-
Notifications
You must be signed in to change notification settings - Fork 1.5k
/
Copy pathcgemm_kernel_8x2_sandy.S
2353 lines (1759 loc) · 55.1 KB
/
cgemm_kernel_8x2_sandy.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) 2013, 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.
**********************************************************************************/
/*********************************************************************
* 2014/07/29 Saar
* BLASTEST : OK
* CTEST : OK
* TEST : OK
*
* 2013/10/28 Saar
* Parameter:
* CGEMM_DEFAULT_UNROLL_N 2
* CGEMM_DEFAULT_UNROLL_M 8
* CGEMM_DEFAULT_P 768
* CGEMM_DEFAULT_Q 512
* A_PR1 512
* B_PR1 512
*
* 2014/07/29 Saar
* Performance at 6192x6192x6192:
* 1 thread: 49 GFLOPS (MKL: 52)
* 2 threads: 99 GFLOPS (MKL: 102)
* 3 threads: 148 GFLOPS (MKL: 150)
* 4 threads: 195 GFLOPS (MKL: 194)
* 8 threads: 354 GFLOPS (MKL: 317)
*
*
*********************************************************************/
#define ASSEMBLER
#include "common.h"
#define OLD_M %rdi
#define OLD_N %rsi
#define M %r13
#define J %r14
#define OLD_K %rdx
#define A %rcx
#define B %r8
#define C %r9
#define LDC %r10
#define I %r11
#define AO %rdi
#define BO %rsi
#define CO1 %r15
#define K %r12
#define BI %rbp
#define SP %rbx
#define BO1 %rdi
#define BO2 %r15
#ifndef WINDOWS_ABI
#define STACKSIZE 96
#else
#define STACKSIZE 320
#define OLD_ALPHA_I 40 + STACKSIZE(%rsp)
#define OLD_A 48 + STACKSIZE(%rsp)
#define OLD_B 56 + STACKSIZE(%rsp)
#define OLD_C 64 + STACKSIZE(%rsp)
#define OLD_LDC 72 + STACKSIZE(%rsp)
#define OLD_OFFSET 80 + STACKSIZE(%rsp)
#endif
#define L_BUFFER_SIZE 8192
#define Ndiv6 24(%rsp)
#define Nmod6 32(%rsp)
#define N 40(%rsp)
#define ALPHA_R 48(%rsp)
#define ALPHA_I 56(%rsp)
#define OFFSET 64(%rsp)
#define KK 72(%rsp)
#define KKK 80(%rsp)
#define BUFFER1 128(%rsp)
#if defined(OS_WINDOWS)
#if L_BUFFER_SIZE > 16384
#define STACK_TOUCH \
movl $ 0, 4096 * 4(%rsp);\
movl $ 0, 4096 * 3(%rsp);\
movl $ 0, 4096 * 2(%rsp);\
movl $ 0, 4096 * 1(%rsp);
#elif L_BUFFER_SIZE > 12288
#define STACK_TOUCH \
movl $ 0, 4096 * 3(%rsp);\
movl $ 0, 4096 * 2(%rsp);\
movl $ 0, 4096 * 1(%rsp);
#elif L_BUFFER_SIZE > 8192
#define STACK_TOUCH \
movl $ 0, 4096 * 2(%rsp);\
movl $ 0, 4096 * 1(%rsp);
#elif L_BUFFER_SIZE > 4096
#define STACK_TOUCH \
movl $ 0, 4096 * 1(%rsp);
#else
#define STACK_TOUCH
#endif
#else
#define STACK_TOUCH
#endif
#if defined(NN) || defined(NT) || defined(TN) || defined(TT)
#define VFMADDPS_YR( y0,y1,y2 ) \
vmulps y1,y2,%ymm2;\
vaddps y0,%ymm2,y0
#define VFMADDPS_YI( y0,y1,y2 ) \
vmulps y1,y2,%ymm3;\
vaddps y0,%ymm3,y0
