blob: 4d13e58cee9fcf80d70e83a87f9537c782c1fd60 [file] [log] [blame]
Guido van Rossum71aa32f1996-01-12 01:34:57 +00001/* Complex math module */
2
3/* much code borrowed from mathmodule.c */
4
Roger E. Masse24070ca1996-12-09 22:59:53 +00005#include "Python.h"
Mark Dickinsonf3718592009-12-21 15:27:41 +00006#include "_math.h"
Christian Heimes53876d92008-04-19 00:31:39 +00007/* we need DBL_MAX, DBL_MIN, DBL_EPSILON, DBL_MANT_DIG and FLT_RADIX from
8 float.h. We assume that FLT_RADIX is either 2 or 16. */
9#include <float.h>
Guido van Rossum71aa32f1996-01-12 01:34:57 +000010
Christian Heimes53876d92008-04-19 00:31:39 +000011#if (FLT_RADIX != 2 && FLT_RADIX != 16)
12#error "Modules/cmathmodule.c expects FLT_RADIX to be 2 or 16"
Guido van Rossum71aa32f1996-01-12 01:34:57 +000013#endif
14
Christian Heimes53876d92008-04-19 00:31:39 +000015#ifndef M_LN2
16#define M_LN2 (0.6931471805599453094) /* natural log of 2 */
17#endif
Guido van Rossum71aa32f1996-01-12 01:34:57 +000018
Christian Heimes53876d92008-04-19 00:31:39 +000019#ifndef M_LN10
20#define M_LN10 (2.302585092994045684) /* natural log of 10 */
21#endif
22
23/*
24 CM_LARGE_DOUBLE is used to avoid spurious overflow in the sqrt, log,
25 inverse trig and inverse hyperbolic trig functions. Its log is used in the
26 evaluation of exp, cos, cosh, sin, sinh, tan, and tanh to avoid unecessary
27 overflow.
28 */
29
30#define CM_LARGE_DOUBLE (DBL_MAX/4.)
31#define CM_SQRT_LARGE_DOUBLE (sqrt(CM_LARGE_DOUBLE))
32#define CM_LOG_LARGE_DOUBLE (log(CM_LARGE_DOUBLE))
33#define CM_SQRT_DBL_MIN (sqrt(DBL_MIN))
34
35/*
36 CM_SCALE_UP is an odd integer chosen such that multiplication by
37 2**CM_SCALE_UP is sufficient to turn a subnormal into a normal.
38 CM_SCALE_DOWN is (-(CM_SCALE_UP+1)/2). These scalings are used to compute
39 square roots accurately when the real and imaginary parts of the argument
40 are subnormal.
41*/
42
43#if FLT_RADIX==2
44#define CM_SCALE_UP (2*(DBL_MANT_DIG/2) + 1)
45#elif FLT_RADIX==16
46#define CM_SCALE_UP (4*DBL_MANT_DIG+1)
47#endif
48#define CM_SCALE_DOWN (-(CM_SCALE_UP+1)/2)
Guido van Rossum71aa32f1996-01-12 01:34:57 +000049
50/* forward declarations */
Christian Heimes53876d92008-04-19 00:31:39 +000051static Py_complex c_asinh(Py_complex);
52static Py_complex c_atanh(Py_complex);
53static Py_complex c_cosh(Py_complex);
54static Py_complex c_sinh(Py_complex);
Jeremy Hylton938ace62002-07-17 16:30:39 +000055static Py_complex c_sqrt(Py_complex);
Christian Heimes53876d92008-04-19 00:31:39 +000056static Py_complex c_tanh(Py_complex);
Raymond Hettingerb67ad7e2004-06-14 07:40:10 +000057static PyObject * math_error(void);
Guido van Rossum71aa32f1996-01-12 01:34:57 +000058
Christian Heimes53876d92008-04-19 00:31:39 +000059/* Code to deal with special values (infinities, NaNs, etc.). */
60
61/* special_type takes a double and returns an integer code indicating
62 the type of the double as follows:
63*/
64
65enum special_types {
66 ST_NINF, /* 0, negative infinity */
67 ST_NEG, /* 1, negative finite number (nonzero) */
68 ST_NZERO, /* 2, -0. */
69 ST_PZERO, /* 3, +0. */
70 ST_POS, /* 4, positive finite number (nonzero) */
71 ST_PINF, /* 5, positive infinity */
Mark Dickinsone5842c12009-08-04 19:25:12 +000072 ST_NAN /* 6, Not a Number */
Christian Heimes53876d92008-04-19 00:31:39 +000073};
74
75static enum special_types
76special_type(double d)
77{
78 if (Py_IS_FINITE(d)) {
79 if (d != 0) {
80 if (copysign(1., d) == 1.)
81 return ST_POS;
82 else
83 return ST_NEG;
84 }
85 else {
86 if (copysign(1., d) == 1.)
87 return ST_PZERO;
88 else
89 return ST_NZERO;
90 }
91 }
92 if (Py_IS_NAN(d))
93 return ST_NAN;
94 if (copysign(1., d) == 1.)
95 return ST_PINF;
96 else
97 return ST_NINF;
98}
99
100#define SPECIAL_VALUE(z, table) \
101 if (!Py_IS_FINITE((z).real) || !Py_IS_FINITE((z).imag)) { \
102 errno = 0; \
103 return table[special_type((z).real)] \
104 [special_type((z).imag)]; \
105 }
106
107#define P Py_MATH_PI
108#define P14 0.25*Py_MATH_PI
109#define P12 0.5*Py_MATH_PI
110#define P34 0.75*Py_MATH_PI
Christian Heimesa342c012008-04-20 21:01:16 +0000111#define INF Py_HUGE_VAL
112#define N Py_NAN
Christian Heimes53876d92008-04-19 00:31:39 +0000113#define U -9.5426319407711027e33 /* unlikely value, used as placeholder */
114
115/* First, the C functions that do the real work. Each of the c_*
116 functions computes and returns the C99 Annex G recommended result
117 and also sets errno as follows: errno = 0 if no floating-point
118 exception is associated with the result; errno = EDOM if C99 Annex
119 G recommends raising divide-by-zero or invalid for this result; and
120 errno = ERANGE where the overflow floating-point signal should be
121 raised.
