1 /*============================================================================
2
3 This C header file is part of the SoftFloat IEC/IEEE Floating-point Arithmetic
4 Package, Release 2b.
5
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'.
15
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.
24
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.
29
30 =============================================================================*/
31
32 #ifndef SOFTFLOAT_H
33 #define SOFTFLOAT_H
34
35 #if defined(HOST_SOLARIS) && defined(NEEDS_LIBSUNMATH)
36 #include <sunmath.h>
37 #endif
38
39 #include <inttypes.h>
40 #include "config.h"
41
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 *----------------------------------------------------------------------------*/
50 typedef uint8_t flag;
51 typedef uint8_t uint8;
52 typedef int8_t int8;
53 typedef int uint16;
54 typedef int int16;
55 typedef unsigned int uint32;
56 typedef signed int int32;
57 typedef uint64_t uint64;
58 typedef int64_t int64;
59
60 /*----------------------------------------------------------------------------
61 | Each of the following `typedef's defines a type that holds integers
62 | of _exactly_ the number of bits specified. For instance, for most
63 | implementation of C, `bits16' and `sbits16' should be `typedef'ed to
64 | `unsigned short int' and `signed short int' (or `short int'), respectively.
65 *----------------------------------------------------------------------------*/
66 typedef uint8_t bits8;
67 typedef int8_t sbits8;
68 typedef uint16_t bits16;
69 typedef int16_t sbits16;
70 typedef uint32_t bits32;
71 typedef int32_t sbits32;
72 typedef uint64_t bits64;
73 typedef int64_t sbits64;
74
75 #define LIT64( a ) a##LL
76 #define INLINE static inline
77
78 /*----------------------------------------------------------------------------
79 | The macro `FLOATX80' must be defined to enable the extended double-precision
80 | floating-point format `floatx80'. If this macro is not defined, the
81 | `floatx80' type will not be defined, and none of the functions that either
82 | input or output the `floatx80' type will be defined. The same applies to
83 | the `FLOAT128' macro and the quadruple-precision format `float128'.
84 *----------------------------------------------------------------------------*/
85 #ifdef CONFIG_SOFTFLOAT
86 /* bit exact soft float support */
87 #define FLOATX80
88 #define FLOAT128
89 #else
90 /* native float support */
91 #if (defined(__i386__) || defined(__x86_64__)) && !defined(_BSD)
92 #define FLOATX80
93 #endif
94 #endif /* !CONFIG_SOFTFLOAT */
95
96 #define STATUS_PARAM , float_status *status
97 #define STATUS(field) status->field
98 #define STATUS_VAR , status
99
100 /*----------------------------------------------------------------------------
101 | Software IEC/IEEE floating-point ordering relations
102 *----------------------------------------------------------------------------*/
103 enum {
104 float_relation_less = -1,
105 float_relation_equal = 0,
106 float_relation_greater = 1,
107 float_relation_unordered = 2
108 };
109
110 #ifdef CONFIG_SOFTFLOAT
111 /*----------------------------------------------------------------------------
112 | Software IEC/IEEE floating-point types.
113 *----------------------------------------------------------------------------*/
114 /* Use structures for soft-float types. This prevents accidentally mixing
115 them with native int/float types. A sufficiently clever compiler and
116 sane ABI should be able to see though these structs. However
117 x86/gcc 3.x seems to struggle a bit, so leave them disabled by default. */
118 //#define USE_SOFTFLOAT_STRUCT_TYPES
119 #ifdef USE_SOFTFLOAT_STRUCT_TYPES
120 typedef struct {
121 uint32_t v;
122 } float32;
123 /* The cast ensures an error if the wrong type is passed. */
124 #define float32_val(x) (((float32)(x)).v)
125 #define make_float32(x) __extension__ ({ float32 f32_val = {x}; f32_val; })
126 typedef struct {
127 uint64_t v;
128 } float64;
129 #define float64_val(x) (((float64)(x)).v)
130 #define make_float64(x) __extension__ ({ float64 f64_val = {x}; f64_val; })
131 #else
132 typedef uint32_t float32;
133 typedef uint64_t float64;
134 #define float32_val(x) (x)
135 #define float64_val(x) (x)
136 #define make_float32(x) (x)
137 #define make_float64(x) (x)
138 #endif
139 #ifdef FLOATX80
140 typedef struct {
141 uint64_t low;
142 uint16_t high;
143 } floatx80;
144 #endif
145 #ifdef FLOAT128
146 typedef struct {
147 #ifdef WORDS_BIGENDIAN
148 uint64_t high, low;
149 #else
150 uint64_t low, high;
151 #endif
152 } float128;
153 #endif
154
155 /*----------------------------------------------------------------------------
156 | Software IEC/IEEE floating-point underflow tininess-detection mode.
