psblastst.f

Go to the documentation of this file.

      SUBROUTINE psoptee( ICTXT, NOUT, SUBPTR, SCODE, SNAME )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER            ICTXT, NOUT, SCODE
*     ..
*     .. Array Arguments ..
      CHARACTER*(*)      SNAME
*     ..
*     .. Subroutine Arguments ..
      EXTERNAL           subptr
*     ..
*
*  Purpose
*  =======
*
*  PSOPTEE  tests  whether  the  PBLAS respond correctly to a bad option
*  argument.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  NOUT    (global input) INTEGER
*          On entry, NOUT specifies the unit number for the output file.
*          When NOUT is 6, output to screen,  when  NOUT is 0, output to
*          stderr. NOUT is only defined for process 0.
*
*  SUBPTR  (global input) SUBROUTINE
*          On entry,  SUBPTR  is  a  subroutine. SUBPTR must be declared
*          EXTERNAL in the calling subroutine.
*
*  SCODE   (global input) INTEGER
*          On entry, SCODE specifies the calling sequence code.
*
*  SNAME   (global input) CHARACTER*(*)
*          On entry,  SNAME  specifies  the subroutine name calling this
*          subprogram.
*
*  Calling sequence encodings
*  ==========================
*
*  code Formal argument list                                Examples
*
*  11   (n,      v1,v2)                                     _SWAP, _COPY
*  12   (n,s1,   v1   )                                     _SCAL, _SCAL
*  13   (n,s1,   v1,v2)                                     _AXPY, _DOT_
*  14   (n,s1,i1,v1   )                                     _AMAX
*  15   (n,u1,   v1   )                                     _ASUM, _NRM2
*
*  21   (     trans,     m,n,s1,m1,v1,s2,v2)                _GEMV
*  22   (uplo,             n,s1,m1,v1,s2,v2)                _SYMV, _HEMV
*  23   (uplo,trans,diag,  n,   m1,v1      )                _TRMV, _TRSV
*  24   (                m,n,s1,v1,v2,m1)                   _GER_
*  25   (uplo,             n,s1,v1,   m1)                   _SYR
*  26   (uplo,             n,u1,v1,   m1)                   _HER
*  27   (uplo,             n,s1,v1,v2,m1)                   _SYR2, _HER2
*
*  31   (          transa,transb,     m,n,k,s1,m1,m2,s2,m3) _GEMM
*  32   (side,uplo,                   m,n,  s1,m1,m2,s2,m3) _SYMM, _HEMM
*  33   (     uplo,trans,               n,k,s1,m1,   s2,m3) _SYRK
*  34   (     uplo,trans,               n,k,u1,m1,   u2,m3) _HERK
*  35   (     uplo,trans,               n,k,s1,m1,m2,s2,m3) _SYR2K
*  36   (     uplo,trans,               n,k,s1,m1,m2,u2,m3) _HER2K
*  37   (                             m,n,  s1,m1,   s2,m3) _TRAN_
*  38   (side,uplo,transa,       diag,m,n,  s1,m1,m2      ) _TRMM, _TRSM
*  39   (          trans,             m,n,  s1,m1,   s2,m3) _GEADD
*  40   (     uplo,trans,             m,n,  s1,m1,   s2,m3) _TRADD
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Local Scalars ..
      INTEGER             APOS
*     ..
*     .. External Subroutines ..
      EXTERNAL            pschkopt
*     ..
*     .. Executable Statements ..
*
*     Level 2 PBLAS
*
      IF( scode.EQ.21 ) THEN
*
*        Check 1st (and only) option
*
         apos = 1
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'A', apos )
*
      ELSE IF( scode.EQ.22 .OR. scode.EQ.25 .OR. scode.EQ.26 .OR.
     $         scode.EQ.27 ) THEN
*
*        Check 1st (and only) option
*
         apos = 1
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'U', apos )
*
      ELSE IF( scode.EQ.23 ) THEN
*
*        Check 1st option
*
         apos = 1
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'U', apos )
*
*        Check 2nd option
*
         apos = 2
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'A', apos )
*
*        Check 3rd option
*
         apos = 3
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'D', apos )
*
*     Level 3 PBLAS
*
      ELSE IF( scode.EQ.31 ) THEN
*
*        Check 1st option
*
         apos = 1
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'A', apos )
*
*        Check 2'nd option
*
         apos = 2
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'B', apos )
*
      ELSE IF( scode.EQ.32 ) THEN
*
*        Check 1st option
*
         apos = 1
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'S', apos )
*
*        Check 2nd option
*
         apos = 2
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'U', apos )
*
      ELSE IF( scode.EQ.33 .OR. scode.EQ.34 .OR. scode.EQ.35 .OR.
     $         scode.EQ.36 .OR. scode.EQ.40 ) THEN
*
*        Check 1st option
*
         apos = 1
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'U', apos )
*
*        Check 2'nd option
*
         apos = 2
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'A', apos )
*
      ELSE IF( scode.EQ.38 ) THEN
*
*        Check 1st option
*
         apos = 1
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'S', apos )
*
*        Check 2nd option
*
         apos = 2
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'U', apos )
*
*        Check 3rd option
*
         apos = 3
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'A', apos )
*
*        Check 4th option
*
         apos = 4
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'D', apos )
*
*
      ELSE IF( scode.EQ.39 ) THEN
*
*        Check 1st option
*
         apos = 1
         CALL pschkopt( ictxt, nout, subptr, scode, sname, 'A', apos )
*
      END IF
*
      RETURN
*
*     End of PSOPTEE
*
      END
      SUBROUTINE pschkopt( ICTXT, NOUT, SUBPTR, SCODE, SNAME, ARGNAM,
     $                     ARGPOS )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1         ARGNAM
      INTEGER             ARGPOS, ICTXT, NOUT, SCODE
*     ..
*     .. Array Arguments ..
      CHARACTER*(*)       SNAME
*     ..
*     .. Subroutine Arguments ..
      EXTERNAL            subptr
*     ..
*
*  Purpose
*  =======
*
*  PSCHKOPT tests the option ARGNAM in any PBLAS routine.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  NOUT    (global input) INTEGER
*          On entry, NOUT specifies the unit number for the output file.
*          When NOUT is 6, output to screen,  when  NOUT is 0, output to
*          stderr. NOUT is only defined for process 0.
*
*  SUBPTR  (global input) SUBROUTINE
*          On entry,  SUBPTR  is  a  subroutine. SUBPTR must be declared
*          EXTERNAL in the calling subroutine.
*
*  SCODE   (global input) INTEGER
*          On entry, SCODE specifies the calling sequence code.
*
*  SNAME   (global input) CHARACTER*(*)
*          On entry,  SNAME  specifies  the subroutine name calling this
*          subprogram.
*
*  ARGNAM  (global input) CHARACTER*(*)
*          On entry,  ARGNAM  specifies  the  name  of  the option to be
*          checked. ARGNAM can either be 'D', 'S', 'A', 'B', or 'U'.
*
*  ARGPOS  (global input) INTEGER
*          On entry, ARGPOS indicates the position of the option ARGNAM
*          to be tested.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Local Scalars ..
      INTEGER            INFOT
*     ..
*     .. External Subroutines ..
      EXTERNAL           pchkpbe, pscallsub, pssetpblas
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      EXTERNAL           lsame
*     ..
*     .. Common Blocks ..
      CHARACTER          DIAG, SIDE, TRANSA, TRANSB, UPLO
      COMMON             /pblasc/diag, side, transa, transb, uplo
*     ..
*     .. Executable Statements ..
*
*     Reiniatilize the dummy arguments to correct values
*
      CALL pssetpblas( ictxt )
*
      IF( lsame( argnam, 'D' ) ) THEN
*
*        Generate bad DIAG option
*
         diag = '/'
*
      ELSE IF( lsame( argnam, 'S' ) ) THEN
*
*        Generate bad SIDE option
*
         side = '/'
*
      ELSE IF( lsame( argnam, 'A' ) ) THEN
*
*        Generate bad TRANSA option
*
         transa = '/'
*
      ELSE IF( lsame( argnam, 'B' ) ) THEN
*
*        Generate bad TRANSB option
*
         transb = '/'
*
      ELSE IF( lsame( argnam, 'U' ) ) THEN
*
*        Generate bad UPLO option
*
         uplo = '/'
*
      END IF
*
*     Set INFOT to the position of the bad dimension argument
*
      infot = argpos
*
*     Call the PBLAS routine
*
      CALL pscallsub( subptr, scode )
      CALL pchkpbe( ictxt, nout, sname, infot )
*
      RETURN
*
*     End of PSCHKOPT
*
      END
      SUBROUTINE psdimee( ICTXT, NOUT, SUBPTR, SCODE, SNAME )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER            ICTXT, NOUT, SCODE
*     ..
*     .. Array Arguments ..
      CHARACTER*(*)      SNAME
*     ..
*     .. Subroutine Arguments ..
      EXTERNAL           subptr
*     ..
*
*  Purpose
*  =======
*
*  PSDIMEE  tests whether the PBLAS respond correctly to a bad dimension
*  argument.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  NOUT    (global input) INTEGER
*          On entry, NOUT specifies the unit number for the output file.
*          When NOUT is 6, output to screen,  when  NOUT is 0, output to
*          stderr. NOUT is only defined for process 0.
*
*  SUBPTR  (global input) SUBROUTINE
*          On entry,  SUBPTR  is  a  subroutine. SUBPTR must be declared
*          EXTERNAL in the calling subroutine.
*
*  SCODE   (global input) INTEGER
*          On entry, SCODE specifies the calling sequence code.
*
*  SNAME   (global input) CHARACTER*(*)
*          On entry,  SNAME  specifies  the subroutine name calling this
*          subprogram.
*
*  Calling sequence encodings
*  ==========================
*
*  code Formal argument list                                Examples
*
*  11   (n,      v1,v2)                                     _SWAP, _COPY
*  12   (n,s1,   v1   )                                     _SCAL, _SCAL
*  13   (n,s1,   v1,v2)                                     _AXPY, _DOT_
*  14   (n,s1,i1,v1   )                                     _AMAX
*  15   (n,u1,   v1   )                                     _ASUM, _NRM2
*
*  21   (     trans,     m,n,s1,m1,v1,s2,v2)                _GEMV
*  22   (uplo,             n,s1,m1,v1,s2,v2)                _SYMV, _HEMV
*  23   (uplo,trans,diag,  n,   m1,v1      )                _TRMV, _TRSV
*  24   (                m,n,s1,v1,v2,m1)                   _GER_
*  25   (uplo,             n,s1,v1,   m1)                   _SYR
*  26   (uplo,             n,u1,v1,   m1)                   _HER
*  27   (uplo,             n,s1,v1,v2,m1)                   _SYR2, _HER2
*
*  31   (          transa,transb,     m,n,k,s1,m1,m2,s2,m3) _GEMM
*  32   (side,uplo,                   m,n,  s1,m1,m2,s2,m3) _SYMM, _HEMM
*  33   (     uplo,trans,               n,k,s1,m1,   s2,m3) _SYRK
*  34   (     uplo,trans,               n,k,u1,m1,   u2,m3) _HERK
*  35   (     uplo,trans,               n,k,s1,m1,m2,s2,m3) _SYR2K
*  36   (     uplo,trans,               n,k,s1,m1,m2,u2,m3) _HER2K
*  37   (                             m,n,  s1,m1,   s2,m3) _TRAN_
*  38   (side,uplo,transa,       diag,m,n,  s1,m1,m2      ) _TRMM, _TRSM
*  39   (          trans,             m,n,  s1,m1,   s2,m3) _GEADD
*  40   (     uplo,trans,             m,n,  s1,m1,   s2,m3) _TRADD
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Local Scalars ..
      INTEGER             APOS
*     ..
*     .. External Subroutines ..
      EXTERNAL            pschkdim
*     ..
*     .. Executable Statements ..
*
*     Level 1 PBLAS
*
      IF( scode.EQ.11 .OR. scode.EQ.12 .OR. scode.EQ.13 .OR.
     $    scode.EQ.14 .OR. scode.EQ.15 ) THEN
*
*        Check 1st (and only) dimension
*
         apos = 1
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'N', apos )
*
*     Level 2 PBLAS
*
      ELSE IF( scode.EQ.21 ) THEN
*
*        Check 1st dimension
*
         apos = 2
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'M', apos )
*
*        Check 2nd dimension
*
         apos = 3
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'N', apos )
*
      ELSE IF( scode.EQ.22 .OR. scode.EQ.25 .OR. scode.EQ.26 .OR.
     $         scode.EQ.27 ) THEN
*
*        Check 1st (and only) dimension
*
         apos = 2
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'N', apos )
*
      ELSE IF( scode.EQ.23 ) THEN
*
*        Check 1st (and only) dimension
*
         apos = 4
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'N', apos )
*
      ELSE IF( scode.EQ.24 ) THEN
*
*        Check 1st dimension
*
         apos = 1
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'M', apos )
*
*        Check 2nd dimension
*
         apos = 2
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'N', apos )
*
*     Level 3 PBLAS
*
      ELSE IF( scode.EQ.31 ) THEN
*
*        Check 1st dimension
*
         apos = 3
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'M', apos )
*
*        Check 2nd dimension
*
         apos = 4
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'N', apos )
*
*        Check 3rd dimension
*
         apos = 5
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'K', apos )
*
      ELSE IF( scode.EQ.32 ) THEN
*
*        Check 1st dimension
*
         apos = 3
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'M', apos )
*
*        Check 2nd dimension
*
         apos = 4
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'N', apos )
*
      ELSE IF( scode.EQ.33 .OR. scode.EQ.34 .OR. scode.EQ.35 .OR.
     $         scode.EQ.36 ) THEN
*
*        Check 1st dimension
*
         apos = 3
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'N', apos )
*
*        Check 2nd dimension
*
         apos = 4
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'K', apos )
*
      ELSE IF( scode.EQ.37 ) THEN
*
*        Check 1st dimension
*
         apos = 1
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'M', apos )
*
*        Check 2nd dimension
*
         apos = 2
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'N', apos )
*
      ELSE IF( scode.EQ.38 ) THEN
*
*        Check 1st dimension
*
         apos = 5
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'M', apos )
*
*        Check 2nd dimension
*
         apos = 6
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'N', apos )
*
      ELSE IF( scode.EQ.39 ) THEN
*
*        Check 1st dimension
*
         apos = 2
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'M', apos )
*
*        Check 2nd dimension
*
         apos = 3
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'N', apos )
*
      ELSE IF( scode.EQ.40 ) THEN
*
*        Check 1st dimension
*
         apos = 3
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'M', apos )
*
*        Check 2nd dimension
*
         apos = 4
         CALL pschkdim( ictxt, nout, subptr, scode, sname, 'N', apos )
*
      END IF
*
      RETURN
*
*     End of PSDIMEE
*
      END
      SUBROUTINE pschkdim( ICTXT, NOUT, SUBPTR, SCODE, SNAME, ARGNAM,
     $                     ARGPOS )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1         ARGNAM
      INTEGER             ARGPOS, ICTXT, NOUT, SCODE
*     ..
*     .. Array Arguments ..
      CHARACTER*(*)       SNAME
*     ..
*     .. Subroutine Arguments ..
      EXTERNAL            subptr
*     ..
*
*  Purpose
*  =======
*
*  PSCHKDIM tests the dimension ARGNAM in any PBLAS routine.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  NOUT    (global input) INTEGER
*          On entry, NOUT specifies the unit number for the output file.
*          When NOUT is 6, output to screen,  when  NOUT is 0, output to
*          stderr. NOUT is only defined for process 0.
*
*  SUBPTR  (global input) SUBROUTINE
*          On entry,  SUBPTR  is  a  subroutine. SUBPTR must be declared
*          EXTERNAL in the calling subroutine.
*
*  SCODE   (global input) INTEGER
*          On entry, SCODE specifies the calling sequence code.
*
*  SNAME   (global input) CHARACTER*(*)
*          On entry,  SNAME  specifies  the subroutine name calling this
*          subprogram.
*
*  ARGNAM  (global input) CHARACTER*(*)
*          On entry,  ARGNAM  specifies  the name of the dimension to be
*          checked. ARGNAM can either be 'M', 'N' or 'K'.
*
*  ARGPOS  (global input) INTEGER
*          On entry, ARGPOS indicates the position of the option ARGNAM
*          to be tested.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Local Scalars ..
      INTEGER            INFOT
*     ..
*     .. External Subroutines ..
      EXTERNAL           pchkpbe, pscallsub, pssetpblas
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      EXTERNAL           LSAME
*     ..
*     .. Common Blocks ..
      INTEGER            KDIM, MDIM, NDIM
      COMMON             /PBLASN/KDIM, MDIM, NDIM
*     ..
*     .. Executable Statements ..
*
*     Reiniatilize the dummy arguments to correct values
*
      CALL pssetpblas( ictxt )
*
      IF( lsame( argnam, 'M' ) ) THEN
*
*        Generate bad MDIM
*
         mdim = -1
*
      ELSE IF( lsame( argnam, 'N' ) ) THEN
*
*        Generate bad NDIM
*
         ndim = -1
*
      ELSE
*
*        Generate bad KDIM
*
         kdim = -1
*
      END IF
*
*     Set INFOT to the position of the bad dimension argument
*
      infot = argpos
*
*     Call the PBLAS routine
*
      CALL pscallsub( subptr, scode )
      CALL pchkpbe( ictxt, nout, sname, infot )
*
      RETURN
*
*     End of PSCHKDIM
*
      END
      SUBROUTINE psvecee( ICTXT, NOUT, SUBPTR, SCODE, SNAME )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER             ICTXT, NOUT, SCODE
*     ..
*     .. Array Arguments ..
      CHARACTER*7         SNAME
*     ..
*     .. Subroutine Arguments ..
      EXTERNAL            subptr
*     ..
*
*  Purpose
*  =======
*
*  PSVECEE  tests  whether  the  PBLAS respond correctly to a bad vector
*  argument.  Each  vector <vec> is described by: <vec>, I<vec>, J<vec>,
*  DESC<vec>,  INC<vec>.   Out   of  all  these,  only  I<vec>,  J<vec>,
*  DESC<vec>, and INC<vec> can be tested.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  NOUT    (global input) INTEGER
*          On entry, NOUT specifies the unit number for the output file.
*          When NOUT is 6, output to screen,  when  NOUT is 0, output to
*          stderr. NOUT is only defined for process 0.
*
*  SUBPTR  (global input) SUBROUTINE
*          On entry,  SUBPTR  is  a  subroutine. SUBPTR must be declared
*          EXTERNAL in the calling subroutine.
*
*  SCODE   (global input) INTEGER
*          On entry, SCODE specifies the calling sequence code.
*
*  SNAME   (global input) CHARACTER*(*)
*          On entry,  SNAME  specifies  the subroutine name calling this
*          subprogram.
*
*  Calling sequence encodings
*  ==========================
*
*  code Formal argument list                                Examples
*
*  11   (n,      v1,v2)                                     _SWAP, _COPY
*  12   (n,s1,   v1   )                                     _SCAL, _SCAL
*  13   (n,s1,   v1,v2)                                     _AXPY, _DOT_
*  14   (n,s1,i1,v1   )                                     _AMAX
*  15   (n,u1,   v1   )                                     _ASUM, _NRM2
*
*  21   (     trans,     m,n,s1,m1,v1,s2,v2)                _GEMV
*  22   (uplo,             n,s1,m1,v1,s2,v2)                _SYMV, _HEMV
*  23   (uplo,trans,diag,  n,   m1,v1      )                _TRMV, _TRSV
*  24   (                m,n,s1,v1,v2,m1)                   _GER_
*  25   (uplo,             n,s1,v1,   m1)                   _SYR
*  26   (uplo,             n,u1,v1,   m1)                   _HER
*  27   (uplo,             n,s1,v1,v2,m1)                   _SYR2, _HER2
*
*  31   (          transa,transb,     m,n,k,s1,m1,m2,s2,m3) _GEMM
*  32   (side,uplo,                   m,n,  s1,m1,m2,s2,m3) _SYMM, _HEMM
*  33   (     uplo,trans,               n,k,s1,m1,   s2,m3) _SYRK
*  34   (     uplo,trans,               n,k,u1,m1,   u2,m3) _HERK
*  35   (     uplo,trans,               n,k,s1,m1,m2,s2,m3) _SYR2K
*  36   (     uplo,trans,               n,k,s1,m1,m2,u2,m3) _HER2K
*  37   (                             m,n,  s1,m1,   s2,m3) _TRAN_
*  38   (side,uplo,transa,       diag,m,n,  s1,m1,m2      ) _TRMM, _TRSM
*  39   (          trans,             m,n,  s1,m1,   s2,m3) _GEADD
*  40   (     uplo,trans,             m,n,  s1,m1,   s2,m3) _TRADD
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Local Scalars ..
      INTEGER             APOS
*     ..
*     .. External Subroutines ..
      EXTERNAL            pschkmat
*     ..
*     .. Executable Statements ..
*
*     Level 1 PBLAS
*
      IF( scode.EQ.11 ) THEN
*
*        Check 1st vector
*
         apos = 2
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'X', apos )
*
*        Check 2nd vector
*
         apos = 7
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'Y', apos )
*
      ELSE IF( scode.EQ.12 .OR. scode.EQ.15 ) THEN
*
*        Check 1st (and only) vector
*
         apos = 3
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'X', apos )
*
      ELSE IF( scode.EQ.13 ) THEN
*
*        Check 1st vector
*
         apos = 3
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'X', apos )
*
*        Check 2nd vector
*
         apos = 8
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'Y', apos )
*
      ELSE IF( scode.EQ.14 ) THEN
*
*        Check 1st (and only) vector
*
         apos = 4
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'X', apos )
*
*     Level 2 PBLAS
*
      ELSE IF( scode.EQ.21 ) THEN
*
*        Check 1st vector
*
         apos = 9
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'X', apos )
*
*        Check 2nd vector
*
         apos = 15
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'Y', apos )
*
      ELSE IF( scode.EQ.22 ) THEN
*
*        Check 1st vector
*
         apos = 8
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'X', apos )
*
*        Check 2nd vector
*
         apos = 14
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'Y', apos )
*
      ELSE IF( scode.EQ.23 ) THEN
*
*        Check 1st (and only) vector
*
         apos = 9
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'X', apos )
*
      ELSE IF( scode.EQ.24 .OR. scode.EQ.27 ) THEN
*
*        Check 1st vector
*
         apos = 4
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'X', apos )
*
*        Check 2nd vector
*
         apos = 9
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'Y', apos )
*
      ELSE IF( scode.EQ.26 .OR. scode.EQ.27 ) THEN
*
*        Check 1'st (and only) vector
*
         apos = 4
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'X', apos )
*
      END IF
*
      RETURN
*
*     End of PSVECEE
*
      END
      SUBROUTINE psmatee( ICTXT, NOUT, SUBPTR, SCODE, SNAME )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER             ICTXT, NOUT, SCODE
*     ..
*     .. Array Arguments ..
      CHARACTER*7         SNAME
*     ..
*     .. Subroutine Arguments ..
      EXTERNAL            subptr
*     ..
*
*  Purpose
*  =======
*
*  PSMATEE  tests  whether  the  PBLAS respond correctly to a bad matrix
*  argument.  Each  matrix <mat> is described by: <mat>, I<mat>, J<mat>,
*  and DESC<mat>.  Out  of  all these, only I<vec>, J<vec> and DESC<mat>
*  can be tested.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  NOUT    (global input) INTEGER
*          On entry, NOUT specifies the unit number for the output file.
*          When NOUT is 6, output to screen,  when  NOUT is 0, output to
*          stderr. NOUT is only defined for process 0.
*
*  SUBPTR  (global input) SUBROUTINE
*          On entry,  SUBPTR  is  a  subroutine. SUBPTR must be declared
*          EXTERNAL in the calling subroutine.
*
*  SCODE   (global input) INTEGER
*          On entry, SCODE specifies the calling sequence code.
*
*  SNAME   (global input) CHARACTER*(*)
*          On entry,  SNAME  specifies  the subroutine name calling this
*          subprogram.
*
*  Calling sequence encodings
*  ==========================
*
*  code Formal argument list                                Examples
*
*  11   (n,      v1,v2)                                     _SWAP, _COPY
*  12   (n,s1,   v1   )                                     _SCAL, _SCAL
*  13   (n,s1,   v1,v2)                                     _AXPY, _DOT_
*  14   (n,s1,i1,v1   )                                     _AMAX
*  15   (n,u1,   v1   )                                     _ASUM, _NRM2
*
*  21   (     trans,     m,n,s1,m1,v1,s2,v2)                _GEMV
*  22   (uplo,             n,s1,m1,v1,s2,v2)                _SYMV, _HEMV
*  23   (uplo,trans,diag,  n,   m1,v1      )                _TRMV, _TRSV
*  24   (                m,n,s1,v1,v2,m1)                   _GER_
*  25   (uplo,             n,s1,v1,   m1)                   _SYR
*  26   (uplo,             n,u1,v1,   m1)                   _HER
*  27   (uplo,             n,s1,v1,v2,m1)                   _SYR2, _HER2
*
*  31   (          transa,transb,     m,n,k,s1,m1,m2,s2,m3) _GEMM
*  32   (side,uplo,                   m,n,  s1,m1,m2,s2,m3) _SYMM, _HEMM
*  33   (     uplo,trans,               n,k,s1,m1,   s2,m3) _SYRK
*  34   (     uplo,trans,               n,k,u1,m1,   u2,m3) _HERK
*  35   (     uplo,trans,               n,k,s1,m1,m2,s2,m3) _SYR2K
*  36   (     uplo,trans,               n,k,s1,m1,m2,u2,m3) _HER2K
*  37   (                             m,n,  s1,m1,   s2,m3) _TRAN_
*  38   (side,uplo,transa,       diag,m,n,  s1,m1,m2      ) _TRMM, _TRSM
*  39   (          trans,             m,n,  s1,m1,   s2,m3) _GEADD
*  40   (     uplo,trans,             m,n,  s1,m1,   s2,m3) _TRADD
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Local Scalars ..
      INTEGER             APOS
*     ..
*     .. External Subroutines ..
      EXTERNAL            pschkmat
*     ..
*     .. Executable Statements ..
*
*     Level 2 PBLAS
*
      IF( scode.EQ.21 .OR. scode.EQ.23 ) THEN
*
*        Check 1st (and only) matrix
*
         apos = 5
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'A', apos )
*
      ELSE IF( scode.EQ.22 ) THEN
*
*        Check 1st (and only) matrix
*
         apos = 4
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'A', apos )
*
      ELSE IF( scode.EQ.24 .OR. scode.EQ.27 ) THEN
*
*        Check 1st (and only) matrix
*
         apos = 14
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'A', apos )
*
      ELSE IF( scode.EQ.25 .OR. scode.EQ.26 ) THEN
*
*        Check 1st (and only) matrix
*
         apos = 9
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'A', apos )
*
*     Level 3 PBLAS
*
      ELSE IF( scode.EQ.31 ) THEN
*
*        Check 1st matrix
*
         apos = 7
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'A', apos )
*
*        Check 2nd matrix
*
         apos = 11
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'B', apos )
*
*        Check 3nd matrix
*
         apos = 16
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'C', apos )
*
      ELSE IF( scode.EQ.32 .OR. scode.EQ.35 .OR. scode.EQ.36 ) THEN
*
*        Check 1st matrix
*
         apos = 6
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'A', apos )
*
*        Check 2nd matrix
*
         apos = 10
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'B', apos )
*
*        Check 3nd matrix
*
         apos = 15
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'C', apos )
*
      ELSE IF( scode.EQ.33 .OR. scode.EQ.34 ) THEN
*
*        Check 1st matrix
*
         apos = 6
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'A', apos )
*
*        Check 2nd matrix
*
         apos = 11
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'C', apos )
*
      ELSE IF( scode.EQ.37 ) THEN
*
*        Check 1st matrix
*
         apos = 4
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'A', apos )
*
*        Check 2nd matrix
*
         apos = 9
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'C', apos )
*
      ELSE IF( scode.EQ.38 ) THEN
*
*        Check 1st matrix
*
         apos = 8
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'A', apos )
*
*        Check 2nd matrix
*
         apos = 12
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'B', apos )
*
      ELSE IF( scode.EQ.39 ) THEN
*
*        Check 1st matrix
*
         apos = 5
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'A', apos )
*
*        Check 2nd matrix
*
         apos = 10
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'C', apos )
*
      ELSE IF( scode.EQ.40 ) THEN
*
*        Check 1st matrix
*
         apos = 6
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'A', apos )
*
*        Check 2nd matrix
*
         apos = 11
         CALL pschkmat( ictxt, nout, subptr, scode, sname, 'C', apos )
*
      END IF
*
      RETURN
*
*     End of PSMATEE
*
      END
      SUBROUTINE pssetpblas( ICTXT )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER            ICTXT
*     ..
*
*  Purpose
*  =======
*
*  PSSETPBLAS initializes *all* the dummy arguments to correct values.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   rsrc_
      parameter( block_cyclic_2d_inb = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
      REAL               ONE
      PARAMETER          ( ONE = 1.0e+0 )
*     ..
*     .. External Subroutines ..
      EXTERNAL           pb_descset2
*     ..
*     .. Common Blocks ..
      CHARACTER*1        DIAG, SIDE, TRANSA, TRANSB, UPLO
      INTEGER            IA, IB, IC, INCX, INCY, ISCLR, IX, IY, JA, JB,
     $                   jc, jx, jy, kdim, mdim, ndim
      REAL               USCLR, SCLR
      INTEGER            DESCA( DLEN_ ), DESCB( DLEN_ ), DESCC( DLEN_ ),
     $                   descx( dlen_ ), descy( dlen_ )
      REAL               A( 2, 2 ), B( 2, 2 ), C( 2, 2 ), X( 2 ), Y( 2 )
      COMMON             /PBLASC/DIAG, SIDE, TRANSA, TRANSB, UPLO
      COMMON             /pblasd/desca, descb, descc, descx, descy
      COMMON             /pblasi/ia, ib, ic, incx, incy, isclr, ix, iy,
     $                   ja, jb, jc, jx, jy
      COMMON             /pblasm/a, b, c
      COMMON             /pblasn/kdim, mdim, ndim
      COMMON             /pblass/sclr, usclr
      COMMON             /pblasv/x, y
*     ..
*     .. Executable Statements ..
*
*     Set default values for options
*
      diag   = 'N'
      side   = 'L'
      transa = 'N'
      transb = 'N'
      uplo   = 'U'
*
*     Set default values for scalars
*
      kdim   = 1
      mdim   = 1
      ndim   = 1
      isclr  = 1
      sclr   = one
      usclr  = one
*
*     Set default values for distributed matrix A
*
      a( 1, 1 ) = one
      a( 2, 1 ) = one
      a( 1, 2 ) = one
      a( 2, 2 ) = one
      ia = 1
      ja = 1
      CALL pb_descset2( desca, 2, 2, 1, 1, 1, 1, 0, 0, ictxt, 2 )
*
*     Set default values for distributed matrix B
*
      b( 1, 1 ) = one
      b( 2, 1 ) = one
      b( 1, 2 ) = one
      b( 2, 2 ) = one
      ib = 1
      jb = 1
      CALL pb_descset2( descb, 2, 2, 1, 1, 1, 1, 0, 0, ictxt, 2 )
*
*     Set default values for distributed matrix C
*
      c( 1, 1 ) = one
      c( 2, 1 ) = one
      c( 1, 2 ) = one
      c( 2, 2 ) = one
      ic = 1
      jc = 1
      CALL pb_descset2( descc, 2, 2, 1, 1, 1, 1, 0, 0, ictxt, 2 )
*
*     Set default values for distributed matrix X
*
      x( 1 ) = one
      x( 2 ) = one
      ix = 1
      jx = 1
      CALL pb_descset2( descx, 2, 1, 1, 1, 1, 1, 0, 0, ictxt, 2 )
      incx = 1
*
*     Set default values for distributed matrix Y
*
      y( 1 ) = one
      y( 2 ) = one
      iy = 1
      jy = 1
      CALL pb_descset2( descy, 2, 1, 1, 1, 1, 1, 0, 0, ictxt, 2 )
      incy = 1
*
      RETURN
*
*     End of PSSETPBLAS
*
      END
      SUBROUTINE pschkmat( ICTXT, NOUT, SUBPTR, SCODE, SNAME, ARGNAM,
     $                     ARGPOS )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1         ARGNAM
      INTEGER             ARGPOS, ICTXT, NOUT, SCODE
*     ..
*     .. Array Arguments ..
      CHARACTER*(*)       SNAME
*     ..
*     .. Subroutine Arguments ..
      EXTERNAL            subptr
*     ..
*
*  Purpose
*  =======
*
*  PSCHKMAT tests the matrix (or vector) ARGNAM in any PBLAS routine.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  NOUT    (global input) INTEGER
*          On entry, NOUT specifies the unit number for the output file.
*          When NOUT is 6, output to screen,  when  NOUT is 0, output to
*          stderr. NOUT is only defined for process 0.
*
*  SUBPTR  (global input) SUBROUTINE
*          On entry,  SUBPTR  is  a  subroutine. SUBPTR must be declared
*          EXTERNAL in the calling subroutine.
*
*  SCODE   (global input) INTEGER
*          On entry, SCODE specifies the calling sequence code.
*
*  SNAME   (global input) CHARACTER*(*)
*          On entry,  SNAME  specifies  the subroutine name calling this
*          subprogram.
*
*  ARGNAM  (global input) CHARACTER*(*)
*          On entry,  ARGNAM  specifies the name of the matrix or vector
*          to be checked.  ARGNAM can either be 'A', 'B' or 'C' when one
*          wants to check a matrix, and 'X' or 'Y' for a vector.
*
*  ARGPOS  (global input) INTEGER
*          On entry, ARGPOS indicates the position of the first argument
*          of the matrix (or vector) ARGNAM.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      parameter( block_cyclic_2d_inb = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
      INTEGER             DESCMULT
      PARAMETER           ( DESCMULT = 100 )
*     ..
*     .. Local Scalars ..
