1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870
|
/* ---------------------------------------------------------------------
*
* -- PBLAS auxiliary routine (version 2.0) --
* University of Tennessee, Knoxville, Oak Ridge National Laboratory,
* and University of California, Berkeley.
* April 1, 1998
*
* ---------------------------------------------------------------------
*/
/*
* Include files
*/
#include "../pblas.h"
#include "../PBpblas.h"
#include "../PBtools.h"
#include "../PBblacs.h"
#include "../PBblas.h"
#ifdef __STDC__
void PB_CInOutV( PBTYP_T * TYPE, char * ROWCOL, int M, int N, int * DESCA,
int K, char * BETA,
char * Y, int IY, int JY, int * DESCY, char * YROC,
char * * TBETA, char * * YAPTR, int * DYA,
int * YAFREE, int * YASUM, int * YAPBY )
#else
void PB_CInOutV( TYPE, ROWCOL, M, N, DESCA, K,
BETA, Y, IY, JY, DESCY, YROC,
TBETA, YAPTR, DYA, YAFREE, YASUM, YAPBY )
/*
* .. Scalar Arguments ..
*/
char * BETA, * ROWCOL, * * TBETA, * YROC;
int * YAPBY, * YAFREE, IY, JY, K, M, N, * YASUM;
PBTYP_T * TYPE;
/*
* .. Array Arguments ..
*/
int * DESCA, * DESCY, * DYA;
char * Y, * * YAPTR;
#endif
{
/*
* Purpose
* =======
*
* PB_CInOutV returns a pointer to an array that contains a one-dimen-
* sional input/output subvector which is replicated over the rows or
* columns of a submatrix described by DESCA. A subvector is specified
* on input to this routine that is reused whenever possible. On return,
* the subvector is specified by a pointer to some data, a descriptor
* array describing its layout, a logical value indicating if this local
* piece of data has been dynamically allocated by this function, a lo-
* gical value specifying if sum reduction should occur, and finally a
* logical value specifying if it is necessary to copy back the alloca-
* ted data to the original data. This routine is specifically designed
* for traditional Level 2 like PBLAS operations using an input/output
* vector such as PxGEMV, PxSYMV ...
*
* Notes
* =====
*
* A description vector is associated with each 2D block-cyclicly dis-
* tributed matrix. This vector stores the information required to
* establish the mapping between a matrix entry and its corresponding
* process and memory location.
*
* In the following comments, the character _ should be read as
* "of the distributed matrix". Let A be a generic term for any 2D
* block cyclicly distributed matrix. Its description vector is DESC_A:
*
* NOTATION STORED IN EXPLANATION
* ---------------- --------------- ------------------------------------
* DTYPE_A (global) DESCA[ DTYPE_ ] The descriptor type.
* CTXT_A (global) DESCA[ CTXT_ ] The BLACS context handle, indicating
* the NPROW x NPCOL BLACS process grid
* A is distributed over. The context
* itself is global, but the handle
* (the integer value) may vary.
* M_A (global) DESCA[ M_ ] The number of rows in the distribu-
* ted matrix A, M_A >= 0.
* N_A (global) DESCA[ N_ ] The number of columns in the distri-
* buted matrix A, N_A >= 0.
* IMB_A (global) DESCA[ IMB_ ] The number of rows of the upper left
* block of the matrix A, IMB_A > 0.
* INB_A (global) DESCA[ INB_ ] The number of columns of the upper
* left block of the matrix A,
* INB_A > 0.
* MB_A (global) DESCA[ MB_ ] The blocking factor used to distri-
* bute the last M_A-IMB_A rows of A,
* MB_A > 0.
* NB_A (global) DESCA[ NB_ ] The blocking factor used to distri-
* bute the last N_A-INB_A columns of
* A, NB_A > 0.
* RSRC_A (global) DESCA[ RSRC_ ] The process row over which the first
* row of the matrix A is distributed,
* NPROW > RSRC_A >= 0.
