1 /* ---------------------------------------------------------------------
2 *
3 * -- PBLAS routine (version 2.0) --
4 * University of Tennessee, Knoxville, Oak Ridge National Laboratory,
5 * and University of California, Berkeley.
6 * April 1, 1998
7 *
8 * ---------------------------------------------------------------------
9 */
10 /*
11 * Include files
12 */
13 #include "pblas.h"
14 #include "PBpblas.h"
15 #include "PBtools.h"
16 #include "PBblacs.h"
17 #include "PBblas.h"
18
19 #ifdef __STDC__
pdtrmm_(F_CHAR_T SIDE,F_CHAR_T UPLO,F_CHAR_T TRANS,F_CHAR_T DIAG,int * M,int * N,double * ALPHA,double * A,int * IA,int * JA,int * DESCA,double * B,int * IB,int * JB,int * DESCB)20 void pdtrmm_( F_CHAR_T SIDE, F_CHAR_T UPLO, F_CHAR_T TRANS, F_CHAR_T DIAG,
21 int * M, int * N, double * ALPHA,
22 double * A, int * IA, int * JA, int * DESCA,
23 double * B, int * IB, int * JB, int * DESCB )
24 #else
25 void pdtrmm_( SIDE, UPLO, TRANS, DIAG, M, N, ALPHA,
26 A, IA, JA, DESCA, B, IB, JB, DESCB )
27 /*
28 * .. Scalar Arguments ..
29 */
30 F_CHAR_T DIAG, SIDE, TRANS, UPLO;
31 int * IA, * IB, * JA, * JB, * M, * N;
32 double * ALPHA;
33 /*
34 * .. Array Arguments ..
35 */
36 int * DESCA, * DESCB;
37 double * A, * B;
38 #endif
39 {
40 /*
41 * Purpose
42 * =======
43 *
44 * PDTRMM performs one of the matrix-matrix operations
45 *
46 * sub( B ) := alpha * op( sub( A ) ) * sub( B ),
47 *
48 * or
49 *
50 * sub( B ) := alpha * sub( B ) * op( sub( A ) ),
51 *
52 * where
53 *
54 * sub( A ) denotes A(IA:IA+M-1,JA:JA+M-1) if SIDE = 'L',
55 * A(IA:IA+N-1,JA:JA+N-1) if SIDE = 'R', and,
56 *
57 * sub( B ) denotes B(IB:IB+M-1,JB:JB+N-1).
58 *
59 * Alpha is a scalar, sub( B ) is an m by n submatrix, sub( A ) is a
60 * unit, or non-unit, upper or lower triangular submatrix and op( X ) is
61 * one of
62 *
63 * op( X ) = X or op( X ) = X'.
64 *
65 * Notes
66 * =====
67 *
68 * A description vector is associated with each 2D block-cyclicly dis-
69 * tributed matrix. This vector stores the information required to
70 * establish the mapping between a matrix entry and its corresponding
71 * process and memory location.
72 *
73 * In the following comments, the character _ should be read as
74 * "of the distributed matrix". Let A be a generic term for any 2D
75 * block cyclicly distributed matrix. Its description vector is DESC_A:
76 *
77 * NOTATION STORED IN EXPLANATION
78 * ---------------- --------------- ------------------------------------
79 * DTYPE_A (global) DESCA[ DTYPE_ ] The descriptor type.
80 * CTXT_A (global) DESCA[ CTXT_ ] The BLACS context handle, indicating
81 * the NPROW x NPCOL BLACS process grid
82 * A is distributed over. The context
83 * itself is global, but the handle
84 * (the integer value) may vary.
85 * M_A (global) DESCA[ M_ ] The number of rows in the distribu-
86 * ted matrix A, M_A >= 0.
87 * N_A (global) DESCA[ N_ ] The number of columns in the distri-
88 * buted matrix A, N_A >= 0.
89 * IMB_A (global) DESCA[ IMB_ ] The number of rows of the upper left
90 * block of the matrix A, IMB_A > 0.
