Blame thirdparty/openblas/xianyi-OpenBLAS-e6e87a2/driver/level2/syr_thread.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 int syr_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *dummy1, FLOAT *buffer, BLASLONG pos){
kusano 2b45e8
kusano 2b45e8
  FLOAT *a, *x;
kusano 2b45e8
  BLASLONG lda, incx;
kusano 2b45e8
  BLASLONG i, m_from, m_to;
kusano 2b45e8
  FLOAT alpha_r;
kusano 2b45e8
#if defined(COMPLEX) && !defined(HER) && !defined(HERREV)
kusano 2b45e8
  FLOAT alpha_i;
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
kusano 2b45e8
  x = (FLOAT *)args -> a;
kusano 2b45e8
  a = (FLOAT *)args -> b;
kusano 2b45e8
kusano 2b45e8
  incx = args -> lda;
kusano 2b45e8
  lda  = args -> ldb;
kusano 2b45e8
kusano 2b45e8
  alpha_r = *((FLOAT *)args -> alpha + 0);
kusano 2b45e8
#if defined(COMPLEX) && !defined(HER) && !defined(HERREV)
kusano 2b45e8
  alpha_i = *((FLOAT *)args -> alpha + 1);
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
  m_from = 0;
kusano 2b45e8
  m_to   = args -> m;
kusano 2b45e8
kusano 2b45e8
  if (range_m) {
kusano 2b45e8
    m_from = *(range_m + 0);
kusano 2b45e8
    m_to   = *(range_m + 1);
kusano 2b45e8
  }
kusano 2b45e8
kusano 2b45e8
  if (incx != 1) {
kusano 2b45e8
#ifndef LOWER
kusano 2b45e8
    COPY_K(m_to, x, incx, buffer, 1);
kusano 2b45e8
#else
kusano 2b45e8
    COPY_K(args -> m - m_from, x + m_from * incx * COMPSIZE, incx, buffer + m_from * COMPSIZE, 1);
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
    x = buffer;
kusano 2b45e8
  }
kusano 2b45e8
kusano 2b45e8
  a += m_from * lda * COMPSIZE;
kusano 2b45e8
kusano 2b45e8
  for (i = m_from; i < m_to; i++){
kusano 2b45e8
#if !defined(HER) && !defined(HERREV)
kusano 2b45e8
#ifndef COMPLEX
kusano 2b45e8
    if (x[i * COMPSIZE] != ZERO) {
kusano 2b45e8
#ifndef LOWER
kusano 2b45e8
      AXPYU_K(i + 1,         0, 0, alpha_r * x[i], x,     1, a,     1, NULL, 0);
kusano 2b45e8
#else
kusano 2b45e8
      AXPYU_K(args -> m - i, 0, 0, alpha_r * x[i], x + i, 1, a + i, 1, NULL, 0);
kusano 2b45e8
#endif
kusano 2b45e8
    }
kusano 2b45e8
#else
kusano 2b45e8
    if ((x[i * COMPSIZE + 0] != ZERO) || (x[i * COMPSIZE + 1] != ZERO)) {
kusano 2b45e8
#ifndef LOWER
kusano 2b45e8
      AXPYU_K(i + 1,         0, 0, 
kusano 2b45e8
	      alpha_r * x[i * COMPSIZE + 0] - alpha_i * x[i * COMPSIZE + 1],
kusano 2b45e8
	      alpha_i * x[i * COMPSIZE + 0] + alpha_r * x[i * COMPSIZE + 1],
kusano 2b45e8
	      x,                1, a,                1, NULL, 0);
kusano 2b45e8
#else
kusano 2b45e8
      AXPYU_K(args -> m - i, 0, 0,
kusano 2b45e8
	      alpha_r * x[i * COMPSIZE + 0] - alpha_i * x[i * COMPSIZE + 1],
kusano 2b45e8
	      alpha_i * x[i * COMPSIZE + 0] + alpha_r * x[i * COMPSIZE + 1],
kusano 2b45e8
	      x + i * COMPSIZE, 1, a + i * COMPSIZE, 1, NULL, 0);
kusano 2b45e8
#endif
kusano 2b45e8
    }
kusano 2b45e8
#endif
kusano 2b45e8
#else
kusano 2b45e8
    if ((x[i * COMPSIZE + 0] != ZERO) || (x[i * COMPSIZE + 1] != ZERO)) {
kusano 2b45e8
#ifndef HERREV
kusano 2b45e8
#ifndef LOWER
kusano 2b45e8
      AXPYU_K(i + 1,         0, 0, 
kusano 2b45e8
	      alpha_r * x[i * COMPSIZE + 0], -alpha_r * x[i * COMPSIZE + 1],
kusano 2b45e8
	      x,                1, a,                1, NULL, 0);
kusano 2b45e8
#else
kusano 2b45e8
      AXPYU_K(args -> m - i, 0, 0,
kusano 2b45e8
	      alpha_r * x[i * COMPSIZE + 0], -alpha_r * x[i * COMPSIZE + 1],
kusano 2b45e8
	      x + i * COMPSIZE, 1, a + i * COMPSIZE, 1, NULL, 0);
kusano 2b45e8
#endif
kusano 2b45e8
#else
kusano 2b45e8
#ifndef LOWER
kusano 2b45e8
      AXPYC_K(i + 1,         0, 0, 
kusano 2b45e8
	      alpha_r * x[i * COMPSIZE + 0],  alpha_r * x[i * COMPSIZE + 1],
kusano 2b45e8
	      x,                1, a,                1, NULL, 0);
kusano 2b45e8
#else
kusano 2b45e8
      AXPYC_K(args -> m - i, 0, 0,
kusano 2b45e8
	      alpha_r * x[i * COMPSIZE + 0],  alpha_r * x[i * COMPSIZE + 1],
kusano 2b45e8
	      x + i * COMPSIZE, 1, a + i * COMPSIZE, 1, NULL, 0);
kusano 2b45e8
#endif
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
    }
kusano 2b45e8
      a[i * COMPSIZE + 1] = ZERO;
kusano 2b45e8
#endif
kusano 2b45e8
      a += lda * COMPSIZE;
kusano 2b45e8
kusano 2b45e8
  }
kusano 2b45e8
  
