Commit | Line | Data |
---|---|---|
b70021f4 | 1 | /* Primitive operations on floating point for GNU Emacs Lisp interpreter. |
3a22ee35 | 2 | Copyright (C) 1988, 1993, 1994 Free Software Foundation, Inc. |
b70021f4 MR |
3 | |
4 | This file is part of GNU Emacs. | |
5 | ||
6 | GNU Emacs is free software; you can redistribute it and/or modify | |
7 | it under the terms of the GNU General Public License as published by | |
4746118a | 8 | the Free Software Foundation; either version 2, or (at your option) |
b70021f4 MR |
9 | any later version. |
10 | ||
11 | GNU Emacs is distributed in the hope that it will be useful, | |
12 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
14 | GNU General Public License for more details. | |
15 | ||
16 | You should have received a copy of the GNU General Public License | |
17 | along with GNU Emacs; see the file COPYING. If not, write to | |
3b7ad313 EN |
18 | the Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
19 | Boston, MA 02111-1307, USA. */ | |
b70021f4 MR |
20 | |
21 | ||
4b6baf5f RS |
22 | /* ANSI C requires only these float functions: |
23 | acos, asin, atan, atan2, ceil, cos, cosh, exp, fabs, floor, fmod, | |
24 | frexp, ldexp, log, log10, modf, pow, sin, sinh, sqrt, tan, tanh. | |
25 | ||
26 | Define HAVE_INVERSE_HYPERBOLIC if you have acosh, asinh, and atanh. | |
27 | Define HAVE_CBRT if you have cbrt. | |
28 | Define HAVE_RINT if you have rint. | |
29 | If you don't define these, then the appropriate routines will be simulated. | |
30 | ||
31 | Define HAVE_MATHERR if on a system supporting the SysV matherr callback. | |
32 | (This should happen automatically.) | |
33 | ||
34 | Define FLOAT_CHECK_ERRNO if the float library routines set errno. | |
35 | This has no effect if HAVE_MATHERR is defined. | |
36 | ||
37 | Define FLOAT_CATCH_SIGILL if the float library routines signal SIGILL. | |
38 | (What systems actually do this? Please let us know.) | |
39 | ||
40 | Define FLOAT_CHECK_DOMAIN if the float library doesn't handle errors by | |
8e6208c5 | 41 | either setting errno, or signaling SIGFPE/SIGILL. Otherwise, domain and |
4b6baf5f RS |
42 | range checking will happen before calling the float routines. This has |
43 | no effect if HAVE_MATHERR is defined (since matherr will be called when | |
44 | a domain error occurs.) | |
45 | */ | |
46 | ||
b70021f4 MR |
47 | #include <signal.h> |
48 | ||
18160b98 | 49 | #include <config.h> |
b70021f4 | 50 | #include "lisp.h" |
e065a56e | 51 | #include "syssignal.h" |
b70021f4 | 52 | |
b70021f4 | 53 | #ifdef LISP_FLOAT_TYPE |
265a9e55 | 54 | |
d137ae2f PE |
55 | #if STDC_HEADERS |
56 | #include <float.h> | |
57 | #endif | |
58 | ||
59 | /* If IEEE_FLOATING_POINT isn't defined, default it from FLT_*. */ | |
60 | #ifndef IEEE_FLOATING_POINT | |
61 | #if (FLT_RADIX == 2 && FLT_MANT_DIG == 24 \ | |
62 | && FLT_MIN_EXP == -125 && FLT_MAX_EXP == 128) | |
63 | #define IEEE_FLOATING_POINT 1 | |
64 | #else | |
65 | #define IEEE_FLOATING_POINT 0 | |
66 | #endif | |
67 | #endif | |
68 | ||
4cd7a373 RS |
69 | /* Work around a problem that happens because math.h on hpux 7 |
70 | defines two static variables--which, in Emacs, are not really static, | |
71 | because `static' is defined as nothing. The problem is that they are | |
72 | defined both here and in lread.c. | |
73 | These macros prevent the name conflict. */ | |
74 | #if defined (HPUX) && !defined (HPUX8) | |
75 | #define _MAXLDBL floatfns_maxldbl | |
76 | #define _NMAXLDBL floatfns_nmaxldbl | |
77 | #endif | |
78 | ||
b70021f4 | 79 | #include <math.h> |
4b6baf5f | 80 | |
32085e8e | 81 | /* This declaration is omitted on some systems, like Ultrix. */ |
7a4720e2 | 82 | #if !defined (HPUX) && defined (HAVE_LOGB) && !defined (logb) |
c26406fe | 83 | extern double logb (); |
7a4720e2 | 84 | #endif /* not HPUX and HAVE_LOGB and no logb macro */ |
c26406fe | 85 | |
4b6baf5f RS |
86 | #if defined(DOMAIN) && defined(SING) && defined(OVERFLOW) |
87 | /* If those are defined, then this is probably a `matherr' machine. */ | |
88 | # ifndef HAVE_MATHERR | |
89 | # define HAVE_MATHERR | |
90 | # endif | |
91 | #endif | |
92 | ||
c0f0a4a2 | 93 | #ifdef NO_MATHERR |
f89182a2 RS |
94 | #undef HAVE_MATHERR |
95 | #endif | |
96 | ||
4b6baf5f RS |
97 | #ifdef HAVE_MATHERR |
98 | # ifdef FLOAT_CHECK_ERRNO | |
99 | # undef FLOAT_CHECK_ERRNO | |
100 | # endif | |
101 | # ifdef FLOAT_CHECK_DOMAIN | |
102 | # undef FLOAT_CHECK_DOMAIN | |
103 | # endif | |
104 | #endif | |
105 | ||
106 | #ifndef NO_FLOAT_CHECK_ERRNO | |
107 | #define FLOAT_CHECK_ERRNO | |
108 | #endif | |
109 | ||
110 | #ifdef FLOAT_CHECK_ERRNO | |
111 | # include <errno.h> | |
265a9e55 JB |
112 | |
113 | extern int errno; | |
4b6baf5f | 114 | #endif |
265a9e55 JB |
115 | |
116 | /* Avoid traps on VMS from sinh and cosh. | |
117 | All the other functions set errno instead. */ | |