91 * INB_A (global) DESCA[ INB_ ] The number of columns of the upper
92 * left block of the matrix A,
93 * INB_A > 0.
94 * MB_A (global) DESCA[ MB_ ] The blocking factor used to distri-
95 * bute the last M_A-IMB_A rows of A,
96 * MB_A > 0.
97 * NB_A (global) DESCA[ NB_ ] The blocking factor used to distri-
98 * bute the last N_A-INB_A columns of
99 * A, NB_A > 0.
100 * RSRC_A (global) DESCA[ RSRC_ ] The process row over which the first
101 * row of the matrix A is distributed,
102 * NPROW > RSRC_A >= 0.
103 * CSRC_A (global) DESCA[ CSRC_ ] The process column over which the
104 * first column of A is distributed.
105 * NPCOL > CSRC_A >= 0.
106 * LLD_A (local) DESCA[ LLD_ ] The leading dimension of the local
107 * array storing the local blocks of
108 * the distributed matrix A,
109 * IF( Lc( 1, N_A ) > 0 )
110 * LLD_A >= MAX( 1, Lr( 1, M_A ) )
111 * ELSE
112 * LLD_A >= 1.
113 *
114 * Let K be the number of rows of a matrix A starting at the global in-
115 * dex IA,i.e, A( IA:IA+K-1, : ). Lr( IA, K ) denotes the number of rows
116 * that the process of row coordinate MYROW ( 0 <= MYROW < NPROW ) would
117 * receive if these K rows were distributed over NPROW processes. If K
118 * is the number of columns of a matrix A starting at the global index
119 * JA, i.e, A( :, JA:JA+K-1, : ), Lc( JA, K ) denotes the number of co-
120 * lumns that the process MYCOL ( 0 <= MYCOL < NPCOL ) would receive if
121 * these K columns were distributed over NPCOL processes.
122 *
123 * The values of Lr() and Lc() may be determined via a call to the func-
124 * tion PB_Cnumroc:
125 * Lr( IA, K ) = PB_Cnumroc( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
126 * Lc( JA, K ) = PB_Cnumroc( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
127 *
128 * Arguments
129 * =========
130 *
131 * SIDE (global input) CHARACTER*1
132 * On entry, SIDE specifies whether op( sub( A ) ) multiplies
133 * sub( B ) from the left or right as follows:
134 *
135 * SIDE = 'L' or 'l' sub( B ) := alpha*op( sub( A ) )*sub( B ),
136 *
137 * SIDE = 'R' or 'r' sub( B ) := alpha*sub( B )*op( sub( A ) ).
138 *
139 * UPLO (global input) CHARACTER*1
140 * On entry, UPLO specifies whether the submatrix sub( A ) is
141 * an upper or lower triangular submatrix as follows:
142 *
143 * UPLO = 'U' or 'u' sub( A ) is an upper triangular
144 * submatrix,
145 *
146 * UPLO = 'L' or 'l' sub( A ) is a lower triangular
147 * submatrix.
148 *
149 * TRANSA (global input) CHARACTER*1
150 * On entry, TRANSA specifies the form of op( sub( A ) ) to be
151 * used in the matrix multiplication as follows:
152 *
153 * TRANSA = 'N' or 'n' op( sub( A ) ) = sub( A ),
154 *
155 * TRANSA = 'T' or 't' op( sub( A ) ) = sub( A )',
156 *
157 * TRANSA = 'C' or 'c' op( sub( A ) ) = sub( A )'.
158 *
159 * DIAG (global input) CHARACTER*1
160 * On entry, DIAG specifies whether or not sub( A ) is unit
161 * triangular as follows:
162 *
163 * DIAG = 'U' or 'u' sub( A ) is assumed to be unit trian-
164 * gular,
165 *
166 * DIAG = 'N' or 'n' sub( A ) is not assumed to be unit tri-
167 * angular.
168 *
169 * M (global input) INTEGER
170 * On entry, M specifies the number of rows of the submatrix
171 * sub( B ). M must be at least zero.
