1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20#include "qemu/osdep.h"
21#include "cpu.h"
22#include "exec/gdbstub.h"
23#include "exec/helper-proto.h"
24#include "qemu/host-utils.h"
25#include "qemu/log.h"
26#include "sysemu/sysemu.h"
27#include "qemu/bitops.h"
28#include "internals.h"
29#include "qemu/crc32c.h"
30#include "exec/exec-all.h"
31#include "exec/cpu_ldst.h"
32#include "qemu/int128.h"
33#include "tcg.h"
34#include "fpu/softfloat.h"
35#include <zlib.h>
36
37
38
39uint64_t HELPER(udiv64)(uint64_t num, uint64_t den)
40{
41 if (den == 0) {
42 return 0;
43 }
44 return num / den;
45}
46
47int64_t HELPER(sdiv64)(int64_t num, int64_t den)
48{
49 if (den == 0) {
50 return 0;
51 }
52 if (num == LLONG_MIN && den == -1) {
53 return LLONG_MIN;
54 }
55 return num / den;
56}
57
58uint64_t HELPER(rbit64)(uint64_t x)
59{
60 return revbit64(x);
61}
62
63
64
65
66
67static inline uint32_t float_rel_to_flags(int res)
68{
69 uint64_t flags;
70 switch (res) {
71 case float_relation_equal:
72 flags = PSTATE_Z | PSTATE_C;
73 break;
74 case float_relation_less:
75 flags = PSTATE_N;
76 break;
77 case float_relation_greater:
78 flags = PSTATE_C;
79 break;
80 case float_relation_unordered:
81 default:
82 flags = PSTATE_C | PSTATE_V;
83 break;
84 }
85 return flags;
86}
87
88uint64_t HELPER(vfp_cmph_a64)(float16 x, float16 y, void *fp_status)
89{
90 return float_rel_to_flags(float16_compare_quiet(x, y, fp_status));
91}
92
93uint64_t HELPER(vfp_cmpeh_a64)(float16 x, float16 y, void *fp_status)
94{
95 return float_rel_to_flags(float16_compare(x, y, fp_status));
96}
97
98uint64_t HELPER(vfp_cmps_a64)(float32 x, float32 y, void *fp_status)
99{
100 return float_rel_to_flags(float32_compare_quiet(x, y, fp_status));
101}
102
103uint64_t HELPER(vfp_cmpes_a64)(float32 x, float32 y, void *fp_status)
104{
105 return float_rel_to_flags(float32_compare(x, y, fp_status));
106}
107
108uint64_t HELPER(vfp_cmpd_a64)(float64 x, float64 y, void *fp_status)
109{
110 return float_rel_to_flags(float64_compare_quiet(x, y, fp_status));
111}
112
113uint64_t HELPER(vfp_cmped_a64)(float64 x, float64 y, void *fp_status)
114{
115 return float_rel_to_flags(float64_compare(x, y, fp_status));
116}
117
118float32 HELPER(vfp_mulxs)(float32 a, float32 b, void *fpstp)
119{
120 float_status *fpst = fpstp;
121
122 a = float32_squash_input_denormal(a, fpst);
123 b = float32_squash_input_denormal(b, fpst);
124
125 if ((float32_is_zero(a) && float32_is_infinity(b)) ||
126 (float32_is_infinity(a) && float32_is_zero(b))) {
127
128 return make_float32((1U << 30) |
129 ((float32_val(a) ^ float32_val(b)) & (1U << 31)));
130 }
131 return float32_mul(a, b, fpst);
132}
133
134float64 HELPER(vfp_mulxd)(float64 a, float64 b, void *fpstp)
135{
136 float_status *fpst = fpstp;
137
138 a = float64_squash_input_denormal(a, fpst);
139 b = float64_squash_input_denormal(b, fpst);
140
141 if ((float64_is_zero(a) && float64_is_infinity(b)) ||
142 (float64_is_infinity(a) && float64_is_zero(b))) {
143
144 return make_float64((1ULL << 62) |
