xref: /openbmc/qemu/include/fpu/softfloat.h (revision e6d34aee)
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 #define LIT64( a ) a##LL
86 
87 /*----------------------------------------------------------------------------
88 | Software IEC/IEEE floating-point ordering relations
89 *----------------------------------------------------------------------------*/
90 enum {
91     float_relation_less      = -1,
92     float_relation_equal     =  0,
93     float_relation_greater   =  1,
94     float_relation_unordered =  2
95 };
96 
97 #include "fpu/softfloat-types.h"
98 
99 static inline void set_float_detect_tininess(int val, float_status *status)
100 {
101     status->float_detect_tininess = val;
102 }
103 static inline void set_float_rounding_mode(int val, float_status *status)
104 {
105     status->float_rounding_mode = val;
106 }
107 static inline void set_float_exception_flags(int val, float_status *status)
108 {
109     status->float_exception_flags = val;
110 }
111 static inline void set_floatx80_rounding_precision(int val,
112                                                    float_status *status)
113 {
114     status->floatx80_rounding_precision = val;
115 }
116 static inline void set_flush_to_zero(flag val, float_status *status)
117 {
118     status->flush_to_zero = val;
119 }
120 static inline void set_flush_inputs_to_zero(flag val, float_status *status)
121 {
122     status->flush_inputs_to_zero = val;
123 }
124 static inline void set_default_nan_mode(flag val, float_status *status)
125 {
126     status->default_nan_mode = val;
127 }
128 static inline void set_snan_bit_is_one(flag val, float_status *status)
129 {
130     status->snan_bit_is_one = val;
131 }
132 static inline int get_float_detect_tininess(float_status *status)
133 {
134     return status->float_detect_tininess;
135 }
136 static inline int get_float_rounding_mode(float_status *status)
137 {
138     return status->float_rounding_mode;
139 }
140 static inline int get_float_exception_flags(float_status *status)
141 {
142     return status->float_exception_flags;
143 }
144 static inline int get_floatx80_rounding_precision(float_status *status)
145 {
146     return status->floatx80_rounding_precision;
147 }
148 static inline flag get_flush_to_zero(float_status *status)
149 {
150     return status->flush_to_zero;
151 }
152 static inline flag get_flush_inputs_to_zero(float_status *status)
153 {
154     return status->flush_inputs_to_zero;
155 }
156 static inline flag get_default_nan_mode(float_status *status)
157 {
158     return status->default_nan_mode;
159 }
160 
161 /*----------------------------------------------------------------------------
162 | Routine to raise any or all of the software IEC/IEEE floating-point
163 | exception flags.
164 *----------------------------------------------------------------------------*/
165 void float_raise(uint8_t flags, float_status *status);
166 
167 /*----------------------------------------------------------------------------
168 | If `a' is denormal and we are in flush-to-zero mode then set the
169 | input-denormal exception and return zero. Otherwise just return the value.
170 *----------------------------------------------------------------------------*/
171 float16 float16_squash_input_denormal(float16 a, float_status *status);
172 float32 float32_squash_input_denormal(float32 a, float_status *status);
173 float64 float64_squash_input_denormal(float64 a, float_status *status);
174 
175 /*----------------------------------------------------------------------------
176 | Options to indicate which negations to perform in float*_muladd()
177 | Using these differs from negating an input or output before calling
178 | the muladd function in that this means that a NaN doesn't have its
179 | sign bit inverted before it is propagated.
180 | We also support halving the result before rounding, as a special
181 | case to support the ARM fused-sqrt-step instruction FRSQRTS.
182 *----------------------------------------------------------------------------*/
183 enum {
184     float_muladd_negate_c = 1,
185     float_muladd_negate_product = 2,
186     float_muladd_negate_result = 4,
187     float_muladd_halve_result = 8,
188 };
189 
190 /*----------------------------------------------------------------------------
191 | Software IEC/IEEE integer-to-floating-point conversion routines.
