.b, z28.b\n"
".inst 0xa0420740 // ld1b { z0.b-z1.b }, pn9.b/Z, [x26, #0x4, MUL VL]\n"
".inst 0xa14206dc // ldnt1b { z20.b, z28.b }, pn9.b/Z, [x22, #0x4, MUL VL]\n"
".inst 0xa0822480 // smopa za0.s, p1/M, p1/M, z4.b, z2.b\n"
".inst 0xa08a2481 // smopa za1.s, p1/M, p1/M, z4.b, z10.b\n"
".inst 0xa08224a2 // smopa za2.s, p1/M, p1/M, z5.b, z2.b\n"
".inst 0xa08a24a3 // smopa za3.s, p1/M, p1/M, z5.b, z10.b\n"
".inst 0xa0430744 // ld1b { z4.b-z5.b }, pn9.b/Z, [x26, #0x6, MUL VL]\n"
"addvl x26, x26, #8\n"
".inst 0xa14306ca // ldnt1b { z2.b, z10.b }, pn9.b/Z, [x22, #0x6, MUL VL]\n"
"addvl x22, x22, #8\n"
"bgt 6b\n"
"7:" // K loop tail
".inst 0xa09027c0 // smopa za0.s, p1/M, p1/M, z30.b, z16.b\n"
".inst 0xa09127c1 // smopa za1.s, p1/M, p1/M, z30.b, z17.b\n"
".inst 0xa09027e2 // smopa za2.s, p1/M, p1/M, z31.b, z16.b\n"
".inst 0xa09127e3 // smopa za3.s, p1/M, p1/M, z31.b, z17.b\n"
".inst 0xa08825c0 // smopa za0.s, p1/M, p1/M, z14.b, z8.b\n"
".inst 0xa08925c1 // smopa za1.s, p1/M, p1/M, z14.b, z9.b\n"
".inst 0xa08825e2 // smopa za2.s, p1/M, p1/M, z15.b, z8.b\n"
".inst 0xa08925e3 // smopa za3.s, p1/M, p1/M, z15.b, z9.b\n"
".inst 0xa0942400 // smopa za0.s, p1/M, p1/M, z0.b, z20.b\n"
".inst 0xa09c2401 // smopa za1.s, p1/M, p1/M, z0.b, z28.b\n"
".inst 0xa0942422 // smopa za2.s, p1/M, p1/M, z1.b, z20.b\n"
".inst 0xa09c2423 // smopa za3.s, p1/M, p1/M, z1.b, z28.b\n"
".inst 0xa0822480 // smopa za0.s, p1/M, p1/M, z4.b, z2.b\n"
".inst 0xa08a2481 // smopa za1.s, p1/M, p1/M, z4.b, z10.b\n"
".inst 0xa08224a2 // smopa za2.s, p1/M, p1/M, z5.b, z2.b\n"
".inst 0xa08a24a3 // smopa za3.s, p1/M, p1/M, z5.b, z10.b\n"
"8:" // K oddments
"cbz x20, 10f\n"
"9:" // K oddments: Loop
".inst 0xa040075e // ld1b { z30.b-z31.b }, pn9.b/Z, [x26]\n"
"subs x20, x20, #0x1\n"
"addvl x26, x26, #2\n"
".inst 0xa04006d0 // ld1b { z16.b-z17.b }, pn9.b/Z, [x22]\n"
"addvl x22, x22, #2\n"
".inst 0xa09027c0 // smopa za0.s, p1/M, p1/M, z30.b, z16.b\n"
".inst 0xa09127c1 // smopa za1.s, p1/M, p1/M, z30.b, z17.b\n"
".inst 0xa09027e2 // smopa za2.s, p1/M, p1/M, z31.b, z16.b\n"
".inst 0xa09127e3 // smopa za3.s, p1/M, p1/M, z31.b, z17.b\n"
"bgt 9b\n"
"10:" // K oddments: End
".inst 0xa040474e // ld1w { z14.s-z15.s }, pn9.b/Z, [x26]\n"
"addvl x26, x26, #2\n"
".inst 0xc09125c0 // addva za0.s, p1/M, p1/M, z14.s\n"
".inst 0xc09125c1 // addva za1.s, p1/M, p1/M, z14.s\n"
".inst 0xc09125e2 // addva za2.s, p1/M, p1/M, z15.s\n"
".inst 0xc09125e3 // addva za3.s, p1/M, p1/M, z15.s\n"
"tbz x15, #1, 14f\n"
"tbz x15, #0, 12f\n"
"mov x12, #0x0\n"
"cntw x19\n"
