LAPACK 3.3.1
Linear Algebra PACKage

dchkpt.f

Go to the documentation of this file.
00001       SUBROUTINE DCHKPT( DOTYPE, NN, NVAL, NNS, NSVAL, THRESH, TSTERR,
00002      $                   A, D, E, B, X, XACT, WORK, RWORK, NOUT )
00003 *
00004 *  -- LAPACK test routine (version 3.1) --
00005 *     Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd..
00006 *     November 2006
00007 *
00008 *     .. Scalar Arguments ..
00009       LOGICAL            TSTERR
00010       INTEGER            NN, NNS, NOUT
00011       DOUBLE PRECISION   THRESH
00012 *     ..
00013 *     .. Array Arguments ..
00014       LOGICAL            DOTYPE( * )
00015       INTEGER            NSVAL( * ), NVAL( * )
00016       DOUBLE PRECISION   A( * ), B( * ), D( * ), E( * ), RWORK( * ),
00017      $                   WORK( * ), X( * ), XACT( * )
00018 *     ..
00019 *
00020 *  Purpose
00021 *  =======
00022 *
00023 *  DCHKPT tests DPTTRF, -TRS, -RFS, and -CON
00024 *
00025 *  Arguments
00026 *  =========
00027 *
00028 *  DOTYPE  (input) LOGICAL array, dimension (NTYPES)
00029 *          The matrix types to be used for testing.  Matrices of type j
00030 *          (for 1 <= j <= NTYPES) are used for testing if DOTYPE(j) =
00031 *          .TRUE.; if DOTYPE(j) = .FALSE., then type j is not used.
00032 *
00033 *  NN      (input) INTEGER
00034 *          The number of values of N contained in the vector NVAL.
00035 *
00036 *  NVAL    (input) INTEGER array, dimension (NN)
00037 *          The values of the matrix dimension N.
00038 *
00039 *  NNS     (input) INTEGER
00040 *          The number of values of NRHS contained in the vector NSVAL.
00041 *
00042 *  NSVAL   (input) INTEGER array, dimension (NNS)
00043 *          The values of the number of right hand sides NRHS.
00044 *
00045 *  THRESH  (input) DOUBLE PRECISION
00046 *          The threshold value for the test ratios.  A result is
00047 *          included in the output file if RESULT >= THRESH.  To have
00048 *          every test ratio printed, use THRESH = 0.
00049 *
00050 *  TSTERR  (input) LOGICAL
00051 *          Flag that indicates whether error exits are to be tested.
00052 *
00053 *  A       (workspace) DOUBLE PRECISION array, dimension (NMAX*2)
00054 *
00055 *  D       (workspace) DOUBLE PRECISION array, dimension (NMAX*2)
00056 *
00057 *  E       (workspace) DOUBLE PRECISION array, dimension (NMAX*2)
00058 *
00059 *  B       (workspace) DOUBLE PRECISION array, dimension (NMAX*NSMAX)
00060 *          where NSMAX is the largest entry in NSVAL.
00061 *
00062 *  X       (workspace) DOUBLE PRECISION array, dimension (NMAX*NSMAX)
00063 *
00064 *  XACT    (workspace) DOUBLE PRECISION array, dimension (NMAX*NSMAX)
00065 *
00066 *  WORK    (workspace) DOUBLE PRECISION array, dimension
00067 *                      (NMAX*max(3,NSMAX))
00068 *
00069 *  RWORK   (workspace) DOUBLE PRECISION array, dimension
00070 *                      (max(NMAX,2*NSMAX))
00071 *
00072 *  NOUT    (input) INTEGER
00073 *          The unit number for output.
00074 *
00075 *  =====================================================================
00076 *
00077 *     .. Parameters ..
00078       DOUBLE PRECISION   ONE, ZERO
00079       PARAMETER          ( ONE = 1.0D+0, ZERO = 0.0D+0 )
00080       INTEGER            NTYPES
00081       PARAMETER          ( NTYPES = 12 )
00082       INTEGER            NTESTS
00083       PARAMETER          ( NTESTS = 7 )
00084 *     ..
00085 *     .. Local Scalars ..
