xref: /openbmc/qemu/target/arm/tcg/translate-neon.c (revision 7d87775f)
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 "translate.h"
25 #include "translate-a32.h"
26 
27 /* Include the generated Neon decoder */
28 #include "decode-neon-dp.c.inc"
29 #include "decode-neon-ls.c.inc"
30 #include "decode-neon-shared.c.inc"
31 
32 static TCGv_ptr vfp_reg_ptr(bool dp, int reg)
33 {
34     TCGv_ptr ret = tcg_temp_new_ptr();
35     tcg_gen_addi_ptr(ret, tcg_env, vfp_reg_offset(dp, reg));
36     return ret;
37 }
38 
39 static void neon_load_element(TCGv_i32 var, int reg, int ele, MemOp mop)
40 {
41     long offset = neon_element_offset(reg, ele, mop & MO_SIZE);
42 
43     switch (mop) {
44     case MO_UB:
45         tcg_gen_ld8u_i32(var, tcg_env, offset);
46         break;
47     case MO_UW:
48         tcg_gen_ld16u_i32(var, tcg_env, offset);
49         break;
50     case MO_UL:
51         tcg_gen_ld_i32(var, tcg_env, offset);
52         break;
53     default:
54         g_assert_not_reached();
55     }
56 }
57 
58 static void neon_load_element64(TCGv_i64 var, int reg, int ele, MemOp mop)
59 {
60     long offset = neon_element_offset(reg, ele, mop & MO_SIZE);
61 
62     switch (mop) {
63     case MO_UB:
64         tcg_gen_ld8u_i64(var, tcg_env, offset);
65         break;
66     case MO_UW:
67         tcg_gen_ld16u_i64(var, tcg_env, offset);
68         break;
69     case MO_UL:
70         tcg_gen_ld32u_i64(var, tcg_env, offset);
71         break;
72     case MO_UQ:
73         tcg_gen_ld_i64(var, tcg_env, offset);
74         break;
75     default:
76         g_assert_not_reached();
77     }
78 }
79 
80 static void neon_store_element(int reg, int ele, MemOp size, TCGv_i32 var)
81 {
82     long offset = neon_element_offset(reg, ele, size);
83 
84     switch (size) {
85     case MO_8:
86         tcg_gen_st8_i32(var, tcg_env, offset);
87         break;
88     case MO_16:
89         tcg_gen_st16_i32(var, tcg_env, offset);
90         break;
91     case MO_32:
92         tcg_gen_st_i32(var, tcg_env, offset);
93         break;
94     default:
95         g_assert_not_reached();
96     }
97 }
98 
99 static void neon_store_element64(int reg, int ele, MemOp size, TCGv_i64 var)
100 {
101     long offset = neon_element_offset(reg, ele, size);
102 
103     switch (size) {
104     case MO_8:
105         tcg_gen_st8_i64(var, tcg_env, offset);
106         break;
107     case MO_16:
108         tcg_gen_st16_i64(var, tcg_env, offset);
109         break;
110     case MO_32:
111         tcg_gen_st32_i64(var, tcg_env, offset);
112         break;
113     case MO_64:
114         tcg_gen_st_i64(var, tcg_env, offset);
115         break;
116     default:
117         g_assert_not_reached();
118     }
119 }
120 
121 static bool do_neon_ddda(DisasContext *s, int q, int vd, int vn, int vm,
122                          int data, gen_helper_gvec_4 *fn_gvec)
123 {
124     /* UNDEF accesses to D16-D31 if they don't exist. */
125     if (((vd | vn | vm) & 0x10) && !dc_isar_feature(aa32_simd_r32, s)) {
126         return false;
127     }
128 
129     /*
130      * UNDEF accesses to odd registers for each bit of Q.
131      * Q will be 0b111 for all Q-reg instructions, otherwise
132      * when we have mixed Q- and D-reg inputs.
133      */
134     if (((vd & 1) * 4 | (vn & 1) * 2 | (vm & 1)) & q) {
135         return false;
136     }
137 
138     if (!vfp_access_check(s)) {
139         return true;
140     }
141 
142     int opr_sz = q ? 16 : 8;
143     tcg_gen_gvec_4_ool(vfp_reg_offset(1, vd),
144                        vfp_reg_offset(1, vn),
145                        vfp_reg_offset(1, vm),
146                        vfp_reg_offset(1, vd),
147                        opr_sz, opr_sz, data, fn_gvec);
148     return true;
149 }
150 
151 static bool do_neon_ddda_env(DisasContext *s, int q, int vd, int vn, int vm,
152                              int data, gen_helper_gvec_4_ptr *fn_gvec)
153 {
154     /* UNDEF accesses to D16-D31 if they don't exist. */
155     if (((vd | vn | vm) & 0x10) && !dc_isar_feature(aa32_simd_r32, s)) {
156         return false;
157     }
158 
159     /*
160      * UNDEF accesses to odd registers for each bit of Q.
161      * Q will be 0b111 for all Q-reg instructions, otherwise
162      * when we have mixed Q- and D-reg inputs.
163      */
164     if (((vd & 1) * 4 | (vn & 1) * 2 | (vm & 1)) & q) {
165         return false;
166     }
167 
168     if (!vfp_access_check(s)) {
169         return true;
170     }
171 
172     int opr_sz = q ? 16 : 8;
173     tcg_gen_gvec_4_ptr(vfp_reg_offset(1, vd),
174                        vfp_reg_offset(1, vn),
175                        vfp_reg_offset(1, vm),
176                        vfp_reg_offset(1, vd),
177                        tcg_env,
178                        opr_sz, opr_sz, data, fn_gvec);
179     return true;
180 }
181 
182 static bool do_neon_ddda_fpst(DisasContext *s, int q, int vd, int vn, int vm,
183                               int data, ARMFPStatusFlavour fp_flavour,
184                               gen_helper_gvec_4_ptr *fn_gvec_ptr)
185 {
186     /* UNDEF accesses to D16-D31 if they don't exist. */
187     if (((vd | vn | vm) & 0x10) && !dc_isar_feature(aa32_simd_r32, s)) {
188         return false;
189     }
190 
191     /*
192      * UNDEF accesses to odd registers for each bit of Q.
193      * Q will be 0b111 for all Q-reg instructions, otherwise
194      * when we have mixed Q- and D-reg inputs.
195      */
196     if (((vd & 1) * 4 | (vn & 1) * 2 | (vm & 1)) & q) {
197         return false;
198     }
199 
200     if (!vfp_access_check(s)) {
201         return true;
202     }
203 
204     int opr_sz = q ? 16 : 8;
205     TCGv_ptr fpst = fpstatus_ptr(fp_flavour);
206 
207     tcg_gen_gvec_4_ptr(vfp_reg_offset(1, vd),
208                        vfp_reg_offset(1, vn),
209                        vfp_reg_offset(1, vm),
210                        vfp_reg_offset(1, vd),
211                        fpst, opr_sz, opr_sz, data, fn_gvec_ptr);
212     return true;
213 }
214 
215 static bool trans_VCMLA(DisasContext *s, arg_VCMLA *a)
216 {
217     if (!dc_isar_feature(aa32_vcma, s)) {
218         return false;
219     }
220     if (a->size == MO_16) {
221         if (!dc_isar_feature(aa32_fp16_arith, s)) {
222             return false;
223         }
224         return do_neon_ddda_fpst(s, a->q * 7, a->vd, a->vn, a->vm, a->rot,
225                                  FPST_STD_F16, gen_helper_gvec_fcmlah);
226     }
227     return do_neon_ddda_fpst(s, a->q * 7, a->vd, a->vn, a->vm, a->rot,
228                              FPST_STD, gen_helper_gvec_fcmlas);
229 }
230 
231 static bool trans_VCADD(DisasContext *s, arg_VCADD *a)
232 {
233     int opr_sz;
234     TCGv_ptr fpst;
235     gen_helper_gvec_3_ptr *fn_gvec_ptr;
236 
237     if (!dc_isar_feature(aa32_vcma, s)
238         || (a->size == MO_16 && !dc_isar_feature(aa32_fp16_arith, s))) {
239         return false;
240     }
241 
242     /* UNDEF accesses to D16-D31 if they don't exist. */
243     if (!dc_isar_feature(aa32_simd_r32, s) &&
244         ((a->vd | a->vn | a->vm) & 0x10)) {
245         return false;
246     }
247 
248     if ((a->vn | a->vm | a->vd) & a->q) {
249         return false;
250     }
251 
252     if (!vfp_access_check(s)) {
253         return true;
254     }
255 
256     opr_sz = (1 + a->q) * 8;
257     fpst = fpstatus_ptr(a->size == MO_16 ? FPST_STD_F16 : FPST_STD);
258     fn_gvec_ptr = (a->size == MO_16) ?
259         gen_helper_gvec_fcaddh : gen_helper_gvec_fcadds;
260     tcg_gen_gvec_3_ptr(vfp_reg_offset(1, a->vd),
261                        vfp_reg_offset(1, a->vn),
262                        vfp_reg_offset(1, a->vm),
263                        fpst, opr_sz, opr_sz, a->rot,
264                        fn_gvec_ptr);
265     return true;
266 }
267 
268 static bool trans_VSDOT(DisasContext *s, arg_VSDOT *a)
269 {
270     if (!dc_isar_feature(aa32_dp, s)) {
271         return false;
272     }
273     return do_neon_ddda(s, a->q * 7, a->vd, a->vn, a->vm, 0,
274                         gen_helper_gvec_sdot_b);
275 }
276 
277 static bool trans_VUDOT(DisasContext *s, arg_VUDOT *a)
278 {
279     if (!dc_isar_feature(aa32_dp, s)) {
280         return false;
281     }
282     return do_neon_ddda(s, a->q * 7, a->vd, a->vn, a->vm, 0,
283                         gen_helper_gvec_udot_b);
284 }
285 
286 static bool trans_VUSDOT(DisasContext *s, arg_VUSDOT *a)
287 {
288     if (!dc_isar_feature(aa32_i8mm, s)) {
289         return false;
290     }
291     return do_neon_ddda(s, a->q * 7, a->vd, a->vn, a->vm, 0,
292                         gen_helper_gvec_usdot_b);
293 }
294 
295 static bool trans_VDOT_b16(DisasContext *s, arg_VDOT_b16 *a)
296 {
297     if (!dc_isar_feature(aa32_bf16, s)) {
298         return false;
299     }
300     return do_neon_ddda_env(s, a->q * 7, a->vd, a->vn, a->vm, 0,
301                             gen_helper_gvec_bfdot);
302 }
303 
304 static bool trans_VFML(DisasContext *s, arg_VFML *a)
305 {
306     int opr_sz;
307 
308     if (!dc_isar_feature(aa32_fhm, s)) {
309         return false;
310     }
311 
312     /* UNDEF accesses to D16-D31 if they don't exist. */
313     if (!dc_isar_feature(aa32_simd_r32, s) &&
314         (a->vd & 0x10)) {
315         return false;
316     }
317 
318     if (a->vd & a->q) {
319         return false;
320     }
321 
322     if (!vfp_access_check(s)) {
323         return true;
324     }
325 
326     opr_sz = (1 + a->q) * 8;
327     tcg_gen_gvec_3_ptr(vfp_reg_offset(1, a->vd),
328                        vfp_reg_offset(a->q, a->vn),
329                        vfp_reg_offset(a->q, a->vm),
330                        tcg_env, opr_sz, opr_sz, a->s, /* is_2 == 0 */
331                        gen_helper_gvec_fmlal_a32);
332     return true;
333 }
334 
335 static bool trans_VCMLA_scalar(DisasContext *s, arg_VCMLA_scalar *a)
336 {
337     int data = (a->index << 2) | a->rot;
338 
339     if (!dc_isar_feature(aa32_vcma, s)) {
340         return false;
341     }
342     if (a->size == MO_16) {
343         if (!dc_isar_feature(aa32_fp16_arith, s)) {
344             return false;
345         }
346         return do_neon_ddda_fpst(s, a->q * 6, a->vd, a->vn, a->vm, data,
347                                  FPST_STD_F16, gen_helper_gvec_fcmlah_idx);
348     }
349     return do_neon_ddda_fpst(s, a->q * 6, a->vd, a->vn, a->vm, data,
350                              FPST_STD, gen_helper_gvec_fcmlas_idx);
351 }
352 
353 static bool trans_VSDOT_scalar(DisasContext *s, arg_VSDOT_scalar *a)
354 {
355     if (!dc_isar_feature(aa32_dp, s)) {
356         return false;
357     }
358     return do_neon_ddda(s, a->q * 6, a->vd, a->vn, a->vm, a->index,
359                         gen_helper_gvec_sdot_idx_b);
360 }
361 
362 static bool trans_VUDOT_scalar(DisasContext *s, arg_VUDOT_scalar *a)
363 {
364     if (!dc_isar_feature(aa32_dp, s)) {
365         return false;
366     }
367     return do_neon_ddda(s, a->q * 6, a->vd, a->vn, a->vm, a->index,
368                         gen_helper_gvec_udot_idx_b);
369 }
370 
371 static bool trans_VUSDOT_scalar(DisasContext *s, arg_VUSDOT_scalar *a)
372 {
373     if (!dc_isar_feature(aa32_i8mm, s)) {
374         return false;
375     }
376     return do_neon_ddda(s, a->q * 6, a->vd, a->vn, a->vm, a->index,
377                         gen_helper_gvec_usdot_idx_b);
378 }
379 
380 static bool trans_VSUDOT_scalar(DisasContext *s, arg_VSUDOT_scalar *a)
381 {
382     if (!dc_isar_feature(aa32_i8mm, s)) {
383         return false;
384     }
385     return do_neon_ddda(s, a->q * 6, a->vd, a->vn, a->vm, a->index,
386                         gen_helper_gvec_sudot_idx_b);
387 }
388 
389 static bool trans_VDOT_b16_scal(DisasContext *s, arg_VDOT_b16_scal *a)
390 {
391     if (!dc_isar_feature(aa32_bf16, s)) {
392         return false;
393     }
394     return do_neon_ddda_env(s, a->q * 6, a->vd, a->vn, a->vm, a->index,
395                             gen_helper_gvec_bfdot_idx);
396 }
397 
398 static bool trans_VFML_scalar(DisasContext *s, arg_VFML_scalar *a)
399 {
400     int opr_sz;
401 
402     if (!dc_isar_feature(aa32_fhm, s)) {
403         return false;
404     }
405 
406     /* UNDEF accesses to D16-D31 if they don't exist. */
407     if (!dc_isar_feature(aa32_simd_r32, s) &&
408         ((a->vd & 0x10) || (a->q && (a->vn & 0x10)))) {
409         return false;
410     }
411 
412     if (a->vd & a->q) {
413         return false;
414     }
415 
416     if (!vfp_access_check(s)) {
417         return true;
418     }
419 
420     opr_sz = (1 + a->q) * 8;
421     tcg_gen_gvec_3_ptr(vfp_reg_offset(1, a->vd),
422                        vfp_reg_offset(a->q, a->vn),
423                        vfp_reg_offset(a->q, a->rm),
424                        tcg_env, opr_sz, opr_sz,
425                        (a->index << 2) | a->s, /* is_2 == 0 */
426                        gen_helper_gvec_fmlal_idx_a32);
427     return true;
428 }
429 
430 static struct {
431     int nregs;
432     int interleave;
433     int spacing;
434 } const neon_ls_element_type[11] = {
435     {1, 4, 1},
436     {1, 4, 2},
437     {4, 1, 1},
438     {2, 2, 2},
439     {1, 3, 1},
440     {1, 3, 2},
441     {3, 1, 1},
442     {1, 1, 1},
443     {1, 2, 1},
444     {1, 2, 2},
445     {2, 1, 1}
446 };
447 
448 static void gen_neon_ldst_base_update(DisasContext *s, int rm, int rn,
449                                       int stride)
450 {
451     if (rm != 15) {
452         TCGv_i32 base;
453 
454         base = load_reg(s, rn);
455         if (rm == 13) {
456             tcg_gen_addi_i32(base, base, stride);
457         } else {
458             TCGv_i32 index;
459             index = load_reg(s, rm);
460             tcg_gen_add_i32(base, base, index);
461         }
462         store_reg(s, rn, base);
463     }
464 }
465 
466 static bool trans_VLDST_multiple(DisasContext *s, arg_VLDST_multiple *a)
467 {
468     /* Neon load/store multiple structures */
469     int nregs, interleave, spacing, reg, n;
470     MemOp mop, align, endian;
471     int mmu_idx = get_mem_index(s);
472     int size = a->size;
473     TCGv_i64 tmp64;
474     TCGv_i32 addr;
475 
476     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
477         return false;
478     }
479 
480     /* UNDEF accesses to D16-D31 if they don't exist */
481     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vd & 0x10)) {
482         return false;
483     }
484     if (a->itype > 10) {
485         return false;
486     }
487     /* Catch UNDEF cases for bad values of align field */
488     switch (a->itype & 0xc) {
489     case 4:
490         if (a->align >= 2) {
491             return false;
492         }
493         break;
494     case 8:
495         if (a->align == 3) {
496             return false;
497         }
498         break;
499     default:
500         break;
501     }
502     nregs = neon_ls_element_type[a->itype].nregs;
503     interleave = neon_ls_element_type[a->itype].interleave;
504     spacing = neon_ls_element_type[a->itype].spacing;
505     if (size == 3 && (interleave | spacing) != 1) {
506         return false;
507     }
508 
509     if (!vfp_access_check(s)) {
510         return true;
511     }
512 
513     /* For our purposes, bytes are always little-endian.  */
514     endian = s->be_data;
515     if (size == 0) {
516         endian = MO_LE;
517     }
518 
519     /* Enforce alignment requested by the instruction */
520     if (a->align) {
521         align = pow2_align(a->align + 2); /* 4 ** a->align */
522     } else {
523         align = s->align_mem ? MO_ALIGN : 0;
524     }
525 
526     /*
527      * Consecutive little-endian elements from a single register
528      * can be promoted to a larger little-endian operation.
