|
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 <ctype.h></ctype.h>
|
|
kusano |
2b45e8 |
#include "common.h"
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
const static FLOAT dp1 = 1.;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
int CNAME(BLASLONG n, BLASLONG k, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG incb, void *buffer){
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
BLASLONG i;
|
|
kusano |
2b45e8 |
FLOAT *gemvbuffer = (FLOAT *)buffer;
|
|
kusano |
2b45e8 |
FLOAT *B = b;
|
|
kusano |
2b45e8 |
BLASLONG length;
|
|
kusano |
2b45e8 |
#if (TRANSA == 2) || (TRANSA == 4)
|
|
kusano |
2b45e8 |
FLOAT _Complex temp;
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
#ifndef UNIT
|
|
kusano |
2b45e8 |
FLOAT ar, ai, br, bi, ratio, den;
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (incb != 1) {
|
|
kusano |
2b45e8 |
B = buffer;
|
|
kusano |
2b45e8 |
gemvbuffer = (FLOAT *)(((BLASLONG)buffer + n * sizeof(FLOAT) * COMPSIZE+ 4095) & ~4095);
|
|
kusano |
2b45e8 |
COPY_K(n, b, incb, buffer, 1);
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
a += (n - 1) * lda * COMPSIZE;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
for (i = n - 1; i >= 0; i--) {
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#if (TRANSA == 2) || (TRANSA == 4)
|
|
kusano |
2b45e8 |
length = n - i - 1;
|
|
kusano |
2b45e8 |
if (length > k) length = k;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (length > 0) {
|
|
kusano |
2b45e8 |
#if TRANSA == 2
|
|
kusano |
2b45e8 |
temp = DOTU_K(length, a + COMPSIZE, 1, B + (i + 1) * COMPSIZE, 1);
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
temp = DOTC_K(length, a + COMPSIZE, 1, B + (i + 1) * COMPSIZE, 1);
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
B[i * 2 + 0] -= CREAL(temp);
|
|
kusano |
2b45e8 |
B[i * 2 + 1] -= CIMAG(temp);
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifndef UNIT
|
|
kusano |
2b45e8 |
#if (TRANSA == 1) || (TRANSA == 3)
|
|
kusano |
2b45e8 |
ar = a[k * 2 + 0];
|
|
kusano |
2b45e8 |
ai = a[k * 2 + 1];
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
ar = a[0];
|
|
kusano |
2b45e8 |
ai = a[1];
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (fabs(ar) >= fabs(ai)){
|
|
kusano |
2b45e8 |
ratio = ai / ar;
|
|
kusano |
2b45e8 |
den = 1./(ar * ( 1 + ratio * ratio));
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
ar = den;
|
|
kusano |
2b45e8 |
#if TRANSA < 3
|
|
kusano |
2b45e8 |
ai = -ratio * den;
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
ai = ratio * den;
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
} else {
|
|
kusano |
2b45e8 |
ratio = ar / ai;
|
|
kusano |
2b45e8 |
den = 1./(ai * ( 1 + ratio * ratio));
|
|
kusano |
2b45e8 |
ar = ratio * den;
|
|
kusano |
2b45e8 |
#if TRANSA < 3
|
|
kusano |
2b45e8 |
ai = -den;
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
ai = den;
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
br = B[i * 2 + 0];
|
|
kusano |
2b45e8 |
bi = B[i * 2 + 1];
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
B[i * 2 + 0] = ar*br - ai*bi;
|
|
kusano |
2b45e8 |
B[i * 2 + 1] = ar*bi + ai*br;
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#if (TRANSA == 1) || (TRANSA == 3)
|
|
kusano |
2b45e8 |
length = i;
|
|
kusano |
2b45e8 |
if (length > k) length = k;
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (length > 0) {
|
|
kusano |
2b45e8 |
#if TRANSA == 1
|
|
kusano |
2b45e8 |
AXPYU_K(length, 0, 0,
|
|
kusano |
2b45e8 |
-B[i * 2 + 0], -B[i * 2 + 1],
|
|
kusano |
2b45e8 |
a + (k - length) * COMPSIZE, 1, B + (i - length) * COMPSIZE, 1, NULL, 0);
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
AXPYC_K(length, 0, 0,
|
|
kusano |
2b45e8 |
-B[i * 2 + 0], -B[i * 2 + 1],
|
|
kusano |
2b45e8 |
a + (k - length) * COMPSIZE, 1, B + (i - length) * COMPSIZE, 1, NULL, 0);
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
a -= lda * COMPSIZE;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
if (incb != 1) {
|
|
kusano |
2b45e8 |
COPY_K(n, buffer, 1, b, incb);
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
return 0;
|
|
kusano |
2b45e8 |
}
|
|
kusano |
2b45e8 |
|