source: mainline/uspace/lib/softfloat/conversion.c@ df6ded8

lfn serial ticket/834-toolchain-update topic/msim-upgrade topic/simplify-dev-export
Last change on this file since df6ded8 was c0c38c7c, checked in by Martin Decky <martin@…>, 10 years ago

software floating point overhaul
use proper type mapping
fix cosine calculation
  • Property mode set to 100644
File size: 27.8 KB
Line 
1/*
2 * Copyright (c) 2005 Josef Cejka
3 * Copyright (c) 2011 Petr Koupy
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 *
10 * - Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * - Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * - The name of the author may not be used to endorse or promote products
16 * derived from this software without specific prior written permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 */
29
30/** @addtogroup softfloat
31 * @{
32 */
33/** @file Conversion of precision and conversion between integers and floats.
34 */
35
36#include "conversion.h"
37#include "comparison.h"
38#include "common.h"
39
40float64 float32_to_float64(float32 a)
41{
42 float64 result;
43 uint64_t frac;
44
45 result.parts.sign = a.parts.sign;
46 result.parts.fraction = a.parts.fraction;
47 result.parts.fraction <<= (FLOAT64_FRACTION_SIZE - FLOAT32_FRACTION_SIZE);
48
49 if ((is_float32_infinity(a)) || (is_float32_nan(a))) {
50 result.parts.exp = FLOAT64_MAX_EXPONENT;
51 // TODO; check if its correct for SigNaNs
52 return result;
53 }
54
55 result.parts.exp = a.parts.exp + ((int) FLOAT64_BIAS - FLOAT32_BIAS);
56 if (a.parts.exp == 0) {
57 /* normalize denormalized numbers */
58
59 if (result.parts.fraction == 0) { /* fix zero */
60 result.parts.exp = 0;
61 return result;
62 }
63
64 frac = result.parts.fraction;
65
66 while (!(frac & FLOAT64_HIDDEN_BIT_MASK)) {
67 frac <<= 1;
68 --result.parts.exp;
69 }
70
71 ++result.parts.exp;
72 result.parts.fraction = frac;
73 }
74
75 return result;
76}
77
78float128 float32_to_float128(float32 a)
79{
80 float128 result;
81 uint64_t frac_hi, frac_lo;
82 uint64_t tmp_hi, tmp_lo;
83
84 result.parts.sign = a.parts.sign;
85 result.parts.frac_hi = 0;
86 result.parts.frac_lo = a.parts.fraction;
87 lshift128(result.parts.frac_hi, result.parts.frac_lo,
88 (FLOAT128_FRACTION_SIZE - FLOAT32_FRACTION_SIZE),
89 &frac_hi, &frac_lo);
90 result.parts.frac_hi = frac_hi;
91 result.parts.frac_lo = frac_lo;
92
93 if ((is_float32_infinity(a)) || (is_float32_nan(a))) {
94 result.parts.exp = FLOAT128_MAX_EXPONENT;
95 // TODO; check if its correct for SigNaNs
96 return result;
97 }
98
99 result.parts.exp = a.parts.exp + ((int) FLOAT128_BIAS - FLOAT32_BIAS);
100 if (a.parts.exp == 0) {
101 /* normalize denormalized numbers */
102
103 if (eq128(result.parts.frac_hi,
104 result.parts.frac_lo, 0x0ll, 0x0ll)) { /* fix zero */
105 result.parts.exp = 0;
106 return result;
107 }
108
109 frac_hi = result.parts.frac_hi;
110 frac_lo = result.parts.frac_lo;
111
112 and128(frac_hi, frac_lo,
113 FLOAT128_HIDDEN_BIT_MASK_HI, FLOAT128_HIDDEN_BIT_MASK_LO,
114 &tmp_hi, &tmp_lo);
115 while (!lt128(0x0ll, 0x0ll, tmp_hi, tmp_lo)) {
116 lshift128(frac_hi, frac_lo, 1, &frac_hi, &frac_lo);
117 --result.parts.exp;
118 }
119
120 ++result.parts.exp;
121 result.parts.frac_hi = frac_hi;
122 result.parts.frac_lo = frac_lo;
123 }
124
125 return result;
126}
127
128float128 float64_to_float128(float64 a)
129{
130 float128 result;
131 uint64_t frac_hi, frac_lo;
132 uint64_t tmp_hi, tmp_lo;
133
134 result.parts.sign = a.parts.sign;
135 result.parts.frac_hi = 0;
136 result.parts.frac_lo = a.parts.fraction;
137 lshift128(result.parts.frac_hi, result.parts.frac_lo,
138 (FLOAT128_FRACTION_SIZE - FLOAT64_FRACTION_SIZE),
139 &frac_hi, &frac_lo);
140 result.parts.frac_hi = frac_hi;
141 result.parts.frac_lo = frac_lo;
142
143 if ((is_float64_infinity(a)) || (is_float64_nan(a))) {
144 result.parts.exp = FLOAT128_MAX_EXPONENT;
145 // TODO; check if its correct for SigNaNs
146 return result;
147 }
