ScaLAPACK 2.1  2.1
ScaLAPACK: Scalable Linear Algebra PACKage
pcgerqrv.f
Go to the documentation of this file.
1  SUBROUTINE pcgerqrv( M, N, A, IA, JA, DESCA, TAU, WORK )
2 *
3 * -- ScaLAPACK routine (version 1.7) --
4 * University of Tennessee, Knoxville, Oak Ridge National Laboratory,
5 * and University of California, Berkeley.
6 * May 28, 2001
7 *
8 * .. Scalar Arguments ..
9  INTEGER IA, JA, M, N
10 * ..
11 * .. Array Arguments ..
12  INTEGER DESCA( * )
13  COMPLEX A( * ), TAU( * ), WORK( * )
14 * ..
15 *
16 * Purpose
17 * =======
18 *
19 * PCGERQRV computes sub( A ) = A(IA:IA+M-1,JA:JA+N-1) from R, Q
20 * computed by PCGERQF.
21 *
22 * Notes
23 * =====
24 *
25 * Each global data object is described by an associated description
26 * vector. This vector stores the information required to establish
27 * the mapping between an object element and its corresponding process
28 * and memory location.
29 *
30 * Let A be a generic term for any 2D block cyclicly distributed array.
31 * Such a global array has an associated description vector DESCA.
32 * In the following comments, the character _ should be read as
33 * "of the global array".
34 *
35 * NOTATION STORED IN EXPLANATION
36 * --------------- -------------- --------------------------------------
37 * DTYPE_A(global) DESCA( DTYPE_ )The descriptor type. In this case,
38 * DTYPE_A = 1.
39 * CTXT_A (global) DESCA( CTXT_ ) The BLACS context handle, indicating
40 * the BLACS process grid A is distribu-
41 * ted over. The context itself is glo-
42 * bal, but the handle (the integer
43 * value) may vary.
44 * M_A (global) DESCA( M_ ) The number of rows in the global
45 * array A.
46 * N_A (global) DESCA( N_ ) The number of columns in the global
47 * array A.
48 * MB_A (global) DESCA( MB_ ) The blocking factor used to distribute
49 * the rows of the array.
50 * NB_A (global) DESCA( NB_ ) The blocking factor used to distribute
51 * the columns of the array.
52 * RSRC_A (global) DESCA( RSRC_ ) The process row over which the first
53 * row of the array A is distributed.
54 * CSRC_A (global) DESCA( CSRC_ ) The process column over which the
55 * first column of the array A is
56 * distributed.
57 * LLD_A (local) DESCA( LLD_ ) The leading dimension of the local
58 * array. LLD_A >= MAX(1,LOCr(M_A)).
59 *
60 * Let K be the number of rows or columns of a distributed matrix,
61 * and assume that its process grid has dimension p x q.
62 * LOCr( K ) denotes the number of elements of K that a process
63 * would receive if K were distributed over the p processes of its
64 * process column.
65 * Similarly, LOCc( K ) denotes the number of elements of K that a
66 * process would receive if K were distributed over the q processes of
67 * its process row.
68 * The values of LOCr() and LOCc() may be determined via a call to the
69 * ScaLAPACK tool function, NUMROC:
70 * LOCr( M ) = NUMROC( M, MB_A, MYROW, RSRC_A, NPROW ),
71 * LOCc( N ) = NUMROC( N, NB_A, MYCOL, CSRC_A, NPCOL ).
72 * An upper bound for these quantities may be computed by:
73 * LOCr( M ) <= ceil( ceil(M/MB_A)/NPROW )*MB_A
74 * LOCc( N ) <= ceil( ceil(N/NB_A)/NPCOL )*NB_A
75 *
76 * Arguments
77 * =========
78 *
79 * M (global input) INTEGER
80 * The number of rows to be operated on, i.e. the number of rows
81 * of the distributed submatrix sub( A ). M >= 0.
82 *
83 * N (global input) INTEGER
84 * The number of columns to be operated on, i.e. the number of
85 * columns of the distributed submatrix sub( A ). N >= 0.
86 *
87 * A (local input/local output) COMPLEX pointer into the
88 * local memory to an array of dimension (LLD_A, LOCc(JA+N-1)).
89 * On entry, sub( A ) contains the the factors R and Q computed
90 * by PCGERQF. On exit, the original matrix is restored.
91 *
92 * IA (global input) INTEGER
93 * The row index in the global array A indicating the first
94 * row of sub( A ).
95 *
96 * JA (global input) INTEGER
97 * The column index in the global array A indicating the
98 * first column of sub( A ).
99 *
100 * DESCA (global and local input) INTEGER array of dimension DLEN_.
