/
generator_packed_gemmnn.h
643 lines (614 loc) · 38.9 KB
/
generator_packed_gemmnn.h
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
/******************************************************************************
** Copyright (c) 2017-2019, Intel Corporation **
** 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 copyright holder 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 COPYRIGHT **
** HOLDER 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. **
******************************************************************************/
/* Alexander Heinecke, Greg Henry (Intel Corp.)
******************************************************************************/
#ifndef GENERATOR_PACKED_GEMMNN_H
#define GENERATOR_PACKED_GEMMNN_H
#include "generator_x86_instructions.h"
#include "generator_common.h"
#include <libxsmm_intrinsics_x86.h>
#if defined(LIBXSMM_OFFLOAD_TARGET)
# pragma offload_attribute(push,target(LIBXSMM_OFFLOAD_TARGET))
#endif
#include <string.h>
#include <stdlib.h>
#include <assert.h>
#include <stdio.h>
#if defined(LIBXSMM_OFFLOAD_TARGET)
# pragma offload_attribute(pop)
#endif
/*
Does C(cm1:cm2,cn1:cn2) <- alpha*A(am1:am2,ak1:ak2)*B(bk1:bk2,bn1:bn2) +
beta *C(cm1:cm2,cn1:cn2)
Obviously, the dimensions must conform
Alpha and Beta are doubles even if the work should be single. Just convert
*/
LIBXSMM_API_INLINE void compact_gemmnn_ (
unsigned int tra, /* 0 if non-transpose */
unsigned int trb, /* 0 if non-transpose */
unsigned int trc, /* 0 if non-transpose */
unsigned int am1,
unsigned int am2,
unsigned int ak1,
unsigned int ak2,
unsigned int bk1,
unsigned int bk2,
unsigned int bn1,
unsigned int bn2,
unsigned int cm1,
unsigned int cm2,
unsigned int cn1,
unsigned int cn2,
double alpha,
unsigned int areg,
unsigned int lda,
unsigned int breg,
unsigned int ldb,
double beta,
unsigned int creg,
unsigned int ldc,
libxsmm_generated_code* io_code,
unsigned int numb,
char regset,
unsigned int iunroll,
unsigned int junroll,
unsigned int loopi,
unsigned int loopj )
{
unsigned int i, j, l, datasz, nloopcnt=0, mloopcnt=0, nborder, mborder, mloopadj = 0;
int aoffset, boffset, coffset; /* Address calcs for the loops */
unsigned int iun=3, jun=3; /* Register blocking sizes */
int a0 = -1, a1 = -1, a2 = -1, a3 = -1, a4 = -1, a5 = -1, a6 = -1, a7 = -1;
int b0 = -1, b1 = -1, b2 = -1, b3 = -1, b4 = -1, b5 = -1, b6 = -1, b7 = -1;
int c00 = -1, c01 = -1, c02 = -1, c03 = -1, c04 = -1, c05 = -1, c06 = -1, c07 = -1;
int c10 = -1, c11 = -1, c12 = -1, c13 = -1, c14 = -1, c15 = -1, c16 = -1, c17 = -1;
int c20 = -1, c21 = -1, c22 = -1, c23 = -1, c24 = -1, c25 = -1, c26 = -1, c27 = -1;
int c30 = -1, c31 = -1, c32 = -1, c33 = -1, c34 = -1, c35 = -1, c36 = -1, c37 = -1;
int c40 = -1, c41 = -1, c42 = -1, c43 = -1, c44 = -1, c45 = -1, c46 = -1, c47 = -1;
int c50 = -1, c51 = -1, c52 = -1, c53 = -1, c54 = -1, c55 = -1, c56 = -1, c57 = -1;
int c60 = -1, c61 = -1, c62 = -1, c63 = -1, c64 = -1, c65 = -1, c66 = -1, c67 = -1;
int c70 = -1, c71 = -1, c72 = -1, c73 = -1, c74 = -1, c75 = -1, c76 = -1, c77 = -1;
int c0 = 0, c1 = 0, c2 = 0, c3 = 0, c4 = 0, c5 = 0, c6 = 0, c7 = 0;
int j0 = 1, j1, j2, j3, j4, j5, j6, j7;
int i0 = 1, i1, i2, i3, i4, i5, i6, i7;
unsigned int maxregblocking = 8, maxreg = 16;
libxsmm_loop_label_tracker l_loop_label_tracker;
/* Test that the dimensions conform */
if ( (am2-am1) != (cm2-cm1) ) {
printf("compact_gemmnn m-dimensions don't conform: %u != %u\n",am2-am1+1,cm2-cm1+1);
exit(-1);
}
if ( (ak2-ak1) != (bk2-bk1) ) {
printf("compact_gemmnn k-dimensions don't conform: %u != %u\n",ak2-ak1+1,bk2-bk1+1);
exit(-1);
}
if ( (bn2-bn1) != (cn2-cn1) ) {
printf("compact_gemmnn n-dimensions don't conform: %u != %u\n",ak2-ak1+1,bk2-bk1+1);
exit(-1);
}
/* See that all dimensions are at least 1 */
if ( am2 < am1) {
printf("compact_gemmnn m-dimension too small: %u\n",am2-am1+1);
exit(-1);
}
if ( ak2 < ak1) {
printf("compact_gemmnn k-dimension too small: %u\n",ak2-ak1+1);
exit(-1);
}
if ( bn2 < bn1) {
printf("compact_gemmnn n-dimension too small: %u\n",bn2-bn1+1);
exit(-1);
}
/* Check that areg, breg, creg is valid */
if ( /*(areg < 0) ||*/ (areg > 15) ) {