192 *----------------------------------------------------------------------------*/
193 
194 float16 int16_to_float16_scalbn(int16_t a, int, float_status *status);
195 float16 int32_to_float16_scalbn(int32_t a, int, float_status *status);
196 float16 int64_to_float16_scalbn(int64_t a, int, float_status *status);
197 float16 uint16_to_float16_scalbn(uint16_t a, int, float_status *status);
198 float16 uint32_to_float16_scalbn(uint32_t a, int, float_status *status);
199 float16 uint64_to_float16_scalbn(uint64_t a, int, float_status *status);
200 
201 float16 int16_to_float16(int16_t a, float_status *status);
202 float16 int32_to_float16(int32_t a, float_status *status);
203 float16 int64_to_float16(int64_t a, float_status *status);
204 float16 uint16_to_float16(uint16_t a, float_status *status);
205 float16 uint32_to_float16(uint32_t a, float_status *status);
206 float16 uint64_to_float16(uint64_t a, float_status *status);
207 
208 float32 int16_to_float32_scalbn(int16_t, int, float_status *status);
209 float32 int32_to_float32_scalbn(int32_t, int, float_status *status);
210 float32 int64_to_float32_scalbn(int64_t, int, float_status *status);
211 float32 uint16_to_float32_scalbn(uint16_t, int, float_status *status);
212 float32 uint32_to_float32_scalbn(uint32_t, int, float_status *status);
213 float32 uint64_to_float32_scalbn(uint64_t, int, float_status *status);
214 
215 float32 int16_to_float32(int16_t, float_status *status);
216 float32 int32_to_float32(int32_t, float_status *status);
217 float32 int64_to_float32(int64_t, float_status *status);
218 float32 uint16_to_float32(uint16_t, float_status *status);
219 float32 uint32_to_float32(uint32_t, float_status *status);
220 float32 uint64_to_float32(uint64_t, float_status *status);
221 
222 float64 int16_to_float64_scalbn(int16_t, int, float_status *status);
223 float64 int32_to_float64_scalbn(int32_t, int, float_status *status);
224 float64 int64_to_float64_scalbn(int64_t, int, float_status *status);
225 float64 uint16_to_float64_scalbn(uint16_t, int, float_status *status);
226 float64 uint32_to_float64_scalbn(uint32_t, int, float_status *status);
227 float64 uint64_to_float64_scalbn(uint64_t, int, float_status *status);
228 
229 float64 int16_to_float64(int16_t, float_status *status);
230 float64 int32_to_float64(int32_t, float_status *status);
231 float64 int64_to_float64(int64_t, float_status *status);
232 float64 uint16_to_float64(uint16_t, float_status *status);
233 float64 uint32_to_float64(uint32_t, float_status *status);
234 float64 uint64_to_float64(uint64_t, float_status *status);
235 
236 floatx80 int32_to_floatx80(int32_t, float_status *status);
237 floatx80 int64_to_floatx80(int64_t, float_status *status);
238 
239 float128 int32_to_float128(int32_t, float_status *status);
240 float128 int64_to_float128(int64_t, float_status *status);
241 float128 uint64_to_float128(uint64_t, float_status *status);
242 
243 /*----------------------------------------------------------------------------
244 | Software half-precision conversion routines.
245 *----------------------------------------------------------------------------*/
246 
247 float16 float32_to_float16(float32, bool ieee, float_status *status);
248 float32 float16_to_float32(float16, bool ieee, float_status *status);
249 float16 float64_to_float16(float64 a, bool ieee, float_status *status);
250 float64 float16_to_float64(float16 a, bool ieee, float_status *status);
251 
252 int16_t float16_to_int16_scalbn(float16, int, int, float_status *status);
253 int32_t float16_to_int32_scalbn(float16, int, int, float_status *status);
254 int64_t float16_to_int64_scalbn(float16, int, int, float_status *status);
255 
256 int16_t float16_to_int16(float16, float_status *status);
257 int32_t float16_to_int32(float16, float_status *status);
258 int64_t float16_to_int64(float16, float_status *status);
259 
260 int16_t float16_to_int16_round_to_zero(float16, float_status *status);
261 int32_t float16_to_int32_round_to_zero(float16, float_status *status);
262 int64_t float16_to_int64_round_to_zero(float16, float_status *status);
263 
264 uint16_t float16_to_uint16_scalbn(float16 a, int, int, float_status *status);
265 uint32_t float16_to_uint32_scalbn(float16 a, int, int, float_status *status);
266 uint64_t float16_to_uint64_scalbn(float16 a, int, int, float_status *status);
267 
268 uint16_t float16_to_uint16(float16 a, float_status *status);
269 uint32_t float16_to_uint32(float16 a, float_status *status);
270 uint64_t float16_to_uint64(float16 a, float_status *status);
271 
272 uint16_t float16_to_uint16_round_to_zero(float16 a, float_status *status);
273 uint32_t float16_to_uint32_round_to_zero(float16 a, float_status *status);
274 uint64_t float16_to_uint64_round_to_zero(float16 a, float_status *status);
275 
276 /*----------------------------------------------------------------------------
277 | Software half-precision operations.
