xref: /qemu/target/arm/tcg/translate-mve.c (revision 104afc68cf6fbd518e202647e996a14dd0c70130)
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 
858 static bool do_2op_scalar(DisasContext *s, arg_2scalar *a,
859                           MVEGenTwoOpScalarFn fn)
860 {
861     TCGv_ptr qd, qn;
862     TCGv_i32 rm;
863 
864     if (!dc_isar_feature(aa32_mve, s) ||
865         !mve_check_qreg_bank(s, a->qd | a->qn) ||
866         !fn) {
867         return false;
868     }
869     if (a->rm == 13 || a->rm == 15) {
870         /* UNPREDICTABLE */
871         return false;
872     }
873     if (!mve_eci_check(s) || !vfp_access_check(s)) {
874         return true;
875     }
876 
877     qd = mve_qreg_ptr(a->qd);
878     qn = mve_qreg_ptr(a->qn);
879     rm = load_reg(s, a->rm);
880     fn(cpu_env, qd, qn, rm);
881     tcg_temp_free_i32(rm);
882     tcg_temp_free_ptr(qd);
883     tcg_temp_free_ptr(qn);
884     mve_update_eci(s);
885     return true;
886 }
887 
888 #define DO_2OP_SCALAR(INSN, FN)                                 \
889     static bool trans_##INSN(DisasContext *s, arg_2scalar *a)   \
890     {                                                           \
891         static MVEGenTwoOpScalarFn * const fns[] = {            \
892             gen_helper_mve_##FN##b,                             \
893             gen_helper_mve_##FN##h,                             \
894             gen_helper_mve_##FN##w,                             \
895             NULL,                                               \
896         };                                                      \
897         return do_2op_scalar(s, a, fns[a->size]);               \
898     }
899 
900 DO_2OP_SCALAR(VADD_scalar, vadd_scalar)
901 DO_2OP_SCALAR(VSUB_scalar, vsub_scalar)
902 DO_2OP_SCALAR(VMUL_scalar, vmul_scalar)
903 DO_2OP_SCALAR(VHADD_S_scalar, vhadds_scalar)
904 DO_2OP_SCALAR(VHADD_U_scalar, vhaddu_scalar)
905 DO_2OP_SCALAR(VHSUB_S_scalar, vhsubs_scalar)
906 DO_2OP_SCALAR(VHSUB_U_scalar, vhsubu_scalar)
907 DO_2OP_SCALAR(VQADD_S_scalar, vqadds_scalar)
908 DO_2OP_SCALAR(VQADD_U_scalar, vqaddu_scalar)
909 DO_2OP_SCALAR(VQSUB_S_scalar, vqsubs_scalar)
910 DO_2OP_SCALAR(VQSUB_U_scalar, vqsubu_scalar)
911 DO_2OP_SCALAR(VQDMULH_scalar, vqdmulh_scalar)
912 DO_2OP_SCALAR(VQRDMULH_scalar, vqrdmulh_scalar)
913 DO_2OP_SCALAR(VBRSR, vbrsr)
914 DO_2OP_SCALAR(VMLA, vmla)
915 DO_2OP_SCALAR(VMLAS, vmlas)
916 DO_2OP_SCALAR(VQDMLAH, vqdmlah)
917 DO_2OP_SCALAR(VQRDMLAH, vqrdmlah)
918 DO_2OP_SCALAR(VQDMLASH, vqdmlash)
919 DO_2OP_SCALAR(VQRDMLASH, vqrdmlash)
920 
921 static bool trans_VQDMULLB_scalar(DisasContext *s, arg_2scalar *a)
922 {
923     static MVEGenTwoOpScalarFn * const fns[] = {
924         NULL,
925         gen_helper_mve_vqdmullb_scalarh,
926         gen_helper_mve_vqdmullb_scalarw,
927         NULL,
928     };
929     if (a->qd == a->qn && a->size == MO_32) {
930         /* UNPREDICTABLE; we choose to undef */
931         return false;
932     }
933     return do_2op_scalar(s, a, fns[a->size]);
934 }
935 
936 static bool trans_VQDMULLT_scalar(DisasContext *s, arg_2scalar *a)
937 {
938     static MVEGenTwoOpScalarFn * const fns[] = {
939         NULL,
940         gen_helper_mve_vqdmullt_scalarh,
941         gen_helper_mve_vqdmullt_scalarw,
942         NULL,
943     };
944     if (a->qd == a->qn && a->size == MO_32) {
945         /* UNPREDICTABLE; we choose to undef */
946         return false;
947     }
948     return do_2op_scalar(s, a, fns[a->size]);
949 }
950 
951 static bool do_long_dual_acc(DisasContext *s, arg_vmlaldav *a,
952                              MVEGenLongDualAccOpFn *fn)
953 {
954     TCGv_ptr qn, qm;
955     TCGv_i64 rda;
956     TCGv_i32 rdalo, rdahi;
957 
958     if (!dc_isar_feature(aa32_mve, s) ||
959         !mve_check_qreg_bank(s, a->qn | a->qm) ||
960         !fn) {
961         return false;
962     }
963     /*
964      * rdahi == 13 is UNPREDICTABLE; rdahi == 15 is a related
965      * encoding; rdalo always has bit 0 clear so cannot be 13 or 15.
966      */
967     if (a->rdahi == 13 || a->rdahi == 15) {
968         return false;
969     }
970     if (!mve_eci_check(s) || !vfp_access_check(s)) {
971         return true;
972     }
973 
974     qn = mve_qreg_ptr(a->qn);
975     qm = mve_qreg_ptr(a->qm);
976 
977     /*
978      * This insn is subject to beat-wise execution. Partial execution
979      * of an A=0 (no-accumulate) insn which does not execute the first
980      * beat must start with the current rda value, not 0.
