xref: /qemu/target/arm/tcg/translate-mve.c (revision 075e7e97e3a042854b8ea2827559891a577b4a6b)
1 /*
2  *  ARM translation: M-profile MVE instructions
3  *
4  *  Copyright (c) 2021 Linaro, Ltd.
5  *
6  * This library is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU Lesser General Public
8  * License as published by the Free Software Foundation; either
9  * version 2.1 of the License, or (at your option) any later version.
10  *
11  * This library is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14  * Lesser General Public License for more details.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
18  */
19 
20 #include "qemu/osdep.h"
21 #include "tcg/tcg-op.h"
22 #include "tcg/tcg-op-gvec.h"
23 #include "exec/exec-all.h"
24 #include "exec/gen-icount.h"
25 #include "translate.h"
26 #include "translate-a32.h"
27 
28 static inline int vidup_imm(DisasContext *s, int x)
29 {
30     return 1 << x;
31 }
32 
33 /* Include the generated decoder */
34 #include "decode-mve.c.inc"
35 
36 typedef void MVEGenLdStFn(TCGv_ptr, TCGv_ptr, TCGv_i32);
37 typedef void MVEGenLdStSGFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32);
38 typedef void MVEGenLdStIlFn(TCGv_ptr, TCGv_i32, TCGv_i32);
39 typedef void MVEGenOneOpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr);
40 typedef void MVEGenTwoOpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_ptr);
41 typedef void MVEGenTwoOpScalarFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32);
42 typedef void MVEGenTwoOpShiftFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32);
43 typedef void MVEGenLongDualAccOpFn(TCGv_i64, TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i64);
44 typedef void MVEGenVADDVFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32);
45 typedef void MVEGenOneOpImmFn(TCGv_ptr, TCGv_ptr, TCGv_i64);
46 typedef void MVEGenVIDUPFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32, TCGv_i32);
47 typedef void MVEGenVIWDUPFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_i32, TCGv_i32, TCGv_i32);
48 typedef void MVEGenCmpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr);
49 typedef void MVEGenScalarCmpFn(TCGv_ptr, TCGv_ptr, TCGv_i32);
50 typedef void MVEGenVABAVFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32);
51 typedef void MVEGenDualAccOpFn(TCGv_i32, TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32);
52 
53 /* Return the offset of a Qn register (same semantics as aa32_vfp_qreg()) */
54 static inline long mve_qreg_offset(unsigned reg)
55 {
56     return offsetof(CPUARMState, vfp.zregs[reg].d[0]);
57 }
58 
59 static TCGv_ptr mve_qreg_ptr(unsigned reg)
60 {
61     TCGv_ptr ret = tcg_temp_new_ptr();
62     tcg_gen_addi_ptr(ret, cpu_env, mve_qreg_offset(reg));
63     return ret;
64 }
65 
66 static bool mve_check_qreg_bank(DisasContext *s, int qmask)
67 {
68     /*
69      * Check whether Qregs are in range. For v8.1M only Q0..Q7
70      * are supported, see VFPSmallRegisterBank().
71      */
72     return qmask < 8;
73 }
74 
75 bool mve_eci_check(DisasContext *s)
76 {
77     /*
78      * This is a beatwise insn: check that ECI is valid (not a
79      * reserved value) and note that we are handling it.
80      * Return true if OK, false if we generated an exception.
81      */
82     s->eci_handled = true;
83     switch (s->eci) {
84     case ECI_NONE:
85     case ECI_A0:
86     case ECI_A0A1:
87     case ECI_A0A1A2:
88     case ECI_A0A1A2B0:
89         return true;
90     default:
91         /* Reserved value: INVSTATE UsageFault */
92         gen_exception_insn(s, s->pc_curr, EXCP_INVSTATE, syn_uncategorized(),
93                            default_exception_el(s));
94         return false;
95     }
96 }
97 
98 void mve_update_eci(DisasContext *s)
99 {
100     /*
101      * The helper function will always update the CPUState field,
102      * so we only need to update the DisasContext field.
103      */
104     if (s->eci) {
105         s->eci = (s->eci == ECI_A0A1A2B0) ? ECI_A0 : ECI_NONE;
106     }
107 }
108 
109 void mve_update_and_store_eci(DisasContext *s)
110 {
111     /*
112      * For insns which don't call a helper function that will call
113      * mve_advance_vpt(), this version updates s->eci and also stores
114      * it out to the CPUState field.
115      */
116     if (s->eci) {
117         mve_update_eci(s);
118         store_cpu_field(tcg_constant_i32(s->eci << 4), condexec_bits);
119     }
120 }
121 
122 static bool mve_skip_first_beat(DisasContext *s)
123 {
124     /* Return true if PSR.ECI says we must skip the first beat of this insn */
125     switch (s->eci) {
126     case ECI_NONE:
127         return false;
128     case ECI_A0:
129     case ECI_A0A1:
130     case ECI_A0A1A2:
131     case ECI_A0A1A2B0:
132         return true;
133     default:
134         g_assert_not_reached();
135     }
136 }
137 
138 static bool do_ldst(DisasContext *s, arg_VLDR_VSTR *a, MVEGenLdStFn *fn,
139                     unsigned msize)
140 {
141     TCGv_i32 addr;
142     uint32_t offset;
143     TCGv_ptr qreg;
144 
145     if (!dc_isar_feature(aa32_mve, s) ||
146         !mve_check_qreg_bank(s, a->qd) ||
147         !fn) {
148         return false;
149     }
150 
151     /* CONSTRAINED UNPREDICTABLE: we choose to UNDEF */
152     if (a->rn == 15 || (a->rn == 13 && a->w)) {
153         return false;
154     }
155 
156     if (!mve_eci_check(s) || !vfp_access_check(s)) {
157         return true;
158     }
159 
160     offset = a->imm << msize;
161     if (!a->a) {
162         offset = -offset;
163     }
164     addr = load_reg(s, a->rn);
165     if (a->p) {
166         tcg_gen_addi_i32(addr, addr, offset);
167     }
168 
169     qreg = mve_qreg_ptr(a->qd);
170     fn(cpu_env, qreg, addr);
171     tcg_temp_free_ptr(qreg);
172 
173     /*
174      * Writeback always happens after the last beat of the insn,
175      * regardless of predication
176      */
177     if (a->w) {
178         if (!a->p) {
179             tcg_gen_addi_i32(addr, addr, offset);
180         }
181         store_reg(s, a->rn, addr);
182     } else {
183         tcg_temp_free_i32(addr);
184     }
185     mve_update_eci(s);
186     return true;
187 }
188 
189 static bool trans_VLDR_VSTR(DisasContext *s, arg_VLDR_VSTR *a)
190 {
191     static MVEGenLdStFn * const ldstfns[4][2] = {
192         { gen_helper_mve_vstrb, gen_helper_mve_vldrb },
193         { gen_helper_mve_vstrh, gen_helper_mve_vldrh },
194         { gen_helper_mve_vstrw, gen_helper_mve_vldrw },
195         { NULL, NULL }
196     };
197     return do_ldst(s, a, ldstfns[a->size][a->l], a->size);
198 }
199 
200 #define DO_VLDST_WIDE_NARROW(OP, SLD, ULD, ST, MSIZE)           \
201     static bool trans_##OP(DisasContext *s, arg_VLDR_VSTR *a)   \
202     {                                                           \
203         static MVEGenLdStFn * const ldstfns[2][2] = {           \
204             { gen_helper_mve_##ST, gen_helper_mve_##SLD },      \
205             { NULL, gen_helper_mve_##ULD },                     \
206         };                                                      \
207         return do_ldst(s, a, ldstfns[a->u][a->l], MSIZE);       \
208     }
209 
210 DO_VLDST_WIDE_NARROW(VLDSTB_H, vldrb_sh, vldrb_uh, vstrb_h, MO_8)
211 DO_VLDST_WIDE_NARROW(VLDSTB_W, vldrb_sw, vldrb_uw, vstrb_w, MO_8)
212 DO_VLDST_WIDE_NARROW(VLDSTH_W, vldrh_sw, vldrh_uw, vstrh_w, MO_16)
213 
214 static bool do_ldst_sg(DisasContext *s, arg_vldst_sg *a, MVEGenLdStSGFn fn)
215 {
216     TCGv_i32 addr;
217     TCGv_ptr qd, qm;
218 
219     if (!dc_isar_feature(aa32_mve, s) ||
220         !mve_check_qreg_bank(s, a->qd | a->qm) ||
221         !fn || a->rn == 15) {
222         /* Rn case is UNPREDICTABLE */
223         return false;
224     }
225 
226     if (!mve_eci_check(s) || !vfp_access_check(s)) {
227         return true;
228     }
229 
230     addr = load_reg(s, a->rn);
231 
232     qd = mve_qreg_ptr(a->qd);
233     qm = mve_qreg_ptr(a->qm);
234     fn(cpu_env, qd, qm, addr);
235     tcg_temp_free_ptr(qd);
236     tcg_temp_free_ptr(qm);
237     tcg_temp_free_i32(addr);
238     mve_update_eci(s);
239     return true;
240 }
241 
242 /*
243  * The naming scheme here is "vldrb_sg_sh == in-memory byte loads
244  * signextended to halfword elements in register". _os_ indicates that
245  * the offsets in Qm should be scaled by the element size.