"11:" // Store to partial result buffer: Store and refill: Loop
".inst 0xa040c5dc // ld1w { z28.s-z31.s }, pn9.b/Z, [x14]\n"
".inst 0xc0860408 // mova { z8.s-z11.s }, za0h.s[x12]\n"
".inst 0xc0840780 // mova za0h.s[x12], { z28.s-z31.s }\n"
".inst 0xc0860434 // mova { z20.s-z23.s }, za1h.s[x12]\n"
".inst 0xa041c5d8 // ld1w { z24.s-z27.s }, pn9.b/Z, [x14, #0x4, MUL VL]\n"
".inst 0xc0840701 // mova za1h.s[x12], { z24.s-z27.s }\n"
".inst 0xc086045c // mova { z28.s-z31.s }, za2h.s[x12]\n"
".inst 0xc0860470 // mova { z16.s-z19.s }, za3h.s[x12]\n"
".inst 0xa042c5d8 // ld1w { z24.s-z27.s }, pn9.b/Z, [x14, #0x8, MUL VL]\n"
".inst 0xc0840702 // mova za2h.s[x12], { z24.s-z27.s }\n"
".inst 0xa043c5cc // ld1w { z12.s-z15.s }, pn9.b/Z, [x14, #0xc, MUL VL]\n"
".inst 0xc0840583 // mova za3h.s[x12], { z12.s-z15.s }\n"
"add x12, x12, #0x4\n"
"cmp x12, x19\n"
".inst 0xa060c5a8 // st1w { z8.s-z11.s }, pn9.b, [x13]\n"
"addvl x14, x14, #16\n"
".inst 0xa061c5b4 // st1w { z20.s-z23.s }, pn9.b, [x13, #0x4, MUL VL]\n"
".inst 0xa062c5bc // st1w { z28.s-z31.s }, pn9.b, [x13, #0x8, MUL VL]\n"
".inst 0xa063c5b0 // st1w { z16.s-z19.s }, pn9.b, [x13, #0xc, MUL VL]\n"
"addvl x13, x13, #16\n"
"blt 11b\n"
"b 24f\n"
"12:" // Store to partial result buffer: Store only
"mov x12, #0x0\n"
"cntw x19\n"
"13:" // Store to partial result buffer: Store only: Loop
".inst 0xc0860410 // mova { z16.s-z19.s }, za0h.s[x12]\n"
".inst 0xc0860424 // mova { z4.s-z7.s }, za1h.s[x12]\n"
".inst 0xa060c5b0 // st1w { z16.s-z19.s }, pn9.b, [x13]\n"
".inst 0xc0860448 // mova { z8.s-z11.s }, za2h.s[x12]\n"
".inst 0xc086046c // mova { z12.s-z15.s }, za3h.s[x12]\n"
".inst 0xa061c5a4 // st1w { z4.s-z7.s }, pn9.b, [x13, #0x4, MUL VL]\n"
"add x12, x12, #0x4\n"
"cmp x12, x19\n"
".inst 0xa062c5a8 // st1w { z8.s-z11.s }, pn9.b, [x13, #0x8, MUL VL]\n"
".inst 0xa063c5ac // st1w { z12.s-z15.s }, pn9.b, [x13, #0xc, MUL VL]\n"
"addvl x13, x13, #16\n"
"blt 13b\n"
"b 24f\n"
"14:" // Store to output array
"ldr x25, [%x[args], %[offsetof_C]]\n"
"add x25, x25, x9\n" // C += n
"sub x24, x11, x10\n"
"ld1rw { z2.s }, p1/Z, [%x[rq], %[offsetof_Requantize32_per_layer_mul]]\n"
"ldr x23, [%x[args], %[offsetof_ldcb]]\n"
"madd x25, x10, x23, x25\n" // C += m * ldc
"ld1rw { z3.s }, p1/Z, [%x[rq], %[offsetof_Requantize32_per_layer_mul]]\n"
"ld1rw { z0.s }, p1/Z, [%x[rq], %[offsetof_Requantize32_per_layer_right_shift]]\n"
"ld1rw { z1.s }, p1/Z, [%x[rq], %[offsetof_Requantize32_per_layer_right_shift]]\n"
"ld1rw { z11.s }, p1/Z, [%x[rq], %[offsetof_Requantize32_c_offset]]\n"
