real_128bit_BLOCK_template.c 491 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
//    This file is part of ELPA.
//
//    The ELPA library was originally created by the ELPA consortium,
//    consisting of the following organizations:
//
//    - Max Planck Computing and Data Facility (MPCDF), formerly known as
//      Rechenzentrum Garching der Max-Planck-Gesellschaft (RZG),
//    - Bergische Universität Wuppertal, Lehrstuhl für angewandte
//      Informatik,
//    - Technische Universität München, Lehrstuhl für Informatik mit
//      Schwerpunkt Wissenschaftliches Rechnen ,
//    - Fritz-Haber-Institut, Berlin, Abt. Theorie,
//    - Max-Plack-Institut für Mathematik in den Naturwissenschaften,
//      Leipzig, Abt. Komplexe Strukutren in Biologie und Kognition,
//      and
//    - IBM Deutschland GmbH
//
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
//    This particular source code file contains additions, changes and
//    enhancements authored by Intel Corporation which is not part of
//    the ELPA consortium.
//
//    More information can be found here:
//    http://elpa.mpcdf.mpg.de/
//
//    ELPA is free software: you can redistribute it and/or modify
//    it under the terms of the version 3 of the license of the
//    GNU Lesser General Public License as published by the Free
//    Software Foundation.
//
//    ELPA is distributed in the hope that it will be useful,
//    but WITHOUT ANY WARRANTY; without even the implied warranty of
//    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
//    GNU Lesser General Public License for more details.
//
//    You should have received a copy of the GNU Lesser General Public License
//    along with ELPA. If not, see <http://www.gnu.org/licenses/>
//
//    ELPA reflects a substantial effort on the part of the original
//    ELPA consortium, and we ask you to respect the spirit of the
//    license that we chose: i.e., please contribute any changes you
//    may have back to the original ELPA library distribution, and keep
//    any derivatives of ELPA under the same license that we chose for
//    the original distribution, the GNU Lesser General Public License.
//
// Author: Andreas Marek, MPCDF, based on the double precision case of A. Heinecke
//
#include "config-f90.h"

50
51
52
53
54
55
#define CONCAT_8ARGS(a, b, c, d, e, f, g, h) CONCAT2_8ARGS(a, b, c, d, e, f, g, h)
#define CONCAT2_8ARGS(a, b, c, d, e, f, g, h) a ## b ## c ## d ## e ## f ## g ## h

#define CONCAT_7ARGS(a, b, c, d, e, f, g) CONCAT2_7ARGS(a, b, c, d, e, f, g)
#define CONCAT2_7ARGS(a, b, c, d, e, f, g) a ## b ## c ## d ## e ## f ## g

56
57
58
59
60
61
62
63
64
65
66
67
#define CONCAT_6ARGS(a, b, c, d, e, f) CONCAT2_6ARGS(a, b, c, d, e, f)
#define CONCAT2_6ARGS(a, b, c, d, e, f) a ## b ## c ## d ## e ## f

#define CONCAT_5ARGS(a, b, c, d, e) CONCAT2_5ARGS(a, b, c, d, e)
#define CONCAT2_5ARGS(a, b, c, d, e) a ## b ## c ## d ## e

#define CONCAT_4ARGS(a, b, c, d) CONCAT2_4ARGS(a, b, c, d)
#define CONCAT2_4ARGS(a, b, c, d) a ## b ## c ## d

#define CONCAT_3ARGS(a, b, c) CONCAT2_3ARGS(a, b, c)
#define CONCAT2_3ARGS(a, b, c) a ## b ## c

68
#if VEC_SET == 128 || VEC_SET == 256 || VEC_SET == 512
69
70
#include <x86intrin.h>
#endif
71
#if VEC_SET == 1281
72
73
74
75
76
77
#include <fjmfunc.h>
#include <emmintrin.h>
#endif
#include <stdio.h>
#include <stdlib.h>

78
79
80
81
82
#ifdef BLOCK6
#define PREFIX hexa
#define BLOCK 6
#endif

83
84
85
86
87
88
89
90
91
92
#ifdef BLOCK4
#define PREFIX quad
#define BLOCK 4
#endif

#ifdef BLOCK2
#define PREFIX double
#define BLOCK 2
#endif

93
#if VEC_SET == 128
94
95
96
#define SIMD_SET SSE
#endif

97
#if VEC_SET == 1281
98
99
#define SIMD_SET SPARC64
#endif
100
101
102
103
104

#if VEC_SET == 256
#define SIMD_SET AVX_AVX2
#endif

105
106
107
108
#if VEC_SET == 512
#define SIMD_SET AVX512
#endif

109
110
#define __forceinline __attribute__((always_inline)) static

111
#if VEC_SET == 128 || VEC_SET == 1281
112
113
#ifdef DOUBLE_PRECISION_REAL
#define offset 2
114
115
116
117
118
119
120
#define __SIMD_DATATYPE __m128d
#define _SIMD_LOAD _mm_load_pd
#define _SIMD_STORE _mm_store_pd
#define _SIMD_ADD _mm_add_pd
#define _SIMD_MUL _mm_mul_pd
#define _SIMD_SUB _mm_sub_pd
#define _SIMD_XOR _mm_xor_pd
121
122
123
#if VEC_SET == 128
#define _SIMD_SET _mm_set_pd
#define _SIMD_SET1 _mm_set1_pd
124
#endif
125
#endif /* DOUBLE_PRECISION_REAL */
126
127
128
129
130
131
132
133
134
#ifdef SINGLE_PRECISION_REAL
#define offset 4
#define __SIMD_DATATYPE __m128
#define _SIMD_LOAD _mm_load_ps
#define _SIMD_STORE _mm_store_ps
#define _SIMD_ADD _mm_add_ps
#define _SIMD_MUL _mm_mul_ps
#define _SIMD_SUB _mm_sub_ps
#define _SIMD_XOR _mm_xor_ps
135
136
137
138
139
#if VEC_SET == 128
#define _SIMD_SET _mm_set_ps
#define _SIMD_SET1 _mm_set1_ps
#endif 
#endif /* SINGLE_PRECISION_REAL */
140
#endif /* VEC_SET == 128 || VEC_SET == 1281 */
141

142
143
144
145
146
147
148
149
#if VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#define offset 4
#define __SIMD_DATATYPE __m256d
#define _SIMD_LOAD _mm256_load_pd
#define _SIMD_STORE _mm256_store_pd
#define _SIMD_ADD _mm256_add_pd
#define _SIMD_MUL _mm256_mul_pd
150
#define _SIMD_SUB _mm256_sub_pd
151
#define _SIMD_SET1 _mm256_set1_pd
152
153
154
155
156
157
#define _SIMD_XOR _mm256_xor_pd
#define _SIMD_BROADCAST _mm256_broadcast_sd
#ifdef HAVE_AVX2
#ifdef __FMA4__
#define __ELPA_USE_FMA__
#define _mm256_FMA_pd(a,b,c) _mm256_macc_pd(a,b,c)
158
159
160
161
#define _mm256_NFMA_pd(a,b,c) _mm256_nmacc_pd(a,b,c)
#error "This should be prop _mm256_msub_pd instead of _mm256_msub"
#define _mm256_FMSUB_pd(a,b,c) _mm256_msub(a,b,c)
#endif /* __FMA4__ */
162
163
164
#ifdef __AVX2__
#define __ELPA_USE_FMA__
#define _mm256_FMA_pd(a,b,c) _mm256_fmadd_pd(a,b,c)
165
166
167
168
#define _mm256_NFMA_pd(a,b,c) _mm256_fnmadd_pd(a,b,c)
#define _mm256_FMSUB_pd(a,b,c) _mm256_fmsub_pd(a,b,c)
#endif /* __AVX2__ */
#ifdef __ELPA_USE_FMA__
169
#define _SIMD_FMA _mm256_FMA_pd
170
171
#define _SIMD_NFMA _mm256_NFMA_pd
#define _SIMD_FMSUB _mm256_FMSUB_pd
172
173
174
175
176
177
178
179
180
181
182
#endif
#endif /* HAVE_AVX2 */
#endif /* DOUBLE_PRECISION_REAL */

#ifdef SINGLE_PRECISION_REAL
#define offset 8
#define __SIMD_DATATYPE __m256
#define _SIMD_LOAD _mm256_load_ps
#define _SIMD_STORE _mm256_store_ps
#define _SIMD_ADD _mm256_add_ps
#define _SIMD_MUL _mm256_mul_ps
183
#define _SIMD_SUB _mm256_sub_ps
184
#define _SIMD_SET1 _mm256_set1_ps
185
186
187
188
189
190
#define _SIMD_XOR _mm256_xor_ps
#define _SIMD_BROADCAST _mm256_broadcast_ss
#ifdef HAVE_AVX2
#ifdef __FMA4__
#define __ELPA_USE_FMA__
#define _mm256_FMA_ps(a,b,c) _mm256_macc_ps(a,b,c)
191
192
193
194
#define _mm256_NFMA_ps(a,b,c) _mm256_nmacc_ps(a,b,c)
#error "This should be prop _mm256_msub_ps instead of _mm256_msub"
#define _mm256_FMSUB_ps(a,b,c) _mm256_msub(a,b,c)
#endif /* __FMA4__ */
195
196
197
#ifdef __AVX2__
#define __ELPA_USE_FMA__
#define _mm256_FMA_ps(a,b,c) _mm256_fmadd_ps(a,b,c)
198
199
200
201
#define _mm256_NFMA_ps(a,b,c) _mm256_fnmadd_ps(a,b,c)
#define _mm256_FMSUB_ps(a,b,c) _mm256_fmsub_ps(a,b,c)
#endif /* __AVX2__ */
#ifdef __ELPA_USE_FMA__
202
#define _SIMD_FMA _mm256_FMA_ps
203
204
#define _SIMD_NFMA _mm256_NFMA_ps
#define _SIMD_FMSUB _mm256_FMSUB_ps
205
206
207
208
209
#endif
#endif /* HAVE_AVX2 */
#endif /* SINGLE_PRECISION_REAL */
#endif /* VEC_SET == 256 */

