1 2 /* 3 =============================================================================== 4 5 This C header file is part of the SoftFloat IEC/IEEE Floating-point 6 Arithmetic Package, Release 2. 7 8 Written by John R. Hauser. This work was made possible in part by the 9 International Computer Science Institute, located at Suite 600, 1947 Center 10 Street, Berkeley, California 94704. Funding was partially provided by the 11 National Science Foundation under grant MIP-9311980. The original version 12 of this code was written as part of a project to build a fixed-point vector 13 processor in collaboration with the University of California at Berkeley, 14 overseen by Profs. Nelson Morgan and John Wawrzynek. More information 15 is available through the Web page `http://HTTP.CS.Berkeley.EDU/~jhauser/ 16 arithmetic/softfloat.html'. 17 18 THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort 19 has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT 20 TIMES RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO 21 PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY 22 AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE. 23 24 Derivative works are acceptable, even for commercial purposes, so long as 25 (1) they include prominent notice that the work is derivative, and (2) they 26 include prominent notice akin to these three paragraphs for those parts of 27 this code that are retained. 28 29 =============================================================================== 30 */ 31 32 #ifndef __SOFTFLOAT_H__ 33 #define __SOFTFLOAT_H__ 34 35 #include <linux/config.h> 36 37 /* 38 ------------------------------------------------------------------------------- 39 The macro `FLOATX80' must be defined to enable the extended double-precision 40 floating-point format `floatx80'. If this macro is not defined, the 41 `floatx80' type will not be defined, and none of the functions that either 42 input or output the `floatx80' type will be defined. 43 ------------------------------------------------------------------------------- 44 */ 45 #ifdef CONFIG_FPE_NWFPE_XP 46 #define FLOATX80 47 #endif 48 49 /* 50 ------------------------------------------------------------------------------- 51 Software IEC/IEEE floating-point types. 52 ------------------------------------------------------------------------------- 53 */ 54 typedef u32 float32; 55 typedef u64 float64; 56 typedef struct { 57 #ifdef __ARMEB__ 58 u16 __padding; 59 u16 high; 60 #else 61 u16 high; 62 u16 __padding; 63 #endif 64 u64 low; 65 } floatx80; 66 67 /* 68 ------------------------------------------------------------------------------- 69 Software IEC/IEEE floating-point underflow tininess-detection mode. 70 ------------------------------------------------------------------------------- 71 */ 72 extern signed char float_detect_tininess; 73 enum { 74 float_tininess_after_rounding = 0, 75 float_tininess_before_rounding = 1 76 }; 77 78 /* 79 ------------------------------------------------------------------------------- 80 Software IEC/IEEE floating-point rounding mode. 81 ------------------------------------------------------------------------------- 82 */ 83 //extern int8 float_rounding_mode; 84 enum { 85 float_round_nearest_even = 0, 86 float_round_to_zero = 1, 87 float_round_down = 2, 88 float_round_up = 3 89 }; 90 91 /* 92 ------------------------------------------------------------------------------- 93 Software IEC/IEEE floating-point exception flags. 94 ------------------------------------------------------------------------------- 95 enum { 96 float_flag_inexact = 1, 97 float_flag_underflow = 2, 98 float_flag_overflow = 4, 99 float_flag_divbyzero = 8, 100 float_flag_invalid = 16 101 }; 102 103 ScottB: November 4, 1998 104 Changed the enumeration to match the bit order in the FPA11. 105 */ 106 107 enum { 108 float_flag_invalid = 1, 109 float_flag_divbyzero = 2, 110 float_flag_overflow = 4, 111 float_flag_underflow = 8, 112 float_flag_inexact = 16 113 }; 114 115 /* 116 ------------------------------------------------------------------------------- 117 Routine to raise any or all of the software IEC/IEEE floating-point 118 exception flags. 119 ------------------------------------------------------------------------------- 120 */ 121 void float_raise( signed char ); 122 123 /* 124 ------------------------------------------------------------------------------- 125 Software IEC/IEEE integer-to-floating-point conversion routines. 126 ------------------------------------------------------------------------------- 127 */ 128 float32 int32_to_float32( struct roundingData *, signed int ); 129 float64 int32_to_float64( signed int ); 130 #ifdef FLOATX80 131 floatx80 int32_to_floatx80( signed int ); 132 #endif 133 134 /* 135 ------------------------------------------------------------------------------- 136 Software IEC/IEEE single-precision conversion routines. 137 ------------------------------------------------------------------------------- 138 */ 139 signed int float32_to_int32( struct roundingData *, float32 ); 140 signed int float32_to_int32_round_to_zero( float32 ); 141 float64 float32_to_float64( float32 ); 142 #ifdef FLOATX80 143 floatx80 float32_to_floatx80( float32 ); 144 #endif 145 146 /* 147 ------------------------------------------------------------------------------- 148 Software IEC/IEEE single-precision operations. 