278 *----------------------------------------------------------------------------*/
279 
280 float16 float16_round_to_int(float16, float_status *status);
281 float16 float16_add(float16, float16, float_status *status);
282 float16 float16_sub(float16, float16, float_status *status);
283 float16 float16_mul(float16, float16, float_status *status);
284 float16 float16_muladd(float16, float16, float16, int, float_status *status);
285 float16 float16_div(float16, float16, float_status *status);
286 float16 float16_scalbn(float16, int, float_status *status);
287 float16 float16_min(float16, float16, float_status *status);
288 float16 float16_max(float16, float16, float_status *status);
289 float16 float16_minnum(float16, float16, float_status *status);
290 float16 float16_maxnum(float16, float16, float_status *status);
291 float16 float16_minnummag(float16, float16, float_status *status);
292 float16 float16_maxnummag(float16, float16, float_status *status);
293 float16 float16_sqrt(float16, float_status *status);
294 int float16_compare(float16, float16, float_status *status);
295 int float16_compare_quiet(float16, float16, float_status *status);
296 
297 int float16_is_quiet_nan(float16, float_status *status);
298 int float16_is_signaling_nan(float16, float_status *status);
299 float16 float16_silence_nan(float16, float_status *status);
300 
301 static inline int float16_is_any_nan(float16 a)
302 {
303     return ((float16_val(a) & ~0x8000) > 0x7c00);
304 }
305 
306 static inline int float16_is_neg(float16 a)
307 {
308     return float16_val(a) >> 15;
309 }
310 
311 static inline int float16_is_infinity(float16 a)
312 {
313     return (float16_val(a) & 0x7fff) == 0x7c00;
314 }
315 
316 static inline int float16_is_zero(float16 a)
317 {
318     return (float16_val(a) & 0x7fff) == 0;
319 }
320 
321 static inline int float16_is_zero_or_denormal(float16 a)
322 {
323     return (float16_val(a) & 0x7c00) == 0;
324 }
325 
326 static inline float16 float16_abs(float16 a)
327 {
328     /* Note that abs does *not* handle NaN specially, nor does
329      * it flush denormal inputs to zero.
330      */
331     return make_float16(float16_val(a) & 0x7fff);
332 }
333 
334 static inline float16 float16_chs(float16 a)
335 {
336     /* Note that chs does *not* handle NaN specially, nor does
337      * it flush denormal inputs to zero.
338      */
339     return make_float16(float16_val(a) ^ 0x8000);
340 }
341 
342 static inline float16 float16_set_sign(float16 a, int sign)
343 {
344     return make_float16((float16_val(a) & 0x7fff) | (sign << 15));
345 }
346 
347 #define float16_zero make_float16(0)
348 #define float16_half make_float16(0x3800)
349 #define float16_one make_float16(0x3c00)
350 #define float16_one_point_five make_float16(0x3e00)
351 #define float16_two make_float16(0x4000)
352 #define float16_three make_float16(0x4200)
353 #define float16_infinity make_float16(0x7c00)
354 
355 /*----------------------------------------------------------------------------
356 | The pattern for a default generated half-precision NaN.
357 *----------------------------------------------------------------------------*/
358 float16 float16_default_nan(float_status *status);
359 
360 /*----------------------------------------------------------------------------
361 | Software IEC/IEEE single-precision conversion routines.
362 *----------------------------------------------------------------------------*/
363 
364 int16_t float32_to_int16_scalbn(float32, int, int, float_status *status);
365 int32_t float32_to_int32_scalbn(float32, int, int, float_status *status);
366 int64_t float32_to_int64_scalbn(float32, int, int, float_status *status);
367 
368 int16_t float32_to_int16(float32, float_status *status);
369 int32_t float32_to_int32(float32, float_status *status);
370 int64_t float32_to_int64(float32, float_status *status);
371 
372 int16_t float32_to_int16_round_to_zero(float32, float_status *status);
373 int32_t float32_to_int32_round_to_zero(float32, float_status *status);
374 int64_t float32_to_int64_round_to_zero(float32, float_status *status);
375 
376 uint16_t float32_to_uint16_scalbn(float32, int, int, float_status *status);
377 uint32_t float32_to_uint32_scalbn(float32, int, int, float_status *status);
378 uint64_t float32_to_uint64_scalbn(float32, int, int, float_status *status);
379 
380 uint16_t float32_to_uint16(float32, float_status *status);
381 uint32_t float32_to_uint32(float32, float_status *status);
382 uint64_t float32_to_uint64(float32, float_status *status);
383 
384 uint16_t float32_to_uint16_round_to_zero(float32, float_status *status);
385 uint32_t float32_to_uint32_round_to_zero(float32, float_status *status);
386 uint64_t float32_to_uint64_round_to_zero(float32, float_status *status);
387 
388 float64 float32_to_float64(float32, float_status *status);
389 floatx80 float32_to_floatx80(float32, float_status *status);
390 float128 float32_to_float128(float32, float_status *status);
391 
392 /*----------------------------------------------------------------------------
393 | Software IEC/IEEE single-precision operations.
