Skip to content

GitLab

Commit 207febce authored by Andreas Marek's avatar Andreas Marek
Browse files

Comment unused code paths in AVX kernels

parent f581320b
Hide whitespace changes
Inline Side-by-side
src/elpa2_kernels/elpa2_kernels_real_avx-avx2_6hv.c
View file @ 207febce
......@@ -272,145 +272,145 @@ void hexa_hh_trafo_real_avx_avx2_6hv(double* q, double* hh, int* pnb, int* pnq,
//#endif
}
#if 0
void hexa_hh_trafo_fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh)
{
int i;
int nb = *pnb;
int nq = *pldq;
int ldq = *pldq;
int ldh = *pldh;
// calculating scalar products to compute
// 6 householder vectors simultaneously
double scalarprods[15];
// scalarprods[0] = s_1_2;
// scalarprods[1] = s_1_3;
// scalarprods[2] = s_2_3;
// scalarprods[3] = s_1_4;
// scalarprods[4] = s_2_4;
// scalarprods[5] = s_3_4;
// scalarprods[6] = s_1_5;
// scalarprods[7] = s_2_5;
// scalarprods[8] = s_3_5;
// scalarprods[9] = s_4_5;
// scalarprods[10] = s_1_6;
// scalarprods[11] = s_2_6;
// scalarprods[12] = s_3_6;
// scalarprods[13] = s_4_6;
// scalarprods[14] = s_5_6;
scalarprods[0] = hh[(ldh+1)];
scalarprods[1] = hh[(ldh*2)+2];
scalarprods[2] = hh[(ldh*2)+1];
scalarprods[3] = hh[(ldh*3)+3];
scalarprods[4] = hh[(ldh*3)+2];
scalarprods[5] = hh[(ldh*3)+1];
scalarprods[6] = hh[(ldh*4)+4];
scalarprods[7] = hh[(ldh*4)+3];
scalarprods[8] = hh[(ldh*4)+2];
scalarprods[9] = hh[(ldh*4)+1];
scalarprods[10] = hh[(ldh*5)+5];
scalarprods[11] = hh[(ldh*5)+4];
scalarprods[12] = hh[(ldh*5)+3];
scalarprods[13] = hh[(ldh*5)+2];
scalarprods[14] = hh[(ldh*5)+1];
// calculate scalar product of first and fourth householder vector
// loop counter = 2
scalarprods[0] += hh[1] * hh[(2+ldh)];
scalarprods[2] += hh[(ldh)+1] * hh[2+(ldh*2)];
scalarprods[5] += hh[(ldh*2)+1] * hh[2+(ldh*3)];
scalarprods[9] += hh[(ldh*3)+1] * hh[2+(ldh*4)];
scalarprods[14] += hh[(ldh*4)+1] * hh[2+(ldh*5)];
// loop counter = 3
scalarprods[0] += hh[2] * hh[(3+ldh)];
scalarprods[2] += hh[(ldh)+2] * hh[3+(ldh*2)];
scalarprods[5] += hh[(ldh*2)+2] * hh[3+(ldh*3)];
scalarprods[9] += hh[(ldh*3)+2] * hh[3+(ldh*4)];
scalarprods[14] += hh[(ldh*4)+2] * hh[3+(ldh*5)];
scalarprods[1] += hh[1] * hh[3+(ldh*2)];
scalarprods[4] += hh[(ldh*1)+1] * hh[3+(ldh*3)];
scalarprods[8] += hh[(ldh*2)+1] * hh[3+(ldh*4)];
scalarprods[13] += hh[(ldh*3)+1] * hh[3+(ldh*5)];