00086       LOGICAL            ZEROT
00087       CHARACTER          DIST, TYPE
00088       CHARACTER*3        PATH
00089       INTEGER            I, IA, IMAT, IN, INFO, IRHS, IX, IZERO, J, K,
00090      $                   KL, KU, LDA, MODE, N, NERRS, NFAIL, NIMAT,
00091      $                   NRHS, NRUN
00092       DOUBLE PRECISION   AINVNM, ANORM, COND, DMAX, RCOND, RCONDC
00093 *     ..
00094 *     .. Local Arrays ..
00095       INTEGER            ISEED( 4 ), ISEEDY( 4 )
00096       DOUBLE PRECISION   RESULT( NTESTS ), Z( 3 )
00097 *     ..
00098 *     .. External Functions ..
00099       INTEGER            IDAMAX
00100       DOUBLE PRECISION   DASUM, DGET06, DLANST
00101       EXTERNAL           IDAMAX, DASUM, DGET06, DLANST
00102 *     ..
00103 *     .. External Subroutines ..
00104       EXTERNAL           ALAERH, ALAHD, ALASUM, DCOPY, DERRGT, DGET04,
00105      $                   DLACPY, DLAPTM, DLARNV, DLATB4, DLATMS, DPTCON,
00106      $                   DPTRFS, DPTT01, DPTT02, DPTT05, DPTTRF, DPTTRS,
00107      $                   DSCAL
00108 *     ..
00109 *     .. Intrinsic Functions ..
00110       INTRINSIC          ABS, MAX
00111 *     ..
00112 *     .. Scalars in Common ..
00113       LOGICAL            LERR, OK
00114       CHARACTER*32       SRNAMT
00115       INTEGER            INFOT, NUNIT
00116 *     ..
00117 *     .. Common blocks ..
00118       COMMON             / INFOC / INFOT, NUNIT, OK, LERR
00119       COMMON             / SRNAMC / SRNAMT
00120 *     ..
00121 *     .. Data statements ..
00122       DATA               ISEEDY / 0, 0, 0, 1 /
00123 *     ..
00124 *     .. Executable Statements ..
00125 *
00126       PATH( 1: 1 ) = 'Double precision'
00127       PATH( 2: 3 ) = 'PT'
00128       NRUN = 0
00129       NFAIL = 0
00130       NERRS = 0
00131       DO 10 I = 1, 4
00132          ISEED( I ) = ISEEDY( I )
00133    10 CONTINUE
00134 *
00135 *     Test the error exits
00136 *
00137       IF( TSTERR )
00138      $   CALL DERRGT( PATH, NOUT )
00139       INFOT = 0
00140 *
00141       DO 110 IN = 1, NN
00142 *
00143 *        Do for each value of N in NVAL.
00144 *
00145          N = NVAL( IN )
00146          LDA = MAX( 1, N )
00147          NIMAT = NTYPES
00148          IF( N.LE.0 )
00149      $      NIMAT = 1
00150 *
00151          DO 100 IMAT = 1, NIMAT
00152 *
00153 *           Do the tests only if DOTYPE( IMAT ) is true.
00154 *
00155             IF( N.GT.0 .AND. .NOT.DOTYPE( IMAT ) )
00156      $         GO TO 100
00157 *
00158 *           Set up parameters with DLATB4.
00159 *
00160             CALL DLATB4( PATH, IMAT, N, N, TYPE, KL, KU, ANORM, MODE,
00161      $                   COND, DIST )
00162 *
00163             ZEROT = IMAT.GE.8 .AND. IMAT.LE.10
00164             IF( IMAT.LE.6 ) THEN
00165 *
00166 *              Type 1-6:  generate a symmetric tridiagonal matrix of
00167 *              known condition number in lower triangular band storage.
00168 *
00169                SRNAMT = 'DLATMS'
00170                CALL DLATMS( N, N, DIST, ISEED, TYPE, RWORK, MODE, COND,
00171      $                      ANORM, KL, KU, 'B', A, 2, WORK, INFO )
00172 *
00173 *              Check the error code from DLATMS.
