1 /*-
2 * SPDX-License-Identifier: BSD-4-Clause
3 *
4 * Copyright (c) 2003 Peter Wemm.
5 * Copyright (c) 1990 Andrew Moore, Talke Studio
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 */
36
37 #ifndef _MACHINE_IEEEFP_H_
38 #define _MACHINE_IEEEFP_H_
39
40 /* Deprecated historical FPU control interface */
41
42 #include <x86/x86_ieeefp.h>
43
44 /*
45 * IEEE floating point type, constant and function definitions.
46 * XXX: {FP,SSE}*FLD and {FP,SSE}*OFF are undocumented pollution.
47 */
48
49 /*
50 * SSE mxcsr register bit-field masks.
51 */
52 #define SSE_STKY_FLD 0x3f /* exception flags */
53 #define SSE_DAZ_FLD 0x40 /* Denormals are zero */
54 #define SSE_MSKS_FLD 0x1f80 /* exception masks field */
55 #define SSE_RND_FLD 0x6000 /* rounding control */
56 #define SSE_FZ_FLD 0x8000 /* flush to zero on underflow */
57
58 /*
59 * SSE mxcsr register bit-field offsets (shift counts).
60 */
61 #define SSE_STKY_OFF 0 /* exception flags offset */
62 #define SSE_DAZ_OFF 6 /* DAZ exception mask offset */
63 #define SSE_MSKS_OFF 7 /* other exception masks offset */
64 #define SSE_RND_OFF 13 /* rounding control offset */
65 #define SSE_FZ_OFF 15 /* flush to zero offset */
66
67 /*
68 * General notes about conflicting SSE vs FP status bits.
69 * This code assumes that software will not fiddle with the control
70 * bits of the SSE and x87 in such a way to get them out of sync and
71 * still expect this to work. Break this at your peril.
72 * Because I based this on the i386 port, the x87 state is used for
73 * the fpget*() functions, and is shadowed into the SSE state for
74 * the fpset*() functions. For dual source fpget*() functions, I
75 * merge the two together. I think.
76 */
77
78 static __inline fp_rnd_t
__fpgetround(void)79 __fpgetround(void)
80 {
81 unsigned short _cw;
82
83 __fnstcw(&_cw);
84 return ((fp_rnd_t)((_cw & FP_RND_FLD) >> FP_RND_OFF));
85 }
86
87 static __inline fp_rnd_t
__fpsetround(fp_rnd_t _m)88 __fpsetround(fp_rnd_t _m)
89 {
90 fp_rnd_t _p;
91 unsigned _mxcsr;
92 unsigned short _cw, _newcw;
93
94 __fnstcw(&_cw);
95 _p = (fp_rnd_t)((_cw & FP_RND_FLD) >> FP_RND_OFF);
96 _newcw = _cw & ~FP_RND_FLD;
97 _newcw |= (_m << FP_RND_OFF) & FP_RND_FLD;
98 __fnldcw(_cw, _newcw);
99 __stmxcsr(&_mxcsr);
100 _mxcsr &= ~SSE_RND_FLD;
101 _mxcsr |= (_m << SSE_RND_OFF) & SSE_RND_FLD;
102 __ldmxcsr(&_mxcsr);
103 return (_p);
104 }
105
106 /*
107 * Get or set the rounding precision for x87 arithmetic operations.
108 * There is no equivalent SSE mode or control.
109 */
110
111 static __inline fp_prec_t
__fpgetprec(void)112 __fpgetprec(void)
113 {
114 unsigned short _cw;
115
116 __fnstcw(&_cw);
117 return ((fp_prec_t)((_cw & FP_PRC_FLD) >> FP_PRC_OFF));
118 }
119
120 static __inline fp_prec_t
__fpsetprec(fp_prec_t _m)121 __fpsetprec(fp_prec_t _m)
122 {
123 fp_prec_t _p;
124 unsigned short _cw, _newcw;
125
126 __fnstcw(&_cw);
127 _p = (fp_prec_t)((_cw & FP_PRC_FLD) >> FP_PRC_OFF);
128 _newcw = _cw & ~FP_PRC_FLD;
129 _newcw |= (_m << FP_PRC_OFF) & FP_PRC_FLD;
130 __fnldcw(_cw, _newcw);
131 return (_p);
132 }
133
134 /*
135 * Get or set the exception mask.
136 * Note that the x87 mask bits are inverted by the API -- a mask bit of 1
137 * means disable for x87 and SSE, but for fp*mask() it means enable.
138 */
139
140 static __inline fp_except_t
__fpgetmask(void)141 __fpgetmask(void)
142 {
143 unsigned short _cw;
144
145 __fnstcw(&_cw);
146 return ((~_cw & FP_MSKS_FLD) >> FP_MSKS_OFF);
147 }
148
149 static __inline fp_except_t
__fpsetmask(fp_except_t _m)150 __fpsetmask(fp_except_t _m)
151 {
152 fp_except_t _p;
153 unsigned _mxcsr;
154 unsigned short _cw, _newcw;
155
156 __fnstcw(&_cw);
157 _p = (~_cw & FP_MSKS_FLD) >> FP_MSKS_OFF;
158 _newcw = _cw & ~FP_MSKS_FLD;
159 _newcw |= (~_m << FP_MSKS_OFF) & FP_MSKS_FLD;
160 __fnldcw(_cw, _newcw);
161 __stmxcsr(&_mxcsr);
162 /* XXX should we clear non-ieee SSE_DAZ_FLD and SSE_FZ_FLD ? */
163 _mxcsr &= ~SSE_MSKS_FLD;
164 _mxcsr |= (~_m << SSE_MSKS_OFF) & SSE_MSKS_FLD;
165 __ldmxcsr(&_mxcsr);
166 return (_p);
167 }
168
169 static __inline fp_except_t
__fpgetsticky(void)170 __fpgetsticky(void)
171 {
172 unsigned _ex, _mxcsr;
173 unsigned short _sw;
174
175 __fnstsw(&_sw);
176 _ex = (_sw & FP_STKY_FLD) >> FP_STKY_OFF;
177 __stmxcsr(&_mxcsr);
178 _ex |= (_mxcsr & SSE_STKY_FLD) >> SSE_STKY_OFF;
179 return ((fp_except_t)_ex);
180 }
181
182 #if !defined(__IEEEFP_NOINLINES__)
183
184 #define fpgetmask() __fpgetmask()
185 #define fpgetprec() __fpgetprec()
186 #define fpgetround() __fpgetround()
187 #define fpgetsticky() __fpgetsticky()
188 #define fpsetmask(m) __fpsetmask(m)
189 #define fpsetprec(m) __fpsetprec(m)
190 #define fpsetround(m) __fpsetround(m)
191
192 #else /* __IEEEFP_NOINLINES__ */
193
194 /* Augment the userland declarations. */
195 __BEGIN_DECLS
196 extern fp_rnd_t fpgetround(void);
197 extern fp_rnd_t fpsetround(fp_rnd_t);
198 extern fp_except_t fpgetmask(void);
199 extern fp_except_t fpsetmask(fp_except_t);
200 extern fp_except_t fpgetsticky(void);
201 extern fp_except_t fpsetsticky(fp_except_t);
202 fp_prec_t fpgetprec(void);
203 fp_prec_t fpsetprec(fp_prec_t);
204 __END_DECLS
205
206 #endif /* !__IEEEFP_NOINLINES__ */
207
208 #endif /* !_MACHINE_IEEEFP_H_ */
209