// loop counter = 4
scalarprods[0] += hh[3] * hh[(4+ldh)];
scalarprods[2] += hh[(ldh)+3] * hh[4+(ldh*2)];
scalarprods[5] += hh[(ldh*2)+3] * hh[4+(ldh*3)];
scalarprods[9] += hh[(ldh*3)+3] * hh[4+(ldh*4)];
scalarprods[14] += hh[(ldh*4)+3] * hh[4+(ldh*5)];
scalarprods[1] += hh[2] * hh[4+(ldh*2)];
scalarprods[4] += hh[(ldh*1)+2] * hh[4+(ldh*3)];
scalarprods[8] += hh[(ldh*2)+2] * hh[4+(ldh*4)];
scalarprods[13] += hh[(ldh*3)+2] * hh[4+(ldh*5)];
scalarprods[3] += hh[1] * hh[4+(ldh*3)];
scalarprods[7] += hh[(ldh)+1] * hh[4+(ldh*4)];
scalarprods[12] += hh[(ldh*2)+1] * hh[4+(ldh*5)];
// loop counter = 5
scalarprods[0] += hh[4] * hh[(5+ldh)];
scalarprods[2] += hh[(ldh)+4] * hh[5+(ldh*2)];
scalarprods[5] += hh[(ldh*2)+4] * hh[5+(ldh*3)];
scalarprods[9] += hh[(ldh*3)+4] * hh[5+(ldh*4)];
scalarprods[14] += hh[(ldh*4)+4] * hh[5+(ldh*5)];
scalarprods[1] += hh[3] * hh[5+(ldh*2)];
scalarprods[4] += hh[(ldh*1)+3] * hh[5+(ldh*3)];
scalarprods[8] += hh[(ldh*2)+3] * hh[5+(ldh*4)];
scalarprods[13] += hh[(ldh*3)+3] * hh[5+(ldh*5)];
scalarprods[3] += hh[2] * hh[5+(ldh*3)];
scalarprods[7] += hh[(ldh)+2] * hh[5+(ldh*4)];
scalarprods[12] += hh[(ldh*2)+2] * hh[5+(ldh*5)];
scalarprods[6] += hh[1] * hh[5+(ldh*4)];
scalarprods[11] += hh[(ldh)+1] * hh[5+(ldh*5)];
#pragma ivdep
for (i = 6; i < nb; i++)
{
scalarprods[0] += hh[i-1] * hh[(i+ldh)];
scalarprods[2] += hh[(ldh)+i-1] * hh[i+(ldh*2)];
scalarprods[5] += hh[(ldh*2)+i-1] * hh[i+(ldh*3)];
scalarprods[9] += hh[(ldh*3)+i-1] * hh[i+(ldh*4)];
scalarprods[14] += hh[(ldh*4)+i-1] * hh[i+(ldh*5)];
scalarprods[1] += hh[i-2] * hh[i+(ldh*2)];
scalarprods[4] += hh[(ldh*1)+i-2] * hh[i+(ldh*3)];
scalarprods[8] += hh[(ldh*2)+i-2] * hh[i+(ldh*4)];
scalarprods[13] += hh[(ldh*3)+i-2] * hh[i+(ldh*5)];
scalarprods[3] += hh[i-3] * hh[i+(ldh*3)];
scalarprods[7] += hh[(ldh)+i-3] * hh[i+(ldh*4)];
scalarprods[12] += hh[(ldh*2)+i-3] * hh[i+(ldh*5)];
scalarprods[6] += hh[i-4] * hh[i+(ldh*4)];
scalarprods[11] += hh[(ldh)+i-4] * hh[i+(ldh*5)];
scalarprods[10] += hh[i-5] * hh[i+(ldh*5)];
}
// Production level kernel calls with padding
//#ifdef __AVX__
for (i = 0; i < nq; i+=8)
{
hh_trafo_kernel_8_AVX_6hv(&q[i], hh, nb, ldq, ldh, scalarprods);
}
//#else
// for (i = 0; i < nq; i+=4)
//#if 0
//void hexa_hh_trafo_fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh)
//{
// int i;
