elpa2_kernels_complex_sse_1hv.cpp 19.6 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 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588
//    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 Naturwissenschaftrn,
//      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.
//
//
// --------------------------------------------------------------------------------------------------
//
// 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>
#include <x86intrin.h>

#define __forceinline __attribute__((always_inline))

#ifdef HAVE_SSE
#undef __AVX__
#endif


extern "C" {

//Forward declaration
static __forceinline void hh_trafo_complex_kernel_6_SSE_1hv(std::complex<double>* q, std::complex<double>* hh, int nb, int ldq);
static __forceinline void hh_trafo_complex_kernel_4_SSE_1hv(std::complex<double>* q, std::complex<double>* hh, int nb, int ldq);
static __forceinline void hh_trafo_complex_kernel_2_SSE_1hv(std::complex<double>* q, std::complex<double>* hh, int nb, int ldq);

#if 0
static __forceinline void hh_trafo_complex_kernel_4_C_1hv(std::complex<double>* q, std::complex<double>* hh, int nb, int ldq)
{
	std::complex<double> x0;
	std::complex<double> x1;
	std::complex<double> x2;
	std::complex<double> x3;
	std::complex<double> h0;
	std::complex<double> tau0;
	int i=0;

	x0 = q[0];
	x1 = q[1];
	x2 = q[2];
	x3 = q[3];

	for (i = 1; i < nb; i++)
	{
		h0 = conj(hh[i]);
		x0 += (q[(i*ldq)+0] * h0);
		x1 += (q[(i*ldq)+1] * h0);
		x2 += (q[(i*ldq)+2] * h0);
		x3 += (q[(i*ldq)+3] * h0);
	}

	tau0 = hh[0];

	h0 = (-1.0)*tau0;

	x0 *= h0;
	x1 *= h0;
	x2 *= h0;
	x3 *= h0;

	q[0] += x0;
	q[1] += x1;
	q[2] += x2;
	q[3] += x3;

	for (i = 1; i < nb; i++)
	{
		h0 = hh[i];
		q[(i*ldq)+0] += (x0*h0);
		q[(i*ldq)+1] += (x1*h0);
		q[(i*ldq)+2] += (x2*h0);
		q[(i*ldq)+3] += (x3*h0);
	}
}
#endif // if 0

void single_hh_trafo_complex_sse_1hv_(std::complex<double>* q, std::complex<double>* hh, int* pnb, int* pnq, int* pldq)
{
	int i;
	int nb = *pnb;
	int nq = *pldq;
	int ldq = *pldq;
	//int ldh = *pldh;

	for (i = 0; i < nq-4; i+=6)
	{
		hh_trafo_complex_kernel_6_SSE_1hv(&q[i], hh, nb, ldq);
	}
	if (nq-i > 2)
	{
		hh_trafo_complex_kernel_4_SSE_1hv(&q[i], hh, nb, ldq);
	}
	else if (nq-i > 0)
	{
		hh_trafo_complex_kernel_2_SSE_1hv(&q[i], hh, nb, ldq);
	}
}

static __forceinline void hh_trafo_complex_kernel_6_SSE_1hv(std::complex<double>* q, std::complex<double>* hh, int nb, int ldq)
{
	double* q_dbl = (double*)q;
	double* hh_dbl = (double*)hh;

	__m128d x1, x2, x3, x4, x5, x6;
	__m128d q1, q2, q3, q4, q5, q6;
	__m128d h1_real, h1_imag;
	__m128d tmp1, tmp2, tmp3, tmp4, tmp5, tmp6;
	int i=0;

