?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
gitlab: tweak comments in edk2/opensbi jobs
[qemu.git] / target / arm / translate-neon.c
blob321c17e2c7e9563555de2a12329369a2aa3a66a3
1 /*
2 * ARM translation: AArch32 Neon instructions
3 *
4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2005-2007 CodeSourcery
6 * Copyright (c) 2007 OpenedHand, Ltd.
7 * Copyright (c) 2020 Linaro, Ltd.
8 *
9 * This library is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
13 *
14 * This library is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
21 */
22
23 #include "qemu/osdep.h"
24 #include "tcg/tcg-op.h"
25 #include "tcg/tcg-op-gvec.h"
26 #include "exec/exec-all.h"
27 #include "exec/gen-icount.h"
28 #include "translate.h"
29 #include "translate-a32.h"
30
31 /* Include the generated Neon decoder */
32 #include "decode-neon-dp.c.inc"
33 #include "decode-neon-ls.c.inc"
34 #include "decode-neon-shared.c.inc"
35
36 static TCGv_ptr vfp_reg_ptr(bool dp, int reg)
37 {
38 TCGv_ptr ret = tcg_temp_new_ptr();
39 tcg_gen_addi_ptr(ret, cpu_env, vfp_reg_offset(dp, reg));
40 return ret;
41 }
42
43 static void neon_load_element(TCGv_i32 var, int reg, int ele, MemOp mop)
44 {
45 long offset = neon_element_offset(reg, ele, mop & MO_SIZE);
46
47 switch (mop) {
48 case MO_UB:
49 tcg_gen_ld8u_i32(var, cpu_env, offset);
50 break;
51 case MO_UW:
52 tcg_gen_ld16u_i32(var, cpu_env, offset);
53 break;
54 case MO_UL:
55 tcg_gen_ld_i32(var, cpu_env, offset);
56 break;
57 default:
58 g_assert_not_reached();
59 }
60 }
61
62 static void neon_load_element64(TCGv_i64 var, int reg, int ele, MemOp mop)
63 {
64 long offset = neon_element_offset(reg, ele, mop & MO_SIZE);
65
66 switch (mop) {
67 case MO_UB:
68 tcg_gen_ld8u_i64(var, cpu_env, offset);
69 break;
70 case MO_UW:
71 tcg_gen_ld16u_i64(var, cpu_env, offset);
72 break;
73 case MO_UL:
74 tcg_gen_ld32u_i64(var, cpu_env, offset);
75 break;
76 case MO_UQ:
77 tcg_gen_ld_i64(var, cpu_env, offset);
78 break;
79 default:
80 g_assert_not_reached();
81 }
82 }
83
84 static void neon_store_element(int reg, int ele, MemOp size, TCGv_i32 var)
85 {
86 long offset = neon_element_offset(reg, ele, size);
87
88 switch (size) {
89 case MO_8:
90 tcg_gen_st8_i32(var, cpu_env, offset);
91 break;
92 case MO_16:
93 tcg_gen_st16_i32(var, cpu_env, offset);
94 break;
95 case MO_32:
96 tcg_gen_st_i32(var, cpu_env, offset);
97 break;
98 default:
99 g_assert_not_reached();
100 }
101 }
102
103 static void neon_store_element64(int reg, int ele, MemOp size, TCGv_i64 var)
104 {
105 long offset = neon_element_offset(reg, ele, size);
106
107 switch (size) {
108 case MO_8:
109 tcg_gen_st8_i64(var, cpu_env, offset);
110 break;
111 case MO_16:
112 tcg_gen_st16_i64(var, cpu_env, offset);
113 break;
114 case MO_32:
115 tcg_gen_st32_i64(var, cpu_env, offset);
116 break;
117 case MO_64:
118 tcg_gen_st_i64(var, cpu_env, offset);
119 break;
120 default:
121 g_assert_not_reached();
122 }
123 }
124
125 static bool do_neon_ddda(DisasContext *s, int q, int vd, int vn, int vm,
126 int data, gen_helper_gvec_4 *fn_gvec)
127 {
128 /* UNDEF accesses to D16-D31 if they don't exist. */
129 if (((vd | vn | vm) & 0x10) && !dc_isar_feature(aa32_simd_r32, s)) {
130 return false;
131 }
132
133 /*
134 * UNDEF accesses to odd registers for each bit of Q.
135 * Q will be 0b111 for all Q-reg instructions, otherwise
136 * when we have mixed Q- and D-reg inputs.
137 */
138 if (((vd & 1) * 4 | (vn & 1) * 2 | (vm & 1)) & q) {
139 return false;
140 }
141
142 if (!vfp_access_check(s)) {
143 return true;
144 }
145
146 int opr_sz = q ? 16 : 8;
147 tcg_gen_gvec_4_ool(vfp_reg_offset(1, vd),
148 vfp_reg_offset(1, vn),
149 vfp_reg_offset(1, vm),
150 vfp_reg_offset(1, vd),
151 opr_sz, opr_sz, data, fn_gvec);
152 return true;
153 }
154
155 static bool do_neon_ddda_fpst(DisasContext *s, int q, int vd, int vn, int vm,
156 int data, ARMFPStatusFlavour fp_flavour,
157 gen_helper_gvec_4_ptr *fn_gvec_ptr)
158 {
159 /* UNDEF accesses to D16-D31 if they don't exist. */
160 if (((vd | vn | vm) & 0x10) && !dc_isar_feature(aa32_simd_r32, s)) {
161 return false;
162 }
163
164 /*
165 * UNDEF accesses to odd registers for each bit of Q.
166 * Q will be 0b111 for all Q-reg instructions, otherwise
167 * when we have mixed Q- and D-reg inputs.
168 */
169 if (((vd & 1) * 4 | (vn & 1) * 2 | (vm & 1)) & q) {
170 return false;
171 }
172
173 if (!vfp_access_check(s)) {
174 return true;
175 }
176
177 int opr_sz = q ? 16 : 8;
178 TCGv_ptr fpst = fpstatus_ptr(fp_flavour);
179
180 tcg_gen_gvec_4_ptr(vfp_reg_offset(1, vd),
181 vfp_reg_offset(1, vn),
182 vfp_reg_offset(1, vm),
183 vfp_reg_offset(1, vd),
184 fpst, opr_sz, opr_sz, data, fn_gvec_ptr);
185 tcg_temp_free_ptr(fpst);
186 return true;
187 }
188
189 static bool trans_VCMLA(DisasContext *s, arg_VCMLA *a)
190 {
191 if (!dc_isar_feature(aa32_vcma, s)) {
192 return false;
193 }
194 if (a->size == MO_16) {
195 if (!dc_isar_feature(aa32_fp16_arith, s)) {
196 return false;
197 }
198 return do_neon_ddda_fpst(s, a->q * 7, a->vd, a->vn, a->vm, a->rot,
199 FPST_STD_F16, gen_helper_gvec_fcmlah);
200 }
201 return do_neon_ddda_fpst(s, a->q * 7, a->vd, a->vn, a->vm, a->rot,
202 FPST_STD, gen_helper_gvec_fcmlas);
203 }
204
205 static bool trans_VCADD(DisasContext *s, arg_VCADD *a)
206 {
207 int opr_sz;
208 TCGv_ptr fpst;
209 gen_helper_gvec_3_ptr *fn_gvec_ptr;
210
211 if (!dc_isar_feature(aa32_vcma, s)
212 || (a->size == MO_16 && !dc_isar_feature(aa32_fp16_arith, s))) {
213 return false;
214 }
215
216 /* UNDEF accesses to D16-D31 if they don't exist. */
217 if (!dc_isar_feature(aa32_simd_r32, s) &&
218 ((a->vd | a->vn | a->vm) & 0x10)) {
219 return false;
220 }
221
222 if ((a->vn | a->vm | a->vd) & a->q) {
223 return false;
224 }
225
226 if (!vfp_access_check(s)) {
227 return true;
228 }
229
230 opr_sz = (1 + a->q) * 8;
231 fpst = fpstatus_ptr(a->size == MO_16 ? FPST_STD_F16 : FPST_STD);
232 fn_gvec_ptr = (a->size == MO_16) ?
233 gen_helper_gvec_fcaddh : gen_helper_gvec_fcadds;
234 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, a->vd),
235 vfp_reg_offset(1, a->vn),
236 vfp_reg_offset(1, a->vm),
237 fpst, opr_sz, opr_sz, a->rot,
238 fn_gvec_ptr);
239 tcg_temp_free_ptr(fpst);
240 return true;
241 }
242
243 static bool trans_VSDOT(DisasContext *s, arg_VSDOT *a)
244 {
245 if (!dc_isar_feature(aa32_dp, s)) {
246 return false;
247 }
248 return do_neon_ddda(s, a->q * 7, a->vd, a->vn, a->vm, 0,
249 gen_helper_gvec_sdot_b);
250 }
251
252 static bool trans_VUDOT(DisasContext *s, arg_VUDOT *a)
253 {
254 if (!dc_isar_feature(aa32_dp, s)) {
255 return false;
256 }
257 return do_neon_ddda(s, a->q * 7, a->vd, a->vn, a->vm, 0,
258 gen_helper_gvec_udot_b);
259 }
260
261 static bool trans_VUSDOT(DisasContext *s, arg_VUSDOT *a)
262 {
263 if (!dc_isar_feature(aa32_i8mm, s)) {
264 return false;
265 }
266 return do_neon_ddda(s, a->q * 7, a->vd, a->vn, a->vm, 0,
267 gen_helper_gvec_usdot_b);
268 }
269
270 static bool trans_VDOT_b16(DisasContext *s, arg_VDOT_b16 *a)
271 {
272 if (!dc_isar_feature(aa32_bf16, s)) {
273 return false;
274 }
275 return do_neon_ddda(s, a->q * 7, a->vd, a->vn, a->vm, 0,
276 gen_helper_gvec_bfdot);
277 }
278
279 static bool trans_VFML(DisasContext *s, arg_VFML *a)
280 {
281 int opr_sz;
282
283 if (!dc_isar_feature(aa32_fhm, s)) {
284 return false;
285 }
286
287 /* UNDEF accesses to D16-D31 if they don't exist. */
288 if (!dc_isar_feature(aa32_simd_r32, s) &&
289 (a->vd & 0x10)) {
290 return false;
291 }
292
293 if (a->vd & a->q) {
294 return false;
295 }
296
297 if (!vfp_access_check(s)) {
298 return true;
299 }
300
301 opr_sz = (1 + a->q) * 8;
302 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, a->vd),
303 vfp_reg_offset(a->q, a->vn),
304 vfp_reg_offset(a->q, a->vm),
305 cpu_env, opr_sz, opr_sz, a->s, /* is_2 == 0 */
306 gen_helper_gvec_fmlal_a32);
307 return true;
308 }
309
310 static bool trans_VCMLA_scalar(DisasContext *s, arg_VCMLA_scalar *a)
311 {
312 int data = (a->index << 2) | a->rot;
313
314 if (!dc_isar_feature(aa32_vcma, s)) {
315 return false;
316 }
317 if (a->size == MO_16) {
318 if (!dc_isar_feature(aa32_fp16_arith, s)) {
319 return false;
320 }
321 return do_neon_ddda_fpst(s, a->q * 6, a->vd, a->vn, a->vm, data,
322 FPST_STD_F16, gen_helper_gvec_fcmlah_idx);
323 }
324 return do_neon_ddda_fpst(s, a->q * 6, a->vd, a->vn, a->vm, data,
325 FPST_STD, gen_helper_gvec_fcmlas_idx);
326 }
327
328 static bool trans_VSDOT_scalar(DisasContext *s, arg_VSDOT_scalar *a)
329 {
330 if (!dc_isar_feature(aa32_dp, s)) {
331 return false;
332 }
333 return do_neon_ddda(s, a->q * 6, a->vd, a->vn, a->vm, a->index,
334 gen_helper_gvec_sdot_idx_b);
335 }
336
337 static bool trans_VUDOT_scalar(DisasContext *s, arg_VUDOT_scalar *a)
338 {
339 if (!dc_isar_feature(aa32_dp, s)) {
340 return false;
341 }
342 return do_neon_ddda(s, a->q * 6, a->vd, a->vn, a->vm, a->index,
343 gen_helper_gvec_udot_idx_b);
344 }
345
346 static bool trans_VUSDOT_scalar(DisasContext *s, arg_VUSDOT_scalar *a)
347 {
348 if (!dc_isar_feature(aa32_i8mm, s)) {
349 return false;
350 }
351 return do_neon_ddda(s, a->q * 6, a->vd, a->vn, a->vm, a->index,
352 gen_helper_gvec_usdot_idx_b);
353 }
354
355 static bool trans_VSUDOT_scalar(DisasContext *s, arg_VSUDOT_scalar *a)
356 {
357 if (!dc_isar_feature(aa32_i8mm, s)) {
358 return false;
359 }
360 return do_neon_ddda(s, a->q * 6, a->vd, a->vn, a->vm, a->index,
361 gen_helper_gvec_sudot_idx_b);
362 }
363
364 static bool trans_VDOT_b16_scal(DisasContext *s, arg_VDOT_b16_scal *a)
365 {
366 if (!dc_isar_feature(aa32_bf16, s)) {
367 return false;
368 }
369 return do_neon_ddda(s, a->q * 6, a->vd, a->vn, a->vm, a->index,
370 gen_helper_gvec_bfdot_idx);
371 }
372
373 static bool trans_VFML_scalar(DisasContext *s, arg_VFML_scalar *a)
374 {
375 int opr_sz;
376
377 if (!dc_isar_feature(aa32_fhm, s)) {
378 return false;
379 }
380
381 /* UNDEF accesses to D16-D31 if they don't exist. */
382 if (!dc_isar_feature(aa32_simd_r32, s) &&
383 ((a->vd & 0x10) || (a->q && (a->vn & 0x10)))) {
384 return false;
385 }
386
387 if (a->vd & a->q) {
388 return false;
389 }
390
391 if (!vfp_access_check(s)) {
392 return true;
393 }
394
395 opr_sz = (1 + a->q) * 8;
396 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, a->vd),
397 vfp_reg_offset(a->q, a->vn),
398 vfp_reg_offset(a->q, a->rm),
399 cpu_env, opr_sz, opr_sz,
400 (a->index << 2) | a->s, /* is_2 == 0 */
401 gen_helper_gvec_fmlal_idx_a32);
402 return true;
403 }
404
405 static struct {
406 int nregs;
407 int interleave;
408 int spacing;
409 } const neon_ls_element_type[11] = {
410 {1, 4, 1},
411 {1, 4, 2},
412 {4, 1, 1},
413 {2, 2, 2},
414 {1, 3, 1},
415 {1, 3, 2},
416 {3, 1, 1},
417 {1, 1, 1},
418 {1, 2, 1},
419 {1, 2, 2},
420 {2, 1, 1}
421 };
422
423 static void gen_neon_ldst_base_update(DisasContext *s, int rm, int rn,
424 int stride)
425 {
426 if (rm != 15) {
427 TCGv_i32 base;
428
429 base = load_reg(s, rn);
430 if (rm == 13) {
431 tcg_gen_addi_i32(base, base, stride);
432 } else {
433 TCGv_i32 index;
434 index = load_reg(s, rm);
435 tcg_gen_add_i32(base, base, index);
436 tcg_temp_free_i32(index);
437 }
438 store_reg(s, rn, base);
439 }
440 }
441
442 static bool trans_VLDST_multiple(DisasContext *s, arg_VLDST_multiple *a)
443 {
444 /* Neon load/store multiple structures */
445 int nregs, interleave, spacing, reg, n;
446 MemOp mop, align, endian;
447 int mmu_idx = get_mem_index(s);
448 int size = a->size;
449 TCGv_i64 tmp64;
450 TCGv_i32 addr;
451
452 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
453 return false;
454 }
455
456 /* UNDEF accesses to D16-D31 if they don't exist */
457 if (!dc_isar_feature(aa32_simd_r32, s) && (a->vd & 0x10)) {
458 return false;
459 }
460 if (a->itype > 10) {
461 return false;
462 }
463 /* Catch UNDEF cases for bad values of align field */
464 switch (a->itype & 0xc) {
465 case 4:
466 if (a->align >= 2) {
467 return false;
468 }
469 break;
470 case 8:
471 if (a->align == 3) {
472 return false;
473 }
474 break;
475 default:
476 break;
477 }
478 nregs = neon_ls_element_type[a->itype].nregs;
479 interleave = neon_ls_element_type[a->itype].interleave;
480 spacing = neon_ls_element_type[a->itype].spacing;
481 if (size == 3 && (interleave | spacing) != 1) {
482 return false;
483 }
484
485 if (!vfp_access_check(s)) {
486 return true;
487 }
488
489 /* For our purposes, bytes are always little-endian. */
490 endian = s->be_data;
491 if (size == 0) {
492 endian = MO_LE;
493 }
494
495 /* Enforce alignment requested by the instruction */
496 if (a->align) {
497 align = pow2_align(a->align + 2); /* 4 ** a->align */
498 } else {
499 align = s->align_mem ? MO_ALIGN : 0;
500 }
501
502 /*
503 * Consecutive little-endian elements from a single register
504 * can be promoted to a larger little-endian operation.
505 */
506 if (interleave == 1 && endian == MO_LE) {
507 /* Retain any natural alignment. */
508 if (align == MO_ALIGN) {
509 align = pow2_align(size);
510 }
511 size = 3;
512 }
513
514 tmp64 = tcg_temp_new_i64();
515 addr = tcg_temp_new_i32();
516 load_reg_var(s, addr, a->rn);
517
518 mop = endian | size | align;
519 for (reg = 0; reg < nregs; reg++) {
520 for (n = 0; n < 8 >> size; n++) {
521 int xs;
522 for (xs = 0; xs < interleave; xs++) {
523 int tt = a->vd + reg + spacing * xs;
524
525 if (a->l) {
526 gen_aa32_ld_internal_i64(s, tmp64, addr, mmu_idx, mop);
527 neon_store_element64(tt, n, size, tmp64);
528 } else {
529 neon_load_element64(tmp64, tt, n, size);
530 gen_aa32_st_internal_i64(s, tmp64, addr, mmu_idx, mop);
531 }
532 tcg_gen_addi_i32(addr, addr, 1 << size);
533
534 /* Subsequent memory operations inherit alignment */
535 mop &= ~MO_AMASK;
536 }
537 }
538 }
539 tcg_temp_free_i32(addr);
540 tcg_temp_free_i64(tmp64);
541
542 gen_neon_ldst_base_update(s, a->rm, a->rn, nregs * interleave * 8);
543 return true;
544 }
545
546 static bool trans_VLD_all_lanes(DisasContext *s, arg_VLD_all_lanes *a)
547 {
548 /* Neon load single structure to all lanes */
549 int reg, stride, vec_size;
550 int vd = a->vd;
551 int size = a->size;
552 int nregs = a->n + 1;
553 TCGv_i32 addr, tmp;
554 MemOp mop, align;
555
556 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
557 return false;
558 }
559
560 /* UNDEF accesses to D16-D31 if they don't exist */
561 if (!dc_isar_feature(aa32_simd_r32, s) && (a->vd & 0x10)) {
562 return false;
563 }
564
565 align = 0;
566 if (size == 3) {
567 if (nregs != 4 || a->a == 0) {
568 return false;
569 }
570 /* For VLD4 size == 3 a == 1 means 32 bits at 16 byte alignment */
571 size = MO_32;
572 align = MO_ALIGN_16;
573 } else if (a->a) {
574 switch (nregs) {
575 case 1:
576 if (size == 0) {
577 return false;
578 }
579 align = MO_ALIGN;
580 break;
581 case 2:
582 align = pow2_align(size + 1);
583 break;
584 case 3:
585 return false;
586 case 4:
587 align = pow2_align(size + 2);
588 break;
589 default:
590 g_assert_not_reached();
591 }
592 }
593
594 if (!vfp_access_check(s)) {
595 return true;
596 }
597
598 /*
599 * VLD1 to all lanes: T bit indicates how many Dregs to write.
600 * VLD2/3/4 to all lanes: T bit indicates register stride.
601 */
602 stride = a->t ? 2 : 1;
603 vec_size = nregs == 1 ? stride * 8 : 8;
604 mop = size | align;
605 tmp = tcg_temp_new_i32();
606 addr = tcg_temp_new_i32();
607 load_reg_var(s, addr, a->rn);
608 for (reg = 0; reg < nregs; reg++) {
609 gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), mop);
610 if ((vd & 1) && vec_size == 16) {
611 /*
612 * We cannot write 16 bytes at once because the
613 * destination is unaligned.
