4 * Copyright (c) 2013 Alexander Graf <agraf@suse.de>
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
28 #include "translate.h"
29 #include "internals.h"
30 #include "qemu/host-utils.h"
32 #include "exec/gen-icount.h"
38 static TCGv_i64 cpu_X[32];
39 static TCGv_i64 cpu_pc;
40 static TCGv_i32 cpu_NF, cpu_ZF, cpu_CF, cpu_VF;
42 /* Load/store exclusive handling */
43 static TCGv_i64 cpu_exclusive_addr;
44 static TCGv_i64 cpu_exclusive_val;
45 static TCGv_i64 cpu_exclusive_high;
46 #ifdef CONFIG_USER_ONLY
47 static TCGv_i64 cpu_exclusive_test;
48 static TCGv_i32 cpu_exclusive_info;
51 static const char *regnames[] = {
52 "x0", "x1", "x2", "x3", "x4", "x5", "x6", "x7",
53 "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15",
54 "x16", "x17", "x18", "x19", "x20", "x21", "x22", "x23",
55 "x24", "x25", "x26", "x27", "x28", "x29", "lr", "sp"
59 A64_SHIFT_TYPE_LSL = 0,
60 A64_SHIFT_TYPE_LSR = 1,
61 A64_SHIFT_TYPE_ASR = 2,
62 A64_SHIFT_TYPE_ROR = 3
65 /* Table based decoder typedefs - used when the relevant bits for decode
66 * are too awkwardly scattered across the instruction (eg SIMD).
68 typedef void AArch64DecodeFn(DisasContext *s, uint32_t insn);
70 typedef struct AArch64DecodeTable {
73 AArch64DecodeFn *disas_fn;
76 /* Function prototype for gen_ functions for calling Neon helpers */
77 typedef void NeonGenOneOpEnvFn(TCGv_i32, TCGv_ptr, TCGv_i32);
78 typedef void NeonGenTwoOpFn(TCGv_i32, TCGv_i32, TCGv_i32);
79 typedef void NeonGenTwoOpEnvFn(TCGv_i32, TCGv_ptr, TCGv_i32, TCGv_i32);
80 typedef void NeonGenTwo64OpFn(TCGv_i64, TCGv_i64, TCGv_i64);
81 typedef void NeonGenTwo64OpEnvFn(TCGv_i64, TCGv_ptr, TCGv_i64, TCGv_i64);
82 typedef void NeonGenNarrowFn(TCGv_i32, TCGv_i64);
83 typedef void NeonGenNarrowEnvFn(TCGv_i32, TCGv_ptr, TCGv_i64);
84 typedef void NeonGenWidenFn(TCGv_i64, TCGv_i32);
85 typedef void NeonGenTwoSingleOPFn(TCGv_i32, TCGv_i32, TCGv_i32, TCGv_ptr);
86 typedef void NeonGenTwoDoubleOPFn(TCGv_i64, TCGv_i64, TCGv_i64, TCGv_ptr);
87 typedef void NeonGenOneOpFn(TCGv_i64, TCGv_i64);
89 /* initialize TCG globals. */
90 void a64_translate_init(void)
94 cpu_pc = tcg_global_mem_new_i64(TCG_AREG0,
95 offsetof(CPUARMState, pc),
97 for (i = 0; i < 32; i++) {
98 cpu_X[i] = tcg_global_mem_new_i64(TCG_AREG0,
99 offsetof(CPUARMState, xregs[i]),
103 cpu_NF = tcg_global_mem_new_i32(TCG_AREG0, offsetof(CPUARMState, NF), "NF");
104 cpu_ZF = tcg_global_mem_new_i32(TCG_AREG0, offsetof(CPUARMState, ZF), "ZF");
105 cpu_CF = tcg_global_mem_new_i32(TCG_AREG0, offsetof(CPUARMState, CF), "CF");
106 cpu_VF = tcg_global_mem_new_i32(TCG_AREG0, offsetof(CPUARMState, VF), "VF");
108 cpu_exclusive_addr = tcg_global_mem_new_i64(TCG_AREG0,
109 offsetof(CPUARMState, exclusive_addr), "exclusive_addr");
110 cpu_exclusive_val = tcg_global_mem_new_i64(TCG_AREG0,
111 offsetof(CPUARMState, exclusive_val), "exclusive_val");
112 cpu_exclusive_high = tcg_global_mem_new_i64(TCG_AREG0,
113 offsetof(CPUARMState, exclusive_high), "exclusive_high");
114 #ifdef CONFIG_USER_ONLY
115 cpu_exclusive_test = tcg_global_mem_new_i64(TCG_AREG0,
116 offsetof(CPUARMState, exclusive_test), "exclusive_test");
117 cpu_exclusive_info = tcg_global_mem_new_i32(TCG_AREG0,
118 offsetof(CPUARMState, exclusive_info), "exclusive_info");
122 void aarch64_cpu_dump_state(CPUState *cs, FILE *f,
123 fprintf_function cpu_fprintf, int flags)
125 ARMCPU *cpu = ARM_CPU(cs);
126 CPUARMState *env = &cpu->env;
127 uint32_t psr = pstate_read(env);
130 cpu_fprintf(f, "PC=%016"PRIx64" SP=%016"PRIx64"\n",
131 env->pc, env->xregs[31]);
132 for (i = 0; i < 31; i++) {
133 cpu_fprintf(f, "X%02d=%016"PRIx64, i, env->xregs[i]);
135 cpu_fprintf(f, "\n");
140 cpu_fprintf(f, "PSTATE=%08x (flags %c%c%c%c)\n",
142 psr & PSTATE_N ? 'N' : '-',
143 psr & PSTATE_Z ? 'Z' : '-',
144 psr & PSTATE_C ? 'C' : '-',
145 psr & PSTATE_V ? 'V' : '-');
146 cpu_fprintf(f, "\n");
148 if (flags & CPU_DUMP_FPU) {
150 for (i = 0; i < numvfpregs; i += 2) {
151 uint64_t vlo = float64_val(env->vfp.regs[i * 2]);
152 uint64_t vhi = float64_val(env->vfp.regs[(i * 2) + 1]);
153 cpu_fprintf(f, "q%02d=%016" PRIx64 ":%016" PRIx64 " ",
155 vlo = float64_val(env->vfp.regs[(i + 1) * 2]);
156 vhi = float64_val(env->vfp.regs[((i + 1) * 2) + 1]);
157 cpu_fprintf(f, "q%02d=%016" PRIx64 ":%016" PRIx64 "\n",
160 cpu_fprintf(f, "FPCR: %08x FPSR: %08x\n",
161 vfp_get_fpcr(env), vfp_get_fpsr(env));
165 void gen_a64_set_pc_im(uint64_t val)
167 tcg_gen_movi_i64(cpu_pc, val);
170 static void gen_exception_internal(int excp)
172 TCGv_i32 tcg_excp = tcg_const_i32(excp);
174 assert(excp_is_internal(excp));
175 gen_helper_exception_internal(cpu_env, tcg_excp);
176 tcg_temp_free_i32(tcg_excp);
179 static void gen_exception(int excp, uint32_t syndrome)
181 TCGv_i32 tcg_excp = tcg_const_i32(excp);
182 TCGv_i32 tcg_syn = tcg_const_i32(syndrome);
184 gen_helper_exception_with_syndrome(cpu_env, tcg_excp, tcg_syn);
185 tcg_temp_free_i32(tcg_syn);
186 tcg_temp_free_i32(tcg_excp);
189 static void gen_exception_internal_insn(DisasContext *s, int offset, int excp)
191 gen_a64_set_pc_im(s->pc - offset);
192 gen_exception_internal(excp);
193 s->is_jmp = DISAS_EXC;
196 static void gen_exception_insn(DisasContext *s, int offset, int excp,
199 gen_a64_set_pc_im(s->pc - offset);
200 gen_exception(excp, syndrome);
201 s->is_jmp = DISAS_EXC;
204 static inline bool use_goto_tb(DisasContext *s, int n, uint64_t dest)
206 /* No direct tb linking with singlestep or deterministic io */
207 if (s->singlestep_enabled || (s->tb->cflags & CF_LAST_IO)) {
211 /* Only link tbs from inside the same guest page */
212 if ((s->tb->pc & TARGET_PAGE_MASK) != (dest & TARGET_PAGE_MASK)) {
219 static inline void gen_goto_tb(DisasContext *s, int n, uint64_t dest)
221 TranslationBlock *tb;
224 if (use_goto_tb(s, n, dest)) {
226 gen_a64_set_pc_im(dest);
227 tcg_gen_exit_tb((intptr_t)tb + n);
228 s->is_jmp = DISAS_TB_JUMP;
230 gen_a64_set_pc_im(dest);
231 if (s->singlestep_enabled) {
232 gen_exception_internal(EXCP_DEBUG);
235 s->is_jmp = DISAS_JUMP;
239 static void unallocated_encoding(DisasContext *s)
241 /* Unallocated and reserved encodings are uncategorized */
242 gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized());
245 #define unsupported_encoding(s, insn) \
247 qemu_log_mask(LOG_UNIMP, \
248 "%s:%d: unsupported instruction encoding 0x%08x " \
249 "at pc=%016" PRIx64 "\n", \
250 __FILE__, __LINE__, insn, s->pc - 4); \
251 unallocated_encoding(s); \
254 static void init_tmp_a64_array(DisasContext *s)
256 #ifdef CONFIG_DEBUG_TCG
258 for (i = 0; i < ARRAY_SIZE(s->tmp_a64); i++) {
259 TCGV_UNUSED_I64(s->tmp_a64[i]);
262 s->tmp_a64_count = 0;
265 static void free_tmp_a64(DisasContext *s)
268 for (i = 0; i < s->tmp_a64_count; i++) {
269 tcg_temp_free_i64(s->tmp_a64[i]);
271 init_tmp_a64_array(s);
274 static TCGv_i64 new_tmp_a64(DisasContext *s)
276 assert(s->tmp_a64_count < TMP_A64_MAX);
277 return s->tmp_a64[s->tmp_a64_count++] = tcg_temp_new_i64();
280 static TCGv_i64 new_tmp_a64_zero(DisasContext *s)
282 TCGv_i64 t = new_tmp_a64(s);
283 tcg_gen_movi_i64(t, 0);
288 * Register access functions
290 * These functions are used for directly accessing a register in where
291 * changes to the final register value are likely to be made. If you
292 * need to use a register for temporary calculation (e.g. index type
293 * operations) use the read_* form.
295 * B1.2.1 Register mappings
297 * In instruction register encoding 31 can refer to ZR (zero register) or
298 * the SP (stack pointer) depending on context. In QEMU's case we map SP
299 * to cpu_X[31] and ZR accesses to a temporary which can be discarded.
300 * This is the point of the _sp forms.
302 static TCGv_i64 cpu_reg(DisasContext *s, int reg)
305 return new_tmp_a64_zero(s);
311 /* register access for when 31 == SP */
312 static TCGv_i64 cpu_reg_sp(DisasContext *s, int reg)
317 /* read a cpu register in 32bit/64bit mode. Returns a TCGv_i64
318 * representing the register contents. This TCGv is an auto-freed
319 * temporary so it need not be explicitly freed, and may be modified.
321 static TCGv_i64 read_cpu_reg(DisasContext *s, int reg, int sf)
323 TCGv_i64 v = new_tmp_a64(s);
326 tcg_gen_mov_i64(v, cpu_X[reg]);
328 tcg_gen_ext32u_i64(v, cpu_X[reg]);
331 tcg_gen_movi_i64(v, 0);
336 static TCGv_i64 read_cpu_reg_sp(DisasContext *s, int reg, int sf)
338 TCGv_i64 v = new_tmp_a64(s);
340 tcg_gen_mov_i64(v, cpu_X[reg]);
342 tcg_gen_ext32u_i64(v, cpu_X[reg]);
347 /* We should have at some point before trying to access an FP register
348 * done the necessary access check, so assert that
349 * (a) we did the check and
350 * (b) we didn't then just plough ahead anyway if it failed.
351 * Print the instruction pattern in the abort message so we can figure
352 * out what we need to fix if a user encounters this problem in the wild.
354 static inline void assert_fp_access_checked(DisasContext *s)
356 #ifdef CONFIG_DEBUG_TCG
357 if (unlikely(!s->fp_access_checked || !s->cpacr_fpen)) {
358 fprintf(stderr, "target-arm: FP access check missing for "
359 "instruction 0x%08x\n", s->insn);
365 /* Return the offset into CPUARMState of an element of specified
366 * size, 'element' places in from the least significant end of
367 * the FP/vector register Qn.
369 static inline int vec_reg_offset(DisasContext *s, int regno,
370 int element, TCGMemOp size)
372 int offs = offsetof(CPUARMState, vfp.regs[regno * 2]);
373 #ifdef HOST_WORDS_BIGENDIAN
374 /* This is complicated slightly because vfp.regs[2n] is
375 * still the low half and vfp.regs[2n+1] the high half
376 * of the 128 bit vector, even on big endian systems.
377 * Calculate the offset assuming a fully bigendian 128 bits,
378 * then XOR to account for the order of the two 64 bit halves.
380 offs += (16 - ((element + 1) * (1 << size)));
383 offs += element * (1 << size);
385 assert_fp_access_checked(s);
389 /* Return the offset into CPUARMState of a slice (from
390 * the least significant end) of FP register Qn (ie
392 * (Note that this is not the same mapping as for A32; see cpu.h)
394 static inline int fp_reg_offset(DisasContext *s, int regno, TCGMemOp size)
396 int offs = offsetof(CPUARMState, vfp.regs[regno * 2]);
397 #ifdef HOST_WORDS_BIGENDIAN
398 offs += (8 - (1 << size));
400 assert_fp_access_checked(s);
404 /* Offset of the high half of the 128 bit vector Qn */
405 static inline int fp_reg_hi_offset(DisasContext *s, int regno)
407 assert_fp_access_checked(s);
408 return offsetof(CPUARMState, vfp.regs[regno * 2 + 1]);
411 /* Convenience accessors for reading and writing single and double
412 * FP registers. Writing clears the upper parts of the associated
413 * 128 bit vector register, as required by the architecture.
414 * Note that unlike the GP register accessors, the values returned
415 * by the read functions must be manually freed.
417 static TCGv_i64 read_fp_dreg(DisasContext *s, int reg)
419 TCGv_i64 v = tcg_temp_new_i64();
421 tcg_gen_ld_i64(v, cpu_env, fp_reg_offset(s, reg, MO_64));
425 static TCGv_i32 read_fp_sreg(DisasContext *s, int reg)
427 TCGv_i32 v = tcg_temp_new_i32();
429 tcg_gen_ld_i32(v, cpu_env, fp_reg_offset(s, reg, MO_32));
433 static void write_fp_dreg(DisasContext *s, int reg, TCGv_i64 v)
435 TCGv_i64 tcg_zero = tcg_const_i64(0);
437 tcg_gen_st_i64(v, cpu_env, fp_reg_offset(s, reg, MO_64));
438 tcg_gen_st_i64(tcg_zero, cpu_env, fp_reg_hi_offset(s, reg));
439 tcg_temp_free_i64(tcg_zero);
442 static void write_fp_sreg(DisasContext *s, int reg, TCGv_i32 v)
444 TCGv_i64 tmp = tcg_temp_new_i64();
446 tcg_gen_extu_i32_i64(tmp, v);
447 write_fp_dreg(s, reg, tmp);
448 tcg_temp_free_i64(tmp);
451 static TCGv_ptr get_fpstatus_ptr(void)
453 TCGv_ptr statusptr = tcg_temp_new_ptr();
456 /* In A64 all instructions (both FP and Neon) use the FPCR;
457 * there is no equivalent of the A32 Neon "standard FPSCR value"
458 * and all operations use vfp.fp_status.
460 offset = offsetof(CPUARMState, vfp.fp_status);
461 tcg_gen_addi_ptr(statusptr, cpu_env, offset);
465 /* Set ZF and NF based on a 64 bit result. This is alas fiddlier
466 * than the 32 bit equivalent.
468 static inline void gen_set_NZ64(TCGv_i64 result)
470 TCGv_i64 flag = tcg_temp_new_i64();
472 tcg_gen_setcondi_i64(TCG_COND_NE, flag, result, 0);
473 tcg_gen_trunc_i64_i32(cpu_ZF, flag);
474 tcg_gen_shri_i64(flag, result, 32);
475 tcg_gen_trunc_i64_i32(cpu_NF, flag);
476 tcg_temp_free_i64(flag);
479 /* Set NZCV as for a logical operation: NZ as per result, CV cleared. */
480 static inline void gen_logic_CC(int sf, TCGv_i64 result)
483 gen_set_NZ64(result);
485 tcg_gen_trunc_i64_i32(cpu_ZF, result);
486 tcg_gen_trunc_i64_i32(cpu_NF, result);
488 tcg_gen_movi_i32(cpu_CF, 0);
489 tcg_gen_movi_i32(cpu_VF, 0);
492 /* dest = T0 + T1; compute C, N, V and Z flags */
493 static void gen_add_CC(int sf, TCGv_i64 dest, TCGv_i64 t0, TCGv_i64 t1)
496 TCGv_i64 result, flag, tmp;
497 result = tcg_temp_new_i64();
498 flag = tcg_temp_new_i64();
499 tmp = tcg_temp_new_i64();
501 tcg_gen_movi_i64(tmp, 0);
502 tcg_gen_add2_i64(result, flag, t0, tmp, t1, tmp);
504 tcg_gen_trunc_i64_i32(cpu_CF, flag);
506 gen_set_NZ64(result);
508 tcg_gen_xor_i64(flag, result, t0);
509 tcg_gen_xor_i64(tmp, t0, t1);
510 tcg_gen_andc_i64(flag, flag, tmp);
511 tcg_temp_free_i64(tmp);
512 tcg_gen_shri_i64(flag, flag, 32);
513 tcg_gen_trunc_i64_i32(cpu_VF, flag);
515 tcg_gen_mov_i64(dest, result);
516 tcg_temp_free_i64(result);
517 tcg_temp_free_i64(flag);
519 /* 32 bit arithmetic */
520 TCGv_i32 t0_32 = tcg_temp_new_i32();
521 TCGv_i32 t1_32 = tcg_temp_new_i32();
522 TCGv_i32 tmp = tcg_temp_new_i32();
524 tcg_gen_movi_i32(tmp, 0);
525 tcg_gen_trunc_i64_i32(t0_32, t0);
526 tcg_gen_trunc_i64_i32(t1_32, t1);
527 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0_32, tmp, t1_32, tmp);
528 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
529 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0_32);
530 tcg_gen_xor_i32(tmp, t0_32, t1_32);
531 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
532 tcg_gen_extu_i32_i64(dest, cpu_NF);
534 tcg_temp_free_i32(tmp);
535 tcg_temp_free_i32(t0_32);
536 tcg_temp_free_i32(t1_32);
540 /* dest = T0 - T1; compute C, N, V and Z flags */
541 static void gen_sub_CC(int sf, TCGv_i64 dest, TCGv_i64 t0, TCGv_i64 t1)
544 /* 64 bit arithmetic */
545 TCGv_i64 result, flag, tmp;
547 result = tcg_temp_new_i64();
548 flag = tcg_temp_new_i64();
549 tcg_gen_sub_i64(result, t0, t1);
551 gen_set_NZ64(result);
553 tcg_gen_setcond_i64(TCG_COND_GEU, flag, t0, t1);
554 tcg_gen_trunc_i64_i32(cpu_CF, flag);
556 tcg_gen_xor_i64(flag, result, t0);
557 tmp = tcg_temp_new_i64();
558 tcg_gen_xor_i64(tmp, t0, t1);
559 tcg_gen_and_i64(flag, flag, tmp);
560 tcg_temp_free_i64(tmp);
561 tcg_gen_shri_i64(flag, flag, 32);
562 tcg_gen_trunc_i64_i32(cpu_VF, flag);
563 tcg_gen_mov_i64(dest, result);
564 tcg_temp_free_i64(flag);
565 tcg_temp_free_i64(result);
567 /* 32 bit arithmetic */
568 TCGv_i32 t0_32 = tcg_temp_new_i32();
569 TCGv_i32 t1_32 = tcg_temp_new_i32();
572 tcg_gen_trunc_i64_i32(t0_32, t0);
573 tcg_gen_trunc_i64_i32(t1_32, t1);
574 tcg_gen_sub_i32(cpu_NF, t0_32, t1_32);
575 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
576 tcg_gen_setcond_i32(TCG_COND_GEU, cpu_CF, t0_32, t1_32);
577 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0_32);
578 tmp = tcg_temp_new_i32();
579 tcg_gen_xor_i32(tmp, t0_32, t1_32);
580 tcg_temp_free_i32(t0_32);
581 tcg_temp_free_i32(t1_32);
582 tcg_gen_and_i32(cpu_VF, cpu_VF, tmp);
583 tcg_temp_free_i32(tmp);
584 tcg_gen_extu_i32_i64(dest, cpu_NF);
588 /* dest = T0 + T1 + CF; do not compute flags. */
589 static void gen_adc(int sf, TCGv_i64 dest, TCGv_i64 t0, TCGv_i64 t1)
591 TCGv_i64 flag = tcg_temp_new_i64();
592 tcg_gen_extu_i32_i64(flag, cpu_CF);
593 tcg_gen_add_i64(dest, t0, t1);
594 tcg_gen_add_i64(dest, dest, flag);
595 tcg_temp_free_i64(flag);
598 tcg_gen_ext32u_i64(dest, dest);
602 /* dest = T0 + T1 + CF; compute C, N, V and Z flags. */
603 static void gen_adc_CC(int sf, TCGv_i64 dest, TCGv_i64 t0, TCGv_i64 t1)
606 TCGv_i64 result, cf_64, vf_64, tmp;
607 result = tcg_temp_new_i64();
608 cf_64 = tcg_temp_new_i64();
609 vf_64 = tcg_temp_new_i64();
610 tmp = tcg_const_i64(0);
612 tcg_gen_extu_i32_i64(cf_64, cpu_CF);
613 tcg_gen_add2_i64(result, cf_64, t0, tmp, cf_64, tmp);
614 tcg_gen_add2_i64(result, cf_64, result, cf_64, t1, tmp);
615 tcg_gen_trunc_i64_i32(cpu_CF, cf_64);
616 gen_set_NZ64(result);
618 tcg_gen_xor_i64(vf_64, result, t0);
619 tcg_gen_xor_i64(tmp, t0, t1);
620 tcg_gen_andc_i64(vf_64, vf_64, tmp);
621 tcg_gen_shri_i64(vf_64, vf_64, 32);
622 tcg_gen_trunc_i64_i32(cpu_VF, vf_64);
624 tcg_gen_mov_i64(dest, result);
626 tcg_temp_free_i64(tmp);
627 tcg_temp_free_i64(vf_64);
628 tcg_temp_free_i64(cf_64);
629 tcg_temp_free_i64(result);
631 TCGv_i32 t0_32, t1_32, tmp;
632 t0_32 = tcg_temp_new_i32();
633 t1_32 = tcg_temp_new_i32();
634 tmp = tcg_const_i32(0);
636 tcg_gen_trunc_i64_i32(t0_32, t0);
637 tcg_gen_trunc_i64_i32(t1_32, t1);
638 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0_32, tmp, cpu_CF, tmp);
639 tcg_gen_add2_i32(cpu_NF, cpu_CF, cpu_NF, cpu_CF, t1_32, tmp);
641 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
642 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0_32);
643 tcg_gen_xor_i32(tmp, t0_32, t1_32);
644 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
645 tcg_gen_extu_i32_i64(dest, cpu_NF);
647 tcg_temp_free_i32(tmp);
648 tcg_temp_free_i32(t1_32);
649 tcg_temp_free_i32(t0_32);
654 * Load/Store generators
658 * Store from GPR register to memory.
660 static void do_gpr_st_memidx(DisasContext *s, TCGv_i64 source,
661 TCGv_i64 tcg_addr, int size, int memidx)
664 tcg_gen_qemu_st_i64(source, tcg_addr, memidx, MO_TE + size);
667 static void do_gpr_st(DisasContext *s, TCGv_i64 source,
668 TCGv_i64 tcg_addr, int size)
670 do_gpr_st_memidx(s, source, tcg_addr, size, get_mem_index(s));
674 * Load from memory to GPR register
676 static void do_gpr_ld_memidx(DisasContext *s, TCGv_i64 dest, TCGv_i64 tcg_addr,
677 int size, bool is_signed, bool extend, int memidx)
679 TCGMemOp memop = MO_TE + size;
687 tcg_gen_qemu_ld_i64(dest, tcg_addr, memidx, memop);
689 if (extend && is_signed) {
691 tcg_gen_ext32u_i64(dest, dest);
695 static void do_gpr_ld(DisasContext *s, TCGv_i64 dest, TCGv_i64 tcg_addr,
696 int size, bool is_signed, bool extend)
698 do_gpr_ld_memidx(s, dest, tcg_addr, size, is_signed, extend,
703 * Store from FP register to memory
705 static void do_fp_st(DisasContext *s, int srcidx, TCGv_i64 tcg_addr, int size)
707 /* This writes the bottom N bits of a 128 bit wide vector to memory */
708 TCGv_i64 tmp = tcg_temp_new_i64();
709 tcg_gen_ld_i64(tmp, cpu_env, fp_reg_offset(s, srcidx, MO_64));
711 tcg_gen_qemu_st_i64(tmp, tcg_addr, get_mem_index(s), MO_TE + size);
713 TCGv_i64 tcg_hiaddr = tcg_temp_new_i64();
714 tcg_gen_qemu_st_i64(tmp, tcg_addr, get_mem_index(s), MO_TEQ);
715 tcg_gen_qemu_st64(tmp, tcg_addr, get_mem_index(s));
716 tcg_gen_ld_i64(tmp, cpu_env, fp_reg_hi_offset(s, srcidx));
717 tcg_gen_addi_i64(tcg_hiaddr, tcg_addr, 8);
718 tcg_gen_qemu_st_i64(tmp, tcg_hiaddr, get_mem_index(s), MO_TEQ);
719 tcg_temp_free_i64(tcg_hiaddr);
722 tcg_temp_free_i64(tmp);
726 * Load from memory to FP register
728 static void do_fp_ld(DisasContext *s, int destidx, TCGv_i64 tcg_addr, int size)
730 /* This always zero-extends and writes to a full 128 bit wide vector */
731 TCGv_i64 tmplo = tcg_temp_new_i64();
735 TCGMemOp memop = MO_TE + size;
736 tmphi = tcg_const_i64(0);
737 tcg_gen_qemu_ld_i64(tmplo, tcg_addr, get_mem_index(s), memop);
740 tmphi = tcg_temp_new_i64();
741 tcg_hiaddr = tcg_temp_new_i64();
743 tcg_gen_qemu_ld_i64(tmplo, tcg_addr, get_mem_index(s), MO_TEQ);
744 tcg_gen_addi_i64(tcg_hiaddr, tcg_addr, 8);
745 tcg_gen_qemu_ld_i64(tmphi, tcg_hiaddr, get_mem_index(s), MO_TEQ);
746 tcg_temp_free_i64(tcg_hiaddr);
749 tcg_gen_st_i64(tmplo, cpu_env, fp_reg_offset(s, destidx, MO_64));
750 tcg_gen_st_i64(tmphi, cpu_env, fp_reg_hi_offset(s, destidx));
752 tcg_temp_free_i64(tmplo);
753 tcg_temp_free_i64(tmphi);
757 * Vector load/store helpers.
759 * The principal difference between this and a FP load is that we don't
760 * zero extend as we are filling a partial chunk of the vector register.
761 * These functions don't support 128 bit loads/stores, which would be
762 * normal load/store operations.
764 * The _i32 versions are useful when operating on 32 bit quantities
765 * (eg for floating point single or using Neon helper functions).
768 /* Get value of an element within a vector register */
769 static void read_vec_element(DisasContext *s, TCGv_i64 tcg_dest, int srcidx,
770 int element, TCGMemOp memop)
772 int vect_off = vec_reg_offset(s, srcidx, element, memop & MO_SIZE);
775 tcg_gen_ld8u_i64(tcg_dest, cpu_env, vect_off);
778 tcg_gen_ld16u_i64(tcg_dest, cpu_env, vect_off);
781 tcg_gen_ld32u_i64(tcg_dest, cpu_env, vect_off);
784 tcg_gen_ld8s_i64(tcg_dest, cpu_env, vect_off);
787 tcg_gen_ld16s_i64(tcg_dest, cpu_env, vect_off);
790 tcg_gen_ld32s_i64(tcg_dest, cpu_env, vect_off);
794 tcg_gen_ld_i64(tcg_dest, cpu_env, vect_off);
797 g_assert_not_reached();
801 static void read_vec_element_i32(DisasContext *s, TCGv_i32 tcg_dest, int srcidx,
802 int element, TCGMemOp memop)
804 int vect_off = vec_reg_offset(s, srcidx, element, memop & MO_SIZE);
807 tcg_gen_ld8u_i32(tcg_dest, cpu_env, vect_off);
810 tcg_gen_ld16u_i32(tcg_dest, cpu_env, vect_off);
813 tcg_gen_ld8s_i32(tcg_dest, cpu_env, vect_off);
816 tcg_gen_ld16s_i32(tcg_dest, cpu_env, vect_off);
820 tcg_gen_ld_i32(tcg_dest, cpu_env, vect_off);
823 g_assert_not_reached();
827 /* Set value of an element within a vector register */
828 static void write_vec_element(DisasContext *s, TCGv_i64 tcg_src, int destidx,
829 int element, TCGMemOp memop)
831 int vect_off = vec_reg_offset(s, destidx, element, memop & MO_SIZE);
834 tcg_gen_st8_i64(tcg_src, cpu_env, vect_off);
837 tcg_gen_st16_i64(tcg_src, cpu_env, vect_off);
840 tcg_gen_st32_i64(tcg_src, cpu_env, vect_off);
843 tcg_gen_st_i64(tcg_src, cpu_env, vect_off);
846 g_assert_not_reached();
850 static void write_vec_element_i32(DisasContext *s, TCGv_i32 tcg_src,
851 int destidx, int element, TCGMemOp memop)
853 int vect_off = vec_reg_offset(s, destidx, element, memop & MO_SIZE);
856 tcg_gen_st8_i32(tcg_src, cpu_env, vect_off);
859 tcg_gen_st16_i32(tcg_src, cpu_env, vect_off);
862 tcg_gen_st_i32(tcg_src, cpu_env, vect_off);
865 g_assert_not_reached();
869 /* Clear the high 64 bits of a 128 bit vector (in general non-quad
870 * vector ops all need to do this).
872 static void clear_vec_high(DisasContext *s, int rd)
874 TCGv_i64 tcg_zero = tcg_const_i64(0);
876 write_vec_element(s, tcg_zero, rd, 1, MO_64);
877 tcg_temp_free_i64(tcg_zero);
880 /* Store from vector register to memory */
881 static void do_vec_st(DisasContext *s, int srcidx, int element,
882 TCGv_i64 tcg_addr, int size)
884 TCGMemOp memop = MO_TE + size;
885 TCGv_i64 tcg_tmp = tcg_temp_new_i64();
887 read_vec_element(s, tcg_tmp, srcidx, element, size);
888 tcg_gen_qemu_st_i64(tcg_tmp, tcg_addr, get_mem_index(s), memop);
890 tcg_temp_free_i64(tcg_tmp);
893 /* Load from memory to vector register */
894 static void do_vec_ld(DisasContext *s, int destidx, int element,
895 TCGv_i64 tcg_addr, int size)
897 TCGMemOp memop = MO_TE + size;
898 TCGv_i64 tcg_tmp = tcg_temp_new_i64();
900 tcg_gen_qemu_ld_i64(tcg_tmp, tcg_addr, get_mem_index(s), memop);
901 write_vec_element(s, tcg_tmp, destidx, element, size);
903 tcg_temp_free_i64(tcg_tmp);
906 /* Check that FP/Neon access is enabled. If it is, return
907 * true. If not, emit code to generate an appropriate exception,
908 * and return false; the caller should not emit any code for
909 * the instruction. Note that this check must happen after all
910 * unallocated-encoding checks (otherwise the syndrome information
911 * for the resulting exception will be incorrect).
913 static inline bool fp_access_check(DisasContext *s)
915 assert(!s->fp_access_checked);
916 s->fp_access_checked = true;
922 gen_exception_insn(s, 4, EXCP_UDEF, syn_fp_access_trap(1, 0xe, false));
927 * This utility function is for doing register extension with an
928 * optional shift. You will likely want to pass a temporary for the
929 * destination register. See DecodeRegExtend() in the ARM ARM.
931 static void ext_and_shift_reg(TCGv_i64 tcg_out, TCGv_i64 tcg_in,
932 int option, unsigned int shift)
934 int extsize = extract32(option, 0, 2);
935 bool is_signed = extract32(option, 2, 1);
940 tcg_gen_ext8s_i64(tcg_out, tcg_in);
943 tcg_gen_ext16s_i64(tcg_out, tcg_in);
946 tcg_gen_ext32s_i64(tcg_out, tcg_in);
949 tcg_gen_mov_i64(tcg_out, tcg_in);
955 tcg_gen_ext8u_i64(tcg_out, tcg_in);
958 tcg_gen_ext16u_i64(tcg_out, tcg_in);
961 tcg_gen_ext32u_i64(tcg_out, tcg_in);
964 tcg_gen_mov_i64(tcg_out, tcg_in);
970 tcg_gen_shli_i64(tcg_out, tcg_out, shift);
974 static inline void gen_check_sp_alignment(DisasContext *s)
976 /* The AArch64 architecture mandates that (if enabled via PSTATE
977 * or SCTLR bits) there is a check that SP is 16-aligned on every
978 * SP-relative load or store (with an exception generated if it is not).
979 * In line with general QEMU practice regarding misaligned accesses,
980 * we omit these checks for the sake of guest program performance.
981 * This function is provided as a hook so we can more easily add these
982 * checks in future (possibly as a "favour catching guest program bugs
983 * over speed" user selectable option).
988 * This provides a simple table based table lookup decoder. It is
989 * intended to be used when the relevant bits for decode are too
990 * awkwardly placed and switch/if based logic would be confusing and
991 * deeply nested. Since it's a linear search through the table, tables
992 * should be kept small.
994 * It returns the first handler where insn & mask == pattern, or
995 * NULL if there is no match.
996 * The table is terminated by an empty mask (i.e. 0)
998 static inline AArch64DecodeFn *lookup_disas_fn(const AArch64DecodeTable *table,
1001 const AArch64DecodeTable *tptr = table;
1003 while (tptr->mask) {
1004 if ((insn & tptr->mask) == tptr->pattern) {
1005 return tptr->disas_fn;
1013 * the instruction disassembly implemented here matches
1014 * the instruction encoding classifications in chapter 3 (C3)
1015 * of the ARM Architecture Reference Manual (DDI0487A_a)
1018 /* C3.2.7 Unconditional branch (immediate)
1020 * +----+-----------+-------------------------------------+
1021 * | op | 0 0 1 0 1 | imm26 |
1022 * +----+-----------+-------------------------------------+
1024 static void disas_uncond_b_imm(DisasContext *s, uint32_t insn)
1026 uint64_t addr = s->pc + sextract32(insn, 0, 26) * 4 - 4;
1028 if (insn & (1 << 31)) {
1029 /* C5.6.26 BL Branch with link */
1030 tcg_gen_movi_i64(cpu_reg(s, 30), s->pc);
1033 /* C5.6.20 B Branch / C5.6.26 BL Branch with link */
1034 gen_goto_tb(s, 0, addr);
1037 /* C3.2.1 Compare & branch (immediate)
1038 * 31 30 25 24 23 5 4 0
1039 * +----+-------------+----+---------------------+--------+
1040 * | sf | 0 1 1 0 1 0 | op | imm19 | Rt |
1041 * +----+-------------+----+---------------------+--------+
1043 static void disas_comp_b_imm(DisasContext *s, uint32_t insn)
1045 unsigned int sf, op, rt;
1050 sf = extract32(insn, 31, 1);
1051 op = extract32(insn, 24, 1); /* 0: CBZ; 1: CBNZ */
1052 rt = extract32(insn, 0, 5);
1053 addr = s->pc + sextract32(insn, 5, 19) * 4 - 4;
1055 tcg_cmp = read_cpu_reg(s, rt, sf);
1056 label_match = gen_new_label();
1058 tcg_gen_brcondi_i64(op ? TCG_COND_NE : TCG_COND_EQ,
1059 tcg_cmp, 0, label_match);
1061 gen_goto_tb(s, 0, s->pc);
1062 gen_set_label(label_match);
1063 gen_goto_tb(s, 1, addr);
1066 /* C3.2.5 Test & branch (immediate)
1067 * 31 30 25 24 23 19 18 5 4 0
1068 * +----+-------------+----+-------+-------------+------+
1069 * | b5 | 0 1 1 0 1 1 | op | b40 | imm14 | Rt |
1070 * +----+-------------+----+-------+-------------+------+
1072 static void disas_test_b_imm(DisasContext *s, uint32_t insn)
1074 unsigned int bit_pos, op, rt;
1079 bit_pos = (extract32(insn, 31, 1) << 5) | extract32(insn, 19, 5);
1080 op = extract32(insn, 24, 1); /* 0: TBZ; 1: TBNZ */
1081 addr = s->pc + sextract32(insn, 5, 14) * 4 - 4;
1082 rt = extract32(insn, 0, 5);
1084 tcg_cmp = tcg_temp_new_i64();
1085 tcg_gen_andi_i64(tcg_cmp, cpu_reg(s, rt), (1ULL << bit_pos));
1086 label_match = gen_new_label();
1087 tcg_gen_brcondi_i64(op ? TCG_COND_NE : TCG_COND_EQ,
1088 tcg_cmp, 0, label_match);
1089 tcg_temp_free_i64(tcg_cmp);
1090 gen_goto_tb(s, 0, s->pc);
1091 gen_set_label(label_match);
1092 gen_goto_tb(s, 1, addr);
1095 /* C3.2.2 / C5.6.19 Conditional branch (immediate)
1096 * 31 25 24 23 5 4 3 0
1097 * +---------------+----+---------------------+----+------+
1098 * | 0 1 0 1 0 1 0 | o1 | imm19 | o0 | cond |
1099 * +---------------+----+---------------------+----+------+
1101 static void disas_cond_b_imm(DisasContext *s, uint32_t insn)
1106 if ((insn & (1 << 4)) || (insn & (1 << 24))) {
1107 unallocated_encoding(s);
1110 addr = s->pc + sextract32(insn, 5, 19) * 4 - 4;
1111 cond = extract32(insn, 0, 4);
1114 /* genuinely conditional branches */
1115 int label_match = gen_new_label();
1116 arm_gen_test_cc(cond, label_match);
1117 gen_goto_tb(s, 0, s->pc);
1118 gen_set_label(label_match);
1119 gen_goto_tb(s, 1, addr);
1121 /* 0xe and 0xf are both "always" conditions */
1122 gen_goto_tb(s, 0, addr);
1127 static void handle_hint(DisasContext *s, uint32_t insn,
1128 unsigned int op1, unsigned int op2, unsigned int crm)
1130 unsigned int selector = crm << 3 | op2;
1133 unallocated_encoding(s);
1141 s->is_jmp = DISAS_WFI;
1145 s->is_jmp = DISAS_WFE;
1149 /* we treat all as NOP at least for now */
1152 /* default specified as NOP equivalent */
1157 static void gen_clrex(DisasContext *s, uint32_t insn)
1159 tcg_gen_movi_i64(cpu_exclusive_addr, -1);
1162 /* CLREX, DSB, DMB, ISB */
1163 static void handle_sync(DisasContext *s, uint32_t insn,
1164 unsigned int op1, unsigned int op2, unsigned int crm)
1167 unallocated_encoding(s);
1178 /* We don't emulate caches so barriers are no-ops */
1181 unallocated_encoding(s);
1186 /* C5.6.130 MSR (immediate) - move immediate to processor state field */
1187 static void handle_msr_i(DisasContext *s, uint32_t insn,
1188 unsigned int op1, unsigned int op2, unsigned int crm)
1190 int op = op1 << 3 | op2;
1192 case 0x05: /* SPSel */
1193 if (s->current_pl == 0) {
1194 unallocated_encoding(s);
1198 case 0x1e: /* DAIFSet */
1199 case 0x1f: /* DAIFClear */
1201 TCGv_i32 tcg_imm = tcg_const_i32(crm);
1202 TCGv_i32 tcg_op = tcg_const_i32(op);
1203 gen_a64_set_pc_im(s->pc - 4);
1204 gen_helper_msr_i_pstate(cpu_env, tcg_op, tcg_imm);
1205 tcg_temp_free_i32(tcg_imm);
1206 tcg_temp_free_i32(tcg_op);
1207 s->is_jmp = DISAS_UPDATE;
1211 unallocated_encoding(s);
1216 static void gen_get_nzcv(TCGv_i64 tcg_rt)
1218 TCGv_i32 tmp = tcg_temp_new_i32();
1219 TCGv_i32 nzcv = tcg_temp_new_i32();
1221 /* build bit 31, N */
1222 tcg_gen_andi_i32(nzcv, cpu_NF, (1 << 31));
1223 /* build bit 30, Z */
1224 tcg_gen_setcondi_i32(TCG_COND_EQ, tmp, cpu_ZF, 0);
1225 tcg_gen_deposit_i32(nzcv, nzcv, tmp, 30, 1);
1226 /* build bit 29, C */
1227 tcg_gen_deposit_i32(nzcv, nzcv, cpu_CF, 29, 1);
1228 /* build bit 28, V */
1229 tcg_gen_shri_i32(tmp, cpu_VF, 31);
1230 tcg_gen_deposit_i32(nzcv, nzcv, tmp, 28, 1);
1231 /* generate result */
1232 tcg_gen_extu_i32_i64(tcg_rt, nzcv);
1234 tcg_temp_free_i32(nzcv);
1235 tcg_temp_free_i32(tmp);
1238 static void gen_set_nzcv(TCGv_i64 tcg_rt)
1241 TCGv_i32 nzcv = tcg_temp_new_i32();
1243 /* take NZCV from R[t] */
1244 tcg_gen_trunc_i64_i32(nzcv, tcg_rt);
1247 tcg_gen_andi_i32(cpu_NF, nzcv, (1 << 31));
1249 tcg_gen_andi_i32(cpu_ZF, nzcv, (1 << 30));
1250 tcg_gen_setcondi_i32(TCG_COND_EQ, cpu_ZF, cpu_ZF, 0);
1252 tcg_gen_andi_i32(cpu_CF, nzcv, (1 << 29));
1253 tcg_gen_shri_i32(cpu_CF, cpu_CF, 29);
1255 tcg_gen_andi_i32(cpu_VF, nzcv, (1 << 28));
1256 tcg_gen_shli_i32(cpu_VF, cpu_VF, 3);
1257 tcg_temp_free_i32(nzcv);
1260 /* C5.6.129 MRS - move from system register
1261 * C5.6.131 MSR (register) - move to system register
1264 * These are all essentially the same insn in 'read' and 'write'
1265 * versions, with varying op0 fields.
1267 static void handle_sys(DisasContext *s, uint32_t insn, bool isread,
1268 unsigned int op0, unsigned int op1, unsigned int op2,
1269 unsigned int crn, unsigned int crm, unsigned int rt)
1271 const ARMCPRegInfo *ri;
1274 ri = get_arm_cp_reginfo(s->cp_regs,
1275 ENCODE_AA64_CP_REG(CP_REG_ARM64_SYSREG_CP,
1276 crn, crm, op0, op1, op2));
1279 /* Unknown register; this might be a guest error or a QEMU
1280 * unimplemented feature.
1282 qemu_log_mask(LOG_UNIMP, "%s access to unsupported AArch64 "
1283 "system register op0:%d op1:%d crn:%d crm:%d op2:%d\n",
1284 isread ? "read" : "write", op0, op1, crn, crm, op2);
1285 unallocated_encoding(s);
1289 /* Check access permissions */
1290 if (!cp_access_ok(s->current_pl, ri, isread)) {
1291 unallocated_encoding(s);
1296 /* Emit code to perform further access permissions checks at
1297 * runtime; this may result in an exception.
1303 gen_a64_set_pc_im(s->pc - 4);
1304 tmpptr = tcg_const_ptr(ri);
1305 syndrome = syn_aa64_sysregtrap(op0, op1, op2, crn, crm, rt, isread);
1306 tcg_syn = tcg_const_i32(syndrome);
1307 gen_helper_access_check_cp_reg(cpu_env, tmpptr, tcg_syn);
1308 tcg_temp_free_ptr(tmpptr);
1309 tcg_temp_free_i32(tcg_syn);
1312 /* Handle special cases first */
1313 switch (ri->type & ~(ARM_CP_FLAG_MASK & ~ARM_CP_SPECIAL)) {
1317 tcg_rt = cpu_reg(s, rt);
1319 gen_get_nzcv(tcg_rt);
1321 gen_set_nzcv(tcg_rt);
1324 case ARM_CP_CURRENTEL:
1325 /* Reads as current EL value from pstate, which is
1326 * guaranteed to be constant by the tb flags.
