xref: /qemu/target/arm/tcg/translate-mve.c (revision bc3087f2531d2ddf0d2d8f842a01db90fb1568c2)
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_VEC(INSN, FN, VECFN)                             \
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_vec(s, a, fns[a->size], VECFN);           \
851     }
852 
853 #define DO_2OP(INSN, FN) DO_2OP_VEC(INSN, FN, NULL)
854 
855 DO_2OP_VEC(VADD, vadd, tcg_gen_gvec_add)
856 DO_2OP_VEC(VSUB, vsub, tcg_gen_gvec_sub)
857 DO_2OP_VEC(VMUL, vmul, tcg_gen_gvec_mul)
858 DO_2OP(VMULH_S, vmulhs)
859 DO_2OP(VMULH_U, vmulhu)
860 DO_2OP(VRMULH_S, vrmulhs)
861 DO_2OP(VRMULH_U, vrmulhu)
862 DO_2OP_VEC(VMAX_S, vmaxs, tcg_gen_gvec_smax)
863 DO_2OP_VEC(VMAX_U, vmaxu, tcg_gen_gvec_umax)
864 DO_2OP_VEC(VMIN_S, vmins, tcg_gen_gvec_smin)
865 DO_2OP_VEC(VMIN_U, vminu, tcg_gen_gvec_umin)
866 DO_2OP(VABD_S, vabds)
867 DO_2OP(VABD_U, vabdu)
868 DO_2OP(VHADD_S, vhadds)
869 DO_2OP(VHADD_U, vhaddu)
870 DO_2OP(VHSUB_S, vhsubs)
871 DO_2OP(VHSUB_U, vhsubu)
872 DO_2OP(VMULL_BS, vmullbs)
873 DO_2OP(VMULL_BU, vmullbu)
874 DO_2OP(VMULL_TS, vmullts)
875 DO_2OP(VMULL_TU, vmulltu)
876 DO_2OP(VQDMULH, vqdmulh)
877 DO_2OP(VQRDMULH, vqrdmulh)
878 DO_2OP(VQADD_S, vqadds)
879 DO_2OP(VQADD_U, vqaddu)
880 DO_2OP(VQSUB_S, vqsubs)
881 DO_2OP(VQSUB_U, vqsubu)
882 DO_2OP(VSHL_S, vshls)
883 DO_2OP(VSHL_U, vshlu)
884 DO_2OP(VRSHL_S, vrshls)
885 DO_2OP(VRSHL_U, vrshlu)
886 DO_2OP(VQSHL_S, vqshls)
887 DO_2OP(VQSHL_U, vqshlu)
888 DO_2OP(VQRSHL_S, vqrshls)
889 DO_2OP(VQRSHL_U, vqrshlu)
890 DO_2OP(VQDMLADH, vqdmladh)
891 DO_2OP(VQDMLADHX, vqdmladhx)
892 DO_2OP(VQRDMLADH, vqrdmladh)
893 DO_2OP(VQRDMLADHX, vqrdmladhx)
894 DO_2OP(VQDMLSDH, vqdmlsdh)
895 DO_2OP(VQDMLSDHX, vqdmlsdhx)
896 DO_2OP(VQRDMLSDH, vqrdmlsdh)
897 DO_2OP(VQRDMLSDHX, vqrdmlsdhx)
898 DO_2OP(VRHADD_S, vrhadds)
899 DO_2OP(VRHADD_U, vrhaddu)
900 /*
901  * VCADD Qd == Qm at size MO_32 is UNPREDICTABLE; we choose not to diagnose
902  * so we can reuse the DO_2OP macro. (Our implementation calculates the
903  * "expected" results in this case.) Similarly for VHCADD.
904  */
905 DO_2OP(VCADD90, vcadd90)
906 DO_2OP(VCADD270, vcadd270)
907 DO_2OP(VHCADD90, vhcadd90)
908 DO_2OP(VHCADD270, vhcadd270)
909 
910 static bool trans_VQDMULLB(DisasContext *s, arg_2op *a)
911 {
912     static MVEGenTwoOpFn * const fns[] = {
913         NULL,
914         gen_helper_mve_vqdmullbh,
915         gen_helper_mve_vqdmullbw,
916         NULL,
917     };
918     if (a->size == MO_32 && (a->qd == a->qm || a->qd == a->qn)) {
919         /* UNPREDICTABLE; we choose to undef */
920         return false;
921     }
922     return do_2op(s, a, fns[a->size]);
923 }
924 
925 static bool trans_VQDMULLT(DisasContext *s, arg_2op *a)
926 {
927     static MVEGenTwoOpFn * const fns[] = {
928         NULL,
929         gen_helper_mve_vqdmullth,
930         gen_helper_mve_vqdmulltw,
931         NULL,
932     };
933     if (a->size == MO_32 && (a->qd == a->qm || a->qd == a->qn)) {
934         /* UNPREDICTABLE; we choose to undef */
935         return false;
936     }
937     return do_2op(s, a, fns[a->size]);
938 }
939 
940 static bool trans_VMULLP_B(DisasContext *s, arg_2op *a)
941 {
942     /*
943      * Note that a->size indicates the output size, ie VMULL.P8
944      * is the 8x8->16 operation and a->size is MO_16; VMULL.P16
945      * is the 16x16->32 operation and a->size is MO_32.
946      */
947     static MVEGenTwoOpFn * const fns[] = {
948         NULL,
949         gen_helper_mve_vmullpbh,
950         gen_helper_mve_vmullpbw,
951         NULL,
952     };
953     return do_2op(s, a, fns[a->size]);
954 }
955 
956 static bool trans_VMULLP_T(DisasContext *s, arg_2op *a)
957 {
958     /* a->size is as for trans_VMULLP_B */
959     static MVEGenTwoOpFn * const fns[] = {
960         NULL,
961         gen_helper_mve_vmullpth,
962         gen_helper_mve_vmullptw,
963         NULL,
964     };
965     return do_2op(s, a, fns[a->size]);
966 }
967 
968 /*
969  * VADC and VSBC: these perform an add-with-carry or subtract-with-carry
970  * of the 32-bit elements in each lane of the input vectors, where the
971  * carry-out of each add is the carry-in of the next.  The initial carry
972  * input is either fixed (0 for VADCI, 1 for VSBCI) or is from FPSCR.C
973  * (for VADC and VSBC); the carry out at the end is written back to FPSCR.C.
974  * These insns are subject to beat-wise execution.  Partial execution
975  * of an I=1 (initial carry input fixed) insn which does not
976  * execute the first beat must start with the current FPSCR.NZCV
977  * value, not the fixed constant input.
