complex_128bit_256bit_512bit_BLOCK_template.c 166 KB
Newer Older
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
//    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
//
//
//    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"

#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

#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

//define instruction set numbers
#define SSE_128 128
70
#define AVX_256 256
71
72
#define NEON_ARCH64_128 1285

73
#if VEC_SET == SSE_128 || VEC_SET == AVX_256 || VEC_SET == 512
74
75
#include <x86intrin.h>
#ifdef BLOCK2
76
#if VEC_SET == SSE_128 
77
78
#include <pmmintrin.h>
#endif
79
#endif
80
81
82

#define __forceinline __attribute__((always_inline))

83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
#endif


#include <complex.h>

#include <stdio.h>
#include <stdlib.h>

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

#ifdef BLOCK1
#define PREFIX single
#define BLOCK 1
#endif

#if VEC_SET == SSE_128
#define SIMD_SET SSE
#endif

105
106
107
108
109
110
#if VEC_SET == AVX_256
#define SIMD_SET AVX_AVX2
#endif

#if VEC_SET == SSE_128

111
112
113
114
115
116
#ifdef DOUBLE_PRECISION_COMPLEX
#define offset 2
#define __SIMD_DATATYPE __m128d
#define _SIMD_LOAD _mm_load_pd
#define _SIMD_LOADU _mm_loadu_pd
#define _SIMD_STORE _mm_store_pd
117
#define _SIMD_STOREU _mm_storeu_pd
118
119
120
121
122
123
#define _SIMD_MUL _mm_mul_pd
#define _SIMD_ADD _mm_add_pd
#define _SIMD_XOR _mm_xor_pd
#define _SIMD_ADDSUB _mm_addsub_pd
#define _SIMD_SHUFFLE _mm_shuffle_pd
#define _SHUFFLE _MM_SHUFFLE2(0,1)
124
125
126

#ifdef __ELPA_USE_FMA__
#define _SIMD_FMSUBADD _mm_maddsub_pd
127
#endif
128
129
#endif /* DOUBLE_PRECISION_COMPLEX */

130
131
132
133
134
135
#ifdef SINGLE_PRECISION_COMPLEX
#define offset 4
#define __SIMD_DATATYPE __m128
#define _SIMD_LOAD _mm_load_ps
#define _SIMD_LOADU _mm_loadu_ps
#define _SIMD_STORE _mm_store_ps
136
#define _SIMD_STOREU _mm_storeu_ps
137
138
139
140
141
142
#define _SIMD_MUL _mm_mul_ps
#define _SIMD_ADD _mm_add_ps
#define _SIMD_XOR _mm_xor_ps
#define _SIMD_ADDSUB _mm_addsub_ps
#define _SIMD_SHUFFLE _mm_shuffle_ps
#define _SHUFFLE 0xb1
143
144
145

#ifdef __ELPA_USE_FMA__
#define _SIMD_FMSUBADD _mm_maddsub_ps
146
147
#endif

148
149
#endif /* SINGLE_PRECISION_COMPLEX */

150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256

#ifdef DOUBLE_PRECISION_COMPLEX
#define offset 4
#define __SIMD_DATATYPE __m256d
#define _SIMD_LOAD _mm256_load_pd
#define _SIMD_LOADU 1
#define _SIMD_STORE _mm256_store_pd
#define _SIMD_STOREU 1
#define _SIMD_MUL _mm256_mul_pd
#define _SIMD_ADD _mm256_add_pd
#define _SIMD_XOR _mm256_xor_pd
#define _SIMD_BROADCAST _mm256_broadcast_sd
165
#define _SIMD_SET1 _mm256_set1_pd
166
167
168
#define _SIMD_ADDSUB _mm256_addsub_pd
#define _SIMD_SHUFFLE _mm256_shuffle_pd
#define _SHUFFLE 0x5
169

170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
#ifdef HAVE_AVX2

#ifdef __FMA4__
#define __ELPA_USE_FMA__
#define _mm256_FMADDSUB_pd(a,b,c) _mm256_maddsub_pd(a,b,c)
#define _mm256_FMSUBADD_pd(a,b,c) _mm256_msubadd_pd(a,b,c)
#endif

#ifdef __AVX2__
#define __ELPA_USE_FMA__
#define _mm256_FMADDSUB_pd(a,b,c) _mm256_fmaddsub_pd(a,b,c)
#define _mm256_FMSUBADD_pd(a,b,c) _mm256_fmsubadd_pd(a,b,c)
#endif

#define _SIMD_FMADDSUB _mm256_FMADDSUB_pd
#define _SIMD_FMSUBADD _mm256_FMSUBADD_pd
Andreas Marek's avatar
Andreas Marek committed
186
#endif /* HAVE_AVX2 */
187
188
189
190
191
192
193
194
195
196
197
198
199
200

#endif /* DOUBLE_PRECISION_COMPLEX */

#ifdef SINGLE_PRECISION_COMPLEX
#define offset 8
#define __SIMD_DATATYPE __m256
#define _SIMD_LOAD _mm256_load_ps
#define _SIMD_LOADU 1
#define _SIMD_STORE _mm256_store_ps
#define _SIMD_STOREU 1
#define _SIMD_MUL _mm256_mul_ps
#define _SIMD_ADD _mm256_add_ps
#define _SIMD_XOR _mm256_xor_ps
#define _SIMD_BROADCAST  _mm256_broadcast_ss
201
#define _SIMD_SET1 _mm256_set1_ps
202
#define _SIMD_ADDSUB _mm256_addsub_ps
203
#define _SIMD_SHUFFLE _mm256_shuffle_ps
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
#define _SHUFFLE 0xb1

#ifdef HAVE_AVX2

#ifdef __FMA4__
#define __ELPA_USE_FMA__
#define _mm256_FMADDSUB_ps(a,b,c) _mm256_maddsub_ps(a,b,c)
#define _mm256_FMSUBADD_ps(a,b,c) _mm256_msubadd_ps(a,b,c)
#endif

#ifdef __AVX2__
#define __ELPA_USE_FMA__
#define _mm256_FMADDSUB_ps(a,b,c) _mm256_fmaddsub_ps(a,b,c)
#define _mm256_FMSUBADD_ps(a,b,c) _mm256_fmsubadd_ps(a,b,c)
#endif

#define _SIMD_FMADDSUB _mm256_FMADDSUB_ps
#define _SIMD_FMSUBADD _mm256_FMSUBADD_ps
Andreas Marek's avatar
Andreas Marek committed
222
#endif /* HAVE_AVX2 */
223
224
225
226
227
228
229

#endif /* SINGLE_PRECISION_COMPLEX */

#endif /* VEC_SET == AVX_256 */



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
#define __forceinline __attribute__((always_inline))

#ifdef HAVE_SSE_INTRINSICS
#undef __AVX__
#endif

#ifdef DOUBLE_PRECISION_COMPLEX
#define WORD_LENGTH double
#define DATA_TYPE double complex
#define DATA_TYPE_PTR double complex*
#define DATA_TYPE_REAL double
#define DATA_TYPE_REAL_PTR double*
#endif

#ifdef SINGLE_PRECISION_COMPLEX
#define WORD_LENGTH single
#define DATA_TYPE float complex
#define DATA_TYPE_PTR float complex*
#define DATA_TYPE_REAL float
#define DATA_TYPE_REAL_PTR float*
#endif


#if VEC_SET  == SSE_128
#ifdef DOUBLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 6
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 12
#endif
#endif /* VEC_SET  == SSE_128 */

264
265
266
267
268
269
270
271
272
273
274
#if VEC_SET  == AVX_256
#ifdef DOUBLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 12
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 24
#endif
#endif /* VEC_SET  == AVX_256 */

275
276
277
278
279
280
281
282
283
//Forward declaration
static __forceinline void CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH)(DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int nb, int ldq 
#ifdef BLOCK1
		                       );
#endif
#ifdef BLOCK2
                                       ,int ldh, DATA_TYPE s);
#endif

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
#if VEC_SET  == SSE_128
#ifdef DOUBLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 5
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 10
#endif
#endif /* VEC_SET  == SSE_128 */


#if VEC_SET  == AVX_256
#ifdef DOUBLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 10
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 20
#endif
#endif /* VEC_SET  == AVX_256 */


static __forceinline void CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH)(DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int nb, int ldq
#ifdef BLOCK1
		                       );
#endif
#ifdef BLOCK2
                                       ,int ldh, DATA_TYPE s);
#endif


317
318
319
320
321
322
323
324
325
326
#if VEC_SET  == SSE_128
#ifdef DOUBLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 4
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 8
#endif
#endif /* VEC_SET  == SSE_128 */
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


