xref: /qemu/target/arm/tcg/translate-mve.c (revision 451f9d66cf114f5263d6bd12a21c33896dd32f80)
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 typedef void MVEGenVCVTRmodeFn(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_i32);
53 
54 /* Return the offset of a Qn register (same semantics as aa32_vfp_qreg()) */
55 static inline long mve_qreg_offset(unsigned reg)
56 {
57     return offsetof(CPUARMState, vfp.zregs[reg].d[0]);
58 }
59 
60 static TCGv_ptr mve_qreg_ptr(unsigned reg)
61 {
62     TCGv_ptr ret = tcg_temp_new_ptr();
63     tcg_gen_addi_ptr(ret, cpu_env, mve_qreg_offset(reg));
64     return ret;
65 }
66 
67 static bool mve_no_predication(DisasContext *s)
68 {
69     /*
70      * Return true if we are executing the entire MVE instruction
71      * with no predication or partial-execution, and so we can safely
72      * use an inline TCG vector implementation.
73      */
74     return s->eci == 0 && s->mve_no_pred;
75 }
76 
77 static bool mve_check_qreg_bank(DisasContext *s, int qmask)
78 {
79     /*
80      * Check whether Qregs are in range. For v8.1M only Q0..Q7
81      * are supported, see VFPSmallRegisterBank().
82      */
83     return qmask < 8;
84 }
85 
86 bool mve_eci_check(DisasContext *s)
87 {
88     /*
89      * This is a beatwise insn: check that ECI is valid (not a
90      * reserved value) and note that we are handling it.
91      * Return true if OK, false if we generated an exception.
92      */
93     s->eci_handled = true;
94     switch (s->eci) {
95     case ECI_NONE:
96     case ECI_A0:
97     case ECI_A0A1:
98     case ECI_A0A1A2:
99     case ECI_A0A1A2B0:
100         return true;
101     default:
102         /* Reserved value: INVSTATE UsageFault */
103         gen_exception_insn(s, s->pc_curr, EXCP_INVSTATE, syn_uncategorized(),
104                            default_exception_el(s));
105         return false;
106     }
107 }
108 
109 void mve_update_eci(DisasContext *s)
110 {
111     /*
112      * The helper function will always update the CPUState field,
113      * so we only need to update the DisasContext field.
114      */
115     if (s->eci) {
116         s->eci = (s->eci == ECI_A0A1A2B0) ? ECI_A0 : ECI_NONE;
117     }
118 }
119 
120 void mve_update_and_store_eci(DisasContext *s)
121 {
122     /*
123      * For insns which don't call a helper function that will call
124      * mve_advance_vpt(), this version updates s->eci and also stores
125      * it out to the CPUState field.
126      */
127     if (s->eci) {
128         mve_update_eci(s);
129         store_cpu_field(tcg_constant_i32(s->eci << 4), condexec_bits);
130     }
131 }
132 
133 static bool mve_skip_first_beat(DisasContext *s)
134 {
135     /* Return true if PSR.ECI says we must skip the first beat of this insn */
136     switch (s->eci) {
137     case ECI_NONE:
138         return false;
139     case ECI_A0:
140     case ECI_A0A1:
141     case ECI_A0A1A2:
142     case ECI_A0A1A2B0:
143         return true;
144     default:
145         g_assert_not_reached();
146     }
147 }
148 
149 static bool do_ldst(DisasContext *s, arg_VLDR_VSTR *a, MVEGenLdStFn *fn,
150                     unsigned msize)
151 {
152     TCGv_i32 addr;
153     uint32_t offset;
154     TCGv_ptr qreg;
155 
156     if (!dc_isar_feature(aa32_mve, s) ||
157         !mve_check_qreg_bank(s, a->qd) ||
158         !fn) {
159         return false;
160     }
161 
162     /* CONSTRAINED UNPREDICTABLE: we choose to UNDEF */
163     if (a->rn == 15 || (a->rn == 13 && a->w)) {
164         return false;
165     }
166 
167     if (!mve_eci_check(s) || !vfp_access_check(s)) {
168         return true;
169     }
170 
171     offset = a->imm << msize;
172     if (!a->a) {
173         offset = -offset;
174     }
175     addr = load_reg(s, a->rn);
176     if (a->p) {
177         tcg_gen_addi_i32(addr, addr, offset);
178     }
179 
180     qreg = mve_qreg_ptr(a->qd);
181     fn(cpu_env, qreg, addr);
182     tcg_temp_free_ptr(qreg);
183 
184     /*
185      * Writeback always happens after the last beat of the insn,
186      * regardless of predication
187      */
188     if (a->w) {
189         if (!a->p) {
190             tcg_gen_addi_i32(addr, addr, offset);
191         }
192         store_reg(s, a->rn, addr);
193     } else {
194         tcg_temp_free_i32(addr);
195     }
196     mve_update_eci(s);
197     return true;
198 }
199 
200 static bool trans_VLDR_VSTR(DisasContext *s, arg_VLDR_VSTR *a)
201 {
202     static MVEGenLdStFn * const ldstfns[4][2] = {
203         { gen_helper_mve_vstrb, gen_helper_mve_vldrb },
204         { gen_helper_mve_vstrh, gen_helper_mve_vldrh },
205         { gen_helper_mve_vstrw, gen_helper_mve_vldrw },
206         { NULL, NULL }
207     };
208     return do_ldst(s, a, ldstfns[a->size][a->l], a->size);
209 }
210 
211 #define DO_VLDST_WIDE_NARROW(OP, SLD, ULD, ST, MSIZE)           \
212     static bool trans_##OP(DisasContext *s, arg_VLDR_VSTR *a)   \
213     {                                                           \
214         static MVEGenLdStFn * const ldstfns[2][2] = {           \
215             { gen_helper_mve_##ST, gen_helper_mve_##SLD },      \
216             { NULL, gen_helper_mve_##ULD },                     \
217         };                                                      \
218         return do_ldst(s, a, ldstfns[a->u][a->l], MSIZE);       \
219     }
220 
221 DO_VLDST_WIDE_NARROW(VLDSTB_H, vldrb_sh, vldrb_uh, vstrb_h, MO_8)
222 DO_VLDST_WIDE_NARROW(VLDSTB_W, vldrb_sw, vldrb_uw, vstrb_w, MO_8)
223 DO_VLDST_WIDE_NARROW(VLDSTH_W, vldrh_sw, vldrh_uw, vstrh_w, MO_16)
224 
225 static bool do_ldst_sg(DisasContext *s, arg_vldst_sg *a, MVEGenLdStSGFn fn)
226 {
227     TCGv_i32 addr;
228     TCGv_ptr qd, qm;
229 
230     if (!dc_isar_feature(aa32_mve, s) ||
231         !mve_check_qreg_bank(s, a->qd | a->qm) ||
232         !fn || a->rn == 15) {
233         /* Rn case is UNPREDICTABLE */
234         return false;
235     }
236 
237     if (!mve_eci_check(s) || !vfp_access_check(s)) {
238         return true;
239     }
240 
241     addr = load_reg(s, a->rn);
242 
243     qd = mve_qreg_ptr(a->qd);
244     qm = mve_qreg_ptr(a->qm);
245     fn(cpu_env, qd, qm, addr);
246     tcg_temp_free_ptr(qd);
247     tcg_temp_free_ptr(qm);
248     tcg_temp_free_i32(addr);
249     mve_update_eci(s);
250     return true;
251 }
252 
253 /*
254  * The naming scheme here is "vldrb_sg_sh == in-memory byte loads
255  * signextended to halfword elements in register". _os_ indicates that
256  * the offsets in Qm should be scaled by the element size.
257  */
258 /* This macro is just to make the arrays more compact in these functions */
259 #define F(N) gen_helper_mve_##N
260 
261 /* VLDRB/VSTRB (ie msize 1) with OS=1 is UNPREDICTABLE; we UNDEF */
262 static bool trans_VLDR_S_sg(DisasContext *s, arg_vldst_sg *a)
263 {
264     static MVEGenLdStSGFn * const fns[2][4][4] = { {
265             { NULL, F(vldrb_sg_sh), F(vldrb_sg_sw), NULL },
266             { NULL, NULL,           F(vldrh_sg_sw), NULL },
267             { NULL, NULL,           NULL,           NULL },
268             { NULL, NULL,           NULL,           NULL }
269         }, {
270             { NULL, NULL,              NULL,              NULL },
271             { NULL, NULL,              F(vldrh_sg_os_sw), NULL },
272             { NULL, NULL,              NULL,              NULL },
273             { NULL, NULL,              NULL,              NULL }
274         }
275     };
276     if (a->qd == a->qm) {
277         return false; /* UNPREDICTABLE */
278     }
279     return do_ldst_sg(s, a, fns[a->os][a->msize][a->size]);
280 }
281 
282 static bool trans_VLDR_U_sg(DisasContext *s, arg_vldst_sg *a)
283 {
284     static MVEGenLdStSGFn * const fns[2][4][4] = { {
285             { F(vldrb_sg_ub), F(vldrb_sg_uh), F(vldrb_sg_uw), NULL },
286             { NULL,           F(vldrh_sg_uh), F(vldrh_sg_uw), NULL },
287             { NULL,           NULL,           F(vldrw_sg_uw), NULL },
288             { NULL,           NULL,           NULL,           F(vldrd_sg_ud) }
289         }, {
290             { NULL, NULL,              NULL,              NULL },
291             { NULL, F(vldrh_sg_os_uh), F(vldrh_sg_os_uw), NULL },
292             { NULL, NULL,              F(vldrw_sg_os_uw), NULL },
293             { NULL, NULL,              NULL,              F(vldrd_sg_os_ud) }
294         }
295     };
296     if (a->qd == a->qm) {
297         return false; /* UNPREDICTABLE */
298     }
299     return do_ldst_sg(s, a, fns[a->os][a->msize][a->size]);
300 }
301 
302 static bool trans_VSTR_sg(DisasContext *s, arg_vldst_sg *a)
303 {
304     static MVEGenLdStSGFn * const fns[2][4][4] = { {
305             { F(vstrb_sg_ub), F(vstrb_sg_uh), F(vstrb_sg_uw), NULL },
306             { NULL,           F(vstrh_sg_uh), F(vstrh_sg_uw), NULL },
307             { NULL,           NULL,           F(vstrw_sg_uw), NULL },
308             { NULL,           NULL,           NULL,           F(vstrd_sg_ud) }
309         }, {
310             { NULL, NULL,              NULL,              NULL },
311             { NULL, F(vstrh_sg_os_uh), F(vstrh_sg_os_uw), NULL },
312             { NULL, NULL,              F(vstrw_sg_os_uw), NULL },
313             { NULL, NULL,              NULL,              F(vstrd_sg_os_ud) }
314         }
315     };
316     return do_ldst_sg(s, a, fns[a->os][a->msize][a->size]);
317 }
318 
319 #undef F
320 
321 static bool do_ldst_sg_imm(DisasContext *s, arg_vldst_sg_imm *a,
322                            MVEGenLdStSGFn *fn, unsigned msize)
323 {
324     uint32_t offset;
325     TCGv_ptr qd, qm;
326 
327     if (!dc_isar_feature(aa32_mve, s) ||
328         !mve_check_qreg_bank(s, a->qd | a->qm) ||
329         !fn) {
330         return false;
331     }
332 
333     if (!mve_eci_check(s) || !vfp_access_check(s)) {
334         return true;
335     }
336 
337     offset = a->imm << msize;
338     if (!a->a) {
339         offset = -offset;
340     }
341 
342     qd = mve_qreg_ptr(a->qd);
343     qm = mve_qreg_ptr(a->qm);
344     fn(cpu_env, qd, qm, tcg_constant_i32(offset));
345     tcg_temp_free_ptr(qd);
346     tcg_temp_free_ptr(qm);
347     mve_update_eci(s);
348     return true;
349 }
350 
351 static bool trans_VLDRW_sg_imm(DisasContext *s, arg_vldst_sg_imm *a)
352 {
353     static MVEGenLdStSGFn * const fns[] = {
354         gen_helper_mve_vldrw_sg_uw,
355         gen_helper_mve_vldrw_sg_wb_uw,
356     };
357     if (a->qd == a->qm) {
358         return false; /* UNPREDICTABLE */
359     }
360     return do_ldst_sg_imm(s, a, fns[a->w], MO_32);
361 }
362 
363 static bool trans_VLDRD_sg_imm(DisasContext *s, arg_vldst_sg_imm *a)
364 {
365     static MVEGenLdStSGFn * const fns[] = {
366         gen_helper_mve_vldrd_sg_ud,
367         gen_helper_mve_vldrd_sg_wb_ud,
368     };
369     if (a->qd == a->qm) {
370         return false; /* UNPREDICTABLE */
371     }
372     return do_ldst_sg_imm(s, a, fns[a->w], MO_64);
373 }
374 
375 static bool trans_VSTRW_sg_imm(DisasContext *s, arg_vldst_sg_imm *a)
376 {
377     static MVEGenLdStSGFn * const fns[] = {
378         gen_helper_mve_vstrw_sg_uw,
379         gen_helper_mve_vstrw_sg_wb_uw,
380     };
381     return do_ldst_sg_imm(s, a, fns[a->w], MO_32);
382 }
383 
384 static bool trans_VSTRD_sg_imm(DisasContext *s, arg_vldst_sg_imm *a)
385 {
386     static MVEGenLdStSGFn * const fns[] = {
387         gen_helper_mve_vstrd_sg_ud,
388         gen_helper_mve_vstrd_sg_wb_ud,
389     };
390     return do_ldst_sg_imm(s, a, fns[a->w], MO_64);
391 }
392 
393 static bool do_vldst_il(DisasContext *s, arg_vldst_il *a, MVEGenLdStIlFn *fn,
394                         int addrinc)
395 {
396     TCGv_i32 rn;
397 
398     if (!dc_isar_feature(aa32_mve, s) ||
399         !mve_check_qreg_bank(s, a->qd) ||
400         !fn || (a->rn == 13 && a->w) || a->rn == 15) {
401         /* Variously UNPREDICTABLE or UNDEF or related-encoding */
402         return false;
403     }
404     if (!mve_eci_check(s) || !vfp_access_check(s)) {
405         return true;
406     }
407 
408     rn = load_reg(s, a->rn);
409     /*
410      * We pass the index of Qd, not a pointer, because the helper must
411      * access multiple Q registers starting at Qd and working up.