210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
#if VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#define offset 8
#define __SIMD_DATATYPE __m512d
#define __SIMD_INTEGER  __m512i
#define _SIMD_LOAD _mm512_load_pd
#define _SIMD_STORE _mm512_store_pd
#define _SIMD_ADD _mm512_add_pd
#define _SIMD_MUL _mm512_mul_pd
#define _SIMD_SUB _mm512_sub_pd
#define _SIMD_SET1 _mm512_set1_pd
#ifdef HAVE_AVX512_XEON
#define _SIMD_XOR _mm512_xor_pd
#endif
#ifdef HAVE_AVX512
#define __ELPA_USE_FMA__
#define _mm512_FMA_pd(a,b,c) _mm512_fmadd_pd(a,b,c)
227
228
#define _mm512_NFMA_pd(a,b,c) _mm512_fnmadd_pd(a,b,c)
#define _mm512_FMSUB_pd(a,b,c) _mm512_fmsub_pd(a,b,c)
229
230
#ifdef __ELPA_USE_FMA__
#define _SIMD_FMA _mm512_FMA_pd
231
232
#define _SIMD_NFMA _mm512_NFMA_pd
#define _SIMD_FMSUB _mm512_FMSUB_pd
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
#endif
#endif /* HAVE_AVX512 */
#endif /* DOUBLE_PRECISION_REAL */

#ifdef SINGLE_PRECISION_REAL
#define offset 16
#define __SIMD_DATATYPE __m512
#define __SIMD_INTEGER  __m512i
#define _SIMD_LOAD _mm512_load_ps
#define _SIMD_STORE _mm512_store_ps
#define _SIMD_ADD _mm512_add_ps
#define _SIMD_MUL _mm512_mul_ps
#define _SIMD_SUB _mm512_sub_ps
#define _SIMD_SET1 _mm512_set1_ps
#ifdef HAVE_AVX512_XEON
#define _SIMD_XOR _mm512_xor_ps
#endif
#ifdef HAVE_AVX512
#define __ELPA_USE_FMA__
#define _mm512_FMA_ps(a,b,c) _mm512_fmadd_ps(a,b,c)
253
254
#define _mm512_NFMA_ps(a,b,c) _mm512_fnmadd_ps(a,b,c)
#define _mm512_FMSUB_ps(a,b,c) _mm512_fmsub_ps(a,b,c)
255
256
#ifdef __ELPA_USE_FMA__
#define _SIMD_FMA _mm512_FMA_ps
257
258
#define _SIMD_NFMA _mm512_NFMA_ps
#define _SIMD_FMSUB _mm512_FMSUB_ps
259
260
261
262
263
#endif
#endif /* HAVE_AVX512 */
#endif /* SINGLE_PRECISION_REAL */
#endif /* VEC_SET == 512 */

264
#ifdef DOUBLE_PRECISION_REAL
265
#define WORD_LENGTH double
266
267
#define DATA_TYPE double
#define DATA_TYPE_PTR double*
268
269
270
271
272
273
274
275
#endif
#ifdef SINGLE_PRECISION_REAL
#define WORD_LENGTH single
#define DATA_TYPE float
#define DATA_TYPE_PTR float*
#endif

#if VEC_SET == 128
276
277
278
#undef __AVX__
#endif

279
280

#if VEC_SET == 128 || VEC_SET == 1281
281
282
//Forward declaration
#ifdef DOUBLE_PRECISION_REAL
283
284
#undef ROW_LENGTH
#define ROW_LENGTH 2
285
#endif
286
287
288
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 4
289
#endif
290
#endif /* VEC_SET == 128 || VEC_SET == 1281 */
291

292
293
294
295
296
297
298
299
300
301
#if VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 4
#endif
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 8
#endif
#endif /* VEC_SET == 256 */
302
303
304
305
306
307
308
309
310
311
312

#if VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 8
#endif
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 16
#endif
#endif /* VEC_SET == 512 */
313
__forceinline void CONCAT_8ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int nb, int ldq, int ldh, 
314
#ifdef BLOCK2
315
	DATA_TYPE s);
316
317
#endif
#ifdef BLOCK4
318
	DATA_TYPE s_1_2, DATA_TYPE s_1_3, DATA_TYPE s_2_3, DATA_TYPE s_1_4, DATA_TYPE s_2_4, DATA_TYPE s_3_4);
319
#endif
320
321
#ifdef BLOCK6
	DATA_TYPE_PTR scalarprods);
322
#endif
323

324
#if VEC_SET == 128 || VEC_SET == 1281
325
326
327
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 4
328
#endif
329
330
331
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 8
332
#endif
333
334
335
336
337
338
339
340
341
342
343
344
#endif /* VEC_SET == 128 || VEC_SET == 1281 */

#if VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 8
#endif
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 16
#endif
#endif /* VEC_SET == 256 */
345
346
347
348
349
350
351
352
353
354
355

#if VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 16
#endif
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 32
#endif
#endif /* VEC_SET == 512 */
356
__forceinline void CONCAT_8ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int nb, int ldq, int ldh, 
357
#ifdef BLOCK2
358
	DATA_TYPE s);
359
360
#endif
#ifdef BLOCK4
361
	DATA_TYPE s_1_2, DATA_TYPE s_1_3, DATA_TYPE s_2_3, DATA_TYPE s_1_4, DATA_TYPE s_2_4, DATA_TYPE s_3_4);
362
#endif
363
364
#ifdef BLOCK6
	DATA_TYPE_PTR scalarprods);
365
#endif
366

367
#if VEC_SET == 128 || VEC_SET == 1281 
368
369
370
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 6
371
372
#endif
#ifdef SINGLE_PRECISION_REAL
373
374
#undef ROW_LENGTH
#define ROW_LENGTH 12
375
#endif
376
377
378
379
380
381
382
383
384
385
386
387
388
#endif /* VEC_SET == 128 || VEC_SET == 1281  */

#if VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 12
#endif
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 24
#endif
#endif /* VEC_SET == 256 */

389
390
391
392
393
394
395
396
397
398
399
#if VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 24
#endif
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 48
#endif
#endif /* VEC_SET == 512 */

400
__forceinline void CONCAT_8ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int nb, int ldq, int ldh,
401
#ifdef BLOCK2
402
	DATA_TYPE s);
403
404
#endif
#ifdef BLOCK4
405
	DATA_TYPE s_1_2, DATA_TYPE s_1_3, DATA_TYPE s_2_3, DATA_TYPE s_1_4, DATA_TYPE s_2_4, DATA_TYPE s_3_4);
406
#endif
407
408
#ifdef BLOCK6
	DATA_TYPE_PTR scalarprods);
409
#endif
410

411
#if VEC_SET == 128 || VEC_SET == 1281 
412
413
414
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 8
415
#endif
416
417
418
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 16
419
#endif
420
421
422
423
424
425
426
427
428
429
430
431
432
#endif /* VEC_SET == 128 || VEC_SET == 1281  */

#if VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 16
#endif
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 32
#endif
#endif /* VEC_SET == 256 */

433
434
435
436
437
438
439
440
441
442
443
#if VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 32
#endif
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 64
#endif
#endif /* VEC_SET == 512 */

444
__forceinline void CONCAT_8ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int nb, int ldq, int ldh, 
445
#ifdef BLOCK2
446
	DATA_TYPE s);
447
448
#endif
#ifdef BLOCK4
449
	DATA_TYPE s_1_2, DATA_TYPE s_1_3, DATA_TYPE s_2_3, DATA_TYPE s_1_4, DATA_TYPE s_2_4, DATA_TYPE s_3_4);
450
#endif
451
452
#ifdef BLOCK6
	DATA_TYPE_PTR scalarprods);
453
#endif
454

455
#if  VEC_SET == 128 || VEC_SET == 1281
456
457
458
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 10
459
#endif
460
461
462
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 20
463
#endif
464
465
466
467
468
469
470
471
472
473
474
475
476
#endif /*  VEC_SET == 128 || VEC_SET == 1281 */

#if  VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 20
#endif
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 40
#endif
#endif /*  VEC_SET == 256 */

477
478
479
480
481
482
483
484
485
486
#if VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 40
#endif
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 80
#endif
#endif /* VEC_SET == 512 */
487

488
__forceinline void CONCAT_8ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int nb, int ldq, int ldh, 
489
#ifdef BLOCK2
490
	DATA_TYPE s);
491
492
#endif
#ifdef BLOCK4
493
494
495
496
	DATA_TYPE s_1_2, DATA_TYPE s_1_3, DATA_TYPE s_2_3, DATA_TYPE s_1_4, DATA_TYPE s_2_4, DATA_TYPE s_3_4);
#endif
#ifdef BLOCK6
	DATA_TYPE_PTR scalarprods);
497
498
#endif

499
#if  VEC_SET == 128 || VEC_SET == 1281
500
#ifdef DOUBLE_PRECISION_REAL
501
502
#undef ROW_LENGTH
#define ROW_LENGTH 12
503
504
#endif
#ifdef SINGLE_PRECISION_REAL
505
506
507
#undef ROW_LENGTH
#define ROW_LENGTH 24
#endif
508
509
510
511
512
513
514
515
516
517
518
519
#endif /* VEC_SET == 128 || VEC_SET == 1281 */

