4 * The code in this source file is derived from release 2a of the SoftFloat
5 * IEC/IEEE Floating-point Arithmetic Package. Those parts of the code (and
6 * some later contributions) are provided under that license, as detailed below.
7 * It has subsequently been modified by contributors to the QEMU Project,
8 * so some portions are provided under:
9 * the SoftFloat-2a license
13 * Any future contributions to this file after December 1st 2014 will be
14 * taken to be licensed under the Softfloat-2a license unless specifically
15 * indicated otherwise.
19 ===============================================================================
20 This C header file is part of the SoftFloat IEC/IEEE Floating-point
21 Arithmetic Package, Release 2a.
23 Written by John R. Hauser. This work was made possible in part by the
24 International Computer Science Institute, located at Suite 600, 1947 Center
25 Street, Berkeley, California 94704. Funding was partially provided by the
26 National Science Foundation under grant MIP-9311980. The original version
27 of this code was written as part of a project to build a fixed-point vector
28 processor in collaboration with the University of California at Berkeley,
29 overseen by Profs. Nelson Morgan and John Wawrzynek. More information
30 is available through the Web page `http://HTTP.CS.Berkeley.EDU/~jhauser/
31 arithmetic/SoftFloat.html'.
33 THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort
34 has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT
35 TIMES RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO
36 PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY
37 AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE.
39 Derivative works are acceptable, even for commercial purposes, so long as
40 (1) they include prominent notice that the work is derivative, and (2) they
41 include prominent notice akin to these four paragraphs for those parts of
42 this code that are retained.
44 ===============================================================================
48 * Copyright (c) 2006, Fabrice Bellard
49 * All rights reserved.
51 * Redistribution and use in source and binary forms, with or without
52 * modification, are permitted provided that the following conditions are met:
54 * 1. Redistributions of source code must retain the above copyright notice,
55 * this list of conditions and the following disclaimer.
57 * 2. Redistributions in binary form must reproduce the above copyright notice,
58 * this list of conditions and the following disclaimer in the documentation
59 * and/or other materials provided with the distribution.
61 * 3. Neither the name of the copyright holder nor the names of its contributors
62 * may be used to endorse or promote products derived from this software without
63 * specific prior written permission.
65 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
66 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
67 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
68 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
69 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
70 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
71 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
72 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
73 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
74 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
75 * THE POSSIBILITY OF SUCH DAMAGE.
78 /* Portions of this work are licensed under the terms of the GNU GPL,
79 * version 2 or later. See the COPYING file in the top-level directory.
85 /*----------------------------------------------------------------------------
86 | Software IEC/IEEE floating-point ordering relations
87 *----------------------------------------------------------------------------*/
90 float_relation_less
= -1,
91 float_relation_equal
= 0,
92 float_relation_greater
= 1,
93 float_relation_unordered
= 2
96 #include "fpu/softfloat-types.h"
97 #include "fpu/softfloat-helpers.h"
98 #include "qemu/int128.h"
100 /*----------------------------------------------------------------------------
101 | Routine to raise any or all of the software IEC/IEEE floating-point
103 *----------------------------------------------------------------------------*/
104 static inline void float_raise(uint16_t flags
, float_status
*status
)
106 status
->float_exception_flags
|= flags
;
109 /*----------------------------------------------------------------------------
110 | If `a' is denormal and we are in flush-to-zero mode then set the
111 | input-denormal exception and return zero. Otherwise just return the value.
112 *----------------------------------------------------------------------------*/
113 float16
float16_squash_input_denormal(float16 a
, float_status
*status
);
114 float32
float32_squash_input_denormal(float32 a
, float_status
*status
);
115 float64
float64_squash_input_denormal(float64 a
, float_status
*status
);
116 bfloat16
bfloat16_squash_input_denormal(bfloat16 a
, float_status
*status
);
118 /*----------------------------------------------------------------------------
119 | Options to indicate which negations to perform in float*_muladd()
120 | Using these differs from negating an input or output before calling
121 | the muladd function in that this means that a NaN doesn't have its
122 | sign bit inverted before it is propagated.
123 | We also support halving the result before rounding, as a special
124 | case to support the ARM fused-sqrt-step instruction FRSQRTS.
125 *----------------------------------------------------------------------------*/
127 float_muladd_negate_c
= 1,
128 float_muladd_negate_product
= 2,
129 float_muladd_negate_result
= 4,
130 float_muladd_halve_result
= 8,
133 /*----------------------------------------------------------------------------
134 | Software IEC/IEEE integer-to-floating-point conversion routines.
135 *----------------------------------------------------------------------------*/
137 float16
int16_to_float16_scalbn(int16_t a
, int, float_status
*status
);
138 float16
int32_to_float16_scalbn(int32_t a
, int, float_status
*status
);
139 float16
int64_to_float16_scalbn(int64_t a
, int, float_status
*status
);
140 float16
uint16_to_float16_scalbn(uint16_t a
, int, float_status
*status
);
141 float16
uint32_to_float16_scalbn(uint32_t a
, int, float_status
*status
);
142 float16
uint64_to_float16_scalbn(uint64_t a
, int, float_status
*status
);
144 float16
int8_to_float16(int8_t a
, float_status
*status
);
145 float16
int16_to_float16(int16_t a
, float_status
*status
);
146 float16
int32_to_float16(int32_t a
, float_status
*status
);
147 float16
int64_to_float16(int64_t a
, float_status
*status
);
148 float16
uint8_to_float16(uint8_t a
, float_status
*status
);
149 float16
uint16_to_float16(uint16_t a
, float_status
*status
);
150 float16
uint32_to_float16(uint32_t a
, float_status
*status
);
151 float16
uint64_to_float16(uint64_t a
, float_status
*status
);
153 float32
int16_to_float32_scalbn(int16_t, int, float_status
*status
);
154 float32
int32_to_float32_scalbn(int32_t, int, float_status
*status
);
155 float32
int64_to_float32_scalbn(int64_t, int, float_status
*status
);
156 float32
uint16_to_float32_scalbn(uint16_t, int, float_status
*status
);
157 float32
uint32_to_float32_scalbn(uint32_t, int, float_status
*status
);
158 float32
uint64_to_float32_scalbn(uint64_t, int, float_status
*status
);
160 float32
int16_to_float32(int16_t, float_status
*status
);
161 float32
int32_to_float32(int32_t, float_status
*status
);
162 float32
int64_to_float32(int64_t, float_status
*status
);
163 float32
uint16_to_float32(uint16_t, float_status
*status
);
164 float32
uint32_to_float32(uint32_t, float_status
*status
);
165 float32
uint64_to_float32(uint64_t, float_status
*status
);
167 float64
int16_to_float64_scalbn(int16_t, int, float_status
*status
);
168 float64
int32_to_float64_scalbn(int32_t, int, float_status
*status
);
169 float64
int64_to_float64_scalbn(int64_t, int, float_status
*status
);
170 float64
uint16_to_float64_scalbn(uint16_t, int, float_status
*status
);
171 float64
uint32_to_float64_scalbn(uint32_t, int, float_status
*status
);
172 float64
uint64_to_float64_scalbn(uint64_t, int, float_status
*status
);
174 float64
int16_to_float64(int16_t, float_status
*status
);
175 float64
int32_to_float64(int32_t, float_status
*status
);
176 float64
int64_to_float64(int64_t, float_status
*status
);
177 float64
uint16_to_float64(uint16_t, float_status
*status
);
178 float64
uint32_to_float64(uint32_t, float_status
*status
);
179 float64
uint64_to_float64(uint64_t, float_status
*status
);
181 floatx80
int32_to_floatx80(int32_t, float_status
*status
);
182 floatx80
int64_to_floatx80(int64_t, float_status
*status
);
184 float128
int32_to_float128(int32_t, float_status
*status
);
185 float128
int64_to_float128(int64_t, float_status
*status
);
186 float128
int128_to_float128(Int128
, float_status
*status
);
187 float128
uint64_to_float128(uint64_t, float_status
*status
);
188 float128
uint128_to_float128(Int128
, float_status
*status
);
190 /*----------------------------------------------------------------------------
191 | Software half-precision conversion routines.
