src/glpmat.c
author Alpar Juttner <alpar@cs.elte.hu>
Mon, 06 Dec 2010 13:09:21 +0100
changeset 1 c445c931472f
permissions -rw-r--r--
Import glpk-4.45

- Generated files and doc/notes are removed
alpar@1
     1
/* glpmat.c */
alpar@1
     2
alpar@1
     3
/***********************************************************************
alpar@1
     4
*  This code is part of GLPK (GNU Linear Programming Kit).
alpar@1
     5
*
alpar@1
     6
*  Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008,
alpar@1
     7
*  2009, 2010 Andrew Makhorin, Department for Applied Informatics,
alpar@1
     8
*  Moscow Aviation Institute, Moscow, Russia. All rights reserved.
alpar@1
     9
*  E-mail: <mao@gnu.org>.
alpar@1
    10
*
alpar@1
    11
*  GLPK is free software: you can redistribute it and/or modify it
alpar@1
    12
*  under the terms of the GNU General Public License as published by
alpar@1
    13
*  the Free Software Foundation, either version 3 of the License, or
alpar@1
    14
*  (at your option) any later version.
alpar@1
    15
*
alpar@1
    16
*  GLPK is distributed in the hope that it will be useful, but WITHOUT
alpar@1
    17
*  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
alpar@1
    18
*  or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
alpar@1
    19
*  License for more details.
alpar@1
    20
*
alpar@1
    21
*  You should have received a copy of the GNU General Public License
alpar@1
    22
*  along with GLPK. If not, see <http://www.gnu.org/licenses/>.
alpar@1
    23
***********************************************************************/
alpar@1
    24
alpar@1
    25
#include "glpenv.h"
alpar@1
    26
#include "glpmat.h"
alpar@1
    27
#include "glpqmd.h"
alpar@1
    28
#include "amd/amd.h"
alpar@1
    29
#include "colamd/colamd.h"
alpar@1
    30
alpar@1
    31
/*----------------------------------------------------------------------
alpar@1
    32
-- check_fvs - check sparse vector in full-vector storage format.
alpar@1
    33
--
alpar@1
    34
-- SYNOPSIS
alpar@1
    35
--
alpar@1
    36
-- #include "glpmat.h"
alpar@1
    37
-- int check_fvs(int n, int nnz, int ind[], double vec[]);
alpar@1
    38
--
alpar@1
    39
-- DESCRIPTION
alpar@1
    40
--
alpar@1
    41
-- The routine check_fvs checks if a given vector of dimension n in
alpar@1
    42
-- full-vector storage format has correct representation.
alpar@1
    43
--
alpar@1
    44
-- RETURNS
alpar@1
    45
--
alpar@1
    46
-- The routine returns one of the following codes:
alpar@1
    47
--
alpar@1
    48
-- 0 - the vector is correct;
alpar@1
    49
-- 1 - the number of elements (n) is negative;
alpar@1
    50
-- 2 - the number of non-zero elements (nnz) is negative;
alpar@1
    51
-- 3 - some element index is out of range;
alpar@1
    52
-- 4 - some element index is duplicate;
alpar@1
    53
-- 5 - some non-zero element is out of pattern. */
alpar@1
    54
alpar@1
    55
int check_fvs(int n, int nnz, int ind[], double vec[])
alpar@1
    56
{     int i, t, ret, *flag = NULL;
alpar@1
    57
      /* check the number of elements */
alpar@1
    58
      if (n < 0)
alpar@1
    59
      {  ret = 1;
alpar@1
    60
         goto done;
alpar@1
    61
      }
alpar@1
    62
      /* check the number of non-zero elements */
alpar@1
    63
      if (nnz < 0)
alpar@1
    64
      {  ret = 2;
alpar@1
    65
         goto done;
alpar@1
    66
      }
alpar@1
    67
      /* check vector indices */
alpar@1
    68
      flag = xcalloc(1+n, sizeof(int));
alpar@1
    69
      for (i = 1; i <= n; i++) flag[i] = 0;
alpar@1
    70
      for (t = 1; t <= nnz; t++)
alpar@1
    71
      {  i = ind[t];
alpar@1
    72
         if (!(1 <= i && i <= n))
alpar@1
    73
         {  ret = 3;
alpar@1
    74
            goto done;
alpar@1
    75
         }
alpar@1
    76
         if (flag[i])
alpar@1
    77
         {  ret = 4;
alpar@1
    78
            goto done;
alpar@1
    79
         }
alpar@1
    80
         flag[i] = 1;
alpar@1
    81
      }
alpar@1
    82
      /* check vector elements */
alpar@1
    83
      for (i = 1; i <= n; i++)
alpar@1
    84
      {  if (!flag[i] && vec[i] != 0.0)
alpar@1
    85
         {  ret = 5;
alpar@1
    86
            goto done;
alpar@1
    87
         }
alpar@1
    88
      }
alpar@1
    89
      /* the vector is ok */
alpar@1
    90
      ret = 0;
alpar@1
    91
done: if (flag != NULL) xfree(flag);
alpar@1
    92
      return ret;
alpar@1
    93
}
alpar@1
    94
alpar@1
    95
/*----------------------------------------------------------------------
alpar@1
    96
-- check_pattern - check pattern of sparse matrix.
alpar@1
    97
--
alpar@1
    98
-- SYNOPSIS
alpar@1
    99
--
alpar@1
   100
-- #include "glpmat.h"
alpar@1
   101
-- int check_pattern(int m, int n, int A_ptr[], int A_ind[]);
alpar@1
   102
--
alpar@1
   103
-- DESCRIPTION
alpar@1
   104
--
alpar@1
   105
-- The routine check_pattern checks the pattern of a given mxn matrix
alpar@1
   106
-- in storage-by-rows format.
