1 | /* Inline math functions for i387 and SSE. |
2 | Copyright (C) 1995-2018 Free Software Foundation, Inc. |
3 | This file is part of the GNU C Library. |
4 | |
5 | The GNU C Library is free software; you can redistribute it and/or |
6 | modify it under the terms of the GNU Lesser General Public |
7 | License as published by the Free Software Foundation; either |
8 | version 2.1 of the License, or (at your option) any later version. |
9 | |
10 | The GNU C Library is distributed in the hope that it will be useful, |
11 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
13 | Lesser General Public License for more details. |
14 | |
15 | You should have received a copy of the GNU Lesser General Public |
16 | License along with the GNU C Library; if not, see |
17 | <http://www.gnu.org/licenses/>. */ |
18 | |
19 | #ifndef _MATH_H |
20 | # error "Never use <bits/mathinline.h> directly; include <math.h> instead." |
21 | #endif |
22 | |
23 | #ifndef __extern_always_inline |
24 | # define __MATH_INLINE __inline |
25 | #else |
26 | # define __MATH_INLINE __extern_always_inline |
27 | #endif |
28 | |
29 | /* Disable x87 inlines when -fpmath=sse is passed and also when we're building |
30 | on x86_64. Older gcc (gcc-3.2 for example) does not define __SSE2_MATH__ |
31 | for x86_64. */ |
32 | #if !defined __SSE2_MATH__ && !defined __x86_64__ |
33 | # if ((!defined __NO_MATH_INLINES || defined __LIBC_INTERNAL_MATH_INLINES) \ |
34 | && defined __OPTIMIZE__) |
35 | |
36 | /* The inline functions do not set errno or raise necessarily the |
37 | correct exceptions. */ |
38 | # undef math_errhandling |
39 | |
40 | /* A macro to define float, double, and long double versions of various |
41 | math functions for the ix87 FPU. FUNC is the function name (which will |
42 | be suffixed with f and l for the float and long double version, |
43 | respectively). OP is the name of the FPU operation. |
44 | We define two sets of macros. The set with the additional NP |
45 | doesn't add a prototype declaration. */ |
46 | |
47 | # ifdef __USE_ISOC99 |
48 | # define __inline_mathop(func, op) \ |
49 | __inline_mathop_ (double, func, op) \ |
50 | __inline_mathop_ (float, __CONCAT(func,f), op) \ |
51 | __inline_mathop_ (long double, __CONCAT(func,l), op) |
52 | # define __inline_mathopNP(func, op) \ |
53 | __inline_mathopNP_ (double, func, op) \ |
54 | __inline_mathopNP_ (float, __CONCAT(func,f), op) \ |
55 | __inline_mathopNP_ (long double, __CONCAT(func,l), op) |
56 | # else |
57 | # define __inline_mathop(func, op) \ |
58 | __inline_mathop_ (double, func, op) |
59 | # define __inline_mathopNP(func, op) \ |
60 | __inline_mathopNP_ (double, func, op) |
61 | # endif |
62 | |
63 | # define __inline_mathop_(float_type, func, op) \ |
64 | __inline_mathop_decl_ (float_type, func, op, "0" (__x)) |
65 | # define __inline_mathopNP_(float_type, func, op) \ |
66 | __inline_mathop_declNP_ (float_type, func, op, "0" (__x)) |
67 | |
68 | |
69 | # ifdef __USE_ISOC99 |
70 | # define __inline_mathop_decl(func, op, params...) \ |
71 | __inline_mathop_decl_ (double, func, op, params) \ |
72 | __inline_mathop_decl_ (float, __CONCAT(func,f), op, params) \ |
73 | __inline_mathop_decl_ (long double, __CONCAT(func,l), op, params) |
74 | # define __inline_mathop_declNP(func, op, params...) \ |
75 | __inline_mathop_declNP_ (double, func, op, params) \ |
76 | __inline_mathop_declNP_ (float, __CONCAT(func,f), op, params) \ |
77 | __inline_mathop_declNP_ (long double, __CONCAT(func,l), op, params) |
78 | # else |
