|
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 |
int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG offset, FLOAT *b){
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
BLASLONG i, ii, j, jj;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
FLOAT data01, data02, data03, data04;
|
|
kusano |
2b45e8 |
FLOAT data05, data06, data07, data08;
|
|
kusano |
2b45e8 |
FLOAT *a1, *a2;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
lda *= 2;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
jj = offset;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
j = (n >> 1);
|
|
kusano |
2b45e8 |
while (j > 0){
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
a1 = a + 0 * lda;
|
|
kusano |
2b45e8 |
a2 = a + 1 * lda;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
i = (m >> 1);
|
|
kusano |
2b45e8 |
ii = 0;
|
|
kusano |
2b45e8 |
while (i > 0) {
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (ii == jj) {
|
|
kusano |
2b45e8 |
#ifndef UNIT
|
|
kusano |
2b45e8 |
data01 = *(a1 + 0);
|
|
kusano |
2b45e8 |
data02 = *(a1 + 1);
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
data03 = *(a1 + 2);
|
|
kusano |
2b45e8 |
data04 = *(a1 + 3);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifndef UNIT
|
|
kusano |
2b45e8 |
data07 = *(a2 + 2);
|
|
kusano |
2b45e8 |
data08 = *(a2 + 3);
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
compinv(b + 0, data01, data02);
|
|
kusano |
2b45e8 |
*(b + 2) = data03;
|
|
kusano |
2b45e8 |
*(b + 3) = data04;
|
|
kusano |
2b45e8 |
compinv(b + 6, data07, data08);
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (ii < jj) {
|
|
kusano |
2b45e8 |
data01 = *(a1 + 0);
|
|
kusano |
2b45e8 |
data02 = *(a1 + 1);
|
|
kusano |
2b45e8 |
data03 = *(a1 + 2);
|
|
kusano |
2b45e8 |
data04 = *(a1 + 3);
|
|
kusano |
2b45e8 |
data05 = *(a2 + 0);
|
|
kusano |
2b45e8 |
data06 = *(a2 + 1);
|
|
kusano |
2b45e8 |
data07 = *(a2 + 2);
|
|
kusano |
2b45e8 |
data08 = *(a2 + 3);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
*(b + 0) = data01;
|
|
kusano |
2b45e8 |
*(b + 1) = data02;
|
|
kusano |
2b45e8 |
*(b + 2) = data03;
|
|
kusano |
2b45e8 |
*(b + 3) = data04;
|
|
kusano |
2b45e8 |
*(b + 4) = data05;
|
|
kusano |
2b45e8 |
*(b + 5) = data06;
|
|
kusano |
2b45e8 |
*(b + 6) = data07;
|
|
kusano |
2b45e8 |
*(b + 7) = data08;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
a1 += 2 * lda;
|
|
kusano |
2b45e8 |
a2 += 2 * lda;
|
|
kusano |
2b45e8 |
b += 8;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
i --;
|
|
kusano |
2b45e8 |
ii += 2;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (m & 1) {
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (ii== jj) {
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifndef UNIT
|
|
kusano |
2b45e8 |
data01 = *(a1 + 0);
|
|
kusano |
2b45e8 |
data02 = *(a1 + 1);
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
data03 = *(a1 + 2);
|
|
kusano |
2b45e8 |
data04 = *(a1 + 3);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
compinv(b + 0, data01, data02);
|
|
kusano |
2b45e8 |
*(b + 2) = data03;
|
|
kusano |
2b45e8 |
*(b + 3) = data04;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (ii < jj) {
|
|
kusano |
2b45e8 |
data01 = *(a1 + 0);
|
|
kusano |
2b45e8 |
data02 = *(a1 + 1);
|
|
kusano |
2b45e8 |
data03 = *(a1 + 2);
|
|
kusano |
2b45e8 |
data04 = *(a1 + 3);
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
*(b + 0) = data01;
|
|
kusano |
2b45e8 |
*(b + 1) = data02;
|
|
kusano |
2b45e8 |
*(b + 2) = data03;
|
|
kusano |
2b45e8 |
*(b + 3) = data04;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
b += 4;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
a += 4;
|
|
kusano |
2b45e8 |
jj += 2;
|
|
kusano |
2b45e8 |
j --;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (n & 1) {
|
|
kusano |
2b45e8 |
a1 = a + 0 * lda;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
i = m;
|
|
kusano |
2b45e8 |
ii = 0;
|
|
kusano |
2b45e8 |
while (i > 0) {
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (ii == jj) {
|
|
kusano |
2b45e8 |
#ifndef UNIT
|
|
kusano |
2b45e8 |
data01 = *(a1 + 0);
|
|
kusano |
2b45e8 |
data02 = *(a1 + 1);
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
compinv(b + 0, data01, data02);
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (ii < jj) {
|
|
kusano |
2b45e8 |
data01 = *(a1 + 0);
|
|
kusano |
2b45e8 |
data02 = *(a1 + 1);
|
|
kusano |
2b45e8 |
*(b + 0) = data01;
|
|
kusano |
2b45e8 |
*(b + 1) = data02;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
a1 += 1 * lda;
|
|
kusano |
2b45e8 |
b += 2;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
i --;
|
|
kusano |
2b45e8 |
ii += 1;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
return 0;
|
|
kusano |
2b45e8 |
}
|