122*/
123
Christian Heimesa342c012008-04-20 21:01:16 +0000124static Py_complex acos_special_values[7][7];
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000125
Tim Peters14e26402001-02-20 20:15:19 +0000126static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000127c_acos(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000128{
Christian Heimes53876d92008-04-19 00:31:39 +0000129 Py_complex s1, s2, r;
130
131 SPECIAL_VALUE(z, acos_special_values);
132
133 if (fabs(z.real) > CM_LARGE_DOUBLE || fabs(z.imag) > CM_LARGE_DOUBLE) {
134 /* avoid unnecessary overflow for large arguments */
135 r.real = atan2(fabs(z.imag), z.real);
136 /* split into cases to make sure that the branch cut has the
137 correct continuity on systems with unsigned zeros */
138 if (z.real < 0.) {
139 r.imag = -copysign(log(hypot(z.real/2., z.imag/2.)) +
140 M_LN2*2., z.imag);
141 } else {
142 r.imag = copysign(log(hypot(z.real/2., z.imag/2.)) +
143 M_LN2*2., -z.imag);
144 }
145 } else {
146 s1.real = 1.-z.real;
147 s1.imag = -z.imag;
148 s1 = c_sqrt(s1);
149 s2.real = 1.+z.real;
150 s2.imag = z.imag;
151 s2 = c_sqrt(s2);
152 r.real = 2.*atan2(s1.real, s2.real);
Mark Dickinsonf3718592009-12-21 15:27:41 +0000153 r.imag = m_asinh(s2.real*s1.imag - s2.imag*s1.real);
Christian Heimes53876d92008-04-19 00:31:39 +0000154 }
155 errno = 0;
156 return r;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000157}
158
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000159PyDoc_STRVAR(c_acos_doc,
Tim Peters14e26402001-02-20 20:15:19 +0000160"acos(x)\n"
161"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000162"Return the arc cosine of x.");
Guido van Rossumc6e22901998-12-04 19:26:43 +0000163
164
Christian Heimesa342c012008-04-20 21:01:16 +0000165static Py_complex acosh_special_values[7][7];
Christian Heimes53876d92008-04-19 00:31:39 +0000166
Tim Peters14e26402001-02-20 20:15:19 +0000167static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000168c_acosh(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000169{
Christian Heimes53876d92008-04-19 00:31:39 +0000170 Py_complex s1, s2, r;
171
172 SPECIAL_VALUE(z, acosh_special_values);
173
174 if (fabs(z.real) > CM_LARGE_DOUBLE || fabs(z.imag) > CM_LARGE_DOUBLE) {
175 /* avoid unnecessary overflow for large arguments */
176 r.real = log(hypot(z.real/2., z.imag/2.)) + M_LN2*2.;
177 r.imag = atan2(z.imag, z.real);
178 } else {
179 s1.real = z.real - 1.;
180 s1.imag = z.imag;
181 s1 = c_sqrt(s1);
182 s2.real = z.real + 1.;
183 s2.imag = z.imag;
184 s2 = c_sqrt(s2);
Mark Dickinsonf3718592009-12-21 15:27:41 +0000185 r.real = m_asinh(s1.real*s2.real + s1.imag*s2.imag);
Christian Heimes53876d92008-04-19 00:31:39 +0000186 r.imag = 2.*atan2(s1.imag, s2.real);
187 }
188 errno = 0;
189 return r;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000190}
191
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000192PyDoc_STRVAR(c_acosh_doc,
Tim Peters14e26402001-02-20 20:15:19 +0000193"acosh(x)\n"
194"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000195"Return the hyperbolic arccosine of x.");
Guido van Rossumc6e22901998-12-04 19:26:43 +0000196
197
Tim Peters14e26402001-02-20 20:15:19 +0000198static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000199c_asin(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000200{
Christian Heimes53876d92008-04-19 00:31:39 +0000201 /* asin(z) = -i asinh(iz) */
202 Py_complex s, r;
203 s.real = -z.imag;
204 s.imag = z.real;
205 s = c_asinh(s);
206 r.real = s.imag;
207 r.imag = -s.real;
208 return r;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000209}
210
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000211PyDoc_STRVAR(c_asin_doc,
Tim Peters14e26402001-02-20 20:15:19 +0000212"asin(x)\n"
213"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000214"Return the arc sine of x.");
Guido van Rossumc6e22901998-12-04 19:26:43 +0000215
216
Christian Heimesa342c012008-04-20 21:01:16 +0000217static Py_complex asinh_special_values[7][7];
Christian Heimes53876d92008-04-19 00:31:39 +0000218
Tim Peters14e26402001-02-20 20:15:19 +0000219static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000220c_asinh(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000221{
Christian Heimes53876d92008-04-19 00:31:39 +0000222 Py_complex s1, s2, r;
223
224 SPECIAL_VALUE(z, asinh_special_values);
225
226 if (fabs(z.real) > CM_LARGE_DOUBLE || fabs(z.imag) > CM_LARGE_DOUBLE) {
227 if (z.imag >= 0.) {
228 r.real = copysign(log(hypot(z.real/2., z.imag/2.)) +
229 M_LN2*2., z.real);
230 } else {
231 r.real = -copysign(log(hypot(z.real/2., z.imag/2.)) +
232 M_LN2*2., -z.real);
233 }
234 r.imag = atan2(z.imag, fabs(z.real));
235 } else {
236 s1.real = 1.+z.imag;
237 s1.imag = -z.real;
238 s1 = c_sqrt(s1);
239 s2.real = 1.-z.imag;
240 s2.imag = z.real;
241 s2 = c_sqrt(s2);
Mark Dickinsonf3718592009-12-21 15:27:41 +0000242 r.real = m_asinh(s1.real*s2.imag-s2.real*s1.imag);
Christian Heimes53876d92008-04-19 00:31:39 +0000243 r.imag = atan2(z.imag, s1.real*s2.real-s1.imag*s2.imag);
244 }
245 errno = 0;
246 return r;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000247}
248
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000249PyDoc_STRVAR(c_asinh_doc,
Tim Peters14e26402001-02-20 20:15:19 +0000250"asinh(x)\n"
251"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000252"Return the hyperbolic arc sine of x.");
Guido van Rossumc6e22901998-12-04 19:26:43 +0000253
254
Tim Peters14e26402001-02-20 20:15:19 +0000255static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000256c_atan(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000257{
Christian Heimes53876d92008-04-19 00:31:39 +0000258 /* atan(z) = -i atanh(iz) */
259 Py_complex s, r;
260 s.real = -z.imag;
261 s.imag = z.real;
262 s = c_atanh(s);
263 r.real = s.imag;
264 r.imag = -s.real;
265 return r;
266}
267
Christian Heimese57950f2008-04-21 13:08:03 +0000268/* Windows screws up atan2 for inf and nan, and alpha Tru64 5.1 doesn't follow
269 C99 for atan2(0., 0.). */
Christian Heimes53876d92008-04-19 00:31:39 +0000270static double
271c_atan2(Py_complex z)
272{
273 if (Py_IS_NAN(z.real) || Py_IS_NAN(z.imag))
274 return Py_NAN;
275 if (Py_IS_INFINITY(z.imag)) {
276 if (Py_IS_INFINITY(z.real)) {
277 if (copysign(1., z.real) == 1.)
278 /* atan2(+-inf, +inf) == +-pi/4 */
279 return copysign(0.25*Py_MATH_PI, z.imag);
280 else
281 /* atan2(+-inf, -inf) == +-pi*3/4 */
282 return copysign(0.75*Py_MATH_PI, z.imag);
283 }
284 /* atan2(+-inf, x) == +-pi/2 for finite x */
285 return copysign(0.5*Py_MATH_PI, z.imag);
286 }
Christian Heimese57950f2008-04-21 13:08:03 +0000287 if (Py_IS_INFINITY(z.real) || z.imag == 0.) {
288 if (copysign(1., z.real) == 1.)
