]> git.proxmox.com Git - mirror_qemu.git/blob - fpu/softfloat.c
softfloat: Use _Generic instead of QEMU_GENERIC
[mirror_qemu.git] / fpu / softfloat.c
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 source 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 /* softfloat (and in particular the code in softfloat-specialize.h) is
83 * target-dependent and needs the TARGET_* macros.
84 */
85 #include "qemu/osdep.h"
86 #include <math.h>
87 #include "qemu/bitops.h"
88 #include "fpu/softfloat.h"
89
90 /* We only need stdlib for abort() */
91
92 /*----------------------------------------------------------------------------
93 | Primitive arithmetic functions, including multi-word arithmetic, and
94 | division and square root approximations. (Can be specialized to target if
95 | desired.)
96 *----------------------------------------------------------------------------*/
97 #include "fpu/softfloat-macros.h"
98
99 /*
100 * Hardfloat
101 *
102 * Fast emulation of guest FP instructions is challenging for two reasons.
103 * First, FP instruction semantics are similar but not identical, particularly
104 * when handling NaNs. Second, emulating at reasonable speed the guest FP
105 * exception flags is not trivial: reading the host's flags register with a
106 * feclearexcept & fetestexcept pair is slow [slightly slower than soft-fp],
107 * and trapping on every FP exception is not fast nor pleasant to work with.
108 *
109 * We address these challenges by leveraging the host FPU for a subset of the
110 * operations. To do this we expand on the idea presented in this paper:
111 *
112 * Guo, Yu-Chuan, et al. "Translating the ARM Neon and VFP instructions in a
113 * binary translator." Software: Practice and Experience 46.12 (2016):1591-1615.
114 *
115 * The idea is thus to leverage the host FPU to (1) compute FP operations
116 * and (2) identify whether FP exceptions occurred while avoiding
117 * expensive exception flag register accesses.
118 *
119 * An important optimization shown in the paper is that given that exception
120 * flags are rarely cleared by the guest, we can avoid recomputing some flags.
121 * This is particularly useful for the inexact flag, which is very frequently
122 * raised in floating-point workloads.
123 *
124 * We optimize the code further by deferring to soft-fp whenever FP exception
125 * detection might get hairy. Two examples: (1) when at least one operand is
126 * denormal/inf/NaN; (2) when operands are not guaranteed to lead to a 0 result
127 * and the result is < the minimum normal.
128 */
129 #define GEN_INPUT_FLUSH__NOCHECK(name, soft_t) \
130 static inline void name(soft_t *a, float_status *s) \
131 { \
132 if (unlikely(soft_t ## _is_denormal(*a))) { \
133 *a = soft_t ## _set_sign(soft_t ## _zero, \
134 soft_t ## _is_neg(*a)); \
135 float_raise(float_flag_input_denormal, s); \
136 } \
137 }
138
139 GEN_INPUT_FLUSH__NOCHECK(float32_input_flush__nocheck, float32)
140 GEN_INPUT_FLUSH__NOCHECK(float64_input_flush__nocheck, float64)
141 #undef GEN_INPUT_FLUSH__NOCHECK
142
143 #define GEN_INPUT_FLUSH1(name, soft_t) \
144 static inline void name(soft_t *a, float_status *s) \
145 { \
146 if (likely(!s->flush_inputs_to_zero)) { \
147 return; \
148 } \
149 soft_t ## _input_flush__nocheck(a, s); \
150 }
151
152 GEN_INPUT_FLUSH1(float32_input_flush1, float32)
153 GEN_INPUT_FLUSH1(float64_input_flush1, float64)
154 #undef GEN_INPUT_FLUSH1
155
156 #define GEN_INPUT_FLUSH2(name, soft_t) \
157 static inline void name(soft_t *a, soft_t *b, float_status *s) \
158 { \
159 if (likely(!s->flush_inputs_to_zero)) { \
160 return; \
161 } \
162 soft_t ## _input_flush__nocheck(a, s); \
163 soft_t ## _input_flush__nocheck(b, s); \
164 }
165
166 GEN_INPUT_FLUSH2(float32_input_flush2, float32)
167 GEN_INPUT_FLUSH2(float64_input_flush2, float64)
168 #undef GEN_INPUT_FLUSH2
169
170 #define GEN_INPUT_FLUSH3(name, soft_t) \
171 static inline void name(soft_t *a, soft_t *b, soft_t *c, float_status *s) \
172 { \
173 if (likely(!s->flush_inputs_to_zero)) { \
174 return; \
175 } \
176 soft_t ## _input_flush__nocheck(a, s); \
177 soft_t ## _input_flush__nocheck(b, s); \
178 soft_t ## _input_flush__nocheck(c, s); \
179 }
180
181 GEN_INPUT_FLUSH3(float32_input_flush3, float32)
182 GEN_INPUT_FLUSH3(float64_input_flush3, float64)
183 #undef GEN_INPUT_FLUSH3
184
185 /*
186 * Choose whether to use fpclassify or float32/64_* primitives in the generated
187 * hardfloat functions. Each combination of number of inputs and float size
188 * gets its own value.
189 */
190 #if defined(__x86_64__)
191 # define QEMU_HARDFLOAT_1F32_USE_FP 0
192 # define QEMU_HARDFLOAT_1F64_USE_FP 1
193 # define QEMU_HARDFLOAT_2F32_USE_FP 0
194 # define QEMU_HARDFLOAT_2F64_USE_FP 1
195 # define QEMU_HARDFLOAT_3F32_USE_FP 0
196 # define QEMU_HARDFLOAT_3F64_USE_FP 1
197 #else
198 # define QEMU_HARDFLOAT_1F32_USE_FP 0
199 # define QEMU_HARDFLOAT_1F64_USE_FP 0
200 # define QEMU_HARDFLOAT_2F32_USE_FP 0
201 # define QEMU_HARDFLOAT_2F64_USE_FP 0
202 # define QEMU_HARDFLOAT_3F32_USE_FP 0
203 # define QEMU_HARDFLOAT_3F64_USE_FP 0
204 #endif
205
206 /*
207 * QEMU_HARDFLOAT_USE_ISINF chooses whether to use isinf() over
208 * float{32,64}_is_infinity when !USE_FP.
209 * On x86_64/aarch64, using the former over the latter can yield a ~6% speedup.
210 * On power64 however, using isinf() reduces fp-bench performance by up to 50%.
211 */
212 #if defined(__x86_64__) || defined(__aarch64__)
213 # define QEMU_HARDFLOAT_USE_ISINF 1
214 #else
215 # define QEMU_HARDFLOAT_USE_ISINF 0
216 #endif
217
218 /*
219 * Some targets clear the FP flags before most FP operations. This prevents
220 * the use of hardfloat, since hardfloat relies on the inexact flag being
221 * already set.
222 */
223 #if defined(TARGET_PPC) || defined(__FAST_MATH__)
224 # if defined(__FAST_MATH__)
225 # warning disabling hardfloat due to -ffast-math: hardfloat requires an exact \
226 IEEE implementation
227 # endif
228 # define QEMU_NO_HARDFLOAT 1
229 # define QEMU_SOFTFLOAT_ATTR QEMU_FLATTEN
230 #else
231 # define QEMU_NO_HARDFLOAT 0
232 # define QEMU_SOFTFLOAT_ATTR QEMU_FLATTEN __attribute__((noinline))
233 #endif
234
235 static inline bool can_use_fpu(const float_status *s)
236 {
237 if (QEMU_NO_HARDFLOAT) {
238 return false;
239 }
240 return likely(s->float_exception_flags & float_flag_inexact &&
241 s->float_rounding_mode == float_round_nearest_even);
242 }
243
244 /*
245 * Hardfloat generation functions. Each operation can have two flavors:
246 * either using softfloat primitives (e.g. float32_is_zero_or_normal) for
247 * most condition checks, or native ones (e.g. fpclassify).
248 *
249 * The flavor is chosen by the callers. Instead of using macros, we rely on the
250 * compiler to propagate constants and inline everything into the callers.
251 *
252 * We only generate functions for operations with two inputs, since only
253 * these are common enough to justify consolidating them into common code.
254 */
255
256 typedef union {
257 float32 s;
258 float h;
259 } union_float32;
260
261 typedef union {
262 float64 s;
263 double h;
264 } union_float64;
265
266 typedef bool (*f32_check_fn)(union_float32 a, union_float32 b);
267 typedef bool (*f64_check_fn)(union_float64 a, union_float64 b);
268
269 typedef float32 (*soft_f32_op2_fn)(float32 a, float32 b, float_status *s);
270 typedef float64 (*soft_f64_op2_fn)(float64 a, float64 b, float_status *s);
271 typedef float (*hard_f32_op2_fn)(float a, float b);
272 typedef double (*hard_f64_op2_fn)(double a, double b);
273
274 /* 2-input is-zero-or-normal */
275 static inline bool f32_is_zon2(union_float32 a, union_float32 b)
276 {
277 if (QEMU_HARDFLOAT_2F32_USE_FP) {
278 /*
279 * Not using a temp variable for consecutive fpclassify calls ends up
280 * generating faster code.
281 */
282 return (fpclassify(a.h) == FP_NORMAL || fpclassify(a.h) == FP_ZERO) &&
283 (fpclassify(b.h) == FP_NORMAL || fpclassify(b.h) == FP_ZERO);
284 }
285 return float32_is_zero_or_normal(a.s) &&
286 float32_is_zero_or_normal(b.s);
287 }
288
289 static inline bool f64_is_zon2(union_float64 a, union_float64 b)
290 {
291 if (QEMU_HARDFLOAT_2F64_USE_FP) {
292 return (fpclassify(a.h) == FP_NORMAL || fpclassify(a.h) == FP_ZERO) &&
293 (fpclassify(b.h) == FP_NORMAL || fpclassify(b.h) == FP_ZERO);
294 }
295 return float64_is_zero_or_normal(a.s) &&
296 float64_is_zero_or_normal(b.s);
297 }
298
299 /* 3-input is-zero-or-normal */
300 static inline
301 bool f32_is_zon3(union_float32 a, union_float32 b, union_float32 c)
302 {
303 if (QEMU_HARDFLOAT_3F32_USE_FP) {
304 return (fpclassify(a.h) == FP_NORMAL || fpclassify(a.h) == FP_ZERO) &&
305 (fpclassify(b.h) == FP_NORMAL || fpclassify(b.h) == FP_ZERO) &&
306 (fpclassify(c.h) == FP_NORMAL || fpclassify(c.h) == FP_ZERO);
307 }
308 return float32_is_zero_or_normal(a.s) &&
309 float32_is_zero_or_normal(b.s) &&
310 float32_is_zero_or_normal(c.s);
311 }
312
313 static inline
314 bool f64_is_zon3(union_float64 a, union_float64 b, union_float64 c)
315 {
316 if (QEMU_HARDFLOAT_3F64_USE_FP) {
317 return (fpclassify(a.h) == FP_NORMAL || fpclassify(a.h) == FP_ZERO) &&
318 (fpclassify(b.h) == FP_NORMAL || fpclassify(b.h) == FP_ZERO) &&
319 (fpclassify(c.h) == FP_NORMAL || fpclassify(c.h) == FP_ZERO);
320 }
321 return float64_is_zero_or_normal(a.s) &&
322 float64_is_zero_or_normal(b.s) &&
323 float64_is_zero_or_normal(c.s);
324 }
325
326 static inline bool f32_is_inf(union_float32 a)
327 {
328 if (QEMU_HARDFLOAT_USE_ISINF) {
329 return isinf(a.h);
330 }
331 return float32_is_infinity(a.s);
332 }
333
334 static inline bool f64_is_inf(union_float64 a)
335 {
336 if (QEMU_HARDFLOAT_USE_ISINF) {
337 return isinf(a.h);
338 }
339 return float64_is_infinity(a.s);
340 }
341
342 static inline float32
343 float32_gen2(float32 xa, float32 xb, float_status *s,
344 hard_f32_op2_fn hard, soft_f32_op2_fn soft,
345 f32_check_fn pre, f32_check_fn post)
346 {
347 union_float32 ua, ub, ur;
348
349 ua.s = xa;
350 ub.s = xb;
351
352 if (unlikely(!can_use_fpu(s))) {
353 goto soft;
354 }
355
356 float32_input_flush2(&ua.s, &ub.s, s);
357 if (unlikely(!pre(ua, ub))) {
358 goto soft;
359 }
360
361 ur.h = hard(ua.h, ub.h);
362 if (unlikely(f32_is_inf(ur))) {
363 float_raise(float_flag_overflow, s);
364 } else if (unlikely(fabsf(ur.h) <= FLT_MIN) && post(ua, ub)) {
365 goto soft;
366 }
367 return ur.s;
368
369 soft:
370 return soft(ua.s, ub.s, s);
371 }
372
373 static inline float64
374 float64_gen2(float64 xa, float64 xb, float_status *s,
375 hard_f64_op2_fn hard, soft_f64_op2_fn soft,
376 f64_check_fn pre, f64_check_fn post)
377 {
378 union_float64 ua, ub, ur;
379
380 ua.s = xa;
381 ub.s = xb;
382
383 if (unlikely(!can_use_fpu(s))) {
384 goto soft;
385 }
386
387 float64_input_flush2(&ua.s, &ub.s, s);
388 if (unlikely(!pre(ua, ub))) {
389 goto soft;
390 }
391
392 ur.h = hard(ua.h, ub.h);
393 if (unlikely(f64_is_inf(ur))) {
394 float_raise(float_flag_overflow, s);
395 } else if (unlikely(fabs(ur.h) <= DBL_MIN) && post(ua, ub)) {
396 goto soft;
397 }
398 return ur.s;
399
400 soft:
401 return soft(ua.s, ub.s, s);
402 }
403
404 /*
405 * Classify a floating point number. Everything above float_class_qnan
406 * is a NaN so cls >= float_class_qnan is any NaN.
407 */
408
409 typedef enum __attribute__ ((__packed__)) {
410 float_class_unclassified,
411 float_class_zero,
412 float_class_normal,
413 float_class_inf,
414 float_class_qnan, /* all NaNs from here */
415 float_class_snan,
416 } FloatClass;
417
418 #define float_cmask(bit) (1u << (bit))
419
420 enum {
421 float_cmask_zero = float_cmask(float_class_zero),
422 float_cmask_normal = float_cmask(float_class_normal),
423 float_cmask_inf = float_cmask(float_class_inf),
424 float_cmask_qnan = float_cmask(float_class_qnan),
425 float_cmask_snan = float_cmask(float_class_snan),
426
427 float_cmask_infzero = float_cmask_zero | float_cmask_inf,
428 float_cmask_anynan = float_cmask_qnan | float_cmask_snan,
429 };
430
431 /* Flags for parts_minmax. */
432 enum {
433 /* Set for minimum; clear for maximum. */
434 minmax_ismin = 1,
435 /* Set for the IEEE 754-2008 minNum() and maxNum() operations. */
436 minmax_isnum = 2,
437 /* Set for the IEEE 754-2008 minNumMag() and minNumMag() operations. */
438 minmax_ismag = 4,
439 };
440
441 /* Simple helpers for checking if, or what kind of, NaN we have */
442 static inline __attribute__((unused)) bool is_nan(FloatClass c)
443 {
444 return unlikely(c >= float_class_qnan);
445 }
446
447 static inline __attribute__((unused)) bool is_snan(FloatClass c)
448 {
449 return c == float_class_snan;
450 }
451
452 static inline __attribute__((unused)) bool is_qnan(FloatClass c)
453 {
454 return c == float_class_qnan;
455 }
456
457 /*
458 * Structure holding all of the decomposed parts of a float.
459 * The exponent is unbiased and the fraction is normalized.
460 *
461 * The fraction words are stored in big-endian word ordering,
462 * so that truncation from a larger format to a smaller format
463 * can be done simply by ignoring subsequent elements.
464 */
465
466 typedef struct {
467 FloatClass cls;
468 bool sign;
469 int32_t exp;
470 union {
471 /* Routines that know the structure may reference the singular name. */
472 uint64_t frac;
473 /*
474 * Routines expanded with multiple structures reference "hi" and "lo"
475 * depending on the operation. In FloatParts64, "hi" and "lo" are
476 * both the same word and aliased here.
477 */
478 uint64_t frac_hi;
479 uint64_t frac_lo;
480 };
481 } FloatParts64;
482
483 typedef struct {
484 FloatClass cls;
485 bool sign;
486 int32_t exp;
487 uint64_t frac_hi;
488 uint64_t frac_lo;
489 } FloatParts128;
490
491 typedef struct {
492 FloatClass cls;
493 bool sign;
494 int32_t exp;
495 uint64_t frac_hi;
496 uint64_t frac_hm; /* high-middle */
497 uint64_t frac_lm; /* low-middle */
498 uint64_t frac_lo;
499 } FloatParts256;
500
501 /* These apply to the most significant word of each FloatPartsN. */
502 #define DECOMPOSED_BINARY_POINT 63
503 #define DECOMPOSED_IMPLICIT_BIT (1ull << DECOMPOSED_BINARY_POINT)
504
505 /* Structure holding all of the relevant parameters for a format.
506 * exp_size: the size of the exponent field
507 * exp_bias: the offset applied to the exponent field
508 * exp_max: the maximum normalised exponent
509 * frac_size: the size of the fraction field
510 * frac_shift: shift to normalise the fraction with DECOMPOSED_BINARY_POINT
511 * The following are computed based the size of fraction
512 * round_mask: bits below lsb which must be rounded
513 * The following optional modifiers are available:
514 * arm_althp: handle ARM Alternative Half Precision
515 */
516 typedef struct {
517 int exp_size;
518 int exp_bias;
519 int exp_max;
520 int frac_size;
521 int frac_shift;
522 bool arm_althp;
523 uint64_t round_mask;
524 } FloatFmt;
525
526 /* Expand fields based on the size of exponent and fraction */
527 #define FLOAT_PARAMS_(E) \
528 .exp_size = E, \
529 .exp_bias = ((1 << E) - 1) >> 1, \
530 .exp_max = (1 << E) - 1
531
532 #define FLOAT_PARAMS(E, F) \
533 FLOAT_PARAMS_(E), \
534 .frac_size = F, \
535 .frac_shift = (-F - 1) & 63, \
536 .round_mask = (1ull << ((-F - 1) & 63)) - 1
537
538 static const FloatFmt float16_params = {
539 FLOAT_PARAMS(5, 10)
540 };
541
542 static const FloatFmt float16_params_ahp = {
543 FLOAT_PARAMS(5, 10),
544 .arm_althp = true
545 };
546
547 static const FloatFmt bfloat16_params = {
548 FLOAT_PARAMS(8, 7)
549 };
550
551 static const FloatFmt float32_params = {
552 FLOAT_PARAMS(8, 23)
553 };
554
555 static const FloatFmt float64_params = {
556 FLOAT_PARAMS(11, 52)
557 };
558
559 static const FloatFmt float128_params = {
560 FLOAT_PARAMS(15, 112)
561 };
562
563 #define FLOATX80_PARAMS(R) \
564 FLOAT_PARAMS_(15), \
565 .frac_size = R == 64 ? 63 : R, \
566 .frac_shift = 0, \
567 .round_mask = R == 64 ? -1 : (1ull << ((-R - 1) & 63)) - 1
568
569 static const FloatFmt floatx80_params[3] = {
570 [floatx80_precision_s] = { FLOATX80_PARAMS(23) },
571 [floatx80_precision_d] = { FLOATX80_PARAMS(52) },
572 [floatx80_precision_x] = { FLOATX80_PARAMS(64) },
573 };
574
575 /* Unpack a float to parts, but do not canonicalize. */
576 static void unpack_raw64(FloatParts64 *r, const FloatFmt *fmt, uint64_t raw)
577 {
578 const int f_size = fmt->frac_size;
579 const int e_size = fmt->exp_size;
580
581 *r = (FloatParts64) {
582 .cls = float_class_unclassified,
583 .sign = extract64(raw, f_size + e_size, 1),
584 .exp = extract64(raw, f_size, e_size),
585 .frac = extract64(raw, 0, f_size)
586 };
587 }
588
589 static inline void float16_unpack_raw(FloatParts64 *p, float16 f)
590 {
591 unpack_raw64(p, &float16_params, f);
592 }
593
594 static inline void bfloat16_unpack_raw(FloatParts64 *p, bfloat16 f)
595 {
596 unpack_raw64(p, &bfloat16_params, f);
597 }
598
599 static inline void float32_unpack_raw(FloatParts64 *p, float32 f)
600 {
601 unpack_raw64(p, &float32_params, f);
602 }
603
604 static inline void float64_unpack_raw(FloatParts64 *p, float64 f)
605 {
606 unpack_raw64(p, &float64_params, f);
607 }
608
609 static void floatx80_unpack_raw(FloatParts128 *p, floatx80 f)
610 {
611 *p = (FloatParts128) {
612 .cls = float_class_unclassified,
613 .sign = extract32(f.high, 15, 1),
614 .exp = extract32(f.high, 0, 15),
615 .frac_hi = f.low
616 };
617 }
618
619 static void float128_unpack_raw(FloatParts128 *p, float128 f)
620 {
621 const int f_size = float128_params.frac_size - 64;
622 const int e_size = float128_params.exp_size;
623
624 *p = (FloatParts128) {
625 .cls = float_class_unclassified,
626 .sign = extract64(f.high, f_size + e_size, 1),
627 .exp = extract64(f.high, f_size, e_size),
628 .frac_hi = extract64(f.high, 0, f_size),
629 .frac_lo = f.low,
630 };
631 }
632
633 /* Pack a float from parts, but do not canonicalize. */
634 static uint64_t pack_raw64(const FloatParts64 *p, const FloatFmt *fmt)
635 {
636 const int f_size = fmt->frac_size;
637 const int e_size = fmt->exp_size;
638 uint64_t ret;
639
640 ret = (uint64_t)p->sign << (f_size + e_size);
641 ret = deposit64(ret, f_size, e_size, p->exp);
642 ret = deposit64(ret, 0, f_size, p->frac);
643 return ret;
644 }
645
646 static inline float16 float16_pack_raw(const FloatParts64 *p)
647 {
648 return make_float16(pack_raw64(p, &float16_params));
649 }
650
651 static inline bfloat16 bfloat16_pack_raw(const FloatParts64 *p)
652 {
653 return pack_raw64(p, &bfloat16_params);
654 }
655
656 static inline float32 float32_pack_raw(const FloatParts64 *p)
657 {
658 return make_float32(pack_raw64(p, &float32_params));
659 }
660
661 static inline float64 float64_pack_raw(const FloatParts64 *p)
662 {
663 return make_float64(pack_raw64(p, &float64_params));
664 }
665
666 static float128 float128_pack_raw(const FloatParts128 *p)
667 {
668 const int f_size = float128_params.frac_size - 64;
669 const int e_size = float128_params.exp_size;
670 uint64_t hi;
671
672 hi = (uint64_t)p->sign << (f_size + e_size);
673 hi = deposit64(hi, f_size, e_size, p->exp);
674 hi = deposit64(hi, 0, f_size, p->frac_hi);
675 return make_float128(hi, p->frac_lo);
676 }
677
678 /*----------------------------------------------------------------------------
679 | Functions and definitions to determine: (1) whether tininess for underflow
680 | is detected before or after rounding by default, (2) what (if anything)
681 | happens when exceptions are raised, (3) how signaling NaNs are distinguished
682 | from quiet NaNs, (4) the default generated quiet NaNs, and (5) how NaNs
683 | are propagated from function inputs to output. These details are target-
684 | specific.
