1
       2
       3
       4
       5
       6
       7
       8
       9
      10
      11
      12
      13
      14
      15
      16
      17
      18
      19
      20
      21
      22
      23
      24
      25
      26
      27
      28
      29
      30
      31
      32
      33
      34
      35
      36
      37
      38
      39
      40
      41
      42
      43
      44
      45
      46
      47
      48
      49
      50
      51
      52
      53
      54
      55
      56
      57
      58
      59
      60
      61
      62
      63
      64
      65
      66
      67
      68
      69
      70
      71
      72
      73
      74
      75
      76
      77
      78
      79
      80
      81
      82
      83
      84
      85
      86
      87
      88
      89
      90
      91
      92
      93
      94
      95
      96
      97
      98
      99
     100
     101
     102
     103
     104
     105
     106
     107
     108
     109
     110
     111
     112
     113
     114
     115
     116
     117
     118
     119
     120
     121
     122
     123
     124
     125
     126
     127
     128
     129
     130
     131
     132
     133
     134
     135
     136
     137
     138
     139
     140
     141
     142
     143
     144
     145
     146
     147
     148
     149
     150
     151
     152
     153
     154
     155
     156
     157
     158
     159
     160
     161
     162
     163
     164
     165
     166
     167
     168
     169
     170
     171
     172
     173
     174
     175
     176
     177
     178
     179
     180
     181
     182
     183
     184
     185
     186
     187
     188
     189
     190
     191
     192
     193
     194
     195
     196
     197
     198
     199
     200
     201
     202
     203
     204
     205
     206
     207
     208
     209
     210
     211
     212
     213
     214
     215
     216
     217
     218
     219
     220
     221
     222
     223
     224
     225
     226
     227
     228
     229
     230
     231
     232
     233
     234
     235
     236
     237
     238
     239
     240
     241
     242
     243
     244
     245
     246
     247
     248
     249
     250
     251
     252
     253
     254
     255
     256
     257
     258
     259
     260
     261
     262
     263
     264
     265
     266
     267
     268
     269
     270
     271
     272
     273
     274
     275
     276
     277
     278
     279
     280
     281
     282
     283
     284
     285
     286
     287
     288
     289
     290
     291
     292
     293
     294
     295
     296
     297
     298
     299
     300
     301
     302
     303
     304
     305
     306
     307
     308
     309
     310
     311
     312
     313
     314
     315
     316
     317
     318
     319
     320
     321
     322
     323
     324
     325
     326
     327
     328
     329
      SUBROUTINE SPPRFS( UPLO, N, NRHS, AP, AFP, B, LDB, X, LDX, FERR,
     $                   BERR, WORK, IWORK, INFO )
*
*  -- LAPACK routine (version 3.2) --
*  -- LAPACK is a software package provided by Univ. of Tennessee,    --
*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
*     November 2006
*
*     Modified to call SLACN2 in place of SLACON, 7 Feb 03, SJH.
*
*     .. Scalar Arguments ..
      CHARACTER          UPLO
      INTEGER            INFO, LDB, LDX, N, NRHS
*     ..
*     .. Array Arguments ..
      INTEGER            IWORK( * )
      REAL               AFP( * ), AP( * ), B( LDB, * ), BERR( * ),
     $                   FERR( * ), WORK( * ), X( LDX, * )
*     ..
*
*  Purpose
*  =======
*
*  SPPRFS improves the computed solution to a system of linear
*  equations when the coefficient matrix is symmetric positive definite
*  and packed, and provides error bounds and backward error estimates
*  for the solution.
*
*  Arguments
*  =========
*
*  UPLO    (input) CHARACTER*1
*          = 'U':  Upper triangle of A is stored;
*          = 'L':  Lower triangle of A is stored.
*
*  N       (input) INTEGER
*          The order of the matrix A.  N >= 0.
*
*  NRHS    (input) INTEGER
*          The number of right hand sides, i.e., the number of columns
*          of the matrices B and X.  NRHS >= 0.
*
*  AP      (input) REAL array, dimension (N*(N+1)/2)
*          The upper or lower triangle of the symmetric matrix A, packed
*          columnwise in a linear array.  The j-th column of A is stored
*          in the array AP as follows:
*          if UPLO = 'U', AP(i + (j-1)*j/2) = A(i,j) for 1<=i<=j;
*          if UPLO = 'L', AP(i + (j-1)*(2n-j)/2) = A(i,j) for j<=i<=n.
*
*  AFP     (input) REAL array, dimension (N*(N+1)/2)
*          The triangular factor U or L from the Cholesky factorization
*          A = U**T*U or A = L*L**T, as computed by SPPTRF/CPPTRF,
*          packed columnwise in a linear array in the same format as A
*          (see AP).
*
*  B       (input) REAL array, dimension (LDB,NRHS)
*          The right hand side matrix B.
*
*  LDB     (input) INTEGER
*          The leading dimension of the array B.  LDB >= max(1,N).
*
*  X       (input/output) REAL array, dimension (LDX,NRHS)
*          On entry, the solution matrix X, as computed by SPPTRS.