978  */
979 static bool trans_VADC(DisasContext *s, arg_2op *a)
980 {
981     return do_2op(s, a, gen_helper_mve_vadc);
982 }
983 
984 static bool trans_VADCI(DisasContext *s, arg_2op *a)
985 {
986     if (mve_skip_first_beat(s)) {
987         return trans_VADC(s, a);
988     }
989     return do_2op(s, a, gen_helper_mve_vadci);
990 }
991 
992 static bool trans_VSBC(DisasContext *s, arg_2op *a)
993 {
994     return do_2op(s, a, gen_helper_mve_vsbc);
995 }
996 
997 static bool trans_VSBCI(DisasContext *s, arg_2op *a)
998 {
999     if (mve_skip_first_beat(s)) {
1000         return trans_VSBC(s, a);
1001     }
1002     return do_2op(s, a, gen_helper_mve_vsbci);
1003 }
1004 
1005 #define DO_2OP_FP(INSN, FN)                                     \
1006     static bool trans_##INSN(DisasContext *s, arg_2op *a)       \
1007     {                                                           \
1008         static MVEGenTwoOpFn * const fns[] = {                  \
1009             NULL,                                               \
1010             gen_helper_mve_##FN##h,                             \
1011             gen_helper_mve_##FN##s,                             \
1012             NULL,                                               \
1013         };                                                      \
1014         if (!dc_isar_feature(aa32_mve_fp, s)) {                 \
1015             return false;                                       \
1016         }                                                       \
1017         return do_2op(s, a, fns[a->size]);                      \
1018     }
1019 
1020 DO_2OP_FP(VADD_fp, vfadd)
1021 DO_2OP_FP(VSUB_fp, vfsub)
1022 DO_2OP_FP(VMUL_fp, vfmul)
1023 DO_2OP_FP(VABD_fp, vfabd)
1024 DO_2OP_FP(VMAXNM, vmaxnm)
1025 DO_2OP_FP(VMINNM, vminnm)
1026 DO_2OP_FP(VCADD90_fp, vfcadd90)
1027 DO_2OP_FP(VCADD270_fp, vfcadd270)
1028 DO_2OP_FP(VFMA, vfma)
1029 DO_2OP_FP(VFMS, vfms)
1030 DO_2OP_FP(VCMUL0, vcmul0)
1031 DO_2OP_FP(VCMUL90, vcmul90)
1032 DO_2OP_FP(VCMUL180, vcmul180)
1033 DO_2OP_FP(VCMUL270, vcmul270)
1034 DO_2OP_FP(VCMLA0, vcmla0)
1035 DO_2OP_FP(VCMLA90, vcmla90)
1036 DO_2OP_FP(VCMLA180, vcmla180)
1037 DO_2OP_FP(VCMLA270, vcmla270)
1038 DO_2OP_FP(VMAXNMA, vmaxnma)
1039 DO_2OP_FP(VMINNMA, vminnma)
1040 
1041 static bool do_2op_scalar(DisasContext *s, arg_2scalar *a,
1042                           MVEGenTwoOpScalarFn fn)
1043 {
1044     TCGv_ptr qd, qn;
1045     TCGv_i32 rm;
1046 
1047     if (!dc_isar_feature(aa32_mve, s) ||
1048         !mve_check_qreg_bank(s, a->qd | a->qn) ||
1049         !fn) {
1050         return false;
1051     }
1052     if (a->rm == 13 || a->rm == 15) {
1053         /* UNPREDICTABLE */
1054         return false;
1055     }
1056     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1057         return true;
1058     }
1059 
1060     qd = mve_qreg_ptr(a->qd);
1061     qn = mve_qreg_ptr(a->qn);
1062     rm = load_reg(s, a->rm);
1063     fn(cpu_env, qd, qn, rm);
1064     tcg_temp_free_i32(rm);
1065     tcg_temp_free_ptr(qd);
1066     tcg_temp_free_ptr(qn);
1067     mve_update_eci(s);
1068     return true;
1069 }
1070 
1071 #define DO_2OP_SCALAR(INSN, FN)                                 \
1072     static bool trans_##INSN(DisasContext *s, arg_2scalar *a)   \
1073     {                                                           \
1074         static MVEGenTwoOpScalarFn * const fns[] = {            \
1075             gen_helper_mve_##FN##b,                             \
1076             gen_helper_mve_##FN##h,                             \
1077             gen_helper_mve_##FN##w,                             \
1078             NULL,                                               \
1079         };                                                      \
1080         return do_2op_scalar(s, a, fns[a->size]);               \
1081     }
1082 
1083 DO_2OP_SCALAR(VADD_scalar, vadd_scalar)
1084 DO_2OP_SCALAR(VSUB_scalar, vsub_scalar)
1085 DO_2OP_SCALAR(VMUL_scalar, vmul_scalar)
1086 DO_2OP_SCALAR(VHADD_S_scalar, vhadds_scalar)
1087 DO_2OP_SCALAR(VHADD_U_scalar, vhaddu_scalar)
1088 DO_2OP_SCALAR(VHSUB_S_scalar, vhsubs_scalar)
1089 DO_2OP_SCALAR(VHSUB_U_scalar, vhsubu_scalar)
1090 DO_2OP_SCALAR(VQADD_S_scalar, vqadds_scalar)
1091 DO_2OP_SCALAR(VQADD_U_scalar, vqaddu_scalar)
1092 DO_2OP_SCALAR(VQSUB_S_scalar, vqsubs_scalar)
1093 DO_2OP_SCALAR(VQSUB_U_scalar, vqsubu_scalar)
1094 DO_2OP_SCALAR(VQDMULH_scalar, vqdmulh_scalar)
1095 DO_2OP_SCALAR(VQRDMULH_scalar, vqrdmulh_scalar)
1096 DO_2OP_SCALAR(VBRSR, vbrsr)
1097 DO_2OP_SCALAR(VMLA, vmla)
1098 DO_2OP_SCALAR(VMLAS, vmlas)
1099 DO_2OP_SCALAR(VQDMLAH, vqdmlah)
1100 DO_2OP_SCALAR(VQRDMLAH, vqrdmlah)
1101 DO_2OP_SCALAR(VQDMLASH, vqdmlash)
1102 DO_2OP_SCALAR(VQRDMLASH, vqrdmlash)
1103 
1104 static bool trans_VQDMULLB_scalar(DisasContext *s, arg_2scalar *a)
1105 {
1106     static MVEGenTwoOpScalarFn * const fns[] = {
1107         NULL,
1108         gen_helper_mve_vqdmullb_scalarh,
1109         gen_helper_mve_vqdmullb_scalarw,
1110         NULL,
1111     };
1112     if (a->qd == a->qn && a->size == MO_32) {
1113         /* UNPREDICTABLE; we choose to undef */
1114         return false;
1115     }
1116     return do_2op_scalar(s, a, fns[a->size]);
1117 }
1118 
1119 static bool trans_VQDMULLT_scalar(DisasContext *s, arg_2scalar *a)
1120 {
1121     static MVEGenTwoOpScalarFn * const fns[] = {
1122         NULL,
1123         gen_helper_mve_vqdmullt_scalarh,
1124         gen_helper_mve_vqdmullt_scalarw,
1125         NULL,
1126     };
1127     if (a->qd == a->qn && a->size == MO_32) {
1128         /* UNPREDICTABLE; we choose to undef */
1129         return false;
1130     }
1131     return do_2op_scalar(s, a, fns[a->size]);
1132 }
1133 
1134 
1135 #define DO_2OP_FP_SCALAR(INSN, FN)                              \
1136     static bool trans_##INSN(DisasContext *s, arg_2scalar *a)   \
1137     {                                                           \
1138         static MVEGenTwoOpScalarFn * const fns[] = {            \
1139             NULL,                                               \
1140             gen_helper_mve_##FN##h,                             \
1141             gen_helper_mve_##FN##s,                             \
1142             NULL,                                               \
1143         };                                                      \
1144         if (!dc_isar_feature(aa32_mve_fp, s)) {                 \
1145             return false;                                       \
1146         }                                                       \
1147         return do_2op_scalar(s, a, fns[a->size]);               \
1148     }
1149 
1150 DO_2OP_FP_SCALAR(VADD_fp_scalar, vfadd_scalar)
1151 DO_2OP_FP_SCALAR(VSUB_fp_scalar, vfsub_scalar)
1152 DO_2OP_FP_SCALAR(VMUL_fp_scalar, vfmul_scalar)
1153 DO_2OP_FP_SCALAR(VFMA_scalar, vfma_scalar)
1154 DO_2OP_FP_SCALAR(VFMAS_scalar, vfmas_scalar)
1155 
1156 static bool do_long_dual_acc(DisasContext *s, arg_vmlaldav *a,
1157                              MVEGenLongDualAccOpFn *fn)
1158 {
1159     TCGv_ptr qn, qm;
1160     TCGv_i64 rda;
1161     TCGv_i32 rdalo, rdahi;
1162 
1163     if (!dc_isar_feature(aa32_mve, s) ||
1164         !mve_check_qreg_bank(s, a->qn | a->qm) ||
1165         !fn) {
1166         return false;
1167     }
1168     /*
1169      * rdahi == 13 is UNPREDICTABLE; rdahi == 15 is a related
1170      * encoding; rdalo always has bit 0 clear so cannot be 13 or 15.
