xref: /qemu/target/arm/tcg/translate-mve.c (revision 3173c0dd933cbd80578bde6aa116f8f519174a2e)
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 #define DO_2OP_FP(INSN, FN)                                     \
835     static bool trans_##INSN(DisasContext *s, arg_2op *a)       \
836     {                                                           \
837         static MVEGenTwoOpFn * const fns[] = {                  \
838             NULL,                                               \
839             gen_helper_mve_##FN##h,                             \
840             gen_helper_mve_##FN##s,                             \
841             NULL,                                               \
842         };                                                      \
843         if (!dc_isar_feature(aa32_mve_fp, s)) {                 \
844             return false;                                       \
845         }                                                       \
846         return do_2op(s, a, fns[a->size]);                      \
847     }
848 
849 DO_2OP_FP(VADD_fp, vfadd)
850 DO_2OP_FP(VSUB_fp, vfsub)
851 DO_2OP_FP(VMUL_fp, vfmul)
852 DO_2OP_FP(VABD_fp, vfabd)
853 DO_2OP_FP(VMAXNM, vmaxnm)
854 DO_2OP_FP(VMINNM, vminnm)
855 DO_2OP_FP(VCADD90_fp, vfcadd90)
856 DO_2OP_FP(VCADD270_fp, vfcadd270)
857 DO_2OP_FP(VFMA, vfma)
858 DO_2OP_FP(VFMS, vfms)
859 
860 static bool do_2op_scalar(DisasContext *s, arg_2scalar *a,
861                           MVEGenTwoOpScalarFn fn)
862 {
863     TCGv_ptr qd, qn;
864     TCGv_i32 rm;
865 
866     if (!dc_isar_feature(aa32_mve, s) ||
867         !mve_check_qreg_bank(s, a->qd | a->qn) ||
868         !fn) {
869         return false;
870     }
871     if (a->rm == 13 || a->rm == 15) {
872         /* UNPREDICTABLE */
873         return false;
874     }
875     if (!mve_eci_check(s) || !vfp_access_check(s)) {
876         return true;
877     }
878 
879     qd = mve_qreg_ptr(a->qd);
880     qn = mve_qreg_ptr(a->qn);
881     rm = load_reg(s, a->rm);
882     fn(cpu_env, qd, qn, rm);
883     tcg_temp_free_i32(rm);
884     tcg_temp_free_ptr(qd);
885     tcg_temp_free_ptr(qn);
886     mve_update_eci(s);
887     return true;
888 }
889 
890 #define DO_2OP_SCALAR(INSN, FN)                                 \
891     static bool trans_##INSN(DisasContext *s, arg_2scalar *a)   \
892     {                                                           \
893         static MVEGenTwoOpScalarFn * const fns[] = {            \
894             gen_helper_mve_##FN##b,                             \
895             gen_helper_mve_##FN##h,                             \
896             gen_helper_mve_##FN##w,                             \
897             NULL,                                               \
898         };                                                      \
899         return do_2op_scalar(s, a, fns[a->size]);               \
900     }
901 
902 DO_2OP_SCALAR(VADD_scalar, vadd_scalar)
903 DO_2OP_SCALAR(VSUB_scalar, vsub_scalar)
904 DO_2OP_SCALAR(VMUL_scalar, vmul_scalar)
905 DO_2OP_SCALAR(VHADD_S_scalar, vhadds_scalar)
906 DO_2OP_SCALAR(VHADD_U_scalar, vhaddu_scalar)
907 DO_2OP_SCALAR(VHSUB_S_scalar, vhsubs_scalar)
908 DO_2OP_SCALAR(VHSUB_U_scalar, vhsubu_scalar)
909 DO_2OP_SCALAR(VQADD_S_scalar, vqadds_scalar)
910 DO_2OP_SCALAR(VQADD_U_scalar, vqaddu_scalar)
911 DO_2OP_SCALAR(VQSUB_S_scalar, vqsubs_scalar)
912 DO_2OP_SCALAR(VQSUB_U_scalar, vqsubu_scalar)
913 DO_2OP_SCALAR(VQDMULH_scalar, vqdmulh_scalar)
914 DO_2OP_SCALAR(VQRDMULH_scalar, vqrdmulh_scalar)
915 DO_2OP_SCALAR(VBRSR, vbrsr)
916 DO_2OP_SCALAR(VMLA, vmla)
917 DO_2OP_SCALAR(VMLAS, vmlas)
918 DO_2OP_SCALAR(VQDMLAH, vqdmlah)
919 DO_2OP_SCALAR(VQRDMLAH, vqrdmlah)
920 DO_2OP_SCALAR(VQDMLASH, vqdmlash)
921 DO_2OP_SCALAR(VQRDMLASH, vqrdmlash)
922 
923 static bool trans_VQDMULLB_scalar(DisasContext *s, arg_2scalar *a)
924 {
925     static MVEGenTwoOpScalarFn * const fns[] = {
926         NULL,
927         gen_helper_mve_vqdmullb_scalarh,
928         gen_helper_mve_vqdmullb_scalarw,
929         NULL,
930     };
931     if (a->qd == a->qn && a->size == MO_32) {
932         /* UNPREDICTABLE; we choose to undef */
933         return false;
934     }
935     return do_2op_scalar(s, a, fns[a->size]);
936 }
937 
938 static bool trans_VQDMULLT_scalar(DisasContext *s, arg_2scalar *a)
939 {
940     static MVEGenTwoOpScalarFn * const fns[] = {
941         NULL,
942         gen_helper_mve_vqdmullt_scalarh,
943         gen_helper_mve_vqdmullt_scalarw,
944         NULL,
945     };
946     if (a->qd == a->qn && a->size == MO_32) {
947         /* UNPREDICTABLE; we choose to undef */
948         return false;
949     }
950     return do_2op_scalar(s, a, fns[a->size]);
951 }
952 
953 static bool do_long_dual_acc(DisasContext *s, arg_vmlaldav *a,
954                              MVEGenLongDualAccOpFn *fn)
955 {
956     TCGv_ptr qn, qm;
957     TCGv_i64 rda;
958     TCGv_i32 rdalo, rdahi;
959 
960     if (!dc_isar_feature(aa32_mve, s) ||
961         !mve_check_qreg_bank(s, a->qn | a->qm) ||
962         !fn) {
963         return false;
964     }
965     /*
966      * rdahi == 13 is UNPREDICTABLE; rdahi == 15 is a related
967      * encoding; rdalo always has bit 0 clear so cannot be 13 or 15.
