pchrddriver.f

Go to the documentation of this file.

      PROGRAM pchrddriver
*
*  -- ScaLAPACK testing driver (version 1.7) --
*     University of Tennessee, Knoxville, Oak Ridge National Laboratory,
*     and University of California, Berkeley.
*     March 13, 2000
*
*  Purpose
*  =======
*
*  PCHRDDRIVER is the main test program for the COMPLEX
*  ScaLAPACK HRD (Hessenberg Reduction) routines.
*
*  The program must be driven by a short data file.  An annotated
*  example of a data file can be obtained by deleting the first 3
*  characters from the following 14 lines:
*  'ScaLAPACK HRD input file'
*  'PVM machine'
*  'HRD.out'            output file name (if any)
*  6                    device out
*  2                    number of problems sizes
*  100 101              values of N
*  2   1                values of ILO
*  99  101              values of IHI
*  3                    number of NB's
*  2 3 5                values of NB
*  7                    number of process grids (ordered pairs of P & Q)
*  1 2 1 4 2 3 8        values of P
*  1 2 4 1 3 2 1        values of Q
*  3.0                  threshold
*
*  Internal Parameters
*  ===================
*
*  TOTMEM   INTEGER, default = 2000000
*           TOTMEM is a machine-specific parameter indicating the
*           maximum amount of available memory in bytes.
*           The user should customize TOTMEM to his platform.  Remember
*           to leave room in memory for the operating system, the BLACS
*           buffer, etc.  For example, on a system with 8 MB of memory
*           per process (e.g., one processor on an Intel iPSC/860), the
*           parameters we use are TOTMEM=6200000 (leaving 1.8 MB for OS,
*           code, BLACS buffer, etc).  However, for PVM, we usually set
*           TOTMEM = 2000000.  Some experimenting with the maximum value
*           of TOTMEM may be required.
*
*  INTGSZ   INTEGER, default = 4 bytes.
*  CPLXSZ   INTEGER, default = 8 bytes.
*           INTGSZ and CPLXSZ indicate the length in bytes on the
*           given platform for an integer and a single precision
*           complex.
*  MEM      COMPLEX array, dimension ( TOTMEM / CPLXSZ )
*
*           All arrays used by SCALAPACK routines are allocated from
*           this array and referenced by pointers.  The integer IPA,
*           for example, is a pointer to the starting element of MEM for
*           the matrix A.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            block_cyclic_2d, CSRC_, ctxt_, dlen_, dtype_,
     $                   lld_, mb_, m_, nb_, n_, rsrc_
      parameter( block_cyclic_2d = 1, dlen_ = 9, dtype_ = 1,
     $                     ctxt_ = 2, m_ = 3, n_ = 4, mb_ = 5, nb_ = 6,
     $                     rsrc_ = 7, csrc_ = 8, lld_ = 9 )
      INTEGER            cplxsz, memsiz, ntests, totmem
      COMPLEX            padval
      parameter( cplxsz = 8, totmem = 2000000,
     $                     memsiz = totmem / cplxsz, ntests = 20,
     $                     padval = ( -9923.0e+0, -9923.0e+0 ) )
*     ..
*     .. Local Scalars ..
      LOGICAL            check
      CHARACTER*6        passed
      CHARACTER*80       outfile
      INTEGER            i, iam, iaseed, ictxt, ihi, ihip, ihlp, ihlq,
     $                   ilcol, ilo, ilrow, info, inlq, imidpad, IPA,
     $                   ipt, ipw, ipostpad, iprepad, itemp, j, k,
     $                   kfail, kpass, kskip, ktests, lcm, lcmq, loff,
     $                   lwork, mycol, myrow, n, nb, ngrids, nmat, nnb,
     $                   nprocs, nout, np, npcol, nprow, nq, workhrd,
     $                   worksiz
      REAL               anorm, fresid, thresh
      DOUBLE PRECISION   nops, tmflops
*     ..
*     .. Local Arrays ..
      INTEGER            desca( dlen_ ), ierr( 1 ), nbval( ntests ),
     $                   nval( ntests ), nvhi( ntests ), nvlo( ntests ),
     $                   pval( ntests ), qval( ntests )
      DOUBLE PRECISION   ctime( 1 ), wtime( 1 )
      COMPLEX            mem( memsiz )
*     ..
*     .. External Subroutines ..