kusano 2b45e8
  return 0;
kusano 2b45e8
}
kusano 2b45e8
kusano 2b45e8
#if !defined(COMPLEX) || defined(HER) || defined(HERREV)
kusano 2b45e8
int CNAME(BLASLONG m, FLOAT  alpha, FLOAT *x, BLASLONG incx, FLOAT *a, BLASLONG lda, FLOAT *buffer, int nthreads){
kusano 2b45e8
#else
kusano 2b45e8
int CNAME(BLASLONG m, FLOAT *alpha, FLOAT *x, BLASLONG incx, FLOAT *a, BLASLONG lda, FLOAT *buffer, int nthreads){
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
  blas_arg_t args;
kusano 2b45e8
  blas_queue_t queue[MAX_CPU_NUMBER];
kusano 2b45e8
  BLASLONG range_m[MAX_CPU_NUMBER + 1];
kusano 2b45e8
kusano 2b45e8
  BLASLONG width, i, num_cpu;
kusano 2b45e8
kusano 2b45e8
  double dnum;
kusano 2b45e8
  int mask = 7;
kusano 2b45e8
kusano 2b45e8
#ifdef SMP
kusano 2b45e8
#ifndef COMPLEX
kusano 2b45e8
#ifdef XDOUBLE
kusano 2b45e8
  int mode  =  BLAS_XDOUBLE | BLAS_REAL;
kusano 2b45e8
#elif defined(DOUBLE)
kusano 2b45e8
  int mode  =  BLAS_DOUBLE  | BLAS_REAL;
kusano 2b45e8
#else
kusano 2b45e8
  int mode  =  BLAS_SINGLE  | BLAS_REAL;
kusano 2b45e8
#endif  
kusano 2b45e8
#else
kusano 2b45e8
#ifdef XDOUBLE
kusano 2b45e8
  int mode  =  BLAS_XDOUBLE | BLAS_COMPLEX;
kusano 2b45e8
#elif defined(DOUBLE)
kusano 2b45e8
  int mode  =  BLAS_DOUBLE  | BLAS_COMPLEX;
kusano 2b45e8
#else
kusano 2b45e8
  int mode  =  BLAS_SINGLE  | BLAS_COMPLEX;
kusano 2b45e8
#endif  
kusano 2b45e8
#endif
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
  args.m = m;
kusano 2b45e8
  
kusano 2b45e8
  args.a = (void *)x;
kusano 2b45e8
  args.b = (void *)a;
kusano 2b45e8
    
kusano 2b45e8
  args.lda = incx;
kusano 2b45e8
  args.ldb = lda;
kusano 2b45e8
#if !defined(COMPLEX) || defined(HER) || defined(HERREV)
kusano 2b45e8
  args.alpha = (void *)α
kusano 2b45e8
#else
kusano 2b45e8
  args.alpha = (void *)alpha;
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
  dnum = (double)m * (double)m / (double)nthreads;
kusano 2b45e8
  num_cpu  = 0;
kusano 2b45e8
  