00174 *
00175                IF( INFO.NE.0 ) THEN
00176                   CALL ALAERH( PATH, 'DLATMS', INFO, 0, ' ', N, N, KL,
00177      $                         KU, -1, IMAT, NFAIL, NERRS, NOUT )
00178                   GO TO 100
00179                END IF
00180                IZERO = 0
00181 *
00182 *              Copy the matrix to D and E.
00183 *
00184                IA = 1
00185                DO 20 I = 1, N - 1
00186                   D( I ) = A( IA )
00187                   E( I ) = A( IA+1 )
00188                   IA = IA + 2
00189    20          CONTINUE
00190                IF( N.GT.0 )
00191      $            D( N ) = A( IA )
00192             ELSE
00193 *
00194 *              Type 7-12:  generate a diagonally dominant matrix with
00195 *              unknown condition number in the vectors D and E.
00196 *
00197                IF( .NOT.ZEROT .OR. .NOT.DOTYPE( 7 ) ) THEN
00198 *
00199 *                 Let D and E have values from [-1,1].
00200 *
00201                   CALL DLARNV( 2, ISEED, N, D )
00202                   CALL DLARNV( 2, ISEED, N-1, E )
00203 *
00204 *                 Make the tridiagonal matrix diagonally dominant.
00205 *
00206                   IF( N.EQ.1 ) THEN
00207                      D( 1 ) = ABS( D( 1 ) )
00208                   ELSE
00209                      D( 1 ) = ABS( D( 1 ) ) + ABS( E( 1 ) )
00210                      D( N ) = ABS( D( N ) ) + ABS( E( N-1 ) )
00211                      DO 30 I = 2, N - 1
00212                         D( I ) = ABS( D( I ) ) + ABS( E( I ) ) +
00213      $                           ABS( E( I-1 ) )
00214    30                CONTINUE
00215                   END IF
00216 *
00217 *                 Scale D and E so the maximum element is ANORM.
00218 *
00219                   IX = IDAMAX( N, D, 1 )
00220                   DMAX = D( IX )
00221                   CALL DSCAL( N, ANORM / DMAX, D, 1 )
00222                   CALL DSCAL( N-1, ANORM / DMAX, E, 1 )
00223 *
00224                ELSE IF( IZERO.GT.0 ) THEN
00225 *
00226 *                 Reuse the last matrix by copying back the zeroed out
00227 *                 elements.
00228 *
00229                   IF( IZERO.EQ.1 ) THEN
00230                      D( 1 ) = Z( 2 )
00231                      IF( N.GT.1 )
00232      $                  E( 1 ) = Z( 3 )
00233                   ELSE IF( IZERO.EQ.N ) THEN
00234                      E( N-1 ) = Z( 1 )
00235                      D( N ) = Z( 2 )
00236                   ELSE
00237                      E( IZERO-1 ) = Z( 1 )
00238                      D( IZERO ) = Z( 2 )
00239                      E( IZERO ) = Z( 3 )
00240                   END IF
00241                END IF
00242 *
00243 *              For types 8-10, set one row and column of the matrix to
00244 *              zero.
00245 *
00246                IZERO = 0
00247                IF( IMAT.EQ.8 ) THEN
00248                   IZERO = 1
00249                   Z( 2 ) = D( 1 )
00250                   D( 1 ) = ZERO
00251                   IF( N.GT.1 ) THEN
00252                      Z( 3 ) = E( 1 )
00253                      E( 1 ) = ZERO
00254                   END IF
00255                ELSE IF( IMAT.EQ.9 ) THEN
00256                   IZERO = N
00257                   IF( N.GT.1 ) THEN
00258                      Z( 1 ) = E( N-1 )
00259                      E( N-1 ) = ZERO
00260                   END IF
00261                   Z( 2 ) = D( N )
00262                   D( N ) = ZERO
00263                ELSE IF( IMAT.EQ.10 ) THEN
00264                   IZERO = ( N+1 ) / 2
00265                   IF( IZERO.GT.1 ) THEN
00266                      Z( 1 ) = E( IZERO-1 )
00267                      E( IZERO-1 ) = ZERO
00268                      Z( 3 ) = E( IZERO )
00269                      E( IZERO ) = ZERO
00270                   END IF
00271                   Z( 2 ) = D( IZERO )
00272                   D( IZERO ) = ZERO
00273                END IF
00274             END IF
00275 *
00276             CALL DCOPY( N, D, 1, D( N+1 ), 1 )
00277             IF( N.GT.1 )
00278      $         CALL DCOPY( N-1, E, 1, E( N+1 ), 1 )
00279 *
00280 *+    TEST 1
00281 *           Factor A as L*D*L' and compute the ratio
00282 *              norm(L*D*L' - A) / (n * norm(A) * EPS )
00283 *
00284             CALL DPTTRF( N, D( N+1 ), E( N+1 ), INFO )
00285 *
00286 *           Check error code from DPTTRF.