529      */
530     if (interleave == 1 && endian == MO_LE) {
531         /* Retain any natural alignment. */
532         if (align == MO_ALIGN) {
533             align = pow2_align(size);
534         }
535         size = 3;
536     }
537 
538     tmp64 = tcg_temp_new_i64();
539     addr = tcg_temp_new_i32();
540     load_reg_var(s, addr, a->rn);
541 
542     mop = endian | size | align;
543     for (reg = 0; reg < nregs; reg++) {
544         for (n = 0; n < 8 >> size; n++) {
545             int xs;
546             for (xs = 0; xs < interleave; xs++) {
547                 int tt = a->vd + reg + spacing * xs;
548 
549                 if (a->l) {
550                     gen_aa32_ld_internal_i64(s, tmp64, addr, mmu_idx, mop);
551                     neon_store_element64(tt, n, size, tmp64);
552                 } else {
553                     neon_load_element64(tmp64, tt, n, size);
554                     gen_aa32_st_internal_i64(s, tmp64, addr, mmu_idx, mop);
555                 }
556                 tcg_gen_addi_i32(addr, addr, 1 << size);
557 
558                 /* Subsequent memory operations inherit alignment */
559                 mop &= ~MO_AMASK;
560             }
561         }
562     }
563 
564     gen_neon_ldst_base_update(s, a->rm, a->rn, nregs * interleave * 8);
565     return true;
566 }
567 
568 static bool trans_VLD_all_lanes(DisasContext *s, arg_VLD_all_lanes *a)
569 {
570     /* Neon load single structure to all lanes */
571     int reg, stride, vec_size;
572     int vd = a->vd;
573     int size = a->size;
574     int nregs = a->n + 1;
575     TCGv_i32 addr, tmp;
576     MemOp mop, align;
577 
578     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
579         return false;
580     }
581 
582     /* UNDEF accesses to D16-D31 if they don't exist */
583     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vd & 0x10)) {
584         return false;
585     }
586 
587     align = 0;
588     if (size == 3) {
589         if (nregs != 4 || a->a == 0) {
590             return false;
591         }
592         /* For VLD4 size == 3 a == 1 means 32 bits at 16 byte alignment */
593         size = MO_32;
594         align = MO_ALIGN_16;
595     } else if (a->a) {
596         switch (nregs) {
597         case 1:
598             if (size == 0) {
599                 return false;
600             }
601             align = MO_ALIGN;
602             break;
603         case 2:
604             align = pow2_align(size + 1);
605             break;
606         case 3:
607             return false;
608         case 4:
609             if (size == 2) {
610                 align = pow2_align(3);
611             } else {
612                 align = pow2_align(size + 2);
613             }
614             break;
615         default:
616             g_assert_not_reached();
617         }
618     }
619 
620     if (!vfp_access_check(s)) {
621         return true;
622     }
623 
624     /*
625      * VLD1 to all lanes: T bit indicates how many Dregs to write.
626      * VLD2/3/4 to all lanes: T bit indicates register stride.
627      */
628     stride = a->t ? 2 : 1;
629     vec_size = nregs == 1 ? stride * 8 : 8;
630     mop = size | align;
631     tmp = tcg_temp_new_i32();
632     addr = tcg_temp_new_i32();
633     load_reg_var(s, addr, a->rn);
634     for (reg = 0; reg < nregs; reg++) {
635         gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), mop);
636         if ((vd & 1) && vec_size == 16) {
637             /*
638              * We cannot write 16 bytes at once because the
639              * destination is unaligned.
640              */
641             tcg_gen_gvec_dup_i32(size, neon_full_reg_offset(vd),
642                                  8, 8, tmp);
643             tcg_gen_gvec_mov(0, neon_full_reg_offset(vd + 1),
644                              neon_full_reg_offset(vd), 8, 8);
645         } else {
646             tcg_gen_gvec_dup_i32(size, neon_full_reg_offset(vd),
647                                  vec_size, vec_size, tmp);
648         }
649         tcg_gen_addi_i32(addr, addr, 1 << size);
650         vd += stride;
651 
652         /* Subsequent memory operations inherit alignment */
653         mop &= ~MO_AMASK;
654     }
655 
656     gen_neon_ldst_base_update(s, a->rm, a->rn, (1 << size) * nregs);
657 
658     return true;
659 }
660 
661 static bool trans_VLDST_single(DisasContext *s, arg_VLDST_single *a)
662 {
663     /* Neon load/store single structure to one lane */
664     int reg;
665     int nregs = a->n + 1;
666     int vd = a->vd;
667     TCGv_i32 addr, tmp;
668     MemOp mop;
669 
670     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
671         return false;
672     }
673 
674     /* UNDEF accesses to D16-D31 if they don't exist */
675     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vd & 0x10)) {
676         return false;
677     }
678 
679     /* Catch the UNDEF cases. This is unavoidably a bit messy. */
680     switch (nregs) {
681     case 1:
682         if (a->stride != 1) {
683             return false;
684         }
685         if (((a->align & (1 << a->size)) != 0) ||
686             (a->size == 2 && (a->align == 1 || a->align == 2))) {
687             return false;
688         }
689         break;
690     case 2:
691         if (a->size == 2 && (a->align & 2) != 0) {
692             return false;
693         }
694         break;
695     case 3:
696         if (a->align != 0) {
697             return false;
698         }
699         break;
700     case 4:
701         if (a->size == 2 && a->align == 3) {
702             return false;
703         }
704         break;
705     default:
706         g_assert_not_reached();
707     }
708     if ((vd + a->stride * (nregs - 1)) > 31) {
709         /*
710          * Attempts to write off the end of the register file are
711          * UNPREDICTABLE; we choose to UNDEF because otherwise we would
712          * access off the end of the array that holds the register data.
713          */
714         return false;
715     }
716 
717     if (!vfp_access_check(s)) {
718         return true;
719     }
720 
721     /* Pick up SCTLR settings */
722     mop = finalize_memop(s, a->size);
723 
724     if (a->align) {
725         MemOp align_op;
726 
727         switch (nregs) {
728         case 1:
729             /* For VLD1, use natural alignment. */
730             align_op = MO_ALIGN;
731             break;
732         case 2:
733             /* For VLD2, use double alignment. */
734             align_op = pow2_align(a->size + 1);
735             break;
736         case 4:
737             if (a->size == MO_32) {
738                 /*
739                  * For VLD4.32, align = 1 is double alignment, align = 2 is
740                  * quad alignment; align = 3 is rejected above.
741                  */
742                 align_op = pow2_align(a->size + a->align);
743             } else {
744                 /* For VLD4.8 and VLD.16, we want quad alignment. */
745                 align_op = pow2_align(a->size + 2);
746             }
747             break;
748         default:
749             /* For VLD3, the alignment field is zero and rejected above. */
750             g_assert_not_reached();
751         }
752 
753         mop = (mop & ~MO_AMASK) | align_op;
754     }
755 
756     tmp = tcg_temp_new_i32();
757     addr = tcg_temp_new_i32();
758     load_reg_var(s, addr, a->rn);
759 
760     for (reg = 0; reg < nregs; reg++) {
761         if (a->l) {
762             gen_aa32_ld_internal_i32(s, tmp, addr, get_mem_index(s), mop);
763             neon_store_element(vd, a->reg_idx, a->size, tmp);
764         } else { /* Store */
765             neon_load_element(tmp, vd, a->reg_idx, a->size);
766             gen_aa32_st_internal_i32(s, tmp, addr, get_mem_index(s), mop);
767         }
768         vd += a->stride;
769         tcg_gen_addi_i32(addr, addr, 1 << a->size);
770 
771         /* Subsequent memory operations inherit alignment */
772         mop &= ~MO_AMASK;
773     }
774 
775     gen_neon_ldst_base_update(s, a->rm, a->rn, (1 << a->size) * nregs);
776 
777     return true;
778 }
779 
780 static bool do_3same(DisasContext *s, arg_3same *a, GVecGen3Fn fn)
781 {
782     int vec_size = a->q ? 16 : 8;
783     int rd_ofs = neon_full_reg_offset(a->vd);
784     int rn_ofs = neon_full_reg_offset(a->vn);
785     int rm_ofs = neon_full_reg_offset(a->vm);
786 
787     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
788         return false;
789     }
790 
791     /* UNDEF accesses to D16-D31 if they don't exist. */
792     if (!dc_isar_feature(aa32_simd_r32, s) &&
793         ((a->vd | a->vn | a->vm) & 0x10)) {
794         return false;
795     }
796 
797     if ((a->vn | a->vm | a->vd) & a->q) {
798         return false;
799     }
800 
801     if (!vfp_access_check(s)) {
802         return true;
803     }
804 
805     fn(a->size, rd_ofs, rn_ofs, rm_ofs, vec_size, vec_size);
806     return true;
807 }
808 
809 #define DO_3SAME(INSN, FUNC)                                            \
810     static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a)        \
811     {                                                                   \
812         return do_3same(s, a, FUNC);                                    \
813     }
814 
815 DO_3SAME(VADD, tcg_gen_gvec_add)
816 DO_3SAME(VSUB, tcg_gen_gvec_sub)
817 DO_3SAME(VAND, tcg_gen_gvec_and)
818 DO_3SAME(VBIC, tcg_gen_gvec_andc)
819 DO_3SAME(VORR, tcg_gen_gvec_or)
820 DO_3SAME(VORN, tcg_gen_gvec_orc)
821 DO_3SAME(VEOR, tcg_gen_gvec_xor)
822 DO_3SAME(VSHL_S, gen_gvec_sshl)
823 DO_3SAME(VSHL_U, gen_gvec_ushl)
824 DO_3SAME(VQADD_S, gen_gvec_sqadd_qc)
825 DO_3SAME(VQADD_U, gen_gvec_uqadd_qc)
826 DO_3SAME(VQSUB_S, gen_gvec_sqsub_qc)
827 DO_3SAME(VQSUB_U, gen_gvec_uqsub_qc)
828 DO_3SAME(VRSHL_S, gen_gvec_srshl)
829 DO_3SAME(VRSHL_U, gen_gvec_urshl)
830 DO_3SAME(VQSHL_S, gen_neon_sqshl)
831 DO_3SAME(VQSHL_U, gen_neon_uqshl)
832 DO_3SAME(VQRSHL_S, gen_neon_sqrshl)
833 DO_3SAME(VQRSHL_U, gen_neon_uqrshl)
834 
835 /* These insns are all gvec_bitsel but with the inputs in various orders. */
836 #define DO_3SAME_BITSEL(INSN, O1, O2, O3)                               \
837     static void gen_##INSN##_3s(unsigned vece, uint32_t rd_ofs,         \
838                                 uint32_t rn_ofs, uint32_t rm_ofs,       \
839                                 uint32_t oprsz, uint32_t maxsz)         \
840     {                                                                   \
841         tcg_gen_gvec_bitsel(vece, rd_ofs, O1, O2, O3, oprsz, maxsz);    \
842     }                                                                   \
843     DO_3SAME(INSN, gen_##INSN##_3s)
844 
845 DO_3SAME_BITSEL(VBSL, rd_ofs, rn_ofs, rm_ofs)
846 DO_3SAME_BITSEL(VBIT, rm_ofs, rn_ofs, rd_ofs)
847 DO_3SAME_BITSEL(VBIF, rm_ofs, rd_ofs, rn_ofs)
848 
849 #define DO_3SAME_NO_SZ_3(INSN, FUNC)                                    \
850     static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a)        \
851     {                                                                   \
852         if (a->size == 3) {                                             \
853             return false;                                               \
854         }                                                               \
855         return do_3same(s, a, FUNC);                                    \
856     }
857 
858 DO_3SAME_NO_SZ_3(VMAX_S, tcg_gen_gvec_smax)
859 DO_3SAME_NO_SZ_3(VMAX_U, tcg_gen_gvec_umax)
860 DO_3SAME_NO_SZ_3(VMIN_S, tcg_gen_gvec_smin)
861 DO_3SAME_NO_SZ_3(VMIN_U, tcg_gen_gvec_umin)
862 DO_3SAME_NO_SZ_3(VMUL, tcg_gen_gvec_mul)
863 DO_3SAME_NO_SZ_3(VMLA, gen_gvec_mla)
864 DO_3SAME_NO_SZ_3(VMLS, gen_gvec_mls)
865 DO_3SAME_NO_SZ_3(VTST, gen_gvec_cmtst)
866 DO_3SAME_NO_SZ_3(VABD_S, gen_gvec_sabd)
867 DO_3SAME_NO_SZ_3(VABA_S, gen_gvec_saba)
868 DO_3SAME_NO_SZ_3(VABD_U, gen_gvec_uabd)
869 DO_3SAME_NO_SZ_3(VABA_U, gen_gvec_uaba)
870 DO_3SAME_NO_SZ_3(VPADD, gen_gvec_addp)
871 DO_3SAME_NO_SZ_3(VPMAX_S, gen_gvec_smaxp)
872 DO_3SAME_NO_SZ_3(VPMIN_S, gen_gvec_sminp)
873 DO_3SAME_NO_SZ_3(VPMAX_U, gen_gvec_umaxp)
874 DO_3SAME_NO_SZ_3(VPMIN_U, gen_gvec_uminp)
875 DO_3SAME_NO_SZ_3(VHADD_S, gen_gvec_shadd)
876 DO_3SAME_NO_SZ_3(VHADD_U, gen_gvec_uhadd)
877 DO_3SAME_NO_SZ_3(VHSUB_S, gen_gvec_shsub)
878 DO_3SAME_NO_SZ_3(VHSUB_U, gen_gvec_uhsub)
879 DO_3SAME_NO_SZ_3(VRHADD_S, gen_gvec_srhadd)
880 DO_3SAME_NO_SZ_3(VRHADD_U, gen_gvec_urhadd)
881 
882 #define DO_3SAME_CMP(INSN, COND)                                        \
883     static void gen_##INSN##_3s(unsigned vece, uint32_t rd_ofs,         \
884                                 uint32_t rn_ofs, uint32_t rm_ofs,       \
885                                 uint32_t oprsz, uint32_t maxsz)         \
886     {                                                                   \
887         tcg_gen_gvec_cmp(COND, vece, rd_ofs, rn_ofs, rm_ofs, oprsz, maxsz); \
888     }                                                                   \
889     DO_3SAME_NO_SZ_3(INSN, gen_##INSN##_3s)
890 
891 DO_3SAME_CMP(VCGT_S, TCG_COND_GT)
892 DO_3SAME_CMP(VCGT_U, TCG_COND_GTU)
893 DO_3SAME_CMP(VCGE_S, TCG_COND_GE)
894 DO_3SAME_CMP(VCGE_U, TCG_COND_GEU)
895 DO_3SAME_CMP(VCEQ, TCG_COND_EQ)
896 
897 #define WRAP_OOL_FN(WRAPNAME, FUNC)                                        \
898     static void WRAPNAME(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,  \
899                          uint32_t rm_ofs, uint32_t oprsz, uint32_t maxsz)  \
900     {                                                                      \
901         tcg_gen_gvec_3_ool(rd_ofs, rn_ofs, rm_ofs, oprsz, maxsz, 0, FUNC); \
902     }
903 
904 WRAP_OOL_FN(gen_VMUL_p_3s, gen_helper_gvec_pmul_b)
905 
906 static bool trans_VMUL_p_3s(DisasContext *s, arg_3same *a)
907 {
908     if (a->size != 0) {
909         return false;
910     }
911     return do_3same(s, a, gen_VMUL_p_3s);
912 }
913 
914 #define DO_VQRDMLAH(INSN, FUNC)                                         \
915     static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a)        \
916     {                                                                   \
917         if (!dc_isar_feature(aa32_rdm, s)) {                            \
918             return false;                                               \
919         }                                                               \
920         if (a->size != 1 && a->size != 2) {                             \
921             return false;                                               \
922         }                                                               \
923         return do_3same(s, a, FUNC);                                    \
924     }
925 
926 DO_VQRDMLAH(VQRDMLAH, gen_gvec_sqrdmlah_qc)
927 DO_VQRDMLAH(VQRDMLSH, gen_gvec_sqrdmlsh_qc)
928 
929 #define DO_SHA1(NAME, FUNC)                                             \
930     WRAP_OOL_FN(gen_##NAME##_3s, FUNC)                                  \
931     static bool trans_##NAME##_3s(DisasContext *s, arg_3same *a)        \
932     {                                                                   \
933         if (!dc_isar_feature(aa32_sha1, s)) {                           \
934             return false;                                               \
935         }                                                               \
936         return do_3same(s, a, gen_##NAME##_3s);                         \
937     }
938 
939 DO_SHA1(SHA1C, gen_helper_crypto_sha1c)
940 DO_SHA1(SHA1P, gen_helper_crypto_sha1p)
941 DO_SHA1(SHA1M, gen_helper_crypto_sha1m)
942 DO_SHA1(SHA1SU0, gen_helper_crypto_sha1su0)
943 
944 #define DO_SHA2(NAME, FUNC)                                             \
945     WRAP_OOL_FN(gen_##NAME##_3s, FUNC)                                  \
946     static bool trans_##NAME##_3s(DisasContext *s, arg_3same *a)        \
947     {                                                                   \
948         if (!dc_isar_feature(aa32_sha2, s)) {                           \
949             return false;                                               \
950         }                                                               \
951         return do_3same(s, a, gen_##NAME##_3s);                         \
952     }
953 
954 DO_SHA2(SHA256H, gen_helper_crypto_sha256h)
955 DO_SHA2(SHA256H2, gen_helper_crypto_sha256h2)
956 DO_SHA2(SHA256SU1, gen_helper_crypto_sha256su1)
957 
958 /*
959  * Some helper functions need to be passed the tcg_env. In order
960  * to use those with the gvec APIs like tcg_gen_gvec_3() we need
961  * to create wrapper functions whose prototype is a NeonGenTwoOpFn()
962  * and which call a NeonGenTwoOpEnvFn().
