The FSF has a new address.
[bpt/guile.git] / libguile / random.c
1 /* Copyright (C) 1999,2000,2001, 2003 Free Software Foundation, Inc.
2 * This library is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU Lesser General Public
4 * License as published by the Free Software Foundation; either
5 * version 2.1 of the License, or (at your option) any later version.
6 *
7 * This library is distributed in the hope that it will be useful,
8 * but WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
10 * Lesser General Public License for more details.
11 *
12 * You should have received a copy of the GNU Lesser General Public
13 * License along with this library; if not, write to the Free Software
14 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
15 */
16
17
18
19 /* Author: Mikael Djurfeldt <djurfeldt@nada.kth.se> */
20
21 #if HAVE_CONFIG_H
22 # include <config.h>
23 #endif
24
25 #include "libguile/_scm.h"
26
27 #include <gmp.h>
28 #include <stdio.h>
29 #include <math.h>
30 #include <string.h>
31 #include "libguile/smob.h"
32 #include "libguile/numbers.h"
33 #include "libguile/feature.h"
34 #include "libguile/strings.h"
35 #include "libguile/unif.h"
36 #include "libguile/srfi-4.h"
37 #include "libguile/vectors.h"
38
39 #include "libguile/validate.h"
40 #include "libguile/random.h"
41
42 \f
43 /*
44 * A plugin interface for RNGs
45 *
46 * Using this interface, it is possible for the application to tell
47 * libguile to use a different RNG. This is desirable if it is
48 * necessary to use the same RNG everywhere in the application in
49 * order to prevent interference, if the application uses RNG
50 * hardware, or if the application has special demands on the RNG.
51 *
52 * Look in random.h and how the default generator is "plugged in" in
53 * scm_init_random().
54 */
55
56 scm_t_rng scm_the_rng;
57
58 \f
59 /*
60 * The prepackaged RNG
61 *
62 * This is the MWC (Multiply With Carry) random number generator
63 * described by George Marsaglia at the Department of Statistics and
64 * Supercomputer Computations Research Institute, The Florida State
65 * University (http://stat.fsu.edu/~geo).
66 *
67 * It uses 64 bits, has a period of 4578426017172946943 (4.6e18), and
68 * passes all tests in the DIEHARD test suite
69 * (http://stat.fsu.edu/~geo/diehard.html)
70 */
71
72 #define A 2131995753UL
73
74 #ifndef M_PI
75 #define M_PI 3.14159265359
76 #endif
77
78 #if SCM_HAVE_T_INT64
79
80 unsigned long
81 scm_i_uniform32 (scm_t_i_rstate *state)
82 {
83 scm_t_int64 x = (scm_t_int64) A * state->w + state->c;
84 scm_t_int32 w = x & 0xffffffffUL;
85 state->w = w;
86 state->c = x >> 32L;
87 return w;
88 }
89
90 #else
91
92 /* ww This is a portable version of the same RNG without 64 bit
93 * * aa arithmetic.
94 * ----
95 * xx It is only intended to provide identical behaviour on
96 * xx platforms without 8 byte longs or long longs until
97 * xx someone has implemented the routine in assembler code.
98 * xxcc
99 * ----
100 * ccww
101 */
102
103 #define L(x) ((x) & 0xffff)
104 #define H(x) ((x) >> 16)
105
106 unsigned long
107 scm_i_uniform32 (scm_t_i_rstate *state)
108 {
109 scm_t_int32 x1 = L (A) * L (state->w);
110 scm_t_int32 x2 = L (A) * H (state->w);
111 scm_t_int32 x3 = H (A) * L (state->w);
112 scm_t_int32 w = L (x1) + L (state->c);
113 scm_t_int32 m = H (x1) + L (x2) + L (x3) + H (state->c) + H (w);