00287 *
00288             IF( INFO.NE.IZERO ) THEN
00289                CALL ALAERH( PATH, 'DPTTRF', INFO, IZERO, ' ', N, N, -1,
00290      $                      -1, -1, IMAT, NFAIL, NERRS, NOUT )
00291                GO TO 100
00292             END IF
00293 *
00294             IF( INFO.GT.0 ) THEN
00295                RCONDC = ZERO
00296                GO TO 90
00297             END IF
00298 *
00299             CALL DPTT01( N, D, E, D( N+1 ), E( N+1 ), WORK,
00300      $                   RESULT( 1 ) )
00301 *
00302 *           Print the test ratio if greater than or equal to THRESH.
00303 *
00304             IF( RESULT( 1 ).GE.THRESH ) THEN
00305                IF( NFAIL.EQ.0 .AND. NERRS.EQ.0 )
00306      $            CALL ALAHD( NOUT, PATH )
00307                WRITE( NOUT, FMT = 9999 )N, IMAT, 1, RESULT( 1 )
00308                NFAIL = NFAIL + 1
00309             END IF
00310             NRUN = NRUN + 1
00311 *
00312 *           Compute RCONDC = 1 / (norm(A) * norm(inv(A))
00313 *
00314 *           Compute norm(A).
00315 *
00316             ANORM = DLANST( '1', N, D, E )
00317 *
00318 *           Use DPTTRS to solve for one column at a time of inv(A),
00319 *           computing the maximum column sum as we go.
00320 *
00321             AINVNM = ZERO
00322             DO 50 I = 1, N
00323                DO 40 J = 1, N
00324                   X( J ) = ZERO
00325    40          CONTINUE
00326                X( I ) = ONE
00327                CALL DPTTRS( N, 1, D( N+1 ), E( N+1 ), X, LDA, INFO )
00328                AINVNM = MAX( AINVNM, DASUM( N, X, 1 ) )
00329    50       CONTINUE
00330             RCONDC = ONE / MAX( ONE, ANORM*AINVNM )
00331 *
00332             DO 80 IRHS = 1, NNS
00333                NRHS = NSVAL( IRHS )
00334 *
00335 *           Generate NRHS random solution vectors.
00336 *
00337                IX = 1
00338                DO 60 J = 1, NRHS
00339                   CALL DLARNV( 2, ISEED, N, XACT( IX ) )
00340                   IX = IX + LDA
00341    60          CONTINUE
00342 *
00343 *           Set the right hand side.
00344 *
00345                CALL DLAPTM( N, NRHS, ONE, D, E, XACT, LDA, ZERO, B,
00346      $                      LDA )
00347 *
00348 *+    TEST 2
00349 *           Solve A*x = b and compute the residual.
00350 *
00351                CALL DLACPY( 'Full', N, NRHS, B, LDA, X, LDA )
00352                CALL DPTTRS( N, NRHS, D( N+1 ), E( N+1 ), X, LDA, INFO )
00353 *
00354 *           Check error code from DPTTRS.
00355 *
00356                IF( INFO.NE.0 )
00357      $            CALL ALAERH( PATH, 'DPTTRS', INFO, 0, ' ', N, N, -1,
00358      $                         -1, NRHS, IMAT, NFAIL, NERRS, NOUT )
00359 *
00360                CALL DLACPY( 'Full', N, NRHS, B, LDA, WORK, LDA )
00361                CALL DPTT02( N, NRHS, D, E, X, LDA, WORK, LDA,
00362      $                      RESULT( 2 ) )
00363 *
00364 *+    TEST 3
00365 *           Check solution from generated exact solution.
