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
|
/*
* 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_SME2)
#include <algorithm>
#include <cstddef>
#include "arm_gemm.hpp"
using arm_gemm::Requantize32;
namespace arm_conv {
namespace depthwise {
void sme2_s8q_planar_3x3_s2_4rows_dot_za_impl(
const int8_t *inptr,
size_t ld_in_row,
size_t ld_in_col,
size_t ld_in_vl,
unsigned int pad_top,
unsigned int valid_input_rows,
unsigned int pad_left,
unsigned int valid_input_cols,
const int8_t *weights,
int8_t **outptrs,
const size_t *outlds,
const size_t *outvllds,
unsigned int output_cols,
unsigned int start_channel,
unsigned int valid_channels,
const arm_gemm::Requantize32 &qp
)
{
struct Args
{
const int8_t *inptr;
size_t ld_in_vl;
long unsigned int pad_top, pad_bottom, pad_left;
const int8_t *weights;
long unsigned int input_cols, output_cols;
int8_t **outptrs;
const size_t *ld_out_cols;
const size_t *ld_out_vls;
long unsigned int current_channel, n_channels;
};
Args args = { inptr, ld_in_vl, pad_top, 9u - std::min(9u, pad_top + valid_input_rows), pad_left, weights, valid_input_cols, output_cols, outptrs, outlds, outvllds, start_channel, valid_channels };
__asm__ __volatile__(
".inst 0xd503477f // SMSTART ZA\n"
"ldr x6, [%x[args], %[offsetof_Args_pad_bottom]]\n"
"mov x22, SP\n"
"mov x21, #0x9\n"
"ptrue p2.b\n"
"ldr x7, [%x[args], %[offsetof_Args_pad_top]]\n"
"sub x20, x22, #0x8\n"
"ldr x17, [%x[args], %[offsetof_Args_n_channels]]\n"
"ld1rh { z29.h }, p2/Z, [%x[qp], %[offsetof_Requantize32_a_offset]]\n"
"and x20, x20, #-0x400\n"
".inst 0x25207812 // ptrue pn10.b\n"
"sub x21, x21, x6\n"
"mov SP, x20\n"
"ldr x16, [%x[args], %[offsetof_Args_current_channel]]\n"
"ld1rw { z8.s }, p2/Z, [%x[qp], %[offsetof_Requantize32_c_offset]]\n"
"str x22, [SP]\n"
"addvl SP, SP, #-6\n"
"ld1rw { z1.s }, p2/Z, [%x[qp], %[offsetof_Requantize32_per_layer_mul]]\n"
"whilelt p1.s, XZR, x17\n"
"whilelt p9.s, XZR, x21\n"
"neg z29.h, p2/M, z29.h\n"
"ld1rw { z0.s }, p2/Z, [%x[qp], %[offsetof_Requantize32_per_layer_right_shift]]\n"
"whilelt p8.s, XZR, x7\n"
"ld1rw { z31.s }, p2/Z, [%x[qp], %[offsetof_Requantize32_minval]]\n"
"eor p8.b, p2/Z, p8.b, p9.b\n"
"ld1rw { z28.s }, p2/Z, [%x[qp], %[offsetof_Requantize32_maxval]]\n"
"1:" // Channel loop
"ldr x20, [%x[qp], %[offsetof_Requantize32_bias]]\n"
"mov z16.s, #0x0\n"
"cbz x20, 2f\n"
"ld1w { z16.s }, p1/Z, [x20, x16, LSL #2]\n"
"2:" // Load bias: Done
"ldr x23, [%x[args], %[offsetof_Args_weights]]\n"
"ld1rh { z22.h }, p2/Z, [%x[qp], %[offsetof_Requantize32_b_offset]]\n"
"mov z27.h, #0x0\n"
"addvl x22, SP, #6\n"
"addvl x22, x22, #-2\n"
"ldr x21, [%x[qp], %[offsetof_Requantize32_per_channel_muls]]\n"
"mov z17.d, z16.d\n"
"mov z18.d, z16.d\n"
"mov z19.d, z16.d\n"
"mov x20, x23\n"
"incw x23\n"
"ld1sb { z25.s }, p2/Z, [x20]\n"
"incw x20, ALL, MUL #3\n"
"ld1sb { z15.s }, p2/Z, [x20]\n"
"incw x20, ALL, MUL #3\n"
"ld1sb { z9.s }, p2/Z, [x20]\n"
"mov x20, x23\n"
"incw x23\n"
"sub z25.h, z25.h, z22.h\n"
"sub z15.h, z15.h, z22.h\n"
"sub z9.h, z9.h, z22.h\n"
"trn1 z24.h, z25.h, z15.h\n"
"ld1sb { z12.s }, p2/Z, [x20]\n"
"incw x20, ALL, MUL #3\n"
"ld1sb { z4.s }, p2/Z, [x20]\n"
"incw x20, ALL, MUL #3\n"
"trn1 z11.h, z9.h, z27.h\n"
"ld1sb { z15.s }, p2/Z, [x20]\n"
"mov x20, x23\n"
"sub z12.h, z12.h, z22.h\n"
"sub z4.h, z4.h, z22.h\n"
"st1h { z24.h }, p2, [x22]\n"
"sub z15.h, z15.h, z22.h\n"
"st1h { z11.h }, p2, [x22, #1, MUL VL]\n"
"addvl x22, x22, #-2\n"
"trn1 z9.h, z12.h, z4.h\n"
"ld1sb { z14.s }, p2/Z, [x20]\n"
"incw x20, ALL, MUL #3\n"
"ld1sb { z10.s }, p2/Z, [x20]\n"
"incw x20, ALL, MUL #3\n"
"trn1 z21.h, z15.h, z27.h\n"
"ld1sb { z30.s }, p2/Z, [x20]\n"
"sub z14.h, z14.h, z22.h\n"
"sub z10.h, z10.h, z22.h\n"
"st1h { z9.h }, p2, [x22]\n"
"sub z30.h, z30.h, z22.h\n"
"st1h { z21.h }, p2, [x22, #1, MUL VL]\n"
"addvl x22, x22, #-2\n"
"trn1 z15.h, z14.h, z10.h\n"
"trn1 z25.h, z30.h, z27.h\n"
"st1h { z15.h }, p2, [x22]\n"
"st1h { z25.h }, p2, [x22, #1, MUL VL]\n"