968      */
969     if (a->rdahi == 13 || a->rdahi == 15) {
970         return false;
971     }
972     if (!mve_eci_check(s) || !vfp_access_check(s)) {
973         return true;
974     }
975 
976     qn = mve_qreg_ptr(a->qn);
977     qm = mve_qreg_ptr(a->qm);
978 
979     /*
980      * This insn is subject to beat-wise execution. Partial execution
981      * of an A=0 (no-accumulate) insn which does not execute the first
982      * beat must start with the current rda value, not 0.
983      */
984     if (a->a || mve_skip_first_beat(s)) {
985         rda = tcg_temp_new_i64();
986         rdalo = load_reg(s, a->rdalo);
987         rdahi = load_reg(s, a->rdahi);
988         tcg_gen_concat_i32_i64(rda, rdalo, rdahi);
989         tcg_temp_free_i32(rdalo);
990         tcg_temp_free_i32(rdahi);
991     } else {
992         rda = tcg_const_i64(0);
993     }
994 
995     fn(rda, cpu_env, qn, qm, rda);
996     tcg_temp_free_ptr(qn);
997     tcg_temp_free_ptr(qm);
998 
999     rdalo = tcg_temp_new_i32();
1000     rdahi = tcg_temp_new_i32();
1001     tcg_gen_extrl_i64_i32(rdalo, rda);
1002     tcg_gen_extrh_i64_i32(rdahi, rda);
1003     store_reg(s, a->rdalo, rdalo);
1004     store_reg(s, a->rdahi, rdahi);
1005     tcg_temp_free_i64(rda);
1006     mve_update_eci(s);
1007     return true;
1008 }
1009 
1010 static bool trans_VMLALDAV_S(DisasContext *s, arg_vmlaldav *a)
1011 {
1012     static MVEGenLongDualAccOpFn * const fns[4][2] = {
1013         { NULL, NULL },
1014         { gen_helper_mve_vmlaldavsh, gen_helper_mve_vmlaldavxsh },
1015         { gen_helper_mve_vmlaldavsw, gen_helper_mve_vmlaldavxsw },
1016         { NULL, NULL },
1017     };
1018     return do_long_dual_acc(s, a, fns[a->size][a->x]);
1019 }
1020 
1021 static bool trans_VMLALDAV_U(DisasContext *s, arg_vmlaldav *a)
1022 {
1023     static MVEGenLongDualAccOpFn * const fns[4][2] = {
1024         { NULL, NULL },
1025         { gen_helper_mve_vmlaldavuh, NULL },
1026         { gen_helper_mve_vmlaldavuw, NULL },
1027         { NULL, NULL },
1028     };
1029     return do_long_dual_acc(s, a, fns[a->size][a->x]);
1030 }
1031 
1032 static bool trans_VMLSLDAV(DisasContext *s, arg_vmlaldav *a)
1033 {
1034     static MVEGenLongDualAccOpFn * const fns[4][2] = {
1035         { NULL, NULL },
1036         { gen_helper_mve_vmlsldavsh, gen_helper_mve_vmlsldavxsh },
1037         { gen_helper_mve_vmlsldavsw, gen_helper_mve_vmlsldavxsw },
1038         { NULL, NULL },
1039     };
1040     return do_long_dual_acc(s, a, fns[a->size][a->x]);
1041 }
1042 
1043 static bool trans_VRMLALDAVH_S(DisasContext *s, arg_vmlaldav *a)
1044 {
1045     static MVEGenLongDualAccOpFn * const fns[] = {
1046         gen_helper_mve_vrmlaldavhsw, gen_helper_mve_vrmlaldavhxsw,
1047     };
1048     return do_long_dual_acc(s, a, fns[a->x]);
1049 }
1050 
1051 static bool trans_VRMLALDAVH_U(DisasContext *s, arg_vmlaldav *a)
1052 {
1053     static MVEGenLongDualAccOpFn * const fns[] = {
1054         gen_helper_mve_vrmlaldavhuw, NULL,
1055     };
1056     return do_long_dual_acc(s, a, fns[a->x]);
1057 }
1058 
1059 static bool trans_VRMLSLDAVH(DisasContext *s, arg_vmlaldav *a)
1060 {
1061     static MVEGenLongDualAccOpFn * const fns[] = {
1062         gen_helper_mve_vrmlsldavhsw, gen_helper_mve_vrmlsldavhxsw,
1063     };
1064     return do_long_dual_acc(s, a, fns[a->x]);
1065 }
1066 
1067 static bool do_dual_acc(DisasContext *s, arg_vmladav *a, MVEGenDualAccOpFn *fn)
1068 {
1069     TCGv_ptr qn, qm;
1070     TCGv_i32 rda;
1071 
1072     if (!dc_isar_feature(aa32_mve, s) ||
1073         !mve_check_qreg_bank(s, a->qn) ||
1074         !fn) {
1075         return false;
1076     }
1077     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1078         return true;
1079     }
1080 
1081     qn = mve_qreg_ptr(a->qn);
1082     qm = mve_qreg_ptr(a->qm);
1083 
1084     /*
1085      * This insn is subject to beat-wise execution. Partial execution
1086      * of an A=0 (no-accumulate) insn which does not execute the first
1087      * beat must start with the current rda value, not 0.