#if  VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 24
#endif
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 48
#endif
#endif /*  VEC_SET == 256 */
520

521
522
523
524
525
526
527
528
529
530
531
#if VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 48
#endif
#ifdef SINGLE_PRECISION_REAL
#undef ROW_LENGTH
#define ROW_LENGTH 96
#endif
#endif /* VEC_SET == 512 */

532
533
534
__forceinline void CONCAT_8ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int nb, int ldq, int ldh,
#ifdef BLOCK2
	DATA_TYPE s);
535
#endif
536
537
538
539
540
541
542
543
#ifdef BLOCK4
	DATA_TYPE s_1_2, DATA_TYPE s_1_3, DATA_TYPE s_2_3, DATA_TYPE s_1_4, DATA_TYPE s_2_4, DATA_TYPE s_3_4);
#endif
#ifdef BLOCK6
	DATA_TYPE_PTR scalarprods);
#endif

void CONCAT_7ARGS(PREFIX,_hh_trafo_real_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int* pnb, int* pnq, int* pldq, int* pldh);
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

/*
!f>#ifdef HAVE_SSE_INTRINSICS
!f> interface
!f>   subroutine double_hh_trafo_real_SSE_2hv_double(q, hh, pnb, pnq, pldq, pldh) &
!f>                                bind(C, name="double_hh_trafo_real_SSE_2hv_double")
!f>        use, intrinsic :: iso_c_binding
!f>        integer(kind=c_int)        :: pnb, pnq, pldq, pldh
!f>        type(c_ptr), value        :: q
!f>        real(kind=c_double)        :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/
/*
!f>#ifdef HAVE_SSE_INTRINSICS
!f> interface
!f>   subroutine double_hh_trafo_real_SSE_2hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f>              bind(C, name="double_hh_trafo_real_SSE_2hv_single")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int) :: pnb, pnq, pldq, pldh
!f>     type(c_ptr), value  :: q
!f>     real(kind=c_float)  :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/

/*
!f>#ifdef HAVE_SPARC64_SSE
!f> interface
!f>   subroutine double_hh_trafo_real_SPARC64_2hv_double(q, hh, pnb, pnq, pldq, pldh) &
!f>              bind(C, name="double_hh_trafo_real_SPARC64_2hv_double")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int) :: pnb, pnq, pldq, pldh
!f>     type(c_ptr), value  :: q
!f>     real(kind=c_double) :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/

/*
!f>#ifdef HAVE_SPARC64_SSE
!f> interface
!f>   subroutine double_hh_trafo_real_SPARC64_2hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f>              bind(C, name="double_hh_trafo_real_SPARC64_2hv_single")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int) :: pnb, pnq, pldq, pldh
!f>     type(c_ptr), value  :: q
!f>     real(kind=c_float)  :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/

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
/*
!f>#if defined(HAVE_AVX) || defined(HAVE_AVX2)
!f> interface
!f>   subroutine double_hh_trafo_real_AVX_AVX2_2hv_double(q, hh, pnb, pnq, pldq, pldh) &
!f>                                bind(C, name="double_hh_trafo_real_AVX_AVX2_2hv_double")
!f>        use, intrinsic :: iso_c_binding
!f>        integer(kind=c_int)        :: pnb, pnq, pldq, pldh
!f>        type(c_ptr), value        :: q
!f>        real(kind=c_double)        :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/

/*
!f>#if defined(HAVE_AVX) || defined(HAVE_AVX2)
!f> interface
!f>   subroutine double_hh_trafo_real_AVX_AVX2_2hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f>                                bind(C, name="double_hh_trafo_real_AVX_AVX2_2hv_single")
!f>        use, intrinsic :: iso_c_binding
!f>        integer(kind=c_int)       :: pnb, pnq, pldq, pldh
!f>        type(c_ptr), value        :: q
!f>        real(kind=c_float)        :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/

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
/*
!f>#if defined(HAVE_AVX512)
!f> interface
!f>   subroutine double_hh_trafo_real_AVX512_2hv_double(q, hh, pnb, pnq, pldq, pldh) &
!f>                             bind(C, name="double_hh_trafo_real_AVX512_2hv_double")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int)     :: pnb, pnq, pldq, pldh
!f>     type(c_ptr), value      :: q
!f>     real(kind=c_double)     :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/
/*
!f>#if defined(HAVE_AVX512)
!f> interface
!f>   subroutine double_hh_trafo_real_AVX512_2hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f>                             bind(C, name="double_hh_trafo_real_AVX512_2hv_single")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int)     :: pnb, pnq, pldq, pldh
!f>     type(c_ptr), value      :: q
!f>     real(kind=c_float)      :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/

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
/*
!f>#ifdef HAVE_SSE_INTRINSICS
!f> interface
!f>   subroutine quad_hh_trafo_real_SSE_4hv_double(q, hh, pnb, pnq, pldq, pldh) &
!f>                                bind(C, name="quad_hh_trafo_real_SSE_4hv_double")
!f>        use, intrinsic :: iso_c_binding
!f>        integer(kind=c_int)        :: pnb, pnq, pldq, pldh
!f>        type(c_ptr), value        :: q
!f>        real(kind=c_double)        :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/
/*
!f>#ifdef HAVE_SSE_INTRINSICS
!f> interface
!f>   subroutine quad_hh_trafo_real_SSE_4hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f>              bind(C, name="quad_hh_trafo_real_SSE_4hv_single")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int) :: pnb, pnq, pldq, pldh
!f>     type(c_ptr), value  :: q
!f>     real(kind=c_float)  :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/

/*
!f>#ifdef HAVE_SPARC64_SSE
!f> interface
!f>   subroutine quad_hh_trafo_real_SPARC64_4hv_double(q, hh, pnb, pnq, pldq, pldh) &
!f>              bind(C, name="quad_hh_trafo_real_SPARC64_4hv_double")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int) :: pnb, pnq, pldq, pldh
!f>     type(c_ptr), value  :: q
!f>     real(kind=c_double) :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/

/*
!f>#ifdef HAVE_SPARC64_SSE
!f> interface
!f>   subroutine quad_hh_trafo_real_SPARC64_4hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f>              bind(C, name="quad_hh_trafo_real_SPARC64_4hv_single")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int) :: pnb, pnq, pldq, pldh
!f>     type(c_ptr), value  :: q
!f>     real(kind=c_float)  :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/
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

/*
!f>#if defined(HAVE_AVX) || defined(HAVE_AVX2)
!f> interface
!f>   subroutine quad_hh_trafo_real_AVX_AVX2_4hv_double(q, hh, pnb, pnq, pldq, pldh) &
!f>                                bind(C, name="quad_hh_trafo_real_AVX_AVX2_4hv_double")
!f>        use, intrinsic :: iso_c_binding
!f>        integer(kind=c_int)        :: pnb, pnq, pldq, pldh
!f>        type(c_ptr), value        :: q
!f>        real(kind=c_double)        :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/
/*
!f>#if defined(HAVE_AVX) || defined(HAVE_AVX2)
!f> interface
!f>   subroutine quad_hh_trafo_real_AVX_AVX2_4hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f>              bind(C, name="quad_hh_trafo_real_AVX_AVX2_4hv_single")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int) :: pnb, pnq, pldq, pldh
!f>     type(c_ptr), value  :: q
!f>     real(kind=c_float)  :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/

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
/*
!f>#if defined(HAVE_AVX512)
!f> interface
!f>   subroutine quad_hh_trafo_real_AVX512_4hv_double(q, hh, pnb, pnq, pldq, pldh) &
!f>                             bind(C, name="quad_hh_trafo_real_AVX512_4hv_double")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int)     :: pnb, pnq, pldq, pldh
!f>     type(c_ptr), value      :: q
!f>     real(kind=c_double)     :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/

/*
!f>#if defined(HAVE_AVX512)
!f> interface
!f>   subroutine quad_hh_trafo_real_AVX512_4hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f>                             bind(C, name="quad_hh_trafo_real_AVX512_4hv_single")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int)     :: pnb, pnq, pldq, pldh
!f>     type(c_ptr), value      :: q
!f>     real(kind=c_float)      :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/
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
/*
!f>#ifdef HAVE_SSE_INTRINSICS
!f> interface
!f>   subroutine hexa_hh_trafo_real_sse_6hv_double(q, hh, pnb, pnq, pldq, pldh) &
!f>                                bind(C, name="hexa_hh_trafo_real_SSE_6hv_double")
!f>        use, intrinsic :: iso_c_binding
!f>        integer(kind=c_int)        :: pnb, pnq, pldq, pldh
!f>        type(c_ptr), value        :: q
!f>        real(kind=c_double)        :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/
/*
!f>#ifdef HAVE_SPARC64_SSE
!f> interface
!f>   subroutine hexa_hh_trafo_real_sparc64_6hv_double(q, hh, pnb, pnq, pldq, pldh) &
!f>                                bind(C, name="hexa_hh_trafo_real_SPARC64_6hv_double")
!f>        use, intrinsic :: iso_c_binding
!f>        integer(kind=c_int)        :: pnb, pnq, pldq, pldh
!f>        type(c_ptr), value        :: q
!f>        real(kind=c_double)        :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/
/*
!f>#ifdef HAVE_SSE_INTRINSICS
!f> interface
!f>   subroutine hexa_hh_trafo_real_sse_6hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f>                                bind(C, name="hexa_hh_trafo_real_SSE_6hv_single")
!f>        use, intrinsic :: iso_c_binding
!f>        integer(kind=c_int)        :: pnb, pnq, pldq, pldh
!f>        type(c_ptr), value        :: q
!f>        real(kind=c_float)        :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/
/*
!f>#ifdef HAVE_SPARC64_SSE
!f> interface
!f>   subroutine hexa_hh_trafo_real_sparc64_6hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f>                                bind(C, name="hexa_hh_trafo_real_SPARC64_6hv_single")
!f>        use, intrinsic :: iso_c_binding
!f>        integer(kind=c_int)        :: pnb, pnq, pldq, pldh
!f>        type(c_ptr), value        :: q
!f>        real(kind=c_float)        :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/