printf("compact_gemmnn A gp register invalid: %u\n",areg);
exit(-1);
}
if ( /*(breg < 0) ||*/ (breg > 15) ) {
printf("compact_gemmnn B gp register invalid: %u\n",breg);
exit(-1);
}
if ( /*(creg < 0) ||*/ (creg > 15) ) {
printf("compact_gemmnn C gp register invalid: %u\n",creg);
exit(-1);
}
if ( (numb == 8) && (regset=='z') ) { datasz = 8; }
else if ( (numb == 16) && (regset=='z') ) { datasz = 4; }
else if ( (numb == 8) && (regset=='y') ) { datasz = 4; }
else if ( (numb == 4) && (regset=='y') ) { datasz = 8; }
else {
printf("compact_gemmnn Unknown number=%u or regset=%c\n",numb,regset);
exit(-1);
}
if ( regset == 'y' ) { iun = 3; jun = 3; maxreg=16; maxregblocking= 7; }
if ( regset == 'z' ) { iun = 5; jun = 4; maxreg=32; maxregblocking= 8; }
if ( iunroll > 0 ) iun = iunroll;
if ( junroll > 0 ) jun = junroll;
/* Make sure values of register blocking are between 1 and maxregblocking */
iun = LIBXSMM_MAX(LIBXSMM_MIN(iun,maxregblocking),1);
jun = LIBXSMM_MAX(LIBXSMM_MIN(jun,maxregblocking),1);
/* CHeck to see the register blocking parameters make sense: */
if ( maxreg < 3 ) {
printf("Sorry, not enough registers available in compact gemm nn\n");
exit(-1);
}
while ( iun+jun+iun*jun > maxreg ) {
if ( (iun >= jun) && (iun > 1) ) --iun;
else if ( (jun >= iun) && (jun > 1) ) --jun;
else {
printf("Seems strange that we can't reduce the registers in compact gemm nn\n");
exit(-1);
}
}
/* Determine if the problem is too small for loops giving this register blocking */
mloopcnt = (int)((am2-am1+1)/iun);
nloopcnt = (int)((bn2-bn1+1)/jun);
if ( mloopcnt < 2 ) loopi = 0;
if ( nloopcnt < 2 ) loopj = 0;
mborder = (am2-am1+1)-mloopcnt*iun;
nborder = (bn2-bn1+1)-nloopcnt*jun;
if ( loopj || loopi ) {
libxsmm_reset_loop_label_tracker ( &l_loop_label_tracker );
}
/* DO register blocking */
a0 = 0;
if ( iun > 1 ) a1 = 1;
if ( iun > 2 ) a2 = 2;
if ( iun > 3 ) a3 = 3;
if ( iun > 4 ) a4 = 4;
if ( iun > 5 ) a5 = 5;
if ( iun > 6 ) a6 = 6;
if ( iun > 7 ) a7 = 7;
b0 = iun;
if ( jun > 1 ) b1 = b0+1;
if ( jun > 2 ) b2 = b0+2;
if ( jun > 3 ) b3 = b0+3;
if ( jun > 4 ) b4 = b0+4;
if ( jun > 5 ) b5 = b0+5;
if ( jun > 6 ) b6 = b0+6;
if ( jun > 7 ) b7 = b0+7;
c00 = iun + jun;
if ( jun > 1 ) c01 = c00 + 1;
if ( jun > 2 ) c02 = c00 + 2;
if ( jun > 3 ) c03 = c00 + 3;
if ( jun > 4 ) c04 = c00 + 4;
if ( jun > 5 ) c05 = c00 + 5;
if ( jun > 6 ) c06 = c00 + 6;
if ( jun > 7 ) c07 = c00 + 7;
if ( iun > 1 ) c10 = c00 + jun;
if ( (iun > 1) && (jun > 1) ) c11 = c10 + 1;
if ( (iun > 1) && (jun > 2) ) c12 = c10 + 2;
if ( (iun > 1) && (jun > 3) ) c13 = c10 + 3;
if ( (iun > 1) && (jun > 4) ) c14 = c10 + 4;
if ( (iun > 1) && (jun > 5) ) c15 = c10 + 5;
if ( (iun > 1) && (jun > 6) ) c16 = c10 + 6;
if ( (iun > 1) && (jun > 7) ) c17 = c10 + 7;
if ( iun > 2 ) c20 = c10 + jun;
if ( (iun > 2) && (jun > 1) ) c21 = c20 + 1;
if ( (iun > 2) && (jun > 2) ) c22 = c20 + 2;
if ( (iun > 2) && (jun > 3) ) c23 = c20 + 3;
if ( (iun > 2) && (jun > 4) ) c24 = c20 + 4;
if ( (iun > 2) && (jun > 5) ) c25 = c20 + 5;
if ( (iun > 2) && (jun > 6) ) c26 = c20 + 6;
if ( (iun > 2) && (jun > 7) ) c27 = c20 + 7;
if ( iun > 3 ) c30 = c20 + jun;
if ( (iun > 3) && (jun > 1) ) c31 = c30 + 1;
if ( (iun > 3) && (jun > 2) ) c32 = c30 + 2;
if ( (iun > 3) && (jun > 3) ) c33 = c30 + 3;
if ( (iun > 3) && (jun > 4) ) c34 = c30 + 4;
if ( (iun > 3) && (jun > 5) ) c35 = c30 + 5;
if ( (iun > 3) && (jun > 6) ) c36 = c30 + 6;
if ( (iun > 3) && (jun > 7) ) c37 = c30 + 7;
if ( iun > 4 ) c40 = c30 + jun;
if ( (iun > 4) && (jun > 1) ) c41 = c40 + 1;
if ( (iun > 4) && (jun > 2) ) c42 = c40 + 2;
if ( (iun > 4) && (jun > 3) ) c43 = c40 + 3;
if ( (iun > 4) && (jun > 4) ) c44 = c40 + 4;
if ( (iun > 4) && (jun > 5) ) c45 = c40 + 5;
if ( (iun > 4) && (jun > 6) ) c46 = c40 + 6;
if ( (iun > 4) && (jun > 7) ) c47 = c40 + 7;
if ( iun > 5 ) c50 = c40 + jun;
if ( (iun > 5) && (jun > 1) ) c51 = c50 + 1;
if ( (iun > 5) && (jun > 2) ) c52 = c50 + 2;
if ( (iun > 5) && (jun > 3) ) c53 = c50 + 3;
if ( (iun > 5) && (jun > 4) ) c54 = c50 + 4;
if ( (iun > 5) && (jun > 5) ) c55 = c50 + 5;
if ( (iun > 5) && (jun > 6) ) c56 = c50 + 6;
if ( (iun > 5) && (jun > 7) ) c57 = c50 + 7;
if ( iun > 6 ) c60 = c50 + jun;
if ( (iun > 6) && (jun > 1) ) c61 = c60 + 1;