1088      */
1089     if (a->a || mve_skip_first_beat(s)) {
1090         rda = load_reg(s, a->rda);
1091     } else {
1092         rda = tcg_const_i32(0);
1093     }
1094 
1095     fn(rda, cpu_env, qn, qm, rda);
1096     store_reg(s, a->rda, rda);
1097     tcg_temp_free_ptr(qn);
1098     tcg_temp_free_ptr(qm);
1099 
1100     mve_update_eci(s);
1101     return true;
1102 }
1103 
1104 #define DO_DUAL_ACC(INSN, FN)                                           \
1105     static bool trans_##INSN(DisasContext *s, arg_vmladav *a)           \
1106     {                                                                   \
1107         static MVEGenDualAccOpFn * const fns[4][2] = {                  \
1108             { gen_helper_mve_##FN##b, gen_helper_mve_##FN##xb },        \
1109             { gen_helper_mve_##FN##h, gen_helper_mve_##FN##xh },        \
1110             { gen_helper_mve_##FN##w, gen_helper_mve_##FN##xw },        \
1111             { NULL, NULL },                                             \
1112         };                                                              \
1113         return do_dual_acc(s, a, fns[a->size][a->x]);                   \
1114     }
1115 
1116 DO_DUAL_ACC(VMLADAV_S, vmladavs)
1117 DO_DUAL_ACC(VMLSDAV, vmlsdav)
1118 
1119 static bool trans_VMLADAV_U(DisasContext *s, arg_vmladav *a)
1120 {
1121     static MVEGenDualAccOpFn * const fns[4][2] = {
1122         { gen_helper_mve_vmladavub, NULL },
1123         { gen_helper_mve_vmladavuh, NULL },
1124         { gen_helper_mve_vmladavuw, NULL },
1125         { NULL, NULL },
1126     };
1127     return do_dual_acc(s, a, fns[a->size][a->x]);
1128 }
1129 
1130 static void gen_vpst(DisasContext *s, uint32_t mask)
1131 {
1132     /*
1133      * Set the VPR mask fields. We take advantage of MASK01 and MASK23
1134      * being adjacent fields in the register.
1135      *
1136      * Updating the masks is not predicated, but it is subject to beat-wise
1137      * execution, and the mask is updated on the odd-numbered beats.
1138      * So if PSR.ECI says we should skip beat 1, we mustn't update the
1139      * 01 mask field.
1140      */
1141     TCGv_i32 vpr = load_cpu_field(v7m.vpr);
1142     switch (s->eci) {
1143     case ECI_NONE:
1144     case ECI_A0:
1145         /* Update both 01 and 23 fields */
1146         tcg_gen_deposit_i32(vpr, vpr,
1147                             tcg_constant_i32(mask | (mask << 4)),
1148                             R_V7M_VPR_MASK01_SHIFT,
1149                             R_V7M_VPR_MASK01_LENGTH + R_V7M_VPR_MASK23_LENGTH);
1150         break;
1151     case ECI_A0A1:
1152     case ECI_A0A1A2:
1153     case ECI_A0A1A2B0:
1154         /* Update only the 23 mask field */
1155         tcg_gen_deposit_i32(vpr, vpr,
1156                             tcg_constant_i32(mask),
1157                             R_V7M_VPR_MASK23_SHIFT, R_V7M_VPR_MASK23_LENGTH);
1158         break;
1159     default:
1160         g_assert_not_reached();
1161     }
1162     store_cpu_field(vpr, v7m.vpr);
1163 }
1164 
1165 static bool trans_VPST(DisasContext *s, arg_VPST *a)
1166 {
1167     /* mask == 0 is a "related encoding" */
1168     if (!dc_isar_feature(aa32_mve, s) || !a->mask) {
1169         return false;
1170     }
1171     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1172         return true;
1173     }
1174     gen_vpst(s, a->mask);
1175     mve_update_and_store_eci(s);
1176     return true;
1177 }
1178 
1179 static bool trans_VPNOT(DisasContext *s, arg_VPNOT *a)
1180 {
1181     /*
1182      * Invert the predicate in VPR.P0. We have call out to
1183      * a helper because this insn itself is beatwise and can
1184      * be predicated.
1185      */
1186     if (!dc_isar_feature(aa32_mve, s)) {
1187         return false;
1188     }
1189     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1190         return true;
1191     }
1192 
1193     gen_helper_mve_vpnot(cpu_env);
1194     mve_update_eci(s);
1195     return true;
1196 }
1197 
1198 static bool trans_VADDV(DisasContext *s, arg_VADDV *a)
1199 {
1200     /* VADDV: vector add across vector */
1201     static MVEGenVADDVFn * const fns[4][2] = {
1202         { gen_helper_mve_vaddvsb, gen_helper_mve_vaddvub },
1203         { gen_helper_mve_vaddvsh, gen_helper_mve_vaddvuh },
1204         { gen_helper_mve_vaddvsw, gen_helper_mve_vaddvuw },
1205         { NULL, NULL }
1206     };
1207     TCGv_ptr qm;
1208     TCGv_i32 rda;
1209 
1210     if (!dc_isar_feature(aa32_mve, s) ||
1211         a->size == 3) {
1212         return false;
1213     }
1214     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1215         return true;
1216     }
1217 
1218     /*
1219      * This insn is subject to beat-wise execution. Partial execution
1220      * of an A=0 (no-accumulate) insn which does not execute the first
1221      * beat must start with the current value of Rda, not zero.
