-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpdgsequ.c
244 lines (201 loc) · 7.29 KB
/
pdgsequ.c
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
/*! \file
Copyright (c) 2003, The Regents of the University of California, through
Lawrence Berkeley National Laboratory (subject to receipt of any required
approvals from U.S. Dept. of Energy)
All rights reserved.
The source code is distributed under BSD license, see the file License.txt
at the top-level directory.
*/
/*! @file
* \brief Computes row and column scalings
*
* File name: pdgsequ.c
* History: Modified from LAPACK routine DGEEQU
*/
#include <math.h>
#include "superlu_ddefs.h"
/*! \brief
<pre>
Purpose
=======
PDGSEQU computes row and column scalings intended to equilibrate an
M-by-N sparse matrix A and reduce its condition number. R returns the row
scale factors and C the column scale factors, chosen to try to make
the largest element in each row and column of the matrix B with
elements B(i,j)=R(i)*A(i,j)*C(j) have absolute value 1.
R(i) and C(j) are restricted to be between SMLNUM = smallest safe
number and BIGNUM = largest safe number. Use of these scaling
factors is not guaranteed to reduce the condition number of A but
works well in practice.
See supermatrix.h for the definition of 'SuperMatrix' structure.
Arguments
=========
A (input) SuperMatrix*
The matrix of dimension (A->nrow, A->ncol) whose equilibration
factors are to be computed. The type of A can be:
Stype = SLU_NR_loc; Dtype = SLU_D; Mtype = SLU_GE.
R (output) double*, size A->nrow
If INFO = 0 or INFO > M, R contains the row scale factors
for A.
C (output) double*, size A->ncol
If INFO = 0, C contains the column scale factors for A.
ROWCND (output) double*
If INFO = 0 or INFO > M, ROWCND contains the ratio of the
smallest R(i) to the largest R(i). If ROWCND >= 0.1 and
AMAX is neither too large nor too small, it is not worth
scaling by R.
COLCND (output) double*
If INFO = 0, COLCND contains the ratio of the smallest
C(i) to the largest C(i). If COLCND >= 0.1, it is not
worth scaling by C.
AMAX (output) double*
Absolute value of largest matrix element. If AMAX is very
close to overflow or very close to underflow, the matrix
should be scaled.
INFO (output) int*
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
> 0: if INFO = i, and i is
<= M: the i-th row of A is exactly zero
> M: the (i-M)-th column of A is exactly zero
GRID (input) gridinof_t*
The 2D process mesh.
=====================================================================
</pre>
*/
void
pdgsequ(SuperMatrix *A, double *r, double *c, double *rowcnd,
double *colcnd, double *amax, int_t *info, gridinfo_t *grid)
{
/* Local variables */
NRformat_loc *Astore;
double *Aval;
int i, j, irow, jcol, m_loc;
double rcmin, rcmax;
double bignum, smlnum;
double tempmax, tempmin;
double *loc_max;
int *r_sizes, *displs;
double *loc_r;
int_t procs;
/* Test the input parameters. */
*info = 0;
if ( A->nrow < 0 || A->ncol < 0 ||
A->Stype != SLU_NR_loc || A->Dtype != SLU_D || A->Mtype != SLU_GE )
*info = -1;
if (*info != 0) {
i = -(*info);
pxerr_dist("pdgsequ", grid, i);
return;
}
/* Quick return if possible */
if ( A->nrow == 0 || A->ncol == 0 ) {
*rowcnd = 1.;
*colcnd = 1.;
*amax = 0.;
return;
}
Astore = A->Store;
Aval = Astore->nzval;
