xref: /freebsd/lib/msun/i387/fenv.c (revision 5e53a4f90f82c4345f277dd87cc9292f26e04a29)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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  * $FreeBSD$
29  */
30 
31 #include <sys/cdefs.h>
32 #include <sys/types.h>
33 #include <machine/npx.h>
34 
35 #define	__fenv_static
36 #include "fenv.h"
37 
38 #ifdef __GNUC_GNU_INLINE__
39 #error "This file must be compiled with C99 'inline' semantics"
40 #endif
41 
42 const fenv_t __fe_dfl_env = {
43 	__INITIAL_NPXCW__,
44 	0x0000,
45 	0x0000,
46 	0x1f80,
47 	0xffffffff,
48 	{ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
49 	  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xff, 0xff }
50 };
51 
52 enum __sse_support __has_sse =
53 #ifdef __SSE__
54 	__SSE_YES;
55 #else
56 	__SSE_UNK;
57 #endif
58 
59 #define	getfl(x)	__asm __volatile("pushfl\n\tpopl %0" : "=mr" (*(x)))
60 #define	setfl(x)	__asm __volatile("pushl %0\n\tpopfl" : : "g" (x))
61 #define	cpuid_dx(x)	__asm __volatile("pushl %%ebx\n\tmovl $1, %%eax\n\t"  \
62 					 "cpuid\n\tpopl %%ebx"		      \
63 					: "=d" (*(x)) : : "eax", "ecx")
64 
65 /*
66  * Test for SSE support on this processor.  We need to do this because
67  * we need to use ldmxcsr/stmxcsr to get correct results if any part
68  * of the program was compiled to use SSE floating-point, but we can't
69  * use SSE on older processors.
70  */
71 int
72 __test_sse(void)
73 {
74 	int flag, nflag;
75 	int dx_features;
76 
77 	/* Am I a 486? */
78 	getfl(&flag);
79 	nflag = flag ^ 0x200000;
80 	setfl(nflag);
81 	getfl(&nflag);
82 	if (flag != nflag) {
83 		/* Not a 486, so CPUID should work. */
84 		cpuid_dx(&dx_features);
85 		if (dx_features & 0x2000000) {
86 			__has_sse = __SSE_YES;
87 			return (1);
88 		}
89 	}
90 	__has_sse = __SSE_NO;
91 	return (0);
92 }
93 
94 extern inline int feclearexcept(int __excepts);
95 extern inline int fegetexceptflag(fexcept_t *__flagp, int __excepts);
96 
97 int
98 fesetexceptflag(const fexcept_t *flagp, int excepts)
99 {
100 	fenv_t env;
101 	__uint32_t mxcsr;
102 
103 	__fnstenv(&env);
104 	env.__status &= ~excepts;
105 	env.__status |= *flagp & excepts;
106 	__fldenv(env);
107 
108 	if (__HAS_SSE()) {
109 		__stmxcsr(&mxcsr);
110 		mxcsr &= ~excepts;
111 		mxcsr |= *flagp & excepts;
112 		__ldmxcsr(mxcsr);
113 	}
114 
115 	return (0);
116 }
117 
118 int
119 feraiseexcept(int excepts)
120 {
121 	fexcept_t ex = excepts;
122 
123 	fesetexceptflag(&ex, excepts);
124 	__fwait();
125 	return (0);
126 }
127 
128 extern inline int fetestexcept(int __excepts);
129 extern inline int fegetround(void);
130 extern inline int fesetround(int __round);
131 
132 int
133 fegetenv(fenv_t *envp)
134 {
135 	__uint32_t mxcsr;
136 
137 	__fnstenv(envp);
138 	/*
139 	 * fnstenv masks all exceptions, so we need to restore
140 	 * the old control word to avoid this side effect.
141 	 */
142 	__fldcw(envp->__control);
143 	if (__HAS_SSE()) {
144 		__stmxcsr(&mxcsr);
145 		__set_mxcsr(*envp, mxcsr);
146 	}
147 	return (0);
148 }
149 
150 int
151 feholdexcept(fenv_t *envp)
152 {
153 	__uint32_t mxcsr;
154 
155 	__fnstenv(envp);
156 	__fnclex();
157 	if (__HAS_SSE()) {
158 		__stmxcsr(&mxcsr);
159 		__set_mxcsr(*envp, mxcsr);
160 		mxcsr &= ~FE_ALL_EXCEPT;
161 		mxcsr |= FE_ALL_EXCEPT << _SSE_EMASK_SHIFT;
162 		__ldmxcsr(mxcsr);
163 	}
164 	return (0);
165 }
166 
167 extern inline int fesetenv(const fenv_t *__envp);
168 
169 int
170 feupdateenv(const fenv_t *envp)
171 {
172 	__uint32_t mxcsr;
173 	__uint16_t status;
174 
175 	__fnstsw(&status);
176 	if (__HAS_SSE())
177 		__stmxcsr(&mxcsr);
178 	else
179 		mxcsr = 0;
180 	fesetenv(envp);
181 	feraiseexcept((mxcsr | status) & FE_ALL_EXCEPT);
182 	return (0);
183 }
184 
185 int
186 __feenableexcept(int mask)
187 {
188 	__uint32_t mxcsr, omask;
189 	__uint16_t control;
190 
191 	mask &= FE_ALL_EXCEPT;
192 	__fnstcw(&control);
193 	if (__HAS_SSE())
194 		__stmxcsr(&mxcsr);
195 	else
196 		mxcsr = 0;
197 	omask = ~(control | mxcsr >> _SSE_EMASK_SHIFT) & FE_ALL_EXCEPT;
198 	control &= ~mask;
199 	__fldcw(control);
200 	if (__HAS_SSE()) {
201 		mxcsr &= ~(mask << _SSE_EMASK_SHIFT);
202 		__ldmxcsr(mxcsr);
203 	}
204 	return (omask);
205 }
206 
207 int
208 __fedisableexcept(int mask)
209 {
210 	__uint32_t mxcsr, omask;
211 	__uint16_t control;
212 
213 	mask &= FE_ALL_EXCEPT;
214 	__fnstcw(&control);
215 	if (__HAS_SSE())
216 		__stmxcsr(&mxcsr);
217 	else
218 		mxcsr = 0;
219 	omask = ~(control | mxcsr >> _SSE_EMASK_SHIFT) & FE_ALL_EXCEPT;
220 	control |= mask;
221 	__fldcw(control);
222 	if (__HAS_SSE()) {
223 		mxcsr |= mask << _SSE_EMASK_SHIFT;
224 		__ldmxcsr(mxcsr);
225 	}
226 	return (omask);
227 }
228 
229 __weak_reference(__feenableexcept, feenableexcept);
230 __weak_reference(__fedisableexcept, fedisableexcept);
231