1222      */
1223     if (a->a || mve_skip_first_beat(s)) {
1224         /* Accumulate input from Rda */
1225         rda = load_reg(s, a->rda);
1226     } else {
1227         /* Accumulate starting at zero */
1228         rda = tcg_const_i32(0);
1229     }
1230 
1231     qm = mve_qreg_ptr(a->qm);
1232     fns[a->size][a->u](rda, cpu_env, qm, rda);
1233     store_reg(s, a->rda, rda);
1234     tcg_temp_free_ptr(qm);
1235 
1236     mve_update_eci(s);
1237     return true;
1238 }
1239 
1240 static bool trans_VADDLV(DisasContext *s, arg_VADDLV *a)
1241 {
1242     /*
1243      * Vector Add Long Across Vector: accumulate the 32-bit
1244      * elements of the vector into a 64-bit result stored in
1245      * a pair of general-purpose registers.
1246      * No need to check Qm's bank: it is only 3 bits in decode.
1247      */
1248     TCGv_ptr qm;
1249     TCGv_i64 rda;
1250     TCGv_i32 rdalo, rdahi;
1251 
1252     if (!dc_isar_feature(aa32_mve, s)) {
1253         return false;
1254     }
1255     /*
1256      * rdahi == 13 is UNPREDICTABLE; rdahi == 15 is a related
1257      * encoding; rdalo always has bit 0 clear so cannot be 13 or 15.
1258      */
1259     if (a->rdahi == 13 || a->rdahi == 15) {
1260         return false;
1261     }
1262     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1263         return true;
1264     }
1265 
1266     /*
1267      * This insn is subject to beat-wise execution. Partial execution
1268      * of an A=0 (no-accumulate) insn which does not execute the first
1269      * beat must start with the current value of RdaHi:RdaLo, not zero.
1270      */
1271     if (a->a || mve_skip_first_beat(s)) {
1272         /* Accumulate input from RdaHi:RdaLo */
1273         rda = tcg_temp_new_i64();
1274         rdalo = load_reg(s, a->rdalo);
1275         rdahi = load_reg(s, a->rdahi);
1276         tcg_gen_concat_i32_i64(rda, rdalo, rdahi);
1277         tcg_temp_free_i32(rdalo);
1278         tcg_temp_free_i32(rdahi);
1279     } else {
1280         /* Accumulate starting at zero */
1281         rda = tcg_const_i64(0);
1282     }
1283 
1284     qm = mve_qreg_ptr(a->qm);
1285     if (a->u) {
1286         gen_helper_mve_vaddlv_u(rda, cpu_env, qm, rda);
1287     } else {
1288         gen_helper_mve_vaddlv_s(rda, cpu_env, qm, rda);
1289     }
1290     tcg_temp_free_ptr(qm);
1291 
1292     rdalo = tcg_temp_new_i32();
1293     rdahi = tcg_temp_new_i32();
1294     tcg_gen_extrl_i64_i32(rdalo, rda);
1295     tcg_gen_extrh_i64_i32(rdahi, rda);
1296     store_reg(s, a->rdalo, rdalo);
1297     store_reg(s, a->rdahi, rdahi);
1298     tcg_temp_free_i64(rda);
1299     mve_update_eci(s);
1300     return true;
1301 }
1302 
1303 static bool do_1imm(DisasContext *s, arg_1imm *a, MVEGenOneOpImmFn *fn)
1304 {
1305     TCGv_ptr qd;
1306     uint64_t imm;
1307 
1308     if (!dc_isar_feature(aa32_mve, s) ||
1309         !mve_check_qreg_bank(s, a->qd) ||
1310         !fn) {
1311         return false;
1312     }
1313     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1314         return true;
1315     }
1316 
1317     imm = asimd_imm_const(a->imm, a->cmode, a->op);
1318 
1319     qd = mve_qreg_ptr(a->qd);
1320     fn(cpu_env, qd, tcg_constant_i64(imm));
1321     tcg_temp_free_ptr(qd);
1322     mve_update_eci(s);
1323     return true;
1324 }
1325 
1326 static bool trans_Vimm_1r(DisasContext *s, arg_1imm *a)
1327 {
1328     /* Handle decode of cmode/op here between VORR/VBIC/VMOV */
1329     MVEGenOneOpImmFn *fn;
1330 
1331     if ((a->cmode & 1) && a->cmode < 12) {
1332         if (a->op) {
1333             /*
1334              * For op=1, the immediate will be inverted by asimd_imm_const(),
1335              * so the VBIC becomes a logical AND operation.
1336              */
1337             fn = gen_helper_mve_vandi;
1338         } else {
1339             fn = gen_helper_mve_vorri;
1340         }
1341     } else {
1342         /* There is one unallocated cmode/op combination in this space */
1343         if (a->cmode == 15 && a->op == 1) {
1344             return false;
1345         }
1346         /* asimd_imm_const() sorts out VMVNI vs VMOVI for us */
1347         fn = gen_helper_mve_vmovi;
1348     }
1349     return do_1imm(s, a, fn);
1350 }
1351 
1352 static bool do_2shift(DisasContext *s, arg_2shift *a, MVEGenTwoOpShiftFn fn,
1353                       bool negateshift)
1354 {
1355     TCGv_ptr qd, qm;
1356     int shift = a->shift;
1357 
1358     if (!dc_isar_feature(aa32_mve, s) ||
1359         !mve_check_qreg_bank(s, a->qd | a->qm) ||
1360         !fn) {
1361         return false;
1362     }
1363     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1364         return true;
1365     }
1366 
1367     /*
1368      * When we handle a right shift insn using a left-shift helper
1369      * which permits a negative shift count to indicate a right-shift,
1370      * we must negate the shift count.
