Disable preadv/pwritev support
[qemu-kvm/fedora.git] / fpu / softfloat-native.h
bloba28c76947236b4f8758731441339e2ca215e328f
1 /* Native implementation of soft float functions */
2 #include <math.h>
4 #if (defined(HOST_BSD) && !defined(__APPLE__)) || defined(HOST_SOLARIS)
5 #include <ieeefp.h>
6 #define fabsf(f) ((float)fabs(f))
7 #else
8 #include <fenv.h>
9 #endif
11 #if defined(__OpenBSD__) || defined(__NetBSD__)
12 #include <sys/param.h>
13 #endif
16 * Define some C99-7.12.3 classification macros and
17 * some C99-.12.4 for Solaris systems OS less than 10,
18 * or Solaris 10 systems running GCC 3.x or less.
19 * Solaris 10 with GCC4 does not need these macros as they
20 * are defined in <iso/math_c99.h> with a compiler directive
22 #if defined(HOST_SOLARIS) && (( HOST_SOLARIS <= 9 ) || ((HOST_SOLARIS >= 10) \
23 && (__GNUC__ < 4))) \
24 || (defined(__OpenBSD__) && (OpenBSD < 200811))
26 * C99 7.12.3 classification macros
27 * and
28 * C99 7.12.14 comparison macros
30 * ... do not work on Solaris 10 using GNU CC 3.4.x.
31 * Try to workaround the missing / broken C99 math macros.
33 #if defined(__OpenBSD__)
34 #define unordered(x, y) (isnan(x) || isnan(y))
35 #endif
37 #ifdef __NetBSD__
38 #ifndef isgreater
39 #define isgreater(x, y) __builtin_isgreater(x, y)
40 #endif
41 #ifndef isgreaterequal
42 #define isgreaterequal(x, y) __builtin_isgreaterequal(x, y)
43 #endif
44 #ifndef isless
45 #define isless(x, y) __builtin_isless(x, y)
46 #endif
47 #ifndef islessequal
48 #define islessequal(x, y) __builtin_islessequal(x, y)
49 #endif
50 #ifndef isunordered
51 #define isunordered(x, y) __builtin_isunordered(x, y)
52 #endif
53 #endif
56 #define isnormal(x) (fpclass(x) >= FP_NZERO)
57 #define isgreater(x, y) ((!unordered(x, y)) && ((x) > (y)))
58 #define isgreaterequal(x, y) ((!unordered(x, y)) && ((x) >= (y)))
59 #define isless(x, y) ((!unordered(x, y)) && ((x) < (y)))
60 #define islessequal(x, y) ((!unordered(x, y)) && ((x) <= (y)))
61 #define isunordered(x,y) unordered(x, y)
62 #endif
64 #if defined(__sun__) && !defined(NEED_LIBSUNMATH)
66 #ifndef isnan
67 # define isnan(x) \
68 (sizeof (x) == sizeof (long double) ? isnan_ld (x) \
69 : sizeof (x) == sizeof (double) ? isnan_d (x) \
70 : isnan_f (x))
71 static inline int isnan_f (float x) { return x != x; }
72 static inline int isnan_d (double x) { return x != x; }
73 static inline int isnan_ld (long double x) { return x != x; }
74 #endif
76 #ifndef isinf
77 # define isinf(x) \
78 (sizeof (x) == sizeof (long double) ? isinf_ld (x) \
79 : sizeof (x) == sizeof (double) ? isinf_d (x) \
80 : isinf_f (x))
81 static inline int isinf_f (float x) { return isnan (x - x); }
82 static inline int isinf_d (double x) { return isnan (x - x); }
83 static inline int isinf_ld (long double x) { return isnan (x - x); }
84 #endif
85 #endif
87 typedef float float32;
88 typedef double float64;
89 #ifdef FLOATX80
90 typedef long double floatx80;
91 #endif
93 typedef union {
94 float32 f;
95 uint32_t i;
96 } float32u;
97 typedef union {
98 float64 f;
99 uint64_t i;
100 } float64u;
101 #ifdef FLOATX80
102 typedef union {
103 floatx80 f;
104 struct {
105 uint64_t low;
106 uint16_t high;
107 } i;
108 } floatx80u;
109 #endif
111 /*----------------------------------------------------------------------------
112 | Software IEC/IEEE floating-point rounding mode.
