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
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
|
/*
* Copyright (c) 2022-2024 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.
*/
#if defined(ARM_COMPUTE_ENABLE_SME)
#include <cstddef>
namespace arm_conv {
namespace winograd {
namespace output_transform {
void sme_fp32_mopa_4x4_3x3(
unsigned int n_channels,
const float* inptr,
size_t matrix_stride,
const float* bptr,
float* const output,
size_t output_row_stride,
size_t output_col_stride,
float output_min,
float output_max
)
{
// The below assembler uses the Kronecker product and the "vec trick" to
// implement the Winograd output transform (y = AT Y A) using the SME
// array. This code REQUIRES that the vectors are 512b long (or longer, if
// we add some predication).
//
// The "vec trick" uses the identity $vec(AT Y A) = (AT (x) AT) vec(Y)$ to
// convert the chain of matrix multiplications into a matrix-vector
// product. We then stack multiple channels of vec(Y) together to allow us
// to perform multiple channels of the transformation simultaneously.
//
// Since the complete matrix (AT (x) AT) is quite big [16 x 36], we compute
// it on the fly. To do so, we store two representations of the matrix AT.
// The first representation (the outer terms) contains, within each quad,
// four coefficients of the matrix AT.
const float outer_terms[32] = {
1, 1, 1, 1,
0, 1, -1, 2,
0, 1, 1, 4,
0, 1, -1, 8,
// The following rows are continuations of the first four rows, and each
// contains two columns of padding values which aren't used in the
// computation but are there to ensure that the coefficients end up in
// the right quads of the vector into which they're read.
1, 0, 0, 0,
-2, 0, 0, 0,
4, 0, 0, 0,
-8, 1, 0, 0
};
// This should be loaded completely into two Z registers.
//
// We can then use by-element FMLA to construct columns of (AT (x) AT) by
// multiplying elements of the outer terms against the following inner
// terms (again split into quads, but expected to be loaded replicated such
// that each of the six required Z registers contains a repeated quad of
// the values).
const float inner_terms[24] = {
1, 0, 0, 0,
1, 1, 1, 1,
1, -1, 1, -1,
1, 2, 4, 8,
1, -2, 4, -8,
0, 0, 0, 1
};
struct Params
{
const float *outer_terms;
const float *inner_terms;
float act_min;
float act_max;
Params(const float *outer_terms,
const float *inner_terms,
float act_min,
float act_max)
: outer_terms(outer_terms), inner_terms(inner_terms),
act_min(act_min), act_max(act_max)
{
}
};
Params params(outer_terms, inner_terms, output_min, output_max);
__asm__ __volatile__(
"ldr x20, [%x[params], %[offsetof_Params_outer_terms]]\n"
".inst 0xd503477f // SMSTART ZA\n"
"ptrue p5.b\n"
"ld1rw { z12.s }, p5/Z, [%x[params], %[offsetof_Params_act_min]]\n"
"ld1rw { z10.s }, p5/Z, [%x[params], %[offsetof_Params_act_max]]\n"
"pfalse p8.b\n"
"ldr x8, [%x[params], %[offsetof_Params_inner_terms]]\n"
"ld1w { z6.s }, p5/Z, [x20]\n"
"ld1w { z7.s }, p5/Z, [x20, #1, MUL VL]\n"
"ld1rqw { z9.s }, p5/Z, [x8]\n"
"ld1rqw { z8.s }, p5/Z, [x8, #16]\n"
"ld1rqw { z15.s }, p5/Z, [x8, #32]\n"
"fmul z11.s, z9.s, z6.s[0]\n"
"fmul z5.s, z9.s, z6.s[1]\n"
"ld1rqw { z4.s }, p5/Z, [x8, #48]\n"
"ld1rqw { z3.s }, p5/Z, [x8, #64]\n"
"ld1rqw { z2.s }, p5/Z, [x8, #80]\n"
"cbz %x[bptr], 1f\n"
"ptrue p8.s\n"
"1:" // Set bias predicate: Done
".inst 0xc00800ff // zero { zad0, zad1, zad2, zad3, zad4, zad5, zad6, zad7 }\n"
"fmov z1.s, #1.0\n"
"mov x25, #0x0\n"
"cntw x24\n"
"cntw x23, ALL, MUL #2\n"
"cntw x22, ALL, MUL #3\n"
"whilelt p4.s, x25, %x[n_channels]\n"
"whilelt p3.s, x24, %x[n_channels]\n"
"ld1w { z31.s }, p4/Z, [%x[inptr], x25, LSL #2]\n"
"ld1w { z30.s }, p3/Z, [%x[inptr], x24, LSL #2]\n"
"whilelt p2.s, x23, %x[n_channels]\n"
"whilelt p1.s, x22, %x[n_channels]\n"
"ld1w { z29.s }, p2/Z, [%x[inptr], x23, LSL #2]\n"
"add x21, %x[inptr], %x[matrix_stride], LSL #2\n"
"and p0.b, p5/Z, p8.b, p4.b\n"
"ld1w { z28.s }, p1/Z, [%x[inptr], x22, LSL #2]\n"
"ld1w { z27.s }, p4/Z, [x21, x25, LSL #2]\n"
"ld1w { z26.s }, p3/Z, [x21, x24, LSL #2]\n"
"ld1w { z25.s }, p2/Z, [x21, x23, LSL #2]\n"
"ld1w { z24.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
"ld1w { z23.s }, p4/Z, [x21, x25, LSL #2]\n"
"ld1w { z22.s }, p3/Z, [x21, x24, LSL #2]\n"
"ld1w { z21.s }, p2/Z, [x21, x23, LSL #2]\n"
"ld1w { z20.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
"ld1w { z19.s }, p4/Z, [x21, x25, LSL #2]\n"
"ld1w { z18.s }, p3/Z, [x21, x24, LSL #2]\n"
"ld1w { z17.s }, p2/Z, [x21, x23, LSL #2]\n"
"ld1w { z16.s }, p1/Z, [x21, x22, LSL #2]\n"
"ld1w { z0.s }, p0/Z, [%x[bptr], x25, LSL #2]\n"
"and p0.b, p5/Z, p8.b, p3.b\n"
".inst 0x8080b420 // fmopa za0.s, p5/M, p5/M, z1.s, z0.s\n"
"ld1w { z0.s }, p0/Z, [%x[bptr], x24, LSL #2]\n"
"and p0.b, p5/Z, p8.b, p2.b\n"
".inst 0x8080b421 // fmopa za1.s, p5/M, p5/M, z1.s, z0.s\n"
"ld1w { z0.s }, p0/Z, [%x[bptr], x23, LSL #2]\n"
"and p0.b, p5/Z, p8.b, p1.b\n"
".inst 0x8080b422 // fmopa za2.s, p5/M, p5/M, z1.s, z0.s\n"