alpar@1
   107
--
alpar@1
   108
-- RETURNS
alpar@1
   109
--
alpar@1
   110
-- The routine returns one of the following codes:
alpar@1
   111
--
alpar@1
   112
-- 0 - the pattern is correct;
alpar@1
   113
-- 1 - the number of rows (m) is negative;
alpar@1
   114
-- 2 - the number of columns (n) is negative;
alpar@1
   115
-- 3 - A_ptr[1] is not 1;
alpar@1
   116
-- 4 - some column index is out of range;
alpar@1
   117
-- 5 - some column indices are duplicate. */
alpar@1
   118
alpar@1
   119
int check_pattern(int m, int n, int A_ptr[], int A_ind[])
alpar@1
   120
{     int i, j, ptr, ret, *flag = NULL;
alpar@1
   121
      /* check the number of rows */
alpar@1
   122
      if (m < 0)
alpar@1
   123
      {  ret = 1;
alpar@1
   124
         goto done;
alpar@1
   125
      }
alpar@1
   126
      /* check the number of columns */
alpar@1
   127
      if (n < 0)
alpar@1
   128
      {  ret = 2;
alpar@1
   129
         goto done;
alpar@1
   130
      }
alpar@1
   131
      /* check location A_ptr[1] */
alpar@1
   132
      if (A_ptr[1] != 1)
alpar@1
   133
      {  ret = 3;
alpar@1
   134
         goto done;
alpar@1
   135
      }
alpar@1
   136
      /* check row patterns */
alpar@1
   137
      flag = xcalloc(1+n, sizeof(int));
alpar@1
   138
      for (j = 1; j <= n; j++) flag[j] = 0;
alpar@1
   139
      for (i = 1; i <= m; i++)
alpar@1
   140
      {  /* check pattern of row i */
alpar@1
   141
         for (ptr = A_ptr[i]; ptr < A_ptr[i+1]; ptr++)
alpar@1
   142
         {  j = A_ind[ptr];
alpar@1
   143
            /* check column index */
alpar@1
   144
            if (!(1 <= j && j <= n))
alpar@1
   145
            {  ret = 4;
alpar@1
   146
               goto done;
alpar@1
   147
            }
alpar@1
   148
            /* check for duplication */
alpar@1
   149
            if (flag[j])
alpar@1
   150
            {  ret = 5;
alpar@1
   151
               goto done;
alpar@1
   152
            }
alpar@1
   153
            flag[j] = 1;
alpar@1
   154
         }
alpar@1
   155
         /* clear flags */
alpar@1
   156
         for (ptr = A_ptr[i]; ptr < A_ptr[i+1]; ptr++)
alpar@1
   157
         {  j = A_ind[ptr];
alpar@1
   158
            flag[j] = 0;
alpar@1
   159
         }
alpar@1
   160
      }
alpar@1
   161
      /* the pattern is ok */
alpar@1
   162
      ret = 0;
alpar@1
   163
done: if (flag != NULL) xfree(flag);
alpar@1
   164
      return ret;
alpar@1
   165
}
alpar@1
   166
alpar@1
   167
/*----------------------------------------------------------------------
alpar@1
   168
-- transpose - transpose sparse matrix.
alpar@1
   169
--
alpar@1
   170
-- *Synopsis*
alpar@1
   171
--
alpar@1
   172
-- #include "glpmat.h"
alpar@1
   173
-- void transpose(int m, int n, int A_ptr[], int A_ind[],
alpar@1
   174
--    double A_val[], int AT_ptr[], int AT_ind[], double AT_val[]);
alpar@1
   175
--
alpar@1
   176
-- *Description*
alpar@1
   177
--
alpar@1
   178
-- For a given mxn sparse matrix A the routine transpose builds a nxm
alpar@1
   179
-- sparse matrix A' which is a matrix transposed to A.
alpar@1
   180
--
alpar@1
   181
-- The arrays A_ptr, A_ind, and A_val specify a given mxn matrix A to
alpar@1
   182
-- be transposed in storage-by-rows format. The parameter A_val can be
alpar@1
   183
-- NULL, in which case numeric values are not copied. The arrays A_ptr,
alpar@1
   184
-- A_ind, and A_val are not changed on exit.
alpar@1
   185
--
alpar@1
   186
-- On entry the arrays AT_ptr, AT_ind, and AT_val must be allocated,
alpar@1
   187
-- but their content is ignored. On exit the routine stores a resultant
alpar@1
   188
-- nxm matrix A' in these arrays in storage-by-rows format. Note that
alpar@1
   189
-- if the parameter A_val is NULL, the array AT_val is not used.
alpar@1
   190
--
alpar@1
   191
-- The routine transpose has a side effect that elements in rows of the
alpar@1
   192
-- resultant matrix A' follow in ascending their column indices. */
alpar@1
   193
alpar@1
   194
void transpose(int m, int n, int A_ptr[], int A_ind[], double A_val[],
alpar@1
   195
      int AT_ptr[], int AT_ind[], double AT_val[])
alpar@1
   196
{     int i, j, t, beg, end, pos, len;
alpar@1
   197
      /* determine row lengths of resultant matrix */
alpar@1
   198
      for (j = 1; j <= n; j++) AT_ptr[j] = 0;
alpar@1
   199
      for (i = 1; i <= m; i++)
alpar@1
   200
      {  beg = A_ptr[i], end = A_ptr[i+1];
alpar@1
   201
         for (t = beg; t < end; t++) AT_ptr[A_ind[t]]++;
alpar@1
   202
      }
alpar@1
   203
      /* set up row pointers of resultant matrix */
alpar@1
   204
      pos = 1;
alpar@1
   205
      for (j = 1; j <= n; j++)
alpar@1
   206
         len = AT_ptr[j], pos += len, AT_ptr[j] = pos;
alpar@1
   207
      AT_ptr[n+1] = pos;
alpar@1
   208
      /* build resultant matrix */
alpar@1
   209
      for (i = m; i >= 1; i--)
alpar@1
   210
      {  beg = A_ptr[i], end = A_ptr[i+1];
alpar@1
   211
         for (t = beg; t < end; t++)
alpar@1
   212
         {  pos = --AT_ptr[A_ind[t]];
alpar@1
   213
            AT_ind[pos] = i;
alpar@1
   214
            if (A_val != NULL) AT_val[pos] = A_val[t];
alpar@1
   215
         }
alpar@1
   216
      }
alpar@1
   217
      return;
alpar@1
   218
}
alpar@1
   219
alpar@1
   220
/*----------------------------------------------------------------------
alpar@1
   221
-- adat_symbolic - compute S = P*A*D*A'*P' (symbolic phase).
alpar@1
   222
--
alpar@1
   223
-- *Synopsis*
alpar@1
   224
--
alpar@1
   225
-- #include "glpmat.h"
alpar@1
   226
-- int *adat_symbolic(int m, int n, int P_per[], int A_ptr[],
alpar@1
   227
--    int A_ind[], int S_ptr[]);
alpar@1
   228
--
alpar@1
   229
-- *Description*
alpar@1
   230
--
alpar@1
   231
-- The routine adat_symbolic implements the symbolic phase to compute
alpar@1
   232
-- symmetric matrix S = P*A*D*A'*P', where P is a permutation matrix,
alpar@1
   233
-- A is a given sparse matrix, D is a diagonal matrix, A' is a matrix
alpar@1
   234
-- transposed to A, P' is an inverse of P.
alpar@1
   235
--
alpar@1
   236
-- The parameter m is the number of rows in A and the order of P.
alpar@1
   237
--
alpar@1
   238
-- The parameter n is the number of columns in A and the order of D.
alpar@1
   239
--
alpar@1
   240
-- The array P_per specifies permutation matrix P. It is not changed on
alpar@1
   241
-- exit.
alpar@1
   242
--
alpar@1
   243
-- The arrays A_ptr and A_ind specify the pattern of matrix A. They are
alpar@1
   244
-- not changed on exit.
alpar@1
   245
--
alpar@1
   246
-- On exit the routine stores the pattern of upper triangular part of
alpar@1
   247
-- matrix S without diagonal elements in the arrays S_ptr and S_ind in
alpar@1
   248
-- storage-by-rows format. The array S_ptr should be allocated on entry,
alpar@1
   249
-- however, its content is ignored. The array S_ind is allocated by the
alpar@1
   250
-- routine itself which returns a pointer to it.