394 *----------------------------------------------------------------------------*/
395 float32 float32_round_to_int(float32, float_status *status);
396 float32 float32_add(float32, float32, float_status *status);
397 float32 float32_sub(float32, float32, float_status *status);
398 float32 float32_mul(float32, float32, float_status *status);
399 float32 float32_div(float32, float32, float_status *status);
400 float32 float32_rem(float32, float32, float_status *status);
401 float32 float32_muladd(float32, float32, float32, int, float_status *status);
402 float32 float32_sqrt(float32, float_status *status);
403 float32 float32_exp2(float32, float_status *status);
404 float32 float32_log2(float32, float_status *status);
405 int float32_eq(float32, float32, float_status *status);
406 int float32_le(float32, float32, float_status *status);
407 int float32_lt(float32, float32, float_status *status);
408 int float32_unordered(float32, float32, float_status *status);
409 int float32_eq_quiet(float32, float32, float_status *status);
410 int float32_le_quiet(float32, float32, float_status *status);
411 int float32_lt_quiet(float32, float32, float_status *status);
412 int float32_unordered_quiet(float32, float32, float_status *status);
413 int float32_compare(float32, float32, float_status *status);
414 int float32_compare_quiet(float32, float32, float_status *status);
415 float32 float32_min(float32, float32, float_status *status);
416 float32 float32_max(float32, float32, float_status *status);
417 float32 float32_minnum(float32, float32, float_status *status);
418 float32 float32_maxnum(float32, float32, float_status *status);
419 float32 float32_minnummag(float32, float32, float_status *status);
420 float32 float32_maxnummag(float32, float32, float_status *status);
421 int float32_is_quiet_nan(float32, float_status *status);
422 int float32_is_signaling_nan(float32, float_status *status);
423 float32 float32_silence_nan(float32, float_status *status);
424 float32 float32_scalbn(float32, int, float_status *status);
425 
426 static inline float32 float32_abs(float32 a)
427 {
428     /* Note that abs does *not* handle NaN specially, nor does
429      * it flush denormal inputs to zero.
430      */
431     return make_float32(float32_val(a) & 0x7fffffff);
432 }
433 
434 static inline float32 float32_chs(float32 a)
435 {
436     /* Note that chs does *not* handle NaN specially, nor does
437      * it flush denormal inputs to zero.
438      */
439     return make_float32(float32_val(a) ^ 0x80000000);
440 }
441 
442 static inline int float32_is_infinity(float32 a)
443 {
444     return (float32_val(a) & 0x7fffffff) == 0x7f800000;
445 }
446 
447 static inline int float32_is_neg(float32 a)
448 {
449     return float32_val(a) >> 31;
450 }
451 
452 static inline int float32_is_zero(float32 a)
453 {
454     return (float32_val(a) & 0x7fffffff) == 0;
455 }
456 
457 static inline int float32_is_any_nan(float32 a)
458 {
459     return ((float32_val(a) & ~(1 << 31)) > 0x7f800000UL);
460 }
461 
462 static inline int float32_is_zero_or_denormal(float32 a)
463 {
464     return (float32_val(a) & 0x7f800000) == 0;
465 }
466 
467 static inline float32 float32_set_sign(float32 a, int sign)
468 {
469     return make_float32((float32_val(a) & 0x7fffffff) | (sign << 31));
470 }
471 
472 #define float32_zero make_float32(0)
473 #define float32_half make_float32(0x3f000000)
474 #define float32_one make_float32(0x3f800000)
475 #define float32_one_point_five make_float32(0x3fc00000)
476 #define float32_two make_float32(0x40000000)
477 #define float32_three make_float32(0x40400000)
478 #define float32_infinity make_float32(0x7f800000)
479 
480 /*----------------------------------------------------------------------------
481 | Packs the sign `zSign', exponent `zExp', and significand `zSig' into a
482 | single-precision floating-point value, returning the result.  After being
483 | shifted into the proper positions, the three fields are simply added
484 | together to form the result.  This means that any integer portion of `zSig'
485 | will be added into the exponent.  Since a properly normalized significand
486 | will have an integer portion equal to 1, the `zExp' input should be 1 less
487 | than the desired result exponent whenever `zSig' is a complete, normalized
488 | significand.
489 *----------------------------------------------------------------------------*/
490 
491 static inline float32 packFloat32(flag zSign, int zExp, uint32_t zSig)
492 {
493     return make_float32(
494           (((uint32_t)zSign) << 31) + (((uint32_t)zExp) << 23) + zSig);
495 }
496 
497 /*----------------------------------------------------------------------------
498 | The pattern for a default generated single-precision NaN.
499 *----------------------------------------------------------------------------*/
500 float32 float32_default_nan(float_status *status);
501 
502 /*----------------------------------------------------------------------------
503 | Software IEC/IEEE double-precision conversion routines.
