2
* ARM NEON vector operations.
4
* Copyright (c) 2007, 2008 CodeSourcery.
5
* Written by Paul Brook
7
* This code is licensed under the GNU GPL v2.
16
#define SIGNBIT (uint32_t)0x80000000
17
#define SIGNBIT64 ((uint64_t)1 << 63)
19
#define SET_QC() env->vfp.xregs[ARM_VFP_FPSCR] = CPSR_Q
21
#define NEON_TYPE1(name, type) \
26
#ifdef HOST_WORDS_BIGENDIAN
27
#define NEON_TYPE2(name, type) \
33
#define NEON_TYPE4(name, type) \
42
#define NEON_TYPE2(name, type) \
48
#define NEON_TYPE4(name, type) \
58
NEON_TYPE4(s8, int8_t)
59
NEON_TYPE4(u8, uint8_t)
60
NEON_TYPE2(s16, int16_t)
61
NEON_TYPE2(u16, uint16_t)
62
NEON_TYPE1(s32, int32_t)
63
NEON_TYPE1(u32, uint32_t)
68
/* Copy from a uint32_t to a vector structure type. */
69
#define NEON_UNPACK(vtype, dest, val) do { \
78
/* Copy from a vector structure type to a uint32_t. */
79
#define NEON_PACK(vtype, dest, val) do { \
89
NEON_FN(vdest.v1, vsrc1.v1, vsrc2.v1);
91
NEON_FN(vdest.v1, vsrc1.v1, vsrc2.v1); \
92
NEON_FN(vdest.v2, vsrc1.v2, vsrc2.v2);
94
NEON_FN(vdest.v1, vsrc1.v1, vsrc2.v1); \
95
NEON_FN(vdest.v2, vsrc1.v2, vsrc2.v2); \
96
NEON_FN(vdest.v3, vsrc1.v3, vsrc2.v3); \
97
NEON_FN(vdest.v4, vsrc1.v4, vsrc2.v4);
99
#define NEON_VOP_BODY(vtype, n) \
105
NEON_UNPACK(vtype, vsrc1, arg1); \
106
NEON_UNPACK(vtype, vsrc2, arg2); \
108
NEON_PACK(vtype, res, vdest); \
112
#define NEON_VOP(name, vtype, n) \
113
uint32_t HELPER(glue(neon_,name))(uint32_t arg1, uint32_t arg2) \
114
NEON_VOP_BODY(vtype, n)
116
#define NEON_VOP_ENV(name, vtype, n) \
117
uint32_t HELPER(glue(neon_,name))(CPUState *env, uint32_t arg1, uint32_t arg2) \
118
NEON_VOP_BODY(vtype, n)
120
/* Pairwise operations. */
121
/* For 32-bit elements each segment only contains a single element, so
122
the elementwise and pairwise operations are the same. */
124
NEON_FN(vdest.v1, vsrc1.v1, vsrc1.v2); \
125
NEON_FN(vdest.v2, vsrc2.v1, vsrc2.v2);
127
NEON_FN(vdest.v1, vsrc1.v1, vsrc1.v2); \
128
NEON_FN(vdest.v2, vsrc1.v3, vsrc1.v4); \
129
NEON_FN(vdest.v3, vsrc2.v1, vsrc2.v2); \
130
NEON_FN(vdest.v4, vsrc2.v3, vsrc2.v4); \
132
#define NEON_POP(name, vtype, n) \
133
uint32_t HELPER(glue(neon_,name))(uint32_t arg1, uint32_t arg2) \
139
NEON_UNPACK(vtype, vsrc1, arg1); \
140
NEON_UNPACK(vtype, vsrc2, arg2); \
142
NEON_PACK(vtype, res, vdest); \
146
/* Unary operators. */
147
#define NEON_VOP1(name, vtype, n) \
148
uint32_t HELPER(glue(neon_,name))(uint32_t arg) \
152
NEON_UNPACK(vtype, vsrc1, arg); \
154
NEON_PACK(vtype, arg, vdest); \
159
#define NEON_USAT(dest, src1, src2, type) do { \
160
uint32_t tmp = (uint32_t)src1 + (uint32_t)src2; \
161
if (tmp != (type)tmp) { \
167
#define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint8_t)
168
NEON_VOP_ENV(qadd_u8, neon_u8, 4)
170
#define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint16_t)
171
NEON_VOP_ENV(qadd_u16, neon_u16, 2)
175
uint32_t HELPER(neon_qadd_u32)(CPUState *env, uint32_t a, uint32_t b)
177
uint32_t res = a + b;
185
uint64_t HELPER(neon_qadd_u64)(CPUState *env, uint64_t src1, uint64_t src2)
197
#define NEON_SSAT(dest, src1, src2, type) do { \
198
int32_t tmp = (uint32_t)src1 + (uint32_t)src2; \
199
if (tmp != (type)tmp) { \
202
tmp = (1 << (sizeof(type) * 8 - 1)) - 1; \
204
tmp = 1 << (sizeof(type) * 8 - 1); \
209
#define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int8_t)
210
NEON_VOP_ENV(qadd_s8, neon_s8, 4)
212
#define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int16_t)
213
NEON_VOP_ENV(qadd_s16, neon_s16, 2)
217
uint32_t HELPER(neon_qadd_s32)(CPUState *env, uint32_t a, uint32_t b)
219
uint32_t res = a + b;
220
if (((res ^ a) & SIGNBIT) && !((a ^ b) & SIGNBIT)) {
222
res = ~(((int32_t)a >> 31) ^ SIGNBIT);
227
uint64_t HELPER(neon_qadd_s64)(CPUState *env, uint64_t src1, uint64_t src2)
232
if (((res ^ src1) & SIGNBIT64) && !((src1 ^ src2) & SIGNBIT64)) {
234
res = ((int64_t)src1 >> 63) ^ ~SIGNBIT64;
239
#define NEON_USAT(dest, src1, src2, type) do { \
240
uint32_t tmp = (uint32_t)src1 - (uint32_t)src2; \
241
if (tmp != (type)tmp) { \
247
#define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint8_t)
248
NEON_VOP_ENV(qsub_u8, neon_u8, 4)
250
#define NEON_FN(dest, src1, src2) NEON_USAT(dest, src1, src2, uint16_t)
251
NEON_VOP_ENV(qsub_u16, neon_u16, 2)
255
uint32_t HELPER(neon_qsub_u32)(CPUState *env, uint32_t a, uint32_t b)
257
uint32_t res = a - b;
265
uint64_t HELPER(neon_qsub_u64)(CPUState *env, uint64_t src1, uint64_t src2)
278
#define NEON_SSAT(dest, src1, src2, type) do { \
279
int32_t tmp = (uint32_t)src1 - (uint32_t)src2; \
280
if (tmp != (type)tmp) { \
283
tmp = (1 << (sizeof(type) * 8 - 1)) - 1; \
285
tmp = 1 << (sizeof(type) * 8 - 1); \
290
#define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int8_t)
291
NEON_VOP_ENV(qsub_s8, neon_s8, 4)
293
#define NEON_FN(dest, src1, src2) NEON_SSAT(dest, src1, src2, int16_t)
294
NEON_VOP_ENV(qsub_s16, neon_s16, 2)
298
uint32_t HELPER(neon_qsub_s32)(CPUState *env, uint32_t a, uint32_t b)
300
uint32_t res = a - b;
301
if (((res ^ a) & SIGNBIT) && ((a ^ b) & SIGNBIT)) {
303
res = ~(((int32_t)a >> 31) ^ SIGNBIT);
308
uint64_t HELPER(neon_qsub_s64)(CPUState *env, uint64_t src1, uint64_t src2)
313
if (((res ^ src1) & SIGNBIT64) && ((src1 ^ src2) & SIGNBIT64)) {
315
res = ((int64_t)src1 >> 63) ^ ~SIGNBIT64;
320
#define NEON_FN(dest, src1, src2) dest = (src1 + src2) >> 1