148
149 result.parts.exp = a.parts.exp + ((int) FLOAT128_BIAS - FLOAT64_BIAS);
150 if (a.parts.exp == 0) {
151 /* normalize denormalized numbers */
152
153 if (eq128(result.parts.frac_hi,
154 result.parts.frac_lo, 0x0ll, 0x0ll)) { /* fix zero */
155 result.parts.exp = 0;
156 return result;
157 }
158
159 frac_hi = result.parts.frac_hi;
160 frac_lo = result.parts.frac_lo;
161
162 and128(frac_hi, frac_lo,
163 FLOAT128_HIDDEN_BIT_MASK_HI, FLOAT128_HIDDEN_BIT_MASK_LO,
164 &tmp_hi, &tmp_lo);
165 while (!lt128(0x0ll, 0x0ll, tmp_hi, tmp_lo)) {
166 lshift128(frac_hi, frac_lo, 1, &frac_hi, &frac_lo);
167 --result.parts.exp;
168 }
169
170 ++result.parts.exp;
171 result.parts.frac_hi = frac_hi;
172 result.parts.frac_lo = frac_lo;
173 }
174
175 return result;
176}
177
178float32 float64_to_float32(float64 a)
179{
180 float32 result;
181 int32_t exp;
182 uint64_t frac;
183
184 result.parts.sign = a.parts.sign;
185
186 if (is_float64_nan(a)) {
187 result.parts.exp = FLOAT32_MAX_EXPONENT;
188
189 if (is_float64_signan(a)) {
190 /* set first bit of fraction nonzero */
191 result.parts.fraction = FLOAT32_HIDDEN_BIT_MASK >> 1;
192 return result;
193 }
194
195 /* fraction nonzero but its first bit is zero */
196 result.parts.fraction = 0x1;
197 return result;
198 }
199
200 if (is_float64_infinity(a)) {
201 result.parts.fraction = 0;
202 result.parts.exp = FLOAT32_MAX_EXPONENT;
203 return result;
204 }
205
206 exp = (int) a.parts.exp - FLOAT64_BIAS + FLOAT32_BIAS;
207
208 if (exp >= FLOAT32_MAX_EXPONENT) {
209 /* FIXME: overflow */
210 result.parts.fraction = 0;
211 result.parts.exp = FLOAT32_MAX_EXPONENT;
212 return result;
213 } else if (exp <= 0) {
214 /* underflow or denormalized */
215
216 result.parts.exp = 0;
217
218 exp *= -1;
219 if (exp > FLOAT32_FRACTION_SIZE) {
220 /* FIXME: underflow */
221 result.parts.fraction = 0;
222 return result;
223 }
224
225 /* denormalized */
226
227 frac = a.parts.fraction;
228 frac |= FLOAT64_HIDDEN_BIT_MASK; /* denormalize and set hidden bit */
229
230 frac >>= (FLOAT64_FRACTION_SIZE - FLOAT32_FRACTION_SIZE + 1);
231
232 while (exp > 0) {
233 --exp;
234 frac >>= 1;
235 }
236 result.parts.fraction = frac;
237
238 return result;
239 }
240
241 result.parts.exp = exp;
242 result.parts.fraction =
243 a.parts.fraction >> (FLOAT64_FRACTION_SIZE - FLOAT32_FRACTION_SIZE);
244 return result;
245}
246
247float32 float128_to_float32(float128 a)
248{
249 float32 result;
250 int32_t exp;
251 uint64_t frac_hi, frac_lo;
252
253 result.parts.sign = a.parts.sign;
254
255 if (is_float128_nan(a)) {
256 result.parts.exp = FLOAT32_MAX_EXPONENT;
257
258 if (is_float128_signan(a)) {
259 /* set first bit of fraction nonzero */
260 result.parts.fraction = FLOAT32_HIDDEN_BIT_MASK >> 1;
261 return result;
262 }
263
264 /* fraction nonzero but its first bit is zero */
265 result.parts.fraction = 0x1;
266 return result;
267 }
268
269 if (is_float128_infinity(a)) {
270 result.parts.fraction = 0;
271 result.parts.exp = FLOAT32_MAX_EXPONENT;
272 return result;
273 }
274
275 exp = (int) a.parts.exp - FLOAT128_BIAS + FLOAT32_BIAS;
276
277 if (exp >= FLOAT32_MAX_EXPONENT) {
278 /* FIXME: overflow */
279 result.parts.fraction = 0;
280 result.parts.exp = FLOAT32_MAX_EXPONENT;
281 return result;
282 } else if (exp <= 0) {
283 /* underflow or denormalized */
284
285 result.parts.exp = 0;
286
287 exp *= -1;
288 if (exp > FLOAT32_FRACTION_SIZE) {
289 /* FIXME: underflow */
290 result.parts.fraction = 0;
291 return result;
292 }
293
294 /* denormalized */
295
296 frac_hi = a.parts.frac_hi;
297 frac_lo = a.parts.frac_lo;
298
299 /* denormalize and set hidden bit */
300 frac_hi |= FLOAT128_HIDDEN_BIT_MASK_HI;
301
302 rshift128(frac_hi, frac_lo,
303 (FLOAT128_FRACTION_SIZE - FLOAT32_FRACTION_SIZE + 1),
304 &frac_hi, &frac_lo);
305
306 while (exp > 0) {
307 --exp;
308 rshift128(frac_hi, frac_lo, 1, &frac_hi, &frac_lo);
309 }
310 result.parts.fraction = frac_lo;
311
312 return result;
313 }
