xref: /openbmc/qemu/target/arm/tcg/translate-vfp.c (revision 29a01a90)
1 /*
2  *  ARM translation: AArch32 VFP instructions
3  *
4  *  Copyright (c) 2003 Fabrice Bellard
5  *  Copyright (c) 2005-2007 CodeSourcery
6  *  Copyright (c) 2007 OpenedHand, Ltd.
7  *  Copyright (c) 2019 Linaro, Ltd.
8  *
9  * This library is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU Lesser General Public
11  * License as published by the Free Software Foundation; either
12  * version 2.1 of the License, or (at your option) any later version.
13  *
14  * This library is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * Lesser General Public License for more details.
18  *
19  * You should have received a copy of the GNU Lesser General Public
20  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
21  */
22 
23 #include "qemu/osdep.h"
24 #include "tcg/tcg-op.h"
25 #include "tcg/tcg-op-gvec.h"
26 #include "exec/exec-all.h"
27 #include "exec/gen-icount.h"
28 #include "translate.h"
29 #include "translate-a32.h"
30 
31 /* Include the generated VFP decoder */
32 #include "decode-vfp.c.inc"
33 #include "decode-vfp-uncond.c.inc"
34 
35 static inline void vfp_load_reg64(TCGv_i64 var, int reg)
36 {
37     tcg_gen_ld_i64(var, cpu_env, vfp_reg_offset(true, reg));
38 }
39 
40 static inline void vfp_store_reg64(TCGv_i64 var, int reg)
41 {
42     tcg_gen_st_i64(var, cpu_env, vfp_reg_offset(true, reg));
43 }
44 
45 static inline void vfp_load_reg32(TCGv_i32 var, int reg)
46 {
47     tcg_gen_ld_i32(var, cpu_env, vfp_reg_offset(false, reg));
48 }
49 
50 static inline void vfp_store_reg32(TCGv_i32 var, int reg)
51 {
52     tcg_gen_st_i32(var, cpu_env, vfp_reg_offset(false, reg));
53 }
54 
55 /*
56  * The imm8 encodes the sign bit, enough bits to represent an exponent in
57  * the range 01....1xx to 10....0xx, and the most significant 4 bits of
58  * the mantissa; see VFPExpandImm() in the v8 ARM ARM.
59  */
60 uint64_t vfp_expand_imm(int size, uint8_t imm8)
61 {
62     uint64_t imm;
63 
64     switch (size) {
65     case MO_64:
66         imm = (extract32(imm8, 7, 1) ? 0x8000 : 0) |
67             (extract32(imm8, 6, 1) ? 0x3fc0 : 0x4000) |
68             extract32(imm8, 0, 6);
69         imm <<= 48;
70         break;
71     case MO_32:
72         imm = (extract32(imm8, 7, 1) ? 0x8000 : 0) |
73             (extract32(imm8, 6, 1) ? 0x3e00 : 0x4000) |
74             (extract32(imm8, 0, 6) << 3);
75         imm <<= 16;
76         break;
77     case MO_16:
78         imm = (extract32(imm8, 7, 1) ? 0x8000 : 0) |
79             (extract32(imm8, 6, 1) ? 0x3000 : 0x4000) |
80             (extract32(imm8, 0, 6) << 6);
81         break;
82     default:
83         g_assert_not_reached();
84     }
85     return imm;
86 }
87 
88 /*
89  * Return the offset of a 16-bit half of the specified VFP single-precision
90  * register. If top is true, returns the top 16 bits; otherwise the bottom
91  * 16 bits.
92  */
93 static inline long vfp_f16_offset(unsigned reg, bool top)
94 {
95     long offs = vfp_reg_offset(false, reg);
96 #if HOST_BIG_ENDIAN
97     if (!top) {
98         offs += 2;
99     }
100 #else
101     if (top) {
102         offs += 2;
103     }
104 #endif
105     return offs;
106 }
107 
108 /*
109  * Generate code for M-profile lazy FP state preservation if needed;
110  * this corresponds to the pseudocode PreserveFPState() function.
111  */
112 static void gen_preserve_fp_state(DisasContext *s, bool skip_context_update)
113 {
114     if (s->v7m_lspact) {
115         /*
116          * Lazy state saving affects external memory and also the NVIC,
117          * so we must mark it as an IO operation for icount (and cause
118          * this to be the last insn in the TB).
119          */
120         if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
121             s->base.is_jmp = DISAS_UPDATE_EXIT;
122             gen_io_start();
123         }
124         gen_helper_v7m_preserve_fp_state(cpu_env);
125         /*
126          * If the preserve_fp_state helper doesn't throw an exception
127          * then it will clear LSPACT; we don't need to repeat this for
128          * any further FP insns in this TB.
129          */
130         s->v7m_lspact = false;
131         /*
132          * The helper might have zeroed VPR, so we do not know the
133          * correct value for the MVE_NO_PRED TB flag any more.
134          * If we're about to create a new fp context then that
135          * will precisely determine the MVE_NO_PRED value (see
136          * gen_update_fp_context()). Otherwise, we must:
137          *  - set s->mve_no_pred to false, so this instruction
138          *    is generated to use helper functions
139          *  - end the TB now, without chaining to the next TB
140          */
141         if (skip_context_update || !s->v7m_new_fp_ctxt_needed) {
142             s->mve_no_pred = false;
143             s->base.is_jmp = DISAS_UPDATE_NOCHAIN;
144         }
145     }
146 }
147 
148 /*
149  * Generate code for M-profile FP context handling: update the
150  * ownership of the FP context, and create a new context if
151  * necessary. This corresponds to the parts of the pseudocode
152  * ExecuteFPCheck() after the inital PreserveFPState() call.
153  */
154 static void gen_update_fp_context(DisasContext *s)
155 {
156     /* Update ownership of FP context: set FPCCR.S to match current state */
157     if (s->v8m_fpccr_s_wrong) {
158         TCGv_i32 tmp;
159 
160         tmp = load_cpu_field(v7m.fpccr[M_REG_S]);
161         if (s->v8m_secure) {
162             tcg_gen_ori_i32(tmp, tmp, R_V7M_FPCCR_S_MASK);
163         } else {
164             tcg_gen_andi_i32(tmp, tmp, ~R_V7M_FPCCR_S_MASK);
165         }
166         store_cpu_field(tmp, v7m.fpccr[M_REG_S]);
167         /* Don't need to do this for any further FP insns in this TB */
168         s->v8m_fpccr_s_wrong = false;
169     }
170 
171     if (s->v7m_new_fp_ctxt_needed) {
172         /*
173          * Create new FP context by updating CONTROL.FPCA, CONTROL.SFPA,
174          * the FPSCR, and VPR.
175          */
176         TCGv_i32 control, fpscr;
177         uint32_t bits = R_V7M_CONTROL_FPCA_MASK;
178 
179         fpscr = load_cpu_field(v7m.fpdscr[s->v8m_secure]);
180         gen_helper_vfp_set_fpscr(cpu_env, fpscr);
181         if (dc_isar_feature(aa32_mve, s)) {
182             store_cpu_field(tcg_constant_i32(0), v7m.vpr);
183         }
184         /*
185          * We just updated the FPSCR and VPR. Some of this state is cached
186          * in the MVE_NO_PRED TB flag. We want to avoid having to end the
187          * TB here, which means we need the new value of the MVE_NO_PRED
188          * flag to be exactly known here and the same for all executions.
189          * Luckily FPDSCR.LTPSIZE is always constant 4 and the VPR is
190          * always set to 0, so the new MVE_NO_PRED flag is always 1
191          * if and only if we have MVE.
192          *
193          * (The other FPSCR state cached in TB flags is VECLEN and VECSTRIDE,
194          * but those do not exist for M-profile, so are not relevant here.)
195          */
196         s->mve_no_pred = dc_isar_feature(aa32_mve, s);
197 
198         if (s->v8m_secure) {
199             bits |= R_V7M_CONTROL_SFPA_MASK;
200         }
201         control = load_cpu_field(v7m.control[M_REG_S]);
202         tcg_gen_ori_i32(control, control, bits);
203         store_cpu_field(control, v7m.control[M_REG_S]);
204         /* Don't need to do this for any further FP insns in this TB */
205         s->v7m_new_fp_ctxt_needed = false;
206     }
207 }
208 
209 /*
210  * Check that VFP access is enabled, A-profile specific version.
211  *
212  * If VFP is enabled, return true. If not, emit code to generate an
213  * appropriate exception and return false.
214  * The ignore_vfp_enabled argument specifies that we should ignore
215  * whether VFP is enabled via FPEXC.EN: this should be true for FMXR/FMRX
216  * accesses to FPSID, FPEXC, MVFR0, MVFR1, MVFR2, and false for all other insns.
217  */
218 static bool vfp_access_check_a(DisasContext *s, bool ignore_vfp_enabled)
219 {
220     if (s->fp_excp_el) {
221         /*
222          * The full syndrome is only used for HSR when HCPTR traps:
223          * For v8, when TA==0, coproc is RES0.
224          * For v7, any use of a Floating-point instruction or access
225          * to a Floating-point Extension register that is trapped to
226          * Hyp mode because of a trap configured in the HCPTR sets
227          * this field to 0xA.
228          */
229         int coproc = arm_dc_feature(s, ARM_FEATURE_V8) ? 0 : 0xa;
230         uint32_t syn = syn_fp_access_trap(1, 0xe, false, coproc);
231 
232         gen_exception_insn_el(s, 0, EXCP_UDEF, syn, s->fp_excp_el);
233         return false;
234     }
235 
236     /*
237      * Note that rebuild_hflags_a32 has already accounted for being in EL0
238      * and the higher EL in A64 mode, etc.  Unlike A64 mode, there do not
239      * appear to be any insns which touch VFP which are allowed.
240      */
241     if (s->sme_trap_nonstreaming) {
242         gen_exception_insn(s, 0, EXCP_UDEF,
243                            syn_smetrap(SME_ET_Streaming,
244                                        curr_insn_len(s) == 2));
245         return false;
246     }
247 
248     if (!s->vfp_enabled && !ignore_vfp_enabled) {
249         assert(!arm_dc_feature(s, ARM_FEATURE_M));
250         unallocated_encoding(s);
251         return false;
252     }
253     return true;
254 }
255 
256 /*
257  * Check that VFP access is enabled, M-profile specific version.
258  *
259  * If VFP is enabled, do the necessary M-profile lazy-FP handling and then
260  * return true. If not, emit code to generate an appropriate exception and
261  * return false.
262  * skip_context_update is true to skip the "update FP context" part of this.
263  */
264 bool vfp_access_check_m(DisasContext *s, bool skip_context_update)
265 {
266     if (s->fp_excp_el) {
267         /*
268          * M-profile mostly catches the "FPU disabled" case early, in
269          * disas_m_nocp(), but a few insns (eg LCTP, WLSTP, DLSTP)
270          * which do coprocessor-checks are outside the large ranges of
271          * the encoding space handled by the patterns in m-nocp.decode,
272          * and for them we may need to raise NOCP here.
273          */
274         gen_exception_insn_el(s, 0, EXCP_NOCP,
275                               syn_uncategorized(), s->fp_excp_el);
276         return false;
277     }
278 
279     /* Handle M-profile lazy FP state mechanics */
280 
281     /* Trigger lazy-state preservation if necessary */
282     gen_preserve_fp_state(s, skip_context_update);
283 
284     if (!skip_context_update) {
285         /* Update ownership of FP context and create new FP context if needed */
286         gen_update_fp_context(s);
287     }
288 
289     return true;
290 }
291 
292 /*
293  * The most usual kind of VFP access check, for everything except
294  * FMXR/FMRX to the always-available special registers.
295  */
296 bool vfp_access_check(DisasContext *s)
297 {
298     if (arm_dc_feature(s, ARM_FEATURE_M)) {
299         return vfp_access_check_m(s, false);
300     } else {
301         return vfp_access_check_a(s, false);
302     }
303 }
304 
305 static bool trans_VSEL(DisasContext *s, arg_VSEL *a)
306 {
307     uint32_t rd, rn, rm;
308     int sz = a->sz;
309 
310     if (!dc_isar_feature(aa32_vsel, s)) {
311         return false;
312     }
313 
314     if (sz == 3 && !dc_isar_feature(aa32_fpdp_v2, s)) {
315         return false;
316     }
317 
318     if (sz == 1 && !dc_isar_feature(aa32_fp16_arith, s)) {
319         return false;
320     }
321 
322     /* UNDEF accesses to D16-D31 if they don't exist */
323     if (sz == 3 && !dc_isar_feature(aa32_simd_r32, s) &&
324         ((a->vm | a->vn | a->vd) & 0x10)) {
325         return false;
326     }
327 
328     rd = a->vd;
329     rn = a->vn;
330     rm = a->vm;
331 
332     if (!vfp_access_check(s)) {
333         return true;
334     }
335 
336     if (sz == 3) {
337         TCGv_i64 frn, frm, dest;
338         TCGv_i64 tmp, zero, zf, nf, vf;
339 
340         zero = tcg_constant_i64(0);
341 
342         frn = tcg_temp_new_i64();
343         frm = tcg_temp_new_i64();
344         dest = tcg_temp_new_i64();
345 
346         zf = tcg_temp_new_i64();
347         nf = tcg_temp_new_i64();
348         vf = tcg_temp_new_i64();
349 
350         tcg_gen_extu_i32_i64(zf, cpu_ZF);
351         tcg_gen_ext_i32_i64(nf, cpu_NF);
352         tcg_gen_ext_i32_i64(vf, cpu_VF);
353 
354         vfp_load_reg64(frn, rn);
355         vfp_load_reg64(frm, rm);
356         switch (a->cc) {
357         case 0: /* eq: Z */
358             tcg_gen_movcond_i64(TCG_COND_EQ, dest, zf, zero, frn, frm);
359             break;
360         case 1: /* vs: V */
361             tcg_gen_movcond_i64(TCG_COND_LT, dest, vf, zero, frn, frm);
362             break;
363         case 2: /* ge: N == V -> N ^ V == 0 */
364             tmp = tcg_temp_new_i64();
365             tcg_gen_xor_i64(tmp, vf, nf);
366             tcg_gen_movcond_i64(TCG_COND_GE, dest, tmp, zero, frn, frm);
367             break;
368         case 3: /* gt: !Z && N == V */
369             tcg_gen_movcond_i64(TCG_COND_NE, dest, zf, zero, frn, frm);
370             tmp = tcg_temp_new_i64();
371             tcg_gen_xor_i64(tmp, vf, nf);
372             tcg_gen_movcond_i64(TCG_COND_GE, dest, tmp, zero, dest, frm);
373             break;
374         }
375         vfp_store_reg64(dest, rd);
376     } else {
377         TCGv_i32 frn, frm, dest;
378         TCGv_i32 tmp, zero;
379 
380         zero = tcg_constant_i32(0);
381 
382         frn = tcg_temp_new_i32();
383         frm = tcg_temp_new_i32();
384         dest = tcg_temp_new_i32();
385         vfp_load_reg32(frn, rn);
386         vfp_load_reg32(frm, rm);
387         switch (a->cc) {
388         case 0: /* eq: Z */
389             tcg_gen_movcond_i32(TCG_COND_EQ, dest, cpu_ZF, zero, frn, frm);
390             break;
391         case 1: /* vs: V */
392             tcg_gen_movcond_i32(TCG_COND_LT, dest, cpu_VF, zero, frn, frm);
393             break;
394         case 2: /* ge: N == V -> N ^ V == 0 */
395             tmp = tcg_temp_new_i32();
396             tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
397             tcg_gen_movcond_i32(TCG_COND_GE, dest, tmp, zero, frn, frm);
398             break;
399         case 3: /* gt: !Z && N == V */
400             tcg_gen_movcond_i32(TCG_COND_NE, dest, cpu_ZF, zero, frn, frm);
401             tmp = tcg_temp_new_i32();
402             tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
403             tcg_gen_movcond_i32(TCG_COND_GE, dest, tmp, zero, dest, frm);
404             break;
405         }
406         /* For fp16 the top half is always zeroes */
407         if (sz == 1) {
408             tcg_gen_andi_i32(dest, dest, 0xffff);
409         }
410         vfp_store_reg32(dest, rd);
411     }
412 
413     return true;
414 }
415 
416 /*
417  * Table for converting the most common AArch32 encoding of
418  * rounding mode to arm_fprounding order (which matches the
419  * common AArch64 order); see ARM ARM pseudocode FPDecodeRM().