321
NEON_VOP(hadd_s8, neon_s8, 4)
322
NEON_VOP(hadd_u8, neon_u8, 4)
323
NEON_VOP(hadd_s16, neon_s16, 2)
324
NEON_VOP(hadd_u16, neon_u16, 2)
327
int32_t HELPER(neon_hadd_s32)(int32_t src1, int32_t src2)
331
dest = (src1 >> 1) + (src2 >> 1);
337
uint32_t HELPER(neon_hadd_u32)(uint32_t src1, uint32_t src2)
341
dest = (src1 >> 1) + (src2 >> 1);
347
#define NEON_FN(dest, src1, src2) dest = (src1 + src2 + 1) >> 1
348
NEON_VOP(rhadd_s8, neon_s8, 4)
349
NEON_VOP(rhadd_u8, neon_u8, 4)
350
NEON_VOP(rhadd_s16, neon_s16, 2)
351
NEON_VOP(rhadd_u16, neon_u16, 2)
354
int32_t HELPER(neon_rhadd_s32)(int32_t src1, int32_t src2)
358
dest = (src1 >> 1) + (src2 >> 1);
359
if ((src1 | src2) & 1)
364
uint32_t HELPER(neon_rhadd_u32)(uint32_t src1, uint32_t src2)
368
dest = (src1 >> 1) + (src2 >> 1);
369
if ((src1 | src2) & 1)
374
#define NEON_FN(dest, src1, src2) dest = (src1 - src2) >> 1
375
NEON_VOP(hsub_s8, neon_s8, 4)
376
NEON_VOP(hsub_u8, neon_u8, 4)
377
NEON_VOP(hsub_s16, neon_s16, 2)
378
NEON_VOP(hsub_u16, neon_u16, 2)
381
int32_t HELPER(neon_hsub_s32)(int32_t src1, int32_t src2)
385
dest = (src1 >> 1) - (src2 >> 1);
386
if ((~src1) & src2 & 1)
391
uint32_t HELPER(neon_hsub_u32)(uint32_t src1, uint32_t src2)
395
dest = (src1 >> 1) - (src2 >> 1);
396
if ((~src1) & src2 & 1)
401
#define NEON_FN(dest, src1, src2) dest = (src1 > src2) ? ~0 : 0
402
NEON_VOP(cgt_s8, neon_s8, 4)
403
NEON_VOP(cgt_u8, neon_u8, 4)
404
NEON_VOP(cgt_s16, neon_s16, 2)
405
NEON_VOP(cgt_u16, neon_u16, 2)
406
NEON_VOP(cgt_s32, neon_s32, 1)
407
NEON_VOP(cgt_u32, neon_u32, 1)
410
#define NEON_FN(dest, src1, src2) dest = (src1 >= src2) ? ~0 : 0
411
NEON_VOP(cge_s8, neon_s8, 4)
412
NEON_VOP(cge_u8, neon_u8, 4)
413
NEON_VOP(cge_s16, neon_s16, 2)
414
NEON_VOP(cge_u16, neon_u16, 2)
415
NEON_VOP(cge_s32, neon_s32, 1)
416
NEON_VOP(cge_u32, neon_u32, 1)
419
#define NEON_FN(dest, src1, src2) dest = (src1 < src2) ? src1 : src2
420
NEON_VOP(min_s8, neon_s8, 4)
421
NEON_VOP(min_u8, neon_u8, 4)
422
NEON_VOP(min_s16, neon_s16, 2)
423
NEON_VOP(min_u16, neon_u16, 2)
424
NEON_VOP(min_s32, neon_s32, 1)
425
NEON_VOP(min_u32, neon_u32, 1)
426
NEON_POP(pmin_s8, neon_s8, 4)
427
NEON_POP(pmin_u8, neon_u8, 4)
428
NEON_POP(pmin_s16, neon_s16, 2)
429
NEON_POP(pmin_u16, neon_u16, 2)
432
#define NEON_FN(dest, src1, src2) dest = (src1 > src2) ? src1 : src2
433
NEON_VOP(max_s8, neon_s8, 4)
434
NEON_VOP(max_u8, neon_u8, 4)
435
NEON_VOP(max_s16, neon_s16, 2)
436
NEON_VOP(max_u16, neon_u16, 2)
437
NEON_VOP(max_s32, neon_s32, 1)
438
NEON_VOP(max_u32, neon_u32, 1)
439
NEON_POP(pmax_s8, neon_s8, 4)
440
NEON_POP(pmax_u8, neon_u8, 4)
441
NEON_POP(pmax_s16, neon_s16, 2)
442
NEON_POP(pmax_u16, neon_u16, 2)
445
#define NEON_FN(dest, src1, src2) \
446
dest = (src1 > src2) ? (src1 - src2) : (src2 - src1)
447
NEON_VOP(abd_s8, neon_s8, 4)
448
NEON_VOP(abd_u8, neon_u8, 4)
449
NEON_VOP(abd_s16, neon_s16, 2)
450
NEON_VOP(abd_u16, neon_u16, 2)
451
NEON_VOP(abd_s32, neon_s32, 1)
452
NEON_VOP(abd_u32, neon_u32, 1)
455
#define NEON_FN(dest, src1, src2) do { \
457
tmp = (int8_t)src2; \
458
if (tmp >= (ssize_t)sizeof(src1) * 8 || \
459
tmp <= -(ssize_t)sizeof(src1) * 8) { \
461
} else if (tmp < 0) { \
462
dest = src1 >> -tmp; \
464
dest = src1 << tmp; \
466
NEON_VOP(shl_u8, neon_u8, 4)
467
NEON_VOP(shl_u16, neon_u16, 2)
468
NEON_VOP(shl_u32, neon_u32, 1)
471
uint64_t HELPER(neon_shl_u64)(uint64_t val, uint64_t shiftop)
473
int8_t shift = (int8_t)shiftop;
474
if (shift >= 64 || shift <= -64) {
476
} else if (shift < 0) {
484
#define NEON_FN(dest, src1, src2) do { \
486
tmp = (int8_t)src2; \
487
if (tmp >= (ssize_t)sizeof(src1) * 8) { \
489
} else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \
490
dest = src1 >> (sizeof(src1) * 8 - 1); \
491
} else if (tmp < 0) { \
492
dest = src1 >> -tmp; \
494
dest = src1 << tmp; \
496
NEON_VOP(shl_s8, neon_s8, 4)
497
NEON_VOP(shl_s16, neon_s16, 2)
498
NEON_VOP(shl_s32, neon_s32, 1)
501
uint64_t HELPER(neon_shl_s64)(uint64_t valop, uint64_t shiftop)
503
int8_t shift = (int8_t)shiftop;
507
} else if (shift <= -64) {
509
} else if (shift < 0) {
517
#define NEON_FN(dest, src1, src2) do { \
519
tmp = (int8_t)src2; \
520
if ((tmp >= (ssize_t)sizeof(src1) * 8) \
521
|| (tmp <= -(ssize_t)sizeof(src1) * 8)) { \
523
} else if (tmp < 0) { \
524
dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \
526
dest = src1 << tmp; \
528
NEON_VOP(rshl_s8, neon_s8, 4)
529
NEON_VOP(rshl_s16, neon_s16, 2)
532
/* The addition of the rounding constant may overflow, so we use an
533
* intermediate 64 bits accumulator. */
534
uint32_t HELPER(neon_rshl_s32)(uint32_t valop, uint32_t shiftop)
537
int32_t val = (int32_t)valop;
538
int8_t shift = (int8_t)shiftop;
539
if ((shift >= 32) || (shift <= -32)) {
541
} else if (shift < 0) {
542
int64_t big_dest = ((int64_t)val + (1 << (-1 - shift)));
543
dest = big_dest >> -shift;
550
/* Handling addition overflow with 64 bits inputs values is more
551
* tricky than with 32 bits values. */
552
uint64_t HELPER(neon_rshl_s64)(uint64_t valop, uint64_t shiftop)
554
int8_t shift = (int8_t)shiftop;
556
if ((shift >= 64) || (shift <= -64)) {
558
} else if (shift < 0) {
559
val >>= (-shift - 1);
560
if (val == INT64_MAX) {
561
/* In this case, it means that the rounding constant is 1,
562
* and the addition would overflow. Return the actual
563
* result directly. */