145 ((float64_val(a) ^ float64_val(b)) & (1ULL << 63)));
146 }
147 return float64_mul(a, b, fpst);
148}
149
150uint64_t HELPER(simd_tbl)(CPUARMState *env, uint64_t result, uint64_t indices,
151 uint32_t rn, uint32_t numregs)
152{
153
154
155
156
157
158
159
160 int shift;
161
162 for (shift = 0; shift < 64; shift += 8) {
163 int index = extract64(indices, shift, 8);
164 if (index < 16 * numregs) {
165
166
167
168
169
170
171 int elt = (rn * 2 + (index >> 3)) % 64;
172 int bitidx = (index & 7) * 8;
173 uint64_t *q = aa64_vfp_qreg(env, elt >> 1);
174 uint64_t val = extract64(q[elt & 1], bitidx, 8);
175
176 result = deposit64(result, shift, 8, val);
177 }
178 }
179 return result;
180}
181
182
183uint64_t HELPER(neon_ceq_f64)(float64 a, float64 b, void *fpstp)
184{
185 float_status *fpst = fpstp;
186 return -float64_eq_quiet(a, b, fpst);
187}
188
189uint64_t HELPER(neon_cge_f64)(float64 a, float64 b, void *fpstp)
190{
191 float_status *fpst = fpstp;
192 return -float64_le(b, a, fpst);
193}
194
195uint64_t HELPER(neon_cgt_f64)(float64 a, float64 b, void *fpstp)
196{
197 float_status *fpst = fpstp;
198 return -float64_lt(b, a, fpst);
199}
200
201
202
203
204
205#define float16_two make_float16(0x4000)
206#define float16_three make_float16(0x4200)
207#define float16_one_point_five make_float16(0x3e00)
208
209#define float32_two make_float32(0x40000000)
210#define float32_three make_float32(0x40400000)
211#define float32_one_point_five make_float32(0x3fc00000)
212
213#define float64_two make_float64(0x4000000000000000ULL)
214#define float64_three make_float64(0x4008000000000000ULL)
215#define float64_one_point_five make_float64(0x3FF8000000000000ULL)
216
217float16 HELPER(recpsf_f16)(float16 a, float16 b, void *fpstp)
218{
219 float_status *fpst = fpstp;
220
221 a = float16_squash_input_denormal(a, fpst);
222 b = float16_squash_input_denormal(b, fpst);
223
224 a = float16_chs(a);
225 if ((float16_is_infinity(a) && float16_is_zero(b)) ||
226 (float16_is_infinity(b) && float16_is_zero(a))) {
227 return float16_two;
228 }
229 return float16_muladd(a, b, float16_two, 0, fpst);
230}
231
232float32 HELPER(recpsf_f32)(float32 a, float32 b, void *fpstp)
233{
234 float_status *fpst = fpstp;
235
236 a = float32_squash_input_denormal(a, fpst);
237 b = float32_squash_input_denormal(b, fpst);
238
239 a = float32_chs(a);
240 if ((float32_is_infinity(a) && float32_is_zero(b)) ||
241 (float32_is_infinity(b) && float32_is_zero(a))) {
242 return float32_two;
243 }
244 return float32_muladd(a, b, float32_two, 0, fpst);
245}
246
247float64 HELPER(recpsf_f64)(float64 a, float64 b, void *fpstp)
248{
249 float_status *fpst = fpstp;
250
251 a = float64_squash_input_denormal(a, fpst);
252 b = float64_squash_input_denormal(b, fpst);
253
254 a = float64_chs(a);
255 if ((float64_is_infinity(a) && float64_is_zero(b)) ||
256 (float64_is_infinity(b) && float64_is_zero(a))) {
257 return float64_two;
258 }
259 return float64_muladd(a, b, float64_two, 0, fpst);
260}
261
