-
Notifications
You must be signed in to change notification settings - Fork 1.5k
/
Copy pathdgemm_kernel_sve_v2x8.S
1570 lines (1208 loc) · 35.8 KB
/
dgemm_kernel_sve_v2x8.S
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
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*******************************************************************************
Copyright (c) 2015, The OpenBLAS Project
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
1. Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in
the documentation and/or other materials provided with the
distribution.
3. Neither the name of the OpenBLAS project nor the names of
its contributors may be used to endorse or promote products
derived from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*******************************************************************************/
/* This is an SVE dgemm kernel with size 2*SVE_LEN x 8.
However, the data layout is the same as for the kernel 1*SVE_LEN x 8.
This means that we sweep two panels of packed A when iterating in a loop over K.
With this approach, we can reuse dgemm_n|tcopy_sve_v1.c packing functions. */
#define ASSEMBLER
#include "common.h"
/* X0 X1 X2 s0 X3 x4 x5 x6 */
/*int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha0,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc )*/
#define origM x0
#define origN x1
#define origK x2
#define origPA x3
#define origPB x4
#define pC x5
#define LDC x6
#define temp x7
#define counterL x8
#define counterI x9
#define counterJ x10
#define pB x11
#define pCRow0 x12
#define pCRow1 x13
#define pCRow2 x14
#define lanes x15
#define pA1 x16
#define pA2 x17
#define alpha x18
#define vec_len x19
#define vec_lenx2 x20
#define alpha0 d10
#define alphaZ z7.d
#define A_PRE_SIZE 1536
#define B_PRE_SIZE 512
#define C_PRE_SIZE 128
// 00 origM
// 01 origN
// 02 origK
// 03 origPA
// 04 origPB
// 05 pC
// 06 origLDC -> LDC
// 07 temp
// 08 counterL
// 09 counterI
// 10 counterJ
// 11 pB
// 12 pCRow0
// 13 pCRow1
// 14 pCRow2
// 15 lanes
// 16 pA1
// 17 pA1
// 18 must save alpha
// 19 must save vec_len
// 20 must save
// 21 must save
// 22 must save
// 23 must save
// 24 must save
// 25 must save
// 26 must save
// 27 must save
// 28 must save
// 29 frame
// 30 link
// 31 sp
//v00 ALPHA -> pA10_0
//v01 pA10_1
//v02 pA20_0
//v03 pA20_1
//v04
//v05
//v06
//v07 ALPHA0
//v08 must save pB0_0
//v09 must save pB0_1
//v10 must save pB0_2
//v11 must save pB0_3
//v12 must save pB0_4
//v13 must save pB0_5
//v14 must save pB0_6
//v15 must save pB0_7
//v16 must save C0
//v17 must save C1
//v18 must save C2
//v19 must save C3
//v20 must save C4
//v21 must save C5
//v22 must save C6