      INTEGER             I, INFOT, NPROW, NPCOL, MYROW, MYCOL
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_gridinfo, pchkpbe, pscallsub, pssetpblas
*     ..
*     .. External Functions ..
      LOGICAL             LSAME
      EXTERNAL            LSAME
*     ..
*     .. Common Blocks ..
      INTEGER            IA, IB, IC, INCX, INCY, ISCLR, IX, IY, JA, JB,
     $                   JC, JX, JY
      INTEGER            DESCA( DLEN_ ), DESCB( DLEN_ ), DESCC( DLEN_ ),
     $                   descx( dlen_ ), descy( dlen_ )
      COMMON             /pblasd/desca, descb, descc, descx, descy
      COMMON             /pblasi/ia, ib, ic, incx, incy, isclr, ix, iy,
     $                   ja, jb, jc, jx, jy
*     ..
*     .. Executable Statements ..
*
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      IF( lsame( argnam, 'A' ) ) THEN
*
*        Check IA. Set all other OK, bad IA
*
         CALL pssetpblas( ictxt )
         ia    = -1
         infot = argpos + 1
         CALL pscallsub( subptr, scode )
         CALL pchkpbe( ictxt, nout, sname, infot )
*
*        Check JA. Set all other OK, bad JA
*
         CALL pssetpblas( ictxt )
         ja    = -1
         infot = argpos + 2
         CALL pscallsub( subptr, scode )
         CALL pchkpbe( ictxt, nout, sname, infot )
*
*        Check DESCA. Set all other OK, bad DESCA
*
         DO 10 i = 1, dlen_
*
*           Set I'th entry of DESCA to incorrect value, rest ok.
*
            CALL pssetpblas( ictxt )
            desca( i ) =  -2
            infot = ( ( argpos + 3 ) * descmult ) + i
            CALL pscallsub( subptr, scode )
            CALL pchkpbe( ictxt, nout, sname, infot )
*
*           Extra tests for RSRCA, CSRCA, LDA
*
            IF( ( i.EQ.rsrc_ ) .OR. ( i.EQ.csrc_ ) .OR.
     $          ( i.EQ.lld_ ) ) THEN
*
               CALL pssetpblas( ictxt )
*
*              Test RSRCA >= NPROW
*
               IF( i.EQ.rsrc_ )
     $            desca( i ) =  nprow
*
*              Test CSRCA >= NPCOL
*
               IF( i.EQ.csrc_ )
     $            desca( i ) =  npcol
*
*              Test LDA >= MAX(1, PB_NUMROC(...)). Set to 1 as mat 2x2.
*
               IF( i.EQ.lld_ ) THEN
                  IF( myrow.EQ.0 .AND.mycol.EQ.0 ) THEN
                     desca( i ) = 1
                  ELSE
                     desca( i ) = 0
                  END IF
               END IF
*
               infot = ( ( argpos + 3 ) * descmult ) + i
               CALL pscallsub( subptr, scode )
               CALL pchkpbe( ictxt, nout, sname, infot )
*
            END IF
*
   10    CONTINUE
*
      ELSE IF( lsame( argnam, 'B' ) ) THEN
*
*        Check IB. Set all other OK, bad IB
*
         CALL pssetpblas( ictxt )
         ib    = -1
         infot = argpos + 1
         CALL pscallsub( subptr, scode )
         CALL pchkpbe( ictxt, nout, sname, infot )
*
*        Check JB. Set all other OK, bad JB
*
         CALL pssetpblas( ictxt )
         jb    = -1
         infot = argpos + 2
         CALL pscallsub( subptr, scode )
         CALL pchkpbe( ictxt, nout, sname, infot )
*
*        Check DESCB. Set all other OK, bad DESCB
*
         DO 20 i = 1, dlen_
*
*           Set I'th entry of DESCB to incorrect value, rest ok.
*
            CALL pssetpblas( ictxt )
            descb( i ) =  -2
            infot = ( ( argpos + 3 ) * descmult ) + i
            CALL pscallsub( subptr, scode )
            CALL pchkpbe( ictxt, nout, sname, infot )
*
*           Extra tests for RSRCB, CSRCB, LDB
*
            IF( ( i.EQ.rsrc_ ) .OR. ( i.EQ.csrc_ ) .OR.
     $          ( i.EQ.lld_ ) ) THEN
*
               CALL pssetpblas( ictxt )
*
*              Test RSRCB >= NPROW
*
               IF( i.EQ.rsrc_ )
     $            descb( i ) =  nprow
*
*              Test CSRCB >= NPCOL
*
               IF( i.EQ.csrc_ )
     $            descb( i ) =  npcol
*
*              Test LDB >= MAX(1, PB_NUMROC(...)). Set to 1 as mat 2x2.
*
               IF( i.EQ.lld_ ) THEN
                  IF( myrow.EQ.0 .AND.mycol.EQ.0 ) THEN
                     descb( i ) = 1
                  ELSE
                     descb( i ) = 0
                  END IF
               END IF
*
               infot = ( ( argpos + 3 ) * descmult ) + i
               CALL pscallsub( subptr, scode )
               CALL pchkpbe( ictxt, nout, sname, infot )
*
            END IF
*
   20    CONTINUE
*
      ELSE IF( lsame( argnam, 'C' ) ) THEN
*
*        Check IC. Set all other OK, bad IC
*
         CALL pssetpblas( ictxt )
         ic    = -1
         infot = argpos + 1
         CALL pscallsub( subptr, scode )
         CALL pchkpbe( ictxt, nout, sname, infot )
*
*        Check JC. Set all other OK, bad JC
*
         CALL pssetpblas( ictxt )
         jc    = -1
         infot = argpos + 2
         CALL pscallsub( subptr, scode )
         CALL pchkpbe( ictxt, nout, sname, infot )
*
*        Check DESCC. Set all other OK, bad DESCC
*
         DO 30 i = 1, dlen_
*
*           Set I'th entry of DESCC to incorrect value, rest ok.
*
            CALL pssetpblas( ictxt )
            descc( i ) =  -2
            infot = ( ( argpos + 3 ) * descmult ) + i
            CALL pscallsub( subptr, scode )
            CALL pchkpbe( ictxt, nout, sname, infot )
*
*           Extra tests for RSRCC, CSRCC, LDC
*
            IF( ( i.EQ.rsrc_ ) .OR. ( i.EQ.csrc_ ) .OR.
     $          ( i.EQ.lld_ ) ) THEN
*
               CALL pssetpblas( ictxt )
*
*              Test RSRCC >= NPROW
*
               IF( i.EQ.rsrc_ )
     $            descc( i ) =  nprow
*
*              Test CSRCC >= NPCOL
*
               IF( i.EQ.csrc_ )
     $            descc( i ) =  npcol
*
*              Test LDC >= MAX(1, PB_NUMROC(...)). Set to 1 as mat 2x2.
*
               IF( i.EQ.lld_ ) THEN
                  IF( myrow.EQ.0 .AND.mycol.EQ.0 ) THEN
                     descc( i ) = 1
                  ELSE
                     descc( i ) = 0
                  END IF
               END IF
*
               infot = ( ( argpos + 3 ) * descmult ) + i
               CALL pscallsub( subptr, scode )
               CALL pchkpbe( ictxt, nout, sname, infot )
*
            END IF
*
   30    CONTINUE
*
      ELSE IF( lsame( argnam, 'X' ) ) THEN
*
*        Check IX. Set all other OK, bad IX
*
         CALL pssetpblas( ictxt )
         ix    = -1
         infot = argpos + 1
         CALL pscallsub( subptr, scode )
         CALL pchkpbe( ictxt, nout, sname, infot )
*
*        Check JX. Set all other OK, bad JX
*
         CALL pssetpblas( ictxt )
         jx    = -1
         infot = argpos + 2
         CALL pscallsub( subptr, scode )
         CALL pchkpbe( ictxt, nout, sname, infot )
*
*        Check DESCX. Set all other OK, bad DESCX
*
         DO 40 i = 1, dlen_
*
*           Set I'th entry of DESCX to incorrect value, rest ok.
*
            CALL pssetpblas( ictxt )
            descx( i ) =  -2
            infot = ( ( argpos + 3 ) * descmult ) + i
            CALL pscallsub( subptr, scode )
            CALL pchkpbe( ictxt, nout, sname, infot )
*
*           Extra tests for RSRCX, CSRCX, LDX
*
            IF( ( i.EQ.rsrc_ ) .OR. ( i.EQ.csrc_ ) .OR.
     $          ( i.EQ.lld_ ) ) THEN
*
               CALL pssetpblas( ictxt )
*
*              Test RSRCX >= NPROW
*
               IF( i.EQ.rsrc_ )
     $            descx( i ) =  nprow
*
*              Test CSRCX >= NPCOL
*
               IF( i.EQ.csrc_ )
     $            descx( i ) =  npcol
*
*              Test LDX >= MAX(1, PB_NUMROC(...)). Set to 1 as mat 2x2.
*
               IF( i.EQ.lld_ ) THEN
                  IF( myrow.EQ.0 .AND.mycol.EQ.0 ) THEN
                     descx( i ) = 1
                  ELSE
                     descx( i ) = 0
                  END IF
               END IF
*
               infot = ( ( argpos + 3 ) * descmult ) + i
               CALL pscallsub( subptr, scode )
               CALL pchkpbe( ictxt, nout, sname, infot )
*
            END IF
*
   40    CONTINUE
*
*        Check INCX. Set all other OK, bad INCX
*
         CALL pssetpblas( ictxt )
         incx  =  -1
         infot = argpos + 4
         CALL pscallsub( subptr, scode )
         CALL pchkpbe( ictxt, nout, sname, infot )
*
      ELSE
*
*        Check IY. Set all other OK, bad IY
*
         CALL pssetpblas( ictxt )
         iy    = -1
         infot = argpos + 1
         CALL pscallsub( subptr, scode )
         CALL pchkpbe( ictxt, nout, sname, infot )
*
*        Check JY. Set all other OK, bad JY
*
         CALL pssetpblas( ictxt )
         jy    = -1
         infot = argpos + 2
         CALL pscallsub( subptr, scode )
         CALL pchkpbe( ictxt, nout, sname, infot )
*
*        Check DESCY. Set all other OK, bad DESCY
*
         DO 50 i = 1, dlen_
*
*           Set I'th entry of DESCY to incorrect value, rest ok.
*
            CALL pssetpblas( ictxt )
            descy( i ) =  -2
            infot = ( ( argpos + 3 ) * descmult ) + i
            CALL pscallsub( subptr, scode )
            CALL pchkpbe( ictxt, nout, sname, infot )
*
*           Extra tests for RSRCY, CSRCY, LDY
*
            IF( ( i.EQ.rsrc_ ) .OR. ( i.EQ.csrc_ ) .OR.
     $          ( i.EQ.lld_ ) ) THEN
*
               CALL pssetpblas( ictxt )
*
*              Test RSRCY >= NPROW
*
               IF( i.EQ.rsrc_ )
     $            descy( i ) = nprow
*
*              Test CSRCY >= NPCOL
*
               IF( i.EQ.csrc_ )
     $            descy( i ) = npcol
*
*              Test LDY >= MAX(1, PB_NUMROC(...)). Set to 1 as mat 2x2.
*
               IF( i.EQ.lld_ ) THEN
                  IF( myrow.EQ.0 .AND.mycol.EQ.0 ) THEN
                     descy( i ) = 1
                  ELSE
                     descy( i ) = 0
                  END IF
               END IF
*
               infot = ( ( argpos + 3 ) * descmult ) + i
               CALL pscallsub( subptr, scode )
               CALL pchkpbe( ictxt, nout, sname, infot )
*
            END IF
*
   50    CONTINUE
*
*        Check INCY. Set all other OK, bad INCY
*
         CALL pssetpblas( ictxt )
         incy =  -1
         infot = argpos + 4
         CALL pscallsub( subptr, scode )
         CALL pchkpbe( ictxt, nout, sname, infot )
*
      END IF
*
      RETURN
*
*     End of PSCHKMAT
*
      END
      SUBROUTINE pscallsub( SUBPTR, SCODE )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER             SCODE
*     ..
*     .. Subroutine Arguments ..
      EXTERNAL            subptr
*     ..
*
*  Purpose
*  =======
*
*  PSCALLSUB calls the subroutine SUBPTR with the calling sequence iden-
*  tified by SCODE.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  SUBPTR  (global input) SUBROUTINE
*          On entry,  SUBPTR  is  a  subroutine. SUBPTR must be declared
*          EXTERNAL in the calling subroutine.
*
*  SCODE   (global input) INTEGER
*          On entry, SCODE specifies the calling sequence code.
*
*  Calling sequence encodings
*  ==========================
*
*  code Formal argument list                                Examples
*
*  11   (n,      v1,v2)                                     _SWAP, _COPY
*  12   (n,s1,   v1   )                                     _SCAL, _SCAL
*  13   (n,s1,   v1,v2)                                     _AXPY, _DOT_
*  14   (n,s1,i1,v1   )                                     _AMAX
*  15   (n,u1,   v1   )                                     _ASUM, _NRM2
*
*  21   (     trans,     m,n,s1,m1,v1,s2,v2)                _GEMV
*  22   (uplo,             n,s1,m1,v1,s2,v2)                _SYMV, _HEMV
*  23   (uplo,trans,diag,  n,   m1,v1      )                _TRMV, _TRSV
*  24   (                m,n,s1,v1,v2,m1)                   _GER_
*  25   (uplo,             n,s1,v1,   m1)                   _SYR
*  26   (uplo,             n,u1,v1,   m1)                   _HER
*  27   (uplo,             n,s1,v1,v2,m1)                   _SYR2, _HER2
*
*  31   (          transa,transb,     m,n,k,s1,m1,m2,s2,m3) _GEMM
*  32   (side,uplo,                   m,n,  s1,m1,m2,s2,m3) _SYMM, _HEMM
*  33   (     uplo,trans,               n,k,s1,m1,   s2,m3) _SYRK
*  34   (     uplo,trans,               n,k,u1,m1,   u2,m3) _HERK
*  35   (     uplo,trans,               n,k,s1,m1,m2,s2,m3) _SYR2K
*  36   (     uplo,trans,               n,k,s1,m1,m2,u2,m3) _HER2K
*  37   (                             m,n,  s1,m1,   s2,m3) _TRAN_
*  38   (side,uplo,transa,       diag,m,n,  s1,m1,m2      ) _TRMM, _TRSM
*  39   (          trans,             m,n,  s1,m1,   s2,m3) _GEADD
*  40   (     uplo,trans,             m,n,  s1,m1,   s2,m3) _TRADD
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      parameter( block_cyclic_2d_inb = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
*     ..
*     .. Common Blocks ..
      CHARACTER*1        DIAG, SIDE, TRANSA, TRANSB, UPLO
      INTEGER            IA, IB, IC, INCX, INCY, ISCLR, IX, IY, JA, JB,
     $                   JC, JX, JY, KDIM, MDIM, NDIM
      REAL               USCLR, SCLR
      INTEGER            DESCA( DLEN_ ), DESCB( DLEN_ ), DESCC( DLEN_ ),
     $                   DESCX( DLEN_ ), DESCY( DLEN_ )
      REAL               A( 2, 2 ), B( 2, 2 ), C( 2, 2 ), X( 2 ), Y( 2 )
      COMMON             /pblasc/diag, side, transa, transb, uplo
      COMMON             /pblasd/desca, descb, descc, descx, descy
      COMMON             /pblasi/ia, ib, ic, incx, incy, isclr, ix, iy,
     $                   ja, jb, jc, jx, jy
      COMMON             /pblasm/a, b, c
      COMMON             /pblasn/kdim, mdim, ndim
      COMMON             /pblass/sclr, usclr
      COMMON             /pblasv/x, y
*     ..
*     .. Executable Statements ..
*
*     Level 1 PBLAS
*
      IF( scode.EQ.11 ) THEN
*
         CALL subptr( ndim, x, ix, jx, descx, incx, y, iy, jy, descy,
     $                incy )
*
      ELSE IF( scode.EQ.12 ) THEN
*
         CALL subptr( ndim, sclr, x, ix, jx, descx, incx )
*
      ELSE IF( scode.EQ.13 ) THEN
*
         CALL subptr( ndim, sclr, x, ix, jx, descx, incx, y, iy, jy,
     $                descy, incy )
*
      ELSE IF( scode.EQ.14 ) THEN
*
         CALL subptr( ndim, sclr, isclr, x, ix, jx, descx, incx )
*
      ELSE IF( scode.EQ.15 ) THEN
*
         CALL subptr( ndim, usclr, x, ix, jx, descx, incx )
*
*     Level 2 PBLAS
*
      ELSE IF( scode.EQ.21 ) THEN
*
         CALL subptr( transa, mdim, ndim, sclr, a, ia, ja, desca, x, ix,
     $                jx, descx, incx, sclr, y, iy, jy, descy, incy )
*
      ELSE IF( scode.EQ.22 ) THEN
*
         CALL subptr( uplo, ndim, sclr, a, ia, ja, desca, x, ix, jx,
     $                descx, incx, sclr, y, iy, jy, descy, incy )
*
      ELSE IF( scode.EQ.23 ) THEN
*
         CALL subptr( uplo, transa, diag, ndim, a, ia, ja, desca, x, ix,
     $                jx, descx, incx )
*
      ELSE IF( scode.EQ.24 ) THEN
*
         CALL subptr( mdim, ndim, sclr, x, ix, jx, descx, incx, y, iy,
     $                jy, descy, incy, a, ia, ja, desca )
*
      ELSE IF( scode.EQ.25 ) THEN
*
         CALL subptr( uplo, ndim, sclr, x, ix, jx, descx, incx, a, ia,
     $                ja, desca )
*
      ELSE IF( scode.EQ.26 ) THEN
*
         CALL subptr( uplo, ndim, usclr, x, ix, jx, descx, incx, a, ia,
     $                ja, desca )
*
      ELSE IF( scode.EQ.27 ) THEN
*
         CALL subptr( uplo, ndim, sclr, x, ix, jx, descx, incx, y, iy,
     $                jy, descy, incy, a, ia, ja, desca )
*
*     Level 3 PBLAS
*
      ELSE IF( scode.EQ.31 ) THEN
*
         CALL subptr( transa, transb, mdim, ndim, kdim, sclr, a, ia, ja,
     $                desca, b, ib, jb, descb, sclr, c, ic, jc, descc )
*
      ELSE IF( scode.EQ.32 ) THEN
*
         CALL subptr( side, uplo, mdim, ndim, sclr, a, ia, ja, desca, b,
     $                ib, jb, descb, sclr, c, ic, jc, descc )
*
      ELSE IF( scode.EQ.33 ) THEN
*
         CALL subptr( uplo, transa, ndim, kdim, sclr, a, ia, ja, desca,
     $                sclr, c, ic, jc, descc )
*
      ELSE IF( scode.EQ.34 ) THEN
*
         CALL subptr( uplo, transa, ndim, kdim, usclr, a, ia, ja, desca,
     $                usclr, c, ic, jc, descc )
*
      ELSE IF( scode.EQ.35 ) THEN
*
         CALL subptr( uplo, transa, ndim, kdim, sclr, a, ia, ja, desca,
     $                b, ib, jb, descb, sclr, c, ic, jc, descc )
*
      ELSE IF( scode.EQ.36 ) THEN
*
         CALL subptr( uplo, transa, ndim, kdim, sclr, a, ia, ja, desca,
     $                b, ib, jb, descb, usclr, c, ic, jc, descc )
*
      ELSE IF( scode.EQ.37 ) THEN
*
         CALL subptr( mdim, ndim, sclr, a, ia, ja, desca, sclr, c, ic,
     $                jc, descc )
*
      ELSE IF( scode.EQ.38 ) THEN
*
         CALL subptr( side, uplo, transa, diag, mdim, ndim, sclr, a, ia,
     $                ja, desca, b, ib, jb, descb )
*
      ELSE IF( scode.EQ.39 ) THEN
*
         CALL subptr( transa, mdim, ndim, sclr, a, ia, ja, desca, sclr,
     $                c, ic, jc, descc )
*
      ELSE IF( scode.EQ.40 ) THEN
*
         CALL subptr( uplo, transa, mdim, ndim, sclr, a, ia, ja, desca,
     $                sclr, c, ic, jc, descc )
*
      END IF
*
      RETURN
*
*     End of PSCALLSUB
*
      END
      SUBROUTINE pserrset( ERR, ERRMAX, XTRUE, X )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      REAL               ERR, ERRMAX, X, XTRUE
*     ..
*
*  Purpose
*  =======
*
*  PSERRSET  computes the absolute difference ERR = |XTRUE - X| and com-
*  pares it with zero. ERRMAX accumulates the absolute error difference.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ERR     (local output) REAL
*          On exit, ERR specifies the absolute difference |XTRUE - X|.
*
*  ERRMAX  (local input/local output) REAL
*          On entry,  ERRMAX  specifies  a previously computed error. On
*          exit ERRMAX is the accumulated error MAX( ERRMAX, ERR ).
*
*  XTRUE   (local input) REAL
*          On entry, XTRUE specifies the true value.
*
*  X       (local input) REAL
*          On entry, X specifies the value to be compared to XTRUE.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. External Functions ..
      REAL               PSDIFF
      EXTERNAL           PSDIFF
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          abs, max
*     ..
*     .. Executable Statements ..
*
      err = abs( psdiff( xtrue, x ) )
*
      errmax = max( errmax, err )
*
      RETURN
*
*     End of PSERRSET
*
      END
      SUBROUTINE pschkvin( ERRMAX, N, X, PX, IX, JX, DESCX, INCX,
     $                     INFO )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER            INCX, INFO, IX, JX, N
      REAL               ERRMAX
*     ..
*     .. Array Arguments ..
      INTEGER            DESCX( * )
      REAL               PX( * ), X( * )
*     ..
*
*  Purpose
*  =======
*
*  PSCHKVIN  checks that the submatrix sub( PX ) remained unchanged. The
*  local  array  entries are compared element by element, and their dif-
*  ference  is tested against 0.0 as well as the epsilon machine. Notice
*  that  this difference should be numerically exactly the zero machine,
*  but  because of the possible fluctuation of some of the data we flag-
*  ged differently a difference less than twice the epsilon machine. The
*  largest error is also returned.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ERRMAX  (global output) REAL
*          On exit,  ERRMAX  specifies the largest absolute element-wise
*          difference between sub( X ) and sub( PX ).
*
*  N       (global input) INTEGER
*          On entry,  N  specifies  the  length of the subvector operand
*          sub( X ). N must be at least zero.
*
*  X       (local input) REAL array
*          On entry, X is an array of  dimension  (DESCX( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PX.
*
*  PX      (local input) REAL array
*          On entry, PX is an array of dimension (DESCX( LLD_ ),*). This
*          array contains the local entries of the matrix PX.
*
*  IX      (global input) INTEGER
*          On entry, IX  specifies X's global row index, which points to
*          the beginning of the submatrix sub( X ).
*
*  JX      (global input) INTEGER
*          On entry, JX  specifies X's global column index, which points
*          to the beginning of the submatrix sub( X ).
*
*  DESCX   (global and local input) INTEGER array
*          On entry, DESCX  is an integer array of dimension DLEN_. This
*          is the array descriptor for the matrix X.
*
*  INCX    (global input) INTEGER
*          On entry,  INCX   specifies  the  global  increment  for  the
*          elements of  X.  Only two values of  INCX   are  supported in
*          this version, namely 1 and M_X. INCX  must not be zero.
*
*  INFO    (global output) INTEGER
*          On exit, if INFO = 0, no error has been found,
*          If INFO > 0, the maximum abolute error found is in (0,eps],
*          If INFO < 0, the maximum abolute error found is in (eps,+oo).
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
      REAL               ZERO
      PARAMETER          ( ZERO = 0.0e+0 )
*     ..
*     .. Local Scalars ..
      LOGICAL            COLREP, ROWREP
      INTEGER            I, IB, ICTXT, ICURCOL, ICURROW, IIX, IN, IXCOL,
     $                   IXROW, J, JB, JJX, JN, KK, LDPX, LDX, LL,
     $                   MYCOL, MYROW, NPCOL, NPROW
      REAL               ERR, EPS
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_gridinfo, pb_infog2l, pserrset, sgamx2d
*     ..
*     .. External Functions ..
      REAL               PSLAMCH
      EXTERNAL           pslamch
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          abs, max, min, mod
*     ..
*     .. Executable Statements ..
*
      info = 0
      errmax = zero
*
*     Quick return if possible
*
      IF( n.LE.0 )
     $   RETURN
*
      ictxt = descx( ctxt_ )
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      eps = pslamch( ictxt, 'eps' )
*
      CALL pb_infog2l( ix, jx, descx, nprow, npcol, myrow, mycol, iix,
     $                 jjx, ixrow, ixcol )
*
      ldx    = descx( m_ )
      ldpx   = descx( lld_ )
      rowrep = ( ixrow.EQ.-1 )
      colrep = ( ixcol.EQ.-1 )
*
      IF( n.EQ.1 ) THEN
*
         IF( ( myrow.EQ.ixrow .OR. rowrep ) .AND.
     $       ( mycol.EQ.ixcol .OR. colrep ) )
     $      CALL pserrset( err, errmax, x( ix+(jx-1)*ldx ),
     $                     px( iix+(jjx-1)*ldpx ) )
*
      ELSE IF( incx.EQ.descx( m_ ) ) THEN
*
*        sub( X ) is a row vector
*
         jb = descx( inb_ ) - jx + 1
         IF( jb.LE.0 )
     $      jb = ( ( -jb ) / descx( nb_ ) + 1 ) * descx( nb_ ) + jb
         jb = min( jb, n )
         jn = jx + jb - 1
*
         IF( myrow.EQ.ixrow .OR. rowrep ) THEN
*
            icurcol = ixcol
            IF( mycol.EQ.icurcol .OR. colrep ) THEN
               DO 10 j = jx, jn
                  CALL pserrset( err, errmax, x( ix+(j-1)*ldx ),
     $                           px( iix+(jjx-1)*ldpx ) )
                  jjx = jjx + 1
   10          CONTINUE
            END IF
            icurcol = mod( icurcol+1, npcol )
*
            DO 30 j = jn+1, jx+n-1, descx( nb_ )
               jb = min( jx+n-j, descx( nb_ ) )
*
               IF( mycol.EQ.icurcol .OR. colrep ) THEN
*
                  DO 20 kk = 0, jb-1
                     CALL pserrset( err, errmax, x( ix+(j+kk-1)*ldx ),
     $                              px( iix+(jjx+kk-1)*ldpx ) )
   20             CONTINUE
*
                  jjx = jjx + jb
*
               END IF
*
               icurcol = mod( icurcol+1, npcol )
*
   30       CONTINUE
*
         END IF
*
      ELSE
*
*        sub( X ) is a column vector
*
         ib = descx( imb_ ) - ix + 1
         IF( ib.LE.0 )
     $      ib = ( ( -ib ) / descx( mb_ ) + 1 ) * descx( mb_ ) + ib
         ib = min( ib, n )
         in = ix + ib - 1
*
         IF( mycol.EQ.ixcol .OR. colrep ) THEN
*
            icurrow = ixrow
            IF( myrow.EQ.icurrow .OR. rowrep ) THEN
               DO 40 i = ix, in
                  CALL pserrset( err, errmax, x( i+(jx-1)*ldx ),
     $                           px( iix+(jjx-1)*ldpx ) )
                  iix = iix + 1
   40          CONTINUE
            END IF
            icurrow = mod( icurrow+1, nprow )
*
            DO 60 i = in+1, ix+n-1, descx( mb_ )
               ib = min( ix+n-i, descx( mb_ ) )
*
               IF( myrow.EQ.icurrow .OR. rowrep ) THEN
*
                  DO 50 kk = 0, ib-1
                     CALL pserrset( err, errmax, x( i+kk+(jx-1)*ldx ),
     $                              px( iix+kk+(jjx-1)*ldpx ) )
   50             CONTINUE
*
                  iix = iix + ib
*
               END IF
*
               icurrow = mod( icurrow+1, nprow )
*
   60       CONTINUE
*
         END IF
*
      END IF
*
      CALL sgamx2d( ictxt, 'All', ' ', 1, 1, errmax, 1, kk, ll, -1,
     $              -1, -1 )
*
      IF( errmax.GT.zero .AND. errmax.LE.eps ) THEN
         info = 1
      ELSE IF( errmax.GT.eps ) THEN
         info = -1
      END IF
*
      RETURN
*
*     End of PSCHKVIN
*
      END
      SUBROUTINE pschkvout( N, X, PX, IX, JX, DESCX, INCX, INFO )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER            INCX, INFO, IX, JX, N
*     ..
*     .. Array Arguments ..
      INTEGER            DESCX( * )
      REAL               PX( * ), X( * )
*     ..
*
*  Purpose
*  =======
*
*  PSCHKVOUT  checks  that the matrix PX \ sub( PX ) remained unchanged.
*  The  local array  entries  are compared element by element, and their
*  difference  is tested against 0.0 as well as the epsilon machine. No-
*  tice that this  difference should be numerically exactly the zero ma-
*  chine, but because  of  the  possible movement of some of the data we
*  flagged differently a difference less than twice the epsilon machine.
*  The largest error is reported.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  N       (global input) INTEGER
*          On entry,  N  specifies  the  length of the subvector operand
*          sub( X ). N must be at least zero.
*
*  X       (local input) REAL array
*          On entry, X is an array of  dimension  (DESCX( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PX.
*
*  PX      (local input) REAL array
*          On entry, PX is an array of dimension (DESCX( LLD_ ),*). This
*          array contains the local entries of the matrix PX.
*
*  IX      (global input) INTEGER
*          On entry, IX  specifies X's global row index, which points to
*          the beginning of the submatrix sub( X ).
*
*  JX      (global input) INTEGER
*          On entry, JX  specifies X's global column index, which points
*          to the beginning of the submatrix sub( X ).
*
*  DESCX   (global and local input) INTEGER array
*          On entry, DESCX  is an integer array of dimension DLEN_. This
*          is the array descriptor for the matrix X.
*
*  INCX    (global input) INTEGER
*          On entry,  INCX   specifies  the  global  increment  for  the
*          elements of  X.  Only two values of  INCX   are  supported in
*          this version, namely 1 and M_X. INCX  must not be zero.
*
*  INFO    (global output) INTEGER
*          On exit, if INFO = 0, no error has been found,
*          If INFO > 0, the maximum abolute error found is in (0,eps],
*          If INFO < 0, the maximum abolute error found is in (eps,+oo).
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
      REAL               ZERO
      PARAMETER          ( ZERO = 0.0e+0 )
*     ..
*     .. Local Scalars ..
      LOGICAL            COLREP, ROWREP
      INTEGER            I, IB, ICTXT, ICURCOL, ICURROW, II, IMBX, INBX,
     $                   J, JB, JJ, KK, LDPX, LDX, LL, MBX, MPALL,
     $                   MYCOL, MYCOLDIST, MYROW, MYROWDIST, NBX, NPCOL,
     $                   nprow, nqall
      REAL               EPS, ERR, ERRMAX
*     ..
*     .. External Subroutines ..
      EXTERNAL           BLACS_GRIDINFO, PSERRSET, SGAMX2D
*     ..
*     .. External Functions ..
      INTEGER            PB_NUMROC
      REAL               PSLAMCH
      EXTERNAL           PSLAMCH, PB_NUMROC
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          abs, max, min, mod
*     ..
*     .. Executable Statements ..
*
      info = 0
      errmax = zero
*
*     Quick return if possible
*
      IF( ( descx( m_ ).LE.0 ).OR.( descx( n_ ).LE.0 ) )
     $   RETURN
*
*     Start the operations
*
      ictxt = descx( ctxt_ )
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      eps = pslamch( ictxt, 'eps' )
*
      mpall   = pb_numroc( descx( m_ ), 1, descx( imb_ ), descx( mb_ ),
     $                     myrow, descx( rsrc_ ), nprow )
      nqall   = pb_numroc( descx( n_ ), 1, descx( inb_ ), descx( nb_ ),
     $                     mycol, descx( csrc_ ), npcol )
*
      mbx     = descx( mb_ )
      nbx     = descx( nb_ )
      ldx     = descx( m_ )
      ldpx    = descx( lld_ )
      icurrow = descx( rsrc_ )
      icurcol = descx( csrc_ )
      rowrep  = ( icurrow.EQ.-1 )
      colrep  = ( icurcol.EQ.-1 )
      IF( myrow.EQ.icurrow .OR. rowrep ) THEN
         imbx = descx( imb_ )
      ELSE
         imbx = mbx
      END IF
      IF( mycol.EQ.icurcol .OR. colrep ) THEN
         inbx = descx( inb_ )
      ELSE
         inbx = nbx
      END IF
      IF( rowrep ) THEN
         myrowdist = 0
      ELSE
         myrowdist = mod( myrow - icurrow + nprow, nprow )
      END IF
      IF( colrep ) THEN
         mycoldist = 0
      ELSE
         mycoldist = mod( mycol - icurcol + npcol, npcol )
      END IF
      ii = 1
      jj = 1
*
      IF( incx.EQ.descx( m_ ) ) THEN
*
*        sub( X ) is a row vector
*
         IF( myrow.EQ.icurrow .OR. rowrep ) THEN
*
            i = 1
            IF( mycoldist.EQ.0 ) THEN
               j = 1
            ELSE
               j = descx( inb_ ) + ( mycoldist - 1 ) * nbx + 1
            END IF
            jb = min( max( 0, descx( n_ ) - j + 1 ), inbx )
            ib = min( descx( m_ ), descx( imb_ ) )
*
            DO 20 kk = 0, jb-1
               DO 10 ll = 0, ib-1
                  IF( i+ll.NE.ix .OR. j+kk.LT.jx .OR. j+kk.GT.jx+n-1 )
     $               CALL pserrset( err, errmax,
     $                              x( i+ll+(j+kk-1)*ldx ),
     $                              px( ii+ll+(jj+kk-1)*ldpx ) )
   10          CONTINUE
   20       CONTINUE
            IF( colrep ) THEN
               j = j + inbx
            ELSE
               j = j + inbx + ( npcol - 1 ) * nbx
            END IF
*
            DO 50 jj = inbx+1, nqall, nbx
               jb = min( nqall-jj+1, nbx )
*
               DO 40 kk = 0, jb-1
                  DO 30 ll = 0, ib-1
                     IF( i+ll.NE.ix .OR. j+kk.LT.jx .OR.
