ScaLAPACK 2.1  2.1
ScaLAPACK: Scalable Linear Algebra PACKage
pdgebrd.f
Go to the documentation of this file.
1  SUBROUTINE pdgebrd( M, N, A, IA, JA, DESCA, D, E, TAUQ, TAUP,
2  $ WORK, LWORK, INFO )
3 *
4 * -- ScaLAPACK routine (version 1.7) --
5 * University of Tennessee, Knoxville, Oak Ridge National Laboratory,
6 * and University of California, Berkeley.
7 * May 25, 2001
8 *
9 * .. Scalar Arguments ..
10  INTEGER IA, INFO, JA, LWORK, M, N
11 * ..
12 * .. Array Arguments ..
13  INTEGER DESCA( * )
14  DOUBLE PRECISION A( * ), D( * ), E( * ), TAUP( * ), TAUQ( * ),
15  $ work( * )
16 * ..
17 *
18 * Purpose
19 * =======
20 *
21 * PDGEBRD reduces a real general M-by-N distributed matrix
22 * sub( A ) = A(IA:IA+M-1,JA:JA+N-1) to upper or lower bidiagonal
23 * form B by an orthogonal transformation: Q' * sub( A ) * P = B.
24 *
25 * If M >= N, B is upper bidiagonal; if M < N, B is lower bidiagonal.
26 *
27 * Notes
28 * =====
29 *
30 * Each global data object is described by an associated description
31 * vector. This vector stores the information required to establish
32 * the mapping between an object element and its corresponding process
33 * and memory location.
34 *
35 * Let A be a generic term for any 2D block cyclicly distributed array.
36 * Such a global array has an associated description vector DESCA.
37 * In the following comments, the character _ should be read as
38 * "of the global array".
39 *
40 * NOTATION STORED IN EXPLANATION
41 * --------------- -------------- --------------------------------------
42 * DTYPE_A(global) DESCA( DTYPE_ )The descriptor type. In this case,
43 * DTYPE_A = 1.
44 * CTXT_A (global) DESCA( CTXT_ ) The BLACS context handle, indicating
45 * the BLACS process grid A is distribu-
46 * ted over. The context itself is glo-
47 * bal, but the handle (the integer
48 * value) may vary.
49 * M_A (global) DESCA( M_ ) The number of rows in the global
50 * array A.
51 * N_A (global) DESCA( N_ ) The number of columns in the global
52 * array A.
53 * MB_A (global) DESCA( MB_ ) The blocking factor used to distribute
54 * the rows of the array.
55 * NB_A (global) DESCA( NB_ ) The blocking factor used to distribute
56 * the columns of the array.
57 * RSRC_A (global) DESCA( RSRC_ ) The process row over which the first
58 * row of the array A is distributed.
59 * CSRC_A (global) DESCA( CSRC_ ) The process column over which the
60 * first column of the array A is
61 * distributed.
62 * LLD_A (local) DESCA( LLD_ ) The leading dimension of the local
63 * array. LLD_A >= MAX(1,LOCr(M_A)).
64 *
65 * Let K be the number of rows or columns of a distributed matrix,
66 * and assume that its process grid has dimension p x q.
67 * LOCr( K ) denotes the number of elements of K that a process
68 * would receive if K were distributed over the p processes of its
69 * process column.
70 * Similarly, LOCc( K ) denotes the number of elements of K that a
71 * process would receive if K were distributed over the q processes of
72 * its process row.
73 * The values of LOCr() and LOCc() may be determined via a call to the
74 * ScaLAPACK tool function, NUMROC:
75 * LOCr( M ) = NUMROC( M, MB_A, MYROW, RSRC_A, NPROW ),
76 * LOCc( N ) = NUMROC( N, NB_A, MYCOL, CSRC_A, NPCOL ).
77 * An upper bound for these quantities may be computed by:
78 * LOCr( M ) <= ceil( ceil(M/MB_A)/NPROW )*MB_A
79 * LOCc( N ) <= ceil( ceil(N/NB_A)/NPCOL )*NB_A
80 *
81 * Arguments
82 * =========
83 *
84 * M (global input) INTEGER
85 * The number of rows to be operated on, i.e. the number of rows
86 * of the distributed submatrix sub( A ). M >= 0.
87 *
88 * N (global input) INTEGER
89 * The number of columns to be operated on, i.e. the number of
90 * columns of the distributed submatrix sub( A ). N >= 0.