"cbz x21, 3f\n"
"ld1w { z1.s }, p1/Z, [x21, x16, LSL #2]\n"
"3:" // Load mul: End
"ldr x20, [%x[qp], %[offsetof_Requantize32_per_channel_right_shifts]]\n"
"cbz x20, 4f\n"
"ld1w { z0.s }, p1/Z, [x20, x16, LSL #2]\n"
"4:" // Load right_shift: End
"ldr x15, [%x[args], %[offsetof_Args_input_cols]]\n"
"mov x22, #0x9\n"
"add x20, x7, x6\n"
"ldr x14, [%x[args], %[offsetof_Args_inptr]]\n"
"lsl x21, %x[ld_in_row], #0x0\n"
"mov x8, #0x0\n"
"ldr x13, [%x[args], %[offsetof_Args_output_cols]]\n"
"sub x22, x22, x20\n"
"sub x20, x15, #0x1\n"
"orr x20, x20, %x[ld_in_col], LSL #16\n"
"madd x21, x21, x7, x14\n"
"orr x20, x17, x20, LSL #22\n"
"lsl x20, x20, #0x0\n"
"5:" // Issue prefetches
"subs x22, x22, #0x1\n"
".inst 0xf8b44abc // rprfm pldstrm, x20, [x21]\n"
"add x21, x21, %x[ld_in_col]\n"
"bgt 5b\n"
"ldr x23, [%x[args], %[offsetof_Args_outptrs]]\n"
"lsl x21, %x[ld_in_row], #0x0\n"
".inst 0xc0040e00 // mova za.d[x8, #0], { z16.d-z19.d }\n"
"mov x22, #0x2\n"
"ldr x20, [%x[args], %[offsetof_Args_ld_out_cols]]\n"
"msub x14, x7, x21, x14\n"
".inst 0xc0040e01 // mova za.d[x8, #1], { z16.d-z19.d }\n"
"ldr x21, [%x[args], %[offsetof_Args_pad_left]]\n"
".inst 0xc0040e02 // mova za.d[x8, #2], { z16.d-z19.d }\n"
"ldp x11, x10, [x23], #0x10\n"
"ldp x9, x28, [x20], #0x10\n"
"ldp x27, x26, [x23], #0x10\n"
"ldp x25, x24, [x20], #0x10\n"
"cbz x21, 7f\n"
"cmp x21, x22\n"
"csel x20, x21, x22, LT\n"
"sub x21, x21, x20\n"
"sub x22, x22, x20\n"
"cbz x21, 7f\n"
".inst 0xc0060c0c // mova { z12.d-z15.d }, za.d[x8, #0]\n"
"and x22, x21, #0x1\n"
"add x21, x21, #0x1\n"
"lsr x21, x21, #0x1\n"
"sub x13, x13, x21\n"
".inst 0xc1a1ac0c // sqdmulh { z12.s-z15.s }, { z12.s-z15.s }, z1.s\n"
".inst 0xc1a0aa2c // srshl { z12.s-z15.s }, { z12.s-z15.s }, z0.s\n"
".inst 0xc1a8ab0c // add { z12.s-z15.s }, { z12.s-z15.s }, z8.s\n"
".inst 0xc1bccfec // sclamp { z12.s-z15.s }, z31.s, z28.s\n"
"6:" // Left padding
"subs x21, x21, #0x1\n"
"st1b { z12.s }, p1, [x11]\n"
"add x11, x11, x9\n"
"st1b { z13.s }, p1, [x10]\n"
"add x10, x10, x28\n"
"st1b { z14.s }, p1, [x27]\n"
"add x27, x27, x25\n"
"st1b { z15.s }, p1, [x26]\n"
"add x26, x26, x24\n"
"bgt 6b\n"
"7:" // Left padding: End
"adds XZR, x7, x6\n"
"bne 12f\n"
"cbz x22, 10f\n"
"cmp x22, #0x1\n"
"sub x15, x15, x22\n"
"beq 9f\n"
"8:" // Unpadded: 2 priming loads
"add x21, x14, %x[ld_in_row]\n"
"ld1sb { z23.s }, p1/Z, [x14]\n"
"addvl x20, SP, #4\n"
"ld1sb { z4.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add x14, x14, %x[ld_in_col]\n"
"ld1sb { z24.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"ld1sb { z5.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"ld1sb { z25.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z23.h, z23.h, z4.h\n"
"ld1sb { z6.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"ld1sb { z26.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z24.h, z24.h, z5.h\n"
"ld1sb { z20.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z25.h, z25.h, z6.h\n"
"ld1sb { z10.s }, p1/Z, [x21]\n"
"add z23.h, z23.h, z29.h\n"
".inst 0xa1402a86 // ld1h { z6.h, z14.h }, pn10.b/Z, [x20]\n"
"trn1 z26.h, z26.h, z20.h\n"
"add z24.h, z24.h, z29.h\n"
"mov z27.d, z10.d\n"
"add z25.h, z25.h, z29.h\n"
"add z26.h, z26.h, z29.h\n"
"add z27.h, z27.h, z29.h\n"
".inst 0xc17616e8 // sdot za.s[x8, 0], { z23.h-z26.h }, z6.h\n"
".inst 0xc17e1708 // sdot za.s[x8, 0], { z24.h-z27.h }, z14.h\n"
"9:" // Unpadded: 1 priming loads
"add x21, x14, %x[ld_in_row]\n"
"ld1sb { z20.s }, p1/Z, [x14]\n"
"addvl x20, SP, #2\n"
"ld1sb { z2.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add x14, x14, %x[ld_in_col]\n"
"ld1sb { z21.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"ld1sb { z25.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"ld1sb { z22.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z20.h, z20.h, z2.h\n"
"ld1sb { z24.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"ld1sb { z23.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z21.h, z21.h, z25.h\n"