172 *
173 * N (global input) INTEGER
174 * On entry, N specifies the number of columns of the submatrix
175 * sub( B ). N must be at least zero.
176 *
177 * ALPHA (global input) DOUBLE PRECISION
178 * On entry, ALPHA specifies the scalar alpha. When ALPHA is
179 * supplied as zero then the local entries of the array B
180 * corresponding to the entries of the submatrix sub( B ) need
181 * not be set on input.
182 *
183 * A (local input) DOUBLE PRECISION array
184 * On entry, A is an array of dimension (LLD_A, Ka), where Ka is
185 * at least Lc( 1, JA+M-1 ) when SIDE = 'L' or 'l' and is at
186 * least Lc( 1, JA+N-1 ) otherwise. Before entry, this array
187 * contains the local entries of the matrix A.
188 * Before entry with UPLO = 'U' or 'u', this array contains the
189 * local entries corresponding to the entries of the upper tri-
190 * angular submatrix sub( A ), and the local entries correspon-
191 * ding to the entries of the strictly lower triangular part of
192 * the submatrix sub( A ) are not referenced.
193 * Before entry with UPLO = 'L' or 'l', this array contains the
194 * local entries corresponding to the entries of the lower tri-
195 * angular submatrix sub( A ), and the local entries correspon-
196 * ding to the entries of the strictly upper triangular part of
197 * the submatrix sub( A ) are not referenced.
198 * Note that when DIAG = 'U' or 'u', the local entries corres-
199 * ponding to the diagonal elements of the submatrix sub( A )
200 * are not referenced either, but are assumed to be unity.
201 *
202 * IA (global input) INTEGER
203 * On entry, IA specifies A's global row index, which points to
204 * the beginning of the submatrix sub( A ).
205 *
206 * JA (global input) INTEGER
207 * On entry, JA specifies A's global column index, which points
208 * to the beginning of the submatrix sub( A ).
209 *
210 * DESCA (global and local input) INTEGER array
211 * On entry, DESCA is an integer array of dimension DLEN_. This
212 * is the array descriptor for the matrix A.
213 *
214 * B (local input/local output) DOUBLE PRECISION array
215 * On entry, B is an array of dimension (LLD_B, Kb), where Kb is
216 * at least Lc( 1, JB+N-1 ). Before entry, this array contains
217 * the local entries of the matrix B.
218 * On exit, the local entries of this array corresponding to the
219 * to the entries of the submatrix sub( B ) are overwritten by
220 * the local entries of the m by n transformed submatrix.
221 *
222 * IB (global input) INTEGER
223 * On entry, IB specifies B's global row index, which points to
224 * the beginning of the submatrix sub( B ).
225 *
226 * JB (global input) INTEGER
227 * On entry, JB specifies B's global column index, which points
228 * to the beginning of the submatrix sub( B ).
229 *
230 * DESCB (global and local input) INTEGER array
231 * On entry, DESCB is an integer array of dimension DLEN_. This
232 * is the array descriptor for the matrix B.
233 *
234 * -- Written on April 1, 1998 by
235 * Antoine Petitet, University of Tennessee, Knoxville 37996, USA.
236 *
237 * ---------------------------------------------------------------------
238 */
239 /*
240 * .. Local Scalars ..
241 */
242 char DiagA, DirB, OpC, OpR, SideOp, TopC, TopR, TranOp, UploA,
243 Var, ctop, ctopsave, rtop, rtopsave;
244 int Ai, Aj, Bi, Bj, ChooseAB, ForceTop, ctxt, info, lside, mycol,
245 myrow, nb, notran, nounit, npcol, nprow, upper;
246 double ABestL, ABestR, Best, tmp1, tmp2, tmp3, tmp4;
247 PBTYP_T * type;
248 /*
249 * .. Local Arrays ..
250 */
251 int Ad[DLEN_], Bd[DLEN_];
252 /* ..
253 * .. Executable Statements ..
