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)(uint32_t x, uint32_t 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)(uint32_t x, uint32_t 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
217uint32_t HELPER(recpsf_f16)(uint32_t a, uint32_t 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
262uint32_t HELPER(rsqrtsf_f16)(uint32_t a, uint32_t 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
369uint32_t HELPER(frecpx_f16)(uint32_t 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_silence_nan(a, fpst);
380 }
381 if (fpst->default_nan_mode) {
382 nan = float16_default_nan(fpst);
383 }
384 return nan;
385 }
386
387 a = float16_squash_input_denormal(a, fpst);
388
389 val16 = float16_val(a);
390 sbit = 0x8000 & val16;
391 exp = extract32(val16, 10, 5);
392
393 if (exp == 0) {
394 return make_float16(deposit32(sbit, 10, 5, 0x1e));
395 } else {
396 return make_float16(deposit32(sbit, 10, 5, ~exp));
397 }
398}
399
400float32 HELPER(frecpx_f32)(float32 a, void *fpstp)
401{
402 float_status *fpst = fpstp;
403 uint32_t val32, sbit;
404 int32_t exp;
405
406 if (float32_is_any_nan(a)) {
407 float32 nan = a;
408 if (float32_is_signaling_nan(a, fpst)) {
409 float_raise(float_flag_invalid, fpst);
410 nan = float32_silence_nan(a, fpst);
411 }
412 if (fpst->default_nan_mode) {
413 nan = float32_default_nan(fpst);
414 }
415 return nan;
416 }
417
418 a = float32_squash_input_denormal(a, fpst);
419
420 val32 = float32_val(a);
421 sbit = 0x80000000ULL & val32;
422 exp = extract32(val32, 23, 8);
423
424 if (exp == 0) {
425 return make_float32(sbit | (0xfe << 23));
426 } else {
427 return make_float32(sbit | (~exp & 0xff) << 23);
428 }
429}
430
431float64 HELPER(frecpx_f64)(float64 a, void *fpstp)
432{
433 float_status *fpst = fpstp;
434 uint64_t val64, sbit;
435 int64_t exp;
436
437 if (float64_is_any_nan(a)) {
438 float64 nan = a;
439 if (float64_is_signaling_nan(a, fpst)) {
440 float_raise(float_flag_invalid, fpst);
441 nan = float64_silence_nan(a, fpst);
442 }
443 if (fpst->default_nan_mode) {
444 nan = float64_default_nan(fpst);
445 }
446 return nan;
447 }
448
449 a = float64_squash_input_denormal(a, fpst);
450
451 val64 = float64_val(a);
452 sbit = 0x8000000000000000ULL & val64;
453 exp = extract64(float64_val(a), 52, 11);
454
455 if (exp == 0) {
456 return make_float64(sbit | (0x7feULL << 52));
457 } else {
458 return make_float64(sbit | (~exp & 0x7ffULL) << 52);
459 }
460}
461
462float32 HELPER(fcvtx_f64_to_f32)(float64 a, CPUARMState *env)
463{
464
465
466
467 float32 r;
468 float_status *fpst = &env->vfp.fp_status;
469 float_status tstat = *fpst;
470 int exflags;
471
472 set_float_rounding_mode(float_round_to_zero, &tstat);
473 set_float_exception_flags(0, &tstat);
474 r = float64_to_float32(a, &tstat);
475 exflags = get_float_exception_flags(&tstat);