412      */
413     fn(cpu_env, tcg_constant_i32(a->qd), rn);
414 
415     if (a->w) {
416         tcg_gen_addi_i32(rn, rn, addrinc);
417         store_reg(s, a->rn, rn);
418     } else {
419         tcg_temp_free_i32(rn);
420     }
421     mve_update_and_store_eci(s);
422     return true;
423 }
424 
425 /* This macro is just to make the arrays more compact in these functions */
426 #define F(N) gen_helper_mve_##N
427 
428 static bool trans_VLD2(DisasContext *s, arg_vldst_il *a)
429 {
430     static MVEGenLdStIlFn * const fns[4][4] = {
431         { F(vld20b), F(vld20h), F(vld20w), NULL, },
432         { F(vld21b), F(vld21h), F(vld21w), NULL, },
433         { NULL, NULL, NULL, NULL },
434         { NULL, NULL, NULL, NULL },
435     };
436     if (a->qd > 6) {
437         return false;
438     }
439     return do_vldst_il(s, a, fns[a->pat][a->size], 32);
440 }
441 
442 static bool trans_VLD4(DisasContext *s, arg_vldst_il *a)
443 {
444     static MVEGenLdStIlFn * const fns[4][4] = {
445         { F(vld40b), F(vld40h), F(vld40w), NULL, },
446         { F(vld41b), F(vld41h), F(vld41w), NULL, },
447         { F(vld42b), F(vld42h), F(vld42w), NULL, },
448         { F(vld43b), F(vld43h), F(vld43w), NULL, },
449     };
450     if (a->qd > 4) {
451         return false;
452     }
453     return do_vldst_il(s, a, fns[a->pat][a->size], 64);
454 }
455 
456 static bool trans_VST2(DisasContext *s, arg_vldst_il *a)
457 {
458     static MVEGenLdStIlFn * const fns[4][4] = {
459         { F(vst20b), F(vst20h), F(vst20w), NULL, },
460         { F(vst21b), F(vst21h), F(vst21w), NULL, },
461         { NULL, NULL, NULL, NULL },
462         { NULL, NULL, NULL, NULL },
463     };
464     if (a->qd > 6) {
465         return false;
466     }
467     return do_vldst_il(s, a, fns[a->pat][a->size], 32);
468 }
469 
470 static bool trans_VST4(DisasContext *s, arg_vldst_il *a)
471 {
472     static MVEGenLdStIlFn * const fns[4][4] = {
473         { F(vst40b), F(vst40h), F(vst40w), NULL, },
474         { F(vst41b), F(vst41h), F(vst41w), NULL, },
475         { F(vst42b), F(vst42h), F(vst42w), NULL, },
476         { F(vst43b), F(vst43h), F(vst43w), NULL, },
477     };
478     if (a->qd > 4) {
479         return false;
480     }
481     return do_vldst_il(s, a, fns[a->pat][a->size], 64);
482 }
483 
484 #undef F
485 
486 static bool trans_VDUP(DisasContext *s, arg_VDUP *a)
487 {
488     TCGv_ptr qd;
489     TCGv_i32 rt;
490 
491     if (!dc_isar_feature(aa32_mve, s) ||
492         !mve_check_qreg_bank(s, a->qd)) {
493         return false;
494     }
495     if (a->rt == 13 || a->rt == 15) {
496         /* UNPREDICTABLE; we choose to UNDEF */
497         return false;
498     }
499     if (!mve_eci_check(s) || !vfp_access_check(s)) {
500         return true;
501     }
502 
503     qd = mve_qreg_ptr(a->qd);
504     rt = load_reg(s, a->rt);
505     tcg_gen_dup_i32(a->size, rt, rt);
506     gen_helper_mve_vdup(cpu_env, qd, rt);
507     tcg_temp_free_ptr(qd);
508     tcg_temp_free_i32(rt);
509     mve_update_eci(s);
510     return true;
511 }
512 
513 static bool do_1op(DisasContext *s, arg_1op *a, MVEGenOneOpFn fn)
514 {
515     TCGv_ptr qd, qm;
516 
517     if (!dc_isar_feature(aa32_mve, s) ||
518         !mve_check_qreg_bank(s, a->qd | a->qm) ||
519         !fn) {
520         return false;
521     }
522 
523     if (!mve_eci_check(s) || !vfp_access_check(s)) {
524         return true;
525     }
526 
527     qd = mve_qreg_ptr(a->qd);
528     qm = mve_qreg_ptr(a->qm);
529     fn(cpu_env, qd, qm);
530     tcg_temp_free_ptr(qd);
531     tcg_temp_free_ptr(qm);
532     mve_update_eci(s);
533     return true;
534 }
535 
536 #define DO_1OP(INSN, FN)                                        \
537     static bool trans_##INSN(DisasContext *s, arg_1op *a)       \
538     {                                                           \
539         static MVEGenOneOpFn * const fns[] = {                  \
540             gen_helper_mve_##FN##b,                             \
541             gen_helper_mve_##FN##h,                             \
542             gen_helper_mve_##FN##w,                             \
543             NULL,                                               \
544         };                                                      \
545         return do_1op(s, a, fns[a->size]);                      \
546     }
547 
548 DO_1OP(VCLZ, vclz)
549 DO_1OP(VCLS, vcls)
550 DO_1OP(VABS, vabs)
551 DO_1OP(VNEG, vneg)
552 DO_1OP(VQABS, vqabs)
553 DO_1OP(VQNEG, vqneg)
554 DO_1OP(VMAXA, vmaxa)
555 DO_1OP(VMINA, vmina)
556 
557 /*
558  * For simple float/int conversions we use the fixed-point
559  * conversion helpers with a zero shift count
560  */
561 #define DO_VCVT(INSN, HFN, SFN)                                         \
562     static void gen_##INSN##h(TCGv_ptr env, TCGv_ptr qd, TCGv_ptr qm)   \
563     {                                                                   \
564         gen_helper_mve_##HFN(env, qd, qm, tcg_constant_i32(0));         \
565     }                                                                   \
566     static void gen_##INSN##s(TCGv_ptr env, TCGv_ptr qd, TCGv_ptr qm)   \
567     {                                                                   \
568         gen_helper_mve_##SFN(env, qd, qm, tcg_constant_i32(0));         \
569     }                                                                   \
570     static bool trans_##INSN(DisasContext *s, arg_1op *a)               \
571     {                                                                   \
572         static MVEGenOneOpFn * const fns[] = {                          \
573             NULL,                                                       \
574             gen_##INSN##h,                                              \
575             gen_##INSN##s,                                              \
576             NULL,                                                       \
577         };                                                              \
578         if (!dc_isar_feature(aa32_mve_fp, s)) {                         \
579             return false;                                               \
580         }                                                               \
581         return do_1op(s, a, fns[a->size]);                              \
582     }
583 
584 DO_VCVT(VCVT_SF, vcvt_sh, vcvt_sf)
585 DO_VCVT(VCVT_UF, vcvt_uh, vcvt_uf)
586 DO_VCVT(VCVT_FS, vcvt_hs, vcvt_fs)
587 DO_VCVT(VCVT_FU, vcvt_hu, vcvt_fu)
588 
589 static bool do_vcvt_rmode(DisasContext *s, arg_1op *a,
590                           enum arm_fprounding rmode, bool u)
591 {
592     /*
593      * Handle VCVT fp to int with specified rounding mode.
594      * This is a 1op fn but we must pass the rounding mode as
595      * an immediate to the helper.