262float16 HELPER(rsqrtsf_f16)(float16 a, float16 b, void *fpstp)
263{
264 float_status *fpst = fpstp;
265
266 a = float16_squash_input_denormal(a, fpst);
267 b = float16_squash_input_denormal(b, fpst);
268
269 a = float16_chs(a);
270 if ((float16_is_infinity(a) && float16_is_zero(b)) ||
271 (float16_is_infinity(b) && float16_is_zero(a))) {
272 return float16_one_point_five;
273 }
274 return float16_muladd(a, b, float16_three, float_muladd_halve_result, fpst);
275}
276
277float32 HELPER(rsqrtsf_f32)(float32 a, float32 b, void *fpstp)
278{
279 float_status *fpst = fpstp;
280
281 a = float32_squash_input_denormal(a, fpst);
282 b = float32_squash_input_denormal(b, fpst);
283
284 a = float32_chs(a);
285 if ((float32_is_infinity(a) && float32_is_zero(b)) ||
286 (float32_is_infinity(b) && float32_is_zero(a))) {
287 return float32_one_point_five;
288 }
289 return float32_muladd(a, b, float32_three, float_muladd_halve_result, fpst);
290}
291
292float64 HELPER(rsqrtsf_f64)(float64 a, float64 b, void *fpstp)
293{
294 float_status *fpst = fpstp;
295
296 a = float64_squash_input_denormal(a, fpst);
297 b = float64_squash_input_denormal(b, fpst);
298
299 a = float64_chs(a);
300 if ((float64_is_infinity(a) && float64_is_zero(b)) ||
301 (float64_is_infinity(b) && float64_is_zero(a))) {
302 return float64_one_point_five;
303 }
304 return float64_muladd(a, b, float64_three, float_muladd_halve_result, fpst);
305}
306
307
308
309
310uint64_t HELPER(neon_addlp_s8)(uint64_t a)
311{
312 uint64_t nsignmask = 0x0080008000800080ULL;
313 uint64_t wsignmask = 0x8000800080008000ULL;
314 uint64_t elementmask = 0x00ff00ff00ff00ffULL;
315 uint64_t tmp1, tmp2;
316 uint64_t res, signres;
317
318
319 tmp1 = a & elementmask;
320 tmp1 ^= nsignmask;
321 tmp1 |= wsignmask;
322 tmp1 = (tmp1 - nsignmask) ^ wsignmask;
323
324 tmp2 = (a >> 8) & elementmask;
325 tmp2 ^= nsignmask;
326 tmp2 |= wsignmask;
327 tmp2 = (tmp2 - nsignmask) ^ wsignmask;
328
329
330
331
332
333 signres = (tmp1 ^ tmp2) & wsignmask;
334 res = (tmp1 & ~wsignmask) + (tmp2 & ~wsignmask);
335 res ^= signres;
336
337 return res;
338}
339
340uint64_t HELPER(neon_addlp_u8)(uint64_t a)
341{
342 uint64_t tmp;
343
344 tmp = a & 0x00ff00ff00ff00ffULL;
345 tmp += (a >> 8) & 0x00ff00ff00ff00ffULL;
346 return tmp;
347}
348
349uint64_t HELPER(neon_addlp_s16)(uint64_t a)
350{
351 int32_t reslo, reshi;
352
353 reslo = (int32_t)(int16_t)a + (int32_t)(int16_t)(a >> 16);
354 reshi = (int32_t)(int16_t)(a >> 32) + (int32_t)(int16_t)(a >> 48);
355
356 return (uint32_t)reslo | (((uint64_t)reshi) << 32);
357}
358
359uint64_t HELPER(neon_addlp_u16)(uint64_t a)
360{
361 uint64_t tmp;
362
363 tmp = a & 0x0000ffff0000ffffULL;
364 tmp += (a >> 16) & 0x0000ffff0000ffffULL;
365 return tmp;
366}
367
368
369float16 HELPER(frecpx_f16)(float16 a, void *fpstp)
370{
371 float_status *fpst = fpstp;
372 uint16_t val16, sbit;
373 int16_t exp;
374
375 if (float16_is_any_nan(a)) {
376 float16 nan = a;
377 if (float16_is_signaling_nan(a, fpst)) {
378 float_raise(float_flag_invalid, fpst);