alpar@1
   251
--
alpar@1
   252
-- *Returns*
alpar@1
   253
--
alpar@1
   254
-- The routine returns a pointer to the array S_ind. */
alpar@1
   255
alpar@1
   256
int *adat_symbolic(int m, int n, int P_per[], int A_ptr[], int A_ind[],
alpar@1
   257
      int S_ptr[])
alpar@1
   258
{     int i, j, t, ii, jj, tt, k, size, len;
alpar@1
   259
      int *S_ind, *AT_ptr, *AT_ind, *ind, *map, *temp;
alpar@1
   260
      /* build the pattern of A', which is a matrix transposed to A, to
alpar@1
   261
         efficiently access A in column-wise manner */
alpar@1
   262
      AT_ptr = xcalloc(1+n+1, sizeof(int));
alpar@1
   263
      AT_ind = xcalloc(A_ptr[m+1], sizeof(int));
alpar@1
   264
      transpose(m, n, A_ptr, A_ind, NULL, AT_ptr, AT_ind, NULL);
alpar@1
   265
      /* allocate the array S_ind */
alpar@1
   266
      size = A_ptr[m+1] - 1;
alpar@1
   267
      if (size < m) size = m;
alpar@1
   268
      S_ind = xcalloc(1+size, sizeof(int));
alpar@1
   269
      /* allocate and initialize working arrays */
alpar@1
   270
      ind = xcalloc(1+m, sizeof(int));
alpar@1
   271
      map = xcalloc(1+m, sizeof(int));
alpar@1
   272
      for (jj = 1; jj <= m; jj++) map[jj] = 0;
alpar@1
   273
      /* compute pattern of S; note that symbolically S = B*B', where
alpar@1
   274
         B = P*A, B' is matrix transposed to B */
alpar@1
   275
      S_ptr[1] = 1;
alpar@1
   276
      for (ii = 1; ii <= m; ii++)
alpar@1
   277
      {  /* compute pattern of ii-th row of S */
alpar@1
   278
         len = 0;
alpar@1
   279
         i = P_per[ii]; /* i-th row of A = ii-th row of B */
alpar@1
   280
         for (t = A_ptr[i]; t < A_ptr[i+1]; t++)
alpar@1
   281
         {  k = A_ind[t];
alpar@1
   282
            /* walk through k-th column of A */
alpar@1
   283
            for (tt = AT_ptr[k]; tt < AT_ptr[k+1]; tt++)
alpar@1
   284
            {  j = AT_ind[tt];
alpar@1
   285
               jj = P_per[m+j]; /* j-th row of A = jj-th row of B */
alpar@1
   286
               /* a[i,k] != 0 and a[j,k] != 0 ergo s[ii,jj] != 0 */
alpar@1
   287
               if (ii < jj && !map[jj]) ind[++len] = jj, map[jj] = 1;
alpar@1
   288
            }
alpar@1
   289
         }
alpar@1
   290
         /* now (ind) is pattern of ii-th row of S */
alpar@1
   291
         S_ptr[ii+1] = S_ptr[ii] + len;
alpar@1
   292
         /* at least (S_ptr[ii+1] - 1) locations should be available in
alpar@1
   293
            the array S_ind */
alpar@1
   294
         if (S_ptr[ii+1] - 1 > size)
alpar@1
   295
         {  temp = S_ind;
alpar@1
   296
            size += size;
alpar@1
   297
            S_ind = xcalloc(1+size, sizeof(int));
alpar@1
   298
            memcpy(&S_ind[1], &temp[1], (S_ptr[ii] - 1) * sizeof(int));
alpar@1
   299
            xfree(temp);
alpar@1
   300
         }
alpar@1
   301
         xassert(S_ptr[ii+1] - 1 <= size);
alpar@1
   302
         /* (ii-th row of S) := (ind) */
alpar@1
   303
         memcpy(&S_ind[S_ptr[ii]], &ind[1], len * sizeof(int));
alpar@1
   304
         /* clear the row pattern map */
alpar@1
   305
         for (t = 1; t <= len; t++) map[ind[t]] = 0;
alpar@1
   306
      }
alpar@1
   307
      /* free working arrays */
alpar@1
   308
      xfree(AT_ptr);
alpar@1
   309
      xfree(AT_ind);
alpar@1
   310
      xfree(ind);
alpar@1
   311
      xfree(map);
alpar@1
   312
      /* reallocate the array S_ind to free unused locations */
alpar@1
   313
      temp = S_ind;
alpar@1
   314
      size = S_ptr[m+1] - 1;
alpar@1
   315
      S_ind = xcalloc(1+size, sizeof(int));
alpar@1
   316
      memcpy(&S_ind[1], &temp[1], size * sizeof(int));
alpar@1
   317
      xfree(temp);
alpar@1
   318
      return S_ind;
alpar@1
   319
}
alpar@1
   320
alpar@1
   321
/*----------------------------------------------------------------------
alpar@1
   322
-- adat_numeric - compute S = P*A*D*A'*P' (numeric phase).
alpar@1
   323
--
alpar@1
   324
-- *Synopsis*
alpar@1
   325
--
alpar@1
   326
-- #include "glpmat.h"
alpar@1
   327
-- void adat_numeric(int m, int n, int P_per[],
alpar@1
   328
--    int A_ptr[], int A_ind[], double A_val[], double D_diag[],
alpar@1
   329
--    int S_ptr[], int S_ind[], double S_val[], double S_diag[]);
alpar@1
   330
--
alpar@1
   331
-- *Description*
alpar@1
   332
--
alpar@1
   333
-- The routine adat_numeric implements the numeric phase to compute
alpar@1
   334
-- symmetric matrix S = P*A*D*A'*P', where P is a permutation matrix,
alpar@1
   335
-- A is a given sparse matrix, D is a diagonal matrix, A' is a matrix
alpar@1
   336
-- transposed to A, P' is an inverse of P.
alpar@1
   337
--
alpar@1
   338
-- The parameter m is the number of rows in A and the order of P.
alpar@1
   339
--
alpar@1
   340
-- The parameter n is the number of columns in A and the order of D.
alpar@1
   341
--
alpar@1
   342
-- The matrix P is specified in the array P_per, which is not changed
alpar@1
   343
-- on exit.
alpar@1
   344
--
alpar@1
   345
-- The matrix A is specified in the arrays A_ptr, A_ind, and A_val in
alpar@1
   346
-- storage-by-rows format. These arrays are not changed on exit.
alpar@1
   347
--
alpar@1
   348
-- Diagonal elements of the matrix D are specified in the array D_diag,
alpar@1
   349
-- where D_diag[0] is not used, D_diag[i] = d[i,i] for i = 1, ..., n.
alpar@1
   350
-- The array D_diag is not changed on exit.