"ld1w { z0.s }, p0/Z, [%x[bptr], x22, LSL #2]\n"
".inst 0x8080b423 // fmopa za3.s, p5/M, p5/M, z1.s, z0.s\n"
"2:" // Loop
".inst 0x809fb560 // fmopa za0.s, p5/M, p5/M, z11.s, z31.s\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
"mov x14, #0xc\n"
".inst 0x809eb561 // fmopa za1.s, p5/M, p5/M, z11.s, z30.s\n"
"ld1w { z31.s }, p4/Z, [x21, x25, LSL #2]\n"
"whilelt p0.s, x25, %x[n_channels]\n"
"add x20, %x[output], %x[output_col_stride], LSL #2\n"
".inst 0x809db562 // fmopa za2.s, p5/M, p5/M, z11.s, z29.s\n"
"ld1w { z30.s }, p3/Z, [x21, x24, LSL #2]\n"
"add x8, %x[output], %x[output_row_stride], LSL #2\n"
".inst 0x809cb563 // fmopa za3.s, p5/M, p5/M, z11.s, z28.s\n"
"fmul z11.s, z9.s, z6.s[2]\n"
"ld1w { z29.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809bb4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z27.s\n"
"ld1w { z28.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x809ab4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z26.s\n"
"ld1w { z27.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8099b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z25.s\n"
"ld1w { z26.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8098b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z24.s\n"
"fmul z5.s, z9.s, z6.s[3]\n"
"ld1w { z25.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8097b560 // fmopa za0.s, p5/M, p5/M, z11.s, z23.s\n"
"ld1w { z24.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8096b561 // fmopa za1.s, p5/M, p5/M, z11.s, z22.s\n"
"ld1w { z23.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8095b562 // fmopa za2.s, p5/M, p5/M, z11.s, z21.s\n"
"ld1w { z22.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8094b563 // fmopa za3.s, p5/M, p5/M, z11.s, z20.s\n"
"fmul z11.s, z9.s, z7.s[0]\n"
"ld1w { z21.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8093b4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z19.s\n"
"ld1w { z20.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8092b4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z18.s\n"
"ld1w { z19.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8091b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z17.s\n"
"ld1w { z18.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8090b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z16.s\n"
"fmul z5.s, z9.s, z7.s[1]\n"
"ld1w { z17.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809fb560 // fmopa za0.s, p5/M, p5/M, z11.s, z31.s\n"
"ld1w { z16.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x809eb561 // fmopa za1.s, p5/M, p5/M, z11.s, z30.s\n"
"ld1w { z31.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x809db562 // fmopa za2.s, p5/M, p5/M, z11.s, z29.s\n"
"ld1w { z30.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x809cb563 // fmopa za3.s, p5/M, p5/M, z11.s, z28.s\n"
"fmul z11.s, z8.s, z6.s[0]\n"
"ld1w { z29.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809bb4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z27.s\n"
"ld1w { z28.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x809ab4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z26.s\n"
"ld1w { z27.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8099b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z25.s\n"
"ld1w { z26.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8098b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z24.s\n"
"fmul z5.s, z8.s, z6.s[1]\n"
"ld1w { z25.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8097b560 // fmopa za0.s, p5/M, p5/M, z11.s, z23.s\n"
"ld1w { z24.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8096b561 // fmopa za1.s, p5/M, p5/M, z11.s, z22.s\n"
"ld1w { z23.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8095b562 // fmopa za2.s, p5/M, p5/M, z11.s, z21.s\n"
"ld1w { z22.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8094b563 // fmopa za3.s, p5/M, p5/M, z11.s, z20.s\n"
"fmul z11.s, z8.s, z6.s[2]\n"
"ld1w { z21.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8093b4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z19.s\n"
"ld1w { z20.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8092b4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z18.s\n"
"ld1w { z19.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8091b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z17.s\n"
"ld1w { z18.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8090b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z16.s\n"
"fmul z5.s, z8.s, z6.s[3]\n"
"ld1w { z17.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809fb560 // fmopa za0.s, p5/M, p5/M, z11.s, z31.s\n"
"ld1w { z16.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x809eb561 // fmopa za1.s, p5/M, p5/M, z11.s, z30.s\n"
"ld1w { z31.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x809db562 // fmopa za2.s, p5/M, p5/M, z11.s, z29.s\n"
"ld1w { z30.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x809cb563 // fmopa za3.s, p5/M, p5/M, z11.s, z28.s\n"