114 scm_t_int32 x4 = H (A) * H (state->w);
115 state->w = w = (L (m) << 16) + L (w);
116 state->c = H (x2) + H (x3) + x4 + H (m);
117 return w;
118 }
119
120 #endif
121
122 void
123 scm_i_init_rstate (scm_t_i_rstate *state, const char *seed, int n)
124 {
125 scm_t_int32 w = 0L;
126 scm_t_int32 c = 0L;
127 int i, m;
128 for (i = 0; i < n; ++i)
129 {
130 m = i % 8;
131 if (m < 4)
132 w += seed[i] << (8 * m);
133 else
134 c += seed[i] << (8 * (m - 4));
135 }
136 if ((w == 0 && c == 0) || (w == -1 && c == A - 1))
137 ++c;
138 state->w = w;
139 state->c = c;
140 }
141
142 scm_t_i_rstate *
143 scm_i_copy_rstate (scm_t_i_rstate *state)
144 {
145 scm_t_rstate *new_state = scm_malloc (scm_the_rng.rstate_size);
146 if (new_state == 0)
147 scm_memory_error ("rstate");
148 return memcpy (new_state, state, scm_the_rng.rstate_size);
149 }
150
151 \f
152 /*
153 * Random number library functions
154 */
155
156 scm_t_rstate *
157 scm_c_make_rstate (const char *seed, int n)
158 {
159 scm_t_rstate *state = scm_malloc (scm_the_rng.rstate_size);
160 if (state == 0)
161 scm_memory_error ("rstate");
162 state->reserved0 = 0;
163 scm_the_rng.init_rstate (state, seed, n);
164 return state;
165 }
166
167
168 scm_t_rstate *
169 scm_c_default_rstate ()
170 #define FUNC_NAME "scm_c_default_rstate"
171 {
172 SCM state = SCM_VARIABLE_REF (scm_var_random_state);
173 if (!SCM_RSTATEP (state))
174 SCM_MISC_ERROR ("*random-state* contains bogus random state", SCM_EOL);
175 return SCM_RSTATE (state);
176 }
177 #undef FUNC_NAME
178
179
180 inline double
181 scm_c_uniform01 (scm_t_rstate *state)
182 {
183 double x = (double) scm_the_rng.random_bits (state) / (double) 0xffffffffUL;
184 return ((x + (double) scm_the_rng.random_bits (state))
185 / (double) 0xffffffffUL);
186 }
187
188 double
189 scm_c_normal01 (scm_t_rstate *state)
190 {
191 if (state->reserved0)
192 {
193 state->reserved0 = 0;
194 return state->reserved1;
195 }
196 else
197 {
198 double r, a, n;
199
200 r = sqrt (-2.0 * log (scm_c_uniform01 (state)));
201 a = 2.0 * M_PI * scm_c_uniform01 (state);
202
203 n = r * sin (a);
204 state->reserved1 = r * cos (a);
205 state->reserved0 = 1;
206
207 return n;
208 }
209 }
210
211 double
212 scm_c_exp1 (scm_t_rstate *state)
213 {
214 return - log (scm_c_uniform01 (state));
215 }
216
217 unsigned char scm_masktab[256];
218
219 unsigned long
220 scm_c_random (scm_t_rstate *state, unsigned long m)
221 {
222 unsigned int r, mask;
223 mask = (m < 0x100
224 ? scm_masktab[m]
225 : (m < 0x10000
226 ? scm_masktab[m >> 8] << 8 | 0xff
227 : (m < 0x1000000
228 ? scm_masktab[m >> 16] << 16 | 0xffff
229 : scm_masktab[m >> 24] << 24 | 0xffffff)));
230 while ((r = scm_the_rng.random_bits (state) & mask) >= m);
231 return r;
232 }
233
234 /*
235 SCM scm_c_random_bignum (scm_t_rstate *state, SCM m)
236
237 Takes a random state (source of random bits) and a bignum m.
238 Returns a bignum b, 0 <= b < m.
239
240 It does this by allocating a bignum b with as many base 65536 digits
241 as m, filling b with random bits (in 32 bit chunks) up to the most
242 significant 1 in m, and, finally checking if the resultant b is too
243 large (>= m). If too large, we simply repeat the process again. (It
244 is important to throw away all generated random bits if b >= m,
245 otherwise we'll end up with a distorted distribution.)