#if VEC_SET  == AVX_256
#ifdef DOUBLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 8
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 16
#endif
#endif /* VEC_SET  == AVX_256 */


static __forceinline void CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH)(DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int nb, int ldq
#ifdef BLOCK1
		                       );
#endif
#ifdef BLOCK2
                                       ,int ldh, DATA_TYPE s);
#endif

#if VEC_SET  == SSE_128
#ifdef DOUBLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 3
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 6
#endif
#endif /* VEC_SET  == SSE_128 */


#if VEC_SET  == AVX_256
#ifdef DOUBLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 6
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 12
#endif
#endif /* VEC_SET  == AVX_256 */


373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
static __forceinline void CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH)(DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int nb, int ldq
#ifdef BLOCK1
		                       );
#endif
#ifdef BLOCK2
                                       ,int ldh, DATA_TYPE s);
#endif

#if VEC_SET  == SSE_128
#ifdef DOUBLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 2
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 4
#endif
#endif /* VEC_SET  == SSE_128 */
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

#if VEC_SET  == AVX_256
#ifdef DOUBLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 4
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 8
#endif
#endif /* VEC_SET  == AVX_256 */


static __forceinline void CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH)(DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int nb, int ldq
#ifdef BLOCK1
		                       );
#endif
#ifdef BLOCK2
                                       ,int ldh, DATA_TYPE s);
#endif

#if VEC_SET  == SSE_128
#ifdef DOUBLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 1
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 2
#endif
#endif /* VEC_SET  == SSE_128 */

#if VEC_SET  == AVX_256
#ifdef DOUBLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 2
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#undef ROW_LENGTH 
#define ROW_LENGTH 4
#endif
#endif /* VEC_SET  == AVX_256 */

434
435
436
437
438
439
440
441
static __forceinline void CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH)(DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int nb, int ldq
#ifdef BLOCK1
		                       );
#endif
#ifdef BLOCK2
                                       ,int ldh, DATA_TYPE s);
#endif

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
/*
!f>#ifdef HAVE_SSE_INTRINSICS
!f> interface
!f>   subroutine single_hh_trafo_complex_SSE_1hv_double(q, hh, pnb, pnq, pldq) &
!f>                             bind(C, name="single_hh_trafo_complex_SSE_1hv_double")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int)     :: pnb, pnq, pldq
!f>     ! complex(kind=c_double_complex)     :: q(*)
!f>     type(c_ptr), value                   :: q
!f>     complex(kind=c_double_complex)     :: hh(pnb,2)
!f>   end subroutine
!f> end interface
!f>#endif
*/

/*
!f>#ifdef HAVE_SSE_INTRINSICS
!f> interface
!f>   subroutine single_hh_trafo_complex_SSE_1hv_single(q, hh, pnb, pnq, pldq) &
!f>                             bind(C, name="single_hh_trafo_complex_SSE_1hv_single")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int)     :: pnb, pnq, pldq
!f>     ! complex(kind=c_float_complex)   :: q(*)
!f>     type(c_ptr), value                :: q
!f>     complex(kind=c_float_complex)   :: hh(pnb,2)
!f>   end subroutine
!f> end interface
!f>#endif
*/

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
/*
!f>#if defined(HAVE_AVX) || defined(HAVE_AVX2)
!f> interface
!f>   subroutine single_hh_trafo_complex_AVX_AVX2_1hv_double(q, hh, pnb, pnq, pldq) &
!f>                             bind(C, name="single_hh_trafo_complex_AVX_AVX2_1hv_double")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int)     :: pnb, pnq, pldq
!f>     ! complex(kind=c_double_complex)     :: q(*)
!f>     type(c_ptr), value                   :: q
!f>     complex(kind=c_double_complex)       :: hh(pnb,2)
!f>   end subroutine
!f> end interface
!f>#endif
*/

/*
!f>#if defined(HAVE_AVX) || defined(HAVE_AVX2)
!f> interface
!f>   subroutine single_hh_trafo_complex_AVX_AVX2_1hv_single(q, hh, pnb, pnq, pldq) &
!f>                             bind(C, name="single_hh_trafo_complex_AVX_AVX2_1hv_single")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int)     :: pnb, pnq, pldq
!f>     ! complex(kind=c_float_complex)   :: q(*)
!f>     type(c_ptr), value              :: q
!f>     complex(kind=c_float_complex)   :: hh(pnb,2)
!f>   end subroutine
!f> end interface
!f>#endif
*/

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
/*
!f>#ifdef HAVE_SSE_INTRINSICS
!f> interface
!f>   subroutine double_hh_trafo_complex_SSE_2hv_double(q, hh, pnb, pnq, pldq, pldh) &
!f>                             bind(C, name="double_hh_trafo_complex_SSE_2hv_double")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int)     :: pnb, pnq, pldq, pldh
!f>     ! complex(kind=c_double_complex)     :: q(*)
!f>     type(c_ptr), value                   :: q
!f>     complex(kind=c_double_complex)     :: hh(pnb,2)
!f>   end subroutine
!f> end interface
!f>#endif
*/

/*
!f>#ifdef HAVE_SSE_INTRINSICS
!f> interface
!f>   subroutine double_hh_trafo_complex_SSE_2hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f>                             bind(C, name="double_hh_trafo_complex_SSE_2hv_single")
!f>     use, intrinsic :: iso_c_binding
!f>     integer(kind=c_int)     :: pnb, pnq, pldq, pldh
!f>     ! complex(kind=c_float_complex)   :: q(*)
!f>     type(c_ptr), value                :: q
!f>     complex(kind=c_float_complex)   :: hh(pnb,2)
!f>   end subroutine
!f> end interface
!f>#endif
*/

Andreas Marek's avatar
Andreas Marek committed
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
/*
!f>#if defined(HAVE_AVX) || defined(HAVE_AVX2)
!f> interface
!f>   subroutine double_hh_trafo_complex_AVX_AVX2_2hv_double(q, hh, pnb, pnq, pldq, pldh) &
!f>                                bind(C, name="double_hh_trafo_complex_AVX_AVX2_2hv_double")
!f>        use, intrinsic :: iso_c_binding
!f>        integer(kind=c_int)        :: pnb, pnq, pldq, pldh
!f>        ! complex(kind=c_double_complex)     :: q(*)
!f>        type(c_ptr), value                     :: q
!f>        complex(kind=c_double_complex)           :: hh(pnb,2)
!f>   end subroutine
!f> end interface
!f>#endif
*/

/*
!f>#if defined(HAVE_AVX) || defined(HAVE_AVX2)
!f> interface
!f>   subroutine double_hh_trafo_complex_AVX_AVX2_2hv_single(q, hh, pnb, pnq, pldq, pldh) &
!f>                                bind(C, name="double_hh_trafo_complex_AVX_AVX2_2hv_single")
!f>        use, intrinsic :: iso_c_binding
!f>        integer(kind=c_int)        :: pnb, pnq, pldq, pldh
!f>        ! complex(kind=c_float_complex)   :: q(*)
!f>        type(c_ptr), value                  :: q
!f>        complex(kind=c_float_complex)        :: hh(pnb,2)
!f>   end subroutine
!f> end interface
!f>#endif
*/

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
void CONCAT_7ARGS(PREFIX,_hh_trafo_complex_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int* pnb, int* pnq, int* pldq
#ifdef BLOCK1
		  )
#endif
#ifdef BLOCK2
                  ,int* pldh)
#endif
{

     int i, worked_on;
     int nb = *pnb;
     int nq = *pldq;
     int ldq = *pldq;
#ifdef BLOCK2
     int ldh = *pldh;

     DATA_TYPE s = conj(hh[(ldh)+1])*1.0;

     for (i = 2; i < nb; i++)
     {
             s += hh[i-1] * conj(hh[(i+ldh)]);
     }
#endif

     worked_on = 0;
589

590
591
#ifdef BLOCK1

592
#if VEC_SET == SSE_128
593
594
595
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 6
#define STEP_SIZE 6
596
#define UPPER_BOUND 5
597
598
599
600
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 12
#define STEP_SIZE 12
601
602
603
604
605
606
607
608
609
610
611
612
613
614
#define UPPER_BOUND 10
#endif
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 12
#define STEP_SIZE 12
#define UPPER_BOUND 10
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 24
#define STEP_SIZE 24
#define UPPER_BOUND 20
615
#endif
616
617
#endif /* VEC_SET == AVX_256 */

618
619
620
621
622
623
624
625
626
627
        for (i = 0; i < nq - UPPER_BOUND; i+= STEP_SIZE)
        {

            CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (&q[i], hh, nb, ldq);
	    worked_on += ROW_LENGTH;
        }
        if (nq == i) {
          return;
        }

628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
#if VEC_SET == SSE_128
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 5
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 10
#endif
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 10
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 20
#endif
#endif /* VEC_SET == AVX_256 */

        if (nq-i == ROW_LENGTH)
        {
            CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (&q[i], hh, nb, ldq);
	    worked_on += ROW_LENGTH;
        }


#if VEC_SET == SSE_128
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 4
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 8
#endif
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 8
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 16
#endif
#endif /* VEC_SET == AVX_256 */

        if (nq-i == ROW_LENGTH)
        {
            CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (&q[i], hh, nb, ldq);
	    worked_on += ROW_LENGTH;
        }

#if VEC_SET == SSE_128
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 3
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 6
#endif
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 6
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 12
#endif
#endif /* VEC_SET == AVX_256 */

        if (nq-i == ROW_LENGTH)
        {
            CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (&q[i], hh, nb, ldq);
	    worked_on += ROW_LENGTH;
        }

#if VEC_SET == SSE_128
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 2
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 4
#endif
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
718
719
720
721
722
723
724
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 4
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 8
#endif
725
726
727
#endif /* VEC_SET == AVX_256 */


728
729
730
731
732
733
        if (nq-i == ROW_LENGTH)
        {
            CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (&q[i], hh, nb, ldq);
	    worked_on += ROW_LENGTH;
        }

734
735
736
737
738
739
740
741
742
743
744
#if VEC_SET == SSE_128
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 1
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 2
#endif
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
745
746
747
748
749
750
751
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 2
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 4
#endif
752
#endif /* VEC_SET == AVX_256 */
753
754
755
756
757
        if (nq-i == ROW_LENGTH)
        {
            CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (&q[i], hh, nb, ldq);
	    worked_on += ROW_LENGTH;
        }
758

759
760
761
762
#endif /* BLOCK1 */

#ifdef BLOCK2

763
#if VEC_SET == SSE_128
764
765
766
767
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 4
#define STEP_SIZE 4
768
#define UPPER_BOUND 3
769
770
771
772
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 8
#define STEP_SIZE 8
773
774
775
776
777
778
779
780
781
782
#define UPPER_BOUND 6
#endif
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 8
#define STEP_SIZE 8
#define UPPER_BOUND 6
783
#endif
784
785
786
787
788
789
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 16
#define STEP_SIZE 16
#define UPPER_BOUND 12
#endif
#endif /* VEC_SET == AVX_256 */
790

791
#if VEC_SET == SSE_128
792
793
794
795
796
    for (i = 0; i < nq - UPPER_BOUND; i+=STEP_SIZE)
    {
         CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, s);
	 worked_on +=ROW_LENGTH;
    }
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
#endif
#if VEC_SET == AVX_256
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 2
#define STEP_SIZE 4
#define UPPER_BOUND 2
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 4
#define STEP_SIZE 8
#define UPPER_BOUND 4
#endif
    for (i = 0; i < nq - UPPER_BOUND; i+=STEP_SIZE)
    {
         CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, s);
         CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (&q[i+ROW_LENGTH], hh, nb, ldq, ldh, s);
	 worked_on +=STEP_SIZE;
    }
#endif
816
817
818
819
820
 
    if (nq == i)
    {
      return;
    }
821
822
#if VEC_SET == SSE_128
    
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
#if VEC_SET == SSE_128
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 3
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 6
#endif
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 6
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 12
#endif
#endif /* VEC_SET == AVX_256 */

    if (nq-i == ROW_LENGTH)
    {
        CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, s);
        worked_on += ROW_LENGTH;
    }

#if VEC_SET == SSE_128
850
851
852
853
854
855
856
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 2
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 4
#endif
857
858
859
860
861
862
863
864
865
866
867
868
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 4
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 8
#endif
#endif /* VEC_SET == AVX_256 */

869
870
871
872
873
874
    if (nq-i == ROW_LENGTH)
    {
        CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, s);
        worked_on += ROW_LENGTH;
    }

875
876
877
878

#endif


879
#if VEC_SET == SSE_128
880
881
882
883
884
885
886
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 1
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 2
#endif
887
888
889
890
891
892
893
894
895
896
897
898
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
#undef ROW_LENGTH
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 2
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 4
#endif
#endif /* VEC_SET == AVX_256 */

899
900
901
902
903
904
905
906
907
908
909
910
911
912
    if (nq-i == ROW_LENGTH)
    {
        CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (&q[i], hh, nb, ldq, ldh, s);
        worked_on += ROW_LENGTH;
    }
#endif /* BLOCK2 */

//#ifdef WITH_DEBUG
    if (worked_on != nq)
    {
      printf("Error in complex SIMD_SET BLOCK BLOCK kernel %d %d\n", worked_on, nq);
      abort();
    }
//#endif
913
914
915
916


}

917
#if VEC_SET == SSE_128
918
919
920
921
922
923
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 6
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 12
#endif
924
925
926
927
928
929
930
931
932
933
934
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
#ifdef DOUBLE_PRECISION_COMPLEX
#define ROW_LENGTH 12
#endif
#ifdef SINGLE_PRECISION_COMPLEX
#define ROW_LENGTH 24
#endif
#endif /* VEC_SET == AVX_256 */

935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
static __forceinline void CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIMD_SET,_,BLOCK,hv_,WORD_LENGTH) (DATA_TYPE_PTR q, DATA_TYPE_PTR hh, int nb, int ldq
#ifdef BLOCK1
		)
#endif
#ifdef BLOCK2
                ,int ldh, DATA_TYPE s)
#endif
{

    DATA_TYPE_REAL_PTR q_dbl = (DATA_TYPE_REAL_PTR)q;
    DATA_TYPE_REAL_PTR hh_dbl = (DATA_TYPE_REAL_PTR)hh;
#ifdef BLOCK2
    DATA_TYPE_REAL_PTR s_dbl = (DATA_TYPE_REAL_PTR)(&s);
#endif

    __SIMD_DATATYPE x1, x2, x3, x4, x5, x6;
    __SIMD_DATATYPE q1, q2, q3, q4, q5, q6;
#ifdef BLOCK2
    __SIMD_DATATYPE y1, y2, y3, y4, y5, y6;
    __SIMD_DATATYPE h2_real, h2_imag;
#endif
    __SIMD_DATATYPE h1_real, h1_imag;
    __SIMD_DATATYPE tmp1, tmp2, tmp3, tmp4, tmp5, tmp6;
    int i=0;

#if VEC_SET == SSE_128
#ifdef DOUBLE_PRECISION_COMPLEX
    __SIMD_DATATYPE sign = (__SIMD_DATATYPE)_mm_set_epi64x(0x8000000000000000, 0x8000000000000000);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
    __SIMD_DATATYPE sign = (__SIMD_DATATYPE)_mm_set_epi32(0x80000000, 0x80000000, 0x80000000, 0x80000000);
#endif
#endif /* VEC_SET == SSE_128 */

969
970
971
972
973
974
975
976
977
#if VEC_SET == AVX_256
#ifdef DOUBLE_PRECISION_COMPLEX
    __SIMD_DATATYPE sign = (__SIMD_DATATYPE)_mm256_set_epi64x(0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
    __SIMD_DATATYPE sign = (__SIMD_DATATYPE)_mm256_set_epi32(0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000);
#endif
#endif /* VEC_SET == AVX_256 */

978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
#ifdef BLOCK2
     x1 = _SIMD_LOAD(&q_dbl[(2*ldq)+0]);
     x2 = _SIMD_LOAD(&q_dbl[(2*ldq)+offset]);
     x3 = _SIMD_LOAD(&q_dbl[(2*ldq)+2*offset]);
     x4 = _SIMD_LOAD(&q_dbl[(2*ldq)+3*offset]);
     x5 = _SIMD_LOAD(&q_dbl[(2*ldq)+4*offset]);
     x6 = _SIMD_LOAD(&q_dbl[(2*ldq)+5*offset]);

#if VEC_SET == SSE_128
#ifdef DOUBLE_PRECISION_COMPLEX
     h2_real = _mm_loaddup_pd(&hh_dbl[(ldh+1)*2]);
     h2_imag = _mm_loaddup_pd(&hh_dbl[((ldh+1)*2)+1]);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
     h2_real = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[(ldh+1)*2]) )));
     h2_imag = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[((ldh+1)*2)+1]) )));
#endif
#endif /* VEC_SET == SSE_128 */

