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
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
/*-*- mode: C; kept-old-versions: 12;  kept-new-versions: 20; -*-
 *
 * srqfn.f -- translated by f2c (version 20031025) and by
 * $Id: f2c-clean,v 1.10 2002/03/28 16:37:27 maechler Exp $
 * plus extended manual code cleaning by Martin Maechler, ETH Zurich
 */

#include <Rmath.h>

/* Table of constant values */

static int c__1 = 1;
static double c_9995 = .9995;


/* BLAS : */
#include <R_ext/BLAS.h>
/* double ddot_(int *, double *, int *, double *, int *); */
/* int   daxpy_(int *, double*, double *, int *, double *, int *); */

#define DDOT(_n_, _X_, _Y_)  F77_CALL(ddot)(_n_, _X_, &c__1, _Y_, &c__1)

/* SparseM -- ./sparskit2.f : */
#include "sparseM.h"

/* Cholesky related : */
#include "cholesky.h"

/* advance declaration : */
static void
slpfn(int *n, int *m, int *nnza, double *a, int *ja, int *ia,
      double *ao, int *jao, int *iao, int *nnzdmax, double *d, int *jd, int *id,
      double *dsub, int *jdsub, int *nsubmax, int *lindx, int *xlindx,
      int *nnzlmax, double *lnz, int *xlnz,
      int *invp, int *perm, int *iwmax, int *iwork,
      int *colcnt, int *snode, int *xsuper, int *split,
      int *tmpmax, double *tmpvec, double *newrhs, int *cachsz,
      int *level, double *x, double *s, double *u,
      double *c, double *y, double *b, double *r,
      double *z, double *w, double *q, int *nnzemax,
      double *e, int *je, int *ie,
      double *dy, double *dx, double *ds, double *dz, double *dw,
      double *dxdz, double *dsdw, double *xi, double *xinv, double *sinv,
      double *ww1, double *ww2, double *small, int *ierr,
      int *maxit, double *timewd);

static void
bound(double *x, double *dx, double *s,
      double *ds, double *z, double *dz, double *w,
      double *dw, int *n, double *beta, double *deltap,
      double *deltad);


/*  called from R's rq.fit.sfn()  in ../R/sfn.R */
int
srqfn_(int *n, int *m, int *nnza,
       double *a, int *ja, int *ia,
       double *ao, int *jao, int *iao,
       int *nnzdmax, double *d, int *jd, int *id,
	double *dsub, int *jdsub,
       int *nnzemax, double *e, int *je, int *ie,
       int *nsubmax, int *lindx, int *xlindx,
       int *nnzlmax, double *lnz, int *xlnz, int *iw,
       int *iwmax, int *iwork, int *xsuper, int *tmpmax,
       double *tmpvec, double *wwm, double *wwn, int *cachsz,
       int *level, double *x, double *s, double *u,
       double *c, double *y, double *b, double *small,
       int *ierr, int *maxit, double *timewd)
{
    /* System generated locals */
    int iw_dim1, wwm_dim1, wwn_dim1;

    /* Parameter adjustments */
    wwn_dim1 = *n;    wwn -= wwn_dim1;
    wwm_dim1 = *m;    wwm -= wwm_dim1;
    iw_dim1 = *m;     iw -= iw_dim1;


