vmxnet3: Drop net_vmxnet3_info.can_receive
[qemu/ar7.git] / target-arm / translate.c
blobe27634f3c8ce006ee90f5a12fa34ab5489491bb6
1 /*
2 * ARM translation
4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2005-2007 CodeSourcery
6 * Copyright (c) 2007 OpenedHand, Ltd.
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2 of the License, or (at your option) any later version.
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
21 #include <stdarg.h>
22 #include <stdlib.h>
23 #include <stdio.h>
24 #include <string.h>
25 #include <inttypes.h>
27 #include "cpu.h"
28 #include "internals.h"
29 #include "disas/disas.h"
30 #include "tcg-op.h"
31 #include "qemu/log.h"
32 #include "qemu/bitops.h"
33 #include "arm_ldst.h"
35 #include "exec/helper-proto.h"
36 #include "exec/helper-gen.h"
38 #include "trace-tcg.h"
41 #define ENABLE_ARCH_4T arm_dc_feature(s, ARM_FEATURE_V4T)
42 #define ENABLE_ARCH_5 arm_dc_feature(s, ARM_FEATURE_V5)
43 /* currently all emulated v5 cores are also v5TE, so don't bother */
44 #define ENABLE_ARCH_5TE arm_dc_feature(s, ARM_FEATURE_V5)
45 #define ENABLE_ARCH_5J 0
46 #define ENABLE_ARCH_6 arm_dc_feature(s, ARM_FEATURE_V6)
47 #define ENABLE_ARCH_6K arm_dc_feature(s, ARM_FEATURE_V6K)
48 #define ENABLE_ARCH_6T2 arm_dc_feature(s, ARM_FEATURE_THUMB2)
49 #define ENABLE_ARCH_7 arm_dc_feature(s, ARM_FEATURE_V7)
50 #define ENABLE_ARCH_8 arm_dc_feature(s, ARM_FEATURE_V8)
52 #define ARCH(x) do { if (!ENABLE_ARCH_##x) goto illegal_op; } while(0)
54 #include "translate.h"
55 static uint32_t gen_opc_condexec_bits[OPC_BUF_SIZE];
57 #if defined(CONFIG_USER_ONLY)
58 #define IS_USER(s) 1
59 #else
60 #define IS_USER(s) (s->user)
61 #endif
63 TCGv_ptr cpu_env;
64 /* We reuse the same 64-bit temporaries for efficiency. */
65 static TCGv_i64 cpu_V0, cpu_V1, cpu_M0;
66 static TCGv_i32 cpu_R[16];
67 static TCGv_i32 cpu_CF, cpu_NF, cpu_VF, cpu_ZF;
68 static TCGv_i64 cpu_exclusive_addr;
69 static TCGv_i64 cpu_exclusive_val;
70 #ifdef CONFIG_USER_ONLY
71 static TCGv_i64 cpu_exclusive_test;
72 static TCGv_i32 cpu_exclusive_info;
73 #endif
75 /* FIXME: These should be removed. */
76 static TCGv_i32 cpu_F0s, cpu_F1s;
77 static TCGv_i64 cpu_F0d, cpu_F1d;
79 #include "exec/gen-icount.h"
81 static const char *regnames[] =
82 { "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7",
83 "r8", "r9", "r10", "r11", "r12", "r13", "r14", "pc" };
85 /* initialize TCG globals. */
86 void arm_translate_init(void)
88 int i;
90 cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env");
92 for (i = 0; i < 16; i++) {
93 cpu_R[i] = tcg_global_mem_new_i32(TCG_AREG0,
94 offsetof(CPUARMState, regs[i]),
95 regnames[i]);
97 cpu_CF = tcg_global_mem_new_i32(TCG_AREG0, offsetof(CPUARMState, CF), "CF");
98 cpu_NF = tcg_global_mem_new_i32(TCG_AREG0, offsetof(CPUARMState, NF), "NF");
99 cpu_VF = tcg_global_mem_new_i32(TCG_AREG0, offsetof(CPUARMState, VF), "VF");
100 cpu_ZF = tcg_global_mem_new_i32(TCG_AREG0, offsetof(CPUARMState, ZF), "ZF");
102 cpu_exclusive_addr = tcg_global_mem_new_i64(TCG_AREG0,
103 offsetof(CPUARMState, exclusive_addr), "exclusive_addr");
104 cpu_exclusive_val = tcg_global_mem_new_i64(TCG_AREG0,
105 offsetof(CPUARMState, exclusive_val), "exclusive_val");
106 #ifdef CONFIG_USER_ONLY
107 cpu_exclusive_test = tcg_global_mem_new_i64(TCG_AREG0,
108 offsetof(CPUARMState, exclusive_test), "exclusive_test");
109 cpu_exclusive_info = tcg_global_mem_new_i32(TCG_AREG0,
110 offsetof(CPUARMState, exclusive_info), "exclusive_info");
111 #endif
113 a64_translate_init();
116 static inline ARMMMUIdx get_a32_user_mem_index(DisasContext *s)
118 /* Return the mmu_idx to use for A32/T32 "unprivileged load/store"
119 * insns:
120 * if PL2, UNPREDICTABLE (we choose to implement as if PL0)
121 * otherwise, access as if at PL0.
123 switch (s->mmu_idx) {
124 case ARMMMUIdx_S1E2: /* this one is UNPREDICTABLE */
125 case ARMMMUIdx_S12NSE0:
126 case ARMMMUIdx_S12NSE1:
127 return ARMMMUIdx_S12NSE0;
128 case ARMMMUIdx_S1E3:
129 case ARMMMUIdx_S1SE0:
130 case ARMMMUIdx_S1SE1:
131 return ARMMMUIdx_S1SE0;
132 case ARMMMUIdx_S2NS:
133 default:
134 g_assert_not_reached();
138 static inline TCGv_i32 load_cpu_offset(int offset)
140 TCGv_i32 tmp = tcg_temp_new_i32();
141 tcg_gen_ld_i32(tmp, cpu_env, offset);
142 return tmp;
145 #define load_cpu_field(name) load_cpu_offset(offsetof(CPUARMState, name))
147 static inline void store_cpu_offset(TCGv_i32 var, int offset)
149 tcg_gen_st_i32(var, cpu_env, offset);
150 tcg_temp_free_i32(var);
153 #define store_cpu_field(var, name) \
154 store_cpu_offset(var, offsetof(CPUARMState, name))
156 /* Set a variable to the value of a CPU register. */
157 static void load_reg_var(DisasContext *s, TCGv_i32 var, int reg)
159 if (reg == 15) {
160 uint32_t addr;
161 /* normally, since we updated PC, we need only to add one insn */
162 if (s->thumb)
163 addr = (long)s->pc + 2;
164 else
165 addr = (long)s->pc + 4;
166 tcg_gen_movi_i32(var, addr);
167 } else {
168 tcg_gen_mov_i32(var, cpu_R[reg]);
172 /* Create a new temporary and set it to the value of a CPU register. */
173 static inline TCGv_i32 load_reg(DisasContext *s, int reg)
175 TCGv_i32 tmp = tcg_temp_new_i32();
176 load_reg_var(s, tmp, reg);
177 return tmp;
180 /* Set a CPU register. The source must be a temporary and will be
181 marked as dead. */
182 static void store_reg(DisasContext *s, int reg, TCGv_i32 var)
184 if (reg == 15) {
185 tcg_gen_andi_i32(var, var, ~1);
186 s->is_jmp = DISAS_JUMP;
188 tcg_gen_mov_i32(cpu_R[reg], var);
189 tcg_temp_free_i32(var);
192 /* Value extensions. */
193 #define gen_uxtb(var) tcg_gen_ext8u_i32(var, var)
194 #define gen_uxth(var) tcg_gen_ext16u_i32(var, var)
195 #define gen_sxtb(var) tcg_gen_ext8s_i32(var, var)
196 #define gen_sxth(var) tcg_gen_ext16s_i32(var, var)
198 #define gen_sxtb16(var) gen_helper_sxtb16(var, var)
199 #define gen_uxtb16(var) gen_helper_uxtb16(var, var)
202 static inline void gen_set_cpsr(TCGv_i32 var, uint32_t mask)
204 TCGv_i32 tmp_mask = tcg_const_i32(mask);
205 gen_helper_cpsr_write(cpu_env, var, tmp_mask);
206 tcg_temp_free_i32(tmp_mask);
208 /* Set NZCV flags from the high 4 bits of var. */
209 #define gen_set_nzcv(var) gen_set_cpsr(var, CPSR_NZCV)
211 static void gen_exception_internal(int excp)
213 TCGv_i32 tcg_excp = tcg_const_i32(excp);
215 assert(excp_is_internal(excp));
216 gen_helper_exception_internal(cpu_env, tcg_excp);
217 tcg_temp_free_i32(tcg_excp);
220 static void gen_exception(int excp, uint32_t syndrome, uint32_t target_el)
222 TCGv_i32 tcg_excp = tcg_const_i32(excp);
223 TCGv_i32 tcg_syn = tcg_const_i32(syndrome);
224 TCGv_i32 tcg_el = tcg_const_i32(target_el);
226 gen_helper_exception_with_syndrome(cpu_env, tcg_excp,
227 tcg_syn, tcg_el);
229 tcg_temp_free_i32(tcg_el);
230 tcg_temp_free_i32(tcg_syn);
231 tcg_temp_free_i32(tcg_excp);
234 static void gen_ss_advance(DisasContext *s)
236 /* If the singlestep state is Active-not-pending, advance to
237 * Active-pending.
239 if (s->ss_active) {
240 s->pstate_ss = 0;
241 gen_helper_clear_pstate_ss(cpu_env);
245 static void gen_step_complete_exception(DisasContext *s)
247 /* We just completed step of an insn. Move from Active-not-pending
248 * to Active-pending, and then also take the swstep exception.
249 * This corresponds to making the (IMPDEF) choice to prioritize
250 * swstep exceptions over asynchronous exceptions taken to an exception
251 * level where debug is disabled. This choice has the advantage that
252 * we do not need to maintain internal state corresponding to the
253 * ISV/EX syndrome bits between completion of the step and generation
254 * of the exception, and our syndrome information is always correct.
256 gen_ss_advance(s);
257 gen_exception(EXCP_UDEF, syn_swstep(s->ss_same_el, 1, s->is_ldex),
258 default_exception_el(s));
259 s->is_jmp = DISAS_EXC;
262 static void gen_smul_dual(TCGv_i32 a, TCGv_i32 b)
264 TCGv_i32 tmp1 = tcg_temp_new_i32();
265 TCGv_i32 tmp2 = tcg_temp_new_i32();
266 tcg_gen_ext16s_i32(tmp1, a);
267 tcg_gen_ext16s_i32(tmp2, b);
268 tcg_gen_mul_i32(tmp1, tmp1, tmp2);
269 tcg_temp_free_i32(tmp2);
270 tcg_gen_sari_i32(a, a, 16);
271 tcg_gen_sari_i32(b, b, 16);
272 tcg_gen_mul_i32(b, b, a);
273 tcg_gen_mov_i32(a, tmp1);
274 tcg_temp_free_i32(tmp1);
277 /* Byteswap each halfword. */
278 static void gen_rev16(TCGv_i32 var)
280 TCGv_i32 tmp = tcg_temp_new_i32();
281 tcg_gen_shri_i32(tmp, var, 8);
282 tcg_gen_andi_i32(tmp, tmp, 0x00ff00ff);
283 tcg_gen_shli_i32(var, var, 8);
284 tcg_gen_andi_i32(var, var, 0xff00ff00);
285 tcg_gen_or_i32(var, var, tmp);
286 tcg_temp_free_i32(tmp);
289 /* Byteswap low halfword and sign extend. */
290 static void gen_revsh(TCGv_i32 var)
292 tcg_gen_ext16u_i32(var, var);
293 tcg_gen_bswap16_i32(var, var);
294 tcg_gen_ext16s_i32(var, var);
297 /* Unsigned bitfield extract. */
298 static void gen_ubfx(TCGv_i32 var, int shift, uint32_t mask)
300 if (shift)
301 tcg_gen_shri_i32(var, var, shift);
302 tcg_gen_andi_i32(var, var, mask);
305 /* Signed bitfield extract. */
306 static void gen_sbfx(TCGv_i32 var, int shift, int width)
308 uint32_t signbit;
310 if (shift)
311 tcg_gen_sari_i32(var, var, shift);
312 if (shift + width < 32) {
313 signbit = 1u << (width - 1);
314 tcg_gen_andi_i32(var, var, (1u << width) - 1);
315 tcg_gen_xori_i32(var, var, signbit);
316 tcg_gen_subi_i32(var, var, signbit);
320 /* Return (b << 32) + a. Mark inputs as dead */
321 static TCGv_i64 gen_addq_msw(TCGv_i64 a, TCGv_i32 b)
323 TCGv_i64 tmp64 = tcg_temp_new_i64();
325 tcg_gen_extu_i32_i64(tmp64, b);
326 tcg_temp_free_i32(b);
327 tcg_gen_shli_i64(tmp64, tmp64, 32);
328 tcg_gen_add_i64(a, tmp64, a);
330 tcg_temp_free_i64(tmp64);
331 return a;
334 /* Return (b << 32) - a. Mark inputs as dead. */
335 static TCGv_i64 gen_subq_msw(TCGv_i64 a, TCGv_i32 b)
337 TCGv_i64 tmp64 = tcg_temp_new_i64();
339 tcg_gen_extu_i32_i64(tmp64, b);
340 tcg_temp_free_i32(b);
341 tcg_gen_shli_i64(tmp64, tmp64, 32);
342 tcg_gen_sub_i64(a, tmp64, a);
344 tcg_temp_free_i64(tmp64);
345 return a;
348 /* 32x32->64 multiply. Marks inputs as dead. */
349 static TCGv_i64 gen_mulu_i64_i32(TCGv_i32 a, TCGv_i32 b)
351 TCGv_i32 lo = tcg_temp_new_i32();
352 TCGv_i32 hi = tcg_temp_new_i32();
353 TCGv_i64 ret;
355 tcg_gen_mulu2_i32(lo, hi, a, b);
356 tcg_temp_free_i32(a);
357 tcg_temp_free_i32(b);
359 ret = tcg_temp_new_i64();
360 tcg_gen_concat_i32_i64(ret, lo, hi);
361 tcg_temp_free_i32(lo);
362 tcg_temp_free_i32(hi);
364 return ret;
367 static TCGv_i64 gen_muls_i64_i32(TCGv_i32 a, TCGv_i32 b)
369 TCGv_i32 lo = tcg_temp_new_i32();
370 TCGv_i32 hi = tcg_temp_new_i32();
371 TCGv_i64 ret;
373 tcg_gen_muls2_i32(lo, hi, a, b);
374 tcg_temp_free_i32(a);
375 tcg_temp_free_i32(b);
377 ret = tcg_temp_new_i64();
378 tcg_gen_concat_i32_i64(ret, lo, hi);
379 tcg_temp_free_i32(lo);
380 tcg_temp_free_i32(hi);
382 return ret;
385 /* Swap low and high halfwords. */
386 static void gen_swap_half(TCGv_i32 var)
388 TCGv_i32 tmp = tcg_temp_new_i32();
389 tcg_gen_shri_i32(tmp, var, 16);
390 tcg_gen_shli_i32(var, var, 16);
391 tcg_gen_or_i32(var, var, tmp);
392 tcg_temp_free_i32(tmp);
395 /* Dual 16-bit add. Result placed in t0 and t1 is marked as dead.
396 tmp = (t0 ^ t1) & 0x8000;
397 t0 &= ~0x8000;
398 t1 &= ~0x8000;
399 t0 = (t0 + t1) ^ tmp;
402 static void gen_add16(TCGv_i32 t0, TCGv_i32 t1)
404 TCGv_i32 tmp = tcg_temp_new_i32();
405 tcg_gen_xor_i32(tmp, t0, t1);
406 tcg_gen_andi_i32(tmp, tmp, 0x8000);
407 tcg_gen_andi_i32(t0, t0, ~0x8000);
408 tcg_gen_andi_i32(t1, t1, ~0x8000);
409 tcg_gen_add_i32(t0, t0, t1);
410 tcg_gen_xor_i32(t0, t0, tmp);
411 tcg_temp_free_i32(tmp);
412 tcg_temp_free_i32(t1);
415 /* Set CF to the top bit of var. */
416 static void gen_set_CF_bit31(TCGv_i32 var)
418 tcg_gen_shri_i32(cpu_CF, var, 31);
421 /* Set N and Z flags from var. */
422 static inline void gen_logic_CC(TCGv_i32 var)
424 tcg_gen_mov_i32(cpu_NF, var);
425 tcg_gen_mov_i32(cpu_ZF, var);
428 /* T0 += T1 + CF. */
429 static void gen_adc(TCGv_i32 t0, TCGv_i32 t1)
431 tcg_gen_add_i32(t0, t0, t1);
432 tcg_gen_add_i32(t0, t0, cpu_CF);
435 /* dest = T0 + T1 + CF. */
436 static void gen_add_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
438 tcg_gen_add_i32(dest, t0, t1);
439 tcg_gen_add_i32(dest, dest, cpu_CF);
442 /* dest = T0 - T1 + CF - 1. */
443 static void gen_sub_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
445 tcg_gen_sub_i32(dest, t0, t1);
446 tcg_gen_add_i32(dest, dest, cpu_CF);
447 tcg_gen_subi_i32(dest, dest, 1);
450 /* dest = T0 + T1. Compute C, N, V and Z flags */
451 static void gen_add_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
453 TCGv_i32 tmp = tcg_temp_new_i32();
454 tcg_gen_movi_i32(tmp, 0);
455 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, t1, tmp);
456 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
457 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
458 tcg_gen_xor_i32(tmp, t0, t1);
459 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
460 tcg_temp_free_i32(tmp);
461 tcg_gen_mov_i32(dest, cpu_NF);
464 /* dest = T0 + T1 + CF. Compute C, N, V and Z flags */
465 static void gen_adc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
467 TCGv_i32 tmp = tcg_temp_new_i32();
468 if (TCG_TARGET_HAS_add2_i32) {
469 tcg_gen_movi_i32(tmp, 0);
470 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, cpu_CF, tmp);
471 tcg_gen_add2_i32(cpu_NF, cpu_CF, cpu_NF, cpu_CF, t1, tmp);
472 } else {
473 TCGv_i64 q0 = tcg_temp_new_i64();
474 TCGv_i64 q1 = tcg_temp_new_i64();
475 tcg_gen_extu_i32_i64(q0, t0);
476 tcg_gen_extu_i32_i64(q1, t1);
477 tcg_gen_add_i64(q0, q0, q1);
478 tcg_gen_extu_i32_i64(q1, cpu_CF);
479 tcg_gen_add_i64(q0, q0, q1);
480 tcg_gen_extr_i64_i32(cpu_NF, cpu_CF, q0);
481 tcg_temp_free_i64(q0);
482 tcg_temp_free_i64(q1);
484 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
485 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
486 tcg_gen_xor_i32(tmp, t0, t1);
487 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
488 tcg_temp_free_i32(tmp);
489 tcg_gen_mov_i32(dest, cpu_NF);
492 /* dest = T0 - T1. Compute C, N, V and Z flags */
493 static void gen_sub_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
495 TCGv_i32 tmp;
496 tcg_gen_sub_i32(cpu_NF, t0, t1);
497 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
498 tcg_gen_setcond_i32(TCG_COND_GEU, cpu_CF, t0, t1);
499 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
500 tmp = tcg_temp_new_i32();
501 tcg_gen_xor_i32(tmp, t0, t1);
502 tcg_gen_and_i32(cpu_VF, cpu_VF, tmp);
503 tcg_temp_free_i32(tmp);
504 tcg_gen_mov_i32(dest, cpu_NF);
507 /* dest = T0 + ~T1 + CF. Compute C, N, V and Z flags */
508 static void gen_sbc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
510 TCGv_i32 tmp = tcg_temp_new_i32();
511 tcg_gen_not_i32(tmp, t1);
512 gen_adc_CC(dest, t0, tmp);
513 tcg_temp_free_i32(tmp);
516 #define GEN_SHIFT(name) \
517 static void gen_##name(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1) \
519 TCGv_i32 tmp1, tmp2, tmp3; \
520 tmp1 = tcg_temp_new_i32(); \
521 tcg_gen_andi_i32(tmp1, t1, 0xff); \
522 tmp2 = tcg_const_i32(0); \
523 tmp3 = tcg_const_i32(0x1f); \
524 tcg_gen_movcond_i32(TCG_COND_GTU, tmp2, tmp1, tmp3, tmp2, t0); \
525 tcg_temp_free_i32(tmp3); \
526 tcg_gen_andi_i32(tmp1, tmp1, 0x1f); \
527 tcg_gen_##name##_i32(dest, tmp2, tmp1); \
528 tcg_temp_free_i32(tmp2); \
529 tcg_temp_free_i32(tmp1); \
531 GEN_SHIFT(shl)
532 GEN_SHIFT(shr)
533 #undef GEN_SHIFT
535 static void gen_sar(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
537 TCGv_i32 tmp1, tmp2;
538 tmp1 = tcg_temp_new_i32();
539 tcg_gen_andi_i32(tmp1, t1, 0xff);
540 tmp2 = tcg_const_i32(0x1f);
541 tcg_gen_movcond_i32(TCG_COND_GTU, tmp1, tmp1, tmp2, tmp2, tmp1);
542 tcg_temp_free_i32(tmp2);
543 tcg_gen_sar_i32(dest, t0, tmp1);
544 tcg_temp_free_i32(tmp1);
547 static void tcg_gen_abs_i32(TCGv_i32 dest, TCGv_i32 src)
549 TCGv_i32 c0 = tcg_const_i32(0);
550 TCGv_i32 tmp = tcg_temp_new_i32();
551 tcg_gen_neg_i32(tmp, src);
552 tcg_gen_movcond_i32(TCG_COND_GT, dest, src, c0, src, tmp);
553 tcg_temp_free_i32(c0);
554 tcg_temp_free_i32(tmp);
557 static void shifter_out_im(TCGv_i32 var, int shift)
559 if (shift == 0) {
560 tcg_gen_andi_i32(cpu_CF, var, 1);
561 } else {
562 tcg_gen_shri_i32(cpu_CF, var, shift);
563 if (shift != 31) {
564 tcg_gen_andi_i32(cpu_CF, cpu_CF, 1);
569 /* Shift by immediate. Includes special handling for shift == 0. */
570 static inline void gen_arm_shift_im(TCGv_i32 var, int shiftop,
571 int shift, int flags)
573 switch (shiftop) {
574 case 0: /* LSL */
575 if (shift != 0) {
576 if (flags)
577 shifter_out_im(var, 32 - shift);
578 tcg_gen_shli_i32(var, var, shift);
580 break;
581 case 1: /* LSR */
582 if (shift == 0) {
583 if (flags) {
584 tcg_gen_shri_i32(cpu_CF, var, 31);
586 tcg_gen_movi_i32(var, 0);
587 } else {
588 if (flags)
589 shifter_out_im(var, shift - 1);
590 tcg_gen_shri_i32(var, var, shift);
592 break;
593 case 2: /* ASR */
594 if (shift == 0)
595 shift = 32;
596 if (flags)
597 shifter_out_im(var, shift - 1);
598 if (shift == 32)
599 shift = 31;
600 tcg_gen_sari_i32(var, var, shift);
601 break;
602 case 3: /* ROR/RRX */
603 if (shift != 0) {
604 if (flags)
605 shifter_out_im(var, shift - 1);
606 tcg_gen_rotri_i32(var, var, shift); break;
607 } else {
608 TCGv_i32 tmp = tcg_temp_new_i32();
609 tcg_gen_shli_i32(tmp, cpu_CF, 31);
610 if (flags)
611 shifter_out_im(var, 0);
612 tcg_gen_shri_i32(var, var, 1);
613 tcg_gen_or_i32(var, var, tmp);
614 tcg_temp_free_i32(tmp);
619 static inline void gen_arm_shift_reg(TCGv_i32 var, int shiftop,
620 TCGv_i32 shift, int flags)
622 if (flags) {
623 switch (shiftop) {
624 case 0: gen_helper_shl_cc(var, cpu_env, var, shift); break;
625 case 1: gen_helper_shr_cc(var, cpu_env, var, shift); break;
626 case 2: gen_helper_sar_cc(var, cpu_env, var, shift); break;
627 case 3: gen_helper_ror_cc(var, cpu_env, var, shift); break;
629 } else {
630 switch (shiftop) {
631 case 0:
632 gen_shl(var, var, shift);
633 break;
634 case 1:
635 gen_shr(var, var, shift);
636 break;
637 case 2:
638 gen_sar(var, var, shift);
639 break;
640 case 3: tcg_gen_andi_i32(shift, shift, 0x1f);
641 tcg_gen_rotr_i32(var, var, shift); break;
644 tcg_temp_free_i32(shift);
647 #define PAS_OP(pfx) \
648 switch (op2) { \
649 case 0: gen_pas_helper(glue(pfx,add16)); break; \
650 case 1: gen_pas_helper(glue(pfx,addsubx)); break; \
651 case 2: gen_pas_helper(glue(pfx,subaddx)); break; \
652 case 3: gen_pas_helper(glue(pfx,sub16)); break; \
653 case 4: gen_pas_helper(glue(pfx,add8)); break; \
654 case 7: gen_pas_helper(glue(pfx,sub8)); break; \
656 static void gen_arm_parallel_addsub(int op1, int op2, TCGv_i32 a, TCGv_i32 b)
658 TCGv_ptr tmp;
660 switch (op1) {
661 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b, tmp)
662 case 1:
663 tmp = tcg_temp_new_ptr();
664 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
665 PAS_OP(s)
666 tcg_temp_free_ptr(tmp);
667 break;
668 case 5:
669 tmp = tcg_temp_new_ptr();
670 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
671 PAS_OP(u)
672 tcg_temp_free_ptr(tmp);
673 break;
674 #undef gen_pas_helper
675 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b)
676 case 2:
677 PAS_OP(q);
678 break;
679 case 3:
680 PAS_OP(sh);
681 break;
682 case 6:
683 PAS_OP(uq);
684 break;
685 case 7:
686 PAS_OP(uh);
687 break;
688 #undef gen_pas_helper
691 #undef PAS_OP
693 /* For unknown reasons Arm and Thumb-2 use arbitrarily different encodings. */
694 #define PAS_OP(pfx) \
695 switch (op1) { \
696 case 0: gen_pas_helper(glue(pfx,add8)); break; \
697 case 1: gen_pas_helper(glue(pfx,add16)); break; \
698 case 2: gen_pas_helper(glue(pfx,addsubx)); break; \
699 case 4: gen_pas_helper(glue(pfx,sub8)); break; \
700 case 5: gen_pas_helper(glue(pfx,sub16)); break; \
701 case 6: gen_pas_helper(glue(pfx,subaddx)); break; \
703 static void gen_thumb2_parallel_addsub(int op1, int op2, TCGv_i32 a, TCGv_i32 b)
705 TCGv_ptr tmp;
707 switch (op2) {
708 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b, tmp)
709 case 0:
710 tmp = tcg_temp_new_ptr();
711 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
712 PAS_OP(s)
713 tcg_temp_free_ptr(tmp);
714 break;
715 case 4:
716 tmp = tcg_temp_new_ptr();
717 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
718 PAS_OP(u)
719 tcg_temp_free_ptr(tmp);
720 break;
721 #undef gen_pas_helper
722 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b)
723 case 1:
724 PAS_OP(q);
725 break;
726 case 2:
727 PAS_OP(sh);
728 break;
729 case 5:
730 PAS_OP(uq);
731 break;
732 case 6:
733 PAS_OP(uh);
734 break;
735 #undef gen_pas_helper
738 #undef PAS_OP
741 * generate a conditional branch based on ARM condition code cc.
742 * This is common between ARM and Aarch64 targets.
744 void arm_gen_test_cc(int cc, TCGLabel *label)
746 TCGv_i32 tmp;
747 TCGLabel *inv;
749 switch (cc) {
750 case 0: /* eq: Z */
751 tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_ZF, 0, label);
752 break;
753 case 1: /* ne: !Z */
754 tcg_gen_brcondi_i32(TCG_COND_NE, cpu_ZF, 0, label);
755 break;
756 case 2: /* cs: C */
757 tcg_gen_brcondi_i32(TCG_COND_NE, cpu_CF, 0, label);
758 break;
759 case 3: /* cc: !C */
760 tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_CF, 0, label);
761 break;
762 case 4: /* mi: N */
763 tcg_gen_brcondi_i32(TCG_COND_LT, cpu_NF, 0, label);
764 break;
765 case 5: /* pl: !N */
766 tcg_gen_brcondi_i32(TCG_COND_GE, cpu_NF, 0, label);
767 break;
768 case 6: /* vs: V */
769 tcg_gen_brcondi_i32(TCG_COND_LT, cpu_VF, 0, label);
770 break;
771 case 7: /* vc: !V */
772 tcg_gen_brcondi_i32(TCG_COND_GE, cpu_VF, 0, label);
773 break;
774 case 8: /* hi: C && !Z */
775 inv = gen_new_label();
776 tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_CF, 0, inv);
777 tcg_gen_brcondi_i32(TCG_COND_NE, cpu_ZF, 0, label);
778 gen_set_label(inv);
779 break;
780 case 9: /* ls: !C || Z */
781 tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_CF, 0, label);
782 tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_ZF, 0, label);
783 break;
784 case 10: /* ge: N == V -> N ^ V == 0 */
785 tmp = tcg_temp_new_i32();
786 tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
787 tcg_gen_brcondi_i32(TCG_COND_GE, tmp, 0, label);
788 tcg_temp_free_i32(tmp);
789 break;
790 case 11: /* lt: N != V -> N ^ V != 0 */
791 tmp = tcg_temp_new_i32();
792 tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
793 tcg_gen_brcondi_i32(TCG_COND_LT, tmp, 0, label);
794 tcg_temp_free_i32(tmp);
795 break;
796 case 12: /* gt: !Z && N == V */
797 inv = gen_new_label();
798 tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_ZF, 0, inv);
799 tmp = tcg_temp_new_i32();
800 tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
801 tcg_gen_brcondi_i32(TCG_COND_GE, tmp, 0, label);
802 tcg_temp_free_i32(tmp);
803 gen_set_label(inv);
804 break;
805 case 13: /* le: Z || N != V */
806 tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_ZF, 0, label);
807 tmp = tcg_temp_new_i32();
808 tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
809 tcg_gen_brcondi_i32(TCG_COND_LT, tmp, 0, label);
810 tcg_temp_free_i32(tmp);
811 break;
812 default:
813 fprintf(stderr, "Bad condition code 0x%x\n", cc);
814 abort();
818 static const uint8_t table_logic_cc[16] = {
819 1, /* and */
820 1, /* xor */
821 0, /* sub */
822 0, /* rsb */
823 0, /* add */
824 0, /* adc */
825 0, /* sbc */
826 0, /* rsc */
827 1, /* andl */
828 1, /* xorl */
829 0, /* cmp */
830 0, /* cmn */
831 1, /* orr */
832 1, /* mov */
833 1, /* bic */
834 1, /* mvn */
837 /* Set PC and Thumb state from an immediate address. */
838 static inline void gen_bx_im(DisasContext *s, uint32_t addr)
840 TCGv_i32 tmp;
842 s->is_jmp = DISAS_UPDATE;
843 if (s->thumb != (addr & 1)) {
844 tmp = tcg_temp_new_i32();
845 tcg_gen_movi_i32(tmp, addr & 1);
846 tcg_gen_st_i32(tmp, cpu_env, offsetof(CPUARMState, thumb));
847 tcg_temp_free_i32(tmp);
849 tcg_gen_movi_i32(cpu_R[15], addr & ~1);
852 /* Set PC and Thumb state from var. var is marked as dead. */
853 static inline void gen_bx(DisasContext *s, TCGv_i32 var)
855 s->is_jmp = DISAS_UPDATE;
856 tcg_gen_andi_i32(cpu_R[15], var, ~1);
857 tcg_gen_andi_i32(var, var, 1);
858 store_cpu_field(var, thumb);
861 /* Variant of store_reg which uses branch&exchange logic when storing
862 to r15 in ARM architecture v7 and above. The source must be a temporary
863 and will be marked as dead. */
864 static inline void store_reg_bx(DisasContext *s, int reg, TCGv_i32 var)
866 if (reg == 15 && ENABLE_ARCH_7) {
867 gen_bx(s, var);
868 } else {
869 store_reg(s, reg, var);
873 /* Variant of store_reg which uses branch&exchange logic when storing
874 * to r15 in ARM architecture v5T and above. This is used for storing
875 * the results of a LDR/LDM/POP into r15, and corresponds to the cases
876 * in the ARM ARM which use the LoadWritePC() pseudocode function. */
877 static inline void store_reg_from_load(DisasContext *s, int reg, TCGv_i32 var)
879 if (reg == 15 && ENABLE_ARCH_5) {
880 gen_bx(s, var);
881 } else {
882 store_reg(s, reg, var);
886 /* Abstractions of "generate code to do a guest load/store for
887 * AArch32", where a vaddr is always 32 bits (and is zero
888 * extended if we're a 64 bit core) and data is also
889 * 32 bits unless specifically doing a 64 bit access.
890 * These functions work like tcg_gen_qemu_{ld,st}* except
891 * that the address argument is TCGv_i32 rather than TCGv.
893 #if TARGET_LONG_BITS == 32
895 #define DO_GEN_LD(SUFF, OPC) \
896 static inline void gen_aa32_ld##SUFF(TCGv_i32 val, TCGv_i32 addr, int index) \
898 tcg_gen_qemu_ld_i32(val, addr, index, OPC); \
901 #define DO_GEN_ST(SUFF, OPC) \
902 static inline void gen_aa32_st##SUFF(TCGv_i32 val, TCGv_i32 addr, int index) \
904 tcg_gen_qemu_st_i32(val, addr, index, OPC); \
907 static inline void gen_aa32_ld64(TCGv_i64 val, TCGv_i32 addr, int index)
909 tcg_gen_qemu_ld_i64(val, addr, index, MO_TEQ);
912 static inline void gen_aa32_st64(TCGv_i64 val, TCGv_i32 addr, int index)
914 tcg_gen_qemu_st_i64(val, addr, index, MO_TEQ);
917 #else
919 #define DO_GEN_LD(SUFF, OPC) \
920 static inline void gen_aa32_ld##SUFF(TCGv_i32 val, TCGv_i32 addr, int index) \
922 TCGv addr64 = tcg_temp_new(); \
923 tcg_gen_extu_i32_i64(addr64, addr); \
924 tcg_gen_qemu_ld_i32(val, addr64, index, OPC); \
925 tcg_temp_free(addr64); \
928 #define DO_GEN_ST(SUFF, OPC) \
929 static inline void gen_aa32_st##SUFF(TCGv_i32 val, TCGv_i32 addr, int index) \
931 TCGv addr64 = tcg_temp_new(); \
932 tcg_gen_extu_i32_i64(addr64, addr); \
933 tcg_gen_qemu_st_i32(val, addr64, index, OPC); \
934 tcg_temp_free(addr64); \
937 static inline void gen_aa32_ld64(TCGv_i64 val, TCGv_i32 addr, int index)
939 TCGv addr64 = tcg_temp_new();
940 tcg_gen_extu_i32_i64(addr64, addr);
941 tcg_gen_qemu_ld_i64(val, addr64, index, MO_TEQ);
942 tcg_temp_free(addr64);
945 static inline void gen_aa32_st64(TCGv_i64 val, TCGv_i32 addr, int index)
947 TCGv addr64 = tcg_temp_new();
948 tcg_gen_extu_i32_i64(addr64, addr);
949 tcg_gen_qemu_st_i64(val, addr64, index, MO_TEQ);
950 tcg_temp_free(addr64);
953 #endif
955 DO_GEN_LD(8s, MO_SB)
956 DO_GEN_LD(8u, MO_UB)
957 DO_GEN_LD(16s, MO_TESW)
958 DO_GEN_LD(16u, MO_TEUW)
959 DO_GEN_LD(32u, MO_TEUL)
960 DO_GEN_ST(8, MO_UB)
961 DO_GEN_ST(16, MO_TEUW)
962 DO_GEN_ST(32, MO_TEUL)
964 static inline void gen_set_pc_im(DisasContext *s, target_ulong val)
966 tcg_gen_movi_i32(cpu_R[15], val);
969 static inline void gen_hvc(DisasContext *s, int imm16)
971 /* The pre HVC helper handles cases when HVC gets trapped
972 * as an undefined insn by runtime configuration (ie before
973 * the insn really executes).
975 gen_set_pc_im(s, s->pc - 4);
976 gen_helper_pre_hvc(cpu_env);
977 /* Otherwise we will treat this as a real exception which
978 * happens after execution of the insn. (The distinction matters
979 * for the PC value reported to the exception handler and also
980 * for single stepping.)
982 s->svc_imm = imm16;
983 gen_set_pc_im(s, s->pc);
984 s->is_jmp = DISAS_HVC;
987 static inline void gen_smc(DisasContext *s)
989 /* As with HVC, we may take an exception either before or after
990 * the insn executes.
992 TCGv_i32 tmp;
994 gen_set_pc_im(s, s->pc - 4);
995 tmp = tcg_const_i32(syn_aa32_smc());
996 gen_helper_pre_smc(cpu_env, tmp);
997 tcg_temp_free_i32(tmp);
998 gen_set_pc_im(s, s->pc);
999 s->is_jmp = DISAS_SMC;
1002 static inline void
1003 gen_set_condexec (DisasContext *s)
1005 if (s->condexec_mask) {
1006 uint32_t val = (s->condexec_cond << 4) | (s->condexec_mask >> 1);
1007 TCGv_i32 tmp = tcg_temp_new_i32();
1008 tcg_gen_movi_i32(tmp, val);
1009 store_cpu_field(tmp, condexec_bits);
1013 static void gen_exception_internal_insn(DisasContext *s, int offset, int excp)
1015 gen_set_condexec(s);
1016 gen_set_pc_im(s, s->pc - offset);
1017 gen_exception_internal(excp);
1018 s->is_jmp = DISAS_JUMP;
1021 static void gen_exception_insn(DisasContext *s, int offset, int excp,
1022 int syn, uint32_t target_el)
1024 gen_set_condexec(s);
1025 gen_set_pc_im(s, s->pc - offset);
1026 gen_exception(excp, syn, target_el);
1027 s->is_jmp = DISAS_JUMP;
1030 /* Force a TB lookup after an instruction that changes the CPU state. */
1031 static inline void gen_lookup_tb(DisasContext *s)
1033 tcg_gen_movi_i32(cpu_R[15], s->pc & ~1);
1034 s->is_jmp = DISAS_UPDATE;
1037 static inline void gen_add_data_offset(DisasContext *s, unsigned int insn,
1038 TCGv_i32 var)
1040 int val, rm, shift, shiftop;
1041 TCGv_i32 offset;
1043 if (!(insn & (1 << 25))) {
1044 /* immediate */
1045 val = insn & 0xfff;
1046 if (!(insn & (1 << 23)))
1047 val = -val;
1048 if (val != 0)
1049 tcg_gen_addi_i32(var, var, val);
1050 } else {
1051 /* shift/register */
1052 rm = (insn) & 0xf;
1053 shift = (insn >> 7) & 0x1f;
1054 shiftop = (insn >> 5) & 3;
1055 offset = load_reg(s, rm);
1056 gen_arm_shift_im(offset, shiftop, shift, 0);
1057 if (!(insn & (1 << 23)))
1058 tcg_gen_sub_i32(var, var, offset);
1059 else
1060 tcg_gen_add_i32(var, var, offset);
1061 tcg_temp_free_i32(offset);
1065 static inline void gen_add_datah_offset(DisasContext *s, unsigned int insn,
1066 int extra, TCGv_i32 var)
1068 int val, rm;
1069 TCGv_i32 offset;
1071 if (insn & (1 << 22)) {
1072 /* immediate */
1073 val = (insn & 0xf) | ((insn >> 4) & 0xf0);
1074 if (!(insn & (1 << 23)))
1075 val = -val;
1076 val += extra;
1077 if (val != 0)
1078 tcg_gen_addi_i32(var, var, val);
1079 } else {
1080 /* register */
1081 if (extra)
1082 tcg_gen_addi_i32(var, var, extra);
1083 rm = (insn) & 0xf;
1084 offset = load_reg(s, rm);
1085 if (!(insn & (1 << 23)))
1086 tcg_gen_sub_i32(var, var, offset);
1087 else
1088 tcg_gen_add_i32(var, var, offset);
1089 tcg_temp_free_i32(offset);
1093 static TCGv_ptr get_fpstatus_ptr(int neon)
1095 TCGv_ptr statusptr = tcg_temp_new_ptr();
1096 int offset;
1097 if (neon) {
1098 offset = offsetof(CPUARMState, vfp.standard_fp_status);
1099 } else {
1100 offset = offsetof(CPUARMState, vfp.fp_status);
1102 tcg_gen_addi_ptr(statusptr, cpu_env, offset);
1103 return statusptr;
1106 #define VFP_OP2(name) \
1107 static inline void gen_vfp_##name(int dp) \
1109 TCGv_ptr fpst = get_fpstatus_ptr(0); \
1110 if (dp) { \
1111 gen_helper_vfp_##name##d(cpu_F0d, cpu_F0d, cpu_F1d, fpst); \
1112 } else { \
1113 gen_helper_vfp_##name##s(cpu_F0s, cpu_F0s, cpu_F1s, fpst); \
1115 tcg_temp_free_ptr(fpst); \
1118 VFP_OP2(add)
1119 VFP_OP2(sub)
1120 VFP_OP2(mul)
1121 VFP_OP2(div)
1123 #undef VFP_OP2
1125 static inline void gen_vfp_F1_mul(int dp)
1127 /* Like gen_vfp_mul() but put result in F1 */
1128 TCGv_ptr fpst = get_fpstatus_ptr(0);
1129 if (dp) {
1130 gen_helper_vfp_muld(cpu_F1d, cpu_F0d, cpu_F1d, fpst);
1131 } else {
1132 gen_helper_vfp_muls(cpu_F1s, cpu_F0s, cpu_F1s, fpst);
1134 tcg_temp_free_ptr(fpst);
1137 static inline void gen_vfp_F1_neg(int dp)
1139 /* Like gen_vfp_neg() but put result in F1 */
1140 if (dp) {
1141 gen_helper_vfp_negd(cpu_F1d, cpu_F0d);
1142 } else {
1143 gen_helper_vfp_negs(cpu_F1s, cpu_F0s);
1147 static inline void gen_vfp_abs(int dp)
1149 if (dp)
1150 gen_helper_vfp_absd(cpu_F0d, cpu_F0d);
1151 else
1152 gen_helper_vfp_abss(cpu_F0s, cpu_F0s);
1155 static inline void gen_vfp_neg(int dp)
1157 if (dp)
1158 gen_helper_vfp_negd(cpu_F0d, cpu_F0d);
1159 else
1160 gen_helper_vfp_negs(cpu_F0s, cpu_F0s);
1163 static inline void gen_vfp_sqrt(int dp)
1165 if (dp)
1166 gen_helper_vfp_sqrtd(cpu_F0d, cpu_F0d, cpu_env);
1167 else
1168 gen_helper_vfp_sqrts(cpu_F0s, cpu_F0s, cpu_env);
1171 static inline void gen_vfp_cmp(int dp)
1173 if (dp)
1174 gen_helper_vfp_cmpd(cpu_F0d, cpu_F1d, cpu_env);
1175 else
1176 gen_helper_vfp_cmps(cpu_F0s, cpu_F1s, cpu_env);
1179 static inline void gen_vfp_cmpe(int dp)
1181 if (dp)
1182 gen_helper_vfp_cmped(cpu_F0d, cpu_F1d, cpu_env);
1183 else
1184 gen_helper_vfp_cmpes(cpu_F0s, cpu_F1s, cpu_env);
1187 static inline void gen_vfp_F1_ld0(int dp)
1189 if (dp)
1190 tcg_gen_movi_i64(cpu_F1d, 0);
1191 else
1192 tcg_gen_movi_i32(cpu_F1s, 0);
1195 #define VFP_GEN_ITOF(name) \
1196 static inline void gen_vfp_##name(int dp, int neon) \
1198 TCGv_ptr statusptr = get_fpstatus_ptr(neon); \
1199 if (dp) { \
1200 gen_helper_vfp_##name##d(cpu_F0d, cpu_F0s, statusptr); \
1201 } else { \
1202 gen_helper_vfp_##name##s(cpu_F0s, cpu_F0s, statusptr); \
1204 tcg_temp_free_ptr(statusptr); \
1207 VFP_GEN_ITOF(uito)
1208 VFP_GEN_ITOF(sito)
1209 #undef VFP_GEN_ITOF
1211 #define VFP_GEN_FTOI(name) \
1212 static inline void gen_vfp_##name(int dp, int neon) \
1214 TCGv_ptr statusptr = get_fpstatus_ptr(neon); \
1215 if (dp) { \
1216 gen_helper_vfp_##name##d(cpu_F0s, cpu_F0d, statusptr); \
1217 } else { \
1218 gen_helper_vfp_##name##s(cpu_F0s, cpu_F0s, statusptr); \
1220 tcg_temp_free_ptr(statusptr); \
1223 VFP_GEN_FTOI(toui)
1224 VFP_GEN_FTOI(touiz)
1225 VFP_GEN_FTOI(tosi)
1226 VFP_GEN_FTOI(tosiz)
1227 #undef VFP_GEN_FTOI
1229 #define VFP_GEN_FIX(name, round) \
1230 static inline void gen_vfp_##name(int dp, int shift, int neon) \
1232 TCGv_i32 tmp_shift = tcg_const_i32(shift); \
1233 TCGv_ptr statusptr = get_fpstatus_ptr(neon); \
1234 if (dp) { \
1235 gen_helper_vfp_##name##d##round(cpu_F0d, cpu_F0d, tmp_shift, \
1236 statusptr); \
1237 } else { \
1238 gen_helper_vfp_##name##s##round(cpu_F0s, cpu_F0s, tmp_shift, \
1239 statusptr); \
1241 tcg_temp_free_i32(tmp_shift); \
1242 tcg_temp_free_ptr(statusptr); \
1244 VFP_GEN_FIX(tosh, _round_to_zero)
1245 VFP_GEN_FIX(tosl, _round_to_zero)
1246 VFP_GEN_FIX(touh, _round_to_zero)
1247 VFP_GEN_FIX(toul, _round_to_zero)
1248 VFP_GEN_FIX(shto, )
1249 VFP_GEN_FIX(slto, )
1250 VFP_GEN_FIX(uhto, )
1251 VFP_GEN_FIX(ulto, )
1252 #undef VFP_GEN_FIX
1254 static inline void gen_vfp_ld(DisasContext *s, int dp, TCGv_i32 addr)
1256 if (dp) {
1257 gen_aa32_ld64(cpu_F0d, addr, get_mem_index(s));
1258 } else {
1259 gen_aa32_ld32u(cpu_F0s, addr, get_mem_index(s));
1263 static inline void gen_vfp_st(DisasContext *s, int dp, TCGv_i32 addr)
1265 if (dp) {
1266 gen_aa32_st64(cpu_F0d, addr, get_mem_index(s));
1267 } else {
1268 gen_aa32_st32(cpu_F0s, addr, get_mem_index(s));
1272 static inline long
1273 vfp_reg_offset (int dp, int reg)
1275 if (dp)
1276 return offsetof(CPUARMState, vfp.regs[reg]);
1277 else if (reg & 1) {
1278 return offsetof(CPUARMState, vfp.regs[reg >> 1])
1279 + offsetof(CPU_DoubleU, l.upper);
1280 } else {
1281 return offsetof(CPUARMState, vfp.regs[reg >> 1])
1282 + offsetof(CPU_DoubleU, l.lower);
1286 /* Return the offset of a 32-bit piece of a NEON register.