91 *
92 * A (local input/local output) DOUBLE PRECISION pointer into the
93 * local memory to an array of dimension (LLD_A,LOCc(JA+N-1)).
94 * On entry, this array contains the local pieces of the
95 * general distributed matrix sub( A ). On exit, if M >= N,
96 * the diagonal and the first superdiagonal of sub( A ) are
97 * overwritten with the upper bidiagonal matrix B; the elements
98 * below the diagonal, with the array TAUQ, represent the
99 * orthogonal matrix Q as a product of elementary reflectors,
100 * and the elements above the first superdiagonal, with the
101 * array TAUP, represent the orthogonal matrix P as a product
102 * of elementary reflectors. If M < N, the diagonal and the
103 * first subdiagonal are overwritten with the lower bidiagonal
104 * matrix B; the elements below the first subdiagonal, with the
105 * array TAUQ, represent the orthogonal matrix Q as a product of
106 * elementary reflectors, and the elements above the diagonal,
107 * with the array TAUP, represent the orthogonal matrix P as a
108 * product of elementary reflectors. See Further Details.
109 *
110 * IA (global input) INTEGER
111 * The row index in the global array A indicating the first
112 * row of sub( A ).
113 *
114 * JA (global input) INTEGER
115 * The column index in the global array A indicating the
116 * first column of sub( A ).
117 *
118 * DESCA (global and local input) INTEGER array of dimension DLEN_.
119 * The array descriptor for the distributed matrix A.
120 *
121 * D (local output) DOUBLE PRECISION array, dimension
122 * LOCc(JA+MIN(M,N)-1) if M >= N; LOCr(IA+MIN(M,N)-1) otherwise.
123 * The distributed diagonal elements of the bidiagonal matrix
124 * B: D(i) = A(i,i). D is tied to the distributed matrix A.
125 *
126 * E (local output) DOUBLE PRECISION array, dimension
127 * LOCr(IA+MIN(M,N)-1) if M >= N; LOCc(JA+MIN(M,N)-2) otherwise.
128 * The distributed off-diagonal elements of the bidiagonal
129 * distributed matrix B:
130 * if m >= n, E(i) = A(i,i+1) for i = 1,2,...,n-1;
131 * if m < n, E(i) = A(i+1,i) for i = 1,2,...,m-1.
132 * E is tied to the distributed matrix A.
133 *
134 * TAUQ (local output) DOUBLE PRECISION array dimension
135 * LOCc(JA+MIN(M,N)-1). The scalar factors of the elementary
136 * reflectors which represent the orthogonal matrix Q. TAUQ
137 * is tied to the distributed matrix A. See Further Details.
138 *
139 * TAUP (local output) DOUBLE PRECISION array, dimension
140 * LOCr(IA+MIN(M,N)-1). The scalar factors of the elementary
141 * reflectors which represent the orthogonal matrix P. TAUP
142 * is tied to the distributed matrix A. See Further Details.
143 *
144 * WORK (local workspace/local output) DOUBLE PRECISION array,
145 * dimension (LWORK)
146 * On exit, WORK( 1 ) returns the minimal and optimal LWORK.
147 *
148 * LWORK (local or global input) INTEGER
149 * The dimension of the array WORK.
150 * LWORK is local input and must be at least
151 * LWORK >= NB*( MpA0 + NqA0 + 1 ) + NqA0
152 *
153 * where NB = MB_A = NB_A,
154 * IROFFA = MOD( IA-1, NB ), ICOFFA = MOD( JA-1, NB ),
155 * IAROW = INDXG2P( IA, NB, MYROW, RSRC_A, NPROW ),
156 * IACOL = INDXG2P( JA, NB, MYCOL, CSRC_A, NPCOL ),
157 * MpA0 = NUMROC( M+IROFFA, NB, MYROW, IAROW, NPROW ),
158 * NqA0 = NUMROC( N+ICOFFA, NB, MYCOL, IACOL, NPCOL ).
159 *
160 * INDXG2P and NUMROC are ScaLAPACK tool functions;
161 * MYROW, MYCOL, NPROW and NPCOL can be determined by calling
162 * the subroutine BLACS_GRIDINFO.
163 *
164 * If LWORK = -1, then LWORK is global input and a workspace
165 * query is assumed; the routine only calculates the minimum
166 * and optimal size for all work arrays. Each of these
167 * values is returned in the first entry of the corresponding
168 * work array, and no error message is issued by PXERBLA.
