|
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 <ctype.h></ctype.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 "XHBMV "
|
|
kusano |
2b45e8 |
#elif defined(DOUBLE)
|
|
kusano |
2b45e8 |
#define ERROR_NAME "ZHBMV "
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
#define ERROR_NAME "CHBMV "
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
static int (*hbmv[])(BLASLONG, BLASLONG, FLOAT, FLOAT, FLOAT *, BLASLONG, FLOAT *, BLASLONG, FLOAT *, BLASLONG, void *) = {
|
|
kusano |
2b45e8 |
#ifdef XDOUBLE
|
|
kusano |
2b45e8 |
xhbmv_U, xhbmv_L, xhbmv_V, xhbmv_M,
|
|
kusano |
2b45e8 |
#elif defined(DOUBLE)
|
|
kusano |
2b45e8 |
zhbmv_U, zhbmv_L, zhbmv_V, zhbmv_M,
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
chbmv_U, chbmv_L, chbmv_V, chbmv_M,
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
};
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifdef SMP
|
|
kusano |
2b45e8 |
static int (*hbmv_thread[])(BLASLONG, BLASLONG, FLOAT *, FLOAT *, BLASLONG, FLOAT *, BLASLONG, FLOAT *, BLASLONG, FLOAT *, int) = {
|
|
kusano |
2b45e8 |
#ifdef XDOUBLE
|
|
kusano |
2b45e8 |
xhbmv_thread_U, xhbmv_thread_L, xhbmv_thread_V, xhbmv_thread_M,
|
|
kusano |
2b45e8 |
#elif defined(DOUBLE)
|
|
kusano |
2b45e8 |
zhbmv_thread_U, zhbmv_thread_L, zhbmv_thread_V, zhbmv_thread_M,
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
chbmv_thread_U, chbmv_thread_L, chbmv_thread_V, chbmv_thread_M,
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
};
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifndef CBLAS
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
void NAME(char *UPLO, blasint *N, blasint *K, FLOAT *ALPHA, FLOAT *a, blasint *LDA,
|
|
kusano |
2b45e8 |
FLOAT *x, blasint *INCX, FLOAT *BETA, FLOAT *y, blasint *INCY){
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
char uplo_arg = *UPLO;
|
|
kusano |
2b45e8 |
blasint n = *N;
|
|
kusano |
2b45e8 |
blasint k = *K;
|
|
kusano |
2b45e8 |
FLOAT alpha_r = ALPHA[0];
|
|
kusano |
2b45e8 |
FLOAT alpha_i = ALPHA[1];
|
|
kusano |
2b45e8 |
blasint lda = *LDA;
|
|
kusano |
2b45e8 |
blasint incx = *INCX;
|
|
kusano |
2b45e8 |
FLOAT beta_r = BETA[0];
|
|
kusano |
2b45e8 |
FLOAT beta_i = BETA[1];
|
|
kusano |
2b45e8 |
blasint incy = *INCY;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
blasint info;
|
|
kusano |
2b45e8 |
int uplo;
|
|
kusano |
2b45e8 |
FLOAT *buffer;
|
|
kusano |
2b45e8 |
#ifdef SMP
|
|
kusano |
2b45e8 |
int nthreads;
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
PRINT_DEBUG_NAME;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
TOUPPER(uplo_arg);
|
|
kusano |
2b45e8 |
uplo = -1;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (uplo_arg == 'U') uplo = 0;
|
|
kusano |
2b45e8 |
if (uplo_arg == 'L') uplo = 1;
|
|
kusano |
2b45e8 |
if (uplo_arg == 'V') uplo = 2;
|
|
kusano |
2b45e8 |
if (uplo_arg == 'M') uplo = 3;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
info = 0;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (incy == 0) info = 11;
|
|
kusano |
2b45e8 |
if (incx == 0) info = 8;
|
|
kusano |
2b45e8 |
if (lda < k + 1) info = 6;
|
|
kusano |
2b45e8 |
if (k < 0) info = 3;
|
|
kusano |
2b45e8 |
if (n < 0) info = 2;
|
|
kusano |
2b45e8 |
if (uplo < 0) info = 1;
|
|
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 |
#else
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