00366 *
00367                CALL DGET04( N, NRHS, X, LDA, XACT, LDA, RCONDC,
00368      $                      RESULT( 3 ) )
00369 *
00370 *+    TESTS 4, 5, and 6
00371 *           Use iterative refinement to improve the solution.
00372 *
00373                SRNAMT = 'DPTRFS'
00374                CALL DPTRFS( N, NRHS, D, E, D( N+1 ), E( N+1 ), B, LDA,
00375      $                      X, LDA, RWORK, RWORK( NRHS+1 ), WORK, INFO )
00376 *
00377 *           Check error code from DPTRFS.
00378 *
00379                IF( INFO.NE.0 )
00380      $            CALL ALAERH( PATH, 'DPTRFS', INFO, 0, ' ', N, N, -1,
00381      $                         -1, NRHS, IMAT, NFAIL, NERRS, NOUT )
00382 *
00383                CALL DGET04( N, NRHS, X, LDA, XACT, LDA, RCONDC,
00384      $                      RESULT( 4 ) )
00385                CALL DPTT05( N, NRHS, D, E, B, LDA, X, LDA, XACT, LDA,
00386      $                      RWORK, RWORK( NRHS+1 ), RESULT( 5 ) )
00387 *
00388 *           Print information about the tests that did not pass the
00389 *           threshold.
00390 *
00391                DO 70 K = 2, 6
00392                   IF( RESULT( K ).GE.THRESH ) THEN
00393                      IF( NFAIL.EQ.0 .AND. NERRS.EQ.0 )
00394      $                  CALL ALAHD( NOUT, PATH )
00395                      WRITE( NOUT, FMT = 9998 )N, NRHS, IMAT, K,
00396      $                  RESULT( K )
00397                      NFAIL = NFAIL + 1
00398                   END IF
00399    70          CONTINUE
00400                NRUN = NRUN + 5
00401    80       CONTINUE
00402 *
00403 *+    TEST 7
00404 *           Estimate the reciprocal of the condition number of the
00405 *           matrix.
00406 *
00407    90       CONTINUE
00408             SRNAMT = 'DPTCON'
00409             CALL DPTCON( N, D( N+1 ), E( N+1 ), ANORM, RCOND, RWORK,
00410      $                   INFO )
00411 *
00412 *           Check error code from DPTCON.
00413 *
00414             IF( INFO.NE.0 )
00415      $         CALL ALAERH( PATH, 'DPTCON', INFO, 0, ' ', N, N, -1, -1,
00416      $                      -1, IMAT, NFAIL, NERRS, NOUT )
00417 *
00418             RESULT( 7 ) = DGET06( RCOND, RCONDC )
00419 *
00420 *           Print the test ratio if greater than or equal to THRESH.
00421 *
00422             IF( RESULT( 7 ).GE.THRESH ) THEN
00423                IF( NFAIL.EQ.0 .AND. NERRS.EQ.0 )
00424      $            CALL ALAHD( NOUT, PATH )
00425                WRITE( NOUT, FMT = 9999 )N, IMAT, 7, RESULT( 7 )
00426                NFAIL = NFAIL + 1
00427             END IF
00428             NRUN = NRUN + 1
00429   100    CONTINUE
00430   110 CONTINUE
00431 *
00432 *     Print a summary of the results.
00433 *
00434       CALL ALASUM( PATH, NOUT, NFAIL, NRUN, NERRS )
00435 *
00436  9999 FORMAT( ' N =', I5, ', type ', I2, ', test ', I2, ', ratio = ',
00437      $      G12.5 )
00438  9998 FORMAT( ' N =', I5, ', NRHS=', I3, ', type ', I2, ', test(', I2,
00439      $      ') = ', G12.5 )
00440       RETURN
00441 *
00442 *     End of DCHKPT
00443 *
00444       END
 All Files Functions