1371      */
1372     if (negateshift) {
1373         shift = -shift;
1374     }
1375 
1376     qd = mve_qreg_ptr(a->qd);
1377     qm = mve_qreg_ptr(a->qm);
1378     fn(cpu_env, qd, qm, tcg_constant_i32(shift));
1379     tcg_temp_free_ptr(qd);
1380     tcg_temp_free_ptr(qm);
1381     mve_update_eci(s);
1382     return true;
1383 }
1384 
1385 #define DO_2SHIFT(INSN, FN, NEGATESHIFT)                         \
1386     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1387     {                                                           \
1388         static MVEGenTwoOpShiftFn * const fns[] = {             \
1389             gen_helper_mve_##FN##b,                             \
1390             gen_helper_mve_##FN##h,                             \
1391             gen_helper_mve_##FN##w,                             \
1392             NULL,                                               \
1393         };                                                      \
1394         return do_2shift(s, a, fns[a->size], NEGATESHIFT);      \
1395     }
1396 
1397 DO_2SHIFT(VSHLI, vshli_u, false)
1398 DO_2SHIFT(VQSHLI_S, vqshli_s, false)
1399 DO_2SHIFT(VQSHLI_U, vqshli_u, false)
1400 DO_2SHIFT(VQSHLUI, vqshlui_s, false)
1401 /* These right shifts use a left-shift helper with negated shift count */
1402 DO_2SHIFT(VSHRI_S, vshli_s, true)
1403 DO_2SHIFT(VSHRI_U, vshli_u, true)
1404 DO_2SHIFT(VRSHRI_S, vrshli_s, true)
1405 DO_2SHIFT(VRSHRI_U, vrshli_u, true)
1406 
1407 DO_2SHIFT(VSRI, vsri, false)
1408 DO_2SHIFT(VSLI, vsli, false)
1409 
1410 static bool do_2shift_scalar(DisasContext *s, arg_shl_scalar *a,
1411                              MVEGenTwoOpShiftFn *fn)
1412 {
1413     TCGv_ptr qda;
1414     TCGv_i32 rm;
1415 
1416     if (!dc_isar_feature(aa32_mve, s) ||
1417         !mve_check_qreg_bank(s, a->qda) ||
1418         a->rm == 13 || a->rm == 15 || !fn) {
1419         /* Rm cases are UNPREDICTABLE */
1420         return false;
1421     }
1422     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1423         return true;
1424     }
1425 
1426     qda = mve_qreg_ptr(a->qda);
1427     rm = load_reg(s, a->rm);
1428     fn(cpu_env, qda, qda, rm);
1429     tcg_temp_free_ptr(qda);
1430     tcg_temp_free_i32(rm);
1431     mve_update_eci(s);
1432     return true;
1433 }
1434 
1435 #define DO_2SHIFT_SCALAR(INSN, FN)                                      \
1436     static bool trans_##INSN(DisasContext *s, arg_shl_scalar *a)        \
1437     {                                                                   \
1438         static MVEGenTwoOpShiftFn * const fns[] = {                     \
1439             gen_helper_mve_##FN##b,                                     \
1440             gen_helper_mve_##FN##h,                                     \
1441             gen_helper_mve_##FN##w,                                     \
1442             NULL,                                                       \
1443         };                                                              \
1444         return do_2shift_scalar(s, a, fns[a->size]);                    \
1445     }
1446 
1447 DO_2SHIFT_SCALAR(VSHL_S_scalar, vshli_s)
1448 DO_2SHIFT_SCALAR(VSHL_U_scalar, vshli_u)
1449 DO_2SHIFT_SCALAR(VRSHL_S_scalar, vrshli_s)
1450 DO_2SHIFT_SCALAR(VRSHL_U_scalar, vrshli_u)
1451 DO_2SHIFT_SCALAR(VQSHL_S_scalar, vqshli_s)
1452 DO_2SHIFT_SCALAR(VQSHL_U_scalar, vqshli_u)
1453 DO_2SHIFT_SCALAR(VQRSHL_S_scalar, vqrshli_s)
1454 DO_2SHIFT_SCALAR(VQRSHL_U_scalar, vqrshli_u)
1455 
1456 #define DO_VSHLL(INSN, FN)                                      \
1457     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1458     {                                                           \
1459         static MVEGenTwoOpShiftFn * const fns[] = {             \
1460             gen_helper_mve_##FN##b,                             \
1461             gen_helper_mve_##FN##h,                             \
1462         };                                                      \
1463         return do_2shift(s, a, fns[a->size], false);            \
1464     }
1465 
1466 DO_VSHLL(VSHLL_BS, vshllbs)
1467 DO_VSHLL(VSHLL_BU, vshllbu)
1468 DO_VSHLL(VSHLL_TS, vshllts)
1469 DO_VSHLL(VSHLL_TU, vshlltu)
1470 
1471 #define DO_2SHIFT_N(INSN, FN)                                   \
1472     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1473     {                                                           \
1474         static MVEGenTwoOpShiftFn * const fns[] = {             \
1475             gen_helper_mve_##FN##b,                             \
1476             gen_helper_mve_##FN##h,                             \
1477         };                                                      \
1478         return do_2shift(s, a, fns[a->size], false);            \
1479     }
1480 
1481 DO_2SHIFT_N(VSHRNB, vshrnb)
1482 DO_2SHIFT_N(VSHRNT, vshrnt)
1483 DO_2SHIFT_N(VRSHRNB, vrshrnb)
1484 DO_2SHIFT_N(VRSHRNT, vrshrnt)
1485 DO_2SHIFT_N(VQSHRNB_S, vqshrnb_s)
1486 DO_2SHIFT_N(VQSHRNT_S, vqshrnt_s)
1487 DO_2SHIFT_N(VQSHRNB_U, vqshrnb_u)
1488 DO_2SHIFT_N(VQSHRNT_U, vqshrnt_u)
1489 DO_2SHIFT_N(VQSHRUNB, vqshrunb)
1490 DO_2SHIFT_N(VQSHRUNT, vqshrunt)
1491 DO_2SHIFT_N(VQRSHRNB_S, vqrshrnb_s)
1492 DO_2SHIFT_N(VQRSHRNT_S, vqrshrnt_s)
1493 DO_2SHIFT_N(VQRSHRNB_U, vqrshrnb_u)
1494 DO_2SHIFT_N(VQRSHRNT_U, vqrshrnt_u)
1495 DO_2SHIFT_N(VQRSHRUNB, vqrshrunb)
1496 DO_2SHIFT_N(VQRSHRUNT, vqrshrunt)
1497 
1498 static bool trans_VSHLC(DisasContext *s, arg_VSHLC *a)
1499 {
1500     /*
1501      * Whole Vector Left Shift with Carry. The carry is taken
1502      * from a general purpose register and written back there.