79 | # define __inline_mathop_decl(func, op, params...) \ |
80 | __inline_mathop_decl_ (double, func, op, params) |
81 | # define __inline_mathop_declNP(func, op, params...) \ |
82 | __inline_mathop_declNP_ (double, func, op, params) |
83 | # endif |
84 | |
85 | # define __inline_mathop_decl_(float_type, func, op, params...) \ |
86 | __MATH_INLINE float_type func (float_type) __THROW; \ |
87 | __inline_mathop_declNP_ (float_type, func, op, params) |
88 | |
89 | # define __inline_mathop_declNP_(float_type, func, op, params...) \ |
90 | __MATH_INLINE float_type __NTH (func (float_type __x)) \ |
91 | { \ |
92 | register float_type __result; \ |
93 | __asm __volatile__ (op : "=t" (__result) : params); \ |
94 | return __result; \ |
95 | } |
96 | |
97 | |
98 | # ifdef __USE_ISOC99 |
99 | # define __inline_mathcode(func, arg, code) \ |
100 | __inline_mathcode_ (double, func, arg, code) \ |
101 | __inline_mathcode_ (float, __CONCAT(func,f), arg, code) \ |
102 | __inline_mathcode_ (long double, __CONCAT(func,l), arg, code) |
103 | # define __inline_mathcodeNP(func, arg, code) \ |
104 | __inline_mathcodeNP_ (double, func, arg, code) \ |
105 | __inline_mathcodeNP_ (float, __CONCAT(func,f), arg, code) \ |
106 | __inline_mathcodeNP_ (long double, __CONCAT(func,l), arg, code) |
107 | # define __inline_mathcode2(func, arg1, arg2, code) \ |
108 | __inline_mathcode2_ (double, func, arg1, arg2, code) \ |
109 | __inline_mathcode2_ (float, __CONCAT(func,f), arg1, arg2, code) \ |
110 | __inline_mathcode2_ (long double, __CONCAT(func,l), arg1, arg2, code) |
111 | # define __inline_mathcodeNP2(func, arg1, arg2, code) \ |
112 | __inline_mathcodeNP2_ (double, func, arg1, arg2, code) \ |
113 | __inline_mathcodeNP2_ (float, __CONCAT(func,f), arg1, arg2, code) \ |
114 | __inline_mathcodeNP2_ (long double, __CONCAT(func,l), arg1, arg2, code) |
115 | # define __inline_mathcode3(func, arg1, arg2, arg3, code) \ |
116 | __inline_mathcode3_ (double, func, arg1, arg2, arg3, code) \ |
117 | __inline_mathcode3_ (float, __CONCAT(func,f), arg1, arg2, arg3, code) \ |
118 | __inline_mathcode3_ (long double, __CONCAT(func,l), arg1, arg2, arg3, code) |
119 | # define __inline_mathcodeNP3(func, arg1, arg2, arg3, code) \ |
120 | __inline_mathcodeNP3_ (double, func, arg1, arg2, arg3, code) \ |
121 | __inline_mathcodeNP3_ (float, __CONCAT(func,f), arg1, arg2, arg3, code) \ |
122 | __inline_mathcodeNP3_ (long double, __CONCAT(func,l), arg1, arg2, arg3, code) |
123 | # else |
124 | # define __inline_mathcode(func, arg, code) \ |
125 | __inline_mathcode_ (double, func, (arg), code) |
126 | # define __inline_mathcodeNP(func, arg, code) \ |
127 | __inline_mathcodeNP_ (double, func, (arg), code) |
128 | # define __inline_mathcode2(func, arg1, arg2, code) \ |
129 | __inline_mathcode2_ (double, func, arg1, arg2, code) |
130 | # define __inline_mathcodeNP2(func, arg1, arg2, code) \ |
131 | __inline_mathcodeNP2_ (double, func, arg1, arg2, code) |
132 | # define __inline_mathcode3(func, arg1, arg2, arg3, code) \ |
133 | __inline_mathcode3_ (double, func, arg1, arg2, arg3, code) |
134 | # define __inline_mathcodeNP3(func, arg1, arg2, arg3, code) \ |
135 | __inline_mathcodeNP3_ (double, func, arg1, arg2, arg3, code) |
136 | # endif |
137 | |
138 | # define __inline_mathcode_(float_type, func, arg, code) \ |
139 | __MATH_INLINE float_type func (float_type) __THROW; \ |
140 | __inline_mathcodeNP_(float_type, func, arg, code) |
141 | |
142 | # define __inline_mathcodeNP_(float_type, func, arg, code) \ |
