1 /*============================================================================
3 This C header file is part of the SoftFloat IEC/IEEE Floating-point Arithmetic
6 Written by John R. Hauser. This work was made possible in part by the
7 International Computer Science Institute, located at Suite 600, 1947 Center
8 Street, Berkeley, California 94704. Funding was partially provided by the
9 National Science Foundation under grant MIP-9311980. The original version
10 of this code was written as part of a project to build a fixed-point vector
11 processor in collaboration with the University of California at Berkeley,
12 overseen by Profs. Nelson Morgan and John Wawrzynek. More information
13 is available through the Web page `http://www.cs.berkeley.edu/~jhauser/
14 arithmetic/SoftFloat.html'.
16 THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort has
17 been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT TIMES
18 RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO PERSONS
19 AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ALL LOSSES,
20 COSTS, OR OTHER PROBLEMS THEY INCUR DUE TO THE SOFTWARE, AND WHO FURTHERMORE
21 EFFECTIVELY INDEMNIFY JOHN HAUSER AND THE INTERNATIONAL COMPUTER SCIENCE
22 INSTITUTE (possibly via similar legal warning) AGAINST ALL LOSSES, COSTS, OR
23 OTHER PROBLEMS INCURRED BY THEIR CUSTOMERS AND CLIENTS DUE TO THE SOFTWARE.
25 Derivative works are acceptable, even for commercial purposes, so long as
26 (1) the source code for the derivative work includes prominent notice that
27 the work is derivative, and (2) the source code includes prominent notice with
28 these four paragraphs for those parts of this code that are retained.
30 =============================================================================*/
35 #if defined(CONFIG_SOLARIS) && defined(CONFIG_NEEDS_LIBSUNMATH)
42 /*----------------------------------------------------------------------------
43 | Each of the following `typedef's defines the most convenient type that holds
44 | integers of at least as many bits as specified. For example, `uint8' should
45 | be the most convenient type that can hold unsigned integers of as many as
46 | 8 bits. The `flag' type must be able to hold either a 0 or 1. For most
47 | implementations of C, `flag', `uint8', and `int8' should all be `typedef'ed
48 | to the same as `int'.
49 *----------------------------------------------------------------------------*/
51 typedef uint8_t uint8
;
57 typedef unsigned int uint32
;
58 typedef signed int int32
;
59 typedef uint64_t uint64
;
60 typedef int64_t int64
;
62 /*----------------------------------------------------------------------------
63 | Each of the following `typedef's defines a type that holds integers
64 | of _exactly_ the number of bits specified. For instance, for most
65 | implementation of C, `bits16' and `sbits16' should be `typedef'ed to
66 | `unsigned short int' and `signed short int' (or `short int'), respectively.
67 *----------------------------------------------------------------------------*/
68 typedef uint8_t bits8
;
69 typedef int8_t sbits8
;
70 typedef uint16_t bits16
;
71 typedef int16_t sbits16
;
72 typedef uint32_t bits32
;
73 typedef int32_t sbits32
;
74 typedef uint64_t bits64
;
75 typedef int64_t sbits64
;
77 #define LIT64( a ) a##LL
78 #define INLINE static inline
80 /*----------------------------------------------------------------------------
81 | The macro `FLOATX80' must be defined to enable the extended double-precision
82 | floating-point format `floatx80'. If this macro is not defined, the
83 | `floatx80' type will not be defined, and none of the functions that either
84 | input or output the `floatx80' type will be defined. The same applies to
85 | the `FLOAT128' macro and the quadruple-precision format `float128'.