*          On exit, the improved solution matrix X.
*
*  LDX     (input) INTEGER
*          The leading dimension of the array X.  LDX >= max(1,N).
*
*  FERR    (output) REAL array, dimension (NRHS)
*          The estimated forward error bound for each solution vector
*          X(j) (the j-th column of the solution matrix X).
*          If XTRUE is the true solution corresponding to X(j), FERR(j)
*          is an estimated upper bound for the magnitude of the largest
*          element in (X(j) - XTRUE) divided by the magnitude of the
*          largest element in X(j).  The estimate is as reliable as
*          the estimate for RCOND, and is almost always a slight
*          overestimate of the true error.
*
*  BERR    (output) REAL array, dimension (NRHS)
*          The componentwise relative backward error of each solution
*          vector X(j) (i.e., the smallest relative change in
*          any element of A or B that makes X(j) an exact solution).
*
*  WORK    (workspace) REAL array, dimension (3*N)
*
*  IWORK   (workspace) INTEGER array, dimension (N)
*
*  INFO    (output) INTEGER
*          = 0:  successful exit
*          < 0:  if INFO = -i, the i-th argument had an illegal value
*
*  Internal Parameters
*  ===================
*
*  ITMAX is the maximum number of steps of iterative refinement.
*
*  =====================================================================
*
*     .. Parameters ..
      INTEGER            ITMAX
      PARAMETER          ( ITMAX = 5 )
      REAL               ZERO
      PARAMETER          ( ZERO = 0.0E+0 )
      REAL               ONE
      PARAMETER          ( ONE = 1.0E+0 )
      REAL               TWO
      PARAMETER          ( TWO = 2.0E+0 )
      REAL               THREE
      PARAMETER          ( THREE = 3.0E+0 )
*     ..
*     .. Local Scalars ..
      LOGICAL            UPPER
      INTEGER            COUNT, I, IK, J, K, KASE, KK, NZ
      REAL               EPS, LSTRES, S, SAFE1, SAFE2, SAFMIN, XK
*     ..
*     .. Local Arrays ..
      INTEGER            ISAVE( 3 )
*     ..
*     .. External Subroutines ..
      EXTERNAL           SAXPY, SCOPY, SLACN2, SPPTRS, SSPMV, XERBLA
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          ABSMAX
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      REAL               SLAMCH
      EXTERNAL           LSAME, SLAMCH
*     ..
*     .. Executable Statements ..
*
*     Test the input parameters.
*
      INFO = 0
      UPPER = LSAME( UPLO, 'U' )
      IF.NOT.UPPER .AND. .NOT.LSAME( UPLO, 'L' ) ) THEN
         INFO = -1
      ELSE IF( N.LT.0 ) THEN
         INFO = -2
      ELSE IF( NRHS.LT.0 ) THEN
         INFO = -3
      ELSE IF( LDB.LT.MAX1, N ) ) THEN
         INFO = -7
      ELSE IF( LDX.LT.MAX1, N ) ) THEN
         INFO = -9
      END IF
      IF( INFO.NE.0 ) THEN
         CALL XERBLA( 'SPPRFS'-INFO )
         RETURN
      END IF
*
*     Quick return if possible
*
      IF( N.EQ.0 .OR. NRHS.EQ.0 ) THEN
         DO 10 J = 1, NRHS
            FERR( J ) = ZERO
            BERR( J ) = ZERO
   10    CONTINUE
         RETURN
      END IF
*
*     NZ = maximum number of nonzero elements in each row of A, plus 1
*
      NZ = N + 1
      EPS = SLAMCH( 'Epsilon' )
      SAFMIN = SLAMCH( 'Safe minimum' )
      SAFE1 = NZ*SAFMIN
      SAFE2 = SAFE1 / EPS
*
*     Do for each right hand side
*
      DO 140 J = 1, NRHS
*
         COUNT = 1
         LSTRES = THREE
   20    CONTINUE
*
*        Loop until stopping criterion is satisfied.
*
*        Compute residual R = B - A * X
*
         CALL SCOPY( N, B( 1, J ), 1, WORK( N+1 ), 1 )
         CALL SSPMV( UPLO, N, -ONE, AP, X( 1, J ), 1, ONE, WORK( N+1 ),
     $               1 )
*
*        Compute componentwise relative backward error from formula
*
*        max(i) ( abs(R(i)) / ( abs(A)*abs(X) + abs(B) )(i) )
*
*        where abs(Z) is the componentwise absolute value of the matrix
*        or vector Z.  If the i-th component of the denominator is less
*        than SAFE2, then SAFE1 is added to the i-th components of the
*        numerator and denominator before dividing.
*
         DO 30 I = 1, N
            WORK( I ) = ABS( B( I, J ) )
   30    CONTINUE
*
*        Compute abs(A)*abs(X) + abs(B).