1171      */
1172     if (a->rdahi == 13 || a->rdahi == 15) {
1173         return false;
1174     }
1175     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1176         return true;
1177     }
1178 
1179     qn = mve_qreg_ptr(a->qn);
1180     qm = mve_qreg_ptr(a->qm);
1181 
1182     /*
1183      * This insn is subject to beat-wise execution. Partial execution
1184      * of an A=0 (no-accumulate) insn which does not execute the first
1185      * beat must start with the current rda value, not 0.
1186      */
1187     if (a->a || mve_skip_first_beat(s)) {
1188         rda = tcg_temp_new_i64();
1189         rdalo = load_reg(s, a->rdalo);
1190         rdahi = load_reg(s, a->rdahi);
1191         tcg_gen_concat_i32_i64(rda, rdalo, rdahi);
1192         tcg_temp_free_i32(rdalo);
1193         tcg_temp_free_i32(rdahi);
1194     } else {
1195         rda = tcg_const_i64(0);
1196     }
1197 
1198     fn(rda, cpu_env, qn, qm, rda);
1199     tcg_temp_free_ptr(qn);
1200     tcg_temp_free_ptr(qm);
1201 
1202     rdalo = tcg_temp_new_i32();
1203     rdahi = tcg_temp_new_i32();
1204     tcg_gen_extrl_i64_i32(rdalo, rda);
1205     tcg_gen_extrh_i64_i32(rdahi, rda);
1206     store_reg(s, a->rdalo, rdalo);
1207     store_reg(s, a->rdahi, rdahi);
1208     tcg_temp_free_i64(rda);
1209     mve_update_eci(s);
1210     return true;
1211 }
1212 
1213 static bool trans_VMLALDAV_S(DisasContext *s, arg_vmlaldav *a)
1214 {
1215     static MVEGenLongDualAccOpFn * const fns[4][2] = {
1216         { NULL, NULL },
1217         { gen_helper_mve_vmlaldavsh, gen_helper_mve_vmlaldavxsh },
1218         { gen_helper_mve_vmlaldavsw, gen_helper_mve_vmlaldavxsw },
1219         { NULL, NULL },
1220     };
1221     return do_long_dual_acc(s, a, fns[a->size][a->x]);
1222 }
1223 
1224 static bool trans_VMLALDAV_U(DisasContext *s, arg_vmlaldav *a)
1225 {
1226     static MVEGenLongDualAccOpFn * const fns[4][2] = {
1227         { NULL, NULL },
1228         { gen_helper_mve_vmlaldavuh, NULL },
1229         { gen_helper_mve_vmlaldavuw, NULL },
1230         { NULL, NULL },
1231     };
1232     return do_long_dual_acc(s, a, fns[a->size][a->x]);
1233 }
1234 
1235 static bool trans_VMLSLDAV(DisasContext *s, arg_vmlaldav *a)
1236 {
1237     static MVEGenLongDualAccOpFn * const fns[4][2] = {
1238         { NULL, NULL },
1239         { gen_helper_mve_vmlsldavsh, gen_helper_mve_vmlsldavxsh },
1240         { gen_helper_mve_vmlsldavsw, gen_helper_mve_vmlsldavxsw },
1241         { NULL, NULL },
1242     };
1243     return do_long_dual_acc(s, a, fns[a->size][a->x]);
1244 }
1245 
1246 static bool trans_VRMLALDAVH_S(DisasContext *s, arg_vmlaldav *a)
1247 {
1248     static MVEGenLongDualAccOpFn * const fns[] = {
1249         gen_helper_mve_vrmlaldavhsw, gen_helper_mve_vrmlaldavhxsw,
1250     };
1251     return do_long_dual_acc(s, a, fns[a->x]);
1252 }
1253 
1254 static bool trans_VRMLALDAVH_U(DisasContext *s, arg_vmlaldav *a)
1255 {
1256     static MVEGenLongDualAccOpFn * const fns[] = {
1257         gen_helper_mve_vrmlaldavhuw, NULL,
1258     };
1259     return do_long_dual_acc(s, a, fns[a->x]);
1260 }
1261 
1262 static bool trans_VRMLSLDAVH(DisasContext *s, arg_vmlaldav *a)
1263 {
1264     static MVEGenLongDualAccOpFn * const fns[] = {
1265         gen_helper_mve_vrmlsldavhsw, gen_helper_mve_vrmlsldavhxsw,
1266     };
1267     return do_long_dual_acc(s, a, fns[a->x]);
1268 }
1269 
1270 static bool do_dual_acc(DisasContext *s, arg_vmladav *a, MVEGenDualAccOpFn *fn)
1271 {
1272     TCGv_ptr qn, qm;
1273     TCGv_i32 rda;
1274 
1275     if (!dc_isar_feature(aa32_mve, s) ||
1276         !mve_check_qreg_bank(s, a->qn) ||
1277         !fn) {
1278         return false;
1279     }
1280     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1281         return true;
1282     }
1283 
1284     qn = mve_qreg_ptr(a->qn);
1285     qm = mve_qreg_ptr(a->qm);
1286 
1287     /*
1288      * This insn is subject to beat-wise execution. Partial execution
1289      * of an A=0 (no-accumulate) insn which does not execute the first
1290      * beat must start with the current rda value, not 0.
1291      */
1292     if (a->a || mve_skip_first_beat(s)) {
1293         rda = load_reg(s, a->rda);
1294     } else {
1295         rda = tcg_const_i32(0);
1296     }
1297 
1298     fn(rda, cpu_env, qn, qm, rda);
1299     store_reg(s, a->rda, rda);
1300     tcg_temp_free_ptr(qn);
1301     tcg_temp_free_ptr(qm);
1302 
1303     mve_update_eci(s);
1304     return true;
1305 }
1306 
1307 #define DO_DUAL_ACC(INSN, FN)                                           \
1308     static bool trans_##INSN(DisasContext *s, arg_vmladav *a)           \
1309     {                                                                   \
1310         static MVEGenDualAccOpFn * const fns[4][2] = {                  \
1311             { gen_helper_mve_##FN##b, gen_helper_mve_##FN##xb },        \
1312             { gen_helper_mve_##FN##h, gen_helper_mve_##FN##xh },        \
1313             { gen_helper_mve_##FN##w, gen_helper_mve_##FN##xw },        \
1314             { NULL, NULL },                                             \
1315         };                                                              \
1316         return do_dual_acc(s, a, fns[a->size][a->x]);                   \
1317     }
1318 
1319 DO_DUAL_ACC(VMLADAV_S, vmladavs)
1320 DO_DUAL_ACC(VMLSDAV, vmlsdav)
1321 
1322 static bool trans_VMLADAV_U(DisasContext *s, arg_vmladav *a)
1323 {
1324     static MVEGenDualAccOpFn * const fns[4][2] = {
1325         { gen_helper_mve_vmladavub, NULL },
1326         { gen_helper_mve_vmladavuh, NULL },
1327         { gen_helper_mve_vmladavuw, NULL },
1328         { NULL, NULL },
1329     };
1330     return do_dual_acc(s, a, fns[a->size][a->x]);
1331 }
1332 
1333 static void gen_vpst(DisasContext *s, uint32_t mask)
1334 {
1335     /*
1336      * Set the VPR mask fields. We take advantage of MASK01 and MASK23
1337      * being adjacent fields in the register.