1328 tcg_rt = cpu_reg(s, rt);
1329 tcg_gen_movi_i64(tcg_rt, s->current_pl << 2);
1332 /* Writes clear the aligned block of memory which rt points into. */
1333 tcg_rt = cpu_reg(s, rt);
1334 gen_helper_dc_zva(cpu_env, tcg_rt);
1340 if (use_icount && (ri->type & ARM_CP_IO)) {
1344 tcg_rt = cpu_reg(s, rt);
1347 if (ri->type & ARM_CP_CONST) {
1348 tcg_gen_movi_i64(tcg_rt, ri->resetvalue);
1349 } else if (ri->readfn) {
1351 tmpptr = tcg_const_ptr(ri);
1352 gen_helper_get_cp_reg64(tcg_rt, cpu_env, tmpptr);
1353 tcg_temp_free_ptr(tmpptr);
1355 tcg_gen_ld_i64(tcg_rt, cpu_env, ri->fieldoffset);
1358 if (ri->type & ARM_CP_CONST) {
1359 /* If not forbidden by access permissions, treat as WI */
1361 } else if (ri->writefn) {
1363 tmpptr = tcg_const_ptr(ri);
1364 gen_helper_set_cp_reg64(cpu_env, tmpptr, tcg_rt);
1365 tcg_temp_free_ptr(tmpptr);
1367 tcg_gen_st_i64(tcg_rt, cpu_env, ri->fieldoffset);
1371 if (use_icount && (ri->type & ARM_CP_IO)) {
1372 /* I/O operations must end the TB here (whether read or write) */
1374 s->is_jmp = DISAS_UPDATE;
1375 } else if (!isread && !(ri->type & ARM_CP_SUPPRESS_TB_END)) {
1376 /* We default to ending the TB on a coprocessor register write,
1377 * but allow this to be suppressed by the register definition
1378 * (usually only necessary to work around guest bugs).
1380 s->is_jmp = DISAS_UPDATE;
1385 * 31 22 21 20 19 18 16 15 12 11 8 7 5 4 0
1386 * +---------------------+---+-----+-----+-------+-------+-----+------+
1387 * | 1 1 0 1 0 1 0 1 0 0 | L | op0 | op1 | CRn | CRm | op2 | Rt |
1388 * +---------------------+---+-----+-----+-------+-------+-----+------+
1390 static void disas_system(DisasContext *s, uint32_t insn)
1392 unsigned int l, op0, op1, crn, crm, op2, rt;
1393 l = extract32(insn, 21, 1);
1394 op0 = extract32(insn, 19, 2);
1395 op1 = extract32(insn, 16, 3);
1396 crn = extract32(insn, 12, 4);
1397 crm = extract32(insn, 8, 4);
1398 op2 = extract32(insn, 5, 3);
1399 rt = extract32(insn, 0, 5);
1402 if (l || rt != 31) {
1403 unallocated_encoding(s);
1407 case 2: /* C5.6.68 HINT */
1408 handle_hint(s, insn, op1, op2, crm);
1410 case 3: /* CLREX, DSB, DMB, ISB */
1411 handle_sync(s, insn, op1, op2, crm);
1413 case 4: /* C5.6.130 MSR (immediate) */
1414 handle_msr_i(s, insn, op1, op2, crm);
1417 unallocated_encoding(s);
1422 handle_sys(s, insn, l, op0, op1, op2, crn, crm, rt);
1425 /* C3.2.3 Exception generation
1427 * 31 24 23 21 20 5 4 2 1 0
1428 * +-----------------+-----+------------------------+-----+----+
1429 * | 1 1 0 1 0 1 0 0 | opc | imm16 | op2 | LL |
1430 * +-----------------------+------------------------+----------+
1432 static void disas_exc(DisasContext *s, uint32_t insn)
1434 int opc = extract32(insn, 21, 3);
1435 int op2_ll = extract32(insn, 0, 5);
1436 int imm16 = extract32(insn, 5, 16);
1440 /* SVC, HVC, SMC; since we don't support the Virtualization
1441 * or TrustZone extensions these all UNDEF except SVC.
1444 unallocated_encoding(s);
1447 gen_exception_insn(s, 0, EXCP_SWI, syn_aa64_svc(imm16));
1451 unallocated_encoding(s);
1455 gen_exception_insn(s, 0, EXCP_BKPT, syn_aa64_bkpt(imm16));
1459 unallocated_encoding(s);
1463 unsupported_encoding(s, insn);
1466 if (op2_ll < 1 || op2_ll > 3) {
1467 unallocated_encoding(s);
1470 /* DCPS1, DCPS2, DCPS3 */
1471 unsupported_encoding(s, insn);
1474 unallocated_encoding(s);
1479 /* C3.2.7 Unconditional branch (register)
1480 * 31 25 24 21 20 16 15 10 9 5 4 0
1481 * +---------------+-------+-------+-------+------+-------+
1482 * | 1 1 0 1 0 1 1 | opc | op2 | op3 | Rn | op4 |
1483 * +---------------+-------+-------+-------+------+-------+
1485 static void disas_uncond_b_reg(DisasContext *s, uint32_t insn)
1487 unsigned int opc, op2, op3, rn, op4;
1489 opc = extract32(insn, 21, 4);
1490 op2 = extract32(insn, 16, 5);
1491 op3 = extract32(insn, 10, 6);
1492 rn = extract32(insn, 5, 5);
1493 op4 = extract32(insn, 0, 5);
1495 if (op4 != 0x0 || op3 != 0x0 || op2 != 0x1f) {
1496 unallocated_encoding(s);
1503 tcg_gen_mov_i64(cpu_pc, cpu_reg(s, rn));
1506 tcg_gen_mov_i64(cpu_pc, cpu_reg(s, rn));
1507 tcg_gen_movi_i64(cpu_reg(s, 30), s->pc);
1510 gen_helper_exception_return(cpu_env);
1511 s->is_jmp = DISAS_JUMP;
1515 unallocated_encoding(s);
1517 unsupported_encoding(s, insn);
1521 unallocated_encoding(s);
1525 s->is_jmp = DISAS_JUMP;
1528 /* C3.2 Branches, exception generating and system instructions */
1529 static void disas_b_exc_sys(DisasContext *s, uint32_t insn)
1531 switch (extract32(insn, 25, 7)) {
1532 case 0x0a: case 0x0b:
1533 case 0x4a: case 0x4b: /* Unconditional branch (immediate) */
1534 disas_uncond_b_imm(s, insn);
1536 case 0x1a: case 0x5a: /* Compare & branch (immediate) */
1537 disas_comp_b_imm(s, insn);
1539 case 0x1b: case 0x5b: /* Test & branch (immediate) */
1540 disas_test_b_imm(s, insn);
1542 case 0x2a: /* Conditional branch (immediate) */
1543 disas_cond_b_imm(s, insn);
1545 case 0x6a: /* Exception generation / System */
1546 if (insn & (1 << 24)) {
1547 disas_system(s, insn);
1552 case 0x6b: /* Unconditional branch (register) */
1553 disas_uncond_b_reg(s, insn);
1556 unallocated_encoding(s);
1562 * Load/Store exclusive instructions are implemented by remembering
1563 * the value/address loaded, and seeing if these are the same
1564 * when the store is performed. This is not actually the architecturally
1565 * mandated semantics, but it works for typical guest code sequences
1566 * and avoids having to monitor regular stores.
1568 * In system emulation mode only one CPU will be running at once, so
1569 * this sequence is effectively atomic. In user emulation mode we
1570 * throw an exception and handle the atomic operation elsewhere.
1572 static void gen_load_exclusive(DisasContext *s, int rt, int rt2,
1573 TCGv_i64 addr, int size, bool is_pair)
1575 TCGv_i64 tmp = tcg_temp_new_i64();
1576 TCGMemOp memop = MO_TE + size;
1578 g_assert(size <= 3);
1579 tcg_gen_qemu_ld_i64(tmp, addr, get_mem_index(s), memop);
1582 TCGv_i64 addr2 = tcg_temp_new_i64();
1583 TCGv_i64 hitmp = tcg_temp_new_i64();
1585 g_assert(size >= 2);
1586 tcg_gen_addi_i64(addr2, addr, 1 << size);
1587 tcg_gen_qemu_ld_i64(hitmp, addr2, get_mem_index(s), memop);
1588 tcg_temp_free_i64(addr2);
1589 tcg_gen_mov_i64(cpu_exclusive_high, hitmp);
1590 tcg_gen_mov_i64(cpu_reg(s, rt2), hitmp);
1591 tcg_temp_free_i64(hitmp);
1594 tcg_gen_mov_i64(cpu_exclusive_val, tmp);
1595 tcg_gen_mov_i64(cpu_reg(s, rt), tmp);
1597 tcg_temp_free_i64(tmp);
1598 tcg_gen_mov_i64(cpu_exclusive_addr, addr);
1601 #ifdef CONFIG_USER_ONLY
1602 static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
1603 TCGv_i64 addr, int size, int is_pair)
1605 tcg_gen_mov_i64(cpu_exclusive_test, addr);
1606 tcg_gen_movi_i32(cpu_exclusive_info,
1607 size | is_pair << 2 | (rd << 4) | (rt << 9) | (rt2 << 14));
1608 gen_exception_internal_insn(s, 4, EXCP_STREX);
1611 static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
1612 TCGv_i64 inaddr, int size, int is_pair)
1614 /* if (env->exclusive_addr == addr && env->exclusive_val == [addr]
1615 * && (!is_pair || env->exclusive_high == [addr + datasize])) {
1618 * [addr + datasize] = {Rt2};
1624 * env->exclusive_addr = -1;
1626 int fail_label = gen_new_label();
1627 int done_label = gen_new_label();
1628 TCGv_i64 addr = tcg_temp_local_new_i64();
1631 /* Copy input into a local temp so it is not trashed when the
1632 * basic block ends at the branch insn.
1634 tcg_gen_mov_i64(addr, inaddr);
1635 tcg_gen_brcond_i64(TCG_COND_NE, addr, cpu_exclusive_addr, fail_label);
1637 tmp = tcg_temp_new_i64();
1638 tcg_gen_qemu_ld_i64(tmp, addr, get_mem_index(s), MO_TE + size);
1639 tcg_gen_brcond_i64(TCG_COND_NE, tmp, cpu_exclusive_val, fail_label);
1640 tcg_temp_free_i64(tmp);
1643 TCGv_i64 addrhi = tcg_temp_new_i64();
1644 TCGv_i64 tmphi = tcg_temp_new_i64();
1646 tcg_gen_addi_i64(addrhi, addr, 1 << size);
1647 tcg_gen_qemu_ld_i64(tmphi, addrhi, get_mem_index(s), MO_TE + size);
1648 tcg_gen_brcond_i64(TCG_COND_NE, tmphi, cpu_exclusive_high, fail_label);
1650 tcg_temp_free_i64(tmphi);
1651 tcg_temp_free_i64(addrhi);
1654 /* We seem to still have the exclusive monitor, so do the store */
1655 tcg_gen_qemu_st_i64(cpu_reg(s, rt), addr, get_mem_index(s), MO_TE + size);
1657 TCGv_i64 addrhi = tcg_temp_new_i64();
1659 tcg_gen_addi_i64(addrhi, addr, 1 << size);
1660 tcg_gen_qemu_st_i64(cpu_reg(s, rt2), addrhi,
1661 get_mem_index(s), MO_TE + size);
1662 tcg_temp_free_i64(addrhi);
1665 tcg_temp_free_i64(addr);
1667 tcg_gen_movi_i64(cpu_reg(s, rd), 0);
1668 tcg_gen_br(done_label);
1669 gen_set_label(fail_label);
1670 tcg_gen_movi_i64(cpu_reg(s, rd), 1);
1671 gen_set_label(done_label);
1672 tcg_gen_movi_i64(cpu_exclusive_addr, -1);
1677 /* C3.3.6 Load/store exclusive
1679 * 31 30 29 24 23 22 21 20 16 15 14 10 9 5 4 0
1680 * +-----+-------------+----+---+----+------+----+-------+------+------+
1681 * | sz | 0 0 1 0 0 0 | o2 | L | o1 | Rs | o0 | Rt2 | Rn | Rt |
1682 * +-----+-------------+----+---+----+------+----+-------+------+------+
1684 * sz: 00 -> 8 bit, 01 -> 16 bit, 10 -> 32 bit, 11 -> 64 bit
1685 * L: 0 -> store, 1 -> load
1686 * o2: 0 -> exclusive, 1 -> not
1687 * o1: 0 -> single register, 1 -> register pair
1688 * o0: 1 -> load-acquire/store-release, 0 -> not
1690 * o0 == 0 AND o2 == 1 is un-allocated
1691 * o1 == 1 is un-allocated except for 32 and 64 bit sizes
1693 static void disas_ldst_excl(DisasContext *s, uint32_t insn)
1695 int rt = extract32(insn, 0, 5);
1696 int rn = extract32(insn, 5, 5);
1697 int rt2 = extract32(insn, 10, 5);
1698 int is_lasr = extract32(insn, 15, 1);
1699 int rs = extract32(insn, 16, 5);
1700 int is_pair = extract32(insn, 21, 1);
1701 int is_store = !extract32(insn, 22, 1);
1702 int is_excl = !extract32(insn, 23, 1);
1703 int size = extract32(insn, 30, 2);
1706 if ((!is_excl && !is_lasr) ||
1707 (is_pair && size < 2)) {
1708 unallocated_encoding(s);
1713 gen_check_sp_alignment(s);
1715 tcg_addr = read_cpu_reg_sp(s, rn, 1);
1717 /* Note that since TCG is single threaded load-acquire/store-release
1718 * semantics require no extra if (is_lasr) { ... } handling.
1723 gen_load_exclusive(s, rt, rt2, tcg_addr, size, is_pair);
1725 gen_store_exclusive(s, rs, rt, rt2, tcg_addr, size, is_pair);
1728 TCGv_i64 tcg_rt = cpu_reg(s, rt);
1730 do_gpr_st(s, tcg_rt, tcg_addr, size);
1732 do_gpr_ld(s, tcg_rt, tcg_addr, size, false, false);
1735 TCGv_i64 tcg_rt2 = cpu_reg(s, rt);
1736 tcg_gen_addi_i64(tcg_addr, tcg_addr, 1 << size);
1738 do_gpr_st(s, tcg_rt2, tcg_addr, size);
1740 do_gpr_ld(s, tcg_rt2, tcg_addr, size, false, false);
1747 * C3.3.5 Load register (literal)
1749 * 31 30 29 27 26 25 24 23 5 4 0
1750 * +-----+-------+---+-----+-------------------+-------+
1751 * | opc | 0 1 1 | V | 0 0 | imm19 | Rt |
1752 * +-----+-------+---+-----+-------------------+-------+
1754 * V: 1 -> vector (simd/fp)
1755 * opc (non-vector): 00 -> 32 bit, 01 -> 64 bit,
1756 * 10-> 32 bit signed, 11 -> prefetch
1757 * opc (vector): 00 -> 32 bit, 01 -> 64 bit, 10 -> 128 bit (11 unallocated)
1759 static void disas_ld_lit(DisasContext *s, uint32_t insn)
1761 int rt = extract32(insn, 0, 5);
1762 int64_t imm = sextract32(insn, 5, 19) << 2;
1763 bool is_vector = extract32(insn, 26, 1);
1764 int opc = extract32(insn, 30, 2);
1765 bool is_signed = false;
1767 TCGv_i64 tcg_rt, tcg_addr;
1771 unallocated_encoding(s);
1775 if (!fp_access_check(s)) {
1780 /* PRFM (literal) : prefetch */
1783 size = 2 + extract32(opc, 0, 1);
1784 is_signed = extract32(opc, 1, 1);
1787 tcg_rt = cpu_reg(s, rt);
1789 tcg_addr = tcg_const_i64((s->pc - 4) + imm);
1791 do_fp_ld(s, rt, tcg_addr, size);
1793 do_gpr_ld(s, tcg_rt, tcg_addr, size, is_signed, false);
1795 tcg_temp_free_i64(tcg_addr);
1799 * C5.6.80 LDNP (Load Pair - non-temporal hint)
1800 * C5.6.81 LDP (Load Pair - non vector)
1801 * C5.6.82 LDPSW (Load Pair Signed Word - non vector)
1802 * C5.6.176 STNP (Store Pair - non-temporal hint)
1803 * C5.6.177 STP (Store Pair - non vector)
1804 * C6.3.165 LDNP (Load Pair of SIMD&FP - non-temporal hint)
1805 * C6.3.165 LDP (Load Pair of SIMD&FP)
1806 * C6.3.284 STNP (Store Pair of SIMD&FP - non-temporal hint)
1807 * C6.3.284 STP (Store Pair of SIMD&FP)
1809 * 31 30 29 27 26 25 24 23 22 21 15 14 10 9 5 4 0
1810 * +-----+-------+---+---+-------+---+-----------------------------+
1811 * | opc | 1 0 1 | V | 0 | index | L | imm7 | Rt2 | Rn | Rt |
1812 * +-----+-------+---+---+-------+---+-------+-------+------+------+
1814 * opc: LDP/STP/LDNP/STNP 00 -> 32 bit, 10 -> 64 bit
1816 * LDP/STP/LDNP/STNP (SIMD) 00 -> 32 bit, 01 -> 64 bit, 10 -> 128 bit
1817 * V: 0 -> GPR, 1 -> Vector
1818 * idx: 00 -> signed offset with non-temporal hint, 01 -> post-index,
1819 * 10 -> signed offset, 11 -> pre-index
1820 * L: 0 -> Store 1 -> Load
1822 * Rt, Rt2 = GPR or SIMD registers to be stored
1823 * Rn = general purpose register containing address
1824 * imm7 = signed offset (multiple of 4 or 8 depending on size)
1826 static void disas_ldst_pair(DisasContext *s, uint32_t insn)
1828 int rt = extract32(insn, 0, 5);
1829 int rn = extract32(insn, 5, 5);
1830 int rt2 = extract32(insn, 10, 5);
1831 int64_t offset = sextract32(insn, 15, 7);
1832 int index = extract32(insn, 23, 2);
1833 bool is_vector = extract32(insn, 26, 1);
1834 bool is_load = extract32(insn, 22, 1);
1835 int opc = extract32(insn, 30, 2);
1837 bool is_signed = false;
1838 bool postindex = false;
1841 TCGv_i64 tcg_addr; /* calculated address */
1845 unallocated_encoding(s);
1852 size = 2 + extract32(opc, 1, 1);
1853 is_signed = extract32(opc, 0, 1);
1854 if (!is_load && is_signed) {
1855 unallocated_encoding(s);
1861 case 1: /* post-index */
1866 /* signed offset with "non-temporal" hint. Since we don't emulate
1867 * caches we don't care about hints to the cache system about
1868 * data access patterns, and handle this identically to plain
1872 /* There is no non-temporal-hint version of LDPSW */
1873 unallocated_encoding(s);
1878 case 2: /* signed offset, rn not updated */
1881 case 3: /* pre-index */
1887 if (is_vector && !fp_access_check(s)) {
1894 gen_check_sp_alignment(s);
1897 tcg_addr = read_cpu_reg_sp(s, rn, 1);
1900 tcg_gen_addi_i64(tcg_addr, tcg_addr, offset);
1905 do_fp_ld(s, rt, tcg_addr, size);
1907 do_fp_st(s, rt, tcg_addr, size);
1910 TCGv_i64 tcg_rt = cpu_reg(s, rt);
1912 do_gpr_ld(s, tcg_rt, tcg_addr, size, is_signed, false);
1914 do_gpr_st(s, tcg_rt, tcg_addr, size);
1917 tcg_gen_addi_i64(tcg_addr, tcg_addr, 1 << size);
1920 do_fp_ld(s, rt2, tcg_addr, size);
1922 do_fp_st(s, rt2, tcg_addr, size);
1925 TCGv_i64 tcg_rt2 = cpu_reg(s, rt2);
1927 do_gpr_ld(s, tcg_rt2, tcg_addr, size, is_signed, false);
1929 do_gpr_st(s, tcg_rt2, tcg_addr, size);
1935 tcg_gen_addi_i64(tcg_addr, tcg_addr, offset - (1 << size));
1937 tcg_gen_subi_i64(tcg_addr, tcg_addr, 1 << size);
1939 tcg_gen_mov_i64(cpu_reg_sp(s, rn), tcg_addr);
1944 * C3.3.8 Load/store (immediate post-indexed)
1945 * C3.3.9 Load/store (immediate pre-indexed)
1946 * C3.3.12 Load/store (unscaled immediate)
1948 * 31 30 29 27 26 25 24 23 22 21 20 12 11 10 9 5 4 0
1949 * +----+-------+---+-----+-----+---+--------+-----+------+------+
1950 * |size| 1 1 1 | V | 0 0 | opc | 0 | imm9 | idx | Rn | Rt |
1951 * +----+-------+---+-----+-----+---+--------+-----+------+------+
1953 * idx = 01 -> post-indexed, 11 pre-indexed, 00 unscaled imm. (no writeback)
1955 * V = 0 -> non-vector
1956 * size: 00 -> 8 bit, 01 -> 16 bit, 10 -> 32 bit, 11 -> 64bit
1957 * opc: 00 -> store, 01 -> loadu, 10 -> loads 64, 11 -> loads 32
1959 static void disas_ldst_reg_imm9(DisasContext *s, uint32_t insn)
1961 int rt = extract32(insn, 0, 5);
1962 int rn = extract32(insn, 5, 5);
1963 int imm9 = sextract32(insn, 12, 9);
1964 int opc = extract32(insn, 22, 2);
1965 int size = extract32(insn, 30, 2);
1966 int idx = extract32(insn, 10, 2);
1967 bool is_signed = false;
1968 bool is_store = false;
1969 bool is_extended = false;
1970 bool is_unpriv = (idx == 2);
1971 bool is_vector = extract32(insn, 26, 1);
1978 size |= (opc & 2) << 1;
1979 if (size > 4 || is_unpriv) {
1980 unallocated_encoding(s);
1983 is_store = ((opc & 1) == 0);
1984 if (!fp_access_check(s)) {
1988 if (size == 3 && opc == 2) {
1989 /* PRFM - prefetch */
1991 unallocated_encoding(s);
1996 if (opc == 3 && size > 1) {
1997 unallocated_encoding(s);
2000 is_store = (opc == 0);
2001 is_signed = opc & (1<<1);
2002 is_extended = (size < 3) && (opc & 1);
2022 gen_check_sp_alignment(s);
2024 tcg_addr = read_cpu_reg_sp(s, rn, 1);
2027 tcg_gen_addi_i64(tcg_addr, tcg_addr, imm9);
2032 do_fp_st(s, rt, tcg_addr, size);
2034 do_fp_ld(s, rt, tcg_addr, size);
2037 TCGv_i64 tcg_rt = cpu_reg(s, rt);
2038 int memidx = is_unpriv ? 1 : get_mem_index(s);
2041 do_gpr_st_memidx(s, tcg_rt, tcg_addr, size, memidx);
2043 do_gpr_ld_memidx(s, tcg_rt, tcg_addr, size,
2044 is_signed, is_extended, memidx);
2049 TCGv_i64 tcg_rn = cpu_reg_sp(s, rn);
2051 tcg_gen_addi_i64(tcg_addr, tcg_addr, imm9);
2053 tcg_gen_mov_i64(tcg_rn, tcg_addr);
2058 * C3.3.10 Load/store (register offset)
2060 * 31 30 29 27 26 25 24 23 22 21 20 16 15 13 12 11 10 9 5 4 0
2061 * +----+-------+---+-----+-----+---+------+-----+--+-----+----+----+
2062 * |size| 1 1 1 | V | 0 0 | opc | 1 | Rm | opt | S| 1 0 | Rn | Rt |
2063 * +----+-------+---+-----+-----+---+------+-----+--+-----+----+----+
2066 * size: 00-> byte, 01 -> 16 bit, 10 -> 32bit, 11 -> 64bit
2067 * opc: 00 -> store, 01 -> loadu, 10 -> loads 64, 11 -> loads 32
2069 * size is opc<1>:size<1:0> so 100 -> 128 bit; 110 and 111 unallocated
2070 * opc<0>: 0 -> store, 1 -> load
2071 * V: 1 -> vector/simd
2072 * opt: extend encoding (see DecodeRegExtend)
2073 * S: if S=1 then scale (essentially index by sizeof(size))
2074 * Rt: register to transfer into/out of
2075 * Rn: address register or SP for base
2076 * Rm: offset register or ZR for offset
2078 static void disas_ldst_reg_roffset(DisasContext *s, uint32_t insn)
2080 int rt = extract32(insn, 0, 5);
2081 int rn = extract32(insn, 5, 5);
2082 int shift = extract32(insn, 12, 1);
2083 int rm = extract32(insn, 16, 5);
2084 int opc = extract32(insn, 22, 2);
2085 int opt = extract32(insn, 13, 3);
2086 int size = extract32(insn, 30, 2);
2087 bool is_signed = false;
2088 bool is_store = false;
2089 bool is_extended = false;
2090 bool is_vector = extract32(insn, 26, 1);
2095 if (extract32(opt, 1, 1) == 0) {
2096 unallocated_encoding(s);
2101 size |= (opc & 2) << 1;
2103 unallocated_encoding(s);
2106 is_store = !extract32(opc, 0, 1);
2107 if (!fp_access_check(s)) {
2111 if (size == 3 && opc == 2) {
2112 /* PRFM - prefetch */
2115 if (opc == 3 && size > 1) {
2116 unallocated_encoding(s);
2119 is_store = (opc == 0);
2120 is_signed = extract32(opc, 1, 1);
2121 is_extended = (size < 3) && extract32(opc, 0, 1);
2125 gen_check_sp_alignment(s);
2127 tcg_addr = read_cpu_reg_sp(s, rn, 1);
2129 tcg_rm = read_cpu_reg(s, rm, 1);
2130 ext_and_shift_reg(tcg_rm, tcg_rm, opt, shift ? size : 0);
2132 tcg_gen_add_i64(tcg_addr, tcg_addr, tcg_rm);
2136 do_fp_st(s, rt, tcg_addr, size);
2138 do_fp_ld(s, rt, tcg_addr, size);
2141 TCGv_i64 tcg_rt = cpu_reg(s, rt);
2143 do_gpr_st(s, tcg_rt, tcg_addr, size);
2145 do_gpr_ld(s, tcg_rt, tcg_addr, size, is_signed, is_extended);
2151 * C3.3.13 Load/store (unsigned immediate)
2153 * 31 30 29 27 26 25 24 23 22 21 10 9 5
2154 * +----+-------+---+-----+-----+------------+-------+------+
2155 * |size| 1 1 1 | V | 0 1 | opc | imm12 | Rn | Rt |
2156 * +----+-------+---+-----+-----+------------+-------+------+
2159 * size: 00-> byte, 01 -> 16 bit, 10 -> 32bit, 11 -> 64bit
2160 * opc: 00 -> store, 01 -> loadu, 10 -> loads 64, 11 -> loads 32
2162 * size is opc<1>:size<1:0> so 100 -> 128 bit; 110 and 111 unallocated
2163 * opc<0>: 0 -> store, 1 -> load
2164 * Rn: base address register (inc SP)
2165 * Rt: target register
2167 static void disas_ldst_reg_unsigned_imm(DisasContext *s, uint32_t insn)
2169 int rt = extract32(insn, 0, 5);
2170 int rn = extract32(insn, 5, 5);
2171 unsigned int imm12 = extract32(insn, 10, 12);
2172 bool is_vector = extract32(insn, 26, 1);
2173 int size = extract32(insn, 30, 2);
2174 int opc = extract32(insn, 22, 2);
2175 unsigned int offset;
2180 bool is_signed = false;
2181 bool is_extended = false;
2184 size |= (opc & 2) << 1;
2186 unallocated_encoding(s);
2189 is_store = !extract32(opc, 0, 1);
2190 if (!fp_access_check(s)) {
2194 if (size == 3 && opc == 2) {
2195 /* PRFM - prefetch */
2198 if (opc == 3 && size > 1) {
2199 unallocated_encoding(s);
2202 is_store = (opc == 0);
2203 is_signed = extract32(opc, 1, 1);
2204 is_extended = (size < 3) && extract32(opc, 0, 1);
2208 gen_check_sp_alignment(s);
2210 tcg_addr = read_cpu_reg_sp(s, rn, 1);
2211 offset = imm12 << size;
2212 tcg_gen_addi_i64(tcg_addr, tcg_addr, offset);
2216 do_fp_st(s, rt, tcg_addr, size);
2218 do_fp_ld(s, rt, tcg_addr, size);
2221 TCGv_i64 tcg_rt = cpu_reg(s, rt);
2223 do_gpr_st(s, tcg_rt, tcg_addr, size);
2225 do_gpr_ld(s, tcg_rt, tcg_addr, size, is_signed, is_extended);
2230 /* Load/store register (all forms) */
2231 static void disas_ldst_reg(DisasContext *s, uint32_t insn)
2233 switch (extract32(insn, 24, 2)) {
2235 if (extract32(insn, 21, 1) == 1 && extract32(insn, 10, 2) == 2) {
2236 disas_ldst_reg_roffset(s, insn);
2238 /* Load/store register (unscaled immediate)
2239 * Load/store immediate pre/post-indexed
2240 * Load/store register unprivileged
2242 disas_ldst_reg_imm9(s, insn);
2246 disas_ldst_reg_unsigned_imm(s, insn);
2249 unallocated_encoding(s);
2254 /* C3.3.1 AdvSIMD load/store multiple structures
2256 * 31 30 29 23 22 21 16 15 12 11 10 9 5 4 0
2257 * +---+---+---------------+---+-------------+--------+------+------+------+
2258 * | 0 | Q | 0 0 1 1 0 0 0 | L | 0 0 0 0 0 0 | opcode | size | Rn | Rt |
2259 * +---+---+---------------+---+-------------+--------+------+------+------+
2261 * C3.3.2 AdvSIMD load/store multiple structures (post-indexed)
2263 * 31 30 29 23 22 21 20 16 15 12 11 10 9 5 4 0
2264 * +---+---+---------------+---+---+---------+--------+------+------+------+
2265 * | 0 | Q | 0 0 1 1 0 0 1 | L | 0 | Rm | opcode | size | Rn | Rt |
2266 * +---+---+---------------+---+---+---------+--------+------+------+------+
2268 * Rt: first (or only) SIMD&FP register to be transferred
2269 * Rn: base address or SP
2270 * Rm (post-index only): post-index register (when !31) or size dependent #imm
2272 static void disas_ldst_multiple_struct(DisasContext *s, uint32_t insn)
2274 int rt = extract32(insn, 0, 5);
2275 int rn = extract32(insn, 5, 5);
2276 int size = extract32(insn, 10, 2);
2277 int opcode = extract32(insn, 12, 4);
2278 bool is_store = !extract32(insn, 22, 1);
2279 bool is_postidx = extract32(insn, 23, 1);
2280 bool is_q = extract32(insn, 30, 1);
2281 TCGv_i64 tcg_addr, tcg_rn;
2283 int ebytes = 1 << size;
2284 int elements = (is_q ? 128 : 64) / (8 << size);
2285 int rpt; /* num iterations */
2286 int selem; /* structure elements */
2289 if (extract32(insn, 31, 1) || extract32(insn, 21, 1)) {
2290 unallocated_encoding(s);
2294 /* From the shared decode logic */
2325 unallocated_encoding(s);
2329 if (size == 3 && !is_q && selem != 1) {
2331 unallocated_encoding(s);
2335 if (!fp_access_check(s)) {
2340 gen_check_sp_alignment(s);
2343 tcg_rn = cpu_reg_sp(s, rn);
2344 tcg_addr = tcg_temp_new_i64();
2345 tcg_gen_mov_i64(tcg_addr, tcg_rn);
2347 for (r = 0; r < rpt; r++) {
2349 for (e = 0; e < elements; e++) {
2350 int tt = (rt + r) % 32;
2352 for (xs = 0; xs < selem; xs++) {
2354 do_vec_st(s, tt, e, tcg_addr, size);
2356 do_vec_ld(s, tt, e, tcg_addr, size);
2358 /* For non-quad operations, setting a slice of the low
2359 * 64 bits of the register clears the high 64 bits (in
2360 * the ARM ARM pseudocode this is implicit in the fact
2361 * that 'rval' is a 64 bit wide variable). We optimize
2362 * by noticing that we only need to do this the first
2363 * time we touch a register.
2365 if (!is_q && e == 0 && (r == 0 || xs == selem - 1)) {
2366 clear_vec_high(s, tt);
2369 tcg_gen_addi_i64(tcg_addr, tcg_addr, ebytes);
2376 int rm = extract32(insn, 16, 5);
2378 tcg_gen_mov_i64(tcg_rn, tcg_addr);
2380 tcg_gen_add_i64(tcg_rn, tcg_rn, cpu_reg(s, rm));
2383 tcg_temp_free_i64(tcg_addr);
2386 /* C3.3.3 AdvSIMD load/store single structure
2388 * 31 30 29 23 22 21 20 16 15 13 12 11 10 9 5 4 0
2389 * +---+---+---------------+-----+-----------+-----+---+------+------+------+
2390 * | 0 | Q | 0 0 1 1 0 1 0 | L R | 0 0 0 0 0 | opc | S | size | Rn | Rt |
2391 * +---+---+---------------+-----+-----------+-----+---+------+------+------+
2393 * C3.3.4 AdvSIMD load/store single structure (post-indexed)
2395 * 31 30 29 23 22 21 20 16 15 13 12 11 10 9 5 4 0
2396 * +---+---+---------------+-----+-----------+-----+---+------+------+------+
2397 * | 0 | Q | 0 0 1 1 0 1 1 | L R | Rm | opc | S | size | Rn | Rt |
2398 * +---+---+---------------+-----+-----------+-----+---+------+------+------+
2400 * Rt: first (or only) SIMD&FP register to be transferred
2401 * Rn: base address or SP
2402 * Rm (post-index only): post-index register (when !31) or size dependent #imm
2403 * index = encoded in Q:S:size dependent on size
2405 * lane_size = encoded in R, opc
2406 * transfer width = encoded in opc, S, size
2408 static void disas_ldst_single_struct(DisasContext *s, uint32_t insn)
2410 int rt = extract32(insn, 0, 5);
2411 int rn = extract32(insn, 5, 5);
2412 int size = extract32(insn, 10, 2);
2413 int S = extract32(insn, 12, 1);
2414 int opc = extract32(insn, 13, 3);
2415 int R = extract32(insn, 21, 1);
2416 int is_load = extract32(insn, 22, 1);
2417 int is_postidx = extract32(insn, 23, 1);
2418 int is_q = extract32(insn, 30, 1);
2420 int scale = extract32(opc, 1, 2);
2421 int selem = (extract32(opc, 0, 1) << 1 | R) + 1;
2422 bool replicate = false;
2423 int index = is_q << 3 | S << 2 | size;
2425 TCGv_i64 tcg_addr, tcg_rn;
2429 if (!is_load || S) {
2430 unallocated_encoding(s);
2439 if (extract32(size, 0, 1)) {
2440 unallocated_encoding(s);
2446 if (extract32(size, 1, 1)) {
2447 unallocated_encoding(s);
2450 if (!extract32(size, 0, 1)) {
2454 unallocated_encoding(s);
2462 g_assert_not_reached();
2465 if (!fp_access_check(s)) {
2469 ebytes = 1 << scale;
2472 gen_check_sp_alignment(s);
2475 tcg_rn = cpu_reg_sp(s, rn);
2476 tcg_addr = tcg_temp_new_i64();
2477 tcg_gen_mov_i64(tcg_addr, tcg_rn);
2479 for (xs = 0; xs < selem; xs++) {
2481 /* Load and replicate to all elements */
2483 TCGv_i64 tcg_tmp = tcg_temp_new_i64();
2485 tcg_gen_qemu_ld_i64(tcg_tmp, tcg_addr,
2486 get_mem_index(s), MO_TE + scale);
2489 mulconst = 0x0101010101010101ULL;
2492 mulconst = 0x0001000100010001ULL;
2495 mulconst = 0x0000000100000001ULL;
2501 g_assert_not_reached();
2504 tcg_gen_muli_i64(tcg_tmp, tcg_tmp, mulconst);
2506 write_vec_element(s, tcg_tmp, rt, 0, MO_64);
2508 write_vec_element(s, tcg_tmp, rt, 1, MO_64);
2510 clear_vec_high(s, rt);
2512 tcg_temp_free_i64(tcg_tmp);
2514 /* Load/store one element per register */
2516 do_vec_ld(s, rt, index, tcg_addr, MO_TE + scale);
2518 do_vec_st(s, rt, index, tcg_addr, MO_TE + scale);
2521 tcg_gen_addi_i64(tcg_addr, tcg_addr, ebytes);
2526 int rm = extract32(insn, 16, 5);
2528 tcg_gen_mov_i64(tcg_rn, tcg_addr);
2530 tcg_gen_add_i64(tcg_rn, tcg_rn, cpu_reg(s, rm));
2533 tcg_temp_free_i64(tcg_addr);
2536 /* C3.3 Loads and stores */
2537 static void disas_ldst(DisasContext *s, uint32_t insn)
2539 switch (extract32(insn, 24, 6)) {
2540 case 0x08: /* Load/store exclusive */
2541 disas_ldst_excl(s, insn);
2543 case 0x18: case 0x1c: /* Load register (literal) */
2544 disas_ld_lit(s, insn);
2546 case 0x28: case 0x29:
2547 case 0x2c: case 0x2d: /* Load/store pair (all forms) */
2548 disas_ldst_pair(s, insn);
2550 case 0x38: case 0x39:
2551 case 0x3c: case 0x3d: /* Load/store register (all forms) */
2552 disas_ldst_reg(s, insn);
2554 case 0x0c: /* AdvSIMD load/store multiple structures */
2555 disas_ldst_multiple_struct(s, insn);
2557 case 0x0d: /* AdvSIMD load/store single structure */
2558 disas_ldst_single_struct(s, insn);
2561 unallocated_encoding(s);
2566 /* C3.4.6 PC-rel. addressing
2567 * 31 30 29 28 24 23 5 4 0
2568 * +----+-------+-----------+-------------------+------+
2569 * | op | immlo | 1 0 0 0 0 | immhi | Rd |
2570 * +----+-------+-----------+-------------------+------+
2572 static void disas_pc_rel_adr(DisasContext *s, uint32_t insn)
2574 unsigned int page, rd;
2578 page = extract32(insn, 31, 1);
2579 /* SignExtend(immhi:immlo) -> offset */
2580 offset = ((int64_t)sextract32(insn, 5, 19) << 2) | extract32(insn, 29, 2);
2581 rd = extract32(insn, 0, 5);
2585 /* ADRP (page based) */
2590 tcg_gen_movi_i64(cpu_reg(s, rd), base + offset);
2594 * C3.4.1 Add/subtract (immediate)
2596 * 31 30 29 28 24 23 22 21 10 9 5 4 0
2597 * +--+--+--+-----------+-----+-------------+-----+-----+
2598 * |sf|op| S| 1 0 0 0 1 |shift| imm12 | Rn | Rd |
2599 * +--+--+--+-----------+-----+-------------+-----+-----+
2601 * sf: 0 -> 32bit, 1 -> 64bit
2602 * op: 0 -> add , 1 -> sub
2604 * shift: 00 -> LSL imm by 0, 01 -> LSL imm by 12
2606 static void disas_add_sub_imm(DisasContext *s, uint32_t insn)
2608 int rd = extract32(insn, 0, 5);
2609 int rn = extract32(insn, 5, 5);
2610 uint64_t imm = extract32(insn, 10, 12);
2611 int shift = extract32(insn, 22, 2);
2612 bool setflags = extract32(insn, 29, 1);
2613 bool sub_op = extract32(insn, 30, 1);
2614 bool is_64bit = extract32(insn, 31, 1);
2616 TCGv_i64 tcg_rn = cpu_reg_sp(s, rn);
2617 TCGv_i64 tcg_rd = setflags ? cpu_reg(s, rd) : cpu_reg_sp(s, rd);
2618 TCGv_i64 tcg_result;
2627 unallocated_encoding(s);
2631 tcg_result = tcg_temp_new_i64();
2634 tcg_gen_subi_i64(tcg_result, tcg_rn, imm);
2636 tcg_gen_addi_i64(tcg_result, tcg_rn, imm);
2639 TCGv_i64 tcg_imm = tcg_const_i64(imm);
2641 gen_sub_CC(is_64bit, tcg_result, tcg_rn, tcg_imm);
2643 gen_add_CC(is_64bit, tcg_result, tcg_rn, tcg_imm);
2645 tcg_temp_free_i64(tcg_imm);
2649 tcg_gen_mov_i64(tcg_rd, tcg_result);
2651 tcg_gen_ext32u_i64(tcg_rd, tcg_result);
2654 tcg_temp_free_i64(tcg_result);
2657 /* The input should be a value in the bottom e bits (with higher
2658 * bits zero); returns that value replicated into every element
2659 * of size e in a 64 bit integer.
2661 static uint64_t bitfield_replicate(uint64_t mask, unsigned int e)
2671 /* Return a value with the bottom len bits set (where 0 < len <= 64) */
2672 static inline uint64_t bitmask64(unsigned int length)
2674 assert(length > 0 && length <= 64);
2675 return ~0ULL >> (64 - length);
2678 /* Simplified variant of pseudocode DecodeBitMasks() for the case where we
2679 * only require the wmask. Returns false if the imms/immr/immn are a reserved
2680 * value (ie should cause a guest UNDEF exception), and true if they are
2681 * valid, in which case the decoded bit pattern is written to result.
2683 static bool logic_imm_decode_wmask(uint64_t *result, unsigned int immn,
2684 unsigned int imms, unsigned int immr)
2687 unsigned e, levels, s, r;
2690 assert(immn < 2 && imms < 64 && immr < 64);
2692 /* The bit patterns we create here are 64 bit patterns which
2693 * are vectors of identical elements of size e = 2, 4, 8, 16, 32 or
2694 * 64 bits each. Each element contains the same value: a run
2695 * of between 1 and e-1 non-zero bits, rotated within the
2696 * element by between 0 and e-1 bits.
2698 * The element size and run length are encoded into immn (1 bit)
2699 * and imms (6 bits) as follows:
2700 * 64 bit elements: immn = 1, imms = <length of run - 1>
2701 * 32 bit elements: immn = 0, imms = 0 : <length of run - 1>
2702 * 16 bit elements: immn = 0, imms = 10 : <length of run - 1>
2703 * 8 bit elements: immn = 0, imms = 110 : <length of run - 1>
2704 * 4 bit elements: immn = 0, imms = 1110 : <length of run - 1>
2705 * 2 bit elements: immn = 0, imms = 11110 : <length of run - 1>
2706 * Notice that immn = 0, imms = 11111x is the only combination
2707 * not covered by one of the above options; this is reserved.
2708 * Further, <length of run - 1> all-ones is a reserved pattern.
2710 * In all cases the rotation is by immr % e (and immr is 6 bits).
2713 /* First determine the element size */
2714 len = 31 - clz32((immn << 6) | (~imms & 0x3f));
2716 /* This is the immn == 0, imms == 0x11111x case */
2726 /* <length of run - 1> mustn't be all-ones. */
2730 /* Create the value of one element: s+1 set bits rotated
2731 * by r within the element (which is e bits wide)...
2733 mask = bitmask64(s + 1);
2734 mask = (mask >> r) | (mask << (e - r));
2735 /* ...then replicate the element over the whole 64 bit value */
2736 mask = bitfield_replicate(mask, e);
2741 /* C3.4.4 Logical (immediate)
2742 * 31 30 29 28 23 22 21 16 15 10 9 5 4 0
2743 * +----+-----+-------------+---+------+------+------+------+
2744 * | sf | opc | 1 0 0 1 0 0 | N | immr | imms | Rn | Rd |
2745 * +----+-----+-------------+---+------+------+------+------+
2747 static void disas_logic_imm(DisasContext *s, uint32_t insn)
2749 unsigned int sf, opc, is_n, immr, imms, rn, rd;
2750 TCGv_i64 tcg_rd, tcg_rn;
2752 bool is_and = false;
2754 sf = extract32(insn, 31, 1);
2755 opc = extract32(insn, 29, 2);
2756 is_n = extract32(insn, 22, 1);
2757 immr = extract32(insn, 16, 6);
2758 imms = extract32(insn, 10, 6);
2759 rn = extract32(insn, 5, 5);
2760 rd = extract32(insn, 0, 5);
2763 unallocated_encoding(s);
2767 if (opc == 0x3) { /* ANDS */
2768 tcg_rd = cpu_reg(s, rd);
2770 tcg_rd = cpu_reg_sp(s, rd);
2772 tcg_rn = cpu_reg(s, rn);
2774 if (!logic_imm_decode_wmask(&wmask, is_n, imms, immr)) {
2775 /* some immediate field values are reserved */
2776 unallocated_encoding(s);
2781 wmask &= 0xffffffff;
2785 case 0x3: /* ANDS */
2787 tcg_gen_andi_i64(tcg_rd, tcg_rn, wmask);
2791 tcg_gen_ori_i64(tcg_rd, tcg_rn, wmask);
2794 tcg_gen_xori_i64(tcg_rd, tcg_rn, wmask);
2797 assert(FALSE); /* must handle all above */
2801 if (!sf && !is_and) {
2802 /* zero extend final result; we know we can skip this for AND
2803 * since the immediate had the high 32 bits clear.