149 ------------------------------------------------------------------------------- 150 */ 151 float32 float32_round_to_int( struct roundingData*, float32 ); 152 float32 float32_add( struct roundingData *, float32, float32 ); 153 float32 float32_sub( struct roundingData *, float32, float32 ); 154 float32 float32_mul( struct roundingData *, float32, float32 ); 155 float32 float32_div( struct roundingData *, float32, float32 ); 156 float32 float32_rem( struct roundingData *, float32, float32 ); 157 float32 float32_sqrt( struct roundingData*, float32 ); 158 char float32_eq( float32, float32 ); 159 char float32_le( float32, float32 ); 160 char float32_lt( float32, float32 ); 161 char float32_eq_signaling( float32, float32 ); 162 char float32_le_quiet( float32, float32 ); 163 char float32_lt_quiet( float32, float32 ); 164 char float32_is_signaling_nan( float32 ); 165 166 /* 167 ------------------------------------------------------------------------------- 168 Software IEC/IEEE double-precision conversion routines. 169 ------------------------------------------------------------------------------- 170 */ 171 signed int float64_to_int32( struct roundingData *, float64 ); 172 signed int float64_to_int32_round_to_zero( float64 ); 173 float32 float64_to_float32( struct roundingData *, float64 ); 174 #ifdef FLOATX80 175 floatx80 float64_to_floatx80( float64 ); 176 #endif 177 178 /* 179 ------------------------------------------------------------------------------- 180 Software IEC/IEEE double-precision operations. 181 ------------------------------------------------------------------------------- 182 */ 183 float64 float64_round_to_int( struct roundingData *, float64 ); 184 float64 float64_add( struct roundingData *, float64, float64 ); 185 float64 float64_sub( struct roundingData *, float64, float64 ); 186 float64 float64_mul( struct roundingData *, float64, float64 ); 187 float64 float64_div( struct roundingData *, float64, float64 ); 188 float64 float64_rem( struct roundingData *, float64, float64 ); 189 float64 float64_sqrt( struct roundingData *, float64 ); 190 char float64_eq( float64, float64 ); 191 char float64_le( float64, float64 ); 192 char float64_lt( float64, float64 ); 193 char float64_eq_signaling( float64, float64 ); 194 char float64_le_quiet( float64, float64 ); 195 char float64_lt_quiet( float64, float64 ); 196 char float64_is_signaling_nan( float64 ); 197 198 #ifdef FLOATX80 199 200 /* 201 ------------------------------------------------------------------------------- 202 Software IEC/IEEE extended double-precision conversion routines. 203 ------------------------------------------------------------------------------- 204 */ 205 signed int floatx80_to_int32( struct roundingData *, floatx80 ); 206 signed int floatx80_to_int32_round_to_zero( floatx80 ); 207 float32 floatx80_to_float32( struct roundingData *, floatx80 ); 208 float64 floatx80_to_float64( struct roundingData *, floatx80 ); 209 210 /* 211 ------------------------------------------------------------------------------- 212 Software IEC/IEEE extended double-precision operations. 213 ------------------------------------------------------------------------------- 214 */ 215 floatx80 floatx80_round_to_int( struct roundingData *, floatx80 ); 216 floatx80 floatx80_add( struct roundingData *, floatx80, floatx80 ); 217 floatx80 floatx80_sub( struct roundingData *, floatx80, floatx80 ); 218 floatx80 floatx80_mul( struct roundingData *, floatx80, floatx80 ); 219 floatx80 floatx80_div( struct roundingData *, floatx80, floatx80 ); 220 floatx80 floatx80_rem( struct roundingData *, floatx80, floatx80 ); 221 floatx80 floatx80_sqrt( struct roundingData *, floatx80 ); 222 char floatx80_eq( floatx80, floatx80 ); 223 char floatx80_le( floatx80, floatx80 ); 224 char floatx80_lt( floatx80, floatx80 ); 225 char floatx80_eq_signaling( floatx80, floatx80 ); 226 char floatx80_le_quiet( floatx80, floatx80 ); 227 char floatx80_lt_quiet( floatx80, floatx80 ); 228 char floatx80_is_signaling_nan( floatx80 ); 229 230 #endif 231 232 static inline flag extractFloat32Sign(float32 a) 233 { 234 return a >> 31; 235 } 236 237 static inline flag float32_eq_nocheck(float32 a, float32 b) 238 { 239 return (a == b) || ((bits32) ((a | b) << 1) == 0); 240 } 241 242 static inline flag float32_lt_nocheck(float32 a, float32 b) 243 { 244 flag aSign, bSign; 245 246 aSign = extractFloat32Sign(a); 247 bSign = extractFloat32Sign(b); 248 if (aSign != bSign) 249 return aSign && ((bits32) ((a | b) << 1) != 0); 250 return (a != b) && (aSign ^ (a < b)); 251 } 252 253 static inline flag extractFloat64Sign(float64 a) 254 { 255 return a >> 63; 256 } 257 258 static inline flag float64_eq_nocheck(float64 a, float64 b) 259 { 260 return (a == b) || ((bits64) ((a | b) << 1) == 0); 261 } 262 263 static inline flag float64_lt_nocheck(float64 a, float64 b) 264 { 265 flag aSign, bSign; 266 267 aSign = extractFloat64Sign(a); 268 bSign = extractFloat64Sign(b); 269 if (aSign != bSign) 270 return aSign && ((bits64) ((a | b) << 1) != 0); 271 return (a != b) && (aSign ^ (a < b)); 272 } 273 274 extern flag float32_is_nan( float32 a ); 275 extern flag float64_is_nan( float64 a ); 276 277 #endif 278