192 *----------------------------------------------------------------------------*/
194 float16
float32_to_float16(float32
, bool ieee
, float_status
*status
);
195 float32
float16_to_float32(float16
, bool ieee
, float_status
*status
);
196 float16
float64_to_float16(float64 a
, bool ieee
, float_status
*status
);
197 float64
float16_to_float64(float16 a
, bool ieee
, float_status
*status
);
199 int8_t float16_to_int8_scalbn(float16
, FloatRoundMode
, int,
200 float_status
*status
);
201 int16_t float16_to_int16_scalbn(float16
, FloatRoundMode
, int, float_status
*);
202 int32_t float16_to_int32_scalbn(float16
, FloatRoundMode
, int, float_status
*);
203 int64_t float16_to_int64_scalbn(float16
, FloatRoundMode
, int, float_status
*);
205 int8_t float16_to_int8(float16
, float_status
*status
);
206 int16_t float16_to_int16(float16
, float_status
*status
);
207 int32_t float16_to_int32(float16
, float_status
*status
);
208 int64_t float16_to_int64(float16
, float_status
*status
);
210 int16_t float16_to_int16_round_to_zero(float16
, float_status
*status
);
211 int32_t float16_to_int32_round_to_zero(float16
, float_status
*status
);
212 int64_t float16_to_int64_round_to_zero(float16
, float_status
*status
);
214 uint8_t float16_to_uint8_scalbn(float16 a
, FloatRoundMode
,
215 int, float_status
*status
);
216 uint16_t float16_to_uint16_scalbn(float16 a
, FloatRoundMode
,
217 int, float_status
*status
);
218 uint32_t float16_to_uint32_scalbn(float16 a
, FloatRoundMode
,
219 int, float_status
*status
);
220 uint64_t float16_to_uint64_scalbn(float16 a
, FloatRoundMode
,
221 int, float_status
*status
);
223 uint8_t float16_to_uint8(float16 a
, float_status
*status
);
224 uint16_t float16_to_uint16(float16 a
, float_status
*status
);
225 uint32_t float16_to_uint32(float16 a
, float_status
*status
);
226 uint64_t float16_to_uint64(float16 a
, float_status
*status
);
228 uint16_t float16_to_uint16_round_to_zero(float16 a
, float_status
*status
);
229 uint32_t float16_to_uint32_round_to_zero(float16 a
, float_status
*status
);
230 uint64_t float16_to_uint64_round_to_zero(float16 a
, float_status
*status
);
232 /*----------------------------------------------------------------------------
233 | Software half-precision operations.
234 *----------------------------------------------------------------------------*/
236 float16
float16_round_to_int(float16
, float_status
*status
);
237 float16
float16_add(float16
, float16
, float_status
*status
);
238 float16
float16_sub(float16
, float16
, float_status
*status
);
239 float16
float16_mul(float16
, float16
, float_status
*status
);
240 float16
float16_muladd(float16
, float16
, float16
, int, float_status
*status
);
241 float16
float16_div(float16
, float16
, float_status
*status
);
242 float16
float16_scalbn(float16
, int, float_status
*status
);
243 float16
float16_min(float16
, float16
, float_status
*status
);
244 float16
float16_max(float16
, float16
, float_status
*status
);
245 float16
float16_minnum(float16
, float16
, float_status
*status
);
246 float16
float16_maxnum(float16
, float16
, float_status
*status
);
247 float16
float16_minnummag(float16
, float16
, float_status
*status
);
248 float16
float16_maxnummag(float16
, float16
, float_status
*status
);
249 float16
float16_minimum_number(float16
, float16
, float_status
*status
);
250 float16
float16_maximum_number(float16
, float16
, float_status
*status
);
251 float16
float16_sqrt(float16
, float_status
*status
);
252 FloatRelation
float16_compare(float16
, float16
, float_status
*status
);
253 FloatRelation
float16_compare_quiet(float16
, float16
, float_status
*status
);
255 bool float16_is_quiet_nan(float16
, float_status
*status
);
256 bool float16_is_signaling_nan(float16
, float_status
*status
);
257 float16
float16_silence_nan(float16
, float_status
*status
);
259 static inline bool float16_is_any_nan(float16 a
)
261 return ((float16_val(a
) & ~0x8000) > 0x7c00);
264 static inline bool float16_is_neg(float16 a
)
266 return float16_val(a
) >> 15;
269 static inline bool float16_is_infinity(float16 a
)
271 return (float16_val(a
) & 0x7fff) == 0x7c00;
274 static inline bool float16_is_zero(float16 a
)
276 return (float16_val(a
) & 0x7fff) == 0;
279 static inline bool float16_is_zero_or_denormal(float16 a
)
281 return (float16_val(a
) & 0x7c00) == 0;
284 static inline bool float16_is_normal(float16 a
)
286 return (((float16_val(a
) >> 10) + 1) & 0x1f) >= 2;
289 static inline float16
float16_abs(float16 a
)
291 /* Note that abs does *not* handle NaN specially, nor does
292 * it flush denormal inputs to zero.
294 return make_float16(float16_val(a
) & 0x7fff);
297 static inline float16
float16_chs(float16 a
)
299 /* Note that chs does *not* handle NaN specially, nor does
300 * it flush denormal inputs to zero.
302 return make_float16(float16_val(a
) ^ 0x8000);
305 static inline float16
float16_set_sign(float16 a
, int sign
)
307 return make_float16((float16_val(a
) & 0x7fff) | (sign
<< 15));
310 static inline bool float16_eq(float16 a
, float16 b
, float_status
*s
)
312 return float16_compare(a
, b
, s
) == float_relation_equal
;
315 static inline bool float16_le(float16 a
, float16 b
, float_status
*s
)
317 return float16_compare(a
, b
, s
) <= float_relation_equal
;
320 static inline bool float16_lt(float16 a
, float16 b
, float_status
*s
)
322 return float16_compare(a
, b
, s
) < float_relation_equal
;
325 static inline bool float16_unordered(float16 a
, float16 b
, float_status
*s
)
327 return float16_compare(a
, b
, s
) == float_relation_unordered
;
330 static inline bool float16_eq_quiet(float16 a
, float16 b
, float_status
*s
)
332 return float16_compare_quiet(a
, b
, s
) == float_relation_equal
;
335 static inline bool float16_le_quiet(float16 a
, float16 b
, float_status
*s
)
337 return float16_compare_quiet(a
, b
, s
) <= float_relation_equal
;
340 static inline bool float16_lt_quiet(float16 a
, float16 b
, float_status
*s
)
342 return float16_compare_quiet(a
, b
, s
) < float_relation_equal
;
345 static inline bool float16_unordered_quiet(float16 a
, float16 b
,
348 return float16_compare_quiet(a
, b
, s
) == float_relation_unordered
;
351 #define float16_zero make_float16(0)
352 #define float16_half make_float16(0x3800)
353 #define float16_one make_float16(0x3c00)
354 #define float16_one_point_five make_float16(0x3e00)
355 #define float16_two make_float16(0x4000)
356 #define float16_three make_float16(0x4200)
357 #define float16_infinity make_float16(0x7c00)
359 /*----------------------------------------------------------------------------
360 | Software bfloat16 conversion routines.