     $                   j+kk.GT.jx+n-1 )
     $                  CALL pserrset( err, errmax,
     $                                 x( i+ll+(j+kk-1)*ldx ),
     $                                 px( ii+ll+(jj+kk-1)*ldpx ) )
   30             CONTINUE
   40          CONTINUE
*
               IF( colrep ) THEN
                  j = j + nbx
               ELSE
                  j = j + npcol * nbx
               END IF
*
   50       CONTINUE
*
            ii = ii + ib
*
         END IF
*
         icurrow = mod( icurrow + 1, nprow )
*
         DO 110 i = descx( imb_ ) + 1, descx( m_ ), mbx
            ib = min( descx( m_ ) - i + 1, mbx )
*
            IF( myrow.EQ.icurrow .OR. rowrep ) THEN
*
               IF( mycoldist.EQ.0 ) THEN
                  j = 1
               ELSE
                  j = descx( inb_ ) + ( mycoldist - 1 ) * nbx + 1
               END IF
*
               jj = 1
               jb = min( max( 0, descx( n_ ) - j + 1 ), inbx )
               DO 70 kk = 0, jb-1
                  DO 60 ll = 0, ib-1
                     IF( i+ll.NE.ix .OR. j+kk.LT.jx .OR.
     $                   j+kk.GT.jx+n-1 )
     $                  CALL pserrset( err, errmax,
     $                                 x( i+ll+(j+kk-1)*ldx ),
     $                                 px( ii+ll+(jj+kk-1)*ldpx ) )
   60             CONTINUE
   70          CONTINUE
               IF( colrep ) THEN
                  j = j + inbx
               ELSE
                  j = j + inbx + ( npcol - 1 ) * nbx
               END IF
*
               DO 100 jj = inbx+1, nqall, nbx
                  jb = min( nqall-jj+1, nbx )
*
                  DO 90 kk = 0, jb-1
                     DO 80 ll = 0, ib-1
                        IF( i+ll.NE.ix .OR. j+kk.LT.jx .OR.
     $                      j+kk.GT.jx+n-1 )
     $                     CALL pserrset( err, errmax,
     $                                    x( i+ll+(j+kk-1)*ldx ),
     $                                    px( ii+ll+(jj+kk-1)*ldpx ) )
   80                CONTINUE
   90             CONTINUE
*
                  IF( colrep ) THEN
                     j = j + nbx
                  ELSE
                     j = j + npcol * nbx
                  END IF
*
  100          CONTINUE
*
               ii = ii + ib
*
            END IF
*
            icurrow = mod( icurrow + 1, nprow )
*
  110    CONTINUE
*
      ELSE
*
*        sub( X ) is a column vector
*
         IF( mycol.EQ.icurcol .OR. colrep ) THEN
*
            j = 1
            IF( myrowdist.EQ.0 ) THEN
               i = 1
            ELSE
               i = descx( imb_ ) + ( myrowdist - 1 ) * mbx + 1
            END IF
            ib = min( max( 0, descx( m_ ) - i + 1 ), imbx )
            jb = min( descx( n_ ), descx( inb_ ) )
*
            DO 130 kk = 0, jb-1
               DO 120 ll = 0, ib-1
                  IF( j+kk.NE.jx .OR. i+ll.LT.ix .OR. i+ll.GT.ix+n-1 )
     $               CALL pserrset( err, errmax,
     $                              x( i+ll+(j+kk-1)*ldx ),
     $                              px( ii+ll+(jj+kk-1)*ldpx ) )
  120          CONTINUE
  130       CONTINUE
            IF( rowrep ) THEN
               i = i + imbx
            ELSE
               i = i + imbx + ( nprow - 1 ) * mbx
            END IF
*
            DO 160 ii = imbx+1, mpall, mbx
               ib = min( mpall-ii+1, mbx )
*
               DO 150 kk = 0, jb-1
                  DO 140 ll = 0, ib-1
                     IF( j+kk.NE.jx .OR. i+ll.LT.ix .OR.
     $                   i+ll.GT.ix+n-1 )
     $                  CALL pserrset( err, errmax,
     $                                 x( i+ll+(j+kk-1)*ldx ),
     $                                 px( ii+ll+(jj+kk-1)*ldpx ) )
  140             CONTINUE
  150          CONTINUE
*
               IF( rowrep ) THEN
                  i = i + mbx
               ELSE
                  i = i + nprow * mbx
               END IF
*
  160       CONTINUE
*
            jj = jj + jb
*
         END IF
*
         icurcol = mod( icurcol + 1, npcol )
*
         DO 220 j = descx( inb_ ) + 1, descx( n_ ), nbx
            jb = min( descx( n_ ) - j + 1, nbx )
*
            IF( mycol.EQ.icurcol .OR. colrep ) THEN
*
               IF( myrowdist.EQ.0 ) THEN
                  i = 1
               ELSE
                  i = descx( imb_ ) + ( myrowdist - 1 ) * mbx + 1
               END IF
*
               ii = 1
               ib = min( max( 0, descx( m_ ) - i + 1 ), imbx )
               DO 180 kk = 0, jb-1
                  DO 170 ll = 0, ib-1
                     IF( j+kk.NE.jx .OR. i+ll.LT.ix .OR.
     $                   i+ll.GT.ix+n-1 )
     $                  CALL pserrset( err, errmax,
     $                                 x( i+ll+(j+kk-1)*ldx ),
     $                                 px( ii+ll+(jj+kk-1)*ldpx ) )
  170             CONTINUE
  180          CONTINUE
               IF( rowrep ) THEN
                  i = i + imbx
               ELSE
                  i = i + imbx + ( nprow - 1 ) * mbx
               END IF
*
               DO 210 ii = imbx+1, mpall, mbx
                  ib = min( mpall-ii+1, mbx )
*
                  DO 200 kk = 0, jb-1
                     DO 190 ll = 0, ib-1
                        IF( j+kk.NE.jx .OR. i+ll.LT.ix .OR.
     $                      i+ll.GT.ix+n-1 )
     $                     CALL pserrset( err, errmax,
     $                                    x( i+ll+(j+kk-1)*ldx ),
     $                                    px( ii+ll+(jj+kk-1)*ldpx ) )
  190                CONTINUE
  200             CONTINUE
*
                  IF( rowrep ) THEN
                     i = i + mbx
                  ELSE
                     i = i + nprow * mbx
                  END IF
*
  210          CONTINUE
*
               jj = jj + jb
*
            END IF
*
            icurcol = mod( icurcol + 1, npcol )
*
  220    CONTINUE
*
      END IF
*
      CALL sgamx2d( ictxt, 'All', ' ', 1, 1, errmax, 1, kk, ll, -1,
     $              -1, -1 )
*
      IF( errmax.GT.zero .AND. errmax.LE.eps ) THEN
         info = 1
      ELSE IF( errmax.GT.eps ) THEN
         info = -1
      END IF
*
      RETURN
*
*     End of PSCHKVOUT
*
      END
      SUBROUTINE pschkmin( ERRMAX, M, N, A, PA, IA, JA, DESCA, INFO )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER            IA, INFO, JA, M, N
      REAL               ERRMAX
*     ..
*     .. Array Arguments ..
      INTEGER            DESCA( * )
      REAL               PA( * ), A( * )
*     ..
*
*  Purpose
*  =======
*
*  PSCHKMIN  checks that the submatrix sub( PA ) remained unchanged. The
*  local  array  entries are compared element by element, and their dif-
*  ference  is tested against 0.0 as well as the epsilon machine. Notice
*  that  this difference should be numerically exactly the zero machine,
*  but  because of the possible fluctuation of some of the data we flag-
*  ged differently a difference less than twice the epsilon machine. The
*  largest error is also returned.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ERRMAX  (global output) REAL
*          On exit,  ERRMAX  specifies the largest absolute element-wise
*          difference between sub( A ) and sub( PA ).
*
*  M       (global input) INTEGER
*          On entry,  M  specifies  the  number of rows of the submatrix
*          operand sub( A ). M must be at least zero.
*
*  N       (global input) INTEGER
*          On entry, N  specifies the number of columns of the submatrix
*          operand sub( A ). N must be at least zero.
*
*  A       (local input) REAL array
*          On entry, A is an array of  dimension  (DESCA( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PA.
*
*  PA      (local input) REAL array
*          On entry, PA is an array of dimension (DESCA( LLD_ ),*). This
*          array contains the local entries of the matrix PA.
*
*  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.
*
*  INFO    (global output) INTEGER
*          On exit, if INFO = 0, no error has been found,
*          If INFO > 0, the maximum abolute error found is in (0,eps],
*          If INFO < 0, the maximum abolute error found is in (eps,+oo).
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
      REAL               ZERO
      PARAMETER          ( ZERO = 0.0e+0 )
*     ..
*     .. Local Scalars ..
      LOGICAL            COLREP, ROWREP
      INTEGER            H, I, IACOL, IAROW, IB, ICTXT, ICURCOL,
     $                   ICURROW, II, IIA, IN, J, JB, JJ, JJA, JN, K,
     $                   KK, LDA, LDPA, LL, MYCOL, MYROW, NPCOL, NPROW
      REAL               ERR, EPS
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_gridinfo, pb_infog2l, pserrset, sgamx2d
*     ..
*     .. External Functions ..
      REAL               PSLAMCH
      EXTERNAL           pslamch
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          abs, max, min, mod
*     ..
*     .. Executable Statements ..
*
      info   = 0
      errmax = zero
*
*     Quick return if posssible
*
      IF( ( m.EQ.0 ).OR.( n.EQ.0 ) )
     $   RETURN
*
*     Start the operations
*
      ictxt = desca( ctxt_ )
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      eps = pslamch( ictxt, 'eps' )
*
      CALL pb_infog2l( ia, ja, desca, nprow, npcol, myrow, mycol, iia,
     $                 jja, iarow, iacol )
*
      ii      = iia
      jj      = jja
      lda     = desca( m_ )
      ldpa    = desca( lld_ )
      icurrow = iarow
      icurcol = iacol
      rowrep  = ( iarow.EQ.-1 )
      colrep  = ( iacol.EQ.-1 )
*
*     Handle the first block of column separately
*
      jb = desca( inb_ ) - ja  + 1
      IF( jb.LE.0 )
     $   jb = ( ( -jb ) / desca( nb_ ) + 1 ) * desca( nb_ ) + jb
      jb = min( jb, n )
      jn = ja + jb - 1
*
      IF( mycol.EQ.icurcol .OR. colrep ) THEN
*
         DO 40 h = 0, jb-1
            ib = desca( imb_ ) - ia  + 1
            IF( ib.LE.0 )
     $         ib = ( ( -ib ) / desca( mb_ ) + 1 ) * desca( mb_ ) + ib
            ib = min( ib, m )
            in = ia + ib - 1
            IF( myrow.EQ.icurrow .OR. rowrep ) THEN
               DO 10 k = 0, ib-1
                  CALL pserrset( err, errmax, a( ia+k+(ja+h-1)*lda ),
     $                           pa( ii+k+(jj+h-1)*ldpa ) )
   10          CONTINUE
               ii = ii + ib
            END IF
            icurrow = mod( icurrow+1, nprow )
*
*           Loop over remaining block of rows
*
            DO 30 i = in+1, ia+m-1, desca( mb_ )
               ib = min( desca( mb_ ), ia+m-i )
               IF( myrow.EQ.icurrow .OR. rowrep ) THEN
                  DO 20 k = 0, ib-1
                     CALL pserrset( err, errmax, a( i+k+(ja+h-1)*lda ),
     $                              pa( ii+k+(jj+h-1)*ldpa ) )
   20             CONTINUE
                  ii = ii + ib
               END IF
               icurrow = mod( icurrow+1, nprow )
   30       CONTINUE
*
            ii = iia
            icurrow = iarow
   40    CONTINUE
*
         jj = jj + jb
*
      END IF
*
      icurcol = mod( icurcol+1, npcol )
*
*     Loop over remaining column blocks
*
      DO 90 j = jn+1, ja+n-1, desca( nb_ )
         jb = min(  desca( nb_ ), ja+n-j )
         IF( mycol.EQ.icurcol .OR. colrep ) THEN
            DO 80 h = 0, jb-1
               ib = desca( imb_ ) - ia  + 1
               IF( ib.LE.0 )
     $            ib = ( ( -ib ) / desca( mb_ ) + 1 )*desca( mb_ ) + ib
               ib = min( ib, m )
               in = ia + ib - 1
               IF( myrow.EQ.icurrow .OR. rowrep ) THEN
                  DO 50 k = 0, ib-1
                     CALL pserrset( err, errmax, a( ia+k+(j+h-1)*lda ),
     $                              pa( ii+k+(jj+h-1)*ldpa ) )
   50             CONTINUE
                  ii = ii + ib
               END IF
               icurrow = mod( icurrow+1, nprow )
*
*              Loop over remaining block of rows
*
               DO 70 i = in+1, ia+m-1, desca( mb_ )
                  ib = min( desca( mb_ ), ia+m-i )
                  IF( myrow.EQ.icurrow .OR. rowrep ) THEN
                     DO 60 k = 0, ib-1
                        CALL pserrset( err, errmax,
     $                                 a( i+k+(j+h-1)*lda ),
     $                                 pa( ii+k+(jj+h-1)*ldpa ) )
   60                CONTINUE
                     ii = ii + ib
                  END IF
                  icurrow = mod( icurrow+1, nprow )
   70          CONTINUE
*
               ii = iia
               icurrow = iarow
   80       CONTINUE
*
            jj = jj + jb
         END IF
*
         icurcol = mod( icurcol+1, npcol )
*
   90 CONTINUE
*
      CALL sgamx2d( ictxt, 'All', ' ', 1, 1, errmax, 1, kk, ll, -1,
     $              -1, -1 )
*
      IF( errmax.GT.zero .AND. errmax.LE.eps ) THEN
         info = 1
      ELSE IF( errmax.GT.eps ) THEN
         info = -1
      END IF
*
      RETURN
*
*     End of PSCHKMIN
*
      END
      SUBROUTINE pschkmout( M, N, A, PA, IA, JA, DESCA, INFO )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER            IA, INFO, JA, M, N
*     ..
*     .. Array Arguments ..
      INTEGER            DESCA( * )
      REAL               A( * ), PA( * )
*     ..
*
*  Purpose
*  =======
*
*  PSCHKMOUT  checks  that the matrix PA \ sub( PA ) remained unchanged.
*  The  local array  entries  are compared element by element, and their
*  difference  is tested against 0.0 as well as the epsilon machine. No-
*  tice that this  difference should be numerically exactly the zero ma-
*  chine, but because  of  the  possible movement of some of the data we
*  flagged differently a difference less than twice the epsilon machine.
*  The largest error is reported.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  M       (global input) INTEGER
*          On entry,  M  specifies  the  number of rows of the submatrix
*          sub( PA ). M must be at least zero.
*
*  N       (global input) INTEGER
*          On entry, N specifies the  number of columns of the submatrix
*          sub( PA ). N must be at least zero.
*
*  A       (local input) REAL array
*          On entry, A is an array of  dimension  (DESCA( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PA.
*
*  PA      (local input) REAL array
*          On entry, PA is an array of dimension (DESCA( LLD_ ),*). This
*          array contains the local entries of the matrix PA.
*
*  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.
*
*  INFO    (global output) INTEGER
*          On exit, if INFO = 0, no error has been found,
*          If INFO > 0, the maximum abolute error found is in (0,eps],
*          If INFO < 0, the maximum abolute error found is in (eps,+oo).
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
      REAL               ZERO
      PARAMETER          ( ZERO = 0.0e+0 )
*     ..
*     .. Local Scalars ..
      LOGICAL            COLREP, ROWREP
      INTEGER            I, IB, ICTXT, ICURCOL, II, IMBA, J, JB, JJ, KK,
     $                   LDA, LDPA, LL, MPALL, MYCOL, MYROW, MYROWDIST,
     $                   NPCOL, NPROW
      REAL               EPS, ERR, ERRMAX
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_gridinfo, pserrset, sgamx2d
*     ..
*     .. External Functions ..
      INTEGER            PB_NUMROC
      REAL               PSLAMCH
      EXTERNAL           PSLAMCH, PB_NUMROC
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          max, min, mod
*     ..
*     .. Executable Statements ..
*
      info = 0
      errmax = zero
*
*     Quick return if possible
*
      IF( ( desca( m_ ).LE.0 ).OR.( desca( n_ ).LE.0 ) )
     $   RETURN
*
*     Start the operations
*
      ictxt = desca( ctxt_ )
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      eps = pslamch( ictxt, 'eps' )
*
      mpall = pb_numroc( desca( m_ ), 1, desca( imb_ ), desca( mb_ ),
     $                   myrow, desca( rsrc_ ), nprow )
*
      lda    = desca( m_ )
      ldpa   = desca( lld_ )
*
      ii = 1
      jj = 1
      rowrep  = ( desca( rsrc_ ).EQ.-1 )
      colrep  = ( desca( csrc_ ).EQ.-1 )
      icurcol = desca( csrc_ )
      IF( myrow.EQ.desca( rsrc_ ) .OR. rowrep ) THEN
         imba = desca( imb_ )
      ELSE
         imba = desca( mb_ )
      END IF
      IF( rowrep ) THEN
         myrowdist = 0
      ELSE
         myrowdist = mod( myrow - desca( rsrc_ ) + nprow, nprow )
      END IF
*
      IF( mycol.EQ.icurcol .OR. colrep ) THEN
*
         j = 1
         IF( myrowdist.EQ.0 ) THEN
            i = 1
         ELSE
            i = desca( imb_ ) + ( myrowdist - 1 ) * desca( mb_ ) + 1
         END IF
         ib = min( max( 0, desca( m_ ) - i + 1 ), imba )
         jb = min( desca( n_ ), desca( inb_ ) )
*
         DO 20 kk = 0, jb-1
            DO 10 ll = 0, ib-1
               IF( i+ll.LT.ia .OR. i+ll.GT.ia+m-1 .OR.
     $             j+kk.LT.ja .OR. j+kk.GT.ja+n-1 )
     $            CALL pserrset( err, errmax, a( i+ll+(j+kk-1)*lda ),
     $                           pa( ii+ll+(jj+kk-1)*ldpa ) )
   10       CONTINUE
   20    CONTINUE
         IF( rowrep ) THEN
            i = i + imba
         ELSE
            i = i + imba + ( nprow - 1 ) * desca( mb_ )
         END IF
*
         DO 50 ii = imba + 1, mpall, desca( mb_ )
            ib = min( mpall-ii+1, desca( mb_ ) )
*
            DO 40 kk = 0, jb-1
               DO 30 ll = 0, ib-1
                  IF( i+ll.LT.ia .OR. i+ll.GT.ia+m-1 .OR.
     $                j+kk.LT.ja .OR. j+kk.GT.ja+n-1 )
     $               CALL pserrset( err, errmax,
     $                              a( i+ll+(j+kk-1)*lda ),
     $                              pa( ii+ll+(jj+kk-1)*ldpa ) )
   30          CONTINUE
   40       CONTINUE
*
            IF( rowrep ) THEN
               i = i + desca( mb_ )
            ELSE
               i = i + nprow * desca( mb_ )
            END IF
*
   50    CONTINUE
*
         jj = jj + jb
*
      END IF
*
      icurcol = mod( icurcol + 1, npcol )
*
      DO 110 j = desca( inb_ ) + 1, desca( n_ ), desca( nb_ )
         jb = min( desca( n_ ) - j + 1, desca( nb_ ) )
*
         IF( mycol.EQ.icurcol .OR. colrep ) THEN
*
            IF( myrowdist.EQ.0 ) THEN
               i = 1
            ELSE
               i = desca( imb_ ) + ( myrowdist - 1 ) * desca( mb_ ) + 1
            END IF
*
            ii = 1
            ib = min( max( 0, desca( m_ ) - i + 1 ), imba )
            DO 70 kk = 0, jb-1
               DO 60 ll = 0, ib-1
                  IF( i+ll.LT.ia .OR. i+ll.GT.ia+m-1 .OR.
     $                j+kk.LT.ja .OR. j+kk.GT.ja+n-1 )
     $               CALL pserrset( err, errmax,
     $                              a( i+ll+(j+kk-1)*lda ),
     $                              pa( ii+ll+(jj+kk-1)*ldpa ) )
   60          CONTINUE
   70       CONTINUE
            IF( rowrep ) THEN
               i = i + imba
            ELSE
               i = i + imba + ( nprow - 1 ) * desca( mb_ )
            END IF
*
            DO 100 ii = imba+1, mpall, desca( mb_ )
               ib = min( mpall-ii+1, desca( mb_ ) )
*
               DO 90 kk = 0, jb-1
                  DO 80 ll = 0, ib-1
                     IF( i+ll.LT.ia .OR. i+ll.GT.ia+m-1 .OR.
     $                   j+kk.LT.ja .OR. j+kk.GT.ja+n-1 )
     $                  CALL pserrset( err, errmax,
     $                                 a( i+ll+(j+kk-1)*lda ),
     $                                 pa( ii+ll+(jj+kk-1)*ldpa ) )
   80             CONTINUE
   90          CONTINUE
*
               IF( rowrep ) THEN
                  i = i + desca( mb_ )
               ELSE
                  i = i + nprow * desca( mb_ )
               END IF
*
  100       CONTINUE
*
            jj = jj + jb
*
         END IF
*
         icurcol = mod( icurcol + 1, npcol )
*                                                           INSERT MODE
  110 CONTINUE
*
      CALL sgamx2d( ictxt, 'All', ' ', 1, 1, errmax, 1, kk, ll, -1,
     $              -1, -1 )
*
      IF( errmax.GT.zero .AND. errmax.LE.eps ) THEN
         info = 1
      ELSE IF( errmax.GT.eps ) THEN
         info = -1
      END IF
*
      RETURN
*
*     End of PSCHKMOUT
*
      END
      SUBROUTINE psmprnt( ICTXT, NOUT, M, N, A, LDA, IRPRNT, ICPRNT,
     $                    CMATNM )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER            ICPRNT, ICTXT, IRPRNT, LDA, M, N, NOUT
*     ..
*     .. Array Arguments ..
      CHARACTER*(*)      CMATNM
      REAL               A( LDA, * )
*     ..
*
*  Purpose
*  =======
*
*  PSMPRNT prints to the standard output an array A of size m by n. Only
*  the process of coordinates ( IRPRNT, ICPRNT ) is printing.
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  NOUT    (global input) INTEGER
*          On entry, NOUT specifies the unit number for the output file.
*          When NOUT is 6, output to screen,  when  NOUT is 0, output to
*          stderr. NOUT is only defined for process 0.
*
*  M       (global input) INTEGER
*          On entry, M  specifies the number of rows of the matrix A.  M
*          must be at least zero.
*
*  N       (global input) INTEGER
*          On entry, N  specifies the number of columns of the matrix A.
*          N must be at least zero.
*
*  A       (local input) REAL array
*          On entry,  A  is an array of dimension (LDA,N). The leading m
*          by n part of this array is printed.
*
*  LDA     (local input) INTEGER
*          On entry, LDA  specifies the leading dimension of  the  local
*          array A to be printed. LDA must be at least MAX( 1, M ).
*
*  IRPRNT  (global input) INTEGER
*          On entry, IRPRNT  specifies the process row coordinate of the
*          printing process.
*
*  ICPRNT  (global input) INTEGER
*          On entry,  ICPRNT  specifies the process column coordinate of
*          the printing process.
*
*  CMATNM  (global input) CHARACTER*(*)
*          On entry, CMATNM specifies the identifier of the matrix to be
*          printed.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Local Scalars ..
      INTEGER            I, J, MYCOL, MYROW, NPCOL, NPROW
*     ..
*     .. External Subroutines ..
      EXTERNAL           BLACS_GRIDINFO
*     ..
*     .. Executable Statements ..
*
*     Quick return if possible
*
      IF( ( m.LE.0 ).OR.( n.LE.0 ) )
     $   RETURN
*
*     Get grid parameters
*
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      IF( myrow.EQ.irprnt .AND. mycol.EQ.icprnt ) THEN
*
         WRITE( nout, fmt = * )
         DO 20 j = 1, n
*
            DO 10 i = 1, m
*
               WRITE( nout, fmt = 9999 ) cmatnm, i, j, a( i, j )
*
   10       CONTINUE
*
   20    CONTINUE
*
      END IF
*
 9999 FORMAT( 1x, a, '(', i6, ',', i6, ')=', e16.8 )
*
      RETURN
*
*     End of PSMPRNT
*
      END
      SUBROUTINE psvprnt( ICTXT, NOUT, N, X, INCX, IRPRNT, ICPRNT,
     $                    CVECNM )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER            ICPRNT, ICTXT, INCX, IRPRNT, N, NOUT
*     ..
*     .. Array Arguments ..
      CHARACTER*(*)      CVECNM
      REAL               X( * )
*     ..
*
*  Purpose
*  =======
*
*  PSVPRNT  prints  to the standard output an vector x of length n. Only
*  the process of coordinates ( IRPRNT, ICPRNT ) is printing.
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  NOUT    (global input) INTEGER
*          On entry, NOUT specifies the unit number for the output file.
*          When NOUT is 6, output to screen,  when  NOUT is 0, output to
*          stderr. NOUT is only defined for process 0.
*
*  N       (global input) INTEGER
*          On entry, N  specifies the length of the vector X.  N must be
*          at least zero.
*
*  X       (global input) REAL array
*          On   entry,   X   is   an   array   of   dimension  at  least
*          ( 1 + ( n - 1 )*abs( INCX ) ).  Before  entry,  the incremen-
*          ted array X must contain the vector x.
*
*  INCX    (global input) INTEGER.
*          On entry, INCX specifies the increment for the elements of X.
*          INCX must not be zero.
*
*  IRPRNT  (global input) INTEGER
*          On entry, IRPRNT  specifies the process row coordinate of the
*          printing process.
*
*  ICPRNT  (global input) INTEGER
*          On entry,  ICPRNT  specifies the process column coordinate of
*          the printing process.
*
*  CVECNM  (global input) CHARACTER*(*)
*          On entry, CVECNM specifies the identifier of the vector to be
*          printed.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Local Scalars ..
      INTEGER            I, MYCOL, MYROW, NPCOL, NPROW
*     ..
*     .. External Subroutines ..
      EXTERNAL           BLACS_GRIDINFO
*     ..
*     .. Executable Statements ..
*
*     Quick return if possible
*
      IF( n.LE.0 )
     $   RETURN
*
*     Get grid parameters
*
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      IF( myrow.EQ.irprnt .AND. mycol.EQ.icprnt ) THEN
*
         WRITE( nout, fmt = * )
         DO 10 i = 1, 1 + ( n-1 )*incx, incx
*
            WRITE( nout, fmt = 9999 ) cvecnm, i, x( i )
*
   10    CONTINUE
*
      END IF
*
 9999 FORMAT( 1x, a, '(', i6, ')=', e16.8 )
*
      RETURN
*
*     End of PSVPRNT
*
      END
      SUBROUTINE psmvch( ICTXT, TRANS, M, N, ALPHA, A, IA, JA, DESCA,
     $                   X, IX, JX, DESCX, INCX, BETA, Y, PY, IY, JY,
     $                   DESCY, INCY, G, ERR, INFO )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1        TRANS
      INTEGER            IA, ICTXT, INCX, INCY, INFO, IX, IY, JA, JX,
     $                   JY, M, N
      REAL               ALPHA, BETA, ERR
*     ..
*     .. Array Arguments ..
      INTEGER            DESCA( * ), DESCX( * ), DESCY( * )
      REAL               A( * ), G( * ), PY( * ), X( * ), Y( * )
*     ..
*
*  Purpose
*  =======
*
*  PSMVCH checks the results of the computational tests.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  TRANS   (global input) CHARACTER*1
*          On entry,  TRANS  specifies which matrix-vector product is to
*          be computed as follows:
*             If TRANS = 'N',
*                sub( Y ) = BETA * sub( Y ) + sub( A )  * sub( X ),
*             otherwise
*                sub( Y ) = BETA * sub( Y ) + sub( A )' * sub( X ).
*
*  M       (global input) INTEGER
*          On entry,  M  specifies  the  number of rows of the submatrix
*          operand matrix A. M must be at least zero.
*
*  N       (global input) INTEGER
*          On entry,  N  specifies  the  number of columns of the subma-
*          trix operand matrix A. N must be at least zero.
*
*  ALPHA   (global input) REAL
*          On entry, ALPHA specifies the scalar alpha.
*
*  A       (local input) REAL array
*          On entry, A is an array of  dimension  (DESCA( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PA.
*
*  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.
*
*  X       (local input) REAL array
*          On entry, X is an array of  dimension  (DESCX( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PX.
*
*  IX      (global input) INTEGER
*          On entry, IX  specifies X's global row index, which points to
*          the beginning of the submatrix sub( X ).
*
*  JX      (global input) INTEGER
*          On entry, JX  specifies X's global column index, which points
*          to the beginning of the submatrix sub( X ).
*
*  DESCX   (global and local input) INTEGER array
*          On entry, DESCX  is an integer array of dimension DLEN_. This
*          is the array descriptor for the matrix X.
*
*  INCX    (global input) INTEGER
*          On entry,  INCX   specifies  the  global  increment  for  the
*          elements of  X.  Only two values of  INCX   are  supported in
*          this version, namely 1 and M_X. INCX  must not be zero.
*
*  BETA    (global input) REAL
*          On entry, BETA specifies the scalar beta.
*
*  Y       (local input/local output) REAL array
*          On entry, Y is an array of  dimension  (DESCY( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PY.
*
*  PY      (local input) REAL array
*          On entry, PY is an array of dimension (DESCY( LLD_ ),*). This
*          array contains the local entries of the matrix PY.
*
*  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.
*
*  INCY    (global input) INTEGER
*          On entry,  INCY   specifies  the  global  increment  for  the
*          elements of  Y.  Only two values of  INCY   are  supported in
*          this version, namely 1 and M_Y. INCY  must not be zero.
*
*  G       (workspace) REAL array
*          On entry, G is an array of dimension at least MAX( M, N ).  G
*          is used to compute the gauges.
*
*  ERR     (global output) REAL
*          On exit, ERR specifies the largest error in absolute value.
*
*  INFO    (global output) INTEGER
*          On exit, if INFO <> 0, the result is less than half accurate.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
      REAL               ZERO, ONE
      parameter( zero = 0.0e+0, one = 1.0e+0 )
*     ..
*     .. Local Scalars ..
      LOGICAL            COLREP, ROWREP, TRAN
      INTEGER            I, IB, ICURCOL, ICURROW, IIY, IN, IOFFA, IOFFX,
     $                   IOFFY, IYCOL, IYROW, J, JB, JJY, JN, KK, LDA,
     $                   LDPY, LDX, LDY, ML, MYCOL, MYROW, NL, NPCOL,
     $                   nprow
      REAL               EPS, ERRI, GTMP, TBETA, YTMP
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_gridinfo, igsum2d, pb_infog2l, sgamx2d
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      REAL               PSLAMCH
      EXTERNAL           lsame, pslamch
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          abs, max, min, mod, sqrt
*     ..
*     .. Executable Statements ..
*
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      eps = pslamch( ictxt, 'eps' )
*
      IF( m.EQ.0 .OR. n.EQ.0 ) THEN
         tbeta = one
      ELSE
         tbeta = beta
      END IF
*
      tran = lsame( trans, 'T' ).OR.lsame( trans, 'C' )
      IF( tran ) THEN
         ml = n
         nl = m
      ELSE
         ml = m
         nl = n
      END IF
*
      lda = max( 1, desca( m_ ) )
      ldx = max( 1, descx( m_ ) )
      ldy = max( 1, descy( m_ ) )
*
*     Compute expected result in Y using data in A, X and Y.
*     Compute gauges in G. This part of the computation is performed
*     by every process in the grid.
*
      ioffy = iy + ( jy - 1 ) * ldy
      DO 30 i = 1, ml
         ytmp = zero
         gtmp = zero
         ioffx = ix + ( jx - 1 ) * ldx
         IF( tran )THEN
            ioffa = ia + ( ja + i - 2 ) * lda
            DO 10 j = 1, nl
               ytmp = ytmp + a( ioffa ) * x( ioffx )
               gtmp = gtmp + abs( a( ioffa ) * x( ioffx ) )
               ioffa = ioffa + 1
               ioffx = ioffx + incx
   10       CONTINUE
         ELSE
            ioffa = ia + i - 1 + ( ja - 1 ) * lda
            DO 20 j = 1, nl
               ytmp = ytmp + a( ioffa ) * x( ioffx )
               gtmp = gtmp + abs( a( ioffa ) * x( ioffx ) )
               ioffa = ioffa + lda
               ioffx = ioffx + incx
   20       CONTINUE
         END IF
         g( i ) = abs( alpha ) * gtmp + abs( tbeta * y( ioffy ) )
         y( ioffy ) = alpha * ytmp + tbeta * y( ioffy )
         ioffy = ioffy + incy
   30 CONTINUE
*
*     Compute the error ratio for this result.
*
      err  = zero
      info = 0
      ldpy = descy( lld_ )
      ioffy = iy + ( jy - 1 ) * ldy
      CALL pb_infog2l( iy, jy, descy, nprow, npcol, myrow, mycol, iiy,
     $                 jjy, iyrow, iycol )
      icurrow = iyrow
      icurcol = iycol
      rowrep  = ( iyrow.EQ.-1 )
      colrep  = ( iycol.EQ.-1 )
*
      IF( incy.EQ.descy( m_ ) ) THEN
*
*        sub( Y ) is a row vector
*
         jb = descy( inb_ ) - jy + 1
         IF( jb.LE.0 )
     $      jb = ( ( -jb ) / descy( nb_ ) + 1 ) * descy( nb_ ) + jb
         jb = min( jb, ml )
         jn = jy + jb - 1
*
         DO 50 j = jy, jn
*
            IF( ( myrow.EQ.icurrow .OR. rowrep ) .AND.