118 | ||
119 | #ifdef VMS | |
120 | #undef cosh | |
121 | #undef sinh | |
122 | #define cosh(x) ((exp(x)+exp(-x))*0.5) | |
123 | #define sinh(x) ((exp(x)-exp(-x))*0.5) | |
124 | #endif /* VMS */ | |
125 | ||
4746118a | 126 | static SIGTYPE float_error (); |
b70021f4 MR |
127 | |
128 | /* Nonzero while executing in floating point. | |
129 | This tells float_error what to do. */ | |
130 | ||
131 | static int in_float; | |
132 | ||
133 | /* If an argument is out of range for a mathematical function, | |
21876236 RS |
134 | here is the actual argument value to use in the error message. |
135 | These variables are used only across the floating point library call | |
136 | so there is no need to staticpro them. */ | |
b70021f4 | 137 | |
4b6baf5f RS |
138 | static Lisp_Object float_error_arg, float_error_arg2; |
139 | ||
140 | static char *float_error_fn_name; | |
b70021f4 | 141 | |
265a9e55 JB |
142 | /* Evaluate the floating point expression D, recording NUM |
143 | as the original argument for error messages. | |
144 | D is normally an assignment expression. | |
f8d83099 JB |
145 | Handle errors which may result in signals or may set errno. |
146 | ||
147 | Note that float_error may be declared to return void, so you can't | |
148 | just cast the zero after the colon to (SIGTYPE) to make the types | |
149 | check properly. */ | |
265a9e55 | 150 | |
4b6baf5f RS |
151 | #ifdef FLOAT_CHECK_ERRNO |
152 | #define IN_FLOAT(d, name, num) \ | |
153 | do { \ | |
154 | float_error_arg = num; \ | |
155 | float_error_fn_name = name; \ | |
156 | in_float = 1; errno = 0; (d); in_float = 0; \ | |
157 | switch (errno) { \ | |
158 | case 0: break; \ | |
159 | case EDOM: domain_error (float_error_fn_name, float_error_arg); \ | |
160 | case ERANGE: range_error (float_error_fn_name, float_error_arg); \ | |
161 | default: arith_error (float_error_fn_name, float_error_arg); \ | |
162 | } \ | |
163 | } while (0) | |
164 | #define IN_FLOAT2(d, name, num, num2) \ | |
165 | do { \ | |
166 | float_error_arg = num; \ | |
167 | float_error_arg2 = num2; \ | |
168 | float_error_fn_name = name; \ | |
169 | in_float = 1; errno = 0; (d); in_float = 0; \ | |
170 | switch (errno) { \ | |
171 | case 0: break; \ | |
172 | case EDOM: domain_error (float_error_fn_name, float_error_arg); \ | |
173 | case ERANGE: range_error (float_error_fn_name, float_error_arg); \ | |
174 | default: arith_error (float_error_fn_name, float_error_arg); \ | |
175 | } \ | |
176 | } while (0) | |
177 | #else | |
f8131ed2 | 178 | #define IN_FLOAT(d, name, num) (in_float = 1, (d), in_float = 0) |
4b6baf5f RS |
179 | #define IN_FLOAT2(d, name, num, num2) (in_float = 1, (d), in_float = 0) |
180 | #endif | |
181 | ||
81a63ccc KH |
182 | /* Convert float to Lisp_Int if it fits, else signal a range error |
183 | using the given arguments. */ | |
184 | #define FLOAT_TO_INT(x, i, name, num) \ | |
185 | do \ | |
186 | { \ | |
7936dcdb RS |
187 | if ((x) >= (((EMACS_INT) 1) << (VALBITS-1)) || \ |
188 | (x) <= - (((EMACS_INT) 1) << (VALBITS-1)) - 1) \ | |
81a63ccc | 189 | range_error (name, num); \ |
e0cb2a68 | 190 | XSETINT (i, (EMACS_INT)(x)); \ |
81a63ccc KH |
191 | } \ |
192 | while (0) | |
193 | #define FLOAT_TO_INT2(x, i, name, num1, num2) \ | |
194 | do \ | |
195 | { \ | |
7936dcdb RS |
196 | if ((x) >= (((EMACS_INT) 1) << (VALBITS-1)) || \ |
197 | (x) <= - (((EMACS_INT) 1) << (VALBITS-1)) - 1) \ | |
81a63ccc | 198 | range_error2 (name, num1, num2); \ |
e0cb2a68 | 199 | XSETINT (i, (EMACS_INT)(x)); \ |
81a63ccc KH |
200 | } \ |
201 | while (0) | |
202 | ||
4b6baf5f RS |
203 | #define arith_error(op,arg) \ |
204 | Fsignal (Qarith_error, Fcons (build_string ((op)), Fcons ((arg), Qnil))) | |
205 | #define range_error(op,arg) \ | |
206 | Fsignal (Qrange_error, Fcons (build_string ((op)), Fcons ((arg), Qnil))) | |
81a63ccc KH |
207 | #define range_error2(op,a1,a2) \ |
208 | Fsignal (Qrange_error, Fcons (build_string ((op)), \ | |
209 | Fcons ((a1), Fcons ((a2), Qnil)))) | |
4b6baf5f RS |
210 | #define domain_error(op,arg) \ |
211 | Fsignal (Qdomain_error, Fcons (build_string ((op)), Fcons ((arg), Qnil))) | |
212 | #define domain_error2(op,a1,a2) \ | |
81a63ccc KH |
213 | Fsignal (Qdomain_error, Fcons (build_string ((op)), \ |
214 | Fcons ((a1), Fcons ((a2), Qnil)))) | |
b70021f4 MR |
215 | |
216 | /* Extract a Lisp number as a `double', or signal an error. */ | |
217 | ||
218 | double | |
219 | extract_float (num) | |
220 | Lisp_Object num; | |
221 | { | |
222 | CHECK_NUMBER_OR_FLOAT (num, 0); | |
223 | ||
207a45c1 | 224 | if (FLOATP (num)) |
b70021f4 MR |
225 | return XFLOAT (num)->data; |
226 | return (double) XINT (num); | |
227 | } | |
c2d4ea74 RS |
228 | \f |
229 | /* Trig functions. */ | |
b70021f4 MR |
230 | |
231 | DEFUN ("acos", Facos, Sacos, 1, 1, 0, | |