981      */
982     if (a->a || mve_skip_first_beat(s)) {
983         rda = tcg_temp_new_i64();
984         rdalo = load_reg(s, a->rdalo);
985         rdahi = load_reg(s, a->rdahi);
986         tcg_gen_concat_i32_i64(rda, rdalo, rdahi);
987         tcg_temp_free_i32(rdalo);
988         tcg_temp_free_i32(rdahi);
989     } else {
990         rda = tcg_const_i64(0);
991     }
992 
993     fn(rda, cpu_env, qn, qm, rda);
994     tcg_temp_free_ptr(qn);
995     tcg_temp_free_ptr(qm);
996 
997     rdalo = tcg_temp_new_i32();
998     rdahi = tcg_temp_new_i32();
999     tcg_gen_extrl_i64_i32(rdalo, rda);
1000     tcg_gen_extrh_i64_i32(rdahi, rda);
1001     store_reg(s, a->rdalo, rdalo);
1002     store_reg(s, a->rdahi, rdahi);
1003     tcg_temp_free_i64(rda);
1004     mve_update_eci(s);
1005     return true;
1006 }
1007 
1008 static bool trans_VMLALDAV_S(DisasContext *s, arg_vmlaldav *a)
1009 {
1010     static MVEGenLongDualAccOpFn * const fns[4][2] = {
1011         { NULL, NULL },
1012         { gen_helper_mve_vmlaldavsh, gen_helper_mve_vmlaldavxsh },
1013         { gen_helper_mve_vmlaldavsw, gen_helper_mve_vmlaldavxsw },
1014         { NULL, NULL },
1015     };
1016     return do_long_dual_acc(s, a, fns[a->size][a->x]);
1017 }
1018 
1019 static bool trans_VMLALDAV_U(DisasContext *s, arg_vmlaldav *a)
1020 {
1021     static MVEGenLongDualAccOpFn * const fns[4][2] = {
1022         { NULL, NULL },
1023         { gen_helper_mve_vmlaldavuh, NULL },
1024         { gen_helper_mve_vmlaldavuw, NULL },
1025         { NULL, NULL },
1026     };
1027     return do_long_dual_acc(s, a, fns[a->size][a->x]);
1028 }
1029 
1030 static bool trans_VMLSLDAV(DisasContext *s, arg_vmlaldav *a)
1031 {
1032     static MVEGenLongDualAccOpFn * const fns[4][2] = {
1033         { NULL, NULL },
1034         { gen_helper_mve_vmlsldavsh, gen_helper_mve_vmlsldavxsh },
1035         { gen_helper_mve_vmlsldavsw, gen_helper_mve_vmlsldavxsw },
1036         { NULL, NULL },
1037     };
1038     return do_long_dual_acc(s, a, fns[a->size][a->x]);
1039 }
1040 
1041 static bool trans_VRMLALDAVH_S(DisasContext *s, arg_vmlaldav *a)
1042 {
1043     static MVEGenLongDualAccOpFn * const fns[] = {
1044         gen_helper_mve_vrmlaldavhsw, gen_helper_mve_vrmlaldavhxsw,
1045     };
1046     return do_long_dual_acc(s, a, fns[a->x]);
1047 }
1048 
1049 static bool trans_VRMLALDAVH_U(DisasContext *s, arg_vmlaldav *a)
1050 {
1051     static MVEGenLongDualAccOpFn * const fns[] = {
1052         gen_helper_mve_vrmlaldavhuw, NULL,
1053     };
1054     return do_long_dual_acc(s, a, fns[a->x]);
1055 }
1056 
1057 static bool trans_VRMLSLDAVH(DisasContext *s, arg_vmlaldav *a)
1058 {
1059     static MVEGenLongDualAccOpFn * const fns[] = {
1060         gen_helper_mve_vrmlsldavhsw, gen_helper_mve_vrmlsldavhxsw,
1061     };
1062     return do_long_dual_acc(s, a, fns[a->x]);
1063 }
1064 
1065 static bool do_dual_acc(DisasContext *s, arg_vmladav *a, MVEGenDualAccOpFn *fn)
1066 {
1067     TCGv_ptr qn, qm;
1068     TCGv_i32 rda;
1069 
1070     if (!dc_isar_feature(aa32_mve, s) ||
1071         !mve_check_qreg_bank(s, a->qn) ||
1072         !fn) {
1073         return false;
1074     }
1075     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1076         return true;
1077     }
1078 
1079     qn = mve_qreg_ptr(a->qn);
1080     qm = mve_qreg_ptr(a->qm);
1081 
1082     /*
1083      * This insn is subject to beat-wise execution. Partial execution
1084      * of an A=0 (no-accumulate) insn which does not execute the first
1085      * beat must start with the current rda value, not 0.