314
315 result.parts.exp = exp;
316 frac_hi = a.parts.frac_hi;
317 frac_lo = a.parts.frac_lo;
318 rshift128(frac_hi, frac_lo,
319 (FLOAT128_FRACTION_SIZE - FLOAT32_FRACTION_SIZE + 1),
320 &frac_hi, &frac_lo);
321 result.parts.fraction = frac_lo;
322 return result;
323}
324
325float64 float128_to_float64(float128 a)
326{
327 float64 result;
328 int32_t exp;
329 uint64_t frac_hi, frac_lo;
330
331 result.parts.sign = a.parts.sign;
332
333 if (is_float128_nan(a)) {
334 result.parts.exp = FLOAT64_MAX_EXPONENT;
335
336 if (is_float128_signan(a)) {
337 /* set first bit of fraction nonzero */
338 result.parts.fraction = FLOAT64_HIDDEN_BIT_MASK >> 1;
339 return result;
340 }
341
342 /* fraction nonzero but its first bit is zero */
343 result.parts.fraction = 0x1;
344 return result;
345 }
346
347 if (is_float128_infinity(a)) {
348 result.parts.fraction = 0;
349 result.parts.exp = FLOAT64_MAX_EXPONENT;
350 return result;
351 }
352
353 exp = (int) a.parts.exp - FLOAT128_BIAS + FLOAT64_BIAS;
354
355 if (exp >= FLOAT64_MAX_EXPONENT) {
356 /* FIXME: overflow */
357 result.parts.fraction = 0;
358 result.parts.exp = FLOAT64_MAX_EXPONENT;
359 return result;
360 } else if (exp <= 0) {
361 /* underflow or denormalized */
362
363 result.parts.exp = 0;
364
365 exp *= -1;
366 if (exp > FLOAT64_FRACTION_SIZE) {
367 /* FIXME: underflow */
368 result.parts.fraction = 0;
369 return result;
370 }
371
372 /* denormalized */
373
374 frac_hi = a.parts.frac_hi;
375 frac_lo = a.parts.frac_lo;
376
377 /* denormalize and set hidden bit */
378 frac_hi |= FLOAT128_HIDDEN_BIT_MASK_HI;
379
380 rshift128(frac_hi, frac_lo,
381 (FLOAT128_FRACTION_SIZE - FLOAT64_FRACTION_SIZE + 1),
382 &frac_hi, &frac_lo);
383
384 while (exp > 0) {
385 --exp;
386 rshift128(frac_hi, frac_lo, 1, &frac_hi, &frac_lo);
387 }
388 result.parts.fraction = frac_lo;
389
390 return result;
391 }
392
393 result.parts.exp = exp;
394 frac_hi = a.parts.frac_hi;
395 frac_lo = a.parts.frac_lo;
396 rshift128(frac_hi, frac_lo,
397 (FLOAT128_FRACTION_SIZE - FLOAT64_FRACTION_SIZE + 1),
398 &frac_hi, &frac_lo);
399 result.parts.fraction = frac_lo;
400 return result;
401}
402
403/** Helper procedure for converting float32 to uint32.
404 *
405 * @param a Floating point number in normalized form
406 * (NaNs or Inf are not checked).
407 * @return Converted unsigned integer.
408 */
409static uint32_t _float32_to_uint32_helper(float32 a)
410{
411 uint32_t frac;
412
413 if (a.parts.exp < FLOAT32_BIAS) {
414 /* TODO: rounding */
415 return 0;
416 }
417
418 frac = a.parts.fraction;
419
420 frac |= FLOAT32_HIDDEN_BIT_MASK;
421 /* shift fraction to left so hidden bit will be the most significant bit */
422 frac <<= 32 - FLOAT32_FRACTION_SIZE - 1;
423
424 frac >>= 32 - (a.parts.exp - FLOAT32_BIAS) - 1;
425 if ((a.parts.sign == 1) && (frac != 0)) {
426 frac = ~frac;
427 ++frac;
428 }
429
430 return frac;
431}
432
433/*
434 * FIXME: Im not sure what to return if overflow/underflow happens
435 * - now its the biggest or the smallest int
436 */
437uint32_t float32_to_uint32(float32 a)
438{
439 if (is_float32_nan(a))
440 return UINT32_MAX;
441
442 if (is_float32_infinity(a) || (a.parts.exp >= (32 + FLOAT32_BIAS))) {
443 if (a.parts.sign)
444 return UINT32_MIN;
445
446 return UINT32_MAX;
447 }
448
449 return _float32_to_uint32_helper(a);
450}
451
452/*
453 * FIXME: Im not sure what to return if overflow/underflow happens
454 * - now its the biggest or the smallest int
455 */
456int32_t float32_to_int32(float32 a)
457{
458 if (is_float32_nan(a))
459 return INT32_MAX;
460
461 if (is_float32_infinity(a) || (a.parts.exp >= (32 + FLOAT32_BIAS))) {
462 if (a.parts.sign)
463 return INT32_MIN;
464
465 return INT32_MAX;
466 }
467
468 return _float32_to_uint32_helper(a);
469}
470
471/** Helper procedure for converting float32 to uint64.
472 *
473 * @param a Floating point number in normalized form
474 * (NaNs or Inf are not checked).
475 * @return Converted unsigned integer.