361 *----------------------------------------------------------------------------*/
363 bfloat16
bfloat16_round_to_int(bfloat16
, float_status
*status
);
364 bfloat16
float32_to_bfloat16(float32
, float_status
*status
);
365 float32
bfloat16_to_float32(bfloat16
, float_status
*status
);
366 bfloat16
float64_to_bfloat16(float64 a
, float_status
*status
);
367 float64
bfloat16_to_float64(bfloat16 a
, float_status
*status
);
369 int8_t bfloat16_to_int8_scalbn(bfloat16
, FloatRoundMode
,
370 int, float_status
*status
);
371 int16_t bfloat16_to_int16_scalbn(bfloat16
, FloatRoundMode
,
372 int, float_status
*status
);
373 int32_t bfloat16_to_int32_scalbn(bfloat16
, FloatRoundMode
,
374 int, float_status
*status
);
375 int64_t bfloat16_to_int64_scalbn(bfloat16
, FloatRoundMode
,
376 int, float_status
*status
);
378 int8_t bfloat16_to_int8(bfloat16
, float_status
*status
);
379 int16_t bfloat16_to_int16(bfloat16
, float_status
*status
);
380 int32_t bfloat16_to_int32(bfloat16
, float_status
*status
);
381 int64_t bfloat16_to_int64(bfloat16
, float_status
*status
);
383 int8_t bfloat16_to_int8_round_to_zero(bfloat16
, float_status
*status
);
384 int16_t bfloat16_to_int16_round_to_zero(bfloat16
, float_status
*status
);
385 int32_t bfloat16_to_int32_round_to_zero(bfloat16
, float_status
*status
);
386 int64_t bfloat16_to_int64_round_to_zero(bfloat16
, float_status
*status
);
388 uint8_t bfloat16_to_uint8_scalbn(bfloat16 a
, FloatRoundMode
,
389 int, float_status
*status
);
390 uint16_t bfloat16_to_uint16_scalbn(bfloat16 a
, FloatRoundMode
,
391 int, float_status
*status
);
392 uint32_t bfloat16_to_uint32_scalbn(bfloat16 a
, FloatRoundMode
,
393 int, float_status
*status
);
394 uint64_t bfloat16_to_uint64_scalbn(bfloat16 a
, FloatRoundMode
,
395 int, float_status
*status
);
397 uint8_t bfloat16_to_uint8(bfloat16 a
, float_status
*status
);
398 uint16_t bfloat16_to_uint16(bfloat16 a
, float_status
*status
);
399 uint32_t bfloat16_to_uint32(bfloat16 a
, float_status
*status
);
400 uint64_t bfloat16_to_uint64(bfloat16 a
, float_status
*status
);
402 uint8_t bfloat16_to_uint8_round_to_zero(bfloat16 a
, float_status
*status
);
403 uint16_t bfloat16_to_uint16_round_to_zero(bfloat16 a
, float_status
*status
);
404 uint32_t bfloat16_to_uint32_round_to_zero(bfloat16 a
, float_status
*status
);
405 uint64_t bfloat16_to_uint64_round_to_zero(bfloat16 a
, float_status
*status
);
407 bfloat16
int8_to_bfloat16_scalbn(int8_t a
, int, float_status
*status
);
408 bfloat16
int16_to_bfloat16_scalbn(int16_t a
, int, float_status
*status
);
409 bfloat16
int32_to_bfloat16_scalbn(int32_t a
, int, float_status
*status
);
410 bfloat16
int64_to_bfloat16_scalbn(int64_t a
, int, float_status
*status
);
411 bfloat16
uint8_to_bfloat16_scalbn(uint8_t a
, int, float_status
*status
);
412 bfloat16
uint16_to_bfloat16_scalbn(uint16_t a
, int, float_status
*status
);
413 bfloat16
uint32_to_bfloat16_scalbn(uint32_t a
, int, float_status
*status
);
414 bfloat16
uint64_to_bfloat16_scalbn(uint64_t a
, int, float_status
*status
);
416 bfloat16
int8_to_bfloat16(int8_t a
, float_status
*status
);
417 bfloat16
int16_to_bfloat16(int16_t a
, float_status
*status
);
418 bfloat16
int32_to_bfloat16(int32_t a
, float_status
*status
);
419 bfloat16
int64_to_bfloat16(int64_t a
, float_status
*status
);
420 bfloat16
uint8_to_bfloat16(uint8_t a
, float_status
*status
);
421 bfloat16
uint16_to_bfloat16(uint16_t a
, float_status
*status
);
422 bfloat16
uint32_to_bfloat16(uint32_t a
, float_status
*status
);
423 bfloat16
uint64_to_bfloat16(uint64_t a
, float_status
*status
);
425 /*----------------------------------------------------------------------------
426 | Software bfloat16 operations.
427 *----------------------------------------------------------------------------*/
429 bfloat16
bfloat16_add(bfloat16
, bfloat16
, float_status
*status
);
430 bfloat16
bfloat16_sub(bfloat16
, bfloat16
, float_status
*status
);
431 bfloat16
bfloat16_mul(bfloat16
, bfloat16
, float_status
*status
);
432 bfloat16
bfloat16_div(bfloat16
, bfloat16
, float_status
*status
);
433 bfloat16
bfloat16_muladd(bfloat16
, bfloat16
, bfloat16
, int,
434 float_status
*status
);
435 float16
bfloat16_scalbn(bfloat16
, int, float_status
*status
);
436 bfloat16
bfloat16_min(bfloat16
, bfloat16
, float_status
*status
);
437 bfloat16
bfloat16_max(bfloat16
, bfloat16
, float_status
*status
);
438 bfloat16
bfloat16_minnum(bfloat16
, bfloat16
, float_status
*status
);
439 bfloat16
bfloat16_maxnum(bfloat16
, bfloat16
, float_status
*status
);
440 bfloat16
bfloat16_minnummag(bfloat16
, bfloat16
, float_status
*status
);
441 bfloat16
bfloat16_maxnummag(bfloat16
, bfloat16
, float_status
*status
);
442 bfloat16
bfloat16_minimum_number(bfloat16
, bfloat16
, float_status
*status
);
443 bfloat16
bfloat16_maximum_number(bfloat16
, bfloat16
, float_status
*status
);
444 bfloat16
bfloat16_sqrt(bfloat16
, float_status
*status
);
445 FloatRelation
bfloat16_compare(bfloat16
, bfloat16
, float_status
*status
);
446 FloatRelation
bfloat16_compare_quiet(bfloat16
, bfloat16
, float_status
*status
);
448 bool bfloat16_is_quiet_nan(bfloat16
, float_status
*status
);
449 bool bfloat16_is_signaling_nan(bfloat16
, float_status
*status
);
450 bfloat16
bfloat16_silence_nan(bfloat16
, float_status
*status
);
451 bfloat16
bfloat16_default_nan(float_status
*status
);
453 static inline bool bfloat16_is_any_nan(bfloat16 a
)
455 return ((a
& ~0x8000) > 0x7F80);
458 static inline bool bfloat16_is_neg(bfloat16 a
)
463 static inline bool bfloat16_is_infinity(bfloat16 a
)
465 return (a
& 0x7fff) == 0x7F80;
468 static inline bool bfloat16_is_zero(bfloat16 a
)
470 return (a
& 0x7fff) == 0;
473 static inline bool bfloat16_is_zero_or_denormal(bfloat16 a
)
475 return (a
& 0x7F80) == 0;
478 static inline bool bfloat16_is_normal(bfloat16 a
)
480 return (((a
>> 7) + 1) & 0xff) >= 2;
483 static inline bfloat16
bfloat16_abs(bfloat16 a
)
485 /* Note that abs does *not* handle NaN specially, nor does
486 * it flush denormal inputs to zero.