1287 zero is the least significant end of the register. */
1288 static inline long
1289 neon_reg_offset (int reg, int n)
1291 int sreg;
1292 sreg = reg * 2 + n;
1293 return vfp_reg_offset(0, sreg);
1296 static TCGv_i32 neon_load_reg(int reg, int pass)
1298 TCGv_i32 tmp = tcg_temp_new_i32();
1299 tcg_gen_ld_i32(tmp, cpu_env, neon_reg_offset(reg, pass));
1300 return tmp;
1303 static void neon_store_reg(int reg, int pass, TCGv_i32 var)
1305 tcg_gen_st_i32(var, cpu_env, neon_reg_offset(reg, pass));
1306 tcg_temp_free_i32(var);
1309 static inline void neon_load_reg64(TCGv_i64 var, int reg)
1311 tcg_gen_ld_i64(var, cpu_env, vfp_reg_offset(1, reg));
1314 static inline void neon_store_reg64(TCGv_i64 var, int reg)
1316 tcg_gen_st_i64(var, cpu_env, vfp_reg_offset(1, reg));
1319 #define tcg_gen_ld_f32 tcg_gen_ld_i32
1320 #define tcg_gen_ld_f64 tcg_gen_ld_i64
1321 #define tcg_gen_st_f32 tcg_gen_st_i32
1322 #define tcg_gen_st_f64 tcg_gen_st_i64
1324 static inline void gen_mov_F0_vreg(int dp, int reg)
1326 if (dp)
1327 tcg_gen_ld_f64(cpu_F0d, cpu_env, vfp_reg_offset(dp, reg));
1328 else
1329 tcg_gen_ld_f32(cpu_F0s, cpu_env, vfp_reg_offset(dp, reg));
1332 static inline void gen_mov_F1_vreg(int dp, int reg)
1334 if (dp)
1335 tcg_gen_ld_f64(cpu_F1d, cpu_env, vfp_reg_offset(dp, reg));
1336 else
1337 tcg_gen_ld_f32(cpu_F1s, cpu_env, vfp_reg_offset(dp, reg));
1340 static inline void gen_mov_vreg_F0(int dp, int reg)
1342 if (dp)
1343 tcg_gen_st_f64(cpu_F0d, cpu_env, vfp_reg_offset(dp, reg));
1344 else
1345 tcg_gen_st_f32(cpu_F0s, cpu_env, vfp_reg_offset(dp, reg));
1348 #define ARM_CP_RW_BIT (1 << 20)
1350 static inline void iwmmxt_load_reg(TCGv_i64 var, int reg)
1352 tcg_gen_ld_i64(var, cpu_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1355 static inline void iwmmxt_store_reg(TCGv_i64 var, int reg)
1357 tcg_gen_st_i64(var, cpu_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1360 static inline TCGv_i32 iwmmxt_load_creg(int reg)
1362 TCGv_i32 var = tcg_temp_new_i32();
1363 tcg_gen_ld_i32(var, cpu_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1364 return var;
1367 static inline void iwmmxt_store_creg(int reg, TCGv_i32 var)
1369 tcg_gen_st_i32(var, cpu_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1370 tcg_temp_free_i32(var);
1373 static inline void gen_op_iwmmxt_movq_wRn_M0(int rn)
1375 iwmmxt_store_reg(cpu_M0, rn);
1378 static inline void gen_op_iwmmxt_movq_M0_wRn(int rn)
1380 iwmmxt_load_reg(cpu_M0, rn);
1383 static inline void gen_op_iwmmxt_orq_M0_wRn(int rn)
1385 iwmmxt_load_reg(cpu_V1, rn);
1386 tcg_gen_or_i64(cpu_M0, cpu_M0, cpu_V1);
1389 static inline void gen_op_iwmmxt_andq_M0_wRn(int rn)
1391 iwmmxt_load_reg(cpu_V1, rn);
1392 tcg_gen_and_i64(cpu_M0, cpu_M0, cpu_V1);
1395 static inline void gen_op_iwmmxt_xorq_M0_wRn(int rn)
1397 iwmmxt_load_reg(cpu_V1, rn);
1398 tcg_gen_xor_i64(cpu_M0, cpu_M0, cpu_V1);
1401 #define IWMMXT_OP(name) \
1402 static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1404 iwmmxt_load_reg(cpu_V1, rn); \
1405 gen_helper_iwmmxt_##name(cpu_M0, cpu_M0, cpu_V1); \
1408 #define IWMMXT_OP_ENV(name) \
1409 static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1411 iwmmxt_load_reg(cpu_V1, rn); \
1412 gen_helper_iwmmxt_##name(cpu_M0, cpu_env, cpu_M0, cpu_V1); \
1415 #define IWMMXT_OP_ENV_SIZE(name) \
1416 IWMMXT_OP_ENV(name##b) \
1417 IWMMXT_OP_ENV(name##w) \
1418 IWMMXT_OP_ENV(name##l)
1420 #define IWMMXT_OP_ENV1(name) \
1421 static inline void gen_op_iwmmxt_##name##_M0(void) \
1423 gen_helper_iwmmxt_##name(cpu_M0, cpu_env, cpu_M0); \
1426 IWMMXT_OP(maddsq)
1427 IWMMXT_OP(madduq)
1428 IWMMXT_OP(sadb)
1429 IWMMXT_OP(sadw)
1430 IWMMXT_OP(mulslw)
1431 IWMMXT_OP(mulshw)
1432 IWMMXT_OP(mululw)
1433 IWMMXT_OP(muluhw)
1434 IWMMXT_OP(macsw)
1435 IWMMXT_OP(macuw)
1437 IWMMXT_OP_ENV_SIZE(unpackl)
1438 IWMMXT_OP_ENV_SIZE(unpackh)
1440 IWMMXT_OP_ENV1(unpacklub)
1441 IWMMXT_OP_ENV1(unpackluw)
1442 IWMMXT_OP_ENV1(unpacklul)
1443 IWMMXT_OP_ENV1(unpackhub)
1444 IWMMXT_OP_ENV1(unpackhuw)
1445 IWMMXT_OP_ENV1(unpackhul)
1446 IWMMXT_OP_ENV1(unpacklsb)
1447 IWMMXT_OP_ENV1(unpacklsw)
1448 IWMMXT_OP_ENV1(unpacklsl)
1449 IWMMXT_OP_ENV1(unpackhsb)
1450 IWMMXT_OP_ENV1(unpackhsw)
1451 IWMMXT_OP_ENV1(unpackhsl)
1453 IWMMXT_OP_ENV_SIZE(cmpeq)
1454 IWMMXT_OP_ENV_SIZE(cmpgtu)
1455 IWMMXT_OP_ENV_SIZE(cmpgts)
1457 IWMMXT_OP_ENV_SIZE(mins)
1458 IWMMXT_OP_ENV_SIZE(minu)
1459 IWMMXT_OP_ENV_SIZE(maxs)
1460 IWMMXT_OP_ENV_SIZE(maxu)
1462 IWMMXT_OP_ENV_SIZE(subn)
1463 IWMMXT_OP_ENV_SIZE(addn)
1464 IWMMXT_OP_ENV_SIZE(subu)
1465 IWMMXT_OP_ENV_SIZE(addu)
1466 IWMMXT_OP_ENV_SIZE(subs)
1467 IWMMXT_OP_ENV_SIZE(adds)
1469 IWMMXT_OP_ENV(avgb0)
1470 IWMMXT_OP_ENV(avgb1)
1471 IWMMXT_OP_ENV(avgw0)
1472 IWMMXT_OP_ENV(avgw1)
1474 IWMMXT_OP_ENV(packuw)
1475 IWMMXT_OP_ENV(packul)
1476 IWMMXT_OP_ENV(packuq)
1477 IWMMXT_OP_ENV(packsw)
1478 IWMMXT_OP_ENV(packsl)
1479 IWMMXT_OP_ENV(packsq)
1481 static void gen_op_iwmmxt_set_mup(void)
1483 TCGv_i32 tmp;
1484 tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1485 tcg_gen_ori_i32(tmp, tmp, 2);
1486 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1489 static void gen_op_iwmmxt_set_cup(void)
1491 TCGv_i32 tmp;
1492 tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1493 tcg_gen_ori_i32(tmp, tmp, 1);
1494 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1497 static void gen_op_iwmmxt_setpsr_nz(void)
1499 TCGv_i32 tmp = tcg_temp_new_i32();
1500 gen_helper_iwmmxt_setpsr_nz(tmp, cpu_M0);
1501 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCASF]);
1504 static inline void gen_op_iwmmxt_addl_M0_wRn(int rn)
1506 iwmmxt_load_reg(cpu_V1, rn);
1507 tcg_gen_ext32u_i64(cpu_V1, cpu_V1);
1508 tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
1511 static inline int gen_iwmmxt_address(DisasContext *s, uint32_t insn,
1512 TCGv_i32 dest)
1514 int rd;
1515 uint32_t offset;
1516 TCGv_i32 tmp;
1518 rd = (insn >> 16) & 0xf;
1519 tmp = load_reg(s, rd);
1521 offset = (insn & 0xff) << ((insn >> 7) & 2);
1522 if (insn & (1 << 24)) {
1523 /* Pre indexed */
1524 if (insn & (1 << 23))
1525 tcg_gen_addi_i32(tmp, tmp, offset);
1526 else
1527 tcg_gen_addi_i32(tmp, tmp, -offset);
1528 tcg_gen_mov_i32(dest, tmp);
1529 if (insn & (1 << 21))
1530 store_reg(s, rd, tmp);
1531 else
1532 tcg_temp_free_i32(tmp);
1533 } else if (insn & (1 << 21)) {
1534 /* Post indexed */
1535 tcg_gen_mov_i32(dest, tmp);
1536 if (insn & (1 << 23))
1537 tcg_gen_addi_i32(tmp, tmp, offset);
1538 else
1539 tcg_gen_addi_i32(tmp, tmp, -offset);
1540 store_reg(s, rd, tmp);
1541 } else if (!(insn & (1 << 23)))
1542 return 1;
1543 return 0;
1546 static inline int gen_iwmmxt_shift(uint32_t insn, uint32_t mask, TCGv_i32 dest)
1548 int rd = (insn >> 0) & 0xf;
1549 TCGv_i32 tmp;
1551 if (insn & (1 << 8)) {
1552 if (rd < ARM_IWMMXT_wCGR0 || rd > ARM_IWMMXT_wCGR3) {
1553 return 1;
1554 } else {
1555 tmp = iwmmxt_load_creg(rd);
1557 } else {
1558 tmp = tcg_temp_new_i32();
1559 iwmmxt_load_reg(cpu_V0, rd);
1560 tcg_gen_extrl_i64_i32(tmp, cpu_V0);
1562 tcg_gen_andi_i32(tmp, tmp, mask);
1563 tcg_gen_mov_i32(dest, tmp);
1564 tcg_temp_free_i32(tmp);
1565 return 0;
1568 /* Disassemble an iwMMXt instruction. Returns nonzero if an error occurred
1569 (ie. an undefined instruction). */
1570 static int disas_iwmmxt_insn(DisasContext *s, uint32_t insn)
1572 int rd, wrd;
1573 int rdhi, rdlo, rd0, rd1, i;
1574 TCGv_i32 addr;
1575 TCGv_i32 tmp, tmp2, tmp3;
1577 if ((insn & 0x0e000e00) == 0x0c000000) {
1578 if ((insn & 0x0fe00ff0) == 0x0c400000) {
1579 wrd = insn & 0xf;
1580 rdlo = (insn >> 12) & 0xf;
1581 rdhi = (insn >> 16) & 0xf;
1582 if (insn & ARM_CP_RW_BIT) { /* TMRRC */
1583 iwmmxt_load_reg(cpu_V0, wrd);
1584 tcg_gen_extrl_i64_i32(cpu_R[rdlo], cpu_V0);
1585 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
1586 tcg_gen_extrl_i64_i32(cpu_R[rdhi], cpu_V0);
1587 } else { /* TMCRR */
1588 tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
1589 iwmmxt_store_reg(cpu_V0, wrd);
1590 gen_op_iwmmxt_set_mup();
1592 return 0;
1595 wrd = (insn >> 12) & 0xf;
1596 addr = tcg_temp_new_i32();
1597 if (gen_iwmmxt_address(s, insn, addr)) {
1598 tcg_temp_free_i32(addr);
1599 return 1;
1601 if (insn & ARM_CP_RW_BIT) {
1602 if ((insn >> 28) == 0xf) { /* WLDRW wCx */
1603 tmp = tcg_temp_new_i32();
1604 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
1605 iwmmxt_store_creg(wrd, tmp);
1606 } else {
1607 i = 1;
1608 if (insn & (1 << 8)) {
1609 if (insn & (1 << 22)) { /* WLDRD */
1610 gen_aa32_ld64(cpu_M0, addr, get_mem_index(s));
1611 i = 0;
1612 } else { /* WLDRW wRd */
1613 tmp = tcg_temp_new_i32();
1614 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
1616 } else {
1617 tmp = tcg_temp_new_i32();
1618 if (insn & (1 << 22)) { /* WLDRH */
1619 gen_aa32_ld16u(tmp, addr, get_mem_index(s));
1620 } else { /* WLDRB */
1621 gen_aa32_ld8u(tmp, addr, get_mem_index(s));
1624 if (i) {
1625 tcg_gen_extu_i32_i64(cpu_M0, tmp);
1626 tcg_temp_free_i32(tmp);
1628 gen_op_iwmmxt_movq_wRn_M0(wrd);
1630 } else {
1631 if ((insn >> 28) == 0xf) { /* WSTRW wCx */
1632 tmp = iwmmxt_load_creg(wrd);
1633 gen_aa32_st32(tmp, addr, get_mem_index(s));
1634 } else {
1635 gen_op_iwmmxt_movq_M0_wRn(wrd);
1636 tmp = tcg_temp_new_i32();
1637 if (insn & (1 << 8)) {
1638 if (insn & (1 << 22)) { /* WSTRD */
1639 gen_aa32_st64(cpu_M0, addr, get_mem_index(s));
1640 } else { /* WSTRW wRd */
1641 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1642 gen_aa32_st32(tmp, addr, get_mem_index(s));
1644 } else {
1645 if (insn & (1 << 22)) { /* WSTRH */
1646 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1647 gen_aa32_st16(tmp, addr, get_mem_index(s));
1648 } else { /* WSTRB */
1649 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1650 gen_aa32_st8(tmp, addr, get_mem_index(s));
1654 tcg_temp_free_i32(tmp);
1656 tcg_temp_free_i32(addr);
1657 return 0;
1660 if ((insn & 0x0f000000) != 0x0e000000)
1661 return 1;
1663 switch (((insn >> 12) & 0xf00) | ((insn >> 4) & 0xff)) {
1664 case 0x000: /* WOR */
1665 wrd = (insn >> 12) & 0xf;
1666 rd0 = (insn >> 0) & 0xf;
1667 rd1 = (insn >> 16) & 0xf;
1668 gen_op_iwmmxt_movq_M0_wRn(rd0);
1669 gen_op_iwmmxt_orq_M0_wRn(rd1);
1670 gen_op_iwmmxt_setpsr_nz();
1671 gen_op_iwmmxt_movq_wRn_M0(wrd);
1672 gen_op_iwmmxt_set_mup();
1673 gen_op_iwmmxt_set_cup();
1674 break;
1675 case 0x011: /* TMCR */
1676 if (insn & 0xf)
1677 return 1;
1678 rd = (insn >> 12) & 0xf;
1679 wrd = (insn >> 16) & 0xf;
1680 switch (wrd) {
1681 case ARM_IWMMXT_wCID:
1682 case ARM_IWMMXT_wCASF:
1683 break;
1684 case ARM_IWMMXT_wCon:
1685 gen_op_iwmmxt_set_cup();
1686 /* Fall through. */
1687 case ARM_IWMMXT_wCSSF:
1688 tmp = iwmmxt_load_creg(wrd);
1689 tmp2 = load_reg(s, rd);
1690 tcg_gen_andc_i32(tmp, tmp, tmp2);
1691 tcg_temp_free_i32(tmp2);
1692 iwmmxt_store_creg(wrd, tmp);
1693 break;
1694 case ARM_IWMMXT_wCGR0:
1695 case ARM_IWMMXT_wCGR1:
1696 case ARM_IWMMXT_wCGR2:
1697 case ARM_IWMMXT_wCGR3:
1698 gen_op_iwmmxt_set_cup();
1699 tmp = load_reg(s, rd);
1700 iwmmxt_store_creg(wrd, tmp);
1701 break;
1702 default:
1703 return 1;
1705 break;
1706 case 0x100: /* WXOR */
1707 wrd = (insn >> 12) & 0xf;
1708 rd0 = (insn >> 0) & 0xf;
1709 rd1 = (insn >> 16) & 0xf;
1710 gen_op_iwmmxt_movq_M0_wRn(rd0);
1711 gen_op_iwmmxt_xorq_M0_wRn(rd1);
1712 gen_op_iwmmxt_setpsr_nz();
1713 gen_op_iwmmxt_movq_wRn_M0(wrd);
1714 gen_op_iwmmxt_set_mup();
1715 gen_op_iwmmxt_set_cup();
1716 break;
1717 case 0x111: /* TMRC */
1718 if (insn & 0xf)
1719 return 1;
1720 rd = (insn >> 12) & 0xf;
1721 wrd = (insn >> 16) & 0xf;
1722 tmp = iwmmxt_load_creg(wrd);
1723 store_reg(s, rd, tmp);
1724 break;
1725 case 0x300: /* WANDN */
1726 wrd = (insn >> 12) & 0xf;
1727 rd0 = (insn >> 0) & 0xf;
1728 rd1 = (insn >> 16) & 0xf;
1729 gen_op_iwmmxt_movq_M0_wRn(rd0);
1730 tcg_gen_neg_i64(cpu_M0, cpu_M0);
1731 gen_op_iwmmxt_andq_M0_wRn(rd1);
1732 gen_op_iwmmxt_setpsr_nz();
1733 gen_op_iwmmxt_movq_wRn_M0(wrd);
1734 gen_op_iwmmxt_set_mup();
1735 gen_op_iwmmxt_set_cup();
1736 break;
1737 case 0x200: /* WAND */
1738 wrd = (insn >> 12) & 0xf;
1739 rd0 = (insn >> 0) & 0xf;
1740 rd1 = (insn >> 16) & 0xf;
1741 gen_op_iwmmxt_movq_M0_wRn(rd0);
1742 gen_op_iwmmxt_andq_M0_wRn(rd1);
1743 gen_op_iwmmxt_setpsr_nz();
1744 gen_op_iwmmxt_movq_wRn_M0(wrd);
1745 gen_op_iwmmxt_set_mup();
1746 gen_op_iwmmxt_set_cup();
1747 break;
1748 case 0x810: case 0xa10: /* WMADD */
1749 wrd = (insn >> 12) & 0xf;
1750 rd0 = (insn >> 0) & 0xf;
1751 rd1 = (insn >> 16) & 0xf;
1752 gen_op_iwmmxt_movq_M0_wRn(rd0);
1753 if (insn & (1 << 21))
1754 gen_op_iwmmxt_maddsq_M0_wRn(rd1);
1755 else
1756 gen_op_iwmmxt_madduq_M0_wRn(rd1);
1757 gen_op_iwmmxt_movq_wRn_M0(wrd);
1758 gen_op_iwmmxt_set_mup();
1759 break;
1760 case 0x10e: case 0x50e: case 0x90e: case 0xd0e: /* WUNPCKIL */
1761 wrd = (insn >> 12) & 0xf;
1762 rd0 = (insn >> 16) & 0xf;
1763 rd1 = (insn >> 0) & 0xf;
1764 gen_op_iwmmxt_movq_M0_wRn(rd0);
1765 switch ((insn >> 22) & 3) {
1766 case 0:
1767 gen_op_iwmmxt_unpacklb_M0_wRn(rd1);
1768 break;
1769 case 1:
1770 gen_op_iwmmxt_unpacklw_M0_wRn(rd1);
1771 break;
1772 case 2:
1773 gen_op_iwmmxt_unpackll_M0_wRn(rd1);
1774 break;
1775 case 3:
1776 return 1;
1778 gen_op_iwmmxt_movq_wRn_M0(wrd);
1779 gen_op_iwmmxt_set_mup();
1780 gen_op_iwmmxt_set_cup();
1781 break;
1782 case 0x10c: case 0x50c: case 0x90c: case 0xd0c: /* WUNPCKIH */
1783 wrd = (insn >> 12) & 0xf;
1784 rd0 = (insn >> 16) & 0xf;
1785 rd1 = (insn >> 0) & 0xf;
1786 gen_op_iwmmxt_movq_M0_wRn(rd0);
1787 switch ((insn >> 22) & 3) {
1788 case 0:
1789 gen_op_iwmmxt_unpackhb_M0_wRn(rd1);
1790 break;
1791 case 1:
1792 gen_op_iwmmxt_unpackhw_M0_wRn(rd1);
1793 break;
1794 case 2:
1795 gen_op_iwmmxt_unpackhl_M0_wRn(rd1);
1796 break;
1797 case 3:
1798 return 1;
1800 gen_op_iwmmxt_movq_wRn_M0(wrd);
1801 gen_op_iwmmxt_set_mup();
1802 gen_op_iwmmxt_set_cup();
1803 break;
1804 case 0x012: case 0x112: case 0x412: case 0x512: /* WSAD */
1805 wrd = (insn >> 12) & 0xf;
1806 rd0 = (insn >> 16) & 0xf;
1807 rd1 = (insn >> 0) & 0xf;
1808 gen_op_iwmmxt_movq_M0_wRn(rd0);
1809 if (insn & (1 << 22))
1810 gen_op_iwmmxt_sadw_M0_wRn(rd1);
1811 else
1812 gen_op_iwmmxt_sadb_M0_wRn(rd1);
1813 if (!(insn & (1 << 20)))
1814 gen_op_iwmmxt_addl_M0_wRn(wrd);
1815 gen_op_iwmmxt_movq_wRn_M0(wrd);
1816 gen_op_iwmmxt_set_mup();
1817 break;
1818 case 0x010: case 0x110: case 0x210: case 0x310: /* WMUL */
1819 wrd = (insn >> 12) & 0xf;
1820 rd0 = (insn >> 16) & 0xf;
1821 rd1 = (insn >> 0) & 0xf;
1822 gen_op_iwmmxt_movq_M0_wRn(rd0);
1823 if (insn & (1 << 21)) {
1824 if (insn & (1 << 20))
1825 gen_op_iwmmxt_mulshw_M0_wRn(rd1);
1826 else
1827 gen_op_iwmmxt_mulslw_M0_wRn(rd1);
1828 } else {
1829 if (insn & (1 << 20))
1830 gen_op_iwmmxt_muluhw_M0_wRn(rd1);
1831 else
1832 gen_op_iwmmxt_mululw_M0_wRn(rd1);
1834 gen_op_iwmmxt_movq_wRn_M0(wrd);
1835 gen_op_iwmmxt_set_mup();
1836 break;
1837 case 0x410: case 0x510: case 0x610: case 0x710: /* WMAC */
1838 wrd = (insn >> 12) & 0xf;
1839 rd0 = (insn >> 16) & 0xf;
1840 rd1 = (insn >> 0) & 0xf;
1841 gen_op_iwmmxt_movq_M0_wRn(rd0);
1842 if (insn & (1 << 21))
1843 gen_op_iwmmxt_macsw_M0_wRn(rd1);
1844 else
1845 gen_op_iwmmxt_macuw_M0_wRn(rd1);
1846 if (!(insn & (1 << 20))) {
1847 iwmmxt_load_reg(cpu_V1, wrd);
1848 tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
1850 gen_op_iwmmxt_movq_wRn_M0(wrd);
1851 gen_op_iwmmxt_set_mup();
1852 break;
1853 case 0x006: case 0x406: case 0x806: case 0xc06: /* WCMPEQ */
1854 wrd = (insn >> 12) & 0xf;
1855 rd0 = (insn >> 16) & 0xf;
1856 rd1 = (insn >> 0) & 0xf;
1857 gen_op_iwmmxt_movq_M0_wRn(rd0);
1858 switch ((insn >> 22) & 3) {
1859 case 0:
1860 gen_op_iwmmxt_cmpeqb_M0_wRn(rd1);
1861 break;
1862 case 1:
1863 gen_op_iwmmxt_cmpeqw_M0_wRn(rd1);
1864 break;
1865 case 2:
1866 gen_op_iwmmxt_cmpeql_M0_wRn(rd1);
1867 break;
1868 case 3:
1869 return 1;
1871 gen_op_iwmmxt_movq_wRn_M0(wrd);
1872 gen_op_iwmmxt_set_mup();
1873 gen_op_iwmmxt_set_cup();
1874 break;
1875 case 0x800: case 0x900: case 0xc00: case 0xd00: /* WAVG2 */
1876 wrd = (insn >> 12) & 0xf;
1877 rd0 = (insn >> 16) & 0xf;
1878 rd1 = (insn >> 0) & 0xf;
1879 gen_op_iwmmxt_movq_M0_wRn(rd0);
1880 if (insn & (1 << 22)) {
1881 if (insn & (1 << 20))
1882 gen_op_iwmmxt_avgw1_M0_wRn(rd1);
1883 else
1884 gen_op_iwmmxt_avgw0_M0_wRn(rd1);
1885 } else {
1886 if (insn & (1 << 20))
1887 gen_op_iwmmxt_avgb1_M0_wRn(rd1);
1888 else
1889 gen_op_iwmmxt_avgb0_M0_wRn(rd1);
1891 gen_op_iwmmxt_movq_wRn_M0(wrd);
1892 gen_op_iwmmxt_set_mup();
1893 gen_op_iwmmxt_set_cup();
1894 break;
1895 case 0x802: case 0x902: case 0xa02: case 0xb02: /* WALIGNR */
1896 wrd = (insn >> 12) & 0xf;
1897 rd0 = (insn >> 16) & 0xf;
1898 rd1 = (insn >> 0) & 0xf;
1899 gen_op_iwmmxt_movq_M0_wRn(rd0);
1900 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCGR0 + ((insn >> 20) & 3));
1901 tcg_gen_andi_i32(tmp, tmp, 7);
1902 iwmmxt_load_reg(cpu_V1, rd1);
1903 gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1, tmp);
1904 tcg_temp_free_i32(tmp);
1905 gen_op_iwmmxt_movq_wRn_M0(wrd);
1906 gen_op_iwmmxt_set_mup();
1907 break;
1908 case 0x601: case 0x605: case 0x609: case 0x60d: /* TINSR */
1909 if (((insn >> 6) & 3) == 3)
1910 return 1;
1911 rd = (insn >> 12) & 0xf;
1912 wrd = (insn >> 16) & 0xf;
1913 tmp = load_reg(s, rd);
1914 gen_op_iwmmxt_movq_M0_wRn(wrd);
1915 switch ((insn >> 6) & 3) {
1916 case 0:
1917 tmp2 = tcg_const_i32(0xff);
1918 tmp3 = tcg_const_i32((insn & 7) << 3);
1919 break;
1920 case 1:
1921 tmp2 = tcg_const_i32(0xffff);
1922 tmp3 = tcg_const_i32((insn & 3) << 4);
1923 break;
1924 case 2:
1925 tmp2 = tcg_const_i32(0xffffffff);
1926 tmp3 = tcg_const_i32((insn & 1) << 5);
1927 break;
1928 default:
1929 TCGV_UNUSED_I32(tmp2);
1930 TCGV_UNUSED_I32(tmp3);
1932 gen_helper_iwmmxt_insr(cpu_M0, cpu_M0, tmp, tmp2, tmp3);
1933 tcg_temp_free_i32(tmp3);
1934 tcg_temp_free_i32(tmp2);
1935 tcg_temp_free_i32(tmp);
1936 gen_op_iwmmxt_movq_wRn_M0(wrd);
1937 gen_op_iwmmxt_set_mup();
1938 break;
1939 case 0x107: case 0x507: case 0x907: case 0xd07: /* TEXTRM */
1940 rd = (insn >> 12) & 0xf;
1941 wrd = (insn >> 16) & 0xf;
1942 if (rd == 15 || ((insn >> 22) & 3) == 3)
1943 return 1;
1944 gen_op_iwmmxt_movq_M0_wRn(wrd);
1945 tmp = tcg_temp_new_i32();
1946 switch ((insn >> 22) & 3) {
1947 case 0:
1948 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 7) << 3);
1949 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1950 if (insn & 8) {
1951 tcg_gen_ext8s_i32(tmp, tmp);
1952 } else {
1953 tcg_gen_andi_i32(tmp, tmp, 0xff);
1955 break;
1956 case 1:
1957 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 3) << 4);
1958 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1959 if (insn & 8) {
1960 tcg_gen_ext16s_i32(tmp, tmp);
1961 } else {
1962 tcg_gen_andi_i32(tmp, tmp, 0xffff);
1964 break;
1965 case 2:
1966 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 1) << 5);
1967 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1968 break;
1970 store_reg(s, rd, tmp);
1971 break;
1972 case 0x117: case 0x517: case 0x917: case 0xd17: /* TEXTRC */
1973 if ((insn & 0x000ff008) != 0x0003f000 || ((insn >> 22) & 3) == 3)
1974 return 1;
1975 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
1976 switch ((insn >> 22) & 3) {
1977 case 0:
1978 tcg_gen_shri_i32(tmp, tmp, ((insn & 7) << 2) + 0);
1979 break;
1980 case 1:
1981 tcg_gen_shri_i32(tmp, tmp, ((insn & 3) << 3) + 4);
1982 break;
1983 case 2:
1984 tcg_gen_shri_i32(tmp, tmp, ((insn & 1) << 4) + 12);
1985 break;
1987 tcg_gen_shli_i32(tmp, tmp, 28);
1988 gen_set_nzcv(tmp);
1989 tcg_temp_free_i32(tmp);
1990 break;
1991 case 0x401: case 0x405: case 0x409: case 0x40d: /* TBCST */
1992 if (((insn >> 6) & 3) == 3)
1993 return 1;
1994 rd = (insn >> 12) & 0xf;
1995 wrd = (insn >> 16) & 0xf;
1996 tmp = load_reg(s, rd);
1997 switch ((insn >> 6) & 3) {
1998 case 0:
1999 gen_helper_iwmmxt_bcstb(cpu_M0, tmp);
2000 break;
2001 case 1:
2002 gen_helper_iwmmxt_bcstw(cpu_M0, tmp);
2003 break;
2004 case 2:
2005 gen_helper_iwmmxt_bcstl(cpu_M0, tmp);
2006 break;
2008 tcg_temp_free_i32(tmp);
2009 gen_op_iwmmxt_movq_wRn_M0(wrd);
2010 gen_op_iwmmxt_set_mup();
2011 break;
2012 case 0x113: case 0x513: case 0x913: case 0xd13: /* TANDC */
2013 if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
2014 return 1;
2015 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
2016 tmp2 = tcg_temp_new_i32();
2017 tcg_gen_mov_i32(tmp2, tmp);
2018 switch ((insn >> 22) & 3) {
2019 case 0:
2020 for (i = 0; i < 7; i ++) {
2021 tcg_gen_shli_i32(tmp2, tmp2, 4);
2022 tcg_gen_and_i32(tmp, tmp, tmp2);
2024 break;
2025 case 1:
2026 for (i = 0; i < 3; i ++) {
2027 tcg_gen_shli_i32(tmp2, tmp2, 8);
2028 tcg_gen_and_i32(tmp, tmp, tmp2);
2030 break;
2031 case 2:
2032 tcg_gen_shli_i32(tmp2, tmp2, 16);
2033 tcg_gen_and_i32(tmp, tmp, tmp2);
2034 break;
2036 gen_set_nzcv(tmp);
2037 tcg_temp_free_i32(tmp2);
2038 tcg_temp_free_i32(tmp);
2039 break;
2040 case 0x01c: case 0x41c: case 0x81c: case 0xc1c: /* WACC */
2041 wrd = (insn >> 12) & 0xf;
2042 rd0 = (insn >> 16) & 0xf;
2043 gen_op_iwmmxt_movq_M0_wRn(rd0);
2044 switch ((insn >> 22) & 3) {
2045 case 0:
2046 gen_helper_iwmmxt_addcb(cpu_M0, cpu_M0);
2047 break;
2048 case 1:
2049 gen_helper_iwmmxt_addcw(cpu_M0, cpu_M0);
2050 break;
2051 case 2:
2052 gen_helper_iwmmxt_addcl(cpu_M0, cpu_M0);
2053 break;
2054 case 3:
2055 return 1;
2057 gen_op_iwmmxt_movq_wRn_M0(wrd);
2058 gen_op_iwmmxt_set_mup();
2059 break;
2060 case 0x115: case 0x515: case 0x915: case 0xd15: /* TORC */
2061 if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
2062 return 1;
2063 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
2064 tmp2 = tcg_temp_new_i32();
2065 tcg_gen_mov_i32(tmp2, tmp);
2066 switch ((insn >> 22) & 3) {
2067 case 0:
2068 for (i = 0; i < 7; i ++) {
2069 tcg_gen_shli_i32(tmp2, tmp2, 4);
2070 tcg_gen_or_i32(tmp, tmp, tmp2);
2072 break;
2073 case 1:
2074 for (i = 0; i < 3; i ++) {
2075 tcg_gen_shli_i32(tmp2, tmp2, 8);
2076 tcg_gen_or_i32(tmp, tmp, tmp2);
2078 break;
2079 case 2:
2080 tcg_gen_shli_i32(tmp2, tmp2, 16);
2081 tcg_gen_or_i32(tmp, tmp, tmp2);
2082 break;
2084 gen_set_nzcv(tmp);
2085 tcg_temp_free_i32(tmp2);
2086 tcg_temp_free_i32(tmp);
2087 break;
2088 case 0x103: case 0x503: case 0x903: case 0xd03: /* TMOVMSK */
2089 rd = (insn >> 12) & 0xf;
2090 rd0 = (insn >> 16) & 0xf;
2091 if ((insn & 0xf) != 0 || ((insn >> 22) & 3) == 3)
2092 return 1;
2093 gen_op_iwmmxt_movq_M0_wRn(rd0);
2094 tmp = tcg_temp_new_i32();
2095 switch ((insn >> 22) & 3) {
2096 case 0:
2097 gen_helper_iwmmxt_msbb(tmp, cpu_M0);
2098 break;
2099 case 1:
2100 gen_helper_iwmmxt_msbw(tmp, cpu_M0);
2101 break;
2102 case 2:
2103 gen_helper_iwmmxt_msbl(tmp, cpu_M0);
2104 break;
2106 store_reg(s, rd, tmp);
2107 break;
2108 case 0x106: case 0x306: case 0x506: case 0x706: /* WCMPGT */
2109 case 0x906: case 0xb06: case 0xd06: case 0xf06:
2110 wrd = (insn >> 12) & 0xf;
2111 rd0 = (insn >> 16) & 0xf;
2112 rd1 = (insn >> 0) & 0xf;
2113 gen_op_iwmmxt_movq_M0_wRn(rd0);
2114 switch ((insn >> 22) & 3) {
2115 case 0:
2116 if (insn & (1 << 21))
2117 gen_op_iwmmxt_cmpgtsb_M0_wRn(rd1);
2118 else
2119 gen_op_iwmmxt_cmpgtub_M0_wRn(rd1);
2120 break;
2121 case 1:
2122 if (insn & (1 << 21))
2123 gen_op_iwmmxt_cmpgtsw_M0_wRn(rd1);
2124 else
2125 gen_op_iwmmxt_cmpgtuw_M0_wRn(rd1);
2126 break;
2127 case 2:
2128 if (insn & (1 << 21))
2129 gen_op_iwmmxt_cmpgtsl_M0_wRn(rd1);
2130 else
2131 gen_op_iwmmxt_cmpgtul_M0_wRn(rd1);
2132 break;
2133 case 3:
2134 return 1;
2136 gen_op_iwmmxt_movq_wRn_M0(wrd);
2137 gen_op_iwmmxt_set_mup();
2138 gen_op_iwmmxt_set_cup();
2139 break;
2140 case 0x00e: case 0x20e: case 0x40e: case 0x60e: /* WUNPCKEL */
2141 case 0x80e: case 0xa0e: case 0xc0e: case 0xe0e:
2142 wrd = (insn >> 12) & 0xf;
2143 rd0 = (insn >> 16) & 0xf;
2144 gen_op_iwmmxt_movq_M0_wRn(rd0);
2145 switch ((insn >> 22) & 3) {
2146 case 0:
2147 if (insn & (1 << 21))
2148 gen_op_iwmmxt_unpacklsb_M0();
2149 else
2150 gen_op_iwmmxt_unpacklub_M0();
2151 break;
2152 case 1:
2153 if (insn & (1 << 21))
2154 gen_op_iwmmxt_unpacklsw_M0();
2155 else
2156 gen_op_iwmmxt_unpackluw_M0();
2157 break;
2158 case 2:
2159 if (insn & (1 << 21))
2160 gen_op_iwmmxt_unpacklsl_M0();
2161 else
2162 gen_op_iwmmxt_unpacklul_M0();
2163 break;
2164 case 3:
2165 return 1;
2167 gen_op_iwmmxt_movq_wRn_M0(wrd);
2168 gen_op_iwmmxt_set_mup();
2169 gen_op_iwmmxt_set_cup();
2170 break;
2171 case 0x00c: case 0x20c: case 0x40c: case 0x60c: /* WUNPCKEH */
2172 case 0x80c: case 0xa0c: case 0xc0c: case 0xe0c:
2173 wrd = (insn >> 12) & 0xf;
2174 rd0 = (insn >> 16) & 0xf;
2175 gen_op_iwmmxt_movq_M0_wRn(rd0);
2176 switch ((insn >> 22) & 3) {
2177 case 0:
2178 if (insn & (1 << 21))
2179 gen_op_iwmmxt_unpackhsb_M0();
2180 else
2181 gen_op_iwmmxt_unpackhub_M0();
2182 break;
2183 case 1:
2184 if (insn & (1 << 21))
2185 gen_op_iwmmxt_unpackhsw_M0();
2186 else
2187 gen_op_iwmmxt_unpackhuw_M0();
2188 break;
2189 case 2:
2190 if (insn & (1 << 21))
2191 gen_op_iwmmxt_unpackhsl_M0();
2192 else
2193 gen_op_iwmmxt_unpackhul_M0();
2194 break;
2195 case 3:
2196 return 1;
2198 gen_op_iwmmxt_movq_wRn_M0(wrd);
2199 gen_op_iwmmxt_set_mup();
2200 gen_op_iwmmxt_set_cup();
2201 break;
2202 case 0x204: case 0x604: case 0xa04: case 0xe04: /* WSRL */
2203 case 0x214: case 0x614: case 0xa14: case 0xe14:
2204 if (((insn >> 22) & 3) == 0)
2205 return 1;
2206 wrd = (insn >> 12) & 0xf;
2207 rd0 = (insn >> 16) & 0xf;
2208 gen_op_iwmmxt_movq_M0_wRn(rd0);
2209 tmp = tcg_temp_new_i32();
2210 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2211 tcg_temp_free_i32(tmp);
2212 return 1;
2214 switch ((insn >> 22) & 3) {
2215 case 1:
2216 gen_helper_iwmmxt_srlw(cpu_M0, cpu_env, cpu_M0, tmp);
2217 break;
2218 case 2:
2219 gen_helper_iwmmxt_srll(cpu_M0, cpu_env, cpu_M0, tmp);
2220 break;
2221 case 3:
2222 gen_helper_iwmmxt_srlq(cpu_M0, cpu_env, cpu_M0, tmp);
2223 break;
2225 tcg_temp_free_i32(tmp);
2226 gen_op_iwmmxt_movq_wRn_M0(wrd);
2227 gen_op_iwmmxt_set_mup();
2228 gen_op_iwmmxt_set_cup();
2229 break;
2230 case 0x004: case 0x404: case 0x804: case 0xc04: /* WSRA */
2231 case 0x014: case 0x414: case 0x814: case 0xc14:
2232 if (((insn >> 22) & 3) == 0)
2233 return 1;
2234 wrd = (insn >> 12) & 0xf;
2235 rd0 = (insn >> 16) & 0xf;
2236 gen_op_iwmmxt_movq_M0_wRn(rd0);
2237 tmp = tcg_temp_new_i32();
2238 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2239 tcg_temp_free_i32(tmp);
2240 return 1;
2242 switch ((insn >> 22) & 3) {
2243 case 1:
2244 gen_helper_iwmmxt_sraw(cpu_M0, cpu_env, cpu_M0, tmp);
2245 break;
2246 case 2:
2247 gen_helper_iwmmxt_sral(cpu_M0, cpu_env, cpu_M0, tmp);
2248 break;
2249 case 3:
2250 gen_helper_iwmmxt_sraq(cpu_M0, cpu_env, cpu_M0, tmp);
2251 break;
2253 tcg_temp_free_i32(tmp);
2254 gen_op_iwmmxt_movq_wRn_M0(wrd);
2255 gen_op_iwmmxt_set_mup();
2256 gen_op_iwmmxt_set_cup();
2257 break;
2258 case 0x104: case 0x504: case 0x904: case 0xd04: /* WSLL */
2259 case 0x114: case 0x514: case 0x914: case 0xd14:
2260 if (((insn >> 22) & 3) == 0)
2261 return 1;
2262 wrd = (insn >> 12) & 0xf;
2263 rd0 = (insn >> 16) & 0xf;
2264 gen_op_iwmmxt_movq_M0_wRn(rd0);
2265 tmp = tcg_temp_new_i32();
2266 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2267 tcg_temp_free_i32(tmp);
2268 return 1;
2270 switch ((insn >> 22) & 3) {
2271 case 1:
2272 gen_helper_iwmmxt_sllw(cpu_M0, cpu_env, cpu_M0, tmp);
2273 break;
2274 case 2:
2275 gen_helper_iwmmxt_slll(cpu_M0, cpu_env, cpu_M0, tmp);
2276 break;
2277 case 3:
2278 gen_helper_iwmmxt_sllq(cpu_M0, cpu_env, cpu_M0, tmp);
2279 break;
2281 tcg_temp_free_i32(tmp);
2282 gen_op_iwmmxt_movq_wRn_M0(wrd);
2283 gen_op_iwmmxt_set_mup();
2284 gen_op_iwmmxt_set_cup();
2285 break;
2286 case 0x304: case 0x704: case 0xb04: case 0xf04: /* WROR */
2287 case 0x314: case 0x714: case 0xb14: case 0xf14:
2288 if (((insn >> 22) & 3) == 0)
2289 return 1;
2290 wrd = (insn >> 12) & 0xf;
2291 rd0 = (insn >> 16) & 0xf;
2292 gen_op_iwmmxt_movq_M0_wRn(rd0);
2293 tmp = tcg_temp_new_i32();
2294 switch ((insn >> 22) & 3) {
2295 case 1:
2296 if (gen_iwmmxt_shift(insn, 0xf, tmp)) {
2297 tcg_temp_free_i32(tmp);
2298 return 1;
2300 gen_helper_iwmmxt_rorw(cpu_M0, cpu_env, cpu_M0, tmp);
2301 break;
2302 case 2:
2303 if (gen_iwmmxt_shift(insn, 0x1f, tmp)) {
2304 tcg_temp_free_i32(tmp);
2305 return 1;
2307 gen_helper_iwmmxt_rorl(cpu_M0, cpu_env, cpu_M0, tmp);
2308 break;
2309 case 3:
2310 if (gen_iwmmxt_shift(insn, 0x3f, tmp)) {
2311 tcg_temp_free_i32(tmp);
2312 return 1;
2314 gen_helper_iwmmxt_rorq(cpu_M0, cpu_env, cpu_M0, tmp);
2315 break;
2317 tcg_temp_free_i32(tmp);
2318 gen_op_iwmmxt_movq_wRn_M0(wrd);
2319 gen_op_iwmmxt_set_mup();
2320 gen_op_iwmmxt_set_cup();
2321 break;
2322 case 0x116: case 0x316: case 0x516: case 0x716: /* WMIN */
2323 case 0x916: case 0xb16: case 0xd16: case 0xf16:
2324 wrd = (insn >> 12) & 0xf;
2325 rd0 = (insn >> 16) & 0xf;
2326 rd1 = (insn >> 0) & 0xf;
2327 gen_op_iwmmxt_movq_M0_wRn(rd0);
2328 switch ((insn >> 22) & 3) {
2329 case 0:
2330 if (insn & (1 << 21))
2331 gen_op_iwmmxt_minsb_M0_wRn(rd1);
2332 else
2333 gen_op_iwmmxt_minub_M0_wRn(rd1);
2334 break;
2335 case 1:
2336 if (insn & (1 << 21))
2337 gen_op_iwmmxt_minsw_M0_wRn(rd1);
2338 else
2339 gen_op_iwmmxt_minuw_M0_wRn(rd1);
2340 break;
2341 case 2:
2342 if (insn & (1 << 21))
2343 gen_op_iwmmxt_minsl_M0_wRn(rd1);
2344 else
2345 gen_op_iwmmxt_minul_M0_wRn(rd1);
2346 break;
2347 case 3:
2348 return 1;
2350 gen_op_iwmmxt_movq_wRn_M0(wrd);
2351 gen_op_iwmmxt_set_mup();
2352 break;
2353 case 0x016: case 0x216: case 0x416: case 0x616: /* WMAX */
2354 case 0x816: case 0xa16: case 0xc16: case 0xe16:
2355 wrd = (insn >> 12) & 0xf;
2356 rd0 = (insn >> 16) & 0xf;
2357 rd1 = (insn >> 0) & 0xf;
2358 gen_op_iwmmxt_movq_M0_wRn(rd0);
2359 switch ((insn >> 22) & 3) {
2360 case 0:
2361 if (insn & (1 << 21))
2362 gen_op_iwmmxt_maxsb_M0_wRn(rd1);
2363 else
2364 gen_op_iwmmxt_maxub_M0_wRn(rd1);
2365 break;
2366 case 1:
2367 if (insn & (1 << 21))
2368 gen_op_iwmmxt_maxsw_M0_wRn(rd1);
2369 else
2370 gen_op_iwmmxt_maxuw_M0_wRn(rd1);
2371 break;
2372 case 2:
2373 if (insn & (1 << 21))
2374 gen_op_iwmmxt_maxsl_M0_wRn(rd1);
2375 else
2376 gen_op_iwmmxt_maxul_M0_wRn(rd1);
2377 break;
2378 case 3:
2379 return 1;
2381 gen_op_iwmmxt_movq_wRn_M0(wrd);
2382 gen_op_iwmmxt_set_mup();
2383 break;
2384 case 0x002: case 0x102: case 0x202: case 0x302: /* WALIGNI */
2385 case 0x402: case 0x502: case 0x602: case 0x702:
2386 wrd = (insn >> 12) & 0xf;
2387 rd0 = (insn >> 16) & 0xf;
2388 rd1 = (insn >> 0) & 0xf;
2389 gen_op_iwmmxt_movq_M0_wRn(rd0);
2390 tmp = tcg_const_i32((insn >> 20) & 3);
2391 iwmmxt_load_reg(cpu_V1, rd1);
2392 gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1, tmp);
2393 tcg_temp_free_i32(tmp);
2394 gen_op_iwmmxt_movq_wRn_M0(wrd);
2395 gen_op_iwmmxt_set_mup();
2396 break;
2397 case 0x01a: case 0x11a: case 0x21a: case 0x31a: /* WSUB */
2398 case 0x41a: case 0x51a: case 0x61a: case 0x71a:
2399 case 0x81a: case 0x91a: case 0xa1a: case 0xb1a:
2400 case 0xc1a: case 0xd1a: case 0xe1a: case 0xf1a:
2401 wrd = (insn >> 12) & 0xf;
2402 rd0 = (insn >> 16) & 0xf;
2403 rd1 = (insn >> 0) & 0xf;
2404 gen_op_iwmmxt_movq_M0_wRn(rd0);
2405 switch ((insn >> 20) & 0xf) {
2406 case 0x0:
2407 gen_op_iwmmxt_subnb_M0_wRn(rd1);
2408 break;
2409 case 0x1:
2410 gen_op_iwmmxt_subub_M0_wRn(rd1);
2411 break;
2412 case 0x3:
2413 gen_op_iwmmxt_subsb_M0_wRn(rd1);
2414 break;
2415 case 0x4:
2416 gen_op_iwmmxt_subnw_M0_wRn(rd1);
2417 break;
2418 case 0x5:
2419 gen_op_iwmmxt_subuw_M0_wRn(rd1);
2420 break;
2421 case 0x7:
2422 gen_op_iwmmxt_subsw_M0_wRn(rd1);
2423 break;
2424 case 0x8:
2425 gen_op_iwmmxt_subnl_M0_wRn(rd1);
2426 break;
2427 case 0x9:
2428 gen_op_iwmmxt_subul_M0_wRn(rd1);
2429 break;
2430 case 0xb:
2431 gen_op_iwmmxt_subsl_M0_wRn(rd1);
2432 break;
2433 default:
2434 return 1;
2436 gen_op_iwmmxt_movq_wRn_M0(wrd);
2437 gen_op_iwmmxt_set_mup();
2438 gen_op_iwmmxt_set_cup();
2439 break;
2440 case 0x01e: case 0x11e: case 0x21e: case 0x31e: /* WSHUFH */
2441 case 0x41e: case 0x51e: case 0x61e: case 0x71e:
2442 case 0x81e: case 0x91e: case 0xa1e: case 0xb1e:
2443 case 0xc1e: case 0xd1e: case 0xe1e: case 0xf1e:
2444 wrd = (insn >> 12) & 0xf;
2445 rd0 = (insn >> 16) & 0xf;
2446 gen_op_iwmmxt_movq_M0_wRn(rd0);
2447 tmp = tcg_const_i32(((insn >> 16) & 0xf0) | (insn & 0x0f));
2448 gen_helper_iwmmxt_shufh(cpu_M0, cpu_env, cpu_M0, tmp);
2449 tcg_temp_free_i32(tmp);
2450 gen_op_iwmmxt_movq_wRn_M0(wrd);
2451 gen_op_iwmmxt_set_mup();
2452 gen_op_iwmmxt_set_cup();
2453 break;
2454 case 0x018: case 0x118: case 0x218: case 0x318: /* WADD */
2455 case 0x418: case 0x518: case 0x618: case 0x718:
2456 case 0x818: case 0x918: case 0xa18: case 0xb18:
2457 case 0xc18: case 0xd18: case 0xe18: case 0xf18:
2458 wrd = (insn >> 12) & 0xf;
2459 rd0 = (insn >> 16) & 0xf;
2460 rd1 = (insn >> 0) & 0xf;
2461 gen_op_iwmmxt_movq_M0_wRn(rd0);
2462 switch ((insn >> 20) & 0xf) {
2463 case 0x0:
2464 gen_op_iwmmxt_addnb_M0_wRn(rd1);
2465 break;
2466 case 0x1:
2467 gen_op_iwmmxt_addub_M0_wRn(rd1);
2468 break;
2469 case 0x3:
2470 gen_op_iwmmxt_addsb_M0_wRn(rd1);
2471 break;
2472 case 0x4:
2473 gen_op_iwmmxt_addnw_M0_wRn(rd1);
2474 break;
2475 case 0x5:
2476 gen_op_iwmmxt_adduw_M0_wRn(rd1);
2477 break;
2478 case 0x7:
2479 gen_op_iwmmxt_addsw_M0_wRn(rd1);
2480 break;
2481 case 0x8:
2482 gen_op_iwmmxt_addnl_M0_wRn(rd1);
2483 break;
2484 case 0x9:
2485 gen_op_iwmmxt_addul_M0_wRn(rd1);
2486 break;
2487 case 0xb:
2488 gen_op_iwmmxt_addsl_M0_wRn(rd1);
2489 break;
2490 default:
2491 return 1;
2493 gen_op_iwmmxt_movq_wRn_M0(wrd);
2494 gen_op_iwmmxt_set_mup();
2495 gen_op_iwmmxt_set_cup();
2496 break;
2497 case 0x008: case 0x108: case 0x208: case 0x308: /* WPACK */
2498 case 0x408: case 0x508: case 0x608: case 0x708:
2499 case 0x808: case 0x908: case 0xa08: case 0xb08:
2500 case 0xc08: case 0xd08: case 0xe08: case 0xf08:
2501 if (!(insn & (1 << 20)) || ((insn >> 22) & 3) == 0)
2502 return 1;
2503 wrd = (insn >> 12) & 0xf;
2504 rd0 = (insn >> 16) & 0xf;
2505 rd1 = (insn >> 0) & 0xf;
2506 gen_op_iwmmxt_movq_M0_wRn(rd0);
2507 switch ((insn >> 22) & 3) {
2508 case 1:
2509 if (insn & (1 << 21))
2510 gen_op_iwmmxt_packsw_M0_wRn(rd1);
2511 else
2512 gen_op_iwmmxt_packuw_M0_wRn(rd1);
2513 break;
2514 case 2:
2515 if (insn & (1 << 21))
2516 gen_op_iwmmxt_packsl_M0_wRn(rd1);
2517 else
2518 gen_op_iwmmxt_packul_M0_wRn(rd1);
2519 break;
2520 case 3:
2521 if (insn & (1 << 21))
2522 gen_op_iwmmxt_packsq_M0_wRn(rd1);
2523 else
2524 gen_op_iwmmxt_packuq_M0_wRn(rd1);
2525 break;
2527 gen_op_iwmmxt_movq_wRn_M0(wrd);
2528 gen_op_iwmmxt_set_mup();
2529 gen_op_iwmmxt_set_cup();
2530 break;
2531 case 0x201: case 0x203: case 0x205: case 0x207:
2532 case 0x209: case 0x20b: case 0x20d: case 0x20f:
2533 case 0x211: case 0x213: case 0x215: case 0x217:
2534 case 0x219: case 0x21b: case 0x21d: case 0x21f:
2535 wrd = (insn >> 5) & 0xf;
2536 rd0 = (insn >> 12) & 0xf;
2537 rd1 = (insn >> 0) & 0xf;
2538 if (rd0 == 0xf || rd1 == 0xf)
2539 return 1;
2540 gen_op_iwmmxt_movq_M0_wRn(wrd);
2541 tmp = load_reg(s, rd0);
2542 tmp2 = load_reg(s, rd1);
2543 switch ((insn >> 16) & 0xf) {
2544 case 0x0: /* TMIA */
2545 gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2546 break;
2547 case 0x8: /* TMIAPH */
2548 gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2549 break;
2550 case 0xc: case 0xd: case 0xe: case 0xf: /* TMIAxy */
2551 if (insn & (1 << 16))
2552 tcg_gen_shri_i32(tmp, tmp, 16);
2553 if (insn & (1 << 17))
2554 tcg_gen_shri_i32(tmp2, tmp2, 16);
2555 gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2556 break;
2557 default:
2558 tcg_temp_free_i32(tmp2);
2559 tcg_temp_free_i32(tmp);
2560 return 1;
2562 tcg_temp_free_i32(tmp2);
2563 tcg_temp_free_i32(tmp);
2564 gen_op_iwmmxt_movq_wRn_M0(wrd);
2565 gen_op_iwmmxt_set_mup();
2566 break;
2567 default:
2568 return 1;
2571 return 0;
2574 /* Disassemble an XScale DSP instruction. Returns nonzero if an error occurred
2575 (ie. an undefined instruction). */
2576 static int disas_dsp_insn(DisasContext *s, uint32_t insn)
2578 int acc, rd0, rd1, rdhi, rdlo;
2579 TCGv_i32 tmp, tmp2;
2581 if ((insn & 0x0ff00f10) == 0x0e200010) {
2582 /* Multiply with Internal Accumulate Format */
2583 rd0 = (insn >> 12) & 0xf;
2584 rd1 = insn & 0xf;
2585 acc = (insn >> 5) & 7;
2587 if (acc != 0)
2588 return 1;
2590 tmp = load_reg(s, rd0);
2591 tmp2 = load_reg(s, rd1);
2592 switch ((insn >> 16) & 0xf) {
2593 case 0x0: /* MIA */
2594 gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2595 break;
2596 case 0x8: /* MIAPH */
2597 gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2598 break;
2599 case 0xc: /* MIABB */
2600 case 0xd: /* MIABT */
2601 case 0xe: /* MIATB */
2602 case 0xf: /* MIATT */
2603 if (insn & (1 << 16))
2604 tcg_gen_shri_i32(tmp, tmp, 16);
2605 if (insn & (1 << 17))
2606 tcg_gen_shri_i32(tmp2, tmp2, 16);
2607 gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2608 break;
2609 default:
2610 return 1;
2612 tcg_temp_free_i32(tmp2);
2613 tcg_temp_free_i32(tmp);
2615 gen_op_iwmmxt_movq_wRn_M0(acc);
2616 return 0;
2619 if ((insn & 0x0fe00ff8) == 0x0c400000) {
2620 /* Internal Accumulator Access Format */
2621 rdhi = (insn >> 16) & 0xf;
2622 rdlo = (insn >> 12) & 0xf;
2623 acc = insn & 7;
2625 if (acc != 0)
2626 return 1;
2628 if (insn & ARM_CP_RW_BIT) { /* MRA */
2629 iwmmxt_load_reg(cpu_V0, acc);
2630 tcg_gen_extrl_i64_i32(cpu_R[rdlo], cpu_V0);
2631 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
2632 tcg_gen_extrl_i64_i32(cpu_R[rdhi], cpu_V0);
2633 tcg_gen_andi_i32(cpu_R[rdhi], cpu_R[rdhi], (1 << (40 - 32)) - 1);
2634 } else { /* MAR */
2635 tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
2636 iwmmxt_store_reg(cpu_V0, acc);
2638 return 0;
2641 return 1;
2644 #define VFP_REG_SHR(x, n) (((n) > 0) ? (x) >> (n) : (x) << -(n))
2645 #define VFP_SREG(insn, bigbit, smallbit) \
2646 ((VFP_REG_SHR(insn, bigbit - 1) & 0x1e) | (((insn) >> (smallbit)) & 1))
2647 #define VFP_DREG(reg, insn, bigbit, smallbit) do { \
2648 if (arm_dc_feature(s, ARM_FEATURE_VFP3)) { \
2649 reg = (((insn) >> (bigbit)) & 0x0f) \
2650 | (((insn) >> ((smallbit) - 4)) & 0x10); \
2651 } else { \
2652 if (insn & (1 << (smallbit))) \
2653 return 1; \
2654 reg = ((insn) >> (bigbit)) & 0x0f; \
2655 }} while (0)
2657 #define VFP_SREG_D(insn) VFP_SREG(insn, 12, 22)
2658 #define VFP_DREG_D(reg, insn) VFP_DREG(reg, insn, 12, 22)
2659 #define VFP_SREG_N(insn) VFP_SREG(insn, 16, 7)
2660 #define VFP_DREG_N(reg, insn) VFP_DREG(reg, insn, 16, 7)
2661 #define VFP_SREG_M(insn) VFP_SREG(insn, 0, 5)
2662 #define VFP_DREG_M(reg, insn) VFP_DREG(reg, insn, 0, 5)
2664 /* Move between integer and VFP cores. */
2665 static TCGv_i32 gen_vfp_mrs(void)
2667 TCGv_i32 tmp = tcg_temp_new_i32();
2668 tcg_gen_mov_i32(tmp, cpu_F0s);
2669 return tmp;
2672 static void gen_vfp_msr(TCGv_i32 tmp)
2674 tcg_gen_mov_i32(cpu_F0s, tmp);
2675 tcg_temp_free_i32(tmp);
2678 static void gen_neon_dup_u8(TCGv_i32 var, int shift)
2680 TCGv_i32 tmp = tcg_temp_new_i32();
2681 if (shift)
2682 tcg_gen_shri_i32(var, var, shift);
2683 tcg_gen_ext8u_i32(var, var);
2684 tcg_gen_shli_i32(tmp, var, 8);
2685 tcg_gen_or_i32(var, var, tmp);
2686 tcg_gen_shli_i32(tmp, var, 16);
2687 tcg_gen_or_i32(var, var, tmp);
2688 tcg_temp_free_i32(tmp);
2691 static void gen_neon_dup_low16(TCGv_i32 var)
2693 TCGv_i32 tmp = tcg_temp_new_i32();
2694 tcg_gen_ext16u_i32(var, var);
2695 tcg_gen_shli_i32(tmp, var, 16);
2696 tcg_gen_or_i32(var, var, tmp);
2697 tcg_temp_free_i32(tmp);
2700 static void gen_neon_dup_high16(TCGv_i32 var)
2702 TCGv_i32 tmp = tcg_temp_new_i32();
2703 tcg_gen_andi_i32(var, var, 0xffff0000);
2704 tcg_gen_shri_i32(tmp, var, 16);
2705 tcg_gen_or_i32(var, var, tmp);
2706 tcg_temp_free_i32(tmp);
2709 static TCGv_i32 gen_load_and_replicate(DisasContext *s, TCGv_i32 addr, int size)
2711 /* Load a single Neon element and replicate into a 32 bit TCG reg */
2712 TCGv_i32 tmp = tcg_temp_new_i32();
2713 switch (size) {
2714 case 0:
2715 gen_aa32_ld8u(tmp, addr, get_mem_index(s));
2716 gen_neon_dup_u8(tmp, 0);
2717 break;
2718 case 1:
2719 gen_aa32_ld16u(tmp, addr, get_mem_index(s));
2720 gen_neon_dup_low16(tmp);
2721 break;
2722 case 2:
2723 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
2724 break;
2725 default: /* Avoid compiler warnings. */
2726 abort();
2728 return tmp;
2731 static int handle_vsel(uint32_t insn, uint32_t rd, uint32_t rn, uint32_t rm,
2732 uint32_t dp)
2734 uint32_t cc = extract32(insn, 20, 2);
2736 if (dp) {
2737 TCGv_i64 frn, frm, dest;
2738 TCGv_i64 tmp, zero, zf, nf, vf;
2740 zero = tcg_const_i64(0);
2742 frn = tcg_temp_new_i64();
2743 frm = tcg_temp_new_i64();
2744 dest = tcg_temp_new_i64();
2746 zf = tcg_temp_new_i64();
2747 nf = tcg_temp_new_i64();
2748 vf = tcg_temp_new_i64();
2750 tcg_gen_extu_i32_i64(zf, cpu_ZF);
2751 tcg_gen_ext_i32_i64(nf, cpu_NF);
2752 tcg_gen_ext_i32_i64(vf, cpu_VF);
2754 tcg_gen_ld_f64(frn, cpu_env, vfp_reg_offset(dp, rn));
2755 tcg_gen_ld_f64(frm, cpu_env, vfp_reg_offset(dp, rm));
2756 switch (cc) {
2757 case 0: /* eq: Z */
2758 tcg_gen_movcond_i64(TCG_COND_EQ, dest, zf, zero,
2759 frn, frm);
2760 break;
2761 case 1: /* vs: V */
2762 tcg_gen_movcond_i64(TCG_COND_LT, dest, vf, zero,
2763 frn, frm);
2764 break;
2765 case 2: /* ge: N == V -> N ^ V == 0 */
2766 tmp = tcg_temp_new_i64();
2767 tcg_gen_xor_i64(tmp, vf, nf);
2768 tcg_gen_movcond_i64(TCG_COND_GE, dest, tmp, zero,
2769 frn, frm);
2770 tcg_temp_free_i64(tmp);
2771 break;
2772 case 3: /* gt: !Z && N == V */
2773 tcg_gen_movcond_i64(TCG_COND_NE, dest, zf, zero,
2774 frn, frm);
2775 tmp = tcg_temp_new_i64();
2776 tcg_gen_xor_i64(tmp, vf, nf);
2777 tcg_gen_movcond_i64(TCG_COND_GE, dest, tmp, zero,
2778 dest, frm);
2779 tcg_temp_free_i64(tmp);
2780 break;
2782 tcg_gen_st_f64(dest, cpu_env, vfp_reg_offset(dp, rd));
2783 tcg_temp_free_i64(frn);
2784 tcg_temp_free_i64(frm);
2785 tcg_temp_free_i64(dest);
2787 tcg_temp_free_i64(zf);
2788 tcg_temp_free_i64(nf);
2789 tcg_temp_free_i64(vf);
2791 tcg_temp_free_i64(zero);
2792 } else {
2793 TCGv_i32 frn, frm, dest;
2794 TCGv_i32 tmp, zero;
2796 zero = tcg_const_i32(0);
2798 frn = tcg_temp_new_i32();
2799 frm = tcg_temp_new_i32();
2800 dest = tcg_temp_new_i32();
2801 tcg_gen_ld_f32(frn, cpu_env, vfp_reg_offset(dp, rn));
2802 tcg_gen_ld_f32(frm, cpu_env, vfp_reg_offset(dp, rm));
2803 switch (cc) {
2804 case 0: /* eq: Z */
2805 tcg_gen_movcond_i32(TCG_COND_EQ, dest, cpu_ZF, zero,
2806 frn, frm);
2807 break;
2808 case 1: /* vs: V */
2809 tcg_gen_movcond_i32(TCG_COND_LT, dest, cpu_VF, zero,
2810 frn, frm);
2811 break;
2812 case 2: /* ge: N == V -> N ^ V == 0 */
2813 tmp = tcg_temp_new_i32();
2814 tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
2815 tcg_gen_movcond_i32(TCG_COND_GE, dest, tmp, zero,
2816 frn, frm);
2817 tcg_temp_free_i32(tmp);
2818 break;
2819 case 3: /* gt: !Z && N == V */
2820 tcg_gen_movcond_i32(TCG_COND_NE, dest, cpu_ZF, zero,
2821 frn, frm);
2822 tmp = tcg_temp_new_i32();
2823 tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
2824 tcg_gen_movcond_i32(TCG_COND_GE, dest, tmp, zero,
2825 dest, frm);
2826 tcg_temp_free_i32(tmp);
2827 break;
2829 tcg_gen_st_f32(dest, cpu_env, vfp_reg_offset(dp, rd));
2830 tcg_temp_free_i32(frn);
2831 tcg_temp_free_i32(frm);
2832 tcg_temp_free_i32(dest);
2834 tcg_temp_free_i32(zero);
2837 return 0;
2840 static int handle_vminmaxnm(uint32_t insn, uint32_t rd, uint32_t rn,
2841 uint32_t rm, uint32_t dp)
2843 uint32_t vmin = extract32(insn, 6, 1);
2844 TCGv_ptr fpst = get_fpstatus_ptr(0);
2846 if (dp) {
2847 TCGv_i64 frn, frm, dest;
2849 frn = tcg_temp_new_i64();
2850 frm = tcg_temp_new_i64();
2851 dest = tcg_temp_new_i64();
2853 tcg_gen_ld_f64(frn, cpu_env, vfp_reg_offset(dp, rn));
2854 tcg_gen_ld_f64(frm, cpu_env, vfp_reg_offset(dp, rm));
2855 if (vmin) {
2856 gen_helper_vfp_minnumd(dest, frn, frm, fpst);
2857 } else {
2858 gen_helper_vfp_maxnumd(dest, frn, frm, fpst);
2860 tcg_gen_st_f64(dest, cpu_env, vfp_reg_offset(dp, rd));
2861 tcg_temp_free_i64(frn);
2862 tcg_temp_free_i64(frm);
2863 tcg_temp_free_i64(dest);
2864 } else {
2865 TCGv_i32 frn, frm, dest;
2867 frn = tcg_temp_new_i32();
2868 frm = tcg_temp_new_i32();
2869 dest = tcg_temp_new_i32();
2871 tcg_gen_ld_f32(frn, cpu_env, vfp_reg_offset(dp, rn));
2872 tcg_gen_ld_f32(frm, cpu_env, vfp_reg_offset(dp, rm));
2873 if (vmin) {
2874 gen_helper_vfp_minnums(dest, frn, frm, fpst);
2875 } else {
2876 gen_helper_vfp_maxnums(dest, frn, frm, fpst);
2878 tcg_gen_st_f32(dest, cpu_env, vfp_reg_offset(dp, rd));
2879 tcg_temp_free_i32(frn);
2880 tcg_temp_free_i32(frm);
2881 tcg_temp_free_i32(dest);
2884 tcg_temp_free_ptr(fpst);
2885 return 0;
2888 static int handle_vrint(uint32_t insn, uint32_t rd, uint32_t rm, uint32_t dp,
2889 int rounding)
2891 TCGv_ptr fpst = get_fpstatus_ptr(0);
2892 TCGv_i32 tcg_rmode;
2894 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rounding));
2895 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
2897 if (dp) {
2898 TCGv_i64 tcg_op;
2899 TCGv_i64 tcg_res;
2900 tcg_op = tcg_temp_new_i64();
2901 tcg_res = tcg_temp_new_i64();
2902 tcg_gen_ld_f64(tcg_op, cpu_env, vfp_reg_offset(dp, rm));
2903 gen_helper_rintd(tcg_res, tcg_op, fpst);
2904 tcg_gen_st_f64(tcg_res, cpu_env, vfp_reg_offset(dp, rd));
2905 tcg_temp_free_i64(tcg_op);
2906 tcg_temp_free_i64(tcg_res);
2907 } else {
2908 TCGv_i32 tcg_op;
2909 TCGv_i32 tcg_res;
2910 tcg_op = tcg_temp_new_i32();
2911 tcg_res = tcg_temp_new_i32();
2912 tcg_gen_ld_f32(tcg_op, cpu_env, vfp_reg_offset(dp, rm));
2913 gen_helper_rints(tcg_res, tcg_op, fpst);
2914 tcg_gen_st_f32(tcg_res, cpu_env, vfp_reg_offset(dp, rd));
2915 tcg_temp_free_i32(tcg_op);
2916 tcg_temp_free_i32(tcg_res);
2919 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
2920 tcg_temp_free_i32(tcg_rmode);
2922 tcg_temp_free_ptr(fpst);
2923 return 0;
2926 static int handle_vcvt(uint32_t insn, uint32_t rd, uint32_t rm, uint32_t dp,
2927 int rounding)
2929 bool is_signed = extract32(insn, 7, 1);
2930 TCGv_ptr fpst = get_fpstatus_ptr(0);
2931 TCGv_i32 tcg_rmode, tcg_shift;
2933 tcg_shift = tcg_const_i32(0);
2935 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rounding));
2936 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
2938 if (dp) {
2939 TCGv_i64 tcg_double, tcg_res;
2940 TCGv_i32 tcg_tmp;
2941 /* Rd is encoded as a single precision register even when the source
2942 * is double precision.