232 | "Return the inverse cosine of ARG.") | |
4b6baf5f RS |
233 | (arg) |
234 | register Lisp_Object arg; | |
b70021f4 | 235 | { |
4b6baf5f RS |
236 | double d = extract_float (arg); |
237 | #ifdef FLOAT_CHECK_DOMAIN | |
238 | if (d > 1.0 || d < -1.0) | |
239 | domain_error ("acos", arg); | |
240 | #endif | |
241 | IN_FLOAT (d = acos (d), "acos", arg); | |
b70021f4 MR |
242 | return make_float (d); |
243 | } | |
244 | ||
c2d4ea74 RS |
245 | DEFUN ("asin", Fasin, Sasin, 1, 1, 0, |
246 | "Return the inverse sine of ARG.") | |
4b6baf5f RS |
247 | (arg) |
248 | register Lisp_Object arg; | |
b70021f4 | 249 | { |
4b6baf5f RS |
250 | double d = extract_float (arg); |
251 | #ifdef FLOAT_CHECK_DOMAIN | |
252 | if (d > 1.0 || d < -1.0) | |
253 | domain_error ("asin", arg); | |
254 | #endif | |
255 | IN_FLOAT (d = asin (d), "asin", arg); | |
b70021f4 MR |
256 | return make_float (d); |
257 | } | |
258 | ||
c2d4ea74 RS |
259 | DEFUN ("atan", Fatan, Satan, 1, 1, 0, |
260 | "Return the inverse tangent of ARG.") | |
4b6baf5f RS |
261 | (arg) |
262 | register Lisp_Object arg; | |
b70021f4 | 263 | { |
4b6baf5f RS |
264 | double d = extract_float (arg); |
265 | IN_FLOAT (d = atan (d), "atan", arg); | |
b70021f4 MR |
266 | return make_float (d); |
267 | } | |
268 | ||
c2d4ea74 RS |
269 | DEFUN ("cos", Fcos, Scos, 1, 1, 0, |
270 | "Return the cosine of ARG.") | |
4b6baf5f RS |
271 | (arg) |
272 | register Lisp_Object arg; | |
b70021f4 | 273 | { |
4b6baf5f RS |
274 | double d = extract_float (arg); |
275 | IN_FLOAT (d = cos (d), "cos", arg); | |
b70021f4 MR |
276 | return make_float (d); |
277 | } | |
278 | ||
c2d4ea74 RS |
279 | DEFUN ("sin", Fsin, Ssin, 1, 1, 0, |
280 | "Return the sine of ARG.") | |
4b6baf5f RS |
281 | (arg) |
282 | register Lisp_Object arg; | |
b70021f4 | 283 | { |
4b6baf5f RS |
284 | double d = extract_float (arg); |
285 | IN_FLOAT (d = sin (d), "sin", arg); | |
b70021f4 MR |
286 | return make_float (d); |
287 | } | |
288 | ||
c2d4ea74 RS |
289 | DEFUN ("tan", Ftan, Stan, 1, 1, 0, |
290 | "Return the tangent of ARG.") | |
4b6baf5f RS |
291 | (arg) |
292 | register Lisp_Object arg; | |
293 | { | |
294 | double d = extract_float (arg); | |
295 | double c = cos (d); | |
296 | #ifdef FLOAT_CHECK_DOMAIN | |
297 | if (c == 0.0) | |
298 | domain_error ("tan", arg); | |
299 | #endif | |
300 | IN_FLOAT (d = sin (d) / c, "tan", arg); | |
b70021f4 MR |
301 | return make_float (d); |
302 | } | |
303 | \f | |
c2d4ea74 RS |
304 | #if 0 /* Leave these out unless we find there's a reason for them. */ |
305 | ||
b70021f4 MR |
306 | DEFUN ("bessel-j0", Fbessel_j0, Sbessel_j0, 1, 1, 0, |
307 | "Return the bessel function j0 of ARG.") | |
4b6baf5f RS |
308 | (arg) |
309 | register Lisp_Object arg; | |
b70021f4 | 310 | { |
4b6baf5f RS |
311 | double d = extract_float (arg); |
312 | IN_FLOAT (d = j0 (d), "bessel-j0", arg); | |
b70021f4 MR |
313 | return make_float (d); |
314 | } | |
315 | ||
316 | DEFUN ("bessel-j1", Fbessel_j1, Sbessel_j1, 1, 1, 0, | |
317 | "Return the bessel function j1 of ARG.") | |
4b6baf5f RS |
318 | (arg) |
319 | register Lisp_Object arg; | |
b70021f4 | 320 | { |
4b6baf5f RS |
321 | double d = extract_float (arg); |
322 | IN_FLOAT (d = j1 (d), "bessel-j1", arg); | |
b70021f4 MR |
323 | return make_float (d); |
324 | } | |
325 | ||
326 | DEFUN ("bessel-jn", Fbessel_jn, Sbessel_jn, 2, 2, 0, | |
327 | "Return the order N bessel function output jn of ARG.\n\ | |
328 | The first arg (the order) is truncated to an integer.") | |
3e670702 EN |
329 | (n, arg) |
330 | register Lisp_Object n, arg; | |
b70021f4 | 331 | { |
3e670702 EN |
332 | int i1 = extract_float (n); |
333 | double f2 = extract_float (arg); | |
b70021f4 | 334 | |
3e670702 | 335 | IN_FLOAT (f2 = jn (i1, f2), "bessel-jn", n); |
b70021f4 MR |
336 | return make_float (f2); |
337 | } | |
338 | ||
339 | DEFUN ("bessel-y0", Fbessel_y0, Sbessel_y0, 1, 1, 0, | |
340 | "Return the bessel function y0 of ARG.") | |
4b6baf5f RS |
341 | (arg) |
342 | register Lisp_Object arg; | |
b70021f4 | 343 | { |
4b6baf5f RS |
344 | double d = extract_float (arg); |
345 | IN_FLOAT (d = y0 (d), "bessel-y0", arg); | |
b70021f4 MR |
346 | return make_float (d); |
347 | } | |
348 | ||
349 | DEFUN ("bessel-y1", Fbessel_y1, Sbessel_y1, 1, 1, 0, | |
350 | "Return the bessel function y1 of ARG.") | |
4b6baf5f RS |
351 | (arg) |
352 | register Lisp_Object arg; | |
b70021f4 | 353 | { |
4b6baf5f RS |
354 | double d = extract_float (arg); |
355 | IN_FLOAT (d = y1 (d), "bessel-y0", arg); | |
b70021f4 MR |
356 | return make_float (d); |
357 | } | |
358 | ||
359 | DEFUN ("bessel-yn", Fbessel_yn, Sbessel_yn, 2, 2, 0, | |
360 | "Return the order N bessel function output yn of ARG.\n\ | |
361 | The first arg (the order) is truncated to an integer.") | |
3e670702 EN |
362 | (n, arg) |
363 | register Lisp_Object n, arg; | |
b70021f4 | 364 | { |