685 *----------------------------------------------------------------------------*/
686 #include "softfloat-specialize.c.inc"
687
688 #define PARTS_GENERIC_64_128(NAME, P) \
689 _Generic((P), FloatParts64 *: parts64_##NAME, \
690 FloatParts128 *: parts128_##NAME)
691
692 #define PARTS_GENERIC_64_128_256(NAME, P) \
693 _Generic((P), FloatParts64 *: parts64_##NAME, \
694 FloatParts128 *: parts128_##NAME, \
695 FloatParts256 *: parts256_##NAME)
696
697 #define parts_default_nan(P, S) PARTS_GENERIC_64_128(default_nan, P)(P, S)
698 #define parts_silence_nan(P, S) PARTS_GENERIC_64_128(silence_nan, P)(P, S)
699
700 static void parts64_return_nan(FloatParts64 *a, float_status *s);
701 static void parts128_return_nan(FloatParts128 *a, float_status *s);
702
703 #define parts_return_nan(P, S) PARTS_GENERIC_64_128(return_nan, P)(P, S)
704
705 static FloatParts64 *parts64_pick_nan(FloatParts64 *a, FloatParts64 *b,
706 float_status *s);
707 static FloatParts128 *parts128_pick_nan(FloatParts128 *a, FloatParts128 *b,
708 float_status *s);
709
710 #define parts_pick_nan(A, B, S) PARTS_GENERIC_64_128(pick_nan, A)(A, B, S)
711
712 static FloatParts64 *parts64_pick_nan_muladd(FloatParts64 *a, FloatParts64 *b,
713 FloatParts64 *c, float_status *s,
714 int ab_mask, int abc_mask);
715 static FloatParts128 *parts128_pick_nan_muladd(FloatParts128 *a,
716 FloatParts128 *b,
717 FloatParts128 *c,
718 float_status *s,
719 int ab_mask, int abc_mask);
720
721 #define parts_pick_nan_muladd(A, B, C, S, ABM, ABCM) \
722 PARTS_GENERIC_64_128(pick_nan_muladd, A)(A, B, C, S, ABM, ABCM)
723
724 static void parts64_canonicalize(FloatParts64 *p, float_status *status,
725 const FloatFmt *fmt);
726 static void parts128_canonicalize(FloatParts128 *p, float_status *status,
727 const FloatFmt *fmt);
728
729 #define parts_canonicalize(A, S, F) \
730 PARTS_GENERIC_64_128(canonicalize, A)(A, S, F)
731
732 static void parts64_uncanon_normal(FloatParts64 *p, float_status *status,
733 const FloatFmt *fmt);
734 static void parts128_uncanon_normal(FloatParts128 *p, float_status *status,
735 const FloatFmt *fmt);
736
737 #define parts_uncanon_normal(A, S, F) \
738 PARTS_GENERIC_64_128(uncanon_normal, A)(A, S, F)
739
740 static void parts64_uncanon(FloatParts64 *p, float_status *status,
741 const FloatFmt *fmt);
742 static void parts128_uncanon(FloatParts128 *p, float_status *status,
743 const FloatFmt *fmt);
744
745 #define parts_uncanon(A, S, F) \
746 PARTS_GENERIC_64_128(uncanon, A)(A, S, F)
747
748 static void parts64_add_normal(FloatParts64 *a, FloatParts64 *b);
749 static void parts128_add_normal(FloatParts128 *a, FloatParts128 *b);
750 static void parts256_add_normal(FloatParts256 *a, FloatParts256 *b);
751
752 #define parts_add_normal(A, B) \
753 PARTS_GENERIC_64_128_256(add_normal, A)(A, B)
754
755 static bool parts64_sub_normal(FloatParts64 *a, FloatParts64 *b);
756 static bool parts128_sub_normal(FloatParts128 *a, FloatParts128 *b);
757 static bool parts256_sub_normal(FloatParts256 *a, FloatParts256 *b);
758
759 #define parts_sub_normal(A, B) \
760 PARTS_GENERIC_64_128_256(sub_normal, A)(A, B)
761
762 static FloatParts64 *parts64_addsub(FloatParts64 *a, FloatParts64 *b,
763 float_status *s, bool subtract);
764 static FloatParts128 *parts128_addsub(FloatParts128 *a, FloatParts128 *b,
765 float_status *s, bool subtract);
766
767 #define parts_addsub(A, B, S, Z) \
768 PARTS_GENERIC_64_128(addsub, A)(A, B, S, Z)
769
770 static FloatParts64 *parts64_mul(FloatParts64 *a, FloatParts64 *b,
771 float_status *s);
772 static FloatParts128 *parts128_mul(FloatParts128 *a, FloatParts128 *b,
773 float_status *s);
774
775 #define parts_mul(A, B, S) \
776 PARTS_GENERIC_64_128(mul, A)(A, B, S)
777
778 static FloatParts64 *parts64_muladd(FloatParts64 *a, FloatParts64 *b,
779 FloatParts64 *c, int flags,
780 float_status *s);
781 static FloatParts128 *parts128_muladd(FloatParts128 *a, FloatParts128 *b,
782 FloatParts128 *c, int flags,
783 float_status *s);
784
785 #define parts_muladd(A, B, C, Z, S) \
786 PARTS_GENERIC_64_128(muladd, A)(A, B, C, Z, S)
787
788 static FloatParts64 *parts64_div(FloatParts64 *a, FloatParts64 *b,
789 float_status *s);
790 static FloatParts128 *parts128_div(FloatParts128 *a, FloatParts128 *b,
791 float_status *s);
792
793 #define parts_div(A, B, S) \
794 PARTS_GENERIC_64_128(div, A)(A, B, S)
795
796 static FloatParts64 *parts64_modrem(FloatParts64 *a, FloatParts64 *b,
797 uint64_t *mod_quot, float_status *s);
798 static FloatParts128 *parts128_modrem(FloatParts128 *a, FloatParts128 *b,
799 uint64_t *mod_quot, float_status *s);
800
801 #define parts_modrem(A, B, Q, S) \
802 PARTS_GENERIC_64_128(modrem, A)(A, B, Q, S)
803
804 static void parts64_sqrt(FloatParts64 *a, float_status *s, const FloatFmt *f);
805 static void parts128_sqrt(FloatParts128 *a, float_status *s, const FloatFmt *f);
806
807 #define parts_sqrt(A, S, F) \
808 PARTS_GENERIC_64_128(sqrt, A)(A, S, F)
809
810 static bool parts64_round_to_int_normal(FloatParts64 *a, FloatRoundMode rm,
811 int scale, int frac_size);
812 static bool parts128_round_to_int_normal(FloatParts128 *a, FloatRoundMode r,
813 int scale, int frac_size);
814
815 #define parts_round_to_int_normal(A, R, C, F) \
816 PARTS_GENERIC_64_128(round_to_int_normal, A)(A, R, C, F)
817
818 static void parts64_round_to_int(FloatParts64 *a, FloatRoundMode rm,
819 int scale, float_status *s,
820 const FloatFmt *fmt);
821 static void parts128_round_to_int(FloatParts128 *a, FloatRoundMode r,
822 int scale, float_status *s,
823 const FloatFmt *fmt);
824
825 #define parts_round_to_int(A, R, C, S, F) \
826 PARTS_GENERIC_64_128(round_to_int, A)(A, R, C, S, F)
827
828 static int64_t parts64_float_to_sint(FloatParts64 *p, FloatRoundMode rmode,
829 int scale, int64_t min, int64_t max,
830 float_status *s);
831 static int64_t parts128_float_to_sint(FloatParts128 *p, FloatRoundMode rmode,
832 int scale, int64_t min, int64_t max,
833 float_status *s);
834
835 #define parts_float_to_sint(P, R, Z, MN, MX, S) \
836 PARTS_GENERIC_64_128(float_to_sint, P)(P, R, Z, MN, MX, S)
837
838 static uint64_t parts64_float_to_uint(FloatParts64 *p, FloatRoundMode rmode,
839 int scale, uint64_t max,
840 float_status *s);
841 static uint64_t parts128_float_to_uint(FloatParts128 *p, FloatRoundMode rmode,
842 int scale, uint64_t max,
843 float_status *s);
844
845 #define parts_float_to_uint(P, R, Z, M, S) \
846 PARTS_GENERIC_64_128(float_to_uint, P)(P, R, Z, M, S)
847
848 static void parts64_sint_to_float(FloatParts64 *p, int64_t a,
849 int scale, float_status *s);
850 static void parts128_sint_to_float(FloatParts128 *p, int64_t a,
851 int scale, float_status *s);
852
853 #define parts_sint_to_float(P, I, Z, S) \
854 PARTS_GENERIC_64_128(sint_to_float, P)(P, I, Z, S)
855
856 static void parts64_uint_to_float(FloatParts64 *p, uint64_t a,
857 int scale, float_status *s);
858 static void parts128_uint_to_float(FloatParts128 *p, uint64_t a,
859 int scale, float_status *s);
860
861 #define parts_uint_to_float(P, I, Z, S) \
862 PARTS_GENERIC_64_128(uint_to_float, P)(P, I, Z, S)
863
864 static FloatParts64 *parts64_minmax(FloatParts64 *a, FloatParts64 *b,
865 float_status *s, int flags);
866 static FloatParts128 *parts128_minmax(FloatParts128 *a, FloatParts128 *b,
867 float_status *s, int flags);
868
869 #define parts_minmax(A, B, S, F) \
870 PARTS_GENERIC_64_128(minmax, A)(A, B, S, F)
871
872 static int parts64_compare(FloatParts64 *a, FloatParts64 *b,
873 float_status *s, bool q);
874 static int parts128_compare(FloatParts128 *a, FloatParts128 *b,
875 float_status *s, bool q);
876
877 #define parts_compare(A, B, S, Q) \
878 PARTS_GENERIC_64_128(compare, A)(A, B, S, Q)
879
880 static void parts64_scalbn(FloatParts64 *a, int n, float_status *s);
881 static void parts128_scalbn(FloatParts128 *a, int n, float_status *s);
882
883 #define parts_scalbn(A, N, S) \
884 PARTS_GENERIC_64_128(scalbn, A)(A, N, S)
885
886 static void parts64_log2(FloatParts64 *a, float_status *s, const FloatFmt *f);
887 static void parts128_log2(FloatParts128 *a, float_status *s, const FloatFmt *f);
888
889 #define parts_log2(A, S, F) \
890 PARTS_GENERIC_64_128(log2, A)(A, S, F)
891
892 /*
893 * Helper functions for softfloat-parts.c.inc, per-size operations.
894 */
895
896 #define FRAC_GENERIC_64_128(NAME, P) \
897 _Generic((P), FloatParts64 *: frac64_##NAME, \
898 FloatParts128 *: frac128_##NAME)
899
900 #define FRAC_GENERIC_64_128_256(NAME, P) \
901 _Generic((P), FloatParts64 *: frac64_##NAME, \
902 FloatParts128 *: frac128_##NAME, \
903 FloatParts256 *: frac256_##NAME)
904
905 static bool frac64_add(FloatParts64 *r, FloatParts64 *a, FloatParts64 *b)
906 {
907 return uadd64_overflow(a->frac, b->frac, &r->frac);
908 }
909
910 static bool frac128_add(FloatParts128 *r, FloatParts128 *a, FloatParts128 *b)
911 {
912 bool c = 0;
913 r->frac_lo = uadd64_carry(a->frac_lo, b->frac_lo, &c);
914 r->frac_hi = uadd64_carry(a->frac_hi, b->frac_hi, &c);
915 return c;
916 }
917
918 static bool frac256_add(FloatParts256 *r, FloatParts256 *a, FloatParts256 *b)
919 {
920 bool c = 0;
921 r->frac_lo = uadd64_carry(a->frac_lo, b->frac_lo, &c);
922 r->frac_lm = uadd64_carry(a->frac_lm, b->frac_lm, &c);
923 r->frac_hm = uadd64_carry(a->frac_hm, b->frac_hm, &c);
924 r->frac_hi = uadd64_carry(a->frac_hi, b->frac_hi, &c);
925 return c;
926 }
927
928 #define frac_add(R, A, B) FRAC_GENERIC_64_128_256(add, R)(R, A, B)
929
930 static bool frac64_addi(FloatParts64 *r, FloatParts64 *a, uint64_t c)
931 {
932 return uadd64_overflow(a->frac, c, &r->frac);
933 }
934
935 static bool frac128_addi(FloatParts128 *r, FloatParts128 *a, uint64_t c)
936 {
937 c = uadd64_overflow(a->frac_lo, c, &r->frac_lo);
938 return uadd64_overflow(a->frac_hi, c, &r->frac_hi);
939 }
940
941 #define frac_addi(R, A, C) FRAC_GENERIC_64_128(addi, R)(R, A, C)
942
943 static void frac64_allones(FloatParts64 *a)
944 {
945 a->frac = -1;
946 }
947
948 static void frac128_allones(FloatParts128 *a)
949 {
950 a->frac_hi = a->frac_lo = -1;
951 }
952
953 #define frac_allones(A) FRAC_GENERIC_64_128(allones, A)(A)
954
955 static int frac64_cmp(FloatParts64 *a, FloatParts64 *b)
956 {
957 return a->frac == b->frac ? 0 : a->frac < b->frac ? -1 : 1;
958 }
959
960 static int frac128_cmp(FloatParts128 *a, FloatParts128 *b)
961 {
962 uint64_t ta = a->frac_hi, tb = b->frac_hi;
963 if (ta == tb) {
964 ta = a->frac_lo, tb = b->frac_lo;
965 if (ta == tb) {
966 return 0;
967 }
968 }
969 return ta < tb ? -1 : 1;
970 }
971
972 #define frac_cmp(A, B) FRAC_GENERIC_64_128(cmp, A)(A, B)
973
974 static void frac64_clear(FloatParts64 *a)
975 {
976 a->frac = 0;
977 }
978
979 static void frac128_clear(FloatParts128 *a)
980 {
981 a->frac_hi = a->frac_lo = 0;
982 }
983
984 #define frac_clear(A) FRAC_GENERIC_64_128(clear, A)(A)
985
986 static bool frac64_div(FloatParts64 *a, FloatParts64 *b)
987 {
988 uint64_t n1, n0, r, q;
989 bool ret;
990
991 /*
992 * We want a 2*N / N-bit division to produce exactly an N-bit
993 * result, so that we do not lose any precision and so that we
994 * do not have to renormalize afterward. If A.frac < B.frac,
995 * then division would produce an (N-1)-bit result; shift A left
996 * by one to produce the an N-bit result, and return true to
997 * decrement the exponent to match.
998 *
999 * The udiv_qrnnd algorithm that we're using requires normalization,
1000 * i.e. the msb of the denominator must be set, which is already true.
1001 */
1002 ret = a->frac < b->frac;
1003 if (ret) {
1004 n0 = a->frac;
1005 n1 = 0;
1006 } else {
1007 n0 = a->frac >> 1;
1008 n1 = a->frac << 63;
1009 }
1010 q = udiv_qrnnd(&r, n0, n1, b->frac);
1011
1012 /* Set lsb if there is a remainder, to set inexact. */
1013 a->frac = q | (r != 0);
1014
1015 return ret;
1016 }
1017
1018 static bool frac128_div(FloatParts128 *a, FloatParts128 *b)
1019 {
1020 uint64_t q0, q1, a0, a1, b0, b1;
1021 uint64_t r0, r1, r2, r3, t0, t1, t2, t3;
1022 bool ret = false;
1023
1024 a0 = a->frac_hi, a1 = a->frac_lo;
1025 b0 = b->frac_hi, b1 = b->frac_lo;
1026
1027 ret = lt128(a0, a1, b0, b1);
1028 if (!ret) {
1029 a1 = shr_double(a0, a1, 1);
1030 a0 = a0 >> 1;
1031 }
1032
1033 /* Use 128/64 -> 64 division as estimate for 192/128 -> 128 division. */
1034 q0 = estimateDiv128To64(a0, a1, b0);
1035
1036 /*
1037 * Estimate is high because B1 was not included (unless B1 == 0).
1038 * Reduce quotient and increase remainder until remainder is non-negative.
1039 * This loop will execute 0 to 2 times.