420  */
421 static const uint8_t fp_decode_rm[] = {
422     FPROUNDING_TIEAWAY,
423     FPROUNDING_TIEEVEN,
424     FPROUNDING_POSINF,
425     FPROUNDING_NEGINF,
426 };
427 
428 static bool trans_VRINT(DisasContext *s, arg_VRINT *a)
429 {
430     uint32_t rd, rm;
431     int sz = a->sz;
432     TCGv_ptr fpst;
433     TCGv_i32 tcg_rmode;
434     int rounding = fp_decode_rm[a->rm];
435 
436     if (!dc_isar_feature(aa32_vrint, s)) {
437         return false;
438     }
439 
440     if (sz == 3 && !dc_isar_feature(aa32_fpdp_v2, s)) {
441         return false;
442     }
443 
444     if (sz == 1 && !dc_isar_feature(aa32_fp16_arith, s)) {
445         return false;
446     }
447 
448     /* UNDEF accesses to D16-D31 if they don't exist */
449     if (sz == 3 && !dc_isar_feature(aa32_simd_r32, s) &&
450         ((a->vm | a->vd) & 0x10)) {
451         return false;
452     }
453 
454     rd = a->vd;
455     rm = a->vm;
456 
457     if (!vfp_access_check(s)) {
458         return true;
459     }
460 
461     if (sz == 1) {
462         fpst = fpstatus_ptr(FPST_FPCR_F16);
463     } else {
464         fpst = fpstatus_ptr(FPST_FPCR);
465     }
466 
467     tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rounding));
468     gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
469 
470     if (sz == 3) {
471         TCGv_i64 tcg_op;
472         TCGv_i64 tcg_res;
473         tcg_op = tcg_temp_new_i64();
474         tcg_res = tcg_temp_new_i64();
475         vfp_load_reg64(tcg_op, rm);
476         gen_helper_rintd(tcg_res, tcg_op, fpst);
477         vfp_store_reg64(tcg_res, rd);
478     } else {
479         TCGv_i32 tcg_op;
480         TCGv_i32 tcg_res;
481         tcg_op = tcg_temp_new_i32();
482         tcg_res = tcg_temp_new_i32();
483         vfp_load_reg32(tcg_op, rm);
484         if (sz == 1) {
485             gen_helper_rinth(tcg_res, tcg_op, fpst);
486         } else {
487             gen_helper_rints(tcg_res, tcg_op, fpst);
488         }
489         vfp_store_reg32(tcg_res, rd);
490     }
491 
492     gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
493     return true;
494 }
495 
496 static bool trans_VCVT(DisasContext *s, arg_VCVT *a)
497 {
498     uint32_t rd, rm;
499     int sz = a->sz;
500     TCGv_ptr fpst;
501     TCGv_i32 tcg_rmode, tcg_shift;
502     int rounding = fp_decode_rm[a->rm];
503     bool is_signed = a->op;
504 
505     if (!dc_isar_feature(aa32_vcvt_dr, s)) {
506         return false;
507     }
508 
509     if (sz == 3 && !dc_isar_feature(aa32_fpdp_v2, s)) {
510         return false;
511     }
512 
513     if (sz == 1 && !dc_isar_feature(aa32_fp16_arith, s)) {
514         return false;
515     }
516 
517     /* UNDEF accesses to D16-D31 if they don't exist */
518     if (sz == 3 && !dc_isar_feature(aa32_simd_r32, s) && (a->vm & 0x10)) {
519         return false;
520     }
521 
522     rd = a->vd;
523     rm = a->vm;
524 
525     if (!vfp_access_check(s)) {
526         return true;
527     }
528 
529     if (sz == 1) {
530         fpst = fpstatus_ptr(FPST_FPCR_F16);
531     } else {
532         fpst = fpstatus_ptr(FPST_FPCR);
533     }
534 
535     tcg_shift = tcg_constant_i32(0);
536 
537     tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rounding));
538     gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
539 
540     if (sz == 3) {
541         TCGv_i64 tcg_double, tcg_res;
542         TCGv_i32 tcg_tmp;
543         tcg_double = tcg_temp_new_i64();
544         tcg_res = tcg_temp_new_i64();
545         tcg_tmp = tcg_temp_new_i32();
546         vfp_load_reg64(tcg_double, rm);
547         if (is_signed) {
548             gen_helper_vfp_tosld(tcg_res, tcg_double, tcg_shift, fpst);
549         } else {
550             gen_helper_vfp_tould(tcg_res, tcg_double, tcg_shift, fpst);
551         }
552         tcg_gen_extrl_i64_i32(tcg_tmp, tcg_res);
553         vfp_store_reg32(tcg_tmp, rd);
554     } else {
555         TCGv_i32 tcg_single, tcg_res;
556         tcg_single = tcg_temp_new_i32();
557         tcg_res = tcg_temp_new_i32();
558         vfp_load_reg32(tcg_single, rm);
559         if (sz == 1) {
560             if (is_signed) {
561                 gen_helper_vfp_toslh(tcg_res, tcg_single, tcg_shift, fpst);
562             } else {
563                 gen_helper_vfp_toulh(tcg_res, tcg_single, tcg_shift, fpst);
564             }
565         } else {
566             if (is_signed) {
567                 gen_helper_vfp_tosls(tcg_res, tcg_single, tcg_shift, fpst);
568             } else {
569                 gen_helper_vfp_touls(tcg_res, tcg_single, tcg_shift, fpst);
570             }
571         }
572         vfp_store_reg32(tcg_res, rd);
573     }
574 
575     gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
576     return true;
577 }
578 
579 bool mve_skip_vmov(DisasContext *s, int vn, int index, int size)
580 {
581     /*
582      * In a CPU with MVE, the VMOV (vector lane to general-purpose register)
583      * and VMOV (general-purpose register to vector lane) insns are not
584      * predicated, but they are subject to beatwise execution if they are
585      * not in an IT block.
586      *
587      * Since our implementation always executes all 4 beats in one tick,
588      * this means only that if PSR.ECI says we should not be executing
589      * the beat corresponding to the lane of the vector register being
590      * accessed then we should skip performing the move, and that we need
591      * to do the usual check for bad ECI state and advance of ECI state.
592      *
593      * Note that if PSR.ECI is non-zero then we cannot be in an IT block.
594      *
595      * Return true if this VMOV scalar <-> gpreg should be skipped because
596      * the MVE PSR.ECI state says we skip the beat where the store happens.
597      */
598 
599     /* Calculate the byte offset into Qn which we're going to access */
600     int ofs = (index << size) + ((vn & 1) * 8);
601 
602     if (!dc_isar_feature(aa32_mve, s)) {
603         return false;
604     }
605 
606     switch (s->eci) {
607     case ECI_NONE:
608         return false;
609     case ECI_A0:
610         return ofs < 4;
611     case ECI_A0A1:
612         return ofs < 8;
613     case ECI_A0A1A2:
614     case ECI_A0A1A2B0:
615         return ofs < 12;
616     default:
617         g_assert_not_reached();
618     }
619 }
620 
621 static bool trans_VMOV_to_gp(DisasContext *s, arg_VMOV_to_gp *a)
622 {
623     /* VMOV scalar to general purpose register */
624     TCGv_i32 tmp;
625 
626     /*
627      * SIZE == MO_32 is a VFP instruction; otherwise NEON. MVE has
628      * all sizes, whether the CPU has fp or not.
629      */
630     if (!dc_isar_feature(aa32_mve, s)) {
631         if (a->size == MO_32
632             ? !dc_isar_feature(aa32_fpsp_v2, s)
633             : !arm_dc_feature(s, ARM_FEATURE_NEON)) {
634             return false;
635         }
636     }
637 
638     /* UNDEF accesses to D16-D31 if they don't exist */
639     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vn & 0x10)) {
640         return false;
641     }
642 
643     if (dc_isar_feature(aa32_mve, s)) {
644         if (!mve_eci_check(s)) {
645             return true;
646         }
647     }
648 
649     if (!vfp_access_check(s)) {
650         return true;
651     }
652 
653     if (!mve_skip_vmov(s, a->vn, a->index, a->size)) {
654         tmp = tcg_temp_new_i32();
655         read_neon_element32(tmp, a->vn, a->index,
656                             a->size | (a->u ? 0 : MO_SIGN));
657         store_reg(s, a->rt, tmp);
658     }
659 
660     if (dc_isar_feature(aa32_mve, s)) {
661         mve_update_and_store_eci(s);
662     }
663     return true;
664 }
665 
666 static bool trans_VMOV_from_gp(DisasContext *s, arg_VMOV_from_gp *a)
667 {
668     /* VMOV general purpose register to scalar */
669     TCGv_i32 tmp;
670 
671     /*
672      * SIZE == MO_32 is a VFP instruction; otherwise NEON. MVE has
673      * all sizes, whether the CPU has fp or not.
674      */
675     if (!dc_isar_feature(aa32_mve, s)) {
676         if (a->size == MO_32
677             ? !dc_isar_feature(aa32_fpsp_v2, s)
678             : !arm_dc_feature(s, ARM_FEATURE_NEON)) {
679             return false;
680         }
681     }
682 
683     /* UNDEF accesses to D16-D31 if they don't exist */
684     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vn & 0x10)) {
685         return false;
686     }
687 
688     if (dc_isar_feature(aa32_mve, s)) {
689         if (!mve_eci_check(s)) {
690             return true;
691         }
692     }
693 
694     if (!vfp_access_check(s)) {
695         return true;
696     }
697 
698     if (!mve_skip_vmov(s, a->vn, a->index, a->size)) {
699         tmp = load_reg(s, a->rt);
700         write_neon_element32(tmp, a->vn, a->index, a->size);
701     }
702 
703     if (dc_isar_feature(aa32_mve, s)) {
704         mve_update_and_store_eci(s);
705     }
706     return true;
707 }
708 
709 static bool trans_VDUP(DisasContext *s, arg_VDUP *a)
710 {
711     /* VDUP (general purpose register) */
712     TCGv_i32 tmp;
713     int size, vec_size;
714 
715     if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
716         return false;
717     }
718 
719     /* UNDEF accesses to D16-D31 if they don't exist */
720     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vn & 0x10)) {
721         return false;
722     }
723 
724     if (a->b && a->e) {
725         return false;
726     }
727 
728     if (a->q && (a->vn & 1)) {
729         return false;
730     }
731 
732     vec_size = a->q ? 16 : 8;
733     if (a->b) {
734         size = 0;
735     } else if (a->e) {
736         size = 1;
737     } else {
738         size = 2;
739     }
740 
741     if (!vfp_access_check(s)) {
742         return true;
743     }
744 
745     tmp = load_reg(s, a->rt);
746     tcg_gen_gvec_dup_i32(size, neon_full_reg_offset(a->vn),
747                          vec_size, vec_size, tmp);
748     return true;
749 }
750 
751 static bool trans_VMSR_VMRS(DisasContext *s, arg_VMSR_VMRS *a)
752 {
753     TCGv_i32 tmp;
754     bool ignore_vfp_enabled = false;
755 
756     if (arm_dc_feature(s, ARM_FEATURE_M)) {
757         /* M profile version was already handled in m-nocp.decode */
758         return false;
759     }
760 
761     if (!dc_isar_feature(aa32_fpsp_v2, s)) {
762         return false;
763     }
764 
765     switch (a->reg) {
766     case ARM_VFP_FPSID:
767         /*
768          * VFPv2 allows access to FPSID from userspace; VFPv3 restricts
769          * all ID registers to privileged access only.
770          */
771         if (IS_USER(s) && dc_isar_feature(aa32_fpsp_v3, s)) {
772             return false;
773         }
774         ignore_vfp_enabled = true;
775         break;
776     case ARM_VFP_MVFR0:
777     case ARM_VFP_MVFR1:
778         if (IS_USER(s) || !arm_dc_feature(s, ARM_FEATURE_MVFR)) {
779             return false;
780         }
781         ignore_vfp_enabled = true;
782         break;
783     case ARM_VFP_MVFR2:
784         if (IS_USER(s) || !arm_dc_feature(s, ARM_FEATURE_V8)) {
785             return false;
786         }
787         ignore_vfp_enabled = true;
788         break;
789     case ARM_VFP_FPSCR:
790         break;
791     case ARM_VFP_FPEXC:
792         if (IS_USER(s)) {
793             return false;
794         }
795         ignore_vfp_enabled = true;
796         break;
797     case ARM_VFP_FPINST:
798     case ARM_VFP_FPINST2:
799         /* Not present in VFPv3 */
800         if (IS_USER(s) || dc_isar_feature(aa32_fpsp_v3, s)) {
801             return false;
802         }
803         break;
804     default:
805         return false;
806     }
807 
808     /*
809      * Call vfp_access_check_a() directly, because we need to tell
810      * it to ignore FPEXC.EN for some register accesses.
811      */
812     if (!vfp_access_check_a(s, ignore_vfp_enabled)) {
813         return true;
814     }
815 
816     if (a->l) {
817         /* VMRS, move VFP special register to gp register */
818         switch (a->reg) {
819         case ARM_VFP_MVFR0:
820         case ARM_VFP_MVFR1:
821         case ARM_VFP_MVFR2:
822         case ARM_VFP_FPSID:
823             if (s->current_el == 1) {
824                 gen_set_condexec(s);
825                 gen_update_pc(s, 0);
826                 gen_helper_check_hcr_el2_trap(cpu_env,
827                                               tcg_constant_i32(a->rt),
828                                               tcg_constant_i32(a->reg));
829             }
830             /* fall through */
831         case ARM_VFP_FPEXC:
832         case ARM_VFP_FPINST:
833         case ARM_VFP_FPINST2:
834             tmp = load_cpu_field(vfp.xregs[a->reg]);
835             break;
836         case ARM_VFP_FPSCR:
837             if (a->rt == 15) {
838                 tmp = load_cpu_field(vfp.xregs[ARM_VFP_FPSCR]);
839                 tcg_gen_andi_i32(tmp, tmp, FPCR_NZCV_MASK);
840             } else {
841                 tmp = tcg_temp_new_i32();
842                 gen_helper_vfp_get_fpscr(tmp, cpu_env);
843             }
844             break;
845         default:
846             g_assert_not_reached();
847         }
848 
849         if (a->rt == 15) {
850             /* Set the 4 flag bits in the CPSR.  */
851             gen_set_nzcv(tmp);
852         } else {
853             store_reg(s, a->rt, tmp);
854         }
855     } else {
856         /* VMSR, move gp register to VFP special register */
857         switch (a->reg) {
858         case ARM_VFP_FPSID:
859         case ARM_VFP_MVFR0:
860         case ARM_VFP_MVFR1:
861         case ARM_VFP_MVFR2:
862             /* Writes are ignored.  */
863             break;
864         case ARM_VFP_FPSCR:
865             tmp = load_reg(s, a->rt);
866             gen_helper_vfp_set_fpscr(cpu_env, tmp);
867             gen_lookup_tb(s);
868             break;
869         case ARM_VFP_FPEXC:
870             /*
871              * TODO: VFP subarchitecture support.