3e670702 EN |
365 | int i1 = extract_float (n); |
366 | double f2 = extract_float (arg); | |
b70021f4 | 367 | |
3e670702 | 368 | IN_FLOAT (f2 = yn (i1, f2), "bessel-yn", n); |
b70021f4 MR |
369 | return make_float (f2); |
370 | } | |
b70021f4 | 371 | |
c2d4ea74 RS |
372 | #endif |
373 | \f | |
374 | #if 0 /* Leave these out unless we see they are worth having. */ | |
b70021f4 MR |
375 | |
376 | DEFUN ("erf", Ferf, Serf, 1, 1, 0, | |
377 | "Return the mathematical error function of ARG.") | |
4b6baf5f RS |
378 | (arg) |
379 | register Lisp_Object arg; | |
b70021f4 | 380 | { |
4b6baf5f RS |
381 | double d = extract_float (arg); |
382 | IN_FLOAT (d = erf (d), "erf", arg); | |
b70021f4 MR |
383 | return make_float (d); |
384 | } | |
385 | ||
386 | DEFUN ("erfc", Ferfc, Serfc, 1, 1, 0, | |
387 | "Return the complementary error function of ARG.") | |
4b6baf5f RS |
388 | (arg) |
389 | register Lisp_Object arg; | |
b70021f4 | 390 | { |
4b6baf5f RS |
391 | double d = extract_float (arg); |
392 | IN_FLOAT (d = erfc (d), "erfc", arg); | |
b70021f4 MR |
393 | return make_float (d); |
394 | } | |
395 | ||
b70021f4 MR |
396 | DEFUN ("log-gamma", Flog_gamma, Slog_gamma, 1, 1, 0, |
397 | "Return the log gamma of ARG.") | |
4b6baf5f RS |
398 | (arg) |
399 | register Lisp_Object arg; | |
b70021f4 | 400 | { |
4b6baf5f RS |
401 | double d = extract_float (arg); |
402 | IN_FLOAT (d = lgamma (d), "log-gamma", arg); | |
b70021f4 MR |
403 | return make_float (d); |
404 | } | |
405 | ||
4b6baf5f | 406 | DEFUN ("cube-root", Fcube_root, Scube_root, 1, 1, 0, |
c2d4ea74 | 407 | "Return the cube root of ARG.") |
4b6baf5f RS |
408 | (arg) |
409 | register Lisp_Object arg; | |
b70021f4 | 410 | { |
4b6baf5f RS |
411 | double d = extract_float (arg); |
412 | #ifdef HAVE_CBRT | |
413 | IN_FLOAT (d = cbrt (d), "cube-root", arg); | |
414 | #else | |
415 | if (d >= 0.0) | |
416 | IN_FLOAT (d = pow (d, 1.0/3.0), "cube-root", arg); | |
417 | else | |
418 | IN_FLOAT (d = -pow (-d, 1.0/3.0), "cube-root", arg); | |
419 | #endif | |
b70021f4 MR |
420 | return make_float (d); |
421 | } | |
422 | ||
706ac90d RS |
423 | #endif |
424 | \f | |
c2d4ea74 RS |
425 | DEFUN ("exp", Fexp, Sexp, 1, 1, 0, |
426 | "Return the exponential base e of ARG.") | |
4b6baf5f RS |
427 | (arg) |
428 | register Lisp_Object arg; | |
429 | { | |
430 | double d = extract_float (arg); | |
431 | #ifdef FLOAT_CHECK_DOMAIN | |
432 | if (d > 709.7827) /* Assume IEEE doubles here */ | |
433 | range_error ("exp", arg); | |
434 | else if (d < -709.0) | |
435 | return make_float (0.0); | |
436 | else | |
437 | #endif | |
438 | IN_FLOAT (d = exp (d), "exp", arg); | |
b70021f4 MR |
439 | return make_float (d); |
440 | } | |
441 | ||
b70021f4 | 442 | DEFUN ("expt", Fexpt, Sexpt, 2, 2, 0, |
4fd72a6a | 443 | "Return the exponential ARG1 ** ARG2.") |
4b6baf5f RS |
444 | (arg1, arg2) |
445 | register Lisp_Object arg1, arg2; | |
b70021f4 MR |
446 | { |
447 | double f1, f2; | |
448 | ||
4b6baf5f RS |
449 | CHECK_NUMBER_OR_FLOAT (arg1, 0); |
450 | CHECK_NUMBER_OR_FLOAT (arg2, 0); | |
207a45c1 KH |
451 | if (INTEGERP (arg1) /* common lisp spec */ |
452 | && INTEGERP (arg2)) /* don't promote, if both are ints */ | |
b70021f4 | 453 | { /* this can be improved by pre-calculating */ |
9a51b24a | 454 | EMACS_INT acc, x, y; /* some binary powers of x then accumulating */ |
4be1d460 RS |
455 | Lisp_Object val; |
456 | ||
4b6baf5f RS |
457 | x = XINT (arg1); |
458 | y = XINT (arg2); | |
b70021f4 MR |
459 | acc = 1; |
460 | ||
461 | if (y < 0) | |
462 | { | |
4b6baf5f RS |
463 | if (x == 1) |
464 | acc = 1; | |
465 | else if (x == -1) | |
466 | acc = (y & 1) ? -1 : 1; | |
467 | else | |
468 | acc = 0; | |
b70021f4 MR |
469 | } |
470 | else | |
471 | { | |
4b6baf5f RS |
472 | while (y > 0) |
473 | { | |
474 | if (y & 1) | |
475 | acc *= x; | |
476 | x *= x; | |
477 | y = (unsigned)y >> 1; | |
478 | } | |
b70021f4 | 479 | } |
e0cb2a68 | 480 | XSETINT (val, acc); |
4be1d460 | 481 | return val; |
b70021f4 | 482 | } |
207a45c1 KH |
483 | f1 = FLOATP (arg1) ? XFLOAT (arg1)->data : XINT (arg1); |
484 | f2 = FLOATP (arg2) ? XFLOAT (arg2)->data : XINT (arg2); | |
4b6baf5f RS |
485 | /* Really should check for overflow, too */ |
486 | if (f1 == 0.0 && f2 == 0.0) | |
487 | f1 = 1.0; | |
488 | #ifdef FLOAT_CHECK_DOMAIN | |
489 | else if ((f1 == 0.0 && f2 < 0.0) || (f1 < 0 && f2 != floor(f2))) | |
490 | domain_error2 ("expt", arg1, arg2); | |
491 | #endif | |
28d849db | 492 | IN_FLOAT2 (f1 = pow (f1, f2), "expt", arg1, arg2); |
b70021f4 MR |
493 | return make_float (f1); |
494 | } | |
c2d4ea74 | 495 | |
56abb480 | 496 | DEFUN ("log", Flog, Slog, 1, 2, 0, |
4b6baf5f RS |
497 | "Return the natural logarithm of ARG.\n\ |
498 | If second optional argument BASE is given, return log ARG using that base.") | |
499 | (arg, base) | |
