Skip to content

GitLab

Commit b57f0f88 authored by Andreas Marek's avatar Andreas Marek
Browse files

Only AVX block1 kernel

parent 28a03265
Hide whitespace changes
Inline Side-by-side
Makefile.am
View file @ b57f0f88
......@@ -794,6 +794,7 @@ EXTRA_DIST = \
src/elpa2/kernels/complex_avx512_1hv_template.c \
src/elpa2/kernels/complex_avx512_2hv_template.c \
src/elpa2/kernels/complex_128bit_256bit_512bit_BLOCK_template.c \
src/elpa2/kernels/complex_avx-avx2_2hv_template.c \
src/elpa2/kernels/complex_template.F90 \
src/elpa2/kernels/real_vsx_4hv_template.c \
src/elpa2/kernels/real_vsx_6hv_template.c \
......
src/elpa2/kernels/complex_128bit_256bit_512bit_BLOCK_template.c
View file @ b57f0f88
......@@ -178,10 +178,9 @@
#define _mm256_FMSUBADD_pd(a,b,c) _mm256_fmsubadd_pd(a,b,c)
#endif
#endif /* HAVE_AVX2 */
#define _SIMD_FMADDSUB _mm256_FMADDSUB_pd
#define _SIMD_FMSUBADD _mm256_FMSUBADD_pd
#endif /* HAVE_AVX2 */
#endif /* DOUBLE_PRECISION_COMPLEX */
......@@ -215,10 +214,9 @@
#define _mm256_FMSUBADD_ps(a,b,c) _mm256_fmsubadd_ps(a,b,c)
#endif
#endif /* HAVE_AVX2 */
#define _SIMD_FMADDSUB _mm256_FMADDSUB_ps
#define _SIMD_FMSUBADD _mm256_FMSUBADD_ps
#endif /* HAVE_AVX2 */
#endif /* SINGLE_PRECISION_COMPLEX */
......@@ -530,36 +528,6 @@ static __forceinline void CONCAT_8ARGS(hh_trafo_complex_kernel_,ROW_LENGTH,_,SIM
!f>#endif
*/
/*
!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
*/
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
)
......
src/elpa2/kernels/complex_avx-avx2_2hv_double_precision.c
View file @ b57f0f88
......@@ -51,7 +51,7 @@
#define VEC_SET AVX_256
#define BLOCK2 1
#include "../../general/precision_macros.h"
#include "complex_128bit_256bit_512bit_BLOCK_template.c"
#include "complex_avx-avx2_2hv_template.c"
#undef VEC_SET
#undef BLOCK2
#undef DOUBLE_PRECISION
......
src/elpa2/kernels/complex_avx-avx2_2hv_single_precision.c
View file @ b57f0f88
......@@ -51,7 +51,7 @@
#define VEC_SET AVX_256
#define BLOCK2 1
#include "../../general/precision_macros.h"
#include "complex_128bit_256bit_512bit_BLOCK_template.c"
#include "complex_avx-avx2_2hv_template.c"
#undef VEC_SET
#undef BLOCK2
#undef SINGLE_PRECISION
......
src/elpa2/kernels/complex_avx-avx2_2hv_template.c 0 → 100644
View file @ b57f0f88
// 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.
//
//
// --------------------------------------------------------------------------------------------------
//
// 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>
#include <stdio.h>
#include <stdlib.h>
#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
#define _AVX_SET1 _mm256_set1_pd
#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
#define _AVX_SET1 _mm256_set1_ps
#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
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);
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) &
!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
*/
#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) &
!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
*/
#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;
int worked_on;
worked_on = 0;
#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
for (i = 0; i < nq-6; i+=8)
{
hh_trafo_complex_kernel_8_AVX_2hv_double(&q[i], hh, nb, ldq, ldh, s);
worked_on += 8;
}
#endif
#ifdef SINGLE_PRECISION_COMPLEX
for (i = 0; i < nq-12; i+=16)
{
hh_trafo_complex_kernel_16_AVX_2hv_single(&q[i], hh, nb, ldq, ldh, s , s);
worked_on += 16;
}
#endif
if (nq-i == 0) {
return;
}
#ifdef DOUBLE_PRECISION_COMPLEX
if (nq-i == 6) {
hh_trafo_complex_kernel_6_AVX_2hv_double(&q[i], hh, nb, ldq, ldh, s);
worked_on += 6;
}
#endif
#ifdef SINGLE_PRECISION_COMPLEX
if (nq-i == 12) {
hh_trafo_complex_kernel_12_AVX_2hv_single(&q[i], hh, nb, ldq, ldh, s, s);
worked_on += 12;
}
#endif
#ifdef DOUBLE_PRECISION_COMPLEX
if (nq-i == 4) {
hh_trafo_complex_kernel_4_AVX_2hv_double(&q[i], hh, nb, ldq, ldh, s);
worked_on += 4;
}
#endif
#ifdef SINGLE_PRECISION_COMPLEX
if (nq-i == 8) {
hh_trafo_complex_kernel_8_AVX_2hv_single(&q[i], hh, nb, ldq, ldh, s, s);
worked_on += 8;
}
#endif
#ifdef DOUBLE_PRECISION_COMPLEX
if (nq-i == 2) {
hh_trafo_complex_kernel_2_AVX_2hv_double(&q[i], hh, nb, ldq, ldh, s);
worked_on += 2;
}
#endif
#ifdef SINGLE_PRECISION_COMPLEX
if (nq-i == 4) {
hh_trafo_complex_kernel_4_AVX_2hv_single(&q[i], hh, nb, ldq, ldh, s, s);
worked_on += 4;
}
#endif
#ifdef WITH_DEBUG
if (worked_on != nq) {
printf("Error in complex avx-avx2 BLOCK 2 kernel \n");
abort();
}
#endif
}
#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
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)
#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]);
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
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