476 */
477static uint64_t _float32_to_uint64_helper(float32 a)
478{
479 uint64_t frac;
480
481 if (a.parts.exp < FLOAT32_BIAS) {
482 // TODO: rounding
483 return 0;
484 }
485
486 frac = a.parts.fraction;
487
488 frac |= FLOAT32_HIDDEN_BIT_MASK;
489 /* shift fraction to left so hidden bit will be the most significant bit */
490 frac <<= 64 - FLOAT32_FRACTION_SIZE - 1;
491
492 frac >>= 64 - (a.parts.exp - FLOAT32_BIAS) - 1;
493 if ((a.parts.sign == 1) && (frac != 0)) {
494 frac = ~frac;
495 ++frac;
496 }
497
498 return frac;
499}
500
501/*
502 * FIXME: Im not sure what to return if overflow/underflow happens
503 * - now its the biggest or the smallest int
504 */
505uint64_t float32_to_uint64(float32 a)
506{
507 if (is_float32_nan(a))
508 return UINT64_MAX;
509
510 if (is_float32_infinity(a) || (a.parts.exp >= (64 + FLOAT32_BIAS))) {
511 if (a.parts.sign)
512 return UINT64_MIN;
513
514 return UINT64_MAX;
515 }
516
517 return _float32_to_uint64_helper(a);
518}
519
520/*
521 * FIXME: Im not sure what to return if overflow/underflow happens
522 * - now its the biggest or the smallest int
523 */
524int64_t float32_to_int64(float32 a)
525{
526 if (is_float32_nan(a))
527 return INT64_MAX;
528
529 if (is_float32_infinity(a) || (a.parts.exp >= (64 + FLOAT32_BIAS))) {
530 if (a.parts.sign)
531 return INT64_MIN;
532
533 return INT64_MAX;
534 }
535
536 return _float32_to_uint64_helper(a);
537}
538
539/** Helper procedure for converting float64 to uint64.
540 *
541 * @param a Floating point number in normalized form
542 * (NaNs or Inf are not checked).
543 * @return Converted unsigned integer.
544 */
545static uint64_t _float64_to_uint64_helper(float64 a)
546{
547 uint64_t frac;
548
549 if (a.parts.exp < FLOAT64_BIAS) {
550 // TODO: rounding
551 return 0;
552 }
553
554 frac = a.parts.fraction;
555
556 frac |= FLOAT64_HIDDEN_BIT_MASK;
557 /* shift fraction to left so hidden bit will be the most significant bit */
558 frac <<= 64 - FLOAT64_FRACTION_SIZE - 1;
559
560 frac >>= 64 - (a.parts.exp - FLOAT64_BIAS) - 1;
561 if ((a.parts.sign == 1) && (frac != 0)) {
562 frac = ~frac;
563 ++frac;
564 }
565
566 return frac;
567}
568
569/*
570 * FIXME: Im not sure what to return if overflow/underflow happens
571 * - now its the biggest or the smallest int
572 */
573uint32_t float64_to_uint32(float64 a)
574{
575 if (is_float64_nan(a))
576 return UINT32_MAX;
577
578 if (is_float64_infinity(a) || (a.parts.exp >= (32 + FLOAT64_BIAS))) {
579 if (a.parts.sign)
580 return UINT32_MIN;
581
582 return UINT32_MAX;
583 }
584
585 return (uint32_t) _float64_to_uint64_helper(a);
586}
587
588/*
589 * FIXME: Im not sure what to return if overflow/underflow happens
590 * - now its the biggest or the smallest int
591 */
592int32_t float64_to_int32(float64 a)
593{
594 if (is_float64_nan(a))
595 return INT32_MAX;
596
597 if (is_float64_infinity(a) || (a.parts.exp >= (32 + FLOAT64_BIAS))) {
598 if (a.parts.sign)
599 return INT32_MIN;
600
601 return INT32_MAX;
602 }
603
604 return (int32_t) _float64_to_uint64_helper(a);
605}
606
607/*
608 * FIXME: Im not sure what to return if overflow/underflow happens
609 * - now its the biggest or the smallest int
610 */
611uint64_t float64_to_uint64(float64 a)
612{
613 if (is_float64_nan(a))
614 return UINT64_MAX;
615
616 if (is_float64_infinity(a) || (a.parts.exp >= (64 + FLOAT64_BIAS))) {
617 if (a.parts.sign)
618 return UINT64_MIN;
619
620 return UINT64_MAX;
621 }
622
623 return _float64_to_uint64_helper(a);
624}
625
626/*
627 * FIXME: Im not sure what to return if overflow/underflow happens
628 * - now its the biggest or the smallest int
629 */
630int64_t float64_to_int64(float64 a)
631{
632 if (is_float64_nan(a))
633 return INT64_MAX;
634
635 if (is_float64_infinity(a) || (a.parts.exp >= (64 + FLOAT64_BIAS))) {
636 if (a.parts.sign)
637 return INT64_MIN;
638
639 return INT64_MAX;
640 }
641
642 return _float64_to_uint64_helper(a);
643}
644
645/** Helper procedure for converting float128 to uint64.
646 *
647 * @param a Floating point number in normalized form
648 * (NaNs or Inf are not checked).
649 * @return Converted unsigned integer.