504 *----------------------------------------------------------------------------*/
505 
506 int16_t float64_to_int16_scalbn(float64, int, int, float_status *status);
507 int32_t float64_to_int32_scalbn(float64, int, int, float_status *status);
508 int64_t float64_to_int64_scalbn(float64, int, int, float_status *status);
509 
510 int16_t float64_to_int16(float64, float_status *status);
511 int32_t float64_to_int32(float64, float_status *status);
512 int64_t float64_to_int64(float64, float_status *status);
513 
514 int16_t float64_to_int16_round_to_zero(float64, float_status *status);
515 int32_t float64_to_int32_round_to_zero(float64, float_status *status);
516 int64_t float64_to_int64_round_to_zero(float64, float_status *status);
517 
518 uint16_t float64_to_uint16_scalbn(float64, int, int, float_status *status);
519 uint32_t float64_to_uint32_scalbn(float64, int, int, float_status *status);
520 uint64_t float64_to_uint64_scalbn(float64, int, int, float_status *status);
521 
522 uint16_t float64_to_uint16(float64, float_status *status);
523 uint32_t float64_to_uint32(float64, float_status *status);
524 uint64_t float64_to_uint64(float64, float_status *status);
525 
526 uint16_t float64_to_uint16_round_to_zero(float64, float_status *status);
527 uint32_t float64_to_uint32_round_to_zero(float64, float_status *status);
528 uint64_t float64_to_uint64_round_to_zero(float64, float_status *status);
529 
530 float32 float64_to_float32(float64, float_status *status);
531 floatx80 float64_to_floatx80(float64, float_status *status);
532 float128 float64_to_float128(float64, float_status *status);
533 
534 /*----------------------------------------------------------------------------
535 | Software IEC/IEEE double-precision operations.
536 *----------------------------------------------------------------------------*/
537 float64 float64_round_to_int(float64, float_status *status);
538 float64 float64_add(float64, float64, float_status *status);
539 float64 float64_sub(float64, float64, float_status *status);
540 float64 float64_mul(float64, float64, float_status *status);
541 float64 float64_div(float64, float64, float_status *status);
542 float64 float64_rem(float64, float64, float_status *status);
543 float64 float64_muladd(float64, float64, float64, int, float_status *status);
544 float64 float64_sqrt(float64, float_status *status);
545 float64 float64_log2(float64, float_status *status);
546 int float64_eq(float64, float64, float_status *status);
547 int float64_le(float64, float64, float_status *status);
548 int float64_lt(float64, float64, float_status *status);
549 int float64_unordered(float64, float64, float_status *status);
550 int float64_eq_quiet(float64, float64, float_status *status);
551 int float64_le_quiet(float64, float64, float_status *status);
552 int float64_lt_quiet(float64, float64, float_status *status);
553 int float64_unordered_quiet(float64, float64, float_status *status);
554 int float64_compare(float64, float64, float_status *status);
555 int float64_compare_quiet(float64, float64, float_status *status);
556 float64 float64_min(float64, float64, float_status *status);
557 float64 float64_max(float64, float64, float_status *status);
558 float64 float64_minnum(float64, float64, float_status *status);
559 float64 float64_maxnum(float64, float64, float_status *status);
560 float64 float64_minnummag(float64, float64, float_status *status);
561 float64 float64_maxnummag(float64, float64, float_status *status);
562 int float64_is_quiet_nan(float64 a, float_status *status);
563 int float64_is_signaling_nan(float64, float_status *status);
564 float64 float64_silence_nan(float64, float_status *status);
565 float64 float64_scalbn(float64, int, float_status *status);
566 
567 static inline float64 float64_abs(float64 a)
568 {
569     /* Note that abs does *not* handle NaN specially, nor does
570      * it flush denormal inputs to zero.
571      */
572     return make_float64(float64_val(a) & 0x7fffffffffffffffLL);
573 }
574 
575 static inline float64 float64_chs(float64 a)
576 {
577     /* Note that chs does *not* handle NaN specially, nor does
578      * it flush denormal inputs to zero.
579      */
580     return make_float64(float64_val(a) ^ 0x8000000000000000LL);
581 }
582 
583 static inline int float64_is_infinity(float64 a)
584 {
585     return (float64_val(a) & 0x7fffffffffffffffLL ) == 0x7ff0000000000000LL;
586 }
587 
588 static inline int float64_is_neg(float64 a)
589 {
590     return float64_val(a) >> 63;
591 }
592 
593 static inline int float64_is_zero(float64 a)
594 {
595     return (float64_val(a) & 0x7fffffffffffffffLL) == 0;
596 }
597 
598 static inline int float64_is_any_nan(float64 a)
599 {
600     return ((float64_val(a) & ~(1ULL << 63)) > 0x7ff0000000000000ULL);
601 }
602 
603 static inline int float64_is_zero_or_denormal(float64 a)
604 {
605     return (float64_val(a) & 0x7ff0000000000000LL) == 0;
606 }
607 
608 static inline float64 float64_set_sign(float64 a, int sign)
609 {
610     return make_float64((float64_val(a) & 0x7fffffffffffffffULL)
611                         | ((int64_t)sign << 63));
612 }
613 
614 #define float64_zero make_float64(0)
615 #define float64_half make_float64(0x3fe0000000000000LL)
616 #define float64_one make_float64(0x3ff0000000000000LL)
617 #define float64_one_point_five make_float64(0x3FF8000000000000ULL)
618 #define float64_two make_float64(0x4000000000000000ULL)
619 #define float64_three make_float64(0x4008000000000000ULL)
620 #define float64_ln2 make_float64(0x3fe62e42fefa39efLL)
621 #define float64_infinity make_float64(0x7ff0000000000000LL)
622 
623 /*----------------------------------------------------------------------------
624 | The pattern for a default generated double-precision NaN.
