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 int16_t bfloat16_to_int16_scalbn(bfloat16
, FloatRoundMode
,
370 int, float_status
*status
);
371 int32_t bfloat16_to_int32_scalbn(bfloat16
, FloatRoundMode
,
372 int, float_status
*status
);
373 int64_t bfloat16_to_int64_scalbn(bfloat16
, FloatRoundMode
,
374 int, float_status
*status
);
376 int16_t bfloat16_to_int16(bfloat16
, float_status
*status
);
377 int32_t bfloat16_to_int32(bfloat16
, float_status
*status
);
378 int64_t bfloat16_to_int64(bfloat16
, float_status
*status
);
380 int16_t bfloat16_to_int16_round_to_zero(bfloat16
, float_status
*status
);
381 int32_t bfloat16_to_int32_round_to_zero(bfloat16
, float_status
*status
);
382 int64_t bfloat16_to_int64_round_to_zero(bfloat16
, float_status
*status
);
384 uint16_t bfloat16_to_uint16_scalbn(bfloat16 a
, FloatRoundMode
,
385 int, float_status
*status
);
386 uint32_t bfloat16_to_uint32_scalbn(bfloat16 a
, FloatRoundMode
,
387 int, float_status
*status
);
388 uint64_t bfloat16_to_uint64_scalbn(bfloat16 a
, FloatRoundMode
,
389 int, float_status
*status
);
391 uint16_t bfloat16_to_uint16(bfloat16 a
, float_status
*status
);
392 uint32_t bfloat16_to_uint32(bfloat16 a
, float_status
*status
);
393 uint64_t bfloat16_to_uint64(bfloat16 a
, float_status
*status
);
395 uint16_t bfloat16_to_uint16_round_to_zero(bfloat16 a
, float_status
*status
);
396 uint32_t bfloat16_to_uint32_round_to_zero(bfloat16 a
, float_status
*status
);
397 uint64_t bfloat16_to_uint64_round_to_zero(bfloat16 a
, float_status
*status
);
399 bfloat16
int16_to_bfloat16_scalbn(int16_t a
, int, float_status
*status
);
400 bfloat16
int32_to_bfloat16_scalbn(int32_t a
, int, float_status
*status
);
401 bfloat16
int64_to_bfloat16_scalbn(int64_t a
, int, float_status
*status
);
402 bfloat16
uint16_to_bfloat16_scalbn(uint16_t a
, int, float_status
*status
);
403 bfloat16
uint32_to_bfloat16_scalbn(uint32_t a
, int, float_status
*status
);
404 bfloat16
uint64_to_bfloat16_scalbn(uint64_t a
, int, float_status
*status
);
406 bfloat16
int16_to_bfloat16(int16_t a
, float_status
*status
);
407 bfloat16
int32_to_bfloat16(int32_t a
, float_status
*status
);
408 bfloat16
int64_to_bfloat16(int64_t a
, float_status
*status
);
409 bfloat16
uint16_to_bfloat16(uint16_t a
, float_status
*status
);
410 bfloat16
uint32_to_bfloat16(uint32_t a
, float_status
*status
);
411 bfloat16
uint64_to_bfloat16(uint64_t a
, float_status
*status
);
413 /*----------------------------------------------------------------------------
414 | Software bfloat16 operations.
415 *----------------------------------------------------------------------------*/
417 bfloat16
bfloat16_add(bfloat16
, bfloat16
, float_status
*status
);
418 bfloat16
bfloat16_sub(bfloat16
, bfloat16
, float_status
*status
);
419 bfloat16
bfloat16_mul(bfloat16
, bfloat16
, float_status
*status
);
420 bfloat16
bfloat16_div(bfloat16
, bfloat16
, float_status
*status
);
421 bfloat16
bfloat16_muladd(bfloat16
, bfloat16
, bfloat16
, int,
422 float_status
*status
);
423 float16
bfloat16_scalbn(bfloat16
, int, float_status
*status
);
424 bfloat16
bfloat16_min(bfloat16
, bfloat16
, float_status
*status
);
425 bfloat16
bfloat16_max(bfloat16
, bfloat16
, float_status
*status
);
426 bfloat16
bfloat16_minnum(bfloat16
, bfloat16
, float_status
*status
);
427 bfloat16
bfloat16_maxnum(bfloat16
, bfloat16
, float_status
*status
);
428 bfloat16
bfloat16_minnummag(bfloat16
, bfloat16
, float_status
*status
);
429 bfloat16
bfloat16_maxnummag(bfloat16
, bfloat16
, float_status
*status
);
430 bfloat16
bfloat16_minimum_number(bfloat16
, bfloat16
, float_status
*status
);
431 bfloat16
bfloat16_maximum_number(bfloat16
, bfloat16
, float_status
*status
);
432 bfloat16
bfloat16_sqrt(bfloat16
, float_status
*status
);
433 FloatRelation
bfloat16_compare(bfloat16
, bfloat16
, float_status
*status
);
434 FloatRelation
bfloat16_compare_quiet(bfloat16
, bfloat16
, float_status
*status
);
436 bool bfloat16_is_quiet_nan(bfloat16
, float_status
*status
);
437 bool bfloat16_is_signaling_nan(bfloat16
, float_status
*status
);
438 bfloat16
bfloat16_silence_nan(bfloat16
, float_status
*status
);
439 bfloat16
bfloat16_default_nan(float_status
*status
);
441 static inline bool bfloat16_is_any_nan(bfloat16 a
)
443 return ((a
& ~0x8000) > 0x7F80);
446 static inline bool bfloat16_is_neg(bfloat16 a
)
451 static inline bool bfloat16_is_infinity(bfloat16 a
)
453 return (a
& 0x7fff) == 0x7F80;
456 static inline bool bfloat16_is_zero(bfloat16 a
)
458 return (a
& 0x7fff) == 0;
461 static inline bool bfloat16_is_zero_or_denormal(bfloat16 a
)
463 return (a
& 0x7F80) == 0;
466 static inline bool bfloat16_is_normal(bfloat16 a
)
468 return (((a
>> 7) + 1) & 0xff) >= 2;
471 static inline bfloat16
bfloat16_abs(bfloat16 a
)
473 /* Note that abs does *not* handle NaN specially, nor does
474 * it flush denormal inputs to zero.
