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 "qemu/osdep.h"
24 #include "internals.h"
25 #include "disas/disas.h"
26 #include "exec/exec-all.h"
27 #include "tcg/tcg-op.h"
28 #include "tcg/tcg-op-gvec.h"
30 #include "qemu/bitops.h"
32 #include "hw/semihosting/semihost.h"
34 #include "exec/helper-proto.h"
35 #include "exec/helper-gen.h"
37 #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 dc_isar_feature(aa32_jazelle, s)
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"
56 #if defined(CONFIG_USER_ONLY)
59 #define IS_USER(s) (s->user)
62 /* We reuse the same 64-bit temporaries for efficiency. */
63 static TCGv_i64 cpu_V0, cpu_V1, cpu_M0;
64 static TCGv_i32 cpu_R[16];
65 TCGv_i32 cpu_CF, cpu_NF, cpu_VF, cpu_ZF;
66 TCGv_i64 cpu_exclusive_addr;
67 TCGv_i64 cpu_exclusive_val;
69 #include "exec/gen-icount.h"
71 static const char * const regnames[] =
72 { "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7",
73 "r8", "r9", "r10", "r11", "r12", "r13", "r14", "pc" };
75 /* Function prototypes for gen_ functions calling Neon helpers. */
76 typedef void NeonGenThreeOpEnvFn(TCGv_i32, TCGv_env, TCGv_i32,
78 /* Function prototypes for gen_ functions for fix point conversions */
79 typedef void VFPGenFixPointFn(TCGv_i32, TCGv_i32, TCGv_i32, TCGv_ptr);
81 /* initialize TCG globals. */
82 void arm_translate_init(void)
86 for (i = 0; i < 16; i++) {
87 cpu_R[i] = tcg_global_mem_new_i32(cpu_env,
88 offsetof(CPUARMState, regs[i]),
91 cpu_CF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, CF), "CF");
92 cpu_NF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, NF), "NF");
93 cpu_VF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, VF), "VF");
94 cpu_ZF = tcg_global_mem_new_i32(cpu_env, offsetof(CPUARMState, ZF), "ZF");
96 cpu_exclusive_addr = tcg_global_mem_new_i64(cpu_env,
97 offsetof(CPUARMState, exclusive_addr), "exclusive_addr");
98 cpu_exclusive_val = tcg_global_mem_new_i64(cpu_env,
99 offsetof(CPUARMState, exclusive_val), "exclusive_val");
101 a64_translate_init();
104 /* Flags for the disas_set_da_iss info argument:
105 * lower bits hold the Rt register number, higher bits are flags.
107 typedef enum ISSInfo {
110 ISSInvalid = (1 << 5),
111 ISSIsAcqRel = (1 << 6),
112 ISSIsWrite = (1 << 7),
113 ISSIs16Bit = (1 << 8),
116 /* Save the syndrome information for a Data Abort */
117 static void disas_set_da_iss(DisasContext *s, MemOp memop, ISSInfo issinfo)
120 int sas = memop & MO_SIZE;
121 bool sse = memop & MO_SIGN;
122 bool is_acqrel = issinfo & ISSIsAcqRel;
123 bool is_write = issinfo & ISSIsWrite;
124 bool is_16bit = issinfo & ISSIs16Bit;
125 int srt = issinfo & ISSRegMask;
127 if (issinfo & ISSInvalid) {
128 /* Some callsites want to conditionally provide ISS info,
129 * eg "only if this was not a writeback"
135 /* For AArch32, insns where the src/dest is R15 never generate
136 * ISS information. Catching that here saves checking at all
142 syn = syn_data_abort_with_iss(0, sas, sse, srt, 0, is_acqrel,
143 0, 0, 0, is_write, 0, is_16bit);
144 disas_set_insn_syndrome(s, syn);
147 static inline int get_a32_user_mem_index(DisasContext *s)
149 /* Return the core mmu_idx to use for A32/T32 "unprivileged load/store"
151 * if PL2, UNPREDICTABLE (we choose to implement as if PL0)
152 * otherwise, access as if at PL0.
154 switch (s->mmu_idx) {
155 case ARMMMUIdx_E2: /* this one is UNPREDICTABLE */
156 case ARMMMUIdx_E10_0:
157 case ARMMMUIdx_E10_1:
158 case ARMMMUIdx_E10_1_PAN:
159 return arm_to_core_mmu_idx(ARMMMUIdx_E10_0);
161 case ARMMMUIdx_SE10_0:
162 case ARMMMUIdx_SE10_1:
163 case ARMMMUIdx_SE10_1_PAN:
164 return arm_to_core_mmu_idx(ARMMMUIdx_SE10_0);
165 case ARMMMUIdx_MUser:
166 case ARMMMUIdx_MPriv:
167 return arm_to_core_mmu_idx(ARMMMUIdx_MUser);
168 case ARMMMUIdx_MUserNegPri:
169 case ARMMMUIdx_MPrivNegPri:
170 return arm_to_core_mmu_idx(ARMMMUIdx_MUserNegPri);
171 case ARMMMUIdx_MSUser:
172 case ARMMMUIdx_MSPriv:
173 return arm_to_core_mmu_idx(ARMMMUIdx_MSUser);
174 case ARMMMUIdx_MSUserNegPri:
175 case ARMMMUIdx_MSPrivNegPri:
176 return arm_to_core_mmu_idx(ARMMMUIdx_MSUserNegPri);
178 g_assert_not_reached();
182 static inline TCGv_i32 load_cpu_offset(int offset)
184 TCGv_i32 tmp = tcg_temp_new_i32();
185 tcg_gen_ld_i32(tmp, cpu_env, offset);
189 #define load_cpu_field(name) load_cpu_offset(offsetof(CPUARMState, name))
191 static inline void store_cpu_offset(TCGv_i32 var, int offset)
193 tcg_gen_st_i32(var, cpu_env, offset);
194 tcg_temp_free_i32(var);
197 #define store_cpu_field(var, name) \
198 store_cpu_offset(var, offsetof(CPUARMState, name))
200 /* The architectural value of PC. */
201 static uint32_t read_pc(DisasContext *s)
203 return s->pc_curr + (s->thumb ? 4 : 8);
206 /* Set a variable to the value of a CPU register. */
207 static void load_reg_var(DisasContext *s, TCGv_i32 var, int reg)
210 tcg_gen_movi_i32(var, read_pc(s));
212 tcg_gen_mov_i32(var, cpu_R[reg]);
216 /* Create a new temporary and set it to the value of a CPU register. */
217 static inline TCGv_i32 load_reg(DisasContext *s, int reg)
219 TCGv_i32 tmp = tcg_temp_new_i32();
220 load_reg_var(s, tmp, reg);
225 * Create a new temp, REG + OFS, except PC is ALIGN(PC, 4).
226 * This is used for load/store for which use of PC implies (literal),
227 * or ADD that implies ADR.
229 static TCGv_i32 add_reg_for_lit(DisasContext *s, int reg, int ofs)
231 TCGv_i32 tmp = tcg_temp_new_i32();
234 tcg_gen_movi_i32(tmp, (read_pc(s) & ~3) + ofs);
236 tcg_gen_addi_i32(tmp, cpu_R[reg], ofs);
241 /* Set a CPU register. The source must be a temporary and will be
243 static void store_reg(DisasContext *s, int reg, TCGv_i32 var)
246 /* In Thumb mode, we must ignore bit 0.
247 * In ARM mode, for ARMv4 and ARMv5, it is UNPREDICTABLE if bits [1:0]
248 * are not 0b00, but for ARMv6 and above, we must ignore bits [1:0].
249 * We choose to ignore [1:0] in ARM mode for all architecture versions.
251 tcg_gen_andi_i32(var, var, s->thumb ? ~1 : ~3);
252 s->base.is_jmp = DISAS_JUMP;
254 tcg_gen_mov_i32(cpu_R[reg], var);
255 tcg_temp_free_i32(var);
259 * Variant of store_reg which applies v8M stack-limit checks before updating
260 * SP. If the check fails this will result in an exception being taken.
261 * We disable the stack checks for CONFIG_USER_ONLY because we have
262 * no idea what the stack limits should be in that case.
263 * If stack checking is not being done this just acts like store_reg().
265 static void store_sp_checked(DisasContext *s, TCGv_i32 var)
267 #ifndef CONFIG_USER_ONLY
268 if (s->v8m_stackcheck) {
269 gen_helper_v8m_stackcheck(cpu_env, var);
272 store_reg(s, 13, var);
275 /* Value extensions. */
276 #define gen_uxtb(var) tcg_gen_ext8u_i32(var, var)
277 #define gen_uxth(var) tcg_gen_ext16u_i32(var, var)
278 #define gen_sxtb(var) tcg_gen_ext8s_i32(var, var)
279 #define gen_sxth(var) tcg_gen_ext16s_i32(var, var)
281 #define gen_sxtb16(var) gen_helper_sxtb16(var, var)
282 #define gen_uxtb16(var) gen_helper_uxtb16(var, var)
285 static inline void gen_set_cpsr(TCGv_i32 var, uint32_t mask)
287 TCGv_i32 tmp_mask = tcg_const_i32(mask);
288 gen_helper_cpsr_write(cpu_env, var, tmp_mask);
289 tcg_temp_free_i32(tmp_mask);
291 /* Set NZCV flags from the high 4 bits of var. */
292 #define gen_set_nzcv(var) gen_set_cpsr(var, CPSR_NZCV)
294 static void gen_exception_internal(int excp)
296 TCGv_i32 tcg_excp = tcg_const_i32(excp);
298 assert(excp_is_internal(excp));
299 gen_helper_exception_internal(cpu_env, tcg_excp);
300 tcg_temp_free_i32(tcg_excp);
303 static void gen_step_complete_exception(DisasContext *s)
305 /* We just completed step of an insn. Move from Active-not-pending
306 * to Active-pending, and then also take the swstep exception.
307 * This corresponds to making the (IMPDEF) choice to prioritize
308 * swstep exceptions over asynchronous exceptions taken to an exception
309 * level where debug is disabled. This choice has the advantage that
310 * we do not need to maintain internal state corresponding to the
311 * ISV/EX syndrome bits between completion of the step and generation
312 * of the exception, and our syndrome information is always correct.
315 gen_swstep_exception(s, 1, s->is_ldex);
316 s->base.is_jmp = DISAS_NORETURN;
319 static void gen_singlestep_exception(DisasContext *s)
321 /* Generate the right kind of exception for singlestep, which is
322 * either the architectural singlestep or EXCP_DEBUG for QEMU's
323 * gdb singlestepping.
326 gen_step_complete_exception(s);
328 gen_exception_internal(EXCP_DEBUG);
332 static inline bool is_singlestepping(DisasContext *s)
334 /* Return true if we are singlestepping either because of
335 * architectural singlestep or QEMU gdbstub singlestep. This does
336 * not include the command line '-singlestep' mode which is rather
337 * misnamed as it only means "one instruction per TB" and doesn't
338 * affect the code we generate.
340 return s->base.singlestep_enabled || s->ss_active;
343 static void gen_smul_dual(TCGv_i32 a, TCGv_i32 b)
345 TCGv_i32 tmp1 = tcg_temp_new_i32();
346 TCGv_i32 tmp2 = tcg_temp_new_i32();
347 tcg_gen_ext16s_i32(tmp1, a);
348 tcg_gen_ext16s_i32(tmp2, b);
349 tcg_gen_mul_i32(tmp1, tmp1, tmp2);
350 tcg_temp_free_i32(tmp2);
351 tcg_gen_sari_i32(a, a, 16);
352 tcg_gen_sari_i32(b, b, 16);
353 tcg_gen_mul_i32(b, b, a);
354 tcg_gen_mov_i32(a, tmp1);
355 tcg_temp_free_i32(tmp1);
358 /* Byteswap each halfword. */
359 static void gen_rev16(TCGv_i32 dest, TCGv_i32 var)
361 TCGv_i32 tmp = tcg_temp_new_i32();
362 TCGv_i32 mask = tcg_const_i32(0x00ff00ff);
363 tcg_gen_shri_i32(tmp, var, 8);
364 tcg_gen_and_i32(tmp, tmp, mask);
365 tcg_gen_and_i32(var, var, mask);
366 tcg_gen_shli_i32(var, var, 8);
367 tcg_gen_or_i32(dest, var, tmp);
368 tcg_temp_free_i32(mask);
369 tcg_temp_free_i32(tmp);
372 /* Byteswap low halfword and sign extend. */
373 static void gen_revsh(TCGv_i32 dest, TCGv_i32 var)
375 tcg_gen_ext16u_i32(var, var);
376 tcg_gen_bswap16_i32(var, var);
377 tcg_gen_ext16s_i32(dest, var);
380 /* 32x32->64 multiply. Marks inputs as dead. */
381 static TCGv_i64 gen_mulu_i64_i32(TCGv_i32 a, TCGv_i32 b)
383 TCGv_i32 lo = tcg_temp_new_i32();
384 TCGv_i32 hi = tcg_temp_new_i32();
387 tcg_gen_mulu2_i32(lo, hi, a, b);
388 tcg_temp_free_i32(a);
389 tcg_temp_free_i32(b);
391 ret = tcg_temp_new_i64();
392 tcg_gen_concat_i32_i64(ret, lo, hi);
393 tcg_temp_free_i32(lo);
394 tcg_temp_free_i32(hi);
399 static TCGv_i64 gen_muls_i64_i32(TCGv_i32 a, TCGv_i32 b)
401 TCGv_i32 lo = tcg_temp_new_i32();
402 TCGv_i32 hi = tcg_temp_new_i32();
405 tcg_gen_muls2_i32(lo, hi, a, b);
406 tcg_temp_free_i32(a);
407 tcg_temp_free_i32(b);
409 ret = tcg_temp_new_i64();
410 tcg_gen_concat_i32_i64(ret, lo, hi);
411 tcg_temp_free_i32(lo);
412 tcg_temp_free_i32(hi);
417 /* Swap low and high halfwords. */
418 static void gen_swap_half(TCGv_i32 var)
420 tcg_gen_rotri_i32(var, var, 16);
423 /* Dual 16-bit add. Result placed in t0 and t1 is marked as dead.
424 tmp = (t0 ^ t1) & 0x8000;
427 t0 = (t0 + t1) ^ tmp;
430 static void gen_add16(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
432 TCGv_i32 tmp = tcg_temp_new_i32();
433 tcg_gen_xor_i32(tmp, t0, t1);
434 tcg_gen_andi_i32(tmp, tmp, 0x8000);
435 tcg_gen_andi_i32(t0, t0, ~0x8000);
436 tcg_gen_andi_i32(t1, t1, ~0x8000);
437 tcg_gen_add_i32(t0, t0, t1);
438 tcg_gen_xor_i32(dest, t0, tmp);
439 tcg_temp_free_i32(tmp);
442 /* Set N and Z flags from var. */
443 static inline void gen_logic_CC(TCGv_i32 var)
445 tcg_gen_mov_i32(cpu_NF, var);
446 tcg_gen_mov_i32(cpu_ZF, var);
449 /* dest = T0 + T1 + CF. */
450 static void gen_add_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
452 tcg_gen_add_i32(dest, t0, t1);
453 tcg_gen_add_i32(dest, dest, cpu_CF);
456 /* dest = T0 - T1 + CF - 1. */
457 static void gen_sub_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
459 tcg_gen_sub_i32(dest, t0, t1);
460 tcg_gen_add_i32(dest, dest, cpu_CF);
461 tcg_gen_subi_i32(dest, dest, 1);
464 /* dest = T0 + T1. Compute C, N, V and Z flags */
465 static void gen_add_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
467 TCGv_i32 tmp = tcg_temp_new_i32();
468 tcg_gen_movi_i32(tmp, 0);
469 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, t1, tmp);
470 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
471 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
472 tcg_gen_xor_i32(tmp, t0, t1);
473 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
474 tcg_temp_free_i32(tmp);
475 tcg_gen_mov_i32(dest, cpu_NF);
478 /* dest = T0 + T1 + CF. Compute C, N, V and Z flags */
479 static void gen_adc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
481 TCGv_i32 tmp = tcg_temp_new_i32();
482 if (TCG_TARGET_HAS_add2_i32) {
483 tcg_gen_movi_i32(tmp, 0);
484 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, cpu_CF, tmp);
485 tcg_gen_add2_i32(cpu_NF, cpu_CF, cpu_NF, cpu_CF, t1, tmp);
487 TCGv_i64 q0 = tcg_temp_new_i64();
488 TCGv_i64 q1 = tcg_temp_new_i64();
489 tcg_gen_extu_i32_i64(q0, t0);
490 tcg_gen_extu_i32_i64(q1, t1);
491 tcg_gen_add_i64(q0, q0, q1);
492 tcg_gen_extu_i32_i64(q1, cpu_CF);
493 tcg_gen_add_i64(q0, q0, q1);
494 tcg_gen_extr_i64_i32(cpu_NF, cpu_CF, q0);
495 tcg_temp_free_i64(q0);
496 tcg_temp_free_i64(q1);
498 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
499 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
500 tcg_gen_xor_i32(tmp, t0, t1);
501 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
502 tcg_temp_free_i32(tmp);
503 tcg_gen_mov_i32(dest, cpu_NF);
506 /* dest = T0 - T1. Compute C, N, V and Z flags */
507 static void gen_sub_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
510 tcg_gen_sub_i32(cpu_NF, t0, t1);
511 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
512 tcg_gen_setcond_i32(TCG_COND_GEU, cpu_CF, t0, t1);
513 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
514 tmp = tcg_temp_new_i32();
515 tcg_gen_xor_i32(tmp, t0, t1);
516 tcg_gen_and_i32(cpu_VF, cpu_VF, tmp);
517 tcg_temp_free_i32(tmp);
518 tcg_gen_mov_i32(dest, cpu_NF);
521 /* dest = T0 + ~T1 + CF. Compute C, N, V and Z flags */
522 static void gen_sbc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
524 TCGv_i32 tmp = tcg_temp_new_i32();
525 tcg_gen_not_i32(tmp, t1);
526 gen_adc_CC(dest, t0, tmp);
527 tcg_temp_free_i32(tmp);
530 #define GEN_SHIFT(name) \
531 static void gen_##name(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1) \
533 TCGv_i32 tmp1, tmp2, tmp3; \
534 tmp1 = tcg_temp_new_i32(); \
535 tcg_gen_andi_i32(tmp1, t1, 0xff); \
536 tmp2 = tcg_const_i32(0); \
537 tmp3 = tcg_const_i32(0x1f); \
538 tcg_gen_movcond_i32(TCG_COND_GTU, tmp2, tmp1, tmp3, tmp2, t0); \
539 tcg_temp_free_i32(tmp3); \
540 tcg_gen_andi_i32(tmp1, tmp1, 0x1f); \
541 tcg_gen_##name##_i32(dest, tmp2, tmp1); \
542 tcg_temp_free_i32(tmp2); \
543 tcg_temp_free_i32(tmp1); \
549 static void gen_sar(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
552 tmp1 = tcg_temp_new_i32();
553 tcg_gen_andi_i32(tmp1, t1, 0xff);
554 tmp2 = tcg_const_i32(0x1f);
555 tcg_gen_movcond_i32(TCG_COND_GTU, tmp1, tmp1, tmp2, tmp2, tmp1);
556 tcg_temp_free_i32(tmp2);
557 tcg_gen_sar_i32(dest, t0, tmp1);
558 tcg_temp_free_i32(tmp1);
561 static void shifter_out_im(TCGv_i32 var, int shift)
563 tcg_gen_extract_i32(cpu_CF, var, shift, 1);
566 /* Shift by immediate. Includes special handling for shift == 0. */
567 static inline void gen_arm_shift_im(TCGv_i32 var, int shiftop,
568 int shift, int flags)
574 shifter_out_im(var, 32 - shift);
575 tcg_gen_shli_i32(var, var, shift);
581 tcg_gen_shri_i32(cpu_CF, var, 31);
583 tcg_gen_movi_i32(var, 0);
586 shifter_out_im(var, shift - 1);
587 tcg_gen_shri_i32(var, var, shift);
594 shifter_out_im(var, shift - 1);
597 tcg_gen_sari_i32(var, var, shift);
599 case 3: /* ROR/RRX */
602 shifter_out_im(var, shift - 1);
603 tcg_gen_rotri_i32(var, var, shift); break;
605 TCGv_i32 tmp = tcg_temp_new_i32();
606 tcg_gen_shli_i32(tmp, cpu_CF, 31);
608 shifter_out_im(var, 0);
609 tcg_gen_shri_i32(var, var, 1);
610 tcg_gen_or_i32(var, var, tmp);
611 tcg_temp_free_i32(tmp);
616 static inline void gen_arm_shift_reg(TCGv_i32 var, int shiftop,
617 TCGv_i32 shift, int flags)
621 case 0: gen_helper_shl_cc(var, cpu_env, var, shift); break;
622 case 1: gen_helper_shr_cc(var, cpu_env, var, shift); break;
623 case 2: gen_helper_sar_cc(var, cpu_env, var, shift); break;
624 case 3: gen_helper_ror_cc(var, cpu_env, var, shift); break;
629 gen_shl(var, var, shift);
632 gen_shr(var, var, shift);
635 gen_sar(var, var, shift);
637 case 3: tcg_gen_andi_i32(shift, shift, 0x1f);
638 tcg_gen_rotr_i32(var, var, shift); break;
641 tcg_temp_free_i32(shift);
645 * Generate a conditional based on ARM condition code cc.
646 * This is common between ARM and Aarch64 targets.