1086      */
1087     if (a->a || mve_skip_first_beat(s)) {
1088         rda = load_reg(s, a->rda);
1089     } else {
1090         rda = tcg_const_i32(0);
1091     }
1092 
1093     fn(rda, cpu_env, qn, qm, rda);
1094     store_reg(s, a->rda, rda);
1095     tcg_temp_free_ptr(qn);
1096     tcg_temp_free_ptr(qm);
1097 
1098     mve_update_eci(s);
1099     return true;
1100 }
1101 
1102 #define DO_DUAL_ACC(INSN, FN)                                           \
1103     static bool trans_##INSN(DisasContext *s, arg_vmladav *a)           \
1104     {                                                                   \
1105         static MVEGenDualAccOpFn * const fns[4][2] = {                  \
1106             { gen_helper_mve_##FN##b, gen_helper_mve_##FN##xb },        \
1107             { gen_helper_mve_##FN##h, gen_helper_mve_##FN##xh },        \
1108             { gen_helper_mve_##FN##w, gen_helper_mve_##FN##xw },        \
1109             { NULL, NULL },                                             \
1110         };                                                              \
1111         return do_dual_acc(s, a, fns[a->size][a->x]);                   \
1112     }
1113 
1114 DO_DUAL_ACC(VMLADAV_S, vmladavs)
1115 DO_DUAL_ACC(VMLSDAV, vmlsdav)
1116 
1117 static bool trans_VMLADAV_U(DisasContext *s, arg_vmladav *a)
1118 {
1119     static MVEGenDualAccOpFn * const fns[4][2] = {
1120         { gen_helper_mve_vmladavub, NULL },
1121         { gen_helper_mve_vmladavuh, NULL },
1122         { gen_helper_mve_vmladavuw, NULL },
1123         { NULL, NULL },
1124     };
1125     return do_dual_acc(s, a, fns[a->size][a->x]);
1126 }
1127 
1128 static void gen_vpst(DisasContext *s, uint32_t mask)
1129 {
1130     /*
1131      * Set the VPR mask fields. We take advantage of MASK01 and MASK23
1132      * being adjacent fields in the register.
1133      *
1134      * Updating the masks is not predicated, but it is subject to beat-wise
1135      * execution, and the mask is updated on the odd-numbered beats.
1136      * So if PSR.ECI says we should skip beat 1, we mustn't update the
1137      * 01 mask field.
1138      */
1139     TCGv_i32 vpr = load_cpu_field(v7m.vpr);
1140     switch (s->eci) {
1141     case ECI_NONE:
1142     case ECI_A0:
1143         /* Update both 01 and 23 fields */
1144         tcg_gen_deposit_i32(vpr, vpr,
1145                             tcg_constant_i32(mask | (mask << 4)),
1146                             R_V7M_VPR_MASK01_SHIFT,
1147                             R_V7M_VPR_MASK01_LENGTH + R_V7M_VPR_MASK23_LENGTH);
1148         break;
1149     case ECI_A0A1:
1150     case ECI_A0A1A2:
1151     case ECI_A0A1A2B0:
1152         /* Update only the 23 mask field */
1153         tcg_gen_deposit_i32(vpr, vpr,
1154                             tcg_constant_i32(mask),
1155                             R_V7M_VPR_MASK23_SHIFT, R_V7M_VPR_MASK23_LENGTH);
1156         break;
1157     default:
1158         g_assert_not_reached();
1159     }
1160     store_cpu_field(vpr, v7m.vpr);
1161 }
1162 
1163 static bool trans_VPST(DisasContext *s, arg_VPST *a)
1164 {
1165     /* mask == 0 is a "related encoding" */
1166     if (!dc_isar_feature(aa32_mve, s) || !a->mask) {
1167         return false;
1168     }
1169     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1170         return true;
1171     }
1172     gen_vpst(s, a->mask);
1173     mve_update_and_store_eci(s);
1174     return true;
1175 }
1176 
1177 static bool trans_VPNOT(DisasContext *s, arg_VPNOT *a)
1178 {
1179     /*
1180      * Invert the predicate in VPR.P0. We have call out to
1181      * a helper because this insn itself is beatwise and can
1182      * be predicated.
1183      */
1184     if (!dc_isar_feature(aa32_mve, s)) {
1185         return false;
1186     }
1187     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1188         return true;
1189     }
1190 
1191     gen_helper_mve_vpnot(cpu_env);
1192     mve_update_eci(s);
1193     return true;
1194 }
1195 
1196 static bool trans_VADDV(DisasContext *s, arg_VADDV *a)
1197 {
1198     /* VADDV: vector add across vector */
1199     static MVEGenVADDVFn * const fns[4][2] = {
1200         { gen_helper_mve_vaddvsb, gen_helper_mve_vaddvub },
1201         { gen_helper_mve_vaddvsh, gen_helper_mve_vaddvuh },
1202         { gen_helper_mve_vaddvsw, gen_helper_mve_vaddvuw },
1203         { NULL, NULL }
1204     };
1205     TCGv_ptr qm;
1206     TCGv_i32 rda;
1207 
1208     if (!dc_isar_feature(aa32_mve, s) ||
1209         a->size == 3) {
1210         return false;
1211     }
1212     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1213         return true;
1214     }
1215 
1216     /*
1217      * This insn is subject to beat-wise execution. Partial execution
1218      * of an A=0 (no-accumulate) insn which does not execute the first
1219      * beat must start with the current value of Rda, not zero.
1220      */
1221     if (a->a || mve_skip_first_beat(s)) {
1222         /* Accumulate input from Rda */
1223         rda = load_reg(s, a->rda);
1224     } else {
1225         /* Accumulate starting at zero */
1226         rda = tcg_const_i32(0);
1227     }
1228 
1229     qm = mve_qreg_ptr(a->qm);
1230     fns[a->size][a->u](rda, cpu_env, qm, rda);
1231     store_reg(s, a->rda, rda);
1232     tcg_temp_free_ptr(qm);
1233 
1234     mve_update_eci(s);
1235     return true;
1236 }
1237 
1238 static bool trans_VADDLV(DisasContext *s, arg_VADDLV *a)
1239 {
1240     /*
1241      * Vector Add Long Across Vector: accumulate the 32-bit
1242      * elements of the vector into a 64-bit result stored in
1243      * a pair of general-purpose registers.