*
         KK = 1
         IF( UPPER ) THEN
            DO 50 K = 1, N
               S = ZERO
               XK = ABS( X( K, J ) )
               IK = KK
               DO 40 I = 1, K - 1
                  WORK( I ) = WORK( I ) + ABS( AP( IK ) )*XK
                  S = S + ABS( AP( IK ) )*ABS( X( I, J ) )
                  IK = IK + 1
   40          CONTINUE
               WORK( K ) = WORK( K ) + ABS( AP( KK+K-1 ) )*XK + S
               KK = KK + K
   50       CONTINUE
         ELSE
            DO 70 K = 1, N
               S = ZERO
               XK = ABS( X( K, J ) )
               WORK( K ) = WORK( K ) + ABS( AP( KK ) )*XK
               IK = KK + 1
               DO 60 I = K + 1, N
                  WORK( I ) = WORK( I ) + ABS( AP( IK ) )*XK
                  S = S + ABS( AP( IK ) )*ABS( X( I, J ) )
                  IK = IK + 1
   60          CONTINUE
               WORK( K ) = WORK( K ) + S
               KK = KK + ( N-K+1 )
   70       CONTINUE
         END IF
         S = ZERO
         DO 80 I = 1, N
            IF( WORK( I ).GT.SAFE2 ) THEN
               S = MAX( S, ABS( WORK( N+I ) ) / WORK( I ) )
            ELSE
               S = MAX( S, ( ABS( WORK( N+I ) )+SAFE1 ) /
     $             ( WORK( I )+SAFE1 ) )
            END IF
   80    CONTINUE
         BERR( J ) = S
*
*        Test stopping criterion. Continue iterating if
*           1) The residual BERR(J) is larger than machine epsilon, and
*           2) BERR(J) decreased by at least a factor of 2 during the
*              last iteration, and
*           3) At most ITMAX iterations tried.
*
         IF( BERR( J ).GT.EPS .AND. TWO*BERR( J ).LE.LSTRES .AND.
     $       COUNT.LE.ITMAX ) THEN
*
*           Update solution and try again.
*
            CALL SPPTRS( UPLO, N, 1, AFP, WORK( N+1 ), N, INFO )
            CALL SAXPY( N, ONE, WORK( N+1 ), 1, X( 1, J ), 1 )
            LSTRES = BERR( J )
            COUNT = COUNT + 1
            GO TO 20
         END IF
*
*        Bound error from formula
*
*        norm(X - XTRUE) / norm(X) .le. FERR =
*        norm( abs(inv(A))*
*           ( abs(R) + NZ*EPS*( abs(A)*abs(X)+abs(B) ))) / norm(X)
*
*        where
*          norm(Z) is the magnitude of the largest component of Z
*          inv(A) is the inverse of A
*          abs(Z) is the componentwise absolute value of the matrix or
*             vector Z
*          NZ is the maximum number of nonzeros in any row of A, plus 1
*          EPS is machine epsilon
*
*        The i-th component of abs(R)+NZ*EPS*(abs(A)*abs(X)+abs(B))
*        is incremented by SAFE1 if the i-th component of
*        abs(A)*abs(X) + abs(B) is less than SAFE2.
*
*        Use SLACN2 to estimate the infinity-norm of the matrix
*           inv(A) * diag(W),
*        where W = abs(R) + NZ*EPS*( abs(A)*abs(X)+abs(B) )))
*
         DO 90 I = 1, N
            IF( WORK( I ).GT.SAFE2 ) THEN
               WORK( I ) = ABS( WORK( N+I ) ) + NZ*EPS*WORK( I )
            ELSE
               WORK( I ) = ABS( WORK( N+I ) ) + NZ*EPS*WORK( I ) + SAFE1
            END IF
   90    CONTINUE
*
         KASE = 0
  100    CONTINUE
         CALL SLACN2( N, WORK( 2*N+1 ), WORK( N+1 ), IWORK, FERR( J ),
     $                KASE, ISAVE )
         IF( KASE.NE.0 ) THEN
            IF( KASE.EQ.1 ) THEN
*
*              Multiply by diag(W)*inv(A**T).
*
               CALL SPPTRS( UPLO, N, 1, AFP, WORK( N+1 ), N, INFO )
               DO 110 I = 1, N
                  WORK( N+I ) = WORK( I )*WORK( N+I )
  110          CONTINUE
            ELSE IF( KASE.EQ.2 ) THEN
*
*              Multiply by inv(A)*diag(W).
*
               DO 120 I = 1, N
                  WORK( N+I ) = WORK( I )*WORK( N+I )
  120          CONTINUE
               CALL SPPTRS( UPLO, N, 1, AFP, WORK( N+1 ), N, INFO )
            END IF
            GO TO 100
         END IF
*
*        Normalize error.
*
         LSTRES = ZERO
         DO 130 I = 1, N
            LSTRES = MAX( LSTRES, ABS( X( I, J ) ) )
  130    CONTINUE
         IF( LSTRES.NE.ZERO )
     $      FERR( J ) = FERR( J ) / LSTRES
*
  140 CONTINUE
*
      RETURN
*
*     End of SPPRFS
*
      END