2944 rd = ((rd << 1) & 0x1e) | ((rd >> 4) & 0x1);
2945 tcg_double = tcg_temp_new_i64();
2946 tcg_res = tcg_temp_new_i64();
2947 tcg_tmp = tcg_temp_new_i32();
2948 tcg_gen_ld_f64(tcg_double, cpu_env, vfp_reg_offset(1, rm));
2949 if (is_signed) {
2950 gen_helper_vfp_tosld(tcg_res, tcg_double, tcg_shift, fpst);
2951 } else {
2952 gen_helper_vfp_tould(tcg_res, tcg_double, tcg_shift, fpst);
2954 tcg_gen_extrl_i64_i32(tcg_tmp, tcg_res);
2955 tcg_gen_st_f32(tcg_tmp, cpu_env, vfp_reg_offset(0, rd));
2956 tcg_temp_free_i32(tcg_tmp);
2957 tcg_temp_free_i64(tcg_res);
2958 tcg_temp_free_i64(tcg_double);
2959 } else {
2960 TCGv_i32 tcg_single, tcg_res;
2961 tcg_single = tcg_temp_new_i32();
2962 tcg_res = tcg_temp_new_i32();
2963 tcg_gen_ld_f32(tcg_single, cpu_env, vfp_reg_offset(0, rm));
2964 if (is_signed) {
2965 gen_helper_vfp_tosls(tcg_res, tcg_single, tcg_shift, fpst);
2966 } else {
2967 gen_helper_vfp_touls(tcg_res, tcg_single, tcg_shift, fpst);
2969 tcg_gen_st_f32(tcg_res, cpu_env, vfp_reg_offset(0, rd));
2970 tcg_temp_free_i32(tcg_res);
2971 tcg_temp_free_i32(tcg_single);
2974 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
2975 tcg_temp_free_i32(tcg_rmode);
2977 tcg_temp_free_i32(tcg_shift);
2979 tcg_temp_free_ptr(fpst);
2981 return 0;
2984 /* Table for converting the most common AArch32 encoding of
2985 * rounding mode to arm_fprounding order (which matches the
2986 * common AArch64 order); see ARM ARM pseudocode FPDecodeRM().
2988 static const uint8_t fp_decode_rm[] = {
2989 FPROUNDING_TIEAWAY,
2990 FPROUNDING_TIEEVEN,
2991 FPROUNDING_POSINF,
2992 FPROUNDING_NEGINF,
2995 static int disas_vfp_v8_insn(DisasContext *s, uint32_t insn)
2997 uint32_t rd, rn, rm, dp = extract32(insn, 8, 1);
2999 if (!arm_dc_feature(s, ARM_FEATURE_V8)) {
3000 return 1;
3003 if (dp) {
3004 VFP_DREG_D(rd, insn);
3005 VFP_DREG_N(rn, insn);
3006 VFP_DREG_M(rm, insn);
3007 } else {
3008 rd = VFP_SREG_D(insn);
3009 rn = VFP_SREG_N(insn);
3010 rm = VFP_SREG_M(insn);
3013 if ((insn & 0x0f800e50) == 0x0e000a00) {
3014 return handle_vsel(insn, rd, rn, rm, dp);
3015 } else if ((insn & 0x0fb00e10) == 0x0e800a00) {
3016 return handle_vminmaxnm(insn, rd, rn, rm, dp);
3017 } else if ((insn & 0x0fbc0ed0) == 0x0eb80a40) {
3018 /* VRINTA, VRINTN, VRINTP, VRINTM */
3019 int rounding = fp_decode_rm[extract32(insn, 16, 2)];
3020 return handle_vrint(insn, rd, rm, dp, rounding);
3021 } else if ((insn & 0x0fbc0e50) == 0x0ebc0a40) {
3022 /* VCVTA, VCVTN, VCVTP, VCVTM */
3023 int rounding = fp_decode_rm[extract32(insn, 16, 2)];
3024 return handle_vcvt(insn, rd, rm, dp, rounding);
3026 return 1;
3029 /* Disassemble a VFP instruction. Returns nonzero if an error occurred
3030 (ie. an undefined instruction). */
3031 static int disas_vfp_insn(DisasContext *s, uint32_t insn)
3033 uint32_t rd, rn, rm, op, i, n, offset, delta_d, delta_m, bank_mask;
3034 int dp, veclen;
3035 TCGv_i32 addr;
3036 TCGv_i32 tmp;
3037 TCGv_i32 tmp2;
3039 if (!arm_dc_feature(s, ARM_FEATURE_VFP)) {
3040 return 1;
3043 /* FIXME: this access check should not take precedence over UNDEF
3044 * for invalid encodings; we will generate incorrect syndrome information
3045 * for attempts to execute invalid vfp/neon encodings with FP disabled.
3047 if (s->fp_excp_el) {
3048 gen_exception_insn(s, 4, EXCP_UDEF,
3049 syn_fp_access_trap(1, 0xe, s->thumb), s->fp_excp_el);
3050 return 0;
3053 if (!s->vfp_enabled) {
3054 /* VFP disabled. Only allow fmxr/fmrx to/from some control regs. */
3055 if ((insn & 0x0fe00fff) != 0x0ee00a10)
3056 return 1;
3057 rn = (insn >> 16) & 0xf;
3058 if (rn != ARM_VFP_FPSID && rn != ARM_VFP_FPEXC && rn != ARM_VFP_MVFR2
3059 && rn != ARM_VFP_MVFR1 && rn != ARM_VFP_MVFR0) {
3060 return 1;
3064 if (extract32(insn, 28, 4) == 0xf) {
3065 /* Encodings with T=1 (Thumb) or unconditional (ARM):
3066 * only used in v8 and above.
3068 return disas_vfp_v8_insn(s, insn);
3071 dp = ((insn & 0xf00) == 0xb00);
3072 switch ((insn >> 24) & 0xf) {
3073 case 0xe:
3074 if (insn & (1 << 4)) {
3075 /* single register transfer */
3076 rd = (insn >> 12) & 0xf;
3077 if (dp) {
3078 int size;
3079 int pass;
3081 VFP_DREG_N(rn, insn);
3082 if (insn & 0xf)
3083 return 1;
3084 if (insn & 0x00c00060
3085 && !arm_dc_feature(s, ARM_FEATURE_NEON)) {
3086 return 1;
3089 pass = (insn >> 21) & 1;
3090 if (insn & (1 << 22)) {
3091 size = 0;
3092 offset = ((insn >> 5) & 3) * 8;
3093 } else if (insn & (1 << 5)) {
3094 size = 1;
3095 offset = (insn & (1 << 6)) ? 16 : 0;
3096 } else {
3097 size = 2;
3098 offset = 0;
3100 if (insn & ARM_CP_RW_BIT) {
3101 /* vfp->arm */
3102 tmp = neon_load_reg(rn, pass);
3103 switch (size) {
3104 case 0:
3105 if (offset)
3106 tcg_gen_shri_i32(tmp, tmp, offset);
3107 if (insn & (1 << 23))
3108 gen_uxtb(tmp);
3109 else
3110 gen_sxtb(tmp);
3111 break;
3112 case 1:
3113 if (insn & (1 << 23)) {
3114 if (offset) {
3115 tcg_gen_shri_i32(tmp, tmp, 16);
3116 } else {
3117 gen_uxth(tmp);
3119 } else {
3120 if (offset) {
3121 tcg_gen_sari_i32(tmp, tmp, 16);
3122 } else {
3123 gen_sxth(tmp);
3126 break;
3127 case 2:
3128 break;
3130 store_reg(s, rd, tmp);
3131 } else {
3132 /* arm->vfp */
3133 tmp = load_reg(s, rd);
3134 if (insn & (1 << 23)) {
3135 /* VDUP */
3136 if (size == 0) {
3137 gen_neon_dup_u8(tmp, 0);
3138 } else if (size == 1) {
3139 gen_neon_dup_low16(tmp);
3141 for (n = 0; n <= pass * 2; n++) {
3142 tmp2 = tcg_temp_new_i32();
3143 tcg_gen_mov_i32(tmp2, tmp);
3144 neon_store_reg(rn, n, tmp2);
3146 neon_store_reg(rn, n, tmp);
3147 } else {
3148 /* VMOV */
3149 switch (size) {
3150 case 0:
3151 tmp2 = neon_load_reg(rn, pass);
3152 tcg_gen_deposit_i32(tmp, tmp2, tmp, offset, 8);
3153 tcg_temp_free_i32(tmp2);
3154 break;
3155 case 1:
3156 tmp2 = neon_load_reg(rn, pass);
3157 tcg_gen_deposit_i32(tmp, tmp2, tmp, offset, 16);
3158 tcg_temp_free_i32(tmp2);
3159 break;
3160 case 2:
3161 break;
3163 neon_store_reg(rn, pass, tmp);
3166 } else { /* !dp */
3167 if ((insn & 0x6f) != 0x00)
3168 return 1;
3169 rn = VFP_SREG_N(insn);
3170 if (insn & ARM_CP_RW_BIT) {
3171 /* vfp->arm */
3172 if (insn & (1 << 21)) {
3173 /* system register */
3174 rn >>= 1;
3176 switch (rn) {
3177 case ARM_VFP_FPSID:
3178 /* VFP2 allows access to FSID from userspace.
3179 VFP3 restricts all id registers to privileged
3180 accesses. */
3181 if (IS_USER(s)
3182 && arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3183 return 1;
3185 tmp = load_cpu_field(vfp.xregs[rn]);
3186 break;
3187 case ARM_VFP_FPEXC:
3188 if (IS_USER(s))
3189 return 1;
3190 tmp = load_cpu_field(vfp.xregs[rn]);
3191 break;
3192 case ARM_VFP_FPINST:
3193 case ARM_VFP_FPINST2:
3194 /* Not present in VFP3. */
3195 if (IS_USER(s)
3196 || arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3197 return 1;
3199 tmp = load_cpu_field(vfp.xregs[rn]);
3200 break;
3201 case ARM_VFP_FPSCR:
3202 if (rd == 15) {
3203 tmp = load_cpu_field(vfp.xregs[ARM_VFP_FPSCR]);
3204 tcg_gen_andi_i32(tmp, tmp, 0xf0000000);
3205 } else {
3206 tmp = tcg_temp_new_i32();
3207 gen_helper_vfp_get_fpscr(tmp, cpu_env);
3209 break;
3210 case ARM_VFP_MVFR2:
3211 if (!arm_dc_feature(s, ARM_FEATURE_V8)) {
3212 return 1;
3214 /* fall through */
3215 case ARM_VFP_MVFR0:
3216 case ARM_VFP_MVFR1:
3217 if (IS_USER(s)
3218 || !arm_dc_feature(s, ARM_FEATURE_MVFR)) {
3219 return 1;
3221 tmp = load_cpu_field(vfp.xregs[rn]);
3222 break;
3223 default:
3224 return 1;
3226 } else {
3227 gen_mov_F0_vreg(0, rn);
3228 tmp = gen_vfp_mrs();
3230 if (rd == 15) {
3231 /* Set the 4 flag bits in the CPSR. */
3232 gen_set_nzcv(tmp);
3233 tcg_temp_free_i32(tmp);
3234 } else {
3235 store_reg(s, rd, tmp);
3237 } else {
3238 /* arm->vfp */
3239 if (insn & (1 << 21)) {
3240 rn >>= 1;
3241 /* system register */
3242 switch (rn) {
3243 case ARM_VFP_FPSID:
3244 case ARM_VFP_MVFR0:
3245 case ARM_VFP_MVFR1:
3246 /* Writes are ignored. */
3247 break;
3248 case ARM_VFP_FPSCR:
3249 tmp = load_reg(s, rd);
3250 gen_helper_vfp_set_fpscr(cpu_env, tmp);
3251 tcg_temp_free_i32(tmp);
3252 gen_lookup_tb(s);
3253 break;
3254 case ARM_VFP_FPEXC:
3255 if (IS_USER(s))
3256 return 1;
3257 /* TODO: VFP subarchitecture support.
3258 * For now, keep the EN bit only */
3259 tmp = load_reg(s, rd);
3260 tcg_gen_andi_i32(tmp, tmp, 1 << 30);
3261 store_cpu_field(tmp, vfp.xregs[rn]);
3262 gen_lookup_tb(s);
3263 break;
3264 case ARM_VFP_FPINST:
3265 case ARM_VFP_FPINST2:
3266 if (IS_USER(s)) {
3267 return 1;
3269 tmp = load_reg(s, rd);
3270 store_cpu_field(tmp, vfp.xregs[rn]);
3271 break;
3272 default:
3273 return 1;
3275 } else {
3276 tmp = load_reg(s, rd);
3277 gen_vfp_msr(tmp);
3278 gen_mov_vreg_F0(0, rn);
3282 } else {
3283 /* data processing */
3284 /* The opcode is in bits 23, 21, 20 and 6. */
3285 op = ((insn >> 20) & 8) | ((insn >> 19) & 6) | ((insn >> 6) & 1);
3286 if (dp) {
3287 if (op == 15) {
3288 /* rn is opcode */
3289 rn = ((insn >> 15) & 0x1e) | ((insn >> 7) & 1);
3290 } else {
3291 /* rn is register number */
3292 VFP_DREG_N(rn, insn);
3295 if (op == 15 && (rn == 15 || ((rn & 0x1c) == 0x18) ||
3296 ((rn & 0x1e) == 0x6))) {
3297 /* Integer or single/half precision destination. */
3298 rd = VFP_SREG_D(insn);
3299 } else {
3300 VFP_DREG_D(rd, insn);
3302 if (op == 15 &&
3303 (((rn & 0x1c) == 0x10) || ((rn & 0x14) == 0x14) ||
3304 ((rn & 0x1e) == 0x4))) {
3305 /* VCVT from int or half precision is always from S reg
3306 * regardless of dp bit. VCVT with immediate frac_bits
3307 * has same format as SREG_M.
3309 rm = VFP_SREG_M(insn);
3310 } else {
3311 VFP_DREG_M(rm, insn);
3313 } else {
3314 rn = VFP_SREG_N(insn);
3315 if (op == 15 && rn == 15) {
3316 /* Double precision destination. */
3317 VFP_DREG_D(rd, insn);
3318 } else {
3319 rd = VFP_SREG_D(insn);
3321 /* NB that we implicitly rely on the encoding for the frac_bits
3322 * in VCVT of fixed to float being the same as that of an SREG_M
3324 rm = VFP_SREG_M(insn);
3327 veclen = s->vec_len;
3328 if (op == 15 && rn > 3)
3329 veclen = 0;
3331 /* Shut up compiler warnings. */
3332 delta_m = 0;
3333 delta_d = 0;
3334 bank_mask = 0;
3336 if (veclen > 0) {
3337 if (dp)
3338 bank_mask = 0xc;
3339 else
3340 bank_mask = 0x18;
3342 /* Figure out what type of vector operation this is. */
3343 if ((rd & bank_mask) == 0) {
3344 /* scalar */
3345 veclen = 0;
3346 } else {
3347 if (dp)
3348 delta_d = (s->vec_stride >> 1) + 1;
3349 else
3350 delta_d = s->vec_stride + 1;
3352 if ((rm & bank_mask) == 0) {
3353 /* mixed scalar/vector */
3354 delta_m = 0;
3355 } else {
3356 /* vector */
3357 delta_m = delta_d;
3362 /* Load the initial operands. */
3363 if (op == 15) {
3364 switch (rn) {
3365 case 16:
3366 case 17:
3367 /* Integer source */
3368 gen_mov_F0_vreg(0, rm);
3369 break;
3370 case 8:
3371 case 9:
3372 /* Compare */
3373 gen_mov_F0_vreg(dp, rd);
3374 gen_mov_F1_vreg(dp, rm);
3375 break;
3376 case 10:
3377 case 11:
3378 /* Compare with zero */
3379 gen_mov_F0_vreg(dp, rd);
3380 gen_vfp_F1_ld0(dp);
3381 break;
3382 case 20:
3383 case 21:
3384 case 22:
3385 case 23:
3386 case 28:
3387 case 29:
3388 case 30:
3389 case 31:
3390 /* Source and destination the same. */
3391 gen_mov_F0_vreg(dp, rd);
3392 break;
3393 case 4:
3394 case 5:
3395 case 6:
3396 case 7:
3397 /* VCVTB, VCVTT: only present with the halfprec extension
3398 * UNPREDICTABLE if bit 8 is set prior to ARMv8
3399 * (we choose to UNDEF)
3401 if ((dp && !arm_dc_feature(s, ARM_FEATURE_V8)) ||
3402 !arm_dc_feature(s, ARM_FEATURE_VFP_FP16)) {
3403 return 1;
3405 if (!extract32(rn, 1, 1)) {
3406 /* Half precision source. */
3407 gen_mov_F0_vreg(0, rm);
3408 break;
3410 /* Otherwise fall through */
3411 default:
3412 /* One source operand. */
3413 gen_mov_F0_vreg(dp, rm);
3414 break;
3416 } else {
3417 /* Two source operands. */
3418 gen_mov_F0_vreg(dp, rn);
3419 gen_mov_F1_vreg(dp, rm);
3422 for (;;) {
3423 /* Perform the calculation. */
3424 switch (op) {
3425 case 0: /* VMLA: fd + (fn * fm) */
3426 /* Note that order of inputs to the add matters for NaNs */
3427 gen_vfp_F1_mul(dp);
3428 gen_mov_F0_vreg(dp, rd);
3429 gen_vfp_add(dp);
3430 break;
3431 case 1: /* VMLS: fd + -(fn * fm) */
3432 gen_vfp_mul(dp);
3433 gen_vfp_F1_neg(dp);
3434 gen_mov_F0_vreg(dp, rd);
3435 gen_vfp_add(dp);
3436 break;
3437 case 2: /* VNMLS: -fd + (fn * fm) */
3438 /* Note that it isn't valid to replace (-A + B) with (B - A)
3439 * or similar plausible looking simplifications
3440 * because this will give wrong results for NaNs.
3442 gen_vfp_F1_mul(dp);
3443 gen_mov_F0_vreg(dp, rd);
3444 gen_vfp_neg(dp);
3445 gen_vfp_add(dp);
3446 break;
3447 case 3: /* VNMLA: -fd + -(fn * fm) */
3448 gen_vfp_mul(dp);
3449 gen_vfp_F1_neg(dp);
3450 gen_mov_F0_vreg(dp, rd);
3451 gen_vfp_neg(dp);
3452 gen_vfp_add(dp);
3453 break;
3454 case 4: /* mul: fn * fm */
3455 gen_vfp_mul(dp);
3456 break;
3457 case 5: /* nmul: -(fn * fm) */
3458 gen_vfp_mul(dp);
3459 gen_vfp_neg(dp);
3460 break;
3461 case 6: /* add: fn + fm */
3462 gen_vfp_add(dp);
3463 break;
3464 case 7: /* sub: fn - fm */
3465 gen_vfp_sub(dp);
3466 break;
3467 case 8: /* div: fn / fm */
3468 gen_vfp_div(dp);
3469 break;
3470 case 10: /* VFNMA : fd = muladd(-fd, fn, fm) */
3471 case 11: /* VFNMS : fd = muladd(-fd, -fn, fm) */
3472 case 12: /* VFMA : fd = muladd( fd, fn, fm) */
3473 case 13: /* VFMS : fd = muladd( fd, -fn, fm) */
3474 /* These are fused multiply-add, and must be done as one
3475 * floating point operation with no rounding between the
3476 * multiplication and addition steps.
3477 * NB that doing the negations here as separate steps is
3478 * correct : an input NaN should come out with its sign bit
3479 * flipped if it is a negated-input.
3481 if (!arm_dc_feature(s, ARM_FEATURE_VFP4)) {
3482 return 1;
3484 if (dp) {
3485 TCGv_ptr fpst;
3486 TCGv_i64 frd;
3487 if (op & 1) {
3488 /* VFNMS, VFMS */
3489 gen_helper_vfp_negd(cpu_F0d, cpu_F0d);
3491 frd = tcg_temp_new_i64();
3492 tcg_gen_ld_f64(frd, cpu_env, vfp_reg_offset(dp, rd));
3493 if (op & 2) {
3494 /* VFNMA, VFNMS */
3495 gen_helper_vfp_negd(frd, frd);
3497 fpst = get_fpstatus_ptr(0);
3498 gen_helper_vfp_muladdd(cpu_F0d, cpu_F0d,
3499 cpu_F1d, frd, fpst);
3500 tcg_temp_free_ptr(fpst);
3501 tcg_temp_free_i64(frd);
3502 } else {
3503 TCGv_ptr fpst;
3504 TCGv_i32 frd;
3505 if (op & 1) {
3506 /* VFNMS, VFMS */
3507 gen_helper_vfp_negs(cpu_F0s, cpu_F0s);
3509 frd = tcg_temp_new_i32();
3510 tcg_gen_ld_f32(frd, cpu_env, vfp_reg_offset(dp, rd));
3511 if (op & 2) {
3512 gen_helper_vfp_negs(frd, frd);
3514 fpst = get_fpstatus_ptr(0);
3515 gen_helper_vfp_muladds(cpu_F0s, cpu_F0s,
3516 cpu_F1s, frd, fpst);
3517 tcg_temp_free_ptr(fpst);
3518 tcg_temp_free_i32(frd);
3520 break;
3521 case 14: /* fconst */
3522 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3523 return 1;
3526 n = (insn << 12) & 0x80000000;
3527 i = ((insn >> 12) & 0x70) | (insn & 0xf);
3528 if (dp) {
3529 if (i & 0x40)
3530 i |= 0x3f80;
3531 else
3532 i |= 0x4000;
3533 n |= i << 16;
3534 tcg_gen_movi_i64(cpu_F0d, ((uint64_t)n) << 32);
3535 } else {
3536 if (i & 0x40)
3537 i |= 0x780;
3538 else
3539 i |= 0x800;
3540 n |= i << 19;
3541 tcg_gen_movi_i32(cpu_F0s, n);
3543 break;
3544 case 15: /* extension space */
3545 switch (rn) {
3546 case 0: /* cpy */
3547 /* no-op */
3548 break;
3549 case 1: /* abs */
3550 gen_vfp_abs(dp);
3551 break;
3552 case 2: /* neg */
3553 gen_vfp_neg(dp);
3554 break;
3555 case 3: /* sqrt */
3556 gen_vfp_sqrt(dp);
3557 break;
3558 case 4: /* vcvtb.f32.f16, vcvtb.f64.f16 */
3559 tmp = gen_vfp_mrs();
3560 tcg_gen_ext16u_i32(tmp, tmp);
3561 if (dp) {
3562 gen_helper_vfp_fcvt_f16_to_f64(cpu_F0d, tmp,
3563 cpu_env);
3564 } else {
3565 gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp,
3566 cpu_env);
3568 tcg_temp_free_i32(tmp);
3569 break;
3570 case 5: /* vcvtt.f32.f16, vcvtt.f64.f16 */
3571 tmp = gen_vfp_mrs();
3572 tcg_gen_shri_i32(tmp, tmp, 16);
3573 if (dp) {
3574 gen_helper_vfp_fcvt_f16_to_f64(cpu_F0d, tmp,
3575 cpu_env);
3576 } else {
3577 gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp,
3578 cpu_env);
3580 tcg_temp_free_i32(tmp);
3581 break;
3582 case 6: /* vcvtb.f16.f32, vcvtb.f16.f64 */
3583 tmp = tcg_temp_new_i32();
3584 if (dp) {
3585 gen_helper_vfp_fcvt_f64_to_f16(tmp, cpu_F0d,
3586 cpu_env);
3587 } else {
3588 gen_helper_vfp_fcvt_f32_to_f16(tmp, cpu_F0s,
3589 cpu_env);
3591 gen_mov_F0_vreg(0, rd);
3592 tmp2 = gen_vfp_mrs();
3593 tcg_gen_andi_i32(tmp2, tmp2, 0xffff0000);
3594 tcg_gen_or_i32(tmp, tmp, tmp2);
3595 tcg_temp_free_i32(tmp2);
3596 gen_vfp_msr(tmp);
3597 break;
3598 case 7: /* vcvtt.f16.f32, vcvtt.f16.f64 */
3599 tmp = tcg_temp_new_i32();
3600 if (dp) {
3601 gen_helper_vfp_fcvt_f64_to_f16(tmp, cpu_F0d,
3602 cpu_env);
3603 } else {
3604 gen_helper_vfp_fcvt_f32_to_f16(tmp, cpu_F0s,
3605 cpu_env);
3607 tcg_gen_shli_i32(tmp, tmp, 16);
3608 gen_mov_F0_vreg(0, rd);
3609 tmp2 = gen_vfp_mrs();
3610 tcg_gen_ext16u_i32(tmp2, tmp2);
3611 tcg_gen_or_i32(tmp, tmp, tmp2);
3612 tcg_temp_free_i32(tmp2);
3613 gen_vfp_msr(tmp);
3614 break;
3615 case 8: /* cmp */
3616 gen_vfp_cmp(dp);
3617 break;
3618 case 9: /* cmpe */
3619 gen_vfp_cmpe(dp);
3620 break;
3621 case 10: /* cmpz */
3622 gen_vfp_cmp(dp);
3623 break;
3624 case 11: /* cmpez */
3625 gen_vfp_F1_ld0(dp);
3626 gen_vfp_cmpe(dp);
3627 break;
3628 case 12: /* vrintr */
3630 TCGv_ptr fpst = get_fpstatus_ptr(0);
3631 if (dp) {
3632 gen_helper_rintd(cpu_F0d, cpu_F0d, fpst);
3633 } else {
3634 gen_helper_rints(cpu_F0s, cpu_F0s, fpst);
3636 tcg_temp_free_ptr(fpst);
3637 break;
3639 case 13: /* vrintz */
3641 TCGv_ptr fpst = get_fpstatus_ptr(0);
3642 TCGv_i32 tcg_rmode;
3643 tcg_rmode = tcg_const_i32(float_round_to_zero);
3644 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
3645 if (dp) {
3646 gen_helper_rintd(cpu_F0d, cpu_F0d, fpst);
3647 } else {
3648 gen_helper_rints(cpu_F0s, cpu_F0s, fpst);
3650 gen_helper_set_rmode(tcg_rmode, tcg_rmode, cpu_env);
3651 tcg_temp_free_i32(tcg_rmode);
3652 tcg_temp_free_ptr(fpst);
3653 break;
3655 case 14: /* vrintx */
3657 TCGv_ptr fpst = get_fpstatus_ptr(0);
3658 if (dp) {
3659 gen_helper_rintd_exact(cpu_F0d, cpu_F0d, fpst);
3660 } else {
3661 gen_helper_rints_exact(cpu_F0s, cpu_F0s, fpst);
3663 tcg_temp_free_ptr(fpst);
3664 break;
3666 case 15: /* single<->double conversion */
3667 if (dp)
3668 gen_helper_vfp_fcvtsd(cpu_F0s, cpu_F0d, cpu_env);
3669 else
3670 gen_helper_vfp_fcvtds(cpu_F0d, cpu_F0s, cpu_env);
3671 break;
3672 case 16: /* fuito */
3673 gen_vfp_uito(dp, 0);
3674 break;
3675 case 17: /* fsito */
3676 gen_vfp_sito(dp, 0);
3677 break;
3678 case 20: /* fshto */
3679 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3680 return 1;
3682 gen_vfp_shto(dp, 16 - rm, 0);
3683 break;
3684 case 21: /* fslto */
3685 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3686 return 1;
3688 gen_vfp_slto(dp, 32 - rm, 0);
3689 break;
3690 case 22: /* fuhto */
3691 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3692 return 1;
3694 gen_vfp_uhto(dp, 16 - rm, 0);
3695 break;
3696 case 23: /* fulto */
3697 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3698 return 1;
3700 gen_vfp_ulto(dp, 32 - rm, 0);
3701 break;
3702 case 24: /* ftoui */
3703 gen_vfp_toui(dp, 0);
3704 break;
3705 case 25: /* ftouiz */
3706 gen_vfp_touiz(dp, 0);
3707 break;
3708 case 26: /* ftosi */
3709 gen_vfp_tosi(dp, 0);
3710 break;
3711 case 27: /* ftosiz */
3712 gen_vfp_tosiz(dp, 0);
3713 break;
3714 case 28: /* ftosh */
3715 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3716 return 1;
3718 gen_vfp_tosh(dp, 16 - rm, 0);
3719 break;
3720 case 29: /* ftosl */
3721 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3722 return 1;
3724 gen_vfp_tosl(dp, 32 - rm, 0);
3725 break;
3726 case 30: /* ftouh */
3727 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3728 return 1;
3730 gen_vfp_touh(dp, 16 - rm, 0);
3731 break;
3732 case 31: /* ftoul */
3733 if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
3734 return 1;
3736 gen_vfp_toul(dp, 32 - rm, 0);
3737 break;
3738 default: /* undefined */
3739 return 1;
3741 break;
3742 default: /* undefined */
3743 return 1;
3746 /* Write back the result. */
3747 if (op == 15 && (rn >= 8 && rn <= 11)) {
3748 /* Comparison, do nothing. */
3749 } else if (op == 15 && dp && ((rn & 0x1c) == 0x18 ||
3750 (rn & 0x1e) == 0x6)) {
3751 /* VCVT double to int: always integer result.
3752 * VCVT double to half precision is always a single
3753 * precision result.
3755 gen_mov_vreg_F0(0, rd);
3756 } else if (op == 15 && rn == 15) {
3757 /* conversion */
3758 gen_mov_vreg_F0(!dp, rd);
3759 } else {
3760 gen_mov_vreg_F0(dp, rd);
3763 /* break out of the loop if we have finished */
3764 if (veclen == 0)
3765 break;
3767 if (op == 15 && delta_m == 0) {
3768 /* single source one-many */
3769 while (veclen--) {
3770 rd = ((rd + delta_d) & (bank_mask - 1))
3771 | (rd & bank_mask);
3772 gen_mov_vreg_F0(dp, rd);
3774 break;
3776 /* Setup the next operands. */
3777 veclen--;
3778 rd = ((rd + delta_d) & (bank_mask - 1))
3779 | (rd & bank_mask);
3781 if (op == 15) {
3782 /* One source operand. */
3783 rm = ((rm + delta_m) & (bank_mask - 1))
3784 | (rm & bank_mask);
3785 gen_mov_F0_vreg(dp, rm);
3786 } else {
3787 /* Two source operands. */
3788 rn = ((rn + delta_d) & (bank_mask - 1))
3789 | (rn & bank_mask);
3790 gen_mov_F0_vreg(dp, rn);
3791 if (delta_m) {
3792 rm = ((rm + delta_m) & (bank_mask - 1))
3793 | (rm & bank_mask);
3794 gen_mov_F1_vreg(dp, rm);
3799 break;
3800 case 0xc:
3801 case 0xd:
3802 if ((insn & 0x03e00000) == 0x00400000) {
3803 /* two-register transfer */
3804 rn = (insn >> 16) & 0xf;
3805 rd = (insn >> 12) & 0xf;
3806 if (dp) {
3807 VFP_DREG_M(rm, insn);
3808 } else {
3809 rm = VFP_SREG_M(insn);
3812 if (insn & ARM_CP_RW_BIT) {
3813 /* vfp->arm */
3814 if (dp) {
3815 gen_mov_F0_vreg(0, rm * 2);
3816 tmp = gen_vfp_mrs();
3817 store_reg(s, rd, tmp);
3818 gen_mov_F0_vreg(0, rm * 2 + 1);
3819 tmp = gen_vfp_mrs();
3820 store_reg(s, rn, tmp);
3821 } else {
3822 gen_mov_F0_vreg(0, rm);
3823 tmp = gen_vfp_mrs();
3824 store_reg(s, rd, tmp);
3825 gen_mov_F0_vreg(0, rm + 1);
3826 tmp = gen_vfp_mrs();
3827 store_reg(s, rn, tmp);
3829 } else {
3830 /* arm->vfp */
3831 if (dp) {
3832 tmp = load_reg(s, rd);
3833 gen_vfp_msr(tmp);
3834 gen_mov_vreg_F0(0, rm * 2);
3835 tmp = load_reg(s, rn);
3836 gen_vfp_msr(tmp);
3837 gen_mov_vreg_F0(0, rm * 2 + 1);
3838 } else {
3839 tmp = load_reg(s, rd);
3840 gen_vfp_msr(tmp);
3841 gen_mov_vreg_F0(0, rm);
3842 tmp = load_reg(s, rn);
3843 gen_vfp_msr(tmp);
3844 gen_mov_vreg_F0(0, rm + 1);
3847 } else {
3848 /* Load/store */
3849 rn = (insn >> 16) & 0xf;
3850 if (dp)
3851 VFP_DREG_D(rd, insn);
3852 else
3853 rd = VFP_SREG_D(insn);
3854 if ((insn & 0x01200000) == 0x01000000) {
3855 /* Single load/store */
3856 offset = (insn & 0xff) << 2;
3857 if ((insn & (1 << 23)) == 0)
3858 offset = -offset;
3859 if (s->thumb && rn == 15) {
3860 /* This is actually UNPREDICTABLE */
3861 addr = tcg_temp_new_i32();
3862 tcg_gen_movi_i32(addr, s->pc & ~2);
3863 } else {
3864 addr = load_reg(s, rn);
3866 tcg_gen_addi_i32(addr, addr, offset);
3867 if (insn & (1 << 20)) {
3868 gen_vfp_ld(s, dp, addr);
3869 gen_mov_vreg_F0(dp, rd);
3870 } else {
3871 gen_mov_F0_vreg(dp, rd);
3872 gen_vfp_st(s, dp, addr);
3874 tcg_temp_free_i32(addr);
3875 } else {
3876 /* load/store multiple */
3877 int w = insn & (1 << 21);
3878 if (dp)
3879 n = (insn >> 1) & 0x7f;
3880 else
3881 n = insn & 0xff;
3883 if (w && !(((insn >> 23) ^ (insn >> 24)) & 1)) {
3884 /* P == U , W == 1 => UNDEF */
3885 return 1;
3887 if (n == 0 || (rd + n) > 32 || (dp && n > 16)) {
3888 /* UNPREDICTABLE cases for bad immediates: we choose to
3889 * UNDEF to avoid generating huge numbers of TCG ops
3891 return 1;
3893 if (rn == 15 && w) {
3894 /* writeback to PC is UNPREDICTABLE, we choose to UNDEF */
3895 return 1;
3898 if (s->thumb && rn == 15) {
3899 /* This is actually UNPREDICTABLE */
3900 addr = tcg_temp_new_i32();
3901 tcg_gen_movi_i32(addr, s->pc & ~2);
3902 } else {
3903 addr = load_reg(s, rn);
3905 if (insn & (1 << 24)) /* pre-decrement */
3906 tcg_gen_addi_i32(addr, addr, -((insn & 0xff) << 2));
3908 if (dp)
3909 offset = 8;
3910 else
3911 offset = 4;
3912 for (i = 0; i < n; i++) {
3913 if (insn & ARM_CP_RW_BIT) {
3914 /* load */
3915 gen_vfp_ld(s, dp, addr);
3916 gen_mov_vreg_F0(dp, rd + i);
3917 } else {
3918 /* store */
3919 gen_mov_F0_vreg(dp, rd + i);
3920 gen_vfp_st(s, dp, addr);
3922 tcg_gen_addi_i32(addr, addr, offset);
3924 if (w) {
3925 /* writeback */
3926 if (insn & (1 << 24))
3927 offset = -offset * n;
3928 else if (dp && (insn & 1))
3929 offset = 4;
3930 else
3931 offset = 0;
3933 if (offset != 0)
3934 tcg_gen_addi_i32(addr, addr, offset);
3935 store_reg(s, rn, addr);
3936 } else {
3937 tcg_temp_free_i32(addr);
3941 break;
3942 default:
3943 /* Should never happen. */
3944 return 1;
3946 return 0;
3949 static inline void gen_goto_tb(DisasContext *s, int n, target_ulong dest)
3951 TranslationBlock *tb;
3953 tb = s->tb;
3954 if ((tb->pc & TARGET_PAGE_MASK) == (dest & TARGET_PAGE_MASK)) {
3955 tcg_gen_goto_tb(n);
3956 gen_set_pc_im(s, dest);
3957 tcg_gen_exit_tb((uintptr_t)tb + n);
3958 } else {
3959 gen_set_pc_im(s, dest);
3960 tcg_gen_exit_tb(0);
3964 static inline void gen_jmp (DisasContext *s, uint32_t dest)
3966 if (unlikely(s->singlestep_enabled || s->ss_active)) {
3967 /* An indirect jump so that we still trigger the debug exception. */
3968 if (s->thumb)
3969 dest |= 1;
3970 gen_bx_im(s, dest);
3971 } else {
3972 gen_goto_tb(s, 0, dest);
3973 s->is_jmp = DISAS_TB_JUMP;
3977 static inline void gen_mulxy(TCGv_i32 t0, TCGv_i32 t1, int x, int y)
3979 if (x)
3980 tcg_gen_sari_i32(t0, t0, 16);
3981 else
3982 gen_sxth(t0);
3983 if (y)
3984 tcg_gen_sari_i32(t1, t1, 16);
3985 else
3986 gen_sxth(t1);
3987 tcg_gen_mul_i32(t0, t0, t1);
3990 /* Return the mask of PSR bits set by a MSR instruction. */
3991 static uint32_t msr_mask(DisasContext *s, int flags, int spsr)
3993 uint32_t mask;
3995 mask = 0;
3996 if (flags & (1 << 0))
3997 mask |= 0xff;
3998 if (flags & (1 << 1))
3999 mask |= 0xff00;
4000 if (flags & (1 << 2))
4001 mask |= 0xff0000;
4002 if (flags & (1 << 3))
4003 mask |= 0xff000000;
4005 /* Mask out undefined bits. */
4006 mask &= ~CPSR_RESERVED;
4007 if (!arm_dc_feature(s, ARM_FEATURE_V4T)) {
4008 mask &= ~CPSR_T;
4010 if (!arm_dc_feature(s, ARM_FEATURE_V5)) {
4011 mask &= ~CPSR_Q; /* V5TE in reality*/
4013 if (!arm_dc_feature(s, ARM_FEATURE_V6)) {
4014 mask &= ~(CPSR_E | CPSR_GE);
4016 if (!arm_dc_feature(s, ARM_FEATURE_THUMB2)) {
4017 mask &= ~CPSR_IT;
4019 /* Mask out execution state and reserved bits. */
4020 if (!spsr) {
4021 mask &= ~(CPSR_EXEC | CPSR_RESERVED);
4023 /* Mask out privileged bits. */
4024 if (IS_USER(s))
4025 mask &= CPSR_USER;
4026 return mask;
4029 /* Returns nonzero if access to the PSR is not permitted. Marks t0 as dead. */
4030 static int gen_set_psr(DisasContext *s, uint32_t mask, int spsr, TCGv_i32 t0)
4032 TCGv_i32 tmp;
4033 if (spsr) {
4034 /* ??? This is also undefined in system mode. */
4035 if (IS_USER(s))
4036 return 1;
4038 tmp = load_cpu_field(spsr);
4039 tcg_gen_andi_i32(tmp, tmp, ~mask);
4040 tcg_gen_andi_i32(t0, t0, mask);
4041 tcg_gen_or_i32(tmp, tmp, t0);
4042 store_cpu_field(tmp, spsr);
4043 } else {
4044 gen_set_cpsr(t0, mask);
4046 tcg_temp_free_i32(t0);
4047 gen_lookup_tb(s);
4048 return 0;
4051 /* Returns nonzero if access to the PSR is not permitted. */
4052 static int gen_set_psr_im(DisasContext *s, uint32_t mask, int spsr, uint32_t val)
4054 TCGv_i32 tmp;
4055 tmp = tcg_temp_new_i32();
4056 tcg_gen_movi_i32(tmp, val);
4057 return gen_set_psr(s, mask, spsr, tmp);
4060 /* Generate an old-style exception return. Marks pc as dead. */
4061 static void gen_exception_return(DisasContext *s, TCGv_i32 pc)
4063 TCGv_i32 tmp;
4064 store_reg(s, 15, pc);
4065 tmp = load_cpu_field(spsr);
4066 gen_set_cpsr(tmp, CPSR_ERET_MASK);
4067 tcg_temp_free_i32(tmp);
4068 s->is_jmp = DISAS_UPDATE;
4071 /* Generate a v6 exception return. Marks both values as dead. */
4072 static void gen_rfe(DisasContext *s, TCGv_i32 pc, TCGv_i32 cpsr)
4074 gen_set_cpsr(cpsr, CPSR_ERET_MASK);
4075 tcg_temp_free_i32(cpsr);
4076 store_reg(s, 15, pc);
4077 s->is_jmp = DISAS_UPDATE;
4080 static void gen_nop_hint(DisasContext *s, int val)
4082 switch (val) {
4083 case 1: /* yield */
4084 gen_set_pc_im(s, s->pc);
4085 s->is_jmp = DISAS_YIELD;
4086 break;
4087 case 3: /* wfi */
4088 gen_set_pc_im(s, s->pc);
4089 s->is_jmp = DISAS_WFI;
4090 break;
4091 case 2: /* wfe */
4092 gen_set_pc_im(s, s->pc);
4093 s->is_jmp = DISAS_WFE;
4094 break;
4095 case 4: /* sev */
4096 case 5: /* sevl */
4097 /* TODO: Implement SEV, SEVL and WFE. May help SMP performance. */
4098 default: /* nop */
4099 break;
4103 #define CPU_V001 cpu_V0, cpu_V0, cpu_V1
4105 static inline void gen_neon_add(int size, TCGv_i32 t0, TCGv_i32 t1)
4107 switch (size) {
4108 case 0: gen_helper_neon_add_u8(t0, t0, t1); break;
4109 case 1: gen_helper_neon_add_u16(t0, t0, t1); break;
4110 case 2: tcg_gen_add_i32(t0, t0, t1); break;
4111 default: abort();
4115 static inline void gen_neon_rsb(int size, TCGv_i32 t0, TCGv_i32 t1)
4117 switch (size) {
4118 case 0: gen_helper_neon_sub_u8(t0, t1, t0); break;
4119 case 1: gen_helper_neon_sub_u16(t0, t1, t0); break;
4120 case 2: tcg_gen_sub_i32(t0, t1, t0); break;
4121 default: return;
4125 /* 32-bit pairwise ops end up the same as the elementwise versions. */
4126 #define gen_helper_neon_pmax_s32 gen_helper_neon_max_s32
4127 #define gen_helper_neon_pmax_u32 gen_helper_neon_max_u32
4128 #define gen_helper_neon_pmin_s32 gen_helper_neon_min_s32
4129 #define gen_helper_neon_pmin_u32 gen_helper_neon_min_u32
4131 #define GEN_NEON_INTEGER_OP_ENV(name) do { \
4132 switch ((size << 1) | u) { \
4133 case 0: \
4134 gen_helper_neon_##name##_s8(tmp, cpu_env, tmp, tmp2); \
4135 break; \
4136 case 1: \
4137 gen_helper_neon_##name##_u8(tmp, cpu_env, tmp, tmp2); \
4138 break; \
4139 case 2: \
4140 gen_helper_neon_##name##_s16(tmp, cpu_env, tmp, tmp2); \
4141 break; \
4142 case 3: \
4143 gen_helper_neon_##name##_u16(tmp, cpu_env, tmp, tmp2); \
4144 break; \
4145 case 4: \
4146 gen_helper_neon_##name##_s32(tmp, cpu_env, tmp, tmp2); \
4147 break; \
4148 case 5: \
4149 gen_helper_neon_##name##_u32(tmp, cpu_env, tmp, tmp2); \
4150 break; \
4151 default: return 1; \
4152 }} while (0)
4154 #define GEN_NEON_INTEGER_OP(name) do { \
4155 switch ((size << 1) | u) { \
4156 case 0: \
4157 gen_helper_neon_##name##_s8(tmp, tmp, tmp2); \
4158 break; \
4159 case 1: \
4160 gen_helper_neon_##name##_u8(tmp, tmp, tmp2); \
4161 break; \
4162 case 2: \
4163 gen_helper_neon_##name##_s16(tmp, tmp, tmp2); \
4164 break; \
4165 case 3: \
4166 gen_helper_neon_##name##_u16(tmp, tmp, tmp2); \
4167 break; \
4168 case 4: \
4169 gen_helper_neon_##name##_s32(tmp, tmp, tmp2); \
4170 break; \
4171 case 5: \
4172 gen_helper_neon_##name##_u32(tmp, tmp, tmp2); \
4173 break; \
4174 default: return 1; \
4175 }} while (0)
4177 static TCGv_i32 neon_load_scratch(int scratch)
4179 TCGv_i32 tmp = tcg_temp_new_i32();
4180 tcg_gen_ld_i32(tmp, cpu_env, offsetof(CPUARMState, vfp.scratch[scratch]));
4181 return tmp;
4184 static void neon_store_scratch(int scratch, TCGv_i32 var)
4186 tcg_gen_st_i32(var, cpu_env, offsetof(CPUARMState, vfp.scratch[scratch]));
4187 tcg_temp_free_i32(var);
4190 static inline TCGv_i32 neon_get_scalar(int size, int reg)
4192 TCGv_i32 tmp;
4193 if (size == 1) {
4194 tmp = neon_load_reg(reg & 7, reg >> 4);
4195 if (reg & 8) {
4196 gen_neon_dup_high16(tmp);
4197 } else {
4198 gen_neon_dup_low16(tmp);
4200 } else {
4201 tmp = neon_load_reg(reg & 15, reg >> 4);
4203 return tmp;
4206 static int gen_neon_unzip(int rd, int rm, int size, int q)
4208 TCGv_i32 tmp, tmp2;
4209 if (!q && size == 2) {
4210 return 1;
4212 tmp = tcg_const_i32(rd);
4213 tmp2 = tcg_const_i32(rm);
4214 if (q) {
4215 switch (size) {
4216 case 0:
4217 gen_helper_neon_qunzip8(cpu_env, tmp, tmp2);
4218 break;
4219 case 1:
4220 gen_helper_neon_qunzip16(cpu_env, tmp, tmp2);
4221 break;
4222 case 2:
4223 gen_helper_neon_qunzip32(cpu_env, tmp, tmp2);
4224 break;
4225 default:
4226 abort();
4228 } else {
4229 switch (size) {
4230 case 0:
4231 gen_helper_neon_unzip8(cpu_env, tmp, tmp2);
4232 break;
4233 case 1:
4234 gen_helper_neon_unzip16(cpu_env, tmp, tmp2);
4235 break;
4236 default:
4237 abort();
4240 tcg_temp_free_i32(tmp);
4241 tcg_temp_free_i32(tmp2);
4242 return 0;
4245 static int gen_neon_zip(int rd, int rm, int size, int q)
4247 TCGv_i32 tmp, tmp2;
4248 if (!q && size == 2) {
4249 return 1;
4251 tmp = tcg_const_i32(rd);
4252 tmp2 = tcg_const_i32(rm);
4253 if (q) {
4254 switch (size) {
4255 case 0:
4256 gen_helper_neon_qzip8(cpu_env, tmp, tmp2);
4257 break;
4258 case 1:
4259 gen_helper_neon_qzip16(cpu_env, tmp, tmp2);
4260 break;
4261 case 2:
4262 gen_helper_neon_qzip32(cpu_env, tmp, tmp2);
4263 break;
4264 default:
4265 abort();
4267 } else {
4268 switch (size) {
4269 case 0:
4270 gen_helper_neon_zip8(cpu_env, tmp, tmp2);
4271 break;
4272 case 1:
4273 gen_helper_neon_zip16(cpu_env, tmp, tmp2);
4274 break;
4275 default:
4276 abort();
4279 tcg_temp_free_i32(tmp);
4280 tcg_temp_free_i32(tmp2);
4281 return 0;
4284 static void gen_neon_trn_u8(TCGv_i32 t0, TCGv_i32 t1)
4286 TCGv_i32 rd, tmp;
4288 rd = tcg_temp_new_i32();
4289 tmp = tcg_temp_new_i32();
4291 tcg_gen_shli_i32(rd, t0, 8);
4292 tcg_gen_andi_i32(rd, rd, 0xff00ff00);
4293 tcg_gen_andi_i32(tmp, t1, 0x00ff00ff);
4294 tcg_gen_or_i32(rd, rd, tmp);
4296 tcg_gen_shri_i32(t1, t1, 8);
4297 tcg_gen_andi_i32(t1, t1, 0x00ff00ff);
4298 tcg_gen_andi_i32(tmp, t0, 0xff00ff00);
4299 tcg_gen_or_i32(t1, t1, tmp);
4300 tcg_gen_mov_i32(t0, rd);
4302 tcg_temp_free_i32(tmp);
4303 tcg_temp_free_i32(rd);
4306 static void gen_neon_trn_u16(TCGv_i32 t0, TCGv_i32 t1)
4308 TCGv_i32 rd, tmp;
4310 rd = tcg_temp_new_i32();
4311 tmp = tcg_temp_new_i32();
4313 tcg_gen_shli_i32(rd, t0, 16);
4314 tcg_gen_andi_i32(tmp, t1, 0xffff);
4315 tcg_gen_or_i32(rd, rd, tmp);
4316 tcg_gen_shri_i32(t1, t1, 16);
4317 tcg_gen_andi_i32(tmp, t0, 0xffff0000);
4318 tcg_gen_or_i32(t1, t1, tmp);
4319 tcg_gen_mov_i32(t0, rd);
4321 tcg_temp_free_i32(tmp);
4322 tcg_temp_free_i32(rd);
4326 static struct {
4327 int nregs;
4328 int interleave;
4329 int spacing;
4330 } neon_ls_element_type[11] = {
4331 {4, 4, 1},
4332 {4, 4, 2},
4333 {4, 1, 1},
4334 {4, 2, 1},
4335 {3, 3, 1},
4336 {3, 3, 2},
4337 {3, 1, 1},
4338 {1, 1, 1},
4339 {2, 2, 1},
4340 {2, 2, 2},
4341 {2, 1, 1}
4344 /* Translate a NEON load/store element instruction. Return nonzero if the
4345 instruction is invalid. */
4346 static int disas_neon_ls_insn(DisasContext *s, uint32_t insn)
4348 int rd, rn, rm;
4349 int op;
4350 int nregs;
4351 int interleave;
4352 int spacing;
4353 int stride;
4354 int size;
4355 int reg;
4356 int pass;
4357 int load;
4358 int shift;
4359 int n;
4360 TCGv_i32 addr;
4361 TCGv_i32 tmp;
4362 TCGv_i32 tmp2;
4363 TCGv_i64 tmp64;
4365 /* FIXME: this access check should not take precedence over UNDEF
4366 * for invalid encodings; we will generate incorrect syndrome information
4367 * for attempts to execute invalid vfp/neon encodings with FP disabled.
