|
ScaLAPACK
2.0.2
ScaLAPACK: Scalable Linear Algebra PACKage
|
|
ScaLAPACK
2.0.2
ScaLAPACK: Scalable Linear Algebra PACKage
|
#include "../pblas.h"#include "../PBpblas.h"#include "../PBtools.h"#include "../PBblacs.h"#include "../PBblas.h"Go to the source code of this file.
Functions/Subroutines | |
| void | PB_Cainfog2l (int M, int N, int I, int J, int *DESC, int NPROW, int NPCOL, int MYROW, int MYCOL, int *IMB1, int *INB1, int *MP, int *NQ, int *II, int *JJ, int *PROW, int *PCOL, int *RPROW, int *RPCOL) |
| void PB_Cainfog2l | ( | int | M, |
| int | N, | ||
| int | I, | ||
| int | J, | ||
| int * | DESC, | ||
| int | NPROW, | ||
| int | NPCOL, | ||
| int | MYROW, | ||
| int | MYCOL, | ||
| int * | IMB1, | ||
| int * | INB1, | ||
| int * | MP, | ||
| int * | NQ, | ||
| int * | II, | ||
| int * | JJ, | ||
| int * | PROW, | ||
| int * | PCOL, | ||
| int * | RPROW, | ||
| int * | RPCOL | ||
| ) |
Definition at line 25 of file PB_Cainfog2l.c.
{
/*
* Purpose
* =======
*
* PB_Cainfog2l computes the starting local row and column indexes II,
* JJ corresponding to the submatrix starting globally at the entry
* pointed by I, J. This routine returns the coordinates in the grid of
* the process owning the matrix entry of global indexes I, J, namely
* PROW and PCOL. In addition, this routine computes the quantities MP
* and NQ, which are respectively the local number of rows and columns
* owned by the process of coordinate MYROW, MYCOL corresponding to the
* global submatrix A(I:I+M-1,J:J+N-1). Finally, the size of the first
* partial block and the relative process coordinates are also returned
* respectively in IMB, INB and RPROW, RPCOL.
*
* 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
* =========
*
* M (global input) INTEGER
* On entry, M specifies the global number of rows of the subma-
* trix. M must be at least zero.
*
* N (global input) INTEGER
* On entry, N specifies the global number of columns of the
* submatrix. N must be at least zero.
*
* I (global input) INTEGER
* On entry, I specifies the global starting row index of the
* submatrix. I must at least zero.
*
* J (global input) INTEGER
* On entry, J specifies the global starting column index of
* the submatrix. J must at least zero.
*
* DESC (global and local input) INTEGER array
* On entry, DESC is an integer array of dimension DLEN_. This
* is the array descriptor of the underlying matrix.
*
* NPROW (global input) INTEGER
* On entry, NPROW specifies the total number of process rows
* over which the matrix is distributed. NPROW must be at least
* one.
*
* NPCOL (global input) INTEGER
* On entry, NPCOL specifies the total number of process columns
* over which the matrix is distributed. NPCOL must be at least
* one.
*
* MYROW (local input) INTEGER
* On entry, MYROW specifies the row coordinate of the process
* whose local index II is determined. MYROW must be at least
* zero and strictly less than NPROW.
*
* MYCOL (local input) INTEGER
* On entry, MYCOL specifies the column coordinate of the pro-
* cess whose local index JJ is determined. MYCOL must be at
* least zero and strictly less than NPCOL.
*
* IMB1 (global output) INTEGER
* On exit, IMB1 specifies the number of rows of the upper left
* block of the submatrix. On exit, IMB1 is less or equal than
* M and greater or equal than MIN( 1, M ).
*
* INB1 (global output) INTEGER
* On exit, INB1 specifies the number of columns of the upper
* left block of the submatrix. On exit, INB1 is less or equal
* than N and greater or equal than MIN( 1, N ).
*
* MP (local output) INTEGER
* On exit, MP specifies the local number of rows of the subma-
* trix, that the processes of row coordinate MYROW own. MP is
* at least zero.
*
* NQ (local output) INTEGER
* On exit, NQ specifies the local number of columns of the
* submatrix, that the processes of column coordinate MYCOL
* own. NQ is at least zero.
*
* II (local output) INTEGER
* On exit, II specifies the local starting row index of the
* submatrix. On exit, II is at least zero.
*
* JJ (local output) INTEGER
* On exit, JJ specifies the local starting column index of
* the submatrix. On exit, II is at least zero.
*
* PROW (global output) INTEGER
* On exit, PROW specifies the row coordinate of the process
* that possesses the first row of the submatrix. On exit, PROW
* is -1 if DESC(RSRC_) is -1 on input, and, at least zero and
* strictly less than NPROW otherwise.