564
val = 0x4000000000000000LL;
575
#define NEON_FN(dest, src1, src2) do { \
577
tmp = (int8_t)src2; \
578
if (tmp >= (ssize_t)sizeof(src1) * 8 || \
579
tmp < -(ssize_t)sizeof(src1) * 8) { \
581
} else if (tmp == -(ssize_t)sizeof(src1) * 8) { \
582
dest = src1 >> (-tmp - 1); \
583
} else if (tmp < 0) { \
584
dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \
586
dest = src1 << tmp; \
588
NEON_VOP(rshl_u8, neon_u8, 4)
589
NEON_VOP(rshl_u16, neon_u16, 2)
592
/* The addition of the rounding constant may overflow, so we use an
593
* intermediate 64 bits accumulator. */
594
uint32_t HELPER(neon_rshl_u32)(uint32_t val, uint32_t shiftop)
597
int8_t shift = (int8_t)shiftop;
598
if (shift >= 32 || shift < -32) {
600
} else if (shift == -32) {
602
} else if (shift < 0) {
603
uint64_t big_dest = ((uint64_t)val + (1 << (-1 - shift)));
604
dest = big_dest >> -shift;
611
/* Handling addition overflow with 64 bits inputs values is more
612
* tricky than with 32 bits values. */
613
uint64_t HELPER(neon_rshl_u64)(uint64_t val, uint64_t shiftop)
615
int8_t shift = (uint8_t)shiftop;
616
if (shift >= 64 || shift < -64) {
618
} else if (shift == -64) {
619
/* Rounding a 1-bit result just preserves that bit. */
621
} else if (shift < 0) {
622
val >>= (-shift - 1);
623
if (val == UINT64_MAX) {
624
/* In this case, it means that the rounding constant is 1,
625
* and the addition would overflow. Return the actual
626
* result directly. */
627
val = 0x8000000000000000ULL;
638
#define NEON_FN(dest, src1, src2) do { \
640
tmp = (int8_t)src2; \
641
if (tmp >= (ssize_t)sizeof(src1) * 8) { \
648
} else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \
650
} else if (tmp < 0) { \
651
dest = src1 >> -tmp; \
653
dest = src1 << tmp; \
654
if ((dest >> tmp) != src1) { \
659
NEON_VOP_ENV(qshl_u8, neon_u8, 4)
660
NEON_VOP_ENV(qshl_u16, neon_u16, 2)
661
NEON_VOP_ENV(qshl_u32, neon_u32, 1)
664
uint64_t HELPER(neon_qshl_u64)(CPUState *env, uint64_t val, uint64_t shiftop)
666
int8_t shift = (int8_t)shiftop;
672
} else if (shift <= -64) {
674
} else if (shift < 0) {
679
if ((val >> shift) != tmp) {
687
#define NEON_FN(dest, src1, src2) do { \
689
tmp = (int8_t)src2; \
690
if (tmp >= (ssize_t)sizeof(src1) * 8) { \
693
dest = (uint32_t)(1 << (sizeof(src1) * 8 - 1)); \
700
} else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \
702
} else if (tmp < 0) { \
703
dest = src1 >> -tmp; \
705
dest = src1 << tmp; \
706
if ((dest >> tmp) != src1) { \
708
dest = (uint32_t)(1 << (sizeof(src1) * 8 - 1)); \
714
NEON_VOP_ENV(qshl_s8, neon_s8, 4)
715
NEON_VOP_ENV(qshl_s16, neon_s16, 2)
716
NEON_VOP_ENV(qshl_s32, neon_s32, 1)
719
uint64_t HELPER(neon_qshl_s64)(CPUState *env, uint64_t valop, uint64_t shiftop)
721
int8_t shift = (uint8_t)shiftop;
726
val = (val >> 63) ^ ~SIGNBIT64;
728
} else if (shift <= -64) {
730
} else if (shift < 0) {
735
if ((val >> shift) != tmp) {
737
val = (tmp >> 63) ^ ~SIGNBIT64;
743
#define NEON_FN(dest, src1, src2) do { \
744
if (src1 & (1 << (sizeof(src1) * 8 - 1))) { \
749
tmp = (int8_t)src2; \
750
if (tmp >= (ssize_t)sizeof(src1) * 8) { \
757
} else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \
759
} else if (tmp < 0) { \
760
dest = src1 >> -tmp; \
762
dest = src1 << tmp; \
763
if ((dest >> tmp) != src1) { \
769
NEON_VOP_ENV(qshlu_s8, neon_u8, 4)
770
NEON_VOP_ENV(qshlu_s16, neon_u16, 2)
773
uint32_t HELPER(neon_qshlu_s32)(CPUState *env, uint32_t valop, uint32_t shiftop)
775
if ((int32_t)valop < 0) {
779
return helper_neon_qshl_u32(env, valop, shiftop);
782
uint64_t HELPER(neon_qshlu_s64)(CPUState *env, uint64_t valop, uint64_t shiftop)
784
if ((int64_t)valop < 0) {
788
return helper_neon_qshl_u64(env, valop, shiftop);
791
/* FIXME: This is wrong. */
792
#define NEON_FN(dest, src1, src2) do { \
794
tmp = (int8_t)src2; \
795
if (tmp >= (ssize_t)sizeof(src1) * 8) { \
802
} else if (tmp < -(ssize_t)sizeof(src1) * 8) { \
804
} else if (tmp == -(ssize_t)sizeof(src1) * 8) { \
805
dest = src1 >> (sizeof(src1) * 8 - 1); \
806
} else if (tmp < 0) { \
807
dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \
809
dest = src1 << tmp; \
810
if ((dest >> tmp) != src1) { \
815
NEON_VOP_ENV(qrshl_u8, neon_u8, 4)
816
NEON_VOP_ENV(qrshl_u16, neon_u16, 2)
819
/* The addition of the rounding constant may overflow, so we use an
820
* intermediate 64 bits accumulator. */
821
uint32_t HELPER(neon_qrshl_u32)(CPUState *env, uint32_t val, uint32_t shiftop)
824
int8_t shift = (int8_t)shiftop;
832
} else if (shift < -32) {
834
} else if (shift == -32) {
836
} else if (shift < 0) {
837
uint64_t big_dest = ((uint64_t)val + (1 << (-1 - shift)));
838
dest = big_dest >> -shift;
841
if ((dest >> shift) != val) {
849
/* Handling addition overflow with 64 bits inputs values is more
850
* tricky than with 32 bits values. */
851
uint64_t HELPER(neon_qrshl_u64)(CPUState *env, uint64_t val, uint64_t shiftop)
853
int8_t shift = (int8_t)shiftop;
859
} else if (shift < -64) {
861
} else if (shift == -64) {
863
} else if (shift < 0) {
864
val >>= (-shift - 1);
865
if (val == UINT64_MAX) {
866
/* In this case, it means that the rounding constant is 1,
867
* and the addition would overflow. Return the actual
868
* result directly. */
869
val = 0x8000000000000000ULL;
877
if ((val >> shift) != tmp) {
885
#define NEON_FN(dest, src1, src2) do { \
887