#define VFMADDPS_R( y0,y1,y2 ) \
vmulps y1,y2,%xmm2;\
vaddps y0,%xmm2,y0
#define VFMADDPS_I( y0,y1,y2 ) \
vmulps y1,y2,%xmm3;\
vaddps y0,%xmm3,y0
#elif defined(RN) || defined(RT) || defined(CN) || defined(CT)
#define VFMADDPS_YR( y0,y1,y2 ) \
vmulps y1,y2,%ymm2;\
vsubps %ymm2,y0,y0
#define VFMADDPS_YI( y0,y1,y2 ) \
vmulps y1,y2,%ymm3;\
vaddps y0,%ymm3,y0
#define VFMADDPS_R( y0,y1,y2 ) \
vmulps y1,y2,%xmm2;\
vsubps %xmm2,y0,y0
#define VFMADDPS_I( y0,y1,y2 ) \
vmulps y1,y2,%xmm3;\
vaddps y0,%xmm3,y0
#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
#define VFMADDPS_YR( y0,y1,y2 ) \
vmulps y1,y2,%ymm2;\
vaddps y0,%ymm2,y0
#define VFMADDPS_YI( y0,y1,y2 ) \
vmulps y1,y2,%ymm3;\
vsubps %ymm3,y0,y0
#define VFMADDPS_R( y0,y1,y2 ) \
vmulps y1,y2,%xmm2;\
vaddps y0,%xmm2,y0
#define VFMADDPS_I( y0,y1,y2 ) \
vmulps y1,y2,%xmm3;\
vsubps %xmm3,y0,y0
#else
#define VFMADDPS_YR( y0,y1,y2 ) \
vmulps y1,y2,%ymm2;\
vsubps %ymm2,y0,y0
#define VFMADDPS_YI( y0,y1,y2 ) \
vmulps y1,y2,%ymm3;\
vsubps %ymm3,y0,y0
#define VFMADDPS_R( y0,y1,y2 ) \
vmulps y1,y2,%xmm2;\
vsubps %xmm2,y0,y0
#define VFMADDPS_I( y0,y1,y2 ) \
vmulps y1,y2,%xmm3;\
vsubps %xmm3,y0,y0
#endif
#define A_PR1 512
#define B_PR1 512
/***************************************************************************************************************************/
.macro KERNEL8x2_1
vmovups -16 * SIZE(AO, %rax, SIZE), %ymm0
vbroadcastss -8 * SIZE(BO, BI, SIZE), %ymm4
vmovups -8 * SIZE(AO, %rax, SIZE), %ymm1
vbroadcastss -7 * SIZE(BO, BI, SIZE), %ymm5
prefetcht0 A_PR1(AO, %rax, SIZE)
VFMADDPS_YR( %ymm8,%ymm4,%ymm0 )
vbroadcastss -6 * SIZE(BO, BI, SIZE), %ymm6
VFMADDPS_YI( %ymm9,%ymm5,%ymm0 )
vbroadcastss -5 * SIZE(BO, BI, SIZE), %ymm7
VFMADDPS_YR( %ymm12,%ymm4,%ymm1 )
VFMADDPS_YI( %ymm13,%ymm5,%ymm1 )
VFMADDPS_YR( %ymm10,%ymm6,%ymm0 )
vbroadcastss -4 * SIZE(BO, BI, SIZE), %ymm4
VFMADDPS_YI( %ymm11,%ymm7,%ymm0 )
vbroadcastss -3 * SIZE(BO, BI, SIZE), %ymm5
VFMADDPS_YR( %ymm14,%ymm6,%ymm1 )
vmovups 0 * SIZE(AO, %rax, SIZE), %ymm0
VFMADDPS_YI( %ymm15,%ymm7,%ymm1 )
vmovups 8 * SIZE(AO, %rax, SIZE), %ymm1
prefetcht0 A_PR1+64(AO, %rax, SIZE)
VFMADDPS_YR( %ymm8,%ymm4,%ymm0 )
vbroadcastss -2 * SIZE(BO, BI, SIZE), %ymm6
VFMADDPS_YI( %ymm9,%ymm5,%ymm0 )
vbroadcastss -1 * SIZE(BO, BI, SIZE), %ymm7
VFMADDPS_YR( %ymm12,%ymm4,%ymm1 )
VFMADDPS_YI( %ymm13,%ymm5,%ymm1 )
VFMADDPS_YR( %ymm10,%ymm6,%ymm0 )
vbroadcastss 0 * SIZE(BO, BI, SIZE), %ymm4
VFMADDPS_YI( %ymm11,%ymm7,%ymm0 )
vbroadcastss 1 * SIZE(BO, BI, SIZE), %ymm5
VFMADDPS_YR( %ymm14,%ymm6,%ymm1 )
vmovups 16 * SIZE(AO, %rax, SIZE), %ymm0
VFMADDPS_YI( %ymm15,%ymm7,%ymm1 )