// int nb = *pnb;
// int nq = *pldq;
// int ldq = *pldq;
// int ldh = *pldh;
//
// // calculating scalar products to compute
// // 6 householder vectors simultaneously
// double scalarprods[15];
//
//// scalarprods[0] = s_1_2;
//// scalarprods[1] = s_1_3;
//// scalarprods[2] = s_2_3;
//// scalarprods[3] = s_1_4;
//// scalarprods[4] = s_2_4;
//// scalarprods[5] = s_3_4;
//// scalarprods[6] = s_1_5;
//// scalarprods[7] = s_2_5;
//// scalarprods[8] = s_3_5;
//// scalarprods[9] = s_4_5;
//// scalarprods[10] = s_1_6;
//// scalarprods[11] = s_2_6;
//// scalarprods[12] = s_3_6;
//// scalarprods[13] = s_4_6;
//// scalarprods[14] = s_5_6;
//
// scalarprods[0] = hh[(ldh+1)];
// scalarprods[1] = hh[(ldh*2)+2];
// scalarprods[2] = hh[(ldh*2)+1];
// scalarprods[3] = hh[(ldh*3)+3];
// scalarprods[4] = hh[(ldh*3)+2];
// scalarprods[5] = hh[(ldh*3)+1];
// scalarprods[6] = hh[(ldh*4)+4];
// scalarprods[7] = hh[(ldh*4)+3];
// scalarprods[8] = hh[(ldh*4)+2];
// scalarprods[9] = hh[(ldh*4)+1];
// scalarprods[10] = hh[(ldh*5)+5];
// scalarprods[11] = hh[(ldh*5)+4];
// scalarprods[12] = hh[(ldh*5)+3];
// scalarprods[13] = hh[(ldh*5)+2];
// scalarprods[14] = hh[(ldh*5)+1];
//
// // calculate scalar product of first and fourth householder vector
// // loop counter = 2
// scalarprods[0] += hh[1] * hh[(2+ldh)];
// scalarprods[2] += hh[(ldh)+1] * hh[2+(ldh*2)];
// scalarprods[5] += hh[(ldh*2)+1] * hh[2+(ldh*3)];
// scalarprods[9] += hh[(ldh*3)+1] * hh[2+(ldh*4)];
// scalarprods[14] += hh[(ldh*4)+1] * hh[2+(ldh*5)];
//
// // loop counter = 3
// scalarprods[0] += hh[2] * hh[(3+ldh)];
// scalarprods[2] += hh[(ldh)+2] * hh[3+(ldh*2)];
// scalarprods[5] += hh[(ldh*2)+2] * hh[3+(ldh*3)];
// scalarprods[9] += hh[(ldh*3)+2] * hh[3+(ldh*4)];
// scalarprods[14] += hh[(ldh*4)+2] * hh[3+(ldh*5)];
//
// scalarprods[1] += hh[1] * hh[3+(ldh*2)];
// scalarprods[4] += hh[(ldh*1)+1] * hh[3+(ldh*3)];
// scalarprods[8] += hh[(ldh*2)+1] * hh[3+(ldh*4)];
// scalarprods[13] += hh[(ldh*3)+1] * hh[3+(ldh*5)];
//
// // loop counter = 4
// scalarprods[0] += hh[3] * hh[(4+ldh)];
// scalarprods[2] += hh[(ldh)+3] * hh[4+(ldh*2)];
// scalarprods[5] += hh[(ldh*2)+3] * hh[4+(ldh*3)];
// scalarprods[9] += hh[(ldh*3)+3] * hh[4+(ldh*4)];
// scalarprods[14] += hh[(ldh*4)+3] * hh[4+(ldh*5)];
//
// scalarprods[1] += hh[2] * hh[4+(ldh*2)];