1338      *
1339      * Updating the masks is not predicated, but it is subject to beat-wise
1340      * execution, and the mask is updated on the odd-numbered beats.
1341      * So if PSR.ECI says we should skip beat 1, we mustn't update the
1342      * 01 mask field.
1343      */
1344     TCGv_i32 vpr = load_cpu_field(v7m.vpr);
1345     switch (s->eci) {
1346     case ECI_NONE:
1347     case ECI_A0:
1348         /* Update both 01 and 23 fields */
1349         tcg_gen_deposit_i32(vpr, vpr,
1350                             tcg_constant_i32(mask | (mask << 4)),
1351                             R_V7M_VPR_MASK01_SHIFT,
1352                             R_V7M_VPR_MASK01_LENGTH + R_V7M_VPR_MASK23_LENGTH);
1353         break;
1354     case ECI_A0A1:
1355     case ECI_A0A1A2:
1356     case ECI_A0A1A2B0:
1357         /* Update only the 23 mask field */
1358         tcg_gen_deposit_i32(vpr, vpr,
1359                             tcg_constant_i32(mask),
1360                             R_V7M_VPR_MASK23_SHIFT, R_V7M_VPR_MASK23_LENGTH);
1361         break;
1362     default:
1363         g_assert_not_reached();
1364     }
1365     store_cpu_field(vpr, v7m.vpr);
1366 }
1367 
1368 static bool trans_VPST(DisasContext *s, arg_VPST *a)
1369 {
1370     /* mask == 0 is a "related encoding" */
1371     if (!dc_isar_feature(aa32_mve, s) || !a->mask) {
1372         return false;
1373     }
1374     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1375         return true;
1376     }
1377     gen_vpst(s, a->mask);
1378     mve_update_and_store_eci(s);
1379     return true;
1380 }
1381 
1382 static bool trans_VPNOT(DisasContext *s, arg_VPNOT *a)
1383 {
1384     /*
1385      * Invert the predicate in VPR.P0. We have call out to
1386      * a helper because this insn itself is beatwise and can
1387      * be predicated.
1388      */
1389     if (!dc_isar_feature(aa32_mve, s)) {
1390         return false;
1391     }
1392     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1393         return true;
1394     }
1395 
1396     gen_helper_mve_vpnot(cpu_env);
1397     /* This insn updates predication bits */
1398     s->base.is_jmp = DISAS_UPDATE_NOCHAIN;
1399     mve_update_eci(s);
1400     return true;
1401 }
1402 
1403 static bool trans_VADDV(DisasContext *s, arg_VADDV *a)
1404 {
1405     /* VADDV: vector add across vector */
1406     static MVEGenVADDVFn * const fns[4][2] = {
1407         { gen_helper_mve_vaddvsb, gen_helper_mve_vaddvub },
1408         { gen_helper_mve_vaddvsh, gen_helper_mve_vaddvuh },
1409         { gen_helper_mve_vaddvsw, gen_helper_mve_vaddvuw },
1410         { NULL, NULL }
1411     };
1412     TCGv_ptr qm;
1413     TCGv_i32 rda;
1414 
1415     if (!dc_isar_feature(aa32_mve, s) ||
1416         a->size == 3) {
1417         return false;
1418     }
1419     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1420         return true;
1421     }
1422 
1423     /*
1424      * This insn is subject to beat-wise execution. Partial execution
1425      * of an A=0 (no-accumulate) insn which does not execute the first
1426      * beat must start with the current value of Rda, not zero.
1427      */
1428     if (a->a || mve_skip_first_beat(s)) {
1429         /* Accumulate input from Rda */
1430         rda = load_reg(s, a->rda);
1431     } else {
1432         /* Accumulate starting at zero */
1433         rda = tcg_const_i32(0);
1434     }
1435 
1436     qm = mve_qreg_ptr(a->qm);
1437     fns[a->size][a->u](rda, cpu_env, qm, rda);
1438     store_reg(s, a->rda, rda);
1439     tcg_temp_free_ptr(qm);
1440 
1441     mve_update_eci(s);
1442     return true;
1443 }
1444 
1445 static bool trans_VADDLV(DisasContext *s, arg_VADDLV *a)
1446 {
1447     /*
1448      * Vector Add Long Across Vector: accumulate the 32-bit
1449      * elements of the vector into a 64-bit result stored in
1450      * a pair of general-purpose registers.
1451      * No need to check Qm's bank: it is only 3 bits in decode.
1452      */
1453     TCGv_ptr qm;
1454     TCGv_i64 rda;
1455     TCGv_i32 rdalo, rdahi;
1456 
1457     if (!dc_isar_feature(aa32_mve, s)) {
1458         return false;
1459     }
1460     /*
1461      * rdahi == 13 is UNPREDICTABLE; rdahi == 15 is a related
1462      * encoding; rdalo always has bit 0 clear so cannot be 13 or 15.
1463      */
1464     if (a->rdahi == 13 || a->rdahi == 15) {
1465         return false;
1466     }
1467     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1468         return true;
1469     }
1470 
1471     /*
1472      * This insn is subject to beat-wise execution. Partial execution
1473      * of an A=0 (no-accumulate) insn which does not execute the first
1474      * beat must start with the current value of RdaHi:RdaLo, not zero.