614 */
615 tcg_gen_gvec_dup_i32(size, neon_full_reg_offset(vd),
616 8, 8, tmp);
617 tcg_gen_gvec_mov(0, neon_full_reg_offset(vd + 1),
618 neon_full_reg_offset(vd), 8, 8);
619 } else {
620 tcg_gen_gvec_dup_i32(size, neon_full_reg_offset(vd),
621 vec_size, vec_size, tmp);
622 }
623 tcg_gen_addi_i32(addr, addr, 1 << size);
624 vd += stride;
625
626 /* Subsequent memory operations inherit alignment */
627 mop &= ~MO_AMASK;
628 }
629 tcg_temp_free_i32(tmp);
630 tcg_temp_free_i32(addr);
631
632 gen_neon_ldst_base_update(s, a->rm, a->rn, (1 << size) * nregs);
633
634 return true;
635 }
636
637 static bool trans_VLDST_single(DisasContext *s, arg_VLDST_single *a)
638 {
639 /* Neon load/store single structure to one lane */
640 int reg;
641 int nregs = a->n + 1;
642 int vd = a->vd;
643 TCGv_i32 addr, tmp;
644 MemOp mop;
645
646 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
647 return false;
648 }
649
650 /* UNDEF accesses to D16-D31 if they don't exist */
651 if (!dc_isar_feature(aa32_simd_r32, s) && (a->vd & 0x10)) {
652 return false;
653 }
654
655 /* Catch the UNDEF cases. This is unavoidably a bit messy. */
656 switch (nregs) {
657 case 1:
658 if (a->stride != 1) {
659 return false;
660 }
661 if (((a->align & (1 << a->size)) != 0) ||
662 (a->size == 2 && (a->align == 1 || a->align == 2))) {
663 return false;
664 }
665 break;
666 case 2:
667 if (a->size == 2 && (a->align & 2) != 0) {
668 return false;
669 }
670 break;
671 case 3:
672 if (a->align != 0) {
673 return false;
674 }
675 break;
676 case 4:
677 if (a->size == 2 && a->align == 3) {
678 return false;
679 }
680 break;
681 default:
682 g_assert_not_reached();
683 }
684 if ((vd + a->stride * (nregs - 1)) > 31) {
685 /*
686 * Attempts to write off the end of the register file are
687 * UNPREDICTABLE; we choose to UNDEF because otherwise we would
688 * access off the end of the array that holds the register data.
689 */
690 return false;
691 }
692
693 if (!vfp_access_check(s)) {
694 return true;
695 }
696
697 /* Pick up SCTLR settings */
698 mop = finalize_memop(s, a->size);
699
700 if (a->align) {
701 MemOp align_op;
702
703 switch (nregs) {
704 case 1:
705 /* For VLD1, use natural alignment. */
706 align_op = MO_ALIGN;
707 break;
708 case 2:
709 /* For VLD2, use double alignment. */
710 align_op = pow2_align(a->size + 1);
711 break;
712 case 4:
713 if (a->size == MO_32) {
714 /*
715 * For VLD4.32, align = 1 is double alignment, align = 2 is
716 * quad alignment; align = 3 is rejected above.
717 */
718 align_op = pow2_align(a->size + a->align);
719 } else {
720 /* For VLD4.8 and VLD.16, we want quad alignment. */
721 align_op = pow2_align(a->size + 2);
722 }
723 break;
724 default:
725 /* For VLD3, the alignment field is zero and rejected above. */
726 g_assert_not_reached();
727 }
728
729 mop = (mop & ~MO_AMASK) | align_op;
730 }
731
732 tmp = tcg_temp_new_i32();
733 addr = tcg_temp_new_i32();
734 load_reg_var(s, addr, a->rn);
735
736 for (reg = 0; reg < nregs; reg++) {
737 if (a->l) {
738 gen_aa32_ld_internal_i32(s, tmp, addr, get_mem_index(s), mop);
739 neon_store_element(vd, a->reg_idx, a->size, tmp);
740 } else { /* Store */
741 neon_load_element(tmp, vd, a->reg_idx, a->size);
742 gen_aa32_st_internal_i32(s, tmp, addr, get_mem_index(s), mop);
743 }
744 vd += a->stride;
745 tcg_gen_addi_i32(addr, addr, 1 << a->size);
746
747 /* Subsequent memory operations inherit alignment */
748 mop &= ~MO_AMASK;
749 }
750 tcg_temp_free_i32(addr);
751 tcg_temp_free_i32(tmp);
752
753 gen_neon_ldst_base_update(s, a->rm, a->rn, (1 << a->size) * nregs);
754
755 return true;
756 }
757
758 static bool do_3same(DisasContext *s, arg_3same *a, GVecGen3Fn fn)
759 {
760 int vec_size = a->q ? 16 : 8;
761 int rd_ofs = neon_full_reg_offset(a->vd);
762 int rn_ofs = neon_full_reg_offset(a->vn);
763 int rm_ofs = neon_full_reg_offset(a->vm);
764
765 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
766 return false;
767 }
768
769 /* UNDEF accesses to D16-D31 if they don't exist. */
770 if (!dc_isar_feature(aa32_simd_r32, s) &&
771 ((a->vd | a->vn | a->vm) & 0x10)) {
772 return false;
773 }
774
775 if ((a->vn | a->vm | a->vd) & a->q) {
776 return false;
777 }
778
779 if (!vfp_access_check(s)) {
780 return true;
781 }
782
783 fn(a->size, rd_ofs, rn_ofs, rm_ofs, vec_size, vec_size);
784 return true;
785 }
786
787 #define DO_3SAME(INSN, FUNC) \
788 static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a) \
789 { \
790 return do_3same(s, a, FUNC); \
791 }
792
793 DO_3SAME(VADD, tcg_gen_gvec_add)
794 DO_3SAME(VSUB, tcg_gen_gvec_sub)
795 DO_3SAME(VAND, tcg_gen_gvec_and)
796 DO_3SAME(VBIC, tcg_gen_gvec_andc)
797 DO_3SAME(VORR, tcg_gen_gvec_or)
798 DO_3SAME(VORN, tcg_gen_gvec_orc)
799 DO_3SAME(VEOR, tcg_gen_gvec_xor)
800 DO_3SAME(VSHL_S, gen_gvec_sshl)
801 DO_3SAME(VSHL_U, gen_gvec_ushl)
802 DO_3SAME(VQADD_S, gen_gvec_sqadd_qc)
803 DO_3SAME(VQADD_U, gen_gvec_uqadd_qc)
804 DO_3SAME(VQSUB_S, gen_gvec_sqsub_qc)
805 DO_3SAME(VQSUB_U, gen_gvec_uqsub_qc)
806
807 /* These insns are all gvec_bitsel but with the inputs in various orders. */
808 #define DO_3SAME_BITSEL(INSN, O1, O2, O3) \
809 static void gen_##INSN##_3s(unsigned vece, uint32_t rd_ofs, \
810 uint32_t rn_ofs, uint32_t rm_ofs, \
811 uint32_t oprsz, uint32_t maxsz) \
812 { \
813 tcg_gen_gvec_bitsel(vece, rd_ofs, O1, O2, O3, oprsz, maxsz); \
814 } \
815 DO_3SAME(INSN, gen_##INSN##_3s)
816
817 DO_3SAME_BITSEL(VBSL, rd_ofs, rn_ofs, rm_ofs)
818 DO_3SAME_BITSEL(VBIT, rm_ofs, rn_ofs, rd_ofs)
819 DO_3SAME_BITSEL(VBIF, rm_ofs, rd_ofs, rn_ofs)
820
821 #define DO_3SAME_NO_SZ_3(INSN, FUNC) \
822 static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a) \
823 { \
824 if (a->size == 3) { \
825 return false; \
826 } \
827 return do_3same(s, a, FUNC); \
828 }
829
830 DO_3SAME_NO_SZ_3(VMAX_S, tcg_gen_gvec_smax)
831 DO_3SAME_NO_SZ_3(VMAX_U, tcg_gen_gvec_umax)
832 DO_3SAME_NO_SZ_3(VMIN_S, tcg_gen_gvec_smin)
833 DO_3SAME_NO_SZ_3(VMIN_U, tcg_gen_gvec_umin)
834 DO_3SAME_NO_SZ_3(VMUL, tcg_gen_gvec_mul)
835 DO_3SAME_NO_SZ_3(VMLA, gen_gvec_mla)
836 DO_3SAME_NO_SZ_3(VMLS, gen_gvec_mls)
837 DO_3SAME_NO_SZ_3(VTST, gen_gvec_cmtst)
838 DO_3SAME_NO_SZ_3(VABD_S, gen_gvec_sabd)
839 DO_3SAME_NO_SZ_3(VABA_S, gen_gvec_saba)
840 DO_3SAME_NO_SZ_3(VABD_U, gen_gvec_uabd)
841 DO_3SAME_NO_SZ_3(VABA_U, gen_gvec_uaba)
842
843 #define DO_3SAME_CMP(INSN, COND) \
844 static void gen_##INSN##_3s(unsigned vece, uint32_t rd_ofs, \
845 uint32_t rn_ofs, uint32_t rm_ofs, \
846 uint32_t oprsz, uint32_t maxsz) \
847 { \
848 tcg_gen_gvec_cmp(COND, vece, rd_ofs, rn_ofs, rm_ofs, oprsz, maxsz); \
849 } \
850 DO_3SAME_NO_SZ_3(INSN, gen_##INSN##_3s)
851
852 DO_3SAME_CMP(VCGT_S, TCG_COND_GT)
853 DO_3SAME_CMP(VCGT_U, TCG_COND_GTU)
854 DO_3SAME_CMP(VCGE_S, TCG_COND_GE)
855 DO_3SAME_CMP(VCGE_U, TCG_COND_GEU)
856 DO_3SAME_CMP(VCEQ, TCG_COND_EQ)
857
858 #define WRAP_OOL_FN(WRAPNAME, FUNC) \
859 static void WRAPNAME(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, \
860 uint32_t rm_ofs, uint32_t oprsz, uint32_t maxsz) \
861 { \
862 tcg_gen_gvec_3_ool(rd_ofs, rn_ofs, rm_ofs, oprsz, maxsz, 0, FUNC); \
863 }
864
865 WRAP_OOL_FN(gen_VMUL_p_3s, gen_helper_gvec_pmul_b)
866
867 static bool trans_VMUL_p_3s(DisasContext *s, arg_3same *a)
868 {
869 if (a->size != 0) {
870 return false;
871 }
872 return do_3same(s, a, gen_VMUL_p_3s);
873 }
874
875 #define DO_VQRDMLAH(INSN, FUNC) \
876 static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a) \
877 { \
878 if (!dc_isar_feature(aa32_rdm, s)) { \
879 return false; \
880 } \
881 if (a->size != 1 && a->size != 2) { \
882 return false; \
883 } \
884 return do_3same(s, a, FUNC); \
885 }
886
887 DO_VQRDMLAH(VQRDMLAH, gen_gvec_sqrdmlah_qc)
888 DO_VQRDMLAH(VQRDMLSH, gen_gvec_sqrdmlsh_qc)
889
890 #define DO_SHA1(NAME, FUNC) \
891 WRAP_OOL_FN(gen_##NAME##_3s, FUNC) \
892 static bool trans_##NAME##_3s(DisasContext *s, arg_3same *a) \
893 { \
894 if (!dc_isar_feature(aa32_sha1, s)) { \
895 return false; \
896 } \
897 return do_3same(s, a, gen_##NAME##_3s); \
898 }
899
900 DO_SHA1(SHA1C, gen_helper_crypto_sha1c)
901 DO_SHA1(SHA1P, gen_helper_crypto_sha1p)
902 DO_SHA1(SHA1M, gen_helper_crypto_sha1m)
903 DO_SHA1(SHA1SU0, gen_helper_crypto_sha1su0)
904
905 #define DO_SHA2(NAME, FUNC) \
906 WRAP_OOL_FN(gen_##NAME##_3s, FUNC) \
907 static bool trans_##NAME##_3s(DisasContext *s, arg_3same *a) \
908 { \
909 if (!dc_isar_feature(aa32_sha2, s)) { \
910 return false; \
911 } \
912 return do_3same(s, a, gen_##NAME##_3s); \
913 }
914
915 DO_SHA2(SHA256H, gen_helper_crypto_sha256h)
916 DO_SHA2(SHA256H2, gen_helper_crypto_sha256h2)
917 DO_SHA2(SHA256SU1, gen_helper_crypto_sha256su1)
918
919 #define DO_3SAME_64(INSN, FUNC) \
920 static void gen_##INSN##_3s(unsigned vece, uint32_t rd_ofs, \
921 uint32_t rn_ofs, uint32_t rm_ofs, \
922 uint32_t oprsz, uint32_t maxsz) \
923 { \
924 static const GVecGen3 op = { .fni8 = FUNC }; \
925 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, oprsz, maxsz, &op); \
926 } \
927 DO_3SAME(INSN, gen_##INSN##_3s)
928
929 #define DO_3SAME_64_ENV(INSN, FUNC) \
930 static void gen_##INSN##_elt(TCGv_i64 d, TCGv_i64 n, TCGv_i64 m) \
931 { \
932 FUNC(d, cpu_env, n, m); \
933 } \
934 DO_3SAME_64(INSN, gen_##INSN##_elt)
935
936 DO_3SAME_64(VRSHL_S64, gen_helper_neon_rshl_s64)
937 DO_3SAME_64(VRSHL_U64, gen_helper_neon_rshl_u64)
938 DO_3SAME_64_ENV(VQSHL_S64, gen_helper_neon_qshl_s64)
939 DO_3SAME_64_ENV(VQSHL_U64, gen_helper_neon_qshl_u64)
940 DO_3SAME_64_ENV(VQRSHL_S64, gen_helper_neon_qrshl_s64)
941 DO_3SAME_64_ENV(VQRSHL_U64, gen_helper_neon_qrshl_u64)
942
943 #define DO_3SAME_32(INSN, FUNC) \
944 static void gen_##INSN##_3s(unsigned vece, uint32_t rd_ofs, \
945 uint32_t rn_ofs, uint32_t rm_ofs, \
946 uint32_t oprsz, uint32_t maxsz) \
947 { \
948 static const GVecGen3 ops[4] = { \
949 { .fni4 = gen_helper_neon_##FUNC##8 }, \
950 { .fni4 = gen_helper_neon_##FUNC##16 }, \
951 { .fni4 = gen_helper_neon_##FUNC##32 }, \
952 { 0 }, \
953 }; \
954 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, oprsz, maxsz, &ops[vece]); \
955 } \
956 static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a) \
957 { \
958 if (a->size > 2) { \
959 return false; \
960 } \
961 return do_3same(s, a, gen_##INSN##_3s); \
962 }
963
964 /*
965 * Some helper functions need to be passed the cpu_env. In order
966 * to use those with the gvec APIs like tcg_gen_gvec_3() we need
967 * to create wrapper functions whose prototype is a NeonGenTwoOpFn()
968 * and which call a NeonGenTwoOpEnvFn().
969 */
970 #define WRAP_ENV_FN(WRAPNAME, FUNC) \
971 static void WRAPNAME(TCGv_i32 d, TCGv_i32 n, TCGv_i32 m) \
972 { \
973 FUNC(d, cpu_env, n, m); \
974 }
975
976 #define DO_3SAME_32_ENV(INSN, FUNC) \
977 WRAP_ENV_FN(gen_##INSN##_tramp8, gen_helper_neon_##FUNC##8); \
978 WRAP_ENV_FN(gen_##INSN##_tramp16, gen_helper_neon_##FUNC##16); \
979 WRAP_ENV_FN(gen_##INSN##_tramp32, gen_helper_neon_##FUNC##32); \
980 static void gen_##INSN##_3s(unsigned vece, uint32_t rd_ofs, \
981 uint32_t rn_ofs, uint32_t rm_ofs, \
982 uint32_t oprsz, uint32_t maxsz) \
983 { \
984 static const GVecGen3 ops[4] = { \
985 { .fni4 = gen_##INSN##_tramp8 }, \
986 { .fni4 = gen_##INSN##_tramp16 }, \
987 { .fni4 = gen_##INSN##_tramp32 }, \
988 { 0 }, \
989 }; \
990 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, oprsz, maxsz, &ops[vece]); \
991 } \
992 static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a) \
993 { \
994 if (a->size > 2) { \
995 return false; \
996 } \
997 return do_3same(s, a, gen_##INSN##_3s); \
998 }
999
1000 DO_3SAME_32(VHADD_S, hadd_s)
1001 DO_3SAME_32(VHADD_U, hadd_u)
1002 DO_3SAME_32(VHSUB_S, hsub_s)
1003 DO_3SAME_32(VHSUB_U, hsub_u)
1004 DO_3SAME_32(VRHADD_S, rhadd_s)
1005 DO_3SAME_32(VRHADD_U, rhadd_u)
1006 DO_3SAME_32(VRSHL_S, rshl_s)
1007 DO_3SAME_32(VRSHL_U, rshl_u)
1008
1009 DO_3SAME_32_ENV(VQSHL_S, qshl_s)
1010 DO_3SAME_32_ENV(VQSHL_U, qshl_u)
1011 DO_3SAME_32_ENV(VQRSHL_S, qrshl_s)
1012 DO_3SAME_32_ENV(VQRSHL_U, qrshl_u)
1013
1014 static bool do_3same_pair(DisasContext *s, arg_3same *a, NeonGenTwoOpFn *fn)
1015 {
1016 /* Operations handled pairwise 32 bits at a time */
1017 TCGv_i32 tmp, tmp2, tmp3;
1018
1019 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1020 return false;
1021 }
1022
1023 /* UNDEF accesses to D16-D31 if they don't exist. */
1024 if (!dc_isar_feature(aa32_simd_r32, s) &&
1025 ((a->vd | a->vn | a->vm) & 0x10)) {
1026 return false;
1027 }
1028
1029 if (a->size == 3) {
1030 return false;
1031 }
1032
1033 if (!vfp_access_check(s)) {
1034 return true;
1035 }
1036
1037 assert(a->q == 0); /* enforced by decode patterns */
1038
1039 /*
1040 * Note that we have to be careful not to clobber the source operands
1041 * in the "vm == vd" case by storing the result of the first pass too
1042 * early. Since Q is 0 there are always just two passes, so instead
1043 * of a complicated loop over each pass we just unroll.
