|
Packit |
67cb25 |
/* blas/source_trsv_c.h
|
|
Packit |
67cb25 |
*
|
|
Packit |
67cb25 |
* Copyright (C) 1996, 1997, 1998, 1999, 2000 Gerard Jungman
|
|
Packit |
67cb25 |
*
|
|
Packit |
67cb25 |
* This program is free software; you can redistribute it and/or modify
|
|
Packit |
67cb25 |
* it under the terms of the GNU General Public License as published by
|
|
Packit |
67cb25 |
* the Free Software Foundation; either version 3 of the License, or (at
|
|
Packit |
67cb25 |
* your option) any later version.
|
|
Packit |
67cb25 |
*
|
|
Packit |
67cb25 |
* This program is distributed in the hope that it will be useful, but
|
|
Packit |
67cb25 |
* WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
Packit |
67cb25 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
Packit |
67cb25 |
* General Public License for more details.
|
|
Packit |
67cb25 |
*
|
|
Packit |
67cb25 |
* You should have received a copy of the GNU General Public License
|
|
Packit |
67cb25 |
* along with this program; if not, write to the Free Software
|
|
Packit |
67cb25 |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
|
|
Packit |
67cb25 |
*/
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
{
|
|
Packit |
67cb25 |
const int conj = (TransA == CblasConjTrans) ? -1 : 1;
|
|
Packit |
67cb25 |
const int Trans = (TransA != CblasConjTrans) ? TransA : CblasTrans;
|
|
Packit |
67cb25 |
const int nonunit = (Diag == CblasNonUnit);
|
|
Packit |
67cb25 |
INDEX i, j;
|
|
Packit |
67cb25 |
INDEX ix, jx;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
CHECK_ARGS9(TRSV,order,Uplo,TransA,Diag,N,A,lda,X,incX);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
if (N == 0)
|
|
Packit |
67cb25 |
return;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* form x := inv( A )*x */
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
if ((order == CblasRowMajor && Trans == CblasNoTrans && Uplo == CblasUpper)
|
|
Packit |
67cb25 |
|| (order == CblasColMajor && Trans == CblasTrans && Uplo == CblasLower)) {
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
ix = OFFSET(N, incX) + incX * (N - 1);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
if (nonunit) {
|
|
Packit |
67cb25 |
const BASE a_real = CONST_REAL(A, lda * (N - 1) + (N - 1));
|
|
Packit |
67cb25 |
const BASE a_imag = conj * CONST_IMAG(A, lda * (N - 1) + (N - 1));
|
|
Packit |
67cb25 |
const BASE x_real = REAL(X, ix);
|
|
Packit |
67cb25 |
const BASE x_imag = IMAG(X, ix);
|
|
Packit |
67cb25 |
const BASE s = xhypot(a_real, a_imag);
|
|
Packit |
67cb25 |
const BASE b_real = a_real / s;
|
|
Packit |
67cb25 |
const BASE b_imag = a_imag / s;
|
|
Packit |
67cb25 |
REAL(X, ix) = (x_real * b_real + x_imag * b_imag) / s;
|
|
Packit |
67cb25 |
IMAG(X, ix) = (x_imag * b_real - b_imag * x_real) / s;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
ix -= incX;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
for (i = N - 1; i > 0 && i--;) {
|
|
Packit |
67cb25 |
BASE tmp_real = REAL(X, ix);
|
|
Packit |
67cb25 |
BASE tmp_imag = IMAG(X, ix);
|
|
Packit |
67cb25 |