596      */
597     TCGv_ptr qd, qm;
598     static MVEGenVCVTRmodeFn * const fns[4][2] = {
599         { NULL, NULL },
600         { gen_helper_mve_vcvt_rm_sh, gen_helper_mve_vcvt_rm_uh },
601         { gen_helper_mve_vcvt_rm_ss, gen_helper_mve_vcvt_rm_us },
602         { NULL, NULL },
603     };
604     MVEGenVCVTRmodeFn *fn = fns[a->size][u];
605 
606     if (!dc_isar_feature(aa32_mve_fp, s) ||
607         !mve_check_qreg_bank(s, a->qd | a->qm) ||
608         !fn) {
609         return false;
610     }
611 
612     if (!mve_eci_check(s) || !vfp_access_check(s)) {
613         return true;
614     }
615 
616     qd = mve_qreg_ptr(a->qd);
617     qm = mve_qreg_ptr(a->qm);
618     fn(cpu_env, qd, qm, tcg_constant_i32(arm_rmode_to_sf(rmode)));
619     tcg_temp_free_ptr(qd);
620     tcg_temp_free_ptr(qm);
621     mve_update_eci(s);
622     return true;
623 }
624 
625 #define DO_VCVT_RMODE(INSN, RMODE, U)                           \
626     static bool trans_##INSN(DisasContext *s, arg_1op *a)       \
627     {                                                           \
628         return do_vcvt_rmode(s, a, RMODE, U);                   \
629     }                                                           \
630 
631 DO_VCVT_RMODE(VCVTAS, FPROUNDING_TIEAWAY, false)
632 DO_VCVT_RMODE(VCVTAU, FPROUNDING_TIEAWAY, true)
633 DO_VCVT_RMODE(VCVTNS, FPROUNDING_TIEEVEN, false)
634 DO_VCVT_RMODE(VCVTNU, FPROUNDING_TIEEVEN, true)
635 DO_VCVT_RMODE(VCVTPS, FPROUNDING_POSINF, false)
636 DO_VCVT_RMODE(VCVTPU, FPROUNDING_POSINF, true)
637 DO_VCVT_RMODE(VCVTMS, FPROUNDING_NEGINF, false)
638 DO_VCVT_RMODE(VCVTMU, FPROUNDING_NEGINF, true)
639 
640 #define DO_VCVT_SH(INSN, FN)                                    \
641     static bool trans_##INSN(DisasContext *s, arg_1op *a)       \
642     {                                                           \
643         if (!dc_isar_feature(aa32_mve_fp, s)) {                 \
644             return false;                                       \
645         }                                                       \
646         return do_1op(s, a, gen_helper_mve_##FN);               \
647     }                                                           \
648 
649 DO_VCVT_SH(VCVTB_SH, vcvtb_sh)
650 DO_VCVT_SH(VCVTT_SH, vcvtt_sh)
651 DO_VCVT_SH(VCVTB_HS, vcvtb_hs)
652 DO_VCVT_SH(VCVTT_HS, vcvtt_hs)
653 
654 #define DO_VRINT(INSN, RMODE)                                           \
655     static void gen_##INSN##h(TCGv_ptr env, TCGv_ptr qd, TCGv_ptr qm)   \
656     {                                                                   \
657         gen_helper_mve_vrint_rm_h(env, qd, qm,                          \
658                                   tcg_constant_i32(arm_rmode_to_sf(RMODE))); \
659     }                                                                   \
660     static void gen_##INSN##s(TCGv_ptr env, TCGv_ptr qd, TCGv_ptr qm)   \
661     {                                                                   \
662         gen_helper_mve_vrint_rm_s(env, qd, qm,                          \
663                                   tcg_constant_i32(arm_rmode_to_sf(RMODE))); \
664     }                                                                   \
665     static bool trans_##INSN(DisasContext *s, arg_1op *a)               \
666     {                                                                   \
667         static MVEGenOneOpFn * const fns[] = {                          \
668             NULL,                                                       \
669             gen_##INSN##h,                                              \
670             gen_##INSN##s,                                              \
671             NULL,                                                       \
672         };                                                              \
673         if (!dc_isar_feature(aa32_mve_fp, s)) {                         \
674             return false;                                               \
675         }                                                               \
676         return do_1op(s, a, fns[a->size]);                              \
677     }
678 
679 DO_VRINT(VRINTN, FPROUNDING_TIEEVEN)
680 DO_VRINT(VRINTA, FPROUNDING_TIEAWAY)
681 DO_VRINT(VRINTZ, FPROUNDING_ZERO)
682 DO_VRINT(VRINTM, FPROUNDING_NEGINF)
683 DO_VRINT(VRINTP, FPROUNDING_POSINF)
684 
685 static bool trans_VRINTX(DisasContext *s, arg_1op *a)
686 {
687     static MVEGenOneOpFn * const fns[] = {
688         NULL,
689         gen_helper_mve_vrintx_h,
690         gen_helper_mve_vrintx_s,
691         NULL,
692     };
693     if (!dc_isar_feature(aa32_mve_fp, s)) {
694         return false;
695     }
696     return do_1op(s, a, fns[a->size]);
697 }
698 
699 /* Narrowing moves: only size 0 and 1 are valid */
700 #define DO_VMOVN(INSN, FN) \
701     static bool trans_##INSN(DisasContext *s, arg_1op *a)       \
702     {                                                           \
703         static MVEGenOneOpFn * const fns[] = {                  \
704             gen_helper_mve_##FN##b,                             \
705             gen_helper_mve_##FN##h,                             \
706             NULL,                                               \
707             NULL,                                               \
708         };                                                      \
709         return do_1op(s, a, fns[a->size]);                      \
710     }
711 
712 DO_VMOVN(VMOVNB, vmovnb)
713 DO_VMOVN(VMOVNT, vmovnt)
714 DO_VMOVN(VQMOVUNB, vqmovunb)
715 DO_VMOVN(VQMOVUNT, vqmovunt)
716 DO_VMOVN(VQMOVN_BS, vqmovnbs)
717 DO_VMOVN(VQMOVN_TS, vqmovnts)
718 DO_VMOVN(VQMOVN_BU, vqmovnbu)
719 DO_VMOVN(VQMOVN_TU, vqmovntu)
720 
721 static bool trans_VREV16(DisasContext *s, arg_1op *a)
722 {
723     static MVEGenOneOpFn * const fns[] = {
724         gen_helper_mve_vrev16b,
725         NULL,
726         NULL,
727         NULL,
728     };
729     return do_1op(s, a, fns[a->size]);
730 }
731 
732 static bool trans_VREV32(DisasContext *s, arg_1op *a)
733 {
734     static MVEGenOneOpFn * const fns[] = {
735         gen_helper_mve_vrev32b,
736         gen_helper_mve_vrev32h,
737         NULL,
738         NULL,
739     };
740     return do_1op(s, a, fns[a->size]);
741 }
742 
743 static bool trans_VREV64(DisasContext *s, arg_1op *a)
744 {
745     static MVEGenOneOpFn * const fns[] = {
746         gen_helper_mve_vrev64b,
747         gen_helper_mve_vrev64h,
748         gen_helper_mve_vrev64w,
749         NULL,
750     };
751     return do_1op(s, a, fns[a->size]);
752 }
753 
754 static bool trans_VMVN(DisasContext *s, arg_1op *a)
755 {
756     return do_1op(s, a, gen_helper_mve_vmvn);
757 }
758 
759 static bool trans_VABS_fp(DisasContext *s, arg_1op *a)
760 {
761     static MVEGenOneOpFn * const fns[] = {
762         NULL,
763         gen_helper_mve_vfabsh,
764         gen_helper_mve_vfabss,
765         NULL,
766     };
767     if (!dc_isar_feature(aa32_mve_fp, s)) {
768         return false;
769     }
770     return do_1op(s, a, fns[a->size]);
771 }
772 
773 static bool trans_VNEG_fp(DisasContext *s, arg_1op *a)
774 {
775     static MVEGenOneOpFn * const fns[] = {
776         NULL,
777         gen_helper_mve_vfnegh,
778         gen_helper_mve_vfnegs,
779         NULL,
780     };
781     if (!dc_isar_feature(aa32_mve_fp, s)) {
782         return false;
783     }
784     return do_1op(s, a, fns[a->size]);
785 }
786 
787 static bool do_2op_vec(DisasContext *s, arg_2op *a, MVEGenTwoOpFn fn,
788                        GVecGen3Fn *vecfn)
789 {
790     TCGv_ptr qd, qn, qm;
791 
792     if (!dc_isar_feature(aa32_mve, s) ||
793         !mve_check_qreg_bank(s, a->qd | a->qn | a->qm) ||
794         !fn) {
795         return false;
796     }
797     if (!mve_eci_check(s) || !vfp_access_check(s)) {
798         return true;
799     }
800 
801     if (vecfn && mve_no_predication(s)) {
802         vecfn(a->size, mve_qreg_offset(a->qd), mve_qreg_offset(a->qn),
803               mve_qreg_offset(a->qm), 16, 16);
804     } else {
805         qd = mve_qreg_ptr(a->qd);
806         qn = mve_qreg_ptr(a->qn);
807         qm = mve_qreg_ptr(a->qm);
808         fn(cpu_env, qd, qn, qm);
809         tcg_temp_free_ptr(qd);
810         tcg_temp_free_ptr(qn);
811         tcg_temp_free_ptr(qm);
812     }
813     mve_update_eci(s);
814     return true;
815 }
816 
817 static bool do_2op(DisasContext *s, arg_2op *a, MVEGenTwoOpFn *fn)
818 {
819     return do_2op_vec(s, a, fn, NULL);
820 }
821 
822 #define DO_LOGIC(INSN, HELPER, VECFN)                           \
823     static bool trans_##INSN(DisasContext *s, arg_2op *a)       \
824     {                                                           \
825         return do_2op_vec(s, a, HELPER, VECFN);                 \
826     }
827 
828 DO_LOGIC(VAND, gen_helper_mve_vand, tcg_gen_gvec_and)
829 DO_LOGIC(VBIC, gen_helper_mve_vbic, tcg_gen_gvec_andc)
830 DO_LOGIC(VORR, gen_helper_mve_vorr, tcg_gen_gvec_or)
831 DO_LOGIC(VORN, gen_helper_mve_vorn, tcg_gen_gvec_orc)
832 DO_LOGIC(VEOR, gen_helper_mve_veor, tcg_gen_gvec_xor)
833 
834 static bool trans_VPSEL(DisasContext *s, arg_2op *a)
835 {
836     /* This insn updates predication bits */
837     s->base.is_jmp = DISAS_UPDATE_NOCHAIN;
838     return do_2op(s, a, gen_helper_mve_vpsel);
839 }
840 
841 #define DO_2OP(INSN, FN) \
842     static bool trans_##INSN(DisasContext *s, arg_2op *a)       \
843     {                                                           \
844         static MVEGenTwoOpFn * const fns[] = {                  \
845             gen_helper_mve_##FN##b,                             \
846             gen_helper_mve_##FN##h,                             \
847             gen_helper_mve_##FN##w,                             \
848             NULL,                                               \
849         };                                                      \
850         return do_2op(s, a, fns[a->size]);                      \
851     }
852 
853 DO_2OP(VADD, vadd)
854 DO_2OP(VSUB, vsub)
855 DO_2OP(VMUL, vmul)
856 DO_2OP(VMULH_S, vmulhs)
857 DO_2OP(VMULH_U, vmulhu)
858 DO_2OP(VRMULH_S, vrmulhs)
859 DO_2OP(VRMULH_U, vrmulhu)
860 DO_2OP(VMAX_S, vmaxs)
861 DO_2OP(VMAX_U, vmaxu)
862 DO_2OP(VMIN_S, vmins)
863 DO_2OP(VMIN_U, vminu)
864 DO_2OP(VABD_S, vabds)
865 DO_2OP(VABD_U, vabdu)
866 DO_2OP(VHADD_S, vhadds)
867 DO_2OP(VHADD_U, vhaddu)
868 DO_2OP(VHSUB_S, vhsubs)
869 DO_2OP(VHSUB_U, vhsubu)
870 DO_2OP(VMULL_BS, vmullbs)
871 DO_2OP(VMULL_BU, vmullbu)
872 DO_2OP(VMULL_TS, vmullts)
873 DO_2OP(VMULL_TU, vmulltu)
874 DO_2OP(VQDMULH, vqdmulh)
875 DO_2OP(VQRDMULH, vqrdmulh)
876 DO_2OP(VQADD_S, vqadds)
877 DO_2OP(VQADD_U, vqaddu)
878 DO_2OP(VQSUB_S, vqsubs)
879 DO_2OP(VQSUB_U, vqsubu)
880 DO_2OP(VSHL_S, vshls)
881 DO_2OP(VSHL_U, vshlu)
882 DO_2OP(VRSHL_S, vrshls)
883 DO_2OP(VRSHL_U, vrshlu)
884 DO_2OP(VQSHL_S, vqshls)
885 DO_2OP(VQSHL_U, vqshlu)
886 DO_2OP(VQRSHL_S, vqrshls)
887 DO_2OP(VQRSHL_U, vqrshlu)
888 DO_2OP(VQDMLADH, vqdmladh)
889 DO_2OP(VQDMLADHX, vqdmladhx)
890 DO_2OP(VQRDMLADH, vqrdmladh)
891 DO_2OP(VQRDMLADHX, vqrdmladhx)
892 DO_2OP(VQDMLSDH, vqdmlsdh)
893 DO_2OP(VQDMLSDHX, vqdmlsdhx)
894 DO_2OP(VQRDMLSDH, vqrdmlsdh)
895 DO_2OP(VQRDMLSDHX, vqrdmlsdhx)
896 DO_2OP(VRHADD_S, vrhadds)
897 DO_2OP(VRHADD_U, vrhaddu)
898 /*
899  * VCADD Qd == Qm at size MO_32 is UNPREDICTABLE; we choose not to diagnose
900  * so we can reuse the DO_2OP macro. (Our implementation calculates the
901  * "expected" results in this case.) Similarly for VHCADD.