"ld1sb { z9.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z22.h, z22.h, z24.h\n"
"ld1sb { z3.s }, p1/Z, [x21]\n"
"add z20.h, z20.h, z29.h\n"
".inst 0xa0402a86 // ld1h { z6.h-z7.h }, pn10.b/Z, [x20]\n"
"trn1 z23.h, z23.h, z9.h\n"
"add z21.h, z21.h, z29.h\n"
"mov z24.d, z3.d\n"
"add z22.h, z22.h, z29.h\n"
"add z23.h, z23.h, z29.h\n"
"add z24.h, z24.h, z29.h\n"
".inst 0xc1761688 // sdot za.s[x8, 0], { z20.h-z23.h }, z6.h\n"
".inst 0xc17716a8 // sdot za.s[x8, 0], { z21.h-z24.h }, z7.h\n"
"10:" // Unpadded: 0 priming loads
"cmp x15, #0x2\n"
".inst 0xa0402be2 // ld1h { z2.h-z3.h }, pn10.b/Z, [SP]\n"
"blt 18f\n"
"add x21, x14, %x[ld_in_row]\n"
"ld1sb { z10.s }, p1/Z, [x14]\n"
"sub x15, x15, #0x2\n"
"ld1sb { z25.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"sub x13, x13, #0x1\n"
"ld1sb { z11.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"lsr x20, x15, #0x1\n"
"ld1sb { z24.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"cmp x20, x13\n"
"ld1sb { z12.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z10.h, z10.h, z25.h\n"
"csel x23, x20, x13, LT\n"
"ld1sb { z30.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add x14, x14, %x[ld_in_col]\n"
"ld1sb { z13.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z11.h, z11.h, z24.h\n"
"and x15, x15, #0x1\n"
"ld1sb { z20.s }, p1/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"sub x13, x13, x23\n"
"ld1sb { z26.s }, p1/Z, [x21]\n"
"trn1 z12.h, z12.h, z30.h\n"
"add z10.h, z10.h, z29.h\n"
"trn1 z13.h, z13.h, z20.h\n"
"add z11.h, z11.h, z29.h\n"
"mov z14.d, z26.d\n"
"add z12.h, z12.h, z29.h\n"
"add z13.h, z13.h, z29.h\n"
"add z14.h, z14.h, z29.h\n"
"cbz x23, 17f\n"
"11:" // Unpadded: Main loop
".inst 0xc1721548 // sdot za.s[x8, 0], { z10.h-z13.h }, z2.h\n"
"addvl x20, SP, #4\n"
"add x22, x14, %x[ld_in_row]\n"
"addvl x21, SP, #2\n"
"subs x23, x23, #0x1\n"
".inst 0xc1731568 // sdot za.s[x8, 0], { z11.h-z14.h }, z3.h\n"
".inst 0xa1402a87 // ld1h { z7.h, z15.h }, pn10.b/Z, [x20]\n"
".inst 0xc1771549 // sdot za.s[x8, 1], { z10.h-z13.h }, z7.h\n"
"ld1sb { z3.s }, p1/Z, [x14]\n"
"add x14, x14, %x[ld_in_col]\n"
"ld1sb { z9.s }, p1/Z, [x22]\n"
"add x22, x22, %x[ld_in_row]\n"
".inst 0xc0060c18 // mova { z24.d-z27.d }, za.d[x8, #0]\n"
"add x20, x14, %x[ld_in_row]\n"
".inst 0xc17f1569 // sdot za.s[x8, 1], { z11.h-z14.h }, z15.h\n"
"ld1sb { z4.s }, p1/Z, [x22]\n"
"add x22, x22, %x[ld_in_row]\n"
"trn1 z3.h, z3.h, z9.h\n"
"ld1sb { z15.s }, p1/Z, [x22]\n"
"add x22, x22, %x[ld_in_row]\n"
"add x8, x8, #0x1\n"
"ld1sb { z5.s }, p1/Z, [x22]\n"
"add x22, x22, %x[ld_in_row]\n"
".inst 0xc1a1ac18 // sqdmulh { z24.s-z27.s }, { z24.s-z27.s }, z1.s\n"
"ld1sb { z10.s }, p1/Z, [x22]\n"
"add x22, x22, %x[ld_in_row]\n"
".inst 0xc0040e02 // mova za.d[x8, #2], { z16.d-z19.d }\n"
"ld1sb { z6.s }, p1/Z, [x22]\n"
"add x22, x22, %x[ld_in_row]\n"
"trn1 z4.h, z4.h, z15.h\n"
"add z3.h, z3.h, z29.h\n"
"ld1sb { z14.s }, p1/Z, [x22]\n"
"add x22, x22, %x[ld_in_row]\n"
"trn1 z5.h, z5.h, z10.h\n"
"ld1sb { z21.s }, p1/Z, [x22]\n"
".inst 0xc1a0aa38 // srshl { z24.s-z27.s }, { z24.s-z27.s }, z0.s\n"
".inst 0xa0402aaa // ld1h { z10.h-z11.h }, pn10.b/Z, [x21]\n"
"trn1 z6.h, z6.h, z14.h\n"
"add z4.h, z4.h, z29.h\n"
"mov z7.d, z21.d\n"
"add z5.h, z5.h, z29.h\n"
".inst 0xc1a8ab18 // add { z24.s-z27.s }, { z24.s-z27.s }, z8.s\n"
"add z6.h, z6.h, z29.h\n"
"add z7.h, z7.h, z29.h\n"
".inst 0xc1bccff8 // sclamp { z24.s-z27.s }, z31.s, z28.s\n"
".inst 0xc17a1468 // sdot za.s[x8, 0], { z3.h-z6.h }, z10.h\n"
"ld1sb { z10.s }, p1/Z, [x14]\n"
"add x14, x14, %x[ld_in_col]\n"
"ld1sb { z22.s }, p1/Z, [x20]\n"
"add x20, x20, %x[ld_in_row]\n"
"st1b { z24.s }, p1, [x11]\n"
"add x11, x11, x9\n"
".inst 0xc17b1488 // sdot za.s[x8, 0], { z4.h-z7.h }, z11.h\n"
"ld1sb { z11.s }, p1/Z, [x20]\n"
"add x20, x20, %x[ld_in_row]\n"
"trn1 z10.h, z10.h, z22.h\n"
"ld1sb { z14.s }, p1/Z, [x20]\n"
"add x20, x20, %x[ld_in_row]\n"