997
998
999
1000
1001
#if VEC_SET == AVX_256
     h2_real = _SIMD_BROADCAST(&hh_dbl[(ldh+1)*2]);
     h2_imag = _SIMD_BROADCAST(&hh_dbl[((ldh+1)*2)+1]);
#endif /* VEC_SET == AVX_256 */

1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
#ifndef __ELPA_USE_FMA__
     // conjugate
     h2_imag = _SIMD_XOR(h2_imag, sign);
#endif

     y1 = _SIMD_LOAD(&q_dbl[0]);
     y2 = _SIMD_LOAD(&q_dbl[offset]);
     y3 = _SIMD_LOAD(&q_dbl[2*offset]);
     y4 = _SIMD_LOAD(&q_dbl[3*offset]);
     y5 = _SIMD_LOAD(&q_dbl[4*offset]);
     y6 = _SIMD_LOAD(&q_dbl[5*offset]);

     tmp1 = _SIMD_MUL(h2_imag, x1);
#ifdef __ELPA_USE_FMA__
1016
     y1 = _SIMD_ADD(y1, _SIMD_FMSUBADD(h2_real, x1, _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
1017
1018
1019
1020
1021
#else
     y1 = _SIMD_ADD(y1, _SIMD_ADDSUB( _SIMD_MUL(h2_real, x1), _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
     tmp2 = _SIMD_MUL(h2_imag, x2);
#ifdef __ELPA_USE_FMA__
1022
     y2 = _SIMD_ADD(y2, _SIMD_FMSUBADD(h2_real, x2, _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
1023
1024
1025
1026
1027
1028
#else
     y2 = _SIMD_ADD(y2, _SIMD_ADDSUB( _SIMD_MUL(h2_real, x2), _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

     tmp3 = _SIMD_MUL(h2_imag, x3);
#ifdef __ELPA_USE_FMA__
1029
     y3 = _SIMD_ADD(y3, _SIMD_FMSUBADD(h2_real, x3, _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
1030
1031
1032
1033
1034
#else
     y3 = _SIMD_ADD(y3, _SIMD_ADDSUB( _SIMD_MUL(h2_real, x3), _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
     tmp4 = _SIMD_MUL(h2_imag, x4);
#ifdef __ELPA_USE_FMA__
1035
     y4 = _SIMD_ADD(y4, _SIMD_FMSUBADD(h2_real, x4, _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
1036
1037
1038
1039
1040
1041
#else
     y4 = _SIMD_ADD(y4, _SIMD_ADDSUB( _SIMD_MUL(h2_real, x4), _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif

     tmp5 = _SIMD_MUL(h2_imag, x5);
#ifdef __ELPA_USE_FMA__
1042
     y5 = _SIMD_ADD(y5, _SIMD_FMSUBADD(h2_real, x5, _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
1043
1044
1045
1046
1047
#else
     y5 = _SIMD_ADD(y5, _SIMD_ADDSUB( _SIMD_MUL(h2_real, x5), _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
#endif
     tmp6 = _SIMD_MUL(h2_imag, x6);
#ifdef __ELPA_USE_FMA__
1048
     y6 = _SIMD_ADD(y6, _SIMD_FMSUBADD(h2_real, x6, _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
#else
     y6 = _SIMD_ADD(y6, _SIMD_ADDSUB( _SIMD_MUL(h2_real, x6), _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
#endif

#endif /* BLOCK2 */

#ifdef BLOCK1
    x1 = _SIMD_LOAD(&q_dbl[0]);
    x2 = _SIMD_LOAD(&q_dbl[offset]);
    x3 = _SIMD_LOAD(&q_dbl[2*offset]);
    x4 = _SIMD_LOAD(&q_dbl[3*offset]);
    x5 = _SIMD_LOAD(&q_dbl[4*offset]);
    x6 = _SIMD_LOAD(&q_dbl[5*offset]);
#endif

    for (i = BLOCK; i < nb; i++)
    {

#if VEC_SET == SSE_128
#ifdef DOUBLE_PRECISION_COMPLEX
        h1_real = _mm_loaddup_pd(&hh_dbl[(i-BLOCK+1)*2]);
        h1_imag = _mm_loaddup_pd(&hh_dbl[((i-BLOCK+1)*2)+1]);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
        h1_real = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[(i-BLOCK+1)*2]) )));
        h1_imag = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[((i-BLOCK+1)*2)+1]) )));
#endif
#endif /* VEC_SET == SSE_128 */

1078
1079
1080
1081
1082
#if VEC_SET == AVX_256
       h1_real = _SIMD_BROADCAST(&hh_dbl[(i-BLOCK+1)*2]);
       h1_imag = _SIMD_BROADCAST(&hh_dbl[((i-BLOCK+1)*2)+1]);
#endif /* VEC_SET == AVX_256 */

1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
#ifndef __ELPA_USE_FMA__
        // conjugate
        h1_imag = _SIMD_XOR(h1_imag, sign);
#endif

        q1 = _SIMD_LOAD(&q_dbl[(2*i*ldq)+0]);
        q2 = _SIMD_LOAD(&q_dbl[(2*i*ldq)+offset]);
        q3 = _SIMD_LOAD(&q_dbl[(2*i*ldq)+2*offset]);
        q4 = _SIMD_LOAD(&q_dbl[(2*i*ldq)+3*offset]);
        q5 = _SIMD_LOAD(&q_dbl[(2*i*ldq)+4*offset]);
        q6 = _SIMD_LOAD(&q_dbl[(2*i*ldq)+5*offset]);

        tmp1 = _SIMD_MUL(h1_imag, q1);
#ifdef __ELPA_USE_FMA__
1097
        x1 = _SIMD_ADD(x1, _SIMD_FMSUBADD(h1_real, q1, _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
1098
1099
1100
1101
1102
#else
        x1 = _SIMD_ADD(x1, _SIMD_ADDSUB( _SIMD_MUL(h1_real, q1), _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
        tmp2 = _SIMD_MUL(h1_imag, q2);
#ifdef __ELPA_USE_FMA__
1103
        x2 = _SIMD_ADD(x2, _SIMD_FMSUBADD(h1_real, q2, _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
1104
1105
1106
1107
1108
#else
        x2 = _SIMD_ADD(x2, _SIMD_ADDSUB( _SIMD_MUL(h1_real, q2), _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif
        tmp3 = _SIMD_MUL(h1_imag, q3);
#ifdef __ELPA_USE_FMA__
1109
        x3 = _SIMD_ADD(x3, _SIMD_FMSUBADD(h1_real, q3, _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
1110
1111
1112
1113
1114
1115
#else
        x3 = _SIMD_ADD(x3, _SIMD_ADDSUB( _SIMD_MUL(h1_real, q3), _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif

        tmp4 = _SIMD_MUL(h1_imag, q4);
#ifdef __ELPA_USE_FMA__
1116
        x4 = _SIMD_ADD(x4, _SIMD_FMSUBADD(h1_real, q4, _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
1117
1118
1119
1120
1121
#else
        x4 = _SIMD_ADD(x4, _SIMD_ADDSUB( _SIMD_MUL(h1_real, q4), _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif
        tmp5 = _SIMD_MUL(h1_imag, q5);
#ifdef __ELPA_USE_FMA__
1122
        x5 = _SIMD_ADD(x5, _SIMD_FMSUBADD(h1_real, q5, _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
1123
1124
1125
1126
1127
#else
        x5 = _SIMD_ADD(x5, _SIMD_ADDSUB( _SIMD_MUL(h1_real, q5), _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
#endif
        tmp6 = _SIMD_MUL(h1_imag, q6);
#ifdef __ELPA_USE_FMA__
1128
        x6 = _SIMD_ADD(x6, _SIMD_FMSUBADD(h1_real, q6, _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
#else
        x6 = _SIMD_ADD(x6, _SIMD_ADDSUB( _SIMD_MUL(h1_real, q6), _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
#endif

#ifdef BLOCK2

#if VEC_SET == SSE_128
#ifdef DOUBLE_PRECISION_COMPLEX
          h2_real = _mm_loaddup_pd(&hh_dbl[(ldh+i)*2]);
          h2_imag = _mm_loaddup_pd(&hh_dbl[((ldh+i)*2)+1]);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
          h2_real = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[(ldh+i)*2]) )));
          h2_imag = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[((ldh+i)*2)+1]) )));
#endif
#endif /* VEC_SET == SSE_128 */

1146
1147
1148
1149
1150
#if VEC_SET == AVX_256
          h2_real = _SIMD_BROADCAST(&hh_dbl[(ldh+i)*2]);
          h2_imag = _SIMD_BROADCAST(&hh_dbl[((ldh+i)*2)+1]);
#endif /* VEC_SET == AVX_256 */