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
/*
!f>#if defined(HAVE_AVX) || defined(HAVE_AVX2)
!f> interface
!f>   subroutine hexa_hh_trafo_real_AVX_AVX2_6hv_double(q, hh, pnb, pnq, pldq, pldh) &
!f>                             bind(C, name="hexa_hh_trafo_real_AVX_AVX2_6hv_double")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int)     :: pnb, pnq, pldq, pldh
!f>     type(c_ptr), value      :: q
!f>     real(kind=c_double)     :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/
/*
!f>#if defined(HAVE_AVX) || defined(HAVE_AVX2)
!f> interface
!f>   subroutine hexa_hh_trafo_real_AVX_AVX2_6hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f>                             bind(C, name="hexa_hh_trafo_real_AVX_AVX2_6hv_single")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int)     :: pnb, pnq, pldq, pldh
!f>     type(c_ptr), value      :: q
!f>     real(kind=c_float)      :: hh(pnb,6)
!f>   end subroutine
!f> end interface
!f>#endif
*/


845
void CONCAT_7ARGS(PREFIX,_hh_trafo_real_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int* pnb, int* pnq, int* pldq, int* pldh)
846
847
848
849
850
851
852
853
{
  int i;
  int nb = *pnb;
  int nq = *pldq;
  int ldq = *pldq;
  int ldh = *pldh;
  int worked_on;

854
855
  worked_on = 0;

856
857
858
#ifdef BLOCK2
  // calculating scalar product to compute
  // 2 householder vectors simultaneously
859
  DATA_TYPE s = hh[(ldh)+1]*1.0;
860
861
862
863
864
#endif

#ifdef BLOCK4
  // calculating scalar products to compute
  // 4 householder vectors simultaneously
865
866
867
868
869
870
871
  DATA_TYPE s_1_2 = hh[(ldh)+1];  
  DATA_TYPE s_1_3 = hh[(ldh*2)+2];
  DATA_TYPE s_2_3 = hh[(ldh*2)+1];
  DATA_TYPE s_1_4 = hh[(ldh*3)+3];
  DATA_TYPE s_2_4 = hh[(ldh*3)+2];
  DATA_TYPE s_3_4 = hh[(ldh*3)+1];

872
873
  // calculate scalar product of first and fourth householder Vector
  // loop counter = 2
874
875
876
  s_1_2 += hh[2-1] * hh[(2+ldh)];          
  s_2_3 += hh[(ldh)+2-1] * hh[2+(ldh*2)];  
  s_3_4 += hh[(ldh*2)+2-1] * hh[2+(ldh*3)];
877
878

  // loop counter = 3
879
880
881
  s_1_2 += hh[3-1] * hh[(3+ldh)];          
  s_2_3 += hh[(ldh)+3-1] * hh[3+(ldh*2)];  
  s_3_4 += hh[(ldh*2)+3-1] * hh[3+(ldh*3)];
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
  s_1_3 += hh[3-2] * hh[3+(ldh*2)];        
  s_2_4 += hh[(ldh*1)+3-2] * hh[3+(ldh*3)];
#endif /* BLOCK4 */

#ifdef BLOCK6
  // calculating scalar products to compute
  // 6 householder vectors simultaneously
  DATA_TYPE scalarprods[15];

  scalarprods[0] = hh[(ldh+1)];  
  scalarprods[1] = hh[(ldh*2)+2];
  scalarprods[2] = hh[(ldh*2)+1];
  scalarprods[3] = hh[(ldh*3)+3];
  scalarprods[4] = hh[(ldh*3)+2];
  scalarprods[5] = hh[(ldh*3)+1];
  scalarprods[6] = hh[(ldh*4)+4];
  scalarprods[7] = hh[(ldh*4)+3];
  scalarprods[8] = hh[(ldh*4)+2];
  scalarprods[9] = hh[(ldh*4)+1];
  scalarprods[10] = hh[(ldh*5)+5];
  scalarprods[11] = hh[(ldh*5)+4];
  scalarprods[12] = hh[(ldh*5)+3];
  scalarprods[13] = hh[(ldh*5)+2];
  scalarprods[14] = hh[(ldh*5)+1];

  // calculate scalar product of first and fourth householder Vector
  // loop counter = 2
  scalarprods[0] += hh[1] * hh[(2+ldh)];           
  scalarprods[2] += hh[(ldh)+1] * hh[2+(ldh*2)];   
  scalarprods[5] += hh[(ldh*2)+1] * hh[2+(ldh*3)]; 
  scalarprods[9] += hh[(ldh*3)+1] * hh[2+(ldh*4)]; 
  scalarprods[14] += hh[(ldh*4)+1] * hh[2+(ldh*5)];

  // loop counter = 3
  scalarprods[0] += hh[2] * hh[(3+ldh)];          
  scalarprods[2] += hh[(ldh)+2] * hh[3+(ldh*2)];  
  scalarprods[5] += hh[(ldh*2)+2] * hh[3+(ldh*3)];
  scalarprods[9] += hh[(ldh*3)+2] * hh[3+(ldh*4)];
  scalarprods[14] += hh[(ldh*4)+2] * hh[3+(ldh*5)];

  scalarprods[1] += hh[1] * hh[3+(ldh*2)];         
  scalarprods[4] += hh[(ldh*1)+1] * hh[3+(ldh*3)]; 
  scalarprods[8] += hh[(ldh*2)+1] * hh[3+(ldh*4)]; 
  scalarprods[13] += hh[(ldh*3)+1] * hh[3+(ldh*5)];

  // loop counter = 4
  scalarprods[0] += hh[3] * hh[(4+ldh)];           
  scalarprods[2] += hh[(ldh)+3] * hh[4+(ldh*2)];   
  scalarprods[5] += hh[(ldh*2)+3] * hh[4+(ldh*3)]; 
  scalarprods[9] += hh[(ldh*3)+3] * hh[4+(ldh*4)]; 
  scalarprods[14] += hh[(ldh*4)+3] * hh[4+(ldh*5)];

  scalarprods[1] += hh[2] * hh[4+(ldh*2)];         
  scalarprods[4] += hh[(ldh*1)+2] * hh[4+(ldh*3)]; 
  scalarprods[8] += hh[(ldh*2)+2] * hh[4+(ldh*4)]; 
  scalarprods[13] += hh[(ldh*3)+2] * hh[4+(ldh*5)];

  scalarprods[3] += hh[1] * hh[4+(ldh*3)];         
  scalarprods[7] += hh[(ldh)+1] * hh[4+(ldh*4)];   
  scalarprods[12] += hh[(ldh*2)+1] * hh[4+(ldh*5)];

  // loop counter = 5
  scalarprods[0] += hh[4] * hh[(5+ldh)];           
  scalarprods[2] += hh[(ldh)+4] * hh[5+(ldh*2)];   
  scalarprods[5] += hh[(ldh*2)+4] * hh[5+(ldh*3)]; 
  scalarprods[9] += hh[(ldh*3)+4] * hh[5+(ldh*4)]; 
  scalarprods[14] += hh[(ldh*4)+4] * hh[5+(ldh*5)];

  scalarprods[1] += hh[3] * hh[5+(ldh*2)];         
  scalarprods[4] += hh[(ldh*1)+3] * hh[5+(ldh*3)]; 
  scalarprods[8] += hh[(ldh*2)+3] * hh[5+(ldh*4)]; 
  scalarprods[13] += hh[(ldh*3)+3] * hh[5+(ldh*5)];

  scalarprods[3] += hh[2] * hh[5+(ldh*3)];         
  scalarprods[7] += hh[(ldh)+2] * hh[5+(ldh*4)];   
  scalarprods[12] += hh[(ldh*2)+2] * hh[5+(ldh*5)];

  scalarprods[6] += hh[1] * hh[5+(ldh*4)];         
  scalarprods[11] += hh[(ldh)+1] * hh[5+(ldh*5)];  


#endif /* BLOCK6 */
965

966
#if VEC_SET == 128 || VEC_SET == 256 || VEC_SET == 512
967
968
969
970
971
972
973
  #pragma ivdep
#endif
  for (i = BLOCK; i < nb; i++)
    {
#ifdef BLOCK2
      s += hh[i-1] * hh[(i+ldh)];
#endif
974

975
#ifdef BLOCK4
976
977
978
      s_1_2 += hh[i-1] * hh[(i+ldh)];           
      s_2_3 += hh[(ldh)+i-1] * hh[i+(ldh*2)];   
      s_3_4 += hh[(ldh*2)+i-1] * hh[i+(ldh*3)]; 
979

