Implementation for the R6RS (rnrs sorting) library.
[bpt/guile.git] / libguile / random.c
1 /* Copyright (C) 1999,2000,2001, 2003, 2005, 2006, 2009 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 License
4 * as published by the Free Software Foundation; either version 3 of
5 * the License, or (at your option) any later version.
6 *
7 * This library is distributed in the hope that it will be useful, but
8 * 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
15 * 02110-1301 USA
16 */
17
18
19
20 /* Author: Mikael Djurfeldt <djurfeldt@nada.kth.se> */
21
22 #ifdef HAVE_CONFIG_H
23 # include <config.h>
24 #endif
25
26 #include "libguile/_scm.h"
27
28 #include <gmp.h>
29 #include <stdio.h>
30 #include <math.h>
31 #include <string.h>
32 #include "libguile/smob.h"
33 #include "libguile/numbers.h"
34 #include "libguile/feature.h"
35 #include "libguile/strings.h"
36 #include "libguile/arrays.h"
37 #include "libguile/srfi-4.h"
38 #include "libguile/vectors.h"
39 #include "libguile/generalized-vectors.h"
40
41 #include "libguile/validate.h"
42 #include "libguile/random.h"
43
44 \f
45 /*
46 * A plugin interface for RNGs
47 *
48 * Using this interface, it is possible for the application to tell
49 * libguile to use a different RNG. This is desirable if it is
50 * necessary to use the same RNG everywhere in the application in
51 * order to prevent interference, if the application uses RNG
52 * hardware, or if the application has special demands on the RNG.
53 *
54 * Look in random.h and how the default generator is "plugged in" in
55 * scm_init_random().
56 */
57
58 scm_t_rng scm_the_rng;
59
60 \f
61 /*
62 * The prepackaged RNG
63 *
64 * This is the MWC (Multiply With Carry) random number generator
65 * described by George Marsaglia at the Department of Statistics and
66 * Supercomputer Computations Research Institute, The Florida State
67 * University (http://stat.fsu.edu/~geo).
68 *
69 * It uses 64 bits, has a period of 4578426017172946943 (4.6e18), and
70 * passes all tests in the DIEHARD test suite
71 * (http://stat.fsu.edu/~geo/diehard.html)
72 */
73
74 #define A 2131995753UL
75
76 #ifndef M_PI
77 #define M_PI 3.14159265359
78 #endif
79
80 #if SCM_HAVE_T_UINT64
81
82 unsigned long
83 scm_i_uniform32 (scm_t_i_rstate *state)
84 {
85 scm_t_uint64 x = (scm_t_uint64) A * state->w + state->c;
86 scm_t_uint32 w = x & 0xffffffffUL;
87 state->w = w;
88 state->c = x >> 32L;
89 return w;
90 }
91
92 #else
93
94 /* ww This is a portable version of the same RNG without 64 bit
95 * * aa arithmetic.
96 * ----
97 * xx It is only intended to provide identical behaviour on
98 * xx platforms without 8 byte longs or long longs until
99 * xx someone has implemented the routine in assembler code.
100 * xxcc
101 * ----
102 * ccww
103 */
104
105 #define L(x) ((x) & 0xffff)
106 #define H(x) ((x) >> 16)
107
108 unsigned long
109 scm_i_uniform32 (scm_t_i_rstate *state)
110 {
111 scm_t_uint32 x1 = L (A) * L (state->w);
112 scm_t_uint32 x2 = L (A) * H (state->w);
113 scm_t_uint32 x3 = H (A) * L (state->w);
114 scm_t_uint32 w = L (x1) + L (state->c);
115 scm_t_uint32 m = H (x1) + L (x2) + L (x3) + H (state->c) + H (w);
116 scm_t_uint32 x4 = H (A) * H (state->w);
117 state->w = w = (L (m) << 16) + L (w);