476 if (exflags & float_flag_inexact) {
477 r = make_float32(float32_val(r) | 1);
478 }
479 exflags |= get_float_exception_flags(fpst);
480 set_float_exception_flags(exflags, fpst);
481 return r;
482}
483
484
485
486
487
488
489
490
491
492uint64_t HELPER(crc32_64)(uint64_t acc, uint64_t val, uint32_t bytes)
493{
494 uint8_t buf[8];
495
496 stq_le_p(buf, val);
497
498
499 return crc32(acc ^ 0xffffffff, buf, bytes) ^ 0xffffffff;
500}
501
502uint64_t HELPER(crc32c_64)(uint64_t acc, uint64_t val, uint32_t bytes)
503{
504 uint8_t buf[8];
505
506 stq_le_p(buf, val);
507
508
509 return crc32c(acc, buf, bytes) ^ 0xffffffff;
510}
511
512
513static uint64_t do_paired_cmpxchg64_le(CPUARMState *env, uint64_t addr,
514 uint64_t new_lo, uint64_t new_hi,
515 bool parallel, uintptr_t ra)
516{
517 Int128 oldv, cmpv, newv;
518 bool success;
519
520 cmpv = int128_make128(env->exclusive_val, env->exclusive_high);
521 newv = int128_make128(new_lo, new_hi);
522
523 if (parallel) {
524#ifndef CONFIG_ATOMIC128
525 cpu_loop_exit_atomic(ENV_GET_CPU(env), ra);
526#else
527 int mem_idx = cpu_mmu_index(env, false);
528 TCGMemOpIdx oi = make_memop_idx(MO_LEQ | MO_ALIGN_16, mem_idx);
529 oldv = helper_atomic_cmpxchgo_le_mmu(env, addr, cmpv, newv, oi, ra);
530 success = int128_eq(oldv, cmpv);
531#endif
532 } else {
533 uint64_t o0, o1;
534
535#ifdef CONFIG_USER_ONLY
536
537 uint64_t *haddr = g2h(addr);
538
539 helper_retaddr = ra;
540 o0 = ldq_le_p(haddr + 0);
541 o1 = ldq_le_p(haddr + 1);
542 oldv = int128_make128(o0, o1);
543
544 success = int128_eq(oldv, cmpv);
545 if (success) {
546 stq_le_p(haddr + 0, int128_getlo(newv));
547 stq_le_p(haddr + 1, int128_gethi(newv));
548 }
549 helper_retaddr = 0;
550#else
551 int mem_idx = cpu_mmu_index(env, false);
552 TCGMemOpIdx oi0 = make_memop_idx(MO_LEQ | MO_ALIGN_16, mem_idx);
553 TCGMemOpIdx oi1 = make_memop_idx(MO_LEQ, mem_idx);
554
555 o0 = helper_le_ldq_mmu(env, addr + 0, oi0, ra);
556 o1 = helper_le_ldq_mmu(env, addr + 8, oi1, ra);
557 oldv = int128_make128(o0, o1);
558
559 success = int128_eq(oldv, cmpv);
560 if (success) {
561 helper_le_stq_mmu(env, addr + 0, int128_getlo(newv), oi1, ra);
562 helper_le_stq_mmu(env, addr + 8, int128_gethi(newv), oi1, ra);
563 }
564#endif
565 }
566
567 return !success;
568}
569
570uint64_t HELPER(paired_cmpxchg64_le)(CPUARMState *env, uint64_t addr,
571 uint64_t new_lo, uint64_t new_hi)
572{
573 return do_paired_cmpxchg64_le(env, addr, new_lo, new_hi, false, GETPC());
574}
575
576uint64_t HELPER(paired_cmpxchg64_le_parallel)(CPUARMState *env, uint64_t addr,
577 uint64_t new_lo, uint64_t new_hi)
578{
579 return do_paired_cmpxchg64_le(env, addr, new_lo, new_hi, true, GETPC());
580}
581
582static uint64_t do_paired_cmpxchg64_be(CPUARMState *env, uint64_t addr,
583 uint64_t new_lo, uint64_t new_hi,