1503      * An imm of 0 means "shift by 32".
1504      */
1505     TCGv_ptr qd;
1506     TCGv_i32 rdm;
1507 
1508     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) {
1509         return false;
1510     }
1511     if (a->rdm == 13 || a->rdm == 15) {
1512         /* CONSTRAINED UNPREDICTABLE: we UNDEF */
1513         return false;
1514     }
1515     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1516         return true;
1517     }
1518 
1519     qd = mve_qreg_ptr(a->qd);
1520     rdm = load_reg(s, a->rdm);
1521     gen_helper_mve_vshlc(rdm, cpu_env, qd, rdm, tcg_constant_i32(a->imm));
1522     store_reg(s, a->rdm, rdm);
1523     tcg_temp_free_ptr(qd);
1524     mve_update_eci(s);
1525     return true;
1526 }
1527 
1528 static bool do_vidup(DisasContext *s, arg_vidup *a, MVEGenVIDUPFn *fn)
1529 {
1530     TCGv_ptr qd;
1531     TCGv_i32 rn;
1532 
1533     /*
1534      * Vector increment/decrement with wrap and duplicate (VIDUP, VDDUP).
1535      * This fills the vector with elements of successively increasing
1536      * or decreasing values, starting from Rn.
1537      */
1538     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) {
1539         return false;
1540     }
1541     if (a->size == MO_64) {
1542         /* size 0b11 is another encoding */
1543         return false;
1544     }
1545     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1546         return true;
1547     }
1548 
1549     qd = mve_qreg_ptr(a->qd);
1550     rn = load_reg(s, a->rn);
1551     fn(rn, cpu_env, qd, rn, tcg_constant_i32(a->imm));
1552     store_reg(s, a->rn, rn);
1553     tcg_temp_free_ptr(qd);
1554     mve_update_eci(s);
1555     return true;
1556 }
1557 
1558 static bool do_viwdup(DisasContext *s, arg_viwdup *a, MVEGenVIWDUPFn *fn)
1559 {
1560     TCGv_ptr qd;
1561     TCGv_i32 rn, rm;
1562 
1563     /*
1564      * Vector increment/decrement with wrap and duplicate (VIWDUp, VDWDUP)
1565      * This fills the vector with elements of successively increasing
1566      * or decreasing values, starting from Rn. Rm specifies a point where
1567      * the count wraps back around to 0. The updated offset is written back
1568      * to Rn.
1569      */
1570     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) {
1571         return false;
1572     }
1573     if (!fn || a->rm == 13 || a->rm == 15) {
1574         /*
1575          * size 0b11 is another encoding; Rm == 13 is UNPREDICTABLE;
1576          * Rm == 13 is VIWDUP, VDWDUP.