157 *----------------------------------------------------------------------------*/
158 enum {
159 float_tininess_after_rounding = 0,
160 float_tininess_before_rounding = 1
161 };
162
163 /*----------------------------------------------------------------------------
164 | Software IEC/IEEE floating-point rounding mode.
165 *----------------------------------------------------------------------------*/
166 enum {
167 float_round_nearest_even = 0,
168 float_round_down = 1,
169 float_round_up = 2,
170 float_round_to_zero = 3
171 };
172
173 /*----------------------------------------------------------------------------
174 | Software IEC/IEEE floating-point exception flags.
175 *----------------------------------------------------------------------------*/
176 enum {
177 float_flag_invalid = 1,
178 float_flag_divbyzero = 4,
179 float_flag_overflow = 8,
180 float_flag_underflow = 16,
181 float_flag_inexact = 32
182 };
183
184 typedef struct float_status {
185 signed char float_detect_tininess;
186 signed char float_rounding_mode;
187 signed char float_exception_flags;
188 #ifdef FLOATX80
189 signed char floatx80_rounding_precision;
190 #endif
191 } float_status;
192
193 void set_float_rounding_mode(int val STATUS_PARAM);
194 void set_float_exception_flags(int val STATUS_PARAM);
get_float_exception_flags(float_status * status)195 INLINE int get_float_exception_flags(float_status *status)
196 {
197 return STATUS(float_exception_flags);
198 }
199 #ifdef FLOATX80
200 void set_floatx80_rounding_precision(int val STATUS_PARAM);
201 #endif
202
203 /*----------------------------------------------------------------------------
204 | Routine to raise any or all of the software IEC/IEEE floating-point
205 | exception flags.
206 *----------------------------------------------------------------------------*/
207 void float_raise( int8 flags STATUS_PARAM);
208
209 /*----------------------------------------------------------------------------
210 | Software IEC/IEEE integer-to-floating-point conversion routines.
211 *----------------------------------------------------------------------------*/
212 float32 int32_to_float32( int STATUS_PARAM );
213 float64 int32_to_float64( int STATUS_PARAM );
214 float32 uint32_to_float32( unsigned int STATUS_PARAM );
215 float64 uint32_to_float64( unsigned int STATUS_PARAM );
216 #ifdef FLOATX80
217 floatx80 int32_to_floatx80( int STATUS_PARAM );
218 #endif
219 #ifdef FLOAT128
220 float128 int32_to_float128( int STATUS_PARAM );
221 #endif
222 float32 int64_to_float32( int64_t STATUS_PARAM );
223 float32 uint64_to_float32( uint64_t STATUS_PARAM );
224 float64 int64_to_float64( int64_t STATUS_PARAM );
225 float64 uint64_to_float64( uint64_t STATUS_PARAM );
226 #ifdef FLOATX80
227 floatx80 int64_to_floatx80( int64_t STATUS_PARAM );
228 #endif
229 #ifdef FLOAT128
230 float128 int64_to_float128( int64_t STATUS_PARAM );
231 #endif
232
233 /*----------------------------------------------------------------------------
234 | Software IEC/IEEE single-precision conversion routines.
235 *----------------------------------------------------------------------------*/
236 int float32_to_int32( float32 STATUS_PARAM );
237 int float32_to_int32_round_to_zero( float32 STATUS_PARAM );
238 unsigned int float32_to_uint32( float32 STATUS_PARAM );
239 unsigned int float32_to_uint32_round_to_zero( float32 STATUS_PARAM );
240 int64_t float32_to_int64( float32 STATUS_PARAM );
241 int64_t float32_to_int64_round_to_zero( float32 STATUS_PARAM );
242 float64 float32_to_float64( float32 STATUS_PARAM );
243 #ifdef FLOATX80
244 floatx80 float32_to_floatx80( float32 STATUS_PARAM );
245 #endif
246 #ifdef FLOAT128
247 float128 float32_to_float128( float32 STATUS_PARAM );
248 #endif
249
250 /*----------------------------------------------------------------------------
251 | Software IEC/IEEE single-precision operations.