648 void arm_test_cc(DisasCompare *cmp, int cc)
679 case 8: /* hi: C && !Z */
680 case 9: /* ls: !C || Z -> !(C && !Z) */
682 value = tcg_temp_new_i32();
684 /* CF is 1 for C, so -CF is an all-bits-set mask for C;
685 ZF is non-zero for !Z; so AND the two subexpressions. */
686 tcg_gen_neg_i32(value, cpu_CF);
687 tcg_gen_and_i32(value, value, cpu_ZF);
690 case 10: /* ge: N == V -> N ^ V == 0 */
691 case 11: /* lt: N != V -> N ^ V != 0 */
692 /* Since we're only interested in the sign bit, == 0 is >= 0. */
694 value = tcg_temp_new_i32();
696 tcg_gen_xor_i32(value, cpu_VF, cpu_NF);
699 case 12: /* gt: !Z && N == V */
700 case 13: /* le: Z || N != V */
702 value = tcg_temp_new_i32();
704 /* (N == V) is equal to the sign bit of ~(NF ^ VF). Propagate
705 * the sign bit then AND with ZF to yield the result. */
706 tcg_gen_xor_i32(value, cpu_VF, cpu_NF);
707 tcg_gen_sari_i32(value, value, 31);
708 tcg_gen_andc_i32(value, cpu_ZF, value);
711 case 14: /* always */
712 case 15: /* always */
713 /* Use the ALWAYS condition, which will fold early.
714 * It doesn't matter what we use for the value. */
715 cond = TCG_COND_ALWAYS;
720 fprintf(stderr, "Bad condition code 0x%x\n", cc);
725 cond = tcg_invert_cond(cond);
731 cmp->value_global = global;
734 void arm_free_cc(DisasCompare *cmp)
736 if (!cmp->value_global) {
737 tcg_temp_free_i32(cmp->value);
741 void arm_jump_cc(DisasCompare *cmp, TCGLabel *label)
743 tcg_gen_brcondi_i32(cmp->cond, cmp->value, 0, label);
746 void arm_gen_test_cc(int cc, TCGLabel *label)
749 arm_test_cc(&cmp, cc);
750 arm_jump_cc(&cmp, label);
754 static inline void gen_set_condexec(DisasContext *s)
756 if (s->condexec_mask) {
757 uint32_t val = (s->condexec_cond << 4) | (s->condexec_mask >> 1);
758 TCGv_i32 tmp = tcg_temp_new_i32();
759 tcg_gen_movi_i32(tmp, val);
760 store_cpu_field(tmp, condexec_bits);
764 static inline void gen_set_pc_im(DisasContext *s, target_ulong val)
766 tcg_gen_movi_i32(cpu_R[15], val);
769 /* Set PC and Thumb state from var. var is marked as dead. */
770 static inline void gen_bx(DisasContext *s, TCGv_i32 var)
772 s->base.is_jmp = DISAS_JUMP;
773 tcg_gen_andi_i32(cpu_R[15], var, ~1);
774 tcg_gen_andi_i32(var, var, 1);
775 store_cpu_field(var, thumb);
779 * Set PC and Thumb state from var. var is marked as dead.
780 * For M-profile CPUs, include logic to detect exception-return
781 * branches and handle them. This is needed for Thumb POP/LDM to PC, LDR to PC,
782 * and BX reg, and no others, and happens only for code in Handler mode.
783 * The Security Extension also requires us to check for the FNC_RETURN
784 * which signals a function return from non-secure state; this can happen
785 * in both Handler and Thread mode.
786 * To avoid having to do multiple comparisons in inline generated code,
787 * we make the check we do here loose, so it will match for EXC_RETURN
788 * in Thread mode. For system emulation do_v7m_exception_exit() checks
789 * for these spurious cases and returns without doing anything (giving
790 * the same behaviour as for a branch to a non-magic address).
792 * In linux-user mode it is unclear what the right behaviour for an
793 * attempted FNC_RETURN should be, because in real hardware this will go
794 * directly to Secure code (ie not the Linux kernel) which will then treat
795 * the error in any way it chooses. For QEMU we opt to make the FNC_RETURN
796 * attempt behave the way it would on a CPU without the security extension,
797 * which is to say "like a normal branch". That means we can simply treat
798 * all branches as normal with no magic address behaviour.
800 static inline void gen_bx_excret(DisasContext *s, TCGv_i32 var)
802 /* Generate the same code here as for a simple bx, but flag via
803 * s->base.is_jmp that we need to do the rest of the work later.
806 #ifndef CONFIG_USER_ONLY
807 if (arm_dc_feature(s, ARM_FEATURE_M_SECURITY) ||
808 (s->v7m_handler_mode && arm_dc_feature(s, ARM_FEATURE_M))) {
809 s->base.is_jmp = DISAS_BX_EXCRET;
814 static inline void gen_bx_excret_final_code(DisasContext *s)
816 /* Generate the code to finish possible exception return and end the TB */
817 TCGLabel *excret_label = gen_new_label();
820 if (arm_dc_feature(s, ARM_FEATURE_M_SECURITY)) {
821 /* Covers FNC_RETURN and EXC_RETURN magic */
822 min_magic = FNC_RETURN_MIN_MAGIC;
824 /* EXC_RETURN magic only */
825 min_magic = EXC_RETURN_MIN_MAGIC;
828 /* Is the new PC value in the magic range indicating exception return? */
829 tcg_gen_brcondi_i32(TCG_COND_GEU, cpu_R[15], min_magic, excret_label);
830 /* No: end the TB as we would for a DISAS_JMP */
831 if (is_singlestepping(s)) {
832 gen_singlestep_exception(s);
834 tcg_gen_exit_tb(NULL, 0);
836 gen_set_label(excret_label);
837 /* Yes: this is an exception return.
838 * At this point in runtime env->regs[15] and env->thumb will hold
839 * the exception-return magic number, which do_v7m_exception_exit()
840 * will read. Nothing else will be able to see those values because
841 * the cpu-exec main loop guarantees that we will always go straight
842 * from raising the exception to the exception-handling code.
844 * gen_ss_advance(s) does nothing on M profile currently but
845 * calling it is conceptually the right thing as we have executed
846 * this instruction (compare SWI, HVC, SMC handling).
849 gen_exception_internal(EXCP_EXCEPTION_EXIT);
852 static inline void gen_bxns(DisasContext *s, int rm)
854 TCGv_i32 var = load_reg(s, rm);
856 /* The bxns helper may raise an EXCEPTION_EXIT exception, so in theory
857 * we need to sync state before calling it, but:
858 * - we don't need to do gen_set_pc_im() because the bxns helper will
859 * always set the PC itself
860 * - we don't need to do gen_set_condexec() because BXNS is UNPREDICTABLE
861 * unless it's outside an IT block or the last insn in an IT block,
862 * so we know that condexec == 0 (already set at the top of the TB)
863 * is correct in the non-UNPREDICTABLE cases, and we can choose
864 * "zeroes the IT bits" as our UNPREDICTABLE behaviour otherwise.
866 gen_helper_v7m_bxns(cpu_env, var);
867 tcg_temp_free_i32(var);
868 s->base.is_jmp = DISAS_EXIT;
871 static inline void gen_blxns(DisasContext *s, int rm)
873 TCGv_i32 var = load_reg(s, rm);
875 /* We don't need to sync condexec state, for the same reason as bxns.
876 * We do however need to set the PC, because the blxns helper reads it.
877 * The blxns helper may throw an exception.
879 gen_set_pc_im(s, s->base.pc_next);
880 gen_helper_v7m_blxns(cpu_env, var);
881 tcg_temp_free_i32(var);
882 s->base.is_jmp = DISAS_EXIT;
885 /* Variant of store_reg which uses branch&exchange logic when storing
886 to r15 in ARM architecture v7 and above. The source must be a temporary
887 and will be marked as dead. */
888 static inline void store_reg_bx(DisasContext *s, int reg, TCGv_i32 var)
890 if (reg == 15 && ENABLE_ARCH_7) {
893 store_reg(s, reg, var);
897 /* Variant of store_reg which uses branch&exchange logic when storing
898 * to r15 in ARM architecture v5T and above. This is used for storing
899 * the results of a LDR/LDM/POP into r15, and corresponds to the cases
900 * in the ARM ARM which use the LoadWritePC() pseudocode function. */
901 static inline void store_reg_from_load(DisasContext *s, int reg, TCGv_i32 var)
903 if (reg == 15 && ENABLE_ARCH_5) {
904 gen_bx_excret(s, var);
906 store_reg(s, reg, var);
910 #ifdef CONFIG_USER_ONLY
911 #define IS_USER_ONLY 1
913 #define IS_USER_ONLY 0
916 /* Abstractions of "generate code to do a guest load/store for
917 * AArch32", where a vaddr is always 32 bits (and is zero
918 * extended if we're a 64 bit core) and data is also
919 * 32 bits unless specifically doing a 64 bit access.
920 * These functions work like tcg_gen_qemu_{ld,st}* except
921 * that the address argument is TCGv_i32 rather than TCGv.
924 static inline TCGv gen_aa32_addr(DisasContext *s, TCGv_i32 a32, MemOp op)
926 TCGv addr = tcg_temp_new();
927 tcg_gen_extu_i32_tl(addr, a32);
929 /* Not needed for user-mode BE32, where we use MO_BE instead. */
930 if (!IS_USER_ONLY && s->sctlr_b && (op & MO_SIZE) < MO_32) {
931 tcg_gen_xori_tl(addr, addr, 4 - (1 << (op & MO_SIZE)));
936 static void gen_aa32_ld_i32(DisasContext *s, TCGv_i32 val, TCGv_i32 a32,
937 int index, MemOp opc)
941 if (arm_dc_feature(s, ARM_FEATURE_M) &&
942 !arm_dc_feature(s, ARM_FEATURE_M_MAIN)) {
946 addr = gen_aa32_addr(s, a32, opc);
947 tcg_gen_qemu_ld_i32(val, addr, index, opc);
951 static void gen_aa32_st_i32(DisasContext *s, TCGv_i32 val, TCGv_i32 a32,
952 int index, MemOp opc)
956 if (arm_dc_feature(s, ARM_FEATURE_M) &&
957 !arm_dc_feature(s, ARM_FEATURE_M_MAIN)) {
961 addr = gen_aa32_addr(s, a32, opc);
962 tcg_gen_qemu_st_i32(val, addr, index, opc);
966 #define DO_GEN_LD(SUFF, OPC) \
967 static inline void gen_aa32_ld##SUFF(DisasContext *s, TCGv_i32 val, \
968 TCGv_i32 a32, int index) \
970 gen_aa32_ld_i32(s, val, a32, index, OPC | s->be_data); \
973 #define DO_GEN_ST(SUFF, OPC) \
974 static inline void gen_aa32_st##SUFF(DisasContext *s, TCGv_i32 val, \
975 TCGv_i32 a32, int index) \
977 gen_aa32_st_i32(s, val, a32, index, OPC | s->be_data); \
980 static inline void gen_aa32_frob64(DisasContext *s, TCGv_i64 val)
982 /* Not needed for user-mode BE32, where we use MO_BE instead. */
983 if (!IS_USER_ONLY && s->sctlr_b) {
984 tcg_gen_rotri_i64(val, val, 32);
988 static void gen_aa32_ld_i64(DisasContext *s, TCGv_i64 val, TCGv_i32 a32,
989 int index, MemOp opc)
991 TCGv addr = gen_aa32_addr(s, a32, opc);
992 tcg_gen_qemu_ld_i64(val, addr, index, opc);
993 gen_aa32_frob64(s, val);
997 static inline void gen_aa32_ld64(DisasContext *s, TCGv_i64 val,
998 TCGv_i32 a32, int index)
1000 gen_aa32_ld_i64(s, val, a32, index, MO_Q | s->be_data);
1003 static void gen_aa32_st_i64(DisasContext *s, TCGv_i64 val, TCGv_i32 a32,
1004 int index, MemOp opc)
1006 TCGv addr = gen_aa32_addr(s, a32, opc);
1008 /* Not needed for user-mode BE32, where we use MO_BE instead. */
1009 if (!IS_USER_ONLY && s->sctlr_b) {
1010 TCGv_i64 tmp = tcg_temp_new_i64();
1011 tcg_gen_rotri_i64(tmp, val, 32);
1012 tcg_gen_qemu_st_i64(tmp, addr, index, opc);
1013 tcg_temp_free_i64(tmp);
1015 tcg_gen_qemu_st_i64(val, addr, index, opc);
1017 tcg_temp_free(addr);
1020 static inline void gen_aa32_st64(DisasContext *s, TCGv_i64 val,
1021 TCGv_i32 a32, int index)
1023 gen_aa32_st_i64(s, val, a32, index, MO_Q | s->be_data);
1026 DO_GEN_LD(8u, MO_UB)
1027 DO_GEN_LD(16u, MO_UW)
1028 DO_GEN_LD(32u, MO_UL)
1030 DO_GEN_ST(16, MO_UW)
1031 DO_GEN_ST(32, MO_UL)
1033 static inline void gen_hvc(DisasContext *s, int imm16)
1035 /* The pre HVC helper handles cases when HVC gets trapped
1036 * as an undefined insn by runtime configuration (ie before
1037 * the insn really executes).
1039 gen_set_pc_im(s, s->pc_curr);
1040 gen_helper_pre_hvc(cpu_env);
1041 /* Otherwise we will treat this as a real exception which
1042 * happens after execution of the insn. (The distinction matters
1043 * for the PC value reported to the exception handler and also
1044 * for single stepping.)
1047 gen_set_pc_im(s, s->base.pc_next);
1048 s->base.is_jmp = DISAS_HVC;
1051 static inline void gen_smc(DisasContext *s)
1053 /* As with HVC, we may take an exception either before or after
1054 * the insn executes.
1058 gen_set_pc_im(s, s->pc_curr);
1059 tmp = tcg_const_i32(syn_aa32_smc());
1060 gen_helper_pre_smc(cpu_env, tmp);
1061 tcg_temp_free_i32(tmp);
1062 gen_set_pc_im(s, s->base.pc_next);
1063 s->base.is_jmp = DISAS_SMC;
1066 static void gen_exception_internal_insn(DisasContext *s, uint32_t pc, int excp)
1068 gen_set_condexec(s);
1069 gen_set_pc_im(s, pc);
1070 gen_exception_internal(excp);
1071 s->base.is_jmp = DISAS_NORETURN;
1074 static void gen_exception_insn(DisasContext *s, uint32_t pc, int excp,
1075 int syn, uint32_t target_el)
1077 gen_set_condexec(s);
1078 gen_set_pc_im(s, pc);
1079 gen_exception(excp, syn, target_el);
1080 s->base.is_jmp = DISAS_NORETURN;
1083 static void gen_exception_bkpt_insn(DisasContext *s, uint32_t syn)
1087 gen_set_condexec(s);
1088 gen_set_pc_im(s, s->pc_curr);
1089 tcg_syn = tcg_const_i32(syn);
1090 gen_helper_exception_bkpt_insn(cpu_env, tcg_syn);
1091 tcg_temp_free_i32(tcg_syn);
1092 s->base.is_jmp = DISAS_NORETURN;
1095 static void unallocated_encoding(DisasContext *s)
1097 /* Unallocated and reserved encodings are uncategorized */
1098 gen_exception_insn(s, s->pc_curr, EXCP_UDEF, syn_uncategorized(),
1099 default_exception_el(s));
1102 /* Force a TB lookup after an instruction that changes the CPU state. */
1103 static inline void gen_lookup_tb(DisasContext *s)
1105 tcg_gen_movi_i32(cpu_R[15], s->base.pc_next);
1106 s->base.is_jmp = DISAS_EXIT;
1109 static inline void gen_hlt(DisasContext *s, int imm)
1111 /* HLT. This has two purposes.
1112 * Architecturally, it is an external halting debug instruction.
1113 * Since QEMU doesn't implement external debug, we treat this as
1114 * it is required for halting debug disabled: it will UNDEF.
1115 * Secondly, "HLT 0x3C" is a T32 semihosting trap instruction,
1116 * and "HLT 0xF000" is an A32 semihosting syscall. These traps
1117 * must trigger semihosting even for ARMv7 and earlier, where
1118 * HLT was an undefined encoding.
1119 * In system mode, we don't allow userspace access to
1120 * semihosting, to provide some semblance of security
1121 * (and for consistency with our 32-bit semihosting).
1123 if (semihosting_enabled() &&
1124 #ifndef CONFIG_USER_ONLY
1125 s->current_el != 0 &&
1127 (imm == (s->thumb ? 0x3c : 0xf000))) {
1128 gen_exception_internal_insn(s, s->pc_curr, EXCP_SEMIHOST);
1132 unallocated_encoding(s);
1135 static TCGv_ptr get_fpstatus_ptr(int neon)
1137 TCGv_ptr statusptr = tcg_temp_new_ptr();
1140 offset = offsetof(CPUARMState, vfp.standard_fp_status);
1142 offset = offsetof(CPUARMState, vfp.fp_status);
1144 tcg_gen_addi_ptr(statusptr, cpu_env, offset);
1148 static inline long vfp_reg_offset(bool dp, unsigned reg)
1151 return offsetof(CPUARMState, vfp.zregs[reg >> 1].d[reg & 1]);
1153 long ofs = offsetof(CPUARMState, vfp.zregs[reg >> 2].d[(reg >> 1) & 1]);
1155 ofs += offsetof(CPU_DoubleU, l.upper);
1157 ofs += offsetof(CPU_DoubleU, l.lower);
1163 /* Return the offset of a 32-bit piece of a NEON register.
1164 zero is the least significant end of the register. */
1166 neon_reg_offset (int reg, int n)
1170 return vfp_reg_offset(0, sreg);
1173 /* Return the offset of a 2**SIZE piece of a NEON register, at index ELE,
1174 * where 0 is the least significant end of the register.
1177 neon_element_offset(int reg, int element, MemOp size)
1179 int element_size = 1 << size;
1180 int ofs = element * element_size;
1181 #ifdef HOST_WORDS_BIGENDIAN
1182 /* Calculate the offset assuming fully little-endian,
1183 * then XOR to account for the order of the 8-byte units.