379 nan = float16_maybe_silence_nan(a, fpst);
380 }
381 if (fpst->default_nan_mode) {
382 nan = float16_default_nan(fpst);
383 }
384 return nan;
385 }
386
387 val16 = float16_val(a);
388 sbit = 0x8000 & val16;
389 exp = extract32(val16, 10, 5);
390
391 if (exp == 0) {
392 return make_float16(deposit32(sbit, 10, 5, 0x1e));
393 } else {
394 return make_float16(deposit32(sbit, 10, 5, ~exp));
395 }
396}
397
398float32 HELPER(frecpx_f32)(float32 a, void *fpstp)
399{
400 float_status *fpst = fpstp;
401 uint32_t val32, sbit;
402 int32_t exp;
403
404 if (float32_is_any_nan(a)) {
405 float32 nan = a;
406 if (float32_is_signaling_nan(a, fpst)) {
407 float_raise(float_flag_invalid, fpst);
408 nan = float32_maybe_silence_nan(a, fpst);
409 }
410 if (fpst->default_nan_mode) {
411 nan = float32_default_nan(fpst);
412 }
413 return nan;
414 }
415
416 val32 = float32_val(a);
417 sbit = 0x80000000ULL & val32;
418 exp = extract32(val32, 23, 8);
419
420 if (exp == 0) {
421 return make_float32(sbit | (0xfe << 23));
422 } else {
423 return make_float32(sbit | (~exp & 0xff) << 23);
424 }
425}
426
427float64 HELPER(frecpx_f64)(float64 a, void *fpstp)
428{
429 float_status *fpst = fpstp;
430 uint64_t val64, sbit;
431 int64_t exp;
432
433 if (float64_is_any_nan(a)) {
434 float64 nan = a;
435 if (float64_is_signaling_nan(a, fpst)) {
436 float_raise(float_flag_invalid, fpst);
437 nan = float64_maybe_silence_nan(a, fpst);
438 }
439 if (fpst->default_nan_mode) {
440 nan = float64_default_nan(fpst);
441 }
442 return nan;
443 }
444
445 val64 = float64_val(a);
446 sbit = 0x8000000000000000ULL & val64;
447 exp = extract64(float64_val(a), 52, 11);
448
449 if (exp == 0) {
450 return make_float64(sbit | (0x7feULL << 52));
451 } else {
452 return make_float64(sbit | (~exp & 0x7ffULL) << 52);
453 }
454}
455
456float32 HELPER(fcvtx_f64_to_f32)(float64 a, CPUARMState *env)
457{
458
459
460
461 float32 r;
462 float_status *fpst = &env->vfp.fp_status;
463 float_status tstat = *fpst;
464 int exflags;
465
466 set_float_rounding_mode(float_round_to_zero, &tstat);
467 set_float_exception_flags(0, &tstat);
468 r = float64_to_float32(a, &tstat);
469 r = float32_maybe_silence_nan(r, &tstat);
470 exflags = get_float_exception_flags(&tstat);
471 if (exflags & float_flag_inexact) {
472 r = make_float32(float32_val(r) | 1);
473 }
474 exflags |= get_float_exception_flags(fpst);
475 set_float_exception_flags(exflags, fpst);
476 return r;
477}
478
479
480
481
482
483
484
485
486
487uint64_t HELPER(crc32_64)(uint64_t acc, uint64_t val, uint32_t bytes)
488{
489 uint8_t buf[8];
490
491 stq_le_p(buf, val);
492
493
494 return crc32(acc ^ 0xffffffff, buf, bytes) ^ 0xffffffff;
495}
496
497uint64_t HELPER(crc32c_64)(uint64_t acc, uint64_t val, uint32_t bytes)
498{
499 uint8_t buf[8];
500
501 stq_le_p(buf, val);
502
503
504 return crc32c(acc, buf, bytes) ^ 0xffffffff;
505}
506
507
508static uint64_t do_paired_cmpxchg64_le(CPUARMState *env, uint64_t addr,
509 uint64_t new_lo, uint64_t new_hi,
510 bool parallel, uintptr_t ra)
511{
512 Int128 oldv, cmpv, newv;