246
247 */
248
249 SCM
250 scm_c_random_bignum (scm_t_rstate *state, SCM m)
251 {
252 SCM result = scm_i_mkbig ();
253 const size_t m_bits = mpz_sizeinbase (SCM_I_BIG_MPZ (m), 2);
254 /* how many bits would only partially fill the last unsigned long? */
255 const size_t end_bits = m_bits % (sizeof (unsigned long) * SCM_CHAR_BIT);
256 unsigned long *random_chunks = NULL;
257 const unsigned long num_full_chunks =
258 m_bits / (sizeof (unsigned long) * SCM_CHAR_BIT);
259 const unsigned long num_chunks = num_full_chunks + ((end_bits) ? 1 : 0);
260
261 /* we know the result will be this big */
262 mpz_realloc2 (SCM_I_BIG_MPZ (result), m_bits);
263
264 random_chunks =
265 (unsigned long *) scm_gc_calloc (num_chunks * sizeof (unsigned long),
266 "random bignum chunks");
267
268 do
269 {
270 unsigned long *current_chunk = random_chunks + (num_chunks - 1);
271 unsigned long chunks_left = num_chunks;
272
273 mpz_set_ui (SCM_I_BIG_MPZ (result), 0);
274
275 if (end_bits)
276 {
277 /* generate a mask with ones in the end_bits position, i.e. if
278 end_bits is 3, then we'd have a mask of ...0000000111 */
279 const unsigned long rndbits = scm_the_rng.random_bits (state);
280 int rshift = (sizeof (unsigned long) * SCM_CHAR_BIT) - end_bits;
281 unsigned long mask = ((unsigned long) ULONG_MAX) >> rshift;
282 unsigned long highest_bits = rndbits & mask;
283 *current_chunk-- = highest_bits;
284 chunks_left--;
285 }
286
287 while (chunks_left)
288 {
289 /* now fill in the remaining unsigned long sized chunks */
290 *current_chunk-- = scm_the_rng.random_bits (state);
291 chunks_left--;
292 }
293 mpz_import (SCM_I_BIG_MPZ (result),
294 num_chunks,
295 -1,
296 sizeof (unsigned long),
297 0,
298 0,
299 random_chunks);
300 /* if result >= m, regenerate it (it is important to regenerate
301 all bits in order not to get a distorted distribution) */
302 } while (mpz_cmp (SCM_I_BIG_MPZ (result), SCM_I_BIG_MPZ (m)) >= 0);
303 scm_gc_free (random_chunks,
304 num_chunks * sizeof (unsigned long),
305 "random bignum chunks");
306 return scm_i_normbig (result);
307 }
308
309 /*
310 * Scheme level representation of random states.
311 */
312
313 scm_t_bits scm_tc16_rstate;
314
315 static SCM
316 make_rstate (scm_t_rstate *state)
317 {
318 SCM_RETURN_NEWSMOB (scm_tc16_rstate, state);
319 }
320
321 static size_t
322 rstate_free (SCM rstate)
323 {
324 free (SCM_RSTATE (rstate));
325 return 0;
326 }
327
328 /*
329 * Scheme level interface.
330 */
331
332 SCM_GLOBAL_VARIABLE_INIT (scm_var_random_state, "*random-state*", scm_seed_to_random_state (scm_from_locale_string ("URL:http://stat.fsu.edu/~geo/diehard.html")));
333
334 SCM_DEFINE (scm_random, "random", 1, 1, 0,
335 (SCM n, SCM state),
336 "Return a number in [0, N).\n"
337 "\n"
338 "Accepts a positive integer or real n and returns a\n"
339 "number of the same type between zero (inclusive) and\n"
340 "N (exclusive). The values returned have a uniform\n"
341 "distribution.\n"
342 "\n"
343 "The optional argument @var{state} must be of the type produced\n"
344 "by @code{seed->random-state}. It defaults to the value of the\n"
345 "variable @var{*random-state*}. This object is used to maintain\n"
346 "the state of the pseudo-random-number generator and is altered\n"
347 "as a side effect of the random operation.")
348 #define FUNC_NAME s_scm_random
349 {
350 if (SCM_UNBNDP (state))
351 state = SCM_VARIABLE_REF (scm_var_random_state);
352 SCM_VALIDATE_RSTATE (2, state);
353 if (SCM_I_INUMP (n))
354 {
355 unsigned long m = SCM_I_INUM (n);
356 SCM_ASSERT_RANGE (1, n, m > 0);
357 return scm_from_ulong (scm_c_random (SCM_RSTATE (state), m));
358 }
359 SCM_VALIDATE_NIM (1, n);
360 if (SCM_REALP (n))
361 return scm_from_double (SCM_REAL_VALUE (n)
362 * scm_c_uniform01 (SCM_RSTATE (state)));
363
364 if (!SCM_BIGP (n))
365 SCM_WRONG_TYPE_ARG (1, n);
366 return scm_c_random_bignum (SCM_RSTATE (state), n);
367 }
368 #undef FUNC_NAME
369
370 SCM_DEFINE (scm_copy_random_state, "copy-random-state", 0, 1, 0,
371 (SCM state),
372 "Return a copy of the random state @var{state}.")
373 #define FUNC_NAME s_scm_copy_random_state
374 {
375 if (SCM_UNBNDP (state))
376 state = SCM_VARIABLE_REF (scm_var_random_state);
377 SCM_VALIDATE_RSTATE (1, state);
378 return make_rstate (scm_the_rng.copy_rstate (SCM_RSTATE (state)));
379 }
380 #undef FUNC_NAME
381
382 SCM_DEFINE (scm_seed_to_random_state, "seed->random-state", 1, 0, 0,
383 (SCM seed),
384 "Return a new random state using @var{seed}.")