1151
1152
1153
1154
1155
1156
1157
#ifndef __ELPA_USE_FMA__
          // conjugate
          h2_imag = _SIMD_XOR(h2_imag, sign);
#endif

          tmp1 = _SIMD_MUL(h2_imag, q1);
#ifdef __ELPA_USE_FMA__
1158
          y1 = _SIMD_ADD(y1, _SIMD_FMSUBADD(h2_real, q1, _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
1159
1160
1161
1162
1163
#else
          y1 = _SIMD_ADD(y1, _SIMD_ADDSUB( _SIMD_MUL(h2_real, q1), _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
          tmp2 = _SIMD_MUL(h2_imag, q2);
#ifdef __ELPA_USE_FMA__
1164
          y2 = _SIMD_ADD(y2, _SIMD_FMSUBADD(h2_real, q2, _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
1165
1166
1167
1168
1169
1170
#else
          y2 = _SIMD_ADD(y2, _SIMD_ADDSUB( _SIMD_MUL(h2_real, q2), _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

          tmp3 = _SIMD_MUL(h2_imag, q3);
#ifdef __ELPA_USE_FMA__
1171
          y3 = _SIMD_ADD(y3, _SIMD_FMSUBADD(h2_real, q3, _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
1172
1173
1174
1175
1176
#else
          y3 = _SIMD_ADD(y3, _SIMD_ADDSUB( _SIMD_MUL(h2_real, q3), _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
          tmp4 = _SIMD_MUL(h2_imag, q4);
#ifdef __ELPA_USE_FMA__
1177
          y4 = _SIMD_ADD(y4, _SIMD_FMSUBADD(h2_real, q4, _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
1178
1179
1180
1181
1182
1183
#else
          y4 = _SIMD_ADD(y4, _SIMD_ADDSUB( _SIMD_MUL(h2_real, q4), _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif

          tmp5 = _SIMD_MUL(h2_imag, q5);
#ifdef __ELPA_USE_FMA__
1184
          y5 = _SIMD_ADD(y5, _SIMD_FMSUBADD(h2_real, q5, _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
1185
1186
1187
1188
1189
#else
          y5 = _SIMD_ADD(y5, _SIMD_ADDSUB( _SIMD_MUL(h2_real, q5), _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
#endif
          tmp6 = _SIMD_MUL(h2_imag, q6);
#ifdef __ELPA_USE_FMA__
1190
          y6 = _SIMD_ADD(y6, _SIMD_FMSUBADD(h2_real, q6, _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
#else
          y6 = _SIMD_ADD(y6, _SIMD_ADDSUB( _SIMD_MUL(h2_real, q6), _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
#endif
	
#endif /* BLOCK2 */

    }

#ifdef BLOCK2

1201
#if VEC_SET == SSE_128
1202
1203
1204
1205
1206
1207
1208
1209
#ifdef DOUBLE_PRECISION_COMPLEX
     h1_real = _mm_loaddup_pd(&hh_dbl[(nb-1)*2]);
     h1_imag = _mm_loaddup_pd(&hh_dbl[((nb-1)*2)+1]);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
     h1_real = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[(nb-1)*2]) )));
     h1_imag = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[((nb-1)*2)+1]) )));
#endif
1210
1211
1212
1213
1214
1215
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
     h1_real = _SIMD_BROADCAST(&hh_dbl[(nb-1)*2]);
     h1_imag = _SIMD_BROADCAST(&hh_dbl[((nb-1)*2)+1]);
#endif /* VEC_SET == AVX_256 */
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230

#ifndef __ELPA_USE_FMA__
     // conjugate
     h1_imag = _SIMD_XOR(h1_imag, sign);
#endif

     q1 = _SIMD_LOAD(&q_dbl[(2*nb*ldq)+0]);
     q2 = _SIMD_LOAD(&q_dbl[(2*nb*ldq)+offset]);
     q3 = _SIMD_LOAD(&q_dbl[(2*nb*ldq)+2*offset]);
     q4 = _SIMD_LOAD(&q_dbl[(2*nb*ldq)+3*offset]);
     q5 = _SIMD_LOAD(&q_dbl[(2*nb*ldq)+4*offset]);
     q6 = _SIMD_LOAD(&q_dbl[(2*nb*ldq)+5*offset]);

     tmp1 = _SIMD_MUL(h1_imag, q1);
#ifdef __ELPA_USE_FMA__
1231
     x1 = _SIMD_ADD(x1, _SIMD_FMSUBADD(h1_real, q1, _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
1232
1233
1234
1235
1236
#else
     x1 = _SIMD_ADD(x1, _SIMD_ADDSUB( _SIMD_MUL(h1_real, q1), _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
     tmp2 = _SIMD_MUL(h1_imag, q2);
#ifdef __ELPA_USE_FMA__
1237
     x2 = _SIMD_ADD(x2, _SIMD_FMSUBADD(h1_real, q2, _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
1238
1239
1240
1241
1242
1243
#else
     x2 = _SIMD_ADD(x2, _SIMD_ADDSUB( _SIMD_MUL(h1_real, q2), _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

     tmp3 = _SIMD_MUL(h1_imag, q3);
#ifdef __ELPA_USE_FMA__
1244
     x3 = _SIMD_ADD(x3, _SIMD_FMSUBADD(h1_real, q3, _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
1245
1246
1247
1248
1249
#else
     x3 = _SIMD_ADD(x3, _SIMD_ADDSUB( _SIMD_MUL(h1_real, q3), _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
     tmp4 = _SIMD_MUL(h1_imag, q4);
#ifdef __ELPA_USE_FMA__
1250
     x4 = _SIMD_ADD(x4, _SIMD_FMSUBADD(h1_real, q4, _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
1251
1252
1253
1254
1255
1256
#else
     x4 = _SIMD_ADD(x4, _SIMD_ADDSUB( _SIMD_MUL(h1_real, q4), _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif

     tmp5 = _SIMD_MUL(h1_imag, q5);
#ifdef __ELPA_USE_FMA__
1257
     x5 = _SIMD_ADD(x5, _SIMD_FMSUBADD(h1_real, q5, _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
1258
1259
1260
1261
1262
#else
     x5 = _SIMD_ADD(x5, _SIMD_ADDSUB( _SIMD_MUL(h1_real, q5), _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
#endif
     tmp6 = _SIMD_MUL(h1_imag, q6);
#ifdef __ELPA_USE_FMA__
1263
     x6 = _SIMD_ADD(x6, _SIMD_FMSUBADD(h1_real, q6, _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
#else
     x6 = _SIMD_ADD(x6, _SIMD_ADDSUB( _SIMD_MUL(h1_real, q6), _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
#endif

#endif /* BLOCK2 */

#if VEC_SET == SSE_128
#ifdef DOUBLE_PRECISION_COMPLEX
    h1_real = _mm_loaddup_pd(&hh_dbl[0]);
    h1_imag = _mm_loaddup_pd(&hh_dbl[1]);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
    h1_real = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[0]) )));
    h1_imag = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[1]) )));
#endif
#endif /*  VEC_SET == SSE_128 */

1281
1282
1283
1284
1285
#if VEC_SET == AVX_256
    h1_real = _SIMD_BROADCAST(&hh_dbl[0]);
    h1_imag = _SIMD_BROADCAST(&hh_dbl[1]);
#endif /* VEC_SET == AVX_256 */

1286
1287
1288
1289
1290
    h1_real = _SIMD_XOR(h1_real, sign);
    h1_imag = _SIMD_XOR(h1_imag, sign);

    tmp1 = _SIMD_MUL(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
1291
    x1 = _SIMD_FMADDSUB(h1_real, x1, _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE));
1292
1293
1294
1295
1296
#else
    x1 = _SIMD_ADDSUB( _SIMD_MUL(h1_real, x1), _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#endif
    tmp2 = _SIMD_MUL(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
1297
    x2 = _SIMD_FMADDSUB(h1_real, x2, _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE));
1298
1299
1300
1301
1302
#else
    x2 = _SIMD_ADDSUB( _SIMD_MUL(h1_real, x2), _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE));
#endif
    tmp3 = _SIMD_MUL(h1_imag, x3);
#ifdef __ELPA_USE_FMA__
1303
    x3 = _SIMD_FMADDSUB(h1_real, x3, _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE));
1304
1305
1306
1307
1308
1309
#else
    x3 = _SIMD_ADDSUB( _SIMD_MUL(h1_real, x3), _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE));
#endif

    tmp4 = _SIMD_MUL(h1_imag, x4);
#ifdef __ELPA_USE_FMA__
1310
    x4 = _SIMD_FMADDSUB(h1_real, x4, _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE));
1311
1312
1313
1314
1315
#else
    x4 = _SIMD_ADDSUB( _SIMD_MUL(h1_real, x4), _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE));
#endif
    tmp5 = _SIMD_MUL(h1_imag, x5);
#ifdef __ELPA_USE_FMA__
1316
    x5 = _SIMD_FMADDSUB(h1_real, x5, _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE));
1317
1318
1319
1320
1321
#else
    x5 = _SIMD_ADDSUB( _SIMD_MUL(h1_real, x5), _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE));
#endif
    tmp6 = _SIMD_MUL(h1_imag, x6);
#ifdef __ELPA_USE_FMA__
1322
    x6 = _SIMD_FMADDSUB(h1_real, x6, _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE));
1323
1324
1325
1326
1327
#else
    x6 = _SIMD_ADDSUB( _SIMD_MUL(h1_real, x6), _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE));
#endif