902  */
903 DO_2OP(VCADD90, vcadd90)
904 DO_2OP(VCADD270, vcadd270)
905 DO_2OP(VHCADD90, vhcadd90)
906 DO_2OP(VHCADD270, vhcadd270)
907 
908 static bool trans_VQDMULLB(DisasContext *s, arg_2op *a)
909 {
910     static MVEGenTwoOpFn * const fns[] = {
911         NULL,
912         gen_helper_mve_vqdmullbh,
913         gen_helper_mve_vqdmullbw,
914         NULL,
915     };
916     if (a->size == MO_32 && (a->qd == a->qm || a->qd == a->qn)) {
917         /* UNPREDICTABLE; we choose to undef */
918         return false;
919     }
920     return do_2op(s, a, fns[a->size]);
921 }
922 
923 static bool trans_VQDMULLT(DisasContext *s, arg_2op *a)
924 {
925     static MVEGenTwoOpFn * const fns[] = {
926         NULL,
927         gen_helper_mve_vqdmullth,
928         gen_helper_mve_vqdmulltw,
929         NULL,
930     };
931     if (a->size == MO_32 && (a->qd == a->qm || a->qd == a->qn)) {
932         /* UNPREDICTABLE; we choose to undef */
933         return false;
934     }
935     return do_2op(s, a, fns[a->size]);
936 }
937 
938 static bool trans_VMULLP_B(DisasContext *s, arg_2op *a)
939 {
940     /*
941      * Note that a->size indicates the output size, ie VMULL.P8
942      * is the 8x8->16 operation and a->size is MO_16; VMULL.P16
943      * is the 16x16->32 operation and a->size is MO_32.
944      */
945     static MVEGenTwoOpFn * const fns[] = {
946         NULL,
947         gen_helper_mve_vmullpbh,
948         gen_helper_mve_vmullpbw,
949         NULL,
950     };
951     return do_2op(s, a, fns[a->size]);
952 }
953 
954 static bool trans_VMULLP_T(DisasContext *s, arg_2op *a)
955 {
956     /* a->size is as for trans_VMULLP_B */
957     static MVEGenTwoOpFn * const fns[] = {
958         NULL,
959         gen_helper_mve_vmullpth,
960         gen_helper_mve_vmullptw,
961         NULL,
962     };
963     return do_2op(s, a, fns[a->size]);
964 }
965 
966 /*
967  * VADC and VSBC: these perform an add-with-carry or subtract-with-carry
968  * of the 32-bit elements in each lane of the input vectors, where the
969  * carry-out of each add is the carry-in of the next.  The initial carry
970  * input is either fixed (0 for VADCI, 1 for VSBCI) or is from FPSCR.C
971  * (for VADC and VSBC); the carry out at the end is written back to FPSCR.C.
972  * These insns are subject to beat-wise execution.  Partial execution
973  * of an I=1 (initial carry input fixed) insn which does not
974  * execute the first beat must start with the current FPSCR.NZCV
975  * value, not the fixed constant input.
976  */
977 static bool trans_VADC(DisasContext *s, arg_2op *a)
978 {
979     return do_2op(s, a, gen_helper_mve_vadc);
980 }
981 
982 static bool trans_VADCI(DisasContext *s, arg_2op *a)
983 {
984     if (mve_skip_first_beat(s)) {
985         return trans_VADC(s, a);
986     }
987     return do_2op(s, a, gen_helper_mve_vadci);
988 }
989 
990 static bool trans_VSBC(DisasContext *s, arg_2op *a)
991 {
992     return do_2op(s, a, gen_helper_mve_vsbc);
993 }
994 
995 static bool trans_VSBCI(DisasContext *s, arg_2op *a)
996 {
997     if (mve_skip_first_beat(s)) {
998         return trans_VSBC(s, a);
999     }
1000     return do_2op(s, a, gen_helper_mve_vsbci);
1001 }
1002 
1003 #define DO_2OP_FP(INSN, FN)                                     \
1004     static bool trans_##INSN(DisasContext *s, arg_2op *a)       \
1005     {                                                           \
1006         static MVEGenTwoOpFn * const fns[] = {                  \
1007             NULL,                                               \
1008             gen_helper_mve_##FN##h,                             \
1009             gen_helper_mve_##FN##s,                             \
1010             NULL,                                               \
1011         };                                                      \
1012         if (!dc_isar_feature(aa32_mve_fp, s)) {                 \
1013             return false;                                       \
1014         }                                                       \
1015         return do_2op(s, a, fns[a->size]);                      \
1016     }
1017 
1018 DO_2OP_FP(VADD_fp, vfadd)
1019 DO_2OP_FP(VSUB_fp, vfsub)
1020 DO_2OP_FP(VMUL_fp, vfmul)
1021 DO_2OP_FP(VABD_fp, vfabd)
1022 DO_2OP_FP(VMAXNM, vmaxnm)
1023 DO_2OP_FP(VMINNM, vminnm)
1024 DO_2OP_FP(VCADD90_fp, vfcadd90)
1025 DO_2OP_FP(VCADD270_fp, vfcadd270)
1026 DO_2OP_FP(VFMA, vfma)
1027 DO_2OP_FP(VFMS, vfms)
1028 DO_2OP_FP(VCMUL0, vcmul0)
1029 DO_2OP_FP(VCMUL90, vcmul90)
1030 DO_2OP_FP(VCMUL180, vcmul180)
1031 DO_2OP_FP(VCMUL270, vcmul270)
1032 DO_2OP_FP(VCMLA0, vcmla0)
1033 DO_2OP_FP(VCMLA90, vcmla90)
1034 DO_2OP_FP(VCMLA180, vcmla180)
1035 DO_2OP_FP(VCMLA270, vcmla270)
1036 DO_2OP_FP(VMAXNMA, vmaxnma)
1037 DO_2OP_FP(VMINNMA, vminnma)
1038 
1039 static bool do_2op_scalar(DisasContext *s, arg_2scalar *a,
1040                           MVEGenTwoOpScalarFn fn)
1041 {
1042     TCGv_ptr qd, qn;
1043     TCGv_i32 rm;
1044 
1045     if (!dc_isar_feature(aa32_mve, s) ||
1046         !mve_check_qreg_bank(s, a->qd | a->qn) ||
1047         !fn) {
1048         return false;
1049     }
1050     if (a->rm == 13 || a->rm == 15) {
1051         /* UNPREDICTABLE */
1052         return false;
1053     }
1054     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1055         return true;
1056     }
1057 
1058     qd = mve_qreg_ptr(a->qd);
1059     qn = mve_qreg_ptr(a->qn);
1060     rm = load_reg(s, a->rm);
1061     fn(cpu_env, qd, qn, rm);
1062     tcg_temp_free_i32(rm);
1063     tcg_temp_free_ptr(qd);
1064     tcg_temp_free_ptr(qn);
1065     mve_update_eci(s);
1066     return true;
1067 }
1068 
1069 #define DO_2OP_SCALAR(INSN, FN)                                 \
1070     static bool trans_##INSN(DisasContext *s, arg_2scalar *a)   \
1071     {                                                           \
1072         static MVEGenTwoOpScalarFn * const fns[] = {            \
1073             gen_helper_mve_##FN##b,                             \
1074             gen_helper_mve_##FN##h,                             \
1075             gen_helper_mve_##FN##w,                             \
1076             NULL,                                               \
1077         };                                                      \
1078         return do_2op_scalar(s, a, fns[a->size]);               \
1079     }
1080 
1081 DO_2OP_SCALAR(VADD_scalar, vadd_scalar)
1082 DO_2OP_SCALAR(VSUB_scalar, vsub_scalar)
1083 DO_2OP_SCALAR(VMUL_scalar, vmul_scalar)
1084 DO_2OP_SCALAR(VHADD_S_scalar, vhadds_scalar)
1085 DO_2OP_SCALAR(VHADD_U_scalar, vhaddu_scalar)
1086 DO_2OP_SCALAR(VHSUB_S_scalar, vhsubs_scalar)
1087 DO_2OP_SCALAR(VHSUB_U_scalar, vhsubu_scalar)
1088 DO_2OP_SCALAR(VQADD_S_scalar, vqadds_scalar)
1089 DO_2OP_SCALAR(VQADD_U_scalar, vqaddu_scalar)
1090 DO_2OP_SCALAR(VQSUB_S_scalar, vqsubs_scalar)
1091 DO_2OP_SCALAR(VQSUB_U_scalar, vqsubu_scalar)
1092 DO_2OP_SCALAR(VQDMULH_scalar, vqdmulh_scalar)
1093 DO_2OP_SCALAR(VQRDMULH_scalar, vqrdmulh_scalar)
1094 DO_2OP_SCALAR(VBRSR, vbrsr)
1095 DO_2OP_SCALAR(VMLA, vmla)
1096 DO_2OP_SCALAR(VMLAS, vmlas)
1097 DO_2OP_SCALAR(VQDMLAH, vqdmlah)
1098 DO_2OP_SCALAR(VQRDMLAH, vqrdmlah)
1099 DO_2OP_SCALAR(VQDMLASH, vqdmlash)
1100 DO_2OP_SCALAR(VQRDMLASH, vqrdmlash)
1101 
1102 static bool trans_VQDMULLB_scalar(DisasContext *s, arg_2scalar *a)
1103 {
1104     static MVEGenTwoOpScalarFn * const fns[] = {
1105         NULL,
1106         gen_helper_mve_vqdmullb_scalarh,
1107         gen_helper_mve_vqdmullb_scalarw,
1108         NULL,
1109     };
1110     if (a->qd == a->qn && a->size == MO_32) {
1111         /* UNPREDICTABLE; we choose to undef */
1112         return false;
1113     }
1114     return do_2op_scalar(s, a, fns[a->size]);
1115 }
1116 
1117 static bool trans_VQDMULLT_scalar(DisasContext *s, arg_2scalar *a)
1118 {
1119     static MVEGenTwoOpScalarFn * const fns[] = {
1120         NULL,
1121         gen_helper_mve_vqdmullt_scalarh,
1122         gen_helper_mve_vqdmullt_scalarw,
1123         NULL,
1124     };
1125     if (a->qd == a->qn && a->size == MO_32) {
1126         /* UNPREDICTABLE; we choose to undef */
1127         return false;
1128     }
1129     return do_2op_scalar(s, a, fns[a->size]);
1130 }
1131 
1132 
1133 #define DO_2OP_FP_SCALAR(INSN, FN)                              \
1134     static bool trans_##INSN(DisasContext *s, arg_2scalar *a)   \
1135     {                                                           \
1136         static MVEGenTwoOpScalarFn * const fns[] = {            \
1137             NULL,                                               \
1138             gen_helper_mve_##FN##h,                             \
1139             gen_helper_mve_##FN##s,                             \
1140             NULL,                                               \
1141         };                                                      \
1142         if (!dc_isar_feature(aa32_mve_fp, s)) {                 \
1143             return false;                                       \
1144         }                                                       \
1145         return do_2op_scalar(s, a, fns[a->size]);               \
1146     }
1147 
1148 DO_2OP_FP_SCALAR(VADD_fp_scalar, vfadd_scalar)
1149 DO_2OP_FP_SCALAR(VSUB_fp_scalar, vfsub_scalar)
1150 DO_2OP_FP_SCALAR(VMUL_fp_scalar, vfmul_scalar)
1151 DO_2OP_FP_SCALAR(VFMA_scalar, vfma_scalar)
1152 DO_2OP_FP_SCALAR(VFMAS_scalar, vfmas_scalar)
1153 
1154 static bool do_long_dual_acc(DisasContext *s, arg_vmlaldav *a,
1155                              MVEGenLongDualAccOpFn *fn)
1156 {
1157     TCGv_ptr qn, qm;
1158     TCGv_i64 rda;
1159     TCGv_i32 rdalo, rdahi;
1160 
1161     if (!dc_isar_feature(aa32_mve, s) ||
1162         !mve_check_qreg_bank(s, a->qn | a->qm) ||
1163         !fn) {
1164         return false;
1165     }
1166     /*
1167      * rdahi == 13 is UNPREDICTABLE; rdahi == 15 is a related
1168      * encoding; rdalo always has bit 0 clear so cannot be 13 or 15.