963  */
964 #define WRAP_ENV_FN(WRAPNAME, FUNC)                                     \
965     static void WRAPNAME(TCGv_i32 d, TCGv_i32 n, TCGv_i32 m)            \
966     {                                                                   \
967         FUNC(d, tcg_env, n, m);                                         \
968     }
969 
970 #define DO_3SAME_VQDMULH(INSN, FUNC)                                    \
971     static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a)        \
972     { return a->size >= 1 && a->size <= 2 && do_3same(s, a, FUNC); }
973 
974 DO_3SAME_VQDMULH(VQDMULH, gen_gvec_sqdmulh_qc)
975 DO_3SAME_VQDMULH(VQRDMULH, gen_gvec_sqrdmulh_qc)
976 
977 #define WRAP_FP_GVEC(WRAPNAME, FPST, FUNC)                              \
978     static void WRAPNAME(unsigned vece, uint32_t rd_ofs,                \
979                          uint32_t rn_ofs, uint32_t rm_ofs,              \
980                          uint32_t oprsz, uint32_t maxsz)                \
981     {                                                                   \
982         TCGv_ptr fpst = fpstatus_ptr(FPST);                             \
983         tcg_gen_gvec_3_ptr(rd_ofs, rn_ofs, rm_ofs, fpst,                \
984                            oprsz, maxsz, 0, FUNC);                      \
985     }
986 
987 #define DO_3S_FP_GVEC(INSN,SFUNC,HFUNC)                                 \
988     WRAP_FP_GVEC(gen_##INSN##_fp32_3s, FPST_STD, SFUNC)                 \
989     WRAP_FP_GVEC(gen_##INSN##_fp16_3s, FPST_STD_F16, HFUNC)             \
990     static bool trans_##INSN##_fp_3s(DisasContext *s, arg_3same *a)     \
991     {                                                                   \
992         if (a->size == MO_16) {                                         \
993             if (!dc_isar_feature(aa32_fp16_arith, s)) {                 \
994                 return false;                                           \
995             }                                                           \
996             return do_3same(s, a, gen_##INSN##_fp16_3s);                \
997         }                                                               \
998         return do_3same(s, a, gen_##INSN##_fp32_3s);                    \
999     }
1000 
1001 
1002 DO_3S_FP_GVEC(VADD, gen_helper_gvec_fadd_s, gen_helper_gvec_fadd_h)
1003 DO_3S_FP_GVEC(VSUB, gen_helper_gvec_fsub_s, gen_helper_gvec_fsub_h)
1004 DO_3S_FP_GVEC(VABD, gen_helper_gvec_fabd_s, gen_helper_gvec_fabd_h)
1005 DO_3S_FP_GVEC(VMUL, gen_helper_gvec_fmul_s, gen_helper_gvec_fmul_h)
1006 DO_3S_FP_GVEC(VCEQ, gen_helper_gvec_fceq_s, gen_helper_gvec_fceq_h)
1007 DO_3S_FP_GVEC(VCGE, gen_helper_gvec_fcge_s, gen_helper_gvec_fcge_h)
1008 DO_3S_FP_GVEC(VCGT, gen_helper_gvec_fcgt_s, gen_helper_gvec_fcgt_h)
1009 DO_3S_FP_GVEC(VACGE, gen_helper_gvec_facge_s, gen_helper_gvec_facge_h)
1010 DO_3S_FP_GVEC(VACGT, gen_helper_gvec_facgt_s, gen_helper_gvec_facgt_h)
1011 DO_3S_FP_GVEC(VMAX, gen_helper_gvec_fmax_s, gen_helper_gvec_fmax_h)
1012 DO_3S_FP_GVEC(VMIN, gen_helper_gvec_fmin_s, gen_helper_gvec_fmin_h)
1013 DO_3S_FP_GVEC(VMLA, gen_helper_gvec_fmla_s, gen_helper_gvec_fmla_h)
1014 DO_3S_FP_GVEC(VMLS, gen_helper_gvec_fmls_s, gen_helper_gvec_fmls_h)
1015 DO_3S_FP_GVEC(VFMA, gen_helper_gvec_vfma_s, gen_helper_gvec_vfma_h)
1016 DO_3S_FP_GVEC(VFMS, gen_helper_gvec_vfms_s, gen_helper_gvec_vfms_h)
1017 DO_3S_FP_GVEC(VRECPS, gen_helper_gvec_recps_nf_s, gen_helper_gvec_recps_nf_h)
1018 DO_3S_FP_GVEC(VRSQRTS, gen_helper_gvec_rsqrts_nf_s, gen_helper_gvec_rsqrts_nf_h)
1019 DO_3S_FP_GVEC(VPADD, gen_helper_gvec_faddp_s, gen_helper_gvec_faddp_h)
1020 DO_3S_FP_GVEC(VPMAX, gen_helper_gvec_fmaxp_s, gen_helper_gvec_fmaxp_h)
1021 DO_3S_FP_GVEC(VPMIN, gen_helper_gvec_fminp_s, gen_helper_gvec_fminp_h)
1022 
1023 WRAP_FP_GVEC(gen_VMAXNM_fp32_3s, FPST_STD, gen_helper_gvec_fmaxnum_s)
1024 WRAP_FP_GVEC(gen_VMAXNM_fp16_3s, FPST_STD_F16, gen_helper_gvec_fmaxnum_h)
1025 WRAP_FP_GVEC(gen_VMINNM_fp32_3s, FPST_STD, gen_helper_gvec_fminnum_s)
1026 WRAP_FP_GVEC(gen_VMINNM_fp16_3s, FPST_STD_F16, gen_helper_gvec_fminnum_h)
1027 
1028 static bool trans_VMAXNM_fp_3s(DisasContext *s, arg_3same *a)
1029 {
1030     if (!arm_dc_feature(s, ARM_FEATURE_V8)) {
1031         return false;
1032     }
1033 
1034     if (a->size == MO_16) {
1035         if (!dc_isar_feature(aa32_fp16_arith, s)) {
1036             return false;
1037         }
1038         return do_3same(s, a, gen_VMAXNM_fp16_3s);
1039     }
1040     return do_3same(s, a, gen_VMAXNM_fp32_3s);
1041 }
1042 
1043 static bool trans_VMINNM_fp_3s(DisasContext *s, arg_3same *a)
1044 {
1045     if (!arm_dc_feature(s, ARM_FEATURE_V8)) {
1046         return false;
1047     }
1048 
1049     if (a->size == MO_16) {
1050         if (!dc_isar_feature(aa32_fp16_arith, s)) {
1051             return false;
1052         }
1053         return do_3same(s, a, gen_VMINNM_fp16_3s);
1054     }
1055     return do_3same(s, a, gen_VMINNM_fp32_3s);
1056 }
1057 
1058 static bool do_vector_2sh(DisasContext *s, arg_2reg_shift *a, GVecGen2iFn *fn)
1059 {
1060     /* Handle a 2-reg-shift insn which can be vectorized. */
1061     int vec_size = a->q ? 16 : 8;
1062     int rd_ofs = neon_full_reg_offset(a->vd);
1063     int rm_ofs = neon_full_reg_offset(a->vm);
1064 
1065     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1066         return false;
1067     }
1068 
1069     /* UNDEF accesses to D16-D31 if they don't exist. */
1070     if (!dc_isar_feature(aa32_simd_r32, s) &&
1071         ((a->vd | a->vm) & 0x10)) {
1072         return false;
1073     }
1074 
1075     if ((a->vm | a->vd) & a->q) {
1076         return false;
1077     }
1078 
1079     if (!vfp_access_check(s)) {
1080         return true;
1081     }
1082 
1083     fn(a->size, rd_ofs, rm_ofs, a->shift, vec_size, vec_size);
1084     return true;
1085 }
1086 
1087 #define DO_2SH(INSN, FUNC)                                              \
1088     static bool trans_##INSN##_2sh(DisasContext *s, arg_2reg_shift *a)  \
1089     {                                                                   \
1090         return do_vector_2sh(s, a, FUNC);                               \
1091     }                                                                   \
1092 
1093 DO_2SH(VSHL, tcg_gen_gvec_shli)
1094 DO_2SH(VSLI, gen_gvec_sli)
1095 DO_2SH(VSRI, gen_gvec_sri)
1096 DO_2SH(VSRA_S, gen_gvec_ssra)
1097 DO_2SH(VSRA_U, gen_gvec_usra)
1098 DO_2SH(VRSHR_S, gen_gvec_srshr)
1099 DO_2SH(VRSHR_U, gen_gvec_urshr)
1100 DO_2SH(VRSRA_S, gen_gvec_srsra)
1101 DO_2SH(VRSRA_U, gen_gvec_ursra)
1102 DO_2SH(VSHR_S, gen_gvec_sshr)
1103 DO_2SH(VSHR_U, gen_gvec_ushr)
1104 DO_2SH(VQSHLU, gen_neon_sqshlui)
1105 DO_2SH(VQSHL_U, gen_neon_uqshli)
1106 DO_2SH(VQSHL_S, gen_neon_sqshli)
1107 
1108 static bool do_2shift_narrow_64(DisasContext *s, arg_2reg_shift *a,
1109                                 NeonGenTwo64OpFn *shiftfn,
1110                                 NeonGenOne64OpEnvFn *narrowfn)
1111 {
1112     /* 2-reg-and-shift narrowing-shift operations, size == 3 case */
1113     TCGv_i64 constimm, rm1, rm2, rd;
1114 
1115     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1116         return false;
1117     }
1118 
1119     /* UNDEF accesses to D16-D31 if they don't exist. */
1120     if (!dc_isar_feature(aa32_simd_r32, s) &&
1121         ((a->vd | a->vm) & 0x10)) {
1122         return false;
1123     }
1124 
1125     if (a->vm & 1) {
1126         return false;
1127     }
1128 
1129     if (!vfp_access_check(s)) {
1130         return true;
1131     }
1132 
1133     /*
1134      * This is always a right shift, and the shiftfn is always a
1135      * left-shift helper, which thus needs the negated shift count.
1136      */
1137     constimm = tcg_constant_i64(-a->shift);
1138     rm1 = tcg_temp_new_i64();
1139     rm2 = tcg_temp_new_i64();
1140     rd = tcg_temp_new_i64();
1141 
1142     /* Load both inputs first to avoid potential overwrite if rm == rd */
1143     read_neon_element64(rm1, a->vm, 0, MO_64);
1144     read_neon_element64(rm2, a->vm, 1, MO_64);
1145 
1146     shiftfn(rm1, rm1, constimm);
1147     narrowfn(rd, tcg_env, rm1);
1148     write_neon_element64(rd, a->vd, 0, MO_32);
1149 
1150     shiftfn(rm2, rm2, constimm);
1151     narrowfn(rd, tcg_env, rm2);
1152     write_neon_element64(rd, a->vd, 1, MO_32);
1153 
1154     return true;
1155 }
1156 
1157 static bool do_2shift_narrow_32(DisasContext *s, arg_2reg_shift *a,
1158                                 NeonGenTwoOpFn *shiftfn,
1159                                 NeonGenOne64OpEnvFn *narrowfn)
1160 {
1161     /* 2-reg-and-shift narrowing-shift operations, size < 3 case */
1162     TCGv_i32 constimm, rm1, rm2, rm3, rm4;
1163     TCGv_i64 rtmp;
1164     uint32_t imm;
1165 
1166     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1167         return false;
1168     }
1169 
1170     /* UNDEF accesses to D16-D31 if they don't exist. */
1171     if (!dc_isar_feature(aa32_simd_r32, s) &&
1172         ((a->vd | a->vm) & 0x10)) {
1173         return false;
1174     }
1175 
1176     if (a->vm & 1) {
1177         return false;
1178     }
1179 
1180     if (!vfp_access_check(s)) {
1181         return true;
1182     }
1183 
1184     /*
1185      * This is always a right shift, and the shiftfn is always a
1186      * left-shift helper, which thus needs the negated shift count
1187      * duplicated into each lane of the immediate value.
1188      */
1189     if (a->size == 1) {
1190         imm = (uint16_t)(-a->shift);
1191         imm |= imm << 16;
1192     } else {
1193         /* size == 2 */
1194         imm = -a->shift;
1195     }
1196     constimm = tcg_constant_i32(imm);
1197 
1198     /* Load all inputs first to avoid potential overwrite */
1199     rm1 = tcg_temp_new_i32();
1200     rm2 = tcg_temp_new_i32();
1201     rm3 = tcg_temp_new_i32();
1202     rm4 = tcg_temp_new_i32();
1203     read_neon_element32(rm1, a->vm, 0, MO_32);
1204     read_neon_element32(rm2, a->vm, 1, MO_32);
1205     read_neon_element32(rm3, a->vm, 2, MO_32);
1206     read_neon_element32(rm4, a->vm, 3, MO_32);
1207     rtmp = tcg_temp_new_i64();
1208 
1209     shiftfn(rm1, rm1, constimm);
1210     shiftfn(rm2, rm2, constimm);
1211 
1212     tcg_gen_concat_i32_i64(rtmp, rm1, rm2);
1213 
1214     narrowfn(rtmp, tcg_env, rtmp);
1215     write_neon_element64(rtmp, a->vd, 0, MO_32);
1216 
1217     shiftfn(rm3, rm3, constimm);
1218     shiftfn(rm4, rm4, constimm);
1219 
1220     tcg_gen_concat_i32_i64(rtmp, rm3, rm4);
1221 
1222     narrowfn(rtmp, tcg_env, rtmp);
1223     write_neon_element64(rtmp, a->vd, 1, MO_32);
1224     return true;
1225 }
1226 
1227 #define DO_2SN_64(INSN, FUNC, NARROWFUNC)                               \
1228     static bool trans_##INSN##_2sh(DisasContext *s, arg_2reg_shift *a)  \
1229     {                                                                   \
1230         return do_2shift_narrow_64(s, a, FUNC, NARROWFUNC);             \
1231     }
1232 #define DO_2SN_32(INSN, FUNC, NARROWFUNC)                               \
1233     static bool trans_##INSN##_2sh(DisasContext *s, arg_2reg_shift *a)  \
1234     {                                                                   \
1235         return do_2shift_narrow_32(s, a, FUNC, NARROWFUNC);             \
1236     }
1237 
1238 static void gen_neon_narrow_u32(TCGv_i64 dest, TCGv_ptr env, TCGv_i64 src)
1239 {
1240     tcg_gen_ext32u_i64(dest, src);
1241 }
1242 
1243 static void gen_neon_narrow_u16(TCGv_i64 dest, TCGv_ptr env, TCGv_i64 src)
1244 {
1245     gen_helper_neon_narrow_u16(dest, src);
1246 }
1247 
1248 static void gen_neon_narrow_u8(TCGv_i64 dest, TCGv_ptr env, TCGv_i64 src)
1249 {
1250     gen_helper_neon_narrow_u8(dest, src);
1251 }
1252 
1253 DO_2SN_64(VSHRN_64, gen_ushl_i64, gen_neon_narrow_u32)
1254 DO_2SN_32(VSHRN_32, gen_ushl_i32, gen_neon_narrow_u16)
1255 DO_2SN_32(VSHRN_16, gen_helper_neon_shl_u16, gen_neon_narrow_u8)
1256 
1257 DO_2SN_64(VRSHRN_64, gen_helper_neon_rshl_u64, gen_neon_narrow_u32)
1258 DO_2SN_32(VRSHRN_32, gen_helper_neon_rshl_u32, gen_neon_narrow_u16)
1259 DO_2SN_32(VRSHRN_16, gen_helper_neon_rshl_u16, gen_neon_narrow_u8)
1260 
1261 DO_2SN_64(VQSHRUN_64, gen_sshl_i64, gen_helper_neon_unarrow_sat32)
1262 DO_2SN_32(VQSHRUN_32, gen_sshl_i32, gen_helper_neon_unarrow_sat16)
1263 DO_2SN_32(VQSHRUN_16, gen_helper_neon_shl_s16, gen_helper_neon_unarrow_sat8)
1264 
1265 DO_2SN_64(VQRSHRUN_64, gen_helper_neon_rshl_s64, gen_helper_neon_unarrow_sat32)
1266 DO_2SN_32(VQRSHRUN_32, gen_helper_neon_rshl_s32, gen_helper_neon_unarrow_sat16)
1267 DO_2SN_32(VQRSHRUN_16, gen_helper_neon_rshl_s16, gen_helper_neon_unarrow_sat8)
1268 DO_2SN_64(VQSHRN_S64, gen_sshl_i64, gen_helper_neon_narrow_sat_s32)
1269 DO_2SN_32(VQSHRN_S32, gen_sshl_i32, gen_helper_neon_narrow_sat_s16)
1270 DO_2SN_32(VQSHRN_S16, gen_helper_neon_shl_s16, gen_helper_neon_narrow_sat_s8)
1271 
1272 DO_2SN_64(VQRSHRN_S64, gen_helper_neon_rshl_s64, gen_helper_neon_narrow_sat_s32)
1273 DO_2SN_32(VQRSHRN_S32, gen_helper_neon_rshl_s32, gen_helper_neon_narrow_sat_s16)
1274 DO_2SN_32(VQRSHRN_S16, gen_helper_neon_rshl_s16, gen_helper_neon_narrow_sat_s8)
1275 
1276 DO_2SN_64(VQSHRN_U64, gen_ushl_i64, gen_helper_neon_narrow_sat_u32)
1277 DO_2SN_32(VQSHRN_U32, gen_ushl_i32, gen_helper_neon_narrow_sat_u16)
1278 DO_2SN_32(VQSHRN_U16, gen_helper_neon_shl_u16, gen_helper_neon_narrow_sat_u8)
1279 
1280 DO_2SN_64(VQRSHRN_U64, gen_helper_neon_rshl_u64, gen_helper_neon_narrow_sat_u32)
1281 DO_2SN_32(VQRSHRN_U32, gen_helper_neon_rshl_u32, gen_helper_neon_narrow_sat_u16)
1282 DO_2SN_32(VQRSHRN_U16, gen_helper_neon_rshl_u16, gen_helper_neon_narrow_sat_u8)
1283 
1284 static bool do_vshll_2sh(DisasContext *s, arg_2reg_shift *a,
1285                          NeonGenWidenFn *widenfn, bool u)
1286 {
1287     TCGv_i64 tmp;
1288     TCGv_i32 rm0, rm1;
1289     uint64_t widen_mask = 0;
1290 
1291     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1292         return false;
1293     }
1294 
1295     /* UNDEF accesses to D16-D31 if they don't exist. */
1296     if (!dc_isar_feature(aa32_simd_r32, s) &&
1297         ((a->vd | a->vm) & 0x10)) {
1298         return false;
1299     }
1300 
1301     if (a->vd & 1) {
1302         return false;
1303     }
1304 
1305     if (!vfp_access_check(s)) {
1306         return true;
1307     }
1308 
1309     /*
1310      * This is a widen-and-shift operation. The shift is always less
1311      * than the width of the source type, so after widening the input
1312      * vector we can simply shift the whole 64-bit widened register,
1313      * and then clear the potential overflow bits resulting from left
1314      * bits of the narrow input appearing as right bits of the left
1315      * neighbour narrow input. Calculate a mask of bits to clear.