118 state->c = H (x2) + H (x3) + x4 + H (m);
119 return w;
120 }
121
122 #endif
123
124 void
125 scm_i_init_rstate (scm_t_i_rstate *state, const char *seed, int n)
126 {
127 scm_t_uint32 w = 0L;
128 scm_t_uint32 c = 0L;
129 int i, m;
130 for (i = 0; i < n; ++i)
131 {
132 m = i % 8;
133 if (m < 4)
134 w += seed[i] << (8 * m);
135 else
136 c += seed[i] << (8 * (m - 4));
137 }
138 if ((w == 0 && c == 0) || (w == -1 && c == A - 1))
139 ++c;
140 state->w = w;
141 state->c = c;
142 }
143
144 scm_t_i_rstate *
145 scm_i_copy_rstate (scm_t_i_rstate *state)
146 {
147 scm_t_rstate *new_state;
148
149 new_state = scm_gc_malloc_pointerless (scm_the_rng.rstate_size,
150 "random-state");
151 return memcpy (new_state, state, scm_the_rng.rstate_size);
152 }
153
154 \f
155 /*
156 * Random number library functions
157 */
158
159 scm_t_rstate *
160 scm_c_make_rstate (const char *seed, int n)
161 {
162 scm_t_rstate *state;
163
164 state = scm_gc_malloc_pointerless (scm_the_rng.rstate_size,
165 "random-state");
166 state->reserved0 = 0;
167 scm_the_rng.init_rstate (state, seed, n);
168 return state;
169 }
170
171
172 scm_t_rstate *
173 scm_c_default_rstate ()
174 #define FUNC_NAME "scm_c_default_rstate"
175 {
176 SCM state = SCM_VARIABLE_REF (scm_var_random_state);
177 if (!SCM_RSTATEP (state))
178 SCM_MISC_ERROR ("*random-state* contains bogus random state", SCM_EOL);
179 return SCM_RSTATE (state);
180 }
181 #undef FUNC_NAME
182
183
184 inline double
185 scm_c_uniform01 (scm_t_rstate *state)
186 {
187 double x = (double) scm_the_rng.random_bits (state) / (double) 0xffffffffUL;
188 return ((x + (double) scm_the_rng.random_bits (state))
189 / (double) 0xffffffffUL);
190 }
191
192 double
193 scm_c_normal01 (scm_t_rstate *state)
194 {
195 if (state->reserved0)
196 {
197 state->reserved0 = 0;
198 return state->reserved1;
199 }
200 else
201 {
202 double r, a, n;
203
204 r = sqrt (-2.0 * log (scm_c_uniform01 (state)));
205 a = 2.0 * M_PI * scm_c_uniform01 (state);
206
207 n = r * sin (a);
208 state->reserved1 = r * cos (a);
209 state->reserved0 = 1;
210
211 return n;
212 }
213 }
214
215 double
216 scm_c_exp1 (scm_t_rstate *state)
217 {
218 return - log (scm_c_uniform01 (state));
219 }
220
221 unsigned char scm_masktab[256];
222
223 unsigned long
224 scm_c_random (scm_t_rstate *state, unsigned long m)
225 {
226 unsigned int r, mask;
227 mask = (m < 0x100
228 ? scm_masktab[m]
229 : (m < 0x10000
230 ? scm_masktab[m >> 8] << 8 | 0xff
231 : (m < 0x1000000
232 ? scm_masktab[m >> 16] << 16 | 0xffff
233 : scm_masktab[m >> 24] << 24 | 0xffffff)));
234 while ((r = scm_the_rng.random_bits (state) & mask) >= m);
235 return r;
236 }
237
238 /*
239 SCM scm_c_random_bignum (scm_t_rstate *state, SCM m)
240
241 Takes a random state (source of random bits) and a bignum m.
242 Returns a bignum b, 0 <= b < m.
243
244 It does this by allocating a bignum b with as many base 65536 digits
245 as m, filling b with random bits (in 32 bit chunks) up to the most
246 significant 1 in m, and, finally checking if the resultant b is too
247 large (>= m). If too large, we simply repeat the process again. (It
248 is important to throw away all generated random bits if b >= m,
249 otherwise we'll end up with a distorted distribution.)
