softfloat.h (10002B)
1 2/* 3=============================================================================== 4 5This C header file is part of the SoftFloat IEC/IEEE Floating-point 6Arithmetic Package, Release 2. 7 8Written by John R. Hauser. This work was made possible in part by the 9International Computer Science Institute, located at Suite 600, 1947 Center 10Street, Berkeley, California 94704. Funding was partially provided by the 11National Science Foundation under grant MIP-9311980. The original version 12of this code was written as part of a project to build a fixed-point vector 13processor in collaboration with the University of California at Berkeley, 14overseen by Profs. Nelson Morgan and John Wawrzynek. More information 15is available through the Web page 16http://www.jhauser.us/arithmetic/SoftFloat-2b/SoftFloat-source.txt 17 18THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort 19has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT 20TIMES RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO 21PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY 22AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE. 23 24Derivative works are acceptable, even for commercial purposes, so long as 25(1) they include prominent notice that the work is derivative, and (2) they 26include prominent notice akin to these three paragraphs for those parts of 27this code that are retained. 28 29=============================================================================== 30*/ 31 32#ifndef __SOFTFLOAT_H__ 33#define __SOFTFLOAT_H__ 34 35 36/* 37------------------------------------------------------------------------------- 38The macro `FLOATX80' must be defined to enable the extended double-precision 39floating-point format `floatx80'. If this macro is not defined, the 40`floatx80' type will not be defined, and none of the functions that either 41input or output the `floatx80' type will be defined. 42------------------------------------------------------------------------------- 43*/ 44#ifdef CONFIG_FPE_NWFPE_XP 45#define FLOATX80 46#endif 47 48/* 49------------------------------------------------------------------------------- 50Software IEC/IEEE floating-point types. 51------------------------------------------------------------------------------- 52*/ 53typedef u32 float32; 54typedef u64 float64; 55typedef struct { 56#ifdef __ARMEB__ 57 u16 __padding; 58 u16 high; 59#else 60 u16 high; 61 u16 __padding; 62#endif 63 u64 low; 64} __attribute__ ((packed,aligned(4))) floatx80; 65 66/* 67------------------------------------------------------------------------------- 68Software IEC/IEEE floating-point underflow tininess-detection mode. 69------------------------------------------------------------------------------- 70*/ 71extern signed char float_detect_tininess; 72enum { 73 float_tininess_after_rounding = 0, 74 float_tininess_before_rounding = 1 75}; 76 77/* 78------------------------------------------------------------------------------- 79Software IEC/IEEE floating-point rounding mode. 80------------------------------------------------------------------------------- 81*/ 82//extern int8 float_rounding_mode; 83enum { 84 float_round_nearest_even = 0, 85 float_round_to_zero = 1, 86 float_round_down = 2, 87 float_round_up = 3 88}; 89 90/* 91------------------------------------------------------------------------------- 92Software IEC/IEEE floating-point exception flags. 93------------------------------------------------------------------------------- 94enum { 95 float_flag_inexact = 1, 96 float_flag_underflow = 2, 97 float_flag_overflow = 4, 98 float_flag_divbyzero = 8, 99 float_flag_invalid = 16 100}; 101 102ScottB: November 4, 1998 103Changed the enumeration to match the bit order in the FPA11. 104*/ 105 106enum { 107 float_flag_invalid = 1, 108 float_flag_divbyzero = 2, 109 float_flag_overflow = 4, 110 float_flag_underflow = 8, 111 float_flag_inexact = 16 112}; 113 114/* 115------------------------------------------------------------------------------- 116Routine to raise any or all of the software IEC/IEEE floating-point 117exception flags. 118------------------------------------------------------------------------------- 119*/ 120void float_raise( signed char ); 121 122/* 123------------------------------------------------------------------------------- 124Software IEC/IEEE integer-to-floating-point conversion routines. 125------------------------------------------------------------------------------- 126*/ 127float32 int32_to_float32( struct roundingData *, signed int ); 128float64 int32_to_float64( signed int ); 129#ifdef FLOATX80 130floatx80 int32_to_floatx80( signed int ); 131#endif 132 133/* 134------------------------------------------------------------------------------- 135Software IEC/IEEE single-precision conversion routines. 136------------------------------------------------------------------------------- 137*/ 138signed int float32_to_int32( struct roundingData *, float32 ); 139signed int float32_to_int32_round_to_zero( float32 ); 140float64 float32_to_float64( float32 ); 141#ifdef FLOATX80 142floatx80 float32_to_floatx80( float32 ); 143#endif 144 145/* 146------------------------------------------------------------------------------- 147Software IEC/IEEE single-precision operations. 