2805 tcg_gen_ext32u_i64(tcg_rd, tcg_rd);
2808 if (opc == 3) { /* ANDS */
2809 gen_logic_CC(sf, tcg_rd);
2814 * C3.4.5 Move wide (immediate)
2816 * 31 30 29 28 23 22 21 20 5 4 0
2817 * +--+-----+-------------+-----+----------------+------+
2818 * |sf| opc | 1 0 0 1 0 1 | hw | imm16 | Rd |
2819 * +--+-----+-------------+-----+----------------+------+
2821 * sf: 0 -> 32 bit, 1 -> 64 bit
2822 * opc: 00 -> N, 10 -> Z, 11 -> K
2823 * hw: shift/16 (0,16, and sf only 32, 48)
2825 static void disas_movw_imm(DisasContext *s, uint32_t insn)
2827 int rd = extract32(insn, 0, 5);
2828 uint64_t imm = extract32(insn, 5, 16);
2829 int sf = extract32(insn, 31, 1);
2830 int opc = extract32(insn, 29, 2);
2831 int pos = extract32(insn, 21, 2) << 4;
2832 TCGv_i64 tcg_rd = cpu_reg(s, rd);
2835 if (!sf && (pos >= 32)) {
2836 unallocated_encoding(s);
2850 tcg_gen_movi_i64(tcg_rd, imm);
2853 tcg_imm = tcg_const_i64(imm);
2854 tcg_gen_deposit_i64(tcg_rd, tcg_rd, tcg_imm, pos, 16);
2855 tcg_temp_free_i64(tcg_imm);
2857 tcg_gen_ext32u_i64(tcg_rd, tcg_rd);
2861 unallocated_encoding(s);
2867 * 31 30 29 28 23 22 21 16 15 10 9 5 4 0
2868 * +----+-----+-------------+---+------+------+------+------+
2869 * | sf | opc | 1 0 0 1 1 0 | N | immr | imms | Rn | Rd |
2870 * +----+-----+-------------+---+------+------+------+------+
2872 static void disas_bitfield(DisasContext *s, uint32_t insn)
2874 unsigned int sf, n, opc, ri, si, rn, rd, bitsize, pos, len;
2875 TCGv_i64 tcg_rd, tcg_tmp;
2877 sf = extract32(insn, 31, 1);
2878 opc = extract32(insn, 29, 2);
2879 n = extract32(insn, 22, 1);
2880 ri = extract32(insn, 16, 6);
2881 si = extract32(insn, 10, 6);
2882 rn = extract32(insn, 5, 5);
2883 rd = extract32(insn, 0, 5);
2884 bitsize = sf ? 64 : 32;
2886 if (sf != n || ri >= bitsize || si >= bitsize || opc > 2) {
2887 unallocated_encoding(s);
2891 tcg_rd = cpu_reg(s, rd);
2892 tcg_tmp = read_cpu_reg(s, rn, sf);
2894 /* OPTME: probably worth recognizing common cases of ext{8,16,32}{u,s} */
2896 if (opc != 1) { /* SBFM or UBFM */
2897 tcg_gen_movi_i64(tcg_rd, 0);
2900 /* do the bit move operation */
2902 /* Wd<s-r:0> = Wn<s:r> */
2903 tcg_gen_shri_i64(tcg_tmp, tcg_tmp, ri);
2905 len = (si - ri) + 1;
2907 /* Wd<32+s-r,32-r> = Wn<s:0> */
2912 tcg_gen_deposit_i64(tcg_rd, tcg_rd, tcg_tmp, pos, len);
2914 if (opc == 0) { /* SBFM - sign extend the destination field */
2915 tcg_gen_shli_i64(tcg_rd, tcg_rd, 64 - (pos + len));
2916 tcg_gen_sari_i64(tcg_rd, tcg_rd, 64 - (pos + len));
2919 if (!sf) { /* zero extend final result */
2920 tcg_gen_ext32u_i64(tcg_rd, tcg_rd);
2925 * 31 30 29 28 23 22 21 20 16 15 10 9 5 4 0
2926 * +----+------+-------------+---+----+------+--------+------+------+
2927 * | sf | op21 | 1 0 0 1 1 1 | N | o0 | Rm | imms | Rn | Rd |
2928 * +----+------+-------------+---+----+------+--------+------+------+
2930 static void disas_extract(DisasContext *s, uint32_t insn)
2932 unsigned int sf, n, rm, imm, rn, rd, bitsize, op21, op0;
2934 sf = extract32(insn, 31, 1);
2935 n = extract32(insn, 22, 1);
2936 rm = extract32(insn, 16, 5);
2937 imm = extract32(insn, 10, 6);
2938 rn = extract32(insn, 5, 5);
2939 rd = extract32(insn, 0, 5);
2940 op21 = extract32(insn, 29, 2);
2941 op0 = extract32(insn, 21, 1);
2942 bitsize = sf ? 64 : 32;
2944 if (sf != n || op21 || op0 || imm >= bitsize) {
2945 unallocated_encoding(s);
2947 TCGv_i64 tcg_rd, tcg_rm, tcg_rn;
2949 tcg_rd = cpu_reg(s, rd);
2952 /* OPTME: we can special case rm==rn as a rotate */
2953 tcg_rm = read_cpu_reg(s, rm, sf);
2954 tcg_rn = read_cpu_reg(s, rn, sf);
2955 tcg_gen_shri_i64(tcg_rm, tcg_rm, imm);
2956 tcg_gen_shli_i64(tcg_rn, tcg_rn, bitsize - imm);
2957 tcg_gen_or_i64(tcg_rd, tcg_rm, tcg_rn);
2959 tcg_gen_ext32u_i64(tcg_rd, tcg_rd);
2962 /* tcg shl_i32/shl_i64 is undefined for 32/64 bit shifts,
2963 * so an extract from bit 0 is a special case.
2966 tcg_gen_mov_i64(tcg_rd, cpu_reg(s, rm));
2968 tcg_gen_ext32u_i64(tcg_rd, cpu_reg(s, rm));
2975 /* C3.4 Data processing - immediate */
2976 static void disas_data_proc_imm(DisasContext *s, uint32_t insn)
2978 switch (extract32(insn, 23, 6)) {
2979 case 0x20: case 0x21: /* PC-rel. addressing */
2980 disas_pc_rel_adr(s, insn);
2982 case 0x22: case 0x23: /* Add/subtract (immediate) */
2983 disas_add_sub_imm(s, insn);
2985 case 0x24: /* Logical (immediate) */
2986 disas_logic_imm(s, insn);
2988 case 0x25: /* Move wide (immediate) */
2989 disas_movw_imm(s, insn);
2991 case 0x26: /* Bitfield */
2992 disas_bitfield(s, insn);
2994 case 0x27: /* Extract */
2995 disas_extract(s, insn);
2998 unallocated_encoding(s);
3003 /* Shift a TCGv src by TCGv shift_amount, put result in dst.
3004 * Note that it is the caller's responsibility to ensure that the
3005 * shift amount is in range (ie 0..31 or 0..63) and provide the ARM
3006 * mandated semantics for out of range shifts.
3008 static void shift_reg(TCGv_i64 dst, TCGv_i64 src, int sf,
3009 enum a64_shift_type shift_type, TCGv_i64 shift_amount)
3011 switch (shift_type) {
3012 case A64_SHIFT_TYPE_LSL:
3013 tcg_gen_shl_i64(dst, src, shift_amount);
3015 case A64_SHIFT_TYPE_LSR:
3016 tcg_gen_shr_i64(dst, src, shift_amount);
3018 case A64_SHIFT_TYPE_ASR:
3020 tcg_gen_ext32s_i64(dst, src);
3022 tcg_gen_sar_i64(dst, sf ? src : dst, shift_amount);
3024 case A64_SHIFT_TYPE_ROR:
3026 tcg_gen_rotr_i64(dst, src, shift_amount);
3029 t0 = tcg_temp_new_i32();
3030 t1 = tcg_temp_new_i32();
3031 tcg_gen_trunc_i64_i32(t0, src);
3032 tcg_gen_trunc_i64_i32(t1, shift_amount);
3033 tcg_gen_rotr_i32(t0, t0, t1);
3034 tcg_gen_extu_i32_i64(dst, t0);
3035 tcg_temp_free_i32(t0);
3036 tcg_temp_free_i32(t1);
3040 assert(FALSE); /* all shift types should be handled */
3044 if (!sf) { /* zero extend final result */
3045 tcg_gen_ext32u_i64(dst, dst);
3049 /* Shift a TCGv src by immediate, put result in dst.
3050 * The shift amount must be in range (this should always be true as the
3051 * relevant instructions will UNDEF on bad shift immediates).
3053 static void shift_reg_imm(TCGv_i64 dst, TCGv_i64 src, int sf,
3054 enum a64_shift_type shift_type, unsigned int shift_i)
3056 assert(shift_i < (sf ? 64 : 32));
3059 tcg_gen_mov_i64(dst, src);
3061 TCGv_i64 shift_const;
3063 shift_const = tcg_const_i64(shift_i);
3064 shift_reg(dst, src, sf, shift_type, shift_const);
3065 tcg_temp_free_i64(shift_const);
3069 /* C3.5.10 Logical (shifted register)
3070 * 31 30 29 28 24 23 22 21 20 16 15 10 9 5 4 0
3071 * +----+-----+-----------+-------+---+------+--------+------+------+
3072 * | sf | opc | 0 1 0 1 0 | shift | N | Rm | imm6 | Rn | Rd |
3073 * +----+-----+-----------+-------+---+------+--------+------+------+
3075 static void disas_logic_reg(DisasContext *s, uint32_t insn)
3077 TCGv_i64 tcg_rd, tcg_rn, tcg_rm;
3078 unsigned int sf, opc, shift_type, invert, rm, shift_amount, rn, rd;
3080 sf = extract32(insn, 31, 1);
3081 opc = extract32(insn, 29, 2);
3082 shift_type = extract32(insn, 22, 2);
3083 invert = extract32(insn, 21, 1);
3084 rm = extract32(insn, 16, 5);
3085 shift_amount = extract32(insn, 10, 6);
3086 rn = extract32(insn, 5, 5);
3087 rd = extract32(insn, 0, 5);
3089 if (!sf && (shift_amount & (1 << 5))) {
3090 unallocated_encoding(s);
3094 tcg_rd = cpu_reg(s, rd);
3096 if (opc == 1 && shift_amount == 0 && shift_type == 0 && rn == 31) {
3097 /* Unshifted ORR and ORN with WZR/XZR is the standard encoding for
3098 * register-register MOV and MVN, so it is worth special casing.
3100 tcg_rm = cpu_reg(s, rm);
3102 tcg_gen_not_i64(tcg_rd, tcg_rm);
3104 tcg_gen_ext32u_i64(tcg_rd, tcg_rd);
3108 tcg_gen_mov_i64(tcg_rd, tcg_rm);
3110 tcg_gen_ext32u_i64(tcg_rd, tcg_rm);
3116 tcg_rm = read_cpu_reg(s, rm, sf);
3119 shift_reg_imm(tcg_rm, tcg_rm, sf, shift_type, shift_amount);
3122 tcg_rn = cpu_reg(s, rn);
3124 switch (opc | (invert << 2)) {
3127 tcg_gen_and_i64(tcg_rd, tcg_rn, tcg_rm);
3130 tcg_gen_or_i64(tcg_rd, tcg_rn, tcg_rm);
3133 tcg_gen_xor_i64(tcg_rd, tcg_rn, tcg_rm);
3137 tcg_gen_andc_i64(tcg_rd, tcg_rn, tcg_rm);
3140 tcg_gen_orc_i64(tcg_rd, tcg_rn, tcg_rm);
3143 tcg_gen_eqv_i64(tcg_rd, tcg_rn, tcg_rm);
3151 tcg_gen_ext32u_i64(tcg_rd, tcg_rd);
3155 gen_logic_CC(sf, tcg_rd);
3160 * C3.5.1 Add/subtract (extended register)
3162 * 31|30|29|28 24|23 22|21|20 16|15 13|12 10|9 5|4 0|
3163 * +--+--+--+-----------+-----+--+-------+------+------+----+----+
3164 * |sf|op| S| 0 1 0 1 1 | opt | 1| Rm |option| imm3 | Rn | Rd |
3165 * +--+--+--+-----------+-----+--+-------+------+------+----+----+
3167 * sf: 0 -> 32bit, 1 -> 64bit
3168 * op: 0 -> add , 1 -> sub
3171 * option: extension type (see DecodeRegExtend)
3172 * imm3: optional shift to Rm
3174 * Rd = Rn + LSL(extend(Rm), amount)
3176 static void disas_add_sub_ext_reg(DisasContext *s, uint32_t insn)
3178 int rd = extract32(insn, 0, 5);
3179 int rn = extract32(insn, 5, 5);
3180 int imm3 = extract32(insn, 10, 3);
3181 int option = extract32(insn, 13, 3);
3182 int rm = extract32(insn, 16, 5);
3183 bool setflags = extract32(insn, 29, 1);
3184 bool sub_op = extract32(insn, 30, 1);
3185 bool sf = extract32(insn, 31, 1);
3187 TCGv_i64 tcg_rm, tcg_rn; /* temps */
3189 TCGv_i64 tcg_result;
3192 unallocated_encoding(s);
3196 /* non-flag setting ops may use SP */
3198 tcg_rd = cpu_reg_sp(s, rd);
3200 tcg_rd = cpu_reg(s, rd);
3202 tcg_rn = read_cpu_reg_sp(s, rn, sf);
3204 tcg_rm = read_cpu_reg(s, rm, sf);
3205 ext_and_shift_reg(tcg_rm, tcg_rm, option, imm3);
3207 tcg_result = tcg_temp_new_i64();
3211 tcg_gen_sub_i64(tcg_result, tcg_rn, tcg_rm);
3213 tcg_gen_add_i64(tcg_result, tcg_rn, tcg_rm);
3217 gen_sub_CC(sf, tcg_result, tcg_rn, tcg_rm);
3219 gen_add_CC(sf, tcg_result, tcg_rn, tcg_rm);
3224 tcg_gen_mov_i64(tcg_rd, tcg_result);
3226 tcg_gen_ext32u_i64(tcg_rd, tcg_result);
3229 tcg_temp_free_i64(tcg_result);
3233 * C3.5.2 Add/subtract (shifted register)
3235 * 31 30 29 28 24 23 22 21 20 16 15 10 9 5 4 0
3236 * +--+--+--+-----------+-----+--+-------+---------+------+------+
3237 * |sf|op| S| 0 1 0 1 1 |shift| 0| Rm | imm6 | Rn | Rd |
3238 * +--+--+--+-----------+-----+--+-------+---------+------+------+
3240 * sf: 0 -> 32bit, 1 -> 64bit
3241 * op: 0 -> add , 1 -> sub
3243 * shift: 00 -> LSL, 01 -> LSR, 10 -> ASR, 11 -> RESERVED
3244 * imm6: Shift amount to apply to Rm before the add/sub
3246 static void disas_add_sub_reg(DisasContext *s, uint32_t insn)
3248 int rd = extract32(insn, 0, 5);
3249 int rn = extract32(insn, 5, 5);
3250 int imm6 = extract32(insn, 10, 6);
3251 int rm = extract32(insn, 16, 5);
3252 int shift_type = extract32(insn, 22, 2);
3253 bool setflags = extract32(insn, 29, 1);
3254 bool sub_op = extract32(insn, 30, 1);
3255 bool sf = extract32(insn, 31, 1);
3257 TCGv_i64 tcg_rd = cpu_reg(s, rd);
3258 TCGv_i64 tcg_rn, tcg_rm;
3259 TCGv_i64 tcg_result;
3261 if ((shift_type == 3) || (!sf && (imm6 > 31))) {
3262 unallocated_encoding(s);
3266 tcg_rn = read_cpu_reg(s, rn, sf);
3267 tcg_rm = read_cpu_reg(s, rm, sf);
3269 shift_reg_imm(tcg_rm, tcg_rm, sf, shift_type, imm6);
3271 tcg_result = tcg_temp_new_i64();
3275 tcg_gen_sub_i64(tcg_result, tcg_rn, tcg_rm);
3277 tcg_gen_add_i64(tcg_result, tcg_rn, tcg_rm);
3281 gen_sub_CC(sf, tcg_result, tcg_rn, tcg_rm);
3283 gen_add_CC(sf, tcg_result, tcg_rn, tcg_rm);
3288 tcg_gen_mov_i64(tcg_rd, tcg_result);
3290 tcg_gen_ext32u_i64(tcg_rd, tcg_result);
3293 tcg_temp_free_i64(tcg_result);
3296 /* C3.5.9 Data-processing (3 source)
3298 31 30 29 28 24 23 21 20 16 15 14 10 9 5 4 0
3299 +--+------+-----------+------+------+----+------+------+------+
3300 |sf| op54 | 1 1 0 1 1 | op31 | Rm | o0 | Ra | Rn | Rd |
3301 +--+------+-----------+------+------+----+------+------+------+
3304 static void disas_data_proc_3src(DisasContext *s, uint32_t insn)
3306 int rd = extract32(insn, 0, 5);
3307 int rn = extract32(insn, 5, 5);
3308 int ra = extract32(insn, 10, 5);
3309 int rm = extract32(insn, 16, 5);
3310 int op_id = (extract32(insn, 29, 3) << 4) |
3311 (extract32(insn, 21, 3) << 1) |
3312 extract32(insn, 15, 1);
3313 bool sf = extract32(insn, 31, 1);
3314 bool is_sub = extract32(op_id, 0, 1);
3315 bool is_high = extract32(op_id, 2, 1);
3316 bool is_signed = false;
3321 /* Note that op_id is sf:op54:op31:o0 so it includes the 32/64 size flag */
3323 case 0x42: /* SMADDL */
3324 case 0x43: /* SMSUBL */
3325 case 0x44: /* SMULH */
3328 case 0x0: /* MADD (32bit) */
3329 case 0x1: /* MSUB (32bit) */
3330 case 0x40: /* MADD (64bit) */
3331 case 0x41: /* MSUB (64bit) */
3332 case 0x4a: /* UMADDL */
3333 case 0x4b: /* UMSUBL */
3334 case 0x4c: /* UMULH */
3337 unallocated_encoding(s);
3342 TCGv_i64 low_bits = tcg_temp_new_i64(); /* low bits discarded */
3343 TCGv_i64 tcg_rd = cpu_reg(s, rd);
3344 TCGv_i64 tcg_rn = cpu_reg(s, rn);
3345 TCGv_i64 tcg_rm = cpu_reg(s, rm);
3348 tcg_gen_muls2_i64(low_bits, tcg_rd, tcg_rn, tcg_rm);
3350 tcg_gen_mulu2_i64(low_bits, tcg_rd, tcg_rn, tcg_rm);
3353 tcg_temp_free_i64(low_bits);
3357 tcg_op1 = tcg_temp_new_i64();
3358 tcg_op2 = tcg_temp_new_i64();
3359 tcg_tmp = tcg_temp_new_i64();
3362 tcg_gen_mov_i64(tcg_op1, cpu_reg(s, rn));
3363 tcg_gen_mov_i64(tcg_op2, cpu_reg(s, rm));
3366 tcg_gen_ext32s_i64(tcg_op1, cpu_reg(s, rn));
3367 tcg_gen_ext32s_i64(tcg_op2, cpu_reg(s, rm));
3369 tcg_gen_ext32u_i64(tcg_op1, cpu_reg(s, rn));
3370 tcg_gen_ext32u_i64(tcg_op2, cpu_reg(s, rm));
3374 if (ra == 31 && !is_sub) {
3375 /* Special-case MADD with rA == XZR; it is the standard MUL alias */
3376 tcg_gen_mul_i64(cpu_reg(s, rd), tcg_op1, tcg_op2);
3378 tcg_gen_mul_i64(tcg_tmp, tcg_op1, tcg_op2);
3380 tcg_gen_sub_i64(cpu_reg(s, rd), cpu_reg(s, ra), tcg_tmp);
3382 tcg_gen_add_i64(cpu_reg(s, rd), cpu_reg(s, ra), tcg_tmp);
3387 tcg_gen_ext32u_i64(cpu_reg(s, rd), cpu_reg(s, rd));
3390 tcg_temp_free_i64(tcg_op1);
3391 tcg_temp_free_i64(tcg_op2);
3392 tcg_temp_free_i64(tcg_tmp);
3395 /* C3.5.3 - Add/subtract (with carry)
3396 * 31 30 29 28 27 26 25 24 23 22 21 20 16 15 10 9 5 4 0
3397 * +--+--+--+------------------------+------+---------+------+-----+
3398 * |sf|op| S| 1 1 0 1 0 0 0 0 | rm | opcode2 | Rn | Rd |
3399 * +--+--+--+------------------------+------+---------+------+-----+
3403 static void disas_adc_sbc(DisasContext *s, uint32_t insn)
3405 unsigned int sf, op, setflags, rm, rn, rd;
3406 TCGv_i64 tcg_y, tcg_rn, tcg_rd;
3408 if (extract32(insn, 10, 6) != 0) {
3409 unallocated_encoding(s);
3413 sf = extract32(insn, 31, 1);
3414 op = extract32(insn, 30, 1);
3415 setflags = extract32(insn, 29, 1);
3416 rm = extract32(insn, 16, 5);
3417 rn = extract32(insn, 5, 5);
3418 rd = extract32(insn, 0, 5);
3420 tcg_rd = cpu_reg(s, rd);
3421 tcg_rn = cpu_reg(s, rn);
3424 tcg_y = new_tmp_a64(s);
3425 tcg_gen_not_i64(tcg_y, cpu_reg(s, rm));
3427 tcg_y = cpu_reg(s, rm);
3431 gen_adc_CC(sf, tcg_rd, tcg_rn, tcg_y);
3433 gen_adc(sf, tcg_rd, tcg_rn, tcg_y);
3437 /* C3.5.4 - C3.5.5 Conditional compare (immediate / register)
3438 * 31 30 29 28 27 26 25 24 23 22 21 20 16 15 12 11 10 9 5 4 3 0
3439 * +--+--+--+------------------------+--------+------+----+--+------+--+-----+
3440 * |sf|op| S| 1 1 0 1 0 0 1 0 |imm5/rm | cond |i/r |o2| Rn |o3|nzcv |
3441 * +--+--+--+------------------------+--------+------+----+--+------+--+-----+
3444 static void disas_cc(DisasContext *s, uint32_t insn)
3446 unsigned int sf, op, y, cond, rn, nzcv, is_imm;
3447 int label_continue = -1;
3448 TCGv_i64 tcg_tmp, tcg_y, tcg_rn;
3450 if (!extract32(insn, 29, 1)) {
3451 unallocated_encoding(s);
3454 if (insn & (1 << 10 | 1 << 4)) {
3455 unallocated_encoding(s);
3458 sf = extract32(insn, 31, 1);
3459 op = extract32(insn, 30, 1);
3460 is_imm = extract32(insn, 11, 1);
3461 y = extract32(insn, 16, 5); /* y = rm (reg) or imm5 (imm) */
3462 cond = extract32(insn, 12, 4);
3463 rn = extract32(insn, 5, 5);
3464 nzcv = extract32(insn, 0, 4);
3466 if (cond < 0x0e) { /* not always */
3467 int label_match = gen_new_label();
3468 label_continue = gen_new_label();
3469 arm_gen_test_cc(cond, label_match);
3471 tcg_tmp = tcg_temp_new_i64();
3472 tcg_gen_movi_i64(tcg_tmp, nzcv << 28);
3473 gen_set_nzcv(tcg_tmp);
3474 tcg_temp_free_i64(tcg_tmp);
3475 tcg_gen_br(label_continue);
3476 gen_set_label(label_match);
3478 /* match, or condition is always */
3480 tcg_y = new_tmp_a64(s);
3481 tcg_gen_movi_i64(tcg_y, y);
3483 tcg_y = cpu_reg(s, y);
3485 tcg_rn = cpu_reg(s, rn);
3487 tcg_tmp = tcg_temp_new_i64();
3489 gen_sub_CC(sf, tcg_tmp, tcg_rn, tcg_y);
3491 gen_add_CC(sf, tcg_tmp, tcg_rn, tcg_y);
3493 tcg_temp_free_i64(tcg_tmp);
3495 if (cond < 0x0e) { /* continue */
3496 gen_set_label(label_continue);
3500 /* C3.5.6 Conditional select
3501 * 31 30 29 28 21 20 16 15 12 11 10 9 5 4 0
3502 * +----+----+---+-----------------+------+------+-----+------+------+
3503 * | sf | op | S | 1 1 0 1 0 1 0 0 | Rm | cond | op2 | Rn | Rd |
3504 * +----+----+---+-----------------+------+------+-----+------+------+
3506 static void disas_cond_select(DisasContext *s, uint32_t insn)
3508 unsigned int sf, else_inv, rm, cond, else_inc, rn, rd;
3509 TCGv_i64 tcg_rd, tcg_src;
3511 if (extract32(insn, 29, 1) || extract32(insn, 11, 1)) {
3512 /* S == 1 or op2<1> == 1 */
3513 unallocated_encoding(s);
3516 sf = extract32(insn, 31, 1);
3517 else_inv = extract32(insn, 30, 1);
3518 rm = extract32(insn, 16, 5);
3519 cond = extract32(insn, 12, 4);
3520 else_inc = extract32(insn, 10, 1);
3521 rn = extract32(insn, 5, 5);
3522 rd = extract32(insn, 0, 5);
3525 /* silly no-op write; until we use movcond we must special-case
3526 * this to avoid a dead temporary across basic blocks.
3531 tcg_rd = cpu_reg(s, rd);
3533 if (cond >= 0x0e) { /* condition "always" */
3534 tcg_src = read_cpu_reg(s, rn, sf);
3535 tcg_gen_mov_i64(tcg_rd, tcg_src);
3537 /* OPTME: we could use movcond here, at the cost of duplicating
3538 * a lot of the arm_gen_test_cc() logic.
3540 int label_match = gen_new_label();
3541 int label_continue = gen_new_label();
3543 arm_gen_test_cc(cond, label_match);
3545 tcg_src = cpu_reg(s, rm);
3547 if (else_inv && else_inc) {
3548 tcg_gen_neg_i64(tcg_rd, tcg_src);
3549 } else if (else_inv) {
3550 tcg_gen_not_i64(tcg_rd, tcg_src);
3551 } else if (else_inc) {
3552 tcg_gen_addi_i64(tcg_rd, tcg_src, 1);
3554 tcg_gen_mov_i64(tcg_rd, tcg_src);
3557 tcg_gen_ext32u_i64(tcg_rd, tcg_rd);
3559 tcg_gen_br(label_continue);
3561 gen_set_label(label_match);
3562 tcg_src = read_cpu_reg(s, rn, sf);
3563 tcg_gen_mov_i64(tcg_rd, tcg_src);
3565 gen_set_label(label_continue);
3569 static void handle_clz(DisasContext *s, unsigned int sf,
3570 unsigned int rn, unsigned int rd)
3572 TCGv_i64 tcg_rd, tcg_rn;
3573 tcg_rd = cpu_reg(s, rd);
3574 tcg_rn = cpu_reg(s, rn);
3577 gen_helper_clz64(tcg_rd, tcg_rn);
3579 TCGv_i32 tcg_tmp32 = tcg_temp_new_i32();
3580 tcg_gen_trunc_i64_i32(tcg_tmp32, tcg_rn);
3581 gen_helper_clz(tcg_tmp32, tcg_tmp32);
3582 tcg_gen_extu_i32_i64(tcg_rd, tcg_tmp32);
3583 tcg_temp_free_i32(tcg_tmp32);
3587 static void handle_cls(DisasContext *s, unsigned int sf,
3588 unsigned int rn, unsigned int rd)
3590 TCGv_i64 tcg_rd, tcg_rn;
3591 tcg_rd = cpu_reg(s, rd);
3592 tcg_rn = cpu_reg(s, rn);
3595 gen_helper_cls64(tcg_rd, tcg_rn);
3597 TCGv_i32 tcg_tmp32 = tcg_temp_new_i32();
3598 tcg_gen_trunc_i64_i32(tcg_tmp32, tcg_rn);
3599 gen_helper_cls32(tcg_tmp32, tcg_tmp32);
3600 tcg_gen_extu_i32_i64(tcg_rd, tcg_tmp32);
3601 tcg_temp_free_i32(tcg_tmp32);
3605 static void handle_rbit(DisasContext *s, unsigned int sf,
3606 unsigned int rn, unsigned int rd)
3608 TCGv_i64 tcg_rd, tcg_rn;
3609 tcg_rd = cpu_reg(s, rd);
3610 tcg_rn = cpu_reg(s, rn);
3613 gen_helper_rbit64(tcg_rd, tcg_rn);
3615 TCGv_i32 tcg_tmp32 = tcg_temp_new_i32();
3616 tcg_gen_trunc_i64_i32(tcg_tmp32, tcg_rn);
3617 gen_helper_rbit(tcg_tmp32, tcg_tmp32);
3618 tcg_gen_extu_i32_i64(tcg_rd, tcg_tmp32);
3619 tcg_temp_free_i32(tcg_tmp32);
3623 /* C5.6.149 REV with sf==1, opcode==3 ("REV64") */
3624 static void handle_rev64(DisasContext *s, unsigned int sf,
3625 unsigned int rn, unsigned int rd)
3628 unallocated_encoding(s);
3631 tcg_gen_bswap64_i64(cpu_reg(s, rd), cpu_reg(s, rn));
3634 /* C5.6.149 REV with sf==0, opcode==2
3635 * C5.6.151 REV32 (sf==1, opcode==2)
3637 static void handle_rev32(DisasContext *s, unsigned int sf,
3638 unsigned int rn, unsigned int rd)
3640 TCGv_i64 tcg_rd = cpu_reg(s, rd);
3643 TCGv_i64 tcg_tmp = tcg_temp_new_i64();
3644 TCGv_i64 tcg_rn = read_cpu_reg(s, rn, sf);
3646 /* bswap32_i64 requires zero high word */
3647 tcg_gen_ext32u_i64(tcg_tmp, tcg_rn);
3648 tcg_gen_bswap32_i64(tcg_rd, tcg_tmp);
3649 tcg_gen_shri_i64(tcg_tmp, tcg_rn, 32);
3650 tcg_gen_bswap32_i64(tcg_tmp, tcg_tmp);
3651 tcg_gen_concat32_i64(tcg_rd, tcg_rd, tcg_tmp);
3653 tcg_temp_free_i64(tcg_tmp);
3655 tcg_gen_ext32u_i64(tcg_rd, cpu_reg(s, rn));
3656 tcg_gen_bswap32_i64(tcg_rd, tcg_rd);
3660 /* C5.6.150 REV16 (opcode==1) */
3661 static void handle_rev16(DisasContext *s, unsigned int sf,
3662 unsigned int rn, unsigned int rd)
3664 TCGv_i64 tcg_rd = cpu_reg(s, rd);
3665 TCGv_i64 tcg_tmp = tcg_temp_new_i64();
3666 TCGv_i64 tcg_rn = read_cpu_reg(s, rn, sf);
3668 tcg_gen_andi_i64(tcg_tmp, tcg_rn, 0xffff);
3669 tcg_gen_bswap16_i64(tcg_rd, tcg_tmp);
3671 tcg_gen_shri_i64(tcg_tmp, tcg_rn, 16);
3672 tcg_gen_andi_i64(tcg_tmp, tcg_tmp, 0xffff);
3673 tcg_gen_bswap16_i64(tcg_tmp, tcg_tmp);
3674 tcg_gen_deposit_i64(tcg_rd, tcg_rd, tcg_tmp, 16, 16);
3677 tcg_gen_shri_i64(tcg_tmp, tcg_rn, 32);
3678 tcg_gen_andi_i64(tcg_tmp, tcg_tmp, 0xffff);
3679 tcg_gen_bswap16_i64(tcg_tmp, tcg_tmp);
3680 tcg_gen_deposit_i64(tcg_rd, tcg_rd, tcg_tmp, 32, 16);
3682 tcg_gen_shri_i64(tcg_tmp, tcg_rn, 48);
3683 tcg_gen_bswap16_i64(tcg_tmp, tcg_tmp);
3684 tcg_gen_deposit_i64(tcg_rd, tcg_rd, tcg_tmp, 48, 16);
3687 tcg_temp_free_i64(tcg_tmp);
3690 /* C3.5.7 Data-processing (1 source)
3691 * 31 30 29 28 21 20 16 15 10 9 5 4 0
3692 * +----+---+---+-----------------+---------+--------+------+------+
3693 * | sf | 1 | S | 1 1 0 1 0 1 1 0 | opcode2 | opcode | Rn | Rd |
3694 * +----+---+---+-----------------+---------+--------+------+------+
3696 static void disas_data_proc_1src(DisasContext *s, uint32_t insn)
3698 unsigned int sf, opcode, rn, rd;
3700 if (extract32(insn, 29, 1) || extract32(insn, 16, 5)) {
3701 unallocated_encoding(s);
3705 sf = extract32(insn, 31, 1);
3706 opcode = extract32(insn, 10, 6);
3707 rn = extract32(insn, 5, 5);
3708 rd = extract32(insn, 0, 5);
3712 handle_rbit(s, sf, rn, rd);
3715 handle_rev16(s, sf, rn, rd);
3718 handle_rev32(s, sf, rn, rd);
3721 handle_rev64(s, sf, rn, rd);
3724 handle_clz(s, sf, rn, rd);
3727 handle_cls(s, sf, rn, rd);
3732 static void handle_div(DisasContext *s, bool is_signed, unsigned int sf,
3733 unsigned int rm, unsigned int rn, unsigned int rd)
3735 TCGv_i64 tcg_n, tcg_m, tcg_rd;
3736 tcg_rd = cpu_reg(s, rd);
3738 if (!sf && is_signed) {
3739 tcg_n = new_tmp_a64(s);
3740 tcg_m = new_tmp_a64(s);
3741 tcg_gen_ext32s_i64(tcg_n, cpu_reg(s, rn));
3742 tcg_gen_ext32s_i64(tcg_m, cpu_reg(s, rm));
3744 tcg_n = read_cpu_reg(s, rn, sf);
3745 tcg_m = read_cpu_reg(s, rm, sf);
3749 gen_helper_sdiv64(tcg_rd, tcg_n, tcg_m);
3751 gen_helper_udiv64(tcg_rd, tcg_n, tcg_m);
3754 if (!sf) { /* zero extend final result */
3755 tcg_gen_ext32u_i64(tcg_rd, tcg_rd);
3759 /* C5.6.115 LSLV, C5.6.118 LSRV, C5.6.17 ASRV, C5.6.154 RORV */
3760 static void handle_shift_reg(DisasContext *s,
3761 enum a64_shift_type shift_type, unsigned int sf,
3762 unsigned int rm, unsigned int rn, unsigned int rd)
3764 TCGv_i64 tcg_shift = tcg_temp_new_i64();
3765 TCGv_i64 tcg_rd = cpu_reg(s, rd);
3766 TCGv_i64 tcg_rn = read_cpu_reg(s, rn, sf);
3768 tcg_gen_andi_i64(tcg_shift, cpu_reg(s, rm), sf ? 63 : 31);
3769 shift_reg(tcg_rd, tcg_rn, sf, shift_type, tcg_shift);
3770 tcg_temp_free_i64(tcg_shift);
3773 /* C3.5.8 Data-processing (2 source)
3774 * 31 30 29 28 21 20 16 15 10 9 5 4 0
3775 * +----+---+---+-----------------+------+--------+------+------+
3776 * | sf | 0 | S | 1 1 0 1 0 1 1 0 | Rm | opcode | Rn | Rd |
3777 * +----+---+---+-----------------+------+--------+------+------+
3779 static void disas_data_proc_2src(DisasContext *s, uint32_t insn)
3781 unsigned int sf, rm, opcode, rn, rd;
3782 sf = extract32(insn, 31, 1);
3783 rm = extract32(insn, 16, 5);
3784 opcode = extract32(insn, 10, 6);
3785 rn = extract32(insn, 5, 5);
3786 rd = extract32(insn, 0, 5);
3788 if (extract32(insn, 29, 1)) {
3789 unallocated_encoding(s);
3795 handle_div(s, false, sf, rm, rn, rd);
3798 handle_div(s, true, sf, rm, rn, rd);
3801 handle_shift_reg(s, A64_SHIFT_TYPE_LSL, sf, rm, rn, rd);
3804 handle_shift_reg(s, A64_SHIFT_TYPE_LSR, sf, rm, rn, rd);
3807 handle_shift_reg(s, A64_SHIFT_TYPE_ASR, sf, rm, rn, rd);
3810 handle_shift_reg(s, A64_SHIFT_TYPE_ROR, sf, rm, rn, rd);
3819 case 23: /* CRC32 */
3820 unsupported_encoding(s, insn);
3823 unallocated_encoding(s);
3828 /* C3.5 Data processing - register */
3829 static void disas_data_proc_reg(DisasContext *s, uint32_t insn)
3831 switch (extract32(insn, 24, 5)) {
3832 case 0x0a: /* Logical (shifted register) */
3833 disas_logic_reg(s, insn);
3835 case 0x0b: /* Add/subtract */
3836 if (insn & (1 << 21)) { /* (extended register) */
3837 disas_add_sub_ext_reg(s, insn);
3839 disas_add_sub_reg(s, insn);
3842 case 0x1b: /* Data-processing (3 source) */
3843 disas_data_proc_3src(s, insn);
3846 switch (extract32(insn, 21, 3)) {
3847 case 0x0: /* Add/subtract (with carry) */
3848 disas_adc_sbc(s, insn);
3850 case 0x2: /* Conditional compare */
3851 disas_cc(s, insn); /* both imm and reg forms */
3853 case 0x4: /* Conditional select */
3854 disas_cond_select(s, insn);
3856 case 0x6: /* Data-processing */
3857 if (insn & (1 << 30)) { /* (1 source) */
3858 disas_data_proc_1src(s, insn);
3859 } else { /* (2 source) */
3860 disas_data_proc_2src(s, insn);
3864 unallocated_encoding(s);
3869 unallocated_encoding(s);
3874 static void handle_fp_compare(DisasContext *s, bool is_double,
3875 unsigned int rn, unsigned int rm,
3876 bool cmp_with_zero, bool signal_all_nans)
3878 TCGv_i64 tcg_flags = tcg_temp_new_i64();
3879 TCGv_ptr fpst = get_fpstatus_ptr();
3882 TCGv_i64 tcg_vn, tcg_vm;
3884 tcg_vn = read_fp_dreg(s, rn);
3885 if (cmp_with_zero) {
3886 tcg_vm = tcg_const_i64(0);
3888 tcg_vm = read_fp_dreg(s, rm);
3890 if (signal_all_nans) {
3891 gen_helper_vfp_cmped_a64(tcg_flags, tcg_vn, tcg_vm, fpst);
3893 gen_helper_vfp_cmpd_a64(tcg_flags, tcg_vn, tcg_vm, fpst);
3895 tcg_temp_free_i64(tcg_vn);
3896 tcg_temp_free_i64(tcg_vm);
3898 TCGv_i32 tcg_vn, tcg_vm;
3900 tcg_vn = read_fp_sreg(s, rn);
3901 if (cmp_with_zero) {
3902 tcg_vm = tcg_const_i32(0);
3904 tcg_vm = read_fp_sreg(s, rm);
3906 if (signal_all_nans) {
3907 gen_helper_vfp_cmpes_a64(tcg_flags, tcg_vn, tcg_vm, fpst);
3909 gen_helper_vfp_cmps_a64(tcg_flags, tcg_vn, tcg_vm, fpst);
3911 tcg_temp_free_i32(tcg_vn);
3912 tcg_temp_free_i32(tcg_vm);
3915 tcg_temp_free_ptr(fpst);
3917 gen_set_nzcv(tcg_flags);
3919 tcg_temp_free_i64(tcg_flags);
3922 /* C3.6.22 Floating point compare
3923 * 31 30 29 28 24 23 22 21 20 16 15 14 13 10 9 5 4 0
3924 * +---+---+---+-----------+------+---+------+-----+---------+------+-------+
3925 * | M | 0 | S | 1 1 1 1 0 | type | 1 | Rm | op | 1 0 0 0 | Rn | op2 |
3926 * +---+---+---+-----------+------+---+------+-----+---------+------+-------+
3928 static void disas_fp_compare(DisasContext *s, uint32_t insn)
3930 unsigned int mos, type, rm, op, rn, opc, op2r;
3932 mos = extract32(insn, 29, 3);
3933 type = extract32(insn, 22, 2); /* 0 = single, 1 = double */
3934 rm = extract32(insn, 16, 5);
3935 op = extract32(insn, 14, 2);
3936 rn = extract32(insn, 5, 5);
3937 opc = extract32(insn, 3, 2);
3938 op2r = extract32(insn, 0, 3);
3940 if (mos || op || op2r || type > 1) {
3941 unallocated_encoding(s);
3945 if (!fp_access_check(s)) {
3949 handle_fp_compare(s, type, rn, rm, opc & 1, opc & 2);
3952 /* C3.6.23 Floating point conditional compare
3953 * 31 30 29 28 24 23 22 21 20 16 15 12 11 10 9 5 4 3 0
3954 * +---+---+---+-----------+------+---+------+------+-----+------+----+------+
3955 * | M | 0 | S | 1 1 1 1 0 | type | 1 | Rm | cond | 0 1 | Rn | op | nzcv |
3956 * +---+---+---+-----------+------+---+------+------+-----+------+----+------+
3958 static void disas_fp_ccomp(DisasContext *s, uint32_t insn)
3960 unsigned int mos, type, rm, cond, rn, op, nzcv;
3962 int label_continue = -1;
3964 mos = extract32(insn, 29, 3);
3965 type = extract32(insn, 22, 2); /* 0 = single, 1 = double */
3966 rm = extract32(insn, 16, 5);
3967 cond = extract32(insn, 12, 4);
3968 rn = extract32(insn, 5, 5);
3969 op = extract32(insn, 4, 1);
3970 nzcv = extract32(insn, 0, 4);
3972 if (mos || type > 1) {
3973 unallocated_encoding(s);
3977 if (!fp_access_check(s)) {
3981 if (cond < 0x0e) { /* not always */
3982 int label_match = gen_new_label();
3983 label_continue = gen_new_label();
3984 arm_gen_test_cc(cond, label_match);
3986 tcg_flags = tcg_const_i64(nzcv << 28);
3987 gen_set_nzcv(tcg_flags);
3988 tcg_temp_free_i64(tcg_flags);
3989 tcg_gen_br(label_continue);
3990 gen_set_label(label_match);
3993 handle_fp_compare(s, type, rn, rm, false, op);
3996 gen_set_label(label_continue);
4000 /* copy src FP register to dst FP register; type specifies single or double */
4001 static void gen_mov_fp2fp(DisasContext *s, int type, int dst, int src)
4004 TCGv_i64 v = read_fp_dreg(s, src);
4005 write_fp_dreg(s, dst, v);
4006 tcg_temp_free_i64(v);
4008 TCGv_i32 v = read_fp_sreg(s, src);
4009 write_fp_sreg(s, dst, v);
4010 tcg_temp_free_i32(v);
4014 /* C3.6.24 Floating point conditional select
4015 * 31 30 29 28 24 23 22 21 20 16 15 12 11 10 9 5 4 0
4016 * +---+---+---+-----------+------+---+------+------+-----+------+------+
4017 * | M | 0 | S | 1 1 1 1 0 | type | 1 | Rm | cond | 1 1 | Rn | Rd |
4018 * +---+---+---+-----------+------+---+------+------+-----+------+------+
4020 static void disas_fp_csel(DisasContext *s, uint32_t insn)
4022 unsigned int mos, type, rm, cond, rn, rd;
4023 int label_continue = -1;
4025 mos = extract32(insn, 29, 3);
4026 type = extract32(insn, 22, 2); /* 0 = single, 1 = double */
4027 rm = extract32(insn, 16, 5);
4028 cond = extract32(insn, 12, 4);
4029 rn = extract32(insn, 5, 5);
4030 rd = extract32(insn, 0, 5);
4032 if (mos || type > 1) {
4033 unallocated_encoding(s);
4037 if (!fp_access_check(s)) {
4041 if (cond < 0x0e) { /* not always */
4042 int label_match = gen_new_label();
4043 label_continue = gen_new_label();
4044 arm_gen_test_cc(cond, label_match);
4046 gen_mov_fp2fp(s, type, rd, rm);
4047 tcg_gen_br(label_continue);
4048 gen_set_label(label_match);
4051 gen_mov_fp2fp(s, type, rd, rn);
4053 if (cond < 0x0e) { /* continue */
4054 gen_set_label(label_continue);
4058 /* C3.6.25 Floating-point data-processing (1 source) - single precision */
4059 static void handle_fp_1src_single(DisasContext *s, int opcode, int rd, int rn)
4065 fpst = get_fpstatus_ptr();
4066 tcg_op = read_fp_sreg(s, rn);
4067 tcg_res = tcg_temp_new_i32();
4070 case 0x0: /* FMOV */
4071 tcg_gen_mov_i32(tcg_res, tcg_op);
4073 case 0x1: /* FABS */
4074 gen_helper_vfp_abss(tcg_res, tcg_op);
4076 case 0x2: /* FNEG */
4077 gen_helper_vfp_negs(tcg_res, tcg_op);
4079 case 0x3: /* FSQRT */
4080 gen_helper_vfp_sqrts(tcg_res, tcg_op, cpu_env);
4082 case 0x8: /* FRINTN */
4083 case 0x9: /* FRINTP */
4084 case 0xa: /* FRINTM */
4085 case 0xb: /* FRINTZ */
4086 case 0xc: /* FRINTA */
4088 TCGv_i32 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(opcode & 7));
4090 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
4091 gen_helper_rints(tcg_res, tcg_op, fpst);
4093 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
4094 tcg_temp_free_i32(tcg_rmode);
4097 case 0xe: /* FRINTX */
4098 gen_helper_rints_exact(tcg_res, tcg_op, fpst);
4100 case 0xf: /* FRINTI */
4101 gen_helper_rints(tcg_res, tcg_op, fpst);
4107 write_fp_sreg(s, rd, tcg_res);
4109 tcg_temp_free_ptr(fpst);
4110 tcg_temp_free_i32(tcg_op);
4111 tcg_temp_free_i32(tcg_res);
4114 /* C3.6.25 Floating-point data-processing (1 source) - double precision */
4115 static void handle_fp_1src_double(DisasContext *s, int opcode, int rd, int rn)
4121 fpst = get_fpstatus_ptr();
4122 tcg_op = read_fp_dreg(s, rn);
4123 tcg_res = tcg_temp_new_i64();
4126 case 0x0: /* FMOV */
4127 tcg_gen_mov_i64(tcg_res, tcg_op);
4129 case 0x1: /* FABS */
4130 gen_helper_vfp_absd(tcg_res, tcg_op);
4132 case 0x2: /* FNEG */
4133 gen_helper_vfp_negd(tcg_res, tcg_op);
4135 case 0x3: /* FSQRT */
4136 gen_helper_vfp_sqrtd(tcg_res, tcg_op, cpu_env);
4138 case 0x8: /* FRINTN */
4139 case 0x9: /* FRINTP */
4140 case 0xa: /* FRINTM */
4141 case 0xb: /* FRINTZ */
4142 case 0xc: /* FRINTA */
4144 TCGv_i32 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(opcode & 7));
4146 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
4147 gen_helper_rintd(tcg_res, tcg_op, fpst);
4149 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
4150 tcg_temp_free_i32(tcg_rmode);
4153 case 0xe: /* FRINTX */
4154 gen_helper_rintd_exact(tcg_res, tcg_op, fpst);
4156 case 0xf: /* FRINTI */
4157 gen_helper_rintd(tcg_res, tcg_op, fpst);
4163 write_fp_dreg(s, rd, tcg_res);
4165 tcg_temp_free_ptr(fpst);
4166 tcg_temp_free_i64(tcg_op);
4167 tcg_temp_free_i64(tcg_res);
4170 static void handle_fp_fcvt(DisasContext *s, int opcode,
4171 int rd, int rn, int dtype, int ntype)
4176 TCGv_i32 tcg_rn = read_fp_sreg(s, rn);
4178 /* Single to double */
4179 TCGv_i64 tcg_rd = tcg_temp_new_i64();
4180 gen_helper_vfp_fcvtds(tcg_rd, tcg_rn, cpu_env);
4181 write_fp_dreg(s, rd, tcg_rd);
4182 tcg_temp_free_i64(tcg_rd);
4184 /* Single to half */
4185 TCGv_i32 tcg_rd = tcg_temp_new_i32();
4186 gen_helper_vfp_fcvt_f32_to_f16(tcg_rd, tcg_rn, cpu_env);
4187 /* write_fp_sreg is OK here because top half of tcg_rd is zero */
4188 write_fp_sreg(s, rd, tcg_rd);
4189 tcg_temp_free_i32(tcg_rd);
4191 tcg_temp_free_i32(tcg_rn);
4196 TCGv_i64 tcg_rn = read_fp_dreg(s, rn);
4197 TCGv_i32 tcg_rd = tcg_temp_new_i32();
4199 /* Double to single */
4200 gen_helper_vfp_fcvtsd(tcg_rd, tcg_rn, cpu_env);
4202 /* Double to half */
4203 gen_helper_vfp_fcvt_f64_to_f16(tcg_rd, tcg_rn, cpu_env);
4204 /* write_fp_sreg is OK here because top half of tcg_rd is zero */
4206 write_fp_sreg(s, rd, tcg_rd);
4207 tcg_temp_free_i32(tcg_rd);
4208 tcg_temp_free_i64(tcg_rn);
4213 TCGv_i32 tcg_rn = read_fp_sreg(s, rn);
4214 tcg_gen_ext16u_i32(tcg_rn, tcg_rn);
4216 /* Half to single */
4217 TCGv_i32 tcg_rd = tcg_temp_new_i32();
4218 gen_helper_vfp_fcvt_f16_to_f32(tcg_rd, tcg_rn, cpu_env);
4219 write_fp_sreg(s, rd, tcg_rd);
4220 tcg_temp_free_i32(tcg_rd);
4222 /* Half to double */
4223 TCGv_i64 tcg_rd = tcg_temp_new_i64();
4224 gen_helper_vfp_fcvt_f16_to_f64(tcg_rd, tcg_rn, cpu_env);
4225 write_fp_dreg(s, rd, tcg_rd);
4226 tcg_temp_free_i64(tcg_rd);
4228 tcg_temp_free_i32(tcg_rn);
4236 /* C3.6.25 Floating point data-processing (1 source)
4237 * 31 30 29 28 24 23 22 21 20 15 14 10 9 5 4 0
4238 * +---+---+---+-----------+------+---+--------+-----------+------+------+
4239 * | M | 0 | S | 1 1 1 1 0 | type | 1 | opcode | 1 0 0 0 0 | Rn | Rd |
4240 * +---+---+---+-----------+------+---+--------+-----------+------+------+
4242 static void disas_fp_1src(DisasContext *s, uint32_t insn)
4244 int type = extract32(insn, 22, 2);
4245 int opcode = extract32(insn, 15, 6);
4246 int rn = extract32(insn, 5, 5);
4247 int rd = extract32(insn, 0, 5);
4250 case 0x4: case 0x5: case 0x7:
4252 /* FCVT between half, single and double precision */
4253 int dtype = extract32(opcode, 0, 2);
4254 if (type == 2 || dtype == type) {
4255 unallocated_encoding(s);
4258 if (!fp_access_check(s)) {
4262 handle_fp_fcvt(s, opcode, rd, rn, dtype, type);
4268 /* 32-to-32 and 64-to-64 ops */
4271 if (!fp_access_check(s)) {
4275 handle_fp_1src_single(s, opcode, rd, rn);
4278 if (!fp_access_check(s)) {
4282 handle_fp_1src_double(s, opcode, rd, rn);
4285 unallocated_encoding(s);
4289 unallocated_encoding(s);
4294 /* C3.6.26 Floating-point data-processing (2 source) - single precision */
4295 static void handle_fp_2src_single(DisasContext *s, int opcode,
4296 int rd, int rn, int rm)
4303 tcg_res = tcg_temp_new_i32();
4304 fpst = get_fpstatus_ptr();
4305 tcg_op1 = read_fp_sreg(s, rn);
4306 tcg_op2 = read_fp_sreg(s, rm);
4309 case 0x0: /* FMUL */
4310 gen_helper_vfp_muls(tcg_res, tcg_op1, tcg_op2, fpst);
4312 case 0x1: /* FDIV */
4313 gen_helper_vfp_divs(tcg_res, tcg_op1, tcg_op2, fpst);
4315 case 0x2: /* FADD */
4316 gen_helper_vfp_adds(tcg_res, tcg_op1, tcg_op2, fpst);
4318 case 0x3: /* FSUB */
4319 gen_helper_vfp_subs(tcg_res, tcg_op1, tcg_op2, fpst);
4321 case 0x4: /* FMAX */
4322 gen_helper_vfp_maxs(tcg_res, tcg_op1, tcg_op2, fpst);
4324 case 0x5: /* FMIN */
4325 gen_helper_vfp_mins(tcg_res, tcg_op1, tcg_op2, fpst);
4327 case 0x6: /* FMAXNM */
4328 gen_helper_vfp_maxnums(tcg_res, tcg_op1, tcg_op2, fpst);
4330 case 0x7: /* FMINNM */
4331 gen_helper_vfp_minnums(tcg_res, tcg_op1, tcg_op2, fpst);
4333 case 0x8: /* FNMUL */
4334 gen_helper_vfp_muls(tcg_res, tcg_op1, tcg_op2, fpst);
4335 gen_helper_vfp_negs(tcg_res, tcg_res);
4339 write_fp_sreg(s, rd, tcg_res);
4341 tcg_temp_free_ptr(fpst);
4342 tcg_temp_free_i32(tcg_op1);
4343 tcg_temp_free_i32(tcg_op2);
4344 tcg_temp_free_i32(tcg_res);
4347 /* C3.6.26 Floating-point data-processing (2 source) - double precision */
4348 static void handle_fp_2src_double(DisasContext *s, int opcode,
4349 int rd, int rn, int rm)
4356 tcg_res = tcg_temp_new_i64();
4357 fpst = get_fpstatus_ptr();
4358 tcg_op1 = read_fp_dreg(s, rn);
4359 tcg_op2 = read_fp_dreg(s, rm);
4362 case 0x0: /* FMUL */
4363 gen_helper_vfp_muld(tcg_res, tcg_op1, tcg_op2, fpst);
4365 case 0x1: /* FDIV */
4366 gen_helper_vfp_divd(tcg_res, tcg_op1, tcg_op2, fpst);
4368 case 0x2: /* FADD */
4369 gen_helper_vfp_addd(tcg_res, tcg_op1, tcg_op2, fpst);
4371 case 0x3: /* FSUB */
4372 gen_helper_vfp_subd(tcg_res, tcg_op1, tcg_op2, fpst);
4374 case 0x4: /* FMAX */
4375 gen_helper_vfp_maxd(tcg_res, tcg_op1, tcg_op2, fpst);
4377 case 0x5: /* FMIN */
4378 gen_helper_vfp_mind(tcg_res, tcg_op1, tcg_op2, fpst);
4380 case 0x6: /* FMAXNM */
4381 gen_helper_vfp_maxnumd(tcg_res, tcg_op1, tcg_op2, fpst);
4383 case 0x7: /* FMINNM */
4384 gen_helper_vfp_minnumd(tcg_res, tcg_op1, tcg_op2, fpst);
4386 case 0x8: /* FNMUL */
4387 gen_helper_vfp_muld(tcg_res, tcg_op1, tcg_op2, fpst);
4388 gen_helper_vfp_negd(tcg_res, tcg_res);
4392 write_fp_dreg(s, rd, tcg_res);
4394 tcg_temp_free_ptr(fpst);
4395 tcg_temp_free_i64(tcg_op1);
4396 tcg_temp_free_i64(tcg_op2);
4397 tcg_temp_free_i64(tcg_res);
4400 /* C3.6.26 Floating point data-processing (2 source)
4401 * 31 30 29 28 24 23 22 21 20 16 15 12 11 10 9 5 4 0
4402 * +---+---+---+-----------+------+---+------+--------+-----+------+------+
4403 * | M | 0 | S | 1 1 1 1 0 | type | 1 | Rm | opcode | 1 0 | Rn | Rd |
4404 * +---+---+---+-----------+------+---+------+--------+-----+------+------+
4406 static void disas_fp_2src(DisasContext *s, uint32_t insn)
4408 int type = extract32(insn, 22, 2);
4409 int rd = extract32(insn, 0, 5);
4410 int rn = extract32(insn, 5, 5);
4411 int rm = extract32(insn, 16, 5);
4412 int opcode = extract32(insn, 12, 4);
4415 unallocated_encoding(s);
4421 if (!fp_access_check(s)) {
4424 handle_fp_2src_single(s, opcode, rd, rn, rm);
4427 if (!fp_access_check(s)) {
4430 handle_fp_2src_double(s, opcode, rd, rn, rm);
4433 unallocated_encoding(s);
4437 /* C3.6.27 Floating-point data-processing (3 source) - single precision */
4438 static void handle_fp_3src_single(DisasContext *s, bool o0, bool o1,
4439 int rd, int rn, int rm, int ra)
4441 TCGv_i32 tcg_op1, tcg_op2, tcg_op3;
4442 TCGv_i32 tcg_res = tcg_temp_new_i32();
4443 TCGv_ptr fpst = get_fpstatus_ptr();
4445 tcg_op1 = read_fp_sreg(s, rn);
4446 tcg_op2 = read_fp_sreg(s, rm);
4447 tcg_op3 = read_fp_sreg(s, ra);
4449 /* These are fused multiply-add, and must be done as one
4450 * floating point operation with no rounding between the
4451 * multiplication and addition steps.