tmp = (int8_t)src2; \
888
if (tmp >= (ssize_t)sizeof(src1) * 8) { \
891
dest = (1 << (sizeof(src1) * 8 - 1)); \
898
} else if (tmp <= -(ssize_t)sizeof(src1) * 8) { \
900
} else if (tmp < 0) { \
901
dest = (src1 + (1 << (-1 - tmp))) >> -tmp; \
903
dest = src1 << tmp; \
904
if ((dest >> tmp) != src1) { \
906
dest = (uint32_t)(1 << (sizeof(src1) * 8 - 1)); \
912
NEON_VOP_ENV(qrshl_s8, neon_s8, 4)
913
NEON_VOP_ENV(qrshl_s16, neon_s16, 2)
916
/* The addition of the rounding constant may overflow, so we use an
917
* intermediate 64 bits accumulator. */
918
uint32_t HELPER(neon_qrshl_s32)(CPUState *env, uint32_t valop, uint32_t shiftop)
921
int32_t val = (int32_t)valop;
922
int8_t shift = (int8_t)shiftop;
926
dest = (val >> 31) ^ ~SIGNBIT;
930
} else if (shift <= -32) {
932
} else if (shift < 0) {
933
int64_t big_dest = ((int64_t)val + (1 << (-1 - shift)));
934
dest = big_dest >> -shift;
937
if ((dest >> shift) != val) {
939
dest = (val >> 31) ^ ~SIGNBIT;
945
/* Handling addition overflow with 64 bits inputs values is more
946
* tricky than with 32 bits values. */
947
uint64_t HELPER(neon_qrshl_s64)(CPUState *env, uint64_t valop, uint64_t shiftop)
949
int8_t shift = (uint8_t)shiftop;
955
val = (val >> 63) ^ ~SIGNBIT64;
957
} else if (shift <= -64) {
959
} else if (shift < 0) {
960
val >>= (-shift - 1);
961
if (val == INT64_MAX) {
962
/* In this case, it means that the rounding constant is 1,
963
* and the addition would overflow. Return the actual
964
* result directly. */
965
val = 0x4000000000000000ULL;
973
if ((val >> shift) != tmp) {
975
val = (tmp >> 63) ^ ~SIGNBIT64;
981
uint32_t HELPER(neon_add_u8)(uint32_t a, uint32_t b)
984
mask = (a ^ b) & 0x80808080u;
987
return (a + b) ^ mask;
990
uint32_t HELPER(neon_add_u16)(uint32_t a, uint32_t b)
993
mask = (a ^ b) & 0x80008000u;
996
return (a + b) ^ mask;
999
#define NEON_FN(dest, src1, src2) dest = src1 + src2
1000
NEON_POP(padd_u8, neon_u8, 4)
1001
NEON_POP(padd_u16, neon_u16, 2)
1004
#define NEON_FN(dest, src1, src2) dest = src1 - src2
1005
NEON_VOP(sub_u8, neon_u8, 4)
1006
NEON_VOP(sub_u16, neon_u16, 2)
1009
#define NEON_FN(dest, src1, src2) dest = src1 * src2
1010
NEON_VOP(mul_u8, neon_u8, 4)
1011
NEON_VOP(mul_u16, neon_u16, 2)
1014
/* Polynomial multiplication is like integer multiplication except the
1015
partial products are XORed, not added. */
1016
uint32_t HELPER(neon_mul_p8)(uint32_t op1, uint32_t op2)
1026
mask |= (0xff << 8);
1027
if (op1 & (1 << 16))
1028
mask |= (0xff << 16);
1029
if (op1 & (1 << 24))
1030
mask |= (0xff << 24);
1031
result ^= op2 & mask;
1032
op1 = (op1 >> 1) & 0x7f7f7f7f;
1033
op2 = (op2 << 1) & 0xfefefefe;
1038
uint64_t HELPER(neon_mull_p8)(uint32_t op1, uint32_t op2)
1040
uint64_t result = 0;
1042
uint64_t op2ex = op2;
1043
op2ex = (op2ex & 0xff) |
1044
((op2ex & 0xff00) << 8) |
1045
((op2ex & 0xff0000) << 16) |
1046
((op2ex & 0xff000000) << 24);
1052
if (op1 & (1 << 8)) {
1053
mask |= (0xffffU << 16);
1055
if (op1 & (1 << 16)) {
1056
mask |= (0xffffULL << 32);
1058
if (op1 & (1 << 24)) {
1059
mask |= (0xffffULL << 48);
1061
result ^= op2ex & mask;
1062
op1 = (op1 >> 1) & 0x7f7f7f7f;
1068
#define NEON_FN(dest, src1, src2) dest = (src1 & src2) ? -1 : 0
1069
NEON_VOP(tst_u8, neon_u8, 4)
1070
NEON_VOP(tst_u16, neon_u16, 2)
1071
NEON_VOP(tst_u32, neon_u32, 1)
1074
#define NEON_FN(dest, src1, src2) dest = (src1 == src2) ? -1 : 0
1075
NEON_VOP(ceq_u8, neon_u8, 4)
1076
NEON_VOP(ceq_u16, neon_u16, 2)
1077
NEON_VOP(ceq_u32, neon_u32, 1)
1080
#define NEON_FN(dest, src, dummy) dest = (src < 0) ? -src : src
1081
NEON_VOP1(abs_s8, neon_s8, 4)
1082
NEON_VOP1(abs_s16, neon_s16, 2)
1085
/* Count Leading Sign/Zero Bits. */
1086
static inline int do_clz8(uint8_t x)
1094
static inline int do_clz16(uint16_t x)
1097
for (n = 16; x; n--)
1102
#define NEON_FN(dest, src, dummy) dest = do_clz8(src)
1103
NEON_VOP1(clz_u8, neon_u8, 4)
1106
#define NEON_FN(dest, src, dummy) dest = do_clz16(src)
1107
NEON_VOP1(clz_u16, neon_u16, 2)
1110
#define NEON_FN(dest, src, dummy) dest = do_clz8((src < 0) ? ~src : src) - 1
1111
NEON_VOP1(cls_s8, neon_s8, 4)
1114
#define NEON_FN(dest, src, dummy) dest = do_clz16((src < 0) ? ~src : src) - 1
1115
NEON_VOP1(cls_s16, neon_s16, 2)
1118
uint32_t HELPER(neon_cls_s32)(uint32_t x)
1123
for (count = 32; x; count--)
1129
uint32_t HELPER(neon_cnt_u8)(uint32_t x)
1131
x = (x & 0x55555555) + ((x >> 1) & 0x55555555);
1132
x = (x & 0x33333333) + ((x >> 2) & 0x33333333);
1133
x = (x & 0x0f0f0f0f) + ((x >> 4) & 0x0f0f0f0f);
1137
#define NEON_QDMULH16(dest, src1, src2, round) do { \
1138
uint32_t tmp = (int32_t)(int16_t) src1 * (int16_t) src2; \
1139
if ((tmp ^ (tmp << 1)) & SIGNBIT) { \
1141
tmp = (tmp >> 31) ^ ~SIGNBIT; \
1146
int32_t old = tmp; \
1148
if ((int32_t)tmp < old) { \
1150
tmp = SIGNBIT - 1; \
1155
#define NEON_FN(dest, src1, src2) NEON_QDMULH16(dest, src1, src2, 0)
1156
NEON_VOP_ENV(qdmulh_s16, neon_s16, 2)
1158
#define NEON_FN(dest, src1, src2) NEON_QDMULH16(dest, src1, src2, 1)
1159
NEON_VOP_ENV(qrdmulh_s16, neon_s16, 2)
1161
#undef NEON_QDMULH16
1163
#define NEON_QDMULH32(dest, src1, src2, round) do { \
1164
uint64_t tmp = (int64_t)(int32_t) src1 * (int32_t) src2; \
1165
if ((tmp ^ (tmp << 1)) & SIGNBIT64) { \
1167
tmp = (tmp >> 63) ^ ~SIGNBIT64; \
1172
int64_t old = tmp; \
1173
tmp += (int64_t)1 << 31; \
1174
if ((int64_t)tmp < old) { \
