ScaLAPACK 2.1  2.1
ScaLAPACK: Scalable Linear Algebra PACKage
pdptlaschk.f
Go to the documentation of this file.
1  SUBROUTINE pdptlaschk( SYMM, UPLO, N, BWL, BWU, NRHS, X, IX, JX,
2  $ DESCX, IASEED, A, IA, JA, DESCA, IBSEED,
3  $ ANORM, RESID, WORK, WORKSIZ )
4 *
5 *
6 * -- ScaLAPACK routine (version 1.7) --
7 * University of Tennessee, Knoxville, Oak Ridge National Laboratory,
8 * and University of California, Berkeley.
9 * November 15, 1997
10 *
11 * .. Scalar Arguments ..
12  CHARACTER SYMM, UPLO
13  INTEGER BWL, BWU, IA, IASEED, IBSEED,
14  $ ix, ja, jx, n, nrhs, worksiz
15  DOUBLE PRECISION ANORM, RESID
16 * ..
17 * .. Array Arguments ..
18  INTEGER DESCA( * ), DESCX( * )
19  DOUBLE PRECISION A( * ), WORK( * ), X( * )
20 * .. External Functions ..
21  LOGICAL LSAME
22 * ..
23 *
24 * Purpose
25 * =======
26 *
27 * PDPTLASCHK computes the residual
28 * || sub( A )*sub( X ) - B || / (|| sub( A ) ||*|| sub( X ) ||*eps*N)
29 * to check the accuracy of the factorization and solve steps in the
30 * LU and Cholesky decompositions, where sub( A ) denotes
31 * A(IA:IA+N-1,JA,JA+N-1), sub( X ) denotes X(IX:IX+N-1, JX:JX+NRHS-1).
32 *
33 * Notes
34 * =====
35 *
36 * Each global data object is described by an associated description
37 * vector. This vector stores the information required to establish
38 * the mapping between an object element and its corresponding process
39 * and memory location.
40 *
41 * Let A be a generic term for any 2D block cyclicly distributed array.
42 * Such a global array has an associated description vector DESCA.
43 * In the following comments, the character _ should be read as
44 * "of the global array".
45 *
46 * NOTATION STORED IN EXPLANATION
47 * --------------- -------------- --------------------------------------
48 * DTYPE_A(global) DESCA( DTYPE_ )The descriptor type. In this case,
49 * DTYPE_A = 1.
50 * CTXT_A (global) DESCA( CTXT_ ) The BLACS context handle, indicating
51 * the BLACS process grid A is distribu-
52 * ted over. The context itself is glo-
53 * bal, but the handle (the integer
54 * value) may vary.
55 * M_A (global) DESCA( M_ ) The number of rows in the global
56 * array A.
57 * N_A (global) DESCA( N_ ) The number of columns in the global
58 * array A.
59 * MB_A (global) DESCA( MB_ ) The blocking factor used to distribute
60 * the rows of the array.
61 * NB_A (global) DESCA( NB_ ) The blocking factor used to distribute
62 * the columns of the array.
63 * RSRC_A (global) DESCA( RSRC_ ) The process row over which the first
64 * row of the array A is distributed.
65 * CSRC_A (global) DESCA( CSRC_ ) The process column over which the
66 * first column of the array A is
67 * distributed.
68 * LLD_A (local) DESCA( LLD_ ) The leading dimension of the local
69 * array. LLD_A >= MAX(1,LOCr(M_A)).
70 *
71 * Let K be the number of rows or columns of a distributed matrix,
72 * and assume that its process grid has dimension p x q.
73 * LOCr( K ) denotes the number of elements of K that a process
74 * would receive if K were distributed over the p processes of its
75 * process column.
76 * Similarly, LOCc( K ) denotes the number of elements of K that a
77 * process would receive if K were distributed over the q processes of
78 * its process row.
79 * The values of LOCr() and LOCc() may be determined via a call to the
80 * ScaLAPACK tool function, NUMROC:
81 * LOCr( M ) = NUMROC( M, MB_A, MYROW, RSRC_A, NPROW ),
82 * LOCc( N ) = NUMROC( N, NB_A, MYCOL, CSRC_A, NPCOL ).
