Blame thirdparty/openblas/xianyi-OpenBLAS-e6e87a2/driver/level3/gemm_thread_n.c

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 <stdlib.h></stdlib.h>
kusano 2b45e8
#include "common.h"
kusano 2b45e8
kusano 2b45e8
int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (*function)(), void *sa, void *sb, BLASLONG nthreads) {
kusano 2b45e8
kusano 2b45e8
  blas_queue_t queue[MAX_CPU_NUMBER];
kusano 2b45e8
  BLASLONG range[MAX_CPU_NUMBER + 1];
kusano 2b45e8
kusano 2b45e8
  BLASLONG width, i, num_cpu;
kusano 2b45e8
kusano 2b45e8
  if (!range_n) {
kusano 2b45e8
    range[0] = 0;
kusano 2b45e8
    i        = arg -> n;
kusano 2b45e8
  } else {
kusano 2b45e8
    range[0] = range_n[0];
kusano 2b45e8
    i        = range_n[1] - range_n[0];
kusano 2b45e8
  }
kusano 2b45e8
  
kusano 2b45e8
  num_cpu  = 0;
kusano 2b45e8
kusano 2b45e8
  while (i > 0){
kusano 2b45e8
    
kusano 2b45e8
    width  = blas_quickdivide(i + nthreads - num_cpu - 1, nthreads - num_cpu);
kusano 2b45e8
kusano 2b45e8
    i -= width;
kusano 2b45e8
    if (i < 0) width = width + i;
kusano 2b45e8
kusano 2b45e8
    range[num_cpu + 1] = range[num_cpu] + width;
kusano 2b45e8
kusano 2b45e8
    queue[num_cpu].mode    = mode;
kusano 2b45e8
    queue[num_cpu].routine = function;
kusano 2b45e8
    queue[num_cpu].args    = arg;
kusano 2b45e8
    queue[num_cpu].range_m = range_m;
kusano 2b45e8
    queue[num_cpu].range_n = &range[num_cpu];
kusano 2b45e8
#if defined(LOONGSON3A)
kusano 2b45e8
    queue[num_cpu].sa      = sa	+ GEMM_OFFSET_A1 * num_cpu;
kusano 2b45e8
    queue[num_cpu].sb      = queue[num_cpu].sa + GEMM_OFFSET_A1 * 5;
kusano 2b45e8
#else
kusano 2b45e8
	queue[num_cpu].sa      = NULL;
kusano 2b45e8
    queue[num_cpu].sb      = NULL;
kusano 2b45e8
#endif
kusano 2b45e8
    queue[num_cpu].next    = &queue[num_cpu + 1];
kusano 2b45e8
    num_cpu ++;
kusano 2b45e8
  }
kusano 2b45e8
  
kusano 2b45e8
  if (num_cpu) {
kusano 2b45e8
#if defined(LOONGSON3A)
kusano 2b45e8
    queue[0].sa = sa;
kusano 2b45e8
    queue[0].sb = sa + GEMM_OFFSET_A1 * 5;
kusano 2b45e8
#else
kusano 2b45e8
	queue[0].sa = sa;
kusano 2b45e8
	queue[0].sb = sb;
kusano 2b45e8
#endif
kusano 2b45e8
    queue[num_cpu - 1].next = NULL;
kusano 2b45e8
    
kusano 2b45e8
    exec_blas(num_cpu,
kusano 2b45e8
	      queue);
kusano 2b45e8
  }
kusano 2b45e8
   
kusano 2b45e8
  return 0;
kusano 2b45e8
}