     $          ( mycol.EQ.icurcol .OR. colrep ) ) THEN
               erri = abs( py( iiy+(jjy-1)*ldpy ) - y( ioffy ) ) / eps
               IF( g( j-jy+1 ).NE.zero )
     $            erri = erri / g( j-jy+1 )
               err = max( err, erri )
               IF( err*sqrt( eps ).GE.one )
     $            info = 1
               jjy = jjy + 1
            END IF
*
            ioffy = ioffy + incy
*
   50    CONTINUE
*
         icurcol = mod( icurcol+1, npcol )
*
         DO 70 j = jn+1, jy+ml-1, descy( nb_ )
            jb = min( jy+ml-j, descy( nb_ ) )
*
            DO 60 kk = 0, jb-1
*
               IF( ( myrow.EQ.icurrow .OR. rowrep ) .AND.
     $             ( mycol.EQ.icurcol .OR. colrep ) ) THEN
                  erri = abs( py( iiy+(jjy-1)*ldpy ) - y( ioffy ) )/eps
                  IF( g( j+kk-jy+1 ).NE.zero )
     $               erri = erri / g( j+kk-jy+1 )
                  err = max( err, erri )
                  IF( err*sqrt( eps ).GE.one )
     $               info = 1
                  jjy = jjy + 1
               END IF
*
               ioffy = ioffy + incy
*
   60       CONTINUE
*
            icurcol = mod( icurcol+1, npcol )
*
   70    CONTINUE
*
      ELSE
*
*        sub( Y ) is a column vector
*
         ib = descy( imb_ ) - iy + 1
         IF( ib.LE.0 )
     $      ib = ( ( -ib ) / descy( mb_ ) + 1 ) * descy( mb_ ) + ib
         ib = min( ib, ml )
         in = iy + ib - 1
*
         DO 80 i = iy, in
*
            IF( ( myrow.EQ.icurrow .OR. rowrep ) .AND.
     $          ( mycol.EQ.icurcol .OR. colrep ) ) THEN
               erri = abs( py( iiy+(jjy-1)*ldpy ) - y( ioffy ) ) / eps
               IF( g( i-iy+1 ).NE.zero )
     $            erri = erri / g( i-iy+1 )
               err = max( err, erri )
               IF( err*sqrt( eps ).GE.one )
     $            info = 1
               iiy = iiy + 1
            END IF
*
            ioffy = ioffy + incy
*
   80    CONTINUE
*
         icurrow = mod( icurrow+1, nprow )
*
         DO 100 i = in+1, iy+ml-1, descy( mb_ )
            ib = min( iy+ml-i, descy( mb_ ) )
*
            DO 90 kk = 0, ib-1
*
               IF( ( myrow.EQ.icurrow .OR. rowrep ) .AND.
     $             ( mycol.EQ.icurcol .OR. colrep ) ) THEN
                  erri = abs( py( iiy+(jjy-1)*ldpy ) - y( ioffy ) )/eps
                  IF( g( i+kk-iy+1 ).NE.zero )
     $               erri = erri / g( i+kk-iy+1 )
                  err = max( err, erri )
                  IF( err*sqrt( eps ).GE.one )
     $               info = 1
                  iiy = iiy + 1
               END IF
*
               ioffy = ioffy + incy
*
   90       CONTINUE
*
            icurrow = mod( icurrow+1, nprow )
*
  100    CONTINUE
*
      END IF
*
*     If INFO = 0, all results are at least half accurate.
*
      CALL igsum2d( ictxt, 'All', ' ', 1, 1, info, 1, -1, mycol )
      CALL sgamx2d( ictxt, 'All', ' ', 1, 1, err, 1, i, j, -1, -1,
     $              mycol )
*
      RETURN
*
*     End of PSMVCH
*
      END
      SUBROUTINE psvmch( ICTXT, UPLO, M, N, ALPHA, X, IX, JX, DESCX,
     $                   INCX, Y, IY, JY, DESCY, INCY, A, PA, IA, JA,
     $                   DESCA, G, ERR, INFO )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1        UPLO
      INTEGER            IA, ICTXT, INCX, INCY, INFO, IX, IY, JA, JX,
     $                   JY, M, N
      REAL               ALPHA, ERR
*     ..
*     .. Array Arguments ..
      INTEGER            DESCA( * ), DESCX( * ), DESCY( * )
      REAL               A( * ), G( * ), PA( * ), X( * ), Y( * )
*     ..
*
*  Purpose
*  =======
*
*  PSVMCH checks the results of the computational tests.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  UPLO    (global input) CHARACTER*1
*          On entry, UPLO specifies which part of the submatrix sub( A )
*          is to be referenced as follows:
*             If UPLO = 'L', only the lower triangular part,
*             If UPLO = 'U', only the upper triangular part,
*             else the entire matrix is to be referenced.
*
*  M       (global input) INTEGER
*          On entry,  M  specifies  the  number of rows of the submatrix
*          operand matrix A. M must be at least zero.
*
*  N       (global input) INTEGER
*          On entry,  N  specifies  the  number of columns of the subma-
*          trix operand matrix A. N must be at least zero.
*
*  ALPHA   (global input) REAL
*          On entry, ALPHA specifies the scalar alpha.
*
*  X       (local input) REAL array
*          On entry, X is an array of  dimension  (DESCX( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PX.
*
*  IX      (global input) INTEGER
*          On entry, IX  specifies X's global row index, which points to
*          the beginning of the submatrix sub( X ).
*
*  JX      (global input) INTEGER
*          On entry, JX  specifies X's global column index, which points
*          to the beginning of the submatrix sub( X ).
*
*  DESCX   (global and local input) INTEGER array
*          On entry, DESCX  is an integer array of dimension DLEN_. This
*          is the array descriptor for the matrix X.
*
*  INCX    (global input) INTEGER
*          On entry,  INCX   specifies  the  global  increment  for  the
*          elements of  X.  Only two values of  INCX   are  supported in
*          this version, namely 1 and M_X. INCX  must not be zero.
*
*  Y       (local input) REAL array
*          On entry, Y is an array of  dimension  (DESCY( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PY.
*
*  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.
*
*  INCY    (global input) INTEGER
*          On entry,  INCY   specifies  the  global  increment  for  the
*          elements of  Y.  Only two values of  INCY   are  supported in
*          this version, namely 1 and M_Y. INCY  must not be zero.
*
*  A       (local input/local output) REAL array
*          On entry, A is an array of  dimension  (DESCA( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PA.
*
*  PA      (local input) REAL array
*          On entry, PA is an array of dimension (DESCA( LLD_ ),*). This
*          array contains the local entries of the matrix PA.
*
*  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.
*
*  G       (workspace) REAL array
*          On entry, G is an array of dimension at least MAX( M, N ).  G
*          is used to compute the gauges.
*
*  ERR     (global output) REAL
*          On exit, ERR specifies the largest error in absolute value.
*
*  INFO    (global output) INTEGER
*          On exit, if INFO <> 0, the result is less than half accurate.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
      REAL               ZERO, ONE
      PARAMETER          ( ZERO = 0.0e+0, one = 1.0e+0 )
*     ..
*     .. Local Scalars ..
      LOGICAL            COLREP, LOWER, ROWREP, UPPER
      INTEGER            I, IACOL, IAROW, IB, IBEG, ICURROW, IEND, IIA,
     $                   in, ioffa, ioffx, ioffy, j, jja, kk, lda, ldpa,
     $                   ldx, ldy, mycol, myrow, npcol, nprow
      REAL               ATMP, EPS, ERRI, GTMP
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_gridinfo, igsum2d, pb_infog2l, sgamx2d
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      REAL               PSLAMCH
      EXTERNAL           LSAME, PSLAMCH
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          abs, max, min, mod, sqrt
*     ..
*     .. Executable Statements ..
*
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      eps = pslamch( ictxt, 'eps' )
*
      upper = lsame( uplo, 'U' )
      lower = lsame( uplo, 'L' )
*
      lda = max( 1, desca( m_ ) )
      ldx = max( 1, descx( m_ ) )
      ldy = max( 1, descy( m_ ) )
*
*     Compute expected result in A using data in A, X and Y.
*     Compute gauges in G. This part of the computation is performed
*     by every process in the grid.
*
      DO 70 j = 1, n
*
         ioffy = iy + ( jy - 1 ) * ldy + ( j - 1 ) * incy
*
         IF( lower ) THEN
            ibeg = j
            iend = m
            DO 10 i = 1, j-1
               g( i ) = zero
   10       CONTINUE
         ELSE IF( upper ) THEN
            ibeg = 1
            iend = j
            DO 20 i = j+1, m
               g( i ) = zero
   20       CONTINUE
         ELSE
            ibeg = 1
            iend = m
         END IF
*
         DO 30 i = ibeg, iend
*
            ioffx = ix + ( jx - 1 ) * ldx + ( i - 1 ) * incx
            ioffa = ia + i - 1 + ( ja + j - 2 ) * lda
            atmp = x( ioffx ) * y( ioffy )
            gtmp = abs( x( ioffx ) * y( ioffy ) )
            g( i ) = abs( alpha ) * gtmp + abs( a( ioffa ) )
            a( ioffa ) = alpha * atmp + a( ioffa )
*
   30    CONTINUE
*
*        Compute the error ratio for this result.
*
         info = 0
         err  = zero
         ldpa = desca( lld_ )
         ioffa = ia + ( ja + j - 2 ) * lda
         CALL pb_infog2l( ia, ja+j-1, desca, nprow, npcol, myrow, mycol,
     $                    iia, jja, iarow, iacol )
         rowrep = ( iarow.EQ.-1 )
         colrep = ( iacol.EQ.-1 )
*
         IF( mycol.EQ.iacol .OR. colrep ) THEN
*
            icurrow = iarow
            ib = desca( imb_ ) - ia + 1
            IF( ib.LE.0 )
     $         ib = ( ( -ib ) / desca( mb_ ) + 1 ) * desca( mb_ ) + ib
            ib = min( ib, m )
            in = ia + ib - 1
*
            DO 40 i = ia, in
*
               IF( myrow.EQ.icurrow .OR. rowrep ) THEN
                  erri = abs( pa( iia+(jja-1)*ldpa ) - a( ioffa ) )/eps
                  IF( g( i-ia+1 ).NE.zero )
     $               erri = erri / g( i-ia+1 )
                  err = max( err, erri )
                  IF( err*sqrt( eps ).GE.one )
     $               info = 1
                  iia = iia + 1
               END IF
*
               ioffa = ioffa + 1
*
   40       CONTINUE
*
            icurrow = mod( icurrow+1, nprow )
*
            DO 60 i = in+1, ia+m-1, desca( mb_ )
               ib = min( ia+m-i, desca( mb_ ) )
*
               DO 50 kk = 0, ib-1
*
                  IF( myrow.EQ.icurrow .OR. rowrep ) THEN
                     erri = abs( pa( iia+(jja-1)*ldpa )-a( ioffa ) )/eps
                     IF( g( i+kk-ia+1 ).NE.zero )
     $                  erri = erri / g( i+kk-ia+1 )
                     err = max( err, erri )
                     IF( err*sqrt( eps ).GE.one )
     $                  info = 1
                     iia = iia + 1
                  END IF
*
                  ioffa = ioffa + 1
*
   50          CONTINUE
*
               icurrow = mod( icurrow+1, nprow )
*
   60       CONTINUE
*
         END IF
*
*        If INFO = 0, all results are at least half accurate.
*
         CALL igsum2d( ictxt, 'All', ' ', 1, 1, info, 1, -1, mycol )
         CALL sgamx2d( ictxt, 'All', ' ', 1, 1, err, 1, i, j, -1, -1,
     $                 mycol )
         IF( info.NE.0 )
     $      GO TO 80
*
   70 CONTINUE
*
   80 CONTINUE
*
      RETURN
*
*     End of PSVMCH
*
      END
      SUBROUTINE psvmch2( ICTXT, UPLO, M, N, ALPHA, X, IX, JX, DESCX,
     $                    INCX, Y, IY, JY, DESCY, INCY, A, PA, IA,
     $                    JA, DESCA, G, ERR, INFO )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1        UPLO
      INTEGER            IA, ICTXT, INCX, INCY, INFO, IX, IY, JA, JX,
     $                   jy, m, n
      REAL               ALPHA, ERR
*     ..
*     .. Array Arguments ..
      INTEGER            DESCA( * ), DESCX( * ), DESCY( * )
      REAL               A( * ), G( * ), PA( * ), X( * ), Y( * )
*     ..
*
*  Purpose
*  =======
*
*  PSVMCH2 checks the results of the computational tests.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  UPLO    (global input) CHARACTER*1
*          On entry, UPLO specifies which part of the submatrix sub( A )
*          is to be referenced as follows:
*             If UPLO = 'L', only the lower triangular part,
*             If UPLO = 'U', only the upper triangular part,
*             else the entire matrix is to be referenced.
*
*  M       (global input) INTEGER
*          On entry,  M  specifies  the  number of rows of the submatrix
*          operand matrix A. M must be at least zero.
*
*  N       (global input) INTEGER
*          On entry,  N  specifies  the  number of columns of the subma-
*          trix operand matrix A. N must be at least zero.
*
*  ALPHA   (global input) REAL
*          On entry, ALPHA specifies the scalar alpha.
*
*  X       (local input) REAL array
*          On entry, X is an array of  dimension  (DESCX( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PX.
*
*  IX      (global input) INTEGER
*          On entry, IX  specifies X's global row index, which points to
*          the beginning of the submatrix sub( X ).
*
*  JX      (global input) INTEGER
*          On entry, JX  specifies X's global column index, which points
*          to the beginning of the submatrix sub( X ).
*
*  DESCX   (global and local input) INTEGER array
*          On entry, DESCX  is an integer array of dimension DLEN_. This
*          is the array descriptor for the matrix X.
*
*  INCX    (global input) INTEGER
*          On entry,  INCX   specifies  the  global  increment  for  the
*          elements of  X.  Only two values of  INCX   are  supported in
*          this version, namely 1 and M_X. INCX  must not be zero.
*
*  Y       (local input) REAL array
*          On entry, Y is an array of  dimension  (DESCY( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PY.
*
*  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.
*
*  INCY    (global input) INTEGER
*          On entry,  INCY   specifies  the  global  increment  for  the
*          elements of  Y.  Only two values of  INCY   are  supported in
*          this version, namely 1 and M_Y. INCY  must not be zero.
*
*  A       (local input/local output) REAL array
*          On entry, A is an array of  dimension  (DESCA( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PA.
*
*  PA      (local input) REAL array
*          On entry, PA is an array of dimension (DESCA( LLD_ ),*). This
*          array contains the local entries of the matrix PA.
*
*  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.
*
*  G       (workspace) REAL array
*          On entry, G is an array of dimension at least MAX( M, N ).  G
*          is used to compute the gauges.
*
*  ERR     (global output) REAL
*          On exit, ERR specifies the largest error in absolute value.
*
*  INFO    (global output) INTEGER
*          On exit, if INFO <> 0, the result is less than half accurate.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
      REAL               ZERO, ONE
      PARAMETER          ( ZERO = 0.0e+0, one = 1.0e+0 )
*     ..
*     .. Local Scalars ..
      LOGICAL            COLREP, LOWER, ROWREP, UPPER
      INTEGER            I, IACOL, IAROW, IB, IBEG, ICURROW, IEND, IIA,
     $                   IN, IOFFA, IOFFXI, IOFFXJ, IOFFYI, IOFFYJ, J,
     $                   JJA, KK, LDA, LDPA, LDX, LDY, MYCOL, MYROW,
     $                   npcol, nprow
      REAL               EPS, ERRI, GTMP, ATMP
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_gridinfo, igsum2d, pb_infog2l, sgamx2d
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      REAL               PSLAMCH
      EXTERNAL           lsame, pslamch
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          abs, max, min, mod, sqrt
*     ..
*     .. Executable Statements ..
*
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      eps = pslamch( ictxt, 'eps' )
*
      upper = lsame( uplo, 'U' )
      lower = lsame( uplo, 'L' )
*
      lda = max( 1, desca( m_ ) )
      ldx = max( 1, descx( m_ ) )
      ldy = max( 1, descy( m_ ) )
*
*     Compute expected result in A using data in A, X and Y.
*     Compute gauges in G. This part of the computation is performed
*     by every process in the grid.
*
      DO 70 j = 1, n
*
         ioffxj = ix + ( jx - 1 ) * ldx + ( j - 1 ) * incx
         ioffyj = iy + ( jy - 1 ) * ldy + ( j - 1 ) * incy
*
         IF( lower ) THEN
            ibeg = j
            iend = m
            DO 10 i = 1, j-1
               g( i ) = zero
   10       CONTINUE
         ELSE IF( upper ) THEN
            ibeg = 1
            iend = j
            DO 20 i = j+1, m
               g( i ) = zero
   20       CONTINUE
         ELSE
            ibeg = 1
            iend = m
         END IF
*
         DO 30 i = ibeg, iend
            ioffa = ia + i - 1 + ( ja + j - 2 ) * lda
            ioffxi = ix + ( jx - 1 ) * ldx + ( i - 1 ) * incx
            ioffyi = iy + ( jy - 1 ) * ldy + ( i - 1 ) * incy
            atmp = x( ioffxi ) * y( ioffyj )
            atmp = atmp + y( ioffyi ) * x( ioffxj )
            gtmp = abs( x( ioffxi ) * y( ioffyj ) )
            gtmp = gtmp + abs( y( ioffyi ) * x( ioffxj ) )
            g( i ) = abs( alpha ) * gtmp + abs( a( ioffa ) )
            a( ioffa ) = alpha*atmp + a( ioffa )
*
   30    CONTINUE
*
*        Compute the error ratio for this result.
*
         info = 0
         err  = zero
         ldpa = desca( lld_ )
         ioffa = ia + ( ja + j - 2 ) * lda
         CALL pb_infog2l( ia, ja+j-1, desca, nprow, npcol, myrow, mycol,
     $                    iia, jja, iarow, iacol )
         rowrep = ( iarow.EQ.-1 )
         colrep = ( iacol.EQ.-1 )
*
         IF( mycol.EQ.iacol .OR. colrep ) THEN
*
            icurrow = iarow
            ib = desca( imb_ ) - ia + 1
            IF( ib.LE.0 )
     $         ib = ( ( -ib ) / desca( mb_ ) + 1 ) * desca( mb_ ) + ib
            ib = min( ib, m )
            in = ia + ib - 1
*
            DO 40 i = ia, in
*
               IF( myrow.EQ.icurrow .OR. rowrep ) THEN
                  erri = abs( pa( iia+(jja-1)*ldpa ) - a( ioffa ) )/eps
                  IF( g( i-ia+1 ).NE.zero )
     $               erri = erri / g( i-ia+1 )
                  err = max( err, erri )
                  IF( err*sqrt( eps ).GE.one )
     $               info = 1
                  iia = iia + 1
               END IF
*
               ioffa = ioffa + 1
*
   40       CONTINUE
*
            icurrow = mod( icurrow+1, nprow )
*
            DO 60 i = in+1, ia+m-1, desca( mb_ )
               ib = min( ia+m-i, desca( mb_ ) )
*
               DO 50 kk = 0, ib-1
*
                  IF( myrow.EQ.icurrow .OR. rowrep ) THEN
                     erri = abs( pa( iia+(jja-1)*ldpa )-a( ioffa ) )/eps
                     IF( g( i+kk-ia+1 ).NE.zero )
     $                  erri = erri / g( i+kk-ia+1 )
                     err = max( err, erri )
                     IF( err*sqrt( eps ).GE.one )
     $                  info = 1
                     iia = iia + 1
                  END IF
*
                  ioffa = ioffa + 1
*
   50          CONTINUE
*
               icurrow = mod( icurrow+1, nprow )
*
   60       CONTINUE
*
         END IF
*
*        If INFO = 0, all results are at least half accurate.
*
         CALL igsum2d( ictxt, 'All', ' ', 1, 1, info, 1, -1, mycol )
         CALL sgamx2d( ictxt, 'All', ' ', 1, 1, err, 1, i, j, -1, -1,
     $                 mycol )
         IF( info.NE.0 )
     $      GO TO 80
*
   70 CONTINUE
*
   80 CONTINUE
*
      RETURN
*
*     End of PSVMCH2
*
      END
      SUBROUTINE psmmch( ICTXT, TRANSA, TRANSB, M, N, K, ALPHA, A, IA,
     $                   JA, DESCA, B, IB, JB, DESCB, BETA, C, PC, IC,
     $                   JC, DESCC, CT, G, ERR, INFO )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1        TRANSA, TRANSB
      INTEGER            IA, IB, IC, ICTXT, INFO, JA, JB, JC, K, M, N
      REAL               ALPHA, BETA, ERR
*     ..
*     .. Array Arguments ..
      INTEGER            DESCA( * ), DESCB( * ), DESCC( * )
      REAL               A( * ), B( * ), C( * ), CT( * ), G( * ),
     $                   PC( * )
*     ..
*
*  Purpose
*  =======
*
*  PSMMCH checks the results of the computational tests.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  TRANSA  (global input) CHARACTER*1
*          On entry, TRANSA specifies if the matrix  operand  A is to be
*          transposed.
*
*  TRANSB  (global input) CHARACTER*1
*          On entry, TRANSB specifies if the matrix  operand  B is to be
*          transposed.
*
*  M       (global input) INTEGER
*          On entry, M specifies the number of rows of C.
*
*  N       (global input) INTEGER
*          On entry, N specifies the number of columns of C.
*
*  K       (global input) INTEGER
*          On entry, K specifies the number of columns (resp. rows) of A
*          when  TRANSA = 'N'  (resp. TRANSA <> 'N')  in PxGEMM, PxSYRK,
*          PxSYR2K, PxHERK and PxHER2K.
*
*  ALPHA   (global input) REAL
*          On entry, ALPHA specifies the scalar alpha.
*
*  A       (local input) REAL array
*          On entry, A is an array of  dimension  (DESCA( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PA.
*
*  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) REAL array
*          On entry, B is an array of  dimension  (DESCB( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PB.
*
*  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) REAL
*          On entry, BETA specifies the scalar beta.
*
*  C       (local input/local output) REAL array
*          On entry, C is an array of  dimension  (DESCC( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PC.
*
*  PC      (local input) REAL array
*          On entry, PC is an array of dimension (DESCC( LLD_ ),*). This
*          array contains the local pieces of the matrix PC.
*
*  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.
*
*  CT      (workspace) REAL array
*          On entry, CT is an array of dimension at least MAX(M,N,K). CT
*          holds a copy of the current column of C.
*
*  G       (workspace) REAL array
*          On entry, G  is  an array of dimension at least MAX(M,N,K). G
*          is used to compute the gauges.
*
*  ERR     (global output) REAL
*          On exit, ERR specifies the largest error in absolute value.
*
*  INFO    (global output) INTEGER
*          On exit, if INFO <> 0, the result is less than half accurate.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
      REAL               ZERO, ONE
      PARAMETER          ( ZERO = 0.0e+0, one = 1.0e+0 )
*     ..
*     .. Local Scalars ..
      LOGICAL            COLREP, ROWREP, TRANA, TRANB
      INTEGER            I, IBB, ICCOL, ICROW, ICURROW, IIC, IN, IOFFA,
     $                   IOFFB, IOFFC, J, JJC, KK, LDA, LDB, LDC, LDPC,
     $                   mycol, myrow, npcol, nprow
      REAL               EPS, ERRI
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_gridinfo, igsum2d, pb_infog2l, sgamx2d
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      REAL               PSLAMCH
      EXTERNAL           LSAME, PSLAMCH
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          abs, max, min, mod, sqrt
*     ..
*     .. Executable Statements ..
*
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      eps = pslamch( ictxt, 'eps' )
*
      trana = lsame( transa, 'T' ).OR.lsame( transa, 'C' )
      tranb = lsame( transb, 'T' ).OR.lsame( transb, 'C' )
*
      lda = max( 1, desca( m_ ) )
      ldb = max( 1, descb( m_ ) )
      ldc = max( 1, descc( m_ ) )
*
*     Compute expected result in C using data in A, B and C.
*     Compute gauges in G. This part of the computation is performed
*     by every process in the grid.
*
      DO 240 j = 1, n
*
         ioffc = ic + ( jc + j - 2 ) * ldc
         DO 10 i = 1, m
            ct( i ) = zero
            g( i )  = zero
   10    CONTINUE
*
         IF( .NOT.trana .AND. .NOT.tranb ) THEN
            DO 30 kk = 1, k
               ioffb = ib + kk - 1 + ( jb + j - 2 ) * ldb
               DO 20 i = 1, m
                  ioffa = ia + i - 1 + ( ja + kk - 2 ) * lda
                  ct( i ) = ct( i ) + a( ioffa ) * b( ioffb )
                  g( i ) = g( i ) + abs( a( ioffa ) ) *
     $                     abs( b( ioffb ) )
   20          CONTINUE
   30       CONTINUE
         ELSE IF( trana .AND. .NOT.tranb ) THEN
            DO 50 kk = 1, k
               ioffb = ib + kk - 1 + ( jb + j - 2 ) * ldb
               DO 40 i = 1, m
                  ioffa = ia + kk - 1 + ( ja + i - 2 ) * lda
                  ct( i ) = ct( i ) + a( ioffa ) * b( ioffb )
                  g( i ) = g( i ) + abs( a( ioffa ) ) *
     $                     abs( b( ioffb ) )
   40          CONTINUE
   50       CONTINUE
         ELSE IF( .NOT.trana .AND. tranb ) THEN
            DO 70 kk = 1, k
               ioffb = ib + j - 1 + ( jb + kk - 2 ) * ldb
               DO 60 i = 1, m
                  ioffa = ia + i - 1 + ( ja + kk - 2 ) * lda
                  ct( i ) = ct( i ) + a( ioffa ) * b( ioffb )
                  g( i ) = g( i ) + abs( a( ioffa ) ) *
     $                     abs( b( ioffb ) )
   60          CONTINUE
   70       CONTINUE
         ELSE IF( trana .AND. tranb ) THEN
            DO 90 kk = 1, k
               ioffb = ib + j - 1 + ( jb + kk - 2 ) * ldb
               DO 80 i = 1, m
                  ioffa = ia + kk - 1 + ( ja + i - 2 ) * lda
                  ct( i ) = ct( i ) + a( ioffa ) * b( ioffb )
                  g( i ) = g( i ) + abs( a( ioffa ) ) *
     $                     abs( b( ioffb ) )
   80          CONTINUE
   90       CONTINUE
         END IF
*
         DO 200 i = 1, m
            ct( i ) = alpha*ct( i ) + beta * c( ioffc )
            g( i ) = abs( alpha )*g( i ) + abs( beta )*abs( c( ioffc ) )
            c( ioffc ) = ct( i )
            ioffc      = ioffc + 1
  200    CONTINUE
*
*        Compute the error ratio for this result.
*
         err  = zero
         info = 0
         ldpc = descc( lld_ )
         ioffc = ic + ( jc + j - 2 ) * ldc
         CALL pb_infog2l( ic, jc+j-1, descc, nprow, npcol, myrow, mycol,
     $                    iic, jjc, icrow, iccol )
         icurrow = icrow
         rowrep  = ( icrow.EQ.-1 )
         colrep  = ( iccol.EQ.-1 )
*
         IF( mycol.EQ.iccol .OR. colrep ) THEN
*
            ibb = descc( imb_ ) - ic + 1
            IF( ibb.LE.0 )
     $         ibb = ( ( -ibb ) / descc( mb_ ) + 1 )*descc( mb_ ) + ibb
            ibb = min( ibb, m )
            in = ic + ibb - 1
*
            DO 210 i = ic, in
*
               IF( myrow.EQ.icurrow .OR. rowrep ) THEN
                  erri = abs( pc( iic+(jjc-1)*ldpc ) -
     $                        c( ioffc ) ) / eps
                  IF( g( i-ic+1 ).NE.zero )
     $               erri = erri / g( i-ic+1 )
                  err = max( err, erri )
                  IF( err*sqrt( eps ).GE.one )
     $               info = 1
                  iic = iic + 1
               END IF
*
               ioffc = ioffc + 1
*
  210       CONTINUE
*
            icurrow = mod( icurrow+1, nprow )
*
            DO 230 i = in+1, ic+m-1, descc( mb_ )
               ibb = min( ic+m-i, descc( mb_ ) )
*
               DO 220 kk = 0, ibb-1
*
                  IF( myrow.EQ.icurrow .OR. rowrep ) THEN
                     erri = abs( pc( iic+(jjc-1)*ldpc ) -
     $                           c( ioffc ) )/eps
                     IF( g( i+kk-ic+1 ).NE.zero )
     $                  erri = erri / g( i+kk-ic+1 )
                     err = max( err, erri )
                     IF( err*sqrt( eps ).GE.one )
     $                  info = 1
                     iic = iic + 1
                  END IF
*
                  ioffc = ioffc + 1
*
  220          CONTINUE
*
               icurrow = mod( icurrow+1, nprow )
*
  230       CONTINUE
*
         END IF
*
*        If INFO = 0, all results are at least half accurate.
*
         CALL igsum2d( ictxt, 'All', ' ', 1, 1, info, 1, -1, mycol )
         CALL sgamx2d( ictxt, 'All', ' ', 1, 1, err, 1, i, j, -1, -1,
     $                 mycol )
         IF( info.NE.0 )
     $      GO TO 250
*
  240 CONTINUE
*
  250 CONTINUE
*
      RETURN
*
*     End of PSMMCH
*
      END
      SUBROUTINE psmmch1( ICTXT, UPLO, TRANS, N, K, ALPHA, A, IA, JA,
     $                    DESCA, BETA, C, PC, IC, JC, DESCC, CT, G,
     $                    ERR, INFO )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1        TRANS, UPLO
      INTEGER            IA, IC, ICTXT, INFO, JA, JC, K, N
      REAL               ALPHA, BETA, ERR
*     ..
*     .. Array Arguments ..
      INTEGER            DESCA( * ), DESCC( * )
      REAL               A( * ), C( * ), CT( * ), G( * ), PC( * )
*     ..
*
*  Purpose
*  =======
*
*  PSMMCH1 checks the results of the computational tests.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  UPLO    (global input) CHARACTER*1
*          On entry,  UPLO  specifies which part of C should contain the
*          result.
*
*  TRANS   (global input) CHARACTER*1
*          On entry,  TRANS  specifies  whether  the  matrix A has to be
*          transposed or not before computing the matrix-matrix product.
*
*  N       (global input) INTEGER
*          On entry, N  specifies  the order  the submatrix operand C. N
*          must be at least zero.
*
*  K       (global input) INTEGER
*          On entry, K specifies the number of columns (resp. rows) of A
*          when  TRANS = 'N'  (resp. TRANS <> 'N').  K  must be at least
*          zero.
*
*  ALPHA   (global input) REAL
*          On entry, ALPHA specifies the scalar alpha.
*
*  A       (local input) REAL array
*          On entry, A is an array of  dimension  (DESCA( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PA.
*
*  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.
*
*  BETA    (global input) REAL
*          On entry, BETA specifies the scalar beta.
*
*  C       (local input/local output) REAL array
*          On entry, C is an array of  dimension  (DESCC( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PC.
*
*  PC      (local input) REAL array
*          On entry, PC is an array of dimension (DESCC( LLD_ ),*). This
*          array contains the local pieces of the matrix PC.
*
*  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.
*
*  CT      (workspace) REAL array
*          On entry, CT is an array of dimension at least MAX(M,N,K). CT
*          holds a copy of the current column of C.
*
*  G       (workspace) REAL array
*          On entry, G  is  an array of dimension at least MAX(M,N,K). G
*          is used to compute the gauges.
*
*  ERR     (global output) REAL
*          On exit, ERR specifies the largest error in absolute value.
*
*  INFO    (global output) INTEGER
*          On exit, if INFO <> 0, the result is less than half accurate.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
      REAL               ZERO, ONE
      PARAMETER          ( ZERO = 0.0e+0, one = 1.0e+0 )
*     ..
*     .. Local Scalars ..
      LOGICAL            COLREP, NOTRAN, ROWREP, TRAN, UPPER
      INTEGER            I, IBB, IBEG, ICCOL, ICROW, ICURROW, IEND, IIC,
     $                   IN, IOFFAK, IOFFAN, IOFFC, J, JJC, KK, LDA,
     $                   LDC, LDPC, MYCOL, MYROW, NPCOL, NPROW
      REAL               EPS, ERRI
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_gridinfo, igsum2d, pb_infog2l, sgamx2d
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      REAL               PSLAMCH
      EXTERNAL           lsame, pslamch
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          abs, max, min, mod, sqrt
*     ..
*     .. Executable Statements ..
*
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      eps = pslamch( ictxt, 'eps' )
*
      upper  = lsame( uplo,  'U' )
      notran = lsame( trans, 'N' )
      tran   = lsame( trans, 'T' )
*
      lda = max( 1, desca( m_ ) )
      ldc = max( 1, descc( m_ ) )
*
*     Compute expected result in C using data in A, B and C.
*     Compute gauges in G. This part of the computation is performed
*     by every process in the grid.
*
      DO 140 j = 1, n
*
         IF( upper ) THEN
            ibeg = 1
            iend = j
         ELSE
            ibeg = j
            iend = n
         END IF
*
         DO 10 i = 1, n
            ct( i ) = zero
            g( i )  = zero
   10    CONTINUE
*
         IF( notran ) THEN
            DO 30 kk = 1, k
               ioffak = ia + j - 1 + ( ja + kk - 2 ) * lda
               DO 20 i = ibeg, iend
                  ioffan = ia + i - 1 + ( ja + kk - 2 ) * lda
                  ct( i ) = ct( i ) + a( ioffak ) * a( ioffan )
                  g( i ) = g( i ) + abs( a( ioffak ) ) *
     $                     abs( a( ioffan ) )
   20          CONTINUE
   30       CONTINUE
         ELSE IF( tran ) THEN
            DO 50 kk = 1, k
               ioffak = ia + kk - 1 + ( ja + j - 2 ) * lda
               DO 40 i = ibeg, iend
                  ioffan = ia + kk - 1 + ( ja + i - 2 ) * lda
                  ct( i ) = ct( i ) + a( ioffak ) * a( ioffan )
                  g( i ) = g( i ) + abs( a( ioffak ) ) *
     $                     abs( a( ioffan ) )
   40          CONTINUE
   50       CONTINUE
         END IF
*
         ioffc = ic + ibeg - 1 + ( jc + j - 2 ) * ldc
*
         DO 100 i = ibeg, iend
            ct( i ) = alpha*ct( i ) + beta * c( ioffc )
            g( i ) = abs( alpha )*g( i ) + abs( beta )*abs( c( ioffc ) )
            c( ioffc ) = ct( i )
            ioffc = ioffc + 1
  100    CONTINUE
*
*        Compute the error ratio for this result.