jx = ix + incX;
|
|
Packit |
67cb25 |
for (j = i + 1; j < N; j++) {
|
|
Packit |
67cb25 |
const BASE Aij_real = CONST_REAL(A, lda * i + j);
|
|
Packit |
67cb25 |
const BASE Aij_imag = conj * CONST_IMAG(A, lda * i + j);
|
|
Packit |
67cb25 |
const BASE x_real = REAL(X, jx);
|
|
Packit |
67cb25 |
const BASE x_imag = IMAG(X, jx);
|
|
Packit |
67cb25 |
tmp_real -= Aij_real * x_real - Aij_imag * x_imag;
|
|
Packit |
67cb25 |
tmp_imag -= Aij_real * x_imag + Aij_imag * x_real;
|
|
Packit |
67cb25 |
jx += incX;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
if (nonunit) {
|
|
Packit |
67cb25 |
const BASE a_real = CONST_REAL(A, lda * i + i);
|
|
Packit |
67cb25 |
const BASE a_imag = conj * CONST_IMAG(A, lda * i + i);
|
|
Packit |
67cb25 |
const BASE s = xhypot(a_real, a_imag);
|
|
Packit |
67cb25 |
const BASE b_real = a_real / s;
|
|
Packit |
67cb25 |
const BASE b_imag = a_imag / s;
|
|
Packit |
67cb25 |
REAL(X, ix) = (tmp_real * b_real + tmp_imag * b_imag) / s;
|
|
Packit |
67cb25 |
IMAG(X, ix) = (tmp_imag * b_real - tmp_real * b_imag) / s;
|
|
Packit |
67cb25 |
} else {
|
|
Packit |
67cb25 |
REAL(X, ix) = tmp_real;
|
|
Packit |
67cb25 |
IMAG(X, ix) = tmp_imag;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
ix -= incX;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
} else if ((order == CblasRowMajor && Trans == CblasNoTrans && Uplo == CblasLower)
|
|
Packit |
67cb25 |
|| (order == CblasColMajor && Trans == CblasTrans && Uplo == CblasUpper)) {
|
|
Packit |
67cb25 |
/* forward substitution */
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
ix = OFFSET(N, incX);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
if (nonunit) {
|
|
Packit |
67cb25 |
const BASE a_real = CONST_REAL(A, lda * 0 + 0);
|
|
Packit |
67cb25 |
const BASE a_imag = conj * CONST_IMAG(A, lda * 0 + 0);
|
|
Packit |
67cb25 |
const BASE x_real = REAL(X, ix);
|
|
Packit |
67cb25 |
const BASE x_imag = IMAG(X, ix);
|
|
Packit |
67cb25 |
const BASE s = xhypot(a_real, a_imag);
|
|
Packit |
67cb25 |
const BASE b_real = a_real / s;
|
|
Packit |
67cb25 |
const BASE b_imag = a_imag / s;
|
|
Packit |
67cb25 |
REAL(X, ix) = (x_real * b_real + x_imag * b_imag) / s;
|
|
Packit |
67cb25 |
IMAG(X, ix) = (x_imag * b_real - b_imag * x_real) / s;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
ix += incX;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
for (i = 1; i < N; i++) {
|
|
Packit |
67cb25 |
BASE tmp_real = REAL(X, ix);
|
|
Packit |
67cb25 |
BASE tmp_imag = IMAG(X, ix);
|
|
Packit |
67cb25 |
jx = OFFSET(N, incX);
|
|
Packit |
67cb25 |
for (j = 0; j < i; j++) {
|
|
Packit |
67cb25 |
const BASE Aij_real = CONST_REAL(A, lda * i + j);
|
|
Packit |
67cb25 |
const BASE Aij_imag = conj * CONST_IMAG(A, lda * i + j);
|
|
Packit |
67cb25 |
const BASE x_real = REAL(X, jx);
|
|
Packit |
67cb25 |
const BASE x_imag = IMAG(X, jx);
|
|
Packit |
67cb25 |
tmp_real -= Aij_real * x_real - Aij_imag * x_imag;
|
|
Packit |
67cb25 |
tmp_imag -= Aij_real * x_imag + Aij_imag * x_real;
|
|
Packit |
67cb25 |
jx += incX;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