491 static inline bfloat16
bfloat16_chs(bfloat16 a
)
493 /* Note that chs does *not* handle NaN specially, nor does
494 * it flush denormal inputs to zero.
499 static inline bfloat16
bfloat16_set_sign(bfloat16 a
, int sign
)
501 return (a
& 0x7fff) | (sign
<< 15);
504 static inline bool bfloat16_eq(bfloat16 a
, bfloat16 b
, float_status
*s
)
506 return bfloat16_compare(a
, b
, s
) == float_relation_equal
;
509 static inline bool bfloat16_le(bfloat16 a
, bfloat16 b
, float_status
*s
)
511 return bfloat16_compare(a
, b
, s
) <= float_relation_equal
;
514 static inline bool bfloat16_lt(bfloat16 a
, bfloat16 b
, float_status
*s
)
516 return bfloat16_compare(a
, b
, s
) < float_relation_equal
;
519 static inline bool bfloat16_unordered(bfloat16 a
, bfloat16 b
, float_status
*s
)
521 return bfloat16_compare(a
, b
, s
) == float_relation_unordered
;
524 static inline bool bfloat16_eq_quiet(bfloat16 a
, bfloat16 b
, float_status
*s
)
526 return bfloat16_compare_quiet(a
, b
, s
) == float_relation_equal
;
529 static inline bool bfloat16_le_quiet(bfloat16 a
, bfloat16 b
, float_status
*s
)
531 return bfloat16_compare_quiet(a
, b
, s
) <= float_relation_equal
;
534 static inline bool bfloat16_lt_quiet(bfloat16 a
, bfloat16 b
, float_status
*s
)
536 return bfloat16_compare_quiet(a
, b
, s
) < float_relation_equal
;
539 static inline bool bfloat16_unordered_quiet(bfloat16 a
, bfloat16 b
,
542 return bfloat16_compare_quiet(a
, b
, s
) == float_relation_unordered
;
545 #define bfloat16_zero 0
546 #define bfloat16_half 0x3f00
547 #define bfloat16_one 0x3f80
548 #define bfloat16_one_point_five 0x3fc0
549 #define bfloat16_two 0x4000
550 #define bfloat16_three 0x4040
551 #define bfloat16_infinity 0x7f80
553 /*----------------------------------------------------------------------------
554 | The pattern for a default generated half-precision NaN.
555 *----------------------------------------------------------------------------*/
556 float16
float16_default_nan(float_status
*status
);
558 /*----------------------------------------------------------------------------
559 | Software IEC/IEEE single-precision conversion routines.
560 *----------------------------------------------------------------------------*/
562 int16_t float32_to_int16_scalbn(float32
, FloatRoundMode
, int, float_status
*);
563 int32_t float32_to_int32_scalbn(float32
, FloatRoundMode
, int, float_status
*);
564 int64_t float32_to_int64_scalbn(float32
, FloatRoundMode
, int, float_status
*);
566 int16_t float32_to_int16(float32
, float_status
*status
);
567 int32_t float32_to_int32(float32
, float_status
*status
);
568 int64_t float32_to_int64(float32
, float_status
*status
);
570 int16_t float32_to_int16_round_to_zero(float32
, float_status
*status
);
571 int32_t float32_to_int32_round_to_zero(float32
, float_status
*status
);
572 int64_t float32_to_int64_round_to_zero(float32
, float_status
*status
);
574 uint16_t float32_to_uint16_scalbn(float32
, FloatRoundMode
, int, float_status
*);
575 uint32_t float32_to_uint32_scalbn(float32
, FloatRoundMode
, int, float_status
*);
576 uint64_t float32_to_uint64_scalbn(float32
, FloatRoundMode
, int, float_status
*);
578 uint16_t float32_to_uint16(float32
, float_status
*status
);
579 uint32_t float32_to_uint32(float32
, float_status
*status
);
580 uint64_t float32_to_uint64(float32
, float_status
*status
);
582 uint16_t float32_to_uint16_round_to_zero(float32
, float_status
*status
);
583 uint32_t float32_to_uint32_round_to_zero(float32
, float_status
*status
);
584 uint64_t float32_to_uint64_round_to_zero(float32
, float_status
*status
);
586 float64
float32_to_float64(float32
, float_status
*status
);
587 floatx80
float32_to_floatx80(float32
, float_status
*status
);
588 float128
float32_to_float128(float32
, float_status
*status
);
590 /*----------------------------------------------------------------------------
591 | Software IEC/IEEE single-precision operations.
592 *----------------------------------------------------------------------------*/
593 float32
float32_round_to_int(float32
, float_status
*status
);
594 float32
float32_add(float32
, float32
, float_status
*status
);
595 float32
float32_sub(float32
, float32
, float_status
*status
);
596 float32
float32_mul(float32
, float32
, float_status
*status
);
597 float32
float32_div(float32
, float32
, float_status
*status
);
598 float32
float32_rem(float32
, float32
, float_status
*status
);
599 float32
float32_muladd(float32
, float32
, float32
, int, float_status
*status
);
600 float32
float32_sqrt(float32
, float_status
*status
);
601 float32
float32_exp2(float32
, float_status
*status
);
602 float32
float32_log2(float32
, float_status
*status
);
603 FloatRelation
float32_compare(float32
, float32
, float_status
*status
);
604 FloatRelation
float32_compare_quiet(float32
, float32
, float_status
*status
);
605 float32
float32_min(float32
, float32
, float_status
*status
);
606 float32
float32_max(float32
, float32
, float_status
*status
);
607 float32
float32_minnum(float32
, float32
, float_status
*status
);
608 float32
float32_maxnum(float32
, float32
, float_status
*status
);
609 float32
float32_minnummag(float32
, float32
, float_status
*status
);
610 float32
float32_maxnummag(float32
, float32
, float_status
*status
);
611 float32
float32_minimum_number(float32
, float32
, float_status
*status
);
612 float32
float32_maximum_number(float32
, float32
, float_status
*status
);
613 bool float32_is_quiet_nan(float32
, float_status
*status
);
614 bool float32_is_signaling_nan(float32
, float_status
*status
);
615 float32
float32_silence_nan(float32
, float_status
*status
);
616 float32
float32_scalbn(float32
, int, float_status
*status
);
618 static inline float32
float32_abs(float32 a
)
620 /* Note that abs does *not* handle NaN specially, nor does
621 * it flush denormal inputs to zero.
623 return make_float32(float32_val(a
) & 0x7fffffff);
626 static inline float32
float32_chs(float32 a
)
628 /* Note that chs does *not* handle NaN specially, nor does
629 * it flush denormal inputs to zero.