4452 * NB that doing the negations here as separate steps is
4453 * correct : an input NaN should come out with its sign bit
4454 * flipped if it is a negated-input.
4457 gen_helper_vfp_negs(tcg_op3, tcg_op3);
4461 gen_helper_vfp_negs(tcg_op1, tcg_op1);
4464 gen_helper_vfp_muladds(tcg_res, tcg_op1, tcg_op2, tcg_op3, fpst);
4466 write_fp_sreg(s, rd, tcg_res);
4468 tcg_temp_free_ptr(fpst);
4469 tcg_temp_free_i32(tcg_op1);
4470 tcg_temp_free_i32(tcg_op2);
4471 tcg_temp_free_i32(tcg_op3);
4472 tcg_temp_free_i32(tcg_res);
4475 /* C3.6.27 Floating-point data-processing (3 source) - double precision */
4476 static void handle_fp_3src_double(DisasContext *s, bool o0, bool o1,
4477 int rd, int rn, int rm, int ra)
4479 TCGv_i64 tcg_op1, tcg_op2, tcg_op3;
4480 TCGv_i64 tcg_res = tcg_temp_new_i64();
4481 TCGv_ptr fpst = get_fpstatus_ptr();
4483 tcg_op1 = read_fp_dreg(s, rn);
4484 tcg_op2 = read_fp_dreg(s, rm);
4485 tcg_op3 = read_fp_dreg(s, ra);
4487 /* These are fused multiply-add, and must be done as one
4488 * floating point operation with no rounding between the
4489 * multiplication and addition steps.
4490 * NB that doing the negations here as separate steps is
4491 * correct : an input NaN should come out with its sign bit
4492 * flipped if it is a negated-input.
4495 gen_helper_vfp_negd(tcg_op3, tcg_op3);
4499 gen_helper_vfp_negd(tcg_op1, tcg_op1);
4502 gen_helper_vfp_muladdd(tcg_res, tcg_op1, tcg_op2, tcg_op3, fpst);
4504 write_fp_dreg(s, rd, tcg_res);
4506 tcg_temp_free_ptr(fpst);
4507 tcg_temp_free_i64(tcg_op1);
4508 tcg_temp_free_i64(tcg_op2);
4509 tcg_temp_free_i64(tcg_op3);
4510 tcg_temp_free_i64(tcg_res);
4513 /* C3.6.27 Floating point data-processing (3 source)
4514 * 31 30 29 28 24 23 22 21 20 16 15 14 10 9 5 4 0
4515 * +---+---+---+-----------+------+----+------+----+------+------+------+
4516 * | M | 0 | S | 1 1 1 1 1 | type | o1 | Rm | o0 | Ra | Rn | Rd |
4517 * +---+---+---+-----------+------+----+------+----+------+------+------+
4519 static void disas_fp_3src(DisasContext *s, uint32_t insn)
4521 int type = extract32(insn, 22, 2);
4522 int rd = extract32(insn, 0, 5);
4523 int rn = extract32(insn, 5, 5);
4524 int ra = extract32(insn, 10, 5);
4525 int rm = extract32(insn, 16, 5);
4526 bool o0 = extract32(insn, 15, 1);
4527 bool o1 = extract32(insn, 21, 1);
4531 if (!fp_access_check(s)) {
4534 handle_fp_3src_single(s, o0, o1, rd, rn, rm, ra);
4537 if (!fp_access_check(s)) {
4540 handle_fp_3src_double(s, o0, o1, rd, rn, rm, ra);
4543 unallocated_encoding(s);
4547 /* C3.6.28 Floating point immediate
4548 * 31 30 29 28 24 23 22 21 20 13 12 10 9 5 4 0
4549 * +---+---+---+-----------+------+---+------------+-------+------+------+
4550 * | M | 0 | S | 1 1 1 1 0 | type | 1 | imm8 | 1 0 0 | imm5 | Rd |
4551 * +---+---+---+-----------+------+---+------------+-------+------+------+
4553 static void disas_fp_imm(DisasContext *s, uint32_t insn)
4555 int rd = extract32(insn, 0, 5);
4556 int imm8 = extract32(insn, 13, 8);
4557 int is_double = extract32(insn, 22, 2);
4561 if (is_double > 1) {
4562 unallocated_encoding(s);
4566 if (!fp_access_check(s)) {
4570 /* The imm8 encodes the sign bit, enough bits to represent
4571 * an exponent in the range 01....1xx to 10....0xx,
4572 * and the most significant 4 bits of the mantissa; see
4573 * VFPExpandImm() in the v8 ARM ARM.
4576 imm = (extract32(imm8, 7, 1) ? 0x8000 : 0) |
4577 (extract32(imm8, 6, 1) ? 0x3fc0 : 0x4000) |
4578 extract32(imm8, 0, 6);
4581 imm = (extract32(imm8, 7, 1) ? 0x8000 : 0) |
4582 (extract32(imm8, 6, 1) ? 0x3e00 : 0x4000) |
4583 (extract32(imm8, 0, 6) << 3);
4587 tcg_res = tcg_const_i64(imm);
4588 write_fp_dreg(s, rd, tcg_res);
4589 tcg_temp_free_i64(tcg_res);
4592 /* Handle floating point <=> fixed point conversions. Note that we can
4593 * also deal with fp <=> integer conversions as a special case (scale == 64)
4594 * OPTME: consider handling that special case specially or at least skipping
4595 * the call to scalbn in the helpers for zero shifts.
4597 static void handle_fpfpcvt(DisasContext *s, int rd, int rn, int opcode,
4598 bool itof, int rmode, int scale, int sf, int type)
4600 bool is_signed = !(opcode & 1);
4601 bool is_double = type;
4602 TCGv_ptr tcg_fpstatus;
4605 tcg_fpstatus = get_fpstatus_ptr();
4607 tcg_shift = tcg_const_i32(64 - scale);
4610 TCGv_i64 tcg_int = cpu_reg(s, rn);
4612 TCGv_i64 tcg_extend = new_tmp_a64(s);
4615 tcg_gen_ext32s_i64(tcg_extend, tcg_int);
4617 tcg_gen_ext32u_i64(tcg_extend, tcg_int);
4620 tcg_int = tcg_extend;
4624 TCGv_i64 tcg_double = tcg_temp_new_i64();
4626 gen_helper_vfp_sqtod(tcg_double, tcg_int,
4627 tcg_shift, tcg_fpstatus);
4629 gen_helper_vfp_uqtod(tcg_double, tcg_int,
4630 tcg_shift, tcg_fpstatus);
4632 write_fp_dreg(s, rd, tcg_double);
4633 tcg_temp_free_i64(tcg_double);
4635 TCGv_i32 tcg_single = tcg_temp_new_i32();
4637 gen_helper_vfp_sqtos(tcg_single, tcg_int,
4638 tcg_shift, tcg_fpstatus);
4640 gen_helper_vfp_uqtos(tcg_single, tcg_int,
4641 tcg_shift, tcg_fpstatus);
4643 write_fp_sreg(s, rd, tcg_single);
4644 tcg_temp_free_i32(tcg_single);
4647 TCGv_i64 tcg_int = cpu_reg(s, rd);
4650 if (extract32(opcode, 2, 1)) {
4651 /* There are too many rounding modes to all fit into rmode,
4652 * so FCVTA[US] is a special case.
4654 rmode = FPROUNDING_TIEAWAY;
4657 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rmode));
4659 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
4662 TCGv_i64 tcg_double = read_fp_dreg(s, rn);
4665 gen_helper_vfp_tosld(tcg_int, tcg_double,
4666 tcg_shift, tcg_fpstatus);
4668 gen_helper_vfp_tosqd(tcg_int, tcg_double,
4669 tcg_shift, tcg_fpstatus);
4673 gen_helper_vfp_tould(tcg_int, tcg_double,
4674 tcg_shift, tcg_fpstatus);
4676 gen_helper_vfp_touqd(tcg_int, tcg_double,
4677 tcg_shift, tcg_fpstatus);
4680 tcg_temp_free_i64(tcg_double);
4682 TCGv_i32 tcg_single = read_fp_sreg(s, rn);
4685 gen_helper_vfp_tosqs(tcg_int, tcg_single,
4686 tcg_shift, tcg_fpstatus);
4688 gen_helper_vfp_touqs(tcg_int, tcg_single,
4689 tcg_shift, tcg_fpstatus);
4692 TCGv_i32 tcg_dest = tcg_temp_new_i32();
4694 gen_helper_vfp_tosls(tcg_dest, tcg_single,
4695 tcg_shift, tcg_fpstatus);
4697 gen_helper_vfp_touls(tcg_dest, tcg_single,
4698 tcg_shift, tcg_fpstatus);
4700 tcg_gen_extu_i32_i64(tcg_int, tcg_dest);
4701 tcg_temp_free_i32(tcg_dest);
4703 tcg_temp_free_i32(tcg_single);
4706 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
4707 tcg_temp_free_i32(tcg_rmode);
4710 tcg_gen_ext32u_i64(tcg_int, tcg_int);
4714 tcg_temp_free_ptr(tcg_fpstatus);
4715 tcg_temp_free_i32(tcg_shift);
4718 /* C3.6.29 Floating point <-> fixed point conversions
4719 * 31 30 29 28 24 23 22 21 20 19 18 16 15 10 9 5 4 0
4720 * +----+---+---+-----------+------+---+-------+--------+-------+------+------+
4721 * | sf | 0 | S | 1 1 1 1 0 | type | 0 | rmode | opcode | scale | Rn | Rd |
4722 * +----+---+---+-----------+------+---+-------+--------+-------+------+------+
4724 static void disas_fp_fixed_conv(DisasContext *s, uint32_t insn)
4726 int rd = extract32(insn, 0, 5);
4727 int rn = extract32(insn, 5, 5);
4728 int scale = extract32(insn, 10, 6);
4729 int opcode = extract32(insn, 16, 3);
4730 int rmode = extract32(insn, 19, 2);
4731 int type = extract32(insn, 22, 2);
4732 bool sbit = extract32(insn, 29, 1);
4733 bool sf = extract32(insn, 31, 1);
4736 if (sbit || (type > 1)
4737 || (!sf && scale < 32)) {
4738 unallocated_encoding(s);
4742 switch ((rmode << 3) | opcode) {
4743 case 0x2: /* SCVTF */
4744 case 0x3: /* UCVTF */
4747 case 0x18: /* FCVTZS */
4748 case 0x19: /* FCVTZU */
4752 unallocated_encoding(s);
4756 if (!fp_access_check(s)) {
4760 handle_fpfpcvt(s, rd, rn, opcode, itof, FPROUNDING_ZERO, scale, sf, type);
4763 static void handle_fmov(DisasContext *s, int rd, int rn, int type, bool itof)
4765 /* FMOV: gpr to or from float, double, or top half of quad fp reg,
4766 * without conversion.
4770 TCGv_i64 tcg_rn = cpu_reg(s, rn);
4776 TCGv_i64 tmp = tcg_temp_new_i64();
4777 tcg_gen_ext32u_i64(tmp, tcg_rn);
4778 tcg_gen_st_i64(tmp, cpu_env, fp_reg_offset(s, rd, MO_64));
4779 tcg_gen_movi_i64(tmp, 0);
4780 tcg_gen_st_i64(tmp, cpu_env, fp_reg_hi_offset(s, rd));
4781 tcg_temp_free_i64(tmp);
4787 TCGv_i64 tmp = tcg_const_i64(0);
4788 tcg_gen_st_i64(tcg_rn, cpu_env, fp_reg_offset(s, rd, MO_64));
4789 tcg_gen_st_i64(tmp, cpu_env, fp_reg_hi_offset(s, rd));
4790 tcg_temp_free_i64(tmp);
4794 /* 64 bit to top half. */
4795 tcg_gen_st_i64(tcg_rn, cpu_env, fp_reg_hi_offset(s, rd));
4799 TCGv_i64 tcg_rd = cpu_reg(s, rd);
4804 tcg_gen_ld32u_i64(tcg_rd, cpu_env, fp_reg_offset(s, rn, MO_32));
4808 tcg_gen_ld_i64(tcg_rd, cpu_env, fp_reg_offset(s, rn, MO_64));
4811 /* 64 bits from top half */
4812 tcg_gen_ld_i64(tcg_rd, cpu_env, fp_reg_hi_offset(s, rn));
4818 /* C3.6.30 Floating point <-> integer conversions
4819 * 31 30 29 28 24 23 22 21 20 19 18 16 15 10 9 5 4 0
4820 * +----+---+---+-----------+------+---+-------+-----+-------------+----+----+
4821 * | sf | 0 | S | 1 1 1 1 0 | type | 1 | rmode | opc | 0 0 0 0 0 0 | Rn | Rd |
4822 * +----+---+---+-----------+------+---+-------+-----+-------------+----+----+
4824 static void disas_fp_int_conv(DisasContext *s, uint32_t insn)
4826 int rd = extract32(insn, 0, 5);
4827 int rn = extract32(insn, 5, 5);
4828 int opcode = extract32(insn, 16, 3);
4829 int rmode = extract32(insn, 19, 2);
4830 int type = extract32(insn, 22, 2);
4831 bool sbit = extract32(insn, 29, 1);
4832 bool sf = extract32(insn, 31, 1);
4835 unallocated_encoding(s);
4841 bool itof = opcode & 1;
4844 unallocated_encoding(s);
4848 switch (sf << 3 | type << 1 | rmode) {
4849 case 0x0: /* 32 bit */
4850 case 0xa: /* 64 bit */
4851 case 0xd: /* 64 bit to top half of quad */
4854 /* all other sf/type/rmode combinations are invalid */
4855 unallocated_encoding(s);
4859 if (!fp_access_check(s)) {
4862 handle_fmov(s, rd, rn, type, itof);
4864 /* actual FP conversions */
4865 bool itof = extract32(opcode, 1, 1);
4867 if (type > 1 || (rmode != 0 && opcode > 1)) {
4868 unallocated_encoding(s);
4872 if (!fp_access_check(s)) {
4875 handle_fpfpcvt(s, rd, rn, opcode, itof, rmode, 64, sf, type);
4879 /* FP-specific subcases of table C3-6 (SIMD and FP data processing)
4880 * 31 30 29 28 25 24 0
4881 * +---+---+---+---------+-----------------------------+
4882 * | | 0 | | 1 1 1 1 | |
4883 * +---+---+---+---------+-----------------------------+
4885 static void disas_data_proc_fp(DisasContext *s, uint32_t insn)
4887 if (extract32(insn, 24, 1)) {
4888 /* Floating point data-processing (3 source) */
4889 disas_fp_3src(s, insn);
4890 } else if (extract32(insn, 21, 1) == 0) {
4891 /* Floating point to fixed point conversions */
4892 disas_fp_fixed_conv(s, insn);
4894 switch (extract32(insn, 10, 2)) {
4896 /* Floating point conditional compare */
4897 disas_fp_ccomp(s, insn);
4900 /* Floating point data-processing (2 source) */
4901 disas_fp_2src(s, insn);
4904 /* Floating point conditional select */
4905 disas_fp_csel(s, insn);
4908 switch (ctz32(extract32(insn, 12, 4))) {
4909 case 0: /* [15:12] == xxx1 */
4910 /* Floating point immediate */
4911 disas_fp_imm(s, insn);
4913 case 1: /* [15:12] == xx10 */
4914 /* Floating point compare */
4915 disas_fp_compare(s, insn);
4917 case 2: /* [15:12] == x100 */
4918 /* Floating point data-processing (1 source) */
4919 disas_fp_1src(s, insn);
4921 case 3: /* [15:12] == 1000 */
4922 unallocated_encoding(s);
4924 default: /* [15:12] == 0000 */
4925 /* Floating point <-> integer conversions */
4926 disas_fp_int_conv(s, insn);
4934 static void do_ext64(DisasContext *s, TCGv_i64 tcg_left, TCGv_i64 tcg_right,
4937 /* Extract 64 bits from the middle of two concatenated 64 bit
4938 * vector register slices left:right. The extracted bits start
4939 * at 'pos' bits into the right (least significant) side.
4940 * We return the result in tcg_right, and guarantee not to
4943 TCGv_i64 tcg_tmp = tcg_temp_new_i64();
4944 assert(pos > 0 && pos < 64);
4946 tcg_gen_shri_i64(tcg_right, tcg_right, pos);
4947 tcg_gen_shli_i64(tcg_tmp, tcg_left, 64 - pos);
4948 tcg_gen_or_i64(tcg_right, tcg_right, tcg_tmp);
4950 tcg_temp_free_i64(tcg_tmp);
4954 * 31 30 29 24 23 22 21 20 16 15 14 11 10 9 5 4 0
4955 * +---+---+-------------+-----+---+------+---+------+---+------+------+
4956 * | 0 | Q | 1 0 1 1 1 0 | op2 | 0 | Rm | 0 | imm4 | 0 | Rn | Rd |
4957 * +---+---+-------------+-----+---+------+---+------+---+------+------+
4959 static void disas_simd_ext(DisasContext *s, uint32_t insn)
4961 int is_q = extract32(insn, 30, 1);
4962 int op2 = extract32(insn, 22, 2);
4963 int imm4 = extract32(insn, 11, 4);
4964 int rm = extract32(insn, 16, 5);
4965 int rn = extract32(insn, 5, 5);
4966 int rd = extract32(insn, 0, 5);
4967 int pos = imm4 << 3;
4968 TCGv_i64 tcg_resl, tcg_resh;
4970 if (op2 != 0 || (!is_q && extract32(imm4, 3, 1))) {
4971 unallocated_encoding(s);
4975 if (!fp_access_check(s)) {
4979 tcg_resh = tcg_temp_new_i64();
4980 tcg_resl = tcg_temp_new_i64();
4982 /* Vd gets bits starting at pos bits into Vm:Vn. This is
4983 * either extracting 128 bits from a 128:128 concatenation, or
4984 * extracting 64 bits from a 64:64 concatenation.
4987 read_vec_element(s, tcg_resl, rn, 0, MO_64);
4989 read_vec_element(s, tcg_resh, rm, 0, MO_64);
4990 do_ext64(s, tcg_resh, tcg_resl, pos);
4992 tcg_gen_movi_i64(tcg_resh, 0);
4999 EltPosns eltposns[] = { {rn, 0}, {rn, 1}, {rm, 0}, {rm, 1} };
5000 EltPosns *elt = eltposns;
5007 read_vec_element(s, tcg_resl, elt->reg, elt->elt, MO_64);
5009 read_vec_element(s, tcg_resh, elt->reg, elt->elt, MO_64);
5012 do_ext64(s, tcg_resh, tcg_resl, pos);
5013 tcg_hh = tcg_temp_new_i64();
5014 read_vec_element(s, tcg_hh, elt->reg, elt->elt, MO_64);
5015 do_ext64(s, tcg_hh, tcg_resh, pos);
5016 tcg_temp_free_i64(tcg_hh);
5020 write_vec_element(s, tcg_resl, rd, 0, MO_64);
5021 tcg_temp_free_i64(tcg_resl);
5022 write_vec_element(s, tcg_resh, rd, 1, MO_64);
5023 tcg_temp_free_i64(tcg_resh);
5027 * 31 30 29 24 23 22 21 20 16 15 14 13 12 11 10 9 5 4 0
5028 * +---+---+-------------+-----+---+------+---+-----+----+-----+------+------+
5029 * | 0 | Q | 0 0 1 1 1 0 | op2 | 0 | Rm | 0 | len | op | 0 0 | Rn | Rd |
5030 * +---+---+-------------+-----+---+------+---+-----+----+-----+------+------+
5032 static void disas_simd_tb(DisasContext *s, uint32_t insn)
5034 int op2 = extract32(insn, 22, 2);
5035 int is_q = extract32(insn, 30, 1);
5036 int rm = extract32(insn, 16, 5);
5037 int rn = extract32(insn, 5, 5);
5038 int rd = extract32(insn, 0, 5);
5039 int is_tblx = extract32(insn, 12, 1);
5040 int len = extract32(insn, 13, 2);
5041 TCGv_i64 tcg_resl, tcg_resh, tcg_idx;
5042 TCGv_i32 tcg_regno, tcg_numregs;
5045 unallocated_encoding(s);
5049 if (!fp_access_check(s)) {
5053 /* This does a table lookup: for every byte element in the input
5054 * we index into a table formed from up to four vector registers,
5055 * and then the output is the result of the lookups. Our helper
5056 * function does the lookup operation for a single 64 bit part of
5059 tcg_resl = tcg_temp_new_i64();
5060 tcg_resh = tcg_temp_new_i64();
5063 read_vec_element(s, tcg_resl, rd, 0, MO_64);
5065 tcg_gen_movi_i64(tcg_resl, 0);
5067 if (is_tblx && is_q) {
5068 read_vec_element(s, tcg_resh, rd, 1, MO_64);
5070 tcg_gen_movi_i64(tcg_resh, 0);
5073 tcg_idx = tcg_temp_new_i64();
5074 tcg_regno = tcg_const_i32(rn);
5075 tcg_numregs = tcg_const_i32(len + 1);
5076 read_vec_element(s, tcg_idx, rm, 0, MO_64);
5077 gen_helper_simd_tbl(tcg_resl, cpu_env, tcg_resl, tcg_idx,
5078 tcg_regno, tcg_numregs);
5080 read_vec_element(s, tcg_idx, rm, 1, MO_64);
5081 gen_helper_simd_tbl(tcg_resh, cpu_env, tcg_resh, tcg_idx,
5082 tcg_regno, tcg_numregs);
5084 tcg_temp_free_i64(tcg_idx);
5085 tcg_temp_free_i32(tcg_regno);
5086 tcg_temp_free_i32(tcg_numregs);
5088 write_vec_element(s, tcg_resl, rd, 0, MO_64);
5089 tcg_temp_free_i64(tcg_resl);
5090 write_vec_element(s, tcg_resh, rd, 1, MO_64);
5091 tcg_temp_free_i64(tcg_resh);
5094 /* C3.6.3 ZIP/UZP/TRN
5095 * 31 30 29 24 23 22 21 20 16 15 14 12 11 10 9 5 4 0
5096 * +---+---+-------------+------+---+------+---+------------------+------+
5097 * | 0 | Q | 0 0 1 1 1 0 | size | 0 | Rm | 0 | opc | 1 0 | Rn | Rd |
5098 * +---+---+-------------+------+---+------+---+------------------+------+
5100 static void disas_simd_zip_trn(DisasContext *s, uint32_t insn)
5102 int rd = extract32(insn, 0, 5);
5103 int rn = extract32(insn, 5, 5);
5104 int rm = extract32(insn, 16, 5);
5105 int size = extract32(insn, 22, 2);
5106 /* opc field bits [1:0] indicate ZIP/UZP/TRN;
5107 * bit 2 indicates 1 vs 2 variant of the insn.
5109 int opcode = extract32(insn, 12, 2);
5110 bool part = extract32(insn, 14, 1);
5111 bool is_q = extract32(insn, 30, 1);
5112 int esize = 8 << size;
5114 int datasize = is_q ? 128 : 64;
5115 int elements = datasize / esize;
5116 TCGv_i64 tcg_res, tcg_resl, tcg_resh;
5118 if (opcode == 0 || (size == 3 && !is_q)) {
5119 unallocated_encoding(s);
5123 if (!fp_access_check(s)) {
5127 tcg_resl = tcg_const_i64(0);
5128 tcg_resh = tcg_const_i64(0);
5129 tcg_res = tcg_temp_new_i64();
5131 for (i = 0; i < elements; i++) {
5133 case 1: /* UZP1/2 */
5135 int midpoint = elements / 2;
5137 read_vec_element(s, tcg_res, rn, 2 * i + part, size);
5139 read_vec_element(s, tcg_res, rm,
5140 2 * (i - midpoint) + part, size);
5144 case 2: /* TRN1/2 */
5146 read_vec_element(s, tcg_res, rm, (i & ~1) + part, size);
5148 read_vec_element(s, tcg_res, rn, (i & ~1) + part, size);
5151 case 3: /* ZIP1/2 */
5153 int base = part * elements / 2;
5155 read_vec_element(s, tcg_res, rm, base + (i >> 1), size);
5157 read_vec_element(s, tcg_res, rn, base + (i >> 1), size);
5162 g_assert_not_reached();
5167 tcg_gen_shli_i64(tcg_res, tcg_res, ofs);
5168 tcg_gen_or_i64(tcg_resl, tcg_resl, tcg_res);
5170 tcg_gen_shli_i64(tcg_res, tcg_res, ofs - 64);
5171 tcg_gen_or_i64(tcg_resh, tcg_resh, tcg_res);
5175 tcg_temp_free_i64(tcg_res);
5177 write_vec_element(s, tcg_resl, rd, 0, MO_64);
5178 tcg_temp_free_i64(tcg_resl);
5179 write_vec_element(s, tcg_resh, rd, 1, MO_64);
5180 tcg_temp_free_i64(tcg_resh);
5183 static void do_minmaxop(DisasContext *s, TCGv_i32 tcg_elt1, TCGv_i32 tcg_elt2,
5184 int opc, bool is_min, TCGv_ptr fpst)
5186 /* Helper function for disas_simd_across_lanes: do a single precision
5187 * min/max operation on the specified two inputs,
5188 * and return the result in tcg_elt1.
5192 gen_helper_vfp_minnums(tcg_elt1, tcg_elt1, tcg_elt2, fpst);
5194 gen_helper_vfp_maxnums(tcg_elt1, tcg_elt1, tcg_elt2, fpst);
5199 gen_helper_vfp_mins(tcg_elt1, tcg_elt1, tcg_elt2, fpst);
5201 gen_helper_vfp_maxs(tcg_elt1, tcg_elt1, tcg_elt2, fpst);
5206 /* C3.6.4 AdvSIMD across lanes
5207 * 31 30 29 28 24 23 22 21 17 16 12 11 10 9 5 4 0
5208 * +---+---+---+-----------+------+-----------+--------+-----+------+------+
5209 * | 0 | Q | U | 0 1 1 1 0 | size | 1 1 0 0 0 | opcode | 1 0 | Rn | Rd |
5210 * +---+---+---+-----------+------+-----------+--------+-----+------+------+
5212 static void disas_simd_across_lanes(DisasContext *s, uint32_t insn)
5214 int rd = extract32(insn, 0, 5);
5215 int rn = extract32(insn, 5, 5);
5216 int size = extract32(insn, 22, 2);
5217 int opcode = extract32(insn, 12, 5);
5218 bool is_q = extract32(insn, 30, 1);
5219 bool is_u = extract32(insn, 29, 1);
5221 bool is_min = false;
5225 TCGv_i64 tcg_res, tcg_elt;
5228 case 0x1b: /* ADDV */
5230 unallocated_encoding(s);
5234 case 0x3: /* SADDLV, UADDLV */
5235 case 0xa: /* SMAXV, UMAXV */
5236 case 0x1a: /* SMINV, UMINV */
5237 if (size == 3 || (size == 2 && !is_q)) {
5238 unallocated_encoding(s);
5242 case 0xc: /* FMAXNMV, FMINNMV */
5243 case 0xf: /* FMAXV, FMINV */
5244 if (!is_u || !is_q || extract32(size, 0, 1)) {
5245 unallocated_encoding(s);
5248 /* Bit 1 of size field encodes min vs max, and actual size is always
5249 * 32 bits: adjust the size variable so following code can rely on it
5251 is_min = extract32(size, 1, 1);
5256 unallocated_encoding(s);
5260 if (!fp_access_check(s)) {
5265 elements = (is_q ? 128 : 64) / esize;
5267 tcg_res = tcg_temp_new_i64();
5268 tcg_elt = tcg_temp_new_i64();
5270 /* These instructions operate across all lanes of a vector
5271 * to produce a single result. We can guarantee that a 64
5272 * bit intermediate is sufficient:
5273 * + for [US]ADDLV the maximum element size is 32 bits, and
5274 * the result type is 64 bits
5275 * + for FMAX*V, FMIN*V, ADDV the intermediate type is the
5276 * same as the element size, which is 32 bits at most
5277 * For the integer operations we can choose to work at 64
5278 * or 32 bits and truncate at the end; for simplicity
5279 * we use 64 bits always. The floating point
5280 * ops do require 32 bit intermediates, though.
5283 read_vec_element(s, tcg_res, rn, 0, size | (is_u ? 0 : MO_SIGN));
5285 for (i = 1; i < elements; i++) {
5286 read_vec_element(s, tcg_elt, rn, i, size | (is_u ? 0 : MO_SIGN));
5289 case 0x03: /* SADDLV / UADDLV */
5290 case 0x1b: /* ADDV */
5291 tcg_gen_add_i64(tcg_res, tcg_res, tcg_elt);
5293 case 0x0a: /* SMAXV / UMAXV */
5294 tcg_gen_movcond_i64(is_u ? TCG_COND_GEU : TCG_COND_GE,
5296 tcg_res, tcg_elt, tcg_res, tcg_elt);
5298 case 0x1a: /* SMINV / UMINV */
5299 tcg_gen_movcond_i64(is_u ? TCG_COND_LEU : TCG_COND_LE,
5301 tcg_res, tcg_elt, tcg_res, tcg_elt);
5305 g_assert_not_reached();
5310 /* Floating point ops which work on 32 bit (single) intermediates.
5311 * Note that correct NaN propagation requires that we do these
5312 * operations in exactly the order specified by the pseudocode.
5314 TCGv_i32 tcg_elt1 = tcg_temp_new_i32();
5315 TCGv_i32 tcg_elt2 = tcg_temp_new_i32();
5316 TCGv_i32 tcg_elt3 = tcg_temp_new_i32();
5317 TCGv_ptr fpst = get_fpstatus_ptr();
5319 assert(esize == 32);
5320 assert(elements == 4);
5322 read_vec_element(s, tcg_elt, rn, 0, MO_32);
5323 tcg_gen_trunc_i64_i32(tcg_elt1, tcg_elt);
5324 read_vec_element(s, tcg_elt, rn, 1, MO_32);
5325 tcg_gen_trunc_i64_i32(tcg_elt2, tcg_elt);
5327 do_minmaxop(s, tcg_elt1, tcg_elt2, opcode, is_min, fpst);
5329 read_vec_element(s, tcg_elt, rn, 2, MO_32);
5330 tcg_gen_trunc_i64_i32(tcg_elt2, tcg_elt);
5331 read_vec_element(s, tcg_elt, rn, 3, MO_32);
5332 tcg_gen_trunc_i64_i32(tcg_elt3, tcg_elt);
5334 do_minmaxop(s, tcg_elt2, tcg_elt3, opcode, is_min, fpst);
5336 do_minmaxop(s, tcg_elt1, tcg_elt2, opcode, is_min, fpst);
5338 tcg_gen_extu_i32_i64(tcg_res, tcg_elt1);
5339 tcg_temp_free_i32(tcg_elt1);
5340 tcg_temp_free_i32(tcg_elt2);
5341 tcg_temp_free_i32(tcg_elt3);
5342 tcg_temp_free_ptr(fpst);
5345 tcg_temp_free_i64(tcg_elt);
5347 /* Now truncate the result to the width required for the final output */
5348 if (opcode == 0x03) {
5349 /* SADDLV, UADDLV: result is 2*esize */
5355 tcg_gen_ext8u_i64(tcg_res, tcg_res);
5358 tcg_gen_ext16u_i64(tcg_res, tcg_res);
5361 tcg_gen_ext32u_i64(tcg_res, tcg_res);
5366 g_assert_not_reached();
5369 write_fp_dreg(s, rd, tcg_res);
5370 tcg_temp_free_i64(tcg_res);
5373 /* C6.3.31 DUP (Element, Vector)
5375 * 31 30 29 21 20 16 15 10 9 5 4 0
5376 * +---+---+-------------------+--------+-------------+------+------+
5377 * | 0 | Q | 0 0 1 1 1 0 0 0 0 | imm5 | 0 0 0 0 0 1 | Rn | Rd |
5378 * +---+---+-------------------+--------+-------------+------+------+
5380 * size: encoded in imm5 (see ARM ARM LowestSetBit())
5382 static void handle_simd_dupe(DisasContext *s, int is_q, int rd, int rn,
5385 int size = ctz32(imm5);
5386 int esize = 8 << size;
5387 int elements = (is_q ? 128 : 64) / esize;
5391 if (size > 3 || (size == 3 && !is_q)) {
5392 unallocated_encoding(s);
5396 if (!fp_access_check(s)) {
5400 index = imm5 >> (size + 1);
5402 tmp = tcg_temp_new_i64();
5403 read_vec_element(s, tmp, rn, index, size);
5405 for (i = 0; i < elements; i++) {
5406 write_vec_element(s, tmp, rd, i, size);
5410 clear_vec_high(s, rd);
5413 tcg_temp_free_i64(tmp);
5416 /* C6.3.31 DUP (element, scalar)
5417 * 31 21 20 16 15 10 9 5 4 0
5418 * +-----------------------+--------+-------------+------+------+
5419 * | 0 1 0 1 1 1 1 0 0 0 0 | imm5 | 0 0 0 0 0 1 | Rn | Rd |
5420 * +-----------------------+--------+-------------+------+------+
5422 static void handle_simd_dupes(DisasContext *s, int rd, int rn,
5425 int size = ctz32(imm5);
5430 unallocated_encoding(s);
5434 if (!fp_access_check(s)) {
5438 index = imm5 >> (size + 1);
5440 /* This instruction just extracts the specified element and
5441 * zero-extends it into the bottom of the destination register.