"fmul z11.s, z8.s, z7.s[0]\n"
"ld1w { z29.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809bb4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z27.s\n"
"ld1w { z28.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x809ab4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z26.s\n"
"ld1w { z27.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8099b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z25.s\n"
"ld1w { z26.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8098b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z24.s\n"
"fmul z5.s, z8.s, z7.s[1]\n"
"ld1w { z25.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8097b560 // fmopa za0.s, p5/M, p5/M, z11.s, z23.s\n"
"ld1w { z24.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8096b561 // fmopa za1.s, p5/M, p5/M, z11.s, z22.s\n"
"ld1w { z23.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8095b562 // fmopa za2.s, p5/M, p5/M, z11.s, z21.s\n"
"ld1w { z22.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8094b563 // fmopa za3.s, p5/M, p5/M, z11.s, z20.s\n"
"fmul z11.s, z15.s, z6.s[0]\n"
"ld1w { z21.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8093b4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z19.s\n"
"ld1w { z20.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8092b4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z18.s\n"
"ld1w { z19.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8091b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z17.s\n"
"ld1w { z18.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8090b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z16.s\n"
"fmul z5.s, z15.s, z6.s[1]\n"
"ld1w { z17.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809fb560 // fmopa za0.s, p5/M, p5/M, z11.s, z31.s\n"
"ld1w { z16.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x809eb561 // fmopa za1.s, p5/M, p5/M, z11.s, z30.s\n"
"ld1w { z31.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x809db562 // fmopa za2.s, p5/M, p5/M, z11.s, z29.s\n"
"ld1w { z30.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x809cb563 // fmopa za3.s, p5/M, p5/M, z11.s, z28.s\n"
"fmul z11.s, z15.s, z6.s[2]\n"
"ld1w { z29.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809bb4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z27.s\n"
"ld1w { z28.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x809ab4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z26.s\n"
"ld1w { z27.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8099b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z25.s\n"
"ld1w { z26.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8098b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z24.s\n"
"fmul z5.s, z15.s, z6.s[3]\n"
"ld1w { z25.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8097b560 // fmopa za0.s, p5/M, p5/M, z11.s, z23.s\n"
"ld1w { z24.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8096b561 // fmopa za1.s, p5/M, p5/M, z11.s, z22.s\n"
"ld1w { z23.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8095b562 // fmopa za2.s, p5/M, p5/M, z11.s, z21.s\n"
"ld1w { z22.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8094b563 // fmopa za3.s, p5/M, p5/M, z11.s, z20.s\n"
"fmul z11.s, z15.s, z7.s[0]\n"
"ld1w { z21.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8093b4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z19.s\n"
"ld1w { z20.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8092b4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z18.s\n"
"ld1w { z19.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8091b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z17.s\n"
"ld1w { z18.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8090b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z16.s\n"
"fmul z5.s, z15.s, z7.s[1]\n"
"ld1w { z17.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809fb560 // fmopa za0.s, p5/M, p5/M, z11.s, z31.s\n"
"ld1w { z16.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x809eb561 // fmopa za1.s, p5/M, p5/M, z11.s, z30.s\n"
"ld1w { z31.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x809db562 // fmopa za2.s, p5/M, p5/M, z11.s, z29.s\n"
"ld1w { z30.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x809cb563 // fmopa za3.s, p5/M, p5/M, z11.s, z28.s\n"
"fmul z11.s, z4.s, z6.s[0]\n"
"ld1w { z29.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809bb4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z27.s\n"
"ld1w { z28.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x809ab4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z26.s\n"
"ld1w { z27.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8099b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z25.s\n"
"ld1w { z26.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8098b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z24.s\n"
"fmul z5.s, z4.s, z6.s[1]\n"
"ld1w { z25.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8097b560 // fmopa za0.s, p5/M, p5/M, z11.s, z23.s\n"