1475      */
1476     if (a->a || mve_skip_first_beat(s)) {
1477         /* Accumulate input from RdaHi:RdaLo */
1478         rda = tcg_temp_new_i64();
1479         rdalo = load_reg(s, a->rdalo);
1480         rdahi = load_reg(s, a->rdahi);
1481         tcg_gen_concat_i32_i64(rda, rdalo, rdahi);
1482         tcg_temp_free_i32(rdalo);
1483         tcg_temp_free_i32(rdahi);
1484     } else {
1485         /* Accumulate starting at zero */
1486         rda = tcg_const_i64(0);
1487     }
1488 
1489     qm = mve_qreg_ptr(a->qm);
1490     if (a->u) {
1491         gen_helper_mve_vaddlv_u(rda, cpu_env, qm, rda);
1492     } else {
1493         gen_helper_mve_vaddlv_s(rda, cpu_env, qm, rda);
1494     }
1495     tcg_temp_free_ptr(qm);
1496 
1497     rdalo = tcg_temp_new_i32();
1498     rdahi = tcg_temp_new_i32();
1499     tcg_gen_extrl_i64_i32(rdalo, rda);
1500     tcg_gen_extrh_i64_i32(rdahi, rda);
1501     store_reg(s, a->rdalo, rdalo);
1502     store_reg(s, a->rdahi, rdahi);
1503     tcg_temp_free_i64(rda);
1504     mve_update_eci(s);
1505     return true;
1506 }
1507 
1508 static bool do_1imm(DisasContext *s, arg_1imm *a, MVEGenOneOpImmFn *fn)
1509 {
1510     TCGv_ptr qd;
1511     uint64_t imm;
1512 
1513     if (!dc_isar_feature(aa32_mve, s) ||
1514         !mve_check_qreg_bank(s, a->qd) ||
1515         !fn) {
1516         return false;
1517     }
1518     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1519         return true;
1520     }
1521 
1522     imm = asimd_imm_const(a->imm, a->cmode, a->op);
1523 
1524     qd = mve_qreg_ptr(a->qd);
1525     fn(cpu_env, qd, tcg_constant_i64(imm));
1526     tcg_temp_free_ptr(qd);
1527     mve_update_eci(s);
1528     return true;
1529 }
1530 
1531 static bool trans_Vimm_1r(DisasContext *s, arg_1imm *a)
1532 {
1533     /* Handle decode of cmode/op here between VORR/VBIC/VMOV */
1534     MVEGenOneOpImmFn *fn;
1535 
1536     if ((a->cmode & 1) && a->cmode < 12) {
1537         if (a->op) {
1538             /*
1539              * For op=1, the immediate will be inverted by asimd_imm_const(),
1540              * so the VBIC becomes a logical AND operation.
1541              */
1542             fn = gen_helper_mve_vandi;
1543         } else {
1544             fn = gen_helper_mve_vorri;
1545         }
1546     } else {
1547         /* There is one unallocated cmode/op combination in this space */
1548         if (a->cmode == 15 && a->op == 1) {
1549             return false;
1550         }
1551         /* asimd_imm_const() sorts out VMVNI vs VMOVI for us */
1552         fn = gen_helper_mve_vmovi;
1553     }
1554     return do_1imm(s, a, fn);
1555 }
1556 
1557 static bool do_2shift(DisasContext *s, arg_2shift *a, MVEGenTwoOpShiftFn fn,
1558                       bool negateshift)
1559 {
1560     TCGv_ptr qd, qm;
1561     int shift = a->shift;
1562 
1563     if (!dc_isar_feature(aa32_mve, s) ||
1564         !mve_check_qreg_bank(s, a->qd | a->qm) ||
1565         !fn) {
1566         return false;
1567     }
1568     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1569         return true;
1570     }
1571 
1572     /*
1573      * When we handle a right shift insn using a left-shift helper
1574      * which permits a negative shift count to indicate a right-shift,
1575      * we must negate the shift count.
1576      */
1577     if (negateshift) {
1578         shift = -shift;
1579     }
1580 
1581     qd = mve_qreg_ptr(a->qd);
1582     qm = mve_qreg_ptr(a->qm);
1583     fn(cpu_env, qd, qm, tcg_constant_i32(shift));
1584     tcg_temp_free_ptr(qd);
1585     tcg_temp_free_ptr(qm);
1586     mve_update_eci(s);
1587     return true;
1588 }
1589 
1590 #define DO_2SHIFT(INSN, FN, NEGATESHIFT)                         \
1591     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1592     {                                                           \
1593         static MVEGenTwoOpShiftFn * const fns[] = {             \
1594             gen_helper_mve_##FN##b,                             \
1595             gen_helper_mve_##FN##h,                             \
1596             gen_helper_mve_##FN##w,                             \
1597             NULL,                                               \
1598         };                                                      \
1599         return do_2shift(s, a, fns[a->size], NEGATESHIFT);      \
1600     }
1601 
1602 DO_2SHIFT(VSHLI, vshli_u, false)
1603 DO_2SHIFT(VQSHLI_S, vqshli_s, false)
1604 DO_2SHIFT(VQSHLI_U, vqshli_u, false)
1605 DO_2SHIFT(VQSHLUI, vqshlui_s, false)
1606 /* These right shifts use a left-shift helper with negated shift count */
1607 DO_2SHIFT(VSHRI_S, vshli_s, true)
1608 DO_2SHIFT(VSHRI_U, vshli_u, true)
1609 DO_2SHIFT(VRSHRI_S, vrshli_s, true)
1610 DO_2SHIFT(VRSHRI_U, vrshli_u, true)
1611 
1612 DO_2SHIFT(VSRI, vsri, false)
1613 DO_2SHIFT(VSLI, vsli, false)
1614 
1615 #define DO_2SHIFT_FP(INSN, FN)                                  \
1616     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1617     {                                                           \
1618         if (!dc_isar_feature(aa32_mve_fp, s)) {                 \
1619             return false;                                       \
1620         }                                                       \
1621         return do_2shift(s, a, gen_helper_mve_##FN, false);     \
1622     }
1623 
1624 DO_2SHIFT_FP(VCVT_SH_fixed, vcvt_sh)
1625 DO_2SHIFT_FP(VCVT_UH_fixed, vcvt_uh)
1626 DO_2SHIFT_FP(VCVT_HS_fixed, vcvt_hs)
1627 DO_2SHIFT_FP(VCVT_HU_fixed, vcvt_hu)
1628 DO_2SHIFT_FP(VCVT_SF_fixed, vcvt_sf)
1629 DO_2SHIFT_FP(VCVT_UF_fixed, vcvt_uf)
1630 DO_2SHIFT_FP(VCVT_FS_fixed, vcvt_fs)
1631 DO_2SHIFT_FP(VCVT_FU_fixed, vcvt_fu)
1632 
1633 static bool do_2shift_scalar(DisasContext *s, arg_shl_scalar *a,
1634                              MVEGenTwoOpShiftFn *fn)
1635 {
1636     TCGv_ptr qda;
1637     TCGv_i32 rm;
1638 
1639     if (!dc_isar_feature(aa32_mve, s) ||
1640         !mve_check_qreg_bank(s, a->qda) ||
1641         a->rm == 13 || a->rm == 15 || !fn) {
1642         /* Rm cases are UNPREDICTABLE */
1643         return false;
1644     }
