1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2004-2005 David Schultz <das@FreeBSD.ORG>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 #ifndef _FENV_H_
30 #define _FENV_H_
31
32 #include <sys/cdefs.h>
33 #include <sys/_types.h>
34 #include <ieeefp.h>
35
36 #ifndef __fenv_static
37 #define __fenv_static static
38 #endif
39
40 typedef __uint16_t fexcept_t;
41
42 /* Exception flags */
43 #define FE_INVALID 0x01
44 #define FE_DENORMAL 0x02
45 #define FE_DIVBYZERO 0x04
46 #define FE_OVERFLOW 0x08
47 #define FE_UNDERFLOW 0x10
48 #define FE_INEXACT 0x20
49 #define FE_ALL_EXCEPT (FE_DIVBYZERO | FE_DENORMAL | FE_INEXACT | \
50 FE_INVALID | FE_OVERFLOW | FE_UNDERFLOW)
51
52 /* Rounding modes */
53 #define FE_TONEAREST 0x0000
54 #define FE_DOWNWARD 0x0400
55 #define FE_UPWARD 0x0800
56 #define FE_TOWARDZERO 0x0c00
57 #define _ROUND_MASK (FE_TONEAREST | FE_DOWNWARD | \
58 FE_UPWARD | FE_TOWARDZERO)
59
60 /*
61 * As compared to the x87 control word, the SSE unit's control word
62 * has the rounding control bits offset by 3 and the exception mask
63 * bits offset by 7.
64 */
65 #define _SSE_ROUND_SHIFT 3
66 #define _SSE_EMASK_SHIFT 7
67
68 #ifdef __i386__
69 /*
70 * To preserve binary compatibility with FreeBSD 5.3, we pack the
71 * mxcsr into some reserved fields, rather than changing sizeof(fenv_t).
72 */
73 typedef struct {
74 __uint16_t __control;
75 __uint16_t __mxcsr_hi;
76 __uint16_t __status;
77 __uint16_t __mxcsr_lo;
78 __uint32_t __tag;
79 char __other[16];
80 } fenv_t;
81 #else /* __amd64__ */
82 typedef struct {
83 struct {
84 __uint32_t __control;
85 __uint32_t __status;
86 __uint32_t __tag;
87 char __other[16];
88 } __x87;
89 __uint32_t __mxcsr;
90 } fenv_t;
91 #endif /* __i386__ */
92
93 __BEGIN_DECLS
94
95 /* Default floating-point environment */
96 extern const fenv_t __fe_dfl_env;
97 #define FE_DFL_ENV (&__fe_dfl_env)
98
99 #define __fldenvx(__env) __asm __volatile("fldenv %0" : : "m" (__env) \
100 : "st", "st(1)", "st(2)", "st(3)", "st(4)", \
101 "st(5)", "st(6)", "st(7)")
102 #define __fwait() __asm __volatile("fwait")
103
104 int fegetenv(fenv_t *__envp);
105 int feholdexcept(fenv_t *__envp);
106 int fesetexceptflag(const fexcept_t *__flagp, int __excepts);
107 int feraiseexcept(int __excepts);
108 int feupdateenv(const fenv_t *__envp);
109
110 __fenv_static inline int
fegetround(void)111 fegetround(void)
112 {
113 __uint16_t __control;
114
115 /*
116 * We assume that the x87 and the SSE unit agree on the
117 * rounding mode. Reading the control word on the x87 turns
118 * out to be about 5 times faster than reading it on the SSE
119 * unit on an Opteron 244.
120 */
121 __fnstcw(&__control);
122 return (__control & _ROUND_MASK);
123 }
124
125 #if __BSD_VISIBLE
126
127 int feenableexcept(int __mask);
128 int fedisableexcept(int __mask);
129
130 /* We currently provide no external definition of fegetexcept(). */
131 static inline int
fegetexcept(void)132 fegetexcept(void)
133 {
134 __uint16_t __control;
135
136 /*
137 * We assume that the masks for the x87 and the SSE unit are
138 * the same.