* CSRC_A (global) DESCA[ CSRC_ ] The process column over which the
* first column of A is distributed.
* NPCOL > CSRC_A >= 0.
* LLD_A (local) DESCA[ LLD_ ] The leading dimension of the local
* array storing the local blocks of
* the distributed matrix A,
* IF( Lc( 1, N_A ) > 0 )
* LLD_A >= MAX( 1, Lr( 1, M_A ) )
* ELSE
* LLD_A >= 1.
*
* Let K be the number of rows of a matrix A starting at the global in-
* dex IA,i.e, A( IA:IA+K-1, : ). Lr( IA, K ) denotes the number of rows
* that the process of row coordinate MYROW ( 0 <= MYROW < NPROW ) would
* receive if these K rows were distributed over NPROW processes. If K
* is the number of columns of a matrix A starting at the global index
* JA, i.e, A( :, JA:JA+K-1, : ), Lc( JA, K ) denotes the number of co-
* lumns that the process MYCOL ( 0 <= MYCOL < NPCOL ) would receive if
* these K columns were distributed over NPCOL processes.
*
* The values of Lr() and Lc() may be determined via a call to the func-
* tion PB_Cnumroc:
* Lr( IA, K ) = PB_Cnumroc( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
* Lc( JA, K ) = PB_Cnumroc( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
* Arguments
* =========
*
* TYPE (local input) pointer to a PBTYP_T structure
* On entry, TYPE is a pointer to a structure of type PBTYP_T,
* that contains type information (See pblas.h).
*
* ROWCOL (global input) pointer to CHAR
* On entry, ROWCOL specifies if this routine should return a
* row or column subvector replicated over the underlying subma-
* trix as follows:
* = 'R' or 'r': A row subvector is returned,
* = 'C' or 'c': A column subvector is returned.
*
* M (global input) INTEGER
* On entry, M specifies the number of rows of the underlying
* submatrix described by DESCA. M must be at least zero.
*
* N (global input) INTEGER
* On entry, N specifies the number of columns of the underlying
* submatrix described by DESCA. N must be at least zero.
*
* DESCA (global and local input/output) INTEGER array
* On entry, DESCA is an integer array of dimension DLEN_. This
* is the array descriptor for the matrix A. EXCEPTIONALLY, THIS
* INTERNAL ROUTINE MAY MODIFY DESCA IN ORDER TO MINIMIZE THE
* AMOUNT OF DATA TO BE MOVED FOR THE VECTOR Y. SEE PxGEMV FOR
* AN EXAMPLE.
*
* K (global input) INTEGER
* On entry, K specifies the length of the non-distributed di-
* mension of the subvector sub( Y ). K must be at least zero.
*
* BETA (global input) pointer to CHAR
* On entry, BETA is a scalar the input subvector sub( Y ) must
* be scaled by.
*
* Y (local input) pointer to CHAR
* On entry, Y is an array of dimension (LLD_Y, Ky), where LLD_Y
* is at least MAX( 1, Lr( K, IY ) ) when YROC is 'R' or 'r'
* and MAX( 1, Lr( 1, IY+Ly-1 ) ) otherwise, and, Ky is at
* least Lc( 1, JY+Ly-1 ) when YROC is 'R' or 'r' and
* Lc( K, JY ) otherwise. Ly is N when ROWCOL is 'R' or 'r' and
* M otherwise. Before entry, this array contains the local
* entries of the matrix Y.
*
* IY (global input) INTEGER
* On entry, IY specifies Y's global row index, which points to
* the beginning of the submatrix sub( Y ).
*
* JY (global input) INTEGER
* On entry, JY specifies Y's global column index, which points
* to the beginning of the submatrix sub( Y ).
*
* DESCY (global and local input) INTEGER array
* On entry, DESCY is an integer array of dimension DLEN_. This
* is the array descriptor for the matrix Y.