289 /* atan2(+-y, +inf) = atan2(+-0, +x) = +-0. */
290 return copysign(0., z.imag);
291 else
292 /* atan2(+-y, -inf) = atan2(+-0., -x) = +-pi. */
293 return copysign(Py_MATH_PI, z.imag);
294 }
Christian Heimes53876d92008-04-19 00:31:39 +0000295 return atan2(z.imag, z.real);
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000296}
297
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000298PyDoc_STRVAR(c_atan_doc,
Tim Peters14e26402001-02-20 20:15:19 +0000299"atan(x)\n"
300"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000301"Return the arc tangent of x.");
Guido van Rossumc6e22901998-12-04 19:26:43 +0000302
303
Christian Heimesa342c012008-04-20 21:01:16 +0000304static Py_complex atanh_special_values[7][7];
Christian Heimes53876d92008-04-19 00:31:39 +0000305
Tim Peters14e26402001-02-20 20:15:19 +0000306static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000307c_atanh(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000308{
Christian Heimes53876d92008-04-19 00:31:39 +0000309 Py_complex r;
310 double ay, h;
311
312 SPECIAL_VALUE(z, atanh_special_values);
313
314 /* Reduce to case where z.real >= 0., using atanh(z) = -atanh(-z). */
315 if (z.real < 0.) {
316 return c_neg(c_atanh(c_neg(z)));
317 }
318
319 ay = fabs(z.imag);
320 if (z.real > CM_SQRT_LARGE_DOUBLE || ay > CM_SQRT_LARGE_DOUBLE) {
321 /*
322 if abs(z) is large then we use the approximation
323 atanh(z) ~ 1/z +/- i*pi/2 (+/- depending on the sign
324 of z.imag)
325 */
326 h = hypot(z.real/2., z.imag/2.); /* safe from overflow */
327 r.real = z.real/4./h/h;
328 /* the two negations in the next line cancel each other out
329 except when working with unsigned zeros: they're there to
330 ensure that the branch cut has the correct continuity on
331 systems that don't support signed zeros */
332 r.imag = -copysign(Py_MATH_PI/2., -z.imag);
333 errno = 0;
334 } else if (z.real == 1. && ay < CM_SQRT_DBL_MIN) {
335 /* C99 standard says: atanh(1+/-0.) should be inf +/- 0i */
336 if (ay == 0.) {
337 r.real = INF;
338 r.imag = z.imag;
339 errno = EDOM;
340 } else {
341 r.real = -log(sqrt(ay)/sqrt(hypot(ay, 2.)));
342 r.imag = copysign(atan2(2., -ay)/2, z.imag);
343 errno = 0;
344 }
345 } else {
Mark Dickinsonf3718592009-12-21 15:27:41 +0000346 r.real = m_log1p(4.*z.real/((1-z.real)*(1-z.real) + ay*ay))/4.;
Christian Heimes53876d92008-04-19 00:31:39 +0000347 r.imag = -atan2(-2.*z.imag, (1-z.real)*(1+z.real) - ay*ay)/2.;
348 errno = 0;
349 }
350 return r;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000351}
352
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000353PyDoc_STRVAR(c_atanh_doc,
Tim Peters14e26402001-02-20 20:15:19 +0000354"atanh(x)\n"
355"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000356"Return the hyperbolic arc tangent of x.");
Guido van Rossumc6e22901998-12-04 19:26:43 +0000357
358
Tim Peters14e26402001-02-20 20:15:19 +0000359static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000360c_cos(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000361{
Christian Heimes53876d92008-04-19 00:31:39 +0000362 /* cos(z) = cosh(iz) */
Guido van Rossum9e720e31996-07-21 02:31:35 +0000363 Py_complex r;
Christian Heimes53876d92008-04-19 00:31:39 +0000364 r.real = -z.imag;
365 r.imag = z.real;
366 r = c_cosh(r);
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000367 return r;
368}
369
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000370PyDoc_STRVAR(c_cos_doc,
Tim Peters14e26402001-02-20 20:15:19 +0000371"cos(x)\n"
Mark Dickinson1bd2e292009-02-28 15:53:24 +0000372"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000373"Return the cosine of x.");
Guido van Rossumc6e22901998-12-04 19:26:43 +0000374
375
Christian Heimes53876d92008-04-19 00:31:39 +0000376/* cosh(infinity + i*y) needs to be dealt with specially */
Christian Heimesa342c012008-04-20 21:01:16 +0000377static Py_complex cosh_special_values[7][7];
Christian Heimes53876d92008-04-19 00:31:39 +0000378
Tim Peters14e26402001-02-20 20:15:19 +0000379static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000380c_cosh(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000381{
Guido van Rossum9e720e31996-07-21 02:31:35 +0000382 Py_complex r;
Christian Heimes53876d92008-04-19 00:31:39 +0000383 double x_minus_one;
384
385 /* special treatment for cosh(+/-inf + iy) if y is not a NaN */
386 if (!Py_IS_FINITE(z.real) || !Py_IS_FINITE(z.imag)) {
387 if (Py_IS_INFINITY(z.real) && Py_IS_FINITE(z.imag) &&
388 (z.imag != 0.)) {
389 if (z.real > 0) {
390 r.real = copysign(INF, cos(z.imag));
391 r.imag = copysign(INF, sin(z.imag));
392 }
393 else {
394 r.real = copysign(INF, cos(z.imag));
395 r.imag = -copysign(INF, sin(z.imag));
396 }
397 }
398 else {
399 r = cosh_special_values[special_type(z.real)]
400 [special_type(z.imag)];
401 }
402 /* need to set errno = EDOM if y is +/- infinity and x is not
403 a NaN */
404 if (Py_IS_INFINITY(z.imag) && !Py_IS_NAN(z.real))
405 errno = EDOM;
406 else
407 errno = 0;
408 return r;
409 }
410
411 if (fabs(z.real) > CM_LOG_LARGE_DOUBLE) {
412 /* deal correctly with cases where cosh(z.real) overflows but
413 cosh(z) does not. */
414 x_minus_one = z.real - copysign(1., z.real);
415 r.real = cos(z.imag) * cosh(x_minus_one) * Py_MATH_E;
416 r.imag = sin(z.imag) * sinh(x_minus_one) * Py_MATH_E;
417 } else {
418 r.real = cos(z.imag) * cosh(z.real);
419 r.imag = sin(z.imag) * sinh(z.real);
420 }
421 /* detect overflow, and set errno accordingly */
422 if (Py_IS_INFINITY(r.real) || Py_IS_INFINITY(r.imag))
423 errno = ERANGE;
424 else
425 errno = 0;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000426 return r;
427}
428
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000429PyDoc_STRVAR(c_cosh_doc,
Tim Peters14e26402001-02-20 20:15:19 +0000430"cosh(x)\n"
Mark Dickinson1bd2e292009-02-28 15:53:24 +0000431"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000432"Return the hyperbolic cosine of x.");
Guido van Rossumc6e22901998-12-04 19:26:43 +0000433
434
Christian Heimes53876d92008-04-19 00:31:39 +0000435/* exp(infinity + i*y) and exp(-infinity + i*y) need special treatment for
436 finite y */
Christian Heimesa342c012008-04-20 21:01:16 +0000437static Py_complex exp_special_values[7][7];
Christian Heimes53876d92008-04-19 00:31:39 +0000438
Tim Peters14e26402001-02-20 20:15:19 +0000439static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000440c_exp(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000441{
Guido van Rossum9e720e31996-07-21 02:31:35 +0000442 Py_complex r;
Christian Heimes53876d92008-04-19 00:31:39 +0000443 double l;
444
445 if (!Py_IS_FINITE(z.real) || !Py_IS_FINITE(z.imag)) {
446 if (Py_IS_INFINITY(z.real) && Py_IS_FINITE(z.imag)
447 && (z.imag != 0.)) {
448 if (z.real > 0) {
449 r.real = copysign(INF, cos(z.imag));
450 r.imag = copysign(INF, sin(z.imag));
451 }
452 else {
453 r.real = copysign(0., cos(z.imag));
454 r.imag = copysign(0., sin(z.imag));
455 }
456 }
457 else {
458 r = exp_special_values[special_type(z.real)]
459 [special_type(z.imag)];
460 }
461 /* need to set errno = EDOM if y is +/- infinity and x is not
462 a NaN and not -infinity */
463 if (Py_IS_INFINITY(z.imag) &&
464 (Py_IS_FINITE(z.real) ||
465 (Py_IS_INFINITY(z.real) && z.real > 0)))
466 errno = EDOM;
467 else
468 errno = 0;
469 return r;
470 }
471
472 if (z.real > CM_LOG_LARGE_DOUBLE) {
473 l = exp(z.real-1.);
474 r.real = l*cos(z.imag)*Py_MATH_E;
475 r.imag = l*sin(z.imag)*Py_MATH_E;
476 } else {
477 l = exp(z.real);
478 r.real = l*cos(z.imag);
479 r.imag = l*sin(z.imag);
480 }
481 /* detect overflow, and set errno accordingly */
482 if (Py_IS_INFINITY(r.real) || Py_IS_INFINITY(r.imag))
483 errno = ERANGE;
484 else
485 errno = 0;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000486 return r;
487}
488
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000489PyDoc_STRVAR(c_exp_doc,
Tim Peters14e26402001-02-20 20:15:19 +0000490"exp(x)\n"
491"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000492"Return the exponential value e**x.");
Guido van Rossumc6e22901998-12-04 19:26:43 +0000493
494
Christian Heimesa342c012008-04-20 21:01:16 +0000495static Py_complex log_special_values[7][7];
Christian Heimes53876d92008-04-19 00:31:39 +0000496
Tim Peters14e26402001-02-20 20:15:19 +0000497static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000498c_log(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000499{
Christian Heimes53876d92008-04-19 00:31:39 +0000500 /*
501 The usual formula for the real part is log(hypot(z.real, z.imag)).
502 There are four situations where this formula is potentially
503 problematic:
504
505 (1) the absolute value of z is subnormal. Then hypot is subnormal,
506 so has fewer than the usual number of bits of accuracy, hence may
507 have large relative error. This then gives a large absolute error
508 in the log. This can be solved by rescaling z by a suitable power
509 of 2.