101 * The array descriptor for the distributed matrix A.
102 *
103 * TAU (local input) COMPLEX, array, dimension LOCr(M_A).
104 * This array contains the scalar factors TAU of the elementary
105 * reflectors computed by PCGERQF. TAU is tied to the dis-
106 * tributed matrix A.
107 *
108 * WORK (local workspace) COMPLEX array, dimension (LWORK)
109 * LWORK = MB_A * ( Mp0 + 2*Nq0 + MB_A ), where
110 * Mp0 = NUMROC( M+IROFF, MB_A, MYROW, IAROW, NPROW ) * NB_A,
111 * Nq0 = NUMROC( N+ICOFF, NB_A, MYCOL, IACOL, NPCOL ) * MB_A,
112 * IROFF = MOD( IA-1, MB_A ), ICOFF = MOD( JA-1, NB_A ),
113 * IAROW = INDXG2P( IA, DESCA( MB_ ), MYROW, DESCA( RSRC_ ),
114 * NPROW ),
115 * IACOL = INDXG2P( JA, DESCA( NB_ ), MYCOL, DESCA( CSRC_ ),
116 * NPCOL ),
117 * and NUMROC, INDXG2P are ScaLAPACK tool functions.
118 *
119 * =====================================================================
120 *
121 * .. Parameters ..
122  INTEGER BLOCK_CYCLIC_2D, CSRC_, CTXT_, DLEN_, DTYPE_,
123  $ LLD_, MB_, M_, NB_, N_, RSRC_
124  parameter( block_cyclic_2d = 1, dlen_ = 9, dtype_ = 1,
125  $ ctxt_ = 2, m_ = 3, n_ = 4, mb_ = 5, nb_ = 6,
126  $ rsrc_ = 7, csrc_ = 8, lld_ = 9 )
127  COMPLEX ONE, ZERO
128  parameter( one = ( 1.0e+0, 0.0e+0 ),
129  $ zero = ( 0.0e+0, 0.0e+0 ) )
130 * ..
131 * .. Local Scalars ..
132  CHARACTER COLBTOP, ROWBTOP
133  INTEGER I, IACOL, IAROW, IB, ICOFF, ICTXT, IIA, IN,
134  $ IPT, IPV, IPW, JJA, JV, K, MYCOL, MYROW, NPCOL,
135  $ NPROW, NQ
136 * ..
137 * .. Local Arrays ..
138  INTEGER DESCV( DLEN_ )
139 * ..
140 * .. External Subroutines ..
141  EXTERNAL blacs_gridinfo, descset, infog2l, pclacpy,
142  $ pclarfb, pclarft, pclaset, pb_topget,
143  $ pb_topset
144 * ..
145 * .. External Functions ..
146  INTEGER ICEIL, NUMROC
147  EXTERNAL iceil, numroc
148 * ..
149 * .. Intrinsic Functions ..
150  INTRINSIC max, min, mod
151 * ..
152 * .. Executable Statements ..
153 *
154 * Get grid parameters
155 *
156  ictxt = desca( ctxt_ )
157  CALL blacs_gridinfo( ictxt, nprow, npcol, myrow, mycol )
158 *
159  k = min( m, n )
160  in = min( iceil( ia+m-k, desca( mb_ ) ) * desca( mb_ ), ia+m-1 )
161 *
162  icoff = mod( ja-1, desca( nb_ ) )
163  CALL infog2l( ia+m-k, ja, desca, nprow, npcol, myrow, mycol,
164  $ iia, jja, iarow, iacol )
165  nq = numroc( n+icoff, desca( nb_ ), mycol, iacol, npcol )
166  ipv = 1
167  ipt = ipv + nq * desca( mb_ )
168  ipw = ipt + desca( mb_ ) * desca( mb_ )
169  CALL pb_topget( ictxt, 'Broadcast', 'Rowwise', rowbtop )
170  CALL pb_topget( ictxt, 'Broadcast', 'Columnwise', colbtop )
171  CALL pb_topset( ictxt, 'Broadcast', 'Rowwise', ' ' )
172  CALL pb_topset( ictxt, 'Broadcast', 'Columnwise', 'I-ring' )
173 *
174  CALL descset( descv, desca( mb_), n + icoff, desca( mb_ ),
175  $ desca( nb_ ), iarow, iacol, ictxt, desca( mb_ ) )
176 *
177 * Handle first block separately
178 *
179  ib = in - ia - m + k + 1
180  jv = 1 + n - k + icoff
181 *
182 * Compute upper triangular matrix T
183 *
184  CALL pclarft( 'Backward', 'Rowwise', n-m+in-ia+1, ib, a, ia+m-k,
185  $ ja, desca, tau, work( ipt ), work( ipw ) )