1244      * No need to check Qm's bank: it is only 3 bits in decode.
1245      */
1246     TCGv_ptr qm;
1247     TCGv_i64 rda;
1248     TCGv_i32 rdalo, rdahi;
1249 
1250     if (!dc_isar_feature(aa32_mve, s)) {
1251         return false;
1252     }
1253     /*
1254      * rdahi == 13 is UNPREDICTABLE; rdahi == 15 is a related
1255      * encoding; rdalo always has bit 0 clear so cannot be 13 or 15.
1256      */
1257     if (a->rdahi == 13 || a->rdahi == 15) {
1258         return false;
1259     }
1260     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1261         return true;
1262     }
1263 
1264     /*
1265      * This insn is subject to beat-wise execution. Partial execution
1266      * of an A=0 (no-accumulate) insn which does not execute the first
1267      * beat must start with the current value of RdaHi:RdaLo, not zero.
1268      */
1269     if (a->a || mve_skip_first_beat(s)) {
1270         /* Accumulate input from RdaHi:RdaLo */
1271         rda = tcg_temp_new_i64();
1272         rdalo = load_reg(s, a->rdalo);
1273         rdahi = load_reg(s, a->rdahi);
1274         tcg_gen_concat_i32_i64(rda, rdalo, rdahi);
1275         tcg_temp_free_i32(rdalo);
1276         tcg_temp_free_i32(rdahi);
1277     } else {
1278         /* Accumulate starting at zero */
1279         rda = tcg_const_i64(0);
1280     }
1281 
1282     qm = mve_qreg_ptr(a->qm);
1283     if (a->u) {
1284         gen_helper_mve_vaddlv_u(rda, cpu_env, qm, rda);
1285     } else {
1286         gen_helper_mve_vaddlv_s(rda, cpu_env, qm, rda);
1287     }
1288     tcg_temp_free_ptr(qm);
1289 
1290     rdalo = tcg_temp_new_i32();
1291     rdahi = tcg_temp_new_i32();
1292     tcg_gen_extrl_i64_i32(rdalo, rda);
1293     tcg_gen_extrh_i64_i32(rdahi, rda);
1294     store_reg(s, a->rdalo, rdalo);
1295     store_reg(s, a->rdahi, rdahi);
1296     tcg_temp_free_i64(rda);
1297     mve_update_eci(s);
1298     return true;
1299 }
1300 
1301 static bool do_1imm(DisasContext *s, arg_1imm *a, MVEGenOneOpImmFn *fn)
1302 {
1303     TCGv_ptr qd;
1304     uint64_t imm;
1305 
1306     if (!dc_isar_feature(aa32_mve, s) ||
1307         !mve_check_qreg_bank(s, a->qd) ||
1308         !fn) {
1309         return false;
1310     }
1311     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1312         return true;
1313     }
1314 
1315     imm = asimd_imm_const(a->imm, a->cmode, a->op);
1316 
1317     qd = mve_qreg_ptr(a->qd);
1318     fn(cpu_env, qd, tcg_constant_i64(imm));
1319     tcg_temp_free_ptr(qd);
1320     mve_update_eci(s);
1321     return true;
1322 }
1323 
1324 static bool trans_Vimm_1r(DisasContext *s, arg_1imm *a)
1325 {
1326     /* Handle decode of cmode/op here between VORR/VBIC/VMOV */
1327     MVEGenOneOpImmFn *fn;
1328 
1329     if ((a->cmode & 1) && a->cmode < 12) {
1330         if (a->op) {
1331             /*
1332              * For op=1, the immediate will be inverted by asimd_imm_const(),
1333              * so the VBIC becomes a logical AND operation.
1334              */
1335             fn = gen_helper_mve_vandi;
1336         } else {
1337             fn = gen_helper_mve_vorri;
1338         }
1339     } else {
1340         /* There is one unallocated cmode/op combination in this space */
1341         if (a->cmode == 15 && a->op == 1) {
1342             return false;
1343         }
1344         /* asimd_imm_const() sorts out VMVNI vs VMOVI for us */
1345         fn = gen_helper_mve_vmovi;
1346     }
1347     return do_1imm(s, a, fn);
1348 }
1349 
1350 static bool do_2shift(DisasContext *s, arg_2shift *a, MVEGenTwoOpShiftFn fn,
1351                       bool negateshift)
1352 {
1353     TCGv_ptr qd, qm;
1354     int shift = a->shift;
1355 
1356     if (!dc_isar_feature(aa32_mve, s) ||
1357         !mve_check_qreg_bank(s, a->qd | a->qm) ||
1358         !fn) {
1359         return false;
1360     }
1361     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1362         return true;
1363     }
1364 
1365     /*
1366      * When we handle a right shift insn using a left-shift helper
1367      * which permits a negative shift count to indicate a right-shift,
1368      * we must negate the shift count.