"ld1w { z24.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8096b561 // fmopa za1.s, p5/M, p5/M, z11.s, z22.s\n"
"ld1w { z23.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8095b562 // fmopa za2.s, p5/M, p5/M, z11.s, z21.s\n"
"ld1w { z22.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8094b563 // fmopa za3.s, p5/M, p5/M, z11.s, z20.s\n"
"fmul z11.s, z4.s, z6.s[2]\n"
"ld1w { z21.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8093b4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z19.s\n"
"ld1w { z20.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8092b4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z18.s\n"
"ld1w { z19.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8091b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z17.s\n"
"ld1w { z18.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8090b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z16.s\n"
"fmul z5.s, z4.s, z6.s[3]\n"
"ld1w { z17.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809fb560 // fmopa za0.s, p5/M, p5/M, z11.s, z31.s\n"
"ld1w { z16.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x809eb561 // fmopa za1.s, p5/M, p5/M, z11.s, z30.s\n"
"ld1w { z31.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x809db562 // fmopa za2.s, p5/M, p5/M, z11.s, z29.s\n"
"ld1w { z30.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x809cb563 // fmopa za3.s, p5/M, p5/M, z11.s, z28.s\n"
"fmul z11.s, z4.s, z7.s[0]\n"
"ld1w { z29.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809bb4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z27.s\n"
"ld1w { z28.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x809ab4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z26.s\n"
"ld1w { z27.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8099b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z25.s\n"
"ld1w { z26.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8098b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z24.s\n"
"fmul z5.s, z4.s, z7.s[1]\n"
"ld1w { z25.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8097b560 // fmopa za0.s, p5/M, p5/M, z11.s, z23.s\n"
"ld1w { z24.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8096b561 // fmopa za1.s, p5/M, p5/M, z11.s, z22.s\n"
"ld1w { z23.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8095b562 // fmopa za2.s, p5/M, p5/M, z11.s, z21.s\n"
"ld1w { z22.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8094b563 // fmopa za3.s, p5/M, p5/M, z11.s, z20.s\n"
"fmul z11.s, z3.s, z6.s[0]\n"
"ld1w { z21.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8093b4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z19.s\n"
"ld1w { z20.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8092b4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z18.s\n"
"ld1w { z19.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8091b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z17.s\n"
"ld1w { z18.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8090b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z16.s\n"
"fmul z5.s, z3.s, z6.s[1]\n"
"ld1w { z17.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809fb560 // fmopa za0.s, p5/M, p5/M, z11.s, z31.s\n"
"ld1w { z16.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x809eb561 // fmopa za1.s, p5/M, p5/M, z11.s, z30.s\n"
"ld1w { z31.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x809db562 // fmopa za2.s, p5/M, p5/M, z11.s, z29.s\n"
"ld1w { z30.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x809cb563 // fmopa za3.s, p5/M, p5/M, z11.s, z28.s\n"
"fmul z11.s, z3.s, z6.s[2]\n"
"ld1w { z29.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809bb4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z27.s\n"
"ld1w { z28.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x809ab4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z26.s\n"
"ld1w { z27.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8099b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z25.s\n"
"ld1w { z26.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8098b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z24.s\n"
"fmul z5.s, z3.s, z6.s[3]\n"
"ld1w { z25.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8097b560 // fmopa za0.s, p5/M, p5/M, z11.s, z23.s\n"
"ld1w { z24.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8096b561 // fmopa za1.s, p5/M, p5/M, z11.s, z22.s\n"
"ld1w { z23.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8095b562 // fmopa za2.s, p5/M, p5/M, z11.s, z21.s\n"
"ld1w { z22.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8094b563 // fmopa za3.s, p5/M, p5/M, z11.s, z20.s\n"
"fmul z11.s, z3.s, z7.s[0]\n"
"ld1w { z21.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8093b4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z19.s\n"
"ld1w { z20.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8092b4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z18.s\n"
