Blame thirdparty/openblas/xianyi-OpenBLAS-e6e87a2/driver/level3/gemm_thread_mn.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
static const int divide_rule[][2] = 
kusano 2b45e8
  {{ 0,  0},
kusano 2b45e8
   { 1,  1}, { 1,  2}, { 1,  3}, { 2,  2},
kusano 2b45e8
   { 1,  5}, { 2,  3}, { 1,  7}, { 2,  4},
kusano 2b45e8
   { 3,  3}, { 2,  5}, { 1, 11}, { 2,  6},
kusano 2b45e8
   { 1, 13}, { 2,  7}, { 3,  5}, { 4,  4},
kusano 2b45e8
   { 1, 17}, { 3,  6}, { 1, 19}, { 4,  5},
kusano 2b45e8
   { 3,  7}, { 2, 11}, { 1, 23}, { 4,  6},
kusano 2b45e8
   { 5,  5}, { 2, 13}, { 3,  9}, { 4,  7},
kusano 2b45e8
   { 1, 29}, { 5,  6}, { 1, 31}, { 4,  8},
kusano 2b45e8
   { 3, 11}, { 2, 17}, { 5,  7}, { 6,  6},
kusano 2b45e8
   { 1, 37}, { 2, 19}, { 3, 13}, { 5,  8},
kusano 2b45e8
   { 1, 41}, { 6,  7}, { 1, 43}, { 4, 11},
kusano 2b45e8
   { 5,  9}, { 2, 23}, { 1, 47}, { 6,  8},
kusano 2b45e8
   { 7,  7}, { 5, 10}, { 3, 17}, { 4, 13},
kusano 2b45e8
   { 1, 53}, { 6,  9}, { 5, 11}, { 7,  8},
kusano 2b45e8
   { 3, 19}, { 2, 29}, { 1, 59}, { 6, 10},
kusano 2b45e8
   { 1, 61}, { 2, 31}, { 7,  9}, { 8,  8},
kusano 2b45e8
};
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
kusano 2b45e8
  BLASLONG range_M[MAX_CPU_NUMBER + 1], range_N[MAX_CPU_NUMBER + 1];
kusano 2b45e8
  BLASLONG procs, total_procs, num_cpu_m, num_cpu_n;
kusano 2b45e8
kusano 2b45e8
  BLASLONG width, i, j;
kusano 2b45e8
  BLASLONG divM, divN;
kusano 2b45e8
kusano 2b45e8
  divM = divide_rule[nthreads][0];
kusano 2b45e8
  divN = divide_rule[nthreads][1];
kusano 2b45e8
kusano 2b45e8
  if (!range_m) {
kusano 2b45e8
    range_M[0] = 0;
kusano 2b45e8
    i          = arg -> m;
kusano 2b45e8
  } else {
kusano 2b45e8
    range_M[0] = range_m[0];
kusano 2b45e8
    i          = range_m[1] - range_m[0];
kusano 2b45e8
  }
kusano 2b45e8
kusano 2b45e8
  num_cpu_m  = 0;
kusano 2b45e8
kusano 2b45e8
  while (i > 0){
kusano 2b45e8
    
kusano 2b45e8
    width  = blas_quickdivide(i + divM - num_cpu_m - 1, divM - num_cpu_m);
kusano 2b45e8
kusano 2b45e8
    i -= width;
kusano 2b45e8
    if (i < 0) width = width + i;
kusano 2b45e8
kusano 2b45e8
    range_M[num_cpu_m + 1] = range_M[num_cpu_m] + width;
kusano 2b45e8
kusano 2b45e8
    num_cpu_m ++;
kusano 2b45e8
  }
kusano 2b45e8
kusano 2b45e8
  if (!range_n) {
kusano 2b45e8
    range_N[0] = 0;
kusano 2b45e8
    i          = arg -> n;
kusano 2b45e8
  } else {
kusano 2b45e8
    range_N[0] = range_n[0];
kusano 2b45e8
    i          = range_n[1] - range_n[0];
kusano 2b45e8
  }
kusano 2b45e8
kusano 2b45e8
  num_cpu_n  = 0;
kusano 2b45e8
kusano 2b45e8
  while (i > 0){
kusano 2b45e8
    
kusano 2b45e8
    width  = blas_quickdivide(i + divN - num_cpu_n - 1, divN - num_cpu_n);
kusano 2b45e8
kusano 2b45e8
    i -= width;
kusano 2b45e8
    if (i < 0) width = width + i;
kusano 2b45e8
kusano 2b45e8
    range_N[num_cpu_n + 1] = range_N[num_cpu_n] + width;
kusano 2b45e8
kusano 2b45e8
    num_cpu_n ++;
kusano 2b45e8
  }
kusano 2b45e8
kusano 2b45e8
  procs = 0;
kusano 2b45e8
kusano 2b45e8
  for (j = 0; j < num_cpu_n; j++) {
kusano 2b45e8
    for (i = 0; i < num_cpu_m; i++) {
kusano 2b45e8
kusano 2b45e8
    queue[procs].mode    = mode;
kusano 2b45e8
    queue[procs].routine = function;
kusano 2b45e8
    queue[procs].args    = arg;
kusano 2b45e8
    queue[procs].range_m = &range_M[i];
kusano 2b45e8
    queue[procs].range_n = &range_N[j];
kusano 2b45e8
    queue[procs].sa      = NULL;
kusano 2b45e8
    queue[procs].sb      = NULL;
kusano 2b45e8
    queue[procs].next    = &queue[procs + 1];
kusano 2b45e8
kusano 2b45e8
    procs ++;
kusano 2b45e8
    }
kusano 2b45e8
  }
kusano 2b45e8
  
kusano 2b45e8
  if (procs) {
kusano 2b45e8
    queue[0].sa = sa;
kusano 2b45e8
    queue[0].sb = sb;
kusano 2b45e8
kusano 2b45e8
    queue[procs - 1].next = NULL;
kusano 2b45e8
    
kusano 2b45e8
    exec_blas(procs, queue);
kusano 2b45e8
  }
kusano 2b45e8
  
kusano 2b45e8
  return 0;
kusano 2b45e8
}