584 bool parallel, uintptr_t ra)
585{
586 Int128 oldv, cmpv, newv;
587 bool success;
588
589
590
591
592 cmpv = int128_make128(env->exclusive_high, env->exclusive_val);
593 newv = int128_make128(new_hi, new_lo);
594
595 if (parallel) {
596#ifndef CONFIG_ATOMIC128
597 cpu_loop_exit_atomic(ENV_GET_CPU(env), ra);
598#else
599 int mem_idx = cpu_mmu_index(env, false);
600 TCGMemOpIdx oi = make_memop_idx(MO_BEQ | MO_ALIGN_16, mem_idx);
601 oldv = helper_atomic_cmpxchgo_be_mmu(env, addr, cmpv, newv, oi, ra);
602 success = int128_eq(oldv, cmpv);
603#endif
604 } else {
605 uint64_t o0, o1;
606
607#ifdef CONFIG_USER_ONLY
608
609 uint64_t *haddr = g2h(addr);
610
611 helper_retaddr = ra;
612 o1 = ldq_be_p(haddr + 0);
613 o0 = ldq_be_p(haddr + 1);
614 oldv = int128_make128(o0, o1);
615
616 success = int128_eq(oldv, cmpv);
617 if (success) {
618 stq_be_p(haddr + 0, int128_gethi(newv));
619 stq_be_p(haddr + 1, int128_getlo(newv));
620 }
621 helper_retaddr = 0;
622#else
623 int mem_idx = cpu_mmu_index(env, false);
624 TCGMemOpIdx oi0 = make_memop_idx(MO_BEQ | MO_ALIGN_16, mem_idx);
625 TCGMemOpIdx oi1 = make_memop_idx(MO_BEQ, mem_idx);
626
627 o1 = helper_be_ldq_mmu(env, addr + 0, oi0, ra);
628 o0 = helper_be_ldq_mmu(env, addr + 8, oi1, ra);
629 oldv = int128_make128(o0, o1);
630
631 success = int128_eq(oldv, cmpv);
632 if (success) {
633 helper_be_stq_mmu(env, addr + 0, int128_gethi(newv), oi1, ra);
634 helper_be_stq_mmu(env, addr + 8, int128_getlo(newv), oi1, ra);
635 }
636#endif
637 }
638
639 return !success;
640}
641
642uint64_t HELPER(paired_cmpxchg64_be)(CPUARMState *env, uint64_t addr,
643 uint64_t new_lo, uint64_t new_hi)
644{
645 return do_paired_cmpxchg64_be(env, addr, new_lo, new_hi, false, GETPC());
646}
647
648uint64_t HELPER(paired_cmpxchg64_be_parallel)(CPUARMState *env, uint64_t addr,
649 uint64_t new_lo, uint64_t new_hi)
650{
651 return do_paired_cmpxchg64_be(env, addr, new_lo, new_hi, true, GETPC());
652}
653
654
655void HELPER(casp_le_parallel)(CPUARMState *env, uint32_t rs, uint64_t addr,
656 uint64_t new_lo, uint64_t new_hi)
657{
658 uintptr_t ra = GETPC();
659#ifndef CONFIG_ATOMIC128
660 cpu_loop_exit_atomic(ENV_GET_CPU(env), ra);
661#else
662 Int128 oldv, cmpv, newv;
663
664 cmpv = int128_make128(env->xregs[rs], env->xregs[rs + 1]);
665 newv = int128_make128(new_lo, new_hi);
666
667 int mem_idx = cpu_mmu_index(env, false);
668 TCGMemOpIdx oi = make_memop_idx(MO_LEQ | MO_ALIGN_16, mem_idx);
669 oldv = helper_atomic_cmpxchgo_le_mmu(env, addr, cmpv, newv, oi, ra);
670
671 env->xregs[rs] = int128_getlo(oldv);
672 env->xregs[rs + 1] = int128_gethi(oldv);
673#endif
674}
675
676void HELPER(casp_be_parallel)(CPUARMState *env, uint32_t rs, uint64_t addr,
677 uint64_t new_hi, uint64_t new_lo)
678{
679 uintptr_t ra = GETPC();
680#ifndef CONFIG_ATOMIC128