385 #define FUNC_NAME s_scm_seed_to_random_state
386 {
387 SCM res;
388 if (SCM_NUMBERP (seed))
389 seed = scm_number_to_string (seed, SCM_UNDEFINED);
390 SCM_VALIDATE_STRING (1, seed);
391 res = make_rstate (scm_c_make_rstate (scm_i_string_chars (seed),
392 scm_i_string_length (seed)));
393 scm_remember_upto_here_1 (seed);
394 return res;
395
396 }
397 #undef FUNC_NAME
398
399 SCM_DEFINE (scm_random_uniform, "random:uniform", 0, 1, 0,
400 (SCM state),
401 "Return a uniformly distributed inexact real random number in\n"
402 "[0,1).")
403 #define FUNC_NAME s_scm_random_uniform
404 {
405 if (SCM_UNBNDP (state))
406 state = SCM_VARIABLE_REF (scm_var_random_state);
407 SCM_VALIDATE_RSTATE (1, state);
408 return scm_from_double (scm_c_uniform01 (SCM_RSTATE (state)));
409 }
410 #undef FUNC_NAME
411
412 SCM_DEFINE (scm_random_normal, "random:normal", 0, 1, 0,
413 (SCM state),
414 "Return an inexact real in a normal distribution. The\n"
415 "distribution used has mean 0 and standard deviation 1. For a\n"
416 "normal distribution with mean m and standard deviation d use\n"
417 "@code{(+ m (* d (random:normal)))}.")
418 #define FUNC_NAME s_scm_random_normal
419 {
420 if (SCM_UNBNDP (state))
421 state = SCM_VARIABLE_REF (scm_var_random_state);
422 SCM_VALIDATE_RSTATE (1, state);
423 return scm_from_double (scm_c_normal01 (SCM_RSTATE (state)));
424 }
425 #undef FUNC_NAME
426
427 static void
428 vector_scale_x (SCM v, double c)
429 {
430 size_t n;
431 if (scm_is_simple_vector (v))
432 {
433 n = SCM_SIMPLE_VECTOR_LENGTH (v);
434 while (n-- > 0)
435 SCM_REAL_VALUE (SCM_SIMPLE_VECTOR_REF (v, n)) *= c;
436 }
437 else
438 {
439 /* must be a f64vector. */
440 scm_t_array_handle handle;
441 size_t i, len;
442 ssize_t inc;
443 double *elts;
444
445 elts = scm_f64vector_writable_elements (v, &handle, &len, &inc);
446
447 for (i = 0; i < len; i++, elts += inc)
448 *elts *= c;
449
450 scm_array_handle_release (&handle);
451 }
452 }
453
454 static double
455 vector_sum_squares (SCM v)
456 {
457 double x, sum = 0.0;
458 size_t n;
459 if (scm_is_simple_vector (v))
460 {
461 n = SCM_SIMPLE_VECTOR_LENGTH (v);
462 while (n-- > 0)
463 {
464 x = SCM_REAL_VALUE (SCM_SIMPLE_VECTOR_REF (v, n));
465 sum += x * x;
466 }
467 }
468 else
469 {
470 /* must be a f64vector. */
471 scm_t_array_handle handle;
472 size_t i, len;
473 ssize_t inc;
474 const double *elts;
475
476 elts = scm_f64vector_elements (v, &handle, &len, &inc);
477
478 for (i = 0; i < len; i++, elts += inc)
479 {
480 x = *elts;
481 sum += x * x;
482 }
483
484 scm_array_handle_release (&handle);
485 }
486 return sum;
487 }
488
489 /* For the uniform distribution on the solid sphere, note that in
490 * this distribution the length r of the vector has cumulative
491 * distribution r^n; i.e., u=r^n is uniform [0,1], so r can be
492 * generated as r=u^(1/n).
493 */
494 SCM_DEFINE (scm_random_solid_sphere_x, "random:solid-sphere!", 1, 1, 0,
495 (SCM v, SCM state),
496 "Fills @var{vect} with inexact real random numbers the sum of\n"
497 "whose squares is less than 1.0. Thinking of @var{vect} as\n"
498 "coordinates in space of dimension @var{n} @math{=}\n"
499 "@code{(vector-length @var{vect})}, the coordinates are\n"
500 "uniformly distributed within the unit @var{n}-sphere.")