"ld1rw { z25.s }, p1/Z, [%x[rq], %[offsetof_Requantize32_minval]]\n"
"ld1rw { z24.s }, p1/Z, [%x[rq], %[offsetof_Requantize32_maxval]]\n"
"tbz x15, #2, 15f\n"
"ldr w20, [%x[args], %[offsetof_n_0]]\n"
"add x20, x20, x9\n"
"ldr x19, [%x[rq], %[offsetof_Requantize32_per_channel_muls]]\n"
"add x19, x19, x20, LSL #2\n"
".inst 0xa0404262 // ld1w { z2.s-z3.s }, p8/Z, [x19]\n"
"ldr x19, [%x[rq], %[offsetof_Requantize32_per_channel_right_shifts]]\n"
"add x19, x19, x20, LSL #2\n"
".inst 0xa0404260 // ld1w { z0.s-z1.s }, p8/Z, [x19]\n"
"15:" // Store to output array: Load per-channel parameters: End
"cntw x22\n"
"whilelt p0.h, x9, x28\n"
"cmp x24, x22\n"
"csel x21, x24, x22, LT\n"
"lsr x20, x21, #0x2\n"
"mov x12, #0x0\n"
"and x19, x21, #0x3\n"
"cbz x20, 17f\n"
"16:" // Store to output array: Accumulator row 0 loop
".inst 0xc086040c // mova { z12.s-z15.s }, za0h.s[x12]\n"
".inst 0xc086043c // mova { z28.s-z31.s }, za1h.s[x12]\n"
".inst 0xc1a2ac0c // sqdmulh { z12.s-z15.s }, { z12.s-z15.s }, z2.s\n"
".inst 0xc1a3ac1c // sqdmulh { z28.s-z31.s }, { z28.s-z31.s }, z3.s\n"
"add x12, x12, #0x4\n"
"cmp x12, x20, LSL #2\n"
".inst 0xc1a0aa2c // srshl { z12.s-z15.s }, { z12.s-z15.s }, z0.s\n"
".inst 0xc1a1aa3c // srshl { z28.s-z31.s }, { z28.s-z31.s }, z1.s\n"
".inst 0xc1abab0c // add { z12.s-z15.s }, { z12.s-z15.s }, z11.s\n"
".inst 0xc1abab1c // add { z28.s-z31.s }, { z28.s-z31.s }, z11.s\n"
".inst 0xc1b8cf2c // sclamp { z12.s-z15.s }, z25.s, z24.s\n"
".inst 0xc1b8cf3c // sclamp { z28.s-z31.s }, z25.s, z24.s\n"
"uzp1 z16.h, z12.h, z28.h\n"
"st1b { z16.h }, p0, [x25]\n"
"add x25, x25, x23\n"
"uzp1 z16.h, z13.h, z29.h\n"
"uzp1 z17.h, z14.h, z30.h\n"
"st1b { z16.h }, p0, [x25]\n"
"add x25, x25, x23\n"
"uzp1 z16.h, z15.h, z31.h\n"
"st1b { z17.h }, p0, [x25]\n"
"add x25, x25, x23\n"
"st1b { z16.h }, p0, [x25]\n"
"add x25, x25, x23\n"
"blt 16b\n"
"17:" // Store to output array: Accumulator row 0 oddments
"cbz x19, 18f\n"
".inst 0xc086041c // mova { z28.s-z31.s }, za0h.s[x12]\n"
".inst 0xc086042c // mova { z12.s-z15.s }, za1h.s[x12]\n"
".inst 0xc1a2ac1c // sqdmulh { z28.s-z31.s }, { z28.s-z31.s }, z2.s\n"
".inst 0xc1a3ac0c // sqdmulh { z12.s-z15.s }, { z12.s-z15.s }, z3.s\n"
"subs x19, x19, #0x1\n"
".inst 0xc1a0aa3c // srshl { z28.s-z31.s }, { z28.s-z31.s }, z0.s\n"
".inst 0xc1a1aa2c // srshl { z12.s-z15.s }, { z12.s-z15.s }, z1.s\n"
".inst 0xc1abab1c // add { z28.s-z31.s }, { z28.s-z31.s }, z11.s\n"
".inst 0xc1abab0c // add { z12.s-z15.s }, { z12.s-z15.s }, z11.s\n"
".inst 0xc1b8cf3c // sclamp { z28.s-z31.s }, z25.s, z24.s\n"