"st1b { z25.s }, p1, [x10]\n"
"add x10, x10, x28\n"
"ld1sb { z12.s }, p1/Z, [x20]\n"
"add x20, x20, %x[ld_in_row]\n"
"st1b { z26.s }, p1, [x27]\n"
"add x27, x27, x25\n"
"ld1sb { z9.s }, p1/Z, [x20]\n"
"add x20, x20, %x[ld_in_row]\n"
"st1b { z27.s }, p1, [x26]\n"
"add x26, x26, x24\n"
"ld1sb { z13.s }, p1/Z, [x20]\n"
"add x20, x20, %x[ld_in_row]\n"
"trn1 z11.h, z11.h, z14.h\n"
"add z10.h, z10.h, z29.h\n"
"ld1sb { z6.s }, p1/Z, [x20]\n"
"add x20, x20, %x[ld_in_row]\n"
"ld1sb { z20.s }, p1/Z, [x20]\n"
"trn1 z12.h, z12.h, z9.h\n"
".inst 0xa0402be2 // ld1h { z2.h-z3.h }, pn10.b/Z, [SP]\n"
"trn1 z13.h, z13.h, z6.h\n"
"add z11.h, z11.h, z29.h\n"
"mov z14.d, z20.d\n"
"add z12.h, z12.h, z29.h\n"
"add z13.h, z13.h, z29.h\n"
"add z14.h, z14.h, z29.h\n"
"bgt 11b\n"
"b 17f\n"
"12:" // Padded
"cbz x22, 15f\n"
"cmp x22, #0x1\n"
"sub x15, x15, x22\n"
"beq 14f\n"
"13:" // Padded: 2 priming loads
"mov x12, #0x0\n"
"add x21, x14, %x[ld_in_row]\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"addvl x20, SP, #4\n"
".inst 0xa1402a83 // ld1h { z3.h, z11.h }, pn10.b/Z, [x20]\n"
"ld1sb { z20.s }, p0/Z, [x14]\n"
"add x14, x14, %x[ld_in_col]\n"
"add z20.h, p0/M, z20.h, z29.h\n"
".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n"
"ld1sb { z22.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z22.h, p0/M, z22.h, z29.h\n"
".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n"
"ld1sb { z21.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z20.h, z20.h, z22.h\n"
"add z21.h, p0/M, z21.h, z29.h\n"
".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n"
"mov x12, #0x4\n"
"ld1sb { z23.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z23.h, p0/M, z23.h, z29.h\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"ld1sb { z22.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z21.h, z21.h, z23.h\n"
"add z22.h, p0/M, z22.h, z29.h\n"
".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n"
"ld1sb { z24.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z24.h, p0/M, z24.h, z29.h\n"
".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n"
"ld1sb { z23.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z22.h, z22.h, z24.h\n"
"add z23.h, p0/M, z23.h, z29.h\n"
".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n"
"mov x12, #0x8\n"
"ld1sb { z25.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z25.h, p0/M, z25.h, z29.h\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"ld1sb { z24.s }, p0/Z, [x21]\n"
"trn1 z23.h, z23.h, z25.h\n"
"add z24.h, p0/M, z24.h, z29.h\n"
".inst 0xc1731688 // sdot za.s[x8, 0], { z20.h-z23.h }, z3.h\n"
"mov z24.d, z24.d\n"
".inst 0xc17b16a8 // sdot za.s[x8, 0], { z21.h-z24.h }, z11.h\n"
"14:" // Padded: 1 priming loads
"mov x12, #0x0\n"
"add x21, x14, %x[ld_in_row]\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"addvl x20, SP, #2\n"
".inst 0xa1402a83 // ld1h { z3.h, z11.h }, pn10.b/Z, [x20]\n"
"ld1sb { z21.s }, p0/Z, [x14]\n"
"add x14, x14, %x[ld_in_col]\n"
"add z21.h, p0/M, z21.h, z29.h\n"
".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n"
"ld1sb { z20.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z20.h, p0/M, z20.h, z29.h\n"
".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n"
"ld1sb { z22.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z21.h, z21.h, z20.h\n"
"add z22.h, p0/M, z22.h, z29.h\n"
".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n"
"mov x12, #0x4\n"
"ld1sb { z20.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z20.h, p0/M, z20.h, z29.h\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"ld1sb { z23.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z22.h, z22.h, z20.h\n"
"add z23.h, p0/M, z23.h, z29.h\n"
".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n"
"ld1sb { z20.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z20.h, p0/M, z20.h, z29.h\n"
".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n"
"ld1sb { z24.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z23.h, z23.h, z20.h\n"