86 *----------------------------------------------------------------------------*/
87 #ifdef CONFIG_SOFTFLOAT
88 /* bit exact soft float support */
92 /* native float support */
93 #if (defined(__i386__) || defined(__x86_64__)) && !defined(CONFIG_BSD)
96 #endif /* !CONFIG_SOFTFLOAT */
98 #define STATUS_PARAM , float_status *status
99 #define STATUS(field) status->field
100 #define STATUS_VAR , status
102 /*----------------------------------------------------------------------------
103 | Software IEC/IEEE floating-point ordering relations
104 *----------------------------------------------------------------------------*/
106 float_relation_less
= -1,
107 float_relation_equal
= 0,
108 float_relation_greater
= 1,
109 float_relation_unordered
= 2
112 #ifdef CONFIG_SOFTFLOAT
113 /*----------------------------------------------------------------------------
114 | Software IEC/IEEE floating-point types.
115 *----------------------------------------------------------------------------*/
116 /* Use structures for soft-float types. This prevents accidentally mixing
117 them with native int/float types. A sufficiently clever compiler and
118 sane ABI should be able to see though these structs. However
119 x86/gcc 3.x seems to struggle a bit, so leave them disabled by default. */
120 //#define USE_SOFTFLOAT_STRUCT_TYPES
121 #ifdef USE_SOFTFLOAT_STRUCT_TYPES
125 /* The cast ensures an error if the wrong type is passed. */
126 #define float32_val(x) (((float32)(x)).v)
127 #define make_float32(x) __extension__ ({ float32 f32_val = {x}; f32_val; })
131 #define float64_val(x) (((float64)(x)).v)
132 #define make_float64(x) __extension__ ({ float64 f64_val = {x}; f64_val; })
134 typedef uint32_t float32
;
135 typedef uint64_t float64
;
136 #define float32_val(x) (x)
137 #define float64_val(x) (x)
138 #define make_float32(x) (x)
139 #define make_float64(x) (x)
149 #ifdef HOST_WORDS_BIGENDIAN
157 /*----------------------------------------------------------------------------
158 | Software IEC/IEEE floating-point underflow tininess-detection mode.
159 *----------------------------------------------------------------------------*/
161 float_tininess_after_rounding
= 0,
162 float_tininess_before_rounding
= 1
165 /*----------------------------------------------------------------------------
166 | Software IEC/IEEE floating-point rounding mode.
167 *----------------------------------------------------------------------------*/
169 float_round_nearest_even
= 0,
170 float_round_down
= 1,
172 float_round_to_zero
= 3
175 /*----------------------------------------------------------------------------
176 | Software IEC/IEEE floating-point exception flags.
177 *----------------------------------------------------------------------------*/
179 float_flag_invalid
= 1,
180 float_flag_divbyzero
= 4,
181 float_flag_overflow
= 8,
182 float_flag_underflow
= 16,
183 float_flag_inexact
= 32,
184 float_flag_input_denormal
= 64
187 typedef struct float_status
{
188 signed char float_detect_tininess
;
189 signed char float_rounding_mode
;
190 signed char float_exception_flags
;
192 signed char floatx80_rounding_precision
;
194 /* should denormalised results go to zero and set the inexact flag? */
196 /* should denormalised inputs go to zero and set the input_denormal flag? */
197 flag flush_inputs_to_zero
;
198 flag default_nan_mode
;
201 void set_float_rounding_mode(int val STATUS_PARAM
);
202 void set_float_exception_flags(int val STATUS_PARAM
);
203 INLINE
void set_flush_to_zero(flag val STATUS_PARAM
)
205 STATUS(flush_to_zero
) = val
;
207 INLINE
void set_flush_inputs_to_zero(flag val STATUS_PARAM
)
209 STATUS(flush_inputs_to_zero
) = val
;
211 INLINE
void set_default_nan_mode(flag val STATUS_PARAM
)
213 STATUS(default_nan_mode
) = val
;
215 INLINE
int get_float_exception_flags(float_status
*status
)
217 return STATUS(float_exception_flags
);
220 void set_floatx80_rounding_precision(int val STATUS_PARAM
);
223 /*----------------------------------------------------------------------------
224 | Routine to raise any or all of the software IEC/IEEE floating-point
226 *----------------------------------------------------------------------------*/
227 void float_raise( int8 flags STATUS_PARAM
);
229 /*----------------------------------------------------------------------------
230 | Software IEC/IEEE integer-to-floating-point conversion routines.
