3 * Copyright (C) 2001, 2007 Brian Gough
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 3 of the License, or (at
8 * your option) any later version.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
21 const int conj = (order == CblasColMajor) ? -1 : 1;
24 const BASE alpha_real = CONST_REAL0(alpha);
25 const BASE alpha_imag = CONST_IMAG0(alpha);
27 const BASE beta_real = CONST_REAL0(beta);
28 const BASE beta_imag = CONST_IMAG0(beta);
30 if ((alpha_real == 0.0 && alpha_imag == 0.0)
31 && (beta_real == 1.0 && beta_imag == 0.0))
34 /* form y := beta*y */
35 if (beta_real == 0.0 && beta_imag == 0.0) {
36 INDEX iy = OFFSET(N, incY);
37 for (i = 0; i < N; i++) {
42 } else if (!(beta_real == 1.0 && beta_imag == 0.0)) {
43 INDEX iy = OFFSET(N, incY);
44 for (i = 0; i < N; i++) {
45 const BASE y_real = REAL(Y, iy);
46 const BASE y_imag = IMAG(Y, iy);
47 const BASE tmpR = y_real * beta_real - y_imag * beta_imag;
48 const BASE tmpI = y_real * beta_imag + y_imag * beta_real;
55 if (alpha_real == 0.0 && alpha_imag == 0.0)
58 /* form y := alpha*A*x + y */
60 if ((order == CblasRowMajor && Uplo == CblasUpper)
61 || (order == CblasColMajor && Uplo == CblasLower)) {
62 INDEX ix = OFFSET(N, incX);
63 INDEX iy = OFFSET(N, incY);
64 for (i = 0; i < N; i++) {
65 BASE x_real = CONST_REAL(X, ix);
66 BASE x_imag = CONST_IMAG(X, ix);
67 BASE temp1_real = alpha_real * x_real - alpha_imag * x_imag;
68 BASE temp1_imag = alpha_real * x_imag + alpha_imag * x_real;
69 BASE temp2_real = 0.0;
70 BASE temp2_imag = 0.0;
71 const INDEX j_min = i + 1;
72 const INDEX j_max = N;
73 INDEX jx = OFFSET(N, incX) + j_min * incX;
74 INDEX jy = OFFSET(N, incY) + j_min * incY;
75 BASE Aii_real = CONST_REAL(A, lda * i + i);
76 /* Aii_imag is zero */
77 REAL(Y, iy) += temp1_real * Aii_real;
78 IMAG(Y, iy) += temp1_imag * Aii_real;
79 for (j = j_min; j < j_max; j++) {
80 BASE Aij_real = CONST_REAL(A, lda * i + j);
81 BASE Aij_imag = conj * CONST_IMAG(A, lda * i + j);
82 REAL(Y, jy) += temp1_real * Aij_real - temp1_imag * (-Aij_imag);
83 IMAG(Y, jy) += temp1_real * (-Aij_imag) + temp1_imag * Aij_real;
84 x_real = CONST_REAL(X, jx);
85 x_imag = CONST_IMAG(X, jx);
86 temp2_real += x_real * Aij_real - x_imag * Aij_imag;
87 temp2_imag += x_real * Aij_imag + x_imag * Aij_real;
91 REAL(Y, iy) += alpha_real * temp2_real - alpha_imag * temp2_imag;
92 IMAG(Y, iy) += alpha_real * temp2_imag + alpha_imag * temp2_real;
96 } else if ((order == CblasRowMajor && Uplo == CblasLower)
97 || (order == CblasColMajor && Uplo == CblasUpper)) {
98 INDEX ix = OFFSET(N, incX) + (N - 1) * incX;
99 INDEX iy = OFFSET(N, incY) + (N - 1) * incY;
100 for (i = N; i > 0 && i--;) {
101 BASE x_real = CONST_REAL(X, ix);
102 BASE x_imag = CONST_IMAG(X, ix);
103 BASE temp1_real = alpha_real * x_real - alpha_imag * x_imag;
104 BASE temp1_imag = alpha_real * x_imag + alpha_imag * x_real;
105 BASE temp2_real = 0.0;
106 BASE temp2_imag = 0.0;
107 const INDEX j_min = 0;
108 const INDEX j_max = i;
109 INDEX jx = OFFSET(N, incX) + j_min * incX;
110 INDEX jy = OFFSET(N, incY) + j_min * incY;
111 BASE Aii_real = CONST_REAL(A, lda * i + i);
112 /* Aii_imag is zero */
113 REAL(Y, iy) += temp1_real * Aii_real;
114 IMAG(Y, iy) += temp1_imag * Aii_real;
116 for (j = j_min; j < j_max; j++) {
117 BASE Aij_real = CONST_REAL(A, lda * i + j);
118 BASE Aij_imag = conj * CONST_IMAG(A, lda * i + j);
119 REAL(Y, jy) += temp1_real * Aij_real - temp1_imag * (-Aij_imag);
120 IMAG(Y, jy) += temp1_real * (-Aij_imag) + temp1_imag * Aij_real;
121 x_real = CONST_REAL(X, jx);
122 x_imag = CONST_IMAG(X, jx);
123 temp2_real += x_real * Aij_real - x_imag * Aij_imag;
124 temp2_imag += x_real * Aij_imag + x_imag * Aij_real;
128 REAL(Y, iy) += alpha_real * temp2_real - alpha_imag * temp2_imag;
129 IMAG(Y, iy) += alpha_real * temp2_imag + alpha_imag * temp2_real;
134 BLAS_ERROR("unrecognized operation");