4369 if (s->fp_excp_el) {
4370 gen_exception_insn(s, 4, EXCP_UDEF,
4371 syn_fp_access_trap(1, 0xe, s->thumb), s->fp_excp_el);
4372 return 0;
4375 if (!s->vfp_enabled)
4376 return 1;
4377 VFP_DREG_D(rd, insn);
4378 rn = (insn >> 16) & 0xf;
4379 rm = insn & 0xf;
4380 load = (insn & (1 << 21)) != 0;
4381 if ((insn & (1 << 23)) == 0) {
4382 /* Load store all elements. */
4383 op = (insn >> 8) & 0xf;
4384 size = (insn >> 6) & 3;
4385 if (op > 10)
4386 return 1;
4387 /* Catch UNDEF cases for bad values of align field */
4388 switch (op & 0xc) {
4389 case 4:
4390 if (((insn >> 5) & 1) == 1) {
4391 return 1;
4393 break;
4394 case 8:
4395 if (((insn >> 4) & 3) == 3) {
4396 return 1;
4398 break;
4399 default:
4400 break;
4402 nregs = neon_ls_element_type[op].nregs;
4403 interleave = neon_ls_element_type[op].interleave;
4404 spacing = neon_ls_element_type[op].spacing;
4405 if (size == 3 && (interleave | spacing) != 1)
4406 return 1;
4407 addr = tcg_temp_new_i32();
4408 load_reg_var(s, addr, rn);
4409 stride = (1 << size) * interleave;
4410 for (reg = 0; reg < nregs; reg++) {
4411 if (interleave > 2 || (interleave == 2 && nregs == 2)) {
4412 load_reg_var(s, addr, rn);
4413 tcg_gen_addi_i32(addr, addr, (1 << size) * reg);
4414 } else if (interleave == 2 && nregs == 4 && reg == 2) {
4415 load_reg_var(s, addr, rn);
4416 tcg_gen_addi_i32(addr, addr, 1 << size);
4418 if (size == 3) {
4419 tmp64 = tcg_temp_new_i64();
4420 if (load) {
4421 gen_aa32_ld64(tmp64, addr, get_mem_index(s));
4422 neon_store_reg64(tmp64, rd);
4423 } else {
4424 neon_load_reg64(tmp64, rd);
4425 gen_aa32_st64(tmp64, addr, get_mem_index(s));
4427 tcg_temp_free_i64(tmp64);
4428 tcg_gen_addi_i32(addr, addr, stride);
4429 } else {
4430 for (pass = 0; pass < 2; pass++) {
4431 if (size == 2) {
4432 if (load) {
4433 tmp = tcg_temp_new_i32();
4434 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
4435 neon_store_reg(rd, pass, tmp);
4436 } else {
4437 tmp = neon_load_reg(rd, pass);
4438 gen_aa32_st32(tmp, addr, get_mem_index(s));
4439 tcg_temp_free_i32(tmp);
4441 tcg_gen_addi_i32(addr, addr, stride);
4442 } else if (size == 1) {
4443 if (load) {
4444 tmp = tcg_temp_new_i32();
4445 gen_aa32_ld16u(tmp, addr, get_mem_index(s));
4446 tcg_gen_addi_i32(addr, addr, stride);
4447 tmp2 = tcg_temp_new_i32();
4448 gen_aa32_ld16u(tmp2, addr, get_mem_index(s));
4449 tcg_gen_addi_i32(addr, addr, stride);
4450 tcg_gen_shli_i32(tmp2, tmp2, 16);
4451 tcg_gen_or_i32(tmp, tmp, tmp2);
4452 tcg_temp_free_i32(tmp2);
4453 neon_store_reg(rd, pass, tmp);
4454 } else {
4455 tmp = neon_load_reg(rd, pass);
4456 tmp2 = tcg_temp_new_i32();
4457 tcg_gen_shri_i32(tmp2, tmp, 16);
4458 gen_aa32_st16(tmp, addr, get_mem_index(s));
4459 tcg_temp_free_i32(tmp);
4460 tcg_gen_addi_i32(addr, addr, stride);
4461 gen_aa32_st16(tmp2, addr, get_mem_index(s));
4462 tcg_temp_free_i32(tmp2);
4463 tcg_gen_addi_i32(addr, addr, stride);
4465 } else /* size == 0 */ {
4466 if (load) {
4467 TCGV_UNUSED_I32(tmp2);
4468 for (n = 0; n < 4; n++) {
4469 tmp = tcg_temp_new_i32();
4470 gen_aa32_ld8u(tmp, addr, get_mem_index(s));
4471 tcg_gen_addi_i32(addr, addr, stride);
4472 if (n == 0) {
4473 tmp2 = tmp;
4474 } else {
4475 tcg_gen_shli_i32(tmp, tmp, n * 8);
4476 tcg_gen_or_i32(tmp2, tmp2, tmp);
4477 tcg_temp_free_i32(tmp);
4480 neon_store_reg(rd, pass, tmp2);
4481 } else {
4482 tmp2 = neon_load_reg(rd, pass);
4483 for (n = 0; n < 4; n++) {
4484 tmp = tcg_temp_new_i32();
4485 if (n == 0) {
4486 tcg_gen_mov_i32(tmp, tmp2);
4487 } else {
4488 tcg_gen_shri_i32(tmp, tmp2, n * 8);
4490 gen_aa32_st8(tmp, addr, get_mem_index(s));
4491 tcg_temp_free_i32(tmp);
4492 tcg_gen_addi_i32(addr, addr, stride);
4494 tcg_temp_free_i32(tmp2);
4499 rd += spacing;
4501 tcg_temp_free_i32(addr);
4502 stride = nregs * 8;
4503 } else {
4504 size = (insn >> 10) & 3;
4505 if (size == 3) {
4506 /* Load single element to all lanes. */
4507 int a = (insn >> 4) & 1;
4508 if (!load) {
4509 return 1;
4511 size = (insn >> 6) & 3;
4512 nregs = ((insn >> 8) & 3) + 1;
4514 if (size == 3) {
4515 if (nregs != 4 || a == 0) {
4516 return 1;
4518 /* For VLD4 size==3 a == 1 means 32 bits at 16 byte alignment */
4519 size = 2;
4521 if (nregs == 1 && a == 1 && size == 0) {
4522 return 1;
4524 if (nregs == 3 && a == 1) {
4525 return 1;
4527 addr = tcg_temp_new_i32();
4528 load_reg_var(s, addr, rn);
4529 if (nregs == 1) {
4530 /* VLD1 to all lanes: bit 5 indicates how many Dregs to write */
4531 tmp = gen_load_and_replicate(s, addr, size);
4532 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 0));
4533 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 1));
4534 if (insn & (1 << 5)) {
4535 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd + 1, 0));
4536 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd + 1, 1));
4538 tcg_temp_free_i32(tmp);
4539 } else {
4540 /* VLD2/3/4 to all lanes: bit 5 indicates register stride */
4541 stride = (insn & (1 << 5)) ? 2 : 1;
4542 for (reg = 0; reg < nregs; reg++) {
4543 tmp = gen_load_and_replicate(s, addr, size);
4544 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 0));
4545 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 1));
4546 tcg_temp_free_i32(tmp);
4547 tcg_gen_addi_i32(addr, addr, 1 << size);
4548 rd += stride;
4551 tcg_temp_free_i32(addr);
4552 stride = (1 << size) * nregs;
4553 } else {
4554 /* Single element. */
4555 int idx = (insn >> 4) & 0xf;
4556 pass = (insn >> 7) & 1;
4557 switch (size) {
4558 case 0:
4559 shift = ((insn >> 5) & 3) * 8;
4560 stride = 1;
4561 break;
4562 case 1:
4563 shift = ((insn >> 6) & 1) * 16;
4564 stride = (insn & (1 << 5)) ? 2 : 1;
4565 break;
4566 case 2:
4567 shift = 0;
4568 stride = (insn & (1 << 6)) ? 2 : 1;
4569 break;
4570 default:
4571 abort();
4573 nregs = ((insn >> 8) & 3) + 1;
4574 /* Catch the UNDEF cases. This is unavoidably a bit messy. */
4575 switch (nregs) {
4576 case 1:
4577 if (((idx & (1 << size)) != 0) ||
4578 (size == 2 && ((idx & 3) == 1 || (idx & 3) == 2))) {
4579 return 1;
4581 break;
4582 case 3:
4583 if ((idx & 1) != 0) {
4584 return 1;
4586 /* fall through */
4587 case 2:
4588 if (size == 2 && (idx & 2) != 0) {
4589 return 1;
4591 break;
4592 case 4:
4593 if ((size == 2) && ((idx & 3) == 3)) {
4594 return 1;
4596 break;
4597 default:
4598 abort();
4600 if ((rd + stride * (nregs - 1)) > 31) {
4601 /* Attempts to write off the end of the register file
4602 * are UNPREDICTABLE; we choose to UNDEF because otherwise
4603 * the neon_load_reg() would write off the end of the array.
4605 return 1;
4607 addr = tcg_temp_new_i32();
4608 load_reg_var(s, addr, rn);
4609 for (reg = 0; reg < nregs; reg++) {
4610 if (load) {
4611 tmp = tcg_temp_new_i32();
4612 switch (size) {
4613 case 0:
4614 gen_aa32_ld8u(tmp, addr, get_mem_index(s));
4615 break;
4616 case 1:
4617 gen_aa32_ld16u(tmp, addr, get_mem_index(s));
4618 break;
4619 case 2:
4620 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
4621 break;
4622 default: /* Avoid compiler warnings. */
4623 abort();
4625 if (size != 2) {
4626 tmp2 = neon_load_reg(rd, pass);
4627 tcg_gen_deposit_i32(tmp, tmp2, tmp,
4628 shift, size ? 16 : 8);
4629 tcg_temp_free_i32(tmp2);
4631 neon_store_reg(rd, pass, tmp);
4632 } else { /* Store */
4633 tmp = neon_load_reg(rd, pass);
4634 if (shift)
4635 tcg_gen_shri_i32(tmp, tmp, shift);
4636 switch (size) {
4637 case 0:
4638 gen_aa32_st8(tmp, addr, get_mem_index(s));
4639 break;
4640 case 1:
4641 gen_aa32_st16(tmp, addr, get_mem_index(s));
4642 break;
4643 case 2:
4644 gen_aa32_st32(tmp, addr, get_mem_index(s));
4645 break;
4647 tcg_temp_free_i32(tmp);
4649 rd += stride;
4650 tcg_gen_addi_i32(addr, addr, 1 << size);
4652 tcg_temp_free_i32(addr);
4653 stride = nregs * (1 << size);
4656 if (rm != 15) {
4657 TCGv_i32 base;
4659 base = load_reg(s, rn);
4660 if (rm == 13) {
4661 tcg_gen_addi_i32(base, base, stride);
4662 } else {
4663 TCGv_i32 index;
4664 index = load_reg(s, rm);
4665 tcg_gen_add_i32(base, base, index);
4666 tcg_temp_free_i32(index);
4668 store_reg(s, rn, base);
4670 return 0;
4673 /* Bitwise select. dest = c ? t : f. Clobbers T and F. */
4674 static void gen_neon_bsl(TCGv_i32 dest, TCGv_i32 t, TCGv_i32 f, TCGv_i32 c)
4676 tcg_gen_and_i32(t, t, c);
4677 tcg_gen_andc_i32(f, f, c);
4678 tcg_gen_or_i32(dest, t, f);
4681 static inline void gen_neon_narrow(int size, TCGv_i32 dest, TCGv_i64 src)
4683 switch (size) {
4684 case 0: gen_helper_neon_narrow_u8(dest, src); break;
4685 case 1: gen_helper_neon_narrow_u16(dest, src); break;
4686 case 2: tcg_gen_extrl_i64_i32(dest, src); break;
4687 default: abort();
4691 static inline void gen_neon_narrow_sats(int size, TCGv_i32 dest, TCGv_i64 src)
4693 switch (size) {
4694 case 0: gen_helper_neon_narrow_sat_s8(dest, cpu_env, src); break;
4695 case 1: gen_helper_neon_narrow_sat_s16(dest, cpu_env, src); break;
4696 case 2: gen_helper_neon_narrow_sat_s32(dest, cpu_env, src); break;
4697 default: abort();
4701 static inline void gen_neon_narrow_satu(int size, TCGv_i32 dest, TCGv_i64 src)
4703 switch (size) {
4704 case 0: gen_helper_neon_narrow_sat_u8(dest, cpu_env, src); break;
4705 case 1: gen_helper_neon_narrow_sat_u16(dest, cpu_env, src); break;
4706 case 2: gen_helper_neon_narrow_sat_u32(dest, cpu_env, src); break;
4707 default: abort();
4711 static inline void gen_neon_unarrow_sats(int size, TCGv_i32 dest, TCGv_i64 src)
4713 switch (size) {
4714 case 0: gen_helper_neon_unarrow_sat8(dest, cpu_env, src); break;
4715 case 1: gen_helper_neon_unarrow_sat16(dest, cpu_env, src); break;
4716 case 2: gen_helper_neon_unarrow_sat32(dest, cpu_env, src); break;
4717 default: abort();
4721 static inline void gen_neon_shift_narrow(int size, TCGv_i32 var, TCGv_i32 shift,
4722 int q, int u)
4724 if (q) {
4725 if (u) {
4726 switch (size) {
4727 case 1: gen_helper_neon_rshl_u16(var, var, shift); break;
4728 case 2: gen_helper_neon_rshl_u32(var, var, shift); break;
4729 default: abort();
4731 } else {
4732 switch (size) {
4733 case 1: gen_helper_neon_rshl_s16(var, var, shift); break;
4734 case 2: gen_helper_neon_rshl_s32(var, var, shift); break;
4735 default: abort();
4738 } else {
4739 if (u) {
4740 switch (size) {
4741 case 1: gen_helper_neon_shl_u16(var, var, shift); break;
4742 case 2: gen_helper_neon_shl_u32(var, var, shift); break;
4743 default: abort();
4745 } else {
4746 switch (size) {
4747 case 1: gen_helper_neon_shl_s16(var, var, shift); break;
4748 case 2: gen_helper_neon_shl_s32(var, var, shift); break;
4749 default: abort();
4755 static inline void gen_neon_widen(TCGv_i64 dest, TCGv_i32 src, int size, int u)
4757 if (u) {
4758 switch (size) {
4759 case 0: gen_helper_neon_widen_u8(dest, src); break;
4760 case 1: gen_helper_neon_widen_u16(dest, src); break;
4761 case 2: tcg_gen_extu_i32_i64(dest, src); break;
4762 default: abort();
4764 } else {
4765 switch (size) {
4766 case 0: gen_helper_neon_widen_s8(dest, src); break;
4767 case 1: gen_helper_neon_widen_s16(dest, src); break;
4768 case 2: tcg_gen_ext_i32_i64(dest, src); break;
4769 default: abort();
4772 tcg_temp_free_i32(src);
4775 static inline void gen_neon_addl(int size)
4777 switch (size) {
4778 case 0: gen_helper_neon_addl_u16(CPU_V001); break;
4779 case 1: gen_helper_neon_addl_u32(CPU_V001); break;
4780 case 2: tcg_gen_add_i64(CPU_V001); break;
4781 default: abort();
4785 static inline void gen_neon_subl(int size)
4787 switch (size) {
4788 case 0: gen_helper_neon_subl_u16(CPU_V001); break;
4789 case 1: gen_helper_neon_subl_u32(CPU_V001); break;
4790 case 2: tcg_gen_sub_i64(CPU_V001); break;
4791 default: abort();
4795 static inline void gen_neon_negl(TCGv_i64 var, int size)
4797 switch (size) {
4798 case 0: gen_helper_neon_negl_u16(var, var); break;
4799 case 1: gen_helper_neon_negl_u32(var, var); break;
4800 case 2:
4801 tcg_gen_neg_i64(var, var);
4802 break;
4803 default: abort();
4807 static inline void gen_neon_addl_saturate(TCGv_i64 op0, TCGv_i64 op1, int size)
4809 switch (size) {
4810 case 1: gen_helper_neon_addl_saturate_s32(op0, cpu_env, op0, op1); break;
4811 case 2: gen_helper_neon_addl_saturate_s64(op0, cpu_env, op0, op1); break;
4812 default: abort();
4816 static inline void gen_neon_mull(TCGv_i64 dest, TCGv_i32 a, TCGv_i32 b,
4817 int size, int u)
4819 TCGv_i64 tmp;
4821 switch ((size << 1) | u) {
4822 case 0: gen_helper_neon_mull_s8(dest, a, b); break;
4823 case 1: gen_helper_neon_mull_u8(dest, a, b); break;
4824 case 2: gen_helper_neon_mull_s16(dest, a, b); break;
4825 case 3: gen_helper_neon_mull_u16(dest, a, b); break;
4826 case 4:
4827 tmp = gen_muls_i64_i32(a, b);
4828 tcg_gen_mov_i64(dest, tmp);
4829 tcg_temp_free_i64(tmp);
4830 break;
4831 case 5:
4832 tmp = gen_mulu_i64_i32(a, b);
4833 tcg_gen_mov_i64(dest, tmp);
4834 tcg_temp_free_i64(tmp);
4835 break;
4836 default: abort();
4839 /* gen_helper_neon_mull_[su]{8|16} do not free their parameters.
4840 Don't forget to clean them now. */
4841 if (size < 2) {
4842 tcg_temp_free_i32(a);
4843 tcg_temp_free_i32(b);
4847 static void gen_neon_narrow_op(int op, int u, int size,
4848 TCGv_i32 dest, TCGv_i64 src)
4850 if (op) {
4851 if (u) {
4852 gen_neon_unarrow_sats(size, dest, src);
4853 } else {
4854 gen_neon_narrow(size, dest, src);
4856 } else {
4857 if (u) {
4858 gen_neon_narrow_satu(size, dest, src);
4859 } else {
4860 gen_neon_narrow_sats(size, dest, src);
4865 /* Symbolic constants for op fields for Neon 3-register same-length.
4866 * The values correspond to bits [11:8,4]; see the ARM ARM DDI0406B
4867 * table A7-9.
4869 #define NEON_3R_VHADD 0
4870 #define NEON_3R_VQADD 1
4871 #define NEON_3R_VRHADD 2
4872 #define NEON_3R_LOGIC 3 /* VAND,VBIC,VORR,VMOV,VORN,VEOR,VBIF,VBIT,VBSL */
4873 #define NEON_3R_VHSUB 4
4874 #define NEON_3R_VQSUB 5
4875 #define NEON_3R_VCGT 6
4876 #define NEON_3R_VCGE 7
4877 #define NEON_3R_VSHL 8
4878 #define NEON_3R_VQSHL 9
4879 #define NEON_3R_VRSHL 10
4880 #define NEON_3R_VQRSHL 11
4881 #define NEON_3R_VMAX 12
4882 #define NEON_3R_VMIN 13
4883 #define NEON_3R_VABD 14
4884 #define NEON_3R_VABA 15
4885 #define NEON_3R_VADD_VSUB 16
4886 #define NEON_3R_VTST_VCEQ 17
4887 #define NEON_3R_VML 18 /* VMLA, VMLAL, VMLS, VMLSL */
4888 #define NEON_3R_VMUL 19
4889 #define NEON_3R_VPMAX 20
4890 #define NEON_3R_VPMIN 21
4891 #define NEON_3R_VQDMULH_VQRDMULH 22
4892 #define NEON_3R_VPADD 23
4893 #define NEON_3R_SHA 24 /* SHA1C,SHA1P,SHA1M,SHA1SU0,SHA256H{2},SHA256SU1 */
4894 #define NEON_3R_VFM 25 /* VFMA, VFMS : float fused multiply-add */
4895 #define NEON_3R_FLOAT_ARITH 26 /* float VADD, VSUB, VPADD, VABD */
4896 #define NEON_3R_FLOAT_MULTIPLY 27 /* float VMLA, VMLS, VMUL */
4897 #define NEON_3R_FLOAT_CMP 28 /* float VCEQ, VCGE, VCGT */
4898 #define NEON_3R_FLOAT_ACMP 29 /* float VACGE, VACGT, VACLE, VACLT */
4899 #define NEON_3R_FLOAT_MINMAX 30 /* float VMIN, VMAX */
4900 #define NEON_3R_FLOAT_MISC 31 /* float VRECPS, VRSQRTS, VMAXNM/MINNM */
4902 static const uint8_t neon_3r_sizes[] = {
4903 [NEON_3R_VHADD] = 0x7,
4904 [NEON_3R_VQADD] = 0xf,
4905 [NEON_3R_VRHADD] = 0x7,
4906 [NEON_3R_LOGIC] = 0xf, /* size field encodes op type */
4907 [NEON_3R_VHSUB] = 0x7,
4908 [NEON_3R_VQSUB] = 0xf,
4909 [NEON_3R_VCGT] = 0x7,
4910 [NEON_3R_VCGE] = 0x7,
4911 [NEON_3R_VSHL] = 0xf,
4912 [NEON_3R_VQSHL] = 0xf,
4913 [NEON_3R_VRSHL] = 0xf,
4914 [NEON_3R_VQRSHL] = 0xf,
4915 [NEON_3R_VMAX] = 0x7,
4916 [NEON_3R_VMIN] = 0x7,
4917 [NEON_3R_VABD] = 0x7,
4918 [NEON_3R_VABA] = 0x7,
4919 [NEON_3R_VADD_VSUB] = 0xf,
4920 [NEON_3R_VTST_VCEQ] = 0x7,
4921 [NEON_3R_VML] = 0x7,
4922 [NEON_3R_VMUL] = 0x7,
4923 [NEON_3R_VPMAX] = 0x7,
4924 [NEON_3R_VPMIN] = 0x7,
4925 [NEON_3R_VQDMULH_VQRDMULH] = 0x6,
4926 [NEON_3R_VPADD] = 0x7,
4927 [NEON_3R_SHA] = 0xf, /* size field encodes op type */
4928 [NEON_3R_VFM] = 0x5, /* size bit 1 encodes op */
4929 [NEON_3R_FLOAT_ARITH] = 0x5, /* size bit 1 encodes op */
4930 [NEON_3R_FLOAT_MULTIPLY] = 0x5, /* size bit 1 encodes op */
4931 [NEON_3R_FLOAT_CMP] = 0x5, /* size bit 1 encodes op */
4932 [NEON_3R_FLOAT_ACMP] = 0x5, /* size bit 1 encodes op */
4933 [NEON_3R_FLOAT_MINMAX] = 0x5, /* size bit 1 encodes op */
4934 [NEON_3R_FLOAT_MISC] = 0x5, /* size bit 1 encodes op */
4937 /* Symbolic constants for op fields for Neon 2-register miscellaneous.
4938 * The values correspond to bits [17:16,10:7]; see the ARM ARM DDI0406B
4939 * table A7-13.
4941 #define NEON_2RM_VREV64 0
4942 #define NEON_2RM_VREV32 1
4943 #define NEON_2RM_VREV16 2
4944 #define NEON_2RM_VPADDL 4
4945 #define NEON_2RM_VPADDL_U 5
4946 #define NEON_2RM_AESE 6 /* Includes AESD */
4947 #define NEON_2RM_AESMC 7 /* Includes AESIMC */
4948 #define NEON_2RM_VCLS 8
4949 #define NEON_2RM_VCLZ 9
4950 #define NEON_2RM_VCNT 10
4951 #define NEON_2RM_VMVN 11
4952 #define NEON_2RM_VPADAL 12
4953 #define NEON_2RM_VPADAL_U 13
4954 #define NEON_2RM_VQABS 14
4955 #define NEON_2RM_VQNEG 15
4956 #define NEON_2RM_VCGT0 16
4957 #define NEON_2RM_VCGE0 17
4958 #define NEON_2RM_VCEQ0 18
4959 #define NEON_2RM_VCLE0 19
4960 #define NEON_2RM_VCLT0 20
4961 #define NEON_2RM_SHA1H 21
4962 #define NEON_2RM_VABS 22
4963 #define NEON_2RM_VNEG 23
4964 #define NEON_2RM_VCGT0_F 24
4965 #define NEON_2RM_VCGE0_F 25
4966 #define NEON_2RM_VCEQ0_F 26
4967 #define NEON_2RM_VCLE0_F 27
4968 #define NEON_2RM_VCLT0_F 28
4969 #define NEON_2RM_VABS_F 30
4970 #define NEON_2RM_VNEG_F 31
4971 #define NEON_2RM_VSWP 32
4972 #define NEON_2RM_VTRN 33
4973 #define NEON_2RM_VUZP 34
4974 #define NEON_2RM_VZIP 35
4975 #define NEON_2RM_VMOVN 36 /* Includes VQMOVN, VQMOVUN */
4976 #define NEON_2RM_VQMOVN 37 /* Includes VQMOVUN */
4977 #define NEON_2RM_VSHLL 38
4978 #define NEON_2RM_SHA1SU1 39 /* Includes SHA256SU0 */
4979 #define NEON_2RM_VRINTN 40
4980 #define NEON_2RM_VRINTX 41
4981 #define NEON_2RM_VRINTA 42
4982 #define NEON_2RM_VRINTZ 43
4983 #define NEON_2RM_VCVT_F16_F32 44
4984 #define NEON_2RM_VRINTM 45
4985 #define NEON_2RM_VCVT_F32_F16 46
4986 #define NEON_2RM_VRINTP 47
4987 #define NEON_2RM_VCVTAU 48
4988 #define NEON_2RM_VCVTAS 49
4989 #define NEON_2RM_VCVTNU 50
4990 #define NEON_2RM_VCVTNS 51
4991 #define NEON_2RM_VCVTPU 52
4992 #define NEON_2RM_VCVTPS 53
4993 #define NEON_2RM_VCVTMU 54
4994 #define NEON_2RM_VCVTMS 55
4995 #define NEON_2RM_VRECPE 56
4996 #define NEON_2RM_VRSQRTE 57
4997 #define NEON_2RM_VRECPE_F 58
4998 #define NEON_2RM_VRSQRTE_F 59
4999 #define NEON_2RM_VCVT_FS 60
5000 #define NEON_2RM_VCVT_FU 61
5001 #define NEON_2RM_VCVT_SF 62
5002 #define NEON_2RM_VCVT_UF 63
5004 static int neon_2rm_is_float_op(int op)
5006 /* Return true if this neon 2reg-misc op is float-to-float */
5007 return (op == NEON_2RM_VABS_F || op == NEON_2RM_VNEG_F ||
5008 (op >= NEON_2RM_VRINTN && op <= NEON_2RM_VRINTZ) ||
5009 op == NEON_2RM_VRINTM ||
5010 (op >= NEON_2RM_VRINTP && op <= NEON_2RM_VCVTMS) ||
5011 op >= NEON_2RM_VRECPE_F);
5014 /* Each entry in this array has bit n set if the insn allows
5015 * size value n (otherwise it will UNDEF). Since unallocated
5016 * op values will have no bits set they always UNDEF.
5018 static const uint8_t neon_2rm_sizes[] = {
5019 [NEON_2RM_VREV64] = 0x7,
5020 [NEON_2RM_VREV32] = 0x3,
5021 [NEON_2RM_VREV16] = 0x1,
5022 [NEON_2RM_VPADDL] = 0x7,
5023 [NEON_2RM_VPADDL_U] = 0x7,
5024 [NEON_2RM_AESE] = 0x1,
5025 [NEON_2RM_AESMC] = 0x1,
5026 [NEON_2RM_VCLS] = 0x7,
5027 [NEON_2RM_VCLZ] = 0x7,
5028 [NEON_2RM_VCNT] = 0x1,
5029 [NEON_2RM_VMVN] = 0x1,
5030 [NEON_2RM_VPADAL] = 0x7,
5031 [NEON_2RM_VPADAL_U] = 0x7,
5032 [NEON_2RM_VQABS] = 0x7,
5033 [NEON_2RM_VQNEG] = 0x7,
5034 [NEON_2RM_VCGT0] = 0x7,
5035 [NEON_2RM_VCGE0] = 0x7,
5036 [NEON_2RM_VCEQ0] = 0x7,
5037 [NEON_2RM_VCLE0] = 0x7,
5038 [NEON_2RM_VCLT0] = 0x7,
5039 [NEON_2RM_SHA1H] = 0x4,
5040 [NEON_2RM_VABS] = 0x7,
5041 [NEON_2RM_VNEG] = 0x7,
5042 [NEON_2RM_VCGT0_F] = 0x4,
5043 [NEON_2RM_VCGE0_F] = 0x4,
5044 [NEON_2RM_VCEQ0_F] = 0x4,
5045 [NEON_2RM_VCLE0_F] = 0x4,
5046 [NEON_2RM_VCLT0_F] = 0x4,
5047 [NEON_2RM_VABS_F] = 0x4,
5048 [NEON_2RM_VNEG_F] = 0x4,
5049 [NEON_2RM_VSWP] = 0x1,
5050 [NEON_2RM_VTRN] = 0x7,
5051 [NEON_2RM_VUZP] = 0x7,
5052 [NEON_2RM_VZIP] = 0x7,
5053 [NEON_2RM_VMOVN] = 0x7,
5054 [NEON_2RM_VQMOVN] = 0x7,
5055 [NEON_2RM_VSHLL] = 0x7,
5056 [NEON_2RM_SHA1SU1] = 0x4,
5057 [NEON_2RM_VRINTN] = 0x4,
5058 [NEON_2RM_VRINTX] = 0x4,
5059 [NEON_2RM_VRINTA] = 0x4,
5060 [NEON_2RM_VRINTZ] = 0x4,
5061 [NEON_2RM_VCVT_F16_F32] = 0x2,
5062 [NEON_2RM_VRINTM] = 0x4,
5063 [NEON_2RM_VCVT_F32_F16] = 0x2,
5064 [NEON_2RM_VRINTP] = 0x4,
5065 [NEON_2RM_VCVTAU] = 0x4,
5066 [NEON_2RM_VCVTAS] = 0x4,
5067 [NEON_2RM_VCVTNU] = 0x4,
5068 [NEON_2RM_VCVTNS] = 0x4,
5069 [NEON_2RM_VCVTPU] = 0x4,
5070 [NEON_2RM_VCVTPS] = 0x4,
5071 [NEON_2RM_VCVTMU] = 0x4,
5072 [NEON_2RM_VCVTMS] = 0x4,
5073 [NEON_2RM_VRECPE] = 0x4,
5074 [NEON_2RM_VRSQRTE] = 0x4,
5075 [NEON_2RM_VRECPE_F] = 0x4,
5076 [NEON_2RM_VRSQRTE_F] = 0x4,
5077 [NEON_2RM_VCVT_FS] = 0x4,
5078 [NEON_2RM_VCVT_FU] = 0x4,
5079 [NEON_2RM_VCVT_SF] = 0x4,
5080 [NEON_2RM_VCVT_UF] = 0x4,
5083 /* Translate a NEON data processing instruction. Return nonzero if the
5084 instruction is invalid.
5085 We process data in a mixture of 32-bit and 64-bit chunks.
5086 Mostly we use 32-bit chunks so we can use normal scalar instructions. */
5088 static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
5090 int op;
5091 int q;
5092 int rd, rn, rm;
5093 int size;
5094 int shift;
5095 int pass;
5096 int count;
5097 int pairwise;
5098 int u;
5099 uint32_t imm, mask;
5100 TCGv_i32 tmp, tmp2, tmp3, tmp4, tmp5;
5101 TCGv_i64 tmp64;
5103 /* FIXME: this access check should not take precedence over UNDEF
5104 * for invalid encodings; we will generate incorrect syndrome information
5105 * for attempts to execute invalid vfp/neon encodings with FP disabled.
5107 if (s->fp_excp_el) {
5108 gen_exception_insn(s, 4, EXCP_UDEF,
5109 syn_fp_access_trap(1, 0xe, s->thumb), s->fp_excp_el);
5110 return 0;
5113 if (!s->vfp_enabled)
5114 return 1;
5115 q = (insn & (1 << 6)) != 0;
5116 u = (insn >> 24) & 1;
5117 VFP_DREG_D(rd, insn);
5118 VFP_DREG_N(rn, insn);
5119 VFP_DREG_M(rm, insn);
5120 size = (insn >> 20) & 3;
5121 if ((insn & (1 << 23)) == 0) {
5122 /* Three register same length. */
5123 op = ((insn >> 7) & 0x1e) | ((insn >> 4) & 1);
5124 /* Catch invalid op and bad size combinations: UNDEF */
5125 if ((neon_3r_sizes[op] & (1 << size)) == 0) {
5126 return 1;
5128 /* All insns of this form UNDEF for either this condition or the
5129 * superset of cases "Q==1"; we catch the latter later.
5131 if (q && ((rd | rn | rm) & 1)) {
5132 return 1;
5135 * The SHA-1/SHA-256 3-register instructions require special treatment
5136 * here, as their size field is overloaded as an op type selector, and
5137 * they all consume their input in a single pass.
5139 if (op == NEON_3R_SHA) {
5140 if (!q) {
5141 return 1;
5143 if (!u) { /* SHA-1 */
5144 if (!arm_dc_feature(s, ARM_FEATURE_V8_SHA1)) {
5145 return 1;
5147 tmp = tcg_const_i32(rd);
5148 tmp2 = tcg_const_i32(rn);
5149 tmp3 = tcg_const_i32(rm);
5150 tmp4 = tcg_const_i32(size);
5151 gen_helper_crypto_sha1_3reg(cpu_env, tmp, tmp2, tmp3, tmp4);
5152 tcg_temp_free_i32(tmp4);
5153 } else { /* SHA-256 */
5154 if (!arm_dc_feature(s, ARM_FEATURE_V8_SHA256) || size == 3) {
5155 return 1;
5157 tmp = tcg_const_i32(rd);
5158 tmp2 = tcg_const_i32(rn);
5159 tmp3 = tcg_const_i32(rm);
5160 switch (size) {
5161 case 0:
5162 gen_helper_crypto_sha256h(cpu_env, tmp, tmp2, tmp3);
5163 break;
5164 case 1:
5165 gen_helper_crypto_sha256h2(cpu_env, tmp, tmp2, tmp3);
5166 break;
5167 case 2:
5168 gen_helper_crypto_sha256su1(cpu_env, tmp, tmp2, tmp3);
5169 break;
5172 tcg_temp_free_i32(tmp);
5173 tcg_temp_free_i32(tmp2);
5174 tcg_temp_free_i32(tmp3);
5175 return 0;
5177 if (size == 3 && op != NEON_3R_LOGIC) {
5178 /* 64-bit element instructions. */
5179 for (pass = 0; pass < (q ? 2 : 1); pass++) {
5180 neon_load_reg64(cpu_V0, rn + pass);
5181 neon_load_reg64(cpu_V1, rm + pass);
5182 switch (op) {
5183 case NEON_3R_VQADD:
5184 if (u) {
5185 gen_helper_neon_qadd_u64(cpu_V0, cpu_env,
5186 cpu_V0, cpu_V1);
5187 } else {
5188 gen_helper_neon_qadd_s64(cpu_V0, cpu_env,
5189 cpu_V0, cpu_V1);
5191 break;
5192 case NEON_3R_VQSUB:
5193 if (u) {
5194 gen_helper_neon_qsub_u64(cpu_V0, cpu_env,
5195 cpu_V0, cpu_V1);
5196 } else {
5197 gen_helper_neon_qsub_s64(cpu_V0, cpu_env,
5198 cpu_V0, cpu_V1);
5200 break;
5201 case NEON_3R_VSHL:
5202 if (u) {
5203 gen_helper_neon_shl_u64(cpu_V0, cpu_V1, cpu_V0);
5204 } else {
5205 gen_helper_neon_shl_s64(cpu_V0, cpu_V1, cpu_V0);
5207 break;
5208 case NEON_3R_VQSHL:
5209 if (u) {
5210 gen_helper_neon_qshl_u64(cpu_V0, cpu_env,
5211 cpu_V1, cpu_V0);
5212 } else {
5213 gen_helper_neon_qshl_s64(cpu_V0, cpu_env,
5214 cpu_V1, cpu_V0);
5216 break;
5217 case NEON_3R_VRSHL:
5218 if (u) {
5219 gen_helper_neon_rshl_u64(cpu_V0, cpu_V1, cpu_V0);
5220 } else {
5221 gen_helper_neon_rshl_s64(cpu_V0, cpu_V1, cpu_V0);
5223 break;
5224 case NEON_3R_VQRSHL:
5225 if (u) {
5226 gen_helper_neon_qrshl_u64(cpu_V0, cpu_env,
5227 cpu_V1, cpu_V0);
5228 } else {
5229 gen_helper_neon_qrshl_s64(cpu_V0, cpu_env,
5230 cpu_V1, cpu_V0);
5232 break;
5233 case NEON_3R_VADD_VSUB:
5234 if (u) {
5235 tcg_gen_sub_i64(CPU_V001);
5236 } else {
5237 tcg_gen_add_i64(CPU_V001);
5239 break;
5240 default:
5241 abort();
5243 neon_store_reg64(cpu_V0, rd + pass);
5245 return 0;
5247 pairwise = 0;
5248 switch (op) {
5249 case NEON_3R_VSHL:
5250 case NEON_3R_VQSHL:
5251 case NEON_3R_VRSHL:
5252 case NEON_3R_VQRSHL:
5254 int rtmp;
5255 /* Shift instruction operands are reversed. */
5256 rtmp = rn;
5257 rn = rm;
5258 rm = rtmp;
5260 break;
5261 case NEON_3R_VPADD:
5262 if (u) {
5263 return 1;
5265 /* Fall through */
5266 case NEON_3R_VPMAX:
5267 case NEON_3R_VPMIN:
5268 pairwise = 1;
5269 break;
5270 case NEON_3R_FLOAT_ARITH:
5271 pairwise = (u && size < 2); /* if VPADD (float) */
5272 break;
5273 case NEON_3R_FLOAT_MINMAX:
5274 pairwise = u; /* if VPMIN/VPMAX (float) */
5275 break;
5276 case NEON_3R_FLOAT_CMP:
5277 if (!u && size) {
5278 /* no encoding for U=0 C=1x */
5279 return 1;
5281 break;
5282 case NEON_3R_FLOAT_ACMP:
5283 if (!u) {
5284 return 1;
5286 break;
5287 case NEON_3R_FLOAT_MISC:
5288 /* VMAXNM/VMINNM in ARMv8 */
5289 if (u && !arm_dc_feature(s, ARM_FEATURE_V8)) {
5290 return 1;
5292 break;
5293 case NEON_3R_VMUL:
5294 if (u && (size != 0)) {
5295 /* UNDEF on invalid size for polynomial subcase */
5296 return 1;
5298 break;
5299 case NEON_3R_VFM:
5300 if (!arm_dc_feature(s, ARM_FEATURE_VFP4) || u) {
5301 return 1;
5303 break;
5304 default:
5305 break;
5308 if (pairwise && q) {
5309 /* All the pairwise insns UNDEF if Q is set */
5310 return 1;
5313 for (pass = 0; pass < (q ? 4 : 2); pass++) {
5315 if (pairwise) {
5316 /* Pairwise. */
5317 if (pass < 1) {
5318 tmp = neon_load_reg(rn, 0);
5319 tmp2 = neon_load_reg(rn, 1);
5320 } else {
5321 tmp = neon_load_reg(rm, 0);
5322 tmp2 = neon_load_reg(rm, 1);
5324 } else {
5325 /* Elementwise. */
5326 tmp = neon_load_reg(rn, pass);
5327 tmp2 = neon_load_reg(rm, pass);
5329 switch (op) {
5330 case NEON_3R_VHADD:
5331 GEN_NEON_INTEGER_OP(hadd);
5332 break;
5333 case NEON_3R_VQADD:
5334 GEN_NEON_INTEGER_OP_ENV(qadd);
5335 break;
5336 case NEON_3R_VRHADD:
5337 GEN_NEON_INTEGER_OP(rhadd);
5338 break;
5339 case NEON_3R_LOGIC: /* Logic ops. */
5340 switch ((u << 2) | size) {
5341 case 0: /* VAND */
5342 tcg_gen_and_i32(tmp, tmp, tmp2);
5343 break;
5344 case 1: /* BIC */
5345 tcg_gen_andc_i32(tmp, tmp, tmp2);
5346 break;
5347 case 2: /* VORR */
5348 tcg_gen_or_i32(tmp, tmp, tmp2);
5349 break;
5350 case 3: /* VORN */
5351 tcg_gen_orc_i32(tmp, tmp, tmp2);
5352 break;
5353 case 4: /* VEOR */
5354 tcg_gen_xor_i32(tmp, tmp, tmp2);
5355 break;
5356 case 5: /* VBSL */
5357 tmp3 = neon_load_reg(rd, pass);
5358 gen_neon_bsl(tmp, tmp, tmp2, tmp3);
5359 tcg_temp_free_i32(tmp3);
5360 break;
5361 case 6: /* VBIT */
5362 tmp3 = neon_load_reg(rd, pass);
5363 gen_neon_bsl(tmp, tmp, tmp3, tmp2);
5364 tcg_temp_free_i32(tmp3);
5365 break;
5366 case 7: /* VBIF */
5367 tmp3 = neon_load_reg(rd, pass);
5368 gen_neon_bsl(tmp, tmp3, tmp, tmp2);
5369 tcg_temp_free_i32(tmp3);
5370 break;
5372 break;
5373 case NEON_3R_VHSUB:
5374 GEN_NEON_INTEGER_OP(hsub);
5375 break;
5376 case NEON_3R_VQSUB:
5377 GEN_NEON_INTEGER_OP_ENV(qsub);
5378 break;
5379 case NEON_3R_VCGT:
5380 GEN_NEON_INTEGER_OP(cgt);
5381 break;
5382 case NEON_3R_VCGE:
5383 GEN_NEON_INTEGER_OP(cge);
5384 break;
5385 case NEON_3R_VSHL:
5386 GEN_NEON_INTEGER_OP(shl);
5387 break;
5388 case NEON_3R_VQSHL:
5389 GEN_NEON_INTEGER_OP_ENV(qshl);
5390 break;
5391 case NEON_3R_VRSHL:
5392 GEN_NEON_INTEGER_OP(rshl);
5393 break;
5394 case NEON_3R_VQRSHL:
5395 GEN_NEON_INTEGER_OP_ENV(qrshl);
5396 break;
5397 case NEON_3R_VMAX:
5398 GEN_NEON_INTEGER_OP(max);
5399 break;
5400 case NEON_3R_VMIN:
5401 GEN_NEON_INTEGER_OP(min);
5402 break;
5403 case NEON_3R_VABD:
5404 GEN_NEON_INTEGER_OP(abd);
5405 break;
5406 case NEON_3R_VABA:
5407 GEN_NEON_INTEGER_OP(abd);
5408 tcg_temp_free_i32(tmp2);
5409 tmp2 = neon_load_reg(rd, pass);
5410 gen_neon_add(size, tmp, tmp2);
5411 break;
5412 case NEON_3R_VADD_VSUB:
5413 if (!u) { /* VADD */
5414 gen_neon_add(size, tmp, tmp2);
5415 } else { /* VSUB */
5416 switch (size) {
5417 case 0: gen_helper_neon_sub_u8(tmp, tmp, tmp2); break;
5418 case 1: gen_helper_neon_sub_u16(tmp, tmp, tmp2); break;
5419 case 2: tcg_gen_sub_i32(tmp, tmp, tmp2); break;
5420 default: abort();
5423 break;
5424 case NEON_3R_VTST_VCEQ:
5425 if (!u) { /* VTST */
5426 switch (size) {
5427 case 0: gen_helper_neon_tst_u8(tmp, tmp, tmp2); break;
5428 case 1: gen_helper_neon_tst_u16(tmp, tmp, tmp2); break;
5429 case 2: gen_helper_neon_tst_u32(tmp, tmp, tmp2); break;
5430 default: abort();
5432 } else { /* VCEQ */
5433 switch (size) {
5434 case 0: gen_helper_neon_ceq_u8(tmp, tmp, tmp2); break;
5435 case 1: gen_helper_neon_ceq_u16(tmp, tmp, tmp2); break;
5436 case 2: gen_helper_neon_ceq_u32(tmp, tmp, tmp2); break;
5437 default: abort();
5440 break;
5441 case NEON_3R_VML: /* VMLA, VMLAL, VMLS,VMLSL */
5442 switch (size) {
5443 case 0: gen_helper_neon_mul_u8(tmp, tmp, tmp2); break;
5444 case 1: gen_helper_neon_mul_u16(tmp, tmp, tmp2); break;
5445 case 2: tcg_gen_mul_i32(tmp, tmp, tmp2); break;
5446 default: abort();
5448 tcg_temp_free_i32(tmp2);
5449 tmp2 = neon_load_reg(rd, pass);
5450 if (u) { /* VMLS */
5451 gen_neon_rsb(size, tmp, tmp2);
5452 } else { /* VMLA */
5453 gen_neon_add(size, tmp, tmp2);
5455 break;
5456 case NEON_3R_VMUL:
5457 if (u) { /* polynomial */
5458 gen_helper_neon_mul_p8(tmp, tmp, tmp2);
5459 } else { /* Integer */
5460 switch (size) {
5461 case 0: gen_helper_neon_mul_u8(tmp, tmp, tmp2); break;
5462 case 1: gen_helper_neon_mul_u16(tmp, tmp, tmp2); break;
5463 case 2: tcg_gen_mul_i32(tmp, tmp, tmp2); break;
5464 default: abort();
5467 break;
5468 case NEON_3R_VPMAX:
5469 GEN_NEON_INTEGER_OP(pmax);
5470 break;
5471 case NEON_3R_VPMIN:
5472 GEN_NEON_INTEGER_OP(pmin);
5473 break;
5474 case NEON_3R_VQDMULH_VQRDMULH: /* Multiply high. */
5475 if (!u) { /* VQDMULH */
5476 switch (size) {
5477 case 1:
5478 gen_helper_neon_qdmulh_s16(tmp, cpu_env, tmp, tmp2);
5479 break;
5480 case 2:
5481 gen_helper_neon_qdmulh_s32(tmp, cpu_env, tmp, tmp2);
5482 break;
5483 default: abort();
5485 } else { /* VQRDMULH */
5486 switch (size) {
5487 case 1:
5488 gen_helper_neon_qrdmulh_s16(tmp, cpu_env, tmp, tmp2);
5489 break;
5490 case 2:
5491 gen_helper_neon_qrdmulh_s32(tmp, cpu_env, tmp, tmp2);
5492 break;
5493 default: abort();
5496 break;
5497 case NEON_3R_VPADD:
5498 switch (size) {
5499 case 0: gen_helper_neon_padd_u8(tmp, tmp, tmp2); break;
5500 case 1: gen_helper_neon_padd_u16(tmp, tmp, tmp2); break;
5501 case 2: tcg_gen_add_i32(tmp, tmp, tmp2); break;
5502 default: abort();
5504 break;
5505 case NEON_3R_FLOAT_ARITH: /* Floating point arithmetic. */
5507 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5508 switch ((u << 2) | size) {
5509 case 0: /* VADD */
5510 case 4: /* VPADD */
5511 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
5512 break;
5513 case 2: /* VSUB */
5514 gen_helper_vfp_subs(tmp, tmp, tmp2, fpstatus);
5515 break;
5516 case 6: /* VABD */
5517 gen_helper_neon_abd_f32(tmp, tmp, tmp2, fpstatus);
5518 break;
5519 default:
5520 abort();
5522 tcg_temp_free_ptr(fpstatus);
5523 break;
5525 case NEON_3R_FLOAT_MULTIPLY:
5527 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5528 gen_helper_vfp_muls(tmp, tmp, tmp2, fpstatus);
5529 if (!u) {
5530 tcg_temp_free_i32(tmp2);
5531 tmp2 = neon_load_reg(rd, pass);
5532 if (size == 0) {
5533 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
5534 } else {
5535 gen_helper_vfp_subs(tmp, tmp2, tmp, fpstatus);
5538 tcg_temp_free_ptr(fpstatus);
5539 break;
5541 case NEON_3R_FLOAT_CMP:
5543 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5544 if (!u) {
5545 gen_helper_neon_ceq_f32(tmp, tmp, tmp2, fpstatus);
5546 } else {
5547 if (size == 0) {
5548 gen_helper_neon_cge_f32(tmp, tmp, tmp2, fpstatus);
5549 } else {
5550 gen_helper_neon_cgt_f32(tmp, tmp, tmp2, fpstatus);
5553 tcg_temp_free_ptr(fpstatus);
5554 break;
5556 case NEON_3R_FLOAT_ACMP:
5558 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5559 if (size == 0) {
5560 gen_helper_neon_acge_f32(tmp, tmp, tmp2, fpstatus);
5561 } else {
5562 gen_helper_neon_acgt_f32(tmp, tmp, tmp2, fpstatus);
5564 tcg_temp_free_ptr(fpstatus);
5565 break;
5567 case NEON_3R_FLOAT_MINMAX:
5569 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5570 if (size == 0) {
5571 gen_helper_vfp_maxs(tmp, tmp, tmp2, fpstatus);
5572 } else {
5573 gen_helper_vfp_mins(tmp, tmp, tmp2, fpstatus);
5575 tcg_temp_free_ptr(fpstatus);
5576 break;
5578 case NEON_3R_FLOAT_MISC:
5579 if (u) {
5580 /* VMAXNM/VMINNM */
5581 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5582 if (size == 0) {
5583 gen_helper_vfp_maxnums(tmp, tmp, tmp2, fpstatus);
5584 } else {
5585 gen_helper_vfp_minnums(tmp, tmp, tmp2, fpstatus);
5587 tcg_temp_free_ptr(fpstatus);
5588 } else {
5589 if (size == 0) {
5590 gen_helper_recps_f32(tmp, tmp, tmp2, cpu_env);
5591 } else {
5592 gen_helper_rsqrts_f32(tmp, tmp, tmp2, cpu_env);
5595 break;
5596 case NEON_3R_VFM:
5598 /* VFMA, VFMS: fused multiply-add */
5599 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5600 TCGv_i32 tmp3 = neon_load_reg(rd, pass);
5601 if (size) {
5602 /* VFMS */
5603 gen_helper_vfp_negs(tmp, tmp);
5605 gen_helper_vfp_muladds(tmp, tmp, tmp2, tmp3, fpstatus);
5606 tcg_temp_free_i32(tmp3);
5607 tcg_temp_free_ptr(fpstatus);
5608 break;
5610 default:
5611 abort();
5613 tcg_temp_free_i32(tmp2);
5615 /* Save the result. For elementwise operations we can put it
5616 straight into the destination register. For pairwise operations
5617 we have to be careful to avoid clobbering the source operands. */
5618 if (pairwise && rd == rm) {
5619 neon_store_scratch(pass, tmp);
5620 } else {
5621 neon_store_reg(rd, pass, tmp);
5624 } /* for pass */
5625 if (pairwise && rd == rm) {
5626 for (pass = 0; pass < (q ? 4 : 2); pass++) {
5627 tmp = neon_load_scratch(pass);
5628 neon_store_reg(rd, pass, tmp);
5631 /* End of 3 register same size operations. */
5632 } else if (insn & (1 << 4)) {
5633 if ((insn & 0x00380080) != 0) {
5634 /* Two registers and shift. */
5635 op = (insn >> 8) & 0xf;
5636 if (insn & (1 << 7)) {
5637 /* 64-bit shift. */
5638 if (op > 7) {
5639 return 1;
5641 size = 3;
5642 } else {
5643 size = 2;
5644 while ((insn & (1 << (size + 19))) == 0)
5645 size--;
5647 shift = (insn >> 16) & ((1 << (3 + size)) - 1);
5648 /* To avoid excessive duplication of ops we implement shift
5649 by immediate using the variable shift operations. */
5650 if (op < 8) {
5651 /* Shift by immediate:
5652 VSHR, VSRA, VRSHR, VRSRA, VSRI, VSHL, VQSHL, VQSHLU. */
5653 if (q && ((rd | rm) & 1)) {
5654 return 1;
5656 if (!u && (op == 4 || op == 6)) {
5657 return 1;
5659 /* Right shifts are encoded as N - shift, where N is the
5660 element size in bits. */
5661 if (op <= 4)
5662 shift = shift - (1 << (size + 3));
5663 if (size == 3) {
5664 count = q + 1;
5665 } else {
5666 count = q ? 4: 2;
5668 switch (size) {
5669 case 0:
5670 imm = (uint8_t) shift;
5671 imm |= imm << 8;
5672 imm |= imm << 16;
5673 break;
5674 case 1:
5675 imm = (uint16_t) shift;
5676 imm |= imm << 16;
5677 break;
5678 case 2:
5679 case 3:
5680 imm = shift;
5681 break;
5682 default:
5683 abort();
5686 for (pass = 0; pass < count; pass++) {
5687 if (size == 3) {
5688 neon_load_reg64(cpu_V0, rm + pass);
5689 tcg_gen_movi_i64(cpu_V1, imm);
5690 switch (op) {
5691 case 0: /* VSHR */
5692 case 1: /* VSRA */
5693 if (u)
5694 gen_helper_neon_shl_u64(cpu_V0, cpu_V0, cpu_V1);
5695 else
5696 gen_helper_neon_shl_s64(cpu_V0, cpu_V0, cpu_V1);
5697 break;
5698 case 2: /* VRSHR */
5699 case 3: /* VRSRA */
5700 if (u)
5701 gen_helper_neon_rshl_u64(cpu_V0, cpu_V0, cpu_V1);
5702 else
5703 gen_helper_neon_rshl_s64(cpu_V0, cpu_V0, cpu_V1);
5704 break;
5705 case 4: /* VSRI */
5706 case 5: /* VSHL, VSLI */
5707 gen_helper_neon_shl_u64(cpu_V0, cpu_V0, cpu_V1);
5708 break;
5709 case 6: /* VQSHLU */
5710 gen_helper_neon_qshlu_s64(cpu_V0, cpu_env,
5711 cpu_V0, cpu_V1);
5712 break;
5713 case 7: /* VQSHL */
5714 if (u) {
5715 gen_helper_neon_qshl_u64(cpu_V0, cpu_env,
5716 cpu_V0, cpu_V1);
5717 } else {
5718 gen_helper_neon_qshl_s64(cpu_V0, cpu_env,
5719 cpu_V0, cpu_V1);
5721 break;
5723 if (op == 1 || op == 3) {
5724 /* Accumulate. */
5725 neon_load_reg64(cpu_V1, rd + pass);
5726 tcg_gen_add_i64(cpu_V0, cpu_V0, cpu_V1);
5727 } else if (op == 4 || (op == 5 && u)) {
5728 /* Insert */
5729 neon_load_reg64(cpu_V1, rd + pass);
5730 uint64_t mask;
5731 if (shift < -63 || shift > 63) {
5732 mask = 0;
5733 } else {
5734 if (op == 4) {
5735 mask = 0xffffffffffffffffull >> -shift;
5736 } else {
5737 mask = 0xffffffffffffffffull << shift;
5740 tcg_gen_andi_i64(cpu_V1, cpu_V1, ~mask);
5741 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
5743 neon_store_reg64(cpu_V0, rd + pass);
5744 } else { /* size < 3 */
5745 /* Operands in T0 and T1. */
5746 tmp = neon_load_reg(rm, pass);
5747 tmp2 = tcg_temp_new_i32();
5748 tcg_gen_movi_i32(tmp2, imm);
5749 switch (op) {
5750 case 0: /* VSHR */
5751 case 1: /* VSRA */
5752 GEN_NEON_INTEGER_OP(shl);
5753 break;
5754 case 2: /* VRSHR */
5755 case 3: /* VRSRA */
5756 GEN_NEON_INTEGER_OP(rshl);
5757 break;
5758 case 4: /* VSRI */
5759 case 5: /* VSHL, VSLI */
5760 switch (size) {
5761 case 0: gen_helper_neon_shl_u8(tmp, tmp, tmp2); break;
5762 case 1: gen_helper_neon_shl_u16(tmp, tmp, tmp2); break;
5763 case 2: gen_helper_neon_shl_u32(tmp, tmp, tmp2); break;
5764 default: abort();
5766 break;
5767 case 6: /* VQSHLU */
5768 switch (size) {
5769 case 0:
5770 gen_helper_neon_qshlu_s8(tmp, cpu_env,
5771 tmp, tmp2);
5772 break;
5773 case 1:
5774 gen_helper_neon_qshlu_s16(tmp, cpu_env,
5775 tmp, tmp2);
5776 break;
5777 case 2:
5778 gen_helper_neon_qshlu_s32(tmp, cpu_env,
5779 tmp, tmp2);
5780 break;
5781 default:
5782 abort();
5784 break;
5785 case 7: /* VQSHL */
5786 GEN_NEON_INTEGER_OP_ENV(qshl);
5787 break;
5789 tcg_temp_free_i32(tmp2);
5791 if (op == 1 || op == 3) {
5792 /* Accumulate. */
5793 tmp2 = neon_load_reg(rd, pass);
5794 gen_neon_add(size, tmp, tmp2);
5795 tcg_temp_free_i32(tmp2);
5796 } else if (op == 4 || (op == 5 && u)) {
5797 /* Insert */
5798 switch (size) {
5799 case 0:
5800 if (op == 4)
5801 mask = 0xff >> -shift;
5802 else
5803 mask = (uint8_t)(0xff << shift);
5804 mask |= mask << 8;
5805 mask |= mask << 16;
5806 break;
5807 case 1:
5808 if (op == 4)
5809 mask = 0xffff >> -shift;
5810 else
5811 mask = (uint16_t)(0xffff << shift);
5812 mask |= mask << 16;
5813 break;
5814 case 2:
5815 if (shift < -31 || shift > 31) {
5816 mask = 0;
5817 } else {
5818 if (op == 4)
5819 mask = 0xffffffffu >> -shift;
5820 else
5821 mask = 0xffffffffu << shift;
5823 break;
5824 default:
5825 abort();
5827 tmp2 = neon_load_reg(rd, pass);
5828 tcg_gen_andi_i32(tmp, tmp, mask);
5829 tcg_gen_andi_i32(tmp2, tmp2, ~mask);
5830 tcg_gen_or_i32(tmp, tmp, tmp2);
5831 tcg_temp_free_i32(tmp2);
5833 neon_store_reg(rd, pass, tmp);
5835 } /* for pass */
5836 } else if (op < 10) {
5837 /* Shift by immediate and narrow:
5838 VSHRN, VRSHRN, VQSHRN, VQRSHRN. */
5839 int input_unsigned = (op == 8) ? !u : u;
5840 if (rm & 1) {
5841 return 1;
5843 shift = shift - (1 << (size + 3));
5844 size++;
5845 if (size == 3) {
5846 tmp64 = tcg_const_i64(shift);
5847 neon_load_reg64(cpu_V0, rm);
5848 neon_load_reg64(cpu_V1, rm + 1);
5849 for (pass = 0; pass < 2; pass++) {
5850 TCGv_i64 in;
5851 if (pass == 0) {
5852 in = cpu_V0;
5853 } else {
5854 in = cpu_V1;
5856 if (q) {
5857 if (input_unsigned) {
5858 gen_helper_neon_rshl_u64(cpu_V0, in, tmp64);
5859 } else {
5860 gen_helper_neon_rshl_s64(cpu_V0, in, tmp64);
5862 } else {
5863 if (input_unsigned) {
5864 gen_helper_neon_shl_u64(cpu_V0, in, tmp64);
5865 } else {
5866 gen_helper_neon_shl_s64(cpu_V0, in, tmp64);
5869 tmp = tcg_temp_new_i32();
5870 gen_neon_narrow_op(op == 8, u, size - 1, tmp, cpu_V0);
5871 neon_store_reg(rd, pass, tmp);
5872 } /* for pass */
5873 tcg_temp_free_i64(tmp64);
5874 } else {
5875 if (size == 1) {
5876 imm = (uint16_t)shift;
5877 imm |= imm << 16;
5878 } else {
5879 /* size == 2 */
5880 imm = (uint32_t)shift;
5882 tmp2 = tcg_const_i32(imm);
5883 tmp4 = neon_load_reg(rm + 1, 0);
5884 tmp5 = neon_load_reg(rm + 1, 1);
5885 for (pass = 0; pass < 2; pass++) {
5886 if (pass == 0) {
5887 tmp = neon_load_reg(rm, 0);
5888 } else {
5889 tmp = tmp4;
5891 gen_neon_shift_narrow(size, tmp, tmp2, q,
5892 input_unsigned);
5893 if (pass == 0) {
5894 tmp3 = neon_load_reg(rm, 1);
5895 } else {
5896 tmp3 = tmp5;
5898 gen_neon_shift_narrow(size, tmp3, tmp2, q,
5899 input_unsigned);
5900 tcg_gen_concat_i32_i64(cpu_V0, tmp, tmp3);
5901 tcg_temp_free_i32(tmp);
5902 tcg_temp_free_i32(tmp3);
5903 tmp = tcg_temp_new_i32();
5904 gen_neon_narrow_op(op == 8, u, size - 1, tmp, cpu_V0);
5905 neon_store_reg(rd, pass, tmp);
5906 } /* for pass */
5907 tcg_temp_free_i32(tmp2);
5909 } else if (op == 10) {
5910 /* VSHLL, VMOVL */
5911 if (q || (rd & 1)) {
5912 return 1;
5914 tmp = neon_load_reg(rm, 0);
5915 tmp2 = neon_load_reg(rm, 1);
5916 for (pass = 0; pass < 2; pass++) {
5917 if (pass == 1)
5918 tmp = tmp2;
5920 gen_neon_widen(cpu_V0, tmp, size, u);
5922 if (shift != 0) {
5923 /* The shift is less than the width of the source
5924 type, so we can just shift the whole register. */
5925 tcg_gen_shli_i64(cpu_V0, cpu_V0, shift);
5926 /* Widen the result of shift: we need to clear
5927 * the potential overflow bits resulting from
5928 * left bits of the narrow input appearing as
5929 * right bits of left the neighbour narrow
5930 * input. */
5931 if (size < 2 || !u) {
5932 uint64_t imm64;
5933 if (size == 0) {
5934 imm = (0xffu >> (8 - shift));
5935 imm |= imm << 16;
5936 } else if (size == 1) {
5937 imm = 0xffff >> (16 - shift);
5938 } else {
5939 /* size == 2 */
5940 imm = 0xffffffff >> (32 - shift);
5942 if (size < 2) {
5943 imm64 = imm | (((uint64_t)imm) << 32);
5944 } else {
5945 imm64 = imm;
5947 tcg_gen_andi_i64(cpu_V0, cpu_V0, ~imm64);
5950 neon_store_reg64(cpu_V0, rd + pass);
5952 } else if (op >= 14) {
5953 /* VCVT fixed-point. */
5954 if (!(insn & (1 << 21)) || (q && ((rd | rm) & 1))) {
5955 return 1;
5957 /* We have already masked out the must-be-1 top bit of imm6,
5958 * hence this 32-shift where the ARM ARM has 64-imm6.