"ld1w { z19.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8091b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z17.s\n"
"ld1w { z18.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8090b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z16.s\n"
"fmul z5.s, z3.s, z7.s[1]\n"
"ld1w { z17.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809fb560 // fmopa za0.s, p5/M, p5/M, z11.s, z31.s\n"
"ld1w { z16.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x809eb561 // fmopa za1.s, p5/M, p5/M, z11.s, z30.s\n"
"ld1w { z31.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x809db562 // fmopa za2.s, p5/M, p5/M, z11.s, z29.s\n"
"ld1w { z30.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x809cb563 // fmopa za3.s, p5/M, p5/M, z11.s, z28.s\n"
"fmul z11.s, z2.s, z6.s[0]\n"
"ld1w { z29.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809bb4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z27.s\n"
"ld1w { z28.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x809ab4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z26.s\n"
"ld1w { z27.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8099b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z25.s\n"
"ld1w { z26.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8098b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z24.s\n"
"fmul z5.s, z2.s, z6.s[1]\n"
"ld1w { z25.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8097b560 // fmopa za0.s, p5/M, p5/M, z11.s, z23.s\n"
"ld1w { z24.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8096b561 // fmopa za1.s, p5/M, p5/M, z11.s, z22.s\n"
"ld1w { z23.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8095b562 // fmopa za2.s, p5/M, p5/M, z11.s, z21.s\n"
"ld1w { z22.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8094b563 // fmopa za3.s, p5/M, p5/M, z11.s, z20.s\n"
"fmul z11.s, z2.s, z6.s[2]\n"
"ld1w { z21.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x8093b4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z19.s\n"
"ld1w { z20.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
".inst 0x8092b4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z18.s\n"
"ld1w { z19.s }, p4/Z, [x21, x25, LSL #2]\n"
".inst 0x8091b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z17.s\n"
"ld1w { z18.s }, p3/Z, [x21, x24, LSL #2]\n"
".inst 0x8090b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z16.s\n"
"fmul z5.s, z2.s, z6.s[3]\n"
"ld1w { z17.s }, p2/Z, [x21, x23, LSL #2]\n"
".inst 0x809fb560 // fmopa za0.s, p5/M, p5/M, z11.s, z31.s\n"
"ld1w { z16.s }, p1/Z, [x21, x22, LSL #2]\n"
".inst 0x809eb561 // fmopa za1.s, p5/M, p5/M, z11.s, z30.s\n"
".inst 0x809db562 // fmopa za2.s, p5/M, p5/M, z11.s, z29.s\n"
".inst 0x809cb563 // fmopa za3.s, p5/M, p5/M, z11.s, z28.s\n"
"fmul z11.s, z2.s, z7.s[0]\n"
".inst 0x809bb4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z27.s\n"
".inst 0x809ab4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z26.s\n"
".inst 0x8099b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z25.s\n"
".inst 0x8098b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z24.s\n"
"fmul z5.s, z2.s, z7.s[1]\n"
".inst 0x8097b560 // fmopa za0.s, p5/M, p5/M, z11.s, z23.s\n"
".inst 0x8096b561 // fmopa za1.s, p5/M, p5/M, z11.s, z22.s\n"
".inst 0x8095b562 // fmopa za2.s, p5/M, p5/M, z11.s, z21.s\n"
".inst 0x8094b563 // fmopa za3.s, p5/M, p5/M, z11.s, z20.s\n"
"fmul z11.s, z9.s, z6.s[0]\n"
".inst 0x8093b4a0 // fmopa za0.s, p5/M, p5/M, z5.s, z19.s\n"
".inst 0x8092b4a1 // fmopa za1.s, p5/M, p5/M, z5.s, z18.s\n"
".inst 0x8091b4a2 // fmopa za2.s, p5/M, p5/M, z5.s, z17.s\n"
".inst 0x8090b4a3 // fmopa za3.s, p5/M, p5/M, z5.s, z16.s\n"
"fmul z5.s, z9.s, z6.s[1]\n"
".inst 0xc082741f // mova z31.s, p5/M, za0h.s[XZR]\n"
".inst 0xc082541c // mova z28.s, p5/M, za0h.s[x14]\n"
"fmin z31.s, p5/M, z31.s, z10.s\n"
".inst 0xc082743b // mova z27.s, p5/M, za0h.s[XZR, #1]\n"
"fmin z28.s, p5/M, z28.s, z10.s\n"
".inst 0xc0825438 // mova z24.s, p5/M, za0h.s[x14, #1]\n"
"fmin z27.s, p5/M, z27.s, z10.s\n"
"mov x13, #0x4\n"
"mov x12, #0x8\n"
".inst 0xc082341e // mova z30.s, p5/M, za0h.s[x13]\n"
"fmin z24.s, p5/M, z24.s, z10.s\n"
".inst 0xc082141d // mova z29.s, p5/M, za0h.s[x12]\n"
"fmax z31.s, p5/M, z31.s, z12.s\n"
"fmin z30.s, p5/M, z30.s, z10.s\n"
".inst 0xc082343a // mova z26.s, p5/M, za0h.s[x13, #1]\n"
"fmin z29.s, p5/M, z29.s, z10.s\n"
"fmax z28.s, p5/M, z28.s, z12.s\n"
".inst 0xc0821439 // mova z25.s, p5/M, za0h.s[x12, #1]\n"
"fmax z27.s, p5/M, z27.s, z12.s\n"
"fmin z26.s, p5/M, z26.s, z10.s\n"
".inst 0xc0827457 // mova z23.s, p5/M, za0h.s[XZR, #2]\n"
"fmin z25.s, p5/M, z25.s, z10.s\n"
"fmax z24.s, p5/M, z24.s, z12.s\n"
".inst 0xc0823456 // mova z22.s, p5/M, za0h.s[x13, #2]\n"
"fmax z30.s, p5/M, z30.s, z12.s\n"
"fmin z23.s, p5/M, z23.s, z10.s\n"
".inst 0xc0821455 // mova z21.s, p5/M, za0h.s[x12, #2]\n"
"fmax z29.s, p5/M, z29.s, z12.s\n"
"fmin z22.s, p5/M, z22.s, z10.s\n"
".inst 0xc0825454 // mova z20.s, p5/M, za0h.s[x14, #2]\n"