*
* PCOL (global output) INTEGER
* On exit, PCOL specifies the column coordinate of the process
* that possesses the first column of the submatrix. On exit,
* PCOL is -1 if DESC(CSRC_) is -1 on input, and, at least zero
* and strictly less than NPCOL otherwise.
*
* RPROW (global output) INTEGER
* On exit, RPROW specifies the relative row coordinate of the
* process that possesses the first row I of the submatrix. On
* exit, RPROW is -1 if DESC(RSRC_) is -1 on input, and, at
* least zero and strictly less than NPROW otherwise.
*
* RPCOL (global output) INTEGER
* On exit, RPCOL specifies the relative column coordinate of
* the process that possesses the first column J of the subma-
* trix. On exit, RPCOL is -1 if DESC(CSRC_) is -1 on input,
* and, at least zero and strictly less than NPCOL otherwise.
*
* -- Written on April 1, 1998 by
* Antoine Petitet, University of Tennessee, Knoxville 37996, USA.
*
* ---------------------------------------------------------------------
*/
/*
* .. Local Scalars ..
*/
int i1, ilocblk, j1, mb, mydist, nb, nblocks, csrc, rsrc;
/* ..
* .. Executable Statements ..
*
*/
/*
* Retrieve the row distribution parameters
*/
mb = DESC[ MB_ ];
rsrc = DESC[ RSRC_ ];
if( ( rsrc == -1 ) || ( NPROW == 1 ) )
{
/*
* The rows are not distributed, or there is just one process row in the grid.
* Therefore, the local and global indexes are the same, as well as the local
* and global number of rows. Finally, the relative row process coordinate is
* zero, since every process owns all rows. Note that the size of the first
* row block can be zero only if M is zero.
*/
*II = I;
if( ( *IMB1 = DESC[IMB_] - I ) <= 0 )
*IMB1 += ( ( -(*IMB1) ) / mb + 1 ) * mb;
*IMB1 = MIN( *IMB1, M );
*MP = M;
*PROW = rsrc;
*RPROW = 0;
}
else
{
/*
* Figure out PROW, II and IMB1 first.
*/
*IMB1 = DESC[IMB_];
if( I < *IMB1 ) /* Is I in first block range ? */
{
/*
* If I is in the first block of rows, then PROW is simply rsrc, II is I in
* this process and zero elsewhere, and the size of the first block is the
* IMB complement.
*/
*PROW = rsrc;
*II = ( ( MYROW == *PROW ) ? I : 0 );
*IMB1 -= I;
}
else
{
/*
* The discussion goes as follows: compute my distance from the source process
* so that within this process coordinate system, the source row process is the
* process such that mydist=0, or equivalently MYROW == rsrc.
*
* Find out the global coordinate of the block of rows I belongs to (nblocks),
* as well as the minimum local number of row blocks that every process has.
*
* when mydist < nblocks - ilocblk * NPROW, I own ilocblk + 1 full blocks,
* when mydist > nblocks - ilocblk * NPROW, I own ilocblk full blocks,
* when mydist = nblocks - ilocblk * NPROW, I own ilocblk full blocks
* but not I, or I own ilocblk + 1 blocks and the entry I refers to.
*/
i1 = I - *IMB1;
if( MYROW == rsrc )
{
/*
* I refers to an entry that is not in the first block, find out which process
* has it.
*/
nblocks = i1 / mb + 1;
*PROW = rsrc + nblocks;
*PROW -= ( *PROW / NPROW ) * NPROW;
/*
* Since mydist = 0 and nblocks - ilocblk * NPROW >= 0, there are only three
* possible cases:
*
* 1) When 0 = mydist = nblocks - ilocblk * NPROW = 0 and I don't own I, in
* which case II = IMB + ( ilocblk - 1 ) * MB. Note that this case cannot
* happen when ilocblk is zero, since nblocks is at least one.
*
* 2) When 0 = mydist = nblocks - ilocblk * NPROW = 0 and I own I, in which
* case I and II can respectively be written as IMB + (nblocks-1)*MB + IL
* and IMB+(ilocblk-1) * MB + IL. That is II = I + (ilocblk - nblocks)*MB.
* Note that this case cannot happen when ilocblk is zero, since nblocks
* is at least one.
*
* 3) mydist = 0 < nblocks - ilocblk * NPROW, the source process owns
* ilocblk+1 full blocks, and therefore II = IMB + ilocblk * MB. Note
* that when ilocblk is zero, II is just IMB.
*/
if( nblocks < NPROW )
{
*II = *IMB1;
}
else
{
ilocblk = nblocks / NPROW;
if( ilocblk * NPROW >= nblocks )
{
*II = ( ( MYROW == *PROW ) ? I + ( ilocblk - nblocks ) * mb :
*IMB1 + ( ilocblk - 1 ) * mb );
}
else
{
*II = *IMB1 + ilocblk * mb;
}
}
}
else
{
/*
* I is not in the first block, find out which process has it.