510
511 (2) the absolute value of z is greater than DBL_MAX (e.g. when both
512 z.real and z.imag are within a factor of 1/sqrt(2) of DBL_MAX)
513 Again, rescaling solves this.
514
515 (3) the absolute value of z is close to 1. In this case it's
516 difficult to achieve good accuracy, at least in part because a
517 change of 1ulp in the real or imaginary part of z can result in a
518 change of billions of ulps in the correctly rounded answer.
519
520 (4) z = 0. The simplest thing to do here is to call the
521 floating-point log with an argument of 0, and let its behaviour
522 (returning -infinity, signaling a floating-point exception, setting
523 errno, or whatever) determine that of c_log. So the usual formula
524 is fine here.
525
526 */
527
Guido van Rossum9e720e31996-07-21 02:31:35 +0000528 Py_complex r;
Christian Heimes53876d92008-04-19 00:31:39 +0000529 double ax, ay, am, an, h;
530
531 SPECIAL_VALUE(z, log_special_values);
532
533 ax = fabs(z.real);
534 ay = fabs(z.imag);
535
536 if (ax > CM_LARGE_DOUBLE || ay > CM_LARGE_DOUBLE) {
537 r.real = log(hypot(ax/2., ay/2.)) + M_LN2;
538 } else if (ax < DBL_MIN && ay < DBL_MIN) {
539 if (ax > 0. || ay > 0.) {
540 /* catch cases where hypot(ax, ay) is subnormal */
541 r.real = log(hypot(ldexp(ax, DBL_MANT_DIG),
542 ldexp(ay, DBL_MANT_DIG))) - DBL_MANT_DIG*M_LN2;
543 }
544 else {
545 /* log(+/-0. +/- 0i) */
546 r.real = -INF;
547 r.imag = atan2(z.imag, z.real);
548 errno = EDOM;
549 return r;
550 }
551 } else {
552 h = hypot(ax, ay);
553 if (0.71 <= h && h <= 1.73) {
554 am = ax > ay ? ax : ay; /* max(ax, ay) */
555 an = ax > ay ? ay : ax; /* min(ax, ay) */
Mark Dickinsonf3718592009-12-21 15:27:41 +0000556 r.real = m_log1p((am-1)*(am+1)+an*an)/2.;
Christian Heimes53876d92008-04-19 00:31:39 +0000557 } else {
558 r.real = log(h);
559 }
560 }
561 r.imag = atan2(z.imag, z.real);
562 errno = 0;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000563 return r;
564}
565
Guido van Rossumc6e22901998-12-04 19:26:43 +0000566
Tim Peters14e26402001-02-20 20:15:19 +0000567static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000568c_log10(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000569{
Guido van Rossum9e720e31996-07-21 02:31:35 +0000570 Py_complex r;
Christian Heimes53876d92008-04-19 00:31:39 +0000571 int errno_save;
572
573 r = c_log(z);
574 errno_save = errno; /* just in case the divisions affect errno */
575 r.real = r.real / M_LN10;
576 r.imag = r.imag / M_LN10;
577 errno = errno_save;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000578 return r;
579}
580
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000581PyDoc_STRVAR(c_log10_doc,
Tim Peters14e26402001-02-20 20:15:19 +0000582"log10(x)\n"
583"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000584"Return the base-10 logarithm of x.");
Guido van Rossumc6e22901998-12-04 19:26:43 +0000585
586
Tim Peters14e26402001-02-20 20:15:19 +0000587static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000588c_sin(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000589{
Christian Heimes53876d92008-04-19 00:31:39 +0000590 /* sin(z) = -i sin(iz) */
591 Py_complex s, r;
592 s.real = -z.imag;
593 s.imag = z.real;
594 s = c_sinh(s);
595 r.real = s.imag;
596 r.imag = -s.real;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000597 return r;
598}
599
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000600PyDoc_STRVAR(c_sin_doc,
Tim Peters14e26402001-02-20 20:15:19 +0000601"sin(x)\n"
602"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000603"Return the sine of x.");
Guido van Rossumc6e22901998-12-04 19:26:43 +0000604
605
Christian Heimes53876d92008-04-19 00:31:39 +0000606/* sinh(infinity + i*y) needs to be dealt with specially */
Christian Heimesa342c012008-04-20 21:01:16 +0000607static Py_complex sinh_special_values[7][7];
Christian Heimes53876d92008-04-19 00:31:39 +0000608
Tim Peters14e26402001-02-20 20:15:19 +0000609static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000610c_sinh(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000611{
Guido van Rossum9e720e31996-07-21 02:31:35 +0000612 Py_complex r;
Christian Heimes53876d92008-04-19 00:31:39 +0000613 double x_minus_one;
614
615 /* special treatment for sinh(+/-inf + iy) if y is finite and
616 nonzero */
617 if (!Py_IS_FINITE(z.real) || !Py_IS_FINITE(z.imag)) {
618 if (Py_IS_INFINITY(z.real) && Py_IS_FINITE(z.imag)
619 && (z.imag != 0.)) {
620 if (z.real > 0) {
621 r.real = copysign(INF, cos(z.imag));
622 r.imag = copysign(INF, sin(z.imag));
623 }
624 else {
625 r.real = -copysign(INF, cos(z.imag));
626 r.imag = copysign(INF, sin(z.imag));
627 }
628 }
629 else {
630 r = sinh_special_values[special_type(z.real)]
631 [special_type(z.imag)];
632 }
633 /* need to set errno = EDOM if y is +/- infinity and x is not
634 a NaN */
635 if (Py_IS_INFINITY(z.imag) && !Py_IS_NAN(z.real))
636 errno = EDOM;
637 else
638 errno = 0;
639 return r;
640 }
641
642 if (fabs(z.real) > CM_LOG_LARGE_DOUBLE) {
643 x_minus_one = z.real - copysign(1., z.real);
644 r.real = cos(z.imag) * sinh(x_minus_one) * Py_MATH_E;
645 r.imag = sin(z.imag) * cosh(x_minus_one) * Py_MATH_E;
646 } else {
647 r.real = cos(z.imag) * sinh(z.real);
648 r.imag = sin(z.imag) * cosh(z.real);
649 }
650 /* detect overflow, and set errno accordingly */
651 if (Py_IS_INFINITY(r.real) || Py_IS_INFINITY(r.imag))
652 errno = ERANGE;
653 else
654 errno = 0;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000655 return r;
656}
657
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000658PyDoc_STRVAR(c_sinh_doc,
Tim Peters14e26402001-02-20 20:15:19 +0000659"sinh(x)\n"
660"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000661"Return the hyperbolic sine of x.");
Guido van Rossumc6e22901998-12-04 19:26:43 +0000662
663
Christian Heimesa342c012008-04-20 21:01:16 +0000664static Py_complex sqrt_special_values[7][7];
Christian Heimes53876d92008-04-19 00:31:39 +0000665
Tim Peters14e26402001-02-20 20:15:19 +0000666static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000667c_sqrt(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000668{
Christian Heimes53876d92008-04-19 00:31:39 +0000669 /*
670 Method: use symmetries to reduce to the case when x = z.real and y
671 = z.imag are nonnegative. Then the real part of the result is
672 given by
673
674 s = sqrt((x + hypot(x, y))/2)
675
676 and the imaginary part is
677
678 d = (y/2)/s
679
680 If either x or y is very large then there's a risk of overflow in
681 computation of the expression x + hypot(x, y). We can avoid this
682 by rewriting the formula for s as:
683
684 s = 2*sqrt(x/8 + hypot(x/8, y/8))
685
686 This costs us two extra multiplications/divisions, but avoids the
687 overhead of checking for x and y large.
