1 /*-
2 * Copyright (c) 2004-2005 David Schultz <das@FreeBSD.ORG>
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 * $FreeBSD: src/lib/msun/amd64/fenv.h,v 1.8 2011/10/10 15:43:09 das Exp $
27 */
28
29 #ifndef _FENV_H_
30 #define _FENV_H_
31
32 #include <openlibm_defs.h>
33 #include "cdefs-compat.h"
34 #include "types-compat.h"
35
36 #ifndef __fenv_static
37 #define __fenv_static static
38 #endif
39
40 typedef struct {
41 struct {
42 uint32_t __control;
43 uint32_t __status;
44 uint32_t __tag;
45 char __other[16];
46 } __x87;
47 uint32_t __mxcsr;
48 } fenv_t;
49
50 typedef uint16_t fexcept_t;
51
52 /* Exception flags */
53 #define FE_INVALID 0x01
54 #define FE_DENORMAL 0x02
55 #define FE_DIVBYZERO 0x04
56 #define FE_OVERFLOW 0x08
57 #define FE_UNDERFLOW 0x10
58 #define FE_INEXACT 0x20
59 #define FE_ALL_EXCEPT (FE_DIVBYZERO | FE_DENORMAL | FE_INEXACT | \
60 FE_INVALID | FE_OVERFLOW | FE_UNDERFLOW)
61
62 /* Rounding modes */
63 #define FE_TONEAREST 0x0000
64 #define FE_DOWNWARD 0x0400
65 #define FE_UPWARD 0x0800
66 #define FE_TOWARDZERO 0x0c00
67 #define _ROUND_MASK (FE_TONEAREST | FE_DOWNWARD | \
68 FE_UPWARD | FE_TOWARDZERO)
69
70 /*
71 * As compared to the x87 control word, the SSE unit's control word
72 * has the rounding control bits offset by 3 and the exception mask
73 * bits offset by 7.
74 */
75 #define _SSE_ROUND_SHIFT 3
76 #define _SSE_EMASK_SHIFT 7
77
78 __BEGIN_DECLS
79
80 /* Default floating-point environment */
81 extern const fenv_t __fe_dfl_env;
82 #define FE_DFL_ENV (&__fe_dfl_env)
83
84 #define __fldcw(__cw) __asm __volatile("fldcw %0" : : "m" (__cw))
85 #define __fldenv(__env) __asm __volatile("fldenv %0" : : "m" (__env))
86 #define __fldenvx(__env) __asm __volatile("fldenv %0" : : "m" (__env) \
87 : "st", "st(1)", "st(2)", "st(3)", "st(4)", \
88 "st(5)", "st(6)", "st(7)")
89 #define __fnclex() __asm __volatile("fnclex")
90 #define __fnstenv(__env) __asm __volatile("fnstenv %0" : "=m" (*(__env)))
91 #define __fnstcw(__cw) __asm __volatile("fnstcw %0" : "=m" (*(__cw)))
92 #define __fnstsw(__sw) __asm __volatile("fnstsw %0" : "=am" (*(__sw)))
93 #define __fwait() __asm __volatile("fwait")
94 #define __ldmxcsr(__csr) __asm __volatile("ldmxcsr %0" : : "m" (__csr))
95 #define __stmxcsr(__csr) __asm __volatile("stmxcsr %0" : "=m" (*(__csr)))
96
97 __fenv_static __attribute__((always_inline)) inline int
feclearexcept(int __excepts)98 feclearexcept(int __excepts)
99 {
100 fenv_t __env;
101
102 if (__excepts == FE_ALL_EXCEPT) {
103 __fnclex();
104 } else {
105 __fnstenv(&__env.__x87);
106 __env.__x87.__status &= ~__excepts;
107 __fldenv(__env.__x87);
108 }
109 __stmxcsr(&__env.__mxcsr);
110 __env.__mxcsr &= ~__excepts;
111 __ldmxcsr(__env.__mxcsr);
112 return (0);
113 }
114
115 __fenv_static inline int
fegetexceptflag(fexcept_t * __flagp,int __excepts)116 fegetexceptflag(fexcept_t *__flagp, int __excepts)