1044 */
1045 tmp = tcg_temp_new_i32();
1046 tmp2 = tcg_temp_new_i32();
1047 tmp3 = tcg_temp_new_i32();
1048
1049 read_neon_element32(tmp, a->vn, 0, MO_32);
1050 read_neon_element32(tmp2, a->vn, 1, MO_32);
1051 fn(tmp, tmp, tmp2);
1052
1053 read_neon_element32(tmp3, a->vm, 0, MO_32);
1054 read_neon_element32(tmp2, a->vm, 1, MO_32);
1055 fn(tmp3, tmp3, tmp2);
1056
1057 write_neon_element32(tmp, a->vd, 0, MO_32);
1058 write_neon_element32(tmp3, a->vd, 1, MO_32);
1059
1060 tcg_temp_free_i32(tmp);
1061 tcg_temp_free_i32(tmp2);
1062 tcg_temp_free_i32(tmp3);
1063 return true;
1064 }
1065
1066 #define DO_3SAME_PAIR(INSN, func) \
1067 static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a) \
1068 { \
1069 static NeonGenTwoOpFn * const fns[] = { \
1070 gen_helper_neon_##func##8, \
1071 gen_helper_neon_##func##16, \
1072 gen_helper_neon_##func##32, \
1073 }; \
1074 if (a->size > 2) { \
1075 return false; \
1076 } \
1077 return do_3same_pair(s, a, fns[a->size]); \
1078 }
1079
1080 /* 32-bit pairwise ops end up the same as the elementwise versions. */
1081 #define gen_helper_neon_pmax_s32 tcg_gen_smax_i32
1082 #define gen_helper_neon_pmax_u32 tcg_gen_umax_i32
1083 #define gen_helper_neon_pmin_s32 tcg_gen_smin_i32
1084 #define gen_helper_neon_pmin_u32 tcg_gen_umin_i32
1085 #define gen_helper_neon_padd_u32 tcg_gen_add_i32
1086
1087 DO_3SAME_PAIR(VPMAX_S, pmax_s)
1088 DO_3SAME_PAIR(VPMIN_S, pmin_s)
1089 DO_3SAME_PAIR(VPMAX_U, pmax_u)
1090 DO_3SAME_PAIR(VPMIN_U, pmin_u)
1091 DO_3SAME_PAIR(VPADD, padd_u)
1092
1093 #define DO_3SAME_VQDMULH(INSN, FUNC) \
1094 WRAP_ENV_FN(gen_##INSN##_tramp16, gen_helper_neon_##FUNC##_s16); \
1095 WRAP_ENV_FN(gen_##INSN##_tramp32, gen_helper_neon_##FUNC##_s32); \
1096 static void gen_##INSN##_3s(unsigned vece, uint32_t rd_ofs, \
1097 uint32_t rn_ofs, uint32_t rm_ofs, \
1098 uint32_t oprsz, uint32_t maxsz) \
1099 { \
1100 static const GVecGen3 ops[2] = { \
1101 { .fni4 = gen_##INSN##_tramp16 }, \
1102 { .fni4 = gen_##INSN##_tramp32 }, \
1103 }; \
1104 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, oprsz, maxsz, &ops[vece - 1]); \
1105 } \
1106 static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a) \
1107 { \
1108 if (a->size != 1 && a->size != 2) { \
1109 return false; \
1110 } \
1111 return do_3same(s, a, gen_##INSN##_3s); \
1112 }
1113
1114 DO_3SAME_VQDMULH(VQDMULH, qdmulh)
1115 DO_3SAME_VQDMULH(VQRDMULH, qrdmulh)
1116
1117 #define WRAP_FP_GVEC(WRAPNAME, FPST, FUNC) \
1118 static void WRAPNAME(unsigned vece, uint32_t rd_ofs, \
1119 uint32_t rn_ofs, uint32_t rm_ofs, \
1120 uint32_t oprsz, uint32_t maxsz) \
1121 { \
1122 TCGv_ptr fpst = fpstatus_ptr(FPST); \
1123 tcg_gen_gvec_3_ptr(rd_ofs, rn_ofs, rm_ofs, fpst, \
1124 oprsz, maxsz, 0, FUNC); \
1125 tcg_temp_free_ptr(fpst); \
1126 }
1127
1128 #define DO_3S_FP_GVEC(INSN,SFUNC,HFUNC) \
1129 WRAP_FP_GVEC(gen_##INSN##_fp32_3s, FPST_STD, SFUNC) \
1130 WRAP_FP_GVEC(gen_##INSN##_fp16_3s, FPST_STD_F16, HFUNC) \
1131 static bool trans_##INSN##_fp_3s(DisasContext *s, arg_3same *a) \
1132 { \
1133 if (a->size == MO_16) { \
1134 if (!dc_isar_feature(aa32_fp16_arith, s)) { \
1135 return false; \
1136 } \
1137 return do_3same(s, a, gen_##INSN##_fp16_3s); \
1138 } \
1139 return do_3same(s, a, gen_##INSN##_fp32_3s); \
1140 }
1141
1142
1143 DO_3S_FP_GVEC(VADD, gen_helper_gvec_fadd_s, gen_helper_gvec_fadd_h)
1144 DO_3S_FP_GVEC(VSUB, gen_helper_gvec_fsub_s, gen_helper_gvec_fsub_h)
1145 DO_3S_FP_GVEC(VABD, gen_helper_gvec_fabd_s, gen_helper_gvec_fabd_h)
1146 DO_3S_FP_GVEC(VMUL, gen_helper_gvec_fmul_s, gen_helper_gvec_fmul_h)
1147 DO_3S_FP_GVEC(VCEQ, gen_helper_gvec_fceq_s, gen_helper_gvec_fceq_h)
1148 DO_3S_FP_GVEC(VCGE, gen_helper_gvec_fcge_s, gen_helper_gvec_fcge_h)
1149 DO_3S_FP_GVEC(VCGT, gen_helper_gvec_fcgt_s, gen_helper_gvec_fcgt_h)
1150 DO_3S_FP_GVEC(VACGE, gen_helper_gvec_facge_s, gen_helper_gvec_facge_h)
1151 DO_3S_FP_GVEC(VACGT, gen_helper_gvec_facgt_s, gen_helper_gvec_facgt_h)
1152 DO_3S_FP_GVEC(VMAX, gen_helper_gvec_fmax_s, gen_helper_gvec_fmax_h)
1153 DO_3S_FP_GVEC(VMIN, gen_helper_gvec_fmin_s, gen_helper_gvec_fmin_h)
1154 DO_3S_FP_GVEC(VMLA, gen_helper_gvec_fmla_s, gen_helper_gvec_fmla_h)
1155 DO_3S_FP_GVEC(VMLS, gen_helper_gvec_fmls_s, gen_helper_gvec_fmls_h)
1156 DO_3S_FP_GVEC(VFMA, gen_helper_gvec_vfma_s, gen_helper_gvec_vfma_h)
1157 DO_3S_FP_GVEC(VFMS, gen_helper_gvec_vfms_s, gen_helper_gvec_vfms_h)
1158 DO_3S_FP_GVEC(VRECPS, gen_helper_gvec_recps_nf_s, gen_helper_gvec_recps_nf_h)
1159 DO_3S_FP_GVEC(VRSQRTS, gen_helper_gvec_rsqrts_nf_s, gen_helper_gvec_rsqrts_nf_h)
1160
1161 WRAP_FP_GVEC(gen_VMAXNM_fp32_3s, FPST_STD, gen_helper_gvec_fmaxnum_s)
1162 WRAP_FP_GVEC(gen_VMAXNM_fp16_3s, FPST_STD_F16, gen_helper_gvec_fmaxnum_h)
1163 WRAP_FP_GVEC(gen_VMINNM_fp32_3s, FPST_STD, gen_helper_gvec_fminnum_s)
1164 WRAP_FP_GVEC(gen_VMINNM_fp16_3s, FPST_STD_F16, gen_helper_gvec_fminnum_h)
1165
1166 static bool trans_VMAXNM_fp_3s(DisasContext *s, arg_3same *a)
1167 {
1168 if (!arm_dc_feature(s, ARM_FEATURE_V8)) {
1169 return false;
1170 }
1171
1172 if (a->size == MO_16) {
1173 if (!dc_isar_feature(aa32_fp16_arith, s)) {
1174 return false;
1175 }
1176 return do_3same(s, a, gen_VMAXNM_fp16_3s);
1177 }
1178 return do_3same(s, a, gen_VMAXNM_fp32_3s);
1179 }
1180
1181 static bool trans_VMINNM_fp_3s(DisasContext *s, arg_3same *a)
1182 {
1183 if (!arm_dc_feature(s, ARM_FEATURE_V8)) {
1184 return false;
1185 }
1186
1187 if (a->size == MO_16) {
1188 if (!dc_isar_feature(aa32_fp16_arith, s)) {
1189 return false;
1190 }
1191 return do_3same(s, a, gen_VMINNM_fp16_3s);
1192 }
1193 return do_3same(s, a, gen_VMINNM_fp32_3s);
1194 }
1195
1196 static bool do_3same_fp_pair(DisasContext *s, arg_3same *a,
1197 gen_helper_gvec_3_ptr *fn)
1198 {
1199 /* FP pairwise operations */
1200 TCGv_ptr fpstatus;
1201
1202 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1203 return false;
1204 }
1205
1206 /* UNDEF accesses to D16-D31 if they don't exist. */
1207 if (!dc_isar_feature(aa32_simd_r32, s) &&
1208 ((a->vd | a->vn | a->vm) & 0x10)) {
1209 return false;
1210 }
1211
1212 if (!vfp_access_check(s)) {
1213 return true;
1214 }
1215
1216 assert(a->q == 0); /* enforced by decode patterns */
1217
1218
1219 fpstatus = fpstatus_ptr(a->size == MO_16 ? FPST_STD_F16 : FPST_STD);
1220 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, a->vd),
1221 vfp_reg_offset(1, a->vn),
1222 vfp_reg_offset(1, a->vm),
1223 fpstatus, 8, 8, 0, fn);
1224 tcg_temp_free_ptr(fpstatus);
1225
1226 return true;
1227 }
1228
1229 /*
1230 * For all the functions using this macro, size == 1 means fp16,
1231 * which is an architecture extension we don't implement yet.
1232 */
1233 #define DO_3S_FP_PAIR(INSN,FUNC) \
1234 static bool trans_##INSN##_fp_3s(DisasContext *s, arg_3same *a) \
1235 { \
1236 if (a->size == MO_16) { \
1237 if (!dc_isar_feature(aa32_fp16_arith, s)) { \
1238 return false; \
1239 } \
1240 return do_3same_fp_pair(s, a, FUNC##h); \
1241 } \
1242 return do_3same_fp_pair(s, a, FUNC##s); \
1243 }
1244
1245 DO_3S_FP_PAIR(VPADD, gen_helper_neon_padd)
1246 DO_3S_FP_PAIR(VPMAX, gen_helper_neon_pmax)
1247 DO_3S_FP_PAIR(VPMIN, gen_helper_neon_pmin)
1248
1249 static bool do_vector_2sh(DisasContext *s, arg_2reg_shift *a, GVecGen2iFn *fn)
1250 {
1251 /* Handle a 2-reg-shift insn which can be vectorized. */
1252 int vec_size = a->q ? 16 : 8;
1253 int rd_ofs = neon_full_reg_offset(a->vd);
1254 int rm_ofs = neon_full_reg_offset(a->vm);
1255
1256 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1257 return false;
1258 }
1259
1260 /* UNDEF accesses to D16-D31 if they don't exist. */
1261 if (!dc_isar_feature(aa32_simd_r32, s) &&
1262 ((a->vd | a->vm) & 0x10)) {
1263 return false;
1264 }
1265
1266 if ((a->vm | a->vd) & a->q) {
1267 return false;
1268 }
1269
1270 if (!vfp_access_check(s)) {
1271 return true;
1272 }
1273
1274 fn(a->size, rd_ofs, rm_ofs, a->shift, vec_size, vec_size);
1275 return true;
1276 }
1277
1278 #define DO_2SH(INSN, FUNC) \
1279 static bool trans_##INSN##_2sh(DisasContext *s, arg_2reg_shift *a) \
1280 { \
1281 return do_vector_2sh(s, a, FUNC); \
1282 } \
1283
1284 DO_2SH(VSHL, tcg_gen_gvec_shli)
1285 DO_2SH(VSLI, gen_gvec_sli)
1286 DO_2SH(VSRI, gen_gvec_sri)
1287 DO_2SH(VSRA_S, gen_gvec_ssra)
1288 DO_2SH(VSRA_U, gen_gvec_usra)
1289 DO_2SH(VRSHR_S, gen_gvec_srshr)
1290 DO_2SH(VRSHR_U, gen_gvec_urshr)
1291 DO_2SH(VRSRA_S, gen_gvec_srsra)
1292 DO_2SH(VRSRA_U, gen_gvec_ursra)
1293
1294 static bool trans_VSHR_S_2sh(DisasContext *s, arg_2reg_shift *a)
1295 {
1296 /* Signed shift out of range results in all-sign-bits */
1297 a->shift = MIN(a->shift, (8 << a->size) - 1);
1298 return do_vector_2sh(s, a, tcg_gen_gvec_sari);
1299 }
1300
1301 static void gen_zero_rd_2sh(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
1302 int64_t shift, uint32_t oprsz, uint32_t maxsz)
1303 {
1304 tcg_gen_gvec_dup_imm(vece, rd_ofs, oprsz, maxsz, 0);
1305 }
1306
1307 static bool trans_VSHR_U_2sh(DisasContext *s, arg_2reg_shift *a)
1308 {
1309 /* Shift out of range is architecturally valid and results in zero. */
1310 if (a->shift >= (8 << a->size)) {
1311 return do_vector_2sh(s, a, gen_zero_rd_2sh);
1312 } else {
1313 return do_vector_2sh(s, a, tcg_gen_gvec_shri);
1314 }
1315 }
1316
1317 static bool do_2shift_env_64(DisasContext *s, arg_2reg_shift *a,
1318 NeonGenTwo64OpEnvFn *fn)
1319 {
1320 /*
1321 * 2-reg-and-shift operations, size == 3 case, where the
1322 * function needs to be passed cpu_env.
1323 */
1324 TCGv_i64 constimm;
1325 int pass;
1326
1327 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1328 return false;
1329 }
1330
1331 /* UNDEF accesses to D16-D31 if they don't exist. */
1332 if (!dc_isar_feature(aa32_simd_r32, s) &&
1333 ((a->vd | a->vm) & 0x10)) {
1334 return false;
1335 }
1336
1337 if ((a->vm | a->vd) & a->q) {
1338 return false;
1339 }
1340
1341 if (!vfp_access_check(s)) {
1342 return true;
1343 }
1344
1345 /*
1346 * To avoid excessive duplication of ops we implement shift
1347 * by immediate using the variable shift operations.
1348 */
1349 constimm = tcg_constant_i64(dup_const(a->size, a->shift));
1350
1351 for (pass = 0; pass < a->q + 1; pass++) {
1352 TCGv_i64 tmp = tcg_temp_new_i64();
1353
1354 read_neon_element64(tmp, a->vm, pass, MO_64);
1355 fn(tmp, cpu_env, tmp, constimm);
1356 write_neon_element64(tmp, a->vd, pass, MO_64);
1357 tcg_temp_free_i64(tmp);
1358 }
1359 return true;
1360 }
1361
1362 static bool do_2shift_env_32(DisasContext *s, arg_2reg_shift *a,
1363 NeonGenTwoOpEnvFn *fn)
1364 {
1365 /*
1366 * 2-reg-and-shift operations, size < 3 case, where the
1367 * helper needs to be passed cpu_env.
1368 */
1369 TCGv_i32 constimm, tmp;
1370 int pass;
1371
1372 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1373 return false;
1374 }
1375
1376 /* UNDEF accesses to D16-D31 if they don't exist. */
1377 if (!dc_isar_feature(aa32_simd_r32, s) &&
1378 ((a->vd | a->vm) & 0x10)) {
1379 return false;
1380 }
1381
1382 if ((a->vm | a->vd) & a->q) {
1383 return false;
1384 }
1385
1386 if (!vfp_access_check(s)) {
1387 return true;
1388 }
1389
1390 /*
1391 * To avoid excessive duplication of ops we implement shift
1392 * by immediate using the variable shift operations.
1393 */
1394 constimm = tcg_constant_i32(dup_const(a->size, a->shift));
1395 tmp = tcg_temp_new_i32();
1396
1397 for (pass = 0; pass < (a->q ? 4 : 2); pass++) {
1398 read_neon_element32(tmp, a->vm, pass, MO_32);
1399 fn(tmp, cpu_env, tmp, constimm);
1400 write_neon_element32(tmp, a->vd, pass, MO_32);
1401 }
1402 tcg_temp_free_i32(tmp);
1403 return true;
1404 }
1405
1406 #define DO_2SHIFT_ENV(INSN, FUNC) \
1407 static bool trans_##INSN##_64_2sh(DisasContext *s, arg_2reg_shift *a) \
1408 { \
1409 return do_2shift_env_64(s, a, gen_helper_neon_##FUNC##64); \
1410 } \
1411 static bool trans_##INSN##_2sh(DisasContext *s, arg_2reg_shift *a) \
1412 { \
1413 static NeonGenTwoOpEnvFn * const fns[] = { \
1414 gen_helper_neon_##FUNC##8, \
1415 gen_helper_neon_##FUNC##16, \
1416 gen_helper_neon_##FUNC##32, \
1417 }; \
1418 assert(a->size < ARRAY_SIZE(fns)); \
1419 return do_2shift_env_32(s, a, fns[a->size]); \
1420 }
1421
1422 DO_2SHIFT_ENV(VQSHLU, qshlu_s)
1423 DO_2SHIFT_ENV(VQSHL_U, qshl_u)
1424 DO_2SHIFT_ENV(VQSHL_S, qshl_s)
1425
1426 static bool do_2shift_narrow_64(DisasContext *s, arg_2reg_shift *a,
1427 NeonGenTwo64OpFn *shiftfn,
1428 NeonGenNarrowEnvFn *narrowfn)
1429 {
1430 /* 2-reg-and-shift narrowing-shift operations, size == 3 case */
1431 TCGv_i64 constimm, rm1, rm2;
1432 TCGv_i32 rd;
1433
1434 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1435 return false;
1436 }
1437
1438 /* UNDEF accesses to D16-D31 if they don't exist. */
1439 if (!dc_isar_feature(aa32_simd_r32, s) &&
1440 ((a->vd | a->vm) & 0x10)) {
1441 return false;
1442 }
1443
1444 if (a->vm & 1) {
1445 return false;
1446 }
1447
1448 if (!vfp_access_check(s)) {
1449 return true;
1450 }
1451
1452 /*
1453 * This is always a right shift, and the shiftfn is always a
1454 * left-shift helper, which thus needs the negated shift count.
1455 */
1456 constimm = tcg_constant_i64(-a->shift);
1457 rm1 = tcg_temp_new_i64();
1458 rm2 = tcg_temp_new_i64();
1459 rd = tcg_temp_new_i32();
1460
1461 /* Load both inputs first to avoid potential overwrite if rm == rd */
1462 read_neon_element64(rm1, a->vm, 0, MO_64);
1463 read_neon_element64(rm2, a->vm, 1, MO_64);
1464
1465 shiftfn(rm1, rm1, constimm);
1466 narrowfn(rd, cpu_env, rm1);
1467 write_neon_element32(rd, a->vd, 0, MO_32);
1468
1469 shiftfn(rm2, rm2, constimm);
1470 narrowfn(rd, cpu_env, rm2);
1471 write_neon_element32(rd, a->vd, 1, MO_32);
1472
1473 tcg_temp_free_i32(rd);
1474 tcg_temp_free_i64(rm1);
1475 tcg_temp_free_i64(rm2);
1476
1477 return true;
1478 }
1479
1480 static bool do_2shift_narrow_32(DisasContext *s, arg_2reg_shift *a,
1481 NeonGenTwoOpFn *shiftfn,
1482 NeonGenNarrowEnvFn *narrowfn)
1483 {
1484 /* 2-reg-and-shift narrowing-shift operations, size < 3 case */
1485 TCGv_i32 constimm, rm1, rm2, rm3, rm4;
1486 TCGv_i64 rtmp;
1487 uint32_t imm;
1488
1489 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1490 return false;
1491 }
1492
1493 /* UNDEF accesses to D16-D31 if they don't exist. */
1494 if (!dc_isar_feature(aa32_simd_r32, s) &&
1495 ((a->vd | a->vm) & 0x10)) {
1496 return false;
1497 }
1498
1499 if (a->vm & 1) {
1500 return false;
1501 }
1502
1503 if (!vfp_access_check(s)) {
1504 return true;
1505 }
1506
1507 /*
1508 * This is always a right shift, and the shiftfn is always a
1509 * left-shift helper, which thus needs the negated shift count
1510 * duplicated into each lane of the immediate value.