void CNAME(enum CBLAS_ORDER order,
|
|
kusano |
2b45e8 |
enum CBLAS_UPLO Uplo,
|
|
kusano |
2b45e8 |
blasint n, blasint k,
|
|
kusano |
2b45e8 |
FLOAT *ALPHA,
|
|
kusano |
2b45e8 |
FLOAT *a, blasint lda,
|
|
kusano |
2b45e8 |
FLOAT *x, blasint incx,
|
|
kusano |
2b45e8 |
FLOAT *BETA,
|
|
kusano |
2b45e8 |
FLOAT *y, blasint incy){
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
FLOAT alpha_r = ALPHA[0];
|
|
kusano |
2b45e8 |
FLOAT alpha_i = ALPHA[1];
|
|
kusano |
2b45e8 |
FLOAT beta_r = BETA[0];
|
|
kusano |
2b45e8 |
FLOAT beta_i = BETA[1];
|
|
kusano |
2b45e8 |
FLOAT *buffer;
|
|
kusano |
2b45e8 |
int uplo;
|
|
kusano |
2b45e8 |
blasint info;
|
|
kusano |
2b45e8 |
#ifdef SMP
|
|
kusano |
2b45e8 |
int nthreads;
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
PRINT_DEBUG_CNAME;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
uplo = -1;
|
|
kusano |
2b45e8 |
info = 0;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (order == CblasColMajor) {
|
|
kusano |
2b45e8 |
if (Uplo == CblasUpper) uplo = 0;
|
|
kusano |
2b45e8 |
if (Uplo == CblasLower) uplo = 1;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
info = -1;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (incy == 0) info = 11;
|
|
kusano |
2b45e8 |
if (incx == 0) info = 8;
|
|
kusano |
2b45e8 |
if (lda < k + 1) info = 6;
|
|
kusano |
2b45e8 |
if (k < 0) info = 3;
|
|
kusano |
2b45e8 |
if (n < 0) info = 2;
|
|
kusano |
2b45e8 |
if (uplo < 0) info = 1;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (order == CblasRowMajor) {
|
|
kusano |
2b45e8 |
if (Uplo == CblasUpper) uplo = 3;
|
|
kusano |
2b45e8 |
if (Uplo == CblasLower) uplo = 2;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
info = -1;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (incy == 0) info = 11;
|
|
kusano |
2b45e8 |
if (incx == 0) info = 8;
|
|
kusano |
2b45e8 |
if (lda < k + 1) info = 6;
|
|
kusano |
2b45e8 |
if (k < 0) info = 3;
|
|
kusano |
2b45e8 |
if (n < 0) info = 2;
|
|
kusano |
2b45e8 |
if (uplo < 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 |
if (n == 0) return;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if ((beta_r != ONE) || (beta_i != ZERO)) SCAL_K(n, 0, 0, beta_r, beta_i, y, abs(incy), NULL, 0, NULL, 0);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if ((alpha_r == ZERO) && (alpha_i == ZERO)) return;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
IDEBUG_START;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
FUNCTION_PROFILE_START();
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (incx < 0 ) x -= (n - 1) * incx * COMPSIZE;
|
|
kusano |
2b45e8 |
if (incy < 0 ) y -= (n - 1) * incy * COMPSIZE;
|
|
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 |
(hbmv[uplo])(n, k, alpha_r, alpha_i, a, lda, x, incx, y, incy, buffer);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifdef SMP
|
|
kusano |
2b45e8 |
} else {
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
(hbmv_thread[uplo])(n, k, ALPHA, a, lda, x, incx, y, incy, buffer, nthreads);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
blas_memory_free(buffer);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
FUNCTION_PROFILE_END(4, n * k / 2 + n, n * k);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
IDEBUG_END;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
return;
|
|
kusano |
2b45e8 |
}
|