625 *----------------------------------------------------------------------------*/
626 float64 float64_default_nan(float_status *status);
627 
628 /*----------------------------------------------------------------------------
629 | Software IEC/IEEE extended double-precision conversion routines.
630 *----------------------------------------------------------------------------*/
631 int32_t floatx80_to_int32(floatx80, float_status *status);
632 int32_t floatx80_to_int32_round_to_zero(floatx80, float_status *status);
633 int64_t floatx80_to_int64(floatx80, float_status *status);
634 int64_t floatx80_to_int64_round_to_zero(floatx80, float_status *status);
635 float32 floatx80_to_float32(floatx80, float_status *status);
636 float64 floatx80_to_float64(floatx80, float_status *status);
637 float128 floatx80_to_float128(floatx80, float_status *status);
638 
639 /*----------------------------------------------------------------------------
640 | The pattern for an extended double-precision inf.
641 *----------------------------------------------------------------------------*/
642 extern const floatx80 floatx80_infinity;
643 
644 /*----------------------------------------------------------------------------
645 | Software IEC/IEEE extended double-precision operations.
646 *----------------------------------------------------------------------------*/
647 floatx80 floatx80_round(floatx80 a, float_status *status);
648 floatx80 floatx80_round_to_int(floatx80, float_status *status);
649 floatx80 floatx80_add(floatx80, floatx80, float_status *status);
650 floatx80 floatx80_sub(floatx80, floatx80, float_status *status);
651 floatx80 floatx80_mul(floatx80, floatx80, float_status *status);
652 floatx80 floatx80_div(floatx80, floatx80, float_status *status);
653 floatx80 floatx80_rem(floatx80, floatx80, float_status *status);
654 floatx80 floatx80_sqrt(floatx80, float_status *status);
655 int floatx80_eq(floatx80, floatx80, float_status *status);
656 int floatx80_le(floatx80, floatx80, float_status *status);
657 int floatx80_lt(floatx80, floatx80, float_status *status);
658 int floatx80_unordered(floatx80, floatx80, float_status *status);
659 int floatx80_eq_quiet(floatx80, floatx80, float_status *status);
660 int floatx80_le_quiet(floatx80, floatx80, float_status *status);
661 int floatx80_lt_quiet(floatx80, floatx80, float_status *status);
662 int floatx80_unordered_quiet(floatx80, floatx80, float_status *status);
663 int floatx80_compare(floatx80, floatx80, float_status *status);
664 int floatx80_compare_quiet(floatx80, floatx80, float_status *status);
665 int floatx80_is_quiet_nan(floatx80, float_status *status);
666 int floatx80_is_signaling_nan(floatx80, float_status *status);
667 floatx80 floatx80_silence_nan(floatx80, float_status *status);
668 floatx80 floatx80_scalbn(floatx80, int, float_status *status);
669 
670 static inline floatx80 floatx80_abs(floatx80 a)
671 {
672     a.high &= 0x7fff;
673     return a;
674 }
675 
676 static inline floatx80 floatx80_chs(floatx80 a)
677 {
678     a.high ^= 0x8000;
679     return a;
680 }
681 
682 static inline int floatx80_is_infinity(floatx80 a)
683 {
684 #if defined(TARGET_M68K)
685     return (a.high & 0x7fff) == floatx80_infinity.high && !(a.low << 1);
686 #else
687     return (a.high & 0x7fff) == floatx80_infinity.high &&
688                        a.low == floatx80_infinity.low;
689 #endif
690 }
691 
692 static inline int floatx80_is_neg(floatx80 a)
693 {
694     return a.high >> 15;
695 }
696 
697 static inline int floatx80_is_zero(floatx80 a)
698 {
699     return (a.high & 0x7fff) == 0 && a.low == 0;
700 }
701 
702 static inline int floatx80_is_zero_or_denormal(floatx80 a)
703 {
704     return (a.high & 0x7fff) == 0;
705 }
706 
707 static inline int floatx80_is_any_nan(floatx80 a)
708 {
709     return ((a.high & 0x7fff) == 0x7fff) && (a.low<<1);
710 }
711 
712 /*----------------------------------------------------------------------------
713 | Return whether the given value is an invalid floatx80 encoding.
714 | Invalid floatx80 encodings arise when the integer bit is not set, but
715 | the exponent is not zero. The only times the integer bit is permitted to
716 | be zero is in subnormal numbers and the value zero.