5960 shift = 32 - shift;
5961 for (pass = 0; pass < (q ? 4 : 2); pass++) {
5962 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, pass));
5963 if (!(op & 1)) {
5964 if (u)
5965 gen_vfp_ulto(0, shift, 1);
5966 else
5967 gen_vfp_slto(0, shift, 1);
5968 } else {
5969 if (u)
5970 gen_vfp_toul(0, shift, 1);
5971 else
5972 gen_vfp_tosl(0, shift, 1);
5974 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, pass));
5976 } else {
5977 return 1;
5979 } else { /* (insn & 0x00380080) == 0 */
5980 int invert;
5981 if (q && (rd & 1)) {
5982 return 1;
5985 op = (insn >> 8) & 0xf;
5986 /* One register and immediate. */
5987 imm = (u << 7) | ((insn >> 12) & 0x70) | (insn & 0xf);
5988 invert = (insn & (1 << 5)) != 0;
5989 /* Note that op = 2,3,4,5,6,7,10,11,12,13 imm=0 is UNPREDICTABLE.
5990 * We choose to not special-case this and will behave as if a
5991 * valid constant encoding of 0 had been given.
5993 switch (op) {
5994 case 0: case 1:
5995 /* no-op */
5996 break;
5997 case 2: case 3:
5998 imm <<= 8;
5999 break;
6000 case 4: case 5:
6001 imm <<= 16;
6002 break;
6003 case 6: case 7:
6004 imm <<= 24;
6005 break;
6006 case 8: case 9:
6007 imm |= imm << 16;
6008 break;
6009 case 10: case 11:
6010 imm = (imm << 8) | (imm << 24);
6011 break;
6012 case 12:
6013 imm = (imm << 8) | 0xff;
6014 break;
6015 case 13:
6016 imm = (imm << 16) | 0xffff;
6017 break;
6018 case 14:
6019 imm |= (imm << 8) | (imm << 16) | (imm << 24);
6020 if (invert)
6021 imm = ~imm;
6022 break;
6023 case 15:
6024 if (invert) {
6025 return 1;
6027 imm = ((imm & 0x80) << 24) | ((imm & 0x3f) << 19)
6028 | ((imm & 0x40) ? (0x1f << 25) : (1 << 30));
6029 break;
6031 if (invert)
6032 imm = ~imm;
6034 for (pass = 0; pass < (q ? 4 : 2); pass++) {
6035 if (op & 1 && op < 12) {
6036 tmp = neon_load_reg(rd, pass);
6037 if (invert) {
6038 /* The immediate value has already been inverted, so
6039 BIC becomes AND. */
6040 tcg_gen_andi_i32(tmp, tmp, imm);
6041 } else {
6042 tcg_gen_ori_i32(tmp, tmp, imm);
6044 } else {
6045 /* VMOV, VMVN. */
6046 tmp = tcg_temp_new_i32();
6047 if (op == 14 && invert) {
6048 int n;
6049 uint32_t val;
6050 val = 0;
6051 for (n = 0; n < 4; n++) {
6052 if (imm & (1 << (n + (pass & 1) * 4)))
6053 val |= 0xff << (n * 8);
6055 tcg_gen_movi_i32(tmp, val);
6056 } else {
6057 tcg_gen_movi_i32(tmp, imm);
6060 neon_store_reg(rd, pass, tmp);
6063 } else { /* (insn & 0x00800010 == 0x00800000) */
6064 if (size != 3) {
6065 op = (insn >> 8) & 0xf;
6066 if ((insn & (1 << 6)) == 0) {
6067 /* Three registers of different lengths. */
6068 int src1_wide;
6069 int src2_wide;
6070 int prewiden;
6071 /* undefreq: bit 0 : UNDEF if size == 0
6072 * bit 1 : UNDEF if size == 1
6073 * bit 2 : UNDEF if size == 2
6074 * bit 3 : UNDEF if U == 1
6075 * Note that [2:0] set implies 'always UNDEF'
6077 int undefreq;
6078 /* prewiden, src1_wide, src2_wide, undefreq */
6079 static const int neon_3reg_wide[16][4] = {
6080 {1, 0, 0, 0}, /* VADDL */
6081 {1, 1, 0, 0}, /* VADDW */
6082 {1, 0, 0, 0}, /* VSUBL */
6083 {1, 1, 0, 0}, /* VSUBW */
6084 {0, 1, 1, 0}, /* VADDHN */
6085 {0, 0, 0, 0}, /* VABAL */
6086 {0, 1, 1, 0}, /* VSUBHN */
6087 {0, 0, 0, 0}, /* VABDL */
6088 {0, 0, 0, 0}, /* VMLAL */
6089 {0, 0, 0, 9}, /* VQDMLAL */
6090 {0, 0, 0, 0}, /* VMLSL */
6091 {0, 0, 0, 9}, /* VQDMLSL */
6092 {0, 0, 0, 0}, /* Integer VMULL */
6093 {0, 0, 0, 1}, /* VQDMULL */
6094 {0, 0, 0, 0xa}, /* Polynomial VMULL */
6095 {0, 0, 0, 7}, /* Reserved: always UNDEF */
6098 prewiden = neon_3reg_wide[op][0];
6099 src1_wide = neon_3reg_wide[op][1];
6100 src2_wide = neon_3reg_wide[op][2];
6101 undefreq = neon_3reg_wide[op][3];
6103 if ((undefreq & (1 << size)) ||
6104 ((undefreq & 8) && u)) {
6105 return 1;
6107 if ((src1_wide && (rn & 1)) ||
6108 (src2_wide && (rm & 1)) ||
6109 (!src2_wide && (rd & 1))) {
6110 return 1;
6113 /* Handle VMULL.P64 (Polynomial 64x64 to 128 bit multiply)
6114 * outside the loop below as it only performs a single pass.
6116 if (op == 14 && size == 2) {
6117 TCGv_i64 tcg_rn, tcg_rm, tcg_rd;
6119 if (!arm_dc_feature(s, ARM_FEATURE_V8_PMULL)) {
6120 return 1;
6122 tcg_rn = tcg_temp_new_i64();
6123 tcg_rm = tcg_temp_new_i64();
6124 tcg_rd = tcg_temp_new_i64();
6125 neon_load_reg64(tcg_rn, rn);
6126 neon_load_reg64(tcg_rm, rm);
6127 gen_helper_neon_pmull_64_lo(tcg_rd, tcg_rn, tcg_rm);
6128 neon_store_reg64(tcg_rd, rd);
6129 gen_helper_neon_pmull_64_hi(tcg_rd, tcg_rn, tcg_rm);
6130 neon_store_reg64(tcg_rd, rd + 1);
6131 tcg_temp_free_i64(tcg_rn);
6132 tcg_temp_free_i64(tcg_rm);
6133 tcg_temp_free_i64(tcg_rd);
6134 return 0;
6137 /* Avoid overlapping operands. Wide source operands are
6138 always aligned so will never overlap with wide
6139 destinations in problematic ways. */
6140 if (rd == rm && !src2_wide) {
6141 tmp = neon_load_reg(rm, 1);
6142 neon_store_scratch(2, tmp);
6143 } else if (rd == rn && !src1_wide) {
6144 tmp = neon_load_reg(rn, 1);
6145 neon_store_scratch(2, tmp);
6147 TCGV_UNUSED_I32(tmp3);
6148 for (pass = 0; pass < 2; pass++) {
6149 if (src1_wide) {
6150 neon_load_reg64(cpu_V0, rn + pass);
6151 TCGV_UNUSED_I32(tmp);
6152 } else {
6153 if (pass == 1 && rd == rn) {
6154 tmp = neon_load_scratch(2);
6155 } else {
6156 tmp = neon_load_reg(rn, pass);
6158 if (prewiden) {
6159 gen_neon_widen(cpu_V0, tmp, size, u);
6162 if (src2_wide) {
6163 neon_load_reg64(cpu_V1, rm + pass);
6164 TCGV_UNUSED_I32(tmp2);
6165 } else {
6166 if (pass == 1 && rd == rm) {
6167 tmp2 = neon_load_scratch(2);
6168 } else {
6169 tmp2 = neon_load_reg(rm, pass);
6171 if (prewiden) {
6172 gen_neon_widen(cpu_V1, tmp2, size, u);
6175 switch (op) {
6176 case 0: case 1: case 4: /* VADDL, VADDW, VADDHN, VRADDHN */
6177 gen_neon_addl(size);
6178 break;
6179 case 2: case 3: case 6: /* VSUBL, VSUBW, VSUBHN, VRSUBHN */
6180 gen_neon_subl(size);
6181 break;
6182 case 5: case 7: /* VABAL, VABDL */
6183 switch ((size << 1) | u) {
6184 case 0:
6185 gen_helper_neon_abdl_s16(cpu_V0, tmp, tmp2);
6186 break;
6187 case 1:
6188 gen_helper_neon_abdl_u16(cpu_V0, tmp, tmp2);
6189 break;
6190 case 2:
6191 gen_helper_neon_abdl_s32(cpu_V0, tmp, tmp2);
6192 break;
6193 case 3:
6194 gen_helper_neon_abdl_u32(cpu_V0, tmp, tmp2);
6195 break;
6196 case 4:
6197 gen_helper_neon_abdl_s64(cpu_V0, tmp, tmp2);
6198 break;
6199 case 5:
6200 gen_helper_neon_abdl_u64(cpu_V0, tmp, tmp2);
6201 break;
6202 default: abort();
6204 tcg_temp_free_i32(tmp2);
6205 tcg_temp_free_i32(tmp);
6206 break;
6207 case 8: case 9: case 10: case 11: case 12: case 13:
6208 /* VMLAL, VQDMLAL, VMLSL, VQDMLSL, VMULL, VQDMULL */
6209 gen_neon_mull(cpu_V0, tmp, tmp2, size, u);
6210 break;
6211 case 14: /* Polynomial VMULL */
6212 gen_helper_neon_mull_p8(cpu_V0, tmp, tmp2);
6213 tcg_temp_free_i32(tmp2);
6214 tcg_temp_free_i32(tmp);
6215 break;
6216 default: /* 15 is RESERVED: caught earlier */
6217 abort();
6219 if (op == 13) {
6220 /* VQDMULL */
6221 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
6222 neon_store_reg64(cpu_V0, rd + pass);
6223 } else if (op == 5 || (op >= 8 && op <= 11)) {
6224 /* Accumulate. */
6225 neon_load_reg64(cpu_V1, rd + pass);
6226 switch (op) {
6227 case 10: /* VMLSL */
6228 gen_neon_negl(cpu_V0, size);
6229 /* Fall through */
6230 case 5: case 8: /* VABAL, VMLAL */
6231 gen_neon_addl(size);
6232 break;
6233 case 9: case 11: /* VQDMLAL, VQDMLSL */
6234 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
6235 if (op == 11) {
6236 gen_neon_negl(cpu_V0, size);
6238 gen_neon_addl_saturate(cpu_V0, cpu_V1, size);
6239 break;
6240 default:
6241 abort();
6243 neon_store_reg64(cpu_V0, rd + pass);
6244 } else if (op == 4 || op == 6) {
6245 /* Narrowing operation. */
6246 tmp = tcg_temp_new_i32();
6247 if (!u) {
6248 switch (size) {
6249 case 0:
6250 gen_helper_neon_narrow_high_u8(tmp, cpu_V0);
6251 break;
6252 case 1:
6253 gen_helper_neon_narrow_high_u16(tmp, cpu_V0);
6254 break;
6255 case 2:
6256 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
6257 tcg_gen_extrl_i64_i32(tmp, cpu_V0);
6258 break;
6259 default: abort();
6261 } else {
6262 switch (size) {
6263 case 0:
6264 gen_helper_neon_narrow_round_high_u8(tmp, cpu_V0);
6265 break;
6266 case 1:
6267 gen_helper_neon_narrow_round_high_u16(tmp, cpu_V0);
6268 break;
6269 case 2:
6270 tcg_gen_addi_i64(cpu_V0, cpu_V0, 1u << 31);
6271 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
6272 tcg_gen_extrl_i64_i32(tmp, cpu_V0);
6273 break;
6274 default: abort();
6277 if (pass == 0) {
6278 tmp3 = tmp;
6279 } else {
6280 neon_store_reg(rd, 0, tmp3);
6281 neon_store_reg(rd, 1, tmp);
6283 } else {
6284 /* Write back the result. */
6285 neon_store_reg64(cpu_V0, rd + pass);
6288 } else {
6289 /* Two registers and a scalar. NB that for ops of this form
6290 * the ARM ARM labels bit 24 as Q, but it is in our variable
6291 * 'u', not 'q'.
6293 if (size == 0) {
6294 return 1;
6296 switch (op) {
6297 case 1: /* Float VMLA scalar */
6298 case 5: /* Floating point VMLS scalar */
6299 case 9: /* Floating point VMUL scalar */
6300 if (size == 1) {
6301 return 1;
6303 /* fall through */
6304 case 0: /* Integer VMLA scalar */
6305 case 4: /* Integer VMLS scalar */
6306 case 8: /* Integer VMUL scalar */
6307 case 12: /* VQDMULH scalar */
6308 case 13: /* VQRDMULH scalar */
6309 if (u && ((rd | rn) & 1)) {
6310 return 1;
6312 tmp = neon_get_scalar(size, rm);
6313 neon_store_scratch(0, tmp);
6314 for (pass = 0; pass < (u ? 4 : 2); pass++) {
6315 tmp = neon_load_scratch(0);
6316 tmp2 = neon_load_reg(rn, pass);
6317 if (op == 12) {
6318 if (size == 1) {
6319 gen_helper_neon_qdmulh_s16(tmp, cpu_env, tmp, tmp2);
6320 } else {
6321 gen_helper_neon_qdmulh_s32(tmp, cpu_env, tmp, tmp2);
6323 } else if (op == 13) {
6324 if (size == 1) {
6325 gen_helper_neon_qrdmulh_s16(tmp, cpu_env, tmp, tmp2);
6326 } else {
6327 gen_helper_neon_qrdmulh_s32(tmp, cpu_env, tmp, tmp2);
6329 } else if (op & 1) {
6330 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6331 gen_helper_vfp_muls(tmp, tmp, tmp2, fpstatus);
6332 tcg_temp_free_ptr(fpstatus);
6333 } else {
6334 switch (size) {
6335 case 0: gen_helper_neon_mul_u8(tmp, tmp, tmp2); break;
6336 case 1: gen_helper_neon_mul_u16(tmp, tmp, tmp2); break;
6337 case 2: tcg_gen_mul_i32(tmp, tmp, tmp2); break;
6338 default: abort();
6341 tcg_temp_free_i32(tmp2);
6342 if (op < 8) {
6343 /* Accumulate. */
6344 tmp2 = neon_load_reg(rd, pass);
6345 switch (op) {
6346 case 0:
6347 gen_neon_add(size, tmp, tmp2);
6348 break;
6349 case 1:
6351 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6352 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
6353 tcg_temp_free_ptr(fpstatus);
6354 break;
6356 case 4:
6357 gen_neon_rsb(size, tmp, tmp2);
6358 break;
6359 case 5:
6361 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6362 gen_helper_vfp_subs(tmp, tmp2, tmp, fpstatus);
6363 tcg_temp_free_ptr(fpstatus);
6364 break;
6366 default:
6367 abort();
6369 tcg_temp_free_i32(tmp2);
6371 neon_store_reg(rd, pass, tmp);
6373 break;
6374 case 3: /* VQDMLAL scalar */
6375 case 7: /* VQDMLSL scalar */
6376 case 11: /* VQDMULL scalar */
6377 if (u == 1) {
6378 return 1;
6380 /* fall through */
6381 case 2: /* VMLAL sclar */
6382 case 6: /* VMLSL scalar */
6383 case 10: /* VMULL scalar */
6384 if (rd & 1) {
6385 return 1;
6387 tmp2 = neon_get_scalar(size, rm);
6388 /* We need a copy of tmp2 because gen_neon_mull
6389 * deletes it during pass 0. */
6390 tmp4 = tcg_temp_new_i32();
6391 tcg_gen_mov_i32(tmp4, tmp2);
6392 tmp3 = neon_load_reg(rn, 1);
6394 for (pass = 0; pass < 2; pass++) {
6395 if (pass == 0) {
6396 tmp = neon_load_reg(rn, 0);
6397 } else {
6398 tmp = tmp3;
6399 tmp2 = tmp4;
6401 gen_neon_mull(cpu_V0, tmp, tmp2, size, u);
6402 if (op != 11) {
6403 neon_load_reg64(cpu_V1, rd + pass);
6405 switch (op) {
6406 case 6:
6407 gen_neon_negl(cpu_V0, size);
6408 /* Fall through */
6409 case 2:
6410 gen_neon_addl(size);
6411 break;
6412 case 3: case 7:
6413 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
6414 if (op == 7) {
6415 gen_neon_negl(cpu_V0, size);
6417 gen_neon_addl_saturate(cpu_V0, cpu_V1, size);
6418 break;
6419 case 10:
6420 /* no-op */
6421 break;
6422 case 11:
6423 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
6424 break;
6425 default:
6426 abort();
6428 neon_store_reg64(cpu_V0, rd + pass);
6432 break;
6433 default: /* 14 and 15 are RESERVED */
6434 return 1;
6437 } else { /* size == 3 */
6438 if (!u) {
6439 /* Extract. */
6440 imm = (insn >> 8) & 0xf;
6442 if (imm > 7 && !q)
6443 return 1;
6445 if (q && ((rd | rn | rm) & 1)) {
6446 return 1;
6449 if (imm == 0) {
6450 neon_load_reg64(cpu_V0, rn);
6451 if (q) {
6452 neon_load_reg64(cpu_V1, rn + 1);
6454 } else if (imm == 8) {
6455 neon_load_reg64(cpu_V0, rn + 1);
6456 if (q) {
6457 neon_load_reg64(cpu_V1, rm);
6459 } else if (q) {
6460 tmp64 = tcg_temp_new_i64();
6461 if (imm < 8) {
6462 neon_load_reg64(cpu_V0, rn);
6463 neon_load_reg64(tmp64, rn + 1);
6464 } else {
6465 neon_load_reg64(cpu_V0, rn + 1);
6466 neon_load_reg64(tmp64, rm);
6468 tcg_gen_shri_i64(cpu_V0, cpu_V0, (imm & 7) * 8);
6469 tcg_gen_shli_i64(cpu_V1, tmp64, 64 - ((imm & 7) * 8));
6470 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
6471 if (imm < 8) {
6472 neon_load_reg64(cpu_V1, rm);
6473 } else {
6474 neon_load_reg64(cpu_V1, rm + 1);
6475 imm -= 8;
6477 tcg_gen_shli_i64(cpu_V1, cpu_V1, 64 - (imm * 8));
6478 tcg_gen_shri_i64(tmp64, tmp64, imm * 8);
6479 tcg_gen_or_i64(cpu_V1, cpu_V1, tmp64);
6480 tcg_temp_free_i64(tmp64);
6481 } else {
6482 /* BUGFIX */
6483 neon_load_reg64(cpu_V0, rn);
6484 tcg_gen_shri_i64(cpu_V0, cpu_V0, imm * 8);
6485 neon_load_reg64(cpu_V1, rm);
6486 tcg_gen_shli_i64(cpu_V1, cpu_V1, 64 - (imm * 8));
6487 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
6489 neon_store_reg64(cpu_V0, rd);
6490 if (q) {
6491 neon_store_reg64(cpu_V1, rd + 1);
6493 } else if ((insn & (1 << 11)) == 0) {
6494 /* Two register misc. */
6495 op = ((insn >> 12) & 0x30) | ((insn >> 7) & 0xf);
6496 size = (insn >> 18) & 3;
6497 /* UNDEF for unknown op values and bad op-size combinations */
6498 if ((neon_2rm_sizes[op] & (1 << size)) == 0) {
6499 return 1;
6501 if ((op != NEON_2RM_VMOVN && op != NEON_2RM_VQMOVN) &&
6502 q && ((rm | rd) & 1)) {
6503 return 1;
6505 switch (op) {
6506 case NEON_2RM_VREV64:
6507 for (pass = 0; pass < (q ? 2 : 1); pass++) {
6508 tmp = neon_load_reg(rm, pass * 2);
6509 tmp2 = neon_load_reg(rm, pass * 2 + 1);
6510 switch (size) {
6511 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
6512 case 1: gen_swap_half(tmp); break;
6513 case 2: /* no-op */ break;
6514 default: abort();
6516 neon_store_reg(rd, pass * 2 + 1, tmp);
6517 if (size == 2) {
6518 neon_store_reg(rd, pass * 2, tmp2);
6519 } else {
6520 switch (size) {
6521 case 0: tcg_gen_bswap32_i32(tmp2, tmp2); break;
6522 case 1: gen_swap_half(tmp2); break;
6523 default: abort();
6525 neon_store_reg(rd, pass * 2, tmp2);
6528 break;
6529 case NEON_2RM_VPADDL: case NEON_2RM_VPADDL_U:
6530 case NEON_2RM_VPADAL: case NEON_2RM_VPADAL_U:
6531 for (pass = 0; pass < q + 1; pass++) {
6532 tmp = neon_load_reg(rm, pass * 2);
6533 gen_neon_widen(cpu_V0, tmp, size, op & 1);
6534 tmp = neon_load_reg(rm, pass * 2 + 1);
6535 gen_neon_widen(cpu_V1, tmp, size, op & 1);
6536 switch (size) {
6537 case 0: gen_helper_neon_paddl_u16(CPU_V001); break;
6538 case 1: gen_helper_neon_paddl_u32(CPU_V001); break;
6539 case 2: tcg_gen_add_i64(CPU_V001); break;
6540 default: abort();
6542 if (op >= NEON_2RM_VPADAL) {
6543 /* Accumulate. */
6544 neon_load_reg64(cpu_V1, rd + pass);
6545 gen_neon_addl(size);
6547 neon_store_reg64(cpu_V0, rd + pass);
6549 break;
6550 case NEON_2RM_VTRN:
6551 if (size == 2) {
6552 int n;
6553 for (n = 0; n < (q ? 4 : 2); n += 2) {
6554 tmp = neon_load_reg(rm, n);
6555 tmp2 = neon_load_reg(rd, n + 1);
6556 neon_store_reg(rm, n, tmp2);
6557 neon_store_reg(rd, n + 1, tmp);
6559 } else {
6560 goto elementwise;
6562 break;
6563 case NEON_2RM_VUZP:
6564 if (gen_neon_unzip(rd, rm, size, q)) {
6565 return 1;
6567 break;
6568 case NEON_2RM_VZIP:
6569 if (gen_neon_zip(rd, rm, size, q)) {
6570 return 1;
6572 break;
6573 case NEON_2RM_VMOVN: case NEON_2RM_VQMOVN:
6574 /* also VQMOVUN; op field and mnemonics don't line up */
6575 if (rm & 1) {
6576 return 1;
6578 TCGV_UNUSED_I32(tmp2);
6579 for (pass = 0; pass < 2; pass++) {
6580 neon_load_reg64(cpu_V0, rm + pass);
6581 tmp = tcg_temp_new_i32();
6582 gen_neon_narrow_op(op == NEON_2RM_VMOVN, q, size,
6583 tmp, cpu_V0);
6584 if (pass == 0) {
6585 tmp2 = tmp;
6586 } else {
6587 neon_store_reg(rd, 0, tmp2);
6588 neon_store_reg(rd, 1, tmp);
6591 break;
6592 case NEON_2RM_VSHLL:
6593 if (q || (rd & 1)) {
6594 return 1;
6596 tmp = neon_load_reg(rm, 0);
6597 tmp2 = neon_load_reg(rm, 1);
6598 for (pass = 0; pass < 2; pass++) {
6599 if (pass == 1)
6600 tmp = tmp2;
6601 gen_neon_widen(cpu_V0, tmp, size, 1);
6602 tcg_gen_shli_i64(cpu_V0, cpu_V0, 8 << size);
6603 neon_store_reg64(cpu_V0, rd + pass);
6605 break;
6606 case NEON_2RM_VCVT_F16_F32:
6607 if (!arm_dc_feature(s, ARM_FEATURE_VFP_FP16) ||
6608 q || (rm & 1)) {
6609 return 1;
6611 tmp = tcg_temp_new_i32();
6612 tmp2 = tcg_temp_new_i32();
6613 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 0));
6614 gen_helper_neon_fcvt_f32_to_f16(tmp, cpu_F0s, cpu_env);
6615 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 1));
6616 gen_helper_neon_fcvt_f32_to_f16(tmp2, cpu_F0s, cpu_env);
6617 tcg_gen_shli_i32(tmp2, tmp2, 16);
6618 tcg_gen_or_i32(tmp2, tmp2, tmp);
6619 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 2));
6620 gen_helper_neon_fcvt_f32_to_f16(tmp, cpu_F0s, cpu_env);
6621 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 3));
6622 neon_store_reg(rd, 0, tmp2);
6623 tmp2 = tcg_temp_new_i32();
6624 gen_helper_neon_fcvt_f32_to_f16(tmp2, cpu_F0s, cpu_env);
6625 tcg_gen_shli_i32(tmp2, tmp2, 16);
6626 tcg_gen_or_i32(tmp2, tmp2, tmp);
6627 neon_store_reg(rd, 1, tmp2);
6628 tcg_temp_free_i32(tmp);
6629 break;
6630 case NEON_2RM_VCVT_F32_F16:
6631 if (!arm_dc_feature(s, ARM_FEATURE_VFP_FP16) ||
6632 q || (rd & 1)) {
6633 return 1;
6635 tmp3 = tcg_temp_new_i32();
6636 tmp = neon_load_reg(rm, 0);
6637 tmp2 = neon_load_reg(rm, 1);
6638 tcg_gen_ext16u_i32(tmp3, tmp);
6639 gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
6640 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 0));
6641 tcg_gen_shri_i32(tmp3, tmp, 16);
6642 gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
6643 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 1));
6644 tcg_temp_free_i32(tmp);
6645 tcg_gen_ext16u_i32(tmp3, tmp2);
6646 gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
6647 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 2));
6648 tcg_gen_shri_i32(tmp3, tmp2, 16);
6649 gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
6650 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 3));
6651 tcg_temp_free_i32(tmp2);
6652 tcg_temp_free_i32(tmp3);
6653 break;
6654 case NEON_2RM_AESE: case NEON_2RM_AESMC:
6655 if (!arm_dc_feature(s, ARM_FEATURE_V8_AES)
6656 || ((rm | rd) & 1)) {
6657 return 1;
6659 tmp = tcg_const_i32(rd);
6660 tmp2 = tcg_const_i32(rm);
6662 /* Bit 6 is the lowest opcode bit; it distinguishes between
6663 * encryption (AESE/AESMC) and decryption (AESD/AESIMC)
6665 tmp3 = tcg_const_i32(extract32(insn, 6, 1));
6667 if (op == NEON_2RM_AESE) {
6668 gen_helper_crypto_aese(cpu_env, tmp, tmp2, tmp3);
6669 } else {
6670 gen_helper_crypto_aesmc(cpu_env, tmp, tmp2, tmp3);
6672 tcg_temp_free_i32(tmp);
6673 tcg_temp_free_i32(tmp2);
6674 tcg_temp_free_i32(tmp3);
6675 break;
6676 case NEON_2RM_SHA1H:
6677 if (!arm_dc_feature(s, ARM_FEATURE_V8_SHA1)
6678 || ((rm | rd) & 1)) {
6679 return 1;
6681 tmp = tcg_const_i32(rd);
6682 tmp2 = tcg_const_i32(rm);
6684 gen_helper_crypto_sha1h(cpu_env, tmp, tmp2);
6686 tcg_temp_free_i32(tmp);
6687 tcg_temp_free_i32(tmp2);
6688 break;
6689 case NEON_2RM_SHA1SU1:
6690 if ((rm | rd) & 1) {
6691 return 1;
6693 /* bit 6 (q): set -> SHA256SU0, cleared -> SHA1SU1 */
6694 if (q) {
6695 if (!arm_dc_feature(s, ARM_FEATURE_V8_SHA256)) {
6696 return 1;
6698 } else if (!arm_dc_feature(s, ARM_FEATURE_V8_SHA1)) {
6699 return 1;
6701 tmp = tcg_const_i32(rd);
6702 tmp2 = tcg_const_i32(rm);
6703 if (q) {
6704 gen_helper_crypto_sha256su0(cpu_env, tmp, tmp2);
6705 } else {
6706 gen_helper_crypto_sha1su1(cpu_env, tmp, tmp2);
6708 tcg_temp_free_i32(tmp);
6709 tcg_temp_free_i32(tmp2);
6710 break;
6711 default:
6712 elementwise:
6713 for (pass = 0; pass < (q ? 4 : 2); pass++) {
6714 if (neon_2rm_is_float_op(op)) {
6715 tcg_gen_ld_f32(cpu_F0s, cpu_env,
6716 neon_reg_offset(rm, pass));
6717 TCGV_UNUSED_I32(tmp);
6718 } else {
6719 tmp = neon_load_reg(rm, pass);
6721 switch (op) {
6722 case NEON_2RM_VREV32:
6723 switch (size) {
6724 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
6725 case 1: gen_swap_half(tmp); break;
6726 default: abort();
6728 break;
6729 case NEON_2RM_VREV16:
6730 gen_rev16(tmp);
6731 break;
6732 case NEON_2RM_VCLS:
6733 switch (size) {
6734 case 0: gen_helper_neon_cls_s8(tmp, tmp); break;
6735 case 1: gen_helper_neon_cls_s16(tmp, tmp); break;
6736 case 2: gen_helper_neon_cls_s32(tmp, tmp); break;
6737 default: abort();
6739 break;
6740 case NEON_2RM_VCLZ:
6741 switch (size) {
6742 case 0: gen_helper_neon_clz_u8(tmp, tmp); break;
6743 case 1: gen_helper_neon_clz_u16(tmp, tmp); break;
6744 case 2: gen_helper_clz(tmp, tmp); break;
6745 default: abort();
6747 break;
6748 case NEON_2RM_VCNT:
6749 gen_helper_neon_cnt_u8(tmp, tmp);
6750 break;
6751 case NEON_2RM_VMVN:
6752 tcg_gen_not_i32(tmp, tmp);
6753 break;
6754 case NEON_2RM_VQABS:
6755 switch (size) {
6756 case 0:
6757 gen_helper_neon_qabs_s8(tmp, cpu_env, tmp);
6758 break;
6759 case 1:
6760 gen_helper_neon_qabs_s16(tmp, cpu_env, tmp);
6761 break;
6762 case 2:
6763 gen_helper_neon_qabs_s32(tmp, cpu_env, tmp);
6764 break;
6765 default: abort();
6767 break;
6768 case NEON_2RM_VQNEG:
6769 switch (size) {
6770 case 0:
6771 gen_helper_neon_qneg_s8(tmp, cpu_env, tmp);
6772 break;
6773 case 1:
6774 gen_helper_neon_qneg_s16(tmp, cpu_env, tmp);
6775 break;
6776 case 2:
6777 gen_helper_neon_qneg_s32(tmp, cpu_env, tmp);
6778 break;
6779 default: abort();
6781 break;
6782 case NEON_2RM_VCGT0: case NEON_2RM_VCLE0:
6783 tmp2 = tcg_const_i32(0);
6784 switch(size) {
6785 case 0: gen_helper_neon_cgt_s8(tmp, tmp, tmp2); break;
6786 case 1: gen_helper_neon_cgt_s16(tmp, tmp, tmp2); break;
6787 case 2: gen_helper_neon_cgt_s32(tmp, tmp, tmp2); break;
6788 default: abort();
6790 tcg_temp_free_i32(tmp2);
6791 if (op == NEON_2RM_VCLE0) {
6792 tcg_gen_not_i32(tmp, tmp);
6794 break;
6795 case NEON_2RM_VCGE0: case NEON_2RM_VCLT0:
6796 tmp2 = tcg_const_i32(0);
6797 switch(size) {
6798 case 0: gen_helper_neon_cge_s8(tmp, tmp, tmp2); break;
6799 case 1: gen_helper_neon_cge_s16(tmp, tmp, tmp2); break;
6800 case 2: gen_helper_neon_cge_s32(tmp, tmp, tmp2); break;
6801 default: abort();
6803 tcg_temp_free_i32(tmp2);
6804 if (op == NEON_2RM_VCLT0) {
6805 tcg_gen_not_i32(tmp, tmp);
6807 break;
6808 case NEON_2RM_VCEQ0:
6809 tmp2 = tcg_const_i32(0);
6810 switch(size) {
6811 case 0: gen_helper_neon_ceq_u8(tmp, tmp, tmp2); break;
6812 case 1: gen_helper_neon_ceq_u16(tmp, tmp, tmp2); break;
6813 case 2: gen_helper_neon_ceq_u32(tmp, tmp, tmp2); break;
6814 default: abort();
6816 tcg_temp_free_i32(tmp2);
6817 break;
6818 case NEON_2RM_VABS:
6819 switch(size) {
6820 case 0: gen_helper_neon_abs_s8(tmp, tmp); break;
6821 case 1: gen_helper_neon_abs_s16(tmp, tmp); break;
6822 case 2: tcg_gen_abs_i32(tmp, tmp); break;
6823 default: abort();
6825 break;
6826 case NEON_2RM_VNEG:
6827 tmp2 = tcg_const_i32(0);
6828 gen_neon_rsb(size, tmp, tmp2);
6829 tcg_temp_free_i32(tmp2);
6830 break;
6831 case NEON_2RM_VCGT0_F:
6833 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6834 tmp2 = tcg_const_i32(0);
6835 gen_helper_neon_cgt_f32(tmp, tmp, tmp2, fpstatus);
6836 tcg_temp_free_i32(tmp2);
6837 tcg_temp_free_ptr(fpstatus);
6838 break;
6840 case NEON_2RM_VCGE0_F:
6842 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6843 tmp2 = tcg_const_i32(0);
6844 gen_helper_neon_cge_f32(tmp, tmp, tmp2, fpstatus);
6845 tcg_temp_free_i32(tmp2);
6846 tcg_temp_free_ptr(fpstatus);
6847 break;
6849 case NEON_2RM_VCEQ0_F:
6851 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6852 tmp2 = tcg_const_i32(0);
6853 gen_helper_neon_ceq_f32(tmp, tmp, tmp2, fpstatus);
6854 tcg_temp_free_i32(tmp2);
6855 tcg_temp_free_ptr(fpstatus);
6856 break;
6858 case NEON_2RM_VCLE0_F:
6860 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6861 tmp2 = tcg_const_i32(0);
6862 gen_helper_neon_cge_f32(tmp, tmp2, tmp, fpstatus);
6863 tcg_temp_free_i32(tmp2);
6864 tcg_temp_free_ptr(fpstatus);
6865 break;
6867 case NEON_2RM_VCLT0_F:
6869 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6870 tmp2 = tcg_const_i32(0);
6871 gen_helper_neon_cgt_f32(tmp, tmp2, tmp, fpstatus);
6872 tcg_temp_free_i32(tmp2);
6873 tcg_temp_free_ptr(fpstatus);
6874 break;
6876 case NEON_2RM_VABS_F:
6877 gen_vfp_abs(0);
6878 break;
6879 case NEON_2RM_VNEG_F:
6880 gen_vfp_neg(0);
6881 break;
6882 case NEON_2RM_VSWP:
6883 tmp2 = neon_load_reg(rd, pass);
6884 neon_store_reg(rm, pass, tmp2);
6885 break;
6886 case NEON_2RM_VTRN:
6887 tmp2 = neon_load_reg(rd, pass);
6888 switch (size) {
6889 case 0: gen_neon_trn_u8(tmp, tmp2); break;
6890 case 1: gen_neon_trn_u16(tmp, tmp2); break;
6891 default: abort();
6893 neon_store_reg(rm, pass, tmp2);
6894 break;
6895 case NEON_2RM_VRINTN:
6896 case NEON_2RM_VRINTA:
6897 case NEON_2RM_VRINTM:
6898 case NEON_2RM_VRINTP:
6899 case NEON_2RM_VRINTZ:
6901 TCGv_i32 tcg_rmode;
6902 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6903 int rmode;
6905 if (op == NEON_2RM_VRINTZ) {
6906 rmode = FPROUNDING_ZERO;
6907 } else {
6908 rmode = fp_decode_rm[((op & 0x6) >> 1) ^ 1];
6911 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rmode));
6912 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
6913 cpu_env);
6914 gen_helper_rints(cpu_F0s, cpu_F0s, fpstatus);
6915 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
6916 cpu_env);
6917 tcg_temp_free_ptr(fpstatus);
6918 tcg_temp_free_i32(tcg_rmode);
6919 break;
6921 case NEON_2RM_VRINTX:
6923 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6924 gen_helper_rints_exact(cpu_F0s, cpu_F0s, fpstatus);
6925 tcg_temp_free_ptr(fpstatus);
6926 break;
6928 case NEON_2RM_VCVTAU:
6929 case NEON_2RM_VCVTAS:
6930 case NEON_2RM_VCVTNU:
6931 case NEON_2RM_VCVTNS:
6932 case NEON_2RM_VCVTPU:
6933 case NEON_2RM_VCVTPS:
6934 case NEON_2RM_VCVTMU:
6935 case NEON_2RM_VCVTMS:
6937 bool is_signed = !extract32(insn, 7, 1);
6938 TCGv_ptr fpst = get_fpstatus_ptr(1);
6939 TCGv_i32 tcg_rmode, tcg_shift;
6940 int rmode = fp_decode_rm[extract32(insn, 8, 2)];
6942 tcg_shift = tcg_const_i32(0);
6943 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rmode));
6944 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
6945 cpu_env);
6947 if (is_signed) {
6948 gen_helper_vfp_tosls(cpu_F0s, cpu_F0s,
6949 tcg_shift, fpst);
6950 } else {
6951 gen_helper_vfp_touls(cpu_F0s, cpu_F0s,
6952 tcg_shift, fpst);
6955 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
6956 cpu_env);
6957 tcg_temp_free_i32(tcg_rmode);
6958 tcg_temp_free_i32(tcg_shift);
6959 tcg_temp_free_ptr(fpst);
6960 break;
6962 case NEON_2RM_VRECPE:
6964 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6965 gen_helper_recpe_u32(tmp, tmp, fpstatus);
6966 tcg_temp_free_ptr(fpstatus);
6967 break;
6969 case NEON_2RM_VRSQRTE:
6971 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6972 gen_helper_rsqrte_u32(tmp, tmp, fpstatus);
6973 tcg_temp_free_ptr(fpstatus);
6974 break;
6976 case NEON_2RM_VRECPE_F:
6978 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6979 gen_helper_recpe_f32(cpu_F0s, cpu_F0s, fpstatus);
6980 tcg_temp_free_ptr(fpstatus);
6981 break;
6983 case NEON_2RM_VRSQRTE_F:
6985 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6986 gen_helper_rsqrte_f32(cpu_F0s, cpu_F0s, fpstatus);
6987 tcg_temp_free_ptr(fpstatus);
6988 break;
6990 case NEON_2RM_VCVT_FS: /* VCVT.F32.S32 */
6991 gen_vfp_sito(0, 1);
6992 break;
6993 case NEON_2RM_VCVT_FU: /* VCVT.F32.U32 */
6994 gen_vfp_uito(0, 1);
6995 break;
6996 case NEON_2RM_VCVT_SF: /* VCVT.S32.F32 */
6997 gen_vfp_tosiz(0, 1);
6998 break;
6999 case NEON_2RM_VCVT_UF: /* VCVT.U32.F32 */
7000 gen_vfp_touiz(0, 1);
7001 break;
7002 default:
7003 /* Reserved op values were caught by the
7004 * neon_2rm_sizes[] check earlier.
7006 abort();
7008 if (neon_2rm_is_float_op(op)) {
7009 tcg_gen_st_f32(cpu_F0s, cpu_env,
7010 neon_reg_offset(rd, pass));
7011 } else {
7012 neon_store_reg(rd, pass, tmp);
7015 break;
7017 } else if ((insn & (1 << 10)) == 0) {
7018 /* VTBL, VTBX. */
7019 int n = ((insn >> 8) & 3) + 1;
7020 if ((rn + n) > 32) {
7021 /* This is UNPREDICTABLE; we choose to UNDEF to avoid the
7022 * helper function running off the end of the register file.
7024 return 1;
7026 n <<= 3;
7027 if (insn & (1 << 6)) {
7028 tmp = neon_load_reg(rd, 0);
7029 } else {
7030 tmp = tcg_temp_new_i32();
7031 tcg_gen_movi_i32(tmp, 0);
7033 tmp2 = neon_load_reg(rm, 0);
7034 tmp4 = tcg_const_i32(rn);
7035 tmp5 = tcg_const_i32(n);
7036 gen_helper_neon_tbl(tmp2, cpu_env, tmp2, tmp, tmp4, tmp5);
7037 tcg_temp_free_i32(tmp);
7038 if (insn & (1 << 6)) {
7039 tmp = neon_load_reg(rd, 1);
7040 } else {
7041 tmp = tcg_temp_new_i32();
7042 tcg_gen_movi_i32(tmp, 0);
7044 tmp3 = neon_load_reg(rm, 1);
7045 gen_helper_neon_tbl(tmp3, cpu_env, tmp3, tmp, tmp4, tmp5);
7046 tcg_temp_free_i32(tmp5);
7047 tcg_temp_free_i32(tmp4);
7048 neon_store_reg(rd, 0, tmp2);
7049 neon_store_reg(rd, 1, tmp3);
7050 tcg_temp_free_i32(tmp);
7051 } else if ((insn & 0x380) == 0) {
7052 /* VDUP */
7053 if ((insn & (7 << 16)) == 0 || (q && (rd & 1))) {
7054 return 1;
7056 if (insn & (1 << 19)) {
7057 tmp = neon_load_reg(rm, 1);
7058 } else {
7059 tmp = neon_load_reg(rm, 0);
7061 if (insn & (1 << 16)) {
7062 gen_neon_dup_u8(tmp, ((insn >> 17) & 3) * 8);
7063 } else if (insn & (1 << 17)) {
7064 if ((insn >> 18) & 1)
7065 gen_neon_dup_high16(tmp);
7066 else
7067 gen_neon_dup_low16(tmp);
7069 for (pass = 0; pass < (q ? 4 : 2); pass++) {
7070 tmp2 = tcg_temp_new_i32();
7071 tcg_gen_mov_i32(tmp2, tmp);
7072 neon_store_reg(rd, pass, tmp2);
7074 tcg_temp_free_i32(tmp);
7075 } else {
7076 return 1;
7080 return 0;
7083 static int disas_coproc_insn(DisasContext *s, uint32_t insn)
7085 int cpnum, is64, crn, crm, opc1, opc2, isread, rt, rt2;
7086 const ARMCPRegInfo *ri;
7088 cpnum = (insn >> 8) & 0xf;
7090 /* First check for coprocessor space used for XScale/iwMMXt insns */
7091 if (arm_dc_feature(s, ARM_FEATURE_XSCALE) && (cpnum < 2)) {
7092 if (extract32(s->c15_cpar, cpnum, 1) == 0) {
7093 return 1;
7095 if (arm_dc_feature(s, ARM_FEATURE_IWMMXT)) {
7096 return disas_iwmmxt_insn(s, insn);
7097 } else if (arm_dc_feature(s, ARM_FEATURE_XSCALE)) {
7098 return disas_dsp_insn(s, insn);
7100 return 1;
7103 /* Otherwise treat as a generic register access */
7104 is64 = (insn & (1 << 25)) == 0;
7105 if (!is64 && ((insn & (1 << 4)) == 0)) {
7106 /* cdp */
7107 return 1;
7110 crm = insn & 0xf;
7111 if (is64) {
7112 crn = 0;
7113 opc1 = (insn >> 4) & 0xf;
7114 opc2 = 0;
7115 rt2 = (insn >> 16) & 0xf;
7116 } else {
7117 crn = (insn >> 16) & 0xf;
7118 opc1 = (insn >> 21) & 7;
7119 opc2 = (insn >> 5) & 7;
7120 rt2 = 0;
7122 isread = (insn >> 20) & 1;
7123 rt = (insn >> 12) & 0xf;
7125 ri = get_arm_cp_reginfo(s->cp_regs,
7126 ENCODE_CP_REG(cpnum, is64, s->ns, crn, crm, opc1, opc2));
7127 if (ri) {
7128 /* Check access permissions */
7129 if (!cp_access_ok(s->current_el, ri, isread)) {
7130 return 1;
7133 if (ri->accessfn ||
7134 (arm_dc_feature(s, ARM_FEATURE_XSCALE) && cpnum < 14)) {
7135 /* Emit code to perform further access permissions checks at
7136 * runtime; this may result in an exception.