872              * For now, keep the EN bit only
873              */
874             tmp = load_reg(s, a->rt);
875             tcg_gen_andi_i32(tmp, tmp, 1 << 30);
876             store_cpu_field(tmp, vfp.xregs[a->reg]);
877             gen_lookup_tb(s);
878             break;
879         case ARM_VFP_FPINST:
880         case ARM_VFP_FPINST2:
881             tmp = load_reg(s, a->rt);
882             store_cpu_field(tmp, vfp.xregs[a->reg]);
883             break;
884         default:
885             g_assert_not_reached();
886         }
887     }
888 
889     return true;
890 }
891 
892 
893 static bool trans_VMOV_half(DisasContext *s, arg_VMOV_single *a)
894 {
895     TCGv_i32 tmp;
896 
897     if (!dc_isar_feature(aa32_fp16_arith, s)) {
898         return false;
899     }
900 
901     if (a->rt == 15) {
902         /* UNPREDICTABLE; we choose to UNDEF */
903         return false;
904     }
905 
906     if (!vfp_access_check(s)) {
907         return true;
908     }
909 
910     if (a->l) {
911         /* VFP to general purpose register */
912         tmp = tcg_temp_new_i32();
913         vfp_load_reg32(tmp, a->vn);
914         tcg_gen_andi_i32(tmp, tmp, 0xffff);
915         store_reg(s, a->rt, tmp);
916     } else {
917         /* general purpose register to VFP */
918         tmp = load_reg(s, a->rt);
919         tcg_gen_andi_i32(tmp, tmp, 0xffff);
920         vfp_store_reg32(tmp, a->vn);
921     }
922 
923     return true;
924 }
925 
926 static bool trans_VMOV_single(DisasContext *s, arg_VMOV_single *a)
927 {
928     TCGv_i32 tmp;
929 
930     if (!dc_isar_feature(aa32_fpsp_v2, s) && !dc_isar_feature(aa32_mve, s)) {
931         return false;
932     }
933 
934     if (!vfp_access_check(s)) {
935         return true;
936     }
937 
938     if (a->l) {
939         /* VFP to general purpose register */
940         tmp = tcg_temp_new_i32();
941         vfp_load_reg32(tmp, a->vn);
942         if (a->rt == 15) {
943             /* Set the 4 flag bits in the CPSR.  */
944             gen_set_nzcv(tmp);
945         } else {
946             store_reg(s, a->rt, tmp);
947         }
948     } else {
949         /* general purpose register to VFP */
950         tmp = load_reg(s, a->rt);
951         vfp_store_reg32(tmp, a->vn);
952     }
953 
954     return true;
955 }
956 
957 static bool trans_VMOV_64_sp(DisasContext *s, arg_VMOV_64_sp *a)
958 {
959     TCGv_i32 tmp;
960 
961     if (!dc_isar_feature(aa32_fpsp_v2, s) && !dc_isar_feature(aa32_mve, s)) {
962         return false;
963     }
964 
965     /*
966      * VMOV between two general-purpose registers and two single precision
967      * floating point registers
968      */
969     if (!vfp_access_check(s)) {
970         return true;
971     }
972 
973     if (a->op) {
974         /* fpreg to gpreg */
975         tmp = tcg_temp_new_i32();
976         vfp_load_reg32(tmp, a->vm);
977         store_reg(s, a->rt, tmp);
978         tmp = tcg_temp_new_i32();
979         vfp_load_reg32(tmp, a->vm + 1);
980         store_reg(s, a->rt2, tmp);
981     } else {
982         /* gpreg to fpreg */
983         tmp = load_reg(s, a->rt);
984         vfp_store_reg32(tmp, a->vm);
985         tmp = load_reg(s, a->rt2);
986         vfp_store_reg32(tmp, a->vm + 1);
987     }
988 
989     return true;
990 }
991 
992 static bool trans_VMOV_64_dp(DisasContext *s, arg_VMOV_64_dp *a)
993 {
994     TCGv_i32 tmp;
995 
996     /*
997      * VMOV between two general-purpose registers and one double precision
998      * floating point register.  Note that this does not require support
999      * for double precision arithmetic.
1000      */
1001     if (!dc_isar_feature(aa32_fpsp_v2, s) && !dc_isar_feature(aa32_mve, s)) {
1002         return false;
1003     }
1004 
1005     /* UNDEF accesses to D16-D31 if they don't exist */
1006     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vm & 0x10)) {
1007         return false;
1008     }
1009 
1010     if (!vfp_access_check(s)) {
1011         return true;
1012     }
1013 
1014     if (a->op) {
1015         /* fpreg to gpreg */
1016         tmp = tcg_temp_new_i32();
1017         vfp_load_reg32(tmp, a->vm * 2);
1018         store_reg(s, a->rt, tmp);
1019         tmp = tcg_temp_new_i32();
1020         vfp_load_reg32(tmp, a->vm * 2 + 1);
1021         store_reg(s, a->rt2, tmp);
1022     } else {
1023         /* gpreg to fpreg */
1024         tmp = load_reg(s, a->rt);
1025         vfp_store_reg32(tmp, a->vm * 2);
1026         tmp = load_reg(s, a->rt2);
1027         vfp_store_reg32(tmp, a->vm * 2 + 1);
1028     }
1029 
1030     return true;
1031 }
1032 
1033 static bool trans_VLDR_VSTR_hp(DisasContext *s, arg_VLDR_VSTR_sp *a)
1034 {
1035     uint32_t offset;
1036     TCGv_i32 addr, tmp;
1037 
1038     if (!dc_isar_feature(aa32_fpsp_v2, s) && !dc_isar_feature(aa32_mve, s)) {
1039         return false;
1040     }
1041 
1042     if (!vfp_access_check(s)) {
1043         return true;
1044     }
1045 
1046     /* imm8 field is offset/2 for fp16, unlike fp32 and fp64 */
1047     offset = a->imm << 1;
1048     if (!a->u) {
1049         offset = -offset;
1050     }
1051 
1052     /* For thumb, use of PC is UNPREDICTABLE.  */
1053     addr = add_reg_for_lit(s, a->rn, offset);
1054     tmp = tcg_temp_new_i32();
1055     if (a->l) {
1056         gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), MO_UW | MO_ALIGN);
1057         vfp_store_reg32(tmp, a->vd);
1058     } else {
1059         vfp_load_reg32(tmp, a->vd);
1060         gen_aa32_st_i32(s, tmp, addr, get_mem_index(s), MO_UW | MO_ALIGN);
1061     }
1062     return true;
1063 }
1064 
1065 static bool trans_VLDR_VSTR_sp(DisasContext *s, arg_VLDR_VSTR_sp *a)
1066 {
1067     uint32_t offset;
1068     TCGv_i32 addr, tmp;
1069 
1070     if (!dc_isar_feature(aa32_fpsp_v2, s) && !dc_isar_feature(aa32_mve, s)) {
1071         return false;
1072     }
1073 
1074     if (!vfp_access_check(s)) {
1075         return true;
1076     }
1077 
1078     offset = a->imm << 2;
1079     if (!a->u) {
1080         offset = -offset;
1081     }
1082 
1083     /* For thumb, use of PC is UNPREDICTABLE.  */
1084     addr = add_reg_for_lit(s, a->rn, offset);
1085     tmp = tcg_temp_new_i32();
1086     if (a->l) {
1087         gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), MO_UL | MO_ALIGN);
1088         vfp_store_reg32(tmp, a->vd);
1089     } else {
1090         vfp_load_reg32(tmp, a->vd);
1091         gen_aa32_st_i32(s, tmp, addr, get_mem_index(s), MO_UL | MO_ALIGN);
1092     }
1093     return true;
1094 }
1095 
1096 static bool trans_VLDR_VSTR_dp(DisasContext *s, arg_VLDR_VSTR_dp *a)
1097 {
1098     uint32_t offset;
1099     TCGv_i32 addr;
1100     TCGv_i64 tmp;
1101 
1102     /* Note that this does not require support for double arithmetic.  */
1103     if (!dc_isar_feature(aa32_fpsp_v2, s) && !dc_isar_feature(aa32_mve, s)) {
1104         return false;
1105     }
1106 
1107     /* UNDEF accesses to D16-D31 if they don't exist */
1108     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vd & 0x10)) {
1109         return false;
1110     }
1111 
1112     if (!vfp_access_check(s)) {
1113         return true;
1114     }
1115 
1116     offset = a->imm << 2;
1117     if (!a->u) {
1118         offset = -offset;
1119     }
1120 
1121     /* For thumb, use of PC is UNPREDICTABLE.  */
1122     addr = add_reg_for_lit(s, a->rn, offset);
1123     tmp = tcg_temp_new_i64();
1124     if (a->l) {
1125         gen_aa32_ld_i64(s, tmp, addr, get_mem_index(s), MO_UQ | MO_ALIGN_4);
1126         vfp_store_reg64(tmp, a->vd);
1127     } else {
1128         vfp_load_reg64(tmp, a->vd);
1129         gen_aa32_st_i64(s, tmp, addr, get_mem_index(s), MO_UQ | MO_ALIGN_4);
1130     }
1131     return true;
1132 }
1133 
1134 static bool trans_VLDM_VSTM_sp(DisasContext *s, arg_VLDM_VSTM_sp *a)
1135 {
1136     uint32_t offset;
1137     TCGv_i32 addr, tmp;
1138     int i, n;
1139 
1140     if (!dc_isar_feature(aa32_fpsp_v2, s) && !dc_isar_feature(aa32_mve, s)) {
1141         return false;
1142     }
1143 
1144     n = a->imm;
1145 
1146     if (n == 0 || (a->vd + n) > 32) {
1147         /*
1148          * UNPREDICTABLE cases for bad immediates: we choose to
1149          * UNDEF to avoid generating huge numbers of TCG ops
1150          */
1151         return false;
1152     }
1153     if (a->rn == 15 && a->w) {
1154         /* writeback to PC is UNPREDICTABLE, we choose to UNDEF */
1155         return false;
1156     }
1157 
1158     s->eci_handled = true;
1159 
1160     if (!vfp_access_check(s)) {
1161         return true;
1162     }
1163 
1164     /* For thumb, use of PC is UNPREDICTABLE.  */
1165     addr = add_reg_for_lit(s, a->rn, 0);
1166     if (a->p) {
1167         /* pre-decrement */
1168         tcg_gen_addi_i32(addr, addr, -(a->imm << 2));
1169     }
1170 
1171     if (s->v8m_stackcheck && a->rn == 13 && a->w) {
1172         /*
1173          * Here 'addr' is the lowest address we will store to,
1174          * and is either the old SP (if post-increment) or
1175          * the new SP (if pre-decrement). For post-increment
1176          * where the old value is below the limit and the new
1177          * value is above, it is UNKNOWN whether the limit check
1178          * triggers; we choose to trigger.
1179          */
1180         gen_helper_v8m_stackcheck(cpu_env, addr);
1181     }
1182 
1183     offset = 4;
1184     tmp = tcg_temp_new_i32();
1185     for (i = 0; i < n; i++) {
1186         if (a->l) {
1187             /* load */
1188             gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), MO_UL | MO_ALIGN);
1189             vfp_store_reg32(tmp, a->vd + i);
1190         } else {
1191             /* store */
1192             vfp_load_reg32(tmp, a->vd + i);
1193             gen_aa32_st_i32(s, tmp, addr, get_mem_index(s), MO_UL | MO_ALIGN);
1194         }
1195         tcg_gen_addi_i32(addr, addr, offset);
1196     }
1197     if (a->w) {
1198         /* writeback */
1199         if (a->p) {
1200             offset = -offset * n;
1201             tcg_gen_addi_i32(addr, addr, offset);
1202         }
1203         store_reg(s, a->rn, addr);
1204     }
1205 
1206     clear_eci_state(s);
1207     return true;
1208 }
1209 
1210 static bool trans_VLDM_VSTM_dp(DisasContext *s, arg_VLDM_VSTM_dp *a)
1211 {
1212     uint32_t offset;
1213     TCGv_i32 addr;
1214     TCGv_i64 tmp;
1215     int i, n;
1216 
1217     /* Note that this does not require support for double arithmetic.  */
1218     if (!dc_isar_feature(aa32_fpsp_v2, s) && !dc_isar_feature(aa32_mve, s)) {
1219         return false;
1220     }
1221 
1222     n = a->imm >> 1;
1223 
1224     if (n == 0 || (a->vd + n) > 32 || n > 16) {
1225         /*
1226          * UNPREDICTABLE cases for bad immediates: we choose to
1227          * UNDEF to avoid generating huge numbers of TCG ops
1228          */
1229         return false;
1230     }
1231     if (a->rn == 15 && a->w) {
1232         /* writeback to PC is UNPREDICTABLE, we choose to UNDEF */
1233         return false;
1234     }
1235 
1236     /* UNDEF accesses to D16-D31 if they don't exist */
1237     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vd + n) > 16) {
1238         return false;
1239     }
1240 
1241     s->eci_handled = true;
1242 
1243     if (!vfp_access_check(s)) {
1244         return true;
1245     }
1246 
1247     /* For thumb, use of PC is UNPREDICTABLE.  */
1248     addr = add_reg_for_lit(s, a->rn, 0);
1249     if (a->p) {
1250         /* pre-decrement */
1251         tcg_gen_addi_i32(addr, addr, -(a->imm << 2));
1252     }
1253 
1254     if (s->v8m_stackcheck && a->rn == 13 && a->w) {
1255         /*
1256          * Here 'addr' is the lowest address we will store to,
1257          * and is either the old SP (if post-increment) or
1258          * the new SP (if pre-decrement). For post-increment
1259          * where the old value is below the limit and the new
1260          * value is above, it is UNKNOWN whether the limit check
1261          * triggers; we choose to trigger.
1262          */
1263         gen_helper_v8m_stackcheck(cpu_env, addr);
1264     }
1265 
1266     offset = 8;
1267     tmp = tcg_temp_new_i64();
1268     for (i = 0; i < n; i++) {
1269         if (a->l) {
1270             /* load */
1271             gen_aa32_ld_i64(s, tmp, addr, get_mem_index(s), MO_UQ | MO_ALIGN_4);
1272             vfp_store_reg64(tmp, a->vd + i);
1273         } else {
1274             /* store */
1275             vfp_load_reg64(tmp, a->vd + i);
1276             gen_aa32_st_i64(s, tmp, addr, get_mem_index(s), MO_UQ | MO_ALIGN_4);
1277         }
1278         tcg_gen_addi_i32(addr, addr, offset);
1279     }
1280     if (a->w) {
1281         /* writeback */
1282         if (a->p) {
1283             offset = -offset * n;
1284         } else if (a->imm & 1) {
1285             offset = 4;
1286         } else {
1287             offset = 0;
1288         }
1289 
1290         if (offset != 0) {
1291             tcg_gen_addi_i32(addr, addr, offset);
1292         }
1293         store_reg(s, a->rn, addr);
1294     }
1295 
1296     clear_eci_state(s);
1297     return true;
1298 }
1299 
1300 /*
1301  * Types for callbacks for do_vfp_3op_sp() and do_vfp_3op_dp().
1302  * The callback should emit code to write a value to vd. If
1303  * do_vfp_3op_{sp,dp}() was passed reads_vd then the TCGv vd
1304  * will contain the old value of the relevant VFP register;
1305  * otherwise it must be written to only.
1306  */
1307 typedef void VFPGen3OpSPFn(TCGv_i32 vd,
1308                            TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst);
1309 typedef void VFPGen3OpDPFn(TCGv_i64 vd,
1310                            TCGv_i64 vn, TCGv_i64 vm, TCGv_ptr fpst);
1311 
1312 /*
1313  * Types for callbacks for do_vfp_2op_sp() and do_vfp_2op_dp().
1314  * The callback should emit code to write a value to vd (which
1315  * should be written to only).