	__m128d sign = (__m128d)_mm_set_epi64x(0x8000000000000000, 0x8000000000000000);

	x1 = _mm_load_pd(&q_dbl[0]);
	x2 = _mm_load_pd(&q_dbl[2]);
	x3 = _mm_load_pd(&q_dbl[4]);
	x4 = _mm_load_pd(&q_dbl[6]);
	x5 = _mm_load_pd(&q_dbl[8]);
	x6 = _mm_load_pd(&q_dbl[10]);

	for (i = 1; i < nb; i++)
	{
		h1_real = _mm_loaddup_pd(&hh_dbl[i*2]);
		h1_imag = _mm_loaddup_pd(&hh_dbl[(i*2)+1]);
#ifndef __ELPA_USE_FMA__
		// conjugate
		h1_imag = _mm_xor_pd(h1_imag, sign);
#endif

		q1 = _mm_load_pd(&q_dbl[(2*i*ldq)+0]);
		q2 = _mm_load_pd(&q_dbl[(2*i*ldq)+2]);
		q3 = _mm_load_pd(&q_dbl[(2*i*ldq)+4]);
		q4 = _mm_load_pd(&q_dbl[(2*i*ldq)+6]);
		q5 = _mm_load_pd(&q_dbl[(2*i*ldq)+8]);
		q6 = _mm_load_pd(&q_dbl[(2*i*ldq)+10]);

		tmp1 = _mm_mul_pd(h1_imag, q1);
#ifdef __ELPA_USE_FMA__
		x1 = _mm_add_pd(x1, _mm_msubadd_pd(h1_real, q1, _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1))));
#else
		x1 = _mm_add_pd(x1, _mm_addsub_pd( _mm_mul_pd(h1_real, q1), _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1))));
#endif
		tmp2 = _mm_mul_pd(h1_imag, q2);
#ifdef __ELPA_USE_FMA__
		x2 = _mm_add_pd(x2, _mm_msubadd_pd(h1_real, q2, _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1))));
#else
		x2 = _mm_add_pd(x2, _mm_addsub_pd( _mm_mul_pd(h1_real, q2), _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1))));
#endif
		tmp3 = _mm_mul_pd(h1_imag, q3);
#ifdef __ELPA_USE_FMA__
		x3 = _mm_add_pd(x3, _mm_msubadd_pd(h1_real, q3, _mm_shuffle_pd(tmp3, tmp3, _MM_SHUFFLE2(0,1))));
#else
		x3 = _mm_add_pd(x3, _mm_addsub_pd( _mm_mul_pd(h1_real, q3), _mm_shuffle_pd(tmp3, tmp3, _MM_SHUFFLE2(0,1))));
#endif
		tmp4 = _mm_mul_pd(h1_imag, q4);
#ifdef __ELPA_USE_FMA__
		x4 = _mm_add_pd(x4, _mm_msubadd_pd(h1_real, q4, _mm_shuffle_pd(tmp4, tmp4, _MM_SHUFFLE2(0,1))));
#else
		x4 = _mm_add_pd(x4, _mm_addsub_pd( _mm_mul_pd(h1_real, q4), _mm_shuffle_pd(tmp4, tmp4, _MM_SHUFFLE2(0,1))));
#endif
		tmp5 = _mm_mul_pd(h1_imag, q5);
#ifdef __ELPA_USE_FMA__
		x5 = _mm_add_pd(x5, _mm_msubadd_pd(h1_real, q5, _mm_shuffle_pd(tmp5, tmp5, _MM_SHUFFLE2(0,1))));
#else
		x5 = _mm_add_pd(x5, _mm_addsub_pd( _mm_mul_pd(h1_real, q5), _mm_shuffle_pd(tmp5, tmp5, _MM_SHUFFLE2(0,1))));
#endif
		tmp6 = _mm_mul_pd(h1_imag, q6);
#ifdef __ELPA_USE_FMA__
		x6 = _mm_add_pd(x6, _mm_msubadd_pd(h1_real, q6, _mm_shuffle_pd(tmp6, tmp6, _MM_SHUFFLE2(0,1))));
#else
		x6 = _mm_add_pd(x6, _mm_addsub_pd( _mm_mul_pd(h1_real, q6), _mm_shuffle_pd(tmp6, tmp6, _MM_SHUFFLE2(0,1))));
#endif
	}

	h1_real = _mm_loaddup_pd(&hh_dbl[0]);
	h1_imag = _mm_loaddup_pd(&hh_dbl[1]);
	h1_real = _mm_xor_pd(h1_real, sign);
	h1_imag = _mm_xor_pd(h1_imag, sign);

