2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012, by the GROMACS development team, led by
5 * David van der Spoel, Berk Hess, Erik Lindahl, and including many
6 * others, as listed in the AUTHORS file in the top-level source
7 * directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 #include "gmx_lapack.h"
40 F77_FUNC(slarfb,SLARFB)(const char *side,
56 int c_dim1, c_offset, t_dim1, t_offset, v_dim1, v_offset, work_dim1,
57 work_offset, i__1, i__2;
63 float minusone = -1.0;
66 v_offset = 1 + v_dim1;
69 t_offset = 1 + t_dim1;
72 c_offset = 1 + c_dim1;
75 work_offset = 1 + work_dim1;
78 if (*m <= 0 || *n <= 0) {
81 if (*trans=='N' || *trans=='n') {
82 *(unsigned char *)transt = 'T';
84 *(unsigned char *)transt = 'N';
87 if (*storev=='C' || *storev=='c') {
89 if (*direct=='F' || *direct=='f') {
90 if (*side=='l' || *side=='L') {
93 for (j = 1; j <= i__1; ++j) {
94 F77_FUNC(scopy,SCOPY)(n, &c__[j + c_dim1], ldc, &work[j * work_dim1 + 1],
98 F77_FUNC(strmm,STRMM)("Right", "Lower", "No transpose", "Unit", n, k, &one,
99 &v[v_offset], ldv, &work[work_offset], ldwork);
103 F77_FUNC(sgemm,SGEMM)("Transpose", "No transpose", n, k, &i__1, &one, &
104 c__[*k + 1 + c_dim1], ldc, &v[*k + 1 + v_dim1],
105 ldv, &one, &work[work_offset], ldwork);
108 F77_FUNC(strmm,STRMM)("Right", "Upper", transt, "Non-unit", n, k, &one, &t[
109 t_offset], ldt, &work[work_offset], ldwork);
113 F77_FUNC(sgemm,SGEMM)("No transpose", "Transpose", &i__1, n, k, &minusone, &
114 v[*k + 1 + v_dim1], ldv, &work[work_offset],
115 ldwork, &one, &c__[*k + 1 + c_dim1], ldc);
118 F77_FUNC(strmm,STRMM)("Right", "Lower", "Transpose", "Unit", n, k, &one, &
119 v[v_offset], ldv, &work[work_offset], ldwork);
122 for (j = 1; j <= i__1; ++j) {
124 for (i__ = 1; i__ <= i__2; ++i__) {
125 c__[j + i__ * c_dim1] -= work[i__ + j * work_dim1];
129 } else if (*side=='r' || *side=='R') {
132 for (j = 1; j <= i__1; ++j) {
133 F77_FUNC(scopy,SCOPY)(m, &c__[j * c_dim1 + 1], &c__1, &work[j *
134 work_dim1 + 1], &c__1);
137 F77_FUNC(strmm,STRMM)("Right", "Lower", "No transpose", "Unit", m, k, &one,
138 &v[v_offset], ldv, &work[work_offset], ldwork);
142 F77_FUNC(sgemm,SGEMM)("No transpose", "No transpose", m, k, &i__1, &
143 one, &c__[(*k + 1) * c_dim1 + 1], ldc, &v[*k +
144 1 + v_dim1], ldv, &one, &work[work_offset],
148 F77_FUNC(strmm,STRMM)("Right", "Upper", trans, "Non-unit", m, k, &one, &t[
149 t_offset], ldt, &work[work_offset], ldwork);
153 F77_FUNC(sgemm,SGEMM)("No transpose", "Transpose", m, &i__1, k, &minusone, &
154 work[work_offset], ldwork, &v[*k + 1 + v_dim1],
155 ldv, &one, &c__[(*k + 1) * c_dim1 + 1], ldc);
158 F77_FUNC(strmm,STRMM)("Right", "Lower", "Transpose", "Unit", m, k, &one, &
159 v[v_offset], ldv, &work[work_offset], ldwork);
162 for (j = 1; j <= i__1; ++j) {
164 for (i__ = 1; i__ <= i__2; ++i__) {
165 c__[i__ + j * c_dim1] -= work[i__ + j * work_dim1];
172 if (*side=='l' || *side=='L') {
174 for (j = 1; j <= i__1; ++j) {
175 F77_FUNC(scopy,SCOPY)(n, &c__[*m - *k + j + c_dim1], ldc, &work[j *
176 work_dim1 + 1], &c__1);
