|
Packit |
67cb25 |
/* blas/source_gemm_c.h
|
|
Packit |
67cb25 |
*
|
|
Packit |
67cb25 |
* Copyright (C) 2001, 2007 Brian Gough
|
|
Packit |
67cb25 |
*
|
|
Packit |
67cb25 |
* This program is free software; you can redistribute it and/or modify
|
|
Packit |
67cb25 |
* it under the terms of the GNU General Public License as published by
|
|
Packit |
67cb25 |
* the Free Software Foundation; either version 3 of the License, or (at
|
|
Packit |
67cb25 |
* your option) any later version.
|
|
Packit |
67cb25 |
*
|
|
Packit |
67cb25 |
* This program is distributed in the hope that it will be useful, but
|
|
Packit |
67cb25 |
* WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
Packit |
67cb25 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
Packit |
67cb25 |
* General Public License for more details.
|
|
Packit |
67cb25 |
*
|
|
Packit |
67cb25 |
* You should have received a copy of the GNU General Public License
|
|
Packit |
67cb25 |
* along with this program; if not, write to the Free Software
|
|
Packit |
67cb25 |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
|
|
Packit |
67cb25 |
*/
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
{
|
|
Packit |
67cb25 |
INDEX i, j, k;
|
|
Packit |
67cb25 |
INDEX n1, n2;
|
|
Packit |
67cb25 |
INDEX ldf, ldg;
|
|
Packit |
67cb25 |
int conjF, conjG, TransF, TransG;
|
|
Packit |
67cb25 |
const BASE *F, *G;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
CHECK_ARGS14(GEMM,Order,TransA,TransB,M,N,K,alpha,A,lda,B,ldb,beta,C,ldc);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
{
|
|
Packit |
67cb25 |
const BASE alpha_real = CONST_REAL0(alpha);
|
|
Packit |
67cb25 |
const BASE alpha_imag = CONST_IMAG0(alpha);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
const BASE beta_real = CONST_REAL0(beta);
|
|
Packit |
67cb25 |
const BASE beta_imag = CONST_IMAG0(beta);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
if ((alpha_real == 0.0 && alpha_imag == 0.0)
|
|
Packit |
67cb25 |
&& (beta_real == 1.0 && beta_imag == 0.0))
|
|
Packit |
67cb25 |
return;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
if (Order == CblasRowMajor) {
|
|
Packit |
67cb25 |
n1 = M;
|
|
Packit |
67cb25 |
n2 = N;
|
|
Packit |
67cb25 |
F = (const BASE *)A;
|
|
Packit |
67cb25 |
ldf = lda;
|
|
Packit |
67cb25 |
conjF = (TransA == CblasConjTrans) ? -1 : 1;
|
|
Packit |
67cb25 |
TransF = (TransA == CblasNoTrans) ? CblasNoTrans : CblasTrans;
|
|
Packit |
67cb25 |
G = (const BASE *)B;
|
|
Packit |
67cb25 |
ldg = ldb;
|
|
Packit |
67cb25 |
conjG = (TransB == CblasConjTrans) ? -1 : 1;
|
|
Packit |
67cb25 |
TransG = (TransB == CblasNoTrans) ? CblasNoTrans : CblasTrans;
|
|
Packit |
67cb25 |
} else {
|
|
Packit |
67cb25 |
n1 = N;
|
|
Packit |
67cb25 |
n2 = M;
|
|
Packit |
67cb25 |
F = (const BASE *)B;
|
|
Packit |
67cb25 |
ldf = ldb;
|
|
Packit |
67cb25 |
conjF = (TransB == CblasConjTrans) ? -1 : 1;
|
|
Packit |
67cb25 |
TransF = (TransB == CblasNoTrans) ? CblasNoTrans : CblasTrans;
|
|
Packit |
67cb25 |
G = (const BASE *)A;
|
|
Packit |
67cb25 |
ldg = lda;
|
|
Packit |
67cb25 |
conjG = (TransA == CblasConjTrans) ? -1 : 1;
|
|
Packit |
67cb25 |
TransG = (TransA == CblasNoTrans) ? CblasNoTrans : CblasTrans;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* form y := beta*y */
|
|
Packit |
67cb25 |
if (beta_real == 0.0 && beta_imag == 0.0) {
|
|
Packit |
67cb25 |
for (i = 0; i < n1; i++) {
|
|
Packit |
67cb25 |
for (j = 0; j < n2; j++) {
|
|
Packit |
67cb25 |
REAL(C, ldc * i + j) = 0.0;
|
|
Packit |
67cb25 |
IMAG(C, ldc * i + j) = 0.0;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
} else if (!(beta_real == 1.0 && beta_imag == 0.0)) {
|
|
Packit |
67cb25 |
for (i = 0; i < n1; i++) {
|
|
Packit |
67cb25 |
for (j = 0; j < n2; j++) {
|
|
Packit |
67cb25 |
const BASE Cij_real = REAL(C, ldc * i + j);
|
|
Packit |
67cb25 |
const BASE Cij_imag = IMAG(C, ldc * i + j);
|
|
Packit |
67cb25 |
REAL(C, ldc * i + j) = beta_real * Cij_real - beta_imag * Cij_imag;
|
|
Packit |
67cb25 |
IMAG(C, ldc * i + j) = beta_real * Cij_imag + beta_imag * Cij_real;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
if (alpha_real == 0.0 && alpha_imag == 0.0)
|
|
Packit |
67cb25 |
return;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
if (TransF == CblasNoTrans && TransG == CblasNoTrans) {
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* form C := alpha*A*B + C */