980
981
982
983
984
      s_1_3 += hh[i-2] * hh[i+(ldh*2)];         
      s_2_4 += hh[(ldh*1)+i-2] * hh[i+(ldh*3)]; 

      s_1_4 += hh[i-3] * hh[i+(ldh*3)];         
#endif /* BLOCK4 */
985

986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
#ifdef BLOCK6
      scalarprods[0] += hh[i-1] * hh[(i+ldh)];           
      scalarprods[2] += hh[(ldh)+i-1] * hh[i+(ldh*2)];   
      scalarprods[5] += hh[(ldh*2)+i-1] * hh[i+(ldh*3)]; 
      scalarprods[9] += hh[(ldh*3)+i-1] * hh[i+(ldh*4)]; 
      scalarprods[14] += hh[(ldh*4)+i-1] * hh[i+(ldh*5)];

      scalarprods[1] += hh[i-2] * hh[i+(ldh*2)];         
      scalarprods[4] += hh[(ldh*1)+i-2] * hh[i+(ldh*3)]; 
      scalarprods[8] += hh[(ldh*2)+i-2] * hh[i+(ldh*4)]; 
      scalarprods[13] += hh[(ldh*3)+i-2] * hh[i+(ldh*5)];

      scalarprods[3] += hh[i-3] * hh[i+(ldh*3)];         
      scalarprods[7] += hh[(ldh)+i-3] * hh[i+(ldh*4)];   
      scalarprods[12] += hh[(ldh*2)+i-3] * hh[i+(ldh*5)];

      scalarprods[6] += hh[i-4] * hh[i+(ldh*4)];         
      scalarprods[11] += hh[(ldh)+i-4] * hh[i+(ldh*5)];  

      scalarprods[10] += hh[i-5] * hh[i+(ldh*5)];        
#endif /* BLOCK6 */
1007
1008
1009
1010

    }

  // Production level kernel calls with padding
1011
1012
#ifdef BLOCK2

1013
#if  VEC_SET == 128 || VEC_SET == 1281
1014
#ifdef DOUBLE_PRECISION_REAL
1015
1016
1017
#define STEP_SIZE 12
#define ROW_LENGTH 12
#define UPPER_BOUND 10
1018
1019
#endif
#ifdef SINGLE_PRECISION_REAL
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
#define STEP_SIZE 24
#define ROW_LENGTH 24
#define UPPER_BOUND 20
#endif
#endif /*  VEC_SET == 128 || VEC_SET == 1281 */

#if  VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#define STEP_SIZE 24
#define ROW_LENGTH 24
#define UPPER_BOUND 20
#endif
#ifdef SINGLE_PRECISION_REAL
#define STEP_SIZE 48
#define ROW_LENGTH 48
#define UPPER_BOUND 40
#endif
1037
#endif /*  VEC_SET == 256 */
1038

1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
#if  VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#define STEP_SIZE 32
#define ROW_LENGTH 32
#define UPPER_BOUND 24
#endif
#ifdef SINGLE_PRECISION_REAL
#define STEP_SIZE 64
#define ROW_LENGTH 64
#define UPPER_BOUND 48
#endif
#endif /*  VEC_SET == 512 */


1053
  for (i = 0; i < nq - UPPER_BOUND; i+= STEP_SIZE )
1054
    {
1055
1056
      CONCAT_6ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_2hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, s);
      worked_on += ROW_LENGTH;
1057
1058
1059
1060
1061
1062
1063
    }

  if (nq == i)
    {
      return;
    }

1064
1065
#undef ROW_LENGTH
#if  VEC_SET == 128 || VEC_SET == 1281
1066
#ifdef DOUBLE_PRECISION_REAL
1067
#define ROW_LENGTH 10
1068
1069
#endif
#ifdef SINGLE_PRECISION_REAL
1070
#define ROW_LENGTH 20
1071
#endif
1072
#endif /*  VEC_SET == 128 || VEC_SET == 1281 */
1073

1074
#if  VEC_SET == 256
1075
#ifdef DOUBLE_PRECISION_REAL
1076
#define ROW_LENGTH 20
1077
1078
#endif
#ifdef SINGLE_PRECISION_REAL
1079
#define ROW_LENGTH 40
1080
#endif
1081
#endif /* VEC_SET == 256 */
1082

1083
1084
1085
1086
1087
1088
1089
1090
1091
#if  VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 24
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 48
#endif
#endif /* VEC_SET == 512 */

1092
  if (nq-i == ROW_LENGTH)
1093
    {
1094
1095
      CONCAT_6ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_2hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, s);
      worked_on += ROW_LENGTH;
1096
    }
1097
1098
1099
1100
1101

#undef ROW_LENGTH
#if  VEC_SET == 128 || VEC_SET == 1281
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 8
1102
#endif
1103
1104
1105
1106
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 16
#endif
#endif /*  VEC_SET == 128 || VEC_SET == 1281 */
1107

1108
1109
1110
1111
#if  VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 16
#endif
1112
#ifdef SINGLE_PRECISION_REAL
1113
1114
1115
1116
#define ROW_LENGTH 32
#endif
#endif /* VEC_SET == 256 */

1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
#if  VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 16
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 32
#endif
#endif /* VEC_SET == 512 */


1127
  if (nq-i == ROW_LENGTH)
1128
    {
1129
1130
      CONCAT_6ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_2hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, s);
      worked_on += ROW_LENGTH;
1131
    }
1132
1133
1134
1135
1136

#undef ROW_LENGTH
#if  VEC_SET == 128 || VEC_SET == 1281
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 6
1137
#endif
1138
1139
1140
1141
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 12
#endif
#endif /*  VEC_SET == 128 || VEC_SET == 1281 */
1142

1143
#if  VEC_SET == 256
1144
#ifdef DOUBLE_PRECISION_REAL
1145
1146
1147
1148
1149
1150
1151
#define ROW_LENGTH 12
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 24
#endif
#endif /* VEC_SET == 256 */

1152
1153
1154
1155
1156
1157
1158
1159
#if  VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 8
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 16
#endif
#endif /* VEC_SET == 512 */
1160
  if (nq-i == ROW_LENGTH)
1161
    {
1162
1163
      CONCAT_6ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_2hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, s);
      worked_on += ROW_LENGTH;
1164
1165
    }

1166
1167
#if VEC_SET == 128 || VEC_SET == 1281 || VEC_SET == 256

1168
1169
1170
1171
1172
#undef ROW_LENGTH
#if  VEC_SET == 128 || VEC_SET == 1281
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 4
#endif
1173
#ifdef SINGLE_PRECISION_REAL
1174
#define ROW_LENGTH 8
1175
#endif
1176
#endif /*  VEC_SET == 128 || VEC_SET == 1281 */
1177

1178
#if  VEC_SET == 256
1179
#ifdef DOUBLE_PRECISION_REAL
1180
1181
1182
1183
1184
1185
1186
1187
1188
#define ROW_LENGTH 8
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 16
#endif
#endif /* VEC_SET == 256 */


  if (nq-i == ROW_LENGTH)
1189
    {
1190
1191
      CONCAT_6ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_2hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, s);
      worked_on += ROW_LENGTH;
1192
    }
1193
1194
1195
1196
1197
1198
1199
1200

#undef ROW_LENGTH
#if  VEC_SET == 128 || VEC_SET == 1281
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 2
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 4
1201
#endif
1202
#endif /*  VEC_SET == 128 || VEC_SET == 1281 */
1203

1204
1205
1206
1207
#if  VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 4
#endif
1208
#ifdef SINGLE_PRECISION_REAL
1209
1210
1211
1212
1213
#define ROW_LENGTH 8
#endif
#endif /* VEC_SET == 256 */

  if (nq-i == ROW_LENGTH)
1214
    {
1215
1216
      CONCAT_6ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_2hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, s);
      worked_on += ROW_LENGTH;
1217
1218
    }

1219
1220
#endif /* VEC_SET == 128 || VEC_SET == 1281 || VEC_SET == 256 */

1221
1222
1223
#endif /* BLOCK2 */

#ifdef BLOCK4
1224
1225
1226
1227


#undef ROW_LENGTH
#if  VEC_SET == 128 || VEC_SET == 1281
1228
#ifdef DOUBLE_PRECISION_REAL
1229
1230
1231
1232
1233
1234
1235
1236
#define ROW_LENGTH 6
#define STEP_SIZE 6
#define UPPER_BOUND 4
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 12
#define STEP_SIZE 12
#define UPPER_BOUND 8
1237
#endif
1238
#endif /*  VEC_SET == 128 || VEC_SET == 1281 */
1239

1240
1241
1242
1243
1244
1245
#if  VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 12
#define STEP_SIZE 12
#define UPPER_BOUND 8
#endif
1246
#ifdef SINGLE_PRECISION_REAL
1247
1248
1249
1250
1251
1252
#define ROW_LENGTH 24
#define STEP_SIZE 24
#define UPPER_BOUND 16
#endif
#endif /* VEC_SET == 256 */

1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
#if  VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 32
#define STEP_SIZE 32
#define UPPER_BOUND 24
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 64
#define STEP_SIZE 64
#define UPPER_BOUND 48
#endif
#endif /* VEC_SET == 512 */
1265
  for (i = 0; i < nq - UPPER_BOUND; i+= STEP_SIZE )
1266
    {
1267
1268
      CONCAT_6ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_4hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, s_1_2, s_1_3, s_2_3, s_1_4, s_2_4, s_3_4);
      worked_on += ROW_LENGTH;
1269
1270
1271
1272
1273
1274
1275
    }

  if (nq == i)
    {
      return;
    }

1276
1277
1278

#undef ROW_LENGTH
#if  VEC_SET == 128 || VEC_SET == 1281
1279
#ifdef DOUBLE_PRECISION_REAL
1280
1281
1282
1283
#define ROW_LENGTH 4
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 8
1284
#endif
1285
#endif /*  VEC_SET == 128 || VEC_SET == 1281 */
1286

1287
1288
1289
1290
#if  VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 8
#endif
1291
#ifdef SINGLE_PRECISION_REAL
1292
1293
1294
1295
#define ROW_LENGTH 16
#endif
#endif /* VEC_SET == 256 */

1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
#if  VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 24
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 48
#endif
#endif /* VEC_SET == 512 */


