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
|
/*****************************************************************************/
/* */
/* Copyright (c) 1990-2008 Morgan Stanley All rights reserved.*/
/* See .../src/LICENSE for terms of distribution. */
/* */
/* */
/*****************************************************************************/
/*
$cd /u/orth/domino
cc -c domino.c -o domino.o
$load domino_a
$load domino_s
'domino.o' _dyld ('_dmd' ;'dmd' ;0 0 0;
'_mmd' ;'mmd' ; 0 0)
*/
#include <stdio.h>
#include <math.h>
#if defined(__cplusplus)
#include <strings.h>
#else
#include <string.h>
#endif
#include <a/f.h>
#include <a/fncdcls.h>
#include <a/fir.h>
#define Ma(i,j) *((F *)a->p+n*i+j)
#define Mb(t,i,j) *((t *)b->p+p*i+j)
#define Mh(i,j) *((F *)h->p+2*i+j)
#define Mz(i,j) *((F *)z->p+p*i+j)
#define vector(t,a,k) *((t *)a->p+k)
#define Vc(k) *((F *)c->p+k)
#define Vfactor(k) *((F *)factor->p+k)
#define Vtvec(k) *((F *)tvec->p+k)
#define Ipp(k) *((I *)pp->p+k)
#define Ipq(k) *((I *)pq->p+k)
/*
For text errors indicating where an error message originates from, use:
#define DOMAIN_ERROR -1
#define LENGTH_ERROR -1
#define RANK_ERROR -1
*/
A dmd(A,A), mmd(A);
static A ls_c(A,A,I,I,I,I);
/*
Domino
------------ The dyadic case.
*/
A dmd(A b,A a)
{
I result_rank;
I m, n, p;
A z;
if(!QA(a)||!QA(b)) ERROUT(ERR_NONDATA);
if ( (It != a->t && Ft != a->t) || (It != b->t && Ft != b->t) )
ERROUT(ERR_TYPE);
if (2 < a->r || 2 < b->r ) ERROUT(ERR_RANK);
if ( 0 == a->r ) result_rank = 0;
else result_rank = a->r - 1;
if ( 0 != b->r ) result_rank += b->r - 1;
if ( 2 == a->r ) {
m = a->d[0]; n = a->d[1];
}
else {
n = 1;
if ( 1 == a->r ) {
m = a->d[0];
}
else {
m = 1;
}
}
if ( m < n ) ERROUT(ERR_DOMAIN);
if ( 2 == b->r ) {
if ( m != b->d[0] ) ERROUT(ERR_LENGTH);
p = b->d[1];
}
else {
p = 1;
if ( ( 1 == b->r && m != b->d[0] ) || ( 0 == b->r && m != 1 ) )
ERROUT(ERR_LENGTH);
}
RESETXCP;
z = ls_c(a,b,m,n,p,0);
CHECKXCP;
if ( 0 == z ) return(0);
z->r = result_rank;
z->d[0]=(2==result_rank||2>b->r)?n:p;
if (2==result_rank) z->d[1]=p;
return(z);
}
/*
------------ The monadic case.
*/
A mmd( A a )
{
I result_rank;
I m, n, p;
A z;
if(!QA(a)) ERROUT(ERR_NONDATA);
if ( It != a->t && Ft != a->t ) ERROUT(ERR_TYPE);
if (2 < a->r ) ERROUT(ERR_RANK);
result_rank = a->r;
if ( 2 == a->r ) {
m = a->d[0];
n = a->d[1];
}
else if ( 1 == a->r ) {
m = a->d[0];
n = 1;
}
else {
m = 1;
n = 1;
}
if ( m < n ) ERROUT(ERR_DOMAIN);
p = m;
z = ls_c(a,0,m,n,p,1);
if ( 0 == z ) return(0);
z->r = result_rank;
if (1==result_rank) z->d[0]=p;
else if (2==result_rank)
{
z->d[0] = n;
z->d[1] = p;
}
return(z);
}
/* ------------ The least squares computation.
Here's the beef. This program is an amalgamation of two of Mike Jenkin's
models for Domino: one is the primitive that appeared in APLSV and VSAPL,
while the other is for the generalized is for the generlized inverse version
that appeared in the original APL2. This program reflects the current state
of Domino in APL2 (as of 7/12/91), except that the complex arithmetic sections
are missing.
