/* --------------------------------------------------------------------- * * -- 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_Cpsyr2kAC( PBTYP_T * TYPE, char * DIRECAB, char * CONJUG, char * UPLO, char * TRANS, int N, int K, char * ALPHA, char * A, int IA, int JA, int * DESCA, char * B, int IB, int JB, int * DESCB, char * BETA, char * C, int IC, int JC, int * DESCC ) #else void PB_Cpsyr2kAC( TYPE, DIRECAB, CONJUG, UPLO, TRANS, N, K, ALPHA, A, IA, JA, DESCA, B, IB, JB, DESCB, BETA, C, IC, JC, DESCC ) /* * .. Scalar Arguments .. */ char * CONJUG, * DIRECAB, * TRANS, * UPLO; int IA, IB, IC, JA, JB, JC, K, N; char * ALPHA, * BETA; PBTYP_T * TYPE; /* * .. Array Arguments .. */ int * DESCA, * DESCB, * DESCC; char * A, * B, * C; #endif { /* * Purpose * ======= * * PB_Cpsyr2kAC performs one of the following symmetric or Hermitian * rank 2k operations * * sub( C ) := alpha*sub( A )*sub( B )' + alpha*sub( B )*sub( A )' + * beta*sub( C ), * or * sub( C ) := alpha*sub( A )*conjg( sub( B ) )' + * conjg( alpha )*sub( B )*conjg( sub( A ) )' + * beta*sub( C ), * or * sub( C ) := alpha*sub( A )'*sub( B ) + alpha*sub( B )'*sub( A ) + * beta*sub( C ), * or * sub( C ) := alpha*conjg( sub( A )' )*sub( B ) + * conjg( alpha )*conjg( sub( B )' )*sub( A ) + * beta*sub( C ), * * where * * sub( C ) denotes C(IC:IC+N-1,JC:JC+N-1), * * sub( A ) denotes A(IA:IA+N-1,JA:JA+K-1) if TRANS = 'N', * A(IA:IA+K-1,JA:JA+N-1) otherwise, and, * * sub( B ) denotes B(IB:IB+N-1,JB:JB+K-1) if TRANS = 'N', * B(IB:IB+K-1,JB:JB+N-1) otherwise. * * Alpha and beta are scalars, sub( C ) is an n by n symmetric or * Hermitian submatrix and sub( A ) and sub( B ) are n by k submatrices * in the first case and k by n submatrices in the second case. * * This is the outer-product algorithm using the logical aggregation * blocking technique. * * 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). * * DIRECAB (global input) pointer to CHAR * On entry, DIRECAB specifies the direction in which the rows * or columns of sub( A ), sub( B ) and sub( C ) should be * looped over as follows: * DIRECAB = 'F' or 'f' forward or increasing, * DIRECAB = 'B' or 'b' backward or decreasing. * * CONJUG (global input) pointer to CHAR * On entry, CONJUG specifies whether sub( C ) is a symmetric or * Hermitian submatrix operand as follows: * CONJUG = 'N' or 'n' sub( C ) is symmetric, * CONJUG = 'Z' or 'z' sub( C ) is Hermitian. * * UPLO (global input) pointer to CHAR * On entry, UPLO specifies whether the local pieces of * the array C containing the upper or lower triangular part * of the submatrix sub( C ) are to be referenced as follows: * UPLO = 'U' or 'u' Only the local pieces corresponding to * the upper triangular part of the * submatrix sub( C ) are referenced, * UPLO = 'L' or 'l' Only the local pieces corresponding to * the lower triangular part of the * submatrix sub( C ) are referenced. * * TRANS (global input) pointer to CHAR * On entry, TRANS specifies the operation to be performed as * follows: * * TRANS = 'N' or 'n' * sub( C ) := alpha*sub( A )*sub( B )' + * alpha*sub( B )*sub( A )' + * beta*sub( C ), * or * sub( C ) := alpha*sub( A )*sub( B )' + * alpha*sub( B )*sub( A )' + * beta*sub( C ), * or * sub( C ) := alpha*sub( A )*conjg( sub( B )' ) + * conjg( alpha )*sub( B )*conjg( sub( A )' ) + * beta*sub( C ), * * TRANS = 'T' or 't' * sub( C ) := alpha*sub( B )'*sub( A ) + * alpha*sub( A )'*sub( B ) + * beta*sub( C ), * or * sub( C ) := alpha*sub( B )'*sub( A ) + * alpha*sub( A )'*sub( B ) + * beta*sub( C ), * * TRANS = 'C' or 'c' * sub( C ) := alpha*sub( B )'*sub( A ) + * alpha*sub( A )'*sub( B ) + * beta*sub( C ), * or * sub( C ) := alpha*conjg( sub( A )' )*sub( B ) + * conjg( alpha )*conjg( sub( B )' )*sub( A ) + * beta*sub( C ). * * N (global input) INTEGER * On entry, N specifies the order of the submatrix sub( C ). * N must be at least zero. * * K (global input) INTEGER * On entry with TRANS = 'N' or 'n', K specifies the number of * columns of the submatrices sub( A ) and sub( B ), and on * entry with TRANS = 'T' or 't' or 'C' or 'c', K specifies the * number of rows of the submatrices sub( A ) and sub( B ). * K must be at least zero. * * ALPHA (global input) pointer to CHAR * On entry, ALPHA specifies the scalar alpha. When ALPHA is * supplied as zero then the local entries of the arrays A * and B corresponding to the entries of the submatrices * sub( A ) and sub( B ) respectively need not be set on input. * * A (local input) pointer to CHAR * On entry, A is an array of dimension (LLD_A, Ka), where Ka is * at least Lc( 1, JA+K-1 ) when TRANS = 'N' or 'n', and is at * least Lc( 1, JA+N-1 ) otherwise. Before entry, this array * contains the local entries of the matrix A. * Before entry with TRANS = 'N' or 'n', this array contains the * local entries corresponding to the entries of the n by k sub- * matrix sub( A ), otherwise the local entries corresponding to * the entries of the k by n submatrix sub( A ). * * IA (global input) INTEGER * On entry, IA specifies A's global row index, which points to * the beginning of the submatrix sub( A ). * * JA (global input) INTEGER * On entry, JA specifies A's global column index, which points * to the beginning of the submatrix sub( A ). * * DESCA (global and local input) INTEGER array * On entry, DESCA is an integer array of dimension DLEN_. This * is the array descriptor for the matrix A. * * B (local input) pointer to CHAR * On entry, B is an array of dimension (LLD_B, Kb), where Kb is * at least Lc( 1, JB+K-1 ) when TRANS = 'N' or 'n', and is at * least Lc( 1, JB+N-1 ) otherwise. Before entry, this array * contains the local entries of the matrix B. * Before entry with TRANS = 'N' or 'n', this array contains the * local entries corresponding to the entries of the n by k sub- * matrix sub( B ), otherwise the local entries corresponding to * the entries of the k by n submatrix sub( B ). * * IB (global input) INTEGER * On entry, IB specifies B's global row index, which points to * the beginning of the submatrix sub( B ). * * JB (global input) INTEGER * On entry, JB specifies B's global column index, which points * to the beginning of the submatrix sub( B ). * * DESCB (global and local input) INTEGER array * On entry, DESCB is an integer array of dimension DLEN_. This * is the array descriptor for the matrix B. * * BETA (global input) pointer to CHAR * On entry, BETA specifies the scalar beta. When BETA is * supplied as zero then the local entries of the array C * corresponding to the entries of the submatrix sub( C ) need * not be set on input. * * C (local input/local output) pointer to CHAR * On entry, C is an array of dimension (LLD_C, Kc), where Kc is * at least Lc( 1, JC+N-1 ). Before entry, this array contains * the local entries of the matrix C. * Before entry with UPLO = 'U' or 'u', this array contains * the local entries corresponding to the upper triangular part * of the symmetric or Hermitian submatrix sub( C ), and the * local entries corresponding to the strictly lower triangular * of sub( C ) are not referenced. On