246  */
247 /* This macro is just to make the arrays more compact in these functions */
248 #define F(N) gen_helper_mve_##N
249 
250 /* VLDRB/VSTRB (ie msize 1) with OS=1 is UNPREDICTABLE; we UNDEF */
251 static bool trans_VLDR_S_sg(DisasContext *s, arg_vldst_sg *a)
252 {
253     static MVEGenLdStSGFn * const fns[2][4][4] = { {
254             { NULL, F(vldrb_sg_sh), F(vldrb_sg_sw), NULL },
255             { NULL, NULL,           F(vldrh_sg_sw), NULL },
256             { NULL, NULL,           NULL,           NULL },
257             { NULL, NULL,           NULL,           NULL }
258         }, {
259             { NULL, NULL,              NULL,              NULL },
260             { NULL, NULL,              F(vldrh_sg_os_sw), NULL },
261             { NULL, NULL,              NULL,              NULL },
262             { NULL, NULL,              NULL,              NULL }
263         }
264     };
265     if (a->qd == a->qm) {
266         return false; /* UNPREDICTABLE */
267     }
268     return do_ldst_sg(s, a, fns[a->os][a->msize][a->size]);
269 }
270 
271 static bool trans_VLDR_U_sg(DisasContext *s, arg_vldst_sg *a)
272 {
273     static MVEGenLdStSGFn * const fns[2][4][4] = { {
274             { F(vldrb_sg_ub), F(vldrb_sg_uh), F(vldrb_sg_uw), NULL },
275             { NULL,           F(vldrh_sg_uh), F(vldrh_sg_uw), NULL },
276             { NULL,           NULL,           F(vldrw_sg_uw), NULL },
277             { NULL,           NULL,           NULL,           F(vldrd_sg_ud) }
278         }, {
279             { NULL, NULL,              NULL,              NULL },
280             { NULL, F(vldrh_sg_os_uh), F(vldrh_sg_os_uw), NULL },
281             { NULL, NULL,              F(vldrw_sg_os_uw), NULL },
282             { NULL, NULL,              NULL,              F(vldrd_sg_os_ud) }
283         }
284     };
285     if (a->qd == a->qm) {
286         return false; /* UNPREDICTABLE */
287     }
288     return do_ldst_sg(s, a, fns[a->os][a->msize][a->size]);
289 }
290 
291 static bool trans_VSTR_sg(DisasContext *s, arg_vldst_sg *a)
292 {
293     static MVEGenLdStSGFn * const fns[2][4][4] = { {
294             { F(vstrb_sg_ub), F(vstrb_sg_uh), F(vstrb_sg_uw), NULL },
295             { NULL,           F(vstrh_sg_uh), F(vstrh_sg_uw), NULL },
296             { NULL,           NULL,           F(vstrw_sg_uw), NULL },
297             { NULL,           NULL,           NULL,           F(vstrd_sg_ud) }
298         }, {
299             { NULL, NULL,              NULL,              NULL },
300             { NULL, F(vstrh_sg_os_uh), F(vstrh_sg_os_uw), NULL },
301             { NULL, NULL,              F(vstrw_sg_os_uw), NULL },
302             { NULL, NULL,              NULL,              F(vstrd_sg_os_ud) }
303         }
304     };
305     return do_ldst_sg(s, a, fns[a->os][a->msize][a->size]);
306 }
307 
308 #undef F
309 
310 static bool do_ldst_sg_imm(DisasContext *s, arg_vldst_sg_imm *a,
311                            MVEGenLdStSGFn *fn, unsigned msize)
312 {
313     uint32_t offset;
314     TCGv_ptr qd, qm;
315 
316     if (!dc_isar_feature(aa32_mve, s) ||
317         !mve_check_qreg_bank(s, a->qd | a->qm) ||
318         !fn) {
319         return false;
320     }
321 
322     if (!mve_eci_check(s) || !vfp_access_check(s)) {
323         return true;
324     }
325 
326     offset = a->imm << msize;
327     if (!a->a) {
328         offset = -offset;
329     }
330 
331     qd = mve_qreg_ptr(a->qd);
332     qm = mve_qreg_ptr(a->qm);
333     fn(cpu_env, qd, qm, tcg_constant_i32(offset));
334     tcg_temp_free_ptr(qd);
335     tcg_temp_free_ptr(qm);
336     mve_update_eci(s);
337     return true;
338 }
339 
340 static bool trans_VLDRW_sg_imm(DisasContext *s, arg_vldst_sg_imm *a)
341 {
342     static MVEGenLdStSGFn * const fns[] = {
343         gen_helper_mve_vldrw_sg_uw,
344         gen_helper_mve_vldrw_sg_wb_uw,
345     };
346     if (a->qd == a->qm) {
347         return false; /* UNPREDICTABLE */
348     }
349     return do_ldst_sg_imm(s, a, fns[a->w], MO_32);
350 }
351 
352 static bool trans_VLDRD_sg_imm(DisasContext *s, arg_vldst_sg_imm *a)
353 {
354     static MVEGenLdStSGFn * const fns[] = {
355         gen_helper_mve_vldrd_sg_ud,
356         gen_helper_mve_vldrd_sg_wb_ud,
357     };
358     if (a->qd == a->qm) {
359         return false; /* UNPREDICTABLE */
360     }
361     return do_ldst_sg_imm(s, a, fns[a->w], MO_64);
362 }
363 
364 static bool trans_VSTRW_sg_imm(DisasContext *s, arg_vldst_sg_imm *a)
365 {
366     static MVEGenLdStSGFn * const fns[] = {
367         gen_helper_mve_vstrw_sg_uw,
368         gen_helper_mve_vstrw_sg_wb_uw,
369     };
370     return do_ldst_sg_imm(s, a, fns[a->w], MO_32);
371 }
372 
373 static bool trans_VSTRD_sg_imm(DisasContext *s, arg_vldst_sg_imm *a)
374 {
375     static MVEGenLdStSGFn * const fns[] = {
376         gen_helper_mve_vstrd_sg_ud,
377         gen_helper_mve_vstrd_sg_wb_ud,
378     };
379     return do_ldst_sg_imm(s, a, fns[a->w], MO_64);
380 }
381 
382 static bool do_vldst_il(DisasContext *s, arg_vldst_il *a, MVEGenLdStIlFn *fn,
383                         int addrinc)
384 {
385     TCGv_i32 rn;
386 
387     if (!dc_isar_feature(aa32_mve, s) ||
388         !mve_check_qreg_bank(s, a->qd) ||
389         !fn || (a->rn == 13 && a->w) || a->rn == 15) {
390         /* Variously UNPREDICTABLE or UNDEF or related-encoding */
391         return false;
392     }
393     if (!mve_eci_check(s) || !vfp_access_check(s)) {
394         return true;
395     }
396 
397     rn = load_reg(s, a->rn);
398     /*
399      * We pass the index of Qd, not a pointer, because the helper must
400      * access multiple Q registers starting at Qd and working up.
401      */
402     fn(cpu_env, tcg_constant_i32(a->qd), rn);
403 
404     if (a->w) {
405         tcg_gen_addi_i32(rn, rn, addrinc);
406         store_reg(s, a->rn, rn);
407     } else {
408         tcg_temp_free_i32(rn);
409     }
410     mve_update_and_store_eci(s);
411     return true;
412 }
413 
414 /* This macro is just to make the arrays more compact in these functions */
415 #define F(N) gen_helper_mve_##N
416 
417 static bool trans_VLD2(DisasContext *s, arg_vldst_il *a)
418 {
419     static MVEGenLdStIlFn * const fns[4][4] = {
420         { F(vld20b), F(vld20h), F(vld20w), NULL, },
421         { F(vld21b), F(vld21h), F(vld21w), NULL, },
422         { NULL, NULL, NULL, NULL },
423         { NULL, NULL, NULL, NULL },
424     };
425     if (a->qd > 6) {
426         return false;
427     }
428     return do_vldst_il(s, a, fns[a->pat][a->size], 32);
429 }
430 
431 static bool trans_VLD4(DisasContext *s, arg_vldst_il *a)
432 {
433     static MVEGenLdStIlFn * const fns[4][4] = {
434         { F(vld40b), F(vld40h), F(vld40w), NULL, },
435         { F(vld41b), F(vld41h), F(vld41w), NULL, },
436         { F(vld42b), F(vld42h), F(vld42w), NULL, },
437         { F(vld43b), F(vld43h), F(vld43w), NULL, },
438     };
439     if (a->qd > 4) {
440         return false;
441     }
442     return do_vldst_il(s, a, fns[a->pat][a->size], 64);
443 }
444 
445 static bool trans_VST2(DisasContext *s, arg_vldst_il *a)
446 {
447     static MVEGenLdStIlFn * const fns[4][4] = {
448         { F(vst20b), F(vst20h), F(vst20w), NULL, },
449         { F(vst21b), F(vst21h), F(vst21w), NULL, },
450         { NULL, NULL, NULL, NULL },
451         { NULL, NULL, NULL, NULL },
452     };
453     if (a->qd > 6) {
454         return false;
455     }
456     return do_vldst_il(s, a, fns[a->pat][a->size], 32);
457 }
458 
459 static bool trans_VST4(DisasContext *s, arg_vldst_il *a)
460 {
461     static MVEGenLdStIlFn * const fns[4][4] = {
462         { F(vst40b), F(vst40h), F(vst40w), NULL, },
463         { F(vst41b), F(vst41h), F(vst41w), NULL, },
464         { F(vst42b), F(vst42h), F(vst42w), NULL, },
465         { F(vst43b), F(vst43h), F(vst43w), NULL, },
466     };
467     if (a->qd > 4) {
468         return false;
469     }
470     return do_vldst_il(s, a, fns[a->pat][a->size], 64);
471 }
472 
473 #undef F
474 
475 static bool trans_VDUP(DisasContext *s, arg_VDUP *a)
476 {
477     TCGv_ptr qd;
478     TCGv_i32 rt;
479 
480     if (!dc_isar_feature(aa32_mve, s) ||
