complex_avx-avx2_2hv_template.c 53.6 KB
Newer Older
1 2 3 4 5 6
//    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
7
//	Rechenzentrum Garching der Max-Planck-Gesellschaft (RZG),
8
//    - Bergische Universität Wuppertal, Lehrstuhl für angewandte
9
//	Informatik,
10
//    - Technische Universität München, Lehrstuhl für Informatik mit
11
//	Schwerpunkt Wissenschaftliches Rechnen ,
12 13
//    - Fritz-Haber-Institut, Berlin, Abt. Theorie,
//    - Max-Plack-Institut für Mathematik in den Naturwissenschaften,
14 15
//	Leipzig, Abt. Komplexe Strukutren in Biologie und Kognition,
//	and
16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35
//    - 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
36
//    along with ELPA.	If not, see <http://www.gnu.org/licenses/>
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
//
//    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.
//
//
// --------------------------------------------------------------------------------------------------
//
// This file contains the compute intensive kernels for the Householder transformations.
// It should be compiled with the highest possible optimization level.
//
// On Intel Nehalem or Intel Westmere or AMD Magny Cours use -O3 -msse3
// On Intel Sandy Bridge use -O3 -mavx
//
// Copyright of the original code rests with the authors inside the ELPA
// consortium. The copyright of any additional modifications shall rest
// with their original authors, but shall adhere to the licensing terms
// distributed along with the original code in the file "COPYING".
//
// Author: Alexander Heinecke (alexander.heinecke@mytum.de)
// Adapted for building a shared-library by Andreas Marek, MPCDF (andreas.marek@mpcdf.mpg.de)
// --------------------------------------------------------------------------------------------------
#include "config-f90.h"

#include <complex.h>
#include <x86intrin.h>
66
#include <stdio.h>
67
#include <stdlib.h>
68 69 70 71 72 73 74 75 76 77 78 79 80

#define __forceinline __attribute__((always_inline))

#ifdef DOUBLE_PRECISION_COMPLEX
#define offset 4
#define __AVX_DATATYPE __m256d
#define _AVX_LOAD _mm256_load_pd
#define _AVX_STORE _mm256_store_pd
#define _AVX_ADD _mm256_add_pd
#define _AVX_MUL _mm256_mul_pd
#define _AVX_ADDSUB _mm256_addsub_pd
#define _AVX_XOR _mm256_xor_pd
#define _AVX_BROADCAST _mm256_broadcast_sd
Lorenz Huedepohl's avatar
Lorenz Huedepohl committed
81
#define _AVX_SET1 _mm256_set1_pd
82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114
#define _AVX_SHUFFLE _mm256_shuffle_pd
#define _SHUFFLE 0x5
#define _CAST _mm256_castpd256_pd128

#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 _AVX_FMADDSUB _mm256_FMADDSUB_pd
#define _AVX_FMSUBADD _mm256_FMSUBADD_pd
#endif
#endif /* DOUBLE_PRECISION_COMPLEX */

#ifdef SINGLE_PRECISION_COMPLEX
#define offset 8
#define __AVX_DATATYPE __m256
#define _AVX_LOAD _mm256_load_ps
#define _AVX_STORE _mm256_store_ps
#define _AVX_ADD _mm256_add_ps
#define _AVX_MUL _mm256_mul_ps
#define _AVX_ADDSUB _mm256_addsub_ps
#define _AVX_XOR _mm256_xor_ps
#define _AVX_BROADCAST _mm256_broadcast_ss
Lorenz Huedepohl's avatar
Lorenz Huedepohl committed
115
#define _AVX_SET1 _mm256_set1_ps
116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146
#define _AVX_SHUFFLE _mm256_shuffle_ps
#define _SHUFFLE 0xb1
#define _CAST _mm256_castps256_ps128
#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 _AVX_FMADDSUB _mm256_FMADDSUB_ps
#define _AVX_FMSUBADD _mm256_FMSUBADD_ps
#endif
#endif /* SINGLE_PRECISION_COMPLEX */

#ifdef DOUBLE_PRECISION_COMPLEX
//Forward declaration
static __forceinline void hh_trafo_complex_kernel_8_AVX_2hv_double(double complex* q, double complex* hh, int nb, int ldq, int ldh, double complex s);
static __forceinline void hh_trafo_complex_kernel_6_AVX_2hv_double(double complex* q, double complex* hh, int nb, int ldq, int ldh, double complex s);
static __forceinline void hh_trafo_complex_kernel_4_AVX_2hv_double(double complex* q, double complex* hh, int nb, int ldq, int ldh, double complex s);
static __forceinline void hh_trafo_complex_kernel_2_AVX_2hv_double(double complex* q, double complex* hh, int nb, int ldq, int ldh, double complex s);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
//Forward declaration
147 148
static __forceinline void hh_trafo_complex_kernel_16_AVX_2hv_single(float complex* q, float complex* hh, int nb, int ldq, int ldh, float complex s, float complex s1);
static __forceinline void hh_trafo_complex_kernel_12_AVX_2hv_single(float complex* q, float complex* hh, int nb, int ldq, int ldh, float complex s, float complex s1);
149 150 151 152 153 154 155 156 157
static __forceinline void hh_trafo_complex_kernel_8_AVX_2hv_single(float complex* q, float complex* hh, int nb, int ldq, int ldh, float complex s, float complex s1);
static __forceinline void hh_trafo_complex_kernel_4_AVX_2hv_single(float complex* q, float complex* hh, int nb, int ldq, int ldh, float complex s, float complex s1);
#endif

#ifdef DOUBLE_PRECISION_COMPLEX
/*
!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) &
158 159 160 161 162 163
!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)
164 165 166 167 168 169 170 171 172 173
!f>   end subroutine
!f> end interface
!f>#endif
*/
#endif
#ifdef SINGLE_PRECISION_COMPLEX
/*
!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) &
174 175 176 177 178 179
!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)
180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197
!f>   end subroutine
!f> end interface
!f>#endif
*/
#endif

#ifdef DOUBLE_PRECISION_COMPLEX
void double_hh_trafo_complex_avx_avx2_2hv_double(double complex* q, double complex* hh, int* pnb, int* pnq, int* pldq, int* pldh)
#endif
#ifdef SINGLE_PRECISION_COMPLEX
void double_hh_trafo_complex_avx_avx2_2hv_single(float complex* q, float complex* hh, int* pnb, int* pnq, int* pldq, int* pldh)
#endif
{
	int i;
	int nb = *pnb;
	int nq = *pldq;
	int ldq = *pldq;
	int ldh = *pldh;
198 199 200
	int worked_on;

	worked_on = 0;
201 202 203 204 205 206 207 208 209 210 211 212 213 214

#ifdef DOUBLE_PRECISION_COMPLEX
	double complex s = conj(hh[(ldh)+1])*1.0;
#endif
#ifdef SINGLE_PRECISION_COMPLEX
	float complex s = conj(hh[(ldh)+1])*1.0f;
#endif

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

#ifdef DOUBLE_PRECISION_COMPLEX
215 216
	for (i = 0; i < nq-6; i+=8)
	{
217
		hh_trafo_complex_kernel_8_AVX_2hv_double(&q[i], hh, nb, ldq, ldh, s);
218
		worked_on += 8;
219
	}
220 221
#endif
#ifdef SINGLE_PRECISION_COMPLEX
222 223 224
	for (i = 0; i < nq-12; i+=16)
	{
		hh_trafo_complex_kernel_16_AVX_2hv_single(&q[i], hh, nb, ldq, ldh, s , s);
225
		worked_on += 16;
226
	}
227
#endif
228
	if (nq-i == 0) {
229
	  return;
230 231
	}
#ifdef DOUBLE_PRECISION_COMPLEX
232
	if (nq-i == 6) {
233 234
		hh_trafo_complex_kernel_6_AVX_2hv_double(&q[i], hh, nb, ldq, ldh, s);
		worked_on += 6;
235
	}
236 237
#endif
#ifdef SINGLE_PRECISION_COMPLEX
238
	if (nq-i == 12) {
239 240
		hh_trafo_complex_kernel_12_AVX_2hv_single(&q[i], hh, nb, ldq, ldh, s, s);
		worked_on += 12;
241
	}
242 243
#endif