      EXTERNAL           blacs_barrier, blacs_exit, blacs_get,
     $                   blacs_gridexit, blacs_gridinit, blacs_gridinfo,
     $                   descinit, igsum2d, blacs_pinfo, pcfillpad,
     $                   pclafchk, pcgehdrv, pcgehrd,
     $                   pchrdinfo, pcmatgen, slboot,
     $                   slcombine, sltimer
*     ..
*     .. External Functions ..
      INTEGER            ilcm, indxg2p, numroc
      REAL               pclange
      EXTERNAL           ilcm, indxg2p, numroc, pclange
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          dble, max
*     ..
*     .. Data statements ..
      DATA               ktests, kpass, kfail, kskip / 4*0 /
*     ..
*     .. Executable Statements ..
*
*     Get starting information
*
      CALL blacs_pinfo( iam, nprocs )
      iaseed = 100
      CALL pchrdinfo( outfile, nout, nmat, nval, nvlo, nvhi, ntests,
     $                nnb, nbval, ntests, ngrids, pval, ntests, qval,
     $                ntests, thresh, mem, iam, nprocs )
      check = ( thresh.GE.0.0e+0 )
*
*     Print headings
*
      IF( iam.EQ.0 ) THEN
         WRITE( nout, fmt = * )
         WRITE( nout, fmt = 9995 )
         WRITE( nout, fmt = 9994 )
         WRITE( nout, fmt = * )
      END IF
*
*     Loop over different process grids
*
      DO 30 i = 1, ngrids
*
         nprow = pval( i )
         npcol = qval( i )
*
*        Make sure grid information is correct
*
         ierr( 1 ) = 0
         IF( nprow.LT.1 ) THEN
            IF( iam.EQ.0 )
     $         WRITE( nout, fmt = 9999 ) 'GRID', 'nprow', nprow
            ierr( 1 ) = 1
         ELSE IF( npcol.LT.1 ) THEN
            IF( iam.EQ.0 )
     $         WRITE( nout, fmt = 9999 ) 'GRID', 'npcol', npcol
            ierr( 1 ) = 1
         ELSE IF( nprow*npcol.GT.nprocs ) THEN
            IF( iam.EQ.0 )
     $         WRITE( nout, fmt = 9998 )nprow*npcol, nprocs
            ierr( 1 ) = 1
         END IF
*
         IF( ierr( 1 ).GT.0 ) THEN
            IF( iam.EQ.0 )
     $         WRITE( nout, fmt = 9997 ) 'grid'
            kskip = kskip + 1
            GO TO 30
         END IF
*
*        Define process grid
*
         CALL blacs_get( -1, 0, ictxt )
         CALL blacs_gridinit( ictxt, 'Row-major', nprow, npcol )
         CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
*
*        Go to bottom of loop if this case doesn't use my process
*
         IF( myrow.GE.nprow .OR. mycol.GE.npcol )
     $      GOTO 30
*
         DO 20 j = 1, nmat
*
            n = nval( j )
            ilo = nvlo( j )
            ihi = nvhi( j )
*
*           Make sure matrix information is correct
*
            ierr( 1 ) = 0
            IF( n.LT.1 ) THEN
               IF( iam.EQ.0 )
     $            WRITE( nout, fmt = 9999 ) 'MATRIX', 'N', n
               ierr( 1 ) = 1
            END IF
*
*           Check all processes for an error
*
            CALL igsum2d( ictxt, 'All', ' ', 1, 1, ierr, 1, -1, 0 )
*
            IF( ierr( 1 ).GT.0 ) THEN
               IF( iam.EQ.0 )
     $            WRITE( nout, fmt = 9997 ) 'matrix'
               kskip = kskip + 1
               GO TO 20
            END IF
*
            DO 10 k = 1, nnb
               nb = nbval( k )
*
*              Make sure nb is legal
*
               ierr( 1 ) = 0
               IF( nb.LT.1 ) THEN
                  ierr( 1 ) = 1
                  IF( iam.EQ.0 )
     $               WRITE( nout, fmt = 9999 ) 'NB', 'NB', nb
               END IF
*
*              Check all processes for an error
*
               CALL igsum2d( ictxt, 'All', ' ', 1, 1, ierr, 1, -1, 0 )
*
               IF( ierr( 1 ).GT.0 ) THEN
                  IF( iam.EQ.0 )
     $               WRITE( nout, fmt = 9997 ) 'NB'
                  kskip = kskip + 1
                  GO TO 10