alpar@1
   351
--
alpar@1
   352
-- The pattern of the upper triangular part of the matrix S without
alpar@1
   353
-- diagonal elements (previously computed by the routine adat_symbolic)
alpar@1
   354
-- is specified in the arrays S_ptr and S_ind, which are not changed on
alpar@1
   355
-- exit. Numeric values of non-diagonal elements of S are stored in
alpar@1
   356
-- corresponding locations of the array S_val, and values of diagonal
alpar@1
   357
-- elements of S are stored in locations S_diag[1], ..., S_diag[n]. */
alpar@1
   358
alpar@1
   359
void adat_numeric(int m, int n, int P_per[],
alpar@1
   360
      int A_ptr[], int A_ind[], double A_val[], double D_diag[],
alpar@1
   361
      int S_ptr[], int S_ind[], double S_val[], double S_diag[])
alpar@1
   362
{     int i, j, t, ii, jj, tt, beg, end, beg1, end1, k;
alpar@1
   363
      double sum, *work;
alpar@1
   364
      work = xcalloc(1+n, sizeof(double));
alpar@1
   365
      for (j = 1; j <= n; j++) work[j] = 0.0;
alpar@1
   366
      /* compute S = B*D*B', where B = P*A, B' is a matrix transposed
alpar@1
   367
         to B */
alpar@1
   368
      for (ii = 1; ii <= m; ii++)
alpar@1
   369
      {  i = P_per[ii]; /* i-th row of A = ii-th row of B */
alpar@1
   370
         /* (work) := (i-th row of A) */
alpar@1
   371
         beg = A_ptr[i], end = A_ptr[i+1];
alpar@1
   372
         for (t = beg; t < end; t++)
alpar@1
   373
            work[A_ind[t]] = A_val[t];
alpar@1
   374
         /* compute ii-th row of S */
alpar@1
   375
         beg = S_ptr[ii], end = S_ptr[ii+1];
alpar@1
   376
         for (t = beg; t < end; t++)
alpar@1
   377
         {  jj = S_ind[t];
alpar@1
   378
            j = P_per[jj]; /* j-th row of A = jj-th row of B */
alpar@1
   379
            /* s[ii,jj] := sum a[i,k] * d[k,k] * a[j,k] */
alpar@1
   380
            sum = 0.0;
alpar@1
   381
            beg1 = A_ptr[j], end1 = A_ptr[j+1];
alpar@1
   382
            for (tt = beg1; tt < end1; tt++)
alpar@1
   383
            {  k = A_ind[tt];
alpar@1
   384
               sum += work[k] * D_diag[k] * A_val[tt];
alpar@1
   385
            }
alpar@1
   386
            S_val[t] = sum;
alpar@1
   387
         }
alpar@1
   388
         /* s[ii,ii] := sum a[i,k] * d[k,k] * a[i,k] */
alpar@1
   389
         sum = 0.0;
alpar@1
   390
         beg = A_ptr[i], end = A_ptr[i+1];
alpar@1
   391
         for (t = beg; t < end; t++)
alpar@1
   392
         {  k = A_ind[t];
alpar@1
   393
            sum += A_val[t] * D_diag[k] * A_val[t];
alpar@1
   394
            work[k] = 0.0;
alpar@1
   395
         }
alpar@1
   396
         S_diag[ii] = sum;
alpar@1
   397
      }
alpar@1
   398
      xfree(work);
alpar@1
   399
      return;
alpar@1
   400
}
alpar@1
   401
alpar@1
   402
/*----------------------------------------------------------------------
alpar@1
   403
-- min_degree - minimum degree ordering.
alpar@1
   404
--
alpar@1
   405
-- *Synopsis*
alpar@1
   406
--
alpar@1
   407
-- #include "glpmat.h"
alpar@1
   408
-- void min_degree(int n, int A_ptr[], int A_ind[], int P_per[]);
alpar@1
   409
--
alpar@1
   410
-- *Description*
alpar@1
   411
--
alpar@1
   412
-- The routine min_degree uses the minimum degree ordering algorithm
alpar@1
   413
-- to find a permutation matrix P for a given sparse symmetric positive
alpar@1
   414
-- matrix A which minimizes the number of non-zeros in upper triangular
alpar@1
   415
-- factor U for Cholesky factorization P*A*P' = U'*U.
alpar@1
   416
--
alpar@1
   417
-- The parameter n is the order of matrices A and P.
alpar@1
   418
--
alpar@1
   419
-- The pattern of the given matrix A is specified on entry in the arrays
alpar@1
   420
-- A_ptr and A_ind in storage-by-rows format. Only the upper triangular
alpar@1
   421
-- part without diagonal elements (which all are assumed to be non-zero)
alpar@1
   422
-- should be specified as if A were upper triangular. The arrays A_ptr
alpar@1
   423
-- and A_ind are not changed on exit.
alpar@1
   424
--
alpar@1
   425
-- The permutation matrix P is stored by the routine in the array P_per
alpar@1
   426
-- on exit.
alpar@1
   427
--
alpar@1
   428
-- *Algorithm*
alpar@1
   429
--
alpar@1
   430
-- The routine min_degree is based on some subroutines from the package
alpar@1
   431
-- SPARSPAK (see comments in the module glpqmd). */
alpar@1
   432
alpar@1
   433
void min_degree(int n, int A_ptr[], int A_ind[], int P_per[])
alpar@1
   434
{     int i, j, ne, t, pos, len;
alpar@1
   435
      int *xadj, *adjncy, *deg, *marker, *rchset, *nbrhd, *qsize,
alpar@1
   436
         *qlink, nofsub;
alpar@1
   437
      /* determine number of non-zeros in complete pattern */
alpar@1
   438
      ne = A_ptr[n+1] - 1;
alpar@1
   439
      ne += ne;
alpar@1
   440
      /* allocate working arrays */
alpar@1
   441
      xadj = xcalloc(1+n+1, sizeof(int));
alpar@1
   442
      adjncy = xcalloc(1+ne, sizeof(int));
alpar@1
   443
      deg = xcalloc(1+n, sizeof(int));
alpar@1
   444
      marker = xcalloc(1+n, sizeof(int));
alpar@1
   445
      rchset = xcalloc(1+n, sizeof(int));
alpar@1
   446
      nbrhd = xcalloc(1+n, sizeof(int));
alpar@1
   447
      qsize = xcalloc(1+n, sizeof(int));
alpar@1
   448
      qlink = xcalloc(1+n, sizeof(int));
alpar@1
   449
      /* determine row lengths in complete pattern */
alpar@1
   450
      for (i = 1; i <= n; i++) xadj[i] = 0;
alpar@1
   451
      for (i = 1; i <= n; i++)
alpar@1
   452
      {  for (t = A_ptr[i]; t < A_ptr[i+1]; t++)
alpar@1
   453
         {  j = A_ind[t];
alpar@1
   454
            xassert(i < j && j <= n);
alpar@1
   455
            xadj[i]++, xadj[j]++;
alpar@1
   456
         }
alpar@1
   457
      }
alpar@1
   458
      /* set up row pointers for complete pattern */
alpar@1
   459