	tmp1 = _mm_mul_pd(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
	x1 = _mm_maddsub_pd(h1_real, x1, _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1)));
#else
	x1 = _mm_addsub_pd( _mm_mul_pd(h1_real, x1), _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1)));
#endif
	tmp2 = _mm_mul_pd(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
	x2 = _mm_maddsub_pd(h1_real, x2, _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1)));
#else
	x2 = _mm_addsub_pd( _mm_mul_pd(h1_real, x2), _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1)));
#endif
	tmp3 = _mm_mul_pd(h1_imag, x3);
#ifdef __ELPA_USE_FMA__
	x3 = _mm_maddsub_pd(h1_real, x3, _mm_shuffle_pd(tmp3, tmp3, _MM_SHUFFLE2(0,1)));
#else
	x3 = _mm_addsub_pd( _mm_mul_pd(h1_real, x3), _mm_shuffle_pd(tmp3, tmp3, _MM_SHUFFLE2(0,1)));
#endif
	tmp4 = _mm_mul_pd(h1_imag, x4);
#ifdef __ELPA_USE_FMA__
	x4 = _mm_maddsub_pd(h1_real, x4, _mm_shuffle_pd(tmp4, tmp4, _MM_SHUFFLE2(0,1)));
#else
	x4 = _mm_addsub_pd( _mm_mul_pd(h1_real, x4), _mm_shuffle_pd(tmp4, tmp4, _MM_SHUFFLE2(0,1)));
#endif
	tmp5 = _mm_mul_pd(h1_imag, x5);
#ifdef __ELPA_USE_FMA__
	x5 = _mm_maddsub_pd(h1_real, x5, _mm_shuffle_pd(tmp5, tmp5, _MM_SHUFFLE2(0,1)));
#else
	x5 = _mm_addsub_pd( _mm_mul_pd(h1_real, x5), _mm_shuffle_pd(tmp5, tmp5, _MM_SHUFFLE2(0,1)));
#endif
	tmp6 = _mm_mul_pd(h1_imag, x6);
#ifdef __ELPA_USE_FMA__
	x6 = _mm_maddsub_pd(h1_real, x6, _mm_shuffle_pd(tmp6, tmp6, _MM_SHUFFLE2(0,1)));
#else
	x6 = _mm_addsub_pd( _mm_mul_pd(h1_real, x6), _mm_shuffle_pd(tmp6, tmp6, _MM_SHUFFLE2(0,1)));
#endif

	q1 = _mm_load_pd(&q_dbl[0]);
	q2 = _mm_load_pd(&q_dbl[2]);
	q3 = _mm_load_pd(&q_dbl[4]);
	q4 = _mm_load_pd(&q_dbl[6]);
	q5 = _mm_load_pd(&q_dbl[8]);
	q6 = _mm_load_pd(&q_dbl[10]);

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

	_mm_store_pd(&q_dbl[0], q1);
	_mm_store_pd(&q_dbl[2], q2);
	_mm_store_pd(&q_dbl[4], q3);
	_mm_store_pd(&q_dbl[6], q4);
	_mm_store_pd(&q_dbl[8], q5);
	_mm_store_pd(&q_dbl[10], q6);

	for (i = 1; i < nb; i++)
	{
		h1_real = _mm_loaddup_pd(&hh_dbl[i*2]);
		h1_imag = _mm_loaddup_pd(&hh_dbl[(i*2)+1]);

		q1 = _mm_load_pd(&q_dbl[(2*i*ldq)+0]);
		q2 = _mm_load_pd(&q_dbl[(2*i*ldq)+2]);
		q3 = _mm_load_pd(&q_dbl[(2*i*ldq)+4]);
		q4 = _mm_load_pd(&q_dbl[(2*i*ldq)+6]);
		q5 = _mm_load_pd(&q_dbl[(2*i*ldq)+8]);
		q6 = _mm_load_pd(&q_dbl[(2*i*ldq)+10]);