"add z24.h, p0/M, z24.h, z29.h\n"
".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n"
"mov x12, #0x8\n"
"ld1sb { z25.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z25.h, p0/M, z25.h, z29.h\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"ld1sb { z20.s }, p0/Z, [x21]\n"
"trn1 z24.h, z24.h, z25.h\n"
"add z20.h, p0/M, z20.h, z29.h\n"
".inst 0xc17316a8 // sdot za.s[x8, 0], { z21.h-z24.h }, z3.h\n"
"mov z25.d, z20.d\n"
".inst 0xc17b16c8 // sdot za.s[x8, 0], { z22.h-z25.h }, z11.h\n"
"15:" // Padded: 0 priming loads
"cmp x15, #0x2\n"
".inst 0xa0402be2 // ld1h { z2.h-z3.h }, pn10.b/Z, [SP]\n"
"blt 18f\n"
"mov x12, #0x0\n"
"add x21, x14, %x[ld_in_row]\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"sub x15, x15, #0x2\n"
"sub x13, x13, #0x1\n"
"lsr x20, x15, #0x1\n"
"cmp x20, x13\n"
"and x15, x15, #0x1\n"
"ld1sb { z10.s }, p0/Z, [x14]\n"
"csel x23, x20, x13, LT\n"
"add x14, x14, %x[ld_in_col]\n"
"sub x13, x13, x23\n"
"add z10.h, p0/M, z10.h, z29.h\n"
".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n"
"ld1sb { z20.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z20.h, p0/M, z20.h, z29.h\n"
".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n"
"ld1sb { z11.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z10.h, z10.h, z20.h\n"
"add z11.h, p0/M, z11.h, z29.h\n"
".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n"
"mov x12, #0x4\n"
"ld1sb { z20.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z20.h, p0/M, z20.h, z29.h\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"ld1sb { z12.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z11.h, z11.h, z20.h\n"
"add z12.h, p0/M, z12.h, z29.h\n"
".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n"
"ld1sb { z20.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z20.h, p0/M, z20.h, z29.h\n"
".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n"
"ld1sb { z13.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z12.h, z12.h, z20.h\n"
"add z13.h, p0/M, z13.h, z29.h\n"
".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n"
"mov x12, #0x8\n"
"ld1sb { z21.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z21.h, p0/M, z21.h, z29.h\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"ld1sb { z20.s }, p0/Z, [x21]\n"
"trn1 z13.h, z13.h, z21.h\n"
"add z20.h, p0/M, z20.h, z29.h\n"
"mov z14.d, z20.d\n"
"cbz x23, 17f\n"
"16:" // Padded: Main loop
".inst 0xc1721548 // sdot za.s[x8, 0], { z10.h-z13.h }, z2.h\n"
"addvl x20, SP, #4\n"
"mov x12, #0x0\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"add x22, x14, %x[ld_in_row]\n"
"addvl x21, SP, #2\n"
"subs x23, x23, #0x1\n"
".inst 0xc1731568 // sdot za.s[x8, 0], { z11.h-z14.h }, z3.h\n"
".inst 0xa0402a82 // ld1h { z2.h-z3.h }, pn10.b/Z, [x20]\n"
".inst 0xc1721549 // sdot za.s[x8, 1], { z10.h-z13.h }, z2.h\n"
"ld1sb { z10.s }, p0/Z, [x14]\n"
"add x14, x14, %x[ld_in_col]\n"
".inst 0xc0060c04 // mova { z4.d-z7.d }, za.d[x8, #0]\n"
"add x20, x14, %x[ld_in_row]\n"
"add z10.h, p0/M, z10.h, z29.h\n"
".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n"
".inst 0xc1731569 // sdot za.s[x8, 1], { z11.h-z14.h }, z3.h\n"
".inst 0xa0402aa2 // ld1h { z2.h-z3.h }, pn10.b/Z, [x21]\n"
"add x8, x8, #0x1\n"
".inst 0xc1a1ac04 // sqdmulh { z4.s-z7.s }, { z4.s-z7.s }, z1.s\n"
".inst 0xc0040e02 // mova za.d[x8, #2], { z16.d-z19.d }\n"
"ld1sb { z26.s }, p0/Z, [x22]\n"
"add x22, x22, %x[ld_in_row]\n"
"add z26.h, p0/M, z26.h, z29.h\n"
".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n"
".inst 0xc1a0aa24 // srshl { z4.s-z7.s }, { z4.s-z7.s }, z0.s\n"
"ld1sb { z11.s }, p0/Z, [x22]\n"
"add x22, x22, %x[ld_in_row]\n"
"trn1 z10.h, z10.h, z26.h\n"
".inst 0xc1a8ab04 // add { z4.s-z7.s }, { z4.s-z7.s }, z8.s\n"
"add z11.h, p0/M, z11.h, z29.h\n"
".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n"
"mov x12, #0x4\n"
"ld1sb { z9.s }, p0/Z, [x22]\n"
"add x22, x22, %x[ld_in_row]\n"