      pos = 1;
alpar@1
   460
      for (i = 1; i <= n; i++)
alpar@1
   461
         len = xadj[i], pos += len, xadj[i] = pos;
alpar@1
   462
      xadj[n+1] = pos;
alpar@1
   463
      xassert(pos - 1 == ne);
alpar@1
   464
      /* construct complete pattern */
alpar@1
   465
      for (i = 1; i <= n; i++)
alpar@1
   466
      {  for (t = A_ptr[i]; t < A_ptr[i+1]; t++)
alpar@1
   467
         {  j = A_ind[t];
alpar@1
   468
            adjncy[--xadj[i]] = j, adjncy[--xadj[j]] = i;
alpar@1
   469
         }
alpar@1
   470
      }
alpar@1
   471
      /* call the main minimimum degree ordering routine */
alpar@1
   472
      genqmd(&n, xadj, adjncy, P_per, P_per + n, deg, marker, rchset,
alpar@1
   473
         nbrhd, qsize, qlink, &nofsub);
alpar@1
   474
      /* make sure that permutation matrix P is correct */
alpar@1
   475
      for (i = 1; i <= n; i++)
alpar@1
   476
      {  j = P_per[i];
alpar@1
   477
         xassert(1 <= j && j <= n);
alpar@1
   478
         xassert(P_per[n+j] == i);
alpar@1
   479
      }
alpar@1
   480
      /* free working arrays */
alpar@1
   481
      xfree(xadj);
alpar@1
   482
      xfree(adjncy);
alpar@1
   483
      xfree(deg);
alpar@1
   484
      xfree(marker);
alpar@1
   485
      xfree(rchset);
alpar@1
   486
      xfree(nbrhd);
alpar@1
   487
      xfree(qsize);
alpar@1
   488
      xfree(qlink);
alpar@1
   489
      return;
alpar@1
   490
}
alpar@1
   491
alpar@1
   492
/**********************************************************************/
alpar@1
   493
alpar@1
   494
void amd_order1(int n, int A_ptr[], int A_ind[], int P_per[])
alpar@1
   495
{     /* approximate minimum degree ordering (AMD) */
alpar@1
   496
      int k, ret;
alpar@1
   497
      double Control[AMD_CONTROL], Info[AMD_INFO];
alpar@1
   498
      /* get the default parameters */
alpar@1
   499
      amd_defaults(Control);
alpar@1
   500
#if 0
alpar@1
   501
      /* and print them */
alpar@1
   502
      amd_control(Control);
alpar@1
   503
#endif
alpar@1
   504
      /* make all indices 0-based */
alpar@1
   505
      for (k = 1; k < A_ptr[n+1]; k++) A_ind[k]--;
alpar@1
   506
      for (k = 1; k <= n+1; k++) A_ptr[k]--;
alpar@1
   507
      /* call the ordering routine */
alpar@1
   508
      ret = amd_order(n, &A_ptr[1], &A_ind[1], &P_per[1], Control, Info)
alpar@1
   509
         ;
alpar@1
   510
#if 0
alpar@1
   511
      amd_info(Info);
alpar@1
   512
#endif
alpar@1
   513
      xassert(ret == AMD_OK || ret == AMD_OK_BUT_JUMBLED);
alpar@1
   514
      /* retsore 1-based indices */
alpar@1
   515
      for (k = 1; k <= n+1; k++) A_ptr[k]++;
alpar@1
   516
      for (k = 1; k < A_ptr[n+1]; k++) A_ind[k]++;
alpar@1
   517
      /* patch up permutation matrix */
alpar@1
   518
      memset(&P_per[n+1], 0, n * sizeof(int));
alpar@1
   519
      for (k = 1; k <= n; k++)
alpar@1
   520
      {  P_per[k]++;
alpar@1
   521
         xassert(1 <= P_per[k] && P_per[k] <= n);
alpar@1
   522
         xassert(P_per[n+P_per[k]] == 0);
alpar@1
   523
         P_per[n+P_per[k]] = k;
alpar@1
   524
      }
alpar@1
   525
      return;
alpar@1
   526
}
alpar@1
   527
alpar@1
   528
/**********************************************************************/
alpar@1
   529
alpar@1
   530
static void *allocate(size_t n, size_t size)
alpar@1
   531
{     void *ptr;
alpar@1
   532
      ptr = xcalloc(n, size);
alpar@1
   533
      memset(ptr, 0, n * size);
alpar@1
   534
      return ptr;
alpar@1
   535
}
alpar@1
   536
alpar@1
   537
static void release(void *ptr)
alpar@1
   538
{     xfree(ptr);
alpar@1
   539
      return;
alpar@1
   540
}
alpar@1
   541
alpar@1
   542
void symamd_ord(int n, int A_ptr[], int A_ind[], int P_per[])
alpar@1
   543
{     /* approximate minimum degree ordering (SYMAMD) */
alpar@1
   544
      int k, ok;
alpar@1
   545
      int stats[COLAMD_STATS];
alpar@1
   546
      /* make all indices 0-based */
alpar@1
   547
      for (k = 1; k < A_ptr[n+1]; k++) A_ind[k]--;
alpar@1
   548
      for (k = 1; k <= n+1; k++) A_ptr[k]--;
alpar@1
   549
      /* call the ordering routine */
alpar@1
   550
      ok = symamd(n, &A_ind[1], &A_ptr[1], &P_per[1], NULL, stats,
alpar@1
   551
         allocate, release);
alpar@1
   552
#if 0
alpar@1
   553
      symamd_report(stats);
alpar@1
   554
#endif
alpar@1
   555
      xassert(ok);
alpar@1
   556
      /* restore 1-based indices */
alpar@1
   557
      for (k = 1; k <= n+1; k++) A_ptr[k]++;
alpar@1
   558
      for (k = 1; k < A_ptr[n+1]; k++) A_ind[k]++;
alpar@1
   559
      /* patch up permutation matrix */
alpar@1
   560
      memset(&P_per[n+1], 0, n * sizeof(int));
alpar@1
   561
      for (k = 1; k <= n; k++)
alpar@1
   562
      {  P_per[k]++;
alpar@1
   563
         xassert(1 <= P_per[k] && P_per[k] <= n);
alpar@1
   564
         xassert(P_per[n+P_per[k]] == 0);
alpar@1
   565
         P_per[n+P_per[k]] = k;
alpar@1
   566
      }
alpar@1
   567
      return;
alpar@1
   568
}
alpar@1
   569
alpar@1
   570
/*----------------------------------------------------------------------
alpar@1
   571
-- chol_symbolic - compute Cholesky factorization (symbolic phase).
alpar@1
   572
--
alpar@1
   573
-- *Synopsis*
alpar@1
   574
--
alpar@1
   575
-- #include "glpmat.h"
alpar@1
   576
-- int *chol_symbolic(int n, int A_ptr[], int A_ind[], int U_ptr[]);
alpar@1
   577
--
alpar@1
   578
-- *Description*
alpar@1
   579
--
alpar@1
   580
-- The routine chol_symbolic implements the symbolic phase of Cholesky
alpar@1
   581
-- factorization A = U'*U, where A is a given sparse symmetric positive
alpar@1
   582
-- definite matrix, U is a resultant upper triangular factor, U' is a
alpar@1
   583
-- matrix transposed to U.
alpar@1
   584
--
alpar@1
   585
-- The parameter n is the order of matrices A and U.