1316  */
1317 typedef void VFPGen2OpSPFn(TCGv_i32 vd, TCGv_i32 vm);
1318 typedef void VFPGen2OpDPFn(TCGv_i64 vd, TCGv_i64 vm);
1319 
1320 /*
1321  * Return true if the specified S reg is in a scalar bank
1322  * (ie if it is s0..s7)
1323  */
1324 static inline bool vfp_sreg_is_scalar(int reg)
1325 {
1326     return (reg & 0x18) == 0;
1327 }
1328 
1329 /*
1330  * Return true if the specified D reg is in a scalar bank
1331  * (ie if it is d0..d3 or d16..d19)
1332  */
1333 static inline bool vfp_dreg_is_scalar(int reg)
1334 {
1335     return (reg & 0xc) == 0;
1336 }
1337 
1338 /*
1339  * Advance the S reg number forwards by delta within its bank
1340  * (ie increment the low 3 bits but leave the rest the same)
1341  */
1342 static inline int vfp_advance_sreg(int reg, int delta)
1343 {
1344     return ((reg + delta) & 0x7) | (reg & ~0x7);
1345 }
1346 
1347 /*
1348  * Advance the D reg number forwards by delta within its bank
1349  * (ie increment the low 2 bits but leave the rest the same)
1350  */
1351 static inline int vfp_advance_dreg(int reg, int delta)
1352 {
1353     return ((reg + delta) & 0x3) | (reg & ~0x3);
1354 }
1355 
1356 /*
1357  * Perform a 3-operand VFP data processing instruction. fn is the
1358  * callback to do the actual operation; this function deals with the
1359  * code to handle looping around for VFP vector processing.
1360  */
1361 static bool do_vfp_3op_sp(DisasContext *s, VFPGen3OpSPFn *fn,
1362                           int vd, int vn, int vm, bool reads_vd)
1363 {
1364     uint32_t delta_m = 0;
1365     uint32_t delta_d = 0;
1366     int veclen = s->vec_len;
1367     TCGv_i32 f0, f1, fd;
1368     TCGv_ptr fpst;
1369 
1370     if (!dc_isar_feature(aa32_fpsp_v2, s)) {
1371         return false;
1372     }
1373 
1374     if (!dc_isar_feature(aa32_fpshvec, s) &&
1375         (veclen != 0 || s->vec_stride != 0)) {
1376         return false;
1377     }
1378 
1379     if (!vfp_access_check(s)) {
1380         return true;
1381     }
1382 
1383     if (veclen > 0) {
1384         /* Figure out what type of vector operation this is.  */
1385         if (vfp_sreg_is_scalar(vd)) {
1386             /* scalar */
1387             veclen = 0;
1388         } else {
1389             delta_d = s->vec_stride + 1;
1390 
1391             if (vfp_sreg_is_scalar(vm)) {
1392                 /* mixed scalar/vector */
1393                 delta_m = 0;
1394             } else {
1395                 /* vector */
1396                 delta_m = delta_d;
1397             }
1398         }
1399     }
1400 
1401     f0 = tcg_temp_new_i32();
1402     f1 = tcg_temp_new_i32();
1403     fd = tcg_temp_new_i32();
1404     fpst = fpstatus_ptr(FPST_FPCR);
1405 
1406     vfp_load_reg32(f0, vn);
1407     vfp_load_reg32(f1, vm);
1408 
1409     for (;;) {
1410         if (reads_vd) {
1411             vfp_load_reg32(fd, vd);
1412         }
1413         fn(fd, f0, f1, fpst);
1414         vfp_store_reg32(fd, vd);
1415 
1416         if (veclen == 0) {
1417             break;
1418         }
1419 
1420         /* Set up the operands for the next iteration */
1421         veclen--;
1422         vd = vfp_advance_sreg(vd, delta_d);
1423         vn = vfp_advance_sreg(vn, delta_d);
1424         vfp_load_reg32(f0, vn);
1425         if (delta_m) {
1426             vm = vfp_advance_sreg(vm, delta_m);
1427             vfp_load_reg32(f1, vm);
1428         }
1429     }
1430     return true;
1431 }
1432 
1433 static bool do_vfp_3op_hp(DisasContext *s, VFPGen3OpSPFn *fn,
1434                           int vd, int vn, int vm, bool reads_vd)
1435 {
1436     /*
1437      * Do a half-precision operation. Functionally this is
1438      * the same as do_vfp_3op_sp(), except:
1439      *  - it uses the FPST_FPCR_F16
1440      *  - it doesn't need the VFP vector handling (fp16 is a
1441      *    v8 feature, and in v8 VFP vectors don't exist)
1442      *  - it does the aa32_fp16_arith feature test
1443      */
1444     TCGv_i32 f0, f1, fd;
1445     TCGv_ptr fpst;
1446 
1447     if (!dc_isar_feature(aa32_fp16_arith, s)) {
1448         return false;
1449     }
1450 
1451     if (s->vec_len != 0 || s->vec_stride != 0) {
1452         return false;
1453     }
1454 
1455     if (!vfp_access_check(s)) {
1456         return true;
1457     }
1458 
1459     f0 = tcg_temp_new_i32();
1460     f1 = tcg_temp_new_i32();
1461     fd = tcg_temp_new_i32();
1462     fpst = fpstatus_ptr(FPST_FPCR_F16);
1463 
1464     vfp_load_reg32(f0, vn);
1465     vfp_load_reg32(f1, vm);
1466 
1467     if (reads_vd) {
1468         vfp_load_reg32(fd, vd);
1469     }
1470     fn(fd, f0, f1, fpst);
1471     vfp_store_reg32(fd, vd);
1472     return true;
1473 }
1474 
1475 static bool do_vfp_3op_dp(DisasContext *s, VFPGen3OpDPFn *fn,
1476                           int vd, int vn, int vm, bool reads_vd)
1477 {
1478     uint32_t delta_m = 0;
1479     uint32_t delta_d = 0;
1480     int veclen = s->vec_len;
1481     TCGv_i64 f0, f1, fd;
1482     TCGv_ptr fpst;
1483 
1484     if (!dc_isar_feature(aa32_fpdp_v2, s)) {
1485         return false;
1486     }
1487 
1488     /* UNDEF accesses to D16-D31 if they don't exist */
1489     if (!dc_isar_feature(aa32_simd_r32, s) && ((vd | vn | vm) & 0x10)) {
1490         return false;
1491     }
1492 
1493     if (!dc_isar_feature(aa32_fpshvec, s) &&
1494         (veclen != 0 || s->vec_stride != 0)) {
1495         return false;
1496     }
1497 
1498     if (!vfp_access_check(s)) {
1499         return true;
1500     }
1501 
1502     if (veclen > 0) {
1503         /* Figure out what type of vector operation this is.  */
1504         if (vfp_dreg_is_scalar(vd)) {
1505             /* scalar */
1506             veclen = 0;
1507         } else {
1508             delta_d = (s->vec_stride >> 1) + 1;
1509 
1510             if (vfp_dreg_is_scalar(vm)) {
1511                 /* mixed scalar/vector */
1512                 delta_m = 0;
1513             } else {
1514                 /* vector */
1515                 delta_m = delta_d;
1516             }
1517         }
1518     }
1519 
1520     f0 = tcg_temp_new_i64();
1521     f1 = tcg_temp_new_i64();
1522     fd = tcg_temp_new_i64();
1523     fpst = fpstatus_ptr(FPST_FPCR);
1524 
1525     vfp_load_reg64(f0, vn);
1526     vfp_load_reg64(f1, vm);
1527 
1528     for (;;) {
1529         if (reads_vd) {
1530             vfp_load_reg64(fd, vd);
1531         }
1532         fn(fd, f0, f1, fpst);
1533         vfp_store_reg64(fd, vd);
1534 
1535         if (veclen == 0) {
1536             break;
1537         }
1538         /* Set up the operands for the next iteration */
1539         veclen--;
1540         vd = vfp_advance_dreg(vd, delta_d);
1541         vn = vfp_advance_dreg(vn, delta_d);
1542         vfp_load_reg64(f0, vn);
1543         if (delta_m) {
1544             vm = vfp_advance_dreg(vm, delta_m);
1545             vfp_load_reg64(f1, vm);
1546         }
1547     }
1548     return true;
1549 }
1550 
1551 static bool do_vfp_2op_sp(DisasContext *s, VFPGen2OpSPFn *fn, int vd, int vm)
1552 {
1553     uint32_t delta_m = 0;
1554     uint32_t delta_d = 0;
1555     int veclen = s->vec_len;
1556     TCGv_i32 f0, fd;
1557 
1558     /* Note that the caller must check the aa32_fpsp_v2 feature. */
1559 
1560     if (!dc_isar_feature(aa32_fpshvec, s) &&
1561         (veclen != 0 || s->vec_stride != 0)) {
1562         return false;
1563     }
1564 
1565     if (!vfp_access_check(s)) {
1566         return true;
1567     }
1568 
1569     if (veclen > 0) {
1570         /* Figure out what type of vector operation this is.  */
1571         if (vfp_sreg_is_scalar(vd)) {
1572             /* scalar */
1573             veclen = 0;
1574         } else {
1575             delta_d = s->vec_stride + 1;
1576 
1577             if (vfp_sreg_is_scalar(vm)) {
1578                 /* mixed scalar/vector */
1579                 delta_m = 0;
1580             } else {
1581                 /* vector */
1582                 delta_m = delta_d;
1583             }
1584         }
1585     }
1586 
1587     f0 = tcg_temp_new_i32();
1588     fd = tcg_temp_new_i32();
1589 
1590     vfp_load_reg32(f0, vm);
1591 
1592     for (;;) {
1593         fn(fd, f0);
1594         vfp_store_reg32(fd, vd);
1595 
1596         if (veclen == 0) {
1597             break;
1598         }
1599 
1600         if (delta_m == 0) {
1601             /* single source one-many */
1602             while (veclen--) {
1603                 vd = vfp_advance_sreg(vd, delta_d);
1604                 vfp_store_reg32(fd, vd);
1605             }
1606             break;
1607         }
1608 
1609         /* Set up the operands for the next iteration */
1610         veclen--;
1611         vd = vfp_advance_sreg(vd, delta_d);
1612         vm = vfp_advance_sreg(vm, delta_m);
1613         vfp_load_reg32(f0, vm);
1614     }
1615     return true;
1616 }
1617 
1618 static bool do_vfp_2op_hp(DisasContext *s, VFPGen2OpSPFn *fn, int vd, int vm)
1619 {
1620     /*
1621      * Do a half-precision operation. Functionally this is
1622      * the same as do_vfp_2op_sp(), except:
1623      *  - it doesn't need the VFP vector handling (fp16 is a
1624      *    v8 feature, and in v8 VFP vectors don't exist)
1625      *  - it does the aa32_fp16_arith feature test
1626      */
1627     TCGv_i32 f0;
1628 
1629     /* Note that the caller must check the aa32_fp16_arith feature */
1630 
1631     if (!dc_isar_feature(aa32_fp16_arith, s)) {
1632         return false;
1633     }
1634 
1635     if (s->vec_len != 0 || s->vec_stride != 0) {
1636         return false;
1637     }
1638 
1639     if (!vfp_access_check(s)) {
1640         return true;
1641     }
1642 
1643     f0 = tcg_temp_new_i32();
1644     vfp_load_reg32(f0, vm);
1645     fn(f0, f0);
1646     vfp_store_reg32(f0, vd);
1647 
1648     return true;
1649 }
1650 
1651 static bool do_vfp_2op_dp(DisasContext *s, VFPGen2OpDPFn *fn, int vd, int vm)
1652 {
1653     uint32_t delta_m = 0;
1654     uint32_t delta_d = 0;
1655     int veclen = s->vec_len;
1656     TCGv_i64 f0, fd;
1657 
1658     /* Note that the caller must check the aa32_fpdp_v2 feature. */
1659 
1660     /* UNDEF accesses to D16-D31 if they don't exist */
1661     if (!dc_isar_feature(aa32_simd_r32, s) && ((vd | vm) & 0x10)) {
1662         return false;
1663     }
1664 
1665     if (!dc_isar_feature(aa32_fpshvec, s) &&
1666         (veclen != 0 || s->vec_stride != 0)) {
1667         return false;
1668     }
1669 
1670     if (!vfp_access_check(s)) {
1671         return true;
1672     }
1673 
1674     if (veclen > 0) {
1675         /* Figure out what type of vector operation this is.  */
1676         if (vfp_dreg_is_scalar(vd)) {
1677             /* scalar */
1678             veclen = 0;
1679         } else {
1680             delta_d = (s->vec_stride >> 1) + 1;
1681 
1682             if (vfp_dreg_is_scalar(vm)) {
1683                 /* mixed scalar/vector */
1684                 delta_m = 0;
1685             } else {
1686                 /* vector */
1687                 delta_m = delta_d;
1688             }
1689         }
1690     }
1691 
1692     f0 = tcg_temp_new_i64();
1693     fd = tcg_temp_new_i64();
1694 
1695     vfp_load_reg64(f0, vm);
1696 
1697     for (;;) {
1698         fn(fd, f0);
1699         vfp_store_reg64(fd, vd);
1700 
1701         if (veclen == 0) {
1702             break;
1703         }
1704 
1705         if (delta_m == 0) {
1706             /* single source one-many */
1707             while (veclen--) {
1708                 vd = vfp_advance_dreg(vd, delta_d);
1709                 vfp_store_reg64(fd, vd);
1710             }
1711             break;
1712         }
1713 
1714         /* Set up the operands for the next iteration */
1715         veclen--;
1716         vd = vfp_advance_dreg(vd, delta_d);
1717         vd = vfp_advance_dreg(vm, delta_m);
1718         vfp_load_reg64(f0, vm);
1719     }
1720     return true;
1721 }
1722 
1723 static void gen_VMLA_hp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst)
1724 {
1725     /* Note that order of inputs to the add matters for NaNs */
1726     TCGv_i32 tmp = tcg_temp_new_i32();
1727 
1728     gen_helper_vfp_mulh(tmp, vn, vm, fpst);
1729     gen_helper_vfp_addh(vd, vd, tmp, fpst);
1730 }
1731 
1732 static bool trans_VMLA_hp(DisasContext *s, arg_VMLA_sp *a)
1733 {
1734     return do_vfp_3op_hp(s, gen_VMLA_hp, a->vd, a->vn, a->vm, true);
1735 }
1736 
1737 static void gen_VMLA_sp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst)
1738 {
1739     /* Note that order of inputs to the add matters for NaNs */
1740     TCGv_i32 tmp = tcg_temp_new_i32();
1741 
1742     gen_helper_vfp_muls(tmp, vn, vm, fpst);
1743     gen_helper_vfp_adds(vd, vd, tmp, fpst);
1744 }
1745 
1746 static bool trans_VMLA_sp(DisasContext *s, arg_VMLA_sp *a)
1747 {
1748     return do_vfp_3op_sp(s, gen_VMLA_sp, a->vd, a->vn, a->vm, true);
1749 }
1750 
1751 static void gen_VMLA_dp(TCGv_i64 vd, TCGv_i64 vn, TCGv_i64 vm, TCGv_ptr fpst)
1752 {
1753     /* Note that order of inputs to the add matters for NaNs */
1754     TCGv_i64 tmp = tcg_temp_new_i64();
1755 
1756     gen_helper_vfp_muld(tmp, vn, vm, fpst);
1757     gen_helper_vfp_addd(vd, vd, tmp, fpst);
1758 }
1759 
1760 static bool trans_VMLA_dp(DisasContext *s, arg_VMLA_dp *a)
1761 {
1762     return do_vfp_3op_dp(s, gen_VMLA_dp, a->vd, a->vn, a->vm, true);
1763 }
1764 
1765 static void gen_VMLS_hp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst)
1766 {
1767     /*
1768      * VMLS: vd = vd + -(vn * vm)
1769      * Note that order of inputs to the add matters for NaNs.