688
689 If both x and y are subnormal then hypot(x, y) may also be
690 subnormal, so will lack full precision. We solve this by rescaling
691 x and y by a sufficiently large power of 2 to ensure that x and y
692 are normal.
693 */
694
695
Guido van Rossum9e720e31996-07-21 02:31:35 +0000696 Py_complex r;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000697 double s,d;
Christian Heimes53876d92008-04-19 00:31:39 +0000698 double ax, ay;
699
700 SPECIAL_VALUE(z, sqrt_special_values);
701
702 if (z.real == 0. && z.imag == 0.) {
703 r.real = 0.;
704 r.imag = z.imag;
705 return r;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000706 }
Christian Heimes53876d92008-04-19 00:31:39 +0000707
708 ax = fabs(z.real);
709 ay = fabs(z.imag);
710
711 if (ax < DBL_MIN && ay < DBL_MIN && (ax > 0. || ay > 0.)) {
712 /* here we catch cases where hypot(ax, ay) is subnormal */
713 ax = ldexp(ax, CM_SCALE_UP);
714 s = ldexp(sqrt(ax + hypot(ax, ldexp(ay, CM_SCALE_UP))),
715 CM_SCALE_DOWN);
716 } else {
717 ax /= 8.;
718 s = 2.*sqrt(ax + hypot(ax, ay/8.));
719 }
720 d = ay/(2.*s);
721
722 if (z.real >= 0.) {
723 r.real = s;
724 r.imag = copysign(d, z.imag);
725 } else {
726 r.real = d;
727 r.imag = copysign(s, z.imag);
728 }
729 errno = 0;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000730 return r;
731}
732
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000733PyDoc_STRVAR(c_sqrt_doc,
Tim Peters14e26402001-02-20 20:15:19 +0000734"sqrt(x)\n"
735"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000736"Return the square root of x.");
Guido van Rossumc6e22901998-12-04 19:26:43 +0000737
738
Tim Peters14e26402001-02-20 20:15:19 +0000739static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000740c_tan(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000741{
Christian Heimes53876d92008-04-19 00:31:39 +0000742 /* tan(z) = -i tanh(iz) */
743 Py_complex s, r;
744 s.real = -z.imag;
745 s.imag = z.real;
746 s = c_tanh(s);
747 r.real = s.imag;
748 r.imag = -s.real;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000749 return r;
750}
751
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000752PyDoc_STRVAR(c_tan_doc,
Tim Peters14e26402001-02-20 20:15:19 +0000753"tan(x)\n"
754"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000755"Return the tangent of x.");
Guido van Rossumc6e22901998-12-04 19:26:43 +0000756
757
Christian Heimes53876d92008-04-19 00:31:39 +0000758/* tanh(infinity + i*y) needs to be dealt with specially */
Christian Heimesa342c012008-04-20 21:01:16 +0000759static Py_complex tanh_special_values[7][7];
Christian Heimes53876d92008-04-19 00:31:39 +0000760
Tim Peters14e26402001-02-20 20:15:19 +0000761static Py_complex
Christian Heimes53876d92008-04-19 00:31:39 +0000762c_tanh(Py_complex z)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000763{
Christian Heimes53876d92008-04-19 00:31:39 +0000764 /* Formula:
765
766 tanh(x+iy) = (tanh(x)(1+tan(y)^2) + i tan(y)(1-tanh(x))^2) /
767 (1+tan(y)^2 tanh(x)^2)
768
769 To avoid excessive roundoff error, 1-tanh(x)^2 is better computed
770 as 1/cosh(x)^2. When abs(x) is large, we approximate 1-tanh(x)^2
771 by 4 exp(-2*x) instead, to avoid possible overflow in the
772 computation of cosh(x).
773
774 */
775
Guido van Rossum9e720e31996-07-21 02:31:35 +0000776 Py_complex r;
Christian Heimes53876d92008-04-19 00:31:39 +0000777 double tx, ty, cx, txty, denom;
778
779 /* special treatment for tanh(+/-inf + iy) if y is finite and
780 nonzero */
781 if (!Py_IS_FINITE(z.real) || !Py_IS_FINITE(z.imag)) {
782 if (Py_IS_INFINITY(z.real) && Py_IS_FINITE(z.imag)
783 && (z.imag != 0.)) {
784 if (z.real > 0) {
785 r.real = 1.0;
786 r.imag = copysign(0.,
787 2.*sin(z.imag)*cos(z.imag));
788 }
789 else {
790 r.real = -1.0;
791 r.imag = copysign(0.,
792 2.*sin(z.imag)*cos(z.imag));
793 }
794 }
795 else {
796 r = tanh_special_values[special_type(z.real)]
797 [special_type(z.imag)];
798 }
799 /* need to set errno = EDOM if z.imag is +/-infinity and
800 z.real is finite */
801 if (Py_IS_INFINITY(z.imag) && Py_IS_FINITE(z.real))
802 errno = EDOM;
803 else
804 errno = 0;
805 return r;
806 }
807
808 /* danger of overflow in 2.*z.imag !*/
809 if (fabs(z.real) > CM_LOG_LARGE_DOUBLE) {
810 r.real = copysign(1., z.real);
811 r.imag = 4.*sin(z.imag)*cos(z.imag)*exp(-2.*fabs(z.real));
812 } else {
813 tx = tanh(z.real);
814 ty = tan(z.imag);
815 cx = 1./cosh(z.real);
816 txty = tx*ty;
817 denom = 1. + txty*txty;
818 r.real = tx*(1.+ty*ty)/denom;
819 r.imag = ((ty/denom)*cx)*cx;
820 }
821 errno = 0;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000822 return r;
823}
824
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000825PyDoc_STRVAR(c_tanh_doc,
Tim Peters14e26402001-02-20 20:15:19 +0000826"tanh(x)\n"
827"\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +0000828"Return the hyperbolic tangent of x.");
Guido van Rossumc6e22901998-12-04 19:26:43 +0000829
Christian Heimes53876d92008-04-19 00:31:39 +0000830
Raymond Hettingerb67ad7e2004-06-14 07:40:10 +0000831static PyObject *
832cmath_log(PyObject *self, PyObject *args)
833{
834 Py_complex x;
835 Py_complex y;
836
837 if (!PyArg_ParseTuple(args, "D|D", &x, &y))
838 return NULL;
839
840 errno = 0;
841 PyFPE_START_PROTECT("complex function", return 0)
842 x = c_log(x);
Georg Brandl86b2fb92008-07-16 03:43:04 +0000843 if (PyTuple_GET_SIZE(args) == 2) {
844 y = c_log(y);
845 x = c_quot(x, y);
846 }
Raymond Hettingerb67ad7e2004-06-14 07:40:10 +0000847 PyFPE_END_PROTECT(x)
848 if (errno != 0)
849 return math_error();
Raymond Hettingerb67ad7e2004-06-14 07:40:10 +0000850 return PyComplex_FromCComplex(x);
851}
852
853PyDoc_STRVAR(cmath_log_doc,
854"log(x[, base]) -> the logarithm of x to the given base.\n\
855If the base not specified, returns the natural logarithm (base e) of x.");
856
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000857
858/* And now the glue to make them available from Python: */
859
Roger E. Masse24070ca1996-12-09 22:59:53 +0000860static PyObject *