83 * An upper bound for these quantities may be computed by:
84 * LOCr( M ) <= ceil( ceil(M/MB_A)/NPROW )*MB_A
85 * LOCc( N ) <= ceil( ceil(N/NB_A)/NPCOL )*NB_A
86 *
87 * Arguments
88 * =========
89 *
90 * SYMM (global input) CHARACTER
91 * if SYMM = 'S', sub( A ) is a symmetric distributed band
92 * matrix, otherwise sub( A ) is a general distributed matrix.
93 *
94 * UPLO (global input) CHARACTER
95 * if SYMM = 'S', then
96 * if UPLO = 'L', the lower half of the matrix is stored
97 * if UPLO = 'U', the upper half of the matrix is stored
98 * if SYMM != 'S' or 'H', then
99 * if UPLO = 'D', the matrix is stable during factorization
100 * without interchanges
101 * if UPLO != 'D', the matrix is general
102 *
103 * N (global input) INTEGER
104 * The number of columns to be operated on, i.e. the number of
105 * columns of the distributed submatrix sub( A ). N >= 0.
106 *
107 * NRHS (global input) INTEGER
108 * The number of right-hand-sides, i.e the number of columns
109 * of the distributed matrix sub( X ). NRHS >= 1.
110 *
111 * X (local input) DOUBLE PRECISION pointer into the local memory
112 * to an array of dimension (LLD_X,LOCq(JX+NRHS-1). This array
113 * contains the local pieces of the answer vector(s) sub( X ) of
114 * sub( A ) sub( X ) - B, split up over a column of processes.
115 *
116 * IX (global input) INTEGER
117 * The row index in the global array X indicating the first
118 * row of sub( X ).
119 *
120 * DESCX (global and local input) INTEGER array of dimension DLEN_.
121 * The array descriptor for the distributed matrix X.
122 *
123 * IASEED (global input) INTEGER
124 * The seed number to generate the original matrix Ao.
125 *
126 * JA (global input) INTEGER
127 * The column index in the global array A indicating the
128 * first column of sub( A ).
129 *
130 * DESCA (global and local input) INTEGER array of dimension DLEN_.
131 * The array descriptor for the distributed matrix A.
132 *
133 * IBSEED (global input) INTEGER
134 * The seed number to generate the original matrix B.
135 *
136 * ANORM (global input) DOUBLE PRECISION
137 * The 1-norm or infinity norm of the distributed matrix
138 * sub( A ).
139 *
140 * RESID (global output) DOUBLE PRECISION
141 * The residual error:
142 * ||sub( A )*sub( X )-B|| / (||sub( A )||*||sub( X )||*eps*N).
143 *
144 * WORK (local workspace) DOUBLE PRECISION array, dimension (LWORK)
145 * IF SYMM='S'
146 * LWORK >= max(5,NB)+2*NB
147 * IF SYMM!='S' or 'H'
148 * LWORK >= max(5,NB)+2*NB
149 *
150 * WORKSIZ (local input) size of WORK.
151 *
152 * =====================================================================
153 *
154 * Code Developer: Andrew J. Cleary, University of Tennessee.
155 * Current address: Lawrence Livermore National Labs.
156 * This version released: August, 2001.
157 *
158 * =====================================================================
159 *
160 * .. Parameters ..
161  DOUBLE PRECISION ZERO, ONE
162  PARAMETER ( ONE = 1.0d+0, zero = 0.0d+0 )
163  INTEGER BLOCK_CYCLIC_2D, CSRC_, CTXT_, DLEN_, DTYPE_,
164  $ lld_, mb_, m_, nb_, n_, rsrc_
165  parameter( block_cyclic_2d = 1, dlen_ = 9, dtype_ = 1,
166  $ ctxt_ = 2, m_ = 3, n_ = 4, mb_ = 5, nb_ = 6,
167  $ rsrc_ = 7, csrc_ = 8, lld_ = 9 )
168  INTEGER INT_ONE
169  PARAMETER ( INT_ONE = 1 )
170 * ..