".inst 0xc1bccfe4 // sclamp { z4.s-z7.s }, z31.s, z28.s\n"
"add z9.h, p0/M, z9.h, z29.h\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"st1b { z4.s }, p1, [x11]\n"
"add x11, x11, x9\n"
"st1b { z5.s }, p1, [x10]\n"
"add x10, x10, x28\n"
"ld1sb { z12.s }, p0/Z, [x22]\n"
"add x22, x22, %x[ld_in_row]\n"
"st1b { z6.s }, p1, [x27]\n"
"add x27, x27, x25\n"
"trn1 z11.h, z11.h, z9.h\n"
"st1b { z7.s }, p1, [x26]\n"
"add x26, x26, x24\n"
"add z12.h, p0/M, z12.h, z29.h\n"
".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n"
"ld1sb { z9.s }, p0/Z, [x22]\n"
"add x22, x22, %x[ld_in_row]\n"
"add z9.h, p0/M, z9.h, z29.h\n"
".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n"
"ld1sb { z13.s }, p0/Z, [x22]\n"
"add x22, x22, %x[ld_in_row]\n"
"trn1 z12.h, z12.h, z9.h\n"
"add z13.h, p0/M, z13.h, z29.h\n"
".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n"
"mov x12, #0x8\n"
"ld1sb { z20.s }, p0/Z, [x22]\n"
"add x22, x22, %x[ld_in_row]\n"
"add z20.h, p0/M, z20.h, z29.h\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"mov x12, #0x0\n"
"ld1sb { z9.s }, p0/Z, [x22]\n"
"trn1 z13.h, z13.h, z20.h\n"
"add z9.h, p0/M, z9.h, z29.h\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
".inst 0xc1721548 // sdot za.s[x8, 0], { z10.h-z13.h }, z2.h\n"
"ld1sb { z10.s }, p0/Z, [x14]\n"
"add x14, x14, %x[ld_in_col]\n"
"mov z14.d, z9.d\n"
"add z10.h, p0/M, z10.h, z29.h\n"
".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n"
".inst 0xc1731568 // sdot za.s[x8, 0], { z11.h-z14.h }, z3.h\n"
".inst 0xa0402be2 // ld1h { z2.h-z3.h }, pn10.b/Z, [SP]\n"
"ld1sb { z25.s }, p0/Z, [x20]\n"
"add x20, x20, %x[ld_in_row]\n"
"add z25.h, p0/M, z25.h, z29.h\n"
".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n"
"ld1sb { z11.s }, p0/Z, [x20]\n"
"add x20, x20, %x[ld_in_row]\n"
"trn1 z10.h, z10.h, z25.h\n"
"add z11.h, p0/M, z11.h, z29.h\n"
".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n"
"mov x12, #0x4\n"
"ld1sb { z15.s }, p0/Z, [x20]\n"
"add x20, x20, %x[ld_in_row]\n"
"add z15.h, p0/M, z15.h, z29.h\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"ld1sb { z12.s }, p0/Z, [x20]\n"
"add x20, x20, %x[ld_in_row]\n"
"trn1 z11.h, z11.h, z15.h\n"
"add z12.h, p0/M, z12.h, z29.h\n"
".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n"
"ld1sb { z4.s }, p0/Z, [x20]\n"
"add x20, x20, %x[ld_in_row]\n"
"add z4.h, p0/M, z4.h, z29.h\n"
".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n"
"ld1sb { z13.s }, p0/Z, [x20]\n"
"add x20, x20, %x[ld_in_row]\n"
"trn1 z12.h, z12.h, z4.h\n"
"add z13.h, p0/M, z13.h, z29.h\n"
".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n"
"mov x12, #0x8\n"
"ld1sb { z4.s }, p0/Z, [x20]\n"
"add x20, x20, %x[ld_in_row]\n"
"add z4.h, p0/M, z4.h, z29.h\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"ld1sb { z26.s }, p0/Z, [x20]\n"
"trn1 z13.h, z13.h, z4.h\n"
"add z26.h, p0/M, z26.h, z29.h\n"
"mov z14.d, z26.d\n"
"bgt 16b\n"
"17:" // Main loop tail
".inst 0xc1721548 // sdot za.s[x8, 0], { z10.h-z13.h }, z2.h\n"
"addvl x22, SP, #4\n"
"mov x12, #0x0\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"add x21, x14, %x[ld_in_row]\n"
"addvl x20, SP, #2\n"
".inst 0xc1731568 // sdot za.s[x8, 0], { z11.h-z14.h }, z3.h\n"
".inst 0xa0402ac2 // ld1h { z2.h-z3.h }, pn10.b/Z, [x22]\n"
".inst 0xc1721549 // sdot za.s[x8, 1], { z10.h-z13.h }, z2.h\n"
"ld1sb { z9.s }, p0/Z, [x14]\n"
"add x14, x14, %x[ld_in_col]\n"
".inst 0xc0060c04 // mova { z4.d-z7.d }, za.d[x8, #0]\n"
"add z9.h, p0/M, z9.h, z29.h\n"
".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n"
".inst 0xc1731569 // sdot za.s[x8, 1], { z11.h-z14.h }, z3.h\n"
".inst 0xa0402a82 // ld1h { z2.h-z3.h }, pn10.b/Z, [x20]\n"
"add x8, x8, #0x1\n"
".inst 0xc1a1ac04 // sqdmulh { z4.s-z7.s }, { z4.s-z7.s }, z1.s\n"
".inst 0xc0040e02 // mova za.d[x8, #2], { z16.d-z19.d }\n"
"ld1sb { z15.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z15.h, p0/M, z15.h, z29.h\n"
".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n"