if ( (iun > 6) && (jun > 2) ) c62 = c60 + 2;
if ( (iun > 6) && (jun > 3) ) c63 = c60 + 3;
if ( (iun > 6) && (jun > 4) ) c64 = c60 + 4;
if ( (iun > 6) && (jun > 5) ) c65 = c60 + 5;
if ( (iun > 6) && (jun > 6) ) c66 = c60 + 6;
if ( (iun > 6) && (jun > 7) ) c67 = c60 + 7;
if ( iun > 7 ) c70 = c60 + jun;
if ( (iun > 7) && (jun > 1) ) c71 = c70 + 1;
if ( (iun > 7) && (jun > 2) ) c72 = c70 + 2;
if ( (iun > 7) && (jun > 3) ) c73 = c70 + 3;
if ( (iun > 7) && (jun > 4) ) c74 = c70 + 4;
if ( (iun > 7) && (jun > 5) ) c75 = c70 + 5;
if ( (iun > 7) && (jun > 6) ) c76 = c70 + 6;
if ( (iun > 7) && (jun > 7) ) c77 = c70 + 7;
#if 0
#define COMPACT_GEMMNN_DEBUG
#endif
#ifdef COMPACT_GEMMNN_DEBUG
printf("iun=%d jun=%d loopi=%d loopj=%d\n",iun,jun,loopi,loopj);
printf("areg=%d breg=%d creg=%d mborder=%d nborder=%d\n",areg,breg,creg,mborder,nborder);
printf("a0:7=%d %d %d %d %d %d %d %d\n",a0,a1,a2,a3,a4,a5,a6,a7);
printf("b0:7=%d %d %d %d %d %d %d %d\n",b0,b1,b2,b3,b4,b5,b6,b7);
printf("c0,0:7=%d %d %d %d %d %d %d %d\n",c00,c01,c02,c03,c04,c05,c06,c07);
if (c10>0) printf("c1,0:7=%d %d %d %d %d %d %d %d\n",c10,c11,c12,c13,c14,c15,c16,c17);
if (c20>0) printf("c2,0:7=%d %d %d %d %d %d %d %d\n",c20,c21,c22,c23,c24,c25,c26,c27);
if (c30>0) printf("c3,0:7=%d %d %d %d %d %d %d %d\n",c30,c31,c32,c33,c34,c35,c36,c37);
if (c40>0) printf("c4,0:7=%d %d %d %d %d %d %d %d\n",c40,c41,c42,c43,c44,c45,c46,c47);
if (c50>0) printf("c5,0:7=%d %d %d %d %d %d %d %d\n",c50,c51,c52,c53,c54,c55,c56,c57);
if (c60>0) printf("c6,0:7=%d %d %d %d %d %d %d %d\n",c60,c61,c62,c63,c64,c65,c66,c67);
if (c70>0) printf("c7,0:7=%d %d %d %d %d %d %d %d\n",c70,c71,c72,c73,c74,c75,c76,c77);
#endif
if ( loopj && (nloopcnt >=2) ) {
#ifdef COMPACT_GEMMNN_DEBUG
printf("Setting up n-loop: loopj=%d nloopcnt=%d\n",loopj,nloopcnt);
#endif
libxsmm_x86_instruction_alu_imm( io_code, LIBXSMM_X86_INSTR_MOVQ, LIBXSMM_X86_GP_REG_RAX, nloopcnt );
libxsmm_x86_instruction_register_jump_back_label( io_code, &l_loop_label_tracker );
}
for ( j = bn1 ; j <= bn2 ; j+=jun ) {
#ifdef COMPACT_GEMMNN_DEBUG
printf("Doing j loop from %d to %d with blocksize %d\n",bn1,bn2,jun);
#endif
if ( ( j <= bn2 ) && ( jun >= 1 ) ) j0 = 1; else j0 = 0;
if ( ( j+1 <= bn2 ) && ( jun >= 2 ) ) j1 = 1; else j1 = 0;
if ( ( j+2 <= bn2 ) && ( jun >= 3 ) ) j2 = 1; else j2 = 0;
if ( ( j+3 <= bn2 ) && ( jun >= 4 ) ) j3 = 1; else j3 = 0;
if ( ( j+4 <= bn2 ) && ( jun >= 5 ) ) j4 = 1; else j4 = 0;
if ( ( j+5 <= bn2 ) && ( jun >= 6 ) ) j5 = 1; else j5 = 0;
if ( ( j+6 <= bn2 ) && ( jun >= 7 ) ) j6 = 1; else j6 = 0;
if ( ( j+7 <= bn2 ) && ( jun >= 8 ) ) j7 = 1; else j7 = 0;
if ( loopj && (j > bn1) && (j + jun -1 <= bn2) ) {
/* Turn everything off, we're really supposed to be in a loop */
j0=0; j1=0; j2=0; j3=0; j4=0; j5=0; j6=0; j7=0;
#ifdef COMPACT_GEMMNN_DEBUG
printf("Emptying n-loop for j=%d\n",j);
#endif
}
if ( loopi && (mloopcnt >=2) && j0 ) {
#ifdef COMPACT_GEMMNN_DEBUG
printf("Setting up m-loop: loopi=%d mloopcnt=%d\n",loopi,mloopcnt);
#endif
libxsmm_x86_instruction_alu_imm( io_code, LIBXSMM_X86_INSTR_MOVQ, LIBXSMM_X86_GP_REG_RCX, mloopcnt );
libxsmm_x86_instruction_register_jump_back_label( io_code, &l_loop_label_tracker );
mloopadj = 1;
}
for ( i = am1; i <= am2 ; i+=iun ) {
if ( ( i <= am2 ) && ( iun >= 1 ) ) i0 = 1; else i0 = 0;
if ( ( i+1 <= am2 ) && ( iun >= 2 ) ) i1 = 1; else i1 = 0;
if ( ( i+2 <= am2 ) && ( iun >= 3 ) ) i2 = 1; else i2 = 0;
if ( ( i+3 <= am2 ) && ( iun >= 4 ) ) i3 = 1; else i3 = 0;
if ( ( i+4 <= am2 ) && ( iun >= 5 ) ) i4 = 1; else i4 = 0;
if ( ( i+5 <= am2 ) && ( iun >= 6 ) ) i5 = 1; else i5 = 0;
if ( ( i+6 <= am2 ) && ( iun >= 7 ) ) i6 = 1; else i6 = 0;
if ( ( i+7 <= am2 ) && ( iun >= 8 ) ) i7 = 1; else i7 = 0;
#ifdef COMPACT_GEMMNN_DEBUG
printf("Doing i loop from %d to %d with blocksize %d (%d,%d,%d,%d,%d,%d,%d,%d)\n",am1,am2,iun,i0,i1,i2,i3,i4,i5,i6,i7);
#endif
if ( loopi && (i > am1) && (i + iun -1 <= am2) ) {
/* Turn everything off, we're really supposed to be in a loop */
i0=0; i1=0; i2=0; i3=0; i4=0; i5=0; i6=0; i7=0;
#ifdef COMPACT_GEMMNN_DEBUG
printf("Emptying m-loop for i=%d j=%d i0=%d j0=%d\n",i,j,i0,j0);
#endif
}
#ifdef COMPACT_GEMMNN_DEBUG