exit, the upper triangular * part of sub( C ) is overwritten by the upper triangular part * of the updated submatrix. * Before entry with UPLO = 'L' or 'l', this array contains * the local entries corresponding to the lower triangular part * of the symmetric or Hermitian submatrix sub( C ), and the * local entries corresponding to the strictly upper triangular * of sub( C ) are not referenced. On exit, the lower triangular * part of sub( C ) is overwritten by the lower triangular part * of the updated submatrix. * Note that the imaginary parts of the local entries corres- * ponding to the diagonal elements of sub( C ) need not be * set, they are assumed to be zero, and on exit they are set * to zero. * * IC (global input) INTEGER * On entry, IC specifies C's global row index, which points to * the beginning of the submatrix sub( C ). * * JC (global input) INTEGER * On entry, JC specifies C's global column index, which points * to the beginning of the submatrix sub( C ). * * DESCC (global and local input) INTEGER array * On entry, DESCC is an integer array of dimension DLEN_. This * is the array descriptor for the matrix C. * * -- Written on April 1, 1998 by * Antoine Petitet, University of Tennessee, Knoxville 37996, USA. * * --------------------------------------------------------------------- */ /* * .. Local Scalars .. */ char GatherDir, ScatterDir, * one, top, * talpha, * tbeta, tran, * zero; int ABm, ABn, Acol, Acurcol, Acurrow, Acurimb1, Acurinb1, Afr, Aii, Aimb, Aimb1, Ainb, Ainb1, AisD, AisR, Ajj, Ald, Amb, Amp, Amp0, Anb, Anq, Anq0, Arow, Aspan, Bcol, Bcurcol, Bcurrow, Bcurimb1, Bcurinb1, Bfr, Bii, Bimb, Bimb1, Binb, Binb1, BisD, BisR, Bjj, Bld, Bmb, Bmp, Bmp0, Bnb, Bnq, Bnq0, Brow, Bspan, Ccsrc, Cimb, Cinb, Cmb, Cnb, Crsrc, WAfr, WACfr, WACld, WACreuse, WACsum, WBfr, WBCfr, WBCld, WBCsum, conjg, ctxt, fwd, k, kb, kbb, kend, kstart, kstep, ktmp, mycol, myrow, notran, npcol, nprow, size, upper; GEMM_T gemm; GSUM2D_T gsum2d; /* * .. Local Arrays .. */ char * Aptr = NULL, * Aptr0 = NULL, * Bptr = NULL, * Bptr0 = NULL, * WA = NULL, * WB = NULL, * WAC = NULL, *WBC = NULL; int Ad0[DLEN_], Bd0[DLEN_], DBUFA[DLEN_], DBUFB[DLEN_], WAd[DLEN_], WBd[DLEN_], WACd [DLEN_], WBCd [DLEN_]; /* .. * .. Executable Statements .. * */ /* * sub( C ) = beta * sub( C ) */ PB_Cplascal( TYPE, UPLO, CONJUG, N, N, BETA, C, IC, JC, DESCC ); /* * Retrieve process grid information */ Cblacs_gridinfo( ( ctxt = DESCA[CTXT_] ), &nprow, &npcol, &myrow, &mycol ); fwd = ( Mupcase( DIRECAB[0] ) == CFORWARD ); conjg = ( Mupcase( CONJUG [0] ) == CCONJG ); upper = ( Mupcase( UPLO [0] ) == CUPPER ); notran = ( Mupcase( TRANS [0] ) == CNOTRAN ); size = TYPE->size; one = TYPE->one; zero = TYPE->zero; gsum2d = TYPE->Cgsum2d; gemm = TYPE->Fgemm; /* * Figure out the loop bounds accordingly to DIRECAB */ kb = pilaenv_( &ctxt, C2F_CHAR( &TYPE->type ) ); if( fwd ) { kstart = 0; kend = ( ( N - 1 ) / kb + 1 ) * kb; kstep = kb; GatherDir = CFORWARD; ScatterDir = CBACKWARD; } else { kstart = ( ( N - 1 ) / kb ) * kb; kend = kstep = -kb; GatherDir = CBACKWARD; ScatterDir = CFORWARD; } /* * Compute conjg( ALPHA ) and transpose parameter for Hermitian case */ if( conjg ) { tran = CCOTRAN; talpha = PB_Cmalloc( size ); PB_Cconjg( TYPE, ALPHA, talpha ); } else { tran = CTRAN; talpha = ALPHA; } /* * Compute local information for sub( A ) and sub( B ) */ if( notran ) { ABm = N; ABn = K; } else { ABm = K; ABn = N; } PB_Cinfog2l( IA, JA, DESCA, nprow, npcol, myrow, mycol, &Aii, &Ajj, &Arow, &Acol ); Aimb = DESCA[IMB_]; Ainb = DESCA[INB_]; Amb = DESCA[MB_ ]; Anb = DESCA[NB_ ]; Ald = DESCA[LLD_]; Aimb1 = PB_Cfirstnb( ABm, IA, Aimb, Amb ); Amp0 = PB_Cnumroc( ABm, 0, Aimb1, Amb, myrow, Arow, nprow ); Ainb1 = PB_Cfirstnb( ABn, JA, Ainb, Anb ); Anq0 = PB_Cnumroc( ABn, 0, Ainb1, Anb, mycol, Acol, npcol ); if( ( Amp0 > 0 ) && ( Anq0 > 0 ) ) Aptr0 = Mptr( A, Aii, Ajj, Ald, size ); PB_Cinfog2l( IB, JB, DESCB, nprow, npcol, myrow, mycol, &Bii, &Bjj, &Brow, &Bcol ); Bimb = DESCB[IMB_]; Binb = DESCB[INB_]; Bmb = DESCB[MB_ ]; Bnb = DESCB[NB_ ]; Bld = DESCB[LLD_]; Bimb1 = PB_Cfirstnb( ABm, IB, Bimb, Bmb ); Bmp0 = PB_Cnumroc( ABm, 0, Bimb1, Bmb, myrow, Brow, nprow ); Binb1 = PB_Cfirstnb( ABn, JB, Binb, Bnb ); Bnq0 = PB_Cnumroc( ABn, 0, Binb1, Bnb, mycol, Bcol, npcol ); if( ( Bmp0 > 0 ) && ( Bnq0 > 0 ) ) Bptr0 = Mptr( B, Bii, Bjj, Bld, size ); if( notran ) { top = *PB_Ctop( &ctxt, COMBINE, ROW, TOP_GET ); Cinb = DESCC[INB_]; Cnb = DESCC[NB_]; Ccsrc = DESCC[CSRC_]; /* * Determine if one can reuse the WAC buffer for the intermediate local products * sub( A ) * sub( B )' and sub( B ) * sub( A )'. */ AisR = ( ( Acol < 0 ) || ( npcol == 1 ) ); BisR = ( ( Bcol < 0 ) || ( npcol == 1 ) ); if( !( AisR ) && !( BisR ) ) { /* * When neither sub( A ) nor sub( B ) are replicated, WAC can be reused if * either sub( A ) spans more than one process column, or, neither sub( A ) * nor sub( B ) span more than one process column, and both operands reside * in the same process column. */ Aspan = PB_Cspan( ABn, 0, Ainb1, Anb, Acol, npcol ); Bspan = PB_Cspan( ABn, 0, Binb1, Bnb, Bcol, npcol ); WACreuse = ( Aspan || ( !( Aspan ) && !( Bspan ) && ( Acol == Bcol ) ) ); } else { /* * Otherwise, WAC can be reused when both operands sub( A ) and sub( B ) are * replicated. */ WACreuse = ( AisR && BisR ); } /* * Furthermore, the ability to reuse WAC requires sub( A ) and sub( B ) to be * either both not row-distributed, or, both row-distributed and aligned. */ AisD = ( ( Arow >= 0 ) && ( nprow > 1 ) ); BisD = ( ( Brow >= 0 ) && ( nprow > 1 ) ); WACreuse = ( WACreuse && ( ( !AisD && !BisD ) || ( ( AisD && BisD ) && ( ( Arow == Brow ) && ( ( ( Aimb1 >= ABm ) && ( Bimb1 >= ABm ) ) || ( ( Aimb1 == Bimb1 ) && ( Amb == Bmb ) ) ) ) ) ) ); tbeta = ( WACreuse ? one : zero ); if( upper ) { for( k = kstart; k != kend; k += kstep ) { kbb = N - k; kbb = MIN( kbb, kb ); ktmp = k + kbb; /* * Accumulate B( IB+k:IB+k+kbb-1, JB:JB+K-1 ) */ PB_CGatherV( TYPE, REUSE, &GatherDir, kbb, ABn, B, IB+k, JB, DESCB, ROW, &Bptr, DBUFB, &Bfr ); /* * Replicate B( IB+k:IB+k+kbb-1, JB:JB+K-1 ) over A( IA:IA+k+kbb-1, JA:JA+K-1 ) */ PB_Cdescset( Ad0, ktmp, ABn, Aimb1, Ainb1, Amb, Anb, Arow, Acol, ctxt, Ald ); PB_CInV( TYPE, NOCONJG, ROW, ktmp, ABn, Ad0, kbb, Bptr, 0, 0, DBUFB, ROW, &WB, WBd, &WBfr ); /* * WAC := A( IA:IA+k+kbb-1, JA:JA+K-1 ) * B( IB+k:IB+k+kbb-1, JB:JB+K-1 )' */ PB_COutV( TYPE, COLUMN, INIT, ktmp, ABn, Ad0, kbb, &WAC, WACd, &WACfr, &WACsum ); WACld = WACd[LLD_]; Amp = PB_Cnumroc( ktmp, 0, Aimb1, Amb, myrow, Arow, nprow ); if( ( Amp > 0 ) && ( Anq0 > 0 ) ) gemm( C2F_CHAR( NOTRAN ), C2F_CHAR( &tran ), &Amp, &kbb, &Anq0, ALPHA, Aptr0, &Ald, WB, &WBd[LLD_], zero, WAC, &WACld ); if( WBfr ) free( WB ); if( Bfr ) free( Bptr ); /* * Accumulate A( IA+k:IA+k+kbb-1, JA:JA+K-1 ) */ PB_CGatherV( TYPE, REUSE, &GatherDir, kbb, ABn, A, IA+k, JA, DESCA, ROW, &Aptr, DBUFA, &Afr ); /* * Replicate A( IA+k:IA+k+kbb-1, JA:JA+K-1 ) over B( IB:IB+k+kbb-1, JB:JB+K-1 ) */ PB_Cdescset( Bd0, ktmp, ABn, Bimb1, Binb1, Bmb, Bnb, Brow, Bcol, ctxt, Bld ); PB_CInV( TYPE, NOCONJG, ROW, ktmp, ABn, Bd0, kbb, Aptr, 0, 0, DBUFA, ROW, &WA, WAd, &WAfr ); /* * WBC := B( IB:IB+k+kbb-1, JB:JB+K-1 ) * A( IA+k:IA+k+kbb-1, JA:JA+K-1 )' */ if( WACreuse ) { WBC = WAC; MDescCopy( WACd, WBCd ); WBCfr = 0; WBCsum = WACsum; } else { PB_COutV( TYPE, COLUMN, INIT, ktmp, ABn, Bd0, kbb, &WBC, WBCd, &WBCfr, &WBCsum ); } WBCld = WBCd[LLD_]; Bmp = PB_Cnumroc( ktmp, 0, Bimb1, Bmb, myrow, Brow, nprow ); if( ( Bmp > 0 ) && ( Bnq0 > 0 ) ) gemm( C2F_CHAR( NOTRAN ), C2F_CHAR( &tran ), &Bmp, &kbb, &Bnq0, talpha, Bptr0, &Bld, WA, &WAd[LLD_], tbeta, WBC, &WBCld ); if( WAfr ) free( WA ); if( Afr ) free( Aptr ); /* * Combine the local copies of WAC when necessary */ if( WACsum ) { WACd[CSRC_] = PB_Cindxg2p( JC + ( fwd ? k : ktmp - 1 ), Cinb, Cnb, Ccsrc, Ccsrc, npcol ); if( Amp > 0 ) gsum2d( ctxt, ROW, &top, Amp, kbb, WAC, WACld, myrow, WACd[CSRC_] ); } /* * Zero lower triangle of WAC( k:k+kbb-1, 0:kbb-1 ) */ if( conjg ) PB_Cplapad( TYPE, LOWER, CONJG, kbb, kbb, zero, zero, WAC, k, 0, WACd ); else if( kbb > 1 ) PB_Cplapad( TYPE, LOWER, NOCONJG, kbb-1, kbb-1, zero, zero, WAC, k+1, 0, WACd ); /* * Combine the local copies of WBC when necessary */ if( !( WACreuse ) ) { if( WBCsum ) { if( WACsum ) { WBCd[CSRC_] = WACd[CSRC_]; } else { WBCd[CSRC_] = PB_Cindxg2p( JC + ( fwd ? k : ktmp - 1 ), Cinb, Cnb, Ccsrc, Ccsrc, npcol ); } if( Bmp > 0 ) gsum2d( ctxt, ROW, &top, Bmp, kbb, WBC, WBCld, myrow, WBCd[CSRC_] ); } /* * Zero lower triangle of WBC( k:k+kbb-1, 0:kbb-1 ) */ if( conjg ) PB_Cplapad( TYPE, LOWER, CONJG, kbb, kbb, zero, zero, WBC, k, 0, WBCd ); else if( kbb > 1 ) PB_Cplapad( TYPE, LOWER, NOCONJG, kbb-1, kbb-1, zero, zero, WBC, k+1, 0, WBCd ); } /* * Add WAC to C( IC:IC+k+kbb-1, JC+k:JC+k+kbb-1 ) */ PB_CScatterV( TYPE, &ScatterDir, ktmp, kbb, WAC, 0, 0, WACd, COLUMN, one, C, IC, JC+k, DESCC, COLUMN ); if( WACfr ) free( WAC ); /* * Add WBC to C( IC:IC+k+kbb-1, JC+k:JC+k+kbb-1 ) */ if( !( WACreuse ) ) { PB_CScatterV( TYPE, &ScatterDir, ktmp, kbb, WBC, 0, 0, WBCd, COLUMN, one, C, IC, JC+k, DESCC, COLUMN ); if( WBCfr ) free( WBC ); } } } else { for( k = kstart; k != kend; k += kstep ) { ktmp = N - k; kbb = MIN( ktmp, kb ); /* * Accumulate B( IB+k:IB+k+kbb-1, JB:JB+K-1 ) */ PB_CGatherV( TYPE, REUSE, &GatherDir, kbb, ABn, B, IB+k, JB, DESCB, ROW, &Bptr, DBUFB, &Bfr ); /* * Replicate B( IB+k:IB+k+kbb-1, JB:JB+K-1 ) over A( IA+k:IA+N-1, JA:JA+K-1 ) */ Acurimb1 = PB_Cfirstnb( ktmp, IA+k, Aimb, Amb ); Acurrow = PB_Cindxg2p( k, Aimb1, Amb, Arow, Arow, nprow ); PB_Cdescset( Ad0, ktmp, ABn, Acurimb1, Ainb1, Amb, Anb, Acurrow, Acol, ctxt, Ald ); PB_CInV( TYPE, NOCONJG, ROW, ktmp, ABn, Ad0, kbb, Bptr, 0, 0, DBUFB, ROW, &WB, WBd, &WBfr ); /* * WAC := A( IA+k:IA+N-1, JA:JA+K-1 ) * B( IB+k:IB+k+kbb-1, JB:JB+K-1 )' */ PB_COutV( TYPE, COLUMN, INIT, ktmp, ABn, Ad0, kbb, &WAC, WACd, &WACfr, &WACsum ); WACld = WACd[LLD_]; Amp = PB_Cnumroc( ktmp, k, Aimb1, Amb, myrow, Arow, nprow ); if( ( Amp > 0 ) && ( Anq0 > 0 ) ) gemm( C2F_CHAR( NOTRAN ), C2F_CHAR( &tran ), &Amp, &kbb, &Anq0, ALPHA, Mptr( Aptr0, Amp0-Amp, 0, Ald, size ), &Ald, WB, &WBd[LLD_], zero, WAC, &WACld ); if( WBfr ) free( WB ); if( Bfr ) free( Bptr ); /* * Accumulate A( IA+k:IA+k+kbb-1, JA:JA+K-1 ) */ PB_CGatherV( TYPE, REUSE, &GatherDir, kbb, ABn, A, IA+k, JA, DESCA, ROW, &Aptr, DBUFA, &Afr ); /* * Replicate A( IA+k:IA+k+kbb-1, JA:JA+K-1 ) over B( IB+k:IB+N-1, JB:JB+K-1 ) */ Bcurimb1 = PB_Cfirstnb( ktmp, IB+k, Bimb, Bmb ); Bcurrow = PB_Cindxg2p( k, Bimb1, Bmb, Brow, Brow, nprow ); PB_Cdescset( Bd0, ktmp, ABn, Bcurimb1, Binb1, Bmb, Bnb, Bcurrow, Bcol, ctxt, Bld ); PB_CInV( TYPE, NOCONJG, ROW, ktmp, ABn, Bd0, kbb, Aptr, 0, 0, DBUFA, ROW, &WA, WAd, &WAfr ); /* * WBC := B( IB+k:IB+N-1, JB:JB+K-1 ) * A( IA+k:IA+k+kbb-1, JA:JA+K-1 )' */ if( WACreuse ) { WBC = WAC; MDescCopy( WACd, WBCd ); WBCfr = 0; WBCsum = WACsum; } else { PB_COutV( TYPE, COLUMN, INIT, ktmp, ABn, Bd0, kbb, &WBC, WBCd, &WBCfr, &WBCsum ); } WBCld = WBCd[LLD_]; Bmp = PB_Cnumroc( ktmp, k, Bimb1, Bmb, myrow, Brow, nprow ); if( ( Bmp > 0 ) && ( Bnq0 > 0 ) ) gemm( C2F_CHAR( NOTRAN ), C2F_CHAR( &tran ), &Bmp, &kbb, &Bnq0, talpha, Mptr( Bptr0, Bmp0-Bmp, 0, Bld, size ), &Bld, WA, &WAd[LLD_], tbeta, WBC, &WBCld ); if( WAfr ) free( WA ); if( Afr ) free( Aptr ); /* * Combine the local copies of WAC when necessary */ if( WACsum ) { WACd[CSRC_] = PB_Cindxg2p( JC + ( fwd ? k : k + kbb - 1 ), Cinb, Cnb, Ccsrc, Ccsrc, npcol ); if( Amp > 0 ) gsum2d( ctxt, ROW, &top, Amp, kbb, WAC, WACld, myrow, WACd[CSRC_] ); } /* * Zero upper triangle of WAC( 0:kbb-1, 0:kbb-1 ) */ if( conjg ) PB_Cplapad( TYPE, UPPER, CONJG, kbb, kbb, zero, zero, WAC, 0, 0, WACd ); else if( kbb > 1 ) PB_Cplapad( TYPE, UPPER, NOCONJG, kbb-1, kbb-1, zero, zero, WAC, 0, 1, WACd ); /* * Combine the local copies of WBC when necessary */ if( !( WACreuse ) ) { if( WBCsum ) { if( WACsum ) { WBCd[CSRC_] = WACd[CSRC_]; } else { WBCd[CSRC_] = PB_Cindxg2p( JC + ( fwd ? k : k+kbb-1 ), Cinb, Cnb, Ccsrc, Ccsrc, npcol ); } if( Bmp > 0 ) gsum2d( ctxt, ROW, &top, Bmp, kbb, WBC, WBCld, myrow, WBCd[CSRC_] ); } /* * Zero upper triangle of WBC( 0:kbb-1, 0:kbb-1 ) */ if( conjg ) PB_Cplapad( TYPE, UPPER, CONJG, kbb, kbb, zero, zero, WBC, 0, 0, WBCd ); else if( kbb > 1 ) PB_Cplapad( TYPE, UPPER, NOCONJG, kbb-1, kbb-1, zero, zero, WBC, 0, 1, WBCd ); } /* * Add WAC to C( IC+k:IC+N-1, JC+k:JC+k+kbb-1 ) */ PB_CScatterV( TYPE, &ScatterDir, ktmp, kbb, WAC, 0, 0, WACd, COLUMN, one, C, IC+k, JC+k, DESCC, COLUMN ); if( WACfr ) free( WAC ); /* * Add WBC to C( IC+k:IC+N-1, JC+k:JC+k+kbb-1 ) */ if( !( WACreuse ) ) { PB_CScatterV( TYPE, &ScatterDir, ktmp, kbb, WBC, 0, 0, WBCd, COLUMN, one, C, IC+k, JC+k, DESCC, COLUMN ); if( WBCfr ) free( WBC ); } } } } else { top = *PB_Ctop( &ctxt, COMBINE, COLUMN, TOP_GET ); Cimb = DESCC[IMB_]; Cmb = DESCC[MB_]; Crsrc = DESCC[RSRC_]; /* * Determine if one can reuse the WAC buffer for the intermediate local products * sub( A )' * sub( B ) and sub( B )' * sub( A ). */ AisR = ( ( Arow < 0 ) || ( nprow == 1 ) ); BisR = ( ( Brow < 0 ) || ( nprow == 1 ) ); /* * When neither sub( A ) nor sub( B ) are replicated, WAC can be reused if * either sub( A ) spans more than one process row, or, neither sub( A ) nor * sub( B ) span more than one process row, and both operands reside in the * same process row. */ if( !( AisR ) && !( BisR ) ) { Aspan = PB_Cspan( ABm, 0, Aimb1, Amb, Arow, nprow ); Bspan = PB_Cspan( ABm, 0, Bimb1, Bmb, Brow, nprow ); WACreuse = ( Aspan || ( !( Aspan ) && !( Bspan ) && ( Arow == Brow ) ) ); } else { /* * Otherwise, WAC can be reused when both operands sub( A ) and sub( B ) are * replicated. */ WACreuse = ( AisR && BisR ); } /* * Furthermore, the ability to reuse WAC requires sub( A ) and sub( B ) to be * either both not column-distributed, or, both column-distributed and aligned. */ AisD = ( ( Acol >= 0 ) && ( npcol > 1 ) ); BisD = ( ( Bcol >= 0 ) && ( npcol > 1 ) ); WACreuse = ( WACreuse && ( ( !AisD && !BisD ) || ( ( AisD && BisD ) && ( ( Acol == Bcol ) && ( ( ( Ainb1 >= ABn ) && ( Binb1 >= ABn ) ) || ( ( Ainb1 == Binb1 ) && ( Anb == Bnb ) ) ) ) ) ) ); tbeta = ( WACreuse ? one : zero ); if( upper ) { for( k = kstart; k != kend; k += kstep ) { ktmp = N - k; kbb = MIN( ktmp, kb ); /* * Accumulate B( IB:IB+K-1, JB+k:JB+k+kbb-1 ) */ PB_CGatherV( TYPE, REUSE, &GatherDir, ABm, kbb, B, IB, JB+k, DESCB, COLUMN, &Bptr, DBUFB, &Bfr ); /* * Replicate B( IB:IB+K-1, JB+k:JB+k+kbb-1 ) over A( IA:IA+K-1, JA+k:JA+N-1 ) */ Acurinb1 = PB_Cfirstnb( ktmp, JA+k, Ainb, Anb ); Acurcol = PB_Cindxg2p( k, Ainb1, Anb, Acol, Acol, npcol ); PB_Cdescset( Ad0, ABm, ktmp, Aimb1, Acurinb1, Amb, Anb, Arow, Acurcol, ctxt, Ald ); PB_CInV( TYPE, NOCONJG, COLUMN, ABm, ktmp, Ad0, kbb, Bptr, 0, 0, DBUFB, COLUMN, &WB, WBd, &WBfr ); /* * WAC := B( IB:IB+K-1, JB+k:JB+k+kbb-1 )' * A( IA:IA+K-1, JA+k:JA+N-1 ) */ PB_COutV( TYPE, ROW, INIT, ABm, ktmp, Ad0, kbb, &WAC, WACd, &WACfr, &WACsum ); WACld = WACd[LLD_]; Anq = PB_Cnumroc( ktmp, k, Ainb1, Anb, mycol, Acol, npcol ); if( ( Anq > 0 ) && ( Amp0 > 0 ) ) gemm( C2F_CHAR( &tran ), C2F_CHAR( NOTRAN ), &kbb, &Anq, &Amp0, talpha, WB, &WBd[LLD_], Mptr( Aptr0, 0, Anq0-Anq, Ald, size ), &Ald, zero, WAC, &WACld ); if( WBfr ) free( WB ); if( Bfr ) free( Bptr ); /* * Accumulate A( IA:IA+K-1, JA+k:JA+k+kbb-1 ) */ PB_CGatherV( TYPE, REUSE, &GatherDir, ABm, kbb, A, IA, JA+k, DESCA, COLUMN, &Aptr, DBUFA, &Afr ); /* * Replicate A( IA:IA+K-1, JA+k:JA+k+kbb-1 ) over B( IB:IB+K-1, JB+k:JB+N-1 ) */ Bcurinb1 = PB_Cfirstnb( ktmp, JB+k, Binb, Bnb ); Bcurcol = PB_Cindxg2p( k, Binb1, Bnb, Bcol, Bcol, npcol ); PB_Cdescset( Bd0, ABm, ktmp, Bimb1, Bcurinb1, Bmb, Bnb, Brow, Bcurcol, ctxt, Bld ); PB_CInV( TYPE, NOCONJG, COLUMN, ABm, ktmp, Bd0, kbb, Aptr, 0, 0, DBUFA, COLUMN, &WA, WAd, &WAfr ); /* * WBC := A( IA:IA+K-1, JA+k:JA+k+kbb-1 )' * B( IB:IB+K-1, JB+k:JB+N-1 ) */ if( WACreuse ) { WBC = WAC; MDescCopy( WACd, WBCd ); WBCfr = 0; WBCsum = WACsum; } else { PB_COutV( TYPE, ROW, INIT, ABm, ktmp, Bd0, kbb, &WBC, WBCd, &WBCfr, &WBCsum ); } WBCld = WBCd[LLD_]; Bnq = PB_Cnumroc( ktmp, k, Binb1, Bnb, mycol, Bcol, npcol ); if( ( Bnq > 0 ) && ( Bmp0 > 0 ) ) gemm( C2F_CHAR( &tran ), C2F_CHAR( NOTRAN ), &kbb, &Bnq, &Bmp0, ALPHA, WA, &WAd[LLD_], Mptr( Bptr0, 0, Bnq0-Bnq, Bld, size ), &Bld, tbeta, WBC, &WBCld ); if( WAfr ) free( WA ); if( Afr ) free( Aptr ); /* * Combine the local copies of WAC when necessary */ if( WACsum ) { WACd[RSRC_] = PB_Cindxg2p( IC + ( fwd ? k : k + kbb - 1 ), Cimb, Cmb, Crsrc, Crsrc, nprow ); if( Anq > 0 ) gsum2d( ctxt, COLUMN, &top, kbb, Anq, WAC, WACld, WACd[RSRC_], mycol ); } /* * Zero lower triangle of WBC( 0:kbb-1, 0:kbb-1 ) */ if( conjg ) PB_Cplapad( TYPE, LOWER, CONJG, kbb, kbb, zero, zero, WAC, 0, 0, WACd ); else if( kbb > 1 ) PB_Cplapad( TYPE, LOWER, NOCONJG, kbb-1, kbb-1, zero, zero, WAC, 1, 0, WACd ); /* * Combine the local copies of WBC when necessary */ if( !( WACreuse ) ) { if( WBCsum ) { if( WACsum ) { WBCd[RSRC_] = WACd[RSRC_]; } else { WBCd[RSRC_] = PB_Cindxg2p( IC + ( fwd ? k : k + kbb - 1 ), Cimb, Cmb, Crsrc, Crsrc, nprow ); } if( Bnq > 0 ) gsum2d( ctxt, COLUMN, &top, kbb, Bnq, WBC, WBCld, WBCd[RSRC_], mycol ); } /* * Zero lower triangle of WBC( 0:kbb-1, 0:kbb-1 ) */ if( conjg ) PB_Cplapad( TYPE, LOWER, CONJG, kbb, kbb, zero, zero, WBC, 0, 0, WBCd ); else if( kbb > 1 ) PB_Cplapad( TYPE, LOWER, NOCONJG, kbb-1, kbb-1, zero, zero, WBC, 1, 0, WBCd ); } /* * Add WAC to C( IC+k:IC+k+kbb-1, JC+k:JC+N-1 ) */ PB_CScatterV( TYPE, &ScatterDir, kbb, ktmp, WAC, 0, 0, WACd, ROW, one, C, IC+k, JC+k, DESCC, ROW ); if( WACfr ) free( WAC ); /* * Add WBC to C( IC+k:IC+k+kbb-1, JC+k:JC+N-1 ) */ if( !