    /* Function Body */
    slpfn(n, m, nnza, a, ja, ia, ao, jao, iao,
	  nnzdmax, d, jd, id, dsub, jdsub, nsubmax,
	  lindx, xlindx, nnzlmax, lnz, xlnz,
	  &iw[iw_dim1], &iw[(iw_dim1 << 1)],
	  iwmax, iwork,
	  &iw[iw_dim1 * 3], &iw[(iw_dim1 << 2)], xsuper, &iw[iw_dim1 * 5],
	  tmpmax, tmpvec, &wwm[(wwm_dim1 << 1)], cachsz, level,
	  x, s, u, c, y, b,
	  &wwn[wwn_dim1    ], &wwn[(wwn_dim1 << 1)],
	  &wwn[wwn_dim1 * 3], &wwn[(wwn_dim1 << 2)],
	  nnzemax, e, je, ie,
	  &wwm[wwm_dim1 * 3], &wwn[wwn_dim1 * 5],
	  &wwn[wwn_dim1 * 6], &wwn[wwn_dim1 * 7],
	  &wwn[(wwn_dim1 << 3)], &wwn[wwn_dim1 * 9],
	  &wwn[wwn_dim1 * 10], &wwn[wwn_dim1 * 11],
	  &wwn[wwn_dim1 * 12], &wwn[wwn_dim1 * 13],
	  &wwn[wwn_dim1 * 14],
	  &wwm[wwm_dim1], small, ierr, maxit, timewd);
    return 0;
} /* srqfn_ */

static void
slpfn(int *n, int *m, int *nnza, double *a, int *ja, int *ia,
      double *ao, int *jao, int *iao, int *nnzdmax, double *d, int *jd, int *id,
      double *dsub, int *jdsub, int *nsubmax, int *lindx, int *xlindx,
      int *nnzlmax, double *lnz, int *xlnz,
      int *invp, int *perm, int *iwmax, int *iwork,
      int *colcnt, int *snode, int *xsuper, int *split,
      int *tmpmax, double *tmpvec, double *newrhs, int *cachsz,
      int *level, double *x, double *s, double *u,
      double *c, double *y, double *b, double *r,
      double *z, double *w, double *q, int *nnzemax,
      double *e, int *je, int *ie,
      double *dy, double *dx, double *ds, double *dz, double *dw,
      double *dxdz, double *dsdw, double *xi, double *xinv, double *sinv,
      double *ww1, double *ww2, double *small, int *ierr,
      int *maxit, double *timewd)
{
/* Sparse implentation of LMS's interior point method via
    Ng-Peyton's sparse Cholesky factorization for sparse
    symmetric positive definite