#ifdef BLOCK2
1328
1329

#if VEC_SET == SSE_128    
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
#ifdef DOUBLE_PRECISION_COMPLEX
     h1_real = _mm_loaddup_pd(&hh_dbl[ldh*2]);
     h1_imag = _mm_loaddup_pd(&hh_dbl[(ldh*2)+1]);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
     h1_real = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[ldh*2]) )));
     h1_imag = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[(ldh*2)+1]) )));
#endif

#ifdef DOUBLE_PRECISION_COMPLEX
     h2_real = _mm_loaddup_pd(&hh_dbl[ldh*2]);
     h2_imag = _mm_loaddup_pd(&hh_dbl[(ldh*2)+1]);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
     h2_real = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[ldh*2]) )));
     h2_imag = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[(ldh*2)+1]) )));
#endif
1347
1348
1349
1350
1351
1352
1353
1354
#endif /* VEC_SET == 128 */

#if VEC_SET == AVX_256
     h1_real = _SIMD_BROADCAST(&hh_dbl[ldh*2]);
     h1_imag = _SIMD_BROADCAST(&hh_dbl[(ldh*2)+1]);
     h2_real = _SIMD_BROADCAST(&hh_dbl[ldh*2]);
     h2_imag = _SIMD_BROADCAST(&hh_dbl[(ldh*2)+1]);
#endif /* VEC_SET == AVX_256 */
1355
1356
1357
1358
1359
1360

     h1_real = _SIMD_XOR(h1_real, sign);
     h1_imag = _SIMD_XOR(h1_imag, sign);
     h2_real = _SIMD_XOR(h2_real, sign);
     h2_imag = _SIMD_XOR(h2_imag, sign);

1361
#if VEC_SET == SSE_128
1362
1363
1364
1365
1366
#ifdef SINGLE_PRECISION_COMPLEX
     tmp2 = _mm_castpd_ps(_mm_load_pd1((double *) s_dbl));
#else
     tmp2 = _SIMD_LOADU(s_dbl);
#endif
1367
#endif /* VEC_SET == SSE_128 */
1368

1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
#if VEC_SET == AVX_256
#ifdef DOUBLE_PRECISION_COMPLEX
     tmp2 = _mm256_set_pd(s_dbl[1], s_dbl[0], s_dbl[1], s_dbl[0]);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
     tmp2 = _mm256_set_ps(s_dbl[1], s_dbl[0], s_dbl[1], s_dbl[0],
                             s_dbl[1], s_dbl[0], s_dbl[1], s_dbl[0]);
#endif
#endif /* VEC_SET == AVX_256 */

     tmp1 = _SIMD_MUL(h2_imag, tmp2);
1380
#ifdef __ELPA_USE_FMA__
1381
     tmp2 = _SIMD_FMSUBADD(h2_real, tmp2, _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE));
1382
1383
1384
1385
#else
     tmp2 = _SIMD_ADDSUB( _SIMD_MUL(h2_real, tmp2), _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#endif

1386
#if VEC_SET == SSE_128
1387
1388
1389
1390
1391
1392
1393
1394
#ifdef DOUBLE_PRECISION_COMPLEX
     h2_real = _mm_movedup_pd(tmp2);
     h2_imag = _mm_set1_pd(tmp2[1]);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
     h2_real = _mm_moveldup_ps(tmp2);
     h2_imag = _mm_movehdup_ps(tmp2);
#endif
1395
1396
1397
1398
1399
1400
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
     h2_real = _SIMD_SET1(tmp2[0]);
     h2_imag = _SIMD_SET1(tmp2[1]);
#endif /* VEC_SET == AVX_256 */
1401
1402
1403

     tmp1 = _SIMD_MUL(h1_imag, y1);
#ifdef __ELPA_USE_FMA__
1404
     y1 = _SIMD_FMSUBADD(h1_real, y1, _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE));
1405
1406
1407
1408
1409
#else
     y1 = _SIMD_ADDSUB( _SIMD_MUL(h1_real, y1), _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#endif
     tmp2 = _SIMD_MUL(h1_imag, y2);
#ifdef __ELPA_USE_FMA__
1410
     y2 = _SIMD_FMSUBADD(h1_real, y2, _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE));
1411
1412
1413
1414
1415
1416
#else
     y2 = _SIMD_ADDSUB( _SIMD_MUL(h1_real, y2), _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE));
#endif

     tmp3 = _SIMD_MUL(h1_imag, y3);
#ifdef __ELPA_USE_FMA__
1417
     y3 = _SIMD_FMSUBADD(h1_real, y3, _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE));
1418
1419
1420
1421
1422
#else
     y3 = _SIMD_ADDSUB( _SIMD_MUL(h1_real, y3), _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE));
#endif
     tmp4 = _SIMD_MUL(h1_imag, y4);
#ifdef __ELPA_USE_FMA__
1423
     y4 = _SIMD_FMSUBADD(h1_real, y4, _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE));
1424
1425
1426
1427
1428
1429
#else
     y4 = _SIMD_ADDSUB( _SIMD_MUL(h1_real, y4), _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE));
#endif

     tmp5 = _SIMD_MUL(h1_imag, y5);
#ifdef __ELPA_USE_FMA__
1430
     y5 = _SIMD_FMSUBADD(h1_real, y5, _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE));
1431
1432
1433
1434
1435
#else
     y5 = _SIMD_ADDSUB( _SIMD_MUL(h1_real, y5), _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE));
#endif
     tmp6 = _SIMD_MUL(h1_imag, y6);
#ifdef __ELPA_USE_FMA__
1436
     y6 = _SIMD_FMSUBADD(h1_real, y6, _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE));
1437
1438
1439
1440
1441
1442
#else
     y6 = _SIMD_ADDSUB( _SIMD_MUL(h1_real, y6), _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE));
#endif

     tmp1 = _SIMD_MUL(h2_imag, x1);
#ifdef __ELPA_USE_FMA__
1443
     y1 = _SIMD_ADD(y1, _SIMD_FMSUBADD(h2_real, x1, _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
1444
1445
1446
1447
1448
#else
     y1 = _SIMD_ADD(y1, _SIMD_ADDSUB( _SIMD_MUL(h2_real, x1), _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
     tmp2 = _SIMD_MUL(h2_imag, x2);
#ifdef __ELPA_USE_FMA__
1449
     y2 = _SIMD_ADD(y2, _SIMD_FMSUBADD(h2_real, x2, _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
1450
1451
1452
1453
1454
1455
#else
     y2 = _SIMD_ADD(y2, _SIMD_ADDSUB( _SIMD_MUL(h2_real, x2), _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

     tmp3 = _SIMD_MUL(h2_imag, x3);
#ifdef __ELPA_USE_FMA__
1456
     y3 = _SIMD_ADD(y3, _SIMD_FMSUBADD(h2_real, x3, _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
1457
1458
1459
1460
1461
#else
     y3 = _SIMD_ADD(y3, _SIMD_ADDSUB( _SIMD_MUL(h2_real, x3), _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
     tmp4 = _SIMD_MUL(h2_imag, x4);
#ifdef __ELPA_USE_FMA__
1462
     y4 = _SIMD_ADD(y4, _SIMD_FMSUBADD(h2_real, x4, _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
1463
1464
1465
1466
1467
1468
#else
     y4 = _SIMD_ADD(y4, _SIMD_ADDSUB( _SIMD_MUL(h2_real, x4), _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif

     tmp5 = _SIMD_MUL(h2_imag, x5);
#ifdef __ELPA_USE_FMA__
1469
     y5 = _SIMD_ADD(y5, _SIMD_FMSUBADD(h2_real, x5, _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
1470
1471
1472
1473
1474
#else
     y5 = _SIMD_ADD(y5, _SIMD_ADDSUB( _SIMD_MUL(h2_real, x5), _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
#endif
     tmp6 = _SIMD_MUL(h2_imag, x6);
#ifdef __ELPA_USE_FMA__
1475
     y6 = _SIMD_ADD(y6, _SIMD_FMSUBADD(h2_real, x6, _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
#else
     y6 = _SIMD_ADD(y6, _SIMD_ADDSUB( _SIMD_MUL(h2_real, x6), _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
#endif