717 | This includes what the Intel software developer's manual calls pseudo-NaNs,
718 | pseudo-infinities and un-normal numbers. It does not include
719 | pseudo-denormals, which must still be correctly handled as inputs even
720 | if they are never generated as outputs.
721 *----------------------------------------------------------------------------*/
722 static inline bool floatx80_invalid_encoding(floatx80 a)
723 {
724     return (a.low & (1ULL << 63)) == 0 && (a.high & 0x7FFF) != 0;
725 }
726 
727 #define floatx80_zero make_floatx80(0x0000, 0x0000000000000000LL)
728 #define floatx80_one make_floatx80(0x3fff, 0x8000000000000000LL)
729 #define floatx80_ln2 make_floatx80(0x3ffe, 0xb17217f7d1cf79acLL)
730 #define floatx80_pi make_floatx80(0x4000, 0xc90fdaa22168c235LL)
731 #define floatx80_half make_floatx80(0x3ffe, 0x8000000000000000LL)
732 
733 /*----------------------------------------------------------------------------
734 | Returns the fraction bits of the extended double-precision floating-point
735 | value `a'.
736 *----------------------------------------------------------------------------*/
737 
738 static inline uint64_t extractFloatx80Frac(floatx80 a)
739 {
740     return a.low;
741 }
742 
743 /*----------------------------------------------------------------------------
744 | Returns the exponent bits of the extended double-precision floating-point
745 | value `a'.
746 *----------------------------------------------------------------------------*/
747 
748 static inline int32_t extractFloatx80Exp(floatx80 a)
749 {
750     return a.high & 0x7FFF;
751 }
752 
753 /*----------------------------------------------------------------------------
754 | Returns the sign bit of the extended double-precision floating-point value
755 | `a'.
756 *----------------------------------------------------------------------------*/
757 
758 static inline flag extractFloatx80Sign(floatx80 a)
759 {
760     return a.high >> 15;
761 }
762 
763 /*----------------------------------------------------------------------------
764 | Packs the sign `zSign', exponent `zExp', and significand `zSig' into an
765 | extended double-precision floating-point value, returning the result.
766 *----------------------------------------------------------------------------*/
767 
768 static inline floatx80 packFloatx80(flag zSign, int32_t zExp, uint64_t zSig)
769 {
770     floatx80 z;
771 
772     z.low = zSig;
773     z.high = (((uint16_t)zSign) << 15) + zExp;
774     return z;
775 }
776 
777 /*----------------------------------------------------------------------------
778 | Normalizes the subnormal extended double-precision floating-point value
779 | represented by the denormalized significand `aSig'.  The normalized exponent
780 | and significand are stored at the locations pointed to by `zExpPtr' and
781 | `zSigPtr', respectively.
782 *----------------------------------------------------------------------------*/
783 
784 void normalizeFloatx80Subnormal(uint64_t aSig, int32_t *zExpPtr,
785                                 uint64_t *zSigPtr);
786 
787 /*----------------------------------------------------------------------------
788 | Takes two extended double-precision floating-point values `a' and `b', one
789 | of which is a NaN, and returns the appropriate NaN result.  If either `a' or
790 | `b' is a signaling NaN, the invalid exception is raised.
791 *----------------------------------------------------------------------------*/
792 
793 floatx80 propagateFloatx80NaN(floatx80 a, floatx80 b, float_status *status);
794 
795 /*----------------------------------------------------------------------------
796 | Takes an abstract floating-point value having sign `zSign', exponent `zExp',
797 | and extended significand formed by the concatenation of `zSig0' and `zSig1',
798 | and returns the proper extended double-precision floating-point value
799 | corresponding to the abstract input.  Ordinarily, the abstract value is
800 | rounded and packed into the extended double-precision format, with the
801 | inexact exception raised if the abstract input cannot be represented
802 | exactly.  However, if the abstract value is too large, the overflow and
803 | inexact exceptions are raised and an infinity or maximal finite value is
804 | returned.  If the abstract value is too small, the input value is rounded to
805 | a subnormal number, and the underflow and inexact exceptions are raised if
806 | the abstract input cannot be represented exactly as a subnormal extended
807 | double-precision floating-point number.
808 |     If `roundingPrecision' is 32 or 64, the result is rounded to the same
809 | number of bits as single or double precision, respectively.  Otherwise, the
810 | result is rounded to the full precision of the extended double-precision
811 | format.
812 |     The input significand must be normalized or smaller.  If the input
813 | significand is not normalized, `zExp' must be 0; in that case, the result
814 | returned is a subnormal number, and it must not require rounding.  The
815 | handling of underflow and overflow follows the IEC/IEEE Standard for Binary
816 | Floating-Point Arithmetic.