 INPUT:
     n -- the number of rows in the coefficient matrix A'
     m -- the number of columns in the coefficient matrix A'
     nnza -- the number of non-zero elements in A'
     a -- an nnza-vector of non-zero values of the design
          matrix (A') stored in csr format
     ja -- an nnza-vector of indices of the non-zero elements of
           the coefficient matrix
     ia -- an (n+1)-vector of pointers to the begining of each
           row in a and ja
     ao -- an nnza-vector of work space for the transpose of
           the design matrix stored in csr format or the
           design matrix stored in csc format
     jao -- an nnza-vector of work space for the indices of the
            transpose of the design matrix
     iao -- an (n+1)-vector of pointers to the begining of each
            column in ao and jao
     nnzdmax -- upper bound of the non-zero elements in AA'
     d -- an nnzdmax-vector of non-zero values of AQ^(-1)
     jd -- an nnzdmax-vector of indices in d
     id -- an (m+1)-vector of pointers to the begining of each
           row in d and jd
     dsub -- the values of e excluding the diagonal elements
     jdsub -- the indices to dsub
     nsubmax -- upper bound of the dimension of lindx
     lindx -- an nsub-vector of integer which contains, in
           column major oder, the row subscripts of the nonzero
           entries in L in a compressed storage format
     xlindx -- an (m+1)-vector of int of pointers for lindx
     nnzlmax -- the upper bound of the non-zero entries in
                L stored in lnz, including the diagonal entries
     lnz -- First contains the non-zero entries of d; later
            contains the entries of the Cholesky factor
     xlnz -- column pointer for L stored in lnz
     invp -- an n-vector of int of inverse permutation vector
     perm -- an n-vector of int of permutation vector
     iw -- int work array of length m
     iwmax -- upper bound of the general purpose int
              working storage iwork; set at 7*m+3
     iwork -- an iwsiz-vector of int as work space
     colcnt -- array of length m, containing the number of
               non-zeros in each column of the factor, including
               the diagonal entries
     snode -- array of length m for recording supernode membership
     xsuper -- array of length m+1 containing the supernode
               partitioning
     split -- an m-vector with splitting of supernodes so that
              they fit into cache
     tmpmax -- upper bound of the dimension of tmpvec
     tmpvec -- a tmpmax-vector of temporary vector
     newrhs -- extra work vector for right-hand side and solution
     cachsz -- size of the cache (in kilobytes) on the target machine
     level -- level of loop unrolling while performing numerical
              factorization
     x -- an n-vector, the initial feasible solution in the primal
              that corresponds to the design matrix A'
     s -- an n-vector
     u -- an n-vector of upper bound for x
     c -- an n-vector, usually the "negative" of
          the pseudo response
     y -- an m-vector, the initial dual solution
     b -- an n-vector, usualy the rhs of the equality constraint
          X'a = (1-tau)X'e in the rq setting
     r -- an n-vector of residuals
     z -- an n-vector of the dual slack variable
     w -- an n-vector
     q -- an n-vector of work array containing the diagonal
          elements of the Q^(-1) matrix
     nnzemax -- upper bound of the non-zero elements in AA'
     e -- an nnzdmax-vector containing the non-zero entries of
          AQ^(-1)A' stored in csr format
     je -- an nnzemax-vector of indices for e
     ie -- an (m+1)-vector of pointers to the begining of each row in e and je
     dy -- work array
     dx -- work array
     ds -- work array
     dz -- work array
     dw -- work array
     dxdz -- work array
     dsdw -- work arry
     xi -- work array
     xinv -- work array
     sinv -- work array
     ww1 -- work array
     ww2 -- work array
     small -- convergence tolerance for interior algorithm
     ierr -- error flag :
       1 -- insufficient storage (work space) when calling extract;
       2 -- nnzd > nnzdmax
       3 -- insufficient storage in iwork when calling ordmmd;
       4 -- insufficient storage in iwork when calling sfinit;
       5 -- nnzl > nnzlmax when calling sfinit
       6 -- nsub > nsubmax when calling sfinit
       7 -- insufficient work space in iwork when calling symfct
       8 -- inconsistancy in input when calling symfct
       9 -- tmpsiz > tmpmax when calling bfinit; increase tmpmax
       10 -- nonpositive diagonal encountered when calling blkfct(),
             the matrix is not positive definite
       11 -- insufficient work storage in tmpvec when calling blkfct()
       12 -- insufficient work storage in iwork  when calling blkfct()
       17 -- tiny diagonals replaced with Inf when calling blkfct()
       		{new error code, was confounded with '10'}

     maxit -- maximal number of iterations; on return: the number of iterations

     timewd -- vector of length 7: [7]: amount of time for this subroutine
               [1:6] time info for the phases of cholfct() only.

 OUTPUT:
     y -- an m-vector of primal solution
*/

    /* Local variables */
    int i, it, nnzd, nnzdsub, nsuper;
    double deltad, deltap, mu, gap;
    double timbeg;


/* Parameter adjustments */
    --sinv;
    --xinv;
    --xi;
    --dsdw;
    --dxdz;
    --dw;
    --dz;
    --ds;
    --dx;
    --q;
    --w;
    --z;
    --r;
    --c;
    --u;
    --s;
    --x;

    --ww1;
    --ww2;
    --dy;
    --b;
    --y;
    --newrhs;

    --perm;
    --invp;

    --dsub;
    --timewd;

    /* Function Body */
    for (i = 1; i <= 7; ++i)
	timewd[i] = 0.;

    nnzd = ie[*m] - 1;
    nnzdsub = nnzd - *m;

/*  Compute the initial gap */

    gap = DDOT(n, &z[1], &x[1]) + DDOT(n, &w[1], &s[1]);

/*  Start iteration */

    it = 0;
    while(gap >= *small && it <= *maxit) {

	++it;

/*  Create the diagonal matrix Q^(-1) stored in q as an n-vector
    and update the residuals in r */

	for (i = 1; i <= *n; ++i) {
	    q[i] = 1. / (z[i] / x[i] + w[i] / s[i]);
	    r[i] = z[i] - w[i];
	}