513 bool success;
514
515 cmpv = int128_make128(env->exclusive_val, env->exclusive_high);
516 newv = int128_make128(new_lo, new_hi);
517
518 if (parallel) {
519#ifndef CONFIG_ATOMIC128
520 cpu_loop_exit_atomic(ENV_GET_CPU(env), ra);
521#else
522 int mem_idx = cpu_mmu_index(env, false);
523 TCGMemOpIdx oi = make_memop_idx(MO_LEQ | MO_ALIGN_16, mem_idx);
524 oldv = helper_atomic_cmpxchgo_le_mmu(env, addr, cmpv, newv, oi, ra);
525 success = int128_eq(oldv, cmpv);
526#endif
527 } else {
528 uint64_t o0, o1;
529
530#ifdef CONFIG_USER_ONLY
531
532 uint64_t *haddr = g2h(addr);
533
534 helper_retaddr = ra;
535 o0 = ldq_le_p(haddr + 0);
536 o1 = ldq_le_p(haddr + 1);
537 oldv = int128_make128(o0, o1);
538
539 success = int128_eq(oldv, cmpv);
540 if (success) {
541 stq_le_p(haddr + 0, int128_getlo(newv));
542 stq_le_p(haddr + 1, int128_gethi(newv));
543 }
544 helper_retaddr = 0;
545#else
546 int mem_idx = cpu_mmu_index(env, false);
547 TCGMemOpIdx oi0 = make_memop_idx(MO_LEQ | MO_ALIGN_16, mem_idx);
548 TCGMemOpIdx oi1 = make_memop_idx(MO_LEQ, mem_idx);
549
550 o0 = helper_le_ldq_mmu(env, addr + 0, oi0, ra);
551 o1 = helper_le_ldq_mmu(env, addr + 8, oi1, ra);
552 oldv = int128_make128(o0, o1);
553
554 success = int128_eq(oldv, cmpv);
555 if (success) {
556 helper_le_stq_mmu(env, addr + 0, int128_getlo(newv), oi1, ra);
557 helper_le_stq_mmu(env, addr + 8, int128_gethi(newv), oi1, ra);
558 }
559#endif
560 }
561
562 return !success;
563}
564
565uint64_t HELPER(paired_cmpxchg64_le)(CPUARMState *env, uint64_t addr,
566 uint64_t new_lo, uint64_t new_hi)
567{
568 return do_paired_cmpxchg64_le(env, addr, new_lo, new_hi, false, GETPC());
569}
570
571uint64_t HELPER(paired_cmpxchg64_le_parallel)(CPUARMState *env, uint64_t addr,
572 uint64_t new_lo, uint64_t new_hi)
573{
574 return do_paired_cmpxchg64_le(env, addr, new_lo, new_hi, true, GETPC());
575}
576
577static uint64_t do_paired_cmpxchg64_be(CPUARMState *env, uint64_t addr,
578 uint64_t new_lo, uint64_t new_hi,
579 bool parallel, uintptr_t ra)
580{
581 Int128 oldv, cmpv, newv;
582 bool success;
583
584
585
586
587 cmpv = int128_make128(env->exclusive_high, env->exclusive_val);
588 newv = int128_make128(new_hi, new_lo);
589
590 if (parallel) {
591#ifndef CONFIG_ATOMIC128
592 cpu_loop_exit_atomic(ENV_GET_CPU(env), ra);
593#else
594 int mem_idx = cpu_mmu_index(env, false);
595 TCGMemOpIdx oi = make_memop_idx(MO_BEQ | MO_ALIGN_16, mem_idx);
596 oldv = helper_atomic_cmpxchgo_be_mmu(env, addr, cmpv, newv, oi, ra);
597 success = int128_eq(oldv, cmpv);
598#endif
599 } else {
600 uint64_t o0, o1;
601
602#ifdef CONFIG_USER_ONLY
603
604 uint64_t *haddr = g2h(addr);
605
606 helper_retaddr = ra;
607 o1 = ldq_be_p(haddr + 0);
608 o0 = ldq_be_p(haddr + 1);
609 oldv = int128_make128(o0, o1);
610
611 success = int128_eq(oldv, cmpv);
612 if (success) {
613 stq_be_p(haddr + 0, int128_gethi(newv));
614 stq_be_p(haddr + 1, int128_getlo(newv));
615 }
616 helper_retaddr = 0;
617#else
618 int mem_idx = cpu_mmu_index(env, false);