231 *----------------------------------------------------------------------------*/
232 float32
int32_to_float32( int STATUS_PARAM
);
233 float64
int32_to_float64( int STATUS_PARAM
);
234 float32
uint32_to_float32( unsigned int STATUS_PARAM
);
235 float64
uint32_to_float64( unsigned int STATUS_PARAM
);
237 floatx80
int32_to_floatx80( int STATUS_PARAM
);
240 float128
int32_to_float128( int STATUS_PARAM
);
242 float32
int64_to_float32( int64_t STATUS_PARAM
);
243 float32
uint64_to_float32( uint64_t STATUS_PARAM
);
244 float64
int64_to_float64( int64_t STATUS_PARAM
);
245 float64
uint64_to_float64( uint64_t STATUS_PARAM
);
247 floatx80
int64_to_floatx80( int64_t STATUS_PARAM
);
250 float128
int64_to_float128( int64_t STATUS_PARAM
);
253 /*----------------------------------------------------------------------------
254 | Software half-precision conversion routines.
255 *----------------------------------------------------------------------------*/
256 bits16
float32_to_float16( float32
, flag STATUS_PARAM
);
257 float32
float16_to_float32( bits16
, flag STATUS_PARAM
);
259 /*----------------------------------------------------------------------------
260 | Software IEC/IEEE single-precision conversion routines.
261 *----------------------------------------------------------------------------*/
262 int float32_to_int16_round_to_zero( float32 STATUS_PARAM
);
263 unsigned int float32_to_uint16_round_to_zero( float32 STATUS_PARAM
);
264 int float32_to_int32( float32 STATUS_PARAM
);
265 int float32_to_int32_round_to_zero( float32 STATUS_PARAM
);
266 unsigned int float32_to_uint32( float32 STATUS_PARAM
);
267 unsigned int float32_to_uint32_round_to_zero( float32 STATUS_PARAM
);
268 int64_t float32_to_int64( float32 STATUS_PARAM
);
269 int64_t float32_to_int64_round_to_zero( float32 STATUS_PARAM
);
270 float64
float32_to_float64( float32 STATUS_PARAM
);
272 floatx80
float32_to_floatx80( float32 STATUS_PARAM
);
275 float128
float32_to_float128( float32 STATUS_PARAM
);
278 /*----------------------------------------------------------------------------
279 | Software IEC/IEEE single-precision operations.
280 *----------------------------------------------------------------------------*/
281 float32
float32_round_to_int( float32 STATUS_PARAM
);
282 float32
float32_add( float32
, float32 STATUS_PARAM
);
283 float32
float32_sub( float32
, float32 STATUS_PARAM
);
284 float32
float32_mul( float32
, float32 STATUS_PARAM
);
285 float32
float32_div( float32
, float32 STATUS_PARAM
);
286 float32
float32_rem( float32
, float32 STATUS_PARAM
);
287 float32
float32_sqrt( float32 STATUS_PARAM
);
288 float32
float32_exp2( float32 STATUS_PARAM
);
289 float32
float32_log2( float32 STATUS_PARAM
);
290 int float32_eq( float32
, float32 STATUS_PARAM
);
291 int float32_le( float32
, float32 STATUS_PARAM
);
292 int float32_lt( float32
, float32 STATUS_PARAM
);
293 int float32_eq_signaling( float32
, float32 STATUS_PARAM
);
294 int float32_le_quiet( float32
, float32 STATUS_PARAM
);
295 int float32_lt_quiet( float32
, float32 STATUS_PARAM
);
296 int float32_compare( float32
, float32 STATUS_PARAM
);
297 int float32_compare_quiet( float32
, float32 STATUS_PARAM
);
298 int float32_is_quiet_nan( float32
);
299 int float32_is_signaling_nan( float32
);
300 float32
float32_maybe_silence_nan( float32
);
301 float32
float32_scalbn( float32
, int STATUS_PARAM
);
303 INLINE float32
float32_abs(float32 a
)
305 /* Note that abs does *not* handle NaN specially, nor does
306 * it flush denormal inputs to zero.