5443 tmp = tcg_temp_new_i64();
5444 read_vec_element(s, tmp, rn, index, size);
5445 write_fp_dreg(s, rd, tmp);
5446 tcg_temp_free_i64(tmp);
5449 /* C6.3.32 DUP (General)
5451 * 31 30 29 21 20 16 15 10 9 5 4 0
5452 * +---+---+-------------------+--------+-------------+------+------+
5453 * | 0 | Q | 0 0 1 1 1 0 0 0 0 | imm5 | 0 0 0 0 1 1 | Rn | Rd |
5454 * +---+---+-------------------+--------+-------------+------+------+
5456 * size: encoded in imm5 (see ARM ARM LowestSetBit())
5458 static void handle_simd_dupg(DisasContext *s, int is_q, int rd, int rn,
5461 int size = ctz32(imm5);
5462 int esize = 8 << size;
5463 int elements = (is_q ? 128 : 64)/esize;
5466 if (size > 3 || ((size == 3) && !is_q)) {
5467 unallocated_encoding(s);
5471 if (!fp_access_check(s)) {
5475 for (i = 0; i < elements; i++) {
5476 write_vec_element(s, cpu_reg(s, rn), rd, i, size);
5479 clear_vec_high(s, rd);
5483 /* C6.3.150 INS (Element)
5485 * 31 21 20 16 15 14 11 10 9 5 4 0
5486 * +-----------------------+--------+------------+---+------+------+
5487 * | 0 1 1 0 1 1 1 0 0 0 0 | imm5 | 0 | imm4 | 1 | Rn | Rd |
5488 * +-----------------------+--------+------------+---+------+------+
5490 * size: encoded in imm5 (see ARM ARM LowestSetBit())
5491 * index: encoded in imm5<4:size+1>
5493 static void handle_simd_inse(DisasContext *s, int rd, int rn,
5496 int size = ctz32(imm5);
5497 int src_index, dst_index;
5501 unallocated_encoding(s);
5505 if (!fp_access_check(s)) {
5509 dst_index = extract32(imm5, 1+size, 5);
5510 src_index = extract32(imm4, size, 4);
5512 tmp = tcg_temp_new_i64();
5514 read_vec_element(s, tmp, rn, src_index, size);
5515 write_vec_element(s, tmp, rd, dst_index, size);
5517 tcg_temp_free_i64(tmp);
5521 /* C6.3.151 INS (General)
5523 * 31 21 20 16 15 10 9 5 4 0
5524 * +-----------------------+--------+-------------+------+------+
5525 * | 0 1 0 0 1 1 1 0 0 0 0 | imm5 | 0 0 0 1 1 1 | Rn | Rd |
5526 * +-----------------------+--------+-------------+------+------+
5528 * size: encoded in imm5 (see ARM ARM LowestSetBit())
5529 * index: encoded in imm5<4:size+1>
5531 static void handle_simd_insg(DisasContext *s, int rd, int rn, int imm5)
5533 int size = ctz32(imm5);
5537 unallocated_encoding(s);
5541 if (!fp_access_check(s)) {
5545 idx = extract32(imm5, 1 + size, 4 - size);
5546 write_vec_element(s, cpu_reg(s, rn), rd, idx, size);
5550 * C6.3.321 UMOV (General)
5551 * C6.3.237 SMOV (General)
5553 * 31 30 29 21 20 16 15 12 10 9 5 4 0
5554 * +---+---+-------------------+--------+-------------+------+------+
5555 * | 0 | Q | 0 0 1 1 1 0 0 0 0 | imm5 | 0 0 1 U 1 1 | Rn | Rd |
5556 * +---+---+-------------------+--------+-------------+------+------+
5558 * U: unsigned when set
5559 * size: encoded in imm5 (see ARM ARM LowestSetBit())
5561 static void handle_simd_umov_smov(DisasContext *s, int is_q, int is_signed,
5562 int rn, int rd, int imm5)
5564 int size = ctz32(imm5);
5568 /* Check for UnallocatedEncodings */
5570 if (size > 2 || (size == 2 && !is_q)) {
5571 unallocated_encoding(s);
5576 || (size < 3 && is_q)
5577 || (size == 3 && !is_q)) {
5578 unallocated_encoding(s);
5583 if (!fp_access_check(s)) {
5587 element = extract32(imm5, 1+size, 4);
5589 tcg_rd = cpu_reg(s, rd);
5590 read_vec_element(s, tcg_rd, rn, element, size | (is_signed ? MO_SIGN : 0));
5591 if (is_signed && !is_q) {
5592 tcg_gen_ext32u_i64(tcg_rd, tcg_rd);
5596 /* C3.6.5 AdvSIMD copy
5597 * 31 30 29 28 21 20 16 15 14 11 10 9 5 4 0
5598 * +---+---+----+-----------------+------+---+------+---+------+------+
5599 * | 0 | Q | op | 0 1 1 1 0 0 0 0 | imm5 | 0 | imm4 | 1 | Rn | Rd |
5600 * +---+---+----+-----------------+------+---+------+---+------+------+
5602 static void disas_simd_copy(DisasContext *s, uint32_t insn)
5604 int rd = extract32(insn, 0, 5);
5605 int rn = extract32(insn, 5, 5);
5606 int imm4 = extract32(insn, 11, 4);
5607 int op = extract32(insn, 29, 1);
5608 int is_q = extract32(insn, 30, 1);
5609 int imm5 = extract32(insn, 16, 5);
5614 handle_simd_inse(s, rd, rn, imm4, imm5);
5616 unallocated_encoding(s);
5621 /* DUP (element - vector) */
5622 handle_simd_dupe(s, is_q, rd, rn, imm5);
5626 handle_simd_dupg(s, is_q, rd, rn, imm5);
5631 handle_simd_insg(s, rd, rn, imm5);
5633 unallocated_encoding(s);
5638 /* UMOV/SMOV (is_q indicates 32/64; imm4 indicates signedness) */
5639 handle_simd_umov_smov(s, is_q, (imm4 == 5), rn, rd, imm5);
5642 unallocated_encoding(s);
5648 /* C3.6.6 AdvSIMD modified immediate
5649 * 31 30 29 28 19 18 16 15 12 11 10 9 5 4 0
5650 * +---+---+----+---------------------+-----+-------+----+---+-------+------+
5651 * | 0 | Q | op | 0 1 1 1 1 0 0 0 0 0 | abc | cmode | o2 | 1 | defgh | Rd |
5652 * +---+---+----+---------------------+-----+-------+----+---+-------+------+
5654 * There are a number of operations that can be carried out here:
5655 * MOVI - move (shifted) imm into register
5656 * MVNI - move inverted (shifted) imm into register
5657 * ORR - bitwise OR of (shifted) imm with register
5658 * BIC - bitwise clear of (shifted) imm with register
5660 static void disas_simd_mod_imm(DisasContext *s, uint32_t insn)
5662 int rd = extract32(insn, 0, 5);
5663 int cmode = extract32(insn, 12, 4);
5664 int cmode_3_1 = extract32(cmode, 1, 3);
5665 int cmode_0 = extract32(cmode, 0, 1);
5666 int o2 = extract32(insn, 11, 1);
5667 uint64_t abcdefgh = extract32(insn, 5, 5) | (extract32(insn, 16, 3) << 5);
5668 bool is_neg = extract32(insn, 29, 1);
5669 bool is_q = extract32(insn, 30, 1);
5671 TCGv_i64 tcg_rd, tcg_imm;
5674 if (o2 != 0 || ((cmode == 0xf) && is_neg && !is_q)) {
5675 unallocated_encoding(s);
5679 if (!fp_access_check(s)) {
5683 /* See AdvSIMDExpandImm() in ARM ARM */
5684 switch (cmode_3_1) {
5685 case 0: /* Replicate(Zeros(24):imm8, 2) */
5686 case 1: /* Replicate(Zeros(16):imm8:Zeros(8), 2) */
5687 case 2: /* Replicate(Zeros(8):imm8:Zeros(16), 2) */
5688 case 3: /* Replicate(imm8:Zeros(24), 2) */
5690 int shift = cmode_3_1 * 8;
5691 imm = bitfield_replicate(abcdefgh << shift, 32);
5694 case 4: /* Replicate(Zeros(8):imm8, 4) */
5695 case 5: /* Replicate(imm8:Zeros(8), 4) */
5697 int shift = (cmode_3_1 & 0x1) * 8;
5698 imm = bitfield_replicate(abcdefgh << shift, 16);
5703 /* Replicate(Zeros(8):imm8:Ones(16), 2) */
5704 imm = (abcdefgh << 16) | 0xffff;
5706 /* Replicate(Zeros(16):imm8:Ones(8), 2) */
5707 imm = (abcdefgh << 8) | 0xff;
5709 imm = bitfield_replicate(imm, 32);
5712 if (!cmode_0 && !is_neg) {
5713 imm = bitfield_replicate(abcdefgh, 8);
5714 } else if (!cmode_0 && is_neg) {
5717 for (i = 0; i < 8; i++) {
5718 if ((abcdefgh) & (1 << i)) {
5719 imm |= 0xffULL << (i * 8);
5722 } else if (cmode_0) {
5724 imm = (abcdefgh & 0x3f) << 48;
5725 if (abcdefgh & 0x80) {
5726 imm |= 0x8000000000000000ULL;
5728 if (abcdefgh & 0x40) {
5729 imm |= 0x3fc0000000000000ULL;
5731 imm |= 0x4000000000000000ULL;
5734 imm = (abcdefgh & 0x3f) << 19;
5735 if (abcdefgh & 0x80) {
5738 if (abcdefgh & 0x40) {
5749 if (cmode_3_1 != 7 && is_neg) {
5753 tcg_imm = tcg_const_i64(imm);
5754 tcg_rd = new_tmp_a64(s);
5756 for (i = 0; i < 2; i++) {
5757 int foffs = i ? fp_reg_hi_offset(s, rd) : fp_reg_offset(s, rd, MO_64);
5759 if (i == 1 && !is_q) {
5760 /* non-quad ops clear high half of vector */
5761 tcg_gen_movi_i64(tcg_rd, 0);
5762 } else if ((cmode & 0x9) == 0x1 || (cmode & 0xd) == 0x9) {
5763 tcg_gen_ld_i64(tcg_rd, cpu_env, foffs);
5766 tcg_gen_and_i64(tcg_rd, tcg_rd, tcg_imm);
5769 tcg_gen_or_i64(tcg_rd, tcg_rd, tcg_imm);
5773 tcg_gen_mov_i64(tcg_rd, tcg_imm);
5775 tcg_gen_st_i64(tcg_rd, cpu_env, foffs);
5778 tcg_temp_free_i64(tcg_imm);
5781 /* C3.6.7 AdvSIMD scalar copy
5782 * 31 30 29 28 21 20 16 15 14 11 10 9 5 4 0
5783 * +-----+----+-----------------+------+---+------+---+------+------+
5784 * | 0 1 | op | 1 1 1 1 0 0 0 0 | imm5 | 0 | imm4 | 1 | Rn | Rd |
5785 * +-----+----+-----------------+------+---+------+---+------+------+
5787 static void disas_simd_scalar_copy(DisasContext *s, uint32_t insn)
5789 int rd = extract32(insn, 0, 5);
5790 int rn = extract32(insn, 5, 5);
5791 int imm4 = extract32(insn, 11, 4);
5792 int imm5 = extract32(insn, 16, 5);
5793 int op = extract32(insn, 29, 1);
5795 if (op != 0 || imm4 != 0) {
5796 unallocated_encoding(s);
5800 /* DUP (element, scalar) */
5801 handle_simd_dupes(s, rd, rn, imm5);
5804 /* C3.6.8 AdvSIMD scalar pairwise
5805 * 31 30 29 28 24 23 22 21 17 16 12 11 10 9 5 4 0
5806 * +-----+---+-----------+------+-----------+--------+-----+------+------+
5807 * | 0 1 | U | 1 1 1 1 0 | size | 1 1 0 0 0 | opcode | 1 0 | Rn | Rd |
5808 * +-----+---+-----------+------+-----------+--------+-----+------+------+
5810 static void disas_simd_scalar_pairwise(DisasContext *s, uint32_t insn)
5812 int u = extract32(insn, 29, 1);
5813 int size = extract32(insn, 22, 2);
5814 int opcode = extract32(insn, 12, 5);
5815 int rn = extract32(insn, 5, 5);
5816 int rd = extract32(insn, 0, 5);
5819 /* For some ops (the FP ones), size[1] is part of the encoding.
5820 * For ADDP strictly it is not but size[1] is always 1 for valid
5823 opcode |= (extract32(size, 1, 1) << 5);
5826 case 0x3b: /* ADDP */
5827 if (u || size != 3) {
5828 unallocated_encoding(s);
5831 if (!fp_access_check(s)) {
5835 TCGV_UNUSED_PTR(fpst);
5837 case 0xc: /* FMAXNMP */
5838 case 0xd: /* FADDP */
5839 case 0xf: /* FMAXP */
5840 case 0x2c: /* FMINNMP */
5841 case 0x2f: /* FMINP */
5842 /* FP op, size[0] is 32 or 64 bit */
5844 unallocated_encoding(s);
5847 if (!fp_access_check(s)) {
5851 size = extract32(size, 0, 1) ? 3 : 2;
5852 fpst = get_fpstatus_ptr();
5855 unallocated_encoding(s);
5860 TCGv_i64 tcg_op1 = tcg_temp_new_i64();
5861 TCGv_i64 tcg_op2 = tcg_temp_new_i64();
5862 TCGv_i64 tcg_res = tcg_temp_new_i64();
5864 read_vec_element(s, tcg_op1, rn, 0, MO_64);
5865 read_vec_element(s, tcg_op2, rn, 1, MO_64);
5868 case 0x3b: /* ADDP */
5869 tcg_gen_add_i64(tcg_res, tcg_op1, tcg_op2);
5871 case 0xc: /* FMAXNMP */
5872 gen_helper_vfp_maxnumd(tcg_res, tcg_op1, tcg_op2, fpst);
5874 case 0xd: /* FADDP */
5875 gen_helper_vfp_addd(tcg_res, tcg_op1, tcg_op2, fpst);
5877 case 0xf: /* FMAXP */
5878 gen_helper_vfp_maxd(tcg_res, tcg_op1, tcg_op2, fpst);
5880 case 0x2c: /* FMINNMP */
5881 gen_helper_vfp_minnumd(tcg_res, tcg_op1, tcg_op2, fpst);
5883 case 0x2f: /* FMINP */
5884 gen_helper_vfp_mind(tcg_res, tcg_op1, tcg_op2, fpst);
5887 g_assert_not_reached();
5890 write_fp_dreg(s, rd, tcg_res);
5892 tcg_temp_free_i64(tcg_op1);
5893 tcg_temp_free_i64(tcg_op2);
5894 tcg_temp_free_i64(tcg_res);
5896 TCGv_i32 tcg_op1 = tcg_temp_new_i32();
5897 TCGv_i32 tcg_op2 = tcg_temp_new_i32();
5898 TCGv_i32 tcg_res = tcg_temp_new_i32();
5900 read_vec_element_i32(s, tcg_op1, rn, 0, MO_32);
5901 read_vec_element_i32(s, tcg_op2, rn, 1, MO_32);
5904 case 0xc: /* FMAXNMP */
5905 gen_helper_vfp_maxnums(tcg_res, tcg_op1, tcg_op2, fpst);
5907 case 0xd: /* FADDP */
5908 gen_helper_vfp_adds(tcg_res, tcg_op1, tcg_op2, fpst);
5910 case 0xf: /* FMAXP */
5911 gen_helper_vfp_maxs(tcg_res, tcg_op1, tcg_op2, fpst);
5913 case 0x2c: /* FMINNMP */
5914 gen_helper_vfp_minnums(tcg_res, tcg_op1, tcg_op2, fpst);
5916 case 0x2f: /* FMINP */
5917 gen_helper_vfp_mins(tcg_res, tcg_op1, tcg_op2, fpst);
5920 g_assert_not_reached();
5923 write_fp_sreg(s, rd, tcg_res);
5925 tcg_temp_free_i32(tcg_op1);
5926 tcg_temp_free_i32(tcg_op2);
5927 tcg_temp_free_i32(tcg_res);
5930 if (!TCGV_IS_UNUSED_PTR(fpst)) {
5931 tcg_temp_free_ptr(fpst);
5936 * Common SSHR[RA]/USHR[RA] - Shift right (optional rounding/accumulate)
5938 * This code is handles the common shifting code and is used by both
5939 * the vector and scalar code.
5941 static void handle_shri_with_rndacc(TCGv_i64 tcg_res, TCGv_i64 tcg_src,
5942 TCGv_i64 tcg_rnd, bool accumulate,
5943 bool is_u, int size, int shift)
5945 bool extended_result = false;
5946 bool round = !TCGV_IS_UNUSED_I64(tcg_rnd);
5948 TCGv_i64 tcg_src_hi;
5950 if (round && size == 3) {
5951 extended_result = true;
5952 ext_lshift = 64 - shift;
5953 tcg_src_hi = tcg_temp_new_i64();
5954 } else if (shift == 64) {
5955 if (!accumulate && is_u) {
5956 /* result is zero */
5957 tcg_gen_movi_i64(tcg_res, 0);
5962 /* Deal with the rounding step */
5964 if (extended_result) {
5965 TCGv_i64 tcg_zero = tcg_const_i64(0);
5967 /* take care of sign extending tcg_res */
5968 tcg_gen_sari_i64(tcg_src_hi, tcg_src, 63);
5969 tcg_gen_add2_i64(tcg_src, tcg_src_hi,
5970 tcg_src, tcg_src_hi,
5973 tcg_gen_add2_i64(tcg_src, tcg_src_hi,
5977 tcg_temp_free_i64(tcg_zero);
5979 tcg_gen_add_i64(tcg_src, tcg_src, tcg_rnd);
5983 /* Now do the shift right */
5984 if (round && extended_result) {
5985 /* extended case, >64 bit precision required */
5986 if (ext_lshift == 0) {
5987 /* special case, only high bits matter */
5988 tcg_gen_mov_i64(tcg_src, tcg_src_hi);
5990 tcg_gen_shri_i64(tcg_src, tcg_src, shift);
5991 tcg_gen_shli_i64(tcg_src_hi, tcg_src_hi, ext_lshift);
5992 tcg_gen_or_i64(tcg_src, tcg_src, tcg_src_hi);
5997 /* essentially shifting in 64 zeros */
5998 tcg_gen_movi_i64(tcg_src, 0);
6000 tcg_gen_shri_i64(tcg_src, tcg_src, shift);
6004 /* effectively extending the sign-bit */
6005 tcg_gen_sari_i64(tcg_src, tcg_src, 63);
6007 tcg_gen_sari_i64(tcg_src, tcg_src, shift);
6013 tcg_gen_add_i64(tcg_res, tcg_res, tcg_src);
6015 tcg_gen_mov_i64(tcg_res, tcg_src);
6018 if (extended_result) {
6019 tcg_temp_free_i64(tcg_src_hi);
6023 /* Common SHL/SLI - Shift left with an optional insert */
6024 static void handle_shli_with_ins(TCGv_i64 tcg_res, TCGv_i64 tcg_src,
6025 bool insert, int shift)
6027 if (insert) { /* SLI */
6028 tcg_gen_deposit_i64(tcg_res, tcg_res, tcg_src, shift, 64 - shift);
6030 tcg_gen_shli_i64(tcg_res, tcg_src, shift);
6034 /* SRI: shift right with insert */
6035 static void handle_shri_with_ins(TCGv_i64 tcg_res, TCGv_i64 tcg_src,
6036 int size, int shift)
6038 int esize = 8 << size;
6040 /* shift count same as element size is valid but does nothing;
6041 * special case to avoid potential shift by 64.
6043 if (shift != esize) {
6044 tcg_gen_shri_i64(tcg_src, tcg_src, shift);
6045 tcg_gen_deposit_i64(tcg_res, tcg_res, tcg_src, 0, esize - shift);
6049 /* SSHR[RA]/USHR[RA] - Scalar shift right (optional rounding/accumulate) */
6050 static void handle_scalar_simd_shri(DisasContext *s,
6051 bool is_u, int immh, int immb,
6052 int opcode, int rn, int rd)
6055 int immhb = immh << 3 | immb;
6056 int shift = 2 * (8 << size) - immhb;
6057 bool accumulate = false;
6059 bool insert = false;
6064 if (!extract32(immh, 3, 1)) {
6065 unallocated_encoding(s);
6069 if (!fp_access_check(s)) {
6074 case 0x02: /* SSRA / USRA (accumulate) */
6077 case 0x04: /* SRSHR / URSHR (rounding) */
6080 case 0x06: /* SRSRA / URSRA (accum + rounding) */
6081 accumulate = round = true;
6083 case 0x08: /* SRI */
6089 uint64_t round_const = 1ULL << (shift - 1);
6090 tcg_round = tcg_const_i64(round_const);
6092 TCGV_UNUSED_I64(tcg_round);
6095 tcg_rn = read_fp_dreg(s, rn);
6096 tcg_rd = (accumulate || insert) ? read_fp_dreg(s, rd) : tcg_temp_new_i64();
6099 handle_shri_with_ins(tcg_rd, tcg_rn, size, shift);
6101 handle_shri_with_rndacc(tcg_rd, tcg_rn, tcg_round,
6102 accumulate, is_u, size, shift);
6105 write_fp_dreg(s, rd, tcg_rd);
6107 tcg_temp_free_i64(tcg_rn);
6108 tcg_temp_free_i64(tcg_rd);
6110 tcg_temp_free_i64(tcg_round);
6114 /* SHL/SLI - Scalar shift left */
6115 static void handle_scalar_simd_shli(DisasContext *s, bool insert,
6116 int immh, int immb, int opcode,
6119 int size = 32 - clz32(immh) - 1;
6120 int immhb = immh << 3 | immb;
6121 int shift = immhb - (8 << size);
6122 TCGv_i64 tcg_rn = new_tmp_a64(s);
6123 TCGv_i64 tcg_rd = new_tmp_a64(s);
6125 if (!extract32(immh, 3, 1)) {
6126 unallocated_encoding(s);
6130 if (!fp_access_check(s)) {
6134 tcg_rn = read_fp_dreg(s, rn);
6135 tcg_rd = insert ? read_fp_dreg(s, rd) : tcg_temp_new_i64();
6137 handle_shli_with_ins(tcg_rd, tcg_rn, insert, shift);
6139 write_fp_dreg(s, rd, tcg_rd);
6141 tcg_temp_free_i64(tcg_rn);
6142 tcg_temp_free_i64(tcg_rd);
6145 /* SQSHRN/SQSHRUN - Saturating (signed/unsigned) shift right with
6146 * (signed/unsigned) narrowing */
6147 static void handle_vec_simd_sqshrn(DisasContext *s, bool is_scalar, bool is_q,
6148 bool is_u_shift, bool is_u_narrow,
6149 int immh, int immb, int opcode,
6152 int immhb = immh << 3 | immb;
6153 int size = 32 - clz32(immh) - 1;
6154 int esize = 8 << size;
6155 int shift = (2 * esize) - immhb;
6156 int elements = is_scalar ? 1 : (64 / esize);
6157 bool round = extract32(opcode, 0, 1);
6158 TCGMemOp ldop = (size + 1) | (is_u_shift ? 0 : MO_SIGN);
6159 TCGv_i64 tcg_rn, tcg_rd, tcg_round;
6160 TCGv_i32 tcg_rd_narrowed;
6163 static NeonGenNarrowEnvFn * const signed_narrow_fns[4][2] = {
6164 { gen_helper_neon_narrow_sat_s8,
6165 gen_helper_neon_unarrow_sat8 },
6166 { gen_helper_neon_narrow_sat_s16,
6167 gen_helper_neon_unarrow_sat16 },
6168 { gen_helper_neon_narrow_sat_s32,
6169 gen_helper_neon_unarrow_sat32 },
6172 static NeonGenNarrowEnvFn * const unsigned_narrow_fns[4] = {
6173 gen_helper_neon_narrow_sat_u8,
6174 gen_helper_neon_narrow_sat_u16,
6175 gen_helper_neon_narrow_sat_u32,
6178 NeonGenNarrowEnvFn *narrowfn;
6184 if (extract32(immh, 3, 1)) {
6185 unallocated_encoding(s);
6189 if (!fp_access_check(s)) {
6194 narrowfn = unsigned_narrow_fns[size];
6196 narrowfn = signed_narrow_fns[size][is_u_narrow ? 1 : 0];
6199 tcg_rn = tcg_temp_new_i64();
6200 tcg_rd = tcg_temp_new_i64();
6201 tcg_rd_narrowed = tcg_temp_new_i32();
6202 tcg_final = tcg_const_i64(0);
6205 uint64_t round_const = 1ULL << (shift - 1);
6206 tcg_round = tcg_const_i64(round_const);
6208 TCGV_UNUSED_I64(tcg_round);
6211 for (i = 0; i < elements; i++) {
6212 read_vec_element(s, tcg_rn, rn, i, ldop);
6213 handle_shri_with_rndacc(tcg_rd, tcg_rn, tcg_round,
6214 false, is_u_shift, size+1, shift);
6215 narrowfn(tcg_rd_narrowed, cpu_env, tcg_rd);
6216 tcg_gen_extu_i32_i64(tcg_rd, tcg_rd_narrowed);
6217 tcg_gen_deposit_i64(tcg_final, tcg_final, tcg_rd, esize * i, esize);
6221 clear_vec_high(s, rd);
6222 write_vec_element(s, tcg_final, rd, 0, MO_64);
6224 write_vec_element(s, tcg_final, rd, 1, MO_64);
6228 tcg_temp_free_i64(tcg_round);
6230 tcg_temp_free_i64(tcg_rn);
6231 tcg_temp_free_i64(tcg_rd);
6232 tcg_temp_free_i32(tcg_rd_narrowed);
6233 tcg_temp_free_i64(tcg_final);
6237 /* SQSHLU, UQSHL, SQSHL: saturating left shifts */
6238 static void handle_simd_qshl(DisasContext *s, bool scalar, bool is_q,
6239 bool src_unsigned, bool dst_unsigned,
6240 int immh, int immb, int rn, int rd)
6242 int immhb = immh << 3 | immb;
6243 int size = 32 - clz32(immh) - 1;
6244 int shift = immhb - (8 << size);
6248 assert(!(scalar && is_q));
6251 if (!is_q && extract32(immh, 3, 1)) {
6252 unallocated_encoding(s);
6256 /* Since we use the variable-shift helpers we must
6257 * replicate the shift count into each element of
6258 * the tcg_shift value.
6262 shift |= shift << 8;
6265 shift |= shift << 16;
6271 g_assert_not_reached();
6275 if (!fp_access_check(s)) {
6280 TCGv_i64 tcg_shift = tcg_const_i64(shift);
6281 static NeonGenTwo64OpEnvFn * const fns[2][2] = {
6282 { gen_helper_neon_qshl_s64, gen_helper_neon_qshlu_s64 },
6283 { NULL, gen_helper_neon_qshl_u64 },
6285 NeonGenTwo64OpEnvFn *genfn = fns[src_unsigned][dst_unsigned];
6286 int maxpass = is_q ? 2 : 1;
6288 for (pass = 0; pass < maxpass; pass++) {
6289 TCGv_i64 tcg_op = tcg_temp_new_i64();
6291 read_vec_element(s, tcg_op, rn, pass, MO_64);
6292 genfn(tcg_op, cpu_env, tcg_op, tcg_shift);
6293 write_vec_element(s, tcg_op, rd, pass, MO_64);
6295 tcg_temp_free_i64(tcg_op);
6297 tcg_temp_free_i64(tcg_shift);
6300 clear_vec_high(s, rd);
6303 TCGv_i32 tcg_shift = tcg_const_i32(shift);
6304 static NeonGenTwoOpEnvFn * const fns[2][2][3] = {
6306 { gen_helper_neon_qshl_s8,
6307 gen_helper_neon_qshl_s16,
6308 gen_helper_neon_qshl_s32 },
6309 { gen_helper_neon_qshlu_s8,
6310 gen_helper_neon_qshlu_s16,
6311 gen_helper_neon_qshlu_s32 }
6313 { NULL, NULL, NULL },
6314 { gen_helper_neon_qshl_u8,
6315 gen_helper_neon_qshl_u16,
6316 gen_helper_neon_qshl_u32 }
6319 NeonGenTwoOpEnvFn *genfn = fns[src_unsigned][dst_unsigned][size];
6320 TCGMemOp memop = scalar ? size : MO_32;
6321 int maxpass = scalar ? 1 : is_q ? 4 : 2;
6323 for (pass = 0; pass < maxpass; pass++) {
6324 TCGv_i32 tcg_op = tcg_temp_new_i32();
6326 read_vec_element_i32(s, tcg_op, rn, pass, memop);
6327 genfn(tcg_op, cpu_env, tcg_op, tcg_shift);
6331 tcg_gen_ext8u_i32(tcg_op, tcg_op);
6334 tcg_gen_ext16u_i32(tcg_op, tcg_op);
6339 g_assert_not_reached();
6341 write_fp_sreg(s, rd, tcg_op);
6343 write_vec_element_i32(s, tcg_op, rd, pass, MO_32);
6346 tcg_temp_free_i32(tcg_op);
6348 tcg_temp_free_i32(tcg_shift);
6350 if (!is_q && !scalar) {
6351 clear_vec_high(s, rd);
6356 /* Common vector code for handling integer to FP conversion */
6357 static void handle_simd_intfp_conv(DisasContext *s, int rd, int rn,
6358 int elements, int is_signed,
6359 int fracbits, int size)
6361 bool is_double = size == 3 ? true : false;
6362 TCGv_ptr tcg_fpst = get_fpstatus_ptr();
6363 TCGv_i32 tcg_shift = tcg_const_i32(fracbits);
6364 TCGv_i64 tcg_int = tcg_temp_new_i64();
6365 TCGMemOp mop = size | (is_signed ? MO_SIGN : 0);
6368 for (pass = 0; pass < elements; pass++) {
6369 read_vec_element(s, tcg_int, rn, pass, mop);
6372 TCGv_i64 tcg_double = tcg_temp_new_i64();
6374 gen_helper_vfp_sqtod(tcg_double, tcg_int,
6375 tcg_shift, tcg_fpst);
6377 gen_helper_vfp_uqtod(tcg_double, tcg_int,
6378 tcg_shift, tcg_fpst);
6380 if (elements == 1) {
6381 write_fp_dreg(s, rd, tcg_double);
6383 write_vec_element(s, tcg_double, rd, pass, MO_64);
6385 tcg_temp_free_i64(tcg_double);
6387 TCGv_i32 tcg_single = tcg_temp_new_i32();
6389 gen_helper_vfp_sqtos(tcg_single, tcg_int,
6390 tcg_shift, tcg_fpst);
6392 gen_helper_vfp_uqtos(tcg_single, tcg_int,
6393 tcg_shift, tcg_fpst);
6395 if (elements == 1) {
6396 write_fp_sreg(s, rd, tcg_single);
6398 write_vec_element_i32(s, tcg_single, rd, pass, MO_32);
6400 tcg_temp_free_i32(tcg_single);
6404 if (!is_double && elements == 2) {
6405 clear_vec_high(s, rd);
6408 tcg_temp_free_i64(tcg_int);
6409 tcg_temp_free_ptr(tcg_fpst);
6410 tcg_temp_free_i32(tcg_shift);
6413 /* UCVTF/SCVTF - Integer to FP conversion */
6414 static void handle_simd_shift_intfp_conv(DisasContext *s, bool is_scalar,
6415 bool is_q, bool is_u,
6416 int immh, int immb, int opcode,
6419 bool is_double = extract32(immh, 3, 1);
6420 int size = is_double ? MO_64 : MO_32;
6422 int immhb = immh << 3 | immb;
6423 int fracbits = (is_double ? 128 : 64) - immhb;
6425 if (!extract32(immh, 2, 2)) {
6426 unallocated_encoding(s);
6433 elements = is_double ? 2 : is_q ? 4 : 2;
6434 if (is_double && !is_q) {
6435 unallocated_encoding(s);
6440 if (!fp_access_check(s)) {
6444 /* immh == 0 would be a failure of the decode logic */
6447 handle_simd_intfp_conv(s, rd, rn, elements, !is_u, fracbits, size);
6450 /* FCVTZS, FVCVTZU - FP to fixedpoint conversion */
6451 static void handle_simd_shift_fpint_conv(DisasContext *s, bool is_scalar,
6452 bool is_q, bool is_u,
6453 int immh, int immb, int rn, int rd)
6455 bool is_double = extract32(immh, 3, 1);
6456 int immhb = immh << 3 | immb;
6457 int fracbits = (is_double ? 128 : 64) - immhb;
6459 TCGv_ptr tcg_fpstatus;
6460 TCGv_i32 tcg_rmode, tcg_shift;
6462 if (!extract32(immh, 2, 2)) {
6463 unallocated_encoding(s);
6467 if (!is_scalar && !is_q && is_double) {
6468 unallocated_encoding(s);
6472 if (!fp_access_check(s)) {
6476 assert(!(is_scalar && is_q));
6478 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(FPROUNDING_ZERO));
6479 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
6480 tcg_fpstatus = get_fpstatus_ptr();
6481 tcg_shift = tcg_const_i32(fracbits);
6484 int maxpass = is_scalar ? 1 : is_q ? 2 : 1;
6486 for (pass = 0; pass < maxpass; pass++) {
6487 TCGv_i64 tcg_op = tcg_temp_new_i64();
6489 read_vec_element(s, tcg_op, rn, pass, MO_64);
6491 gen_helper_vfp_touqd(tcg_op, tcg_op, tcg_shift, tcg_fpstatus);
6493 gen_helper_vfp_tosqd(tcg_op, tcg_op, tcg_shift, tcg_fpstatus);
6495 write_vec_element(s, tcg_op, rd, pass, MO_64);
6496 tcg_temp_free_i64(tcg_op);
6499 clear_vec_high(s, rd);
6502 int maxpass = is_scalar ? 1 : is_q ? 4 : 2;
6503 for (pass = 0; pass < maxpass; pass++) {
6504 TCGv_i32 tcg_op = tcg_temp_new_i32();
6506 read_vec_element_i32(s, tcg_op, rn, pass, MO_32);
6508 gen_helper_vfp_touls(tcg_op, tcg_op, tcg_shift, tcg_fpstatus);
6510 gen_helper_vfp_tosls(tcg_op, tcg_op, tcg_shift, tcg_fpstatus);
6513 write_fp_sreg(s, rd, tcg_op);
6515 write_vec_element_i32(s, tcg_op, rd, pass, MO_32);
6517 tcg_temp_free_i32(tcg_op);
6519 if (!is_q && !is_scalar) {
6520 clear_vec_high(s, rd);
6524 tcg_temp_free_ptr(tcg_fpstatus);
6525 tcg_temp_free_i32(tcg_shift);
6526 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
6527 tcg_temp_free_i32(tcg_rmode);
6530 /* C3.6.9 AdvSIMD scalar shift by immediate
6531 * 31 30 29 28 23 22 19 18 16 15 11 10 9 5 4 0
6532 * +-----+---+-------------+------+------+--------+---+------+------+
6533 * | 0 1 | U | 1 1 1 1 1 0 | immh | immb | opcode | 1 | Rn | Rd |
6534 * +-----+---+-------------+------+------+--------+---+------+------+
6536 * This is the scalar version so it works on a fixed sized registers
6538 static void disas_simd_scalar_shift_imm(DisasContext *s, uint32_t insn)
6540 int rd = extract32(insn, 0, 5);
6541 int rn = extract32(insn, 5, 5);
6542 int opcode = extract32(insn, 11, 5);
6543 int immb = extract32(insn, 16, 3);
6544 int immh = extract32(insn, 19, 4);
6545 bool is_u = extract32(insn, 29, 1);
6548 unallocated_encoding(s);
6553 case 0x08: /* SRI */
6555 unallocated_encoding(s);
6559 case 0x00: /* SSHR / USHR */
6560 case 0x02: /* SSRA / USRA */
6561 case 0x04: /* SRSHR / URSHR */
6562 case 0x06: /* SRSRA / URSRA */
6563 handle_scalar_simd_shri(s, is_u, immh, immb, opcode, rn, rd);
6565 case 0x0a: /* SHL / SLI */
6566 handle_scalar_simd_shli(s, is_u, immh, immb, opcode, rn, rd);
6568 case 0x1c: /* SCVTF, UCVTF */
6569 handle_simd_shift_intfp_conv(s, true, false, is_u, immh, immb,
6572 case 0x10: /* SQSHRUN, SQSHRUN2 */
6573 case 0x11: /* SQRSHRUN, SQRSHRUN2 */
6575 unallocated_encoding(s);
6578 handle_vec_simd_sqshrn(s, true, false, false, true,
6579 immh, immb, opcode, rn, rd);
6581 case 0x12: /* SQSHRN, SQSHRN2, UQSHRN */
6582 case 0x13: /* SQRSHRN, SQRSHRN2, UQRSHRN, UQRSHRN2 */
6583 handle_vec_simd_sqshrn(s, true, false, is_u, is_u,
6584 immh, immb, opcode, rn, rd);
6586 case 0xc: /* SQSHLU */
6588 unallocated_encoding(s);
6591 handle_simd_qshl(s, true, false, false, true, immh, immb, rn, rd);
6593 case 0xe: /* SQSHL, UQSHL */
6594 handle_simd_qshl(s, true, false, is_u, is_u, immh, immb, rn, rd);
6596 case 0x1f: /* FCVTZS, FCVTZU */
6597 handle_simd_shift_fpint_conv(s, true, false, is_u, immh, immb, rn, rd);
6600 unallocated_encoding(s);
6605 /* C3.6.10 AdvSIMD scalar three different
6606 * 31 30 29 28 24 23 22 21 20 16 15 12 11 10 9 5 4 0
6607 * +-----+---+-----------+------+---+------+--------+-----+------+------+
6608 * | 0 1 | U | 1 1 1 1 0 | size | 1 | Rm | opcode | 0 0 | Rn | Rd |
6609 * +-----+---+-----------+------+---+------+--------+-----+------+------+
6611 static void disas_simd_scalar_three_reg_diff(DisasContext *s, uint32_t insn)
6613 bool is_u = extract32(insn, 29, 1);
6614 int size = extract32(insn, 22, 2);
6615 int opcode = extract32(insn, 12, 4);
6616 int rm = extract32(insn, 16, 5);
6617 int rn = extract32(insn, 5, 5);
6618 int rd = extract32(insn, 0, 5);
6621 unallocated_encoding(s);
6626 case 0x9: /* SQDMLAL, SQDMLAL2 */
6627 case 0xb: /* SQDMLSL, SQDMLSL2 */
6628 case 0xd: /* SQDMULL, SQDMULL2 */
6629 if (size == 0 || size == 3) {
6630 unallocated_encoding(s);
6635 unallocated_encoding(s);
6639 if (!fp_access_check(s)) {
6644 TCGv_i64 tcg_op1 = tcg_temp_new_i64();
6645 TCGv_i64 tcg_op2 = tcg_temp_new_i64();
6646 TCGv_i64 tcg_res = tcg_temp_new_i64();
6648 read_vec_element(s, tcg_op1, rn, 0, MO_32 | MO_SIGN);
6649 read_vec_element(s, tcg_op2, rm, 0, MO_32 | MO_SIGN);
6651 tcg_gen_mul_i64(tcg_res, tcg_op1, tcg_op2);
6652 gen_helper_neon_addl_saturate_s64(tcg_res, cpu_env, tcg_res, tcg_res);
6655 case 0xd: /* SQDMULL, SQDMULL2 */
6657 case 0xb: /* SQDMLSL, SQDMLSL2 */
6658 tcg_gen_neg_i64(tcg_res, tcg_res);
6660 case 0x9: /* SQDMLAL, SQDMLAL2 */
6661 read_vec_element(s, tcg_op1, rd, 0, MO_64);
6662 gen_helper_neon_addl_saturate_s64(tcg_res, cpu_env,
6666 g_assert_not_reached();
6669 write_fp_dreg(s, rd, tcg_res);
6671 tcg_temp_free_i64(tcg_op1);
6672 tcg_temp_free_i64(tcg_op2);
6673 tcg_temp_free_i64(tcg_res);
6675 TCGv_i32 tcg_op1 = tcg_temp_new_i32();
6676 TCGv_i32 tcg_op2 = tcg_temp_new_i32();
6677 TCGv_i64 tcg_res = tcg_temp_new_i64();
6679 read_vec_element_i32(s, tcg_op1, rn, 0, MO_16);
6680 read_vec_element_i32(s, tcg_op2, rm, 0, MO_16);
6682 gen_helper_neon_mull_s16(tcg_res, tcg_op1, tcg_op2);
6683 gen_helper_neon_addl_saturate_s32(tcg_res, cpu_env, tcg_res, tcg_res);
6686 case 0xd: /* SQDMULL, SQDMULL2 */
6688 case 0xb: /* SQDMLSL, SQDMLSL2 */
6689 gen_helper_neon_negl_u32(tcg_res, tcg_res);
6691 case 0x9: /* SQDMLAL, SQDMLAL2 */
6693 TCGv_i64 tcg_op3 = tcg_temp_new_i64();
6694 read_vec_element(s, tcg_op3, rd, 0, MO_32);
6695 gen_helper_neon_addl_saturate_s32(tcg_res, cpu_env,
6697 tcg_temp_free_i64(tcg_op3);
6701 g_assert_not_reached();
6704 tcg_gen_ext32u_i64(tcg_res, tcg_res);
6705 write_fp_dreg(s, rd, tcg_res);
6707 tcg_temp_free_i32(tcg_op1);
6708 tcg_temp_free_i32(tcg_op2);
6709 tcg_temp_free_i64(tcg_res);
6713 static void handle_3same_64(DisasContext *s, int opcode, bool u,
6714 TCGv_i64 tcg_rd, TCGv_i64 tcg_rn, TCGv_i64 tcg_rm)
6716 /* Handle 64x64->64 opcodes which are shared between the scalar
6717 * and vector 3-same groups. We cover every opcode where size == 3
6718 * is valid in either the three-reg-same (integer, not pairwise)
6719 * or scalar-three-reg-same groups. (Some opcodes are not yet
6725 case 0x1: /* SQADD */
6727 gen_helper_neon_qadd_u64(tcg_rd, cpu_env, tcg_rn, tcg_rm);
6729 gen_helper_neon_qadd_s64(tcg_rd, cpu_env, tcg_rn, tcg_rm);
6732 case 0x5: /* SQSUB */
6734 gen_helper_neon_qsub_u64(tcg_rd, cpu_env, tcg_rn, tcg_rm);
6736 gen_helper_neon_qsub_s64(tcg_rd, cpu_env, tcg_rn, tcg_rm);
6739 case 0x6: /* CMGT, CMHI */
6740 /* 64 bit integer comparison, result = test ? (2^64 - 1) : 0.
6741 * We implement this using setcond (test) and then negating.
6743 cond = u ? TCG_COND_GTU : TCG_COND_GT;
6745 tcg_gen_setcond_i64(cond, tcg_rd, tcg_rn, tcg_rm);
6746 tcg_gen_neg_i64(tcg_rd, tcg_rd);
6748 case 0x7: /* CMGE, CMHS */
6749 cond = u ? TCG_COND_GEU : TCG_COND_GE;
6751 case 0x11: /* CMTST, CMEQ */
6756 /* CMTST : test is "if (X & Y != 0)". */
6757 tcg_gen_and_i64(tcg_rd, tcg_rn, tcg_rm);
6758 tcg_gen_setcondi_i64(TCG_COND_NE, tcg_rd, tcg_rd, 0);
6759 tcg_gen_neg_i64(tcg_rd, tcg_rd);
6761 case 0x8: /* SSHL, USHL */
6763 gen_helper_neon_shl_u64(tcg_rd, tcg_rn, tcg_rm);
6765 gen_helper_neon_shl_s64(tcg_rd, tcg_rn, tcg_rm);
6768 case 0x9: /* SQSHL, UQSHL */
6770 gen_helper_neon_qshl_u64(tcg_rd, cpu_env, tcg_rn, tcg_rm);
6772 gen_helper_neon_qshl_s64(tcg_rd, cpu_env, tcg_rn, tcg_rm);
6775 case 0xa: /* SRSHL, URSHL */
6777 gen_helper_neon_rshl_u64(tcg_rd, tcg_rn, tcg_rm);
6779 gen_helper_neon_rshl_s64(tcg_rd, tcg_rn, tcg_rm);
6782 case 0xb: /* SQRSHL, UQRSHL */
6784 gen_helper_neon_qrshl_u64(tcg_rd, cpu_env, tcg_rn, tcg_rm);
6786 gen_helper_neon_qrshl_s64(tcg_rd, cpu_env, tcg_rn, tcg_rm);
6789 case 0x10: /* ADD, SUB */
6791 tcg_gen_sub_i64(tcg_rd, tcg_rn, tcg_rm);
6793 tcg_gen_add_i64(tcg_rd, tcg_rn, tcg_rm);
6797 g_assert_not_reached();
6801 /* Handle the 3-same-operands float operations; shared by the scalar
6802 * and vector encodings. The caller must filter out any encodings
6803 * not allocated for the encoding it is dealing with.