*
         err  = zero
         info = 0
         ldpc = descc( lld_ )
         ioffc = ic + ( jc + j - 2 ) * ldc
         CALL pb_infog2l( ic, jc+j-1, descc, nprow, npcol, myrow, mycol,
     $                    iic, jjc, icrow, iccol )
         icurrow = icrow
         rowrep  = ( icrow.EQ.-1 )
         colrep  = ( iccol.EQ.-1 )
*
         IF( mycol.EQ.iccol .OR. colrep ) THEN
*
            ibb = descc( imb_ ) - ic + 1
            IF( ibb.LE.0 )
     $         ibb = ( ( -ibb ) / descc( mb_ ) + 1 )*descc( mb_ ) + ibb
            ibb = min( ibb, n )
            in = ic + ibb - 1
*
            DO 110 i = ic, in
*
               IF( myrow.EQ.icurrow .OR. rowrep ) THEN
                  erri = abs( pc( iic+(jjc-1)*ldpc ) -
     $                        c( ioffc ) ) / eps
                  IF( g( i-ic+1 ).NE.zero )
     $               erri = erri / g( i-ic+1 )
                  err = max( err, erri )
                  IF( err*sqrt( eps ).GE.one )
     $               info = 1
                  iic = iic + 1
               END IF
*
               ioffc = ioffc + 1
*
  110       CONTINUE
*
            icurrow = mod( icurrow+1, nprow )
*
            DO 130 i = in+1, ic+n-1, descc( mb_ )
               ibb = min( ic+n-i, descc( mb_ ) )
*
               DO 120 kk = 0, ibb-1
*
                  IF( myrow.EQ.icurrow .OR. rowrep ) THEN
                     erri = abs( pc( iic+(jjc-1)*ldpc ) -
     $                           c( ioffc ) )/eps
                     IF( g( i+kk-ic+1 ).NE.zero )
     $                  erri = erri / g( i+kk-ic+1 )
                     err = max( err, erri )
                     IF( err*sqrt( eps ).GE.one )
     $                  info = 1
                     iic = iic + 1
                  END IF
*
                  ioffc = ioffc + 1
*
  120          CONTINUE
*
               icurrow = mod( icurrow+1, nprow )
*
  130       CONTINUE
*
         END IF
*
*        If INFO = 0, all results are at least half accurate.
*
         CALL igsum2d( ictxt, 'All', ' ', 1, 1, info, 1, -1, mycol )
         CALL sgamx2d( ictxt, 'All', ' ', 1, 1, err, 1, i, j, -1, -1,
     $                 mycol )
         IF( info.NE.0 )
     $      GO TO 150
*
  140 CONTINUE
*
  150 CONTINUE
*
      RETURN
*
*     End of PSMMCH1
*
      END
      SUBROUTINE psmmch2( ICTXT, UPLO, TRANS, N, K, ALPHA, A, IA, JA,
     $                    DESCA, B, IB, JB, DESCB, BETA, C, PC, IC,
     $                    JC, DESCC, CT, G, ERR, INFO )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1        TRANS, UPLO
      INTEGER            IA, IB, IC, ICTXT, INFO, JA, JB, JC, K, N
      REAL               ALPHA, BETA, ERR
*     ..
*     .. Array Arguments ..
      INTEGER            DESCA( * ), DESCB( * ), DESCC( * )
      REAL               A( * ), B( * ), C( * ), CT( * ), G( * ),
     $                   pc( * )
*     ..
*
*  Purpose
*  =======
*
*  PSMMCH2 checks the results of the computational tests.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  UPLO    (global input) CHARACTER*1
*          On entry,  UPLO  specifies which part of C should contain the
*          result.
*
*  TRANS   (global input) CHARACTER*1
*          On entry,  TRANS  specifies whether the matrices A and B have
*          to  be  transposed  or not before computing the matrix-matrix
*          product.
*
*  N       (global input) INTEGER
*          On entry, N  specifies  the order  the submatrix operand C. N
*          must be at least zero.
*
*  K       (global input) INTEGER
*          On entry, K specifies the number of columns (resp. rows) of A
*          and B when  TRANS = 'N' (resp. TRANS <> 'N').  K  must  be at
*          least zero.
*
*  ALPHA   (global input) REAL
*          On entry, ALPHA specifies the scalar alpha.
*
*  A       (local input) REAL array
*          On entry, A is an array of  dimension  (DESCA( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PA.
*
*  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) REAL array
*          On entry, B is an array of  dimension  (DESCB( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PB.
*
*  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) REAL
*          On entry, BETA specifies the scalar beta.
*
*  C       (local input/local output) REAL array
*          On entry, C is an array of  dimension  (DESCC( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PC.
*
*  PC      (local input) REAL array
*          On entry, PC is an array of dimension (DESCC( LLD_ ),*). This
*          array contains the local pieces of the matrix PC.
*
*  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.
*
*  CT      (workspace) REAL array
*          On entry, CT is an array of dimension at least MAX(M,N,K). CT
*          holds a copy of the current column of C.
*
*  G       (workspace) REAL array
*          On entry, G  is  an array of dimension at least MAX(M,N,K). G
*          is used to compute the gauges.
*
*  ERR     (global output) REAL
*          On exit, ERR specifies the largest error in absolute value.
*
*  INFO    (global output) INTEGER
*          On exit, if INFO <> 0, the result is less than half accurate.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
      REAL               ZERO, ONE
      PARAMETER          ( ZERO = 0.0e+0, one = 1.0e+0 )
*     ..
*     .. Local Scalars ..
      LOGICAL            COLREP, NOTRAN, ROWREP, TRAN, UPPER
      INTEGER            I, IBB, IBEG, ICCOL, ICROW, ICURROW, IEND, IIC,
     $                   IN, IOFFAK, IOFFAN, IOFFBK, IOFFBN, IOFFC, J,
     $                   JJC, KK, LDA, LDB, LDC, LDPC, MYCOL, MYROW,
     $                   NPCOL, NPROW
      REAL               EPS, ERRI
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_gridinfo, igsum2d, pb_infog2l, sgamx2d
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      REAL               PSLAMCH
      EXTERNAL           LSAME, PSLAMCH
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          abs, max, min, mod, sqrt
*     ..
*     .. Executable Statements ..
*
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      eps = pslamch( ictxt, 'eps' )
*
      upper = lsame( uplo, 'U' )
      notran = lsame( trans, 'N' )
      tran = lsame( trans, 'T' )
*
      lda = max( 1, desca( m_ ) )
      ldb = max( 1, descb( m_ ) )
      ldc = max( 1, descc( m_ ) )
*
*     Compute expected result in C using data in A, B and C.
*     Compute gauges in G. This part of the computation is performed
*     by every process in the grid.
*
      DO 140 j = 1, n
*
         IF( upper ) THEN
            ibeg = 1
            iend = j
         ELSE
            ibeg = j
            iend = n
         END IF
*
         DO 10 i = 1, n
            ct( i ) = zero
            g( i )  = zero
   10    CONTINUE
*
         IF( notran ) THEN
            DO 30 kk = 1, k
               ioffak = ia + j - 1 + ( ja + kk - 2 ) * lda
               ioffbk = ib + j - 1 + ( jb + kk - 2 ) * ldb
               DO 20 i = ibeg, iend
                  ioffan = ia + i - 1 + ( ja + kk - 2 ) * lda
                  ioffbn = ib + i - 1 + ( jb + kk - 2 ) * ldb
                  ct( i ) = ct( i ) + alpha * (
     $                      a( ioffan ) * b( ioffbk ) +
     $                      b( ioffbn ) * a( ioffak ) )
                  g( i ) = g( i ) + abs( alpha ) * (
     $                     abs( a( ioffan ) ) * abs( b( ioffbk ) ) +
     $                     abs( b( ioffbn ) ) * abs( a( ioffak ) ) )
   20          CONTINUE
   30       CONTINUE
         ELSE IF( tran ) THEN
            DO 50 kk = 1, k
               ioffak = ia + kk - 1 + ( ja + j - 2 ) * lda
               ioffbk = ib + kk - 1 + ( jb + j - 2 ) * ldb
               DO 40 i = ibeg, iend
                  ioffan = ia + kk - 1 + ( ja + i - 2 ) * lda
                  ioffbn = ib + kk - 1 + ( jb + i - 2 ) * ldb
                  ct( i ) = ct( i ) + alpha * (
     $                      a( ioffan ) * b( ioffbk ) +
     $                      b( ioffbn ) * a( ioffak ) )
                  g( i ) = g( i ) + abs( alpha ) * (
     $                     abs( a( ioffan ) ) * abs( b( ioffbk ) ) +
     $                     abs( b( ioffbn ) ) * abs( a( ioffak ) ) )
   40          CONTINUE
   50       CONTINUE
         END IF
*
         ioffc = ic + ibeg - 1 + ( jc + j - 2 ) * ldc
*
         DO 100 i = ibeg, iend
            ct( i ) = ct( i ) + beta * c( ioffc )
            g( i ) = g( i ) + abs( beta )*abs( c( ioffc ) )
            c( ioffc ) = ct( i )
            ioffc = ioffc + 1
  100    CONTINUE
*
*        Compute the error ratio for this result.
*
         err  = zero
         info = 0
         ldpc = descc( lld_ )
         ioffc = ic + ( jc + j - 2 ) * ldc
         CALL pb_infog2l( ic, jc+j-1, descc, nprow, npcol, myrow, mycol,
     $                    iic, jjc, icrow, iccol )
         icurrow = icrow
         rowrep  = ( icrow.EQ.-1 )
         colrep  = ( iccol.EQ.-1 )
*
         IF( mycol.EQ.iccol .OR. colrep ) THEN
*
            ibb = descc( imb_ ) - ic + 1
            IF( ibb.LE.0 )
     $         ibb = ( ( -ibb ) / descc( mb_ ) + 1 )*descc( mb_ ) + ibb
            ibb = min( ibb, n )
            in = ic + ibb - 1
*
            DO 110 i = ic, in
*
               IF( myrow.EQ.icurrow .OR. rowrep ) THEN
                  erri = abs( pc( iic+(jjc-1)*ldpc ) -
     $                        c( ioffc ) ) / eps
                  IF( g( i-ic+1 ).NE.zero )
     $               erri = erri / g( i-ic+1 )
                  err = max( err, erri )
                  IF( err*sqrt( eps ).GE.one )
     $               info = 1
                  iic = iic + 1
               END IF
*
               ioffc = ioffc + 1
*
  110       CONTINUE
*
            icurrow = mod( icurrow+1, nprow )
*
            DO 130 i = in+1, ic+n-1, descc( mb_ )
               ibb = min( ic+n-i, descc( mb_ ) )
*
               DO 120 kk = 0, ibb-1
*
                  IF( myrow.EQ.icurrow .OR. rowrep ) THEN
                     erri = abs( pc( iic+(jjc-1)*ldpc ) -
     $                           c( ioffc ) )/eps
                     IF( g( i+kk-ic+1 ).NE.zero )
     $                  erri = erri / g( i+kk-ic+1 )
                     err = max( err, erri )
                     IF( err*sqrt( eps ).GE.one )
     $                  info = 1
                     iic = iic + 1
                  END IF
*
                  ioffc = ioffc + 1
*
  120          CONTINUE
*
               icurrow = mod( icurrow+1, nprow )
*
  130       CONTINUE
*
         END IF
*
*        If INFO = 0, all results are at least half accurate.
*
         CALL igsum2d( ictxt, 'All', ' ', 1, 1, info, 1, -1, mycol )
         CALL sgamx2d( ictxt, 'All', ' ', 1, 1, err, 1, i, j, -1, -1,
     $                 mycol )
         IF( info.NE.0 )
     $      GO TO 150
*
  140 CONTINUE
*
  150 CONTINUE
*
      RETURN
*
*     End of PSMMCH2
*
      END
      SUBROUTINE psmmch3( UPLO, TRANS, M, N, ALPHA, A, IA, JA, DESCA,
     $                    BETA, C, PC, IC, JC, DESCC, ERR, INFO )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1        TRANS, UPLO
      INTEGER            IA, IC, INFO, JA, JC, M, N
      REAL               ALPHA, BETA, ERR
*     ..
*     .. Array Arguments ..
      INTEGER            DESCA( * ), DESCC( * )
      REAL               A( * ), C( * ), PC( * )
*     ..
*
*  Purpose
*  =======
*
*  PSMMCH3 checks the results of the computational tests.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  UPLO    (global input) CHARACTER*1
*          On entry,  UPLO  specifies which part of C should contain the
*          result.
*
*  TRANS   (global input) CHARACTER*1
*          On entry,  TRANS  specifies  whether  the  matrix A has to be
*          transposed  or not  before computing the  matrix-matrix addi-
*          tion.
*
*  M       (global input) INTEGER
*          On entry, M specifies the number of rows of C.
*
*  N       (global input) INTEGER
*          On entry, N specifies the number of columns of C.
*
*  ALPHA   (global input) REAL
*          On entry, ALPHA specifies the scalar alpha.
*
*  A       (local input) REAL array
*          On entry, A is an array of  dimension  (DESCA( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PA.
*
*  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.
*
*  BETA    (global input) REAL
*          On entry, BETA specifies the scalar beta.
*
*  C       (local input/local output) REAL array
*          On entry, C is an array of  dimension  (DESCC( M_ ),*).  This
*          array contains a local copy of the initial entire matrix PC.
*
*  PC      (local input) REAL array
*          On entry, PC is an array of dimension (DESCC( LLD_ ),*). This
*          array contains the local pieces of the matrix PC.
*
*  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.
*
*  ERR     (global output) REAL
*          On exit, ERR specifies the largest error in absolute value.
*
*  INFO    (global output) INTEGER
*          On exit, if INFO <> 0, the result is less than half accurate.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
      REAL               ZERO
      PARAMETER          ( ZERO = 0.0e+0 )
*     ..
*     .. Local Scalars ..
      LOGICAL            COLREP, LOWER, NOTRAN, ROWREP, UPPER
      INTEGER            I, ICCOL, ICROW, ICTXT, IIC, IOFFA, IOFFC, J,
     $                   JJC, LDA, LDC, LDPC, MYCOL, MYROW, NPCOL,
     $                   NPROW
      REAL               ERR0, ERRI, PREC
*     ..
*     .. External Subroutines ..
      EXTERNAL           BLACS_GRIDINFO, IGSUM2D, PB_INFOG2L,
     $                   pserraxpby, sgamx2d
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      REAL               PSLAMCH
      EXTERNAL           LSAME, PSLAMCH
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          abs, max
*     ..
*     .. Executable Statements ..
*
      ictxt = descc( ctxt_ )
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      prec   = pslamch( ictxt, 'eps' )
*
      upper  = lsame( uplo,  'U' )
      lower  = lsame( uplo,  'L' )
      notran = lsame( trans, 'N' )
*
*     Compute expected result in C using data in A and C. This part of
*     the computation is performed by every process in the grid.
*
      info   = 0
      err    = zero
*
      lda    = max( 1, desca( m_   ) )
      ldc    = max( 1, descc( m_   ) )
      ldpc   = max( 1, descc( lld_ ) )
      rowrep = ( descc( rsrc_ ).EQ.-1 )
      colrep = ( descc( csrc_ ).EQ.-1 )
*
      IF( notran ) THEN
*
         DO 20 j = jc, jc + n - 1
*
            ioffc = ic + ( j  - 1          ) * ldc
            ioffa = ia + ( ja - 1 + j - jc ) * lda
*
            DO 10 i = ic, ic + m - 1
*
               IF( upper ) THEN
                  IF( ( j - jc ).GE.( i - ic ) ) THEN
                     CALL pserraxpby( erri, alpha, a( ioffa ), beta,
     $                                c( ioffc ), prec )
                  ELSE
                     erri = zero
                  END IF
               ELSE IF( lower ) THEN
                  IF( ( j - jc ).LE.( i - ic ) ) THEN
                     CALL pserraxpby( erri, alpha, a( ioffa ), beta,
     $                                c( ioffc ), prec )
                  ELSE
                     erri = zero
                  END IF
               ELSE
                  CALL pserraxpby( erri, alpha, a( ioffa ), beta,
     $                             c( ioffc ), prec )
               END IF
*
               CALL pb_infog2l( i, j, descc, nprow, npcol, myrow, mycol,
     $                          iic, jjc, icrow, iccol )
               IF( ( myrow.EQ.icrow .OR. rowrep ) .AND.
     $             ( mycol.EQ.iccol .OR. colrep ) ) THEN
                  err0 = abs( pc( iic+(jjc-1)*ldpc )-c( ioffc ) )
                  IF( err0.GT.erri )
     $               info = 1
                  err = max( err, err0 )
               END IF
*
               ioffa = ioffa + 1
               ioffc = ioffc + 1
*
   10       CONTINUE
*
   20    CONTINUE
*
      ELSE
*
         DO 40 j = jc, jc + n - 1
*
            ioffc = ic +              ( j  - 1 ) * ldc
            ioffa = ia + ( j - jc ) + ( ja - 1 ) * lda
*
            DO 30 i = ic, ic + m - 1
*
               IF( upper ) THEN
                  IF( ( j - jc ).GE.( i - ic ) ) THEN
                     CALL pserraxpby( erri, alpha, a( ioffa ), beta,
     $                               c( ioffc ), prec )
                  ELSE
                     erri = zero
                  END IF
               ELSE IF( lower ) THEN
                  IF( ( j - jc ).LE.( i - ic ) ) THEN
                     CALL pserraxpby( erri, alpha, a( ioffa ), beta,
     $                               c( ioffc ), prec )
                  ELSE
                     erri = zero
                  END IF
               ELSE
                  CALL pserraxpby( erri, alpha, a( ioffa ), beta,
     $                            c( ioffc ), prec )
               END IF
*
               CALL pb_infog2l( i, j, descc, nprow, npcol, myrow, mycol,
     $                          iic, jjc, icrow, iccol )
               IF( ( myrow.EQ.icrow .OR. rowrep ) .AND.
     $             ( mycol.EQ.iccol .OR. colrep ) ) THEN
                  err0 = abs( pc( iic+(jjc-1)*ldpc )-c( ioffc ) )
                  IF( err0.GT.erri )
     $               info = 1
                  err = max( err, err0 )
               END IF
*
               ioffc = ioffc + 1
               ioffa = ioffa + lda
*
   30       CONTINUE
*
   40    CONTINUE
*
      END IF
*
*     If INFO = 0, all results are at least half accurate.
*
      CALL igsum2d( ictxt, 'All', ' ', 1, 1, info, 1, -1, mycol )
      CALL sgamx2d( ictxt, 'All', ' ', 1, 1, err, 1, i, j, -1, -1,
     $              mycol )
*
      RETURN
*
*     End of PSMMCH3
*
      END
      SUBROUTINE pserraxpby( ERRBND, ALPHA, X, BETA, Y, PREC )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      REAL               ALPHA, BETA, ERRBND, PREC, X, Y
*     ..
*
*  Purpose
*  =======
*
*  PSERRAXPBY  serially  computes  y := beta*y + alpha * x and returns a
*  scaled relative acceptable error bound on the result.
*
*  Arguments
*  =========
*
*  ERRBND  (global output) REAL
*          On exit, ERRBND  specifies the scaled relative acceptable er-
*          ror bound.
*
*  ALPHA   (global input) REAL
*          On entry, ALPHA specifies the scalar alpha.
*
*  X       (global input) REAL
*          On entry, X  specifies the scalar x to be scaled.
*
*  BETA    (global input) REAL
*          On entry, BETA specifies the scalar beta.
*
*  Y       (global input/global output) REAL
*          On entry,  Y  specifies  the scalar y to be added. On exit, Y
*          contains the resulting scalar y.
*
*  PREC    (global input) REAL
*          On entry, PREC specifies the machine precision.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      REAL               ONE, TWO, ZERO
      PARAMETER          ( ONE = 1.0e+0, two = 2.0e+0,
     $                   zero = 0.0e+0 )
*     ..
*     .. Local Scalars ..
      REAL               ADDBND, FACT, SUMPOS, SUMNEG, TMP
*     ..
*     .. Intrinsic Functions ..
*     ..
*     .. Executable Statements ..
*
      SUMPOS = zero
      sumneg = zero
      fact = one + two * prec
      addbnd = two * two * two * prec
*
      tmp = alpha * x
      IF( tmp.GE.zero ) THEN
         sumpos = sumpos + tmp * fact
      ELSE
         sumneg = sumneg - tmp * fact
      END IF
*
      tmp = beta * y
      IF( tmp.GE.zero ) THEN
         sumpos = sumpos + tmp * fact
      ELSE
         sumneg = sumneg - tmp * fact
      END IF
*
      y = ( beta * y ) + ( alpha * x )
*
      errbnd = addbnd * max( sumpos, sumneg )
*
      RETURN
*
*     End of PSERRAXPBY
*
      END
      REAL               FUNCTION PSLAMCH( ICTXT, CMACH )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1        cmach
      INTEGER            ictxt
*     ..
*
*  Purpose
*  =======
*
*  PSLAMCH determines single precision machine parameters.
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  CMACH   (global input) CHARACTER*1
*          On entry, CMACH specifies the value to be returned by PSLAMCH
*          as follows:
*             = 'E' or 'e',   PSLAMCH := eps,
*             = 'S' or 's ,   PSLAMCH := sfmin,
*             = 'B' or 'b',   PSLAMCH := base,
*             = 'P' or 'p',   PSLAMCH := eps*base,
*             = 'N' or 'n',   PSLAMCH := t,
*             = 'R' or 'r',   PSLAMCH := rnd,
*             = 'M' or 'm',   PSLAMCH := emin,
*             = 'U' or 'u',   PSLAMCH := rmin,
*             = 'L' or 'l',   PSLAMCH := emax,
*             = 'O' or 'o',   PSLAMCH := rmax,
*
*          where
*
*          eps   = relative machine precision,
*          sfmin = safe minimum, such that 1/sfmin does not overflow,
*          base  = base of the machine,
*          prec  = eps*base,
*          t     = number of (base) digits in the mantissa,
*          rnd   = 1.0 when rounding occurs in addition, 0.0 otherwise,
*          emin  = minimum exponent before (gradual) underflow,
*          rmin  = underflow threshold - base**(emin-1),
*          emax  = largest exponent before overflow,
*          rmax  = overflow threshold  - (base**emax)*(1-eps).
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Local Scalars ..
      CHARACTER*1        top
      INTEGER            idumm
      REAL               temp
*     ..
*     .. External Subroutines ..
      EXTERNAL           pb_topget, sgamn2d, sgamx2d
*     ..
*     .. External Functions ..
      LOGICAL            lsame
      REAL               slamch
      EXTERNAL           lsame, slamch
*     ..
*     .. Executable Statements ..
*
      temp = slamch( cmach )
*
      IF( lsame( cmach, 'E' ).OR.lsame( cmach, 'S' ).OR.
     $    lsame( cmach, 'M' ).OR.lsame( cmach, 'U' ) ) THEN
         CALL pb_topget( ictxt, 'Combine', 'All', top )
         idumm = 0
         CALL sgamx2d( ictxt, 'All', top, 1, 1, temp, 1, idumm,
     $                 idumm, -1, -1, idumm )
      ELSE IF( lsame( cmach, 'L' ).OR.lsame( cmach, 'O' ) ) THEN
         CALL pb_topget( ictxt, 'Combine', 'All', top )
         idumm = 0
         CALL sgamn2d( ictxt, 'All', top, 1, 1, temp, 1, idumm,
     $                 idumm, -1, -1, idumm )
      END IF
*
      pslamch = temp
*
      RETURN
*
*     End of PSLAMCH
*
      END
      SUBROUTINE pslaset( UPLO, M, N, ALPHA, BETA, A, IA, JA, DESCA )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1        UPLO
      INTEGER            IA, JA, M, N
      REAL               ALPHA, BETA
*     ..
*     .. Array Arguments ..
      INTEGER            DESCA( * )
      REAL               A( * )
*     ..
*
*  Purpose
*  =======
*
*  PSLASET  initializes an m by n submatrix A(IA:IA+M-1,JA:JA+N-1) deno-
*  ted  by  sub( A )  to beta on the diagonal and alpha on the offdiago-
*  nals.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  UPLO    (global input) CHARACTER*1
*          On entry, UPLO specifies the part  of  the submatrix sub( A )
*          to be set:
*             = 'L' or 'l':   Lower triangular part is set; the strictly
*                      upper triangular part of sub( A ) is not changed;
*             = 'U' or 'u':   Upper triangular part is set; the strictly
*                      lower triangular part of sub( A ) is not changed;
*             Otherwise:  All of the matrix sub( A ) is set.
*
*  M       (global input) INTEGER
*          On entry,  M  specifies the number of rows of  the  submatrix
*          sub( A ). M  must be at least zero.
*
*  N       (global input) INTEGER
*          On entry, N  specifies the number of columns of the submatrix
*          sub( A ). N must be at least zero.
*
*  ALPHA   (global input) REAL
*          On entry,  ALPHA  specifies the scalar alpha, i.e., the cons-
*          tant to which the offdiagonal elements are to be set.
*
*  BETA    (global input) REAL
*          On entry, BETA  specifies the scalar beta, i.e., the constant
*          to which the diagonal elements are to be set.
*
*  A       (local input/local output) REAL array
*          On entry, A is an array of dimension (LLD_A, Ka), where Ka is
*          at least Lc( 1, JA+N-1 ).  Before  entry, this array contains
*          the local entries of the matrix  A  to be  set.  On exit, the
*          leading m by n submatrix sub( A ) is set as follows:
*
*          if UPLO = 'U',  A(IA+i-1,JA+j-1) = ALPHA, 1<=i<=j-1, 1<=j<=N,
*          if UPLO = 'L',  A(IA+i-1,JA+j-1) = ALPHA, j+1<=i<=M, 1<=j<=N,
*          otherwise,      A(IA+i-1,JA+j-1) = ALPHA, 1<=i<=M,   1<=j<=N,
*                                                      and IA+i.NE.JA+j,
*          and, for all UPLO,  A(IA+i-1,JA+i-1) = BETA,  1<=i<=min(M,N).
*
*  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.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
*     ..
*     .. Local Scalars ..
      LOGICAL            GODOWN, GOLEFT, ISCOLREP, ISROWREP, LOWER,
     $                   UPPER
      INTEGER            IACOL, IAROW, ICTXT, IIA, IIMAX, ILOW, IMB1,
     $                   IMBLOC, INB1, INBLOC, IOFFA, IOFFD, IUPP, JJA,
     $                   JJMAX, JOFFA, JOFFD, LCMT, LCMT00, LDA, LMBLOC,
     $                   LNBLOC, LOW, M1, MB, MBLKD, MBLKS, MBLOC, MP,
     $                   MRCOL, MRROW, MYCOL, MYROW, N1, NB, NBLKD,
     $                   NBLKS, NBLOC, NPCOL, NPROW, NQ, PMB, QNB, TMP1,
     $                   UPP
*     ..
*     .. Local Arrays ..
      INTEGER            DESCA2( DLEN_ )
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_gridinfo, pb_ainfog2l, pb_binfo,
     $                   pb_desctrans, pb_slaset
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      EXTERNAL           lsame
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          min
*     ..
*     .. Executable Statements ..
*
      IF( m.EQ.0 .OR. n.EQ.0 )
     $   RETURN
*
*     Convert descriptor
*
      CALL pb_desctrans( desca, desca2 )
*
*     Get grid parameters
*
      ictxt = desca2( ctxt_ )
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      CALL pb_ainfog2l( m, n, ia, ja, desca2, nprow, npcol, myrow,
     $                  mycol, imb1, inb1, mp, nq, iia, jja, iarow,
     $                  iacol, mrrow, mrcol )
*
      IF( mp.LE.0 .OR. nq.LE.0 )
     $   RETURN
*
      isrowrep = ( desca2( rsrc_ ).LT.0 )
      iscolrep = ( desca2( csrc_ ).LT.0 )
      lda      = desca2( lld_ )
*
      upper = .NOT.( lsame( uplo, 'L' ) )
      lower = .NOT.( lsame( uplo, 'U' ) )
*
      IF( ( ( lower.AND.upper ).AND.( alpha.EQ.beta ) ).OR.
     $    (   isrowrep        .AND.  iscolrep        ) ) THEN
         IF( ( mp.GT.0 ).AND.( nq.GT.0 ) )
     $      CALL pb_slaset( uplo, mp, nq, 0, alpha, beta,
     $                      a( iia + ( jja - 1 ) * lda ), lda )
         RETURN
      END IF
*
*     Initialize LCMT00, MBLKS, NBLKS, IMBLOC, INBLOC, LMBLOC, LNBLOC,
*     ILOW, LOW, IUPP, and UPP.
*
      mb = desca2( mb_ )
      nb = desca2( nb_ )
      CALL pb_binfo( 0, mp, nq, imb1, inb1, mb, nb, mrrow, mrcol,
     $               lcmt00, mblks, nblks, imbloc, inbloc, lmbloc,
     $               lnbloc, ilow, low, iupp, upp )
*
      ioffa = iia - 1
      joffa = jja - 1
      iimax = ioffa + mp
      jjmax = joffa + nq
*
      IF( isrowrep ) THEN
         pmb = mb
      ELSE
         pmb = nprow * mb
      END IF
      IF( iscolrep ) THEN
         qnb = nb
      ELSE
         qnb = npcol * nb
      END IF
*
      m1 = mp
      n1 = nq
*
*     Handle the first block of rows or columns separately, and update
*     LCMT00, MBLKS and NBLKS.