681 cpu_loop_exit_atomic(ENV_GET_CPU(env), ra);
682#else
683 Int128 oldv, cmpv, newv;
684
685 cmpv = int128_make128(env->xregs[rs + 1], env->xregs[rs]);
686 newv = int128_make128(new_lo, new_hi);
687
688 int mem_idx = cpu_mmu_index(env, false);
689 TCGMemOpIdx oi = make_memop_idx(MO_LEQ | MO_ALIGN_16, mem_idx);
690 oldv = helper_atomic_cmpxchgo_be_mmu(env, addr, cmpv, newv, oi, ra);
691
692 env->xregs[rs + 1] = int128_getlo(oldv);
693 env->xregs[rs] = int128_gethi(oldv);
694#endif
695}
696
697
698
699
700
701#define ADVSIMD_HELPER(name, suffix) HELPER(glue(glue(advsimd_, name), suffix))
702
703#define ADVSIMD_HALFOP(name) \
704uint32_t ADVSIMD_HELPER(name, h)(uint32_t a, uint32_t b, void *fpstp) \
705{ \
706 float_status *fpst = fpstp; \
707 return float16_ ## name(a, b, fpst); \
708}
709
710ADVSIMD_HALFOP(add)
711ADVSIMD_HALFOP(sub)
712ADVSIMD_HALFOP(mul)
713ADVSIMD_HALFOP(div)
714ADVSIMD_HALFOP(min)
715ADVSIMD_HALFOP(max)
716ADVSIMD_HALFOP(minnum)
717ADVSIMD_HALFOP(maxnum)
718
719#define ADVSIMD_TWOHALFOP(name) \
720uint32_t ADVSIMD_HELPER(name, 2h)(uint32_t two_a, uint32_t two_b, void *fpstp) \
721{ \
722 float16 a1, a2, b1, b2; \
723 uint32_t r1, r2; \
724 float_status *fpst = fpstp; \
725 a1 = extract32(two_a, 0, 16); \
726 a2 = extract32(two_a, 16, 16); \
727 b1 = extract32(two_b, 0, 16); \
728 b2 = extract32(two_b, 16, 16); \
729 r1 = float16_ ## name(a1, b1, fpst); \
730 r2 = float16_ ## name(a2, b2, fpst); \
731 return deposit32(r1, 16, 16, r2); \
732}
733
734ADVSIMD_TWOHALFOP(add)
735ADVSIMD_TWOHALFOP(sub)
736ADVSIMD_TWOHALFOP(mul)
737ADVSIMD_TWOHALFOP(div)
738ADVSIMD_TWOHALFOP(min)
739ADVSIMD_TWOHALFOP(max)
740ADVSIMD_TWOHALFOP(minnum)
741ADVSIMD_TWOHALFOP(maxnum)
742
743
744static float16 float16_mulx(float16 a, float16 b, void *fpstp)
745{
746 float_status *fpst = fpstp;
747
748 a = float16_squash_input_denormal(a, fpst);
749 b = float16_squash_input_denormal(b, fpst);
750
751 if ((float16_is_zero(a) && float16_is_infinity(b)) ||
752 (float16_is_infinity(a) && float16_is_zero(b))) {
753
754 return make_float16((1U << 14) |
755 ((float16_val(a) ^ float16_val(b)) & (1U << 15)));
756 }
757 return float16_mul(a, b, fpst);
758}
759
760ADVSIMD_HALFOP(mulx)
761ADVSIMD_TWOHALFOP(mulx)
762
763
764uint32_t HELPER(advsimd_muladdh)(uint32_t a, uint32_t b, uint32_t c,
765 void *fpstp)
766{
767 float_status *fpst = fpstp;
768 return float16_muladd(a, b, c, 0, fpst);
769}
770
771uint32_t HELPER(advsimd_muladd2h)(uint32_t two_a, uint32_t two_b,
772 uint32_t two_c, void *fpstp)
773{
774 float_status *fpst = fpstp;
775 float16 a1, a2, b1, b2, c1, c2;
776 uint32_t r1, r2;
777 a1 = extract32(two_a, 0, 16);
778 a2 = extract32(two_a, 16, 16);
779 b1 = extract32(two_b, 0, 16);
780 b2 = extract32(two_b, 16, 16);
781 c1 = extract32(two_c, 0, 16);
782 c2 = extract32(two_c, 16, 16);