#endif /* BLOCK2 */

    q1 = _SIMD_LOAD(&q_dbl[0]);
    q2 = _SIMD_LOAD(&q_dbl[offset]);
    q3 = _SIMD_LOAD(&q_dbl[2*offset]);
    q4 = _SIMD_LOAD(&q_dbl[3*offset]);
    q5 = _SIMD_LOAD(&q_dbl[4*offset]);
    q6 = _SIMD_LOAD(&q_dbl[5*offset]);

1489
#ifdef BLOCK1
1490
1491
1492
1493
1494
1495
    q1 = _SIMD_ADD(q1, x1);
    q2 = _SIMD_ADD(q2, x2);
    q3 = _SIMD_ADD(q3, x3);
    q4 = _SIMD_ADD(q4, x4);
    q5 = _SIMD_ADD(q5, x5);
    q6 = _SIMD_ADD(q6, x6);
1496
#endif
1497

1498
1499
1500
1501
1502
1503
1504
1505
1506

#ifdef BLOCK2
    q1 = _SIMD_ADD(q1, y1);
    q2 = _SIMD_ADD(q2, y2);
    q3 = _SIMD_ADD(q3, y3);
    q4 = _SIMD_ADD(q4, y4);
    q5 = _SIMD_ADD(q5, y5);
    q6 = _SIMD_ADD(q6, y6);
#endif
1507

1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
    _SIMD_STORE(&q_dbl[0], q1);
    _SIMD_STORE(&q_dbl[offset], q2);
    _SIMD_STORE(&q_dbl[2*offset], q3);
    _SIMD_STORE(&q_dbl[3*offset], q4);
    _SIMD_STORE(&q_dbl[4*offset], q5);
    _SIMD_STORE(&q_dbl[5*offset], q6);


#ifdef BLOCK2

1518
#if VEC_SET == SSE_128
1519
1520
1521
1522
1523
1524
1525
1526
#ifdef DOUBLE_PRECISION_COMPLEX
     h2_real = _mm_loaddup_pd(&hh_dbl[(ldh+1)*2]);
     h2_imag = _mm_loaddup_pd(&hh_dbl[((ldh+1)*2)+1]);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
     h2_real = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[(ldh+1)*2]) )));
     h2_imag = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[((ldh+1)*2)+1]) )));
#endif
1527
1528
1529
1530
1531
1532
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
     h2_real = _SIMD_BROADCAST(&hh_dbl[(ldh+1)*2]);
     h2_imag = _SIMD_BROADCAST(&hh_dbl[((ldh+1)*2)+1]);
#endif /* VEC_SET == AVX_256 */
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549

     q1 = _SIMD_LOAD(&q_dbl[(ldq*2)+0]);
     q2 = _SIMD_LOAD(&q_dbl[(ldq*2)+offset]);
     q3 = _SIMD_LOAD(&q_dbl[(ldq*2)+2*offset]);
     q4 = _SIMD_LOAD(&q_dbl[(ldq*2)+3*offset]);
     q5 = _SIMD_LOAD(&q_dbl[(ldq*2)+4*offset]);
     q6 = _SIMD_LOAD(&q_dbl[(ldq*2)+5*offset]);

     q1 = _SIMD_ADD(q1, x1);
     q2 = _SIMD_ADD(q2, x2);
     q3 = _SIMD_ADD(q3, x3);
     q4 = _SIMD_ADD(q4, x4);
     q5 = _SIMD_ADD(q5, x5);
     q6 = _SIMD_ADD(q6, x6);

     tmp1 = _SIMD_MUL(h2_imag, y1);
#ifdef __ELPA_USE_FMA__
1550
     q1 = _SIMD_ADD(q1, _SIMD_FMSUBADD(h2_real, y1, _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
1551
1552
1553
1554
1555
#else
     q1 = _SIMD_ADD(q1, _SIMD_ADDSUB( _SIMD_MUL(h2_real, y1), _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
     tmp2 = _SIMD_MUL(h2_imag, y2);
#ifdef __ELPA_USE_FMA__
1556
     q2 = _SIMD_ADD(q2, _SIMD_FMSUBADD(h2_real, y2, _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
1557
1558
1559
1560
1561
1562
#else
     q2 = _SIMD_ADD(q2, _SIMD_ADDSUB( _SIMD_MUL(h2_real, y2), _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

     tmp3 = _SIMD_MUL(h2_imag, y3);
#ifdef __ELPA_USE_FMA__
1563
     q3 = _SIMD_ADD(q3, _SIMD_FMSUBADD(h2_real, y3, _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
1564
1565
1566
1567
1568
#else
     q3 = _SIMD_ADD(q3, _SIMD_ADDSUB( _SIMD_MUL(h2_real, y3), _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
     tmp4 = _SIMD_MUL(h2_imag, y4);
#ifdef __ELPA_USE_FMA__
1569
     q4 = _SIMD_ADD(q4, _SIMD_FMSUBADD(h2_real, y4, _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
1570
1571
1572
1573
1574
1575
#else
     q4 = _SIMD_ADD(q4, _SIMD_ADDSUB( _SIMD_MUL(h2_real, y4), _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif

     tmp5 = _SIMD_MUL(h2_imag, y5);
#ifdef __ELPA_USE_FMA__
1576
     q5 = _SIMD_ADD(q5, _SIMD_FMSUBADD(h2_real, y5, _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
1577
1578
1579
1580
1581
#else
     q5 = _SIMD_ADD(q5, _SIMD_ADDSUB( _SIMD_MUL(h2_real, y5), _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
#endif
     tmp6 = _SIMD_MUL(h2_imag, y6);
#ifdef __ELPA_USE_FMA__
1582
     q6 = _SIMD_ADD(q6, _SIMD_FMSUBADD(h2_real, y6, _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
#else
     q6 = _SIMD_ADD(q6, _SIMD_ADDSUB( _SIMD_MUL(h2_real, y6), _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
#endif

     _SIMD_STORE(&q_dbl[(ldq*2)+0], q1);
     _SIMD_STORE(&q_dbl[(ldq*2)+offset], q2);
     _SIMD_STORE(&q_dbl[(ldq*2)+2*offset], q3);
     _SIMD_STORE(&q_dbl[(ldq*2)+3*offset], q4);
     _SIMD_STORE(&q_dbl[(ldq*2)+4*offset], q5);
     _SIMD_STORE(&q_dbl[(ldq*2)+5*offset], q6);

#endif /* BLOCK2 */


    for (i = BLOCK; i < nb; i++)
    {

#if VEC_SET == SSE_128
#ifdef DOUBLE_PRECISION_COMPLEX
        h1_real = _mm_loaddup_pd(&hh_dbl[(i-BLOCK+1)*2]);
        h1_imag = _mm_loaddup_pd(&hh_dbl[((i-BLOCK+1)*2)+1]);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
        h1_real = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[(i-BLOCK+1)*2]) )));
        h1_imag = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[((i-BLOCK+1)*2)+1]) )));
#endif
#endif /* VEC_SET == SSE_128 */

1611
1612
1613
1614
1615
#if VEC_SET == AVX_256
	h1_real = _SIMD_BROADCAST(&hh_dbl[(i-BLOCK+1)*2]);
        h1_imag = _SIMD_BROADCAST(&hh_dbl[((i-BLOCK+1)*2)+1]);
#endif /* VEC_SET == AVX_256 */

1616
1617
1618
1619
1620
1621
1622
        q1 = _SIMD_LOAD(&q_dbl[(2*i*ldq)+0]);
        q2 = _SIMD_LOAD(&q_dbl[(2*i*ldq)+offset]);
        q3 = _SIMD_LOAD(&q_dbl[(2*i*ldq)+2*offset]);
        q4 = _SIMD_LOAD(&q_dbl[(2*i*ldq)+3*offset]);
        q5 = _SIMD_LOAD(&q_dbl[(2*i*ldq)+4*offset]);
        q6 = _SIMD_LOAD(&q_dbl[(2*i*ldq)+5*offset]);