1040 */
1041 mul128By64To192(b0, b1, q0, &t0, &t1, &t2);
1042 sub192(a0, a1, 0, t0, t1, t2, &r0, &r1, &r2);
1043 while (r0 != 0) {
1044 q0--;
1045 add192(r0, r1, r2, 0, b0, b1, &r0, &r1, &r2);
1046 }
1047
1048 /* Repeat using the remainder, producing a second word of quotient. */
1049 q1 = estimateDiv128To64(r1, r2, b0);
1050 mul128By64To192(b0, b1, q1, &t1, &t2, &t3);
1051 sub192(r1, r2, 0, t1, t2, t3, &r1, &r2, &r3);
1052 while (r1 != 0) {
1053 q1--;
1054 add192(r1, r2, r3, 0, b0, b1, &r1, &r2, &r3);
1055 }
1056
1057 /* Any remainder indicates inexact; set sticky bit. */
1058 q1 |= (r2 | r3) != 0;
1059
1060 a->frac_hi = q0;
1061 a->frac_lo = q1;
1062 return ret;
1063 }
1064
1065 #define frac_div(A, B) FRAC_GENERIC_64_128(div, A)(A, B)
1066
1067 static bool frac64_eqz(FloatParts64 *a)
1068 {
1069 return a->frac == 0;
1070 }
1071
1072 static bool frac128_eqz(FloatParts128 *a)
1073 {
1074 return (a->frac_hi | a->frac_lo) == 0;
1075 }
1076
1077 #define frac_eqz(A) FRAC_GENERIC_64_128(eqz, A)(A)
1078
1079 static void frac64_mulw(FloatParts128 *r, FloatParts64 *a, FloatParts64 *b)
1080 {
1081 mulu64(&r->frac_lo, &r->frac_hi, a->frac, b->frac);
1082 }
1083
1084 static void frac128_mulw(FloatParts256 *r, FloatParts128 *a, FloatParts128 *b)
1085 {
1086 mul128To256(a->frac_hi, a->frac_lo, b->frac_hi, b->frac_lo,
1087 &r->frac_hi, &r->frac_hm, &r->frac_lm, &r->frac_lo);
1088 }
1089
1090 #define frac_mulw(R, A, B) FRAC_GENERIC_64_128(mulw, A)(R, A, B)
1091
1092 static void frac64_neg(FloatParts64 *a)
1093 {
1094 a->frac = -a->frac;
1095 }
1096
1097 static void frac128_neg(FloatParts128 *a)
1098 {
1099 bool c = 0;
1100 a->frac_lo = usub64_borrow(0, a->frac_lo, &c);
1101 a->frac_hi = usub64_borrow(0, a->frac_hi, &c);
1102 }
1103
1104 static void frac256_neg(FloatParts256 *a)
1105 {
1106 bool c = 0;
1107 a->frac_lo = usub64_borrow(0, a->frac_lo, &c);
1108 a->frac_lm = usub64_borrow(0, a->frac_lm, &c);
1109 a->frac_hm = usub64_borrow(0, a->frac_hm, &c);
1110 a->frac_hi = usub64_borrow(0, a->frac_hi, &c);
1111 }
1112
1113 #define frac_neg(A) FRAC_GENERIC_64_128_256(neg, A)(A)
1114
1115 static int frac64_normalize(FloatParts64 *a)
1116 {
1117 if (a->frac) {
1118 int shift = clz64(a->frac);
1119 a->frac <<= shift;
1120 return shift;
1121 }
1122 return 64;
1123 }
1124
1125 static int frac128_normalize(FloatParts128 *a)
1126 {
1127 if (a->frac_hi) {
1128 int shl = clz64(a->frac_hi);
1129 a->frac_hi = shl_double(a->frac_hi, a->frac_lo, shl);
1130 a->frac_lo <<= shl;
1131 return shl;
1132 } else if (a->frac_lo) {
1133 int shl = clz64(a->frac_lo);
1134 a->frac_hi = a->frac_lo << shl;
1135 a->frac_lo = 0;
1136 return shl + 64;
1137 }
1138 return 128;
1139 }
1140
1141 static int frac256_normalize(FloatParts256 *a)
1142 {
1143 uint64_t a0 = a->frac_hi, a1 = a->frac_hm;
1144 uint64_t a2 = a->frac_lm, a3 = a->frac_lo;
1145 int ret, shl;
1146
1147 if (likely(a0)) {
1148 shl = clz64(a0);
1149 if (shl == 0) {
1150 return 0;
1151 }
1152 ret = shl;
1153 } else {
1154 if (a1) {
1155 ret = 64;
1156 a0 = a1, a1 = a2, a2 = a3, a3 = 0;
1157 } else if (a2) {
1158 ret = 128;
1159 a0 = a2, a1 = a3, a2 = 0, a3 = 0;
1160 } else if (a3) {
1161 ret = 192;
1162 a0 = a3, a1 = 0, a2 = 0, a3 = 0;
1163 } else {
1164 ret = 256;
1165 a0 = 0, a1 = 0, a2 = 0, a3 = 0;
1166 goto done;
1167 }
1168 shl = clz64(a0);
1169 if (shl == 0) {
1170 goto done;
1171 }
1172 ret += shl;
1173 }
1174
1175 a0 = shl_double(a0, a1, shl);
1176 a1 = shl_double(a1, a2, shl);
1177 a2 = shl_double(a2, a3, shl);
1178 a3 <<= shl;
1179
1180 done:
1181 a->frac_hi = a0;
1182 a->frac_hm = a1;
1183 a->frac_lm = a2;
1184 a->frac_lo = a3;
1185 return ret;
1186 }
1187
1188 #define frac_normalize(A) FRAC_GENERIC_64_128_256(normalize, A)(A)
1189
1190 static void frac64_modrem(FloatParts64 *a, FloatParts64 *b, uint64_t *mod_quot)
1191 {
1192 uint64_t a0, a1, b0, t0, t1, q, quot;
1193 int exp_diff = a->exp - b->exp;
1194 int shift;
1195
1196 a0 = a->frac;
1197 a1 = 0;
1198
1199 if (exp_diff < -1) {
1200 if (mod_quot) {
1201 *mod_quot = 0;
1202 }
1203 return;
1204 }
1205 if (exp_diff == -1) {
1206 a0 >>= 1;
1207 exp_diff = 0;
1208 }
1209
1210 b0 = b->frac;
1211 quot = q = b0 <= a0;
1212 if (q) {
1213 a0 -= b0;
1214 }
1215
1216 exp_diff -= 64;
1217 while (exp_diff > 0) {
1218 q = estimateDiv128To64(a0, a1, b0);
1219 q = q > 2 ? q - 2 : 0;
1220 mul64To128(b0, q, &t0, &t1);
1221 sub128(a0, a1, t0, t1, &a0, &a1);
1222 shortShift128Left(a0, a1, 62, &a0, &a1);
1223 exp_diff -= 62;
1224 quot = (quot << 62) + q;
1225 }
1226
1227 exp_diff += 64;
1228 if (exp_diff > 0) {
1229 q = estimateDiv128To64(a0, a1, b0);
1230 q = q > 2 ? (q - 2) >> (64 - exp_diff) : 0;
1231 mul64To128(b0, q << (64 - exp_diff), &t0, &t1);
1232 sub128(a0, a1, t0, t1, &a0, &a1);
1233 shortShift128Left(0, b0, 64 - exp_diff, &t0, &t1);
1234 while (le128(t0, t1, a0, a1)) {
1235 ++q;
1236 sub128(a0, a1, t0, t1, &a0, &a1);
1237 }
1238 quot = (exp_diff < 64 ? quot << exp_diff : 0) + q;
1239 } else {
1240 t0 = b0;
1241 t1 = 0;
1242 }
1243
1244 if (mod_quot) {
1245 *mod_quot = quot;
1246 } else {
1247 sub128(t0, t1, a0, a1, &t0, &t1);
1248 if (lt128(t0, t1, a0, a1) ||
1249 (eq128(t0, t1, a0, a1) && (q & 1))) {
1250 a0 = t0;
1251 a1 = t1;
1252 a->sign = !a->sign;
1253 }
1254 }
1255
1256 if (likely(a0)) {
1257 shift = clz64(a0);
1258 shortShift128Left(a0, a1, shift, &a0, &a1);
1259 } else if (likely(a1)) {
1260 shift = clz64(a1);
1261 a0 = a1 << shift;
1262 a1 = 0;
1263 shift += 64;
1264 } else {
1265 a->cls = float_class_zero;
1266 return;
1267 }
1268
1269 a->exp = b->exp + exp_diff - shift;
1270 a->frac = a0 | (a1 != 0);
1271 }
1272
1273 static void frac128_modrem(FloatParts128 *a, FloatParts128 *b,
1274 uint64_t *mod_quot)
1275 {
1276 uint64_t a0, a1, a2, b0, b1, t0, t1, t2, q, quot;
1277 int exp_diff = a->exp - b->exp;
1278 int shift;
1279
1280 a0 = a->frac_hi;
1281 a1 = a->frac_lo;
1282 a2 = 0;
1283
1284 if (exp_diff < -1) {
1285 if (mod_quot) {
1286 *mod_quot = 0;
1287 }
1288 return;
1289 }
1290 if (exp_diff == -1) {
1291 shift128Right(a0, a1, 1, &a0, &a1);
1292 exp_diff = 0;
1293 }
1294
1295 b0 = b->frac_hi;
1296 b1 = b->frac_lo;
1297
1298 quot = q = le128(b0, b1, a0, a1);
1299 if (q) {
1300 sub128(a0, a1, b0, b1, &a0, &a1);
1301 }
1302
1303 exp_diff -= 64;
1304 while (exp_diff > 0) {
1305 q = estimateDiv128To64(a0, a1, b0);
1306 q = q > 4 ? q - 4 : 0;
1307 mul128By64To192(b0, b1, q, &t0, &t1, &t2);
1308 sub192(a0, a1, a2, t0, t1, t2, &a0, &a1, &a2);
1309 shortShift192Left(a0, a1, a2, 61, &a0, &a1, &a2);
1310 exp_diff -= 61;
1311 quot = (quot << 61) + q;
1312 }
1313
1314 exp_diff += 64;
1315 if (exp_diff > 0) {
1316 q = estimateDiv128To64(a0, a1, b0);
1317 q = q > 4 ? (q - 4) >> (64 - exp_diff) : 0;
1318 mul128By64To192(b0, b1, q << (64 - exp_diff), &t0, &t1, &t2);
1319 sub192(a0, a1, a2, t0, t1, t2, &a0, &a1, &a2);
1320 shortShift192Left(0, b0, b1, 64 - exp_diff, &t0, &t1, &t2);
1321 while (le192(t0, t1, t2, a0, a1, a2)) {
1322 ++q;
1323 sub192(a0, a1, a2, t0, t1, t2, &a0, &a1, &a2);
1324 }
1325 quot = (exp_diff < 64 ? quot << exp_diff : 0) + q;
1326 } else {
1327 t0 = b0;
1328 t1 = b1;
1329 t2 = 0;
1330 }
1331
1332 if (mod_quot) {
1333 *mod_quot = quot;
1334 } else {
1335 sub192(t0, t1, t2, a0, a1, a2, &t0, &t1, &t2);
1336 if (lt192(t0, t1, t2, a0, a1, a2) ||
1337 (eq192(t0, t1, t2, a0, a1, a2) && (q & 1))) {
1338 a0 = t0;
1339 a1 = t1;
1340 a2 = t2;
1341 a->sign = !a->sign;
1342 }
1343 }
1344
1345 if (likely(a0)) {
1346 shift = clz64(a0);
1347 shortShift192Left(a0, a1, a2, shift, &a0, &a1, &a2);
1348 } else if (likely(a1)) {
1349 shift = clz64(a1);
1350 shortShift128Left(a1, a2, shift, &a0, &a1);
1351 a2 = 0;
1352 shift += 64;
1353 } else if (likely(a2)) {
1354 shift = clz64(a2);
1355 a0 = a2 << shift;
1356 a1 = a2 = 0;
1357 shift += 128;
1358 } else {
1359 a->cls = float_class_zero;
1360 return;
1361 }
1362
1363 a->exp = b->exp + exp_diff - shift;
1364 a->frac_hi = a0;
1365 a->frac_lo = a1 | (a2 != 0);
1366 }
1367
1368 #define frac_modrem(A, B, Q) FRAC_GENERIC_64_128(modrem, A)(A, B, Q)
1369
1370 static void frac64_shl(FloatParts64 *a, int c)
1371 {
1372 a->frac <<= c;
1373 }
1374
1375 static void frac128_shl(FloatParts128 *a, int c)
1376 {
1377 uint64_t a0 = a->frac_hi, a1 = a->frac_lo;
1378
1379 if (c & 64) {
1380 a0 = a1, a1 = 0;
1381 }
1382
1383 c &= 63;
1384 if (c) {
1385 a0 = shl_double(a0, a1, c);
1386 a1 = a1 << c;
1387 }
1388
1389 a->frac_hi = a0;
1390 a->frac_lo = a1;
1391 }
1392
1393 #define frac_shl(A, C) FRAC_GENERIC_64_128(shl, A)(A, C)
1394
1395 static void frac64_shr(FloatParts64 *a, int c)
1396 {
1397 a->frac >>= c;
1398 }
1399
1400 static void frac128_shr(FloatParts128 *a, int c)
1401 {
1402 uint64_t a0 = a->frac_hi, a1 = a->frac_lo;
1403
1404 if (c & 64) {
1405 a1 = a0, a0 = 0;
1406 }
1407
1408 c &= 63;
1409 if (c) {
1410 a1 = shr_double(a0, a1, c);
1411 a0 = a0 >> c;
1412 }
1413
1414 a->frac_hi = a0;
1415 a->frac_lo = a1;
1416 }
1417
1418 #define frac_shr(A, C) FRAC_GENERIC_64_128(shr, A)(A, C)
1419
1420 static void frac64_shrjam(FloatParts64 *a, int c)
1421 {
1422 uint64_t a0 = a->frac;
1423
1424 if (likely(c != 0)) {
1425 if (likely(c < 64)) {
1426 a0 = (a0 >> c) | (shr_double(a0, 0, c) != 0);
1427 } else {
1428 a0 = a0 != 0;
1429 }
1430 a->frac = a0;
1431 }
1432 }
1433
1434 static void frac128_shrjam(FloatParts128 *a, int c)
1435 {
1436 uint64_t a0 = a->frac_hi, a1 = a->frac_lo;
1437 uint64_t sticky = 0;
1438
1439 if (unlikely(c == 0)) {
1440 return;
1441 } else if (likely(c < 64)) {
1442 /* nothing */
1443 } else if (likely(c < 128)) {
1444 sticky = a1;
1445 a1 = a0;
1446 a0 = 0;
1447 c &= 63;
1448 if (c == 0) {
1449 goto done;
1450 }
1451 } else {
1452 sticky = a0 | a1;
1453 a0 = a1 = 0;
1454 goto done;
1455 }
1456
1457 sticky |= shr_double(a1, 0, c);
1458 a1 = shr_double(a0, a1, c);
1459 a0 = a0 >> c;
1460
1461 done:
1462 a->frac_lo = a1 | (sticky != 0);
1463 a->frac_hi = a0;
1464 }
1465
1466 static void frac256_shrjam(FloatParts256 *a, int c)
1467 {
1468 uint64_t a0 = a->frac_hi, a1 = a->frac_hm;
1469 uint64_t a2 = a->frac_lm, a3 = a->frac_lo;
1470 uint64_t sticky = 0;
1471
1472 if (unlikely(c == 0)) {
1473 return;
1474 } else if (likely(c < 64)) {
1475 /* nothing */
1476 } else if (likely(c < 256)) {
1477 if (unlikely(c & 128)) {
1478 sticky |= a2 | a3;
1479 a3 = a1, a2 = a0, a1 = 0, a0 = 0;
1480 }
1481 if (unlikely(c & 64)) {
1482 sticky |= a3;
1483 a3 = a2, a2 = a1, a1 = a0, a0 = 0;
1484 }
1485 c &= 63;
1486 if (c == 0) {
1487 goto done;
1488 }
1489 } else {
1490 sticky = a0 | a1 | a2 | a3;
1491 a0 = a1 = a2 = a3 = 0;
1492 goto done;
1493 }
1494
1495 sticky |= shr_double(a3, 0, c);
1496 a3 = shr_double(a2, a3, c);
1497 a2 = shr_double(a1, a2, c);
1498 a1 = shr_double(a0, a1, c);
1499 a0 = a0 >> c;
1500
1501 done:
1502 a->frac_lo = a3 | (sticky != 0);
1503 a->frac_lm = a2;
1504 a->frac_hm = a1;
1505 a->frac_hi = a0;
1506 }
1507
1508 #define frac_shrjam(A, C) FRAC_GENERIC_64_128_256(shrjam, A)(A, C)
1509
1510 static bool frac64_sub(FloatParts64 *r, FloatParts64 *a, FloatParts64 *b)
1511 {
1512 return usub64_overflow(a->frac, b->frac, &r->frac);
1513 }
1514
1515 static bool frac128_sub(FloatParts128 *r, FloatParts128 *a, FloatParts128 *b)
1516 {
1517 bool c = 0;
1518 r->frac_lo = usub64_borrow(a->frac_lo, b->frac_lo, &c);
1519 r->frac_hi = usub64_borrow(a->frac_hi, b->frac_hi, &c);
1520 return c;
1521 }
1522
1523 static bool frac256_sub(FloatParts256 *r, FloatParts256 *a, FloatParts256 *b)
1524 {
1525 bool c = 0;
1526 r->frac_lo = usub64_borrow(a->frac_lo, b->frac_lo, &c);
1527 r->frac_lm = usub64_borrow(a->frac_lm, b->frac_lm, &c);
1528 r->frac_hm = usub64_borrow(a->frac_hm, b->frac_hm, &c);
1529 r->frac_hi = usub64_borrow(a->frac_hi, b->frac_hi, &c);
1530 return c;
1531 }
1532
1533 #define frac_sub(R, A, B) FRAC_GENERIC_64_128_256(sub, R)(R, A, B)
1534
1535 static void frac64_truncjam(FloatParts64 *r, FloatParts128 *a)
1536 {
1537 r->frac = a->frac_hi | (a->frac_lo != 0);
1538 }
1539
1540 static void frac128_truncjam(FloatParts128 *r, FloatParts256 *a)
1541 {
1542 r->frac_hi = a->frac_hi;
1543 r->frac_lo = a->frac_hm | ((a->frac_lm | a->frac_lo) != 0);
1544 }
1545
1546 #define frac_truncjam(R, A) FRAC_GENERIC_64_128(truncjam, R)(R, A)
1547
1548 static void frac64_widen(FloatParts128 *r, FloatParts64 *a)
1549 {
1550 r->frac_hi = a->frac;
1551 r->frac_lo = 0;
1552 }
1553
1554 static void frac128_widen(FloatParts256 *r, FloatParts128 *a)
1555 {
1556 r->frac_hi = a->frac_hi;
1557 r->frac_hm = a->frac_lo;
1558 r->frac_lm = 0;
1559 r->frac_lo = 0;
1560 }
1561
1562 #define frac_widen(A, B) FRAC_GENERIC_64_128(widen, B)(A, B)
1563
1564 /*
1565 * Reciprocal sqrt table. 1 bit of exponent, 6-bits of mantessa.
1566 * From https://git.musl-libc.org/cgit/musl/tree/src/math/sqrt_data.c
1567 * and thus MIT licenced.
1568 */
1569 static const uint16_t rsqrt_tab[128] = {
1570 0xb451, 0xb2f0, 0xb196, 0xb044, 0xaef9, 0xadb6, 0xac79, 0xab43,
1571 0xaa14, 0xa8eb, 0xa7c8, 0xa6aa, 0xa592, 0xa480, 0xa373, 0xa26b,
1572 0xa168, 0xa06a, 0x9f70, 0x9e7b, 0x9d8a, 0x9c9d, 0x9bb5, 0x9ad1,
1573 0x99f0, 0x9913, 0x983a, 0x9765, 0x9693, 0x95c4, 0x94f8, 0x9430,
1574 0x936b, 0x92a9, 0x91ea, 0x912e, 0x9075, 0x8fbe, 0x8f0a, 0x8e59,
1575 0x8daa, 0x8cfe, 0x8c54, 0x8bac, 0x8b07, 0x8a64, 0x89c4, 0x8925,
1576 0x8889, 0x87ee, 0x8756, 0x86c0, 0x862b, 0x8599, 0x8508, 0x8479,
1577 0x83ec, 0x8361, 0x82d8, 0x8250, 0x81c9, 0x8145, 0x80c2, 0x8040,
1578 0xff02, 0xfd0e, 0xfb25, 0xf947, 0xf773, 0xf5aa, 0xf3ea, 0xf234,
1579 0xf087, 0xeee3, 0xed47, 0xebb3, 0xea27, 0xe8a3, 0xe727, 0xe5b2,
1580 0xe443, 0xe2dc, 0xe17a, 0xe020, 0xdecb, 0xdd7d, 0xdc34, 0xdaf1,
1581 0xd9b3, 0xd87b, 0xd748, 0xd61a, 0xd4f1, 0xd3cd, 0xd2ad, 0xd192,
1582 0xd07b, 0xcf69, 0xce5b, 0xcd51, 0xcc4a, 0xcb48, 0xca4a, 0xc94f,
1583 0xc858, 0xc764, 0xc674, 0xc587, 0xc49d, 0xc3b7, 0xc2d4, 0xc1f4,
1584 0xc116, 0xc03c, 0xbf65, 0xbe90, 0xbdbe, 0xbcef, 0xbc23, 0xbb59,
1585 0xba91, 0xb9cc, 0xb90a, 0xb84a, 0xb78c, 0xb6d0, 0xb617, 0xb560,
1586 };
1587
1588 #define partsN(NAME) glue(glue(glue(parts,N),_),NAME)
1589 #define FloatPartsN glue(FloatParts,N)
1590 #define FloatPartsW glue(FloatParts,W)
1591
1592 #define N 64
1593 #define W 128
1594
1595 #include "softfloat-parts-addsub.c.inc"
1596 #include "softfloat-parts.c.inc"
1597
1598 #undef N
1599 #undef W
1600 #define N 128
1601 #define W 256
1602
1603 #include "softfloat-parts-addsub.c.inc"
1604 #include "softfloat-parts.c.inc"
1605
1606 #undef N
1607 #undef W
1608 #define N 256
1609
1610 #include "softfloat-parts-addsub.c.inc"
1611
1612 #undef N
1613 #undef W
1614 #undef partsN
1615 #undef FloatPartsN
1616 #undef FloatPartsW
1617
1618 /*
1619 * Pack/unpack routines with a specific FloatFmt.