alpar@1
   586
--
alpar@1
   587
-- The pattern of the given matrix A is specified on entry in the arrays
alpar@1
   588
-- A_ptr and A_ind in storage-by-rows format. Only the upper triangular
alpar@1
   589
-- part without diagonal elements (which all are assumed to be non-zero)
alpar@1
   590
-- should be specified as if A were upper triangular. The arrays A_ptr
alpar@1
   591
-- and A_ind are not changed on exit.
alpar@1
   592
--
alpar@1
   593
-- The pattern of the matrix U without diagonal elements (which all are
alpar@1
   594
-- assumed to be non-zero) is stored on exit from the routine in the
alpar@1
   595
-- arrays U_ptr and U_ind in storage-by-rows format. The array U_ptr
alpar@1
   596
-- should be allocated on entry, however, its content is ignored. The
alpar@1
   597
-- array U_ind is allocated by the routine which returns a pointer to it
alpar@1
   598
-- on exit.
alpar@1
   599
--
alpar@1
   600
-- *Returns*
alpar@1
   601
--
alpar@1
   602
-- The routine returns a pointer to the array U_ind.
alpar@1
   603
--
alpar@1
   604
-- *Method*
alpar@1
   605
--
alpar@1
   606
-- The routine chol_symbolic computes the pattern of the matrix U in a
alpar@1
   607
-- row-wise manner. No pivoting is used.
alpar@1
   608
--
alpar@1
   609
-- It is known that to compute the pattern of row k of the matrix U we
alpar@1
   610
-- need to merge the pattern of row k of the matrix A and the patterns
alpar@1
   611
-- of each row i of U, where u[i,k] is non-zero (these rows are already
alpar@1
   612
-- computed and placed above row k).
alpar@1
   613
--
alpar@1
   614
-- However, to reduce the number of rows to be merged the routine uses
alpar@1
   615
-- an advanced algorithm proposed in:
alpar@1
   616
--
alpar@1
   617
-- D.J.Rose, R.E.Tarjan, and G.S.Lueker. Algorithmic aspects of vertex
alpar@1
   618
-- elimination on graphs. SIAM J. Comput. 5, 1976, 266-83.
alpar@1
   619
--
alpar@1
   620
-- The authors of the cited paper show that we have the same result if
alpar@1
   621
-- we merge row k of the matrix A and such rows of the matrix U (among
alpar@1
   622
-- rows 1, ..., k-1) whose leftmost non-diagonal non-zero element is
alpar@1
   623
-- placed in k-th column. This feature signficantly reduces the number
alpar@1
   624
-- of rows to be merged, especially on the final steps, where rows of
alpar@1
   625
-- the matrix U become quite dense.
alpar@1
   626
--
alpar@1
   627
-- To determine rows, which should be merged on k-th step, for a fixed
alpar@1
   628
-- time the routine uses linked lists of row numbers of the matrix U.
alpar@1
   629
-- Location head[k] contains the number of a first row, whose leftmost
alpar@1
   630
-- non-diagonal non-zero element is placed in column k, and location
alpar@1
   631
-- next[i] contains the number of a next row with the same property as
alpar@1
   632
-- row i. */
alpar@1
   633
alpar@1
   634
int *chol_symbolic(int n, int A_ptr[], int A_ind[], int U_ptr[])
alpar@1
   635
{     int i, j, k, t, len, size, beg, end, min_j, *U_ind, *head, *next,
alpar@1
   636
         *ind, *map, *temp;
alpar@1
   637
      /* initially we assume that on computing the pattern of U fill-in
alpar@1
   638
         will double the number of non-zeros in A */
alpar@1
   639
      size = A_ptr[n+1] - 1;
alpar@1
   640
      if (size < n) size = n;
alpar@1
   641
      size += size;
alpar@1
   642
      U_ind = xcalloc(1+size, sizeof(int));
alpar@1
   643
      /* allocate and initialize working arrays */
alpar@1
   644
      head = xcalloc(1+n, sizeof(int));
alpar@1
   645
      for (i = 1; i <= n; i++) head[i] = 0;
alpar@1
   646
      next = xcalloc(1+n, sizeof(int));
alpar@1
   647
      ind = xcalloc(1+n, sizeof(int));
alpar@1
   648
      map = xcalloc(1+n, sizeof(int));
alpar@1
   649
      for (j = 1; j <= n; j++) map[j] = 0;
alpar@1
   650
      /* compute the pattern of matrix U */
alpar@1
   651
      U_ptr[1] = 1;
alpar@1
   652
      for (k = 1; k <= n; k++)
alpar@1
   653
      {  /* compute the pattern of k-th row of U, which is the union of
alpar@1
   654
            k-th row of A and those rows of U (among 1, ..., k-1) whose
alpar@1
   655
            leftmost non-diagonal non-zero is placed in k-th column */
alpar@1
   656
         /* (ind) := (k-th row of A) */
alpar@1
   657
         len = A_ptr[k+1] - A_ptr[k];
alpar@1
   658
         memcpy(&ind[1], &A_ind[A_ptr[k]], len * sizeof(int));
alpar@1
   659
         for (t = 1; t <= len; t++)
alpar@1
   660
         {  j = ind[t];
alpar@1
   661
            xassert(k < j && j <= n);
alpar@1
   662
            map[j] = 1;
alpar@1
   663
         }
alpar@1
   664
         /* walk through rows of U whose leftmost non-diagonal non-zero
alpar@1
   665
            is placed in k-th column */
alpar@1
   666
         for (i = head[k]; i != 0; i = next[i])
alpar@1
   667
         {  /* (ind) := (ind) union (i-th row of U) */
alpar@1
   668
            beg = U_ptr[i], end = U_ptr[i+1];
alpar@1
   669
            for (t = beg; t < end; t++)
alpar@1
   670
            {  j = U_ind[t];
alpar@1
   671
               if (j > k && !map[j]) ind[++len] = j, map[j] = 1;
alpar@1
   672
            }
alpar@1
   673
         }
alpar@1
   674
         /* now (ind) is the pattern of k-th row of U */
alpar@1
   675
         U_ptr[k+1] = U_ptr[k] + len;
alpar@1
   676
         /* at least (U_ptr[k+1] - 1) locations should be available in
alpar@1
   677
            the array U_ind */
alpar@1
   678
         if (U_ptr[k+1] - 1 > size)
alpar@1
   679
         {  temp = U_ind;
alpar@1
   680
            size += size;
alpar@1
   681
            U_ind = xcalloc(1+size, sizeof(int));
alpar@1
   682
            memcpy(&U_ind[1], &temp[1], (U_ptr[k] - 1) * sizeof(int));
alpar@1
   683
            xfree(temp);
alpar@1
   684
         }
alpar@1
   685
         xassert(U_ptr[k+1] - 1 <= size);
alpar@1
   686
         /* (k-th row of U) := (ind) */
alpar@1
   687
         memcpy(&U_ind[U_ptr[k]], &ind[1], len * sizeof(int));
alpar@1
   688
         /* determine column index of leftmost non-diagonal non-zero in
alpar@1
   689
            k-th row of U and clear the row pattern map */
alpar@1
   690
         min_j = n + 1;
alpar@1
   691
         for (t = 1; t <= len; t++)
alpar@1
   692
         {  j = ind[t], map[j] = 0;
alpar@1
   693
            if (min_j > j) min_j = j;
alpar@1
   694
         }
alpar@1
   695
         /* include k-th row into corresponding linked list */
alpar@1
   696
         if (min_j <= n) next[k] = head[min_j], head[min_j] = k;
alpar@1
   697
      }
alpar@1
   698
      /* free working arrays */
alpar@1
   699
      xfree(head);
alpar@1
   700
      xfree(next);
alpar@1
   701
      xfree(ind);
alpar@1
   702
      xfree(map);
alpar@1
   703
      /* reallocate the array U_ind to free unused locations */
alpar@1
   704
      temp = U_ind;
alpar@1
   705
      size = U_ptr[n+1] - 1;
alpar@1
   706
      U_ind = xcalloc(1+size, sizeof(int));
alpar@1
   707
      memcpy(&U_ind[1], &temp[1], size * sizeof(int));
alpar@1
   708
      xfree(temp);
alpar@1
   709
      return U_ind;
alpar@1
   710
}
alpar@1
   711
alpar@1
   712
/*----------------------------------------------------------------------
alpar@1
   713
-- chol_numeric - compute Cholesky factorization (numeric phase).