1623
        tmp1 = _SIMD_MUL(h1_imag, x1);
1624
#ifdef __ELPA_USE_FMA__
1625
        q1 = _SIMD_ADD(q1, _SIMD_FMADDSUB(h1_real, x1, _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
1626
1627
1628
1629
1630
#else
        q1 = _SIMD_ADD(q1, _SIMD_ADDSUB( _SIMD_MUL(h1_real, x1), _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
        tmp2 = _SIMD_MUL(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
1631
        q2 = _SIMD_ADD(q2, _SIMD_FMADDSUB(h1_real, x2, _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
1632
1633
1634
1635
1636
#else
        q2 = _SIMD_ADD(q2, _SIMD_ADDSUB( _SIMD_MUL(h1_real, x2), _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif
        tmp3 = _SIMD_MUL(h1_imag, x3);
#ifdef __ELPA_USE_FMA__
1637
        q3 = _SIMD_ADD(q3, _SIMD_FMADDSUB(h1_real, x3, _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
1638
1639
1640
1641
1642
1643
#else
        q3 = _SIMD_ADD(q3, _SIMD_ADDSUB( _SIMD_MUL(h1_real, x3), _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif

         tmp4 = _SIMD_MUL(h1_imag, x4);
#ifdef __ELPA_USE_FMA__
1644
         q4 = _SIMD_ADD(q4, _SIMD_FMADDSUB(h1_real, x4, _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
1645
1646
1647
1648
1649
#else
         q4 = _SIMD_ADD(q4, _SIMD_ADDSUB( _SIMD_MUL(h1_real, x4), _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif
         tmp5 = _SIMD_MUL(h1_imag, x5);
#ifdef __ELPA_USE_FMA__
1650
         q5 = _SIMD_ADD(q5, _SIMD_FMADDSUB(h1_real, x5, _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
1651
1652
1653
1654
1655
#else
         q5 = _SIMD_ADD(q5, _SIMD_ADDSUB( _SIMD_MUL(h1_real, x5), _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
#endif
         tmp6 = _SIMD_MUL(h1_imag, x6);
#ifdef __ELPA_USE_FMA__
1656
         q6 = _SIMD_ADD(q6, _SIMD_FMADDSUB(h1_real, x6, _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
1657
1658
1659
1660
1661
1662
#else
         q6 = _SIMD_ADD(q6, _SIMD_ADDSUB( _SIMD_MUL(h1_real, x6), _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
#endif

#ifdef BLOCK2

1663
#if VEC_SET == SSE_128
1664
1665
1666
1667
1668
1669
1670
1671
#ifdef DOUBLE_PRECISION_COMPLEX
          h2_real = _mm_loaddup_pd(&hh_dbl[(ldh+i)*2]);
          h2_imag = _mm_loaddup_pd(&hh_dbl[((ldh+i)*2)+1]);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
          h2_real = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[(ldh+i)*2]) )));
          h2_imag = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[((ldh+i)*2)+1]) )));
#endif
1672
1673
1674
1675
1676
1677
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
	  h2_real = _SIMD_BROADCAST(&hh_dbl[(ldh+i)*2]);
          h2_imag = _SIMD_BROADCAST(&hh_dbl[((ldh+i)*2)+1]);
#endif /* VEC_SET == AVX_256 */
1678
1679
1680

          tmp1 = _SIMD_MUL(h2_imag, y1);
#ifdef __ELPA_USE_FMA__
1681
          q1 = _SIMD_ADD(q1, _SIMD_FMSUBADD(h2_real, y1, _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
1682
1683
1684
1685
1686
#else
          q1 = _SIMD_ADD(q1, _SIMD_ADDSUB( _SIMD_MUL(h2_real, y1), _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
          tmp2 = _SIMD_MUL(h2_imag, y2);
#ifdef __ELPA_USE_FMA__
1687
          q2 = _SIMD_ADD(q2, _SIMD_FMSUBADD(h2_real, y2, _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
1688
1689
1690
1691
1692
1693
#else
          q2 = _SIMD_ADD(q2, _SIMD_ADDSUB( _SIMD_MUL(h2_real, y2), _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

          tmp3 = _SIMD_MUL(h2_imag, y3);
#ifdef __ELPA_USE_FMA__
1694
          q3 = _SIMD_ADD(q3, _SIMD_FMSUBADD(h2_real, y3, _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
1695
1696
1697
1698
1699
#else
          q3 = _SIMD_ADD(q3, _SIMD_ADDSUB( _SIMD_MUL(h2_real, y3), _SIMD_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
          tmp4 = _SIMD_MUL(h2_imag, y4);
#ifdef __ELPA_USE_FMA__
1700
          q4 = _SIMD_ADD(q4, _SIMD_FMSUBADD(h2_real, y4, _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
1701
1702
1703
1704
1705
1706
#else
          q4 = _SIMD_ADD(q4, _SIMD_ADDSUB( _SIMD_MUL(h2_real, y4), _SIMD_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif

          tmp5 = _SIMD_MUL(h2_imag, y5);
#ifdef __ELPA_USE_FMA__
1707
          q5 = _SIMD_ADD(q5, _SIMD_FMSUBADD(h2_real, y5, _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
1708
1709
1710
1711
1712
#else
          q5 = _SIMD_ADD(q5, _SIMD_ADDSUB( _SIMD_MUL(h2_real, y5), _SIMD_SHUFFLE(tmp5, tmp5, _SHUFFLE)));
#endif
          tmp6 = _SIMD_MUL(h2_imag, y6);
#ifdef __ELPA_USE_FMA__
1713
          q6 = _SIMD_ADD(q6, _SIMD_FMSUBADD(h2_real, y6, _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
#else
          q6 = _SIMD_ADD(q6, _SIMD_ADDSUB( _SIMD_MUL(h2_real, y6), _SIMD_SHUFFLE(tmp6, tmp6, _SHUFFLE)));
#endif

#endif /* BLOCK2 */


         _SIMD_STORE(&q_dbl[(2*i*ldq)+0], q1);
         _SIMD_STORE(&q_dbl[(2*i*ldq)+offset], q2);
         _SIMD_STORE(&q_dbl[(2*i*ldq)+2*offset], q3);
         _SIMD_STORE(&q_dbl[(2*i*ldq)+3*offset], q4);
         _SIMD_STORE(&q_dbl[(2*i*ldq)+4*offset], q5);
         _SIMD_STORE(&q_dbl[(2*i*ldq)+5*offset], q6);
    }
#ifdef BLOCK2
1729
1730

#if VEC_SET == SSE_128     
1731
1732
1733
1734
1735
1736
1737
1738
#ifdef DOUBLE_PRECISION_COMPLEX
     h1_real = _mm_loaddup_pd(&hh_dbl[(nb-1)*2]);
     h1_imag = _mm_loaddup_pd(&hh_dbl[((nb-1)*2)+1]);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
     h1_real = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[(nb-1)*2]) )));
     h1_imag = _mm_moveldup_ps(_mm_castpd_ps(_mm_loaddup_pd( (double *)(&hh_dbl[((nb-1)*2)+1]) )));
#endif
1739
1740
1741
1742
1743
1744
#endif /* VEC_SET == SSE_128 */

#if VEC_SET == AVX_256
     h1_real = _SIMD_BROADCAST(&hh_dbl[(nb-1)*2]);
     h1_imag = _SIMD_BROADCAST(&hh_dbl[((nb-1)*2)+1]);
#endif /* VEC_SET == AVX_256 */
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754

     q1 = _SIMD_LOAD(&q_dbl[(2*nb*ldq)+0]);
     q2 = _SIMD_LOAD(&q_dbl[(2*nb*ldq)+offset]);
     q3 = _SIMD_LOAD(&q_dbl[(2*nb*ldq)+2*offset]);
     q4 = _SIMD_LOAD(&q_dbl[(2*nb*ldq)+3*offset]);
     q5 = _SIMD_LOAD(&q_dbl[(2*nb*ldq)+4*offset]);
     q6 = _SIMD_LOAD(&q_dbl[(2*nb*ldq)+5*offset]);

     tmp1 = _SIMD_MUL(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
1755
     q1 = _SIMD_ADD(q1, _SIMD_FMSUBADD(h1_real, x1, _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
1756
1757
1758
1759
1760
#else
     q1 = _SIMD_ADD(q1, _SIMD_ADDSUB( _SIMD_MUL(h1_real, x1), _SIMD_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
     tmp2 = _SIMD_MUL(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
1761
     q2 = _SIMD_ADD(q2, _SIMD_FMSUBADD(h1_real, x2, _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
1762
1763
1764
1765
1766
1767
#else
     q2 = _SIMD_ADD(q2, _SIMD_ADDSUB( _SIMD_MUL(h1_real, x2), _SIMD_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

     tmp3 = _SIMD_MUL(h1_imag, x3);
#ifdef __ELPA_USE_FMA__