//v23 must save C7
//v24 must save C8
//v25 must save C9
//v26 must save C10
//v27 must save C11
//v28 must save C12
//v29 must save C13
//v30 must save C14
//v31 must save C15
/*******************************************************************************
* Macro definitions
*******************************************************************************/
.macro INITv2x8
dup z16.d, #0
dup z17.d, #0
dup z18.d, #0
dup z19.d, #0
dup z20.d, #0
dup z21.d, #0
dup z22.d, #0
dup z23.d, #0
dup z24.d, #0
dup z25.d, #0
dup z26.d, #0
dup z27.d, #0
dup z28.d, #0
dup z29.d, #0
dup z30.d, #0
dup z31.d, #0
.endm
.macro KERNELv2x8_I
ld1d z0.d, p0/z, [pA1]
ld1d z1.d, p0/z, [pA2]
ld1d z2.d, p0/z, [pA1, vec_len, lsl #3]
ld1d z3.d, p0/z, [pA2, vec_len, lsl #3]
add pA1, pA1, vec_len, lsl #4 // pA1 = pA1 + vec_len * 8 *2
add pA2, pA2, vec_len, lsl #4 // pA1 = pA1 + vec_len * 8 *2
ld1rd z8.d, p0/z, [pB]
ld1rd z9.d, p0/z, [pB, 8]
ld1rd z10.d, p0/z, [pB, 16]
ld1rd z11.d, p0/z, [pB, 24]
ld1rd z12.d, p0/z, [pB, 32]
ld1rd z13.d, p0/z, [pB, 40]
ld1rd z14.d, p0/z, [pB, 48]
ld1rd z15.d, p0/z, [pB, 56]
add pB, pB, 64
fmla z16.d, p0/m, z0.d, z8.d
fmla z17.d, p0/m, z1.d, z8.d
ld1rd z8.d, p0/z, [pB]
fmla z18.d, p0/m, z0.d, z9.d
fmla z19.d, p0/m, z1.d, z9.d
ld1rd z9.d, p0/z, [pB, 8]
fmla z20.d, p0/m, z0.d, z10.d
fmla z21.d, p0/m, z1.d, z10.d
ld1rd z10.d, p0/z, [pB, 16]
fmla z22.d, p0/m, z0.d, z11.d
fmla z23.d, p0/m, z1.d, z11.d
ld1rd z11.d, p0/z, [pB, 24]
fmla z24.d, p0/m, z0.d, z12.d
fmla z25.d, p0/m, z1.d, z12.d
ld1rd z12.d, p0/z, [pB, 32]
fmla z26.d, p0/m, z0.d, z13.d
fmla z27.d, p0/m, z1.d, z13.d
ld1rd z13.d, p0/z, [pB, 40]
fmla z28.d, p0/m, z0.d, z14.d
fmla z29.d, p0/m, z1.d, z14.d
ld1rd z14.d, p0/z, [pB, 48]
fmla z30.d, p0/m, z0.d, z15.d
fmla z31.d, p0/m, z1.d, z15.d
ld1rd z15.d, p0/z, [pB, 56]
add pB, pB, 64
.endm
.macro KERNELv2x8_M1
ld1d z2.d, p0/z, [pA1]
ld1d z3.d, p0/z, [pA2]
add pA1, pA1, vec_len, lsl #3 // pA1 = pA1 + vec_len * 8
add pA2, pA2, vec_len, lsl #3 // pA1 = pA1 + vec_len * 8
fmla z16.d, p0/m, z0.d, z8.d
fmla z17.d, p0/m, z1.d, z8.d
ld1rd z8.d, p0/z, [pB]
fmla z18.d, p0/m, z0.d, z9.d
fmla z19.d, p0/m, z1.d, z9.d
ld1rd z9.d, p0/z, [pB, 8]
fmla z20.d, p0/m, z0.d, z10.d
fmla z21.d, p0/m, z1.d, z10.d
ld1rd z10.d, p0/z, [pB, 16]
fmla z22.d, p0/m, z0.d, z11.d
fmla z23.d, p0/m, z1.d, z11.d
ld1rd z11.d, p0/z, [pB, 24]
fmla z24.d, p0/m, z0.d, z12.d
fmla z25.d, p0/m, z1.d, z12.d
ld1rd z12.d, p0/z, [pB, 32]
fmla z26.d, p0/m, z0.d, z13.d
fmla z27.d, p0/m, z1.d, z13.d
ld1rd z13.d, p0/z, [pB, 40]
fmla z28.d, p0/m, z0.d, z14.d
fmla z29.d, p0/m, z1.d, z14.d
ld1rd z14.d, p0/z, [pB, 48]
fmla z30.d, p0/m, z0.d, z15.d
fmla z31.d, p0/m, z1.d, z15.d