1306
  if (nq-i == ROW_LENGTH )
1307
    {
1308
1309
      CONCAT_6ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_4hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, s_1_2, s_1_3, s_2_3, s_1_4, s_2_4, s_3_4);
      worked_on += ROW_LENGTH;
1310
1311
    }

1312
1313
#undef ROW_LENGTH
#if  VEC_SET == 128 || VEC_SET == 1281
1314
#ifdef DOUBLE_PRECISION_REAL
1315
#define ROW_LENGTH 2
1316
#endif
1317
1318
1319
1320
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 4
#endif
#endif /*  VEC_SET == 128 || VEC_SET == 1281 */
1321

1322
1323
1324
1325
#if  VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 4
#endif
1326
#ifdef SINGLE_PRECISION_REAL
1327
#define ROW_LENGTH 8
1328
#endif
1329
1330
#endif /* VEC_SET == 256 */

1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
#if  VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 16
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 32
#endif
#endif /* VEC_SET == 512 */

   if (nq-i == ROW_LENGTH )
     {
       CONCAT_6ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_4hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, s_1_2, s_1_3, s_2_3, s_1_4, s_2_4, s_3_4);
       worked_on += ROW_LENGTH;
     }

#if VEC_SET == 512

#if  VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 8
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 16
#endif
#endif /* VEC_SET == 512 */

1357
1358
1359
1360
1361
   if (nq-i == ROW_LENGTH )
     {
       CONCAT_6ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_4hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, s_1_2, s_1_3, s_2_3, s_1_4, s_2_4, s_3_4);
       worked_on += ROW_LENGTH;
     }
1362

1363
1364
#endif /* VEC_SET == 512 */

1365
1366
#endif /* BLOCK4 */

1367
#ifdef BLOCK6
1368
1369
1370

#undef ROW_LENGTH
#if  VEC_SET == 128 || VEC_SET == 1281
1371
#ifdef DOUBLE_PRECISION_REAL
1372
1373
1374
#define ROW_LENGTH 4
#define STEP_SIZE 4
#define UPPER_BOUND 2
1375
1376
#endif
#ifdef SINGLE_PRECISION_REAL
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
#define ROW_LENGTH 8
#define STEP_SIZE 8
#define UPPER_BOUND 4
#endif
#endif /*  VEC_SET == 128 || VEC_SET == 1281 */

#if  VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 8
#define STEP_SIZE 8
#define UPPER_BOUND 4
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 16
#define STEP_SIZE 16
#define UPPER_BOUND 8
1393
#endif
1394
1395
1396
1397
1398
1399
1400
#endif /* VEC_SET == 256 */

  for (i = 0; i < nq - UPPER_BOUND; i+= STEP_SIZE)
    { 
      CONCAT_6ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_6hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, scalarprods);
      worked_on += ROW_LENGTH;
    }
1401
1402
1403
1404
    if (nq == i)
      {
        return;
      }
1405
1406
1407

#undef ROW_LENGTH
#if  VEC_SET == 128 || VEC_SET == 1281
1408
#ifdef DOUBLE_PRECISION_REAL
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
#define ROW_LENGTH 2
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 4
#endif
#endif /*  VEC_SET == 128 || VEC_SET == 1281 */

#if  VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 4
1419
1420
#endif
#ifdef SINGLE_PRECISION_REAL
1421
1422
1423
1424
1425
#define ROW_LENGTH 8
#endif
#endif /* VEC_SET == 256 */

    if (nq -i == ROW_LENGTH )
1426
      {
1427
1428
        CONCAT_6ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_6hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, scalarprods);
        worked_on += ROW_LENGTH;
1429
1430
1431
1432
      }
  
#endif /* BLOCK6 */

1433
1434
1435
#ifdef WITH_DEBUG
  if (worked_on != nq)
    {
1436
      printf("Error in real SIMD_SET BLOCK BLOCK kernel %d %d\n", worked_on, nq);
1437
1438
1439
1440
1441
      abort();
    }
#endif
}

1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
#undef ROW_LENGTH
#if  VEC_SET == 128 || VEC_SET == 1281
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 12
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 24
#endif
#endif /*  VEC_SET == 128 || VEC_SET == 1281 */

#if  VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 24
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 48
#endif
#endif /* VEC_SET == 256 */
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469

#if  VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
#define ROW_LENGTH 48
#endif
#ifdef SINGLE_PRECISION_REAL
#define ROW_LENGTH 96
#endif
#endif /* VEC_SET == 512 */

1470
1471
/*
 * Unrolled kernel that computes
1472
 * ROW_LENGTH rows of Q simultaneously, a
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
 * matrix Vector product with two householder
 */
#ifdef BLOCK2
/*
 * vectors + a rank 2 update is performed
 */
#endif
#ifdef BLOCK4
/*
 * vectors + a rank 1 update is performed
 */
#endif
1485
__forceinline void CONCAT_8ARGS(hh_trafo_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int nb, int ldq, int ldh,
1486
#ifdef BLOCK2
1487
               DATA_TYPE s)
1488
1489
#endif
#ifdef BLOCK4
1490
1491
1492
1493
               DATA_TYPE s_1_2, DATA_TYPE s_1_3, DATA_TYPE s_2_3, DATA_TYPE s_1_4, DATA_TYPE s_2_4, DATA_TYPE s_3_4)
#endif 
#ifdef BLOCK6
               DATA_TYPE_PTR scalarprods)
1494
1495
1496
1497
1498
1499
1500
1501
#endif
  {
#ifdef BLOCK2
    /////////////////////////////////////////////////////
    // Matrix Vector Multiplication, Q [10 x nb+1] * hh
    // hh contains two householder vectors, with offset 1
    /////////////////////////////////////////////////////
#endif
1502
#if defined(BLOCK4) || defined(BLOCK6)
1503
1504
1505
1506
1507
1508
1509
1510
1511
    /////////////////////////////////////////////////////
    // Matrix Vector Multiplication, Q [10 x nb+3] * hh
    // hh contains four householder vectors
    /////////////////////////////////////////////////////
#endif

    int i;

#ifdef BLOCK2
1512
#if VEC_SET == 128
1513
1514
    // Needed bit mask for floating point sign flip
#ifdef DOUBLE_PRECISION_REAL
1515
1516
1517
1518
1519
    __SIMD_DATATYPE sign = (__SIMD_DATATYPE)_mm_set1_epi64x(0x8000000000000000LL);
#endif
#ifdef SINGLE_PRECISION_REAL
    __SIMD_DATATYPE sign = _mm_castsi128_ps(_mm_set_epi32(0x80000000, 0x80000000, 0x80000000, 0x80000000));
#endif
1520
#endif /* VEC_SET == 128 */
1521
1522
1523
1524

#if  VEC_SET == 256
#ifdef DOUBLE_PRECISION_REAL
        __SIMD_DATATYPE sign = (__SIMD_DATATYPE)_mm256_set1_epi64x(0x8000000000000000);
1525
1526
#endif
#ifdef SINGLE_PRECISION_REAL
1527
1528
1529
1530
        __SIMD_DATATYPE sign = (__SIMD_DATATYPE)_mm256_set1_epi32(0x80000000);
#endif
#endif /* VEC_SET == 256 */

1531
1532
#if  VEC_SET == 512
#ifdef DOUBLE_PRECISION_REAL
1533
        __SIMD_DATATYPE sign = (__SIMD_DATATYPE)_mm512_set1_epi64(0x8000000000000000);
1534
1535
1536
1537
1538
1539
#endif
#ifdef SINGLE_PRECISION_REAL
        __SIMD_DATATYPE sign = (__SIMD_DATATYPE)_mm512_set1_epi32(0x80000000);
#endif
#endif /* VEC_SET == 512 */

1540
1541
1542
1543
1544
1545
1546
    __SIMD_DATATYPE x1 = _SIMD_LOAD(&q[ldq]);
    __SIMD_DATATYPE x2 = _SIMD_LOAD(&q[ldq+offset]);
    __SIMD_DATATYPE x3 = _SIMD_LOAD(&q[ldq+2*offset]);
    __SIMD_DATATYPE x4 = _SIMD_LOAD(&q[ldq+3*offset]);
    __SIMD_DATATYPE x5 = _SIMD_LOAD(&q[ldq+4*offset]);
    __SIMD_DATATYPE x6 = _SIMD_LOAD(&q[ldq+5*offset]);

1547
1548
#if VEC_SET == 128 || VEC_SET == 512
    __SIMD_DATATYPE h1 = _SIMD_SET1(hh[ldh+1]);
1549
1550
#endif
#if VEC_SET == 1281
1551
    __SIMD_DATATYPE h1 = _SIMD_SET(hh[ldh+1], hh[ldh+1]);
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
#endif
#if VEC_SET == 256
    __SIMD_DATATYPE h1 = _SIMD_BROADCAST(&hh[ldh+1]);
#endif
 