Thomas Woutersf3f33dc2000-07-21 06:00:07 +0000861math_error(void)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000862{
863 if (errno == EDOM)
Roger E. Masse24070ca1996-12-09 22:59:53 +0000864 PyErr_SetString(PyExc_ValueError, "math domain error");
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000865 else if (errno == ERANGE)
Roger E. Masse24070ca1996-12-09 22:59:53 +0000866 PyErr_SetString(PyExc_OverflowError, "math range error");
867 else /* Unexpected math error */
Tim Peters14e26402001-02-20 20:15:19 +0000868 PyErr_SetFromErrno(PyExc_ValueError);
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000869 return NULL;
870}
871
Roger E. Masse24070ca1996-12-09 22:59:53 +0000872static PyObject *
Peter Schneider-Kampf1ca8982000-07-10 09:31:34 +0000873math_1(PyObject *args, Py_complex (*func)(Py_complex))
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000874{
Christian Heimes53876d92008-04-19 00:31:39 +0000875 Py_complex x,r ;
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000876 if (!PyArg_ParseTuple(args, "D", &x))
877 return NULL;
878 errno = 0;
Christian Heimes53876d92008-04-19 00:31:39 +0000879 PyFPE_START_PROTECT("complex function", return 0);
880 r = (*func)(x);
881 PyFPE_END_PROTECT(r);
882 if (errno == EDOM) {
883 PyErr_SetString(PyExc_ValueError, "math domain error");
884 return NULL;
885 }
886 else if (errno == ERANGE) {
887 PyErr_SetString(PyExc_OverflowError, "math range error");
888 return NULL;
889 }
890 else {
891 return PyComplex_FromCComplex(r);
892 }
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000893}
894
895#define FUNC1(stubname, func) \
Peter Schneider-Kampf1ca8982000-07-10 09:31:34 +0000896 static PyObject * stubname(PyObject *self, PyObject *args) { \
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000897 return math_1(args, func); \
898 }
899
900FUNC1(cmath_acos, c_acos)
901FUNC1(cmath_acosh, c_acosh)
902FUNC1(cmath_asin, c_asin)
903FUNC1(cmath_asinh, c_asinh)
904FUNC1(cmath_atan, c_atan)
905FUNC1(cmath_atanh, c_atanh)
906FUNC1(cmath_cos, c_cos)
907FUNC1(cmath_cosh, c_cosh)
908FUNC1(cmath_exp, c_exp)
Guido van Rossum71aa32f1996-01-12 01:34:57 +0000909FUNC1(cmath_log10, c_log10)
910FUNC1(cmath_sin, c_sin)
911FUNC1(cmath_sinh, c_sinh)
912FUNC1(cmath_sqrt, c_sqrt)
913FUNC1(cmath_tan, c_tan)
914FUNC1(cmath_tanh, c_tanh)
915
Christian Heimes53876d92008-04-19 00:31:39 +0000916static PyObject *
917cmath_phase(PyObject *self, PyObject *args)
918{
919 Py_complex z;
920 double phi;
921 if (!PyArg_ParseTuple(args, "D:phase", &z))
922 return NULL;
923 errno = 0;
924 PyFPE_START_PROTECT("arg function", return 0)
925 phi = c_atan2(z);
Alexandre Vassalottibee32532008-05-16 18:15:12 +0000926 PyFPE_END_PROTECT(phi)
Christian Heimes53876d92008-04-19 00:31:39 +0000927 if (errno != 0)
928 return math_error();
929 else
930 return PyFloat_FromDouble(phi);
931}
932
933PyDoc_STRVAR(cmath_phase_doc,
934"phase(z) -> float\n\n\
935Return argument, also known as the phase angle, of a complex.");
936
937static PyObject *
938cmath_polar(PyObject *self, PyObject *args)
939{
940 Py_complex z;
941 double r, phi;
942 if (!PyArg_ParseTuple(args, "D:polar", &z))
943 return NULL;
944 PyFPE_START_PROTECT("polar function", return 0)
945 phi = c_atan2(z); /* should not cause any exception */
946 r = c_abs(z); /* sets errno to ERANGE on overflow; otherwise 0 */
947 PyFPE_END_PROTECT(r)
948 if (errno != 0)
949 return math_error();
950 else
951 return Py_BuildValue("dd", r, phi);
952}
953
954PyDoc_STRVAR(cmath_polar_doc,
955"polar(z) -> r: float, phi: float\n\n\
956Convert a complex from rectangular coordinates to polar coordinates. r is\n\
957the distance from 0 and phi the phase angle.");
958
959/*
960 rect() isn't covered by the C99 standard, but it's not too hard to
961 figure out 'spirit of C99' rules for special value handing:
962
963 rect(x, t) should behave like exp(log(x) + it) for positive-signed x
964 rect(x, t) should behave like -exp(log(-x) + it) for negative-signed x
965 rect(nan, t) should behave like exp(nan + it), except that rect(nan, 0)
966 gives nan +- i0 with the sign of the imaginary part unspecified.
967
968*/
969
Christian Heimesa342c012008-04-20 21:01:16 +0000970static Py_complex rect_special_values[7][7];
Christian Heimes53876d92008-04-19 00:31:39 +0000971
972static PyObject *
973cmath_rect(PyObject *self, PyObject *args)
974{
975 Py_complex z;
976 double r, phi;
977 if (!PyArg_ParseTuple(args, "dd:rect", &r, &phi))
978 return NULL;
979 errno = 0;
980 PyFPE_START_PROTECT("rect function", return 0)
981
982 /* deal with special values */
983 if (!Py_IS_FINITE(r) || !Py_IS_FINITE(phi)) {
984 /* if r is +/-infinity and phi is finite but nonzero then
985 result is (+-INF +-INF i), but we need to compute cos(phi)
986 and sin(phi) to figure out the signs. */
987 if (Py_IS_INFINITY(r) && (Py_IS_FINITE(phi)
988 && (phi != 0.))) {
989 if (r > 0) {
990 z.real = copysign(INF, cos(phi));
991 z.imag = copysign(INF, sin(phi));
992 }
993 else {
994 z.real = -copysign(INF, cos(phi));
995 z.imag = -copysign(INF, sin(phi));
996 }
997 }
998 else {
999 z = rect_special_values[special_type(r)]
1000 [special_type(phi)];
1001 }
1002 /* need to set errno = EDOM if r is a nonzero number and phi
1003 is infinite */
1004 if (r != 0. && !Py_IS_NAN(r) && Py_IS_INFINITY(phi))
1005 errno = EDOM;
1006 else
1007 errno = 0;
1008 }
1009 else {
1010 z.real = r * cos(phi);
1011 z.imag = r * sin(phi);
1012 errno = 0;
1013 }
1014
1015 PyFPE_END_PROTECT(z)
1016 if (errno != 0)
1017 return math_error();
1018 else
1019 return PyComplex_FromCComplex(z);
1020}
1021
1022PyDoc_STRVAR(cmath_rect_doc,
1023"rect(r, phi) -> z: complex\n\n\
1024Convert from polar coordinates to rectangular coordinates.");
1025
1026static PyObject *
1027cmath_isnan(PyObject *self, PyObject *args)
1028{
1029 Py_complex z;
1030 if (!PyArg_ParseTuple(args, "D:isnan", &z))
1031 return NULL;
1032 return PyBool_FromLong(Py_IS_NAN(z.real) || Py_IS_NAN(z.imag));