308 return make_float32(float32_val(a
) & 0x7fffffff);
311 INLINE float32
float32_chs(float32 a
)
313 /* Note that chs does *not* handle NaN specially, nor does
314 * it flush denormal inputs to zero.
316 return make_float32(float32_val(a
) ^ 0x80000000);
319 INLINE
int float32_is_infinity(float32 a
)
321 return (float32_val(a
) & 0x7fffffff) == 0x7f800000;
324 INLINE
int float32_is_neg(float32 a
)
326 return float32_val(a
) >> 31;
329 INLINE
int float32_is_zero(float32 a
)
331 return (float32_val(a
) & 0x7fffffff) == 0;
334 INLINE
int float32_is_any_nan(float32 a
)
336 return ((float32_val(a
) & ~(1 << 31)) > 0x7f800000UL
);
339 INLINE
int float32_is_zero_or_denormal(float32 a
)
341 return (float32_val(a
) & 0x7f800000) == 0;
344 #define float32_zero make_float32(0)
345 #define float32_one make_float32(0x3f800000)
346 #define float32_ln2 make_float32(0x3f317218)
348 /*----------------------------------------------------------------------------
349 | Software IEC/IEEE double-precision conversion routines.
350 *----------------------------------------------------------------------------*/
351 int float64_to_int16_round_to_zero( float64 STATUS_PARAM
);
352 unsigned int float64_to_uint16_round_to_zero( float64 STATUS_PARAM
);
353 int float64_to_int32( float64 STATUS_PARAM
);
354 int float64_to_int32_round_to_zero( float64 STATUS_PARAM
);
355 unsigned int float64_to_uint32( float64 STATUS_PARAM
);
356 unsigned int float64_to_uint32_round_to_zero( float64 STATUS_PARAM
);
357 int64_t float64_to_int64( float64 STATUS_PARAM
);
358 int64_t float64_to_int64_round_to_zero( float64 STATUS_PARAM
);
359 uint64_t float64_to_uint64 (float64 a STATUS_PARAM
);
360 uint64_t float64_to_uint64_round_to_zero (float64 a STATUS_PARAM
);
361 float32
float64_to_float32( float64 STATUS_PARAM
);
363 floatx80
float64_to_floatx80( float64 STATUS_PARAM
);
366 float128
float64_to_float128( float64 STATUS_PARAM
);
369 /*----------------------------------------------------------------------------
370 | Software IEC/IEEE double-precision operations.
371 *----------------------------------------------------------------------------*/
372 float64
float64_round_to_int( float64 STATUS_PARAM
);
373 float64
float64_trunc_to_int( float64 STATUS_PARAM
);
374 float64
float64_add( float64
, float64 STATUS_PARAM
);
375 float64
float64_sub( float64
, float64 STATUS_PARAM
);
376 float64
float64_mul( float64
, float64 STATUS_PARAM
);
377 float64
float64_div( float64
, float64 STATUS_PARAM
);
378 float64
float64_rem( float64
, float64 STATUS_PARAM
);
379 float64
float64_sqrt( float64 STATUS_PARAM
);
380 float64
float64_log2( float64 STATUS_PARAM
);
381 int float64_eq( float64
, float64 STATUS_PARAM
);
382 int float64_le( float64
, float64 STATUS_PARAM
);
383 int float64_lt( float64
, float64 STATUS_PARAM
);
384 int float64_eq_signaling( float64
, float64 STATUS_PARAM
);
385 int float64_le_quiet( float64
, float64 STATUS_PARAM
);
386 int float64_lt_quiet( float64
, float64 STATUS_PARAM
);
387 int float64_compare( float64
, float64 STATUS_PARAM
);
388 int float64_compare_quiet( float64
, float64 STATUS_PARAM
);
389 int float64_is_quiet_nan( float64 a
);
390 int float64_is_signaling_nan( float64
);
391 float64
float64_maybe_silence_nan( float64
);
392 float64
float64_scalbn( float64
, int STATUS_PARAM
);
394 INLINE float64
float64_abs(float64 a
)
396 /* Note that abs does *not* handle NaN specially, nor does
397 * it flush denormal inputs to zero.