1316      */
1317     if ((a->shift != 0) && (a->size < 2 || u)) {
1318         int esize = 8 << a->size;
1319         widen_mask = MAKE_64BIT_MASK(0, esize);
1320         widen_mask >>= esize - a->shift;
1321         widen_mask = dup_const(a->size + 1, widen_mask);
1322     }
1323 
1324     rm0 = tcg_temp_new_i32();
1325     rm1 = tcg_temp_new_i32();
1326     read_neon_element32(rm0, a->vm, 0, MO_32);
1327     read_neon_element32(rm1, a->vm, 1, MO_32);
1328     tmp = tcg_temp_new_i64();
1329 
1330     widenfn(tmp, rm0);
1331     if (a->shift != 0) {
1332         tcg_gen_shli_i64(tmp, tmp, a->shift);
1333         tcg_gen_andi_i64(tmp, tmp, ~widen_mask);
1334     }
1335     write_neon_element64(tmp, a->vd, 0, MO_64);
1336 
1337     widenfn(tmp, rm1);
1338     if (a->shift != 0) {
1339         tcg_gen_shli_i64(tmp, tmp, a->shift);
1340         tcg_gen_andi_i64(tmp, tmp, ~widen_mask);
1341     }
1342     write_neon_element64(tmp, a->vd, 1, MO_64);
1343     return true;
1344 }
1345 
1346 static bool trans_VSHLL_S_2sh(DisasContext *s, arg_2reg_shift *a)
1347 {
1348     static NeonGenWidenFn * const widenfn[] = {
1349         gen_helper_neon_widen_s8,
1350         gen_helper_neon_widen_s16,
1351         tcg_gen_ext_i32_i64,
1352     };
1353     return do_vshll_2sh(s, a, widenfn[a->size], false);
1354 }
1355 
1356 static bool trans_VSHLL_U_2sh(DisasContext *s, arg_2reg_shift *a)
1357 {
1358     static NeonGenWidenFn * const widenfn[] = {
1359         gen_helper_neon_widen_u8,
1360         gen_helper_neon_widen_u16,
1361         tcg_gen_extu_i32_i64,
1362     };
1363     return do_vshll_2sh(s, a, widenfn[a->size], true);
1364 }
1365 
1366 static bool do_fp_2sh(DisasContext *s, arg_2reg_shift *a,
1367                       gen_helper_gvec_2_ptr *fn)
1368 {
1369     /* FP operations in 2-reg-and-shift group */
1370     int vec_size = a->q ? 16 : 8;
1371     int rd_ofs = neon_full_reg_offset(a->vd);
1372     int rm_ofs = neon_full_reg_offset(a->vm);
1373     TCGv_ptr fpst;
1374 
1375     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1376         return false;
1377     }
1378 
1379     if (a->size == MO_16) {
1380         if (!dc_isar_feature(aa32_fp16_arith, s)) {
1381             return false;
1382         }
1383     }
1384 
1385     /* UNDEF accesses to D16-D31 if they don't exist. */
1386     if (!dc_isar_feature(aa32_simd_r32, s) &&
1387         ((a->vd | a->vm) & 0x10)) {
1388         return false;
1389     }
1390 
1391     if ((a->vm | a->vd) & a->q) {
1392         return false;
1393     }
1394 
1395     if (!vfp_access_check(s)) {
1396         return true;
1397     }
1398 
1399     fpst = fpstatus_ptr(a->size == MO_16 ? FPST_STD_F16 : FPST_STD);
1400     tcg_gen_gvec_2_ptr(rd_ofs, rm_ofs, fpst, vec_size, vec_size, a->shift, fn);
1401     return true;
1402 }
1403 
1404 #define DO_FP_2SH(INSN, FUNC)                                           \
1405     static bool trans_##INSN##_2sh(DisasContext *s, arg_2reg_shift *a)  \
1406     {                                                                   \
1407         return do_fp_2sh(s, a, FUNC);                                   \
1408     }
1409 
1410 DO_FP_2SH(VCVT_SF, gen_helper_gvec_vcvt_sf)
1411 DO_FP_2SH(VCVT_UF, gen_helper_gvec_vcvt_uf)
1412 DO_FP_2SH(VCVT_FS, gen_helper_gvec_vcvt_fs)
1413 DO_FP_2SH(VCVT_FU, gen_helper_gvec_vcvt_fu)
1414 
1415 DO_FP_2SH(VCVT_SH, gen_helper_gvec_vcvt_sh)
1416 DO_FP_2SH(VCVT_UH, gen_helper_gvec_vcvt_uh)
1417 DO_FP_2SH(VCVT_HS, gen_helper_gvec_vcvt_hs)
1418 DO_FP_2SH(VCVT_HU, gen_helper_gvec_vcvt_hu)
1419 
1420 static bool do_1reg_imm(DisasContext *s, arg_1reg_imm *a,
1421                         GVecGen2iFn *fn)
1422 {
1423     uint64_t imm;
1424     int reg_ofs, vec_size;
1425 
1426     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1427         return false;
1428     }
1429 
1430     /* UNDEF accesses to D16-D31 if they don't exist. */
1431     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vd & 0x10)) {
1432         return false;
1433     }
1434 
1435     if (a->vd & a->q) {
1436         return false;
1437     }
1438 
1439     if (!vfp_access_check(s)) {
1440         return true;
1441     }
1442 
1443     reg_ofs = neon_full_reg_offset(a->vd);
1444     vec_size = a->q ? 16 : 8;
1445     imm = asimd_imm_const(a->imm, a->cmode, a->op);
1446 
1447     fn(MO_64, reg_ofs, reg_ofs, imm, vec_size, vec_size);
1448     return true;
1449 }
1450 
1451 static void gen_VMOV_1r(unsigned vece, uint32_t dofs, uint32_t aofs,
1452                         int64_t c, uint32_t oprsz, uint32_t maxsz)
1453 {
1454     tcg_gen_gvec_dup_imm(MO_64, dofs, oprsz, maxsz, c);
1455 }
1456 
1457 static bool trans_Vimm_1r(DisasContext *s, arg_1reg_imm *a)
1458 {
1459     /* Handle decode of cmode/op here between VORR/VBIC/VMOV */
1460     GVecGen2iFn *fn;
1461 
1462     if ((a->cmode & 1) && a->cmode < 12) {
1463         /* for op=1, the imm will be inverted, so BIC becomes AND. */
1464         fn = a->op ? tcg_gen_gvec_andi : tcg_gen_gvec_ori;
1465     } else {
1466         /* There is one unallocated cmode/op combination in this space */
1467         if (a->cmode == 15 && a->op == 1) {
1468             return false;
1469         }
1470         fn = gen_VMOV_1r;
1471     }
1472     return do_1reg_imm(s, a, fn);
1473 }
1474 
1475 static bool do_prewiden_3d(DisasContext *s, arg_3diff *a,
1476                            NeonGenWidenFn *widenfn,
1477                            NeonGenTwo64OpFn *opfn,
1478                            int src1_mop, int src2_mop)
1479 {
1480     /* 3-regs different lengths, prewidening case (VADDL/VSUBL/VAADW/VSUBW) */
1481     TCGv_i64 rn0_64, rn1_64, rm_64;
1482 
1483     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1484         return false;
1485     }
1486 
1487     /* UNDEF accesses to D16-D31 if they don't exist. */
1488     if (!dc_isar_feature(aa32_simd_r32, s) &&
1489         ((a->vd | a->vn | a->vm) & 0x10)) {
1490         return false;
1491     }
1492 
1493     if (!opfn) {
1494         /* size == 3 case, which is an entirely different insn group */
1495         return false;
1496     }
1497 
1498     if ((a->vd & 1) || (src1_mop == MO_UQ && (a->vn & 1))) {
1499         return false;
1500     }
1501 
1502     if (!vfp_access_check(s)) {
1503         return true;
1504     }
1505 
1506     rn0_64 = tcg_temp_new_i64();
1507     rn1_64 = tcg_temp_new_i64();
1508     rm_64 = tcg_temp_new_i64();
1509 
1510     if (src1_mop >= 0) {
1511         read_neon_element64(rn0_64, a->vn, 0, src1_mop);
1512     } else {
1513         TCGv_i32 tmp = tcg_temp_new_i32();
1514         read_neon_element32(tmp, a->vn, 0, MO_32);
1515         widenfn(rn0_64, tmp);
1516     }
1517     if (src2_mop >= 0) {
1518         read_neon_element64(rm_64, a->vm, 0, src2_mop);
1519     } else {
1520         TCGv_i32 tmp = tcg_temp_new_i32();
1521         read_neon_element32(tmp, a->vm, 0, MO_32);
1522         widenfn(rm_64, tmp);
1523     }
1524 
1525     opfn(rn0_64, rn0_64, rm_64);
1526 
1527     /*
1528      * Load second pass inputs before storing the first pass result, to
1529      * avoid incorrect results if a narrow input overlaps with the result.
1530      */
1531     if (src1_mop >= 0) {
1532         read_neon_element64(rn1_64, a->vn, 1, src1_mop);
1533     } else {
1534         TCGv_i32 tmp = tcg_temp_new_i32();
1535         read_neon_element32(tmp, a->vn, 1, MO_32);
1536         widenfn(rn1_64, tmp);
1537     }
1538     if (src2_mop >= 0) {
1539         read_neon_element64(rm_64, a->vm, 1, src2_mop);
1540     } else {
1541         TCGv_i32 tmp = tcg_temp_new_i32();
1542         read_neon_element32(tmp, a->vm, 1, MO_32);
1543         widenfn(rm_64, tmp);
1544     }
1545 
1546     write_neon_element64(rn0_64, a->vd, 0, MO_64);
1547 
1548     opfn(rn1_64, rn1_64, rm_64);
1549     write_neon_element64(rn1_64, a->vd, 1, MO_64);
1550 
1551     return true;
1552 }
1553 
1554 #define DO_PREWIDEN(INSN, S, OP, SRC1WIDE, SIGN)                        \
1555     static bool trans_##INSN##_3d(DisasContext *s, arg_3diff *a)        \
1556     {                                                                   \
1557         static NeonGenWidenFn * const widenfn[] = {                     \
1558             gen_helper_neon_widen_##S##8,                               \
1559             gen_helper_neon_widen_##S##16,                              \
1560             NULL, NULL,                                                 \
1561         };                                                              \
1562         static NeonGenTwo64OpFn * const addfn[] = {                     \
1563             gen_helper_neon_##OP##l_u16,                                \
1564             gen_helper_neon_##OP##l_u32,                                \
1565             tcg_gen_##OP##_i64,                                         \
1566             NULL,                                                       \
1567         };                                                              \
1568         int narrow_mop = a->size == MO_32 ? MO_32 | SIGN : -1;          \
1569         return do_prewiden_3d(s, a, widenfn[a->size], addfn[a->size],   \
1570                               SRC1WIDE ? MO_UQ : narrow_mop,             \
1571                               narrow_mop);                              \
1572     }
1573 
1574 DO_PREWIDEN(VADDL_S, s, add, false, MO_SIGN)
1575 DO_PREWIDEN(VADDL_U, u, add, false, 0)
1576 DO_PREWIDEN(VSUBL_S, s, sub, false, MO_SIGN)
1577 DO_PREWIDEN(VSUBL_U, u, sub, false, 0)
1578 DO_PREWIDEN(VADDW_S, s, add, true, MO_SIGN)
1579 DO_PREWIDEN(VADDW_U, u, add, true, 0)
1580 DO_PREWIDEN(VSUBW_S, s, sub, true, MO_SIGN)
1581 DO_PREWIDEN(VSUBW_U, u, sub, true, 0)
1582 
1583 static bool do_narrow_3d(DisasContext *s, arg_3diff *a,
1584                          NeonGenTwo64OpFn *opfn, NeonGenNarrowFn *narrowfn)
1585 {
1586     /* 3-regs different lengths, narrowing (VADDHN/VSUBHN/VRADDHN/VRSUBHN) */
1587     TCGv_i64 rn_64, rm_64;
1588     TCGv_i32 rd0, rd1;
1589 
1590     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1591         return false;
1592     }
1593 
1594     /* UNDEF accesses to D16-D31 if they don't exist. */
1595     if (!dc_isar_feature(aa32_simd_r32, s) &&
1596         ((a->vd | a->vn | a->vm) & 0x10)) {
1597         return false;
1598     }
1599 
1600     if (!opfn || !narrowfn) {
1601         /* size == 3 case, which is an entirely different insn group */
1602         return false;
1603     }
1604 
1605     if ((a->vn | a->vm) & 1) {
1606         return false;
1607     }
1608 
1609     if (!vfp_access_check(s)) {
1610         return true;
1611     }
1612 
1613     rn_64 = tcg_temp_new_i64();
1614     rm_64 = tcg_temp_new_i64();
1615     rd0 = tcg_temp_new_i32();
1616     rd1 = tcg_temp_new_i32();
1617 
1618     read_neon_element64(rn_64, a->vn, 0, MO_64);
1619     read_neon_element64(rm_64, a->vm, 0, MO_64);
1620 
1621     opfn(rn_64, rn_64, rm_64);
1622 
1623     narrowfn(rd0, rn_64);
1624 
1625     read_neon_element64(rn_64, a->vn, 1, MO_64);
1626     read_neon_element64(rm_64, a->vm, 1, MO_64);
1627 
1628     opfn(rn_64, rn_64, rm_64);
1629 
1630     narrowfn(rd1, rn_64);
1631 
1632     write_neon_element32(rd0, a->vd, 0, MO_32);
1633     write_neon_element32(rd1, a->vd, 1, MO_32);
1634 
1635     return true;
1636 }
1637 
1638 #define DO_NARROW_3D(INSN, OP, NARROWTYPE, EXTOP)                       \
1639     static bool trans_##INSN##_3d(DisasContext *s, arg_3diff *a)        \
1640     {                                                                   \
1641         static NeonGenTwo64OpFn * const addfn[] = {                     \
1642             gen_helper_neon_##OP##l_u16,                                \
1643             gen_helper_neon_##OP##l_u32,                                \
1644             tcg_gen_##OP##_i64,                                         \
1645             NULL,                                                       \
1646         };                                                              \
1647         static NeonGenNarrowFn * const narrowfn[] = {                   \
1648             gen_helper_neon_##NARROWTYPE##_high_u8,                     \
1649             gen_helper_neon_##NARROWTYPE##_high_u16,                    \
1650             EXTOP,                                                      \
1651             NULL,                                                       \
1652         };                                                              \
1653         return do_narrow_3d(s, a, addfn[a->size], narrowfn[a->size]);   \
1654     }
1655 
1656 static void gen_narrow_round_high_u32(TCGv_i32 rd, TCGv_i64 rn)
1657 {
1658     tcg_gen_addi_i64(rn, rn, 1u << 31);
1659     tcg_gen_extrh_i64_i32(rd, rn);
1660 }
1661 
1662 DO_NARROW_3D(VADDHN, add, narrow, tcg_gen_extrh_i64_i32)
1663 DO_NARROW_3D(VSUBHN, sub, narrow, tcg_gen_extrh_i64_i32)
1664 DO_NARROW_3D(VRADDHN, add, narrow_round, gen_narrow_round_high_u32)
1665 DO_NARROW_3D(VRSUBHN, sub, narrow_round, gen_narrow_round_high_u32)
1666 
1667 static bool do_long_3d(DisasContext *s, arg_3diff *a,
1668                        NeonGenTwoOpWidenFn *opfn,
1669                        NeonGenTwo64OpFn *accfn)
1670 {
1671     /*
1672      * 3-regs different lengths, long operations.
1673      * These perform an operation on two inputs that returns a double-width
1674      * result, and then possibly perform an accumulation operation of
1675      * that result into the double-width destination.