179 F77_FUNC(strmm,STRMM)("Right", "Upper", "No transpose", "Unit", n, k, &one,
180 &v[*m - *k + 1 + v_dim1], ldv, &work[work_offset],
184 F77_FUNC(sgemm,SGEMM)("Transpose", "No transpose", n, k, &i__1, &one, &
185 c__[c_offset], ldc, &v[v_offset], ldv, &one, &
186 work[work_offset], ldwork);
189 F77_FUNC(strmm,STRMM)("Right", "Lower", transt, "Non-unit", n, k, &one, &t[
190 t_offset], ldt, &work[work_offset], ldwork);
195 F77_FUNC(sgemm,SGEMM)("No transpose", "Transpose", &i__1, n, k, &minusone, &
196 v[v_offset], ldv, &work[work_offset], ldwork, &
197 one, &c__[c_offset], ldc)
201 F77_FUNC(strmm,STRMM)("Right", "Upper", "Transpose", "Unit", n, k, &one, &
202 v[*m - *k + 1 + v_dim1], ldv, &work[work_offset],
206 for (j = 1; j <= i__1; ++j) {
208 for (i__ = 1; i__ <= i__2; ++i__) {
209 c__[*m - *k + j + i__ * c_dim1] -= work[i__ + j *
214 } else if (*side=='r' || *side=='R') {
216 for (j = 1; j <= i__1; ++j) {
217 F77_FUNC(scopy,SCOPY)(m, &c__[(*n - *k + j) * c_dim1 + 1], &c__1, &work[
218 j * work_dim1 + 1], &c__1);
221 F77_FUNC(strmm,STRMM)("Right", "Upper", "No transpose", "Unit", m, k, &one,
222 &v[*n - *k + 1 + v_dim1], ldv, &work[work_offset],
226 F77_FUNC(sgemm,SGEMM)("No transpose", "No transpose", m, k, &i__1, &
227 one, &c__[c_offset], ldc, &v[v_offset], ldv, &
228 one, &work[work_offset], ldwork);
230 F77_FUNC(strmm,STRMM)("Right", "Lower", trans, "Non-unit", m, k, &one, &t[
231 t_offset], ldt, &work[work_offset], ldwork);
234 F77_FUNC(sgemm,SGEMM)("No transpose", "Transpose", m, &i__1, k, &minusone, &
235 work[work_offset], ldwork, &v[v_offset], ldv, &
236 one, &c__[c_offset], ldc)
239 F77_FUNC(strmm,STRMM)("Right", "Upper", "Transpose", "Unit", m, k, &one, &
240 v[*n - *k + 1 + v_dim1], ldv, &work[work_offset],
243 for (j = 1; j <= i__1; ++j) {
245 for (i__ = 1; i__ <= i__2; ++i__) {
246 c__[i__ + (*n - *k + j) * c_dim1] -= work[i__ + j *
253 } else if (*storev=='r' || *storev=='R') {
254 if (*direct=='F' || *direct=='f') {
255 if (*side=='l' || *side=='L') {
257 for (j = 1; j <= i__1; ++j) {
258 F77_FUNC(scopy,SCOPY)(n, &c__[j + c_dim1], ldc, &work[j * work_dim1 + 1],
261 F77_FUNC(strmm,STRMM)("Right", "Upper", "Transpose", "Unit", n, k, &one, &
262 v[v_offset], ldv, &work[work_offset], ldwork);
265 F77_FUNC(sgemm,SGEMM)("Transpose", "Transpose", n, k, &i__1, &one, &
266 c__[*k + 1 + c_dim1], ldc, &v[(*k + 1) * v_dim1 +
267 1], ldv, &one, &work[work_offset], ldwork);
270 F77_FUNC(strmm,STRMM)("Right", "Upper", transt, "Non-unit", n, k, &one, &t[
271 t_offset], ldt, &work[work_offset], ldwork);
275 F77_FUNC(sgemm,SGEMM)("Transpose", "Transpose", &i__1, n, k, &minusone, &v[(
276 *k + 1) * v_dim1 + 1], ldv, &work[work_offset],
277 ldwork, &one, &c__[*k + 1 + c_dim1], ldc);
280 F77_FUNC(strmm,STRMM)("Right", "Upper", "No transpose", "Unit", n, k, &one,
281 &v[v_offset], ldv, &work[work_offset], ldwork);
284 for (j = 1; j <= i__1; ++j) {
286 for (i__ = 1; i__ <= i__2; ++i__) {
287 c__[j + i__ * c_dim1] -= work[i__ + j * work_dim1];
291 } else if (*side=='r' || *side=='R') {
294 for (j = 1; j <= i__1; ++j) {