1770      */
1771     TCGv_i32 tmp = tcg_temp_new_i32();
1772 
1773     gen_helper_vfp_mulh(tmp, vn, vm, fpst);
1774     gen_helper_vfp_negh(tmp, tmp);
1775     gen_helper_vfp_addh(vd, vd, tmp, fpst);
1776 }
1777 
1778 static bool trans_VMLS_hp(DisasContext *s, arg_VMLS_sp *a)
1779 {
1780     return do_vfp_3op_hp(s, gen_VMLS_hp, a->vd, a->vn, a->vm, true);
1781 }
1782 
1783 static void gen_VMLS_sp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst)
1784 {
1785     /*
1786      * VMLS: vd = vd + -(vn * vm)
1787      * Note that order of inputs to the add matters for NaNs.
1788      */
1789     TCGv_i32 tmp = tcg_temp_new_i32();
1790 
1791     gen_helper_vfp_muls(tmp, vn, vm, fpst);
1792     gen_helper_vfp_negs(tmp, tmp);
1793     gen_helper_vfp_adds(vd, vd, tmp, fpst);
1794 }
1795 
1796 static bool trans_VMLS_sp(DisasContext *s, arg_VMLS_sp *a)
1797 {
1798     return do_vfp_3op_sp(s, gen_VMLS_sp, a->vd, a->vn, a->vm, true);
1799 }
1800 
1801 static void gen_VMLS_dp(TCGv_i64 vd, TCGv_i64 vn, TCGv_i64 vm, TCGv_ptr fpst)
1802 {
1803     /*
1804      * VMLS: vd = vd + -(vn * vm)
1805      * Note that order of inputs to the add matters for NaNs.
1806      */
1807     TCGv_i64 tmp = tcg_temp_new_i64();
1808 
1809     gen_helper_vfp_muld(tmp, vn, vm, fpst);
1810     gen_helper_vfp_negd(tmp, tmp);
1811     gen_helper_vfp_addd(vd, vd, tmp, fpst);
1812 }
1813 
1814 static bool trans_VMLS_dp(DisasContext *s, arg_VMLS_dp *a)
1815 {
1816     return do_vfp_3op_dp(s, gen_VMLS_dp, a->vd, a->vn, a->vm, true);
1817 }
1818 
1819 static void gen_VNMLS_hp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst)
1820 {
1821     /*
1822      * VNMLS: -fd + (fn * fm)
1823      * Note that it isn't valid to replace (-A + B) with (B - A) or similar
1824      * plausible looking simplifications because this will give wrong results
1825      * for NaNs.
1826      */
1827     TCGv_i32 tmp = tcg_temp_new_i32();
1828 
1829     gen_helper_vfp_mulh(tmp, vn, vm, fpst);
1830     gen_helper_vfp_negh(vd, vd);
1831     gen_helper_vfp_addh(vd, vd, tmp, fpst);
1832 }
1833 
1834 static bool trans_VNMLS_hp(DisasContext *s, arg_VNMLS_sp *a)
1835 {
1836     return do_vfp_3op_hp(s, gen_VNMLS_hp, a->vd, a->vn, a->vm, true);
1837 }
1838 
1839 static void gen_VNMLS_sp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst)
1840 {
1841     /*
1842      * VNMLS: -fd + (fn * fm)
1843      * Note that it isn't valid to replace (-A + B) with (B - A) or similar
1844      * plausible looking simplifications because this will give wrong results
1845      * for NaNs.
1846      */
1847     TCGv_i32 tmp = tcg_temp_new_i32();
1848 
1849     gen_helper_vfp_muls(tmp, vn, vm, fpst);
1850     gen_helper_vfp_negs(vd, vd);
1851     gen_helper_vfp_adds(vd, vd, tmp, fpst);
1852 }
1853 
1854 static bool trans_VNMLS_sp(DisasContext *s, arg_VNMLS_sp *a)
1855 {
1856     return do_vfp_3op_sp(s, gen_VNMLS_sp, a->vd, a->vn, a->vm, true);
1857 }
1858 
1859 static void gen_VNMLS_dp(TCGv_i64 vd, TCGv_i64 vn, TCGv_i64 vm, TCGv_ptr fpst)
1860 {
1861     /*
1862      * VNMLS: -fd + (fn * fm)
1863      * Note that it isn't valid to replace (-A + B) with (B - A) or similar
1864      * plausible looking simplifications because this will give wrong results
1865      * for NaNs.
1866      */
1867     TCGv_i64 tmp = tcg_temp_new_i64();
1868 
1869     gen_helper_vfp_muld(tmp, vn, vm, fpst);
1870     gen_helper_vfp_negd(vd, vd);
1871     gen_helper_vfp_addd(vd, vd, tmp, fpst);
1872 }
1873 
1874 static bool trans_VNMLS_dp(DisasContext *s, arg_VNMLS_dp *a)
1875 {
1876     return do_vfp_3op_dp(s, gen_VNMLS_dp, a->vd, a->vn, a->vm, true);
1877 }
1878 
1879 static void gen_VNMLA_hp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst)
1880 {
1881     /* VNMLA: -fd + -(fn * fm) */
1882     TCGv_i32 tmp = tcg_temp_new_i32();
1883 
1884     gen_helper_vfp_mulh(tmp, vn, vm, fpst);
1885     gen_helper_vfp_negh(tmp, tmp);
1886     gen_helper_vfp_negh(vd, vd);
1887     gen_helper_vfp_addh(vd, vd, tmp, fpst);
1888 }
1889 
1890 static bool trans_VNMLA_hp(DisasContext *s, arg_VNMLA_sp *a)
1891 {
1892     return do_vfp_3op_hp(s, gen_VNMLA_hp, a->vd, a->vn, a->vm, true);
1893 }
1894 
1895 static void gen_VNMLA_sp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst)
1896 {
1897     /* VNMLA: -fd + -(fn * fm) */
1898     TCGv_i32 tmp = tcg_temp_new_i32();
1899 
1900     gen_helper_vfp_muls(tmp, vn, vm, fpst);
1901     gen_helper_vfp_negs(tmp, tmp);
1902     gen_helper_vfp_negs(vd, vd);
1903     gen_helper_vfp_adds(vd, vd, tmp, fpst);
1904 }
1905 
1906 static bool trans_VNMLA_sp(DisasContext *s, arg_VNMLA_sp *a)
1907 {
1908     return do_vfp_3op_sp(s, gen_VNMLA_sp, a->vd, a->vn, a->vm, true);
1909 }
1910 
1911 static void gen_VNMLA_dp(TCGv_i64 vd, TCGv_i64 vn, TCGv_i64 vm, TCGv_ptr fpst)
1912 {
1913     /* VNMLA: -fd + (fn * fm) */
1914     TCGv_i64 tmp = tcg_temp_new_i64();
1915 
1916     gen_helper_vfp_muld(tmp, vn, vm, fpst);
1917     gen_helper_vfp_negd(tmp, tmp);
1918     gen_helper_vfp_negd(vd, vd);
1919     gen_helper_vfp_addd(vd, vd, tmp, fpst);
1920 }
1921 
1922 static bool trans_VNMLA_dp(DisasContext *s, arg_VNMLA_dp *a)
1923 {
1924     return do_vfp_3op_dp(s, gen_VNMLA_dp, a->vd, a->vn, a->vm, true);
1925 }
1926 
1927 static bool trans_VMUL_hp(DisasContext *s, arg_VMUL_sp *a)
1928 {
1929     return do_vfp_3op_hp(s, gen_helper_vfp_mulh, a->vd, a->vn, a->vm, false);
1930 }
1931 
1932 static bool trans_VMUL_sp(DisasContext *s, arg_VMUL_sp *a)
1933 {
1934     return do_vfp_3op_sp(s, gen_helper_vfp_muls, a->vd, a->vn, a->vm, false);
1935 }
1936 
1937 static bool trans_VMUL_dp(DisasContext *s, arg_VMUL_dp *a)
1938 {
1939     return do_vfp_3op_dp(s, gen_helper_vfp_muld, a->vd, a->vn, a->vm, false);
1940 }
1941 
1942 static void gen_VNMUL_hp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst)
1943 {
1944     /* VNMUL: -(fn * fm) */
1945     gen_helper_vfp_mulh(vd, vn, vm, fpst);
1946     gen_helper_vfp_negh(vd, vd);
1947 }
1948 
1949 static bool trans_VNMUL_hp(DisasContext *s, arg_VNMUL_sp *a)
1950 {
1951     return do_vfp_3op_hp(s, gen_VNMUL_hp, a->vd, a->vn, a->vm, false);
1952 }
1953 
1954 static void gen_VNMUL_sp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst)
1955 {
1956     /* VNMUL: -(fn * fm) */
1957     gen_helper_vfp_muls(vd, vn, vm, fpst);
1958     gen_helper_vfp_negs(vd, vd);
1959 }
1960 
1961 static bool trans_VNMUL_sp(DisasContext *s, arg_VNMUL_sp *a)
1962 {
1963     return do_vfp_3op_sp(s, gen_VNMUL_sp, a->vd, a->vn, a->vm, false);
1964 }
1965 
1966 static void gen_VNMUL_dp(TCGv_i64 vd, TCGv_i64 vn, TCGv_i64 vm, TCGv_ptr fpst)
1967 {
1968     /* VNMUL: -(fn * fm) */
1969     gen_helper_vfp_muld(vd, vn, vm, fpst);
1970     gen_helper_vfp_negd(vd, vd);
1971 }
1972 
1973 static bool trans_VNMUL_dp(DisasContext *s, arg_VNMUL_dp *a)
1974 {
1975     return do_vfp_3op_dp(s, gen_VNMUL_dp, a->vd, a->vn, a->vm, false);
1976 }
1977 
1978 static bool trans_VADD_hp(DisasContext *s, arg_VADD_sp *a)
1979 {
1980     return do_vfp_3op_hp(s, gen_helper_vfp_addh, a->vd, a->vn, a->vm, false);
1981 }
1982 
1983 static bool trans_VADD_sp(DisasContext *s, arg_VADD_sp *a)
1984 {
1985     return do_vfp_3op_sp(s, gen_helper_vfp_adds, a->vd, a->vn, a->vm, false);
1986 }
1987 
1988 static bool trans_VADD_dp(DisasContext *s, arg_VADD_dp *a)
1989 {
1990     return do_vfp_3op_dp(s, gen_helper_vfp_addd, a->vd, a->vn, a->vm, false);
1991 }
1992 
1993 static bool trans_VSUB_hp(DisasContext *s, arg_VSUB_sp *a)
1994 {
1995     return do_vfp_3op_hp(s, gen_helper_vfp_subh, a->vd, a->vn, a->vm, false);
1996 }
1997 
1998 static bool trans_VSUB_sp(DisasContext *s, arg_VSUB_sp *a)
1999 {
2000     return do_vfp_3op_sp(s, gen_helper_vfp_subs, a->vd, a->vn, a->vm, false);
2001 }
2002 
2003 static bool trans_VSUB_dp(DisasContext *s, arg_VSUB_dp *a)
2004 {
2005     return do_vfp_3op_dp(s, gen_helper_vfp_subd, a->vd, a->vn, a->vm, false);
2006 }
2007 
2008 static bool trans_VDIV_hp(DisasContext *s, arg_VDIV_sp *a)
2009 {
2010     return do_vfp_3op_hp(s, gen_helper_vfp_divh, a->vd, a->vn, a->vm, false);
2011 }
2012 
2013 static bool trans_VDIV_sp(DisasContext *s, arg_VDIV_sp *a)
2014 {
2015     return do_vfp_3op_sp(s, gen_helper_vfp_divs, a->vd, a->vn, a->vm, false);
2016 }
2017 
2018 static bool trans_VDIV_dp(DisasContext *s, arg_VDIV_dp *a)
2019 {
2020     return do_vfp_3op_dp(s, gen_helper_vfp_divd, a->vd, a->vn, a->vm, false);
2021 }
2022 
2023 static bool trans_VMINNM_hp(DisasContext *s, arg_VMINNM_sp *a)
2024 {
2025     if (!dc_isar_feature(aa32_vminmaxnm, s)) {
2026         return false;
2027     }
2028     return do_vfp_3op_hp(s, gen_helper_vfp_minnumh,
2029                          a->vd, a->vn, a->vm, false);
2030 }
2031 
2032 static bool trans_VMAXNM_hp(DisasContext *s, arg_VMAXNM_sp *a)
2033 {
2034     if (!dc_isar_feature(aa32_vminmaxnm, s)) {
2035         return false;
2036     }
2037     return do_vfp_3op_hp(s, gen_helper_vfp_maxnumh,
2038                          a->vd, a->vn, a->vm, false);
2039 }
2040 
2041 static bool trans_VMINNM_sp(DisasContext *s, arg_VMINNM_sp *a)
2042 {
2043     if (!dc_isar_feature(aa32_vminmaxnm, s)) {
2044         return false;
2045     }
2046     return do_vfp_3op_sp(s, gen_helper_vfp_minnums,
2047                          a->vd, a->vn, a->vm, false);
2048 }
2049 
2050 static bool trans_VMAXNM_sp(DisasContext *s, arg_VMAXNM_sp *a)
2051 {
2052     if (!dc_isar_feature(aa32_vminmaxnm, s)) {
2053         return false;
2054     }
2055     return do_vfp_3op_sp(s, gen_helper_vfp_maxnums,
2056                          a->vd, a->vn, a->vm, false);
2057 }
2058 
2059 static bool trans_VMINNM_dp(DisasContext *s, arg_VMINNM_dp *a)
2060 {
2061     if (!dc_isar_feature(aa32_vminmaxnm, s)) {
2062         return false;
2063     }
2064     return do_vfp_3op_dp(s, gen_helper_vfp_minnumd,
2065                          a->vd, a->vn, a->vm, false);
2066 }
2067 
2068 static bool trans_VMAXNM_dp(DisasContext *s, arg_VMAXNM_dp *a)
2069 {
2070     if (!dc_isar_feature(aa32_vminmaxnm, s)) {
2071         return false;
2072     }
2073     return do_vfp_3op_dp(s, gen_helper_vfp_maxnumd,
2074                          a->vd, a->vn, a->vm, false);
2075 }
2076 
2077 static bool do_vfm_hp(DisasContext *s, arg_VFMA_sp *a, bool neg_n, bool neg_d)
2078 {
2079     /*
2080      * VFNMA : fd = muladd(-fd,  fn, fm)
2081      * VFNMS : fd = muladd(-fd, -fn, fm)
2082      * VFMA  : fd = muladd( fd,  fn, fm)
2083      * VFMS  : fd = muladd( fd, -fn, fm)
2084      *
2085      * These are fused multiply-add, and must be done as one floating
2086      * point operation with no rounding between the multiplication and
2087      * addition steps.  NB that doing the negations here as separate
2088      * steps is correct : an input NaN should come out with its sign
2089      * bit flipped if it is a negated-input.
2090      */
2091     TCGv_ptr fpst;
2092     TCGv_i32 vn, vm, vd;
2093 
2094     /*
2095      * Present in VFPv4 only, and only with the FP16 extension.
2096      * Note that we can't rely on the SIMDFMAC check alone, because
2097      * in a Neon-no-VFP core that ID register field will be non-zero.