244
#ifdef DOUBLE_PRECISION_COMPLEX
245
	if (nq-i == 4) {
246
		hh_trafo_complex_kernel_4_AVX_2hv_double(&q[i], hh, nb, ldq, ldh, s);
247
		worked_on += 4;
248
	}
249 250
#endif
#ifdef SINGLE_PRECISION_COMPLEX
251
	if (nq-i == 8) {
252 253
		hh_trafo_complex_kernel_8_AVX_2hv_single(&q[i], hh, nb, ldq, ldh, s, s);
		worked_on += 8;
254
	}
255 256
#endif
#ifdef DOUBLE_PRECISION_COMPLEX
257
	if (nq-i == 2) {
258 259 260 261 262
		hh_trafo_complex_kernel_2_AVX_2hv_double(&q[i], hh, nb, ldq, ldh, s);
		worked_on += 2;
	}
#endif
#ifdef SINGLE_PRECISION_COMPLEX
263
	if (nq-i == 4) {
264
		hh_trafo_complex_kernel_4_AVX_2hv_single(&q[i], hh, nb, ldq, ldh, s, s);
265 266
		worked_on += 4;
	}
267
#endif
268
	if (worked_on != nq) {
269 270
		printf("Error in complex avx-avx2 BLOCK 2 kernel \n");
		abort();
271 272 273 274 275 276 277
	}
}

#ifdef DOUBLE_PRECISION_COMPLEX
static __forceinline void hh_trafo_complex_kernel_8_AVX_2hv_double(double complex* q, double complex* hh, int nb, int ldq, int ldh, double complex s)
#endif
#ifdef SINGLE_PRECISION_COMPLEX
278
static __forceinline void hh_trafo_complex_kernel_16_AVX_2hv_single(float complex* q, float complex* hh, int nb, int ldq, int ldh, float complex s, float complex s1)
279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353
#endif
{

#ifdef DOUBLE_PRECISION_COMPLEX
	double* q_dbl = (double*)q;
	double* hh_dbl = (double*)hh;
	double* s_dbl = (double*)(&s);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
	float* q_dbl = (float*)q;
	float* hh_dbl = (float*)hh;
	float* s_dbl = (float*)(&s);
#endif
	__AVX_DATATYPE x1, x2, x3, x4;
	__AVX_DATATYPE y1, y2, y3, y4;
	__AVX_DATATYPE q1, q2, q3, q4;
	__AVX_DATATYPE h1_real, h1_imag, h2_real, h2_imag;
	__AVX_DATATYPE tmp1, tmp2, tmp3, tmp4;
	int i=0;

#ifdef DOUBLE_PRECISION_COMPLEX
	__AVX_DATATYPE sign = (__AVX_DATATYPE)_mm256_set_epi64x(0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
	__AVX_DATATYPE sign = (__AVX_DATATYPE)_mm256_set_epi32(0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000);
#endif

	x1 = _AVX_LOAD(&q_dbl[(2*ldq)+0]);
	x2 = _AVX_LOAD(&q_dbl[(2*ldq)+offset]);
	x3 = _AVX_LOAD(&q_dbl[(2*ldq)+2*offset]);
	x4 = _AVX_LOAD(&q_dbl[(2*ldq)+3*offset]);
	h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+1)*2]);
	h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+1)*2)+1]);
#ifndef __ELPA_USE_FMA__
	// conjugate
	h2_imag = _AVX_XOR(h2_imag, sign);
#endif

	y1 = _AVX_LOAD(&q_dbl[0]);
	y2 = _AVX_LOAD(&q_dbl[offset]);
	y3 = _AVX_LOAD(&q_dbl[2*offset]);
	y4 = _AVX_LOAD(&q_dbl[3*offset]);

	tmp1 = _AVX_MUL(h2_imag, x1);
#ifdef __ELPA_USE_FMA__
	y1 = _AVX_ADD(y1, _AVX_FMSUBADD(h2_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	y1 = _AVX_ADD(y1, _AVX_ADDSUB( _AVX_MUL(h2_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
	tmp2 = _AVX_MUL(h2_imag, x2);
#ifdef __ELPA_USE_FMA__
	y2 = _AVX_ADD(y2, _AVX_FMSUBADD(h2_real, x2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
	y2 = _AVX_ADD(y2, _AVX_ADDSUB( _AVX_MUL(h2_real, x2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

	tmp3 = _AVX_MUL(h2_imag, x3);
#ifdef __ELPA_USE_FMA__
	y3 = _AVX_ADD(y3, _AVX_FMSUBADD(h2_real, x3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
	y3 = _AVX_ADD(y3, _AVX_ADDSUB( _AVX_MUL(h2_real, x3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
	tmp4 = _AVX_MUL(h2_imag, x4);
#ifdef __ELPA_USE_FMA__
	y4 = _AVX_ADD(y4, _AVX_FMSUBADD(h2_real, x4, _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#else
	y4 = _AVX_ADD(y4, _AVX_ADDSUB( _AVX_MUL(h2_real, x4), _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif

	for (i = 2; i < nb; i++)
	{
		q1 = _AVX_LOAD(&q_dbl[(2*i*ldq)+0]);
		q2 = _AVX_LOAD(&q_dbl[(2*i*ldq)+offset]);
		q3 = _AVX_LOAD(&q_dbl[(2*i*ldq)+2*offset]);
		q4 = _AVX_LOAD(&q_dbl[(2*i*ldq)+3*offset]);
354

355
		h1_real = _AVX_BROADCAST(&hh_dbl[(i-1)*2]);
356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501
		h1_imag = _AVX_BROADCAST(&hh_dbl[((i-1)*2)+1]);
#ifndef __ELPA_USE_FMA__
		// conjugate
		h1_imag = _AVX_XOR(h1_imag, sign);
#endif

		tmp1 = _AVX_MUL(h1_imag, q1);
#ifdef __ELPA_USE_FMA__
		x1 = _AVX_ADD(x1, _AVX_FMSUBADD(h1_real, q1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		x1 = _AVX_ADD(x1, _AVX_ADDSUB( _AVX_MUL(h1_real, q1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
		tmp2 = _AVX_MUL(h1_imag, q2);
#ifdef __ELPA_USE_FMA__
		x2 = _AVX_ADD(x2, _AVX_FMSUBADD(h1_real, q2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
		x2 = _AVX_ADD(x2, _AVX_ADDSUB( _AVX_MUL(h1_real, q2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

		tmp3 = _AVX_MUL(h1_imag, q3);
#ifdef __ELPA_USE_FMA__
		x3 = _AVX_ADD(x3, _AVX_FMSUBADD(h1_real, q3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
		x3 = _AVX_ADD(x3, _AVX_ADDSUB( _AVX_MUL(h1_real, q3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
		tmp4 = _AVX_MUL(h1_imag, q4);
#ifdef __ELPA_USE_FMA__
		x4 = _AVX_ADD(x4, _AVX_FMSUBADD(h1_real, q4, _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#else
		x4 = _AVX_ADD(x4, _AVX_ADDSUB( _AVX_MUL(h1_real, q4), _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif

		h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+i)*2]);
		h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+i)*2)+1]);
#ifndef __ELPA_USE_FMA__
		// conjugate
		h2_imag = _AVX_XOR(h2_imag, sign);
#endif

		tmp1 = _AVX_MUL(h2_imag, q1);
#ifdef __ELPA_USE_FMA__
		y1 = _AVX_ADD(y1, _AVX_FMSUBADD(h2_real, q1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		y1 = _AVX_ADD(y1, _AVX_ADDSUB( _AVX_MUL(h2_real, q1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
		tmp2 = _AVX_MUL(h2_imag, q2);
#ifdef __ELPA_USE_FMA__
		y2 = _AVX_ADD(y2, _AVX_FMSUBADD(h2_real, q2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
		y2 = _AVX_ADD(y2, _AVX_ADDSUB( _AVX_MUL(h2_real, q2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

		tmp3 = _AVX_MUL(h2_imag, q3);
#ifdef __ELPA_USE_FMA__
		y3 = _AVX_ADD(y3, _AVX_FMSUBADD(h2_real, q3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
		y3 = _AVX_ADD(y3, _AVX_ADDSUB( _AVX_MUL(h2_real, q3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
		tmp4 = _AVX_MUL(h2_imag, q4);
#ifdef __ELPA_USE_FMA__
		y4 = _AVX_ADD(y4, _AVX_FMSUBADD(h2_real, q4, _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#else
		y4 = _AVX_ADD(y4, _AVX_ADDSUB( _AVX_MUL(h2_real, q4), _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif
	}

	h1_real = _AVX_BROADCAST(&hh_dbl[(nb-1)*2]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[((nb-1)*2)+1]);
#ifndef __ELPA_USE_FMA__
	// conjugate
	h1_imag = _AVX_XOR(h1_imag, sign);
#endif

	q1 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+0]);
	q2 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+offset]);
	q3 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+2*offset]);
	q4 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+3*offset]);