817 *----------------------------------------------------------------------------*/
818 
819 floatx80 roundAndPackFloatx80(int8_t roundingPrecision, flag zSign,
820                               int32_t zExp, uint64_t zSig0, uint64_t zSig1,
821                               float_status *status);
822 
823 /*----------------------------------------------------------------------------
824 | Takes an abstract floating-point value having sign `zSign', exponent
825 | `zExp', and significand formed by the concatenation of `zSig0' and `zSig1',
826 | and returns the proper extended double-precision floating-point value
827 | corresponding to the abstract input.  This routine is just like
828 | `roundAndPackFloatx80' except that the input significand does not have to be
829 | normalized.
830 *----------------------------------------------------------------------------*/
831 
832 floatx80 normalizeRoundAndPackFloatx80(int8_t roundingPrecision,
833                                        flag zSign, int32_t zExp,
834                                        uint64_t zSig0, uint64_t zSig1,
835                                        float_status *status);
836 
837 /*----------------------------------------------------------------------------
838 | The pattern for a default generated extended double-precision NaN.
839 *----------------------------------------------------------------------------*/
840 floatx80 floatx80_default_nan(float_status *status);
841 
842 /*----------------------------------------------------------------------------
843 | Software IEC/IEEE quadruple-precision conversion routines.
844 *----------------------------------------------------------------------------*/
845 int32_t float128_to_int32(float128, float_status *status);
846 int32_t float128_to_int32_round_to_zero(float128, float_status *status);
847 int64_t float128_to_int64(float128, float_status *status);
848 int64_t float128_to_int64_round_to_zero(float128, float_status *status);
849 uint64_t float128_to_uint64(float128, float_status *status);
850 uint64_t float128_to_uint64_round_to_zero(float128, float_status *status);
851 uint32_t float128_to_uint32_round_to_zero(float128, float_status *status);
852 float32 float128_to_float32(float128, float_status *status);
853 float64 float128_to_float64(float128, float_status *status);
854 floatx80 float128_to_floatx80(float128, float_status *status);
855 
856 /*----------------------------------------------------------------------------
857 | Software IEC/IEEE quadruple-precision operations.
858 *----------------------------------------------------------------------------*/
859 float128 float128_round_to_int(float128, float_status *status);
860 float128 float128_add(float128, float128, float_status *status);
861 float128 float128_sub(float128, float128, float_status *status);
862 float128 float128_mul(float128, float128, float_status *status);
863 float128 float128_div(float128, float128, float_status *status);
864 float128 float128_rem(float128, float128, float_status *status);
865 float128 float128_sqrt(float128, float_status *status);
866 int float128_eq(float128, float128, float_status *status);
867 int float128_le(float128, float128, float_status *status);
868 int float128_lt(float128, float128, float_status *status);
869 int float128_unordered(float128, float128, float_status *status);
870 int float128_eq_quiet(float128, float128, float_status *status);
871 int float128_le_quiet(float128, float128, float_status *status);
872 int float128_lt_quiet(float128, float128, float_status *status);
873 int float128_unordered_quiet(float128, float128, float_status *status);
874 int float128_compare(float128, float128, float_status *status);
875 int float128_compare_quiet(float128, float128, float_status *status);
876 int float128_is_quiet_nan(float128, float_status *status);
877 int float128_is_signaling_nan(float128, float_status *status);
878 float128 float128_silence_nan(float128, float_status *status);
879 float128 float128_scalbn(float128, int, float_status *status);
880 
881 static inline float128 float128_abs(float128 a)
882 {
883     a.high &= 0x7fffffffffffffffLL;
884     return a;
885 }
886 
887 static inline float128 float128_chs(float128 a)
888 {
889     a.high ^= 0x8000000000000000LL;
890     return a;
891 }
892 
893 static inline int float128_is_infinity(float128 a)
894 {
895     return (a.high & 0x7fffffffffffffffLL) == 0x7fff000000000000LL && a.low == 0;
896 }
897 
898 static inline int float128_is_neg(float128 a)
899 {
900     return a.high >> 63;
901 }
902 
903 static inline int float128_is_zero(float128 a)
904 {
905     return (a.high & 0x7fffffffffffffffLL) == 0 && a.low == 0;
906 }
907 
908 static inline int float128_is_zero_or_denormal(float128 a)
909 {
910     return (a.high & 0x7fff000000000000LL) == 0;
911 }
912 
913 static inline int float128_is_any_nan(float128 a)
914 {
915     return ((a.high >> 48) & 0x7fff) == 0x7fff &&
916         ((a.low != 0) || ((a.high & 0xffffffffffffLL) != 0));
917 }
918 
919 #define float128_zero make_float128(0, 0)
920 
921 /*----------------------------------------------------------------------------
922 | The pattern for a default generated quadruple-precision NaN.
923 *----------------------------------------------------------------------------*/
924 float128 float128_default_nan(float_status *status);
925 
926 #endif /* SOFTFLOAT_H */
927