1511 */
1512 if (a->size == 1) {
1513 imm = (uint16_t)(-a->shift);
1514 imm |= imm << 16;
1515 } else {
1516 /* size == 2 */
1517 imm = -a->shift;
1518 }
1519 constimm = tcg_constant_i32(imm);
1520
1521 /* Load all inputs first to avoid potential overwrite */
1522 rm1 = tcg_temp_new_i32();
1523 rm2 = tcg_temp_new_i32();
1524 rm3 = tcg_temp_new_i32();
1525 rm4 = tcg_temp_new_i32();
1526 read_neon_element32(rm1, a->vm, 0, MO_32);
1527 read_neon_element32(rm2, a->vm, 1, MO_32);
1528 read_neon_element32(rm3, a->vm, 2, MO_32);
1529 read_neon_element32(rm4, a->vm, 3, MO_32);
1530 rtmp = tcg_temp_new_i64();
1531
1532 shiftfn(rm1, rm1, constimm);
1533 shiftfn(rm2, rm2, constimm);
1534
1535 tcg_gen_concat_i32_i64(rtmp, rm1, rm2);
1536 tcg_temp_free_i32(rm2);
1537
1538 narrowfn(rm1, cpu_env, rtmp);
1539 write_neon_element32(rm1, a->vd, 0, MO_32);
1540 tcg_temp_free_i32(rm1);
1541
1542 shiftfn(rm3, rm3, constimm);
1543 shiftfn(rm4, rm4, constimm);
1544
1545 tcg_gen_concat_i32_i64(rtmp, rm3, rm4);
1546 tcg_temp_free_i32(rm4);
1547
1548 narrowfn(rm3, cpu_env, rtmp);
1549 tcg_temp_free_i64(rtmp);
1550 write_neon_element32(rm3, a->vd, 1, MO_32);
1551 tcg_temp_free_i32(rm3);
1552 return true;
1553 }
1554
1555 #define DO_2SN_64(INSN, FUNC, NARROWFUNC) \
1556 static bool trans_##INSN##_2sh(DisasContext *s, arg_2reg_shift *a) \
1557 { \
1558 return do_2shift_narrow_64(s, a, FUNC, NARROWFUNC); \
1559 }
1560 #define DO_2SN_32(INSN, FUNC, NARROWFUNC) \
1561 static bool trans_##INSN##_2sh(DisasContext *s, arg_2reg_shift *a) \
1562 { \
1563 return do_2shift_narrow_32(s, a, FUNC, NARROWFUNC); \
1564 }
1565
1566 static void gen_neon_narrow_u32(TCGv_i32 dest, TCGv_ptr env, TCGv_i64 src)
1567 {
1568 tcg_gen_extrl_i64_i32(dest, src);
1569 }
1570
1571 static void gen_neon_narrow_u16(TCGv_i32 dest, TCGv_ptr env, TCGv_i64 src)
1572 {
1573 gen_helper_neon_narrow_u16(dest, src);
1574 }
1575
1576 static void gen_neon_narrow_u8(TCGv_i32 dest, TCGv_ptr env, TCGv_i64 src)
1577 {
1578 gen_helper_neon_narrow_u8(dest, src);
1579 }
1580
1581 DO_2SN_64(VSHRN_64, gen_ushl_i64, gen_neon_narrow_u32)
1582 DO_2SN_32(VSHRN_32, gen_ushl_i32, gen_neon_narrow_u16)
1583 DO_2SN_32(VSHRN_16, gen_helper_neon_shl_u16, gen_neon_narrow_u8)
1584
1585 DO_2SN_64(VRSHRN_64, gen_helper_neon_rshl_u64, gen_neon_narrow_u32)
1586 DO_2SN_32(VRSHRN_32, gen_helper_neon_rshl_u32, gen_neon_narrow_u16)
1587 DO_2SN_32(VRSHRN_16, gen_helper_neon_rshl_u16, gen_neon_narrow_u8)
1588
1589 DO_2SN_64(VQSHRUN_64, gen_sshl_i64, gen_helper_neon_unarrow_sat32)
1590 DO_2SN_32(VQSHRUN_32, gen_sshl_i32, gen_helper_neon_unarrow_sat16)
1591 DO_2SN_32(VQSHRUN_16, gen_helper_neon_shl_s16, gen_helper_neon_unarrow_sat8)
1592
1593 DO_2SN_64(VQRSHRUN_64, gen_helper_neon_rshl_s64, gen_helper_neon_unarrow_sat32)
1594 DO_2SN_32(VQRSHRUN_32, gen_helper_neon_rshl_s32, gen_helper_neon_unarrow_sat16)
1595 DO_2SN_32(VQRSHRUN_16, gen_helper_neon_rshl_s16, gen_helper_neon_unarrow_sat8)
1596 DO_2SN_64(VQSHRN_S64, gen_sshl_i64, gen_helper_neon_narrow_sat_s32)
1597 DO_2SN_32(VQSHRN_S32, gen_sshl_i32, gen_helper_neon_narrow_sat_s16)
1598 DO_2SN_32(VQSHRN_S16, gen_helper_neon_shl_s16, gen_helper_neon_narrow_sat_s8)
1599
1600 DO_2SN_64(VQRSHRN_S64, gen_helper_neon_rshl_s64, gen_helper_neon_narrow_sat_s32)
1601 DO_2SN_32(VQRSHRN_S32, gen_helper_neon_rshl_s32, gen_helper_neon_narrow_sat_s16)
1602 DO_2SN_32(VQRSHRN_S16, gen_helper_neon_rshl_s16, gen_helper_neon_narrow_sat_s8)
1603
1604 DO_2SN_64(VQSHRN_U64, gen_ushl_i64, gen_helper_neon_narrow_sat_u32)
1605 DO_2SN_32(VQSHRN_U32, gen_ushl_i32, gen_helper_neon_narrow_sat_u16)
1606 DO_2SN_32(VQSHRN_U16, gen_helper_neon_shl_u16, gen_helper_neon_narrow_sat_u8)
1607
1608 DO_2SN_64(VQRSHRN_U64, gen_helper_neon_rshl_u64, gen_helper_neon_narrow_sat_u32)
1609 DO_2SN_32(VQRSHRN_U32, gen_helper_neon_rshl_u32, gen_helper_neon_narrow_sat_u16)
1610 DO_2SN_32(VQRSHRN_U16, gen_helper_neon_rshl_u16, gen_helper_neon_narrow_sat_u8)
1611
1612 static bool do_vshll_2sh(DisasContext *s, arg_2reg_shift *a,
1613 NeonGenWidenFn *widenfn, bool u)
1614 {
1615 TCGv_i64 tmp;
1616 TCGv_i32 rm0, rm1;
1617 uint64_t widen_mask = 0;
1618
1619 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1620 return false;
1621 }
1622
1623 /* UNDEF accesses to D16-D31 if they don't exist. */
1624 if (!dc_isar_feature(aa32_simd_r32, s) &&
1625 ((a->vd | a->vm) & 0x10)) {
1626 return false;
1627 }
1628
1629 if (a->vd & 1) {
1630 return false;
1631 }
1632
1633 if (!vfp_access_check(s)) {
1634 return true;
1635 }
1636
1637 /*
1638 * This is a widen-and-shift operation. The shift is always less
1639 * than the width of the source type, so after widening the input
1640 * vector we can simply shift the whole 64-bit widened register,
1641 * and then clear the potential overflow bits resulting from left
1642 * bits of the narrow input appearing as right bits of the left
1643 * neighbour narrow input. Calculate a mask of bits to clear.
1644 */
1645 if ((a->shift != 0) && (a->size < 2 || u)) {
1646 int esize = 8 << a->size;
1647 widen_mask = MAKE_64BIT_MASK(0, esize);
1648 widen_mask >>= esize - a->shift;
1649 widen_mask = dup_const(a->size + 1, widen_mask);
1650 }
1651
1652 rm0 = tcg_temp_new_i32();
1653 rm1 = tcg_temp_new_i32();
1654 read_neon_element32(rm0, a->vm, 0, MO_32);
1655 read_neon_element32(rm1, a->vm, 1, MO_32);
1656 tmp = tcg_temp_new_i64();
1657
1658 widenfn(tmp, rm0);
1659 tcg_temp_free_i32(rm0);
1660 if (a->shift != 0) {
1661 tcg_gen_shli_i64(tmp, tmp, a->shift);
1662 tcg_gen_andi_i64(tmp, tmp, ~widen_mask);
1663 }
1664 write_neon_element64(tmp, a->vd, 0, MO_64);
1665
1666 widenfn(tmp, rm1);
1667 tcg_temp_free_i32(rm1);
1668 if (a->shift != 0) {
1669 tcg_gen_shli_i64(tmp, tmp, a->shift);
1670 tcg_gen_andi_i64(tmp, tmp, ~widen_mask);
1671 }
1672 write_neon_element64(tmp, a->vd, 1, MO_64);
1673 tcg_temp_free_i64(tmp);
1674 return true;
1675 }
1676
1677 static bool trans_VSHLL_S_2sh(DisasContext *s, arg_2reg_shift *a)
1678 {
1679 static NeonGenWidenFn * const widenfn[] = {
1680 gen_helper_neon_widen_s8,
1681 gen_helper_neon_widen_s16,
1682 tcg_gen_ext_i32_i64,
1683 };
1684 return do_vshll_2sh(s, a, widenfn[a->size], false);
1685 }
1686
1687 static bool trans_VSHLL_U_2sh(DisasContext *s, arg_2reg_shift *a)
1688 {
1689 static NeonGenWidenFn * const widenfn[] = {
1690 gen_helper_neon_widen_u8,
1691 gen_helper_neon_widen_u16,
1692 tcg_gen_extu_i32_i64,
1693 };
1694 return do_vshll_2sh(s, a, widenfn[a->size], true);
1695 }
1696
1697 static bool do_fp_2sh(DisasContext *s, arg_2reg_shift *a,
1698 gen_helper_gvec_2_ptr *fn)
1699 {
1700 /* FP operations in 2-reg-and-shift group */
1701 int vec_size = a->q ? 16 : 8;
1702 int rd_ofs = neon_full_reg_offset(a->vd);
1703 int rm_ofs = neon_full_reg_offset(a->vm);
1704 TCGv_ptr fpst;
1705
1706 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1707 return false;
1708 }
1709
1710 if (a->size == MO_16) {
1711 if (!dc_isar_feature(aa32_fp16_arith, s)) {
1712 return false;
1713 }
1714 }
1715
1716 /* UNDEF accesses to D16-D31 if they don't exist. */
1717 if (!dc_isar_feature(aa32_simd_r32, s) &&
1718 ((a->vd | a->vm) & 0x10)) {
1719 return false;
1720 }
1721
1722 if ((a->vm | a->vd) & a->q) {
1723 return false;
1724 }
1725
1726 if (!vfp_access_check(s)) {
1727 return true;
1728 }
1729
1730 fpst = fpstatus_ptr(a->size == MO_16 ? FPST_STD_F16 : FPST_STD);
1731 tcg_gen_gvec_2_ptr(rd_ofs, rm_ofs, fpst, vec_size, vec_size, a->shift, fn);
1732 tcg_temp_free_ptr(fpst);
1733 return true;
1734 }
1735
1736 #define DO_FP_2SH(INSN, FUNC) \
1737 static bool trans_##INSN##_2sh(DisasContext *s, arg_2reg_shift *a) \
1738 { \
1739 return do_fp_2sh(s, a, FUNC); \
1740 }
1741
1742 DO_FP_2SH(VCVT_SF, gen_helper_gvec_vcvt_sf)
1743 DO_FP_2SH(VCVT_UF, gen_helper_gvec_vcvt_uf)
1744 DO_FP_2SH(VCVT_FS, gen_helper_gvec_vcvt_fs)
1745 DO_FP_2SH(VCVT_FU, gen_helper_gvec_vcvt_fu)
1746
1747 DO_FP_2SH(VCVT_SH, gen_helper_gvec_vcvt_sh)
1748 DO_FP_2SH(VCVT_UH, gen_helper_gvec_vcvt_uh)
1749 DO_FP_2SH(VCVT_HS, gen_helper_gvec_vcvt_hs)
1750 DO_FP_2SH(VCVT_HU, gen_helper_gvec_vcvt_hu)
1751
1752 static bool do_1reg_imm(DisasContext *s, arg_1reg_imm *a,
1753 GVecGen2iFn *fn)
1754 {
1755 uint64_t imm;
1756 int reg_ofs, vec_size;
1757
1758 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1759 return false;
1760 }
1761
1762 /* UNDEF accesses to D16-D31 if they don't exist. */
1763 if (!dc_isar_feature(aa32_simd_r32, s) && (a->vd & 0x10)) {
1764 return false;
1765 }
1766
1767 if (a->vd & a->q) {
1768 return false;
1769 }
1770
1771 if (!vfp_access_check(s)) {
1772 return true;
1773 }
1774
1775 reg_ofs = neon_full_reg_offset(a->vd);
1776 vec_size = a->q ? 16 : 8;
1777 imm = asimd_imm_const(a->imm, a->cmode, a->op);
1778
1779 fn(MO_64, reg_ofs, reg_ofs, imm, vec_size, vec_size);
1780 return true;
1781 }
1782
1783 static void gen_VMOV_1r(unsigned vece, uint32_t dofs, uint32_t aofs,
1784 int64_t c, uint32_t oprsz, uint32_t maxsz)
1785 {
1786 tcg_gen_gvec_dup_imm(MO_64, dofs, oprsz, maxsz, c);
1787 }
1788
1789 static bool trans_Vimm_1r(DisasContext *s, arg_1reg_imm *a)
1790 {
1791 /* Handle decode of cmode/op here between VORR/VBIC/VMOV */
1792 GVecGen2iFn *fn;
1793
1794 if ((a->cmode & 1) && a->cmode < 12) {
1795 /* for op=1, the imm will be inverted, so BIC becomes AND. */
1796 fn = a->op ? tcg_gen_gvec_andi : tcg_gen_gvec_ori;
1797 } else {
1798 /* There is one unallocated cmode/op combination in this space */
1799 if (a->cmode == 15 && a->op == 1) {
1800 return false;
1801 }
1802 fn = gen_VMOV_1r;
1803 }
1804 return do_1reg_imm(s, a, fn);
1805 }
1806
1807 static bool do_prewiden_3d(DisasContext *s, arg_3diff *a,
1808 NeonGenWidenFn *widenfn,
1809 NeonGenTwo64OpFn *opfn,
1810 int src1_mop, int src2_mop)
1811 {
1812 /* 3-regs different lengths, prewidening case (VADDL/VSUBL/VAADW/VSUBW) */
1813 TCGv_i64 rn0_64, rn1_64, rm_64;
1814
1815 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1816 return false;
1817 }
1818
1819 /* UNDEF accesses to D16-D31 if they don't exist. */
1820 if (!dc_isar_feature(aa32_simd_r32, s) &&
1821 ((a->vd | a->vn | a->vm) & 0x10)) {
1822 return false;
1823 }
1824
1825 if (!opfn) {
1826 /* size == 3 case, which is an entirely different insn group */
1827 return false;
1828 }
1829
1830 if ((a->vd & 1) || (src1_mop == MO_UQ && (a->vn & 1))) {
1831 return false;
1832 }
1833
1834 if (!vfp_access_check(s)) {
1835 return true;
1836 }
1837
1838 rn0_64 = tcg_temp_new_i64();
1839 rn1_64 = tcg_temp_new_i64();
1840 rm_64 = tcg_temp_new_i64();
1841
1842 if (src1_mop >= 0) {
1843 read_neon_element64(rn0_64, a->vn, 0, src1_mop);
1844 } else {
1845 TCGv_i32 tmp = tcg_temp_new_i32();
1846 read_neon_element32(tmp, a->vn, 0, MO_32);
1847 widenfn(rn0_64, tmp);
1848 tcg_temp_free_i32(tmp);
1849 }
1850 if (src2_mop >= 0) {
1851 read_neon_element64(rm_64, a->vm, 0, src2_mop);
1852 } else {
1853 TCGv_i32 tmp = tcg_temp_new_i32();
1854 read_neon_element32(tmp, a->vm, 0, MO_32);
1855 widenfn(rm_64, tmp);
1856 tcg_temp_free_i32(tmp);
1857 }
1858
1859 opfn(rn0_64, rn0_64, rm_64);
1860
1861 /*
1862 * Load second pass inputs before storing the first pass result, to
1863 * avoid incorrect results if a narrow input overlaps with the result.
1864 */
1865 if (src1_mop >= 0) {
1866 read_neon_element64(rn1_64, a->vn, 1, src1_mop);
1867 } else {
1868 TCGv_i32 tmp = tcg_temp_new_i32();
1869 read_neon_element32(tmp, a->vn, 1, MO_32);
1870 widenfn(rn1_64, tmp);
1871 tcg_temp_free_i32(tmp);
1872 }
1873 if (src2_mop >= 0) {
1874 read_neon_element64(rm_64, a->vm, 1, src2_mop);
1875 } else {
1876 TCGv_i32 tmp = tcg_temp_new_i32();
1877 read_neon_element32(tmp, a->vm, 1, MO_32);
1878 widenfn(rm_64, tmp);
1879 tcg_temp_free_i32(tmp);
1880 }
1881
1882 write_neon_element64(rn0_64, a->vd, 0, MO_64);
1883
1884 opfn(rn1_64, rn1_64, rm_64);
1885 write_neon_element64(rn1_64, a->vd, 1, MO_64);
1886
1887 tcg_temp_free_i64(rn0_64);
1888 tcg_temp_free_i64(rn1_64);
1889 tcg_temp_free_i64(rm_64);
1890
1891 return true;
1892 }
1893
1894 #define DO_PREWIDEN(INSN, S, OP, SRC1WIDE, SIGN) \
1895 static bool trans_##INSN##_3d(DisasContext *s, arg_3diff *a) \
1896 { \
1897 static NeonGenWidenFn * const widenfn[] = { \
1898 gen_helper_neon_widen_##S##8, \
1899 gen_helper_neon_widen_##S##16, \
1900 NULL, NULL, \
1901 }; \
1902 static NeonGenTwo64OpFn * const addfn[] = { \
1903 gen_helper_neon_##OP##l_u16, \
1904 gen_helper_neon_##OP##l_u32, \
1905 tcg_gen_##OP##_i64, \
1906 NULL, \
1907 }; \
1908 int narrow_mop = a->size == MO_32 ? MO_32 | SIGN : -1; \
1909 return do_prewiden_3d(s, a, widenfn[a->size], addfn[a->size], \
1910 SRC1WIDE ? MO_UQ : narrow_mop, \
1911 narrow_mop); \
1912 }
1913
1914 DO_PREWIDEN(VADDL_S, s, add, false, MO_SIGN)
1915 DO_PREWIDEN(VADDL_U, u, add, false, 0)
1916 DO_PREWIDEN(VSUBL_S, s, sub, false, MO_SIGN)
1917 DO_PREWIDEN(VSUBL_U, u, sub, false, 0)
1918 DO_PREWIDEN(VADDW_S, s, add, true, MO_SIGN)
1919 DO_PREWIDEN(VADDW_U, u, add, true, 0)
1920 DO_PREWIDEN(VSUBW_S, s, sub, true, MO_SIGN)
1921 DO_PREWIDEN(VSUBW_U, u, sub, true, 0)
1922
1923 static bool do_narrow_3d(DisasContext *s, arg_3diff *a,
1924 NeonGenTwo64OpFn *opfn, NeonGenNarrowFn *narrowfn)
1925 {
1926 /* 3-regs different lengths, narrowing (VADDHN/VSUBHN/VRADDHN/VRSUBHN) */
1927 TCGv_i64 rn_64, rm_64;
1928 TCGv_i32 rd0, rd1;
1929
1930 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1931 return false;
1932 }
1933
1934 /* UNDEF accesses to D16-D31 if they don't exist. */
1935 if (!dc_isar_feature(aa32_simd_r32, s) &&
1936 ((a->vd | a->vn | a->vm) & 0x10)) {
1937 return false;
1938 }
1939
1940 if (!opfn || !narrowfn) {
1941 /* size == 3 case, which is an entirely different insn group */
1942 return false;
1943 }
1944
1945 if ((a->vn | a->vm) & 1) {
1946 return false;
1947 }
1948
1949 if (!vfp_access_check(s)) {
1950 return true;
1951 }
1952
1953 rn_64 = tcg_temp_new_i64();
1954 rm_64 = tcg_temp_new_i64();
1955 rd0 = tcg_temp_new_i32();
1956 rd1 = tcg_temp_new_i32();
1957
1958 read_neon_element64(rn_64, a->vn, 0, MO_64);
1959 read_neon_element64(rm_64, a->vm, 0, MO_64);
1960
1961 opfn(rn_64, rn_64, rm_64);
1962
1963 narrowfn(rd0, rn_64);
1964
1965 read_neon_element64(rn_64, a->vn, 1, MO_64);
1966 read_neon_element64(rm_64, a->vm, 1, MO_64);
1967
1968 opfn(rn_64, rn_64, rm_64);
1969
1970 narrowfn(rd1, rn_64);
1971
1972 write_neon_element32(rd0, a->vd, 0, MO_32);
1973 write_neon_element32(rd1, a->vd, 1, MO_32);
1974
1975 tcg_temp_free_i32(rd0);
1976 tcg_temp_free_i32(rd1);
1977 tcg_temp_free_i64(rn_64);
1978 tcg_temp_free_i64(rm_64);
1979
1980 return true;
1981 }
1982
1983 #define DO_NARROW_3D(INSN, OP, NARROWTYPE, EXTOP) \
1984 static bool trans_##INSN##_3d(DisasContext *s, arg_3diff *a) \
1985 { \
1986 static NeonGenTwo64OpFn * const addfn[] = { \
1987 gen_helper_neon_##OP##l_u16, \
1988 gen_helper_neon_##OP##l_u32, \
1989 tcg_gen_##OP##_i64, \
1990 NULL, \
1991 }; \
1992 static NeonGenNarrowFn * const narrowfn[] = { \
1993 gen_helper_neon_##NARROWTYPE##_high_u8, \
1994 gen_helper_neon_##NARROWTYPE##_high_u16, \
1995 EXTOP, \
1996 NULL, \
1997 }; \
1998 return do_narrow_3d(s, a, addfn[a->size], narrowfn[a->size]); \
1999 }
2000
2001 static void gen_narrow_round_high_u32(TCGv_i32 rd, TCGv_i64 rn)
2002 {
2003 tcg_gen_addi_i64(rn, rn, 1u << 31);
2004 tcg_gen_extrh_i64_i32(rd, rn);
2005 }
2006
2007 DO_NARROW_3D(VADDHN, add, narrow, tcg_gen_extrh_i64_i32)
2008 DO_NARROW_3D(VSUBHN, sub, narrow, tcg_gen_extrh_i64_i32)
2009 DO_NARROW_3D(VRADDHN, add, narrow_round, gen_narrow_round_high_u32)
2010 DO_NARROW_3D(VRSUBHN, sub, narrow_round, gen_narrow_round_high_u32)
2011
2012 static bool do_long_3d(DisasContext *s, arg_3diff *a,
2013 NeonGenTwoOpWidenFn *opfn,
2014 NeonGenTwo64OpFn *accfn)
2015 {
2016 /*
2017 * 3-regs different lengths, long operations.