kusano 2b45e8
#ifndef LOWER
kusano 2b45e8
kusano 2b45e8
  range_m[MAX_CPU_NUMBER] = m;
kusano 2b45e8
  i          = 0;
kusano 2b45e8
    
kusano 2b45e8
  while (i < m){
kusano 2b45e8
    
kusano 2b45e8
    if (nthreads - num_cpu > 1) {
kusano 2b45e8
      
kusano 2b45e8
      double di = (double)(m - i);
kusano 2b45e8
      if (di * di - dnum > 0) {
kusano 2b45e8
	width = ((BLASLONG)(-sqrt(di * di - dnum) + di) + mask) & ~mask;
kusano 2b45e8
      } else {
kusano 2b45e8
	width = m - i;
kusano 2b45e8
      }
kusano 2b45e8
kusano 2b45e8
      if (width < 16) width = 16;
kusano 2b45e8
      if (width > m - i) width = m - i;
kusano 2b45e8
	
kusano 2b45e8
    } else {
kusano 2b45e8
      width = m - i;
kusano 2b45e8
    }
kusano 2b45e8
    
kusano 2b45e8
    range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width;
kusano 2b45e8
      
kusano 2b45e8
    queue[num_cpu].mode    = mode;
kusano 2b45e8
    queue[num_cpu].routine = syr_kernel;
kusano 2b45e8
    queue[num_cpu].args    = &args;
kusano 2b45e8
    queue[num_cpu].range_m = &range_m[MAX_CPU_NUMBER - num_cpu - 1];
kusano 2b45e8
    queue[num_cpu].range_n = NULL;
kusano 2b45e8
    queue[num_cpu].sa      = NULL;
kusano 2b45e8
    queue[num_cpu].sb      = NULL;
kusano 2b45e8
    queue[num_cpu].next    = &queue[num_cpu + 1];
kusano 2b45e8
    
kusano 2b45e8
    num_cpu ++;
kusano 2b45e8
    i += width;
kusano 2b45e8
  }
kusano 2b45e8
  
kusano 2b45e8
#else
kusano 2b45e8
kusano 2b45e8
  range_m[0] = 0;
kusano 2b45e8
  i          = 0;
kusano 2b45e8
    
kusano 2b45e8
  while (i < m){
kusano 2b45e8
    
kusano 2b45e8
    if (nthreads - num_cpu > 1) {
kusano 2b45e8
      
kusano 2b45e8
      double di = (double)(m - i);
kusano 2b45e8
      if (di * di - dnum > 0) {
kusano 2b45e8
	width = ((BLASLONG)(-sqrt(di * di - dnum) + di) + mask) & ~mask;
kusano 2b45e8
      } else {
kusano 2b45e8
	width = m - i;
kusano 2b45e8
      }
kusano 2b45e8
kusano 2b45e8
      if (width < 16) width = 16;
kusano 2b45e8
      if (width > m - i) width = m - i;
kusano 2b45e8
	
kusano 2b45e8
    } else {
kusano 2b45e8
      width = m - i;
kusano 2b45e8
    }
kusano 2b45e8
    
kusano 2b45e8
    range_m[num_cpu + 1] = range_m[num_cpu] + width;
kusano 2b45e8
      
kusano 2b45e8
    queue[num_cpu].mode    = mode;
kusano 2b45e8
    queue[num_cpu].routine = syr_kernel;
kusano 2b45e8
    queue[num_cpu].args    = &args;
kusano 2b45e8
    queue[num_cpu].range_m = &range_m[num_cpu];
kusano 2b45e8
    queue[num_cpu].range_n = NULL;
kusano 2b45e8
    queue[num_cpu].sa      = NULL;
kusano 2b45e8
    queue[num_cpu].sb      = NULL;
kusano 2b45e8
    queue[num_cpu].next    = &queue[num_cpu + 1];
kusano 2b45e8
    
kusano 2b45e8
    num_cpu ++;
kusano 2b45e8
    i += width;
kusano 2b45e8
  }
kusano 2b45e8
  
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
  if (num_cpu) {
kusano 2b45e8
    queue[0].sa = NULL;
kusano 2b45e8
    queue[0].sb = buffer;
kusano 2b45e8
    
kusano 2b45e8
    queue[num_cpu - 1].next = NULL;
kusano 2b45e8
    
kusano 2b45e8
    exec_blas(num_cpu, queue);
kusano 2b45e8
  }
kusano 2b45e8
   
kusano 2b45e8
  return 0;
kusano 2b45e8
}