// scalarprods[4] += hh[(ldh*1)+2] * hh[4+(ldh*3)];
// scalarprods[8] += hh[(ldh*2)+2] * hh[4+(ldh*4)];
// scalarprods[13] += hh[(ldh*3)+2] * hh[4+(ldh*5)];
//
// scalarprods[3] += hh[1] * hh[4+(ldh*3)];
// scalarprods[7] += hh[(ldh)+1] * hh[4+(ldh*4)];
// scalarprods[12] += hh[(ldh*2)+1] * hh[4+(ldh*5)];
//
// // loop counter = 5
// scalarprods[0] += hh[4] * hh[(5+ldh)];
// scalarprods[2] += hh[(ldh)+4] * hh[5+(ldh*2)];
// scalarprods[5] += hh[(ldh*2)+4] * hh[5+(ldh*3)];
// scalarprods[9] += hh[(ldh*3)+4] * hh[5+(ldh*4)];
// scalarprods[14] += hh[(ldh*4)+4] * hh[5+(ldh*5)];
//
// scalarprods[1] += hh[3] * hh[5+(ldh*2)];
// scalarprods[4] += hh[(ldh*1)+3] * hh[5+(ldh*3)];
// scalarprods[8] += hh[(ldh*2)+3] * hh[5+(ldh*4)];
// scalarprods[13] += hh[(ldh*3)+3] * hh[5+(ldh*5)];
//
// scalarprods[3] += hh[2] * hh[5+(ldh*3)];
// scalarprods[7] += hh[(ldh)+2] * hh[5+(ldh*4)];
// scalarprods[12] += hh[(ldh*2)+2] * hh[5+(ldh*5)];
//
// scalarprods[6] += hh[1] * hh[5+(ldh*4)];
// scalarprods[11] += hh[(ldh)+1] * hh[5+(ldh*5)];
//
// #pragma ivdep
// for (i = 6; i < nb; i++)
// {
// hh_trafo_kernel_4_SSE_6hv(&q[i], hh, nb, ldq, ldh, scalarprods);
// scalarprods[0] += hh[i-1] * hh[(i+ldh)];
// scalarprods[2] += hh[(ldh)+i-1] * hh[i+(ldh*2)];
// scalarprods[5] += hh[(ldh*2)+i-1] * hh[i+(ldh*3)];
// scalarprods[9] += hh[(ldh*3)+i-1] * hh[i+(ldh*4)];
// scalarprods[14] += hh[(ldh*4)+i-1] * hh[i+(ldh*5)];
//
// scalarprods[1] += hh[i-2] * hh[i+(ldh*2)];
// scalarprods[4] += hh[(ldh*1)+i-2] * hh[i+(ldh*3)];
// scalarprods[8] += hh[(ldh*2)+i-2] * hh[i+(ldh*4)];
// scalarprods[13] += hh[(ldh*3)+i-2] * hh[i+(ldh*5)];
//
// scalarprods[3] += hh[i-3] * hh[i+(ldh*3)];
// scalarprods[7] += hh[(ldh)+i-3] * hh[i+(ldh*4)];
// scalarprods[12] += hh[(ldh*2)+i-3] * hh[i+(ldh*5)];
//
// scalarprods[6] += hh[i-4] * hh[i+(ldh*4)];
// scalarprods[11] += hh[(ldh)+i-4] * hh[i+(ldh*5)];
//
// scalarprods[10] += hh[i-5] * hh[i+(ldh*5)];
// }
//
//
// // Production level kernel calls with padding
////#ifdef __AVX__
// for (i = 0; i < nq; i+=8)
// {
// hh_trafo_kernel_8_AVX_6hv(&q[i], hh, nb, ldq, ldh, scalarprods);
// }
////#else
//// for (i = 0; i < nq; i+=4)
//// {
//// hh_trafo_kernel_4_SSE_6hv(&q[i], hh, nb, ldq, ldh, scalarprods);
//// }
////#endif
//}
//#endif
}
#endif
/**
* Unrolled kernel that computes
......