399 return make_float64(float64_val(a
) & 0x7fffffffffffffffLL
);
402 INLINE float64
float64_chs(float64 a
)
404 /* Note that chs does *not* handle NaN specially, nor does
405 * it flush denormal inputs to zero.
407 return make_float64(float64_val(a
) ^ 0x8000000000000000LL
);
410 INLINE
int float64_is_infinity(float64 a
)
412 return (float64_val(a
) & 0x7fffffffffffffffLL
) == 0x7ff0000000000000LL
;
415 INLINE
int float64_is_neg(float64 a
)
417 return float64_val(a
) >> 63;
420 INLINE
int float64_is_zero(float64 a
)
422 return (float64_val(a
) & 0x7fffffffffffffffLL
) == 0;
425 INLINE
int float64_is_any_nan(float64 a
)
427 return ((float64_val(a
) & ~(1ULL << 63)) > 0x7ff0000000000000ULL
);
430 #define float64_zero make_float64(0)
431 #define float64_one make_float64(0x3ff0000000000000LL)
432 #define float64_ln2 make_float64(0x3fe62e42fefa39efLL)
436 /*----------------------------------------------------------------------------
437 | Software IEC/IEEE extended double-precision conversion routines.
438 *----------------------------------------------------------------------------*/
439 int floatx80_to_int32( floatx80 STATUS_PARAM
);
440 int floatx80_to_int32_round_to_zero( floatx80 STATUS_PARAM
);
441 int64_t floatx80_to_int64( floatx80 STATUS_PARAM
);
442 int64_t floatx80_to_int64_round_to_zero( floatx80 STATUS_PARAM
);
443 float32
floatx80_to_float32( floatx80 STATUS_PARAM
);
444 float64
floatx80_to_float64( floatx80 STATUS_PARAM
);
446 float128
floatx80_to_float128( floatx80 STATUS_PARAM
);
449 /*----------------------------------------------------------------------------
450 | Software IEC/IEEE extended double-precision operations.
451 *----------------------------------------------------------------------------*/
452 floatx80
floatx80_round_to_int( floatx80 STATUS_PARAM
);
453 floatx80
floatx80_add( floatx80
, floatx80 STATUS_PARAM
);
454 floatx80
floatx80_sub( floatx80
, floatx80 STATUS_PARAM
);
455 floatx80
floatx80_mul( floatx80
, floatx80 STATUS_PARAM
);
456 floatx80
floatx80_div( floatx80
, floatx80 STATUS_PARAM
);
457 floatx80
floatx80_rem( floatx80
, floatx80 STATUS_PARAM
);
458 floatx80
floatx80_sqrt( floatx80 STATUS_PARAM
);
459 int floatx80_eq( floatx80
, floatx80 STATUS_PARAM
);
460 int floatx80_le( floatx80
, floatx80 STATUS_PARAM
);
461 int floatx80_lt( floatx80
, floatx80 STATUS_PARAM
);
462 int floatx80_eq_signaling( floatx80
, floatx80 STATUS_PARAM
);
463 int floatx80_le_quiet( floatx80
, floatx80 STATUS_PARAM
);
464 int floatx80_lt_quiet( floatx80
, floatx80 STATUS_PARAM
);
465 int floatx80_is_quiet_nan( floatx80
);
466 int floatx80_is_signaling_nan( floatx80
);
467 floatx80
floatx80_maybe_silence_nan( floatx80
);
468 floatx80
floatx80_scalbn( floatx80
, int STATUS_PARAM
);
470 INLINE floatx80
floatx80_abs(floatx80 a
)
476 INLINE floatx80
floatx80_chs(floatx80 a
)
482 INLINE
int floatx80_is_infinity(floatx80 a
)
484 return (a
.high
& 0x7fff) == 0x7fff && a
.low
== 0;
487 INLINE
int floatx80_is_neg(floatx80 a
)
492 INLINE
int floatx80_is_zero(floatx80 a
)
494 return (a
.high
& 0x7fff) == 0 && a
.low
== 0;
497 INLINE
int floatx80_is_any_nan(floatx80 a
)
499 return ((a
.high
& 0x7fff) == 0x7fff) && (a
.low
<<1);
506 /*----------------------------------------------------------------------------
507 | Software IEC/IEEE quadruple-precision conversion routines.