		tmp1 = _mm_mul_pd(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
		q1 = _mm_add_pd(q1, _mm_maddsub_pd(h1_real, x1, _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1))));
#else
		q1 = _mm_add_pd(q1, _mm_addsub_pd( _mm_mul_pd(h1_real, x1), _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1))));
#endif
		tmp2 = _mm_mul_pd(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
		q2 = _mm_add_pd(q2, _mm_maddsub_pd(h1_real, x2, _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1))));
#else
		q2 = _mm_add_pd(q2, _mm_addsub_pd( _mm_mul_pd(h1_real, x2), _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1))));
#endif
		tmp3 = _mm_mul_pd(h1_imag, x3);
#ifdef __ELPA_USE_FMA__
		q3 = _mm_add_pd(q3, _mm_maddsub_pd(h1_real, x3, _mm_shuffle_pd(tmp3, tmp3, _MM_SHUFFLE2(0,1))));
#else
		q3 = _mm_add_pd(q3, _mm_addsub_pd( _mm_mul_pd(h1_real, x3), _mm_shuffle_pd(tmp3, tmp3, _MM_SHUFFLE2(0,1))));
#endif
		tmp4 = _mm_mul_pd(h1_imag, x4);
#ifdef __ELPA_USE_FMA__
		q4 = _mm_add_pd(q4, _mm_maddsub_pd(h1_real, x4, _mm_shuffle_pd(tmp4, tmp4, _MM_SHUFFLE2(0,1))));
#else
		q4 = _mm_add_pd(q4, _mm_addsub_pd( _mm_mul_pd(h1_real, x4), _mm_shuffle_pd(tmp4, tmp4, _MM_SHUFFLE2(0,1))));
#endif
		tmp5 = _mm_mul_pd(h1_imag, x5);
#ifdef __ELPA_USE_FMA__
		q5 = _mm_add_pd(q5, _mm_maddsub_pd(h1_real, x5, _mm_shuffle_pd(tmp5, tmp5, _MM_SHUFFLE2(0,1))));
#else
		q5 = _mm_add_pd(q5, _mm_addsub_pd( _mm_mul_pd(h1_real, x5), _mm_shuffle_pd(tmp5, tmp5, _MM_SHUFFLE2(0,1))));
#endif
		tmp6 = _mm_mul_pd(h1_imag, x6);
#ifdef __ELPA_USE_FMA__
		q6 = _mm_add_pd(q6, _mm_maddsub_pd(h1_real, x6, _mm_shuffle_pd(tmp6, tmp6, _MM_SHUFFLE2(0,1))));
#else
		q6 = _mm_add_pd(q6, _mm_addsub_pd( _mm_mul_pd(h1_real, x6), _mm_shuffle_pd(tmp6, tmp6, _MM_SHUFFLE2(0,1))));
#endif

		_mm_store_pd(&q_dbl[(2*i*ldq)+0], q1);
		_mm_store_pd(&q_dbl[(2*i*ldq)+2], q2);
		_mm_store_pd(&q_dbl[(2*i*ldq)+4], q3);
		_mm_store_pd(&q_dbl[(2*i*ldq)+6], q4);
		_mm_store_pd(&q_dbl[(2*i*ldq)+8], q5);
		_mm_store_pd(&q_dbl[(2*i*ldq)+10], q6);
	}
}

static __forceinline void hh_trafo_complex_kernel_4_SSE_1hv(std::complex<double>* q, std::complex<double>* hh, int nb, int ldq)
{
	double* q_dbl = (double*)q;
	double* hh_dbl = (double*)hh;