src/elpa2_kernels/elpa2_kernels_real_sse_4hv.c
View file @ 207febce
......@@ -163,65 +163,67 @@ void quad_hh_trafo_real_sse_4hv(double* q, double* hh, int* pnb, int* pnq, int*
}
}
#if 0
void quad_hh_trafo_fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh)
{
int i;
int nb = *pnb;
int nq = *pldq;
int ldq = *pldq;
int ldh = *pldh;
// calculating scalar products to compute
// 4 householder vectors simultaneously
double s_1_2 = hh[(ldh)+1];
double s_1_3 = hh[(ldh*2)+2];
double s_2_3 = hh[(ldh*2)+1];
double s_1_4 = hh[(ldh*3)+3];
double s_2_4 = hh[(ldh*3)+2];
double s_3_4 = hh[(ldh*3)+1];
// calculate scalar product of first and fourth householder vector
// loop counter = 2
s_1_2 += hh[2-1] * hh[(2+ldh)];
s_2_3 += hh[(ldh)+2-1] * hh[2+(ldh*2)];
s_3_4 += hh[(ldh*2)+2-1] * hh[2+(ldh*3)];
// loop counter = 3
s_1_2 += hh[3-1] * hh[(3+ldh)];
s_2_3 += hh[(ldh)+3-1] * hh[3+(ldh*2)];
s_3_4 += hh[(ldh*2)+3-1] * hh[3+(ldh*3)];
s_1_3 += hh[3-2] * hh[3+(ldh*2)];
s_2_4 += hh[(ldh*1)+3-2] * hh[3+(ldh*3)];
#pragma ivdep
for (i = 4; i < nb; i++)
{
s_1_2 += hh[i-1] * hh[(i+ldh)];
s_2_3 += hh[(ldh)+i-1] * hh[i+(ldh*2)];
s_3_4 += hh[(ldh*2)+i-1] * hh[i+(ldh*3)];
s_1_3 += hh[i-2] * hh[i+(ldh*2)];
s_2_4 += hh[(ldh*1)+i-2] * hh[i+(ldh*3)];
//#if 0
//void quad_hh_trafo_fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh)
//{
// int i;
// int nb = *pnb;
// int nq = *pldq;
// int ldq = *pldq;
// int ldh = *pldh;
//
// // calculating scalar products to compute
// // 4 householder vectors simultaneously
// double s_1_2 = hh[(ldh)+1];
// double s_1_3 = hh[(ldh*2)+2];
// double s_2_3 = hh[(ldh*2)+1];
// double s_1_4 = hh[(ldh*3)+3];
// double s_2_4 = hh[(ldh*3)+2];
// double s_3_4 = hh[(ldh*3)+1];
//
// // calculate scalar product of first and fourth householder vector
// // loop counter = 2
// s_1_2 += hh[2-1] * hh[(2+ldh)];
// s_2_3 += hh[(ldh)+2-1] * hh[2+(ldh*2)];
// s_3_4 += hh[(ldh*2)+2-1] * hh[2+(ldh*3)];
//
// // loop counter = 3
// s_1_2 += hh[3-1] * hh[(3+ldh)];
// s_2_3 += hh[(ldh)+3-1] * hh[3+(ldh*2)];
// s_3_4 += hh[(ldh*2)+3-1] * hh[3+(ldh*3)];
//
// s_1_3 += hh[3-2] * hh[3+(ldh*2)];
// s_2_4 += hh[(ldh*1)+3-2] * hh[3+(ldh*3)];
//
// #pragma ivdep
// for (i = 4; i < nb; i++)
// {
// s_1_2 += hh[i-1] * hh[(i+ldh)];
// s_2_3 += hh[(ldh)+i-1] * hh[i+(ldh*2)];
// s_3_4 += hh[(ldh*2)+i-1] * hh[i+(ldh*3)];
//
// s_1_3 += hh[i-2] * hh[i+(ldh*2)];
// s_2_4 += hh[(ldh*1)+i-2] * hh[i+(ldh*3)];
//
// s_1_4 += hh[i-3] * hh[i+(ldh*3)];
// }
//
// // Production level kernel calls with padding
//#ifdef __AVX__
// for (i = 0; i < nq; i+=12)
// {
// hh_trafo_kernel_12_AVX_4hv(&q[i], hh, nb, ldq, ldh, s_1_2, s_1_3, s_2_3, s_1_4, s_2_4, s_3_4);
// }
//#else
// for (i = 0; i < nq; i+=6)
// {
// hh_trafo_kernel_6_SSE_4hv(&q[i], hh, nb, ldq, ldh, s_1_2, s_1_3, s_2_3, s_1_4, s_2_4, s_3_4);
// }
//#endif
//}
//#endif
s_1_4 += hh[i-3] * hh[i+(ldh*3)];
}
// Production level kernel calls with padding
#ifdef __AVX__
for (i = 0; i < nq; i+=12)
{
hh_trafo_kernel_12_AVX_4hv(&q[i], hh, nb, ldq, ldh, s_1_2, s_1_3, s_2_3, s_1_4, s_2_4, s_3_4);
}
#else
for (i = 0; i < nq; i+=6)
{
hh_trafo_kernel_6_SSE_4hv(&q[i], hh, nb, ldq, ldh, s_1_2, s_1_3, s_2_3, s_1_4, s_2_4, s_3_4);
}
#endif
}
#endif
/**
* Unrolled kernel that computes
* 6 rows of Q simultaneously, a
......