650 */
651static uint64_t _float128_to_uint64_helper(float128 a)
652{
653 uint64_t frac_hi, frac_lo;
654
655 if (a.parts.exp < FLOAT128_BIAS) {
656 // TODO: rounding
657 return 0;
658 }
659
660 frac_hi = a.parts.frac_hi;
661 frac_lo = a.parts.frac_lo;
662
663 frac_hi |= FLOAT128_HIDDEN_BIT_MASK_HI;
664 /* shift fraction to left so hidden bit will be the most significant bit */
665 lshift128(frac_hi, frac_lo,
666 (128 - FLOAT128_FRACTION_SIZE - 1), &frac_hi, &frac_lo);
667
668 rshift128(frac_hi, frac_lo,
669 (128 - (a.parts.exp - FLOAT128_BIAS) - 1), &frac_hi, &frac_lo);
670 if ((a.parts.sign == 1) && !eq128(frac_hi, frac_lo, 0x0ll, 0x0ll)) {
671 not128(frac_hi, frac_lo, &frac_hi, &frac_lo);
672 add128(frac_hi, frac_lo, 0x0ll, 0x1ll, &frac_hi, &frac_lo);
673 }
674
675 return frac_lo;
676}
677
678/*
679 * FIXME: Im not sure what to return if overflow/underflow happens
680 * - now its the biggest or the smallest int
681 */
682uint32_t float128_to_uint32(float128 a)
683{
684 if (is_float128_nan(a))
685 return UINT32_MAX;
686
687 if (is_float128_infinity(a) || (a.parts.exp >= (32 + FLOAT128_BIAS))) {
688 if (a.parts.sign)
689 return UINT32_MIN;
690
691 return UINT32_MAX;
692 }
693
694 return (uint32_t) _float128_to_uint64_helper(a);
695}
696
697/*
698 * FIXME: Im not sure what to return if overflow/underflow happens
699 * - now its the biggest or the smallest int
700 */
701int32_t float128_to_int32(float128 a)
702{
703 if (is_float128_nan(a))
704 return INT32_MAX;
705
706 if (is_float128_infinity(a) || (a.parts.exp >= (32 + FLOAT128_BIAS))) {
707 if (a.parts.sign)
708 return INT32_MIN;
709
710 return INT32_MAX;
711 }
712
713 return (int32_t) _float128_to_uint64_helper(a);
714}
715
716/*
717 * FIXME: Im not sure what to return if overflow/underflow happens
718 * - now its the biggest or the smallest int
719 */
720uint64_t float128_to_uint64(float128 a)
721{
722 if (is_float128_nan(a))
723 return UINT64_MAX;
724
725 if (is_float128_infinity(a) || (a.parts.exp >= (64 + FLOAT128_BIAS))) {
726 if (a.parts.sign)
727 return UINT64_MIN;
728
729 return UINT64_MAX;
730 }
731
732 return _float128_to_uint64_helper(a);
733}
734
735/*
736 * FIXME: Im not sure what to return if overflow/underflow happens
737 * - now its the biggest or the smallest int
738 */
739int64_t float128_to_int64(float128 a)
740{
741 if (is_float128_nan(a))
742 return INT64_MAX;
743
744 if (is_float128_infinity(a) || (a.parts.exp >= (64 + FLOAT128_BIAS))) {
745 if (a.parts.sign)
746 return INT64_MIN;
747
748 return INT64_MAX;
749 }
750
751 return _float128_to_uint64_helper(a);
752}
753
754float32 uint32_to_float32(uint32_t i)
755{
756 int counter;
757 int32_t exp;
758 float32 result;
759
760 result.parts.sign = 0;
761 result.parts.fraction = 0;
762
763 counter = count_zeroes32(i);
764
765 exp = FLOAT32_BIAS + 32 - counter - 1;
766
767 if (counter == 32) {
768 result.bin = 0;
769 return result;
770 }
771
772 if (counter > 0) {
773 i <<= counter - 1;
774 } else {
775 i >>= 1;
776 }
777
778 round_float32(&exp, &i);
779
780 result.parts.fraction = i >> (32 - FLOAT32_FRACTION_SIZE - 2);
781 result.parts.exp = exp;
782
783 return result;
784}
785
786float32 int32_to_float32(int32_t i)
787{
788 float32 result;
789
790 if (i < 0)
791 result = uint32_to_float32((uint32_t) (-i));
792 else
793 result = uint32_to_float32((uint32_t) i);
794
795 result.parts.sign = i < 0;
796
797 return result;
798}
799
800float32 uint64_to_float32(uint64_t i)
801{
802 int counter;
803 int32_t exp;
804 uint32_t j;
805 float32 result;
806
807 result.parts.sign = 0;
808 result.parts.fraction = 0;
809
810 counter = count_zeroes64(i);
811
812 exp = FLOAT32_BIAS + 64 - counter - 1;
813
814 if (counter == 64) {
815 result.bin = 0;
816 return result;
817 }
818
819 /* Shift all to the first 31 bits (31st will be hidden 1) */
820 if (counter > 33) {
821 i <<= counter - 1 - 32;
822 } else {
823 i >>= 1 + 32 - counter;
824 }
825
826 j = (uint32_t) i;
827 round_float32(&exp, &j);
828
829 result.parts.fraction = j >> (32 - FLOAT32_FRACTION_SIZE - 2);