113 *----------------------------------------------------------------------------*/
114 #if (defined(HOST_BSD) && !defined(__APPLE__)) || defined(HOST_SOLARIS)
115 #if defined(__OpenBSD__)
116 #define FE_RM FP_RM
117 #define FE_RP FP_RP
118 #define FE_RZ FP_RZ
119 #endif
120 enum {
121 float_round_nearest_even = FP_RN,
122 float_round_down = FP_RM,
123 float_round_up = FP_RP,
124 float_round_to_zero = FP_RZ
126 #elif defined(__arm__)
127 enum {
128 float_round_nearest_even = 0,
129 float_round_down = 1,
130 float_round_up = 2,
131 float_round_to_zero = 3
133 #else
134 enum {
135 float_round_nearest_even = FE_TONEAREST,
136 float_round_down = FE_DOWNWARD,
137 float_round_up = FE_UPWARD,
138 float_round_to_zero = FE_TOWARDZERO
140 #endif
142 typedef struct float_status {
143 int float_rounding_mode;
144 #ifdef FLOATX80
145 int floatx80_rounding_precision;
146 #endif
147 } float_status;
149 void set_float_rounding_mode(int val STATUS_PARAM);
150 #ifdef FLOATX80
151 void set_floatx80_rounding_precision(int val STATUS_PARAM);
152 #endif
154 /*----------------------------------------------------------------------------
155 | Software IEC/IEEE integer-to-floating-point conversion routines.
156 *----------------------------------------------------------------------------*/
157 float32 int32_to_float32( int STATUS_PARAM);
158 float32 uint32_to_float32( unsigned int STATUS_PARAM);
159 float64 int32_to_float64( int STATUS_PARAM);
160 float64 uint32_to_float64( unsigned int STATUS_PARAM);
161 #ifdef FLOATX80
162 floatx80 int32_to_floatx80( int STATUS_PARAM);
163 #endif
164 #ifdef FLOAT128
165 float128 int32_to_float128( int STATUS_PARAM);
166 #endif
167 float32 int64_to_float32( int64_t STATUS_PARAM);
168 float32 uint64_to_float32( uint64_t STATUS_PARAM);
169 float64 int64_to_float64( int64_t STATUS_PARAM);
170 float64 uint64_to_float64( uint64_t v STATUS_PARAM);
171 #ifdef FLOATX80
172 floatx80 int64_to_floatx80( int64_t STATUS_PARAM);
173 #endif
174 #ifdef FLOAT128
175 float128 int64_to_float128( int64_t STATUS_PARAM);
176 #endif
178 /*----------------------------------------------------------------------------
179 | Software IEC/IEEE single-precision conversion routines.
180 *----------------------------------------------------------------------------*/
181 int float32_to_int32( float32 STATUS_PARAM);
182 int float32_to_int32_round_to_zero( float32 STATUS_PARAM);
183 unsigned int float32_to_uint32( float32 a STATUS_PARAM);
184 unsigned int float32_to_uint32_round_to_zero( float32 a STATUS_PARAM);
185 int64_t float32_to_int64( float32 STATUS_PARAM);
186 int64_t float32_to_int64_round_to_zero( float32 STATUS_PARAM);
187 float64 float32_to_float64( float32 STATUS_PARAM);
188 #ifdef FLOATX80
189 floatx80 float32_to_floatx80( float32 STATUS_PARAM);
190 #endif
191 #ifdef FLOAT128
192 float128 float32_to_float128( float32 STATUS_PARAM);
193 #endif
195 /*----------------------------------------------------------------------------
196 | Software IEC/IEEE single-precision operations.