	tmp1 = _AVX_MUL(h1_imag, q1);
#ifdef __ELPA_USE_FMA__
	x1 = _AVX_ADD(x1, _AVX_FMSUBADD(h1_real, q1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	x1 = _AVX_ADD(x1, _AVX_ADDSUB( _AVX_MUL(h1_real, q1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
	tmp2 = _AVX_MUL(h1_imag, q2);
#ifdef __ELPA_USE_FMA__
	x2 = _AVX_ADD(x2, _AVX_FMSUBADD(h1_real, q2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
	x2 = _AVX_ADD(x2, _AVX_ADDSUB( _AVX_MUL(h1_real, q2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

	tmp3 = _AVX_MUL(h1_imag, q3);
#ifdef __ELPA_USE_FMA__
	x3 = _AVX_ADD(x3, _AVX_FMSUBADD(h1_real, q3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
	x3 = _AVX_ADD(x3, _AVX_ADDSUB( _AVX_MUL(h1_real, q3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
	tmp4 = _AVX_MUL(h1_imag, q4);
#ifdef __ELPA_USE_FMA__
	x4 = _AVX_ADD(x4, _AVX_FMSUBADD(h1_real, q4, _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#else
	x4 = _AVX_ADD(x4, _AVX_ADDSUB( _AVX_MUL(h1_real, q4), _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif

	h1_real = _AVX_BROADCAST(&hh_dbl[0]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[1]);
	h1_real = _AVX_XOR(h1_real, sign);
	h1_imag = _AVX_XOR(h1_imag, sign);

	tmp1 = _AVX_MUL(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
	x1 = _AVX_FMADDSUB(h1_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#else
	x1 = _AVX_ADDSUB( _AVX_MUL(h1_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#endif
	tmp2 = _AVX_MUL(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
	x2 = _AVX_FMADDSUB(h1_real, x2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE));
#else
	x2 = _AVX_ADDSUB( _AVX_MUL(h1_real, x2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE));
#endif

	tmp3 = _AVX_MUL(h1_imag, x3);
#ifdef __ELPA_USE_FMA__
	x3 = _AVX_FMADDSUB(h1_real, x3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE));
#else
	x3 = _AVX_ADDSUB( _AVX_MUL(h1_real, x3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE));
#endif
	tmp4 = _AVX_MUL(h1_imag, x4);
#ifdef __ELPA_USE_FMA__
	x4 = _AVX_FMADDSUB(h1_real, x4, _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE));
#else
	x4 = _AVX_ADDSUB( _AVX_MUL(h1_real, x4), _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE));
#endif

	h1_real = _AVX_BROADCAST(&hh_dbl[ldh*2]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[(ldh*2)+1]);
	h2_real = _AVX_BROADCAST(&hh_dbl[ldh*2]);
	h2_imag = _AVX_BROADCAST(&hh_dbl[(ldh*2)+1]);

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

#ifdef DOUBLE_PRECISION_COMPLEX
502
	tmp2 = _mm256_set_pd(s_dbl[1], s_dbl[0], s_dbl[1], s_dbl[0]);
503 504
#endif
#ifdef SINGLE_PRECISION_COMPLEX
505 506
	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]);
507 508 509 510 511 512 513 514 515
#endif

	tmp1 = _AVX_MUL(h2_imag, tmp2);
#ifdef __ELPA_USE_FMA__
	tmp2 = _AVX_FMADDSUB(h2_real, tmp2, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#else
	tmp2 = _AVX_ADDSUB( _AVX_MUL(h2_real, tmp2), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#endif

516 517
	h2_real = _AVX_SET1(tmp2[0]);
	h2_imag = _AVX_SET1(tmp2[1]);
518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598

	tmp1 = _AVX_MUL(h1_imag, y1);
#ifdef __ELPA_USE_FMA__
	y1 = _AVX_FMADDSUB(h1_real, y1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#else
	y1 = _AVX_ADDSUB( _AVX_MUL(h1_real, y1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#endif
	tmp2 = _AVX_MUL(h1_imag, y2);
#ifdef __ELPA_USE_FMA__
	y2 = _AVX_FMADDSUB(h1_real, y2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE));
#else
	y2 = _AVX_ADDSUB( _AVX_MUL(h1_real, y2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE));
#endif

	tmp3 = _AVX_MUL(h1_imag, y3);
#ifdef __ELPA_USE_FMA__
	y3 = _AVX_FMADDSUB(h1_real, y3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE));
#else
	y3 = _AVX_ADDSUB( _AVX_MUL(h1_real, y3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE));
#endif
	tmp4 = _AVX_MUL(h1_imag, y4);
#ifdef __ELPA_USE_FMA__
	y4 = _AVX_FMADDSUB(h1_real, y4, _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE));
#else
	y4 = _AVX_ADDSUB( _AVX_MUL(h1_real, y4), _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE));
#endif

	tmp1 = _AVX_MUL(h2_imag, x1);
#ifdef __ELPA_USE_FMA__
	y1 = _AVX_ADD(y1, _AVX_FMADDSUB(h2_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	y1 = _AVX_ADD(y1, _AVX_ADDSUB( _AVX_MUL(h2_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
	tmp2 = _AVX_MUL(h2_imag, x2);
#ifdef __ELPA_USE_FMA__
	y2 = _AVX_ADD(y2, _AVX_FMADDSUB(h2_real, x2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
	y2 = _AVX_ADD(y2, _AVX_ADDSUB( _AVX_MUL(h2_real, x2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

	tmp3 = _AVX_MUL(h2_imag, x3);
#ifdef __ELPA_USE_FMA__
	y3 = _AVX_ADD(y3, _AVX_FMADDSUB(h2_real, x3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
	y3 = _AVX_ADD(y3, _AVX_ADDSUB( _AVX_MUL(h2_real, x3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
	tmp4 = _AVX_MUL(h2_imag, x4);
#ifdef __ELPA_USE_FMA__
	y4 = _AVX_ADD(y4, _AVX_FMADDSUB(h2_real, x4, _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#else
	y4 = _AVX_ADD(y4, _AVX_ADDSUB( _AVX_MUL(h2_real, x4), _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif

	q1 = _AVX_LOAD(&q_dbl[0]);
	q2 = _AVX_LOAD(&q_dbl[offset]);
	q3 = _AVX_LOAD(&q_dbl[2*offset]);
	q4 = _AVX_LOAD(&q_dbl[3*offset]);

	q1 = _AVX_ADD(q1, y1);
	q2 = _AVX_ADD(q2, y2);
	q3 = _AVX_ADD(q3, y3);
	q4 = _AVX_ADD(q4, y4);


	_AVX_STORE(&q_dbl[0], q1);
	_AVX_STORE(&q_dbl[offset], q2);
	_AVX_STORE(&q_dbl[2*offset], q3);
	_AVX_STORE(&q_dbl[3*offset], q4);

	h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+1)*2]);
	h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+1)*2)+1]);

	q1 = _AVX_LOAD(&q_dbl[(ldq*2)+0]);
	q2 = _AVX_LOAD(&q_dbl[(ldq*2)+offset]);
	q3 = _AVX_LOAD(&q_dbl[(ldq*2)+2*offset]);
	q4 = _AVX_LOAD(&q_dbl[(ldq*2)+3*offset]);

	q1 = _AVX_ADD(q1, x1);
	q2 = _AVX_ADD(q2, x2);
	q3 = _AVX_ADD(q3, x3);
	q4 = _AVX_ADD(q4, x4);
599

600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629
	tmp1 = _AVX_MUL(h2_imag, y1);
#ifdef __ELPA_USE_FMA__
	q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h2_real, y1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h2_real, y1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
	tmp2 = _AVX_MUL(h2_imag, y2);
#ifdef __ELPA_USE_FMA_
	q2 = _AVX_ADD(q2, _AVX_FMADDSUB(h2_real, y2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
	q2 = _AVX_ADD(q2, _AVX_ADDSUB( _AVX_MUL(h2_real, y2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

	tmp3 = _AVX_MUL(h2_imag, y3);
#ifdef __ELPA_USE_FMA__
	q3 = _AVX_ADD(q3, _AVX_FMADDSUB(h2_real, y3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
	q3 = _AVX_ADD(q3, _AVX_ADDSUB( _AVX_MUL(h2_real, y3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
	tmp4 = _AVX_MUL(h2_imag, y4);
#ifdef __ELPA_USE_FMA__
	q4 = _AVX_ADD(q4, _AVX_FMADDSUB(h2_real, y4, _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#else
	q4 = _AVX_ADD(q4, _AVX_ADDSUB( _AVX_MUL(h2_real, y4), _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif

	_AVX_STORE(&q_dbl[(ldq*2)+0], q1);
	_AVX_STORE(&q_dbl[(ldq*2)+offset], q2);
	_AVX_STORE(&q_dbl[(ldq*2)+2*offset], q3);
	_AVX_STORE(&q_dbl[(ldq*2)+3*offset], q4);
630

631 632 633 634 635 636 637
	for (i = 2; i < nb; i++)
	{

		q1 = _AVX_LOAD(&q_dbl[(2*i*ldq)+0]);
		q2 = _AVX_LOAD(&q_dbl[(2*i*ldq)+offset]);
		q3 = _AVX_LOAD(&q_dbl[(2*i*ldq)+2*offset]);
		q4 = _AVX_LOAD(&q_dbl[(2*i*ldq)+3*offset]);
638

639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743
		h1_real = _AVX_BROADCAST(&hh_dbl[(i-1)*2]);
		h1_imag = _AVX_BROADCAST(&hh_dbl[((i-1)*2)+1]);

		tmp1 = _AVX_MUL(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
		q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h1_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h1_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
		tmp2 = _AVX_MUL(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
		q2 = _AVX_ADD(q2, _AVX_FMADDSUB(h1_real, x2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
		q2 = _AVX_ADD(q2, _AVX_ADDSUB( _AVX_MUL(h1_real, x2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

		tmp3 = _AVX_MUL(h1_imag, x3);
#ifdef __ELPA_USE_FMA__
		q3 = _AVX_ADD(q3, _AVX_FMADDSUB(h1_real, x3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
		q3 = _AVX_ADD(q3, _AVX_ADDSUB( _AVX_MUL(h1_real, x3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
		tmp4 = _AVX_MUL(h1_imag, x4);
#ifdef __ELPA_USE_FMA__
		q4 = _AVX_ADD(q4, _AVX_FMADDSUB(h1_real, x4, _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#else
		q4 = _AVX_ADD(q4, _AVX_ADDSUB( _AVX_MUL(h1_real, x4), _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif

		h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+i)*2]);
		h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+i)*2)+1]);

		tmp1 = _AVX_MUL(h2_imag, y1);
#ifdef __ELPA_USE_FMA__
		q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h2_real, y1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h2_real, y1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
		tmp2 = _AVX_MUL(h2_imag, y2);
#ifdef __ELPA_USE_FMA__
		q2 = _AVX_ADD(q2, _AVX_FMADDSUB(h2_real, y2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
		q2 = _AVX_ADD(q2, _AVX_ADDSUB( _AVX_MUL(h2_real, y2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

		tmp3 = _AVX_MUL(h2_imag, y3);
#ifdef __ELPA_USE_FMA__
		q3 = _AVX_ADD(q3, _AVX_FMADDSUB(h2_real, y3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
		q3 = _AVX_ADD(q3, _AVX_ADDSUB( _AVX_MUL(h2_real, y3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
		tmp4 = _AVX_MUL(h2_imag, y4);
#ifdef __ELPA_USE_FMA__
		q4 = _AVX_ADD(q4, _AVX_FMADDSUB(h2_real, y4, _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#else
		q4 = _AVX_ADD(q4, _AVX_ADDSUB( _AVX_MUL(h2_real, y4), _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif

		_AVX_STORE(&q_dbl[(2*i*ldq)+0], q1);
		_AVX_STORE(&q_dbl[(2*i*ldq)+offset], q2);
		_AVX_STORE(&q_dbl[(2*i*ldq)+2*offset], q3);
		_AVX_STORE(&q_dbl[(2*i*ldq)+3*offset], q4);
	}
	h1_real = _AVX_BROADCAST(&hh_dbl[(nb-1)*2]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[((nb-1)*2)+1]);

	q1 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+0]);
	q2 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+offset]);
	q3 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+2*offset]);
	q4 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+3*offset]);

	tmp1 = _AVX_MUL(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
	q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h1_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h1_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
	tmp2 = _AVX_MUL(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
	q2 = _AVX_ADD(q2, _AVX_FMADDSUB(h1_real, x2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
	q2 = _AVX_ADD(q2, _AVX_ADDSUB( _AVX_MUL(h1_real, x2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

	tmp3 = _AVX_MUL(h1_imag, x3);
#ifdef __ELPA_USE_FMA__
	q3 = _AVX_ADD(q3, _AVX_FMADDSUB(h1_real, x3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
	q3 = _AVX_ADD(q3, _AVX_ADDSUB( _AVX_MUL(h1_real, x3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
	tmp4 = _AVX_MUL(h1_imag, x4);
#ifdef __ELPA_USE_FMA__
	q4 = _AVX_ADD(q4, _AVX_FMADDSUB(h1_real, x4, _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#else
	q4 = _AVX_ADD(q4, _AVX_ADDSUB( _AVX_MUL(h1_real, x4), _AVX_SHUFFLE(tmp4, tmp4, _SHUFFLE)));
#endif

	_AVX_STORE(&q_dbl[(2*nb*ldq)+0], q1);
	_AVX_STORE(&q_dbl[(2*nb*ldq)+offset], q2);
	_AVX_STORE(&q_dbl[(2*nb*ldq)+2*offset], q3);
	_AVX_STORE(&q_dbl[(2*nb*ldq)+3*offset], q4);
}

#ifdef DOUBLE_PRECISION_COMPLEX
static __forceinline void hh_trafo_complex_kernel_6_AVX_2hv_double(double complex* q, double complex* hh, int nb, int ldq, int ldh, double complex s)
744 745 746 747 748
#endif
#ifdef SINGLE_PRECISION_COMPLEX
static __forceinline void hh_trafo_complex_kernel_12_AVX_2hv_single(float complex* q, float complex* hh, int nb, int ldq, int ldh, float complex s, float complex s1)
#endif

749
{
750
#ifdef DOUBLE_PRECISION_COMPLEX
751 752 753
	double* q_dbl = (double*)q;
	double* hh_dbl = (double*)hh;
	double* s_dbl = (double*)(&s);
754 755 756 757 758 759
#endif
#ifdef SINGLE_PRECISION_COMPLEX
	float* q_dbl = (float*)q;
	float* hh_dbl = (float*)hh;
	float* s_dbl = (float*)(&s);
#endif
760 761 762 763 764 765 766
	__AVX_DATATYPE x1, x2, x3;
	__AVX_DATATYPE y1, y2, y3;
	__AVX_DATATYPE q1, q2, q3;
	__AVX_DATATYPE h1_real, h1_imag, h2_real, h2_imag;
	__AVX_DATATYPE tmp1, tmp2, tmp3;
	int i=0;

767
#ifdef DOUBLE_PRECISION_COMPLEX
768
	__AVX_DATATYPE sign = (__AVX_DATATYPE)_mm256_set_epi64x(0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000);
769 770 771 772
#endif
#ifdef SINGLE_PRECISION_COMPLEX
	__AVX_DATATYPE sign = (__AVX_DATATYPE)_mm256_set_epi32(0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000);
#endif
773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930
	x1 = _AVX_LOAD(&q_dbl[(2*ldq)+0]);
	x2 = _AVX_LOAD(&q_dbl[(2*ldq)+offset]);
	x3 = _AVX_LOAD(&q_dbl[(2*ldq)+2*offset]);

	h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+1)*2]);
	h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+1)*2)+1]);
#ifndef __ELPA_USE_FMA__
	// conjugate
	h2_imag = _AVX_XOR(h2_imag, sign);
#endif

	y1 = _AVX_LOAD(&q_dbl[0]);
	y2 = _AVX_LOAD(&q_dbl[offset]);
	y3 = _AVX_LOAD(&q_dbl[2*offset]);

	tmp1 = _AVX_MUL(h2_imag, x1);
#ifdef __ELPA_USE_FMA__
	y1 = _AVX_ADD(y1, _AVX_FMSUBADD(h2_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	y1 = _AVX_ADD(y1, _AVX_ADDSUB( _AVX_MUL(h2_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
	tmp2 = _AVX_MUL(h2_imag, x2);
#ifdef __ELPA_USE_FMA__
	y2 = _AVX_ADD(y2, _AVX_FMSUBADD(h2_real, x2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
	y2 = _AVX_ADD(y2, _AVX_ADDSUB( _AVX_MUL(h2_real, x2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif
	tmp3 = _AVX_MUL(h2_imag, x3);
#ifdef __ELPA_USE_FMA__
	y3 = _AVX_ADD(y3, _AVX_FMSUBADD(h2_real, x3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
	y3 = _AVX_ADD(y3, _AVX_ADDSUB( _AVX_MUL(h2_real, x3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif

	for (i = 2; i < nb; i++)
	{
		q1 = _AVX_LOAD(&q_dbl[(2*i*ldq)+0]);
		q2 = _AVX_LOAD(&q_dbl[(2*i*ldq)+offset]);
		q3 = _AVX_LOAD(&q_dbl[(2*i*ldq)+2*offset]);