171 * .. Local Scalars ..
172  INTEGER IACOL, IAROW, ICTXT,
173  $ IIA, IIX, IPB, IPW,
174  $ ixcol, ixrow, j, jja, jjx, lda,
175  $ mycol, myrow, nb, np, npcol, nprow, nq
176  INTEGER I, START
177  INTEGER BW, INFO, IPPRODUCT, WORK_MIN
178  DOUBLE PRECISION DIVISOR, EPS, RESID1, NORMX
179 * ..
180 * .. Local Arrays ..
181 * ..
182 * .. External Subroutines ..
183  EXTERNAL blacs_gridinfo, dgamx2d, dgebr2d,
184  $ dgebs2d, dgemm, dgerv2d, dgesd2d,
185  $ dgsum2d, dlaset, pbdtran, pdmatgen
186 * ..
187 * .. External Functions ..
188  INTEGER IDAMAX, NUMROC
189  DOUBLE PRECISION PDLAMCH
190  EXTERNAL idamax, numroc, pdlamch
191 * ..
192 * .. Intrinsic Functions ..
193  INTRINSIC abs, dble, max, min, mod
194 * ..
195 * .. Executable Statements ..
196 *
197 * Get needed initial parameters
198 *
199  ictxt = desca( ctxt_ )
200  nb = desca( nb_ )
201 *
202  IF( lsame( symm, 'S' ) ) THEN
203  bw = bwl
204  start = 1
205  work_min = max(5,nb)+2*nb
206  ELSE
207  bw = max(bwl, bwu)
208  IF( lsame( uplo, 'D' )) THEN
209  start = 1
210  ELSE
211  start = 2
212  ENDIF
213  work_min = max(5,nb)+2*nb
214  ENDIF
215 *
216  IF ( worksiz .LT. work_min ) THEN
217  CALL pxerbla( ictxt, 'PDTLASCHK', -18 )
218  RETURN
219  END IF
220 *
221  CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
222 *
223  eps = pdlamch( ictxt, 'eps' )
224  resid = 0.0d+0
225  divisor = anorm * eps * dble( n )
226 *
227  CALL infog2l( ia, ja, desca, nprow, npcol, myrow, mycol, iia, jja,
228  $ iarow, iacol )
229  CALL infog2l( ix, jx, descx, nprow, npcol, myrow, mycol, iix, jjx,
230  $ ixrow, ixcol )
231  np = numroc( (2), desca( mb_ ), myrow, 0, nprow )
232  nq = numroc( n, desca( nb_ ), mycol, 0, npcol )
233 *
234  ipb = 1
235  ipproduct = 1 + desca( nb_ )
236  ipw = 1 + 2*desca( nb_ )
237 *
238  lda = desca( lld_ )
239 *
240 * Regenerate A
241 *
242  IF( lsame( symm, 'S' )) THEN
243  CALL pdbmatgen( ictxt, uplo, 'D', bw, bw, n, bw+1,
244  $ desca( nb_ ), a, desca( lld_ ), 0, 0,
245  $ iaseed, myrow, mycol, nprow, npcol )
246  ELSE
247 *
248  CALL pdbmatgen( ictxt, 'N', uplo, bwl, bwu, n,
249  $ desca( mb_ ), desca( nb_ ), a,
250  $ desca( lld_ ), 0, 0, iaseed, myrow,
251  $ mycol, nprow, npcol )
252  ENDIF
253  IF( lsame( uplo, 'U' ) ) THEN
254 *
255 *
256 * Matrix formed above has the diagonals shifted from what was
257 * input to the tridiagonal routine. Shift them back.