169 *
170 * INFO (global output) INTEGER
171 * = 0: successful exit
172 * < 0: If the i-th argument is an array and the j-entry had
173 * an illegal value, then INFO = -(i*100+j), if the i-th
174 * argument is a scalar and had an illegal value, then
175 * INFO = -i.
176 *
177 * Further Details
178 * ===============
179 *
180 * The matrices Q and P are represented as products of elementary
181 * reflectors:
182 *
183 * If m >= n,
184 *
185 * Q = H(1) H(2) . . . H(n) and P = G(1) G(2) . . . G(n-1)
186 *
187 * Each H(i) and G(i) has the form:
188 *
189 * H(i) = I - tauq * v * v' and G(i) = I - taup * u * u'
190 *
191 * where tauq and taup are real scalars, and v and u are real vectors;
192 * v(1:i-1) = 0, v(i) = 1, and v(i+1:m) is stored on exit in
193 * A(ia+i:ia+m-1,ja+i-1);
194 * u(1:i) = 0, u(i+1) = 1, and u(i+2:n) is stored on exit in
195 * A(ia+i-1,ja+i+1:ja+n-1);
196 * tauq is stored in TAUQ(ja+i-1) and taup in TAUP(ia+i-1).
197 *
198 * If m < n,
199 *
200 * Q = H(1) H(2) . . . H(m-1) and P = G(1) G(2) . . . G(m)
201 *
202 * Each H(i) and G(i) has the form:
203 *
204 * H(i) = I - tauq * v * v' and G(i) = I - taup * u * u'
205 *
206 * where tauq and taup are real scalars, and v and u are real vectors;
207 * v(1:i) = 0, v(i+1) = 1, and v(i+2:m) is stored on exit in
208 * A(ia+i+1:ia+m-1,ja+i-1);
209 * u(1:i-1) = 0, u(i) = 1, and u(i+1:n) is stored on exit in
210 * A(ia+i-1,ja+i:ja+n-1);
211 * tauq is stored in TAUQ(ja+i-1) and taup in TAUP(ia+i-1).
212 *
213 * The contents of sub( A ) on exit are illustrated by the following
214 * examples:
215 *
216 * m = 6 and n = 5 (m > n): m = 5 and n = 6 (m < n):
217 *
218 * ( d e u1 u1 u1 ) ( d u1 u1 u1 u1 u1 )
219 * ( v1 d e u2 u2 ) ( e d u2 u2 u2 u2 )
220 * ( v1 v2 d e u3 ) ( v1 e d u3 u3 u3 )
221 * ( v1 v2 v3 d e ) ( v1 v2 e d u4 u4 )
222 * ( v1 v2 v3 v4 d ) ( v1 v2 v3 e d u5 )
223 * ( v1 v2 v3 v4 v5 )
224 *
225 * where d and e denote diagonal and off-diagonal elements of B, vi
226 * denotes an element of the vector defining H(i), and ui an element of
227 * the vector defining G(i).
228 *
229 * Alignment requirements
230 * ======================
231 *
232 * The distributed submatrix sub( A ) must verify some alignment proper-
233 * ties, namely the following expressions should be true:
234 * ( MB_A.EQ.NB_A .AND. IROFFA.EQ.ICOFFA )
235 *
236 * =====================================================================
237 *
238 * .. Parameters ..
239  INTEGER BLOCK_CYCLIC_2D, CSRC_, CTXT_, DLEN_, DTYPE_,
240  $ lld_, mb_, m_, nb_, n_, rsrc_
241  parameter( block_cyclic_2d = 1, dlen_ = 9, dtype_ = 1,
242  $ ctxt_ = 2, m_ = 3, n_ = 4, mb_ = 5, nb_ = 6,
243  $ rsrc_ = 7, csrc_ = 8, lld_ = 9 )
244  DOUBLE PRECISION ONE
245  parameter( one = 1.0d+0 )
246 * ..
247 * .. Local Scalars ..
248  LOGICAL LQUERY
249  CHARACTER COLCTOP, ROWCTOP
250  INTEGER I, IACOL, IAROW, ICTXT, IINFO, IOFF, IPW, IPY,
251  $ iw, j, jb, js, jw, k, l, lwmin, mn, mp, mycol,
252  $ myrow, nb, npcol, nprow, nq
253 * ..