481         !mve_check_qreg_bank(s, a->qd)) {
482         return false;
483     }
484     if (a->rt == 13 || a->rt == 15) {
485         /* UNPREDICTABLE; we choose to UNDEF */
486         return false;
487     }
488     if (!mve_eci_check(s) || !vfp_access_check(s)) {
489         return true;
490     }
491 
492     qd = mve_qreg_ptr(a->qd);
493     rt = load_reg(s, a->rt);
494     tcg_gen_dup_i32(a->size, rt, rt);
495     gen_helper_mve_vdup(cpu_env, qd, rt);
496     tcg_temp_free_ptr(qd);
497     tcg_temp_free_i32(rt);
498     mve_update_eci(s);
499     return true;
500 }
501 
502 static bool do_1op(DisasContext *s, arg_1op *a, MVEGenOneOpFn fn)
503 {
504     TCGv_ptr qd, qm;
505 
506     if (!dc_isar_feature(aa32_mve, s) ||
507         !mve_check_qreg_bank(s, a->qd | a->qm) ||
508         !fn) {
509         return false;
510     }
511 
512     if (!mve_eci_check(s) || !vfp_access_check(s)) {
513         return true;
514     }
515 
516     qd = mve_qreg_ptr(a->qd);
517     qm = mve_qreg_ptr(a->qm);
518     fn(cpu_env, qd, qm);
519     tcg_temp_free_ptr(qd);
520     tcg_temp_free_ptr(qm);
521     mve_update_eci(s);
522     return true;
523 }
524 
525 #define DO_1OP(INSN, FN)                                        \
526     static bool trans_##INSN(DisasContext *s, arg_1op *a)       \
527     {                                                           \
528         static MVEGenOneOpFn * const fns[] = {                  \
529             gen_helper_mve_##FN##b,                             \
530             gen_helper_mve_##FN##h,                             \
531             gen_helper_mve_##FN##w,                             \
532             NULL,                                               \
533         };                                                      \
534         return do_1op(s, a, fns[a->size]);                      \
535     }
536 
537 DO_1OP(VCLZ, vclz)
538 DO_1OP(VCLS, vcls)
539 DO_1OP(VABS, vabs)
540 DO_1OP(VNEG, vneg)
541 DO_1OP(VQABS, vqabs)
542 DO_1OP(VQNEG, vqneg)
543 DO_1OP(VMAXA, vmaxa)
544 DO_1OP(VMINA, vmina)
545 
546 /* Narrowing moves: only size 0 and 1 are valid */
547 #define DO_VMOVN(INSN, FN) \
548     static bool trans_##INSN(DisasContext *s, arg_1op *a)       \
549     {                                                           \
550         static MVEGenOneOpFn * const fns[] = {                  \
551             gen_helper_mve_##FN##b,                             \
552             gen_helper_mve_##FN##h,                             \
553             NULL,                                               \
554             NULL,                                               \
555         };                                                      \
556         return do_1op(s, a, fns[a->size]);                      \
557     }
558 
559 DO_VMOVN(VMOVNB, vmovnb)
560 DO_VMOVN(VMOVNT, vmovnt)
561 DO_VMOVN(VQMOVUNB, vqmovunb)
562 DO_VMOVN(VQMOVUNT, vqmovunt)
563 DO_VMOVN(VQMOVN_BS, vqmovnbs)
564 DO_VMOVN(VQMOVN_TS, vqmovnts)
565 DO_VMOVN(VQMOVN_BU, vqmovnbu)
566 DO_VMOVN(VQMOVN_TU, vqmovntu)
567 
568 static bool trans_VREV16(DisasContext *s, arg_1op *a)
569 {
570     static MVEGenOneOpFn * const fns[] = {
571         gen_helper_mve_vrev16b,
572         NULL,
573         NULL,
574         NULL,
575     };
576     return do_1op(s, a, fns[a->size]);
577 }
578 
579 static bool trans_VREV32(DisasContext *s, arg_1op *a)
580 {
581     static MVEGenOneOpFn * const fns[] = {
582         gen_helper_mve_vrev32b,
583         gen_helper_mve_vrev32h,
584         NULL,
585         NULL,
586     };
587     return do_1op(s, a, fns[a->size]);
588 }
589 
590 static bool trans_VREV64(DisasContext *s, arg_1op *a)
591 {
592     static MVEGenOneOpFn * const fns[] = {
593         gen_helper_mve_vrev64b,
594         gen_helper_mve_vrev64h,
595         gen_helper_mve_vrev64w,
596         NULL,
597     };
598     return do_1op(s, a, fns[a->size]);
599 }
600 
601 static bool trans_VMVN(DisasContext *s, arg_1op *a)
602 {
603     return do_1op(s, a, gen_helper_mve_vmvn);
604 }
605 
606 static bool trans_VABS_fp(DisasContext *s, arg_1op *a)
607 {
608     static MVEGenOneOpFn * const fns[] = {
609         NULL,
610         gen_helper_mve_vfabsh,
611         gen_helper_mve_vfabss,
612         NULL,
613     };
614     if (!dc_isar_feature(aa32_mve_fp, s)) {
615         return false;
616     }
617     return do_1op(s, a, fns[a->size]);
618 }
619 
620 static bool trans_VNEG_fp(DisasContext *s, arg_1op *a)
621 {
622     static MVEGenOneOpFn * const fns[] = {
623         NULL,
624         gen_helper_mve_vfnegh,
625         gen_helper_mve_vfnegs,
626         NULL,
627     };
628     if (!dc_isar_feature(aa32_mve_fp, s)) {
629         return false;
630     }
631     return do_1op(s, a, fns[a->size]);
632 }
633 
634 static bool do_2op(DisasContext *s, arg_2op *a, MVEGenTwoOpFn fn)
635 {
636     TCGv_ptr qd, qn, qm;
637 
638     if (!dc_isar_feature(aa32_mve, s) ||
639         !mve_check_qreg_bank(s, a->qd | a->qn | a->qm) ||
640         !fn) {
641         return false;
642     }
643     if (!mve_eci_check(s) || !vfp_access_check(s)) {
644         return true;
645     }
646 
647     qd = mve_qreg_ptr(a->qd);
648     qn = mve_qreg_ptr(a->qn);
649     qm = mve_qreg_ptr(a->qm);
650     fn(cpu_env, qd, qn, qm);
651     tcg_temp_free_ptr(qd);
652     tcg_temp_free_ptr(qn);
653     tcg_temp_free_ptr(qm);
654     mve_update_eci(s);
655     return true;
656 }
657 
658 #define DO_LOGIC(INSN, HELPER)                                  \
659     static bool trans_##INSN(DisasContext *s, arg_2op *a)       \
660     {                                                           \
661         return do_2op(s, a, HELPER);                            \
662     }
663 
664 DO_LOGIC(VAND, gen_helper_mve_vand)
665 DO_LOGIC(VBIC, gen_helper_mve_vbic)
666 DO_LOGIC(VORR, gen_helper_mve_vorr)
667 DO_LOGIC(VORN, gen_helper_mve_vorn)
668 DO_LOGIC(VEOR, gen_helper_mve_veor)
669 
670 DO_LOGIC(VPSEL, gen_helper_mve_vpsel)
671 
672 #define DO_2OP(INSN, FN) \
673     static bool trans_##INSN(DisasContext *s, arg_2op *a)       \
674     {                                                           \
675         static MVEGenTwoOpFn * const fns[] = {                  \
676             gen_helper_mve_##FN##b,                             \
677             gen_helper_mve_##FN##h,                             \
678             gen_helper_mve_##FN##w,                             \
679             NULL,                                               \
680         };                                                      \
681         return do_2op(s, a, fns[a->size]);                      \
682     }
683 
684 DO_2OP(VADD, vadd)
685 DO_2OP(VSUB, vsub)
686 DO_2OP(VMUL, vmul)
687 DO_2OP(VMULH_S, vmulhs)
688 DO_2OP(VMULH_U, vmulhu)
689 DO_2OP(VRMULH_S, vrmulhs)
690 DO_2OP(VRMULH_U, vrmulhu)
691 DO_2OP(VMAX_S, vmaxs)
692 DO_2OP(VMAX_U, vmaxu)
693 DO_2OP(VMIN_S, vmins)
694 DO_2OP(VMIN_U, vminu)
695 DO_2OP(VABD_S, vabds)
696 DO_2OP(VABD_U, vabdu)
697 DO_2OP(VHADD_S, vhadds)
698 DO_2OP(VHADD_U, vhaddu)
699 DO_2OP(VHSUB_S, vhsubs)
700 DO_2OP(VHSUB_U, vhsubu)
701 DO_2OP(VMULL_BS, vmullbs)
702 DO_2OP(VMULL_BU, vmullbu)
703 DO_2OP(VMULL_TS, vmullts)
704 DO_2OP(VMULL_TU, vmulltu)
705 DO_2OP(VQDMULH, vqdmulh)
706 DO_2OP(VQRDMULH, vqrdmulh)
707 DO_2OP(VQADD_S, vqadds)
708 DO_2OP(VQADD_U, vqaddu)
709 DO_2OP(VQSUB_S, vqsubs)
710 DO_2OP(VQSUB_U, vqsubu)
711 DO_2OP(VSHL_S, vshls)
712 DO_2OP(VSHL_U, vshlu)
713 DO_2OP(VRSHL_S, vrshls)
714 DO_2OP(VRSHL_U, vrshlu)
715 DO_2OP(VQSHL_S, vqshls)
716 DO_2OP(VQSHL_U, vqshlu)
717 DO_2OP(VQRSHL_S, vqrshls)
718 DO_2OP(VQRSHL_U, vqrshlu)
719 DO_2OP(VQDMLADH, vqdmladh)
720 DO_2OP(VQDMLADHX, vqdmladhx)
721 DO_2OP(VQRDMLADH, vqrdmladh)
722 DO_2OP(VQRDMLADHX, vqrdmladhx)
723 DO_2OP(VQDMLSDH, vqdmlsdh)
724 DO_2OP(VQDMLSDHX, vqdmlsdhx)
725 DO_2OP(VQRDMLSDH, vqrdmlsdh)
726 DO_2OP(VQRDMLSDHX, vqrdmlsdhx)
727 DO_2OP(VRHADD_S, vrhadds)
728 DO_2OP(VRHADD_U, vrhaddu)
729 /*
730  * VCADD Qd == Qm at size MO_32 is UNPREDICTABLE; we choose not to diagnose
731  * so we can reuse the DO_2OP macro. (Our implementation calculates the
732  * "expected" results in this case.) Similarly for VHCADD.