1645     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1646         return true;
1647     }
1648 
1649     qda = mve_qreg_ptr(a->qda);
1650     rm = load_reg(s, a->rm);
1651     fn(cpu_env, qda, qda, rm);
1652     tcg_temp_free_ptr(qda);
1653     tcg_temp_free_i32(rm);
1654     mve_update_eci(s);
1655     return true;
1656 }
1657 
1658 #define DO_2SHIFT_SCALAR(INSN, FN)                                      \
1659     static bool trans_##INSN(DisasContext *s, arg_shl_scalar *a)        \
1660     {                                                                   \
1661         static MVEGenTwoOpShiftFn * const fns[] = {                     \
1662             gen_helper_mve_##FN##b,                                     \
1663             gen_helper_mve_##FN##h,                                     \
1664             gen_helper_mve_##FN##w,                                     \
1665             NULL,                                                       \
1666         };                                                              \
1667         return do_2shift_scalar(s, a, fns[a->size]);                    \
1668     }
1669 
1670 DO_2SHIFT_SCALAR(VSHL_S_scalar, vshli_s)
1671 DO_2SHIFT_SCALAR(VSHL_U_scalar, vshli_u)
1672 DO_2SHIFT_SCALAR(VRSHL_S_scalar, vrshli_s)
1673 DO_2SHIFT_SCALAR(VRSHL_U_scalar, vrshli_u)
1674 DO_2SHIFT_SCALAR(VQSHL_S_scalar, vqshli_s)
1675 DO_2SHIFT_SCALAR(VQSHL_U_scalar, vqshli_u)
1676 DO_2SHIFT_SCALAR(VQRSHL_S_scalar, vqrshli_s)
1677 DO_2SHIFT_SCALAR(VQRSHL_U_scalar, vqrshli_u)
1678 
1679 #define DO_VSHLL(INSN, FN)                                      \
1680     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1681     {                                                           \
1682         static MVEGenTwoOpShiftFn * const fns[] = {             \
1683             gen_helper_mve_##FN##b,                             \
1684             gen_helper_mve_##FN##h,                             \
1685         };                                                      \
1686         return do_2shift(s, a, fns[a->size], false);            \
1687     }
1688 
1689 DO_VSHLL(VSHLL_BS, vshllbs)
1690 DO_VSHLL(VSHLL_BU, vshllbu)
1691 DO_VSHLL(VSHLL_TS, vshllts)
1692 DO_VSHLL(VSHLL_TU, vshlltu)
1693 
1694 #define DO_2SHIFT_N(INSN, FN)                                   \
1695     static bool trans_##INSN(DisasContext *s, arg_2shift *a)    \
1696     {                                                           \
1697         static MVEGenTwoOpShiftFn * const fns[] = {             \
1698             gen_helper_mve_##FN##b,                             \
1699             gen_helper_mve_##FN##h,                             \
1700         };                                                      \
1701         return do_2shift(s, a, fns[a->size], false);            \
1702     }
1703 
1704 DO_2SHIFT_N(VSHRNB, vshrnb)
1705 DO_2SHIFT_N(VSHRNT, vshrnt)
1706 DO_2SHIFT_N(VRSHRNB, vrshrnb)
1707 DO_2SHIFT_N(VRSHRNT, vrshrnt)
1708 DO_2SHIFT_N(VQSHRNB_S, vqshrnb_s)
1709 DO_2SHIFT_N(VQSHRNT_S, vqshrnt_s)
1710 DO_2SHIFT_N(VQSHRNB_U, vqshrnb_u)
1711 DO_2SHIFT_N(VQSHRNT_U, vqshrnt_u)
1712 DO_2SHIFT_N(VQSHRUNB, vqshrunb)
1713 DO_2SHIFT_N(VQSHRUNT, vqshrunt)
1714 DO_2SHIFT_N(VQRSHRNB_S, vqrshrnb_s)
1715 DO_2SHIFT_N(VQRSHRNT_S, vqrshrnt_s)
1716 DO_2SHIFT_N(VQRSHRNB_U, vqrshrnb_u)
1717 DO_2SHIFT_N(VQRSHRNT_U, vqrshrnt_u)
1718 DO_2SHIFT_N(VQRSHRUNB, vqrshrunb)
1719 DO_2SHIFT_N(VQRSHRUNT, vqrshrunt)
1720 
1721 static bool trans_VSHLC(DisasContext *s, arg_VSHLC *a)
1722 {
1723     /*
1724      * Whole Vector Left Shift with Carry. The carry is taken
1725      * from a general purpose register and written back there.
1726      * An imm of 0 means "shift by 32".
1727      */
1728     TCGv_ptr qd;
1729     TCGv_i32 rdm;
1730 
1731     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) {
1732         return false;
1733     }
1734     if (a->rdm == 13 || a->rdm == 15) {
1735         /* CONSTRAINED UNPREDICTABLE: we UNDEF */
1736         return false;
1737     }
1738     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1739         return true;
1740     }
1741 
1742     qd = mve_qreg_ptr(a->qd);
1743     rdm = load_reg(s, a->rdm);
1744     gen_helper_mve_vshlc(rdm, cpu_env, qd, rdm, tcg_constant_i32(a->imm));
1745     store_reg(s, a->rdm, rdm);
1746     tcg_temp_free_ptr(qd);
1747     mve_update_eci(s);
1748     return true;
1749 }
1750 
1751 static bool do_vidup(DisasContext *s, arg_vidup *a, MVEGenVIDUPFn *fn)
1752 {
1753     TCGv_ptr qd;
1754     TCGv_i32 rn;
1755 
1756     /*
1757      * Vector increment/decrement with wrap and duplicate (VIDUP, VDDUP).
1758      * This fills the vector with elements of successively increasing
1759      * or decreasing values, starting from Rn.
1760      */
1761     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) {
1762         return false;
1763     }
1764     if (a->size == MO_64) {
1765         /* size 0b11 is another encoding */
1766         return false;
1767     }
1768     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1769         return true;
1770     }
1771 
1772     qd = mve_qreg_ptr(a->qd);
1773     rn = load_reg(s, a->rn);
1774     fn(rn, cpu_env, qd, rn, tcg_constant_i32(a->imm));
1775     store_reg(s, a->rn, rn);
1776     tcg_temp_free_ptr(qd);
1777     mve_update_eci(s);
1778     return true;
1779 }
1780 
1781 static bool do_viwdup(DisasContext *s, arg_viwdup *a, MVEGenVIWDUPFn *fn)
1782 {
1783     TCGv_ptr qd;
1784     TCGv_i32 rn, rm;
1785 
1786     /*
1787      * Vector increment/decrement with wrap and duplicate (VIWDUp, VDWDUP)
1788      * This fills the vector with elements of successively increasing
1789      * or decreasing values, starting from Rn. Rm specifies a point where
1790      * the count wraps back around to 0. The updated offset is written back
1791      * to Rn.
1792      */
1793     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd)) {
1794         return false;
1795     }
1796     if (!fn || a->rm == 13 || a->rm == 15) {
1797         /*
1798          * size 0b11 is another encoding; Rm == 13 is UNPREDICTABLE;
1799          * Rm == 13 is VIWDUP, VDWDUP.