479 static inline bfloat16
bfloat16_chs(bfloat16 a
)
481 /* Note that chs does *not* handle NaN specially, nor does
482 * it flush denormal inputs to zero.
487 static inline bfloat16
bfloat16_set_sign(bfloat16 a
, int sign
)
489 return (a
& 0x7fff) | (sign
<< 15);
492 static inline bool bfloat16_eq(bfloat16 a
, bfloat16 b
, float_status
*s
)
494 return bfloat16_compare(a
, b
, s
) == float_relation_equal
;
497 static inline bool bfloat16_le(bfloat16 a
, bfloat16 b
, float_status
*s
)
499 return bfloat16_compare(a
, b
, s
) <= float_relation_equal
;
502 static inline bool bfloat16_lt(bfloat16 a
, bfloat16 b
, float_status
*s
)
504 return bfloat16_compare(a
, b
, s
) < float_relation_equal
;
507 static inline bool bfloat16_unordered(bfloat16 a
, bfloat16 b
, float_status
*s
)
509 return bfloat16_compare(a
, b
, s
) == float_relation_unordered
;
512 static inline bool bfloat16_eq_quiet(bfloat16 a
, bfloat16 b
, float_status
*s
)
514 return bfloat16_compare_quiet(a
, b
, s
) == float_relation_equal
;
517 static inline bool bfloat16_le_quiet(bfloat16 a
, bfloat16 b
, float_status
*s
)
519 return bfloat16_compare_quiet(a
, b
, s
) <= float_relation_equal
;
522 static inline bool bfloat16_lt_quiet(bfloat16 a
, bfloat16 b
, float_status
*s
)
524 return bfloat16_compare_quiet(a
, b
, s
) < float_relation_equal
;
527 static inline bool bfloat16_unordered_quiet(bfloat16 a
, bfloat16 b
,
530 return bfloat16_compare_quiet(a
, b
, s
) == float_relation_unordered
;
533 #define bfloat16_zero 0
534 #define bfloat16_half 0x3f00
535 #define bfloat16_one 0x3f80
536 #define bfloat16_one_point_five 0x3fc0
537 #define bfloat16_two 0x4000
538 #define bfloat16_three 0x4040
539 #define bfloat16_infinity 0x7f80
541 /*----------------------------------------------------------------------------
542 | The pattern for a default generated half-precision NaN.
543 *----------------------------------------------------------------------------*/
544 float16
float16_default_nan(float_status
*status
);
546 /*----------------------------------------------------------------------------
547 | Software IEC/IEEE single-precision conversion routines.
548 *----------------------------------------------------------------------------*/
550 int16_t float32_to_int16_scalbn(float32
, FloatRoundMode
, int, float_status
*);
551 int32_t float32_to_int32_scalbn(float32
, FloatRoundMode
, int, float_status
*);
552 int64_t float32_to_int64_scalbn(float32
, FloatRoundMode
, int, float_status
*);
554 int16_t float32_to_int16(float32
, float_status
*status
);
555 int32_t float32_to_int32(float32
, float_status
*status
);
556 int64_t float32_to_int64(float32
, float_status
*status
);
558 int16_t float32_to_int16_round_to_zero(float32
, float_status
*status
);
559 int32_t float32_to_int32_round_to_zero(float32
, float_status
*status
);
560 int64_t float32_to_int64_round_to_zero(float32
, float_status
*status
);
562 uint16_t float32_to_uint16_scalbn(float32
, FloatRoundMode
, int, float_status
*);
563 uint32_t float32_to_uint32_scalbn(float32
, FloatRoundMode
, int, float_status
*);
564 uint64_t float32_to_uint64_scalbn(float32
, FloatRoundMode
, int, float_status
*);
566 uint16_t float32_to_uint16(float32
, float_status
*status
);
567 uint32_t float32_to_uint32(float32
, float_status
*status
);
568 uint64_t float32_to_uint64(float32
, float_status
*status
);
570 uint16_t float32_to_uint16_round_to_zero(float32
, float_status
*status
);
571 uint32_t float32_to_uint32_round_to_zero(float32
, float_status
*status
);
572 uint64_t float32_to_uint64_round_to_zero(float32
, float_status
*status
);
574 float64
float32_to_float64(float32
, float_status
*status
);
575 floatx80
float32_to_floatx80(float32
, float_status
*status
);
576 float128
float32_to_float128(float32
, float_status
*status
);
578 /*----------------------------------------------------------------------------
579 | Software IEC/IEEE single-precision operations.