    __SIMD_DATATYPE h2;
#ifdef __ELPA_USE_FMA__
    __SIMD_DATATYPE q1 = _SIMD_LOAD(q);
    __SIMD_DATATYPE y1 = _SIMD_FMA(x1, h1, q1);
    __SIMD_DATATYPE q2 = _SIMD_LOAD(&q[offset]);
    __SIMD_DATATYPE y2 = _SIMD_FMA(x2, h1, q2);
    __SIMD_DATATYPE q3 = _SIMD_LOAD(&q[2*offset]);
    __SIMD_DATATYPE y3 = _SIMD_FMA(x3, h1, q3);
    __SIMD_DATATYPE q4 = _SIMD_LOAD(&q[3*offset]);
    __SIMD_DATATYPE y4 = _SIMD_FMA(x4, h1, q4);
    __SIMD_DATATYPE q5 = _SIMD_LOAD(&q[4*offset]);
    __SIMD_DATATYPE y5 = _SIMD_FMA(x5, h1, q5);
    __SIMD_DATATYPE q6 = _SIMD_LOAD(&q[5*offset]);
    __SIMD_DATATYPE y6 = _SIMD_FMA(x6, h1, q6);
#else
    __SIMD_DATATYPE q1 = _SIMD_LOAD(q);
    __SIMD_DATATYPE y1 = _SIMD_ADD(q1, _SIMD_MUL(x1, h1));
    __SIMD_DATATYPE q2 = _SIMD_LOAD(&q[offset]);
    __SIMD_DATATYPE y2 = _SIMD_ADD(q2, _SIMD_MUL(x2, h1));
    __SIMD_DATATYPE q3 = _SIMD_LOAD(&q[2*offset]);
    __SIMD_DATATYPE y3 = _SIMD_ADD(q3, _SIMD_MUL(x3, h1));
    __SIMD_DATATYPE q4 = _SIMD_LOAD(&q[3*offset]);
    __SIMD_DATATYPE y4 = _SIMD_ADD(q4, _SIMD_MUL(x4, h1));
    __SIMD_DATATYPE q5 = _SIMD_LOAD(&q[4*offset]);
    __SIMD_DATATYPE y5 = _SIMD_ADD(q5, _SIMD_MUL(x5, h1));
    __SIMD_DATATYPE q6 = _SIMD_LOAD(&q[5*offset]);
    __SIMD_DATATYPE y6 = _SIMD_ADD(q6, _SIMD_MUL(x6, h1));
#endif
1585
1586
1587
#endif /* BLOCK2 */

#ifdef BLOCK4
1588
1589
1590
1591
1592
    __SIMD_DATATYPE a1_1 = _SIMD_LOAD(&q[ldq*3]);
    __SIMD_DATATYPE a2_1 = _SIMD_LOAD(&q[ldq*2]);
    __SIMD_DATATYPE a3_1 = _SIMD_LOAD(&q[ldq]);  
    __SIMD_DATATYPE a4_1 = _SIMD_LOAD(&q[0]);    

1593
#if VEC_SET == 128 || VEC_SET == 512
1594
1595
1596
1597
1598
1599
    __SIMD_DATATYPE h_2_1 = _SIMD_SET1(hh[ldh+1]);    
    __SIMD_DATATYPE h_3_2 = _SIMD_SET1(hh[(ldh*2)+1]);
    __SIMD_DATATYPE h_3_1 = _SIMD_SET1(hh[(ldh*2)+2]);
    __SIMD_DATATYPE h_4_3 = _SIMD_SET1(hh[(ldh*3)+1]);
    __SIMD_DATATYPE h_4_2 = _SIMD_SET1(hh[(ldh*3)+2]);
    __SIMD_DATATYPE h_4_1 = _SIMD_SET1(hh[(ldh*3)+3]);
1600
1601
1602
#endif

#if VEC_SET == 1281
1603
1604
1605
1606
1607
1608
    __SIMD_DATATYPE h_2_1 = _SIMD_SET(hh[ldh+1], hh[ldh+1]);
    __SIMD_DATATYPE h_3_2 = _SIMD_SET(hh[(ldh*2)+1], hh[(ldh*2)+1]);
    __SIMD_DATATYPE h_3_1 = _SIMD_SET(hh[(ldh*2)+2], hh[(ldh*2)+2]);
    __SIMD_DATATYPE h_4_3 = _SIMD_SET(hh[(ldh*3)+1], hh[(ldh*3)+1]);
    __SIMD_DATATYPE h_4_2 = _SIMD_SET(hh[(ldh*3)+2], hh[(ldh*3)+2]);
    __SIMD_DATATYPE h_4_1 = _SIMD_SET(hh[(ldh*3)+3], hh[(ldh*3)+3]);
1609
1610
#endif

1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
#if VEC_SET == 256
    __SIMD_DATATYPE h_2_1 = _SIMD_BROADCAST(&hh[ldh+1]);
    __SIMD_DATATYPE h_3_2 = _SIMD_BROADCAST(&hh[(ldh*2)+1]);
    __SIMD_DATATYPE h_3_1 = _SIMD_BROADCAST(&hh[(ldh*2)+2]);
    __SIMD_DATATYPE h_4_3 = _SIMD_BROADCAST(&hh[(ldh*3)+1]);
    __SIMD_DATATYPE h_4_2 = _SIMD_BROADCAST(&hh[(ldh*3)+2]);
    __SIMD_DATATYPE h_4_1 = _SIMD_BROADCAST(&hh[(ldh*3)+3]);
#endif

#ifdef __ELPA_USE_FMA__
    register __SIMD_DATATYPE w1 = _SIMD_FMA(a3_1, h_4_3, a4_1);
    w1 = _SIMD_FMA(a2_1, h_4_2, w1);
    w1 = _SIMD_FMA(a1_1, h_4_1, w1);
    register __SIMD_DATATYPE z1 = _SIMD_FMA(a2_1, h_3_2, a3_1);
    z1 = _SIMD_FMA(a1_1, h_3_1, z1);
    register __SIMD_DATATYPE y1 = _SIMD_FMA(a1_1, h_2_1, a2_1);
    register __SIMD_DATATYPE x1 = a1_1;
#else
1629
1630
1631
1632
1633
1634
1635
    register __SIMD_DATATYPE w1 = _SIMD_ADD(a4_1, _SIMD_MUL(a3_1, h_4_3));
    w1 = _SIMD_ADD(w1, _SIMD_MUL(a2_1, h_4_2));                          
    w1 = _SIMD_ADD(w1, _SIMD_MUL(a1_1, h_4_1));                          
    register __SIMD_DATATYPE z1 = _SIMD_ADD(a3_1, _SIMD_MUL(a2_1, h_3_2));
    z1 = _SIMD_ADD(z1, _SIMD_MUL(a1_1, h_3_1));                          
    register __SIMD_DATATYPE y1 = _SIMD_ADD(a2_1, _SIMD_MUL(a1_1, h_2_1));
    register __SIMD_DATATYPE x1 = a1_1;
1636
#endif /* __ELPA_USE_FMA__ */
1637
1638
1639
1640
1641
1642

    __SIMD_DATATYPE a1_2 = _SIMD_LOAD(&q[(ldq*3)+offset]);                  
    __SIMD_DATATYPE a2_2 = _SIMD_LOAD(&q[(ldq*2)+offset]);
    __SIMD_DATATYPE a3_2 = _SIMD_LOAD(&q[ldq+offset]);
    __SIMD_DATATYPE a4_2 = _SIMD_LOAD(&q[0+offset]);

1643
1644
1645
1646
1647
1648
1649
1650
1651
#ifdef __ELPA_USE_FMA__
    register __SIMD_DATATYPE w2 = _SIMD_FMA(a3_2, h_4_3, a4_2);
    w2 = _SIMD_FMA(a2_2, h_4_2, w2);
    w2 = _SIMD_FMA(a1_2, h_4_1, w2);
    register __SIMD_DATATYPE z2 = _SIMD_FMA(a2_2, h_3_2, a3_2);
    z2 = _SIMD_FMA(a1_2, h_3_1, z2);
    register __SIMD_DATATYPE y2 = _SIMD_FMA(a1_2, h_2_1, a2_2);
    register __SIMD_DATATYPE x2 = a1_2;
#else
1652
1653
1654
1655
1656
1657
1658
    register __SIMD_DATATYPE w2 = _SIMD_ADD(a4_2, _SIMD_MUL(a3_2, h_4_3));
    w2 = _SIMD_ADD(w2, _SIMD_MUL(a2_2, h_4_2));
    w2 = _SIMD_ADD(w2, _SIMD_MUL(a1_2, h_4_1));
    register __SIMD_DATATYPE z2 = _SIMD_ADD(a3_2, _SIMD_MUL(a2_2, h_3_2));
    z2 = _SIMD_ADD(z2, _SIMD_MUL(a1_2, h_3_1));
    register __SIMD_DATATYPE y2 = _SIMD_ADD(a2_2, _SIMD_MUL(a1_2, h_2_1));
    register __SIMD_DATATYPE x2 = a1_2;
1659
#endif /* __ELPA_USE_FMA__ */
1660
1661
1662
1663
1664
1665

    __SIMD_DATATYPE a1_3 = _SIMD_LOAD(&q[(ldq*3)+2*offset]);
    __SIMD_DATATYPE a2_3 = _SIMD_LOAD(&q[(ldq*2)+2*offset]);
    __SIMD_DATATYPE a3_3 = _SIMD_LOAD(&q[ldq+2*offset]);
    __SIMD_DATATYPE a4_3 = _SIMD_LOAD(&q[0+2*offset]);

1666
1667
1668
1669
1670
1671
1672
1673
1674
#ifdef __ELPA_USE_FMA__
    register __SIMD_DATATYPE w3 = _SIMD_FMA(a3_3, h_4_3, a4_3);
    w3 = _SIMD_FMA(a2_3, h_4_2, w3);
    w3 = _SIMD_FMA(a1_3, h_4_1, w3);
    register __SIMD_DATATYPE z3 = _SIMD_FMA(a2_3, h_3_2, a3_3);
    z3 = _SIMD_FMA(a1_3, h_3_1, z3);
    register __SIMD_DATATYPE y3 = _SIMD_FMA(a1_3, h_2_1, a2_3);
    register __SIMD_DATATYPE x3 = a1_3;
#else
1675
1676
1677
1678
1679
1680
1681
    register __SIMD_DATATYPE w3 = _SIMD_ADD(a4_3, _SIMD_MUL(a3_3, h_4_3));
    w3 = _SIMD_ADD(w3, _SIMD_MUL(a2_3, h_4_2));
    w3 = _SIMD_ADD(w3, _SIMD_MUL(a1_3, h_4_1));
    register __SIMD_DATATYPE z3 = _SIMD_ADD(a3_3, _SIMD_MUL(a2_3, h_3_2));
    z3 = _SIMD_ADD(z3, _SIMD_MUL(a1_3, h_3_1));
    register __SIMD_DATATYPE y3 = _SIMD_ADD(a2_3, _SIMD_MUL(a1_3, h_2_1));
    register __SIMD_DATATYPE x3 = a1_3;
1682
#endif /* __ELPA_USE_FMA__ */
1683
1684
1685
1686
1687
1688