250
251 */
252
253 SCM
254 scm_c_random_bignum (scm_t_rstate *state, SCM m)
255 {
256 SCM result = scm_i_mkbig ();
257 const size_t m_bits = mpz_sizeinbase (SCM_I_BIG_MPZ (m), 2);
258 /* how many bits would only partially fill the last unsigned long? */
259 const size_t end_bits = m_bits % (sizeof (unsigned long) * SCM_CHAR_BIT);
260 unsigned long *random_chunks = NULL;
261 const unsigned long num_full_chunks =
262 m_bits / (sizeof (unsigned long) * SCM_CHAR_BIT);
263 const unsigned long num_chunks = num_full_chunks + ((end_bits) ? 1 : 0);
264
265 /* we know the result will be this big */
266 mpz_realloc2 (SCM_I_BIG_MPZ (result), m_bits);
267
268 random_chunks =
269 (unsigned long *) scm_gc_calloc (num_chunks * sizeof (unsigned long),
270 "random bignum chunks");
271
272 do
273 {
274 unsigned long *current_chunk = random_chunks + (num_chunks - 1);
275 unsigned long chunks_left = num_chunks;
276
277 mpz_set_ui (SCM_I_BIG_MPZ (result), 0);
278
279 if (end_bits)
280 {
281 /* generate a mask with ones in the end_bits position, i.e. if
282 end_bits is 3, then we'd have a mask of ...0000000111 */
283 const unsigned long rndbits = scm_the_rng.random_bits (state);
284 int rshift = (sizeof (unsigned long) * SCM_CHAR_BIT) - end_bits;
285 unsigned long mask = ((unsigned long) ULONG_MAX) >> rshift;
286 unsigned long highest_bits = rndbits & mask;
287 *current_chunk-- = highest_bits;
288 chunks_left--;
289 }
290
291 while (chunks_left)
292 {
293 /* now fill in the remaining unsigned long sized chunks */
294 *current_chunk-- = scm_the_rng.random_bits (state);
295 chunks_left--;
296 }
297 mpz_import (SCM_I_BIG_MPZ (result),
298 num_chunks,
299 -1,
300 sizeof (unsigned long),
301 0,
302 0,
303 random_chunks);
304 /* if result >= m, regenerate it (it is important to regenerate
305 all bits in order not to get a distorted distribution) */
306 } while (mpz_cmp (SCM_I_BIG_MPZ (result), SCM_I_BIG_MPZ (m)) >= 0);
307 scm_gc_free (random_chunks,
308 num_chunks * sizeof (unsigned long),
309 "random bignum chunks");
310 return scm_i_normbig (result);
311 }
312
313 /*
314 * Scheme level representation of random states.
315 */
316
317 scm_t_bits scm_tc16_rstate;
318
319 static SCM
320 make_rstate (scm_t_rstate *state)
321 {
322 SCM_RETURN_NEWSMOB (scm_tc16_rstate, state);
323 }
324
325
326 /*
327 * Scheme level interface.
328 */
329
330 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")));
331
332 SCM_DEFINE (scm_random, "random", 1, 1, 0,
333 (SCM n, SCM state),
334 "Return a number in [0, N).\n"
335 "\n"
336 "Accepts a positive integer or real n and returns a\n"
337 "number of the same type between zero (inclusive) and\n"
338 "N (exclusive). The values returned have a uniform\n"
339 "distribution.\n"
340 "\n"
341 "The optional argument @var{state} must be of the type produced\n"
342 "by @code{seed->random-state}. It defaults to the value of the\n"
343 "variable @var{*random-state*}. This object is used to maintain\n"
344 "the state of the pseudo-random-number generator and is altered\n"
345 "as a side effect of the random operation.")
346 #define FUNC_NAME s_scm_random
347 {
348 if (SCM_UNBNDP (state))
349 state = SCM_VARIABLE_REF (scm_var_random_state);
350 SCM_VALIDATE_RSTATE (2, state);
351 if (SCM_I_INUMP (n))
352 {
353 unsigned long m = SCM_I_INUM (n);
354 SCM_ASSERT_RANGE (1, n, m > 0);
355 return scm_from_ulong (scm_c_random (SCM_RSTATE (state), m));
356 }
357 SCM_VALIDATE_NIM (1, n);
358 if (SCM_REALP (n))
359 return scm_from_double (SCM_REAL_VALUE (n)
360 * scm_c_uniform01 (SCM_RSTATE (state)));
361
362 if (!SCM_BIGP (n))
363 SCM_WRONG_TYPE_ARG (1, n);
364 return scm_c_random_bignum (SCM_RSTATE (state), n);
365 }
366 #undef FUNC_NAME
367
368 SCM_DEFINE (scm_copy_random_state, "copy-random-state", 0, 1, 0,
369 (SCM state),
370 "Return a copy of the random state @var{state}.")
371 #define FUNC_NAME s_scm_copy_random_state
372 {
373 if (SCM_UNBNDP (state))
374 state = SCM_VARIABLE_REF (scm_var_random_state);
375 SCM_VALIDATE_RSTATE (1, state);
376 return make_rstate (scm_the_rng.copy_rstate (SCM_RSTATE (state)));
377 }
378 #undef FUNC_NAME
379
380 SCM_DEFINE (scm_seed_to_random_state, "seed->random-state", 1, 0, 0,
381 (SCM seed),
382 "Return a new random state using @var{seed}.")