1169      */
1170     if (a->rdahi == 13 || a->rdahi == 15) {
1171         return false;
1172     }
1173     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1174         return true;
1175     }
1176 
1177     qn = mve_qreg_ptr(a->qn);
1178     qm = mve_qreg_ptr(a->qm);
1179 
1180     /*
1181      * This insn is subject to beat-wise execution. Partial execution
1182      * of an A=0 (no-accumulate) insn which does not execute the first
1183      * beat must start with the current rda value, not 0.
1184      */
1185     if (a->a || mve_skip_first_beat(s)) {
1186         rda = tcg_temp_new_i64();
1187         rdalo = load_reg(s, a->rdalo);
1188         rdahi = load_reg(s, a->rdahi);
1189         tcg_gen_concat_i32_i64(rda, rdalo, rdahi);
1190         tcg_temp_free_i32(rdalo);
1191         tcg_temp_free_i32(rdahi);
1192     } else {
1193         rda = tcg_const_i64(0);
1194     }
1195 
1196     fn(rda, cpu_env, qn, qm, rda);
1197     tcg_temp_free_ptr(qn);
1198     tcg_temp_free_ptr(qm);
1199 
1200     rdalo = tcg_temp_new_i32();
1201     rdahi = tcg_temp_new_i32();
1202     tcg_gen_extrl_i64_i32(rdalo, rda);
1203     tcg_gen_extrh_i64_i32(rdahi, rda);
1204     store_reg(s, a->rdalo, rdalo);
1205     store_reg(s, a->rdahi, rdahi);
1206     tcg_temp_free_i64(rda);
1207     mve_update_eci(s);
1208     return true;
1209 }
1210 
1211 static bool trans_VMLALDAV_S(DisasContext *s, arg_vmlaldav *a)
1212 {
1213     static MVEGenLongDualAccOpFn * const fns[4][2] = {
1214         { NULL, NULL },
1215         { gen_helper_mve_vmlaldavsh, gen_helper_mve_vmlaldavxsh },
1216         { gen_helper_mve_vmlaldavsw, gen_helper_mve_vmlaldavxsw },
1217         { NULL, NULL },
1218     };
1219     return do_long_dual_acc(s, a, fns[a->size][a->x]);
1220 }
1221 
1222 static bool trans_VMLALDAV_U(DisasContext *s, arg_vmlaldav *a)
1223 {
1224     static MVEGenLongDualAccOpFn * const fns[4][2] = {
1225         { NULL, NULL },
1226         { gen_helper_mve_vmlaldavuh, NULL },
1227         { gen_helper_mve_vmlaldavuw, NULL },
1228         { NULL, NULL },
1229     };
1230     return do_long_dual_acc(s, a, fns[a->size][a->x]);
1231 }
1232 
1233 static bool trans_VMLSLDAV(DisasContext *s, arg_vmlaldav *a)
1234 {
1235     static MVEGenLongDualAccOpFn * const fns[4][2] = {
1236         { NULL, NULL },
1237         { gen_helper_mve_vmlsldavsh, gen_helper_mve_vmlsldavxsh },
1238         { gen_helper_mve_vmlsldavsw, gen_helper_mve_vmlsldavxsw },
1239         { NULL, NULL },
1240     };
1241     return do_long_dual_acc(s, a, fns[a->size][a->x]);
1242 }
1243 
1244 static bool trans_VRMLALDAVH_S(DisasContext *s, arg_vmlaldav *a)
1245 {
1246     static MVEGenLongDualAccOpFn * const fns[] = {
1247         gen_helper_mve_vrmlaldavhsw, gen_helper_mve_vrmlaldavhxsw,
1248     };
1249     return do_long_dual_acc(s, a, fns[a->x]);
1250 }
1251 
1252 static bool trans_VRMLALDAVH_U(DisasContext *s, arg_vmlaldav *a)
1253 {
1254     static MVEGenLongDualAccOpFn * const fns[] = {
1255         gen_helper_mve_vrmlaldavhuw, NULL,
1256     };
1257     return do_long_dual_acc(s, a, fns[a->x]);
1258 }
1259 
1260 static bool trans_VRMLSLDAVH(DisasContext *s, arg_vmlaldav *a)
1261 {
1262     static MVEGenLongDualAccOpFn * const fns[] = {
1263         gen_helper_mve_vrmlsldavhsw, gen_helper_mve_vrmlsldavhxsw,
1264     };
1265     return do_long_dual_acc(s, a, fns[a->x]);
1266 }
1267 
1268 static bool do_dual_acc(DisasContext *s, arg_vmladav *a, MVEGenDualAccOpFn *fn)
1269 {
1270     TCGv_ptr qn, qm;
1271     TCGv_i32 rda;
1272 
1273     if (!dc_isar_feature(aa32_mve, s) ||
1274         !mve_check_qreg_bank(s, a->qn) ||
1275         !fn) {
1276         return false;
1277     }
1278     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1279         return true;
1280     }
1281 
1282     qn = mve_qreg_ptr(a->qn);
1283     qm = mve_qreg_ptr(a->qm);
1284 
1285     /*
1286      * This insn is subject to beat-wise execution. Partial execution
1287      * of an A=0 (no-accumulate) insn which does not execute the first
1288      * beat must start with the current rda value, not 0.
1289      */
1290     if (a->a || mve_skip_first_beat(s)) {
1291         rda = load_reg(s, a->rda);
1292     } else {
1293         rda = tcg_const_i32(0);
1294     }
1295 
1296     fn(rda, cpu_env, qn, qm, rda);
1297     store_reg(s, a->rda, rda);
1298     tcg_temp_free_ptr(qn);
1299     tcg_temp_free_ptr(qm);
1300 
1301     mve_update_eci(s);
1302     return true;
1303 }
1304 
1305 #define DO_DUAL_ACC(INSN, FN)                                           \
1306     static bool trans_##INSN(DisasContext *s, arg_vmladav *a)           \
1307     {                                                                   \
1308         static MVEGenDualAccOpFn * const fns[4][2] = {                  \
1309             { gen_helper_mve_##FN##b, gen_helper_mve_##FN##xb },        \
1310             { gen_helper_mve_##FN##h, gen_helper_mve_##FN##xh },        \
1311             { gen_helper_mve_##FN##w, gen_helper_mve_##FN##xw },        \
1312             { NULL, NULL },                                             \
1313         };                                                              \
1314         return do_dual_acc(s, a, fns[a->size][a->x]);                   \
1315     }
1316 
1317 DO_DUAL_ACC(VMLADAV_S, vmladavs)
1318 DO_DUAL_ACC(VMLSDAV, vmlsdav)
1319 
1320 static bool trans_VMLADAV_U(DisasContext *s, arg_vmladav *a)
1321 {
1322     static MVEGenDualAccOpFn * const fns[4][2] = {
1323         { gen_helper_mve_vmladavub, NULL },
1324         { gen_helper_mve_vmladavuh, NULL },
1325         { gen_helper_mve_vmladavuw, NULL },
1326         { NULL, NULL },
1327     };
1328     return do_dual_acc(s, a, fns[a->size][a->x]);
1329 }
1330 
1331 static void gen_vpst(DisasContext *s, uint32_t mask)
1332 {
1333     /*
1334      * Set the VPR mask fields. We take advantage of MASK01 and MASK23
1335      * being adjacent fields in the register.
1336      *
1337      * Updating the masks is not predicated, but it is subject to beat-wise
1338      * execution, and the mask is updated on the odd-numbered beats.
1339      * So if PSR.ECI says we should skip beat 1, we mustn't update the
1340      * 01 mask field.
1341      */
1342     TCGv_i32 vpr = load_cpu_field(v7m.vpr);
1343     switch (s->eci) {
1344     case ECI_NONE:
1345     case ECI_A0:
1346         /* Update both 01 and 23 fields */
1347         tcg_gen_deposit_i32(vpr, vpr,
1348                             tcg_constant_i32(mask | (mask << 4)),
1349                             R_V7M_VPR_MASK01_SHIFT,
1350                             R_V7M_VPR_MASK01_LENGTH + R_V7M_VPR_MASK23_LENGTH);
1351         break;
1352     case ECI_A0A1:
1353     case ECI_A0A1A2:
1354     case ECI_A0A1A2B0:
1355         /* Update only the 23 mask field */
1356         tcg_gen_deposit_i32(vpr, vpr,
1357                             tcg_constant_i32(mask),
1358                             R_V7M_VPR_MASK23_SHIFT, R_V7M_VPR_MASK23_LENGTH);
1359         break;
1360     default:
1361         g_assert_not_reached();
1362     }
1363     store_cpu_field(vpr, v7m.vpr);
1364 }
1365 
1366 static bool trans_VPST(DisasContext *s, arg_VPST *a)
1367 {
1368     /* mask == 0 is a "related encoding" */
1369     if (!dc_isar_feature(aa32_mve, s) || !a->mask) {
1370         return false;
1371     }
1372     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1373         return true;
1374     }
1375     gen_vpst(s, a->mask);
1376     mve_update_and_store_eci(s);
1377     return true;
1378 }
1379 
1380 static bool trans_VPNOT(DisasContext *s, arg_VPNOT *a)
1381 {
1382     /*
1383      * Invert the predicate in VPR.P0. We have call out to
1384      * a helper because this insn itself is beatwise and can
1385      * be predicated.
