|
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 |
|
|
kusano |
2b45e8 |
#undef TIMING
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#define BETA_OPERATION(M_FROM, M_TO, N_FROM, N_TO, BETA, C, LDC) \
|
|
kusano |
2b45e8 |
GEMM_BETA((M_TO) - (M_FROM), (N_TO - N_FROM), 0, \
|
|
kusano |
2b45e8 |
BETA[0], BETA[1], NULL, 0, NULL, 0, \
|
|
kusano |
2b45e8 |
(FLOAT *)(C) + (M_FROM) + (N_FROM) * (LDC) * COMPSIZE, LDC)
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifndef RSIDE
|
|
kusano |
2b45e8 |
#ifndef LOWER
|
|
kusano |
2b45e8 |
#define ICOPYB_OPERATION(M, N, A, LDA, X, Y, BUFFER) HEMM3M_IUCOPYB(M, N, A, LDA, Y, X, BUFFER)
|
|
kusano |
2b45e8 |
#define ICOPYR_OPERATION(M, N, A, LDA, X, Y, BUFFER) HEMM3M_IUCOPYR(M, N, A, LDA, Y, X, BUFFER)
|
|
kusano |
2b45e8 |
#define ICOPYI_OPERATION(M, N, A, LDA, X, Y, BUFFER) HEMM3M_IUCOPYI(M, N, A, LDA, Y, X, BUFFER)
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
#define ICOPYB_OPERATION(M, N, A, LDA, X, Y, BUFFER) HEMM3M_ILCOPYB(M, N, A, LDA, Y, X, BUFFER)
|
|
kusano |
2b45e8 |
#define ICOPYR_OPERATION(M, N, A, LDA, X, Y, BUFFER) HEMM3M_ILCOPYR(M, N, A, LDA, Y, X, BUFFER)
|
|
kusano |
2b45e8 |
#define ICOPYI_OPERATION(M, N, A, LDA, X, Y, BUFFER) HEMM3M_ILCOPYI(M, N, A, LDA, Y, X, BUFFER)
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifdef RSIDE
|
|
kusano |
2b45e8 |
#ifndef LOWER
|
|
kusano |
2b45e8 |
#define OCOPYB_OPERATION(M, N, A, LDA, ALPHA_R, ALPHA_I, X, Y, BUFFER) \
|
|
kusano |
2b45e8 |
HEMM3M_OUCOPYB(M, N, A, LDA, Y, X, ALPHA_R, ALPHA_I, BUFFER)
|
|
kusano |
2b45e8 |
#define OCOPYR_OPERATION(M, N, A, LDA, ALPHA_R, ALPHA_I, X, Y, BUFFER) \
|
|
kusano |
2b45e8 |
HEMM3M_OUCOPYR(M, N, A, LDA, Y, X, ALPHA_R, ALPHA_I, BUFFER)
|
|
kusano |
2b45e8 |
#define OCOPYI_OPERATION(M, N, A, LDA, ALPHA_R, ALPHA_I, X, Y, BUFFER) \
|
|
kusano |
2b45e8 |
HEMM3M_OUCOPYI(M, N, A, LDA, Y, X, ALPHA_R, ALPHA_I, BUFFER)
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
#define OCOPYB_OPERATION(M, N, A, LDA, ALPHA_R, ALPHA_I, X, Y, BUFFER) \
|
|
kusano |
2b45e8 |
HEMM3M_OLCOPYB(M, N, A, LDA, Y, X, ALPHA_R, ALPHA_I, BUFFER)
|
|
kusano |
2b45e8 |
#define OCOPYR_OPERATION(M, N, A, LDA, ALPHA_R, ALPHA_I, X, Y, BUFFER) \
|
|
kusano |
2b45e8 |
HEMM3M_OLCOPYR(M, N, A, LDA, Y, X, ALPHA_R, ALPHA_I, BUFFER)
|
|
kusano |
2b45e8 |
#define OCOPYI_OPERATION(M, N, A, LDA, ALPHA_R, ALPHA_I, X, Y, BUFFER) \
|
|
kusano |
2b45e8 |
HEMM3M_OLCOPYI(M, N, A, LDA, Y, X, ALPHA_R, ALPHA_I, BUFFER)
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifndef RSIDE
|
|
kusano |
2b45e8 |
#define K args -> m
|
|
kusano |
2b45e8 |
#ifndef LOWER
|
|
kusano |
2b45e8 |
#define GEMM3M_LOCAL HEMM3M_LU
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
#define GEMM3M_LOCAL HEMM3M_LL
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
#define K args -> n
|
|
kusano |
2b45e8 |
#ifndef LOWER
|
|
kusano |
2b45e8 |
#define GEMM3M_LOCAL HEMM3M_RU
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
#define GEMM3M_LOCAL HEMM3M_RL
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifdef THREADED_LEVEL3
|
|
kusano |
2b45e8 |
#include "level3_gemm3m_thread.c"
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
#include "gemm3m_level3.c"
|
|
kusano |
2b45e8 |
#endif
|