"fmax z26.s, p5/M, z26.s, z12.s\n"
"fmin z21.s, p5/M, z21.s, z10.s\n"
".inst 0xc0827473 // mova z19.s, p5/M, za0h.s[XZR, #3]\n"
"fmax z25.s, p5/M, z25.s, z12.s\n"
"fmin z20.s, p5/M, z20.s, z10.s\n"
".inst 0xc0823472 // mova z18.s, p5/M, za0h.s[x13, #3]\n"
"fmax z23.s, p5/M, z23.s, z12.s\n"
"fmin z19.s, p5/M, z19.s, z10.s\n"
".inst 0xc0821471 // mova z17.s, p5/M, za0h.s[x12, #3]\n"
"fmax z22.s, p5/M, z22.s, z12.s\n"
"fmin z18.s, p5/M, z18.s, z10.s\n"
".inst 0xc0825470 // mova z16.s, p5/M, za0h.s[x14, #3]\n"
"fmax z21.s, p5/M, z21.s, z12.s\n"
"fmin z17.s, p5/M, z17.s, z10.s\n"
"fmax z20.s, p5/M, z20.s, z12.s\n"
"fmin z16.s, p5/M, z16.s, z10.s\n"
"st1w { z31.s }, p0, [%x[output], x25, LSL #2]\n"
"fmax z19.s, p5/M, z19.s, z12.s\n"
"st1w { z30.s }, p0, [x20, x25, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"fmax z18.s, p5/M, z18.s, z12.s\n"
"st1w { z29.s }, p0, [x20, x25, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"fmax z17.s, p5/M, z17.s, z12.s\n"
"st1w { z28.s }, p0, [x20, x25, LSL #2]\n"
"add x20, x8, %x[output_col_stride], LSL #2\n"
"fmax z16.s, p5/M, z16.s, z12.s\n"
"st1w { z27.s }, p0, [x8, x25, LSL #2]\n"
"add x8, x8, %x[output_row_stride], LSL #2\n"
"st1w { z26.s }, p0, [x20, x25, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z25.s }, p0, [x20, x25, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z24.s }, p0, [x20, x25, LSL #2]\n"
"add x20, x8, %x[output_col_stride], LSL #2\n"
"st1w { z23.s }, p0, [x8, x25, LSL #2]\n"
"add x8, x8, %x[output_row_stride], LSL #2\n"
"st1w { z22.s }, p0, [x20, x25, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z21.s }, p0, [x20, x25, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z20.s }, p0, [x20, x25, LSL #2]\n"
"add x20, x8, %x[output_col_stride], LSL #2\n"
"st1w { z19.s }, p0, [x8, x25, LSL #2]\n"
"st1w { z18.s }, p0, [x20, x25, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z17.s }, p0, [x20, x25, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z16.s }, p0, [x20, x25, LSL #2]\n"
"whilelt p0.s, x24, %x[n_channels]\n"
"b.none 3f\n"
".inst 0xc082749f // mova z31.s, p5/M, za1h.s[XZR]\n"
".inst 0xc082349e // mova z30.s, p5/M, za1h.s[x13]\n"
"fmin z31.s, p5/M, z31.s, z10.s\n"
".inst 0xc082149d // mova z29.s, p5/M, za1h.s[x12]\n"
"fmin z30.s, p5/M, z30.s, z10.s\n"
".inst 0xc082549c // mova z28.s, p5/M, za1h.s[x14]\n"
"fmin z29.s, p5/M, z29.s, z10.s\n"
".inst 0xc08274bb // mova z27.s, p5/M, za1h.s[XZR, #1]\n"
"fmin z28.s, p5/M, z28.s, z10.s\n"
".inst 0xc08234ba // mova z26.s, p5/M, za1h.s[x13, #1]\n"
"fmax z31.s, p5/M, z31.s, z12.s\n"
"fmin z27.s, p5/M, z27.s, z10.s\n"
".inst 0xc08214b9 // mova z25.s, p5/M, za1h.s[x12, #1]\n"
"fmax z30.s, p5/M, z30.s, z12.s\n"
"fmin z26.s, p5/M, z26.s, z10.s\n"
".inst 0xc08254b8 // mova z24.s, p5/M, za1h.s[x14, #1]\n"
"fmax z29.s, p5/M, z29.s, z12.s\n"
"fmin z25.s, p5/M, z25.s, z10.s\n"
".inst 0xc08274d7 // mova z23.s, p5/M, za1h.s[XZR, #2]\n"
"fmax z28.s, p5/M, z28.s, z12.s\n"
"fmin z24.s, p5/M, z24.s, z10.s\n"
".inst 0xc08234d6 // mova z22.s, p5/M, za1h.s[x13, #2]\n"
"fmax z27.s, p5/M, z27.s, z12.s\n"
"fmin z23.s, p5/M, z23.s, z10.s\n"
".inst 0xc08214d5 // mova z21.s, p5/M, za1h.s[x12, #2]\n"
"fmax z26.s, p5/M, z26.s, z12.s\n"
"fmin z22.s, p5/M, z22.s, z10.s\n"
"add x20, %x[output], %x[output_col_stride], LSL #2\n"
".inst 0xc08254d4 // mova z20.s, p5/M, za1h.s[x14, #2]\n"
"fmax z25.s, p5/M, z25.s, z12.s\n"
"fmin z21.s, p5/M, z21.s, z10.s\n"
"add x8, %x[output], %x[output_row_stride], LSL #2\n"
".inst 0xc08274f3 // mova z19.s, p5/M, za1h.s[XZR, #3]\n"
"fmax z24.s, p5/M, z24.s, z12.s\n"
"fmin z20.s, p5/M, z20.s, z10.s\n"
".inst 0xc08234f2 // mova z18.s, p5/M, za1h.s[x13, #3]\n"
"fmax z23.s, p5/M, z23.s, z12.s\n"
"fmin z19.s, p5/M, z19.s, z10.s\n"
".inst 0xc08214f1 // mova z17.s, p5/M, za1h.s[x12, #3]\n"
"fmax z22.s, p5/M, z22.s, z12.s\n"
"fmin z18.s, p5/M, z18.s, z10.s\n"
".inst 0xc08254f0 // mova z16.s, p5/M, za1h.s[x14, #3]\n"
"fmax z21.s, p5/M, z21.s, z12.s\n"
"fmin z17.s, p5/M, z17.s, z10.s\n"
"fmax z20.s, p5/M, z20.s, z12.s\n"
"fmin z16.s, p5/M, z16.s, z10.s\n"
"st1w { z31.s }, p0, [%x[output], x24, LSL #2]\n"
"fmax z19.s, p5/M, z19.s, z12.s\n"
"st1w { z30.s }, p0, [x20, x24, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"fmax z18.s, p5/M, z18.s, z12.s\n"
"st1w { z29.s }, p0, [x20, x24, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"fmax z17.s, p5/M, z17.s, z12.s\n"
"st1w { z28.s }, p0, [x20, x24, LSL #2]\n"
"add x20, x8, %x[output_col_stride], LSL #2\n"
"fmax z16.s, p5/M, z16.s, z12.s\n"
"st1w { z27.s }, p0, [x8, x24, LSL #2]\n"
"add x8, x8, %x[output_row_stride], LSL #2\n"
"st1w { z26.s }, p0, [x20, x24, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z25.s }, p0, [x20, x24, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z24.s }, p0, [x20, x24, LSL #2]\n"
"add x20, x8, %x[output_col_stride], LSL #2\n"
"st1w { z23.s }, p0, [x8, x24, LSL #2]\n"
"add x8, x8, %x[output_row_stride], LSL #2\n"