254 *
255 */
256 lside = ( ( SideOp = Mupcase( F2C_CHAR( SIDE )[0] ) ) == CLEFT );
257 upper = ( ( UploA = Mupcase( F2C_CHAR( UPLO )[0] ) ) == CUPPER );
258 notran = ( ( TranOp = Mupcase( F2C_CHAR( TRANS )[0] ) ) == CNOTRAN );
259 nounit = ( ( DiagA = Mupcase( F2C_CHAR( DIAG )[0] ) ) == CNOUNIT );
260 PB_CargFtoC( *IA, *JA, DESCA, &Ai, &Aj, Ad );
261 PB_CargFtoC( *IB, *JB, DESCB, &Bi, &Bj, Bd );
262 #ifndef NO_ARGCHK
263 /*
264 * Test the input parameters
265 */
266 Cblacs_gridinfo( ( ctxt = Ad[CTXT_] ), &nprow, &npcol, &myrow, &mycol );
267 if( !( info = ( ( nprow == -1 ) ? -( 1101 + CTXT_ ) : 0 ) ) )
268 {
269 if( ( !lside ) && ( SideOp != CRIGHT ) )
270 {
271 PB_Cwarn( ctxt, __LINE__, "PDTRMM", "Illegal SIDE = %c\n", SideOp );
272 info = -1;
273 }
274 else if( ( !upper ) && ( UploA != CLOWER ) )
275 {
276 PB_Cwarn( ctxt, __LINE__, "PDTRMM", "Illegal UPLO = %c\n", UploA );
277 info = -2;
278 }
279 else if( ( !notran ) && ( TranOp != CTRAN ) && ( TranOp != CCOTRAN ) )
280 {
281 PB_Cwarn( ctxt, __LINE__, "PDTRMM", "Illegal TRANS = %c\n", TranOp );
282 info = -3;
283 }
284 if( ( !nounit ) && ( DiagA != CUNIT ) )
285 {
286 PB_Cwarn( ctxt, __LINE__, "PDTRMM",
287 "Illegal DIAG = %c\n", DiagA );
288 info = -4;
289 }
290 if( lside )
291 PB_Cchkmat( ctxt, "PDTRMM", "A", *M, 5, *M, 5, Ai, Aj, Ad, 11,
292 &info );
293 else
294 PB_Cchkmat( ctxt, "PDTRMM", "A", *N, 6, *N, 6, Ai, Aj, Ad, 11,
295 &info );
296 PB_Cchkmat( ctxt, "PDTRMM", "B", *M, 5, *N, 6, Bi, Bj, Bd, 15,
297 &info );
298 }
299 if( info ) { PB_Cabort( ctxt, "PDTRMM", info ); return; }
300 #endif
301 /*
302 * Quick return if possible
303 */
304 if( *M == 0 || *N == 0 ) return;
305 /*
306 * Get type structure
307 */
308 type = PB_Cdtypeset();
309 /*
310 * And when alpha is zero
311 */
312 if( ALPHA[REAL_PART] == ZERO )
313 {
314 PB_Cplapad( type, ALL, NOCONJG, *M, *N, type->zero, type->zero,
315 ((char *) B), Bi, Bj, Bd );
316 return;
317 }
318 /*
319 * Start the operations
320 */
321 #ifdef NO_ARGCHK
322 Cblacs_gridinfo( ( ctxt = Ad[CTXT_] ), &nprow, &npcol, &myrow, &mycol );
323 #endif
324 /*
325 * Algorithm selection is based on approximation of the communication volume
326 * for distributed and aligned operands.