508 *----------------------------------------------------------------------------*/
509 int float128_to_int32( float128 STATUS_PARAM
);
510 int float128_to_int32_round_to_zero( float128 STATUS_PARAM
);
511 int64_t float128_to_int64( float128 STATUS_PARAM
);
512 int64_t float128_to_int64_round_to_zero( float128 STATUS_PARAM
);
513 float32
float128_to_float32( float128 STATUS_PARAM
);
514 float64
float128_to_float64( float128 STATUS_PARAM
);
516 floatx80
float128_to_floatx80( float128 STATUS_PARAM
);
519 /*----------------------------------------------------------------------------
520 | Software IEC/IEEE quadruple-precision operations.
521 *----------------------------------------------------------------------------*/
522 float128
float128_round_to_int( float128 STATUS_PARAM
);
523 float128
float128_add( float128
, float128 STATUS_PARAM
);
524 float128
float128_sub( float128
, float128 STATUS_PARAM
);
525 float128
float128_mul( float128
, float128 STATUS_PARAM
);
526 float128
float128_div( float128
, float128 STATUS_PARAM
);
527 float128
float128_rem( float128
, float128 STATUS_PARAM
);
528 float128
float128_sqrt( float128 STATUS_PARAM
);
529 int float128_eq( float128
, float128 STATUS_PARAM
);
530 int float128_le( float128
, float128 STATUS_PARAM
);
531 int float128_lt( float128
, float128 STATUS_PARAM
);
532 int float128_eq_signaling( float128
, float128 STATUS_PARAM
);
533 int float128_le_quiet( float128
, float128 STATUS_PARAM
);
534 int float128_lt_quiet( float128
, float128 STATUS_PARAM
);
535 int float128_compare( float128
, float128 STATUS_PARAM
);
536 int float128_compare_quiet( float128
, float128 STATUS_PARAM
);
537 int float128_is_quiet_nan( float128
);
538 int float128_is_signaling_nan( float128
);
539 float128
float128_maybe_silence_nan( float128
);
540 float128
float128_scalbn( float128
, int STATUS_PARAM
);
542 INLINE float128
float128_abs(float128 a
)
544 a
.high
&= 0x7fffffffffffffffLL
;
548 INLINE float128
float128_chs(float128 a
)
550 a
.high
^= 0x8000000000000000LL
;
554 INLINE
int float128_is_infinity(float128 a
)
556 return (a
.high
& 0x7fffffffffffffffLL
) == 0x7fff000000000000LL
&& a
.low
== 0;
559 INLINE
int float128_is_neg(float128 a
)
564 INLINE
int float128_is_zero(float128 a
)
566 return (a
.high
& 0x7fffffffffffffffLL
) == 0 && a
.low
== 0;
569 INLINE
int float128_is_any_nan(float128 a
)
571 return ((a
.high
>> 48) & 0x7fff) == 0x7fff &&
572 ((a
.low
!= 0) || ((a
.high
& 0xffffffffffffLL
) != 0));
577 #else /* CONFIG_SOFTFLOAT */
579 #include "softfloat-native.h"
581 #endif /* !CONFIG_SOFTFLOAT */
583 #endif /* !SOFTFLOAT_H */