".inst 0xc1a0aa24 // srshl { z4.s-z7.s }, { z4.s-z7.s }, z0.s\n"
"ld1sb { z10.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z9.h, z9.h, z15.h\n"
".inst 0xc1a8ab04 // add { z4.s-z7.s }, { z4.s-z7.s }, z8.s\n"
"add z10.h, p0/M, z10.h, z29.h\n"
".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n"
"mov x12, #0x4\n"
"ld1sb { z15.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
".inst 0xc1bccfe4 // sclamp { z4.s-z7.s }, z31.s, z28.s\n"
"add z15.h, p0/M, z15.h, z29.h\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"st1b { z4.s }, p1, [x11]\n"
"add x11, x11, x9\n"
"st1b { z5.s }, p1, [x10]\n"
"add x10, x10, x28\n"
"ld1sb { z11.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"st1b { z6.s }, p1, [x27]\n"
"add x27, x27, x25\n"
"trn1 z10.h, z10.h, z15.h\n"
"st1b { z7.s }, p1, [x26]\n"
"add x26, x26, x24\n"
"add z11.h, p0/M, z11.h, z29.h\n"
".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n"
"ld1sb { z5.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z5.h, p0/M, z5.h, z29.h\n"
".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n"
"ld1sb { z12.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z11.h, z11.h, z5.h\n"
"add z12.h, p0/M, z12.h, z29.h\n"
".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n"
"mov x12, #0x8\n"
"ld1sb { z15.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z15.h, p0/M, z15.h, z29.h\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"ld1sb { z5.s }, p0/Z, [x21]\n"
"trn1 z12.h, z12.h, z15.h\n"
"add z5.h, p0/M, z5.h, z29.h\n"
".inst 0xc1721528 // sdot za.s[x8, 0], { z9.h-z12.h }, z2.h\n"
"mov z13.d, z5.d\n"
".inst 0xc1731548 // sdot za.s[x8, 0], { z10.h-z13.h }, z3.h\n"
".inst 0xa0402be2 // ld1h { z2.h-z3.h }, pn10.b/Z, [SP]\n"
"18:" // Main loop skip tail
"cbz x15, 19f\n" // Skip remainder inputs
"mov x12, #0x0\n"
"add x21, x14, %x[ld_in_row]\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"addvl x20, SP, #4\n"
"sub x13, x13, #0x1\n"
"ld1sb { z21.s }, p0/Z, [x14]\n"
"add z21.h, p0/M, z21.h, z29.h\n"
".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n"
"ld1sb { z15.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z15.h, p0/M, z15.h, z29.h\n"
".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n"
"ld1sb { z22.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z21.h, z21.h, z15.h\n"
"add z22.h, p0/M, z22.h, z29.h\n"
".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n"
"mov x12, #0x4\n"
"ld1sb { z12.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z12.h, p0/M, z12.h, z29.h\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"ld1sb { z23.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z22.h, z22.h, z12.h\n"
"add z23.h, p0/M, z23.h, z29.h\n"
".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n"
"ld1sb { z20.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z20.h, p0/M, z20.h, z29.h\n"
".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n"
"ld1sb { z24.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"trn1 z23.h, z23.h, z20.h\n"
"add z24.h, p0/M, z24.h, z29.h\n"
".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n"
"mov x12, #0x8\n"
"ld1sb { z30.s }, p0/Z, [x21]\n"
"add x21, x21, %x[ld_in_row]\n"
"add z30.h, p0/M, z30.h, z29.h\n"
".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n"
"ld1sb { z6.s }, p0/Z, [x21]\n"
"trn1 z24.h, z24.h, z30.h\n"
"add z6.h, p0/M, z6.h, z29.h\n"
".inst 0xc17216a8 // sdot za.s[x8, 0], { z21.h-z24.h }, z2.h\n"
"mov z25.d, z6.d\n"
".inst 0xc17316c8 // sdot za.s[x8, 0], { z22.h-z25.h }, z3.h\n"
".inst 0xa1402a85 // ld1h { z5.h, z13.h }, pn10.b/Z, [x20]\n"
".inst 0xc17516a9 // sdot za.s[x8, 1], { z21.h-z24.h }, z5.h\n"
".inst 0xc0060c04 // mova { z4.d-z7.d }, za.d[x8, #0]\n"
".inst 0xc17d16c9 // sdot za.s[x8, 1], { z22.h-z25.h }, z13.h\n"
"add x8, x8, #0x1\n"
".inst 0xc1a1ac04 // sqdmulh { z4.s-z7.s }, { z4.s-z7.s }, z1.s\n"
".inst 0xc0040e02 // mova za.d[x8, #2], { z16.d-z19.d }\n"