ld1rd z15.d, p0/z, [pB, 56]
add pB, pB, 64
.endm
.macro KERNELv2x8_M2
ld1d z0.d, p0/z, [pA1]
ld1d z1.d, p0/z, [pA2]
add pA1, pA1, vec_len, lsl #3 // pA1 = pA1 + vec_len * 2 * 8
add pA2, pA2, vec_len, lsl #3 // pA1 = pA1 + vec_len * 2 * 8
fmla z16.d, p0/m, z2.d, z8.d
fmla z17.d, p0/m, z3.d, z8.d
ld1rd z8.d, p0/z, [pB]
fmla z18.d, p0/m, z2.d, z9.d
fmla z19.d, p0/m, z3.d, z9.d
ld1rd z9.d, p0/z, [pB, 8]
fmla z20.d, p0/m, z2.d, z10.d
fmla z21.d, p0/m, z3.d, z10.d
ld1rd z10.d, p0/z, [pB, 16]
fmla z22.d, p0/m, z2.d, z11.d
fmla z23.d, p0/m, z3.d, z11.d
ld1rd z11.d, p0/z, [pB, 24]
fmla z24.d, p0/m, z2.d, z12.d
fmla z25.d, p0/m, z3.d, z12.d
ld1rd z12.d, p0/z, [pB, 32]
fmla z26.d, p0/m, z2.d, z13.d
fmla z27.d, p0/m, z3.d, z13.d
ld1rd z13.d, p0/z, [pB, 40]
fmla z28.d, p0/m, z2.d, z14.d
fmla z29.d, p0/m, z3.d, z14.d
ld1rd z14.d, p0/z, [pB, 48]
fmla z30.d, p0/m, z2.d, z15.d
fmla z31.d, p0/m, z3.d, z15.d
ld1rd z15.d, p0/z, [pB, 56]
add pB, pB, 64
.endm
.macro KERNELv2x8_E
fmla z16.d, p0/m, z2.d, z8.d
fmla z17.d, p0/m, z3.d, z8.d
fmla z18.d, p0/m, z2.d, z9.d
fmla z19.d, p0/m, z3.d, z9.d
fmla z20.d, p0/m, z2.d, z10.d
fmla z21.d, p0/m, z3.d, z10.d
fmla z22.d, p0/m, z2.d, z11.d
fmla z23.d, p0/m, z3.d, z11.d
fmla z24.d, p0/m, z2.d, z12.d
fmla z25.d, p0/m, z3.d, z12.d
fmla z26.d, p0/m, z2.d, z13.d
fmla z27.d, p0/m, z3.d, z13.d
fmla z28.d, p0/m, z2.d, z14.d
fmla z29.d, p0/m, z3.d, z14.d
fmla z30.d, p0/m, z2.d, z15.d
fmla z31.d, p0/m, z3.d, z15.d
.endm
.macro KERNELv2x8_SUB
ld1d z0.d, p0/z, [pA1]
ld1d z1.d, p0/z, [pA2]
add pA1, pA1, vec_len, lsl #3 // pA1 = pA1 + vec_len * 8
add pA2, pA2, vec_len, lsl #3 // pA1 = pA1 + vec_len * 8
ld1rd z8.d, p0/z, [pB]
ld1rd z9.d, p0/z, [pB, 8]
ld1rd z10.d, p0/z, [pB, 16]
ld1rd z11.d, p0/z, [pB, 24]
ld1rd z12.d, p0/z, [pB, 32]
ld1rd z13.d, p0/z, [pB, 40]
ld1rd z14.d, p0/z, [pB, 48]
ld1rd z15.d, p0/z, [pB, 56]
add pB, pB, 64
fmla z16.d, p0/m, z0.d, z8.d
fmla z17.d, p0/m, z1.d, z8.d
fmla z18.d, p0/m, z0.d, z9.d
fmla z19.d, p0/m, z1.d, z9.d
fmla z20.d, p0/m, z0.d, z10.d
fmla z21.d, p0/m, z1.d, z10.d
fmla z22.d, p0/m, z0.d, z11.d
fmla z23.d, p0/m, z1.d, z11.d
fmla z24.d, p0/m, z0.d, z12.d
fmla z25.d, p0/m, z1.d, z12.d
fmla z26.d, p0/m, z0.d, z13.d
fmla z27.d, p0/m, z1.d, z13.d
fmla z28.d, p0/m, z0.d, z14.d
fmla z29.d, p0/m, z1.d, z14.d
fmla z30.d, p0/m, z0.d, z15.d
fmla z31.d, p0/m, z1.d, z15.d
.endm
.macro SAVEv2x8
add pCRow1, pCRow0, LDC
ld1d z8.d, p0/z, [pCRow0]
ld1d z9.d, p0/z, [pCRow0, #1, mul vl]
fmla z8.d, p0/m, z16.d, alphaZ
fmla z9.d, p0/m, z17.d, alphaZ
st1d z8.d, p0, [pCRow0]
st1d z9.d, p0, [pCRow0, #1, mul vl]
add pCRow2, pCRow1, LDC
ld1d z10.d, p0/z, [pCRow1]
ld1d z11.d, p0/z, [pCRow1, #1, mul vl]