631 return make_float32(float32_val(a
) ^ 0x80000000);
634 static inline bool float32_is_infinity(float32 a
)
636 return (float32_val(a
) & 0x7fffffff) == 0x7f800000;
639 static inline bool float32_is_neg(float32 a
)
641 return float32_val(a
) >> 31;
644 static inline bool float32_is_zero(float32 a
)
646 return (float32_val(a
) & 0x7fffffff) == 0;
649 static inline bool float32_is_any_nan(float32 a
)
651 return ((float32_val(a
) & ~(1 << 31)) > 0x7f800000UL
);
654 static inline bool float32_is_zero_or_denormal(float32 a
)
656 return (float32_val(a
) & 0x7f800000) == 0;
659 static inline bool float32_is_normal(float32 a
)
661 return (((float32_val(a
) >> 23) + 1) & 0xff) >= 2;
664 static inline bool float32_is_denormal(float32 a
)
666 return float32_is_zero_or_denormal(a
) && !float32_is_zero(a
);
669 static inline bool float32_is_zero_or_normal(float32 a
)
671 return float32_is_normal(a
) || float32_is_zero(a
);
674 static inline float32
float32_set_sign(float32 a
, int sign
)
676 return make_float32((float32_val(a
) & 0x7fffffff) | (sign
<< 31));
679 static inline bool float32_eq(float32 a
, float32 b
, float_status
*s
)
681 return float32_compare(a
, b
, s
) == float_relation_equal
;
684 static inline bool float32_le(float32 a
, float32 b
, float_status
*s
)
686 return float32_compare(a
, b
, s
) <= float_relation_equal
;
689 static inline bool float32_lt(float32 a
, float32 b
, float_status
*s
)
691 return float32_compare(a
, b
, s
) < float_relation_equal
;
694 static inline bool float32_unordered(float32 a
, float32 b
, float_status
*s
)
696 return float32_compare(a
, b
, s
) == float_relation_unordered
;
699 static inline bool float32_eq_quiet(float32 a
, float32 b
, float_status
*s
)
701 return float32_compare_quiet(a
, b
, s
) == float_relation_equal
;
704 static inline bool float32_le_quiet(float32 a
, float32 b
, float_status
*s
)
706 return float32_compare_quiet(a
, b
, s
) <= float_relation_equal
;
709 static inline bool float32_lt_quiet(float32 a
, float32 b
, float_status
*s
)
711 return float32_compare_quiet(a
, b
, s
) < float_relation_equal
;
714 static inline bool float32_unordered_quiet(float32 a
, float32 b
,
717 return float32_compare_quiet(a
, b
, s
) == float_relation_unordered
;
720 #define float32_zero make_float32(0)
721 #define float32_half make_float32(0x3f000000)
722 #define float32_one make_float32(0x3f800000)
723 #define float32_one_point_five make_float32(0x3fc00000)
724 #define float32_two make_float32(0x40000000)
725 #define float32_three make_float32(0x40400000)
726 #define float32_infinity make_float32(0x7f800000)
728 /*----------------------------------------------------------------------------
729 | Packs the sign `zSign', exponent `zExp', and significand `zSig' into a
730 | single-precision floating-point value, returning the result. After being
731 | shifted into the proper positions, the three fields are simply added
732 | together to form the result. This means that any integer portion of `zSig'
733 | will be added into the exponent. Since a properly normalized significand
734 | will have an integer portion equal to 1, the `zExp' input should be 1 less
735 | than the desired result exponent whenever `zSig' is a complete, normalized
737 *----------------------------------------------------------------------------*/
739 static inline float32
packFloat32(bool zSign
, int zExp
, uint32_t zSig
)
742 (((uint32_t)zSign
) << 31) + (((uint32_t)zExp
) << 23) + zSig
);
745 /*----------------------------------------------------------------------------
746 | The pattern for a default generated single-precision NaN.
747 *----------------------------------------------------------------------------*/
748 float32
float32_default_nan(float_status
*status
);
750 /*----------------------------------------------------------------------------
751 | Software IEC/IEEE double-precision conversion routines.
752 *----------------------------------------------------------------------------*/
754 int16_t float64_to_int16_scalbn(float64
, FloatRoundMode
, int, float_status
*);
755 int32_t float64_to_int32_scalbn(float64
, FloatRoundMode
, int, float_status
*);
756 int64_t float64_to_int64_scalbn(float64
, FloatRoundMode
, int, float_status
*);
758 int16_t float64_to_int16(float64
, float_status
*status
);
759 int32_t float64_to_int32(float64
, float_status
*status
);
760 int64_t float64_to_int64(float64
, float_status
*status
);
762 int16_t float64_to_int16_round_to_zero(float64
, float_status
*status
);
763 int32_t float64_to_int32_round_to_zero(float64
, float_status
*status
);
764 int64_t float64_to_int64_round_to_zero(float64
, float_status
*status
);
766 int32_t float64_to_int32_modulo(float64
, FloatRoundMode
, float_status
*status
);
767 int64_t float64_to_int64_modulo(float64
, FloatRoundMode
, float_status
*status
);
769 uint16_t float64_to_uint16_scalbn(float64
, FloatRoundMode
, int, float_status
*);
770 uint32_t float64_to_uint32_scalbn(float64
, FloatRoundMode
, int, float_status
*);
771 uint64_t float64_to_uint64_scalbn(float64
, FloatRoundMode
, int, float_status
*);
773 uint16_t float64_to_uint16(float64
, float_status
*status
);
774 uint32_t float64_to_uint32(float64
, float_status
*status
);
775 uint64_t float64_to_uint64(float64
, float_status
*status
);
777 uint16_t float64_to_uint16_round_to_zero(float64
, float_status
*status
);
778 uint32_t float64_to_uint32_round_to_zero(float64
, float_status
*status
);
779 uint64_t float64_to_uint64_round_to_zero(float64
, float_status
*status
);
781 float32
float64_to_float32(float64
, float_status
*status
);
782 floatx80
float64_to_floatx80(float64
, float_status
*status
);
783 float128
float64_to_float128(float64
, float_status
*status
);
785 /*----------------------------------------------------------------------------
786 | Software IEC/IEEE double-precision operations.
787 *----------------------------------------------------------------------------*/
788 float64
float64_round_to_int(float64
, float_status
*status
);
789 float64
float64_add(float64
, float64
, float_status
*status
);
790 float64
float64_sub(float64
, float64
, float_status
*status
);
791 float64
float64_mul(float64
, float64
, float_status
*status
);
792 float64
float64_div(float64
, float64
, float_status
*status
);
793 float64
float64_rem(float64
, float64
, float_status
*status
);
794 float64
float64_muladd(float64
, float64
, float64
, int, float_status
*status
);
795 float64
float64_sqrt(float64
, float_status
*status
);
796 float64
float64_log2(float64
, float_status
*status
);
797 FloatRelation
float64_compare(float64
, float64
, float_status
*status
);
798 FloatRelation
float64_compare_quiet(float64
, float64
, float_status
*status
);
799 float64
float64_min(float64
, float64
, float_status
*status
);
800 float64
float64_max(float64
, float64
, float_status
*status
);
801 float64
float64_minnum(float64
, float64
, float_status
*status
);
802 float64
float64_maxnum(float64
, float64
, float_status
*status
);
803 float64
float64_minnummag(float64
, float64
, float_status
*status
);
804 float64
float64_maxnummag(float64
, float64
, float_status
*status
);
805 float64
float64_minimum_number(float64
, float64
, float_status
*status
);
806 float64
float64_maximum_number(float64
, float64
, float_status
*status
);
807 bool float64_is_quiet_nan(float64 a
, float_status
*status
);
808 bool float64_is_signaling_nan(float64
, float_status
*status
);
809 float64
float64_silence_nan(float64
, float_status
*status
);
810 float64
float64_scalbn(float64
, int, float_status
*status
);
812 static inline float64
float64_abs(float64 a
)
814 /* Note that abs does *not* handle NaN specially, nor does
815 * it flush denormal inputs to zero.