/*  Obtain AQ^(-1) and store in d,jd,id in csr format */

	F77_CALL(amudia)(m, &c__1, ao, jao, iao, &q[1], d, jd, id);

/*  Obtain AQ^(-1)A' and store in e,je,ie in csr format */

	F77_CALL(amub)(m, m, &c__1, d, jd, id,
		       a, ja, ia, e, je, ie, nnzemax, iwork, ierr);

	if (*ierr != 0) { *ierr = 2; goto L100; }

/*  Extract the non-diagonal structure of e,je,ie and store in dsub,jdsub */

	i = *nnzemax + 1;
	F77_CALL(extract)(e, je, ie, &dsub[1], jdsub, m, nnzemax, &i, ierr);

	if (*ierr != 0) { *ierr = 1; goto L100; }

/* Compute b - Ax + AQ^(-1)r and store it in c in two steps
   First: store Ax in ww2 */
	F77_CALL(amux)(m, &x[1], &ww2[1], ao, jao, iao);

/*  Second: save AQ^(-1)r in c temporarily */

	F77_CALL(amux)(m, &r[1], &c[1], d, jd, id);
	for (i = 1; i <= *m; ++i) {
	    c[i] = b[i] - ww2[i] + c[i];
	}

/*  Compute dy = (AQ^(-1)A')^(-1)(b-Ax+AQ^(-1)r); result returned via dy */

	/* chlfct: perform Cholesky's decomposition of e,je,ie */

	F77_CALL(chlfct)(m, xlindx, lindx, &invp[1], &perm[1], iwork, &nnzdsub,
			 jdsub, colcnt, &nsuper, snode, xsuper, nnzlmax,
			 nsubmax, xlnz, lnz, ie, je, e, cachsz, tmpmax,
			 level, tmpvec, split, ierr, &it, &timewd[1]);

	if (*ierr != 0) { /* return the chlfct() error code in {3:12}: */ goto L100; }

	for (i = 1; i <= *m; ++i)
	    newrhs[i] = c[perm[i]];

	/* blkslv: Numerical solution for the new rhs stored in b */
	timbeg = F77_CALL(gtimer)();
	F77_CALL(blkslv)(&nsuper, xsuper, xlindx, lindx, xlnz, lnz, &newrhs[1]);
	timewd[7] += F77_CALL(gtimer)() - timbeg;
	for (i = 1; i <= *m; ++i)
	    dy[i] = newrhs[invp[i]];

/*  Compute dx = Q^(-1)(A'dy - r), ds = -dx, dz  and dw */

	F77_CALL(amux)(n, &dy[1], &dx[1], a, ja, ia);
	for (i = 1; i <= *n; ++i) {
	    dx[i] = q[i] * (dx[i] - r[i]);
	    ds[i] = -dx[i];
	    dz[i] = -z[i] * (dx[i] / x[i] + 1.);
	    dw[i] = -w[i] * (ds[i] / s[i] + 1.);
	}

/* Compute the maximum allowable step lengths */

	bound(&x[1], &dx[1], &s[1], &ds[1], &z[1], &dz[1], &w[1], &dw[1],
	      n, &c_9995, &deltap, &deltad);

	if (deltap * deltad < 1.) {

	    /* Update mu */
	    double
		g =		      DDOT(n, &z[1], &x[1]) +
		    deltap *	      DDOT(n, &z[1], &dx[1]) +
		    deltad *	      DDOT(n, &dz[1], &x[1]) +
		    deltad * deltap * DDOT(n, &dz[1], &dx[1]) +
				      DDOT(n, &w[1], &s[1]) +
		    deltap *	      DDOT(n, &w[1], &ds[1]) +
		    deltad *	      DDOT(n, &dw[1], &s[1]) +
		    deltad * deltap * DDOT(n, &dw[1], &ds[1]);

	    mu = DDOT(n, &z[1], &x[1]) + DDOT(n, &w[1], &s[1]);
	    g /= mu;
	    mu *= g * g * g / (2. * (*n));