*/
nblocks = i1 / mb + 1;
*PROW = rsrc + nblocks;
*PROW -= ( *PROW / NPROW ) * NPROW;
/*
* Compute my distance from the source process so that within this process
* coordinate system, the source process is the process such that mydist=0.
*/
if( ( mydist = MYROW - rsrc ) < 0 ) mydist += NPROW;
/*
* When mydist < nblocks - ilocblk * NPROW, I own ilocblk + 1 full blocks of
* size MB since I am not the source process, i.e. II = ( ilocblk + 1 ) * MB.
* When mydist >= nblocks - ilocblk * NPROW and I don't own I, I own ilocblk
* full blocks of size MB, i.e. II = ilocblk * MB, otherwise I own ilocblk
* blocks and I, in which case I can be written as IMB + (nblocks-1)*MB + IL
* and II = ilocblk*MB + IL = I - IMB + ( ilocblk - nblocks + 1 )*MB.
*/
if( nblocks < NPROW )
{
mydist -= nblocks;
*II = ( ( mydist < 0 ) ? mb : ( ( MYROW == *PROW ) ?
i1 + ( 1 - nblocks ) * mb : 0 ) );
}
else
{
ilocblk = nblocks / NPROW;
mydist -= nblocks - ilocblk * NPROW;
*II = ( ( mydist < 0 ) ? ( ilocblk + 1 ) * mb :
( ( MYROW == *PROW ) ?
( ilocblk - nblocks + 1 ) * mb + i1 :
ilocblk * mb ) );
}
}
/*
* Update the size of first block
*/
*IMB1 = nblocks * mb - i1;
}
/*
* Now everything is just like M, I=0, IMB1, MB, PROW, NPROW. The discussion
* goes as follows: compute my distance from the source process PROW so that
* within this process coordinate system, the source process is the process
* such that mydist = 0. Figure out MP.
*/
if( M <= *IMB1 )
{
/*
* M <= IMB1: if I am the source process, i.e. I own I (mydist = 0), MP is M
* and 0 otherwise.
*/
*MP = ( ( MYROW == *PROW ) ? M : 0 );
}
else
{
/*
* Find out how many full blocks are globally (nblocks) and locally (ilocblk)
* in those M entries
*/
nblocks = ( M - *IMB1 ) / mb + 1;
if( MYROW == *PROW )
{
/*
* Since mydist = 0 and nblocks - ilocblk * NPROW >= 0, there are only two
* possible cases:
*
* 1) When mydist = nblocks - ilocblk * NPROW = 0, that is NPROW divides
* the global number of full blocks, then the source process PROW owns
* one more block than the other processes; and M can be rewritten as
* M = IMB1 + (nblocks-1) * NB + LNB with LNB >= 0 size of the last block.
* Similarly, the local value MP corresponding to M can be written as
* MP = IMB1 + (ilocblk-1) * MB + LMB = M + ( ilocblk-1 - (nblocks-1) )*MB.
* Note that this case cannot happen when ilocblk is zero, since nblocks
* is at least one.
*
* 2) mydist = 0 < nblocks - ilocblk * NPROW, the source process only owns
* full blocks, and therefore MP = IMB1 + ilocblk * MB. Note that when
* ilocblk is zero, MP is just IMB1.
*/
if( nblocks < NPROW )
{
*MP = *IMB1;
}
else
{
ilocblk = nblocks / NPROW;
*MP = ( ( nblocks - ilocblk * NPROW ) ?
*IMB1 + ilocblk * mb :
M + ( ilocblk - nblocks ) * mb );
}
}
else
{
/*
* Compute my distance from the source process so that within this process
* coordinate system, the source process is the process such that mydist=0.
*/
if( ( mydist = MYROW - *PROW ) < 0 ) mydist += NPROW;
/*
* When mydist < nblocks - ilocblk * NPROW, I own ilocblk + 1 full blocks of
* size MB since I am not the source process,
*
* when mydist > nblocks - ilocblk * NPROW, I own ilocblk full blocks of
* size MB since I am not the source process,
*
* when mydist = nblocks - ilocblk * NPROW,
* either the last block is not full and I own it, in which case
* M = IMB1 + (nblocks - 1)*MB + LMB with LNB the size of the last block
* such that MB > LMB > 0; the local value MP corresponding to M is given
* by MP = ilocblk * MB + LMB = M - IMB1 + ( ilocblk - nblocks + 1 ) * MB;
* or the last block is full and I am the first process owning only ilocblk
* full blocks of size MB, that is M = IMB + ( nblocks - 1 ) * MB and
* MP = ilocblk * MB = M - IMB + ( ilocblk - nblocks + 1 ) * MB.