*
      godown = ( lcmt00.GT.iupp )
      goleft = ( lcmt00.LT.ilow )
*
      IF( .NOT.godown .AND. .NOT.goleft ) THEN
*
*        LCMT00 >= ILOW && LCMT00 <= IUPP
*
         goleft = ( ( lcmt00 - ( iupp - upp + pmb ) ).LT.ilow )
         godown = .NOT.goleft
*
         CALL pb_slaset( uplo, imbloc, inbloc, lcmt00, alpha, beta,
     $                   a( iia+joffa*lda ), lda )
         IF( godown ) THEN
            IF( upper .AND. nq.GT.inbloc )
     $         CALL pb_slaset( 'All', imbloc, nq-inbloc, 0, alpha,
     $                         alpha, a( iia+(joffa+inbloc)*lda ), lda )
            iia = iia + imbloc
            m1  = m1 - imbloc
         ELSE
            IF( lower .AND. mp.GT.imbloc )
     $         CALL pb_slaset( 'All', mp-imbloc, inbloc, 0, alpha,
     $                         alpha, a( iia+imbloc+joffa*lda ), lda )
            jja = jja + inbloc
            n1  = n1 - inbloc
         END IF
*
      END IF
*
      IF( godown ) THEN
*
         lcmt00 = lcmt00 - ( iupp - upp + pmb )
         mblks  = mblks - 1
         ioffa  = ioffa + imbloc
*
   10    CONTINUE
         IF( mblks.GT.0 .AND. lcmt00.GT.upp ) THEN
            lcmt00 = lcmt00 - pmb
            mblks  = mblks - 1
            ioffa  = ioffa + mb
            GO TO 10
         END IF
*
         tmp1 = min( ioffa, iimax ) - iia + 1
         IF( upper .AND. tmp1.GT.0 ) THEN
            CALL pb_slaset( 'All', tmp1, n1, 0, alpha, alpha,
     $                      a( iia+joffa*lda ), lda )
            iia = iia + tmp1
            m1  = m1 - tmp1
         END IF
*
         IF( mblks.LE.0 )
     $      RETURN
*
         lcmt  = lcmt00
         mblkd = mblks
         ioffd = ioffa
*
         mbloc = mb
   20    CONTINUE
         IF( mblkd.GT.0 .AND. lcmt.GE.ilow ) THEN
            IF( mblkd.EQ.1 )
     $         mbloc = lmbloc
            CALL pb_slaset( uplo, mbloc, inbloc, lcmt, alpha, beta,
     $                      a( ioffd+1+joffa*lda ), lda )
            lcmt00 = lcmt
            lcmt   = lcmt - pmb
            mblks  = mblkd
            mblkd  = mblkd - 1
            ioffa  = ioffd
            ioffd  = ioffd + mbloc
            GO TO 20
         END IF
*
         tmp1 = m1 - ioffd + iia - 1
         IF( lower .AND. tmp1.GT.0 )
     $      CALL pb_slaset( 'ALL', tmp1, inbloc, 0, alpha, alpha,
     $                      a( ioffd+1+joffa*lda ), lda )
*
         tmp1   = ioffa - iia + 1
         m1     = m1 - tmp1
         n1     = n1 - inbloc
         lcmt00 = lcmt00 + low - ilow + qnb
         nblks  = nblks - 1
         joffa  = joffa + inbloc
*
         IF( upper .AND. tmp1.GT.0 .AND. n1.GT.0 )
     $      CALL pb_slaset( 'ALL', tmp1, n1, 0, alpha, alpha,
     $                      a( iia+joffa*lda ), lda )
*
         iia = ioffa + 1
         jja = joffa + 1
*
      ELSE IF( goleft ) THEN
*
         lcmt00 = lcmt00 + low - ilow + qnb
         nblks  = nblks - 1
         joffa  = joffa + inbloc
*
   30    CONTINUE
         IF( nblks.GT.0 .AND. lcmt00.LT.low ) THEN
            lcmt00 = lcmt00 + qnb
            nblks  = nblks - 1
            joffa  = joffa + nb
            GO TO 30
         END IF
*
         tmp1 = min( joffa, jjmax ) - jja + 1
         IF( lower .AND. tmp1.GT.0 ) THEN
            CALL pb_slaset( 'All', m1, tmp1, 0, alpha, alpha,
     $                      a( iia+(jja-1)*lda ), lda )
            jja = jja + tmp1
            n1  = n1 - tmp1
         END IF
*
         IF( nblks.LE.0 )
     $      RETURN
*
         lcmt  = lcmt00
         nblkd = nblks
         joffd = joffa
*
         nbloc = nb
   40    CONTINUE
         IF( nblkd.GT.0 .AND. lcmt.LE.iupp ) THEN
            IF( nblkd.EQ.1 )
     $         nbloc = lnbloc
            CALL pb_slaset( uplo, imbloc, nbloc, lcmt, alpha, beta,
     $                      a( iia+joffd*lda ), lda )
            lcmt00 = lcmt
            lcmt   = lcmt + qnb
            nblks  = nblkd
            nblkd  = nblkd - 1
            joffa  = joffd
            joffd  = joffd + nbloc
            GO TO 40
         END IF
*
         tmp1 = n1 - joffd + jja - 1
         IF( upper .AND. tmp1.GT.0 )
     $      CALL pb_slaset( 'All', imbloc, tmp1, 0, alpha, alpha,
     $                      a( iia+joffd*lda ), lda )
*
         tmp1   = joffa - jja + 1
         m1     = m1 - imbloc
         n1     = n1 - tmp1
         lcmt00 = lcmt00 - ( iupp - upp + pmb )
         mblks  = mblks - 1
         ioffa  = ioffa + imbloc
*
         IF( lower .AND. m1.GT.0 .AND. tmp1.GT.0 )
     $      CALL pb_slaset( 'All', m1, tmp1, 0, alpha, alpha,
     $                      a( ioffa+1+(jja-1)*lda ), lda )
*
         iia = ioffa + 1
         jja = joffa + 1
*
      END IF
*
      nbloc = nb
   50 CONTINUE
      IF( nblks.GT.0 ) THEN
         IF( nblks.EQ.1 )
     $      nbloc = lnbloc
   60    CONTINUE
         IF( mblks.GT.0 .AND. lcmt00.GT.upp ) THEN
            lcmt00 = lcmt00 - pmb
            mblks  = mblks - 1
            ioffa  = ioffa + mb
            GO TO 60
         END IF
*
         tmp1 = min( ioffa, iimax ) - iia + 1
         IF( upper .AND. tmp1.GT.0 ) THEN
            CALL pb_slaset( 'All', tmp1, n1, 0, alpha, alpha,
     $                      a( iia+joffa*lda ), lda )
            iia = iia + tmp1
            m1  = m1 - tmp1
         END IF
*
         IF( mblks.LE.0 )
     $      RETURN
*
         lcmt  = lcmt00
         mblkd = mblks
         ioffd = ioffa
*
         mbloc = mb
   70    CONTINUE
         IF( mblkd.GT.0 .AND. lcmt.GE.low ) THEN
            IF( mblkd.EQ.1 )
     $         mbloc = lmbloc
            CALL pb_slaset( uplo, mbloc, nbloc, lcmt, alpha, beta,
     $                      a( ioffd+1+joffa*lda ), lda )
            lcmt00 = lcmt
            lcmt   = lcmt - pmb
            mblks  = mblkd
            mblkd  = mblkd - 1
            ioffa  = ioffd
            ioffd  = ioffd + mbloc
            GO TO 70
         END IF
*
         tmp1 = m1 - ioffd + iia - 1
         IF( lower .AND. tmp1.GT.0 )
     $      CALL pb_slaset( 'All', tmp1, nbloc, 0, alpha, alpha,
     $                      a( ioffd+1+joffa*lda ), lda )
*
         tmp1   = min( ioffa, iimax )  - iia + 1
         m1     = m1 - tmp1
         n1     = n1 - nbloc
         lcmt00 = lcmt00 + qnb
         nblks  = nblks - 1
         joffa  = joffa + nbloc
*
         IF( upper .AND. tmp1.GT.0 .AND. n1.GT.0 )
     $      CALL pb_slaset( 'All', tmp1, n1, 0, alpha, alpha,
     $                      a( iia+joffa*lda ), lda )
*
         iia = ioffa + 1
         jja = joffa + 1
*
         GO TO 50
*
      END IF
*
      RETURN
*
*     End of PSLASET
*
      END
      SUBROUTINE pslascal( TYPE, M, N, ALPHA, A, IA, JA, DESCA )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1        TYPE
      INTEGER            IA, JA, M, N
      REAL               ALPHA
*     ..
*     .. Array Arguments ..
      INTEGER            DESCA( * )
      REAL               A( * )
*     ..
*
*  Purpose
*  =======
*
*  PSLASCAL  scales the  m by n submatrix A(IA:IA+M-1,JA:JA+N-1) denoted
*  by sub( A ) by the scalar alpha. TYPE  specifies if sub( A ) is full,
*  upper triangular, lower triangular or upper Hessenberg.
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  TYPE    (global input) CHARACTER*1
*          On entry,  TYPE  specifies the type of the input submatrix as
*          follows:
*             = 'L' or 'l':  sub( A ) is a lower triangular matrix,
*             = 'U' or 'u':  sub( A ) is an upper triangular matrix,
*             = 'H' or 'h':  sub( A ) is an upper Hessenberg matrix,
*             otherwise sub( A ) is a  full matrix.
*
*  M       (global input) INTEGER
*          On entry,  M  specifies the number of rows of  the  submatrix
*          sub( A ). M  must be at least zero.
*
*  N       (global input) INTEGER
*          On entry, N  specifies the number of columns of the submatrix
*          sub( A ). N  must be at least zero.
*
*  ALPHA   (global input) REAL
*          On entry, ALPHA specifies the scalar alpha.
*
*  A       (local input/local output) REAL array
*          On entry, A is an array of dimension (LLD_A, Ka), where Ka is
*          at least Lc( 1, JA+N-1 ).  Before  entry, this array contains
*          the local entries of the matrix  A.
*          On exit, the local entries of this array corresponding to the
*          to  the entries of the submatrix sub( A ) are  overwritten by
*          the local entries of the m by n scaled submatrix.
*
*  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.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
*     ..
*     .. Local Scalars ..
      CHARACTER*1        UPLO
      LOGICAL            GODOWN, GOLEFT, LOWER, UPPER
      INTEGER            IACOL, IAROW, ICTXT, IIA, IIMAX, ILOW, IMB1,
     $                   IMBLOC, INB1, INBLOC, IOFFA, IOFFD, ITYPE,
     $                   IUPP, JJA, JJMAX, JOFFA, JOFFD, LCMT, LCMT00,
     $                   LDA, LMBLOC, LNBLOC, LOW, M1, MB, MBLKD, MBLKS,
     $                   MBLOC, MP, MRCOL, MRROW, MYCOL, MYROW, N1, NB,
     $                   NBLKD, NBLKS, NBLOC, NPCOL, NPROW, NQ, PMB,
     $                   QNB, TMP1, UPP
*     ..
*     .. Local Arrays ..
      INTEGER            DESCA2( DLEN_ )
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_gridinfo, pb_ainfog2l, pb_binfo,
     $                   pb_desctrans, pb_infog2l, pb_slascal
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      INTEGER            PB_NUMROC
      EXTERNAL           lsame, pb_numroc
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          min
*     ..
*     .. Executable Statements ..
*
*     Convert descriptor
*
      CALL pb_desctrans( desca, desca2 )
*
*     Get grid parameters
*
      ictxt = desca2( ctxt_ )
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
*     Quick return if possible
*
      IF( m.EQ.0 .OR. n.EQ.0 )
     $   RETURN
*
      IF( lsame( TYPE, 'L' ) ) then
         itype = 1
         uplo  = TYPE
         upper = .false.
         lower = .true.
         ioffd = 0
      ELSE IF( lsame( TYPE, 'U' ) ) then
         itype = 2
         uplo  = TYPE
         upper = .true.
         lower = .false.
         ioffd = 0
      ELSE IF( lsame( TYPE, 'H' ) ) then
         itype = 3
         uplo  = 'U'
         upper = .true.
         lower = .false.
         ioffd = 1
      ELSE
         itype = 0
         uplo  = 'A'
         upper = .true.
         lower = .true.
         ioffd = 0
      END IF
*
*     Compute local indexes
*
      IF( itype.EQ.0 ) THEN
*
*        Full matrix
*
         CALL pb_infog2l( ia, ja, desca2, nprow, npcol, myrow, mycol,
     $                    iia, jja, iarow, iacol )
         mp = pb_numroc( m, ia, desca2( imb_ ), desca2( mb_ ), myrow,
     $                   desca2( rsrc_ ), nprow )
         nq = pb_numroc( n, ja, desca2( inb_ ), desca2( nb_ ), mycol,
     $                   desca2( csrc_ ), npcol )
*
         IF( mp.LE.0 .OR. nq.LE.0 )
     $      RETURN
*
         lda   = desca2( lld_ )
         ioffa = iia + ( jja - 1 ) * lda
*
         CALL pb_slascal( 'All', mp, nq, 0, alpha, a( ioffa ), lda )
*
      ELSE
*
*        Trapezoidal matrix
*
         CALL pb_ainfog2l( m, n, ia, ja, desca2, nprow, npcol, myrow,
     $                     mycol, imb1, inb1, mp, nq, iia, jja, iarow,
     $                     iacol, mrrow, mrcol )
*
         IF( mp.LE.0 .OR. nq.LE.0 )
     $      RETURN
*
*        Initialize LCMT00, MBLKS, NBLKS, IMBLOC, INBLOC, LMBLOC,
*        LNBLOC, ILOW, LOW, IUPP, and UPP.
*
         mb  = desca2( mb_ )
         nb  = desca2( nb_ )
         lda = desca2( lld_ )
*
         CALL pb_binfo( ioffd, mp, nq, imb1, inb1, mb, nb, mrrow,
     $                  mrcol, lcmt00, mblks, nblks, imbloc, inbloc,
     $                  lmbloc, lnbloc, ilow, low, iupp, upp )
*
         m1    = mp
         n1    = nq
         ioffa = iia - 1
         joffa = jja - 1
         iimax = ioffa + mp
         jjmax = joffa + nq
*
         IF( desca2( rsrc_ ).LT.0 ) THEN
            pmb = mb
         ELSE
            pmb = nprow * mb
         END IF
         IF( desca2( csrc_ ).LT.0 ) THEN
            qnb = nb
         ELSE
            qnb = npcol * nb
         END IF
*
*        Handle the first block of rows or columns separately, and
*        update LCMT00, MBLKS and NBLKS.
*
         godown = ( lcmt00.GT.iupp )
         goleft = ( lcmt00.LT.ilow )
*
         IF( .NOT.godown .AND. .NOT.goleft ) THEN
*
*           LCMT00 >= ILOW && LCMT00 <= IUPP
*
            goleft = ( ( lcmt00 - ( iupp - upp + pmb ) ).LT.ilow )
            godown = .NOT.goleft
*
            CALL pb_slascal( uplo, imbloc, inbloc, lcmt00, alpha,
     $                       a( iia+joffa*lda ), lda )
            IF( godown ) THEN
               IF( upper .AND. nq.GT.inbloc )
     $            CALL pb_slascal( 'All', imbloc, nq-inbloc, 0, alpha,
     $                             a( iia+(joffa+inbloc)*lda ), lda )
               iia = iia + imbloc
               m1  = m1 - imbloc
            ELSE
               IF( lower .AND. mp.GT.imbloc )
     $            CALL pb_slascal( 'All', mp-imbloc, inbloc, 0, alpha,
     $                             a( iia+imbloc+joffa*lda ), lda )
               jja = jja + inbloc
               n1  = n1 - inbloc
            END IF
*
         END IF
*
         IF( godown ) THEN
*
            lcmt00 = lcmt00 - ( iupp - upp + pmb )
            mblks  = mblks - 1
            ioffa  = ioffa + imbloc
*
   10       CONTINUE
            IF( mblks.GT.0 .AND. lcmt00.GT.upp ) THEN
               lcmt00 = lcmt00 - pmb
               mblks  = mblks - 1
               ioffa  = ioffa + mb
               GO TO 10
            END IF
*
            tmp1 = min( ioffa, iimax ) - iia + 1
            IF( upper .AND. tmp1.GT.0 ) THEN
               CALL pb_slascal( 'All', tmp1, n1, 0, alpha,
     $                          a( iia+joffa*lda ), lda )
               iia = iia + tmp1
               m1  = m1 - tmp1
            END IF
*
            IF( mblks.LE.0 )
     $         RETURN
*
            lcmt  = lcmt00
            mblkd = mblks
            ioffd = ioffa
*
            mbloc = mb
   20       CONTINUE
            IF( mblkd.GT.0 .AND. lcmt.GE.ilow ) THEN
               IF( mblkd.EQ.1 )
     $            mbloc = lmbloc
               CALL pb_slascal( uplo, mbloc, inbloc, lcmt, alpha,
     $                          a( ioffd+1+joffa*lda ), lda )
               lcmt00 = lcmt
               lcmt   = lcmt - pmb
               mblks  = mblkd
               mblkd  = mblkd - 1
               ioffa  = ioffd
               ioffd  = ioffd + mbloc
               GO TO 20
            END IF
*
            tmp1 = m1 - ioffd + iia - 1
            IF( lower .AND. tmp1.GT.0 )
     $         CALL pb_slascal( 'All', tmp1, inbloc, 0, alpha,
     $                          a( ioffd+1+joffa*lda ), lda )
*
            tmp1   = ioffa - iia + 1
            m1     = m1 - tmp1
            n1     = n1 - inbloc
            lcmt00 = lcmt00 + low - ilow + qnb
            nblks  = nblks - 1
            joffa  = joffa + inbloc
*
            IF( upper .AND. tmp1.GT.0 .AND. n1.GT.0 )
     $         CALL pb_slascal( 'All', tmp1, n1, 0, alpha,
     $                          a( iia+joffa*lda ), lda )
*
            iia = ioffa + 1
            jja = joffa + 1
*
         ELSE IF( goleft ) THEN
*
            lcmt00 = lcmt00 + low - ilow + qnb
            nblks  = nblks - 1
            joffa  = joffa + inbloc
*
   30       CONTINUE
            IF( nblks.GT.0 .AND. lcmt00.LT.low ) THEN
               lcmt00 = lcmt00 + qnb
               nblks  = nblks - 1
               joffa  = joffa + nb
               GO TO 30
            END IF
*
            tmp1 = min( joffa, jjmax ) - jja + 1
            IF( lower .AND. tmp1.GT.0 ) THEN
               CALL pb_slascal( 'All', m1, tmp1, 0, alpha,
     $                          a( iia+(jja-1)*lda ), lda )
               jja = jja + tmp1
               n1  = n1 - tmp1
            END IF
*
            IF( nblks.LE.0 )
     $         RETURN
*
            lcmt  = lcmt00
            nblkd = nblks
            joffd = joffa
*
            nbloc = nb
   40       CONTINUE
            IF( nblkd.GT.0 .AND. lcmt.LE.iupp ) THEN
               IF( nblkd.EQ.1 )
     $            nbloc = lnbloc
               CALL pb_slascal( uplo, imbloc, nbloc, lcmt, alpha,
     $                          a( iia+joffd*lda ), lda )
               lcmt00 = lcmt
               lcmt   = lcmt + qnb
               nblks  = nblkd
               nblkd  = nblkd - 1
               joffa  = joffd
               joffd  = joffd + nbloc
               GO TO 40
            END IF
*
            tmp1 = n1 - joffd + jja - 1
            IF( upper .AND. tmp1.GT.0 )
     $         CALL pb_slascal( 'All', imbloc, tmp1, 0, alpha,
     $                          a( iia+joffd*lda ), lda )
*
            tmp1   = joffa - jja + 1
            m1     = m1 - imbloc
            n1     = n1 - tmp1
            lcmt00 = lcmt00 - ( iupp - upp + pmb )
            mblks  = mblks - 1
            ioffa  = ioffa + imbloc
*
            IF( lower .AND. m1.GT.0 .AND. tmp1.GT.0 )
     $         CALL pb_slascal( 'All', m1, tmp1, 0, alpha,
     $                          a( ioffa+1+(jja-1)*lda ), lda )
*
            iia = ioffa + 1
            jja = joffa + 1
*
         END IF
*
         nbloc = nb
   50    CONTINUE
         IF( nblks.GT.0 ) THEN
            IF( nblks.EQ.1 )
     $         nbloc = lnbloc
   60       CONTINUE
            IF( mblks.GT.0 .AND. lcmt00.GT.upp ) THEN
               lcmt00 = lcmt00 - pmb
               mblks  = mblks - 1
               ioffa  = ioffa + mb
               GO TO 60
            END IF
*
            tmp1 = min( ioffa, iimax ) - iia + 1
            IF( upper .AND. tmp1.GT.0 ) THEN
               CALL pb_slascal( 'All', tmp1, n1, 0, alpha,
     $                          a( iia+joffa*lda ), lda )
               iia = iia + tmp1
               m1  = m1 - tmp1
            END IF
*
            IF( mblks.LE.0 )
     $         RETURN
*
            lcmt  = lcmt00
            mblkd = mblks
            ioffd = ioffa
*
            mbloc = mb
   70       CONTINUE
            IF( mblkd.GT.0 .AND. lcmt.GE.low ) THEN
               IF( mblkd.EQ.1 )
     $            mbloc = lmbloc
               CALL pb_slascal( uplo, mbloc, nbloc, lcmt, alpha,
     $                          a( ioffd+1+joffa*lda ), lda )
               lcmt00 = lcmt
               lcmt   = lcmt - pmb
               mblks  = mblkd
               mblkd  = mblkd - 1
               ioffa  = ioffd
               ioffd  = ioffd + mbloc
               GO TO 70
            END IF
*
            tmp1 = m1 - ioffd + iia - 1
            IF( lower .AND. tmp1.GT.0 )
     $         CALL pb_slascal( 'All', tmp1, nbloc, 0, alpha,
     $                          a( ioffd+1+joffa*lda ), lda )
*
            tmp1   = min( ioffa, iimax )  - iia + 1
            m1     = m1 - tmp1
            n1     = n1 - nbloc
            lcmt00 = lcmt00 + qnb
            nblks  = nblks - 1
            joffa  = joffa + nbloc
*
            IF( upper .AND. tmp1.GT.0 .AND. n1.GT.0 )
     $         CALL pb_slascal( 'All', tmp1, n1, 0, alpha,
     $                          a( iia+joffa*lda ), lda )
*
            iia = ioffa + 1
            jja = joffa + 1
*
            GO TO 50
*
         END IF
*
      END IF
*
      RETURN
*
*     End of PSLASCAL
*
      END
      SUBROUTINE pslagen( INPLACE, AFORM, DIAG, OFFA, M, N, IA, JA,
     $                    DESCA, IASEED, A, LDA )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      LOGICAL            inplace
      CHARACTER*1        aform, diag
      INTEGER            ia, iaseed, ja, lda, m, n, offa
*     ..
*     .. Array Arguments ..
      INTEGER            desca( * )
      REAL               A( LDA, * )
*     ..
*
*  Purpose
*  =======
*
*  PSLAGEN  generates  (or regenerates)  a  submatrix  sub( A ) denoting
*  A(IA:IA+M-1,JA:JA+N-1).
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  INPLACE (global input) LOGICAL
*          On entry, INPLACE specifies if the matrix should be generated
*          in place or not. If INPLACE is .TRUE., the local random array
*          to be generated  will start in memory at the local memory lo-
*          cation A( 1, 1 ),  otherwise it will start at the local posi-
*          tion induced by IA and JA.
*
*  AFORM   (global input) CHARACTER*1
*          On entry, AFORM specifies the type of submatrix to be genera-
*          ted as follows:
*             AFORM = 'S', sub( A ) is a symmetric matrix,
*             AFORM = 'H', sub( A ) is a Hermitian matrix,
*             AFORM = 'T', sub( A ) is overrwritten  with  the transpose
*                          of what would normally be generated,
*             AFORM = 'C', sub( A ) is overwritten  with  the  conjugate
*                          transpose  of  what would normally be genera-
*                          ted.
*             AFORM = 'N', a random submatrix is generated.
*
*  DIAG    (global input) CHARACTER*1
*          On entry, DIAG specifies if the generated submatrix is diago-
*          nally dominant or not as follows:
*             DIAG = 'D' : sub( A ) is diagonally dominant,
*             DIAG = 'N' : sub( A ) is not diagonally dominant.
*
*  OFFA    (global input) INTEGER
*          On entry, OFFA  specifies  the  offdiagonal of the underlying
*          matrix A(1:DESCA(M_),1:DESCA(N_)) of interest when the subma-
*          trix is symmetric, Hermitian or diagonally dominant. OFFA = 0
*          specifies the main diagonal,  OFFA > 0  specifies a subdiago-
*          nal,  and OFFA < 0 specifies a superdiagonal (see further de-
*          tails).
*
*  M       (global input) INTEGER
*          On entry, M specifies the global number of matrix rows of the
*          submatrix sub( A ) to be generated. M must be at least zero.
*
*  N       (global input) INTEGER
*          On entry,  N specifies the global number of matrix columns of
*          the  submatrix  sub( A )  to be generated. N must be at least
*          zero.
*
*  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.
*
*  IASEED  (global input) INTEGER
*          On entry, IASEED  specifies  the  seed number to generate the
*          matrix A. IASEED must be at least zero.
*
*  A       (local output) REAL array
*          On entry, A is an array of dimension (LLD_A, Ka), where Ka is
*          at least Lc( 1, JA+N-1 ).  On  exit, this array  contains the
*          local entries of the randomly generated submatrix sub( A ).
*
*  LDA     (local input) INTEGER
*          On entry,  LDA  specifies  the local leading dimension of the
*          array A. When INPLACE is .FALSE., LDA is usually DESCA(LLD_).
*          This restriction is however not enforced, and this subroutine
*          requires only that LDA >= MAX( 1, Mp ) where
*
*          Mp = PB_NUMROC( M, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW ).
*
*          PB_NUMROC  is  a ScaLAPACK tool function; MYROW, MYCOL, NPROW
*          and NPCOL  can  be determined by calling the BLACS subroutine
*          BLACS_GRIDINFO.
*
*  Further Details
*  ===============
*
*  OFFD  is  tied  to  the matrix described by  DESCA, as opposed to the
*  piece that is currently  (re)generated.  This is a global information
*  independent from the distribution  parameters.  Below are examples of
*  the meaning of OFFD for a global 7 by 5 matrix:
*
*  ---------------------------------------------------------------------
*  OFFD   |  0 -1 -2 -3 -4         0 -1 -2 -3 -4          0 -1 -2 -3 -4
*  -------|-------------------------------------------------------------
*         |     | OFFD=-1          |   OFFD=0                 OFFD=2
*         |     V                  V
*  0      |  .  d  .  .  .      -> d  .  .  .  .          .  .  .  .  .
*  1      |  .  .  d  .  .         .  d  .  .  .          .  .  .  .  .
*  2      |  .  .  .  d  .         .  .  d  .  .       -> d  .  .  .  .
*  3      |  .  .  .  .  d         .  .  .  d  .          .  d  .  .  .
*  4      |  .  .  .  .  .         .  .  .  .  d          .  .  d  .  .
*  5      |  .  .  .  .  .         .  .  .  .  .          .  .  .  d  .
*  6      |  .  .  .  .  .         .  .  .  .  .          .  .  .  .  d
*  ---------------------------------------------------------------------
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
      INTEGER            JMP_1, JMP_COL, JMP_IMBV, JMP_INBV, JMP_LEN,
     $                   JMP_MB, JMP_NB, JMP_NPIMBLOC, JMP_NPMB,
     $                   JMP_NQINBLOC, JMP_NQNB, JMP_ROW
      PARAMETER          ( JMP_1 = 1, jmp_row = 2, jmp_col = 3,
     $                   jmp_mb = 4, jmp_imbv = 5, jmp_npmb = 6,
     $                   jmp_npimbloc = 7, jmp_nb = 8, jmp_inbv = 9,
     $                   jmp_nqnb = 10, jmp_nqinbloc = 11,
     $                   jmp_len = 11 )
*     ..
*     .. Local Scalars ..
      LOGICAL            DIAGDO, SYMM, HERM, NOTRAN
      INTEGER            CSRC, I, IACOL, IAROW, ICTXT, IIA, ILOCBLK,
     $                   ILOCOFF, ILOW, IMB, IMB1, IMBLOC, IMBVIR, INB,
     $                   inb1, inbloc, inbvir, info, ioffda, itmp, iupp,
     $                   ivir, jja, jlocblk, jlocoff, jvir, lcmt00,
     $                   lmbloc, lnbloc, low, maxmn, mb, mblks, mp,
     $                   mrcol, mrrow, mycdist, mycol, myrdist, myrow,
     $                   nb, nblks, npcol, nprow, nq, nvir, rsrc, upp
      REAL               ALPHA
*     ..
*     .. Local Arrays ..
      INTEGER            DESCA2( DLEN_ ), IMULADD( 4, JMP_LEN ),
     $                   IRAN( 2 ), JMP( JMP_LEN ), MULADD0( 4 )
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_gridinfo, pb_ainfog2l, pb_binfo,
     $                   pb_chkmat, pb_desctrans, pb_initjmp,
     $                   pb_initmuladd, pb_jump, pb_jumpit, pb_locinfo,
     $                   pb_setlocran, pb_setran, pb_slagen, psladom,
     $                   pxerbla
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      EXTERNAL           LSAME
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          MAX, MIN, REAL
*     ..
*     .. Data Statements ..
      DATA               ( muladd0( i ), i = 1, 4 ) / 20077, 16838,
     $                   12345, 0 /
*     ..
*     .. Executable Statements ..
*
*     Convert descriptor
*
      CALL pb_desctrans( desca, desca2 )
*
*     Test the input arguments
*
      ictxt = desca2( ctxt_ )
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
*     Test the input parameters
*
      info = 0
      IF( nprow.EQ.-1 ) THEN
         info = -( 1000 + ctxt_ )
      ELSE
         symm   = lsame( aform, 'S' )
         herm   = lsame( aform, 'H' )
         notran = lsame( aform, 'N' )
         diagdo = lsame( diag, 'D' )
         IF( .NOT.( symm.OR.herm.OR.notran ) .AND.
     $       .NOT.( lsame( aform, 'T' )    ) .AND.
     $       .NOT.( lsame( aform, 'C' )    ) ) THEN
            info = -2
         ELSE IF( ( .NOT.diagdo ) .AND.
     $            ( .NOT.lsame( diag, 'N' ) ) ) THEN
            info = -3
         END IF
         CALL pb_chkmat( ictxt, m, 5, n, 6, ia, ja, desca2, 10, info )
      END IF
*
      IF( info.NE.0 ) THEN
         CALL pxerbla( ictxt, 'PSLAGEN', -info )
         RETURN
      END IF
*
*     Quick return if possible
*
      IF( ( m.LE.0 ).OR.( n.LE.0 ) )
     $   RETURN
*
*     Start the operations
*
      mb   = desca2( mb_   )
      nb   = desca2( nb_   )
      imb  = desca2( imb_  )
      inb  = desca2( inb_  )
      rsrc = desca2( rsrc_ )
      csrc = desca2( csrc_ )
*
*     Figure out local information about the distributed matrix operand
*
      CALL pb_ainfog2l( m, n, ia, ja, desca2, nprow, npcol, myrow,
     $                  mycol, imb1, inb1, mp, nq, iia, jja, iarow,
     $                  iacol, mrrow, mrcol )
*
*     Decide where the entries shall be stored in memory
*
      IF( inplace ) THEN
         iia = 1
         jja = 1
      END IF
*
*     Initialize LCMT00, MBLKS, NBLKS, IMBLOC, INBLOC, LMBLOC, LNBLOC,
*     ILOW, LOW, IUPP, and UPP.
*
      ioffda = ja + offa - ia
      CALL pb_binfo( ioffda, mp, nq, imb1, inb1, mb, nb, mrrow,
     $               mrcol, lcmt00, mblks, nblks, imbloc, inbloc,
     $               lmbloc, lnbloc, ilow, low, iupp, upp )
*
*     Initialize ILOCBLK, ILOCOFF, MYRDIST, JLOCBLK, JLOCOFF, MYCDIST
*     This values correspond to the square virtual underlying matrix
*     of size MAX( M_ + MAX( 0, -OFFA ), N_ + MAX( 0, OFFA ) ) used
*     to set up the random sequence. For practical purposes, the size
*     of this virtual matrix is upper bounded by M_ + N_ - 1.
*
      itmp   = max( 0, -offa )
      ivir   = ia  + itmp
      imbvir = imb + itmp
      nvir   = desca2( m_ ) + itmp
*
      CALL pb_locinfo( ivir, imbvir, mb, myrow, rsrc, nprow, ilocblk,
     $                 ilocoff, myrdist )
*
      itmp   = max( 0, offa )
      jvir   = ja  + itmp
      inbvir = inb + itmp
      nvir   = max( max( nvir, desca2( n_ ) + itmp ),
     $              desca2( m_ ) + desca2( n_ ) - 1 )
*
      CALL pb_locinfo( jvir, inbvir, nb, mycol, csrc, npcol, jlocblk,
     $                 jlocoff, mycdist )
*
      IF( symm .OR. herm .OR. notran ) THEN
*
         CALL pb_initjmp( .true., nvir, imbvir, inbvir, imbloc, inbloc,
     $                    mb, nb, rsrc, csrc, nprow, npcol, 1, jmp )
*
*        Compute constants to jump JMP( * ) numbers in the sequence
*
         CALL pb_initmuladd( muladd0, jmp, imuladd )
*
*        Compute and set the random value corresponding to A( IA, JA )
*
         CALL pb_setlocran( iaseed, ilocblk, jlocblk, ilocoff, jlocoff,
     $                      myrdist, mycdist, nprow, npcol, jmp,
     $                      imuladd, iran )
*
         CALL pb_slagen( 'Lower', aform, a( iia, jja ), lda, lcmt00,
     $                   iran, mblks, imbloc, mb, lmbloc, nblks, inbloc,
     $                   nb, lnbloc, jmp, imuladd )
*
      END IF
*
      IF( symm .OR. herm .OR. ( .NOT. notran ) ) THEN
*
         CALL pb_initjmp( .false., nvir, imbvir, inbvir, imbloc, inbloc,
     $                    mb, nb, rsrc, csrc, nprow, npcol, 1, jmp )
*
*        Compute constants to jump JMP( * ) numbers in the sequence
*
         CALL pb_initmuladd( muladd0, jmp, imuladd )
*
*        Compute and set the random value corresponding to A( IA, JA )
*
         CALL pb_setlocran( iaseed, ilocblk, jlocblk, ilocoff, jlocoff,
     $                      myrdist, mycdist, nprow, npcol, jmp,
     $                      imuladd, iran )
*
         CALL pb_slagen( 'Upper', aform, a( iia, jja ), lda, lcmt00,
     $                   iran, mblks, imbloc, mb, lmbloc, nblks, inbloc,
     $                   nb, lnbloc, jmp, imuladd )
*
      END IF
*
      IF( diagdo ) THEN
*
         maxmn = max( desca2( m_ ), desca2( n_ ) )
         alpha = real( maxmn )
*
         IF( ioffda.GE.0 ) THEN
            CALL psladom( inplace, min( max( 0, m-ioffda ), n ), alpha,
     $                    a, min( ia+ioffda, ia+m-1 ), ja, desca )
         ELSE
            CALL psladom( inplace, min( m, max( 0, n+ioffda ) ), alpha,
     $                    a, ia, min( ja-ioffda, ja+n-1 ), desca )
         END IF
*
      END IF
*
      RETURN
*
*     End of PSLAGEN
*
      END
      SUBROUTINE psladom( INPLACE, N, ALPHA, A, IA, JA, DESCA )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      LOGICAL            INPLACE
      INTEGER            IA, JA, N
      REAL               ALPHA
*     ..
*     .. Array Arguments ..
      INTEGER            DESCA( * )
      REAL               A( * )
*     ..
*
*  Purpose
*  =======
*
*  PSLADOM  adds alpha to the diagonal entries  of  an  n by n submatrix
*  sub( A ) denoting A( IA:IA+N-1, JA:JA+N-1 ).
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  INPLACE (global input) LOGICAL
*          On entry, INPLACE specifies if the matrix should be generated
*          in place or not. If INPLACE is .TRUE., the local random array
*          to be generated  will start in memory at the local memory lo-
*          cation A( 1, 1 ),  otherwise it will start at the local posi-
*          tion induced by IA and JA.
*
*  N       (global input) INTEGER
*          On entry,  N  specifies  the  global  order  of the submatrix
*          sub( A ) to be modified. N must be at least zero.
*
*  ALPHA   (global input) REAL
*          On entry, ALPHA specifies the scalar alpha.
*
*  A       (local input/local output) REAL array
*          On entry, A is an array of dimension (LLD_A, Ka), where Ka is
*          at least Lc( 1, JA+N-1 ).  Before  entry, this array contains
*          the local entries of the matrix A. On exit, the local entries
*          of this array corresponding to the main diagonal of  sub( A )
*          have been updated.
*
*  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.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
*     ..
*     .. Local Scalars ..
      LOGICAL            GODOWN, GOLEFT
      INTEGER            I, IACOL, IAROW, ICTXT, IIA, IJOFFA, ILOW,
     $                   IMB1, IMBLOC, INB1, INBLOC, IOFFA, IOFFD, IUPP,
     $                   JJA, JOFFA, JOFFD, LCMT, LCMT00, LDA, LDAP1,
     $                   LMBLOC, LNBLOC, LOW, MB, MBLKD, MBLKS, MBLOC,
     $                   MRCOL, MRROW, MYCOL, MYROW, NB, NBLKD, NBLKS,
     $                   NBLOC, NP, NPCOL, NPROW, NQ, PMB, QNB, UPP
      REAL               ATMP
*     ..
*     .. Local Scalars ..