2018 * These perform an operation on two inputs that returns a double-width
2019 * result, and then possibly perform an accumulation operation of
2020 * that result into the double-width destination.
2021 */
2022 TCGv_i64 rd0, rd1, tmp;
2023 TCGv_i32 rn, rm;
2024
2025 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2026 return false;
2027 }
2028
2029 /* UNDEF accesses to D16-D31 if they don't exist. */
2030 if (!dc_isar_feature(aa32_simd_r32, s) &&
2031 ((a->vd | a->vn | a->vm) & 0x10)) {
2032 return false;
2033 }
2034
2035 if (!opfn) {
2036 /* size == 3 case, which is an entirely different insn group */
2037 return false;
2038 }
2039
2040 if (a->vd & 1) {
2041 return false;
2042 }
2043
2044 if (!vfp_access_check(s)) {
2045 return true;
2046 }
2047
2048 rd0 = tcg_temp_new_i64();
2049 rd1 = tcg_temp_new_i64();
2050
2051 rn = tcg_temp_new_i32();
2052 rm = tcg_temp_new_i32();
2053 read_neon_element32(rn, a->vn, 0, MO_32);
2054 read_neon_element32(rm, a->vm, 0, MO_32);
2055 opfn(rd0, rn, rm);
2056
2057 read_neon_element32(rn, a->vn, 1, MO_32);
2058 read_neon_element32(rm, a->vm, 1, MO_32);
2059 opfn(rd1, rn, rm);
2060 tcg_temp_free_i32(rn);
2061 tcg_temp_free_i32(rm);
2062
2063 /* Don't store results until after all loads: they might overlap */
2064 if (accfn) {
2065 tmp = tcg_temp_new_i64();
2066 read_neon_element64(tmp, a->vd, 0, MO_64);
2067 accfn(rd0, tmp, rd0);
2068 read_neon_element64(tmp, a->vd, 1, MO_64);
2069 accfn(rd1, tmp, rd1);
2070 tcg_temp_free_i64(tmp);
2071 }
2072
2073 write_neon_element64(rd0, a->vd, 0, MO_64);
2074 write_neon_element64(rd1, a->vd, 1, MO_64);
2075 tcg_temp_free_i64(rd0);
2076 tcg_temp_free_i64(rd1);
2077
2078 return true;
2079 }
2080
2081 static bool trans_VABDL_S_3d(DisasContext *s, arg_3diff *a)
2082 {
2083 static NeonGenTwoOpWidenFn * const opfn[] = {
2084 gen_helper_neon_abdl_s16,
2085 gen_helper_neon_abdl_s32,
2086 gen_helper_neon_abdl_s64,
2087 NULL,
2088 };
2089
2090 return do_long_3d(s, a, opfn[a->size], NULL);
2091 }
2092
2093 static bool trans_VABDL_U_3d(DisasContext *s, arg_3diff *a)
2094 {
2095 static NeonGenTwoOpWidenFn * const opfn[] = {
2096 gen_helper_neon_abdl_u16,
2097 gen_helper_neon_abdl_u32,
2098 gen_helper_neon_abdl_u64,
2099 NULL,
2100 };
2101
2102 return do_long_3d(s, a, opfn[a->size], NULL);
2103 }
2104
2105 static bool trans_VABAL_S_3d(DisasContext *s, arg_3diff *a)
2106 {
2107 static NeonGenTwoOpWidenFn * const opfn[] = {
2108 gen_helper_neon_abdl_s16,
2109 gen_helper_neon_abdl_s32,
2110 gen_helper_neon_abdl_s64,
2111 NULL,
2112 };
2113 static NeonGenTwo64OpFn * const addfn[] = {
2114 gen_helper_neon_addl_u16,
2115 gen_helper_neon_addl_u32,
2116 tcg_gen_add_i64,
2117 NULL,
2118 };
2119
2120 return do_long_3d(s, a, opfn[a->size], addfn[a->size]);
2121 }
2122
2123 static bool trans_VABAL_U_3d(DisasContext *s, arg_3diff *a)
2124 {
2125 static NeonGenTwoOpWidenFn * const opfn[] = {
2126 gen_helper_neon_abdl_u16,
2127 gen_helper_neon_abdl_u32,
2128 gen_helper_neon_abdl_u64,
2129 NULL,
2130 };
2131 static NeonGenTwo64OpFn * const addfn[] = {
2132 gen_helper_neon_addl_u16,
2133 gen_helper_neon_addl_u32,
2134 tcg_gen_add_i64,
2135 NULL,
2136 };
2137
2138 return do_long_3d(s, a, opfn[a->size], addfn[a->size]);
2139 }
2140
2141 static void gen_mull_s32(TCGv_i64 rd, TCGv_i32 rn, TCGv_i32 rm)
2142 {
2143 TCGv_i32 lo = tcg_temp_new_i32();
2144 TCGv_i32 hi = tcg_temp_new_i32();
2145
2146 tcg_gen_muls2_i32(lo, hi, rn, rm);
2147 tcg_gen_concat_i32_i64(rd, lo, hi);
2148
2149 tcg_temp_free_i32(lo);
2150 tcg_temp_free_i32(hi);
2151 }
2152
2153 static void gen_mull_u32(TCGv_i64 rd, TCGv_i32 rn, TCGv_i32 rm)
2154 {
2155 TCGv_i32 lo = tcg_temp_new_i32();
2156 TCGv_i32 hi = tcg_temp_new_i32();
2157
2158 tcg_gen_mulu2_i32(lo, hi, rn, rm);
2159 tcg_gen_concat_i32_i64(rd, lo, hi);
2160
2161 tcg_temp_free_i32(lo);
2162 tcg_temp_free_i32(hi);
2163 }
2164
2165 static bool trans_VMULL_S_3d(DisasContext *s, arg_3diff *a)
2166 {
2167 static NeonGenTwoOpWidenFn * const opfn[] = {
2168 gen_helper_neon_mull_s8,
2169 gen_helper_neon_mull_s16,
2170 gen_mull_s32,
2171 NULL,
2172 };
2173
2174 return do_long_3d(s, a, opfn[a->size], NULL);
2175 }
2176
2177 static bool trans_VMULL_U_3d(DisasContext *s, arg_3diff *a)
2178 {
2179 static NeonGenTwoOpWidenFn * const opfn[] = {
2180 gen_helper_neon_mull_u8,
2181 gen_helper_neon_mull_u16,
2182 gen_mull_u32,
2183 NULL,
2184 };
2185
2186 return do_long_3d(s, a, opfn[a->size], NULL);
2187 }
2188
2189 #define DO_VMLAL(INSN,MULL,ACC) \
2190 static bool trans_##INSN##_3d(DisasContext *s, arg_3diff *a) \
2191 { \
2192 static NeonGenTwoOpWidenFn * const opfn[] = { \
2193 gen_helper_neon_##MULL##8, \
2194 gen_helper_neon_##MULL##16, \
2195 gen_##MULL##32, \
2196 NULL, \
2197 }; \
2198 static NeonGenTwo64OpFn * const accfn[] = { \
2199 gen_helper_neon_##ACC##l_u16, \
2200 gen_helper_neon_##ACC##l_u32, \
2201 tcg_gen_##ACC##_i64, \
2202 NULL, \
2203 }; \
2204 return do_long_3d(s, a, opfn[a->size], accfn[a->size]); \
2205 }
2206
2207 DO_VMLAL(VMLAL_S,mull_s,add)
2208 DO_VMLAL(VMLAL_U,mull_u,add)
2209 DO_VMLAL(VMLSL_S,mull_s,sub)
2210 DO_VMLAL(VMLSL_U,mull_u,sub)
2211
2212 static void gen_VQDMULL_16(TCGv_i64 rd, TCGv_i32 rn, TCGv_i32 rm)
2213 {
2214 gen_helper_neon_mull_s16(rd, rn, rm);
2215 gen_helper_neon_addl_saturate_s32(rd, cpu_env, rd, rd);
2216 }
2217
2218 static void gen_VQDMULL_32(TCGv_i64 rd, TCGv_i32 rn, TCGv_i32 rm)
2219 {
2220 gen_mull_s32(rd, rn, rm);
2221 gen_helper_neon_addl_saturate_s64(rd, cpu_env, rd, rd);
2222 }
2223
2224 static bool trans_VQDMULL_3d(DisasContext *s, arg_3diff *a)
2225 {
2226 static NeonGenTwoOpWidenFn * const opfn[] = {
2227 NULL,
2228 gen_VQDMULL_16,
2229 gen_VQDMULL_32,
2230 NULL,
2231 };
2232
2233 return do_long_3d(s, a, opfn[a->size], NULL);
2234 }
2235
2236 static void gen_VQDMLAL_acc_16(TCGv_i64 rd, TCGv_i64 rn, TCGv_i64 rm)
2237 {
2238 gen_helper_neon_addl_saturate_s32(rd, cpu_env, rn, rm);
2239 }
2240
2241 static void gen_VQDMLAL_acc_32(TCGv_i64 rd, TCGv_i64 rn, TCGv_i64 rm)
2242 {
2243 gen_helper_neon_addl_saturate_s64(rd, cpu_env, rn, rm);
2244 }
2245
2246 static bool trans_VQDMLAL_3d(DisasContext *s, arg_3diff *a)
2247 {
2248 static NeonGenTwoOpWidenFn * const opfn[] = {
2249 NULL,
2250 gen_VQDMULL_16,
2251 gen_VQDMULL_32,
2252 NULL,
2253 };
2254 static NeonGenTwo64OpFn * const accfn[] = {
2255 NULL,
2256 gen_VQDMLAL_acc_16,
2257 gen_VQDMLAL_acc_32,
2258 NULL,
2259 };
2260
2261 return do_long_3d(s, a, opfn[a->size], accfn[a->size]);
2262 }
2263
2264 static void gen_VQDMLSL_acc_16(TCGv_i64 rd, TCGv_i64 rn, TCGv_i64 rm)
2265 {
2266 gen_helper_neon_negl_u32(rm, rm);
2267 gen_helper_neon_addl_saturate_s32(rd, cpu_env, rn, rm);
2268 }
2269
2270 static void gen_VQDMLSL_acc_32(TCGv_i64 rd, TCGv_i64 rn, TCGv_i64 rm)
2271 {
2272 tcg_gen_neg_i64(rm, rm);
2273 gen_helper_neon_addl_saturate_s64(rd, cpu_env, rn, rm);
2274 }
2275
2276 static bool trans_VQDMLSL_3d(DisasContext *s, arg_3diff *a)
2277 {
2278 static NeonGenTwoOpWidenFn * const opfn[] = {
2279 NULL,
2280 gen_VQDMULL_16,
2281 gen_VQDMULL_32,
2282 NULL,
2283 };
2284 static NeonGenTwo64OpFn * const accfn[] = {
2285 NULL,
2286 gen_VQDMLSL_acc_16,
2287 gen_VQDMLSL_acc_32,
2288 NULL,
2289 };
2290
2291 return do_long_3d(s, a, opfn[a->size], accfn[a->size]);
2292 }
2293
2294 static bool trans_VMULL_P_3d(DisasContext *s, arg_3diff *a)
2295 {
2296 gen_helper_gvec_3 *fn_gvec;
2297
2298 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2299 return false;
2300 }
2301
2302 /* UNDEF accesses to D16-D31 if they don't exist. */
2303 if (!dc_isar_feature(aa32_simd_r32, s) &&
2304 ((a->vd | a->vn | a->vm) & 0x10)) {
2305 return false;
2306 }
2307
2308 if (a->vd & 1) {
2309 return false;
2310 }
2311
2312 switch (a->size) {
2313 case 0:
2314 fn_gvec = gen_helper_neon_pmull_h;
2315 break;
2316 case 2:
2317 if (!dc_isar_feature(aa32_pmull, s)) {
2318 return false;
2319 }
2320 fn_gvec = gen_helper_gvec_pmull_q;
2321 break;
2322 default:
2323 return false;
2324 }
2325
2326 if (!vfp_access_check(s)) {
2327 return true;
2328 }
2329
2330 tcg_gen_gvec_3_ool(neon_full_reg_offset(a->vd),
2331 neon_full_reg_offset(a->vn),
2332 neon_full_reg_offset(a->vm),
2333 16, 16, 0, fn_gvec);
2334 return true;
2335 }
2336
2337 static void gen_neon_dup_low16(TCGv_i32 var)
2338 {
2339 TCGv_i32 tmp = tcg_temp_new_i32();
2340 tcg_gen_ext16u_i32(var, var);
2341 tcg_gen_shli_i32(tmp, var, 16);
2342 tcg_gen_or_i32(var, var, tmp);
2343 tcg_temp_free_i32(tmp);
2344 }
2345
2346 static void gen_neon_dup_high16(TCGv_i32 var)
2347 {
2348 TCGv_i32 tmp = tcg_temp_new_i32();
2349 tcg_gen_andi_i32(var, var, 0xffff0000);
2350 tcg_gen_shri_i32(tmp, var, 16);
2351 tcg_gen_or_i32(var, var, tmp);
2352 tcg_temp_free_i32(tmp);
2353 }
2354
2355 static inline TCGv_i32 neon_get_scalar(int size, int reg)
2356 {
2357 TCGv_i32 tmp = tcg_temp_new_i32();
2358 if (size == MO_16) {
2359 read_neon_element32(tmp, reg & 7, reg >> 4, MO_32);
2360 if (reg & 8) {
2361 gen_neon_dup_high16(tmp);
2362 } else {
2363 gen_neon_dup_low16(tmp);
2364 }
2365 } else {
2366 read_neon_element32(tmp, reg & 15, reg >> 4, MO_32);
2367 }
2368 return tmp;
2369 }
2370
2371 static bool do_2scalar(DisasContext *s, arg_2scalar *a,
2372 NeonGenTwoOpFn *opfn, NeonGenTwoOpFn *accfn)
2373 {
2374 /*
2375 * Two registers and a scalar: perform an operation between
2376 * the input elements and the scalar, and then possibly
2377 * perform an accumulation operation of that result into the
2378 * destination.
2379 */
2380 TCGv_i32 scalar, tmp;
2381 int pass;
2382
2383 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2384 return false;
2385 }
2386
2387 /* UNDEF accesses to D16-D31 if they don't exist. */
2388 if (!dc_isar_feature(aa32_simd_r32, s) &&
2389 ((a->vd | a->vn | a->vm) & 0x10)) {
2390 return false;
2391 }
2392
2393 if (!opfn) {
2394 /* Bad size (including size == 3, which is a different insn group) */
2395 return false;
2396 }
2397
2398 if (a->q && ((a->vd | a->vn) & 1)) {
2399 return false;
2400 }
2401
2402 if (!vfp_access_check(s)) {
2403 return true;
2404 }
2405
2406 scalar = neon_get_scalar(a->size, a->vm);
2407 tmp = tcg_temp_new_i32();
2408
2409 for (pass = 0; pass < (a->q ? 4 : 2); pass++) {
2410 read_neon_element32(tmp, a->vn, pass, MO_32);
2411 opfn(tmp, tmp, scalar);
2412 if (accfn) {
2413 TCGv_i32 rd = tcg_temp_new_i32();
2414 read_neon_element32(rd, a->vd, pass, MO_32);
2415 accfn(tmp, rd, tmp);
2416 tcg_temp_free_i32(rd);
2417 }
2418 write_neon_element32(tmp, a->vd, pass, MO_32);
2419 }
2420 tcg_temp_free_i32(tmp);
2421 tcg_temp_free_i32(scalar);
2422 return true;
2423 }
2424
2425 static bool trans_VMUL_2sc(DisasContext *s, arg_2scalar *a)
2426 {
2427 static NeonGenTwoOpFn * const opfn[] = {
2428 NULL,
2429 gen_helper_neon_mul_u16,
2430 tcg_gen_mul_i32,
2431 NULL,
2432 };
2433
2434 return do_2scalar(s, a, opfn[a->size], NULL);
2435 }
2436
2437 static bool trans_VMLA_2sc(DisasContext *s, arg_2scalar *a)
2438 {
2439 static NeonGenTwoOpFn * const opfn[] = {
2440 NULL,
2441 gen_helper_neon_mul_u16,
2442 tcg_gen_mul_i32,
2443 NULL,
2444 };
2445 static NeonGenTwoOpFn * const accfn[] = {
2446 NULL,
2447 gen_helper_neon_add_u16,
2448 tcg_gen_add_i32,
2449 NULL,
2450 };
2451
2452 return do_2scalar(s, a, opfn[a->size], accfn[a->size]);
2453 }
2454
2455 static bool trans_VMLS_2sc(DisasContext *s, arg_2scalar *a)
2456 {
2457 static NeonGenTwoOpFn * const opfn[] = {
2458 NULL,
2459 gen_helper_neon_mul_u16,
2460 tcg_gen_mul_i32,
2461 NULL,
2462 };
2463 static NeonGenTwoOpFn * const accfn[] = {
2464 NULL,
2465 gen_helper_neon_sub_u16,
2466 tcg_gen_sub_i32,
2467 NULL,
2468 };
2469
2470 return do_2scalar(s, a, opfn[a->size], accfn[a->size]);
2471 }
2472
2473 static bool do_2scalar_fp_vec(DisasContext *s, arg_2scalar *a,
2474 gen_helper_gvec_3_ptr *fn)
2475 {
2476 /* Two registers and a scalar, using gvec */
2477 int vec_size = a->q ? 16 : 8;
2478 int rd_ofs = neon_full_reg_offset(a->vd);
2479 int rn_ofs = neon_full_reg_offset(a->vn);
2480 int rm_ofs;
2481 int idx;
2482 TCGv_ptr fpstatus;
2483
2484 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2485 return false;
2486 }
2487
2488 /* UNDEF accesses to D16-D31 if they don't exist. */
2489 if (!dc_isar_feature(aa32_simd_r32, s) &&
2490 ((a->vd | a->vn | a->vm) & 0x10)) {
2491 return false;
2492 }
2493
2494 if (!fn) {
2495 /* Bad size (including size == 3, which is a different insn group) */
2496 return false;
2497 }
2498
2499 if (a->q && ((a->vd | a->vn) & 1)) {
2500 return false;
2501 }
2502
2503 if (!vfp_access_check(s)) {
2504 return true;
2505 }
2506
2507 /* a->vm is M:Vm, which encodes both register and index */
2508 idx = extract32(a->vm, a->size + 2, 2);
2509 a->vm = extract32(a->vm, 0, a->size + 2);
2510 rm_ofs = neon_full_reg_offset(a->vm);
2511
2512 fpstatus = fpstatus_ptr(a->size == 1 ? FPST_STD_F16 : FPST_STD);
2513 tcg_gen_gvec_3_ptr(rd_ofs, rn_ofs, rm_ofs, fpstatus,
2514 vec_size, vec_size, idx, fn);
2515 tcg_temp_free_ptr(fpstatus);
2516 return true;
2517 }
2518
2519 #define DO_VMUL_F_2sc(NAME, FUNC) \
2520 static bool trans_##NAME##_F_2sc(DisasContext *s, arg_2scalar *a) \
2521 { \
2522 static gen_helper_gvec_3_ptr * const opfn[] = { \
2523 NULL, \
2524 gen_helper_##FUNC##_h, \
2525 gen_helper_##FUNC##_s, \
2526 NULL, \
2527 }; \
2528 if (a->size == MO_16 && !dc_isar_feature(aa32_fp16_arith, s)) { \
2529 return false; \
2530 } \
2531 return do_2scalar_fp_vec(s, a, opfn[a->size]); \
2532 }
2533
2534 DO_VMUL_F_2sc(VMUL, gvec_fmul_idx)
2535 DO_VMUL_F_2sc(VMLA, gvec_fmla_nf_idx)
2536 DO_VMUL_F_2sc(VMLS, gvec_fmls_nf_idx)
2537
2538 WRAP_ENV_FN(gen_VQDMULH_16, gen_helper_neon_qdmulh_s16)
2539 WRAP_ENV_FN(gen_VQDMULH_32, gen_helper_neon_qdmulh_s32)
2540 WRAP_ENV_FN(gen_VQRDMULH_16, gen_helper_neon_qrdmulh_s16)
2541 WRAP_ENV_FN(gen_VQRDMULH_32, gen_helper_neon_qrdmulh_s32)
2542
2543 static bool trans_VQDMULH_2sc(DisasContext *s, arg_2scalar *a)
2544 {
2545 static NeonGenTwoOpFn * const opfn[] = {
2546 NULL,
2547 gen_VQDMULH_16,
2548 gen_VQDMULH_32,
2549 NULL,
2550 };
2551
2552 return do_2scalar(s, a, opfn[a->size], NULL);
2553 }
2554
2555 static bool trans_VQRDMULH_2sc(DisasContext *s, arg_2scalar *a)
2556 {
2557 static NeonGenTwoOpFn * const opfn[] = {
2558 NULL,
2559 gen_VQRDMULH_16,
2560 gen_VQRDMULH_32,
2561 NULL,
2562 };
2563
2564 return do_2scalar(s, a, opfn[a->size], NULL);
2565 }
2566
2567 static bool do_vqrdmlah_2sc(DisasContext *s, arg_2scalar *a,
2568 NeonGenThreeOpEnvFn *opfn)
2569 {
2570 /*
2571 * VQRDMLAH/VQRDMLSH: this is like do_2scalar, but the opfn
2572 * performs a kind of fused op-then-accumulate using a helper
2573 * function that takes all of rd, rn and the scalar at once.