	__m128d x1, x2, x3, x4;
	__m128d q1, q2, q3, q4;
	__m128d h1_real, h1_imag;
	__m128d tmp1, tmp2, tmp3, tmp4;
	int i=0;

	__m128d sign = (__m128d)_mm_set_epi64x(0x8000000000000000, 0x8000000000000000);

	x1 = _mm_load_pd(&q_dbl[0]);
	x2 = _mm_load_pd(&q_dbl[2]);
	x3 = _mm_load_pd(&q_dbl[4]);
	x4 = _mm_load_pd(&q_dbl[6]);

	for (i = 1; i < nb; i++)
	{
		h1_real = _mm_loaddup_pd(&hh_dbl[i*2]);
		h1_imag = _mm_loaddup_pd(&hh_dbl[(i*2)+1]);
#ifndef __ELPA_USE_FMA__
		// conjugate
		h1_imag = _mm_xor_pd(h1_imag, sign);
#endif

		q1 = _mm_load_pd(&q_dbl[(2*i*ldq)+0]);
		q2 = _mm_load_pd(&q_dbl[(2*i*ldq)+2]);
		q3 = _mm_load_pd(&q_dbl[(2*i*ldq)+4]);
		q4 = _mm_load_pd(&q_dbl[(2*i*ldq)+6]);

		tmp1 = _mm_mul_pd(h1_imag, q1);
#ifdef __ELPA_USE_FMA__
		x1 = _mm_add_pd(x1, _mm_msubadd_pd(h1_real, q1, _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1))));
#else
		x1 = _mm_add_pd(x1, _mm_addsub_pd( _mm_mul_pd(h1_real, q1), _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1))));
#endif
		tmp2 = _mm_mul_pd(h1_imag, q2);
#ifdef __ELPA_USE_FMA__
		x2 = _mm_add_pd(x2, _mm_msubadd_pd(h1_real, q2, _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1))));
#else
		x2 = _mm_add_pd(x2, _mm_addsub_pd( _mm_mul_pd(h1_real, q2), _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1))));
#endif
		tmp3 = _mm_mul_pd(h1_imag, q3);
#ifdef __ELPA_USE_FMA__
		x3 = _mm_add_pd(x3, _mm_msubadd_pd(h1_real, q3, _mm_shuffle_pd(tmp3, tmp3, _MM_SHUFFLE2(0,1))));
#else
		x3 = _mm_add_pd(x3, _mm_addsub_pd( _mm_mul_pd(h1_real, q3), _mm_shuffle_pd(tmp3, tmp3, _MM_SHUFFLE2(0,1))));
#endif
		tmp4 = _mm_mul_pd(h1_imag, q4);
#ifdef __ELPA_USE_FMA__
		x4 = _mm_add_pd(x4, _mm_msubadd_pd(h1_real, q4, _mm_shuffle_pd(tmp4, tmp4, _MM_SHUFFLE2(0,1))));
#else
		x4 = _mm_add_pd(x4, _mm_addsub_pd( _mm_mul_pd(h1_real, q4), _mm_shuffle_pd(tmp4, tmp4, _MM_SHUFFLE2(0,1))));
#endif
	}

	h1_real = _mm_loaddup_pd(&hh_dbl[0]);
	h1_imag = _mm_loaddup_pd(&hh_dbl[1]);
	h1_real = _mm_xor_pd(h1_real, sign);
	h1_imag = _mm_xor_pd(h1_imag, sign);

	tmp1 = _mm_mul_pd(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
	x1 = _mm_maddsub_pd(h1_real, x1, _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1)));
#else
	x1 = _mm_addsub_pd( _mm_mul_pd(h1_real, x1), _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1)));
#endif
	tmp2 = _mm_mul_pd(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
	x2 = _mm_maddsub_pd(h1_real, x2, _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1)));
#else
	x2 = _mm_addsub_pd( _mm_mul_pd(h1_real, x2), _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1)));
#endif
	tmp3 = _mm_mul_pd(h1_imag, x3);
#ifdef __ELPA_USE_FMA__
	x3 = _mm_maddsub_pd(h1_real, x3, _mm_shuffle_pd(tmp3, tmp3, _MM_SHUFFLE2(0,1)));
#else
	x3 = _mm_addsub_pd( _mm_mul_pd(h1_real, x3), _mm_shuffle_pd(tmp3, tmp3, _MM_SHUFFLE2(0,1)));
#endif
	tmp4 = _mm_mul_pd(h1_imag, x4);
#ifdef __ELPA_USE_FMA__
	x4 = _mm_maddsub_pd(h1_real, x4, _mm_shuffle_pd(tmp4, tmp4, _MM_SHUFFLE2(0,1)));
#else
	x4 = _mm_addsub_pd( _mm_mul_pd(h1_real, x4), _mm_shuffle_pd(tmp4, tmp4, _MM_SHUFFLE2(0,1)));
#endif