7137 * Note that on XScale all cp0..c13 registers do an access check
7138 * call in order to handle c15_cpar.
7140 TCGv_ptr tmpptr;
7141 TCGv_i32 tcg_syn;
7142 uint32_t syndrome;
7144 /* Note that since we are an implementation which takes an
7145 * exception on a trapped conditional instruction only if the
7146 * instruction passes its condition code check, we can take
7147 * advantage of the clause in the ARM ARM that allows us to set
7148 * the COND field in the instruction to 0xE in all cases.
7149 * We could fish the actual condition out of the insn (ARM)
7150 * or the condexec bits (Thumb) but it isn't necessary.
7152 switch (cpnum) {
7153 case 14:
7154 if (is64) {
7155 syndrome = syn_cp14_rrt_trap(1, 0xe, opc1, crm, rt, rt2,
7156 isread, s->thumb);
7157 } else {
7158 syndrome = syn_cp14_rt_trap(1, 0xe, opc1, opc2, crn, crm,
7159 rt, isread, s->thumb);
7161 break;
7162 case 15:
7163 if (is64) {
7164 syndrome = syn_cp15_rrt_trap(1, 0xe, opc1, crm, rt, rt2,
7165 isread, s->thumb);
7166 } else {
7167 syndrome = syn_cp15_rt_trap(1, 0xe, opc1, opc2, crn, crm,
7168 rt, isread, s->thumb);
7170 break;
7171 default:
7172 /* ARMv8 defines that only coprocessors 14 and 15 exist,
7173 * so this can only happen if this is an ARMv7 or earlier CPU,
7174 * in which case the syndrome information won't actually be
7175 * guest visible.
7177 assert(!arm_dc_feature(s, ARM_FEATURE_V8));
7178 syndrome = syn_uncategorized();
7179 break;
7182 gen_set_pc_im(s, s->pc - 4);
7183 tmpptr = tcg_const_ptr(ri);
7184 tcg_syn = tcg_const_i32(syndrome);
7185 gen_helper_access_check_cp_reg(cpu_env, tmpptr, tcg_syn);
7186 tcg_temp_free_ptr(tmpptr);
7187 tcg_temp_free_i32(tcg_syn);
7190 /* Handle special cases first */
7191 switch (ri->type & ~(ARM_CP_FLAG_MASK & ~ARM_CP_SPECIAL)) {
7192 case ARM_CP_NOP:
7193 return 0;
7194 case ARM_CP_WFI:
7195 if (isread) {
7196 return 1;
7198 gen_set_pc_im(s, s->pc);
7199 s->is_jmp = DISAS_WFI;
7200 return 0;
7201 default:
7202 break;
7205 if ((s->tb->cflags & CF_USE_ICOUNT) && (ri->type & ARM_CP_IO)) {
7206 gen_io_start();
7209 if (isread) {
7210 /* Read */
7211 if (is64) {
7212 TCGv_i64 tmp64;
7213 TCGv_i32 tmp;
7214 if (ri->type & ARM_CP_CONST) {
7215 tmp64 = tcg_const_i64(ri->resetvalue);
7216 } else if (ri->readfn) {
7217 TCGv_ptr tmpptr;
7218 tmp64 = tcg_temp_new_i64();
7219 tmpptr = tcg_const_ptr(ri);
7220 gen_helper_get_cp_reg64(tmp64, cpu_env, tmpptr);
7221 tcg_temp_free_ptr(tmpptr);
7222 } else {
7223 tmp64 = tcg_temp_new_i64();
7224 tcg_gen_ld_i64(tmp64, cpu_env, ri->fieldoffset);
7226 tmp = tcg_temp_new_i32();
7227 tcg_gen_extrl_i64_i32(tmp, tmp64);
7228 store_reg(s, rt, tmp);
7229 tcg_gen_shri_i64(tmp64, tmp64, 32);
7230 tmp = tcg_temp_new_i32();
7231 tcg_gen_extrl_i64_i32(tmp, tmp64);
7232 tcg_temp_free_i64(tmp64);
7233 store_reg(s, rt2, tmp);
7234 } else {
7235 TCGv_i32 tmp;
7236 if (ri->type & ARM_CP_CONST) {
7237 tmp = tcg_const_i32(ri->resetvalue);
7238 } else if (ri->readfn) {
7239 TCGv_ptr tmpptr;
7240 tmp = tcg_temp_new_i32();
7241 tmpptr = tcg_const_ptr(ri);
7242 gen_helper_get_cp_reg(tmp, cpu_env, tmpptr);
7243 tcg_temp_free_ptr(tmpptr);
7244 } else {
7245 tmp = load_cpu_offset(ri->fieldoffset);
7247 if (rt == 15) {
7248 /* Destination register of r15 for 32 bit loads sets
7249 * the condition codes from the high 4 bits of the value
7251 gen_set_nzcv(tmp);
7252 tcg_temp_free_i32(tmp);
7253 } else {
7254 store_reg(s, rt, tmp);
7257 } else {
7258 /* Write */
7259 if (ri->type & ARM_CP_CONST) {
7260 /* If not forbidden by access permissions, treat as WI */
7261 return 0;
7264 if (is64) {
7265 TCGv_i32 tmplo, tmphi;
7266 TCGv_i64 tmp64 = tcg_temp_new_i64();
7267 tmplo = load_reg(s, rt);
7268 tmphi = load_reg(s, rt2);
7269 tcg_gen_concat_i32_i64(tmp64, tmplo, tmphi);
7270 tcg_temp_free_i32(tmplo);
7271 tcg_temp_free_i32(tmphi);
7272 if (ri->writefn) {
7273 TCGv_ptr tmpptr = tcg_const_ptr(ri);
7274 gen_helper_set_cp_reg64(cpu_env, tmpptr, tmp64);
7275 tcg_temp_free_ptr(tmpptr);
7276 } else {
7277 tcg_gen_st_i64(tmp64, cpu_env, ri->fieldoffset);
7279 tcg_temp_free_i64(tmp64);
7280 } else {
7281 if (ri->writefn) {
7282 TCGv_i32 tmp;
7283 TCGv_ptr tmpptr;
7284 tmp = load_reg(s, rt);
7285 tmpptr = tcg_const_ptr(ri);
7286 gen_helper_set_cp_reg(cpu_env, tmpptr, tmp);
7287 tcg_temp_free_ptr(tmpptr);
7288 tcg_temp_free_i32(tmp);
7289 } else {
7290 TCGv_i32 tmp = load_reg(s, rt);
7291 store_cpu_offset(tmp, ri->fieldoffset);
7296 if ((s->tb->cflags & CF_USE_ICOUNT) && (ri->type & ARM_CP_IO)) {
7297 /* I/O operations must end the TB here (whether read or write) */
7298 gen_io_end();
7299 gen_lookup_tb(s);
7300 } else if (!isread && !(ri->type & ARM_CP_SUPPRESS_TB_END)) {
7301 /* We default to ending the TB on a coprocessor register write,
7302 * but allow this to be suppressed by the register definition
7303 * (usually only necessary to work around guest bugs).
7305 gen_lookup_tb(s);
7308 return 0;
7311 /* Unknown register; this might be a guest error or a QEMU
7312 * unimplemented feature.
7314 if (is64) {
7315 qemu_log_mask(LOG_UNIMP, "%s access to unsupported AArch32 "
7316 "64 bit system register cp:%d opc1: %d crm:%d "
7317 "(%s)\n",
7318 isread ? "read" : "write", cpnum, opc1, crm,
7319 s->ns ? "non-secure" : "secure");
7320 } else {
7321 qemu_log_mask(LOG_UNIMP, "%s access to unsupported AArch32 "
7322 "system register cp:%d opc1:%d crn:%d crm:%d opc2:%d "
7323 "(%s)\n",
7324 isread ? "read" : "write", cpnum, opc1, crn, crm, opc2,
7325 s->ns ? "non-secure" : "secure");
7328 return 1;
7332 /* Store a 64-bit value to a register pair. Clobbers val. */
7333 static void gen_storeq_reg(DisasContext *s, int rlow, int rhigh, TCGv_i64 val)
7335 TCGv_i32 tmp;
7336 tmp = tcg_temp_new_i32();
7337 tcg_gen_extrl_i64_i32(tmp, val);
7338 store_reg(s, rlow, tmp);
7339 tmp = tcg_temp_new_i32();
7340 tcg_gen_shri_i64(val, val, 32);
7341 tcg_gen_extrl_i64_i32(tmp, val);
7342 store_reg(s, rhigh, tmp);
7345 /* load a 32-bit value from a register and perform a 64-bit accumulate. */
7346 static void gen_addq_lo(DisasContext *s, TCGv_i64 val, int rlow)
7348 TCGv_i64 tmp;
7349 TCGv_i32 tmp2;
7351 /* Load value and extend to 64 bits. */
7352 tmp = tcg_temp_new_i64();
7353 tmp2 = load_reg(s, rlow);
7354 tcg_gen_extu_i32_i64(tmp, tmp2);
7355 tcg_temp_free_i32(tmp2);
7356 tcg_gen_add_i64(val, val, tmp);
7357 tcg_temp_free_i64(tmp);
7360 /* load and add a 64-bit value from a register pair. */
7361 static void gen_addq(DisasContext *s, TCGv_i64 val, int rlow, int rhigh)
7363 TCGv_i64 tmp;
7364 TCGv_i32 tmpl;
7365 TCGv_i32 tmph;
7367 /* Load 64-bit value rd:rn. */
7368 tmpl = load_reg(s, rlow);
7369 tmph = load_reg(s, rhigh);
7370 tmp = tcg_temp_new_i64();
7371 tcg_gen_concat_i32_i64(tmp, tmpl, tmph);
7372 tcg_temp_free_i32(tmpl);
7373 tcg_temp_free_i32(tmph);
7374 tcg_gen_add_i64(val, val, tmp);
7375 tcg_temp_free_i64(tmp);
7378 /* Set N and Z flags from hi|lo. */
7379 static void gen_logicq_cc(TCGv_i32 lo, TCGv_i32 hi)
7381 tcg_gen_mov_i32(cpu_NF, hi);
7382 tcg_gen_or_i32(cpu_ZF, lo, hi);
7385 /* Load/Store exclusive instructions are implemented by remembering
7386 the value/address loaded, and seeing if these are the same
7387 when the store is performed. This should be sufficient to implement
7388 the architecturally mandated semantics, and avoids having to monitor
7389 regular stores.
7391 In system emulation mode only one CPU will be running at once, so
7392 this sequence is effectively atomic. In user emulation mode we
7393 throw an exception and handle the atomic operation elsewhere. */
7394 static void gen_load_exclusive(DisasContext *s, int rt, int rt2,
7395 TCGv_i32 addr, int size)
7397 TCGv_i32 tmp = tcg_temp_new_i32();
7399 s->is_ldex = true;
7401 switch (size) {
7402 case 0:
7403 gen_aa32_ld8u(tmp, addr, get_mem_index(s));
7404 break;
7405 case 1:
7406 gen_aa32_ld16u(tmp, addr, get_mem_index(s));
7407 break;
7408 case 2:
7409 case 3:
7410 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
7411 break;
7412 default:
7413 abort();
7416 if (size == 3) {
7417 TCGv_i32 tmp2 = tcg_temp_new_i32();
7418 TCGv_i32 tmp3 = tcg_temp_new_i32();
7420 tcg_gen_addi_i32(tmp2, addr, 4);
7421 gen_aa32_ld32u(tmp3, tmp2, get_mem_index(s));
7422 tcg_temp_free_i32(tmp2);
7423 tcg_gen_concat_i32_i64(cpu_exclusive_val, tmp, tmp3);
7424 store_reg(s, rt2, tmp3);
7425 } else {
7426 tcg_gen_extu_i32_i64(cpu_exclusive_val, tmp);
7429 store_reg(s, rt, tmp);
7430 tcg_gen_extu_i32_i64(cpu_exclusive_addr, addr);
7433 static void gen_clrex(DisasContext *s)
7435 tcg_gen_movi_i64(cpu_exclusive_addr, -1);
7438 #ifdef CONFIG_USER_ONLY
7439 static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
7440 TCGv_i32 addr, int size)
7442 tcg_gen_extu_i32_i64(cpu_exclusive_test, addr);
7443 tcg_gen_movi_i32(cpu_exclusive_info,
7444 size | (rd << 4) | (rt << 8) | (rt2 << 12));
7445 gen_exception_internal_insn(s, 4, EXCP_STREX);
7447 #else
7448 static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
7449 TCGv_i32 addr, int size)
7451 TCGv_i32 tmp;
7452 TCGv_i64 val64, extaddr;
7453 TCGLabel *done_label;
7454 TCGLabel *fail_label;
7456 /* if (env->exclusive_addr == addr && env->exclusive_val == [addr]) {
7457 [addr] = {Rt};
7458 {Rd} = 0;
7459 } else {
7460 {Rd} = 1;
7461 } */
7462 fail_label = gen_new_label();
7463 done_label = gen_new_label();
7464 extaddr = tcg_temp_new_i64();
7465 tcg_gen_extu_i32_i64(extaddr, addr);
7466 tcg_gen_brcond_i64(TCG_COND_NE, extaddr, cpu_exclusive_addr, fail_label);
7467 tcg_temp_free_i64(extaddr);
7469 tmp = tcg_temp_new_i32();
7470 switch (size) {
7471 case 0:
7472 gen_aa32_ld8u(tmp, addr, get_mem_index(s));
7473 break;
7474 case 1:
7475 gen_aa32_ld16u(tmp, addr, get_mem_index(s));
7476 break;
7477 case 2:
7478 case 3:
7479 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
7480 break;
7481 default:
7482 abort();
7485 val64 = tcg_temp_new_i64();
7486 if (size == 3) {
7487 TCGv_i32 tmp2 = tcg_temp_new_i32();
7488 TCGv_i32 tmp3 = tcg_temp_new_i32();
7489 tcg_gen_addi_i32(tmp2, addr, 4);
7490 gen_aa32_ld32u(tmp3, tmp2, get_mem_index(s));
7491 tcg_temp_free_i32(tmp2);
7492 tcg_gen_concat_i32_i64(val64, tmp, tmp3);
7493 tcg_temp_free_i32(tmp3);
7494 } else {
7495 tcg_gen_extu_i32_i64(val64, tmp);
7497 tcg_temp_free_i32(tmp);
7499 tcg_gen_brcond_i64(TCG_COND_NE, val64, cpu_exclusive_val, fail_label);
7500 tcg_temp_free_i64(val64);
7502 tmp = load_reg(s, rt);
7503 switch (size) {
7504 case 0:
7505 gen_aa32_st8(tmp, addr, get_mem_index(s));
7506 break;
7507 case 1:
7508 gen_aa32_st16(tmp, addr, get_mem_index(s));
7509 break;
7510 case 2:
7511 case 3:
7512 gen_aa32_st32(tmp, addr, get_mem_index(s));
7513 break;
7514 default:
7515 abort();
7517 tcg_temp_free_i32(tmp);
7518 if (size == 3) {
7519 tcg_gen_addi_i32(addr, addr, 4);
7520 tmp = load_reg(s, rt2);
7521 gen_aa32_st32(tmp, addr, get_mem_index(s));
7522 tcg_temp_free_i32(tmp);
7524 tcg_gen_movi_i32(cpu_R[rd], 0);
7525 tcg_gen_br(done_label);
7526 gen_set_label(fail_label);
7527 tcg_gen_movi_i32(cpu_R[rd], 1);
7528 gen_set_label(done_label);
7529 tcg_gen_movi_i64(cpu_exclusive_addr, -1);
7531 #endif
7533 /* gen_srs:
7534 * @env: CPUARMState
7535 * @s: DisasContext
7536 * @mode: mode field from insn (which stack to store to)
7537 * @amode: addressing mode (DA/IA/DB/IB), encoded as per P,U bits in ARM insn
7538 * @writeback: true if writeback bit set
7540 * Generate code for the SRS (Store Return State) insn.
7542 static void gen_srs(DisasContext *s,
7543 uint32_t mode, uint32_t amode, bool writeback)
7545 int32_t offset;
7546 TCGv_i32 addr = tcg_temp_new_i32();
7547 TCGv_i32 tmp = tcg_const_i32(mode);
7548 gen_helper_get_r13_banked(addr, cpu_env, tmp);
7549 tcg_temp_free_i32(tmp);
7550 switch (amode) {
7551 case 0: /* DA */
7552 offset = -4;
7553 break;
7554 case 1: /* IA */
7555 offset = 0;
7556 break;
7557 case 2: /* DB */
7558 offset = -8;
7559 break;
7560 case 3: /* IB */
7561 offset = 4;
7562 break;
7563 default:
7564 abort();
7566 tcg_gen_addi_i32(addr, addr, offset);
7567 tmp = load_reg(s, 14);
7568 gen_aa32_st32(tmp, addr, get_mem_index(s));
7569 tcg_temp_free_i32(tmp);
7570 tmp = load_cpu_field(spsr);
7571 tcg_gen_addi_i32(addr, addr, 4);
7572 gen_aa32_st32(tmp, addr, get_mem_index(s));
7573 tcg_temp_free_i32(tmp);
7574 if (writeback) {
7575 switch (amode) {
7576 case 0:
7577 offset = -8;
7578 break;
7579 case 1:
7580 offset = 4;
7581 break;
7582 case 2:
7583 offset = -4;
7584 break;
7585 case 3:
7586 offset = 0;
7587 break;
7588 default:
7589 abort();
7591 tcg_gen_addi_i32(addr, addr, offset);
7592 tmp = tcg_const_i32(mode);
7593 gen_helper_set_r13_banked(cpu_env, tmp, addr);
7594 tcg_temp_free_i32(tmp);
7596 tcg_temp_free_i32(addr);
7599 static void disas_arm_insn(DisasContext *s, unsigned int insn)
7601 unsigned int cond, val, op1, i, shift, rm, rs, rn, rd, sh;
7602 TCGv_i32 tmp;
7603 TCGv_i32 tmp2;
7604 TCGv_i32 tmp3;
7605 TCGv_i32 addr;
7606 TCGv_i64 tmp64;
7608 /* M variants do not implement ARM mode. */
7609 if (arm_dc_feature(s, ARM_FEATURE_M)) {
7610 goto illegal_op;
7612 cond = insn >> 28;
7613 if (cond == 0xf){
7614 /* In ARMv3 and v4 the NV condition is UNPREDICTABLE; we
7615 * choose to UNDEF. In ARMv5 and above the space is used
7616 * for miscellaneous unconditional instructions.
7618 ARCH(5);
7620 /* Unconditional instructions. */
7621 if (((insn >> 25) & 7) == 1) {
7622 /* NEON Data processing. */
7623 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
7624 goto illegal_op;
7627 if (disas_neon_data_insn(s, insn)) {
7628 goto illegal_op;
7630 return;
7632 if ((insn & 0x0f100000) == 0x04000000) {
7633 /* NEON load/store. */
7634 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
7635 goto illegal_op;
7638 if (disas_neon_ls_insn(s, insn)) {
7639 goto illegal_op;
7641 return;
7643 if ((insn & 0x0f000e10) == 0x0e000a00) {
7644 /* VFP. */
7645 if (disas_vfp_insn(s, insn)) {
7646 goto illegal_op;
7648 return;
7650 if (((insn & 0x0f30f000) == 0x0510f000) ||
7651 ((insn & 0x0f30f010) == 0x0710f000)) {
7652 if ((insn & (1 << 22)) == 0) {
7653 /* PLDW; v7MP */
7654 if (!arm_dc_feature(s, ARM_FEATURE_V7MP)) {
7655 goto illegal_op;
7658 /* Otherwise PLD; v5TE+ */
7659 ARCH(5TE);
7660 return;
7662 if (((insn & 0x0f70f000) == 0x0450f000) ||
7663 ((insn & 0x0f70f010) == 0x0650f000)) {
7664 ARCH(7);
7665 return; /* PLI; V7 */
7667 if (((insn & 0x0f700000) == 0x04100000) ||
7668 ((insn & 0x0f700010) == 0x06100000)) {
7669 if (!arm_dc_feature(s, ARM_FEATURE_V7MP)) {
7670 goto illegal_op;
7672 return; /* v7MP: Unallocated memory hint: must NOP */
7675 if ((insn & 0x0ffffdff) == 0x01010000) {
7676 ARCH(6);
7677 /* setend */
7678 if (((insn >> 9) & 1) != s->bswap_code) {
7679 /* Dynamic endianness switching not implemented. */
7680 qemu_log_mask(LOG_UNIMP, "arm: unimplemented setend\n");
7681 goto illegal_op;
7683 return;
7684 } else if ((insn & 0x0fffff00) == 0x057ff000) {
7685 switch ((insn >> 4) & 0xf) {
7686 case 1: /* clrex */
7687 ARCH(6K);
7688 gen_clrex(s);
7689 return;
7690 case 4: /* dsb */
7691 case 5: /* dmb */
7692 case 6: /* isb */
7693 ARCH(7);
7694 /* We don't emulate caches so these are a no-op. */
7695 return;
7696 default:
7697 goto illegal_op;
7699 } else if ((insn & 0x0e5fffe0) == 0x084d0500) {
7700 /* srs */
7701 if (IS_USER(s)) {
7702 goto illegal_op;
7704 ARCH(6);
7705 gen_srs(s, (insn & 0x1f), (insn >> 23) & 3, insn & (1 << 21));
7706 return;
7707 } else if ((insn & 0x0e50ffe0) == 0x08100a00) {
7708 /* rfe */
7709 int32_t offset;
7710 if (IS_USER(s))
7711 goto illegal_op;
7712 ARCH(6);
7713 rn = (insn >> 16) & 0xf;
7714 addr = load_reg(s, rn);
7715 i = (insn >> 23) & 3;
7716 switch (i) {
7717 case 0: offset = -4; break; /* DA */
7718 case 1: offset = 0; break; /* IA */
7719 case 2: offset = -8; break; /* DB */
7720 case 3: offset = 4; break; /* IB */
7721 default: abort();
7723 if (offset)
7724 tcg_gen_addi_i32(addr, addr, offset);
7725 /* Load PC into tmp and CPSR into tmp2. */
7726 tmp = tcg_temp_new_i32();
7727 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
7728 tcg_gen_addi_i32(addr, addr, 4);
7729 tmp2 = tcg_temp_new_i32();
7730 gen_aa32_ld32u(tmp2, addr, get_mem_index(s));
7731 if (insn & (1 << 21)) {
7732 /* Base writeback. */
7733 switch (i) {
7734 case 0: offset = -8; break;
7735 case 1: offset = 4; break;
7736 case 2: offset = -4; break;
7737 case 3: offset = 0; break;
7738 default: abort();
7740 if (offset)
7741 tcg_gen_addi_i32(addr, addr, offset);
7742 store_reg(s, rn, addr);
7743 } else {
7744 tcg_temp_free_i32(addr);
7746 gen_rfe(s, tmp, tmp2);
7747 return;
7748 } else if ((insn & 0x0e000000) == 0x0a000000) {
7749 /* branch link and change to thumb (blx <offset>) */
7750 int32_t offset;
7752 val = (uint32_t)s->pc;
7753 tmp = tcg_temp_new_i32();
7754 tcg_gen_movi_i32(tmp, val);
7755 store_reg(s, 14, tmp);
7756 /* Sign-extend the 24-bit offset */
7757 offset = (((int32_t)insn) << 8) >> 8;
7758 /* offset * 4 + bit24 * 2 + (thumb bit) */
7759 val += (offset << 2) | ((insn >> 23) & 2) | 1;
7760 /* pipeline offset */
7761 val += 4;
7762 /* protected by ARCH(5); above, near the start of uncond block */
7763 gen_bx_im(s, val);
7764 return;
7765 } else if ((insn & 0x0e000f00) == 0x0c000100) {
7766 if (arm_dc_feature(s, ARM_FEATURE_IWMMXT)) {
7767 /* iWMMXt register transfer. */
7768 if (extract32(s->c15_cpar, 1, 1)) {
7769 if (!disas_iwmmxt_insn(s, insn)) {
7770 return;
7774 } else if ((insn & 0x0fe00000) == 0x0c400000) {
7775 /* Coprocessor double register transfer. */
7776 ARCH(5TE);
7777 } else if ((insn & 0x0f000010) == 0x0e000010) {
7778 /* Additional coprocessor register transfer. */
7779 } else if ((insn & 0x0ff10020) == 0x01000000) {
7780 uint32_t mask;
7781 uint32_t val;
7782 /* cps (privileged) */
7783 if (IS_USER(s))
7784 return;
7785 mask = val = 0;
7786 if (insn & (1 << 19)) {
7787 if (insn & (1 << 8))
7788 mask |= CPSR_A;
7789 if (insn & (1 << 7))
7790 mask |= CPSR_I;
7791 if (insn & (1 << 6))
7792 mask |= CPSR_F;
7793 if (insn & (1 << 18))
7794 val |= mask;
7796 if (insn & (1 << 17)) {
7797 mask |= CPSR_M;
7798 val |= (insn & 0x1f);
7800 if (mask) {
7801 gen_set_psr_im(s, mask, 0, val);
7803 return;
7805 goto illegal_op;
7807 if (cond != 0xe) {
7808 /* if not always execute, we generate a conditional jump to
7809 next instruction */
7810 s->condlabel = gen_new_label();
7811 arm_gen_test_cc(cond ^ 1, s->condlabel);
7812 s->condjmp = 1;
7814 if ((insn & 0x0f900000) == 0x03000000) {
7815 if ((insn & (1 << 21)) == 0) {
7816 ARCH(6T2);
7817 rd = (insn >> 12) & 0xf;
7818 val = ((insn >> 4) & 0xf000) | (insn & 0xfff);
7819 if ((insn & (1 << 22)) == 0) {
7820 /* MOVW */
7821 tmp = tcg_temp_new_i32();
7822 tcg_gen_movi_i32(tmp, val);
7823 } else {
7824 /* MOVT */
7825 tmp = load_reg(s, rd);
7826 tcg_gen_ext16u_i32(tmp, tmp);
7827 tcg_gen_ori_i32(tmp, tmp, val << 16);
7829 store_reg(s, rd, tmp);
7830 } else {
7831 if (((insn >> 12) & 0xf) != 0xf)
7832 goto illegal_op;
7833 if (((insn >> 16) & 0xf) == 0) {
7834 gen_nop_hint(s, insn & 0xff);
7835 } else {
7836 /* CPSR = immediate */
7837 val = insn & 0xff;
7838 shift = ((insn >> 8) & 0xf) * 2;
7839 if (shift)
7840 val = (val >> shift) | (val << (32 - shift));
7841 i = ((insn & (1 << 22)) != 0);
7842 if (gen_set_psr_im(s, msr_mask(s, (insn >> 16) & 0xf, i),
7843 i, val)) {
7844 goto illegal_op;
7848 } else if ((insn & 0x0f900000) == 0x01000000
7849 && (insn & 0x00000090) != 0x00000090) {
7850 /* miscellaneous instructions */
7851 op1 = (insn >> 21) & 3;
7852 sh = (insn >> 4) & 0xf;
7853 rm = insn & 0xf;
7854 switch (sh) {
7855 case 0x0: /* move program status register */
7856 if (op1 & 1) {
7857 /* PSR = reg */
7858 tmp = load_reg(s, rm);
7859 i = ((op1 & 2) != 0);
7860 if (gen_set_psr(s, msr_mask(s, (insn >> 16) & 0xf, i), i, tmp))
7861 goto illegal_op;
7862 } else {
7863 /* reg = PSR */
7864 rd = (insn >> 12) & 0xf;
7865 if (op1 & 2) {
7866 if (IS_USER(s))
7867 goto illegal_op;
7868 tmp = load_cpu_field(spsr);
7869 } else {
7870 tmp = tcg_temp_new_i32();
7871 gen_helper_cpsr_read(tmp, cpu_env);
7873 store_reg(s, rd, tmp);
7875 break;
7876 case 0x1:
7877 if (op1 == 1) {
7878 /* branch/exchange thumb (bx). */
7879 ARCH(4T);
7880 tmp = load_reg(s, rm);
7881 gen_bx(s, tmp);
7882 } else if (op1 == 3) {
7883 /* clz */
7884 ARCH(5);
7885 rd = (insn >> 12) & 0xf;
7886 tmp = load_reg(s, rm);
7887 gen_helper_clz(tmp, tmp);
7888 store_reg(s, rd, tmp);
7889 } else {
7890 goto illegal_op;
7892 break;
7893 case 0x2:
7894 if (op1 == 1) {
7895 ARCH(5J); /* bxj */
7896 /* Trivial implementation equivalent to bx. */
7897 tmp = load_reg(s, rm);
7898 gen_bx(s, tmp);
7899 } else {
7900 goto illegal_op;
7902 break;
7903 case 0x3:
7904 if (op1 != 1)
7905 goto illegal_op;
7907 ARCH(5);
7908 /* branch link/exchange thumb (blx) */
7909 tmp = load_reg(s, rm);
7910 tmp2 = tcg_temp_new_i32();
7911 tcg_gen_movi_i32(tmp2, s->pc);
7912 store_reg(s, 14, tmp2);
7913 gen_bx(s, tmp);
7914 break;
7915 case 0x4:
7917 /* crc32/crc32c */
7918 uint32_t c = extract32(insn, 8, 4);
7920 /* Check this CPU supports ARMv8 CRC instructions.
7921 * op1 == 3 is UNPREDICTABLE but handle as UNDEFINED.
7922 * Bits 8, 10 and 11 should be zero.
7924 if (!arm_dc_feature(s, ARM_FEATURE_CRC) || op1 == 0x3 ||
7925 (c & 0xd) != 0) {
7926 goto illegal_op;
7929 rn = extract32(insn, 16, 4);
7930 rd = extract32(insn, 12, 4);
7932 tmp = load_reg(s, rn);
7933 tmp2 = load_reg(s, rm);
7934 if (op1 == 0) {
7935 tcg_gen_andi_i32(tmp2, tmp2, 0xff);
7936 } else if (op1 == 1) {
7937 tcg_gen_andi_i32(tmp2, tmp2, 0xffff);
7939 tmp3 = tcg_const_i32(1 << op1);
7940 if (c & 0x2) {
7941 gen_helper_crc32c(tmp, tmp, tmp2, tmp3);
7942 } else {
7943 gen_helper_crc32(tmp, tmp, tmp2, tmp3);
7945 tcg_temp_free_i32(tmp2);
7946 tcg_temp_free_i32(tmp3);
7947 store_reg(s, rd, tmp);
7948 break;
7950 case 0x5: /* saturating add/subtract */
7951 ARCH(5TE);
7952 rd = (insn >> 12) & 0xf;
7953 rn = (insn >> 16) & 0xf;
7954 tmp = load_reg(s, rm);
7955 tmp2 = load_reg(s, rn);
7956 if (op1 & 2)
7957 gen_helper_double_saturate(tmp2, cpu_env, tmp2);
7958 if (op1 & 1)
7959 gen_helper_sub_saturate(tmp, cpu_env, tmp, tmp2);
7960 else
7961 gen_helper_add_saturate(tmp, cpu_env, tmp, tmp2);
7962 tcg_temp_free_i32(tmp2);
7963 store_reg(s, rd, tmp);
7964 break;
7965 case 7:
7967 int imm16 = extract32(insn, 0, 4) | (extract32(insn, 8, 12) << 4);
7968 switch (op1) {
7969 case 1:
7970 /* bkpt */
7971 ARCH(5);
7972 gen_exception_insn(s, 4, EXCP_BKPT,
7973 syn_aa32_bkpt(imm16, false),
7974 default_exception_el(s));
7975 break;
7976 case 2:
7977 /* Hypervisor call (v7) */
7978 ARCH(7);
7979 if (IS_USER(s)) {
7980 goto illegal_op;
7982 gen_hvc(s, imm16);
7983 break;
7984 case 3:
7985 /* Secure monitor call (v6+) */
7986 ARCH(6K);
7987 if (IS_USER(s)) {
7988 goto illegal_op;
7990 gen_smc(s);
7991 break;
7992 default:
7993 goto illegal_op;
7995 break;
7997 case 0x8: /* signed multiply */
7998 case 0xa:
7999 case 0xc:
8000 case 0xe:
8001 ARCH(5TE);
8002 rs = (insn >> 8) & 0xf;
8003 rn = (insn >> 12) & 0xf;
8004 rd = (insn >> 16) & 0xf;
8005 if (op1 == 1) {
8006 /* (32 * 16) >> 16 */
8007 tmp = load_reg(s, rm);
8008 tmp2 = load_reg(s, rs);
8009 if (sh & 4)
8010 tcg_gen_sari_i32(tmp2, tmp2, 16);
8011 else
8012 gen_sxth(tmp2);
8013 tmp64 = gen_muls_i64_i32(tmp, tmp2);
8014 tcg_gen_shri_i64(tmp64, tmp64, 16);
8015 tmp = tcg_temp_new_i32();
8016 tcg_gen_extrl_i64_i32(tmp, tmp64);
8017 tcg_temp_free_i64(tmp64);
8018 if ((sh & 2) == 0) {
8019 tmp2 = load_reg(s, rn);
8020 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
8021 tcg_temp_free_i32(tmp2);
8023 store_reg(s, rd, tmp);
8024 } else {
8025 /* 16 * 16 */
8026 tmp = load_reg(s, rm);
8027 tmp2 = load_reg(s, rs);
8028 gen_mulxy(tmp, tmp2, sh & 2, sh & 4);
8029 tcg_temp_free_i32(tmp2);
8030 if (op1 == 2) {
8031 tmp64 = tcg_temp_new_i64();
8032 tcg_gen_ext_i32_i64(tmp64, tmp);
8033 tcg_temp_free_i32(tmp);
8034 gen_addq(s, tmp64, rn, rd);
8035 gen_storeq_reg(s, rn, rd, tmp64);
8036 tcg_temp_free_i64(tmp64);
8037 } else {
8038 if (op1 == 0) {
8039 tmp2 = load_reg(s, rn);
8040 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
8041 tcg_temp_free_i32(tmp2);
8043 store_reg(s, rd, tmp);
8046 break;
8047 default:
8048 goto illegal_op;
8050 } else if (((insn & 0x0e000000) == 0 &&
8051 (insn & 0x00000090) != 0x90) ||
8052 ((insn & 0x0e000000) == (1 << 25))) {
8053 int set_cc, logic_cc, shiftop;
8055 op1 = (insn >> 21) & 0xf;
8056 set_cc = (insn >> 20) & 1;
8057 logic_cc = table_logic_cc[op1] & set_cc;
8059 /* data processing instruction */
8060 if (insn & (1 << 25)) {
8061 /* immediate operand */
8062 val = insn & 0xff;
8063 shift = ((insn >> 8) & 0xf) * 2;
8064 if (shift) {
8065 val = (val >> shift) | (val << (32 - shift));
8067 tmp2 = tcg_temp_new_i32();
8068 tcg_gen_movi_i32(tmp2, val);
8069 if (logic_cc && shift) {
8070 gen_set_CF_bit31(tmp2);
8072 } else {
8073 /* register */
8074 rm = (insn) & 0xf;
8075 tmp2 = load_reg(s, rm);
8076 shiftop = (insn >> 5) & 3;
8077 if (!(insn & (1 << 4))) {
8078 shift = (insn >> 7) & 0x1f;
8079 gen_arm_shift_im(tmp2, shiftop, shift, logic_cc);
8080 } else {
8081 rs = (insn >> 8) & 0xf;
8082 tmp = load_reg(s, rs);
8083 gen_arm_shift_reg(tmp2, shiftop, tmp, logic_cc);
8086 if (op1 != 0x0f && op1 != 0x0d) {
8087 rn = (insn >> 16) & 0xf;
8088 tmp = load_reg(s, rn);
8089 } else {
8090 TCGV_UNUSED_I32(tmp);
8092 rd = (insn >> 12) & 0xf;
8093 switch(op1) {
8094 case 0x00:
8095 tcg_gen_and_i32(tmp, tmp, tmp2);
8096 if (logic_cc) {
8097 gen_logic_CC(tmp);
8099 store_reg_bx(s, rd, tmp);
8100 break;
8101 case 0x01:
8102 tcg_gen_xor_i32(tmp, tmp, tmp2);
8103 if (logic_cc) {
8104 gen_logic_CC(tmp);
8106 store_reg_bx(s, rd, tmp);
8107 break;
8108 case 0x02:
8109 if (set_cc && rd == 15) {
8110 /* SUBS r15, ... is used for exception return. */
8111 if (IS_USER(s)) {
8112 goto illegal_op;
8114 gen_sub_CC(tmp, tmp, tmp2);
8115 gen_exception_return(s, tmp);
8116 } else {
8117 if (set_cc) {
8118 gen_sub_CC(tmp, tmp, tmp2);
8119 } else {
8120 tcg_gen_sub_i32(tmp, tmp, tmp2);
8122 store_reg_bx(s, rd, tmp);
8124 break;
8125 case 0x03:
8126 if (set_cc) {
8127 gen_sub_CC(tmp, tmp2, tmp);
8128 } else {
8129 tcg_gen_sub_i32(tmp, tmp2, tmp);
8131 store_reg_bx(s, rd, tmp);
8132 break;
8133 case 0x04:
8134 if (set_cc) {
8135 gen_add_CC(tmp, tmp, tmp2);
8136 } else {
8137 tcg_gen_add_i32(tmp, tmp, tmp2);
8139 store_reg_bx(s, rd, tmp);
8140 break;
8141 case 0x05:
8142 if (set_cc) {
8143 gen_adc_CC(tmp, tmp, tmp2);
8144 } else {
8145 gen_add_carry(tmp, tmp, tmp2);
8147 store_reg_bx(s, rd, tmp);
8148 break;
8149 case 0x06:
8150 if (set_cc) {
8151 gen_sbc_CC(tmp, tmp, tmp2);
8152 } else {
8153 gen_sub_carry(tmp, tmp, tmp2);
8155 store_reg_bx(s, rd, tmp);
8156 break;
8157 case 0x07:
8158 if (set_cc) {
8159 gen_sbc_CC(tmp, tmp2, tmp);
8160 } else {
8161 gen_sub_carry(tmp, tmp2, tmp);
8163 store_reg_bx(s, rd, tmp);
8164 break;
8165 case 0x08:
8166 if (set_cc) {
8167 tcg_gen_and_i32(tmp, tmp, tmp2);
8168 gen_logic_CC(tmp);
8170 tcg_temp_free_i32(tmp);
8171 break;
8172 case 0x09:
8173 if (set_cc) {
8174 tcg_gen_xor_i32(tmp, tmp, tmp2);
8175 gen_logic_CC(tmp);
8177 tcg_temp_free_i32(tmp);
8178 break;
8179 case 0x0a:
8180 if (set_cc) {
8181 gen_sub_CC(tmp, tmp, tmp2);
8183 tcg_temp_free_i32(tmp);
8184 break;
8185 case 0x0b:
8186 if (set_cc) {
8187 gen_add_CC(tmp, tmp, tmp2);
8189 tcg_temp_free_i32(tmp);
8190 break;
8191 case 0x0c:
8192 tcg_gen_or_i32(tmp, tmp, tmp2);
8193 if (logic_cc) {
8194 gen_logic_CC(tmp);
8196 store_reg_bx(s, rd, tmp);
8197 break;
8198 case 0x0d:
8199 if (logic_cc && rd == 15) {
8200 /* MOVS r15, ... is used for exception return. */
8201 if (IS_USER(s)) {
8202 goto illegal_op;
8204 gen_exception_return(s, tmp2);
8205 } else {
8206 if (logic_cc) {
8207 gen_logic_CC(tmp2);
8209 store_reg_bx(s, rd, tmp2);
8211 break;
8212 case 0x0e:
8213 tcg_gen_andc_i32(tmp, tmp, tmp2);
8214 if (logic_cc) {
8215 gen_logic_CC(tmp);
8217 store_reg_bx(s, rd, tmp);
8218 break;
8219 default:
8220 case 0x0f:
8221 tcg_gen_not_i32(tmp2, tmp2);
8222 if (logic_cc) {
8223 gen_logic_CC(tmp2);
8225 store_reg_bx(s, rd, tmp2);
8226 break;
8228 if (op1 != 0x0f && op1 != 0x0d) {
8229 tcg_temp_free_i32(tmp2);
8231 } else {
8232 /* other instructions */
8233 op1 = (insn >> 24) & 0xf;
8234 switch(op1) {
8235 case 0x0:
8236 case 0x1:
8237 /* multiplies, extra load/stores */
8238 sh = (insn >> 5) & 3;
8239 if (sh == 0) {
8240 if (op1 == 0x0) {
8241 rd = (insn >> 16) & 0xf;
8242 rn = (insn >> 12) & 0xf;
8243 rs = (insn >> 8) & 0xf;
8244 rm = (insn) & 0xf;
8245 op1 = (insn >> 20) & 0xf;
8246 switch (op1) {
8247 case 0: case 1: case 2: case 3: case 6:
8248 /* 32 bit mul */
8249 tmp = load_reg(s, rs);
8250 tmp2 = load_reg(s, rm);
8251 tcg_gen_mul_i32(tmp, tmp, tmp2);
8252 tcg_temp_free_i32(tmp2);
8253 if (insn & (1 << 22)) {
8254 /* Subtract (mls) */
8255 ARCH(6T2);
8256 tmp2 = load_reg(s, rn);
8257 tcg_gen_sub_i32(tmp, tmp2, tmp);
8258 tcg_temp_free_i32(tmp2);
8259 } else if (insn & (1 << 21)) {
8260 /* Add */
8261 tmp2 = load_reg(s, rn);
8262 tcg_gen_add_i32(tmp, tmp, tmp2);
8263 tcg_temp_free_i32(tmp2);
8265 if (insn & (1 << 20))
8266 gen_logic_CC(tmp);
8267 store_reg(s, rd, tmp);
8268 break;
8269 case 4:
8270 /* 64 bit mul double accumulate (UMAAL) */
8271 ARCH(6);
8272 tmp = load_reg(s, rs);
8273 tmp2 = load_reg(s, rm);
8274 tmp64 = gen_mulu_i64_i32(tmp, tmp2);
8275 gen_addq_lo(s, tmp64, rn);
8276 gen_addq_lo(s, tmp64, rd);
8277 gen_storeq_reg(s, rn, rd, tmp64);
8278 tcg_temp_free_i64(tmp64);
8279 break;
8280 case 8: case 9: case 10: case 11:
8281 case 12: case 13: case 14: case 15:
8282 /* 64 bit mul: UMULL, UMLAL, SMULL, SMLAL. */
8283 tmp = load_reg(s, rs);
8284 tmp2 = load_reg(s, rm);
8285 if (insn & (1 << 22)) {
8286 tcg_gen_muls2_i32(tmp, tmp2, tmp, tmp2);
8287 } else {
8288 tcg_gen_mulu2_i32(tmp, tmp2, tmp, tmp2);
8290 if (insn & (1 << 21)) { /* mult accumulate */
8291 TCGv_i32 al = load_reg(s, rn);
8292 TCGv_i32 ah = load_reg(s, rd);
8293 tcg_gen_add2_i32(tmp, tmp2, tmp, tmp2, al, ah);
8294 tcg_temp_free_i32(al);
8295 tcg_temp_free_i32(ah);
8297 if (insn & (1 << 20)) {
8298 gen_logicq_cc(tmp, tmp2);
8300 store_reg(s, rn, tmp);
8301 store_reg(s, rd, tmp2);
8302 break;
8303 default:
8304 goto illegal_op;
8306 } else {
8307 rn = (insn >> 16) & 0xf;
8308 rd = (insn >> 12) & 0xf;
8309 if (insn & (1 << 23)) {
8310 /* load/store exclusive */
8311 int op2 = (insn >> 8) & 3;
8312 op1 = (insn >> 21) & 0x3;
8314 switch (op2) {
8315 case 0: /* lda/stl */
8316 if (op1 == 1) {
8317 goto illegal_op;
8319 ARCH(8);
8320 break;
8321 case 1: /* reserved */
8322 goto illegal_op;
8323 case 2: /* ldaex/stlex */
8324 ARCH(8);
8325 break;
8326 case 3: /* ldrex/strex */
8327 if (op1) {
8328 ARCH(6K);
8329 } else {
8330 ARCH(6);
8332 break;
8335 addr = tcg_temp_local_new_i32();
8336 load_reg_var(s, addr, rn);
8338 /* Since the emulation does not have barriers,
8339 the acquire/release semantics need no special
8340 handling */
8341 if (op2 == 0) {
8342 if (insn & (1 << 20)) {
8343 tmp = tcg_temp_new_i32();
8344 switch (op1) {
8345 case 0: /* lda */
8346 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
8347 break;
8348 case 2: /* ldab */
8349 gen_aa32_ld8u(tmp, addr, get_mem_index(s));
8350 break;
8351 case 3: /* ldah */
8352 gen_aa32_ld16u(tmp, addr, get_mem_index(s));
8353 break;
8354 default:
8355 abort();
8357 store_reg(s, rd, tmp);
8358 } else {
8359 rm = insn & 0xf;
8360 tmp = load_reg(s, rm);
8361 switch (op1) {
8362 case 0: /* stl */
8363 gen_aa32_st32(tmp, addr, get_mem_index(s));
8364 break;
8365 case 2: /* stlb */
8366 gen_aa32_st8(tmp, addr, get_mem_index(s));
8367 break;
8368 case 3: /* stlh */
8369 gen_aa32_st16(tmp, addr, get_mem_index(s));
8370 break;
8371 default:
8372 abort();
8374 tcg_temp_free_i32(tmp);
8376 } else if (insn & (1 << 20)) {
8377 switch (op1) {
8378 case 0: /* ldrex */
8379 gen_load_exclusive(s, rd, 15, addr, 2);
8380 break;
8381 case 1: /* ldrexd */
8382 gen_load_exclusive(s, rd, rd + 1, addr, 3);
8383 break;
8384 case 2: /* ldrexb */
8385 gen_load_exclusive(s, rd, 15, addr, 0);
8386 break;
8387 case 3: /* ldrexh */
8388 gen_load_exclusive(s, rd, 15, addr, 1);
8389 break;
8390 default:
8391 abort();
8393 } else {
8394 rm = insn & 0xf;
8395 switch (op1) {
8396 case 0: /* strex */
8397 gen_store_exclusive(s, rd, rm, 15, addr, 2);
8398 break;
8399 case 1: /* strexd */
8400 gen_store_exclusive(s, rd, rm, rm + 1, addr, 3);
8401 break;
8402 case 2: /* strexb */
8403 gen_store_exclusive(s, rd, rm, 15, addr, 0);
8404 break;
8405 case 3: /* strexh */
8406 gen_store_exclusive(s, rd, rm, 15, addr, 1);
8407 break;
8408 default:
8409 abort();
8412 tcg_temp_free_i32(addr);
8413 } else {
8414 /* SWP instruction */
8415 rm = (insn) & 0xf;
8417 /* ??? This is not really atomic. However we know
8418 we never have multiple CPUs running in parallel,
8419 so it is good enough. */
8420 addr = load_reg(s, rn);
8421 tmp = load_reg(s, rm);
8422 tmp2 = tcg_temp_new_i32();
8423 if (insn & (1 << 22)) {
8424 gen_aa32_ld8u(tmp2, addr, get_mem_index(s));
8425 gen_aa32_st8(tmp, addr, get_mem_index(s));
8426 } else {
8427 gen_aa32_ld32u(tmp2, addr, get_mem_index(s));
8428 gen_aa32_st32(tmp, addr, get_mem_index(s));
8430 tcg_temp_free_i32(tmp);
8431 tcg_temp_free_i32(addr);
8432 store_reg(s, rd, tmp2);
8435 } else {
8436 int address_offset;
8437 bool load = insn & (1 << 20);
8438 bool doubleword = false;
8439 /* Misc load/store */
8440 rn = (insn >> 16) & 0xf;
8441 rd = (insn >> 12) & 0xf;
8443 if (!load && (sh & 2)) {
8444 /* doubleword */
8445 ARCH(5TE);
8446 if (rd & 1) {
8447 /* UNPREDICTABLE; we choose to UNDEF */
8448 goto illegal_op;
8450 load = (sh & 1) == 0;
8451 doubleword = true;
8454 addr = load_reg(s, rn);
8455 if (insn & (1 << 24))
8456 gen_add_datah_offset(s, insn, 0, addr);
8457 address_offset = 0;
8459 if (doubleword) {
8460 if (!load) {
8461 /* store */
8462 tmp = load_reg(s, rd);
8463 gen_aa32_st32(tmp, addr, get_mem_index(s));
8464 tcg_temp_free_i32(tmp);
8465 tcg_gen_addi_i32(addr, addr, 4);
8466 tmp = load_reg(s, rd + 1);
8467 gen_aa32_st32(tmp, addr, get_mem_index(s));
8468 tcg_temp_free_i32(tmp);
8469 } else {
8470 /* load */
8471 tmp = tcg_temp_new_i32();
8472 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
8473 store_reg(s, rd, tmp);
8474 tcg_gen_addi_i32(addr, addr, 4);
8475 tmp = tcg_temp_new_i32();
8476 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
8477 rd++;
8479 address_offset = -4;
8480 } else if (load) {
8481 /* load */
8482 tmp = tcg_temp_new_i32();
8483 switch (sh) {
8484 case 1:
8485 gen_aa32_ld16u(tmp, addr, get_mem_index(s));
8486 break;
8487 case 2:
8488 gen_aa32_ld8s(tmp, addr, get_mem_index(s));
8489 break;
8490 default:
8491 case 3:
8492 gen_aa32_ld16s(tmp, addr, get_mem_index(s));
8493 break;
8495 } else {
8496 /* store */
8497 tmp = load_reg(s, rd);
8498 gen_aa32_st16(tmp, addr, get_mem_index(s));
8499 tcg_temp_free_i32(tmp);
8501 /* Perform base writeback before the loaded value to
8502 ensure correct behavior with overlapping index registers.
8503 ldrd with base writeback is is undefined if the
8504 destination and index registers overlap. */
8505 if (!(insn & (1 << 24))) {
8506 gen_add_datah_offset(s, insn, address_offset, addr);
8507 store_reg(s, rn, addr);
8508 } else if (insn & (1 << 21)) {
8509 if (address_offset)
8510 tcg_gen_addi_i32(addr, addr, address_offset);
8511 store_reg(s, rn, addr);
8512 } else {
8513 tcg_temp_free_i32(addr);
8515 if (load) {
8516 /* Complete the load. */
8517 store_reg(s, rd, tmp);
8520 break;
8521 case 0x4:
8522 case 0x5:
8523 goto do_ldst;
8524 case 0x6:
8525 case 0x7:
8526 if (insn & (1 << 4)) {
8527 ARCH(6);
8528 /* Armv6 Media instructions. */
8529 rm = insn & 0xf;
8530 rn = (insn >> 16) & 0xf;
8531 rd = (insn >> 12) & 0xf;
8532 rs = (insn >> 8) & 0xf;
8533 switch ((insn >> 23) & 3) {
8534 case 0: /* Parallel add/subtract. */
8535 op1 = (insn >> 20) & 7;
8536 tmp = load_reg(s, rn);
8537 tmp2 = load_reg(s, rm);
8538 sh = (insn >> 5) & 7;
8539 if ((op1 & 3) == 0 || sh == 5 || sh == 6)
8540 goto illegal_op;
8541 gen_arm_parallel_addsub(op1, sh, tmp, tmp2);
8542 tcg_temp_free_i32(tmp2);
8543 store_reg(s, rd, tmp);
8544 break;
8545 case 1:
8546 if ((insn & 0x00700020) == 0) {
8547 /* Halfword pack. */
8548 tmp = load_reg(s, rn);
8549 tmp2 = load_reg(s, rm);
8550 shift = (insn >> 7) & 0x1f;
8551 if (insn & (1 << 6)) {
8552 /* pkhtb */
8553 if (shift == 0)
8554 shift = 31;
8555 tcg_gen_sari_i32(tmp2, tmp2, shift);
8556 tcg_gen_andi_i32(tmp, tmp, 0xffff0000);
8557 tcg_gen_ext16u_i32(tmp2, tmp2);
8558 } else {
8559 /* pkhbt */
8560 if (shift)
8561 tcg_gen_shli_i32(tmp2, tmp2, shift);
8562 tcg_gen_ext16u_i32(tmp, tmp);
8563 tcg_gen_andi_i32(tmp2, tmp2, 0xffff0000);
8565 tcg_gen_or_i32(tmp, tmp, tmp2);
8566 tcg_temp_free_i32(tmp2);
8567 store_reg(s, rd, tmp);
8568 } else if ((insn & 0x00200020) == 0x00200000) {
8569 /* [us]sat */
8570 tmp = load_reg(s, rm);
8571 shift = (insn >> 7) & 0x1f;
8572 if (insn & (1 << 6)) {
8573 if (shift == 0)
8574 shift = 31;
8575 tcg_gen_sari_i32(tmp, tmp, shift);
8576 } else {
8577 tcg_gen_shli_i32(tmp, tmp, shift);
8579 sh = (insn >> 16) & 0x1f;
8580 tmp2 = tcg_const_i32(sh);
8581 if (insn & (1 << 22))
8582 gen_helper_usat(tmp, cpu_env, tmp, tmp2);
8583 else
8584 gen_helper_ssat(tmp, cpu_env, tmp, tmp2);
8585 tcg_temp_free_i32(tmp2);
8586 store_reg(s, rd, tmp);
8587 } else if ((insn & 0x00300fe0) == 0x00200f20) {
8588 /* [us]sat16 */
8589 tmp = load_reg(s, rm);
8590 sh = (insn >> 16) & 0x1f;
8591 tmp2 = tcg_const_i32(sh);
8592 if (insn & (1 << 22))
8593 gen_helper_usat16(tmp, cpu_env, tmp, tmp2);
8594 else
8595 gen_helper_ssat16(tmp, cpu_env, tmp, tmp2);
8596 tcg_temp_free_i32(tmp2);
8597 store_reg(s, rd, tmp);
8598 } else if ((insn & 0x00700fe0) == 0x00000fa0) {
8599 /* Select bytes. */
8600 tmp = load_reg(s, rn);
8601 tmp2 = load_reg(s, rm);
8602 tmp3 = tcg_temp_new_i32();
8603 tcg_gen_ld_i32(tmp3, cpu_env, offsetof(CPUARMState, GE));
8604 gen_helper_sel_flags(tmp, tmp3, tmp, tmp2);
8605 tcg_temp_free_i32(tmp3);
8606 tcg_temp_free_i32(tmp2);
8607 store_reg(s, rd, tmp);
8608 } else if ((insn & 0x000003e0) == 0x00000060) {
8609 tmp = load_reg(s, rm);
8610 shift = (insn >> 10) & 3;
8611 /* ??? In many cases it's not necessary to do a
8612 rotate, a shift is sufficient. */
8613 if (shift != 0)
8614 tcg_gen_rotri_i32(tmp, tmp, shift * 8);
8615 op1 = (insn >> 20) & 7;
8616 switch (op1) {
8617 case 0: gen_sxtb16(tmp); break;
8618 case 2: gen_sxtb(tmp); break;
8619 case 3: gen_sxth(tmp); break;
8620 case 4: gen_uxtb16(tmp); break;
8621 case 6: gen_uxtb(tmp); break;
8622 case 7: gen_uxth(tmp); break;
8623 default: goto illegal_op;
8625 if (rn != 15) {
8626 tmp2 = load_reg(s, rn);
8627 if ((op1 & 3) == 0) {
8628 gen_add16(tmp, tmp2);
8629 } else {
8630 tcg_gen_add_i32(tmp, tmp, tmp2);
8631 tcg_temp_free_i32(tmp2);
8634 store_reg(s, rd, tmp);
8635 } else if ((insn & 0x003f0f60) == 0x003f0f20) {
8636 /* rev */
8637 tmp = load_reg(s, rm);
8638 if (insn & (1 << 22)) {
8639 if (insn & (1 << 7)) {
8640 gen_revsh(tmp);
8641 } else {
8642 ARCH(6T2);
8643 gen_helper_rbit(tmp, tmp);
8645 } else {
8646 if (insn & (1 << 7))
8647 gen_rev16(tmp);
8648 else
8649 tcg_gen_bswap32_i32(tmp, tmp);
8651 store_reg(s, rd, tmp);
8652 } else {
8653 goto illegal_op;
8655 break;
8656 case 2: /* Multiplies (Type 3). */
8657 switch ((insn >> 20) & 0x7) {
8658 case 5:
8659 if (((insn >> 6) ^ (insn >> 7)) & 1) {
8660 /* op2 not 00x or 11x : UNDEF */
8661 goto illegal_op;
8663 /* Signed multiply most significant [accumulate].