1185 if (element_size < 8) {
1186 ofs ^= 8 - element_size;
1189 return neon_reg_offset(reg, 0) + ofs;
1192 static TCGv_i32 neon_load_reg(int reg, int pass)
1194 TCGv_i32 tmp = tcg_temp_new_i32();
1195 tcg_gen_ld_i32(tmp, cpu_env, neon_reg_offset(reg, pass));
1199 static void neon_load_element(TCGv_i32 var, int reg, int ele, MemOp mop)
1201 long offset = neon_element_offset(reg, ele, mop & MO_SIZE);
1205 tcg_gen_ld8u_i32(var, cpu_env, offset);
1208 tcg_gen_ld16u_i32(var, cpu_env, offset);
1211 tcg_gen_ld_i32(var, cpu_env, offset);
1214 g_assert_not_reached();
1218 static void neon_load_element64(TCGv_i64 var, int reg, int ele, MemOp mop)
1220 long offset = neon_element_offset(reg, ele, mop & MO_SIZE);
1224 tcg_gen_ld8u_i64(var, cpu_env, offset);
1227 tcg_gen_ld16u_i64(var, cpu_env, offset);
1230 tcg_gen_ld32u_i64(var, cpu_env, offset);
1233 tcg_gen_ld_i64(var, cpu_env, offset);
1236 g_assert_not_reached();
1240 static void neon_store_reg(int reg, int pass, TCGv_i32 var)
1242 tcg_gen_st_i32(var, cpu_env, neon_reg_offset(reg, pass));
1243 tcg_temp_free_i32(var);
1246 static void neon_store_element(int reg, int ele, MemOp size, TCGv_i32 var)
1248 long offset = neon_element_offset(reg, ele, size);
1252 tcg_gen_st8_i32(var, cpu_env, offset);
1255 tcg_gen_st16_i32(var, cpu_env, offset);
1258 tcg_gen_st_i32(var, cpu_env, offset);
1261 g_assert_not_reached();
1265 static void neon_store_element64(int reg, int ele, MemOp size, TCGv_i64 var)
1267 long offset = neon_element_offset(reg, ele, size);
1271 tcg_gen_st8_i64(var, cpu_env, offset);
1274 tcg_gen_st16_i64(var, cpu_env, offset);
1277 tcg_gen_st32_i64(var, cpu_env, offset);
1280 tcg_gen_st_i64(var, cpu_env, offset);
1283 g_assert_not_reached();
1287 static inline void neon_load_reg64(TCGv_i64 var, int reg)
1289 tcg_gen_ld_i64(var, cpu_env, vfp_reg_offset(1, reg));
1292 static inline void neon_store_reg64(TCGv_i64 var, int reg)
1294 tcg_gen_st_i64(var, cpu_env, vfp_reg_offset(1, reg));
1297 static inline void neon_load_reg32(TCGv_i32 var, int reg)
1299 tcg_gen_ld_i32(var, cpu_env, vfp_reg_offset(false, reg));
1302 static inline void neon_store_reg32(TCGv_i32 var, int reg)
1304 tcg_gen_st_i32(var, cpu_env, vfp_reg_offset(false, reg));
1307 static TCGv_ptr vfp_reg_ptr(bool dp, int reg)
1309 TCGv_ptr ret = tcg_temp_new_ptr();
1310 tcg_gen_addi_ptr(ret, cpu_env, vfp_reg_offset(dp, reg));
1314 #define ARM_CP_RW_BIT (1 << 20)
1316 /* Include the VFP and Neon decoders */
1317 #include "translate-vfp.inc.c"
1318 #include "translate-neon.inc.c"
1320 static inline void iwmmxt_load_reg(TCGv_i64 var, int reg)
1322 tcg_gen_ld_i64(var, cpu_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1325 static inline void iwmmxt_store_reg(TCGv_i64 var, int reg)
1327 tcg_gen_st_i64(var, cpu_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1330 static inline TCGv_i32 iwmmxt_load_creg(int reg)
1332 TCGv_i32 var = tcg_temp_new_i32();
1333 tcg_gen_ld_i32(var, cpu_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1337 static inline void iwmmxt_store_creg(int reg, TCGv_i32 var)
1339 tcg_gen_st_i32(var, cpu_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1340 tcg_temp_free_i32(var);
1343 static inline void gen_op_iwmmxt_movq_wRn_M0(int rn)
1345 iwmmxt_store_reg(cpu_M0, rn);
1348 static inline void gen_op_iwmmxt_movq_M0_wRn(int rn)
1350 iwmmxt_load_reg(cpu_M0, rn);
1353 static inline void gen_op_iwmmxt_orq_M0_wRn(int rn)
1355 iwmmxt_load_reg(cpu_V1, rn);
1356 tcg_gen_or_i64(cpu_M0, cpu_M0, cpu_V1);
1359 static inline void gen_op_iwmmxt_andq_M0_wRn(int rn)
1361 iwmmxt_load_reg(cpu_V1, rn);
1362 tcg_gen_and_i64(cpu_M0, cpu_M0, cpu_V1);
1365 static inline void gen_op_iwmmxt_xorq_M0_wRn(int rn)
1367 iwmmxt_load_reg(cpu_V1, rn);
1368 tcg_gen_xor_i64(cpu_M0, cpu_M0, cpu_V1);
1371 #define IWMMXT_OP(name) \
1372 static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1374 iwmmxt_load_reg(cpu_V1, rn); \
1375 gen_helper_iwmmxt_##name(cpu_M0, cpu_M0, cpu_V1); \
1378 #define IWMMXT_OP_ENV(name) \
1379 static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1381 iwmmxt_load_reg(cpu_V1, rn); \
1382 gen_helper_iwmmxt_##name(cpu_M0, cpu_env, cpu_M0, cpu_V1); \
1385 #define IWMMXT_OP_ENV_SIZE(name) \
1386 IWMMXT_OP_ENV(name##b) \
1387 IWMMXT_OP_ENV(name##w) \
1388 IWMMXT_OP_ENV(name##l)
1390 #define IWMMXT_OP_ENV1(name) \
1391 static inline void gen_op_iwmmxt_##name##_M0(void) \
1393 gen_helper_iwmmxt_##name(cpu_M0, cpu_env, cpu_M0); \
1407 IWMMXT_OP_ENV_SIZE(unpackl)
1408 IWMMXT_OP_ENV_SIZE(unpackh)
1410 IWMMXT_OP_ENV1(unpacklub)
1411 IWMMXT_OP_ENV1(unpackluw)
1412 IWMMXT_OP_ENV1(unpacklul)
1413 IWMMXT_OP_ENV1(unpackhub)
1414 IWMMXT_OP_ENV1(unpackhuw)
1415 IWMMXT_OP_ENV1(unpackhul)
1416 IWMMXT_OP_ENV1(unpacklsb)
1417 IWMMXT_OP_ENV1(unpacklsw)
1418 IWMMXT_OP_ENV1(unpacklsl)
1419 IWMMXT_OP_ENV1(unpackhsb)
1420 IWMMXT_OP_ENV1(unpackhsw)
1421 IWMMXT_OP_ENV1(unpackhsl)
1423 IWMMXT_OP_ENV_SIZE(cmpeq)
1424 IWMMXT_OP_ENV_SIZE(cmpgtu)
1425 IWMMXT_OP_ENV_SIZE(cmpgts)
1427 IWMMXT_OP_ENV_SIZE(mins)
1428 IWMMXT_OP_ENV_SIZE(minu)
1429 IWMMXT_OP_ENV_SIZE(maxs)
1430 IWMMXT_OP_ENV_SIZE(maxu)
1432 IWMMXT_OP_ENV_SIZE(subn)
1433 IWMMXT_OP_ENV_SIZE(addn)
1434 IWMMXT_OP_ENV_SIZE(subu)
1435 IWMMXT_OP_ENV_SIZE(addu)
1436 IWMMXT_OP_ENV_SIZE(subs)
1437 IWMMXT_OP_ENV_SIZE(adds)
1439 IWMMXT_OP_ENV(avgb0)
1440 IWMMXT_OP_ENV(avgb1)
1441 IWMMXT_OP_ENV(avgw0)
1442 IWMMXT_OP_ENV(avgw1)
1444 IWMMXT_OP_ENV(packuw)
1445 IWMMXT_OP_ENV(packul)
1446 IWMMXT_OP_ENV(packuq)
1447 IWMMXT_OP_ENV(packsw)
1448 IWMMXT_OP_ENV(packsl)
1449 IWMMXT_OP_ENV(packsq)
1451 static void gen_op_iwmmxt_set_mup(void)
1454 tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1455 tcg_gen_ori_i32(tmp, tmp, 2);
1456 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1459 static void gen_op_iwmmxt_set_cup(void)
1462 tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1463 tcg_gen_ori_i32(tmp, tmp, 1);
1464 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1467 static void gen_op_iwmmxt_setpsr_nz(void)
1469 TCGv_i32 tmp = tcg_temp_new_i32();
1470 gen_helper_iwmmxt_setpsr_nz(tmp, cpu_M0);
1471 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCASF]);
1474 static inline void gen_op_iwmmxt_addl_M0_wRn(int rn)
1476 iwmmxt_load_reg(cpu_V1, rn);
1477 tcg_gen_ext32u_i64(cpu_V1, cpu_V1);
1478 tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
1481 static inline int gen_iwmmxt_address(DisasContext *s, uint32_t insn,
1488 rd = (insn >> 16) & 0xf;
1489 tmp = load_reg(s, rd);
1491 offset = (insn & 0xff) << ((insn >> 7) & 2);
1492 if (insn & (1 << 24)) {
1494 if (insn & (1 << 23))
1495 tcg_gen_addi_i32(tmp, tmp, offset);
1497 tcg_gen_addi_i32(tmp, tmp, -offset);
1498 tcg_gen_mov_i32(dest, tmp);
1499 if (insn & (1 << 21))
1500 store_reg(s, rd, tmp);
1502 tcg_temp_free_i32(tmp);
1503 } else if (insn & (1 << 21)) {
1505 tcg_gen_mov_i32(dest, tmp);
1506 if (insn & (1 << 23))
1507 tcg_gen_addi_i32(tmp, tmp, offset);
1509 tcg_gen_addi_i32(tmp, tmp, -offset);
1510 store_reg(s, rd, tmp);
1511 } else if (!(insn & (1 << 23)))
1516 static inline int gen_iwmmxt_shift(uint32_t insn, uint32_t mask, TCGv_i32 dest)
1518 int rd = (insn >> 0) & 0xf;
1521 if (insn & (1 << 8)) {
1522 if (rd < ARM_IWMMXT_wCGR0 || rd > ARM_IWMMXT_wCGR3) {
1525 tmp = iwmmxt_load_creg(rd);
1528 tmp = tcg_temp_new_i32();
1529 iwmmxt_load_reg(cpu_V0, rd);
1530 tcg_gen_extrl_i64_i32(tmp, cpu_V0);
1532 tcg_gen_andi_i32(tmp, tmp, mask);
1533 tcg_gen_mov_i32(dest, tmp);
1534 tcg_temp_free_i32(tmp);
1538 /* Disassemble an iwMMXt instruction. Returns nonzero if an error occurred
1539 (ie. an undefined instruction). */
1540 static int disas_iwmmxt_insn(DisasContext *s, uint32_t insn)
1543 int rdhi, rdlo, rd0, rd1, i;
1545 TCGv_i32 tmp, tmp2, tmp3;
1547 if ((insn & 0x0e000e00) == 0x0c000000) {
1548 if ((insn & 0x0fe00ff0) == 0x0c400000) {
1550 rdlo = (insn >> 12) & 0xf;
1551 rdhi = (insn >> 16) & 0xf;
1552 if (insn & ARM_CP_RW_BIT) { /* TMRRC */
1553 iwmmxt_load_reg(cpu_V0, wrd);
1554 tcg_gen_extrl_i64_i32(cpu_R[rdlo], cpu_V0);
1555 tcg_gen_extrh_i64_i32(cpu_R[rdhi], cpu_V0);
1556 } else { /* TMCRR */
1557 tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
1558 iwmmxt_store_reg(cpu_V0, wrd);
1559 gen_op_iwmmxt_set_mup();
1564 wrd = (insn >> 12) & 0xf;
1565 addr = tcg_temp_new_i32();
1566 if (gen_iwmmxt_address(s, insn, addr)) {
1567 tcg_temp_free_i32(addr);
1570 if (insn & ARM_CP_RW_BIT) {
1571 if ((insn >> 28) == 0xf) { /* WLDRW wCx */
1572 tmp = tcg_temp_new_i32();
1573 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
1574 iwmmxt_store_creg(wrd, tmp);
1577 if (insn & (1 << 8)) {
1578 if (insn & (1 << 22)) { /* WLDRD */
1579 gen_aa32_ld64(s, cpu_M0, addr, get_mem_index(s));
1581 } else { /* WLDRW wRd */
1582 tmp = tcg_temp_new_i32();
1583 gen_aa32_ld32u(s, tmp, addr, get_mem_index(s));
1586 tmp = tcg_temp_new_i32();
1587 if (insn & (1 << 22)) { /* WLDRH */
1588 gen_aa32_ld16u(s, tmp, addr, get_mem_index(s));
1589 } else { /* WLDRB */
1590 gen_aa32_ld8u(s, tmp, addr, get_mem_index(s));
1594 tcg_gen_extu_i32_i64(cpu_M0, tmp);
1595 tcg_temp_free_i32(tmp);
1597 gen_op_iwmmxt_movq_wRn_M0(wrd);
1600 if ((insn >> 28) == 0xf) { /* WSTRW wCx */
1601 tmp = iwmmxt_load_creg(wrd);
1602 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
1604 gen_op_iwmmxt_movq_M0_wRn(wrd);
1605 tmp = tcg_temp_new_i32();
1606 if (insn & (1 << 8)) {
1607 if (insn & (1 << 22)) { /* WSTRD */
1608 gen_aa32_st64(s, cpu_M0, addr, get_mem_index(s));
1609 } else { /* WSTRW wRd */
1610 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1611 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
1614 if (insn & (1 << 22)) { /* WSTRH */
1615 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1616 gen_aa32_st16(s, tmp, addr, get_mem_index(s));
1617 } else { /* WSTRB */
1618 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1619 gen_aa32_st8(s, tmp, addr, get_mem_index(s));
1623 tcg_temp_free_i32(tmp);
1625 tcg_temp_free_i32(addr);
1629 if ((insn & 0x0f000000) != 0x0e000000)
1632 switch (((insn >> 12) & 0xf00) | ((insn >> 4) & 0xff)) {
1633 case 0x000: /* WOR */
1634 wrd = (insn >> 12) & 0xf;
1635 rd0 = (insn >> 0) & 0xf;
1636 rd1 = (insn >> 16) & 0xf;
1637 gen_op_iwmmxt_movq_M0_wRn(rd0);
1638 gen_op_iwmmxt_orq_M0_wRn(rd1);
1639 gen_op_iwmmxt_setpsr_nz();
1640 gen_op_iwmmxt_movq_wRn_M0(wrd);
1641 gen_op_iwmmxt_set_mup();
1642 gen_op_iwmmxt_set_cup();
1644 case 0x011: /* TMCR */
1647 rd = (insn >> 12) & 0xf;
1648 wrd = (insn >> 16) & 0xf;
1650 case ARM_IWMMXT_wCID:
1651 case ARM_IWMMXT_wCASF:
1653 case ARM_IWMMXT_wCon:
1654 gen_op_iwmmxt_set_cup();
1656 case ARM_IWMMXT_wCSSF:
1657 tmp = iwmmxt_load_creg(wrd);
1658 tmp2 = load_reg(s, rd);
1659 tcg_gen_andc_i32(tmp, tmp, tmp2);
1660 tcg_temp_free_i32(tmp2);
1661 iwmmxt_store_creg(wrd, tmp);
1663 case ARM_IWMMXT_wCGR0:
1664 case ARM_IWMMXT_wCGR1:
1665 case ARM_IWMMXT_wCGR2:
1666 case ARM_IWMMXT_wCGR3:
1667 gen_op_iwmmxt_set_cup();
1668 tmp = load_reg(s, rd);
1669 iwmmxt_store_creg(wrd, tmp);
1675 case 0x100: /* WXOR */
1676 wrd = (insn >> 12) & 0xf;
1677 rd0 = (insn >> 0) & 0xf;
1678 rd1 = (insn >> 16) & 0xf;
1679 gen_op_iwmmxt_movq_M0_wRn(rd0);
1680 gen_op_iwmmxt_xorq_M0_wRn(rd1);
1681 gen_op_iwmmxt_setpsr_nz();
1682 gen_op_iwmmxt_movq_wRn_M0(wrd);
1683 gen_op_iwmmxt_set_mup();
1684 gen_op_iwmmxt_set_cup();
1686 case 0x111: /* TMRC */
1689 rd = (insn >> 12) & 0xf;
1690 wrd = (insn >> 16) & 0xf;
1691 tmp = iwmmxt_load_creg(wrd);
1692 store_reg(s, rd, tmp);
1694 case 0x300: /* WANDN */
1695 wrd = (insn >> 12) & 0xf;
1696 rd0 = (insn >> 0) & 0xf;
1697 rd1 = (insn >> 16) & 0xf;
1698 gen_op_iwmmxt_movq_M0_wRn(rd0);
1699 tcg_gen_neg_i64(cpu_M0, cpu_M0);
1700 gen_op_iwmmxt_andq_M0_wRn(rd1);
1701 gen_op_iwmmxt_setpsr_nz();
1702 gen_op_iwmmxt_movq_wRn_M0(wrd);
1703 gen_op_iwmmxt_set_mup();
1704 gen_op_iwmmxt_set_cup();
1706 case 0x200: /* WAND */
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_andq_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();
1717 case 0x810: case 0xa10: /* WMADD */
1718 wrd = (insn >> 12) & 0xf;
1719 rd0 = (insn >> 0) & 0xf;
1720 rd1 = (insn >> 16) & 0xf;
1721 gen_op_iwmmxt_movq_M0_wRn(rd0);
1722 if (insn & (1 << 21))
1723 gen_op_iwmmxt_maddsq_M0_wRn(rd1);
1725 gen_op_iwmmxt_madduq_M0_wRn(rd1);
1726 gen_op_iwmmxt_movq_wRn_M0(wrd);
1727 gen_op_iwmmxt_set_mup();
1729 case 0x10e: case 0x50e: case 0x90e: case 0xd0e: /* WUNPCKIL */
1730 wrd = (insn >> 12) & 0xf;
1731 rd0 = (insn >> 16) & 0xf;
1732 rd1 = (insn >> 0) & 0xf;
1733 gen_op_iwmmxt_movq_M0_wRn(rd0);
1734 switch ((insn >> 22) & 3) {
1736 gen_op_iwmmxt_unpacklb_M0_wRn(rd1);
1739 gen_op_iwmmxt_unpacklw_M0_wRn(rd1);
1742 gen_op_iwmmxt_unpackll_M0_wRn(rd1);
1747 gen_op_iwmmxt_movq_wRn_M0(wrd);
1748 gen_op_iwmmxt_set_mup();
1749 gen_op_iwmmxt_set_cup();
1751 case 0x10c: case 0x50c: case 0x90c: case 0xd0c: /* WUNPCKIH */
1752 wrd = (insn >> 12) & 0xf;
1753 rd0 = (insn >> 16) & 0xf;
1754 rd1 = (insn >> 0) & 0xf;
1755 gen_op_iwmmxt_movq_M0_wRn(rd0);
1756 switch ((insn >> 22) & 3) {
1758 gen_op_iwmmxt_unpackhb_M0_wRn(rd1);
1761 gen_op_iwmmxt_unpackhw_M0_wRn(rd1);
1764 gen_op_iwmmxt_unpackhl_M0_wRn(rd1);
1769 gen_op_iwmmxt_movq_wRn_M0(wrd);
1770 gen_op_iwmmxt_set_mup();
1771 gen_op_iwmmxt_set_cup();
1773 case 0x012: case 0x112: case 0x412: case 0x512: /* WSAD */
1774 wrd = (insn >> 12) & 0xf;
1775 rd0 = (insn >> 16) & 0xf;
1776 rd1 = (insn >> 0) & 0xf;
1777 gen_op_iwmmxt_movq_M0_wRn(rd0);
1778 if (insn & (1 << 22))
1779 gen_op_iwmmxt_sadw_M0_wRn(rd1);
1781 gen_op_iwmmxt_sadb_M0_wRn(rd1);
1782 if (!(insn & (1 << 20)))
1783 gen_op_iwmmxt_addl_M0_wRn(wrd);
1784 gen_op_iwmmxt_movq_wRn_M0(wrd);
1785 gen_op_iwmmxt_set_mup();
1787 case 0x010: case 0x110: case 0x210: case 0x310: /* WMUL */
1788 wrd = (insn >> 12) & 0xf;
1789 rd0 = (insn >> 16) & 0xf;
1790 rd1 = (insn >> 0) & 0xf;
1791 gen_op_iwmmxt_movq_M0_wRn(rd0);
1792 if (insn & (1 << 21)) {
1793 if (insn & (1 << 20))
1794 gen_op_iwmmxt_mulshw_M0_wRn(rd1);
1796 gen_op_iwmmxt_mulslw_M0_wRn(rd1);
1798 if (insn & (1 << 20))
1799 gen_op_iwmmxt_muluhw_M0_wRn(rd1);
1801 gen_op_iwmmxt_mululw_M0_wRn(rd1);
1803 gen_op_iwmmxt_movq_wRn_M0(wrd);
1804 gen_op_iwmmxt_set_mup();
1806 case 0x410: case 0x510: case 0x610: case 0x710: /* WMAC */
1807 wrd = (insn >> 12) & 0xf;
1808 rd0 = (insn >> 16) & 0xf;
1809 rd1 = (insn >> 0) & 0xf;
1810 gen_op_iwmmxt_movq_M0_wRn(rd0);
1811 if (insn & (1 << 21))
1812 gen_op_iwmmxt_macsw_M0_wRn(rd1);
1814 gen_op_iwmmxt_macuw_M0_wRn(rd1);
1815 if (!(insn & (1 << 20))) {
1816 iwmmxt_load_reg(cpu_V1, wrd);
1817 tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
1819 gen_op_iwmmxt_movq_wRn_M0(wrd);
1820 gen_op_iwmmxt_set_mup();
1822 case 0x006: case 0x406: case 0x806: case 0xc06: /* WCMPEQ */
1823 wrd = (insn >> 12) & 0xf;
1824 rd0 = (insn >> 16) & 0xf;
1825 rd1 = (insn >> 0) & 0xf;
1826 gen_op_iwmmxt_movq_M0_wRn(rd0);
1827 switch ((insn >> 22) & 3) {
1829 gen_op_iwmmxt_cmpeqb_M0_wRn(rd1);
1832 gen_op_iwmmxt_cmpeqw_M0_wRn(rd1);
1835 gen_op_iwmmxt_cmpeql_M0_wRn(rd1);
1840 gen_op_iwmmxt_movq_wRn_M0(wrd);
1841 gen_op_iwmmxt_set_mup();
1842 gen_op_iwmmxt_set_cup();
1844 case 0x800: case 0x900: case 0xc00: case 0xd00: /* WAVG2 */
1845 wrd = (insn >> 12) & 0xf;
1846 rd0 = (insn >> 16) & 0xf;
1847 rd1 = (insn >> 0) & 0xf;
1848 gen_op_iwmmxt_movq_M0_wRn(rd0);
1849 if (insn & (1 << 22)) {
1850 if (insn & (1 << 20))
1851 gen_op_iwmmxt_avgw1_M0_wRn(rd1);
1853 gen_op_iwmmxt_avgw0_M0_wRn(rd1);
1855 if (insn & (1 << 20))
1856 gen_op_iwmmxt_avgb1_M0_wRn(rd1);
1858 gen_op_iwmmxt_avgb0_M0_wRn(rd1);
1860 gen_op_iwmmxt_movq_wRn_M0(wrd);
1861 gen_op_iwmmxt_set_mup();
1862 gen_op_iwmmxt_set_cup();
1864 case 0x802: case 0x902: case 0xa02: case 0xb02: /* WALIGNR */
1865 wrd = (insn >> 12) & 0xf;
1866 rd0 = (insn >> 16) & 0xf;
1867 rd1 = (insn >> 0) & 0xf;
1868 gen_op_iwmmxt_movq_M0_wRn(rd0);
1869 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCGR0 + ((insn >> 20) & 3));
1870 tcg_gen_andi_i32(tmp, tmp, 7);
1871 iwmmxt_load_reg(cpu_V1, rd1);
1872 gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1, tmp);
1873 tcg_temp_free_i32(tmp);
1874 gen_op_iwmmxt_movq_wRn_M0(wrd);
1875 gen_op_iwmmxt_set_mup();
1877 case 0x601: case 0x605: case 0x609: case 0x60d: /* TINSR */
1878 if (((insn >> 6) & 3) == 3)
1880 rd = (insn >> 12) & 0xf;
1881 wrd = (insn >> 16) & 0xf;
1882 tmp = load_reg(s, rd);
1883 gen_op_iwmmxt_movq_M0_wRn(wrd);
1884 switch ((insn >> 6) & 3) {
1886 tmp2 = tcg_const_i32(0xff);
1887 tmp3 = tcg_const_i32((insn & 7) << 3);
1890 tmp2 = tcg_const_i32(0xffff);
1891 tmp3 = tcg_const_i32((insn & 3) << 4);
1894 tmp2 = tcg_const_i32(0xffffffff);
1895 tmp3 = tcg_const_i32((insn & 1) << 5);
1901 gen_helper_iwmmxt_insr(cpu_M0, cpu_M0, tmp, tmp2, tmp3);
1902 tcg_temp_free_i32(tmp3);
1903 tcg_temp_free_i32(tmp2);
1904 tcg_temp_free_i32(tmp);
1905 gen_op_iwmmxt_movq_wRn_M0(wrd);
1906 gen_op_iwmmxt_set_mup();
1908 case 0x107: case 0x507: case 0x907: case 0xd07: /* TEXTRM */
1909 rd = (insn >> 12) & 0xf;
1910 wrd = (insn >> 16) & 0xf;
1911 if (rd == 15 || ((insn >> 22) & 3) == 3)
1913 gen_op_iwmmxt_movq_M0_wRn(wrd);
1914 tmp = tcg_temp_new_i32();
1915 switch ((insn >> 22) & 3) {
1917 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 7) << 3);
1918 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1920 tcg_gen_ext8s_i32(tmp, tmp);