619 TCGMemOpIdx oi0 = make_memop_idx(MO_BEQ | MO_ALIGN_16, mem_idx);
620 TCGMemOpIdx oi1 = make_memop_idx(MO_BEQ, mem_idx);
621
622 o1 = helper_be_ldq_mmu(env, addr + 0, oi0, ra);
623 o0 = helper_be_ldq_mmu(env, addr + 8, oi1, ra);
624 oldv = int128_make128(o0, o1);
625
626 success = int128_eq(oldv, cmpv);
627 if (success) {
628 helper_be_stq_mmu(env, addr + 0, int128_gethi(newv), oi1, ra);
629 helper_be_stq_mmu(env, addr + 8, int128_getlo(newv), oi1, ra);
630 }
631#endif
632 }
633
634 return !success;
635}
636
637uint64_t HELPER(paired_cmpxchg64_be)(CPUARMState *env, uint64_t addr,
638 uint64_t new_lo, uint64_t new_hi)
639{
640 return do_paired_cmpxchg64_be(env, addr, new_lo, new_hi, false, GETPC());
641}
642
643uint64_t HELPER(paired_cmpxchg64_be_parallel)(CPUARMState *env, uint64_t addr,
644 uint64_t new_lo, uint64_t new_hi)
645{
646 return do_paired_cmpxchg64_be(env, addr, new_lo, new_hi, true, GETPC());
647}
648
649
650
651
652
653#define ADVSIMD_HELPER(name, suffix) HELPER(glue(glue(advsimd_, name), suffix))
654
655#define ADVSIMD_HALFOP(name) \
656float16 ADVSIMD_HELPER(name, h)(float16 a, float16 b, void *fpstp) \
657{ \
658 float_status *fpst = fpstp; \
659 return float16_ ## name(a, b, fpst); \
660}
661
662ADVSIMD_HALFOP(add)
663ADVSIMD_HALFOP(sub)
664ADVSIMD_HALFOP(mul)
665ADVSIMD_HALFOP(div)
666ADVSIMD_HALFOP(min)
667ADVSIMD_HALFOP(max)
668ADVSIMD_HALFOP(minnum)
669ADVSIMD_HALFOP(maxnum)
670
671#define ADVSIMD_TWOHALFOP(name) \
672uint32_t ADVSIMD_HELPER(name, 2h)(uint32_t two_a, uint32_t two_b, void *fpstp) \
673{ \
674 float16 a1, a2, b1, b2; \
675 uint32_t r1, r2; \
676 float_status *fpst = fpstp; \
677 a1 = extract32(two_a, 0, 16); \
678 a2 = extract32(two_a, 16, 16); \
679 b1 = extract32(two_b, 0, 16); \
680 b2 = extract32(two_b, 16, 16); \
681 r1 = float16_ ## name(a1, b1, fpst); \
682 r2 = float16_ ## name(a2, b2, fpst); \
683 return deposit32(r1, 16, 16, r2); \
684}
685
686ADVSIMD_TWOHALFOP(add)
687ADVSIMD_TWOHALFOP(sub)
688ADVSIMD_TWOHALFOP(mul)
689ADVSIMD_TWOHALFOP(div)
690ADVSIMD_TWOHALFOP(min)
691ADVSIMD_TWOHALFOP(max)
692ADVSIMD_TWOHALFOP(minnum)
693ADVSIMD_TWOHALFOP(maxnum)
694
695
696static float16 float16_mulx(float16 a, float16 b, void *fpstp)
697{
698 float_status *fpst = fpstp;
699
700 a = float16_squash_input_denormal(a, fpst);
701 b = float16_squash_input_denormal(b, fpst);
702
703 if ((float16_is_zero(a) && float16_is_infinity(b)) ||
704 (float16_is_infinity(a) && float16_is_zero(b))) {
705
706 return make_float16((1U << 14) |
707 ((float16_val(a) ^ float16_val(b)) & (1U << 15)));
708 }
709 return float16_mul(a, b, fpst);
710}
711
712ADVSIMD_HALFOP(mulx)
713ADVSIMD_TWOHALFOP(mulx)
714
715
716float16 HELPER(advsimd_muladdh)(float16 a, float16 b, float16 c, void *fpstp)
717{
718 float_status *fpst = fpstp;
719 return float16_muladd(a, b, c, 0, fpst);
720}
721
722uint32_t HELPER(advsimd_muladd2h)(uint32_t two_a, uint32_t two_b,
723 uint32_t two_c, void *fpstp)
724{
725 float_status *fpst = fpstp;
726 float16 a1, a2, b1, b2, c1, c2;
727 uint32_t r1, r2;