8664 (SMMUL, SMMLA, SMMLS) */
8665 tmp = load_reg(s, rm);
8666 tmp2 = load_reg(s, rs);
8667 tmp64 = gen_muls_i64_i32(tmp, tmp2);
8669 if (rd != 15) {
8670 tmp = load_reg(s, rd);
8671 if (insn & (1 << 6)) {
8672 tmp64 = gen_subq_msw(tmp64, tmp);
8673 } else {
8674 tmp64 = gen_addq_msw(tmp64, tmp);
8677 if (insn & (1 << 5)) {
8678 tcg_gen_addi_i64(tmp64, tmp64, 0x80000000u);
8680 tcg_gen_shri_i64(tmp64, tmp64, 32);
8681 tmp = tcg_temp_new_i32();
8682 tcg_gen_extrl_i64_i32(tmp, tmp64);
8683 tcg_temp_free_i64(tmp64);
8684 store_reg(s, rn, tmp);
8685 break;
8686 case 0:
8687 case 4:
8688 /* SMLAD, SMUAD, SMLSD, SMUSD, SMLALD, SMLSLD */
8689 if (insn & (1 << 7)) {
8690 goto illegal_op;
8692 tmp = load_reg(s, rm);
8693 tmp2 = load_reg(s, rs);
8694 if (insn & (1 << 5))
8695 gen_swap_half(tmp2);
8696 gen_smul_dual(tmp, tmp2);
8697 if (insn & (1 << 22)) {
8698 /* smlald, smlsld */
8699 TCGv_i64 tmp64_2;
8701 tmp64 = tcg_temp_new_i64();
8702 tmp64_2 = tcg_temp_new_i64();
8703 tcg_gen_ext_i32_i64(tmp64, tmp);
8704 tcg_gen_ext_i32_i64(tmp64_2, tmp2);
8705 tcg_temp_free_i32(tmp);
8706 tcg_temp_free_i32(tmp2);
8707 if (insn & (1 << 6)) {
8708 tcg_gen_sub_i64(tmp64, tmp64, tmp64_2);
8709 } else {
8710 tcg_gen_add_i64(tmp64, tmp64, tmp64_2);
8712 tcg_temp_free_i64(tmp64_2);
8713 gen_addq(s, tmp64, rd, rn);
8714 gen_storeq_reg(s, rd, rn, tmp64);
8715 tcg_temp_free_i64(tmp64);
8716 } else {
8717 /* smuad, smusd, smlad, smlsd */
8718 if (insn & (1 << 6)) {
8719 /* This subtraction cannot overflow. */
8720 tcg_gen_sub_i32(tmp, tmp, tmp2);
8721 } else {
8722 /* This addition cannot overflow 32 bits;
8723 * however it may overflow considered as a
8724 * signed operation, in which case we must set
8725 * the Q flag.
8727 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
8729 tcg_temp_free_i32(tmp2);
8730 if (rd != 15)
8732 tmp2 = load_reg(s, rd);
8733 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
8734 tcg_temp_free_i32(tmp2);
8736 store_reg(s, rn, tmp);
8738 break;
8739 case 1:
8740 case 3:
8741 /* SDIV, UDIV */
8742 if (!arm_dc_feature(s, ARM_FEATURE_ARM_DIV)) {
8743 goto illegal_op;
8745 if (((insn >> 5) & 7) || (rd != 15)) {
8746 goto illegal_op;
8748 tmp = load_reg(s, rm);
8749 tmp2 = load_reg(s, rs);
8750 if (insn & (1 << 21)) {
8751 gen_helper_udiv(tmp, tmp, tmp2);
8752 } else {
8753 gen_helper_sdiv(tmp, tmp, tmp2);
8755 tcg_temp_free_i32(tmp2);
8756 store_reg(s, rn, tmp);
8757 break;
8758 default:
8759 goto illegal_op;
8761 break;
8762 case 3:
8763 op1 = ((insn >> 17) & 0x38) | ((insn >> 5) & 7);
8764 switch (op1) {
8765 case 0: /* Unsigned sum of absolute differences. */
8766 ARCH(6);
8767 tmp = load_reg(s, rm);
8768 tmp2 = load_reg(s, rs);
8769 gen_helper_usad8(tmp, tmp, tmp2);
8770 tcg_temp_free_i32(tmp2);
8771 if (rd != 15) {
8772 tmp2 = load_reg(s, rd);
8773 tcg_gen_add_i32(tmp, tmp, tmp2);
8774 tcg_temp_free_i32(tmp2);
8776 store_reg(s, rn, tmp);
8777 break;
8778 case 0x20: case 0x24: case 0x28: case 0x2c:
8779 /* Bitfield insert/clear. */
8780 ARCH(6T2);
8781 shift = (insn >> 7) & 0x1f;
8782 i = (insn >> 16) & 0x1f;
8783 if (i < shift) {
8784 /* UNPREDICTABLE; we choose to UNDEF */
8785 goto illegal_op;
8787 i = i + 1 - shift;
8788 if (rm == 15) {
8789 tmp = tcg_temp_new_i32();
8790 tcg_gen_movi_i32(tmp, 0);
8791 } else {
8792 tmp = load_reg(s, rm);
8794 if (i != 32) {
8795 tmp2 = load_reg(s, rd);
8796 tcg_gen_deposit_i32(tmp, tmp2, tmp, shift, i);
8797 tcg_temp_free_i32(tmp2);
8799 store_reg(s, rd, tmp);
8800 break;
8801 case 0x12: case 0x16: case 0x1a: case 0x1e: /* sbfx */
8802 case 0x32: case 0x36: case 0x3a: case 0x3e: /* ubfx */
8803 ARCH(6T2);
8804 tmp = load_reg(s, rm);
8805 shift = (insn >> 7) & 0x1f;
8806 i = ((insn >> 16) & 0x1f) + 1;
8807 if (shift + i > 32)
8808 goto illegal_op;
8809 if (i < 32) {
8810 if (op1 & 0x20) {
8811 gen_ubfx(tmp, shift, (1u << i) - 1);
8812 } else {
8813 gen_sbfx(tmp, shift, i);
8816 store_reg(s, rd, tmp);
8817 break;
8818 default:
8819 goto illegal_op;
8821 break;
8823 break;
8825 do_ldst:
8826 /* Check for undefined extension instructions
8827 * per the ARM Bible IE:
8828 * xxxx 0111 1111 xxxx xxxx xxxx 1111 xxxx
8830 sh = (0xf << 20) | (0xf << 4);
8831 if (op1 == 0x7 && ((insn & sh) == sh))
8833 goto illegal_op;
8835 /* load/store byte/word */
8836 rn = (insn >> 16) & 0xf;
8837 rd = (insn >> 12) & 0xf;
8838 tmp2 = load_reg(s, rn);
8839 if ((insn & 0x01200000) == 0x00200000) {
8840 /* ldrt/strt */
8841 i = get_a32_user_mem_index(s);
8842 } else {
8843 i = get_mem_index(s);
8845 if (insn & (1 << 24))
8846 gen_add_data_offset(s, insn, tmp2);
8847 if (insn & (1 << 20)) {
8848 /* load */
8849 tmp = tcg_temp_new_i32();
8850 if (insn & (1 << 22)) {
8851 gen_aa32_ld8u(tmp, tmp2, i);
8852 } else {
8853 gen_aa32_ld32u(tmp, tmp2, i);
8855 } else {
8856 /* store */
8857 tmp = load_reg(s, rd);
8858 if (insn & (1 << 22)) {
8859 gen_aa32_st8(tmp, tmp2, i);
8860 } else {
8861 gen_aa32_st32(tmp, tmp2, i);
8863 tcg_temp_free_i32(tmp);
8865 if (!(insn & (1 << 24))) {
8866 gen_add_data_offset(s, insn, tmp2);
8867 store_reg(s, rn, tmp2);
8868 } else if (insn & (1 << 21)) {
8869 store_reg(s, rn, tmp2);
8870 } else {
8871 tcg_temp_free_i32(tmp2);
8873 if (insn & (1 << 20)) {
8874 /* Complete the load. */
8875 store_reg_from_load(s, rd, tmp);
8877 break;
8878 case 0x08:
8879 case 0x09:
8881 int j, n, loaded_base;
8882 bool exc_return = false;
8883 bool is_load = extract32(insn, 20, 1);
8884 bool user = false;
8885 TCGv_i32 loaded_var;
8886 /* load/store multiple words */
8887 /* XXX: store correct base if write back */
8888 if (insn & (1 << 22)) {
8889 /* LDM (user), LDM (exception return) and STM (user) */
8890 if (IS_USER(s))
8891 goto illegal_op; /* only usable in supervisor mode */
8893 if (is_load && extract32(insn, 15, 1)) {
8894 exc_return = true;
8895 } else {
8896 user = true;
8899 rn = (insn >> 16) & 0xf;
8900 addr = load_reg(s, rn);
8902 /* compute total size */
8903 loaded_base = 0;
8904 TCGV_UNUSED_I32(loaded_var);
8905 n = 0;
8906 for(i=0;i<16;i++) {
8907 if (insn & (1 << i))
8908 n++;
8910 /* XXX: test invalid n == 0 case ? */
8911 if (insn & (1 << 23)) {
8912 if (insn & (1 << 24)) {
8913 /* pre increment */
8914 tcg_gen_addi_i32(addr, addr, 4);
8915 } else {
8916 /* post increment */
8918 } else {
8919 if (insn & (1 << 24)) {
8920 /* pre decrement */
8921 tcg_gen_addi_i32(addr, addr, -(n * 4));
8922 } else {
8923 /* post decrement */
8924 if (n != 1)
8925 tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
8928 j = 0;
8929 for(i=0;i<16;i++) {
8930 if (insn & (1 << i)) {
8931 if (is_load) {
8932 /* load */
8933 tmp = tcg_temp_new_i32();
8934 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
8935 if (user) {
8936 tmp2 = tcg_const_i32(i);
8937 gen_helper_set_user_reg(cpu_env, tmp2, tmp);
8938 tcg_temp_free_i32(tmp2);
8939 tcg_temp_free_i32(tmp);
8940 } else if (i == rn) {
8941 loaded_var = tmp;
8942 loaded_base = 1;
8943 } else {
8944 store_reg_from_load(s, i, tmp);
8946 } else {
8947 /* store */
8948 if (i == 15) {
8949 /* special case: r15 = PC + 8 */
8950 val = (long)s->pc + 4;
8951 tmp = tcg_temp_new_i32();
8952 tcg_gen_movi_i32(tmp, val);
8953 } else if (user) {
8954 tmp = tcg_temp_new_i32();
8955 tmp2 = tcg_const_i32(i);
8956 gen_helper_get_user_reg(tmp, cpu_env, tmp2);
8957 tcg_temp_free_i32(tmp2);
8958 } else {
8959 tmp = load_reg(s, i);
8961 gen_aa32_st32(tmp, addr, get_mem_index(s));
8962 tcg_temp_free_i32(tmp);
8964 j++;
8965 /* no need to add after the last transfer */
8966 if (j != n)
8967 tcg_gen_addi_i32(addr, addr, 4);
8970 if (insn & (1 << 21)) {
8971 /* write back */
8972 if (insn & (1 << 23)) {
8973 if (insn & (1 << 24)) {
8974 /* pre increment */
8975 } else {
8976 /* post increment */
8977 tcg_gen_addi_i32(addr, addr, 4);
8979 } else {
8980 if (insn & (1 << 24)) {
8981 /* pre decrement */
8982 if (n != 1)
8983 tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
8984 } else {
8985 /* post decrement */
8986 tcg_gen_addi_i32(addr, addr, -(n * 4));
8989 store_reg(s, rn, addr);
8990 } else {
8991 tcg_temp_free_i32(addr);
8993 if (loaded_base) {
8994 store_reg(s, rn, loaded_var);
8996 if (exc_return) {
8997 /* Restore CPSR from SPSR. */
8998 tmp = load_cpu_field(spsr);
8999 gen_set_cpsr(tmp, CPSR_ERET_MASK);
9000 tcg_temp_free_i32(tmp);
9001 s->is_jmp = DISAS_UPDATE;
9004 break;
9005 case 0xa:
9006 case 0xb:
9008 int32_t offset;
9010 /* branch (and link) */
9011 val = (int32_t)s->pc;
9012 if (insn & (1 << 24)) {
9013 tmp = tcg_temp_new_i32();
9014 tcg_gen_movi_i32(tmp, val);
9015 store_reg(s, 14, tmp);
9017 offset = sextract32(insn << 2, 0, 26);
9018 val += offset + 4;
9019 gen_jmp(s, val);
9021 break;
9022 case 0xc:
9023 case 0xd:
9024 case 0xe:
9025 if (((insn >> 8) & 0xe) == 10) {
9026 /* VFP. */
9027 if (disas_vfp_insn(s, insn)) {
9028 goto illegal_op;
9030 } else if (disas_coproc_insn(s, insn)) {
9031 /* Coprocessor. */
9032 goto illegal_op;
9034 break;
9035 case 0xf:
9036 /* swi */
9037 gen_set_pc_im(s, s->pc);
9038 s->svc_imm = extract32(insn, 0, 24);
9039 s->is_jmp = DISAS_SWI;
9040 break;
9041 default:
9042 illegal_op:
9043 gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized(),
9044 default_exception_el(s));
9045 break;
9050 /* Return true if this is a Thumb-2 logical op. */
9051 static int
9052 thumb2_logic_op(int op)
9054 return (op < 8);
9057 /* Generate code for a Thumb-2 data processing operation. If CONDS is nonzero
9058 then set condition code flags based on the result of the operation.
9059 If SHIFTER_OUT is nonzero then set the carry flag for logical operations
9060 to the high bit of T1.
9061 Returns zero if the opcode is valid. */
9063 static int
9064 gen_thumb2_data_op(DisasContext *s, int op, int conds, uint32_t shifter_out,
9065 TCGv_i32 t0, TCGv_i32 t1)
9067 int logic_cc;
9069 logic_cc = 0;
9070 switch (op) {
9071 case 0: /* and */
9072 tcg_gen_and_i32(t0, t0, t1);
9073 logic_cc = conds;
9074 break;
9075 case 1: /* bic */
9076 tcg_gen_andc_i32(t0, t0, t1);
9077 logic_cc = conds;
9078 break;
9079 case 2: /* orr */
9080 tcg_gen_or_i32(t0, t0, t1);
9081 logic_cc = conds;
9082 break;
9083 case 3: /* orn */
9084 tcg_gen_orc_i32(t0, t0, t1);
9085 logic_cc = conds;
9086 break;
9087 case 4: /* eor */
9088 tcg_gen_xor_i32(t0, t0, t1);
9089 logic_cc = conds;
9090 break;
9091 case 8: /* add */
9092 if (conds)
9093 gen_add_CC(t0, t0, t1);
9094 else
9095 tcg_gen_add_i32(t0, t0, t1);
9096 break;
9097 case 10: /* adc */
9098 if (conds)
9099 gen_adc_CC(t0, t0, t1);
9100 else
9101 gen_adc(t0, t1);
9102 break;
9103 case 11: /* sbc */
9104 if (conds) {
9105 gen_sbc_CC(t0, t0, t1);
9106 } else {
9107 gen_sub_carry(t0, t0, t1);
9109 break;
9110 case 13: /* sub */
9111 if (conds)
9112 gen_sub_CC(t0, t0, t1);
9113 else
9114 tcg_gen_sub_i32(t0, t0, t1);
9115 break;
9116 case 14: /* rsb */
9117 if (conds)
9118 gen_sub_CC(t0, t1, t0);
9119 else
9120 tcg_gen_sub_i32(t0, t1, t0);
9121 break;
9122 default: /* 5, 6, 7, 9, 12, 15. */
9123 return 1;
9125 if (logic_cc) {
9126 gen_logic_CC(t0);
9127 if (shifter_out)
9128 gen_set_CF_bit31(t1);
9130 return 0;
9133 /* Translate a 32-bit thumb instruction. Returns nonzero if the instruction
9134 is not legal. */
9135 static int disas_thumb2_insn(CPUARMState *env, DisasContext *s, uint16_t insn_hw1)
9137 uint32_t insn, imm, shift, offset;
9138 uint32_t rd, rn, rm, rs;
9139 TCGv_i32 tmp;
9140 TCGv_i32 tmp2;
9141 TCGv_i32 tmp3;
9142 TCGv_i32 addr;
9143 TCGv_i64 tmp64;
9144 int op;
9145 int shiftop;
9146 int conds;
9147 int logic_cc;
9149 if (!(arm_dc_feature(s, ARM_FEATURE_THUMB2)
9150 || arm_dc_feature(s, ARM_FEATURE_M))) {
9151 /* Thumb-1 cores may need to treat bl and blx as a pair of
9152 16-bit instructions to get correct prefetch abort behavior. */
9153 insn = insn_hw1;
9154 if ((insn & (1 << 12)) == 0) {
9155 ARCH(5);
9156 /* Second half of blx. */
9157 offset = ((insn & 0x7ff) << 1);
9158 tmp = load_reg(s, 14);
9159 tcg_gen_addi_i32(tmp, tmp, offset);
9160 tcg_gen_andi_i32(tmp, tmp, 0xfffffffc);
9162 tmp2 = tcg_temp_new_i32();
9163 tcg_gen_movi_i32(tmp2, s->pc | 1);
9164 store_reg(s, 14, tmp2);
9165 gen_bx(s, tmp);
9166 return 0;
9168 if (insn & (1 << 11)) {
9169 /* Second half of bl. */
9170 offset = ((insn & 0x7ff) << 1) | 1;
9171 tmp = load_reg(s, 14);
9172 tcg_gen_addi_i32(tmp, tmp, offset);
9174 tmp2 = tcg_temp_new_i32();
9175 tcg_gen_movi_i32(tmp2, s->pc | 1);
9176 store_reg(s, 14, tmp2);
9177 gen_bx(s, tmp);
9178 return 0;
9180 if ((s->pc & ~TARGET_PAGE_MASK) == 0) {
9181 /* Instruction spans a page boundary. Implement it as two
9182 16-bit instructions in case the second half causes an
9183 prefetch abort. */
9184 offset = ((int32_t)insn << 21) >> 9;
9185 tcg_gen_movi_i32(cpu_R[14], s->pc + 2 + offset);
9186 return 0;
9188 /* Fall through to 32-bit decode. */
9191 insn = arm_lduw_code(env, s->pc, s->bswap_code);
9192 s->pc += 2;
9193 insn |= (uint32_t)insn_hw1 << 16;
9195 if ((insn & 0xf800e800) != 0xf000e800) {
9196 ARCH(6T2);
9199 rn = (insn >> 16) & 0xf;
9200 rs = (insn >> 12) & 0xf;
9201 rd = (insn >> 8) & 0xf;
9202 rm = insn & 0xf;
9203 switch ((insn >> 25) & 0xf) {
9204 case 0: case 1: case 2: case 3:
9205 /* 16-bit instructions. Should never happen. */
9206 abort();
9207 case 4:
9208 if (insn & (1 << 22)) {
9209 /* Other load/store, table branch. */
9210 if (insn & 0x01200000) {
9211 /* Load/store doubleword. */
9212 if (rn == 15) {
9213 addr = tcg_temp_new_i32();
9214 tcg_gen_movi_i32(addr, s->pc & ~3);
9215 } else {
9216 addr = load_reg(s, rn);
9218 offset = (insn & 0xff) * 4;
9219 if ((insn & (1 << 23)) == 0)
9220 offset = -offset;
9221 if (insn & (1 << 24)) {
9222 tcg_gen_addi_i32(addr, addr, offset);
9223 offset = 0;
9225 if (insn & (1 << 20)) {
9226 /* ldrd */
9227 tmp = tcg_temp_new_i32();
9228 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
9229 store_reg(s, rs, tmp);
9230 tcg_gen_addi_i32(addr, addr, 4);
9231 tmp = tcg_temp_new_i32();
9232 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
9233 store_reg(s, rd, tmp);
9234 } else {
9235 /* strd */
9236 tmp = load_reg(s, rs);
9237 gen_aa32_st32(tmp, addr, get_mem_index(s));
9238 tcg_temp_free_i32(tmp);
9239 tcg_gen_addi_i32(addr, addr, 4);
9240 tmp = load_reg(s, rd);
9241 gen_aa32_st32(tmp, addr, get_mem_index(s));
9242 tcg_temp_free_i32(tmp);
9244 if (insn & (1 << 21)) {
9245 /* Base writeback. */
9246 if (rn == 15)
9247 goto illegal_op;
9248 tcg_gen_addi_i32(addr, addr, offset - 4);
9249 store_reg(s, rn, addr);
9250 } else {
9251 tcg_temp_free_i32(addr);
9253 } else if ((insn & (1 << 23)) == 0) {
9254 /* Load/store exclusive word. */
9255 addr = tcg_temp_local_new_i32();
9256 load_reg_var(s, addr, rn);
9257 tcg_gen_addi_i32(addr, addr, (insn & 0xff) << 2);
9258 if (insn & (1 << 20)) {
9259 gen_load_exclusive(s, rs, 15, addr, 2);
9260 } else {
9261 gen_store_exclusive(s, rd, rs, 15, addr, 2);
9263 tcg_temp_free_i32(addr);
9264 } else if ((insn & (7 << 5)) == 0) {
9265 /* Table Branch. */
9266 if (rn == 15) {
9267 addr = tcg_temp_new_i32();
9268 tcg_gen_movi_i32(addr, s->pc);
9269 } else {
9270 addr = load_reg(s, rn);
9272 tmp = load_reg(s, rm);
9273 tcg_gen_add_i32(addr, addr, tmp);
9274 if (insn & (1 << 4)) {
9275 /* tbh */
9276 tcg_gen_add_i32(addr, addr, tmp);
9277 tcg_temp_free_i32(tmp);
9278 tmp = tcg_temp_new_i32();
9279 gen_aa32_ld16u(tmp, addr, get_mem_index(s));
9280 } else { /* tbb */
9281 tcg_temp_free_i32(tmp);
9282 tmp = tcg_temp_new_i32();
9283 gen_aa32_ld8u(tmp, addr, get_mem_index(s));
9285 tcg_temp_free_i32(addr);
9286 tcg_gen_shli_i32(tmp, tmp, 1);
9287 tcg_gen_addi_i32(tmp, tmp, s->pc);
9288 store_reg(s, 15, tmp);
9289 } else {
9290 int op2 = (insn >> 6) & 0x3;
9291 op = (insn >> 4) & 0x3;
9292 switch (op2) {
9293 case 0:
9294 goto illegal_op;
9295 case 1:
9296 /* Load/store exclusive byte/halfword/doubleword */
9297 if (op == 2) {
9298 goto illegal_op;
9300 ARCH(7);
9301 break;
9302 case 2:
9303 /* Load-acquire/store-release */
9304 if (op == 3) {
9305 goto illegal_op;
9307 /* Fall through */
9308 case 3:
9309 /* Load-acquire/store-release exclusive */
9310 ARCH(8);
9311 break;
9313 addr = tcg_temp_local_new_i32();
9314 load_reg_var(s, addr, rn);
9315 if (!(op2 & 1)) {
9316 if (insn & (1 << 20)) {
9317 tmp = tcg_temp_new_i32();
9318 switch (op) {
9319 case 0: /* ldab */
9320 gen_aa32_ld8u(tmp, addr, get_mem_index(s));
9321 break;
9322 case 1: /* ldah */
9323 gen_aa32_ld16u(tmp, addr, get_mem_index(s));
9324 break;
9325 case 2: /* lda */
9326 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
9327 break;
9328 default:
9329 abort();
9331 store_reg(s, rs, tmp);
9332 } else {
9333 tmp = load_reg(s, rs);
9334 switch (op) {
9335 case 0: /* stlb */
9336 gen_aa32_st8(tmp, addr, get_mem_index(s));
9337 break;
9338 case 1: /* stlh */
9339 gen_aa32_st16(tmp, addr, get_mem_index(s));
9340 break;
9341 case 2: /* stl */
9342 gen_aa32_st32(tmp, addr, get_mem_index(s));
9343 break;
9344 default:
9345 abort();
9347 tcg_temp_free_i32(tmp);
9349 } else if (insn & (1 << 20)) {
9350 gen_load_exclusive(s, rs, rd, addr, op);
9351 } else {
9352 gen_store_exclusive(s, rm, rs, rd, addr, op);
9354 tcg_temp_free_i32(addr);
9356 } else {
9357 /* Load/store multiple, RFE, SRS. */
9358 if (((insn >> 23) & 1) == ((insn >> 24) & 1)) {
9359 /* RFE, SRS: not available in user mode or on M profile */
9360 if (IS_USER(s) || arm_dc_feature(s, ARM_FEATURE_M)) {
9361 goto illegal_op;
9363 if (insn & (1 << 20)) {
9364 /* rfe */
9365 addr = load_reg(s, rn);
9366 if ((insn & (1 << 24)) == 0)
9367 tcg_gen_addi_i32(addr, addr, -8);
9368 /* Load PC into tmp and CPSR into tmp2. */
9369 tmp = tcg_temp_new_i32();
9370 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
9371 tcg_gen_addi_i32(addr, addr, 4);
9372 tmp2 = tcg_temp_new_i32();
9373 gen_aa32_ld32u(tmp2, addr, get_mem_index(s));
9374 if (insn & (1 << 21)) {
9375 /* Base writeback. */
9376 if (insn & (1 << 24)) {
9377 tcg_gen_addi_i32(addr, addr, 4);
9378 } else {
9379 tcg_gen_addi_i32(addr, addr, -4);
9381 store_reg(s, rn, addr);
9382 } else {
9383 tcg_temp_free_i32(addr);
9385 gen_rfe(s, tmp, tmp2);
9386 } else {
9387 /* srs */
9388 gen_srs(s, (insn & 0x1f), (insn & (1 << 24)) ? 1 : 2,
9389 insn & (1 << 21));
9391 } else {
9392 int i, loaded_base = 0;
9393 TCGv_i32 loaded_var;
9394 /* Load/store multiple. */
9395 addr = load_reg(s, rn);
9396 offset = 0;
9397 for (i = 0; i < 16; i++) {
9398 if (insn & (1 << i))
9399 offset += 4;
9401 if (insn & (1 << 24)) {
9402 tcg_gen_addi_i32(addr, addr, -offset);
9405 TCGV_UNUSED_I32(loaded_var);
9406 for (i = 0; i < 16; i++) {
9407 if ((insn & (1 << i)) == 0)
9408 continue;
9409 if (insn & (1 << 20)) {
9410 /* Load. */
9411 tmp = tcg_temp_new_i32();
9412 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
9413 if (i == 15) {
9414 gen_bx(s, tmp);
9415 } else if (i == rn) {
9416 loaded_var = tmp;
9417 loaded_base = 1;
9418 } else {
9419 store_reg(s, i, tmp);
9421 } else {
9422 /* Store. */
9423 tmp = load_reg(s, i);
9424 gen_aa32_st32(tmp, addr, get_mem_index(s));
9425 tcg_temp_free_i32(tmp);
9427 tcg_gen_addi_i32(addr, addr, 4);
9429 if (loaded_base) {
9430 store_reg(s, rn, loaded_var);
9432 if (insn & (1 << 21)) {
9433 /* Base register writeback. */
9434 if (insn & (1 << 24)) {
9435 tcg_gen_addi_i32(addr, addr, -offset);
9437 /* Fault if writeback register is in register list. */
9438 if (insn & (1 << rn))
9439 goto illegal_op;
9440 store_reg(s, rn, addr);
9441 } else {
9442 tcg_temp_free_i32(addr);
9446 break;
9447 case 5:
9449 op = (insn >> 21) & 0xf;
9450 if (op == 6) {
9451 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9452 goto illegal_op;
9454 /* Halfword pack. */
9455 tmp = load_reg(s, rn);
9456 tmp2 = load_reg(s, rm);
9457 shift = ((insn >> 10) & 0x1c) | ((insn >> 6) & 0x3);
9458 if (insn & (1 << 5)) {
9459 /* pkhtb */
9460 if (shift == 0)
9461 shift = 31;
9462 tcg_gen_sari_i32(tmp2, tmp2, shift);
9463 tcg_gen_andi_i32(tmp, tmp, 0xffff0000);
9464 tcg_gen_ext16u_i32(tmp2, tmp2);
9465 } else {
9466 /* pkhbt */
9467 if (shift)
9468 tcg_gen_shli_i32(tmp2, tmp2, shift);
9469 tcg_gen_ext16u_i32(tmp, tmp);
9470 tcg_gen_andi_i32(tmp2, tmp2, 0xffff0000);
9472 tcg_gen_or_i32(tmp, tmp, tmp2);
9473 tcg_temp_free_i32(tmp2);
9474 store_reg(s, rd, tmp);
9475 } else {
9476 /* Data processing register constant shift. */
9477 if (rn == 15) {
9478 tmp = tcg_temp_new_i32();
9479 tcg_gen_movi_i32(tmp, 0);
9480 } else {
9481 tmp = load_reg(s, rn);
9483 tmp2 = load_reg(s, rm);
9485 shiftop = (insn >> 4) & 3;
9486 shift = ((insn >> 6) & 3) | ((insn >> 10) & 0x1c);
9487 conds = (insn & (1 << 20)) != 0;
9488 logic_cc = (conds && thumb2_logic_op(op));
9489 gen_arm_shift_im(tmp2, shiftop, shift, logic_cc);
9490 if (gen_thumb2_data_op(s, op, conds, 0, tmp, tmp2))
9491 goto illegal_op;
9492 tcg_temp_free_i32(tmp2);
9493 if (rd != 15) {
9494 store_reg(s, rd, tmp);
9495 } else {
9496 tcg_temp_free_i32(tmp);
9499 break;
9500 case 13: /* Misc data processing. */
9501 op = ((insn >> 22) & 6) | ((insn >> 7) & 1);
9502 if (op < 4 && (insn & 0xf000) != 0xf000)
9503 goto illegal_op;
9504 switch (op) {
9505 case 0: /* Register controlled shift. */
9506 tmp = load_reg(s, rn);
9507 tmp2 = load_reg(s, rm);
9508 if ((insn & 0x70) != 0)
9509 goto illegal_op;
9510 op = (insn >> 21) & 3;
9511 logic_cc = (insn & (1 << 20)) != 0;
9512 gen_arm_shift_reg(tmp, op, tmp2, logic_cc);
9513 if (logic_cc)
9514 gen_logic_CC(tmp);
9515 store_reg_bx(s, rd, tmp);
9516 break;
9517 case 1: /* Sign/zero extend. */
9518 op = (insn >> 20) & 7;
9519 switch (op) {
9520 case 0: /* SXTAH, SXTH */
9521 case 1: /* UXTAH, UXTH */
9522 case 4: /* SXTAB, SXTB */
9523 case 5: /* UXTAB, UXTB */
9524 break;
9525 case 2: /* SXTAB16, SXTB16 */
9526 case 3: /* UXTAB16, UXTB16 */
9527 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9528 goto illegal_op;
9530 break;
9531 default:
9532 goto illegal_op;
9534 if (rn != 15) {
9535 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9536 goto illegal_op;
9539 tmp = load_reg(s, rm);
9540 shift = (insn >> 4) & 3;
9541 /* ??? In many cases it's not necessary to do a
9542 rotate, a shift is sufficient. */
9543 if (shift != 0)
9544 tcg_gen_rotri_i32(tmp, tmp, shift * 8);
9545 op = (insn >> 20) & 7;
9546 switch (op) {
9547 case 0: gen_sxth(tmp); break;
9548 case 1: gen_uxth(tmp); break;
9549 case 2: gen_sxtb16(tmp); break;
9550 case 3: gen_uxtb16(tmp); break;
9551 case 4: gen_sxtb(tmp); break;
9552 case 5: gen_uxtb(tmp); break;
9553 default:
9554 g_assert_not_reached();
9556 if (rn != 15) {
9557 tmp2 = load_reg(s, rn);
9558 if ((op >> 1) == 1) {
9559 gen_add16(tmp, tmp2);
9560 } else {
9561 tcg_gen_add_i32(tmp, tmp, tmp2);
9562 tcg_temp_free_i32(tmp2);
9565 store_reg(s, rd, tmp);
9566 break;
9567 case 2: /* SIMD add/subtract. */
9568 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9569 goto illegal_op;
9571 op = (insn >> 20) & 7;
9572 shift = (insn >> 4) & 7;
9573 if ((op & 3) == 3 || (shift & 3) == 3)
9574 goto illegal_op;
9575 tmp = load_reg(s, rn);
9576 tmp2 = load_reg(s, rm);
9577 gen_thumb2_parallel_addsub(op, shift, tmp, tmp2);
9578 tcg_temp_free_i32(tmp2);
9579 store_reg(s, rd, tmp);
9580 break;
9581 case 3: /* Other data processing. */
9582 op = ((insn >> 17) & 0x38) | ((insn >> 4) & 7);
9583 if (op < 4) {
9584 /* Saturating add/subtract. */
9585 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9586 goto illegal_op;
9588 tmp = load_reg(s, rn);
9589 tmp2 = load_reg(s, rm);
9590 if (op & 1)
9591 gen_helper_double_saturate(tmp, cpu_env, tmp);
9592 if (op & 2)
9593 gen_helper_sub_saturate(tmp, cpu_env, tmp2, tmp);
9594 else
9595 gen_helper_add_saturate(tmp, cpu_env, tmp, tmp2);
9596 tcg_temp_free_i32(tmp2);
9597 } else {
9598 switch (op) {
9599 case 0x0a: /* rbit */
9600 case 0x08: /* rev */
9601 case 0x09: /* rev16 */
9602 case 0x0b: /* revsh */
9603 case 0x18: /* clz */
9604 break;
9605 case 0x10: /* sel */
9606 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9607 goto illegal_op;
9609 break;
9610 case 0x20: /* crc32/crc32c */
9611 case 0x21:
9612 case 0x22:
9613 case 0x28:
9614 case 0x29:
9615 case 0x2a:
9616 if (!arm_dc_feature(s, ARM_FEATURE_CRC)) {
9617 goto illegal_op;
9619 break;
9620 default:
9621 goto illegal_op;
9623 tmp = load_reg(s, rn);
9624 switch (op) {
9625 case 0x0a: /* rbit */
9626 gen_helper_rbit(tmp, tmp);
9627 break;
9628 case 0x08: /* rev */
9629 tcg_gen_bswap32_i32(tmp, tmp);
9630 break;
9631 case 0x09: /* rev16 */
9632 gen_rev16(tmp);
9633 break;
9634 case 0x0b: /* revsh */
9635 gen_revsh(tmp);
9636 break;
9637 case 0x10: /* sel */
9638 tmp2 = load_reg(s, rm);
9639 tmp3 = tcg_temp_new_i32();
9640 tcg_gen_ld_i32(tmp3, cpu_env, offsetof(CPUARMState, GE));
9641 gen_helper_sel_flags(tmp, tmp3, tmp, tmp2);
9642 tcg_temp_free_i32(tmp3);
9643 tcg_temp_free_i32(tmp2);
9644 break;
9645 case 0x18: /* clz */
9646 gen_helper_clz(tmp, tmp);
9647 break;
9648 case 0x20:
9649 case 0x21:
9650 case 0x22:
9651 case 0x28:
9652 case 0x29:
9653 case 0x2a:
9655 /* crc32/crc32c */
9656 uint32_t sz = op & 0x3;
9657 uint32_t c = op & 0x8;
9659 tmp2 = load_reg(s, rm);
9660 if (sz == 0) {
9661 tcg_gen_andi_i32(tmp2, tmp2, 0xff);
9662 } else if (sz == 1) {
9663 tcg_gen_andi_i32(tmp2, tmp2, 0xffff);
9665 tmp3 = tcg_const_i32(1 << sz);
9666 if (c) {
9667 gen_helper_crc32c(tmp, tmp, tmp2, tmp3);
9668 } else {
9669 gen_helper_crc32(tmp, tmp, tmp2, tmp3);
9671 tcg_temp_free_i32(tmp2);
9672 tcg_temp_free_i32(tmp3);
9673 break;
9675 default:
9676 g_assert_not_reached();
9679 store_reg(s, rd, tmp);
9680 break;
9681 case 4: case 5: /* 32-bit multiply. Sum of absolute differences. */
9682 switch ((insn >> 20) & 7) {
9683 case 0: /* 32 x 32 -> 32 */
9684 case 7: /* Unsigned sum of absolute differences. */
9685 break;
9686 case 1: /* 16 x 16 -> 32 */
9687 case 2: /* Dual multiply add. */
9688 case 3: /* 32 * 16 -> 32msb */
9689 case 4: /* Dual multiply subtract. */
9690 case 5: case 6: /* 32 * 32 -> 32msb (SMMUL, SMMLA, SMMLS) */
9691 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9692 goto illegal_op;
9694 break;
9696 op = (insn >> 4) & 0xf;
9697 tmp = load_reg(s, rn);
9698 tmp2 = load_reg(s, rm);
9699 switch ((insn >> 20) & 7) {
9700 case 0: /* 32 x 32 -> 32 */
9701 tcg_gen_mul_i32(tmp, tmp, tmp2);
9702 tcg_temp_free_i32(tmp2);
9703 if (rs != 15) {
9704 tmp2 = load_reg(s, rs);
9705 if (op)
9706 tcg_gen_sub_i32(tmp, tmp2, tmp);
9707 else
9708 tcg_gen_add_i32(tmp, tmp, tmp2);
9709 tcg_temp_free_i32(tmp2);
9711 break;
9712 case 1: /* 16 x 16 -> 32 */
9713 gen_mulxy(tmp, tmp2, op & 2, op & 1);
9714 tcg_temp_free_i32(tmp2);
9715 if (rs != 15) {
9716 tmp2 = load_reg(s, rs);
9717 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
9718 tcg_temp_free_i32(tmp2);
9720 break;
9721 case 2: /* Dual multiply add. */
9722 case 4: /* Dual multiply subtract. */
9723 if (op)
9724 gen_swap_half(tmp2);
9725 gen_smul_dual(tmp, tmp2);
9726 if (insn & (1 << 22)) {
9727 /* This subtraction cannot overflow. */
9728 tcg_gen_sub_i32(tmp, tmp, tmp2);
9729 } else {
9730 /* This addition cannot overflow 32 bits;
9731 * however it may overflow considered as a signed
9732 * operation, in which case we must set the Q flag.
9734 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
9736 tcg_temp_free_i32(tmp2);
9737 if (rs != 15)
9739 tmp2 = load_reg(s, rs);
9740 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
9741 tcg_temp_free_i32(tmp2);
9743 break;
9744 case 3: /* 32 * 16 -> 32msb */
9745 if (op)
9746 tcg_gen_sari_i32(tmp2, tmp2, 16);
9747 else
9748 gen_sxth(tmp2);
9749 tmp64 = gen_muls_i64_i32(tmp, tmp2);
9750 tcg_gen_shri_i64(tmp64, tmp64, 16);
9751 tmp = tcg_temp_new_i32();
9752 tcg_gen_extrl_i64_i32(tmp, tmp64);
9753 tcg_temp_free_i64(tmp64);
9754 if (rs != 15)
9756 tmp2 = load_reg(s, rs);
9757 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
9758 tcg_temp_free_i32(tmp2);
9760 break;
9761 case 5: case 6: /* 32 * 32 -> 32msb (SMMUL, SMMLA, SMMLS) */
9762 tmp64 = gen_muls_i64_i32(tmp, tmp2);
9763 if (rs != 15) {
9764 tmp = load_reg(s, rs);
9765 if (insn & (1 << 20)) {
9766 tmp64 = gen_addq_msw(tmp64, tmp);
9767 } else {
9768 tmp64 = gen_subq_msw(tmp64, tmp);
9771 if (insn & (1 << 4)) {
9772 tcg_gen_addi_i64(tmp64, tmp64, 0x80000000u);
9774 tcg_gen_shri_i64(tmp64, tmp64, 32);
9775 tmp = tcg_temp_new_i32();
9776 tcg_gen_extrl_i64_i32(tmp, tmp64);
9777 tcg_temp_free_i64(tmp64);
9778 break;
9779 case 7: /* Unsigned sum of absolute differences. */
9780 gen_helper_usad8(tmp, tmp, tmp2);
9781 tcg_temp_free_i32(tmp2);
9782 if (rs != 15) {
9783 tmp2 = load_reg(s, rs);
9784 tcg_gen_add_i32(tmp, tmp, tmp2);
9785 tcg_temp_free_i32(tmp2);
9787 break;
9789 store_reg(s, rd, tmp);
9790 break;
9791 case 6: case 7: /* 64-bit multiply, Divide. */
9792 op = ((insn >> 4) & 0xf) | ((insn >> 16) & 0x70);
9793 tmp = load_reg(s, rn);
9794 tmp2 = load_reg(s, rm);
9795 if ((op & 0x50) == 0x10) {
9796 /* sdiv, udiv */
9797 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DIV)) {
9798 goto illegal_op;
9800 if (op & 0x20)
9801 gen_helper_udiv(tmp, tmp, tmp2);
9802 else
9803 gen_helper_sdiv(tmp, tmp, tmp2);
9804 tcg_temp_free_i32(tmp2);
9805 store_reg(s, rd, tmp);
9806 } else if ((op & 0xe) == 0xc) {
9807 /* Dual multiply accumulate long. */
9808 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9809 tcg_temp_free_i32(tmp);
9810 tcg_temp_free_i32(tmp2);
9811 goto illegal_op;
9813 if (op & 1)
9814 gen_swap_half(tmp2);
9815 gen_smul_dual(tmp, tmp2);
9816 if (op & 0x10) {
9817 tcg_gen_sub_i32(tmp, tmp, tmp2);
9818 } else {
9819 tcg_gen_add_i32(tmp, tmp, tmp2);
9821 tcg_temp_free_i32(tmp2);
9822 /* BUGFIX */
9823 tmp64 = tcg_temp_new_i64();
9824 tcg_gen_ext_i32_i64(tmp64, tmp);
9825 tcg_temp_free_i32(tmp);
9826 gen_addq(s, tmp64, rs, rd);
9827 gen_storeq_reg(s, rs, rd, tmp64);
9828 tcg_temp_free_i64(tmp64);
9829 } else {
9830 if (op & 0x20) {
9831 /* Unsigned 64-bit multiply */
9832 tmp64 = gen_mulu_i64_i32(tmp, tmp2);
9833 } else {
9834 if (op & 8) {
9835 /* smlalxy */
9836 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9837 tcg_temp_free_i32(tmp2);
9838 tcg_temp_free_i32(tmp);
9839 goto illegal_op;
9841 gen_mulxy(tmp, tmp2, op & 2, op & 1);
9842 tcg_temp_free_i32(tmp2);
9843 tmp64 = tcg_temp_new_i64();
9844 tcg_gen_ext_i32_i64(tmp64, tmp);
9845 tcg_temp_free_i32(tmp);
9846 } else {
9847 /* Signed 64-bit multiply */
9848 tmp64 = gen_muls_i64_i32(tmp, tmp2);
9851 if (op & 4) {
9852 /* umaal */
9853 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9854 tcg_temp_free_i64(tmp64);
9855 goto illegal_op;
9857 gen_addq_lo(s, tmp64, rs);
9858 gen_addq_lo(s, tmp64, rd);
9859 } else if (op & 0x40) {
9860 /* 64-bit accumulate. */
9861 gen_addq(s, tmp64, rs, rd);
9863 gen_storeq_reg(s, rs, rd, tmp64);
9864 tcg_temp_free_i64(tmp64);
9866 break;
9868 break;
9869 case 6: case 7: case 14: case 15:
9870 /* Coprocessor. */
9871 if (((insn >> 24) & 3) == 3) {
9872 /* Translate into the equivalent ARM encoding. */
9873 insn = (insn & 0xe2ffffff) | ((insn & (1 << 28)) >> 4) | (1 << 28);
9874 if (disas_neon_data_insn(s, insn)) {
9875 goto illegal_op;
9877 } else if (((insn >> 8) & 0xe) == 10) {
9878 if (disas_vfp_insn(s, insn)) {
9879 goto illegal_op;
9881 } else {
9882 if (insn & (1 << 28))
9883 goto illegal_op;
9884 if (disas_coproc_insn(s, insn)) {
9885 goto illegal_op;
9888 break;
9889 case 8: case 9: case 10: case 11:
9890 if (insn & (1 << 15)) {
9891 /* Branches, misc control. */
9892 if (insn & 0x5000) {
9893 /* Unconditional branch. */
9894 /* signextend(hw1[10:0]) -> offset[:12]. */
9895 offset = ((int32_t)insn << 5) >> 9 & ~(int32_t)0xfff;
9896 /* hw1[10:0] -> offset[11:1]. */
9897 offset |= (insn & 0x7ff) << 1;
9898 /* (~hw2[13, 11] ^ offset[24]) -> offset[23,22]
9899 offset[24:22] already have the same value because of the
9900 sign extension above. */
9901 offset ^= ((~insn) & (1 << 13)) << 10;
9902 offset ^= ((~insn) & (1 << 11)) << 11;
9904 if (insn & (1 << 14)) {
9905 /* Branch and link. */
9906 tcg_gen_movi_i32(cpu_R[14], s->pc | 1);
9909 offset += s->pc;
9910 if (insn & (1 << 12)) {
9911 /* b/bl */
9912 gen_jmp(s, offset);
9913 } else {
9914 /* blx */
9915 offset &= ~(uint32_t)2;
9916 /* thumb2 bx, no need to check */
9917 gen_bx_im(s, offset);
9919 } else if (((insn >> 23) & 7) == 7) {
9920 /* Misc control */
9921 if (insn & (1 << 13))
9922 goto illegal_op;
9924 if (insn & (1 << 26)) {
9925 if (!(insn & (1 << 20))) {
9926 /* Hypervisor call (v7) */
9927 int imm16 = extract32(insn, 16, 4) << 12
9928 | extract32(insn, 0, 12);
9929 ARCH(7);
9930 if (IS_USER(s)) {
9931 goto illegal_op;
9933 gen_hvc(s, imm16);
9934 } else {
9935 /* Secure monitor call (v6+) */
9936 ARCH(6K);
9937 if (IS_USER(s)) {
9938 goto illegal_op;
9940 gen_smc(s);
9942 } else {
9943 op = (insn >> 20) & 7;
9944 switch (op) {
9945 case 0: /* msr cpsr. */
9946 if (arm_dc_feature(s, ARM_FEATURE_M)) {
9947 tmp = load_reg(s, rn);
9948 addr = tcg_const_i32(insn & 0xff);
9949 gen_helper_v7m_msr(cpu_env, addr, tmp);
9950 tcg_temp_free_i32(addr);
9951 tcg_temp_free_i32(tmp);
9952 gen_lookup_tb(s);
9953 break;
9955 /* fall through */
9956 case 1: /* msr spsr. */
9957 if (arm_dc_feature(s, ARM_FEATURE_M)) {
9958 goto illegal_op;
9960 tmp = load_reg(s, rn);
9961 if (gen_set_psr(s,
9962 msr_mask(s, (insn >> 8) & 0xf, op == 1),
9963 op == 1, tmp))
9964 goto illegal_op;
9965 break;
9966 case 2: /* cps, nop-hint. */
9967 if (((insn >> 8) & 7) == 0) {
9968 gen_nop_hint(s, insn & 0xff);
9970 /* Implemented as NOP in user mode. */
9971 if (IS_USER(s))
9972 break;
9973 offset = 0;
9974 imm = 0;
9975 if (insn & (1 << 10)) {
9976 if (insn & (1 << 7))
9977 offset |= CPSR_A;
9978 if (insn & (1 << 6))
9979 offset |= CPSR_I;
9980 if (insn & (1 << 5))
9981 offset |= CPSR_F;
9982 if (insn & (1 << 9))
9983 imm = CPSR_A | CPSR_I | CPSR_F;
9985 if (insn & (1 << 8)) {
9986 offset |= 0x1f;
9987 imm |= (insn & 0x1f);
9989 if (offset) {
9990 gen_set_psr_im(s, offset, 0, imm);
9992 break;
9993 case 3: /* Special control operations. */
9994 ARCH(7);
9995 op = (insn >> 4) & 0xf;
9996 switch (op) {
9997 case 2: /* clrex */
9998 gen_clrex(s);
9999 break;
10000 case 4: /* dsb */
10001 case 5: /* dmb */
10002 case 6: /* isb */
10003 /* These execute as NOPs. */
10004 break;
10005 default:
10006 goto illegal_op;
10008 break;
10009 case 4: /* bxj */
10010 /* Trivial implementation equivalent to bx. */
10011 tmp = load_reg(s, rn);
10012 gen_bx(s, tmp);
10013 break;
10014 case 5: /* Exception return. */
10015 if (IS_USER(s)) {
10016 goto illegal_op;
10018 if (rn != 14 || rd != 15) {
10019 goto illegal_op;
10021 tmp = load_reg(s, rn);
10022 tcg_gen_subi_i32(tmp, tmp, insn & 0xff);
10023 gen_exception_return(s, tmp);
10024 break;
10025 case 6: /* mrs cpsr. */
10026 tmp = tcg_temp_new_i32();
10027 if (arm_dc_feature(s, ARM_FEATURE_M)) {
10028 addr = tcg_const_i32(insn & 0xff);
10029 gen_helper_v7m_mrs(tmp, cpu_env, addr);
10030 tcg_temp_free_i32(addr);
10031 } else {
10032 gen_helper_cpsr_read(tmp, cpu_env);
10034 store_reg(s, rd, tmp);
10035 break;
10036 case 7: /* mrs spsr. */
10037 /* Not accessible in user mode. */
10038 if (IS_USER(s) || arm_dc_feature(s, ARM_FEATURE_M)) {
10039 goto illegal_op;
10041 tmp = load_cpu_field(spsr);
10042 store_reg(s, rd, tmp);
10043 break;
10046 } else {
10047 /* Conditional branch. */
10048 op = (insn >> 22) & 0xf;
10049 /* Generate a conditional jump to next instruction. */
10050 s->condlabel = gen_new_label();
10051 arm_gen_test_cc(op ^ 1, s->condlabel);
10052 s->condjmp = 1;
10054 /* offset[11:1] = insn[10:0] */
10055 offset = (insn & 0x7ff) << 1;
10056 /* offset[17:12] = insn[21:16]. */
10057 offset |= (insn & 0x003f0000) >> 4;
10058 /* offset[31:20] = insn[26]. */
10059 offset |= ((int32_t)((insn << 5) & 0x80000000)) >> 11;
10060 /* offset[18] = insn[13]. */
10061 offset |= (insn & (1 << 13)) << 5;
10062 /* offset[19] = insn[11]. */
10063 offset |= (insn & (1 << 11)) << 8;
10065 /* jump to the offset */
10066 gen_jmp(s, s->pc + offset);
10068 } else {
10069 /* Data processing immediate. */
10070 if (insn & (1 << 25)) {
10071 if (insn & (1 << 24)) {
10072 if (insn & (1 << 20))
10073 goto illegal_op;
10074 /* Bitfield/Saturate. */
10075 op = (insn >> 21) & 7;
10076 imm = insn & 0x1f;
10077 shift = ((insn >> 6) & 3) | ((insn >> 10) & 0x1c);
10078 if (rn == 15) {
10079 tmp = tcg_temp_new_i32();
10080 tcg_gen_movi_i32(tmp, 0);
10081 } else {
10082 tmp = load_reg(s, rn);
10084 switch (op) {
10085 case 2: /* Signed bitfield extract. */
10086 imm++;
10087 if (shift + imm > 32)
10088 goto illegal_op;
10089 if (imm < 32)
10090 gen_sbfx(tmp, shift, imm);
10091 break;
10092 case 6: /* Unsigned bitfield extract. */
10093 imm++;
10094 if (shift + imm > 32)
10095 goto illegal_op;
10096 if (imm < 32)
10097 gen_ubfx(tmp, shift, (1u << imm) - 1);
10098 break;
10099 case 3: /* Bitfield insert/clear. */
10100 if (imm < shift)
10101 goto illegal_op;
10102 imm = imm + 1 - shift;
10103 if (imm != 32) {
10104 tmp2 = load_reg(s, rd);
10105 tcg_gen_deposit_i32(tmp, tmp2, tmp, shift, imm);
10106 tcg_temp_free_i32(tmp2);
10108 break;
10109 case 7:
10110 goto illegal_op;
10111 default: /* Saturate. */
10112 if (shift) {
10113 if (op & 1)
10114 tcg_gen_sari_i32(tmp, tmp, shift);
10115 else
10116 tcg_gen_shli_i32(tmp, tmp, shift);
10118 tmp2 = tcg_const_i32(imm);
10119 if (op & 4) {
10120 /* Unsigned. */
10121 if ((op & 1) && shift == 0) {
10122 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10123 tcg_temp_free_i32(tmp);
10124 tcg_temp_free_i32(tmp2);
10125 goto illegal_op;
10127 gen_helper_usat16(tmp, cpu_env, tmp, tmp2);
10128 } else {
10129 gen_helper_usat(tmp, cpu_env, tmp, tmp2);
10131 } else {
10132 /* Signed. */
10133 if ((op & 1) && shift == 0) {
10134 if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
10135 tcg_temp_free_i32(tmp);
10136 tcg_temp_free_i32(tmp2);
10137 goto illegal_op;
10139 gen_helper_ssat16(tmp, cpu_env, tmp, tmp2);
10140 } else {
10141 gen_helper_ssat(tmp, cpu_env, tmp, tmp2);
10144 tcg_temp_free_i32(tmp2);
10145 break;
10147 store_reg(s, rd, tmp);
10148 } else {
10149 imm = ((insn & 0x04000000) >> 15)
10150 | ((insn & 0x7000) >> 4) | (insn & 0xff);
10151 if (insn & (1 << 22)) {
10152 /* 16-bit immediate. */
10153 imm |= (insn >> 4) & 0xf000;
10154 if (insn & (1 << 23)) {
10155 /* movt */
10156 tmp = load_reg(s, rd);
10157 tcg_gen_ext16u_i32(tmp, tmp);
10158 tcg_gen_ori_i32(tmp, tmp, imm << 16);
10159 } else {
10160 /* movw */
10161 tmp = tcg_temp_new_i32();
10162 tcg_gen_movi_i32(tmp, imm);
10164 } else {
10165 /* Add/sub 12-bit immediate. */
10166 if (rn == 15) {
10167 offset = s->pc & ~(uint32_t)3;
10168 if (insn & (1 << 23))
10169 offset -= imm;
10170 else
10171 offset += imm;
10172 tmp = tcg_temp_new_i32();
10173 tcg_gen_movi_i32(tmp, offset);
10174 } else {
10175 tmp = load_reg(s, rn);
10176 if (insn & (1 << 23))
10177 tcg_gen_subi_i32(tmp, tmp, imm);
10178 else
10179 tcg_gen_addi_i32(tmp, tmp, imm);
10182 store_reg(s, rd, tmp);
10184 } else {
10185 int shifter_out = 0;
10186 /* modified 12-bit immediate. */
10187 shift = ((insn & 0x04000000) >> 23) | ((insn & 0x7000) >> 12);
10188 imm = (insn & 0xff);
10189 switch (shift) {
10190 case 0: /* XY */
10191 /* Nothing to do. */
10192 break;
10193 case 1: /* 00XY00XY */
10194 imm |= imm << 16;
10195 break;
10196 case 2: /* XY00XY00 */
10197 imm |= imm << 16;
10198 imm <<= 8;
10199 break;
10200 case 3: /* XYXYXYXY */
10201 imm |= imm << 16;
10202 imm |= imm << 8;
10203 break;
10204 default: /* Rotated constant. */
10205 shift = (shift << 1) | (imm >> 7);
10206 imm |= 0x80;
10207 imm = imm << (32 - shift);
10208 shifter_out = 1;
10209 break;
10211 tmp2 = tcg_temp_new_i32();
10212 tcg_gen_movi_i32(tmp2, imm);
10213 rn = (insn >> 16) & 0xf;
10214 if (rn == 15) {
10215 tmp = tcg_temp_new_i32();
10216 tcg_gen_movi_i32(tmp, 0);
10217 } else {
10218 tmp = load_reg(s, rn);
10220 op = (insn >> 21) & 0xf;
10221 if (gen_thumb2_data_op(s, op, (insn & (1 << 20)) != 0,
10222 shifter_out, tmp, tmp2))
10223 goto illegal_op;
10224 tcg_temp_free_i32(tmp2);
10225 rd = (insn >> 8) & 0xf;
10226 if (rd != 15) {
10227 store_reg(s, rd, tmp);
10228 } else {
10229 tcg_temp_free_i32(tmp);
10233 break;
10234 case 12: /* Load/store single data item. */
10236 int postinc = 0;
10237 int writeback = 0;
10238 int memidx;
10239 if ((insn & 0x01100000) == 0x01000000) {
10240 if (disas_neon_ls_insn(s, insn)) {
10241 goto illegal_op;
10243 break;
10245 op = ((insn >> 21) & 3) | ((insn >> 22) & 4);
10246 if (rs == 15) {
10247 if (!(insn & (1 << 20))) {
10248 goto illegal_op;
10250 if (op != 2) {
10251 /* Byte or halfword load space with dest == r15 : memory hints.
