|
kusano |
2b45e8 |
/*********************************************************************/
|
|
kusano |
2b45e8 |
/* Copyright 2009, 2010 The University of Texas at Austin. */
|
|
kusano |
2b45e8 |
/* All rights reserved. */
|
|
kusano |
2b45e8 |
/* */
|
|
kusano |
2b45e8 |
/* Redistribution and use in source and binary forms, with or */
|
|
kusano |
2b45e8 |
/* without modification, are permitted provided that the following */
|
|
kusano |
2b45e8 |
/* conditions are met: */
|
|
kusano |
2b45e8 |
/* */
|
|
kusano |
2b45e8 |
/* 1. Redistributions of source code must retain the above */
|
|
kusano |
2b45e8 |
/* copyright notice, this list of conditions and the following */
|
|
kusano |
2b45e8 |
/* disclaimer. */
|
|
kusano |
2b45e8 |
/* */
|
|
kusano |
2b45e8 |
/* 2. Redistributions in binary form must reproduce the above */
|
|
kusano |
2b45e8 |
/* copyright notice, this list of conditions and the following */
|
|
kusano |
2b45e8 |
/* disclaimer in the documentation and/or other materials */
|
|
kusano |
2b45e8 |
/* provided with the distribution. */
|
|
kusano |
2b45e8 |
/* */
|
|
kusano |
2b45e8 |
/* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */
|
|
kusano |
2b45e8 |
/* AUSTIN ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, */
|
|
kusano |
2b45e8 |
/* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */
|
|
kusano |
2b45e8 |
/* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */
|
|
kusano |
2b45e8 |
/* DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY OF TEXAS AT */
|
|
kusano |
2b45e8 |
/* AUSTIN OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */
|
|
kusano |
2b45e8 |
/* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES */
|
|
kusano |
2b45e8 |
/* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE */
|
|
kusano |
2b45e8 |
/* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR */
|
|
kusano |
2b45e8 |
/* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF */
|
|
kusano |
2b45e8 |
/* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT */
|
|
kusano |
2b45e8 |
/* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT */
|
|
kusano |
2b45e8 |
/* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */
|
|
kusano |
2b45e8 |
/* POSSIBILITY OF SUCH DAMAGE. */
|
|
kusano |
2b45e8 |
/* */
|
|
kusano |
2b45e8 |
/* The views and conclusions contained in the software and */
|
|
kusano |
2b45e8 |
/* documentation are those of the authors and should not be */
|
|
kusano |
2b45e8 |
/* interpreted as representing official policies, either expressed */
|
|
kusano |
2b45e8 |
/* or implied, of The University of Texas at Austin. */
|
|
kusano |
2b45e8 |
/*********************************************************************/
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#include <stdio.h></stdio.h>
|
|
kusano |
2b45e8 |
#include "common.h"
|
|
kusano |
2b45e8 |
#ifdef FUNCTION_PROFILE
|
|
kusano |
2b45e8 |
#include "functable.h"
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifdef XDOUBLE
|
|
kusano |
2b45e8 |
#define ERROR_NAME "QGER "
|
|
kusano |
2b45e8 |
#elif defined DOUBLE
|
|
kusano |
2b45e8 |
#define ERROR_NAME "DGER "
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
#define ERROR_NAME "SGER "
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#define GER GERU_K
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#if defined XDOUBLE
|
|
kusano |
2b45e8 |
#define GER_THREAD qger_thread
|
|
kusano |
2b45e8 |
#elif defined DOUBLE
|
|
kusano |
2b45e8 |
#define GER_THREAD dger_thread
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
#define GER_THREAD sger_thread
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifndef CBLAS
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
void NAME(blasint *M, blasint *N, FLOAT *Alpha,
|
|
kusano |
2b45e8 |
FLOAT *x, blasint *INCX,
|
|
kusano |
2b45e8 |
FLOAT *y, blasint *INCY,
|
|
kusano |
2b45e8 |
FLOAT *a, blasint *LDA){
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
blasint m = *M;
|
|
kusano |
2b45e8 |
blasint n = *N;
|
|
kusano |
2b45e8 |
FLOAT alpha = *Alpha;
|
|
kusano |
2b45e8 |
blasint incx = *INCX;
|
|
kusano |
2b45e8 |