2574 */
2575 TCGv_i32 scalar, rn, rd;
2576 int pass;
2577
2578 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2579 return false;
2580 }
2581
2582 if (!dc_isar_feature(aa32_rdm, s)) {
2583 return false;
2584 }
2585
2586 /* UNDEF accesses to D16-D31 if they don't exist. */
2587 if (!dc_isar_feature(aa32_simd_r32, s) &&
2588 ((a->vd | a->vn | a->vm) & 0x10)) {
2589 return false;
2590 }
2591
2592 if (!opfn) {
2593 /* Bad size (including size == 3, which is a different insn group) */
2594 return false;
2595 }
2596
2597 if (a->q && ((a->vd | a->vn) & 1)) {
2598 return false;
2599 }
2600
2601 if (!vfp_access_check(s)) {
2602 return true;
2603 }
2604
2605 scalar = neon_get_scalar(a->size, a->vm);
2606 rn = tcg_temp_new_i32();
2607 rd = tcg_temp_new_i32();
2608
2609 for (pass = 0; pass < (a->q ? 4 : 2); pass++) {
2610 read_neon_element32(rn, a->vn, pass, MO_32);
2611 read_neon_element32(rd, a->vd, pass, MO_32);
2612 opfn(rd, cpu_env, rn, scalar, rd);
2613 write_neon_element32(rd, a->vd, pass, MO_32);
2614 }
2615 tcg_temp_free_i32(rn);
2616 tcg_temp_free_i32(rd);
2617 tcg_temp_free_i32(scalar);
2618
2619 return true;
2620 }
2621
2622 static bool trans_VQRDMLAH_2sc(DisasContext *s, arg_2scalar *a)
2623 {
2624 static NeonGenThreeOpEnvFn *opfn[] = {
2625 NULL,
2626 gen_helper_neon_qrdmlah_s16,
2627 gen_helper_neon_qrdmlah_s32,
2628 NULL,
2629 };
2630 return do_vqrdmlah_2sc(s, a, opfn[a->size]);
2631 }
2632
2633 static bool trans_VQRDMLSH_2sc(DisasContext *s, arg_2scalar *a)
2634 {
2635 static NeonGenThreeOpEnvFn *opfn[] = {
2636 NULL,
2637 gen_helper_neon_qrdmlsh_s16,
2638 gen_helper_neon_qrdmlsh_s32,
2639 NULL,
2640 };
2641 return do_vqrdmlah_2sc(s, a, opfn[a->size]);
2642 }
2643
2644 static bool do_2scalar_long(DisasContext *s, arg_2scalar *a,
2645 NeonGenTwoOpWidenFn *opfn,
2646 NeonGenTwo64OpFn *accfn)
2647 {
2648 /*
2649 * Two registers and a scalar, long operations: perform an
2650 * operation on the input elements and the scalar which produces
2651 * a double-width result, and then possibly perform an accumulation
2652 * operation of that result into the destination.
2653 */
2654 TCGv_i32 scalar, rn;
2655 TCGv_i64 rn0_64, rn1_64;
2656
2657 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2658 return false;
2659 }
2660
2661 /* UNDEF accesses to D16-D31 if they don't exist. */
2662 if (!dc_isar_feature(aa32_simd_r32, s) &&
2663 ((a->vd | a->vn | a->vm) & 0x10)) {
2664 return false;
2665 }
2666
2667 if (!opfn) {
2668 /* Bad size (including size == 3, which is a different insn group) */
2669 return false;
2670 }
2671
2672 if (a->vd & 1) {
2673 return false;
2674 }
2675
2676 if (!vfp_access_check(s)) {
2677 return true;
2678 }
2679
2680 scalar = neon_get_scalar(a->size, a->vm);
2681
2682 /* Load all inputs before writing any outputs, in case of overlap */
2683 rn = tcg_temp_new_i32();
2684 read_neon_element32(rn, a->vn, 0, MO_32);
2685 rn0_64 = tcg_temp_new_i64();
2686 opfn(rn0_64, rn, scalar);
2687
2688 read_neon_element32(rn, a->vn, 1, MO_32);
2689 rn1_64 = tcg_temp_new_i64();
2690 opfn(rn1_64, rn, scalar);
2691 tcg_temp_free_i32(rn);
2692 tcg_temp_free_i32(scalar);
2693
2694 if (accfn) {
2695 TCGv_i64 t64 = tcg_temp_new_i64();
2696 read_neon_element64(t64, a->vd, 0, MO_64);
2697 accfn(rn0_64, t64, rn0_64);
2698 read_neon_element64(t64, a->vd, 1, MO_64);
2699 accfn(rn1_64, t64, rn1_64);
2700 tcg_temp_free_i64(t64);
2701 }
2702
2703 write_neon_element64(rn0_64, a->vd, 0, MO_64);
2704 write_neon_element64(rn1_64, a->vd, 1, MO_64);
2705 tcg_temp_free_i64(rn0_64);
2706 tcg_temp_free_i64(rn1_64);
2707 return true;
2708 }
2709
2710 static bool trans_VMULL_S_2sc(DisasContext *s, arg_2scalar *a)
2711 {
2712 static NeonGenTwoOpWidenFn * const opfn[] = {
2713 NULL,
2714 gen_helper_neon_mull_s16,
2715 gen_mull_s32,
2716 NULL,
2717 };
2718
2719 return do_2scalar_long(s, a, opfn[a->size], NULL);
2720 }
2721
2722 static bool trans_VMULL_U_2sc(DisasContext *s, arg_2scalar *a)
2723 {
2724 static NeonGenTwoOpWidenFn * const opfn[] = {
2725 NULL,
2726 gen_helper_neon_mull_u16,
2727 gen_mull_u32,
2728 NULL,
2729 };
2730
2731 return do_2scalar_long(s, a, opfn[a->size], NULL);
2732 }
2733
2734 #define DO_VMLAL_2SC(INSN, MULL, ACC) \
2735 static bool trans_##INSN##_2sc(DisasContext *s, arg_2scalar *a) \
2736 { \
2737 static NeonGenTwoOpWidenFn * const opfn[] = { \
2738 NULL, \
2739 gen_helper_neon_##MULL##16, \
2740 gen_##MULL##32, \
2741 NULL, \
2742 }; \
2743 static NeonGenTwo64OpFn * const accfn[] = { \
2744 NULL, \
2745 gen_helper_neon_##ACC##l_u32, \
2746 tcg_gen_##ACC##_i64, \
2747 NULL, \
2748 }; \
2749 return do_2scalar_long(s, a, opfn[a->size], accfn[a->size]); \
2750 }
2751
2752 DO_VMLAL_2SC(VMLAL_S, mull_s, add)
2753 DO_VMLAL_2SC(VMLAL_U, mull_u, add)
2754 DO_VMLAL_2SC(VMLSL_S, mull_s, sub)
2755 DO_VMLAL_2SC(VMLSL_U, mull_u, sub)
2756
2757 static bool trans_VQDMULL_2sc(DisasContext *s, arg_2scalar *a)
2758 {
2759 static NeonGenTwoOpWidenFn * const opfn[] = {
2760 NULL,
2761 gen_VQDMULL_16,
2762 gen_VQDMULL_32,
2763 NULL,
2764 };
2765
2766 return do_2scalar_long(s, a, opfn[a->size], NULL);
2767 }
2768
2769 static bool trans_VQDMLAL_2sc(DisasContext *s, arg_2scalar *a)
2770 {
2771 static NeonGenTwoOpWidenFn * const opfn[] = {
2772 NULL,
2773 gen_VQDMULL_16,
2774 gen_VQDMULL_32,
2775 NULL,
2776 };
2777 static NeonGenTwo64OpFn * const accfn[] = {
2778 NULL,
2779 gen_VQDMLAL_acc_16,
2780 gen_VQDMLAL_acc_32,
2781 NULL,
2782 };
2783
2784 return do_2scalar_long(s, a, opfn[a->size], accfn[a->size]);
2785 }
2786
2787 static bool trans_VQDMLSL_2sc(DisasContext *s, arg_2scalar *a)
2788 {
2789 static NeonGenTwoOpWidenFn * const opfn[] = {
2790 NULL,
2791 gen_VQDMULL_16,
2792 gen_VQDMULL_32,
2793 NULL,
2794 };
2795 static NeonGenTwo64OpFn * const accfn[] = {
2796 NULL,
2797 gen_VQDMLSL_acc_16,
2798 gen_VQDMLSL_acc_32,
2799 NULL,
2800 };
2801
2802 return do_2scalar_long(s, a, opfn[a->size], accfn[a->size]);
2803 }
2804
2805 static bool trans_VEXT(DisasContext *s, arg_VEXT *a)
2806 {
2807 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2808 return false;
2809 }
2810
2811 /* UNDEF accesses to D16-D31 if they don't exist. */
2812 if (!dc_isar_feature(aa32_simd_r32, s) &&
2813 ((a->vd | a->vn | a->vm) & 0x10)) {
2814 return false;
2815 }
2816
2817 if ((a->vn | a->vm | a->vd) & a->q) {
2818 return false;
2819 }
2820
2821 if (a->imm > 7 && !a->q) {
2822 return false;
2823 }
2824
2825 if (!vfp_access_check(s)) {
2826 return true;
2827 }
2828
2829 if (!a->q) {
2830 /* Extract 64 bits from <Vm:Vn> */
2831 TCGv_i64 left, right, dest;
2832
2833 left = tcg_temp_new_i64();
2834 right = tcg_temp_new_i64();
2835 dest = tcg_temp_new_i64();
2836
2837 read_neon_element64(right, a->vn, 0, MO_64);
2838 read_neon_element64(left, a->vm, 0, MO_64);
2839 tcg_gen_extract2_i64(dest, right, left, a->imm * 8);
2840 write_neon_element64(dest, a->vd, 0, MO_64);
2841
2842 tcg_temp_free_i64(left);
2843 tcg_temp_free_i64(right);
2844 tcg_temp_free_i64(dest);
2845 } else {
2846 /* Extract 128 bits from <Vm+1:Vm:Vn+1:Vn> */
2847 TCGv_i64 left, middle, right, destleft, destright;
2848
2849 left = tcg_temp_new_i64();
2850 middle = tcg_temp_new_i64();
2851 right = tcg_temp_new_i64();
2852 destleft = tcg_temp_new_i64();
2853 destright = tcg_temp_new_i64();
2854
2855 if (a->imm < 8) {
2856 read_neon_element64(right, a->vn, 0, MO_64);
2857 read_neon_element64(middle, a->vn, 1, MO_64);
2858 tcg_gen_extract2_i64(destright, right, middle, a->imm * 8);
2859 read_neon_element64(left, a->vm, 0, MO_64);
2860 tcg_gen_extract2_i64(destleft, middle, left, a->imm * 8);
2861 } else {
2862 read_neon_element64(right, a->vn, 1, MO_64);
2863 read_neon_element64(middle, a->vm, 0, MO_64);
2864 tcg_gen_extract2_i64(destright, right, middle, (a->imm - 8) * 8);
2865 read_neon_element64(left, a->vm, 1, MO_64);
2866 tcg_gen_extract2_i64(destleft, middle, left, (a->imm - 8) * 8);
2867 }
2868
2869 write_neon_element64(destright, a->vd, 0, MO_64);
2870 write_neon_element64(destleft, a->vd, 1, MO_64);
2871
2872 tcg_temp_free_i64(destright);
2873 tcg_temp_free_i64(destleft);
2874 tcg_temp_free_i64(right);
2875 tcg_temp_free_i64(middle);
2876 tcg_temp_free_i64(left);
2877 }
2878 return true;
2879 }
2880
2881 static bool trans_VTBL(DisasContext *s, arg_VTBL *a)
2882 {
2883 TCGv_i64 val, def;
2884 TCGv_i32 desc;
2885
2886 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2887 return false;
2888 }
2889
2890 /* UNDEF accesses to D16-D31 if they don't exist. */
2891 if (!dc_isar_feature(aa32_simd_r32, s) &&
2892 ((a->vd | a->vn | a->vm) & 0x10)) {
2893 return false;
2894 }
2895
2896 if ((a->vn + a->len + 1) > 32) {
2897 /*
2898 * This is UNPREDICTABLE; we choose to UNDEF to avoid the
2899 * helper function running off the end of the register file.
2900 */
2901 return false;
2902 }
2903
2904 if (!vfp_access_check(s)) {
2905 return true;
2906 }
2907
2908 desc = tcg_constant_i32((a->vn << 2) | a->len);
2909 def = tcg_temp_new_i64();
2910 if (a->op) {
2911 read_neon_element64(def, a->vd, 0, MO_64);
2912 } else {
2913 tcg_gen_movi_i64(def, 0);
2914 }
2915 val = tcg_temp_new_i64();
2916 read_neon_element64(val, a->vm, 0, MO_64);
2917
2918 gen_helper_neon_tbl(val, cpu_env, desc, val, def);
2919 write_neon_element64(val, a->vd, 0, MO_64);
2920
2921 tcg_temp_free_i64(def);
2922 tcg_temp_free_i64(val);
2923 return true;
2924 }
2925
2926 static bool trans_VDUP_scalar(DisasContext *s, arg_VDUP_scalar *a)
2927 {
2928 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2929 return false;
2930 }
2931
2932 /* UNDEF accesses to D16-D31 if they don't exist. */
2933 if (!dc_isar_feature(aa32_simd_r32, s) &&
2934 ((a->vd | a->vm) & 0x10)) {
2935 return false;
2936 }
2937
2938 if (a->vd & a->q) {
2939 return false;
2940 }
2941
2942 if (!vfp_access_check(s)) {
2943 return true;
2944 }
2945
2946 tcg_gen_gvec_dup_mem(a->size, neon_full_reg_offset(a->vd),
2947 neon_element_offset(a->vm, a->index, a->size),
2948 a->q ? 16 : 8, a->q ? 16 : 8);
2949 return true;
2950 }
2951
2952 static bool trans_VREV64(DisasContext *s, arg_VREV64 *a)
2953 {
2954 int pass, half;
2955 TCGv_i32 tmp[2];
2956
2957 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2958 return false;
2959 }
2960
2961 /* UNDEF accesses to D16-D31 if they don't exist. */
2962 if (!dc_isar_feature(aa32_simd_r32, s) &&
2963 ((a->vd | a->vm) & 0x10)) {
2964 return false;
2965 }
2966
2967 if ((a->vd | a->vm) & a->q) {
2968 return false;
2969 }
2970
2971 if (a->size == 3) {
2972 return false;
2973 }
2974
2975 if (!vfp_access_check(s)) {
2976 return true;
2977 }
2978
2979 tmp[0] = tcg_temp_new_i32();
2980 tmp[1] = tcg_temp_new_i32();
2981
2982 for (pass = 0; pass < (a->q ? 2 : 1); pass++) {
2983 for (half = 0; half < 2; half++) {
2984 read_neon_element32(tmp[half], a->vm, pass * 2 + half, MO_32);
2985 switch (a->size) {
2986 case 0:
2987 tcg_gen_bswap32_i32(tmp[half], tmp[half]);
2988 break;
2989 case 1:
2990 gen_swap_half(tmp[half], tmp[half]);
2991 break;
2992 case 2:
2993 break;
2994 default:
2995 g_assert_not_reached();
2996 }
2997 }
2998 write_neon_element32(tmp[1], a->vd, pass * 2, MO_32);
2999 write_neon_element32(tmp[0], a->vd, pass * 2 + 1, MO_32);
3000 }
3001
3002 tcg_temp_free_i32(tmp[0]);
3003 tcg_temp_free_i32(tmp[1]);
3004 return true;
3005 }
3006
3007 static bool do_2misc_pairwise(DisasContext *s, arg_2misc *a,
3008 NeonGenWidenFn *widenfn,
3009 NeonGenTwo64OpFn *opfn,
3010 NeonGenTwo64OpFn *accfn)
3011 {
3012 /*
3013 * Pairwise long operations: widen both halves of the pair,
3014 * combine the pairs with the opfn, and then possibly accumulate
3015 * into the destination with the accfn.