"st1w { z22.s }, p0, [x20, x24, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z21.s }, p0, [x20, x24, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z20.s }, p0, [x20, x24, LSL #2]\n"
"add x20, x8, %x[output_col_stride], LSL #2\n"
"st1w { z19.s }, p0, [x8, x24, LSL #2]\n"
"st1w { z18.s }, p0, [x20, x24, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z17.s }, p0, [x20, x24, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z16.s }, p0, [x20, x24, LSL #2]\n"
"whilelt p0.s, x23, %x[n_channels]\n"
"b.none 3f\n"
".inst 0xc082751f // mova z31.s, p5/M, za2h.s[XZR]\n"
".inst 0xc082351e // mova z30.s, p5/M, za2h.s[x13]\n"
"fmin z31.s, p5/M, z31.s, z10.s\n"
".inst 0xc082151d // mova z29.s, p5/M, za2h.s[x12]\n"
"fmin z30.s, p5/M, z30.s, z10.s\n"
".inst 0xc082551c // mova z28.s, p5/M, za2h.s[x14]\n"
"fmin z29.s, p5/M, z29.s, z10.s\n"
".inst 0xc082753b // mova z27.s, p5/M, za2h.s[XZR, #1]\n"
"fmin z28.s, p5/M, z28.s, z10.s\n"
".inst 0xc082353a // mova z26.s, p5/M, za2h.s[x13, #1]\n"
"fmax z31.s, p5/M, z31.s, z12.s\n"
"fmin z27.s, p5/M, z27.s, z10.s\n"
".inst 0xc0821539 // mova z25.s, p5/M, za2h.s[x12, #1]\n"
"fmax z30.s, p5/M, z30.s, z12.s\n"
"fmin z26.s, p5/M, z26.s, z10.s\n"
".inst 0xc0825538 // mova z24.s, p5/M, za2h.s[x14, #1]\n"
"fmax z29.s, p5/M, z29.s, z12.s\n"
"fmin z25.s, p5/M, z25.s, z10.s\n"
".inst 0xc0827557 // mova z23.s, p5/M, za2h.s[XZR, #2]\n"
"fmax z28.s, p5/M, z28.s, z12.s\n"
"fmin z24.s, p5/M, z24.s, z10.s\n"
".inst 0xc0823556 // mova z22.s, p5/M, za2h.s[x13, #2]\n"
"fmax z27.s, p5/M, z27.s, z12.s\n"
"fmin z23.s, p5/M, z23.s, z10.s\n"
".inst 0xc0821555 // mova z21.s, p5/M, za2h.s[x12, #2]\n"
"fmax z26.s, p5/M, z26.s, z12.s\n"
"fmin z22.s, p5/M, z22.s, z10.s\n"
"add x20, %x[output], %x[output_col_stride], LSL #2\n"
".inst 0xc0825554 // mova z20.s, p5/M, za2h.s[x14, #2]\n"
"fmax z25.s, p5/M, z25.s, z12.s\n"
"fmin z21.s, p5/M, z21.s, z10.s\n"
"add x8, %x[output], %x[output_row_stride], LSL #2\n"
".inst 0xc0827573 // mova z19.s, p5/M, za2h.s[XZR, #3]\n"
"fmax z24.s, p5/M, z24.s, z12.s\n"
"fmin z20.s, p5/M, z20.s, z10.s\n"
".inst 0xc0823572 // mova z18.s, p5/M, za2h.s[x13, #3]\n"
"fmax z23.s, p5/M, z23.s, z12.s\n"
"fmin z19.s, p5/M, z19.s, z10.s\n"
".inst 0xc0821571 // mova z17.s, p5/M, za2h.s[x12, #3]\n"
"fmax z22.s, p5/M, z22.s, z12.s\n"
"fmin z18.s, p5/M, z18.s, z10.s\n"
".inst 0xc0825570 // mova z16.s, p5/M, za2h.s[x14, #3]\n"
"fmax z21.s, p5/M, z21.s, z12.s\n"
"fmin z17.s, p5/M, z17.s, z10.s\n"
"fmax z20.s, p5/M, z20.s, z12.s\n"
"fmin z16.s, p5/M, z16.s, z10.s\n"
"st1w { z31.s }, p0, [%x[output], x23, LSL #2]\n"
"fmax z19.s, p5/M, z19.s, z12.s\n"
"st1w { z30.s }, p0, [x20, x23, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"fmax z18.s, p5/M, z18.s, z12.s\n"
"st1w { z29.s }, p0, [x20, x23, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"fmax z17.s, p5/M, z17.s, z12.s\n"
"st1w { z28.s }, p0, [x20, x23, LSL #2]\n"
"add x20, x8, %x[output_col_stride], LSL #2\n"
"fmax z16.s, p5/M, z16.s, z12.s\n"
"st1w { z27.s }, p0, [x8, x23, LSL #2]\n"
"add x8, x8, %x[output_row_stride], LSL #2\n"
"st1w { z26.s }, p0, [x20, x23, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z25.s }, p0, [x20, x23, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z24.s }, p0, [x20, x23, LSL #2]\n"
"add x20, x8, %x[output_col_stride], LSL #2\n"
"st1w { z23.s }, p0, [x8, x23, LSL #2]\n"
"add x8, x8, %x[output_row_stride], LSL #2\n"
"st1w { z22.s }, p0, [x20, x23, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z21.s }, p0, [x20, x23, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z20.s }, p0, [x20, x23, LSL #2]\n"
"add x20, x8, %x[output_col_stride], LSL #2\n"
"st1w { z19.s }, p0, [x8, x23, LSL #2]\n"
"st1w { z18.s }, p0, [x20, x23, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z17.s }, p0, [x20, x23, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z16.s }, p0, [x20, x23, LSL #2]\n"
"whilelt p0.s, x22, %x[n_channels]\n"
"b.none 3f\n"
"fmov z1.s, #1.0\n"
".inst 0xc082759f // mova z31.s, p5/M, za3h.s[XZR]\n"
".inst 0xc082359e // mova z30.s, p5/M, za3h.s[x13]\n"
"fmin z31.s, p5/M, z31.s, z10.s\n"
".inst 0xc082159d // mova z29.s, p5/M, za3h.s[x12]\n"
"fmin z30.s, p5/M, z30.s, z10.s\n"
".inst 0xc082559c // mova z28.s, p5/M, za3h.s[x14]\n"
"fmin z29.s, p5/M, z29.s, z10.s\n"
".inst 0xc08275bb // mova z27.s, p5/M, za3h.s[XZR, #1]\n"
"fmin z28.s, p5/M, z28.s, z10.s\n"
".inst 0xc08235ba // mova z26.s, p5/M, za3h.s[x13, #1]\n"
"fmax z31.s, p5/M, z31.s, z12.s\n"
"fmin z27.s, p5/M, z27.s, z10.s\n"
".inst 0xc08215b9 // mova z25.s, p5/M, za3h.s[x12, #1]\n"
"fmax z30.s, p5/M, z30.s, z12.s\n"
"fmin z26.s, p5/M, z26.s, z10.s\n"
".inst 0xc08255b8 // mova z24.s, p5/M, za3h.s[x14, #1]\n"
"fmax z29.s, p5/M, z29.s, z12.s\n"
"fmin z25.s, p5/M, z25.s, z10.s\n"
".inst 0xc08275d7 // mova z23.s, p5/M, za3h.s[XZR, #2]\n"
"fmax z28.s, p5/M, z28.s, z12.s\n"
"fmin z24.s, p5/M, z24.s, z10.s\n"
".inst 0xc08235d6 // mova z22.s, p5/M, za3h.s[x13, #2]\n"
"fmax z27.s, p5/M, z27.s, z12.s\n"
"fmin z23.s, p5/M, z23.s, z10.s\n"