6805 static void handle_3same_float(DisasContext *s, int size, int elements,
6806 int fpopcode, int rd, int rn, int rm)
6809 TCGv_ptr fpst = get_fpstatus_ptr();
6811 for (pass = 0; pass < elements; pass++) {
6814 TCGv_i64 tcg_op1 = tcg_temp_new_i64();
6815 TCGv_i64 tcg_op2 = tcg_temp_new_i64();
6816 TCGv_i64 tcg_res = tcg_temp_new_i64();
6818 read_vec_element(s, tcg_op1, rn, pass, MO_64);
6819 read_vec_element(s, tcg_op2, rm, pass, MO_64);
6822 case 0x39: /* FMLS */
6823 /* As usual for ARM, separate negation for fused multiply-add */
6824 gen_helper_vfp_negd(tcg_op1, tcg_op1);
6826 case 0x19: /* FMLA */
6827 read_vec_element(s, tcg_res, rd, pass, MO_64);
6828 gen_helper_vfp_muladdd(tcg_res, tcg_op1, tcg_op2,
6831 case 0x18: /* FMAXNM */
6832 gen_helper_vfp_maxnumd(tcg_res, tcg_op1, tcg_op2, fpst);
6834 case 0x1a: /* FADD */
6835 gen_helper_vfp_addd(tcg_res, tcg_op1, tcg_op2, fpst);
6837 case 0x1b: /* FMULX */
6838 gen_helper_vfp_mulxd(tcg_res, tcg_op1, tcg_op2, fpst);
6840 case 0x1c: /* FCMEQ */
6841 gen_helper_neon_ceq_f64(tcg_res, tcg_op1, tcg_op2, fpst);
6843 case 0x1e: /* FMAX */
6844 gen_helper_vfp_maxd(tcg_res, tcg_op1, tcg_op2, fpst);
6846 case 0x1f: /* FRECPS */
6847 gen_helper_recpsf_f64(tcg_res, tcg_op1, tcg_op2, fpst);
6849 case 0x38: /* FMINNM */
6850 gen_helper_vfp_minnumd(tcg_res, tcg_op1, tcg_op2, fpst);
6852 case 0x3a: /* FSUB */
6853 gen_helper_vfp_subd(tcg_res, tcg_op1, tcg_op2, fpst);
6855 case 0x3e: /* FMIN */
6856 gen_helper_vfp_mind(tcg_res, tcg_op1, tcg_op2, fpst);
6858 case 0x3f: /* FRSQRTS */
6859 gen_helper_rsqrtsf_f64(tcg_res, tcg_op1, tcg_op2, fpst);
6861 case 0x5b: /* FMUL */
6862 gen_helper_vfp_muld(tcg_res, tcg_op1, tcg_op2, fpst);
6864 case 0x5c: /* FCMGE */
6865 gen_helper_neon_cge_f64(tcg_res, tcg_op1, tcg_op2, fpst);
6867 case 0x5d: /* FACGE */
6868 gen_helper_neon_acge_f64(tcg_res, tcg_op1, tcg_op2, fpst);
6870 case 0x5f: /* FDIV */
6871 gen_helper_vfp_divd(tcg_res, tcg_op1, tcg_op2, fpst);
6873 case 0x7a: /* FABD */
6874 gen_helper_vfp_subd(tcg_res, tcg_op1, tcg_op2, fpst);
6875 gen_helper_vfp_absd(tcg_res, tcg_res);
6877 case 0x7c: /* FCMGT */
6878 gen_helper_neon_cgt_f64(tcg_res, tcg_op1, tcg_op2, fpst);
6880 case 0x7d: /* FACGT */
6881 gen_helper_neon_acgt_f64(tcg_res, tcg_op1, tcg_op2, fpst);
6884 g_assert_not_reached();
6887 write_vec_element(s, tcg_res, rd, pass, MO_64);
6889 tcg_temp_free_i64(tcg_res);
6890 tcg_temp_free_i64(tcg_op1);
6891 tcg_temp_free_i64(tcg_op2);
6894 TCGv_i32 tcg_op1 = tcg_temp_new_i32();
6895 TCGv_i32 tcg_op2 = tcg_temp_new_i32();
6896 TCGv_i32 tcg_res = tcg_temp_new_i32();
6898 read_vec_element_i32(s, tcg_op1, rn, pass, MO_32);
6899 read_vec_element_i32(s, tcg_op2, rm, pass, MO_32);
6902 case 0x39: /* FMLS */
6903 /* As usual for ARM, separate negation for fused multiply-add */
6904 gen_helper_vfp_negs(tcg_op1, tcg_op1);
6906 case 0x19: /* FMLA */
6907 read_vec_element_i32(s, tcg_res, rd, pass, MO_32);
6908 gen_helper_vfp_muladds(tcg_res, tcg_op1, tcg_op2,
6911 case 0x1a: /* FADD */
6912 gen_helper_vfp_adds(tcg_res, tcg_op1, tcg_op2, fpst);
6914 case 0x1b: /* FMULX */
6915 gen_helper_vfp_mulxs(tcg_res, tcg_op1, tcg_op2, fpst);
6917 case 0x1c: /* FCMEQ */
6918 gen_helper_neon_ceq_f32(tcg_res, tcg_op1, tcg_op2, fpst);
6920 case 0x1e: /* FMAX */
6921 gen_helper_vfp_maxs(tcg_res, tcg_op1, tcg_op2, fpst);
6923 case 0x1f: /* FRECPS */
6924 gen_helper_recpsf_f32(tcg_res, tcg_op1, tcg_op2, fpst);
6926 case 0x18: /* FMAXNM */
6927 gen_helper_vfp_maxnums(tcg_res, tcg_op1, tcg_op2, fpst);
6929 case 0x38: /* FMINNM */
6930 gen_helper_vfp_minnums(tcg_res, tcg_op1, tcg_op2, fpst);
6932 case 0x3a: /* FSUB */
6933 gen_helper_vfp_subs(tcg_res, tcg_op1, tcg_op2, fpst);
6935 case 0x3e: /* FMIN */
6936 gen_helper_vfp_mins(tcg_res, tcg_op1, tcg_op2, fpst);
6938 case 0x3f: /* FRSQRTS */
6939 gen_helper_rsqrtsf_f32(tcg_res, tcg_op1, tcg_op2, fpst);
6941 case 0x5b: /* FMUL */
6942 gen_helper_vfp_muls(tcg_res, tcg_op1, tcg_op2, fpst);
6944 case 0x5c: /* FCMGE */
6945 gen_helper_neon_cge_f32(tcg_res, tcg_op1, tcg_op2, fpst);
6947 case 0x5d: /* FACGE */
6948 gen_helper_neon_acge_f32(tcg_res, tcg_op1, tcg_op2, fpst);
6950 case 0x5f: /* FDIV */
6951 gen_helper_vfp_divs(tcg_res, tcg_op1, tcg_op2, fpst);
6953 case 0x7a: /* FABD */
6954 gen_helper_vfp_subs(tcg_res, tcg_op1, tcg_op2, fpst);
6955 gen_helper_vfp_abss(tcg_res, tcg_res);
6957 case 0x7c: /* FCMGT */
6958 gen_helper_neon_cgt_f32(tcg_res, tcg_op1, tcg_op2, fpst);
6960 case 0x7d: /* FACGT */
6961 gen_helper_neon_acgt_f32(tcg_res, tcg_op1, tcg_op2, fpst);
6964 g_assert_not_reached();
6967 if (elements == 1) {
6968 /* scalar single so clear high part */
6969 TCGv_i64 tcg_tmp = tcg_temp_new_i64();
6971 tcg_gen_extu_i32_i64(tcg_tmp, tcg_res);
6972 write_vec_element(s, tcg_tmp, rd, pass, MO_64);
6973 tcg_temp_free_i64(tcg_tmp);
6975 write_vec_element_i32(s, tcg_res, rd, pass, MO_32);
6978 tcg_temp_free_i32(tcg_res);
6979 tcg_temp_free_i32(tcg_op1);
6980 tcg_temp_free_i32(tcg_op2);
6984 tcg_temp_free_ptr(fpst);
6986 if ((elements << size) < 4) {
6987 /* scalar, or non-quad vector op */
6988 clear_vec_high(s, rd);
6992 /* C3.6.11 AdvSIMD scalar three same
6993 * 31 30 29 28 24 23 22 21 20 16 15 11 10 9 5 4 0
6994 * +-----+---+-----------+------+---+------+--------+---+------+------+
6995 * | 0 1 | U | 1 1 1 1 0 | size | 1 | Rm | opcode | 1 | Rn | Rd |
6996 * +-----+---+-----------+------+---+------+--------+---+------+------+
6998 static void disas_simd_scalar_three_reg_same(DisasContext *s, uint32_t insn)
7000 int rd = extract32(insn, 0, 5);
7001 int rn = extract32(insn, 5, 5);
7002 int opcode = extract32(insn, 11, 5);
7003 int rm = extract32(insn, 16, 5);
7004 int size = extract32(insn, 22, 2);
7005 bool u = extract32(insn, 29, 1);
7008 if (opcode >= 0x18) {
7009 /* Floating point: U, size[1] and opcode indicate operation */
7010 int fpopcode = opcode | (extract32(size, 1, 1) << 5) | (u << 6);
7012 case 0x1b: /* FMULX */
7013 case 0x1f: /* FRECPS */
7014 case 0x3f: /* FRSQRTS */
7015 case 0x5d: /* FACGE */
7016 case 0x7d: /* FACGT */
7017 case 0x1c: /* FCMEQ */
7018 case 0x5c: /* FCMGE */
7019 case 0x7c: /* FCMGT */
7020 case 0x7a: /* FABD */
7023 unallocated_encoding(s);
7027 if (!fp_access_check(s)) {
7031 handle_3same_float(s, extract32(size, 0, 1), 1, fpopcode, rd, rn, rm);
7036 case 0x1: /* SQADD, UQADD */
7037 case 0x5: /* SQSUB, UQSUB */
7038 case 0x9: /* SQSHL, UQSHL */
7039 case 0xb: /* SQRSHL, UQRSHL */
7041 case 0x8: /* SSHL, USHL */
7042 case 0xa: /* SRSHL, URSHL */
7043 case 0x6: /* CMGT, CMHI */
7044 case 0x7: /* CMGE, CMHS */
7045 case 0x11: /* CMTST, CMEQ */
7046 case 0x10: /* ADD, SUB (vector) */
7048 unallocated_encoding(s);
7052 case 0x16: /* SQDMULH, SQRDMULH (vector) */
7053 if (size != 1 && size != 2) {
7054 unallocated_encoding(s);
7059 unallocated_encoding(s);
7063 if (!fp_access_check(s)) {
7067 tcg_rd = tcg_temp_new_i64();
7070 TCGv_i64 tcg_rn = read_fp_dreg(s, rn);
7071 TCGv_i64 tcg_rm = read_fp_dreg(s, rm);
7073 handle_3same_64(s, opcode, u, tcg_rd, tcg_rn, tcg_rm);
7074 tcg_temp_free_i64(tcg_rn);
7075 tcg_temp_free_i64(tcg_rm);
7077 /* Do a single operation on the lowest element in the vector.
7078 * We use the standard Neon helpers and rely on 0 OP 0 == 0 with
7079 * no side effects for all these operations.
7080 * OPTME: special-purpose helpers would avoid doing some
7081 * unnecessary work in the helper for the 8 and 16 bit cases.
7083 NeonGenTwoOpEnvFn *genenvfn;
7084 TCGv_i32 tcg_rn = tcg_temp_new_i32();
7085 TCGv_i32 tcg_rm = tcg_temp_new_i32();
7086 TCGv_i32 tcg_rd32 = tcg_temp_new_i32();
7088 read_vec_element_i32(s, tcg_rn, rn, 0, size);
7089 read_vec_element_i32(s, tcg_rm, rm, 0, size);
7092 case 0x1: /* SQADD, UQADD */
7094 static NeonGenTwoOpEnvFn * const fns[3][2] = {
7095 { gen_helper_neon_qadd_s8, gen_helper_neon_qadd_u8 },
7096 { gen_helper_neon_qadd_s16, gen_helper_neon_qadd_u16 },
7097 { gen_helper_neon_qadd_s32, gen_helper_neon_qadd_u32 },
7099 genenvfn = fns[size][u];
7102 case 0x5: /* SQSUB, UQSUB */
7104 static NeonGenTwoOpEnvFn * const fns[3][2] = {
7105 { gen_helper_neon_qsub_s8, gen_helper_neon_qsub_u8 },
7106 { gen_helper_neon_qsub_s16, gen_helper_neon_qsub_u16 },
7107 { gen_helper_neon_qsub_s32, gen_helper_neon_qsub_u32 },
7109 genenvfn = fns[size][u];
7112 case 0x9: /* SQSHL, UQSHL */
7114 static NeonGenTwoOpEnvFn * const fns[3][2] = {
7115 { gen_helper_neon_qshl_s8, gen_helper_neon_qshl_u8 },
7116 { gen_helper_neon_qshl_s16, gen_helper_neon_qshl_u16 },
7117 { gen_helper_neon_qshl_s32, gen_helper_neon_qshl_u32 },
7119 genenvfn = fns[size][u];
7122 case 0xb: /* SQRSHL, UQRSHL */
7124 static NeonGenTwoOpEnvFn * const fns[3][2] = {
7125 { gen_helper_neon_qrshl_s8, gen_helper_neon_qrshl_u8 },
7126 { gen_helper_neon_qrshl_s16, gen_helper_neon_qrshl_u16 },
7127 { gen_helper_neon_qrshl_s32, gen_helper_neon_qrshl_u32 },
7129 genenvfn = fns[size][u];
7132 case 0x16: /* SQDMULH, SQRDMULH */
7134 static NeonGenTwoOpEnvFn * const fns[2][2] = {
7135 { gen_helper_neon_qdmulh_s16, gen_helper_neon_qrdmulh_s16 },
7136 { gen_helper_neon_qdmulh_s32, gen_helper_neon_qrdmulh_s32 },
7138 assert(size == 1 || size == 2);
7139 genenvfn = fns[size - 1][u];
7143 g_assert_not_reached();
7146 genenvfn(tcg_rd32, cpu_env, tcg_rn, tcg_rm);
7147 tcg_gen_extu_i32_i64(tcg_rd, tcg_rd32);
7148 tcg_temp_free_i32(tcg_rd32);
7149 tcg_temp_free_i32(tcg_rn);
7150 tcg_temp_free_i32(tcg_rm);
7153 write_fp_dreg(s, rd, tcg_rd);
7155 tcg_temp_free_i64(tcg_rd);
7158 static void handle_2misc_64(DisasContext *s, int opcode, bool u,
7159 TCGv_i64 tcg_rd, TCGv_i64 tcg_rn,
7160 TCGv_i32 tcg_rmode, TCGv_ptr tcg_fpstatus)
7162 /* Handle 64->64 opcodes which are shared between the scalar and
7163 * vector 2-reg-misc groups. We cover every integer opcode where size == 3
7164 * is valid in either group and also the double-precision fp ops.
7165 * The caller only need provide tcg_rmode and tcg_fpstatus if the op
7171 case 0x4: /* CLS, CLZ */
7173 gen_helper_clz64(tcg_rd, tcg_rn);
7175 gen_helper_cls64(tcg_rd, tcg_rn);
7179 /* This opcode is shared with CNT and RBIT but we have earlier
7180 * enforced that size == 3 if and only if this is the NOT insn.
7182 tcg_gen_not_i64(tcg_rd, tcg_rn);
7184 case 0x7: /* SQABS, SQNEG */
7186 gen_helper_neon_qneg_s64(tcg_rd, cpu_env, tcg_rn);
7188 gen_helper_neon_qabs_s64(tcg_rd, cpu_env, tcg_rn);
7191 case 0xa: /* CMLT */
7192 /* 64 bit integer comparison against zero, result is
7193 * test ? (2^64 - 1) : 0. We implement via setcond(!test) and
7198 tcg_gen_setcondi_i64(cond, tcg_rd, tcg_rn, 0);
7199 tcg_gen_neg_i64(tcg_rd, tcg_rd);
7201 case 0x8: /* CMGT, CMGE */
7202 cond = u ? TCG_COND_GE : TCG_COND_GT;
7204 case 0x9: /* CMEQ, CMLE */
7205 cond = u ? TCG_COND_LE : TCG_COND_EQ;
7207 case 0xb: /* ABS, NEG */
7209 tcg_gen_neg_i64(tcg_rd, tcg_rn);
7211 TCGv_i64 tcg_zero = tcg_const_i64(0);
7212 tcg_gen_neg_i64(tcg_rd, tcg_rn);
7213 tcg_gen_movcond_i64(TCG_COND_GT, tcg_rd, tcg_rn, tcg_zero,
7215 tcg_temp_free_i64(tcg_zero);
7218 case 0x2f: /* FABS */
7219 gen_helper_vfp_absd(tcg_rd, tcg_rn);
7221 case 0x6f: /* FNEG */
7222 gen_helper_vfp_negd(tcg_rd, tcg_rn);
7224 case 0x7f: /* FSQRT */
7225 gen_helper_vfp_sqrtd(tcg_rd, tcg_rn, cpu_env);
7227 case 0x1a: /* FCVTNS */
7228 case 0x1b: /* FCVTMS */
7229 case 0x1c: /* FCVTAS */
7230 case 0x3a: /* FCVTPS */
7231 case 0x3b: /* FCVTZS */
7233 TCGv_i32 tcg_shift = tcg_const_i32(0);
7234 gen_helper_vfp_tosqd(tcg_rd, tcg_rn, tcg_shift, tcg_fpstatus);
7235 tcg_temp_free_i32(tcg_shift);
7238 case 0x5a: /* FCVTNU */
7239 case 0x5b: /* FCVTMU */
7240 case 0x5c: /* FCVTAU */
7241 case 0x7a: /* FCVTPU */
7242 case 0x7b: /* FCVTZU */
7244 TCGv_i32 tcg_shift = tcg_const_i32(0);
7245 gen_helper_vfp_touqd(tcg_rd, tcg_rn, tcg_shift, tcg_fpstatus);
7246 tcg_temp_free_i32(tcg_shift);
7249 case 0x18: /* FRINTN */
7250 case 0x19: /* FRINTM */
7251 case 0x38: /* FRINTP */
7252 case 0x39: /* FRINTZ */
7253 case 0x58: /* FRINTA */
7254 case 0x79: /* FRINTI */
7255 gen_helper_rintd(tcg_rd, tcg_rn, tcg_fpstatus);
7257 case 0x59: /* FRINTX */
7258 gen_helper_rintd_exact(tcg_rd, tcg_rn, tcg_fpstatus);
7261 g_assert_not_reached();
7265 static void handle_2misc_fcmp_zero(DisasContext *s, int opcode,
7266 bool is_scalar, bool is_u, bool is_q,
7267 int size, int rn, int rd)
7269 bool is_double = (size == 3);
7272 if (!fp_access_check(s)) {
7276 fpst = get_fpstatus_ptr();
7279 TCGv_i64 tcg_op = tcg_temp_new_i64();
7280 TCGv_i64 tcg_zero = tcg_const_i64(0);
7281 TCGv_i64 tcg_res = tcg_temp_new_i64();
7282 NeonGenTwoDoubleOPFn *genfn;
7287 case 0x2e: /* FCMLT (zero) */
7290 case 0x2c: /* FCMGT (zero) */
7291 genfn = gen_helper_neon_cgt_f64;
7293 case 0x2d: /* FCMEQ (zero) */
7294 genfn = gen_helper_neon_ceq_f64;
7296 case 0x6d: /* FCMLE (zero) */
7299 case 0x6c: /* FCMGE (zero) */
7300 genfn = gen_helper_neon_cge_f64;
7303 g_assert_not_reached();
7306 for (pass = 0; pass < (is_scalar ? 1 : 2); pass++) {
7307 read_vec_element(s, tcg_op, rn, pass, MO_64);
7309 genfn(tcg_res, tcg_zero, tcg_op, fpst);
7311 genfn(tcg_res, tcg_op, tcg_zero, fpst);
7313 write_vec_element(s, tcg_res, rd, pass, MO_64);
7316 clear_vec_high(s, rd);
7319 tcg_temp_free_i64(tcg_res);
7320 tcg_temp_free_i64(tcg_zero);
7321 tcg_temp_free_i64(tcg_op);
7323 TCGv_i32 tcg_op = tcg_temp_new_i32();
7324 TCGv_i32 tcg_zero = tcg_const_i32(0);
7325 TCGv_i32 tcg_res = tcg_temp_new_i32();
7326 NeonGenTwoSingleOPFn *genfn;
7328 int pass, maxpasses;
7331 case 0x2e: /* FCMLT (zero) */
7334 case 0x2c: /* FCMGT (zero) */
7335 genfn = gen_helper_neon_cgt_f32;
7337 case 0x2d: /* FCMEQ (zero) */
7338 genfn = gen_helper_neon_ceq_f32;
7340 case 0x6d: /* FCMLE (zero) */
7343 case 0x6c: /* FCMGE (zero) */
7344 genfn = gen_helper_neon_cge_f32;
7347 g_assert_not_reached();
7353 maxpasses = is_q ? 4 : 2;
7356 for (pass = 0; pass < maxpasses; pass++) {
7357 read_vec_element_i32(s, tcg_op, rn, pass, MO_32);
7359 genfn(tcg_res, tcg_zero, tcg_op, fpst);
7361 genfn(tcg_res, tcg_op, tcg_zero, fpst);
7364 write_fp_sreg(s, rd, tcg_res);
7366 write_vec_element_i32(s, tcg_res, rd, pass, MO_32);
7369 tcg_temp_free_i32(tcg_res);
7370 tcg_temp_free_i32(tcg_zero);
7371 tcg_temp_free_i32(tcg_op);
7372 if (!is_q && !is_scalar) {
7373 clear_vec_high(s, rd);
7377 tcg_temp_free_ptr(fpst);
7380 static void handle_2misc_reciprocal(DisasContext *s, int opcode,
7381 bool is_scalar, bool is_u, bool is_q,
7382 int size, int rn, int rd)
7384 bool is_double = (size == 3);
7385 TCGv_ptr fpst = get_fpstatus_ptr();
7388 TCGv_i64 tcg_op = tcg_temp_new_i64();
7389 TCGv_i64 tcg_res = tcg_temp_new_i64();
7392 for (pass = 0; pass < (is_scalar ? 1 : 2); pass++) {
7393 read_vec_element(s, tcg_op, rn, pass, MO_64);
7395 case 0x3d: /* FRECPE */
7396 gen_helper_recpe_f64(tcg_res, tcg_op, fpst);
7398 case 0x3f: /* FRECPX */
7399 gen_helper_frecpx_f64(tcg_res, tcg_op, fpst);
7401 case 0x7d: /* FRSQRTE */
7402 gen_helper_rsqrte_f64(tcg_res, tcg_op, fpst);
7405 g_assert_not_reached();
7407 write_vec_element(s, tcg_res, rd, pass, MO_64);
7410 clear_vec_high(s, rd);
7413 tcg_temp_free_i64(tcg_res);
7414 tcg_temp_free_i64(tcg_op);
7416 TCGv_i32 tcg_op = tcg_temp_new_i32();
7417 TCGv_i32 tcg_res = tcg_temp_new_i32();
7418 int pass, maxpasses;
7423 maxpasses = is_q ? 4 : 2;
7426 for (pass = 0; pass < maxpasses; pass++) {
7427 read_vec_element_i32(s, tcg_op, rn, pass, MO_32);
7430 case 0x3c: /* URECPE */
7431 gen_helper_recpe_u32(tcg_res, tcg_op, fpst);
7433 case 0x3d: /* FRECPE */
7434 gen_helper_recpe_f32(tcg_res, tcg_op, fpst);
7436 case 0x3f: /* FRECPX */
7437 gen_helper_frecpx_f32(tcg_res, tcg_op, fpst);
7439 case 0x7d: /* FRSQRTE */
7440 gen_helper_rsqrte_f32(tcg_res, tcg_op, fpst);
7443 g_assert_not_reached();
7447 write_fp_sreg(s, rd, tcg_res);
7449 write_vec_element_i32(s, tcg_res, rd, pass, MO_32);
7452 tcg_temp_free_i32(tcg_res);
7453 tcg_temp_free_i32(tcg_op);
7454 if (!is_q && !is_scalar) {
7455 clear_vec_high(s, rd);
7458 tcg_temp_free_ptr(fpst);
7461 static void handle_2misc_narrow(DisasContext *s, bool scalar,
7462 int opcode, bool u, bool is_q,
7463 int size, int rn, int rd)
7465 /* Handle 2-reg-misc ops which are narrowing (so each 2*size element
7466 * in the source becomes a size element in the destination).
7469 TCGv_i32 tcg_res[2];
7470 int destelt = is_q ? 2 : 0;
7471 int passes = scalar ? 1 : 2;
7474 tcg_res[1] = tcg_const_i32(0);
7477 for (pass = 0; pass < passes; pass++) {
7478 TCGv_i64 tcg_op = tcg_temp_new_i64();
7479 NeonGenNarrowFn *genfn = NULL;
7480 NeonGenNarrowEnvFn *genenvfn = NULL;
7483 read_vec_element(s, tcg_op, rn, pass, size + 1);
7485 read_vec_element(s, tcg_op, rn, pass, MO_64);
7487 tcg_res[pass] = tcg_temp_new_i32();
7490 case 0x12: /* XTN, SQXTUN */
7492 static NeonGenNarrowFn * const xtnfns[3] = {
7493 gen_helper_neon_narrow_u8,
7494 gen_helper_neon_narrow_u16,
7495 tcg_gen_trunc_i64_i32,
7497 static NeonGenNarrowEnvFn * const sqxtunfns[3] = {
7498 gen_helper_neon_unarrow_sat8,
7499 gen_helper_neon_unarrow_sat16,
7500 gen_helper_neon_unarrow_sat32,
7503 genenvfn = sqxtunfns[size];
7505 genfn = xtnfns[size];
7509 case 0x14: /* SQXTN, UQXTN */
7511 static NeonGenNarrowEnvFn * const fns[3][2] = {
7512 { gen_helper_neon_narrow_sat_s8,
7513 gen_helper_neon_narrow_sat_u8 },
7514 { gen_helper_neon_narrow_sat_s16,
7515 gen_helper_neon_narrow_sat_u16 },
7516 { gen_helper_neon_narrow_sat_s32,
7517 gen_helper_neon_narrow_sat_u32 },
7519 genenvfn = fns[size][u];
7522 case 0x16: /* FCVTN, FCVTN2 */
7523 /* 32 bit to 16 bit or 64 bit to 32 bit float conversion */
7525 gen_helper_vfp_fcvtsd(tcg_res[pass], tcg_op, cpu_env);
7527 TCGv_i32 tcg_lo = tcg_temp_new_i32();
7528 TCGv_i32 tcg_hi = tcg_temp_new_i32();
7529 tcg_gen_trunc_i64_i32(tcg_lo, tcg_op);
7530 gen_helper_vfp_fcvt_f32_to_f16(tcg_lo, tcg_lo, cpu_env);
7531 tcg_gen_shri_i64(tcg_op, tcg_op, 32);
7532 tcg_gen_trunc_i64_i32(tcg_hi, tcg_op);
7533 gen_helper_vfp_fcvt_f32_to_f16(tcg_hi, tcg_hi, cpu_env);
7534 tcg_gen_deposit_i32(tcg_res[pass], tcg_lo, tcg_hi, 16, 16);
7535 tcg_temp_free_i32(tcg_lo);
7536 tcg_temp_free_i32(tcg_hi);
7539 case 0x56: /* FCVTXN, FCVTXN2 */
7540 /* 64 bit to 32 bit float conversion
7541 * with von Neumann rounding (round to odd)
7544 gen_helper_fcvtx_f64_to_f32(tcg_res[pass], tcg_op, cpu_env);
7547 g_assert_not_reached();
7551 genfn(tcg_res[pass], tcg_op);
7552 } else if (genenvfn) {
7553 genenvfn(tcg_res[pass], cpu_env, tcg_op);
7556 tcg_temp_free_i64(tcg_op);
7559 for (pass = 0; pass < 2; pass++) {
7560 write_vec_element_i32(s, tcg_res[pass], rd, destelt + pass, MO_32);
7561 tcg_temp_free_i32(tcg_res[pass]);
7564 clear_vec_high(s, rd);
7568 /* Remaining saturating accumulating ops */
7569 static void handle_2misc_satacc(DisasContext *s, bool is_scalar, bool is_u,
7570 bool is_q, int size, int rn, int rd)
7572 bool is_double = (size == 3);
7575 TCGv_i64 tcg_rn = tcg_temp_new_i64();
7576 TCGv_i64 tcg_rd = tcg_temp_new_i64();
7579 for (pass = 0; pass < (is_scalar ? 1 : 2); pass++) {
7580 read_vec_element(s, tcg_rn, rn, pass, MO_64);
7581 read_vec_element(s, tcg_rd, rd, pass, MO_64);
7583 if (is_u) { /* USQADD */
7584 gen_helper_neon_uqadd_s64(tcg_rd, cpu_env, tcg_rn, tcg_rd);
7585 } else { /* SUQADD */
7586 gen_helper_neon_sqadd_u64(tcg_rd, cpu_env, tcg_rn, tcg_rd);
7588 write_vec_element(s, tcg_rd, rd, pass, MO_64);
7591 clear_vec_high(s, rd);
7594 tcg_temp_free_i64(tcg_rd);
7595 tcg_temp_free_i64(tcg_rn);
7597 TCGv_i32 tcg_rn = tcg_temp_new_i32();
7598 TCGv_i32 tcg_rd = tcg_temp_new_i32();
7599 int pass, maxpasses;
7604 maxpasses = is_q ? 4 : 2;
7607 for (pass = 0; pass < maxpasses; pass++) {
7609 read_vec_element_i32(s, tcg_rn, rn, pass, size);
7610 read_vec_element_i32(s, tcg_rd, rd, pass, size);
7612 read_vec_element_i32(s, tcg_rn, rn, pass, MO_32);
7613 read_vec_element_i32(s, tcg_rd, rd, pass, MO_32);
7616 if (is_u) { /* USQADD */
7619 gen_helper_neon_uqadd_s8(tcg_rd, cpu_env, tcg_rn, tcg_rd);
7622 gen_helper_neon_uqadd_s16(tcg_rd, cpu_env, tcg_rn, tcg_rd);
7625 gen_helper_neon_uqadd_s32(tcg_rd, cpu_env, tcg_rn, tcg_rd);
7628 g_assert_not_reached();
7630 } else { /* SUQADD */
7633 gen_helper_neon_sqadd_u8(tcg_rd, cpu_env, tcg_rn, tcg_rd);
7636 gen_helper_neon_sqadd_u16(tcg_rd, cpu_env, tcg_rn, tcg_rd);
7639 gen_helper_neon_sqadd_u32(tcg_rd, cpu_env, tcg_rn, tcg_rd);
7642 g_assert_not_reached();
7647 TCGv_i64 tcg_zero = tcg_const_i64(0);
7648 write_vec_element(s, tcg_zero, rd, 0, MO_64);
7649 tcg_temp_free_i64(tcg_zero);
7651 write_vec_element_i32(s, tcg_rd, rd, pass, MO_32);
7655 clear_vec_high(s, rd);
7658 tcg_temp_free_i32(tcg_rd);
7659 tcg_temp_free_i32(tcg_rn);
7663 /* C3.6.12 AdvSIMD scalar two reg misc
7664 * 31 30 29 28 24 23 22 21 17 16 12 11 10 9 5 4 0
7665 * +-----+---+-----------+------+-----------+--------+-----+------+------+
7666 * | 0 1 | U | 1 1 1 1 0 | size | 1 0 0 0 0 | opcode | 1 0 | Rn | Rd |
7667 * +-----+---+-----------+------+-----------+--------+-----+------+------+
7669 static void disas_simd_scalar_two_reg_misc(DisasContext *s, uint32_t insn)
7671 int rd = extract32(insn, 0, 5);
7672 int rn = extract32(insn, 5, 5);
7673 int opcode = extract32(insn, 12, 5);
7674 int size = extract32(insn, 22, 2);
7675 bool u = extract32(insn, 29, 1);
7676 bool is_fcvt = false;
7679 TCGv_ptr tcg_fpstatus;
7682 case 0x3: /* USQADD / SUQADD*/
7683 if (!fp_access_check(s)) {
7686 handle_2misc_satacc(s, true, u, false, size, rn, rd);
7688 case 0x7: /* SQABS / SQNEG */
7690 case 0xa: /* CMLT */
7692 unallocated_encoding(s);
7696 case 0x8: /* CMGT, CMGE */
7697 case 0x9: /* CMEQ, CMLE */
7698 case 0xb: /* ABS, NEG */
7700 unallocated_encoding(s);
7704 case 0x12: /* SQXTUN */
7706 unallocated_encoding(s);
7710 case 0x14: /* SQXTN, UQXTN */
7712 unallocated_encoding(s);
7715 if (!fp_access_check(s)) {
7718 handle_2misc_narrow(s, true, opcode, u, false, size, rn, rd);
7723 /* Floating point: U, size[1] and opcode indicate operation;
7724 * size[0] indicates single or double precision.
7726 opcode |= (extract32(size, 1, 1) << 5) | (u << 6);
7727 size = extract32(size, 0, 1) ? 3 : 2;
7729 case 0x2c: /* FCMGT (zero) */
7730 case 0x2d: /* FCMEQ (zero) */
7731 case 0x2e: /* FCMLT (zero) */
7732 case 0x6c: /* FCMGE (zero) */
7733 case 0x6d: /* FCMLE (zero) */
7734 handle_2misc_fcmp_zero(s, opcode, true, u, true, size, rn, rd);
7736 case 0x1d: /* SCVTF */
7737 case 0x5d: /* UCVTF */
7739 bool is_signed = (opcode == 0x1d);
7740 if (!fp_access_check(s)) {
7743 handle_simd_intfp_conv(s, rd, rn, 1, is_signed, 0, size);
7746 case 0x3d: /* FRECPE */
7747 case 0x3f: /* FRECPX */
7748 case 0x7d: /* FRSQRTE */
7749 if (!fp_access_check(s)) {
7752 handle_2misc_reciprocal(s, opcode, true, u, true, size, rn, rd);
7754 case 0x1a: /* FCVTNS */
7755 case 0x1b: /* FCVTMS */
7756 case 0x3a: /* FCVTPS */
7757 case 0x3b: /* FCVTZS */
7758 case 0x5a: /* FCVTNU */
7759 case 0x5b: /* FCVTMU */
7760 case 0x7a: /* FCVTPU */
7761 case 0x7b: /* FCVTZU */
7763 rmode = extract32(opcode, 5, 1) | (extract32(opcode, 0, 1) << 1);
7765 case 0x1c: /* FCVTAS */
7766 case 0x5c: /* FCVTAU */
7767 /* TIEAWAY doesn't fit in the usual rounding mode encoding */
7769 rmode = FPROUNDING_TIEAWAY;
7771 case 0x56: /* FCVTXN, FCVTXN2 */
7773 unallocated_encoding(s);
7776 if (!fp_access_check(s)) {
7779 handle_2misc_narrow(s, true, opcode, u, false, size - 1, rn, rd);
7782 unallocated_encoding(s);
7787 unallocated_encoding(s);
7791 if (!fp_access_check(s)) {
7796 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rmode));
7797 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
7798 tcg_fpstatus = get_fpstatus_ptr();
7800 TCGV_UNUSED_I32(tcg_rmode);
7801 TCGV_UNUSED_PTR(tcg_fpstatus);
7805 TCGv_i64 tcg_rn = read_fp_dreg(s, rn);
7806 TCGv_i64 tcg_rd = tcg_temp_new_i64();
7808 handle_2misc_64(s, opcode, u, tcg_rd, tcg_rn, tcg_rmode, tcg_fpstatus);
7809 write_fp_dreg(s, rd, tcg_rd);
7810 tcg_temp_free_i64(tcg_rd);
7811 tcg_temp_free_i64(tcg_rn);
7813 TCGv_i32 tcg_rn = tcg_temp_new_i32();
7814 TCGv_i32 tcg_rd = tcg_temp_new_i32();
7816 read_vec_element_i32(s, tcg_rn, rn, 0, size);
7819 case 0x7: /* SQABS, SQNEG */
7821 NeonGenOneOpEnvFn *genfn;
7822 static NeonGenOneOpEnvFn * const fns[3][2] = {
7823 { gen_helper_neon_qabs_s8, gen_helper_neon_qneg_s8 },
7824 { gen_helper_neon_qabs_s16, gen_helper_neon_qneg_s16 },
7825 { gen_helper_neon_qabs_s32, gen_helper_neon_qneg_s32 },
7827 genfn = fns[size][u];
7828 genfn(tcg_rd, cpu_env, tcg_rn);
7831 case 0x1a: /* FCVTNS */
7832 case 0x1b: /* FCVTMS */
7833 case 0x1c: /* FCVTAS */
7834 case 0x3a: /* FCVTPS */
7835 case 0x3b: /* FCVTZS */
7837 TCGv_i32 tcg_shift = tcg_const_i32(0);
7838 gen_helper_vfp_tosls(tcg_rd, tcg_rn, tcg_shift, tcg_fpstatus);
7839 tcg_temp_free_i32(tcg_shift);
7842 case 0x5a: /* FCVTNU */
7843 case 0x5b: /* FCVTMU */
7844 case 0x5c: /* FCVTAU */
7845 case 0x7a: /* FCVTPU */
7846 case 0x7b: /* FCVTZU */
7848 TCGv_i32 tcg_shift = tcg_const_i32(0);
7849 gen_helper_vfp_touls(tcg_rd, tcg_rn, tcg_shift, tcg_fpstatus);
7850 tcg_temp_free_i32(tcg_shift);
7854 g_assert_not_reached();
7857 write_fp_sreg(s, rd, tcg_rd);
7858 tcg_temp_free_i32(tcg_rd);
7859 tcg_temp_free_i32(tcg_rn);
7863 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
7864 tcg_temp_free_i32(tcg_rmode);
7865 tcg_temp_free_ptr(tcg_fpstatus);
7869 /* SSHR[RA]/USHR[RA] - Vector shift right (optional rounding/accumulate) */
7870 static void handle_vec_simd_shri(DisasContext *s, bool is_q, bool is_u,
7871 int immh, int immb, int opcode, int rn, int rd)
7873 int size = 32 - clz32(immh) - 1;
7874 int immhb = immh << 3 | immb;
7875 int shift = 2 * (8 << size) - immhb;
7876 bool accumulate = false;
7878 bool insert = false;
7879 int dsize = is_q ? 128 : 64;
7880 int esize = 8 << size;
7881 int elements = dsize/esize;
7882 TCGMemOp memop = size | (is_u ? 0 : MO_SIGN);
7883 TCGv_i64 tcg_rn = new_tmp_a64(s);
7884 TCGv_i64 tcg_rd = new_tmp_a64(s);
7888 if (extract32(immh, 3, 1) && !is_q) {
7889 unallocated_encoding(s);
7893 if (size > 3 && !is_q) {
7894 unallocated_encoding(s);
7898 if (!fp_access_check(s)) {
7903 case 0x02: /* SSRA / USRA (accumulate) */
7906 case 0x04: /* SRSHR / URSHR (rounding) */
7909 case 0x06: /* SRSRA / URSRA (accum + rounding) */
7910 accumulate = round = true;
7912 case 0x08: /* SRI */
7918 uint64_t round_const = 1ULL << (shift - 1);
7919 tcg_round = tcg_const_i64(round_const);
7921 TCGV_UNUSED_I64(tcg_round);
7924 for (i = 0; i < elements; i++) {
7925 read_vec_element(s, tcg_rn, rn, i, memop);
7926 if (accumulate || insert) {
7927 read_vec_element(s, tcg_rd, rd, i, memop);
7931 handle_shri_with_ins(tcg_rd, tcg_rn, size, shift);
7933 handle_shri_with_rndacc(tcg_rd, tcg_rn, tcg_round,
7934 accumulate, is_u, size, shift);
7937 write_vec_element(s, tcg_rd, rd, i, size);
7941 clear_vec_high(s, rd);
7945 tcg_temp_free_i64(tcg_round);
7949 /* SHL/SLI - Vector shift left */
7950 static void handle_vec_simd_shli(DisasContext *s, bool is_q, bool insert,
7951 int immh, int immb, int opcode, int rn, int rd)
7953 int size = 32 - clz32(immh) - 1;
7954 int immhb = immh << 3 | immb;
7955 int shift = immhb - (8 << size);
7956 int dsize = is_q ? 128 : 64;
7957 int esize = 8 << size;
7958 int elements = dsize/esize;
7959 TCGv_i64 tcg_rn = new_tmp_a64(s);
7960 TCGv_i64 tcg_rd = new_tmp_a64(s);
7963 if (extract32(immh, 3, 1) && !is_q) {
7964 unallocated_encoding(s);
7968 if (size > 3 && !is_q) {
7969 unallocated_encoding(s);
7973 if (!fp_access_check(s)) {
7977 for (i = 0; i < elements; i++) {
7978 read_vec_element(s, tcg_rn, rn, i, size);
7980 read_vec_element(s, tcg_rd, rd, i, size);
7983 handle_shli_with_ins(tcg_rd, tcg_rn, insert, shift);
7985 write_vec_element(s, tcg_rd, rd, i, size);
7989 clear_vec_high(s, rd);
7993 /* USHLL/SHLL - Vector shift left with widening */
7994 static void handle_vec_simd_wshli(DisasContext *s, bool is_q, bool is_u,
7995 int immh, int immb, int opcode, int rn, int rd)
7997 int size = 32 - clz32(immh) - 1;
7998 int immhb = immh << 3 | immb;
7999 int shift = immhb - (8 << size);
8001 int esize = 8 << size;
8002 int elements = dsize/esize;
8003 TCGv_i64 tcg_rn = new_tmp_a64(s);
8004 TCGv_i64 tcg_rd = new_tmp_a64(s);
8008 unallocated_encoding(s);
8012 if (!fp_access_check(s)) {
8016 /* For the LL variants the store is larger than the load,
8017 * so if rd == rn we would overwrite parts of our input.
8018 * So load everything right now and use shifts in the main loop.
8020 read_vec_element(s, tcg_rn, rn, is_q ? 1 : 0, MO_64);
8022 for (i = 0; i < elements; i++) {
8023 tcg_gen_shri_i64(tcg_rd, tcg_rn, i * esize);
8024 ext_and_shift_reg(tcg_rd, tcg_rd, size | (!is_u << 2), 0);
8025 tcg_gen_shli_i64(tcg_rd, tcg_rd, shift);
8026 write_vec_element(s, tcg_rd, rd, i, size + 1);
8030 /* SHRN/RSHRN - Shift right with narrowing (and potential rounding) */
8031 static void handle_vec_simd_shrn(DisasContext *s, bool is_q,
8032 int immh, int immb, int opcode, int rn, int rd)
8034 int immhb = immh << 3 | immb;
8035 int size = 32 - clz32(immh) - 1;
8037 int esize = 8 << size;
8038 int elements = dsize/esize;
8039 int shift = (2 * esize) - immhb;
8040 bool round = extract32(opcode, 0, 1);
8041 TCGv_i64 tcg_rn, tcg_rd, tcg_final;
8045 if (extract32(immh, 3, 1)) {
8046 unallocated_encoding(s);
8050 if (!fp_access_check(s)) {
8054 tcg_rn = tcg_temp_new_i64();
8055 tcg_rd = tcg_temp_new_i64();
8056 tcg_final = tcg_temp_new_i64();
8057 read_vec_element(s, tcg_final, rd, is_q ? 1 : 0, MO_64);
8060 uint64_t round_const = 1ULL << (shift - 1);
8061 tcg_round = tcg_const_i64(round_const);
8063 TCGV_UNUSED_I64(tcg_round);
8066 for (i = 0; i < elements; i++) {
8067 read_vec_element(s, tcg_rn, rn, i, size+1);
8068 handle_shri_with_rndacc(tcg_rd, tcg_rn, tcg_round,
8069 false, true, size+1, shift);
8071 tcg_gen_deposit_i64(tcg_final, tcg_final, tcg_rd, esize * i, esize);
8075 clear_vec_high(s, rd);
8076 write_vec_element(s, tcg_final, rd, 0, MO_64);
8078 write_vec_element(s, tcg_final, rd, 1, MO_64);
8082 tcg_temp_free_i64(tcg_round);
8084 tcg_temp_free_i64(tcg_rn);
8085 tcg_temp_free_i64(tcg_rd);
8086 tcg_temp_free_i64(tcg_final);
8091 /* C3.6.14 AdvSIMD shift by immediate
8092 * 31 30 29 28 23 22 19 18 16 15 11 10 9 5 4 0
8093 * +---+---+---+-------------+------+------+--------+---+------+------+
8094 * | 0 | Q | U | 0 1 1 1 1 0 | immh | immb | opcode | 1 | Rn | Rd |
8095 * +---+---+---+-------------+------+------+--------+---+------+------+
8097 static void disas_simd_shift_imm(DisasContext *s, uint32_t insn)
8099 int rd = extract32(insn, 0, 5);
8100 int rn = extract32(insn, 5, 5);
8101 int opcode = extract32(insn, 11, 5);
8102 int immb = extract32(insn, 16, 3);
8103 int immh = extract32(insn, 19, 4);
8104 bool is_u = extract32(insn, 29, 1);
8105 bool is_q = extract32(insn, 30, 1);
8108 case 0x08: /* SRI */
8110 unallocated_encoding(s);
8114 case 0x00: /* SSHR / USHR */
8115 case 0x02: /* SSRA / USRA (accumulate) */
8116 case 0x04: /* SRSHR / URSHR (rounding) */
8117 case 0x06: /* SRSRA / URSRA (accum + rounding) */
8118 handle_vec_simd_shri(s, is_q, is_u, immh, immb, opcode, rn, rd);
8120 case 0x0a: /* SHL / SLI */
8121 handle_vec_simd_shli(s, is_q, is_u, immh, immb, opcode, rn, rd);
8123 case 0x10: /* SHRN */
8124 case 0x11: /* RSHRN / SQRSHRUN */
8126 handle_vec_simd_sqshrn(s, false, is_q, false, true, immh, immb,
8129 handle_vec_simd_shrn(s, is_q, immh, immb, opcode, rn, rd);
8132 case 0x12: /* SQSHRN / UQSHRN */
8133 case 0x13: /* SQRSHRN / UQRSHRN */
8134 handle_vec_simd_sqshrn(s, false, is_q, is_u, is_u, immh, immb,
8137 case 0x14: /* SSHLL / USHLL */
8138 handle_vec_simd_wshli(s, is_q, is_u, immh, immb, opcode, rn, rd);
8140 case 0x1c: /* SCVTF / UCVTF */
8141 handle_simd_shift_intfp_conv(s, false, is_q, is_u, immh, immb,
8144 case 0xc: /* SQSHLU */
8146 unallocated_encoding(s);
8149 handle_simd_qshl(s, false, is_q, false, true, immh, immb, rn, rd);
8151 case 0xe: /* SQSHL, UQSHL */
8152 handle_simd_qshl(s, false, is_q, is_u, is_u, immh, immb, rn, rd);
8154 case 0x1f: /* FCVTZS/ FCVTZU */
8155 handle_simd_shift_fpint_conv(s, false, is_q, is_u, immh, immb, rn, rd);
8158 unallocated_encoding(s);
8163 /* Generate code to do a "long" addition or subtraction, ie one done in
8164 * TCGv_i64 on vector lanes twice the width specified by size.