*/
if( nblocks < NPROW )
{
mydist -= nblocks;
*MP = ( ( mydist < 0 ) ? mb : ( ( mydist > 0 ) ? 0 :
M - *IMB1 + mb * ( 1 - nblocks ) ) );
}
else
{
ilocblk = nblocks / NPROW;
mydist -= nblocks - ilocblk * NPROW;
*MP = ( ( mydist < 0 ) ? ( ilocblk + 1 ) * mb :
( ( mydist > 0 ) ? ilocblk * mb :
M - *IMB1 + mb * ( ilocblk - nblocks + 1 ) ) );
}
}
}
/*
* Finally figure out IMB1 and RPROW. Note that IMB1 can be zero when M = 0.
*/
*IMB1 = MIN( *IMB1, M );
if( ( *RPROW = MYROW - *PROW ) < 0 ) *RPROW += NPROW;
}
/*
* Idem for the columns
*/
nb = DESC[ NB_ ];
csrc = DESC[ CSRC_ ];
if( ( csrc == -1 ) || ( NPCOL == 1 ) )
{
*JJ = J;
if( ( *INB1 = DESC[INB_] - J ) <= 0 )
*INB1 += ( ( -(*INB1) ) / nb + 1 ) * nb;
*INB1 = MIN( *INB1, N );
*NQ = N;
*PCOL = csrc;
*RPCOL = 0;
}
else
{
*INB1 = DESC[INB_];
if( J < *INB1 )
{
*PCOL = csrc;
*JJ = ( ( MYCOL == *PCOL ) ? J : 0 );
*INB1 -= J;
}
else
{
j1 = J - *INB1;
if( MYCOL == csrc )
{
nblocks = j1 / nb + 1;
*PCOL = csrc + nblocks;
*PCOL -= ( *PCOL / NPCOL ) * NPCOL;
if( nblocks < NPCOL )
{
*JJ = *INB1;
}
else
{
ilocblk = nblocks / NPCOL;
if( ilocblk * NPCOL >= nblocks )
{
*JJ = ( ( MYCOL == *PCOL ) ? J + ( ilocblk - nblocks ) * nb :
*INB1 + ( ilocblk - 1 ) * nb );
}
else
{
*JJ = *INB1 + ilocblk * nb;
}
}
}
else
{
nblocks = j1 / nb + 1;
*PCOL = csrc + nblocks;
*PCOL -= ( *PCOL / NPCOL ) * NPCOL;
if( ( mydist = MYCOL - csrc ) < 0 ) mydist += NPCOL;
if( nblocks < NPCOL )
{
mydist -= nblocks;
*JJ = ( ( mydist < 0 ) ? nb : ( ( MYCOL == *PCOL ) ?
j1 + ( 1 - nblocks ) * nb : 0 ) );
}
else
{
ilocblk = nblocks / NPCOL;
mydist -= nblocks - ilocblk * NPCOL;
*JJ = ( ( mydist < 0 ) ? ( ilocblk + 1 ) * nb :
( ( MYCOL == *PCOL ) ?
( ilocblk - nblocks + 1 ) * nb + j1 :
ilocblk * nb ) );
}
}
*INB1 = nblocks * nb - j1;
}
if( N <= *INB1 )
{
*NQ = ( ( MYCOL == *PCOL ) ? N : 0 );
}
else
{
nblocks = ( N - *INB1 ) / nb + 1;
if( MYCOL == *PCOL )
{
if( nblocks < NPCOL )
{
*NQ = *INB1;
}
else
{
ilocblk = nblocks / NPCOL;
*NQ = ( ( nblocks - ilocblk * NPCOL ) ?
*INB1 + ilocblk * nb :
N + ( ilocblk - nblocks ) * nb );
}
}
else
{
if( ( mydist = MYCOL - *PCOL ) < 0 ) mydist += NPCOL;
if( nblocks < NPCOL )
{
mydist -= nblocks;
*NQ = ( ( mydist < 0 ) ? nb : ( ( mydist > 0 ) ? 0 :
N - *INB1 + nb * ( 1 - nblocks ) ) );
}
else
{
ilocblk = nblocks / NPCOL;
mydist -= nblocks - ilocblk * NPCOL;
*NQ = ( ( mydist < 0 ) ? ( ilocblk + 1 ) * nb :
( ( mydist > 0 ) ? ilocblk * nb :
N - *INB1 + nb * ( ilocblk - nblocks + 1 ) ) );
}
}
}
*INB1 = MIN( *INB1, N );
if( ( *RPCOL = MYCOL - *PCOL ) < 0 ) *RPCOL += NPCOL;
}
/*
* End of PB_Cainfog2l
*/
}