580 *----------------------------------------------------------------------------*/
581 float32
float32_round_to_int(float32
, float_status
*status
);
582 float32
float32_add(float32
, float32
, float_status
*status
);
583 float32
float32_sub(float32
, float32
, float_status
*status
);
584 float32
float32_mul(float32
, float32
, float_status
*status
);
585 float32
float32_div(float32
, float32
, float_status
*status
);
586 float32
float32_rem(float32
, float32
, float_status
*status
);
587 float32
float32_muladd(float32
, float32
, float32
, int, float_status
*status
);
588 float32
float32_sqrt(float32
, float_status
*status
);
589 float32
float32_exp2(float32
, float_status
*status
);
590 float32
float32_log2(float32
, float_status
*status
);
591 FloatRelation
float32_compare(float32
, float32
, float_status
*status
);
592 FloatRelation
float32_compare_quiet(float32
, float32
, float_status
*status
);
593 float32
float32_min(float32
, float32
, float_status
*status
);
594 float32
float32_max(float32
, float32
, float_status
*status
);
595 float32
float32_minnum(float32
, float32
, float_status
*status
);
596 float32
float32_maxnum(float32
, float32
, float_status
*status
);
597 float32
float32_minnummag(float32
, float32
, float_status
*status
);
598 float32
float32_maxnummag(float32
, float32
, float_status
*status
);
599 float32
float32_minimum_number(float32
, float32
, float_status
*status
);
600 float32
float32_maximum_number(float32
, float32
, float_status
*status
);
601 bool float32_is_quiet_nan(float32
, float_status
*status
);
602 bool float32_is_signaling_nan(float32
, float_status
*status
);
603 float32
float32_silence_nan(float32
, float_status
*status
);
604 float32
float32_scalbn(float32
, int, float_status
*status
);
606 static inline float32
float32_abs(float32 a
)
608 /* Note that abs does *not* handle NaN specially, nor does
609 * it flush denormal inputs to zero.
611 return make_float32(float32_val(a
) & 0x7fffffff);
614 static inline float32
float32_chs(float32 a
)
616 /* Note that chs does *not* handle NaN specially, nor does
617 * it flush denormal inputs to zero.
619 return make_float32(float32_val(a
) ^ 0x80000000);
622 static inline bool float32_is_infinity(float32 a
)
624 return (float32_val(a
) & 0x7fffffff) == 0x7f800000;
627 static inline bool float32_is_neg(float32 a
)
629 return float32_val(a
) >> 31;
632 static inline bool float32_is_zero(float32 a
)
634 return (float32_val(a
) & 0x7fffffff) == 0;
637 static inline bool float32_is_any_nan(float32 a
)
639 return ((float32_val(a
) & ~(1 << 31)) > 0x7f800000UL
);
642 static inline bool float32_is_zero_or_denormal(float32 a
)
644 return (float32_val(a
) & 0x7f800000) == 0;
647 static inline bool float32_is_normal(float32 a
)
649 return (((float32_val(a
) >> 23) + 1) & 0xff) >= 2;
652 static inline bool float32_is_denormal(float32 a
)
654 return float32_is_zero_or_denormal(a
) && !float32_is_zero(a
);
657 static inline bool float32_is_zero_or_normal(float32 a
)
659 return float32_is_normal(a
) || float32_is_zero(a
);
662 static inline float32
float32_set_sign(float32 a
, int sign
)
664 return make_float32((float32_val(a
) & 0x7fffffff) | (sign
<< 31));
667 static inline bool float32_eq(float32 a
, float32 b
, float_status
*s
)
669 return float32_compare(a
, b
, s
) == float_relation_equal
;
672 static inline bool float32_le(float32 a
, float32 b
, float_status
*s
)
674 return float32_compare(a
, b
, s
) <= float_relation_equal
;
677 static inline bool float32_lt(float32 a
, float32 b
, float_status
*s
)
679 return float32_compare(a
, b
, s
) < float_relation_equal
;
682 static inline bool float32_unordered(float32 a
, float32 b
, float_status
*s
)
684 return float32_compare(a
, b
, s
) == float_relation_unordered
;
687 static inline bool float32_eq_quiet(float32 a
, float32 b
, float_status
*s
)
689 return float32_compare_quiet(a
, b
, s
) == float_relation_equal
;
692 static inline bool float32_le_quiet(float32 a
, float32 b
, float_status
*s
)
694 return float32_compare_quiet(a
, b
, s
) <= float_relation_equal
;
697 static inline bool float32_lt_quiet(float32 a
, float32 b
, float_status
*s
)
699 return float32_compare_quiet(a
, b
, s
) < float_relation_equal
;
702 static inline bool float32_unordered_quiet(float32 a
, float32 b
,
705 return float32_compare_quiet(a
, b
, s
) == float_relation_unordered
;
708 #define float32_zero make_float32(0)
709 #define float32_half make_float32(0x3f000000)
710 #define float32_one make_float32(0x3f800000)
711 #define float32_one_point_five make_float32(0x3fc00000)
712 #define float32_two make_float32(0x40000000)
713 #define float32_three make_float32(0x40400000)
714 #define float32_infinity make_float32(0x7f800000)
716 /*----------------------------------------------------------------------------
717 | Packs the sign `zSign', exponent `zExp', and significand `zSig' into a
718 | single-precision floating-point value, returning the result. After being
719 | shifted into the proper positions, the three fields are simply added
720 | together to form the result. This means that any integer portion of `zSig'
721 | will be added into the exponent. Since a properly normalized significand
722 | will have an integer portion equal to 1, the `zExp' input should be 1 less
723 | than the desired result exponent whenever `zSig' is a complete, normalized
725 *----------------------------------------------------------------------------*/
727 static inline float32
packFloat32(bool zSign
, int zExp
, uint32_t zSig
)
730 (((uint32_t)zSign
) << 31) + (((uint32_t)zExp
) << 23) + zSig
);
733 /*----------------------------------------------------------------------------
734 | The pattern for a default generated single-precision NaN.
735 *----------------------------------------------------------------------------*/
736 float32
float32_default_nan(float_status
*status
);
738 /*----------------------------------------------------------------------------
739 | Software IEC/IEEE double-precision conversion routines.
740 *----------------------------------------------------------------------------*/
742 int16_t float64_to_int16_scalbn(float64
, FloatRoundMode
, int, float_status
*);
743 int32_t float64_to_int32_scalbn(float64
, FloatRoundMode
, int, float_status
*);
744 int64_t float64_to_int64_scalbn(float64
, FloatRoundMode
, int, float_status
*);
746 int16_t float64_to_int16(float64
, float_status
*status
);
747 int32_t float64_to_int32(float64
, float_status
*status
);
748 int64_t float64_to_int64(float64
, float_status
*status
);
750 int16_t float64_to_int16_round_to_zero(float64
, float_status
*status
);
751 int32_t float64_to_int32_round_to_zero(float64
, float_status
*status
);
752 int64_t float64_to_int64_round_to_zero(float64
, float_status
*status
);
754 uint16_t float64_to_uint16_scalbn(float64
, FloatRoundMode
, int, float_status
*);
755 uint32_t float64_to_uint32_scalbn(float64
, FloatRoundMode
, int, float_status
*);
756 uint64_t float64_to_uint64_scalbn(float64
, FloatRoundMode
, int, float_status
*);
758 uint16_t float64_to_uint16(float64
, float_status
*status
);
759 uint32_t float64_to_uint32(float64
, float_status
*status
);
760 uint64_t float64_to_uint64(float64
, float_status
*status
);
762 uint16_t float64_to_uint16_round_to_zero(float64
, float_status
*status
);
763 uint32_t float64_to_uint32_round_to_zero(float64
, float_status
*status
);
764 uint64_t float64_to_uint64_round_to_zero(float64
, float_status
*status
);
766 float32
float64_to_float32(float64
, float_status
*status
);
767 floatx80
float64_to_floatx80(float64
, float_status
*status
);
768 float128
float64_to_float128(float64
, float_status
*status
);
770 /*----------------------------------------------------------------------------
771 | Software IEC/IEEE double-precision operations.
772 *----------------------------------------------------------------------------*/
773 float64
float64_round_to_int(float64
, float_status
*status
);
774 float64
float64_add(float64
, float64
, float_status
*status
);
775 float64
float64_sub(float64
, float64
, float_status
*status
);
776 float64
float64_mul(float64
, float64
, float_status
*status
);
777 float64
float64_div(float64
, float64
, float_status
*status
);
778 float64
float64_rem(float64
, float64
, float_status
*status
);
779 float64
float64_muladd(float64
, float64
, float64
, int, float_status
*status
);
780 float64
float64_sqrt(float64
, float_status
*status
);
781 float64
float64_log2(float64
, float_status
*status
);
782 FloatRelation
float64_compare(float64
, float64
, float_status
*status
);
783 FloatRelation
float64_compare_quiet(float64
, float64
, float_status
*status
);
784 float64
float64_min(float64
, float64
, float_status
*status
);
785 float64
float64_max(float64
, float64
, float_status
*status
);
786 float64
float64_minnum(float64
, float64
, float_status
*status
);
787 float64
float64_maxnum(float64
, float64
, float_status
*status
);
788 float64
float64_minnummag(float64
, float64
, float_status
*status
);
789 float64
float64_maxnummag(float64
, float64
, float_status
*status
);
790 float64
float64_minimum_number(float64
, float64
, float_status
*status
);
791 float64
float64_maximum_number(float64
, float64
, float_status
*status
);
792 bool float64_is_quiet_nan(float64 a
, float_status
*status
);
793 bool float64_is_signaling_nan(float64
, float_status
*status
);
794 float64
float64_silence_nan(float64
, float_status
*status
);
795 float64
float64_scalbn(float64
, int, float_status
*status
);
797 static inline float64
float64_abs(float64 a
)
799 /* Note that abs does *not* handle NaN specially, nor does
800 * it flush denormal inputs to zero.
802 return make_float64(float64_val(a
) & 0x7fffffffffffffffLL
);
805 static inline float64
float64_chs(float64 a
)
807 /* Note that chs does *not* handle NaN specially, nor does
808 * it flush denormal inputs to zero.
810 return make_float64(float64_val(a
) ^ 0x8000000000000000LL
);
813 static inline bool float64_is_infinity(float64 a
)
815 return (float64_val(a
) & 0x7fffffffffffffffLL
) == 0x7ff0000000000000LL
;
818 static inline bool float64_is_neg(float64 a
)
820 return float64_val(a
) >> 63;
823 static inline bool float64_is_zero(float64 a
)
825 return (float64_val(a
) & 0x7fffffffffffffffLL
) == 0;
828 static inline bool float64_is_any_nan(float64 a
)
830 return ((float64_val(a
) & ~(1ULL << 63)) > 0x7ff0000000000000ULL
);
833 static inline bool float64_is_zero_or_denormal(float64 a
)
835 return (float64_val(a
) & 0x7ff0000000000000LL
) == 0;
838 static inline bool float64_is_normal(float64 a
)
840 return (((float64_val(a
) >> 52) + 1) & 0x7ff) >= 2;
843 static inline bool float64_is_denormal(float64 a
)
845 return float64_is_zero_or_denormal(a
) && !float64_is_zero(a
);
848 static inline bool float64_is_zero_or_normal(float64 a
)
850 return float64_is_normal(a
) || float64_is_zero(a
);
853 static inline float64
float64_set_sign(float64 a
, int sign
)
855 return make_float64((float64_val(a
) & 0x7fffffffffffffffULL
)
856 | ((int64_t)sign
<< 63));
859 static inline bool float64_eq(float64 a
, float64 b
, float_status
*s
)
861 return float64_compare(a
, b
, s
) == float_relation_equal
;
864 static inline bool float64_le(float64 a
, float64 b
, float_status
*s
)
866 return float64_compare(a
, b
, s
) <= float_relation_equal
;
869 static inline bool float64_lt(float64 a
, float64 b
, float_status
*s
)
871 return float64_compare(a
, b
, s
) < float_relation_equal
;
874 static inline bool float64_unordered(float64 a
, float64 b
, float_status
*s
)
876 return float64_compare(a
, b
, s
) == float_relation_unordered
;
879 static inline bool float64_eq_quiet(float64 a
, float64 b
, float_status
*s
)
881 return float64_compare_quiet(a
, b
, s
) == float_relation_equal
;
884 static inline bool float64_le_quiet(float64 a
, float64 b
, float_status
*s
)
886 return float64_compare_quiet(a
, b
, s
) <= float_relation_equal
;
889 static inline bool float64_lt_quiet(float64 a
, float64 b
, float_status
*s
)
891 return float64_compare_quiet(a
, b
, s
) < float_relation_equal
;
894 static inline bool float64_unordered_quiet(float64 a
, float64 b
,
897 return float64_compare_quiet(a
, b
, s
) == float_relation_unordered
;
900 #define float64_zero make_float64(0)
901 #define float64_half make_float64(0x3fe0000000000000LL)
902 #define float64_one make_float64(0x3ff0000000000000LL)
903 #define float64_one_point_five make_float64(0x3FF8000000000000ULL)
904 #define float64_two make_float64(0x4000000000000000ULL)
905 #define float64_three make_float64(0x4008000000000000ULL)
906 #define float64_ln2 make_float64(0x3fe62e42fefa39efLL)
907 #define float64_infinity make_float64(0x7ff0000000000000LL)
909 /*----------------------------------------------------------------------------
910 | The pattern for a default generated double-precision NaN.
911 *----------------------------------------------------------------------------*/
912 float64
float64_default_nan(float_status
*status
);
914 /*----------------------------------------------------------------------------
915 | Software IEC/IEEE double-precision operations, rounding to single precision,
916 | returning a result in double precision, with only one rounding step.
917 *----------------------------------------------------------------------------*/
919 float64
float64r32_add(float64
, float64
, float_status
*status
);
920 float64
float64r32_sub(float64
, float64
, float_status
*status
);
921 float64
float64r32_mul(float64
, float64
, float_status
*status
);
922 float64
float64r32_div(float64
, float64
, float_status
*status
);
923 float64
float64r32_muladd(float64
, float64
, float64
, int, float_status
*status
);
924 float64
float64r32_sqrt(float64
, float_status
*status
);
926 /*----------------------------------------------------------------------------
927 | Software IEC/IEEE extended double-precision conversion routines.
928 *----------------------------------------------------------------------------*/
929 int32_t floatx80_to_int32(floatx80
, float_status
*status
);
930 int32_t floatx80_to_int32_round_to_zero(floatx80
, float_status
*status
);
931 int64_t floatx80_to_int64(floatx80
, float_status
*status
);
932 int64_t floatx80_to_int64_round_to_zero(floatx80
, float_status
*status
);
933 float32
floatx80_to_float32(floatx80
, float_status
*status
);
934 float64
floatx80_to_float64(floatx80
, float_status
*status
);
935 float128
floatx80_to_float128(floatx80
, float_status
*status
);
937 /*----------------------------------------------------------------------------
938 | The pattern for an extended double-precision inf.
939 *----------------------------------------------------------------------------*/
940 extern const floatx80 floatx80_infinity
;
942 /*----------------------------------------------------------------------------
943 | Software IEC/IEEE extended double-precision operations.
944 *----------------------------------------------------------------------------*/
945 floatx80
floatx80_round(floatx80 a
, float_status
*status
);
946 floatx80
floatx80_round_to_int(floatx80
, float_status
*status
);
947 floatx80
floatx80_add(floatx80
, floatx80
, float_status
*status
);
948 floatx80
floatx80_sub(floatx80
, floatx80
, float_status
*status
);
949 floatx80
floatx80_mul(floatx80
, floatx80
, float_status
*status
);
950 floatx80
floatx80_div(floatx80
, floatx80
, float_status
*status
);
951 floatx80
floatx80_modrem(floatx80
, floatx80
, bool, uint64_t *,
952 float_status
*status
);
953 floatx80
floatx80_mod(floatx80
, floatx80
, float_status
*status
);
954 floatx80
floatx80_rem(floatx80
, floatx80
, float_status
*status
);
955 floatx80
floatx80_sqrt(floatx80
, float_status
*status
);
956 FloatRelation
floatx80_compare(floatx80
, floatx80
, float_status
*status
);
957 FloatRelation
floatx80_compare_quiet(floatx80
, floatx80
, float_status
*status
);
958 int floatx80_is_quiet_nan(floatx80
, float_status
*status
);
959 int floatx80_is_signaling_nan(floatx80
, float_status
*status
);
960 floatx80
floatx80_silence_nan(floatx80
, float_status
*status
);
961 floatx80
floatx80_scalbn(floatx80
, int, float_status
*status
);
963 static inline floatx80
floatx80_abs(floatx80 a
)
969 static inline floatx80
floatx80_chs(floatx80 a
)
975 static inline bool floatx80_is_infinity(floatx80 a
)
977 #if defined(TARGET_M68K)
978 return (a
.high
& 0x7fff) == floatx80_infinity
.high
&& !(a
.low
<< 1);
980 return (a
.high
& 0x7fff) == floatx80_infinity
.high
&&
981 a
.low
== floatx80_infinity
.low
;
985 static inline bool floatx80_is_neg(floatx80 a
)
990 static inline bool floatx80_is_zero(floatx80 a
)
992 return (a
.high
& 0x7fff) == 0 && a
.low
== 0;
995 static inline bool floatx80_is_zero_or_denormal(floatx80 a
)
997 return (a
.high
& 0x7fff) == 0;
1000 static inline bool floatx80_is_any_nan(floatx80 a
)
1002 return ((a
.high
& 0x7fff) == 0x7fff) && (a
.low
<<1);
1005 static inline bool floatx80_eq(floatx80 a
, floatx80 b
, float_status
*s
)
1007 return floatx80_compare(a
, b
, s
) == float_relation_equal
;
1010 static inline bool floatx80_le(floatx80 a
, floatx80 b
, float_status
*s
)
1012 return floatx80_compare(a
, b
, s
) <= float_relation_equal
;
1015 static inline bool floatx80_lt(floatx80 a
, floatx80 b
, float_status
*s
)
1017 return floatx80_compare(a
, b
, s
) < float_relation_equal
;
1020 static inline bool floatx80_unordered(floatx80 a
, floatx80 b
, float_status
*s
)
1022 return floatx80_compare(a
, b
, s
) == float_relation_unordered
;
1025 static inline bool floatx80_eq_quiet(floatx80 a
, floatx80 b
, float_status
*s
)
1027 return floatx80_compare_quiet(a
, b
, s
) == float_relation_equal
;
1030 static inline bool floatx80_le_quiet(floatx80 a
, floatx80 b
, float_status
*s
)
1032 return floatx80_compare_quiet(a
, b
, s
) <= float_relation_equal
;
1035 static inline bool floatx80_lt_quiet(floatx80 a
, floatx80 b
, float_status
*s
)
1037 return floatx80_compare_quiet(a
, b
, s
) < float_relation_equal
;
1040 static inline bool floatx80_unordered_quiet(floatx80 a
, floatx80 b
,
1043 return floatx80_compare_quiet(a
, b
, s
) == float_relation_unordered
;
1046 /*----------------------------------------------------------------------------
1047 | Return whether the given value is an invalid floatx80 encoding.
1048 | Invalid floatx80 encodings arise when the integer bit is not set, but
1049 | the exponent is not zero. The only times the integer bit is permitted to
1050 | be zero is in subnormal numbers and the value zero.
1051 | This includes what the Intel software developer's manual calls pseudo-NaNs,
1052 | pseudo-infinities and un-normal numbers. It does not include
1053 | pseudo-denormals, which must still be correctly handled as inputs even
1054 | if they are never generated as outputs.
1055 *----------------------------------------------------------------------------*/
1056 static inline bool floatx80_invalid_encoding(floatx80 a
)
1058 #if defined(TARGET_M68K)
1059 /*-------------------------------------------------------------------------
1060 | With m68k, the explicit integer bit can be zero in the case of:
1061 | - zeros (exp == 0, mantissa == 0)
1062 | - denormalized numbers (exp == 0, mantissa != 0)
1063 | - unnormalized numbers (exp != 0, exp < 0x7FFF)
1064 | - infinities (exp == 0x7FFF, mantissa == 0)
1065 | - not-a-numbers (exp == 0x7FFF, mantissa != 0)
1067 | For infinities and NaNs, the explicit integer bit can be either one or
1070 | The IEEE 754 standard does not define a zero integer bit. Such a number
1071 | is an unnormalized number. Hardware does not directly support
1072 | denormalized and unnormalized numbers, but implicitly supports them by
1073 | trapping them as unimplemented data types, allowing efficient conversion
1076 | See "M68000 FAMILY PROGRAMMER’S REFERENCE MANUAL",
1077 | "1.6 FLOATING-POINT DATA TYPES"
1078 *------------------------------------------------------------------------*/
1081 return (a
.low
& (1ULL << 63)) == 0 && (a
.high
& 0x7FFF) != 0;
1085 #define floatx80_zero make_floatx80(0x0000, 0x0000000000000000LL)
1086 #define floatx80_zero_init make_floatx80_init(0x0000, 0x0000000000000000LL)
1087 #define floatx80_one make_floatx80(0x3fff, 0x8000000000000000LL)
1088 #define floatx80_ln2 make_floatx80(0x3ffe, 0xb17217f7d1cf79acLL)
1089 #define floatx80_pi make_floatx80(0x4000, 0xc90fdaa22168c235LL)
1090 #define floatx80_half make_floatx80(0x3ffe, 0x8000000000000000LL)
1092 /*----------------------------------------------------------------------------
1093 | Returns the fraction bits of the extended double-precision floating-point
1095 *----------------------------------------------------------------------------*/
1097 static inline uint64_t extractFloatx80Frac(floatx80 a
)
1102 /*----------------------------------------------------------------------------
1103 | Returns the exponent bits of the extended double-precision floating-point
1105 *----------------------------------------------------------------------------*/
1107 static inline int32_t extractFloatx80Exp(floatx80 a
)
1109 return a
.high
& 0x7FFF;
1112 /*----------------------------------------------------------------------------
1113 | Returns the sign bit of the extended double-precision floating-point value
1115 *----------------------------------------------------------------------------*/
1117 static inline bool extractFloatx80Sign(floatx80 a
)
1119 return a
.high
>> 15;
1122 /*----------------------------------------------------------------------------
1123 | Packs the sign `zSign', exponent `zExp', and significand `zSig' into an
1124 | extended double-precision floating-point value, returning the result.
1125 *----------------------------------------------------------------------------*/
1127 static inline floatx80
packFloatx80(bool zSign
, int32_t zExp
, uint64_t zSig
)
1132 z
.high
= (((uint16_t)zSign
) << 15) + zExp
;
1136 /*----------------------------------------------------------------------------
1137 | Normalizes the subnormal extended double-precision floating-point value
1138 | represented by the denormalized significand `aSig'. The normalized exponent
1139 | and significand are stored at the locations pointed to by `zExpPtr' and
1140 | `zSigPtr', respectively.
1141 *----------------------------------------------------------------------------*/
1143 void normalizeFloatx80Subnormal(uint64_t aSig
, int32_t *zExpPtr
,
1146 /*----------------------------------------------------------------------------
1147 | Takes two extended double-precision floating-point values `a' and `b', one
1148 | of which is a NaN, and returns the appropriate NaN result. If either `a' or
1149 | `b' is a signaling NaN, the invalid exception is raised.
1150 *----------------------------------------------------------------------------*/
1152 floatx80
propagateFloatx80NaN(floatx80 a
, floatx80 b
, float_status
*status
);
1154 /*----------------------------------------------------------------------------
1155 | Takes an abstract floating-point value having sign `zSign', exponent `zExp',
1156 | and extended significand formed by the concatenation of `zSig0' and `zSig1',
1157 | and returns the proper extended double-precision floating-point value
1158 | corresponding to the abstract input. Ordinarily, the abstract value is
1159 | rounded and packed into the extended double-precision format, with the
1160 | inexact exception raised if the abstract input cannot be represented
1161 | exactly. However, if the abstract value is too large, the overflow and
1162 | inexact exceptions are raised and an infinity or maximal finite value is
1163 | returned. If the abstract value is too small, the input value is rounded to
1164 | a subnormal number, and the underflow and inexact exceptions are raised if
1165 | the abstract input cannot be represented exactly as a subnormal extended
1166 | double-precision floating-point number.
1167 | If `roundingPrecision' is 32 or 64, the result is rounded to the same
1168 | number of bits as single or double precision, respectively. Otherwise, the
1169 | result is rounded to the full precision of the extended double-precision
1171 | The input significand must be normalized or smaller. If the input
1172 | significand is not normalized, `zExp' must be 0; in that case, the result
1173 | returned is a subnormal number, and it must not require rounding. The
1174 | handling of underflow and overflow follows the IEC/IEEE Standard for Binary
1175 | Floating-Point Arithmetic.
1176 *----------------------------------------------------------------------------*/
1178 floatx80
roundAndPackFloatx80(FloatX80RoundPrec roundingPrecision
, bool zSign
,
1179 int32_t zExp
, uint64_t zSig0
, uint64_t zSig1
,
1180 float_status
*status
);
1182 /*----------------------------------------------------------------------------
1183 | Takes an abstract floating-point value having sign `zSign', exponent
1184 | `zExp', and significand formed by the concatenation of `zSig0' and `zSig1',
1185 | and returns the proper extended double-precision floating-point value
1186 | corresponding to the abstract input. This routine is just like
1187 | `roundAndPackFloatx80' except that the input significand does not have to be
1189 *----------------------------------------------------------------------------*/
1191 floatx80
normalizeRoundAndPackFloatx80(FloatX80RoundPrec roundingPrecision
,
1192 bool zSign
, int32_t zExp
,
1193 uint64_t zSig0
, uint64_t zSig1
,
1194 float_status
*status
);
1196 /*----------------------------------------------------------------------------
1197 | The pattern for a default generated extended double-precision NaN.
1198 *----------------------------------------------------------------------------*/
1199 floatx80
floatx80_default_nan(float_status
*status
);
1201 /*----------------------------------------------------------------------------
1202 | Software IEC/IEEE quadruple-precision conversion routines.
1203 *----------------------------------------------------------------------------*/
1204 int32_t float128_to_int32(float128
, float_status
*status
);
1205 int32_t float128_to_int32_round_to_zero(float128
, float_status
*status
);
1206 int64_t float128_to_int64(float128
, float_status
*status
);
1207 Int128
float128_to_int128(float128
, float_status
*status
);
1208 int64_t float128_to_int64_round_to_zero(float128
, float_status
*status
);
1209 Int128
float128_to_int128_round_to_zero(float128
, float_status
*status
);
1210 uint64_t float128_to_uint64(float128
, float_status
*status
);
1211 Int128
float128_to_uint128(float128
, float_status
*status
);
1212 uint64_t float128_to_uint64_round_to_zero(float128
, float_status
*status
);
1213 Int128
float128_to_uint128_round_to_zero(float128
, float_status
*status
);
1214 uint32_t float128_to_uint32(float128
, float_status
*status
);
1215 uint32_t float128_to_uint32_round_to_zero(float128
, float_status
*status
);
1216 float32
float128_to_float32(float128
, float_status
*status
);
1217 float64
float128_to_float64(float128
, float_status
*status
);
1218 floatx80
float128_to_floatx80(float128
, float_status
*status
);
1220 /*----------------------------------------------------------------------------
1221 | Software IEC/IEEE quadruple-precision operations.