1369      */
1370     if (negateshift) {
1371         shift = -shift;
1372     }
1373 
1374     qd = mve_qreg_ptr(a->qd);
1375     qm = mve_qreg_ptr(a->qm);
1376     fn(cpu_env, qd, qm, tcg_constant_i32(shift));
1377     tcg_temp_free_ptr(qd);
1378     tcg_temp_free_ptr(qm);
1379     mve_update_eci(s);
1380     return true;
1381 }
1382 
1383 #define DO_2SHIFT(INSN, FN, NEGATESHIFT)                         \
1384     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1385     {                                                           \
1386         static MVEGenTwoOpShiftFn * const fns[] = {             \
1387             gen_helper_mve_##FN##b,                             \
1388             gen_helper_mve_##FN##h,                             \
1389             gen_helper_mve_##FN##w,                             \
1390             NULL,                                               \
1391         };                                                      \
1392         return do_2shift(s, a, fns[a->size], NEGATESHIFT);      \
1393     }
1394 
1395 DO_2SHIFT(VSHLI, vshli_u, false)
1396 DO_2SHIFT(VQSHLI_S, vqshli_s, false)
1397 DO_2SHIFT(VQSHLI_U, vqshli_u, false)
1398 DO_2SHIFT(VQSHLUI, vqshlui_s, false)
1399 /* These right shifts use a left-shift helper with negated shift count */
1400 DO_2SHIFT(VSHRI_S, vshli_s, true)
1401 DO_2SHIFT(VSHRI_U, vshli_u, true)
1402 DO_2SHIFT(VRSHRI_S, vrshli_s, true)
1403 DO_2SHIFT(VRSHRI_U, vrshli_u, true)
1404 
1405 DO_2SHIFT(VSRI, vsri, false)
1406 DO_2SHIFT(VSLI, vsli, false)
1407 
1408 static bool do_2shift_scalar(DisasContext *s, arg_shl_scalar *a,
1409                              MVEGenTwoOpShiftFn *fn)
1410 {
1411     TCGv_ptr qda;
1412     TCGv_i32 rm;
1413 
1414     if (!dc_isar_feature(aa32_mve, s) ||
1415         !mve_check_qreg_bank(s, a->qda) ||
1416         a->rm == 13 || a->rm == 15 || !fn) {
1417         /* Rm cases are UNPREDICTABLE */
1418         return false;
1419     }
1420     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1421         return true;
1422     }
1423 
1424     qda = mve_qreg_ptr(a->qda);
1425     rm = load_reg(s, a->rm);
1426     fn(cpu_env, qda, qda, rm);
1427     tcg_temp_free_ptr(qda);
1428     tcg_temp_free_i32(rm);
1429     mve_update_eci(s);
1430     return true;
1431 }
1432 
1433 #define DO_2SHIFT_SCALAR(INSN, FN)                                      \
1434     static bool trans_##INSN(DisasContext *s, arg_shl_scalar *a)        \
1435     {                                                                   \
1436         static MVEGenTwoOpShiftFn * const fns[] = {                     \
1437             gen_helper_mve_##FN##b,                                     \
1438             gen_helper_mve_##FN##h,                                     \
1439             gen_helper_mve_##FN##w,                                     \
1440             NULL,                                                       \
1441         };                                                              \
1442         return do_2shift_scalar(s, a, fns[a->size]);                    \
1443     }
1444 
1445 DO_2SHIFT_SCALAR(VSHL_S_scalar, vshli_s)
1446 DO_2SHIFT_SCALAR(VSHL_U_scalar, vshli_u)
1447 DO_2SHIFT_SCALAR(VRSHL_S_scalar, vrshli_s)
1448 DO_2SHIFT_SCALAR(VRSHL_U_scalar, vrshli_u)
1449 DO_2SHIFT_SCALAR(VQSHL_S_scalar, vqshli_s)
1450 DO_2SHIFT_SCALAR(VQSHL_U_scalar, vqshli_u)
1451 DO_2SHIFT_SCALAR(VQRSHL_S_scalar, vqrshli_s)
1452 DO_2SHIFT_SCALAR(VQRSHL_U_scalar, vqrshli_u)
1453 
1454 #define DO_VSHLL(INSN, FN)                                      \
1455     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1456     {                                                           \
1457         static MVEGenTwoOpShiftFn * const fns[] = {             \
1458             gen_helper_mve_##FN##b,                             \
1459             gen_helper_mve_##FN##h,                             \
1460         };                                                      \
1461         return do_2shift(s, a, fns[a->size], false);            \
1462     }
1463 
1464 DO_VSHLL(VSHLL_BS, vshllbs)
1465 DO_VSHLL(VSHLL_BU, vshllbu)
1466 DO_VSHLL(VSHLL_TS, vshllts)
1467 DO_VSHLL(VSHLL_TU, vshlltu)
1468 
1469 #define DO_2SHIFT_N(INSN, FN)                                   \
1470     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1471     {                                                           \
1472         static MVEGenTwoOpShiftFn * const fns[] = {             \
1473             gen_helper_mve_##FN##b,                             \
1474             gen_helper_mve_##FN##h,                             \
1475         };                                                      \
1476         return do_2shift(s, a, fns[a->size], false);            \
1477     }
1478 
1479 DO_2SHIFT_N(VSHRNB, vshrnb)
1480 DO_2SHIFT_N(VSHRNT, vshrnt)
1481 DO_2SHIFT_N(VRSHRNB, vrshrnb)
1482 DO_2SHIFT_N(VRSHRNT, vrshrnt)
1483 DO_2SHIFT_N(VQSHRNB_S, vqshrnb_s)
1484 DO_2SHIFT_N(VQSHRNT_S, vqshrnt_s)
1485 DO_2SHIFT_N(VQSHRNB_U, vqshrnb_u)
1486 DO_2SHIFT_N(VQSHRNT_U, vqshrnt_u)
1487 DO_2SHIFT_N(VQSHRUNB, vqshrunb)
1488 DO_2SHIFT_N(VQSHRUNT, vqshrunt)
1489 DO_2SHIFT_N(VQRSHRNB_S, vqrshrnb_s)
1490 DO_2SHIFT_N(VQRSHRNT_S, vqrshrnt_s)
1491 DO_2SHIFT_N(VQRSHRNB_U, vqrshrnb_u)
1492 DO_2SHIFT_N(VQRSHRNT_U, vqrshrnt_u)
1493 DO_2SHIFT_N(VQRSHRUNB, vqrshrunb)
1494 DO_2SHIFT_N(VQRSHRUNT, vqrshrunt)
1495 
1496 static bool trans_VSHLC(DisasContext *s, arg_VSHLC *a)
1497 {
1498     /*
1499      * Whole Vector Left Shift with Carry. The carry is taken
1500      * from a general purpose register and written back there.