817 return make_float64(float64_val(a
) & 0x7fffffffffffffffLL
);
820 static inline float64
float64_chs(float64 a
)
822 /* Note that chs does *not* handle NaN specially, nor does
823 * it flush denormal inputs to zero.
825 return make_float64(float64_val(a
) ^ 0x8000000000000000LL
);
828 static inline bool float64_is_infinity(float64 a
)
830 return (float64_val(a
) & 0x7fffffffffffffffLL
) == 0x7ff0000000000000LL
;
833 static inline bool float64_is_neg(float64 a
)
835 return float64_val(a
) >> 63;
838 static inline bool float64_is_zero(float64 a
)
840 return (float64_val(a
) & 0x7fffffffffffffffLL
) == 0;
843 static inline bool float64_is_any_nan(float64 a
)
845 return ((float64_val(a
) & ~(1ULL << 63)) > 0x7ff0000000000000ULL
);
848 static inline bool float64_is_zero_or_denormal(float64 a
)
850 return (float64_val(a
) & 0x7ff0000000000000LL
) == 0;
853 static inline bool float64_is_normal(float64 a
)
855 return (((float64_val(a
) >> 52) + 1) & 0x7ff) >= 2;
858 static inline bool float64_is_denormal(float64 a
)
860 return float64_is_zero_or_denormal(a
) && !float64_is_zero(a
);
863 static inline bool float64_is_zero_or_normal(float64 a
)
865 return float64_is_normal(a
) || float64_is_zero(a
);
868 static inline float64
float64_set_sign(float64 a
, int sign
)
870 return make_float64((float64_val(a
) & 0x7fffffffffffffffULL
)
871 | ((int64_t)sign
<< 63));
874 static inline bool float64_eq(float64 a
, float64 b
, float_status
*s
)
876 return float64_compare(a
, b
, s
) == float_relation_equal
;
879 static inline bool float64_le(float64 a
, float64 b
, float_status
*s
)
881 return float64_compare(a
, b
, s
) <= float_relation_equal
;
884 static inline bool float64_lt(float64 a
, float64 b
, float_status
*s
)
886 return float64_compare(a
, b
, s
) < float_relation_equal
;
889 static inline bool float64_unordered(float64 a
, float64 b
, float_status
*s
)
891 return float64_compare(a
, b
, s
) == float_relation_unordered
;
894 static inline bool float64_eq_quiet(float64 a
, float64 b
, float_status
*s
)
896 return float64_compare_quiet(a
, b
, s
) == float_relation_equal
;
899 static inline bool float64_le_quiet(float64 a
, float64 b
, float_status
*s
)
901 return float64_compare_quiet(a
, b
, s
) <= float_relation_equal
;
904 static inline bool float64_lt_quiet(float64 a
, float64 b
, float_status
*s
)
906 return float64_compare_quiet(a
, b
, s
) < float_relation_equal
;
909 static inline bool float64_unordered_quiet(float64 a
, float64 b
,
912 return float64_compare_quiet(a
, b
, s
) == float_relation_unordered
;
915 #define float64_zero make_float64(0)
916 #define float64_half make_float64(0x3fe0000000000000LL)
917 #define float64_one make_float64(0x3ff0000000000000LL)
918 #define float64_one_point_five make_float64(0x3FF8000000000000ULL)
919 #define float64_two make_float64(0x4000000000000000ULL)
920 #define float64_three make_float64(0x4008000000000000ULL)
921 #define float64_ln2 make_float64(0x3fe62e42fefa39efLL)
922 #define float64_infinity make_float64(0x7ff0000000000000LL)
924 /*----------------------------------------------------------------------------
925 | The pattern for a default generated double-precision NaN.
926 *----------------------------------------------------------------------------*/
927 float64
float64_default_nan(float_status
*status
);
929 /*----------------------------------------------------------------------------
930 | Software IEC/IEEE double-precision operations, rounding to single precision,
931 | returning a result in double precision, with only one rounding step.
932 *----------------------------------------------------------------------------*/
934 float64
float64r32_add(float64
, float64
, float_status
*status
);
935 float64
float64r32_sub(float64
, float64
, float_status
*status
);
936 float64
float64r32_mul(float64
, float64
, float_status
*status
);
937 float64
float64r32_div(float64
, float64
, float_status
*status
);
938 float64
float64r32_muladd(float64
, float64
, float64
, int, float_status
*status
);
939 float64
float64r32_sqrt(float64
, float_status
*status
);
941 /*----------------------------------------------------------------------------
942 | Software IEC/IEEE extended double-precision conversion routines.
943 *----------------------------------------------------------------------------*/
944 int32_t floatx80_to_int32(floatx80
, float_status
*status
);
945 int32_t floatx80_to_int32_round_to_zero(floatx80
, float_status
*status
);
946 int64_t floatx80_to_int64(floatx80
, float_status
*status
);
947 int64_t floatx80_to_int64_round_to_zero(floatx80
, float_status
*status
);
948 float32
floatx80_to_float32(floatx80
, float_status
*status
);
949 float64
floatx80_to_float64(floatx80
, float_status
*status
);
950 float128
floatx80_to_float128(floatx80
, float_status
*status
);
952 /*----------------------------------------------------------------------------
953 | The pattern for an extended double-precision inf.
954 *----------------------------------------------------------------------------*/
955 extern const floatx80 floatx80_infinity
;
957 /*----------------------------------------------------------------------------
958 | Software IEC/IEEE extended double-precision operations.
959 *----------------------------------------------------------------------------*/
960 floatx80
floatx80_round(floatx80 a
, float_status
*status
);
961 floatx80
floatx80_round_to_int(floatx80
, float_status
*status
);
962 floatx80
floatx80_add(floatx80
, floatx80
, float_status
*status
);
963 floatx80
floatx80_sub(floatx80
, floatx80
, float_status
*status
);
964 floatx80
floatx80_mul(floatx80
, floatx80
, float_status
*status
);
965 floatx80
floatx80_div(floatx80
, floatx80
, float_status
*status
);
966 floatx80
floatx80_modrem(floatx80
, floatx80
, bool, uint64_t *,
967 float_status
*status
);
968 floatx80
floatx80_mod(floatx80
, floatx80
, float_status
*status
);
969 floatx80
floatx80_rem(floatx80
, floatx80
, float_status
*status
);
970 floatx80
floatx80_sqrt(floatx80
, float_status
*status
);
971 FloatRelation
floatx80_compare(floatx80
, floatx80
, float_status
*status
);
972 FloatRelation
floatx80_compare_quiet(floatx80
, floatx80
, float_status
*status
);
973 int floatx80_is_quiet_nan(floatx80
, float_status
*status
);
974 int floatx80_is_signaling_nan(floatx80
, float_status
*status
);
975 floatx80
floatx80_silence_nan(floatx80
, float_status
*status
);
976 floatx80
floatx80_scalbn(floatx80
, int, float_status
*status
);
978 static inline floatx80
floatx80_abs(floatx80 a
)
984 static inline floatx80
floatx80_chs(floatx80 a
)
990 static inline bool floatx80_is_infinity(floatx80 a
)
992 #if defined(TARGET_M68K)
993 return (a
.high
& 0x7fff) == floatx80_infinity
.high
&& !(a
.low
<< 1);
995 return (a
.high
& 0x7fff) == floatx80_infinity
.high
&&
996 a
.low
== floatx80_infinity
.low
;
1000 static inline bool floatx80_is_neg(floatx80 a
)
1002 return a
.high
>> 15;
1005 static inline bool floatx80_is_zero(floatx80 a
)
1007 return (a
.high
& 0x7fff) == 0 && a
.low
== 0;
1010 static inline bool floatx80_is_zero_or_denormal(floatx80 a
)
1012 return (a
.high
& 0x7fff) == 0;
1015 static inline bool floatx80_is_any_nan(floatx80 a
)
1017 return ((a
.high
& 0x7fff) == 0x7fff) && (a
.low
<<1);
1020 static inline bool floatx80_eq(floatx80 a
, floatx80 b
, float_status
*s
)
1022 return floatx80_compare(a
, b
, s
) == float_relation_equal
;
1025 static inline bool floatx80_le(floatx80 a
, floatx80 b
, float_status
*s
)
1027 return floatx80_compare(a
, b
, s
) <= float_relation_equal
;
1030 static inline bool floatx80_lt(floatx80 a
, floatx80 b
, float_status
*s
)
1032 return floatx80_compare(a
, b
, s
) < float_relation_equal
;
1035 static inline bool floatx80_unordered(floatx80 a
, floatx80 b
, float_status
*s
)
1037 return floatx80_compare(a
, b
, s
) == float_relation_unordered
;
1040 static inline bool floatx80_eq_quiet(floatx80 a
, floatx80 b
, float_status
*s
)
1042 return floatx80_compare_quiet(a
, b
, s
) == float_relation_equal
;
1045 static inline bool floatx80_le_quiet(floatx80 a
, floatx80 b
, float_status
*s
)
1047 return floatx80_compare_quiet(a
, b
, s
) <= float_relation_equal
;
1050 static inline bool floatx80_lt_quiet(floatx80 a
, floatx80 b
, float_status
*s
)
1052 return floatx80_compare_quiet(a
, b
, s
) < float_relation_equal
;
1055 static inline bool floatx80_unordered_quiet(floatx80 a
, floatx80 b
,
1058 return floatx80_compare_quiet(a
, b
, s
) == float_relation_unordered
;
1061 /*----------------------------------------------------------------------------
1062 | Return whether the given value is an invalid floatx80 encoding.
1063 | Invalid floatx80 encodings arise when the integer bit is not set, but
1064 | the exponent is not zero. The only times the integer bit is permitted to
1065 | be zero is in subnormal numbers and the value zero.
1066 | This includes what the Intel software developer's manual calls pseudo-NaNs,
1067 | pseudo-infinities and un-normal numbers. It does not include
1068 | pseudo-denormals, which must still be correctly handled as inputs even
1069 | if they are never generated as outputs.
1070 *----------------------------------------------------------------------------*/
1071 static inline bool floatx80_invalid_encoding(floatx80 a
)
1073 #if defined(TARGET_M68K)
1074 /*-------------------------------------------------------------------------
1075 | With m68k, the explicit integer bit can be zero in the case of:
1076 | - zeros (exp == 0, mantissa == 0)
1077 | - denormalized numbers (exp == 0, mantissa != 0)
1078 | - unnormalized numbers (exp != 0, exp < 0x7FFF)
1079 | - infinities (exp == 0x7FFF, mantissa == 0)
1080 | - not-a-numbers (exp == 0x7FFF, mantissa != 0)
1082 | For infinities and NaNs, the explicit integer bit can be either one or
1085 | The IEEE 754 standard does not define a zero integer bit. Such a number
1086 | is an unnormalized number. Hardware does not directly support
1087 | denormalized and unnormalized numbers, but implicitly supports them by
1088 | trapping them as unimplemented data types, allowing efficient conversion
1091 | See "M68000 FAMILY PROGRAMMER’S REFERENCE MANUAL",
1092 | "1.6 FLOATING-POINT DATA TYPES"
1093 *------------------------------------------------------------------------*/
1096 return (a
.low
& (1ULL << 63)) == 0 && (a
.high
& 0x7FFF) != 0;
1100 #define floatx80_zero make_floatx80(0x0000, 0x0000000000000000LL)
1101 #define floatx80_zero_init make_floatx80_init(0x0000, 0x0000000000000000LL)
1102 #define floatx80_one make_floatx80(0x3fff, 0x8000000000000000LL)
1103 #define floatx80_ln2 make_floatx80(0x3ffe, 0xb17217f7d1cf79acLL)
1104 #define floatx80_pi make_floatx80(0x4000, 0xc90fdaa22168c235LL)
1105 #define floatx80_half make_floatx80(0x3ffe, 0x8000000000000000LL)
1107 /*----------------------------------------------------------------------------
1108 | Returns the fraction bits of the extended double-precision floating-point
1110 *----------------------------------------------------------------------------*/
1112 static inline uint64_t extractFloatx80Frac(floatx80 a
)
1117 /*----------------------------------------------------------------------------
1118 | Returns the exponent bits of the extended double-precision floating-point
1120 *----------------------------------------------------------------------------*/
1122 static inline int32_t extractFloatx80Exp(floatx80 a
)
1124 return a
.high
& 0x7FFF;
1127 /*----------------------------------------------------------------------------
1128 | Returns the sign bit of the extended double-precision floating-point value
1130 *----------------------------------------------------------------------------*/
1132 static inline bool extractFloatx80Sign(floatx80 a
)
1134 return a
.high
>> 15;
1137 /*----------------------------------------------------------------------------
1138 | Packs the sign `zSign', exponent `zExp', and significand `zSig' into an
1139 | extended double-precision floating-point value, returning the result.
1140 *----------------------------------------------------------------------------*/
1142 static inline floatx80
packFloatx80(bool zSign
, int32_t zExp
, uint64_t zSig
)
1147 z
.high
= (((uint16_t)zSign
) << 15) + zExp
;
1151 /*----------------------------------------------------------------------------
1152 | Normalizes the subnormal extended double-precision floating-point value
1153 | represented by the denormalized significand `aSig'. The normalized exponent
1154 | and significand are stored at the locations pointed to by `zExpPtr' and
1155 | `zSigPtr', respectively.
1156 *----------------------------------------------------------------------------*/
1158 void normalizeFloatx80Subnormal(uint64_t aSig
, int32_t *zExpPtr
,
1161 /*----------------------------------------------------------------------------
1162 | Takes two extended double-precision floating-point values `a' and `b', one
1163 | of which is a NaN, and returns the appropriate NaN result. If either `a' or
1164 | `b' is a signaling NaN, the invalid exception is raised.
1165 *----------------------------------------------------------------------------*/
1167 floatx80
propagateFloatx80NaN(floatx80 a
, floatx80 b
, float_status
*status
);
1169 /*----------------------------------------------------------------------------
1170 | Takes an abstract floating-point value having sign `zSign', exponent `zExp',
1171 | and extended significand formed by the concatenation of `zSig0' and `zSig1',
1172 | and returns the proper extended double-precision floating-point value
1173 | corresponding to the abstract input. Ordinarily, the abstract value is
1174 | rounded and packed into the extended double-precision format, with the
1175 | inexact exception raised if the abstract input cannot be represented
1176 | exactly. However, if the abstract value is too large, the overflow and
1177 | inexact exceptions are raised and an infinity or maximal finite value is
1178 | returned. If the abstract value is too small, the input value is rounded to
1179 | a subnormal number, and the underflow and inexact exceptions are raised if
1180 | the abstract input cannot be represented exactly as a subnormal extended
1181 | double-precision floating-point number.
1182 | If `roundingPrecision' is 32 or 64, the result is rounded to the same
1183 | number of bits as single or double precision, respectively. Otherwise, the
1184 | result is rounded to the full precision of the extended double-precision
1186 | The input significand must be normalized or smaller. If the input
1187 | significand is not normalized, `zExp' must be 0; in that case, the result
1188 | returned is a subnormal number, and it must not require rounding. The
1189 | handling of underflow and overflow follows the IEC/IEEE Standard for Binary
1190 | Floating-Point Arithmetic.