1676      */
1677     TCGv_i64 rd0, rd1, tmp;
1678     TCGv_i32 rn, rm;
1679 
1680     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1681         return false;
1682     }
1683 
1684     /* UNDEF accesses to D16-D31 if they don't exist. */
1685     if (!dc_isar_feature(aa32_simd_r32, s) &&
1686         ((a->vd | a->vn | a->vm) & 0x10)) {
1687         return false;
1688     }
1689 
1690     if (!opfn) {
1691         /* size == 3 case, which is an entirely different insn group */
1692         return false;
1693     }
1694 
1695     if (a->vd & 1) {
1696         return false;
1697     }
1698 
1699     if (!vfp_access_check(s)) {
1700         return true;
1701     }
1702 
1703     rd0 = tcg_temp_new_i64();
1704     rd1 = tcg_temp_new_i64();
1705 
1706     rn = tcg_temp_new_i32();
1707     rm = tcg_temp_new_i32();
1708     read_neon_element32(rn, a->vn, 0, MO_32);
1709     read_neon_element32(rm, a->vm, 0, MO_32);
1710     opfn(rd0, rn, rm);
1711 
1712     read_neon_element32(rn, a->vn, 1, MO_32);
1713     read_neon_element32(rm, a->vm, 1, MO_32);
1714     opfn(rd1, rn, rm);
1715 
1716     /* Don't store results until after all loads: they might overlap */
1717     if (accfn) {
1718         tmp = tcg_temp_new_i64();
1719         read_neon_element64(tmp, a->vd, 0, MO_64);
1720         accfn(rd0, tmp, rd0);
1721         read_neon_element64(tmp, a->vd, 1, MO_64);
1722         accfn(rd1, tmp, rd1);
1723     }
1724 
1725     write_neon_element64(rd0, a->vd, 0, MO_64);
1726     write_neon_element64(rd1, a->vd, 1, MO_64);
1727 
1728     return true;
1729 }
1730 
1731 static bool trans_VABDL_S_3d(DisasContext *s, arg_3diff *a)
1732 {
1733     static NeonGenTwoOpWidenFn * const opfn[] = {
1734         gen_helper_neon_abdl_s16,
1735         gen_helper_neon_abdl_s32,
1736         gen_helper_neon_abdl_s64,
1737         NULL,
1738     };
1739 
1740     return do_long_3d(s, a, opfn[a->size], NULL);
1741 }
1742 
1743 static bool trans_VABDL_U_3d(DisasContext *s, arg_3diff *a)
1744 {
1745     static NeonGenTwoOpWidenFn * const opfn[] = {
1746         gen_helper_neon_abdl_u16,
1747         gen_helper_neon_abdl_u32,
1748         gen_helper_neon_abdl_u64,
1749         NULL,
1750     };
1751 
1752     return do_long_3d(s, a, opfn[a->size], NULL);
1753 }
1754 
1755 static bool trans_VABAL_S_3d(DisasContext *s, arg_3diff *a)
1756 {
1757     static NeonGenTwoOpWidenFn * const opfn[] = {
1758         gen_helper_neon_abdl_s16,
1759         gen_helper_neon_abdl_s32,
1760         gen_helper_neon_abdl_s64,
1761         NULL,
1762     };
1763     static NeonGenTwo64OpFn * const addfn[] = {
1764         gen_helper_neon_addl_u16,
1765         gen_helper_neon_addl_u32,
1766         tcg_gen_add_i64,
1767         NULL,
1768     };
1769 
1770     return do_long_3d(s, a, opfn[a->size], addfn[a->size]);
1771 }
1772 
1773 static bool trans_VABAL_U_3d(DisasContext *s, arg_3diff *a)
1774 {
1775     static NeonGenTwoOpWidenFn * const opfn[] = {
1776         gen_helper_neon_abdl_u16,
1777         gen_helper_neon_abdl_u32,
1778         gen_helper_neon_abdl_u64,
1779         NULL,
1780     };
1781     static NeonGenTwo64OpFn * const addfn[] = {
1782         gen_helper_neon_addl_u16,
1783         gen_helper_neon_addl_u32,
1784         tcg_gen_add_i64,
1785         NULL,
1786     };
1787 
1788     return do_long_3d(s, a, opfn[a->size], addfn[a->size]);
1789 }
1790 
1791 static void gen_mull_s32(TCGv_i64 rd, TCGv_i32 rn, TCGv_i32 rm)
1792 {
1793     TCGv_i32 lo = tcg_temp_new_i32();
1794     TCGv_i32 hi = tcg_temp_new_i32();
1795 
1796     tcg_gen_muls2_i32(lo, hi, rn, rm);
1797     tcg_gen_concat_i32_i64(rd, lo, hi);
1798 }
1799 
1800 static void gen_mull_u32(TCGv_i64 rd, TCGv_i32 rn, TCGv_i32 rm)
1801 {
1802     TCGv_i32 lo = tcg_temp_new_i32();
1803     TCGv_i32 hi = tcg_temp_new_i32();
1804 
1805     tcg_gen_mulu2_i32(lo, hi, rn, rm);
1806     tcg_gen_concat_i32_i64(rd, lo, hi);
1807 }
1808 
1809 static bool trans_VMULL_S_3d(DisasContext *s, arg_3diff *a)
1810 {
1811     static NeonGenTwoOpWidenFn * const opfn[] = {
1812         gen_helper_neon_mull_s8,
1813         gen_helper_neon_mull_s16,
1814         gen_mull_s32,
1815         NULL,
1816     };
1817 
1818     return do_long_3d(s, a, opfn[a->size], NULL);
1819 }
1820 
1821 static bool trans_VMULL_U_3d(DisasContext *s, arg_3diff *a)
1822 {
1823     static NeonGenTwoOpWidenFn * const opfn[] = {
1824         gen_helper_neon_mull_u8,
1825         gen_helper_neon_mull_u16,
1826         gen_mull_u32,
1827         NULL,
1828     };
1829 
1830     return do_long_3d(s, a, opfn[a->size], NULL);
1831 }
1832 
1833 #define DO_VMLAL(INSN,MULL,ACC)                                         \
1834     static bool trans_##INSN##_3d(DisasContext *s, arg_3diff *a)        \
1835     {                                                                   \
1836         static NeonGenTwoOpWidenFn * const opfn[] = {                   \
1837             gen_helper_neon_##MULL##8,                                  \
1838             gen_helper_neon_##MULL##16,                                 \
1839             gen_##MULL##32,                                             \
1840             NULL,                                                       \
1841         };                                                              \
1842         static NeonGenTwo64OpFn * const accfn[] = {                     \
1843             gen_helper_neon_##ACC##l_u16,                               \
1844             gen_helper_neon_##ACC##l_u32,                               \
1845             tcg_gen_##ACC##_i64,                                        \
1846             NULL,                                                       \
1847         };                                                              \
1848         return do_long_3d(s, a, opfn[a->size], accfn[a->size]);         \
1849     }
1850 
1851 DO_VMLAL(VMLAL_S,mull_s,add)
1852 DO_VMLAL(VMLAL_U,mull_u,add)
1853 DO_VMLAL(VMLSL_S,mull_s,sub)
1854 DO_VMLAL(VMLSL_U,mull_u,sub)
1855 
1856 static void gen_VQDMULL_16(TCGv_i64 rd, TCGv_i32 rn, TCGv_i32 rm)
1857 {
1858     gen_helper_neon_mull_s16(rd, rn, rm);
1859     gen_helper_neon_addl_saturate_s32(rd, tcg_env, rd, rd);
1860 }
1861 
1862 static void gen_VQDMULL_32(TCGv_i64 rd, TCGv_i32 rn, TCGv_i32 rm)
1863 {
1864     gen_mull_s32(rd, rn, rm);
1865     gen_helper_neon_addl_saturate_s64(rd, tcg_env, rd, rd);
1866 }
1867 
1868 static bool trans_VQDMULL_3d(DisasContext *s, arg_3diff *a)
1869 {
1870     static NeonGenTwoOpWidenFn * const opfn[] = {
1871         NULL,
1872         gen_VQDMULL_16,
1873         gen_VQDMULL_32,
1874         NULL,
1875     };
1876 
1877     return do_long_3d(s, a, opfn[a->size], NULL);
1878 }
1879 
1880 static void gen_VQDMLAL_acc_16(TCGv_i64 rd, TCGv_i64 rn, TCGv_i64 rm)
1881 {
1882     gen_helper_neon_addl_saturate_s32(rd, tcg_env, rn, rm);
1883 }
1884 
1885 static void gen_VQDMLAL_acc_32(TCGv_i64 rd, TCGv_i64 rn, TCGv_i64 rm)
1886 {
1887     gen_helper_neon_addl_saturate_s64(rd, tcg_env, rn, rm);
1888 }
1889 
1890 static bool trans_VQDMLAL_3d(DisasContext *s, arg_3diff *a)
1891 {
1892     static NeonGenTwoOpWidenFn * const opfn[] = {
1893         NULL,
1894         gen_VQDMULL_16,
1895         gen_VQDMULL_32,
1896         NULL,
1897     };
1898     static NeonGenTwo64OpFn * const accfn[] = {
1899         NULL,
1900         gen_VQDMLAL_acc_16,
1901         gen_VQDMLAL_acc_32,
1902         NULL,
1903     };
1904 
1905     return do_long_3d(s, a, opfn[a->size], accfn[a->size]);
1906 }
1907 
1908 static void gen_VQDMLSL_acc_16(TCGv_i64 rd, TCGv_i64 rn, TCGv_i64 rm)
1909 {
1910     gen_helper_neon_negl_u32(rm, rm);
1911     gen_helper_neon_addl_saturate_s32(rd, tcg_env, rn, rm);
1912 }
1913 
1914 static void gen_VQDMLSL_acc_32(TCGv_i64 rd, TCGv_i64 rn, TCGv_i64 rm)
1915 {
1916     tcg_gen_neg_i64(rm, rm);
1917     gen_helper_neon_addl_saturate_s64(rd, tcg_env, rn, rm);
1918 }
1919 
1920 static bool trans_VQDMLSL_3d(DisasContext *s, arg_3diff *a)
1921 {
1922     static NeonGenTwoOpWidenFn * const opfn[] = {
1923         NULL,
1924         gen_VQDMULL_16,
1925         gen_VQDMULL_32,
1926         NULL,
1927     };
1928     static NeonGenTwo64OpFn * const accfn[] = {
1929         NULL,
1930         gen_VQDMLSL_acc_16,
1931         gen_VQDMLSL_acc_32,
1932         NULL,
1933     };
1934 
1935     return do_long_3d(s, a, opfn[a->size], accfn[a->size]);
1936 }
1937 
1938 static bool trans_VMULL_P_3d(DisasContext *s, arg_3diff *a)
1939 {
1940     gen_helper_gvec_3 *fn_gvec;
1941 
1942     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
1943         return false;
1944     }
1945 
1946     /* UNDEF accesses to D16-D31 if they don't exist. */
1947     if (!dc_isar_feature(aa32_simd_r32, s) &&
1948         ((a->vd | a->vn | a->vm) & 0x10)) {
1949         return false;
1950     }
1951 
1952     if (a->vd & 1) {
1953         return false;
1954     }
1955 
1956     switch (a->size) {
1957     case 0:
1958         fn_gvec = gen_helper_neon_pmull_h;
1959         break;
1960     case 2:
1961         if (!dc_isar_feature(aa32_pmull, s)) {
1962             return false;
1963         }
1964         fn_gvec = gen_helper_gvec_pmull_q;
1965         break;
1966     default:
1967         return false;
1968     }
1969 
1970     if (!vfp_access_check(s)) {
1971         return true;
1972     }
1973 
1974     tcg_gen_gvec_3_ool(neon_full_reg_offset(a->vd),
1975                        neon_full_reg_offset(a->vn),
1976                        neon_full_reg_offset(a->vm),
1977                        16, 16, 0, fn_gvec);
1978     return true;
1979 }
1980 
1981 static void gen_neon_dup_low16(TCGv_i32 var)
1982 {
1983     TCGv_i32 tmp = tcg_temp_new_i32();
1984     tcg_gen_ext16u_i32(var, var);
1985     tcg_gen_shli_i32(tmp, var, 16);
1986     tcg_gen_or_i32(var, var, tmp);
1987 }
1988 
1989 static void gen_neon_dup_high16(TCGv_i32 var)
1990 {
1991     TCGv_i32 tmp = tcg_temp_new_i32();
1992     tcg_gen_andi_i32(var, var, 0xffff0000);
1993     tcg_gen_shri_i32(tmp, var, 16);
1994     tcg_gen_or_i32(var, var, tmp);
1995 }
1996 
1997 static inline TCGv_i32 neon_get_scalar(int size, int reg)
1998 {
1999     TCGv_i32 tmp = tcg_temp_new_i32();
2000     if (size == MO_16) {
2001         read_neon_element32(tmp, reg & 7, reg >> 4, MO_32);
2002         if (reg & 8) {
2003             gen_neon_dup_high16(tmp);
2004         } else {
2005             gen_neon_dup_low16(tmp);
2006         }
2007     } else {
2008         read_neon_element32(tmp, reg & 15, reg >> 4, MO_32);
2009     }
2010     return tmp;
2011 }
2012 
2013 static bool do_2scalar(DisasContext *s, arg_2scalar *a,
2014                        NeonGenTwoOpFn *opfn, NeonGenTwoOpFn *accfn)
2015 {
2016     /*
2017      * Two registers and a scalar: perform an operation between
2018      * the input elements and the scalar, and then possibly
2019      * perform an accumulation operation of that result into the
2020      * destination.
2021      */
2022     TCGv_i32 scalar, tmp;
2023     int pass;
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         /* Bad size (including size == 3, which is a different insn group) */
2037         return false;
2038     }
2039 
2040     if (a->q && ((a->vd | a->vn) & 1)) {
2041         return false;
2042     }
2043 
2044     if (!vfp_access_check(s)) {
2045         return true;
2046     }
2047 
2048     scalar = neon_get_scalar(a->size, a->vm);
2049     tmp = tcg_temp_new_i32();
2050 
2051     for (pass = 0; pass < (a->q ? 4 : 2); pass++) {
2052         read_neon_element32(tmp, a->vn, pass, MO_32);
2053         opfn(tmp, tmp, scalar);
2054         if (accfn) {
2055             TCGv_i32 rd = tcg_temp_new_i32();
2056             read_neon_element32(rd, a->vd, pass, MO_32);
2057             accfn(tmp, rd, tmp);
2058         }
2059         write_neon_element32(tmp, a->vd, pass, MO_32);
2060     }
2061     return true;
2062 }
2063 
2064 static bool trans_VMUL_2sc(DisasContext *s, arg_2scalar *a)
2065 {
2066     static NeonGenTwoOpFn * const opfn[] = {
2067         NULL,
2068         gen_helper_neon_mul_u16,
2069         tcg_gen_mul_i32,
2070         NULL,
2071     };
2072 
2073     return do_2scalar(s, a, opfn[a->size], NULL);
2074 }
2075 
2076 static bool trans_VMLA_2sc(DisasContext *s, arg_2scalar *a)
2077 {
2078     static NeonGenTwoOpFn * const opfn[] = {
2079         NULL,
2080         gen_helper_neon_mul_u16,
2081         tcg_gen_mul_i32,
2082         NULL,
2083     };
2084     static NeonGenTwoOpFn * const accfn[] = {
2085         NULL,
2086         gen_helper_neon_add_u16,
2087         tcg_gen_add_i32,
2088         NULL,
2089     };
2090 
2091     return do_2scalar(s, a, opfn[a->size], accfn[a->size]);
2092 }
2093 
2094 static bool trans_VMLS_2sc(DisasContext *s, arg_2scalar *a)
2095 {
2096     static NeonGenTwoOpFn * const opfn[] = {
2097         NULL,
2098         gen_helper_neon_mul_u16,
2099         tcg_gen_mul_i32,
2100         NULL,
2101     };
2102     static NeonGenTwoOpFn * const accfn[] = {
2103         NULL,
2104         gen_helper_neon_sub_u16,
2105         tcg_gen_sub_i32,
2106         NULL,
2107     };
2108 
2109     return do_2scalar(s, a, opfn[a->size], accfn[a->size]);
2110 }
2111 
2112 static bool do_2scalar_fp_vec(DisasContext *s, arg_2scalar *a,
2113                               gen_helper_gvec_3_ptr *fn)
2114 {
2115     /* Two registers and a scalar, using gvec */
2116     int vec_size = a->q ? 16 : 8;
2117     int rd_ofs = neon_full_reg_offset(a->vd);
2118     int rn_ofs = neon_full_reg_offset(a->vn);
2119     int rm_ofs;
2120     int idx;
2121     TCGv_ptr fpstatus;
2122 
2123     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2124         return false;
2125     }
2126 
2127     /* UNDEF accesses to D16-D31 if they don't exist. */
2128     if (!dc_isar_feature(aa32_simd_r32, s) &&
2129         ((a->vd | a->vn | a->vm) & 0x10)) {
2130         return false;
2131     }
2132 
2133     if (!fn) {
2134         /* Bad size (including size == 3, which is a different insn group) */
2135         return false;
2136     }
2137 
2138     if (a->q && ((a->vd | a->vn) & 1)) {
2139         return false;
2140     }
2141 
2142     if (!vfp_access_check(s)) {
2143         return true;
2144     }
2145 
2146     /* a->vm is M:Vm, which encodes both register and index */
2147     idx = extract32(a->vm, a->size + 2, 2);
2148     a->vm = extract32(a->vm, 0, a->size + 2);
2149     rm_ofs = neon_full_reg_offset(a->vm);
2150 
2151     fpstatus = fpstatus_ptr(a->size == 1 ? FPST_STD_F16 : FPST_STD);
2152     tcg_gen_gvec_3_ptr(rd_ofs, rn_ofs, rm_ofs, fpstatus,
2153                        vec_size, vec_size, idx, fn);
2154     return true;
2155 }
2156 
2157 #define DO_VMUL_F_2sc(NAME, FUNC)                                       \
2158     static bool trans_##NAME##_F_2sc(DisasContext *s, arg_2scalar *a)   \
2159     {                                                                   \
2160         static gen_helper_gvec_3_ptr * const opfn[] = {                 \
2161             NULL,                                                       \
2162             gen_helper_##FUNC##_h,                                      \
2163             gen_helper_##FUNC##_s,                                      \
2164             NULL,                                                       \
2165         };                                                              \
2166         if (a->size == MO_16 && !dc_isar_feature(aa32_fp16_arith, s)) { \
2167             return false;                                               \
2168         }                                                               \
2169         return do_2scalar_fp_vec(s, a, opfn[a->size]);                  \
2170     }
2171 
2172 DO_VMUL_F_2sc(VMUL, gvec_fmul_idx)
2173 DO_VMUL_F_2sc(VMLA, gvec_fmla_nf_idx)
2174 DO_VMUL_F_2sc(VMLS, gvec_fmls_nf_idx)
2175 
2176 WRAP_ENV_FN(gen_VQDMULH_16, gen_helper_neon_qdmulh_s16)
2177 WRAP_ENV_FN(gen_VQDMULH_32, gen_helper_neon_qdmulh_s32)
2178 WRAP_ENV_FN(gen_VQRDMULH_16, gen_helper_neon_qrdmulh_s16)
2179 WRAP_ENV_FN(gen_VQRDMULH_32, gen_helper_neon_qrdmulh_s32)
2180 
2181 static bool trans_VQDMULH_2sc(DisasContext *s, arg_2scalar *a)
2182 {
2183     static NeonGenTwoOpFn * const opfn[] = {
2184         NULL,
2185         gen_VQDMULH_16,
2186         gen_VQDMULH_32,
2187         NULL,
2188     };
2189 
2190     return do_2scalar(s, a, opfn[a->size], NULL);
2191 }
2192 
2193 static bool trans_VQRDMULH_2sc(DisasContext *s, arg_2scalar *a)
2194 {
2195     static NeonGenTwoOpFn * const opfn[] = {
2196         NULL,
2197         gen_VQRDMULH_16,
2198         gen_VQRDMULH_32,
2199         NULL,
2200     };
2201 
2202     return do_2scalar(s, a, opfn[a->size], NULL);
2203 }
2204 
2205 static bool do_vqrdmlah_2sc(DisasContext *s, arg_2scalar *a,
2206                             NeonGenThreeOpEnvFn *opfn)
2207 {
2208     /*
2209      * VQRDMLAH/VQRDMLSH: this is like do_2scalar, but the opfn
2210      * performs a kind of fused op-then-accumulate using a helper
2211      * function that takes all of rd, rn and the scalar at once.
2212      */
2213     TCGv_i32 scalar, rn, rd;
2214     int pass;
2215 
2216     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2217         return false;
2218     }
2219 
2220     if (!dc_isar_feature(aa32_rdm, s)) {
2221         return false;
2222     }
2223 
2224     /* UNDEF accesses to D16-D31 if they don't exist. */
2225     if (!dc_isar_feature(aa32_simd_r32, s) &&
2226         ((a->vd | a->vn | a->vm) & 0x10)) {
2227         return false;
2228     }
2229 
2230     if (!opfn) {
2231         /* Bad size (including size == 3, which is a different insn group) */
2232         return false;
2233     }
2234 
2235     if (a->q && ((a->vd | a->vn) & 1)) {
2236         return false;
2237     }
2238 
2239     if (!vfp_access_check(s)) {
2240         return true;
2241     }
2242 
2243     scalar = neon_get_scalar(a->size, a->vm);
2244     rn = tcg_temp_new_i32();
2245     rd = tcg_temp_new_i32();
2246 
2247     for (pass = 0; pass < (a->q ? 4 : 2); pass++) {
2248         read_neon_element32(rn, a->vn, pass, MO_32);
2249         read_neon_element32(rd, a->vd, pass, MO_32);
2250         opfn(rd, tcg_env, rn, scalar, rd);
2251         write_neon_element32(rd, a->vd, pass, MO_32);
2252     }
2253     return true;
2254 }
2255 
2256 static bool trans_VQRDMLAH_2sc(DisasContext *s, arg_2scalar *a)
2257 {
2258     static NeonGenThreeOpEnvFn *opfn[] = {
2259         NULL,
2260         gen_helper_neon_qrdmlah_s16,
2261         gen_helper_neon_qrdmlah_s32,
2262         NULL,
2263     };
2264     return do_vqrdmlah_2sc(s, a, opfn[a->size]);
2265 }
2266 
2267 static bool trans_VQRDMLSH_2sc(DisasContext *s, arg_2scalar *a)
2268 {
2269     static NeonGenThreeOpEnvFn *opfn[] = {
2270         NULL,
2271         gen_helper_neon_qrdmlsh_s16,
2272         gen_helper_neon_qrdmlsh_s32,
2273         NULL,
2274     };
2275     return do_vqrdmlah_2sc(s, a, opfn[a->size]);
2276 }
2277 
2278 static bool do_2scalar_long(DisasContext *s, arg_2scalar *a,
2279                             NeonGenTwoOpWidenFn *opfn,
2280                             NeonGenTwo64OpFn *accfn)
2281 {
2282     /*
2283      * Two registers and a scalar, long operations: perform an
2284      * operation on the input elements and the scalar which produces
2285      * a double-width result, and then possibly perform an accumulation
2286      * operation of that result into the destination.