383 #define FUNC_NAME s_scm_seed_to_random_state
384 {
385 SCM res;
386 if (SCM_NUMBERP (seed))
387 seed = scm_number_to_string (seed, SCM_UNDEFINED);
388 SCM_VALIDATE_STRING (1, seed);
389 res = make_rstate (scm_c_make_rstate (scm_i_string_chars (seed),
390 scm_i_string_length (seed)));
391 scm_remember_upto_here_1 (seed);
392 return res;
393
394 }
395 #undef FUNC_NAME
396
397 SCM_DEFINE (scm_random_uniform, "random:uniform", 0, 1, 0,
398 (SCM state),
399 "Return a uniformly distributed inexact real random number in\n"
400 "[0,1).")
401 #define FUNC_NAME s_scm_random_uniform
402 {
403 if (SCM_UNBNDP (state))
404 state = SCM_VARIABLE_REF (scm_var_random_state);
405 SCM_VALIDATE_RSTATE (1, state);
406 return scm_from_double (scm_c_uniform01 (SCM_RSTATE (state)));
407 }
408 #undef FUNC_NAME
409
410 SCM_DEFINE (scm_random_normal, "random:normal", 0, 1, 0,
411 (SCM state),
412 "Return an inexact real in a normal distribution. The\n"
413 "distribution used has mean 0 and standard deviation 1. For a\n"
414 "normal distribution with mean m and standard deviation d use\n"
415 "@code{(+ m (* d (random:normal)))}.")
416 #define FUNC_NAME s_scm_random_normal
417 {
418 if (SCM_UNBNDP (state))
419 state = SCM_VARIABLE_REF (scm_var_random_state);
420 SCM_VALIDATE_RSTATE (1, state);
421 return scm_from_double (scm_c_normal01 (SCM_RSTATE (state)));
422 }
423 #undef FUNC_NAME
424
425 static void
426 vector_scale_x (SCM v, double c)
427 {
428 size_t n;
429 if (scm_is_simple_vector (v))
430 {
431 n = SCM_SIMPLE_VECTOR_LENGTH (v);
432 while (n-- > 0)
433 SCM_REAL_VALUE (SCM_SIMPLE_VECTOR_REF (v, n)) *= c;
434 }
435 else
436 {
437 /* must be a f64vector. */
438 scm_t_array_handle handle;
439 size_t i, len;
440 ssize_t inc;
441 double *elts;
442
443 elts = scm_f64vector_writable_elements (v, &handle, &len, &inc);
444
445 for (i = 0; i < len; i++, elts += inc)
446 *elts *= c;
447
448 scm_array_handle_release (&handle);
449 }
450 }
451
452 static double
453 vector_sum_squares (SCM v)
454 {
455 double x, sum = 0.0;
456 size_t n;
457 if (scm_is_simple_vector (v))
458 {
459 n = SCM_SIMPLE_VECTOR_LENGTH (v);
460 while (n-- > 0)
461 {
462 x = SCM_REAL_VALUE (SCM_SIMPLE_VECTOR_REF (v, n));
463 sum += x * x;
464 }
465 }
466 else
467 {
468 /* must be a f64vector. */
469 scm_t_array_handle handle;
470 size_t i, len;
471 ssize_t inc;
472 const double *elts;
473
474 elts = scm_f64vector_elements (v, &handle, &len, &inc);
475
476 for (i = 0; i < len; i++, elts += inc)
477 {
478 x = *elts;
479 sum += x * x;
480 }
481
482 scm_array_handle_release (&handle);
483 }
484 return sum;
485 }
486
487 /* For the uniform distribution on the solid sphere, note that in
488 * this distribution the length r of the vector has cumulative
489 * distribution r^n; i.e., u=r^n is uniform [0,1], so r can be
490 * generated as r=u^(1/n).
491 */
492 SCM_DEFINE (scm_random_solid_sphere_x, "random:solid-sphere!", 1, 1, 0,
493 (SCM v, SCM state),
494 "Fills @var{vect} with inexact real random numbers the sum of\n"
495 "whose squares is less than 1.0. Thinking of @var{vect} as\n"
496 "coordinates in space of dimension @var{n} @math{=}\n"
497 "@code{(vector-length @var{vect})}, the coordinates are\n"
498 "uniformly distributed within the unit @var{n}-sphere.")