1577          */
1578         return false;
1579     }
1580     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1581         return true;
1582     }
1583 
1584     qd = mve_qreg_ptr(a->qd);
1585     rn = load_reg(s, a->rn);
1586     rm = load_reg(s, a->rm);
1587     fn(rn, cpu_env, qd, rn, rm, tcg_constant_i32(a->imm));
1588     store_reg(s, a->rn, rn);
1589     tcg_temp_free_ptr(qd);
1590     tcg_temp_free_i32(rm);
1591     mve_update_eci(s);
1592     return true;
1593 }
1594 
1595 static bool trans_VIDUP(DisasContext *s, arg_vidup *a)
1596 {
1597     static MVEGenVIDUPFn * const fns[] = {
1598         gen_helper_mve_vidupb,
1599         gen_helper_mve_viduph,
1600         gen_helper_mve_vidupw,
1601         NULL,
1602     };
1603     return do_vidup(s, a, fns[a->size]);
1604 }
1605 
1606 static bool trans_VDDUP(DisasContext *s, arg_vidup *a)
1607 {
1608     static MVEGenVIDUPFn * const fns[] = {
1609         gen_helper_mve_vidupb,
1610         gen_helper_mve_viduph,
1611         gen_helper_mve_vidupw,
1612         NULL,
1613     };
1614     /* VDDUP is just like VIDUP but with a negative immediate */
1615     a->imm = -a->imm;
1616     return do_vidup(s, a, fns[a->size]);
1617 }
1618 
1619 static bool trans_VIWDUP(DisasContext *s, arg_viwdup *a)
1620 {
1621     static MVEGenVIWDUPFn * const fns[] = {
1622         gen_helper_mve_viwdupb,
1623         gen_helper_mve_viwduph,
1624         gen_helper_mve_viwdupw,
1625         NULL,
1626     };
1627     return do_viwdup(s, a, fns[a->size]);
1628 }
1629 
1630 static bool trans_VDWDUP(DisasContext *s, arg_viwdup *a)
1631 {
1632     static MVEGenVIWDUPFn * const fns[] = {
1633         gen_helper_mve_vdwdupb,
1634         gen_helper_mve_vdwduph,
1635         gen_helper_mve_vdwdupw,
1636         NULL,
1637     };
1638     return do_viwdup(s, a, fns[a->size]);
1639 }
1640 
1641 static bool do_vcmp(DisasContext *s, arg_vcmp *a, MVEGenCmpFn *fn)
1642 {
1643     TCGv_ptr qn, qm;
1644 
1645     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qm) ||
1646         !fn) {
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     qm = mve_qreg_ptr(a->qm);
1655     fn(cpu_env, qn, qm);
1656     tcg_temp_free_ptr(qn);
1657     tcg_temp_free_ptr(qm);
1658     if (a->mask) {
1659         /* VPT */
1660         gen_vpst(s, a->mask);
1661     }
1662     mve_update_eci(s);
1663     return true;
1664 }
1665 
1666 static bool do_vcmp_scalar(DisasContext *s, arg_vcmp_scalar *a,
1667                            MVEGenScalarCmpFn *fn)
1668 {
1669     TCGv_ptr qn;
1670     TCGv_i32 rm;
1671 
1672     if (!dc_isar_feature(aa32_mve, s) || !fn || a->rm == 13) {
1673         return false;
1674     }
1675     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1676         return true;
1677     }
1678 
1679     qn = mve_qreg_ptr(a->qn);
1680     if (a->rm == 15) {
1681         /* Encoding Rm=0b1111 means "constant zero" */
1682         rm = tcg_constant_i32(0);
1683     } else {
1684         rm = load_reg(s, a->rm);
1685     }
1686     fn(cpu_env, qn, rm);
1687     tcg_temp_free_ptr(qn);
1688     tcg_temp_free_i32(rm);
1689     if (a->mask) {
1690         /* VPT */
1691         gen_vpst(s, a->mask);
1692     }
1693     mve_update_eci(s);
1694     return true;
1695 }
1696 
1697 #define DO_VCMP(INSN, FN)                                       \
1698     static bool trans_##INSN(DisasContext *s, arg_vcmp *a)      \
1699     {                                                           \
1700         static MVEGenCmpFn * const fns[] = {                    \
1701             gen_helper_mve_##FN##b,                             \
1702             gen_helper_mve_##FN##h,                             \
1703             gen_helper_mve_##FN##w,                             \
1704             NULL,                                               \
1705         };                                                      \
1706         return do_vcmp(s, a, fns[a->size]);                     \
1707     }                                                           \
1708     static bool trans_##INSN##_scalar(DisasContext *s,          \
1709                                       arg_vcmp_scalar *a)       \
1710     {                                                           \
1711         static MVEGenScalarCmpFn * const fns[] = {              \
1712             gen_helper_mve_##FN##_scalarb,                      \
1713             gen_helper_mve_##FN##_scalarh,                      \
1714             gen_helper_mve_##FN##_scalarw,                      \
1715             NULL,                                               \
1716         };                                                      \
1717         return do_vcmp_scalar(s, a, fns[a->size]);              \
1718     }
1719 
1720 DO_VCMP(VCMPEQ, vcmpeq)
1721 DO_VCMP(VCMPNE, vcmpne)
1722 DO_VCMP(VCMPCS, vcmpcs)
1723 DO_VCMP(VCMPHI, vcmphi)
1724 DO_VCMP(VCMPGE, vcmpge)
1725 DO_VCMP(VCMPLT, vcmplt)
1726 DO_VCMP(VCMPGT, vcmpgt)
1727 DO_VCMP(VCMPLE, vcmple)
1728 
1729 static bool do_vmaxv(DisasContext *s, arg_vmaxv *a, MVEGenVADDVFn fn)
1730 {
1731     /*
1732      * MIN/MAX operations across a vector: compute the min or
1733      * max of the initial value in a general purpose register
1734      * and all the elements in the vector, and store it back
1735      * into the general purpose register.