blasint incy = *INCY;
|
|
kusano |
2b45e8 |
blasint lda = *LDA;
|
|
kusano |
2b45e8 |
FLOAT *buffer;
|
|
kusano |
2b45e8 |
#ifdef SMP
|
|
kusano |
2b45e8 |
int nthreads;
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
blasint info;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
PRINT_DEBUG_NAME;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
info = 0;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (lda < MAX(1,m)) info = 9;
|
|
kusano |
2b45e8 |
if (incy == 0) info = 7;
|
|
kusano |
2b45e8 |
if (incx == 0) info = 5;
|
|
kusano |
2b45e8 |
if (n < 0) info = 2;
|
|
kusano |
2b45e8 |
if (m < 0) info = 1;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (info){
|
|
kusano |
2b45e8 |
BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
|
|
kusano |
2b45e8 |
return;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
void CNAME(enum CBLAS_ORDER order,
|
|
kusano |
2b45e8 |
blasint m, blasint n,
|
|
kusano |
2b45e8 |
FLOAT alpha,
|
|
kusano |
2b45e8 |
FLOAT *x, blasint incx,
|
|
kusano |
2b45e8 |
FLOAT *y, blasint incy,
|
|
kusano |
2b45e8 |
FLOAT *a, blasint lda) {
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
FLOAT *buffer;
|
|
kusano |
2b45e8 |
blasint info, t;
|
|
kusano |
2b45e8 |
#ifdef SMP
|
|
kusano |
2b45e8 |
int nthreads;
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
PRINT_DEBUG_CNAME;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
info = 0;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (order == CblasColMajor) {
|
|
kusano |
2b45e8 |
info = -1;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (lda < MAX(1,m)) info = 9;
|
|
kusano |
2b45e8 |
if (incy == 0) info = 7;
|
|
kusano |
2b45e8 |
if (incx == 0) info = 5;
|
|
kusano |
2b45e8 |
if (n < 0) info = 2;
|
|
kusano |
2b45e8 |
if (m < 0) info = 1;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (order == CblasRowMajor) {
|
|
kusano |
2b45e8 |
info = -1;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
t = n;
|
|
kusano |
2b45e8 |
n = m;
|
|
kusano |
2b45e8 |
m = t;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
t = incx;
|
|
kusano |
2b45e8 |
incx = incy;
|
|
kusano |
2b45e8 |
incy = t;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
buffer = x;
|
|
kusano |
2b45e8 |
x = y;
|
|
kusano |
2b45e8 |
y = buffer;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (lda < MAX(1,m)) info = 9;
|
|
kusano |
2b45e8 |
if (incy == 0) info = 7;
|
|
kusano |
2b45e8 |
if (incx == 0) info = 5;
|
|
kusano |
2b45e8 |
if (n < 0) info = 2;
|
|
kusano |
2b45e8 |
if (m < 0) info = 1;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (info >= 0) {
|
|
kusano |
2b45e8 |
BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
|
|
kusano |
2b45e8 |
return;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
/* Quick return if possible. */
|
|
kusano |
2b45e8 |
if (m == 0 || n == 0) return;
|
|
kusano |
2b45e8 |
if (alpha == 0.) return;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
IDEBUG_START;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
FUNCTION_PROFILE_START();
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (incy < 0) y -= (n - 1) * incy;
|
|
kusano |
2b45e8 |
if (incx < 0) x -= (m - 1) * incx;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
buffer = (FLOAT *)blas_memory_alloc(1);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifdef SMP
|
|
kusano |
2b45e8 |
nthreads = num_cpu_avail(2);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (nthreads == 1) {
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
GER(m, n, 0, alpha, x, incx, y, incy, a, lda, buffer);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifdef SMP
|
|
kusano |
2b45e8 |
} else {
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
GER_THREAD(m, n, alpha, x, incx, y, incy, a, lda, buffer, nthreads);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
blas_memory_free(buffer);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
FUNCTION_PROFILE_END(1, m * n + m + n, 2 * m * n);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
IDEBUG_END;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
return;
|
|
kusano |
2b45e8 |
}
|