vmovups 24 * SIZE(AO, %rax, SIZE), %ymm1
prefetcht0 A_PR1+128(AO, %rax, SIZE)
VFMADDPS_YR( %ymm8,%ymm4,%ymm0 )
vbroadcastss 2 * SIZE(BO, BI, SIZE), %ymm6
VFMADDPS_YI( %ymm9,%ymm5,%ymm0 )
vbroadcastss 3 * SIZE(BO, BI, SIZE), %ymm7
VFMADDPS_YR( %ymm12,%ymm4,%ymm1 )
VFMADDPS_YI( %ymm13,%ymm5,%ymm1 )
VFMADDPS_YR( %ymm10,%ymm6,%ymm0 )
vbroadcastss 4 * SIZE(BO, BI, SIZE), %ymm4
VFMADDPS_YI( %ymm11,%ymm7,%ymm0 )
vbroadcastss 5 * SIZE(BO, BI, SIZE), %ymm5
VFMADDPS_YR( %ymm14,%ymm6,%ymm1 )
vmovups 32 * SIZE(AO, %rax, SIZE), %ymm0
VFMADDPS_YI( %ymm15,%ymm7,%ymm1 )
vmovups 40 * SIZE(AO, %rax, SIZE), %ymm1
prefetcht0 A_PR1+192(AO, %rax, SIZE)
VFMADDPS_YR( %ymm8,%ymm4,%ymm0 )
vbroadcastss 6 * SIZE(BO, BI, SIZE), %ymm6
VFMADDPS_YI( %ymm9,%ymm5,%ymm0 )
vbroadcastss 7 * SIZE(BO, BI, SIZE), %ymm7
VFMADDPS_YR( %ymm12,%ymm4,%ymm1 )
VFMADDPS_YI( %ymm13,%ymm5,%ymm1 )
VFMADDPS_YR( %ymm10,%ymm6,%ymm0 )
VFMADDPS_YI( %ymm11,%ymm7,%ymm0 )
addq $ 16, BI
VFMADDPS_YR( %ymm14,%ymm6,%ymm1 )
VFMADDPS_YI( %ymm15,%ymm7,%ymm1 )
addq $ 64, %rax
.endm
.macro KERNEL8x2_SUB
vmovups -16 * SIZE(AO, %rax, SIZE), %ymm0
vmovups -8 * SIZE(AO, %rax, SIZE), %ymm1
vbroadcastss -8 * SIZE(BO, BI, SIZE), %ymm4
vbroadcastss -7 * SIZE(BO, BI, SIZE), %ymm5
VFMADDPS_YR( %ymm8,%ymm4,%ymm0 )
vbroadcastss -6 * SIZE(BO, BI, SIZE), %ymm6
VFMADDPS_YI( %ymm9,%ymm5,%ymm0 )
vbroadcastss -5 * SIZE(BO, BI, SIZE), %ymm7
VFMADDPS_YR( %ymm12,%ymm4,%ymm1 )
VFMADDPS_YI( %ymm13,%ymm5,%ymm1 )
VFMADDPS_YR( %ymm10,%ymm6,%ymm0 )
VFMADDPS_YI( %ymm11,%ymm7,%ymm0 )
VFMADDPS_YR( %ymm14,%ymm6,%ymm1 )
VFMADDPS_YI( %ymm15,%ymm7,%ymm1 )
addq $ 4 , BI
addq $ 16, %rax
.endm
.macro SAVE8x2
vbroadcastss ALPHA_R, %ymm0
vbroadcastss ALPHA_I, %ymm1
// swap high and low 64 bytes
vshufps $ 0xb1, %ymm9 , %ymm9, %ymm9
vshufps $ 0xb1, %ymm11, %ymm11, %ymm11
vshufps $ 0xb1, %ymm13, %ymm13, %ymm13
vshufps $ 0xb1, %ymm15, %ymm15, %ymm15
#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
defined(NR) || defined(NC) || defined(TR) || defined(TC)
vaddsubps %ymm9, %ymm8 , %ymm8
vaddsubps %ymm11,%ymm10, %ymm10
vaddsubps %ymm13,%ymm12, %ymm12
vaddsubps %ymm15,%ymm14, %ymm14
vshufps $ 0xb1, %ymm8 , %ymm8, %ymm9
vshufps $ 0xb1, %ymm10, %ymm10, %ymm11
vshufps $ 0xb1, %ymm12, %ymm12, %ymm13
vshufps $ 0xb1, %ymm14, %ymm14, %ymm15
#else
vaddsubps %ymm8, %ymm9 ,%ymm9
vaddsubps %ymm10, %ymm11,%ymm11
vaddsubps %ymm12, %ymm13,%ymm13
vaddsubps %ymm14, %ymm15,%ymm15
vmovaps %ymm9, %ymm8
vmovaps %ymm11, %ymm10
vmovaps %ymm13, %ymm12
vmovaps %ymm15, %ymm14
// swap high and low 64 bytes
vshufps $ 0xb1, %ymm9 , %ymm9, %ymm9