fmla z10.d, p0/m, z18.d, alphaZ
fmla z11.d, p0/m, z19.d, alphaZ
st1d z10.d, p0, [pCRow1]
st1d z11.d, p0, [pCRow1, #1, mul vl]
add pCRow1, pCRow2, LDC
ld1d z12.d, p0/z, [pCRow2]
ld1d z13.d, p0/z, [pCRow2, #1, mul vl]
fmla z12.d, p0/m, z20.d, alphaZ
fmla z13.d, p0/m, z21.d, alphaZ
st1d z12.d, p0, [pCRow2]
st1d z13.d, p0, [pCRow2, #1, mul vl]
add pCRow2, pCRow1, LDC
ld1d z14.d, p0/z, [pCRow1]
ld1d z15.d, p0/z, [pCRow1, #1, mul vl]
fmla z14.d, p0/m, z22.d, alphaZ
fmla z15.d, p0/m, z23.d, alphaZ
st1d z14.d, p0, [pCRow1]
st1d z15.d, p0, [pCRow1, #1, mul vl]
add pCRow1, pCRow2, LDC
ld1d z8.d, p0/z, [pCRow2]
ld1d z9.d, p0/z, [pCRow2, #1, mul vl]
fmla z8.d, p0/m, z24.d, alphaZ
fmla z9.d, p0/m, z25.d, alphaZ
st1d z8.d, p0, [pCRow2]
st1d z9.d, p0, [pCRow2, #1, mul vl]
add pCRow2, pCRow1, LDC
ld1d z10.d, p0/z, [pCRow1]
ld1d z11.d, p0/z, [pCRow1, #1, mul vl]
fmla z10.d, p0/m, z26.d, alphaZ
fmla z11.d, p0/m, z27.d, alphaZ
st1d z10.d, p0, [pCRow1]
st1d z11.d, p0, [pCRow1, #1, mul vl]
add pCRow1, pCRow2, LDC
ld1d z12.d, p0/z, [pCRow2]
ld1d z13.d, p0/z, [pCRow2, #1, mul vl]
fmla z12.d, p0/m, z28.d, alphaZ
fmla z13.d, p0/m, z29.d, alphaZ
st1d z12.d, p0, [pCRow2]
st1d z13.d, p0, [pCRow2, #1, mul vl]
ld1d z14.d, p0/z, [pCRow1]
ld1d z15.d, p0/z, [pCRow1, #1, mul vl]
fmla z14.d, p0/m, z30.d, alphaZ
fmla z15.d, p0/m, z31.d, alphaZ
st1d z14.d, p0, [pCRow1]
st1d z15.d, p0, [pCRow1, #1, mul vl]
add pCRow0, pCRow0, vec_len, lsl #4 // pC = pC + vec_len * 8 * 2
.endm
.macro INITv2x4
dup z16.d, #0
dup z17.d, #0
dup z18.d, #0
dup z19.d, #0
dup z20.d, #0
dup z21.d, #0
dup z22.d, #0
dup z23.d, #0
.endm
.macro KERNELv2x4_SUB
ld1d z0.d, p0/z, [pA1]
ld1d z1.d, p0/z, [pA2]
add pA1, pA1, vec_len, lsl #3 // pA1 = pA1 + vec_len * 8
add pA2, pA2, vec_len, lsl #3 // pA1 = pA1 + vec_len * 8
ld1rd z8.d, p0/z, [pB]
ld1rd z9.d, p0/z, [pB, 8]
ld1rd z10.d, p0/z, [pB, 16]
ld1rd z11.d, p0/z, [pB, 24]
add pB, pB, 32
fmla z16.d, p0/m, z0.d, z8.d
fmla z17.d, p0/m, z1.d, z8.d
fmla z18.d, p0/m, z0.d, z9.d
fmla z19.d, p0/m, z1.d, z9.d
fmla z20.d, p0/m, z0.d, z10.d
fmla z21.d, p0/m, z1.d, z10.d
fmla z22.d, p0/m, z0.d, z11.d
fmla z23.d, p0/m, z1.d, z11.d
.endm
.macro SAVEv2x4
add pCRow1, pCRow0, LDC
ld1d z8.d, p0/z, [pCRow0]
ld1d z9.d, p0/z, [pCRow0, #1, mul vl]
fmla z8.d, p0/m, z16.d, alphaZ
fmla z9.d, p0/m, z17.d, alphaZ
st1d z8.d, p0, [pCRow0]
st1d z9.d, p0, [pCRow0, #1, mul vl]
add pCRow2, pCRow1, LDC
ld1d z10.d, p0/z, [pCRow1]
ld1d z11.d, p0/z, [pCRow1, #1, mul vl]
fmla z10.d, p0/m, z18.d, alphaZ
fmla z11.d, p0/m, z19.d, alphaZ
st1d z10.d, p0, [pCRow1]
st1d z11.d, p0, [pCRow1, #1, mul vl]
add pCRow1, pCRow2, LDC
ld1d z12.d, p0/z, [pCRow2]
ld1d z13.d, p0/z, [pCRow2, #1, mul vl]