1386      */
1387     if (!dc_isar_feature(aa32_mve, s)) {
1388         return false;
1389     }
1390     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1391         return true;
1392     }
1393 
1394     gen_helper_mve_vpnot(cpu_env);
1395     /* This insn updates predication bits */
1396     s->base.is_jmp = DISAS_UPDATE_NOCHAIN;
1397     mve_update_eci(s);
1398     return true;
1399 }
1400 
1401 static bool trans_VADDV(DisasContext *s, arg_VADDV *a)
1402 {
1403     /* VADDV: vector add across vector */
1404     static MVEGenVADDVFn * const fns[4][2] = {
1405         { gen_helper_mve_vaddvsb, gen_helper_mve_vaddvub },
1406         { gen_helper_mve_vaddvsh, gen_helper_mve_vaddvuh },
1407         { gen_helper_mve_vaddvsw, gen_helper_mve_vaddvuw },
1408         { NULL, NULL }
1409     };
1410     TCGv_ptr qm;
1411     TCGv_i32 rda;
1412 
1413     if (!dc_isar_feature(aa32_mve, s) ||
1414         a->size == 3) {
1415         return false;
1416     }
1417     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1418         return true;
1419     }
1420 
1421     /*
1422      * This insn is subject to beat-wise execution. Partial execution
1423      * of an A=0 (no-accumulate) insn which does not execute the first
1424      * beat must start with the current value of Rda, not zero.
1425      */
1426     if (a->a || mve_skip_first_beat(s)) {
1427         /* Accumulate input from Rda */
1428         rda = load_reg(s, a->rda);
1429     } else {
1430         /* Accumulate starting at zero */
1431         rda = tcg_const_i32(0);
1432     }
1433 
1434     qm = mve_qreg_ptr(a->qm);
1435     fns[a->size][a->u](rda, cpu_env, qm, rda);
1436     store_reg(s, a->rda, rda);
1437     tcg_temp_free_ptr(qm);
1438 
1439     mve_update_eci(s);
1440     return true;
1441 }
1442 
1443 static bool trans_VADDLV(DisasContext *s, arg_VADDLV *a)
1444 {
1445     /*
1446      * Vector Add Long Across Vector: accumulate the 32-bit
1447      * elements of the vector into a 64-bit result stored in
1448      * a pair of general-purpose registers.
1449      * No need to check Qm's bank: it is only 3 bits in decode.
1450      */
1451     TCGv_ptr qm;
1452     TCGv_i64 rda;
1453     TCGv_i32 rdalo, rdahi;
1454 
1455     if (!dc_isar_feature(aa32_mve, s)) {
1456         return false;
1457     }
1458     /*
1459      * rdahi == 13 is UNPREDICTABLE; rdahi == 15 is a related
1460      * encoding; rdalo always has bit 0 clear so cannot be 13 or 15.
1461      */
1462     if (a->rdahi == 13 || a->rdahi == 15) {
1463         return false;
1464     }
1465     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1466         return true;
1467     }
1468 
1469     /*
1470      * This insn is subject to beat-wise execution. Partial execution
1471      * of an A=0 (no-accumulate) insn which does not execute the first
1472      * beat must start with the current value of RdaHi:RdaLo, not zero.
1473      */
1474     if (a->a || mve_skip_first_beat(s)) {
1475         /* Accumulate input from RdaHi:RdaLo */
1476         rda = tcg_temp_new_i64();
1477         rdalo = load_reg(s, a->rdalo);
1478         rdahi = load_reg(s, a->rdahi);
1479         tcg_gen_concat_i32_i64(rda, rdalo, rdahi);
1480         tcg_temp_free_i32(rdalo);
1481         tcg_temp_free_i32(rdahi);
1482     } else {
1483         /* Accumulate starting at zero */
1484         rda = tcg_const_i64(0);
1485     }
1486 
1487     qm = mve_qreg_ptr(a->qm);
1488     if (a->u) {
1489         gen_helper_mve_vaddlv_u(rda, cpu_env, qm, rda);
1490     } else {
1491         gen_helper_mve_vaddlv_s(rda, cpu_env, qm, rda);
1492     }
1493     tcg_temp_free_ptr(qm);
1494 
1495     rdalo = tcg_temp_new_i32();
1496     rdahi = tcg_temp_new_i32();
1497     tcg_gen_extrl_i64_i32(rdalo, rda);
1498     tcg_gen_extrh_i64_i32(rdahi, rda);
1499     store_reg(s, a->rdalo, rdalo);
1500     store_reg(s, a->rdahi, rdahi);
1501     tcg_temp_free_i64(rda);
1502     mve_update_eci(s);
1503     return true;
1504 }
1505 
1506 static bool do_1imm(DisasContext *s, arg_1imm *a, MVEGenOneOpImmFn *fn)
1507 {
1508     TCGv_ptr qd;
1509     uint64_t imm;
1510 
1511     if (!dc_isar_feature(aa32_mve, s) ||
1512         !mve_check_qreg_bank(s, a->qd) ||
1513         !fn) {
1514         return false;
1515     }
1516     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1517         return true;
1518     }
1519 
1520     imm = asimd_imm_const(a->imm, a->cmode, a->op);
1521 
1522     qd = mve_qreg_ptr(a->qd);
1523     fn(cpu_env, qd, tcg_constant_i64(imm));
1524     tcg_temp_free_ptr(qd);
1525     mve_update_eci(s);
1526     return true;
1527 }
1528 
1529 static bool trans_Vimm_1r(DisasContext *s, arg_1imm *a)
1530 {
1531     /* Handle decode of cmode/op here between VORR/VBIC/VMOV */
1532     MVEGenOneOpImmFn *fn;
1533 
1534     if ((a->cmode & 1) && a->cmode < 12) {
1535         if (a->op) {
1536             /*
1537              * For op=1, the immediate will be inverted by asimd_imm_const(),
1538              * so the VBIC becomes a logical AND operation.
1539              */
1540             fn = gen_helper_mve_vandi;
1541         } else {
1542             fn = gen_helper_mve_vorri;
1543         }
1544     } else {
1545         /* There is one unallocated cmode/op combination in this space */
1546         if (a->cmode == 15 && a->op == 1) {
1547             return false;
1548         }
1549         /* asimd_imm_const() sorts out VMVNI vs VMOVI for us */
1550         fn = gen_helper_mve_vmovi;
1551     }
1552     return do_1imm(s, a, fn);
1553 }
1554 
1555 static bool do_2shift(DisasContext *s, arg_2shift *a, MVEGenTwoOpShiftFn fn,
1556                       bool negateshift)
1557 {
1558     TCGv_ptr qd, qm;
1559     int shift = a->shift;
1560 
1561     if (!dc_isar_feature(aa32_mve, s) ||
1562         !mve_check_qreg_bank(s, a->qd | a->qm) ||
1563         !fn) {
1564         return false;
1565     }
1566     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1567         return true;
1568     }
1569 
1570     /*
1571      * When we handle a right shift insn using a left-shift helper
1572      * which permits a negative shift count to indicate a right-shift,
1573      * we must negate the shift count.
1574      */
1575     if (negateshift) {
1576         shift = -shift;
1577     }
1578 
1579     qd = mve_qreg_ptr(a->qd);
1580     qm = mve_qreg_ptr(a->qm);
1581     fn(cpu_env, qd, qm, tcg_constant_i32(shift));
1582     tcg_temp_free_ptr(qd);
1583     tcg_temp_free_ptr(qm);
1584     mve_update_eci(s);
1585     return true;
1586 }
1587 
1588 #define DO_2SHIFT(INSN, FN, NEGATESHIFT)                         \
1589     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1590     {                                                           \
1591         static MVEGenTwoOpShiftFn * const fns[] = {             \
1592             gen_helper_mve_##FN##b,                             \
1593             gen_helper_mve_##FN##h,                             \
1594             gen_helper_mve_##FN##w,                             \
1595             NULL,                                               \
1596         };                                                      \
1597         return do_2shift(s, a, fns[a->size], NEGATESHIFT);      \
1598     }
1599 
1600 DO_2SHIFT(VSHLI, vshli_u, false)
1601 DO_2SHIFT(VQSHLI_S, vqshli_s, false)
1602 DO_2SHIFT(VQSHLI_U, vqshli_u, false)
1603 DO_2SHIFT(VQSHLUI, vqshlui_s, false)
1604 /* These right shifts use a left-shift helper with negated shift count */
1605 DO_2SHIFT(VSHRI_S, vshli_s, true)
1606 DO_2SHIFT(VSHRI_U, vshli_u, true)
1607 DO_2SHIFT(VRSHRI_S, vrshli_s, true)
1608 DO_2SHIFT(VRSHRI_U, vrshli_u, true)
1609 
1610 DO_2SHIFT(VSRI, vsri, false)
1611 DO_2SHIFT(VSLI, vsli, false)
1612 
1613 #define DO_2SHIFT_FP(INSN, FN)                                  \
1614     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1615     {                                                           \
1616         if (!dc_isar_feature(aa32_mve_fp, s)) {                 \
1617             return false;                                       \
1618         }                                                       \
1619         return do_2shift(s, a, gen_helper_mve_##FN, false);     \
1620     }
1621 
1622 DO_2SHIFT_FP(VCVT_SH_fixed, vcvt_sh)
1623 DO_2SHIFT_FP(VCVT_UH_fixed, vcvt_uh)
1624 DO_2SHIFT_FP(VCVT_HS_fixed, vcvt_hs)
1625 DO_2SHIFT_FP(VCVT_HU_fixed, vcvt_hu)
1626 DO_2SHIFT_FP(VCVT_SF_fixed, vcvt_sf)
1627 DO_2SHIFT_FP(VCVT_UF_fixed, vcvt_uf)
1628 DO_2SHIFT_FP(VCVT_FS_fixed, vcvt_fs)
1629 DO_2SHIFT_FP(VCVT_FU_fixed, vcvt_fu)
1630 
1631 static bool do_2shift_scalar(DisasContext *s, arg_shl_scalar *a,
1632                              MVEGenTwoOpShiftFn *fn)
1633 {
1634     TCGv_ptr qda;
1635     TCGv_i32 rm;
1636 
1637     if (!dc_isar_feature(aa32_mve, s) ||
1638         !mve_check_qreg_bank(s, a->qda) ||
1639         a->rm == 13 || a->rm == 15 || !fn) {
1640         /* Rm cases are UNPREDICTABLE */
1641         return false;
1642     }
1643     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1644         return true;
1645     }
1646 
1647     qda = mve_qreg_ptr(a->qda);
1648     rm = load_reg(s, a->rm);
1649     fn(cpu_env, qda, qda, rm);
1650     tcg_temp_free_ptr(qda);
1651     tcg_temp_free_i32(rm);
1652     mve_update_eci(s);
1653     return true;
1654 }
1655 
1656 #define DO_2SHIFT_SCALAR(INSN, FN)                                      \
1657     static bool trans_##INSN(DisasContext *s, arg_shl_scalar *a)        \
1658     {                                                                   \
1659         static MVEGenTwoOpShiftFn * const fns[] = {                     \
1660             gen_helper_mve_##FN##b,                                     \
1661             gen_helper_mve_##FN##h,                                     \
1662             gen_helper_mve_##FN##w,                                     \
1663             NULL,                                                       \
1664         };                                                              \
1665         return do_2shift_scalar(s, a, fns[a->size]);                    \
1666     }
1667 
1668 DO_2SHIFT_SCALAR(VSHL_S_scalar, vshli_s)
1669 DO_2SHIFT_SCALAR(VSHL_U_scalar, vshli_u)
1670 DO_2SHIFT_SCALAR(VRSHL_S_scalar, vrshli_s)
1671 DO_2SHIFT_SCALAR(VRSHL_U_scalar, vrshli_u)
1672 DO_2SHIFT_SCALAR(VQSHL_S_scalar, vqshli_s)
1673 DO_2SHIFT_SCALAR(VQSHL_U_scalar, vqshli_u)
1674 DO_2SHIFT_SCALAR(VQRSHL_S_scalar, vqrshli_s)
1675 DO_2SHIFT_SCALAR(VQRSHL_U_scalar, vqrshli_u)
1676 
1677 #define DO_VSHLL(INSN, FN)                                      \
1678     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1679     {                                                           \
1680         static MVEGenTwoOpShiftFn * const fns[] = {             \
1681             gen_helper_mve_##FN##b,                             \
1682             gen_helper_mve_##FN##h,                             \
1683         };                                                      \
1684         return do_2shift(s, a, fns[a->size], false);            \
1685     }
1686 
1687 DO_VSHLL(VSHLL_BS, vshllbs)
1688 DO_VSHLL(VSHLL_BU, vshllbu)
1689 DO_VSHLL(VSHLL_TS, vshllts)
1690 DO_VSHLL(VSHLL_TU, vshlltu)
1691 
1692 #define DO_2SHIFT_N(INSN, FN)                                   \
1693     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1694     {                                                           \