327 *
328 * ABestR, ABestL : both operands sub( A ) and sub( B ) are communicated
329 * ( N >> M when SIDE is left and M >> N otherwise )
330 * Best : only sub( B ) is communicated
331 * ( M >> N when SIDE is left and N >> M otherwise )
332 */
333 if( lside )
334 {
335 if( notran )
336 {
337 tmp1 = DNROC( *M, Ad[MB_], nprow ); tmp4 = DNROC( *N, Bd[NB_], npcol );
338 ABestR = (double)(*M) *
339 ( ( ( ( Ad[CSRC_] == -1 ) || ( npcol == 1 ) ) ? ZERO : tmp1 / TWO ) +
340 ( ( ( Bd[RSRC_] == -1 ) || ( nprow == 1 ) ) ? ZERO : tmp4 ) );
341 tmp1 = DNROC( *M, Ad[MB_], nprow ); tmp2 = DNROC( *M, Ad[NB_], npcol );
342 tmp4 = DNROC( *M, Bd[MB_], nprow );
343 Best = (double)(*N) *
344 ( CBRATIO * ( npcol == 1 ? ZERO : tmp1 ) +
345 ( nprow == 1 ? ZERO : tmp2 ) + MAX( tmp2, tmp4 ) );
346 ChooseAB = ( ( 1.1 * ABestR ) <= Best );
347 }
348 else
349 {
350 tmp1 = DNROC( *M, Ad[MB_], nprow ); tmp2 = DNROC( *M, Ad[NB_], npcol );
351 tmp4 = DNROC( *N, Bd[NB_], npcol );
352 ABestL = (double)(*M) *
353 ( ( ( ( Ad[CSRC_] == -1 ) || ( npcol == 1 ) ) ? ZERO : tmp1 / TWO ) +
354 CBRATIO *
355 ( ( ( Bd[RSRC_] == -1 ) || ( nprow == 1 ) ) ? ZERO : tmp4 ) );
356 ABestR = (double)(*M) *
357 ( ( ( ( Ad[CSRC_] == -1 ) || ( npcol == 1 ) ) ? ZERO : tmp1 / TWO ) +
358 ( ( ( Bd[RSRC_] == -1 ) || ( nprow == 1 ) ) ? ZERO : tmp4 ) +
359 MAX( tmp2, tmp1 ) / TWO );
360 tmp1 = DNROC( *M, Ad[MB_], nprow ); tmp2 = DNROC( *M, Ad[NB_], npcol );
361 tmp4 = DNROC( *M, Bd[MB_], nprow );
362 Best = (double)(*N) *
363 ( ( ( ( Bd[CSRC_] == -1 ) || ( npcol == 1 ) ) ? ZERO : tmp1 ) +
364 CBRATIO * ( nprow == 1 ? ZERO : tmp2 ) + MAX( tmp2, tmp4 ) );
365 ChooseAB = ( ( ( 1.1 * ABestL ) <= Best ) ||
366 ( ( 1.1 * ABestR ) <= Best ) );
367 }
368 }
369 else
370 {
371 if( notran )
372 {
373 tmp2 = DNROC( *N, Ad[NB_], npcol ); tmp3 = DNROC( *M, Bd[MB_], nprow );
374 ABestR = (double)(*N) *
375 ( ( ( ( Ad[RSRC_] == -1 ) || ( nprow == 1 ) ) ? ZERO : tmp2 / TWO ) +
376 ( ( ( Bd[CSRC_] == -1 ) || ( npcol == 1 ) ) ? ZERO : tmp3 ) );
377 tmp1 = DNROC( *N, Ad[MB_], nprow ); tmp2 = DNROC( *N, Ad[NB_], npcol );
378 tmp3 = DNROC( *N, Bd[NB_], npcol );
379 Best = (double)(*M) *
380 ( CBRATIO * ( nprow == 1 ? ZERO : tmp2 ) +
381 ( npcol == 1 ? ZERO : tmp1 ) + MAX( tmp1, tmp3 ) );
382 ChooseAB = ( ( 1.1 * ABestR ) <= Best );
383 }
384 else
385 {
386 tmp1 = DNROC( *N, Ad[MB_], nprow ); tmp2 = DNROC( *N, Ad[NB_], npcol );
387 tmp3 = DNROC( *M, Bd[MB_], nprow );
388 ABestL = (double)(*N) *
389 ( ( ( ( Ad[RSRC_] == -1 ) || ( nprow == 1 ) ) ? ZERO : tmp2 / TWO ) +