               END IF
*
               np = numroc( n, nb, myrow, 0, nprow )
               nq = numroc( n, nb, mycol, 0, npcol )
               IF( check ) THEN
                  iprepad  = max( nb, np )
                  imidpad  = nb
                  ipostpad = max( nb, nq )
               ELSE
                  iprepad  = 0
                  imidpad  = 0
                  ipostpad = 0
               END IF
*
*              Initialize the array descriptor for the matrix A
*
               CALL descinit( desca, n, n, nb, nb, 0, 0, ictxt,
     $                        max( 1, np ) + imidpad, info )
*
*              Check all processes for an error
*
               CALL igsum2d( ictxt, 'All', ' ', 1, 1, ierr, 1, -1, 0 )
*
               IF( ierr( 1 ).LT.0 ) THEN
                  IF( iam.EQ.0 )
     $               WRITE( nout, fmt = 9997 ) 'descriptor'
                  kskip = kskip + 1
                  GO TO 10
               END IF
*
*              Assign pointers into MEM for SCALAPACK arrays, A is
*              allocated starting at position MEM( IPREPAD+1 )
*
               ipa = iprepad + 1
               ipt = ipa + desca( lld_ )*nq + ipostpad + iprepad
               ipw = ipt + nq + ipostpad + iprepad
*
*              Calculate the amount of workspace required for the
*              reduction
*
               ihip = numroc( ihi, nb, myrow, desca( rsrc_ ), nprow )
               loff = mod( ilo-1, nb )
               ilrow = indxg2p( ilo, nb, myrow, desca( rsrc_ ), nprow )
               ilcol = indxg2p( ilo, nb, mycol, desca( csrc_ ), npcol )
               ihlp = numroc( ihi-ilo+loff+1, nb, myrow, ilrow, nprow )
               inlq = numroc( n-ilo+loff+1, nb, mycol, ilcol, npcol )
               lwork = nb*( nb + max( ihip+1, ihlp+inlq ) )
               workhrd = lwork + ipostpad
               worksiz = workhrd
*
*              Figure the amount of workspace required by the check
*
               IF( check ) THEN
                  lcm = ilcm( nprow, npcol )
                  lcmq = lcm / npcol
                  ihlq = numroc( ihi-ilo+loff+1, nb, mycol, ilcol,
     $                           npcol )
                  itemp = nb*max( ihlp+inlq, ihlq+max( ihip,
     $                    ihlp+numroc( numroc( ihi-ilo+loff+1, nb, 0, 0,
     $                    npcol ), nb, 0, 0, lcmq ) ) )
                  worksiz = max( nb*nb + nb*ihlp + itemp, nb * np ) +
     $                           ipostpad
               END IF
*
*              Check for adequate memory for problem size
*
               ierr( 1 ) = 0
               IF( ipw+worksiz.GT.memsiz ) THEN
                  IF( iam.EQ.0 )
     $               WRITE( nout, fmt = 9996 ) 'Hessenberg reduction',
     $                      ( ipw+worksiz )*cplxsz
                  ierr( 1 ) = 1
               END IF
*
*              Check all processes for an error
*
               CALL igsum2d( ictxt, 'All', ' ', 1, 1, ierr, 1, -1, 0 )
*
               IF( ierr( 1 ).GT.0 ) THEN
                  IF( iam.EQ.0 )
     $               WRITE( nout, fmt = 9997 ) 'MEMORY'
                  kskip = kskip + 1
                  GO TO 10
               END IF
*
*              Generate A
*
               CALL pcmatgen( ictxt, 'No', 'No', desca( m_ ),
     $                        desca( n_ ), desca( mb_ ), desca( nb_ ),
     $                        mem( ipa ), desca( lld_ ), desca( rsrc_ ),
     $                        desca( csrc_ ),
     $                        iaseed, 0, np, 0, nq, myrow, mycol,
     $                        nprow, npcol )
*
*              Need Infinity-norm of A for checking
*
               IF( check ) THEN
                  CALL pcfillpad( ictxt, np, nq, mem( ipa-iprepad ),
     $                            desca( lld_ ), iprepad, ipostpad,