fmla z12.d, p0/m, z20.d, alphaZ
fmla z13.d, p0/m, z21.d, alphaZ
st1d z12.d, p0, [pCRow2]
st1d z13.d, p0, [pCRow2, #1, mul vl]
ld1d z14.d, p0/z, [pCRow1]
ld1d z15.d, p0/z, [pCRow1, #1, mul vl]
fmla z14.d, p0/m, z22.d, alphaZ
fmla z15.d, p0/m, z23.d, alphaZ
st1d z14.d, p0, [pCRow1]
st1d z15.d, p0, [pCRow1, #1, mul vl]
add pCRow0, pCRow0, vec_len, lsl #4 // pC = pC + vec_len * 8 * 2
.endm
.macro INITv2x2
dup z16.d, #0
dup z17.d, #0
dup z18.d, #0
dup z19.d, #0
.endm
.macro KERNELv2x2_SUB
ld1d z0.d, p0/z, [pA1]
ld1d z1.d, p0/z, [pA2]
add pA1, pA1, vec_len, lsl #3 // pA1 = pA1 + vec_len * 8
add pA2, pA2, vec_len, lsl #3 // pA1 = pA1 + vec_len * 8
ld1rd z8.d, p0/z, [pB]
ld1rd z9.d, p0/z, [pB, 8]
add pB, pB, 16
fmla z16.d, p0/m, z0.d, z8.d
fmla z17.d, p0/m, z1.d, z8.d
fmla z18.d, p0/m, z0.d, z9.d
fmla z19.d, p0/m, z1.d, z9.d
.endm
.macro SAVEv2x2
add pCRow1, pCRow0, LDC
ld1d z8.d, p0/z, [pCRow0]
ld1d z9.d, p0/z, [pCRow0, #1, mul vl]
fmla z8.d, p0/m, z16.d, alphaZ
fmla z9.d, p0/m, z17.d, alphaZ
st1d z8.d, p0, [pCRow0]
st1d z9.d, p0, [pCRow0, #1, mul vl]
ld1d z10.d, p0/z, [pCRow1]
ld1d z11.d, p0/z, [pCRow1, #1, mul vl]
fmla z10.d, p0/m, z18.d, alphaZ
fmla z11.d, p0/m, z19.d, alphaZ
st1d z10.d, p0, [pCRow1]
st1d z11.d, p0, [pCRow1, #1, mul vl]
add pCRow0, pCRow0, vec_len, lsl #4 // pC = pC + vec_len * 8 * 2
.endm
.macro INITv2x1
dup z16.d, #0
dup z17.d, #0
.endm
.macro KERNELv2x1_SUB
ld1d z0.d, p0/z, [pA1]
ld1d z1.d, p0/z, [pA2]
add pA1, pA1, vec_len, lsl #3 // pA1 = pA1 + vec_len * 8
add pA2, pA2, vec_len, lsl #3 // pA1 = pA1 + vec_len * 8
ld1rd z8.d, p0/z, [pB]
add pB, pB, 8
fmla z16.d, p0/m, z0.d, z8.d
fmla z17.d, p0/m, z1.d, z8.d
.endm
.macro SAVEv2x1
add pCRow1, pCRow0, LDC
ld1d z8.d, p0/z, [pCRow0]
ld1d z9.d, p0/z, [pCRow0, #1, mul vl]
fmla z8.d, p0/m, z16.d, alphaZ
fmla z9.d, p0/m, z17.d, alphaZ
st1d z8.d, p0, [pCRow0]
st1d z9.d, p0, [pCRow0, #1, mul vl]
add pCRow0, pCRow0, vec_len, lsl #4 // pC = pC + vec_len * 8 * 2
.endm
.macro INITv1x8
dup z16.d, #0
dup z17.d, #0
dup z18.d, #0
dup z19.d, #0
dup z20.d, #0
dup z21.d, #0
dup z22.d, #0
dup z23.d, #0
.endm
.macro KERNELv1x8_I
ld1d z0.d, p1/z, [pA1]
ld1d z1.d, p1/z, [pA1, lanes, lsl #3] // next one
add pA1, pA1, lanes, lsl #4 // pA1 = pA1 + lanes * 2 * 8
ld1rd z8.d, p0/z, [pB]
ld1rd z9.d, p0/z, [pB, 8]
ld1rd z10.d, p0/z, [pB, 16]
ld1rd z11.d, p0/z, [pB, 24]
ld1rd z12.d, p0/z, [pB, 32]
ld1rd z13.d, p0/z, [pB, 40]
ld1rd z14.d, p0/z, [pB, 48]
ld1rd z15.d, p0/z, [pB, 56]
add pB, pB, 64
fmla z16.d, p1/m, z0.d, z8.d
ld1rd z8.d, p0/z, [pB]
fmla z17.d, p1/m, z0.d, z9.d
ld1rd z9.d, p0/z, [pB, 8]
fmla z18.d, p1/m, z0.d, z10.d
ld1rd z10.d, p0/z, [pB, 16]
fmla z19.d, p1/m, z0.d, z11.d