	    /* Compute dxdz and dsdw */

	    for (i = 1; i <= *n; ++i) {
		dxdz[i] = dx[i] * dz[i];
		dsdw[i] = ds[i] * dw[i];
		xinv[i] = 1. / x[i];
		sinv[i] = 1. / s[i];
		xi[i] = xinv[i] * dxdz[i] - sinv[i] * dsdw[i] -
		    mu * (xinv[i] - sinv[i]);
		ww1[i] = q[i] * xi[i];
	    }

	    /* Compute AQ^(-1)(dxdz - dsdw - mu(X^(-1) - S^(-1))) and
	       store it in ww2 temporarily */

	    F77_CALL(amux)(m, &ww1[1], &ww2[1], ao, jao, iao);
	    for (i = 1; i <= *m; ++i) {
		c[i] += ww2[i];
	    }

	    /* Compute dy and return the result in dy */

	    for (i = 1; i <= *m; ++i) {
		newrhs[i] = c[perm[i]];
	    }

	    /* Call blkslv: Numerical solution for the new rhs stored in b */

	    timbeg = F77_CALL(gtimer)();
	    F77_CALL(blkslv)(&nsuper, xsuper, xlindx, lindx, xlnz, lnz, &newrhs[1]);
	    timewd[7] += F77_CALL(gtimer)() - timbeg;
	    for (i = 1; i <= *m; ++i) {
		dy[i] = newrhs[invp[i]];
	    }

	    /* Compute dx = Q^(-1)(A'dy - r + mu(X^(-1) - S^(-1)) -dxdz + dsdw),
	       ds = -dx, dz  and dw */

	    F77_CALL(amux)(n, &dy[1], &dx[1], a, ja, ia);
	    for (i = 1; i <= *n; ++i) {
		dx[i] = q[i] * (dx[i] - xi[i] - r[i]);
		ds[i] = -dx[i];
		dz[i] = -z[i] + xinv[i] * (mu - z[i] * dx[i] - dxdz[i]);
		dw[i] = -w[i] + sinv[i] * (mu - w[i] * ds[i] - dsdw[i]);
	    }

	    /* Compute the maximum allowable step lengths */

	    bound(&x[1], &dx[1], &s[1], &ds[1], &z[1], &dz[1], &w[1], &dw[1],
		  n, &c_9995, &deltap, &deltad);
	}

	/* Take the step */

	F77_CALL(daxpy)(n, &deltap, &dx[1], &c__1, &x[1], &c__1);
	F77_CALL(daxpy)(n, &deltap, &ds[1], &c__1, &s[1], &c__1);
	F77_CALL(daxpy)(n, &deltad, &dw[1], &c__1, &w[1], &c__1);
	F77_CALL(daxpy)(n, &deltad, &dz[1], &c__1, &z[1], &c__1);
	F77_CALL(daxpy)(m, &deltad, &dy[1], &c__1, &y[1], &c__1);

	gap = DDOT(n, &z[1], &x[1]) + DDOT(n, &w[1], &s[1]);

    } /* end {while} iterations */


L100:
    *maxit = it;
    return;
} /* slpfn */

static void
bound(double *x, double *dx, double *s,
      double *ds, double *z, double *dz, double *w,
      double *dw, int *n, double *beta, double *deltap,
      double *deltad)
{
    int i;

    *deltap = R_PosInf;
    *deltad = R_PosInf;
    for (i = 0; i < *n; ++i) {
	if (dx[i] < 0.) *deltap = fmin2(*deltap, -x[i] / dx[i]);
	if (ds[i] < 0.) *deltap = fmin2(*deltap, -s[i] / ds[i]);
	if (dz[i] < 0.) *deltad = fmin2(*deltad, -z[i] / dz[i]);
	if (dw[i] < 0.) *deltad = fmin2(*deltad, -w[i] / dw[i]);
    }

    *deltap = fmin2(1., *beta * *deltap);
    *deltad = fmin2(1., *beta * *deltad);
    return;
} /* bound */