*/
static A ls_c(A a0,A b,I m,I n,I p,I monadic)
/*
m,n,p : scalars set by the driver
a0 : the right argument, a matrix of shape m,n
b : the left argument in the dyadic case, a matrix of shape m,p
z : the result, a matrix of shape n,p
monadic : a flag used to signal generation of the identity matrix
*/
{
I i, j, i0, j0, d[MAXR], l, pi, pj;
F eps, mv, mmv, s, sa, st, t, tolerance, t0, t1, t2, t3, t4, t5, v;
A a, c, factor, h, pp, pq, tvec, z;
/*
Make a copy of a0 because it will be modified. Remember that a0 can be
either integer or real, and always copy to real.
*/
a = ga(Ft, a0->r, a0->n, a0->d );
switch( a0->t ) {
case It:
for ( i = 0 ; i < a0->n ; ++i )
vector(F,a,i) = vector(I,a0,i);
break;
case Ft:
for ( i = 0 ; i < a0->n ; ++i )
vector(F,a,i) = vector(F,a0,i);
break;
};
/*
Initialize the temps and the result.
*/
d[0] = m;
pq = ga(It, 1, d[0], d );
for ( i = 0 ; i < m ; ++i ) Ipq(i) = i;
c = ga(Ft, 1, d[0], d );
d[0] = n;
pp = ga(It, 1, d[0], d );
for ( i = 0 ; i < n ; ++i ) Ipp(i) = i;
factor = ga(Ft, 1, d[0], d );
d[1] = 2;
h = ga(Ft, 2, d[0]*d[1], d );
d[1] = p;
z = ga(Ft, 2, d[0]*d[1], d );
if ( n <= m ) d[0] = m;
tvec = ga(Ft, 1, d[0], d );
/*
tolerance : APL2 uses 1e-16, Jenkin's original paper uses 16e13.
*/
mmv=0.0;
for ( i = 0 ; i < m ; ++i ) {
t0 = 0.0;
for ( j = 0 ; j < n ; ++j ) {
sa = Ma(i,j);
if ( 0 > sa ) sa = -sa;
t0 = t0 + sa;
}
if ( mmv < t0 ) mmv = t0;
}
tolerance = 1e-13;
eps = tolerance*mmv;
/*
Here is the scaling from the APLSV/VSAPL model:
*/
for ( i = 0 ; i < m ; ++i ) {
Vtvec(i)=0.0;
for ( j = 0 ; j < n ; ++j ) {
t1 = Ma(i,j);
if ( 0 > t1 ) t1 = -t1;
if ( Vtvec(i)<t1 ) Vtvec(i) = t1;
}
if ( 0.0 == Vtvec(i) ) Vtvec(i) = 1.0;
}
for ( j = 0 ; j < n ; ++j ) {
Vfactor(j) = 0.0;
for ( i = 0 ; i < m ; ++i ) {
sa = Ma(i,j);
if ( 0 > sa ) sa = -sa;
t2 = sa / Vtvec(i);
if ( Vfactor(j)<t2 ) Vfactor(j) = t2;
}
if ( 0.0 == Vfactor(j) ) Vfactor(j) = 1.0;
sa = 1.0 / Vfactor(j);
Vfactor(j) = sa;
}
for ( i = 0 ; i < m ; ++i ) {
for ( j = 0 ; j < n ; ++j ) {
Ma(i,j) = Ma(i,j)*Vfactor(j);
}
}
/*
Apply the Householder transformations to zero out the matrix a below
the diagonal.
This is the BIG loop.
*/
for ( j = 0 ; j < n ; ++j ) {
mmv = 0.0;
pj = 0;
for ( j0 = j ; j0 < n ; ++j0 ) {
mv = 0.0;
for ( i0 = j ; i0 < m ; ++i0 ) {
t3 = Ma(i0,j0);
if ( 0 > t3 ) t3 = -t3;
if ( mv < t3 ) {
mv = t3;
}
}
if ( mmv < mv ) {
mmv = mv;
pj = j0;
}
}
if( eps >= mmv ) { /* There is no rank deficient case. */
dc(a); dc(c); dc(factor); dc(h); dc(pp); dc(pq); dc(tvec); dc(z);
ERROUT(ERR_DOMAIN);
}
if ( j != pj ) {
i = Ipp(pj);
Ipp(pj) = Ipp(j);
Ipp(j) = i;
for ( i = 0 ; i < m ; ++i ) {
s = Ma(i,pj);
Ma(i,pj) = Ma(i,j);
Ma(i,j) = s;
}
}
/*
The following row interchange is from the APLSV/VSAPL model.