1800          */
1801         return false;
1802     }
1803     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1804         return true;
1805     }
1806 
1807     qd = mve_qreg_ptr(a->qd);
1808     rn = load_reg(s, a->rn);
1809     rm = load_reg(s, a->rm);
1810     fn(rn, cpu_env, qd, rn, rm, tcg_constant_i32(a->imm));
1811     store_reg(s, a->rn, rn);
1812     tcg_temp_free_ptr(qd);
1813     tcg_temp_free_i32(rm);
1814     mve_update_eci(s);
1815     return true;
1816 }
1817 
1818 static bool trans_VIDUP(DisasContext *s, arg_vidup *a)
1819 {
1820     static MVEGenVIDUPFn * const fns[] = {
1821         gen_helper_mve_vidupb,
1822         gen_helper_mve_viduph,
1823         gen_helper_mve_vidupw,
1824         NULL,
1825     };
1826     return do_vidup(s, a, fns[a->size]);
1827 }
1828 
1829 static bool trans_VDDUP(DisasContext *s, arg_vidup *a)
1830 {
1831     static MVEGenVIDUPFn * const fns[] = {
1832         gen_helper_mve_vidupb,
1833         gen_helper_mve_viduph,
1834         gen_helper_mve_vidupw,
1835         NULL,
1836     };
1837     /* VDDUP is just like VIDUP but with a negative immediate */
1838     a->imm = -a->imm;
1839     return do_vidup(s, a, fns[a->size]);
1840 }
1841 
1842 static bool trans_VIWDUP(DisasContext *s, arg_viwdup *a)
1843 {
1844     static MVEGenVIWDUPFn * const fns[] = {
1845         gen_helper_mve_viwdupb,
1846         gen_helper_mve_viwduph,
1847         gen_helper_mve_viwdupw,
1848         NULL,
1849     };
1850     return do_viwdup(s, a, fns[a->size]);
1851 }
1852 
1853 static bool trans_VDWDUP(DisasContext *s, arg_viwdup *a)
1854 {
1855     static MVEGenVIWDUPFn * const fns[] = {
1856         gen_helper_mve_vdwdupb,
1857         gen_helper_mve_vdwduph,
1858         gen_helper_mve_vdwdupw,
1859         NULL,
1860     };
1861     return do_viwdup(s, a, fns[a->size]);
1862 }
1863 
1864 static bool do_vcmp(DisasContext *s, arg_vcmp *a, MVEGenCmpFn *fn)
1865 {
1866     TCGv_ptr qn, qm;
1867 
1868     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qm) ||
1869         !fn) {
1870         return false;
1871     }
1872     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1873         return true;
1874     }
1875 
1876     qn = mve_qreg_ptr(a->qn);
1877     qm = mve_qreg_ptr(a->qm);
1878     fn(cpu_env, qn, qm);
1879     tcg_temp_free_ptr(qn);
1880     tcg_temp_free_ptr(qm);
1881     if (a->mask) {
1882         /* VPT */
1883         gen_vpst(s, a->mask);
1884     }
1885     /* This insn updates predication bits */
1886     s->base.is_jmp = DISAS_UPDATE_NOCHAIN;
1887     mve_update_eci(s);
1888     return true;
1889 }
1890 
1891 static bool do_vcmp_scalar(DisasContext *s, arg_vcmp_scalar *a,
1892                            MVEGenScalarCmpFn *fn)
1893 {
1894     TCGv_ptr qn;
1895     TCGv_i32 rm;
1896 
1897     if (!dc_isar_feature(aa32_mve, s) || !fn || a->rm == 13) {
1898         return false;
1899     }
1900     if (!mve_eci_check(s) || !vfp_access_check(s)) {
1901         return true;
1902     }
1903 
1904     qn = mve_qreg_ptr(a->qn);
1905     if (a->rm == 15) {
1906         /* Encoding Rm=0b1111 means "constant zero" */
1907         rm = tcg_constant_i32(0);
1908     } else {
1909         rm = load_reg(s, a->rm);
1910     }
1911     fn(cpu_env, qn, rm);
1912     tcg_temp_free_ptr(qn);
1913     tcg_temp_free_i32(rm);
1914     if (a->mask) {
1915         /* VPT */
1916         gen_vpst(s, a->mask);
1917     }
1918     /* This insn updates predication bits */
1919     s->base.is_jmp = DISAS_UPDATE_NOCHAIN;
1920     mve_update_eci(s);
1921     return true;
1922 }
1923 
1924 #define DO_VCMP(INSN, FN)                                       \
1925     static bool trans_##INSN(DisasContext *s, arg_vcmp *a)      \
1926     {                                                           \
1927         static MVEGenCmpFn * const fns[] = {                    \
1928             gen_helper_mve_##FN##b,                             \
1929             gen_helper_mve_##FN##h,                             \
1930             gen_helper_mve_##FN##w,                             \
1931             NULL,                                               \
1932         };                                                      \
1933         return do_vcmp(s, a, fns[a->size]);                     \
1934     }                                                           \
1935     static bool trans_##INSN##_scalar(DisasContext *s,          \
1936                                       arg_vcmp_scalar *a)       \
1937     {                                                           \
1938         static MVEGenScalarCmpFn * const fns[] = {              \
1939             gen_helper_mve_##FN##_scalarb,                      \
1940             gen_helper_mve_##FN##_scalarh,                      \
1941             gen_helper_mve_##FN##_scalarw,                      \
1942             NULL,                                               \
1943         };                                                      \
1944         return do_vcmp_scalar(s, a, fns[a->size]);              \
1945     }
1946 
1947 DO_VCMP(VCMPEQ, vcmpeq)
1948 DO_VCMP(VCMPNE, vcmpne)
1949 DO_VCMP(VCMPCS, vcmpcs)
1950 DO_VCMP(VCMPHI, vcmphi)
1951 DO_VCMP(VCMPGE, vcmpge)
1952 DO_VCMP(VCMPLT, vcmplt)
1953 DO_VCMP(VCMPGT, vcmpgt)
1954 DO_VCMP(VCMPLE, vcmple)
1955 
1956 #define DO_VCMP_FP(INSN, FN)                                    \
1957     static bool trans_##INSN(DisasContext *s, arg_vcmp *a)      \
1958     {                                                           \
1959         static MVEGenCmpFn * const fns[] = {                    \
1960             NULL,                                               \
1961             gen_helper_mve_##FN##h,                             \
1962             gen_helper_mve_##FN##s,                             \
1963             NULL,                                               \
1964         };                                                      \
1965         if (!dc_isar_feature(aa32_mve_fp, s)) {                 \
1966             return false;                                       \
1967         }                                                       \
1968         return do_vcmp(s, a, fns[a->size]);                     \
1969     }                                                           \
1970     static bool trans_##INSN##_scalar(DisasContext *s,          \
1971                                       arg_vcmp_scalar *a)       \
1972     {                                                           \
1973         static MVEGenScalarCmpFn * const fns[] = {              \
1974             NULL,                                               \
1975             gen_helper_mve_##FN##_scalarh,                      \
1976             gen_helper_mve_##FN##_scalars,                      \
1977             NULL,                                               \
1978         };                                                      \
1979         if (!dc_isar_feature(aa32_mve_fp, s)) {                 \
1980             return false;                                       \
1981         }                                                       \
1982         return do_vcmp_scalar(s, a, fns[a->size]);              \
1983     }
1984 
1985 DO_VCMP_FP(VCMPEQ_fp, vfcmpeq)
1986 DO_VCMP_FP(VCMPNE_fp, vfcmpne)
1987 DO_VCMP_FP(VCMPGE_fp, vfcmpge)
1988 DO_VCMP_FP(VCMPLT_fp, vfcmplt)
1989 DO_VCMP_FP(VCMPGT_fp, vfcmpgt)
1990 DO_VCMP_FP(VCMPLE_fp, vfcmple)
1991 
1992 static bool do_vmaxv(DisasContext *s, arg_vmaxv *a, MVEGenVADDVFn fn)
1993 {
1994     /*
1995      * MIN/MAX operations across a vector: compute the min or
1996      * max of the initial value in a general purpose register
1997      * and all the elements in the vector, and store it back
1998      * into the general purpose register.