      INTEGER            DESCA2( DLEN_ )
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_gridinfo, pb_ainfog2l, pb_binfo,
     $                   pb_desctrans
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          abs, max, min
*     ..
*     .. Executable Statements ..
*
*     Convert descriptor
*
      CALL pb_desctrans( desca, desca2 )
*
*     Get grid parameters
*
      ictxt = desca2( ctxt_ )
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
      IF( n.EQ.0 )
     $   RETURN
*
      CALL pb_ainfog2l( n, n, ia, ja, desca2, nprow, npcol, myrow,
     $                  mycol, imb1, inb1, np, nq, iia, jja, iarow,
     $                  iacol, mrrow, mrcol )
*
*     Decide where the entries shall be stored in memory
*
      IF( inplace ) THEN
         iia = 1
         jja = 1
      END IF
*
*     Initialize LCMT00, MBLKS, NBLKS, IMBLOC, INBLOC, LMBLOC, LNBLOC,
*     ILOW, LOW, IUPP, and UPP.
*
      mb = desca2( mb_ )
      nb = desca2( nb_ )
*
      CALL pb_binfo( 0, np, nq, imb1, inb1, mb, nb, mrrow, mrcol,
     $               lcmt00, mblks, nblks, imbloc, inbloc, lmbloc,
     $               lnbloc, ilow, low, iupp, upp )
*
      ioffa  = iia - 1
      joffa  = jja - 1
      lda    = desca2( lld_ )
      ldap1  = lda + 1
*
      IF( desca2( rsrc_ ).LT.0 ) THEN
         pmb = mb
      ELSE
         pmb = nprow * mb
      END IF
      IF( desca2( csrc_ ).LT.0 ) THEN
         qnb = nb
      ELSE
         qnb = npcol * nb
      END IF
*
*     Handle the first block of rows or columns separately, and update
*     LCMT00, MBLKS and NBLKS.
*
      godown = ( lcmt00.GT.iupp )
      goleft = ( lcmt00.LT.ilow )
*
      IF( .NOT.godown .AND. .NOT.goleft ) THEN
*
*        LCMT00 >= ILOW && LCMT00 <= IUPP
*
         IF( lcmt00.GE.0 ) THEN
            ijoffa = ioffa+lcmt00 + ( joffa - 1 ) * lda
            DO 10 i = 1, min( inbloc, max( 0, imbloc - lcmt00 ) )
               atmp = a( ijoffa + i*ldap1 )
               a( ijoffa + i*ldap1 ) = abs( atmp ) + alpha
   10       CONTINUE
         ELSE
            ijoffa = ioffa + ( joffa - lcmt00 - 1 ) * lda
            DO 20 i = 1, min( imbloc, max( 0, inbloc + lcmt00 ) )
               atmp = a( ijoffa + i*ldap1 )
               a( ijoffa + i*ldap1 ) = abs( atmp ) + alpha
   20       CONTINUE
         END IF
         goleft = ( ( lcmt00 - ( iupp - upp + pmb ) ).LT.ilow )
         godown = .NOT.goleft
*
      END IF
*
      IF( godown ) THEN
*
         lcmt00 = lcmt00 - ( iupp - upp + pmb )
         mblks  = mblks - 1
         ioffa  = ioffa + imbloc
*
   30    CONTINUE
         IF( mblks.GT.0 .AND. lcmt00.GT.upp ) THEN
            lcmt00 = lcmt00 - pmb
            mblks  = mblks - 1
            ioffa  = ioffa + mb
            GO TO 30
         END IF
*
         lcmt  = lcmt00
         mblkd = mblks
         ioffd = ioffa
*
         mbloc = mb
   40    CONTINUE
         IF( mblkd.GT.0 .AND. lcmt.GE.ilow ) THEN
            IF( mblkd.EQ.1 )
     $         mbloc = lmbloc
            IF( lcmt.GE.0 ) THEN
               ijoffa = ioffd + lcmt + ( joffa - 1 ) * lda
               DO 50 i = 1, min( inbloc, max( 0, mbloc - lcmt ) )
                  atmp = a( ijoffa + i*ldap1 )
                  a( ijoffa + i*ldap1 ) = abs( atmp ) + alpha
   50          CONTINUE
            ELSE
               ijoffa = ioffd + ( joffa - lcmt - 1 ) * lda
               DO 60 i = 1, min( mbloc, max( 0, inbloc + lcmt ) )
                  atmp = a( ijoffa + i*ldap1 )
                  a( ijoffa + i*ldap1 ) = abs( atmp ) + alpha
   60          CONTINUE
            END IF
            lcmt00 = lcmt
            lcmt   = lcmt - pmb
            mblks  = mblkd
            mblkd  = mblkd - 1
            ioffa  = ioffd
            ioffd  = ioffd + mbloc
            GO TO 40
         END IF
*
         lcmt00 = lcmt00 + low - ilow + qnb
         nblks  = nblks - 1
         joffa  = joffa + inbloc
*
      ELSE IF( goleft ) THEN
*
         lcmt00 = lcmt00 + low - ilow + qnb
         nblks  = nblks - 1
         joffa  = joffa + inbloc
*
   70    CONTINUE
         IF( nblks.GT.0 .AND. lcmt00.LT.low ) THEN
            lcmt00 = lcmt00 + qnb
            nblks  = nblks - 1
            joffa  = joffa + nb
            GO TO 70
         END IF
*
         lcmt  = lcmt00
         nblkd = nblks
         joffd = joffa
*
         nbloc = nb
   80    CONTINUE
         IF( nblkd.GT.0 .AND. lcmt.LE.iupp ) THEN
            IF( nblkd.EQ.1 )
     $         nbloc = lnbloc
            IF( lcmt.GE.0 ) THEN
               ijoffa = ioffa + lcmt + ( joffd - 1 ) * lda
               DO 90 i = 1, min( nbloc, max( 0, imbloc - lcmt ) )
                  atmp = a( ijoffa + i*ldap1 )
                  a( ijoffa + i*ldap1 ) = abs( atmp ) + alpha
   90          CONTINUE
            ELSE
               ijoffa = ioffa + ( joffd - lcmt - 1 ) * lda
               DO 100 i = 1, min( imbloc, max( 0, nbloc + lcmt ) )
                  atmp = a( ijoffa + i*ldap1 )
                  a( ijoffa + i*ldap1 ) = abs( atmp ) + alpha
  100          CONTINUE
            END IF
            lcmt00 = lcmt
            lcmt   = lcmt + qnb
            nblks  = nblkd
            nblkd  = nblkd - 1
            joffa  = joffd
            joffd  = joffd + nbloc
            GO TO 80
         END IF
*
         lcmt00 = lcmt00 - ( iupp - upp + pmb )
         mblks  = mblks - 1
         ioffa  = ioffa + imbloc
*
      END IF
*
      nbloc = nb
  110 CONTINUE
      IF( nblks.GT.0 ) THEN
         IF( nblks.EQ.1 )
     $      nbloc = lnbloc
  120    CONTINUE
         IF( mblks.GT.0 .AND. lcmt00.GT.upp ) THEN
            lcmt00 = lcmt00 - pmb
            mblks  = mblks - 1
            ioffa  = ioffa + mb
            GO TO 120
         END IF
*
         lcmt  = lcmt00
         mblkd = mblks
         ioffd = ioffa
*
         mbloc = mb
  130    CONTINUE
         IF( mblkd.GT.0 .AND. lcmt.GE.low ) THEN
            IF( mblkd.EQ.1 )
     $         mbloc = lmbloc
            IF( lcmt.GE.0 ) THEN
               ijoffa = ioffd + lcmt + ( joffa - 1 ) * lda
               DO 140 i = 1, min( nbloc, max( 0, mbloc - lcmt ) )
                  atmp = a( ijoffa + i*ldap1 )
                  a( ijoffa + i*ldap1 ) = abs( atmp ) + alpha
  140          CONTINUE
            ELSE
               ijoffa = ioffd + ( joffa - lcmt - 1 ) * lda
               DO 150 i = 1, min( mbloc, max( 0, nbloc + lcmt ) )
                  atmp = a( ijoffa + i*ldap1 )
                  a( ijoffa + i*ldap1 ) = abs( atmp ) + alpha
  150          CONTINUE
            END IF
            lcmt00 = lcmt
            lcmt   = lcmt - pmb
            mblks  = mblkd
            mblkd  = mblkd - 1
            ioffa  = ioffd
            ioffd  = ioffd + mbloc
            GO TO 130
         END IF
*
         lcmt00 = lcmt00 + qnb
         nblks  = nblks - 1
         joffa  = joffa + nbloc
         GO TO 110
*
      END IF
*
      RETURN
*
*     End of PSLADOM
*
      END
      SUBROUTINE pb_pslaprnt( M, N, A, IA, JA, DESCA, IRPRNT, ICPRNT,
     $                        CMATNM, NOUT, WORK )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER            IA, ICPRNT, IRPRNT, JA, M, N, NOUT
*     ..
*     .. Array Arguments ..
      CHARACTER*(*)      CMATNM
      INTEGER            DESCA( * )
      REAL               A( * ), WORK( * )
*     ..
*
*  Purpose
*  =======
*
*  PB_PSLAPRNT  prints to the standard output a submatrix sub( A ) deno-
*  ting A(IA:IA+M-1,JA:JA+N-1). The local pieces are sent and printed by
*  the process of coordinates (IRPRNT, ICPRNT).
*
*  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 DESCA:
*
*  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_NUMROC:
*  Lr( IA, K ) = PB_NUMROC( K, IA, IMB_A, MB_A, MYROW, RSRC_A, NPROW )
*  Lc( JA, K ) = PB_NUMROC( K, JA, INB_A, NB_A, MYCOL, CSRC_A, NPCOL )
*
*  Arguments
*  =========
*
*  M       (global input) INTEGER
*          On entry,  M  specifies the number of rows of  the  submatrix
*          sub( A ). M  must be at least zero.
*
*  N       (global input) INTEGER
*          On entry, N  specifies the number of columns of the submatrix
*          sub( A ). N must be at least zero.
*
*  A       (local input) REAL array
*          On entry, A is an array of dimension (LLD_A, Ka), where Ka is
*          at least Lc( 1, JA+N-1 ).  Before  entry, this array contains
*          the local entries of the matrix 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.
*
*  IRPRNT  (global input) INTEGER
*          On entry, IRPRNT specifies the row index of the printing pro-
*          cess.
*
*  ICPRNT  (global input) INTEGER
*          On entry, ICPRNT specifies the  column  index of the printing
*          process.
*
*  CMATNM  (global input) CHARACTER*(*)
*          On entry, CMATNM is the name of the matrix to be printed.
*
*  NOUT    (global input) INTEGER
*          On entry, NOUT specifies the output unit number. When NOUT is
*          equal to 6, the submatrix is printed on the screen.
*
*  WORK    (local workspace) REAL array
*          On entry, WORK is a work array of dimension at least equal to
*          MAX( IMB_A, MB_A ).
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      PARAMETER          ( BLOCK_CYCLIC_2D_INB = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
*     ..
*     .. Local Scalars ..
      INTEGER            MYCOL, MYROW, NPCOL, NPROW, PCOL, PROW
*     ..
*     .. Local Arrays ..
      INTEGER            DESCA2( DLEN_ )
*     ..
*     .. External Subroutines ..
      EXTERNAL           BLACS_GRIDINFO, PB_DESCTRANS, PB_PSLAPRN2
*     ..
*     .. Executable Statements ..
*
*     Quick return if possible
*
      IF( ( m.LE.0 ).OR.( n.LE.0 ) )
     $   RETURN
*
*     Convert descriptor
*
      CALL pb_desctrans( desca, desca2 )
*
      CALL blacs_gridinfo( desca2( ctxt_ ), nprow, npcol, myrow, mycol )
*
      IF( desca2( rsrc_ ).GE.0 ) THEN
         IF( desca2( csrc_ ).GE.0 ) THEN
            CALL pb_pslaprn2( m, n, a, ia, ja, desca2, irprnt, icprnt,
     $                        cmatnm, nout, desca2( rsrc_ ),
     $                        desca2( csrc_ ), work )
         ELSE
            DO 10 pcol = 0, npcol - 1
               IF( ( myrow.EQ.irprnt ).AND.( mycol.EQ.icprnt ) )
     $            WRITE( nout, * ) 'Colum-replicated array -- ' ,
     $                             'copy in process column: ', pcol
               CALL pb_pslaprn2( m, n, a, ia, ja, desca2, irprnt,
     $                           icprnt, cmatnm, nout, desca2( rsrc_ ),
     $                           pcol, work )
   10       CONTINUE
         END IF
      ELSE
         IF( desca2( csrc_ ).GE.0 ) THEN
            DO 20 prow = 0, nprow - 1
               IF( ( myrow.EQ.irprnt ).AND.( mycol.EQ.icprnt ) )
     $            WRITE( nout, * ) 'Row-replicated array -- ' ,
     $                             'copy in process row: ', prow
               CALL pb_pslaprn2( m, n, a, ia, ja, desca2, irprnt,
     $                           icprnt, cmatnm, nout, prow,
     $                           desca2( csrc_ ), work )
   20       CONTINUE
         ELSE
            DO 40 prow = 0, nprow - 1
               DO 30 pcol = 0, npcol - 1
                  IF( ( myrow.EQ.irprnt ).AND.( mycol.EQ.icprnt ) )
     $               WRITE( nout, * ) 'Replicated array -- ' ,
     $                      'copy in process (', prow, ',', pcol, ')'
                  CALL pb_pslaprn2( m, n, a, ia, ja, desca2, irprnt,
     $                              icprnt, cmatnm, nout, prow, pcol,
     $                              work )
   30          CONTINUE
   40       CONTINUE
         END IF
      END IF
*
      RETURN
*
*     End of PB_PSLAPRNT
*
      END
      SUBROUTINE pb_pslaprn2( M, N, A, IA, JA, DESCA, IRPRNT, ICPRNT,
     $                        CMATNM, NOUT, PROW, PCOL, WORK )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER            IA, ICPRNT, IRPRNT, JA, M, N, NOUT, PCOL, PROW
*     ..
*     .. Array Arguments ..
      CHARACTER*(*)      CMATNM
      INTEGER            DESCA( * )
      REAL               A( * ), WORK( * )
*     ..
*
*     .. Parameters ..
      INTEGER            BLOCK_CYCLIC_2D_INB, CSRC_, CTXT_, DLEN_,
     $                   DTYPE_, IMB_, INB_, LLD_, MB_, M_, NB_, N_,
     $                   RSRC_
      parameter( block_cyclic_2d_inb = 2, dlen_ = 11,
     $                   dtype_ = 1, ctxt_ = 2, m_ = 3, n_ = 4,
     $                   imb_ = 5, inb_ = 6, mb_ = 7, nb_ = 8,
     $                   rsrc_ = 9, csrc_ = 10, lld_ = 11 )
*     ..
*     .. Local Scalars ..
      LOGICAL            AISCOLREP, AISROWREP
      INTEGER            H, I, IACOL, IAROW, IB, ICTXT, ICURCOL,
     $                   ICURROW, II, IIA, IN, J, JB, JJ, JJA, JN, K,
     $                   LDA, LDW, MYCOL, MYROW, NPCOL, NPROW
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_barrier, blacs_gridinfo, pb_infog2l,
     $                   sgerv2d, sgesd2d
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          min
*     ..
*     .. Executable Statements ..
*
*     Get grid parameters
*
      ictxt = desca( ctxt_ )
      CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
      CALL pb_infog2l( ia, ja, desca, nprow, npcol, myrow, mycol,
     $                 iia, jja, iarow, iacol )
      ii = iia
      jj = jja
      IF( desca( rsrc_ ).LT.0 ) THEN
         aisrowrep = .true.
         iarow     = prow
         icurrow   = prow
      ELSE
         aisrowrep = .false.
         icurrow   = iarow
      END IF
      IF( desca( csrc_ ).LT.0 ) THEN
         aiscolrep = .true.
         iacol     = pcol
         icurcol   = pcol
      ELSE
         aiscolrep = .false.
         icurcol   = iacol
      END IF
      lda = desca( lld_ )
      ldw = max( desca( imb_ ), desca( mb_ ) )
*
*     Handle the first block of column separately
*
      jb = desca( inb_ ) - ja + 1
      IF( jb.LE.0 )
     $   jb = ( (-jb) / desca( nb_ ) + 1 ) * desca( nb_ ) + jb
      jb = min( jb, n )
      jn = ja+jb-1
      DO 60 h = 0, jb-1
         ib = desca( imb_ ) - ia + 1
         IF( ib.LE.0 )
     $      ib = ( (-ib) / desca( mb_ ) + 1 ) * desca( mb_ ) + ib
         ib = min( ib, m )
         in = ia+ib-1
         IF( icurrow.EQ.irprnt .AND. icurcol.EQ.icprnt ) THEN
            IF( myrow.EQ.irprnt .AND. mycol.EQ.icprnt ) THEN
               DO 10 k = 0, ib-1
                  WRITE( nout, fmt = 9999 )
     $                   cmatnm, ia+k, ja+h, a( ii+k+(jj+h-1)*lda )
   10          CONTINUE
            END IF
         ELSE
            IF( myrow.EQ.icurrow .AND. mycol.EQ.icurcol ) THEN
               CALL sgesd2d( ictxt, ib, 1, a( ii+(jj+h-1)*lda ), lda,
     $                       irprnt, icprnt )
            ELSE IF( myrow.EQ.irprnt .AND. mycol.EQ.icprnt ) THEN
               CALL sgerv2d( ictxt, ib, 1, work, ldw, icurrow, icurcol )
               DO 20 k = 1, ib
                  WRITE( nout, fmt = 9999 )
     $                   cmatnm, ia+k-1, ja+h, work( k )
   20          CONTINUE
            END IF
         END IF
         IF( myrow.EQ.icurrow )
     $      ii = ii + ib
         IF( .NOT.aisrowrep )
     $      icurrow = mod( icurrow+1, nprow )
         CALL blacs_barrier( ictxt, 'All' )
*
*        Loop over remaining block of rows
*
         DO 50 i = in+1, ia+m-1, desca( mb_ )
            ib = min( desca( mb_ ), ia+m-i )
            IF( icurrow.EQ.irprnt .AND. icurcol.EQ.icprnt ) THEN
               IF( myrow.EQ.irprnt .AND. mycol.EQ.icprnt ) THEN
                  DO 30 k = 0, ib-1
                     WRITE( nout, fmt = 9999 )
     $                      cmatnm, i+k, ja+h, a( ii+k+(jj+h-1)*lda )
   30             CONTINUE
               END IF
            ELSE
               IF( myrow.EQ.icurrow .AND. mycol.EQ.icurcol ) THEN
                  CALL sgesd2d( ictxt, ib, 1, a( ii+(jj+h-1)*lda ),
     $                          lda, irprnt, icprnt )
               ELSE IF( myrow.EQ.irprnt .AND. mycol.EQ.icprnt ) THEN
                  CALL sgerv2d( ictxt, ib, 1, work, ldw, icurrow,
     $                          icurcol )
                  DO 40 k = 1, ib
                     WRITE( nout, fmt = 9999 )
     $                      cmatnm, i+k-1, ja+h, work( k )
   40             CONTINUE
               END IF
            END IF
            IF( myrow.EQ.icurrow )
     $         ii = ii + ib
            IF( .NOT.aisrowrep )
     $         icurrow = mod( icurrow+1, nprow )
            CALL blacs_barrier( ictxt, 'All' )
   50    CONTINUE
*
         ii = iia
         icurrow = iarow
   60 CONTINUE
*
      IF( mycol.EQ.icurcol )
     $   jj = jj + jb
      IF( .NOT.aiscolrep )
     $   icurcol = mod( icurcol+1, npcol )
      CALL blacs_barrier( ictxt, 'All' )
*
*     Loop over remaining column blocks
*
      DO 130 j = jn+1, ja+n-1, desca( nb_ )
         jb = min(  desca( nb_ ), ja+n-j )
         DO 120 h = 0, jb-1
            ib = desca( imb_ )-ia+1
            IF( ib.LE.0 )
     $         ib = ( (-ib) / desca( mb_ ) + 1 ) * desca( mb_ ) + ib
            ib = min( ib, m )
            in = ia+ib-1
            IF( icurrow.EQ.irprnt .AND. icurcol.EQ.icprnt ) THEN
               IF( myrow.EQ.irprnt .AND. mycol.EQ.icprnt ) THEN
                  DO 70 k = 0, ib-1
                     WRITE( nout, fmt = 9999 )
     $                      cmatnm, ia+k, j+h, a( ii+k+(jj+h-1)*lda )
   70             CONTINUE
               END IF
            ELSE
               IF( myrow.EQ.icurrow .AND. mycol.EQ.icurcol ) THEN
                  CALL sgesd2d( ictxt, ib, 1, a( ii+(jj+h-1)*lda ),
     $                          lda, irprnt, icprnt )
               ELSE IF( myrow.EQ.irprnt .AND. mycol.EQ.icprnt ) THEN
                  CALL sgerv2d( ictxt, ib, 1, work, ldw, icurrow,
     $                          icurcol )
                  DO 80 k = 1, ib
                     WRITE( nout, fmt = 9999 )
     $                      cmatnm, ia+k-1, j+h, work( k )
   80             CONTINUE
               END IF
            END IF
            IF( myrow.EQ.icurrow )
     $         ii = ii + ib
            icurrow = mod( icurrow+1, nprow )
            CALL blacs_barrier( ictxt, 'All' )
*
*           Loop over remaining block of rows
*
            DO 110 i = in+1, ia+m-1, desca( mb_ )
               ib = min( desca( mb_ ), ia+m-i )
               IF( icurrow.EQ.irprnt .AND. icurcol.EQ.icprnt ) THEN
                  IF( myrow.EQ.irprnt .AND. mycol.EQ.icprnt ) THEN
                     DO 90 k = 0, ib-1
                        WRITE( nout, fmt = 9999 )
     $                         cmatnm, i+k, j+h, a( ii+k+(jj+h-1)*lda )
   90                CONTINUE
                  END IF
               ELSE
                  IF( myrow.EQ.icurrow .AND. mycol.EQ.icurcol ) THEN
                     CALL sgesd2d( ictxt, ib, 1, a( ii+(jj+h-1)*lda ),
     $                             lda, irprnt, icprnt )
                   ELSE IF( myrow.EQ.irprnt .AND. mycol.EQ.icprnt ) THEN
                     CALL sgerv2d( ictxt, ib, 1, work, ldw, icurrow,
     $                             icurcol )
                     DO 100 k = 1, ib
                        WRITE( nout, fmt = 9999 )
     $                         cmatnm, i+k-1, j+h, work( k )
  100                CONTINUE
                  END IF
               END IF
               IF( myrow.EQ.icurrow )
     $            ii = ii + ib
               IF( .NOT.aisrowrep )
     $            icurrow = mod( icurrow+1, nprow )
               CALL blacs_barrier( ictxt, 'All' )
  110       CONTINUE
*
            ii = iia
            icurrow = iarow
  120    CONTINUE
*
         IF( mycol.EQ.icurcol )
     $      jj = jj + jb
         IF( .NOT.aiscolrep )
     $      icurcol = mod( icurcol+1, npcol )
         CALL blacs_barrier( ictxt, 'All' )
*
  130 CONTINUE
*
 9999 FORMAT( 1x, a, '(', i6, ',', i6, ')=', e16.8 )
*
      RETURN
*
*     End of PB_PSLAPRN2
*
      END
      SUBROUTINE pb_sfillpad( ICTXT, M, N, A, LDA, IPRE, IPOST, CHKVAL )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER            ICTXT, IPOST, IPRE, LDA, M, N
      REAL               CHKVAL
*     ..
*     .. Array Arguments ..
      REAL               A( * )
*     ..
*
*  Purpose
*  =======
*
*  PB_SFILLPAD surrounds a two dimensional local array with a guard-zone
*  initialized to the value CHKVAL. The user may later call the  routine
*  PB_SCHEKPAD to discover if the guardzone has been violated. There are
*  three guardzones. The first is a buffer of size  IPRE  that is before
*  the start of the array. The second is the buffer of size IPOST  which
*  is after the end of the array to be padded. Finally, there is a guard
*  zone inside every column of the array to be padded, in  the  elements
*  of A(M+1:LDA, J).
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  M       (local input) INTEGER
*          On entry, M  specifies the number of rows in the local  array
*          A.  M must be at least zero.
*
*  N       (local input) INTEGER
*          On entry, N  specifies the number of columns in the local ar-
*          ray A. N must be at least zero.
*
*  A       (local input/local output) REAL array
*          On entry,  A  is an array of dimension (LDA,N). On exit, this
*          array is the padded array.
*
*  LDA     (local input) INTEGER
*          On entry,  LDA  specifies  the leading dimension of the local
*          array to be padded. LDA must be at least MAX( 1, M ).
*
*  IPRE    (local input) INTEGER
*          On entry, IPRE specifies the size of  the  guard zone  to put
*          before the start of the padded array.
*
*  IPOST   (local input) INTEGER
*          On entry, IPOST specifies the size of the  guard zone  to put
*          after the end of the padded array.
*
*  CHKVAL  (local input) REAL
*          On entry, CHKVAL specifies the value to pad the array with.
*
*  -- Written on April 1, 1998 by
*     R. Clint Whaley, University of Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Local Scalars ..
      INTEGER            I, J, K
*     ..
*     .. Executable Statements ..
*
*     Put check buffer in front of A
*
      IF( IPRE.GT.0 ) THEN
         DO 10 I = 1, ipre
            a( i ) = chkval
   10    CONTINUE
      ELSE
         WRITE( *, fmt = '(A)' )
     $          'WARNING no pre-guardzone in PB_SFILLPAD'
      END IF
*
*     Put check buffer in back of A
*
      IF( ipost.GT.0 ) THEN
         j = ipre+lda*n+1
         DO 20 i = j, j+ipost-1
            a( i ) = chkval
   20    CONTINUE
      ELSE
         WRITE( *, fmt = '(A)' )
     $          'WARNING no post-guardzone in PB_SFILLPAD'
      END IF
*
*     Put check buffer in all (LDA-M) gaps
*
      IF( lda.GT.m ) THEN
         k = ipre + m + 1
         DO 40 j = 1, n
            DO 30 i = k, k + ( lda - m ) - 1
               a( i ) = chkval
   30       CONTINUE
            k = k + lda
   40    CONTINUE
      END IF
*
      RETURN
*
*     End of PB_SFILLPAD
*
      END
      SUBROUTINE pb_schekpad( ICTXT, MESS, M, N, A, LDA, IPRE, IPOST,
     $                        CHKVAL )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER            ICTXT, IPOST, IPRE, LDA, M, N
      REAL               CHKVAL
*     ..
*     .. Array Arguments ..
      CHARACTER*(*)      MESS
      REAL               A( * )
*     ..
*
*  Purpose
*  =======
*
*  PB_SCHEKPAD checks that the padding around a local array has not been
*  overwritten since the call to PB_SFILLPAD.  Three types of errors are
*  reported:
*
*  1) Overwrite in pre-guardzone.  This indicates a memory overwrite has
*  occurred in the  first  IPRE  elements which form a buffer before the
*  beginning of A. Therefore, the error message:
*     'Overwrite in  pre-guardzone: loc(  5) =         18.00000'
*  tells that the 5th element of the IPRE long buffer has been overwrit-
*  ten with the value 18, where it should still have the value CHKVAL.
*
*  2) Overwrite in post-guardzone. This indicates a memory overwrite has
*  occurred in the last IPOST elements which form a buffer after the end
*  of A. Error reports are refered from the end of A.  Therefore,
*     'Overwrite in post-guardzone: loc( 19) =         24.00000'
*  tells  that the  19th element after the end of A was overwritten with
*  the value 24, where it should still have the value of CHKVAL.
*
*  3) Overwrite in lda-m gap.  Tells you elements between M and LDA were
*  overwritten.  So,
*     'Overwrite in lda-m gap: A( 12,  3) =         22.00000'
*  tells  that the element at the 12th row and 3rd column of A was over-
*  written with the value of 22, where it should still have the value of
*  CHKVAL.
*
*  Arguments
*  =========
*
*  ICTXT   (local input) INTEGER
*          On entry,  ICTXT  specifies the BLACS context handle, indica-
*          ting the global  context of the operation. The context itself
*          is global, but the value of ICTXT is local.
*
*  MESS    (local input) CHARACTER*(*)
*          On entry, MESS is a ttring containing a user-defined message.
*
*  M       (local input) INTEGER
*          On entry, M  specifies the number of rows in the local  array
*          A.  M must be at least zero.
*
*  N       (local input) INTEGER
*          On entry, N  specifies the number of columns in the local ar-
*          ray A. N must be at least zero.
*
*  A       (local input) REAL array
*          On entry,  A  is an array of dimension (LDA,N).
*
*  LDA     (local input) INTEGER
*          On entry,  LDA  specifies  the leading dimension of the local
*          array to be padded. LDA must be at least MAX( 1, M ).
*
*  IPRE    (local input) INTEGER
*          On entry, IPRE specifies the size of  the  guard zone  to put
*          before the start of the padded array.
*
*  IPOST   (local input) INTEGER
*          On entry, IPOST specifies the size of the  guard zone  to put
*          after the end of the padded array.
*
*  CHKVAL  (local input) REAL
*          On entry, CHKVAL specifies the value to pad the array with.
*
*
*  -- Written on April 1, 1998 by
*     R. Clint Whaley, University of Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Local Scalars ..
      CHARACTER*1        TOP
      INTEGER            I, IAM, IDUMM, INFO, J, K, MYCOL, MYROW, NPCOL,
     $                   NPROW
*     ..
*     .. External Subroutines ..
      EXTERNAL           BLACS_GRIDINFO, IGAMX2D, PB_TOPGET
*     ..
*     .. Executable Statements ..
*
*     Get grid parameters
*
      CALL BLACS_GRIDINFO( ICTXT, NPROW, NPCOL, MYROW, MYCOL )
      IAM  = myrow*npcol + mycol
      info = -1
*
*     Check buffer in front of A
*
      IF( ipre.GT.0 ) THEN
         DO 10 i = 1, ipre
            IF( a( i ).NE.chkval ) THEN
               WRITE( *, fmt = 9998 ) myrow, mycol, mess, ' pre', i,
     $                                a( i )
               info = iam
            END IF
   10    CONTINUE
      ELSE
         WRITE( *, fmt = * ) 'WARNING no pre-guardzone in PB_SCHEKPAD'
      END IF
*
*     Check buffer after A
*
      IF( ipost.GT.0 ) THEN
         j = ipre+lda*n+1
         DO 20 i = j, j+ipost-1
            IF( a( i ).NE.chkval ) THEN
               WRITE( *, fmt = 9998 ) myrow, mycol, mess, 'post',
     $                                i-j+1, a( i )
               info = iam
            END IF
   20    CONTINUE
      ELSE
         WRITE( *, fmt = * )
     $          'WARNING no post-guardzone buffer in PB_SCHEKPAD'
      END IF
*
*     Check all (LDA-M) gaps
*
      IF( lda.GT.m ) THEN
         k = ipre + m + 1
         DO 40 j = 1, n
            DO 30 i = k, k + (lda-m) - 1
               IF( a( i ).NE.chkval ) THEN
                  WRITE( *, fmt = 9997 ) myrow, mycol, mess,
     $               i-ipre-lda*(j-1), j, a( i )
                  info = iam
               END IF
   30       CONTINUE
            k = k + lda
   40    CONTINUE
      END IF
*
      CALL pb_topget( ictxt, 'Combine', 'All', top )
      CALL igamx2d( ictxt, 'All', top, 1, 1, info, 1, idumm, idumm, -1,
     $              0, 0 )
      IF( iam.EQ.0 .AND. info.GE.0 ) THEN
         WRITE( *, fmt = 9999 ) info / npcol, mod( info, npcol ), mess
      END IF
*
 9999 FORMAT( '{', i5, ',', i5, '}:  Memory overwrite in ', a )
 9998 FORMAT( '{', i5, ',', i5, '}:  ', a, ' memory overwrite in ',
     $        a4, '-guardzone: loc(', i3, ') = ', g11.4 )
 9997 FORMAT( '{', i5, ',', i5, '}: ', a, ' memory overwrite in ',
     $        'lda-m gap: loc(', i3, ',', i3, ') = ', g11.4 )
*
      RETURN
*
*     End of PB_SCHEKPAD
*
      END
      SUBROUTINE pb_slaset( UPLO, M, N, IOFFD, ALPHA, BETA, A, LDA )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1        UPLO
      INTEGER            IOFFD, LDA, M, N
      REAL               ALPHA, BETA
*     ..
*     .. Array Arguments ..
      REAL               A( LDA, * )
*     ..
*
*  Purpose
*  =======
*
*  PB_SLASET initializes a two-dimensional array A to beta on the diago-
*  nal specified by IOFFD and alpha on the offdiagonals.
*
*  Arguments
*  =========
*
*  UPLO    (global input) CHARACTER*1
*          On entry,  UPLO  specifies  which trapezoidal part of the ar-
*          ray A is to be set as follows:
*             = 'L' or 'l':   Lower triangular part is set; the strictly
*                             upper triangular part of A is not changed,
*             = 'U' or 'u':   Upper triangular part is set; the strictly
*                             lower triangular part of A is not changed,
*             = 'D' or 'd'    Only the diagonal of A is set,
*             Otherwise:      All of the array A is set.
*
*  M       (input) INTEGER
*          On entry,  M  specifies the number of rows of the array A.  M
*          must be at least zero.
*
*  N       (input) INTEGER
*          On entry,  N  specifies the number of columns of the array A.
*          N must be at least zero.
*
*  IOFFD   (input) INTEGER
*          On entry, IOFFD specifies the position of the offdiagonal de-
*          limiting the upper and lower trapezoidal part of A as follows
*          (see the notes below):
*
*             IOFFD = 0  specifies the main diagonal A( i, i ),
*                        with i = 1 ... MIN( M, N ),
*             IOFFD > 0  specifies the subdiagonal   A( i+IOFFD, i ),
*                        with i = 1 ... MIN( M-IOFFD, N ),
*             IOFFD < 0  specifies the superdiagonal A( i, i-IOFFD ),
*                        with i = 1 ... MIN( M, N+IOFFD ).