1501      * An imm of 0 means "shift by 32".
1502      */
1503     TCGv_ptr qd;
1504     TCGv_i32 rdm;
1505 
1506     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) {
1507         return false;
1508     }
1509     if (a->rdm == 13 || a->rdm == 15) {
1510         /* CONSTRAINED UNPREDICTABLE: we UNDEF */
1511         return false;
1512     }
1513     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1514         return true;
1515     }
1516 
1517     qd = mve_qreg_ptr(a->qd);
1518     rdm = load_reg(s, a->rdm);
1519     gen_helper_mve_vshlc(rdm, cpu_env, qd, rdm, tcg_constant_i32(a->imm));
1520     store_reg(s, a->rdm, rdm);
1521     tcg_temp_free_ptr(qd);
1522     mve_update_eci(s);
1523     return true;
1524 }
1525 
1526 static bool do_vidup(DisasContext *s, arg_vidup *a, MVEGenVIDUPFn *fn)
1527 {
1528     TCGv_ptr qd;
1529     TCGv_i32 rn;
1530 
1531     /*
1532      * Vector increment/decrement with wrap and duplicate (VIDUP, VDDUP).
1533      * This fills the vector with elements of successively increasing
1534      * or decreasing values, starting from Rn.
1535      */
1536     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) {
1537         return false;
1538     }
1539     if (a->size == MO_64) {
1540         /* size 0b11 is another encoding */
1541         return false;
1542     }
1543     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1544         return true;
1545     }
1546 
1547     qd = mve_qreg_ptr(a->qd);
1548     rn = load_reg(s, a->rn);
1549     fn(rn, cpu_env, qd, rn, tcg_constant_i32(a->imm));
1550     store_reg(s, a->rn, rn);
1551     tcg_temp_free_ptr(qd);
1552     mve_update_eci(s);
1553     return true;
1554 }
1555 
1556 static bool do_viwdup(DisasContext *s, arg_viwdup *a, MVEGenVIWDUPFn *fn)
1557 {
1558     TCGv_ptr qd;
1559     TCGv_i32 rn, rm;
1560 
1561     /*
1562      * Vector increment/decrement with wrap and duplicate (VIWDUp, VDWDUP)
1563      * This fills the vector with elements of successively increasing
1564      * or decreasing values, starting from Rn. Rm specifies a point where
1565      * the count wraps back around to 0. The updated offset is written back
1566      * to Rn.
1567      */
1568     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) {
1569         return false;
1570     }
1571     if (!fn || a->rm == 13 || a->rm == 15) {
1572         /*
1573          * size 0b11 is another encoding; Rm == 13 is UNPREDICTABLE;
1574          * Rm == 13 is VIWDUP, VDWDUP.
1575          */
1576         return false;
1577     }
1578     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1579         return true;
1580     }
1581 
1582     qd = mve_qreg_ptr(a->qd);
1583     rn = load_reg(s, a->rn);
1584     rm = load_reg(s, a->rm);
1585     fn(rn, cpu_env, qd, rn, rm, tcg_constant_i32(a->imm));
1586     store_reg(s, a->rn, rn);
1587     tcg_temp_free_ptr(qd);
1588     tcg_temp_free_i32(rm);
1589     mve_update_eci(s);
1590     return true;
1591 }
1592 
1593 static bool trans_VIDUP(DisasContext *s, arg_vidup *a)
1594 {
1595     static MVEGenVIDUPFn * const fns[] = {
1596         gen_helper_mve_vidupb,
1597         gen_helper_mve_viduph,
1598         gen_helper_mve_vidupw,
1599         NULL,
1600     };
1601     return do_vidup(s, a, fns[a->size]);
1602 }
1603 
1604 static bool trans_VDDUP(DisasContext *s, arg_vidup *a)
1605 {
1606     static MVEGenVIDUPFn * const fns[] = {
1607         gen_helper_mve_vidupb,
1608         gen_helper_mve_viduph,
1609         gen_helper_mve_vidupw,
1610         NULL,
1611     };
1612     /* VDDUP is just like VIDUP but with a negative immediate */
1613     a->imm = -a->imm;
1614     return do_vidup(s, a, fns[a->size]);
1615 }
1616 
1617 static bool trans_VIWDUP(DisasContext *s, arg_viwdup *a)
1618 {
1619     static MVEGenVIWDUPFn * const fns[] = {
1620         gen_helper_mve_viwdupb,
1621         gen_helper_mve_viwduph,
1622         gen_helper_mve_viwdupw,
1623         NULL,
1624     };
1625     return do_viwdup(s, a, fns[a->size]);