254 * .. Local Arrays ..
255  INTEGER DESCWX( DLEN_ ), DESCWY( DLEN_ ), IDUM1( 1 ),
256  $ idum2( 1 )
257 * ..
258 * .. External Subroutines ..
259  EXTERNAL blacs_gridinfo, chk1mat, descset, pchk1mat,
260  $ pdelset, pdgebd2, pdgemm, pdlabrd,
261  $ pb_topget, pb_topset, pxerbla
262 * ..
263 * .. External Functions ..
264  INTEGER INDXG2L, INDXG2P, NUMROC
265  EXTERNAL indxg2l, indxg2p, numroc
266 * ..
267 * .. Intrinsic Functions ..
268  INTRINSIC dble, max, min, mod
269 * ..
270 * .. Executable Statements ..
271 *
272 * Get grid parameters
273 *
274  ictxt = desca( ctxt_ )
275  CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
276 *
277 * Test the input parameters
278 *
279  info = 0
280  IF( nprow.EQ.-1 ) THEN
281  info = -(600+ctxt_)
282  ELSE
283  CALL chk1mat( m, 1, n, 2, ia, ja, desca, 6, info )
284  IF( info.EQ.0 ) THEN
285  nb = desca( mb_ )
286  ioff = mod( ia-1, desca( mb_ ) )
287  iarow = indxg2p( ia, nb, myrow, desca( rsrc_ ), nprow )
288  iacol = indxg2p( ja, nb, mycol, desca( csrc_ ), npcol )
289  mp = numroc( m+ioff, nb, myrow, iarow, nprow )
290  nq = numroc( n+ioff, nb, mycol, iacol, npcol )
291  lwmin = nb*( mp+nq+1 ) + nq
292 *
293  work( 1 ) = dble( lwmin )
294  lquery = ( lwork.EQ.-1 )
295  IF( ioff.NE.mod( ja-1, desca( nb_ ) ) ) THEN
296  info = -5
297  ELSE IF( nb.NE.desca( nb_ ) ) THEN
298  info = -(600+nb_)
299  ELSE IF( lwork.LT.lwmin .AND. .NOT.lquery ) THEN
300  info = -12
301  END IF
302  END IF
303  IF( lquery ) THEN
304  idum1( 1 ) = -1
305  ELSE
306  idum1( 1 ) = 1
307  END IF
308  idum2( 1 ) = 12
309  CALL pchk1mat( m, 1, n, 2, ia, ja, desca, 6, 1, idum1, idum2,
310  $ info )
311  END IF
312 *
313  IF( info.LT.0 ) THEN
314  CALL pxerbla( ictxt, 'PDGEBRD', -info )
315  RETURN
316  ELSE IF( lquery ) THEN
317  RETURN
318  END IF
319 *
320 * Quick return if possible
321 *
322  mn = min( m, n )
323  IF( mn.EQ.0 )
324  $ RETURN
325 *
326 * Initialize parameters.
327 *
328  CALL pb_topget( ictxt, 'Combine', 'Columnwise', colctop )
329  CALL pb_topget( ictxt, 'Combine', 'Rowwise', rowctop )
330  CALL pb_topset( ictxt, 'Combine', 'Columnwise', '1-tree' )
331  CALL pb_topset( ictxt, 'Combine', 'Rowwise', '1-tree' )
332 *
333  ipy = mp * nb + 1
334  ipw = nq * nb + ipy
335 *
336  CALL descset( descwx, m+ioff, nb, nb, nb, iarow, iacol, ictxt,
337  $ max( 1, mp ) )
338  CALL descset( descwy, nb, n+ioff, nb, nb, iarow, iacol, ictxt,
339  $ nb )
340 *
341  mp = numroc( m+ia-1, nb, myrow, desca( rsrc_ ), nprow )
342  nq = numroc( n+ja-1, nb, mycol, desca( csrc_ ), npcol )
343  k = 1
344  jb = nb - ioff
345  iw = ioff + 1
346  jw = ioff + 1
347 *
348  DO 10 l = 1, mn+ioff-nb, nb
349  i = ia + k - 1
350  j = ja + k - 1
351 *
352 * Reduce rows and columns i:i+nb-1 to bidiagonal form and return
353 * the matrices X and Y which are needed to update the unreduced
354 * part of the matrix.