1620 */
1621
1622 static void float16a_unpack_canonical(FloatParts64 *p, float16 f,
1623 float_status *s, const FloatFmt *params)
1624 {
1625 float16_unpack_raw(p, f);
1626 parts_canonicalize(p, s, params);
1627 }
1628
1629 static void float16_unpack_canonical(FloatParts64 *p, float16 f,
1630 float_status *s)
1631 {
1632 float16a_unpack_canonical(p, f, s, &float16_params);
1633 }
1634
1635 static void bfloat16_unpack_canonical(FloatParts64 *p, bfloat16 f,
1636 float_status *s)
1637 {
1638 bfloat16_unpack_raw(p, f);
1639 parts_canonicalize(p, s, &bfloat16_params);
1640 }
1641
1642 static float16 float16a_round_pack_canonical(FloatParts64 *p,
1643 float_status *s,
1644 const FloatFmt *params)
1645 {
1646 parts_uncanon(p, s, params);
1647 return float16_pack_raw(p);
1648 }
1649
1650 static float16 float16_round_pack_canonical(FloatParts64 *p,
1651 float_status *s)
1652 {
1653 return float16a_round_pack_canonical(p, s, &float16_params);
1654 }
1655
1656 static bfloat16 bfloat16_round_pack_canonical(FloatParts64 *p,
1657 float_status *s)
1658 {
1659 parts_uncanon(p, s, &bfloat16_params);
1660 return bfloat16_pack_raw(p);
1661 }
1662
1663 static void float32_unpack_canonical(FloatParts64 *p, float32 f,
1664 float_status *s)
1665 {
1666 float32_unpack_raw(p, f);
1667 parts_canonicalize(p, s, &float32_params);
1668 }
1669
1670 static float32 float32_round_pack_canonical(FloatParts64 *p,
1671 float_status *s)
1672 {
1673 parts_uncanon(p, s, &float32_params);
1674 return float32_pack_raw(p);
1675 }
1676
1677 static void float64_unpack_canonical(FloatParts64 *p, float64 f,
1678 float_status *s)
1679 {
1680 float64_unpack_raw(p, f);
1681 parts_canonicalize(p, s, &float64_params);
1682 }
1683
1684 static float64 float64_round_pack_canonical(FloatParts64 *p,
1685 float_status *s)
1686 {
1687 parts_uncanon(p, s, &float64_params);
1688 return float64_pack_raw(p);
1689 }
1690
1691 static void float128_unpack_canonical(FloatParts128 *p, float128 f,
1692 float_status *s)
1693 {
1694 float128_unpack_raw(p, f);
1695 parts_canonicalize(p, s, &float128_params);
1696 }
1697
1698 static float128 float128_round_pack_canonical(FloatParts128 *p,
1699 float_status *s)
1700 {
1701 parts_uncanon(p, s, &float128_params);
1702 return float128_pack_raw(p);
1703 }
1704
1705 /* Returns false if the encoding is invalid. */
1706 static bool floatx80_unpack_canonical(FloatParts128 *p, floatx80 f,
1707 float_status *s)
1708 {
1709 /* Ensure rounding precision is set before beginning. */
1710 switch (s->floatx80_rounding_precision) {
1711 case floatx80_precision_x:
1712 case floatx80_precision_d:
1713 case floatx80_precision_s:
1714 break;
1715 default:
1716 g_assert_not_reached();
1717 }
1718
1719 if (unlikely(floatx80_invalid_encoding(f))) {
1720 float_raise(float_flag_invalid, s);
1721 return false;
1722 }
1723
1724 floatx80_unpack_raw(p, f);
1725
1726 if (likely(p->exp != floatx80_params[floatx80_precision_x].exp_max)) {
1727 parts_canonicalize(p, s, &floatx80_params[floatx80_precision_x]);
1728 } else {
1729 /* The explicit integer bit is ignored, after invalid checks. */
1730 p->frac_hi &= MAKE_64BIT_MASK(0, 63);
1731 p->cls = (p->frac_hi == 0 ? float_class_inf
1732 : parts_is_snan_frac(p->frac_hi, s)
1733 ? float_class_snan : float_class_qnan);
1734 }
1735 return true;
1736 }
1737
1738 static floatx80 floatx80_round_pack_canonical(FloatParts128 *p,
1739 float_status *s)
1740 {
1741 const FloatFmt *fmt = &floatx80_params[s->floatx80_rounding_precision];
1742 uint64_t frac;
1743 int exp;
1744
1745 switch (p->cls) {
1746 case float_class_normal:
1747 if (s->floatx80_rounding_precision == floatx80_precision_x) {
1748 parts_uncanon_normal(p, s, fmt);
1749 frac = p->frac_hi;
1750 exp = p->exp;
1751 } else {
1752 FloatParts64 p64;
1753
1754 p64.sign = p->sign;
1755 p64.exp = p->exp;
1756 frac_truncjam(&p64, p);
1757 parts_uncanon_normal(&p64, s, fmt);
1758 frac = p64.frac;
1759 exp = p64.exp;
1760 }
1761 if (exp != fmt->exp_max) {
1762 break;
1763 }
1764 /* rounded to inf -- fall through to set frac correctly */
1765
1766 case float_class_inf:
1767 /* x86 and m68k differ in the setting of the integer bit. */
1768 frac = floatx80_infinity_low;
1769 exp = fmt->exp_max;
1770 break;
1771
1772 case float_class_zero:
1773 frac = 0;
1774 exp = 0;
1775 break;
1776
1777 case float_class_snan:
1778 case float_class_qnan:
1779 /* NaNs have the integer bit set. */
1780 frac = p->frac_hi | (1ull << 63);
1781 exp = fmt->exp_max;
1782 break;
1783
1784 default:
1785 g_assert_not_reached();
1786 }
1787
1788 return packFloatx80(p->sign, exp, frac);
1789 }
1790
1791 /*
1792 * Addition and subtraction
1793 */
1794
1795 static float16 QEMU_FLATTEN
1796 float16_addsub(float16 a, float16 b, float_status *status, bool subtract)
1797 {
1798 FloatParts64 pa, pb, *pr;
1799
1800 float16_unpack_canonical(&pa, a, status);
1801 float16_unpack_canonical(&pb, b, status);
1802 pr = parts_addsub(&pa, &pb, status, subtract);
1803
1804 return float16_round_pack_canonical(pr, status);
1805 }
1806
1807 float16 float16_add(float16 a, float16 b, float_status *status)
1808 {
1809 return float16_addsub(a, b, status, false);
1810 }
1811
1812 float16 float16_sub(float16 a, float16 b, float_status *status)
1813 {
1814 return float16_addsub(a, b, status, true);
1815 }
1816
1817 static float32 QEMU_SOFTFLOAT_ATTR
1818 soft_f32_addsub(float32 a, float32 b, float_status *status, bool subtract)
1819 {
1820 FloatParts64 pa, pb, *pr;
1821
1822 float32_unpack_canonical(&pa, a, status);
1823 float32_unpack_canonical(&pb, b, status);
1824 pr = parts_addsub(&pa, &pb, status, subtract);
1825
1826 return float32_round_pack_canonical(pr, status);
1827 }
1828
1829 static float32 soft_f32_add(float32 a, float32 b, float_status *status)
1830 {
1831 return soft_f32_addsub(a, b, status, false);
1832 }
1833
1834 static float32 soft_f32_sub(float32 a, float32 b, float_status *status)
1835 {
1836 return soft_f32_addsub(a, b, status, true);
1837 }
1838
1839 static float64 QEMU_SOFTFLOAT_ATTR
1840 soft_f64_addsub(float64 a, float64 b, float_status *status, bool subtract)
1841 {
1842 FloatParts64 pa, pb, *pr;
1843
1844 float64_unpack_canonical(&pa, a, status);
1845 float64_unpack_canonical(&pb, b, status);
1846 pr = parts_addsub(&pa, &pb, status, subtract);
1847
1848 return float64_round_pack_canonical(pr, status);
1849 }
1850
1851 static float64 soft_f64_add(float64 a, float64 b, float_status *status)
1852 {
1853 return soft_f64_addsub(a, b, status, false);
1854 }
1855
1856 static float64 soft_f64_sub(float64 a, float64 b, float_status *status)
1857 {
1858 return soft_f64_addsub(a, b, status, true);
1859 }
1860
1861 static float hard_f32_add(float a, float b)
1862 {
1863 return a + b;
1864 }
1865
1866 static float hard_f32_sub(float a, float b)
1867 {
1868 return a - b;
1869 }
1870
1871 static double hard_f64_add(double a, double b)
1872 {
1873 return a + b;
1874 }
1875
1876 static double hard_f64_sub(double a, double b)
1877 {
1878 return a - b;
1879 }
1880
1881 static bool f32_addsubmul_post(union_float32 a, union_float32 b)
1882 {
1883 if (QEMU_HARDFLOAT_2F32_USE_FP) {
1884 return !(fpclassify(a.h) == FP_ZERO && fpclassify(b.h) == FP_ZERO);
1885 }
1886 return !(float32_is_zero(a.s) && float32_is_zero(b.s));
1887 }
1888
1889 static bool f64_addsubmul_post(union_float64 a, union_float64 b)
1890 {
1891 if (QEMU_HARDFLOAT_2F64_USE_FP) {
1892 return !(fpclassify(a.h) == FP_ZERO && fpclassify(b.h) == FP_ZERO);
1893 } else {
1894 return !(float64_is_zero(a.s) && float64_is_zero(b.s));
1895 }
1896 }
1897
1898 static float32 float32_addsub(float32 a, float32 b, float_status *s,
1899 hard_f32_op2_fn hard, soft_f32_op2_fn soft)
1900 {
1901 return float32_gen2(a, b, s, hard, soft,
1902 f32_is_zon2, f32_addsubmul_post);
1903 }
1904
1905 static float64 float64_addsub(float64 a, float64 b, float_status *s,
1906 hard_f64_op2_fn hard, soft_f64_op2_fn soft)
1907 {
1908 return float64_gen2(a, b, s, hard, soft,
1909 f64_is_zon2, f64_addsubmul_post);
1910 }
1911
1912 float32 QEMU_FLATTEN
1913 float32_add(float32 a, float32 b, float_status *s)
1914 {
1915 return float32_addsub(a, b, s, hard_f32_add, soft_f32_add);
1916 }
1917
1918 float32 QEMU_FLATTEN
1919 float32_sub(float32 a, float32 b, float_status *s)
1920 {
1921 return float32_addsub(a, b, s, hard_f32_sub, soft_f32_sub);
1922 }
1923
1924 float64 QEMU_FLATTEN
1925 float64_add(float64 a, float64 b, float_status *s)
1926 {
1927 return float64_addsub(a, b, s, hard_f64_add, soft_f64_add);
1928 }
1929
1930 float64 QEMU_FLATTEN
1931 float64_sub(float64 a, float64 b, float_status *s)
1932 {
1933 return float64_addsub(a, b, s, hard_f64_sub, soft_f64_sub);
1934 }
1935
1936 static bfloat16 QEMU_FLATTEN
1937 bfloat16_addsub(bfloat16 a, bfloat16 b, float_status *status, bool subtract)
1938 {
1939 FloatParts64 pa, pb, *pr;
1940
1941 bfloat16_unpack_canonical(&pa, a, status);
1942 bfloat16_unpack_canonical(&pb, b, status);
1943 pr = parts_addsub(&pa, &pb, status, subtract);
1944
1945 return bfloat16_round_pack_canonical(pr, status);
1946 }
1947
1948 bfloat16 bfloat16_add(bfloat16 a, bfloat16 b, float_status *status)
1949 {
1950 return bfloat16_addsub(a, b, status, false);
1951 }
1952
1953 bfloat16 bfloat16_sub(bfloat16 a, bfloat16 b, float_status *status)
1954 {
1955 return bfloat16_addsub(a, b, status, true);
1956 }
1957
1958 static float128 QEMU_FLATTEN
1959 float128_addsub(float128 a, float128 b, float_status *status, bool subtract)
1960 {
1961 FloatParts128 pa, pb, *pr;
1962
1963 float128_unpack_canonical(&pa, a, status);
1964 float128_unpack_canonical(&pb, b, status);
1965 pr = parts_addsub(&pa, &pb, status, subtract);
1966
1967 return float128_round_pack_canonical(pr, status);
1968 }
1969
1970 float128 float128_add(float128 a, float128 b, float_status *status)
1971 {
1972 return float128_addsub(a, b, status, false);
1973 }
1974
1975 float128 float128_sub(float128 a, float128 b, float_status *status)
1976 {
1977 return float128_addsub(a, b, status, true);
1978 }
1979
1980 static floatx80 QEMU_FLATTEN
1981 floatx80_addsub(floatx80 a, floatx80 b, float_status *status, bool subtract)
1982 {
1983 FloatParts128 pa, pb, *pr;
1984
1985 if (!floatx80_unpack_canonical(&pa, a, status) ||
1986 !floatx80_unpack_canonical(&pb, b, status)) {
1987 return floatx80_default_nan(status);
1988 }
1989
1990 pr = parts_addsub(&pa, &pb, status, subtract);
1991 return floatx80_round_pack_canonical(pr, status);
1992 }
1993
1994 floatx80 floatx80_add(floatx80 a, floatx80 b, float_status *status)
1995 {
1996 return floatx80_addsub(a, b, status, false);
1997 }
1998
1999 floatx80 floatx80_sub(floatx80 a, floatx80 b, float_status *status)
2000 {
2001 return floatx80_addsub(a, b, status, true);
2002 }
2003
2004 /*
2005 * Multiplication
2006 */
2007
2008 float16 QEMU_FLATTEN float16_mul(float16 a, float16 b, float_status *status)
2009 {
2010 FloatParts64 pa, pb, *pr;
2011
2012 float16_unpack_canonical(&pa, a, status);
2013 float16_unpack_canonical(&pb, b, status);
2014 pr = parts_mul(&pa, &pb, status);
2015
2016 return float16_round_pack_canonical(pr, status);
2017 }
2018
2019 static float32 QEMU_SOFTFLOAT_ATTR
2020 soft_f32_mul(float32 a, float32 b, float_status *status)
2021 {
2022 FloatParts64 pa, pb, *pr;
2023
2024 float32_unpack_canonical(&pa, a, status);
2025 float32_unpack_canonical(&pb, b, status);
2026 pr = parts_mul(&pa, &pb, status);
2027
2028 return float32_round_pack_canonical(pr, status);
2029 }
2030
2031 static float64 QEMU_SOFTFLOAT_ATTR
2032 soft_f64_mul(float64 a, float64 b, float_status *status)
2033 {
2034 FloatParts64 pa, pb, *pr;
2035
2036 float64_unpack_canonical(&pa, a, status);
2037 float64_unpack_canonical(&pb, b, status);
2038 pr = parts_mul(&pa, &pb, status);
2039
2040 return float64_round_pack_canonical(pr, status);
2041 }
2042
2043 static float hard_f32_mul(float a, float b)
2044 {
2045 return a * b;
2046 }
2047
2048 static double hard_f64_mul(double a, double b)
2049 {
2050 return a * b;
2051 }
2052
2053 float32 QEMU_FLATTEN
2054 float32_mul(float32 a, float32 b, float_status *s)
2055 {
2056 return float32_gen2(a, b, s, hard_f32_mul, soft_f32_mul,
2057 f32_is_zon2, f32_addsubmul_post);
2058 }
2059
2060 float64 QEMU_FLATTEN
2061 float64_mul(float64 a, float64 b, float_status *s)
2062 {
2063 return float64_gen2(a, b, s, hard_f64_mul, soft_f64_mul,
2064 f64_is_zon2, f64_addsubmul_post);
2065 }
2066
2067 bfloat16 QEMU_FLATTEN
2068 bfloat16_mul(bfloat16 a, bfloat16 b, float_status *status)
2069 {
2070 FloatParts64 pa, pb, *pr;
2071
2072 bfloat16_unpack_canonical(&pa, a, status);
2073 bfloat16_unpack_canonical(&pb, b, status);
2074 pr = parts_mul(&pa, &pb, status);
2075
2076 return bfloat16_round_pack_canonical(pr, status);
2077 }
2078
2079 float128 QEMU_FLATTEN
2080 float128_mul(float128 a, float128 b, float_status *status)
2081 {
2082 FloatParts128 pa, pb, *pr;
2083
2084 float128_unpack_canonical(&pa, a, status);
2085 float128_unpack_canonical(&pb, b, status);
2086 pr = parts_mul(&pa, &pb, status);
2087
2088 return float128_round_pack_canonical(pr, status);
2089 }
2090
2091 floatx80 QEMU_FLATTEN
2092 floatx80_mul(floatx80 a, floatx80 b, float_status *status)
2093 {
2094 FloatParts128 pa, pb, *pr;
2095
2096 if (!floatx80_unpack_canonical(&pa, a, status) ||
2097 !floatx80_unpack_canonical(&pb, b, status)) {
2098 return floatx80_default_nan(status);
2099 }
2100
2101 pr = parts_mul(&pa, &pb, status);
2102 return floatx80_round_pack_canonical(pr, status);
2103 }
2104
2105 /*
2106 * Fused multiply-add
2107 */
2108
2109 float16 QEMU_FLATTEN float16_muladd(float16 a, float16 b, float16 c,
2110 int flags, float_status *status)
2111 {
2112 FloatParts64 pa, pb, pc, *pr;
2113
2114 float16_unpack_canonical(&pa, a, status);
2115 float16_unpack_canonical(&pb, b, status);
2116 float16_unpack_canonical(&pc, c, status);
2117 pr = parts_muladd(&pa, &pb, &pc, flags, status);
2118
2119 return float16_round_pack_canonical(pr, status);
2120 }
2121
2122 static float32 QEMU_SOFTFLOAT_ATTR
2123 soft_f32_muladd(float32 a, float32 b, float32 c, int flags,
2124 float_status *status)
2125 {
2126 FloatParts64 pa, pb, pc, *pr;
2127
2128 float32_unpack_canonical(&pa, a, status);
2129 float32_unpack_canonical(&pb, b, status);
2130 float32_unpack_canonical(&pc, c, status);
2131 pr = parts_muladd(&pa, &pb, &pc, flags, status);
2132
2133 return float32_round_pack_canonical(pr, status);
2134 }
2135
2136 static float64 QEMU_SOFTFLOAT_ATTR
2137 soft_f64_muladd(float64 a, float64 b, float64 c, int flags,
2138 float_status *status)
2139 {
2140 FloatParts64 pa, pb, pc, *pr;
2141
2142 float64_unpack_canonical(&pa, a, status);
2143 float64_unpack_canonical(&pb, b, status);
2144 float64_unpack_canonical(&pc, c, status);
2145 pr = parts_muladd(&pa, &pb, &pc, flags, status);
2146
2147 return float64_round_pack_canonical(pr, status);
2148 }
2149
2150 static bool force_soft_fma;
2151
2152 float32 QEMU_FLATTEN
2153 float32_muladd(float32 xa, float32 xb, float32 xc, int flags, float_status *s)
2154 {
2155 union_float32 ua, ub, uc, ur;
2156
2157 ua.s = xa;
2158 ub.s = xb;
2159 uc.s = xc;
2160
2161 if (unlikely(!can_use_fpu(s))) {
2162 goto soft;
2163 }
2164 if (unlikely(flags & float_muladd_halve_result)) {
2165 goto soft;
2166 }
2167
2168 float32_input_flush3(&ua.s, &ub.s, &uc.s, s);
2169 if (unlikely(!f32_is_zon3(ua, ub, uc))) {
2170 goto soft;
2171 }
2172
2173 if (unlikely(force_soft_fma)) {
2174 goto soft;
2175 }
2176
2177 /*
2178 * When (a || b) == 0, there's no need to check for under/over flow,
2179 * since we know the addend is (normal || 0) and the product is 0.
2180 */
2181 if (float32_is_zero(ua.s) || float32_is_zero(ub.s)) {
2182 union_float32 up;
2183 bool prod_sign;
2184
2185 prod_sign = float32_is_neg(ua.s) ^ float32_is_neg(ub.s);
2186 prod_sign ^= !!(flags & float_muladd_negate_product);
2187 up.s = float32_set_sign(float32_zero, prod_sign);
2188
2189 if (flags & float_muladd_negate_c) {
2190 uc.h = -uc.h;
2191 }
2192 ur.h = up.h + uc.h;
2193 } else {
2194 union_float32 ua_orig = ua;
2195 union_float32 uc_orig = uc;
2196
2197 if (flags & float_muladd_negate_product) {
2198 ua.h = -ua.h;
2199 }
2200 if (flags & float_muladd_negate_c) {
2201 uc.h = -uc.h;
2202 }
2203
2204 ur.h = fmaf(ua.h, ub.h, uc.h);
2205
2206 if (unlikely(f32_is_inf(ur))) {
2207 float_raise(float_flag_overflow, s);
2208 } else if (unlikely(fabsf(ur.h) <= FLT_MIN)) {
2209 ua = ua_orig;
2210 uc = uc_orig;
2211 goto soft;
2212 }
2213 }
2214 if (flags & float_muladd_negate_result) {
2215 return float32_chs(ur.s);
2216 }
2217 return ur.s;
2218
2219 soft:
2220 return soft_f32_muladd(ua.s, ub.s, uc.s, flags, s);
2221 }
2222
2223 float64 QEMU_FLATTEN
2224 float64_muladd(float64 xa, float64 xb, float64 xc, int flags, float_status *s)
2225 {
2226 union_float64 ua, ub, uc, ur;
2227
2228 ua.s = xa;
2229 ub.s = xb;
2230 uc.s = xc;
2231
2232 if (unlikely(!can_use_fpu(s))) {
2233 goto soft;
2234 }
2235 if (unlikely(flags & float_muladd_halve_result)) {
2236 goto soft;
2237 }
2238
2239 float64_input_flush3(&ua.s, &ub.s, &uc.s, s);
2240 if (unlikely(!f64_is_zon3(ua, ub, uc))) {
2241 goto soft;
2242 }
2243
2244 if (unlikely(force_soft_fma)) {
2245 goto soft;
2246 }
2247
2248 /*
2249 * When (a || b) == 0, there's no need to check for under/over flow,
2250 * since we know the addend is (normal || 0) and the product is 0.
2251 */
2252 if (float64_is_zero(ua.s) || float64_is_zero(ub.s)) {
2253 union_float64 up;
2254 bool prod_sign;
2255
2256 prod_sign = float64_is_neg(ua.s) ^ float64_is_neg(ub.s);
2257 prod_sign ^= !!(flags & float_muladd_negate_product);
2258 up.s = float64_set_sign(float64_zero, prod_sign);
2259
2260 if (flags & float_muladd_negate_c) {
2261 uc.h = -uc.h;
2262 }
2263 ur.h = up.h + uc.h;
2264 } else {
2265 union_float64 ua_orig = ua;
2266 union_float64 uc_orig = uc;
2267
2268 if (flags & float_muladd_negate_product) {
2269 ua.h = -ua.h;
2270 }
2271 if (flags & float_muladd_negate_c) {
2272 uc.h = -uc.h;
2273 }
2274
2275 ur.h = fma(ua.h, ub.h, uc.h);
2276
2277 if (unlikely(f64_is_inf(ur))) {
2278 float_raise(float_flag_overflow, s);
2279 } else if (unlikely(fabs(ur.h) <= FLT_MIN)) {
2280 ua = ua_orig;
2281 uc = uc_orig;
2282 goto soft;
2283 }
2284 }
2285 if (flags & float_muladd_negate_result) {
2286 return float64_chs(ur.s);
2287 }
2288 return ur.s;
2289
2290 soft:
2291 return soft_f64_muladd(ua.s, ub.s, uc.s, flags, s);
2292 }
2293
2294 bfloat16 QEMU_FLATTEN bfloat16_muladd(bfloat16 a, bfloat16 b, bfloat16 c,
2295 int flags, float_status *status)
2296 {
2297 FloatParts64 pa, pb, pc, *pr;
2298
2299 bfloat16_unpack_canonical(&pa, a, status);
2300 bfloat16_unpack_canonical(&pb, b, status);
2301 bfloat16_unpack_canonical(&pc, c, status);
2302 pr = parts_muladd(&pa, &pb, &pc, flags, status);
2303
2304 return bfloat16_round_pack_canonical(pr, status);
2305 }
2306
2307 float128 QEMU_FLATTEN float128_muladd(float128 a, float128 b, float128 c,
2308 int flags, float_status *status)
2309 {
2310 FloatParts128 pa, pb, pc, *pr;
2311
2312 float128_unpack_canonical(&pa, a, status);
2313 float128_unpack_canonical(&pb, b, status);
2314 float128_unpack_canonical(&pc, c, status);
2315 pr = parts_muladd(&pa, &pb, &pc, flags, status);
2316
2317 return float128_round_pack_canonical(pr, status);
2318 }
2319
2320 /*
2321 * Division
2322 */
2323
2324 float16 float16_div(float16 a, float16 b, float_status *status)
2325 {
2326 FloatParts64 pa, pb, *pr;
2327
2328 float16_unpack_canonical(&pa, a, status);
2329 float16_unpack_canonical(&pb, b, status);
2330 pr = parts_div(&pa, &pb, status);
2331
2332 return float16_round_pack_canonical(pr, status);
2333 }
2334
2335 static float32 QEMU_SOFTFLOAT_ATTR
2336 soft_f32_div(float32 a, float32 b, float_status *status)
2337 {
2338 FloatParts64 pa, pb, *pr;
2339
2340 float32_unpack_canonical(&pa, a, status);
2341 float32_unpack_canonical(&pb, b, status);
2342 pr = parts_div(&pa, &pb, status);
2343
2344 return float32_round_pack_canonical(pr, status);
2345 }
2346
2347 static float64 QEMU_SOFTFLOAT_ATTR
2348 soft_f64_div(float64 a, float64 b, float_status *status)
2349 {
2350 FloatParts64 pa, pb, *pr;
2351
2352 float64_unpack_canonical(&pa, a, status);
2353 float64_unpack_canonical(&pb, b, status);
2354 pr = parts_div(&pa, &pb, status);
2355
2356 return float64_round_pack_canonical(pr, status);
2357 }
2358
2359 static float hard_f32_div(float a, float b)
2360 {
2361 return a / b;
2362 }
2363
2364 static double hard_f64_div(double a, double b)
2365 {
2366 return a / b;
2367 }
2368
2369 static bool f32_div_pre(union_float32 a, union_float32 b)
2370 {
2371 if (QEMU_HARDFLOAT_2F32_USE_FP) {
2372 return (fpclassify(a.h) == FP_NORMAL || fpclassify(a.h) == FP_ZERO) &&
2373 fpclassify(b.h) == FP_NORMAL;
2374 }
2375 return float32_is_zero_or_normal(a.s) && float32_is_normal(b.s);
2376 }
2377
2378 static bool f64_div_pre(union_float64 a, union_float64 b)
2379 {
2380 if (QEMU_HARDFLOAT_2F64_USE_FP) {
2381 return (fpclassify(a.h) == FP_NORMAL || fpclassify(a.h) == FP_ZERO) &&
2382 fpclassify(b.h) == FP_NORMAL;
2383 }
2384 return float64_is_zero_or_normal(a.s) && float64_is_normal(b.s);
2385 }
2386
2387 static bool f32_div_post(union_float32 a, union_float32 b)
2388 {
2389 if (QEMU_HARDFLOAT_2F32_USE_FP) {
2390 return fpclassify(a.h) != FP_ZERO;
2391 }
2392 return !float32_is_zero(a.s);
2393 }
2394
2395 static bool f64_div_post(union_float64 a, union_float64 b)
2396 {
2397 if (QEMU_HARDFLOAT_2F64_USE_FP) {
2398 return fpclassify(a.h) != FP_ZERO;
2399 }
2400 return !float64_is_zero(a.s);
2401 }
2402
2403 float32 QEMU_FLATTEN
2404 float32_div(float32 a, float32 b, float_status *s)
2405 {
2406 return float32_gen2(a, b, s, hard_f32_div, soft_f32_div,
2407 f32_div_pre, f32_div_post);
2408 }
2409
2410 float64 QEMU_FLATTEN
2411 float64_div(float64 a, float64 b, float_status *s)
2412 {
2413 return float64_gen2(a, b, s, hard_f64_div, soft_f64_div,
2414 f64_div_pre, f64_div_post);
2415 }
2416
2417 bfloat16 QEMU_FLATTEN
2418 bfloat16_div(bfloat16 a, bfloat16 b, float_status *status)
2419 {
2420 FloatParts64 pa, pb, *pr;
2421
2422 bfloat16_unpack_canonical(&pa, a, status);
2423 bfloat16_unpack_canonical(&pb, b, status);
2424 pr = parts_div(&pa, &pb, status);
2425
2426 return bfloat16_round_pack_canonical(pr, status);
2427 }
2428
2429 float128 QEMU_FLATTEN
2430 float128_div(float128 a, float128 b, float_status *status)
2431 {
2432 FloatParts128 pa, pb, *pr;
2433
2434 float128_unpack_canonical(&pa, a, status);
2435 float128_unpack_canonical(&pb, b, status);
2436 pr = parts_div(&pa, &pb, status);
2437
2438 return float128_round_pack_canonical(pr, status);
2439 }
2440
2441 floatx80 floatx80_div(floatx80 a, floatx80 b, float_status *status)
2442 {
2443 FloatParts128 pa, pb, *pr;
2444
2445 if (!floatx80_unpack_canonical(&pa, a, status) ||
2446 !floatx80_unpack_canonical(&pb, b, status)) {
2447 return floatx80_default_nan(status);
2448 }
2449
2450 pr = parts_div(&pa, &pb, status);
2451 return floatx80_round_pack_canonical(pr, status);
2452 }
2453
2454 /*
2455 * Remainder
2456 */
2457
2458 float32 float32_rem(float32 a, float32 b, float_status *status)
2459 {
2460 FloatParts64 pa, pb, *pr;
2461
2462 float32_unpack_canonical(&pa, a, status);
2463 float32_unpack_canonical(&pb, b, status);
2464 pr = parts_modrem(&pa, &pb, NULL, status);
2465
2466 return float32_round_pack_canonical(pr, status);
2467 }
2468
2469 float64 float64_rem(float64 a, float64 b, float_status *status)
2470 {
2471 FloatParts64 pa, pb, *pr;
2472
2473 float64_unpack_canonical(&pa, a, status);
2474 float64_unpack_canonical(&pb, b, status);
2475 pr = parts_modrem(&pa, &pb, NULL, status);
2476
2477 return float64_round_pack_canonical(pr, status);
2478 }
2479
2480 float128 float128_rem(float128 a, float128 b, float_status *status)
2481 {
2482 FloatParts128 pa, pb, *pr;
2483
2484 float128_unpack_canonical(&pa, a, status);
2485 float128_unpack_canonical(&pb, b, status);
2486 pr = parts_modrem(&pa, &pb, NULL, status);
2487
2488 return float128_round_pack_canonical(pr, status);
2489 }
2490
2491 /*
2492 * Returns the remainder of the extended double-precision floating-point value
2493 * `a' with respect to the corresponding value `b'.