143 | __MATH_INLINE float_type __NTH (func (float_type arg)) \ |
144 | { \ |
145 | code; \ |
146 | } |
147 | |
148 | |
149 | # define __inline_mathcode2_(float_type, func, arg1, arg2, code) \ |
150 | __MATH_INLINE float_type func (float_type, float_type) __THROW; \ |
151 | __inline_mathcodeNP2_ (float_type, func, arg1, arg2, code) |
152 | |
153 | # define __inline_mathcodeNP2_(float_type, func, arg1, arg2, code) \ |
154 | __MATH_INLINE float_type __NTH (func (float_type arg1, float_type arg2)) \ |
155 | { \ |
156 | code; \ |
157 | } |
158 | |
159 | # define __inline_mathcode3_(float_type, func, arg1, arg2, arg3, code) \ |
160 | __MATH_INLINE float_type func (float_type, float_type, float_type) __THROW; \ |
161 | __inline_mathcodeNP3_(float_type, func, arg1, arg2, arg3, code) |
162 | |
163 | # define __inline_mathcodeNP3_(float_type, func, arg1, arg2, arg3, code) \ |
164 | __MATH_INLINE float_type __NTH (func (float_type arg1, float_type arg2, \ |
165 | float_type arg3)) \ |
166 | { \ |
167 | code; \ |
168 | } |
169 | # endif |
170 | |
171 | |
172 | # if !defined __NO_MATH_INLINES && defined __OPTIMIZE__ |
173 | /* Miscellaneous functions */ |
174 | |
175 | /* __FAST_MATH__ is defined by gcc -ffast-math. */ |
176 | # ifdef __FAST_MATH__ |
177 | /* Optimized inline implementation, sometimes with reduced precision |
178 | and/or argument range. */ |
179 | |
180 | # if __GNUC_PREREQ (3, 5) |
181 | # define __expm1_code \ |
182 | register long double __temp; \ |
183 | __temp = __builtin_expm1l (__x); \ |
184 | return __temp ? __temp : __x |
185 | # else |
186 | # define __expm1_code \ |
187 | register long double __value; \ |
188 | register long double __exponent; \ |
189 | register long double __temp; \ |
190 | __asm __volatile__ \ |
191 | ("fldl2e # e^x - 1 = 2^(x * log2(e)) - 1\n\t" \ |
192 | "fmul %%st(1) # x * log2(e)\n\t" \ |
193 | "fst %%st(1)\n\t" \ |
194 | "frndint # int(x * log2(e))\n\t" \ |
195 | "fxch\n\t" \ |
196 | "fsub %%st(1) # fract(x * log2(e))\n\t" \ |
197 | "f2xm1 # 2^(fract(x * log2(e))) - 1\n\t" \ |
198 | "fscale # 2^(x * log2(e)) - 2^(int(x * log2(e)))\n\t" \ |
199 | : "=t" (__value), "=u" (__exponent) : "0" (__x)); \ |
200 | __asm __volatile__ \ |
201 | ("fscale # 2^int(x * log2(e))\n\t" \ |
202 | : "=t" (__temp) : "0" (1.0), "u" (__exponent)); \ |
203 | __temp -= 1.0; \ |
204 | __temp += __value; \ |
205 | return __temp ? __temp : __x |
206 | # endif |
207 | __inline_mathcodeNP_ (long double, __expm1l, __x, __expm1_code) |
208 | |
209 | # if __GNUC_PREREQ (3, 4) |
210 | __inline_mathcodeNP_ (long double, __expl, __x, return __builtin_expl (__x)) |
211 | # else |
212 | # define __exp_code \ |
213 | register long double __value; \ |
214 | register long double __exponent; \ |
215 | __asm __volatile__ \ |
216 | ("fldl2e # e^x = 2^(x * log2(e))\n\t" \ |
217 | "fmul %%st(1) # x * log2(e)\n\t" \ |
218 | "fst %%st(1)\n\t" \ |
219 | "frndint # int(x * log2(e))\n\t" \ |
220 | "fxch\n\t" \ |
221 | "fsub %%st(1) # fract(x * log2(e))\n\t" \ |
222 | "f2xm1 # 2^(fract(x * log2(e))) - 1\n\t" \ |
223 | : "=t" (__value), "=u" (__exponent) : "0" (__x)); \ |
224 | __value += 1.0; \ |
225 | __asm __volatile__ \ |
226 | ("fscale" \ |
227 | : "=t" (__value) : "0" (__value), "u" (__exponent)); \ |
228 | return __value |
229 | __inline_mathcodeNP (exp, __x, __exp_code) |
230 | __inline_mathcodeNP_ (long double, __expl, __x, __exp_code) |
231 | # endif |
232 | # endif /* __FAST_MATH__ */ |
233 | |
234 | |
235 | # ifdef __FAST_MATH__ |
236 | # if !__GNUC_PREREQ (3,3) |