1626 }
1627 
1628 static bool trans_VDWDUP(DisasContext *s, arg_viwdup *a)
1629 {
1630     static MVEGenVIWDUPFn * const fns[] = {
1631         gen_helper_mve_vdwdupb,
1632         gen_helper_mve_vdwduph,
1633         gen_helper_mve_vdwdupw,
1634         NULL,
1635     };
1636     return do_viwdup(s, a, fns[a->size]);
1637 }
1638 
1639 static bool do_vcmp(DisasContext *s, arg_vcmp *a, MVEGenCmpFn *fn)
1640 {
1641     TCGv_ptr qn, qm;
1642 
1643     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qm) ||
1644         !fn) {
1645         return false;
1646     }
1647     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1648         return true;
1649     }
1650 
1651     qn = mve_qreg_ptr(a->qn);
1652     qm = mve_qreg_ptr(a->qm);
1653     fn(cpu_env, qn, qm);
1654     tcg_temp_free_ptr(qn);
1655     tcg_temp_free_ptr(qm);
1656     if (a->mask) {
1657         /* VPT */
1658         gen_vpst(s, a->mask);
1659     }
1660     mve_update_eci(s);
1661     return true;
1662 }
1663 
1664 static bool do_vcmp_scalar(DisasContext *s, arg_vcmp_scalar *a,
1665                            MVEGenScalarCmpFn *fn)
1666 {
1667     TCGv_ptr qn;
1668     TCGv_i32 rm;
1669 
1670     if (!dc_isar_feature(aa32_mve, s) || !fn || a->rm == 13) {
1671         return false;
1672     }
1673     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1674         return true;
1675     }
1676 
1677     qn = mve_qreg_ptr(a->qn);
1678     if (a->rm == 15) {
1679         /* Encoding Rm=0b1111 means "constant zero" */
1680         rm = tcg_constant_i32(0);
1681     } else {
1682         rm = load_reg(s, a->rm);
1683     }
1684     fn(cpu_env, qn, rm);
1685     tcg_temp_free_ptr(qn);
1686     tcg_temp_free_i32(rm);
1687     if (a->mask) {
1688         /* VPT */
1689         gen_vpst(s, a->mask);
1690     }
1691     mve_update_eci(s);
1692     return true;
1693 }
1694 
1695 #define DO_VCMP(INSN, FN)                                       \
1696     static bool trans_##INSN(DisasContext *s, arg_vcmp *a)      \
1697     {                                                           \
1698         static MVEGenCmpFn * const fns[] = {                    \
1699             gen_helper_mve_##FN##b,                             \
1700             gen_helper_mve_##FN##h,                             \
1701             gen_helper_mve_##FN##w,                             \
1702             NULL,                                               \
1703         };                                                      \
1704         return do_vcmp(s, a, fns[a->size]);                     \
1705     }                                                           \
1706     static bool trans_##INSN##_scalar(DisasContext *s,          \
1707                                       arg_vcmp_scalar *a)       \
1708     {                                                           \
1709         static MVEGenScalarCmpFn * const fns[] = {              \
1710             gen_helper_mve_##FN##_scalarb,                      \
1711             gen_helper_mve_##FN##_scalarh,                      \
1712             gen_helper_mve_##FN##_scalarw,                      \
1713             NULL,                                               \
1714         };                                                      \
1715         return do_vcmp_scalar(s, a, fns[a->size]);              \
1716     }
1717 
1718 DO_VCMP(VCMPEQ, vcmpeq)
1719 DO_VCMP(VCMPNE, vcmpne)
1720 DO_VCMP(VCMPCS, vcmpcs)
1721 DO_VCMP(VCMPHI, vcmphi)
1722 DO_VCMP(VCMPGE, vcmpge)
1723 DO_VCMP(VCMPLT, vcmplt)
1724 DO_VCMP(VCMPGT, vcmpgt)
1725 DO_VCMP(VCMPLE, vcmple)
1726 
1727 static bool do_vmaxv(DisasContext *s, arg_vmaxv *a, MVEGenVADDVFn fn)
1728 {
1729     /*
1730      * MIN/MAX operations across a vector: compute the min or
1731      * max of the initial value in a general purpose register
1732      * and all the elements in the vector, and store it back
1733      * into the general purpose register.