2098      */
2099     if (!dc_isar_feature(aa32_fp16_arith, s) ||
2100         !dc_isar_feature(aa32_simdfmac, s) ||
2101         !dc_isar_feature(aa32_fpsp_v2, s)) {
2102         return false;
2103     }
2104 
2105     if (s->vec_len != 0 || s->vec_stride != 0) {
2106         return false;
2107     }
2108 
2109     if (!vfp_access_check(s)) {
2110         return true;
2111     }
2112 
2113     vn = tcg_temp_new_i32();
2114     vm = tcg_temp_new_i32();
2115     vd = tcg_temp_new_i32();
2116 
2117     vfp_load_reg32(vn, a->vn);
2118     vfp_load_reg32(vm, a->vm);
2119     if (neg_n) {
2120         /* VFNMS, VFMS */
2121         gen_helper_vfp_negh(vn, vn);
2122     }
2123     vfp_load_reg32(vd, a->vd);
2124     if (neg_d) {
2125         /* VFNMA, VFNMS */
2126         gen_helper_vfp_negh(vd, vd);
2127     }
2128     fpst = fpstatus_ptr(FPST_FPCR_F16);
2129     gen_helper_vfp_muladdh(vd, vn, vm, vd, fpst);
2130     vfp_store_reg32(vd, a->vd);
2131     return true;
2132 }
2133 
2134 static bool do_vfm_sp(DisasContext *s, arg_VFMA_sp *a, bool neg_n, bool neg_d)
2135 {
2136     /*
2137      * VFNMA : fd = muladd(-fd,  fn, fm)
2138      * VFNMS : fd = muladd(-fd, -fn, fm)
2139      * VFMA  : fd = muladd( fd,  fn, fm)
2140      * VFMS  : fd = muladd( fd, -fn, fm)
2141      *
2142      * These are fused multiply-add, and must be done as one floating
2143      * point operation with no rounding between the multiplication and
2144      * addition steps.  NB that doing the negations here as separate
2145      * steps is correct : an input NaN should come out with its sign
2146      * bit flipped if it is a negated-input.
2147      */
2148     TCGv_ptr fpst;
2149     TCGv_i32 vn, vm, vd;
2150 
2151     /*
2152      * Present in VFPv4 only.
2153      * Note that we can't rely on the SIMDFMAC check alone, because
2154      * in a Neon-no-VFP core that ID register field will be non-zero.
2155      */
2156     if (!dc_isar_feature(aa32_simdfmac, s) ||
2157         !dc_isar_feature(aa32_fpsp_v2, s)) {
2158         return false;
2159     }
2160     /*
2161      * In v7A, UNPREDICTABLE with non-zero vector length/stride; from
2162      * v8A, must UNDEF. We choose to UNDEF for both v7A and v8A.
2163      */
2164     if (s->vec_len != 0 || s->vec_stride != 0) {
2165         return false;
2166     }
2167 
2168     if (!vfp_access_check(s)) {
2169         return true;
2170     }
2171 
2172     vn = tcg_temp_new_i32();
2173     vm = tcg_temp_new_i32();
2174     vd = tcg_temp_new_i32();
2175 
2176     vfp_load_reg32(vn, a->vn);
2177     vfp_load_reg32(vm, a->vm);
2178     if (neg_n) {
2179         /* VFNMS, VFMS */
2180         gen_helper_vfp_negs(vn, vn);
2181     }
2182     vfp_load_reg32(vd, a->vd);
2183     if (neg_d) {
2184         /* VFNMA, VFNMS */
2185         gen_helper_vfp_negs(vd, vd);
2186     }
2187     fpst = fpstatus_ptr(FPST_FPCR);
2188     gen_helper_vfp_muladds(vd, vn, vm, vd, fpst);
2189     vfp_store_reg32(vd, a->vd);
2190     return true;
2191 }
2192 
2193 static bool do_vfm_dp(DisasContext *s, arg_VFMA_dp *a, bool neg_n, bool neg_d)
2194 {
2195     /*
2196      * VFNMA : fd = muladd(-fd,  fn, fm)
2197      * VFNMS : fd = muladd(-fd, -fn, fm)
2198      * VFMA  : fd = muladd( fd,  fn, fm)
2199      * VFMS  : fd = muladd( fd, -fn, fm)
2200      *
2201      * These are fused multiply-add, and must be done as one floating
2202      * point operation with no rounding between the multiplication and
2203      * addition steps.  NB that doing the negations here as separate
2204      * steps is correct : an input NaN should come out with its sign
2205      * bit flipped if it is a negated-input.
2206      */
2207     TCGv_ptr fpst;
2208     TCGv_i64 vn, vm, vd;
2209 
2210     /*
2211      * Present in VFPv4 only.
2212      * Note that we can't rely on the SIMDFMAC check alone, because
2213      * in a Neon-no-VFP core that ID register field will be non-zero.
2214      */
2215     if (!dc_isar_feature(aa32_simdfmac, s) ||
2216         !dc_isar_feature(aa32_fpdp_v2, s)) {
2217         return false;
2218     }
2219     /*
2220      * In v7A, UNPREDICTABLE with non-zero vector length/stride; from
2221      * v8A, must UNDEF. We choose to UNDEF for both v7A and v8A.
2222      */
2223     if (s->vec_len != 0 || s->vec_stride != 0) {
2224         return false;
2225     }
2226 
2227     /* UNDEF accesses to D16-D31 if they don't exist. */
2228     if (!dc_isar_feature(aa32_simd_r32, s) &&
2229         ((a->vd | a->vn | a->vm) & 0x10)) {
2230         return false;
2231     }
2232 
2233     if (!vfp_access_check(s)) {
2234         return true;
2235     }
2236 
2237     vn = tcg_temp_new_i64();
2238     vm = tcg_temp_new_i64();
2239     vd = tcg_temp_new_i64();
2240 
2241     vfp_load_reg64(vn, a->vn);
2242     vfp_load_reg64(vm, a->vm);
2243     if (neg_n) {
2244         /* VFNMS, VFMS */
2245         gen_helper_vfp_negd(vn, vn);
2246     }
2247     vfp_load_reg64(vd, a->vd);
2248     if (neg_d) {
2249         /* VFNMA, VFNMS */
2250         gen_helper_vfp_negd(vd, vd);
2251     }
2252     fpst = fpstatus_ptr(FPST_FPCR);
2253     gen_helper_vfp_muladdd(vd, vn, vm, vd, fpst);
2254     vfp_store_reg64(vd, a->vd);
2255     return true;
2256 }
2257 
2258 #define MAKE_ONE_VFM_TRANS_FN(INSN, PREC, NEGN, NEGD)                   \
2259     static bool trans_##INSN##_##PREC(DisasContext *s,                  \
2260                                       arg_##INSN##_##PREC *a)           \
2261     {                                                                   \
2262         return do_vfm_##PREC(s, a, NEGN, NEGD);                         \
2263     }
2264 
2265 #define MAKE_VFM_TRANS_FNS(PREC) \
2266     MAKE_ONE_VFM_TRANS_FN(VFMA, PREC, false, false) \
2267     MAKE_ONE_VFM_TRANS_FN(VFMS, PREC, true, false) \
2268     MAKE_ONE_VFM_TRANS_FN(VFNMA, PREC, false, true) \
2269     MAKE_ONE_VFM_TRANS_FN(VFNMS, PREC, true, true)
2270 
2271 MAKE_VFM_TRANS_FNS(hp)
2272 MAKE_VFM_TRANS_FNS(sp)
2273 MAKE_VFM_TRANS_FNS(dp)
2274 
2275 static bool trans_VMOV_imm_hp(DisasContext *s, arg_VMOV_imm_sp *a)
2276 {
2277     if (!dc_isar_feature(aa32_fp16_arith, s)) {
2278         return false;
2279     }
2280 
2281     if (s->vec_len != 0 || s->vec_stride != 0) {
2282         return false;
2283     }
2284 
2285     if (!vfp_access_check(s)) {
2286         return true;
2287     }
2288 
2289     vfp_store_reg32(tcg_constant_i32(vfp_expand_imm(MO_16, a->imm)), a->vd);
2290     return true;
2291 }
2292 
2293 static bool trans_VMOV_imm_sp(DisasContext *s, arg_VMOV_imm_sp *a)
2294 {
2295     uint32_t delta_d = 0;
2296     int veclen = s->vec_len;
2297     TCGv_i32 fd;
2298     uint32_t vd;
2299 
2300     vd = a->vd;
2301 
2302     if (!dc_isar_feature(aa32_fpsp_v3, s)) {
2303         return false;
2304     }
2305 
2306     if (!dc_isar_feature(aa32_fpshvec, s) &&
2307         (veclen != 0 || s->vec_stride != 0)) {
2308         return false;
2309     }
2310 
2311     if (!vfp_access_check(s)) {
2312         return true;
2313     }
2314 
2315     if (veclen > 0) {
2316         /* Figure out what type of vector operation this is.  */
2317         if (vfp_sreg_is_scalar(vd)) {
2318             /* scalar */
2319             veclen = 0;
2320         } else {
2321             delta_d = s->vec_stride + 1;
2322         }
2323     }
2324 
2325     fd = tcg_constant_i32(vfp_expand_imm(MO_32, a->imm));
2326 
2327     for (;;) {
2328         vfp_store_reg32(fd, vd);
2329 
2330         if (veclen == 0) {
2331             break;
2332         }
2333 
2334         /* Set up the operands for the next iteration */
2335         veclen--;
2336         vd = vfp_advance_sreg(vd, delta_d);
2337     }
2338 
2339     return true;
2340 }
2341 
2342 static bool trans_VMOV_imm_dp(DisasContext *s, arg_VMOV_imm_dp *a)
2343 {
2344     uint32_t delta_d = 0;
2345     int veclen = s->vec_len;
2346     TCGv_i64 fd;
2347     uint32_t vd;
2348 
2349     vd = a->vd;
2350 
2351     if (!dc_isar_feature(aa32_fpdp_v3, s)) {
2352         return false;
2353     }
2354 
2355     /* UNDEF accesses to D16-D31 if they don't exist. */
2356     if (!dc_isar_feature(aa32_simd_r32, s) && (vd & 0x10)) {
2357         return false;
2358     }
2359 
2360     if (!dc_isar_feature(aa32_fpshvec, s) &&
2361         (veclen != 0 || s->vec_stride != 0)) {
2362         return false;
2363     }
2364 
2365     if (!vfp_access_check(s)) {
2366         return true;
2367     }
2368 
2369     if (veclen > 0) {
2370         /* Figure out what type of vector operation this is.  */
2371         if (vfp_dreg_is_scalar(vd)) {
2372             /* scalar */
2373             veclen = 0;
2374         } else {
2375             delta_d = (s->vec_stride >> 1) + 1;
2376         }
2377     }
2378 
2379     fd = tcg_constant_i64(vfp_expand_imm(MO_64, a->imm));
2380 
2381     for (;;) {
2382         vfp_store_reg64(fd, vd);
2383 
2384         if (veclen == 0) {
2385             break;
2386         }
2387 
2388         /* Set up the operands for the next iteration */
2389         veclen--;
2390         vd = vfp_advance_dreg(vd, delta_d);
2391     }
2392 
2393     return true;
2394 }
2395 
2396 #define DO_VFP_2OP(INSN, PREC, FN, CHECK)                       \
2397     static bool trans_##INSN##_##PREC(DisasContext *s,          \
2398                                       arg_##INSN##_##PREC *a)   \
2399     {                                                           \
2400         if (!dc_isar_feature(CHECK, s)) {                       \
2401             return false;                                       \
2402         }                                                       \
2403         return do_vfp_2op_##PREC(s, FN, a->vd, a->vm);          \
2404     }
2405 
2406 #define DO_VFP_VMOV(INSN, PREC, FN)                             \
2407     static bool trans_##INSN##_##PREC(DisasContext *s,          \
2408                                       arg_##INSN##_##PREC *a)   \
2409     {                                                           \
2410         if (!dc_isar_feature(aa32_fp##PREC##_v2, s) &&          \
2411             !dc_isar_feature(aa32_mve, s)) {                    \
2412             return false;                                       \
2413         }                                                       \
2414         return do_vfp_2op_##PREC(s, FN, a->vd, a->vm);          \
2415     }
2416 
2417 DO_VFP_VMOV(VMOV_reg, sp, tcg_gen_mov_i32)
2418 DO_VFP_VMOV(VMOV_reg, dp, tcg_gen_mov_i64)
2419 
2420 DO_VFP_2OP(VABS, hp, gen_helper_vfp_absh, aa32_fp16_arith)
2421 DO_VFP_2OP(VABS, sp, gen_helper_vfp_abss, aa32_fpsp_v2)
2422 DO_VFP_2OP(VABS, dp, gen_helper_vfp_absd, aa32_fpdp_v2)
2423 
2424 DO_VFP_2OP(VNEG, hp, gen_helper_vfp_negh, aa32_fp16_arith)
2425 DO_VFP_2OP(VNEG, sp, gen_helper_vfp_negs, aa32_fpsp_v2)
2426 DO_VFP_2OP(VNEG, dp, gen_helper_vfp_negd, aa32_fpdp_v2)
2427 
2428 static void gen_VSQRT_hp(TCGv_i32 vd, TCGv_i32 vm)
2429 {
2430     gen_helper_vfp_sqrth(vd, vm, cpu_env);
2431 }
2432 
2433 static void gen_VSQRT_sp(TCGv_i32 vd, TCGv_i32 vm)
2434 {
2435     gen_helper_vfp_sqrts(vd, vm, cpu_env);
2436 }
2437 
2438 static void gen_VSQRT_dp(TCGv_i64 vd, TCGv_i64 vm)
2439 {
2440     gen_helper_vfp_sqrtd(vd, vm, cpu_env);
2441 }
2442 
2443 DO_VFP_2OP(VSQRT, hp, gen_VSQRT_hp, aa32_fp16_arith)
2444 DO_VFP_2OP(VSQRT, sp, gen_VSQRT_sp, aa32_fpsp_v2)
2445 DO_VFP_2OP(VSQRT, dp, gen_VSQRT_dp, aa32_fpdp_v2)
2446 
2447 static bool trans_VCMP_hp(DisasContext *s, arg_VCMP_sp *a)
2448 {
2449     TCGv_i32 vd, vm;
2450 
2451     if (!dc_isar_feature(aa32_fp16_arith, s)) {
2452         return false;
2453     }
2454 
2455     /* Vm/M bits must be zero for the Z variant */
2456     if (a->z && a->vm != 0) {
2457         return false;
2458     }
2459 
2460     if (!vfp_access_check(s)) {
2461         return true;
2462     }
2463 
2464     vd = tcg_temp_new_i32();
2465     vm = tcg_temp_new_i32();
2466 
2467     vfp_load_reg32(vd, a->vd);
2468     if (a->z) {
2469         tcg_gen_movi_i32(vm, 0);
2470     } else {
2471         vfp_load_reg32(vm, a->vm);
2472     }
2473 
2474     if (a->e) {
2475         gen_helper_vfp_cmpeh(vd, vm, cpu_env);
2476     } else {
2477         gen_helper_vfp_cmph(vd, vm, cpu_env);
2478     }
2479     return true;
2480 }
2481 
2482 static bool trans_VCMP_sp(DisasContext *s, arg_VCMP_sp *a)
2483 {
2484     TCGv_i32 vd, vm;
2485 
2486     if (!dc_isar_feature(aa32_fpsp_v2, s)) {
2487         return false;
2488     }
2489 
2490     /* Vm/M bits must be zero for the Z variant */
2491     if (a->z && a->vm != 0) {
2492         return false;
2493     }
2494 
2495     if (!vfp_access_check(s)) {
2496         return true;
2497     }
2498 
2499     vd = tcg_temp_new_i32();
2500     vm = tcg_temp_new_i32();
2501 
2502     vfp_load_reg32(vd, a->vd);