783 r1 = float16_muladd(a1, b1, c1, 0, fpst);
784 r2 = float16_muladd(a2, b2, c2, 0, fpst);
785 return deposit32(r1, 16, 16, r2);
786}
787
788
789
790
791
792
793
794#define ADVSIMD_CMPRES(test) (test) ? 0xffff : 0
795
796uint32_t HELPER(advsimd_ceq_f16)(uint32_t a, uint32_t b, void *fpstp)
797{
798 float_status *fpst = fpstp;
799 int compare = float16_compare_quiet(a, b, fpst);
800 return ADVSIMD_CMPRES(compare == float_relation_equal);
801}
802
803uint32_t HELPER(advsimd_cge_f16)(uint32_t a, uint32_t b, void *fpstp)
804{
805 float_status *fpst = fpstp;
806 int compare = float16_compare(a, b, fpst);
807 return ADVSIMD_CMPRES(compare == float_relation_greater ||
808 compare == float_relation_equal);
809}
810
811uint32_t HELPER(advsimd_cgt_f16)(uint32_t a, uint32_t b, void *fpstp)
812{
813 float_status *fpst = fpstp;
814 int compare = float16_compare(a, b, fpst);
815 return ADVSIMD_CMPRES(compare == float_relation_greater);
816}
817
818uint32_t HELPER(advsimd_acge_f16)(uint32_t a, uint32_t b, void *fpstp)
819{
820 float_status *fpst = fpstp;
821 float16 f0 = float16_abs(a);
822 float16 f1 = float16_abs(b);
823 int compare = float16_compare(f0, f1, fpst);
824 return ADVSIMD_CMPRES(compare == float_relation_greater ||
825 compare == float_relation_equal);
826}
827
828uint32_t HELPER(advsimd_acgt_f16)(uint32_t a, uint32_t b, void *fpstp)
829{
830 float_status *fpst = fpstp;
831 float16 f0 = float16_abs(a);
832 float16 f1 = float16_abs(b);
833 int compare = float16_compare(f0, f1, fpst);
834 return ADVSIMD_CMPRES(compare == float_relation_greater);
835}
836
837
838uint32_t HELPER(advsimd_rinth_exact)(uint32_t x, void *fp_status)
839{
840 return float16_round_to_int(x, fp_status);
841}
842
843uint32_t HELPER(advsimd_rinth)(uint32_t x, void *fp_status)
844{
845 int old_flags = get_float_exception_flags(fp_status), new_flags;
846 float16 ret;
847
848 ret = float16_round_to_int(x, fp_status);
849
850
851 if (!(old_flags & float_flag_inexact)) {
852 new_flags = get_float_exception_flags(fp_status);
853 set_float_exception_flags(new_flags & ~float_flag_inexact, fp_status);
854 }
855
856 return ret;
857}
858
859
860
861
862
863
864
865
866
867uint32_t HELPER(advsimd_f16tosinth)(uint32_t a, void *fpstp)
868{
869 float_status *fpst = fpstp;
870
871
872 if (float16_is_any_nan(a)) {
873 float_raise(float_flag_invalid, fpst);
874 return 0;
875 }
876 return float16_to_int16(a, fpst);
877}
878
879uint32_t HELPER(advsimd_f16touinth)(uint32_t a, void *fpstp)
880{
881 float_status *fpst = fpstp;
882
883
884 if (float16_is_any_nan(a)) {
885 float_raise(float_flag_invalid, fpst);
886 return 0;
887 }
888 return float16_to_uint16(a, fpst);
889}
890
891
892
893
894
895uint32_t HELPER(sqrt_f16)(uint32_t a, void *fpstp)
896{
897 float_status *s = fpstp;
898
899 return float16_sqrt(a, s);
900}
901
902
903