197 *----------------------------------------------------------------------------*/
198 float32 float32_round_to_int( float32 STATUS_PARAM);
199 INLINE float32 float32_add( float32 a, float32 b STATUS_PARAM)
201 return a + b;
203 INLINE float32 float32_sub( float32 a, float32 b STATUS_PARAM)
205 return a - b;
207 INLINE float32 float32_mul( float32 a, float32 b STATUS_PARAM)
209 return a * b;
211 INLINE float32 float32_div( float32 a, float32 b STATUS_PARAM)
213 return a / b;
215 float32 float32_rem( float32, float32 STATUS_PARAM);
216 float32 float32_sqrt( float32 STATUS_PARAM);
217 INLINE int float32_eq( float32 a, float32 b STATUS_PARAM)
219 return a == b;
221 INLINE int float32_le( float32 a, float32 b STATUS_PARAM)
223 return a <= b;
225 INLINE int float32_lt( float32 a, float32 b STATUS_PARAM)
227 return a < b;
229 INLINE int float32_eq_signaling( float32 a, float32 b STATUS_PARAM)
231 return a <= b && a >= b;
233 INLINE int float32_le_quiet( float32 a, float32 b STATUS_PARAM)
235 return islessequal(a, b);
237 INLINE int float32_lt_quiet( float32 a, float32 b STATUS_PARAM)
239 return isless(a, b);
241 INLINE int float32_unordered( float32 a, float32 b STATUS_PARAM)
243 return isunordered(a, b);
246 int float32_compare( float32, float32 STATUS_PARAM );
247 int float32_compare_quiet( float32, float32 STATUS_PARAM );
248 int float32_is_signaling_nan( float32 );
249 int float32_is_nan( float32 );
251 INLINE float32 float32_abs(float32 a)
253 return fabsf(a);
256 INLINE float32 float32_chs(float32 a)
258 return -a;
261 INLINE float32 float32_is_infinity(float32 a)
263 return fpclassify(a) == FP_INFINITE;
266 INLINE float32 float32_is_neg(float32 a)
268 float32u u;
269 u.f = a;
270 return u.i >> 31;
273 INLINE float32 float32_is_zero(float32 a)
275 return fpclassify(a) == FP_ZERO;
278 INLINE float32 float32_scalbn(float32 a, int n)
280 return scalbnf(a, n);
283 /*----------------------------------------------------------------------------
284 | Software IEC/IEEE double-precision conversion routines.
285 *----------------------------------------------------------------------------*/
286 int float64_to_int32( float64 STATUS_PARAM );
287 int float64_to_int32_round_to_zero( float64 STATUS_PARAM );
288 unsigned int float64_to_uint32( float64 STATUS_PARAM );
289 unsigned int float64_to_uint32_round_to_zero( float64 STATUS_PARAM );
290 int64_t float64_to_int64( float64 STATUS_PARAM );
291 int64_t float64_to_int64_round_to_zero( float64 STATUS_PARAM );
292 uint64_t float64_to_uint64( float64 STATUS_PARAM );
293 uint64_t float64_to_uint64_round_to_zero( float64 STATUS_PARAM );
294 float32 float64_to_float32( float64 STATUS_PARAM );
295 #ifdef FLOATX80
296 floatx80 float64_to_floatx80( float64 STATUS_PARAM );
297 #endif
298 #ifdef FLOAT128
299 float128 float64_to_float128( float64 STATUS_PARAM );
300 #endif
302 /*----------------------------------------------------------------------------
303 | Software IEC/IEEE double-precision operations.
304 *----------------------------------------------------------------------------*/
305 float64 float64_round_to_int( float64 STATUS_PARAM );
306 float64 float64_trunc_to_int( float64 STATUS_PARAM );
307 INLINE float64 float64_add( float64 a, float64 b STATUS_PARAM)
309 return a + b;
311 INLINE float64 float64_sub( float64 a, float64 b STATUS_PARAM)
313 return a - b;
315 INLINE float64 float64_mul( float64 a, float64 b STATUS_PARAM)
317 return a * b;
319 INLINE float64 float64_div( float64 a, float64 b STATUS_PARAM)
321 return a / b;
323 float64 float64_rem( float64, float64 STATUS_PARAM );
324 float64 float64_sqrt( float64 STATUS_PARAM );
325 INLINE int float64_eq( float64 a, float64 b STATUS_PARAM)
327 return a == b;
329 INLINE int float64_le( float64 a, float64 b STATUS_PARAM)
331 return a <= b;
333 INLINE int float64_lt( float64 a, float64 b STATUS_PARAM)
335 return a < b;
337 INLINE int float64_eq_signaling( float64 a, float64 b STATUS_PARAM)
339 return a <= b && a >= b;
341 INLINE int float64_le_quiet( float64 a, float64 b STATUS_PARAM)
343 return islessequal(a, b);
345 INLINE int float64_lt_quiet( float64 a, float64 b STATUS_PARAM)
347 return isless(a, b);
350 INLINE int float64_unordered( float64 a, float64 b STATUS_PARAM)
352 return isunordered(a, b);
355 int float64_compare( float64, float64 STATUS_PARAM );
356 int float64_compare_quiet( float64, float64 STATUS_PARAM );
357 int float64_is_signaling_nan( float64 );
358 int float64_is_nan( float64 );
360 INLINE float64 float64_abs(float64 a)
362 return fabs(a);
365 INLINE float64 float64_chs(float64 a)
367 return -a;
370 INLINE float64 float64_is_infinity(float64 a)
372 return fpclassify(a) == FP_INFINITE;
375 INLINE float64 float64_is_neg(float64 a)
377 float64u u;
378 u.f = a;
379 return u.i >> 63;
382 INLINE float64 float64_is_zero(float64 a)
384 return fpclassify(a) == FP_ZERO;
387 INLINE float64 float64_scalbn(float64 a, int n)
389 return scalbn(a, n);
392 #ifdef FLOATX80
394 /*----------------------------------------------------------------------------
395 | Software IEC/IEEE extended double-precision conversion routines.