if (i0 && j0 ) printf("Loading A into %d with tra=%d lda=%d i=%d j=%d numb=%d datasz=%d regset=%c areg=%d\n",a0,tra,lda,i,ak1,numb,datasz, regset,areg);
#endif
if (i0 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i, ak1, a0, numb, datasz, regset, areg );
#ifdef COMPACT_GEMMNN_DEBUG
if (i0 && j0 ) printf("Loaded A into %d with tra=%d lda=%d i=%d j=%d numb=%d datasz=%d regset=%c areg=%d\n",a0,tra,lda,i,ak1,numb,datasz, regset,areg);
#endif
if (i0 && j0) compact_load_matrix_gen_ ( io_code, trb, ldb, bk1, j, b0, numb, datasz, regset, breg );
if (i0 && j0) compact_mult_two_nums_ ( io_code, a0, b0, c00, numb, regset );
if (i1 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i+1, ak1, a1, numb, datasz, regset, areg );
if (i1 && j0) compact_mult_two_nums_ ( io_code, a1, b0, c10, numb, regset );
if (i2 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i+2, ak1, a2, numb, datasz, regset, areg );
if (i2 && j0) compact_mult_two_nums_ ( io_code, a2, b0, c20, numb, regset );
if (i3 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i+3, ak1, a3, numb, datasz, regset, areg );
if (i3 && j0) compact_mult_two_nums_ ( io_code, a3, b0, c30, numb, regset );
if (i4 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i+4, ak1, a4, numb, datasz, regset, areg );
if (i4 && j0) compact_mult_two_nums_ ( io_code, a4, b0, c40, numb, regset );
if (i5 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i+5, ak1, a5, numb, datasz, regset, areg );
if (i5 && j0) compact_mult_two_nums_ ( io_code, a5, b0, c50, numb, regset );
if (i6 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i+6, ak1, a6, numb, datasz, regset, areg );
if (i6 && j0) compact_mult_two_nums_ ( io_code, a6, b0, c60, numb, regset );
if (i7 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i+7, ak1, a7, numb, datasz, regset, areg );
if (i7 && j0) compact_mult_two_nums_ ( io_code, a7, b0, c70, numb, regset );
if (i0 && j1) compact_load_matrix_gen_ ( io_code, trb, ldb, bk1, j+1, b1, numb, datasz, regset, breg );
if (i0 && j1) compact_mult_two_nums_ ( io_code, a0, b1, c01, numb, regset );
if (i0 && j2) compact_load_matrix_gen_ ( io_code, trb, ldb, bk1, j+2, b2, numb, datasz, regset, breg );
if (i0 && j2) compact_mult_two_nums_ ( io_code, a0, b2, c02, numb, regset );
if (i0 && j3) compact_load_matrix_gen_ ( io_code, trb, ldb, bk1, j+3, b3, numb, datasz, regset, breg );
if (i0 && j3) compact_mult_two_nums_ ( io_code, a0, b3, c03, numb, regset );
if (i0 && j4) compact_load_matrix_gen_ ( io_code, trb, ldb, bk1, j+4, b4, numb, datasz, regset, breg );
if (i0 && j4) compact_mult_two_nums_ ( io_code, a0, b4, c04, numb, regset );
if (i0 && j5) compact_load_matrix_gen_ ( io_code, trb, ldb, bk1, j+5, b5, numb, datasz, regset, breg );
if (i0 && j5) compact_mult_two_nums_ ( io_code, a0, b5, c05, numb, regset );
if (i0 && j6) compact_load_matrix_gen_ ( io_code, trb, ldb, bk1, j+6, b6, numb, datasz, regset, breg );
if (i0 && j6) compact_mult_two_nums_ ( io_code, a0, b6, c06, numb, regset );
if (i0 && j7) compact_load_matrix_gen_ ( io_code, trb, ldb, bk1, j+7, b7, numb, datasz, regset, breg );
if (i0 && j7) compact_mult_two_nums_ ( io_code, a0, b7, c07, numb, regset );
if (i1 && j1) compact_mult_two_nums_ ( io_code, a1, b1, c11, numb, regset );
if (i1 && j2) compact_mult_two_nums_ ( io_code, a1, b2, c12, numb, regset );
if (i1 && j3) compact_mult_two_nums_ ( io_code, a1, b3, c13, numb, regset );
if (i1 && j4) compact_mult_two_nums_ ( io_code, a1, b4, c14, numb, regset );
if (i1 && j5) compact_mult_two_nums_ ( io_code, a1, b5, c15, numb, regset );
if (i1 && j6) compact_mult_two_nums_ ( io_code, a1, b6, c16, numb, regset );
if (i1 && j7) compact_mult_two_nums_ ( io_code, a1, b7, c17, numb, regset );
if (i2 && j1) compact_mult_two_nums_ ( io_code, a2, b1, c21, numb, regset );
if (i2 && j2) compact_mult_two_nums_ ( io_code, a2, b2, c22, numb, regset );
if (i2 && j3) compact_mult_two_nums_ ( io_code, a2, b3, c23, numb, regset );
if (i2 && j4) compact_mult_two_nums_ ( io_code, a2, b4, c24, numb, regset );
if (i2 && j5) compact_mult_two_nums_ ( io_code, a2, b5, c25, numb, regset );
if (i2 && j6) compact_mult_two_nums_ ( io_code, a2, b6, c26, numb, regset );