295 F77_FUNC(scopy,SCOPY)(m, &c__[j * c_dim1 + 1], &c__1, &work[j *
296 work_dim1 + 1], &c__1);
299 F77_FUNC(strmm,STRMM)("Right", "Upper", "Transpose", "Unit", m, k, &one, &
300 v[v_offset], ldv, &work[work_offset], ldwork);
304 F77_FUNC(sgemm,SGEMM)("No transpose", "Transpose", m, k, &i__1, &one, &
305 c__[(*k + 1) * c_dim1 + 1], ldc, &v[(*k + 1) *
306 v_dim1 + 1], ldv, &one, &work[work_offset],
310 F77_FUNC(strmm,STRMM)("Right", "Upper", trans, "Non-unit", m, k, &one, &t[
311 t_offset], ldt, &work[work_offset], ldwork);
316 F77_FUNC(sgemm,SGEMM)("No transpose", "No transpose", m, &i__1, k, &
317 minusone, &work[work_offset], ldwork, &v[(*k + 1) *
318 v_dim1 + 1], ldv, &one, &c__[(*k + 1) * c_dim1
321 F77_FUNC(strmm,STRMM)("Right", "Upper", "No transpose", "Unit", m, k, &one,
322 &v[v_offset], ldv, &work[work_offset], ldwork);
324 for (j = 1; j <= i__1; ++j) {
326 for (i__ = 1; i__ <= i__2; ++i__) {
327 c__[i__ + j * c_dim1] -= work[i__ + j * work_dim1];
335 if (*side=='l' || *side=='L') {
338 for (j = 1; j <= i__1; ++j) {
339 F77_FUNC(scopy,SCOPY)(n, &c__[*m - *k + j + c_dim1], ldc, &work[j *
340 work_dim1 + 1], &c__1);
343 F77_FUNC(strmm,STRMM)("Right", "Lower", "Transpose", "Unit", n, k, &one, &
344 v[(*m - *k + 1) * v_dim1 + 1], ldv, &work[work_offset]
349 F77_FUNC(sgemm,SGEMM)("Transpose", "Transpose", n, k, &i__1, &one, &
350 c__[c_offset], ldc, &v[v_offset], ldv, &one, &
351 work[work_offset], ldwork);
354 F77_FUNC(strmm,STRMM)("Right", "Lower", transt, "Non-unit", n, k, &one, &t[
355 t_offset], ldt, &work[work_offset], ldwork);
360 F77_FUNC(sgemm,SGEMM)("Transpose", "Transpose", &i__1, n, k, &minusone, &v[
361 v_offset], ldv, &work[work_offset], ldwork, &
362 one, &c__[c_offset], ldc);
365 F77_FUNC(strmm,STRMM)("Right", "Lower", "No transpose", "Unit", n, k, &one,
366 &v[(*m - *k + 1) * v_dim1 + 1], ldv, &work[
367 work_offset], ldwork);
370 for (j = 1; j <= i__1; ++j) {
372 for (i__ = 1; i__ <= i__2; ++i__) {
373 c__[*m - *k + j + i__ * c_dim1] -= work[i__ + j *
378 } else if (*side=='r' || *side=='R') {
381 for (j = 1; j <= i__1; ++j) {
382 F77_FUNC(scopy,SCOPY)(m, &c__[(*n - *k + j) * c_dim1 + 1], &c__1, &work[
383 j * work_dim1 + 1], &c__1);
386 F77_FUNC(strmm,STRMM)("Right", "Lower", "Transpose", "Unit", m, k, &one, &
387 v[(*n - *k + 1) * v_dim1 + 1], ldv, &work[work_offset]
392 F77_FUNC(sgemm,SGEMM)("No transpose", "Transpose", m, k, &i__1, &one, &
393 c__[c_offset], ldc, &v[v_offset], ldv, &one, &
394 work[work_offset], ldwork);
397 F77_FUNC(strmm,STRMM)("Right", "Lower", trans, "Non-unit", m, k, &one, &t[
398 t_offset], ldt, &work[work_offset], ldwork);
403 F77_FUNC(sgemm,SGEMM)("No transpose", "No transpose", m, &i__1, k, &
404 minusone, &work[work_offset], ldwork, &v[v_offset],
405 ldv, &one, &c__[c_offset], ldc);
408 F77_FUNC(strmm,STRMM)("Right", "Lower", "No transpose", "Unit", m, k, &one,
409 &v[(*n - *k + 1) * v_dim1 + 1], ldv, &work[
410 work_offset], ldwork);
413 for (j = 1; j <= i__1; ++j) {
415 for (i__ = 1; i__ <= i__2; ++i__) {
416 c__[i__ + (*n - *k + j) * c_dim1] -= work[i__ + j *