".inst 0xc1b8cf2c // sclamp { z12.s-z15.s }, z25.s, z24.s\n"
"uzp1 z16.h, z28.h, z12.h\n"
"st1b { z16.h }, p0, [x25]\n"
"add x25, x25, x23\n"
"beq 18f\n"
"subs x19, x19, #0x1\n"
"uzp1 z16.h, z29.h, z13.h\n"
"st1b { z16.h }, p0, [x25]\n"
"add x25, x25, x23\n"
"beq 18f\n"
"uzp1 z16.h, z30.h, z14.h\n"
"st1b { z16.h }, p0, [x25]\n"
"add x25, x25, x23\n"
"18:" // Store to output array: Accumulator row 0 oddments: End
"subs x24, x24, x21\n"
"beq 22f\n"
"whilelt p0.h, x9, x28\n"
"cmp x24, x22\n"
"csel x19, x24, x22, LT\n"
"lsr x20, x19, #0x2\n"
"mov x12, #0x0\n"
"and x19, x19, #0x3\n"
"cbz x20, 20f\n"
"19:" // Store to output array: Accumulator row 1 loop
".inst 0xc0860444 // mova { z4.s-z7.s }, za2h.s[x12]\n"
".inst 0xc0860470 // mova { z16.s-z19.s }, za3h.s[x12]\n"
".inst 0xc1a2ac04 // sqdmulh { z4.s-z7.s }, { z4.s-z7.s }, z2.s\n"
".inst 0xc1a3ac10 // sqdmulh { z16.s-z19.s }, { z16.s-z19.s }, z3.s\n"
"add x12, x12, #0x4\n"
"cmp x12, x20, LSL #2\n"
".inst 0xc1a0aa24 // srshl { z4.s-z7.s }, { z4.s-z7.s }, z0.s\n"
".inst 0xc1a1aa30 // srshl { z16.s-z19.s }, { z16.s-z19.s }, z1.s\n"
".inst 0xc1abab04 // add { z4.s-z7.s }, { z4.s-z7.s }, z11.s\n"
".inst 0xc1abab10 // add { z16.s-z19.s }, { z16.s-z19.s }, z11.s\n"
".inst 0xc1b8cf24 // sclamp { z4.s-z7.s }, z25.s, z24.s\n"
".inst 0xc1b8cf30 // sclamp { z16.s-z19.s }, z25.s, z24.s\n"
"uzp1 z16.h, z4.h, z16.h\n"
"st1b { z16.h }, p0, [x25]\n"
"add x25, x25, x23\n"
"uzp1 z16.h, z5.h, z17.h\n"
"uzp1 z17.h, z6.h, z18.h\n"
"st1b { z16.h }, p0, [x25]\n"
"add x25, x25, x23\n"
"uzp1 z16.h, z7.h, z19.h\n"
"st1b { z17.h }, p0, [x25]\n"
"add x25, x25, x23\n"
"st1b { z16.h }, p0, [x25]\n"
"add x25, x25, x23\n"
"blt 19b\n"
"20:" // Store to output array: Accumulator row 1 oddments
"cbz x19, 21f\n"
".inst 0xc0860454 // mova { z20.s-z23.s }, za2h.s[x12]\n"
".inst 0xc0860470 // mova { z16.s-z19.s }, za3h.s[x12]\n"
".inst 0xc1a2ac14 // sqdmulh { z20.s-z23.s }, { z20.s-z23.s }, z2.s\n"
".inst 0xc1a3ac10 // sqdmulh { z16.s-z19.s }, { z16.s-z19.s }, z3.s\n"
"subs x19, x19, #0x1\n"
".inst 0xc1a0aa34 // srshl { z20.s-z23.s }, { z20.s-z23.s }, z0.s\n"
".inst 0xc1a1aa30 // srshl { z16.s-z19.s }, { z16.s-z19.s }, z1.s\n"
".inst 0xc1abab14 // add { z20.s-z23.s }, { z20.s-z23.s }, z11.s\n"
".inst 0xc1abab10 // add { z16.s-z19.s }, { z16.s-z19.s }, z11.s\n"
".inst 0xc1b8cf34 // sclamp { z20.s-z23.s }, z25.s, z24.s\n"
".inst 0xc1b8cf30 // sclamp { z16.s-z19.s }, z25.s, z24.s\n"
"uzp1 z16.h, z20.h, z16.h\n"
"st1b { z16.h }, p0, [x25]\n"
"add x25, x25, x23\n"
"beq 21f\n"
"subs x19, x19, #0x1\n"