*
* YROC (global input) pointer to CHAR
* On entry, YROC specifies the orientation of the subvector
* sub( Y ). When YROC is 'R' or 'r', sub( Y ) is a row vector,
* and a column vector otherwise.
*
* TBETA (local output) pointer to pointer to CHAR
* On exit, * TBETA is a scalar to be used locally to scale the
* data pointed to by * YAPTR, in order to obtain the correct
* result in the original data sub( Y ).
*
* YAPTR (local output) pointer to pointer to CHAR
* On exit, * YAPTR is an array containing the same data as the
* subvector sub( Y ) which is replicated over the rows or co-
* lumns of the underlying matrix as specified by ROWCOL and
* DESCA.
*
* DYA (global and local output) INTEGER array
* On exit, DYA is a descriptor array of dimension DLEN_ descri-
* bing the data layout of the data pointed to by * YAPTR.
*
* YAFREE (local output) INTEGER
* On exit, YAFREE specifies if it was possible to reuse the
* subvector sub( Y ), i.e., if some dynamic memory was alloca-
* ted for the data pointed to by * YAPTR or not. When YAFREE is
* zero, no dynamic memory was allocated. Otherwise, some dyna-
* mic memory was allocated by this function that one MUST re-
* lease as soon as possible.
*
* YASUM (global output) INTEGER
* On exit, YASUM specifies if a global sum reduction should be
* performed to obtain the correct sub( Y ). When YASUM is zero,
* no reduction is to be performed, otherwise reduction should
* occur.
*
* YAPBY (global output) INTEGER
* On exit, YAPBY specifies if the data pointed to by * YAPTR
* must be move back onto sub( Y ) to obtain the correct result.
* When YAPBY is zero, no supplementary data movement is neces-
* sary, otherwise a data redistribution should occur.
*
* -- Written on April 1, 1998 by
* Antoine Petitet, University of Tennessee, Knoxville 37996, USA.
*
* ---------------------------------------------------------------------
*/
/*
* .. Local Scalars ..
*/
int Acol, Aimb, Ainb, AisD, AisR, Amb, Amp, Anb, Anq, Arow, Ycol,
Yii, Yimb, Yimb1, Yinb, Yinb1, YisD, YisR, YisRow, Yjj, Yld,
Ymb, Ymp, Ynb, Ynq, Yrow, ctxt, izero=0, nprow, myrow, npcol,
mycol;
/* ..
* .. Executable Statements ..
*
*/
/*
* Initialize the output parameters to a default value
*/
*YAFREE = 0;
*YASUM = 0;
*YAPBY = 0;
*YAPTR = NULL;
*TBETA = BETA;
/*
* Quick return if possible
*/
if( ( M <= 0 ) || ( N <= 0 ) || ( K <= 0 ) )
{
if( Mupcase( ROWCOL[0] ) == CROW )
{
PB_Cdescset( DYA, K, N, 1, DESCA[INB_], 1, DESCA[NB_], DESCA[RSRC_],
DESCA[CSRC_], DESCA[CTXT_], 1 );
}
else
{
PB_Cdescset( DYA, M, K, DESCA[IMB_], 1, DESCA[MB_], 1, DESCA[RSRC_],
DESCA[CSRC_], DESCA[CTXT_], DESCA[LLD_] );
}
return;
}
/*
* Retrieve process grid information
*/
Cblacs_gridinfo( ( ctxt = DESCY[CTXT_] ), &nprow, &npcol, &myrow, &mycol );
/*
* Retrieve sub( Y )'s local information: Yii, Yjj, Yrow, Ycol
*/
Minfog2l( IY, JY, DESCY, nprow, npcol, myrow, mycol, Yii, Yjj, Yrow, Ycol );
/*
* Is sub( Y ) distributed or not, replicated or not ?
