1 /*- 2 * Copyright (c) 2003 Peter Wemm. 3 * Copyright (c) 1990 Andrew Moore, Talke Studio 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 3. Neither the name of the University nor the names of its contributors 15 * may be used to endorse or promote products derived from this software 16 * without specific prior written permission. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 * 30 * from: @(#) ieeefp.h 1.0 (Berkeley) 9/23/93 31 * $FreeBSD: src/sys/amd64/include/ieeefp.h,v 1.14 2005/04/12 23:12:00 jhb Exp $ 32 */ 33 34 /* 35 * IEEE floating point type and constant definitions. 36 */ 37 38 #ifndef _MACHINE_IEEEFP_H_ 39 #define _MACHINE_IEEEFP_H_ 40 41 #if !defined(_SYS_CDEFS_H) && !defined(_SYS_CDEFS_H_) && !defined(_CDEFS_H_) 42 #error this file needs sys/cdefs.h as a prerequisite 43 #endif 44 45 /* 46 * FP rounding modes 47 */ 48 typedef enum { 49 FP_RN=0, /* round to nearest */ 50 FP_RM, /* round down to minus infinity */ 51 FP_RP, /* round up to plus infinity */ 52 FP_RZ /* truncate */ 53 } fp_rnd_t; 54 55 /* 56 * FP precision modes 57 */ 58 typedef enum { 59 FP_PS=0, /* 24 bit (single-precision) */ 60 FP_PRS, /* reserved */ 61 FP_PD, /* 53 bit (double-precision) */ 62 FP_PE /* 64 bit (extended-precision) */ 63 } fp_prec_t; 64 65 #define fp_except_t int 66 67 /* 68 * FP exception masks 69 */ 70 #define FP_X_INV 0x01 /* invalid operation */ 71 #define FP_X_DNML 0x02 /* denormal */ 72 #define FP_X_DZ 0x04 /* zero divide */ 73 #define FP_X_OFL 0x08 /* overflow */ 74 #define FP_X_UFL 0x10 /* underflow */ 75 #define FP_X_IMP 0x20 /* (im)precision */ 76 #define FP_X_STK 0x40 /* stack fault */ 77 78 /* 79 * FP registers 80 */ 81 #define FP_MSKS_REG 0 /* exception masks */ 82 #define FP_PRC_REG 0 /* precision */ 83 #define FP_RND_REG 0 /* direction */ 84 #define FP_STKY_REG 1 /* sticky flags */ 85 86 /* 87 * FP register bit field masks 88 */ 89 #define FP_MSKS_FLD 0x3f /* exception masks field */ 90 #define FP_PRC_FLD 0x300 /* precision control field */ 91 #define FP_RND_FLD 0xc00 /* round control field */ 92 #define FP_STKY_FLD 0x3f /* sticky flags field */ 93 94 /* 95 * SSE mxcsr register bit field masks 96 */ 97 #define SSE_STKY_FLD 0x3f /* exception flags */ 98 #define SSE_DAZ_FLD 0x40 /* Denormals are zero */ 99 #define SSE_MSKS_FLD 0x1f80 /* exception masks field */ 100 #define SSE_RND_FLD 0x6000 /* rounding control */ 101 #define SSE_FZ_FLD 0x8000 /* flush to zero on underflow */ 102 103 /* 104 * FP register bit field offsets 105 */ 106 #define FP_MSKS_OFF 0 /* exception masks offset */ 107 #define FP_PRC_OFF 8 /* precision control offset */ 108 #define FP_RND_OFF 10 /* round control offset */ 109 #define FP_STKY_OFF 0 /* sticky flags offset */ 110 111 /* 112 * SSE mxcsr register bit field offsets 113 */ 114 #define SSE_STKY_OFF 0 /* exception flags offset */ 115 #define SSE_DAZ_OFF 6 /* DAZ exception mask offset */ 116 #define SSE_MSKS_OFF 7 /* other exception masks offset */ 117 #define SSE_RND_OFF 13 /* rounding control offset */ 118 #define SSE_FZ_OFF 15 /* flush to zero offset */ 119 120 #if (defined(__GNUCLIKE_ASM) && defined(__CC_SUPPORTS___INLINE__)) || defined(_WIN32) \ 121 && !defined(__cplusplus) 122 123 #define __fldenv(addr) __asm __volatile("fldenv %0" : : "m" (*(addr))) 124 #define __fnstenv(addr) __asm __volatile("fnstenv %0" : "=m" (*(addr))) 125 #define __fldcw(addr) __asm __volatile("fldcw %0" : : "m" (*(addr))) 126 #define __fnstcw(addr) __asm __volatile("fnstcw %0" : "=m" (*(addr))) 127 #define __fnstsw(addr) __asm __volatile("fnstsw %0" : "=m" (*(addr))) 128 #define __ldmxcsr(addr) __asm __volatile("ldmxcsr %0" : : "m" (*(addr))) 129 #define __stmxcsr(addr) __asm __volatile("stmxcsr %0" : "=m" (*(addr))) 130 131 /* 132 * General notes about conflicting SSE vs FP status bits. 133 * This code assumes that software will not fiddle with the control 134 * bits of the SSE and x87 in such a way to get them out of sync and 135 * still expect this to work. Break this at your peril. 