vshufps $ 0xb1, %ymm11, %ymm11, %ymm11
vshufps $ 0xb1, %ymm13, %ymm13, %ymm13
vshufps $ 0xb1, %ymm15, %ymm15, %ymm15
#endif
// multiply with ALPHA_R
vmulps %ymm8 , %ymm0, %ymm8
vmulps %ymm10, %ymm0, %ymm10
vmulps %ymm12, %ymm0, %ymm12
vmulps %ymm14, %ymm0, %ymm14
// multiply with ALPHA_I
vmulps %ymm9 , %ymm1, %ymm9
vmulps %ymm11, %ymm1, %ymm11
vmulps %ymm13, %ymm1, %ymm13
vmulps %ymm15, %ymm1, %ymm15
vaddsubps %ymm9, %ymm8 , %ymm8
vaddsubps %ymm11,%ymm10, %ymm10
vaddsubps %ymm13,%ymm12, %ymm12
vaddsubps %ymm15,%ymm14, %ymm14
#ifndef TRMMKERNEL
vaddps (CO1), %ymm8 , %ymm8
vaddps 8 * SIZE(CO1), %ymm12, %ymm12
vaddps (CO1, LDC), %ymm10, %ymm10
vaddps 8 * SIZE(CO1, LDC), %ymm14, %ymm14
#endif
vmovups %ymm8 , (CO1)
vmovups %ymm12 , 8 * SIZE(CO1)
vmovups %ymm10 , (CO1, LDC)
vmovups %ymm14 , 8 * SIZE(CO1, LDC)
prefetcht0 64(CO1)
prefetcht0 64(CO1, LDC)
.endm
/***************************************************************************************************************************/
.macro KERNEL4x2_SUB
vmovups -16 * SIZE(AO, %rax, SIZE), %xmm0
vbroadcastss -8 * SIZE(BO, BI, SIZE), %xmm4
VFMADDPS_R( %xmm8,%xmm4,%xmm0 )
vmovups -12 * SIZE(AO, %rax, SIZE), %xmm1
VFMADDPS_R( %xmm12,%xmm4,%xmm1 )
vbroadcastss -7 * SIZE(BO, BI, SIZE), %xmm5
VFMADDPS_I( %xmm9,%xmm5,%xmm0 )
VFMADDPS_I( %xmm13,%xmm5,%xmm1 )
vbroadcastss -6 * SIZE(BO, BI, SIZE), %xmm6
VFMADDPS_R( %xmm10,%xmm6,%xmm0 )
VFMADDPS_R( %xmm14,%xmm6,%xmm1 )
vbroadcastss -5 * SIZE(BO, BI, SIZE), %xmm7
VFMADDPS_I( %xmm11,%xmm7,%xmm0 )
VFMADDPS_I( %xmm15,%xmm7,%xmm1 )
addq $ 4, BI
addq $ 8, %rax
.endm
.macro SAVE4x2
vbroadcastss ALPHA_R, %xmm0
vbroadcastss ALPHA_I, %xmm1
// swap high and low 64 bytes
vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
vshufps $ 0xb1, %xmm11, %xmm11, %xmm11
vshufps $ 0xb1, %xmm13, %xmm13, %xmm13
vshufps $ 0xb1, %xmm15, %xmm15, %xmm15
#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
defined(NR) || defined(NC) || defined(TR) || defined(TC)
vaddsubps %xmm9, %xmm8 , %xmm8
vaddsubps %xmm11,%xmm10, %xmm10
vaddsubps %xmm13,%xmm12, %xmm12
vaddsubps %xmm15,%xmm14, %xmm14
vshufps $ 0xb1, %xmm8 , %xmm8, %xmm9
vshufps $ 0xb1, %xmm10, %xmm10, %xmm11
vshufps $ 0xb1, %xmm12, %xmm12, %xmm13
vshufps $ 0xb1, %xmm14, %xmm14, %xmm15
#else
vaddsubps %xmm8, %xmm9 ,%xmm9
vaddsubps %xmm10, %xmm11,%xmm11
vaddsubps %xmm12, %xmm13,%xmm13
vaddsubps %xmm14, %xmm15,%xmm15
vmovaps %xmm9, %xmm8
vmovaps %xmm11, %xmm10
vmovaps %xmm13, %xmm12
vmovaps %xmm15, %xmm14
// swap high and low 64 bytes
vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
vshufps $ 0xb1, %xmm11, %xmm11, %xmm11
vshufps $ 0xb1, %xmm13, %xmm13, %xmm13