733  */
734 DO_2OP(VCADD90, vcadd90)
735 DO_2OP(VCADD270, vcadd270)
736 DO_2OP(VHCADD90, vhcadd90)
737 DO_2OP(VHCADD270, vhcadd270)
738 
739 static bool trans_VQDMULLB(DisasContext *s, arg_2op *a)
740 {
741     static MVEGenTwoOpFn * const fns[] = {
742         NULL,
743         gen_helper_mve_vqdmullbh,
744         gen_helper_mve_vqdmullbw,
745         NULL,
746     };
747     if (a->size == MO_32 && (a->qd == a->qm || a->qd == a->qn)) {
748         /* UNPREDICTABLE; we choose to undef */
749         return false;
750     }
751     return do_2op(s, a, fns[a->size]);
752 }
753 
754 static bool trans_VQDMULLT(DisasContext *s, arg_2op *a)
755 {
756     static MVEGenTwoOpFn * const fns[] = {
757         NULL,
758         gen_helper_mve_vqdmullth,
759         gen_helper_mve_vqdmulltw,
760         NULL,
761     };
762     if (a->size == MO_32 && (a->qd == a->qm || a->qd == a->qn)) {
763         /* UNPREDICTABLE; we choose to undef */
764         return false;
765     }
766     return do_2op(s, a, fns[a->size]);
767 }
768 
769 static bool trans_VMULLP_B(DisasContext *s, arg_2op *a)
770 {
771     /*
772      * Note that a->size indicates the output size, ie VMULL.P8
773      * is the 8x8->16 operation and a->size is MO_16; VMULL.P16
774      * is the 16x16->32 operation and a->size is MO_32.
775      */
776     static MVEGenTwoOpFn * const fns[] = {
777         NULL,
778         gen_helper_mve_vmullpbh,
779         gen_helper_mve_vmullpbw,
780         NULL,
781     };
782     return do_2op(s, a, fns[a->size]);
783 }
784 
785 static bool trans_VMULLP_T(DisasContext *s, arg_2op *a)
786 {
787     /* a->size is as for trans_VMULLP_B */
788     static MVEGenTwoOpFn * const fns[] = {
789         NULL,
790         gen_helper_mve_vmullpth,
791         gen_helper_mve_vmullptw,
792         NULL,
793     };
794     return do_2op(s, a, fns[a->size]);
795 }
796 
797 /*
798  * VADC and VSBC: these perform an add-with-carry or subtract-with-carry
799  * of the 32-bit elements in each lane of the input vectors, where the
800  * carry-out of each add is the carry-in of the next.  The initial carry
801  * input is either fixed (0 for VADCI, 1 for VSBCI) or is from FPSCR.C
802  * (for VADC and VSBC); the carry out at the end is written back to FPSCR.C.
803  * These insns are subject to beat-wise execution.  Partial execution
804  * of an I=1 (initial carry input fixed) insn which does not
805  * execute the first beat must start with the current FPSCR.NZCV
806  * value, not the fixed constant input.
807  */
808 static bool trans_VADC(DisasContext *s, arg_2op *a)
809 {
810     return do_2op(s, a, gen_helper_mve_vadc);
811 }
812 
813 static bool trans_VADCI(DisasContext *s, arg_2op *a)
814 {
815     if (mve_skip_first_beat(s)) {
816         return trans_VADC(s, a);
817     }
818     return do_2op(s, a, gen_helper_mve_vadci);
819 }
820 
821 static bool trans_VSBC(DisasContext *s, arg_2op *a)
822 {
823     return do_2op(s, a, gen_helper_mve_vsbc);
824 }
825 
826 static bool trans_VSBCI(DisasContext *s, arg_2op *a)
827 {
828     if (mve_skip_first_beat(s)) {
829         return trans_VSBC(s, a);
830     }
831     return do_2op(s, a, gen_helper_mve_vsbci);
832 }
833 
834 static bool do_2op_scalar(DisasContext *s, arg_2scalar *a,
835                           MVEGenTwoOpScalarFn fn)
836 {
837     TCGv_ptr qd, qn;
838     TCGv_i32 rm;
839 
840     if (!dc_isar_feature(aa32_mve, s) ||
841         !mve_check_qreg_bank(s, a->qd | a->qn) ||
842         !fn) {
843         return false;
844     }
845     if (a->rm == 13 || a->rm == 15) {
846         /* UNPREDICTABLE */
847         return false;
848     }
849     if (!mve_eci_check(s) || !vfp_access_check(s)) {
850         return true;
851     }
852 
853     qd = mve_qreg_ptr(a->qd);
854     qn = mve_qreg_ptr(a->qn);
855     rm = load_reg(s, a->rm);
856     fn(cpu_env, qd, qn, rm);
857     tcg_temp_free_i32(rm);
858     tcg_temp_free_ptr(qd);
859     tcg_temp_free_ptr(qn);
860     mve_update_eci(s);
861     return true;
862 }
863 
864 #define DO_2OP_SCALAR(INSN, FN) \
865     static bool trans_##INSN(DisasContext *s, arg_2scalar *a)   \
866     {                                                           \
867         static MVEGenTwoOpScalarFn * const fns[] = {            \
868             gen_helper_mve_##FN##b,                             \
869             gen_helper_mve_##FN##h,                             \
870             gen_helper_mve_##FN##w,                             \
871             NULL,                                               \
872         };                                                      \
873         return do_2op_scalar(s, a, fns[a->size]);               \
874     }
875 
876 DO_2OP_SCALAR(VADD_scalar, vadd_scalar)
877 DO_2OP_SCALAR(VSUB_scalar, vsub_scalar)
878 DO_2OP_SCALAR(VMUL_scalar, vmul_scalar)
879 DO_2OP_SCALAR(VHADD_S_scalar, vhadds_scalar)
880 DO_2OP_SCALAR(VHADD_U_scalar, vhaddu_scalar)
881 DO_2OP_SCALAR(VHSUB_S_scalar, vhsubs_scalar)
882 DO_2OP_SCALAR(VHSUB_U_scalar, vhsubu_scalar)
883 DO_2OP_SCALAR(VQADD_S_scalar, vqadds_scalar)
884 DO_2OP_SCALAR(VQADD_U_scalar, vqaddu_scalar)
885 DO_2OP_SCALAR(VQSUB_S_scalar, vqsubs_scalar)
886 DO_2OP_SCALAR(VQSUB_U_scalar, vqsubu_scalar)
887 DO_2OP_SCALAR(VQDMULH_scalar, vqdmulh_scalar)
888 DO_2OP_SCALAR(VQRDMULH_scalar, vqrdmulh_scalar)
889 DO_2OP_SCALAR(VBRSR, vbrsr)
890 DO_2OP_SCALAR(VMLA, vmla)
891 DO_2OP_SCALAR(VMLAS, vmlas)
892 DO_2OP_SCALAR(VQDMLAH, vqdmlah)
893 DO_2OP_SCALAR(VQRDMLAH, vqrdmlah)
894 DO_2OP_SCALAR(VQDMLASH, vqdmlash)
895 DO_2OP_SCALAR(VQRDMLASH, vqrdmlash)
896 
897 static bool trans_VQDMULLB_scalar(DisasContext *s, arg_2scalar *a)
898 {
899     static MVEGenTwoOpScalarFn * const fns[] = {
900         NULL,
901         gen_helper_mve_vqdmullb_scalarh,
902         gen_helper_mve_vqdmullb_scalarw,
903         NULL,
904     };
905     if (a->qd == a->qn && a->size == MO_32) {
906         /* UNPREDICTABLE; we choose to undef */
907         return false;
908     }
909     return do_2op_scalar(s, a, fns[a->size]);
910 }
911 
912 static bool trans_VQDMULLT_scalar(DisasContext *s, arg_2scalar *a)
913 {
914     static MVEGenTwoOpScalarFn * const fns[] = {
915         NULL,
916         gen_helper_mve_vqdmullt_scalarh,
917         gen_helper_mve_vqdmullt_scalarw,
918         NULL,
919     };
920     if (a->qd == a->qn && a->size == MO_32) {
921         /* UNPREDICTABLE; we choose to undef */
922         return false;
923     }
924     return do_2op_scalar(s, a, fns[a->size]);
925 }
926 
927 static bool do_long_dual_acc(DisasContext *s, arg_vmlaldav *a,
928                              MVEGenLongDualAccOpFn *fn)
929 {
930     TCGv_ptr qn, qm;
931     TCGv_i64 rda;
932     TCGv_i32 rdalo, rdahi;
933 
934     if (!dc_isar_feature(aa32_mve, s) ||
935         !mve_check_qreg_bank(s, a->qn | a->qm) ||
936         !fn) {
937         return false;
938     }
939     /*
940      * rdahi == 13 is UNPREDICTABLE; rdahi == 15 is a related
941      * encoding; rdalo always has bit 0 clear so cannot be 13 or 15.