2494 * If 'mod' is false, the operation is performed according to the IEC/IEEE
2495 * Standard for Binary Floating-Point Arithmetic. If 'mod' is true, return
2496 * the remainder based on truncating the quotient toward zero instead and
2497 * *quotient is set to the low 64 bits of the absolute value of the integer
2498 * quotient.
2499 */
2500 floatx80 floatx80_modrem(floatx80 a, floatx80 b, bool mod,
2501 uint64_t *quotient, float_status *status)
2502 {
2503 FloatParts128 pa, pb, *pr;
2504
2505 *quotient = 0;
2506 if (!floatx80_unpack_canonical(&pa, a, status) ||
2507 !floatx80_unpack_canonical(&pb, b, status)) {
2508 return floatx80_default_nan(status);
2509 }
2510 pr = parts_modrem(&pa, &pb, mod ? quotient : NULL, status);
2511
2512 return floatx80_round_pack_canonical(pr, status);
2513 }
2514
2515 floatx80 floatx80_rem(floatx80 a, floatx80 b, float_status *status)
2516 {
2517 uint64_t quotient;
2518 return floatx80_modrem(a, b, false, &quotient, status);
2519 }
2520
2521 floatx80 floatx80_mod(floatx80 a, floatx80 b, float_status *status)
2522 {
2523 uint64_t quotient;
2524 return floatx80_modrem(a, b, true, &quotient, status);
2525 }
2526
2527 /*
2528 * Float to Float conversions
2529 *
2530 * Returns the result of converting one float format to another. The
2531 * conversion is performed according to the IEC/IEEE Standard for
2532 * Binary Floating-Point Arithmetic.
2533 *
2534 * Usually this only needs to take care of raising invalid exceptions
2535 * and handling the conversion on NaNs.
2536 */
2537
2538 static void parts_float_to_ahp(FloatParts64 *a, float_status *s)
2539 {
2540 switch (a->cls) {
2541 case float_class_qnan:
2542 case float_class_snan:
2543 /*
2544 * There is no NaN in the destination format. Raise Invalid
2545 * and return a zero with the sign of the input NaN.
2546 */
2547 float_raise(float_flag_invalid, s);
2548 a->cls = float_class_zero;
2549 break;
2550
2551 case float_class_inf:
2552 /*
2553 * There is no Inf in the destination format. Raise Invalid
2554 * and return the maximum normal with the correct sign.
2555 */
2556 float_raise(float_flag_invalid, s);
2557 a->cls = float_class_normal;
2558 a->exp = float16_params_ahp.exp_max;
2559 a->frac = MAKE_64BIT_MASK(float16_params_ahp.frac_shift,
2560 float16_params_ahp.frac_size + 1);
2561 break;
2562
2563 case float_class_normal:
2564 case float_class_zero:
2565 break;
2566
2567 default:
2568 g_assert_not_reached();
2569 }
2570 }
2571
2572 static void parts64_float_to_float(FloatParts64 *a, float_status *s)
2573 {
2574 if (is_nan(a->cls)) {
2575 parts_return_nan(a, s);
2576 }
2577 }
2578
2579 static void parts128_float_to_float(FloatParts128 *a, float_status *s)
2580 {
2581 if (is_nan(a->cls)) {
2582 parts_return_nan(a, s);
2583 }
2584 }
2585
2586 #define parts_float_to_float(P, S) \
2587 PARTS_GENERIC_64_128(float_to_float, P)(P, S)
2588
2589 static void parts_float_to_float_narrow(FloatParts64 *a, FloatParts128 *b,
2590 float_status *s)
2591 {
2592 a->cls = b->cls;
2593 a->sign = b->sign;
2594 a->exp = b->exp;
2595
2596 if (a->cls == float_class_normal) {
2597 frac_truncjam(a, b);
2598 } else if (is_nan(a->cls)) {
2599 /* Discard the low bits of the NaN. */
2600 a->frac = b->frac_hi;
2601 parts_return_nan(a, s);
2602 }
2603 }
2604
2605 static void parts_float_to_float_widen(FloatParts128 *a, FloatParts64 *b,
2606 float_status *s)
2607 {
2608 a->cls = b->cls;
2609 a->sign = b->sign;
2610 a->exp = b->exp;
2611 frac_widen(a, b);
2612
2613 if (is_nan(a->cls)) {
2614 parts_return_nan(a, s);
2615 }
2616 }
2617
2618 float32 float16_to_float32(float16 a, bool ieee, float_status *s)
2619 {
2620 const FloatFmt *fmt16 = ieee ? &float16_params : &float16_params_ahp;
2621 FloatParts64 p;
2622
2623 float16a_unpack_canonical(&p, a, s, fmt16);
2624 parts_float_to_float(&p, s);
2625 return float32_round_pack_canonical(&p, s);
2626 }
2627
2628 float64 float16_to_float64(float16 a, bool ieee, float_status *s)
2629 {
2630 const FloatFmt *fmt16 = ieee ? &float16_params : &float16_params_ahp;
2631 FloatParts64 p;
2632
2633 float16a_unpack_canonical(&p, a, s, fmt16);
2634 parts_float_to_float(&p, s);
2635 return float64_round_pack_canonical(&p, s);
2636 }
2637
2638 float16 float32_to_float16(float32 a, bool ieee, float_status *s)
2639 {
2640 FloatParts64 p;
2641 const FloatFmt *fmt;
2642
2643 float32_unpack_canonical(&p, a, s);
2644 if (ieee) {
2645 parts_float_to_float(&p, s);
2646 fmt = &float16_params;
2647 } else {
2648 parts_float_to_ahp(&p, s);
2649 fmt = &float16_params_ahp;
2650 }
2651 return float16a_round_pack_canonical(&p, s, fmt);
2652 }
2653
2654 static float64 QEMU_SOFTFLOAT_ATTR
2655 soft_float32_to_float64(float32 a, float_status *s)
2656 {
2657 FloatParts64 p;
2658
2659 float32_unpack_canonical(&p, a, s);
2660 parts_float_to_float(&p, s);
2661 return float64_round_pack_canonical(&p, s);
2662 }
2663
2664 float64 float32_to_float64(float32 a, float_status *s)
2665 {
2666 if (likely(float32_is_normal(a))) {
2667 /* Widening conversion can never produce inexact results. */
2668 union_float32 uf;
2669 union_float64 ud;
2670 uf.s = a;
2671 ud.h = uf.h;
2672 return ud.s;
2673 } else if (float32_is_zero(a)) {
2674 return float64_set_sign(float64_zero, float32_is_neg(a));
2675 } else {
2676 return soft_float32_to_float64(a, s);
2677 }
2678 }
2679
2680 float16 float64_to_float16(float64 a, bool ieee, float_status *s)
2681 {
2682 FloatParts64 p;
2683 const FloatFmt *fmt;
2684
2685 float64_unpack_canonical(&p, a, s);
2686 if (ieee) {
2687 parts_float_to_float(&p, s);
2688 fmt = &float16_params;
2689 } else {
2690 parts_float_to_ahp(&p, s);
2691 fmt = &float16_params_ahp;
2692 }
2693 return float16a_round_pack_canonical(&p, s, fmt);
2694 }
2695
2696 float32 float64_to_float32(float64 a, float_status *s)
2697 {
2698 FloatParts64 p;
2699
2700 float64_unpack_canonical(&p, a, s);
2701 parts_float_to_float(&p, s);
2702 return float32_round_pack_canonical(&p, s);
2703 }
2704
2705 float32 bfloat16_to_float32(bfloat16 a, float_status *s)
2706 {
2707 FloatParts64 p;
2708
2709 bfloat16_unpack_canonical(&p, a, s);
2710 parts_float_to_float(&p, s);
2711 return float32_round_pack_canonical(&p, s);
2712 }
2713
2714 float64 bfloat16_to_float64(bfloat16 a, float_status *s)
2715 {
2716 FloatParts64 p;
2717
2718 bfloat16_unpack_canonical(&p, a, s);
2719 parts_float_to_float(&p, s);
2720 return float64_round_pack_canonical(&p, s);
2721 }
2722
2723 bfloat16 float32_to_bfloat16(float32 a, float_status *s)
2724 {
2725 FloatParts64 p;
2726
2727 float32_unpack_canonical(&p, a, s);
2728 parts_float_to_float(&p, s);
2729 return bfloat16_round_pack_canonical(&p, s);
2730 }
2731
2732 bfloat16 float64_to_bfloat16(float64 a, float_status *s)
2733 {
2734 FloatParts64 p;
2735
2736 float64_unpack_canonical(&p, a, s);
2737 parts_float_to_float(&p, s);
2738 return bfloat16_round_pack_canonical(&p, s);
2739 }
2740
2741 float32 float128_to_float32(float128 a, float_status *s)
2742 {
2743 FloatParts64 p64;
2744 FloatParts128 p128;
2745
2746 float128_unpack_canonical(&p128, a, s);
2747 parts_float_to_float_narrow(&p64, &p128, s);
2748 return float32_round_pack_canonical(&p64, s);
2749 }
2750
2751 float64 float128_to_float64(float128 a, float_status *s)
2752 {
2753 FloatParts64 p64;
2754 FloatParts128 p128;
2755
2756 float128_unpack_canonical(&p128, a, s);
2757 parts_float_to_float_narrow(&p64, &p128, s);
2758 return float64_round_pack_canonical(&p64, s);
2759 }
2760
2761 float128 float32_to_float128(float32 a, float_status *s)
2762 {
2763 FloatParts64 p64;
2764 FloatParts128 p128;
2765
2766 float32_unpack_canonical(&p64, a, s);
2767 parts_float_to_float_widen(&p128, &p64, s);
2768 return float128_round_pack_canonical(&p128, s);
2769 }
2770
2771 float128 float64_to_float128(float64 a, float_status *s)
2772 {
2773 FloatParts64 p64;
2774 FloatParts128 p128;
2775
2776 float64_unpack_canonical(&p64, a, s);
2777 parts_float_to_float_widen(&p128, &p64, s);
2778 return float128_round_pack_canonical(&p128, s);
2779 }
2780
2781 float32 floatx80_to_float32(floatx80 a, float_status *s)
2782 {
2783 FloatParts64 p64;
2784 FloatParts128 p128;
2785
2786 if (floatx80_unpack_canonical(&p128, a, s)) {
2787 parts_float_to_float_narrow(&p64, &p128, s);
2788 } else {
2789 parts_default_nan(&p64, s);
2790 }
2791 return float32_round_pack_canonical(&p64, s);
2792 }
2793
2794 float64 floatx80_to_float64(floatx80 a, float_status *s)
2795 {
2796 FloatParts64 p64;
2797 FloatParts128 p128;
2798
2799 if (floatx80_unpack_canonical(&p128, a, s)) {
2800 parts_float_to_float_narrow(&p64, &p128, s);
2801 } else {
2802 parts_default_nan(&p64, s);
2803 }
2804 return float64_round_pack_canonical(&p64, s);
2805 }
2806
2807 float128 floatx80_to_float128(floatx80 a, float_status *s)
2808 {
2809 FloatParts128 p;
2810
2811 if (floatx80_unpack_canonical(&p, a, s)) {
2812 parts_float_to_float(&p, s);
2813 } else {
2814 parts_default_nan(&p, s);
2815 }
2816 return float128_round_pack_canonical(&p, s);
2817 }
2818
2819 floatx80 float32_to_floatx80(float32 a, float_status *s)
2820 {
2821 FloatParts64 p64;
2822 FloatParts128 p128;
2823
2824 float32_unpack_canonical(&p64, a, s);
2825 parts_float_to_float_widen(&p128, &p64, s);
2826 return floatx80_round_pack_canonical(&p128, s);
2827 }
2828
2829 floatx80 float64_to_floatx80(float64 a, float_status *s)
2830 {
2831 FloatParts64 p64;
2832 FloatParts128 p128;
2833
2834 float64_unpack_canonical(&p64, a, s);
2835 parts_float_to_float_widen(&p128, &p64, s);
2836 return floatx80_round_pack_canonical(&p128, s);
2837 }
2838
2839 floatx80 float128_to_floatx80(float128 a, float_status *s)
2840 {
2841 FloatParts128 p;
2842
2843 float128_unpack_canonical(&p, a, s);
2844 parts_float_to_float(&p, s);
2845 return floatx80_round_pack_canonical(&p, s);
2846 }
2847
2848 /*
2849 * Round to integral value
2850 */
2851
2852 float16 float16_round_to_int(float16 a, float_status *s)
2853 {
2854 FloatParts64 p;
2855
2856 float16_unpack_canonical(&p, a, s);
2857 parts_round_to_int(&p, s->float_rounding_mode, 0, s, &float16_params);
2858 return float16_round_pack_canonical(&p, s);
2859 }
2860
2861 float32 float32_round_to_int(float32 a, float_status *s)
2862 {
2863 FloatParts64 p;
2864
2865 float32_unpack_canonical(&p, a, s);
2866 parts_round_to_int(&p, s->float_rounding_mode, 0, s, &float32_params);
2867 return float32_round_pack_canonical(&p, s);
2868 }
2869
2870 float64 float64_round_to_int(float64 a, float_status *s)
2871 {
2872 FloatParts64 p;
2873
2874 float64_unpack_canonical(&p, a, s);
2875 parts_round_to_int(&p, s->float_rounding_mode, 0, s, &float64_params);
2876 return float64_round_pack_canonical(&p, s);
2877 }
2878
2879 bfloat16 bfloat16_round_to_int(bfloat16 a, float_status *s)
2880 {
2881 FloatParts64 p;
2882
2883 bfloat16_unpack_canonical(&p, a, s);
2884 parts_round_to_int(&p, s->float_rounding_mode, 0, s, &bfloat16_params);
2885 return bfloat16_round_pack_canonical(&p, s);
2886 }
2887
2888 float128 float128_round_to_int(float128 a, float_status *s)
2889 {
2890 FloatParts128 p;
2891
2892 float128_unpack_canonical(&p, a, s);
2893 parts_round_to_int(&p, s->float_rounding_mode, 0, s, &float128_params);
2894 return float128_round_pack_canonical(&p, s);
2895 }
2896
2897 floatx80 floatx80_round_to_int(floatx80 a, float_status *status)
2898 {
2899 FloatParts128 p;
2900
2901 if (!floatx80_unpack_canonical(&p, a, status)) {
2902 return floatx80_default_nan(status);
2903 }
2904
2905 parts_round_to_int(&p, status->float_rounding_mode, 0, status,
2906 &floatx80_params[status->floatx80_rounding_precision]);
2907 return floatx80_round_pack_canonical(&p, status);
2908 }
2909
2910 /*
2911 * Floating-point to signed integer conversions
2912 */
2913
2914 int8_t float16_to_int8_scalbn(float16 a, FloatRoundMode rmode, int scale,
2915 float_status *s)
2916 {
2917 FloatParts64 p;
2918
2919 float16_unpack_canonical(&p, a, s);
2920 return parts_float_to_sint(&p, rmode, scale, INT8_MIN, INT8_MAX, s);
2921 }
2922
2923 int16_t float16_to_int16_scalbn(float16 a, FloatRoundMode rmode, int scale,
2924 float_status *s)
2925 {
2926 FloatParts64 p;
2927
2928 float16_unpack_canonical(&p, a, s);
2929 return parts_float_to_sint(&p, rmode, scale, INT16_MIN, INT16_MAX, s);
2930 }
2931
2932 int32_t float16_to_int32_scalbn(float16 a, FloatRoundMode rmode, int scale,
2933 float_status *s)
2934 {
2935 FloatParts64 p;
2936
2937 float16_unpack_canonical(&p, a, s);
2938 return parts_float_to_sint(&p, rmode, scale, INT32_MIN, INT32_MAX, s);
2939 }
2940
2941 int64_t float16_to_int64_scalbn(float16 a, FloatRoundMode rmode, int scale,
2942 float_status *s)
2943 {
2944 FloatParts64 p;
2945
2946 float16_unpack_canonical(&p, a, s);
2947 return parts_float_to_sint(&p, rmode, scale, INT64_MIN, INT64_MAX, s);
2948 }
2949
2950 int16_t float32_to_int16_scalbn(float32 a, FloatRoundMode rmode, int scale,
2951 float_status *s)
2952 {
2953 FloatParts64 p;
2954
2955 float32_unpack_canonical(&p, a, s);
2956 return parts_float_to_sint(&p, rmode, scale, INT16_MIN, INT16_MAX, s);
2957 }
2958
2959 int32_t float32_to_int32_scalbn(float32 a, FloatRoundMode rmode, int scale,
2960 float_status *s)
2961 {
2962 FloatParts64 p;
2963
2964 float32_unpack_canonical(&p, a, s);
2965 return parts_float_to_sint(&p, rmode, scale, INT32_MIN, INT32_MAX, s);
2966 }
2967
2968 int64_t float32_to_int64_scalbn(float32 a, FloatRoundMode rmode, int scale,
2969 float_status *s)
2970 {
2971 FloatParts64 p;
2972
2973 float32_unpack_canonical(&p, a, s);
2974 return parts_float_to_sint(&p, rmode, scale, INT64_MIN, INT64_MAX, s);
2975 }
2976
2977 int16_t float64_to_int16_scalbn(float64 a, FloatRoundMode rmode, int scale,
2978 float_status *s)
2979 {
2980 FloatParts64 p;
2981
2982 float64_unpack_canonical(&p, a, s);
2983 return parts_float_to_sint(&p, rmode, scale, INT16_MIN, INT16_MAX, s);
2984 }
2985
2986 int32_t float64_to_int32_scalbn(float64 a, FloatRoundMode rmode, int scale,
2987 float_status *s)
2988 {
2989 FloatParts64 p;
2990
2991 float64_unpack_canonical(&p, a, s);
2992 return parts_float_to_sint(&p, rmode, scale, INT32_MIN, INT32_MAX, s);
2993 }
2994
2995 int64_t float64_to_int64_scalbn(float64 a, FloatRoundMode rmode, int scale,
2996 float_status *s)
2997 {
2998 FloatParts64 p;
2999
3000 float64_unpack_canonical(&p, a, s);
3001 return parts_float_to_sint(&p, rmode, scale, INT64_MIN, INT64_MAX, s);
3002 }
3003
3004 int16_t bfloat16_to_int16_scalbn(bfloat16 a, FloatRoundMode rmode, int scale,
3005 float_status *s)
3006 {
3007 FloatParts64 p;
3008
3009 bfloat16_unpack_canonical(&p, a, s);
3010 return parts_float_to_sint(&p, rmode, scale, INT16_MIN, INT16_MAX, s);
3011 }
3012
3013 int32_t bfloat16_to_int32_scalbn(bfloat16 a, FloatRoundMode rmode, int scale,
3014 float_status *s)
3015 {
3016 FloatParts64 p;
3017
3018 bfloat16_unpack_canonical(&p, a, s);
3019 return parts_float_to_sint(&p, rmode, scale, INT32_MIN, INT32_MAX, s);
3020 }
3021
3022 int64_t bfloat16_to_int64_scalbn(bfloat16 a, FloatRoundMode rmode, int scale,
3023 float_status *s)
3024 {
3025 FloatParts64 p;
3026
3027 bfloat16_unpack_canonical(&p, a, s);
3028 return parts_float_to_sint(&p, rmode, scale, INT64_MIN, INT64_MAX, s);
3029 }
3030
3031 static int32_t float128_to_int32_scalbn(float128 a, FloatRoundMode rmode,
3032 int scale, float_status *s)
3033 {
3034 FloatParts128 p;
3035
3036 float128_unpack_canonical(&p, a, s);
3037 return parts_float_to_sint(&p, rmode, scale, INT32_MIN, INT32_MAX, s);
3038 }
3039
3040 static int64_t float128_to_int64_scalbn(float128 a, FloatRoundMode rmode,
3041 int scale, float_status *s)
3042 {
3043 FloatParts128 p;
3044
3045 float128_unpack_canonical(&p, a, s);
3046 return parts_float_to_sint(&p, rmode, scale, INT64_MIN, INT64_MAX, s);
3047 }
3048
3049 static int32_t floatx80_to_int32_scalbn(floatx80 a, FloatRoundMode rmode,
3050 int scale, float_status *s)
3051 {
3052 FloatParts128 p;
3053
3054 if (!floatx80_unpack_canonical(&p, a, s)) {
3055 parts_default_nan(&p, s);
3056 }
3057 return parts_float_to_sint(&p, rmode, scale, INT32_MIN, INT32_MAX, s);
3058 }
3059
3060 static int64_t floatx80_to_int64_scalbn(floatx80 a, FloatRoundMode rmode,
3061 int scale, float_status *s)
3062 {
3063 FloatParts128 p;
3064
3065 if (!floatx80_unpack_canonical(&p, a, s)) {
3066 parts_default_nan(&p, s);
3067 }
3068 return parts_float_to_sint(&p, rmode, scale, INT64_MIN, INT64_MAX, s);
3069 }
3070
3071 int8_t float16_to_int8(float16 a, float_status *s)
3072 {
3073 return float16_to_int8_scalbn(a, s->float_rounding_mode, 0, s);
3074 }
3075
3076 int16_t float16_to_int16(float16 a, float_status *s)
3077 {
3078 return float16_to_int16_scalbn(a, s->float_rounding_mode, 0, s);
3079 }
3080
3081 int32_t float16_to_int32(float16 a, float_status *s)
3082 {
3083 return float16_to_int32_scalbn(a, s->float_rounding_mode, 0, s);
3084 }
3085
3086 int64_t float16_to_int64(float16 a, float_status *s)
3087 {
3088 return float16_to_int64_scalbn(a, s->float_rounding_mode, 0, s);
3089 }
3090
3091 int16_t float32_to_int16(float32 a, float_status *s)
3092 {
3093 return float32_to_int16_scalbn(a, s->float_rounding_mode, 0, s);
3094 }
3095
3096 int32_t float32_to_int32(float32 a, float_status *s)
3097 {
3098 return float32_to_int32_scalbn(a, s->float_rounding_mode, 0, s);
3099 }
3100
3101 int64_t float32_to_int64(float32 a, float_status *s)
3102 {
3103 return float32_to_int64_scalbn(a, s->float_rounding_mode, 0, s);
3104 }
3105
3106 int16_t float64_to_int16(float64 a, float_status *s)