	q1 = _mm_load_pd(&q_dbl[0]);
	q2 = _mm_load_pd(&q_dbl[2]);
	q3 = _mm_load_pd(&q_dbl[4]);
	q4 = _mm_load_pd(&q_dbl[6]);

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

	_mm_store_pd(&q_dbl[0], q1);
	_mm_store_pd(&q_dbl[2], q2);
	_mm_store_pd(&q_dbl[4], q3);
	_mm_store_pd(&q_dbl[6], q4);

	for (i = 1; i < nb; i++)
	{
		h1_real = _mm_loaddup_pd(&hh_dbl[i*2]);
		h1_imag = _mm_loaddup_pd(&hh_dbl[(i*2)+1]);

		q1 = _mm_load_pd(&q_dbl[(2*i*ldq)+0]);
		q2 = _mm_load_pd(&q_dbl[(2*i*ldq)+2]);
		q3 = _mm_load_pd(&q_dbl[(2*i*ldq)+4]);
		q4 = _mm_load_pd(&q_dbl[(2*i*ldq)+6]);

		tmp1 = _mm_mul_pd(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
		q1 = _mm_add_pd(q1, _mm_maddsub_pd(h1_real, x1, _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1))));
#else
		q1 = _mm_add_pd(q1, _mm_addsub_pd( _mm_mul_pd(h1_real, x1), _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1))));
#endif
		tmp2 = _mm_mul_pd(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
		q2 = _mm_add_pd(q2, _mm_maddsub_pd(h1_real, x2, _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1))));
#else
		q2 = _mm_add_pd(q2, _mm_addsub_pd( _mm_mul_pd(h1_real, x2), _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1))));
#endif
		tmp3 = _mm_mul_pd(h1_imag, x3);
#ifdef __ELPA_USE_FMA__
		q3 = _mm_add_pd(q3, _mm_maddsub_pd(h1_real, x3, _mm_shuffle_pd(tmp3, tmp3, _MM_SHUFFLE2(0,1))));
#else
		q3 = _mm_add_pd(q3, _mm_addsub_pd( _mm_mul_pd(h1_real, x3), _mm_shuffle_pd(tmp3, tmp3, _MM_SHUFFLE2(0,1))));
#endif
		tmp4 = _mm_mul_pd(h1_imag, x4);
#ifdef __ELPA_USE_FMA__
		q4 = _mm_add_pd(q4, _mm_maddsub_pd(h1_real, x4, _mm_shuffle_pd(tmp4, tmp4, _MM_SHUFFLE2(0,1))));
#else
		q4 = _mm_add_pd(q4, _mm_addsub_pd( _mm_mul_pd(h1_real, x4), _mm_shuffle_pd(tmp4, tmp4, _MM_SHUFFLE2(0,1))));
#endif

		_mm_store_pd(&q_dbl[(2*i*ldq)+0], q1);
		_mm_store_pd(&q_dbl[(2*i*ldq)+2], q2);
		_mm_store_pd(&q_dbl[(2*i*ldq)+4], q3);
		_mm_store_pd(&q_dbl[(2*i*ldq)+6], q4);
	}
}

static __forceinline void hh_trafo_complex_kernel_2_SSE_1hv(std::complex<double>* q, std::complex<double>* hh, int nb, int ldq)
{
	double* q_dbl = (double*)q;
	double* hh_dbl = (double*)hh;