2503     if (a->z) {
2504         tcg_gen_movi_i32(vm, 0);
2505     } else {
2506         vfp_load_reg32(vm, a->vm);
2507     }
2508 
2509     if (a->e) {
2510         gen_helper_vfp_cmpes(vd, vm, cpu_env);
2511     } else {
2512         gen_helper_vfp_cmps(vd, vm, cpu_env);
2513     }
2514     return true;
2515 }
2516 
2517 static bool trans_VCMP_dp(DisasContext *s, arg_VCMP_dp *a)
2518 {
2519     TCGv_i64 vd, vm;
2520 
2521     if (!dc_isar_feature(aa32_fpdp_v2, s)) {
2522         return false;
2523     }
2524 
2525     /* Vm/M bits must be zero for the Z variant */
2526     if (a->z && a->vm != 0) {
2527         return false;
2528     }
2529 
2530     /* UNDEF accesses to D16-D31 if they don't exist. */
2531     if (!dc_isar_feature(aa32_simd_r32, s) && ((a->vd | a->vm) & 0x10)) {
2532         return false;
2533     }
2534 
2535     if (!vfp_access_check(s)) {
2536         return true;
2537     }
2538 
2539     vd = tcg_temp_new_i64();
2540     vm = tcg_temp_new_i64();
2541 
2542     vfp_load_reg64(vd, a->vd);
2543     if (a->z) {
2544         tcg_gen_movi_i64(vm, 0);
2545     } else {
2546         vfp_load_reg64(vm, a->vm);
2547     }
2548 
2549     if (a->e) {
2550         gen_helper_vfp_cmped(vd, vm, cpu_env);
2551     } else {
2552         gen_helper_vfp_cmpd(vd, vm, cpu_env);
2553     }
2554     return true;
2555 }
2556 
2557 static bool trans_VCVT_f32_f16(DisasContext *s, arg_VCVT_f32_f16 *a)
2558 {
2559     TCGv_ptr fpst;
2560     TCGv_i32 ahp_mode;
2561     TCGv_i32 tmp;
2562 
2563     if (!dc_isar_feature(aa32_fp16_spconv, s)) {
2564         return false;
2565     }
2566 
2567     if (!vfp_access_check(s)) {
2568         return true;
2569     }
2570 
2571     fpst = fpstatus_ptr(FPST_FPCR);
2572     ahp_mode = get_ahp_flag();
2573     tmp = tcg_temp_new_i32();
2574     /* The T bit tells us if we want the low or high 16 bits of Vm */
2575     tcg_gen_ld16u_i32(tmp, cpu_env, vfp_f16_offset(a->vm, a->t));
2576     gen_helper_vfp_fcvt_f16_to_f32(tmp, tmp, fpst, ahp_mode);
2577     vfp_store_reg32(tmp, a->vd);
2578     return true;
2579 }
2580 
2581 static bool trans_VCVT_f64_f16(DisasContext *s, arg_VCVT_f64_f16 *a)
2582 {
2583     TCGv_ptr fpst;
2584     TCGv_i32 ahp_mode;
2585     TCGv_i32 tmp;
2586     TCGv_i64 vd;
2587 
2588     if (!dc_isar_feature(aa32_fpdp_v2, s)) {
2589         return false;
2590     }
2591 
2592     if (!dc_isar_feature(aa32_fp16_dpconv, s)) {
2593         return false;
2594     }
2595 
2596     /* UNDEF accesses to D16-D31 if they don't exist. */
2597     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vd  & 0x10)) {
2598         return false;
2599     }
2600 
2601     if (!vfp_access_check(s)) {
2602         return true;
2603     }
2604 
2605     fpst = fpstatus_ptr(FPST_FPCR);
2606     ahp_mode = get_ahp_flag();
2607     tmp = tcg_temp_new_i32();
2608     /* The T bit tells us if we want the low or high 16 bits of Vm */
2609     tcg_gen_ld16u_i32(tmp, cpu_env, vfp_f16_offset(a->vm, a->t));
2610     vd = tcg_temp_new_i64();
2611     gen_helper_vfp_fcvt_f16_to_f64(vd, tmp, fpst, ahp_mode);
2612     vfp_store_reg64(vd, a->vd);
2613     return true;
2614 }
2615 
2616 static bool trans_VCVT_b16_f32(DisasContext *s, arg_VCVT_b16_f32 *a)
2617 {
2618     TCGv_ptr fpst;
2619     TCGv_i32 tmp;
2620 
2621     if (!dc_isar_feature(aa32_bf16, s)) {
2622         return false;
2623     }
2624 
2625     if (!vfp_access_check(s)) {
2626         return true;
2627     }
2628 
2629     fpst = fpstatus_ptr(FPST_FPCR);
2630     tmp = tcg_temp_new_i32();
2631 
2632     vfp_load_reg32(tmp, a->vm);
2633     gen_helper_bfcvt(tmp, tmp, fpst);
2634     tcg_gen_st16_i32(tmp, cpu_env, vfp_f16_offset(a->vd, a->t));
2635     return true;
2636 }
2637 
2638 static bool trans_VCVT_f16_f32(DisasContext *s, arg_VCVT_f16_f32 *a)
2639 {
2640     TCGv_ptr fpst;
2641     TCGv_i32 ahp_mode;
2642     TCGv_i32 tmp;
2643 
2644     if (!dc_isar_feature(aa32_fp16_spconv, s)) {
2645         return false;
2646     }
2647 
2648     if (!vfp_access_check(s)) {
2649         return true;
2650     }
2651 
2652     fpst = fpstatus_ptr(FPST_FPCR);
2653     ahp_mode = get_ahp_flag();
2654     tmp = tcg_temp_new_i32();
2655 
2656     vfp_load_reg32(tmp, a->vm);
2657     gen_helper_vfp_fcvt_f32_to_f16(tmp, tmp, fpst, ahp_mode);
2658     tcg_gen_st16_i32(tmp, cpu_env, vfp_f16_offset(a->vd, a->t));
2659     return true;
2660 }
2661 
2662 static bool trans_VCVT_f16_f64(DisasContext *s, arg_VCVT_f16_f64 *a)
2663 {
2664     TCGv_ptr fpst;
2665     TCGv_i32 ahp_mode;
2666     TCGv_i32 tmp;
2667     TCGv_i64 vm;
2668 
2669     if (!dc_isar_feature(aa32_fpdp_v2, s)) {
2670         return false;
2671     }
2672 
2673     if (!dc_isar_feature(aa32_fp16_dpconv, s)) {
2674         return false;
2675     }
2676 
2677     /* UNDEF accesses to D16-D31 if they don't exist. */
2678     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vm  & 0x10)) {
2679         return false;
2680     }
2681 
2682     if (!vfp_access_check(s)) {
2683         return true;
2684     }
2685 
2686     fpst = fpstatus_ptr(FPST_FPCR);
2687     ahp_mode = get_ahp_flag();
2688     tmp = tcg_temp_new_i32();
2689     vm = tcg_temp_new_i64();
2690 
2691     vfp_load_reg64(vm, a->vm);
2692     gen_helper_vfp_fcvt_f64_to_f16(tmp, vm, fpst, ahp_mode);
2693     tcg_gen_st16_i32(tmp, cpu_env, vfp_f16_offset(a->vd, a->t));
2694     return true;
2695 }
2696 
2697 static bool trans_VRINTR_hp(DisasContext *s, arg_VRINTR_sp *a)
2698 {
2699     TCGv_ptr fpst;
2700     TCGv_i32 tmp;
2701 
2702     if (!dc_isar_feature(aa32_fp16_arith, s)) {
2703         return false;
2704     }
2705 
2706     if (!vfp_access_check(s)) {
2707         return true;
2708     }
2709 
2710     tmp = tcg_temp_new_i32();
2711     vfp_load_reg32(tmp, a->vm);
2712     fpst = fpstatus_ptr(FPST_FPCR_F16);
2713     gen_helper_rinth(tmp, tmp, fpst);
2714     vfp_store_reg32(tmp, a->vd);
2715     return true;
2716 }
2717 
2718 static bool trans_VRINTR_sp(DisasContext *s, arg_VRINTR_sp *a)
2719 {
2720     TCGv_ptr fpst;
2721     TCGv_i32 tmp;
2722 
2723     if (!dc_isar_feature(aa32_vrint, s)) {
2724         return false;
2725     }
2726 
2727     if (!vfp_access_check(s)) {
2728         return true;
2729     }
2730 
2731     tmp = tcg_temp_new_i32();
2732     vfp_load_reg32(tmp, a->vm);
2733     fpst = fpstatus_ptr(FPST_FPCR);
2734     gen_helper_rints(tmp, tmp, fpst);
2735     vfp_store_reg32(tmp, a->vd);
2736     return true;
2737 }
2738 
2739 static bool trans_VRINTR_dp(DisasContext *s, arg_VRINTR_dp *a)
2740 {
2741     TCGv_ptr fpst;
2742     TCGv_i64 tmp;
2743 
2744     if (!dc_isar_feature(aa32_fpdp_v2, s)) {
2745         return false;
2746     }
2747 
2748     if (!dc_isar_feature(aa32_vrint, s)) {
2749         return false;
2750     }
2751 
2752     /* UNDEF accesses to D16-D31 if they don't exist. */
2753     if (!dc_isar_feature(aa32_simd_r32, s) && ((a->vd | a->vm) & 0x10)) {
2754         return false;
2755     }
2756 
2757     if (!vfp_access_check(s)) {
2758         return true;
2759     }
2760 
2761     tmp = tcg_temp_new_i64();
2762     vfp_load_reg64(tmp, a->vm);
2763     fpst = fpstatus_ptr(FPST_FPCR);
2764     gen_helper_rintd(tmp, tmp, fpst);
2765     vfp_store_reg64(tmp, a->vd);
2766     return true;
2767 }
2768 
2769 static bool trans_VRINTZ_hp(DisasContext *s, arg_VRINTZ_sp *a)
2770 {
2771     TCGv_ptr fpst;
2772     TCGv_i32 tmp;
2773     TCGv_i32 tcg_rmode;
2774 
2775     if (!dc_isar_feature(aa32_fp16_arith, s)) {
2776         return false;
2777     }
2778 
2779     if (!vfp_access_check(s)) {
2780         return true;
2781     }
2782 
2783     tmp = tcg_temp_new_i32();
2784     vfp_load_reg32(tmp, a->vm);
2785     fpst = fpstatus_ptr(FPST_FPCR_F16);
2786     tcg_rmode = tcg_const_i32(float_round_to_zero);
2787     gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
2788     gen_helper_rinth(tmp, tmp, fpst);
2789     gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
2790     vfp_store_reg32(tmp, a->vd);
2791     return true;
2792 }
2793 
2794 static bool trans_VRINTZ_sp(DisasContext *s, arg_VRINTZ_sp *a)
2795 {
2796     TCGv_ptr fpst;
2797     TCGv_i32 tmp;
2798     TCGv_i32 tcg_rmode;
2799 
2800     if (!dc_isar_feature(aa32_vrint, s)) {
2801         return false;
2802     }
2803 
2804     if (!vfp_access_check(s)) {
2805         return true;
2806     }
2807 
2808     tmp = tcg_temp_new_i32();
2809     vfp_load_reg32(tmp, a->vm);
2810     fpst = fpstatus_ptr(FPST_FPCR);
2811     tcg_rmode = tcg_const_i32(float_round_to_zero);
2812     gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
2813     gen_helper_rints(tmp, tmp, fpst);
2814     gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
2815     vfp_store_reg32(tmp, a->vd);
2816     return true;
2817 }
2818 
2819 static bool trans_VRINTZ_dp(DisasContext *s, arg_VRINTZ_dp *a)
2820 {
2821     TCGv_ptr fpst;
2822     TCGv_i64 tmp;
2823     TCGv_i32 tcg_rmode;
2824 
2825     if (!dc_isar_feature(aa32_fpdp_v2, s)) {
2826         return false;
2827     }
2828 
2829     if (!dc_isar_feature(aa32_vrint, s)) {
2830         return false;
2831     }
2832 
2833     /* UNDEF accesses to D16-D31 if they don't exist. */
2834     if (!dc_isar_feature(aa32_simd_r32, s) && ((a->vd | a->vm) & 0x10)) {
2835         return false;
2836     }
2837 
2838     if (!vfp_access_check(s)) {
2839         return true;
2840     }
2841 
2842     tmp = tcg_temp_new_i64();
2843     vfp_load_reg64(tmp, a->vm);
2844     fpst = fpstatus_ptr(FPST_FPCR);
2845     tcg_rmode = tcg_const_i32(float_round_to_zero);
2846     gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
2847     gen_helper_rintd(tmp, tmp, fpst);
2848     gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst);
2849     vfp_store_reg64(tmp, a->vd);
2850     return true;
2851 }
2852 
2853 static bool trans_VRINTX_hp(DisasContext *s, arg_VRINTX_sp *a)
2854 {
2855     TCGv_ptr fpst;
2856     TCGv_i32 tmp;
2857 
2858     if (!dc_isar_feature(aa32_fp16_arith, s)) {
2859         return false;
2860     }
2861 
2862     if (!vfp_access_check(s)) {
2863         return true;
2864     }
2865 
2866     tmp = tcg_temp_new_i32();
2867     vfp_load_reg32(tmp, a->vm);
2868     fpst = fpstatus_ptr(FPST_FPCR_F16);
2869     gen_helper_rinth_exact(tmp, tmp, fpst);
2870     vfp_store_reg32(tmp, a->vd);
2871     return true;
2872 }
2873 
2874 static bool trans_VRINTX_sp(DisasContext *s, arg_VRINTX_sp *a)
2875 {
2876     TCGv_ptr fpst;
2877     TCGv_i32 tmp;
2878 
2879     if (!dc_isar_feature(aa32_vrint, s)) {
2880         return false;
2881     }
2882 
2883     if (!vfp_access_check(s)) {
2884         return true;
2885     }
2886 
2887     tmp = tcg_temp_new_i32();
2888     vfp_load_reg32(tmp, a->vm);
2889     fpst = fpstatus_ptr(FPST_FPCR);
2890     gen_helper_rints_exact(tmp, tmp, fpst);
2891     vfp_store_reg32(tmp, a->vd);
2892     return true;
2893 }
2894 
2895 static bool trans_VRINTX_dp(DisasContext *s, arg_VRINTX_dp *a)
2896 {
2897     TCGv_ptr fpst;
2898     TCGv_i64 tmp;
2899 
2900     if (!dc_isar_feature(aa32_fpdp_v2, s)) {
2901         return false;
2902     }
2903 
2904     if (!dc_isar_feature(aa32_vrint, s)) {
2905         return false;
2906     }
2907 
2908     /* UNDEF accesses to D16-D31 if they don't exist. */
2909     if (!dc_isar_feature(aa32_simd_r32, s) && ((a->vd | a->vm) & 0x10)) {
2910         return false;
2911     }
2912 
2913     if (!vfp_access_check(s)) {
2914         return true;
2915     }
2916 
2917     tmp = tcg_temp_new_i64();
2918     vfp_load_reg64(tmp, a->vm);
2919     fpst = fpstatus_ptr(FPST_FPCR);
2920     gen_helper_rintd_exact(tmp, tmp, fpst);
2921     vfp_store_reg64(tmp, a->vd);
2922     return true;
2923 }
2924 
2925 static bool trans_VCVT_sp(DisasContext *s, arg_VCVT_sp *a)
2926 {
2927     TCGv_i64 vd;
2928     TCGv_i32 vm;
2929 
2930     if (!dc_isar_feature(aa32_fpdp_v2, s)) {
2931         return false;
2932     }
2933 
2934     /* UNDEF accesses to D16-D31 if they don't exist. */
2935     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vd & 0x10)) {
2936         return false;
2937     }
2938 
2939     if (!vfp_access_check(s)) {
2940         return true;
2941     }
2942 
2943     vm = tcg_temp_new_i32();
2944     vd = tcg_temp_new_i64();
2945     vfp_load_reg32(vm, a->vm);
2946     gen_helper_vfp_fcvtds(vd, vm, cpu_env);
2947     vfp_store_reg64(vd, a->vd);
2948     return true;
2949 }
2950 
2951 static bool trans_VCVT_dp(DisasContext *s, arg_VCVT_dp *a)
2952 {
2953     TCGv_i64 vm;
2954     TCGv_i32 vd;
2955 
2956     if (!dc_isar_feature(aa32_fpdp_v2, s)) {
2957         return false;
2958     }
2959 
2960     /* UNDEF accesses to D16-D31 if they don't exist. */
2961     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vm & 0x10)) {
2962         return false;
2963     }
