|
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 |
#include "symcopy.h"
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
int CNAME(BLASLONG m, BLASLONG offset, FLOAT alpha, FLOAT *a, BLASLONG lda,
|
|
kusano |
2b45e8 |
FLOAT *x, BLASLONG incx, FLOAT *y, BLASLONG incy, FLOAT *buffer){
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
BLASLONG is, min_i;
|
|
kusano |
2b45e8 |
FLOAT *X = x;
|
|
kusano |
2b45e8 |
FLOAT *Y = y;
|
|
kusano |
2b45e8 |
FLOAT *symbuffer = buffer;
|
|
kusano |
2b45e8 |
FLOAT *gemvbuffer = (FLOAT *)(((BLASLONG)buffer + SYMV_P * SYMV_P * sizeof(FLOAT) + 4095) & ~4095);
|
|
kusano |
2b45e8 |
FLOAT *bufferY = gemvbuffer;
|
|
kusano |
2b45e8 |
FLOAT *bufferX = gemvbuffer;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (incy != 1) {
|
|
kusano |
2b45e8 |
Y = bufferY;
|
|
kusano |
2b45e8 |
bufferX = (FLOAT *)(((BLASLONG)bufferY + m * sizeof(FLOAT) + 4095) & ~4095);
|
|
kusano |
2b45e8 |
gemvbuffer = bufferX;
|
|
kusano |
2b45e8 |
COPY_K(m, y, incy, Y, 1);
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (incx != 1) {
|
|
kusano |
2b45e8 |
X = bufferX;
|
|
kusano |
2b45e8 |
gemvbuffer = (FLOAT *)(((BLASLONG)bufferX + m * sizeof(FLOAT) + 4095) & ~4095);
|
|
kusano |
2b45e8 |
COPY_K(m, x, incx, X, 1);
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifndef LOWER
|
|
kusano |
2b45e8 |
for(is = m - offset; is < m; is += SYMV_P){
|
|
kusano |
2b45e8 |
min_i = MIN(m - is, SYMV_P);
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
for(is = 0; is < offset; is += SYMV_P){
|
|
kusano |
2b45e8 |
min_i = MIN(offset - is, SYMV_P);
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifndef LOWER
|
|
kusano |
2b45e8 |
if (is >0){
|
|
kusano |
2b45e8 |
GEMV_T(is, min_i, 0, alpha,
|
|
kusano |
2b45e8 |
a + is * lda, lda,
|
|
kusano |
2b45e8 |
X, 1,
|
|
kusano |
2b45e8 |
Y + is, 1, gemvbuffer);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
GEMV_N(is, min_i, 0, alpha,
|
|
kusano |
2b45e8 |
a + is * lda, lda,
|
|
kusano |
2b45e8 |
X + is, 1,
|
|
kusano |
2b45e8 |
Y, 1, gemvbuffer);
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifdef LOWER
|
|
kusano |
2b45e8 |
SYMCOPY_L(min_i, a + is + is * lda, lda, symbuffer);
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
SYMCOPY_U(min_i, a + is + is * lda, lda, symbuffer);
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
GEMV_N(min_i, min_i, 0, alpha,
|
|
kusano |
2b45e8 |
symbuffer, min_i,
|
|
kusano |
2b45e8 |
X + is, 1,
|
|
kusano |
2b45e8 |
Y + is, 1, gemvbuffer);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifdef LOWER
|
|
kusano |
2b45e8 |
if (m - is > min_i){
|
|
kusano |
2b45e8 |
GEMV_T(m - is - min_i, min_i, 0, alpha,
|
|
kusano |
2b45e8 |
a + (is + min_i) + is * lda, lda,
|
|
kusano |
2b45e8 |
X + (is + min_i), 1,
|
|
kusano |
2b45e8 |
Y + is, 1, gemvbuffer);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
GEMV_N(m - is - min_i, min_i, 0, alpha,
|
|
kusano |
2b45e8 |
a + (is + min_i) + is * lda, lda,
|
|
kusano |
2b45e8 |
X + is, 1,
|
|
kusano |
2b45e8 |
Y + (is + min_i), 1, gemvbuffer);
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
} /* end of is */
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (incy != 1) {
|
|
kusano |
2b45e8 |
COPY_K(m, Y, 1, y, incy);
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
return 0;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|