ld1rd z11.d, p0/z, [pB, 24]
fmla z20.d, p1/m, z0.d, z12.d
ld1rd z12.d, p0/z, [pB, 32]
fmla z21.d, p1/m, z0.d, z13.d
ld1rd z13.d, p0/z, [pB, 40]
fmla z22.d, p1/m, z0.d, z14.d
ld1rd z14.d, p0/z, [pB, 48]
fmla z23.d, p1/m, z0.d, z15.d
ld1rd z15.d, p0/z, [pB, 56]
add pB, pB, 64
.endm
.macro KERNELv1x8_M1
ld1d z1.d, p1/z, [pA1]
add pA1, pA1, lanes, lsl #3 // pA1 = pA1 + lanes * 8
fmla z16.d, p1/m, z0.d, z8.d
ld1rd z8.d, p0/z, [pB]
fmla z17.d, p1/m, z0.d, z9.d
ld1rd z9.d, p0/z, [pB, 8]
fmla z18.d, p1/m, z0.d, z10.d
ld1rd z10.d, p0/z, [pB, 16]
fmla z19.d, p1/m, z0.d, z11.d
ld1rd z11.d, p0/z, [pB, 24]
fmla z20.d, p1/m, z0.d, z12.d
ld1rd z12.d, p0/z, [pB, 32]
fmla z21.d, p1/m, z0.d, z13.d
ld1rd z13.d, p0/z, [pB, 40]
fmla z22.d, p1/m, z0.d, z14.d
ld1rd z14.d, p0/z, [pB, 48]
fmla z23.d, p1/m, z0.d, z15.d
ld1rd z15.d, p0/z, [pB, 56]
add pB, pB, 64
.endm
.macro KERNELv1x8_M2
ld1d z0.d, p1/z, [pA1]
add pA1, pA1, lanes, lsl #3 // pA1 = pA1 + lanes * 8
fmla z16.d, p1/m, z1.d, z8.d
ld1rd z8.d, p0/z, [pB]
fmla z17.d, p1/m, z1.d, z9.d
ld1rd z9.d, p0/z, [pB, 8]
fmla z18.d, p1/m, z1.d, z10.d
ld1rd z10.d, p0/z, [pB, 16]
fmla z19.d, p1/m, z1.d, z11.d
ld1rd z11.d, p0/z, [pB, 24]
fmla z20.d, p1/m, z1.d, z12.d
ld1rd z12.d, p0/z, [pB, 32]
fmla z21.d, p1/m, z1.d, z13.d
ld1rd z13.d, p0/z, [pB, 40]
fmla z22.d, p1/m, z1.d, z14.d
ld1rd z14.d, p0/z, [pB, 48]
fmla z23.d, p1/m, z1.d, z15.d
ld1rd z15.d, p0/z, [pB, 56]
add pB, pB, 64
.endm
.macro KERNELv1x8_E
fmla z16.d, p1/m, z1.d, z8.d
fmla z17.d, p1/m, z1.d, z9.d
fmla z18.d, p1/m, z1.d, z10.d
fmla z19.d, p1/m, z1.d, z11.d
fmla z20.d, p1/m, z1.d, z12.d
fmla z21.d, p1/m, z1.d, z13.d
fmla z22.d, p1/m, z1.d, z14.d
fmla z23.d, p1/m, z1.d, z15.d
.endm
.macro KERNELv1x8_SUB
ld1d z0.d, p1/z, [pA1]
add pA1, pA1, lanes, lsl #3 // pA1 = pA1 + lanes * 8
ld1rd z8.d, p0/z, [pB]
ld1rd z9.d, p0/z, [pB, 8]
ld1rd z10.d, p0/z, [pB, 16]
ld1rd z11.d, p0/z, [pB, 24]
ld1rd z12.d, p0/z, [pB, 32]
ld1rd z13.d, p0/z, [pB, 40]
ld1rd z14.d, p0/z, [pB, 48]
ld1rd z15.d, p0/z, [pB, 56]
add pB, pB, 64
fmla z16.d, p1/m, z0.d, z8.d
fmla z17.d, p1/m, z0.d, z9.d
fmla z18.d, p1/m, z0.d, z10.d
fmla z19.d, p1/m, z0.d, z11.d
fmla z20.d, p1/m, z0.d, z12.d
fmla z21.d, p1/m, z0.d, z13.d
fmla z22.d, p1/m, z0.d, z14.d
fmla z23.d, p1/m, z0.d, z15.d
.endm
.macro SAVEv1x8
add pCRow1, pCRow0, LDC
ld1d z24.d, p1/z, [pCRow0]
fmla z24.d, p1/m, z16.d, alphaZ
st1d z24.d, p1, [pCRow0]
add pCRow2, pCRow1, LDC
ld1d z25.d, p1/z, [pCRow1]
fmla z25.d, p1/m, z17.d, alphaZ
st1d z25.d, p1, [pCRow1]
add pCRow1, pCRow2, LDC
ld1d z26.d, p1/z, [pCRow2]
fmla z26.d, p1/m, z18.d, alphaZ
st1d z26.d, p1, [pCRow2]
add pCRow2, pCRow1, LDC
ld1d z27.d, p1/z, [pCRow1]