src/elpa2_kernels/elpa2_kernels_real_sse_6hv.c
View file @ 207febce
......@@ -243,160 +243,160 @@ void hexa_hh_trafo_real_sse_6hv(double* q, double* hh, int* pnb, int* pnq, int*
}
}
#if 0
void hexa_hh_trafo_fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh)
{
int i;
int nb = *pnb;
int nq = *pldq;
int ldq = *pldq;
int ldh = *pldh;
// calculating scalar products to compute
// 6 householder vectors simultaneously
double scalarprods[15];
// scalarprods[0] = s_1_2;
// scalarprods[1] = s_1_3;
// scalarprods[2] = s_2_3;
// scalarprods[3] = s_1_4;
// scalarprods[4] = s_2_4;
// scalarprods[5] = s_3_4;
// scalarprods[6] = s_1_5;
// scalarprods[7] = s_2_5;
// scalarprods[8] = s_3_5;
// scalarprods[9] = s_4_5;
// scalarprods[10] = s_1_6;
// scalarprods[11] = s_2_6;
// scalarprods[12] = s_3_6;
// scalarprods[13] = s_4_6;
// scalarprods[14] = s_5_6;
scalarprods[0] = hh[(ldh+1)];
scalarprods[1] = hh[(ldh*2)+2];
scalarprods[2] = hh[(ldh*2)+1];
scalarprods[3] = hh[(ldh*3)+3];
scalarprods[4] = hh[(ldh*3)+2];
scalarprods[5] = hh[(ldh*3)+1];
scalarprods[6] = hh[(ldh*4)+4];
scalarprods[7] = hh[(ldh*4)+3];
scalarprods[8] = hh[(ldh*4)+2];
scalarprods[9] = hh[(ldh*4)+1];
scalarprods[10] = hh[(ldh*5)+5];
scalarprods[11] = hh[(ldh*5)+4];
scalarprods[12] = hh[(ldh*5)+3];
scalarprods[13] = hh[(ldh*5)+2];
scalarprods[14] = hh[(ldh*5)+1];
// calculate scalar product of first and fourth householder vector
// loop counter = 2
scalarprods[0] += hh[1] * hh[(2+ldh)];
scalarprods[2] += hh[(ldh)+1] * hh[2+(ldh*2)];
scalarprods[5] += hh[(ldh*2)+1] * hh[2+(ldh*3)];
scalarprods[9] += hh[(ldh*3)+1] * hh[2+(ldh*4)];
scalarprods[14] += hh[(ldh*4)+1] * hh[2+(ldh*5)];
// loop counter = 3
scalarprods[0] += hh[2] * hh[(3+ldh)];
scalarprods[2] += hh[(ldh)+2] * hh[3+(ldh*2)];
scalarprods[5] += hh[(ldh*2)+2] * hh[3+(ldh*3)];
scalarprods[9] += hh[(ldh*3)+2] * hh[3+(ldh*4)];
scalarprods[14] += hh[(ldh*4)+2] * hh[3+(ldh*5)];
scalarprods[1] += hh[1] * hh[3+(ldh*2)];
scalarprods[4] += hh[(ldh*1)+1] * hh[3+(ldh*3)];
scalarprods[8] += hh[(ldh*2)+1] * hh[3+(ldh*4)];
scalarprods[13] += hh[(ldh*3)+1] * hh[3+(ldh*5)];
// loop counter = 4
scalarprods[0] += hh[3] * hh[(4+ldh)];
scalarprods[2] += hh[(ldh)+3] * hh[4+(ldh*2)];
scalarprods[5] += hh[(ldh*2)+3] * hh[4+(ldh*3)];
scalarprods[9] += hh[(ldh*3)+3] * hh[4+(ldh*4)];