if (nonunit) {
|
|
Packit |
67cb25 |
const BASE a_real = CONST_REAL(A, lda * i + i);
|
|
Packit |
67cb25 |
const BASE a_imag = conj * CONST_IMAG(A, lda * i + i);
|
|
Packit |
67cb25 |
const BASE s = xhypot(a_real, a_imag);
|
|
Packit |
67cb25 |
const BASE b_real = a_real / s;
|
|
Packit |
67cb25 |
const BASE b_imag = a_imag / s;
|
|
Packit |
67cb25 |
REAL(X, ix) = (tmp_real * b_real + tmp_imag * b_imag) / s;
|
|
Packit |
67cb25 |
IMAG(X, ix) = (tmp_imag * b_real - tmp_real * b_imag) / s;
|
|
Packit |
67cb25 |
} else {
|
|
Packit |
67cb25 |
REAL(X, ix) = tmp_real;
|
|
Packit |
67cb25 |
IMAG(X, ix) = tmp_imag;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
ix += incX;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
} else if ((order == CblasRowMajor && Trans == CblasTrans && Uplo == CblasUpper)
|
|
Packit |
67cb25 |
|| (order == CblasColMajor && Trans == CblasNoTrans && Uplo == CblasLower)) {
|
|
Packit |
67cb25 |
/* form x := inv( A' )*x */
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* forward substitution */
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
ix = OFFSET(N, incX);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
if (nonunit) {
|
|
Packit |
67cb25 |
const BASE a_real = CONST_REAL(A, lda * 0 + 0);
|
|
Packit |
67cb25 |
const BASE a_imag = conj * CONST_IMAG(A, lda * 0 + 0);
|
|
Packit |
67cb25 |
const BASE x_real = REAL(X, ix);
|
|
Packit |
67cb25 |
const BASE x_imag = IMAG(X, ix);
|
|
Packit |
67cb25 |
const BASE s = xhypot(a_real, a_imag);
|
|
Packit |
67cb25 |
const BASE b_real = a_real / s;
|
|
Packit |
67cb25 |
const BASE b_imag = a_imag / s;
|
|
Packit |
67cb25 |
REAL(X, ix) = (x_real * b_real + x_imag * b_imag) / s;
|
|
Packit |
67cb25 |
IMAG(X, ix) = (x_imag * b_real - b_imag * x_real) / s;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
ix += incX;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
for (i = 1; i < N; i++) {
|
|
Packit |
67cb25 |
BASE tmp_real = REAL(X, ix);
|
|
Packit |
67cb25 |
BASE tmp_imag = IMAG(X, ix);
|
|
Packit |
67cb25 |
jx = OFFSET(N, incX);
|
|
Packit |
67cb25 |
for (j = 0; j < i; j++) {
|
|
Packit |
67cb25 |
const BASE Aij_real = CONST_REAL(A, lda * j + i);
|
|
Packit |
67cb25 |
const BASE Aij_imag = conj * CONST_IMAG(A, lda * j + i);
|
|
Packit |
67cb25 |
const BASE x_real = REAL(X, jx);
|
|
Packit |
67cb25 |
const BASE x_imag = IMAG(X, jx);
|
|
Packit |
67cb25 |
tmp_real -= Aij_real * x_real - Aij_imag * x_imag;
|
|
Packit |
67cb25 |
tmp_imag -= Aij_real * x_imag + Aij_imag * x_real;
|
|
Packit |
67cb25 |
jx += incX;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
if (nonunit) {
|
|
Packit |
67cb25 |
const BASE a_real = CONST_REAL(A, lda * i + i);
|
|
Packit |
67cb25 |
const BASE a_imag = conj * CONST_IMAG(A, lda * i + i);
|
|
Packit |
67cb25 |
const BASE s = xhypot(a_real, a_imag);
|
|
Packit |
67cb25 |
const BASE b_real = a_real / s;
|
|
Packit |
67cb25 |
const BASE b_imag = a_imag / s;
|
|
Packit |
67cb25 |
REAL(X, ix) = (tmp_real * b_real + tmp_imag * b_imag) / s;
|
|
Packit |