1999      */
2000     TCGv_ptr qm;
2001     TCGv_i32 rda;
2002 
2003     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qm) ||
2004         !fn || a->rda == 13 || a->rda == 15) {
2005         /* Rda cases are UNPREDICTABLE */
2006         return false;
2007     }
2008     if (!mve_eci_check(s) || !vfp_access_check(s)) {
2009         return true;
2010     }
2011 
2012     qm = mve_qreg_ptr(a->qm);
2013     rda = load_reg(s, a->rda);
2014     fn(rda, cpu_env, qm, rda);
2015     store_reg(s, a->rda, rda);
2016     tcg_temp_free_ptr(qm);
2017     mve_update_eci(s);
2018     return true;
2019 }
2020 
2021 #define DO_VMAXV(INSN, FN)                                      \
2022     static bool trans_##INSN(DisasContext *s, arg_vmaxv *a)     \
2023     {                                                           \
2024         static MVEGenVADDVFn * const fns[] = {                  \
2025             gen_helper_mve_##FN##b,                             \
2026             gen_helper_mve_##FN##h,                             \
2027             gen_helper_mve_##FN##w,                             \
2028             NULL,                                               \
2029         };                                                      \
2030         return do_vmaxv(s, a, fns[a->size]);                    \
2031     }
2032 
2033 DO_VMAXV(VMAXV_S, vmaxvs)
2034 DO_VMAXV(VMAXV_U, vmaxvu)
2035 DO_VMAXV(VMAXAV, vmaxav)
2036 DO_VMAXV(VMINV_S, vminvs)
2037 DO_VMAXV(VMINV_U, vminvu)
2038 DO_VMAXV(VMINAV, vminav)
2039 
2040 #define DO_VMAXV_FP(INSN, FN)                                   \
2041     static bool trans_##INSN(DisasContext *s, arg_vmaxv *a)     \
2042     {                                                           \
2043         static MVEGenVADDVFn * const fns[] = {                  \
2044             NULL,                                               \
2045             gen_helper_mve_##FN##h,                             \
2046             gen_helper_mve_##FN##s,                             \
2047             NULL,                                               \
2048         };                                                      \
2049         if (!dc_isar_feature(aa32_mve_fp, s)) {                 \
2050             return false;                                       \
2051         }                                                       \
2052         return do_vmaxv(s, a, fns[a->size]);                    \
2053     }
2054 
2055 DO_VMAXV_FP(VMAXNMV, vmaxnmv)
2056 DO_VMAXV_FP(VMINNMV, vminnmv)
2057 DO_VMAXV_FP(VMAXNMAV, vmaxnmav)
2058 DO_VMAXV_FP(VMINNMAV, vminnmav)
2059 
2060 static bool do_vabav(DisasContext *s, arg_vabav *a, MVEGenVABAVFn *fn)
2061 {
2062     /* Absolute difference accumulated across vector */
2063     TCGv_ptr qn, qm;
2064     TCGv_i32 rda;
2065 
2066     if (!dc_isar_feature(aa32_mve, s) ||
2067         !mve_check_qreg_bank(s, a->qm | a->qn) ||
2068         !fn || a->rda == 13 || a->rda == 15) {
2069         /* Rda cases are UNPREDICTABLE */
2070         return false;
2071     }
2072     if (!mve_eci_check(s) || !vfp_access_check(s)) {
2073         return true;
2074     }
2075 
2076     qm = mve_qreg_ptr(a->qm);
2077     qn = mve_qreg_ptr(a->qn);
2078     rda = load_reg(s, a->rda);
2079     fn(rda, cpu_env, qn, qm, rda);
2080     store_reg(s, a->rda, rda);
2081     tcg_temp_free_ptr(qm);
2082     tcg_temp_free_ptr(qn);
2083     mve_update_eci(s);
2084     return true;
2085 }
2086 
2087 #define DO_VABAV(INSN, FN)                                      \
2088     static bool trans_##INSN(DisasContext *s, arg_vabav *a)     \
2089     {                                                           \
2090         static MVEGenVABAVFn * const fns[] = {                  \
2091             gen_helper_mve_##FN##b,                             \
2092             gen_helper_mve_##FN##h,                             \
2093             gen_helper_mve_##FN##w,                             \
2094             NULL,                                               \
2095         };                                                      \
2096         return do_vabav(s, a, fns[a->size]);                    \
2097     }
2098 
2099 DO_VABAV(VABAV_S, vabavs)
2100 DO_VABAV(VABAV_U, vabavu)
2101 
2102 static bool trans_VMOV_to_2gp(DisasContext *s, arg_VMOV_to_2gp *a)
2103 {
2104     /*
2105      * VMOV two 32-bit vector lanes to two general-purpose registers.
2106      * This insn is not predicated but it is subject to beat-wise
2107      * execution if it is not in an IT block. For us this means
2108      * only that if PSR.ECI says we should not be executing the beat
2109      * corresponding to the lane of the vector register being accessed
2110      * then we should skip perfoming the move, and that we need to do
2111      * the usual check for bad ECI state and advance of ECI state.
2112      * (If PSR.ECI is non-zero then we cannot be in an IT block.)
2113      */
2114     TCGv_i32 tmp;
2115     int vd;
2116 
2117     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd) ||
2118         a->rt == 13 || a->rt == 15 || a->rt2 == 13 || a->rt2 == 15 ||
2119         a->rt == a->rt2) {
2120         /* Rt/Rt2 cases are UNPREDICTABLE */
2121         return false;
2122     }
2123     if (!mve_eci_check(s) || !vfp_access_check(s)) {
2124         return true;
2125     }
2126 
2127     /* Convert Qreg index to Dreg for read_neon_element32() etc */
2128     vd = a->qd * 2;
2129 
2130     if (!mve_skip_vmov(s, vd, a->idx, MO_32)) {
2131         tmp = tcg_temp_new_i32();
2132         read_neon_element32(tmp, vd, a->idx, MO_32);
2133         store_reg(s, a->rt, tmp);
2134     }
2135     if (!mve_skip_vmov(s, vd + 1, a->idx, MO_32)) {
2136         tmp = tcg_temp_new_i32();
2137         read_neon_element32(tmp, vd + 1, a->idx, MO_32);
2138         store_reg(s, a->rt2, tmp);
2139     }
2140 
2141     mve_update_and_store_eci(s);
2142     return true;
2143 }
2144 
2145 static bool trans_VMOV_from_2gp(DisasContext *s, arg_VMOV_to_2gp *a)
2146 {
2147     /*
2148      * VMOV two general-purpose registers to two 32-bit vector lanes.
2149      * This insn is not predicated but it is subject to beat-wise
2150      * execution if it is not in an IT block. For us this means
2151      * only that if PSR.ECI says we should not be executing the beat
2152      * corresponding to the lane of the vector register being accessed
2153      * then we should skip perfoming the move, and that we need to do
2154      * the usual check for bad ECI state and advance of ECI state.
2155      * (If PSR.ECI is non-zero then we cannot be in an IT block.)
2156      */
2157     TCGv_i32 tmp;
2158     int vd;
2159 
2160     if (!dc_isar_feature(aa32_mve, s) || !mve_check_qreg_bank(s, a->qd) ||
2161         a->rt == 13 || a->rt == 15 || a->rt2 == 13 || a->rt2 == 15) {
2162         /* Rt/Rt2 cases are UNPREDICTABLE */
2163         return false;
2164     }
2165     if (!mve_eci_check(s) || !vfp_access_check(s)) {
2166         return true;
2167     }
2168 
2169     /* Convert Qreg idx to Dreg for read_neon_element32() etc */
2170     vd = a->qd * 2;
2171 
2172     if (!mve_skip_vmov(s, vd, a->idx, MO_32)) {
2173         tmp = load_reg(s, a->rt);
2174         write_neon_element32(tmp, vd, a->idx, MO_32);
2175         tcg_temp_free_i32(tmp);
2176     }
2177     if (!mve_skip_vmov(s, vd + 1, a->idx, MO_32)) {
2178         tmp = load_reg(s, a->rt2);
2179         write_neon_element32(tmp, vd + 1, a->idx, MO_32);
2180         tcg_temp_free_i32(tmp);
2181     }
2182 
2183     mve_update_and_store_eci(s);
2184     return true;
2185 }
2186