"uzp1 z16.h, z21.h, z17.h\n"
"st1b { z16.h }, p0, [x25]\n"
"add x25, x25, x23\n"
"beq 21f\n"
"uzp1 z16.h, z22.h, z18.h\n"
"st1b { z16.h }, p0, [x25]\n"
"21:" // Store to output array: Accumulator row 1 oddments: End
"22:" // Store to output array: End
"tbz x15, #0, 24f\n"
"mov x12, #0x0\n"
"cntw x19\n"
"23:" // Store to output array: Refill accumulators: Loop
".inst 0xa040c5d0 // ld1w { z16.s-z19.s }, pn9.b/Z, [x14]\n"
".inst 0xc0840600 // mova za0h.s[x12], { z16.s-z19.s }\n"
".inst 0xa041c5d0 // ld1w { z16.s-z19.s }, pn9.b/Z, [x14, #0x4, MUL VL]\n"
".inst 0xc0840601 // mova za1h.s[x12], { z16.s-z19.s }\n"
".inst 0xa042c5d0 // ld1w { z16.s-z19.s }, pn9.b/Z, [x14, #0x8, MUL VL]\n"
".inst 0xc0840602 // mova za2h.s[x12], { z16.s-z19.s }\n"
".inst 0xa043c5c4 // ld1w { z4.s-z7.s }, pn9.b/Z, [x14, #0xc, MUL VL]\n"
".inst 0xc0840483 // mova za3h.s[x12], { z4.s-z7.s }\n"
"add x12, x12, #0x4\n"
"cmp x12, x19\n"
"addvl x14, x14, #16\n"
"blt 23b\n"
"24:" // End block
"incw x9, ALL, MUL #2\n"
"cmp x9, x28\n"
"blt 3b\n"
"incw x10, ALL, MUL #2\n"
"cmp x10, x11\n"
"mov x9, #0x0\n"
"mov x27, x26\n"
"blt 3b\n"
".inst 0xd503467f // SMSTOP\n"
:
: [args] "r" (&args), [offsetof_A] "I" (offsetof(KernelArgs, A)), [offsetof_B] "I" (offsetof(KernelArgs, B)), [offsetof_C] "I" (offsetof(KernelArgs, C)), [offsetof_K] "I" (offsetof(KernelArgs, K)), [offsetof_M] "I" (offsetof(KernelArgs, M)), [offsetof_N] "I" (offsetof(KernelArgs, N)), [offsetof_Requantize32_c_offset] "I" (offsetof(Requantize32, c_offset)), [offsetof_Requantize32_maxval] "I" (offsetof(Requantize32, maxval)), [offsetof_Requantize32_minval] "I" (offsetof(Requantize32, minval)), [offsetof_Requantize32_per_channel_muls] "I" (offsetof(Requantize32, per_channel_muls)), [offsetof_Requantize32_per_channel_right_shifts] "I" (offsetof(Requantize32, per_channel_right_shifts)), [offsetof_Requantize32_per_layer_mul] "I" (offsetof(Requantize32, per_layer_mul)), [offsetof_Requantize32_per_layer_right_shift] "I" (offsetof(Requantize32, per_layer_right_shift)), [offsetof_accumulator_buffer] "I" (offsetof(KernelArgs, accumulator_buffer)), [offsetof_bias] "I" (offsetof(KernelArgs, bias)), [offsetof_flags] "I" (offsetof(KernelArgs, flags)), [offsetof_kstride_bytes] "I" (offsetof(KernelArgs, kstride_bytes)), [offsetof_ldcb] "I" (offsetof(KernelArgs, ldcb)), [offsetof_n_0] "I" (offsetof(KernelArgs, n_0)), [rq] "r" (&rq)
: "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
);
}
} // namespace arm_gemm
#endif // ARM_COMPUTE_ENABLE_SME2
#endif // __ARM_FEATURE_SVE
|