*/
t = 0.0;
pi = 0;
for ( i0 = j; i0 < m ; ++i0 ) {
t4 = Ma(i0,j);
if ( 0 > t4 ) t4 = -t4;
if ( t < t4 ) {
t = t4;
pi = i0;
}
}
if ( j != pi ) {
Ipq(j) = Ipq(pi);
for ( j0 = j ; j0 < n ; ++j0 ) {
s = Ma(j,j0);
Ma(j,j0) = Ma(pi,j0);
Ma(pi,j0) = s;
}
}
/*
Now do the i-th transformation (in place).
*/
t = 0.0;
for ( i0 = j ; i0 < m ; ++i0 ) {
t5 = Ma(i0,j);
if ( 0 > t5 ) t5 = -t5;
if ( t < t5 ) t = t5;
}
v = 0.0;
for ( i0 = j ; i0 < m ; ++i0 ) {
s = Ma(i0,j) / t;
v = v + s*s;
}
NAN_CHECK(v, t*sqrt(v))
if ( Ma(j,j) < 0 ) v = -v;
/*
Save the essential values and adjust the diagonal element.
*/
Mh(j,0) = v;
Mh(j,1) = Ma(j,j);
Ma(j,j) = -v;
/*
Apply the transformation (in place).
*/
for ( j0 = j + 1 ; j0 < n ; ++j0 ) {
Vtvec(j0) = Ma(j,j0);
Ma(j,j0) = Mh(j,1) * Ma(j,j0);
for ( i0 = j + 1 ; i0 < m ; ++i0 ) {
Ma(j,j0) = Ma(j,j0) + Ma(i0,j) * Ma(i0,j0);
}
Ma(j,j0) = - Ma(j,j0) / Mh(j,0);
}
s = Mh(j,0) + Mh(j,1);
for ( j0 = j + 1 ; j0 < n ; ++j0 ) {
t = (Vtvec(j0) - Ma(j,j0)) / s;
for ( i0 = j + 1 ; i0 < m ; ++i0 ) {
Ma(i0,j0) = Ma(i0,j0) - Ma(i0,j) * t;
}
}
}
/*
Build the solutions.
First apply the same transformations to the righthand side as were applied
to a, one column at a time.
The formation of the vector c shows why we process each column of the righthand
side separately, for otherwise in the monadic case we would have to form very
large identity matrices when the argument matrix had many rows.
As with the argument a, the argument b might be integer and we always copy
to a real vector c. The vector c should be created at the top of the loop in C.
*/
for ( i = 0 ; i < n ; ++i )
for ( j = 0; j < p ; ++j ) Mz(i,j) = 0.0;
/*
The next loop is the outer loop for the solution builder.
*/
for ( l = 0 ; l < p ; ++l ) {
if ( monadic ) {
for ( i = 0 ; i < m ; ++i ) Vc(i) = 0.0;
Vc(l) = 1.0;
}
else {
switch ( b->t ) {
case It:
for ( i = 0 ; i < m ; ++i ) Vc(i) = Mb(I,i,l);
break;
case Ft:
for ( i = 0 ; i < m ; ++i ) Vc(i) = Mb(F,i,l);
break;
}
}
for ( j = 0 ; j < n ; ++j ) {
if ( j != Ipq(j) ) {
s = Vc(j);
Vc(j) = Vc(Ipq(j));
Vc(Ipq(j)) = s;
}
/*
Apply the transformation (in place).
*/
st = Vc(j);
Vc(j) = Mh(j,1) * Vc(j);
for ( i0 = j + 1 ; i0 < m ; ++i0 ) {
Vc(j) = Vc(j)+ Ma(i0,j) * Vc(i0);
};
Vc(j) = -Vc(j) / Mh(j,0);
s = Mh(j,0) + Mh(j,1);
t = (st-Vc(j)) / s;
for ( i0 = j + 1 ; i0 < m ; ++i0 ) {
Vc(i0) = Vc(i0) - Ma(i0,j) * t;
}
}
/*
Backsolve the n-by-n triangular system.
*/
for ( j = n - 1 ; 0 <= j ; --j ) {
s = 0.0;
for ( j0 = j ; j0 < n ; ++j0 ) {
s = s + Ma(j,j0) * Mz(Ipp(j0),l);
}
Mz(Ipp(j),l) = (Vc(j)-s) / Ma(j,j);
}
}
/*
Adjust for APLSV/VSAPL scaling.
*/
for ( i = 0 ; i < n ; ++i )
for ( j = 0 ; j < p ; ++j ) {
Mz(i,j) = Mz(i,j) * Vfactor(i);
}
/* Remove the temps.
*/
dc(a); dc(c); dc(factor); dc(h); dc(pp); dc(pq); dc(tvec);
return(z);
}
|