8166 static void gen_neon_addl(int size, bool is_sub, TCGv_i64 tcg_res,
8167 TCGv_i64 tcg_op1, TCGv_i64 tcg_op2)
8169 static NeonGenTwo64OpFn * const fns[3][2] = {
8170 { gen_helper_neon_addl_u16, gen_helper_neon_subl_u16 },
8171 { gen_helper_neon_addl_u32, gen_helper_neon_subl_u32 },
8172 { tcg_gen_add_i64, tcg_gen_sub_i64 },
8174 NeonGenTwo64OpFn *genfn;
8177 genfn = fns[size][is_sub];
8178 genfn(tcg_res, tcg_op1, tcg_op2);
8181 static void handle_3rd_widening(DisasContext *s, int is_q, int is_u, int size,
8182 int opcode, int rd, int rn, int rm)
8184 /* 3-reg-different widening insns: 64 x 64 -> 128 */
8185 TCGv_i64 tcg_res[2];
8188 tcg_res[0] = tcg_temp_new_i64();
8189 tcg_res[1] = tcg_temp_new_i64();
8191 /* Does this op do an adding accumulate, a subtracting accumulate,
8192 * or no accumulate at all?
8210 read_vec_element(s, tcg_res[0], rd, 0, MO_64);
8211 read_vec_element(s, tcg_res[1], rd, 1, MO_64);
8214 /* size == 2 means two 32x32->64 operations; this is worth special
8215 * casing because we can generally handle it inline.
8218 for (pass = 0; pass < 2; pass++) {
8219 TCGv_i64 tcg_op1 = tcg_temp_new_i64();
8220 TCGv_i64 tcg_op2 = tcg_temp_new_i64();
8221 TCGv_i64 tcg_passres;
8222 TCGMemOp memop = MO_32 | (is_u ? 0 : MO_SIGN);
8224 int elt = pass + is_q * 2;
8226 read_vec_element(s, tcg_op1, rn, elt, memop);
8227 read_vec_element(s, tcg_op2, rm, elt, memop);
8230 tcg_passres = tcg_res[pass];
8232 tcg_passres = tcg_temp_new_i64();
8236 case 0: /* SADDL, SADDL2, UADDL, UADDL2 */
8237 tcg_gen_add_i64(tcg_passres, tcg_op1, tcg_op2);
8239 case 2: /* SSUBL, SSUBL2, USUBL, USUBL2 */
8240 tcg_gen_sub_i64(tcg_passres, tcg_op1, tcg_op2);
8242 case 5: /* SABAL, SABAL2, UABAL, UABAL2 */
8243 case 7: /* SABDL, SABDL2, UABDL, UABDL2 */
8245 TCGv_i64 tcg_tmp1 = tcg_temp_new_i64();
8246 TCGv_i64 tcg_tmp2 = tcg_temp_new_i64();
8248 tcg_gen_sub_i64(tcg_tmp1, tcg_op1, tcg_op2);
8249 tcg_gen_sub_i64(tcg_tmp2, tcg_op2, tcg_op1);
8250 tcg_gen_movcond_i64(is_u ? TCG_COND_GEU : TCG_COND_GE,
8252 tcg_op1, tcg_op2, tcg_tmp1, tcg_tmp2);
8253 tcg_temp_free_i64(tcg_tmp1);
8254 tcg_temp_free_i64(tcg_tmp2);
8257 case 8: /* SMLAL, SMLAL2, UMLAL, UMLAL2 */
8258 case 10: /* SMLSL, SMLSL2, UMLSL, UMLSL2 */
8259 case 12: /* UMULL, UMULL2, SMULL, SMULL2 */
8260 tcg_gen_mul_i64(tcg_passres, tcg_op1, tcg_op2);
8262 case 9: /* SQDMLAL, SQDMLAL2 */
8263 case 11: /* SQDMLSL, SQDMLSL2 */
8264 case 13: /* SQDMULL, SQDMULL2 */
8265 tcg_gen_mul_i64(tcg_passres, tcg_op1, tcg_op2);
8266 gen_helper_neon_addl_saturate_s64(tcg_passres, cpu_env,
8267 tcg_passres, tcg_passres);
8270 g_assert_not_reached();
8273 if (opcode == 9 || opcode == 11) {
8274 /* saturating accumulate ops */
8276 tcg_gen_neg_i64(tcg_passres, tcg_passres);
8278 gen_helper_neon_addl_saturate_s64(tcg_res[pass], cpu_env,
8279 tcg_res[pass], tcg_passres);
8280 } else if (accop > 0) {
8281 tcg_gen_add_i64(tcg_res[pass], tcg_res[pass], tcg_passres);
8282 } else if (accop < 0) {
8283 tcg_gen_sub_i64(tcg_res[pass], tcg_res[pass], tcg_passres);
8287 tcg_temp_free_i64(tcg_passres);
8290 tcg_temp_free_i64(tcg_op1);
8291 tcg_temp_free_i64(tcg_op2);
8294 /* size 0 or 1, generally helper functions */
8295 for (pass = 0; pass < 2; pass++) {
8296 TCGv_i32 tcg_op1 = tcg_temp_new_i32();
8297 TCGv_i32 tcg_op2 = tcg_temp_new_i32();
8298 TCGv_i64 tcg_passres;
8299 int elt = pass + is_q * 2;
8301 read_vec_element_i32(s, tcg_op1, rn, elt, MO_32);
8302 read_vec_element_i32(s, tcg_op2, rm, elt, MO_32);
8305 tcg_passres = tcg_res[pass];
8307 tcg_passres = tcg_temp_new_i64();
8311 case 0: /* SADDL, SADDL2, UADDL, UADDL2 */
8312 case 2: /* SSUBL, SSUBL2, USUBL, USUBL2 */
8314 TCGv_i64 tcg_op2_64 = tcg_temp_new_i64();
8315 static NeonGenWidenFn * const widenfns[2][2] = {
8316 { gen_helper_neon_widen_s8, gen_helper_neon_widen_u8 },
8317 { gen_helper_neon_widen_s16, gen_helper_neon_widen_u16 },
8319 NeonGenWidenFn *widenfn = widenfns[size][is_u];
8321 widenfn(tcg_op2_64, tcg_op2);
8322 widenfn(tcg_passres, tcg_op1);
8323 gen_neon_addl(size, (opcode == 2), tcg_passres,
8324 tcg_passres, tcg_op2_64);
8325 tcg_temp_free_i64(tcg_op2_64);
8328 case 5: /* SABAL, SABAL2, UABAL, UABAL2 */
8329 case 7: /* SABDL, SABDL2, UABDL, UABDL2 */
8332 gen_helper_neon_abdl_u16(tcg_passres, tcg_op1, tcg_op2);
8334 gen_helper_neon_abdl_s16(tcg_passres, tcg_op1, tcg_op2);
8338 gen_helper_neon_abdl_u32(tcg_passres, tcg_op1, tcg_op2);
8340 gen_helper_neon_abdl_s32(tcg_passres, tcg_op1, tcg_op2);
8344 case 8: /* SMLAL, SMLAL2, UMLAL, UMLAL2 */
8345 case 10: /* SMLSL, SMLSL2, UMLSL, UMLSL2 */
8346 case 12: /* UMULL, UMULL2, SMULL, SMULL2 */
8349 gen_helper_neon_mull_u8(tcg_passres, tcg_op1, tcg_op2);
8351 gen_helper_neon_mull_s8(tcg_passres, tcg_op1, tcg_op2);
8355 gen_helper_neon_mull_u16(tcg_passres, tcg_op1, tcg_op2);
8357 gen_helper_neon_mull_s16(tcg_passres, tcg_op1, tcg_op2);
8361 case 9: /* SQDMLAL, SQDMLAL2 */
8362 case 11: /* SQDMLSL, SQDMLSL2 */
8363 case 13: /* SQDMULL, SQDMULL2 */
8365 gen_helper_neon_mull_s16(tcg_passres, tcg_op1, tcg_op2);
8366 gen_helper_neon_addl_saturate_s32(tcg_passres, cpu_env,
8367 tcg_passres, tcg_passres);
8369 case 14: /* PMULL */
8371 gen_helper_neon_mull_p8(tcg_passres, tcg_op1, tcg_op2);
8374 g_assert_not_reached();
8376 tcg_temp_free_i32(tcg_op1);
8377 tcg_temp_free_i32(tcg_op2);
8380 if (opcode == 9 || opcode == 11) {
8381 /* saturating accumulate ops */
8383 gen_helper_neon_negl_u32(tcg_passres, tcg_passres);
8385 gen_helper_neon_addl_saturate_s32(tcg_res[pass], cpu_env,
8389 gen_neon_addl(size, (accop < 0), tcg_res[pass],
8390 tcg_res[pass], tcg_passres);
8392 tcg_temp_free_i64(tcg_passres);
8397 write_vec_element(s, tcg_res[0], rd, 0, MO_64);
8398 write_vec_element(s, tcg_res[1], rd, 1, MO_64);
8399 tcg_temp_free_i64(tcg_res[0]);
8400 tcg_temp_free_i64(tcg_res[1]);
8403 static void handle_3rd_wide(DisasContext *s, int is_q, int is_u, int size,
8404 int opcode, int rd, int rn, int rm)
8406 TCGv_i64 tcg_res[2];
8407 int part = is_q ? 2 : 0;
8410 for (pass = 0; pass < 2; pass++) {
8411 TCGv_i64 tcg_op1 = tcg_temp_new_i64();
8412 TCGv_i32 tcg_op2 = tcg_temp_new_i32();
8413 TCGv_i64 tcg_op2_wide = tcg_temp_new_i64();
8414 static NeonGenWidenFn * const widenfns[3][2] = {
8415 { gen_helper_neon_widen_s8, gen_helper_neon_widen_u8 },
8416 { gen_helper_neon_widen_s16, gen_helper_neon_widen_u16 },
8417 { tcg_gen_ext_i32_i64, tcg_gen_extu_i32_i64 },
8419 NeonGenWidenFn *widenfn = widenfns[size][is_u];
8421 read_vec_element(s, tcg_op1, rn, pass, MO_64);
8422 read_vec_element_i32(s, tcg_op2, rm, part + pass, MO_32);
8423 widenfn(tcg_op2_wide, tcg_op2);
8424 tcg_temp_free_i32(tcg_op2);
8425 tcg_res[pass] = tcg_temp_new_i64();
8426 gen_neon_addl(size, (opcode == 3),
8427 tcg_res[pass], tcg_op1, tcg_op2_wide);
8428 tcg_temp_free_i64(tcg_op1);
8429 tcg_temp_free_i64(tcg_op2_wide);
8432 for (pass = 0; pass < 2; pass++) {
8433 write_vec_element(s, tcg_res[pass], rd, pass, MO_64);
8434 tcg_temp_free_i64(tcg_res[pass]);
8438 static void do_narrow_high_u32(TCGv_i32 res, TCGv_i64 in)
8440 tcg_gen_shri_i64(in, in, 32);
8441 tcg_gen_trunc_i64_i32(res, in);
8444 static void do_narrow_round_high_u32(TCGv_i32 res, TCGv_i64 in)
8446 tcg_gen_addi_i64(in, in, 1U << 31);
8447 do_narrow_high_u32(res, in);
8450 static void handle_3rd_narrowing(DisasContext *s, int is_q, int is_u, int size,
8451 int opcode, int rd, int rn, int rm)
8453 TCGv_i32 tcg_res[2];
8454 int part = is_q ? 2 : 0;
8457 for (pass = 0; pass < 2; pass++) {
8458 TCGv_i64 tcg_op1 = tcg_temp_new_i64();
8459 TCGv_i64 tcg_op2 = tcg_temp_new_i64();
8460 TCGv_i64 tcg_wideres = tcg_temp_new_i64();
8461 static NeonGenNarrowFn * const narrowfns[3][2] = {
8462 { gen_helper_neon_narrow_high_u8,
8463 gen_helper_neon_narrow_round_high_u8 },
8464 { gen_helper_neon_narrow_high_u16,
8465 gen_helper_neon_narrow_round_high_u16 },
8466 { do_narrow_high_u32, do_narrow_round_high_u32 },
8468 NeonGenNarrowFn *gennarrow = narrowfns[size][is_u];
8470 read_vec_element(s, tcg_op1, rn, pass, MO_64);
8471 read_vec_element(s, tcg_op2, rm, pass, MO_64);
8473 gen_neon_addl(size, (opcode == 6), tcg_wideres, tcg_op1, tcg_op2);
8475 tcg_temp_free_i64(tcg_op1);
8476 tcg_temp_free_i64(tcg_op2);
8478 tcg_res[pass] = tcg_temp_new_i32();
8479 gennarrow(tcg_res[pass], tcg_wideres);
8480 tcg_temp_free_i64(tcg_wideres);
8483 for (pass = 0; pass < 2; pass++) {
8484 write_vec_element_i32(s, tcg_res[pass], rd, pass + part, MO_32);
8485 tcg_temp_free_i32(tcg_res[pass]);
8488 clear_vec_high(s, rd);
8492 static void handle_pmull_64(DisasContext *s, int is_q, int rd, int rn, int rm)
8494 /* PMULL of 64 x 64 -> 128 is an odd special case because it
8495 * is the only three-reg-diff instruction which produces a
8496 * 128-bit wide result from a single operation. However since
8497 * it's possible to calculate the two halves more or less
8498 * separately we just use two helper calls.
8500 TCGv_i64 tcg_op1 = tcg_temp_new_i64();
8501 TCGv_i64 tcg_op2 = tcg_temp_new_i64();
8502 TCGv_i64 tcg_res = tcg_temp_new_i64();
8504 read_vec_element(s, tcg_op1, rn, is_q, MO_64);
8505 read_vec_element(s, tcg_op2, rm, is_q, MO_64);
8506 gen_helper_neon_pmull_64_lo(tcg_res, tcg_op1, tcg_op2);
8507 write_vec_element(s, tcg_res, rd, 0, MO_64);
8508 gen_helper_neon_pmull_64_hi(tcg_res, tcg_op1, tcg_op2);
8509 write_vec_element(s, tcg_res, rd, 1, MO_64);
8511 tcg_temp_free_i64(tcg_op1);
8512 tcg_temp_free_i64(tcg_op2);
8513 tcg_temp_free_i64(tcg_res);
8516 /* C3.6.15 AdvSIMD three different
8517 * 31 30 29 28 24 23 22 21 20 16 15 12 11 10 9 5 4 0
8518 * +---+---+---+-----------+------+---+------+--------+-----+------+------+
8519 * | 0 | Q | U | 0 1 1 1 0 | size | 1 | Rm | opcode | 0 0 | Rn | Rd |
8520 * +---+---+---+-----------+------+---+------+--------+-----+------+------+
8522 static void disas_simd_three_reg_diff(DisasContext *s, uint32_t insn)
8524 /* Instructions in this group fall into three basic classes
8525 * (in each case with the operation working on each element in
8526 * the input vectors):
8527 * (1) widening 64 x 64 -> 128 (with possibly Vd as an extra
8529 * (2) wide 64 x 128 -> 128
8530 * (3) narrowing 128 x 128 -> 64
8531 * Here we do initial decode, catch unallocated cases and
8532 * dispatch to separate functions for each class.
8534 int is_q = extract32(insn, 30, 1);
8535 int is_u = extract32(insn, 29, 1);
8536 int size = extract32(insn, 22, 2);
8537 int opcode = extract32(insn, 12, 4);
8538 int rm = extract32(insn, 16, 5);
8539 int rn = extract32(insn, 5, 5);
8540 int rd = extract32(insn, 0, 5);
8543 case 1: /* SADDW, SADDW2, UADDW, UADDW2 */
8544 case 3: /* SSUBW, SSUBW2, USUBW, USUBW2 */
8545 /* 64 x 128 -> 128 */
8547 unallocated_encoding(s);
8550 if (!fp_access_check(s)) {
8553 handle_3rd_wide(s, is_q, is_u, size, opcode, rd, rn, rm);
8555 case 4: /* ADDHN, ADDHN2, RADDHN, RADDHN2 */
8556 case 6: /* SUBHN, SUBHN2, RSUBHN, RSUBHN2 */
8557 /* 128 x 128 -> 64 */
8559 unallocated_encoding(s);
8562 if (!fp_access_check(s)) {
8565 handle_3rd_narrowing(s, is_q, is_u, size, opcode, rd, rn, rm);
8567 case 14: /* PMULL, PMULL2 */
8568 if (is_u || size == 1 || size == 2) {
8569 unallocated_encoding(s);
8573 if (!arm_dc_feature(s, ARM_FEATURE_V8_AES)) {
8574 unallocated_encoding(s);
8577 if (!fp_access_check(s)) {
8580 handle_pmull_64(s, is_q, rd, rn, rm);
8584 case 9: /* SQDMLAL, SQDMLAL2 */
8585 case 11: /* SQDMLSL, SQDMLSL2 */
8586 case 13: /* SQDMULL, SQDMULL2 */
8587 if (is_u || size == 0) {
8588 unallocated_encoding(s);
8592 case 0: /* SADDL, SADDL2, UADDL, UADDL2 */
8593 case 2: /* SSUBL, SSUBL2, USUBL, USUBL2 */
8594 case 5: /* SABAL, SABAL2, UABAL, UABAL2 */
8595 case 7: /* SABDL, SABDL2, UABDL, UABDL2 */
8596 case 8: /* SMLAL, SMLAL2, UMLAL, UMLAL2 */
8597 case 10: /* SMLSL, SMLSL2, UMLSL, UMLSL2 */
8598 case 12: /* SMULL, SMULL2, UMULL, UMULL2 */
8599 /* 64 x 64 -> 128 */
8601 unallocated_encoding(s);
8605 if (!fp_access_check(s)) {
8609 handle_3rd_widening(s, is_q, is_u, size, opcode, rd, rn, rm);
8612 /* opcode 15 not allocated */
8613 unallocated_encoding(s);
8618 /* Logic op (opcode == 3) subgroup of C3.6.16. */
8619 static void disas_simd_3same_logic(DisasContext *s, uint32_t insn)
8621 int rd = extract32(insn, 0, 5);
8622 int rn = extract32(insn, 5, 5);
8623 int rm = extract32(insn, 16, 5);
8624 int size = extract32(insn, 22, 2);
8625 bool is_u = extract32(insn, 29, 1);
8626 bool is_q = extract32(insn, 30, 1);
8627 TCGv_i64 tcg_op1, tcg_op2, tcg_res[2];
8630 if (!fp_access_check(s)) {
8634 tcg_op1 = tcg_temp_new_i64();
8635 tcg_op2 = tcg_temp_new_i64();
8636 tcg_res[0] = tcg_temp_new_i64();
8637 tcg_res[1] = tcg_temp_new_i64();
8639 for (pass = 0; pass < (is_q ? 2 : 1); pass++) {
8640 read_vec_element(s, tcg_op1, rn, pass, MO_64);
8641 read_vec_element(s, tcg_op2, rm, pass, MO_64);
8646 tcg_gen_and_i64(tcg_res[pass], tcg_op1, tcg_op2);
8649 tcg_gen_andc_i64(tcg_res[pass], tcg_op1, tcg_op2);
8652 tcg_gen_or_i64(tcg_res[pass], tcg_op1, tcg_op2);
8655 tcg_gen_orc_i64(tcg_res[pass], tcg_op1, tcg_op2);
8660 /* B* ops need res loaded to operate on */
8661 read_vec_element(s, tcg_res[pass], rd, pass, MO_64);
8666 tcg_gen_xor_i64(tcg_res[pass], tcg_op1, tcg_op2);
8668 case 1: /* BSL bitwise select */
8669 tcg_gen_xor_i64(tcg_op1, tcg_op1, tcg_op2);
8670 tcg_gen_and_i64(tcg_op1, tcg_op1, tcg_res[pass]);
8671 tcg_gen_xor_i64(tcg_res[pass], tcg_op2, tcg_op1);
8673 case 2: /* BIT, bitwise insert if true */
8674 tcg_gen_xor_i64(tcg_op1, tcg_op1, tcg_res[pass]);
8675 tcg_gen_and_i64(tcg_op1, tcg_op1, tcg_op2);
8676 tcg_gen_xor_i64(tcg_res[pass], tcg_res[pass], tcg_op1);
8678 case 3: /* BIF, bitwise insert if false */
8679 tcg_gen_xor_i64(tcg_op1, tcg_op1, tcg_res[pass]);
8680 tcg_gen_andc_i64(tcg_op1, tcg_op1, tcg_op2);
8681 tcg_gen_xor_i64(tcg_res[pass], tcg_res[pass], tcg_op1);
8687 write_vec_element(s, tcg_res[0], rd, 0, MO_64);
8689 tcg_gen_movi_i64(tcg_res[1], 0);
8691 write_vec_element(s, tcg_res[1], rd, 1, MO_64);
8693 tcg_temp_free_i64(tcg_op1);
8694 tcg_temp_free_i64(tcg_op2);
8695 tcg_temp_free_i64(tcg_res[0]);
8696 tcg_temp_free_i64(tcg_res[1]);
8699 /* Helper functions for 32 bit comparisons */
8700 static void gen_max_s32(TCGv_i32 res, TCGv_i32 op1, TCGv_i32 op2)
8702 tcg_gen_movcond_i32(TCG_COND_GE, res, op1, op2, op1, op2);
8705 static void gen_max_u32(TCGv_i32 res, TCGv_i32 op1, TCGv_i32 op2)
8707 tcg_gen_movcond_i32(TCG_COND_GEU, res, op1, op2, op1, op2);
8710 static void gen_min_s32(TCGv_i32 res, TCGv_i32 op1, TCGv_i32 op2)
8712 tcg_gen_movcond_i32(TCG_COND_LE, res, op1, op2, op1, op2);
8715 static void gen_min_u32(TCGv_i32 res, TCGv_i32 op1, TCGv_i32 op2)
8717 tcg_gen_movcond_i32(TCG_COND_LEU, res, op1, op2, op1, op2);
8720 /* Pairwise op subgroup of C3.6.16.
8722 * This is called directly or via the handle_3same_float for float pairwise
8723 * operations where the opcode and size are calculated differently.
8725 static void handle_simd_3same_pair(DisasContext *s, int is_q, int u, int opcode,
8726 int size, int rn, int rm, int rd)
8731 /* Floating point operations need fpst */
8732 if (opcode >= 0x58) {
8733 fpst = get_fpstatus_ptr();
8735 TCGV_UNUSED_PTR(fpst);
8738 if (!fp_access_check(s)) {
8742 /* These operations work on the concatenated rm:rn, with each pair of
8743 * adjacent elements being operated on to produce an element in the result.
8746 TCGv_i64 tcg_res[2];
8748 for (pass = 0; pass < 2; pass++) {
8749 TCGv_i64 tcg_op1 = tcg_temp_new_i64();
8750 TCGv_i64 tcg_op2 = tcg_temp_new_i64();
8751 int passreg = (pass == 0) ? rn : rm;
8753 read_vec_element(s, tcg_op1, passreg, 0, MO_64);
8754 read_vec_element(s, tcg_op2, passreg, 1, MO_64);
8755 tcg_res[pass] = tcg_temp_new_i64();
8758 case 0x17: /* ADDP */
8759 tcg_gen_add_i64(tcg_res[pass], tcg_op1, tcg_op2);
8761 case 0x58: /* FMAXNMP */
8762 gen_helper_vfp_maxnumd(tcg_res[pass], tcg_op1, tcg_op2, fpst);
8764 case 0x5a: /* FADDP */
8765 gen_helper_vfp_addd(tcg_res[pass], tcg_op1, tcg_op2, fpst);
8767 case 0x5e: /* FMAXP */
8768 gen_helper_vfp_maxd(tcg_res[pass], tcg_op1, tcg_op2, fpst);
8770 case 0x78: /* FMINNMP */
8771 gen_helper_vfp_minnumd(tcg_res[pass], tcg_op1, tcg_op2, fpst);
8773 case 0x7e: /* FMINP */
8774 gen_helper_vfp_mind(tcg_res[pass], tcg_op1, tcg_op2, fpst);
8777 g_assert_not_reached();
8780 tcg_temp_free_i64(tcg_op1);
8781 tcg_temp_free_i64(tcg_op2);
8784 for (pass = 0; pass < 2; pass++) {
8785 write_vec_element(s, tcg_res[pass], rd, pass, MO_64);
8786 tcg_temp_free_i64(tcg_res[pass]);
8789 int maxpass = is_q ? 4 : 2;
8790 TCGv_i32 tcg_res[4];
8792 for (pass = 0; pass < maxpass; pass++) {
8793 TCGv_i32 tcg_op1 = tcg_temp_new_i32();
8794 TCGv_i32 tcg_op2 = tcg_temp_new_i32();
8795 NeonGenTwoOpFn *genfn = NULL;
8796 int passreg = pass < (maxpass / 2) ? rn : rm;
8797 int passelt = (is_q && (pass & 1)) ? 2 : 0;
8799 read_vec_element_i32(s, tcg_op1, passreg, passelt, MO_32);
8800 read_vec_element_i32(s, tcg_op2, passreg, passelt + 1, MO_32);
8801 tcg_res[pass] = tcg_temp_new_i32();
8804 case 0x17: /* ADDP */
8806 static NeonGenTwoOpFn * const fns[3] = {
8807 gen_helper_neon_padd_u8,
8808 gen_helper_neon_padd_u16,
8814 case 0x14: /* SMAXP, UMAXP */
8816 static NeonGenTwoOpFn * const fns[3][2] = {
8817 { gen_helper_neon_pmax_s8, gen_helper_neon_pmax_u8 },
8818 { gen_helper_neon_pmax_s16, gen_helper_neon_pmax_u16 },
8819 { gen_max_s32, gen_max_u32 },
8821 genfn = fns[size][u];
8824 case 0x15: /* SMINP, UMINP */
8826 static NeonGenTwoOpFn * const fns[3][2] = {
8827 { gen_helper_neon_pmin_s8, gen_helper_neon_pmin_u8 },
8828 { gen_helper_neon_pmin_s16, gen_helper_neon_pmin_u16 },
8829 { gen_min_s32, gen_min_u32 },
8831 genfn = fns[size][u];
8834 /* The FP operations are all on single floats (32 bit) */
8835 case 0x58: /* FMAXNMP */
8836 gen_helper_vfp_maxnums(tcg_res[pass], tcg_op1, tcg_op2, fpst);
8838 case 0x5a: /* FADDP */
8839 gen_helper_vfp_adds(tcg_res[pass], tcg_op1, tcg_op2, fpst);
8841 case 0x5e: /* FMAXP */
8842 gen_helper_vfp_maxs(tcg_res[pass], tcg_op1, tcg_op2, fpst);
8844 case 0x78: /* FMINNMP */
8845 gen_helper_vfp_minnums(tcg_res[pass], tcg_op1, tcg_op2, fpst);
8847 case 0x7e: /* FMINP */
8848 gen_helper_vfp_mins(tcg_res[pass], tcg_op1, tcg_op2, fpst);
8851 g_assert_not_reached();
8854 /* FP ops called directly, otherwise call now */
8856 genfn(tcg_res[pass], tcg_op1, tcg_op2);
8859 tcg_temp_free_i32(tcg_op1);
8860 tcg_temp_free_i32(tcg_op2);
8863 for (pass = 0; pass < maxpass; pass++) {
8864 write_vec_element_i32(s, tcg_res[pass], rd, pass, MO_32);
8865 tcg_temp_free_i32(tcg_res[pass]);
8868 clear_vec_high(s, rd);
8872 if (!TCGV_IS_UNUSED_PTR(fpst)) {
8873 tcg_temp_free_ptr(fpst);
8877 /* Floating point op subgroup of C3.6.16. */
8878 static void disas_simd_3same_float(DisasContext *s, uint32_t insn)
8880 /* For floating point ops, the U, size[1] and opcode bits
8881 * together indicate the operation. size[0] indicates single
8884 int fpopcode = extract32(insn, 11, 5)
8885 | (extract32(insn, 23, 1) << 5)
8886 | (extract32(insn, 29, 1) << 6);
8887 int is_q = extract32(insn, 30, 1);
8888 int size = extract32(insn, 22, 1);
8889 int rm = extract32(insn, 16, 5);
8890 int rn = extract32(insn, 5, 5);
8891 int rd = extract32(insn, 0, 5);
8893 int datasize = is_q ? 128 : 64;
8894 int esize = 32 << size;
8895 int elements = datasize / esize;
8897 if (size == 1 && !is_q) {
8898 unallocated_encoding(s);
8903 case 0x58: /* FMAXNMP */
8904 case 0x5a: /* FADDP */
8905 case 0x5e: /* FMAXP */
8906 case 0x78: /* FMINNMP */
8907 case 0x7e: /* FMINP */
8908 if (size && !is_q) {
8909 unallocated_encoding(s);
8912 handle_simd_3same_pair(s, is_q, 0, fpopcode, size ? MO_64 : MO_32,
8915 case 0x1b: /* FMULX */
8916 case 0x1f: /* FRECPS */
8917 case 0x3f: /* FRSQRTS */
8918 case 0x5d: /* FACGE */
8919 case 0x7d: /* FACGT */
8920 case 0x19: /* FMLA */
8921 case 0x39: /* FMLS */
8922 case 0x18: /* FMAXNM */
8923 case 0x1a: /* FADD */
8924 case 0x1c: /* FCMEQ */
8925 case 0x1e: /* FMAX */
8926 case 0x38: /* FMINNM */
8927 case 0x3a: /* FSUB */
8928 case 0x3e: /* FMIN */
8929 case 0x5b: /* FMUL */
8930 case 0x5c: /* FCMGE */
8931 case 0x5f: /* FDIV */
8932 case 0x7a: /* FABD */
8933 case 0x7c: /* FCMGT */
8934 if (!fp_access_check(s)) {
8938 handle_3same_float(s, size, elements, fpopcode, rd, rn, rm);
8941 unallocated_encoding(s);
8946 /* Integer op subgroup of C3.6.16. */
8947 static void disas_simd_3same_int(DisasContext *s, uint32_t insn)
8949 int is_q = extract32(insn, 30, 1);
8950 int u = extract32(insn, 29, 1);
8951 int size = extract32(insn, 22, 2);
8952 int opcode = extract32(insn, 11, 5);
8953 int rm = extract32(insn, 16, 5);
8954 int rn = extract32(insn, 5, 5);
8955 int rd = extract32(insn, 0, 5);
8959 case 0x13: /* MUL, PMUL */
8960 if (u && size != 0) {
8961 unallocated_encoding(s);
8965 case 0x0: /* SHADD, UHADD */
8966 case 0x2: /* SRHADD, URHADD */
8967 case 0x4: /* SHSUB, UHSUB */
8968 case 0xc: /* SMAX, UMAX */
8969 case 0xd: /* SMIN, UMIN */
8970 case 0xe: /* SABD, UABD */
8971 case 0xf: /* SABA, UABA */
8972 case 0x12: /* MLA, MLS */
8974 unallocated_encoding(s);
8978 case 0x16: /* SQDMULH, SQRDMULH */
8979 if (size == 0 || size == 3) {
8980 unallocated_encoding(s);
8985 if (size == 3 && !is_q) {
8986 unallocated_encoding(s);
8992 if (!fp_access_check(s)) {
8997 for (pass = 0; pass < (is_q ? 2 : 1); pass++) {
8998 TCGv_i64 tcg_op1 = tcg_temp_new_i64();
8999 TCGv_i64 tcg_op2 = tcg_temp_new_i64();
9000 TCGv_i64 tcg_res = tcg_temp_new_i64();
9002 read_vec_element(s, tcg_op1, rn, pass, MO_64);
9003 read_vec_element(s, tcg_op2, rm, pass, MO_64);
9005 handle_3same_64(s, opcode, u, tcg_res, tcg_op1, tcg_op2);
9007 write_vec_element(s, tcg_res, rd, pass, MO_64);
9009 tcg_temp_free_i64(tcg_res);
9010 tcg_temp_free_i64(tcg_op1);
9011 tcg_temp_free_i64(tcg_op2);
9014 for (pass = 0; pass < (is_q ? 4 : 2); pass++) {
9015 TCGv_i32 tcg_op1 = tcg_temp_new_i32();
9016 TCGv_i32 tcg_op2 = tcg_temp_new_i32();
9017 TCGv_i32 tcg_res = tcg_temp_new_i32();
9018 NeonGenTwoOpFn *genfn = NULL;
9019 NeonGenTwoOpEnvFn *genenvfn = NULL;
9021 read_vec_element_i32(s, tcg_op1, rn, pass, MO_32);
9022 read_vec_element_i32(s, tcg_op2, rm, pass, MO_32);
9025 case 0x0: /* SHADD, UHADD */
9027 static NeonGenTwoOpFn * const fns[3][2] = {
9028 { gen_helper_neon_hadd_s8, gen_helper_neon_hadd_u8 },
9029 { gen_helper_neon_hadd_s16, gen_helper_neon_hadd_u16 },
9030 { gen_helper_neon_hadd_s32, gen_helper_neon_hadd_u32 },
9032 genfn = fns[size][u];
9035 case 0x1: /* SQADD, UQADD */
9037 static NeonGenTwoOpEnvFn * const fns[3][2] = {
9038 { gen_helper_neon_qadd_s8, gen_helper_neon_qadd_u8 },
9039 { gen_helper_neon_qadd_s16, gen_helper_neon_qadd_u16 },
9040 { gen_helper_neon_qadd_s32, gen_helper_neon_qadd_u32 },
9042 genenvfn = fns[size][u];
9045 case 0x2: /* SRHADD, URHADD */
9047 static NeonGenTwoOpFn * const fns[3][2] = {
9048 { gen_helper_neon_rhadd_s8, gen_helper_neon_rhadd_u8 },
9049 { gen_helper_neon_rhadd_s16, gen_helper_neon_rhadd_u16 },
9050 { gen_helper_neon_rhadd_s32, gen_helper_neon_rhadd_u32 },
9052 genfn = fns[size][u];
9055 case 0x4: /* SHSUB, UHSUB */
9057 static NeonGenTwoOpFn * const fns[3][2] = {
9058 { gen_helper_neon_hsub_s8, gen_helper_neon_hsub_u8 },
9059 { gen_helper_neon_hsub_s16, gen_helper_neon_hsub_u16 },
9060 { gen_helper_neon_hsub_s32, gen_helper_neon_hsub_u32 },
9062 genfn = fns[size][u];
9065 case 0x5: /* SQSUB, UQSUB */
9067 static NeonGenTwoOpEnvFn * const fns[3][2] = {
9068 { gen_helper_neon_qsub_s8, gen_helper_neon_qsub_u8 },
9069 { gen_helper_neon_qsub_s16, gen_helper_neon_qsub_u16 },
9070 { gen_helper_neon_qsub_s32, gen_helper_neon_qsub_u32 },
9072 genenvfn = fns[size][u];
9075 case 0x6: /* CMGT, CMHI */
9077 static NeonGenTwoOpFn * const fns[3][2] = {
9078 { gen_helper_neon_cgt_s8, gen_helper_neon_cgt_u8 },
9079 { gen_helper_neon_cgt_s16, gen_helper_neon_cgt_u16 },
9080 { gen_helper_neon_cgt_s32, gen_helper_neon_cgt_u32 },
9082 genfn = fns[size][u];
9085 case 0x7: /* CMGE, CMHS */
9087 static NeonGenTwoOpFn * const fns[3][2] = {
9088 { gen_helper_neon_cge_s8, gen_helper_neon_cge_u8 },
9089 { gen_helper_neon_cge_s16, gen_helper_neon_cge_u16 },
9090 { gen_helper_neon_cge_s32, gen_helper_neon_cge_u32 },
9092 genfn = fns[size][u];
9095 case 0x8: /* SSHL, USHL */
9097 static NeonGenTwoOpFn * const fns[3][2] = {
9098 { gen_helper_neon_shl_s8, gen_helper_neon_shl_u8 },
9099 { gen_helper_neon_shl_s16, gen_helper_neon_shl_u16 },
9100 { gen_helper_neon_shl_s32, gen_helper_neon_shl_u32 },
9102 genfn = fns[size][u];
9105 case 0x9: /* SQSHL, UQSHL */
9107 static NeonGenTwoOpEnvFn * const fns[3][2] = {
9108 { gen_helper_neon_qshl_s8, gen_helper_neon_qshl_u8 },
9109 { gen_helper_neon_qshl_s16, gen_helper_neon_qshl_u16 },
9110 { gen_helper_neon_qshl_s32, gen_helper_neon_qshl_u32 },
9112 genenvfn = fns[size][u];
9115 case 0xa: /* SRSHL, URSHL */
9117 static NeonGenTwoOpFn * const fns[3][2] = {
9118 { gen_helper_neon_rshl_s8, gen_helper_neon_rshl_u8 },
9119 { gen_helper_neon_rshl_s16, gen_helper_neon_rshl_u16 },
9120 { gen_helper_neon_rshl_s32, gen_helper_neon_rshl_u32 },
9122 genfn = fns[size][u];
9125 case 0xb: /* SQRSHL, UQRSHL */
9127 static NeonGenTwoOpEnvFn * const fns[3][2] = {
9128 { gen_helper_neon_qrshl_s8, gen_helper_neon_qrshl_u8 },
9129 { gen_helper_neon_qrshl_s16, gen_helper_neon_qrshl_u16 },
9130 { gen_helper_neon_qrshl_s32, gen_helper_neon_qrshl_u32 },
9132 genenvfn = fns[size][u];
9135 case 0xc: /* SMAX, UMAX */
9137 static NeonGenTwoOpFn * const fns[3][2] = {
9138 { gen_helper_neon_max_s8, gen_helper_neon_max_u8 },
9139 { gen_helper_neon_max_s16, gen_helper_neon_max_u16 },
9140 { gen_max_s32, gen_max_u32 },
9142 genfn = fns[size][u];
9146 case 0xd: /* SMIN, UMIN */
9148 static NeonGenTwoOpFn * const fns[3][2] = {
9149 { gen_helper_neon_min_s8, gen_helper_neon_min_u8 },
9150 { gen_helper_neon_min_s16, gen_helper_neon_min_u16 },
9151 { gen_min_s32, gen_min_u32 },
9153 genfn = fns[size][u];
9156 case 0xe: /* SABD, UABD */
9157 case 0xf: /* SABA, UABA */
9159 static NeonGenTwoOpFn * const fns[3][2] = {
9160 { gen_helper_neon_abd_s8, gen_helper_neon_abd_u8 },
9161 { gen_helper_neon_abd_s16, gen_helper_neon_abd_u16 },
9162 { gen_helper_neon_abd_s32, gen_helper_neon_abd_u32 },
9164 genfn = fns[size][u];
9167 case 0x10: /* ADD, SUB */
9169 static NeonGenTwoOpFn * const fns[3][2] = {
9170 { gen_helper_neon_add_u8, gen_helper_neon_sub_u8 },
9171 { gen_helper_neon_add_u16, gen_helper_neon_sub_u16 },
9172 { tcg_gen_add_i32, tcg_gen_sub_i32 },
9174 genfn = fns[size][u];
9177 case 0x11: /* CMTST, CMEQ */
9179 static NeonGenTwoOpFn * const fns[3][2] = {
9180 { gen_helper_neon_tst_u8, gen_helper_neon_ceq_u8 },
9181 { gen_helper_neon_tst_u16, gen_helper_neon_ceq_u16 },
9182 { gen_helper_neon_tst_u32, gen_helper_neon_ceq_u32 },
9184 genfn = fns[size][u];
9187 case 0x13: /* MUL, PMUL */
9191 genfn = gen_helper_neon_mul_p8;
9194 /* fall through : MUL */
9195 case 0x12: /* MLA, MLS */
9197 static NeonGenTwoOpFn * const fns[3] = {
9198 gen_helper_neon_mul_u8,
9199 gen_helper_neon_mul_u16,
9205 case 0x16: /* SQDMULH, SQRDMULH */
9207 static NeonGenTwoOpEnvFn * const fns[2][2] = {
9208 { gen_helper_neon_qdmulh_s16, gen_helper_neon_qrdmulh_s16 },
9209 { gen_helper_neon_qdmulh_s32, gen_helper_neon_qrdmulh_s32 },
9211 assert(size == 1 || size == 2);
9212 genenvfn = fns[size - 1][u];
9216 g_assert_not_reached();
9220 genenvfn(tcg_res, cpu_env, tcg_op1, tcg_op2);
9222 genfn(tcg_res, tcg_op1, tcg_op2);
9225 if (opcode == 0xf || opcode == 0x12) {
9226 /* SABA, UABA, MLA, MLS: accumulating ops */
9227 static NeonGenTwoOpFn * const fns[3][2] = {
9228 { gen_helper_neon_add_u8, gen_helper_neon_sub_u8 },
9229 { gen_helper_neon_add_u16, gen_helper_neon_sub_u16 },
9230 { tcg_gen_add_i32, tcg_gen_sub_i32 },
9232 bool is_sub = (opcode == 0x12 && u); /* MLS */
9234 genfn = fns[size][is_sub];
9235 read_vec_element_i32(s, tcg_op1, rd, pass, MO_32);
9236 genfn(tcg_res, tcg_op1, tcg_res);
9239 write_vec_element_i32(s, tcg_res, rd, pass, MO_32);
9241 tcg_temp_free_i32(tcg_res);
9242 tcg_temp_free_i32(tcg_op1);
9243 tcg_temp_free_i32(tcg_op2);
9248 clear_vec_high(s, rd);
9252 /* C3.6.16 AdvSIMD three same
9253 * 31 30 29 28 24 23 22 21 20 16 15 11 10 9 5 4 0
9254 * +---+---+---+-----------+------+---+------+--------+---+------+------+
9255 * | 0 | Q | U | 0 1 1 1 0 | size | 1 | Rm | opcode | 1 | Rn | Rd |
9256 * +---+---+---+-----------+------+---+------+--------+---+------+------+
9258 static void disas_simd_three_reg_same(DisasContext *s, uint32_t insn)
9260 int opcode = extract32(insn, 11, 5);
9263 case 0x3: /* logic ops */
9264 disas_simd_3same_logic(s, insn);
9266 case 0x17: /* ADDP */
9267 case 0x14: /* SMAXP, UMAXP */
9268 case 0x15: /* SMINP, UMINP */
9270 /* Pairwise operations */
9271 int is_q = extract32(insn, 30, 1);
9272 int u = extract32(insn, 29, 1);
9273 int size = extract32(insn, 22, 2);
9274 int rm = extract32(insn, 16, 5);
9275 int rn = extract32(insn, 5, 5);
9276 int rd = extract32(insn, 0, 5);
9277 if (opcode == 0x17) {
9278 if (u || (size == 3 && !is_q)) {
9279 unallocated_encoding(s);
9284 unallocated_encoding(s);
9288 handle_simd_3same_pair(s, is_q, u, opcode, size, rn, rm, rd);
9292 /* floating point ops, sz[1] and U are part of opcode */
9293 disas_simd_3same_float(s, insn);
9296 disas_simd_3same_int(s, insn);
9301 static void handle_2misc_widening(DisasContext *s, int opcode, bool is_q,
9302 int size, int rn, int rd)
9304 /* Handle 2-reg-misc ops which are widening (so each size element
9305 * in the source becomes a 2*size element in the destination.
9306 * The only instruction like this is FCVTL.
9311 /* 32 -> 64 bit fp conversion */
9312 TCGv_i64 tcg_res[2];
9313 int srcelt = is_q ? 2 : 0;
9315 for (pass = 0; pass < 2; pass++) {
9316 TCGv_i32 tcg_op = tcg_temp_new_i32();
9317 tcg_res[pass] = tcg_temp_new_i64();
9319 read_vec_element_i32(s, tcg_op, rn, srcelt + pass, MO_32);
9320 gen_helper_vfp_fcvtds(tcg_res[pass], tcg_op, cpu_env);
9321 tcg_temp_free_i32(tcg_op);
9323 for (pass = 0; pass < 2; pass++) {
9324 write_vec_element(s, tcg_res[pass], rd, pass, MO_64);
9325 tcg_temp_free_i64(tcg_res[pass]);
9328 /* 16 -> 32 bit fp conversion */
9329 int srcelt = is_q ? 4 : 0;
9330 TCGv_i32 tcg_res[4];
9332 for (pass = 0; pass < 4; pass++) {
9333 tcg_res[pass] = tcg_temp_new_i32();
9335 read_vec_element_i32(s, tcg_res[pass], rn, srcelt + pass, MO_16);
9336 gen_helper_vfp_fcvt_f16_to_f32(tcg_res[pass], tcg_res[pass],
9339 for (pass = 0; pass < 4; pass++) {
9340 write_vec_element_i32(s, tcg_res[pass], rd, pass, MO_32);
9341 tcg_temp_free_i32(tcg_res[pass]);
9346 static void handle_rev(DisasContext *s, int opcode, bool u,
9347 bool is_q, int size, int rn, int rd)
9349 int op = (opcode << 1) | u;
9350 int opsz = op + size;
9351 int grp_size = 3 - opsz;
9352 int dsize = is_q ? 128 : 64;
9356 unallocated_encoding(s);
9360 if (!fp_access_check(s)) {
9365 /* Special case bytes, use bswap op on each group of elements */
9366 int groups = dsize / (8 << grp_size);
9368 for (i = 0; i < groups; i++) {
9369 TCGv_i64 tcg_tmp = tcg_temp_new_i64();
9371 read_vec_element(s, tcg_tmp, rn, i, grp_size);
9374 tcg_gen_bswap16_i64(tcg_tmp, tcg_tmp);
9377 tcg_gen_bswap32_i64(tcg_tmp, tcg_tmp);
9380 tcg_gen_bswap64_i64(tcg_tmp, tcg_tmp);
9383 g_assert_not_reached();
9385 write_vec_element(s, tcg_tmp, rd, i, grp_size);
9386 tcg_temp_free_i64(tcg_tmp);
9389 clear_vec_high(s, rd);
9392 int revmask = (1 << grp_size) - 1;
9393 int esize = 8 << size;
9394 int elements = dsize / esize;
9395 TCGv_i64 tcg_rn = tcg_temp_new_i64();
9396 TCGv_i64 tcg_rd = tcg_const_i64(0);
9397 TCGv_i64 tcg_rd_hi = tcg_const_i64(0);
9399 for (i = 0; i < elements; i++) {
9400 int e_rev = (i & 0xf) ^ revmask;
9401 int off = e_rev * esize;
9402 read_vec_element(s, tcg_rn, rn, i, size);
9404 tcg_gen_deposit_i64(tcg_rd_hi, tcg_rd_hi,
9405 tcg_rn, off - 64, esize);
9407 tcg_gen_deposit_i64(tcg_rd, tcg_rd, tcg_rn, off, esize);
9410 write_vec_element(s, tcg_rd, rd, 0, MO_64);
9411 write_vec_element(s, tcg_rd_hi, rd, 1, MO_64);
9413 tcg_temp_free_i64(tcg_rd_hi);
9414 tcg_temp_free_i64(tcg_rd);
9415 tcg_temp_free_i64(tcg_rn);
9419 static void handle_2misc_pairwise(DisasContext *s, int opcode, bool u,
9420 bool is_q, int size, int rn, int rd)
9422 /* Implement the pairwise operations from 2-misc:
9423 * SADDLP, UADDLP, SADALP, UADALP.