1695         static MVEGenTwoOpShiftFn * const fns[] = {             \
1696             gen_helper_mve_##FN##b,                             \
1697             gen_helper_mve_##FN##h,                             \
1698         };                                                      \
1699         return do_2shift(s, a, fns[a->size], false);            \
1700     }
1701 
1702 DO_2SHIFT_N(VSHRNB, vshrnb)
1703 DO_2SHIFT_N(VSHRNT, vshrnt)
1704 DO_2SHIFT_N(VRSHRNB, vrshrnb)
1705 DO_2SHIFT_N(VRSHRNT, vrshrnt)
1706 DO_2SHIFT_N(VQSHRNB_S, vqshrnb_s)
1707 DO_2SHIFT_N(VQSHRNT_S, vqshrnt_s)
1708 DO_2SHIFT_N(VQSHRNB_U, vqshrnb_u)
1709 DO_2SHIFT_N(VQSHRNT_U, vqshrnt_u)
1710 DO_2SHIFT_N(VQSHRUNB, vqshrunb)
1711 DO_2SHIFT_N(VQSHRUNT, vqshrunt)
1712 DO_2SHIFT_N(VQRSHRNB_S, vqrshrnb_s)
1713 DO_2SHIFT_N(VQRSHRNT_S, vqrshrnt_s)
1714 DO_2SHIFT_N(VQRSHRNB_U, vqrshrnb_u)
1715 DO_2SHIFT_N(VQRSHRNT_U, vqrshrnt_u)
1716 DO_2SHIFT_N(VQRSHRUNB, vqrshrunb)
1717 DO_2SHIFT_N(VQRSHRUNT, vqrshrunt)
1718 
1719 static bool trans_VSHLC(DisasContext *s, arg_VSHLC *a)
1720 {
1721     /*
1722      * Whole Vector Left Shift with Carry. The carry is taken
1723      * from a general purpose register and written back there.
1724      * An imm of 0 means "shift by 32".
1725      */
1726     TCGv_ptr qd;
1727     TCGv_i32 rdm;
1728 
1729     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) {
1730         return false;
1731     }
1732     if (a->rdm == 13 || a->rdm == 15) {
1733         /* CONSTRAINED UNPREDICTABLE: we UNDEF */
1734         return false;
1735     }
1736     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1737         return true;
1738     }
1739 
1740     qd = mve_qreg_ptr(a->qd);
1741     rdm = load_reg(s, a->rdm);
1742     gen_helper_mve_vshlc(rdm, cpu_env, qd, rdm, tcg_constant_i32(a->imm));
1743     store_reg(s, a->rdm, rdm);
1744     tcg_temp_free_ptr(qd);
1745     mve_update_eci(s);
1746     return true;
1747 }
1748 
1749 static bool do_vidup(DisasContext *s, arg_vidup *a, MVEGenVIDUPFn *fn)
1750 {
1751     TCGv_ptr qd;
1752     TCGv_i32 rn;
1753 
1754     /*
1755      * Vector increment/decrement with wrap and duplicate (VIDUP, VDDUP).
1756      * This fills the vector with elements of successively increasing
1757      * or decreasing values, starting from Rn.
1758      */
1759     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) {
1760         return false;
1761     }
1762     if (a->size == MO_64) {
1763         /* size 0b11 is another encoding */
1764         return false;
1765     }
1766     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1767         return true;
1768     }
1769 
1770     qd = mve_qreg_ptr(a->qd);
1771     rn = load_reg(s, a->rn);
1772     fn(rn, cpu_env, qd, rn, tcg_constant_i32(a->imm));
1773     store_reg(s, a->rn, rn);
1774     tcg_temp_free_ptr(qd);
1775     mve_update_eci(s);
1776     return true;
1777 }
1778 
1779 static bool do_viwdup(DisasContext *s, arg_viwdup *a, MVEGenVIWDUPFn *fn)
1780 {
1781     TCGv_ptr qd;
1782     TCGv_i32 rn, rm;
1783 
1784     /*
1785      * Vector increment/decrement with wrap and duplicate (VIWDUp, VDWDUP)
1786      * This fills the vector with elements of successively increasing
1787      * or decreasing values, starting from Rn. Rm specifies a point where
1788      * the count wraps back around to 0. The updated offset is written back
1789      * to Rn.
1790      */
1791     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) {
1792         return false;
1793     }
1794     if (!fn || a->rm == 13 || a->rm == 15) {
1795         /*
1796          * size 0b11 is another encoding; Rm == 13 is UNPREDICTABLE;
1797          * Rm == 13 is VIWDUP, VDWDUP.
1798          */
1799         return false;
1800     }
1801     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1802         return true;
1803     }
1804 
1805     qd = mve_qreg_ptr(a->qd);
1806     rn = load_reg(s, a->rn);
1807     rm = load_reg(s, a->rm);
1808     fn(rn, cpu_env, qd, rn, rm, tcg_constant_i32(a->imm));
1809     store_reg(s, a->rn, rn);
1810     tcg_temp_free_ptr(qd);
1811     tcg_temp_free_i32(rm);
1812     mve_update_eci(s);
1813     return true;
1814 }
1815 
1816 static bool trans_VIDUP(DisasContext *s, arg_vidup *a)
1817 {
1818     static MVEGenVIDUPFn * const fns[] = {
1819         gen_helper_mve_vidupb,
1820         gen_helper_mve_viduph,
1821         gen_helper_mve_vidupw,
1822         NULL,
1823     };
1824     return do_vidup(s, a, fns[a->size]);
1825 }
1826 
1827 static bool trans_VDDUP(DisasContext *s, arg_vidup *a)
1828 {
1829     static MVEGenVIDUPFn * const fns[] = {
1830         gen_helper_mve_vidupb,
1831         gen_helper_mve_viduph,
1832         gen_helper_mve_vidupw,
1833         NULL,
1834     };
1835     /* VDDUP is just like VIDUP but with a negative immediate */
1836     a->imm = -a->imm;
1837     return do_vidup(s, a, fns[a->size]);
1838 }
1839 
1840 static bool trans_VIWDUP(DisasContext *s, arg_viwdup *a)
1841 {
1842     static MVEGenVIWDUPFn * const fns[] = {
1843         gen_helper_mve_viwdupb,
1844         gen_helper_mve_viwduph,
1845         gen_helper_mve_viwdupw,
1846         NULL,
1847     };
1848     return do_viwdup(s, a, fns[a->size]);
1849 }
1850 
1851 static bool trans_VDWDUP(DisasContext *s, arg_viwdup *a)
1852 {
1853     static MVEGenVIWDUPFn * const fns[] = {
1854         gen_helper_mve_vdwdupb,
1855         gen_helper_mve_vdwduph,
1856         gen_helper_mve_vdwdupw,
1857         NULL,
1858     };
1859     return do_viwdup(s, a, fns[a->size]);
1860 }
1861 
1862 static bool do_vcmp(DisasContext *s, arg_vcmp *a, MVEGenCmpFn *fn)
1863 {
1864     TCGv_ptr qn, qm;
1865 
1866     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qm) ||
1867         !fn) {
1868         return false;
1869     }
1870     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1871         return true;
1872     }
1873 
1874     qn = mve_qreg_ptr(a->qn);
1875     qm = mve_qreg_ptr(a->qm);
1876     fn(cpu_env, qn, qm);
1877     tcg_temp_free_ptr(qn);
1878     tcg_temp_free_ptr(qm);
1879     if (a->mask) {
1880         /* VPT */
1881         gen_vpst(s, a->mask);
1882     }
1883     /* This insn updates predication bits */
1884     s->base.is_jmp = DISAS_UPDATE_NOCHAIN;
1885     mve_update_eci(s);
1886     return true;
1887 }
1888 
1889 static bool do_vcmp_scalar(DisasContext *s, arg_vcmp_scalar *a,
1890                            MVEGenScalarCmpFn *fn)
1891 {
1892     TCGv_ptr qn;
1893     TCGv_i32 rm;
1894 
1895     if (!dc_isar_feature(aa32_mve, s) || !fn || a->rm == 13) {
1896         return false;
1897     }
1898     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1899         return true;
1900     }
1901 
1902     qn = mve_qreg_ptr(a->qn);
1903     if (a->rm == 15) {
1904         /* Encoding Rm=0b1111 means "constant zero" */
1905         rm = tcg_constant_i32(0);
1906     } else {
1907         rm = load_reg(s, a->rm);
1908     }
1909     fn(cpu_env, qn, rm);
1910     tcg_temp_free_ptr(qn);
1911     tcg_temp_free_i32(rm);
1912     if (a->mask) {
1913         /* VPT */
1914         gen_vpst(s, a->mask);
1915     }
1916     /* This insn updates predication bits */
1917     s->base.is_jmp = DISAS_UPDATE_NOCHAIN;
1918     mve_update_eci(s);
1919     return true;
1920 }
1921 
1922 #define DO_VCMP(INSN, FN)                                       \
1923     static bool trans_##INSN(DisasContext *s, arg_vcmp *a)      \
1924     {                                                           \
1925         static MVEGenCmpFn * const fns[] = {                    \
1926             gen_helper_mve_##FN##b,                             \
1927             gen_helper_mve_##FN##h,                             \
1928             gen_helper_mve_##FN##w,                             \
1929             NULL,                                               \
1930         };                                                      \
1931         return do_vcmp(s, a, fns[a->size]);                     \
1932     }                                                           \
1933     static bool trans_##INSN##_scalar(DisasContext *s,          \
1934                                       arg_vcmp_scalar *a)       \
1935     {                                                           \
1936         static MVEGenScalarCmpFn * const fns[] = {              \
1937             gen_helper_mve_##FN##_scalarb,                      \
1938             gen_helper_mve_##FN##_scalarh,                      \
1939             gen_helper_mve_##FN##_scalarw,                      \
1940             NULL,                                               \
1941         };                                                      \
1942         return do_vcmp_scalar(s, a, fns[a->size]);              \
1943     }
1944 
1945 DO_VCMP(VCMPEQ, vcmpeq)
1946 DO_VCMP(VCMPNE, vcmpne)
1947 DO_VCMP(VCMPCS, vcmpcs)
1948 DO_VCMP(VCMPHI, vcmphi)
1949 DO_VCMP(VCMPGE, vcmpge)
1950 DO_VCMP(VCMPLT, vcmplt)
1951 DO_VCMP(VCMPGT, vcmpgt)
1952 DO_VCMP(VCMPLE, vcmple)
1953 
1954 #define DO_VCMP_FP(INSN, FN)                                    \
1955     static bool trans_##INSN(DisasContext *s, arg_vcmp *a)      \
1956     {                                                           \
1957         static MVEGenCmpFn * const fns[] = {                    \
1958             NULL,                                               \
1959             gen_helper_mve_##FN##h,                             \
1960             gen_helper_mve_##FN##s,                             \
1961             NULL,                                               \
1962         };                                                      \
1963         if (!dc_isar_feature(aa32_mve_fp, s)) {                 \
1964             return false;                                       \
1965         }                                                       \
1966         return do_vcmp(s, a, fns[a->size]);                     \
1967     }                                                           \
1968     static bool trans_##INSN##_scalar(DisasContext *s,          \
1969                                       arg_vcmp_scalar *a)       \
1970     {                                                           \
1971         static MVEGenScalarCmpFn * const fns[] = {              \
1972             NULL,                                               \
1973             gen_helper_mve_##FN##_scalarh,                      \
1974             gen_helper_mve_##FN##_scalars,                      \
1975             NULL,                                               \
1976         };                                                      \
1977         if (!dc_isar_feature(aa32_mve_fp, s)) {                 \
1978             return false;                                       \
1979         }                                                       \
1980         return do_vcmp_scalar(s, a, fns[a->size]);              \
1981     }
1982 
1983 DO_VCMP_FP(VCMPEQ_fp, vfcmpeq)
1984 DO_VCMP_FP(VCMPNE_fp, vfcmpne)
1985 DO_VCMP_FP(VCMPGE_fp, vfcmpge)
1986 DO_VCMP_FP(VCMPLT_fp, vfcmplt)
1987 DO_VCMP_FP(VCMPGT_fp, vfcmpgt)
1988 DO_VCMP_FP(VCMPLE_fp, vfcmple)
1989 
1990 static bool do_vmaxv(DisasContext *s, arg_vmaxv *a, MVEGenVADDVFn fn)
1991 {
1992     /*
1993      * MIN/MAX operations across a vector: compute the min or
1994      * max of the initial value in a general purpose register
1995      * and all the elements in the vector, and store it back
1996      * into the general purpose register.