fmla z27.d, p1/m, z19.d, alphaZ
st1d z27.d, p1, [pCRow1]
add pCRow1, pCRow2, LDC
ld1d z28.d, p1/z, [pCRow2]
fmla z28.d, p1/m, z20.d, alphaZ
st1d z28.d, p1, [pCRow2]
add pCRow2, pCRow1, LDC
ld1d z29.d, p1/z, [pCRow1]
fmla z29.d, p1/m, z21.d, alphaZ
st1d z29.d, p1, [pCRow1]
add pCRow1, pCRow2, LDC
ld1d z30.d, p1/z, [pCRow2]
fmla z30.d, p1/m, z22.d, alphaZ
st1d z30.d, p1, [pCRow2]
ld1d z31.d, p1/z, [pCRow1]
fmla z31.d, p1/m, z23.d, alphaZ
st1d z31.d, p1, [pCRow1]
add pCRow0, pCRow0, lanes, lsl #3 // pC = pC + lanes * 8
.endm
/******************************************************************************/
.macro INITv1x4
dup z16.d, #0
dup z17.d, #0
dup z18.d, #0
dup z19.d, #0
.endm
.macro KERNELv1x4_SUB
ld1d z0.d, p1/z, [pA1]
add pA1, pA1, lanes, lsl #3 // pA1 = pA1 + lanes * 8
ld1rd z8.d, p0/z, [pB]
ld1rd z9.d, p0/z, [pB, 8]
ld1rd z10.d, p0/z, [pB, 16]
ld1rd z11.d, p0/z, [pB, 24]
add pB, pB, 32
fmla z16.d, p1/m, z0.d, z8.d
fmla z17.d, p1/m, z0.d, z9.d
fmla z18.d, p1/m, z0.d, z10.d
fmla z19.d, p1/m, z0.d, z11.d
.endm
.macro SAVEv1x4
add pCRow1, pCRow0, LDC
ld1d z24.d, p1/z, [pCRow0]
fmla z24.d, p1/m, z16.d, alphaZ
st1d z24.d, p1, [pCRow0]
add pCRow2, pCRow1, LDC
ld1d z25.d, p1/z, [pCRow1]
fmla z25.d, p1/m, z17.d, alphaZ
st1d z25.d, p1, [pCRow1]
add pCRow1, pCRow2, LDC
ld1d z26.d, p1/z, [pCRow2]
fmla z26.d, p1/m, z18.d, alphaZ
st1d z26.d, p1, [pCRow2]
ld1d z27.d, p1/z, [pCRow1]
fmla z27.d, p1/m, z19.d, alphaZ
st1d z27.d, p1, [pCRow1]
add pCRow0, pCRow0, lanes, lsl #3 // pC = pC + lanes * 8
.endm
/******************************************************************************/
.macro INITv1x2
dup z16.d, #0
dup z17.d, #0
.endm
.macro KERNELv1x2_SUB
ld1d z0.d, p1/z, [pA1]
add pA1, pA1, lanes, lsl #3 // pA1 = pA1 + lanes * 8
ld1rd z8.d, p0/z, [pB]
ld1rd z9.d, p0/z, [pB, 8]
add pB, pB, 16
fmla z16.d, p1/m, z0.d, z8.d
fmla z17.d, p1/m, z0.d, z9.d
.endm
.macro SAVEv1x2
add pCRow1, pCRow0, LDC
ld1d z24.d, p1/z, [pCRow0]
fmla z24.d, p1/m, z16.d, alphaZ
st1d z24.d, p1, [pCRow0]
ld1d z25.d, p1/z, [pCRow1]
fmla z25.d, p1/m, z17.d, alphaZ
st1d z25.d, p1, [pCRow1]
add pCRow0, pCRow0, lanes, lsl #3 // pC = pC + lanes * 8
.endm
/******************************************************************************/
.macro INITv1x1
dup z16.d, #0
.endm
.macro KERNELv1x1_SUB
ld1d z0.d, p1/z, [pA1]
add pA1, pA1, lanes, lsl #3 // pA1 = pA1 + lanes * 8
ld1rd z8.d, p0/z, [pB]
add pB, pB, 8
fmla z16.d, p1/m, z0.d, z8.d
.endm
.macro SAVEv1x1
ld1d z24.d, p1/z, [pCRow0]
fmla z24.d, p1/m, z16.d, alphaZ
st1d z24.d, p1, [pCRow0]
add pCRow0, pCRow0, lanes, lsl #3 // pC = pC + lanes * 8
.endm
/*******************************************************************************