*/
if( ( YisRow = ( Mupcase( YROC[0] ) == CROW ) ) != 0 )
{
YisD = ( ( Ycol >= 0 ) && ( npcol > 1 ) );
YisR = ( ( Yrow == -1 ) || ( nprow == 1 ) );
}
else
{
YisD = ( ( Yrow >= 0 ) && ( nprow > 1 ) );
YisR = ( ( Ycol == -1 ) || ( npcol == 1 ) );
}
Aimb = DESCA[IMB_ ]; Ainb = DESCA[INB_ ];
Amb = DESCA[MB_ ]; Anb = DESCA[NB_ ];
Arow = DESCA[RSRC_]; Acol = DESCA[CSRC_];
if( Mupcase( ROWCOL[0] ) == CROW )
{
/*
* Want a row vector
*/
AisR = ( ( Arow < 0 ) || ( nprow == 1 ) );
if( YisRow )
{
/*
* It is possible to reuse sub( Y ) iff sub( Y ) is already a row vector.
*/
AisD = ( ( Acol >= 0 ) && ( npcol > 1 ) );
Yinb = DESCY[INB_]; Ynb = DESCY[NB_];
Mfirstnb( Yinb1, N, JY, Yinb, Ynb );
/*
* sub( Y ) is aligned with A (reuse condition) iff both operands are not
* distributed, or both of them are distributed and start in the same process
* column and either N is smaller than the first blocksize of sub( Y ) and A,
* or their column blocking factors match.
*/
if( ( !AisD && !YisD ) ||
( ( AisD && YisD ) &&
( ( Acol == Ycol ) &&
( ( ( Ainb >= N ) && ( Yinb1 >= N ) ) ||
( ( Ainb == Yinb1 ) && ( Anb == Ynb ) ) ) ) ) )
{
Mnumroc( Ynq, N, 0, Yinb1, Ynb, mycol, Ycol, npcol );
Ymp = ( YisR ? K : ( ( myrow == Yrow ) ? K : 0 ) );
if( YisR )
{
/*
* If sub( Y ) is replicated, there is no need to move sub( Y ) after the
* operation (*YAPBY = 0), and it can be reused.
*/
*YAPBY = 0;
Yld = DESCY[ LLD_ ];
if( Ynq > 0 )
*YAPTR = Mptr( Y, Yii, Yjj, Yld, TYPE->size );
if( AisR )
{
/*
* If A is replicated as well, use BETA in every process row, and do not combine
* the local results.
*/
*TBETA = BETA;
*YASUM = 0;
}
else
{
/*
* Otherwise, use BETA in process row Arow and zero elsewhere. Reduce the local
* result if there is more than one row in the process grid.
*/
*TBETA = ( ( myrow == Arow ) ? BETA : TYPE->zero );
*YASUM = ( nprow > 1 );
/*
* If some process rows do not own any entries of A, better set sub( Y ) to zero
* in those processes.
*/
Mnumroc( Amp, M, 0, Aimb, Amb, myrow, Arow, nprow );
if( Amp <= 0 )
TYPE->Ftzscal( C2F_CHAR( ALL ), &K, &Ynq, &izero, *TBETA,
*YAPTR, &Yld );
}
}
else
{
/*
* sub( Y ) is not replicated, the descriptor of A may need to be modified ...
*/
if( AisR )
{
/*
* If A is replicated, use only the copy in the process row where sub( Y )
* resides -> modify DESCA !!!
*/
*TBETA = BETA;
*YASUM = 0;
*YAPBY = 0;
Yld = DESCY[ LLD_ ];
DESCA[ IMB_ ] = M;
DESCA[ RSRC_ ] = Yrow;
if( ( Ynq > 0 ) && ( Ymp > 0 ) )
*YAPTR = Mptr( Y, Yii, Yjj, Yld, TYPE->size );
}
else
{
if( Mspan( M, 0, Aimb, Amb, Arow, nprow ) )
{
/*
* Otherwise, A is not replicated, let assume in addition that it spans more
* than one process row
*/
*YASUM = ( nprow > 1 );
*YAPBY = 0;
if( myrow == Yrow )
{
/*
* Reuse sub( Y ). If there is no entries of A in the process row where sub( Y )
* resides, better scale it by BETA immediately.