if (i2 && j7) compact_mult_two_nums_ ( io_code, a2, b7, c27, numb, regset );
if (i3 && j1) compact_mult_two_nums_ ( io_code, a3, b1, c31, numb, regset );
if (i3 && j2) compact_mult_two_nums_ ( io_code, a3, b2, c32, numb, regset );
if (i3 && j3) compact_mult_two_nums_ ( io_code, a3, b3, c33, numb, regset );
if (i3 && j4) compact_mult_two_nums_ ( io_code, a3, b4, c34, numb, regset );
if (i3 && j5) compact_mult_two_nums_ ( io_code, a3, b5, c35, numb, regset );
if (i3 && j6) compact_mult_two_nums_ ( io_code, a3, b6, c36, numb, regset );
if (i3 && j7) compact_mult_two_nums_ ( io_code, a3, b7, c37, numb, regset );
if (i4 && j1) compact_mult_two_nums_ ( io_code, a4, b1, c41, numb, regset );
if (i4 && j2) compact_mult_two_nums_ ( io_code, a4, b2, c42, numb, regset );
if (i4 && j3) compact_mult_two_nums_ ( io_code, a4, b3, c43, numb, regset );
if (i4 && j4) compact_mult_two_nums_ ( io_code, a4, b4, c44, numb, regset );
if (i4 && j5) compact_mult_two_nums_ ( io_code, a4, b5, c45, numb, regset );
if (i4 && j6) compact_mult_two_nums_ ( io_code, a4, b6, c46, numb, regset );
if (i4 && j7) compact_mult_two_nums_ ( io_code, a4, b7, c47, numb, regset );
if (i5 && j1) compact_mult_two_nums_ ( io_code, a5, b1, c51, numb, regset );
if (i5 && j2) compact_mult_two_nums_ ( io_code, a5, b2, c52, numb, regset );
if (i5 && j3) compact_mult_two_nums_ ( io_code, a5, b3, c53, numb, regset );
if (i5 && j4) compact_mult_two_nums_ ( io_code, a5, b4, c54, numb, regset );
if (i5 && j5) compact_mult_two_nums_ ( io_code, a5, b5, c55, numb, regset );
if (i5 && j6) compact_mult_two_nums_ ( io_code, a5, b6, c56, numb, regset );
if (i5 && j7) compact_mult_two_nums_ ( io_code, a5, b7, c57, numb, regset );
if (i6 && j1) compact_mult_two_nums_ ( io_code, a6, b1, c61, numb, regset );
if (i6 && j2) compact_mult_two_nums_ ( io_code, a6, b2, c62, numb, regset );
if (i6 && j3) compact_mult_two_nums_ ( io_code, a6, b3, c63, numb, regset );
if (i6 && j4) compact_mult_two_nums_ ( io_code, a6, b4, c64, numb, regset );
if (i6 && j5) compact_mult_two_nums_ ( io_code, a6, b5, c65, numb, regset );
if (i6 && j6) compact_mult_two_nums_ ( io_code, a6, b6, c66, numb, regset );
if (i6 && j7) compact_mult_two_nums_ ( io_code, a6, b7, c67, numb, regset );
if (i7 && j1) compact_mult_two_nums_ ( io_code, a7, b1, c71, numb, regset );
if (i7 && j2) compact_mult_two_nums_ ( io_code, a7, b2, c72, numb, regset );
if (i7 && j3) compact_mult_two_nums_ ( io_code, a7, b3, c73, numb, regset );
if (i7 && j4) compact_mult_two_nums_ ( io_code, a7, b4, c74, numb, regset );
if (i7 && j5) compact_mult_two_nums_ ( io_code, a7, b5, c75, numb, regset );
if (i7 && j6) compact_mult_two_nums_ ( io_code, a7, b6, c76, numb, regset );
if (i7 && j7) compact_mult_two_nums_ ( io_code, a7, b7, c77, numb, regset );
for ( l = ak1+1 ; l <= ak2; l++ ) {
#ifdef COMPACT_GEMMNN_DEBUG
printf("Doing l loop from %d to %d\n",ak1+1,ak2);
#endif
if (i0 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i, l, a0, numb, datasz, regset, areg );
if (i0 && j0) compact_load_matrix_gen_ ( io_code, trb, ldb, l-ak1+bk1, j, b0, numb, datasz, regset, breg);
if (i0 && j0) compact_fma_cplusab_ ( io_code, c00, a0, b0, numb, regset );
if (i1 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i+1, l, a1, numb, datasz, regset, areg );
if (i1 && j0) compact_fma_cplusab_ ( io_code, c10, a1, b0, numb, regset );
if (i2 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i+2, l, a2, numb, datasz, regset, areg );
if (i2 && j0) compact_fma_cplusab_ ( io_code, c20, a2, b0, numb, regset );
if (i3 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i+3, l, a3, numb, datasz, regset, areg );
if (i3 && j0) compact_fma_cplusab_ ( io_code, c30, a3, b0, numb, regset );
if (i4 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i+4, l, a4, numb, datasz, regset, areg );
if (i4 && j0) compact_fma_cplusab_ ( io_code, c40, a4, b0, numb, regset );
if (i5 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i+5, l, a5, numb, datasz, regset, areg );
if (i5 && j0) compact_fma_cplusab_ ( io_code, c50, a5, b0, numb, regset );
if (i6 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i+6, l, a6, numb, datasz, regset, areg );