scalarprods[14] += hh[(ldh*4)+3] * hh[4+(ldh*5)];
scalarprods[1] += hh[2] * hh[4+(ldh*2)];
scalarprods[4] += hh[(ldh*1)+2] * hh[4+(ldh*3)];
scalarprods[8] += hh[(ldh*2)+2] * hh[4+(ldh*4)];
scalarprods[13] += hh[(ldh*3)+2] * hh[4+(ldh*5)];
scalarprods[3] += hh[1] * hh[4+(ldh*3)];
scalarprods[7] += hh[(ldh)+1] * hh[4+(ldh*4)];
scalarprods[12] += hh[(ldh*2)+1] * hh[4+(ldh*5)];
// loop counter = 5
scalarprods[0] += hh[4] * hh[(5+ldh)];
scalarprods[2] += hh[(ldh)+4] * hh[5+(ldh*2)];
scalarprods[5] += hh[(ldh*2)+4] * hh[5+(ldh*3)];
scalarprods[9] += hh[(ldh*3)+4] * hh[5+(ldh*4)];
scalarprods[14] += hh[(ldh*4)+4] * hh[5+(ldh*5)];
scalarprods[1] += hh[3] * hh[5+(ldh*2)];
scalarprods[4] += hh[(ldh*1)+3] * hh[5+(ldh*3)];
scalarprods[8] += hh[(ldh*2)+3] * hh[5+(ldh*4)];
scalarprods[13] += hh[(ldh*3)+3] * hh[5+(ldh*5)];
scalarprods[3] += hh[2] * hh[5+(ldh*3)];
scalarprods[7] += hh[(ldh)+2] * hh[5+(ldh*4)];
scalarprods[12] += hh[(ldh*2)+2] * hh[5+(ldh*5)];
scalarprods[6] += hh[1] * hh[5+(ldh*4)];
scalarprods[11] += hh[(ldh)+1] * hh[5+(ldh*5)];
#pragma ivdep
for (i = 6; i < nb; i++)
{
scalarprods[0] += hh[i-1] * hh[(i+ldh)];
scalarprods[2] += hh[(ldh)+i-1] * hh[i+(ldh*2)];
scalarprods[5] += hh[(ldh*2)+i-1] * hh[i+(ldh*3)];
scalarprods[9] += hh[(ldh*3)+i-1] * hh[i+(ldh*4)];
scalarprods[14] += hh[(ldh*4)+i-1] * hh[i+(ldh*5)];
scalarprods[1] += hh[i-2] * hh[i+(ldh*2)];
scalarprods[4] += hh[(ldh*1)+i-2] * hh[i+(ldh*3)];
scalarprods[8] += hh[(ldh*2)+i-2] * hh[i+(ldh*4)];
scalarprods[13] += hh[(ldh*3)+i-2] * hh[i+(ldh*5)];
scalarprods[3] += hh[i-3] * hh[i+(ldh*3)];
scalarprods[7] += hh[(ldh)+i-3] * hh[i+(ldh*4)];
scalarprods[12] += hh[(ldh*2)+i-3] * hh[i+(ldh*5)];
scalarprods[6] += hh[i-4] * hh[i+(ldh*4)];
scalarprods[11] += hh[(ldh)+i-4] * hh[i+(ldh*5)];
scalarprods[10] += hh[i-5] * hh[i+(ldh*5)];
}
// printf("s_1_2: %f\n", scalarprods[0]);
// printf("s_1_3: %f\n", scalarprods[1]);
// printf("s_2_3: %f\n", scalarprods[2]);
// printf("s_1_4: %f\n", scalarprods[3]);
// printf("s_2_4: %f\n", scalarprods[4]);
// printf("s_3_4: %f\n", scalarprods[5]);
// printf("s_1_5: %f\n", scalarprods[6]);
// printf("s_2_5: %f\n", scalarprods[7]);
// printf("s_3_5: %f\n", scalarprods[8]);
// printf("s_4_5: %f\n", scalarprods[9]);
// printf("s_1_6: %f\n", scalarprods[10]);
// printf("s_2_6: %f\n", scalarprods[11]);