*/
*TBETA = BETA;
Yld = DESCY[ LLD_ ];
if( Ynq > 0 )
{
*YAPTR = Mptr( Y, Yii, Yjj, Yld, TYPE->size );
Mnumroc( Amp, M, 0, Aimb, Amb, myrow, Arow, nprow );
if( Amp <= 0 )
TYPE->Ftzscal( C2F_CHAR( ALL ), &K, &Ynq, &izero,
*TBETA, *YAPTR, &Yld );
}
}
else
{
/*
* Allocate space in the other process rows and initialize to zero.
*/
*TBETA = TYPE->zero;
Yld = MAX( 1, K );
if( Ynq > 0 )
{
*YAPTR = PB_Cmalloc( K * Ynq * TYPE->size );
*YAFREE = 1;
TYPE->Ftzpad( C2F_CHAR( ALL ), C2F_CHAR( NOCONJG ),
&K, &Ynq, &izero, *TBETA, *TBETA,
*YAPTR, &Yld );
}
}
}
else
{
/*
* A spans only one process row
*/
if( Yrow == Arow )
{
/*
* A and sub( Y ) resides in the same process row
*/
*TBETA = BETA;
*YASUM = 0;
*YAPBY = 0;
Yld = DESCY[ LLD_ ];
if( ( myrow == Yrow ) && ( Ynq > 0 ) )
*YAPTR = Mptr( Y, Yii, Yjj, Yld, TYPE->size );
}
else
{
/*
* If sub( Y ) resides in another process row, then allocate zero-data in
* process row where A resides, and set *YAPBY to 1, so that this data will be
* added (moved) after the local operation has been performed.
*/
*TBETA = TYPE->zero;
*YASUM = 0;
*YAPBY = 1;
Yrow = Arow;
Yld = MAX( 1, K );
if( myrow == Arow )
{
if( Ynq > 0 )
{
*YAPTR = PB_Cmalloc( K * Ynq * TYPE->size );
*YAFREE = 1;
TYPE->Ftzpad( C2F_CHAR( ALL ),
C2F_CHAR( NOCONJG ), &K, &Ynq,
&izero, *TBETA, *TBETA, *YAPTR,
&Yld );
}
}
}
}
}
}
/*
* Describe the resulting operand. Note that when reduction should occur, Yrow
* contains the destination row. Assuming every process row needs the result,
* Yrow is then -1.
*/
MDescSet( DYA, K, N, K, Yinb1, 1, Ynb, Yrow, Ycol, ctxt, Yld );
return;
}
}
/*
* sub( Y ) cannot be reused, set TBETA to zero for the local operation, and
* force YAPBY to 1 for the later update of sub( Y ).
*/
*TBETA = TYPE->zero;
*YAPBY = 1;
Mnumroc( Anq, N, 0, Ainb, Anb, mycol, Acol, npcol );
Yld = MAX( 1, K );
if( YisR )
{
/*
* If sub( Y ) is replicated, allocate space in every process row owning some
* columns of A and initialize it to zero. There may be some wasted space
* (suppose A was residing in just one row), however, it is hoped that moving
* back this data to sub( Y ) will then be cheaper ...
*/
*YASUM = ( AisR ? 0 : ( nprow > 1 ) );
Arow = -1;
if( Anq > 0 )
{
*YAPTR = PB_Cmalloc( K * Anq * TYPE->size );
*YAFREE = 1;
TYPE->Ftzpad( C2F_CHAR( ALL ), C2F_CHAR( NOCONJG ), &K, &Anq,
&izero, *TBETA, *TBETA, *YAPTR, &Yld );
}
}
else
{
/*
* sub( Y ) resides within only one process row
*/
if( AisR )
{
/*
* If A is replicated, then modify sub( A ) so that only one process row will
* compute the result before moving it back to sub( Y ).