alpar@1
   714
--
alpar@1
   715
-- *Synopsis*
alpar@1
   716
--
alpar@1
   717
-- #include "glpmat.h"
alpar@1
   718
-- int chol_numeric(int n,
alpar@1
   719
--    int A_ptr[], int A_ind[], double A_val[], double A_diag[],
alpar@1
   720
--    int U_ptr[], int U_ind[], double U_val[], double U_diag[]);
alpar@1
   721
--
alpar@1
   722
-- *Description*
alpar@1
   723
--
alpar@1
   724
-- The routine chol_symbolic implements the numeric phase of Cholesky
alpar@1
   725
-- factorization A = U'*U, where A is a given sparse symmetric positive
alpar@1
   726
-- definite matrix, U is a resultant upper triangular factor, U' is a
alpar@1
   727
-- matrix transposed to U.
alpar@1
   728
--
alpar@1
   729
-- The parameter n is the order of matrices A and U.
alpar@1
   730
--
alpar@1
   731
-- Upper triangular part of the matrix A without diagonal elements is
alpar@1
   732
-- specified in the arrays A_ptr, A_ind, and A_val in storage-by-rows
alpar@1
   733
-- format. Diagonal elements of A are specified in the array A_diag,
alpar@1
   734
-- where A_diag[0] is not used, A_diag[i] = a[i,i] for i = 1, ..., n.
alpar@1
   735
-- The arrays A_ptr, A_ind, A_val, and A_diag are not changed on exit.
alpar@1
   736
--
alpar@1
   737
-- The pattern of the matrix U without diagonal elements (previously
alpar@1
   738
-- computed with the routine chol_symbolic) is specified in the arrays
alpar@1
   739
-- U_ptr and U_ind, which are not changed on exit. Numeric values of
alpar@1
   740
-- non-diagonal elements of U are stored in corresponding locations of
alpar@1
   741
-- the array U_val, and values of diagonal elements of U are stored in
alpar@1
   742
-- locations U_diag[1], ..., U_diag[n].
alpar@1
   743
--
alpar@1
   744
-- *Returns*
alpar@1
   745
--
alpar@1
   746
-- The routine returns the number of non-positive diagonal elements of
alpar@1
   747
-- the matrix U which have been replaced by a huge positive number (see
alpar@1
   748
-- the method description below). Zero return code means the matrix A
alpar@1
   749
-- has been successfully factorized.
alpar@1
   750
--
alpar@1
   751
-- *Method*
alpar@1
   752
--
alpar@1
   753
-- The routine chol_numeric computes the matrix U in a row-wise manner
alpar@1
   754
-- using standard gaussian elimination technique. No pivoting is used.
alpar@1
   755
--
alpar@1
   756
-- Initially the routine sets U = A, and before k-th elimination step
alpar@1
   757
-- the matrix U is the following:
alpar@1
   758
--
alpar@1
   759
--       1       k         n
alpar@1
   760
--    1  x x x x x x x x x x
alpar@1
   761
--       . x x x x x x x x x
alpar@1
   762
--       . . x x x x x x x x
alpar@1
   763
--       . . . x x x x x x x
alpar@1
   764
--    k  . . . . * * * * * *
alpar@1
   765
--       . . . . * * * * * *
alpar@1
   766
--       . . . . * * * * * *
alpar@1
   767
--       . . . . * * * * * *
alpar@1
   768
--       . . . . * * * * * *
alpar@1
   769
--    n  . . . . * * * * * *
alpar@1
   770
--
alpar@1
   771
-- where 'x' are elements of already computed rows, '*' are elements of
alpar@1
   772
-- the active submatrix. (Note that the lower triangular part of the
alpar@1
   773
-- active submatrix being symmetric is not stored and diagonal elements
alpar@1
   774
-- are stored separately in the array U_diag.)
alpar@1
   775
--
alpar@1
   776
-- The matrix A is assumed to be positive definite. However, if it is
alpar@1
   777
-- close to semi-definite, on some elimination step a pivot u[k,k] may
alpar@1
   778
-- happen to be non-positive due to round-off errors. In this case the
alpar@1
   779
-- routine uses a technique proposed in:
alpar@1
   780
--
alpar@1
   781
-- S.J.Wright. The Cholesky factorization in interior-point and barrier
alpar@1
   782
-- methods. Preprint MCS-P600-0596, Mathematics and Computer Science
alpar@1
   783
-- Division, Argonne National Laboratory, Argonne, Ill., May 1996.