1176
tmp = SIGNBIT64 - 1; \
1181
#define NEON_FN(dest, src1, src2) NEON_QDMULH32(dest, src1, src2, 0)
1182
NEON_VOP_ENV(qdmulh_s32, neon_s32, 1)
1184
#define NEON_FN(dest, src1, src2) NEON_QDMULH32(dest, src1, src2, 1)
1185
NEON_VOP_ENV(qrdmulh_s32, neon_s32, 1)
1187
#undef NEON_QDMULH32
1189
uint32_t HELPER(neon_narrow_u8)(uint64_t x)
1191
return (x & 0xffu) | ((x >> 8) & 0xff00u) | ((x >> 16) & 0xff0000u)
1192
| ((x >> 24) & 0xff000000u);
1195
uint32_t HELPER(neon_narrow_u16)(uint64_t x)
1197
return (x & 0xffffu) | ((x >> 16) & 0xffff0000u);
1200
uint32_t HELPER(neon_narrow_high_u8)(uint64_t x)
1202
return ((x >> 8) & 0xff) | ((x >> 16) & 0xff00)
1203
| ((x >> 24) & 0xff0000) | ((x >> 32) & 0xff000000);
1206
uint32_t HELPER(neon_narrow_high_u16)(uint64_t x)
1208
return ((x >> 16) & 0xffff) | ((x >> 32) & 0xffff0000);
1211
uint32_t HELPER(neon_narrow_round_high_u8)(uint64_t x)
1213
x &= 0xff80ff80ff80ff80ull;
1214
x += 0x0080008000800080ull;
1215
return ((x >> 8) & 0xff) | ((x >> 16) & 0xff00)
1216
| ((x >> 24) & 0xff0000) | ((x >> 32) & 0xff000000);
1219
uint32_t HELPER(neon_narrow_round_high_u16)(uint64_t x)
1221
x &= 0xffff8000ffff8000ull;
1222
x += 0x0000800000008000ull;
1223
return ((x >> 16) & 0xffff) | ((x >> 32) & 0xffff0000);
1226
uint32_t HELPER(neon_unarrow_sat8)(CPUState *env, uint64_t x)
1242
res |= (uint32_t)d << (n / 2); \
1253
uint32_t HELPER(neon_narrow_sat_u8)(CPUState *env, uint64_t x)
1266
res |= (uint32_t)d << (n / 2);
1276
uint32_t HELPER(neon_narrow_sat_s8)(CPUState *env, uint64_t x)
1283
if (s != (int8_t)s) { \
1284
d = (s >> 15) ^ 0x7f; \
1289
res |= (uint32_t)d << (n / 2);
1299
uint32_t HELPER(neon_unarrow_sat16)(CPUState *env, uint64_t x)
1304
if (low & 0x80000000) {
1307
} else if (low > 0xffff) {
1312
if (high & 0x80000000) {
1315
} else if (high > 0xffff) {
1319
return low | (high << 16);
1322
uint32_t HELPER(neon_narrow_sat_u16)(CPUState *env, uint64_t x)
1332
if (high > 0xffff) {
1336
return low | (high << 16);
1339
uint32_t HELPER(neon_narrow_sat_s16)(CPUState *env, uint64_t x)
1344
if (low != (int16_t)low) {
1345
low = (low >> 31) ^ 0x7fff;
1349
if (high != (int16_t)high) {
1350
high = (high >> 31) ^ 0x7fff;
1353
return (uint16_t)low | (high << 16);
1356
uint32_t HELPER(neon_unarrow_sat32)(CPUState *env, uint64_t x)
1358
if (x & 0x8000000000000000ull) {
1362
if (x > 0xffffffffu) {
1369
uint32_t HELPER(neon_narrow_sat_u32)(CPUState *env, uint64_t x)
1371
if (x > 0xffffffffu) {
1378
uint32_t HELPER(neon_narrow_sat_s32)(CPUState *env, uint64_t x)
1380
if ((int64_t)x != (int32_t)x) {
1382
return ((int64_t)x >> 63) ^ 0x7fffffff;
1387
uint64_t HELPER(neon_widen_u8)(uint32_t x)
1392
tmp = (uint8_t)(x >> 8);
1394
tmp = (uint8_t)(x >> 16);
1396
tmp = (uint8_t)(x >> 24);
1401
uint64_t HELPER(neon_widen_s8)(uint32_t x)
1405
ret = (uint16_t)(int8_t)x;
1406
tmp = (uint16_t)(int8_t)(x >> 8);
1408
tmp = (uint16_t)(int8_t)(x >> 16);
1410
tmp = (uint16_t)(int8_t)(x >> 24);
1415
uint64_t HELPER(neon_widen_u16)(uint32_t x)
1417
uint64_t high = (uint16_t)(x >> 16);
1418
return ((uint16_t)x) | (high << 32);
1421
uint64_t HELPER(neon_widen_s16)(uint32_t x)
1423
uint64_t high = (int16_t)(x >> 16);
1424
return ((uint32_t)(int16_t)x) | (high << 32);
1427
uint64_t HELPER(neon_addl_u16)(uint64_t a, uint64_t b)
1430
mask = (a ^ b) & 0x8000800080008000ull;
1431
a &= ~0x8000800080008000ull;
1432
b &= ~0x8000800080008000ull;
1433
return (a + b) ^ mask;
1436
uint64_t HELPER(neon_addl_u32)(uint64_t a, uint64_t b)
1439
mask = (a ^ b) & 0x8000000080000000ull;
1440
a &= ~0x8000000080000000ull;
1441
b &= ~0x8000000080000000ull;
1442
return (a + b) ^ mask;
1445
uint64_t HELPER(neon_paddl_u16)(uint64_t a, uint64_t b)
1450
tmp = a & 0x0000ffff0000ffffull;
1451
tmp += (a >> 16) & 0x0000ffff0000ffffull;
1452
tmp2 = b & 0xffff0000ffff0000ull;
1453
tmp2 += (b << 16) & 0xffff0000ffff0000ull;
1454
return ( tmp & 0xffff)
1455
| ((tmp >> 16) & 0xffff0000ull)
1456
| ((tmp2 << 16) & 0xffff00000000ull)
1457
| ( tmp2 & 0xffff000000000000ull);
1460
uint64_t HELPER(neon_paddl_u32)(uint64_t a, uint64_t b)
1462
uint32_t low = a + (a >> 32);
1463
uint32_t high = b + (b >> 32);
1464
return low + ((uint64_t)high << 32);
1467
uint64_t HELPER(neon_subl_u16)(uint64_t a, uint64_t b)
1470
mask = (a ^ ~b) & 0x8000800080008000ull;
1471
a |= 0x8000800080008000ull;
1472
b &= ~0x8000800080008000ull;
1473
return (a - b) ^ mask;
1476
uint64_t HELPER(neon_subl_u32)(uint64_t a, uint64_t b)
1479
mask = (a ^ ~b) & 0x8000000080000000ull;
1480
a |= 0x8000000080000000ull;
1481
b &= ~0x8000000080000000ull;
1482
return (a - b) ^ mask;
1485
uint64_t HELPER(neon_addl_saturate_s32)(CPUState *env, uint64_t a, uint64_t b)
1493
if (((low ^ x) & SIGNBIT) && !((x ^ y) & SIGNBIT)) {
1495
low = ((int32_t)x >> 31) ^ ~SIGNBIT;
1500
if (((high ^ x) & SIGNBIT) && !((x ^ y) & SIGNBIT)) {
1502
high = ((int32_t)x >> 31) ^ ~SIGNBIT;
1504
return low | ((uint64_t)high << 32);
1507
uint64_t HELPER(neon_addl_saturate_s64)(CPUState *env, uint64_t a, uint64_t b)
1512
if (((result ^ a) & SIGNBIT64) && !((a ^ b) & SIGNBIT64)) {
1514
result = ((int64_t)a >> 63) ^ ~SIGNBIT64;
1519
/* We have to do the arithmetic in a larger type than
1520
* the input type, because for example with a signed 32 bit
1521
* op the absolute difference can overflow a signed 32 bit value.