390 CBRATIO *
391 ( ( ( Bd[CSRC_] == -1 ) || ( npcol == 1 ) ) ? ZERO : tmp3 ) );
392 ABestR = (double)(*N) *
393 ( ( ( ( Ad[RSRC_] == -1 ) || ( nprow == 1 ) ) ? ZERO : tmp2 / TWO ) +
394 ( ( ( Bd[CSRC_] == -1 ) || ( npcol == 1 ) ) ? ZERO : tmp3 ) +
395 MAX( tmp2, tmp1 ) / TWO );
396 tmp1 = DNROC( *N, Ad[MB_], nprow ); tmp2 = DNROC( *N, Ad[NB_], npcol );
397 tmp3 = DNROC( *N, Bd[NB_], npcol );
398 Best = (double)(*M) *
399 ( ( ( ( Bd[RSRC_] == -1 ) || ( nprow == 1 ) ) ? ZERO : tmp2 ) +
400 CBRATIO * ( npcol == 1 ? ZERO : tmp1 ) + MAX( tmp1, tmp3 ) );
401 ChooseAB = ( ( ( 1.1 * ABestL ) <= Best ) ||
402 ( ( 1.1 * ABestR ) <= Best ) );
403 }
404 }
405 /*
406 * BLACS topologies are enforced iff M and N are strictly greater than the
407 * logical block size returned by pilaenv_. Otherwise, it is assumed that the
408 * routine calling this routine has already selected an adequate topology.
409 */
410 nb = pilaenv_( &ctxt, C2F_CHAR( &type->type ) );
411 ForceTop = ( ( *M > nb ) && ( *N > nb ) );
412
413 if( ChooseAB )
414 {
415 if( lside )
416 {
417 if( notran )
418 {
419 OpR = CBCAST; OpC = CBCAST; Var = CRIGHT;
420 if( upper ) { TopR = TopC = CTOP_IRING; }
421 else { TopR = TopC = CTOP_DRING; }
422 }
423 else
424 {
425 if( ABestL <= ABestR )
426 {
427 OpR = CBCAST; OpC = CCOMBINE; Var = CLEFT;
428 if( upper ) { TopR = CTOP_DRING; TopC = CTOP_IRING; }
429 else { TopR = CTOP_IRING; TopC = CTOP_DRING; }
430 }
431 else
432 {
433 OpR = CBCAST; OpC = CBCAST; Var = CRIGHT;
434 if( upper ) { TopR = TopC = CTOP_DRING; }
435 else { TopR = TopC = CTOP_IRING; }
436 }
437 }
438 }
439 else
440 {
441 if( notran )
442 {
443 OpR = CBCAST; OpC = CBCAST; Var = CRIGHT;
444 if( upper ) { TopR = TopC = CTOP_DRING; }
445 else { TopR = TopC = CTOP_IRING; }
446 }
447 else
448 {
449 if( ABestL <= ABestR )
450 {
451 OpR = CCOMBINE; OpC = CBCAST; Var = CLEFT;
452 if( upper ) { TopR = CTOP_DRING; TopC = CTOP_IRING; }
453 else { TopR = CTOP_IRING; TopC = CTOP_DRING; }
454 }
455 else
456 {
457 OpR = CBCAST; OpC = CBCAST; Var = CRIGHT;
458 if( upper ) { TopR = TopC = CTOP_IRING; }
459 else { TopR = TopC = CTOP_DRING; }
460 }
461 }
462 }
463
464 rtop = *PB_Ctop( &ctxt, &OpR, ROW, TOP_GET );
465 ctop = *PB_Ctop( &ctxt, &OpC, COLUMN, TOP_GET );
466
467 if( ForceTop )
468 {
469 if( ( rtopsave = rtop ) != TopR )
470 rtop = *PB_Ctop( &ctxt, &OpR, ROW, &TopR );
471 if( ( ctopsave = ctop ) != TopC )
472 ctop = *PB_Ctop( &ctxt, &OpC, COLUMN, &TopC );
473 /*