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
for (k = 0; k < K; k++) {
|
|
Packit |
67cb25 |
for (i = 0; i < n1; i++) {
|
|
Packit |
67cb25 |
const BASE Fik_real = CONST_REAL(F, ldf * i + k);
|
|
Packit |
67cb25 |
const BASE Fik_imag = conjF * CONST_IMAG(F, ldf * i + k);
|
|
Packit |
67cb25 |
const BASE temp_real = alpha_real * Fik_real - alpha_imag * Fik_imag;
|
|
Packit |
67cb25 |
const BASE temp_imag = alpha_real * Fik_imag + alpha_imag * Fik_real;
|
|
Packit |
67cb25 |
if (!(temp_real == 0.0 && temp_imag == 0.0)) {
|
|
Packit |
67cb25 |
for (j = 0; j < n2; j++) {
|
|
Packit |
67cb25 |
const BASE Gkj_real = CONST_REAL(G, ldg * k + j);
|
|
Packit |
67cb25 |
const BASE Gkj_imag = conjG * CONST_IMAG(G, ldg * k + j);
|
|
Packit |
67cb25 |
REAL(C, ldc * i + j) += temp_real * Gkj_real - temp_imag * Gkj_imag;
|
|
Packit |
67cb25 |
IMAG(C, ldc * i + j) += temp_real * Gkj_imag + temp_imag * Gkj_real;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
} else if (TransF == CblasNoTrans && TransG == CblasTrans) {
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* form C := alpha*A*B' + C */
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
for (i = 0; i < n1; i++) {
|
|
Packit |
67cb25 |
for (j = 0; j < n2; j++) {
|
|
Packit |
67cb25 |
BASE temp_real = 0.0;
|
|
Packit |
67cb25 |
BASE temp_imag = 0.0;
|
|
Packit |
67cb25 |
for (k = 0; k < K; k++) {
|
|
Packit |
67cb25 |
const BASE Fik_real = CONST_REAL(F, ldf * i + k);
|
|
Packit |
67cb25 |
const BASE Fik_imag = conjF * CONST_IMAG(F, ldf * i + k);
|
|
Packit |
67cb25 |
const BASE Gjk_real = CONST_REAL(G, ldg * j + k);
|
|
Packit |
67cb25 |
const BASE Gjk_imag = conjG * CONST_IMAG(G, ldg * j + k);
|
|
Packit |
67cb25 |
temp_real += Fik_real * Gjk_real - Fik_imag * Gjk_imag;
|
|
Packit |
67cb25 |
temp_imag += Fik_real * Gjk_imag + Fik_imag * Gjk_real;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
REAL(C, ldc * i + j) += alpha_real * temp_real - alpha_imag * temp_imag;
|
|
Packit |
67cb25 |
IMAG(C, ldc * i + j) += alpha_real * temp_imag + alpha_imag * temp_real;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
} else if (TransF == CblasTrans && TransG == CblasNoTrans) {
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
for (k = 0; k < K; k++) {
|
|
Packit |
67cb25 |
for (i = 0; i < n1; i++) {
|
|
Packit |
67cb25 |
const BASE Fki_real = CONST_REAL(F, ldf * k + i);
|
|
Packit |
67cb25 |
const BASE Fki_imag = conjF * CONST_IMAG(F, ldf * k + i);
|
|
Packit |
67cb25 |
const BASE temp_real = alpha_real * Fki_real - alpha_imag * Fki_imag;
|
|
Packit |
67cb25 |
const BASE temp_imag = alpha_real * Fki_imag + alpha_imag * Fki_real;
|
|
Packit |
67cb25 |
if (!(temp_real == 0.0 && temp_imag == 0.0)) {
|
|
Packit |
67cb25 |
for (j = 0; j < n2; j++) {
|
|
Packit |
67cb25 |
const BASE Gkj_real = CONST_REAL(G, ldg * k + j);
|
|
Packit |
67cb25 |
const BASE Gkj_imag = conjG * CONST_IMAG(G, ldg * k + j);
|
|
Packit |
67cb25 |
REAL(C, ldc * i + j) += temp_real * Gkj_real - temp_imag * Gkj_imag;
|
|
Packit |
67cb25 |
IMAG(C, ldc * i + j) += temp_real * Gkj_imag + temp_imag * Gkj_real;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
} else if (TransF == CblasTrans && TransG == CblasTrans) {
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
for (i = 0; i < n1; i++) {
|
|
Packit |
67cb25 |
for (j = 0; j < n2; j++) {
|
|
Packit |
67cb25 |
BASE temp_real = 0.0;
|
|
Packit |
67cb25 |
BASE temp_imag = 0.0;
|
|
Packit |
67cb25 |
for (k = 0; k < K; k++) {
|
|
Packit |
67cb25 |
const BASE Fki_real = CONST_REAL(F, ldf * k + i);
|
|
Packit |
67cb25 |
const BASE Fki_imag = conjF * CONST_IMAG(F, ldf * k + i);
|
|
Packit |
67cb25 |
const BASE Gjk_real = CONST_REAL(G, ldg * j + k);
|
|
Packit |
67cb25 |
const BASE Gjk_imag = conjG * CONST_IMAG(G, ldg * j + k);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
temp_real += Fki_real * Gjk_real - Fki_imag * Gjk_imag;
|
|
Packit |
67cb25 |
temp_imag += Fki_real * Gjk_imag + Fki_imag * Gjk_real;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
REAL(C, ldc * i + j) += alpha_real * temp_real - alpha_imag * temp_imag;
|
|
Packit |
67cb25 |
IMAG(C, ldc * i + j) += alpha_real * temp_imag + alpha_imag * temp_real;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
} else {
|
|
Packit |
67cb25 |
BLAS_ERROR("unrecognized operation");
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
}
|