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 /* 71 * General notes about conflicting SSE vs FP status bits. 72 * This code assumes that software will not fiddle with the control 73 * bits of the SSE and x87 in such a way to get them out of sync and 74 * still expect this to work. Break this at your peril. 75 * Because I based this on the i386 port, the x87 state is used for 76 * the fpget*() functions, and is shadowed into the SSE state for 77 * the fpset*() functions. For dual source fpget*() functions, I 78 * merge the two together. I think. 79 */ 80 81 static __inline fp_rnd_t 82 __fpgetround(void) 83 { 84 unsigned short _cw; 85 86 __fnstcw(&_cw); 87 return ((fp_rnd_t)((_cw & FP_RND_FLD) >> FP_RND_OFF)); 88 } 89 90 static __inline fp_rnd_t 91 __fpsetround(fp_rnd_t _m) 92 { 93 fp_rnd_t _p; 94 unsigned _mxcsr; 95 unsigned short _cw, _newcw; 96 97 __fnstcw(&_cw); 98 _p = (fp_rnd_t)((_cw & FP_RND_FLD) >> FP_RND_OFF); 99 _newcw = _cw & ~FP_RND_FLD; 100 _newcw |= (_m << FP_RND_OFF) & FP_RND_FLD; 101 __fnldcw(_cw, _newcw); 102 __stmxcsr(&_mxcsr); 103 _mxcsr &= ~SSE_RND_FLD; 104 _mxcsr |= (_m << SSE_RND_OFF) & SSE_RND_FLD; 105 __ldmxcsr(&_mxcsr); 106 return (_p); 107 } 108 109 /* 110 * Get or set the rounding precision for x87 arithmetic operations. 111 * There is no equivalent SSE mode or control. 112 */ 113 114 static __inline fp_prec_t 115 __fpgetprec(void) 116 { 117 unsigned short _cw; 118 119 __fnstcw(&_cw); 120 return ((fp_prec_t)((_cw & FP_PRC_FLD) >> FP_PRC_OFF)); 121 } 122 123 static __inline fp_prec_t 124 __fpsetprec(fp_prec_t _m) 125 { 126 fp_prec_t _p; 127 unsigned short _cw, _newcw; 128 129 __fnstcw(&_cw); 130 _p = (fp_prec_t)((_cw & FP_PRC_FLD) >> FP_PRC_OFF); 131 _newcw = _cw & ~FP_PRC_FLD; 132 _newcw |= (_m << FP_PRC_OFF) & FP_PRC_FLD; 133 __fnldcw(_cw, _newcw); 134 return (_p); 135 } 136 137 /* 138 * Get or set the exception mask. 139 * Note that the x87 mask bits are inverted by the API -- a mask bit of 1 140 * means disable for x87 and SSE, but for fp*mask() it means enable. 141 */ 142 143 static __inline fp_except_t 144 __fpgetmask(void) 145 { 146 unsigned short _cw; 147 148 __fnstcw(&_cw); 149 return ((~_cw & FP_MSKS_FLD) >> FP_MSKS_OFF); 150 } 151 152 static __inline fp_except_t 153 __fpsetmask(fp_except_t _m) 154 { 155 fp_except_t _p; 156 unsigned _mxcsr; 157 unsigned short _cw, _newcw; 158 159 __fnstcw(&_cw); 160 _p = (~_cw & FP_MSKS_FLD) >> FP_MSKS_OFF; 161 _newcw = _cw & ~FP_MSKS_FLD; 162 _newcw |= (~_m << FP_MSKS_OFF) & FP_MSKS_FLD; 163 __fnldcw(_cw, _newcw); 164 __stmxcsr(&_mxcsr); 165 /* XXX should we clear non-ieee SSE_DAZ_FLD and SSE_FZ_FLD ? */ 166 _mxcsr &= ~SSE_MSKS_FLD; 167 _mxcsr |= (~_m << SSE_MSKS_OFF) & SSE_MSKS_FLD; 168 __ldmxcsr(&_mxcsr); 169 return (_p); 170 } 171 172 static __inline fp_except_t 173 __fpgetsticky(void) 174 { 175 unsigned _ex, _mxcsr; 176 unsigned short _sw; 177 178 __fnstsw(&_sw); 179 _ex = (_sw & FP_STKY_FLD) >> FP_STKY_OFF; 180 __stmxcsr(&_mxcsr); 181 _ex |= (_mxcsr & SSE_STKY_FLD) >> SSE_STKY_OFF; 182 return ((fp_except_t)_ex); 183 } 184 185 #if !defined(__IEEEFP_NOINLINES__) 186 187 #define fpgetmask() __fpgetmask() 188 #define fpgetprec() __fpgetprec() 189 #define fpgetround() __fpgetround() 190 #define fpgetsticky() __fpgetsticky() 191 #define fpsetmask(m) __fpsetmask(m) 192 #define fpsetprec(m) __fpsetprec(m) 193 #define fpsetround(m) __fpsetround(m) 194 195 #else /* __IEEEFP_NOINLINES__ */ 196 197 /* Augment the userland declarations. */ 198 __BEGIN_DECLS 199 extern fp_rnd_t fpgetround(void); 200 extern fp_rnd_t fpsetround(fp_rnd_t); 201 extern fp_except_t fpgetmask(void); 202 extern fp_except_t fpsetmask(fp_except_t); 203 extern fp_except_t fpgetsticky(void); 204 extern fp_except_t fpsetsticky(fp_except_t); 205 fp_prec_t fpgetprec(void); 206 fp_prec_t fpsetprec(fp_prec_t); 207 __END_DECLS 208 209 #endif /* !__IEEEFP_NOINLINES__ */ 210 211 #endif /* !_MACHINE_IEEEFP_H_ */ 212