10252 * Catch them early so we don't emit pointless addressing code.
10253 * This space is a mix of:
10254 * PLD/PLDW/PLI, which we implement as NOPs (note that unlike
10255 * the ARM encodings, PLDW space doesn't UNDEF for non-v7MP
10256 * cores)
10257 * unallocated hints, which must be treated as NOPs
10258 * UNPREDICTABLE space, which we NOP or UNDEF depending on
10259 * which is easiest for the decoding logic
10260 * Some space which must UNDEF
10262 int op1 = (insn >> 23) & 3;
10263 int op2 = (insn >> 6) & 0x3f;
10264 if (op & 2) {
10265 goto illegal_op;
10267 if (rn == 15) {
10268 /* UNPREDICTABLE, unallocated hint or
10269 * PLD/PLDW/PLI (literal)
10271 return 0;
10273 if (op1 & 1) {
10274 return 0; /* PLD/PLDW/PLI or unallocated hint */
10276 if ((op2 == 0) || ((op2 & 0x3c) == 0x30)) {
10277 return 0; /* PLD/PLDW/PLI or unallocated hint */
10279 /* UNDEF space, or an UNPREDICTABLE */
10280 return 1;
10283 memidx = get_mem_index(s);
10284 if (rn == 15) {
10285 addr = tcg_temp_new_i32();
10286 /* PC relative. */
10287 /* s->pc has already been incremented by 4. */
10288 imm = s->pc & 0xfffffffc;
10289 if (insn & (1 << 23))
10290 imm += insn & 0xfff;
10291 else
10292 imm -= insn & 0xfff;
10293 tcg_gen_movi_i32(addr, imm);
10294 } else {
10295 addr = load_reg(s, rn);
10296 if (insn & (1 << 23)) {
10297 /* Positive offset. */
10298 imm = insn & 0xfff;
10299 tcg_gen_addi_i32(addr, addr, imm);
10300 } else {
10301 imm = insn & 0xff;
10302 switch ((insn >> 8) & 0xf) {
10303 case 0x0: /* Shifted Register. */
10304 shift = (insn >> 4) & 0xf;
10305 if (shift > 3) {
10306 tcg_temp_free_i32(addr);
10307 goto illegal_op;
10309 tmp = load_reg(s, rm);
10310 if (shift)
10311 tcg_gen_shli_i32(tmp, tmp, shift);
10312 tcg_gen_add_i32(addr, addr, tmp);
10313 tcg_temp_free_i32(tmp);
10314 break;
10315 case 0xc: /* Negative offset. */
10316 tcg_gen_addi_i32(addr, addr, -imm);
10317 break;
10318 case 0xe: /* User privilege. */
10319 tcg_gen_addi_i32(addr, addr, imm);
10320 memidx = get_a32_user_mem_index(s);
10321 break;
10322 case 0x9: /* Post-decrement. */
10323 imm = -imm;
10324 /* Fall through. */
10325 case 0xb: /* Post-increment. */
10326 postinc = 1;
10327 writeback = 1;
10328 break;
10329 case 0xd: /* Pre-decrement. */
10330 imm = -imm;
10331 /* Fall through. */
10332 case 0xf: /* Pre-increment. */
10333 tcg_gen_addi_i32(addr, addr, imm);
10334 writeback = 1;
10335 break;
10336 default:
10337 tcg_temp_free_i32(addr);
10338 goto illegal_op;
10342 if (insn & (1 << 20)) {
10343 /* Load. */
10344 tmp = tcg_temp_new_i32();
10345 switch (op) {
10346 case 0:
10347 gen_aa32_ld8u(tmp, addr, memidx);
10348 break;
10349 case 4:
10350 gen_aa32_ld8s(tmp, addr, memidx);
10351 break;
10352 case 1:
10353 gen_aa32_ld16u(tmp, addr, memidx);
10354 break;
10355 case 5:
10356 gen_aa32_ld16s(tmp, addr, memidx);
10357 break;
10358 case 2:
10359 gen_aa32_ld32u(tmp, addr, memidx);
10360 break;
10361 default:
10362 tcg_temp_free_i32(tmp);
10363 tcg_temp_free_i32(addr);
10364 goto illegal_op;
10366 if (rs == 15) {
10367 gen_bx(s, tmp);
10368 } else {
10369 store_reg(s, rs, tmp);
10371 } else {
10372 /* Store. */
10373 tmp = load_reg(s, rs);
10374 switch (op) {
10375 case 0:
10376 gen_aa32_st8(tmp, addr, memidx);
10377 break;
10378 case 1:
10379 gen_aa32_st16(tmp, addr, memidx);
10380 break;
10381 case 2:
10382 gen_aa32_st32(tmp, addr, memidx);
10383 break;
10384 default:
10385 tcg_temp_free_i32(tmp);
10386 tcg_temp_free_i32(addr);
10387 goto illegal_op;
10389 tcg_temp_free_i32(tmp);
10391 if (postinc)
10392 tcg_gen_addi_i32(addr, addr, imm);
10393 if (writeback) {
10394 store_reg(s, rn, addr);
10395 } else {
10396 tcg_temp_free_i32(addr);
10399 break;
10400 default:
10401 goto illegal_op;
10403 return 0;
10404 illegal_op:
10405 return 1;
10408 static void disas_thumb_insn(CPUARMState *env, DisasContext *s)
10410 uint32_t val, insn, op, rm, rn, rd, shift, cond;
10411 int32_t offset;
10412 int i;
10413 TCGv_i32 tmp;
10414 TCGv_i32 tmp2;
10415 TCGv_i32 addr;
10417 if (s->condexec_mask) {
10418 cond = s->condexec_cond;
10419 if (cond != 0x0e) { /* Skip conditional when condition is AL. */
10420 s->condlabel = gen_new_label();
10421 arm_gen_test_cc(cond ^ 1, s->condlabel);
10422 s->condjmp = 1;
10426 insn = arm_lduw_code(env, s->pc, s->bswap_code);
10427 s->pc += 2;
10429 switch (insn >> 12) {
10430 case 0: case 1:
10432 rd = insn & 7;
10433 op = (insn >> 11) & 3;
10434 if (op == 3) {
10435 /* add/subtract */
10436 rn = (insn >> 3) & 7;
10437 tmp = load_reg(s, rn);
10438 if (insn & (1 << 10)) {
10439 /* immediate */
10440 tmp2 = tcg_temp_new_i32();
10441 tcg_gen_movi_i32(tmp2, (insn >> 6) & 7);
10442 } else {
10443 /* reg */
10444 rm = (insn >> 6) & 7;
10445 tmp2 = load_reg(s, rm);
10447 if (insn & (1 << 9)) {
10448 if (s->condexec_mask)
10449 tcg_gen_sub_i32(tmp, tmp, tmp2);
10450 else
10451 gen_sub_CC(tmp, tmp, tmp2);
10452 } else {
10453 if (s->condexec_mask)
10454 tcg_gen_add_i32(tmp, tmp, tmp2);
10455 else
10456 gen_add_CC(tmp, tmp, tmp2);
10458 tcg_temp_free_i32(tmp2);
10459 store_reg(s, rd, tmp);
10460 } else {
10461 /* shift immediate */
10462 rm = (insn >> 3) & 7;
10463 shift = (insn >> 6) & 0x1f;
10464 tmp = load_reg(s, rm);
10465 gen_arm_shift_im(tmp, op, shift, s->condexec_mask == 0);
10466 if (!s->condexec_mask)
10467 gen_logic_CC(tmp);
10468 store_reg(s, rd, tmp);
10470 break;
10471 case 2: case 3:
10472 /* arithmetic large immediate */
10473 op = (insn >> 11) & 3;
10474 rd = (insn >> 8) & 0x7;
10475 if (op == 0) { /* mov */
10476 tmp = tcg_temp_new_i32();
10477 tcg_gen_movi_i32(tmp, insn & 0xff);
10478 if (!s->condexec_mask)
10479 gen_logic_CC(tmp);
10480 store_reg(s, rd, tmp);
10481 } else {
10482 tmp = load_reg(s, rd);
10483 tmp2 = tcg_temp_new_i32();
10484 tcg_gen_movi_i32(tmp2, insn & 0xff);
10485 switch (op) {
10486 case 1: /* cmp */
10487 gen_sub_CC(tmp, tmp, tmp2);
10488 tcg_temp_free_i32(tmp);
10489 tcg_temp_free_i32(tmp2);
10490 break;
10491 case 2: /* add */
10492 if (s->condexec_mask)
10493 tcg_gen_add_i32(tmp, tmp, tmp2);
10494 else
10495 gen_add_CC(tmp, tmp, tmp2);
10496 tcg_temp_free_i32(tmp2);
10497 store_reg(s, rd, tmp);
10498 break;
10499 case 3: /* sub */
10500 if (s->condexec_mask)
10501 tcg_gen_sub_i32(tmp, tmp, tmp2);
10502 else
10503 gen_sub_CC(tmp, tmp, tmp2);
10504 tcg_temp_free_i32(tmp2);
10505 store_reg(s, rd, tmp);
10506 break;
10509 break;
10510 case 4:
10511 if (insn & (1 << 11)) {
10512 rd = (insn >> 8) & 7;
10513 /* load pc-relative. Bit 1 of PC is ignored. */
10514 val = s->pc + 2 + ((insn & 0xff) * 4);
10515 val &= ~(uint32_t)2;
10516 addr = tcg_temp_new_i32();
10517 tcg_gen_movi_i32(addr, val);
10518 tmp = tcg_temp_new_i32();
10519 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
10520 tcg_temp_free_i32(addr);
10521 store_reg(s, rd, tmp);
10522 break;
10524 if (insn & (1 << 10)) {
10525 /* data processing extended or blx */
10526 rd = (insn & 7) | ((insn >> 4) & 8);
10527 rm = (insn >> 3) & 0xf;
10528 op = (insn >> 8) & 3;
10529 switch (op) {
10530 case 0: /* add */
10531 tmp = load_reg(s, rd);
10532 tmp2 = load_reg(s, rm);
10533 tcg_gen_add_i32(tmp, tmp, tmp2);
10534 tcg_temp_free_i32(tmp2);
10535 store_reg(s, rd, tmp);
10536 break;
10537 case 1: /* cmp */
10538 tmp = load_reg(s, rd);
10539 tmp2 = load_reg(s, rm);
10540 gen_sub_CC(tmp, tmp, tmp2);
10541 tcg_temp_free_i32(tmp2);
10542 tcg_temp_free_i32(tmp);
10543 break;
10544 case 2: /* mov/cpy */
10545 tmp = load_reg(s, rm);
10546 store_reg(s, rd, tmp);
10547 break;
10548 case 3:/* branch [and link] exchange thumb register */
10549 tmp = load_reg(s, rm);
10550 if (insn & (1 << 7)) {
10551 ARCH(5);
10552 val = (uint32_t)s->pc | 1;
10553 tmp2 = tcg_temp_new_i32();
10554 tcg_gen_movi_i32(tmp2, val);
10555 store_reg(s, 14, tmp2);
10557 /* already thumb, no need to check */
10558 gen_bx(s, tmp);
10559 break;
10561 break;
10564 /* data processing register */
10565 rd = insn & 7;
10566 rm = (insn >> 3) & 7;
10567 op = (insn >> 6) & 0xf;
10568 if (op == 2 || op == 3 || op == 4 || op == 7) {
10569 /* the shift/rotate ops want the operands backwards */
10570 val = rm;
10571 rm = rd;
10572 rd = val;
10573 val = 1;
10574 } else {
10575 val = 0;
10578 if (op == 9) { /* neg */
10579 tmp = tcg_temp_new_i32();
10580 tcg_gen_movi_i32(tmp, 0);
10581 } else if (op != 0xf) { /* mvn doesn't read its first operand */
10582 tmp = load_reg(s, rd);
10583 } else {
10584 TCGV_UNUSED_I32(tmp);
10587 tmp2 = load_reg(s, rm);
10588 switch (op) {
10589 case 0x0: /* and */
10590 tcg_gen_and_i32(tmp, tmp, tmp2);
10591 if (!s->condexec_mask)
10592 gen_logic_CC(tmp);
10593 break;
10594 case 0x1: /* eor */
10595 tcg_gen_xor_i32(tmp, tmp, tmp2);
10596 if (!s->condexec_mask)
10597 gen_logic_CC(tmp);
10598 break;
10599 case 0x2: /* lsl */
10600 if (s->condexec_mask) {
10601 gen_shl(tmp2, tmp2, tmp);
10602 } else {
10603 gen_helper_shl_cc(tmp2, cpu_env, tmp2, tmp);
10604 gen_logic_CC(tmp2);
10606 break;
10607 case 0x3: /* lsr */
10608 if (s->condexec_mask) {
10609 gen_shr(tmp2, tmp2, tmp);
10610 } else {
10611 gen_helper_shr_cc(tmp2, cpu_env, tmp2, tmp);
10612 gen_logic_CC(tmp2);
10614 break;
10615 case 0x4: /* asr */
10616 if (s->condexec_mask) {
10617 gen_sar(tmp2, tmp2, tmp);
10618 } else {
10619 gen_helper_sar_cc(tmp2, cpu_env, tmp2, tmp);
10620 gen_logic_CC(tmp2);
10622 break;
10623 case 0x5: /* adc */
10624 if (s->condexec_mask) {
10625 gen_adc(tmp, tmp2);
10626 } else {
10627 gen_adc_CC(tmp, tmp, tmp2);
10629 break;
10630 case 0x6: /* sbc */
10631 if (s->condexec_mask) {
10632 gen_sub_carry(tmp, tmp, tmp2);
10633 } else {
10634 gen_sbc_CC(tmp, tmp, tmp2);
10636 break;
10637 case 0x7: /* ror */
10638 if (s->condexec_mask) {
10639 tcg_gen_andi_i32(tmp, tmp, 0x1f);
10640 tcg_gen_rotr_i32(tmp2, tmp2, tmp);
10641 } else {
10642 gen_helper_ror_cc(tmp2, cpu_env, tmp2, tmp);
10643 gen_logic_CC(tmp2);
10645 break;
10646 case 0x8: /* tst */
10647 tcg_gen_and_i32(tmp, tmp, tmp2);
10648 gen_logic_CC(tmp);
10649 rd = 16;
10650 break;
10651 case 0x9: /* neg */
10652 if (s->condexec_mask)
10653 tcg_gen_neg_i32(tmp, tmp2);
10654 else
10655 gen_sub_CC(tmp, tmp, tmp2);
10656 break;
10657 case 0xa: /* cmp */
10658 gen_sub_CC(tmp, tmp, tmp2);
10659 rd = 16;
10660 break;
10661 case 0xb: /* cmn */
10662 gen_add_CC(tmp, tmp, tmp2);
10663 rd = 16;
10664 break;
10665 case 0xc: /* orr */
10666 tcg_gen_or_i32(tmp, tmp, tmp2);
10667 if (!s->condexec_mask)
10668 gen_logic_CC(tmp);
10669 break;
10670 case 0xd: /* mul */
10671 tcg_gen_mul_i32(tmp, tmp, tmp2);
10672 if (!s->condexec_mask)
10673 gen_logic_CC(tmp);
10674 break;
10675 case 0xe: /* bic */
10676 tcg_gen_andc_i32(tmp, tmp, tmp2);
10677 if (!s->condexec_mask)
10678 gen_logic_CC(tmp);
10679 break;
10680 case 0xf: /* mvn */
10681 tcg_gen_not_i32(tmp2, tmp2);
10682 if (!s->condexec_mask)
10683 gen_logic_CC(tmp2);
10684 val = 1;
10685 rm = rd;
10686 break;
10688 if (rd != 16) {
10689 if (val) {
10690 store_reg(s, rm, tmp2);
10691 if (op != 0xf)
10692 tcg_temp_free_i32(tmp);
10693 } else {
10694 store_reg(s, rd, tmp);
10695 tcg_temp_free_i32(tmp2);
10697 } else {
10698 tcg_temp_free_i32(tmp);
10699 tcg_temp_free_i32(tmp2);
10701 break;
10703 case 5:
10704 /* load/store register offset. */
10705 rd = insn & 7;
10706 rn = (insn >> 3) & 7;
10707 rm = (insn >> 6) & 7;
10708 op = (insn >> 9) & 7;
10709 addr = load_reg(s, rn);
10710 tmp = load_reg(s, rm);
10711 tcg_gen_add_i32(addr, addr, tmp);
10712 tcg_temp_free_i32(tmp);
10714 if (op < 3) { /* store */
10715 tmp = load_reg(s, rd);
10716 } else {
10717 tmp = tcg_temp_new_i32();
10720 switch (op) {
10721 case 0: /* str */
10722 gen_aa32_st32(tmp, addr, get_mem_index(s));
10723 break;
10724 case 1: /* strh */
10725 gen_aa32_st16(tmp, addr, get_mem_index(s));
10726 break;
10727 case 2: /* strb */
10728 gen_aa32_st8(tmp, addr, get_mem_index(s));
10729 break;
10730 case 3: /* ldrsb */
10731 gen_aa32_ld8s(tmp, addr, get_mem_index(s));
10732 break;
10733 case 4: /* ldr */
10734 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
10735 break;
10736 case 5: /* ldrh */
10737 gen_aa32_ld16u(tmp, addr, get_mem_index(s));
10738 break;
10739 case 6: /* ldrb */
10740 gen_aa32_ld8u(tmp, addr, get_mem_index(s));
10741 break;
10742 case 7: /* ldrsh */
10743 gen_aa32_ld16s(tmp, addr, get_mem_index(s));
10744 break;
10746 if (op >= 3) { /* load */
10747 store_reg(s, rd, tmp);
10748 } else {
10749 tcg_temp_free_i32(tmp);
10751 tcg_temp_free_i32(addr);
10752 break;
10754 case 6:
10755 /* load/store word immediate offset */
10756 rd = insn & 7;
10757 rn = (insn >> 3) & 7;
10758 addr = load_reg(s, rn);
10759 val = (insn >> 4) & 0x7c;
10760 tcg_gen_addi_i32(addr, addr, val);
10762 if (insn & (1 << 11)) {
10763 /* load */
10764 tmp = tcg_temp_new_i32();
10765 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
10766 store_reg(s, rd, tmp);
10767 } else {
10768 /* store */
10769 tmp = load_reg(s, rd);
10770 gen_aa32_st32(tmp, addr, get_mem_index(s));
10771 tcg_temp_free_i32(tmp);
10773 tcg_temp_free_i32(addr);
10774 break;
10776 case 7:
10777 /* load/store byte immediate offset */
10778 rd = insn & 7;
10779 rn = (insn >> 3) & 7;
10780 addr = load_reg(s, rn);
10781 val = (insn >> 6) & 0x1f;
10782 tcg_gen_addi_i32(addr, addr, val);
10784 if (insn & (1 << 11)) {
10785 /* load */
10786 tmp = tcg_temp_new_i32();
10787 gen_aa32_ld8u(tmp, addr, get_mem_index(s));
10788 store_reg(s, rd, tmp);
10789 } else {
10790 /* store */
10791 tmp = load_reg(s, rd);
10792 gen_aa32_st8(tmp, addr, get_mem_index(s));
10793 tcg_temp_free_i32(tmp);
10795 tcg_temp_free_i32(addr);
10796 break;
10798 case 8:
10799 /* load/store halfword immediate offset */
10800 rd = insn & 7;
10801 rn = (insn >> 3) & 7;
10802 addr = load_reg(s, rn);
10803 val = (insn >> 5) & 0x3e;
10804 tcg_gen_addi_i32(addr, addr, val);
10806 if (insn & (1 << 11)) {
10807 /* load */
10808 tmp = tcg_temp_new_i32();
10809 gen_aa32_ld16u(tmp, addr, get_mem_index(s));
10810 store_reg(s, rd, tmp);
10811 } else {
10812 /* store */
10813 tmp = load_reg(s, rd);
10814 gen_aa32_st16(tmp, addr, get_mem_index(s));
10815 tcg_temp_free_i32(tmp);
10817 tcg_temp_free_i32(addr);
10818 break;
10820 case 9:
10821 /* load/store from stack */
10822 rd = (insn >> 8) & 7;
10823 addr = load_reg(s, 13);
10824 val = (insn & 0xff) * 4;
10825 tcg_gen_addi_i32(addr, addr, val);
10827 if (insn & (1 << 11)) {
10828 /* load */
10829 tmp = tcg_temp_new_i32();
10830 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
10831 store_reg(s, rd, tmp);
10832 } else {
10833 /* store */
10834 tmp = load_reg(s, rd);
10835 gen_aa32_st32(tmp, addr, get_mem_index(s));
10836 tcg_temp_free_i32(tmp);
10838 tcg_temp_free_i32(addr);
10839 break;
10841 case 10:
10842 /* add to high reg */
10843 rd = (insn >> 8) & 7;
10844 if (insn & (1 << 11)) {
10845 /* SP */
10846 tmp = load_reg(s, 13);
10847 } else {
10848 /* PC. bit 1 is ignored. */
10849 tmp = tcg_temp_new_i32();
10850 tcg_gen_movi_i32(tmp, (s->pc + 2) & ~(uint32_t)2);
10852 val = (insn & 0xff) * 4;
10853 tcg_gen_addi_i32(tmp, tmp, val);
10854 store_reg(s, rd, tmp);
10855 break;
10857 case 11:
10858 /* misc */
10859 op = (insn >> 8) & 0xf;
10860 switch (op) {
10861 case 0:
10862 /* adjust stack pointer */
10863 tmp = load_reg(s, 13);
10864 val = (insn & 0x7f) * 4;
10865 if (insn & (1 << 7))
10866 val = -(int32_t)val;
10867 tcg_gen_addi_i32(tmp, tmp, val);
10868 store_reg(s, 13, tmp);
10869 break;
10871 case 2: /* sign/zero extend. */
10872 ARCH(6);
10873 rd = insn & 7;
10874 rm = (insn >> 3) & 7;
10875 tmp = load_reg(s, rm);
10876 switch ((insn >> 6) & 3) {
10877 case 0: gen_sxth(tmp); break;
10878 case 1: gen_sxtb(tmp); break;
10879 case 2: gen_uxth(tmp); break;
10880 case 3: gen_uxtb(tmp); break;
10882 store_reg(s, rd, tmp);
10883 break;
10884 case 4: case 5: case 0xc: case 0xd:
10885 /* push/pop */
10886 addr = load_reg(s, 13);
10887 if (insn & (1 << 8))
10888 offset = 4;
10889 else
10890 offset = 0;
10891 for (i = 0; i < 8; i++) {
10892 if (insn & (1 << i))
10893 offset += 4;
10895 if ((insn & (1 << 11)) == 0) {
10896 tcg_gen_addi_i32(addr, addr, -offset);
10898 for (i = 0; i < 8; i++) {
10899 if (insn & (1 << i)) {
10900 if (insn & (1 << 11)) {
10901 /* pop */
10902 tmp = tcg_temp_new_i32();
10903 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
10904 store_reg(s, i, tmp);
10905 } else {
10906 /* push */
10907 tmp = load_reg(s, i);
10908 gen_aa32_st32(tmp, addr, get_mem_index(s));
10909 tcg_temp_free_i32(tmp);
10911 /* advance to the next address. */
10912 tcg_gen_addi_i32(addr, addr, 4);
10915 TCGV_UNUSED_I32(tmp);
10916 if (insn & (1 << 8)) {
10917 if (insn & (1 << 11)) {
10918 /* pop pc */
10919 tmp = tcg_temp_new_i32();
10920 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
10921 /* don't set the pc until the rest of the instruction
10922 has completed */
10923 } else {
10924 /* push lr */
10925 tmp = load_reg(s, 14);
10926 gen_aa32_st32(tmp, addr, get_mem_index(s));
10927 tcg_temp_free_i32(tmp);
10929 tcg_gen_addi_i32(addr, addr, 4);
10931 if ((insn & (1 << 11)) == 0) {
10932 tcg_gen_addi_i32(addr, addr, -offset);
10934 /* write back the new stack pointer */
10935 store_reg(s, 13, addr);
10936 /* set the new PC value */
10937 if ((insn & 0x0900) == 0x0900) {
10938 store_reg_from_load(s, 15, tmp);
10940 break;
10942 case 1: case 3: case 9: case 11: /* czb */
10943 rm = insn & 7;
10944 tmp = load_reg(s, rm);
10945 s->condlabel = gen_new_label();
10946 s->condjmp = 1;
10947 if (insn & (1 << 11))
10948 tcg_gen_brcondi_i32(TCG_COND_EQ, tmp, 0, s->condlabel);
10949 else
10950 tcg_gen_brcondi_i32(TCG_COND_NE, tmp, 0, s->condlabel);
10951 tcg_temp_free_i32(tmp);
10952 offset = ((insn & 0xf8) >> 2) | (insn & 0x200) >> 3;
10953 val = (uint32_t)s->pc + 2;
10954 val += offset;
10955 gen_jmp(s, val);
10956 break;
10958 case 15: /* IT, nop-hint. */
10959 if ((insn & 0xf) == 0) {
10960 gen_nop_hint(s, (insn >> 4) & 0xf);
10961 break;
10963 /* If Then. */
10964 s->condexec_cond = (insn >> 4) & 0xe;
10965 s->condexec_mask = insn & 0x1f;
10966 /* No actual code generated for this insn, just setup state. */
10967 break;
10969 case 0xe: /* bkpt */
10971 int imm8 = extract32(insn, 0, 8);
10972 ARCH(5);
10973 gen_exception_insn(s, 2, EXCP_BKPT, syn_aa32_bkpt(imm8, true),
10974 default_exception_el(s));
10975 break;
10978 case 0xa: /* rev */
10979 ARCH(6);
10980 rn = (insn >> 3) & 0x7;
10981 rd = insn & 0x7;
10982 tmp = load_reg(s, rn);
10983 switch ((insn >> 6) & 3) {
10984 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
10985 case 1: gen_rev16(tmp); break;
10986 case 3: gen_revsh(tmp); break;
10987 default: goto illegal_op;
10989 store_reg(s, rd, tmp);
10990 break;
10992 case 6:
10993 switch ((insn >> 5) & 7) {
10994 case 2:
10995 /* setend */
10996 ARCH(6);
10997 if (((insn >> 3) & 1) != s->bswap_code) {
10998 /* Dynamic endianness switching not implemented. */
10999 qemu_log_mask(LOG_UNIMP, "arm: unimplemented setend\n");
11000 goto illegal_op;
11002 break;
11003 case 3:
11004 /* cps */
11005 ARCH(6);
11006 if (IS_USER(s)) {
11007 break;
11009 if (arm_dc_feature(s, ARM_FEATURE_M)) {
11010 tmp = tcg_const_i32((insn & (1 << 4)) != 0);
11011 /* FAULTMASK */
11012 if (insn & 1) {
11013 addr = tcg_const_i32(19);
11014 gen_helper_v7m_msr(cpu_env, addr, tmp);
11015 tcg_temp_free_i32(addr);
11017 /* PRIMASK */
11018 if (insn & 2) {
11019 addr = tcg_const_i32(16);
11020 gen_helper_v7m_msr(cpu_env, addr, tmp);
11021 tcg_temp_free_i32(addr);
11023 tcg_temp_free_i32(tmp);
11024 gen_lookup_tb(s);
11025 } else {
11026 if (insn & (1 << 4)) {
11027 shift = CPSR_A | CPSR_I | CPSR_F;
11028 } else {
11029 shift = 0;
11031 gen_set_psr_im(s, ((insn & 7) << 6), 0, shift);
11033 break;
11034 default:
11035 goto undef;
11037 break;
11039 default:
11040 goto undef;
11042 break;
11044 case 12:
11046 /* load/store multiple */
11047 TCGv_i32 loaded_var;
11048 TCGV_UNUSED_I32(loaded_var);
11049 rn = (insn >> 8) & 0x7;
11050 addr = load_reg(s, rn);
11051 for (i = 0; i < 8; i++) {
11052 if (insn & (1 << i)) {
11053 if (insn & (1 << 11)) {
11054 /* load */
11055 tmp = tcg_temp_new_i32();
11056 gen_aa32_ld32u(tmp, addr, get_mem_index(s));
11057 if (i == rn) {
11058 loaded_var = tmp;
11059 } else {
11060 store_reg(s, i, tmp);
11062 } else {
11063 /* store */
11064 tmp = load_reg(s, i);
11065 gen_aa32_st32(tmp, addr, get_mem_index(s));
11066 tcg_temp_free_i32(tmp);
11068 /* advance to the next address */
11069 tcg_gen_addi_i32(addr, addr, 4);
11072 if ((insn & (1 << rn)) == 0) {
11073 /* base reg not in list: base register writeback */
11074 store_reg(s, rn, addr);
11075 } else {
11076 /* base reg in list: if load, complete it now */
11077 if (insn & (1 << 11)) {
11078 store_reg(s, rn, loaded_var);
11080 tcg_temp_free_i32(addr);
11082 break;
11084 case 13:
11085 /* conditional branch or swi */
11086 cond = (insn >> 8) & 0xf;
11087 if (cond == 0xe)
11088 goto undef;
11090 if (cond == 0xf) {
11091 /* swi */
11092 gen_set_pc_im(s, s->pc);
11093 s->svc_imm = extract32(insn, 0, 8);
11094 s->is_jmp = DISAS_SWI;
11095 break;
11097 /* generate a conditional jump to next instruction */
11098 s->condlabel = gen_new_label();
11099 arm_gen_test_cc(cond ^ 1, s->condlabel);
11100 s->condjmp = 1;
11102 /* jump to the offset */
11103 val = (uint32_t)s->pc + 2;
11104 offset = ((int32_t)insn << 24) >> 24;
11105 val += offset << 1;
11106 gen_jmp(s, val);
11107 break;
11109 case 14:
11110 if (insn & (1 << 11)) {
11111 if (disas_thumb2_insn(env, s, insn))
11112 goto undef32;
11113 break;
11115 /* unconditional branch */
11116 val = (uint32_t)s->pc;
11117 offset = ((int32_t)insn << 21) >> 21;
11118 val += (offset << 1) + 2;
11119 gen_jmp(s, val);
11120 break;
11122 case 15:
11123 if (disas_thumb2_insn(env, s, insn))
11124 goto undef32;
11125 break;
11127 return;
11128 undef32:
11129 gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized(),
11130 default_exception_el(s));
11131 return;
11132 illegal_op:
11133 undef:
11134 gen_exception_insn(s, 2, EXCP_UDEF, syn_uncategorized(),
11135 default_exception_el(s));
11138 /* generate intermediate code in gen_opc_buf and gen_opparam_buf for
11139 basic block 'tb'. If search_pc is TRUE, also generate PC
11140 information for each intermediate instruction. */
11141 static inline void gen_intermediate_code_internal(ARMCPU *cpu,
11142 TranslationBlock *tb,
11143 bool search_pc)
11145 CPUState *cs = CPU(cpu);
11146 CPUARMState *env = &cpu->env;
11147 DisasContext dc1, *dc = &dc1;
11148 CPUBreakpoint *bp;
11149 int j, lj;
11150 target_ulong pc_start;
11151 target_ulong next_page_start;
11152 int num_insns;
11153 int max_insns;
11155 /* generate intermediate code */
11157 /* The A64 decoder has its own top level loop, because it doesn't need
11158 * the A32/T32 complexity to do with conditional execution/IT blocks/etc.
11160 if (ARM_TBFLAG_AARCH64_STATE(tb->flags)) {
11161 gen_intermediate_code_internal_a64(cpu, tb, search_pc);
11162 return;
11165 pc_start = tb->pc;
11167 dc->tb = tb;
11169 dc->is_jmp = DISAS_NEXT;
11170 dc->pc = pc_start;
11171 dc->singlestep_enabled = cs->singlestep_enabled;
11172 dc->condjmp = 0;
11174 dc->aarch64 = 0;
11175 dc->el3_is_aa64 = arm_el_is_aa64(env, 3);
11176 dc->thumb = ARM_TBFLAG_THUMB(tb->flags);
11177 dc->bswap_code = ARM_TBFLAG_BSWAP_CODE(tb->flags);
11178 dc->condexec_mask = (ARM_TBFLAG_CONDEXEC(tb->flags) & 0xf) << 1;
11179 dc->condexec_cond = ARM_TBFLAG_CONDEXEC(tb->flags) >> 4;
11180 dc->mmu_idx = ARM_TBFLAG_MMUIDX(tb->flags);
11181 dc->current_el = arm_mmu_idx_to_el(dc->mmu_idx);
11182 #if !defined(CONFIG_USER_ONLY)
11183 dc->user = (dc->current_el == 0);
11184 #endif
11185 dc->ns = ARM_TBFLAG_NS(tb->flags);
11186 dc->fp_excp_el = ARM_TBFLAG_FPEXC_EL(tb->flags);
11187 dc->vfp_enabled = ARM_TBFLAG_VFPEN(tb->flags);
11188 dc->vec_len = ARM_TBFLAG_VECLEN(tb->flags);
11189 dc->vec_stride = ARM_TBFLAG_VECSTRIDE(tb->flags);
11190 dc->c15_cpar = ARM_TBFLAG_XSCALE_CPAR(tb->flags);
11191 dc->cp_regs = cpu->cp_regs;
11192 dc->features = env->features;
11194 /* Single step state. The code-generation logic here is:
11195 * SS_ACTIVE == 0:
11196 * generate code with no special handling for single-stepping (except
11197 * that anything that can make us go to SS_ACTIVE == 1 must end the TB;
11198 * this happens anyway because those changes are all system register or
11199 * PSTATE writes).
11200 * SS_ACTIVE == 1, PSTATE.SS == 1: (active-not-pending)
11201 * emit code for one insn
11202 * emit code to clear PSTATE.SS
11203 * emit code to generate software step exception for completed step
11204 * end TB (as usual for having generated an exception)
11205 * SS_ACTIVE == 1, PSTATE.SS == 0: (active-pending)
11206 * emit code to generate a software step exception
11207 * end the TB
11209 dc->ss_active = ARM_TBFLAG_SS_ACTIVE(tb->flags);
11210 dc->pstate_ss = ARM_TBFLAG_PSTATE_SS(tb->flags);
11211 dc->is_ldex = false;
11212 dc->ss_same_el = false; /* Can't be true since EL_d must be AArch64 */
11214 cpu_F0s = tcg_temp_new_i32();
11215 cpu_F1s = tcg_temp_new_i32();
11216 cpu_F0d = tcg_temp_new_i64();
11217 cpu_F1d = tcg_temp_new_i64();
11218 cpu_V0 = cpu_F0d;
11219 cpu_V1 = cpu_F1d;
11220 /* FIXME: cpu_M0 can probably be the same as cpu_V0. */
11221 cpu_M0 = tcg_temp_new_i64();
11222 next_page_start = (pc_start & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE;
11223 lj = -1;
11224 num_insns = 0;
11225 max_insns = tb->cflags & CF_COUNT_MASK;
11226 if (max_insns == 0)
11227 max_insns = CF_COUNT_MASK;
11229 gen_tb_start(tb);
11231 tcg_clear_temp_count();
11233 /* A note on handling of the condexec (IT) bits:
11235 * We want to avoid the overhead of having to write the updated condexec
11236 * bits back to the CPUARMState for every instruction in an IT block. So:
11237 * (1) if the condexec bits are not already zero then we write
11238 * zero back into the CPUARMState now. This avoids complications trying
11239 * to do it at the end of the block. (For example if we don't do this
11240 * it's hard to identify whether we can safely skip writing condexec
11241 * at the end of the TB, which we definitely want to do for the case
11242 * where a TB doesn't do anything with the IT state at all.)
11243 * (2) if we are going to leave the TB then we call gen_set_condexec()
11244 * which will write the correct value into CPUARMState if zero is wrong.
11245 * This is done both for leaving the TB at the end, and for leaving
11246 * it because of an exception we know will happen, which is done in
11247 * gen_exception_insn(). The latter is necessary because we need to
11248 * leave the TB with the PC/IT state just prior to execution of the
11249 * instruction which caused the exception.
11250 * (3) if we leave the TB unexpectedly (eg a data abort on a load)
11251 * then the CPUARMState will be wrong and we need to reset it.
11252 * This is handled in the same way as restoration of the
11253 * PC in these situations: we will be called again with search_pc=1
11254 * and generate a mapping of the condexec bits for each PC in
11255 * gen_opc_condexec_bits[]. restore_state_to_opc() then uses
11256 * this to restore the condexec bits.
11258 * Note that there are no instructions which can read the condexec
11259 * bits, and none which can write non-static values to them, so
11260 * we don't need to care about whether CPUARMState is correct in the
11261 * middle of a TB.
11264 /* Reset the conditional execution bits immediately. This avoids
11265 complications trying to do it at the end of the block. */
11266 if (dc->condexec_mask || dc->condexec_cond)
11268 TCGv_i32 tmp = tcg_temp_new_i32();
11269 tcg_gen_movi_i32(tmp, 0);
11270 store_cpu_field(tmp, condexec_bits);
11272 do {
11273 #ifdef CONFIG_USER_ONLY
11274 /* Intercept jump to the magic kernel page. */
11275 if (dc->pc >= 0xffff0000) {
11276 /* We always get here via a jump, so know we are not in a
11277 conditional execution block. */
11278 gen_exception_internal(EXCP_KERNEL_TRAP);
11279 dc->is_jmp = DISAS_UPDATE;
11280 break;
11282 #else
11283 if (dc->pc >= 0xfffffff0 && arm_dc_feature(dc, ARM_FEATURE_M)) {
11284 /* We always get here via a jump, so know we are not in a
11285 conditional execution block. */
11286 gen_exception_internal(EXCP_EXCEPTION_EXIT);
11287 dc->is_jmp = DISAS_UPDATE;
11288 break;
11290 #endif
11292 if (unlikely(!QTAILQ_EMPTY(&cs->breakpoints))) {
11293 QTAILQ_FOREACH(bp, &cs->breakpoints, entry) {
11294 if (bp->pc == dc->pc) {
11295 gen_exception_internal_insn(dc, 0, EXCP_DEBUG);
11296 /* Advance PC so that clearing the breakpoint will
11297 invalidate this TB. */
11298 dc->pc += 2;
11299 goto done_generating;
11303 if (search_pc) {
11304 j = tcg_op_buf_count();
11305 if (lj < j) {
11306 lj++;
11307 while (lj < j)
11308 tcg_ctx.gen_opc_instr_start[lj++] = 0;
11310 tcg_ctx.gen_opc_pc[lj] = dc->pc;
11311 gen_opc_condexec_bits[lj] = (dc->condexec_cond << 4) | (dc->condexec_mask >> 1);
11312 tcg_ctx.gen_opc_instr_start[lj] = 1;
11313 tcg_ctx.gen_opc_icount[lj] = num_insns;
11316 if (num_insns + 1 == max_insns && (tb->cflags & CF_LAST_IO))
11317 gen_io_start();
11319 if (unlikely(qemu_loglevel_mask(CPU_LOG_TB_OP | CPU_LOG_TB_OP_OPT))) {
11320 tcg_gen_debug_insn_start(dc->pc);
11323 if (dc->ss_active && !dc->pstate_ss) {
11324 /* Singlestep state is Active-pending.
11325 * If we're in this state at the start of a TB then either
11326 * a) we just took an exception to an EL which is being debugged
11327 * and this is the first insn in the exception handler
11328 * b) debug exceptions were masked and we just unmasked them
11329 * without changing EL (eg by clearing PSTATE.D)
11330 * In either case we're going to take a swstep exception in the
11331 * "did not step an insn" case, and so the syndrome ISV and EX
11332 * bits should be zero.
11334 assert(num_insns == 0);
11335 gen_exception(EXCP_UDEF, syn_swstep(dc->ss_same_el, 0, 0),
11336 default_exception_el(dc));
11337 goto done_generating;
11340 if (dc->thumb) {
11341 disas_thumb_insn(env, dc);
11342 if (dc->condexec_mask) {
11343 dc->condexec_cond = (dc->condexec_cond & 0xe)
11344 | ((dc->condexec_mask >> 4) & 1);
11345 dc->condexec_mask = (dc->condexec_mask << 1) & 0x1f;
11346 if (dc->condexec_mask == 0) {
11347 dc->condexec_cond = 0;
11350 } else {
11351 unsigned int insn = arm_ldl_code(env, dc->pc, dc->bswap_code);
11352 dc->pc += 4;
11353 disas_arm_insn(dc, insn);
11356 if (dc->condjmp && !dc->is_jmp) {
11357 gen_set_label(dc->condlabel);
11358 dc->condjmp = 0;
11361 if (tcg_check_temp_count()) {
11362 fprintf(stderr, "TCG temporary leak before "TARGET_FMT_lx"\n",
11363 dc->pc);
11366 /* Translation stops when a conditional branch is encountered.
11367 * Otherwise the subsequent code could get translated several times.
11368 * Also stop translation when a page boundary is reached. This
11369 * ensures prefetch aborts occur at the right place. */
11370 num_insns ++;
11371 } while (!dc->is_jmp && !tcg_op_buf_full() &&
11372 !cs->singlestep_enabled &&
11373 !singlestep &&
11374 !dc->ss_active &&
11375 dc->pc < next_page_start &&
11376 num_insns < max_insns);
11378 if (tb->cflags & CF_LAST_IO) {
11379 if (dc->condjmp) {
11380 /* FIXME: This can theoretically happen with self-modifying
11381 code. */
11382 cpu_abort(cs, "IO on conditional branch instruction");
11384 gen_io_end();
11387 /* At this stage dc->condjmp will only be set when the skipped
11388 instruction was a conditional branch or trap, and the PC has
11389 already been written. */
11390 if (unlikely(cs->singlestep_enabled || dc->ss_active)) {
11391 /* Make sure the pc is updated, and raise a debug exception. */
11392 if (dc->condjmp) {
11393 gen_set_condexec(dc);
11394 if (dc->is_jmp == DISAS_SWI) {
11395 gen_ss_advance(dc);
11396 gen_exception(EXCP_SWI, syn_aa32_svc(dc->svc_imm, dc->thumb),
11397 default_exception_el(dc));
11398 } else if (dc->is_jmp == DISAS_HVC) {
11399 gen_ss_advance(dc);
11400 gen_exception(EXCP_HVC, syn_aa32_hvc(dc->svc_imm), 2);
11401 } else if (dc->is_jmp == DISAS_SMC) {
11402 gen_ss_advance(dc);
11403 gen_exception(EXCP_SMC, syn_aa32_smc(), 3);
11404 } else if (dc->ss_active) {
11405 gen_step_complete_exception(dc);
11406 } else {
11407 gen_exception_internal(EXCP_DEBUG);
11409 gen_set_label(dc->condlabel);
11411 if (dc->condjmp || !dc->is_jmp) {
11412 gen_set_pc_im(dc, dc->pc);
11413 dc->condjmp = 0;
11415 gen_set_condexec(dc);
11416 if (dc->is_jmp == DISAS_SWI && !dc->condjmp) {
11417 gen_ss_advance(dc);
11418 gen_exception(EXCP_SWI, syn_aa32_svc(dc->svc_imm, dc->thumb),
11419 default_exception_el(dc));
11420 } else if (dc->is_jmp == DISAS_HVC && !dc->condjmp) {
11421 gen_ss_advance(dc);
11422 gen_exception(EXCP_HVC, syn_aa32_hvc(dc->svc_imm), 2);
11423 } else if (dc->is_jmp == DISAS_SMC && !dc->condjmp) {
11424 gen_ss_advance(dc);
11425 gen_exception(EXCP_SMC, syn_aa32_smc(), 3);
11426 } else if (dc->ss_active) {
11427 gen_step_complete_exception(dc);
11428 } else {
11429 /* FIXME: Single stepping a WFI insn will not halt
11430 the CPU. */
11431 gen_exception_internal(EXCP_DEBUG);
11433 } else {
11434 /* While branches must always occur at the end of an IT block,
11435 there are a few other things that can cause us to terminate
11436 the TB in the middle of an IT block:
11437 - Exception generating instructions (bkpt, swi, undefined).
11438 - Page boundaries.
11439 - Hardware watchpoints.
11440 Hardware breakpoints have already been handled and skip this code.
11442 gen_set_condexec(dc);
11443 switch(dc->is_jmp) {
11444 case DISAS_NEXT:
11445 gen_goto_tb(dc, 1, dc->pc);
11446 break;
11447 default:
11448 case DISAS_JUMP:
11449 case DISAS_UPDATE:
11450 /* indicate that the hash table must be used to find the next TB */
11451 tcg_gen_exit_tb(0);
11452 break;
11453 case DISAS_TB_JUMP:
11454 /* nothing more to generate */
11455 break;
11456 case DISAS_WFI:
11457 gen_helper_wfi(cpu_env);
11458 /* The helper doesn't necessarily throw an exception, but we
11459 * must go back to the main loop to check for interrupts anyway.
11461 tcg_gen_exit_tb(0);
11462 break;
11463 case DISAS_WFE:
11464 gen_helper_wfe(cpu_env);
11465 break;
11466 case DISAS_YIELD:
11467 gen_helper_yield(cpu_env);
11468 break;
11469 case DISAS_SWI:
11470 gen_exception(EXCP_SWI, syn_aa32_svc(dc->svc_imm, dc->thumb),
11471 default_exception_el(dc));
11472 break;
11473 case DISAS_HVC:
11474 gen_exception(EXCP_HVC, syn_aa32_hvc(dc->svc_imm), 2);
11475 break;
11476 case DISAS_SMC:
11477 gen_exception(EXCP_SMC, syn_aa32_smc(), 3);
11478 break;
11480 if (dc->condjmp) {
11481 gen_set_label(dc->condlabel);
11482 gen_set_condexec(dc);
11483 gen_goto_tb(dc, 1, dc->pc);
11484 dc->condjmp = 0;
11488 done_generating:
11489 gen_tb_end(tb, num_insns);
11491 #ifdef DEBUG_DISAS
11492 if (qemu_loglevel_mask(CPU_LOG_TB_IN_ASM)) {
11493 qemu_log("----------------\n");
11494 qemu_log("IN: %s\n", lookup_symbol(pc_start));
11495 log_target_disas(cs, pc_start, dc->pc - pc_start,
11496 dc->thumb | (dc->bswap_code << 1));
11497 qemu_log("\n");
11499 #endif
11500 if (search_pc) {
11501 j = tcg_op_buf_count();
11502 lj++;
11503 while (lj <= j)
11504 tcg_ctx.gen_opc_instr_start[lj++] = 0;
11505 } else {
11506 tb->size = dc->pc - pc_start;
11507 tb->icount = num_insns;
11511 void gen_intermediate_code(CPUARMState *env, TranslationBlock *tb)
11513 gen_intermediate_code_internal(arm_env_get_cpu(env), tb, false);
11516 void gen_intermediate_code_pc(CPUARMState *env, TranslationBlock *tb)
11518 gen_intermediate_code_internal(arm_env_get_cpu(env), tb, true);
11521 static const char *cpu_mode_names[16] = {
11522 "usr", "fiq", "irq", "svc", "???", "???", "mon", "abt",
11523 "???", "???", "hyp", "und", "???", "???", "???", "sys"
11526 void arm_cpu_dump_state(CPUState *cs, FILE *f, fprintf_function cpu_fprintf,
11527 int flags)
11529 ARMCPU *cpu = ARM_CPU(cs);
11530 CPUARMState *env = &cpu->env;
11531 int i;
11532 uint32_t psr;
11534 if (is_a64(env)) {
11535 aarch64_cpu_dump_state(cs, f, cpu_fprintf, flags);
11536 return;
11539 for(i=0;i<16;i++) {
11540 cpu_fprintf(f, "R%02d=%08x", i, env->regs[i]);
11541 if ((i % 4) == 3)
11542 cpu_fprintf(f, "\n");
11543 else
11544 cpu_fprintf(f, " ");
11546 psr = cpsr_read(env);
11547 cpu_fprintf(f, "PSR=%08x %c%c%c%c %c %s%d\n",
11548 psr,
11549 psr & (1 << 31) ? 'N' : '-',
11550 psr & (1 << 30) ? 'Z' : '-',
11551 psr & (1 << 29) ? 'C' : '-',
11552 psr & (1 << 28) ? 'V' : '-',
11553 psr & CPSR_T ? 'T' : 'A',
11554 cpu_mode_names[psr & 0xf], (psr & 0x10) ? 32 : 26);
11556 if (flags & CPU_DUMP_FPU) {
11557 int numvfpregs = 0;
11558 if (arm_feature(env, ARM_FEATURE_VFP)) {
11559 numvfpregs += 16;
11561 if (arm_feature(env, ARM_FEATURE_VFP3)) {
11562 numvfpregs += 16;
11564 for (i = 0; i < numvfpregs; i++) {
11565 uint64_t v = float64_val(env->vfp.regs[i]);
11566 cpu_fprintf(f, "s%02d=%08x s%02d=%08x d%02d=%016" PRIx64 "\n",
11567 i * 2, (uint32_t)v,
11568 i * 2 + 1, (uint32_t)(v >> 32),
11569 i, v);
11571 cpu_fprintf(f, "FPSCR: %08x\n", (int)env->vfp.xregs[ARM_VFP_FPSCR]);
11575 void restore_state_to_opc(CPUARMState *env, TranslationBlock *tb, int pc_pos)
11577 if (is_a64(env)) {
11578 env->pc = tcg_ctx.gen_opc_pc[pc_pos];
11579 env->condexec_bits = 0;
11580 } else {
11581 env->regs[15] = tcg_ctx.gen_opc_pc[pc_pos];
11582 env->condexec_bits = gen_opc_condexec_bits[pc_pos];