830 result.parts.exp = exp;
831 return result;
832}
833
834float32 int64_to_float32(int64_t i)
835{
836 float32 result;
837
838 if (i < 0)
839 result = uint64_to_float32((uint64_t) (-i));
840 else
841 result = uint64_to_float32((uint64_t) i);
842
843 result.parts.sign = i < 0;
844
845 return result;
846}
847
848float64 uint32_to_float64(uint32_t i)
849{
850 int counter;
851 int32_t exp;
852 float64 result;
853 uint64_t frac;
854
855 result.parts.sign = 0;
856 result.parts.fraction = 0;
857
858 counter = count_zeroes32(i);
859
860 exp = FLOAT64_BIAS + 32 - counter - 1;
861
862 if (counter == 32) {
863 result.bin = 0;
864 return result;
865 }
866
867 frac = i;
868 frac <<= counter + 32 - 1;
869
870 round_float64(&exp, &frac);
871
872 result.parts.fraction = frac >> (64 - FLOAT64_FRACTION_SIZE - 2);
873 result.parts.exp = exp;
874
875 return result;
876}
877
878float64 int32_to_float64(int32_t i)
879{
880 float64 result;
881
882 if (i < 0)
883 result = uint32_to_float64((uint32_t) (-i));
884 else
885 result = uint32_to_float64((uint32_t) i);
886
887 result.parts.sign = i < 0;
888
889 return result;
890}
891
892float64 uint64_to_float64(uint64_t i)
893{
894 int counter;
895 int32_t exp;
896 float64 result;
897
898 result.parts.sign = 0;
899 result.parts.fraction = 0;
900
901 counter = count_zeroes64(i);
902
903 exp = FLOAT64_BIAS + 64 - counter - 1;
904
905 if (counter == 64) {
906 result.bin = 0;
907 return result;
908 }
909
910 if (counter > 0) {
911 i <<= counter - 1;
912 } else {
913 i >>= 1;
914 }
915
916 round_float64(&exp, &i);
917
918 result.parts.fraction = i >> (64 - FLOAT64_FRACTION_SIZE - 2);
919 result.parts.exp = exp;
920 return result;
921}
922
923float64 int64_to_float64(int64_t i)
924{
925 float64 result;
926
927 if (i < 0)
928 result = uint64_to_float64((uint64_t) (-i));
929 else
930 result = uint64_to_float64((uint64_t) i);
931
932 result.parts.sign = i < 0;
933
934 return result;
935}
936
937float128 uint32_to_float128(uint32_t i)
938{
939 int counter;
940 int32_t exp;
941 float128 result;
942 uint64_t frac_hi, frac_lo;
943
944 result.parts.sign = 0;
945 result.parts.frac_hi = 0;
946 result.parts.frac_lo = 0;
947
948 counter = count_zeroes32(i);
949
950 exp = FLOAT128_BIAS + 32 - counter - 1;
951
952 if (counter == 32) {
953 result.bin.hi = 0;
954 result.bin.lo = 0;
955 return result;
956 }
957
958 frac_hi = 0;
959 frac_lo = i;
960 lshift128(frac_hi, frac_lo, (counter + 96 - 1), &frac_hi, &frac_lo);
961
962 round_float128(&exp, &frac_hi, &frac_lo);
963
964 rshift128(frac_hi, frac_lo,
965 (128 - FLOAT128_FRACTION_SIZE - 2), &frac_hi, &frac_lo);
966 result.parts.frac_hi = frac_hi;
967 result.parts.frac_lo = frac_lo;
968 result.parts.exp = exp;
969
970 return result;
971}
972
973float128 int32_to_float128(int32_t i)
974{
975 float128 result;
976
977 if (i < 0)
978 result = uint32_to_float128((uint32_t) (-i));
979 else
980 result = uint32_to_float128((uint32_t) i);
981
982 result.parts.sign = i < 0;
983
984 return result;
985}
986
987
988float128 uint64_to_float128(uint64_t i)
989{
990 int counter;
991 int32_t exp;
992 float128 result;
993 uint64_t frac_hi, frac_lo;
994
995 result.parts.sign = 0;
996 result.parts.frac_hi = 0;
997 result.parts.frac_lo = 0;
998
999 counter = count_zeroes64(i);
1000
1001 exp = FLOAT128_BIAS + 64 - counter - 1;
1002
1003 if (counter == 64) {
1004 result.bin.hi = 0;
1005 result.bin.lo = 0;
1006 return result;
1007 }
1008
1009 frac_hi = 0;
1010 frac_lo = i;
1011 lshift128(frac_hi, frac_lo, (counter + 64 - 1), &frac_hi, &frac_lo);
1012
1013 round_float128(&exp, &frac_hi, &frac_lo);
1014
1015 rshift128(frac_hi, frac_lo,
1016 (128 - FLOAT128_FRACTION_SIZE - 2), &frac_hi, &frac_lo);
1017 result.parts.frac_hi = frac_hi;
1018 result.parts.frac_lo = frac_lo;
1019 result.parts.exp = exp;
1020
1021 return result;
1022}
1023
1024float128 int64_to_float128(int64_t i)
1025{
1026 float128 result;
1027
1028 if (i < 0)
1029 result = uint64_to_float128((uint64_t) (-i));
1030 else
1031 result = uint64_to_float128((uint64_t) i);