186 *
187 * Copy Householder vectors into workspace
188 *
189  CALL pclacpy( 'All', ib, n-m+in-ia+1, a, ia+m-k, ja, desca,
190  $ work( ipv ), 1, icoff+1, descv )
191  CALL pclaset( 'Upper', ib, ib, zero, one, work( ipv ), 1, jv,
192  $ descv )
193 *
194 * Zeoes the strict lower triangular part of sub( A ) to get block
195 * column of R
196 *
197  CALL pclaset( 'All', ib, n-k, zero, zero, a, ia+m-k, ja,
198  $ desca )
199  CALL pclaset( 'Lower', ib-1, ib, zero, zero, a, ia+m-k+1,
200  $ ja+n-k, desca )
201 *
202 * Apply block Householder transformation
203 *
204  CALL pclarfb( 'Right', 'Conjugate transpose', 'Backward',
205  $ 'Rowwise', in-ia+1, n-m+in-ia+1, ib, work( ipv ), 1,
206  $ icoff+1, descv, work( ipt ), a, ia, ja, desca,
207  $ work( ipw ) )
208 *
209  descv( rsrc_ ) = mod( descv( rsrc_ ) + 1, nprow )
210 *
211 * Loop over the remaining row blocks
212 *
213  DO 10 i = in+1, ia+m-1, desca( mb_ )
214  ib = min( ia+m-i, desca( mb_ ) )
215  jv = 1 + n - m + i - ia + icoff
216 *
217 * Compute upper triangular matrix T
218 *
219  CALL pclarft( 'Backward', 'Rowwise', n-m+i+ib-ia, ib, a, i, ja,
220  $ desca, tau, work( ipt ), work( ipw ) )
221 *
222 * Copy Householder vectors into workspace
223 *
224  CALL pclacpy( 'All', ib, n-m+i+ib-ia, a, i, ja, desca,
225  $ work( ipv ), 1, icoff+1, descv )
226  CALL pclaset( 'Upper', ib, ib, zero, one, work( ipv ), 1, jv,
227  $ descv )
228 *
229 * Zeoes the strict Lower triangular part of sub( A ) to get
230 * block column of R
231 *
232  CALL pclaset( 'All', ib, n-m+i-ia, zero, zero, a, i, ja,
233  $ desca )
234  CALL pclaset( 'Lower', ib-1, ib, zero, zero, a, i+1,
235  $ ja+n-m+i-ia, desca )
236 *
237 * Apply block Householder transformation
238 *
239  CALL pclarfb( 'Right', 'Conjugate transpose', 'Backward',
240  $ 'Rowwise', i+ib-ia, n-m+i+ib-ia, ib, work( ipv ),
241  $ 1, icoff+1, descv, work( ipt ), a, ia, ja, desca,
242  $ work( ipw ) )
243 *
244  descv( rsrc_ ) = mod( descv( rsrc_ ) + 1, nprow )
245 *
246  10 CONTINUE
247 *
248  CALL pb_topset( ictxt, 'Broadcast', 'Rowwise', rowbtop )
249  CALL pb_topset( ictxt, 'Broadcast', 'Columnwise', colbtop )
250 *
251  RETURN
252 *
253 * End of PCGERQRV
254 *
255  END
max
#define max(A, B)
Definition: pcgemr.c:180
infog2l
subroutine infog2l(GRINDX, GCINDX, DESC, NPROW, NPCOL, MYROW, MYCOL, LRINDX, LCINDX, RSRC, CSRC)
Definition: infog2l.f:3
pcgerqrv
subroutine pcgerqrv(M, N, A, IA, JA, DESCA, TAU, WORK)
Definition: pcgerqrv.f:2
pclarfb
subroutine pclarfb(SIDE, TRANS, DIRECT, STOREV, M, N, K, V, IV, JV, DESCV, T, C, IC, JC, DESCC, WORK)
Definition: pclarfb.f:3
descset
subroutine descset(DESC, M, N, MB, NB, IRSRC, ICSRC, ICTXT, LLD)
Definition: descset.f:3
pclaset
subroutine pclaset(UPLO, M, N, ALPHA, BETA, A, IA, JA, DESCA)
Definition: pcblastst.f:7508
pclacpy
subroutine pclacpy(UPLO, M, N, A, IA, JA, DESCA, B, IB, JB, DESCB)
Definition: pclacpy.f:3
pclarft
subroutine pclarft(DIRECT, STOREV, N, K, V, IV, JV, DESCV, TAU, T, WORK)
Definition: pclarft.f:3
min
#define min(A, B)
Definition: pcgemr.c:181