* End of macro definitions
*******************************************************************************/
PROLOGUE
.align 5
add sp, sp, #-(11 * 16)
stp d8, d9, [sp, #(0 * 16)]
stp d10, d11, [sp, #(1 * 16)]
stp d12, d13, [sp, #(2 * 16)]
stp d14, d15, [sp, #(3 * 16)]
stp d16, d17, [sp, #(4 * 16)]
stp x18, x19, [sp, #(5 * 16)]
stp x20, x21, [sp, #(6 * 16)]
stp x22, x23, [sp, #(7 * 16)]
stp x24, x25, [sp, #(8 * 16)]
stp x26, x27, [sp, #(9 * 16)]
str x28, [sp, #(10 * 16)]
fmov alpha, d0
dup alphaZ, alpha
cntd vec_len
lsl vec_lenx2, vec_len, #1
lsl LDC, LDC, #3 // ldc = ldc * 8
ptrue p0.d // create true predicate
mov pB, origPB
// Loop over N
mov counterJ, origN
asr counterJ, counterJ, #3 // J = J / 8
cmp counterJ, #0
ble .Ldgemm_kernel_L4_BEGIN
/******************************************************************************/
/* Repeat this as long as there are 8 left in N */
.align 5
.Ldgemm_kernel_L8_BEGIN:
mov pCRow0, pC
add pC, pC, LDC, lsl #3 // add 8 x LDC
mov pA1, origPA // pA1 = start of A array
.Ldgemm_kernel_L8_Mv2_BEGIN:
mov counterI, #0
cmp origM, vec_lenx2 // Check if M < 2*SVE_LEN
blt .Ldgemm_kernel_L8_Mv1_BEGIN
mov counterI, origM
/* Until we have at least 2*SVE_LEN iters left in M, we do them with V2*8 kernel */
mul temp, vec_len, origK // generate address of pA2
add pA2, pA1, temp, lsl #3 // pA1 = start of A array
.align 5
.Ldgemm_kernel_L8_Mv2_20:
mov pB, origPB
INITv2x8 // fill with zeros
asr counterL , origK, #3 // L = K / 8
cmp counterL , #2 // is there at least 4 to do?
blt .Ldgemm_kernel_L8_Mv2_32
KERNELv2x8_I
KERNELv2x8_M2
KERNELv2x8_M1
KERNELv2x8_M2
KERNELv2x8_M1
KERNELv2x8_M2
KERNELv2x8_M1
KERNELv2x8_M2
subs counterL, counterL, #2 // subtract 2
ble .Ldgemm_kernel_L8_Mv2_22a
.align 5
.Ldgemm_kernel_L8_Mv2_22:
KERNELv2x8_M1
KERNELv2x8_M2
KERNELv2x8_M1
KERNELv2x8_M2
KERNELv2x8_M1
KERNELv2x8_M2
KERNELv2x8_M1
KERNELv2x8_M2
subs counterL, counterL, #1
bgt .Ldgemm_kernel_L8_Mv2_22
.align 5
.Ldgemm_kernel_L8_Mv2_22a:
KERNELv2x8_M1
KERNELv2x8_M2
KERNELv2x8_M1
KERNELv2x8_M2
KERNELv2x8_M1
KERNELv2x8_M2
KERNELv2x8_M1
KERNELv2x8_E
b .Ldgemm_kernel_L8_Mv2_44
.align 5
.Ldgemm_kernel_L8_Mv2_32:
tst counterL, #1
ble .Ldgemm_kernel_L8_Mv2_40
KERNELv2x8_I
KERNELv2x8_M2
KERNELv2x8_M1
KERNELv2x8_M2
KERNELv2x8_M1
KERNELv2x8_M2
KERNELv2x8_M1
KERNELv2x8_E
b .Ldgemm_kernel_L8_Mv2_44
.Ldgemm_kernel_L8_Mv2_40:
INITv2x8
.Ldgemm_kernel_L8_Mv2_44:
ands counterL , origK, #7
ble .Ldgemm_kernel_L8_Mv2_100
.align 5
.Ldgemm_kernel_L8_Mv2_46:
KERNELv2x8_SUB
subs counterL, counterL, #1
bne .Ldgemm_kernel_L8_Mv2_46
.Ldgemm_kernel_L8_Mv2_100:
SAVEv2x8
mov pA1, pA2 // pA1 = pA2
mul temp, vec_len, origK // generate address of pA2