		h1_real = _AVX_BROADCAST(&hh_dbl[(i-1)*2]);
		h1_imag = _AVX_BROADCAST(&hh_dbl[((i-1)*2)+1]);
#ifndef __ELPA_USE_FMA__
		// conjugate
		h1_imag = _AVX_XOR(h1_imag, sign);
#endif

		tmp1 = _AVX_MUL(h1_imag, q1);
#ifdef __ELPA_USE_FMA__
		x1 = _AVX_ADD(x1, _AVX_FMSUBADD(h1_real, q1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		x1 = _AVX_ADD(x1, _AVX_ADDSUB( _AVX_MUL(h1_real, q1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
		tmp2 = _AVX_MUL(h1_imag, q2);
#ifdef __ELPA_USE_FMA__
		x2 = _AVX_ADD(x2, _AVX_FMSUBADD(h1_real, q2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
		x2 = _AVX_ADD(x2, _AVX_ADDSUB( _AVX_MUL(h1_real, q2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif
		tmp3 = _AVX_MUL(h1_imag, q3);
#ifdef __ELPA_USE_FMA__
		x3 = _AVX_ADD(x3, _AVX_FMSUBADD(h1_real, q3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
		x3 = _AVX_ADD(x3, _AVX_ADDSUB( _AVX_MUL(h1_real, q3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif

		h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+i)*2]);
		h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+i)*2)+1]);
#ifndef __ELPA_USE_FMA__
		// conjugate
		h2_imag = _AVX_XOR(h2_imag, sign);
#endif

		tmp1 = _AVX_MUL(h2_imag, q1);
#ifdef __ELPA_USE_FMA__
		y1 = _AVX_ADD(y1, _AVX_FMSUBADD(h2_real, q1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		y1 = _AVX_ADD(y1, _AVX_ADDSUB( _AVX_MUL(h2_real, q1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
		tmp2 = _AVX_MUL(h2_imag, q2);
#ifdef __ELPA_USE_FMA__
		y2 = _AVX_ADD(y2, _AVX_FMSUBADD(h2_real, q2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
		y2 = _AVX_ADD(y2, _AVX_ADDSUB( _AVX_MUL(h2_real, q2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif
		tmp3 = _AVX_MUL(h2_imag, q3);
#ifdef __ELPA_USE_FMA__
		y3 = _AVX_ADD(y3, _AVX_FMSUBADD(h2_real, q3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
		y3 = _AVX_ADD(y3, _AVX_ADDSUB( _AVX_MUL(h2_real, q3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif
	}

	h1_real = _AVX_BROADCAST(&hh_dbl[(nb-1)*2]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[((nb-1)*2)+1]);
#ifndef __ELPA_USE_FMA__
	// conjugate
	h1_imag = _AVX_XOR(h1_imag, sign);
#endif

	q1 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+0]);
	q2 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+offset]);
	q3 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+2*offset]);

	tmp1 = _AVX_MUL(h1_imag, q1);
#ifdef __ELPA_USE_FMA__
	x1 = _AVX_ADD(x1, _AVX_FMSUBADD(h1_real, q1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	x1 = _AVX_ADD(x1, _AVX_ADDSUB( _AVX_MUL(h1_real, q1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
	tmp2 = _AVX_MUL(h1_imag, q2);
#ifdef __ELPA_USE_FMA__
	x2 = _AVX_ADD(x2, _AVX_FMSUBADD(h1_real, q2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
	x2 = _AVX_ADD(x2, _AVX_ADDSUB( _AVX_MUL(h1_real, q2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif
	tmp3 = _AVX_MUL(h1_imag, q3);
#ifdef __ELPA_USE_FMA__
	x3 = _AVX_ADD(x3, _AVX_FMSUBADD(h1_real, q3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
	x3 = _AVX_ADD(x3, _AVX_ADDSUB( _AVX_MUL(h1_real, q3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif

	h1_real = _AVX_BROADCAST(&hh_dbl[0]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[1]);
	h1_real = _AVX_XOR(h1_real, sign);
	h1_imag = _AVX_XOR(h1_imag, sign);

	tmp1 = _AVX_MUL(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
	x1 = _AVX_FMADDSUB(h1_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#else
	x1 = _AVX_ADDSUB( _AVX_MUL(h1_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#endif
	tmp2 = _AVX_MUL(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
	x2 = _AVX_FMADDSUB(h1_real, x2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE));
#else
	x2 = _AVX_ADDSUB( _AVX_MUL(h1_real, x2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE));
#endif
	tmp3 = _AVX_MUL(h1_imag, x3);
#ifdef __ELPA_USE_FMA__
	x3 = _AVX_FMADDSUB(h1_real, x3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE));
#else
	x3 = _AVX_ADDSUB( _AVX_MUL(h1_real, x3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE));
#endif

	h1_real = _AVX_BROADCAST(&hh_dbl[ldh*2]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[(ldh*2)+1]);
	h2_real = _AVX_BROADCAST(&hh_dbl[ldh*2]);
	h2_imag = _AVX_BROADCAST(&hh_dbl[(ldh*2)+1]);

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

#ifdef DOUBLE_PRECISION_COMPLEX
931
	tmp2 = _mm256_set_pd(s_dbl[1], s_dbl[0], s_dbl[1], s_dbl[0]);
932 933
#endif
#ifdef SINGLE_PRECISION_COMPLEX
934 935
	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]);
936
#endif
937

938 939 940 941 942 943
	tmp1 = _AVX_MUL(h2_imag, tmp2);
#ifdef __ELPA_USE_FMA__
	tmp2 = _AVX_FMADDSUB(h2_real, tmp2, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#else
	tmp2 = _AVX_ADDSUB( _AVX_MUL(h2_real, tmp2), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#endif
944 945
	h2_real = _AVX_SET1(tmp2[0]);
	h2_imag = _AVX_SET1(tmp2[1]);
946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 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 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119

	tmp1 = _AVX_MUL(h1_imag, y1);
#ifdef __ELPA_USE_FMA__
	y1 = _AVX_FMADDSUB(h1_real, y1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#else
	y1 = _AVX_ADDSUB( _AVX_MUL(h1_real, y1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#endif
	tmp2 = _AVX_MUL(h1_imag, y2);
#ifdef __ELPA_USE_FMA__
	y2 = _AVX_FMADDSUB(h1_real, y2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE));
#else
	y2 = _AVX_ADDSUB( _AVX_MUL(h1_real, y2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE));
#endif
	tmp3 = _AVX_MUL(h1_imag, y3);
#ifdef __ELPA_USE_FMA__
	y3 = _AVX_FMADDSUB(h1_real, y3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE));
#else
	y3 = _AVX_ADDSUB( _AVX_MUL(h1_real, y3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE));
#endif

	tmp1 = _AVX_MUL(h2_imag, x1);
#ifdef __ELPA_USE_FMA__
	y1 = _AVX_ADD(y1, _AVX_FMADDSUB(h2_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	y1 = _AVX_ADD(y1, _AVX_ADDSUB( _AVX_MUL(h2_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
	tmp2 = _AVX_MUL(h2_imag, x2);
#ifdef __ELPA_USE_FMA__
	y2 = _AVX_ADD(y2, _AVX_FMADDSUB(h2_real, x2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
	y2 = _AVX_ADD(y2, _AVX_ADDSUB( _AVX_MUL(h2_real, x2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif
	tmp3 = _AVX_MUL(h2_imag, x3);
#ifdef __ELPA_USE_FMA__
	y3 = _AVX_ADD(y3, _AVX_FMADDSUB(h2_real, x3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
	y3 = _AVX_ADD(y3, _AVX_ADDSUB( _AVX_MUL(h2_real, x3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif

	q1 = _AVX_LOAD(&q_dbl[0]);
	q2 = _AVX_LOAD(&q_dbl[offset]);
	q3 = _AVX_LOAD(&q_dbl[2*offset]);

	q1 = _AVX_ADD(q1, y1);
	q2 = _AVX_ADD(q2, y2);
	q3 = _AVX_ADD(q3, y3);