     $                            padval )
                  CALL pcfillpad( ictxt, nq, 1, mem( ipt-iprepad ),
     $                            nq, iprepad, ipostpad, padval )
                  CALL pcfillpad( ictxt, worksiz-ipostpad, 1,
     $                            mem( ipw-iprepad ), worksiz-ipostpad,
     $                            iprepad, ipostpad, padval )
                  anorm = pclange( 'I', n, n, mem( ipa ), 1, 1, desca,
     $                             mem( ipw ) )
                  CALL pcchekpad( ictxt, 'PCLANGE', np, nq,
     $                            mem( ipa-iprepad ), desca( lld_ ),
     $                            iprepad, ipostpad, padval )
                  CALL pcchekpad( ictxt, 'PCLANGE',
     $                            worksiz-ipostpad, 1,
     $                            mem( ipw-iprepad ), worksiz-ipostpad,
     $                            iprepad, ipostpad, padval )
                  CALL pcfillpad( ictxt, workhrd-ipostpad, 1,
     $                            mem( ipw-iprepad ), workhrd-ipostpad,
     $                            iprepad, ipostpad, padval )
               END IF
*
               CALL slboot()
               CALL blacs_barrier( ictxt, 'All' )
               CALL sltimer( 1 )
*
*              Reduce Hessenberg form
*
               CALL pcgehrd( n, ilo, ihi, mem( ipa ), 1, 1, desca,
     $                       mem( ipt ), mem( ipw ), lwork, info )
               CALL sltimer( 1 )
*
               IF( check ) THEN
*
*                 Check for memory overwrite
*
                  CALL pcchekpad( ictxt, 'PCGEHRD', np, nq,
     $                            mem( ipa-iprepad ), desca( lld_ ),
     $                            iprepad, ipostpad, padval )
                  CALL pcchekpad( ictxt, 'PCGEHRD', nq, 1,
     $                            mem( ipt-iprepad ), nq, iprepad,
     $                            ipostpad, padval )
                  CALL pcchekpad( ictxt, 'PCGEHRD', workhrd-ipostpad,
     $                            1, mem( ipw-iprepad ),
     $                            workhrd-ipostpad, iprepad,
     $                            ipostpad, padval )
                  CALL pcfillpad( ictxt, worksiz-ipostpad, 1,
     $                            mem( ipw-iprepad ), worksiz-ipostpad,
     $                            iprepad, ipostpad, padval )
*
*                 Compute fctres = ||A - Q H Q'|| / (||A||*N*eps)
*
                  CALL pcgehdrv( n, ilo, ihi, mem( ipa ), 1, 1, desca,
     $                           mem( ipt ), mem( ipw ) )
                  CALL pclafchk( 'No', 'No', n, n, mem( ipa ), 1, 1,
     $                           desca, iaseed, anorm, fresid,
     $                           mem( ipw ) )
*
*                 Check for memory overwrite
*
                  CALL pcchekpad( ictxt, 'pcgehdrv', NP, NQ,
     $                            MEM( IPA-IPREPAD ), DESCA( LLD_ ),
     $                            IPREPAD, IPOSTPAD, PADVAL )
                  CALL PCCHEKPAD( ICTXT, 'pcgehdrv', NQ, 1,
     $                            MEM( IPT-IPREPAD ), NQ, IPREPAD,
     $                            IPOSTPAD, PADVAL )
                  CALL PCCHEKPAD( ICTXT, 'pcgehdrv',
     $                            WORKSIZ-IPOSTPAD, 1,
     $                            MEM( IPW-IPREPAD ), WORKSIZ-IPOSTPAD,
     $                            IPREPAD, IPOSTPAD, PADVAL )
*
*                 Test residual and detect NaN result
*
.LE..AND..EQ.                  IF( FRESIDTHRESH  FRESID-FRESID0.0E+0 )
     $                THEN
                     KPASS = KPASS + 1
                     PASSED = 'passed'
                  ELSE
.EQ..AND..EQ.                     IF( MYROW0  MYCOL0 )
     $                  WRITE( NOUT, FMT = 9986 ) FRESID
                     KFAIL = KFAIL + 1