9424 * These all add pairs of elements in the input to produce a
9425 * double-width result element in the output (possibly accumulating).
9427 bool accum = (opcode == 0x6);
9428 int maxpass = is_q ? 2 : 1;
9430 TCGv_i64 tcg_res[2];
9433 /* 32 + 32 -> 64 op */
9434 TCGMemOp memop = size + (u ? 0 : MO_SIGN);
9436 for (pass = 0; pass < maxpass; pass++) {
9437 TCGv_i64 tcg_op1 = tcg_temp_new_i64();
9438 TCGv_i64 tcg_op2 = tcg_temp_new_i64();
9440 tcg_res[pass] = tcg_temp_new_i64();
9442 read_vec_element(s, tcg_op1, rn, pass * 2, memop);
9443 read_vec_element(s, tcg_op2, rn, pass * 2 + 1, memop);
9444 tcg_gen_add_i64(tcg_res[pass], tcg_op1, tcg_op2);
9446 read_vec_element(s, tcg_op1, rd, pass, MO_64);
9447 tcg_gen_add_i64(tcg_res[pass], tcg_res[pass], tcg_op1);
9450 tcg_temp_free_i64(tcg_op1);
9451 tcg_temp_free_i64(tcg_op2);
9454 for (pass = 0; pass < maxpass; pass++) {
9455 TCGv_i64 tcg_op = tcg_temp_new_i64();
9456 NeonGenOneOpFn *genfn;
9457 static NeonGenOneOpFn * const fns[2][2] = {
9458 { gen_helper_neon_addlp_s8, gen_helper_neon_addlp_u8 },
9459 { gen_helper_neon_addlp_s16, gen_helper_neon_addlp_u16 },
9462 genfn = fns[size][u];
9464 tcg_res[pass] = tcg_temp_new_i64();
9466 read_vec_element(s, tcg_op, rn, pass, MO_64);
9467 genfn(tcg_res[pass], tcg_op);
9470 read_vec_element(s, tcg_op, rd, pass, MO_64);
9472 gen_helper_neon_addl_u16(tcg_res[pass],
9473 tcg_res[pass], tcg_op);
9475 gen_helper_neon_addl_u32(tcg_res[pass],
9476 tcg_res[pass], tcg_op);
9479 tcg_temp_free_i64(tcg_op);
9483 tcg_res[1] = tcg_const_i64(0);
9485 for (pass = 0; pass < 2; pass++) {
9486 write_vec_element(s, tcg_res[pass], rd, pass, MO_64);
9487 tcg_temp_free_i64(tcg_res[pass]);
9491 static void handle_shll(DisasContext *s, bool is_q, int size, int rn, int rd)
9493 /* Implement SHLL and SHLL2 */
9495 int part = is_q ? 2 : 0;
9496 TCGv_i64 tcg_res[2];
9498 for (pass = 0; pass < 2; pass++) {
9499 static NeonGenWidenFn * const widenfns[3] = {
9500 gen_helper_neon_widen_u8,
9501 gen_helper_neon_widen_u16,
9502 tcg_gen_extu_i32_i64,
9504 NeonGenWidenFn *widenfn = widenfns[size];
9505 TCGv_i32 tcg_op = tcg_temp_new_i32();
9507 read_vec_element_i32(s, tcg_op, rn, part + pass, MO_32);
9508 tcg_res[pass] = tcg_temp_new_i64();
9509 widenfn(tcg_res[pass], tcg_op);
9510 tcg_gen_shli_i64(tcg_res[pass], tcg_res[pass], 8 << size);
9512 tcg_temp_free_i32(tcg_op);
9515 for (pass = 0; pass < 2; pass++) {
9516 write_vec_element(s, tcg_res[pass], rd, pass, MO_64);
9517 tcg_temp_free_i64(tcg_res[pass]);
9521 /* C3.6.17 AdvSIMD two reg misc
9522 * 31 30 29 28 24 23 22 21 17 16 12 11 10 9 5 4 0
9523 * +---+---+---+-----------+------+-----------+--------+-----+------+------+
9524 * | 0 | Q | U | 0 1 1 1 0 | size | 1 0 0 0 0 | opcode | 1 0 | Rn | Rd |
9525 * +---+---+---+-----------+------+-----------+--------+-----+------+------+
9527 static void disas_simd_two_reg_misc(DisasContext *s, uint32_t insn)
9529 int size = extract32(insn, 22, 2);
9530 int opcode = extract32(insn, 12, 5);
9531 bool u = extract32(insn, 29, 1);
9532 bool is_q = extract32(insn, 30, 1);
9533 int rn = extract32(insn, 5, 5);
9534 int rd = extract32(insn, 0, 5);
9535 bool need_fpstatus = false;
9536 bool need_rmode = false;
9539 TCGv_ptr tcg_fpstatus;
9542 case 0x0: /* REV64, REV32 */
9543 case 0x1: /* REV16 */
9544 handle_rev(s, opcode, u, is_q, size, rn, rd);
9546 case 0x5: /* CNT, NOT, RBIT */
9547 if (u && size == 0) {
9548 /* NOT: adjust size so we can use the 64-bits-at-a-time loop. */
9551 } else if (u && size == 1) {
9554 } else if (!u && size == 0) {
9558 unallocated_encoding(s);
9560 case 0x12: /* XTN, XTN2, SQXTUN, SQXTUN2 */
9561 case 0x14: /* SQXTN, SQXTN2, UQXTN, UQXTN2 */
9563 unallocated_encoding(s);
9566 if (!fp_access_check(s)) {
9570 handle_2misc_narrow(s, false, opcode, u, is_q, size, rn, rd);
9572 case 0x4: /* CLS, CLZ */
9574 unallocated_encoding(s);
9578 case 0x2: /* SADDLP, UADDLP */
9579 case 0x6: /* SADALP, UADALP */
9581 unallocated_encoding(s);
9584 if (!fp_access_check(s)) {
9587 handle_2misc_pairwise(s, opcode, u, is_q, size, rn, rd);
9589 case 0x13: /* SHLL, SHLL2 */
9590 if (u == 0 || size == 3) {
9591 unallocated_encoding(s);
9594 if (!fp_access_check(s)) {
9597 handle_shll(s, is_q, size, rn, rd);
9599 case 0xa: /* CMLT */
9601 unallocated_encoding(s);
9605 case 0x8: /* CMGT, CMGE */
9606 case 0x9: /* CMEQ, CMLE */
9607 case 0xb: /* ABS, NEG */
9608 if (size == 3 && !is_q) {
9609 unallocated_encoding(s);
9613 case 0x3: /* SUQADD, USQADD */
9614 if (size == 3 && !is_q) {
9615 unallocated_encoding(s);
9618 if (!fp_access_check(s)) {
9621 handle_2misc_satacc(s, false, u, is_q, size, rn, rd);
9623 case 0x7: /* SQABS, SQNEG */
9624 if (size == 3 && !is_q) {
9625 unallocated_encoding(s);
9633 /* Floating point: U, size[1] and opcode indicate operation;
9634 * size[0] indicates single or double precision.
9636 int is_double = extract32(size, 0, 1);
9637 opcode |= (extract32(size, 1, 1) << 5) | (u << 6);
9638 size = is_double ? 3 : 2;
9640 case 0x2f: /* FABS */
9641 case 0x6f: /* FNEG */
9642 if (size == 3 && !is_q) {
9643 unallocated_encoding(s);
9647 case 0x1d: /* SCVTF */
9648 case 0x5d: /* UCVTF */
9650 bool is_signed = (opcode == 0x1d) ? true : false;
9651 int elements = is_double ? 2 : is_q ? 4 : 2;
9652 if (is_double && !is_q) {
9653 unallocated_encoding(s);
9656 if (!fp_access_check(s)) {
9659 handle_simd_intfp_conv(s, rd, rn, elements, is_signed, 0, size);
9662 case 0x2c: /* FCMGT (zero) */
9663 case 0x2d: /* FCMEQ (zero) */
9664 case 0x2e: /* FCMLT (zero) */
9665 case 0x6c: /* FCMGE (zero) */
9666 case 0x6d: /* FCMLE (zero) */
9667 if (size == 3 && !is_q) {
9668 unallocated_encoding(s);
9671 handle_2misc_fcmp_zero(s, opcode, false, u, is_q, size, rn, rd);
9673 case 0x7f: /* FSQRT */
9674 if (size == 3 && !is_q) {
9675 unallocated_encoding(s);
9679 case 0x1a: /* FCVTNS */
9680 case 0x1b: /* FCVTMS */
9681 case 0x3a: /* FCVTPS */
9682 case 0x3b: /* FCVTZS */
9683 case 0x5a: /* FCVTNU */
9684 case 0x5b: /* FCVTMU */
9685 case 0x7a: /* FCVTPU */
9686 case 0x7b: /* FCVTZU */
9687 need_fpstatus = true;
9689 rmode = extract32(opcode, 5, 1) | (extract32(opcode, 0, 1) << 1);
9690 if (size == 3 && !is_q) {
9691 unallocated_encoding(s);
9695 case 0x5c: /* FCVTAU */
9696 case 0x1c: /* FCVTAS */
9697 need_fpstatus = true;
9699 rmode = FPROUNDING_TIEAWAY;
9700 if (size == 3 && !is_q) {
9701 unallocated_encoding(s);
9705 case 0x3c: /* URECPE */
9707 unallocated_encoding(s);
9711 case 0x3d: /* FRECPE */
9712 case 0x7d: /* FRSQRTE */
9713 if (size == 3 && !is_q) {
9714 unallocated_encoding(s);
9717 if (!fp_access_check(s)) {
9720 handle_2misc_reciprocal(s, opcode, false, u, is_q, size, rn, rd);
9722 case 0x56: /* FCVTXN, FCVTXN2 */
9724 unallocated_encoding(s);
9728 case 0x16: /* FCVTN, FCVTN2 */
9729 /* handle_2misc_narrow does a 2*size -> size operation, but these
9730 * instructions encode the source size rather than dest size.
9732 if (!fp_access_check(s)) {
9735 handle_2misc_narrow(s, false, opcode, 0, is_q, size - 1, rn, rd);
9737 case 0x17: /* FCVTL, FCVTL2 */
9738 if (!fp_access_check(s)) {
9741 handle_2misc_widening(s, opcode, is_q, size, rn, rd);
9743 case 0x18: /* FRINTN */
9744 case 0x19: /* FRINTM */
9745 case 0x38: /* FRINTP */
9746 case 0x39: /* FRINTZ */
9748 rmode = extract32(opcode, 5, 1) | (extract32(opcode, 0, 1) << 1);
9750 case 0x59: /* FRINTX */
9751 case 0x79: /* FRINTI */
9752 need_fpstatus = true;
9753 if (size == 3 && !is_q) {
9754 unallocated_encoding(s);
9758 case 0x58: /* FRINTA */
9760 rmode = FPROUNDING_TIEAWAY;
9761 need_fpstatus = true;
9762 if (size == 3 && !is_q) {
9763 unallocated_encoding(s);
9767 case 0x7c: /* URSQRTE */
9769 unallocated_encoding(s);
9772 need_fpstatus = true;
9775 unallocated_encoding(s);
9781 unallocated_encoding(s);
9785 if (!fp_access_check(s)) {
9789 if (need_fpstatus) {
9790 tcg_fpstatus = get_fpstatus_ptr();
9792 TCGV_UNUSED_PTR(tcg_fpstatus);
9795 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rmode));
9796 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
9798 TCGV_UNUSED_I32(tcg_rmode);
9802 /* All 64-bit element operations can be shared with scalar 2misc */
9805 for (pass = 0; pass < (is_q ? 2 : 1); pass++) {
9806 TCGv_i64 tcg_op = tcg_temp_new_i64();
9807 TCGv_i64 tcg_res = tcg_temp_new_i64();
9809 read_vec_element(s, tcg_op, rn, pass, MO_64);
9811 handle_2misc_64(s, opcode, u, tcg_res, tcg_op,
9812 tcg_rmode, tcg_fpstatus);
9814 write_vec_element(s, tcg_res, rd, pass, MO_64);
9816 tcg_temp_free_i64(tcg_res);
9817 tcg_temp_free_i64(tcg_op);
9822 for (pass = 0; pass < (is_q ? 4 : 2); pass++) {
9823 TCGv_i32 tcg_op = tcg_temp_new_i32();
9824 TCGv_i32 tcg_res = tcg_temp_new_i32();
9827 read_vec_element_i32(s, tcg_op, rn, pass, MO_32);
9830 /* Special cases for 32 bit elements */
9832 case 0xa: /* CMLT */
9833 /* 32 bit integer comparison against zero, result is
9834 * test ? (2^32 - 1) : 0. We implement via setcond(test)
9839 tcg_gen_setcondi_i32(cond, tcg_res, tcg_op, 0);
9840 tcg_gen_neg_i32(tcg_res, tcg_res);
9842 case 0x8: /* CMGT, CMGE */
9843 cond = u ? TCG_COND_GE : TCG_COND_GT;
9845 case 0x9: /* CMEQ, CMLE */
9846 cond = u ? TCG_COND_LE : TCG_COND_EQ;
9850 gen_helper_clz32(tcg_res, tcg_op);
9852 gen_helper_cls32(tcg_res, tcg_op);
9855 case 0x7: /* SQABS, SQNEG */
9857 gen_helper_neon_qneg_s32(tcg_res, cpu_env, tcg_op);
9859 gen_helper_neon_qabs_s32(tcg_res, cpu_env, tcg_op);
9862 case 0xb: /* ABS, NEG */
9864 tcg_gen_neg_i32(tcg_res, tcg_op);
9866 TCGv_i32 tcg_zero = tcg_const_i32(0);
9867 tcg_gen_neg_i32(tcg_res, tcg_op);
9868 tcg_gen_movcond_i32(TCG_COND_GT, tcg_res, tcg_op,
9869 tcg_zero, tcg_op, tcg_res);
9870 tcg_temp_free_i32(tcg_zero);
9873 case 0x2f: /* FABS */
9874 gen_helper_vfp_abss(tcg_res, tcg_op);
9876 case 0x6f: /* FNEG */
9877 gen_helper_vfp_negs(tcg_res, tcg_op);
9879 case 0x7f: /* FSQRT */
9880 gen_helper_vfp_sqrts(tcg_res, tcg_op, cpu_env);
9882 case 0x1a: /* FCVTNS */
9883 case 0x1b: /* FCVTMS */
9884 case 0x1c: /* FCVTAS */
9885 case 0x3a: /* FCVTPS */
9886 case 0x3b: /* FCVTZS */
9888 TCGv_i32 tcg_shift = tcg_const_i32(0);
9889 gen_helper_vfp_tosls(tcg_res, tcg_op,
9890 tcg_shift, tcg_fpstatus);
9891 tcg_temp_free_i32(tcg_shift);
9894 case 0x5a: /* FCVTNU */
9895 case 0x5b: /* FCVTMU */
9896 case 0x5c: /* FCVTAU */
9897 case 0x7a: /* FCVTPU */
9898 case 0x7b: /* FCVTZU */
9900 TCGv_i32 tcg_shift = tcg_const_i32(0);
9901 gen_helper_vfp_touls(tcg_res, tcg_op,
9902 tcg_shift, tcg_fpstatus);
9903 tcg_temp_free_i32(tcg_shift);
9906 case 0x18: /* FRINTN */
9907 case 0x19: /* FRINTM */
9908 case 0x38: /* FRINTP */
9909 case 0x39: /* FRINTZ */
9910 case 0x58: /* FRINTA */
9911 case 0x79: /* FRINTI */
9912 gen_helper_rints(tcg_res, tcg_op, tcg_fpstatus);
9914 case 0x59: /* FRINTX */
9915 gen_helper_rints_exact(tcg_res, tcg_op, tcg_fpstatus);
9917 case 0x7c: /* URSQRTE */
9918 gen_helper_rsqrte_u32(tcg_res, tcg_op, tcg_fpstatus);
9921 g_assert_not_reached();
9924 /* Use helpers for 8 and 16 bit elements */
9926 case 0x5: /* CNT, RBIT */
9927 /* For these two insns size is part of the opcode specifier
9928 * (handled earlier); they always operate on byte elements.
9931 gen_helper_neon_rbit_u8(tcg_res, tcg_op);
9933 gen_helper_neon_cnt_u8(tcg_res, tcg_op);
9936 case 0x7: /* SQABS, SQNEG */
9938 NeonGenOneOpEnvFn *genfn;
9939 static NeonGenOneOpEnvFn * const fns[2][2] = {
9940 { gen_helper_neon_qabs_s8, gen_helper_neon_qneg_s8 },
9941 { gen_helper_neon_qabs_s16, gen_helper_neon_qneg_s16 },
9943 genfn = fns[size][u];
9944 genfn(tcg_res, cpu_env, tcg_op);
9947 case 0x8: /* CMGT, CMGE */
9948 case 0x9: /* CMEQ, CMLE */
9949 case 0xa: /* CMLT */
9951 static NeonGenTwoOpFn * const fns[3][2] = {
9952 { gen_helper_neon_cgt_s8, gen_helper_neon_cgt_s16 },
9953 { gen_helper_neon_cge_s8, gen_helper_neon_cge_s16 },
9954 { gen_helper_neon_ceq_u8, gen_helper_neon_ceq_u16 },
9956 NeonGenTwoOpFn *genfn;
9959 TCGv_i32 tcg_zero = tcg_const_i32(0);
9961 /* comp = index into [CMGT, CMGE, CMEQ, CMLE, CMLT] */
9962 comp = (opcode - 0x8) * 2 + u;
9963 /* ...but LE, LT are implemented as reverse GE, GT */
9964 reverse = (comp > 2);
9968 genfn = fns[comp][size];
9970 genfn(tcg_res, tcg_zero, tcg_op);
9972 genfn(tcg_res, tcg_op, tcg_zero);
9974 tcg_temp_free_i32(tcg_zero);
9977 case 0xb: /* ABS, NEG */
9979 TCGv_i32 tcg_zero = tcg_const_i32(0);
9981 gen_helper_neon_sub_u16(tcg_res, tcg_zero, tcg_op);
9983 gen_helper_neon_sub_u8(tcg_res, tcg_zero, tcg_op);
9985 tcg_temp_free_i32(tcg_zero);
9988 gen_helper_neon_abs_s16(tcg_res, tcg_op);
9990 gen_helper_neon_abs_s8(tcg_res, tcg_op);
9994 case 0x4: /* CLS, CLZ */
9997 gen_helper_neon_clz_u8(tcg_res, tcg_op);
9999 gen_helper_neon_clz_u16(tcg_res, tcg_op);
10003 gen_helper_neon_cls_s8(tcg_res, tcg_op);
10005 gen_helper_neon_cls_s16(tcg_res, tcg_op);
10010 g_assert_not_reached();
10014 write_vec_element_i32(s, tcg_res, rd, pass, MO_32);
10016 tcg_temp_free_i32(tcg_res);
10017 tcg_temp_free_i32(tcg_op);
10021 clear_vec_high(s, rd);
10025 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
10026 tcg_temp_free_i32(tcg_rmode);
10028 if (need_fpstatus) {
10029 tcg_temp_free_ptr(tcg_fpstatus);
10033 /* C3.6.13 AdvSIMD scalar x indexed element
10034 * 31 30 29 28 24 23 22 21 20 19 16 15 12 11 10 9 5 4 0
10035 * +-----+---+-----------+------+---+---+------+-----+---+---+------+------+
10036 * | 0 1 | U | 1 1 1 1 1 | size | L | M | Rm | opc | H | 0 | Rn | Rd |
10037 * +-----+---+-----------+------+---+---+------+-----+---+---+------+------+
10038 * C3.6.18 AdvSIMD vector x indexed element
10039 * 31 30 29 28 24 23 22 21 20 19 16 15 12 11 10 9 5 4 0
10040 * +---+---+---+-----------+------+---+---+------+-----+---+---+------+------+
10041 * | 0 | Q | U | 0 1 1 1 1 | size | L | M | Rm | opc | H | 0 | Rn | Rd |
10042 * +---+---+---+-----------+------+---+---+------+-----+---+---+------+------+
10044 static void disas_simd_indexed(DisasContext *s, uint32_t insn)
10046 /* This encoding has two kinds of instruction:
10047 * normal, where we perform elt x idxelt => elt for each
10048 * element in the vector
10049 * long, where we perform elt x idxelt and generate a result of
10050 * double the width of the input element
10051 * The long ops have a 'part' specifier (ie come in INSN, INSN2 pairs).
10053 bool is_scalar = extract32(insn, 28, 1);
10054 bool is_q = extract32(insn, 30, 1);
10055 bool u = extract32(insn, 29, 1);
10056 int size = extract32(insn, 22, 2);
10057 int l = extract32(insn, 21, 1);
10058 int m = extract32(insn, 20, 1);
10059 /* Note that the Rm field here is only 4 bits, not 5 as it usually is */
10060 int rm = extract32(insn, 16, 4);
10061 int opcode = extract32(insn, 12, 4);
10062 int h = extract32(insn, 11, 1);
10063 int rn = extract32(insn, 5, 5);
10064 int rd = extract32(insn, 0, 5);
10065 bool is_long = false;
10066 bool is_fp = false;
10071 case 0x0: /* MLA */
10072 case 0x4: /* MLS */
10073 if (!u || is_scalar) {
10074 unallocated_encoding(s);
10078 case 0x2: /* SMLAL, SMLAL2, UMLAL, UMLAL2 */
10079 case 0x6: /* SMLSL, SMLSL2, UMLSL, UMLSL2 */
10080 case 0xa: /* SMULL, SMULL2, UMULL, UMULL2 */
10082 unallocated_encoding(s);
10087 case 0x3: /* SQDMLAL, SQDMLAL2 */
10088 case 0x7: /* SQDMLSL, SQDMLSL2 */
10089 case 0xb: /* SQDMULL, SQDMULL2 */
10092 case 0xc: /* SQDMULH */
10093 case 0xd: /* SQRDMULH */
10095 unallocated_encoding(s);
10099 case 0x8: /* MUL */
10100 if (u || is_scalar) {
10101 unallocated_encoding(s);
10105 case 0x1: /* FMLA */
10106 case 0x5: /* FMLS */
10108 unallocated_encoding(s);
10112 case 0x9: /* FMUL, FMULX */
10113 if (!extract32(size, 1, 1)) {
10114 unallocated_encoding(s);
10120 unallocated_encoding(s);
10125 /* low bit of size indicates single/double */
10126 size = extract32(size, 0, 1) ? 3 : 2;
10128 index = h << 1 | l;
10131 unallocated_encoding(s);
10140 index = h << 2 | l << 1 | m;
10143 index = h << 1 | l;
10147 unallocated_encoding(s);
10152 if (!fp_access_check(s)) {
10157 fpst = get_fpstatus_ptr();
10159 TCGV_UNUSED_PTR(fpst);
10163 TCGv_i64 tcg_idx = tcg_temp_new_i64();
10166 assert(is_fp && is_q && !is_long);
10168 read_vec_element(s, tcg_idx, rm, index, MO_64);
10170 for (pass = 0; pass < (is_scalar ? 1 : 2); pass++) {
10171 TCGv_i64 tcg_op = tcg_temp_new_i64();
10172 TCGv_i64 tcg_res = tcg_temp_new_i64();
10174 read_vec_element(s, tcg_op, rn, pass, MO_64);
10177 case 0x5: /* FMLS */
10178 /* As usual for ARM, separate negation for fused multiply-add */
10179 gen_helper_vfp_negd(tcg_op, tcg_op);
10181 case 0x1: /* FMLA */
10182 read_vec_element(s, tcg_res, rd, pass, MO_64);
10183 gen_helper_vfp_muladdd(tcg_res, tcg_op, tcg_idx, tcg_res, fpst);
10185 case 0x9: /* FMUL, FMULX */
10187 gen_helper_vfp_mulxd(tcg_res, tcg_op, tcg_idx, fpst);
10189 gen_helper_vfp_muld(tcg_res, tcg_op, tcg_idx, fpst);
10193 g_assert_not_reached();
10196 write_vec_element(s, tcg_res, rd, pass, MO_64);
10197 tcg_temp_free_i64(tcg_op);
10198 tcg_temp_free_i64(tcg_res);
10202 clear_vec_high(s, rd);
10205 tcg_temp_free_i64(tcg_idx);
10206 } else if (!is_long) {
10207 /* 32 bit floating point, or 16 or 32 bit integer.
10208 * For the 16 bit scalar case we use the usual Neon helpers and
10209 * rely on the fact that 0 op 0 == 0 with no side effects.
10211 TCGv_i32 tcg_idx = tcg_temp_new_i32();
10212 int pass, maxpasses;
10217 maxpasses = is_q ? 4 : 2;
10220 read_vec_element_i32(s, tcg_idx, rm, index, size);
10222 if (size == 1 && !is_scalar) {
10223 /* The simplest way to handle the 16x16 indexed ops is to duplicate
10224 * the index into both halves of the 32 bit tcg_idx and then use
10225 * the usual Neon helpers.
10227 tcg_gen_deposit_i32(tcg_idx, tcg_idx, tcg_idx, 16, 16);
10230 for (pass = 0; pass < maxpasses; pass++) {
10231 TCGv_i32 tcg_op = tcg_temp_new_i32();
10232 TCGv_i32 tcg_res = tcg_temp_new_i32();
10234 read_vec_element_i32(s, tcg_op, rn, pass, is_scalar ? size : MO_32);
10237 case 0x0: /* MLA */
10238 case 0x4: /* MLS */
10239 case 0x8: /* MUL */
10241 static NeonGenTwoOpFn * const fns[2][2] = {
10242 { gen_helper_neon_add_u16, gen_helper_neon_sub_u16 },
10243 { tcg_gen_add_i32, tcg_gen_sub_i32 },
10245 NeonGenTwoOpFn *genfn;
10246 bool is_sub = opcode == 0x4;
10249 gen_helper_neon_mul_u16(tcg_res, tcg_op, tcg_idx);
10251 tcg_gen_mul_i32(tcg_res, tcg_op, tcg_idx);
10253 if (opcode == 0x8) {
10256 read_vec_element_i32(s, tcg_op, rd, pass, MO_32);
10257 genfn = fns[size - 1][is_sub];
10258 genfn(tcg_res, tcg_op, tcg_res);
10261 case 0x5: /* FMLS */
10262 /* As usual for ARM, separate negation for fused multiply-add */
10263 gen_helper_vfp_negs(tcg_op, tcg_op);
10265 case 0x1: /* FMLA */
10266 read_vec_element_i32(s, tcg_res, rd, pass, MO_32);
10267 gen_helper_vfp_muladds(tcg_res, tcg_op, tcg_idx, tcg_res, fpst);
10269 case 0x9: /* FMUL, FMULX */
10271 gen_helper_vfp_mulxs(tcg_res, tcg_op, tcg_idx, fpst);
10273 gen_helper_vfp_muls(tcg_res, tcg_op, tcg_idx, fpst);
10276 case 0xc: /* SQDMULH */
10278 gen_helper_neon_qdmulh_s16(tcg_res, cpu_env,
10281 gen_helper_neon_qdmulh_s32(tcg_res, cpu_env,
10285 case 0xd: /* SQRDMULH */
10287 gen_helper_neon_qrdmulh_s16(tcg_res, cpu_env,
10290 gen_helper_neon_qrdmulh_s32(tcg_res, cpu_env,
10295 g_assert_not_reached();
10299 write_fp_sreg(s, rd, tcg_res);
10301 write_vec_element_i32(s, tcg_res, rd, pass, MO_32);
10304 tcg_temp_free_i32(tcg_op);
10305 tcg_temp_free_i32(tcg_res);
10308 tcg_temp_free_i32(tcg_idx);
10311 clear_vec_high(s, rd);
10314 /* long ops: 16x16->32 or 32x32->64 */
10315 TCGv_i64 tcg_res[2];
10317 bool satop = extract32(opcode, 0, 1);
10318 TCGMemOp memop = MO_32;
10325 TCGv_i64 tcg_idx = tcg_temp_new_i64();
10327 read_vec_element(s, tcg_idx, rm, index, memop);
10329 for (pass = 0; pass < (is_scalar ? 1 : 2); pass++) {
10330 TCGv_i64 tcg_op = tcg_temp_new_i64();
10331 TCGv_i64 tcg_passres;
10337 passelt = pass + (is_q * 2);
10340 read_vec_element(s, tcg_op, rn, passelt, memop);
10342 tcg_res[pass] = tcg_temp_new_i64();
10344 if (opcode == 0xa || opcode == 0xb) {
10345 /* Non-accumulating ops */
10346 tcg_passres = tcg_res[pass];
10348 tcg_passres = tcg_temp_new_i64();
10351 tcg_gen_mul_i64(tcg_passres, tcg_op, tcg_idx);
10352 tcg_temp_free_i64(tcg_op);
10355 /* saturating, doubling */
10356 gen_helper_neon_addl_saturate_s64(tcg_passres, cpu_env,
10357 tcg_passres, tcg_passres);
10360 if (opcode == 0xa || opcode == 0xb) {
10364 /* Accumulating op: handle accumulate step */
10365 read_vec_element(s, tcg_res[pass], rd, pass, MO_64);
10368 case 0x2: /* SMLAL, SMLAL2, UMLAL, UMLAL2 */
10369 tcg_gen_add_i64(tcg_res[pass], tcg_res[pass], tcg_passres);
10371 case 0x6: /* SMLSL, SMLSL2, UMLSL, UMLSL2 */
10372 tcg_gen_sub_i64(tcg_res[pass], tcg_res[pass], tcg_passres);
10374 case 0x7: /* SQDMLSL, SQDMLSL2 */
10375 tcg_gen_neg_i64(tcg_passres, tcg_passres);
10377 case 0x3: /* SQDMLAL, SQDMLAL2 */
10378 gen_helper_neon_addl_saturate_s64(tcg_res[pass], cpu_env,
10383 g_assert_not_reached();
10385 tcg_temp_free_i64(tcg_passres);
10387 tcg_temp_free_i64(tcg_idx);
10390 clear_vec_high(s, rd);
10393 TCGv_i32 tcg_idx = tcg_temp_new_i32();
10396 read_vec_element_i32(s, tcg_idx, rm, index, size);
10399 /* The simplest way to handle the 16x16 indexed ops is to
10400 * duplicate the index into both halves of the 32 bit tcg_idx
10401 * and then use the usual Neon helpers.
10403 tcg_gen_deposit_i32(tcg_idx, tcg_idx, tcg_idx, 16, 16);
10406 for (pass = 0; pass < (is_scalar ? 1 : 2); pass++) {
10407 TCGv_i32 tcg_op = tcg_temp_new_i32();
10408 TCGv_i64 tcg_passres;
10411 read_vec_element_i32(s, tcg_op, rn, pass, size);
10413 read_vec_element_i32(s, tcg_op, rn,
10414 pass + (is_q * 2), MO_32);
10417 tcg_res[pass] = tcg_temp_new_i64();
10419 if (opcode == 0xa || opcode == 0xb) {
10420 /* Non-accumulating ops */
10421 tcg_passres = tcg_res[pass];
10423 tcg_passres = tcg_temp_new_i64();
10426 if (memop & MO_SIGN) {
10427 gen_helper_neon_mull_s16(tcg_passres, tcg_op, tcg_idx);
10429 gen_helper_neon_mull_u16(tcg_passres, tcg_op, tcg_idx);
10432 gen_helper_neon_addl_saturate_s32(tcg_passres, cpu_env,
10433 tcg_passres, tcg_passres);
10435 tcg_temp_free_i32(tcg_op);
10437 if (opcode == 0xa || opcode == 0xb) {
10441 /* Accumulating op: handle accumulate step */
10442 read_vec_element(s, tcg_res[pass], rd, pass, MO_64);
10445 case 0x2: /* SMLAL, SMLAL2, UMLAL, UMLAL2 */
10446 gen_helper_neon_addl_u32(tcg_res[pass], tcg_res[pass],
10449 case 0x6: /* SMLSL, SMLSL2, UMLSL, UMLSL2 */
10450 gen_helper_neon_subl_u32(tcg_res[pass], tcg_res[pass],
10453 case 0x7: /* SQDMLSL, SQDMLSL2 */
10454 gen_helper_neon_negl_u32(tcg_passres, tcg_passres);
10456 case 0x3: /* SQDMLAL, SQDMLAL2 */
10457 gen_helper_neon_addl_saturate_s32(tcg_res[pass], cpu_env,
10462 g_assert_not_reached();
10464 tcg_temp_free_i64(tcg_passres);
10466 tcg_temp_free_i32(tcg_idx);
10469 tcg_gen_ext32u_i64(tcg_res[0], tcg_res[0]);
10474 tcg_res[1] = tcg_const_i64(0);
10477 for (pass = 0; pass < 2; pass++) {
10478 write_vec_element(s, tcg_res[pass], rd, pass, MO_64);
10479 tcg_temp_free_i64(tcg_res[pass]);
10483 if (!TCGV_IS_UNUSED_PTR(fpst)) {
10484 tcg_temp_free_ptr(fpst);
10488 /* C3.6.19 Crypto AES
10489 * 31 24 23 22 21 17 16 12 11 10 9 5 4 0
10490 * +-----------------+------+-----------+--------+-----+------+------+
10491 * | 0 1 0 0 1 1 1 0 | size | 1 0 1 0 0 | opcode | 1 0 | Rn | Rd |
10492 * +-----------------+------+-----------+--------+-----+------+------+
10494 static void disas_crypto_aes(DisasContext *s, uint32_t insn)
10496 unsupported_encoding(s, insn);
10499 /* C3.6.20 Crypto three-reg SHA
10500 * 31 24 23 22 21 20 16 15 14 12 11 10 9 5 4 0
10501 * +-----------------+------+---+------+---+--------+-----+------+------+
10502 * | 0 1 0 1 1 1 1 0 | size | 0 | Rm | 0 | opcode | 0 0 | Rn | Rd |
10503 * +-----------------+------+---+------+---+--------+-----+------+------+
10505 static void disas_crypto_three_reg_sha(DisasContext *s, uint32_t insn)
10507 unsupported_encoding(s, insn);
10510 /* C3.6.21 Crypto two-reg SHA
10511 * 31 24 23 22 21 17 16 12 11 10 9 5 4 0
10512 * +-----------------+------+-----------+--------+-----+------+------+
10513 * | 0 1 0 1 1 1 1 0 | size | 1 0 1 0 0 | opcode | 1 0 | Rn | Rd |
10514 * +-----------------+------+-----------+--------+-----+------+------+
10516 static void disas_crypto_two_reg_sha(DisasContext *s, uint32_t insn)
10518 unsupported_encoding(s, insn);
10521 /* C3.6 Data processing - SIMD, inc Crypto
10523 * As the decode gets a little complex we are using a table based
10524 * approach for this part of the decode.
10526 static const AArch64DecodeTable data_proc_simd[] = {
10527 /* pattern , mask , fn */
10528 { 0x0e200400, 0x9f200400, disas_simd_three_reg_same },
10529 { 0x0e200000, 0x9f200c00, disas_simd_three_reg_diff },
10530 { 0x0e200800, 0x9f3e0c00, disas_simd_two_reg_misc },
10531 { 0x0e300800, 0x9f3e0c00, disas_simd_across_lanes },
10532 { 0x0e000400, 0x9fe08400, disas_simd_copy },
10533 { 0x0f000000, 0x9f000400, disas_simd_indexed }, /* vector indexed */
10534 /* simd_mod_imm decode is a subset of simd_shift_imm, so must precede it */
10535 { 0x0f000400, 0x9ff80400, disas_simd_mod_imm },
10536 { 0x0f000400, 0x9f800400, disas_simd_shift_imm },
10537 { 0x0e000000, 0xbf208c00, disas_simd_tb },
10538 { 0x0e000800, 0xbf208c00, disas_simd_zip_trn },
10539 { 0x2e000000, 0xbf208400, disas_simd_ext },
10540 { 0x5e200400, 0xdf200400, disas_simd_scalar_three_reg_same },
10541 { 0x5e200000, 0xdf200c00, disas_simd_scalar_three_reg_diff },
10542 { 0x5e200800, 0xdf3e0c00, disas_simd_scalar_two_reg_misc },
10543 { 0x5e300800, 0xdf3e0c00, disas_simd_scalar_pairwise },
10544 { 0x5e000400, 0xdfe08400, disas_simd_scalar_copy },
10545 { 0x5f000000, 0xdf000400, disas_simd_indexed }, /* scalar indexed */
10546 { 0x5f000400, 0xdf800400, disas_simd_scalar_shift_imm },
10547 { 0x4e280800, 0xff3e0c00, disas_crypto_aes },
10548 { 0x5e000000, 0xff208c00, disas_crypto_three_reg_sha },
10549 { 0x5e280800, 0xff3e0c00, disas_crypto_two_reg_sha },
10550 { 0x00000000, 0x00000000, NULL }
10553 static void disas_data_proc_simd(DisasContext *s, uint32_t insn)
10555 /* Note that this is called with all non-FP cases from
10556 * table C3-6 so it must UNDEF for entries not specifically
10557 * allocated to instructions in that table.
10559 AArch64DecodeFn *fn = lookup_disas_fn(&data_proc_simd[0], insn);
10563 unallocated_encoding(s);
10567 /* C3.6 Data processing - SIMD and floating point */
10568 static void disas_data_proc_simd_fp(DisasContext *s, uint32_t insn)
10570 if (extract32(insn, 28, 1) == 1 && extract32(insn, 30, 1) == 0) {
10571 disas_data_proc_fp(s, insn);
10573 /* SIMD, including crypto */
10574 disas_data_proc_simd(s, insn);
10578 /* C3.1 A64 instruction index by encoding */
10579 static void disas_a64_insn(CPUARMState *env, DisasContext *s)
10583 insn = arm_ldl_code(env, s->pc, s->bswap_code);
10587 s->fp_access_checked = false;
10589 switch (extract32(insn, 25, 4)) {
10590 case 0x0: case 0x1: case 0x2: case 0x3: /* UNALLOCATED */
10591 unallocated_encoding(s);
10593 case 0x8: case 0x9: /* Data processing - immediate */
10594 disas_data_proc_imm(s, insn);
10596 case 0xa: case 0xb: /* Branch, exception generation and system insns */
10597 disas_b_exc_sys(s, insn);
10602 case 0xe: /* Loads and stores */
10603 disas_ldst(s, insn);
10606 case 0xd: /* Data processing - register */
10607 disas_data_proc_reg(s, insn);
10610 case 0xf: /* Data processing - SIMD and floating point */
10611 disas_data_proc_simd_fp(s, insn);
10614 assert(FALSE); /* all 15 cases should be handled above */
10618 /* if we allocated any temporaries, free them here */
10622 void gen_intermediate_code_internal_a64(ARMCPU *cpu,
10623 TranslationBlock *tb,
10626 CPUState *cs = CPU(cpu);
10627 CPUARMState *env = &cpu->env;
10628 DisasContext dc1, *dc = &dc1;
10630 uint16_t *gen_opc_end;
10632 target_ulong pc_start;
10633 target_ulong next_page_start;
10641 gen_opc_end = tcg_ctx.gen_opc_buf + OPC_MAX_SIZE;
10643 dc->is_jmp = DISAS_NEXT;
10645 dc->singlestep_enabled = cs->singlestep_enabled;
10650 dc->bswap_code = 0;
10651 dc->condexec_mask = 0;
10652 dc->condexec_cond = 0;
10653 #if !defined(CONFIG_USER_ONLY)
10654 dc->user = (ARM_TBFLAG_AA64_EL(tb->flags) == 0);
10656 dc->cpacr_fpen = ARM_TBFLAG_AA64_FPEN(tb->flags);
10658 dc->vec_stride = 0;
10659 dc->cp_regs = cpu->cp_regs;
10660 dc->current_pl = arm_current_pl(env);
10661 dc->features = env->features;
10663 init_tmp_a64_array(dc);
10665 next_page_start = (pc_start & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE;
10668 max_insns = tb->cflags & CF_COUNT_MASK;
10669 if (max_insns == 0) {
10670 max_insns = CF_COUNT_MASK;
10675 tcg_clear_temp_count();
10678 if (unlikely(!QTAILQ_EMPTY(&cs->breakpoints))) {
10679 QTAILQ_FOREACH(bp, &cs->breakpoints, entry) {
10680 if (bp->pc == dc->pc) {
10681 gen_exception_internal_insn(dc, 0, EXCP_DEBUG);
10682 /* Advance PC so that clearing the breakpoint will
10683 invalidate this TB. */
10685 goto done_generating;
10691 j = tcg_ctx.gen_opc_ptr - tcg_ctx.gen_opc_buf;
10695 tcg_ctx.gen_opc_instr_start[lj++] = 0;
10698 tcg_ctx.gen_opc_pc[lj] = dc->pc;
10699 tcg_ctx.gen_opc_instr_start[lj] = 1;
10700 tcg_ctx.gen_opc_icount[lj] = num_insns;
10703 if (num_insns + 1 == max_insns && (tb->cflags & CF_LAST_IO)) {
10707 if (unlikely(qemu_loglevel_mask(CPU_LOG_TB_OP | CPU_LOG_TB_OP_OPT))) {
10708 tcg_gen_debug_insn_start(dc->pc);
10711 disas_a64_insn(env, dc);
10713 if (tcg_check_temp_count()) {
10714 fprintf(stderr, "TCG temporary leak before "TARGET_FMT_lx"\n",
10718 /* Translation stops when a conditional branch is encountered.
10719 * Otherwise the subsequent code could get translated several times.
10720 * Also stop translation when a page boundary is reached. This
10721 * ensures prefetch aborts occur at the right place.
10724 } while (!dc->is_jmp && tcg_ctx.gen_opc_ptr < gen_opc_end &&
10725 !cs->singlestep_enabled &&
10727 dc->pc < next_page_start &&
10728 num_insns < max_insns);
10730 if (tb->cflags & CF_LAST_IO) {
10734 if (unlikely(cs->singlestep_enabled) && dc->is_jmp != DISAS_EXC) {
10735 /* Note that this means single stepping WFI doesn't halt the CPU.
10736 * For conditional branch insns this is harmless unreachable code as
10737 * gen_goto_tb() has already handled emitting the debug exception
10738 * (and thus a tb-jump is not possible when singlestepping).
10740 assert(dc->is_jmp != DISAS_TB_JUMP);
10741 if (dc->is_jmp != DISAS_JUMP) {
10742 gen_a64_set_pc_im(dc->pc);
10744 gen_exception_internal(EXCP_DEBUG);
10746 switch (dc->is_jmp) {
10748 gen_goto_tb(dc, 1, dc->pc);
10752 gen_a64_set_pc_im(dc->pc);
10755 /* indicate that the hash table must be used to find the next TB */
10756 tcg_gen_exit_tb(0);
10758 case DISAS_TB_JUMP:
10763 gen_a64_set_pc_im(dc->pc);
10764 gen_helper_wfe(cpu_env);
10767 /* This is a special case because we don't want to just halt the CPU
10768 * if trying to debug across a WFI.
10770 gen_a64_set_pc_im(dc->pc);
10771 gen_helper_wfi(cpu_env);
10777 gen_tb_end(tb, num_insns);
10778 *tcg_ctx.gen_opc_ptr = INDEX_op_end;
10781 if (qemu_loglevel_mask(CPU_LOG_TB_IN_ASM)) {
10782 qemu_log("----------------\n");
10783 qemu_log("IN: %s\n", lookup_symbol(pc_start));
10784 log_target_disas(env, pc_start, dc->pc - pc_start,
10785 4 | (dc->bswap_code << 1));
10790 j = tcg_ctx.gen_opc_ptr - tcg_ctx.gen_opc_buf;
10793 tcg_ctx.gen_opc_instr_start[lj++] = 0;
10796 tb->size = dc->pc - pc_start;
10797 tb->icount = num_insns;