	_AVX_STORE(&q_dbl[0], q1);
	_AVX_STORE(&q_dbl[offset], q2);
	_AVX_STORE(&q_dbl[2*offset], q3);

	h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+1)*2]);
	h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+1)*2)+1]);

	q1 = _AVX_LOAD(&q_dbl[(ldq*2)+0]);
	q2 = _AVX_LOAD(&q_dbl[(ldq*2)+offset]);
	q3 = _AVX_LOAD(&q_dbl[(ldq*2)+2*offset]);

	q1 = _AVX_ADD(q1, x1);
	q2 = _AVX_ADD(q2, x2);
	q3 = _AVX_ADD(q3, x3);

	tmp1 = _AVX_MUL(h2_imag, y1);
#ifdef __ELPA_USE_FMA__
	q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h2_real, y1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h2_real, y1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
	tmp2 = _AVX_MUL(h2_imag, y2);
#ifdef __FMA4_
	q2 = _AVX_ADD(q2, _AVX_FMADDSUB(h2_real, y2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
	q2 = _AVX_ADD(q2, _AVX_ADDSUB( _AVX_MUL(h2_real, y2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif
	tmp3 = _AVX_MUL(h2_imag, y3);
#ifdef __ELPA_USE_FMA__
	q3 = _AVX_ADD(q3, _AVX_FMADDSUB(h2_real, y3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
	q3 = _AVX_ADD(q3, _AVX_ADDSUB( _AVX_MUL(h2_real, y3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif

	_AVX_STORE(&q_dbl[(ldq*2)+0], q1);
	_AVX_STORE(&q_dbl[(ldq*2)+offset], q2);
	_AVX_STORE(&q_dbl[(ldq*2)+2*offset], q3);

	for (i = 2; i < nb; i++)
	{
		q1 = _AVX_LOAD(&q_dbl[(2*i*ldq)+0]);
		q2 = _AVX_LOAD(&q_dbl[(2*i*ldq)+offset]);
		q3 = _AVX_LOAD(&q_dbl[(2*i*ldq)+2*offset]);

		h1_real = _AVX_BROADCAST(&hh_dbl[(i-1)*2]);
		h1_imag = _AVX_BROADCAST(&hh_dbl[((i-1)*2)+1]);

		tmp1 = _AVX_MUL(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
		q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h1_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h1_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
		tmp2 = _AVX_MUL(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
		q2 = _AVX_ADD(q2, _AVX_FMADDSUB(h1_real, x2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
		q2 = _AVX_ADD(q2, _AVX_ADDSUB( _AVX_MUL(h1_real, x2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif
		tmp3 = _AVX_MUL(h1_imag, x3);
#ifdef __ELPA_USE_FMA__
		q3 = _AVX_ADD(q3, _AVX_FMADDSUB(h1_real, x3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
		q3 = _AVX_ADD(q3, _AVX_ADDSUB( _AVX_MUL(h1_real, x3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif

		h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+i)*2]);
		h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+i)*2)+1]);

		tmp1 = _AVX_MUL(h2_imag, y1);
#ifdef __ELPA_USE_FMA__
		q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h2_real, y1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h2_real, y1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
		tmp2 = _AVX_MUL(h2_imag, y2);
#ifdef __ELPA_USE_FMA__
		q2 = _AVX_ADD(q2, _AVX_FMADDSUB(h2_real, y2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
		q2 = _AVX_ADD(q2, _AVX_ADDSUB( _AVX_MUL(h2_real, y2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif
		tmp3 = _AVX_MUL(h2_imag, y3);
#ifdef __ELPA_USE_FMA__
		q3 = _AVX_ADD(q3, _AVX_FMADDSUB(h2_real, y3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
		q3 = _AVX_ADD(q3, _AVX_ADDSUB( _AVX_MUL(h2_real, y3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif

		_AVX_STORE(&q_dbl[(2*i*ldq)+0], q1);
		_AVX_STORE(&q_dbl[(2*i*ldq)+offset], q2);
		_AVX_STORE(&q_dbl[(2*i*ldq)+2*offset], q3);
	}
	h1_real = _AVX_BROADCAST(&hh_dbl[(nb-1)*2]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[((nb-1)*2)+1]);

	q1 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+0]);
	q2 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+offset]);
	q3 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+2*offset]);

	tmp1 = _AVX_MUL(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
	q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h1_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h1_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
	tmp2 = _AVX_MUL(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
	q2 = _AVX_ADD(q2, _AVX_FMADDSUB(h1_real, x2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
	q2 = _AVX_ADD(q2, _AVX_ADDSUB( _AVX_MUL(h1_real, x2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif
	tmp3 = _AVX_MUL(h1_imag, x3);
#ifdef __ELPA_USE_FMA__
	q3 = _AVX_ADD(q3, _AVX_FMADDSUB(h1_real, x3, _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#else
	q3 = _AVX_ADD(q3, _AVX_ADDSUB( _AVX_MUL(h1_real, x3), _AVX_SHUFFLE(tmp3, tmp3, _SHUFFLE)));
#endif

	_AVX_STORE(&q_dbl[(2*nb*ldq)+0], q1);
	_AVX_STORE(&q_dbl[(2*nb*ldq)+offset], q2);
	_AVX_STORE(&q_dbl[(2*nb*ldq)+2*offset], q3);
}

#ifdef DOUBLE_PRECISION_COMPLEX
static __forceinline void hh_trafo_complex_kernel_4_AVX_2hv_double(double complex* q, double complex* hh, int nb, int ldq, int ldh, double complex s)
#endif
#ifdef SINGLE_PRECISION_COMPLEX
1120
static __forceinline void hh_trafo_complex_kernel_8_AVX_2hv_single(float complex* q, float complex* hh, int nb, int ldq, int ldh, float complex s, float complex s1)
1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149
#endif

{
#ifdef DOUBLE_PRECISION_COMPLEX
	double* q_dbl = (double*)q;
	double* hh_dbl = (double*)hh;
	double* s_dbl = (double*)(&s);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
	float* q_dbl = (float*)q;
	float* hh_dbl = (float*)hh;
	float* s_dbl = (float*)(&s);
#endif
	__AVX_DATATYPE x1, x2;
	__AVX_DATATYPE y1, y2;
	__AVX_DATATYPE q1, q2;
	__AVX_DATATYPE h1_real, h1_imag, h2_real, h2_imag;
	__AVX_DATATYPE tmp1, tmp2;
	int i=0;

#ifdef DOUBLE_PRECISION_COMPLEX
	__AVX_DATATYPE sign = (__AVX_DATATYPE)_mm256_set_epi64x(0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000);
#endif
#ifdef SINGLE_PRECISION_COMPLEX
	__AVX_DATATYPE sign = (__AVX_DATATYPE)_mm256_set_epi32(0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000);
#endif

	x1 = _AVX_LOAD(&q_dbl[(2*ldq)+0]);
	x2 = _AVX_LOAD(&q_dbl[(2*ldq)+offset]);
1150

1151 1152 1153 1154 1155 1156 1157 1158 1159
	h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+1)*2]);
	h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+1)*2)+1]);
#ifndef __ELPA_USE_FMA__
	// conjugate
	h2_imag = _AVX_XOR(h2_imag, sign);
#endif

	y1 = _AVX_LOAD(&q_dbl[0]);
	y2 = _AVX_LOAD(&q_dbl[offset]);
1160

1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178
	tmp1 = _AVX_MUL(h2_imag, x1);
#ifdef __ELPA_USE_FMA__
	y1 = _AVX_ADD(y1, _AVX_FMSUBADD(h2_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	y1 = _AVX_ADD(y1, _AVX_ADDSUB( _AVX_MUL(h2_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

	tmp2 = _AVX_MUL(h2_imag, x2);
#ifdef __ELPA_USE_FMA__
	y2 = _AVX_ADD(y2, _AVX_FMSUBADD(h2_real, x2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
	y2 = _AVX_ADD(y2, _AVX_ADDSUB( _AVX_MUL(h2_real, x2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

	for (i = 2; i < nb; i++)
	{
		q1 = _AVX_LOAD(&q_dbl[(2*i*ldq)+0]);
		q2 = _AVX_LOAD(&q_dbl[(2*i*ldq)+offset]);
1179