3016 */
3017 int pass;
3018
3019 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
3020 return false;
3021 }
3022
3023 /* UNDEF accesses to D16-D31 if they don't exist. */
3024 if (!dc_isar_feature(aa32_simd_r32, s) &&
3025 ((a->vd | a->vm) & 0x10)) {
3026 return false;
3027 }
3028
3029 if ((a->vd | a->vm) & a->q) {
3030 return false;
3031 }
3032
3033 if (!widenfn) {
3034 return false;
3035 }
3036
3037 if (!vfp_access_check(s)) {
3038 return true;
3039 }
3040
3041 for (pass = 0; pass < a->q + 1; pass++) {
3042 TCGv_i32 tmp;
3043 TCGv_i64 rm0_64, rm1_64, rd_64;
3044
3045 rm0_64 = tcg_temp_new_i64();
3046 rm1_64 = tcg_temp_new_i64();
3047 rd_64 = tcg_temp_new_i64();
3048
3049 tmp = tcg_temp_new_i32();
3050 read_neon_element32(tmp, a->vm, pass * 2, MO_32);
3051 widenfn(rm0_64, tmp);
3052 read_neon_element32(tmp, a->vm, pass * 2 + 1, MO_32);
3053 widenfn(rm1_64, tmp);
3054 tcg_temp_free_i32(tmp);
3055
3056 opfn(rd_64, rm0_64, rm1_64);
3057 tcg_temp_free_i64(rm0_64);
3058 tcg_temp_free_i64(rm1_64);
3059
3060 if (accfn) {
3061 TCGv_i64 tmp64 = tcg_temp_new_i64();
3062 read_neon_element64(tmp64, a->vd, pass, MO_64);
3063 accfn(rd_64, tmp64, rd_64);
3064 tcg_temp_free_i64(tmp64);
3065 }
3066 write_neon_element64(rd_64, a->vd, pass, MO_64);
3067 tcg_temp_free_i64(rd_64);
3068 }
3069 return true;
3070 }
3071
3072 static bool trans_VPADDL_S(DisasContext *s, arg_2misc *a)
3073 {
3074 static NeonGenWidenFn * const widenfn[] = {
3075 gen_helper_neon_widen_s8,
3076 gen_helper_neon_widen_s16,
3077 tcg_gen_ext_i32_i64,
3078 NULL,
3079 };
3080 static NeonGenTwo64OpFn * const opfn[] = {
3081 gen_helper_neon_paddl_u16,
3082 gen_helper_neon_paddl_u32,
3083 tcg_gen_add_i64,
3084 NULL,
3085 };
3086
3087 return do_2misc_pairwise(s, a, widenfn[a->size], opfn[a->size], NULL);
3088 }
3089
3090 static bool trans_VPADDL_U(DisasContext *s, arg_2misc *a)
3091 {
3092 static NeonGenWidenFn * const widenfn[] = {
3093 gen_helper_neon_widen_u8,
3094 gen_helper_neon_widen_u16,
3095 tcg_gen_extu_i32_i64,
3096 NULL,
3097 };
3098 static NeonGenTwo64OpFn * const opfn[] = {
3099 gen_helper_neon_paddl_u16,
3100 gen_helper_neon_paddl_u32,
3101 tcg_gen_add_i64,
3102 NULL,
3103 };
3104
3105 return do_2misc_pairwise(s, a, widenfn[a->size], opfn[a->size], NULL);
3106 }
3107
3108 static bool trans_VPADAL_S(DisasContext *s, arg_2misc *a)
3109 {
3110 static NeonGenWidenFn * const widenfn[] = {
3111 gen_helper_neon_widen_s8,
3112 gen_helper_neon_widen_s16,
3113 tcg_gen_ext_i32_i64,
3114 NULL,
3115 };
3116 static NeonGenTwo64OpFn * const opfn[] = {
3117 gen_helper_neon_paddl_u16,
3118 gen_helper_neon_paddl_u32,
3119 tcg_gen_add_i64,
3120 NULL,
3121 };
3122 static NeonGenTwo64OpFn * const accfn[] = {
3123 gen_helper_neon_addl_u16,
3124 gen_helper_neon_addl_u32,
3125 tcg_gen_add_i64,
3126 NULL,
3127 };
3128
3129 return do_2misc_pairwise(s, a, widenfn[a->size], opfn[a->size],
3130 accfn[a->size]);
3131 }
3132
3133 static bool trans_VPADAL_U(DisasContext *s, arg_2misc *a)
3134 {
3135 static NeonGenWidenFn * const widenfn[] = {
3136 gen_helper_neon_widen_u8,
3137 gen_helper_neon_widen_u16,
3138 tcg_gen_extu_i32_i64,
3139 NULL,
3140 };
3141 static NeonGenTwo64OpFn * const opfn[] = {
3142 gen_helper_neon_paddl_u16,
3143 gen_helper_neon_paddl_u32,
3144 tcg_gen_add_i64,
3145 NULL,
3146 };
3147 static NeonGenTwo64OpFn * const accfn[] = {
3148 gen_helper_neon_addl_u16,
3149 gen_helper_neon_addl_u32,
3150 tcg_gen_add_i64,
3151 NULL,
3152 };
3153
3154 return do_2misc_pairwise(s, a, widenfn[a->size], opfn[a->size],
3155 accfn[a->size]);
3156 }
3157
3158 typedef void ZipFn(TCGv_ptr, TCGv_ptr);
3159
3160 static bool do_zip_uzp(DisasContext *s, arg_2misc *a,
3161 ZipFn *fn)
3162 {
3163 TCGv_ptr pd, pm;
3164
3165 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
3166 return false;
3167 }
3168
3169 /* UNDEF accesses to D16-D31 if they don't exist. */
3170 if (!dc_isar_feature(aa32_simd_r32, s) &&
3171 ((a->vd | a->vm) & 0x10)) {
3172 return false;
3173 }
3174
3175 if ((a->vd | a->vm) & a->q) {
3176 return false;
3177 }
3178
3179 if (!fn) {
3180 /* Bad size or size/q combination */
3181 return false;
3182 }
3183
3184 if (!vfp_access_check(s)) {
3185 return true;
3186 }
3187
3188 pd = vfp_reg_ptr(true, a->vd);
3189 pm = vfp_reg_ptr(true, a->vm);
3190 fn(pd, pm);
3191 tcg_temp_free_ptr(pd);
3192 tcg_temp_free_ptr(pm);
3193 return true;
3194 }
3195
3196 static bool trans_VUZP(DisasContext *s, arg_2misc *a)
3197 {
3198 static ZipFn * const fn[2][4] = {
3199 {
3200 gen_helper_neon_unzip8,
3201 gen_helper_neon_unzip16,
3202 NULL,
3203 NULL,
3204 }, {
3205 gen_helper_neon_qunzip8,
3206 gen_helper_neon_qunzip16,
3207 gen_helper_neon_qunzip32,
3208 NULL,
3209 }
3210 };
3211 return do_zip_uzp(s, a, fn[a->q][a->size]);
3212 }
3213
3214 static bool trans_VZIP(DisasContext *s, arg_2misc *a)
3215 {
3216 static ZipFn * const fn[2][4] = {
3217 {
3218 gen_helper_neon_zip8,
3219 gen_helper_neon_zip16,
3220 NULL,
3221 NULL,
3222 }, {
3223 gen_helper_neon_qzip8,
3224 gen_helper_neon_qzip16,
3225 gen_helper_neon_qzip32,
3226 NULL,
3227 }
3228 };
3229 return do_zip_uzp(s, a, fn[a->q][a->size]);
3230 }
3231
3232 static bool do_vmovn(DisasContext *s, arg_2misc *a,
3233 NeonGenNarrowEnvFn *narrowfn)
3234 {
3235 TCGv_i64 rm;
3236 TCGv_i32 rd0, rd1;
3237
3238 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
3239 return false;
3240 }
3241
3242 /* UNDEF accesses to D16-D31 if they don't exist. */
3243 if (!dc_isar_feature(aa32_simd_r32, s) &&
3244 ((a->vd | a->vm) & 0x10)) {
3245 return false;
3246 }
3247
3248 if (a->vm & 1) {
3249 return false;
3250 }
3251
3252 if (!narrowfn) {
3253 return false;
3254 }
3255
3256 if (!vfp_access_check(s)) {
3257 return true;
3258 }
3259
3260 rm = tcg_temp_new_i64();
3261 rd0 = tcg_temp_new_i32();
3262 rd1 = tcg_temp_new_i32();
3263
3264 read_neon_element64(rm, a->vm, 0, MO_64);
3265 narrowfn(rd0, cpu_env, rm);
3266 read_neon_element64(rm, a->vm, 1, MO_64);
3267 narrowfn(rd1, cpu_env, rm);
3268 write_neon_element32(rd0, a->vd, 0, MO_32);
3269 write_neon_element32(rd1, a->vd, 1, MO_32);
3270 tcg_temp_free_i32(rd0);
3271 tcg_temp_free_i32(rd1);
3272 tcg_temp_free_i64(rm);
3273 return true;
3274 }
3275
3276 #define DO_VMOVN(INSN, FUNC) \
3277 static bool trans_##INSN(DisasContext *s, arg_2misc *a) \
3278 { \
3279 static NeonGenNarrowEnvFn * const narrowfn[] = { \
3280 FUNC##8, \
3281 FUNC##16, \
3282 FUNC##32, \
3283 NULL, \
3284 }; \
3285 return do_vmovn(s, a, narrowfn[a->size]); \
3286 }
3287
3288 DO_VMOVN(VMOVN, gen_neon_narrow_u)
3289 DO_VMOVN(VQMOVUN, gen_helper_neon_unarrow_sat)
3290 DO_VMOVN(VQMOVN_S, gen_helper_neon_narrow_sat_s)
3291 DO_VMOVN(VQMOVN_U, gen_helper_neon_narrow_sat_u)
3292
3293 static bool trans_VSHLL(DisasContext *s, arg_2misc *a)
3294 {
3295 TCGv_i32 rm0, rm1;
3296 TCGv_i64 rd;
3297 static NeonGenWidenFn * const widenfns[] = {
3298 gen_helper_neon_widen_u8,
3299 gen_helper_neon_widen_u16,
3300 tcg_gen_extu_i32_i64,
3301 NULL,
3302 };
3303 NeonGenWidenFn *widenfn = widenfns[a->size];
3304
3305 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
3306 return false;
3307 }
3308
3309 /* UNDEF accesses to D16-D31 if they don't exist. */
3310 if (!dc_isar_feature(aa32_simd_r32, s) &&
3311 ((a->vd | a->vm) & 0x10)) {
3312 return false;
3313 }
3314
3315 if (a->vd & 1) {
3316 return false;
3317 }
3318
3319 if (!widenfn) {
3320 return false;
3321 }
3322
3323 if (!vfp_access_check(s)) {
3324 return true;
3325 }
3326
3327 rd = tcg_temp_new_i64();
3328 rm0 = tcg_temp_new_i32();
3329 rm1 = tcg_temp_new_i32();
3330
3331 read_neon_element32(rm0, a->vm, 0, MO_32);
3332 read_neon_element32(rm1, a->vm, 1, MO_32);
3333
3334 widenfn(rd, rm0);
3335 tcg_gen_shli_i64(rd, rd, 8 << a->size);
3336 write_neon_element64(rd, a->vd, 0, MO_64);
3337 widenfn(rd, rm1);
3338 tcg_gen_shli_i64(rd, rd, 8 << a->size);
3339 write_neon_element64(rd, a->vd, 1, MO_64);
3340
3341 tcg_temp_free_i64(rd);
3342 tcg_temp_free_i32(rm0);
3343 tcg_temp_free_i32(rm1);
3344 return true;
3345 }
3346
3347 static bool trans_VCVT_B16_F32(DisasContext *s, arg_2misc *a)
3348 {
3349 TCGv_ptr fpst;
3350 TCGv_i64 tmp;
3351 TCGv_i32 dst0, dst1;
3352
3353 if (!dc_isar_feature(aa32_bf16, s)) {
3354 return false;
3355 }
3356
3357 /* UNDEF accesses to D16-D31 if they don't exist. */
3358 if (!dc_isar_feature(aa32_simd_r32, s) &&
3359 ((a->vd | a->vm) & 0x10)) {
3360 return false;
3361 }
3362
3363 if ((a->vm & 1) || (a->size != 1)) {
3364 return false;
3365 }
3366
3367 if (!vfp_access_check(s)) {
3368 return true;
3369 }
3370
3371 fpst = fpstatus_ptr(FPST_STD);
3372 tmp = tcg_temp_new_i64();
3373 dst0 = tcg_temp_new_i32();
3374 dst1 = tcg_temp_new_i32();
3375
3376 read_neon_element64(tmp, a->vm, 0, MO_64);
3377 gen_helper_bfcvt_pair(dst0, tmp, fpst);
3378
3379 read_neon_element64(tmp, a->vm, 1, MO_64);
3380 gen_helper_bfcvt_pair(dst1, tmp, fpst);
3381
3382 write_neon_element32(dst0, a->vd, 0, MO_32);
3383 write_neon_element32(dst1, a->vd, 1, MO_32);
3384
3385 tcg_temp_free_i64(tmp);
3386 tcg_temp_free_i32(dst0);
3387 tcg_temp_free_i32(dst1);
3388 tcg_temp_free_ptr(fpst);
3389 return true;
3390 }
3391
3392 static bool trans_VCVT_F16_F32(DisasContext *s, arg_2misc *a)
3393 {
3394 TCGv_ptr fpst;
3395 TCGv_i32 ahp, tmp, tmp2, tmp3;
3396
3397 if (!arm_dc_feature(s, ARM_FEATURE_NEON) ||
3398 !dc_isar_feature(aa32_fp16_spconv, s)) {
3399 return false;
3400 }
3401
3402 /* UNDEF accesses to D16-D31 if they don't exist. */
3403 if (!dc_isar_feature(aa32_simd_r32, s) &&
3404 ((a->vd | a->vm) & 0x10)) {
3405 return false;
3406 }
3407
3408 if ((a->vm & 1) || (a->size != 1)) {
3409 return false;
3410 }
3411
3412 if (!vfp_access_check(s)) {
3413 return true;
3414 }
3415
3416 fpst = fpstatus_ptr(FPST_STD);
3417 ahp = get_ahp_flag();
3418 tmp = tcg_temp_new_i32();
3419 read_neon_element32(tmp, a->vm, 0, MO_32);
3420 gen_helper_vfp_fcvt_f32_to_f16(tmp, tmp, fpst, ahp);
3421 tmp2 = tcg_temp_new_i32();
3422 read_neon_element32(tmp2, a->vm, 1, MO_32);
3423 gen_helper_vfp_fcvt_f32_to_f16(tmp2, tmp2, fpst, ahp);
3424 tcg_gen_shli_i32(tmp2, tmp2, 16);
3425 tcg_gen_or_i32(tmp2, tmp2, tmp);
3426 read_neon_element32(tmp, a->vm, 2, MO_32);
3427 gen_helper_vfp_fcvt_f32_to_f16(tmp, tmp, fpst, ahp);
3428 tmp3 = tcg_temp_new_i32();
3429 read_neon_element32(tmp3, a->vm, 3, MO_32);
3430 write_neon_element32(tmp2, a->vd, 0, MO_32);
3431 tcg_temp_free_i32(tmp2);
3432 gen_helper_vfp_fcvt_f32_to_f16(tmp3, tmp3, fpst, ahp);
3433 tcg_gen_shli_i32(tmp3, tmp3, 16);
3434 tcg_gen_or_i32(tmp3, tmp3, tmp);
3435 write_neon_element32(tmp3, a->vd, 1, MO_32);
3436 tcg_temp_free_i32(tmp3);
3437 tcg_temp_free_i32(tmp);
3438 tcg_temp_free_i32(ahp);
3439 tcg_temp_free_ptr(fpst);
3440
3441 return true;
3442 }
3443
3444 static bool trans_VCVT_F32_F16(DisasContext *s, arg_2misc *a)
3445 {
3446 TCGv_ptr fpst;
3447 TCGv_i32 ahp, tmp, tmp2, tmp3;
3448
3449 if (!arm_dc_feature(s, ARM_FEATURE_NEON) ||
3450 !dc_isar_feature(aa32_fp16_spconv, s)) {
3451 return false;
3452 }
3453
3454 /* UNDEF accesses to D16-D31 if they don't exist. */
3455 if (!dc_isar_feature(aa32_simd_r32, s) &&
3456 ((a->vd | a->vm) & 0x10)) {
3457 return false;
3458 }
3459
3460 if ((a->vd & 1) || (a->size != 1)) {
3461 return false;
3462 }
3463
3464 if (!vfp_access_check(s)) {
3465 return true;
3466 }
3467
3468 fpst = fpstatus_ptr(FPST_STD);
3469 ahp = get_ahp_flag();
3470 tmp3 = tcg_temp_new_i32();
3471 tmp2 = tcg_temp_new_i32();
3472 tmp = tcg_temp_new_i32();
3473 read_neon_element32(tmp, a->vm, 0, MO_32);
3474 read_neon_element32(tmp2, a->vm, 1, MO_32);
3475 tcg_gen_ext16u_i32(tmp3, tmp);
3476 gen_helper_vfp_fcvt_f16_to_f32(tmp3, tmp3, fpst, ahp);
3477 write_neon_element32(tmp3, a->vd, 0, MO_32);
3478 tcg_gen_shri_i32(tmp, tmp, 16);
3479 gen_helper_vfp_fcvt_f16_to_f32(tmp, tmp, fpst, ahp);
3480 write_neon_element32(tmp, a->vd, 1, MO_32);
3481 tcg_temp_free_i32(tmp);
3482 tcg_gen_ext16u_i32(tmp3, tmp2);
3483 gen_helper_vfp_fcvt_f16_to_f32(tmp3, tmp3, fpst, ahp);
3484 write_neon_element32(tmp3, a->vd, 2, MO_32);
3485 tcg_temp_free_i32(tmp3);
3486 tcg_gen_shri_i32(tmp2, tmp2, 16);
3487 gen_helper_vfp_fcvt_f16_to_f32(tmp2, tmp2, fpst, ahp);
3488 write_neon_element32(tmp2, a->vd, 3, MO_32);
3489 tcg_temp_free_i32(tmp2);
3490 tcg_temp_free_i32(ahp);
3491 tcg_temp_free_ptr(fpst);
3492
3493 return true;
3494 }
3495
3496 static bool do_2misc_vec(DisasContext *s, arg_2misc *a, GVecGen2Fn *fn)
3497 {
3498 int vec_size = a->q ? 16 : 8;
3499 int rd_ofs = neon_full_reg_offset(a->vd);
3500 int rm_ofs = neon_full_reg_offset(a->vm);
3501
3502 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
3503 return false;
3504 }
3505
3506 /* UNDEF accesses to D16-D31 if they don't exist. */
3507 if (!dc_isar_feature(aa32_simd_r32, s) &&
3508 ((a->vd | a->vm) & 0x10)) {
3509 return false;
3510 }
3511
3512 if (a->size == 3) {
3513 return false;
3514 }
3515
3516 if ((a->vd | a->vm) & a->q) {
3517 return false;
3518 }
3519
3520 if (!vfp_access_check(s)) {
3521 return true;
3522 }
3523
3524 fn(a->size, rd_ofs, rm_ofs, vec_size, vec_size);
3525
3526 return true;
3527 }
3528
3529 #define DO_2MISC_VEC(INSN, FN) \
3530 static bool trans_##INSN(DisasContext *s, arg_2misc *a) \
3531 { \
3532 return do_2misc_vec(s, a, FN); \
3533 }
3534
3535 DO_2MISC_VEC(VNEG, tcg_gen_gvec_neg)
3536 DO_2MISC_VEC(VABS, tcg_gen_gvec_abs)
3537 DO_2MISC_VEC(VCEQ0, gen_gvec_ceq0)
3538 DO_2MISC_VEC(VCGT0, gen_gvec_cgt0)
3539 DO_2MISC_VEC(VCLE0, gen_gvec_cle0)
3540 DO_2MISC_VEC(VCGE0, gen_gvec_cge0)
3541 DO_2MISC_VEC(VCLT0, gen_gvec_clt0)
3542
3543 static bool trans_VMVN(DisasContext *s, arg_2misc *a)
3544 {
3545 if (a->size != 0) {
3546 return false;
3547 }
3548 return do_2misc_vec(s, a, tcg_gen_gvec_not);
3549 }
3550
3551 #define WRAP_2M_3_OOL_FN(WRAPNAME, FUNC, DATA) \
3552 static void WRAPNAME(unsigned vece, uint32_t rd_ofs, \
3553 uint32_t rm_ofs, uint32_t oprsz, \
3554 uint32_t maxsz) \
3555 { \
3556 tcg_gen_gvec_3_ool(rd_ofs, rd_ofs, rm_ofs, oprsz, maxsz, \
3557 DATA, FUNC); \
3558 }
3559
3560 #define WRAP_2M_2_OOL_FN(WRAPNAME, FUNC, DATA) \
3561 static void WRAPNAME(unsigned vece, uint32_t rd_ofs, \
3562 uint32_t rm_ofs, uint32_t oprsz, \
3563 uint32_t maxsz) \
3564 { \
3565 tcg_gen_gvec_2_ool(rd_ofs, rm_ofs, oprsz, maxsz, DATA, FUNC); \
3566 }
3567