2964 
2965     if (!vfp_access_check(s)) {
2966         return true;
2967     }
2968 
2969     vd = tcg_temp_new_i32();
2970     vm = tcg_temp_new_i64();
2971     vfp_load_reg64(vm, a->vm);
2972     gen_helper_vfp_fcvtsd(vd, vm, cpu_env);
2973     vfp_store_reg32(vd, a->vd);
2974     return true;
2975 }
2976 
2977 static bool trans_VCVT_int_hp(DisasContext *s, arg_VCVT_int_sp *a)
2978 {
2979     TCGv_i32 vm;
2980     TCGv_ptr fpst;
2981 
2982     if (!dc_isar_feature(aa32_fp16_arith, s)) {
2983         return false;
2984     }
2985 
2986     if (!vfp_access_check(s)) {
2987         return true;
2988     }
2989 
2990     vm = tcg_temp_new_i32();
2991     vfp_load_reg32(vm, a->vm);
2992     fpst = fpstatus_ptr(FPST_FPCR_F16);
2993     if (a->s) {
2994         /* i32 -> f16 */
2995         gen_helper_vfp_sitoh(vm, vm, fpst);
2996     } else {
2997         /* u32 -> f16 */
2998         gen_helper_vfp_uitoh(vm, vm, fpst);
2999     }
3000     vfp_store_reg32(vm, a->vd);
3001     return true;
3002 }
3003 
3004 static bool trans_VCVT_int_sp(DisasContext *s, arg_VCVT_int_sp *a)
3005 {
3006     TCGv_i32 vm;
3007     TCGv_ptr fpst;
3008 
3009     if (!dc_isar_feature(aa32_fpsp_v2, s)) {
3010         return false;
3011     }
3012 
3013     if (!vfp_access_check(s)) {
3014         return true;
3015     }
3016 
3017     vm = tcg_temp_new_i32();
3018     vfp_load_reg32(vm, a->vm);
3019     fpst = fpstatus_ptr(FPST_FPCR);
3020     if (a->s) {
3021         /* i32 -> f32 */
3022         gen_helper_vfp_sitos(vm, vm, fpst);
3023     } else {
3024         /* u32 -> f32 */
3025         gen_helper_vfp_uitos(vm, vm, fpst);
3026     }
3027     vfp_store_reg32(vm, a->vd);
3028     return true;
3029 }
3030 
3031 static bool trans_VCVT_int_dp(DisasContext *s, arg_VCVT_int_dp *a)
3032 {
3033     TCGv_i32 vm;
3034     TCGv_i64 vd;
3035     TCGv_ptr fpst;
3036 
3037     if (!dc_isar_feature(aa32_fpdp_v2, s)) {
3038         return false;
3039     }
3040 
3041     /* UNDEF accesses to D16-D31 if they don't exist. */
3042     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vd & 0x10)) {
3043         return false;
3044     }
3045 
3046     if (!vfp_access_check(s)) {
3047         return true;
3048     }
3049 
3050     vm = tcg_temp_new_i32();
3051     vd = tcg_temp_new_i64();
3052     vfp_load_reg32(vm, a->vm);
3053     fpst = fpstatus_ptr(FPST_FPCR);
3054     if (a->s) {
3055         /* i32 -> f64 */
3056         gen_helper_vfp_sitod(vd, vm, fpst);
3057     } else {
3058         /* u32 -> f64 */
3059         gen_helper_vfp_uitod(vd, vm, fpst);
3060     }
3061     vfp_store_reg64(vd, a->vd);
3062     return true;
3063 }
3064 
3065 static bool trans_VJCVT(DisasContext *s, arg_VJCVT *a)
3066 {
3067     TCGv_i32 vd;
3068     TCGv_i64 vm;
3069 
3070     if (!dc_isar_feature(aa32_fpdp_v2, s)) {
3071         return false;
3072     }
3073 
3074     if (!dc_isar_feature(aa32_jscvt, s)) {
3075         return false;
3076     }
3077 
3078     /* UNDEF accesses to D16-D31 if they don't exist. */
3079     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vm & 0x10)) {
3080         return false;
3081     }
3082 
3083     if (!vfp_access_check(s)) {
3084         return true;
3085     }
3086 
3087     vm = tcg_temp_new_i64();
3088     vd = tcg_temp_new_i32();
3089     vfp_load_reg64(vm, a->vm);
3090     gen_helper_vjcvt(vd, vm, cpu_env);
3091     vfp_store_reg32(vd, a->vd);
3092     return true;
3093 }
3094 
3095 static bool trans_VCVT_fix_hp(DisasContext *s, arg_VCVT_fix_sp *a)
3096 {
3097     TCGv_i32 vd, shift;
3098     TCGv_ptr fpst;
3099     int frac_bits;
3100 
3101     if (!dc_isar_feature(aa32_fp16_arith, s)) {
3102         return false;
3103     }
3104 
3105     if (!vfp_access_check(s)) {
3106         return true;
3107     }
3108 
3109     frac_bits = (a->opc & 1) ? (32 - a->imm) : (16 - a->imm);
3110 
3111     vd = tcg_temp_new_i32();
3112     vfp_load_reg32(vd, a->vd);
3113 
3114     fpst = fpstatus_ptr(FPST_FPCR_F16);
3115     shift = tcg_constant_i32(frac_bits);
3116 
3117     /* Switch on op:U:sx bits */
3118     switch (a->opc) {
3119     case 0:
3120         gen_helper_vfp_shtoh_round_to_nearest(vd, vd, shift, fpst);
3121         break;
3122     case 1:
3123         gen_helper_vfp_sltoh_round_to_nearest(vd, vd, shift, fpst);
3124         break;
3125     case 2:
3126         gen_helper_vfp_uhtoh_round_to_nearest(vd, vd, shift, fpst);
3127         break;
3128     case 3:
3129         gen_helper_vfp_ultoh_round_to_nearest(vd, vd, shift, fpst);
3130         break;
3131     case 4:
3132         gen_helper_vfp_toshh_round_to_zero(vd, vd, shift, fpst);
3133         break;
3134     case 5:
3135         gen_helper_vfp_toslh_round_to_zero(vd, vd, shift, fpst);
3136         break;
3137     case 6:
3138         gen_helper_vfp_touhh_round_to_zero(vd, vd, shift, fpst);
3139         break;
3140     case 7:
3141         gen_helper_vfp_toulh_round_to_zero(vd, vd, shift, fpst);
3142         break;
3143     default:
3144         g_assert_not_reached();
3145     }
3146 
3147     vfp_store_reg32(vd, a->vd);
3148     return true;
3149 }
3150 
3151 static bool trans_VCVT_fix_sp(DisasContext *s, arg_VCVT_fix_sp *a)
3152 {
3153     TCGv_i32 vd, shift;
3154     TCGv_ptr fpst;
3155     int frac_bits;
3156 
3157     if (!dc_isar_feature(aa32_fpsp_v3, s)) {
3158         return false;
3159     }
3160 
3161     if (!vfp_access_check(s)) {
3162         return true;
3163     }
3164 
3165     frac_bits = (a->opc & 1) ? (32 - a->imm) : (16 - a->imm);
3166 
3167     vd = tcg_temp_new_i32();
3168     vfp_load_reg32(vd, a->vd);
3169 
3170     fpst = fpstatus_ptr(FPST_FPCR);
3171     shift = tcg_constant_i32(frac_bits);
3172 
3173     /* Switch on op:U:sx bits */
3174     switch (a->opc) {
3175     case 0:
3176         gen_helper_vfp_shtos_round_to_nearest(vd, vd, shift, fpst);
3177         break;
3178     case 1:
3179         gen_helper_vfp_sltos_round_to_nearest(vd, vd, shift, fpst);
3180         break;
3181     case 2:
3182         gen_helper_vfp_uhtos_round_to_nearest(vd, vd, shift, fpst);
3183         break;
3184     case 3:
3185         gen_helper_vfp_ultos_round_to_nearest(vd, vd, shift, fpst);
3186         break;
3187     case 4:
3188         gen_helper_vfp_toshs_round_to_zero(vd, vd, shift, fpst);
3189         break;
3190     case 5:
3191         gen_helper_vfp_tosls_round_to_zero(vd, vd, shift, fpst);
3192         break;
3193     case 6:
3194         gen_helper_vfp_touhs_round_to_zero(vd, vd, shift, fpst);
3195         break;
3196     case 7:
3197         gen_helper_vfp_touls_round_to_zero(vd, vd, shift, fpst);
3198         break;
3199     default:
3200         g_assert_not_reached();
3201     }
3202 
3203     vfp_store_reg32(vd, a->vd);
3204     return true;
3205 }
3206 
3207 static bool trans_VCVT_fix_dp(DisasContext *s, arg_VCVT_fix_dp *a)
3208 {
3209     TCGv_i64 vd;
3210     TCGv_i32 shift;
3211     TCGv_ptr fpst;
3212     int frac_bits;
3213 
3214     if (!dc_isar_feature(aa32_fpdp_v3, s)) {
3215         return false;
3216     }
3217 
3218     /* UNDEF accesses to D16-D31 if they don't exist. */
3219     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vd & 0x10)) {
3220         return false;
3221     }
3222 
3223     if (!vfp_access_check(s)) {
3224         return true;
3225     }
3226 
3227     frac_bits = (a->opc & 1) ? (32 - a->imm) : (16 - a->imm);
3228 
3229     vd = tcg_temp_new_i64();
3230     vfp_load_reg64(vd, a->vd);
3231 
3232     fpst = fpstatus_ptr(FPST_FPCR);
3233     shift = tcg_constant_i32(frac_bits);
3234 
3235     /* Switch on op:U:sx bits */
3236     switch (a->opc) {
3237     case 0:
3238         gen_helper_vfp_shtod_round_to_nearest(vd, vd, shift, fpst);
3239         break;
3240     case 1:
3241         gen_helper_vfp_sltod_round_to_nearest(vd, vd, shift, fpst);
3242         break;
3243     case 2:
3244         gen_helper_vfp_uhtod_round_to_nearest(vd, vd, shift, fpst);
3245         break;
3246     case 3:
3247         gen_helper_vfp_ultod_round_to_nearest(vd, vd, shift, fpst);
3248         break;
3249     case 4:
3250         gen_helper_vfp_toshd_round_to_zero(vd, vd, shift, fpst);
3251         break;
3252     case 5:
3253         gen_helper_vfp_tosld_round_to_zero(vd, vd, shift, fpst);
3254         break;
3255     case 6:
3256         gen_helper_vfp_touhd_round_to_zero(vd, vd, shift, fpst);
3257         break;
3258     case 7:
3259         gen_helper_vfp_tould_round_to_zero(vd, vd, shift, fpst);
3260         break;
3261     default:
3262         g_assert_not_reached();
3263     }
3264 
3265     vfp_store_reg64(vd, a->vd);
3266     return true;
3267 }
3268 
3269 static bool trans_VCVT_hp_int(DisasContext *s, arg_VCVT_sp_int *a)
3270 {
3271     TCGv_i32 vm;
3272     TCGv_ptr fpst;
3273 
3274     if (!dc_isar_feature(aa32_fp16_arith, s)) {
3275         return false;
3276     }
3277 
3278     if (!vfp_access_check(s)) {
3279         return true;
3280     }
3281 
3282     fpst = fpstatus_ptr(FPST_FPCR_F16);
3283     vm = tcg_temp_new_i32();
3284     vfp_load_reg32(vm, a->vm);
3285 
3286     if (a->s) {
3287         if (a->rz) {
3288             gen_helper_vfp_tosizh(vm, vm, fpst);
3289         } else {
3290             gen_helper_vfp_tosih(vm, vm, fpst);
3291         }
3292     } else {
3293         if (a->rz) {
3294             gen_helper_vfp_touizh(vm, vm, fpst);
3295         } else {
3296             gen_helper_vfp_touih(vm, vm, fpst);
3297         }
3298     }
3299     vfp_store_reg32(vm, a->vd);
3300     return true;
3301 }
3302 
3303 static bool trans_VCVT_sp_int(DisasContext *s, arg_VCVT_sp_int *a)
3304 {
3305     TCGv_i32 vm;
3306     TCGv_ptr fpst;
3307 
3308     if (!dc_isar_feature(aa32_fpsp_v2, s)) {
3309         return false;
3310     }
3311 
3312     if (!vfp_access_check(s)) {
3313         return true;
3314     }
3315 
3316     fpst = fpstatus_ptr(FPST_FPCR);
3317     vm = tcg_temp_new_i32();
3318     vfp_load_reg32(vm, a->vm);
3319 
3320     if (a->s) {
3321         if (a->rz) {
3322             gen_helper_vfp_tosizs(vm, vm, fpst);
3323         } else {
3324             gen_helper_vfp_tosis(vm, vm, fpst);
3325         }
3326     } else {
3327         if (a->rz) {
3328             gen_helper_vfp_touizs(vm, vm, fpst);
3329         } else {
3330             gen_helper_vfp_touis(vm, vm, fpst);
3331         }
3332     }
3333     vfp_store_reg32(vm, a->vd);
3334     return true;
3335 }
3336 
3337 static bool trans_VCVT_dp_int(DisasContext *s, arg_VCVT_dp_int *a)
3338 {
3339     TCGv_i32 vd;
3340     TCGv_i64 vm;
3341     TCGv_ptr fpst;
3342 
3343     if (!dc_isar_feature(aa32_fpdp_v2, s)) {
3344         return false;
3345     }
3346 
3347     /* UNDEF accesses to D16-D31 if they don't exist. */
3348     if (!dc_isar_feature(aa32_simd_r32, s) && (a->vm & 0x10)) {
3349         return false;
3350     }
3351 
3352     if (!vfp_access_check(s)) {
3353         return true;
3354     }
3355 
3356     fpst = fpstatus_ptr(FPST_FPCR);
3357     vm = tcg_temp_new_i64();
3358     vd = tcg_temp_new_i32();
3359     vfp_load_reg64(vm, a->vm);
3360 
3361     if (a->s) {
3362         if (a->rz) {
3363             gen_helper_vfp_tosizd(vd, vm, fpst);
3364         } else {
3365             gen_helper_vfp_tosid(vd, vm, fpst);
3366         }
3367     } else {
3368         if (a->rz) {
3369             gen_helper_vfp_touizd(vd, vm, fpst);
3370         } else {
3371             gen_helper_vfp_touid(vd, vm, fpst);
3372         }
3373     }
3374     vfp_store_reg32(vd, a->vd);
3375     return true;
3376 }
3377 
3378 static bool trans_VINS(DisasContext *s, arg_VINS *a)
3379 {
3380     TCGv_i32 rd, rm;
3381 
3382     if (!dc_isar_feature(aa32_fp16_arith, s)) {
3383         return false;
3384     }
3385 
3386     if (s->vec_len != 0 || s->vec_stride != 0) {
3387         return false;
3388     }
3389 
3390     if (!vfp_access_check(s)) {
3391         return true;
3392     }
3393 
3394     /* Insert low half of Vm into high half of Vd */
3395     rm = tcg_temp_new_i32();
3396     rd = tcg_temp_new_i32();
3397     vfp_load_reg32(rm, a->vm);
3398     vfp_load_reg32(rd, a->vd);
3399     tcg_gen_deposit_i32(rd, rd, rm, 16, 16);
3400     vfp_store_reg32(rd, a->vd);
3401     return true;
3402 }
3403 
3404 static bool trans_VMOVX(DisasContext *s, arg_VINS *a)
3405 {
3406     TCGv_i32 rm;
3407 
3408     if (!dc_isar_feature(aa32_fp16_arith, s)) {
3409         return false;
3410     }
3411 
3412     if (s->vec_len != 0 || s->vec_stride != 0) {
3413         return false;
3414     }
3415 
3416     if (!vfp_access_check(s)) {
3417         return true;
3418     }
3419 
3420     /* Set Vd to high half of Vm */
3421     rm = tcg_temp_new_i32();
3422     vfp_load_reg32(rm, a->vm);
3423     tcg_gen_shri_i32(rm, rm, 16);
3424     vfp_store_reg32(rm, a->vd);
3425     return true;
3426 }
3427