1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277
		h1_real = _AVX_BROADCAST(&hh_dbl[(i-1)*2]);
		h1_imag = _AVX_BROADCAST(&hh_dbl[((i-1)*2)+1]);
#ifndef __ELPA_USE_FMA__
		// conjugate
		h1_imag = _AVX_XOR(h1_imag, sign);
#endif

		tmp1 = _AVX_MUL(h1_imag, q1);
#ifdef __ELPA_USE_FMA__
		x1 = _AVX_ADD(x1, _AVX_FMSUBADD(h1_real, q1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		x1 = _AVX_ADD(x1, _AVX_ADDSUB( _AVX_MUL(h1_real, q1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

		tmp2 = _AVX_MUL(h1_imag, q2);
#ifdef __ELPA_USE_FMA__
		x2 = _AVX_ADD(x2, _AVX_FMSUBADD(h1_real, q2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
		x2 = _AVX_ADD(x2, _AVX_ADDSUB( _AVX_MUL(h1_real, q2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

		h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+i)*2]);
		h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+i)*2)+1]);
#ifndef __ELPA_USE_FMA__
		// conjugate
		h2_imag = _AVX_XOR(h2_imag, sign);
#endif

		tmp1 = _AVX_MUL(h2_imag, q1);
#ifdef __ELPA_USE_FMA__
		y1 = _AVX_ADD(y1, _AVX_FMSUBADD(h2_real, q1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		y1 = _AVX_ADD(y1, _AVX_ADDSUB( _AVX_MUL(h2_real, q1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

		tmp2 = _AVX_MUL(h2_imag, q2);
#ifdef __ELPA_USE_FMA__
		y2 = _AVX_ADD(y2, _AVX_FMSUBADD(h2_real, q2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
		y2 = _AVX_ADD(y2, _AVX_ADDSUB( _AVX_MUL(h2_real, q2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

	}

	h1_real = _AVX_BROADCAST(&hh_dbl[(nb-1)*2]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[((nb-1)*2)+1]);
#ifndef __ELPA_USE_FMA__
	// conjugate
	h1_imag = _AVX_XOR(h1_imag, sign);
#endif

	q1 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+0]);
	q2 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+offset]);

	tmp1 = _AVX_MUL(h1_imag, q1);
#ifdef __ELPA_USE_FMA__
	x1 = _AVX_ADD(x1, _AVX_FMSUBADD(h1_real, q1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	x1 = _AVX_ADD(x1, _AVX_ADDSUB( _AVX_MUL(h1_real, q1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

	tmp2 = _AVX_MUL(h1_imag, q2);
#ifdef __ELPA_USE_FMA__
	x2 = _AVX_ADD(x2, _AVX_FMSUBADD(h1_real, q2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
	x2 = _AVX_ADD(x2, _AVX_ADDSUB( _AVX_MUL(h1_real, q2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

	h1_real = _AVX_BROADCAST(&hh_dbl[0]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[1]);
	h1_real = _AVX_XOR(h1_real, sign);
	h1_imag = _AVX_XOR(h1_imag, sign);

	tmp1 = _AVX_MUL(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
	x1 = _AVX_FMADDSUB(h1_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#else
	x1 = _AVX_ADDSUB( _AVX_MUL(h1_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#endif

	tmp2 = _AVX_MUL(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
	x2 = _AVX_FMADDSUB(h1_real, x2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE));
#else
	x2 = _AVX_ADDSUB( _AVX_MUL(h1_real, x2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE));
#endif

	h1_real = _AVX_BROADCAST(&hh_dbl[ldh*2]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[(ldh*2)+1]);
	h2_real = _AVX_BROADCAST(&hh_dbl[ldh*2]);
	h2_imag = _AVX_BROADCAST(&hh_dbl[(ldh*2)+1]);

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

#ifdef DOUBLE_PRECISION_COMPLEX
1278
	tmp2 = _mm256_set_pd(s_dbl[1], s_dbl[0], s_dbl[1], s_dbl[0]);
1279 1280
#endif
#ifdef SINGLE_PRECISION_COMPLEX
1281 1282
	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]);
1283 1284 1285 1286 1287 1288 1289 1290
#endif

	tmp1 = _AVX_MUL(h2_imag, tmp2);
#ifdef __ELPA_USE_FMA__
	tmp2 = _AVX_FMADDSUB(h2_real, tmp2, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#else
	tmp2 = _AVX_ADDSUB( _AVX_MUL(h2_real, tmp2), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#endif
1291 1292
	h2_real = _AVX_SET1(tmp2[0]);
	h2_imag = _AVX_SET1(tmp2[1]);
1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322

	tmp1 = _AVX_MUL(h1_imag, y1);
#ifdef __ELPA_USE_FMA__
	y1 = _AVX_FMADDSUB(h1_real, y1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#else
	y1 = _AVX_ADDSUB( _AVX_MUL(h1_real, y1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#endif

	tmp2 = _AVX_MUL(h1_imag, y2);
#ifdef __ELPA_USE_FMA__
	y2 = _AVX_FMADDSUB(h1_real, y2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE));
#else
	y2 = _AVX_ADDSUB( _AVX_MUL(h1_real, y2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE));
#endif

	tmp1 = _AVX_MUL(h2_imag, x1);
#ifdef __ELPA_USE_FMA__
	y1 = _AVX_ADD(y1, _AVX_FMADDSUB(h2_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	y1 = _AVX_ADD(y1, _AVX_ADDSUB( _AVX_MUL(h2_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
	tmp2 = _AVX_MUL(h2_imag, x2);
#ifdef __ELPA_USE_FMA__
	y2 = _AVX_ADD(y2, _AVX_FMADDSUB(h2_real, x2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
	y2 = _AVX_ADD(y2, _AVX_ADDSUB( _AVX_MUL(h2_real, x2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

	q1 = _AVX_LOAD(&q_dbl[0]);
	q2 = _AVX_LOAD(&q_dbl[offset]);
1323

1324 1325
	q1 = _AVX_ADD(q1, y1);
	q2 = _AVX_ADD(q2, y2);
1326

1327 1328
	_AVX_STORE(&q_dbl[0], q1);
	_AVX_STORE(&q_dbl[offset], q2);
1329

1330 1331 1332 1333 1334
	h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+1)*2]);
	h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+1)*2)+1]);

	q1 = _AVX_LOAD(&q_dbl[(ldq*2)+0]);
	q2 = _AVX_LOAD(&q_dbl[(ldq*2)+offset]);
1335

1336 1337
	q1 = _AVX_ADD(q1, x1);
	q2 = _AVX_ADD(q2, x2);
1338

1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354
	tmp1 = _AVX_MUL(h2_imag, y1);
#ifdef __ELPA_USE_FMA__
	q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h2_real, y1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h2_real, y1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

	tmp2 = _AVX_MUL(h2_imag, y2);
#ifdef __FMA4_
	q2 = _AVX_ADD(q2, _AVX_FMADDSUB(h2_real, y2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
	q2 = _AVX_ADD(q2, _AVX_ADDSUB( _AVX_MUL(h2_real, y2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

	_AVX_STORE(&q_dbl[(ldq*2)+0], q1);
	_AVX_STORE(&q_dbl[(ldq*2)+offset], q2);
1355

1356 1357 1358 1359
	for (i = 2; i < nb; i++)
	{
		q1 = _AVX_LOAD(&q_dbl[(2*i*ldq)+0]);
		q2 = _AVX_LOAD(&q_dbl[(2*i*ldq)+offset]);
1360

1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422
		h1_real = _AVX_BROADCAST(&hh_dbl[(i-1)*2]);
		h1_imag = _AVX_BROADCAST(&hh_dbl[((i-1)*2)+1]);

		tmp1 = _AVX_MUL(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
		q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h1_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h1_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

		tmp2 = _AVX_MUL(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
		q2 = _AVX_ADD(q2, _AVX_FMADDSUB(h1_real, x2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
		q2 = _AVX_ADD(q2, _AVX_ADDSUB( _AVX_MUL(h1_real, x2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif
		h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+i)*2]);
		h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+i)*2)+1]);

		tmp1 = _AVX_MUL(h2_imag, y1);
#ifdef __ELPA_USE_FMA__
		q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h2_real, y1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h2_real, y1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

		tmp2 = _AVX_MUL(h2_imag, y2);
#ifdef __ELPA_USE_FMA__
		q2 = _AVX_ADD(q2, _AVX_FMADDSUB(h2_real, y2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
		q2 = _AVX_ADD(q2, _AVX_ADDSUB( _AVX_MUL(h2_real, y2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

		_AVX_STORE(&q_dbl[(2*i*ldq)+0], q1);
		_AVX_STORE(&q_dbl[(2*i*ldq)+offset], q2);
	}
	h1_real = _AVX_BROADCAST(&hh_dbl[(nb-1)*2]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[((nb-1)*2)+1]);

	q1 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+0]);
	q2 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+offset]);

	tmp1 = _AVX_MUL(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
	q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h1_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h1_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

	tmp2 = _AVX_MUL(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
	q2 = _AVX_ADD(q2, _AVX_FMADDSUB(h1_real, x2, _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#else
	q2 = _AVX_ADD(q2, _AVX_ADDSUB( _AVX_MUL(h1_real, x2), _AVX_SHUFFLE(tmp2, tmp2, _SHUFFLE)));
#endif

	_AVX_STORE(&q_dbl[(2*nb*ldq)+0], q1);
	_AVX_STORE(&q_dbl[(2*nb*ldq)+offset], q2);
}