3107 {
3108 return float64_to_int16_scalbn(a, s->float_rounding_mode, 0, s);
3109 }
3110
3111 int32_t float64_to_int32(float64 a, float_status *s)
3112 {
3113 return float64_to_int32_scalbn(a, s->float_rounding_mode, 0, s);
3114 }
3115
3116 int64_t float64_to_int64(float64 a, float_status *s)
3117 {
3118 return float64_to_int64_scalbn(a, s->float_rounding_mode, 0, s);
3119 }
3120
3121 int32_t float128_to_int32(float128 a, float_status *s)
3122 {
3123 return float128_to_int32_scalbn(a, s->float_rounding_mode, 0, s);
3124 }
3125
3126 int64_t float128_to_int64(float128 a, float_status *s)
3127 {
3128 return float128_to_int64_scalbn(a, s->float_rounding_mode, 0, s);
3129 }
3130
3131 int32_t floatx80_to_int32(floatx80 a, float_status *s)
3132 {
3133 return floatx80_to_int32_scalbn(a, s->float_rounding_mode, 0, s);
3134 }
3135
3136 int64_t floatx80_to_int64(floatx80 a, float_status *s)
3137 {
3138 return floatx80_to_int64_scalbn(a, s->float_rounding_mode, 0, s);
3139 }
3140
3141 int16_t float16_to_int16_round_to_zero(float16 a, float_status *s)
3142 {
3143 return float16_to_int16_scalbn(a, float_round_to_zero, 0, s);
3144 }
3145
3146 int32_t float16_to_int32_round_to_zero(float16 a, float_status *s)
3147 {
3148 return float16_to_int32_scalbn(a, float_round_to_zero, 0, s);
3149 }
3150
3151 int64_t float16_to_int64_round_to_zero(float16 a, float_status *s)
3152 {
3153 return float16_to_int64_scalbn(a, float_round_to_zero, 0, s);
3154 }
3155
3156 int16_t float32_to_int16_round_to_zero(float32 a, float_status *s)
3157 {
3158 return float32_to_int16_scalbn(a, float_round_to_zero, 0, s);
3159 }
3160
3161 int32_t float32_to_int32_round_to_zero(float32 a, float_status *s)
3162 {
3163 return float32_to_int32_scalbn(a, float_round_to_zero, 0, s);
3164 }
3165
3166 int64_t float32_to_int64_round_to_zero(float32 a, float_status *s)
3167 {
3168 return float32_to_int64_scalbn(a, float_round_to_zero, 0, s);
3169 }
3170
3171 int16_t float64_to_int16_round_to_zero(float64 a, float_status *s)
3172 {
3173 return float64_to_int16_scalbn(a, float_round_to_zero, 0, s);
3174 }
3175
3176 int32_t float64_to_int32_round_to_zero(float64 a, float_status *s)
3177 {
3178 return float64_to_int32_scalbn(a, float_round_to_zero, 0, s);
3179 }
3180
3181 int64_t float64_to_int64_round_to_zero(float64 a, float_status *s)
3182 {
3183 return float64_to_int64_scalbn(a, float_round_to_zero, 0, s);
3184 }
3185
3186 int32_t float128_to_int32_round_to_zero(float128 a, float_status *s)
3187 {
3188 return float128_to_int32_scalbn(a, float_round_to_zero, 0, s);
3189 }
3190
3191 int64_t float128_to_int64_round_to_zero(float128 a, float_status *s)
3192 {
3193 return float128_to_int64_scalbn(a, float_round_to_zero, 0, s);
3194 }
3195
3196 int32_t floatx80_to_int32_round_to_zero(floatx80 a, float_status *s)
3197 {
3198 return floatx80_to_int32_scalbn(a, float_round_to_zero, 0, s);
3199 }
3200
3201 int64_t floatx80_to_int64_round_to_zero(floatx80 a, float_status *s)
3202 {
3203 return floatx80_to_int64_scalbn(a, float_round_to_zero, 0, s);
3204 }
3205
3206 int16_t bfloat16_to_int16(bfloat16 a, float_status *s)
3207 {
3208 return bfloat16_to_int16_scalbn(a, s->float_rounding_mode, 0, s);
3209 }
3210
3211 int32_t bfloat16_to_int32(bfloat16 a, float_status *s)
3212 {
3213 return bfloat16_to_int32_scalbn(a, s->float_rounding_mode, 0, s);
3214 }
3215
3216 int64_t bfloat16_to_int64(bfloat16 a, float_status *s)
3217 {
3218 return bfloat16_to_int64_scalbn(a, s->float_rounding_mode, 0, s);
3219 }
3220
3221 int16_t bfloat16_to_int16_round_to_zero(bfloat16 a, float_status *s)
3222 {
3223 return bfloat16_to_int16_scalbn(a, float_round_to_zero, 0, s);
3224 }
3225
3226 int32_t bfloat16_to_int32_round_to_zero(bfloat16 a, float_status *s)
3227 {
3228 return bfloat16_to_int32_scalbn(a, float_round_to_zero, 0, s);
3229 }
3230
3231 int64_t bfloat16_to_int64_round_to_zero(bfloat16 a, float_status *s)
3232 {
3233 return bfloat16_to_int64_scalbn(a, float_round_to_zero, 0, s);
3234 }
3235
3236 /*
3237 * Floating-point to unsigned integer conversions
3238 */
3239
3240 uint8_t float16_to_uint8_scalbn(float16 a, FloatRoundMode rmode, int scale,
3241 float_status *s)
3242 {
3243 FloatParts64 p;
3244
3245 float16_unpack_canonical(&p, a, s);
3246 return parts_float_to_uint(&p, rmode, scale, UINT8_MAX, s);
3247 }
3248
3249 uint16_t float16_to_uint16_scalbn(float16 a, FloatRoundMode rmode, int scale,
3250 float_status *s)
3251 {
3252 FloatParts64 p;
3253
3254 float16_unpack_canonical(&p, a, s);
3255 return parts_float_to_uint(&p, rmode, scale, UINT16_MAX, s);
3256 }
3257
3258 uint32_t float16_to_uint32_scalbn(float16 a, FloatRoundMode rmode, int scale,
3259 float_status *s)
3260 {
3261 FloatParts64 p;
3262
3263 float16_unpack_canonical(&p, a, s);
3264 return parts_float_to_uint(&p, rmode, scale, UINT32_MAX, s);
3265 }
3266
3267 uint64_t float16_to_uint64_scalbn(float16 a, FloatRoundMode rmode, int scale,
3268 float_status *s)
3269 {
3270 FloatParts64 p;
3271
3272 float16_unpack_canonical(&p, a, s);
3273 return parts_float_to_uint(&p, rmode, scale, UINT64_MAX, s);
3274 }
3275
3276 uint16_t float32_to_uint16_scalbn(float32 a, FloatRoundMode rmode, int scale,
3277 float_status *s)
3278 {
3279 FloatParts64 p;
3280
3281 float32_unpack_canonical(&p, a, s);
3282 return parts_float_to_uint(&p, rmode, scale, UINT16_MAX, s);
3283 }
3284
3285 uint32_t float32_to_uint32_scalbn(float32 a, FloatRoundMode rmode, int scale,
3286 float_status *s)
3287 {
3288 FloatParts64 p;
3289
3290 float32_unpack_canonical(&p, a, s);
3291 return parts_float_to_uint(&p, rmode, scale, UINT32_MAX, s);
3292 }
3293
3294 uint64_t float32_to_uint64_scalbn(float32 a, FloatRoundMode rmode, int scale,
3295 float_status *s)
3296 {
3297 FloatParts64 p;
3298
3299 float32_unpack_canonical(&p, a, s);
3300 return parts_float_to_uint(&p, rmode, scale, UINT64_MAX, s);
3301 }
3302
3303 uint16_t float64_to_uint16_scalbn(float64 a, FloatRoundMode rmode, int scale,
3304 float_status *s)
3305 {
3306 FloatParts64 p;
3307
3308 float64_unpack_canonical(&p, a, s);
3309 return parts_float_to_uint(&p, rmode, scale, UINT16_MAX, s);
3310 }
3311
3312 uint32_t float64_to_uint32_scalbn(float64 a, FloatRoundMode rmode, int scale,
3313 float_status *s)
3314 {
3315 FloatParts64 p;
3316
3317 float64_unpack_canonical(&p, a, s);
3318 return parts_float_to_uint(&p, rmode, scale, UINT32_MAX, s);
3319 }
3320
3321 uint64_t float64_to_uint64_scalbn(float64 a, FloatRoundMode rmode, int scale,
3322 float_status *s)
3323 {
3324 FloatParts64 p;
3325
3326 float64_unpack_canonical(&p, a, s);
3327 return parts_float_to_uint(&p, rmode, scale, UINT64_MAX, s);
3328 }
3329
3330 uint16_t bfloat16_to_uint16_scalbn(bfloat16 a, FloatRoundMode rmode,
3331 int scale, float_status *s)
3332 {
3333 FloatParts64 p;
3334
3335 bfloat16_unpack_canonical(&p, a, s);
3336 return parts_float_to_uint(&p, rmode, scale, UINT16_MAX, s);
3337 }
3338
3339 uint32_t bfloat16_to_uint32_scalbn(bfloat16 a, FloatRoundMode rmode,
3340 int scale, float_status *s)
3341 {
3342 FloatParts64 p;
3343
3344 bfloat16_unpack_canonical(&p, a, s);
3345 return parts_float_to_uint(&p, rmode, scale, UINT32_MAX, s);
3346 }
3347
3348 uint64_t bfloat16_to_uint64_scalbn(bfloat16 a, FloatRoundMode rmode,
3349 int scale, float_status *s)
3350 {
3351 FloatParts64 p;
3352
3353 bfloat16_unpack_canonical(&p, a, s);
3354 return parts_float_to_uint(&p, rmode, scale, UINT64_MAX, s);
3355 }
3356
3357 static uint32_t float128_to_uint32_scalbn(float128 a, FloatRoundMode rmode,
3358 int scale, float_status *s)
3359 {
3360 FloatParts128 p;
3361
3362 float128_unpack_canonical(&p, a, s);
3363 return parts_float_to_uint(&p, rmode, scale, UINT32_MAX, s);
3364 }
3365
3366 static uint64_t float128_to_uint64_scalbn(float128 a, FloatRoundMode rmode,
3367 int scale, float_status *s)
3368 {
3369 FloatParts128 p;
3370
3371 float128_unpack_canonical(&p, a, s);
3372 return parts_float_to_uint(&p, rmode, scale, UINT64_MAX, s);
3373 }
3374
3375 uint8_t float16_to_uint8(float16 a, float_status *s)
3376 {
3377 return float16_to_uint8_scalbn(a, s->float_rounding_mode, 0, s);
3378 }
3379
3380 uint16_t float16_to_uint16(float16 a, float_status *s)
3381 {
3382 return float16_to_uint16_scalbn(a, s->float_rounding_mode, 0, s);
3383 }
3384
3385 uint32_t float16_to_uint32(float16 a, float_status *s)
3386 {
3387 return float16_to_uint32_scalbn(a, s->float_rounding_mode, 0, s);
3388 }
3389
3390 uint64_t float16_to_uint64(float16 a, float_status *s)
3391 {
3392 return float16_to_uint64_scalbn(a, s->float_rounding_mode, 0, s);
3393 }
3394
3395 uint16_t float32_to_uint16(float32 a, float_status *s)
3396 {
3397 return float32_to_uint16_scalbn(a, s->float_rounding_mode, 0, s);
3398 }
3399
3400 uint32_t float32_to_uint32(float32 a, float_status *s)
3401 {
3402 return float32_to_uint32_scalbn(a, s->float_rounding_mode, 0, s);
3403 }
3404
3405 uint64_t float32_to_uint64(float32 a, float_status *s)
3406 {
3407 return float32_to_uint64_scalbn(a, s->float_rounding_mode, 0, s);
3408 }
3409
3410 uint16_t float64_to_uint16(float64 a, float_status *s)
3411 {
3412 return float64_to_uint16_scalbn(a, s->float_rounding_mode, 0, s);
3413 }
3414
3415 uint32_t float64_to_uint32(float64 a, float_status *s)
3416 {
3417 return float64_to_uint32_scalbn(a, s->float_rounding_mode, 0, s);
3418 }
3419
3420 uint64_t float64_to_uint64(float64 a, float_status *s)
3421 {
3422 return float64_to_uint64_scalbn(a, s->float_rounding_mode, 0, s);
3423 }
3424
3425 uint32_t float128_to_uint32(float128 a, float_status *s)
3426 {
3427 return float128_to_uint32_scalbn(a, s->float_rounding_mode, 0, s);
3428 }
3429
3430 uint64_t float128_to_uint64(float128 a, float_status *s)
3431 {
3432 return float128_to_uint64_scalbn(a, s->float_rounding_mode, 0, s);
3433 }
3434
3435 uint16_t float16_to_uint16_round_to_zero(float16 a, float_status *s)
3436 {
3437 return float16_to_uint16_scalbn(a, float_round_to_zero, 0, s);
3438 }
3439
3440 uint32_t float16_to_uint32_round_to_zero(float16 a, float_status *s)
3441 {
3442 return float16_to_uint32_scalbn(a, float_round_to_zero, 0, s);
3443 }
3444
3445 uint64_t float16_to_uint64_round_to_zero(float16 a, float_status *s)
3446 {
3447 return float16_to_uint64_scalbn(a, float_round_to_zero, 0, s);
3448 }
3449
3450 uint16_t float32_to_uint16_round_to_zero(float32 a, float_status *s)
3451 {
3452 return float32_to_uint16_scalbn(a, float_round_to_zero, 0, s);
3453 }
3454
3455 uint32_t float32_to_uint32_round_to_zero(float32 a, float_status *s)
3456 {
3457 return float32_to_uint32_scalbn(a, float_round_to_zero, 0, s);
3458 }
3459
3460 uint64_t float32_to_uint64_round_to_zero(float32 a, float_status *s)
3461 {
3462 return float32_to_uint64_scalbn(a, float_round_to_zero, 0, s);
3463 }
3464
3465 uint16_t float64_to_uint16_round_to_zero(float64 a, float_status *s)
3466 {
3467 return float64_to_uint16_scalbn(a, float_round_to_zero, 0, s);
3468 }
3469
3470 uint32_t float64_to_uint32_round_to_zero(float64 a, float_status *s)
3471 {
3472 return float64_to_uint32_scalbn(a, float_round_to_zero, 0, s);
3473 }
3474
3475 uint64_t float64_to_uint64_round_to_zero(float64 a, float_status *s)
3476 {
3477 return float64_to_uint64_scalbn(a, float_round_to_zero, 0, s);
3478 }
3479
3480 uint32_t float128_to_uint32_round_to_zero(float128 a, float_status *s)
3481 {
3482 return float128_to_uint32_scalbn(a, float_round_to_zero, 0, s);
3483 }
3484
3485 uint64_t float128_to_uint64_round_to_zero(float128 a, float_status *s)
3486 {
3487 return float128_to_uint64_scalbn(a, float_round_to_zero, 0, s);
3488 }
3489
3490 uint16_t bfloat16_to_uint16(bfloat16 a, float_status *s)
3491 {
3492 return bfloat16_to_uint16_scalbn(a, s->float_rounding_mode, 0, s);
3493 }
3494
3495 uint32_t bfloat16_to_uint32(bfloat16 a, float_status *s)
3496 {
3497 return bfloat16_to_uint32_scalbn(a, s->float_rounding_mode, 0, s);
3498 }
3499
3500 uint64_t bfloat16_to_uint64(bfloat16 a, float_status *s)
3501 {
3502 return bfloat16_to_uint64_scalbn(a, s->float_rounding_mode, 0, s);
3503 }
3504
3505 uint16_t bfloat16_to_uint16_round_to_zero(bfloat16 a, float_status *s)
3506 {
3507 return bfloat16_to_uint16_scalbn(a, float_round_to_zero, 0, s);
3508 }
3509
3510 uint32_t bfloat16_to_uint32_round_to_zero(bfloat16 a, float_status *s)
3511 {
3512 return bfloat16_to_uint32_scalbn(a, float_round_to_zero, 0, s);
3513 }
3514
3515 uint64_t bfloat16_to_uint64_round_to_zero(bfloat16 a, float_status *s)
3516 {
3517 return bfloat16_to_uint64_scalbn(a, float_round_to_zero, 0, s);
3518 }
3519
3520 /*
3521 * Signed integer to floating-point conversions
3522 */
3523
3524 float16 int64_to_float16_scalbn(int64_t a, int scale, float_status *status)
3525 {
3526 FloatParts64 p;
3527
3528 parts_sint_to_float(&p, a, scale, status);
3529 return float16_round_pack_canonical(&p, status);
3530 }
3531
3532 float16 int32_to_float16_scalbn(int32_t a, int scale, float_status *status)
3533 {
3534 return int64_to_float16_scalbn(a, scale, status);
3535 }
3536
3537 float16 int16_to_float16_scalbn(int16_t a, int scale, float_status *status)
3538 {
3539 return int64_to_float16_scalbn(a, scale, status);
3540 }
3541
3542 float16 int64_to_float16(int64_t a, float_status *status)
3543 {
3544 return int64_to_float16_scalbn(a, 0, status);
3545 }
3546
3547 float16 int32_to_float16(int32_t a, float_status *status)
3548 {
3549 return int64_to_float16_scalbn(a, 0, status);
3550 }
3551
3552 float16 int16_to_float16(int16_t a, float_status *status)
3553 {
3554 return int64_to_float16_scalbn(a, 0, status);
3555 }
3556
3557 float16 int8_to_float16(int8_t a, float_status *status)
3558 {
3559 return int64_to_float16_scalbn(a, 0, status);
3560 }
3561
3562 float32 int64_to_float32_scalbn(int64_t a, int scale, float_status *status)
3563 {
3564 FloatParts64 p;
3565
3566 /* Without scaling, there are no overflow concerns. */
3567 if (likely(scale == 0) && can_use_fpu(status)) {
3568 union_float32 ur;
3569 ur.h = a;
3570 return ur.s;
3571 }
3572
3573 parts64_sint_to_float(&p, a, scale, status);
3574 return float32_round_pack_canonical(&p, status);
3575 }
3576
3577 float32 int32_to_float32_scalbn(int32_t a, int scale, float_status *status)
3578 {
3579 return int64_to_float32_scalbn(a, scale, status);
3580 }
3581
3582 float32 int16_to_float32_scalbn(int16_t a, int scale, float_status *status)
3583 {
3584 return int64_to_float32_scalbn(a, scale, status);
3585 }
3586
3587 float32 int64_to_float32(int64_t a, float_status *status)
3588 {
3589 return int64_to_float32_scalbn(a, 0, status);
3590 }
3591
3592 float32 int32_to_float32(int32_t a, float_status *status)
3593 {
3594 return int64_to_float32_scalbn(a, 0, status);
3595 }
3596
3597 float32 int16_to_float32(int16_t a, float_status *status)
3598 {
3599 return int64_to_float32_scalbn(a, 0, status);
3600 }
3601
3602 float64 int64_to_float64_scalbn(int64_t a, int scale, float_status *status)
3603 {
3604 FloatParts64 p;
3605
3606 /* Without scaling, there are no overflow concerns. */
3607 if (likely(scale == 0) && can_use_fpu(status)) {
3608 union_float64 ur;
3609 ur.h = a;
3610 return ur.s;
3611 }
3612
3613 parts_sint_to_float(&p, a, scale, status);
3614 return float64_round_pack_canonical(&p, status);
3615 }
3616
3617 float64 int32_to_float64_scalbn(int32_t a, int scale, float_status *status)
3618 {
3619 return int64_to_float64_scalbn(a, scale, status);
3620 }
3621
3622 float64 int16_to_float64_scalbn(int16_t a, int scale, float_status *status)
3623 {
3624 return int64_to_float64_scalbn(a, scale, status);
3625 }
3626
3627 float64 int64_to_float64(int64_t a, float_status *status)
3628 {
3629 return int64_to_float64_scalbn(a, 0, status);
3630 }
3631
3632 float64 int32_to_float64(int32_t a, float_status *status)
3633 {
3634 return int64_to_float64_scalbn(a, 0, status);
3635 }
3636
3637 float64 int16_to_float64(int16_t a, float_status *status)
3638 {
3639 return int64_to_float64_scalbn(a, 0, status);
3640 }
3641
3642 bfloat16 int64_to_bfloat16_scalbn(int64_t a, int scale, float_status *status)
3643 {
3644 FloatParts64 p;
3645
3646 parts_sint_to_float(&p, a, scale, status);
3647 return bfloat16_round_pack_canonical(&p, status);
3648 }
3649
3650 bfloat16 int32_to_bfloat16_scalbn(int32_t a, int scale, float_status *status)
3651 {
3652 return int64_to_bfloat16_scalbn(a, scale, status);
3653 }
3654
3655 bfloat16 int16_to_bfloat16_scalbn(int16_t a, int scale, float_status *status)
3656 {
3657 return int64_to_bfloat16_scalbn(a, scale, status);
3658 }
3659
3660 bfloat16 int64_to_bfloat16(int64_t a, float_status *status)
3661 {
3662 return int64_to_bfloat16_scalbn(a, 0, status);
3663 }
3664
3665 bfloat16 int32_to_bfloat16(int32_t a, float_status *status)
3666 {
3667 return int64_to_bfloat16_scalbn(a, 0, status);
3668 }
3669
3670 bfloat16 int16_to_bfloat16(int16_t a, float_status *status)
3671 {
3672 return int64_to_bfloat16_scalbn(a, 0, status);
3673 }
3674
3675 float128 int64_to_float128(int64_t a, float_status *status)
3676 {
3677 FloatParts128 p;
3678
3679 parts_sint_to_float(&p, a, 0, status);
3680 return float128_round_pack_canonical(&p, status);
3681 }
3682
3683 float128 int32_to_float128(int32_t a, float_status *status)
3684 {
3685 return int64_to_float128(a, status);
3686 }
3687
3688 floatx80 int64_to_floatx80(int64_t a, float_status *status)
3689 {
3690 FloatParts128 p;
3691
3692 parts_sint_to_float(&p, a, 0, status);
3693 return floatx80_round_pack_canonical(&p, status);
3694 }
3695
3696 floatx80 int32_to_floatx80(int32_t a, float_status *status)