vshufps $ 0xb1, %xmm15, %xmm15, %xmm15
#endif
// multiply with ALPHA_R
vmulps %xmm8 , %xmm0, %xmm8
vmulps %xmm10, %xmm0, %xmm10
vmulps %xmm12, %xmm0, %xmm12
vmulps %xmm14, %xmm0, %xmm14
// multiply with ALPHA_I
vmulps %xmm9 , %xmm1, %xmm9
vmulps %xmm11, %xmm1, %xmm11
vmulps %xmm13, %xmm1, %xmm13
vmulps %xmm15, %xmm1, %xmm15
vaddsubps %xmm9, %xmm8 , %xmm8
vaddsubps %xmm11,%xmm10, %xmm10
vaddsubps %xmm13,%xmm12, %xmm12
vaddsubps %xmm15,%xmm14, %xmm14
#ifndef TRMMKERNEL
vaddps (CO1), %xmm8 , %xmm8
vaddps 4 * SIZE(CO1), %xmm12, %xmm12
vaddps (CO1, LDC), %xmm10, %xmm10
vaddps 4 * SIZE(CO1, LDC), %xmm14, %xmm14
#endif
vmovups %xmm8 , (CO1)
vmovups %xmm12 , 4 * SIZE(CO1)
vmovups %xmm10 , (CO1, LDC)
vmovups %xmm14 , 4 * SIZE(CO1, LDC)
.endm
/************************************************************************************************/
.macro KERNEL2x2_SUB
vmovups -16 * SIZE(AO, %rax, SIZE), %xmm0
vbroadcastss -8 * SIZE(BO, BI, SIZE), %xmm4
VFMADDPS_R( %xmm8,%xmm4,%xmm0 )
vbroadcastss -7 * SIZE(BO, BI, SIZE), %xmm5
VFMADDPS_I( %xmm9,%xmm5,%xmm0 )
vbroadcastss -6 * SIZE(BO, BI, SIZE), %xmm6
VFMADDPS_R( %xmm10,%xmm6,%xmm0 )
vbroadcastss -5 * SIZE(BO, BI, SIZE), %xmm7
VFMADDPS_I( %xmm11,%xmm7,%xmm0 )
addq $ 4, BI
addq $ 4, %rax
.endm
.macro SAVE2x2
vbroadcastss ALPHA_R, %xmm0
vbroadcastss ALPHA_I, %xmm1
// swap high and low 4 bytes
vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
vshufps $ 0xb1, %xmm11, %xmm11, %xmm11
#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
defined(NR) || defined(NC) || defined(TR) || defined(TC)
vaddsubps %xmm9, %xmm8 , %xmm8
vaddsubps %xmm11,%xmm10, %xmm10
vshufps $ 0xb1, %xmm8 , %xmm8, %xmm9
vshufps $ 0xb1, %xmm10, %xmm10, %xmm11
#else
vaddsubps %xmm8, %xmm9 ,%xmm9
vaddsubps %xmm10, %xmm11,%xmm11
vmovaps %xmm9, %xmm8
vmovaps %xmm11, %xmm10
// swap high and low 4 bytes
vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
vshufps $ 0xb1, %xmm11, %xmm11, %xmm11
#endif
// multiply with ALPHA_R
vmulps %xmm8 , %xmm0, %xmm8
vmulps %xmm10, %xmm0, %xmm10
// multiply with ALPHA_I
vmulps %xmm9 , %xmm1, %xmm9
vmulps %xmm11, %xmm1, %xmm11
vaddsubps %xmm9, %xmm8 , %xmm8
vaddsubps %xmm11,%xmm10, %xmm10
#ifndef TRMMKERNEL
vaddps (CO1), %xmm8 , %xmm8
vaddps (CO1, LDC), %xmm10, %xmm10
#endif
vmovups %xmm8 , (CO1)
vmovups %xmm10 , (CO1, LDC)
.endm
/************************************************************************************************/
.macro KERNEL1x2_SUB
vmovsd -16 * SIZE(AO, %rax, SIZE), %xmm0
vbroadcastss -8 * SIZE(BO, BI, SIZE), %xmm4
VFMADDPS_R( %xmm8,%xmm4,%xmm0 )
vbroadcastss -7 * SIZE(BO, BI, SIZE), %xmm5