500 | register Lisp_Object arg, base; | |
b70021f4 | 501 | { |
4b6baf5f | 502 | double d = extract_float (arg); |
56abb480 | 503 | |
4b6baf5f RS |
504 | #ifdef FLOAT_CHECK_DOMAIN |
505 | if (d <= 0.0) | |
506 | domain_error2 ("log", arg, base); | |
507 | #endif | |
56abb480 | 508 | if (NILP (base)) |
4b6baf5f | 509 | IN_FLOAT (d = log (d), "log", arg); |
56abb480 JB |
510 | else |
511 | { | |
512 | double b = extract_float (base); | |
513 | ||
4b6baf5f RS |
514 | #ifdef FLOAT_CHECK_DOMAIN |
515 | if (b <= 0.0 || b == 1.0) | |
516 | domain_error2 ("log", arg, base); | |
517 | #endif | |
518 | if (b == 10.0) | |
519 | IN_FLOAT2 (d = log10 (d), "log", arg, base); | |
520 | else | |
f8131ed2 | 521 | IN_FLOAT2 (d = log (d) / log (b), "log", arg, base); |
56abb480 | 522 | } |
b70021f4 MR |
523 | return make_float (d); |
524 | } | |
525 | ||
c2d4ea74 RS |
526 | DEFUN ("log10", Flog10, Slog10, 1, 1, 0, |
527 | "Return the logarithm base 10 of ARG.") | |
4b6baf5f RS |
528 | (arg) |
529 | register Lisp_Object arg; | |
b70021f4 | 530 | { |
4b6baf5f RS |
531 | double d = extract_float (arg); |
532 | #ifdef FLOAT_CHECK_DOMAIN | |
533 | if (d <= 0.0) | |
534 | domain_error ("log10", arg); | |
535 | #endif | |
536 | IN_FLOAT (d = log10 (d), "log10", arg); | |
c2d4ea74 RS |
537 | return make_float (d); |
538 | } | |
539 | ||
b70021f4 MR |
540 | DEFUN ("sqrt", Fsqrt, Ssqrt, 1, 1, 0, |
541 | "Return the square root of ARG.") | |
4b6baf5f RS |
542 | (arg) |
543 | register Lisp_Object arg; | |
b70021f4 | 544 | { |
4b6baf5f RS |
545 | double d = extract_float (arg); |
546 | #ifdef FLOAT_CHECK_DOMAIN | |
547 | if (d < 0.0) | |
548 | domain_error ("sqrt", arg); | |
549 | #endif | |
550 | IN_FLOAT (d = sqrt (d), "sqrt", arg); | |
b70021f4 MR |
551 | return make_float (d); |
552 | } | |
c2d4ea74 | 553 | \f |
706ac90d | 554 | #if 0 /* Not clearly worth adding. */ |
b70021f4 | 555 | |
c2d4ea74 RS |
556 | DEFUN ("acosh", Facosh, Sacosh, 1, 1, 0, |
557 | "Return the inverse hyperbolic cosine of ARG.") | |
4b6baf5f RS |
558 | (arg) |
559 | register Lisp_Object arg; | |
b70021f4 | 560 | { |
4b6baf5f RS |
561 | double d = extract_float (arg); |
562 | #ifdef FLOAT_CHECK_DOMAIN | |
563 | if (d < 1.0) | |
564 | domain_error ("acosh", arg); | |
565 | #endif | |
566 | #ifdef HAVE_INVERSE_HYPERBOLIC | |
567 | IN_FLOAT (d = acosh (d), "acosh", arg); | |
568 | #else | |
569 | IN_FLOAT (d = log (d + sqrt (d*d - 1.0)), "acosh", arg); | |
570 | #endif | |
c2d4ea74 RS |
571 | return make_float (d); |
572 | } | |
573 | ||
574 | DEFUN ("asinh", Fasinh, Sasinh, 1, 1, 0, | |
575 | "Return the inverse hyperbolic sine of ARG.") | |
4b6baf5f RS |
576 | (arg) |
577 | register Lisp_Object arg; | |
c2d4ea74 | 578 | { |
4b6baf5f RS |
579 | double d = extract_float (arg); |
580 | #ifdef HAVE_INVERSE_HYPERBOLIC | |
581 | IN_FLOAT (d = asinh (d), "asinh", arg); | |
582 | #else | |
583 | IN_FLOAT (d = log (d + sqrt (d*d + 1.0)), "asinh", arg); | |
584 | #endif | |
c2d4ea74 RS |
585 | return make_float (d); |
586 | } | |
587 | ||
588 | DEFUN ("atanh", Fatanh, Satanh, 1, 1, 0, | |
589 | "Return the inverse hyperbolic tangent of ARG.") | |
4b6baf5f RS |
590 | (arg) |
591 | register Lisp_Object arg; | |
c2d4ea74 | 592 | { |
4b6baf5f RS |
593 | double d = extract_float (arg); |
594 | #ifdef FLOAT_CHECK_DOMAIN | |
595 | if (d >= 1.0 || d <= -1.0) | |
596 | domain_error ("atanh", arg); | |
597 | #endif | |
598 | #ifdef HAVE_INVERSE_HYPERBOLIC | |
599 | IN_FLOAT (d = atanh (d), "atanh", arg); | |
600 | #else | |
601 | IN_FLOAT (d = 0.5 * log ((1.0 + d) / (1.0 - d)), "atanh", arg); | |
602 | #endif | |
c2d4ea74 RS |
603 | return make_float (d); |
604 | } | |
605 | ||
606 | DEFUN ("cosh", Fcosh, Scosh, 1, 1, 0, | |
607 | "Return the hyperbolic cosine of ARG.") | |
4b6baf5f RS |
608 | (arg) |
609 | register Lisp_Object arg; | |
c2d4ea74 | 610 | { |
4b6baf5f RS |
611 | double d = extract_float (arg); |
612 | #ifdef FLOAT_CHECK_DOMAIN | |
613 | if (d > 710.0 || d < -710.0) | |
614 | range_error ("cosh", arg); | |
615 | #endif | |
616 | IN_FLOAT (d = cosh (d), "cosh", arg); | |
c2d4ea74 RS |
617 | return make_float (d); |
618 | } | |
619 | ||
620 | DEFUN ("sinh", Fsinh, Ssinh, 1, 1, 0, | |
621 | "Return the hyperbolic sine of ARG.") | |
4b6baf5f RS |
622 | (arg) |
623 | register Lisp_Object arg; | |
c2d4ea74 | 624 | { |
4b6baf5f RS |
625 | double d = extract_float (arg); |
626 | #ifdef FLOAT_CHECK_DOMAIN | |
627 | if (d > 710.0 || d < -710.0) | |
628 | range_error ("sinh", arg); | |
629 | #endif | |
630 | IN_FLOAT (d = sinh (d), "sinh", arg); | |
b70021f4 MR |
631 | return make_float (d); |
632 | } | |
633 | ||
634 | DEFUN ("tanh", Ftanh, Stanh, 1, 1, 0, | |
635 | "Return the hyperbolic tangent of ARG.") | |
4b6baf5f RS |
636 | (arg) |
637 | register Lisp_Object arg; | |
b70021f4 | 638 | { |
4b6baf5f RS |