237 | __inline_mathopNP (sqrt, "fsqrt" ) |
238 | __inline_mathopNP_ (long double, __sqrtl, "fsqrt" ) |
239 | # define __libc_sqrtl(n) __sqrtl (n) |
240 | # else |
241 | # define __libc_sqrtl(n) __builtin_sqrtl (n) |
242 | # endif |
243 | # endif |
244 | |
245 | # if __GNUC_PREREQ (2, 8) |
246 | __inline_mathcodeNP_ (double, fabs, __x, return __builtin_fabs (__x)) |
247 | # ifdef __USE_ISOC99 |
248 | __inline_mathcodeNP_ (float, fabsf, __x, return __builtin_fabsf (__x)) |
249 | __inline_mathcodeNP_ (long double, fabsl, __x, return __builtin_fabsl (__x)) |
250 | # endif |
251 | __inline_mathcodeNP_ (long double, __fabsl, __x, return __builtin_fabsl (__x)) |
252 | # else |
253 | __inline_mathop (fabs, "fabs" ) |
254 | __inline_mathop_ (long double, __fabsl, "fabs" ) |
255 | # endif |
256 | |
257 | __inline_mathcode_ (long double, __sgn1l, __x, \ |
258 | __extension__ union { long double __xld; unsigned int __xi[3]; } __n = \ |
259 | { __xld: __x }; \ |
260 | __n.__xi[2] = (__n.__xi[2] & 0x8000) | 0x3fff; \ |
261 | __n.__xi[1] = 0x80000000; \ |
262 | __n.__xi[0] = 0; \ |
263 | return __n.__xld) |
264 | |
265 | |
266 | # ifdef __FAST_MATH__ |
267 | /* The argument range of the inline version of sinhl is slightly reduced. */ |
268 | __inline_mathcodeNP (sinh, __x, \ |
269 | register long double __exm1 = __expm1l (__fabsl (__x)); \ |
270 | return 0.5 * (__exm1 / (__exm1 + 1.0) + __exm1) * __sgn1l (__x)) |
271 | |
272 | __inline_mathcodeNP (cosh, __x, \ |
273 | register long double __ex = __expl (__x); \ |
274 | return 0.5 * (__ex + 1.0 / __ex)) |
275 | |
276 | __inline_mathcodeNP (tanh, __x, \ |
277 | register long double __exm1 = __expm1l (-__fabsl (__x + __x)); \ |
278 | return __exm1 / (__exm1 + 2.0) * __sgn1l (-__x)) |
279 | # endif |
280 | |
281 | |
282 | /* Optimized versions for some non-standardized functions. */ |
283 | # ifdef __USE_ISOC99 |
284 | |
285 | # ifdef __FAST_MATH__ |
286 | __inline_mathcodeNP (expm1, __x, __expm1_code) |
287 | |
288 | /* The argument range of the inline version of asinhl is slightly reduced. */ |
289 | __inline_mathcodeNP (asinh, __x, \ |
290 | register long double __y = __fabsl (__x); \ |
291 | return (log1pl (__y * __y / (__libc_sqrtl (__y * __y + 1.0) + 1.0) + __y) \ |
292 | * __sgn1l (__x))) |
293 | |
294 | __inline_mathcodeNP (acosh, __x, \ |
295 | return logl (__x + __libc_sqrtl (__x - 1.0) * __libc_sqrtl (__x + 1.0))) |
296 | |
297 | __inline_mathcodeNP (atanh, __x, \ |
298 | register long double __y = __fabsl (__x); \ |
299 | return -0.5 * log1pl (-(__y + __y) / (1.0 + __y)) * __sgn1l (__x)) |
300 | |
301 | /* The argument range of the inline version of hypotl is slightly reduced. */ |
302 | __inline_mathcodeNP2 (hypot, __x, __y, |
303 | return __libc_sqrtl (__x * __x + __y * __y)) |
304 | |
305 | # endif |
306 | # endif |
307 | |
308 | |
309 | /* Undefine some of the large macros which are not used anymore. */ |
310 | # ifdef __FAST_MATH__ |
311 | # undef __expm1_code |
312 | # undef __exp_code |
313 | # endif /* __FAST_MATH__ */ |
314 | |
315 | # endif /* __NO_MATH_INLINES */ |
316 | |
317 | |
318 | /* This code is used internally in the GNU libc. */ |
319 | # ifdef __LIBC_INTERNAL_MATH_INLINES |
320 | __inline_mathcode2_ (long double, __ieee754_atan2l, __y, __x, |
321 | register long double __value; |
322 | __asm __volatile__ ("fpatan\n\t" |
323 | : "=t" (__value) |
324 | : "0" (__x), "u" (__y) : "st(1)" ); |
325 | return __value;) |
326 | # endif |
327 | |
328 | #endif /* !__SSE2_MATH__ && !__x86_64__ */ |
329 | |