942      */
943     if (a->rdahi == 13 || a->rdahi == 15) {
944         return false;
945     }
946     if (!mve_eci_check(s) || !vfp_access_check(s)) {
947         return true;
948     }
949 
950     qn = mve_qreg_ptr(a->qn);
951     qm = mve_qreg_ptr(a->qm);
952 
953     /*
954      * This insn is subject to beat-wise execution. Partial execution
955      * of an A=0 (no-accumulate) insn which does not execute the first
956      * beat must start with the current rda value, not 0.
957      */
958     if (a->a || mve_skip_first_beat(s)) {
959         rda = tcg_temp_new_i64();
960         rdalo = load_reg(s, a->rdalo);
961         rdahi = load_reg(s, a->rdahi);
962         tcg_gen_concat_i32_i64(rda, rdalo, rdahi);
963         tcg_temp_free_i32(rdalo);
964         tcg_temp_free_i32(rdahi);
965     } else {
966         rda = tcg_const_i64(0);
967     }
968 
969     fn(rda, cpu_env, qn, qm, rda);
970     tcg_temp_free_ptr(qn);
971     tcg_temp_free_ptr(qm);
972 
973     rdalo = tcg_temp_new_i32();
974     rdahi = tcg_temp_new_i32();
975     tcg_gen_extrl_i64_i32(rdalo, rda);
976     tcg_gen_extrh_i64_i32(rdahi, rda);
977     store_reg(s, a->rdalo, rdalo);
978     store_reg(s, a->rdahi, rdahi);
979     tcg_temp_free_i64(rda);
980     mve_update_eci(s);
981     return true;
982 }
983 
984 static bool trans_VMLALDAV_S(DisasContext *s, arg_vmlaldav *a)
985 {
986     static MVEGenLongDualAccOpFn * const fns[4][2] = {
987         { NULL, NULL },
988         { gen_helper_mve_vmlaldavsh, gen_helper_mve_vmlaldavxsh },
989         { gen_helper_mve_vmlaldavsw, gen_helper_mve_vmlaldavxsw },
990         { NULL, NULL },
991     };
992     return do_long_dual_acc(s, a, fns[a->size][a->x]);
993 }
994 
995 static bool trans_VMLALDAV_U(DisasContext *s, arg_vmlaldav *a)
996 {
997     static MVEGenLongDualAccOpFn * const fns[4][2] = {
998         { NULL, NULL },
999         { gen_helper_mve_vmlaldavuh, NULL },
1000         { gen_helper_mve_vmlaldavuw, NULL },
1001         { NULL, NULL },
1002     };
1003     return do_long_dual_acc(s, a, fns[a->size][a->x]);
1004 }
1005 
1006 static bool trans_VMLSLDAV(DisasContext *s, arg_vmlaldav *a)
1007 {
1008     static MVEGenLongDualAccOpFn * const fns[4][2] = {
1009         { NULL, NULL },
1010         { gen_helper_mve_vmlsldavsh, gen_helper_mve_vmlsldavxsh },
1011         { gen_helper_mve_vmlsldavsw, gen_helper_mve_vmlsldavxsw },
1012         { NULL, NULL },
1013     };
1014     return do_long_dual_acc(s, a, fns[a->size][a->x]);
1015 }
1016 
1017 static bool trans_VRMLALDAVH_S(DisasContext *s, arg_vmlaldav *a)
1018 {
1019     static MVEGenLongDualAccOpFn * const fns[] = {
1020         gen_helper_mve_vrmlaldavhsw, gen_helper_mve_vrmlaldavhxsw,
1021     };
1022     return do_long_dual_acc(s, a, fns[a->x]);
1023 }
1024 
1025 static bool trans_VRMLALDAVH_U(DisasContext *s, arg_vmlaldav *a)
1026 {
1027     static MVEGenLongDualAccOpFn * const fns[] = {
1028         gen_helper_mve_vrmlaldavhuw, NULL,
1029     };
1030     return do_long_dual_acc(s, a, fns[a->x]);
1031 }
1032 
1033 static bool trans_VRMLSLDAVH(DisasContext *s, arg_vmlaldav *a)
1034 {
1035     static MVEGenLongDualAccOpFn * const fns[] = {
1036         gen_helper_mve_vrmlsldavhsw, gen_helper_mve_vrmlsldavhxsw,
1037     };
1038     return do_long_dual_acc(s, a, fns[a->x]);
1039 }
1040 
1041 static bool do_dual_acc(DisasContext *s, arg_vmladav *a, MVEGenDualAccOpFn *fn)
1042 {
1043     TCGv_ptr qn, qm;
1044     TCGv_i32 rda;
1045 
1046     if (!dc_isar_feature(aa32_mve, s) ||
1047         !mve_check_qreg_bank(s, a->qn) ||
1048         !fn) {
1049         return false;
1050     }
1051     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1052         return true;
1053     }
1054 
1055     qn = mve_qreg_ptr(a->qn);
1056     qm = mve_qreg_ptr(a->qm);
1057 
1058     /*
1059      * This insn is subject to beat-wise execution. Partial execution
1060      * of an A=0 (no-accumulate) insn which does not execute the first
1061      * beat must start with the current rda value, not 0.
1062      */
1063     if (a->a || mve_skip_first_beat(s)) {
1064         rda = load_reg(s, a->rda);
1065     } else {
1066         rda = tcg_const_i32(0);
1067     }
1068 
1069     fn(rda, cpu_env, qn, qm, rda);
1070     store_reg(s, a->rda, rda);
1071     tcg_temp_free_ptr(qn);
1072     tcg_temp_free_ptr(qm);
1073 
1074     mve_update_eci(s);
1075     return true;
1076 }
1077 
1078 #define DO_DUAL_ACC(INSN, FN)                                           \
1079     static bool trans_##INSN(DisasContext *s, arg_vmladav *a)           \
1080     {                                                                   \
1081         static MVEGenDualAccOpFn * const fns[4][2] = {                  \
1082             { gen_helper_mve_##FN##b, gen_helper_mve_##FN##xb },        \
1083             { gen_helper_mve_##FN##h, gen_helper_mve_##FN##xh },        \
1084             { gen_helper_mve_##FN##w, gen_helper_mve_##FN##xw },        \
1085             { NULL, NULL },                                             \
1086         };                                                              \
1087         return do_dual_acc(s, a, fns[a->size][a->x]);                   \
1088     }
1089 
1090 DO_DUAL_ACC(VMLADAV_S, vmladavs)
1091 DO_DUAL_ACC(VMLSDAV, vmlsdav)
1092 
1093 static bool trans_VMLADAV_U(DisasContext *s, arg_vmladav *a)
1094 {
1095     static MVEGenDualAccOpFn * const fns[4][2] = {
1096         { gen_helper_mve_vmladavub, NULL },
1097         { gen_helper_mve_vmladavuh, NULL },
1098         { gen_helper_mve_vmladavuw, NULL },
1099         { NULL, NULL },
1100     };
1101     return do_dual_acc(s, a, fns[a->size][a->x]);
1102 }
1103 
1104 static void gen_vpst(DisasContext *s, uint32_t mask)
1105 {
1106     /*
1107      * Set the VPR mask fields. We take advantage of MASK01 and MASK23
1108      * being adjacent fields in the register.
1109      *
1110      * Updating the masks is not predicated, but it is subject to beat-wise
1111      * execution, and the mask is updated on the odd-numbered beats.
1112      * So if PSR.ECI says we should skip beat 1, we mustn't update the
1113      * 01 mask field.
1114      */
1115     TCGv_i32 vpr = load_cpu_field(v7m.vpr);
1116     switch (s->eci) {
1117     case ECI_NONE:
1118     case ECI_A0:
1119         /* Update both 01 and 23 fields */
1120         tcg_gen_deposit_i32(vpr, vpr,
1121                             tcg_constant_i32(mask | (mask << 4)),
1122                             R_V7M_VPR_MASK01_SHIFT,
1123                             R_V7M_VPR_MASK01_LENGTH + R_V7M_VPR_MASK23_LENGTH);
1124         break;
1125     case ECI_A0A1:
1126     case ECI_A0A1A2:
1127     case ECI_A0A1A2B0:
1128         /* Update only the 23 mask field */
1129         tcg_gen_deposit_i32(vpr, vpr,
1130                             tcg_constant_i32(mask),
1131                             R_V7M_VPR_MASK23_SHIFT, R_V7M_VPR_MASK23_LENGTH);
1132         break;
1133     default:
1134         g_assert_not_reached();
1135     }
1136     store_cpu_field(vpr, v7m.vpr);
1137 }
1138 
1139 static bool trans_VPST(DisasContext *s, arg_VPST *a)
1140 {
1141     /* mask == 0 is a "related encoding" */
1142     if (!dc_isar_feature(aa32_mve, s) || !a->mask) {
1143         return false;
1144     }
1145     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1146         return true;
1147     }
1148     gen_vpst(s, a->mask);
1149     mve_update_and_store_eci(s);
1150     return true;
1151 }
1152 
1153 static bool trans_VPNOT(DisasContext *s, arg_VPNOT *a)
1154 {
1155     /*
1156      * Invert the predicate in VPR.P0. We have call out to
1157      * a helper because this insn itself is beatwise and can
1158      * be predicated.