".inst 0xc08215d5 // mova z21.s, p5/M, za3h.s[x12, #2]\n"
"fmax z26.s, p5/M, z26.s, z12.s\n"
"fmin z22.s, p5/M, z22.s, z10.s\n"
".inst 0xc08255d4 // mova z20.s, p5/M, za3h.s[x14, #2]\n"
"fmax z25.s, p5/M, z25.s, z12.s\n"
"fmin z21.s, p5/M, z21.s, z10.s\n"
"add x20, %x[output], %x[output_col_stride], LSL #2\n"
".inst 0xc08275f3 // mova z19.s, p5/M, za3h.s[XZR, #3]\n"
"fmax z24.s, p5/M, z24.s, z12.s\n"
"fmin z20.s, p5/M, z20.s, z10.s\n"
"add x8, %x[output], %x[output_row_stride], LSL #2\n"
".inst 0xc08235f2 // mova z18.s, p5/M, za3h.s[x13, #3]\n"
"fmax z23.s, p5/M, z23.s, z12.s\n"
"fmin z19.s, p5/M, z19.s, z10.s\n"
"incw x25, ALL, MUL #4\n"
".inst 0xc08215f1 // mova z17.s, p5/M, za3h.s[x12, #3]\n"
"fmax z22.s, p5/M, z22.s, z12.s\n"
"fmin z18.s, p5/M, z18.s, z10.s\n"
"incw x24, ALL, MUL #4\n"
".inst 0xc08255f0 // mova z16.s, p5/M, za3h.s[x14, #3]\n"
"fmax z21.s, p5/M, z21.s, z12.s\n"
"fmin z17.s, p5/M, z17.s, z10.s\n"
"incw x23, ALL, MUL #4\n"
".inst 0xc00800ff // zero { zad0, zad1, zad2, zad3, zad4, zad5, zad6, zad7 }\n"
"fmax z20.s, p5/M, z20.s, z12.s\n"
"fmin z16.s, p5/M, z16.s, z10.s\n"
"add x21, %x[inptr], %x[matrix_stride], LSL #2\n"
"fmax z19.s, p5/M, z19.s, z12.s\n"
"st1w { z31.s }, p0, [%x[output], x22, LSL #2]\n"
"fmax z18.s, p5/M, z18.s, z12.s\n"
"st1w { z30.s }, p0, [x20, x22, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"fmax z17.s, p5/M, z17.s, z12.s\n"
"st1w { z29.s }, p0, [x20, x22, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"fmax z16.s, p5/M, z16.s, z12.s\n"
"st1w { z28.s }, p0, [x20, x22, LSL #2]\n"
"add x20, x8, %x[output_col_stride], LSL #2\n"
"st1w { z27.s }, p0, [x8, x22, LSL #2]\n"
"add x8, x8, %x[output_row_stride], LSL #2\n"
"st1w { z26.s }, p0, [x20, x22, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z25.s }, p0, [x20, x22, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z24.s }, p0, [x20, x22, LSL #2]\n"
"add x20, x8, %x[output_col_stride], LSL #2\n"
"st1w { z23.s }, p0, [x8, x22, LSL #2]\n"
"add x8, x8, %x[output_row_stride], LSL #2\n"
"st1w { z22.s }, p0, [x20, x22, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z21.s }, p0, [x20, x22, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z20.s }, p0, [x20, x22, LSL #2]\n"
"add x20, x8, %x[output_col_stride], LSL #2\n"
"st1w { z19.s }, p0, [x8, x22, LSL #2]\n"
"st1w { z18.s }, p0, [x20, x22, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z17.s }, p0, [x20, x22, LSL #2]\n"
"add x20, x20, %x[output_col_stride], LSL #2\n"
"st1w { z16.s }, p0, [x20, x22, LSL #2]\n"
"incw x22, ALL, MUL #4\n"
"whilelt p1.s, x22, %x[n_channels]\n"
"ld1w { z28.s }, p1/Z, [%x[inptr], x22, LSL #2]\n"
"ld1w { z24.s }, p1/Z, [x21, x22, LSL #2]\n"
"whilelt p2.s, x23, %x[n_channels]\n"
"whilelt p3.s, x24, %x[n_channels]\n"
"ld1w { z30.s }, p3/Z, [%x[inptr], x24, LSL #2]\n"
"whilelt p4.s, x25, %x[n_channels]\n"
"ld1w { z31.s }, p4/Z, [%x[inptr], x25, LSL #2]\n"
"and p0.b, p5/Z, p8.b, p4.b\n"
"ld1w { z29.s }, p2/Z, [%x[inptr], x23, LSL #2]\n"
"ld1w { z27.s }, p4/Z, [x21, x25, LSL #2]\n"
"ld1w { z26.s }, p3/Z, [x21, x24, LSL #2]\n"
"ld1w { z25.s }, p2/Z, [x21, x23, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
"ld1w { z23.s }, p4/Z, [x21, x25, LSL #2]\n"
"ld1w { z22.s }, p3/Z, [x21, x24, LSL #2]\n"
"ld1w { z21.s }, p2/Z, [x21, x23, LSL #2]\n"
"ld1w { z20.s }, p1/Z, [x21, x22, LSL #2]\n"
"add x21, x21, %x[matrix_stride], LSL #2\n"
"ld1w { z19.s }, p4/Z, [x21, x25, LSL #2]\n"
"ld1w { z18.s }, p3/Z, [x21, x24, LSL #2]\n"
"ld1w { z17.s }, p2/Z, [x21, x23, LSL #2]\n"
"ld1w { z16.s }, p1/Z, [x21, x22, LSL #2]\n"
"ld1w { z0.s }, p0/Z, [%x[bptr], x25, LSL #2]\n"
"and p0.b, p5/Z, p8.b, p3.b\n"
".inst 0x8080b420 // fmopa za0.s, p5/M, p5/M, z1.s, z0.s\n"
"ld1w { z0.s }, p0/Z, [%x[bptr], x24, LSL #2]\n"
"and p0.b, p5/Z, p8.b, p2.b\n"
".inst 0x8080b421 // fmopa za1.s, p5/M, p5/M, z1.s, z0.s\n"
"ld1w { z0.s }, p0/Z, [%x[bptr], x23, LSL #2]\n"
"and p0.b, p5/Z, p8.b, p1.b\n"
".inst 0x8080b422 // fmopa za2.s, p5/M, p5/M, z1.s, z0.s\n"
"ld1w { z0.s }, p0/Z, [%x[bptr], x22, LSL #2]\n"
".inst 0x8080b423 // fmopa za3.s, p5/M, p5/M, z1.s, z0.s\n"
"b.any 2b\n"
"3:" // End
".inst 0xd503467f // SMSTOP\n"
:
: [bptr] "r" (bptr), [inptr] "r" (inptr), [matrix_stride] "r" (matrix_stride), [n_channels] "r" (n_channels), [offsetof_Params_act_max] "I" (offsetof(Params, act_max)), [offsetof_Params_act_min] "I" (offsetof(Params, act_min)), [offsetof_Params_inner_terms] "I" (offsetof(Params, inner_terms)), [offsetof_Params_outer_terms] "I" (offsetof(Params, outer_terms)), [output] "r" (output), [output_col_stride] "r" (output_col_stride), [output_row_stride] "r" (output_row_stride), [params] "r" (¶ms)
: "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p8", "x12", "x13", "x14", "x8", "x20", "x21", "x22", "x23", "x24", "x25", "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 output_transform
} // namespace winograd
} // namespace arm_conv
#endif // defined(ARM_COMPUTE_ENABLE_SME)
|