                     PASSED = 'failed'
                  END IF
               ELSE
*
*                 Don't perform the checking, only the timing operation
*
                  KPASS = KPASS + 1
                  FRESID = FRESID - FRESID
                  PASSED = 'bypass'
               END IF
*
*              Gather max. of all CPU and WALL clock timings
*
               CALL SLCOMBINE( ICTXT, 'all', '>', 'w', 1, 1, WTIME )
               CALL SLCOMBINE( ICTXT, 'all', '>', 'c', 1, 1, CTIME )
*
*              Print results
*
.EQ..AND..EQ.               IF( MYROW0  MYCOL0 ) THEN
*
*                 HRD requires 40/3 * N^3 floating point ops. (flops)
*                 more precisely,
*                 HRD requires 16/3*(IHI-ILO)^3+8*IHI*(IHI-ILO)^2 flops
*
                  NOPS = DBLE( IHI-ILO )
                  NOPS = NOPS * NOPS *
     $                   ( 8.0D0*DBLE( IHI ) + (16.0D0/3.0D0)*NOPS )
                  NOPS = NOPS / 1.0D+6
*
*                 Print WALL time
*
.GT.                  IF( WTIME( 1 )0.0D+0 ) THEN
                     TMFLOPS = NOPS / WTIME( 1 )
                  ELSE
                     TMFLOPS = 0.0D+0
                  END IF
.GE.                  IF( WTIME( 1 )0.0D+0 )
     $               WRITE( NOUT, FMT = 9993 ) 'wall',  N, ILO, IHI, NB,
     $                      NPROW, NPCOL, WTIME( 1 ), TMFLOPS, FRESID,
     $                      PASSED
*
*                 Print CPU time
*
.GT.                  IF( CTIME( 1 )0.0D+0 ) THEN
                     TMFLOPS = NOPS / CTIME( 1 )
                  ELSE
                     TMFLOPS = 0.0D+0
                  END IF
.GE.                  IF( CTIME( 1 )0.0D+0 )
     $               WRITE( NOUT, FMT = 9993 ) 'cpu ', N, ILO, IHI, NB,
     $                      NPROW, NPCOL, CTIME( 1 ), TMFLOPS, FRESID,
     $                      PASSED
               END IF
   10       CONTINUE
   20    CONTINUE
*
         CALL BLACS_GRIDEXIT( ICTXT )
   30 CONTINUE
*
*     Print ending messages and close output file
*
.EQ.      IF( IAM0 ) THEN
         KTESTS = KPASS + KFAIL + KSKIP
         WRITE( NOUT, FMT = * )
         WRITE( NOUT, FMT = 9992 ) KTESTS
         IF( CHECK ) THEN
            WRITE( NOUT, FMT = 9991 ) KPASS
            WRITE( NOUT, FMT = 9989 ) KFAIL
         ELSE
            WRITE( NOUT, FMT = 9990 ) KPASS
         END IF
         WRITE( NOUT, FMT = 9988 ) KSKIP
         WRITE( NOUT, FMT = * )
         WRITE( NOUT, FMT = * )
         WRITE( NOUT, FMT = 9987 )
.NE..AND..NE.         IF( NOUT6  NOUT0 )
     $      CLOSE( NOUT )
      END IF
*
      CALL BLACS_EXIT( 0 )
*
 9999 FORMAT( 'illegal ', A6, ': ', A5, ' = ', I3,
     $      '; it should be at least 1' )
 9998 FORMAT( 'illegal grid: nprow*npcol = ', I4, '. it can be at most',
     $      I4 )
 9997 FORMAT( 'bad ', A6, ' parameters: going on to next test case.' )
 9996 FORMAT( 'unable to perform ', A, ': need totmem of at least',
     $      I11 )
 9995 FORMAT( 'time      n    ilo    ihi  nb     p     q  hrd time ',
     $        '     mflops residual  check' )
 9994 FORMAT( '---- ------ ------ ------ --- ----- ----- --------- ',
     $        '----------- -------- ------' )
 9993 FORMAT( A4, 1X, I6, 1X, I6, 1X, I6, 1X, I3, 1X, I5, 1X, I5, 1X,
     $        F9.2, 1X, F11.2, 1X, F8.2, 1X, A6 )
 9992 FORMAT( 'finished', I4, ' tests, with the following results:' )
 9991 FORMAT( I5, ' tests completed and passed residual checks.' )
 9990 FORMAT( I5, ' tests completed without checking.' )
 9989 FORMAT( I5, ' tests completed and failed residual checks.' )
 9988 FORMAT( I5, ' tests skipped because of illegal input values.' )
 9987 FORMAT( 'END OF TESTS.' )
 9986 FORMAT( '||A - Q*H*Q''|| / (||A|| * N * eps) = ', G25.7 )
*
      STOP
*
*     End of PCHRDDRIVER
*
      END