258 *
259 * Send elements to neighboring processors
260 *
261  IF( mycol.LT.npcol-1 ) THEN
262  CALL dgesd2d( ictxt, 1, 1,
263  $ a( start+( desca( nb_ )-1 )*lda ),
264  $ lda, myrow, mycol+1 )
265  ENDIF
266 *
267 * Shift local elements
268 *
269  DO 230 i=desca( nb_ )-1,0,-1
270  a( start+(i+1)*lda ) = a( start+(i)*lda )
271  230 CONTINUE
272 *
273 * Receive elements from neighboring processors
274 *
275  IF( mycol.GT.0 ) THEN
276  CALL dgerv2d( ictxt, 1, 1, a( start), lda,
277  $ myrow, mycol-1 )
278  ENDIF
279 *
280  ENDIF
281 *
282 * Loop over the rhs
283 *
284  resid = 0.0
285 *
286  DO 40 j = 1, nrhs
287 *
288 * Multiply A * current column of X
289 *
290 *
291  CALL pdpbdcmv( bw+1, bw, uplo, n, a, 1, desca,
292  $ 1, x( 1 + (j-1)*descx( lld_ )), 1, descx,
293  $ work( ipproduct ), work( ipw ), (bw+2)*bw, info )
294 *
295 *
296 * Regenerate column of B
297 *
298  CALL pdmatgen( descx( ctxt_ ), 'No', 'No', descx( m_ ),
299  $ descx( n_ ), descx( mb_ ), descx( nb_ ),
300  $ work( ipb ), descx( lld_ ), descx( rsrc_ ),
301  $ descx( csrc_ ), ibseed, 0, nq, j-1, 1, mycol,
302  $ myrow, npcol, nprow )
303 *
304 * Figure || A * X - B || & || X ||
305 *
306  CALL pdaxpy( n, -one, work( ipproduct ), 1, 1, descx, 1,
307  $ work( ipb ), 1, 1, descx, 1 )
308 *
309  CALL pdnrm2( n, normx,
310  $ x, 1, j, descx, 1 )
311 *
312  CALL pdnrm2( n, resid1,
313  $ work( ipb ), 1, 1, descx, 1 )
314 *
315 *
316 * Calculate residual = ||Ax-b|| / (||x||*||A||*eps*N)
317 *
318  resid1 = resid1 / ( normx*divisor )
319 *
320  resid = max( resid, resid1 )
321 *
322  40 CONTINUE
323 *
324  RETURN
325 *
326 * End of PDTLASCHK
327 *
328  END
pdptlaschk
subroutine pdptlaschk(SYMM, UPLO, N, BWL, BWU, NRHS, X, IX, JX, DESCX, IASEED, A, IA, JA, DESCA, IBSEED, ANORM, RESID, WORK, WORKSIZ)
Definition: pdptlaschk.f:4
max
#define max(A, B)
Definition: pcgemr.c:180
pdpbdcmv
subroutine pdpbdcmv(LDBW, BW, UPLO, N, A, JA, DESCA, NRHS, B, IB, DESCB, X, WORK, LWORK, INFO)
Definition: pdpbmv1.f:3
infog2l
subroutine infog2l(GRINDX, GCINDX, DESC, NPROW, NPCOL, MYROW, MYCOL, LRINDX, LCINDX, RSRC, CSRC)
Definition: infog2l.f:3
pbdtran
subroutine pbdtran(ICONTXT, ADIST, TRANS, M, N, NB, A, LDA, BETA, C, LDC, IAROW, IACOL, ICROW, ICCOL, WORK)
Definition: pbdtran.f:3
pdmatgen
subroutine pdmatgen(ICTXT, AFORM, DIAG, M, N, MB, NB, A, LDA, IAROW, IACOL, ISEED, IROFF, IRNUM, ICOFF, ICNUM, MYROW, MYCOL, NPROW, NPCOL)
Definition: pdmatgen.f:4
pdbmatgen
subroutine pdbmatgen(ICTXT, AFORM, AFORM2, BWL, BWU, N, MB, NB, A, LDA, IAROW, IACOL, ISEED, MYROW, MYCOL, NPROW, NPCOL)
Definition: pdbmatgen.f:5
pxerbla
subroutine pxerbla(ICTXT, SRNAME, INFO)
Definition: pxerbla.f:2
min
#define min(A, B)
Definition: pcgemr.c:181