VFMADDPS_I( %xmm9,%xmm5,%xmm0 )
vbroadcastss -6 * SIZE(BO, BI, SIZE), %xmm6
VFMADDPS_R( %xmm10,%xmm6,%xmm0 )
vbroadcastss -5 * SIZE(BO, BI, SIZE), %xmm7
VFMADDPS_I( %xmm11,%xmm7,%xmm0 )
addq $ 4, BI
addq $ 2, %rax
.endm
.macro SAVE1x2
vbroadcastss ALPHA_R, %xmm0
vbroadcastss ALPHA_I, %xmm1
// swap high and low 64 bytes
vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
vshufps $ 0xb1, %xmm11, %xmm11, %xmm11
#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
defined(NR) || defined(NC) || defined(TR) || defined(TC)
vaddsubps %xmm9, %xmm8 , %xmm8
vaddsubps %xmm11,%xmm10, %xmm10
vshufps $ 0xb1, %xmm8 , %xmm8, %xmm9
vshufps $ 0xb1, %xmm10, %xmm10, %xmm11
#else
vaddsubps %xmm8, %xmm9 ,%xmm9
vaddsubps %xmm10, %xmm11,%xmm11
vmovaps %xmm9, %xmm8
vmovaps %xmm11, %xmm10
// swap high and low 64 bytes
vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
vshufps $ 0xb1, %xmm11, %xmm11, %xmm11
#endif
// multiply with ALPHA_R
vmulps %xmm8 , %xmm0, %xmm8
vmulps %xmm10, %xmm0, %xmm10
// multiply with ALPHA_I
vmulps %xmm9 , %xmm1, %xmm9
vmulps %xmm11, %xmm1, %xmm11
vaddsubps %xmm9, %xmm8 , %xmm8
vaddsubps %xmm11,%xmm10, %xmm10
#ifndef TRMMKERNEL
vmovsd (CO1), %xmm14
vaddps %xmm14, %xmm8 , %xmm8
vmovsd (CO1, LDC), %xmm15
vaddps %xmm15, %xmm10, %xmm10
#endif
vmovsd %xmm8 , (CO1)
vmovsd %xmm10 , (CO1, LDC)
.endm
/************************************************************************************************/
.macro KERNEL8x1_SUB
vmovups -16 * SIZE(AO, %rax, SIZE), %ymm0
vmovups -8 * SIZE(AO, %rax, SIZE), %ymm1
vbroadcastss -4 * SIZE(BO, BI, SIZE), %ymm4
VFMADDPS_YR( %ymm8,%ymm4,%ymm0 )
VFMADDPS_YR( %ymm12,%ymm4,%ymm1 )
vbroadcastss -3 * SIZE(BO, BI, SIZE), %ymm5
VFMADDPS_YI( %ymm9,%ymm5,%ymm0 )
VFMADDPS_YI( %ymm13,%ymm5,%ymm1 )
addq $ 2 , BI
addq $ 16, %rax
.endm
.macro SAVE8x1
vbroadcastss ALPHA_R, %ymm0
vbroadcastss ALPHA_I, %ymm1
// swap high and low 64 bytes
vshufps $ 0xb1, %ymm9 , %ymm9, %ymm9
vshufps $ 0xb1, %ymm13, %ymm13, %ymm13
#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
defined(NR) || defined(NC) || defined(TR) || defined(TC)
vaddsubps %ymm9, %ymm8 , %ymm8
vaddsubps %ymm13,%ymm12, %ymm12
vshufps $ 0xb1, %ymm8 , %ymm8, %ymm9
vshufps $ 0xb1, %ymm12, %ymm12, %ymm13
#else
vaddsubps %ymm8, %ymm9 ,%ymm9
vaddsubps %ymm12, %ymm13,%ymm13
vmovaps %ymm9, %ymm8
vmovaps %ymm13, %ymm12
// swap high and low 64 bytes
vshufps $ 0xb1, %ymm9 , %ymm9, %ymm9
vshufps $ 0xb1, %ymm13, %ymm13, %ymm13
#endif
// multiply with ALPHA_R
vmulps %ymm8 , %ymm0, %ymm8
vmulps %ymm12, %ymm0, %ymm12
// multiply with ALPHA_I
vmulps %ymm9 , %ymm1, %ymm9