1159      */
1160     if (!dc_isar_feature(aa32_mve, s)) {
1161         return false;
1162     }
1163     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1164         return true;
1165     }
1166 
1167     gen_helper_mve_vpnot(cpu_env);
1168     mve_update_eci(s);
1169     return true;
1170 }
1171 
1172 static bool trans_VADDV(DisasContext *s, arg_VADDV *a)
1173 {
1174     /* VADDV: vector add across vector */
1175     static MVEGenVADDVFn * const fns[4][2] = {
1176         { gen_helper_mve_vaddvsb, gen_helper_mve_vaddvub },
1177         { gen_helper_mve_vaddvsh, gen_helper_mve_vaddvuh },
1178         { gen_helper_mve_vaddvsw, gen_helper_mve_vaddvuw },
1179         { NULL, NULL }
1180     };
1181     TCGv_ptr qm;
1182     TCGv_i32 rda;
1183 
1184     if (!dc_isar_feature(aa32_mve, s) ||
1185         a->size == 3) {
1186         return false;
1187     }
1188     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1189         return true;
1190     }
1191 
1192     /*
1193      * This insn is subject to beat-wise execution. Partial execution
1194      * of an A=0 (no-accumulate) insn which does not execute the first
1195      * beat must start with the current value of Rda, not zero.
1196      */
1197     if (a->a || mve_skip_first_beat(s)) {
1198         /* Accumulate input from Rda */
1199         rda = load_reg(s, a->rda);
1200     } else {
1201         /* Accumulate starting at zero */
1202         rda = tcg_const_i32(0);
1203     }
1204 
1205     qm = mve_qreg_ptr(a->qm);
1206     fns[a->size][a->u](rda, cpu_env, qm, rda);
1207     store_reg(s, a->rda, rda);
1208     tcg_temp_free_ptr(qm);
1209 
1210     mve_update_eci(s);
1211     return true;
1212 }
1213 
1214 static bool trans_VADDLV(DisasContext *s, arg_VADDLV *a)
1215 {
1216     /*
1217      * Vector Add Long Across Vector: accumulate the 32-bit
1218      * elements of the vector into a 64-bit result stored in
1219      * a pair of general-purpose registers.
1220      * No need to check Qm's bank: it is only 3 bits in decode.
1221      */
1222     TCGv_ptr qm;
1223     TCGv_i64 rda;
1224     TCGv_i32 rdalo, rdahi;
1225 
1226     if (!dc_isar_feature(aa32_mve, s)) {
1227         return false;
1228     }
1229     /*
1230      * rdahi == 13 is UNPREDICTABLE; rdahi == 15 is a related
1231      * encoding; rdalo always has bit 0 clear so cannot be 13 or 15.
1232      */
1233     if (a->rdahi == 13 || a->rdahi == 15) {
1234         return false;
1235     }
1236     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1237         return true;
1238     }
1239 
1240     /*
1241      * This insn is subject to beat-wise execution. Partial execution
1242      * of an A=0 (no-accumulate) insn which does not execute the first
1243      * beat must start with the current value of RdaHi:RdaLo, not zero.
1244      */
1245     if (a->a || mve_skip_first_beat(s)) {
1246         /* Accumulate input from RdaHi:RdaLo */
1247         rda = tcg_temp_new_i64();
1248         rdalo = load_reg(s, a->rdalo);
1249         rdahi = load_reg(s, a->rdahi);
1250         tcg_gen_concat_i32_i64(rda, rdalo, rdahi);
1251         tcg_temp_free_i32(rdalo);
1252         tcg_temp_free_i32(rdahi);
1253     } else {
1254         /* Accumulate starting at zero */
1255         rda = tcg_const_i64(0);
1256     }
1257 
1258     qm = mve_qreg_ptr(a->qm);
1259     if (a->u) {
1260         gen_helper_mve_vaddlv_u(rda, cpu_env, qm, rda);
1261     } else {
1262         gen_helper_mve_vaddlv_s(rda, cpu_env, qm, rda);
1263     }
1264     tcg_temp_free_ptr(qm);
1265 
1266     rdalo = tcg_temp_new_i32();
1267     rdahi = tcg_temp_new_i32();
1268     tcg_gen_extrl_i64_i32(rdalo, rda);
1269     tcg_gen_extrh_i64_i32(rdahi, rda);
1270     store_reg(s, a->rdalo, rdalo);
1271     store_reg(s, a->rdahi, rdahi);
1272     tcg_temp_free_i64(rda);
1273     mve_update_eci(s);
1274     return true;
1275 }
1276 
1277 static bool do_1imm(DisasContext *s, arg_1imm *a, MVEGenOneOpImmFn *fn)
1278 {
1279     TCGv_ptr qd;
1280     uint64_t imm;
1281 
1282     if (!dc_isar_feature(aa32_mve, s) ||
1283         !mve_check_qreg_bank(s, a->qd) ||
1284         !fn) {
1285         return false;
1286     }
1287     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1288         return true;
1289     }
1290 
1291     imm = asimd_imm_const(a->imm, a->cmode, a->op);
1292 
1293     qd = mve_qreg_ptr(a->qd);
1294     fn(cpu_env, qd, tcg_constant_i64(imm));
1295     tcg_temp_free_ptr(qd);
1296     mve_update_eci(s);
1297     return true;
1298 }
1299 
1300 static bool trans_Vimm_1r(DisasContext *s, arg_1imm *a)
1301 {
1302     /* Handle decode of cmode/op here between VORR/VBIC/VMOV */
1303     MVEGenOneOpImmFn *fn;
1304 
1305     if ((a->cmode & 1) && a->cmode < 12) {
1306         if (a->op) {
1307             /*
1308              * For op=1, the immediate will be inverted by asimd_imm_const(),
1309              * so the VBIC becomes a logical AND operation.
1310              */
1311             fn = gen_helper_mve_vandi;
1312         } else {
1313             fn = gen_helper_mve_vorri;
1314         }
1315     } else {
1316         /* There is one unallocated cmode/op combination in this space */
1317         if (a->cmode == 15 && a->op == 1) {
1318             return false;
1319         }
1320         /* asimd_imm_const() sorts out VMVNI vs VMOVI for us */
1321         fn = gen_helper_mve_vmovi;
1322     }
1323     return do_1imm(s, a, fn);
1324 }
1325 
1326 static bool do_2shift(DisasContext *s, arg_2shift *a, MVEGenTwoOpShiftFn fn,
1327                       bool negateshift)
1328 {
1329     TCGv_ptr qd, qm;
1330     int shift = a->shift;
1331 
1332     if (!dc_isar_feature(aa32_mve, s) ||
1333         !mve_check_qreg_bank(s, a->qd | a->qm) ||
1334         !fn) {
1335         return false;
1336     }
1337     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1338         return true;
1339     }
1340 
1341     /*
1342      * When we handle a right shift insn using a left-shift helper
1343      * which permits a negative shift count to indicate a right-shift,
1344      * we must negate the shift count.
1345      */
1346     if (negateshift) {
1347         shift = -shift;
1348     }
1349 
1350     qd = mve_qreg_ptr(a->qd);
1351     qm = mve_qreg_ptr(a->qm);
1352     fn(cpu_env, qd, qm, tcg_constant_i32(shift));
1353     tcg_temp_free_ptr(qd);
1354     tcg_temp_free_ptr(qm);
1355     mve_update_eci(s);
1356     return true;
1357 }
1358 
1359 #define DO_2SHIFT(INSN, FN, NEGATESHIFT)                         \
1360     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1361     {                                                           \
1362         static MVEGenTwoOpShiftFn * const fns[] = {             \
1363             gen_helper_mve_##FN##b,                             \
1364             gen_helper_mve_##FN##h,                             \
1365             gen_helper_mve_##FN##w,                             \
1366             NULL,                                               \
1367         };                                                      \
1368         return do_2shift(s, a, fns[a->size], NEGATESHIFT);      \
1369     }
1370 
1371 DO_2SHIFT(VSHLI, vshli_u, false)
1372 DO_2SHIFT(VQSHLI_S, vqshli_s, false)
1373 DO_2SHIFT(VQSHLI_U, vqshli_u, false)
1374 DO_2SHIFT(VQSHLUI, vqshlui_s, false)
1375 /* These right shifts use a left-shift helper with negated shift count */
1376 DO_2SHIFT(VSHRI_S, vshli_s, true)
1377 DO_2SHIFT(VSHRI_U, vshli_u, true)
1378 DO_2SHIFT(VRSHRI_S, vrshli_s, true)
1379 DO_2SHIFT(VRSHRI_U, vrshli_u, true)
1380 
1381 DO_2SHIFT(VSRI, vsri, false)
1382 DO_2SHIFT(VSLI, vsli, false)
1383 
1384 static bool do_2shift_scalar(DisasContext *s, arg_shl_scalar *a,
1385                              MVEGenTwoOpShiftFn *fn)
1386 {
1387     TCGv_ptr qda;
1388     TCGv_i32 rm;
1389 
1390     if (!dc_isar_feature(aa32_mve, s) ||
1391         !mve_check_qreg_bank(s, a->qda) ||
1392         a->rm == 13 || a->rm == 15 || !fn) {
1393         /* Rm cases are UNPREDICTABLE */
1394         return false;
1395     }
1396     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1397         return true;
1398     }
1399 
1400     qda = mve_qreg_ptr(a->qda);
1401     rm = load_reg(s, a->rm);
1402     fn(cpu_env, qda, qda, rm);
1403     tcg_temp_free_ptr(qda);
1404     tcg_temp_free_i32(rm);
1405     mve_update_eci(s);
1406     return true;
1407 }
1408 
1409 #define DO_2SHIFT_SCALAR(INSN, FN)                                      \
1410     static bool trans_##INSN(DisasContext *s, arg_shl_scalar *a)        \
1411     {                                                                   \
1412         static MVEGenTwoOpShiftFn * const fns[] = {                     \
1413             gen_helper_mve_##FN##b,                                     \
1414             gen_helper_mve_##FN##h,                                     \
1415             gen_helper_mve_##FN##w,                                     \
1416             NULL,                                                       \
1417         };                                                              \
1418         return do_2shift_scalar(s, a, fns[a->size]);                    \
1419     }
1420 
1421 DO_2SHIFT_SCALAR(VSHL_S_scalar, vshli_s)
1422 DO_2SHIFT_SCALAR(VSHL_U_scalar, vshli_u)
1423 DO_2SHIFT_SCALAR(VRSHL_S_scalar, vrshli_s)
1424 DO_2SHIFT_SCALAR(VRSHL_U_scalar, vrshli_u)
1425 DO_2SHIFT_SCALAR(VQSHL_S_scalar, vqshli_s)
1426 DO_2SHIFT_SCALAR(VQSHL_U_scalar, vqshli_u)
1427 DO_2SHIFT_SCALAR(VQRSHL_S_scalar, vqrshli_s)
1428 DO_2SHIFT_SCALAR(VQRSHL_U_scalar, vqrshli_u)
1429 
1430 #define DO_VSHLL(INSN, FN)                                      \
1431     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1432     {                                                           \