252 *----------------------------------------------------------------------------*/
253 float32 float32_round_to_int( float32 STATUS_PARAM );
254 float32 float32_add( float32, float32 STATUS_PARAM );
255 float32 float32_sub( float32, float32 STATUS_PARAM );
256 float32 float32_mul( float32, float32 STATUS_PARAM );
257 float32 float32_div( float32, float32 STATUS_PARAM );
258 float32 float32_rem( float32, float32 STATUS_PARAM );
259 float32 float32_sqrt( float32 STATUS_PARAM );
260 int float32_eq( float32, float32 STATUS_PARAM );
261 int float32_le( float32, float32 STATUS_PARAM );
262 int float32_lt( float32, float32 STATUS_PARAM );
263 int float32_eq_signaling( float32, float32 STATUS_PARAM );
264 int float32_le_quiet( float32, float32 STATUS_PARAM );
265 int float32_lt_quiet( float32, float32 STATUS_PARAM );
266 int float32_compare( float32, float32 STATUS_PARAM );
267 int float32_compare_quiet( float32, float32 STATUS_PARAM );
268 int float32_is_nan( float32 );
269 int float32_is_signaling_nan( float32 );
270 float32 float32_scalbn( float32, int STATUS_PARAM );
271
float32_abs(float32 a)272 INLINE float32 float32_abs(float32 a)
273 {
274 return make_float32(float32_val(a) & 0x7fffffff);
275 }
276
float32_chs(float32 a)277 INLINE float32 float32_chs(float32 a)
278 {
279 return make_float32(float32_val(a) ^ 0x80000000);
280 }
281
282 #define float32_zero make_float32(0)
283
284 /*----------------------------------------------------------------------------
285 | Software IEC/IEEE double-precision conversion routines.
286 *----------------------------------------------------------------------------*/
287 int float64_to_int32( float64 STATUS_PARAM );
288 int float64_to_int32_round_to_zero( float64 STATUS_PARAM );
289 unsigned int float64_to_uint32( float64 STATUS_PARAM );
290 unsigned int float64_to_uint32_round_to_zero( float64 STATUS_PARAM );
291 int64_t float64_to_int64( float64 STATUS_PARAM );
292 int64_t float64_to_int64_round_to_zero( float64 STATUS_PARAM );
293 uint64_t float64_to_uint64 (float64 a STATUS_PARAM);
294 uint64_t float64_to_uint64_round_to_zero (float64 a STATUS_PARAM);
295 float32 float64_to_float32( float64 STATUS_PARAM );
296 #ifdef FLOATX80
297 floatx80 float64_to_floatx80( float64 STATUS_PARAM );
298 #endif
299 #ifdef FLOAT128
300 float128 float64_to_float128( float64 STATUS_PARAM );
301 #endif
302
303 /*----------------------------------------------------------------------------
304 | Software IEC/IEEE double-precision operations.
305 *----------------------------------------------------------------------------*/
306 float64 float64_round_to_int( float64 STATUS_PARAM );
307 float64 float64_trunc_to_int( float64 STATUS_PARAM );
308 float64 float64_add( float64, float64 STATUS_PARAM );
309 float64 float64_sub( float64, float64 STATUS_PARAM );
310 float64 float64_mul( float64, float64 STATUS_PARAM );
311 float64 float64_div( float64, float64 STATUS_PARAM );
312 float64 float64_rem( float64, float64 STATUS_PARAM );
313 float64 float64_sqrt( float64 STATUS_PARAM );
314 int float64_eq( float64, float64 STATUS_PARAM );
315 int float64_le( float64, float64 STATUS_PARAM );
316 int float64_lt( float64, float64 STATUS_PARAM );
317 int float64_eq_signaling( float64, float64 STATUS_PARAM );
318 int float64_le_quiet( float64, float64 STATUS_PARAM );
319 int float64_lt_quiet( float64, float64 STATUS_PARAM );
320 int float64_compare( float64, float64 STATUS_PARAM );
321 int float64_compare_quiet( float64, float64 STATUS_PARAM );
322 int float64_is_nan( float64 a );
323 int float64_is_signaling_nan( float64 );
324 float64 float64_scalbn( float64, int STATUS_PARAM );
325
float64_abs(float64 a)326 INLINE float64 float64_abs(float64 a)
327 {
328 return make_float64(float64_val(a) & 0x7fffffffffffffffLL);
329 }
330
float64_chs(float64 a)331 INLINE float64 float64_chs(float64 a)
332 {
333 return make_float64(float64_val(a) ^ 0x8000000000000000LL);
334 }
335
336 #define float64_zero make_float64(0)
337
338 #ifdef FLOATX80
339
340 /*----------------------------------------------------------------------------
341 | Software IEC/IEEE extended double-precision conversion routines.
342 *----------------------------------------------------------------------------*/
343 int floatx80_to_int32( floatx80 STATUS_PARAM );
344 int floatx80_to_int32_round_to_zero( floatx80 STATUS_PARAM );
345 int64_t floatx80_to_int64( floatx80 STATUS_PARAM );
346 int64_t floatx80_to_int64_round_to_zero( floatx80 STATUS_PARAM );
347 float32 floatx80_to_float32( floatx80 STATUS_PARAM );
348 float64 floatx80_to_float64( floatx80 STATUS_PARAM );
349 #ifdef FLOAT128
350 float128 floatx80_to_float128( floatx80 STATUS_PARAM );
351 #endif
352
353 /*----------------------------------------------------------------------------
354 | Software IEC/IEEE extended double-precision operations.