1032
1033 result.parts.sign = i < 0;
1034
1035 return result;
1036}
1037
1038#ifdef float32_t
1039
1040float32_t __floatsisf(int32_t i)
1041{
1042 float32_u res;
1043 res.data = int32_to_float32(i);
1044
1045 return res.val;
1046}
1047
1048float32_t __floatdisf(int64_t i)
1049{
1050 float32_u res;
1051 res.data = int64_to_float32(i);
1052
1053 return res.val;
1054}
1055
1056float32_t __floatunsisf(uint32_t i)
1057{
1058 float32_u res;
1059 res.data = uint32_to_float32(i);
1060
1061 return res.val;
1062}
1063
1064float32_t __floatundisf(uint64_t i)
1065{
1066 float32_u res;
1067 res.data = uint64_to_float32(i);
1068
1069 return res.val;
1070}
1071
1072int32_t __fixsfsi(float32_t a)
1073{
1074 float32_u ua;
1075 ua.val = a;
1076
1077 return float32_to_int32(ua.data);
1078}
1079
1080int64_t __fixsfdi(float32_t a)
1081{
1082 float32_u ua;
1083 ua.val = a;
1084
1085 return float32_to_int64(ua.data);
1086}
1087
1088uint32_t __fixunssfsi(float32_t a)
1089{
1090 float32_u ua;
1091 ua.val = a;
1092
1093 return float32_to_uint32(ua.data);
1094}
1095
1096uint64_t __fixunssfdi(float32_t a)
1097{
1098 float32_u ua;
1099 ua.val = a;
1100
1101 return float32_to_uint64(ua.data);
1102}
1103
1104int32_t __aeabi_f2iz(float32_t a)
1105{
1106 float32_u ua;
1107 ua.val = a;
1108
1109 return float32_to_int32(ua.data);
1110}
1111
1112uint32_t __aeabi_f2uiz(float32_t a)
1113{
1114 float32_u ua;
1115 ua.val = a;
1116
1117 return float32_to_uint32(ua.data);
1118}
1119
1120float32_t __aeabi_i2f(int32_t i)
1121{
1122 float32_u res;
1123 res.data = int32_to_float32(i);
1124
1125 return res.val;
1126}
1127
1128float32_t __aeabi_l2f(int64_t i)
1129{
1130 float32_u res;
1131 res.data = int64_to_float32(i);
1132
1133 return res.val;
1134}
1135
1136float32_t __aeabi_ui2f(uint32_t i)
1137{
1138 float32_u res;
1139 res.data = uint32_to_float32(i);
1140
1141 return res.val;
1142}
1143
1144float32_t __aeabi_ul2f(uint64_t i)
1145{
1146 float32_u res;
1147 res.data = uint64_to_float32(i);
1148
1149 return res.val;
1150}
1151
1152#endif
1153
1154#ifdef float64_t
1155
1156float64_t __floatsidf(int32_t i)
1157{
1158 float64_u res;
1159 res.data = int32_to_float64(i);
1160
1161 return res.val;
1162}
1163
1164float64_t __floatdidf(int64_t i)
1165{
1166 float64_u res;
1167 res.data = int64_to_float64(i);
1168
1169 return res.val;
1170}
1171
1172float64_t __floatunsidf(uint32_t i)
1173{
1174 float64_u res;
1175 res.data = uint32_to_float64(i);
1176
1177 return res.val;
1178}
1179
1180float64_t __floatundidf(uint64_t i)
1181{
1182 float64_u res;
1183 res.data = uint64_to_float64(i);
1184
1185 return res.val;
1186}
1187
1188uint32_t __fixunsdfsi(float64_t a)
1189{
1190 float64_u ua;
1191 ua.val = a;
1192
1193 return float64_to_uint32(ua.data);
1194}
1195
1196uint64_t __fixunsdfdi(float64_t a)
1197{
1198 float64_u ua;
1199 ua.val = a;
1200
1201 return float64_to_uint64(ua.data);
1202}
1203
1204int32_t __fixdfsi(float64_t a)
1205{
1206 float64_u ua;
1207 ua.val = a;
1208
1209 return float64_to_int32(ua.data);
1210}
1211
1212int64_t __fixdfdi(float64_t a)
1213{
1214 float64_u ua;
1215 ua.val = a;
1216
1217 return float64_to_int64(ua.data);
1218}
1219
1220float64_t __aeabi_i2d(int32_t i)
1221{
1222 float64_u res;
1223 res.data = int32_to_float64(i);
1224
1225 return res.val;
1226}
1227
1228float64_t __aeabi_ui2d(uint32_t i)
1229{
1230 float64_u res;
1231 res.data = uint32_to_float64(i);
1232
1233 return res.val;
1234}
1235
1236float64_t __aeabi_l2d(int64_t i)
1237{
1238 float64_u res;
1239 res.data = int64_to_float64(i);
1240
1241 return res.val;
1242}
1243
1244int32_t __aeabi_d2iz(float64_t a)
1245{
1246 float64_u ua;
1247 ua.val = a;
1248
1249 return float64_to_int32(ua.data);
1250}
1251
1252int64_t __aeabi_d2lz(float64_t a)
1253{
1254 float64_u ua;
1255 ua.val = a;
1256
1257 return float64_to_int64(ua.data);
1258}
1259
1260uint32_t __aeabi_d2uiz(float64_t a)
1261{
1262 float64_u ua;
1263 ua.val = a;
1264
1265 return float64_to_uint32(ua.data);
1266}
1267
1268#endif
1269