1922 tcg_gen_andi_i32(tmp, tmp, 0xff);
1926 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 3) << 4);
1927 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1929 tcg_gen_ext16s_i32(tmp, tmp);
1931 tcg_gen_andi_i32(tmp, tmp, 0xffff);
1935 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 1) << 5);
1936 tcg_gen_extrl_i64_i32(tmp, cpu_M0);
1939 store_reg(s, rd, tmp);
1941 case 0x117: case 0x517: case 0x917: case 0xd17: /* TEXTRC */
1942 if ((insn & 0x000ff008) != 0x0003f000 || ((insn >> 22) & 3) == 3)
1944 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
1945 switch ((insn >> 22) & 3) {
1947 tcg_gen_shri_i32(tmp, tmp, ((insn & 7) << 2) + 0);
1950 tcg_gen_shri_i32(tmp, tmp, ((insn & 3) << 3) + 4);
1953 tcg_gen_shri_i32(tmp, tmp, ((insn & 1) << 4) + 12);
1956 tcg_gen_shli_i32(tmp, tmp, 28);
1958 tcg_temp_free_i32(tmp);
1960 case 0x401: case 0x405: case 0x409: case 0x40d: /* TBCST */
1961 if (((insn >> 6) & 3) == 3)
1963 rd = (insn >> 12) & 0xf;
1964 wrd = (insn >> 16) & 0xf;
1965 tmp = load_reg(s, rd);
1966 switch ((insn >> 6) & 3) {
1968 gen_helper_iwmmxt_bcstb(cpu_M0, tmp);
1971 gen_helper_iwmmxt_bcstw(cpu_M0, tmp);
1974 gen_helper_iwmmxt_bcstl(cpu_M0, tmp);
1977 tcg_temp_free_i32(tmp);
1978 gen_op_iwmmxt_movq_wRn_M0(wrd);
1979 gen_op_iwmmxt_set_mup();
1981 case 0x113: case 0x513: case 0x913: case 0xd13: /* TANDC */
1982 if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
1984 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
1985 tmp2 = tcg_temp_new_i32();
1986 tcg_gen_mov_i32(tmp2, tmp);
1987 switch ((insn >> 22) & 3) {
1989 for (i = 0; i < 7; i ++) {
1990 tcg_gen_shli_i32(tmp2, tmp2, 4);
1991 tcg_gen_and_i32(tmp, tmp, tmp2);
1995 for (i = 0; i < 3; i ++) {
1996 tcg_gen_shli_i32(tmp2, tmp2, 8);
1997 tcg_gen_and_i32(tmp, tmp, tmp2);
2001 tcg_gen_shli_i32(tmp2, tmp2, 16);
2002 tcg_gen_and_i32(tmp, tmp, tmp2);
2006 tcg_temp_free_i32(tmp2);
2007 tcg_temp_free_i32(tmp);
2009 case 0x01c: case 0x41c: case 0x81c: case 0xc1c: /* WACC */
2010 wrd = (insn >> 12) & 0xf;
2011 rd0 = (insn >> 16) & 0xf;
2012 gen_op_iwmmxt_movq_M0_wRn(rd0);
2013 switch ((insn >> 22) & 3) {
2015 gen_helper_iwmmxt_addcb(cpu_M0, cpu_M0);
2018 gen_helper_iwmmxt_addcw(cpu_M0, cpu_M0);
2021 gen_helper_iwmmxt_addcl(cpu_M0, cpu_M0);
2026 gen_op_iwmmxt_movq_wRn_M0(wrd);
2027 gen_op_iwmmxt_set_mup();
2029 case 0x115: case 0x515: case 0x915: case 0xd15: /* TORC */
2030 if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
2032 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
2033 tmp2 = tcg_temp_new_i32();
2034 tcg_gen_mov_i32(tmp2, tmp);
2035 switch ((insn >> 22) & 3) {
2037 for (i = 0; i < 7; i ++) {
2038 tcg_gen_shli_i32(tmp2, tmp2, 4);
2039 tcg_gen_or_i32(tmp, tmp, tmp2);
2043 for (i = 0; i < 3; i ++) {
2044 tcg_gen_shli_i32(tmp2, tmp2, 8);
2045 tcg_gen_or_i32(tmp, tmp, tmp2);
2049 tcg_gen_shli_i32(tmp2, tmp2, 16);
2050 tcg_gen_or_i32(tmp, tmp, tmp2);
2054 tcg_temp_free_i32(tmp2);
2055 tcg_temp_free_i32(tmp);
2057 case 0x103: case 0x503: case 0x903: case 0xd03: /* TMOVMSK */
2058 rd = (insn >> 12) & 0xf;
2059 rd0 = (insn >> 16) & 0xf;
2060 if ((insn & 0xf) != 0 || ((insn >> 22) & 3) == 3)
2062 gen_op_iwmmxt_movq_M0_wRn(rd0);
2063 tmp = tcg_temp_new_i32();
2064 switch ((insn >> 22) & 3) {
2066 gen_helper_iwmmxt_msbb(tmp, cpu_M0);
2069 gen_helper_iwmmxt_msbw(tmp, cpu_M0);
2072 gen_helper_iwmmxt_msbl(tmp, cpu_M0);
2075 store_reg(s, rd, tmp);
2077 case 0x106: case 0x306: case 0x506: case 0x706: /* WCMPGT */
2078 case 0x906: case 0xb06: case 0xd06: case 0xf06:
2079 wrd = (insn >> 12) & 0xf;
2080 rd0 = (insn >> 16) & 0xf;
2081 rd1 = (insn >> 0) & 0xf;
2082 gen_op_iwmmxt_movq_M0_wRn(rd0);
2083 switch ((insn >> 22) & 3) {
2085 if (insn & (1 << 21))
2086 gen_op_iwmmxt_cmpgtsb_M0_wRn(rd1);
2088 gen_op_iwmmxt_cmpgtub_M0_wRn(rd1);
2091 if (insn & (1 << 21))
2092 gen_op_iwmmxt_cmpgtsw_M0_wRn(rd1);
2094 gen_op_iwmmxt_cmpgtuw_M0_wRn(rd1);
2097 if (insn & (1 << 21))
2098 gen_op_iwmmxt_cmpgtsl_M0_wRn(rd1);
2100 gen_op_iwmmxt_cmpgtul_M0_wRn(rd1);
2105 gen_op_iwmmxt_movq_wRn_M0(wrd);
2106 gen_op_iwmmxt_set_mup();
2107 gen_op_iwmmxt_set_cup();
2109 case 0x00e: case 0x20e: case 0x40e: case 0x60e: /* WUNPCKEL */
2110 case 0x80e: case 0xa0e: case 0xc0e: case 0xe0e:
2111 wrd = (insn >> 12) & 0xf;
2112 rd0 = (insn >> 16) & 0xf;
2113 gen_op_iwmmxt_movq_M0_wRn(rd0);
2114 switch ((insn >> 22) & 3) {
2116 if (insn & (1 << 21))
2117 gen_op_iwmmxt_unpacklsb_M0();
2119 gen_op_iwmmxt_unpacklub_M0();
2122 if (insn & (1 << 21))
2123 gen_op_iwmmxt_unpacklsw_M0();
2125 gen_op_iwmmxt_unpackluw_M0();
2128 if (insn & (1 << 21))
2129 gen_op_iwmmxt_unpacklsl_M0();
2131 gen_op_iwmmxt_unpacklul_M0();
2136 gen_op_iwmmxt_movq_wRn_M0(wrd);
2137 gen_op_iwmmxt_set_mup();
2138 gen_op_iwmmxt_set_cup();
2140 case 0x00c: case 0x20c: case 0x40c: case 0x60c: /* WUNPCKEH */
2141 case 0x80c: case 0xa0c: case 0xc0c: case 0xe0c:
2142 wrd = (insn >> 12) & 0xf;
2143 rd0 = (insn >> 16) & 0xf;
2144 gen_op_iwmmxt_movq_M0_wRn(rd0);
2145 switch ((insn >> 22) & 3) {
2147 if (insn & (1 << 21))
2148 gen_op_iwmmxt_unpackhsb_M0();
2150 gen_op_iwmmxt_unpackhub_M0();
2153 if (insn & (1 << 21))
2154 gen_op_iwmmxt_unpackhsw_M0();
2156 gen_op_iwmmxt_unpackhuw_M0();
2159 if (insn & (1 << 21))
2160 gen_op_iwmmxt_unpackhsl_M0();
2162 gen_op_iwmmxt_unpackhul_M0();
2167 gen_op_iwmmxt_movq_wRn_M0(wrd);
2168 gen_op_iwmmxt_set_mup();
2169 gen_op_iwmmxt_set_cup();
2171 case 0x204: case 0x604: case 0xa04: case 0xe04: /* WSRL */
2172 case 0x214: case 0x614: case 0xa14: case 0xe14:
2173 if (((insn >> 22) & 3) == 0)
2175 wrd = (insn >> 12) & 0xf;
2176 rd0 = (insn >> 16) & 0xf;
2177 gen_op_iwmmxt_movq_M0_wRn(rd0);
2178 tmp = tcg_temp_new_i32();
2179 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2180 tcg_temp_free_i32(tmp);
2183 switch ((insn >> 22) & 3) {
2185 gen_helper_iwmmxt_srlw(cpu_M0, cpu_env, cpu_M0, tmp);
2188 gen_helper_iwmmxt_srll(cpu_M0, cpu_env, cpu_M0, tmp);
2191 gen_helper_iwmmxt_srlq(cpu_M0, cpu_env, cpu_M0, tmp);
2194 tcg_temp_free_i32(tmp);
2195 gen_op_iwmmxt_movq_wRn_M0(wrd);
2196 gen_op_iwmmxt_set_mup();
2197 gen_op_iwmmxt_set_cup();
2199 case 0x004: case 0x404: case 0x804: case 0xc04: /* WSRA */
2200 case 0x014: case 0x414: case 0x814: case 0xc14:
2201 if (((insn >> 22) & 3) == 0)
2203 wrd = (insn >> 12) & 0xf;
2204 rd0 = (insn >> 16) & 0xf;
2205 gen_op_iwmmxt_movq_M0_wRn(rd0);
2206 tmp = tcg_temp_new_i32();
2207 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2208 tcg_temp_free_i32(tmp);
2211 switch ((insn >> 22) & 3) {
2213 gen_helper_iwmmxt_sraw(cpu_M0, cpu_env, cpu_M0, tmp);
2216 gen_helper_iwmmxt_sral(cpu_M0, cpu_env, cpu_M0, tmp);
2219 gen_helper_iwmmxt_sraq(cpu_M0, cpu_env, cpu_M0, tmp);
2222 tcg_temp_free_i32(tmp);
2223 gen_op_iwmmxt_movq_wRn_M0(wrd);
2224 gen_op_iwmmxt_set_mup();
2225 gen_op_iwmmxt_set_cup();
2227 case 0x104: case 0x504: case 0x904: case 0xd04: /* WSLL */
2228 case 0x114: case 0x514: case 0x914: case 0xd14:
2229 if (((insn >> 22) & 3) == 0)
2231 wrd = (insn >> 12) & 0xf;
2232 rd0 = (insn >> 16) & 0xf;
2233 gen_op_iwmmxt_movq_M0_wRn(rd0);
2234 tmp = tcg_temp_new_i32();
2235 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2236 tcg_temp_free_i32(tmp);
2239 switch ((insn >> 22) & 3) {
2241 gen_helper_iwmmxt_sllw(cpu_M0, cpu_env, cpu_M0, tmp);
2244 gen_helper_iwmmxt_slll(cpu_M0, cpu_env, cpu_M0, tmp);
2247 gen_helper_iwmmxt_sllq(cpu_M0, cpu_env, cpu_M0, tmp);
2250 tcg_temp_free_i32(tmp);
2251 gen_op_iwmmxt_movq_wRn_M0(wrd);
2252 gen_op_iwmmxt_set_mup();
2253 gen_op_iwmmxt_set_cup();
2255 case 0x304: case 0x704: case 0xb04: case 0xf04: /* WROR */
2256 case 0x314: case 0x714: case 0xb14: case 0xf14:
2257 if (((insn >> 22) & 3) == 0)
2259 wrd = (insn >> 12) & 0xf;
2260 rd0 = (insn >> 16) & 0xf;
2261 gen_op_iwmmxt_movq_M0_wRn(rd0);
2262 tmp = tcg_temp_new_i32();
2263 switch ((insn >> 22) & 3) {
2265 if (gen_iwmmxt_shift(insn, 0xf, tmp)) {
2266 tcg_temp_free_i32(tmp);
2269 gen_helper_iwmmxt_rorw(cpu_M0, cpu_env, cpu_M0, tmp);
2272 if (gen_iwmmxt_shift(insn, 0x1f, tmp)) {
2273 tcg_temp_free_i32(tmp);
2276 gen_helper_iwmmxt_rorl(cpu_M0, cpu_env, cpu_M0, tmp);
2279 if (gen_iwmmxt_shift(insn, 0x3f, tmp)) {
2280 tcg_temp_free_i32(tmp);
2283 gen_helper_iwmmxt_rorq(cpu_M0, cpu_env, cpu_M0, tmp);
2286 tcg_temp_free_i32(tmp);
2287 gen_op_iwmmxt_movq_wRn_M0(wrd);
2288 gen_op_iwmmxt_set_mup();
2289 gen_op_iwmmxt_set_cup();
2291 case 0x116: case 0x316: case 0x516: case 0x716: /* WMIN */
2292 case 0x916: case 0xb16: case 0xd16: case 0xf16:
2293 wrd = (insn >> 12) & 0xf;
2294 rd0 = (insn >> 16) & 0xf;
2295 rd1 = (insn >> 0) & 0xf;
2296 gen_op_iwmmxt_movq_M0_wRn(rd0);
2297 switch ((insn >> 22) & 3) {
2299 if (insn & (1 << 21))
2300 gen_op_iwmmxt_minsb_M0_wRn(rd1);
2302 gen_op_iwmmxt_minub_M0_wRn(rd1);
2305 if (insn & (1 << 21))
2306 gen_op_iwmmxt_minsw_M0_wRn(rd1);
2308 gen_op_iwmmxt_minuw_M0_wRn(rd1);
2311 if (insn & (1 << 21))
2312 gen_op_iwmmxt_minsl_M0_wRn(rd1);
2314 gen_op_iwmmxt_minul_M0_wRn(rd1);
2319 gen_op_iwmmxt_movq_wRn_M0(wrd);
2320 gen_op_iwmmxt_set_mup();
2322 case 0x016: case 0x216: case 0x416: case 0x616: /* WMAX */
2323 case 0x816: case 0xa16: case 0xc16: case 0xe16:
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) {
2330 if (insn & (1 << 21))
2331 gen_op_iwmmxt_maxsb_M0_wRn(rd1);
2333 gen_op_iwmmxt_maxub_M0_wRn(rd1);
2336 if (insn & (1 << 21))
2337 gen_op_iwmmxt_maxsw_M0_wRn(rd1);
2339 gen_op_iwmmxt_maxuw_M0_wRn(rd1);
2342 if (insn & (1 << 21))
2343 gen_op_iwmmxt_maxsl_M0_wRn(rd1);
2345 gen_op_iwmmxt_maxul_M0_wRn(rd1);
2350 gen_op_iwmmxt_movq_wRn_M0(wrd);
2351 gen_op_iwmmxt_set_mup();
2353 case 0x002: case 0x102: case 0x202: case 0x302: /* WALIGNI */
2354 case 0x402: case 0x502: case 0x602: case 0x702:
2355 wrd = (insn >> 12) & 0xf;
2356 rd0 = (insn >> 16) & 0xf;
2357 rd1 = (insn >> 0) & 0xf;
2358 gen_op_iwmmxt_movq_M0_wRn(rd0);
2359 tmp = tcg_const_i32((insn >> 20) & 3);
2360 iwmmxt_load_reg(cpu_V1, rd1);
2361 gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1, tmp);
2362 tcg_temp_free_i32(tmp);
2363 gen_op_iwmmxt_movq_wRn_M0(wrd);
2364 gen_op_iwmmxt_set_mup();
2366 case 0x01a: case 0x11a: case 0x21a: case 0x31a: /* WSUB */
2367 case 0x41a: case 0x51a: case 0x61a: case 0x71a:
2368 case 0x81a: case 0x91a: case 0xa1a: case 0xb1a:
2369 case 0xc1a: case 0xd1a: case 0xe1a: case 0xf1a:
2370 wrd = (insn >> 12) & 0xf;
2371 rd0 = (insn >> 16) & 0xf;
2372 rd1 = (insn >> 0) & 0xf;
2373 gen_op_iwmmxt_movq_M0_wRn(rd0);
2374 switch ((insn >> 20) & 0xf) {
2376 gen_op_iwmmxt_subnb_M0_wRn(rd1);
2379 gen_op_iwmmxt_subub_M0_wRn(rd1);
2382 gen_op_iwmmxt_subsb_M0_wRn(rd1);
2385 gen_op_iwmmxt_subnw_M0_wRn(rd1);
2388 gen_op_iwmmxt_subuw_M0_wRn(rd1);
2391 gen_op_iwmmxt_subsw_M0_wRn(rd1);
2394 gen_op_iwmmxt_subnl_M0_wRn(rd1);
2397 gen_op_iwmmxt_subul_M0_wRn(rd1);
2400 gen_op_iwmmxt_subsl_M0_wRn(rd1);
2405 gen_op_iwmmxt_movq_wRn_M0(wrd);
2406 gen_op_iwmmxt_set_mup();
2407 gen_op_iwmmxt_set_cup();
2409 case 0x01e: case 0x11e: case 0x21e: case 0x31e: /* WSHUFH */
2410 case 0x41e: case 0x51e: case 0x61e: case 0x71e:
2411 case 0x81e: case 0x91e: case 0xa1e: case 0xb1e:
2412 case 0xc1e: case 0xd1e: case 0xe1e: case 0xf1e:
2413 wrd = (insn >> 12) & 0xf;
2414 rd0 = (insn >> 16) & 0xf;
2415 gen_op_iwmmxt_movq_M0_wRn(rd0);
2416 tmp = tcg_const_i32(((insn >> 16) & 0xf0) | (insn & 0x0f));
2417 gen_helper_iwmmxt_shufh(cpu_M0, cpu_env, cpu_M0, tmp);
2418 tcg_temp_free_i32(tmp);
2419 gen_op_iwmmxt_movq_wRn_M0(wrd);
2420 gen_op_iwmmxt_set_mup();
2421 gen_op_iwmmxt_set_cup();
2423 case 0x018: case 0x118: case 0x218: case 0x318: /* WADD */
2424 case 0x418: case 0x518: case 0x618: case 0x718:
2425 case 0x818: case 0x918: case 0xa18: case 0xb18:
2426 case 0xc18: case 0xd18: case 0xe18: case 0xf18:
2427 wrd = (insn >> 12) & 0xf;
2428 rd0 = (insn >> 16) & 0xf;
2429 rd1 = (insn >> 0) & 0xf;
2430 gen_op_iwmmxt_movq_M0_wRn(rd0);
2431 switch ((insn >> 20) & 0xf) {
2433 gen_op_iwmmxt_addnb_M0_wRn(rd1);
2436 gen_op_iwmmxt_addub_M0_wRn(rd1);
2439 gen_op_iwmmxt_addsb_M0_wRn(rd1);
2442 gen_op_iwmmxt_addnw_M0_wRn(rd1);
2445 gen_op_iwmmxt_adduw_M0_wRn(rd1);
2448 gen_op_iwmmxt_addsw_M0_wRn(rd1);
2451 gen_op_iwmmxt_addnl_M0_wRn(rd1);
2454 gen_op_iwmmxt_addul_M0_wRn(rd1);
2457 gen_op_iwmmxt_addsl_M0_wRn(rd1);
2462 gen_op_iwmmxt_movq_wRn_M0(wrd);
2463 gen_op_iwmmxt_set_mup();
2464 gen_op_iwmmxt_set_cup();
2466 case 0x008: case 0x108: case 0x208: case 0x308: /* WPACK */
2467 case 0x408: case 0x508: case 0x608: case 0x708:
2468 case 0x808: case 0x908: case 0xa08: case 0xb08:
2469 case 0xc08: case 0xd08: case 0xe08: case 0xf08:
2470 if (!(insn & (1 << 20)) || ((insn >> 22) & 3) == 0)
2472 wrd = (insn >> 12) & 0xf;
2473 rd0 = (insn >> 16) & 0xf;
2474 rd1 = (insn >> 0) & 0xf;
2475 gen_op_iwmmxt_movq_M0_wRn(rd0);
2476 switch ((insn >> 22) & 3) {
2478 if (insn & (1 << 21))
2479 gen_op_iwmmxt_packsw_M0_wRn(rd1);
2481 gen_op_iwmmxt_packuw_M0_wRn(rd1);
2484 if (insn & (1 << 21))
2485 gen_op_iwmmxt_packsl_M0_wRn(rd1);
2487 gen_op_iwmmxt_packul_M0_wRn(rd1);
2490 if (insn & (1 << 21))
2491 gen_op_iwmmxt_packsq_M0_wRn(rd1);
2493 gen_op_iwmmxt_packuq_M0_wRn(rd1);
2496 gen_op_iwmmxt_movq_wRn_M0(wrd);
2497 gen_op_iwmmxt_set_mup();
2498 gen_op_iwmmxt_set_cup();
2500 case 0x201: case 0x203: case 0x205: case 0x207:
2501 case 0x209: case 0x20b: case 0x20d: case 0x20f:
2502 case 0x211: case 0x213: case 0x215: case 0x217:
2503 case 0x219: case 0x21b: case 0x21d: case 0x21f:
2504 wrd = (insn >> 5) & 0xf;
2505 rd0 = (insn >> 12) & 0xf;
2506 rd1 = (insn >> 0) & 0xf;
2507 if (rd0 == 0xf || rd1 == 0xf)
2509 gen_op_iwmmxt_movq_M0_wRn(wrd);
2510 tmp = load_reg(s, rd0);
2511 tmp2 = load_reg(s, rd1);
2512 switch ((insn >> 16) & 0xf) {
2513 case 0x0: /* TMIA */
2514 gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2516 case 0x8: /* TMIAPH */
2517 gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2519 case 0xc: case 0xd: case 0xe: case 0xf: /* TMIAxy */
2520 if (insn & (1 << 16))
2521 tcg_gen_shri_i32(tmp, tmp, 16);
2522 if (insn & (1 << 17))
2523 tcg_gen_shri_i32(tmp2, tmp2, 16);
2524 gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2527 tcg_temp_free_i32(tmp2);
2528 tcg_temp_free_i32(tmp);
2531 tcg_temp_free_i32(tmp2);
2532 tcg_temp_free_i32(tmp);
2533 gen_op_iwmmxt_movq_wRn_M0(wrd);
2534 gen_op_iwmmxt_set_mup();
2543 /* Disassemble an XScale DSP instruction. Returns nonzero if an error occurred
2544 (ie. an undefined instruction). */
2545 static int disas_dsp_insn(DisasContext *s, uint32_t insn)
2547 int acc, rd0, rd1, rdhi, rdlo;
2550 if ((insn & 0x0ff00f10) == 0x0e200010) {
2551 /* Multiply with Internal Accumulate Format */
2552 rd0 = (insn >> 12) & 0xf;
2554 acc = (insn >> 5) & 7;
2559 tmp = load_reg(s, rd0);
2560 tmp2 = load_reg(s, rd1);
2561 switch ((insn >> 16) & 0xf) {
2563 gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2565 case 0x8: /* MIAPH */
2566 gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2568 case 0xc: /* MIABB */
2569 case 0xd: /* MIABT */
2570 case 0xe: /* MIATB */
2571 case 0xf: /* MIATT */
2572 if (insn & (1 << 16))
2573 tcg_gen_shri_i32(tmp, tmp, 16);
2574 if (insn & (1 << 17))
2575 tcg_gen_shri_i32(tmp2, tmp2, 16);
2576 gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2581 tcg_temp_free_i32(tmp2);
2582 tcg_temp_free_i32(tmp);
2584 gen_op_iwmmxt_movq_wRn_M0(acc);
2588 if ((insn & 0x0fe00ff8) == 0x0c400000) {
2589 /* Internal Accumulator Access Format */
2590 rdhi = (insn >> 16) & 0xf;
2591 rdlo = (insn >> 12) & 0xf;
2597 if (insn & ARM_CP_RW_BIT) { /* MRA */
2598 iwmmxt_load_reg(cpu_V0, acc);
2599 tcg_gen_extrl_i64_i32(cpu_R[rdlo], cpu_V0);
2600 tcg_gen_extrh_i64_i32(cpu_R[rdhi], cpu_V0);
2601 tcg_gen_andi_i32(cpu_R[rdhi], cpu_R[rdhi], (1 << (40 - 32)) - 1);
2603 tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
2604 iwmmxt_store_reg(cpu_V0, acc);
2612 #define VFP_REG_SHR(x, n) (((n) > 0) ? (x) >> (n) : (x) << -(n))
2613 #define VFP_DREG(reg, insn, bigbit, smallbit) do { \
2614 if (dc_isar_feature(aa32_simd_r32, s)) { \
2615 reg = (((insn) >> (bigbit)) & 0x0f) \
2616 | (((insn) >> ((smallbit) - 4)) & 0x10); \
2618 if (insn & (1 << (smallbit))) \
2620 reg = ((insn) >> (bigbit)) & 0x0f; \
2623 #define VFP_DREG_D(reg, insn) VFP_DREG(reg, insn, 12, 22)
2624 #define VFP_DREG_N(reg, insn) VFP_DREG(reg, insn, 16, 7)
2625 #define VFP_DREG_M(reg, insn) VFP_DREG(reg, insn, 0, 5)
2627 static void gen_neon_dup_low16(TCGv_i32 var)
2629 TCGv_i32 tmp = tcg_temp_new_i32();
2630 tcg_gen_ext16u_i32(var, var);
2631 tcg_gen_shli_i32(tmp, var, 16);
2632 tcg_gen_or_i32(var, var, tmp);
2633 tcg_temp_free_i32(tmp);
2636 static void gen_neon_dup_high16(TCGv_i32 var)
2638 TCGv_i32 tmp = tcg_temp_new_i32();
2639 tcg_gen_andi_i32(var, var, 0xffff0000);
2640 tcg_gen_shri_i32(tmp, var, 16);
2641 tcg_gen_or_i32(var, var, tmp);
2642 tcg_temp_free_i32(tmp);
2645 static inline bool use_goto_tb(DisasContext *s, target_ulong dest)
2647 #ifndef CONFIG_USER_ONLY
2648 return (s->base.tb->pc & TARGET_PAGE_MASK) == (dest & TARGET_PAGE_MASK) ||
2649 ((s->base.pc_next - 1) & TARGET_PAGE_MASK) == (dest & TARGET_PAGE_MASK);
2655 static void gen_goto_ptr(void)
2657 tcg_gen_lookup_and_goto_ptr();
2660 /* This will end the TB but doesn't guarantee we'll return to
2661 * cpu_loop_exec. Any live exit_requests will be processed as we
2662 * enter the next TB.