355 *----------------------------------------------------------------------------*/
356 floatx80 floatx80_round_to_int( floatx80 STATUS_PARAM );
357 floatx80 floatx80_add( floatx80, floatx80 STATUS_PARAM );
358 floatx80 floatx80_sub( floatx80, floatx80 STATUS_PARAM );
359 floatx80 floatx80_mul( floatx80, floatx80 STATUS_PARAM );
360 floatx80 floatx80_div( floatx80, floatx80 STATUS_PARAM );
361 floatx80 floatx80_rem( floatx80, floatx80 STATUS_PARAM );
362 floatx80 floatx80_sqrt( floatx80 STATUS_PARAM );
363 int floatx80_eq( floatx80, floatx80 STATUS_PARAM );
364 int floatx80_le( floatx80, floatx80 STATUS_PARAM );
365 int floatx80_lt( floatx80, floatx80 STATUS_PARAM );
366 int floatx80_eq_signaling( floatx80, floatx80 STATUS_PARAM );
367 int floatx80_le_quiet( floatx80, floatx80 STATUS_PARAM );
368 int floatx80_lt_quiet( floatx80, floatx80 STATUS_PARAM );
369 int floatx80_is_nan( floatx80 );
370 int floatx80_is_signaling_nan( floatx80 );
371 floatx80 floatx80_scalbn( floatx80, int STATUS_PARAM );
372
floatx80_abs(floatx80 a)373 INLINE floatx80 floatx80_abs(floatx80 a)
374 {
375 a.high &= 0x7fff;
376 return a;
377 }
378
floatx80_chs(floatx80 a)379 INLINE floatx80 floatx80_chs(floatx80 a)
380 {
381 a.high ^= 0x8000;
382 return a;
383 }
384
385 #endif
386
387 #ifdef FLOAT128
388
389 /*----------------------------------------------------------------------------
390 | Software IEC/IEEE quadruple-precision conversion routines.
391 *----------------------------------------------------------------------------*/
392 int float128_to_int32( float128 STATUS_PARAM );
393 int float128_to_int32_round_to_zero( float128 STATUS_PARAM );
394 int64_t float128_to_int64( float128 STATUS_PARAM );
395 int64_t float128_to_int64_round_to_zero( float128 STATUS_PARAM );
396 float32 float128_to_float32( float128 STATUS_PARAM );
397 float64 float128_to_float64( float128 STATUS_PARAM );
398 #ifdef FLOATX80
399 floatx80 float128_to_floatx80( float128 STATUS_PARAM );
400 #endif
401
402 /*----------------------------------------------------------------------------
403 | Software IEC/IEEE quadruple-precision operations.
404 *----------------------------------------------------------------------------*/
405 float128 float128_round_to_int( float128 STATUS_PARAM );
406 float128 float128_add( float128, float128 STATUS_PARAM );
407 float128 float128_sub( float128, float128 STATUS_PARAM );
408 float128 float128_mul( float128, float128 STATUS_PARAM );
409 float128 float128_div( float128, float128 STATUS_PARAM );
410 float128 float128_rem( float128, float128 STATUS_PARAM );
411 float128 float128_sqrt( float128 STATUS_PARAM );
412 int float128_eq( float128, float128 STATUS_PARAM );
413 int float128_le( float128, float128 STATUS_PARAM );
414 int float128_lt( float128, float128 STATUS_PARAM );
415 int float128_eq_signaling( float128, float128 STATUS_PARAM );
416 int float128_le_quiet( float128, float128 STATUS_PARAM );
417 int float128_lt_quiet( float128, float128 STATUS_PARAM );
418 int float128_compare( float128, float128 STATUS_PARAM );
419 int float128_compare_quiet( float128, float128 STATUS_PARAM );
420 int float128_is_nan( float128 );
421 int float128_is_signaling_nan( float128 );
422 float128 float128_scalbn( float128, int STATUS_PARAM );
423
float128_abs(float128 a)424 INLINE float128 float128_abs(float128 a)
425 {
426 a.high &= 0x7fffffffffffffffLL;
427 return a;
428 }
429
float128_chs(float128 a)430 INLINE float128 float128_chs(float128 a)
431 {
432 a.high ^= 0x8000000000000000LL;
433 return a;
434 }
435
436 #endif
437
438 #else /* CONFIG_SOFTFLOAT */
439
440 #include "softfloat-native.h"
441
442 #endif /* !CONFIG_SOFTFLOAT */
443
444 #endif /* !SOFTFLOAT_H */
445