136 * Because I based this on the i386 port, the x87 state is used for 137 * the fpget*() functions, and is shadowed into the SSE state for 138 * the fpset*() functions. For dual source fpget*() functions, I 139 * merge the two together. I think. 140 */ 141 142 /* Set rounding control */ 143 static __inline__ fp_rnd_t 144 __fpgetround(void) 145 { 146 unsigned short _cw; 147 148 __fnstcw(&_cw); 149 return ((_cw & FP_RND_FLD) >> FP_RND_OFF); 150 } 151 152 static __inline__ fp_rnd_t 153 __fpsetround(fp_rnd_t _m) 154 { 155 unsigned short _cw; 156 unsigned int _mxcsr; 157 fp_rnd_t _p; 158 159 __fnstcw(&_cw); 160 _p = (_cw & FP_RND_FLD) >> FP_RND_OFF; 161 _cw &= ~FP_RND_FLD; 162 _cw |= (_m << FP_RND_OFF) & FP_RND_FLD; 163 __fldcw(&_cw); 164 __stmxcsr(&_mxcsr); 165 _mxcsr &= ~SSE_RND_FLD; 166 _mxcsr |= (_m << SSE_RND_OFF) & SSE_RND_FLD; 167 __ldmxcsr(&_mxcsr); 168 return (_p); 169 } 170 171 /* 172 * Set precision for fadd/fsub/fsqrt etc x87 instructions 173 * There is no equivalent SSE mode or control. 174 */ 175 static __inline__ fp_prec_t 176 __fpgetprec(void) 177 { 178 unsigned short _cw; 179 180 __fnstcw(&_cw); 181 return ((_cw & FP_PRC_FLD) >> FP_PRC_OFF); 182 } 183 184 static __inline__ fp_prec_t 185 __fpsetprec(fp_rnd_t _m) 186 { 187 unsigned short _cw; 188 fp_prec_t _p; 189 190 __fnstcw(&_cw); 191 _p = (_cw & FP_PRC_FLD) >> FP_PRC_OFF; 192 _cw &= ~FP_PRC_FLD; 193 _cw |= (_m << FP_PRC_OFF) & FP_PRC_FLD; 194 __fldcw(&_cw); 195 return (_p); 196 } 197 198 /* 199 * Look at the exception masks 200 * Note that x87 masks are inverse of the fp*() functions 201 * API. ie: mask = 1 means disable for x87 and SSE, but 202 * for the fp*() api, mask = 1 means enabled. 203 */ 204 static __inline__ fp_except_t 205 __fpgetmask(void) 206 { 207 unsigned short _cw; 208 209 __fnstcw(&_cw); 210 return ((~_cw) & FP_MSKS_FLD); 211 } 212 213 static __inline__ fp_except_t 214 __fpsetmask(fp_except_t _m) 215 { 216 unsigned short _cw; 217 unsigned int _mxcsr; 218 fp_except_t _p; 219 220 __fnstcw(&_cw); 221 _p = (~_cw) & FP_MSKS_FLD; 222 _cw &= ~FP_MSKS_FLD; 223 _cw |= (~_m) & FP_MSKS_FLD; 224 __fldcw(&_cw); 225 __stmxcsr(&_mxcsr); 226 /* XXX should we clear non-ieee SSE_DAZ_FLD and SSE_FZ_FLD ? */ 227 _mxcsr &= ~SSE_MSKS_FLD; 228 _mxcsr |= ((~_m) << SSE_MSKS_OFF) & SSE_MSKS_FLD; 229 __ldmxcsr(&_mxcsr); 230 return (_p); 231 } 232 233 /* See which sticky exceptions are pending, and reset them */ 234 static __inline__ fp_except_t 235 __fpgetsticky(void) 236 { 237 unsigned short _sw; 238 unsigned int _mxcsr; 239 fp_except_t _ex; 240 241 __fnstsw(&_sw); 242 _ex = _sw & FP_STKY_FLD; 243 __stmxcsr(&_mxcsr); 244 _ex |= _mxcsr & SSE_STKY_FLD; 245 return (_ex); 246 } 247 248 #endif /* __GNUCLIKE_ASM && __CC_SUPPORTS___INLINE__ && !__cplusplus */ 249 250 #if !defined(__IEEEFP_NOINLINES__) && !defined(__cplusplus) \ 251 && defined(__GNUCLIKE_ASM) && defined(__CC_SUPPORTS___INLINE__) 252 253 #define fpgetround() __fpgetround() 254 #define fpsetround(_m) __fpsetround(_m) 255 #define fpgetprec() __fpgetprec() 256 #define fpsetprec(_m) __fpsetprec(_m) 257 #define fpgetmask() __fpgetmask() 258 #define fpsetmask(_m) __fpsetmask(_m) 259 #define fpgetsticky() __fpgetsticky() 260 261 /* Suppress prototypes in the MI header. */ 262 #define _IEEEFP_INLINED_ 1 263 264 #else /* !__IEEEFP_NOINLINES__ && !__cplusplus && __GNUCLIKE_ASM 265 && __CC_SUPPORTS___INLINE__ */ 266 267 /* Augment the userland declarations */ 268 __BEGIN_DECLS 269 extern fp_prec_t fpgetprec(void); 270 extern fp_prec_t fpsetprec(fp_prec_t); 271 __END_DECLS 272 273 #endif /* !__IEEEFP_NOINLINES__ && !__cplusplus && __GNUCLIKE_ASM 274 && __CC_SUPPORTS___INLINE__ */ 275 276 #endif /* !_MACHINE_IEEEFP_H_ */ 277