    __SIMD_DATATYPE a1_4 = _SIMD_LOAD(&q[(ldq*3)+3*offset]);
    __SIMD_DATATYPE a2_4 = _SIMD_LOAD(&q[(ldq*2)+3*offset]);
    __SIMD_DATATYPE a3_4 = _SIMD_LOAD(&q[ldq+3*offset]);
    __SIMD_DATATYPE a4_4 = _SIMD_LOAD(&q[0+3*offset]);

1689
1690
1691
1692
1693
1694
1695
1696
1697
#ifdef __ELPA_USE_FMA__
    register __SIMD_DATATYPE w4 = _SIMD_FMA(a3_4, h_4_3, a4_4);
    w4 = _SIMD_FMA(a2_4, h_4_2, w4);
    w4 = _SIMD_FMA(a1_4, h_4_1, w4);
    register __SIMD_DATATYPE z4 = _SIMD_FMA(a2_4, h_3_2, a3_4);
    z4 = _SIMD_FMA(a1_4, h_3_1, z4);
    register __SIMD_DATATYPE y4 = _SIMD_FMA(a1_4, h_2_1, a2_4);
    register __SIMD_DATATYPE x4 = a1_4;
#else
1698
1699
1700
1701
1702
1703
1704
    register __SIMD_DATATYPE w4 = _SIMD_ADD(a4_4, _SIMD_MUL(a3_4, h_4_3));
    w4 = _SIMD_ADD(w4, _SIMD_MUL(a2_4, h_4_2));
    w4 = _SIMD_ADD(w4, _SIMD_MUL(a1_4, h_4_1));
    register __SIMD_DATATYPE z4 = _SIMD_ADD(a3_4, _SIMD_MUL(a2_4, h_3_2));
    z4 = _SIMD_ADD(z4, _SIMD_MUL(a1_4, h_3_1));
    register __SIMD_DATATYPE y4 = _SIMD_ADD(a2_4, _SIMD_MUL(a1_4, h_2_1));
    register __SIMD_DATATYPE x4 = a1_4;
1705
#endif /* __ELPA_USE_FMA__ */
1706
1707
1708
1709
1710
1711

    __SIMD_DATATYPE a1_5 = _SIMD_LOAD(&q[(ldq*3)+4*offset]);
    __SIMD_DATATYPE a2_5 = _SIMD_LOAD(&q[(ldq*2)+4*offset]);
    __SIMD_DATATYPE a3_5 = _SIMD_LOAD(&q[ldq+4*offset]);
    __SIMD_DATATYPE a4_5 = _SIMD_LOAD(&q[0+4*offset]);

1712
1713
1714
1715
1716
1717
1718
1719
1720
#ifdef __ELPA_USE_FMA__
    register __SIMD_DATATYPE w5 = _SIMD_FMA(a3_5, h_4_3, a4_5);
    w5 = _SIMD_FMA(a2_5, h_4_2, w5);
    w5 = _SIMD_FMA(a1_5, h_4_1, w5);
    register __SIMD_DATATYPE z5 = _SIMD_FMA(a2_5, h_3_2, a3_5);
    z5 = _SIMD_FMA(a1_5, h_3_1, z5);
    register __SIMD_DATATYPE y5 = _SIMD_FMA(a1_5, h_2_1, a2_5);
    register __SIMD_DATATYPE x5 = a1_5;
#else
1721
1722
1723
1724
1725
1726
1727
    register __SIMD_DATATYPE w5 = _SIMD_ADD(a4_5, _SIMD_MUL(a3_5, h_4_3));
    w5 = _SIMD_ADD(w5, _SIMD_MUL(a2_5, h_4_2));
    w5 = _SIMD_ADD(w5, _SIMD_MUL(a1_5, h_4_1));
    register __SIMD_DATATYPE z5 = _SIMD_ADD(a3_5, _SIMD_MUL(a2_5, h_3_2));
    z5 = _SIMD_ADD(z5, _SIMD_MUL(a1_5, h_3_1));
    register __SIMD_DATATYPE y5 = _SIMD_ADD(a2_5, _SIMD_MUL(a1_5, h_2_1));
    register __SIMD_DATATYPE x5 = a1_5;
1728
#endif /* __ELPA_USE_FMA__ */
1729
1730
1731
1732
1733
1734

    __SIMD_DATATYPE a1_6 = _SIMD_LOAD(&q[(ldq*3)+5*offset]);
    __SIMD_DATATYPE a2_6 = _SIMD_LOAD(&q[(ldq*2)+5*offset]);
    __SIMD_DATATYPE a3_6 = _SIMD_LOAD(&q[ldq+5*offset]);
    __SIMD_DATATYPE a4_6 = _SIMD_LOAD(&q[0+5*offset]);

1735
1736
1737
1738
1739
1740
1741
1742
1743
#ifdef __ELPA_USE_FMA__
    register __SIMD_DATATYPE w6 = _SIMD_FMA(a3_6, h_4_3, a4_6);
    w6 = _SIMD_FMA(a2_6, h_4_2, w6);
    w6 = _SIMD_FMA(a1_6, h_4_1, w6);
    register __SIMD_DATATYPE z6 = _SIMD_FMA(a2_6, h_3_2, a3_6);
    z6 = _SIMD_FMA(a1_6, h_3_1, z6);
    register __SIMD_DATATYPE y6 = _SIMD_FMA(a1_6, h_2_1, a2_6);
    register __SIMD_DATATYPE x6 = a1_6;
#else
1744
1745
1746
1747
1748
1749
1750
    register __SIMD_DATATYPE w6 = _SIMD_ADD(a4_6, _SIMD_MUL(a3_6, h_4_3));
    w6 = _SIMD_ADD(w6, _SIMD_MUL(a2_6, h_4_2));
    w6 = _SIMD_ADD(w6, _SIMD_MUL(a1_6, h_4_1));
    register __SIMD_DATATYPE z6 = _SIMD_ADD(a3_6, _SIMD_MUL(a2_6, h_3_2));
    z6 = _SIMD_ADD(z6, _SIMD_MUL(a1_6, h_3_1));
    register __SIMD_DATATYPE y6 = _SIMD_ADD(a2_6, _SIMD_MUL(a1_6, h_2_1));
    register __SIMD_DATATYPE x6 = a1_6;
1751
#endif /* __ELPA_USE_FMA__ */
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763

    __SIMD_DATATYPE q1;
    __SIMD_DATATYPE q2;
    __SIMD_DATATYPE q3;
    __SIMD_DATATYPE q4;
    __SIMD_DATATYPE q5;
    __SIMD_DATATYPE q6;

    __SIMD_DATATYPE h1;
    __SIMD_DATATYPE h2;
    __SIMD_DATATYPE h3;
    __SIMD_DATATYPE h4;
1764
1765
#endif /* BLOCK4 */

1766
1767
#ifdef BLOCK6
    
1768
1769
1770
1771
1772
1773
1774
1775
    __SIMD_DATATYPE a1_1 = _SIMD_LOAD(&q[ldq*5]);
    __SIMD_DATATYPE a2_1 = _SIMD_LOAD(&q[ldq*4]);
    __SIMD_DATATYPE a3_1 = _SIMD_LOAD(&q[ldq*3]);
    __SIMD_DATATYPE a4_1 = _SIMD_LOAD(&q[ldq*2]);
    __SIMD_DATATYPE a5_1 = _SIMD_LOAD(&q[ldq]);  
    __SIMD_DATATYPE a6_1 = _SIMD_LOAD(&q[0]);    

#if VEC_SET == 128
1776
1777
1778
1779
1780
    __SIMD_DATATYPE h_6_5 = _SIMD_SET1(hh[(ldh*5)+1]);
    __SIMD_DATATYPE h_6_4 = _SIMD_SET1(hh[(ldh*5)+2]);
    __SIMD_DATATYPE h_6_3 = _SIMD_SET1(hh[(ldh*5)+3]);
    __SIMD_DATATYPE h_6_2 = _SIMD_SET1(hh[(ldh*5)+4]);
    __SIMD_DATATYPE h_6_1 = _SIMD_SET1(hh[(ldh*5)+5]);
1781
1782
1783
#endif

#if VEC_SET == 1281
1784
1785
1786
1787
1788
    __SIMD_DATATYPE h_6_5 = _SIMD_SET(hh[(ldh*5)+1], hh[(ldh*5)+1]);
    __SIMD_DATATYPE h_6_4 = _SIMD_SET(hh[(ldh*5)+2], hh[(ldh*5)+2]);
    __SIMD_DATATYPE h_6_3 = _SIMD_SET(hh[(ldh*5)+3], hh[(ldh*5)+3]);
    __SIMD_DATATYPE h_6_2 = _SIMD_SET(hh[(ldh*5)+4], hh[(ldh*5)+4]);
    __SIMD_DATATYPE h_6_1 = _SIMD_SET(hh[(ldh*5)+5], hh[(ldh*5)+5]);
1789
1790
#endif

1791
1792
1793
1794
1795
1796
1797