*/
*YASUM = 0;
DESCA[ IMB_ ] = M;
if( YisRow )
{
/*
* Choose different process row than Yrow for better performance (more links)
* of the later move-back phase.
*/
DESCA[RSRC_] = Arow = MModSub1( Yrow, nprow );
}
else
{
DESCA[RSRC_] = Arow = 0;
}
if( ( myrow == Arow ) && ( Anq > 0 ) )
{
*YAPTR = PB_Cmalloc( K * Anq * TYPE->size );
*YAFREE = 1;
TYPE->Ftzpad( C2F_CHAR( ALL ), C2F_CHAR( NOCONJG ), &K, &Anq,
&izero, *TBETA, *TBETA, *YAPTR, &Yld );
}
}
else
{
if( Mspan( M, 0, Aimb, Amb, Arow, nprow ) )
{
/*
* If A is not replicated, and spans more than just one process row, then
* allocate space in every process row and zero it.
*/
*YASUM = ( nprow > 1 );
if( Anq > 0 )
{
*YAPTR = PB_Cmalloc( K * Anq * TYPE->size );
*YAFREE = 1;
TYPE->Ftzpad( C2F_CHAR( ALL ), C2F_CHAR( NOCONJG ), &K, &Anq,
&izero, *TBETA, *TBETA, *YAPTR, &K );
}
}
else
{
/*
* If A is not replicated, and spans only one process row, then allocate space
* within that process row and zero it.
*/
*YASUM = 0;
if( ( myrow == Arow ) && ( Anq > 0 ) )
{
*YAPTR = PB_Cmalloc( K * Anq * TYPE->size );
*YAFREE = 1;
TYPE->Ftzpad( C2F_CHAR( ALL ), C2F_CHAR( NOCONJG ), &K, &Anq,
&izero, *TBETA, *TBETA, *YAPTR, &K );
}
}
}
}
/*
* Describe the resulting operand. Note that when reduction should occur, Arow
* contains the destination row. Assuming every process row needs the result,
* Arow is then -1.
*/
MDescSet( DYA, K, N, K, Ainb, 1, Anb, Arow, Acol, ctxt, Yld );
}
else
{
/*
* Want a column vector
*/
AisR = ( ( Acol < 0 ) || ( npcol == 1 ) );
if( !YisRow )
{
/*
* It is possible to reuse sub( Y ) iff sub( Y ) is already a column vector.
*/
AisD = ( ( Arow >= 0 ) && ( nprow > 1 ) );
Yimb = DESCY[IMB_]; Ymb = DESCY[MB_];
Mfirstnb( Yimb1, M, IY, Yimb, Ymb );
/*
* sub( Y ) is aligned with A (reuse condition) iff both operands are not
* distributed, or both of them are distributed and start in the same process
* row and either M is smaller than the first blocksize of sub( Y ) and A, or
* their row blocking factors match.
*/
if( ( !AisD && !YisD ) ||
( ( AisD && YisD ) &&
( ( Arow == Yrow ) &&
( ( ( Aimb >= M ) && ( Yimb1 >= M ) ) ||
( ( Aimb == Yimb1 ) && ( Amb == Ymb ) ) ) ) ) )
{
Mnumroc( Ymp, M, 0, Yimb1, Ymb, myrow, Yrow, nprow );
Ynq = ( YisR ? K : ( ( mycol == Ycol ) ? K : 0 ) );
if( YisR )
{
/*
* If sub( Y ) is replicated, there is no need to move sub( Y ) after the
* operation (*YAPBY = 0), and it can be reused.
*/
*YAPBY = 0;
Yld = DESCY[ LLD_ ];
if( Ymp > 0 )
*YAPTR = Mptr( Y, Yii, Yjj, Yld, TYPE->size );
if( AisR )
{
/*
* If A is replicated as well, use BETA in every process column, and do not
* combine the local results.