2287      */
2288     TCGv_i32 scalar, rn;
2289     TCGv_i64 rn0_64, rn1_64;
2290 
2291     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2292         return false;
2293     }
2294 
2295     /* UNDEF accesses to D16-D31 if they don't exist. */
2296     if (!dc_isar_feature(aa32_simd_r32, s) &&
2297         ((a->vd | a->vn | a->vm) & 0x10)) {
2298         return false;
2299     }
2300 
2301     if (!opfn) {
2302         /* Bad size (including size == 3, which is a different insn group) */
2303         return false;
2304     }
2305 
2306     if (a->vd & 1) {
2307         return false;
2308     }
2309 
2310     if (!vfp_access_check(s)) {
2311         return true;
2312     }
2313 
2314     scalar = neon_get_scalar(a->size, a->vm);
2315 
2316     /* Load all inputs before writing any outputs, in case of overlap */
2317     rn = tcg_temp_new_i32();
2318     read_neon_element32(rn, a->vn, 0, MO_32);
2319     rn0_64 = tcg_temp_new_i64();
2320     opfn(rn0_64, rn, scalar);
2321 
2322     read_neon_element32(rn, a->vn, 1, MO_32);
2323     rn1_64 = tcg_temp_new_i64();
2324     opfn(rn1_64, rn, scalar);
2325 
2326     if (accfn) {
2327         TCGv_i64 t64 = tcg_temp_new_i64();
2328         read_neon_element64(t64, a->vd, 0, MO_64);
2329         accfn(rn0_64, t64, rn0_64);
2330         read_neon_element64(t64, a->vd, 1, MO_64);
2331         accfn(rn1_64, t64, rn1_64);
2332     }
2333 
2334     write_neon_element64(rn0_64, a->vd, 0, MO_64);
2335     write_neon_element64(rn1_64, a->vd, 1, MO_64);
2336     return true;
2337 }
2338 
2339 static bool trans_VMULL_S_2sc(DisasContext *s, arg_2scalar *a)
2340 {
2341     static NeonGenTwoOpWidenFn * const opfn[] = {
2342         NULL,
2343         gen_helper_neon_mull_s16,
2344         gen_mull_s32,
2345         NULL,
2346     };
2347 
2348     return do_2scalar_long(s, a, opfn[a->size], NULL);
2349 }
2350 
2351 static bool trans_VMULL_U_2sc(DisasContext *s, arg_2scalar *a)
2352 {
2353     static NeonGenTwoOpWidenFn * const opfn[] = {
2354         NULL,
2355         gen_helper_neon_mull_u16,
2356         gen_mull_u32,
2357         NULL,
2358     };
2359 
2360     return do_2scalar_long(s, a, opfn[a->size], NULL);
2361 }
2362 
2363 #define DO_VMLAL_2SC(INSN, MULL, ACC)                                   \
2364     static bool trans_##INSN##_2sc(DisasContext *s, arg_2scalar *a)     \
2365     {                                                                   \
2366         static NeonGenTwoOpWidenFn * const opfn[] = {                   \
2367             NULL,                                                       \
2368             gen_helper_neon_##MULL##16,                                 \
2369             gen_##MULL##32,                                             \
2370             NULL,                                                       \
2371         };                                                              \
2372         static NeonGenTwo64OpFn * const accfn[] = {                     \
2373             NULL,                                                       \
2374             gen_helper_neon_##ACC##l_u32,                               \
2375             tcg_gen_##ACC##_i64,                                        \
2376             NULL,                                                       \
2377         };                                                              \
2378         return do_2scalar_long(s, a, opfn[a->size], accfn[a->size]);    \
2379     }
2380 
2381 DO_VMLAL_2SC(VMLAL_S, mull_s, add)
2382 DO_VMLAL_2SC(VMLAL_U, mull_u, add)
2383 DO_VMLAL_2SC(VMLSL_S, mull_s, sub)
2384 DO_VMLAL_2SC(VMLSL_U, mull_u, sub)
2385 
2386 static bool trans_VQDMULL_2sc(DisasContext *s, arg_2scalar *a)
2387 {
2388     static NeonGenTwoOpWidenFn * const opfn[] = {
2389         NULL,
2390         gen_VQDMULL_16,
2391         gen_VQDMULL_32,
2392         NULL,
2393     };
2394 
2395     return do_2scalar_long(s, a, opfn[a->size], NULL);
2396 }
2397 
2398 static bool trans_VQDMLAL_2sc(DisasContext *s, arg_2scalar *a)
2399 {
2400     static NeonGenTwoOpWidenFn * const opfn[] = {
2401         NULL,
2402         gen_VQDMULL_16,
2403         gen_VQDMULL_32,
2404         NULL,
2405     };
2406     static NeonGenTwo64OpFn * const accfn[] = {
2407         NULL,
2408         gen_VQDMLAL_acc_16,
2409         gen_VQDMLAL_acc_32,
2410         NULL,
2411     };
2412 
2413     return do_2scalar_long(s, a, opfn[a->size], accfn[a->size]);
2414 }
2415 
2416 static bool trans_VQDMLSL_2sc(DisasContext *s, arg_2scalar *a)
2417 {
2418     static NeonGenTwoOpWidenFn * const opfn[] = {
2419         NULL,
2420         gen_VQDMULL_16,
2421         gen_VQDMULL_32,
2422         NULL,
2423     };
2424     static NeonGenTwo64OpFn * const accfn[] = {
2425         NULL,
2426         gen_VQDMLSL_acc_16,
2427         gen_VQDMLSL_acc_32,
2428         NULL,
2429     };
2430 
2431     return do_2scalar_long(s, a, opfn[a->size], accfn[a->size]);
2432 }
2433 
2434 static bool trans_VEXT(DisasContext *s, arg_VEXT *a)
2435 {
2436     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2437         return false;
2438     }
2439 
2440     /* UNDEF accesses to D16-D31 if they don't exist. */
2441     if (!dc_isar_feature(aa32_simd_r32, s) &&
2442         ((a->vd | a->vn | a->vm) & 0x10)) {
2443         return false;
2444     }
2445 
2446     if ((a->vn | a->vm | a->vd) & a->q) {
2447         return false;
2448     }
2449 
2450     if (a->imm > 7 && !a->q) {
2451         return false;
2452     }
2453 
2454     if (!vfp_access_check(s)) {
2455         return true;
2456     }
2457 
2458     if (!a->q) {
2459         /* Extract 64 bits from <Vm:Vn> */
2460         TCGv_i64 left, right, dest;
2461 
2462         left = tcg_temp_new_i64();
2463         right = tcg_temp_new_i64();
2464         dest = tcg_temp_new_i64();
2465 
2466         read_neon_element64(right, a->vn, 0, MO_64);
2467         read_neon_element64(left, a->vm, 0, MO_64);
2468         tcg_gen_extract2_i64(dest, right, left, a->imm * 8);
2469         write_neon_element64(dest, a->vd, 0, MO_64);
2470     } else {
2471         /* Extract 128 bits from <Vm+1:Vm:Vn+1:Vn> */
2472         TCGv_i64 left, middle, right, destleft, destright;
2473 
2474         left = tcg_temp_new_i64();
2475         middle = tcg_temp_new_i64();
2476         right = tcg_temp_new_i64();
2477         destleft = tcg_temp_new_i64();
2478         destright = tcg_temp_new_i64();
2479 
2480         if (a->imm < 8) {
2481             read_neon_element64(right, a->vn, 0, MO_64);
2482             read_neon_element64(middle, a->vn, 1, MO_64);
2483             tcg_gen_extract2_i64(destright, right, middle, a->imm * 8);
2484             read_neon_element64(left, a->vm, 0, MO_64);
2485             tcg_gen_extract2_i64(destleft, middle, left, a->imm * 8);
2486         } else {
2487             read_neon_element64(right, a->vn, 1, MO_64);
2488             read_neon_element64(middle, a->vm, 0, MO_64);
2489             tcg_gen_extract2_i64(destright, right, middle, (a->imm - 8) * 8);
2490             read_neon_element64(left, a->vm, 1, MO_64);
2491             tcg_gen_extract2_i64(destleft, middle, left, (a->imm - 8) * 8);
2492         }
2493 
2494         write_neon_element64(destright, a->vd, 0, MO_64);
2495         write_neon_element64(destleft, a->vd, 1, MO_64);
2496     }
2497     return true;
2498 }
2499 
2500 static bool trans_VTBL(DisasContext *s, arg_VTBL *a)
2501 {
2502     TCGv_i64 val, def;
2503     TCGv_i32 desc;
2504 
2505     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2506         return false;
2507     }
2508 
2509     /* UNDEF accesses to D16-D31 if they don't exist. */
2510     if (!dc_isar_feature(aa32_simd_r32, s) &&
2511         ((a->vd | a->vn | a->vm) & 0x10)) {
2512         return false;
2513     }
2514 
2515     if ((a->vn + a->len + 1) > 32) {
2516         /*
2517          * This is UNPREDICTABLE; we choose to UNDEF to avoid the
2518          * helper function running off the end of the register file.
2519          */
2520         return false;
2521     }
2522 
2523     if (!vfp_access_check(s)) {
2524         return true;
2525     }
2526 
2527     desc = tcg_constant_i32((a->vn << 2) | a->len);
2528     def = tcg_temp_new_i64();
2529     if (a->op) {
2530         read_neon_element64(def, a->vd, 0, MO_64);
2531     } else {
2532         tcg_gen_movi_i64(def, 0);
2533     }
2534     val = tcg_temp_new_i64();
2535     read_neon_element64(val, a->vm, 0, MO_64);
2536 
2537     gen_helper_neon_tbl(val, tcg_env, desc, val, def);
2538     write_neon_element64(val, a->vd, 0, MO_64);
2539     return true;
2540 }
2541 
2542 static bool trans_VDUP_scalar(DisasContext *s, arg_VDUP_scalar *a)
2543 {
2544     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2545         return false;
2546     }
2547 
2548     /* UNDEF accesses to D16-D31 if they don't exist. */
2549     if (!dc_isar_feature(aa32_simd_r32, s) &&
2550         ((a->vd | a->vm) & 0x10)) {
2551         return false;
2552     }
2553 
2554     if (a->vd & a->q) {
2555         return false;
2556     }
2557 
2558     if (!vfp_access_check(s)) {
2559         return true;
2560     }
2561 
2562     tcg_gen_gvec_dup_mem(a->size, neon_full_reg_offset(a->vd),
2563                          neon_element_offset(a->vm, a->index, a->size),
2564                          a->q ? 16 : 8, a->q ? 16 : 8);
2565     return true;
2566 }
2567 
2568 static bool trans_VREV64(DisasContext *s, arg_VREV64 *a)
2569 {
2570     int pass, half;
2571     TCGv_i32 tmp[2];
2572 
2573     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2574         return false;
2575     }
2576 
2577     /* UNDEF accesses to D16-D31 if they don't exist. */
2578     if (!dc_isar_feature(aa32_simd_r32, s) &&
2579         ((a->vd | a->vm) & 0x10)) {
2580         return false;
2581     }
2582 
2583     if ((a->vd | a->vm) & a->q) {
2584         return false;
2585     }
2586 
2587     if (a->size == 3) {
2588         return false;
2589     }
2590 
2591     if (!vfp_access_check(s)) {
2592         return true;
2593     }
2594 
2595     tmp[0] = tcg_temp_new_i32();
2596     tmp[1] = tcg_temp_new_i32();
2597 
2598     for (pass = 0; pass < (a->q ? 2 : 1); pass++) {
2599         for (half = 0; half < 2; half++) {
2600             read_neon_element32(tmp[half], a->vm, pass * 2 + half, MO_32);
2601             switch (a->size) {
2602             case 0:
2603                 tcg_gen_bswap32_i32(tmp[half], tmp[half]);
2604                 break;
2605             case 1:
2606                 gen_swap_half(tmp[half], tmp[half]);
2607                 break;
2608             case 2:
2609                 break;
2610             default:
2611                 g_assert_not_reached();
2612             }
2613         }
2614         write_neon_element32(tmp[1], a->vd, pass * 2, MO_32);
2615         write_neon_element32(tmp[0], a->vd, pass * 2 + 1, MO_32);
2616     }
2617     return true;
2618 }
2619 
2620 static bool do_2misc_pairwise(DisasContext *s, arg_2misc *a,
2621                               NeonGenWidenFn *widenfn,
2622                               NeonGenTwo64OpFn *opfn,
2623                               NeonGenTwo64OpFn *accfn)
2624 {
2625     /*
2626      * Pairwise long operations: widen both halves of the pair,
2627      * combine the pairs with the opfn, and then possibly accumulate
2628      * into the destination with the accfn.