".inst 0xc1a0aa24 // srshl { z4.s-z7.s }, { z4.s-z7.s }, z0.s\n"
".inst 0xc1a8ab04 // add { z4.s-z7.s }, { z4.s-z7.s }, z8.s\n"
".inst 0xc1bccfe4 // sclamp { z4.s-z7.s }, z31.s, z28.s\n"
"st1b { z4.s }, p1, [x11]\n"
"add x11, x11, x9\n"
"st1b { z5.s }, p1, [x10]\n"
"add x10, x10, x28\n"
"st1b { z6.s }, p1, [x27]\n"
"add x27, x27, x25\n"
"st1b { z7.s }, p1, [x26]\n"
"add x26, x26, x24\n"
"19:" // Tail input: End
"cbz x13, 21f\n"
"20:" // Right padding loop
".inst 0xc0060c04 // mova { z4.d-z7.d }, za.d[x8, #0]\n"
"add x8, x8, #0x1\n"
"subs x13, x13, #0x1\n"
".inst 0xc0040e02 // mova za.d[x8, #2], { z16.d-z19.d }\n"
".inst 0xc1a1ac04 // sqdmulh { z4.s-z7.s }, { z4.s-z7.s }, z1.s\n"
".inst 0xc1a0aa24 // srshl { z4.s-z7.s }, { z4.s-z7.s }, z0.s\n"
".inst 0xc1a8ab04 // add { z4.s-z7.s }, { z4.s-z7.s }, z8.s\n"
".inst 0xc1bccfe4 // sclamp { z4.s-z7.s }, z31.s, z28.s\n"
"st1b { z4.s }, p1, [x11]\n"
"add x11, x11, x9\n"
"st1b { z5.s }, p1, [x10]\n"
"add x10, x10, x28\n"
"st1b { z6.s }, p1, [x27]\n"
"add x27, x27, x25\n"
"st1b { z7.s }, p1, [x26]\n"
"add x26, x26, x24\n"
"bgt 20b\n"
"21:" // End
"ldr x20, [%x[args], %[offsetof_Args_weights]]\n"
"incw x16\n"
"whilelt p1.s, x16, x17\n"
"incw x20, ALL, MUL #9\n"
"str x20, [%x[args], %[offsetof_Args_weights]]\n"
"ldr x21, [%x[args], %[offsetof_Args_ld_in_vl]]\n"
"ldr x20, [%x[args], %[offsetof_Args_inptr]]\n"
"add x20, x20, x21\n"
"str x20, [%x[args], %[offsetof_Args_inptr]]\n"
"ldr x25, [%x[args], %[offsetof_Args_outptrs]]\n"
"ldr x24, [%x[args], %[offsetof_Args_ld_out_vls]]\n"
"ldp x23, x22, [x25, #0x0]\n"
"ldp x21, x20, [x24, #0x0]\n"
"add x23, x23, x21\n"
"add x22, x22, x20\n"
"stp x23, x22, [x25, #0x0]\n"
"ldp x23, x22, [x25, #0x10]\n"
"ldp x21, x20, [x24, #0x10]\n"
"add x23, x23, x21\n"
"add x22, x22, x20\n"
"stp x23, x22, [x25, #0x10]\n"
"b.any 1b\n"
"addvl SP, SP, #6\n"
".inst 0xd503467f // SMSTOP\n"
"ldr x20, [SP, #0x0]\n"
"mov SP, x20\n"
:
: [args] "r" (&args), [ld_in_col] "r" (ld_in_col), [ld_in_row] "r" (ld_in_row), [offsetof_Args_current_channel] "I" (offsetof(Args, current_channel)), [offsetof_Args_inptr] "I" (offsetof(Args, inptr)), [offsetof_Args_input_cols] "I" (offsetof(Args, input_cols)), [offsetof_Args_ld_in_vl] "I" (offsetof(Args, ld_in_vl)), [offsetof_Args_ld_out_cols] "I" (offsetof(Args, ld_out_cols)), [offsetof_Args_ld_out_vls] "I" (offsetof(Args, ld_out_vls)), [offsetof_Args_n_channels] "I" (offsetof(Args, n_channels)), [offsetof_Args_outptrs] "I" (offsetof(Args, outptrs)), [offsetof_Args_output_cols] "I" (offsetof(Args, output_cols)), [offsetof_Args_pad_bottom] "I" (offsetof(Args, pad_bottom)), [offsetof_Args_pad_left] "I" (offsetof(Args, pad_left)), [offsetof_Args_pad_top] "I" (offsetof(Args, pad_top)), [offsetof_Args_weights] "I" (offsetof(Args, weights)), [offsetof_Requantize32_a_offset] "I" (offsetof(arm_gemm::Requantize32, a_offset)), [offsetof_Requantize32_b_offset] "I" (offsetof(arm_gemm::Requantize32, b_offset)), [offsetof_Requantize32_bias] "I" (offsetof(arm_gemm::Requantize32, bias)), [offsetof_Requantize32_c_offset] "I" (offsetof(arm_gemm::Requantize32, c_offset)), [offsetof_Requantize32_maxval] "I" (offsetof(arm_gemm::Requantize32, maxval)), [offsetof_Requantize32_minval] "I" (offsetof(arm_gemm::Requantize32, minval)), [offsetof_Requantize32_per_channel_muls] "I" (offsetof(arm_gemm::Requantize32, per_channel_muls)), [offsetof_Requantize32_per_channel_right_shifts] "I" (offsetof(arm_gemm::Requantize32, per_channel_right_shifts)), [offsetof_Requantize32_per_layer_mul] "I" (offsetof(arm_gemm::Requantize32, per_layer_mul)), [offsetof_Requantize32_per_layer_right_shift] "I" (offsetof(arm_gemm::Requantize32, per_layer_right_shift)), [qp] "r" (&qp)
: "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x6", "x7", "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "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 depthwise
} // namespace arm_conv
#endif // defined(ARM_COMPUTE_ENABLE_SME2)
|