1433         static MVEGenTwoOpShiftFn * const fns[] = {             \
1434             gen_helper_mve_##FN##b,                             \
1435             gen_helper_mve_##FN##h,                             \
1436         };                                                      \
1437         return do_2shift(s, a, fns[a->size], false);            \
1438     }
1439 
1440 DO_VSHLL(VSHLL_BS, vshllbs)
1441 DO_VSHLL(VSHLL_BU, vshllbu)
1442 DO_VSHLL(VSHLL_TS, vshllts)
1443 DO_VSHLL(VSHLL_TU, vshlltu)
1444 
1445 #define DO_2SHIFT_N(INSN, FN)                                   \
1446     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1447     {                                                           \
1448         static MVEGenTwoOpShiftFn * const fns[] = {             \
1449             gen_helper_mve_##FN##b,                             \
1450             gen_helper_mve_##FN##h,                             \
1451         };                                                      \
1452         return do_2shift(s, a, fns[a->size], false);            \
1453     }
1454 
1455 DO_2SHIFT_N(VSHRNB, vshrnb)
1456 DO_2SHIFT_N(VSHRNT, vshrnt)
1457 DO_2SHIFT_N(VRSHRNB, vrshrnb)
1458 DO_2SHIFT_N(VRSHRNT, vrshrnt)
1459 DO_2SHIFT_N(VQSHRNB_S, vqshrnb_s)
1460 DO_2SHIFT_N(VQSHRNT_S, vqshrnt_s)
1461 DO_2SHIFT_N(VQSHRNB_U, vqshrnb_u)
1462 DO_2SHIFT_N(VQSHRNT_U, vqshrnt_u)
1463 DO_2SHIFT_N(VQSHRUNB, vqshrunb)
1464 DO_2SHIFT_N(VQSHRUNT, vqshrunt)
1465 DO_2SHIFT_N(VQRSHRNB_S, vqrshrnb_s)
1466 DO_2SHIFT_N(VQRSHRNT_S, vqrshrnt_s)
1467 DO_2SHIFT_N(VQRSHRNB_U, vqrshrnb_u)
1468 DO_2SHIFT_N(VQRSHRNT_U, vqrshrnt_u)
1469 DO_2SHIFT_N(VQRSHRUNB, vqrshrunb)
1470 DO_2SHIFT_N(VQRSHRUNT, vqrshrunt)
1471 
1472 static bool trans_VSHLC(DisasContext *s, arg_VSHLC *a)
1473 {
1474     /*
1475      * Whole Vector Left Shift with Carry. The carry is taken
1476      * from a general purpose register and written back there.
1477      * An imm of 0 means "shift by 32".
1478      */
1479     TCGv_ptr qd;
1480     TCGv_i32 rdm;
1481 
1482     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) {
1483         return false;
1484     }
1485     if (a->rdm == 13 || a->rdm == 15) {
1486         /* CONSTRAINED UNPREDICTABLE: we UNDEF */
1487         return false;
1488     }
1489     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1490         return true;
1491     }
1492 
1493     qd = mve_qreg_ptr(a->qd);
1494     rdm = load_reg(s, a->rdm);
1495     gen_helper_mve_vshlc(rdm, cpu_env, qd, rdm, tcg_constant_i32(a->imm));
1496     store_reg(s, a->rdm, rdm);
1497     tcg_temp_free_ptr(qd);
1498     mve_update_eci(s);
1499     return true;
1500 }
1501 
1502 static bool do_vidup(DisasContext *s, arg_vidup *a, MVEGenVIDUPFn *fn)
1503 {
1504     TCGv_ptr qd;
1505     TCGv_i32 rn;
1506 
1507     /*
1508      * Vector increment/decrement with wrap and duplicate (VIDUP, VDDUP).
1509      * This fills the vector with elements of successively increasing
1510      * or decreasing values, starting from Rn.
1511      */
1512     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) {
1513         return false;
1514     }
1515     if (a->size == MO_64) {
1516         /* size 0b11 is another encoding */
1517         return false;
1518     }
1519     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1520         return true;
1521     }
1522 
1523     qd = mve_qreg_ptr(a->qd);
1524     rn = load_reg(s, a->rn);
1525     fn(rn, cpu_env, qd, rn, tcg_constant_i32(a->imm));
1526     store_reg(s, a->rn, rn);
1527     tcg_temp_free_ptr(qd);
1528     mve_update_eci(s);
1529     return true;
1530 }
1531 
1532 static bool do_viwdup(DisasContext *s, arg_viwdup *a, MVEGenVIWDUPFn *fn)
1533 {
1534     TCGv_ptr qd;
1535     TCGv_i32 rn, rm;
1536 
1537     /*
1538      * Vector increment/decrement with wrap and duplicate (VIWDUp, VDWDUP)
1539      * This fills the vector with elements of successively increasing
1540      * or decreasing values, starting from Rn. Rm specifies a point where
1541      * the count wraps back around to 0. The updated offset is written back
1542      * to Rn.
1543      */
1544     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) {
1545         return false;
1546     }
1547     if (!fn || a->rm == 13 || a->rm == 15) {
1548         /*
1549          * size 0b11 is another encoding; Rm == 13 is UNPREDICTABLE;
1550          * Rm == 13 is VIWDUP, VDWDUP.
1551          */
1552         return false;
1553     }
1554     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1555         return true;
1556     }
1557 
1558     qd = mve_qreg_ptr(a->qd);
1559     rn = load_reg(s, a->rn);
1560     rm = load_reg(s, a->rm);
1561     fn(rn, cpu_env, qd, rn, rm, tcg_constant_i32(a->imm));
1562     store_reg(s, a->rn, rn);
1563     tcg_temp_free_ptr(qd);
1564     tcg_temp_free_i32(rm);
1565     mve_update_eci(s);
1566     return true;
1567 }
1568 
1569 static bool trans_VIDUP(DisasContext *s, arg_vidup *a)
1570 {
1571     static MVEGenVIDUPFn * const fns[] = {
1572         gen_helper_mve_vidupb,
1573         gen_helper_mve_viduph,
1574         gen_helper_mve_vidupw,
1575         NULL,
1576     };
1577     return do_vidup(s, a, fns[a->size]);
1578 }
1579 
1580 static bool trans_VDDUP(DisasContext *s, arg_vidup *a)
1581 {
1582     static MVEGenVIDUPFn * const fns[] = {
1583         gen_helper_mve_vidupb,
1584         gen_helper_mve_viduph,
1585         gen_helper_mve_vidupw,
1586         NULL,
1587     };
1588     /* VDDUP is just like VIDUP but with a negative immediate */
1589     a->imm = -a->imm;
1590     return do_vidup(s, a, fns[a->size]);
1591 }
1592 
1593 static bool trans_VIWDUP(DisasContext *s, arg_viwdup *a)
1594 {
1595     static MVEGenVIWDUPFn * const fns[] = {
1596         gen_helper_mve_viwdupb,
1597         gen_helper_mve_viwduph,
1598         gen_helper_mve_viwdupw,
1599         NULL,
1600     };
1601     return do_viwdup(s, a, fns[a->size]);
1602 }
1603 
1604 static bool trans_VDWDUP(DisasContext *s, arg_viwdup *a)
1605 {
1606     static MVEGenVIWDUPFn * const fns[] = {
1607         gen_helper_mve_vdwdupb,
1608         gen_helper_mve_vdwduph,
1609         gen_helper_mve_vdwdupw,
1610         NULL,
1611     };
1612     return do_viwdup(s, a, fns[a->size]);
1613 }
1614 
1615 static bool do_vcmp(DisasContext *s, arg_vcmp *a, MVEGenCmpFn *fn)
1616 {
1617     TCGv_ptr qn, qm;
1618 
1619     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qm) ||
1620         !fn) {
1621         return false;
1622     }
1623     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1624         return true;
1625     }
1626 
1627     qn = mve_qreg_ptr(a->qn);
1628     qm = mve_qreg_ptr(a->qm);
1629     fn(cpu_env, qn, qm);
1630     tcg_temp_free_ptr(qn);
1631     tcg_temp_free_ptr(qm);
1632     if (a->mask) {
1633         /* VPT */
1634         gen_vpst(s, a->mask);
1635     }
1636     mve_update_eci(s);
1637     return true;
1638 }
1639 
1640 static bool do_vcmp_scalar(DisasContext *s, arg_vcmp_scalar *a,
1641                            MVEGenScalarCmpFn *fn)
1642 {
1643     TCGv_ptr qn;
1644     TCGv_i32 rm;
1645 
1646     if (!dc_isar_feature(aa32_mve, s) || !fn || a->rm == 13) {
1647         return false;
1648     }
1649     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1650         return true;
1651     }
1652 
1653     qn = mve_qreg_ptr(a->qn);
1654     if (a->rm == 15) {
1655         /* Encoding Rm=0b1111 means "constant zero" */
1656         rm = tcg_constant_i32(0);
1657     } else {
1658         rm = load_reg(s, a->rm);
1659     }
1660     fn(cpu_env, qn, rm);
1661     tcg_temp_free_ptr(qn);
1662     tcg_temp_free_i32(rm);
1663     if (a->mask) {
1664         /* VPT */
1665         gen_vpst(s, a->mask);
1666     }
1667     mve_update_eci(s);
1668     return true;
1669 }
1670 
1671 #define DO_VCMP(INSN, FN)                                       \
1672     static bool trans_##INSN(DisasContext *s, arg_vcmp *a)      \
1673     {                                                           \
1674         static MVEGenCmpFn * const fns[] = {                    \
1675             gen_helper_mve_##FN##b,                             \
1676             gen_helper_mve_##FN##h,                             \
1677             gen_helper_mve_##FN##w,                             \
1678             NULL,                                               \
1679         };                                                      \
1680         return do_vcmp(s, a, fns[a->size]);                     \
1681     }                                                           \
1682     static bool trans_##INSN##_scalar(DisasContext *s,          \
1683                                       arg_vcmp_scalar *a)       \
1684     {                                                           \
1685         static MVEGenScalarCmpFn * const fns[] = {              \
1686             gen_helper_mve_##FN##_scalarb,                      \
1687             gen_helper_mve_##FN##_scalarh,                      \
1688             gen_helper_mve_##FN##_scalarw,                      \
1689             NULL,                                               \
1690         };                                                      \
1691         return do_vcmp_scalar(s, a, fns[a->size]);              \
1692     }
1693 
1694 DO_VCMP(VCMPEQ, vcmpeq)
1695 DO_VCMP(VCMPNE, vcmpne)
1696 DO_VCMP(VCMPCS, vcmpcs)
1697 DO_VCMP(VCMPHI, vcmphi)
1698 DO_VCMP(VCMPGE, vcmpge)
1699 DO_VCMP(VCMPLT, vcmplt)
1700 DO_VCMP(VCMPGT, vcmpgt)
1701 DO_VCMP(VCMPLE, vcmple)
1702 
1703 static bool do_vmaxv(DisasContext *s, arg_vmaxv *a, MVEGenVADDVFn fn)
1704 {
1705     /*
1706      * MIN/MAX operations across a vector: compute the min or
1707      * max of the initial value in a general purpose register
1708      * and all the elements in the vector, and store it back
1709      * into the general purpose register.