1033}
1034
1035PyDoc_STRVAR(cmath_isnan_doc,
1036"isnan(z) -> bool\n\
1037Checks if the real or imaginary part of z not a number (NaN)");
1038
1039static PyObject *
1040cmath_isinf(PyObject *self, PyObject *args)
1041{
1042 Py_complex z;
1043 if (!PyArg_ParseTuple(args, "D:isnan", &z))
1044 return NULL;
1045 return PyBool_FromLong(Py_IS_INFINITY(z.real) ||
1046 Py_IS_INFINITY(z.imag));
1047}
1048
1049PyDoc_STRVAR(cmath_isinf_doc,
1050"isinf(z) -> bool\n\
1051Checks if the real or imaginary part of z is infinite.");
1052
Guido van Rossum71aa32f1996-01-12 01:34:57 +00001053
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001054PyDoc_STRVAR(module_doc,
Tim Peters14e26402001-02-20 20:15:19 +00001055"This module is always available. It provides access to mathematical\n"
Martin v. Löwis14f8b4c2002-06-13 20:33:02 +00001056"functions for complex numbers.");
Guido van Rossumc6e22901998-12-04 19:26:43 +00001057
Roger E. Masse24070ca1996-12-09 22:59:53 +00001058static PyMethodDef cmath_methods[] = {
Tim Peters14e26402001-02-20 20:15:19 +00001059 {"acos", cmath_acos, METH_VARARGS, c_acos_doc},
1060 {"acosh", cmath_acosh, METH_VARARGS, c_acosh_doc},
1061 {"asin", cmath_asin, METH_VARARGS, c_asin_doc},
1062 {"asinh", cmath_asinh, METH_VARARGS, c_asinh_doc},
1063 {"atan", cmath_atan, METH_VARARGS, c_atan_doc},
1064 {"atanh", cmath_atanh, METH_VARARGS, c_atanh_doc},
1065 {"cos", cmath_cos, METH_VARARGS, c_cos_doc},
1066 {"cosh", cmath_cosh, METH_VARARGS, c_cosh_doc},
1067 {"exp", cmath_exp, METH_VARARGS, c_exp_doc},
Christian Heimes53876d92008-04-19 00:31:39 +00001068 {"isinf", cmath_isinf, METH_VARARGS, cmath_isinf_doc},
1069 {"isnan", cmath_isnan, METH_VARARGS, cmath_isnan_doc},
Raymond Hettingerb67ad7e2004-06-14 07:40:10 +00001070 {"log", cmath_log, METH_VARARGS, cmath_log_doc},
Tim Peters14e26402001-02-20 20:15:19 +00001071 {"log10", cmath_log10, METH_VARARGS, c_log10_doc},
Christian Heimes53876d92008-04-19 00:31:39 +00001072 {"phase", cmath_phase, METH_VARARGS, cmath_phase_doc},
1073 {"polar", cmath_polar, METH_VARARGS, cmath_polar_doc},
1074 {"rect", cmath_rect, METH_VARARGS, cmath_rect_doc},
Tim Peters14e26402001-02-20 20:15:19 +00001075 {"sin", cmath_sin, METH_VARARGS, c_sin_doc},
1076 {"sinh", cmath_sinh, METH_VARARGS, c_sinh_doc},
1077 {"sqrt", cmath_sqrt, METH_VARARGS, c_sqrt_doc},
1078 {"tan", cmath_tan, METH_VARARGS, c_tan_doc},
1079 {"tanh", cmath_tanh, METH_VARARGS, c_tanh_doc},
Guido van Rossum71aa32f1996-01-12 01:34:57 +00001080 {NULL, NULL} /* sentinel */
1081};
1082
Martin v. Löwis1a214512008-06-11 05:26:20 +00001083
1084static struct PyModuleDef cmathmodule = {
1085 PyModuleDef_HEAD_INIT,
1086 "cmath",
1087 module_doc,
1088 -1,
1089 cmath_methods,
1090 NULL,
1091 NULL,
1092 NULL,
1093 NULL
1094};
1095
Mark Hammondfe51c6d2002-08-02 02:27:13 +00001096PyMODINIT_FUNC
Martin v. Löwis1a214512008-06-11 05:26:20 +00001097PyInit_cmath(void)
Guido van Rossum71aa32f1996-01-12 01:34:57 +00001098{
Fred Drakef4e34842002-04-01 03:45:06 +00001099 PyObject *m;
Tim Peters14e26402001-02-20 20:15:19 +00001100
Martin v. Löwis1a214512008-06-11 05:26:20 +00001101 m = PyModule_Create(&cmathmodule);
Neal Norwitz1ac754f2006-01-19 06:09:39 +00001102 if (m == NULL)
Martin v. Löwis1a214512008-06-11 05:26:20 +00001103 return NULL;
Fred Drakef4e34842002-04-01 03:45:06 +00001104
1105 PyModule_AddObject(m, "pi",
Christian Heimes53876d92008-04-19 00:31:39 +00001106 PyFloat_FromDouble(Py_MATH_PI));
1107 PyModule_AddObject(m, "e", PyFloat_FromDouble(Py_MATH_E));
Christian Heimesa342c012008-04-20 21:01:16 +00001108
1109 /* initialize special value tables */
1110
1111#define INIT_SPECIAL_VALUES(NAME, BODY) { Py_complex* p = (Py_complex*)NAME; BODY }
1112#define C(REAL, IMAG) p->real = REAL; p->imag = IMAG; ++p;
1113
1114 INIT_SPECIAL_VALUES(acos_special_values, {
1115 C(P34,INF) C(P,INF) C(P,INF) C(P,-INF) C(P,-INF) C(P34,-INF) C(N,INF)
1116 C(P12,INF) C(U,U) C(U,U) C(U,U) C(U,U) C(P12,-INF) C(N,N)
1117 C(P12,INF) C(U,U) C(P12,0.) C(P12,-0.) C(U,U) C(P12,-INF) C(P12,N)
1118 C(P12,INF) C(U,U) C(P12,0.) C(P12,-0.) C(U,U) C(P12,-INF) C(P12,N)
1119 C(P12,INF) C(U,U) C(U,U) C(U,U) C(U,U) C(P12,-INF) C(N,N)
1120 C(P14,INF) C(0.,INF) C(0.,INF) C(0.,-INF) C(0.,-INF) C(P14,-INF) C(N,INF)
1121 C(N,INF) C(N,N) C(N,N) C(N,N) C(N,N) C(N,-INF) C(N,N)
1122 })
1123
1124 INIT_SPECIAL_VALUES(acosh_special_values, {
1125 C(INF,-P34) C(INF,-P) C(INF,-P) C(INF,P) C(INF,P) C(INF,P34) C(INF,N)
1126 C(INF,-P12) C(U,U) C(U,U) C(U,U) C(U,U) C(INF,P12) C(N,N)
1127 C(INF,-P12) C(U,U) C(0.,-P12) C(0.,P12) C(U,U) C(INF,P12) C(N,N)
1128 C(INF,-P12) C(U,U) C(0.,-P12) C(0.,P12) C(U,U) C(INF,P12) C(N,N)
1129 C(INF,-P12) C(U,U) C(U,U) C(U,U) C(U,U) C(INF,P12) C(N,N)
1130 C(INF,-P14) C(INF,-0.) C(INF,-0.) C(INF,0.) C(INF,0.) C(INF,P14) C(INF,N)
1131 C(INF,N) C(N,N) C(N,N) C(N,N) C(N,N) C(INF,N) C(N,N)
1132 })
1133
1134 INIT_SPECIAL_VALUES(asinh_special_values, {
1135 C(-INF,-P14) C(-INF,-0.) C(-INF,-0.) C(-INF,0.) C(-INF,0.) C(-INF,P14) C(-INF,N)
1136 C(-INF,-P12) C(U,U) C(U,U) C(U,U) C(U,U) C(-INF,P12) C(N,N)
1137 C(-INF,-P12) C(U,U) C(-0.,-0.) C(-0.,0.) C(U,U) C(-INF,P12) C(N,N)
1138 C(INF,-P12) C(U,U) C(0.,-0.) C(0.,0.) C(U,U) C(INF,P12) C(N,N)
1139 C(INF,-P12) C(U,U) C(U,U) C(U,U) C(U,U) C(INF,P12) C(N,N)
1140 C(INF,-P14) C(INF,-0.) C(INF,-0.) C(INF,0.) C(INF,0.) C(INF,P14) C(INF,N)
1141 C(INF,N) C(N,N) C(N,-0.) C(N,0.) C(N,N) C(INF,N) C(N,N)
1142 })
1143
1144 INIT_SPECIAL_VALUES(atanh_special_values, {
1145 C(-0.,-P12) C(-0.,-P12) C(-0.,-P12) C(-0.,P12) C(-0.,P12) C(-0.,P12) C(-0.,N)
1146 C(-0.,-P12) C(U,U) C(U,U) C(U,U) C(U,U) C(-0.,P12) C(N,N)
1147 C(-0.,-P12) C(U,U) C(-0.,-0.) C(-0.,0.) C(U,U) C(-0.,P12) C(-0.,N)
1148 C(0.,-P12) C(U,U) C(0.,-0.) C(0.,0.) C(U,U) C(0.,P12) C(0.,N)
1149 C(0.,-P12) C(U,U) C(U,U) C(U,U) C(U,U) C(0.,P12) C(N,N)
1150 C(0.,-P12) C(0.,-P12) C(0.,-P12) C(0.,P12) C(0.,P12) C(0.,P12) C(0.,N)
1151 C(0.,-P12) C(N,N) C(N,N) C(N,N) C(N,N) C(0.,P12) C(N,N)
1152 })