1191 *----------------------------------------------------------------------------*/
1193 floatx80
roundAndPackFloatx80(FloatX80RoundPrec roundingPrecision
, bool zSign
,
1194 int32_t zExp
, uint64_t zSig0
, uint64_t zSig1
,
1195 float_status
*status
);
1197 /*----------------------------------------------------------------------------
1198 | Takes an abstract floating-point value having sign `zSign', exponent
1199 | `zExp', and significand formed by the concatenation of `zSig0' and `zSig1',
1200 | and returns the proper extended double-precision floating-point value
1201 | corresponding to the abstract input. This routine is just like
1202 | `roundAndPackFloatx80' except that the input significand does not have to be
1204 *----------------------------------------------------------------------------*/
1206 floatx80
normalizeRoundAndPackFloatx80(FloatX80RoundPrec roundingPrecision
,
1207 bool zSign
, int32_t zExp
,
1208 uint64_t zSig0
, uint64_t zSig1
,
1209 float_status
*status
);
1211 /*----------------------------------------------------------------------------
1212 | The pattern for a default generated extended double-precision NaN.
1213 *----------------------------------------------------------------------------*/
1214 floatx80
floatx80_default_nan(float_status
*status
);
1216 /*----------------------------------------------------------------------------
1217 | Software IEC/IEEE quadruple-precision conversion routines.
1218 *----------------------------------------------------------------------------*/
1219 int32_t float128_to_int32(float128
, float_status
*status
);
1220 int32_t float128_to_int32_round_to_zero(float128
, float_status
*status
);
1221 int64_t float128_to_int64(float128
, float_status
*status
);
1222 Int128
float128_to_int128(float128
, float_status
*status
);
1223 int64_t float128_to_int64_round_to_zero(float128
, float_status
*status
);
1224 Int128
float128_to_int128_round_to_zero(float128
, float_status
*status
);
1225 uint64_t float128_to_uint64(float128
, float_status
*status
);
1226 Int128
float128_to_uint128(float128
, float_status
*status
);
1227 uint64_t float128_to_uint64_round_to_zero(float128
, float_status
*status
);
1228 Int128
float128_to_uint128_round_to_zero(float128
, float_status
*status
);
1229 uint32_t float128_to_uint32(float128
, float_status
*status
);
1230 uint32_t float128_to_uint32_round_to_zero(float128
, float_status
*status
);
1231 float32
float128_to_float32(float128
, float_status
*status
);
1232 float64
float128_to_float64(float128
, float_status
*status
);
1233 floatx80
float128_to_floatx80(float128
, float_status
*status
);
1235 /*----------------------------------------------------------------------------
1236 | Software IEC/IEEE quadruple-precision operations.
1237 *----------------------------------------------------------------------------*/
1238 float128
float128_round_to_int(float128
, float_status
*status
);
1239 float128
float128_add(float128
, float128
, float_status
*status
);
1240 float128
float128_sub(float128
, float128
, float_status
*status
);
1241 float128
float128_mul(float128
, float128
, float_status
*status
);
1242 float128
float128_muladd(float128
, float128
, float128
, int,
1243 float_status
*status
);
1244 float128
float128_div(float128
, float128
, float_status
*status
);
1245 float128
float128_rem(float128
, float128
, float_status
*status
);
1246 float128
float128_sqrt(float128
, float_status
*status
);
1247 FloatRelation
float128_compare(float128
, float128
, float_status
*status
);
1248 FloatRelation
float128_compare_quiet(float128
, float128
, float_status
*status
);
1249 float128
float128_min(float128
, float128
, float_status
*status
);
1250 float128
float128_max(float128
, float128
, float_status
*status
);
1251 float128
float128_minnum(float128
, float128
, float_status
*status
);
1252 float128
float128_maxnum(float128
, float128
, float_status
*status
);
1253 float128
float128_minnummag(float128
, float128
, float_status
*status
);
1254 float128
float128_maxnummag(float128
, float128
, float_status
*status
);
1255 float128
float128_minimum_number(float128
, float128
, float_status
*status
);
1256 float128
float128_maximum_number(float128
, float128
, float_status
*status
);
1257 bool float128_is_quiet_nan(float128
, float_status
*status
);
1258 bool float128_is_signaling_nan(float128
, float_status
*status
);
1259 float128
float128_silence_nan(float128
, float_status
*status
);
1260 float128
float128_scalbn(float128
, int, float_status
*status
);
1262 static inline float128
float128_abs(float128 a
)
1264 a
.high
&= 0x7fffffffffffffffLL
;
1268 static inline float128
float128_chs(float128 a
)
1270 a
.high
^= 0x8000000000000000LL
;
1274 static inline bool float128_is_infinity(float128 a
)
1276 return (a
.high
& 0x7fffffffffffffffLL
) == 0x7fff000000000000LL
&& a
.low
== 0;
1279 static inline bool float128_is_neg(float128 a
)
1281 return a
.high
>> 63;
1284 static inline bool float128_is_zero(float128 a
)
1286 return (a
.high
& 0x7fffffffffffffffLL
) == 0 && a
.low
== 0;
1289 static inline bool float128_is_zero_or_denormal(float128 a
)
1291 return (a
.high
& 0x7fff000000000000LL
) == 0;
1294 static inline bool float128_is_normal(float128 a
)
1296 return (((a
.high
>> 48) + 1) & 0x7fff) >= 2;
1299 static inline bool float128_is_denormal(float128 a
)
1301 return float128_is_zero_or_denormal(a
) && !float128_is_zero(a
);
1304 static inline bool float128_is_any_nan(float128 a
)
1306 return ((a
.high
>> 48) & 0x7fff) == 0x7fff &&
1307 ((a
.low
!= 0) || ((a
.high
& 0xffffffffffffLL
) != 0));
1310 static inline bool float128_eq(float128 a
, float128 b
, float_status
*s
)
1312 return float128_compare(a
, b
, s
) == float_relation_equal
;
1315 static inline bool float128_le(float128 a
, float128 b
, float_status
*s
)
1317 return float128_compare(a
, b
, s
) <= float_relation_equal
;
1320 static inline bool float128_lt(float128 a
, float128 b
, float_status
*s
)
1322 return float128_compare(a
, b
, s
) < float_relation_equal
;
1325 static inline bool float128_unordered(float128 a
, float128 b
, float_status
*s
)
1327 return float128_compare(a
, b
, s
) == float_relation_unordered
;
1330 static inline bool float128_eq_quiet(float128 a
, float128 b
, float_status
*s
)
1332 return float128_compare_quiet(a
, b
, s
) == float_relation_equal
;
1335 static inline bool float128_le_quiet(float128 a
, float128 b
, float_status
*s
)
1337 return float128_compare_quiet(a
, b
, s
) <= float_relation_equal
;
1340 static inline bool float128_lt_quiet(float128 a
, float128 b
, float_status
*s
)
1342 return float128_compare_quiet(a
, b
, s
) < float_relation_equal
;
1345 static inline bool float128_unordered_quiet(float128 a
, float128 b
,
1348 return float128_compare_quiet(a
, b
, s
) == float_relation_unordered
;
1351 #define float128_zero make_float128(0, 0)
1353 /*----------------------------------------------------------------------------
1354 | The pattern for a default generated quadruple-precision NaN.
1355 *----------------------------------------------------------------------------*/
1356 float128
float128_default_nan(float_status
*status
);
1358 #endif /* SOFTFLOAT_H */