639 | double d = extract_float (arg); |
640 | IN_FLOAT (d = tanh (d), "tanh", arg); | |
b70021f4 MR |
641 | return make_float (d); |
642 | } | |
c2d4ea74 | 643 | #endif |
b70021f4 MR |
644 | \f |
645 | DEFUN ("abs", Fabs, Sabs, 1, 1, 0, | |
646 | "Return the absolute value of ARG.") | |
4b6baf5f RS |
647 | (arg) |
648 | register Lisp_Object arg; | |
b70021f4 | 649 | { |
4b6baf5f | 650 | CHECK_NUMBER_OR_FLOAT (arg, 0); |
b70021f4 | 651 | |
207a45c1 | 652 | if (FLOATP (arg)) |
4b6baf5f RS |
653 | IN_FLOAT (arg = make_float (fabs (XFLOAT (arg)->data)), "abs", arg); |
654 | else if (XINT (arg) < 0) | |
db37cb37 | 655 | XSETINT (arg, - XINT (arg)); |
b70021f4 | 656 | |
4b6baf5f | 657 | return arg; |
b70021f4 MR |
658 | } |
659 | ||
660 | DEFUN ("float", Ffloat, Sfloat, 1, 1, 0, | |
661 | "Return the floating point number equal to ARG.") | |
4b6baf5f RS |
662 | (arg) |
663 | register Lisp_Object arg; | |
b70021f4 | 664 | { |
4b6baf5f | 665 | CHECK_NUMBER_OR_FLOAT (arg, 0); |
b70021f4 | 666 | |
207a45c1 | 667 | if (INTEGERP (arg)) |
4b6baf5f | 668 | return make_float ((double) XINT (arg)); |
b70021f4 | 669 | else /* give 'em the same float back */ |
4b6baf5f | 670 | return arg; |
b70021f4 MR |
671 | } |
672 | ||
673 | DEFUN ("logb", Flogb, Slogb, 1, 1, 0, | |
1a3ac8b9 | 674 | "Returns largest integer <= the base 2 log of the magnitude of ARG.\n\ |
b70021f4 | 675 | This is the same as the exponent of a float.") |
4b6baf5f RS |
676 | (arg) |
677 | Lisp_Object arg; | |
b70021f4 | 678 | { |
340176df | 679 | Lisp_Object val; |
a7bf3c54 | 680 | EMACS_INT value; |
5bf54166 | 681 | double f = extract_float (arg); |
340176df | 682 | |
6694b327 KH |
683 | if (f == 0.0) |
684 | value = -(VALMASK >> 1); | |
685 | else | |
686 | { | |
6d3c6adb | 687 | #ifdef HAVE_LOGB |
6694b327 | 688 | IN_FLOAT (value = logb (f), "logb", arg); |
6d3c6adb JB |
689 | #else |
690 | #ifdef HAVE_FREXP | |
c8bf6cf3 KH |
691 | int ivalue; |
692 | IN_FLOAT (frexp (f, &ivalue), "logb", arg); | |
693 | value = ivalue - 1; | |
c26406fe | 694 | #else |
6694b327 KH |
695 | int i; |
696 | double d; | |
697 | if (f < 0.0) | |
698 | f = -f; | |
699 | value = -1; | |
700 | while (f < 0.5) | |
701 | { | |
702 | for (i = 1, d = 0.5; d * d >= f; i += i) | |
703 | d *= d; | |
704 | f /= d; | |
705 | value -= i; | |
706 | } | |
707 | while (f >= 1.0) | |
708 | { | |
709 | for (i = 1, d = 2.0; d * d <= f; i += i) | |
710 | d *= d; | |
711 | f /= d; | |
712 | value += i; | |
713 | } | |
6d3c6adb | 714 | #endif |
340176df | 715 | #endif |
6694b327 | 716 | } |
e0cb2a68 | 717 | XSETINT (val, value); |
c26406fe | 718 | return val; |
b70021f4 MR |
719 | } |
720 | ||
fc2157cb PE |
721 | #endif /* LISP_FLOAT_TYPE */ |
722 | ||
723 | ||
acbbacbe PE |
724 | /* the rounding functions */ |
725 | ||
726 | static Lisp_Object | |
727 | rounding_driver (arg, divisor, double_round, int_round2, name) | |
fc2157cb | 728 | register Lisp_Object arg, divisor; |
acbbacbe PE |
729 | double (*double_round) (); |
730 | EMACS_INT (*int_round2) (); | |
731 | char *name; | |
b70021f4 | 732 | { |
4b6baf5f | 733 | CHECK_NUMBER_OR_FLOAT (arg, 0); |
b70021f4 | 734 | |
fc2157cb PE |
735 | if (! NILP (divisor)) |
736 | { | |
9a51b24a | 737 | EMACS_INT i1, i2; |
fc2157cb PE |
738 | |
739 | CHECK_NUMBER_OR_FLOAT (divisor, 1); | |
740 | ||
741 | #ifdef LISP_FLOAT_TYPE | |
207a45c1 | 742 | if (FLOATP (arg) || FLOATP (divisor)) |
fc2157cb PE |
743 | { |
744 | double f1, f2; | |
745 | ||
207a45c1 KH |
746 | f1 = FLOATP (arg) ? XFLOAT (arg)->data : XINT (arg); |
747 | f2 = (FLOATP (divisor) ? XFLOAT (divisor)->data : XINT (divisor)); | |
d137ae2f | 748 | if (! IEEE_FLOATING_POINT && f2 == 0) |
fc2157cb PE |
749 | Fsignal (Qarith_error, Qnil); |
750 | ||
acbbacbe PE |
751 | IN_FLOAT2 (f1 = (*double_round) (f1 / f2), name, arg, divisor); |
752 | FLOAT_TO_INT2 (f1, arg, name, arg, divisor); | |
fc2157cb PE |
753 | return arg; |
754 | } | |
755 | #endif | |
756 | ||
757 | i1 = XINT (arg); | |
758 | i2 = XINT (divisor); | |
759 | ||
760 | if (i2 == 0) | |
761 | Fsignal (Qarith_error, Qnil); | |
762 | ||
acbbacbe | 763 | XSETINT (arg, (*int_round2) (i1, i2)); |
fc2157cb PE |
764 | return arg; |
765 | } | |
766 | ||
767 | #ifdef LISP_FLOAT_TYPE | |
207a45c1 | 768 | if (FLOATP (arg)) |
81a63ccc KH |
769 | { |
770 | double d; | |
acbbacbe PE |
771 | |
772 | IN_FLOAT (d = (*double_round) (XFLOAT (arg)->data), name, arg); | |
773 | FLOAT_TO_INT (d, arg, name, arg); | |
81a63ccc | 774 | } |
fc2157cb | 775 | #endif |
b70021f4 | 776 | |
4b6baf5f | 777 | return arg; |
b70021f4 MR |
778 | } |
779 | ||
acbbacbe PE |
780 | /* With C's /, the result is implementation-defined if either operand |
781 | is negative, so take care with negative operands in the following | |
782 | integer functions. */ | |
783 | ||
784 | static EMACS_INT | |
785 | ceiling2 (i1, i2) | |