	__m128d x1, x2;
	__m128d q1, q2;
	__m128d h1_real, h1_imag;
	__m128d tmp1, tmp2;
	int i=0;

	__m128d sign = (__m128d)_mm_set_epi64x(0x8000000000000000, 0x8000000000000000);

	x1 = _mm_load_pd(&q_dbl[0]);
	x2 = _mm_load_pd(&q_dbl[2]);

	for (i = 1; i < nb; i++)
	{
		h1_real = _mm_loaddup_pd(&hh_dbl[i*2]);
		h1_imag = _mm_loaddup_pd(&hh_dbl[(i*2)+1]);
#ifndef __ELPA_USE_FMA__
		// conjugate
		h1_imag = _mm_xor_pd(h1_imag, sign);
#endif

		q1 = _mm_load_pd(&q_dbl[(2*i*ldq)+0]);
		q2 = _mm_load_pd(&q_dbl[(2*i*ldq)+2]);

		tmp1 = _mm_mul_pd(h1_imag, q1);
#ifdef __ELPA_USE_FMA__
		x1 = _mm_add_pd(x1, _mm_msubadd_pd(h1_real, q1, _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1))));
#else
		x1 = _mm_add_pd(x1, _mm_addsub_pd( _mm_mul_pd(h1_real, q1), _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1))));
#endif
		tmp2 = _mm_mul_pd(h1_imag, q2);
#ifdef __ELPA_USE_FMA__
		x2 = _mm_add_pd(x2, _mm_msubadd_pd(h1_real, q2, _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1))));
#else
		x2 = _mm_add_pd(x2, _mm_addsub_pd( _mm_mul_pd(h1_real, q2), _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1))));
#endif
	}

	h1_real = _mm_loaddup_pd(&hh_dbl[0]);
	h1_imag = _mm_loaddup_pd(&hh_dbl[1]);
	h1_real = _mm_xor_pd(h1_real, sign);
	h1_imag = _mm_xor_pd(h1_imag, sign);

	tmp1 = _mm_mul_pd(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
	x1 = _mm_maddsub_pd(h1_real, x1, _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1)));
#else
	x1 = _mm_addsub_pd( _mm_mul_pd(h1_real, x1), _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1)));
#endif
	tmp2 = _mm_mul_pd(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
	x2 = _mm_maddsub_pd(h1_real, x2, _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1)));
#else
	x2 = _mm_addsub_pd( _mm_mul_pd(h1_real, x2), _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1)));
#endif

	q1 = _mm_load_pd(&q_dbl[0]);
	q2 = _mm_load_pd(&q_dbl[2]);

	q1 = _mm_add_pd(q1, x1);
	q2 = _mm_add_pd(q2, x2);

	_mm_store_pd(&q_dbl[0], q1);
	_mm_store_pd(&q_dbl[2], q2);

	for (i = 1; i < nb; i++)
	{
		h1_real = _mm_loaddup_pd(&hh_dbl[i*2]);
		h1_imag = _mm_loaddup_pd(&hh_dbl[(i*2)+1]);

		q1 = _mm_load_pd(&q_dbl[(2*i*ldq)+0]);
		q2 = _mm_load_pd(&q_dbl[(2*i*ldq)+2]);

		tmp1 = _mm_mul_pd(h1_imag, x1);
#ifdef __ELPA_USE_FMA__
		q1 = _mm_add_pd(q1, _mm_maddsub_pd(h1_real, x1, _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1))));
#else
		q1 = _mm_add_pd(q1, _mm_addsub_pd( _mm_mul_pd(h1_real, x1), _mm_shuffle_pd(tmp1, tmp1, _MM_SHUFFLE2(0,1))));
#endif
		tmp2 = _mm_mul_pd(h1_imag, x2);
#ifdef __ELPA_USE_FMA__
		q2 = _mm_add_pd(q2, _mm_maddsub_pd(h1_real, x2, _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1))));
#else
		q2 = _mm_add_pd(q2, _mm_addsub_pd( _mm_mul_pd(h1_real, x2), _mm_shuffle_pd(tmp2, tmp2, _MM_SHUFFLE2(0,1))));
#endif

		_mm_store_pd(&q_dbl[(2*i*ldq)+0], q1);
		_mm_store_pd(&q_dbl[(2*i*ldq)+2], q2);
	}
}
} // extern C