*
*  ALPHA   (input) REAL
*          On entry,  ALPHA specifies the value to which the offdiagonal
*          array elements are set to.
*
*  BETA    (input) REAL
*          On entry, BETA  specifies the value to which the diagonal ar-
*          ray elements are set to.
*
*  A       (input/output) REAL array
*          On entry, A is an array of dimension  (LDA,N).  Before  entry
*          with UPLO = 'U' or 'u', the leading m by n part of the  array
*          A  must  contain  the upper trapezoidal part of the matrix as
*          specified by IOFFD to be set, and  the  strictly lower trape-
*          zoidal  part of A is not referenced; When IUPLO = 'L' or 'l',
*          the leading m by n part of  the  array  A  must  contain  the
*          lower trapezoidal part of the matrix as specified by IOFFD to
*          be set,  and  the  strictly  upper  trapezoidal part of  A is
*          not referenced.
*
*  LDA     (input) INTEGER
*          On entry, LDA specifies the leading dimension of the array A.
*          LDA must be at least max( 1, M ).
*
*  Notes
*  =====
*                           N                                    N
*             ----------------------------                  -----------
*            |       d                    |                |           |
*          M |         d        'U'       |                |      'U'  |
*            |  'L'     'D'               |                |d          |
*            |             d              |              M |  d        |
*             ----------------------------                 |   'D'     |
*                                                          |      d    |
*               IOFFD < 0                                  | 'L'    d  |
*                                                          |          d|
*                  N                                       |           |
*             -----------                                   -----------
*            |    d   'U'|
*            |      d    |                                   IOFFD > 0
*          M |       'D' |
*            |          d|                              N
*            |  'L'      |                 ----------------------------
*            |           |                |          'U'               |
*            |           |                |d                           |
*            |           |                | 'D'                        |
*            |           |                |    d                       |
*            |           |                |'L'   d                     |
*             -----------                  ----------------------------
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Local Scalars ..
      INTEGER            I, J, JTMP, MN
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      EXTERNAL           LSAME
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          MAX, MIN
*     ..
*     .. Executable Statements ..
*
*     Quick return if possible
*
      IF( M.LE.0 .OR. N.LE.0 )
     $   RETURN
*
*     Start the operations
*
      IF( LSAME( UPLO, 'L' ) ) THEN
*
*        Set the diagonal to BETA and the strictly lower triangular
*        part of the array to ALPHA.
*
         mn = max( 0, -ioffd )
         DO 20 j = 1, min( mn, n )
            DO 10 i = 1, m
               a( i, j ) = alpha
   10       CONTINUE
   20    CONTINUE
         DO 40 j = mn + 1, min( m - ioffd, n )
            jtmp = j + ioffd
            a( jtmp, j ) = beta
            DO 30 i = jtmp + 1, m
               a( i, j ) = alpha
   30       CONTINUE
   40    CONTINUE
*
      ELSE IF( lsame( uplo, 'U' ) ) THEN
*
*        Set the diagonal to BETA and the strictly upper triangular
*        part of the array to ALPHA.
*
         mn = min( m - ioffd, n )
         DO 60 j = max( 0, -ioffd ) + 1, mn
            jtmp = j + ioffd
            DO 50 i = 1, jtmp - 1
               a( i, j ) = alpha
   50       CONTINUE
            a( jtmp, j ) = beta
   60    CONTINUE
         DO 80 j = max( 0, mn ) + 1, n
            DO 70 i = 1, m
               a( i, j ) = alpha
   70       CONTINUE
   80    CONTINUE
*
      ELSE IF( lsame( uplo, 'D' ) ) THEN
*
*        Set the array to BETA on the diagonal.
*
         DO 90 j = max( 0, -ioffd ) + 1, min( m - ioffd, n )
            a( j + ioffd, j ) = beta
   90    CONTINUE
*
      ELSE
*
*        Set the array to BETA on the diagonal and ALPHA on the
*        offdiagonal.
*
         DO 110 j = 1, n
            DO 100 i = 1, m
               a( i, j ) = alpha
  100       CONTINUE
  110    CONTINUE
         IF( alpha.NE.beta .AND. ioffd.LT.m .AND. ioffd.GT.-n ) THEN
            DO 120 j = max( 0, -ioffd ) + 1, min( m - ioffd, n )
               a( j + ioffd, j ) = beta
  120       CONTINUE
         END IF
*
      END IF
*
      RETURN
*
*     End of PB_SLASET
*
      END
      SUBROUTINE pb_slascal( UPLO, M, N, IOFFD, ALPHA, A, LDA )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1        UPLO
      INTEGER            IOFFD, LDA, M, N
      REAL               ALPHA
*     ..
*     .. Array Arguments ..
      REAL               A( LDA, * )
*     ..
*
*  Purpose
*  =======
*
*  PB_SLASCAL scales a two-dimensional array A by the scalar alpha.
*
*  Arguments
*  =========
*
*  UPLO    (input) CHARACTER*1
*          On entry,  UPLO  specifies  which trapezoidal part of the ar-
*          ray A is to be scaled as follows:
*             = 'L' or 'l':          the lower trapezoid of A is scaled,
*             = 'U' or 'u':          the upper trapezoid of A is scaled,
*             = 'D' or 'd':       diagonal specified by IOFFD is scaled,
*             Otherwise:                   all of the array A is scaled.
*
*  M       (input) INTEGER
*          On entry,  M  specifies the number of rows of the array A.  M
*          must be at least zero.
*
*  N       (input) INTEGER
*          On entry,  N  specifies the number of columns of the array A.
*          N must be at least zero.
*
*  IOFFD   (input) INTEGER
*          On entry, IOFFD specifies the position of the offdiagonal de-
*          limiting the upper and lower trapezoidal part of A as follows
*          (see the notes below):
*
*             IOFFD = 0  specifies the main diagonal A( i, i ),
*                        with i = 1 ... MIN( M, N ),
*             IOFFD > 0  specifies the subdiagonal   A( i+IOFFD, i ),
*                        with i = 1 ... MIN( M-IOFFD, N ),
*             IOFFD < 0  specifies the superdiagonal A( i, i-IOFFD ),
*                        with i = 1 ... MIN( M, N+IOFFD ).
*
*  ALPHA   (input) REAL
*          On entry, ALPHA specifies the scalar alpha.
*
*  A       (input/output) REAL array
*          On entry, A is an array of dimension  (LDA,N).  Before  entry
*          with  UPLO = 'U' or 'u', the leading m by n part of the array
*          A must contain the upper trapezoidal  part  of the matrix  as
*          specified by  IOFFD to be scaled, and the strictly lower tra-
*          pezoidal part of A is not referenced; When UPLO = 'L' or 'l',
*          the leading m by n part of the array A must contain the lower
*          trapezoidal  part  of  the matrix as specified by IOFFD to be
*          scaled,  and  the strictly upper trapezoidal part of A is not
*          referenced. On exit, the entries of the  trapezoid part of  A
*          determined by UPLO and IOFFD are scaled.
*
*  LDA     (input) INTEGER
*          On entry, LDA specifies the leading dimension of the array A.
*          LDA must be at least max( 1, M ).
*
*  Notes
*  =====
*                           N                                    N
*             ----------------------------                  -----------
*            |       d                    |                |           |
*          M |         d        'U'       |                |      'U'  |
*            |  'L'     'D'               |                |d          |
*            |             d              |              M |  d        |
*             ----------------------------                 |   'D'     |
*                                                          |      d    |
*              IOFFD < 0                                   | 'L'    d  |
*                                                          |          d|
*                  N                                       |           |
*             -----------                                   -----------
*            |    d   'U'|
*            |      d    |                                   IOFFD > 0
*          M |       'D' |
*            |          d|                              N
*            |  'L'      |                 ----------------------------
*            |           |                |          'U'               |
*            |           |                |d                           |
*            |           |                | 'D'                        |
*            |           |                |    d                       |
*            |           |                |'L'   d                     |
*             -----------                  ----------------------------
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Local Scalars ..
      INTEGER            I, J, JTMP, MN
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      EXTERNAL           LSAME
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          MAX, MIN
*     ..
*     .. Executable Statements ..
*
*     Quick return if possible
*
      IF( M.LE.0 .OR. N.LE.0 )
     $   RETURN
*
*     Start the operations
*
      IF( LSAME( UPLO, 'L' ) ) THEN
*
*        Scales the lower triangular part of the array by ALPHA.
*
         MN = max( 0, -ioffd )
         DO 20 j = 1, min( mn, n )
            DO 10 i = 1, m
               a( i, j ) = alpha * a( i, j )
   10       CONTINUE
   20    CONTINUE
         DO 40 j = mn + 1, min( m - ioffd, n )
            DO 30 i = j + ioffd, m
               a( i, j ) = alpha * a( i, j )
   30       CONTINUE
   40    CONTINUE
*
      ELSE IF( lsame( uplo, 'U' ) ) THEN
*
*        Scales the upper triangular part of the array by ALPHA.
*
         mn = min( m - ioffd, n )
         DO 60 j = max( 0, -ioffd ) + 1, mn
            DO 50 i = 1, j + ioffd
               a( i, j ) = alpha * a( i, j )
   50       CONTINUE
   60    CONTINUE
         DO 80 j = max( 0, mn ) + 1, n
            DO 70 i = 1, m
               a( i, j ) = alpha * a( i, j )
   70       CONTINUE
   80    CONTINUE
*
      ELSE IF( lsame( uplo, 'D' ) ) THEN
*
*        Scales the diagonal entries by ALPHA.
*
         DO 90 j = max( 0, -ioffd ) + 1, min( m - ioffd, n )
            jtmp = j + ioffd
            a( jtmp, j ) = alpha * a( jtmp, j )
   90    CONTINUE
*
      ELSE
*
*        Scales the entire array by ALPHA.
*
         DO 110 j = 1, n
            DO 100 i = 1, m
               a( i, j ) = alpha * a( i, j )
  100       CONTINUE
  110    CONTINUE
*
      END IF
*
      RETURN
*
*     End of PB_SLASCAL
*
      END
      SUBROUTINE pb_slagen( UPLO, AFORM, A, LDA, LCMT00, IRAN, MBLKS,
     $                      IMBLOC, MB, LMBLOC, NBLKS, INBLOC, NB,
     $                      LNBLOC, JMP, IMULADD )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      CHARACTER*1        UPLO, AFORM
      INTEGER            IMBLOC, INBLOC, LCMT00, LDA, LMBLOC, LNBLOC,
     $                   mb, mblks, nb, nblks
*     ..
*     .. Array Arguments ..
      INTEGER            IMULADD( 4, * ), IRAN( * ), JMP( * )
      REAL               A( LDA, * )
*     ..
*
*  Purpose
*  =======
*
*  PB_SLAGEN locally initializes an array A.
*
*  Arguments
*  =========
*
*  UPLO    (global input) CHARACTER*1
*          On entry, UPLO  specifies whether the lower (UPLO='L') trape-
*          zoidal part or the upper (UPLO='U') trapezoidal part is to be
*          generated  when  the  matrix  to be generated is symmetric or
*          Hermitian. For  all  the  other values of AFORM, the value of
*          this input argument is ignored.
*
*  AFORM   (global input) CHARACTER*1
*          On entry, AFORM specifies the type of submatrix to be genera-
*          ted as follows:
*             AFORM = 'S', sub( A ) is a symmetric matrix,
*             AFORM = 'H', sub( A ) is a Hermitian matrix,
*             AFORM = 'T', sub( A ) is overrwritten  with  the transpose
*                          of what would normally be generated,
*             AFORM = 'C', sub( A ) is overwritten  with  the  conjugate
*                          transpose  of  what would normally be genera-
*                          ted.
*             AFORM = 'N', a random submatrix is generated.
*
*  A       (local output) REAL array
*          On entry,  A  is  an array of dimension (LLD_A, *).  On exit,
*          this array contains the local entries of the randomly genera-
*          ted submatrix sub( A ).
*
*  LDA     (local input) INTEGER
*          On entry,  LDA  specifies  the local leading dimension of the
*          array A. LDA must be at least one.
*
*  LCMT00  (global input) INTEGER
*          On entry, LCMT00 is the LCM value specifying the off-diagonal
*          of the underlying matrix of interest. LCMT00=0 specifies  the
*          main diagonal, LCMT00 > 0 specifies a subdiagonal, LCMT00 < 0
*          specifies superdiagonals.
*
*  IRAN    (local input) INTEGER array
*          On entry, IRAN  is an array of dimension 2 containing respec-
*          tively the 16-lower and 16-higher bits of the encoding of the
*          entry of  the  random  sequence  corresponding locally to the
*          first local array entry to generate. Usually,  this  array is
*          computed by PB_SETLOCRAN.
*
*  MBLKS   (local input) INTEGER
*          On entry, MBLKS specifies the local number of blocks of rows.
*          MBLKS is at least zero.
*
*  IMBLOC  (local input) INTEGER
*          On entry, IMBLOC specifies  the  number of rows (size) of the
*          local uppest  blocks. IMBLOC is at least zero.
*
*  MB      (global input) INTEGER
*          On entry, MB  specifies the blocking factor used to partition
*          the rows of the matrix.  MB  must be at least one.
*
*  LMBLOC  (local input) INTEGER
*          On entry, LMBLOC specifies the number of  rows  (size) of the
*          local lowest blocks. LMBLOC is at least zero.
*
*  NBLKS   (local input) INTEGER
*          On entry,  NBLKS  specifies the local number of blocks of co-
*          lumns. NBLKS is at least zero.
*
*  INBLOC  (local input) INTEGER
*          On entry,  INBLOC  specifies the number of columns (size)  of
*          the local leftmost blocks. INBLOC is at least zero.
*
*  NB      (global input) INTEGER
*          On entry, NB  specifies the blocking factor used to partition
*          the the columns of the matrix.  NB  must be at least one.
*
*  LNBLOC  (local input) INTEGER
*          On entry,  LNBLOC  specifies  the number of columns (size) of
*          the local rightmost blocks. LNBLOC is at least zero.
*
*  JMP     (local input) INTEGER array
*          On entry, JMP is an array of dimension JMP_LEN containing the
*          different jump values used by the random matrix generator.
*
*  IMULADD (local input) INTEGER array
*          On entry, IMULADD is an array of dimension (4, JMP_LEN).  The
*          jth  column  of this array contains the encoded initial cons-
*          tants a_j and c_j to  jump  from X( n ) to  X( n + JMP( j ) )
*          (= a_j * X( n ) + c_j) in the random sequence. IMULADD(1:2,j)
*          contains respectively the 16-lower and 16-higher bits of  the
*          constant a_j, and IMULADD(3:4,j)  contains  the 16-lower  and
*          16-higher bits of the constant c_j.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            JMP_1, JMP_COL, JMP_IMBV, JMP_INBV, JMP_LEN,
     $                   JMP_MB, JMP_NB, JMP_NPIMBLOC, JMP_NPMB,
     $                   JMP_NQINBLOC, JMP_NQNB, JMP_ROW
      PARAMETER          ( JMP_1 = 1, jmp_row = 2, jmp_col = 3,
     $                   jmp_mb = 4, jmp_imbv = 5, jmp_npmb = 6,
     $                   jmp_npimbloc = 7, jmp_nb = 8, jmp_inbv = 9,
     $                   jmp_nqnb = 10, jmp_nqinbloc = 11,
     $                   jmp_len = 11 )
*     ..
*     .. Local Scalars ..
      INTEGER            I, IB, IBLK, II, IK, ITMP, JB, JBLK, JJ, JK,
     $                   JTMP, LCMTC, LCMTR, LOW, MNB, UPP
      REAL               DUMMY
*     ..
*     .. Local Arrays ..
      INTEGER            IB0( 2 ), IB1( 2 ), IB2( 2 ), IB3( 2 )
*     ..
*     .. External Subroutines ..
      EXTERNAL           PB_JUMPIT
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      REAL               PB_SRAND
      EXTERNAL           LSAME, PB_SRAND
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          max, min
*     ..
*     .. Executable Statements ..
*
      DO 10 i = 1, 2
         ib1( i ) = iran( i )
         ib2( i ) = iran( i )
         ib3( i ) = iran( i )
   10 CONTINUE
*
      IF( lsame( aform, 'N' ) ) THEN
*
*        Generate random matrix
*
         jj = 1
*
         DO 50 jblk = 1, nblks
*
            IF( jblk.EQ.1 ) THEN
               jb = inbloc
            ELSE IF( jblk.EQ.nblks ) THEN
               jb = lnbloc
            ELSE
               jb = nb
            END IF
*
            DO 40 jk = jj, jj + jb - 1
*
               ii = 1
*
               DO 30 iblk = 1, mblks
*
                  IF( iblk.EQ.1 ) THEN
                     ib = imbloc
                  ELSE IF( iblk.EQ.mblks ) THEN
                     ib = lmbloc
                  ELSE
                     ib = mb
                  END IF
*
*                 Blocks are IB by JB
*
                  DO 20 ik = ii, ii + ib - 1
                     a( ik, jk ) = pb_srand( 0 )
   20             CONTINUE
*
                  ii = ii + ib
*
                  IF( iblk.EQ.1 ) THEN
*
*                    Jump IMBLOC + ( NPROW - 1 ) * MB rows
*
                     CALL pb_jumpit( imuladd( 1, jmp_npimbloc ), ib1,
     $                               ib0 )
*
                  ELSE
*
*                    Jump NPROW * MB rows
*
                     CALL pb_jumpit( imuladd( 1, jmp_npmb ), ib1, ib0 )
*
                  END IF
*
                  ib1( 1 ) = ib0( 1 )
                  ib1( 2 ) = ib0( 2 )
*
   30          CONTINUE
*
*              Jump one column
*
               CALL pb_jumpit( imuladd( 1, jmp_col ), ib2, ib0 )
*
               ib1( 1 ) = ib0( 1 )
               ib1( 2 ) = ib0( 2 )
               ib2( 1 ) = ib0( 1 )
               ib2( 2 ) = ib0( 2 )
*
   40       CONTINUE
*
            jj = jj + jb
*
            IF( jblk.EQ.1 ) THEN
*
*              Jump INBLOC + ( NPCOL - 1 ) * NB columns
*
               CALL pb_jumpit( imuladd( 1, jmp_nqinbloc ), ib3, ib0 )
*
            ELSE
*
*              Jump NPCOL * NB columns
*
               CALL pb_jumpit( imuladd( 1, jmp_nqnb ), ib3, ib0 )
*
            END IF
*
            ib1( 1 ) = ib0( 1 )
            ib1( 2 ) = ib0( 2 )
            ib2( 1 ) = ib0( 1 )
            ib2( 2 ) = ib0( 2 )
            ib3( 1 ) = ib0( 1 )
            ib3( 2 ) = ib0( 2 )
*
   50    CONTINUE
*
      ELSE IF( lsame( aform, 'T' ) .OR. lsame( aform, 'C' ) ) THEN
*
*        Generate the transpose of the matrix that would be normally
*        generated.
*
         ii = 1
*
         DO 90 iblk = 1, mblks
*
            IF( iblk.EQ.1 ) THEN
               ib = imbloc
            ELSE IF( iblk.EQ.mblks ) THEN
               ib = lmbloc
            ELSE
               ib = mb
            END IF
*
            DO 80 ik = ii, ii + ib - 1
*
               jj = 1
*
               DO 70 jblk = 1, nblks
*
                  IF( jblk.EQ.1 ) THEN
                     jb = inbloc
                  ELSE IF( jblk.EQ.nblks ) THEN
                     jb = lnbloc
                  ELSE
                     jb = nb
                  END IF
*
*                 Blocks are IB by JB
*
                  DO 60 jk = jj, jj + jb - 1
                     a( ik, jk ) = pb_srand( 0 )
   60             CONTINUE
*
                  jj = jj + jb
*
                  IF( jblk.EQ.1 ) THEN
*
*                    Jump INBLOC + ( NPCOL - 1 ) * NB columns
*
                     CALL pb_jumpit( imuladd( 1, jmp_nqinbloc ), ib1,
     $                               ib0 )
*
                  ELSE
*
*                    Jump NPCOL * NB columns
*
                     CALL pb_jumpit( imuladd( 1, jmp_nqnb ), ib1, ib0 )
*
                  END IF
*
                  ib1( 1 ) = ib0( 1 )
                  ib1( 2 ) = ib0( 2 )
*
   70          CONTINUE
*
*              Jump one row
*
               CALL pb_jumpit( imuladd( 1, jmp_row ), ib2, ib0 )
*
               ib1( 1 ) = ib0( 1 )
               ib1( 2 ) = ib0( 2 )
               ib2( 1 ) = ib0( 1 )
               ib2( 2 ) = ib0( 2 )
*
   80       CONTINUE
*
            ii = ii + ib
*
            IF( iblk.EQ.1 ) THEN
*
*              Jump IMBLOC + ( NPROW - 1 ) * MB rows
*
               CALL pb_jumpit( imuladd( 1, jmp_npimbloc ), ib3, ib0 )
*
            ELSE
*
*              Jump NPROW * MB rows
*
               CALL pb_jumpit( imuladd( 1, jmp_npmb ), ib3, ib0 )
*
            END IF
*
            ib1( 1 ) = ib0( 1 )
            ib1( 2 ) = ib0( 2 )
            ib2( 1 ) = ib0( 1 )
            ib2( 2 ) = ib0( 2 )
            ib3( 1 ) = ib0( 1 )
            ib3( 2 ) = ib0( 2 )
*
   90    CONTINUE
*
      ELSE IF( ( lsame( aform, 'S' ) ).OR.( lsame( aform, 'H' ) ) ) THEN
*
*        Generate a symmetric matrix
*
         IF( lsame( uplo, 'L' ) ) THEN
*
*           generate lower trapezoidal part
*
            jj = 1
            lcmtc = lcmt00
*
            DO 170 jblk = 1, nblks
*
               IF( jblk.EQ.1 ) THEN
                  jb  = inbloc
                  low = 1 - inbloc
               ELSE IF( jblk.EQ.nblks ) THEN
                  jb = lnbloc
                  low = 1 - nb
               ELSE
                  jb  = nb
                  low = 1 - nb
               END IF
*
               DO 160 jk = jj, jj + jb - 1
*
                  ii = 1
                  lcmtr = lcmtc
*
                  DO 150 iblk = 1, mblks
*
                     IF( iblk.EQ.1 ) THEN
                        ib  = imbloc
                        upp = imbloc - 1
                     ELSE IF( iblk.EQ.mblks ) THEN
                        ib  = lmbloc
                        upp = mb - 1
                     ELSE
                        ib  = mb
                        upp = mb - 1
                     END IF
*
*                    Blocks are IB by JB
*
                     IF( lcmtr.GT.upp ) THEN
*
                        DO 100 ik = ii, ii + ib - 1
                           dummy = pb_srand( 0 )
  100                   CONTINUE
*
                     ELSE IF( lcmtr.GE.low ) THEN
*
                        jtmp = jk - jj + 1
                        mnb  = max( 0, -lcmtr )
*
                        IF( jtmp.LE.min( mnb, jb ) ) THEN
*
                           DO 110 ik = ii, ii + ib - 1
                              a( ik, jk ) = pb_srand( 0 )
  110                      CONTINUE
*
                        ELSE IF( ( jtmp.GE.( mnb + 1 )         ) .AND.
     $                           ( jtmp.LE.min( ib-lcmtr, jb ) ) ) THEN
*
                           itmp = ii + jtmp + lcmtr - 1
*
                           DO 120 ik = ii, itmp - 1
                              dummy = pb_srand( 0 )
  120                      CONTINUE
*
                           DO 130 ik = itmp, ii + ib - 1
                              a( ik, jk ) = pb_srand( 0 )
  130                      CONTINUE
*
                        END IF
*
                     ELSE
*
                        DO 140 ik = ii, ii + ib - 1
                           a( ik, jk ) = pb_srand( 0 )
  140                   CONTINUE
*
                     END IF
*
                     ii = ii + ib
*
                     IF( iblk.EQ.1 ) THEN
*
*                       Jump IMBLOC + ( NPROW - 1 ) * MB rows
*
                        lcmtr = lcmtr - jmp( jmp_npimbloc )
                        CALL pb_jumpit( imuladd( 1, jmp_npimbloc ), ib1,
     $                                  ib0 )
*
                     ELSE
*
*                       Jump NPROW * MB rows
*
                        lcmtr = lcmtr - jmp( jmp_npmb )
                        CALL pb_jumpit( imuladd( 1, jmp_npmb ), ib1,
     $                                  ib0 )
*
                     END IF
*
                     ib1( 1 ) = ib0( 1 )
                     ib1( 2 ) = ib0( 2 )
*
  150             CONTINUE
*
*                 Jump one column
*
                  CALL pb_jumpit( imuladd( 1, jmp_col ), ib2, ib0 )
*
                  ib1( 1 ) = ib0( 1 )
                  ib1( 2 ) = ib0( 2 )
                  ib2( 1 ) = ib0( 1 )
                  ib2( 2 ) = ib0( 2 )
*
  160          CONTINUE
*
               jj = jj + jb
*
               IF( jblk.EQ.1 ) THEN
*
*                 Jump INBLOC + ( NPCOL - 1 ) * NB columns
*
                  lcmtc = lcmtc + jmp( jmp_nqinbloc )
                  CALL pb_jumpit( imuladd( 1, jmp_nqinbloc ), ib3, ib0 )
*
               ELSE
*
*                 Jump NPCOL * NB columns
*
                  lcmtc = lcmtc + jmp( jmp_nqnb )
                  CALL pb_jumpit( imuladd( 1, jmp_nqnb ), ib3, ib0 )
*
               END IF
*
               ib1( 1 ) = ib0( 1 )
               ib1( 2 ) = ib0( 2 )
               ib2( 1 ) = ib0( 1 )
               ib2( 2 ) = ib0( 2 )
               ib3( 1 ) = ib0( 1 )
               ib3( 2 ) = ib0( 2 )
*
  170       CONTINUE
*
         ELSE
*
*           generate upper trapezoidal part
*
            ii = 1
            lcmtr = lcmt00
*
            DO 250 iblk = 1, mblks
*
               IF( iblk.EQ.1 ) THEN
                  ib  = imbloc
                  upp = imbloc - 1
               ELSE IF( iblk.EQ.mblks ) THEN
                  ib  = lmbloc
                  upp = mb - 1
               ELSE
                  ib  = mb
                  upp = mb - 1
               END IF
*
               DO 240 ik = ii, ii + ib - 1
*
                  jj = 1
                  lcmtc = lcmtr
*
                  DO 230 jblk = 1, nblks
*
                     IF( jblk.EQ.1 ) THEN
                        jb  = inbloc
                        low = 1 - inbloc
                     ELSE IF( jblk.EQ.nblks ) THEN
                        jb  = lnbloc
                        low = 1 - nb
                     ELSE
                        jb  = nb
                        low = 1 - nb
                     END IF
*
*                    Blocks are IB by JB
*
                     IF( lcmtc.LT.low ) THEN
*
                        DO 180 jk = jj, jj + jb - 1
                           dummy = pb_srand( 0 )
  180                   CONTINUE
*
                     ELSE IF( lcmtc.LE.upp ) THEN
*
                        itmp = ik - ii + 1
                        mnb  = max( 0, lcmtc )
*
                        IF( itmp.LE.min( mnb, ib ) ) THEN
*
                           DO 190 jk = jj, jj + jb - 1
                              a( ik, jk ) = pb_srand( 0 )
  190                      CONTINUE
*
                        ELSE IF( ( itmp.GE.( mnb + 1 )         ) .AND.
     $                           ( itmp.LE.min( jb+lcmtc, ib ) ) ) THEN
*
                           jtmp = jj + itmp - lcmtc - 1
*
                           DO 200 jk = jj, jtmp - 1
                              dummy = pb_srand( 0 )
  200                      CONTINUE
*
                           DO 210 jk = jtmp, jj + jb - 1
                              a( ik, jk ) = pb_srand( 0 )
  210                      CONTINUE
*
                        END IF
*
                     ELSE
*
                        DO 220 jk = jj, jj + jb - 1
                           a( ik, jk ) = pb_srand( 0 )
  220                   CONTINUE
*
                     END IF
*
                     jj = jj + jb
*
                     IF( jblk.EQ.1 ) THEN
*
*                       Jump INBLOC + ( NPCOL - 1 ) * NB columns
*
                        lcmtc = lcmtc + jmp( jmp_nqinbloc )
                        CALL pb_jumpit( imuladd( 1, jmp_nqinbloc ), ib1,
     $                                  ib0 )
*
                     ELSE
*
*                       Jump NPCOL * NB columns
*
                        lcmtc = lcmtc + jmp( jmp_nqnb )
                        CALL pb_jumpit( imuladd( 1, jmp_nqnb ), ib1,
     $                                  ib0 )
*
                     END IF
*
                     ib1( 1 ) = ib0( 1 )
                     ib1( 2 ) = ib0( 2 )
*
  230             CONTINUE
*
*                 Jump one row
*
                  CALL pb_jumpit( imuladd( 1, jmp_row ), ib2, ib0 )
*
                  ib1( 1 ) = ib0( 1 )
                  ib1( 2 ) = ib0( 2 )
                  ib2( 1 ) = ib0( 1 )
                  ib2( 2 ) = ib0( 2 )
*
  240          CONTINUE
*
               ii = ii + ib
*
               IF( iblk.EQ.1 ) THEN
*
*                 Jump IMBLOC + ( NPROW - 1 ) * MB rows
*
                  lcmtr = lcmtr - jmp( jmp_npimbloc )
                  CALL pb_jumpit( imuladd( 1, jmp_npimbloc ), ib3, ib0 )
*
               ELSE
*
*                 Jump NPROW * MB rows
*
                  lcmtr = lcmtr - jmp( jmp_npmb )
                  CALL pb_jumpit( imuladd( 1, jmp_npmb ), ib3, ib0 )
*
               END IF
*
               ib1( 1 ) = ib0( 1 )
               ib1( 2 ) = ib0( 2 )
               ib2( 1 ) = ib0( 1 )
               ib2( 2 ) = ib0( 2 )
               ib3( 1 ) = ib0( 1 )
               ib3( 2 ) = ib0( 2 )
*
  250       CONTINUE
*
         END IF
*
      END IF
*
      RETURN
*
*     End of PB_SLAGEN
*
      END
      REAL               FUNCTION PB_SRAND( IDUMM )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER            idumm
*     ..
*
*  Purpose
*  =======
*
*  PB_SRAND generates the next number in the random sequence. This func-
*  tion ensures that this number will be in the interval ( -1.0, 1.0 ).
*
*  Arguments
*  =========
*
*  IDUMM   (local input) INTEGER
*          This argument is ignored, but necessary to a FORTRAN 77 func-
*          tion.
*
*  Further Details
*  ===============
*
*  On entry, the array IRAND stored in the common block  RANCOM contains
*  the information (2 integers)  required to generate the next number in
*  the sequence X( n ). This number is computed as
*
*     X( n ) = ( 2^16 * IRAND( 2 ) + IRAND( 1 ) ) / d,
*
*  where the constant d is the  largest  32 bit  positive  integer.  The
*  array  IRAND  is  then  updated for the generation of the next number
*  X( n+1 ) in the random sequence as follows X( n+1 ) = a * X( n ) + c.
*  The constants  a  and c  should have been preliminarily stored in the
*  array  IACS  as  2 pairs of integers. The initial set up of IRAND and
*  IACS is performed by the routine PB_SETRAN.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      REAL               one, two
      PARAMETER          ( one = 1.0e+0, two = 2.0e+0 )
*     ..
*     .. External Functions ..
      REAL               pb_sran
      EXTERNAL           pb_sran
*     ..
*     .. Executable Statements ..
*
      pb_srand = one - two * pb_sran( idumm )
*
      RETURN
*
*     End of PB_SRAND
*
      END
      REAL               function pb_sran( idumm )
*
*  -- PBLAS test routine (version 2.0) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     April 1, 1998
*
*     .. Scalar Arguments ..
      INTEGER            idumm
*     ..
*
*  Purpose
*  =======
*
*  PB_SRAN generates the next number in the random sequence.
*
*  Arguments
*  =========
*
*  IDUMM   (local input) INTEGER
*          This argument is ignored, but necessary to a FORTRAN 77 func-
*          tion.
*
*  Further Details
*  ===============
*
*  On entry, the array IRAND stored in the common block  RANCOM contains
*  the information (2 integers)  required to generate the next number in
*  the sequence X( n ). This number is computed as
*
*     X( n ) = ( 2^16 * IRAND( 2 ) + IRAND( 1 ) ) / d,
*
*  where the constant d is the  largest  32 bit  positive  integer.  The
*  array  IRAND  is  then  updated for the generation of the next number
*  X( n+1 ) in the random sequence as follows X( n+1 ) = a * X( n ) + c.
*  The constants  a  and c  should have been preliminarily stored in the
*  array  IACS  as  2 pairs of integers. The initial set up of IRAND and
*  IACS is performed by the routine PB_SETRAN.
*
*  -- Written on April 1, 1998 by
*     Antoine Petitet, University  of  Tennessee, Knoxville 37996, USA.
*
*  =====================================================================
*
*     .. Parameters ..
      REAL               divfac, pow16
      PARAMETER          ( divfac = 2.147483648e+9,
     $                   pow16 = 6.5536e+4 )
*     ..
*     .. Local Arrays ..
      INTEGER            j( 2 )
*     ..
*     .. External Subroutines ..
      EXTERNAL           pb_ladd, pb_lmul
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          real
*     ..
*     .. Common Blocks ..
      INTEGER            iacs( 4 ), irand( 2 )
      common             /rancom/ irand, iacs
*     ..
*     .. Save Statements ..
      SAVE               /rancom/
*     ..
*     .. Executable Statements ..
*
      pb_sran = ( real( irand( 1 ) ) + pow16 * real( irand( 2 ) ) ) /
     $            divfac
*
      CALL pb_lmul( irand, iacs, j )
      CALL pb_ladd( j, iacs( 3 ), irand )
*
      RETURN
*
*     End of PB_SRAN
*
      END