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 * from: @(#) ieeefp.h 1.0 (Berkeley) 9/23/93 37 * $FreeBSD$ 38 */ 39 40 #ifndef _MACHINE_IEEEFP_H_ 41 #define _MACHINE_IEEEFP_H_ 42 43 /* Deprecated historical FPU control interface */ 44 45 #include <x86/x86_ieeefp.h> 46 47 /* 48 * IEEE floating point type, constant and function definitions. 49 * XXX: {FP,SSE}*FLD and {FP,SSE}*OFF are undocumented pollution. 50 */ 51 52 /* 53 * SSE mxcsr register bit-field masks. 54 */ 55 #define SSE_STKY_FLD 0x3f /* exception flags */ 56 #define SSE_DAZ_FLD 0x40 /* Denormals are zero */ 57 #define SSE_MSKS_FLD 0x1f80 /* exception masks field */ 58 #define SSE_RND_FLD 0x6000 /* rounding control */ 59 #define SSE_FZ_FLD 0x8000 /* flush to zero on underflow */ 60 61 /* 62 * SSE mxcsr register bit-field offsets (shift counts). 63 */ 64 #define SSE_STKY_OFF 0 /* exception flags offset */ 65 #define SSE_DAZ_OFF 6 /* DAZ exception mask offset */ 66 #define SSE_MSKS_OFF 7 /* other exception masks offset */ 67 #define SSE_RND_OFF 13 /* rounding control offset */ 68 #define SSE_FZ_OFF 15 /* flush to zero offset */ 69 70 #ifdef __GNUCLIKE_ASM 71 72 /* 73 * General notes about conflicting SSE vs FP status bits. 74 * This code assumes that software will not fiddle with the control 75 * bits of the SSE and x87 in such a way to get them out of sync and 76 * still expect this to work. Break this at your peril. 77 * Because I based this on the i386 port, the x87 state is used for 78 * the fpget*() functions, and is shadowed into the SSE state for 79 * the fpset*() functions. For dual source fpget*() functions, I 80 * merge the two together. I think. 81 */ 82 83 static __inline fp_rnd_t 84 __fpgetround(void) 85 { 86 unsigned short _cw; 87 88 __fnstcw(&_cw); 89 return ((fp_rnd_t)((_cw & FP_RND_FLD) >> FP_RND_OFF)); 90 } 91 92 static __inline fp_rnd_t 93 __fpsetround(fp_rnd_t _m) 94 { 95 fp_rnd_t _p; 96 unsigned _mxcsr; 97 unsigned short _cw, _newcw; 98 99 __fnstcw(&_cw); 100 _p = (fp_rnd_t)((_cw & FP_RND_FLD) >> FP_RND_OFF); 101 _newcw = _cw & ~FP_RND_FLD; 102 _newcw |= (_m << FP_RND_OFF) & FP_RND_FLD; 103 __fnldcw(_cw, _newcw); 104 __stmxcsr(&_mxcsr); 105 _mxcsr &= ~SSE_RND_FLD; 106 _mxcsr |= (_m << SSE_RND_OFF) & SSE_RND_FLD; 107 __ldmxcsr(&_mxcsr); 108 return (_p); 109 } 110 111 /* 112 * Get or set the rounding precision for x87 arithmetic operations. 113 * There is no equivalent SSE mode or control. 114 */ 115 116 static __inline fp_prec_t 117 __fpgetprec(void) 118 { 119 unsigned short _cw; 120 121 __fnstcw(&_cw); 122 return ((fp_prec_t)((_cw & FP_PRC_FLD) >> FP_PRC_OFF)); 123 } 124 125 static __inline fp_prec_t 126 __fpsetprec(fp_prec_t _m) 127 { 128 fp_prec_t _p; 129 unsigned short _cw, _newcw; 130 131 __fnstcw(&_cw); 132 _p = (fp_prec_t)((_cw & FP_PRC_FLD) >> FP_PRC_OFF); 133 _newcw = _cw & ~FP_PRC_FLD; 134 _newcw |= (_m << FP_PRC_OFF) & FP_PRC_FLD; 135 __fnldcw(_cw, _newcw); 136 return (_p); 137 } 138 139 /* 140 * Get or set the exception mask. 141 * Note that the x87 mask bits are inverted by the API -- a mask bit of 1 142 * means disable for x87 and SSE, but for fp*mask() it means enable. 143 */ 144 145 static __inline fp_except_t 146 __fpgetmask(void) 147 { 148 unsigned short _cw; 149 150 __fnstcw(&_cw); 151 return ((~_cw & FP_MSKS_FLD) >> FP_MSKS_OFF); 152 } 153 154 static __inline fp_except_t 155 __fpsetmask(fp_except_t _m) 156 { 157 fp_except_t _p; 158 unsigned _mxcsr; 159 unsigned short _cw, _newcw; 160 161 __fnstcw(&_cw); 162 _p = (~_cw & FP_MSKS_FLD) >> FP_MSKS_OFF; 163 _newcw = _cw & ~FP_MSKS_FLD; 164 _newcw |= (~_m << FP_MSKS_OFF) & FP_MSKS_FLD; 165 __fnldcw(_cw, _newcw); 166 __stmxcsr(&_mxcsr); 167 /* XXX should we clear non-ieee SSE_DAZ_FLD and SSE_FZ_FLD ? */ 168 _mxcsr &= ~SSE_MSKS_FLD; 169 _mxcsr |= (~_m << SSE_MSKS_OFF) & SSE_MSKS_FLD; 170 __ldmxcsr(&_mxcsr); 171 return (_p); 172 } 173 174 static __inline fp_except_t 175 __fpgetsticky(void) 176 { 177 unsigned _ex, _mxcsr; 178 unsigned short _sw; 179 180 __fnstsw(&_sw); 181 _ex = (_sw & FP_STKY_FLD) >> FP_STKY_OFF; 182 __stmxcsr(&_mxcsr); 183 _ex |= (_mxcsr & SSE_STKY_FLD) >> SSE_STKY_OFF; 184 return ((fp_except_t)_ex); 185 } 186 187 #endif /* __GNUCLIKE_ASM */ 188 189 #if !defined(__IEEEFP_NOINLINES__) && defined(__GNUCLIKE_ASM) 190 191 #define fpgetmask() __fpgetmask() 192 #define fpgetprec() __fpgetprec() 193 #define fpgetround() __fpgetround() 194 #define fpgetsticky() __fpgetsticky() 195 #define fpsetmask(m) __fpsetmask(m) 196 #define fpsetprec(m) __fpsetprec(m) 197 #define fpsetround(m) __fpsetround(m) 198 199 #else /* !(!__IEEEFP_NOINLINES__ && __GNUCLIKE_ASM) */ 200 201 /* Augment the userland declarations. */ 202 __BEGIN_DECLS 203 extern fp_rnd_t fpgetround(void); 204 extern fp_rnd_t fpsetround(fp_rnd_t); 205 extern fp_except_t fpgetmask(void); 206 extern fp_except_t fpsetmask(fp_except_t); 207 extern fp_except_t fpgetsticky(void); 208 extern fp_except_t fpsetsticky(fp_except_t); 209 fp_prec_t fpgetprec(void); 210 fp_prec_t fpsetprec(fp_prec_t); 211 __END_DECLS 212 213 #endif /* !__IEEEFP_NOINLINES__ && __GNUCLIKE_ASM */ 214 215 #endif /* !_MACHINE_IEEEFP_H_ */ 216