1523
#define DO_ABD(dest, x, y, intype, arithtype) do { \
1524
arithtype tmp_x = (intype)(x); \
1525
arithtype tmp_y = (intype)(y); \
1526
dest = ((tmp_x > tmp_y) ? tmp_x - tmp_y : tmp_y - tmp_x); \
1529
uint64_t HELPER(neon_abdl_u16)(uint32_t a, uint32_t b)
1533
DO_ABD(result, a, b, uint8_t, uint32_t);
1534
DO_ABD(tmp, a >> 8, b >> 8, uint8_t, uint32_t);
1535
result |= tmp << 16;
1536
DO_ABD(tmp, a >> 16, b >> 16, uint8_t, uint32_t);
1537
result |= tmp << 32;
1538
DO_ABD(tmp, a >> 24, b >> 24, uint8_t, uint32_t);
1539
result |= tmp << 48;
1543
uint64_t HELPER(neon_abdl_s16)(uint32_t a, uint32_t b)
1547
DO_ABD(result, a, b, int8_t, int32_t);
1548
DO_ABD(tmp, a >> 8, b >> 8, int8_t, int32_t);
1549
result |= tmp << 16;
1550
DO_ABD(tmp, a >> 16, b >> 16, int8_t, int32_t);
1551
result |= tmp << 32;
1552
DO_ABD(tmp, a >> 24, b >> 24, int8_t, int32_t);
1553
result |= tmp << 48;
1557
uint64_t HELPER(neon_abdl_u32)(uint32_t a, uint32_t b)
1561
DO_ABD(result, a, b, uint16_t, uint32_t);
1562
DO_ABD(tmp, a >> 16, b >> 16, uint16_t, uint32_t);
1563
return result | (tmp << 32);
1566
uint64_t HELPER(neon_abdl_s32)(uint32_t a, uint32_t b)
1570
DO_ABD(result, a, b, int16_t, int32_t);
1571
DO_ABD(tmp, a >> 16, b >> 16, int16_t, int32_t);
1572
return result | (tmp << 32);
1575
uint64_t HELPER(neon_abdl_u64)(uint32_t a, uint32_t b)
1578
DO_ABD(result, a, b, uint32_t, uint64_t);
1582
uint64_t HELPER(neon_abdl_s64)(uint32_t a, uint32_t b)
1585
DO_ABD(result, a, b, int32_t, int64_t);
1590
/* Widening multiply. Named type is the source type. */
1591
#define DO_MULL(dest, x, y, type1, type2) do { \
1594
dest = (type2)((type2)tmp_x * (type2)tmp_y); \
1597
uint64_t HELPER(neon_mull_u8)(uint32_t a, uint32_t b)
1602
DO_MULL(result, a, b, uint8_t, uint16_t);
1603
DO_MULL(tmp, a >> 8, b >> 8, uint8_t, uint16_t);
1604
result |= tmp << 16;
1605
DO_MULL(tmp, a >> 16, b >> 16, uint8_t, uint16_t);
1606
result |= tmp << 32;
1607
DO_MULL(tmp, a >> 24, b >> 24, uint8_t, uint16_t);
1608
result |= tmp << 48;
1612
uint64_t HELPER(neon_mull_s8)(uint32_t a, uint32_t b)
1617
DO_MULL(result, a, b, int8_t, uint16_t);
1618
DO_MULL(tmp, a >> 8, b >> 8, int8_t, uint16_t);
1619
result |= tmp << 16;
1620
DO_MULL(tmp, a >> 16, b >> 16, int8_t, uint16_t);
1621
result |= tmp << 32;
1622
DO_MULL(tmp, a >> 24, b >> 24, int8_t, uint16_t);
1623
result |= tmp << 48;
1627
uint64_t HELPER(neon_mull_u16)(uint32_t a, uint32_t b)
1632
DO_MULL(result, a, b, uint16_t, uint32_t);
1633
DO_MULL(tmp, a >> 16, b >> 16, uint16_t, uint32_t);
1634
return result | (tmp << 32);
1637
uint64_t HELPER(neon_mull_s16)(uint32_t a, uint32_t b)
1642
DO_MULL(result, a, b, int16_t, uint32_t);
1643
DO_MULL(tmp, a >> 16, b >> 16, int16_t, uint32_t);
1644
return result | (tmp << 32);
1647
uint64_t HELPER(neon_negl_u16)(uint64_t x)
1651
result = (uint16_t)-x;
1653
result |= (uint64_t)tmp << 16;
1655
result |= (uint64_t)tmp << 32;
1657
result |= (uint64_t)tmp << 48;
1661
uint64_t HELPER(neon_negl_u32)(uint64_t x)
1664
uint32_t high = -(x >> 32);
1665
return low | ((uint64_t)high << 32);
1668
/* FIXME: There should be a native op for this. */
1669
uint64_t HELPER(neon_negl_u64)(uint64_t x)
1674
/* Saturnating sign manuipulation. */
1675
/* ??? Make these use NEON_VOP1 */
1676
#define DO_QABS8(x) do { \
1677
if (x == (int8_t)0x80) { \
1680
} else if (x < 0) { \
1683
uint32_t HELPER(neon_qabs_s8)(CPUState *env, uint32_t x)
1686
NEON_UNPACK(neon_s8, vec, x);
1691
NEON_PACK(neon_s8, x, vec);
1696
#define DO_QNEG8(x) do { \
1697
if (x == (int8_t)0x80) { \
1703
uint32_t HELPER(neon_qneg_s8)(CPUState *env, uint32_t x)
1706
NEON_UNPACK(neon_s8, vec, x);
1711
NEON_PACK(neon_s8, x, vec);
1716
#define DO_QABS16(x) do { \
1717
if (x == (int16_t)0x8000) { \
1720
} else if (x < 0) { \
1723
uint32_t HELPER(neon_qabs_s16)(CPUState *env, uint32_t x)
1726
NEON_UNPACK(neon_s16, vec, x);
1729
NEON_PACK(neon_s16, x, vec);
1734
#define DO_QNEG16(x) do { \
1735
if (x == (int16_t)0x8000) { \
1741
uint32_t HELPER(neon_qneg_s16)(CPUState *env, uint32_t x)
1744
NEON_UNPACK(neon_s16, vec, x);
1747
NEON_PACK(neon_s16, x, vec);
1752
uint32_t HELPER(neon_qabs_s32)(CPUState *env, uint32_t x)
1757
} else if ((int32_t)x < 0) {
1763
uint32_t HELPER(neon_qneg_s32)(CPUState *env, uint32_t x)
1774
/* NEON Float helpers. */
1775
uint32_t HELPER(neon_min_f32)(uint32_t a, uint32_t b, void *fpstp)
1777
float_status *fpst = fpstp;
1778
return float32_val(float32_min(make_float32(a), make_float32(b), fpst));
1781
uint32_t HELPER(neon_max_f32)(uint32_t a, uint32_t b, void *fpstp)
1783
float_status *fpst = fpstp;
1784
return float32_val(float32_max(make_float32(a), make_float32(b), fpst));
1787
uint32_t HELPER(neon_abd_f32)(uint32_t a, uint32_t b, void *fpstp)
1789
float_status *fpst = fpstp;
1790
float32 f0 = make_float32(a);
1791
float32 f1 = make_float32(b);
1792
return float32_val(float32_abs(float32_sub(f0, f1, fpst)));
1795
/* Floating point comparisons produce an integer result.