vmulps %ymm13, %ymm1, %ymm13
vaddsubps %ymm9, %ymm8 , %ymm8
vaddsubps %ymm13,%ymm12, %ymm12
#ifndef TRMMKERNEL
vaddps (CO1), %ymm8 , %ymm8
vaddps 8 * SIZE(CO1), %ymm12, %ymm12
#endif
vmovups %ymm8 , (CO1)
vmovups %ymm12 , 8 * SIZE(CO1)
.endm
/************************************************************************************************/
.macro KERNEL4x1_SUB
vmovups -16 * SIZE(AO, %rax, SIZE), %xmm0
vbroadcastss -4 * SIZE(BO, BI, SIZE), %xmm4
VFMADDPS_R( %xmm8,%xmm4,%xmm0 )
vmovups -12 * SIZE(AO, %rax, SIZE), %xmm1
VFMADDPS_R( %xmm12,%xmm4,%xmm1 )
vbroadcastss -3 * SIZE(BO, BI, SIZE), %xmm5
VFMADDPS_I( %xmm9,%xmm5,%xmm0 )
VFMADDPS_I( %xmm13,%xmm5,%xmm1 )
addq $ 2, BI
addq $ 8, %rax
.endm
.macro SAVE4x1
vbroadcastss ALPHA_R, %xmm0
vbroadcastss ALPHA_I, %xmm1
// swap high and low 4 bytes
vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
vshufps $ 0xb1, %xmm13, %xmm13, %xmm13
#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
defined(NR) || defined(NC) || defined(TR) || defined(TC)
vaddsubps %xmm9, %xmm8 , %xmm8
vaddsubps %xmm13,%xmm12, %xmm12
vshufps $ 0xb1, %xmm8 , %xmm8, %xmm9
vshufps $ 0xb1, %xmm12, %xmm12, %xmm13
#else
vaddsubps %xmm8, %xmm9 ,%xmm9
vaddsubps %xmm12, %xmm13,%xmm13
vmovaps %xmm9, %xmm8
vmovaps %xmm13, %xmm12
// swap high and low 4 bytes
vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
vshufps $ 0xb1, %xmm13, %xmm13, %xmm13
#endif
// multiply with ALPHA_R
vmulps %xmm8 , %xmm0, %xmm8
vmulps %xmm12, %xmm0, %xmm12
// multiply with ALPHA_I
vmulps %xmm9 , %xmm1, %xmm9
vmulps %xmm13, %xmm1, %xmm13
vaddsubps %xmm9, %xmm8 , %xmm8
vaddsubps %xmm13,%xmm12, %xmm12
#ifndef TRMMKERNEL
vaddps (CO1), %xmm8 , %xmm8
vaddps 4 * SIZE(CO1), %xmm12, %xmm12
#endif
vmovups %xmm8 , (CO1)
vmovups %xmm12 , 4 * SIZE(CO1)
.endm
/************************************************************************************************/
.macro KERNEL2x1_SUB
vmovups -16 * SIZE(AO, %rax, SIZE), %xmm0
vbroadcastss -4 * SIZE(BO, BI, SIZE), %xmm4
VFMADDPS_R( %xmm8,%xmm4,%xmm0 )
vbroadcastss -3 * SIZE(BO, BI, SIZE), %xmm5
VFMADDPS_I( %xmm9,%xmm5,%xmm0 )
addq $ 2, BI
addq $ 4, %rax
.endm
.macro SAVE2x1
vbroadcastss ALPHA_R, %xmm0
vbroadcastss ALPHA_I, %xmm1
// swap high and low 64 bytes
vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
defined(NR) || defined(NC) || defined(TR) || defined(TC)
vaddsubps %xmm9, %xmm8 , %xmm8
vshufps $ 0xb1, %xmm8 , %xmm8, %xmm9
#else
vaddsubps %xmm8, %xmm9 ,%xmm9
vmovaps %xmm9, %xmm8
// swap high and low 64 bytes
vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
#endif
// multiply with ALPHA_R