1270#ifdef float128_t
1271
1272float128_t __floatsitf(int32_t i)
1273{
1274 float128_u res;
1275 res.data = int32_to_float128(i);
1276
1277 return res.val;
1278}
1279
1280float128_t __floatditf(int64_t i)
1281{
1282 float128_u res;
1283 res.data = int64_to_float128(i);
1284
1285 return res.val;
1286}
1287
1288float128_t __floatunsitf(uint32_t i)
1289{
1290 float128_u res;
1291 res.data = uint32_to_float128(i);
1292
1293 return res.val;
1294}
1295
1296float128_t __floatunditf(uint64_t i)
1297{
1298 float128_u res;
1299 res.data = uint64_to_float128(i);
1300
1301 return res.val;
1302}
1303
1304int32_t __fixtfsi(float128_t a)
1305{
1306 float128_u ua;
1307 ua.val = a;
1308
1309 return float128_to_int32(ua.data);
1310}
1311
1312int64_t __fixtfdi(float128_t a)
1313{
1314 float128_u ua;
1315 ua.val = a;
1316
1317 return float128_to_uint64(ua.data);
1318}
1319
1320uint32_t __fixunstfsi(float128_t a)
1321{
1322 float128_u ua;
1323 ua.val = a;
1324
1325 return float128_to_uint32(ua.data);
1326}
1327
1328uint64_t __fixunstfdi(float128_t a)
1329{
1330 float128_u ua;
1331 ua.val = a;
1332
1333 return float128_to_uint64(ua.data);
1334}
1335
1336int32_t _Qp_qtoi(float128_t *a)
1337{
1338 return __fixtfsi(*a);
1339}
1340
1341int64_t _Qp_qtox(float128_t *a)
1342{
1343 return __fixunstfdi(*a);
1344}
1345
1346uint32_t _Qp_qtoui(float128_t *a)
1347{
1348 return __fixunstfsi(*a);
1349}
1350
1351uint64_t _Qp_qtoux(float128_t *a)
1352{
1353 return __fixunstfdi(*a);
1354}
1355
1356void _Qp_itoq(float128_t *c, int32_t a)
1357{
1358 *c = __floatsitf(a);
1359}
1360
1361void _Qp_xtoq(float128_t *c, int64_t a)
1362{
1363 *c = __floatditf(a);
1364}
1365
1366void _Qp_uitoq(float128_t *c, uint32_t a)
1367{
1368 *c = __floatunsitf(a);
1369}
1370
1371void _Qp_uxtoq(float128_t *c, uint64_t a)
1372{
1373 *c = __floatunditf(a);
1374}
1375
1376#endif
1377
1378#if (defined(float32_t) && defined(float64_t))
1379
1380float32_t __truncdfsf2(float64_t a)
1381{
1382 float64_u ua;
1383 ua.val = a;
1384
1385 float32_u res;
1386 res.data = float64_to_float32(ua.data);
1387
1388 return res.val;
1389}
1390
1391float64_t __extendsfdf2(float32_t a)
1392{
1393 float32_u ua;
1394 ua.val = a;
1395
1396 float64_u res;
1397 res.data = float32_to_float64(ua.data);
1398
1399 return res.val;
1400}
1401
1402float64_t __aeabi_f2d(float32_t a)
1403{
1404 float32_u ua;
1405 ua.val = a;
1406
1407 float64_u res;
1408 res.data = float32_to_float64(ua.data);
1409
1410 return res.val;
1411}
1412
1413float32_t __aeabi_d2f(float64_t a)
1414{
1415 float64_u ua;
1416 ua.val = a;
1417
1418 float32_u res;
1419 res.data = float64_to_float32(ua.data);
1420
1421 return res.val;
1422}
1423
1424#endif
1425
1426#if (defined(float32_t) && defined(float128_t))
1427
1428float32_t __trunctfsf2(float128_t a)
1429{
1430 float128_u ua;
1431 ua.val = a;
1432
1433 float32_u res;
1434 res.data = float128_to_float32(ua.data);
1435
1436 return res.val;
1437}
1438
1439float128_t __extendsftf2(float32_t a)
1440{
1441 float32_u ua;
1442 ua.val = a;
1443
1444 float128_u res;
1445 res.data = float32_to_float128(ua.data);
1446
1447 return res.val;
1448}
1449
1450void _Qp_stoq(float128_t *c, float32_t a)
1451{
1452 *c = __extendsftf2(a);
1453}
1454
1455float32_t _Qp_qtos(float128_t *a)
1456{
1457 return __trunctfsf2(*a);
1458}
1459
1460#endif
1461
1462#if (defined(float64_t) && defined(float128_t))
1463
1464float64_t __trunctfdf2(float128_t a)
1465{
1466 float128_u ua;
1467 ua.val = a;
1468
1469 float64_u res;
1470 res.data = float128_to_float64(ua.data);
1471
1472 return res.val;
1473}
1474
1475float128_t __extenddftf2(float64_t a)
1476{
1477 float64_u ua;
1478 ua.val = a;
1479
1480 float128_u res;
1481 res.data = float64_to_float128(ua.data);
1482
1483 return res.val;
1484}
1485
1486void _Qp_dtoq(float128_t *c, float64_t a)
1487{
1488 *c = __extenddftf2(a);
1489}
1490
1491float64_t _Qp_qtod(float128_t *a)
1492{
1493 return __trunctfdf2(*a);
1494}
1495
1496#endif
1497
1498/** @}
1499 */
Note: See TracBrowser for help on using the repository browser.