3697 {
3698 return int64_to_floatx80(a, status);
3699 }
3700
3701 /*
3702 * Unsigned Integer to floating-point conversions
3703 */
3704
3705 float16 uint64_to_float16_scalbn(uint64_t a, int scale, float_status *status)
3706 {
3707 FloatParts64 p;
3708
3709 parts_uint_to_float(&p, a, scale, status);
3710 return float16_round_pack_canonical(&p, status);
3711 }
3712
3713 float16 uint32_to_float16_scalbn(uint32_t a, int scale, float_status *status)
3714 {
3715 return uint64_to_float16_scalbn(a, scale, status);
3716 }
3717
3718 float16 uint16_to_float16_scalbn(uint16_t a, int scale, float_status *status)
3719 {
3720 return uint64_to_float16_scalbn(a, scale, status);
3721 }
3722
3723 float16 uint64_to_float16(uint64_t a, float_status *status)
3724 {
3725 return uint64_to_float16_scalbn(a, 0, status);
3726 }
3727
3728 float16 uint32_to_float16(uint32_t a, float_status *status)
3729 {
3730 return uint64_to_float16_scalbn(a, 0, status);
3731 }
3732
3733 float16 uint16_to_float16(uint16_t a, float_status *status)
3734 {
3735 return uint64_to_float16_scalbn(a, 0, status);
3736 }
3737
3738 float16 uint8_to_float16(uint8_t a, float_status *status)
3739 {
3740 return uint64_to_float16_scalbn(a, 0, status);
3741 }
3742
3743 float32 uint64_to_float32_scalbn(uint64_t a, int scale, float_status *status)
3744 {
3745 FloatParts64 p;
3746
3747 /* Without scaling, there are no overflow concerns. */
3748 if (likely(scale == 0) && can_use_fpu(status)) {
3749 union_float32 ur;
3750 ur.h = a;
3751 return ur.s;
3752 }
3753
3754 parts_uint_to_float(&p, a, scale, status);
3755 return float32_round_pack_canonical(&p, status);
3756 }
3757
3758 float32 uint32_to_float32_scalbn(uint32_t a, int scale, float_status *status)
3759 {
3760 return uint64_to_float32_scalbn(a, scale, status);
3761 }
3762
3763 float32 uint16_to_float32_scalbn(uint16_t a, int scale, float_status *status)
3764 {
3765 return uint64_to_float32_scalbn(a, scale, status);
3766 }
3767
3768 float32 uint64_to_float32(uint64_t a, float_status *status)
3769 {
3770 return uint64_to_float32_scalbn(a, 0, status);
3771 }
3772
3773 float32 uint32_to_float32(uint32_t a, float_status *status)
3774 {
3775 return uint64_to_float32_scalbn(a, 0, status);
3776 }
3777
3778 float32 uint16_to_float32(uint16_t a, float_status *status)
3779 {
3780 return uint64_to_float32_scalbn(a, 0, status);
3781 }
3782
3783 float64 uint64_to_float64_scalbn(uint64_t a, int scale, float_status *status)
3784 {
3785 FloatParts64 p;
3786
3787 /* Without scaling, there are no overflow concerns. */
3788 if (likely(scale == 0) && can_use_fpu(status)) {
3789 union_float64 ur;
3790 ur.h = a;
3791 return ur.s;
3792 }
3793
3794 parts_uint_to_float(&p, a, scale, status);
3795 return float64_round_pack_canonical(&p, status);
3796 }
3797
3798 float64 uint32_to_float64_scalbn(uint32_t a, int scale, float_status *status)
3799 {
3800 return uint64_to_float64_scalbn(a, scale, status);
3801 }
3802
3803 float64 uint16_to_float64_scalbn(uint16_t a, int scale, float_status *status)
3804 {
3805 return uint64_to_float64_scalbn(a, scale, status);
3806 }
3807
3808 float64 uint64_to_float64(uint64_t a, float_status *status)
3809 {
3810 return uint64_to_float64_scalbn(a, 0, status);
3811 }
3812
3813 float64 uint32_to_float64(uint32_t a, float_status *status)
3814 {
3815 return uint64_to_float64_scalbn(a, 0, status);
3816 }
3817
3818 float64 uint16_to_float64(uint16_t a, float_status *status)
3819 {
3820 return uint64_to_float64_scalbn(a, 0, status);
3821 }
3822
3823 bfloat16 uint64_to_bfloat16_scalbn(uint64_t a, int scale, float_status *status)
3824 {
3825 FloatParts64 p;
3826
3827 parts_uint_to_float(&p, a, scale, status);
3828 return bfloat16_round_pack_canonical(&p, status);
3829 }
3830
3831 bfloat16 uint32_to_bfloat16_scalbn(uint32_t a, int scale, float_status *status)
3832 {
3833 return uint64_to_bfloat16_scalbn(a, scale, status);
3834 }
3835
3836 bfloat16 uint16_to_bfloat16_scalbn(uint16_t a, int scale, float_status *status)
3837 {
3838 return uint64_to_bfloat16_scalbn(a, scale, status);
3839 }
3840
3841 bfloat16 uint64_to_bfloat16(uint64_t a, float_status *status)
3842 {
3843 return uint64_to_bfloat16_scalbn(a, 0, status);
3844 }
3845
3846 bfloat16 uint32_to_bfloat16(uint32_t a, float_status *status)
3847 {
3848 return uint64_to_bfloat16_scalbn(a, 0, status);
3849 }
3850
3851 bfloat16 uint16_to_bfloat16(uint16_t a, float_status *status)
3852 {
3853 return uint64_to_bfloat16_scalbn(a, 0, status);
3854 }
3855
3856 float128 uint64_to_float128(uint64_t a, float_status *status)
3857 {
3858 FloatParts128 p;
3859
3860 parts_uint_to_float(&p, a, 0, status);
3861 return float128_round_pack_canonical(&p, status);
3862 }
3863
3864 /*
3865 * Minimum and maximum
3866 */
3867
3868 static float16 float16_minmax(float16 a, float16 b, float_status *s, int flags)
3869 {
3870 FloatParts64 pa, pb, *pr;
3871
3872 float16_unpack_canonical(&pa, a, s);
3873 float16_unpack_canonical(&pb, b, s);
3874 pr = parts_minmax(&pa, &pb, s, flags);
3875
3876 return float16_round_pack_canonical(pr, s);
3877 }
3878
3879 static bfloat16 bfloat16_minmax(bfloat16 a, bfloat16 b,
3880 float_status *s, int flags)
3881 {
3882 FloatParts64 pa, pb, *pr;
3883
3884 bfloat16_unpack_canonical(&pa, a, s);
3885 bfloat16_unpack_canonical(&pb, b, s);
3886 pr = parts_minmax(&pa, &pb, s, flags);
3887
3888 return bfloat16_round_pack_canonical(pr, s);
3889 }
3890
3891 static float32 float32_minmax(float32 a, float32 b, float_status *s, int flags)
3892 {
3893 FloatParts64 pa, pb, *pr;
3894
3895 float32_unpack_canonical(&pa, a, s);
3896 float32_unpack_canonical(&pb, b, s);
3897 pr = parts_minmax(&pa, &pb, s, flags);
3898
3899 return float32_round_pack_canonical(pr, s);
3900 }
3901
3902 static float64 float64_minmax(float64 a, float64 b, float_status *s, int flags)
3903 {
3904 FloatParts64 pa, pb, *pr;
3905
3906 float64_unpack_canonical(&pa, a, s);
3907 float64_unpack_canonical(&pb, b, s);
3908 pr = parts_minmax(&pa, &pb, s, flags);
3909
3910 return float64_round_pack_canonical(pr, s);
3911 }
3912
3913 static float128 float128_minmax(float128 a, float128 b,
3914 float_status *s, int flags)
3915 {
3916 FloatParts128 pa, pb, *pr;
3917
3918 float128_unpack_canonical(&pa, a, s);
3919 float128_unpack_canonical(&pb, b, s);
3920 pr = parts_minmax(&pa, &pb, s, flags);
3921
3922 return float128_round_pack_canonical(pr, s);
3923 }
3924
3925 #define MINMAX_1(type, name, flags) \
3926 type type##_##name(type a, type b, float_status *s) \
3927 { return type##_minmax(a, b, s, flags); }
3928
3929 #define MINMAX_2(type) \
3930 MINMAX_1(type, max, 0) \
3931 MINMAX_1(type, maxnum, minmax_isnum) \
3932 MINMAX_1(type, maxnummag, minmax_isnum | minmax_ismag) \
3933 MINMAX_1(type, min, minmax_ismin) \
3934 MINMAX_1(type, minnum, minmax_ismin | minmax_isnum) \
3935 MINMAX_1(type, minnummag, minmax_ismin | minmax_isnum | minmax_ismag)
3936
3937 MINMAX_2(float16)
3938 MINMAX_2(bfloat16)
3939 MINMAX_2(float32)
3940 MINMAX_2(float64)
3941 MINMAX_2(float128)
3942
3943 #undef MINMAX_1
3944 #undef MINMAX_2
3945
3946 /*
3947 * Floating point compare
3948 */
3949
3950 static FloatRelation QEMU_FLATTEN
3951 float16_do_compare(float16 a, float16 b, float_status *s, bool is_quiet)
3952 {
3953 FloatParts64 pa, pb;
3954
3955 float16_unpack_canonical(&pa, a, s);
3956 float16_unpack_canonical(&pb, b, s);
3957 return parts_compare(&pa, &pb, s, is_quiet);
3958 }
3959
3960 FloatRelation float16_compare(float16 a, float16 b, float_status *s)
3961 {
3962 return float16_do_compare(a, b, s, false);
3963 }
3964
3965 FloatRelation float16_compare_quiet(float16 a, float16 b, float_status *s)
3966 {
3967 return float16_do_compare(a, b, s, true);
3968 }
3969
3970 static FloatRelation QEMU_SOFTFLOAT_ATTR
3971 float32_do_compare(float32 a, float32 b, float_status *s, bool is_quiet)
3972 {
3973 FloatParts64 pa, pb;
3974
3975 float32_unpack_canonical(&pa, a, s);
3976 float32_unpack_canonical(&pb, b, s);
3977 return parts_compare(&pa, &pb, s, is_quiet);
3978 }
3979
3980 static FloatRelation QEMU_FLATTEN
3981 float32_hs_compare(float32 xa, float32 xb, float_status *s, bool is_quiet)
3982 {
3983 union_float32 ua, ub;
3984
3985 ua.s = xa;
3986 ub.s = xb;
3987
3988 if (QEMU_NO_HARDFLOAT) {
3989 goto soft;
3990 }
3991
3992 float32_input_flush2(&ua.s, &ub.s, s);
3993 if (isgreaterequal(ua.h, ub.h)) {
3994 if (isgreater(ua.h, ub.h)) {
3995 return float_relation_greater;
3996 }
3997 return float_relation_equal;
3998 }
3999 if (likely(isless(ua.h, ub.h))) {
4000 return float_relation_less;
4001 }
4002 /*
4003 * The only condition remaining is unordered.
4004 * Fall through to set flags.
4005 */
4006 soft:
4007 return float32_do_compare(ua.s, ub.s, s, is_quiet);
4008 }
4009
4010 FloatRelation float32_compare(float32 a, float32 b, float_status *s)
4011 {
4012 return float32_hs_compare(a, b, s, false);
4013 }
4014
4015 FloatRelation float32_compare_quiet(float32 a, float32 b, float_status *s)
4016 {
4017 return float32_hs_compare(a, b, s, true);
4018 }
4019
4020 static FloatRelation QEMU_SOFTFLOAT_ATTR
4021 float64_do_compare(float64 a, float64 b, float_status *s, bool is_quiet)
4022 {
4023 FloatParts64 pa, pb;
4024
4025 float64_unpack_canonical(&pa, a, s);
4026 float64_unpack_canonical(&pb, b, s);
4027 return parts_compare(&pa, &pb, s, is_quiet);
4028 }
4029
4030 static FloatRelation QEMU_FLATTEN
4031 float64_hs_compare(float64 xa, float64 xb, float_status *s, bool is_quiet)
4032 {
4033 union_float64 ua, ub;
4034
4035 ua.s = xa;
4036 ub.s = xb;
4037
4038 if (QEMU_NO_HARDFLOAT) {
4039 goto soft;
4040 }
4041
4042 float64_input_flush2(&ua.s, &ub.s, s);
4043 if (isgreaterequal(ua.h, ub.h)) {
4044 if (isgreater(ua.h, ub.h)) {
4045 return float_relation_greater;
4046 }
4047 return float_relation_equal;
4048 }
4049 if (likely(isless(ua.h, ub.h))) {
4050 return float_relation_less;
4051 }
4052 /*
4053 * The only condition remaining is unordered.
4054 * Fall through to set flags.
4055 */
4056 soft:
4057 return float64_do_compare(ua.s, ub.s, s, is_quiet);
4058 }
4059
4060 FloatRelation float64_compare(float64 a, float64 b, float_status *s)
4061 {
4062 return float64_hs_compare(a, b, s, false);
4063 }
4064
4065 FloatRelation float64_compare_quiet(float64 a, float64 b, float_status *s)
4066 {
4067 return float64_hs_compare(a, b, s, true);
4068 }
4069
4070 static FloatRelation QEMU_FLATTEN
4071 bfloat16_do_compare(bfloat16 a, bfloat16 b, float_status *s, bool is_quiet)
4072 {
4073 FloatParts64 pa, pb;
4074
4075 bfloat16_unpack_canonical(&pa, a, s);
4076 bfloat16_unpack_canonical(&pb, b, s);
4077 return parts_compare(&pa, &pb, s, is_quiet);
4078 }
4079
4080 FloatRelation bfloat16_compare(bfloat16 a, bfloat16 b, float_status *s)
4081 {
4082 return bfloat16_do_compare(a, b, s, false);
4083 }
4084
4085 FloatRelation bfloat16_compare_quiet(bfloat16 a, bfloat16 b, float_status *s)
4086 {
4087 return bfloat16_do_compare(a, b, s, true);
4088 }
4089
4090 static FloatRelation QEMU_FLATTEN
4091 float128_do_compare(float128 a, float128 b, float_status *s, bool is_quiet)
4092 {
4093 FloatParts128 pa, pb;
4094
4095 float128_unpack_canonical(&pa, a, s);
4096 float128_unpack_canonical(&pb, b, s);
4097 return parts_compare(&pa, &pb, s, is_quiet);
4098 }
4099
4100 FloatRelation float128_compare(float128 a, float128 b, float_status *s)
4101 {
4102 return float128_do_compare(a, b, s, false);
4103 }
4104
4105 FloatRelation float128_compare_quiet(float128 a, float128 b, float_status *s)
4106 {
4107 return float128_do_compare(a, b, s, true);
4108 }
4109
4110 static FloatRelation QEMU_FLATTEN
4111 floatx80_do_compare(floatx80 a, floatx80 b, float_status *s, bool is_quiet)
4112 {
4113 FloatParts128 pa, pb;
4114
4115 if (!floatx80_unpack_canonical(&pa, a, s) ||
4116 !floatx80_unpack_canonical(&pb, b, s)) {
4117 return float_relation_unordered;
4118 }
4119 return parts_compare(&pa, &pb, s, is_quiet);
4120 }
4121
4122 FloatRelation floatx80_compare(floatx80 a, floatx80 b, float_status *s)
4123 {
4124 return floatx80_do_compare(a, b, s, false);
4125 }
4126
4127 FloatRelation floatx80_compare_quiet(floatx80 a, floatx80 b, float_status *s)
4128 {
4129 return floatx80_do_compare(a, b, s, true);
4130 }
4131
4132 /*
4133 * Scale by 2**N
4134 */
4135
4136 float16 float16_scalbn(float16 a, int n, float_status *status)
4137 {
4138 FloatParts64 p;
4139
4140 float16_unpack_canonical(&p, a, status);
4141 parts_scalbn(&p, n, status);
4142 return float16_round_pack_canonical(&p, status);
4143 }
4144
4145 float32 float32_scalbn(float32 a, int n, float_status *status)
4146 {
4147 FloatParts64 p;
4148
4149 float32_unpack_canonical(&p, a, status);
4150 parts_scalbn(&p, n, status);
4151 return float32_round_pack_canonical(&p, status);
4152 }
4153
4154 float64 float64_scalbn(float64 a, int n, float_status *status)
4155 {
4156 FloatParts64 p;
4157
4158 float64_unpack_canonical(&p, a, status);
4159 parts_scalbn(&p, n, status);
4160 return float64_round_pack_canonical(&p, status);
4161 }
4162
4163 bfloat16 bfloat16_scalbn(bfloat16 a, int n, float_status *status)
4164 {
4165 FloatParts64 p;
4166
4167 bfloat16_unpack_canonical(&p, a, status);
4168 parts_scalbn(&p, n, status);
4169 return bfloat16_round_pack_canonical(&p, status);
4170 }
4171
4172 float128 float128_scalbn(float128 a, int n, float_status *status)
4173 {
4174 FloatParts128 p;
4175
4176 float128_unpack_canonical(&p, a, status);
4177 parts_scalbn(&p, n, status);
4178 return float128_round_pack_canonical(&p, status);
4179 }
4180
4181 floatx80 floatx80_scalbn(floatx80 a, int n, float_status *status)
4182 {
4183 FloatParts128 p;
4184
4185 if (!floatx80_unpack_canonical(&p, a, status)) {
4186 return floatx80_default_nan(status);
4187 }
4188 parts_scalbn(&p, n, status);
4189 return floatx80_round_pack_canonical(&p, status);
4190 }
4191
4192 /*
4193 * Square Root
4194 */
4195
4196 float16 QEMU_FLATTEN float16_sqrt(float16 a, float_status *status)
4197 {
4198 FloatParts64 p;
4199
4200 float16_unpack_canonical(&p, a, status);
4201 parts_sqrt(&p, status, &float16_params);
4202 return float16_round_pack_canonical(&p, status);
4203 }
4204
4205 static float32 QEMU_SOFTFLOAT_ATTR
4206 soft_f32_sqrt(float32 a, float_status *status)
4207 {
4208 FloatParts64 p;
4209
4210 float32_unpack_canonical(&p, a, status);
4211 parts_sqrt(&p, status, &float32_params);
4212 return float32_round_pack_canonical(&p, status);
4213 }
4214
4215 static float64 QEMU_SOFTFLOAT_ATTR
4216 soft_f64_sqrt(float64 a, float_status *status)
4217 {
4218 FloatParts64 p;
4219
4220 float64_unpack_canonical(&p, a, status);
4221 parts_sqrt(&p, status, &float64_params);
4222 return float64_round_pack_canonical(&p, status);
4223 }
4224
4225 float32 QEMU_FLATTEN float32_sqrt(float32 xa, float_status *s)
4226 {
4227 union_float32 ua, ur;
4228
4229 ua.s = xa;
4230 if (unlikely(!can_use_fpu(s))) {
4231 goto soft;
4232 }
4233
4234 float32_input_flush1(&ua.s, s);
4235 if (QEMU_HARDFLOAT_1F32_USE_FP) {
4236 if (unlikely(!(fpclassify(ua.h) == FP_NORMAL ||
4237 fpclassify(ua.h) == FP_ZERO) ||
4238 signbit(ua.h))) {
4239 goto soft;
4240 }
4241 } else if (unlikely(!float32_is_zero_or_normal(ua.s) ||
4242 float32_is_neg(ua.s))) {
4243 goto soft;
4244 }
4245 ur.h = sqrtf(ua.h);
4246 return ur.s;
4247
4248 soft:
4249 return soft_f32_sqrt(ua.s, s);
4250 }
4251
4252 float64 QEMU_FLATTEN float64_sqrt(float64 xa, float_status *s)
4253 {
4254 union_float64 ua, ur;
4255
4256 ua.s = xa;
4257 if (unlikely(!can_use_fpu(s))) {
4258 goto soft;
4259 }
4260
4261 float64_input_flush1(&ua.s, s);
4262 if (QEMU_HARDFLOAT_1F64_USE_FP) {
4263 if (unlikely(!(fpclassify(ua.h) == FP_NORMAL ||
4264 fpclassify(ua.h) == FP_ZERO) ||
4265 signbit(ua.h))) {
4266 goto soft;
4267 }
4268 } else if (unlikely(!float64_is_zero_or_normal(ua.s) ||
4269 float64_is_neg(ua.s))) {
4270 goto soft;
4271 }
4272 ur.h = sqrt(ua.h);
4273 return ur.s;
4274
4275 soft:
4276 return soft_f64_sqrt(ua.s, s);
4277 }
4278
4279 bfloat16 QEMU_FLATTEN bfloat16_sqrt(bfloat16 a, float_status *status)
4280 {
4281 FloatParts64 p;
4282
4283 bfloat16_unpack_canonical(&p, a, status);
4284 parts_sqrt(&p, status, &bfloat16_params);
4285 return bfloat16_round_pack_canonical(&p, status);
4286 }
4287
4288 float128 QEMU_FLATTEN float128_sqrt(float128 a, float_status *status)
4289 {
4290 FloatParts128 p;
4291
4292 float128_unpack_canonical(&p, a, status);
4293 parts_sqrt(&p, status, &float128_params);
4294 return float128_round_pack_canonical(&p, status);
4295 }
4296
4297 floatx80 floatx80_sqrt(floatx80 a, float_status *s)
4298 {
4299 FloatParts128 p;
4300
4301 if (!floatx80_unpack_canonical(&p, a, s)) {
4302 return floatx80_default_nan(s);
4303 }
4304 parts_sqrt(&p, s, &floatx80_params[s->floatx80_rounding_precision]);
4305 return floatx80_round_pack_canonical(&p, s);
4306 }
4307
4308 /*
4309 * log2
4310 */
4311 float32 float32_log2(float32 a, float_status *status)
4312 {
4313 FloatParts64 p;
4314
4315 float32_unpack_canonical(&p, a, status);
4316 parts_log2(&p, status, &float32_params);
4317 return float32_round_pack_canonical(&p, status);
4318 }
4319
4320 float64 float64_log2(float64 a, float_status *status)
4321 {
4322 FloatParts64 p;
4323
4324 float64_unpack_canonical(&p, a, status);
4325 parts_log2(&p, status, &float64_params);
4326 return float64_round_pack_canonical(&p, status);
4327 }
4328
4329 /*----------------------------------------------------------------------------
4330 | The pattern for a default generated NaN.