396 *----------------------------------------------------------------------------*/
397 int floatx80_to_int32( floatx80 STATUS_PARAM );
398 int floatx80_to_int32_round_to_zero( floatx80 STATUS_PARAM );
399 int64_t floatx80_to_int64( floatx80 STATUS_PARAM);
400 int64_t floatx80_to_int64_round_to_zero( floatx80 STATUS_PARAM);
401 float32 floatx80_to_float32( floatx80 STATUS_PARAM );
402 float64 floatx80_to_float64( floatx80 STATUS_PARAM );
403 #ifdef FLOAT128
404 float128 floatx80_to_float128( floatx80 STATUS_PARAM );
405 #endif
407 /*----------------------------------------------------------------------------
408 | Software IEC/IEEE extended double-precision operations.
409 *----------------------------------------------------------------------------*/
410 floatx80 floatx80_round_to_int( floatx80 STATUS_PARAM );
411 INLINE floatx80 floatx80_add( floatx80 a, floatx80 b STATUS_PARAM)
413 return a + b;
415 INLINE floatx80 floatx80_sub( floatx80 a, floatx80 b STATUS_PARAM)
417 return a - b;
419 INLINE floatx80 floatx80_mul( floatx80 a, floatx80 b STATUS_PARAM)
421 return a * b;
423 INLINE floatx80 floatx80_div( floatx80 a, floatx80 b STATUS_PARAM)
425 return a / b;
427 floatx80 floatx80_rem( floatx80, floatx80 STATUS_PARAM );
428 floatx80 floatx80_sqrt( floatx80 STATUS_PARAM );
429 INLINE int floatx80_eq( floatx80 a, floatx80 b STATUS_PARAM)
431 return a == b;
433 INLINE int floatx80_le( floatx80 a, floatx80 b STATUS_PARAM)
435 return a <= b;
437 INLINE int floatx80_lt( floatx80 a, floatx80 b STATUS_PARAM)
439 return a < b;
441 INLINE int floatx80_eq_signaling( floatx80 a, floatx80 b STATUS_PARAM)
443 return a <= b && a >= b;
445 INLINE int floatx80_le_quiet( floatx80 a, floatx80 b STATUS_PARAM)
447 return islessequal(a, b);
449 INLINE int floatx80_lt_quiet( floatx80 a, floatx80 b STATUS_PARAM)
451 return isless(a, b);
454 INLINE int floatx80_unordered( floatx80 a, floatx80 b STATUS_PARAM)
456 return isunordered(a, b);
459 int floatx80_compare( floatx80, floatx80 STATUS_PARAM );
460 int floatx80_compare_quiet( floatx80, floatx80 STATUS_PARAM );
461 int floatx80_is_signaling_nan( floatx80 );
462 int floatx80_is_nan( floatx80 );
464 INLINE floatx80 floatx80_abs(floatx80 a)
466 return fabsl(a);
469 INLINE floatx80 floatx80_chs(floatx80 a)
471 return -a;
474 INLINE floatx80 floatx80_is_infinity(floatx80 a)
476 return fpclassify(a) == FP_INFINITE;
479 INLINE floatx80 floatx80_is_neg(floatx80 a)
481 floatx80u u;
482 u.f = a;
483 return u.i.high >> 15;
486 INLINE floatx80 floatx80_is_zero(floatx80 a)
488 return fpclassify(a) == FP_ZERO;
491 INLINE floatx80 floatx80_scalbn(floatx80 a, int n)
493 return scalbnl(a, n);
496 #endif