499 #define FUNC_NAME s_scm_random_solid_sphere_x
500 {
501 if (SCM_UNBNDP (state))
502 state = SCM_VARIABLE_REF (scm_var_random_state);
503 SCM_VALIDATE_RSTATE (2, state);
504 scm_random_normal_vector_x (v, state);
505 vector_scale_x (v,
506 pow (scm_c_uniform01 (SCM_RSTATE (state)),
507 1.0 / scm_c_generalized_vector_length (v))
508 / sqrt (vector_sum_squares (v)));
509 return SCM_UNSPECIFIED;
510 }
511 #undef FUNC_NAME
512
513 SCM_DEFINE (scm_random_hollow_sphere_x, "random:hollow-sphere!", 1, 1, 0,
514 (SCM v, SCM state),
515 "Fills vect with inexact real random numbers\n"
516 "the sum of whose squares is equal to 1.0.\n"
517 "Thinking of vect as coordinates in space of\n"
518 "dimension n = (vector-length vect), the coordinates\n"
519 "are uniformly distributed over the surface of the\n"
520 "unit n-sphere.")
521 #define FUNC_NAME s_scm_random_hollow_sphere_x
522 {
523 if (SCM_UNBNDP (state))
524 state = SCM_VARIABLE_REF (scm_var_random_state);
525 SCM_VALIDATE_RSTATE (2, state);
526 scm_random_normal_vector_x (v, state);
527 vector_scale_x (v, 1 / sqrt (vector_sum_squares (v)));
528 return SCM_UNSPECIFIED;
529 }
530 #undef FUNC_NAME
531
532
533 SCM_DEFINE (scm_random_normal_vector_x, "random:normal-vector!", 1, 1, 0,
534 (SCM v, SCM state),
535 "Fills vect with inexact real random numbers that are\n"
536 "independent and standard normally distributed\n"
537 "(i.e., with mean 0 and variance 1).")
538 #define FUNC_NAME s_scm_random_normal_vector_x
539 {
540 long i;
541 scm_t_array_handle handle;
542 scm_t_array_dim *dim;
543
544 if (SCM_UNBNDP (state))
545 state = SCM_VARIABLE_REF (scm_var_random_state);
546 SCM_VALIDATE_RSTATE (2, state);
547
548 scm_generalized_vector_get_handle (v, &handle);
549 dim = scm_array_handle_dims (&handle);
550
551 if (scm_is_vector (v))
552 {
553 SCM *elts = scm_array_handle_writable_elements (&handle);
554 for (i = dim->lbnd; i <= dim->ubnd; i++, elts += dim->inc)
555 *elts = scm_from_double (scm_c_normal01 (SCM_RSTATE (state)));
556 }
557 else
558 {
559 /* must be a f64vector. */
560 double *elts = scm_array_handle_f64_writable_elements (&handle);
561 for (i = dim->lbnd; i <= dim->ubnd; i++, elts += dim->inc)
562 *elts = scm_c_normal01 (SCM_RSTATE (state));
563 }
564
565 scm_array_handle_release (&handle);
566
567 return SCM_UNSPECIFIED;
568 }
569 #undef FUNC_NAME
570
571 SCM_DEFINE (scm_random_exp, "random:exp", 0, 1, 0,
572 (SCM state),
573 "Return an inexact real in an exponential distribution with mean\n"
574 "1. For an exponential distribution with mean u use (* u\n"
575 "(random:exp)).")
576 #define FUNC_NAME s_scm_random_exp
577 {
578 if (SCM_UNBNDP (state))
579 state = SCM_VARIABLE_REF (scm_var_random_state);
580 SCM_VALIDATE_RSTATE (1, state);
581 return scm_from_double (scm_c_exp1 (SCM_RSTATE (state)));
582 }
583 #undef FUNC_NAME
584
585 void
586 scm_init_random ()
587 {
588 int i, m;
589 /* plug in default RNG */
590 scm_t_rng rng =
591 {
592 sizeof (scm_t_i_rstate),
593 (unsigned long (*)()) scm_i_uniform32,
594 (void (*)()) scm_i_init_rstate,
595 (scm_t_rstate *(*)()) scm_i_copy_rstate
596 };
597 scm_the_rng = rng;
598
599 scm_tc16_rstate = scm_make_smob_type ("random-state", 0);
600
601 for (m = 1; m <= 0x100; m <<= 1)
602 for (i = m >> 1; i < m; ++i)
603 scm_masktab[i] = m - 1;
604
605 #include "libguile/random.x"
606
607 scm_add_feature ("random");
608 }
609
610 /*
611 Local Variables:
612 c-file-style: "gnu"
613 End:
614 */