786 | EMACS_INT i1, i2; | |
787 | { | |
788 | return (i2 < 0 | |
789 | ? (i1 < 0 ? ((-1 - i1) / -i2) + 1 : - (i1 / -i2)) | |
790 | : (i1 <= 0 ? - (-i1 / i2) : ((i1 - 1) / i2) + 1)); | |
791 | } | |
792 | ||
793 | static EMACS_INT | |
794 | floor2 (i1, i2) | |
795 | EMACS_INT i1, i2; | |
796 | { | |
797 | return (i2 < 0 | |
798 | ? (i1 <= 0 ? -i1 / -i2 : -1 - ((i1 - 1) / -i2)) | |
799 | : (i1 < 0 ? -1 - ((-1 - i1) / i2) : i1 / i2)); | |
800 | } | |
801 | ||
802 | static EMACS_INT | |
803 | truncate2 (i1, i2) | |
804 | EMACS_INT i1, i2; | |
805 | { | |
806 | return (i2 < 0 | |
807 | ? (i1 < 0 ? -i1 / -i2 : - (i1 / -i2)) | |
808 | : (i1 < 0 ? - (-i1 / i2) : i1 / i2)); | |
809 | } | |
810 | ||
811 | static EMACS_INT | |
812 | round2 (i1, i2) | |
813 | EMACS_INT i1, i2; | |
814 | { | |
815 | /* The C language's division operator gives us one remainder R, but | |
816 | we want the remainder R1 on the other side of 0 if R1 is closer | |
817 | to 0 than R is; because we want to round to even, we also want R1 | |
818 | if R and R1 are the same distance from 0 and if C's quotient is | |
819 | odd. */ | |
820 | EMACS_INT q = i1 / i2; | |
821 | EMACS_INT r = i1 % i2; | |
822 | EMACS_INT abs_r = r < 0 ? -r : r; | |
823 | EMACS_INT abs_r1 = (i2 < 0 ? -i2 : i2) - abs_r; | |
824 | return q + (abs_r + (q & 1) <= abs_r1 ? 0 : (i2 ^ r) < 0 ? -1 : 1); | |
825 | } | |
826 | ||
4b5878a8 KH |
827 | #ifndef HAVE_RINT |
828 | static double | |
829 | rint (d) | |
830 | double d; | |
831 | { | |
1b65c684 | 832 | return floor (d + 0.5); |
4b5878a8 KH |
833 | } |
834 | #endif | |
835 | ||
acbbacbe PE |
836 | static double |
837 | double_identity (d) | |
838 | double d; | |
839 | { | |
840 | return d; | |
841 | } | |
842 | ||
843 | DEFUN ("ceiling", Fceiling, Sceiling, 1, 2, 0, | |
844 | "Return the smallest integer no less than ARG. (Round toward +inf.)\n\ | |
845 | With optional DIVISOR, return the smallest integer no less than ARG/DIVISOR.") | |
846 | (arg, divisor) | |
847 | Lisp_Object arg, divisor; | |
848 | { | |
849 | return rounding_driver (arg, divisor, ceil, ceiling2, "ceiling"); | |
850 | } | |
851 | ||
852 | DEFUN ("floor", Ffloor, Sfloor, 1, 2, 0, | |
853 | "Return the largest integer no greater than ARG. (Round towards -inf.)\n\ | |
854 | With optional DIVISOR, return the largest integer no greater than ARG/DIVISOR.") | |
855 | (arg, divisor) | |
856 | Lisp_Object arg, divisor; | |
857 | { | |
858 | return rounding_driver (arg, divisor, floor, floor2, "floor"); | |
859 | } | |
860 | ||
861 | DEFUN ("round", Fround, Sround, 1, 2, 0, | |
862 | "Return the nearest integer to ARG.\n\ | |
863 | With optional DIVISOR, return the nearest integer to ARG/DIVISOR.") | |
864 | (arg, divisor) | |
865 | Lisp_Object arg, divisor; | |
866 | { | |
867 | return rounding_driver (arg, divisor, rint, round2, "round"); | |
868 | } | |
869 | ||
870 | DEFUN ("truncate", Ftruncate, Struncate, 1, 2, 0, | |
871 | "Truncate a floating point number to an int.\n\ | |
872 | Rounds ARG toward zero.\n\ | |
873 | With optional DIVISOR, truncate ARG/DIVISOR.") | |
874 | (arg, divisor) | |
875 | Lisp_Object arg, divisor; | |
876 | { | |
877 | return rounding_driver (arg, divisor, double_identity, truncate2, | |
878 | "truncate"); | |
879 | } | |
880 | ||
fc2157cb PE |
881 | #ifdef LISP_FLOAT_TYPE |
882 | ||
d137ae2f PE |
883 | Lisp_Object |
884 | fmod_float (x, y) | |
885 | register Lisp_Object x, y; | |
886 | { | |
887 | double f1, f2; | |
888 | ||
889 | f1 = FLOATP (x) ? XFLOAT (x)->data : XINT (x); | |
890 | f2 = FLOATP (y) ? XFLOAT (y)->data : XINT (y); | |
891 | ||
892 | if (! IEEE_FLOATING_POINT && f2 == 0) | |
893 | Fsignal (Qarith_error, Qnil); | |
894 | ||
895 | /* If the "remainder" comes out with the wrong sign, fix it. */ | |
896 | IN_FLOAT2 ((f1 = fmod (f1, f2), | |
897 | f1 = (f2 < 0 ? f1 > 0 : f1 < 0) ? f1 + f2 : f1), | |
898 | "mod", x, y); | |
899 | return make_float (f1); | |
900 | } | |
4b6baf5f | 901 | \f |
4b6baf5f RS |
902 | /* It's not clear these are worth adding. */ |
903 | ||
904 | DEFUN ("fceiling", Ffceiling, Sfceiling, 1, 1, 0, | |
905 | "Return the smallest integer no less than ARG, as a float.\n\ | |
906 | \(Round toward +inf.\)") | |
907 | (arg) | |
908 | register Lisp_Object arg; | |
909 | { | |
910 | double d = extract_float (arg); | |
911 | IN_FLOAT (d = ceil (d), "fceiling", arg); | |
912 | return make_float (d); | |
913 | } | |
914 | ||
915 | DEFUN ("ffloor", Fffloor, Sffloor, 1, 1, 0, | |
916 | "Return the largest integer no greater than ARG, as a float.\n\ | |
917 | \(Round towards -inf.\)") | |
918 | (arg) | |
919 | register Lisp_Object arg; | |
920 | { | |
921 | double d = extract_float (arg); | |
922 | IN_FLOAT (d = floor (d), "ffloor", arg); | |
923 | return make_float (d); | |
924 | } | |
b70021f4 | 925 | |
4b6baf5f RS |
926 | DEFUN ("fround", Ffround, Sfround, 1, 1, 0, |