2664 static void gen_goto_tb(DisasContext *s, int n, target_ulong dest)
2666 if (use_goto_tb(s, dest)) {
2668 gen_set_pc_im(s, dest);
2669 tcg_gen_exit_tb(s->base.tb, n);
2671 gen_set_pc_im(s, dest);
2674 s->base.is_jmp = DISAS_NORETURN;
2677 static inline void gen_jmp (DisasContext *s, uint32_t dest)
2679 if (unlikely(is_singlestepping(s))) {
2680 /* An indirect jump so that we still trigger the debug exception. */
2681 gen_set_pc_im(s, dest);
2682 s->base.is_jmp = DISAS_JUMP;
2684 gen_goto_tb(s, 0, dest);
2688 static inline void gen_mulxy(TCGv_i32 t0, TCGv_i32 t1, int x, int y)
2691 tcg_gen_sari_i32(t0, t0, 16);
2695 tcg_gen_sari_i32(t1, t1, 16);
2698 tcg_gen_mul_i32(t0, t0, t1);
2701 /* Return the mask of PSR bits set by a MSR instruction. */
2702 static uint32_t msr_mask(DisasContext *s, int flags, int spsr)
2706 if (flags & (1 << 0)) {
2709 if (flags & (1 << 1)) {
2712 if (flags & (1 << 2)) {
2715 if (flags & (1 << 3)) {
2719 /* Mask out undefined and reserved bits. */
2720 mask &= aarch32_cpsr_valid_mask(s->features, s->isar);
2722 /* Mask out execution state. */
2727 /* Mask out privileged bits. */
2734 /* Returns nonzero if access to the PSR is not permitted. Marks t0 as dead. */
2735 static int gen_set_psr(DisasContext *s, uint32_t mask, int spsr, TCGv_i32 t0)
2739 /* ??? This is also undefined in system mode. */
2743 tmp = load_cpu_field(spsr);
2744 tcg_gen_andi_i32(tmp, tmp, ~mask);
2745 tcg_gen_andi_i32(t0, t0, mask);
2746 tcg_gen_or_i32(tmp, tmp, t0);
2747 store_cpu_field(tmp, spsr);
2749 gen_set_cpsr(t0, mask);
2751 tcg_temp_free_i32(t0);
2756 /* Returns nonzero if access to the PSR is not permitted. */
2757 static int gen_set_psr_im(DisasContext *s, uint32_t mask, int spsr, uint32_t val)
2760 tmp = tcg_temp_new_i32();
2761 tcg_gen_movi_i32(tmp, val);
2762 return gen_set_psr(s, mask, spsr, tmp);
2765 static bool msr_banked_access_decode(DisasContext *s, int r, int sysm, int rn,
2766 int *tgtmode, int *regno)
2768 /* Decode the r and sysm fields of MSR/MRS banked accesses into
2769 * the target mode and register number, and identify the various
2770 * unpredictable cases.
2771 * MSR (banked) and MRS (banked) are CONSTRAINED UNPREDICTABLE if:
2772 * + executed in user mode
2773 * + using R15 as the src/dest register
2774 * + accessing an unimplemented register
2775 * + accessing a register that's inaccessible at current PL/security state*
2776 * + accessing a register that you could access with a different insn
2777 * We choose to UNDEF in all these cases.
2778 * Since we don't know which of the various AArch32 modes we are in
2779 * we have to defer some checks to runtime.
2780 * Accesses to Monitor mode registers from Secure EL1 (which implies
2781 * that EL3 is AArch64) must trap to EL3.
2783 * If the access checks fail this function will emit code to take
2784 * an exception and return false. Otherwise it will return true,
2785 * and set *tgtmode and *regno appropriately.
2787 int exc_target = default_exception_el(s);
2789 /* These instructions are present only in ARMv8, or in ARMv7 with the
2790 * Virtualization Extensions.
2792 if (!arm_dc_feature(s, ARM_FEATURE_V8) &&
2793 !arm_dc_feature(s, ARM_FEATURE_EL2)) {
2797 if (IS_USER(s) || rn == 15) {
2801 /* The table in the v8 ARM ARM section F5.2.3 describes the encoding
2802 * of registers into (r, sysm).
2805 /* SPSRs for other modes */
2807 case 0xe: /* SPSR_fiq */
2808 *tgtmode = ARM_CPU_MODE_FIQ;
2810 case 0x10: /* SPSR_irq */
2811 *tgtmode = ARM_CPU_MODE_IRQ;
2813 case 0x12: /* SPSR_svc */
2814 *tgtmode = ARM_CPU_MODE_SVC;
2816 case 0x14: /* SPSR_abt */
2817 *tgtmode = ARM_CPU_MODE_ABT;
2819 case 0x16: /* SPSR_und */
2820 *tgtmode = ARM_CPU_MODE_UND;
2822 case 0x1c: /* SPSR_mon */
2823 *tgtmode = ARM_CPU_MODE_MON;
2825 case 0x1e: /* SPSR_hyp */
2826 *tgtmode = ARM_CPU_MODE_HYP;
2828 default: /* unallocated */
2831 /* We arbitrarily assign SPSR a register number of 16. */
2834 /* general purpose registers for other modes */
2836 case 0x0 ... 0x6: /* 0b00xxx : r8_usr ... r14_usr */
2837 *tgtmode = ARM_CPU_MODE_USR;
2840 case 0x8 ... 0xe: /* 0b01xxx : r8_fiq ... r14_fiq */
2841 *tgtmode = ARM_CPU_MODE_FIQ;
2844 case 0x10 ... 0x11: /* 0b1000x : r14_irq, r13_irq */
2845 *tgtmode = ARM_CPU_MODE_IRQ;
2846 *regno = sysm & 1 ? 13 : 14;
2848 case 0x12 ... 0x13: /* 0b1001x : r14_svc, r13_svc */
2849 *tgtmode = ARM_CPU_MODE_SVC;
2850 *regno = sysm & 1 ? 13 : 14;
2852 case 0x14 ... 0x15: /* 0b1010x : r14_abt, r13_abt */
2853 *tgtmode = ARM_CPU_MODE_ABT;
2854 *regno = sysm & 1 ? 13 : 14;
2856 case 0x16 ... 0x17: /* 0b1011x : r14_und, r13_und */
2857 *tgtmode = ARM_CPU_MODE_UND;
2858 *regno = sysm & 1 ? 13 : 14;
2860 case 0x1c ... 0x1d: /* 0b1110x : r14_mon, r13_mon */
2861 *tgtmode = ARM_CPU_MODE_MON;
2862 *regno = sysm & 1 ? 13 : 14;
2864 case 0x1e ... 0x1f: /* 0b1111x : elr_hyp, r13_hyp */
2865 *tgtmode = ARM_CPU_MODE_HYP;
2866 /* Arbitrarily pick 17 for ELR_Hyp (which is not a banked LR!) */
2867 *regno = sysm & 1 ? 13 : 17;
2869 default: /* unallocated */
2874 /* Catch the 'accessing inaccessible register' cases we can detect
2875 * at translate time.
2878 case ARM_CPU_MODE_MON:
2879 if (!arm_dc_feature(s, ARM_FEATURE_EL3) || s->ns) {
2882 if (s->current_el == 1) {
2883 /* If we're in Secure EL1 (which implies that EL3 is AArch64)
2884 * then accesses to Mon registers trap to EL3
2890 case ARM_CPU_MODE_HYP:
2892 * SPSR_hyp and r13_hyp can only be accessed from Monitor mode
2893 * (and so we can forbid accesses from EL2 or below). elr_hyp
2894 * can be accessed also from Hyp mode, so forbid accesses from
2897 if (!arm_dc_feature(s, ARM_FEATURE_EL2) || s->current_el < 2 ||
2898 (s->current_el < 3 && *regno != 17)) {
2909 /* If we get here then some access check did not pass */
2910 gen_exception_insn(s, s->pc_curr, EXCP_UDEF,
2911 syn_uncategorized(), exc_target);
2915 static void gen_msr_banked(DisasContext *s, int r, int sysm, int rn)
2917 TCGv_i32 tcg_reg, tcg_tgtmode, tcg_regno;
2918 int tgtmode = 0, regno = 0;
2920 if (!msr_banked_access_decode(s, r, sysm, rn, &tgtmode, ®no)) {
2924 /* Sync state because msr_banked() can raise exceptions */
2925 gen_set_condexec(s);
2926 gen_set_pc_im(s, s->pc_curr);
2927 tcg_reg = load_reg(s, rn);
2928 tcg_tgtmode = tcg_const_i32(tgtmode);
2929 tcg_regno = tcg_const_i32(regno);
2930 gen_helper_msr_banked(cpu_env, tcg_reg, tcg_tgtmode, tcg_regno);
2931 tcg_temp_free_i32(tcg_tgtmode);
2932 tcg_temp_free_i32(tcg_regno);
2933 tcg_temp_free_i32(tcg_reg);
2934 s->base.is_jmp = DISAS_UPDATE;
2937 static void gen_mrs_banked(DisasContext *s, int r, int sysm, int rn)
2939 TCGv_i32 tcg_reg, tcg_tgtmode, tcg_regno;
2940 int tgtmode = 0, regno = 0;
2942 if (!msr_banked_access_decode(s, r, sysm, rn, &tgtmode, ®no)) {
2946 /* Sync state because mrs_banked() can raise exceptions */
2947 gen_set_condexec(s);
2948 gen_set_pc_im(s, s->pc_curr);
2949 tcg_reg = tcg_temp_new_i32();
2950 tcg_tgtmode = tcg_const_i32(tgtmode);
2951 tcg_regno = tcg_const_i32(regno);
2952 gen_helper_mrs_banked(tcg_reg, cpu_env, tcg_tgtmode, tcg_regno);
2953 tcg_temp_free_i32(tcg_tgtmode);
2954 tcg_temp_free_i32(tcg_regno);
2955 store_reg(s, rn, tcg_reg);
2956 s->base.is_jmp = DISAS_UPDATE;
2959 /* Store value to PC as for an exception return (ie don't
2960 * mask bits). The subsequent call to gen_helper_cpsr_write_eret()
2961 * will do the masking based on the new value of the Thumb bit.
2963 static void store_pc_exc_ret(DisasContext *s, TCGv_i32 pc)
2965 tcg_gen_mov_i32(cpu_R[15], pc);
2966 tcg_temp_free_i32(pc);
2969 /* Generate a v6 exception return. Marks both values as dead. */
2970 static void gen_rfe(DisasContext *s, TCGv_i32 pc, TCGv_i32 cpsr)
2972 store_pc_exc_ret(s, pc);
2973 /* The cpsr_write_eret helper will mask the low bits of PC
2974 * appropriately depending on the new Thumb bit, so it must
2975 * be called after storing the new PC.
2977 if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
2980 gen_helper_cpsr_write_eret(cpu_env, cpsr);
2981 tcg_temp_free_i32(cpsr);
2982 /* Must exit loop to check un-masked IRQs */
2983 s->base.is_jmp = DISAS_EXIT;
2986 /* Generate an old-style exception return. Marks pc as dead. */
2987 static void gen_exception_return(DisasContext *s, TCGv_i32 pc)
2989 gen_rfe(s, pc, load_cpu_field(spsr));
2992 #define CPU_V001 cpu_V0, cpu_V0, cpu_V1
2994 static inline void gen_neon_add(int size, TCGv_i32 t0, TCGv_i32 t1)
2997 case 0: gen_helper_neon_add_u8(t0, t0, t1); break;
2998 case 1: gen_helper_neon_add_u16(t0, t0, t1); break;
2999 case 2: tcg_gen_add_i32(t0, t0, t1); break;
3004 static inline void gen_neon_rsb(int size, TCGv_i32 t0, TCGv_i32 t1)
3007 case 0: gen_helper_neon_sub_u8(t0, t1, t0); break;
3008 case 1: gen_helper_neon_sub_u16(t0, t1, t0); break;
3009 case 2: tcg_gen_sub_i32(t0, t1, t0); break;
3014 static TCGv_i32 neon_load_scratch(int scratch)
3016 TCGv_i32 tmp = tcg_temp_new_i32();
3017 tcg_gen_ld_i32(tmp, cpu_env, offsetof(CPUARMState, vfp.scratch[scratch]));
3021 static void neon_store_scratch(int scratch, TCGv_i32 var)
3023 tcg_gen_st_i32(var, cpu_env, offsetof(CPUARMState, vfp.scratch[scratch]));
3024 tcg_temp_free_i32(var);
3027 static inline TCGv_i32 neon_get_scalar(int size, int reg)
3031 tmp = neon_load_reg(reg & 7, reg >> 4);
3033 gen_neon_dup_high16(tmp);
3035 gen_neon_dup_low16(tmp);
3038 tmp = neon_load_reg(reg & 15, reg >> 4);
3043 static int gen_neon_unzip(int rd, int rm, int size, int q)
3047 if (!q && size == 2) {
3050 pd = vfp_reg_ptr(true, rd);
3051 pm = vfp_reg_ptr(true, rm);
3055 gen_helper_neon_qunzip8(pd, pm);
3058 gen_helper_neon_qunzip16(pd, pm);
3061 gen_helper_neon_qunzip32(pd, pm);
3069 gen_helper_neon_unzip8(pd, pm);
3072 gen_helper_neon_unzip16(pd, pm);
3078 tcg_temp_free_ptr(pd);
3079 tcg_temp_free_ptr(pm);
3083 static int gen_neon_zip(int rd, int rm, int size, int q)
3087 if (!q && size == 2) {
3090 pd = vfp_reg_ptr(true, rd);
3091 pm = vfp_reg_ptr(true, rm);
3095 gen_helper_neon_qzip8(pd, pm);
3098 gen_helper_neon_qzip16(pd, pm);
3101 gen_helper_neon_qzip32(pd, pm);
3109 gen_helper_neon_zip8(pd, pm);
3112 gen_helper_neon_zip16(pd, pm);
3118 tcg_temp_free_ptr(pd);
3119 tcg_temp_free_ptr(pm);
3123 static void gen_neon_trn_u8(TCGv_i32 t0, TCGv_i32 t1)
3127 rd = tcg_temp_new_i32();
3128 tmp = tcg_temp_new_i32();
3130 tcg_gen_shli_i32(rd, t0, 8);
3131 tcg_gen_andi_i32(rd, rd, 0xff00ff00);
3132 tcg_gen_andi_i32(tmp, t1, 0x00ff00ff);
3133 tcg_gen_or_i32(rd, rd, tmp);
3135 tcg_gen_shri_i32(t1, t1, 8);
3136 tcg_gen_andi_i32(t1, t1, 0x00ff00ff);
3137 tcg_gen_andi_i32(tmp, t0, 0xff00ff00);
3138 tcg_gen_or_i32(t1, t1, tmp);
3139 tcg_gen_mov_i32(t0, rd);
3141 tcg_temp_free_i32(tmp);
3142 tcg_temp_free_i32(rd);
3145 static void gen_neon_trn_u16(TCGv_i32 t0, TCGv_i32 t1)
3149 rd = tcg_temp_new_i32();
3150 tmp = tcg_temp_new_i32();
3152 tcg_gen_shli_i32(rd, t0, 16);
3153 tcg_gen_andi_i32(tmp, t1, 0xffff);
3154 tcg_gen_or_i32(rd, rd, tmp);
3155 tcg_gen_shri_i32(t1, t1, 16);
3156 tcg_gen_andi_i32(tmp, t0, 0xffff0000);
3157 tcg_gen_or_i32(t1, t1, tmp);
3158 tcg_gen_mov_i32(t0, rd);
3160 tcg_temp_free_i32(tmp);
3161 tcg_temp_free_i32(rd);
3164 static inline void gen_neon_narrow(int size, TCGv_i32 dest, TCGv_i64 src)
3167 case 0: gen_helper_neon_narrow_u8(dest, src); break;
3168 case 1: gen_helper_neon_narrow_u16(dest, src); break;
3169 case 2: tcg_gen_extrl_i64_i32(dest, src); break;
3174 static inline void gen_neon_narrow_sats(int size, TCGv_i32 dest, TCGv_i64 src)
3177 case 0: gen_helper_neon_narrow_sat_s8(dest, cpu_env, src); break;
3178 case 1: gen_helper_neon_narrow_sat_s16(dest, cpu_env, src); break;
3179 case 2: gen_helper_neon_narrow_sat_s32(dest, cpu_env, src); break;
3184 static inline void gen_neon_narrow_satu(int size, TCGv_i32 dest, TCGv_i64 src)
3187 case 0: gen_helper_neon_narrow_sat_u8(dest, cpu_env, src); break;
3188 case 1: gen_helper_neon_narrow_sat_u16(dest, cpu_env, src); break;
3189 case 2: gen_helper_neon_narrow_sat_u32(dest, cpu_env, src); break;
3194 static inline void gen_neon_unarrow_sats(int size, TCGv_i32 dest, TCGv_i64 src)
3197 case 0: gen_helper_neon_unarrow_sat8(dest, cpu_env, src); break;
3198 case 1: gen_helper_neon_unarrow_sat16(dest, cpu_env, src); break;
3199 case 2: gen_helper_neon_unarrow_sat32(dest, cpu_env, src); break;
3204 static inline void gen_neon_widen(TCGv_i64 dest, TCGv_i32 src, int size, int u)
3208 case 0: gen_helper_neon_widen_u8(dest, src); break;
3209 case 1: gen_helper_neon_widen_u16(dest, src); break;
3210 case 2: tcg_gen_extu_i32_i64(dest, src); break;
3215 case 0: gen_helper_neon_widen_s8(dest, src); break;
3216 case 1: gen_helper_neon_widen_s16(dest, src); break;
3217 case 2: tcg_gen_ext_i32_i64(dest, src); break;
3221 tcg_temp_free_i32(src);
3224 static inline void gen_neon_addl(int size)
3227 case 0: gen_helper_neon_addl_u16(CPU_V001); break;
3228 case 1: gen_helper_neon_addl_u32(CPU_V001); break;
3229 case 2: tcg_gen_add_i64(CPU_V001); break;
3234 static inline void gen_neon_subl(int size)
3237 case 0: gen_helper_neon_subl_u16(CPU_V001); break;
3238 case 1: gen_helper_neon_subl_u32(CPU_V001); break;
3239 case 2: tcg_gen_sub_i64(CPU_V001); break;
3244 static inline void gen_neon_negl(TCGv_i64 var, int size)
3247 case 0: gen_helper_neon_negl_u16(var, var); break;
3248 case 1: gen_helper_neon_negl_u32(var, var); break;
3250 tcg_gen_neg_i64(var, var);
3256 static inline void gen_neon_addl_saturate(TCGv_i64 op0, TCGv_i64 op1, int size)
3259 case 1: gen_helper_neon_addl_saturate_s32(op0, cpu_env, op0, op1); break;
3260 case 2: gen_helper_neon_addl_saturate_s64(op0, cpu_env, op0, op1); break;
3265 static inline void gen_neon_mull(TCGv_i64 dest, TCGv_i32 a, TCGv_i32 b,
3270 switch ((size << 1) | u) {
3271 case 0: gen_helper_neon_mull_s8(dest, a, b); break;
3272 case 1: gen_helper_neon_mull_u8(dest, a, b); break;
3273 case 2: gen_helper_neon_mull_s16(dest, a, b); break;
3274 case 3: gen_helper_neon_mull_u16(dest, a, b); break;
3276 tmp = gen_muls_i64_i32(a, b);
3277 tcg_gen_mov_i64(dest, tmp);
3278 tcg_temp_free_i64(tmp);
3281 tmp = gen_mulu_i64_i32(a, b);
3282 tcg_gen_mov_i64(dest, tmp);
3283 tcg_temp_free_i64(tmp);
3288 /* gen_helper_neon_mull_[su]{8|16} do not free their parameters.