2629      */
2630     int pass;
2631 
2632     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2633         return false;
2634     }
2635 
2636     /* UNDEF accesses to D16-D31 if they don't exist. */
2637     if (!dc_isar_feature(aa32_simd_r32, s) &&
2638         ((a->vd | a->vm) & 0x10)) {
2639         return false;
2640     }
2641 
2642     if ((a->vd | a->vm) & a->q) {
2643         return false;
2644     }
2645 
2646     if (!widenfn) {
2647         return false;
2648     }
2649 
2650     if (!vfp_access_check(s)) {
2651         return true;
2652     }
2653 
2654     for (pass = 0; pass < a->q + 1; pass++) {
2655         TCGv_i32 tmp;
2656         TCGv_i64 rm0_64, rm1_64, rd_64;
2657 
2658         rm0_64 = tcg_temp_new_i64();
2659         rm1_64 = tcg_temp_new_i64();
2660         rd_64 = tcg_temp_new_i64();
2661 
2662         tmp = tcg_temp_new_i32();
2663         read_neon_element32(tmp, a->vm, pass * 2, MO_32);
2664         widenfn(rm0_64, tmp);
2665         read_neon_element32(tmp, a->vm, pass * 2 + 1, MO_32);
2666         widenfn(rm1_64, tmp);
2667 
2668         opfn(rd_64, rm0_64, rm1_64);
2669 
2670         if (accfn) {
2671             TCGv_i64 tmp64 = tcg_temp_new_i64();
2672             read_neon_element64(tmp64, a->vd, pass, MO_64);
2673             accfn(rd_64, tmp64, rd_64);
2674         }
2675         write_neon_element64(rd_64, a->vd, pass, MO_64);
2676     }
2677     return true;
2678 }
2679 
2680 static bool trans_VPADDL_S(DisasContext *s, arg_2misc *a)
2681 {
2682     static NeonGenWidenFn * const widenfn[] = {
2683         gen_helper_neon_widen_s8,
2684         gen_helper_neon_widen_s16,
2685         tcg_gen_ext_i32_i64,
2686         NULL,
2687     };
2688     static NeonGenTwo64OpFn * const opfn[] = {
2689         gen_helper_neon_paddl_u16,
2690         gen_helper_neon_paddl_u32,
2691         tcg_gen_add_i64,
2692         NULL,
2693     };
2694 
2695     return do_2misc_pairwise(s, a, widenfn[a->size], opfn[a->size], NULL);
2696 }
2697 
2698 static bool trans_VPADDL_U(DisasContext *s, arg_2misc *a)
2699 {
2700     static NeonGenWidenFn * const widenfn[] = {
2701         gen_helper_neon_widen_u8,
2702         gen_helper_neon_widen_u16,
2703         tcg_gen_extu_i32_i64,
2704         NULL,
2705     };
2706     static NeonGenTwo64OpFn * const opfn[] = {
2707         gen_helper_neon_paddl_u16,
2708         gen_helper_neon_paddl_u32,
2709         tcg_gen_add_i64,
2710         NULL,
2711     };
2712 
2713     return do_2misc_pairwise(s, a, widenfn[a->size], opfn[a->size], NULL);
2714 }
2715 
2716 static bool trans_VPADAL_S(DisasContext *s, arg_2misc *a)
2717 {
2718     static NeonGenWidenFn * const widenfn[] = {
2719         gen_helper_neon_widen_s8,
2720         gen_helper_neon_widen_s16,
2721         tcg_gen_ext_i32_i64,
2722         NULL,
2723     };
2724     static NeonGenTwo64OpFn * const opfn[] = {
2725         gen_helper_neon_paddl_u16,
2726         gen_helper_neon_paddl_u32,
2727         tcg_gen_add_i64,
2728         NULL,
2729     };
2730     static NeonGenTwo64OpFn * const accfn[] = {
2731         gen_helper_neon_addl_u16,
2732         gen_helper_neon_addl_u32,
2733         tcg_gen_add_i64,
2734         NULL,
2735     };
2736 
2737     return do_2misc_pairwise(s, a, widenfn[a->size], opfn[a->size],
2738                              accfn[a->size]);
2739 }
2740 
2741 static bool trans_VPADAL_U(DisasContext *s, arg_2misc *a)
2742 {
2743     static NeonGenWidenFn * const widenfn[] = {
2744         gen_helper_neon_widen_u8,
2745         gen_helper_neon_widen_u16,
2746         tcg_gen_extu_i32_i64,
2747         NULL,
2748     };
2749     static NeonGenTwo64OpFn * const opfn[] = {
2750         gen_helper_neon_paddl_u16,
2751         gen_helper_neon_paddl_u32,
2752         tcg_gen_add_i64,
2753         NULL,
2754     };
2755     static NeonGenTwo64OpFn * const accfn[] = {
2756         gen_helper_neon_addl_u16,
2757         gen_helper_neon_addl_u32,
2758         tcg_gen_add_i64,
2759         NULL,
2760     };
2761 
2762     return do_2misc_pairwise(s, a, widenfn[a->size], opfn[a->size],
2763                              accfn[a->size]);
2764 }
2765 
2766 typedef void ZipFn(TCGv_ptr, TCGv_ptr);
2767 
2768 static bool do_zip_uzp(DisasContext *s, arg_2misc *a,
2769                        ZipFn *fn)
2770 {
2771     TCGv_ptr pd, pm;
2772 
2773     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2774         return false;
2775     }
2776 
2777     /* UNDEF accesses to D16-D31 if they don't exist. */
2778     if (!dc_isar_feature(aa32_simd_r32, s) &&
2779         ((a->vd | a->vm) & 0x10)) {
2780         return false;
2781     }
2782 
2783     if ((a->vd | a->vm) & a->q) {
2784         return false;
2785     }
2786 
2787     if (!fn) {
2788         /* Bad size or size/q combination */
2789         return false;
2790     }
2791 
2792     if (!vfp_access_check(s)) {
2793         return true;
2794     }
2795 
2796     pd = vfp_reg_ptr(true, a->vd);
2797     pm = vfp_reg_ptr(true, a->vm);
2798     fn(pd, pm);
2799     return true;
2800 }
2801 
2802 static bool trans_VUZP(DisasContext *s, arg_2misc *a)
2803 {
2804     static ZipFn * const fn[2][4] = {
2805         {
2806             gen_helper_neon_unzip8,
2807             gen_helper_neon_unzip16,
2808             NULL,
2809             NULL,
2810         }, {
2811             gen_helper_neon_qunzip8,
2812             gen_helper_neon_qunzip16,
2813             gen_helper_neon_qunzip32,
2814             NULL,
2815         }
2816     };
2817     return do_zip_uzp(s, a, fn[a->q][a->size]);
2818 }
2819 
2820 static bool trans_VZIP(DisasContext *s, arg_2misc *a)
2821 {
2822     static ZipFn * const fn[2][4] = {
2823         {
2824             gen_helper_neon_zip8,
2825             gen_helper_neon_zip16,
2826             NULL,
2827             NULL,
2828         }, {
2829             gen_helper_neon_qzip8,
2830             gen_helper_neon_qzip16,
2831             gen_helper_neon_qzip32,
2832             NULL,
2833         }
2834     };
2835     return do_zip_uzp(s, a, fn[a->q][a->size]);
2836 }
2837 
2838 static bool do_vmovn(DisasContext *s, arg_2misc *a,
2839                      NeonGenOne64OpEnvFn *narrowfn)
2840 {
2841     TCGv_i64 rm, rd0, rd1;
2842 
2843     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2844         return false;
2845     }
2846 
2847     /* UNDEF accesses to D16-D31 if they don't exist. */
2848     if (!dc_isar_feature(aa32_simd_r32, s) &&
2849         ((a->vd | a->vm) & 0x10)) {
2850         return false;
2851     }
2852 
2853     if (a->vm & 1) {
2854         return false;
2855     }
2856 
2857     if (!narrowfn) {
2858         return false;
2859     }
2860 
2861     if (!vfp_access_check(s)) {
2862         return true;
2863     }
2864 
2865     rm = tcg_temp_new_i64();
2866     rd0 = tcg_temp_new_i64();
2867     rd1 = tcg_temp_new_i64();
2868 
2869     read_neon_element64(rm, a->vm, 0, MO_64);
2870     narrowfn(rd0, tcg_env, rm);
2871     read_neon_element64(rm, a->vm, 1, MO_64);
2872     narrowfn(rd1, tcg_env, rm);
2873     write_neon_element64(rd0, a->vd, 0, MO_32);
2874     write_neon_element64(rd1, a->vd, 1, MO_32);
2875     return true;
2876 }
2877 
2878 #define DO_VMOVN(INSN, FUNC)                                    \
2879     static bool trans_##INSN(DisasContext *s, arg_2misc *a)     \
2880     {                                                           \
2881         static NeonGenOne64OpEnvFn * const narrowfn[] = {       \
2882             FUNC##8,                                            \
2883             FUNC##16,                                           \
2884             FUNC##32,                                           \
2885             NULL,                                               \
2886         };                                                      \
2887         return do_vmovn(s, a, narrowfn[a->size]);               \
2888     }
2889 
2890 DO_VMOVN(VMOVN, gen_neon_narrow_u)
2891 DO_VMOVN(VQMOVUN, gen_helper_neon_unarrow_sat)
2892 DO_VMOVN(VQMOVN_S, gen_helper_neon_narrow_sat_s)
2893 DO_VMOVN(VQMOVN_U, gen_helper_neon_narrow_sat_u)
2894 
2895 static bool trans_VSHLL(DisasContext *s, arg_2misc *a)
2896 {
2897     TCGv_i32 rm0, rm1;
2898     TCGv_i64 rd;
2899     static NeonGenWidenFn * const widenfns[] = {
2900         gen_helper_neon_widen_u8,
2901         gen_helper_neon_widen_u16,
2902         tcg_gen_extu_i32_i64,
2903         NULL,
2904     };
2905     NeonGenWidenFn *widenfn = widenfns[a->size];
2906 
2907     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
2908         return false;
2909     }
2910 
2911     /* UNDEF accesses to D16-D31 if they don't exist. */
2912     if (!dc_isar_feature(aa32_simd_r32, s) &&
2913         ((a->vd | a->vm) & 0x10)) {
2914         return false;
2915     }
2916 
2917     if (a->vd & 1) {
2918         return false;
2919     }
2920 
2921     if (!widenfn) {
2922         return false;
2923     }
2924 
2925     if (!vfp_access_check(s)) {
2926         return true;
2927     }
2928 
2929     rd = tcg_temp_new_i64();
2930     rm0 = tcg_temp_new_i32();
2931     rm1 = tcg_temp_new_i32();
2932 
2933     read_neon_element32(rm0, a->vm, 0, MO_32);
2934     read_neon_element32(rm1, a->vm, 1, MO_32);
2935 
2936     widenfn(rd, rm0);
2937     tcg_gen_shli_i64(rd, rd, 8 << a->size);
2938     write_neon_element64(rd, a->vd, 0, MO_64);
2939     widenfn(rd, rm1);
2940     tcg_gen_shli_i64(rd, rd, 8 << a->size);
2941     write_neon_element64(rd, a->vd, 1, MO_64);
2942     return true;
2943 }
2944 
2945 static bool trans_VCVT_B16_F32(DisasContext *s, arg_2misc *a)
2946 {
2947     TCGv_ptr fpst;
2948     TCGv_i64 tmp;
2949     TCGv_i32 dst0, dst1;
2950 
2951     if (!dc_isar_feature(aa32_bf16, s)) {
2952         return false;
2953     }
2954 
2955     /* UNDEF accesses to D16-D31 if they don't exist. */
2956     if (!dc_isar_feature(aa32_simd_r32, s) &&
2957         ((a->vd | a->vm) & 0x10)) {
2958         return false;
2959     }
2960 
2961     if ((a->vm & 1) || (a->size != 1)) {
2962         return false;
2963     }
2964 
2965     if (!vfp_access_check(s)) {
2966         return true;
2967     }
2968 
2969     fpst = fpstatus_ptr(FPST_STD);
2970     tmp = tcg_temp_new_i64();
2971     dst0 = tcg_temp_new_i32();
2972     dst1 = tcg_temp_new_i32();
2973 
2974     read_neon_element64(tmp, a->vm, 0, MO_64);
2975     gen_helper_bfcvt_pair(dst0, tmp, fpst);
2976 
2977     read_neon_element64(tmp, a->vm, 1, MO_64);
2978     gen_helper_bfcvt_pair(dst1, tmp, fpst);
2979 
2980     write_neon_element32(dst0, a->vd, 0, MO_32);
2981     write_neon_element32(dst1, a->vd, 1, MO_32);
2982     return true;
2983 }
2984 
2985 static bool trans_VCVT_F16_F32(DisasContext *s, arg_2misc *a)
2986 {
2987     TCGv_ptr fpst;
2988     TCGv_i32 ahp, tmp, tmp2, tmp3;
2989 
2990     if (!arm_dc_feature(s, ARM_FEATURE_NEON) ||
2991         !dc_isar_feature(aa32_fp16_spconv, s)) {
2992         return false;
2993     }
2994 
2995     /* UNDEF accesses to D16-D31 if they don't exist. */
2996     if (!dc_isar_feature(aa32_simd_r32, s) &&
2997         ((a->vd | a->vm) & 0x10)) {
2998         return false;
2999     }
3000 
3001     if ((a->vm & 1) || (a->size != 1)) {
3002         return false;
3003     }
3004 
3005     if (!vfp_access_check(s)) {
3006         return true;
3007     }
3008 
3009     fpst = fpstatus_ptr(FPST_STD);
3010     ahp = get_ahp_flag();
3011     tmp = tcg_temp_new_i32();
3012     read_neon_element32(tmp, a->vm, 0, MO_32);
3013     gen_helper_vfp_fcvt_f32_to_f16(tmp, tmp, fpst, ahp);
3014     tmp2 = tcg_temp_new_i32();
3015     read_neon_element32(tmp2, a->vm, 1, MO_32);
3016     gen_helper_vfp_fcvt_f32_to_f16(tmp2, tmp2, fpst, ahp);
3017     tcg_gen_shli_i32(tmp2, tmp2, 16);
3018     tcg_gen_or_i32(tmp2, tmp2, tmp);
3019     read_neon_element32(tmp, a->vm, 2, MO_32);
3020     gen_helper_vfp_fcvt_f32_to_f16(tmp, tmp, fpst, ahp);
3021     tmp3 = tcg_temp_new_i32();
3022     read_neon_element32(tmp3, a->vm, 3, MO_32);
3023     write_neon_element32(tmp2, a->vd, 0, MO_32);
3024     gen_helper_vfp_fcvt_f32_to_f16(tmp3, tmp3, fpst, ahp);
3025     tcg_gen_shli_i32(tmp3, tmp3, 16);
3026     tcg_gen_or_i32(tmp3, tmp3, tmp);
3027     write_neon_element32(tmp3, a->vd, 1, MO_32);
3028     return true;
3029 }
3030 
3031 static bool trans_VCVT_F32_F16(DisasContext *s, arg_2misc *a)
3032 {
3033     TCGv_ptr fpst;
3034     TCGv_i32 ahp, tmp, tmp2, tmp3;
3035 
3036     if (!arm_dc_feature(s, ARM_FEATURE_NEON) ||
3037         !dc_isar_feature(aa32_fp16_spconv, s)) {
3038         return false;
3039     }
3040 
3041     /* UNDEF accesses to D16-D31 if they don't exist. */
3042     if (!dc_isar_feature(aa32_simd_r32, s) &&
3043         ((a->vd | a->vm) & 0x10)) {
3044         return false;
3045     }
3046 
3047     if ((a->vd & 1) || (a->size != 1)) {
3048         return false;
3049     }
3050 
3051     if (!vfp_access_check(s)) {
3052         return true;
3053     }
3054 
3055     fpst = fpstatus_ptr(FPST_STD);
3056     ahp = get_ahp_flag();
3057     tmp3 = tcg_temp_new_i32();
3058     tmp2 = tcg_temp_new_i32();
3059     tmp = tcg_temp_new_i32();
3060     read_neon_element32(tmp, a->vm, 0, MO_32);
3061     read_neon_element32(tmp2, a->vm, 1, MO_32);
3062     tcg_gen_ext16u_i32(tmp3, tmp);
3063     gen_helper_vfp_fcvt_f16_to_f32(tmp3, tmp3, fpst, ahp);
3064     write_neon_element32(tmp3, a->vd, 0, MO_32);
3065     tcg_gen_shri_i32(tmp, tmp, 16);
3066     gen_helper_vfp_fcvt_f16_to_f32(tmp, tmp, fpst, ahp);
3067     write_neon_element32(tmp, a->vd, 1, MO_32);
3068     tcg_gen_ext16u_i32(tmp3, tmp2);
3069     gen_helper_vfp_fcvt_f16_to_f32(tmp3, tmp3, fpst, ahp);
3070     write_neon_element32(tmp3, a->vd, 2, MO_32);
3071     tcg_gen_shri_i32(tmp2, tmp2, 16);
3072     gen_helper_vfp_fcvt_f16_to_f32(tmp2, tmp2, fpst, ahp);
3073     write_neon_element32(tmp2, a->vd, 3, MO_32);
3074     return true;
3075 }
3076 
3077 static bool do_2misc_vec(DisasContext *s, arg_2misc *a, GVecGen2Fn *fn)
3078 {
3079     int vec_size = a->q ? 16 : 8;
3080     int rd_ofs = neon_full_reg_offset(a->vd);
3081     int rm_ofs = neon_full_reg_offset(a->vm);
3082 
3083     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
3084         return false;
3085     }
3086 
3087     /* UNDEF accesses to D16-D31 if they don't exist. */
3088     if (!dc_isar_feature(aa32_simd_r32, s) &&
3089         ((a->vd | a->vm) & 0x10)) {
3090         return false;
3091     }
3092 
3093     if (a->size == 3) {
3094         return false;
3095     }
3096 
3097     if ((a->vd | a->vm) & a->q) {
3098         return false;
3099     }
3100 
3101     if (!vfp_access_check(s)) {
3102         return true;
3103     }
3104 
3105     fn(a->size, rd_ofs, rm_ofs, vec_size, vec_size);
3106 
3107     return true;
3108 }
3109 
3110 #define DO_2MISC_VEC(INSN, FN)                                  \
3111     static bool trans_##INSN(DisasContext *s, arg_2misc *a)     \
3112     {                                                           \
3113         return do_2misc_vec(s, a, FN);                          \
3114     }
3115 
3116 DO_2MISC_VEC(VNEG, tcg_gen_gvec_neg)
3117 DO_2MISC_VEC(VABS, tcg_gen_gvec_abs)
3118 DO_2MISC_VEC(VCEQ0, gen_gvec_ceq0)
3119 DO_2MISC_VEC(VCGT0, gen_gvec_cgt0)
3120 DO_2MISC_VEC(VCLE0, gen_gvec_cle0)
3121 DO_2MISC_VEC(VCGE0, gen_gvec_cge0)
3122 DO_2MISC_VEC(VCLT0, gen_gvec_clt0)
3123 
3124 static bool trans_VMVN(DisasContext *s, arg_2misc *a)
3125 {
3126     if (a->size != 0) {
3127         return false;
3128     }
3129     return do_2misc_vec(s, a, tcg_gen_gvec_not);
3130 }
3131 
3132 #define WRAP_2M_3_OOL_FN(WRAPNAME, FUNC, DATA)                          \
3133     static void WRAPNAME(unsigned vece, uint32_t rd_ofs,                \
3134                          uint32_t rm_ofs, uint32_t oprsz,               \
3135                          uint32_t maxsz)                                \
3136     {                                                                   \
3137         tcg_gen_gvec_3_ool(rd_ofs, rd_ofs, rm_ofs, oprsz, maxsz,        \
3138                            DATA, FUNC);                                 \
3139     }
3140 
3141 #define WRAP_2M_2_OOL_FN(WRAPNAME, FUNC, DATA)                          \
3142     static void WRAPNAME(unsigned vece, uint32_t rd_ofs,                \
3143                          uint32_t rm_ofs, uint32_t oprsz,               \
3144                          uint32_t maxsz)                                \
3145     {                                                                   \
3146         tcg_gen_gvec_2_ool(rd_ofs, rm_ofs, oprsz, maxsz, DATA, FUNC);   \
3147     }
3148 
3149 WRAP_2M_3_OOL_FN(gen_AESE, gen_helper_crypto_aese, 0)
3150 WRAP_2M_3_OOL_FN(gen_AESD, gen_helper_crypto_aesd, 0)
3151 WRAP_2M_2_OOL_FN(gen_AESMC, gen_helper_crypto_aesmc, 0)
3152 WRAP_2M_2_OOL_FN(gen_AESIMC, gen_helper_crypto_aesimc, 0)
3153 WRAP_2M_2_OOL_FN(gen_SHA1H, gen_helper_crypto_sha1h, 0)
3154 WRAP_2M_2_OOL_FN(gen_SHA1SU1, gen_helper_crypto_sha1su1, 0)
3155 WRAP_2M_2_OOL_FN(gen_SHA256SU0, gen_helper_crypto_sha256su0, 0)
3156 
3157 #define DO_2M_CRYPTO(INSN, FEATURE, SIZE)                       \
3158     static bool trans_##INSN(DisasContext *s, arg_2misc *a)     \
3159     {                                                           \
3160         if (!dc_isar_feature(FEATURE, s) || a->size != SIZE) {  \
3161             return false;                                       \