139 */
140 __fnstcw(&__control);
141 return (~__control & FE_ALL_EXCEPT);
142 }
143
144 #endif /* __BSD_VISIBLE */
145
146 #ifdef __i386__
147
148 /* After testing for SSE support once, we cache the result in __has_sse. */
149 enum __sse_support { __SSE_YES, __SSE_NO, __SSE_UNK };
150 extern enum __sse_support __has_sse;
151 int __test_sse(void);
152 #ifdef __SSE__
153 #define __HAS_SSE() 1
154 #else
155 #define __HAS_SSE() (__has_sse == __SSE_YES || \
156 (__has_sse == __SSE_UNK && __test_sse()))
157 #endif
158
159 #define __get_mxcsr(env) (((env).__mxcsr_hi << 16) | \
160 ((env).__mxcsr_lo))
161 #define __set_mxcsr(env, x) do { \
162 (env).__mxcsr_hi = (__uint32_t)(x) >> 16; \
163 (env).__mxcsr_lo = (__uint16_t)(x); \
164 } while (0)
165
166 __fenv_static inline int
feclearexcept(int __excepts)167 feclearexcept(int __excepts)
168 {
169 fenv_t __env;
170 __uint32_t __mxcsr;
171
172 if (__excepts == FE_ALL_EXCEPT) {
173 __fnclex();
174 } else {
175 __fnstenv(&__env);
176 __env.__status &= ~__excepts;
177 __fldenv(&__env);
178 }
179 if (__HAS_SSE()) {
180 __stmxcsr(&__mxcsr);
181 __mxcsr &= ~__excepts;
182 __ldmxcsr(&__mxcsr);
183 }
184 return (0);
185 }
186
187 __fenv_static inline int
fegetexceptflag(fexcept_t * __flagp,int __excepts)188 fegetexceptflag(fexcept_t *__flagp, int __excepts)
189 {
190 __uint32_t __mxcsr;
191 __uint16_t __status;
192
193 __fnstsw(&__status);
194 if (__HAS_SSE())
195 __stmxcsr(&__mxcsr);
196 else
197 __mxcsr = 0;
198 *__flagp = (__mxcsr | __status) & __excepts;
199 return (0);
200 }
201
202 __fenv_static inline int
fetestexcept(int __excepts)203 fetestexcept(int __excepts)
204 {
205 __uint32_t __mxcsr;
206 __uint16_t __status;
207
208 __fnstsw(&__status);
209 if (__HAS_SSE())
210 __stmxcsr(&__mxcsr);
211 else
212 __mxcsr = 0;
213 return ((__status | __mxcsr) & __excepts);
214 }
215
216 __fenv_static inline int
fesetround(int __round)217 fesetround(int __round)
218 {
219 __uint32_t __mxcsr;
220 __uint16_t __control;
221
222 if (__round & ~_ROUND_MASK)
223 return (-1);
224
225 __fnstcw(&__control);
226 __control &= ~_ROUND_MASK;
227 __control |= __round;
228 __fldcw(&__control);
229
230 if (__HAS_SSE()) {
231 __stmxcsr(&__mxcsr);
232 __mxcsr &= ~(_ROUND_MASK << _SSE_ROUND_SHIFT);
233 __mxcsr |= __round << _SSE_ROUND_SHIFT;
234 __ldmxcsr(&__mxcsr);
235 }
236
237 return (0);
238 }
239
240 __fenv_static inline int
fesetenv(const fenv_t * __envp)241 fesetenv(const fenv_t *__envp)
242 {
243 fenv_t __env = *__envp;
244 __uint32_t __mxcsr;
245
246 __mxcsr = __get_mxcsr(__env);
247 __set_mxcsr(__env, 0xffffffff);
248 /*
249 * XXX Using fldenvx() instead of fldenv() tells the compiler that this
250 * instruction clobbers the i387 register stack. This happens because
251 * we restore the tag word from the saved environment. Normally, this
252 * would happen anyway and we wouldn't care, because the ABI allows
253 * function calls to clobber the i387 regs. However, fesetenv() is
254 * inlined, so we need to be more careful.
255 */
256 __fldenvx(__env);
257 if (__HAS_SSE())
258 __ldmxcsr(&__mxcsr);
259 return (0);
260 }
261
262 #else /* __amd64__ */
263
264 __fenv_static inline int
feclearexcept(int __excepts)265 feclearexcept(int __excepts)
266 {
267 fenv_t __env;
268
269 if (__excepts == FE_ALL_EXCEPT) {
270 __fnclex();
271 } else {
272 __fnstenv(&__env.__x87);
273 __env.__x87.__status &= ~__excepts;
274 __fldenv(&__env.__x87);
275 }
276 __stmxcsr(&__env.__mxcsr);
277 __env.__mxcsr &= ~__excepts;
278 __ldmxcsr(&__env.__mxcsr);
279 return (0);
280 }
281
282 __fenv_static inline int
fegetexceptflag(fexcept_t * __flagp,int __excepts)283 fegetexceptflag(fexcept_t *__flagp, int __excepts)
284 {
285 __uint32_t __mxcsr;
286 __uint16_t __status;
287
288 __stmxcsr(&__mxcsr);
289 __fnstsw(&__status);
290 *__flagp = (__mxcsr | __status) & __excepts;
291 return (0);
292 }
293
294 __fenv_static inline int
fetestexcept(int __excepts)295 fetestexcept(int __excepts)
296 {
297 __uint32_t __mxcsr;
298 __uint16_t __status;
299
300 __stmxcsr(&__mxcsr);
301 __fnstsw(&__status);
302 return ((__status | __mxcsr) & __excepts);
303 }
304
305 __fenv_static inline int
fesetround(int __round)306 fesetround(int __round)
307 {
308 __uint32_t __mxcsr;
309 __uint16_t __control;
310
311 if (__round & ~_ROUND_MASK)
312 return (-1);
313
314 __fnstcw(&__control);
315 __control &= ~_ROUND_MASK;
316 __control |= __round;
317 __fldcw(&__control);
318
319 __stmxcsr(&__mxcsr);
320 __mxcsr &= ~(_ROUND_MASK << _SSE_ROUND_SHIFT);
321 __mxcsr |= __round << _SSE_ROUND_SHIFT;
322 __ldmxcsr(&__mxcsr);
323
324 return (0);
325 }
326
327 __fenv_static inline int
fesetenv(const fenv_t * __envp)328 fesetenv(const fenv_t *__envp)
329 {
330
331 /*
332 * XXX Using fldenvx() instead of fldenv() tells the compiler that this
333 * instruction clobbers the i387 register stack. This happens because
334 * we restore the tag word from the saved environment. Normally, this
335 * would happen anyway and we wouldn't care, because the ABI allows
336 * function calls to clobber the i387 regs. However, fesetenv() is
337 * inlined, so we need to be more careful.
338 */
339 __fldenvx(__envp->__x87);
340 __ldmxcsr(&__envp->__mxcsr);
341 return (0);
342 }
343
344 #endif /* __i386__ */
345
346 __END_DECLS
347
348 #endif /* !_FENV_H_ */
349