#ifdef DOUBLE_PRECISION_COMPLEX
static __forceinline void hh_trafo_complex_kernel_2_AVX_2hv_double(double complex* q, double complex* hh, int nb, int ldq, int ldh, double complex s)
1423 1424 1425 1426 1427
#endif
#ifdef SINGLE_PRECISION_COMPLEX
static __forceinline void hh_trafo_complex_kernel_4_AVX_2hv_single(float complex* q, float complex* hh, int nb, int ldq, int ldh, float complex s, float complex s1)
#endif

1428
{
1429
#ifdef DOUBLE_PRECISION_COMPLEX
1430 1431 1432
	double* q_dbl = (double*)q;
	double* hh_dbl = (double*)hh;
	double* s_dbl = (double*)(&s);
1433 1434 1435 1436 1437 1438
#endif
#ifdef SINGLE_PRECISION_COMPLEX
	float* q_dbl = (float*)q;
	float* hh_dbl = (float*)hh;
	float* s_dbl = (float*)(&s);
#endif
1439 1440 1441 1442 1443 1444
	__AVX_DATATYPE x1;
	__AVX_DATATYPE y1;
	__AVX_DATATYPE q1;
	__AVX_DATATYPE h1_real, h1_imag, h2_real, h2_imag;
	__AVX_DATATYPE tmp1;
	int i=0;
1445
#ifdef DOUBLE_PRECISION_COMPLEX
1446
	__AVX_DATATYPE sign = (__AVX_DATATYPE)_mm256_set_epi64x(0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000);
1447 1448 1449 1450
#endif
#ifdef SINGLE_PRECISION_COMPLEX
	__AVX_DATATYPE sign = (__AVX_DATATYPE)_mm256_set_epi32(0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000);
#endif
1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539
	x1 = _AVX_LOAD(&q_dbl[(2*ldq)+0]);

	h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+1)*2]);
	h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+1)*2)+1]);
#ifndef __ELPA_USE_FMA__
	// conjugate
	h2_imag = _AVX_XOR(h2_imag, sign);
#endif

	y1 = _AVX_LOAD(&q_dbl[0]);

	tmp1 = _AVX_MUL(h2_imag, x1);
#ifdef __ELPA_USE_FMA__
	y1 = _AVX_ADD(y1, _AVX_FMSUBADD(h2_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	y1 = _AVX_ADD(y1, _AVX_ADDSUB( _AVX_MUL(h2_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

	for (i = 2; i < nb; i++)
	{
		q1 = _AVX_LOAD(&q_dbl[(2*i*ldq)+0]);

		h1_real = _AVX_BROADCAST(&hh_dbl[(i-1)*2]);
		h1_imag = _AVX_BROADCAST(&hh_dbl[((i-1)*2)+1]);
#ifndef __ELPA_USE_FMA__
		// conjugate
		h1_imag = _AVX_XOR(h1_imag, sign);
#endif

		tmp1 = _AVX_MUL(h1_imag, q1);
#ifdef __ELPA_USE_FMA__
		x1 = _AVX_ADD(x1, _AVX_FMSUBADD(h1_real, q1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		x1 = _AVX_ADD(x1, _AVX_ADDSUB( _AVX_MUL(h1_real, q1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

		h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+i)*2]);
		h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+i)*2)+1]);
#ifndef __ELPA_USE_FMA__
		// conjugate
		h2_imag = _AVX_XOR(h2_imag, sign);
#endif

		tmp1 = _AVX_MUL(h2_imag, q1);
#ifdef __ELPA_USE_FMA__
		y1 = _AVX_ADD(y1, _AVX_FMSUBADD(h2_real, q1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		y1 = _AVX_ADD(y1, _AVX_ADDSUB( _AVX_MUL(h2_real, q1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif
	}

	h1_real = _AVX_BROADCAST(&hh_dbl[(nb-1)*2]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[((nb-1)*2)+1]);
#ifndef __ELPA_USE_FMA__
	// conjugate
	h1_imag = _AVX_XOR(h1_imag, sign);
#endif

	q1 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+0]);

	tmp1 = _AVX_MUL(h1_imag, q1);
#ifdef __ELPA_USE_FMA__
	x1 = _AVX_ADD(x1, _AVX_FMSUBADD(h1_real, q1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	x1 = _AVX_ADD(x1, _AVX_ADDSUB( _AVX_MUL(h1_real, q1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

	h1_real = _AVX_BROADCAST(&hh_dbl[0]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[1]);
	h1_real = _AVX_XOR(h1_real, sign);
	h1_imag = _AVX_XOR(h1_imag, sign);

	tmp1 = _AVX_MUL(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
	x1 = _AVX_FMADDSUB(h1_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#else
	x1 = _AVX_ADDSUB( _AVX_MUL(h1_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#endif

	h1_real = _AVX_BROADCAST(&hh_dbl[ldh*2]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[(ldh*2)+1]);
	h2_real = _AVX_BROADCAST(&hh_dbl[ldh*2]);
	h2_imag = _AVX_BROADCAST(&hh_dbl[(ldh*2)+1]);

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

1540
	__AVX_DATATYPE tmp2;
1541
#ifdef DOUBLE_PRECISION_COMPLEX
1542
	tmp2 = _mm256_set_pd(s_dbl[1], s_dbl[0], s_dbl[1], s_dbl[0]);
1543 1544
#endif
#ifdef SINGLE_PRECISION_COMPLEX
1545 1546
	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]);
1547 1548 1549 1550 1551 1552 1553
#endif
	tmp1 = _AVX_MUL(h2_imag, tmp2);
#ifdef __ELPA_USE_FMA__
	tmp2 = _AVX_FMADDSUB(h2_real, tmp2, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#else
	tmp2 = _AVX_ADDSUB( _AVX_MUL(h2_real, tmp2), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#endif
1554 1555
	h2_real = _AVX_SET1(tmp2[0]);
	h2_imag = _AVX_SET1(tmp2[1]);
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 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 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632

	tmp1 = _AVX_MUL(h1_imag, y1);
#ifdef __ELPA_USE_FMA__
	y1 = _AVX_FMADDSUB(h1_real, y1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#else
	y1 = _AVX_ADDSUB( _AVX_MUL(h1_real, y1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE));
#endif

	tmp1 = _AVX_MUL(h2_imag, x1);
#ifdef __ELPA_USE_FMA__
	y1 = _AVX_ADD(y1, _AVX_FMADDSUB(h2_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	y1 = _AVX_ADD(y1, _AVX_ADDSUB( _AVX_MUL(h2_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

	q1 = _AVX_LOAD(&q_dbl[0]);

	q1 = _AVX_ADD(q1, y1);

	_AVX_STORE(&q_dbl[0], q1);

	h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+1)*2]);
	h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+1)*2)+1]);

	q1 = _AVX_LOAD(&q_dbl[(ldq*2)+0]);

	q1 = _AVX_ADD(q1, x1);

	tmp1 = _AVX_MUL(h2_imag, y1);
#ifdef __ELPA_USE_FMA__
	q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h2_real, y1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h2_real, y1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

	_AVX_STORE(&q_dbl[(ldq*2)+0], q1);

	for (i = 2; i < nb; i++)
	{
		q1 = _AVX_LOAD(&q_dbl[(2*i*ldq)+0]);

		h1_real = _AVX_BROADCAST(&hh_dbl[(i-1)*2]);
		h1_imag = _AVX_BROADCAST(&hh_dbl[((i-1)*2)+1]);

		tmp1 = _AVX_MUL(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
		q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h1_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h1_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

		h2_real = _AVX_BROADCAST(&hh_dbl[(ldh+i)*2]);
		h2_imag = _AVX_BROADCAST(&hh_dbl[((ldh+i)*2)+1]);

		tmp1 = _AVX_MUL(h2_imag, y1);
#ifdef __ELPA_USE_FMA__
		q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h2_real, y1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
		q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h2_real, y1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

		_AVX_STORE(&q_dbl[(2*i*ldq)+0], q1);
	}
	h1_real = _AVX_BROADCAST(&hh_dbl[(nb-1)*2]);
	h1_imag = _AVX_BROADCAST(&hh_dbl[((nb-1)*2)+1]);

	q1 = _AVX_LOAD(&q_dbl[(2*nb*ldq)+0]);

	tmp1 = _AVX_MUL(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
	q1 = _AVX_ADD(q1, _AVX_FMADDSUB(h1_real, x1, _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#else
	q1 = _AVX_ADD(q1, _AVX_ADDSUB( _AVX_MUL(h1_real, x1), _AVX_SHUFFLE(tmp1, tmp1, _SHUFFLE)));
#endif

	_AVX_STORE(&q_dbl[(2