1796
* Note that EQ doesn't signal InvalidOp for QNaNs but GE and GT do.
1797
* Softfloat routines return 0/1, which we convert to the 0/-1 Neon requires.
1799
uint32_t HELPER(neon_ceq_f32)(uint32_t a, uint32_t b, void *fpstp)
1801
float_status *fpst = fpstp;
1802
return -float32_eq_quiet(make_float32(a), make_float32(b), fpst);
1805
uint32_t HELPER(neon_cge_f32)(uint32_t a, uint32_t b, void *fpstp)
1807
float_status *fpst = fpstp;
1808
return -float32_le(make_float32(b), make_float32(a), fpst);
1811
uint32_t HELPER(neon_cgt_f32)(uint32_t a, uint32_t b, void *fpstp)
1813
float_status *fpst = fpstp;
1814
return -float32_lt(make_float32(b), make_float32(a), fpst);
1817
uint32_t HELPER(neon_acge_f32)(uint32_t a, uint32_t b, void *fpstp)
1819
float_status *fpst = fpstp;
1820
float32 f0 = float32_abs(make_float32(a));
1821
float32 f1 = float32_abs(make_float32(b));
1822
return -float32_le(f1, f0, fpst);
1825
uint32_t HELPER(neon_acgt_f32)(uint32_t a, uint32_t b, void *fpstp)
1827
float_status *fpst = fpstp;
1828
float32 f0 = float32_abs(make_float32(a));
1829
float32 f1 = float32_abs(make_float32(b));
1830
return -float32_lt(f1, f0, fpst);
1833
#define ELEM(V, N, SIZE) (((V) >> ((N) * (SIZE))) & ((1ull << (SIZE)) - 1))
1835
void HELPER(neon_qunzip8)(CPUState *env, uint32_t rd, uint32_t rm)
1837
uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1838
uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1839
uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1840
uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1841
uint64_t d0 = ELEM(zd0, 0, 8) | (ELEM(zd0, 2, 8) << 8)
1842
| (ELEM(zd0, 4, 8) << 16) | (ELEM(zd0, 6, 8) << 24)
1843
| (ELEM(zd1, 0, 8) << 32) | (ELEM(zd1, 2, 8) << 40)
1844
| (ELEM(zd1, 4, 8) << 48) | (ELEM(zd1, 6, 8) << 56);
1845
uint64_t d1 = ELEM(zm0, 0, 8) | (ELEM(zm0, 2, 8) << 8)
1846
| (ELEM(zm0, 4, 8) << 16) | (ELEM(zm0, 6, 8) << 24)
1847
| (ELEM(zm1, 0, 8) << 32) | (ELEM(zm1, 2, 8) << 40)
1848
| (ELEM(zm1, 4, 8) << 48) | (ELEM(zm1, 6, 8) << 56);
1849
uint64_t m0 = ELEM(zd0, 1, 8) | (ELEM(zd0, 3, 8) << 8)
1850
| (ELEM(zd0, 5, 8) << 16) | (ELEM(zd0, 7, 8) << 24)
1851
| (ELEM(zd1, 1, 8) << 32) | (ELEM(zd1, 3, 8) << 40)
1852
| (ELEM(zd1, 5, 8) << 48) | (ELEM(zd1, 7, 8) << 56);
1853
uint64_t m1 = ELEM(zm0, 1, 8) | (ELEM(zm0, 3, 8) << 8)
1854
| (ELEM(zm0, 5, 8) << 16) | (ELEM(zm0, 7, 8) << 24)
1855
| (ELEM(zm1, 1, 8) << 32) | (ELEM(zm1, 3, 8) << 40)
1856
| (ELEM(zm1, 5, 8) << 48) | (ELEM(zm1, 7, 8) << 56);
1857
env->vfp.regs[rm] = make_float64(m0);
1858
env->vfp.regs[rm + 1] = make_float64(m1);
1859
env->vfp.regs[rd] = make_float64(d0);
1860
env->vfp.regs[rd + 1] = make_float64(d1);
1863
void HELPER(neon_qunzip16)(CPUState *env, uint32_t rd, uint32_t rm)
1865
uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1866
uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1867
uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1868
uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1869
uint64_t d0 = ELEM(zd0, 0, 16) | (ELEM(zd0, 2, 16) << 16)
1870
| (ELEM(zd1, 0, 16) << 32) | (ELEM(zd1, 2, 16) << 48);
1871
uint64_t d1 = ELEM(zm0, 0, 16) | (ELEM(zm0, 2, 16) << 16)
1872
| (ELEM(zm1, 0, 16) << 32) | (ELEM(zm1, 2, 16) << 48);
1873
uint64_t m0 = ELEM(zd0, 1, 16) | (ELEM(zd0, 3, 16) << 16)
1874
| (ELEM(zd1, 1, 16) << 32) | (ELEM(zd1, 3, 16) << 48);
1875
uint64_t m1 = ELEM(zm0, 1, 16) | (ELEM(zm0, 3, 16) << 16)
1876
| (ELEM(zm1, 1, 16) << 32) | (ELEM(zm1, 3, 16) << 48);
1877
env->vfp.regs[rm] = make_float64(m0);
1878
env->vfp.regs[rm + 1] = make_float64(m1);
1879
env->vfp.regs[rd] = make_float64(d0);
1880
env->vfp.regs[rd + 1] = make_float64(d1);
1883
void HELPER(neon_qunzip32)(CPUState *env, uint32_t rd, uint32_t rm)
1885
uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1886
uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1887
uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1888
uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1889
uint64_t d0 = ELEM(zd0, 0, 32) | (ELEM(zd1, 0, 32) << 32);
1890
uint64_t d1 = ELEM(zm0, 0, 32) | (ELEM(zm1, 0, 32) << 32);
1891
uint64_t m0 = ELEM(zd0, 1, 32) | (ELEM(zd1, 1, 32) << 32);
1892
uint64_t m1 = ELEM(zm0, 1, 32) | (ELEM(zm1, 1, 32) << 32);
1893
env->vfp.regs[rm] = make_float64(m0);
1894
env->vfp.regs[rm + 1] = make_float64(m1);
1895
env->vfp.regs[rd] = make_float64(d0);
1896
env->vfp.regs[rd + 1] = make_float64(d1);
1899
void HELPER(neon_unzip8)(CPUState *env, uint32_t rd, uint32_t rm)
1901
uint64_t zm = float64_val(env->vfp.regs[rm]);
1902
uint64_t zd = float64_val(env->vfp.regs[rd]);
1903
uint64_t d0 = ELEM(zd, 0, 8) | (ELEM(zd, 2, 8) << 8)
1904
| (ELEM(zd, 4, 8) << 16) | (ELEM(zd, 6, 8) << 24)
1905
| (ELEM(zm, 0, 8) << 32) | (ELEM(zm, 2, 8) << 40)
1906
| (ELEM(zm, 4, 8) << 48) | (ELEM(zm, 6, 8) << 56);
1907
uint64_t m0 = ELEM(zd, 1, 8) | (ELEM(zd, 3, 8) << 8)
1908
| (ELEM(zd, 5, 8) << 16) | (ELEM(zd, 7, 8) << 24)
1909
| (ELEM(zm, 1, 8) << 32) | (ELEM(zm, 3, 8) << 40)
1910