*/
*TBETA = BETA;
*YASUM = 0;
}
else
{
/*
* Otherwise, use BETA in process column Acol and zero elsewhere. Reduce the
* local result if there is more than one column in the process grid.
*/
*TBETA = ( ( mycol == Acol ) ? BETA : TYPE->zero );
*YASUM = ( npcol > 1 );
/*
* If some process columns do not own any entries of A, better set sub( Y ) to
* zero in those processes.
*/
Mnumroc( Anq, N, 0, Ainb, Anb, mycol, Acol, npcol );
if( Anq <= 0 )
TYPE->Ftzscal( C2F_CHAR( ALL ), &Ymp, &K, &izero, *TBETA,
*YAPTR, &Yld );
}
}
else
{
/*
* sub( Y ) is not replicated, the descriptor of A may need to be modified ...
*/
if( AisR )
{
/*
* If A is replicated, use only the copy in the process column where sub( Y )
* resides -> modify DESCA !!!
*/
*TBETA = BETA;
*YASUM = 0;
*YAPBY = 0;
Yld = DESCY[ LLD_ ];
DESCA[ INB_ ] = N;
DESCA[ CSRC_ ] = Ycol;
if( ( Ymp > 0 ) && ( Ynq > 0 ) )
*YAPTR = Mptr( Y, Yii, Yjj, Yld, TYPE->size );
}
else
{
if( Mspan( N, 0, Ainb, Anb, Acol, npcol ) )
{
/*
* Otherwise, A is not replicated, let assume in addition that it spans more
* than one process column
*/
*YASUM = ( npcol > 1 );
*YAPBY = 0;
if( mycol == Ycol )
{
/*
* Reuse sub( Y ). If there is no entries of A in the process column where
* sub( Y ) resides, better scale it by BETA immediately.
*/
*TBETA = BETA;
Yld = DESCY[ LLD_ ];
if( Ymp > 0 )
{
*YAPTR = Mptr( Y, Yii, Yjj, Yld, TYPE->size );
Mnumroc( Anq, N, 0, Ainb, Anb, mycol, Acol, npcol );
if( Anq <= 0 )
TYPE->Ftzscal( C2F_CHAR( ALL ), &Ymp, &K, &izero,
*TBETA, *YAPTR, &Yld );
}
}
else
{
/*
* Allocate space in the other process columns and initialize to zero.
*/
*TBETA = TYPE->zero;
Yld = MAX( 1, Ymp );
if( Ymp > 0 )
{
*YAPTR = PB_Cmalloc( Ymp * K * TYPE->size );
*YAFREE = 1;
TYPE->Ftzpad( C2F_CHAR( ALL ), C2F_CHAR( NOCONJG ),
&Ymp, &K, &izero, *TBETA, *TBETA,
*YAPTR, &Yld );
}
}
}
else
{
/*
* A spans only one process column
*/
if( Ycol == Acol )
{
/*
* A and sub( Y ) resides in the same process column
*/
*TBETA = BETA;
*YASUM = 0;
*YAPBY = 0;
Yld = DESCY[ LLD_ ];
if( ( mycol == Ycol ) && ( Ymp > 0 ) )
*YAPTR = Mptr( Y, Yii, Yjj, Yld, TYPE->size );
}
else
{
/*
* If sub( Y ) resides in another process column, then allocate zero-data in
* process column where A resides, and set *YAPBY to 1, so that this data will
* be added (moved) after the local operation has been performed.
*/
*TBETA = TYPE->zero;
*YASUM = 0;
*YAPBY = 1;
Ycol = Acol;
Yld = MAX( 1, Ymp ) ;
if( mycol == Acol )
{
if( Ymp > 0 )
{
*YAPTR = PB_Cmalloc( Ymp * K * TYPE->size );
*YAFREE = 1;
TYPE->Ftzpad( C2F_CHAR( ALL ),
C2F_CHAR( NOCONJG ), &Ymp, &K,
&izero, *TBETA, *TBETA, *YAPTR,
&Yld );
}
}
}
}
}
}
/*
* Describe the resulting operand. Note that when reduction should occur, Ycol
* contains the destination column. Assuming every process column needs the
* result, Ycol is then -1.
