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