474 * Remove the next 4 lines when the BLACS combine operations support ring
475 * topologies
476 */
477 if( OpR == CCOMBINE )
478 rtop = *PB_Ctop( &ctxt, &OpR, ROW, TOP_DEFAULT );
479 if( OpC == CCOMBINE )
480 ctop = *PB_Ctop( &ctxt, &OpC, COLUMN, TOP_DEFAULT );
481 }
482
483 PB_CptrmmAB( type, &Var, &SideOp, &UploA, ( notran ? NOTRAN : TRAN ),
484 &DiagA, *M, *N, ((char *)ALPHA), ((char *)A), Ai, Aj, Ad,
485 ((char *)B), Bi, Bj, Bd );
486 }
487 else
488 {
489 if( ( lside && notran ) || ( !( lside ) && !( notran ) ) )
490 {
491 OpR = CCOMBINE; OpC = CBCAST;
492 rtop = *PB_Ctop( &ctxt, &OpR, ROW, TOP_GET );
493 ctop = *PB_Ctop( &ctxt, &OpC, COLUMN, TOP_GET );
494
495 if( ForceTop )
496 {
497 rtopsave = rtop;
498 ctopsave = ctop;
499 /*
500 * No clear winner for the ring topologies, so that if a ring topology is
501 * already selected, keep it.
502 */
503 if( ( rtop != CTOP_DRING ) && ( rtop != CTOP_IRING ) &&
504 ( rtop != CTOP_SRING ) )
505 rtop = *PB_Ctop( &ctxt, &OpR, ROW, TOP_SRING );
506 ctop = *PB_Ctop( &ctxt, &OpC, COLUMN, TOP_DEFAULT );
507 /*
508 * Remove the next line when the BLACS combine operations support ring
509 * topologies
510 */
511 rtop = *PB_Ctop( &ctxt, &OpR, ROW, TOP_DEFAULT );
512 }
513 }
514 else
515 {
516 OpR = CBCAST; OpC = CCOMBINE;
517 rtop = *PB_Ctop( &ctxt, &OpR, ROW, TOP_GET );
518 ctop = *PB_Ctop( &ctxt, &OpC, COLUMN, TOP_GET );
519
520 if( ForceTop )
521 {
522 rtopsave = rtop;
523 ctopsave = ctop;
524 /*
525 * No clear winner for the ring topologies, so that if a ring topology is
526 * already selected, keep it.
527 */
528 if( ( ctop != CTOP_DRING ) && ( ctop != CTOP_IRING ) &&
529 ( ctop != CTOP_SRING ) )
530 ctop = *PB_Ctop( &ctxt, &OpC, COLUMN, TOP_SRING );
531 rtop = *PB_Ctop( &ctxt, &OpR, ROW, TOP_DEFAULT );
532 /*
533 * Remove the next line when the BLACS combine operations support ring
534 * topologies
535 */
536 ctop = *PB_Ctop( &ctxt, &OpC, COLUMN, TOP_DEFAULT );
537 }
538 }
539
540 if( lside )
541 DirB = ( rtop == CTOP_DRING ? CBACKWARD : CFORWARD );
542 else
543 DirB = ( ctop == CTOP_DRING ? CBACKWARD : CFORWARD );
544
545 PB_CptrmmB( type, &DirB, &SideOp, &UploA, ( notran ? NOTRAN : TRAN ),
546 &DiagA, *M, *N, ((char *)ALPHA), ((char *)A), Ai, Aj, Ad,
547 ((char *)B), Bi, Bj, Bd );
548 }
549 /*
550 * Restore the BLACS topologies when necessary.
551 */
552 if( ForceTop )
553 {
554 rtopsave = *PB_Ctop( &ctxt, &OpR, ROW, &rtopsave );
555 ctopsave = *PB_Ctop( &ctxt, &OpC, COLUMN, &ctopsave );
556 }
557 /*
558 * End of PDTRMM
559 */
560 }
561