1734      */
1735     TCGv_ptr qm;
1736     TCGv_i32 rda;
1737 
1738     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qm) ||
1739         !fn || a->rda == 13 || a->rda == 15) {
1740         /* Rda cases are UNPREDICTABLE */
1741         return false;
1742     }
1743     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1744         return true;
1745     }
1746 
1747     qm = mve_qreg_ptr(a->qm);
1748     rda = load_reg(s, a->rda);
1749     fn(rda, cpu_env, qm, rda);
1750     store_reg(s, a->rda, rda);
1751     tcg_temp_free_ptr(qm);
1752     mve_update_eci(s);
1753     return true;
1754 }
1755 
1756 #define DO_VMAXV(INSN, FN)                                      \
1757     static bool trans_##INSN(DisasContext *s, arg_vmaxv *a)     \
1758     {                                                           \
1759         static MVEGenVADDVFn * const fns[] = {                  \
1760             gen_helper_mve_##FN##b,                             \
1761             gen_helper_mve_##FN##h,                             \
1762             gen_helper_mve_##FN##w,                             \
1763             NULL,                                               \
1764         };                                                      \
1765         return do_vmaxv(s, a, fns[a->size]);                    \
1766     }
1767 
1768 DO_VMAXV(VMAXV_S, vmaxvs)
1769 DO_VMAXV(VMAXV_U, vmaxvu)
1770 DO_VMAXV(VMAXAV, vmaxav)
1771 DO_VMAXV(VMINV_S, vminvs)
1772 DO_VMAXV(VMINV_U, vminvu)
1773 DO_VMAXV(VMINAV, vminav)
1774 
1775 static bool do_vabav(DisasContext *s, arg_vabav *a, MVEGenVABAVFn *fn)
1776 {
1777     /* Absolute difference accumulated across vector */
1778     TCGv_ptr qn, qm;
1779     TCGv_i32 rda;
1780 
1781     if (!dc_isar_feature(aa32_mve, s) ||
1782         !mve_check_qreg_bank(s, a->qm | a->qn) ||
1783         !fn || a->rda == 13 || a->rda == 15) {
1784         /* Rda cases are UNPREDICTABLE */
1785         return false;
1786     }
1787     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1788         return true;
1789     }
1790 
1791     qm = mve_qreg_ptr(a->qm);
1792     qn = mve_qreg_ptr(a->qn);
1793     rda = load_reg(s, a->rda);
1794     fn(rda, cpu_env, qn, qm, rda);
1795     store_reg(s, a->rda, rda);
1796     tcg_temp_free_ptr(qm);
1797     tcg_temp_free_ptr(qn);
1798     mve_update_eci(s);
1799     return true;
1800 }
1801 
1802 #define DO_VABAV(INSN, FN)                                      \
1803     static bool trans_##INSN(DisasContext *s, arg_vabav *a)     \
1804     {                                                           \
1805         static MVEGenVABAVFn * const fns[] = {                  \
1806             gen_helper_mve_##FN##b,                             \
1807             gen_helper_mve_##FN##h,                             \
1808             gen_helper_mve_##FN##w,                             \
1809             NULL,                                               \
1810         };                                                      \
1811         return do_vabav(s, a, fns[a->size]);                    \
1812     }
1813 
1814 DO_VABAV(VABAV_S, vabavs)
1815 DO_VABAV(VABAV_U, vabavu)
1816 
1817 static bool trans_VMOV_to_2gp(DisasContext *s, arg_VMOV_to_2gp *a)
1818 {
1819     /*
1820      * VMOV two 32-bit vector lanes to two general-purpose registers.
1821      * This insn is not predicated but it is subject to beat-wise
1822      * execution if it is not in an IT block. For us this means
1823      * only that if PSR.ECI says we should not be executing the beat
1824      * corresponding to the lane of the vector register being accessed
1825      * then we should skip perfoming the move, and that we need to do
1826      * the usual check for bad ECI state and advance of ECI state.
1827      * (If PSR.ECI is non-zero then we cannot be in an IT block.)
1828      */
1829     TCGv_i32 tmp;
1830     int vd;
1831 
1832     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd) ||
1833         a->rt == 13 || a->rt == 15 || a->rt2 == 13 || a->rt2 == 15 ||
1834         a->rt == a->rt2) {
1835         /* Rt/Rt2 cases are UNPREDICTABLE */
1836         return false;
1837     }
1838     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1839         return true;
1840     }
1841 
1842     /* Convert Qreg index to Dreg for read_neon_element32() etc */
1843     vd = a->qd * 2;
1844 
1845     if (!mve_skip_vmov(s, vd, a->idx, MO_32)) {
1846         tmp = tcg_temp_new_i32();
1847         read_neon_element32(tmp, vd, a->idx, MO_32);
1848         store_reg(s, a->rt, tmp);
1849     }
1850     if (!mve_skip_vmov(s, vd + 1, a->idx, MO_32)) {
1851         tmp = tcg_temp_new_i32();
1852         read_neon_element32(tmp, vd + 1, a->idx, MO_32);
1853         store_reg(s, a->rt2, tmp);
1854     }
1855 
1856     mve_update_and_store_eci(s);
1857     return true;
1858 }
1859 
1860 static bool trans_VMOV_from_2gp(DisasContext *s, arg_VMOV_to_2gp *a)
1861 {
1862     /*
1863      * VMOV two general-purpose registers to two 32-bit vector lanes.
1864      * This insn is not predicated but it is subject to beat-wise
1865      * execution if it is not in an IT block. For us this means
1866      * only that if PSR.ECI says we should not be executing the beat
1867      * corresponding to the lane of the vector register being accessed
1868      * then we should skip perfoming the move, and that we need to do
1869      * the usual check for bad ECI state and advance of ECI state.
1870      * (If PSR.ECI is non-zero then we cannot be in an IT block.)
1871      */
1872     TCGv_i32 tmp;
1873     int vd;
1874 
1875     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd) ||
1876         a->rt == 13 || a->rt == 15 || a->rt2 == 13 || a->rt2 == 15) {
1877         /* Rt/Rt2 cases are UNPREDICTABLE */
1878         return false;
1879     }
1880     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1881         return true;
1882     }
1883 
1884     /* Convert Qreg idx to Dreg for read_neon_element32() etc */
1885     vd = a->qd * 2;
1886 
1887     if (!mve_skip_vmov(s, vd, a->idx, MO_32)) {
1888         tmp = load_reg(s, a->rt);
1889         write_neon_element32(tmp, vd, a->idx, MO_32);
1890         tcg_temp_free_i32(tmp);
1891     }
1892     if (!mve_skip_vmov(s, vd + 1, a->idx, MO_32)) {
1893         tmp = load_reg(s, a->rt2);
1894         write_neon_element32(tmp, vd + 1, a->idx, MO_32);
1895         tcg_temp_free_i32(tmp);
1896     }
1897 
1898     mve_update_and_store_eci(s);
1899     return true;
1900 }
1901