m_loc = Astore->m_loc;
/* Get machine constants. */
smlnum = dmach_dist("S");
bignum = 1. / smlnum;
/* Compute row scale factors. */
for (i = 0; i < A->nrow; ++i) r[i] = 0.;
/* Find the maximum element in each row. */
irow = Astore->fst_row;
for (i = 0; i < m_loc; ++i) {
for (j = Astore->rowptr[i]; j < Astore->rowptr[i+1]; ++j)
r[irow] = SUPERLU_MAX( r[irow], fabs(Aval[j]) );
++irow;
}
/* Find the maximum and minimum scale factors. */
rcmin = bignum;
rcmax = 0.;
for (i = Astore->fst_row; i < Astore->fst_row + m_loc; ++i) {
rcmax = SUPERLU_MAX(rcmax, r[i]);
rcmin = SUPERLU_MIN(rcmin, r[i]);
}
/* Get the global MAX and MIN for R */
tempmax = rcmax;
tempmin = rcmin;
MPI_Allreduce( &tempmax, &rcmax,
1, MPI_DOUBLE, MPI_MAX, grid->comm);
MPI_Allreduce( &tempmin, &rcmin,
1, MPI_DOUBLE, MPI_MIN, grid->comm);
*amax = rcmax;
if (rcmin == 0.) {
/* Find the first zero scale factor and return an error code. */
for (i = 0; i < A->nrow; ++i)
if (r[i] == 0.) {
*info = i + 1;
return;
}
} else {
/* Invert the scale factors. */
for (i = 0; i < A->nrow; ++i)
r[i] = 1. / SUPERLU_MIN( SUPERLU_MAX( r[i], smlnum ), bignum );
/* Compute ROWCND = min(R(I)) / max(R(I)) */
*rowcnd = SUPERLU_MAX( rcmin, smlnum ) / SUPERLU_MIN( rcmax, bignum );
}
/* Compute column scale factors */
for (j = 0; j < A->ncol; ++j) c[j] = 0.;
/* Find the maximum element in each column, assuming the row
scalings computed above. */
irow = Astore->fst_row;
for (i = 0; i < m_loc; ++i) {
for (j = Astore->rowptr[i]; j < Astore->rowptr[i+1]; ++j) {
jcol = Astore->colind[j];
c[jcol] = SUPERLU_MAX( c[jcol], fabs(Aval[j]) * r[irow] );
}
++irow;
}
/* Find the global maximum for c[j] */
if ( !(loc_max = doubleMalloc_dist(A->ncol)))
ABORT("Malloc fails for loc_max[].");
for (j = 0; j < A->ncol; ++j) loc_max[j] = c[j];
MPI_Allreduce(loc_max, c, A->ncol, MPI_DOUBLE, MPI_MAX, grid->comm);
SUPERLU_FREE(loc_max);
/* Find the maximum and minimum scale factors. */
rcmin = bignum;
rcmax = 0.;
for (j = 0; j < A->ncol; ++j) {
rcmax = SUPERLU_MAX(rcmax, c[j]);
rcmin = SUPERLU_MIN(rcmin, c[j]);
}
if (rcmin == 0.) {
/* Find the first zero scale factor and return an error code. */
for (j = 0; j < A->ncol; ++j)
if ( c[j] == 0. ) {
*info = A->nrow + j + 1;
return;
}
} else {
/* Invert the scale factors. */
for (j = 0; j < A->ncol; ++j)
c[j] = 1. / SUPERLU_MIN( SUPERLU_MAX( c[j], smlnum ), bignum);
/* Compute COLCND = min(C(J)) / max(C(J)) */
*colcnd = SUPERLU_MAX( rcmin, smlnum ) / SUPERLU_MIN( rcmax, bignum );
}
/* gather R from each process to get the global R. */
procs = grid->nprow * grid->npcol;
if ( !(r_sizes = SUPERLU_MALLOC(2 * procs * sizeof(int))))
ABORT("Malloc fails for r_sizes[].");
displs = r_sizes + procs;
if ( !(loc_r = doubleMalloc_dist(m_loc)))
ABORT("Malloc fails for loc_r[].");
j = Astore->fst_row;
for (i = 0; i < m_loc; ++i) loc_r[i] = r[j++];
/* First gather the size of each piece. */
MPI_Allgather(&m_loc, 1, MPI_INT, r_sizes, 1, MPI_INT, grid->comm);
/* Set up the displacements for allgatherv */
displs[0] = 0;
for (i = 1; i < procs; ++i) displs[i] = displs[i-1] + r_sizes[i-1];
/* Now gather the actual data */
MPI_Allgatherv(loc_r, m_loc, MPI_DOUBLE, r, r_sizes, displs,
MPI_DOUBLE, grid->comm);
SUPERLU_FREE(r_sizes);
SUPERLU_FREE(loc_r);
return;
} /* pdgsequ */