1997      */
1998     TCGv_ptr qm;
1999     TCGv_i32 rda;
2000 
2001     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qm) ||
2002         !fn || a->rda == 13 || a->rda == 15) {
2003         /* Rda cases are UNPREDICTABLE */
2004         return false;
2005     }
2006     if (!mve_eci_check(s) || !vfp_access_check(s)) {
2007         return true;
2008     }
2009 
2010     qm = mve_qreg_ptr(a->qm);
2011     rda = load_reg(s, a->rda);
2012     fn(rda, cpu_env, qm, rda);
2013     store_reg(s, a->rda, rda);
2014     tcg_temp_free_ptr(qm);
2015     mve_update_eci(s);
2016     return true;
2017 }
2018 
2019 #define DO_VMAXV(INSN, FN)                                      \
2020     static bool trans_##INSN(DisasContext *s, arg_vmaxv *a)     \
2021     {                                                           \
2022         static MVEGenVADDVFn * const fns[] = {                  \
2023             gen_helper_mve_##FN##b,                             \
2024             gen_helper_mve_##FN##h,                             \
2025             gen_helper_mve_##FN##w,                             \
2026             NULL,                                               \
2027         };                                                      \
2028         return do_vmaxv(s, a, fns[a->size]);                    \
2029     }
2030 
2031 DO_VMAXV(VMAXV_S, vmaxvs)
2032 DO_VMAXV(VMAXV_U, vmaxvu)
2033 DO_VMAXV(VMAXAV, vmaxav)
2034 DO_VMAXV(VMINV_S, vminvs)
2035 DO_VMAXV(VMINV_U, vminvu)
2036 DO_VMAXV(VMINAV, vminav)
2037 
2038 #define DO_VMAXV_FP(INSN, FN)                                   \
2039     static bool trans_##INSN(DisasContext *s, arg_vmaxv *a)     \
2040     {                                                           \
2041         static MVEGenVADDVFn * const fns[] = {                  \
2042             NULL,                                               \
2043             gen_helper_mve_##FN##h,                             \
2044             gen_helper_mve_##FN##s,                             \
2045             NULL,                                               \
2046         };                                                      \
2047         if (!dc_isar_feature(aa32_mve_fp, s)) {                 \
2048             return false;                                       \
2049         }                                                       \
2050         return do_vmaxv(s, a, fns[a->size]);                    \
2051     }
2052 
2053 DO_VMAXV_FP(VMAXNMV, vmaxnmv)
2054 DO_VMAXV_FP(VMINNMV, vminnmv)
2055 DO_VMAXV_FP(VMAXNMAV, vmaxnmav)
2056 DO_VMAXV_FP(VMINNMAV, vminnmav)
2057 
2058 static bool do_vabav(DisasContext *s, arg_vabav *a, MVEGenVABAVFn *fn)
2059 {
2060     /* Absolute difference accumulated across vector */
2061     TCGv_ptr qn, qm;
2062     TCGv_i32 rda;
2063 
2064     if (!dc_isar_feature(aa32_mve, s) ||
2065         !mve_check_qreg_bank(s, a->qm | a->qn) ||
2066         !fn || a->rda == 13 || a->rda == 15) {
2067         /* Rda cases are UNPREDICTABLE */
2068         return false;
2069     }
2070     if (!mve_eci_check(s) || !vfp_access_check(s)) {
2071         return true;
2072     }
2073 
2074     qm = mve_qreg_ptr(a->qm);
2075     qn = mve_qreg_ptr(a->qn);
2076     rda = load_reg(s, a->rda);
2077     fn(rda, cpu_env, qn, qm, rda);
2078     store_reg(s, a->rda, rda);
2079     tcg_temp_free_ptr(qm);
2080     tcg_temp_free_ptr(qn);
2081     mve_update_eci(s);
2082     return true;
2083 }
2084 
2085 #define DO_VABAV(INSN, FN)                                      \
2086     static bool trans_##INSN(DisasContext *s, arg_vabav *a)     \
2087     {                                                           \
2088         static MVEGenVABAVFn * const fns[] = {                  \
2089             gen_helper_mve_##FN##b,                             \
2090             gen_helper_mve_##FN##h,                             \
2091             gen_helper_mve_##FN##w,                             \
2092             NULL,                                               \
2093         };                                                      \
2094         return do_vabav(s, a, fns[a->size]);                    \
2095     }
2096 
2097 DO_VABAV(VABAV_S, vabavs)
2098 DO_VABAV(VABAV_U, vabavu)
2099 
2100 static bool trans_VMOV_to_2gp(DisasContext *s, arg_VMOV_to_2gp *a)
2101 {
2102     /*
2103      * VMOV two 32-bit vector lanes to two general-purpose registers.
2104      * This insn is not predicated but it is subject to beat-wise
2105      * execution if it is not in an IT block. For us this means
2106      * only that if PSR.ECI says we should not be executing the beat
2107      * corresponding to the lane of the vector register being accessed
2108      * then we should skip perfoming the move, and that we need to do
2109      * the usual check for bad ECI state and advance of ECI state.
2110      * (If PSR.ECI is non-zero then we cannot be in an IT block.)
2111      */
2112     TCGv_i32 tmp;
2113     int vd;
2114 
2115     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd) ||
2116         a->rt == 13 || a->rt == 15 || a->rt2 == 13 || a->rt2 == 15 ||
2117         a->rt == a->rt2) {
2118         /* Rt/Rt2 cases are UNPREDICTABLE */
2119         return false;
2120     }
2121     if (!mve_eci_check(s) || !vfp_access_check(s)) {
2122         return true;
2123     }
2124 
2125     /* Convert Qreg index to Dreg for read_neon_element32() etc */
2126     vd = a->qd * 2;
2127 
2128     if (!mve_skip_vmov(s, vd, a->idx, MO_32)) {
2129         tmp = tcg_temp_new_i32();
2130         read_neon_element32(tmp, vd, a->idx, MO_32);
2131         store_reg(s, a->rt, tmp);
2132     }
2133     if (!mve_skip_vmov(s, vd + 1, a->idx, MO_32)) {
2134         tmp = tcg_temp_new_i32();
2135         read_neon_element32(tmp, vd + 1, a->idx, MO_32);
2136         store_reg(s, a->rt2, tmp);
2137     }
2138 
2139     mve_update_and_store_eci(s);
2140     return true;
2141 }
2142 
2143 static bool trans_VMOV_from_2gp(DisasContext *s, arg_VMOV_to_2gp *a)
2144 {
2145     /*
2146      * VMOV two general-purpose registers to two 32-bit vector lanes.
2147      * This insn is not predicated but it is subject to beat-wise
2148      * execution if it is not in an IT block. For us this means
2149      * only that if PSR.ECI says we should not be executing the beat
2150      * corresponding to the lane of the vector register being accessed
2151      * then we should skip perfoming the move, and that we need to do
2152      * the usual check for bad ECI state and advance of ECI state.
2153      * (If PSR.ECI is non-zero then we cannot be in an IT block.)
2154      */
2155     TCGv_i32 tmp;
2156     int vd;
2157 
2158     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd) ||
2159         a->rt == 13 || a->rt == 15 || a->rt2 == 13 || a->rt2 == 15) {
2160         /* Rt/Rt2 cases are UNPREDICTABLE */
2161         return false;
2162     }
2163     if (!mve_eci_check(s) || !vfp_access_check(s)) {
2164         return true;
2165     }
2166 
2167     /* Convert Qreg idx to Dreg for read_neon_element32() etc */
2168     vd = a->qd * 2;
2169 
2170     if (!mve_skip_vmov(s, vd, a->idx, MO_32)) {
2171         tmp = load_reg(s, a->rt);
2172         write_neon_element32(tmp, vd, a->idx, MO_32);
2173         tcg_temp_free_i32(tmp);
2174     }
2175     if (!mve_skip_vmov(s, vd + 1, a->idx, MO_32)) {
2176         tmp = load_reg(s, a->rt2);
2177         write_neon_element32(tmp, vd + 1, a->idx, MO_32);
2178         tcg_temp_free_i32(tmp);
2179     }
2180 
2181     mve_update_and_store_eci(s);
2182     return true;
2183 }
2184