67cb25 |
IMAG(X, ix) = (tmp_imag * b_real - tmp_real * b_imag) / s;
|
|
Packit |
67cb25 |
} else {
|
|
Packit |
67cb25 |
REAL(X, ix) = tmp_real;
|
|
Packit |
67cb25 |
IMAG(X, ix) = tmp_imag;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
ix += incX;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
} else if ((order == CblasRowMajor && Trans == CblasTrans && Uplo == CblasLower)
|
|
Packit |
67cb25 |
|| (order == CblasColMajor && Trans == CblasNoTrans && Uplo == CblasUpper)) {
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
/* backsubstitution */
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
ix = OFFSET(N, incX) + incX * (N - 1);
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
if (nonunit) {
|
|
Packit |
67cb25 |
const BASE a_real = CONST_REAL(A, lda * (N - 1) + (N - 1));
|
|
Packit |
67cb25 |
const BASE a_imag = conj * CONST_IMAG(A, lda * (N - 1) + (N - 1));
|
|
Packit |
67cb25 |
const BASE x_real = REAL(X, ix);
|
|
Packit |
67cb25 |
const BASE x_imag = IMAG(X, ix);
|
|
Packit |
67cb25 |
const BASE s = xhypot(a_real, a_imag);
|
|
Packit |
67cb25 |
const BASE b_real = a_real / s;
|
|
Packit |
67cb25 |
const BASE b_imag = a_imag / s;
|
|
Packit |
67cb25 |
REAL(X, ix) = (x_real * b_real + x_imag * b_imag) / s;
|
|
Packit |
67cb25 |
IMAG(X, ix) = (x_imag * b_real - b_imag * x_real) / s;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
ix -= incX;
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
for (i = N - 1; i > 0 && i--;) {
|
|
Packit |
67cb25 |
BASE tmp_real = REAL(X, ix);
|
|
Packit |
67cb25 |
BASE tmp_imag = IMAG(X, ix);
|
|
Packit |
67cb25 |
jx = ix + incX;
|
|
Packit |
67cb25 |
for (j = i + 1; j < N; j++) {
|
|
Packit |
67cb25 |
const BASE Aij_real = CONST_REAL(A, lda * j + i);
|
|
Packit |
67cb25 |
const BASE Aij_imag = conj * CONST_IMAG(A, lda * j + i);
|
|
Packit |
67cb25 |
const BASE x_real = REAL(X, jx);
|
|
Packit |
67cb25 |
const BASE x_imag = IMAG(X, jx);
|
|
Packit |
67cb25 |
tmp_real -= Aij_real * x_real - Aij_imag * x_imag;
|
|
Packit |
67cb25 |
tmp_imag -= Aij_real * x_imag + Aij_imag * x_real;
|
|
Packit |
67cb25 |
jx += incX;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
|
|
Packit |
67cb25 |
if (nonunit) {
|
|
Packit |
67cb25 |
const BASE a_real = CONST_REAL(A, lda * i + i);
|
|
Packit |
67cb25 |
const BASE a_imag = conj * CONST_IMAG(A, lda * i + i);
|
|
Packit |
67cb25 |
const BASE s = xhypot(a_real, a_imag);
|
|
Packit |
67cb25 |
const BASE b_real = a_real / s;
|
|
Packit |
67cb25 |
const BASE b_imag = a_imag / s;
|
|
Packit |
67cb25 |
REAL(X, ix) = (tmp_real * b_real + tmp_imag * b_imag) / s;
|
|
Packit |
67cb25 |
IMAG(X, ix) = (tmp_imag * b_real - tmp_real * b_imag) / s;
|
|
Packit |
67cb25 |
} else {
|
|
Packit |
67cb25 |
REAL(X, ix) = tmp_real;
|
|
Packit |
67cb25 |
IMAG(X, ix) = tmp_imag;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
ix -= incX;
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
} else {
|
|
Packit |
67cb25 |
BLAS_ERROR("unrecognized operation");
|
|
Packit |
67cb25 |
}
|
|
Packit |
67cb25 |
}
|