148------------------------------------------------------------------------------- 149*/ 150float32 float32_round_to_int( struct roundingData*, float32 ); 151float32 float32_add( struct roundingData *, float32, float32 ); 152float32 float32_sub( struct roundingData *, float32, float32 ); 153float32 float32_mul( struct roundingData *, float32, float32 ); 154float32 float32_div( struct roundingData *, float32, float32 ); 155float32 float32_rem( struct roundingData *, float32, float32 ); 156float32 float32_sqrt( struct roundingData*, float32 ); 157char float32_eq( float32, float32 ); 158char float32_le( float32, float32 ); 159char float32_lt( float32, float32 ); 160char float32_eq_signaling( float32, float32 ); 161char float32_le_quiet( float32, float32 ); 162char float32_lt_quiet( float32, float32 ); 163char float32_is_signaling_nan( float32 ); 164 165/* 166------------------------------------------------------------------------------- 167Software IEC/IEEE double-precision conversion routines. 168------------------------------------------------------------------------------- 169*/ 170signed int float64_to_int32( struct roundingData *, float64 ); 171signed int float64_to_int32_round_to_zero( float64 ); 172float32 float64_to_float32( struct roundingData *, float64 ); 173#ifdef FLOATX80 174floatx80 float64_to_floatx80( float64 ); 175#endif 176 177/* 178------------------------------------------------------------------------------- 179Software IEC/IEEE double-precision operations. 180------------------------------------------------------------------------------- 181*/ 182float64 float64_round_to_int( struct roundingData *, float64 ); 183float64 float64_add( struct roundingData *, float64, float64 ); 184float64 float64_sub( struct roundingData *, float64, float64 ); 185float64 float64_mul( struct roundingData *, float64, float64 ); 186float64 float64_div( struct roundingData *, float64, float64 ); 187float64 float64_rem( struct roundingData *, float64, float64 ); 188float64 float64_sqrt( struct roundingData *, float64 ); 189char float64_eq( float64, float64 ); 190char float64_le( float64, float64 ); 191char float64_lt( float64, float64 ); 192char float64_eq_signaling( float64, float64 ); 193char float64_le_quiet( float64, float64 ); 194char float64_lt_quiet( float64, float64 ); 195char float64_is_signaling_nan( float64 ); 196 197#ifdef FLOATX80 198 199/* 200------------------------------------------------------------------------------- 201Software IEC/IEEE extended double-precision conversion routines. 202------------------------------------------------------------------------------- 203*/ 204signed int floatx80_to_int32( struct roundingData *, floatx80 ); 205signed int floatx80_to_int32_round_to_zero( floatx80 ); 206float32 floatx80_to_float32( struct roundingData *, floatx80 ); 207float64 floatx80_to_float64( struct roundingData *, floatx80 ); 208 209/* 210------------------------------------------------------------------------------- 211Software IEC/IEEE extended double-precision operations. 212------------------------------------------------------------------------------- 213*/ 214floatx80 floatx80_round_to_int( struct roundingData *, floatx80 ); 215floatx80 floatx80_add( struct roundingData *, floatx80, floatx80 ); 216floatx80 floatx80_sub( struct roundingData *, floatx80, floatx80 ); 217floatx80 floatx80_mul( struct roundingData *, floatx80, floatx80 ); 218floatx80 floatx80_div( struct roundingData *, floatx80, floatx80 ); 219floatx80 floatx80_rem( struct roundingData *, floatx80, floatx80 ); 220floatx80 floatx80_sqrt( struct roundingData *, floatx80 ); 221char floatx80_eq( floatx80, floatx80 ); 222char floatx80_le( floatx80, floatx80 ); 223char floatx80_lt( floatx80, floatx80 ); 224char floatx80_eq_signaling( floatx80, floatx80 ); 225char floatx80_le_quiet( floatx80, floatx80 ); 226char floatx80_lt_quiet( floatx80, floatx80 ); 227char floatx80_is_signaling_nan( floatx80 ); 228 229extern flag floatx80_is_nan(floatx80); 230 231#endif 232 233static inline flag extractFloat32Sign(float32 a) 234{ 235 return a >> 31; 236} 237 238static inline flag float32_eq_nocheck(float32 a, float32 b) 239{ 240 return (a == b) || ((bits32) ((a | b) << 1) == 0); 241} 242 243static inline flag float32_lt_nocheck(float32 a, float32 b) 244{ 245 flag aSign, bSign; 246 247 aSign = extractFloat32Sign(a); 248 bSign = extractFloat32Sign(b); 249 if (aSign != bSign) 250 return aSign && ((bits32) ((a | b) << 1) != 0); 251 return (a != b) && (aSign ^ (a < b)); 252} 253 254static inline flag extractFloat64Sign(float64 a) 255{ 256 return a >> 63; 257} 258 259static inline flag float64_eq_nocheck(float64 a, float64 b) 260{ 261 return (a == b) || ((bits64) ((a | b) << 1) == 0); 262} 263 264static inline flag float64_lt_nocheck(float64 a, float64 b) 265{ 266 flag aSign, bSign; 267 268 aSign = extractFloat64Sign(a); 269 bSign = extractFloat64Sign(b); 270 if (aSign != bSign) 271 return aSign && ((bits64) ((a | b) << 1) != 0); 272 return (a != b) && (aSign ^ (a < b)); 273} 274 275extern flag float32_is_nan( float32 a ); 276extern flag float64_is_nan( float64 a ); 277 278extern int32 float64_to_uint32( struct roundingData *roundData, float64 a ); 279extern int32 float64_to_uint32_round_to_zero( float64 a ); 280 281#endif