1153
1154 INIT_SPECIAL_VALUES(cosh_special_values, {
1155 C(INF,N) C(U,U) C(INF,0.) C(INF,-0.) C(U,U) C(INF,N) C(INF,N)
1156 C(N,N) C(U,U) C(U,U) C(U,U) C(U,U) C(N,N) C(N,N)
1157 C(N,0.) C(U,U) C(1.,0.) C(1.,-0.) C(U,U) C(N,0.) C(N,0.)
1158 C(N,0.) C(U,U) C(1.,-0.) C(1.,0.) C(U,U) C(N,0.) C(N,0.)
1159 C(N,N) C(U,U) C(U,U) C(U,U) C(U,U) C(N,N) C(N,N)
1160 C(INF,N) C(U,U) C(INF,-0.) C(INF,0.) C(U,U) C(INF,N) C(INF,N)
1161 C(N,N) C(N,N) C(N,0.) C(N,0.) C(N,N) C(N,N) C(N,N)
1162 })
1163
1164 INIT_SPECIAL_VALUES(exp_special_values, {
1165 C(0.,0.) C(U,U) C(0.,-0.) C(0.,0.) C(U,U) C(0.,0.) C(0.,0.)
1166 C(N,N) C(U,U) C(U,U) C(U,U) C(U,U) C(N,N) C(N,N)
1167 C(N,N) C(U,U) C(1.,-0.) C(1.,0.) C(U,U) C(N,N) C(N,N)
1168 C(N,N) C(U,U) C(1.,-0.) C(1.,0.) C(U,U) C(N,N) C(N,N)
1169 C(N,N) C(U,U) C(U,U) C(U,U) C(U,U) C(N,N) C(N,N)
1170 C(INF,N) C(U,U) C(INF,-0.) C(INF,0.) C(U,U) C(INF,N) C(INF,N)
1171 C(N,N) C(N,N) C(N,-0.) C(N,0.) C(N,N) C(N,N) C(N,N)
1172 })
1173
1174 INIT_SPECIAL_VALUES(log_special_values, {
1175 C(INF,-P34) C(INF,-P) C(INF,-P) C(INF,P) C(INF,P) C(INF,P34) C(INF,N)
1176 C(INF,-P12) C(U,U) C(U,U) C(U,U) C(U,U) C(INF,P12) C(N,N)
1177 C(INF,-P12) C(U,U) C(-INF,-P) C(-INF,P) C(U,U) C(INF,P12) C(N,N)
1178 C(INF,-P12) C(U,U) C(-INF,-0.) C(-INF,0.) C(U,U) C(INF,P12) C(N,N)
1179 C(INF,-P12) C(U,U) C(U,U) C(U,U) C(U,U) C(INF,P12) C(N,N)
1180 C(INF,-P14) C(INF,-0.) C(INF,-0.) C(INF,0.) C(INF,0.) C(INF,P14) C(INF,N)
1181 C(INF,N) C(N,N) C(N,N) C(N,N) C(N,N) C(INF,N) C(N,N)
1182 })
1183
1184 INIT_SPECIAL_VALUES(sinh_special_values, {
1185 C(INF,N) C(U,U) C(-INF,-0.) C(-INF,0.) C(U,U) C(INF,N) C(INF,N)
1186 C(N,N) C(U,U) C(U,U) C(U,U) C(U,U) C(N,N) C(N,N)
1187 C(0.,N) C(U,U) C(-0.,-0.) C(-0.,0.) C(U,U) C(0.,N) C(0.,N)
1188 C(0.,N) C(U,U) C(0.,-0.) C(0.,0.) C(U,U) C(0.,N) C(0.,N)
1189 C(N,N) C(U,U) C(U,U) C(U,U) C(U,U) C(N,N) C(N,N)
1190 C(INF,N) C(U,U) C(INF,-0.) C(INF,0.) C(U,U) C(INF,N) C(INF,N)
1191 C(N,N) C(N,N) C(N,-0.) C(N,0.) C(N,N) C(N,N) C(N,N)
1192 })
1193
1194 INIT_SPECIAL_VALUES(sqrt_special_values, {
1195 C(INF,-INF) C(0.,-INF) C(0.,-INF) C(0.,INF) C(0.,INF) C(INF,INF) C(N,INF)
1196 C(INF,-INF) C(U,U) C(U,U) C(U,U) C(U,U) C(INF,INF) C(N,N)
1197 C(INF,-INF) C(U,U) C(0.,-0.) C(0.,0.) C(U,U) C(INF,INF) C(N,N)
1198 C(INF,-INF) C(U,U) C(0.,-0.) C(0.,0.) C(U,U) C(INF,INF) C(N,N)
1199 C(INF,-INF) C(U,U) C(U,U) C(U,U) C(U,U) C(INF,INF) C(N,N)
1200 C(INF,-INF) C(INF,-0.) C(INF,-0.) C(INF,0.) C(INF,0.) C(INF,INF) C(INF,N)
1201 C(INF,-INF) C(N,N) C(N,N) C(N,N) C(N,N) C(INF,INF) C(N,N)
1202 })
1203
1204 INIT_SPECIAL_VALUES(tanh_special_values, {
1205 C(-1.,0.) C(U,U) C(-1.,-0.) C(-1.,0.) C(U,U) C(-1.,0.) C(-1.,0.)
1206 C(N,N) C(U,U) C(U,U) C(U,U) C(U,U) C(N,N) C(N,N)
1207 C(N,N) C(U,U) C(-0.,-0.) C(-0.,0.) C(U,U) C(N,N) C(N,N)
1208 C(N,N) C(U,U) C(0.,-0.) C(0.,0.) C(U,U) C(N,N) C(N,N)
1209 C(N,N) C(U,U) C(U,U) C(U,U) C(U,U) C(N,N) C(N,N)
1210 C(1.,0.) C(U,U) C(1.,-0.) C(1.,0.) C(U,U) C(1.,0.) C(1.,0.)
1211 C(N,N) C(N,N) C(N,-0.) C(N,0.) C(N,N) C(N,N) C(N,N)
1212 })
1213
1214 INIT_SPECIAL_VALUES(rect_special_values, {
1215 C(INF,N) C(U,U) C(-INF,0.) C(-INF,-0.) C(U,U) C(INF,N) C(INF,N)
1216 C(N,N) C(U,U) C(U,U) C(U,U) C(U,U) C(N,N) C(N,N)
1217 C(0.,0.) C(U,U) C(-0.,0.) C(-0.,-0.) C(U,U) C(0.,0.) C(0.,0.)
1218 C(0.,0.) C(U,U) C(0.,-0.) C(0.,0.) C(U,U) C(0.,0.) C(0.,0.)
1219 C(N,N) C(U,U) C(U,U) C(U,U) C(U,U) C(N,N) C(N,N)
1220 C(INF,N) C(U,U) C(INF,-0.) C(INF,0.) C(U,U) C(INF,N) C(INF,N)
1221 C(N,N) C(N,N) C(N,0.) C(N,0.) C(N,N) C(N,N) C(N,N)
1222 })
Martin v. Löwis1a214512008-06-11 05:26:20 +00001223 return m;
Guido van Rossum71aa32f1996-01-12 01:34:57 +00001224}