| (ELEM(zm, 5, 8) << 48) | (ELEM(zm, 7, 8) << 56);
1911
env->vfp.regs[rm] = make_float64(m0);
1912
env->vfp.regs[rd] = make_float64(d0);
1915
void HELPER(neon_unzip16)(CPUState *env, uint32_t rd, uint32_t rm)
1917
uint64_t zm = float64_val(env->vfp.regs[rm]);
1918
uint64_t zd = float64_val(env->vfp.regs[rd]);
1919
uint64_t d0 = ELEM(zd, 0, 16) | (ELEM(zd, 2, 16) << 16)
1920
| (ELEM(zm, 0, 16) << 32) | (ELEM(zm, 2, 16) << 48);
1921
uint64_t m0 = ELEM(zd, 1, 16) | (ELEM(zd, 3, 16) << 16)
1922
| (ELEM(zm, 1, 16) << 32) | (ELEM(zm, 3, 16) << 48);
1923
env->vfp.regs[rm] = make_float64(m0);
1924
env->vfp.regs[rd] = make_float64(d0);
1927
void HELPER(neon_qzip8)(CPUState *env, uint32_t rd, uint32_t rm)
1929
uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1930
uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1931
uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1932
uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1933
uint64_t d0 = ELEM(zd0, 0, 8) | (ELEM(zm0, 0, 8) << 8)
1934
| (ELEM(zd0, 1, 8) << 16) | (ELEM(zm0, 1, 8) << 24)
1935
| (ELEM(zd0, 2, 8) << 32) | (ELEM(zm0, 2, 8) << 40)
1936
| (ELEM(zd0, 3, 8) << 48) | (ELEM(zm0, 3, 8) << 56);
1937
uint64_t d1 = ELEM(zd0, 4, 8) | (ELEM(zm0, 4, 8) << 8)
1938
| (ELEM(zd0, 5, 8) << 16) | (ELEM(zm0, 5, 8) << 24)
1939
| (ELEM(zd0, 6, 8) << 32) | (ELEM(zm0, 6, 8) << 40)
1940
| (ELEM(zd0, 7, 8) << 48) | (ELEM(zm0, 7, 8) << 56);
1941
uint64_t m0 = ELEM(zd1, 0, 8) | (ELEM(zm1, 0, 8) << 8)
1942
| (ELEM(zd1, 1, 8) << 16) | (ELEM(zm1, 1, 8) << 24)
1943
| (ELEM(zd1, 2, 8) << 32) | (ELEM(zm1, 2, 8) << 40)
1944
| (ELEM(zd1, 3, 8) << 48) | (ELEM(zm1, 3, 8) << 56);
1945
uint64_t m1 = ELEM(zd1, 4, 8) | (ELEM(zm1, 4, 8) << 8)
1946
| (ELEM(zd1, 5, 8) << 16) | (ELEM(zm1, 5, 8) << 24)
1947
| (ELEM(zd1, 6, 8) << 32) | (ELEM(zm1, 6, 8) << 40)
1948
| (ELEM(zd1, 7, 8) << 48) | (ELEM(zm1, 7, 8) << 56);
1949
env->vfp.regs[rm] = make_float64(m0);
1950
env->vfp.regs[rm + 1] = make_float64(m1);
1951
env->vfp.regs[rd] = make_float64(d0);
1952
env->vfp.regs[rd + 1] = make_float64(d1);
1955
void HELPER(neon_qzip16)(CPUState *env, uint32_t rd, uint32_t rm)
1957
uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1958
uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1959
uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1960
uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1961
uint64_t d0 = ELEM(zd0, 0, 16) | (ELEM(zm0, 0, 16) << 16)
1962
| (ELEM(zd0, 1, 16) << 32) | (ELEM(zm0, 1, 16) << 48);
1963
uint64_t d1 = ELEM(zd0, 2, 16) | (ELEM(zm0, 2, 16) << 16)
1964
| (ELEM(zd0, 3, 16) << 32) | (ELEM(zm0, 3, 16) << 48);
1965
uint64_t m0 = ELEM(zd1, 0, 16) | (ELEM(zm1, 0, 16) << 16)
1966
| (ELEM(zd1, 1, 16) << 32) | (ELEM(zm1, 1, 16) << 48);
1967
uint64_t m1 = ELEM(zd1, 2, 16) | (ELEM(zm1, 2, 16) << 16)
1968
| (ELEM(zd1, 3, 16) << 32) | (ELEM(zm1, 3, 16) << 48);
1969
env->vfp.regs[rm] = make_float64(m0);
1970
env->vfp.regs[rm + 1] = make_float64(m1);
1971
env->vfp.regs[rd] = make_float64(d0);
1972
env->vfp.regs[rd + 1] = make_float64(d1);
1975
void HELPER(neon_qzip32)(CPUState *env, uint32_t rd, uint32_t rm)
1977
uint64_t zm0 = float64_val(env->vfp.regs[rm]);
1978
uint64_t zm1 = float64_val(env->vfp.regs[rm + 1]);
1979
uint64_t zd0 = float64_val(env->vfp.regs[rd]);
1980
uint64_t zd1 = float64_val(env->vfp.regs[rd + 1]);
1981
uint64_t d0 = ELEM(zd0, 0, 32) | (ELEM(zm0, 0, 32) << 32);
1982
uint64_t d1 = ELEM(zd0, 1, 32) | (ELEM(zm0, 1, 32) << 32);
1983
uint64_t m0 = ELEM(zd1, 0, 32) | (ELEM(zm1, 0, 32) << 32);
1984
uint64_t m1 = ELEM(zd1, 1, 32) | (ELEM(zm1, 1, 32) << 32);
1985
env->vfp.regs[rm] = make_float64(m0);
1986
env->vfp.regs[rm + 1] = make_float64(m1);
1987
env->vfp.regs[rd] = make_float64(d0);
1988
env->vfp.regs[rd + 1] = make_float64(d1);
1991
void HELPER(neon_zip8)(CPUState *env, uint32_t rd, uint32_t rm)
1993
uint64_t zm = float64_val(env->vfp.regs[rm]);
1994
uint64_t zd = float64_val(env->vfp.regs[rd]);
1995
uint64_t d0 = ELEM(zd, 0, 8) | (ELEM(zm, 0, 8) << 8)
1996
| (ELEM(zd, 1, 8) << 16) | (ELEM(zm, 1, 8) << 24)
1997
| (ELEM(zd, 2, 8) << 32) | (ELEM(zm, 2, 8) << 40)
1998
| (ELEM(zd, 3, 8) << 48) | (ELEM(zm, 3, 8) << 56);
1999
uint64_t m0 = ELEM(zd, 4, 8) | (ELEM(zm, 4, 8) << 8)
2000
| (ELEM(zd, 5, 8) << 16) | (ELEM(zm, 5, 8) << 24)
2001
| (ELEM(zd, 6, 8) << 32) | (ELEM(zm, 6, 8) << 40)
2002
| (ELEM(zd, 7, 8) << 48) | (ELEM(zm, 7, 8) << 56);
2003
env->vfp.regs[rm] = make_float64(m0);
2004
env->vfp.regs[rd] = make_float64(d0);
2007
void HELPER(neon_zip16)(CPUState *env, uint32_t rd, uint32_t rm)
2009
uint64_t zm = float64_val(env->vfp.regs[rm]);
2010
uint64_t zd = float64_val(env->vfp.regs[rd]);
2011
uint64_t d0 = ELEM(zd, 0, 16) | (ELEM(zm, 0, 16) << 16)
2012
| (ELEM(zd, 1, 16) << 32) | (ELEM(zm, 1, 16) << 48);
2013
uint64_t m0 = ELEM(zd, 2, 16) | (ELEM(zm, 2, 16) << 16)
2014
| (ELEM(zd, 3, 16) << 32) | (ELEM(zm, 3, 16) << 48);
2015
env->vfp.regs[rm] = make_float64(m0);
2016
env->vfp.regs[rd] = make_float64(d0);