501 #define FUNC_NAME s_scm_random_solid_sphere_x
502 {
503 if (SCM_UNBNDP (state))
504 state = SCM_VARIABLE_REF (scm_var_random_state);
505 SCM_VALIDATE_RSTATE (2, state);
506 scm_random_normal_vector_x (v, state);
507 vector_scale_x (v,
508 pow (scm_c_uniform01 (SCM_RSTATE (state)),
509 1.0 / scm_c_generalized_vector_length (v))
510 / sqrt (vector_sum_squares (v)));
511 return SCM_UNSPECIFIED;
512 }
513 #undef FUNC_NAME
514
515 SCM_DEFINE (scm_random_hollow_sphere_x, "random:hollow-sphere!", 1, 1, 0,
516 (SCM v, SCM state),
517 "Fills vect with inexact real random numbers\n"
518 "the sum of whose squares is equal to 1.0.\n"
519 "Thinking of vect as coordinates in space of\n"
520 "dimension n = (vector-length vect), the coordinates\n"
521 "are uniformly distributed over the surface of the\n"
522 "unit n-sphere.")
523 #define FUNC_NAME s_scm_random_hollow_sphere_x
524 {
525 if (SCM_UNBNDP (state))
526 state = SCM_VARIABLE_REF (scm_var_random_state);
527 SCM_VALIDATE_RSTATE (2, state);
528 scm_random_normal_vector_x (v, state);
529 vector_scale_x (v, 1 / sqrt (vector_sum_squares (v)));
530 return SCM_UNSPECIFIED;
531 }
532 #undef FUNC_NAME
533
534
535 SCM_DEFINE (scm_random_normal_vector_x, "random:normal-vector!", 1, 1, 0,
536 (SCM v, SCM state),
537 "Fills vect with inexact real random numbers that are\n"
538 "independent and standard normally distributed\n"
539 "(i.e., with mean 0 and variance 1).")
540 #define FUNC_NAME s_scm_random_normal_vector_x
541 {
542 long i;
543 scm_t_array_handle handle;
544 scm_t_array_dim *dim;
545
546 if (SCM_UNBNDP (state))
547 state = SCM_VARIABLE_REF (scm_var_random_state);
548 SCM_VALIDATE_RSTATE (2, state);
549
550 scm_generalized_vector_get_handle (v, &handle);
551 dim = scm_array_handle_dims (&handle);
552
553 if (scm_is_vector (v))
554 {
555 SCM *elts = scm_array_handle_writable_elements (&handle);
556 for (i = dim->lbnd; i <= dim->ubnd; i++, elts += dim->inc)
557 *elts = scm_from_double (scm_c_normal01 (SCM_RSTATE (state)));
558 }
559 else
560 {
561 /* must be a f64vector. */
562 double *elts = scm_array_handle_f64_writable_elements (&handle);
563 for (i = dim->lbnd; i <= dim->ubnd; i++, elts += dim->inc)
564 *elts = scm_c_normal01 (SCM_RSTATE (state));
565 }
566
567 scm_array_handle_release (&handle);
568
569 return SCM_UNSPECIFIED;
570 }
571 #undef FUNC_NAME
572
573 SCM_DEFINE (scm_random_exp, "random:exp", 0, 1, 0,
574 (SCM state),
575 "Return an inexact real in an exponential distribution with mean\n"
576 "1. For an exponential distribution with mean u use (* u\n"
577 "(random:exp)).")
578 #define FUNC_NAME s_scm_random_exp
579 {
580 if (SCM_UNBNDP (state))
581 state = SCM_VARIABLE_REF (scm_var_random_state);
582 SCM_VALIDATE_RSTATE (1, state);
583 return scm_from_double (scm_c_exp1 (SCM_RSTATE (state)));
584 }
585 #undef FUNC_NAME
586
587 void
588 scm_init_random ()
589 {
590 int i, m;
591 /* plug in default RNG */
592 scm_t_rng rng =
593 {
594 sizeof (scm_t_i_rstate),
595 (unsigned long (*)()) scm_i_uniform32,
596 (void (*)()) scm_i_init_rstate,
597 (scm_t_rstate *(*)()) scm_i_copy_rstate
598 };
599 scm_the_rng = rng;
600
601 scm_tc16_rstate = scm_make_smob_type ("random-state", 0);
602 scm_set_smob_free (scm_tc16_rstate, rstate_free);
603
604 for (m = 1; m <= 0x100; m <<= 1)
605 for (i = m >> 1; i < m; ++i)
606 scm_masktab[i] = m - 1;
607
608 #include "libguile/random.x"
609
610 scm_add_feature ("random");
611 }
612
613 /*
614 Local Variables:
615 c-file-style: "gnu"
616 End:
617 */