355 *
356  CALL pdlabrd( m-k+1, n-k+1, jb, a, i, j, desca, d, e, tauq,
357  $ taup, work, iw, jw, descwx, work( ipy ), iw,
358  $ jw, descwy, work( ipw ) )
359 *
360 * Update the trailing submatrix A(i+nb:ia+m-1,j+nb:ja+n-1), using
361 * an update of the form A := A - V*Y' - X*U'.
362 *
363  CALL pdgemm( 'No transpose', 'No transpose', m-k-jb+1,
364  $ n-k-jb+1, jb, -one, a, i+jb, j, desca,
365  $ work( ipy ), iw, jw+jb, descwy, one, a, i+jb,
366  $ j+jb, desca )
367  CALL pdgemm( 'No transpose', 'No transpose', m-k-jb+1,
368  $ n-k-jb+1, jb, -one, work, iw+jb, jw, descwx, a, i,
369  $ j+jb, desca, one, a, i+jb, j+jb, desca )
370 *
371 * Copy last off-diagonal elements of B back into sub( A ).
372 *
373  IF( m.GE.n ) THEN
374  js = min( indxg2l( i+jb-1, nb, 0, desca( rsrc_ ), nprow ),
375  $ mp )
376  IF( js.GT.0 )
377  $ CALL pdelset( a, i+jb-1, j+jb, desca, e( js ) )
378  ELSE
379  js = min( indxg2l( j+jb-1, nb, 0, desca( csrc_ ), npcol ),
380  $ nq )
381  IF( js.GT.0 )
382  $ CALL pdelset( a, i+jb, j+jb-1, desca, e( js ) )
383  END IF
384 *
385  k = k + jb
386  jb = nb
387  iw = 1
388  jw = 1
389  descwx( m_ ) = descwx( m_ ) - jb
390  descwx( rsrc_ ) = mod( descwx( rsrc_ ) + 1, nprow )
391  descwx( csrc_ ) = mod( descwx( csrc_ ) + 1, npcol )
392  descwy( n_ ) = descwy( n_ ) - jb
393  descwy( rsrc_ ) = mod( descwy( rsrc_ ) + 1, nprow )
394  descwy( csrc_ ) = mod( descwy( csrc_ ) + 1, npcol )
395 *
396  10 CONTINUE
397 *
398 * Use unblocked code to reduce the remainder of the matrix.
399 *
400  CALL pdgebd2( m-k+1, n-k+1, a, ia+k-1, ja+k-1, desca, d, e, tauq,
401  $ taup, work, lwork, iinfo )
402 *
403  CALL pb_topset( ictxt, 'Combine', 'Columnwise', colctop )
404  CALL pb_topset( ictxt, 'Combine', 'Rowwise', rowctop )
405 *
406  work( 1 ) = dble( lwmin )
407 *
408  RETURN
409 *
410 * End of PDGEBRD
411 *
412  END
pdgebd2
subroutine pdgebd2(M, N, A, IA, JA, DESCA, D, E, TAUQ, TAUP, WORK, LWORK, INFO)
Definition: pdgebd2.f:3
max
#define max(A, B)
Definition: pcgemr.c:180
pchk1mat
subroutine pchk1mat(MA, MAPOS0, NA, NAPOS0, IA, JA, DESCA, DESCAPOS0, NEXTRA, EX, EXPOS, INFO)
Definition: pchkxmat.f:3
pdgebrd
subroutine pdgebrd(M, N, A, IA, JA, DESCA, D, E, TAUQ, TAUP, WORK, LWORK, INFO)
Definition: pdgebrd.f:3
descset
subroutine descset(DESC, M, N, MB, NB, IRSRC, ICSRC, ICTXT, LLD)
Definition: descset.f:3
pdlabrd
subroutine pdlabrd(M, N, NB, A, IA, JA, DESCA, D, E, TAUQ, TAUP, X, IX, JX, DESCX, Y, IY, JY, DESCY, WORK)
Definition: pdlabrd.f:3
chk1mat
subroutine chk1mat(MA, MAPOS0, NA, NAPOS0, IA, JA, DESCA, DESCAPOS0, INFO)
Definition: chk1mat.f:3
pxerbla
subroutine pxerbla(ICTXT, SRNAME, INFO)
Definition: pxerbla.f:2
pdelset
subroutine pdelset(A, IA, JA, DESCA, ALPHA)
Definition: pdelset.f:2
min
#define min(A, B)
Definition: pcgemr.c:181