728 a1 = extract32(two_a, 0, 16);
729 a2 = extract32(two_a, 16, 16);
730 b1 = extract32(two_b, 0, 16);
731 b2 = extract32(two_b, 16, 16);
732 c1 = extract32(two_c, 0, 16);
733 c2 = extract32(two_c, 16, 16);
734 r1 = float16_muladd(a1, b1, c1, 0, fpst);
735 r2 = float16_muladd(a2, b2, c2, 0, fpst);
736 return deposit32(r1, 16, 16, r2);
737}
738
739
740
741
742
743
744
745#define ADVSIMD_CMPRES(test) (test) ? 0xffff : 0
746
747uint32_t HELPER(advsimd_ceq_f16)(float16 a, float16 b, void *fpstp)
748{
749 float_status *fpst = fpstp;
750 int compare = float16_compare_quiet(a, b, fpst);
751 return ADVSIMD_CMPRES(compare == float_relation_equal);
752}
753
754uint32_t HELPER(advsimd_cge_f16)(float16 a, float16 b, void *fpstp)
755{
756 float_status *fpst = fpstp;
757 int compare = float16_compare(a, b, fpst);
758 return ADVSIMD_CMPRES(compare == float_relation_greater ||
759 compare == float_relation_equal);
760}
761
762uint32_t HELPER(advsimd_cgt_f16)(float16 a, float16 b, void *fpstp)
763{
764 float_status *fpst = fpstp;
765 int compare = float16_compare(a, b, fpst);
766 return ADVSIMD_CMPRES(compare == float_relation_greater);
767}
768
769uint32_t HELPER(advsimd_acge_f16)(float16 a, float16 b, void *fpstp)
770{
771 float_status *fpst = fpstp;
772 float16 f0 = float16_abs(a);
773 float16 f1 = float16_abs(b);
774 int compare = float16_compare(f0, f1, fpst);
775 return ADVSIMD_CMPRES(compare == float_relation_greater ||
776 compare == float_relation_equal);
777}
778
779uint32_t HELPER(advsimd_acgt_f16)(float16 a, float16 b, void *fpstp)
780{
781 float_status *fpst = fpstp;
782 float16 f0 = float16_abs(a);
783 float16 f1 = float16_abs(b);
784 int compare = float16_compare(f0, f1, fpst);
785 return ADVSIMD_CMPRES(compare == float_relation_greater);
786}
787
788
789float16 HELPER(advsimd_rinth_exact)(float16 x, void *fp_status)
790{
791 return float16_round_to_int(x, fp_status);
792}
793
794float16 HELPER(advsimd_rinth)(float16 x, void *fp_status)
795{
796 int old_flags = get_float_exception_flags(fp_status), new_flags;
797 float16 ret;
798
799 ret = float16_round_to_int(x, fp_status);
800
801
802 if (!(old_flags & float_flag_inexact)) {
803 new_flags = get_float_exception_flags(fp_status);
804 set_float_exception_flags(new_flags & ~float_flag_inexact, fp_status);
805 }
806
807 return ret;
808}
809
810
811
812
813
814
815
816
817
818uint32_t HELPER(advsimd_f16tosinth)(float16 a, void *fpstp)
819{
820 float_status *fpst = fpstp;
821
822
823 if (float16_is_any_nan(a)) {
824 float_raise(float_flag_invalid, fpst);
825 return 0;
826 }
827 return float16_to_int16(a, fpst);
828}
829
830uint32_t HELPER(advsimd_f16touinth)(float16 a, void *fpstp)
831{
832 float_status *fpst = fpstp;
833
834
835 if (float16_is_any_nan(a)) {
836 float_raise(float_flag_invalid, fpst);
837 return 0;
838 }
839 return float16_to_uint16(a, fpst);
840}
841
842
843
844
845
846float16 HELPER(sqrt_f16)(float16 a, void *fpstp)
847{
848 float_status *s = fpstp;
849
850 return float16_sqrt(a, s);
851}
852
853
854