3568 WRAP_2M_3_OOL_FN(gen_AESE, gen_helper_crypto_aese, 0)
3569 WRAP_2M_3_OOL_FN(gen_AESD, gen_helper_crypto_aese, 1)
3570 WRAP_2M_2_OOL_FN(gen_AESMC, gen_helper_crypto_aesmc, 0)
3571 WRAP_2M_2_OOL_FN(gen_AESIMC, gen_helper_crypto_aesmc, 1)
3572 WRAP_2M_2_OOL_FN(gen_SHA1H, gen_helper_crypto_sha1h, 0)
3573 WRAP_2M_2_OOL_FN(gen_SHA1SU1, gen_helper_crypto_sha1su1, 0)
3574 WRAP_2M_2_OOL_FN(gen_SHA256SU0, gen_helper_crypto_sha256su0, 0)
3575
3576 #define DO_2M_CRYPTO(INSN, FEATURE, SIZE) \
3577 static bool trans_##INSN(DisasContext *s, arg_2misc *a) \
3578 { \
3579 if (!dc_isar_feature(FEATURE, s) || a->size != SIZE) { \
3580 return false; \
3581 } \
3582 return do_2misc_vec(s, a, gen_##INSN); \
3583 }
3584
3585 DO_2M_CRYPTO(AESE, aa32_aes, 0)
3586 DO_2M_CRYPTO(AESD, aa32_aes, 0)
3587 DO_2M_CRYPTO(AESMC, aa32_aes, 0)
3588 DO_2M_CRYPTO(AESIMC, aa32_aes, 0)
3589 DO_2M_CRYPTO(SHA1H, aa32_sha1, 2)
3590 DO_2M_CRYPTO(SHA1SU1, aa32_sha1, 2)
3591 DO_2M_CRYPTO(SHA256SU0, aa32_sha2, 2)
3592
3593 static bool do_2misc(DisasContext *s, arg_2misc *a, NeonGenOneOpFn *fn)
3594 {
3595 TCGv_i32 tmp;
3596 int pass;
3597
3598 /* Handle a 2-reg-misc operation by iterating 32 bits at a time */
3599 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
3600 return false;
3601 }
3602
3603 /* UNDEF accesses to D16-D31 if they don't exist. */
3604 if (!dc_isar_feature(aa32_simd_r32, s) &&
3605 ((a->vd | a->vm) & 0x10)) {
3606 return false;
3607 }
3608
3609 if (!fn) {
3610 return false;
3611 }
3612
3613 if ((a->vd | a->vm) & a->q) {
3614 return false;
3615 }
3616
3617 if (!vfp_access_check(s)) {
3618 return true;
3619 }
3620
3621 tmp = tcg_temp_new_i32();
3622 for (pass = 0; pass < (a->q ? 4 : 2); pass++) {
3623 read_neon_element32(tmp, a->vm, pass, MO_32);
3624 fn(tmp, tmp);
3625 write_neon_element32(tmp, a->vd, pass, MO_32);
3626 }
3627 tcg_temp_free_i32(tmp);
3628
3629 return true;
3630 }
3631
3632 static bool trans_VREV32(DisasContext *s, arg_2misc *a)
3633 {
3634 static NeonGenOneOpFn * const fn[] = {
3635 tcg_gen_bswap32_i32,
3636 gen_swap_half,
3637 NULL,
3638 NULL,
3639 };
3640 return do_2misc(s, a, fn[a->size]);
3641 }
3642
3643 static bool trans_VREV16(DisasContext *s, arg_2misc *a)
3644 {
3645 if (a->size != 0) {
3646 return false;
3647 }
3648 return do_2misc(s, a, gen_rev16);
3649 }
3650
3651 static bool trans_VCLS(DisasContext *s, arg_2misc *a)
3652 {
3653 static NeonGenOneOpFn * const fn[] = {
3654 gen_helper_neon_cls_s8,
3655 gen_helper_neon_cls_s16,
3656 gen_helper_neon_cls_s32,
3657 NULL,
3658 };
3659 return do_2misc(s, a, fn[a->size]);
3660 }
3661
3662 static void do_VCLZ_32(TCGv_i32 rd, TCGv_i32 rm)
3663 {
3664 tcg_gen_clzi_i32(rd, rm, 32);
3665 }
3666
3667 static bool trans_VCLZ(DisasContext *s, arg_2misc *a)
3668 {
3669 static NeonGenOneOpFn * const fn[] = {
3670 gen_helper_neon_clz_u8,
3671 gen_helper_neon_clz_u16,
3672 do_VCLZ_32,
3673 NULL,
3674 };
3675 return do_2misc(s, a, fn[a->size]);
3676 }
3677
3678 static bool trans_VCNT(DisasContext *s, arg_2misc *a)
3679 {
3680 if (a->size != 0) {
3681 return false;
3682 }
3683 return do_2misc(s, a, gen_helper_neon_cnt_u8);
3684 }
3685
3686 static void gen_VABS_F(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3687 uint32_t oprsz, uint32_t maxsz)
3688 {
3689 tcg_gen_gvec_andi(vece, rd_ofs, rm_ofs,
3690 vece == MO_16 ? 0x7fff : 0x7fffffff,
3691 oprsz, maxsz);
3692 }
3693
3694 static bool trans_VABS_F(DisasContext *s, arg_2misc *a)
3695 {
3696 if (a->size == MO_16) {
3697 if (!dc_isar_feature(aa32_fp16_arith, s)) {
3698 return false;
3699 }
3700 } else if (a->size != MO_32) {
3701 return false;
3702 }
3703 return do_2misc_vec(s, a, gen_VABS_F);
3704 }
3705
3706 static void gen_VNEG_F(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3707 uint32_t oprsz, uint32_t maxsz)
3708 {
3709 tcg_gen_gvec_xori(vece, rd_ofs, rm_ofs,
3710 vece == MO_16 ? 0x8000 : 0x80000000,
3711 oprsz, maxsz);
3712 }
3713
3714 static bool trans_VNEG_F(DisasContext *s, arg_2misc *a)
3715 {
3716 if (a->size == MO_16) {
3717 if (!dc_isar_feature(aa32_fp16_arith, s)) {
3718 return false;
3719 }
3720 } else if (a->size != MO_32) {
3721 return false;
3722 }
3723 return do_2misc_vec(s, a, gen_VNEG_F);
3724 }
3725
3726 static bool trans_VRECPE(DisasContext *s, arg_2misc *a)
3727 {
3728 if (a->size != 2) {
3729 return false;
3730 }
3731 return do_2misc(s, a, gen_helper_recpe_u32);
3732 }
3733
3734 static bool trans_VRSQRTE(DisasContext *s, arg_2misc *a)
3735 {
3736 if (a->size != 2) {
3737 return false;
3738 }
3739 return do_2misc(s, a, gen_helper_rsqrte_u32);
3740 }
3741
3742 #define WRAP_1OP_ENV_FN(WRAPNAME, FUNC) \
3743 static void WRAPNAME(TCGv_i32 d, TCGv_i32 m) \
3744 { \
3745 FUNC(d, cpu_env, m); \
3746 }
3747
3748 WRAP_1OP_ENV_FN(gen_VQABS_s8, gen_helper_neon_qabs_s8)
3749 WRAP_1OP_ENV_FN(gen_VQABS_s16, gen_helper_neon_qabs_s16)
3750 WRAP_1OP_ENV_FN(gen_VQABS_s32, gen_helper_neon_qabs_s32)
3751 WRAP_1OP_ENV_FN(gen_VQNEG_s8, gen_helper_neon_qneg_s8)
3752 WRAP_1OP_ENV_FN(gen_VQNEG_s16, gen_helper_neon_qneg_s16)
3753 WRAP_1OP_ENV_FN(gen_VQNEG_s32, gen_helper_neon_qneg_s32)
3754
3755 static bool trans_VQABS(DisasContext *s, arg_2misc *a)
3756 {
3757 static NeonGenOneOpFn * const fn[] = {
3758 gen_VQABS_s8,
3759 gen_VQABS_s16,
3760 gen_VQABS_s32,
3761 NULL,
3762 };
3763 return do_2misc(s, a, fn[a->size]);
3764 }
3765
3766 static bool trans_VQNEG(DisasContext *s, arg_2misc *a)
3767 {
3768 static NeonGenOneOpFn * const fn[] = {
3769 gen_VQNEG_s8,
3770 gen_VQNEG_s16,
3771 gen_VQNEG_s32,
3772 NULL,
3773 };
3774 return do_2misc(s, a, fn[a->size]);
3775 }
3776
3777 #define DO_2MISC_FP_VEC(INSN, HFUNC, SFUNC) \
3778 static void gen_##INSN(unsigned vece, uint32_t rd_ofs, \
3779 uint32_t rm_ofs, \
3780 uint32_t oprsz, uint32_t maxsz) \
3781 { \
3782 static gen_helper_gvec_2_ptr * const fns[4] = { \
3783 NULL, HFUNC, SFUNC, NULL, \
3784 }; \
3785 TCGv_ptr fpst; \
3786 fpst = fpstatus_ptr(vece == MO_16 ? FPST_STD_F16 : FPST_STD); \
3787 tcg_gen_gvec_2_ptr(rd_ofs, rm_ofs, fpst, oprsz, maxsz, 0, \
3788 fns[vece]); \
3789 tcg_temp_free_ptr(fpst); \
3790 } \
3791 static bool trans_##INSN(DisasContext *s, arg_2misc *a) \
3792 { \
3793 if (a->size == MO_16) { \
3794 if (!dc_isar_feature(aa32_fp16_arith, s)) { \
3795 return false; \
3796 } \
3797 } else if (a->size != MO_32) { \
3798 return false; \
3799 } \
3800 return do_2misc_vec(s, a, gen_##INSN); \
3801 }
3802
3803 DO_2MISC_FP_VEC(VRECPE_F, gen_helper_gvec_frecpe_h, gen_helper_gvec_frecpe_s)
3804 DO_2MISC_FP_VEC(VRSQRTE_F, gen_helper_gvec_frsqrte_h, gen_helper_gvec_frsqrte_s)
3805 DO_2MISC_FP_VEC(VCGT0_F, gen_helper_gvec_fcgt0_h, gen_helper_gvec_fcgt0_s)
3806 DO_2MISC_FP_VEC(VCGE0_F, gen_helper_gvec_fcge0_h, gen_helper_gvec_fcge0_s)
3807 DO_2MISC_FP_VEC(VCEQ0_F, gen_helper_gvec_fceq0_h, gen_helper_gvec_fceq0_s)
3808 DO_2MISC_FP_VEC(VCLT0_F, gen_helper_gvec_fclt0_h, gen_helper_gvec_fclt0_s)
3809 DO_2MISC_FP_VEC(VCLE0_F, gen_helper_gvec_fcle0_h, gen_helper_gvec_fcle0_s)
3810 DO_2MISC_FP_VEC(VCVT_FS, gen_helper_gvec_sstoh, gen_helper_gvec_sitos)
3811 DO_2MISC_FP_VEC(VCVT_FU, gen_helper_gvec_ustoh, gen_helper_gvec_uitos)
3812 DO_2MISC_FP_VEC(VCVT_SF, gen_helper_gvec_tosszh, gen_helper_gvec_tosizs)
3813 DO_2MISC_FP_VEC(VCVT_UF, gen_helper_gvec_touszh, gen_helper_gvec_touizs)
3814
3815 DO_2MISC_FP_VEC(VRINTX_impl, gen_helper_gvec_vrintx_h, gen_helper_gvec_vrintx_s)
3816
3817 static bool trans_VRINTX(DisasContext *s, arg_2misc *a)
3818 {
3819 if (!arm_dc_feature(s, ARM_FEATURE_V8)) {
3820 return false;
3821 }
3822 return trans_VRINTX_impl(s, a);
3823 }
3824
3825 #define DO_VEC_RMODE(INSN, RMODE, OP) \
3826 static void gen_##INSN(unsigned vece, uint32_t rd_ofs, \
3827 uint32_t rm_ofs, \
3828 uint32_t oprsz, uint32_t maxsz) \
3829 { \
3830 static gen_helper_gvec_2_ptr * const fns[4] = { \
3831 NULL, \
3832 gen_helper_gvec_##OP##h, \
3833 gen_helper_gvec_##OP##s, \
3834 NULL, \
3835 }; \
3836 TCGv_ptr fpst; \
3837 fpst = fpstatus_ptr(vece == 1 ? FPST_STD_F16 : FPST_STD); \
3838 tcg_gen_gvec_2_ptr(rd_ofs, rm_ofs, fpst, oprsz, maxsz, \
3839 arm_rmode_to_sf(RMODE), fns[vece]); \
3840 tcg_temp_free_ptr(fpst); \
3841 } \
3842 static bool trans_##INSN(DisasContext *s, arg_2misc *a) \
3843 { \
3844 if (!arm_dc_feature(s, ARM_FEATURE_V8)) { \
3845 return false; \
3846 } \
3847 if (a->size == MO_16) { \
3848 if (!dc_isar_feature(aa32_fp16_arith, s)) { \
3849 return false; \
3850 } \
3851 } else if (a->size != MO_32) { \
3852 return false; \
3853 } \
3854 return do_2misc_vec(s, a, gen_##INSN); \
3855 }
3856
3857 DO_VEC_RMODE(VCVTAU, FPROUNDING_TIEAWAY, vcvt_rm_u)
3858 DO_VEC_RMODE(VCVTAS, FPROUNDING_TIEAWAY, vcvt_rm_s)
3859 DO_VEC_RMODE(VCVTNU, FPROUNDING_TIEEVEN, vcvt_rm_u)
3860 DO_VEC_RMODE(VCVTNS, FPROUNDING_TIEEVEN, vcvt_rm_s)
3861 DO_VEC_RMODE(VCVTPU, FPROUNDING_POSINF, vcvt_rm_u)
3862 DO_VEC_RMODE(VCVTPS, FPROUNDING_POSINF, vcvt_rm_s)
3863 DO_VEC_RMODE(VCVTMU, FPROUNDING_NEGINF, vcvt_rm_u)
3864 DO_VEC_RMODE(VCVTMS, FPROUNDING_NEGINF, vcvt_rm_s)
3865
3866 DO_VEC_RMODE(VRINTN, FPROUNDING_TIEEVEN, vrint_rm_)
3867 DO_VEC_RMODE(VRINTA, FPROUNDING_TIEAWAY, vrint_rm_)
3868 DO_VEC_RMODE(VRINTZ, FPROUNDING_ZERO, vrint_rm_)
3869 DO_VEC_RMODE(VRINTM, FPROUNDING_NEGINF, vrint_rm_)
3870 DO_VEC_RMODE(VRINTP, FPROUNDING_POSINF, vrint_rm_)
3871
3872 static bool trans_VSWP(DisasContext *s, arg_2misc *a)
3873 {
3874 TCGv_i64 rm, rd;
3875 int pass;
3876
3877 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
3878 return false;
3879 }
3880
3881 /* UNDEF accesses to D16-D31 if they don't exist. */
3882 if (!dc_isar_feature(aa32_simd_r32, s) &&
3883 ((a->vd | a->vm) & 0x10)) {
3884 return false;
3885 }
3886
3887 if (a->size != 0) {
3888 return false;
3889 }
3890
3891 if ((a->vd | a->vm) & a->q) {
3892 return false;
3893 }
3894
3895 if (!vfp_access_check(s)) {
3896 return true;
3897 }
3898
3899 rm = tcg_temp_new_i64();
3900 rd = tcg_temp_new_i64();
3901 for (pass = 0; pass < (a->q ? 2 : 1); pass++) {
3902 read_neon_element64(rm, a->vm, pass, MO_64);
3903 read_neon_element64(rd, a->vd, pass, MO_64);
3904 write_neon_element64(rm, a->vd, pass, MO_64);
3905 write_neon_element64(rd, a->vm, pass, MO_64);
3906 }
3907 tcg_temp_free_i64(rm);
3908 tcg_temp_free_i64(rd);
3909
3910 return true;
3911 }
3912 static void gen_neon_trn_u8(TCGv_i32 t0, TCGv_i32 t1)
3913 {
3914 TCGv_i32 rd, tmp;
3915
3916 rd = tcg_temp_new_i32();
3917 tmp = tcg_temp_new_i32();
3918
3919 tcg_gen_shli_i32(rd, t0, 8);
3920 tcg_gen_andi_i32(rd, rd, 0xff00ff00);
3921 tcg_gen_andi_i32(tmp, t1, 0x00ff00ff);
3922 tcg_gen_or_i32(rd, rd, tmp);
3923
3924 tcg_gen_shri_i32(t1, t1, 8);
3925 tcg_gen_andi_i32(t1, t1, 0x00ff00ff);
3926 tcg_gen_andi_i32(tmp, t0, 0xff00ff00);
3927 tcg_gen_or_i32(t1, t1, tmp);
3928 tcg_gen_mov_i32(t0, rd);
3929
3930 tcg_temp_free_i32(tmp);
3931 tcg_temp_free_i32(rd);
3932 }
3933
3934 static void gen_neon_trn_u16(TCGv_i32 t0, TCGv_i32 t1)
3935 {
3936 TCGv_i32 rd, tmp;
3937
3938 rd = tcg_temp_new_i32();
3939 tmp = tcg_temp_new_i32();
3940
3941 tcg_gen_shli_i32(rd, t0, 16);
3942 tcg_gen_andi_i32(tmp, t1, 0xffff);
3943 tcg_gen_or_i32(rd, rd, tmp);
3944 tcg_gen_shri_i32(t1, t1, 16);
3945 tcg_gen_andi_i32(tmp, t0, 0xffff0000);
3946 tcg_gen_or_i32(t1, t1, tmp);
3947 tcg_gen_mov_i32(t0, rd);
3948
3949 tcg_temp_free_i32(tmp);
3950 tcg_temp_free_i32(rd);
3951 }
3952
3953 static bool trans_VTRN(DisasContext *s, arg_2misc *a)
3954 {
3955 TCGv_i32 tmp, tmp2;
3956 int pass;
3957
3958 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
3959 return false;
3960 }
3961
3962 /* UNDEF accesses to D16-D31 if they don't exist. */
3963 if (!dc_isar_feature(aa32_simd_r32, s) &&
3964 ((a->vd | a->vm) & 0x10)) {
3965 return false;
3966 }
3967
3968 if ((a->vd | a->vm) & a->q) {
3969 return false;
3970 }
3971
3972 if (a->size == 3) {
3973 return false;
3974 }
3975
3976 if (!vfp_access_check(s)) {
3977 return true;
3978 }
3979
3980 tmp = tcg_temp_new_i32();
3981 tmp2 = tcg_temp_new_i32();
3982 if (a->size == MO_32) {
3983 for (pass = 0; pass < (a->q ? 4 : 2); pass += 2) {
3984 read_neon_element32(tmp, a->vm, pass, MO_32);
3985 read_neon_element32(tmp2, a->vd, pass + 1, MO_32);
3986 write_neon_element32(tmp2, a->vm, pass, MO_32);
3987 write_neon_element32(tmp, a->vd, pass + 1, MO_32);
3988 }
3989 } else {
3990 for (pass = 0; pass < (a->q ? 4 : 2); pass++) {
3991 read_neon_element32(tmp, a->vm, pass, MO_32);
3992 read_neon_element32(tmp2, a->vd, pass, MO_32);
3993 if (a->size == MO_8) {
3994 gen_neon_trn_u8(tmp, tmp2);
3995 } else {
3996 gen_neon_trn_u16(tmp, tmp2);
3997 }
3998 write_neon_element32(tmp2, a->vm, pass, MO_32);
3999 write_neon_element32(tmp, a->vd, pass, MO_32);
4000 }
4001 }
4002 tcg_temp_free_i32(tmp);
4003 tcg_temp_free_i32(tmp2);
4004 return true;
4005 }
4006
4007 static bool trans_VSMMLA(DisasContext *s, arg_VSMMLA *a)
4008 {
4009 if (!dc_isar_feature(aa32_i8mm, s)) {
4010 return false;
4011 }
4012 return do_neon_ddda(s, 7, a->vd, a->vn, a->vm, 0,
4013 gen_helper_gvec_smmla_b);
4014 }
4015
4016 static bool trans_VUMMLA(DisasContext *s, arg_VUMMLA *a)
4017 {
4018 if (!dc_isar_feature(aa32_i8mm, s)) {
4019 return false;
4020 }
4021 return do_neon_ddda(s, 7, a->vd, a->vn, a->vm, 0,
4022 gen_helper_gvec_ummla_b);
4023 }
4024
4025 static bool trans_VUSMMLA(DisasContext *s, arg_VUSMMLA *a)
4026 {
4027 if (!dc_isar_feature(aa32_i8mm, s)) {
4028 return false;
4029 }
4030 return do_neon_ddda(s, 7, a->vd, a->vn, a->vm, 0,
4031 gen_helper_gvec_usmmla_b);
4032 }
4033
4034 static bool trans_VMMLA_b16(DisasContext *s, arg_VMMLA_b16 *a)
4035 {
4036 if (!dc_isar_feature(aa32_bf16, s)) {
4037 return false;
4038 }
4039 return do_neon_ddda(s, 7, a->vd, a->vn, a->vm, 0,
4040 gen_helper_gvec_bfmmla);
4041 }
4042
4043 static bool trans_VFMA_b16(DisasContext *s, arg_VFMA_b16 *a)
4044 {
4045 if (!dc_isar_feature(aa32_bf16, s)) {
4046 return false;
4047 }
4048 return do_neon_ddda_fpst(s, 7, a->vd, a->vn, a->vm, a->q, FPST_STD,
4049 gen_helper_gvec_bfmlal);
4050 }
4051
4052 static bool trans_VFMA_b16_scal(DisasContext *s, arg_VFMA_b16_scal *a)
4053 {
4054 if (!dc_isar_feature(aa32_bf16, s)) {
4055 return false;
4056 }
4057 return do_neon_ddda_fpst(s, 6, a->vd, a->vn, a->vm,
4058 (a->index << 1) | a->q, FPST_STD,
4059 gen_helper_gvec_bfmlal_idx);
4060 }
QEmu