// printf("s_3_6: %f\n", scalarprods[12]);
// printf("s_4_6: %f\n", scalarprods[13]);
// printf("s_5_6: %f\n", scalarprods[14]);
// Production level kernel calls with padding
#ifdef __AVX__
for (i = 0; i < nq; i+=8)
{
hh_trafo_kernel_8_AVX_6hv(&q[i], hh, nb, ldq, ldh, scalarprods);
}
#else
for (i = 0; i < nq; i+=4)
{
hh_trafo_kernel_4_SSE_6hv(&q[i], hh, nb, ldq, ldh, scalarprods);
}
#endif
}
#endif
//#if 0
//void hexa_hh_trafo_fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh)
//{
// int i;
// int nb = *pnb;
// int nq = *pldq;
// int ldq = *pldq;
// int ldh = *pldh;
//
// // calculating scalar products to compute
// // 6 householder vectors simultaneously
// double scalarprods[15];
//
//// scalarprods[0] = s_1_2;
//// scalarprods[1] = s_1_3;
//// scalarprods[2] = s_2_3;
//// scalarprods[3] = s_1_4;
//// scalarprods[4] = s_2_4;
//// scalarprods[5] = s_3_4;
//// scalarprods[6] = s_1_5;
//// scalarprods[7] = s_2_5;
//// scalarprods[8] = s_3_5;
//// scalarprods[9] = s_4_5;
//// scalarprods[10] = s_1_6;
//// scalarprods[11] = s_2_6;
//// scalarprods[12] = s_3_6;
//// scalarprods[13] = s_4_6;
//// scalarprods[14] = s_5_6;
//
// scalarprods[0] = hh[(ldh+1)];
// scalarprods[1] = hh[(ldh*2)+2];
// scalarprods[2] = hh[(ldh*2)+1];
// scalarprods[3] = hh[(ldh*3)+3];
// scalarprods[4] = hh[(ldh*3)+2];
// scalarprods[5] = hh[(ldh*3)+1];
// scalarprods[6] = hh[(ldh*4)+4];
// scalarprods[7] = hh[(ldh*4)+3];
// scalarprods[8] = hh[(ldh*4)+2];
// scalarprods[9] = hh[(ldh*4)+1];
// scalarprods[10] = hh[(ldh*5)+5];
// scalarprods[11] = hh[(ldh*5)+4];
// scalarprods[12] = hh[(ldh*5)+3];
// scalarprods[13] = hh[(ldh*5)+2];
// scalarprods[14] = hh[(ldh*5)+1];
//
// // calculate scalar product of first and fourth householder vector
// // loop counter = 2
// scalarprods[0] += hh[1] * hh[(2+ldh)];
// scalarprods[2] += hh[(ldh)+1] * hh[2+(ldh*2)];
// scalarprods[5] += hh[(ldh*2)+1] * hh[2+(ldh*3)];
// scalarprods[9] += hh[(ldh*3)+1] * hh[2+(ldh*4)];
// scalarprods[14] += hh[(ldh*4)+1] * hh[2+(ldh*5)];
//
// // loop counter = 3
// scalarprods[0] += hh[2] * hh[(3+ldh)];
// scalarprods[2] += hh[(ldh)+2] * hh[3+(ldh*2)];
// scalarprods[5] += hh[(ldh*2)+2] * hh[3+(ldh*3)];
// scalarprods[9] += hh[(ldh*3)+2] * hh[3+(ldh*4)];
// scalarprods[14] += hh[(ldh*4)+2] * hh[3+(ldh*5)];
//