alpar@1
   784
--
alpar@1
   785
-- The routine just replaces non-positive u[k,k] by a huge positive
alpar@1
   786
-- number. This involves non-diagonal elements in k-th row of U to be
alpar@1
   787
-- close to zero that, in turn, involves k-th component of a solution
alpar@1
   788
-- vector to be close to zero. Note, however, that this technique works
alpar@1
   789
-- only if the system A*x = b is consistent. */
alpar@1
   790
alpar@1
   791
int chol_numeric(int n,
alpar@1
   792
      int A_ptr[], int A_ind[], double A_val[], double A_diag[],
alpar@1
   793
      int U_ptr[], int U_ind[], double U_val[], double U_diag[])
alpar@1
   794
{     int i, j, k, t, t1, beg, end, beg1, end1, count = 0;
alpar@1
   795
      double ukk, uki, *work;
alpar@1
   796
      work = xcalloc(1+n, sizeof(double));
alpar@1
   797
      for (j = 1; j <= n; j++) work[j] = 0.0;
alpar@1
   798
      /* U := (upper triangle of A) */
alpar@1
   799
      /* note that the upper traingle of A is a subset of U */
alpar@1
   800
      for (i = 1; i <= n; i++)
alpar@1
   801
      {  beg = A_ptr[i], end = A_ptr[i+1];
alpar@1
   802
         for (t = beg; t < end; t++)
alpar@1
   803
            j = A_ind[t], work[j] = A_val[t];
alpar@1
   804
         beg = U_ptr[i], end = U_ptr[i+1];
alpar@1
   805
         for (t = beg; t < end; t++)
alpar@1
   806
            j = U_ind[t], U_val[t] = work[j], work[j] = 0.0;
alpar@1
   807
         U_diag[i] = A_diag[i];
alpar@1
   808
      }
alpar@1
   809
      /* main elimination loop */
alpar@1
   810
      for (k = 1; k <= n; k++)
alpar@1
   811
      {  /* transform k-th row of U */
alpar@1
   812
         ukk = U_diag[k];
alpar@1
   813
         if (ukk > 0.0)
alpar@1
   814
            U_diag[k] = ukk = sqrt(ukk);
alpar@1
   815
         else
alpar@1
   816
            U_diag[k] = ukk = DBL_MAX, count++;
alpar@1
   817
         /* (work) := (transformed k-th row) */
alpar@1
   818
         beg = U_ptr[k], end = U_ptr[k+1];
alpar@1
   819
         for (t = beg; t < end; t++)
alpar@1
   820
            work[U_ind[t]] = (U_val[t] /= ukk);
alpar@1
   821
         /* transform other rows of U */
alpar@1
   822
         for (t = beg; t < end; t++)
alpar@1
   823
         {  i = U_ind[t];
alpar@1
   824
            xassert(i > k);
alpar@1
   825
            /* (i-th row) := (i-th row) - u[k,i] * (k-th row) */
alpar@1
   826
            uki = work[i];
alpar@1
   827
            beg1 = U_ptr[i], end1 = U_ptr[i+1];
alpar@1
   828
            for (t1 = beg1; t1 < end1; t1++)
alpar@1
   829
               U_val[t1] -= uki * work[U_ind[t1]];
alpar@1
   830
            U_diag[i] -= uki * uki;
alpar@1
   831
         }
alpar@1
   832
         /* (work) := 0 */
alpar@1
   833
         for (t = beg; t < end; t++)
alpar@1
   834
            work[U_ind[t]] = 0.0;
alpar@1
   835
      }
alpar@1
   836
      xfree(work);
alpar@1
   837
      return count;
alpar@1
   838
}
alpar@1
   839
alpar@1
   840
/*----------------------------------------------------------------------
alpar@1
   841
-- u_solve - solve upper triangular system U*x = b.
alpar@1
   842
--
alpar@1
   843
-- *Synopsis*
alpar@1
   844
--
alpar@1
   845
-- #include "glpmat.h"
alpar@1
   846
-- void u_solve(int n, int U_ptr[], int U_ind[], double U_val[],
alpar@1
   847
--    double U_diag[], double x[]);
alpar@1
   848
--
alpar@1
   849
-- *Description*
alpar@1
   850
--
alpar@1
   851
-- The routine u_solve solves an linear system U*x = b, where U is an
alpar@1
   852
-- upper triangular matrix.
alpar@1
   853
--
alpar@1
   854
-- The parameter n is the order of matrix U.
alpar@1
   855
--
alpar@1
   856
-- The matrix U without diagonal elements is specified in the arrays
alpar@1
   857
-- U_ptr, U_ind, and U_val in storage-by-rows format. Diagonal elements
alpar@1
   858
-- of U are specified in the array U_diag, where U_diag[0] is not used,
alpar@1
   859
-- U_diag[i] = u[i,i] for i = 1, ..., n. All these four arrays are not
alpar@1
   860
-- changed on exit.
alpar@1
   861
--
alpar@1
   862
-- The right-hand side vector b is specified on entry in the array x,
alpar@1
   863
-- where x[0] is not used, and x[i] = b[i] for i = 1, ..., n. On exit
alpar@1
   864
-- the routine stores computed components of the vector of unknowns x
alpar@1
   865
-- in the array x in the same manner. */
alpar@1
   866
alpar@1
   867
void u_solve(int n, int U_ptr[], int U_ind[], double U_val[],
alpar@1
   868
      double U_diag[], double x[])
alpar@1
   869
{     int i, t, beg, end;
alpar@1
   870
      double temp;
alpar@1
   871
      for (i = n; i >= 1; i--)
alpar@1
   872
      {  temp = x[i];
alpar@1
   873
         beg = U_ptr[i], end = U_ptr[i+1];
alpar@1
   874
         for (t = beg; t < end; t++)
alpar@1
   875
            temp -= U_val[t] * x[U_ind[t]];
alpar@1
   876
         xassert(U_diag[i] != 0.0);
alpar@1
   877
         x[i] = temp / U_diag[i];
alpar@1
   878
      }
alpar@1
   879
      return;
alpar@1
   880
}
alpar@1
   881
alpar@1
   882
/*----------------------------------------------------------------------
alpar@1
   883
-- ut_solve - solve lower triangular system U'*x = b.
alpar@1
   884
--
alpar@1
   885
-- *Synopsis*
alpar@1
   886
--
alpar@1
   887
-- #include "glpmat.h"
alpar@1
   888
-- void ut_solve(int n, int U_ptr[], int U_ind[], double U_val[],
alpar@1
   889
--    double U_diag[], double x[]);
alpar@1
   890
--
alpar@1
   891
-- *Description*
alpar@1
   892
--
alpar@1
   893
-- The routine ut_solve solves an linear system U'*x = b, where U is a
alpar@1
   894
-- matrix transposed to an upper triangular matrix.
alpar@1
   895
--
alpar@1
   896
-- The parameter n is the order of matrix U.
alpar@1
   897
--
alpar@1
   898
-- The matrix U without diagonal elements is specified in the arrays
alpar@1
   899
-- U_ptr, U_ind, and U_val in storage-by-rows format. Diagonal elements
alpar@1
   900
-- of U are specified in the array U_diag, where U_diag[0] is not used,
alpar@1
   901
-- U_diag[i] = u[i,i] for i = 1, ..., n. All these four arrays are not
alpar@1
   902
-- changed on exit.
alpar@1
   903
--
alpar@1
   904
-- The right-hand side vector b is specified on entry in the array x,
alpar@1
   905
-- where x[0] is not used, and x[i] = b[i] for i = 1, ..., n. On exit
alpar@1
   906
-- the routine stores computed components of the vector of unknowns x
alpar@1
   907
-- in the array x in the same manner. */
alpar@1
   908
alpar@1
   909
void ut_solve(int n, int U_ptr[], int U_ind[], double U_val[],
alpar@1
   910
      double U_diag[], double x[])
alpar@1
   911
{     int i, t, beg, end;
alpar@1
   912
      double temp;
alpar@1
   913
      for (i = 1; i <= n; i++)
alpar@1
   914
      {  xassert(U_diag[i] != 0.0);
alpar@1
   915
         temp = (x[i] /= U_diag[i]);
alpar@1
   916
         if (temp == 0.0) continue;
alpar@1
   917
         beg = U_ptr[i], end = U_ptr[i+1];
alpar@1
   918
         for (t = beg; t < end; t++)
alpar@1
   919
            x[U_ind[t]] -= U_val[t] * temp;
alpar@1
   920
      }
alpar@1
   921
      return;
alpar@1
   922
}
alpar@1
   923
alpar@1
   924
/* eof */