*/
MDescSet( DYA, M, K, Yimb1, K, Ymb, 1, Yrow, Ycol, ctxt, Yld );
return;
}
}
/*
* sub( Y ) cannot be reused, set TBETA to zero for the local operation, and
* force YAPBY to 1 for the later update of sub( Y ).
*/
*TBETA = TYPE->zero;
*YAPBY = 1;
Mnumroc( Amp, M, 0, Aimb, Amb, myrow, Arow, nprow );
Yld = MAX( 1, Amp );
if( YisR )
{
/*
* If sub( Y ) is replicated, allocate space in every process column owning some
* rows of A and initialize it to zero. There may be some wasted space (suppose
* A was residing in just one column), however, it is hoped that moving back
* this data to sub( Y ) will then be cheaper ...
*/
*YASUM = ( AisR ? 0 : ( npcol > 1 ) );
Acol = -1;
if( Amp > 0 )
{
*YAPTR = PB_Cmalloc( Amp * K * TYPE->size );
*YAFREE = 1;
TYPE->Ftzpad( C2F_CHAR( ALL ), C2F_CHAR( NOCONJG ), &Amp, &K,
&izero, *TBETA, *TBETA, *YAPTR, &Yld );
}
}
else
{
/*
* sub( Y ) resides within only one process column
*/
if( AisR )
{
/*
* If A is replicated, then modify sub( A ) so that only one process column will
* compute the result before moving it back to sub( Y ).
*/
*YASUM = 0;
DESCA[ INB_ ] = N;
if( YisRow )
{
DESCA[ CSRC_ ] = Acol = 0;
}
else
{
/*
* Choose different process column than Ycol for better performance (more links)
* of the later move-back phase.
*/
DESCA[ CSRC_ ] = Acol = MModSub1( Ycol, npcol );
}
if( ( mycol == Acol ) && ( Amp > 0 ) )
{
*YAPTR = PB_Cmalloc( Amp * K * TYPE->size );
*YAFREE = 1;
TYPE->Ftzpad( C2F_CHAR( ALL ), C2F_CHAR( NOCONJG ), &Amp, &K,
&izero, *TBETA, *TBETA, *YAPTR, &Yld );
}
}
else
{
if( Mspan( N, 0, Ainb, Anb, Acol, npcol ) )
{
/*
* If A is not replicated, and spans more than just one process column, then
* allocate space in every process column and zero it.
*/
*YASUM = ( npcol > 1 );
if( Amp > 0 )
{
*YAPTR = PB_Cmalloc( Amp * K * TYPE->size );
*YAFREE = 1;
TYPE->Ftzpad( C2F_CHAR( ALL ), C2F_CHAR( NOCONJG ), &Amp, &K,
&izero, *TBETA, *TBETA, *YAPTR, &Yld );
}
}
else
{
/*
* If A is not replicated, and spans only one process column, then allocate
* space within that process column and zero it.
*/
*YASUM = 0;
if( ( mycol == Acol ) && ( Amp > 0 ) )
{
*YAPTR = PB_Cmalloc( Amp * K * TYPE->size );
*YAFREE = 1;
TYPE->Ftzpad( C2F_CHAR( ALL ), C2F_CHAR( NOCONJG ), &Amp, &K,
&izero, *TBETA, *TBETA, *YAPTR, &Yld );
}
}
}
}
/*
* Describe the resulting operand. Note that when reduction should occur, Acol
* contains the destination column. Assuming every process column needs the
* result, Acol is then -1.
*/
MDescSet( DYA, M, K, Aimb, K, Amb, 1, Arow, Acol, ctxt, Yld );
}
/*
* End of PB_CInOutV
*/
}
|