if (i6 && j0) compact_fma_cplusab_ ( io_code, c60, a6, b0, numb, regset );
if (i7 && j0) compact_load_matrix_gen_ ( io_code, tra, lda, i+7, l, a7, numb, datasz, regset, areg );
if (i7 && j0) compact_fma_cplusab_ ( io_code, c70, a7, b0, numb, regset );
if (i0 && j1) compact_load_matrix_gen_ ( io_code, trb, ldb, l-ak1+bk1, j+1, b1, numb, datasz, regset, breg);
if (i0 && j1) compact_fma_cplusab_ ( io_code, c01, a0, b1, numb, regset );
if (i0 && j2) compact_load_matrix_gen_ ( io_code, trb, ldb, l-ak1+bk1, j+2, b2, numb, datasz, regset, breg);
if (i0 && j2) compact_fma_cplusab_ ( io_code, c02, a0, b2, numb, regset );
if (i0 && j3) compact_load_matrix_gen_ ( io_code, trb, ldb, l-ak1+bk1, j+3, b3, numb, datasz, regset, breg);
if (i0 && j3) compact_fma_cplusab_ ( io_code, c03, a0, b3, numb, regset );
if (i0 && j4) compact_load_matrix_gen_ ( io_code, trb, ldb, l-ak1+bk1, j+4, b4, numb, datasz, regset, breg);
if (i0 && j4) compact_fma_cplusab_ ( io_code, c04, a0, b4, numb, regset );
if (i0 && j5) compact_load_matrix_gen_ ( io_code, trb, ldb, l-ak1+bk1, j+5, b5, numb, datasz, regset, breg);
if (i0 && j5) compact_fma_cplusab_ ( io_code, c05, a0, b5, numb, regset );
if (i0 && j6) compact_load_matrix_gen_ ( io_code, trb, ldb, l-ak1+bk1, j+6, b6, numb, datasz, regset, breg);
if (i0 && j6) compact_fma_cplusab_ ( io_code, c06, a0, b6, numb, regset );
if (i0 && j7) compact_load_matrix_gen_ ( io_code, trb, ldb, l-ak1+bk1, j+7, b7, numb, datasz, regset, breg);
if (i0 && j7) compact_fma_cplusab_ ( io_code, c07, a0, b7, numb, regset );
if (i1 && j1) compact_fma_cplusab_ ( io_code, c11, a1, b1, numb, regset );
if (i1 && j2) compact_fma_cplusab_ ( io_code, c12, a1, b2, numb, regset );
if (i1 && j3) compact_fma_cplusab_ ( io_code, c13, a1, b3, numb, regset );
if (i1 && j4) compact_fma_cplusab_ ( io_code, c14, a1, b4, numb, regset );
if (i1 && j5) compact_fma_cplusab_ ( io_code, c15, a1, b5, numb, regset );
if (i1 && j6) compact_fma_cplusab_ ( io_code, c16, a1, b6, numb, regset );
if (i1 && j7) compact_fma_cplusab_ ( io_code, c17, a1, b7, numb, regset );
if (i2 && j1) compact_fma_cplusab_ ( io_code, c21, a2, b1, numb, regset );
if (i2 && j2) compact_fma_cplusab_ ( io_code, c22, a2, b2, numb, regset );
if (i2 && j3) compact_fma_cplusab_ ( io_code, c23, a2, b3, numb, regset );
if (i2 && j4) compact_fma_cplusab_ ( io_code, c24, a2, b4, numb, regset );
if (i2 && j5) compact_fma_cplusab_ ( io_code, c25, a2, b5, numb, regset );
if (i2 && j6) compact_fma_cplusab_ ( io_code, c26, a2, b6, numb, regset );
if (i2 && j7) compact_fma_cplusab_ ( io_code, c27, a2, b7, numb, regset );
if (i3 && j1) compact_fma_cplusab_ ( io_code, c31, a3, b1, numb, regset );
if (i3 && j2) compact_fma_cplusab_ ( io_code, c32, a3, b2, numb, regset );
if (i3 && j3) compact_fma_cplusab_ ( io_code, c33, a3, b3, numb, regset );
if (i3 && j4) compact_fma_cplusab_ ( io_code, c34, a3, b4, numb, regset );
if (i3 && j5) compact_fma_cplusab_ ( io_code, c35, a3, b5, numb, regset );
if (i3 && j6) compact_fma_cplusab_ ( io_code, c36, a3, b6, numb, regset );
if (i3 && j7) compact_fma_cplusab_ ( io_code, c37, a3, b7, numb, regset );
if (i4 && j1) compact_fma_cplusab_ ( io_code, c41, a4, b1, numb, regset );
if (i4 && j2) compact_fma_cplusab_ ( io_code, c42, a4, b2, numb, regset );
if (i4 && j3) compact_fma_cplusab_ ( io_code, c43, a4, b3, numb, regset );
if (i4 && j4) compact_fma_cplusab_ ( io_code, c44, a4, b4, numb, regset );
if (i4 && j5) compact_fma_cplusab_ ( io_code, c45, a4, b5, numb, regset );
if (i4 && j6) compact_fma_cplusab_ ( io_code, c46, a4, b6, numb, regset );
if (i4 && j7) compact_fma_cplusab_ ( io_code, c47, a4, b7, numb, regset );
if (i5 && j1) compact_fma_cplusab_ ( io_code, c51, a5, b1, numb, regset );
if (i5 && j2) compact_fma_cplusab_ ( io_code, c52, a5, b2, numb, regset );
if (i5 && j3) compact_fma_cplusab_ ( io_code, c53, a5, b3, numb, regset );
if (i5 && j4) compact_fma_cplusab_ ( io_code, c54, a5, b4, numb, regset );
if (i5 && j5) compact_fma_cplusab_ ( io_code, c55, a5, b5, numb, regset );
if (i5 && j6) compact_fma_cplusab_ ( io_code, c56, a5, b6, numb, regset );
if (i5 && j7) compact_fma_cplusab_ ( io_code, c57, a5, b7, numb, regset );