3289 Don't forget to clean them now. */
3291 tcg_temp_free_i32(a);
3292 tcg_temp_free_i32(b);
3296 static void gen_neon_narrow_op(int op, int u, int size,
3297 TCGv_i32 dest, TCGv_i64 src)
3301 gen_neon_unarrow_sats(size, dest, src);
3303 gen_neon_narrow(size, dest, src);
3307 gen_neon_narrow_satu(size, dest, src);
3309 gen_neon_narrow_sats(size, dest, src);
3314 /* Symbolic constants for op fields for Neon 2-register miscellaneous.
3315 * The values correspond to bits [17:16,10:7]; see the ARM ARM DDI0406B
3318 #define NEON_2RM_VREV64 0
3319 #define NEON_2RM_VREV32 1
3320 #define NEON_2RM_VREV16 2
3321 #define NEON_2RM_VPADDL 4
3322 #define NEON_2RM_VPADDL_U 5
3323 #define NEON_2RM_AESE 6 /* Includes AESD */
3324 #define NEON_2RM_AESMC 7 /* Includes AESIMC */
3325 #define NEON_2RM_VCLS 8
3326 #define NEON_2RM_VCLZ 9
3327 #define NEON_2RM_VCNT 10
3328 #define NEON_2RM_VMVN 11
3329 #define NEON_2RM_VPADAL 12
3330 #define NEON_2RM_VPADAL_U 13
3331 #define NEON_2RM_VQABS 14
3332 #define NEON_2RM_VQNEG 15
3333 #define NEON_2RM_VCGT0 16
3334 #define NEON_2RM_VCGE0 17
3335 #define NEON_2RM_VCEQ0 18
3336 #define NEON_2RM_VCLE0 19
3337 #define NEON_2RM_VCLT0 20
3338 #define NEON_2RM_SHA1H 21
3339 #define NEON_2RM_VABS 22
3340 #define NEON_2RM_VNEG 23
3341 #define NEON_2RM_VCGT0_F 24
3342 #define NEON_2RM_VCGE0_F 25
3343 #define NEON_2RM_VCEQ0_F 26
3344 #define NEON_2RM_VCLE0_F 27
3345 #define NEON_2RM_VCLT0_F 28
3346 #define NEON_2RM_VABS_F 30
3347 #define NEON_2RM_VNEG_F 31
3348 #define NEON_2RM_VSWP 32
3349 #define NEON_2RM_VTRN 33
3350 #define NEON_2RM_VUZP 34
3351 #define NEON_2RM_VZIP 35
3352 #define NEON_2RM_VMOVN 36 /* Includes VQMOVN, VQMOVUN */
3353 #define NEON_2RM_VQMOVN 37 /* Includes VQMOVUN */
3354 #define NEON_2RM_VSHLL 38
3355 #define NEON_2RM_SHA1SU1 39 /* Includes SHA256SU0 */
3356 #define NEON_2RM_VRINTN 40
3357 #define NEON_2RM_VRINTX 41
3358 #define NEON_2RM_VRINTA 42
3359 #define NEON_2RM_VRINTZ 43
3360 #define NEON_2RM_VCVT_F16_F32 44
3361 #define NEON_2RM_VRINTM 45
3362 #define NEON_2RM_VCVT_F32_F16 46
3363 #define NEON_2RM_VRINTP 47
3364 #define NEON_2RM_VCVTAU 48
3365 #define NEON_2RM_VCVTAS 49
3366 #define NEON_2RM_VCVTNU 50
3367 #define NEON_2RM_VCVTNS 51
3368 #define NEON_2RM_VCVTPU 52
3369 #define NEON_2RM_VCVTPS 53
3370 #define NEON_2RM_VCVTMU 54
3371 #define NEON_2RM_VCVTMS 55
3372 #define NEON_2RM_VRECPE 56
3373 #define NEON_2RM_VRSQRTE 57
3374 #define NEON_2RM_VRECPE_F 58
3375 #define NEON_2RM_VRSQRTE_F 59
3376 #define NEON_2RM_VCVT_FS 60
3377 #define NEON_2RM_VCVT_FU 61
3378 #define NEON_2RM_VCVT_SF 62
3379 #define NEON_2RM_VCVT_UF 63
3381 static bool neon_2rm_is_v8_op(int op)
3383 /* Return true if this neon 2reg-misc op is ARMv8 and up */
3385 case NEON_2RM_VRINTN:
3386 case NEON_2RM_VRINTA:
3387 case NEON_2RM_VRINTM:
3388 case NEON_2RM_VRINTP:
3389 case NEON_2RM_VRINTZ:
3390 case NEON_2RM_VRINTX:
3391 case NEON_2RM_VCVTAU:
3392 case NEON_2RM_VCVTAS:
3393 case NEON_2RM_VCVTNU:
3394 case NEON_2RM_VCVTNS:
3395 case NEON_2RM_VCVTPU:
3396 case NEON_2RM_VCVTPS:
3397 case NEON_2RM_VCVTMU:
3398 case NEON_2RM_VCVTMS:
3405 /* Each entry in this array has bit n set if the insn allows
3406 * size value n (otherwise it will UNDEF). Since unallocated
3407 * op values will have no bits set they always UNDEF.
3409 static const uint8_t neon_2rm_sizes[] = {
3410 [NEON_2RM_VREV64] = 0x7,
3411 [NEON_2RM_VREV32] = 0x3,
3412 [NEON_2RM_VREV16] = 0x1,
3413 [NEON_2RM_VPADDL] = 0x7,
3414 [NEON_2RM_VPADDL_U] = 0x7,
3415 [NEON_2RM_AESE] = 0x1,
3416 [NEON_2RM_AESMC] = 0x1,
3417 [NEON_2RM_VCLS] = 0x7,
3418 [NEON_2RM_VCLZ] = 0x7,
3419 [NEON_2RM_VCNT] = 0x1,
3420 [NEON_2RM_VMVN] = 0x1,
3421 [NEON_2RM_VPADAL] = 0x7,
3422 [NEON_2RM_VPADAL_U] = 0x7,
3423 [NEON_2RM_VQABS] = 0x7,
3424 [NEON_2RM_VQNEG] = 0x7,
3425 [NEON_2RM_VCGT0] = 0x7,
3426 [NEON_2RM_VCGE0] = 0x7,
3427 [NEON_2RM_VCEQ0] = 0x7,
3428 [NEON_2RM_VCLE0] = 0x7,
3429 [NEON_2RM_VCLT0] = 0x7,
3430 [NEON_2RM_SHA1H] = 0x4,
3431 [NEON_2RM_VABS] = 0x7,
3432 [NEON_2RM_VNEG] = 0x7,
3433 [NEON_2RM_VCGT0_F] = 0x4,
3434 [NEON_2RM_VCGE0_F] = 0x4,
3435 [NEON_2RM_VCEQ0_F] = 0x4,
3436 [NEON_2RM_VCLE0_F] = 0x4,
3437 [NEON_2RM_VCLT0_F] = 0x4,
3438 [NEON_2RM_VABS_F] = 0x4,
3439 [NEON_2RM_VNEG_F] = 0x4,
3440 [NEON_2RM_VSWP] = 0x1,
3441 [NEON_2RM_VTRN] = 0x7,
3442 [NEON_2RM_VUZP] = 0x7,
3443 [NEON_2RM_VZIP] = 0x7,
3444 [NEON_2RM_VMOVN] = 0x7,
3445 [NEON_2RM_VQMOVN] = 0x7,
3446 [NEON_2RM_VSHLL] = 0x7,
3447 [NEON_2RM_SHA1SU1] = 0x4,
3448 [NEON_2RM_VRINTN] = 0x4,
3449 [NEON_2RM_VRINTX] = 0x4,
3450 [NEON_2RM_VRINTA] = 0x4,
3451 [NEON_2RM_VRINTZ] = 0x4,
3452 [NEON_2RM_VCVT_F16_F32] = 0x2,
3453 [NEON_2RM_VRINTM] = 0x4,
3454 [NEON_2RM_VCVT_F32_F16] = 0x2,
3455 [NEON_2RM_VRINTP] = 0x4,
3456 [NEON_2RM_VCVTAU] = 0x4,
3457 [NEON_2RM_VCVTAS] = 0x4,
3458 [NEON_2RM_VCVTNU] = 0x4,
3459 [NEON_2RM_VCVTNS] = 0x4,
3460 [NEON_2RM_VCVTPU] = 0x4,
3461 [NEON_2RM_VCVTPS] = 0x4,
3462 [NEON_2RM_VCVTMU] = 0x4,
3463 [NEON_2RM_VCVTMS] = 0x4,
3464 [NEON_2RM_VRECPE] = 0x4,
3465 [NEON_2RM_VRSQRTE] = 0x4,
3466 [NEON_2RM_VRECPE_F] = 0x4,
3467 [NEON_2RM_VRSQRTE_F] = 0x4,
3468 [NEON_2RM_VCVT_FS] = 0x4,
3469 [NEON_2RM_VCVT_FU] = 0x4,
3470 [NEON_2RM_VCVT_SF] = 0x4,
3471 [NEON_2RM_VCVT_UF] = 0x4,
3474 static void gen_gvec_fn3_qc(uint32_t rd_ofs, uint32_t rn_ofs, uint32_t rm_ofs,
3475 uint32_t opr_sz, uint32_t max_sz,
3476 gen_helper_gvec_3_ptr *fn)
3478 TCGv_ptr qc_ptr = tcg_temp_new_ptr();
3480 tcg_gen_addi_ptr(qc_ptr, cpu_env, offsetof(CPUARMState, vfp.qc));
3481 tcg_gen_gvec_3_ptr(rd_ofs, rn_ofs, rm_ofs, qc_ptr,
3482 opr_sz, max_sz, 0, fn);
3483 tcg_temp_free_ptr(qc_ptr);
3486 void gen_gvec_sqrdmlah_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
3487 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
3489 static gen_helper_gvec_3_ptr * const fns[2] = {
3490 gen_helper_gvec_qrdmlah_s16, gen_helper_gvec_qrdmlah_s32
3492 tcg_debug_assert(vece >= 1 && vece <= 2);
3493 gen_gvec_fn3_qc(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, fns[vece - 1]);
3496 void gen_gvec_sqrdmlsh_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
3497 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
3499 static gen_helper_gvec_3_ptr * const fns[2] = {
3500 gen_helper_gvec_qrdmlsh_s16, gen_helper_gvec_qrdmlsh_s32
3502 tcg_debug_assert(vece >= 1 && vece <= 2);
3503 gen_gvec_fn3_qc(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, fns[vece - 1]);
3506 #define GEN_CMP0(NAME, COND) \
3507 static void gen_##NAME##0_i32(TCGv_i32 d, TCGv_i32 a) \
3509 tcg_gen_setcondi_i32(COND, d, a, 0); \
3510 tcg_gen_neg_i32(d, d); \
3512 static void gen_##NAME##0_i64(TCGv_i64 d, TCGv_i64 a) \
3514 tcg_gen_setcondi_i64(COND, d, a, 0); \
3515 tcg_gen_neg_i64(d, d); \
3517 static void gen_##NAME##0_vec(unsigned vece, TCGv_vec d, TCGv_vec a) \
3519 TCGv_vec zero = tcg_const_zeros_vec_matching(d); \
3520 tcg_gen_cmp_vec(COND, vece, d, a, zero); \
3521 tcg_temp_free_vec(zero); \
3523 void gen_gvec_##NAME##0(unsigned vece, uint32_t d, uint32_t m, \
3524 uint32_t opr_sz, uint32_t max_sz) \
3526 const GVecGen2 op[4] = { \
3527 { .fno = gen_helper_gvec_##NAME##0_b, \
3528 .fniv = gen_##NAME##0_vec, \
3529 .opt_opc = vecop_list_cmp, \
3531 { .fno = gen_helper_gvec_##NAME##0_h, \
3532 .fniv = gen_##NAME##0_vec, \
3533 .opt_opc = vecop_list_cmp, \
3535 { .fni4 = gen_##NAME##0_i32, \
3536 .fniv = gen_##NAME##0_vec, \
3537 .opt_opc = vecop_list_cmp, \
3539 { .fni8 = gen_##NAME##0_i64, \
3540 .fniv = gen_##NAME##0_vec, \
3541 .opt_opc = vecop_list_cmp, \
3542 .prefer_i64 = TCG_TARGET_REG_BITS == 64, \
3545 tcg_gen_gvec_2(d, m, opr_sz, max_sz, &op[vece]); \
3548 static const TCGOpcode vecop_list_cmp[] = {
3552 GEN_CMP0(ceq, TCG_COND_EQ)
3553 GEN_CMP0(cle, TCG_COND_LE)
3554 GEN_CMP0(cge, TCG_COND_GE)
3555 GEN_CMP0(clt, TCG_COND_LT)
3556 GEN_CMP0(cgt, TCG_COND_GT)
3560 static void gen_ssra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3562 tcg_gen_vec_sar8i_i64(a, a, shift);
3563 tcg_gen_vec_add8_i64(d, d, a);
3566 static void gen_ssra16_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3568 tcg_gen_vec_sar16i_i64(a, a, shift);
3569 tcg_gen_vec_add16_i64(d, d, a);
3572 static void gen_ssra32_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
3574 tcg_gen_sari_i32(a, a, shift);
3575 tcg_gen_add_i32(d, d, a);
3578 static void gen_ssra64_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3580 tcg_gen_sari_i64(a, a, shift);
3581 tcg_gen_add_i64(d, d, a);
3584 static void gen_ssra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
3586 tcg_gen_sari_vec(vece, a, a, sh);
3587 tcg_gen_add_vec(vece, d, d, a);
3590 void gen_gvec_ssra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3591 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
3593 static const TCGOpcode vecop_list[] = {
3594 INDEX_op_sari_vec, INDEX_op_add_vec, 0
3596 static const GVecGen2i ops[4] = {
3597 { .fni8 = gen_ssra8_i64,
3598 .fniv = gen_ssra_vec,
3599 .fno = gen_helper_gvec_ssra_b,
3601 .opt_opc = vecop_list,
3603 { .fni8 = gen_ssra16_i64,
3604 .fniv = gen_ssra_vec,
3605 .fno = gen_helper_gvec_ssra_h,
3607 .opt_opc = vecop_list,
3609 { .fni4 = gen_ssra32_i32,
3610 .fniv = gen_ssra_vec,
3611 .fno = gen_helper_gvec_ssra_s,
3613 .opt_opc = vecop_list,
3615 { .fni8 = gen_ssra64_i64,
3616 .fniv = gen_ssra_vec,
3617 .fno = gen_helper_gvec_ssra_b,
3618 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3619 .opt_opc = vecop_list,
3624 /* tszimm encoding produces immediates in the range [1..esize]. */
3625 tcg_debug_assert(shift > 0);
3626 tcg_debug_assert(shift <= (8 << vece));
3629 * Shifts larger than the element size are architecturally valid.
3630 * Signed results in all sign bits.
3632 shift = MIN(shift, (8 << vece) - 1);
3633 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
3636 static void gen_usra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3638 tcg_gen_vec_shr8i_i64(a, a, shift);
3639 tcg_gen_vec_add8_i64(d, d, a);
3642 static void gen_usra16_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3644 tcg_gen_vec_shr16i_i64(a, a, shift);
3645 tcg_gen_vec_add16_i64(d, d, a);
3648 static void gen_usra32_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
3650 tcg_gen_shri_i32(a, a, shift);
3651 tcg_gen_add_i32(d, d, a);
3654 static void gen_usra64_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
3656 tcg_gen_shri_i64(a, a, shift);
3657 tcg_gen_add_i64(d, d, a);
3660 static void gen_usra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
3662 tcg_gen_shri_vec(vece, a, a, sh);
3663 tcg_gen_add_vec(vece, d, d, a);
3666 void gen_gvec_usra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3667 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
3669 static const TCGOpcode vecop_list[] = {
3670 INDEX_op_shri_vec, INDEX_op_add_vec, 0
3672 static const GVecGen2i ops[4] = {
3673 { .fni8 = gen_usra8_i64,
3674 .fniv = gen_usra_vec,
3675 .fno = gen_helper_gvec_usra_b,
3677 .opt_opc = vecop_list,
3679 { .fni8 = gen_usra16_i64,
3680 .fniv = gen_usra_vec,
3681 .fno = gen_helper_gvec_usra_h,
3683 .opt_opc = vecop_list,
3685 { .fni4 = gen_usra32_i32,
3686 .fniv = gen_usra_vec,
3687 .fno = gen_helper_gvec_usra_s,
3689 .opt_opc = vecop_list,
3691 { .fni8 = gen_usra64_i64,
3692 .fniv = gen_usra_vec,
3693 .fno = gen_helper_gvec_usra_d,
3694 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3696 .opt_opc = vecop_list,
3700 /* tszimm encoding produces immediates in the range [1..esize]. */
3701 tcg_debug_assert(shift > 0);
3702 tcg_debug_assert(shift <= (8 << vece));
3705 * Shifts larger than the element size are architecturally valid.
3706 * Unsigned results in all zeros as input to accumulate: nop.
3708 if (shift < (8 << vece)) {
3709 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
3711 /* Nop, but we do need to clear the tail. */
3712 tcg_gen_gvec_mov(vece, rd_ofs, rd_ofs, opr_sz, max_sz);
3717 * Shift one less than the requested amount, and the low bit is
3718 * the rounding bit. For the 8 and 16-bit operations, because we
3719 * mask the low bit, we can perform a normal integer shift instead
3720 * of a vector shift.
3722 static void gen_srshr8_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3724 TCGv_i64 t = tcg_temp_new_i64();
3726 tcg_gen_shri_i64(t, a, sh - 1);
3727 tcg_gen_andi_i64(t, t, dup_const(MO_8, 1));
3728 tcg_gen_vec_sar8i_i64(d, a, sh);
3729 tcg_gen_vec_add8_i64(d, d, t);
3730 tcg_temp_free_i64(t);
3733 static void gen_srshr16_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3735 TCGv_i64 t = tcg_temp_new_i64();
3737 tcg_gen_shri_i64(t, a, sh - 1);
3738 tcg_gen_andi_i64(t, t, dup_const(MO_16, 1));
3739 tcg_gen_vec_sar16i_i64(d, a, sh);
3740 tcg_gen_vec_add16_i64(d, d, t);
3741 tcg_temp_free_i64(t);
3744 static void gen_srshr32_i32(TCGv_i32 d, TCGv_i32 a, int32_t sh)
3746 TCGv_i32 t = tcg_temp_new_i32();
3748 tcg_gen_extract_i32(t, a, sh - 1, 1);
3749 tcg_gen_sari_i32(d, a, sh);
3750 tcg_gen_add_i32(d, d, t);
3751 tcg_temp_free_i32(t);
3754 static void gen_srshr64_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3756 TCGv_i64 t = tcg_temp_new_i64();
3758 tcg_gen_extract_i64(t, a, sh - 1, 1);
3759 tcg_gen_sari_i64(d, a, sh);
3760 tcg_gen_add_i64(d, d, t);
3761 tcg_temp_free_i64(t);
3764 static void gen_srshr_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
3766 TCGv_vec t = tcg_temp_new_vec_matching(d);
3767 TCGv_vec ones = tcg_temp_new_vec_matching(d);
3769 tcg_gen_shri_vec(vece, t, a, sh - 1);
3770 tcg_gen_dupi_vec(vece, ones, 1);
3771 tcg_gen_and_vec(vece, t, t, ones);
3772 tcg_gen_sari_vec(vece, d, a, sh);
3773 tcg_gen_add_vec(vece, d, d, t);
3775 tcg_temp_free_vec(t);
3776 tcg_temp_free_vec(ones);
3779 void gen_gvec_srshr(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3780 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
3782 static const TCGOpcode vecop_list[] = {
3783 INDEX_op_shri_vec, INDEX_op_sari_vec, INDEX_op_add_vec, 0
3785 static const GVecGen2i ops[4] = {
3786 { .fni8 = gen_srshr8_i64,
3787 .fniv = gen_srshr_vec,
3788 .fno = gen_helper_gvec_srshr_b,
3789 .opt_opc = vecop_list,
3791 { .fni8 = gen_srshr16_i64,
3792 .fniv = gen_srshr_vec,
3793 .fno = gen_helper_gvec_srshr_h,
3794 .opt_opc = vecop_list,
3796 { .fni4 = gen_srshr32_i32,
3797 .fniv = gen_srshr_vec,
3798 .fno = gen_helper_gvec_srshr_s,
3799 .opt_opc = vecop_list,
3801 { .fni8 = gen_srshr64_i64,
3802 .fniv = gen_srshr_vec,
3803 .fno = gen_helper_gvec_srshr_d,
3804 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3805 .opt_opc = vecop_list,
3809 /* tszimm encoding produces immediates in the range [1..esize] */
3810 tcg_debug_assert(shift > 0);
3811 tcg_debug_assert(shift <= (8 << vece));
3813 if (shift == (8 << vece)) {
3815 * Shifts larger than the element size are architecturally valid.