3162         }                                                       \
3163         return do_2misc_vec(s, a, gen_##INSN);                  \
3164     }
3165 
3166 DO_2M_CRYPTO(AESE, aa32_aes, 0)
3167 DO_2M_CRYPTO(AESD, aa32_aes, 0)
3168 DO_2M_CRYPTO(AESMC, aa32_aes, 0)
3169 DO_2M_CRYPTO(AESIMC, aa32_aes, 0)
3170 DO_2M_CRYPTO(SHA1H, aa32_sha1, 2)
3171 DO_2M_CRYPTO(SHA1SU1, aa32_sha1, 2)
3172 DO_2M_CRYPTO(SHA256SU0, aa32_sha2, 2)
3173 
3174 static bool do_2misc(DisasContext *s, arg_2misc *a, NeonGenOneOpFn *fn)
3175 {
3176     TCGv_i32 tmp;
3177     int pass;
3178 
3179     /* Handle a 2-reg-misc operation by iterating 32 bits at a time */
3180     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
3181         return false;
3182     }
3183 
3184     /* UNDEF accesses to D16-D31 if they don't exist. */
3185     if (!dc_isar_feature(aa32_simd_r32, s) &&
3186         ((a->vd | a->vm) & 0x10)) {
3187         return false;
3188     }
3189 
3190     if (!fn) {
3191         return false;
3192     }
3193 
3194     if ((a->vd | a->vm) & a->q) {
3195         return false;
3196     }
3197 
3198     if (!vfp_access_check(s)) {
3199         return true;
3200     }
3201 
3202     tmp = tcg_temp_new_i32();
3203     for (pass = 0; pass < (a->q ? 4 : 2); pass++) {
3204         read_neon_element32(tmp, a->vm, pass, MO_32);
3205         fn(tmp, tmp);
3206         write_neon_element32(tmp, a->vd, pass, MO_32);
3207     }
3208     return true;
3209 }
3210 
3211 static bool trans_VREV32(DisasContext *s, arg_2misc *a)
3212 {
3213     static NeonGenOneOpFn * const fn[] = {
3214         tcg_gen_bswap32_i32,
3215         gen_swap_half,
3216         NULL,
3217         NULL,
3218     };
3219     return do_2misc(s, a, fn[a->size]);
3220 }
3221 
3222 static bool trans_VREV16(DisasContext *s, arg_2misc *a)
3223 {
3224     if (a->size != 0) {
3225         return false;
3226     }
3227     return do_2misc(s, a, gen_rev16);
3228 }
3229 
3230 static bool trans_VCLS(DisasContext *s, arg_2misc *a)
3231 {
3232     static NeonGenOneOpFn * const fn[] = {
3233         gen_helper_neon_cls_s8,
3234         gen_helper_neon_cls_s16,
3235         gen_helper_neon_cls_s32,
3236         NULL,
3237     };
3238     return do_2misc(s, a, fn[a->size]);
3239 }
3240 
3241 static void do_VCLZ_32(TCGv_i32 rd, TCGv_i32 rm)
3242 {
3243     tcg_gen_clzi_i32(rd, rm, 32);
3244 }
3245 
3246 static bool trans_VCLZ(DisasContext *s, arg_2misc *a)
3247 {
3248     static NeonGenOneOpFn * const fn[] = {
3249         gen_helper_neon_clz_u8,
3250         gen_helper_neon_clz_u16,
3251         do_VCLZ_32,
3252         NULL,
3253     };
3254     return do_2misc(s, a, fn[a->size]);
3255 }
3256 
3257 static bool trans_VCNT(DisasContext *s, arg_2misc *a)
3258 {
3259     if (a->size != 0) {
3260         return false;
3261     }
3262     return do_2misc(s, a, gen_helper_neon_cnt_u8);
3263 }
3264 
3265 static void gen_VABS_F(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3266                        uint32_t oprsz, uint32_t maxsz)
3267 {
3268     tcg_gen_gvec_andi(vece, rd_ofs, rm_ofs,
3269                       vece == MO_16 ? 0x7fff : 0x7fffffff,
3270                       oprsz, maxsz);
3271 }
3272 
3273 static bool trans_VABS_F(DisasContext *s, arg_2misc *a)
3274 {
3275     if (a->size == MO_16) {
3276         if (!dc_isar_feature(aa32_fp16_arith, s)) {
3277             return false;
3278         }
3279     } else if (a->size != MO_32) {
3280         return false;
3281     }
3282     return do_2misc_vec(s, a, gen_VABS_F);
3283 }
3284 
3285 static void gen_VNEG_F(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3286                        uint32_t oprsz, uint32_t maxsz)
3287 {
3288     tcg_gen_gvec_xori(vece, rd_ofs, rm_ofs,
3289                       vece == MO_16 ? 0x8000 : 0x80000000,
3290                       oprsz, maxsz);
3291 }
3292 
3293 static bool trans_VNEG_F(DisasContext *s, arg_2misc *a)
3294 {
3295     if (a->size == MO_16) {
3296         if (!dc_isar_feature(aa32_fp16_arith, s)) {
3297             return false;
3298         }
3299     } else if (a->size != MO_32) {
3300         return false;
3301     }
3302     return do_2misc_vec(s, a, gen_VNEG_F);
3303 }
3304 
3305 static bool trans_VRECPE(DisasContext *s, arg_2misc *a)
3306 {
3307     if (a->size != 2) {
3308         return false;
3309     }
3310     return do_2misc(s, a, gen_helper_recpe_u32);
3311 }
3312 
3313 static bool trans_VRSQRTE(DisasContext *s, arg_2misc *a)
3314 {
3315     if (a->size != 2) {
3316         return false;
3317     }
3318     return do_2misc(s, a, gen_helper_rsqrte_u32);
3319 }
3320 
3321 #define WRAP_1OP_ENV_FN(WRAPNAME, FUNC) \
3322     static void WRAPNAME(TCGv_i32 d, TCGv_i32 m)        \
3323     {                                                   \
3324         FUNC(d, tcg_env, m);                            \
3325     }
3326 
3327 WRAP_1OP_ENV_FN(gen_VQABS_s8, gen_helper_neon_qabs_s8)
3328 WRAP_1OP_ENV_FN(gen_VQABS_s16, gen_helper_neon_qabs_s16)
3329 WRAP_1OP_ENV_FN(gen_VQABS_s32, gen_helper_neon_qabs_s32)
3330 WRAP_1OP_ENV_FN(gen_VQNEG_s8, gen_helper_neon_qneg_s8)
3331 WRAP_1OP_ENV_FN(gen_VQNEG_s16, gen_helper_neon_qneg_s16)
3332 WRAP_1OP_ENV_FN(gen_VQNEG_s32, gen_helper_neon_qneg_s32)
3333 
3334 static bool trans_VQABS(DisasContext *s, arg_2misc *a)
3335 {
3336     static NeonGenOneOpFn * const fn[] = {
3337         gen_VQABS_s8,
3338         gen_VQABS_s16,
3339         gen_VQABS_s32,
3340         NULL,
3341     };
3342     return do_2misc(s, a, fn[a->size]);
3343 }
3344 
3345 static bool trans_VQNEG(DisasContext *s, arg_2misc *a)
3346 {
3347     static NeonGenOneOpFn * const fn[] = {
3348         gen_VQNEG_s8,
3349         gen_VQNEG_s16,
3350         gen_VQNEG_s32,
3351         NULL,
3352     };
3353     return do_2misc(s, a, fn[a->size]);
3354 }
3355 
3356 #define DO_2MISC_FP_VEC(INSN, HFUNC, SFUNC)                             \
3357     static void gen_##INSN(unsigned vece, uint32_t rd_ofs,              \
3358                            uint32_t rm_ofs,                             \
3359                            uint32_t oprsz, uint32_t maxsz)              \
3360     {                                                                   \
3361         static gen_helper_gvec_2_ptr * const fns[4] = {                 \
3362             NULL, HFUNC, SFUNC, NULL,                                   \
3363         };                                                              \
3364         TCGv_ptr fpst;                                                  \
3365         fpst = fpstatus_ptr(vece == MO_16 ? FPST_STD_F16 : FPST_STD);   \
3366         tcg_gen_gvec_2_ptr(rd_ofs, rm_ofs, fpst, oprsz, maxsz, 0,       \
3367                            fns[vece]);                                  \
3368     }                                                                   \
3369     static bool trans_##INSN(DisasContext *s, arg_2misc *a)             \
3370     {                                                                   \
3371         if (a->size == MO_16) {                                         \
3372             if (!dc_isar_feature(aa32_fp16_arith, s)) {                 \
3373                 return false;                                           \
3374             }                                                           \
3375         } else if (a->size != MO_32) {                                  \
3376             return false;                                               \
3377         }                                                               \
3378         return do_2misc_vec(s, a, gen_##INSN);                          \
3379     }
3380 
3381 DO_2MISC_FP_VEC(VRECPE_F, gen_helper_gvec_frecpe_h, gen_helper_gvec_frecpe_s)
3382 DO_2MISC_FP_VEC(VRSQRTE_F, gen_helper_gvec_frsqrte_h, gen_helper_gvec_frsqrte_s)
3383 DO_2MISC_FP_VEC(VCGT0_F, gen_helper_gvec_fcgt0_h, gen_helper_gvec_fcgt0_s)
3384 DO_2MISC_FP_VEC(VCGE0_F, gen_helper_gvec_fcge0_h, gen_helper_gvec_fcge0_s)
3385 DO_2MISC_FP_VEC(VCEQ0_F, gen_helper_gvec_fceq0_h, gen_helper_gvec_fceq0_s)
3386 DO_2MISC_FP_VEC(VCLT0_F, gen_helper_gvec_fclt0_h, gen_helper_gvec_fclt0_s)
3387 DO_2MISC_FP_VEC(VCLE0_F, gen_helper_gvec_fcle0_h, gen_helper_gvec_fcle0_s)
3388 DO_2MISC_FP_VEC(VCVT_FS, gen_helper_gvec_sstoh, gen_helper_gvec_sitos)
3389 DO_2MISC_FP_VEC(VCVT_FU, gen_helper_gvec_ustoh, gen_helper_gvec_uitos)
3390 DO_2MISC_FP_VEC(VCVT_SF, gen_helper_gvec_tosszh, gen_helper_gvec_tosizs)
3391 DO_2MISC_FP_VEC(VCVT_UF, gen_helper_gvec_touszh, gen_helper_gvec_touizs)
3392 
3393 DO_2MISC_FP_VEC(VRINTX_impl, gen_helper_gvec_vrintx_h, gen_helper_gvec_vrintx_s)
3394 
3395 static bool trans_VRINTX(DisasContext *s, arg_2misc *a)
3396 {
3397     if (!arm_dc_feature(s, ARM_FEATURE_V8)) {
3398         return false;
3399     }
3400     return trans_VRINTX_impl(s, a);
3401 }
3402 
3403 #define DO_VEC_RMODE(INSN, RMODE, OP)                                   \
3404     static void gen_##INSN(unsigned vece, uint32_t rd_ofs,              \
3405                            uint32_t rm_ofs,                             \
3406                            uint32_t oprsz, uint32_t maxsz)              \
3407     {                                                                   \
3408         static gen_helper_gvec_2_ptr * const fns[4] = {                 \
3409             NULL,                                                       \
3410             gen_helper_gvec_##OP##h,                                    \
3411             gen_helper_gvec_##OP##s,                                    \
3412             NULL,                                                       \
3413         };                                                              \
3414         TCGv_ptr fpst;                                                  \
3415         fpst = fpstatus_ptr(vece == 1 ? FPST_STD_F16 : FPST_STD);       \
3416         tcg_gen_gvec_2_ptr(rd_ofs, rm_ofs, fpst, oprsz, maxsz,          \
3417                            arm_rmode_to_sf(RMODE), fns[vece]);          \
3418     }                                                                   \
3419     static bool trans_##INSN(DisasContext *s, arg_2misc *a)             \
3420     {                                                                   \
3421         if (!arm_dc_feature(s, ARM_FEATURE_V8)) {                       \
3422             return false;                                               \
3423         }                                                               \
3424         if (a->size == MO_16) {                                         \
3425             if (!dc_isar_feature(aa32_fp16_arith, s)) {                 \
3426                 return false;                                           \
3427             }                                                           \
3428         } else if (a->size != MO_32) {                                  \
3429             return false;                                               \
3430         }                                                               \
3431         return do_2misc_vec(s, a, gen_##INSN);                          \
3432     }
3433 
3434 DO_VEC_RMODE(VCVTAU, FPROUNDING_TIEAWAY, vcvt_rm_u)
3435 DO_VEC_RMODE(VCVTAS, FPROUNDING_TIEAWAY, vcvt_rm_s)
3436 DO_VEC_RMODE(VCVTNU, FPROUNDING_TIEEVEN, vcvt_rm_u)
3437 DO_VEC_RMODE(VCVTNS, FPROUNDING_TIEEVEN, vcvt_rm_s)
3438 DO_VEC_RMODE(VCVTPU, FPROUNDING_POSINF, vcvt_rm_u)
3439 DO_VEC_RMODE(VCVTPS, FPROUNDING_POSINF, vcvt_rm_s)
3440 DO_VEC_RMODE(VCVTMU, FPROUNDING_NEGINF, vcvt_rm_u)
3441 DO_VEC_RMODE(VCVTMS, FPROUNDING_NEGINF, vcvt_rm_s)
3442 
3443 DO_VEC_RMODE(VRINTN, FPROUNDING_TIEEVEN, vrint_rm_)
3444 DO_VEC_RMODE(VRINTA, FPROUNDING_TIEAWAY, vrint_rm_)
3445 DO_VEC_RMODE(VRINTZ, FPROUNDING_ZERO, vrint_rm_)
3446 DO_VEC_RMODE(VRINTM, FPROUNDING_NEGINF, vrint_rm_)
3447 DO_VEC_RMODE(VRINTP, FPROUNDING_POSINF, vrint_rm_)
3448 
3449 static bool trans_VSWP(DisasContext *s, arg_2misc *a)
3450 {
3451     TCGv_i64 rm, rd;
3452     int pass;
3453 
3454     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
3455         return false;
3456     }
3457 
3458     /* UNDEF accesses to D16-D31 if they don't exist. */
3459     if (!dc_isar_feature(aa32_simd_r32, s) &&
3460         ((a->vd | a->vm) & 0x10)) {
3461         return false;
3462     }
3463 
3464     if (a->size != 0) {
3465         return false;
3466     }
3467 
3468     if ((a->vd | a->vm) & a->q) {
3469         return false;
3470     }
3471 
3472     if (!vfp_access_check(s)) {
3473         return true;
3474     }
3475 
3476     rm = tcg_temp_new_i64();
3477     rd = tcg_temp_new_i64();
3478     for (pass = 0; pass < (a->q ? 2 : 1); pass++) {
3479         read_neon_element64(rm, a->vm, pass, MO_64);
3480         read_neon_element64(rd, a->vd, pass, MO_64);
3481         write_neon_element64(rm, a->vd, pass, MO_64);
3482         write_neon_element64(rd, a->vm, pass, MO_64);
3483     }
3484     return true;
3485 }
3486 
3487 static void gen_neon_trn_u8(TCGv_i32 t0, TCGv_i32 t1)
3488 {
3489     TCGv_i32 rd, tmp;
3490 
3491     rd = tcg_temp_new_i32();
3492     tmp = tcg_temp_new_i32();
3493 
3494     tcg_gen_shli_i32(rd, t0, 8);
3495     tcg_gen_andi_i32(rd, rd, 0xff00ff00);
3496     tcg_gen_andi_i32(tmp, t1, 0x00ff00ff);
3497     tcg_gen_or_i32(rd, rd, tmp);
3498 
3499     tcg_gen_shri_i32(t1, t1, 8);
3500     tcg_gen_andi_i32(t1, t1, 0x00ff00ff);
3501     tcg_gen_andi_i32(tmp, t0, 0xff00ff00);
3502     tcg_gen_or_i32(t1, t1, tmp);
3503     tcg_gen_mov_i32(t0, rd);
3504 }
3505 
3506 static void gen_neon_trn_u16(TCGv_i32 t0, TCGv_i32 t1)
3507 {
3508     TCGv_i32 rd, tmp;
3509 
3510     rd = tcg_temp_new_i32();
3511     tmp = tcg_temp_new_i32();
3512 
3513     tcg_gen_shli_i32(rd, t0, 16);
3514     tcg_gen_andi_i32(tmp, t1, 0xffff);
3515     tcg_gen_or_i32(rd, rd, tmp);
3516     tcg_gen_shri_i32(t1, t1, 16);
3517     tcg_gen_andi_i32(tmp, t0, 0xffff0000);
3518     tcg_gen_or_i32(t1, t1, tmp);
3519     tcg_gen_mov_i32(t0, rd);
3520 }
3521 
3522 static bool trans_VTRN(DisasContext *s, arg_2misc *a)
3523 {
3524     TCGv_i32 tmp, tmp2;
3525     int pass;
3526 
3527     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
3528         return false;
3529     }
3530 
3531     /* UNDEF accesses to D16-D31 if they don't exist. */
3532     if (!dc_isar_feature(aa32_simd_r32, s) &&
3533         ((a->vd | a->vm) & 0x10)) {
3534         return false;
3535     }
3536 
3537     if ((a->vd | a->vm) & a->q) {
3538         return false;
3539     }
3540 
3541     if (a->size == 3) {
3542         return false;
3543     }
3544 
3545     if (!vfp_access_check(s)) {
3546         return true;
3547     }
3548 
3549     tmp = tcg_temp_new_i32();
3550     tmp2 = tcg_temp_new_i32();
3551     if (a->size == MO_32) {
3552         for (pass = 0; pass < (a->q ? 4 : 2); pass += 2) {
3553             read_neon_element32(tmp, a->vm, pass, MO_32);
3554             read_neon_element32(tmp2, a->vd, pass + 1, MO_32);
3555             write_neon_element32(tmp2, a->vm, pass, MO_32);
3556             write_neon_element32(tmp, a->vd, pass + 1, MO_32);
3557         }
3558     } else {
3559         for (pass = 0; pass < (a->q ? 4 : 2); pass++) {
3560             read_neon_element32(tmp, a->vm, pass, MO_32);
3561             read_neon_element32(tmp2, a->vd, pass, MO_32);
3562             if (a->size == MO_8) {
3563                 gen_neon_trn_u8(tmp, tmp2);
3564             } else {
3565                 gen_neon_trn_u16(tmp, tmp2);
3566             }
3567             write_neon_element32(tmp2, a->vm, pass, MO_32);
3568             write_neon_element32(tmp, a->vd, pass, MO_32);
3569         }
3570     }
3571     return true;
3572 }
3573 
3574 static bool trans_VSMMLA(DisasContext *s, arg_VSMMLA *a)
3575 {
3576     if (!dc_isar_feature(aa32_i8mm, s)) {
3577         return false;
3578     }
3579     return do_neon_ddda(s, 7, a->vd, a->vn, a->vm, 0,
3580                         gen_helper_gvec_smmla_b);
3581 }
3582 
3583 static bool trans_VUMMLA(DisasContext *s, arg_VUMMLA *a)
3584 {
3585     if (!dc_isar_feature(aa32_i8mm, s)) {
3586         return false;
3587     }
3588     return do_neon_ddda(s, 7, a->vd, a->vn, a->vm, 0,
3589                         gen_helper_gvec_ummla_b);
3590 }
3591 
3592 static bool trans_VUSMMLA(DisasContext *s, arg_VUSMMLA *a)
3593 {
3594     if (!dc_isar_feature(aa32_i8mm, s)) {
3595         return false;
3596     }
3597     return do_neon_ddda(s, 7, a->vd, a->vn, a->vm, 0,
3598                         gen_helper_gvec_usmmla_b);
3599 }
3600 
3601 static bool trans_VMMLA_b16(DisasContext *s, arg_VMMLA_b16 *a)
3602 {
3603     if (!dc_isar_feature(aa32_bf16, s)) {
3604         return false;
3605     }
3606     return do_neon_ddda_env(s, 7, a->vd, a->vn, a->vm, 0,
3607                             gen_helper_gvec_bfmmla);
3608 }
3609 
3610 static bool trans_VFMA_b16(DisasContext *s, arg_VFMA_b16 *a)
3611 {
3612     if (!dc_isar_feature(aa32_bf16, s)) {
3613         return false;
3614     }
3615     return do_neon_ddda_fpst(s, 7, a->vd, a->vn, a->vm, a->q, FPST_STD,
3616                              gen_helper_gvec_bfmlal);
3617 }
3618 
3619 static bool trans_VFMA_b16_scal(DisasContext *s, arg_VFMA_b16_scal *a)
3620 {
3621     if (!dc_isar_feature(aa32_bf16, s)) {
3622         return false;
3623     }
3624     return do_neon_ddda_fpst(s, 6, a->vd, a->vn, a->vm,
3625                              (a->index << 1) | a->q, FPST_STD,
3626                              gen_helper_gvec_bfmlal_idx);
3627 }
3628