if (i6 && j1) compact_fma_cplusab_ ( io_code, c61, a6, b1, numb, regset );
if (i6 && j2) compact_fma_cplusab_ ( io_code, c62, a6, b2, numb, regset );
if (i6 && j3) compact_fma_cplusab_ ( io_code, c63, a6, b3, numb, regset );
if (i6 && j4) compact_fma_cplusab_ ( io_code, c64, a6, b4, numb, regset );
if (i6 && j5) compact_fma_cplusab_ ( io_code, c65, a6, b5, numb, regset );
if (i6 && j6) compact_fma_cplusab_ ( io_code, c66, a6, b6, numb, regset );
if (i6 && j7) compact_fma_cplusab_ ( io_code, c67, a6, b7, numb, regset );
if (i7 && j1) compact_fma_cplusab_ ( io_code, c71, a7, b1, numb, regset );
if (i7 && j2) compact_fma_cplusab_ ( io_code, c72, a7, b2, numb, regset );
if (i7 && j3) compact_fma_cplusab_ ( io_code, c73, a7, b3, numb, regset );
if (i7 && j4) compact_fma_cplusab_ ( io_code, c74, a7, b4, numb, regset );
if (i7 && j5) compact_fma_cplusab_ ( io_code, c75, a7, b5, numb, regset );
if (i7 && j6) compact_fma_cplusab_ ( io_code, c76, a7, b6, numb, regset );
if (i7 && j7) compact_fma_cplusab_ ( io_code, c77, a7, b7, numb, regset );
} /* Inner loop */
/* Storing into C, do it one column at a time and reuse some regs */
for ( l = j ; l <= LIBXSMM_MIN(j+jun-1,bn2); l++ ) {
#ifdef COMPACT_GEMMNN_DEBUG
printf("Doing j wrap-up storage from %d to %d\n",j,LIBXSMM_MIN(j+jun-1,bn2));
#endif
if (l== j ) { c0=c00 ; c1=c10; c2=c20; c3=c30; c4=c40; c5=c50; c6=c60; c7=c70; }
if (l==j+1) { c0=c01 ; c1=c11; c2=c21; c3=c31; c4=c41; c5=c51; c6=c61; c7=c71; }
if (l==j+2) { c0=c02 ; c1=c12; c2=c22; c3=c32; c4=c42; c5=c52; c6=c62; c7=c72; }
if (l==j+3) { c0=c03 ; c1=c13; c2=c23; c3=c33; c4=c43; c5=c53; c6=c63; c7=c73; }
if (l==j+4) { c0=c04 ; c1=c14; c2=c24; c3=c34; c4=c44; c5=c54; c6=c64; c7=c74; }
if (l==j+5) { c0=c05 ; c1=c15; c2=c25; c3=c35; c4=c45; c5=c55; c6=c65; c7=c75; }
if (l==j+6) { c0=c06 ; c1=c16; c2=c26; c3=c36; c4=c46; c5=c56; c6=c66; c7=c76; }
if (l==j+7) { c0=c07 ; c1=c17; c2=c27; c3=c37; c4=c47; c5=c57; c6=c67; c7=c77; }
if ( beta == 1.0 ) {
if (i0 && j0) compact_load_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1, l-bn1+cn1, a0, numb, datasz, regset, creg );
if (i1 && j0) compact_load_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1+1, l-bn1+cn1, a1, numb, datasz, regset, creg );
if (i2 && j0) compact_load_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1+2, l-bn1+cn1, a2, numb, datasz, regset, creg );
if (i3 && j0) compact_load_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1+3, l-bn1+cn1, a3, numb, datasz, regset, creg );
if (i4 && j0) compact_load_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1+4, l-bn1+cn1, a4, numb, datasz, regset, creg );
if (i5 && j0) compact_load_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1+5, l-bn1+cn1, a5, numb, datasz, regset, creg );
if (i6 && j0) compact_load_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1+6, l-bn1+cn1, a6, numb, datasz, regset, creg );
if (i7 && j0) compact_load_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1+7, l-bn1+cn1, a7, numb, datasz, regset, creg );
} else if ( (beta == 0.0) && (alpha != 1.0) ) {
if (i0 && j0) compact_set_zero_( io_code, a0, numb, datasz, regset );
if (i1 && j0) compact_set_zero_( io_code, a1, numb, datasz, regset );
if (i2 && j0) compact_set_zero_( io_code, a2, numb, datasz, regset );
if (i3 && j0) compact_set_zero_( io_code, a3, numb, datasz, regset );
if (i4 && j0) compact_set_zero_( io_code, a4, numb, datasz, regset );
if (i5 && j0) compact_set_zero_( io_code, a5, numb, datasz, regset );
if (i6 && j0) compact_set_zero_( io_code, a6, numb, datasz, regset );
if (i7 && j0) compact_set_zero_( io_code, a7, numb, datasz, regset );
}
if ( alpha == -1.0 ) {
if (i0 && j0) compact_sub_two_nums_ ( io_code, a0, c0, c0, numb, regset );
if (i1 && j0) compact_sub_two_nums_ ( io_code, a1, c1, c1, numb, regset );
if (i2 && j0) compact_sub_two_nums_ ( io_code, a2, c2, c2, numb, regset );
if (i3 && j0) compact_sub_two_nums_ ( io_code, a3, c3, c3, numb, regset );
if (i4 && j0) compact_sub_two_nums_ ( io_code, a4, c4, c4, numb, regset );
if (i5 && j0) compact_sub_two_nums_ ( io_code, a5, c5, c5, numb, regset );