vmulps %xmm8 , %xmm0, %xmm8
// multiply with ALPHA_I
vmulps %xmm9 , %xmm1, %xmm9
vaddsubps %xmm9, %xmm8 , %xmm8
#ifndef TRMMKERNEL
vaddps (CO1), %xmm8 , %xmm8
#endif
vmovups %xmm8 , (CO1)
.endm
/************************************************************************************************/
.macro KERNEL1x1_SUB
vmovsd -16 * SIZE(AO, %rax, SIZE), %xmm0
vbroadcastss -4 * SIZE(BO, BI, SIZE), %xmm4
VFMADDPS_R( %xmm8,%xmm4,%xmm0 )
vbroadcastss -3 * SIZE(BO, BI, SIZE), %xmm5
VFMADDPS_I( %xmm9,%xmm5,%xmm0 )
addq $ 2, BI
addq $ 2, %rax
.endm
.macro SAVE1x1
vbroadcastss ALPHA_R, %xmm0
vbroadcastss ALPHA_I, %xmm1
// swap high and low 64 bytes
vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
defined(NR) || defined(NC) || defined(TR) || defined(TC)
vaddsubps %xmm9, %xmm8 , %xmm8
vshufps $ 0xb1, %xmm8 , %xmm8, %xmm9
#else
vaddsubps %xmm8, %xmm9 ,%xmm9
vmovaps %xmm9, %xmm8
// swap high and low 64 bytes
vshufps $ 0xb1, %xmm9 , %xmm9, %xmm9
#endif
// multiply with ALPHA_R
vmulps %xmm8 , %xmm0, %xmm8
// multiply with ALPHA_I
vmulps %xmm9 , %xmm1, %xmm9
vaddsubps %xmm9, %xmm8 , %xmm8
#ifndef TRMMKERNEL
vmovsd (CO1), %xmm14
vaddps %xmm14, %xmm8 , %xmm8
#endif
vmovsd %xmm8 , (CO1)
.endm
/************************************************************************************************/
PROLOGUE
PROFCODE
subq $ STACKSIZE, %rsp
movq %rbx, (%rsp)
movq %rbp, 8(%rsp)
movq %r12, 16(%rsp)
movq %r13, 24(%rsp)
movq %r14, 32(%rsp)
movq %r15, 40(%rsp)
vzeroupper
#ifdef WINDOWS_ABI
movq %rdi, 48(%rsp)
movq %rsi, 56(%rsp)
vmovups %xmm6, 64(%rsp)
vmovups %xmm7, 80(%rsp)
vmovups %xmm8, 96(%rsp)
vmovups %xmm9, 112(%rsp)
vmovups %xmm10, 128(%rsp)
vmovups %xmm11, 144(%rsp)
vmovups %xmm12, 160(%rsp)
vmovups %xmm13, 176(%rsp)
vmovups %xmm14, 192(%rsp)
vmovups %xmm15, 208(%rsp)
movq ARG1, OLD_M
movq ARG2, OLD_N
movq ARG3, OLD_K
movq OLD_A, A
movq OLD_B, B
movq OLD_C, C
movq OLD_LDC, LDC
#ifdef TRMMKERNEL
movsd OLD_OFFSET, %xmm12
#endif
vmovaps %xmm3, %xmm0
vmovsd OLD_ALPHA_I, %xmm1
#else
movq STACKSIZE + 8(%rsp), LDC
#ifdef TRMMKERNEL
movsd STACKSIZE + 16(%rsp), %xmm12
#endif
#endif
movq %rsp, SP # save old stack
subq $ 128 + L_BUFFER_SIZE, %rsp
andq $ -4096, %rsp # align stack
STACK_TOUCH
cmpq $ 0, OLD_M
je .L999
cmpq $ 0, OLD_N
je .L999
cmpq $ 0, OLD_K
je .L999
movq OLD_M, M
movq OLD_N, N
movq OLD_K, K
vmovss %xmm0, ALPHA_R
vmovss %xmm1, ALPHA_I
salq $ ZBASE_SHIFT, LDC
movq N, %rax
xorq %rdx, %rdx
movq $ 2, %rdi
divq %rdi // N / 2
movq %rax, Ndiv6 // N / 2
movq %rdx, Nmod6 // N % 2
#ifdef TRMMKERNEL
vmovsd %xmm12, OFFSET
vmovsd %xmm12, KK
#ifndef LEFT
negq KK
#endif
#endif
.L2_0: