xref: /openbmc/qemu/include/fpu/softfloat.h (revision 0ca117a7)
1 /*
2  * QEMU float support
3  *
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
10  *  the BSD license
11  *  GPL-v2-or-later
12  *
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.
16  */
17 
18 /*
19 ===============================================================================
20 This C header file is part of the SoftFloat IEC/IEEE Floating-point
21 Arithmetic Package, Release 2a.
22 
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'.
32 
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.
38 
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.
43 
44 ===============================================================================
45 */
46 
47 /* BSD licensing:
48  * Copyright (c) 2006, Fabrice Bellard
49  * All rights reserved.
50  *
51  * Redistribution and use in source and binary forms, with or without
52  * modification, are permitted provided that the following conditions are met:
53  *
54  * 1. Redistributions of source code must retain the above copyright notice,
55  * this list of conditions and the following disclaimer.
56  *
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.
60  *
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.
64  *
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.
76  */
77 
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.
80  */
81 
82 #ifndef SOFTFLOAT_H
83 #define SOFTFLOAT_H
84 
85 /*----------------------------------------------------------------------------
86 | Software IEC/IEEE floating-point ordering relations
87 *----------------------------------------------------------------------------*/
88 
89 typedef enum {
90     float_relation_less      = -1,
91     float_relation_equal     =  0,
92     float_relation_greater   =  1,
93     float_relation_unordered =  2
94 } FloatRelation;
95 
96 #include "fpu/softfloat-types.h"
97 #include "fpu/softfloat-helpers.h"
98 
99 /*----------------------------------------------------------------------------
100 | Routine to raise any or all of the software IEC/IEEE floating-point
101 | exception flags.
102 *----------------------------------------------------------------------------*/
103 static inline void float_raise(uint8_t flags, float_status *status)
104 {
105     status->float_exception_flags |= flags;
106 }
107 
108 /*----------------------------------------------------------------------------
109 | If `a' is denormal and we are in flush-to-zero mode then set the
110 | input-denormal exception and return zero. Otherwise just return the value.
111 *----------------------------------------------------------------------------*/
112 float16 float16_squash_input_denormal(float16 a, float_status *status);
113 float32 float32_squash_input_denormal(float32 a, float_status *status);
114 float64 float64_squash_input_denormal(float64 a, float_status *status);
115 bfloat16 bfloat16_squash_input_denormal(bfloat16 a, float_status *status);
116 
117 /*----------------------------------------------------------------------------
118 | Options to indicate which negations to perform in float*_muladd()
119 | Using these differs from negating an input or output before calling
120 | the muladd function in that this means that a NaN doesn't have its
121 | sign bit inverted before it is propagated.
122 | We also support halving the result before rounding, as a special
123 | case to support the ARM fused-sqrt-step instruction FRSQRTS.
124 *----------------------------------------------------------------------------*/
125 enum {
126     float_muladd_negate_c = 1,
127     float_muladd_negate_product = 2,
128     float_muladd_negate_result = 4,
129     float_muladd_halve_result = 8,
130 };
131 
132 /*----------------------------------------------------------------------------
133 | Software IEC/IEEE integer-to-floating-point conversion routines.
134 *----------------------------------------------------------------------------*/
135 
136 float16 int16_to_float16_scalbn(int16_t a, int, float_status *status);
137 float16 int32_to_float16_scalbn(int32_t a, int, float_status *status);
138 float16 int64_to_float16_scalbn(int64_t a, int, float_status *status);
139 float16 uint16_to_float16_scalbn(uint16_t a, int, float_status *status);
140 float16 uint32_to_float16_scalbn(uint32_t a, int, float_status *status);
141 float16 uint64_to_float16_scalbn(uint64_t a, int, float_status *status);
142 
143 float16 int8_to_float16(int8_t a, float_status *status);
144 float16 int16_to_float16(int16_t a, float_status *status);
145 float16 int32_to_float16(int32_t a, float_status *status);
146 float16 int64_to_float16(int64_t a, float_status *status);
147 float16 uint8_to_float16(uint8_t a, float_status *status);
148 float16 uint16_to_float16(uint16_t a, float_status *status);
149 float16 uint32_to_float16(uint32_t a, float_status *status);
150 float16 uint64_to_float16(uint64_t a, float_status *status);
151 
152 float32 int16_to_float32_scalbn(int16_t, int, float_status *status);
153 float32 int32_to_float32_scalbn(int32_t, int, float_status *status);
154 float32 int64_to_float32_scalbn(int64_t, int, float_status *status);
155 float32 uint16_to_float32_scalbn(uint16_t, int, float_status *status);
156 float32 uint32_to_float32_scalbn(uint32_t, int, float_status *status);
157 float32 uint64_to_float32_scalbn(uint64_t, int, float_status *status);
158 
159 float32 int16_to_float32(int16_t, float_status *status);
160 float32 int32_to_float32(int32_t, float_status *status);
161 float32 int64_to_float32(int64_t, float_status *status);
162 float32 uint16_to_float32(uint16_t, float_status *status);
163 float32 uint32_to_float32(uint32_t, float_status *status);
164 float32 uint64_to_float32(uint64_t, float_status *status);
165 
166 float64 int16_to_float64_scalbn(int16_t, int, float_status *status);
167 float64 int32_to_float64_scalbn(int32_t, int, float_status *status);
168 float64 int64_to_float64_scalbn(int64_t, int, float_status *status);
169 float64 uint16_to_float64_scalbn(uint16_t, int, float_status *status);
170 float64 uint32_to_float64_scalbn(uint32_t, int, float_status *status);
171 float64 uint64_to_float64_scalbn(uint64_t, int, float_status *status);
172 
173 float64 int16_to_float64(int16_t, float_status *status);
174 float64 int32_to_float64(int32_t, float_status *status);
175 float64 int64_to_float64(int64_t, float_status *status);
176 float64 uint16_to_float64(uint16_t, float_status *status);
177 float64 uint32_to_float64(uint32_t, float_status *status);
178 float64 uint64_to_float64(uint64_t, float_status *status);
179 
180 floatx80 int32_to_floatx80(int32_t, float_status *status);
181 floatx80 int64_to_floatx80(int64_t, float_status *status);
182 
183 float128 int32_to_float128(int32_t, float_status *status);
184 float128 int64_to_float128(int64_t, float_status *status);
185 float128 uint64_to_float128(uint64_t, float_status *status);
186 
187 /*----------------------------------------------------------------------------
188 | Software half-precision conversion routines.
189 *----------------------------------------------------------------------------*/
190 
191 float16 float32_to_float16(float32, bool ieee, float_status *status);
192 float32 float16_to_float32(float16, bool ieee, float_status *status);
193 float16 float64_to_float16(float64 a, bool ieee, float_status *status);
194 float64 float16_to_float64(float16 a, bool ieee, float_status *status);
195 
196 int8_t  float16_to_int8_scalbn(float16, FloatRoundMode, int,
197                                float_status *status);
198 int16_t float16_to_int16_scalbn(float16, FloatRoundMode, int, float_status *);
199 int32_t float16_to_int32_scalbn(float16, FloatRoundMode, int, float_status *);
200 int64_t float16_to_int64_scalbn(float16, FloatRoundMode, int, float_status *);
201 
202 int8_t  float16_to_int8(float16, float_status *status);
203 int16_t float16_to_int16(float16, float_status *status);
204 int32_t float16_to_int32(float16, float_status *status);
205 int64_t float16_to_int64(float16, float_status *status);
206 
207 int16_t float16_to_int16_round_to_zero(float16, float_status *status);
208 int32_t float16_to_int32_round_to_zero(float16, float_status *status);
209 int64_t float16_to_int64_round_to_zero(float16, float_status *status);
210 
211 uint8_t float16_to_uint8_scalbn(float16 a, FloatRoundMode,
212                                 int, float_status *status);
213 uint16_t float16_to_uint16_scalbn(float16 a, FloatRoundMode,
214                                   int, float_status *status);
215 uint32_t float16_to_uint32_scalbn(float16 a, FloatRoundMode,
216                                   int, float_status *status);
217 uint64_t float16_to_uint64_scalbn(float16 a, FloatRoundMode,
218                                   int, float_status *status);
219 
220 uint8_t  float16_to_uint8(float16 a, float_status *status);
221 uint16_t float16_to_uint16(float16 a, float_status *status);
222 uint32_t float16_to_uint32(float16 a, float_status *status);
223 uint64_t float16_to_uint64(float16 a, float_status *status);
224 
225 uint16_t float16_to_uint16_round_to_zero(float16 a, float_status *status);
226 uint32_t float16_to_uint32_round_to_zero(float16 a, float_status *status);
227 uint64_t float16_to_uint64_round_to_zero(float16 a, float_status *status);
228 
229 /*----------------------------------------------------------------------------
230 | Software half-precision operations.
231 *----------------------------------------------------------------------------*/
232 
233 float16 float16_round_to_int(float16, float_status *status);
234 float16 float16_add(float16, float16, float_status *status);
235 float16 float16_sub(float16, float16, float_status *status);
236 float16 float16_mul(float16, float16, float_status *status);
237 float16 float16_muladd(float16, float16, float16, int, float_status *status);
238 float16 float16_div(float16, float16, float_status *status);
239 float16 float16_scalbn(float16, int, float_status *status);
240 float16 float16_min(float16, float16, float_status *status);
241 float16 float16_max(float16, float16, float_status *status);
242 float16 float16_minnum(float16, float16, float_status *status);
243 float16 float16_maxnum(float16, float16, float_status *status);
244 float16 float16_minnummag(float16, float16, float_status *status);
245 float16 float16_maxnummag(float16, float16, float_status *status);
246 float16 float16_minimum_number(float16, float16, float_status *status);
247 float16 float16_maximum_number(float16, float16, float_status *status);
248 float16 float16_sqrt(float16, float_status *status);
249 FloatRelation float16_compare(float16, float16, float_status *status);
250 FloatRelation float16_compare_quiet(float16, float16, float_status *status);
251 
252 bool float16_is_quiet_nan(float16, float_status *status);
253 bool float16_is_signaling_nan(float16, float_status *status);
254 float16 float16_silence_nan(float16, float_status *status);
255 
256 static inline bool float16_is_any_nan(float16 a)
257 {
258     return ((float16_val(a) & ~0x8000) > 0x7c00);
259 }
260 
261 static inline bool float16_is_neg(float16 a)
262 {
263     return float16_val(a) >> 15;
264 }
265 
266 static inline bool float16_is_infinity(float16 a)
267 {
268     return (float16_val(a) & 0x7fff) == 0x7c00;
269 }
270 
271 static inline bool float16_is_zero(float16 a)
272 {
273     return (float16_val(a) & 0x7fff) == 0;
274 }
275 
276 static inline bool float16_is_zero_or_denormal(float16 a)
277 {
278     return (float16_val(a) & 0x7c00) == 0;
279 }
280 
281 static inline bool float16_is_normal(float16 a)
282 {
283     return (((float16_val(a) >> 10) + 1) & 0x1f) >= 2;
284 }
285 
286 static inline float16 float16_abs(float16 a)
287 {
288     /* Note that abs does *not* handle NaN specially, nor does
289      * it flush denormal inputs to zero.
290      */
291     return make_float16(float16_val(a) & 0x7fff);
292 }
293 
294 static inline float16 float16_chs(float16 a)
295 {
296     /* Note that chs does *not* handle NaN specially, nor does
297      * it flush denormal inputs to zero.
298      */
299     return make_float16(float16_val(a) ^ 0x8000);
300 }
301 
302 static inline float16 float16_set_sign(float16 a, int sign)
303 {
304     return make_float16((float16_val(a) & 0x7fff) | (sign << 15));
305 }
306 
307 static inline bool float16_eq(float16 a, float16 b, float_status *s)
308 {
309     return float16_compare(a, b, s) == float_relation_equal;
310 }
311 
312 static inline bool float16_le(float16 a, float16 b, float_status *s)
313 {
314     return float16_compare(a, b, s) <= float_relation_equal;
315 }
316 
317 static inline bool float16_lt(float16 a, float16 b, float_status *s)
318 {
319     return float16_compare(a, b, s) < float_relation_equal;
320 }
321 
322 static inline bool float16_unordered(float16 a, float16 b, float_status *s)
323 {
324     return float16_compare(a, b, s) == float_relation_unordered;
325 }
326 
327 static inline bool float16_eq_quiet(float16 a, float16 b, float_status *s)
328 {
329     return float16_compare_quiet(a, b, s) == float_relation_equal;
330 }
331 
332 static inline bool float16_le_quiet(float16 a, float16 b, float_status *s)
333 {
334     return float16_compare_quiet(a, b, s) <= float_relation_equal;
335 }
336 
337 static inline bool float16_lt_quiet(float16 a, float16 b, float_status *s)
338 {
339     return float16_compare_quiet(a, b, s) < float_relation_equal;
340 }
341 
342 static inline bool float16_unordered_quiet(float16 a, float16 b,
343                                            float_status *s)
344 {
345     return float16_compare_quiet(a, b, s) == float_relation_unordered;
346 }
347 
348 #define float16_zero make_float16(0)
349 #define float16_half make_float16(0x3800)
350 #define float16_one make_float16(0x3c00)
351 #define float16_one_point_five make_float16(0x3e00)
352 #define float16_two make_float16(0x4000)
353 #define float16_three make_float16(0x4200)
354 #define float16_infinity make_float16(0x7c00)
355 
356 /*----------------------------------------------------------------------------
357 | Software bfloat16 conversion routines.
358 *----------------------------------------------------------------------------*/
359 
360 bfloat16 bfloat16_round_to_int(bfloat16, float_status *status);
361 bfloat16 float32_to_bfloat16(float32, float_status *status);
362 float32 bfloat16_to_float32(bfloat16, float_status *status);
363 bfloat16 float64_to_bfloat16(float64 a, float_status *status);
364 float64 bfloat16_to_float64(bfloat16 a, float_status *status);
365 
366 int16_t bfloat16_to_int16_scalbn(bfloat16, FloatRoundMode,
367                                  int, float_status *status);
368 int32_t bfloat16_to_int32_scalbn(bfloat16, FloatRoundMode,
369                                  int, float_status *status);
370 int64_t bfloat16_to_int64_scalbn(bfloat16, FloatRoundMode,
371                                  int, float_status *status);
372 
373 int16_t bfloat16_to_int16(bfloat16, float_status *status);
374 int32_t bfloat16_to_int32(bfloat16, float_status *status);
375 int64_t bfloat16_to_int64(bfloat16, float_status *status);
376 
377 int16_t bfloat16_to_int16_round_to_zero(bfloat16, float_status *status);
378 int32_t bfloat16_to_int32_round_to_zero(bfloat16, float_status *status);
379 int64_t bfloat16_to_int64_round_to_zero(bfloat16, float_status *status);
380 
381 uint16_t bfloat16_to_uint16_scalbn(bfloat16 a, FloatRoundMode,
382                                    int, float_status *status);
383 uint32_t bfloat16_to_uint32_scalbn(bfloat16 a, FloatRoundMode,
384                                    int, float_status *status);
385 uint64_t bfloat16_to_uint64_scalbn(bfloat16 a, FloatRoundMode,
386                                    int, float_status *status);
387 
388 uint16_t bfloat16_to_uint16(bfloat16 a, float_status *status);
389 uint32_t bfloat16_to_uint32(bfloat16 a, float_status *status);
390 uint64_t bfloat16_to_uint64(bfloat16 a, float_status *status);
391 
392 uint16_t bfloat16_to_uint16_round_to_zero(bfloat16 a, float_status *status);
393 uint32_t bfloat16_to_uint32_round_to_zero(bfloat16 a, float_status *status);
394 uint64_t bfloat16_to_uint64_round_to_zero(bfloat16 a, float_status *status);
395 
396 bfloat16 int16_to_bfloat16_scalbn(int16_t a, int, float_status *status);
397 bfloat16 int32_to_bfloat16_scalbn(int32_t a, int, float_status *status);
398 bfloat16 int64_to_bfloat16_scalbn(int64_t a, int, float_status *status);
399 bfloat16 uint16_to_bfloat16_scalbn(uint16_t a, int, float_status *status);
400 bfloat16 uint32_to_bfloat16_scalbn(uint32_t a, int, float_status *status);
401 bfloat16 uint64_to_bfloat16_scalbn(uint64_t a, int, float_status *status);
402 
403 bfloat16 int16_to_bfloat16(int16_t a, float_status *status);
404 bfloat16 int32_to_bfloat16(int32_t a, float_status *status);
405 bfloat16 int64_to_bfloat16(int64_t a, float_status *status);
406 bfloat16 uint16_to_bfloat16(uint16_t a, float_status *status);
407 bfloat16 uint32_to_bfloat16(uint32_t a, float_status *status);
408 bfloat16 uint64_to_bfloat16(uint64_t a, float_status *status);
409 
410 /*----------------------------------------------------------------------------
411 | Software bfloat16 operations.
412 *----------------------------------------------------------------------------*/
413 
414 bfloat16 bfloat16_add(bfloat16, bfloat16, float_status *status);
415 bfloat16 bfloat16_sub(bfloat16, bfloat16, float_status *status);
416 bfloat16 bfloat16_mul(bfloat16, bfloat16, float_status *status);
417 bfloat16 bfloat16_div(bfloat16, bfloat16, float_status *status);
418 bfloat16 bfloat16_muladd(bfloat16, bfloat16, bfloat16, int,
419                          float_status *status);
420 float16 bfloat16_scalbn(bfloat16, int, float_status *status);
421 bfloat16 bfloat16_min(bfloat16, bfloat16, float_status *status);
422 bfloat16 bfloat16_max(bfloat16, bfloat16, float_status *status);
423 bfloat16 bfloat16_minnum(bfloat16, bfloat16, float_status *status);
424 bfloat16 bfloat16_maxnum(bfloat16, bfloat16, float_status *status);
425 bfloat16 bfloat16_minnummag(bfloat16, bfloat16, float_status *status);
426 bfloat16 bfloat16_maxnummag(bfloat16, bfloat16, float_status *status);
427 bfloat16 bfloat16_minimum_number(bfloat16, bfloat16, float_status *status);
428 bfloat16 bfloat16_maximum_number(bfloat16, bfloat16, float_status *status);
429 bfloat16 bfloat16_sqrt(bfloat16, float_status *status);
430 FloatRelation bfloat16_compare(bfloat16, bfloat16, float_status *status);
431 FloatRelation bfloat16_compare_quiet(bfloat16, bfloat16, float_status *status);
432 
433 bool bfloat16_is_quiet_nan(bfloat16, float_status *status);
434 bool bfloat16_is_signaling_nan(bfloat16, float_status *status);
435 bfloat16 bfloat16_silence_nan(bfloat16, float_status *status);
436 bfloat16 bfloat16_default_nan(float_status *status);
437 
438 static inline bool bfloat16_is_any_nan(bfloat16 a)
439 {
440     return ((a & ~0x8000) > 0x7F80);
441 }
442 
443 static inline bool bfloat16_is_neg(bfloat16 a)
444 {
445     return a >> 15;
446 }
447 
448 static inline bool bfloat16_is_infinity(bfloat16 a)
449 {
450     return (a & 0x7fff) == 0x7F80;
451 }
452 
453 static inline bool bfloat16_is_zero(bfloat16 a)
454 {
455     return (a & 0x7fff) == 0;
456 }
457 
458 static inline bool bfloat16_is_zero_or_denormal(bfloat16 a)
459 {
460     return (a & 0x7F80) == 0;
461 }
462 
463 static inline bool bfloat16_is_normal(bfloat16 a)
464 {
465     return (((a >> 7) + 1) & 0xff) >= 2;
466 }
467 
468 static inline bfloat16 bfloat16_abs(bfloat16 a)
469 {
470     /* Note that abs does *not* handle NaN specially, nor does
471      * it flush denormal inputs to zero.
472      */
473     return a & 0x7fff;
474 }
475 
476 static inline bfloat16 bfloat16_chs(bfloat16 a)
477 {
478     /* Note that chs does *not* handle NaN specially, nor does
479      * it flush denormal inputs to zero.
480      */
481     return a ^ 0x8000;
482 }
483 
484 static inline bfloat16 bfloat16_set_sign(bfloat16 a, int sign)
485 {
486     return (a & 0x7fff) | (sign << 15);
487 }
488 
489 static inline bool bfloat16_eq(bfloat16 a, bfloat16 b, float_status *s)
490 {
491     return bfloat16_compare(a, b, s) == float_relation_equal;
492 }
493 
494 static inline bool bfloat16_le(bfloat16 a, bfloat16 b, float_status *s)
495 {
496     return bfloat16_compare(a, b, s) <= float_relation_equal;
497 }
498 
499 static inline bool bfloat16_lt(bfloat16 a, bfloat16 b, float_status *s)
500 {
501     return bfloat16_compare(a, b, s) < float_relation_equal;
502 }
503 
504 static inline bool bfloat16_unordered(bfloat16 a, bfloat16 b, float_status *s)
505 {
506     return bfloat16_compare(a, b, s) == float_relation_unordered;
507 }
508 
509 static inline bool bfloat16_eq_quiet(bfloat16 a, bfloat16 b, float_status *s)
510 {
511     return bfloat16_compare_quiet(a, b, s) == float_relation_equal;
512 }
513 
514 static inline bool bfloat16_le_quiet(bfloat16 a, bfloat16 b, float_status *s)
515 {
516     return bfloat16_compare_quiet(a, b, s) <= float_relation_equal;
517 }
518 
519 static inline bool bfloat16_lt_quiet(bfloat16 a, bfloat16 b, float_status *s)
520 {
521     return bfloat16_compare_quiet(a, b, s) < float_relation_equal;
522 }
523 
524 static inline bool bfloat16_unordered_quiet(bfloat16 a, bfloat16 b,
525                                            float_status *s)
526 {
527     return bfloat16_compare_quiet(a, b, s) == float_relation_unordered;
528 }
529 
530 #define bfloat16_zero 0
531 #define bfloat16_half 0x3f00
532 #define bfloat16_one 0x3f80
533 #define bfloat16_one_point_five 0x3fc0
534 #define bfloat16_two 0x4000
535 #define bfloat16_three 0x4040
536 #define bfloat16_infinity 0x7f80
537 
538 /*----------------------------------------------------------------------------
539 | The pattern for a default generated half-precision NaN.
540 *----------------------------------------------------------------------------*/
541 float16 float16_default_nan(float_status *status);
542 
543 /*----------------------------------------------------------------------------
544 | Software IEC/IEEE single-precision conversion routines.
545 *----------------------------------------------------------------------------*/
546 
547 int16_t float32_to_int16_scalbn(float32, FloatRoundMode, int, float_status *);
548 int32_t float32_to_int32_scalbn(float32, FloatRoundMode, int, float_status *);
549 int64_t float32_to_int64_scalbn(float32, FloatRoundMode, int, float_status *);
550 
551 int16_t float32_to_int16(float32, float_status *status);
552 int32_t float32_to_int32(float32, float_status *status);
553 int64_t float32_to_int64(float32, float_status *status);
554 
555 int16_t float32_to_int16_round_to_zero(float32, float_status *status);
556 int32_t float32_to_int32_round_to_zero(float32, float_status *status);
557 int64_t float32_to_int64_round_to_zero(float32, float_status *status);
558 
559 uint16_t float32_to_uint16_scalbn(float32, FloatRoundMode, int, float_status *);
560 uint32_t float32_to_uint32_scalbn(float32, FloatRoundMode, int, float_status *);
561 uint64_t float32_to_uint64_scalbn(float32, FloatRoundMode, int, float_status *);
562 
563 uint16_t float32_to_uint16(float32, float_status *status);
564 uint32_t float32_to_uint32(float32, float_status *status);
565 uint64_t float32_to_uint64(float32, float_status *status);
566 
567 uint16_t float32_to_uint16_round_to_zero(float32, float_status *status);
568 uint32_t float32_to_uint32_round_to_zero(float32, float_status *status);
569 uint64_t float32_to_uint64_round_to_zero(float32, float_status *status);
570 
571 float64 float32_to_float64(float32, float_status *status);
572 floatx80 float32_to_floatx80(float32, float_status *status);
573 float128 float32_to_float128(float32, float_status *status);
574 
575 /*----------------------------------------------------------------------------
576 | Software IEC/IEEE single-precision operations.
577 *----------------------------------------------------------------------------*/
578 float32 float32_round_to_int(float32, float_status *status);
579 float32 float32_add(float32, float32, float_status *status);
580 float32 float32_sub(float32, float32, float_status *status);
581 float32 float32_mul(float32, float32, float_status *status);
582 float32 float32_div(float32, float32, float_status *status);
583 float32 float32_rem(float32, float32, float_status *status);
584 float32 float32_muladd(float32, float32, float32, int, float_status *status);
585 float32 float32_sqrt(float32, float_status *status);
586 float32 float32_exp2(float32, float_status *status);
587 float32 float32_log2(float32, float_status *status);
588 FloatRelation float32_compare(float32, float32, float_status *status);
589 FloatRelation float32_compare_quiet(float32, float32, float_status *status);
590 float32 float32_min(float32, float32, float_status *status);
591 float32 float32_max(float32, float32, float_status *status);
592 float32 float32_minnum(float32, float32, float_status *status);
593 float32 float32_maxnum(float32, float32, float_status *status);
594 float32 float32_minnummag(float32, float32, float_status *status);
595 float32 float32_maxnummag(float32, float32, float_status *status);
596 float32 float32_minimum_number(float32, float32, float_status *status);
597 float32 float32_maximum_number(float32, float32, float_status *status);
598 bool float32_is_quiet_nan(float32, float_status *status);
599 bool float32_is_signaling_nan(float32, float_status *status);
600 float32 float32_silence_nan(float32, float_status *status);
601 float32 float32_scalbn(float32, int, float_status *status);
602 
603 static inline float32 float32_abs(float32 a)
604 {
605     /* Note that abs does *not* handle NaN specially, nor does
606      * it flush denormal inputs to zero.
607      */
608     return make_float32(float32_val(a) & 0x7fffffff);
609 }
610 
611 static inline float32 float32_chs(float32 a)
612 {
613     /* Note that chs does *not* handle NaN specially, nor does
614      * it flush denormal inputs to zero.
615      */
616     return make_float32(float32_val(a) ^ 0x80000000);
617 }
618 
619 static inline bool float32_is_infinity(float32 a)
620 {
621     return (float32_val(a) & 0x7fffffff) == 0x7f800000;
622 }
623 
624 static inline bool float32_is_neg(float32 a)
625 {
626     return float32_val(a) >> 31;
627 }
628 
629 static inline bool float32_is_zero(float32 a)
630 {
631     return (float32_val(a) & 0x7fffffff) == 0;
632 }
633 
634 static inline bool float32_is_any_nan(float32 a)
635 {
636     return ((float32_val(a) & ~(1 << 31)) > 0x7f800000UL);
637 }
638 
639 static inline bool float32_is_zero_or_denormal(float32 a)
640 {
641     return (float32_val(a) & 0x7f800000) == 0;
642 }
643 
644 static inline bool float32_is_normal(float32 a)
645 {
646     return (((float32_val(a) >> 23) + 1) & 0xff) >= 2;
647 }
648 
649 static inline bool float32_is_denormal(float32 a)
650 {
651     return float32_is_zero_or_denormal(a) && !float32_is_zero(a);
652 }
653 
654 static inline bool float32_is_zero_or_normal(float32 a)
655 {
656     return float32_is_normal(a) || float32_is_zero(a);
657 }
658 
659 static inline float32 float32_set_sign(float32 a, int sign)
660 {
661     return make_float32((float32_val(a) & 0x7fffffff) | (sign << 31));
662 }
663 
664 static inline bool float32_eq(float32 a, float32 b, float_status *s)
665 {
666     return float32_compare(a, b, s) == float_relation_equal;
667 }
668 
669 static inline bool float32_le(float32 a, float32 b, float_status *s)
670 {
671     return float32_compare(a, b, s) <= float_relation_equal;
672 }
673 
674 static inline bool float32_lt(float32 a, float32 b, float_status *s)
675 {
676     return float32_compare(a, b, s) < float_relation_equal;
677 }
678 
679 static inline bool float32_unordered(float32 a, float32 b, float_status *s)
680 {
681     return float32_compare(a, b, s) == float_relation_unordered;
682 }
683 
684 static inline bool float32_eq_quiet(float32 a, float32 b, float_status *s)
685 {
686     return float32_compare_quiet(a, b, s) == float_relation_equal;
687 }
688 
689 static inline bool float32_le_quiet(float32 a, float32 b, float_status *s)
690 {
691     return float32_compare_quiet(a, b, s) <= float_relation_equal;
692 }
693 
694 static inline bool float32_lt_quiet(float32 a, float32 b, float_status *s)
695 {
696     return float32_compare_quiet(a, b, s) < float_relation_equal;
697 }
698 
699 static inline bool float32_unordered_quiet(float32 a, float32 b,
700                                            float_status *s)
701 {
702     return float32_compare_quiet(a, b, s) == float_relation_unordered;
703 }
704 
705 #define float32_zero make_float32(0)
706 #define float32_half make_float32(0x3f000000)
707 #define float32_one make_float32(0x3f800000)
708 #define float32_one_point_five make_float32(0x3fc00000)
709 #define float32_two make_float32(0x40000000)
710 #define float32_three make_float32(0x40400000)
711 #define float32_infinity make_float32(0x7f800000)
712 
713 /*----------------------------------------------------------------------------
714 | Packs the sign `zSign', exponent `zExp', and significand `zSig' into a
715 | single-precision floating-point value, returning the result.  After being
716 | shifted into the proper positions, the three fields are simply added
717 | together to form the result.  This means that any integer portion of `zSig'
718 | will be added into the exponent.  Since a properly normalized significand
719 | will have an integer portion equal to 1, the `zExp' input should be 1 less
720 | than the desired result exponent whenever `zSig' is a complete, normalized
721 | significand.
722 *----------------------------------------------------------------------------*/
723 
724 static inline float32 packFloat32(bool zSign, int zExp, uint32_t zSig)
725 {
726     return make_float32(
727           (((uint32_t)zSign) << 31) + (((uint32_t)zExp) << 23) + zSig);
728 }
729 
730 /*----------------------------------------------------------------------------
731 | The pattern for a default generated single-precision NaN.
732 *----------------------------------------------------------------------------*/
733 float32 float32_default_nan(float_status *status);
734 
735 /*----------------------------------------------------------------------------
736 | Software IEC/IEEE double-precision conversion routines.
737 *----------------------------------------------------------------------------*/
738 
739 int16_t float64_to_int16_scalbn(float64, FloatRoundMode, int, float_status *);
740 int32_t float64_to_int32_scalbn(float64, FloatRoundMode, int, float_status *);
741 int64_t float64_to_int64_scalbn(float64, FloatRoundMode, int, float_status *);
742 
743 int16_t float64_to_int16(float64, float_status *status);
744 int32_t float64_to_int32(float64, float_status *status);
745 int64_t float64_to_int64(float64, float_status *status);
746 
747 int16_t float64_to_int16_round_to_zero(float64, float_status *status);
748 int32_t float64_to_int32_round_to_zero(float64, float_status *status);
749 int64_t float64_to_int64_round_to_zero(float64, float_status *status);
750 
751 uint16_t float64_to_uint16_scalbn(float64, FloatRoundMode, int, float_status *);
752 uint32_t float64_to_uint32_scalbn(float64, FloatRoundMode, int, float_status *);
753 uint64_t float64_to_uint64_scalbn(float64, FloatRoundMode, int, float_status *);
754 
755 uint16_t float64_to_uint16(float64, float_status *status);
756 uint32_t float64_to_uint32(float64, float_status *status);
757 uint64_t float64_to_uint64(float64, float_status *status);
758 
759 uint16_t float64_to_uint16_round_to_zero(float64, float_status *status);
760 uint32_t float64_to_uint32_round_to_zero(float64, float_status *status);
761 uint64_t float64_to_uint64_round_to_zero(float64, float_status *status);
762 
763 float32 float64_to_float32(float64, float_status *status);
764 floatx80 float64_to_floatx80(float64, float_status *status);
765 float128 float64_to_float128(float64, float_status *status);
766 
767 /*----------------------------------------------------------------------------
768 | Software IEC/IEEE double-precision operations.
769 *----------------------------------------------------------------------------*/
770 float64 float64_round_to_int(float64, float_status *status);
771 float64 float64_add(float64, float64, float_status *status);
772 float64 float64_sub(float64, float64, float_status *status);
773 float64 float64_mul(float64, float64, float_status *status);
774 float64 float64_div(float64, float64, float_status *status);
775 float64 float64_rem(float64, float64, float_status *status);
776 float64 float64_muladd(float64, float64, float64, int, float_status *status);
777 float64 float64_sqrt(float64, float_status *status);
778 float64 float64_log2(float64, float_status *status);
779 FloatRelation float64_compare(float64, float64, float_status *status);
780 FloatRelation float64_compare_quiet(float64, float64, float_status *status);
781 float64 float64_min(float64, float64, float_status *status);
782 float64 float64_max(float64, float64, float_status *status);
783 float64 float64_minnum(float64, float64, float_status *status);
784 float64 float64_maxnum(float64, float64, float_status *status);
785 float64 float64_minnummag(float64, float64, float_status *status);
786 float64 float64_maxnummag(float64, float64, float_status *status);
787 float64 float64_minimum_number(float64, float64, float_status *status);
788 float64 float64_maximum_number(float64, float64, float_status *status);
789 bool float64_is_quiet_nan(float64 a, float_status *status);
790 bool float64_is_signaling_nan(float64, float_status *status);
791 float64 float64_silence_nan(float64, float_status *status);
792 float64 float64_scalbn(float64, int, float_status *status);
793 
794 static inline float64 float64_abs(float64 a)
795 {
796     /* Note that abs does *not* handle NaN specially, nor does
797      * it flush denormal inputs to zero.
798      */
799     return make_float64(float64_val(a) & 0x7fffffffffffffffLL);
800 }
801 
802 static inline float64 float64_chs(float64 a)
803 {
804     /* Note that chs does *not* handle NaN specially, nor does
805      * it flush denormal inputs to zero.
806      */
807     return make_float64(float64_val(a) ^ 0x8000000000000000LL);
808 }
809 
810 static inline bool float64_is_infinity(float64 a)
811 {
812     return (float64_val(a) & 0x7fffffffffffffffLL ) == 0x7ff0000000000000LL;
813 }
814 
815 static inline bool float64_is_neg(float64 a)
816 {
817     return float64_val(a) >> 63;
818 }
819 
820 static inline bool float64_is_zero(float64 a)
821 {
822     return (float64_val(a) & 0x7fffffffffffffffLL) == 0;
823 }
824 
825 static inline bool float64_is_any_nan(float64 a)
826 {
827     return ((float64_val(a) & ~(1ULL << 63)) > 0x7ff0000000000000ULL);
828 }
829 
830 static inline bool float64_is_zero_or_denormal(float64 a)
831 {
832     return (float64_val(a) & 0x7ff0000000000000LL) == 0;
833 }
834 
835 static inline bool float64_is_normal(float64 a)
836 {
837     return (((float64_val(a) >> 52) + 1) & 0x7ff) >= 2;
838 }
839 
840 static inline bool float64_is_denormal(float64 a)
841 {
842     return float64_is_zero_or_denormal(a) && !float64_is_zero(a);
843 }
844 
845 static inline bool float64_is_zero_or_normal(float64 a)
846 {
847     return float64_is_normal(a) || float64_is_zero(a);
848 }
849 
850 static inline float64 float64_set_sign(float64 a, int sign)
851 {
852     return make_float64((float64_val(a) & 0x7fffffffffffffffULL)
853                         | ((int64_t)sign << 63));
854 }
855 
856 static inline bool float64_eq(float64 a, float64 b, float_status *s)
857 {
858     return float64_compare(a, b, s) == float_relation_equal;
859 }
860 
861 static inline bool float64_le(float64 a, float64 b, float_status *s)
862 {
863     return float64_compare(a, b, s) <= float_relation_equal;
864 }
865 
866 static inline bool float64_lt(float64 a, float64 b, float_status *s)
867 {
868     return float64_compare(a, b, s) < float_relation_equal;
869 }
870 
871 static inline bool float64_unordered(float64 a, float64 b, float_status *s)
872 {
873     return float64_compare(a, b, s) == float_relation_unordered;
874 }
875 
876 static inline bool float64_eq_quiet(float64 a, float64 b, float_status *s)
877 {
878     return float64_compare_quiet(a, b, s) == float_relation_equal;
879 }
880 
881 static inline bool float64_le_quiet(float64 a, float64 b, float_status *s)
882 {
883     return float64_compare_quiet(a, b, s) <= float_relation_equal;
884 }
885 
886 static inline bool float64_lt_quiet(float64 a, float64 b, float_status *s)
887 {
888     return float64_compare_quiet(a, b, s) < float_relation_equal;
889 }
890 
891 static inline bool float64_unordered_quiet(float64 a, float64 b,
892                                            float_status *s)
893 {
894     return float64_compare_quiet(a, b, s) == float_relation_unordered;
895 }
896 
897 #define float64_zero make_float64(0)
898 #define float64_half make_float64(0x3fe0000000000000LL)
899 #define float64_one make_float64(0x3ff0000000000000LL)
900 #define float64_one_point_five make_float64(0x3FF8000000000000ULL)
901 #define float64_two make_float64(0x4000000000000000ULL)
902 #define float64_three make_float64(0x4008000000000000ULL)
903 #define float64_ln2 make_float64(0x3fe62e42fefa39efLL)
904 #define float64_infinity make_float64(0x7ff0000000000000LL)
905 
906 /*----------------------------------------------------------------------------
907 | The pattern for a default generated double-precision NaN.
908 *----------------------------------------------------------------------------*/
909 float64 float64_default_nan(float_status *status);
910 
911 /*----------------------------------------------------------------------------
912 | Software IEC/IEEE extended double-precision conversion routines.
913 *----------------------------------------------------------------------------*/
914 int32_t floatx80_to_int32(floatx80, float_status *status);
915 int32_t floatx80_to_int32_round_to_zero(floatx80, float_status *status);
916 int64_t floatx80_to_int64(floatx80, float_status *status);
917 int64_t floatx80_to_int64_round_to_zero(floatx80, float_status *status);
918 float32 floatx80_to_float32(floatx80, float_status *status);
919 float64 floatx80_to_float64(floatx80, float_status *status);
920 float128 floatx80_to_float128(floatx80, float_status *status);
921 
922 /*----------------------------------------------------------------------------
923 | The pattern for an extended double-precision inf.
924 *----------------------------------------------------------------------------*/
925 extern const floatx80 floatx80_infinity;
926 
927 /*----------------------------------------------------------------------------
928 | Software IEC/IEEE extended double-precision operations.
929 *----------------------------------------------------------------------------*/
930 floatx80 floatx80_round(floatx80 a, float_status *status);
931 floatx80 floatx80_round_to_int(floatx80, float_status *status);
932 floatx80 floatx80_add(floatx80, floatx80, float_status *status);
933 floatx80 floatx80_sub(floatx80, floatx80, float_status *status);
934 floatx80 floatx80_mul(floatx80, floatx80, float_status *status);
935 floatx80 floatx80_div(floatx80, floatx80, float_status *status);
936 floatx80 floatx80_modrem(floatx80, floatx80, bool, uint64_t *,
937                          float_status *status);
938 floatx80 floatx80_mod(floatx80, floatx80, float_status *status);
939 floatx80 floatx80_rem(floatx80, floatx80, float_status *status);
940 floatx80 floatx80_sqrt(floatx80, float_status *status);
941 FloatRelation floatx80_compare(floatx80, floatx80, float_status *status);
942 FloatRelation floatx80_compare_quiet(floatx80, floatx80, float_status *status);
943 int floatx80_is_quiet_nan(floatx80, float_status *status);
944 int floatx80_is_signaling_nan(floatx80, float_status *status);
945 floatx80 floatx80_silence_nan(floatx80, float_status *status);
946 floatx80 floatx80_scalbn(floatx80, int, float_status *status);
947 
948 static inline floatx80 floatx80_abs(floatx80 a)
949 {
950     a.high &= 0x7fff;
951     return a;
952 }
953 
954 static inline floatx80 floatx80_chs(floatx80 a)
955 {
956     a.high ^= 0x8000;
957     return a;
958 }
959 
960 static inline bool floatx80_is_infinity(floatx80 a)
961 {
962 #if defined(TARGET_M68K)
963     return (a.high & 0x7fff) == floatx80_infinity.high && !(a.low << 1);
964 #else
965     return (a.high & 0x7fff) == floatx80_infinity.high &&
966                        a.low == floatx80_infinity.low;
967 #endif
968 }
969 
970 static inline bool floatx80_is_neg(floatx80 a)
971 {
972     return a.high >> 15;
973 }
974 
975 static inline bool floatx80_is_zero(floatx80 a)
976 {
977     return (a.high & 0x7fff) == 0 && a.low == 0;
978 }
979 
980 static inline bool floatx80_is_zero_or_denormal(floatx80 a)
981 {
982     return (a.high & 0x7fff) == 0;
983 }
984 
985 static inline bool floatx80_is_any_nan(floatx80 a)
986 {
987     return ((a.high & 0x7fff) == 0x7fff) && (a.low<<1);
988 }
989 
990 static inline bool floatx80_eq(floatx80 a, floatx80 b, float_status *s)
991 {
992     return floatx80_compare(a, b, s) == float_relation_equal;
993 }
994 
995 static inline bool floatx80_le(floatx80 a, floatx80 b, float_status *s)
996 {
997     return floatx80_compare(a, b, s) <= float_relation_equal;
998 }
999 
1000 static inline bool floatx80_lt(floatx80 a, floatx80 b, float_status *s)
1001 {
1002     return floatx80_compare(a, b, s) < float_relation_equal;
1003 }
1004 
1005 static inline bool floatx80_unordered(floatx80 a, floatx80 b, float_status *s)
1006 {
1007     return floatx80_compare(a, b, s) == float_relation_unordered;
1008 }
1009 
1010 static inline bool floatx80_eq_quiet(floatx80 a, floatx80 b, float_status *s)
1011 {
1012     return floatx80_compare_quiet(a, b, s) == float_relation_equal;
1013 }
1014 
1015 static inline bool floatx80_le_quiet(floatx80 a, floatx80 b, float_status *s)
1016 {
1017     return floatx80_compare_quiet(a, b, s) <= float_relation_equal;
1018 }
1019 
1020 static inline bool floatx80_lt_quiet(floatx80 a, floatx80 b, float_status *s)
1021 {
1022     return floatx80_compare_quiet(a, b, s) < float_relation_equal;
1023 }
1024 
1025 static inline bool floatx80_unordered_quiet(floatx80 a, floatx80 b,
1026                                            float_status *s)
1027 {
1028     return floatx80_compare_quiet(a, b, s) == float_relation_unordered;
1029 }
1030 
1031 /*----------------------------------------------------------------------------
1032 | Return whether the given value is an invalid floatx80 encoding.
1033 | Invalid floatx80 encodings arise when the integer bit is not set, but
1034 | the exponent is not zero. The only times the integer bit is permitted to
1035 | be zero is in subnormal numbers and the value zero.
1036 | This includes what the Intel software developer's manual calls pseudo-NaNs,
1037 | pseudo-infinities and un-normal numbers. It does not include
1038 | pseudo-denormals, which must still be correctly handled as inputs even
1039 | if they are never generated as outputs.
1040 *----------------------------------------------------------------------------*/
1041 static inline bool floatx80_invalid_encoding(floatx80 a)
1042 {
1043 #if defined(TARGET_M68K)
1044     /*-------------------------------------------------------------------------
1045     | With m68k, the explicit integer bit can be zero in the case of:
1046     | - zeros                (exp == 0, mantissa == 0)
1047     | - denormalized numbers (exp == 0, mantissa != 0)
1048     | - unnormalized numbers (exp != 0, exp < 0x7FFF)
1049     | - infinities           (exp == 0x7FFF, mantissa == 0)
1050     | - not-a-numbers        (exp == 0x7FFF, mantissa != 0)
1051     |
1052     | For infinities and NaNs, the explicit integer bit can be either one or
1053     | zero.
1054     |
1055     | The IEEE 754 standard does not define a zero integer bit. Such a number
1056     | is an unnormalized number. Hardware does not directly support
1057     | denormalized and unnormalized numbers, but implicitly supports them by
1058     | trapping them as unimplemented data types, allowing efficient conversion
1059     | in software.
1060     |
1061     | See "M68000 FAMILY PROGRAMMER’S REFERENCE MANUAL",
1062     |     "1.6 FLOATING-POINT DATA TYPES"
1063     *------------------------------------------------------------------------*/
1064     return false;
1065 #else
1066     return (a.low & (1ULL << 63)) == 0 && (a.high & 0x7FFF) != 0;
1067 #endif
1068 }
1069 
1070 #define floatx80_zero make_floatx80(0x0000, 0x0000000000000000LL)
1071 #define floatx80_zero_init make_floatx80_init(0x0000, 0x0000000000000000LL)
1072 #define floatx80_one make_floatx80(0x3fff, 0x8000000000000000LL)
1073 #define floatx80_ln2 make_floatx80(0x3ffe, 0xb17217f7d1cf79acLL)
1074 #define floatx80_pi make_floatx80(0x4000, 0xc90fdaa22168c235LL)
1075 #define floatx80_half make_floatx80(0x3ffe, 0x8000000000000000LL)
1076 
1077 /*----------------------------------------------------------------------------
1078 | Returns the fraction bits of the extended double-precision floating-point
1079 | value `a'.
1080 *----------------------------------------------------------------------------*/
1081 
1082 static inline uint64_t extractFloatx80Frac(floatx80 a)
1083 {
1084     return a.low;
1085 }
1086 
1087 /*----------------------------------------------------------------------------
1088 | Returns the exponent bits of the extended double-precision floating-point
1089 | value `a'.
1090 *----------------------------------------------------------------------------*/
1091 
1092 static inline int32_t extractFloatx80Exp(floatx80 a)
1093 {
1094     return a.high & 0x7FFF;
1095 }
1096 
1097 /*----------------------------------------------------------------------------
1098 | Returns the sign bit of the extended double-precision floating-point value
1099 | `a'.
1100 *----------------------------------------------------------------------------*/
1101 
1102 static inline bool extractFloatx80Sign(floatx80 a)
1103 {
1104     return a.high >> 15;
1105 }
1106 
1107 /*----------------------------------------------------------------------------
1108 | Packs the sign `zSign', exponent `zExp', and significand `zSig' into an
1109 | extended double-precision floating-point value, returning the result.
1110 *----------------------------------------------------------------------------*/
1111 
1112 static inline floatx80 packFloatx80(bool zSign, int32_t zExp, uint64_t zSig)
1113 {
1114     floatx80 z;
1115 
1116     z.low = zSig;
1117     z.high = (((uint16_t)zSign) << 15) + zExp;
1118     return z;
1119 }
1120 
1121 /*----------------------------------------------------------------------------
1122 | Normalizes the subnormal extended double-precision floating-point value
1123 | represented by the denormalized significand `aSig'.  The normalized exponent
1124 | and significand are stored at the locations pointed to by `zExpPtr' and
1125 | `zSigPtr', respectively.
1126 *----------------------------------------------------------------------------*/
1127 
1128 void normalizeFloatx80Subnormal(uint64_t aSig, int32_t *zExpPtr,
1129                                 uint64_t *zSigPtr);
1130 
1131 /*----------------------------------------------------------------------------
1132 | Takes two extended double-precision floating-point values `a' and `b', one
1133 | of which is a NaN, and returns the appropriate NaN result.  If either `a' or
1134 | `b' is a signaling NaN, the invalid exception is raised.
1135 *----------------------------------------------------------------------------*/
1136 
1137 floatx80 propagateFloatx80NaN(floatx80 a, floatx80 b, float_status *status);
1138 
1139 /*----------------------------------------------------------------------------
1140 | Takes an abstract floating-point value having sign `zSign', exponent `zExp',
1141 | and extended significand formed by the concatenation of `zSig0' and `zSig1',
1142 | and returns the proper extended double-precision floating-point value
1143 | corresponding to the abstract input.  Ordinarily, the abstract value is
1144 | rounded and packed into the extended double-precision format, with the
1145 | inexact exception raised if the abstract input cannot be represented
1146 | exactly.  However, if the abstract value is too large, the overflow and
1147 | inexact exceptions are raised and an infinity or maximal finite value is
1148 | returned.  If the abstract value is too small, the input value is rounded to
1149 | a subnormal number, and the underflow and inexact exceptions are raised if
1150 | the abstract input cannot be represented exactly as a subnormal extended
1151 | double-precision floating-point number.
1152 |     If `roundingPrecision' is 32 or 64, the result is rounded to the same
1153 | number of bits as single or double precision, respectively.  Otherwise, the
1154 | result is rounded to the full precision of the extended double-precision
1155 | format.
1156 |     The input significand must be normalized or smaller.  If the input
1157 | significand is not normalized, `zExp' must be 0; in that case, the result
1158 | returned is a subnormal number, and it must not require rounding.  The
1159 | handling of underflow and overflow follows the IEC/IEEE Standard for Binary
1160 | Floating-Point Arithmetic.
1161 *----------------------------------------------------------------------------*/
1162 
1163 floatx80 roundAndPackFloatx80(FloatX80RoundPrec roundingPrecision, bool zSign,
1164                               int32_t zExp, uint64_t zSig0, uint64_t zSig1,
1165                               float_status *status);
1166 
1167 /*----------------------------------------------------------------------------
1168 | Takes an abstract floating-point value having sign `zSign', exponent
1169 | `zExp', and significand formed by the concatenation of `zSig0' and `zSig1',
1170 | and returns the proper extended double-precision floating-point value
1171 | corresponding to the abstract input.  This routine is just like
1172 | `roundAndPackFloatx80' except that the input significand does not have to be
1173 | normalized.
1174 *----------------------------------------------------------------------------*/
1175 
1176 floatx80 normalizeRoundAndPackFloatx80(FloatX80RoundPrec roundingPrecision,
1177                                        bool zSign, int32_t zExp,
1178                                        uint64_t zSig0, uint64_t zSig1,
1179                                        float_status *status);
1180 
1181 /*----------------------------------------------------------------------------
1182 | The pattern for a default generated extended double-precision NaN.
1183 *----------------------------------------------------------------------------*/
1184 floatx80 floatx80_default_nan(float_status *status);
1185 
1186 /*----------------------------------------------------------------------------
1187 | Software IEC/IEEE quadruple-precision conversion routines.
1188 *----------------------------------------------------------------------------*/
1189 int32_t float128_to_int32(float128, float_status *status);
1190 int32_t float128_to_int32_round_to_zero(float128, float_status *status);
1191 int64_t float128_to_int64(float128, float_status *status);
1192 int64_t float128_to_int64_round_to_zero(float128, float_status *status);
1193 uint64_t float128_to_uint64(float128, float_status *status);
1194 uint64_t float128_to_uint64_round_to_zero(float128, float_status *status);
1195 uint32_t float128_to_uint32(float128, float_status *status);
1196 uint32_t float128_to_uint32_round_to_zero(float128, float_status *status);
1197 float32 float128_to_float32(float128, float_status *status);
1198 float64 float128_to_float64(float128, float_status *status);
1199 floatx80 float128_to_floatx80(float128, float_status *status);
1200 
1201 /*----------------------------------------------------------------------------
1202 | Software IEC/IEEE quadruple-precision operations.
1203 *----------------------------------------------------------------------------*/
1204 float128 float128_round_to_int(float128, float_status *status);
1205 float128 float128_add(float128, float128, float_status *status);
1206 float128 float128_sub(float128, float128, float_status *status);
1207 float128 float128_mul(float128, float128, float_status *status);
1208 float128 float128_muladd(float128, float128, float128, int,
1209                          float_status *status);
1210 float128 float128_div(float128, float128, float_status *status);
1211 float128 float128_rem(float128, float128, float_status *status);
1212 float128 float128_sqrt(float128, float_status *status);
1213 FloatRelation float128_compare(float128, float128, float_status *status);
1214 FloatRelation float128_compare_quiet(float128, float128, float_status *status);
1215 float128 float128_min(float128, float128, float_status *status);
1216 float128 float128_max(float128, float128, float_status *status);
1217 float128 float128_minnum(float128, float128, float_status *status);
1218 float128 float128_maxnum(float128, float128, float_status *status);
1219 float128 float128_minnummag(float128, float128, float_status *status);
1220 float128 float128_maxnummag(float128, float128, float_status *status);
1221 float128 float128_minimum_number(float128, float128, float_status *status);
1222 float128 float128_maximum_number(float128, float128, float_status *status);
1223 bool float128_is_quiet_nan(float128, float_status *status);
1224 bool float128_is_signaling_nan(float128, float_status *status);
1225 float128 float128_silence_nan(float128, float_status *status);
1226 float128 float128_scalbn(float128, int, float_status *status);
1227 
1228 static inline float128 float128_abs(float128 a)
1229 {
1230     a.high &= 0x7fffffffffffffffLL;
1231     return a;
1232 }
1233 
1234 static inline float128 float128_chs(float128 a)
1235 {
1236     a.high ^= 0x8000000000000000LL;
1237     return a;
1238 }
1239 
1240 static inline bool float128_is_infinity(float128 a)
1241 {
1242     return (a.high & 0x7fffffffffffffffLL) == 0x7fff000000000000LL && a.low == 0;
1243 }
1244 
1245 static inline bool float128_is_neg(float128 a)
1246 {
1247     return a.high >> 63;
1248 }
1249 
1250 static inline bool float128_is_zero(float128 a)
1251 {
1252     return (a.high & 0x7fffffffffffffffLL) == 0 && a.low == 0;
1253 }
1254 
1255 static inline bool float128_is_zero_or_denormal(float128 a)
1256 {
1257     return (a.high & 0x7fff000000000000LL) == 0;
1258 }
1259 
1260 static inline bool float128_is_normal(float128 a)
1261 {
1262     return (((a.high >> 48) + 1) & 0x7fff) >= 2;
1263 }
1264 
1265 static inline bool float128_is_denormal(float128 a)
1266 {
1267     return float128_is_zero_or_denormal(a) && !float128_is_zero(a);
1268 }
1269 
1270 static inline bool float128_is_any_nan(float128 a)
1271 {
1272     return ((a.high >> 48) & 0x7fff) == 0x7fff &&
1273         ((a.low != 0) || ((a.high & 0xffffffffffffLL) != 0));
1274 }
1275 
1276 static inline bool float128_eq(float128 a, float128 b, float_status *s)
1277 {
1278     return float128_compare(a, b, s) == float_relation_equal;
1279 }
1280 
1281 static inline bool float128_le(float128 a, float128 b, float_status *s)
1282 {
1283     return float128_compare(a, b, s) <= float_relation_equal;
1284 }
1285 
1286 static inline bool float128_lt(float128 a, float128 b, float_status *s)
1287 {
1288     return float128_compare(a, b, s) < float_relation_equal;
1289 }
1290 
1291 static inline bool float128_unordered(float128 a, float128 b, float_status *s)
1292 {
1293     return float128_compare(a, b, s) == float_relation_unordered;
1294 }
1295 
1296 static inline bool float128_eq_quiet(float128 a, float128 b, float_status *s)
1297 {
1298     return float128_compare_quiet(a, b, s) == float_relation_equal;
1299 }
1300 
1301 static inline bool float128_le_quiet(float128 a, float128 b, float_status *s)
1302 {
1303     return float128_compare_quiet(a, b, s) <= float_relation_equal;
1304 }
1305 
1306 static inline bool float128_lt_quiet(float128 a, float128 b, float_status *s)
1307 {
1308     return float128_compare_quiet(a, b, s) < float_relation_equal;
1309 }
1310 
1311 static inline bool float128_unordered_quiet(float128 a, float128 b,
1312                                            float_status *s)
1313 {
1314     return float128_compare_quiet(a, b, s) == float_relation_unordered;
1315 }
1316 
1317 #define float128_zero make_float128(0, 0)
1318 
1319 /*----------------------------------------------------------------------------
1320 | The pattern for a default generated quadruple-precision NaN.
1321 *----------------------------------------------------------------------------*/
1322 float128 float128_default_nan(float_status *status);
1323 
1324 #endif /* SOFTFLOAT_H */
1325