1736      */
1737     TCGv_ptr qm;
1738     TCGv_i32 rda;
1739 
1740     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qm) ||
1741         !fn || a->rda == 13 || a->rda == 15) {
1742         /* Rda cases are UNPREDICTABLE */
1743         return false;
1744     }
1745     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1746         return true;
1747     }
1748 
1749     qm = mve_qreg_ptr(a->qm);
1750     rda = load_reg(s, a->rda);
1751     fn(rda, cpu_env, qm, rda);
1752     store_reg(s, a->rda, rda);
1753     tcg_temp_free_ptr(qm);
1754     mve_update_eci(s);
1755     return true;
1756 }
1757 
1758 #define DO_VMAXV(INSN, FN)                                      \
1759     static bool trans_##INSN(DisasContext *s, arg_vmaxv *a)     \
1760     {                                                           \
1761         static MVEGenVADDVFn * const fns[] = {                  \
1762             gen_helper_mve_##FN##b,                             \
1763             gen_helper_mve_##FN##h,                             \
1764             gen_helper_mve_##FN##w,                             \
1765             NULL,                                               \
1766         };                                                      \
1767         return do_vmaxv(s, a, fns[a->size]);                    \
1768     }
1769 
1770 DO_VMAXV(VMAXV_S, vmaxvs)
1771 DO_VMAXV(VMAXV_U, vmaxvu)
1772 DO_VMAXV(VMAXAV, vmaxav)
1773 DO_VMAXV(VMINV_S, vminvs)
1774 DO_VMAXV(VMINV_U, vminvu)
1775 DO_VMAXV(VMINAV, vminav)
1776 
1777 static bool do_vabav(DisasContext *s, arg_vabav *a, MVEGenVABAVFn *fn)
1778 {
1779     /* Absolute difference accumulated across vector */
1780     TCGv_ptr qn, qm;
1781     TCGv_i32 rda;
1782 
1783     if (!dc_isar_feature(aa32_mve, s) ||
1784         !mve_check_qreg_bank(s, a->qm | a->qn) ||
1785         !fn || a->rda == 13 || a->rda == 15) {
1786         /* Rda cases are UNPREDICTABLE */
1787         return false;
1788     }
1789     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1790         return true;
1791     }
1792 
1793     qm = mve_qreg_ptr(a->qm);
1794     qn = mve_qreg_ptr(a->qn);
1795     rda = load_reg(s, a->rda);
1796     fn(rda, cpu_env, qn, qm, rda);
1797     store_reg(s, a->rda, rda);
1798     tcg_temp_free_ptr(qm);
1799     tcg_temp_free_ptr(qn);
1800     mve_update_eci(s);
1801     return true;
1802 }
1803 
1804 #define DO_VABAV(INSN, FN)                                      \
1805     static bool trans_##INSN(DisasContext *s, arg_vabav *a)     \
1806     {                                                           \
1807         static MVEGenVABAVFn * const fns[] = {                  \
1808             gen_helper_mve_##FN##b,                             \
1809             gen_helper_mve_##FN##h,                             \
1810             gen_helper_mve_##FN##w,                             \
1811             NULL,                                               \
1812         };                                                      \
1813         return do_vabav(s, a, fns[a->size]);                    \
1814     }
1815 
1816 DO_VABAV(VABAV_S, vabavs)
1817 DO_VABAV(VABAV_U, vabavu)
1818 
1819 static bool trans_VMOV_to_2gp(DisasContext *s, arg_VMOV_to_2gp *a)
1820 {
1821     /*
1822      * VMOV two 32-bit vector lanes to two general-purpose registers.
1823      * This insn is not predicated but it is subject to beat-wise
1824      * execution if it is not in an IT block. For us this means
1825      * only that if PSR.ECI says we should not be executing the beat
1826      * corresponding to the lane of the vector register being accessed
1827      * then we should skip perfoming the move, and that we need to do
1828      * the usual check for bad ECI state and advance of ECI state.
1829      * (If PSR.ECI is non-zero then we cannot be in an IT block.)
1830      */
1831     TCGv_i32 tmp;
1832     int vd;
1833 
1834     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd) ||
1835         a->rt == 13 || a->rt == 15 || a->rt2 == 13 || a->rt2 == 15 ||
1836         a->rt == a->rt2) {
1837         /* Rt/Rt2 cases are UNPREDICTABLE */
1838         return false;
1839     }
1840     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1841         return true;
1842     }
1843 
1844     /* Convert Qreg index to Dreg for read_neon_element32() etc */
1845     vd = a->qd * 2;
1846 
1847     if (!mve_skip_vmov(s, vd, a->idx, MO_32)) {
1848         tmp = tcg_temp_new_i32();
1849         read_neon_element32(tmp, vd, a->idx, MO_32);
1850         store_reg(s, a->rt, tmp);
1851     }
1852     if (!mve_skip_vmov(s, vd + 1, a->idx, MO_32)) {
1853         tmp = tcg_temp_new_i32();
1854         read_neon_element32(tmp, vd + 1, a->idx, MO_32);
1855         store_reg(s, a->rt2, tmp);
1856     }
1857 
1858     mve_update_and_store_eci(s);
1859     return true;
1860 }
1861 
1862 static bool trans_VMOV_from_2gp(DisasContext *s, arg_VMOV_to_2gp *a)
1863 {
1864     /*
1865      * VMOV two general-purpose registers to two 32-bit vector lanes.
1866      * This insn is not predicated but it is subject to beat-wise
1867      * execution if it is not in an IT block. For us this means
1868      * only that if PSR.ECI says we should not be executing the beat
1869      * corresponding to the lane of the vector register being accessed
1870      * then we should skip perfoming the move, and that we need to do
1871      * the usual check for bad ECI state and advance of ECI state.
1872      * (If PSR.ECI is non-zero then we cannot be in an IT block.)
1873      */
1874     TCGv_i32 tmp;
1875     int vd;
1876 
1877     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd) ||
1878         a->rt == 13 || a->rt == 15 || a->rt2 == 13 || a->rt2 == 15) {
1879         /* Rt/Rt2 cases are UNPREDICTABLE */
1880         return false;
1881     }
1882     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1883         return true;
1884     }
1885 
1886     /* Convert Qreg idx to Dreg for read_neon_element32() etc */
1887     vd = a->qd * 2;
1888 
1889     if (!mve_skip_vmov(s, vd, a->idx, MO_32)) {
1890         tmp = load_reg(s, a->rt);
1891         write_neon_element32(tmp, vd, a->idx, MO_32);
1892         tcg_temp_free_i32(tmp);
1893     }
1894     if (!mve_skip_vmov(s, vd + 1, a->idx, MO_32)) {
1895         tmp = load_reg(s, a->rt2);
1896         write_neon_element32(tmp, vd + 1, a->idx, MO_32);
1897         tcg_temp_free_i32(tmp);
1898     }
1899 
1900     mve_update_and_store_eci(s);
1901     return true;
1902 }
1903