117 {
118 uint32_t __mxcsr;
119 uint16_t __status;
120
121 __stmxcsr(&__mxcsr);
122 __fnstsw(&__status);
123 *__flagp = (__mxcsr | __status) & __excepts;
124 return (0);
125 }
126
127 OLM_DLLEXPORT int fesetexceptflag(const fexcept_t *__flagp, int __excepts);
128 OLM_DLLEXPORT int feraiseexcept(int __excepts);
129
130 __fenv_static __attribute__((always_inline)) inline int
fetestexcept(int __excepts)131 fetestexcept(int __excepts)
132 {
133 uint32_t __mxcsr;
134 uint16_t __status;
135
136 __stmxcsr(&__mxcsr);
137 __fnstsw(&__status);
138 return ((__status | __mxcsr) & __excepts);
139 }
140
141 __fenv_static inline int
fegetround(void)142 fegetround(void)
143 {
144 uint16_t __control;
145
146 /*
147 * We assume that the x87 and the SSE unit agree on the
148 * rounding mode. Reading the control word on the x87 turns
149 * out to be about 5 times faster than reading it on the SSE
150 * unit on an Opteron 244.
151 */
152 __fnstcw(&__control);
153 return (__control & _ROUND_MASK);
154 }
155
156 __fenv_static inline int
fesetround(int __round)157 fesetround(int __round)
158 {
159 uint32_t __mxcsr;
160 uint16_t __control;
161
162 if (__round & ~_ROUND_MASK)
163 return (-1);
164
165 __fnstcw(&__control);
166 __control &= ~_ROUND_MASK;
167 __control |= __round;
168 __fldcw(__control);
169
170 __stmxcsr(&__mxcsr);
171 __mxcsr &= ~(_ROUND_MASK << _SSE_ROUND_SHIFT);
172 __mxcsr |= __round << _SSE_ROUND_SHIFT;
173 __ldmxcsr(__mxcsr);
174
175 return (0);
176 }
177
178 OLM_DLLEXPORT int fegetenv(fenv_t *__envp);
179 OLM_DLLEXPORT int feholdexcept(fenv_t *__envp);
180
181 __fenv_static inline int
fesetenv(const fenv_t * __envp)182 fesetenv(const fenv_t *__envp)
183 {
184
185 /*
186 * XXX Using fldenvx() instead of fldenv() tells the compiler that this
187 * instruction clobbers the i387 register stack. This happens because
188 * we restore the tag word from the saved environment. Normally, this
189 * would happen anyway and we wouldn't care, because the ABI allows
190 * function calls to clobber the i387 regs. However, fesetenv() is
191 * inlined, so we need to be more careful.
192 */
193 __fldenvx(__envp->__x87);
194 __ldmxcsr(__envp->__mxcsr);
195 return (0);
196 }
197
198 OLM_DLLEXPORT int feupdateenv(const fenv_t *__envp);
199
200 #if __BSD_VISIBLE
201
202 OLM_DLLEXPORT int feenableexcept(int __mask);
203 OLM_DLLEXPORT int fedisableexcept(int __mask);
204
205 /* We currently provide no external definition of fegetexcept(). */
206 static inline int
fegetexcept(void)207 fegetexcept(void)
208 {
209 uint16_t __control;
210
211 /*
212 * We assume that the masks for the x87 and the SSE unit are
213 * the same.
214 */
215 __fnstcw(&__control);
216 return (~__control & FE_ALL_EXCEPT);
217 }
218
219 #endif /* __BSD_VISIBLE */
220
221 __END_DECLS
222
223 #endif /* !_FENV_H_ */
224