3816 * Signed results in all sign bits. With rounding, this produces
3817 * (-1 + 1) >> 1 == 0, or (0 + 1) >> 1 == 0.
3820 tcg_gen_gvec_dup_imm(vece, rd_ofs, opr_sz, max_sz, 0);
3822 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
3826 static void gen_srsra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3828 TCGv_i64 t = tcg_temp_new_i64();
3830 gen_srshr8_i64(t, a, sh);
3831 tcg_gen_vec_add8_i64(d, d, t);
3832 tcg_temp_free_i64(t);
3835 static void gen_srsra16_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3837 TCGv_i64 t = tcg_temp_new_i64();
3839 gen_srshr16_i64(t, a, sh);
3840 tcg_gen_vec_add16_i64(d, d, t);
3841 tcg_temp_free_i64(t);
3844 static void gen_srsra32_i32(TCGv_i32 d, TCGv_i32 a, int32_t sh)
3846 TCGv_i32 t = tcg_temp_new_i32();
3848 gen_srshr32_i32(t, a, sh);
3849 tcg_gen_add_i32(d, d, t);
3850 tcg_temp_free_i32(t);
3853 static void gen_srsra64_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3855 TCGv_i64 t = tcg_temp_new_i64();
3857 gen_srshr64_i64(t, a, sh);
3858 tcg_gen_add_i64(d, d, t);
3859 tcg_temp_free_i64(t);
3862 static void gen_srsra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
3864 TCGv_vec t = tcg_temp_new_vec_matching(d);
3866 gen_srshr_vec(vece, t, a, sh);
3867 tcg_gen_add_vec(vece, d, d, t);
3868 tcg_temp_free_vec(t);
3871 void gen_gvec_srsra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3872 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
3874 static const TCGOpcode vecop_list[] = {
3875 INDEX_op_shri_vec, INDEX_op_sari_vec, INDEX_op_add_vec, 0
3877 static const GVecGen2i ops[4] = {
3878 { .fni8 = gen_srsra8_i64,
3879 .fniv = gen_srsra_vec,
3880 .fno = gen_helper_gvec_srsra_b,
3881 .opt_opc = vecop_list,
3884 { .fni8 = gen_srsra16_i64,
3885 .fniv = gen_srsra_vec,
3886 .fno = gen_helper_gvec_srsra_h,
3887 .opt_opc = vecop_list,
3890 { .fni4 = gen_srsra32_i32,
3891 .fniv = gen_srsra_vec,
3892 .fno = gen_helper_gvec_srsra_s,
3893 .opt_opc = vecop_list,
3896 { .fni8 = gen_srsra64_i64,
3897 .fniv = gen_srsra_vec,
3898 .fno = gen_helper_gvec_srsra_d,
3899 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3900 .opt_opc = vecop_list,
3905 /* tszimm encoding produces immediates in the range [1..esize] */
3906 tcg_debug_assert(shift > 0);
3907 tcg_debug_assert(shift <= (8 << vece));
3910 * Shifts larger than the element size are architecturally valid.
3911 * Signed results in all sign bits. With rounding, this produces
3912 * (-1 + 1) >> 1 == 0, or (0 + 1) >> 1 == 0.
3913 * I.e. always zero. With accumulation, this leaves D unchanged.
3915 if (shift == (8 << vece)) {
3916 /* Nop, but we do need to clear the tail. */
3917 tcg_gen_gvec_mov(vece, rd_ofs, rd_ofs, opr_sz, max_sz);
3919 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
3923 static void gen_urshr8_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3925 TCGv_i64 t = tcg_temp_new_i64();
3927 tcg_gen_shri_i64(t, a, sh - 1);
3928 tcg_gen_andi_i64(t, t, dup_const(MO_8, 1));
3929 tcg_gen_vec_shr8i_i64(d, a, sh);
3930 tcg_gen_vec_add8_i64(d, d, t);
3931 tcg_temp_free_i64(t);
3934 static void gen_urshr16_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3936 TCGv_i64 t = tcg_temp_new_i64();
3938 tcg_gen_shri_i64(t, a, sh - 1);
3939 tcg_gen_andi_i64(t, t, dup_const(MO_16, 1));
3940 tcg_gen_vec_shr16i_i64(d, a, sh);
3941 tcg_gen_vec_add16_i64(d, d, t);
3942 tcg_temp_free_i64(t);
3945 static void gen_urshr32_i32(TCGv_i32 d, TCGv_i32 a, int32_t sh)
3947 TCGv_i32 t = tcg_temp_new_i32();
3949 tcg_gen_extract_i32(t, a, sh - 1, 1);
3950 tcg_gen_shri_i32(d, a, sh);
3951 tcg_gen_add_i32(d, d, t);
3952 tcg_temp_free_i32(t);
3955 static void gen_urshr64_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
3957 TCGv_i64 t = tcg_temp_new_i64();
3959 tcg_gen_extract_i64(t, a, sh - 1, 1);
3960 tcg_gen_shri_i64(d, a, sh);
3961 tcg_gen_add_i64(d, d, t);
3962 tcg_temp_free_i64(t);
3965 static void gen_urshr_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t shift)
3967 TCGv_vec t = tcg_temp_new_vec_matching(d);
3968 TCGv_vec ones = tcg_temp_new_vec_matching(d);
3970 tcg_gen_shri_vec(vece, t, a, shift - 1);
3971 tcg_gen_dupi_vec(vece, ones, 1);
3972 tcg_gen_and_vec(vece, t, t, ones);
3973 tcg_gen_shri_vec(vece, d, a, shift);
3974 tcg_gen_add_vec(vece, d, d, t);
3976 tcg_temp_free_vec(t);
3977 tcg_temp_free_vec(ones);
3980 void gen_gvec_urshr(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
3981 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
3983 static const TCGOpcode vecop_list[] = {
3984 INDEX_op_shri_vec, INDEX_op_add_vec, 0
3986 static const GVecGen2i ops[4] = {
3987 { .fni8 = gen_urshr8_i64,
3988 .fniv = gen_urshr_vec,
3989 .fno = gen_helper_gvec_urshr_b,
3990 .opt_opc = vecop_list,
3992 { .fni8 = gen_urshr16_i64,
3993 .fniv = gen_urshr_vec,
3994 .fno = gen_helper_gvec_urshr_h,
3995 .opt_opc = vecop_list,
3997 { .fni4 = gen_urshr32_i32,
3998 .fniv = gen_urshr_vec,
3999 .fno = gen_helper_gvec_urshr_s,
4000 .opt_opc = vecop_list,
4002 { .fni8 = gen_urshr64_i64,
4003 .fniv = gen_urshr_vec,
4004 .fno = gen_helper_gvec_urshr_d,
4005 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4006 .opt_opc = vecop_list,
4010 /* tszimm encoding produces immediates in the range [1..esize] */
4011 tcg_debug_assert(shift > 0);
4012 tcg_debug_assert(shift <= (8 << vece));
4014 if (shift == (8 << vece)) {
4016 * Shifts larger than the element size are architecturally valid.
4017 * Unsigned results in zero. With rounding, this produces a
4018 * copy of the most significant bit.
4020 tcg_gen_gvec_shri(vece, rd_ofs, rm_ofs, shift - 1, opr_sz, max_sz);
4022 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
4026 static void gen_ursra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
4028 TCGv_i64 t = tcg_temp_new_i64();
4031 tcg_gen_vec_shr8i_i64(t, a, 7);
4033 gen_urshr8_i64(t, a, sh);
4035 tcg_gen_vec_add8_i64(d, d, t);
4036 tcg_temp_free_i64(t);
4039 static void gen_ursra16_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
4041 TCGv_i64 t = tcg_temp_new_i64();
4044 tcg_gen_vec_shr16i_i64(t, a, 15);
4046 gen_urshr16_i64(t, a, sh);
4048 tcg_gen_vec_add16_i64(d, d, t);
4049 tcg_temp_free_i64(t);
4052 static void gen_ursra32_i32(TCGv_i32 d, TCGv_i32 a, int32_t sh)
4054 TCGv_i32 t = tcg_temp_new_i32();
4057 tcg_gen_shri_i32(t, a, 31);
4059 gen_urshr32_i32(t, a, sh);
4061 tcg_gen_add_i32(d, d, t);
4062 tcg_temp_free_i32(t);
4065 static void gen_ursra64_i64(TCGv_i64 d, TCGv_i64 a, int64_t sh)
4067 TCGv_i64 t = tcg_temp_new_i64();
4070 tcg_gen_shri_i64(t, a, 63);
4072 gen_urshr64_i64(t, a, sh);
4074 tcg_gen_add_i64(d, d, t);
4075 tcg_temp_free_i64(t);
4078 static void gen_ursra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
4080 TCGv_vec t = tcg_temp_new_vec_matching(d);
4082 if (sh == (8 << vece)) {
4083 tcg_gen_shri_vec(vece, t, a, sh - 1);
4085 gen_urshr_vec(vece, t, a, sh);
4087 tcg_gen_add_vec(vece, d, d, t);
4088 tcg_temp_free_vec(t);
4091 void gen_gvec_ursra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
4092 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
4094 static const TCGOpcode vecop_list[] = {
4095 INDEX_op_shri_vec, INDEX_op_add_vec, 0
4097 static const GVecGen2i ops[4] = {
4098 { .fni8 = gen_ursra8_i64,
4099 .fniv = gen_ursra_vec,
4100 .fno = gen_helper_gvec_ursra_b,
4101 .opt_opc = vecop_list,
4104 { .fni8 = gen_ursra16_i64,
4105 .fniv = gen_ursra_vec,
4106 .fno = gen_helper_gvec_ursra_h,
4107 .opt_opc = vecop_list,
4110 { .fni4 = gen_ursra32_i32,
4111 .fniv = gen_ursra_vec,
4112 .fno = gen_helper_gvec_ursra_s,
4113 .opt_opc = vecop_list,
4116 { .fni8 = gen_ursra64_i64,
4117 .fniv = gen_ursra_vec,
4118 .fno = gen_helper_gvec_ursra_d,
4119 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4120 .opt_opc = vecop_list,
4125 /* tszimm encoding produces immediates in the range [1..esize] */
4126 tcg_debug_assert(shift > 0);
4127 tcg_debug_assert(shift <= (8 << vece));
4129 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
4132 static void gen_shr8_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
4134 uint64_t mask = dup_const(MO_8, 0xff >> shift);
4135 TCGv_i64 t = tcg_temp_new_i64();
4137 tcg_gen_shri_i64(t, a, shift);
4138 tcg_gen_andi_i64(t, t, mask);
4139 tcg_gen_andi_i64(d, d, ~mask);
4140 tcg_gen_or_i64(d, d, t);
4141 tcg_temp_free_i64(t);
4144 static void gen_shr16_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
4146 uint64_t mask = dup_const(MO_16, 0xffff >> shift);
4147 TCGv_i64 t = tcg_temp_new_i64();
4149 tcg_gen_shri_i64(t, a, shift);
4150 tcg_gen_andi_i64(t, t, mask);
4151 tcg_gen_andi_i64(d, d, ~mask);
4152 tcg_gen_or_i64(d, d, t);
4153 tcg_temp_free_i64(t);
4156 static void gen_shr32_ins_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
4158 tcg_gen_shri_i32(a, a, shift);
4159 tcg_gen_deposit_i32(d, d, a, 0, 32 - shift);
4162 static void gen_shr64_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
4164 tcg_gen_shri_i64(a, a, shift);
4165 tcg_gen_deposit_i64(d, d, a, 0, 64 - shift);
4168 static void gen_shr_ins_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
4170 TCGv_vec t = tcg_temp_new_vec_matching(d);
4171 TCGv_vec m = tcg_temp_new_vec_matching(d);
4173 tcg_gen_dupi_vec(vece, m, MAKE_64BIT_MASK((8 << vece) - sh, sh));
4174 tcg_gen_shri_vec(vece, t, a, sh);
4175 tcg_gen_and_vec(vece, d, d, m);
4176 tcg_gen_or_vec(vece, d, d, t);
4178 tcg_temp_free_vec(t);
4179 tcg_temp_free_vec(m);
4182 void gen_gvec_sri(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
4183 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
4185 static const TCGOpcode vecop_list[] = { INDEX_op_shri_vec, 0 };
4186 const GVecGen2i ops[4] = {
4187 { .fni8 = gen_shr8_ins_i64,
4188 .fniv = gen_shr_ins_vec,
4189 .fno = gen_helper_gvec_sri_b,
4191 .opt_opc = vecop_list,
4193 { .fni8 = gen_shr16_ins_i64,
4194 .fniv = gen_shr_ins_vec,
4195 .fno = gen_helper_gvec_sri_h,
4197 .opt_opc = vecop_list,
4199 { .fni4 = gen_shr32_ins_i32,
4200 .fniv = gen_shr_ins_vec,
4201 .fno = gen_helper_gvec_sri_s,
4203 .opt_opc = vecop_list,
4205 { .fni8 = gen_shr64_ins_i64,
4206 .fniv = gen_shr_ins_vec,
4207 .fno = gen_helper_gvec_sri_d,
4208 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4210 .opt_opc = vecop_list,
4214 /* tszimm encoding produces immediates in the range [1..esize]. */
4215 tcg_debug_assert(shift > 0);
4216 tcg_debug_assert(shift <= (8 << vece));
4218 /* Shift of esize leaves destination unchanged. */
4219 if (shift < (8 << vece)) {
4220 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
4222 /* Nop, but we do need to clear the tail. */
4223 tcg_gen_gvec_mov(vece, rd_ofs, rd_ofs, opr_sz, max_sz);
4227 static void gen_shl8_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
4229 uint64_t mask = dup_const(MO_8, 0xff << shift);
4230 TCGv_i64 t = tcg_temp_new_i64();
4232 tcg_gen_shli_i64(t, a, shift);
4233 tcg_gen_andi_i64(t, t, mask);
4234 tcg_gen_andi_i64(d, d, ~mask);
4235 tcg_gen_or_i64(d, d, t);
4236 tcg_temp_free_i64(t);
4239 static void gen_shl16_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
4241 uint64_t mask = dup_const(MO_16, 0xffff << shift);
4242 TCGv_i64 t = tcg_temp_new_i64();
4244 tcg_gen_shli_i64(t, a, shift);
4245 tcg_gen_andi_i64(t, t, mask);
4246 tcg_gen_andi_i64(d, d, ~mask);
4247 tcg_gen_or_i64(d, d, t);
4248 tcg_temp_free_i64(t);
4251 static void gen_shl32_ins_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
4253 tcg_gen_deposit_i32(d, d, a, shift, 32 - shift);
4256 static void gen_shl64_ins_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
4258 tcg_gen_deposit_i64(d, d, a, shift, 64 - shift);
4261 static void gen_shl_ins_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t sh)
4263 TCGv_vec t = tcg_temp_new_vec_matching(d);
4264 TCGv_vec m = tcg_temp_new_vec_matching(d);
4266 tcg_gen_shli_vec(vece, t, a, sh);
4267 tcg_gen_dupi_vec(vece, m, MAKE_64BIT_MASK(0, sh));
4268 tcg_gen_and_vec(vece, d, d, m);
4269 tcg_gen_or_vec(vece, d, d, t);
4271 tcg_temp_free_vec(t);
4272 tcg_temp_free_vec(m);
4275 void gen_gvec_sli(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
4276 int64_t shift, uint32_t opr_sz, uint32_t max_sz)
4278 static const TCGOpcode vecop_list[] = { INDEX_op_shli_vec, 0 };
4279 const GVecGen2i ops[4] = {
4280 { .fni8 = gen_shl8_ins_i64,
4281 .fniv = gen_shl_ins_vec,
4282 .fno = gen_helper_gvec_sli_b,
4284 .opt_opc = vecop_list,
4286 { .fni8 = gen_shl16_ins_i64,
4287 .fniv = gen_shl_ins_vec,
4288 .fno = gen_helper_gvec_sli_h,
4290 .opt_opc = vecop_list,
4292 { .fni4 = gen_shl32_ins_i32,
4293 .fniv = gen_shl_ins_vec,
4294 .fno = gen_helper_gvec_sli_s,
4296 .opt_opc = vecop_list,
4298 { .fni8 = gen_shl64_ins_i64,
4299 .fniv = gen_shl_ins_vec,
4300 .fno = gen_helper_gvec_sli_d,
4301 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4303 .opt_opc = vecop_list,
4307 /* tszimm encoding produces immediates in the range [0..esize-1]. */
4308 tcg_debug_assert(shift >= 0);
4309 tcg_debug_assert(shift < (8 << vece));
4312 tcg_gen_gvec_mov(vece, rd_ofs, rm_ofs, opr_sz, max_sz);
4314 tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
4318 static void gen_mla8_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4320 gen_helper_neon_mul_u8(a, a, b);
4321 gen_helper_neon_add_u8(d, d, a);
4324 static void gen_mls8_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4326 gen_helper_neon_mul_u8(a, a, b);
4327 gen_helper_neon_sub_u8(d, d, a);
4330 static void gen_mla16_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4332 gen_helper_neon_mul_u16(a, a, b);
4333 gen_helper_neon_add_u16(d, d, a);
4336 static void gen_mls16_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4338 gen_helper_neon_mul_u16(a, a, b);
4339 gen_helper_neon_sub_u16(d, d, a);
4342 static void gen_mla32_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4344 tcg_gen_mul_i32(a, a, b);
4345 tcg_gen_add_i32(d, d, a);
4348 static void gen_mls32_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4350 tcg_gen_mul_i32(a, a, b);
4351 tcg_gen_sub_i32(d, d, a);
4354 static void gen_mla64_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
4356 tcg_gen_mul_i64(a, a, b);
4357 tcg_gen_add_i64(d, d, a);
4360 static void gen_mls64_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
4362 tcg_gen_mul_i64(a, a, b);
4363 tcg_gen_sub_i64(d, d, a);
4366 static void gen_mla_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
4368 tcg_gen_mul_vec(vece, a, a, b);
4369 tcg_gen_add_vec(vece, d, d, a);
4372 static void gen_mls_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
4374 tcg_gen_mul_vec(vece, a, a, b);
4375 tcg_gen_sub_vec(vece, d, d, a);
4378 /* Note that while NEON does not support VMLA and VMLS as 64-bit ops,
4379 * these tables are shared with AArch64 which does support them.
4381 void gen_gvec_mla(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4382 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4384 static const TCGOpcode vecop_list[] = {
4385 INDEX_op_mul_vec, INDEX_op_add_vec, 0
4387 static const GVecGen3 ops[4] = {
4388 { .fni4 = gen_mla8_i32,
4389 .fniv = gen_mla_vec,
4391 .opt_opc = vecop_list,
4393 { .fni4 = gen_mla16_i32,
4394 .fniv = gen_mla_vec,
4396 .opt_opc = vecop_list,
4398 { .fni4 = gen_mla32_i32,
4399 .fniv = gen_mla_vec,
4401 .opt_opc = vecop_list,
4403 { .fni8 = gen_mla64_i64,
4404 .fniv = gen_mla_vec,
4405 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4407 .opt_opc = vecop_list,
4410 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4413 void gen_gvec_mls(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4414 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4416 static const TCGOpcode vecop_list[] = {
4417 INDEX_op_mul_vec, INDEX_op_sub_vec, 0
4419 static const GVecGen3 ops[4] = {
4420 { .fni4 = gen_mls8_i32,
4421 .fniv = gen_mls_vec,
4423 .opt_opc = vecop_list,
4425 { .fni4 = gen_mls16_i32,
4426 .fniv = gen_mls_vec,
4428 .opt_opc = vecop_list,
4430 { .fni4 = gen_mls32_i32,
4431 .fniv = gen_mls_vec,
4433 .opt_opc = vecop_list,
4435 { .fni8 = gen_mls64_i64,
4436 .fniv = gen_mls_vec,
4437 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4439 .opt_opc = vecop_list,
4442 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4445 /* CMTST : test is "if (X & Y != 0)". */
4446 static void gen_cmtst_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4448 tcg_gen_and_i32(d, a, b);
4449 tcg_gen_setcondi_i32(TCG_COND_NE, d, d, 0);
4450 tcg_gen_neg_i32(d, d);
4453 void gen_cmtst_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
4455 tcg_gen_and_i64(d, a, b);
4456 tcg_gen_setcondi_i64(TCG_COND_NE, d, d, 0);
4457 tcg_gen_neg_i64(d, d);
4460 static void gen_cmtst_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
4462 tcg_gen_and_vec(vece, d, a, b);
4463 tcg_gen_dupi_vec(vece, a, 0);
4464 tcg_gen_cmp_vec(TCG_COND_NE, vece, d, d, a);
4467 void gen_gvec_cmtst(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4468 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4470 static const TCGOpcode vecop_list[] = { INDEX_op_cmp_vec, 0 };
4471 static const GVecGen3 ops[4] = {
4472 { .fni4 = gen_helper_neon_tst_u8,
4473 .fniv = gen_cmtst_vec,
4474 .opt_opc = vecop_list,
4476 { .fni4 = gen_helper_neon_tst_u16,
4477 .fniv = gen_cmtst_vec,
4478 .opt_opc = vecop_list,
4480 { .fni4 = gen_cmtst_i32,
4481 .fniv = gen_cmtst_vec,
4482 .opt_opc = vecop_list,
4484 { .fni8 = gen_cmtst_i64,
4485 .fniv = gen_cmtst_vec,
4486 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
4487 .opt_opc = vecop_list,
4490 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4493 void gen_ushl_i32(TCGv_i32 dst, TCGv_i32 src, TCGv_i32 shift)
4495 TCGv_i32 lval = tcg_temp_new_i32();
4496 TCGv_i32 rval = tcg_temp_new_i32();
4497 TCGv_i32 lsh = tcg_temp_new_i32();
4498 TCGv_i32 rsh = tcg_temp_new_i32();
4499 TCGv_i32 zero = tcg_const_i32(0);
4500 TCGv_i32 max = tcg_const_i32(32);
4503 * Rely on the TCG guarantee that out of range shifts produce
4504 * unspecified results, not undefined behaviour (i.e. no trap).