( WACreuse ) ) { PB_CScatterV( TYPE, &ScatterDir, kbb, ktmp, WBC, 0, 0, WBCd, ROW, one, C, IC+k, JC+k, DESCC, ROW ); if( WBCfr ) free( WBC ); } } } else { for( k = kstart; k != kend; k += kstep ) { kbb = N - k; kbb = MIN( kbb, kb ); ktmp = k + kbb; /* * Accumulate B( IB:IB+K-1, JB+k:JB+k+kbb-1 ) */ PB_CGatherV( TYPE, REUSE, &GatherDir, ABm, kbb, B, IB, JB+k, DESCB, COLUMN, &Bptr, DBUFB, &Bfr ); /* * Replicate B( IB:IB+K-1, JB+k:JB+k+kbb-1 ) over A( IA:IA+K-1, JA:JA+k+kbb-1 ) */ PB_Cdescset( Ad0, ABm, ktmp, Aimb1, Ainb1, Amb, Anb, Arow, Acol, ctxt, Ald ); PB_CInV( TYPE, NOCONJG, COLUMN, ABm, ktmp, Ad0, kbb, Bptr, 0, 0, DBUFB, COLUMN, &WB, WBd, &WBfr ); /* * WAC := B( IB:IB+K-1, JB+k:JB+k+kbb-1 )' * A( IA:IA+K-1, JA:JA+k+kbb-1 ) */ PB_COutV( TYPE, ROW, INIT, ABm, ktmp, Ad0, kbb, &WAC, WACd, &WACfr, &WACsum ); WACld = WACd[LLD_]; Anq = PB_Cnumroc( ktmp, 0, Ainb1, Anb, mycol, Acol, npcol ); if( ( Anq > 0 ) && ( Amp0 > 0 ) ) gemm( C2F_CHAR( &tran ), C2F_CHAR( NOTRAN ), &kbb, &Anq, &Amp0, talpha, WB, &WBd[LLD_], Aptr0, &Ald, zero, WAC, &WACld ); if( WBfr ) free( WB ); if( Bfr ) free( Bptr ); /* * Accumulate A( IA:IA+K-1, JA+k:JA+k+kbb-1 ) */ PB_CGatherV( TYPE, REUSE, &GatherDir, ABm, kbb, A, IA, JA+k, DESCA, COLUMN, &Aptr, DBUFA, &Afr ); /* * Replicate A( IA:IA+K-1, JA+k:JA+k+kbb-1 ) over B( IB:IB+K-1, JB:JB+k+kbb-1 ) */ PB_Cdescset( Bd0, ABm, ktmp, Bimb1, Binb1, Bmb, Bnb, Brow, Bcol, ctxt, Bld ); PB_CInV( TYPE, NOCONJG, COLUMN, ABm, ktmp, Bd0, kbb, Aptr, 0, 0, DBUFA, COLUMN, &WA, WAd, &WAfr ); /* * WBC := A( IA:IA+K-1, JA+k:JA+k+kbb-1 )' * B( IB:IB+K-1, JB:JB+k+kbb-1 ) */ if( WACreuse ) { WBC = WAC; MDescCopy( WACd, WBCd ); WBCfr = 0; WBCsum = WACsum; } else { PB_COutV( TYPE, ROW, INIT, ABm, ktmp, Bd0, kbb, &WBC, WBCd, &WBCfr, &WBCsum ); } WBCld = WBCd[LLD_]; Bnq = PB_Cnumroc( ktmp, 0, Binb1, Bnb, mycol, Bcol, npcol ); if( ( Bnq > 0 ) && ( Bmp0 > 0 ) ) gemm( C2F_CHAR( &tran ), C2F_CHAR( NOTRAN ), &kbb, &Bnq, &Bmp0, ALPHA, WA, &WAd[LLD_], Bptr0, &Bld, tbeta, WBC, &WBCld ); if( WAfr ) free( WA ); if( Afr ) free( Aptr ); /* * Combine the local copies of WAC when necessary */ if( WACsum ) { WACd[RSRC_] = PB_Cindxg2p( IC + ( fwd ? k : ktmp - 1 ), Cimb, Cmb, Crsrc, Crsrc, nprow ); if( Anq > 0 ) gsum2d( ctxt, COLUMN, &top, kbb, Anq, WAC, WACld, WACd[RSRC_], mycol ); } /* * Zero upper triangle of WBC( 0:kbb-1, k:k+kbb-1 ) */ if( conjg ) PB_Cplapad( TYPE, UPPER, CONJG, kbb, kbb, zero, zero, WAC, 0, k, WACd ); else if( kbb > 1 ) PB_Cplapad( TYPE, UPPER, NOCONJG, kbb-1, kbb-1, zero, zero, WAC, 0, k+1, WACd ); /* * Combine the local copies of WBC when necessary */ if( !( WACreuse ) ) { if( WBCsum ) { if( WACsum ) { WBCd[RSRC_] = WACd[RSRC_]; } else { WBCd[RSRC_] = PB_Cindxg2p( IC + ( fwd ? k : ktmp - 1 ), Cimb, Cmb, Crsrc, Crsrc, nprow ); } if( Bnq > 0 ) gsum2d( ctxt, COLUMN, &top, kbb, Bnq, WBC, WBCld, WBCd[RSRC_], mycol ); } /* * Zero upper triangle of WBC( 0:kbb-1, k:k+kbb-1 ) */ if( conjg ) PB_Cplapad( TYPE, UPPER, CONJG, kbb, kbb, zero, zero, WBC, 0, k, WBCd ); else if( kbb > 1 ) PB_Cplapad( TYPE, UPPER, NOCONJG, kbb-1, kbb-1, zero, zero, WBC, 0, k+1, WBCd ); } /* * Add WAC to C( IC+k:IC+k+kbb-1, JC:JC+k+kbb-1 ) */ PB_CScatterV( TYPE, &ScatterDir, kbb, ktmp, WAC, 0, 0, WACd, ROW, one, C, IC+k, JC, DESCC, ROW ); if( WACfr ) free( WAC ); /* * Add WBC to C( IC+k:IC+k+kbb-1, JC:JC+k+kbb-1 ) */ if( !( WACreuse ) ) { PB_CScatterV( TYPE, &ScatterDir, kbb, ktmp, WBC, 0, 0, WBCd, ROW, one, C, IC+k, JC, DESCC, ROW ); if( WBCfr ) free( WBC ); } } } } if( conjg ) free( talpha ); /* * End of PB_Cpsyr2kAC */ }