4331 *----------------------------------------------------------------------------*/
4332
4333 float16 float16_default_nan(float_status *status)
4334 {
4335 FloatParts64 p;
4336
4337 parts_default_nan(&p, status);
4338 p.frac >>= float16_params.frac_shift;
4339 return float16_pack_raw(&p);
4340 }
4341
4342 float32 float32_default_nan(float_status *status)
4343 {
4344 FloatParts64 p;
4345
4346 parts_default_nan(&p, status);
4347 p.frac >>= float32_params.frac_shift;
4348 return float32_pack_raw(&p);
4349 }
4350
4351 float64 float64_default_nan(float_status *status)
4352 {
4353 FloatParts64 p;
4354
4355 parts_default_nan(&p, status);
4356 p.frac >>= float64_params.frac_shift;
4357 return float64_pack_raw(&p);
4358 }
4359
4360 float128 float128_default_nan(float_status *status)
4361 {
4362 FloatParts128 p;
4363
4364 parts_default_nan(&p, status);
4365 frac_shr(&p, float128_params.frac_shift);
4366 return float128_pack_raw(&p);
4367 }
4368
4369 bfloat16 bfloat16_default_nan(float_status *status)
4370 {
4371 FloatParts64 p;
4372
4373 parts_default_nan(&p, status);
4374 p.frac >>= bfloat16_params.frac_shift;
4375 return bfloat16_pack_raw(&p);
4376 }
4377
4378 /*----------------------------------------------------------------------------
4379 | Returns a quiet NaN from a signalling NaN for the floating point value `a'.
4380 *----------------------------------------------------------------------------*/
4381
4382 float16 float16_silence_nan(float16 a, float_status *status)
4383 {
4384 FloatParts64 p;
4385
4386 float16_unpack_raw(&p, a);
4387 p.frac <<= float16_params.frac_shift;
4388 parts_silence_nan(&p, status);
4389 p.frac >>= float16_params.frac_shift;
4390 return float16_pack_raw(&p);
4391 }
4392
4393 float32 float32_silence_nan(float32 a, float_status *status)
4394 {
4395 FloatParts64 p;
4396
4397 float32_unpack_raw(&p, a);
4398 p.frac <<= float32_params.frac_shift;
4399 parts_silence_nan(&p, status);
4400 p.frac >>= float32_params.frac_shift;
4401 return float32_pack_raw(&p);
4402 }
4403
4404 float64 float64_silence_nan(float64 a, float_status *status)
4405 {
4406 FloatParts64 p;
4407
4408 float64_unpack_raw(&p, a);
4409 p.frac <<= float64_params.frac_shift;
4410 parts_silence_nan(&p, status);
4411 p.frac >>= float64_params.frac_shift;
4412 return float64_pack_raw(&p);
4413 }
4414
4415 bfloat16 bfloat16_silence_nan(bfloat16 a, float_status *status)
4416 {
4417 FloatParts64 p;
4418
4419 bfloat16_unpack_raw(&p, a);
4420 p.frac <<= bfloat16_params.frac_shift;
4421 parts_silence_nan(&p, status);
4422 p.frac >>= bfloat16_params.frac_shift;
4423 return bfloat16_pack_raw(&p);
4424 }
4425
4426 float128 float128_silence_nan(float128 a, float_status *status)
4427 {
4428 FloatParts128 p;
4429
4430 float128_unpack_raw(&p, a);
4431 frac_shl(&p, float128_params.frac_shift);
4432 parts_silence_nan(&p, status);
4433 frac_shr(&p, float128_params.frac_shift);
4434 return float128_pack_raw(&p);
4435 }
4436
4437 /*----------------------------------------------------------------------------
4438 | If `a' is denormal and we are in flush-to-zero mode then set the
4439 | input-denormal exception and return zero. Otherwise just return the value.
4440 *----------------------------------------------------------------------------*/
4441
4442 static bool parts_squash_denormal(FloatParts64 p, float_status *status)
4443 {
4444 if (p.exp == 0 && p.frac != 0) {
4445 float_raise(float_flag_input_denormal, status);
4446 return true;
4447 }
4448
4449 return false;
4450 }
4451
4452 float16 float16_squash_input_denormal(float16 a, float_status *status)
4453 {
4454 if (status->flush_inputs_to_zero) {
4455 FloatParts64 p;
4456
4457 float16_unpack_raw(&p, a);
4458 if (parts_squash_denormal(p, status)) {
4459 return float16_set_sign(float16_zero, p.sign);
4460 }
4461 }
4462 return a;
4463 }
4464
4465 float32 float32_squash_input_denormal(float32 a, float_status *status)
4466 {
4467 if (status->flush_inputs_to_zero) {
4468 FloatParts64 p;
4469
4470 float32_unpack_raw(&p, a);
4471 if (parts_squash_denormal(p, status)) {
4472 return float32_set_sign(float32_zero, p.sign);
4473 }
4474 }
4475 return a;
4476 }
4477
4478 float64 float64_squash_input_denormal(float64 a, float_status *status)
4479 {
4480 if (status->flush_inputs_to_zero) {
4481 FloatParts64 p;
4482
4483 float64_unpack_raw(&p, a);
4484 if (parts_squash_denormal(p, status)) {
4485 return float64_set_sign(float64_zero, p.sign);
4486 }
4487 }
4488 return a;
4489 }
4490
4491 bfloat16 bfloat16_squash_input_denormal(bfloat16 a, float_status *status)
4492 {
4493 if (status->flush_inputs_to_zero) {
4494 FloatParts64 p;
4495
4496 bfloat16_unpack_raw(&p, a);
4497 if (parts_squash_denormal(p, status)) {
4498 return bfloat16_set_sign(bfloat16_zero, p.sign);
4499 }
4500 }
4501 return a;
4502 }
4503
4504 /*----------------------------------------------------------------------------
4505 | Normalizes the subnormal extended double-precision floating-point value
4506 | represented by the denormalized significand `aSig'. The normalized exponent
4507 | and significand are stored at the locations pointed to by `zExpPtr' and
4508 | `zSigPtr', respectively.
4509 *----------------------------------------------------------------------------*/
4510
4511 void normalizeFloatx80Subnormal(uint64_t aSig, int32_t *zExpPtr,
4512 uint64_t *zSigPtr)
4513 {
4514 int8_t shiftCount;
4515
4516 shiftCount = clz64(aSig);
4517 *zSigPtr = aSig<<shiftCount;
4518 *zExpPtr = 1 - shiftCount;
4519 }
4520
4521 /*----------------------------------------------------------------------------
4522 | Takes an abstract floating-point value having sign `zSign', exponent `zExp',
4523 | and extended significand formed by the concatenation of `zSig0' and `zSig1',
4524 | and returns the proper extended double-precision floating-point value
4525 | corresponding to the abstract input. Ordinarily, the abstract value is
4526 | rounded and packed into the extended double-precision format, with the
4527 | inexact exception raised if the abstract input cannot be represented
4528 | exactly. However, if the abstract value is too large, the overflow and
4529 | inexact exceptions are raised and an infinity or maximal finite value is
4530 | returned. If the abstract value is too small, the input value is rounded to
4531 | a subnormal number, and the underflow and inexact exceptions are raised if
4532 | the abstract input cannot be represented exactly as a subnormal extended
4533 | double-precision floating-point number.
4534 | If `roundingPrecision' is floatx80_precision_s or floatx80_precision_d,
4535 | the result is rounded to the same number of bits as single or double
4536 | precision, respectively. Otherwise, the result is rounded to the full
4537 | precision of the extended double-precision format.
4538 | The input significand must be normalized or smaller. If the input
4539 | significand is not normalized, `zExp' must be 0; in that case, the result
4540 | returned is a subnormal number, and it must not require rounding. The
4541 | handling of underflow and overflow follows the IEC/IEEE Standard for Binary
4542 | Floating-Point Arithmetic.
4543 *----------------------------------------------------------------------------*/
4544
4545 floatx80 roundAndPackFloatx80(FloatX80RoundPrec roundingPrecision, bool zSign,
4546 int32_t zExp, uint64_t zSig0, uint64_t zSig1,
4547 float_status *status)
4548 {
4549 FloatRoundMode roundingMode;
4550 bool roundNearestEven, increment, isTiny;
4551 int64_t roundIncrement, roundMask, roundBits;
4552
4553 roundingMode = status->float_rounding_mode;
4554 roundNearestEven = ( roundingMode == float_round_nearest_even );
4555 switch (roundingPrecision) {
4556 case floatx80_precision_x:
4557 goto precision80;
4558 case floatx80_precision_d:
4559 roundIncrement = UINT64_C(0x0000000000000400);
4560 roundMask = UINT64_C(0x00000000000007FF);
4561 break;
4562 case floatx80_precision_s:
4563 roundIncrement = UINT64_C(0x0000008000000000);
4564 roundMask = UINT64_C(0x000000FFFFFFFFFF);
4565 break;
4566 default:
4567 g_assert_not_reached();
4568 }
4569 zSig0 |= ( zSig1 != 0 );
4570 switch (roundingMode) {
4571 case float_round_nearest_even:
4572 case float_round_ties_away:
4573 break;
4574 case float_round_to_zero:
4575 roundIncrement = 0;
4576 break;
4577 case float_round_up:
4578 roundIncrement = zSign ? 0 : roundMask;
4579 break;
4580 case float_round_down:
4581 roundIncrement = zSign ? roundMask : 0;
4582 break;
4583 default:
4584 abort();
4585 }
4586 roundBits = zSig0 & roundMask;
4587 if ( 0x7FFD <= (uint32_t) ( zExp - 1 ) ) {
4588 if ( ( 0x7FFE < zExp )
4589 || ( ( zExp == 0x7FFE ) && ( zSig0 + roundIncrement < zSig0 ) )
4590 ) {
4591 goto overflow;
4592 }
4593 if ( zExp <= 0 ) {
4594 if (status->flush_to_zero) {
4595 float_raise(float_flag_output_denormal, status);
4596 return packFloatx80(zSign, 0, 0);
4597 }
4598 isTiny = status->tininess_before_rounding
4599 || (zExp < 0 )
4600 || (zSig0 <= zSig0 + roundIncrement);
4601 shift64RightJamming( zSig0, 1 - zExp, &zSig0 );
4602 zExp = 0;
4603 roundBits = zSig0 & roundMask;
4604 if (isTiny && roundBits) {
4605 float_raise(float_flag_underflow, status);
4606 }
4607 if (roundBits) {
4608 float_raise(float_flag_inexact, status);
4609 }
4610 zSig0 += roundIncrement;
4611 if ( (int64_t) zSig0 < 0 ) zExp = 1;
4612 roundIncrement = roundMask + 1;
4613 if ( roundNearestEven && ( roundBits<<1 == roundIncrement ) ) {
4614 roundMask |= roundIncrement;
4615 }
4616 zSig0 &= ~ roundMask;
4617 return packFloatx80( zSign, zExp, zSig0 );
4618 }
4619 }
4620 if (roundBits) {
4621 float_raise(float_flag_inexact, status);
4622 }
4623 zSig0 += roundIncrement;
4624 if ( zSig0 < roundIncrement ) {
4625 ++zExp;
4626 zSig0 = UINT64_C(0x8000000000000000);
4627 }
4628 roundIncrement = roundMask + 1;
4629 if ( roundNearestEven && ( roundBits<<1 == roundIncrement ) ) {
4630 roundMask |= roundIncrement;
4631 }
4632 zSig0 &= ~ roundMask;
4633 if ( zSig0 == 0 ) zExp = 0;
4634 return packFloatx80( zSign, zExp, zSig0 );
4635 precision80:
4636 switch (roundingMode) {
4637 case float_round_nearest_even:
4638 case float_round_ties_away:
4639 increment = ((int64_t)zSig1 < 0);
4640 break;
4641 case float_round_to_zero:
4642 increment = 0;
4643 break;
4644 case float_round_up:
4645 increment = !zSign && zSig1;
4646 break;
4647 case float_round_down:
4648 increment = zSign && zSig1;
4649 break;
4650 default:
4651 abort();
4652 }
4653 if ( 0x7FFD <= (uint32_t) ( zExp - 1 ) ) {
4654 if ( ( 0x7FFE < zExp )
4655 || ( ( zExp == 0x7FFE )
4656 && ( zSig0 == UINT64_C(0xFFFFFFFFFFFFFFFF) )
4657 && increment
4658 )
4659 ) {
4660 roundMask = 0;
4661 overflow:
4662 float_raise(float_flag_overflow | float_flag_inexact, status);
4663 if ( ( roundingMode == float_round_to_zero )
4664 || ( zSign && ( roundingMode == float_round_up ) )
4665 || ( ! zSign && ( roundingMode == float_round_down ) )
4666 ) {
4667 return packFloatx80( zSign, 0x7FFE, ~ roundMask );
4668 }
4669 return packFloatx80(zSign,
4670 floatx80_infinity_high,
4671 floatx80_infinity_low);
4672 }
4673 if ( zExp <= 0 ) {
4674 isTiny = status->tininess_before_rounding
4675 || (zExp < 0)
4676 || !increment
4677 || (zSig0 < UINT64_C(0xFFFFFFFFFFFFFFFF));
4678 shift64ExtraRightJamming( zSig0, zSig1, 1 - zExp, &zSig0, &zSig1 );
4679 zExp = 0;
4680 if (isTiny && zSig1) {
4681 float_raise(float_flag_underflow, status);
4682 }
4683 if (zSig1) {
4684 float_raise(float_flag_inexact, status);
4685 }
4686 switch (roundingMode) {
4687 case float_round_nearest_even:
4688 case float_round_ties_away:
4689 increment = ((int64_t)zSig1 < 0);
4690 break;
4691 case float_round_to_zero:
4692 increment = 0;
4693 break;
4694 case float_round_up:
4695 increment = !zSign && zSig1;
4696 break;
4697 case float_round_down:
4698 increment = zSign && zSig1;
4699 break;
4700 default:
4701 abort();
4702 }
4703 if ( increment ) {
4704 ++zSig0;
4705 if (!(zSig1 << 1) && roundNearestEven) {
4706 zSig0 &= ~1;
4707 }
4708 if ( (int64_t) zSig0 < 0 ) zExp = 1;
4709 }
4710 return packFloatx80( zSign, zExp, zSig0 );
4711 }
4712 }
4713 if (zSig1) {
4714 float_raise(float_flag_inexact, status);
4715 }
4716 if ( increment ) {
4717 ++zSig0;
4718 if ( zSig0 == 0 ) {
4719 ++zExp;
4720 zSig0 = UINT64_C(0x8000000000000000);
4721 }
4722 else {
4723 if (!(zSig1 << 1) && roundNearestEven) {
4724 zSig0 &= ~1;
4725 }
4726 }
4727 }
4728 else {
4729 if ( zSig0 == 0 ) zExp = 0;
4730 }
4731 return packFloatx80( zSign, zExp, zSig0 );
4732
4733 }
4734
4735 /*----------------------------------------------------------------------------
4736 | Takes an abstract floating-point value having sign `zSign', exponent
4737 | `zExp', and significand formed by the concatenation of `zSig0' and `zSig1',
4738 | and returns the proper extended double-precision floating-point value
4739 | corresponding to the abstract input. This routine is just like
4740 | `roundAndPackFloatx80' except that the input significand does not have to be
4741 | normalized.
4742 *----------------------------------------------------------------------------*/
4743
4744 floatx80 normalizeRoundAndPackFloatx80(FloatX80RoundPrec roundingPrecision,
4745 bool zSign, int32_t zExp,
4746 uint64_t zSig0, uint64_t zSig1,
4747 float_status *status)
4748 {
4749 int8_t shiftCount;
4750
4751 if ( zSig0 == 0 ) {
4752 zSig0 = zSig1;
4753 zSig1 = 0;
4754 zExp -= 64;
4755 }
4756 shiftCount = clz64(zSig0);
4757 shortShift128Left( zSig0, zSig1, shiftCount, &zSig0, &zSig1 );
4758 zExp -= shiftCount;
4759 return roundAndPackFloatx80(roundingPrecision, zSign, zExp,
4760 zSig0, zSig1, status);
4761
4762 }
4763
4764 /*----------------------------------------------------------------------------
4765 | Returns the binary exponential of the single-precision floating-point value
4766 | `a'. The operation is performed according to the IEC/IEEE Standard for
4767 | Binary Floating-Point Arithmetic.
4768 |
4769 | Uses the following identities:
4770 |
4771 | 1. -------------------------------------------------------------------------
4772 | x x*ln(2)
4773 | 2 = e
4774 |
4775 | 2. -------------------------------------------------------------------------
4776 | 2 3 4 5 n
4777 | x x x x x x x
4778 | e = 1 + --- + --- + --- + --- + --- + ... + --- + ...
4779 | 1! 2! 3! 4! 5! n!
4780 *----------------------------------------------------------------------------*/
4781
4782 static const float64 float32_exp2_coefficients[15] =
4783 {
4784 const_float64( 0x3ff0000000000000ll ), /* 1 */
4785 const_float64( 0x3fe0000000000000ll ), /* 2 */
4786 const_float64( 0x3fc5555555555555ll ), /* 3 */
4787 const_float64( 0x3fa5555555555555ll ), /* 4 */
4788 const_float64( 0x3f81111111111111ll ), /* 5 */
4789 const_float64( 0x3f56c16c16c16c17ll ), /* 6 */
4790 const_float64( 0x3f2a01a01a01a01all ), /* 7 */
4791 const_float64( 0x3efa01a01a01a01all ), /* 8 */
4792 const_float64( 0x3ec71de3a556c734ll ), /* 9 */
4793 const_float64( 0x3e927e4fb7789f5cll ), /* 10 */
4794 const_float64( 0x3e5ae64567f544e4ll ), /* 11 */
4795 const_float64( 0x3e21eed8eff8d898ll ), /* 12 */
4796 const_float64( 0x3de6124613a86d09ll ), /* 13 */
4797 const_float64( 0x3da93974a8c07c9dll ), /* 14 */
4798 const_float64( 0x3d6ae7f3e733b81fll ), /* 15 */
4799 };
4800
4801 float32 float32_exp2(float32 a, float_status *status)
4802 {
4803 FloatParts64 xp, xnp, tp, rp;
4804 int i;
4805
4806 float32_unpack_canonical(&xp, a, status);
4807 if (unlikely(xp.cls != float_class_normal)) {
4808 switch (xp.cls) {
4809 case float_class_snan:
4810 case float_class_qnan:
4811 parts_return_nan(&xp, status);
4812 return float32_round_pack_canonical(&xp, status);
4813 case float_class_inf:
4814 return xp.sign ? float32_zero : a;
4815 case float_class_zero:
4816 return float32_one;
4817 default:
4818 break;
4819 }
4820 g_assert_not_reached();
4821 }
4822
4823 float_raise(float_flag_inexact, status);
4824
4825 float64_unpack_canonical(&tp, float64_ln2, status);
4826 xp = *parts_mul(&xp, &tp, status);
4827 xnp = xp;
4828
4829 float64_unpack_canonical(&rp, float64_one, status);
4830 for (i = 0 ; i < 15 ; i++) {
4831 float64_unpack_canonical(&tp, float32_exp2_coefficients[i], status);
4832 rp = *parts_muladd(&tp, &xp, &rp, 0, status);
4833 xnp = *parts_mul(&xnp, &xp, status);
4834 }
4835
4836 return float32_round_pack_canonical(&rp, status);
4837 }
4838
4839 /*----------------------------------------------------------------------------
4840 | Rounds the extended double-precision floating-point value `a'
4841 | to the precision provided by floatx80_rounding_precision and returns the
4842 | result as an extended double-precision floating-point value.
4843 | The operation is performed according to the IEC/IEEE Standard for Binary
4844 | Floating-Point Arithmetic.
4845 *----------------------------------------------------------------------------*/
4846
4847 floatx80 floatx80_round(floatx80 a, float_status *status)
4848 {
4849 FloatParts128 p;
4850
4851 if (!floatx80_unpack_canonical(&p, a, status)) {
4852 return floatx80_default_nan(status);
4853 }
4854 return floatx80_round_pack_canonical(&p, status);
4855 }
4856
4857 static void __attribute__((constructor)) softfloat_init(void)
4858 {
4859 union_float64 ua, ub, uc, ur;
4860
4861 if (QEMU_NO_HARDFLOAT) {
4862 return;
4863 }
4864 /*
4865 * Test that the host's FMA is not obviously broken. For example,
4866 * glibc < 2.23 can perform an incorrect FMA on certain hosts; see
4867 * https://sourceware.org/bugzilla/show_bug.cgi?id=13304
4868 */
4869 ua.s = 0x0020000000000001ULL;
4870 ub.s = 0x3ca0000000000000ULL;
4871 uc.s = 0x0020000000000000ULL;
4872 ur.h = fma(ua.h, ub.h, uc.h);
4873 if (ur.s != 0x0020000000000001ULL) {
4874 force_soft_fma = true;
4875 }
4876 }