if (i6 && j0) compact_sub_two_nums_ ( io_code, a6, c6, c6, numb, regset );
if (i7 && j0) compact_sub_two_nums_ ( io_code, a7, c7, c7, numb, regset );
} else if ( (beta != 0.0) && (alpha==1.0) ) {
if (i0 && j0) compact_add_two_nums_ ( io_code, a0, c0, c0, numb, regset );
if (i1 && j0) compact_add_two_nums_ ( io_code, a1, c1, c1, numb, regset );
if (i2 && j0) compact_add_two_nums_ ( io_code, a2, c2, c2, numb, regset );
if (i3 && j0) compact_add_two_nums_ ( io_code, a3, c3, c3, numb, regset );
if (i4 && j0) compact_add_two_nums_ ( io_code, a4, c4, c4, numb, regset );
if (i5 && j0) compact_add_two_nums_ ( io_code, a5, c5, c5, numb, regset );
if (i6 && j0) compact_add_two_nums_ ( io_code, a6, c6, c6, numb, regset );
if (i7 && j0) compact_add_two_nums_ ( io_code, a7, c7, c7, numb, regset );
}
if (i0 && j0) compact_store_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1, l-bn1+cn1, c0, numb, datasz, regset, creg );
if (i1 && j0) compact_store_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1+1, l-bn1+cn1, c1, numb, datasz, regset, creg );
if (i2 && j0) compact_store_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1+2, l-bn1+cn1, c2, numb, datasz, regset, creg );
if (i3 && j0) compact_store_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1+3, l-bn1+cn1, c3, numb, datasz, regset, creg );
if (i4 && j0) compact_store_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1+4, l-bn1+cn1, c4, numb, datasz, regset, creg );
if (i5 && j0) compact_store_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1+5, l-bn1+cn1, c5, numb, datasz, regset, creg );
if (i6 && j0) compact_store_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1+6, l-bn1+cn1, c6, numb, datasz, regset, creg );
if (i7 && j0) compact_store_matrix_gen_ ( io_code, trc, ldc, i-am1+cm1+7, l-bn1+cn1, c7, numb, datasz, regset, creg );
} /* Store the results */
if ( loopi && j0 ) {
aoffset = datasz*iun*numb;
coffset = datasz*iun*numb;
if ( i == am1 ) {
#ifdef COMPACT_GEMMNN_DEBUG
printf("Should be putting in a m-jump soon: i=%d j=%d i0=%d j0=%d am1=%d am2=%d\n",i,j,i0,j0,am1,am2);
#endif
libxsmm_x86_instruction_alu_imm( io_code, LIBXSMM_X86_INSTR_ADDQ, areg, aoffset );
libxsmm_x86_instruction_alu_imm( io_code, LIBXSMM_X86_INSTR_ADDQ, creg, coffset );
libxsmm_x86_instruction_alu_imm( io_code, LIBXSMM_X86_INSTR_SUBQ, LIBXSMM_X86_GP_REG_RCX, 1 );
libxsmm_x86_instruction_jump_back_to_label( io_code, LIBXSMM_X86_INSTR_JG, &l_loop_label_tracker );
}
if ( (am2-i+1 < 2*iun) && ((mborder > 0) || (j + jun - 1 < bn2)) && (mloopadj==1) ) {
#ifdef COMPACT_GEMMNN_DEBUG
printf("Finished with m-loop, doing clean-up: i=%d i0=%d j0=%d mborder=%d j=%d jun=%d bn2=%d\n",i,i0,j0,mborder,j,jun,bn2);
#endif
aoffset = datasz*iun*numb*mloopcnt;
coffset = datasz*iun*numb*mloopcnt;
libxsmm_x86_instruction_alu_imm( io_code, LIBXSMM_X86_INSTR_SUBQ, areg, aoffset );
libxsmm_x86_instruction_alu_imm( io_code, LIBXSMM_X86_INSTR_SUBQ, creg, coffset );
mloopadj = 0;
}
i0 = 1; /* Turn everything back on again */
}
} /* M-loop */
if ( loopj ) {
coffset = ldc*datasz*jun*numb;
boffset = ldb*datasz*jun*numb;
if ( j == bn1 ) {
#ifdef COMPACT_GEMMNN_DEBUG
printf("Should be putting in a n-jump soon: j=%d bn1=%d bn2=%d\n",j,bn1,bn2);
#endif
libxsmm_x86_instruction_alu_imm( io_code, LIBXSMM_X86_INSTR_ADDQ, creg, coffset );
libxsmm_x86_instruction_alu_imm( io_code, LIBXSMM_X86_INSTR_ADDQ, breg, boffset );
libxsmm_x86_instruction_alu_imm( io_code, LIBXSMM_X86_INSTR_SUBQ, LIBXSMM_X86_GP_REG_RAX, 1 );
libxsmm_x86_instruction_jump_back_to_label( io_code, LIBXSMM_X86_INSTR_JG, &l_loop_label_tracker );
}
if ( (bn2-j+1 < 2*jun) && (nborder > 0) ) {
#ifdef COMPACT_GEMMNN_DEBUG
printf("Finished with n-loop, doing clean-up, j=%d\n",j);
#endif
coffset = ldc*datasz*jun*numb*nloopcnt;
boffset = ldb*datasz*jun*numb*nloopcnt;
libxsmm_x86_instruction_alu_imm( io_code, LIBXSMM_X86_INSTR_SUBQ, creg, coffset );
libxsmm_x86_instruction_alu_imm( io_code, LIBXSMM_X86_INSTR_SUBQ, breg, boffset );
}
j0 = 1; /* Turn everything back on again */
}
} /* N-loop */
#ifdef COMPACT_GEMMNN_DEBUG
printf("Inlined Compact GEMM code pointer ends at: %u\n",io_code->code_size);
#endif
}
#endif /*GENERATOR_PACKED_GEMMNN_H*/