1222 *----------------------------------------------------------------------------*/
1223 float128
float128_round_to_int(float128
, float_status
*status
);
1224 float128
float128_add(float128
, float128
, float_status
*status
);
1225 float128
float128_sub(float128
, float128
, float_status
*status
);
1226 float128
float128_mul(float128
, float128
, float_status
*status
);
1227 float128
float128_muladd(float128
, float128
, float128
, int,
1228 float_status
*status
);
1229 float128
float128_div(float128
, float128
, float_status
*status
);
1230 float128
float128_rem(float128
, float128
, float_status
*status
);
1231 float128
float128_sqrt(float128
, float_status
*status
);
1232 FloatRelation
float128_compare(float128
, float128
, float_status
*status
);
1233 FloatRelation
float128_compare_quiet(float128
, float128
, float_status
*status
);
1234 float128
float128_min(float128
, float128
, float_status
*status
);
1235 float128
float128_max(float128
, float128
, float_status
*status
);
1236 float128
float128_minnum(float128
, float128
, float_status
*status
);
1237 float128
float128_maxnum(float128
, float128
, float_status
*status
);
1238 float128
float128_minnummag(float128
, float128
, float_status
*status
);
1239 float128
float128_maxnummag(float128
, float128
, float_status
*status
);
1240 float128
float128_minimum_number(float128
, float128
, float_status
*status
);
1241 float128
float128_maximum_number(float128
, float128
, float_status
*status
);
1242 bool float128_is_quiet_nan(float128
, float_status
*status
);
1243 bool float128_is_signaling_nan(float128
, float_status
*status
);
1244 float128
float128_silence_nan(float128
, float_status
*status
);
1245 float128
float128_scalbn(float128
, int, float_status
*status
);
1247 static inline float128
float128_abs(float128 a
)
1249 a
.high
&= 0x7fffffffffffffffLL
;
1253 static inline float128
float128_chs(float128 a
)
1255 a
.high
^= 0x8000000000000000LL
;
1259 static inline bool float128_is_infinity(float128 a
)
1261 return (a
.high
& 0x7fffffffffffffffLL
) == 0x7fff000000000000LL
&& a
.low
== 0;
1264 static inline bool float128_is_neg(float128 a
)
1266 return a
.high
>> 63;
1269 static inline bool float128_is_zero(float128 a
)
1271 return (a
.high
& 0x7fffffffffffffffLL
) == 0 && a
.low
== 0;
1274 static inline bool float128_is_zero_or_denormal(float128 a
)
1276 return (a
.high
& 0x7fff000000000000LL
) == 0;
1279 static inline bool float128_is_normal(float128 a
)
1281 return (((a
.high
>> 48) + 1) & 0x7fff) >= 2;
1284 static inline bool float128_is_denormal(float128 a
)
1286 return float128_is_zero_or_denormal(a
) && !float128_is_zero(a
);
1289 static inline bool float128_is_any_nan(float128 a
)
1291 return ((a
.high
>> 48) & 0x7fff) == 0x7fff &&
1292 ((a
.low
!= 0) || ((a
.high
& 0xffffffffffffLL
) != 0));
1295 static inline bool float128_eq(float128 a
, float128 b
, float_status
*s
)
1297 return float128_compare(a
, b
, s
) == float_relation_equal
;
1300 static inline bool float128_le(float128 a
, float128 b
, float_status
*s
)
1302 return float128_compare(a
, b
, s
) <= float_relation_equal
;
1305 static inline bool float128_lt(float128 a
, float128 b
, float_status
*s
)
1307 return float128_compare(a
, b
, s
) < float_relation_equal
;
1310 static inline bool float128_unordered(float128 a
, float128 b
, float_status
*s
)
1312 return float128_compare(a
, b
, s
) == float_relation_unordered
;
1315 static inline bool float128_eq_quiet(float128 a
, float128 b
, float_status
*s
)
1317 return float128_compare_quiet(a
, b
, s
) == float_relation_equal
;
1320 static inline bool float128_le_quiet(float128 a
, float128 b
, float_status
*s
)
1322 return float128_compare_quiet(a
, b
, s
) <= float_relation_equal
;
1325 static inline bool float128_lt_quiet(float128 a
, float128 b
, float_status
*s
)
1327 return float128_compare_quiet(a
, b
, s
) < float_relation_equal
;
1330 static inline bool float128_unordered_quiet(float128 a
, float128 b
,
1333 return float128_compare_quiet(a
, b
, s
) == float_relation_unordered
;
1336 #define float128_zero make_float128(0, 0)
1338 /*----------------------------------------------------------------------------
1339 | The pattern for a default generated quadruple-precision NaN.
1340 *----------------------------------------------------------------------------*/
1341 float128
float128_default_nan(float_status
*status
);
1343 #endif /* SOFTFLOAT_H */