1710      */
1711     TCGv_ptr qm;
1712     TCGv_i32 rda;
1713 
1714     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qm) ||
1715         !fn || a->rda == 13 || a->rda == 15) {
1716         /* Rda cases are UNPREDICTABLE */
1717         return false;
1718     }
1719     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1720         return true;
1721     }
1722 
1723     qm = mve_qreg_ptr(a->qm);
1724     rda = load_reg(s, a->rda);
1725     fn(rda, cpu_env, qm, rda);
1726     store_reg(s, a->rda, rda);
1727     tcg_temp_free_ptr(qm);
1728     mve_update_eci(s);
1729     return true;
1730 }
1731 
1732 #define DO_VMAXV(INSN, FN)                                      \
1733     static bool trans_##INSN(DisasContext *s, arg_vmaxv *a)     \
1734     {                                                           \
1735         static MVEGenVADDVFn * const fns[] = {                  \
1736             gen_helper_mve_##FN##b,                             \
1737             gen_helper_mve_##FN##h,                             \
1738             gen_helper_mve_##FN##w,                             \
1739             NULL,                                               \
1740         };                                                      \
1741         return do_vmaxv(s, a, fns[a->size]);                    \
1742     }
1743 
1744 DO_VMAXV(VMAXV_S, vmaxvs)
1745 DO_VMAXV(VMAXV_U, vmaxvu)
1746 DO_VMAXV(VMAXAV, vmaxav)
1747 DO_VMAXV(VMINV_S, vminvs)
1748 DO_VMAXV(VMINV_U, vminvu)
1749 DO_VMAXV(VMINAV, vminav)
1750 
1751 static bool do_vabav(DisasContext *s, arg_vabav *a, MVEGenVABAVFn *fn)
1752 {
1753     /* Absolute difference accumulated across vector */
1754     TCGv_ptr qn, qm;
1755     TCGv_i32 rda;
1756 
1757     if (!dc_isar_feature(aa32_mve, s) ||
1758         !mve_check_qreg_bank(s, a->qm | a->qn) ||
1759         !fn || a->rda == 13 || a->rda == 15) {
1760         /* Rda cases are UNPREDICTABLE */
1761         return false;
1762     }
1763     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1764         return true;
1765     }
1766 
1767     qm = mve_qreg_ptr(a->qm);
1768     qn = mve_qreg_ptr(a->qn);
1769     rda = load_reg(s, a->rda);
1770     fn(rda, cpu_env, qn, qm, rda);
1771     store_reg(s, a->rda, rda);
1772     tcg_temp_free_ptr(qm);
1773     tcg_temp_free_ptr(qn);
1774     mve_update_eci(s);
1775     return true;
1776 }
1777 
1778 #define DO_VABAV(INSN, FN)                                      \
1779     static bool trans_##INSN(DisasContext *s, arg_vabav *a)     \
1780     {                                                           \
1781         static MVEGenVABAVFn * const fns[] = {                  \
1782             gen_helper_mve_##FN##b,                             \
1783             gen_helper_mve_##FN##h,                             \
1784             gen_helper_mve_##FN##w,                             \
1785             NULL,                                               \
1786         };                                                      \
1787         return do_vabav(s, a, fns[a->size]);                    \
1788     }
1789 
1790 DO_VABAV(VABAV_S, vabavs)
1791 DO_VABAV(VABAV_U, vabavu)
1792 
1793 static bool trans_VMOV_to_2gp(DisasContext *s, arg_VMOV_to_2gp *a)
1794 {
1795     /*
1796      * VMOV two 32-bit vector lanes to two general-purpose registers.
1797      * This insn is not predicated but it is subject to beat-wise
1798      * execution if it is not in an IT block. For us this means
1799      * only that if PSR.ECI says we should not be executing the beat
1800      * corresponding to the lane of the vector register being accessed
1801      * then we should skip perfoming the move, and that we need to do
1802      * the usual check for bad ECI state and advance of ECI state.
1803      * (If PSR.ECI is non-zero then we cannot be in an IT block.)
1804      */
1805     TCGv_i32 tmp;
1806     int vd;
1807 
1808     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd) ||
1809         a->rt == 13 || a->rt == 15 || a->rt2 == 13 || a->rt2 == 15 ||
1810         a->rt == a->rt2) {
1811         /* Rt/Rt2 cases are UNPREDICTABLE */
1812         return false;
1813     }
1814     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1815         return true;
1816     }
1817 
1818     /* Convert Qreg index to Dreg for read_neon_element32() etc */
1819     vd = a->qd * 2;
1820 
1821     if (!mve_skip_vmov(s, vd, a->idx, MO_32)) {
1822         tmp = tcg_temp_new_i32();
1823         read_neon_element32(tmp, vd, a->idx, MO_32);
1824         store_reg(s, a->rt, tmp);
1825     }
1826     if (!mve_skip_vmov(s, vd + 1, a->idx, MO_32)) {
1827         tmp = tcg_temp_new_i32();
1828         read_neon_element32(tmp, vd + 1, a->idx, MO_32);
1829         store_reg(s, a->rt2, tmp);
1830     }
1831 
1832     mve_update_and_store_eci(s);
1833     return true;
1834 }
1835 
1836 static bool trans_VMOV_from_2gp(DisasContext *s, arg_VMOV_to_2gp *a)
1837 {
1838     /*
1839      * VMOV two general-purpose registers to two 32-bit vector lanes.
1840      * This insn is not predicated but it is subject to beat-wise
1841      * execution if it is not in an IT block. For us this means
1842      * only that if PSR.ECI says we should not be executing the beat
1843      * corresponding to the lane of the vector register being accessed
1844      * then we should skip perfoming the move, and that we need to do
1845      * the usual check for bad ECI state and advance of ECI state.
1846      * (If PSR.ECI is non-zero then we cannot be in an IT block.)
1847      */
1848     TCGv_i32 tmp;
1849     int vd;
1850 
1851     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd) ||
1852         a->rt == 13 || a->rt == 15 || a->rt2 == 13 || a->rt2 == 15) {
1853         /* Rt/Rt2 cases are UNPREDICTABLE */
1854         return false;
1855     }
1856     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1857         return true;
1858     }
1859 
1860     /* Convert Qreg idx to Dreg for read_neon_element32() etc */
1861     vd = a->qd * 2;
1862 
1863     if (!mve_skip_vmov(s, vd, a->idx, MO_32)) {
1864         tmp = load_reg(s, a->rt);
1865         write_neon_element32(tmp, vd, a->idx, MO_32);
1866         tcg_temp_free_i32(tmp);
1867     }
1868     if (!mve_skip_vmov(s, vd + 1, a->idx, MO_32)) {
1869         tmp = load_reg(s, a->rt2);
1870         write_neon_element32(tmp, vd + 1, a->idx, MO_32);
1871         tcg_temp_free_i32(tmp);
1872     }
1873 
1874     mve_update_and_store_eci(s);
1875     return true;
1876 }
1877