927 | "Return the nearest integer to ARG, as a float.") | |
928 | (arg) | |
929 | register Lisp_Object arg; | |
930 | { | |
931 | double d = extract_float (arg); | |
892ed7e0 | 932 | IN_FLOAT (d = rint (d), "fround", arg); |
4b6baf5f RS |
933 | return make_float (d); |
934 | } | |
935 | ||
936 | DEFUN ("ftruncate", Fftruncate, Sftruncate, 1, 1, 0, | |
937 | "Truncate a floating point number to an integral float value.\n\ | |
938 | Rounds the value toward zero.") | |
939 | (arg) | |
940 | register Lisp_Object arg; | |
941 | { | |
942 | double d = extract_float (arg); | |
943 | if (d >= 0.0) | |
944 | IN_FLOAT (d = floor (d), "ftruncate", arg); | |
945 | else | |
a3fc5236 | 946 | IN_FLOAT (d = ceil (d), "ftruncate", arg); |
4b6baf5f | 947 | return make_float (d); |
b70021f4 MR |
948 | } |
949 | \f | |
4b6baf5f | 950 | #ifdef FLOAT_CATCH_SIGILL |
4746118a | 951 | static SIGTYPE |
b70021f4 MR |
952 | float_error (signo) |
953 | int signo; | |
954 | { | |
955 | if (! in_float) | |
956 | fatal_error_signal (signo); | |
957 | ||
6df54671 | 958 | #ifdef BSD_SYSTEM |
b70021f4 MR |
959 | #ifdef BSD4_1 |
960 | sigrelse (SIGILL); | |
961 | #else /* not BSD4_1 */ | |
e065a56e | 962 | sigsetmask (SIGEMPTYMASK); |
b70021f4 | 963 | #endif /* not BSD4_1 */ |
265a9e55 JB |
964 | #else |
965 | /* Must reestablish handler each time it is called. */ | |
966 | signal (SIGILL, float_error); | |
6df54671 | 967 | #endif /* BSD_SYSTEM */ |
b70021f4 MR |
968 | |
969 | in_float = 0; | |
970 | ||
971 | Fsignal (Qarith_error, Fcons (float_error_arg, Qnil)); | |
972 | } | |
973 | ||
4b6baf5f RS |
974 | /* Another idea was to replace the library function `infnan' |
975 | where SIGILL is signaled. */ | |
976 | ||
977 | #endif /* FLOAT_CATCH_SIGILL */ | |
978 | ||
979 | #ifdef HAVE_MATHERR | |
980 | int | |
981 | matherr (x) | |
982 | struct exception *x; | |
983 | { | |
984 | Lisp_Object args; | |
985 | if (! in_float) | |
986 | /* Not called from emacs-lisp float routines; do the default thing. */ | |
987 | return 0; | |
988 | if (!strcmp (x->name, "pow")) | |
989 | x->name = "expt"; | |
990 | ||
991 | args | |
992 | = Fcons (build_string (x->name), | |
993 | Fcons (make_float (x->arg1), | |
994 | ((!strcmp (x->name, "log") || !strcmp (x->name, "pow")) | |
995 | ? Fcons (make_float (x->arg2), Qnil) | |
996 | : Qnil))); | |
997 | switch (x->type) | |
998 | { | |
999 | case DOMAIN: Fsignal (Qdomain_error, args); break; | |
1000 | case SING: Fsignal (Qsingularity_error, args); break; | |
1001 | case OVERFLOW: Fsignal (Qoverflow_error, args); break; | |
1002 | case UNDERFLOW: Fsignal (Qunderflow_error, args); break; | |
1003 | default: Fsignal (Qarith_error, args); break; | |
1004 | } | |
1005 | return (1); /* don't set errno or print a message */ | |
1006 | } | |
1007 | #endif /* HAVE_MATHERR */ | |
1008 | ||
b70021f4 MR |
1009 | init_floatfns () |
1010 | { | |
4b6baf5f | 1011 | #ifdef FLOAT_CATCH_SIGILL |
b70021f4 | 1012 | signal (SIGILL, float_error); |
4b6baf5f | 1013 | #endif |
b70021f4 MR |
1014 | in_float = 0; |
1015 | } | |
1016 | ||
fc2157cb PE |
1017 | #else /* not LISP_FLOAT_TYPE */ |
1018 | ||
1019 | init_floatfns () | |
1020 | {} | |
1021 | ||
1022 | #endif /* not LISP_FLOAT_TYPE */ | |
1023 | ||
b70021f4 MR |
1024 | syms_of_floatfns () |
1025 | { | |
fc2157cb | 1026 | #ifdef LISP_FLOAT_TYPE |
b70021f4 | 1027 | defsubr (&Sacos); |
b70021f4 | 1028 | defsubr (&Sasin); |
b70021f4 | 1029 | defsubr (&Satan); |
c2d4ea74 RS |
1030 | defsubr (&Scos); |
1031 | defsubr (&Ssin); | |
1032 | defsubr (&Stan); | |
1033 | #if 0 | |
1034 | defsubr (&Sacosh); | |
1035 | defsubr (&Sasinh); | |
b70021f4 | 1036 | defsubr (&Satanh); |
c2d4ea74 RS |
1037 | defsubr (&Scosh); |
1038 | defsubr (&Ssinh); | |
1039 | defsubr (&Stanh); | |
b70021f4 MR |
1040 | defsubr (&Sbessel_y0); |
1041 | defsubr (&Sbessel_y1); | |
1042 | defsubr (&Sbessel_yn); | |
1043 | defsubr (&Sbessel_j0); | |
1044 | defsubr (&Sbessel_j1); | |
1045 | defsubr (&Sbessel_jn); | |
b70021f4 MR |
1046 | defsubr (&Serf); |
1047 | defsubr (&Serfc); | |
c2d4ea74 | 1048 | defsubr (&Slog_gamma); |
4b6baf5f | 1049 | defsubr (&Scube_root); |
892ed7e0 | 1050 | #endif |
4b6baf5f RS |
1051 | defsubr (&Sfceiling); |
1052 | defsubr (&Sffloor); | |
1053 | defsubr (&Sfround); | |
1054 | defsubr (&Sftruncate); | |
b70021f4 | 1055 | defsubr (&Sexp); |
c2d4ea74 | 1056 | defsubr (&Sexpt); |
b70021f4 MR |
1057 | defsubr (&Slog); |
1058 | defsubr (&Slog10); | |
b70021f4 | 1059 | defsubr (&Ssqrt); |
b70021f4 MR |
1060 | |
1061 | defsubr (&Sabs); | |
1062 | defsubr (&Sfloat); | |
1063 | defsubr (&Slogb); | |
acbbacbe | 1064 | #endif /* LISP_FLOAT_TYPE */ |
b70021f4 | 1065 | defsubr (&Sceiling); |
acbbacbe | 1066 | defsubr (&Sfloor); |
b70021f4 MR |
1067 | defsubr (&Sround); |
1068 | defsubr (&Struncate); | |
1069 | } |