4505 * Discard out-of-range results after the fact.
4507 tcg_gen_ext8s_i32(lsh, shift);
4508 tcg_gen_neg_i32(rsh, lsh);
4509 tcg_gen_shl_i32(lval, src, lsh);
4510 tcg_gen_shr_i32(rval, src, rsh);
4511 tcg_gen_movcond_i32(TCG_COND_LTU, dst, lsh, max, lval, zero);
4512 tcg_gen_movcond_i32(TCG_COND_LTU, dst, rsh, max, rval, dst);
4514 tcg_temp_free_i32(lval);
4515 tcg_temp_free_i32(rval);
4516 tcg_temp_free_i32(lsh);
4517 tcg_temp_free_i32(rsh);
4518 tcg_temp_free_i32(zero);
4519 tcg_temp_free_i32(max);
4522 void gen_ushl_i64(TCGv_i64 dst, TCGv_i64 src, TCGv_i64 shift)
4524 TCGv_i64 lval = tcg_temp_new_i64();
4525 TCGv_i64 rval = tcg_temp_new_i64();
4526 TCGv_i64 lsh = tcg_temp_new_i64();
4527 TCGv_i64 rsh = tcg_temp_new_i64();
4528 TCGv_i64 zero = tcg_const_i64(0);
4529 TCGv_i64 max = tcg_const_i64(64);
4532 * Rely on the TCG guarantee that out of range shifts produce
4533 * unspecified results, not undefined behaviour (i.e. no trap).
4534 * Discard out-of-range results after the fact.
4536 tcg_gen_ext8s_i64(lsh, shift);
4537 tcg_gen_neg_i64(rsh, lsh);
4538 tcg_gen_shl_i64(lval, src, lsh);
4539 tcg_gen_shr_i64(rval, src, rsh);
4540 tcg_gen_movcond_i64(TCG_COND_LTU, dst, lsh, max, lval, zero);
4541 tcg_gen_movcond_i64(TCG_COND_LTU, dst, rsh, max, rval, dst);
4543 tcg_temp_free_i64(lval);
4544 tcg_temp_free_i64(rval);
4545 tcg_temp_free_i64(lsh);
4546 tcg_temp_free_i64(rsh);
4547 tcg_temp_free_i64(zero);
4548 tcg_temp_free_i64(max);
4551 static void gen_ushl_vec(unsigned vece, TCGv_vec dst,
4552 TCGv_vec src, TCGv_vec shift)
4554 TCGv_vec lval = tcg_temp_new_vec_matching(dst);
4555 TCGv_vec rval = tcg_temp_new_vec_matching(dst);
4556 TCGv_vec lsh = tcg_temp_new_vec_matching(dst);
4557 TCGv_vec rsh = tcg_temp_new_vec_matching(dst);
4560 tcg_gen_neg_vec(vece, rsh, shift);
4562 tcg_gen_mov_vec(lsh, shift);
4564 msk = tcg_temp_new_vec_matching(dst);
4565 tcg_gen_dupi_vec(vece, msk, 0xff);
4566 tcg_gen_and_vec(vece, lsh, shift, msk);
4567 tcg_gen_and_vec(vece, rsh, rsh, msk);
4568 tcg_temp_free_vec(msk);
4572 * Rely on the TCG guarantee that out of range shifts produce
4573 * unspecified results, not undefined behaviour (i.e. no trap).
4574 * Discard out-of-range results after the fact.
4576 tcg_gen_shlv_vec(vece, lval, src, lsh);
4577 tcg_gen_shrv_vec(vece, rval, src, rsh);
4579 max = tcg_temp_new_vec_matching(dst);
4580 tcg_gen_dupi_vec(vece, max, 8 << vece);
4583 * The choice of LT (signed) and GEU (unsigned) are biased toward
4584 * the instructions of the x86_64 host. For MO_8, the whole byte
4585 * is significant so we must use an unsigned compare; otherwise we
4586 * have already masked to a byte and so a signed compare works.
4587 * Other tcg hosts have a full set of comparisons and do not care.
4590 tcg_gen_cmp_vec(TCG_COND_GEU, vece, lsh, lsh, max);
4591 tcg_gen_cmp_vec(TCG_COND_GEU, vece, rsh, rsh, max);
4592 tcg_gen_andc_vec(vece, lval, lval, lsh);
4593 tcg_gen_andc_vec(vece, rval, rval, rsh);
4595 tcg_gen_cmp_vec(TCG_COND_LT, vece, lsh, lsh, max);
4596 tcg_gen_cmp_vec(TCG_COND_LT, vece, rsh, rsh, max);
4597 tcg_gen_and_vec(vece, lval, lval, lsh);
4598 tcg_gen_and_vec(vece, rval, rval, rsh);
4600 tcg_gen_or_vec(vece, dst, lval, rval);
4602 tcg_temp_free_vec(max);
4603 tcg_temp_free_vec(lval);
4604 tcg_temp_free_vec(rval);
4605 tcg_temp_free_vec(lsh);
4606 tcg_temp_free_vec(rsh);
4609 void gen_gvec_ushl(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4610 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4612 static const TCGOpcode vecop_list[] = {
4613 INDEX_op_neg_vec, INDEX_op_shlv_vec,
4614 INDEX_op_shrv_vec, INDEX_op_cmp_vec, 0
4616 static const GVecGen3 ops[4] = {
4617 { .fniv = gen_ushl_vec,
4618 .fno = gen_helper_gvec_ushl_b,
4619 .opt_opc = vecop_list,
4621 { .fniv = gen_ushl_vec,
4622 .fno = gen_helper_gvec_ushl_h,
4623 .opt_opc = vecop_list,
4625 { .fni4 = gen_ushl_i32,
4626 .fniv = gen_ushl_vec,
4627 .opt_opc = vecop_list,
4629 { .fni8 = gen_ushl_i64,
4630 .fniv = gen_ushl_vec,
4631 .opt_opc = vecop_list,
4634 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4637 void gen_sshl_i32(TCGv_i32 dst, TCGv_i32 src, TCGv_i32 shift)
4639 TCGv_i32 lval = tcg_temp_new_i32();
4640 TCGv_i32 rval = tcg_temp_new_i32();
4641 TCGv_i32 lsh = tcg_temp_new_i32();
4642 TCGv_i32 rsh = tcg_temp_new_i32();
4643 TCGv_i32 zero = tcg_const_i32(0);
4644 TCGv_i32 max = tcg_const_i32(31);
4647 * Rely on the TCG guarantee that out of range shifts produce
4648 * unspecified results, not undefined behaviour (i.e. no trap).
4649 * Discard out-of-range results after the fact.
4651 tcg_gen_ext8s_i32(lsh, shift);
4652 tcg_gen_neg_i32(rsh, lsh);
4653 tcg_gen_shl_i32(lval, src, lsh);
4654 tcg_gen_umin_i32(rsh, rsh, max);
4655 tcg_gen_sar_i32(rval, src, rsh);
4656 tcg_gen_movcond_i32(TCG_COND_LEU, lval, lsh, max, lval, zero);
4657 tcg_gen_movcond_i32(TCG_COND_LT, dst, lsh, zero, rval, lval);
4659 tcg_temp_free_i32(lval);
4660 tcg_temp_free_i32(rval);
4661 tcg_temp_free_i32(lsh);
4662 tcg_temp_free_i32(rsh);
4663 tcg_temp_free_i32(zero);
4664 tcg_temp_free_i32(max);
4667 void gen_sshl_i64(TCGv_i64 dst, TCGv_i64 src, TCGv_i64 shift)
4669 TCGv_i64 lval = tcg_temp_new_i64();
4670 TCGv_i64 rval = tcg_temp_new_i64();
4671 TCGv_i64 lsh = tcg_temp_new_i64();
4672 TCGv_i64 rsh = tcg_temp_new_i64();
4673 TCGv_i64 zero = tcg_const_i64(0);
4674 TCGv_i64 max = tcg_const_i64(63);
4677 * Rely on the TCG guarantee that out of range shifts produce
4678 * unspecified results, not undefined behaviour (i.e. no trap).
4679 * Discard out-of-range results after the fact.
4681 tcg_gen_ext8s_i64(lsh, shift);
4682 tcg_gen_neg_i64(rsh, lsh);
4683 tcg_gen_shl_i64(lval, src, lsh);
4684 tcg_gen_umin_i64(rsh, rsh, max);
4685 tcg_gen_sar_i64(rval, src, rsh);
4686 tcg_gen_movcond_i64(TCG_COND_LEU, lval, lsh, max, lval, zero);
4687 tcg_gen_movcond_i64(TCG_COND_LT, dst, lsh, zero, rval, lval);
4689 tcg_temp_free_i64(lval);
4690 tcg_temp_free_i64(rval);
4691 tcg_temp_free_i64(lsh);
4692 tcg_temp_free_i64(rsh);
4693 tcg_temp_free_i64(zero);
4694 tcg_temp_free_i64(max);
4697 static void gen_sshl_vec(unsigned vece, TCGv_vec dst,
4698 TCGv_vec src, TCGv_vec shift)
4700 TCGv_vec lval = tcg_temp_new_vec_matching(dst);
4701 TCGv_vec rval = tcg_temp_new_vec_matching(dst);
4702 TCGv_vec lsh = tcg_temp_new_vec_matching(dst);
4703 TCGv_vec rsh = tcg_temp_new_vec_matching(dst);
4704 TCGv_vec tmp = tcg_temp_new_vec_matching(dst);
4707 * Rely on the TCG guarantee that out of range shifts produce
4708 * unspecified results, not undefined behaviour (i.e. no trap).
4709 * Discard out-of-range results after the fact.
4711 tcg_gen_neg_vec(vece, rsh, shift);
4713 tcg_gen_mov_vec(lsh, shift);
4715 tcg_gen_dupi_vec(vece, tmp, 0xff);
4716 tcg_gen_and_vec(vece, lsh, shift, tmp);
4717 tcg_gen_and_vec(vece, rsh, rsh, tmp);
4720 /* Bound rsh so out of bound right shift gets -1. */
4721 tcg_gen_dupi_vec(vece, tmp, (8 << vece) - 1);
4722 tcg_gen_umin_vec(vece, rsh, rsh, tmp);
4723 tcg_gen_cmp_vec(TCG_COND_GT, vece, tmp, lsh, tmp);
4725 tcg_gen_shlv_vec(vece, lval, src, lsh);
4726 tcg_gen_sarv_vec(vece, rval, src, rsh);
4728 /* Select in-bound left shift. */
4729 tcg_gen_andc_vec(vece, lval, lval, tmp);
4731 /* Select between left and right shift. */
4733 tcg_gen_dupi_vec(vece, tmp, 0);
4734 tcg_gen_cmpsel_vec(TCG_COND_LT, vece, dst, lsh, tmp, rval, lval);
4736 tcg_gen_dupi_vec(vece, tmp, 0x80);
4737 tcg_gen_cmpsel_vec(TCG_COND_LT, vece, dst, lsh, tmp, lval, rval);
4740 tcg_temp_free_vec(lval);
4741 tcg_temp_free_vec(rval);
4742 tcg_temp_free_vec(lsh);
4743 tcg_temp_free_vec(rsh);
4744 tcg_temp_free_vec(tmp);
4747 void gen_gvec_sshl(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4748 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4750 static const TCGOpcode vecop_list[] = {
4751 INDEX_op_neg_vec, INDEX_op_umin_vec, INDEX_op_shlv_vec,
4752 INDEX_op_sarv_vec, INDEX_op_cmp_vec, INDEX_op_cmpsel_vec, 0
4754 static const GVecGen3 ops[4] = {
4755 { .fniv = gen_sshl_vec,
4756 .fno = gen_helper_gvec_sshl_b,
4757 .opt_opc = vecop_list,
4759 { .fniv = gen_sshl_vec,
4760 .fno = gen_helper_gvec_sshl_h,
4761 .opt_opc = vecop_list,
4763 { .fni4 = gen_sshl_i32,
4764 .fniv = gen_sshl_vec,
4765 .opt_opc = vecop_list,
4767 { .fni8 = gen_sshl_i64,
4768 .fniv = gen_sshl_vec,
4769 .opt_opc = vecop_list,
4772 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4775 static void gen_uqadd_vec(unsigned vece, TCGv_vec t, TCGv_vec sat,
4776 TCGv_vec a, TCGv_vec b)
4778 TCGv_vec x = tcg_temp_new_vec_matching(t);
4779 tcg_gen_add_vec(vece, x, a, b);
4780 tcg_gen_usadd_vec(vece, t, a, b);
4781 tcg_gen_cmp_vec(TCG_COND_NE, vece, x, x, t);
4782 tcg_gen_or_vec(vece, sat, sat, x);
4783 tcg_temp_free_vec(x);
4786 void gen_gvec_uqadd_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4787 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4789 static const TCGOpcode vecop_list[] = {
4790 INDEX_op_usadd_vec, INDEX_op_cmp_vec, INDEX_op_add_vec, 0
4792 static const GVecGen4 ops[4] = {
4793 { .fniv = gen_uqadd_vec,
4794 .fno = gen_helper_gvec_uqadd_b,
4796 .opt_opc = vecop_list,
4798 { .fniv = gen_uqadd_vec,
4799 .fno = gen_helper_gvec_uqadd_h,
4801 .opt_opc = vecop_list,
4803 { .fniv = gen_uqadd_vec,
4804 .fno = gen_helper_gvec_uqadd_s,
4806 .opt_opc = vecop_list,
4808 { .fniv = gen_uqadd_vec,
4809 .fno = gen_helper_gvec_uqadd_d,
4811 .opt_opc = vecop_list,
4814 tcg_gen_gvec_4(rd_ofs, offsetof(CPUARMState, vfp.qc),
4815 rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4818 static void gen_sqadd_vec(unsigned vece, TCGv_vec t, TCGv_vec sat,
4819 TCGv_vec a, TCGv_vec b)
4821 TCGv_vec x = tcg_temp_new_vec_matching(t);
4822 tcg_gen_add_vec(vece, x, a, b);
4823 tcg_gen_ssadd_vec(vece, t, a, b);
4824 tcg_gen_cmp_vec(TCG_COND_NE, vece, x, x, t);
4825 tcg_gen_or_vec(vece, sat, sat, x);
4826 tcg_temp_free_vec(x);
4829 void gen_gvec_sqadd_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4830 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4832 static const TCGOpcode vecop_list[] = {
4833 INDEX_op_ssadd_vec, INDEX_op_cmp_vec, INDEX_op_add_vec, 0
4835 static const GVecGen4 ops[4] = {
4836 { .fniv = gen_sqadd_vec,
4837 .fno = gen_helper_gvec_sqadd_b,
4838 .opt_opc = vecop_list,
4841 { .fniv = gen_sqadd_vec,
4842 .fno = gen_helper_gvec_sqadd_h,
4843 .opt_opc = vecop_list,
4846 { .fniv = gen_sqadd_vec,
4847 .fno = gen_helper_gvec_sqadd_s,
4848 .opt_opc = vecop_list,
4851 { .fniv = gen_sqadd_vec,
4852 .fno = gen_helper_gvec_sqadd_d,
4853 .opt_opc = vecop_list,
4857 tcg_gen_gvec_4(rd_ofs, offsetof(CPUARMState, vfp.qc),
4858 rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4861 static void gen_uqsub_vec(unsigned vece, TCGv_vec t, TCGv_vec sat,
4862 TCGv_vec a, TCGv_vec b)
4864 TCGv_vec x = tcg_temp_new_vec_matching(t);
4865 tcg_gen_sub_vec(vece, x, a, b);
4866 tcg_gen_ussub_vec(vece, t, a, b);
4867 tcg_gen_cmp_vec(TCG_COND_NE, vece, x, x, t);
4868 tcg_gen_or_vec(vece, sat, sat, x);
4869 tcg_temp_free_vec(x);
4872 void gen_gvec_uqsub_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4873 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4875 static const TCGOpcode vecop_list[] = {
4876 INDEX_op_ussub_vec, INDEX_op_cmp_vec, INDEX_op_sub_vec, 0
4878 static const GVecGen4 ops[4] = {
4879 { .fniv = gen_uqsub_vec,
4880 .fno = gen_helper_gvec_uqsub_b,
4881 .opt_opc = vecop_list,
4884 { .fniv = gen_uqsub_vec,
4885 .fno = gen_helper_gvec_uqsub_h,
4886 .opt_opc = vecop_list,
4889 { .fniv = gen_uqsub_vec,
4890 .fno = gen_helper_gvec_uqsub_s,
4891 .opt_opc = vecop_list,
4894 { .fniv = gen_uqsub_vec,
4895 .fno = gen_helper_gvec_uqsub_d,
4896 .opt_opc = vecop_list,
4900 tcg_gen_gvec_4(rd_ofs, offsetof(CPUARMState, vfp.qc),
4901 rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4904 static void gen_sqsub_vec(unsigned vece, TCGv_vec t, TCGv_vec sat,
4905 TCGv_vec a, TCGv_vec b)
4907 TCGv_vec x = tcg_temp_new_vec_matching(t);
4908 tcg_gen_sub_vec(vece, x, a, b);
4909 tcg_gen_sssub_vec(vece, t, a, b);
4910 tcg_gen_cmp_vec(TCG_COND_NE, vece, x, x, t);
4911 tcg_gen_or_vec(vece, sat, sat, x);
4912 tcg_temp_free_vec(x);
4915 void gen_gvec_sqsub_qc(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4916 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4918 static const TCGOpcode vecop_list[] = {
4919 INDEX_op_sssub_vec, INDEX_op_cmp_vec, INDEX_op_sub_vec, 0
4921 static const GVecGen4 ops[4] = {
4922 { .fniv = gen_sqsub_vec,
4923 .fno = gen_helper_gvec_sqsub_b,
4924 .opt_opc = vecop_list,
4927 { .fniv = gen_sqsub_vec,
4928 .fno = gen_helper_gvec_sqsub_h,
4929 .opt_opc = vecop_list,
4932 { .fniv = gen_sqsub_vec,
4933 .fno = gen_helper_gvec_sqsub_s,
4934 .opt_opc = vecop_list,
4937 { .fniv = gen_sqsub_vec,
4938 .fno = gen_helper_gvec_sqsub_d,
4939 .opt_opc = vecop_list,
4943 tcg_gen_gvec_4(rd_ofs, offsetof(CPUARMState, vfp.qc),
4944 rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
4947 static void gen_sabd_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
4949 TCGv_i32 t = tcg_temp_new_i32();
4951 tcg_gen_sub_i32(t, a, b);
4952 tcg_gen_sub_i32(d, b, a);
4953 tcg_gen_movcond_i32(TCG_COND_LT, d, a, b, d, t);
4954 tcg_temp_free_i32(t);
4957 static void gen_sabd_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
4959 TCGv_i64 t = tcg_temp_new_i64();
4961 tcg_gen_sub_i64(t, a, b);
4962 tcg_gen_sub_i64(d, b, a);
4963 tcg_gen_movcond_i64(TCG_COND_LT, d, a, b, d, t);
4964 tcg_temp_free_i64(t);
4967 static void gen_sabd_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
4969 TCGv_vec t = tcg_temp_new_vec_matching(d);
4971 tcg_gen_smin_vec(vece, t, a, b);
4972 tcg_gen_smax_vec(vece, d, a, b);
4973 tcg_gen_sub_vec(vece, d, d, t);
4974 tcg_temp_free_vec(t);
4977 void gen_gvec_sabd(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
4978 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
4980 static const TCGOpcode vecop_list[] = {
4981 INDEX_op_sub_vec, INDEX_op_smin_vec, INDEX_op_smax_vec, 0
4983 static const GVecGen3 ops[4] = {
4984 { .fniv = gen_sabd_vec,
4985 .fno = gen_helper_gvec_sabd_b,
4986 .opt_opc = vecop_list,
4988 { .fniv = gen_sabd_vec,
4989 .fno = gen_helper_gvec_sabd_h,
4990 .opt_opc = vecop_list,
4992 { .fni4 = gen_sabd_i32,
4993 .fniv = gen_sabd_vec,
4994 .fno = gen_helper_gvec_sabd_s,
4995 .opt_opc = vecop_list,
4997 { .fni8 = gen_sabd_i64,
4998 .fniv = gen_sabd_vec,
4999 .fno = gen_helper_gvec_sabd_d,
5000 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
5001 .opt_opc = vecop_list,
5004 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
5007 static void gen_uabd_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
5009 TCGv_i32 t = tcg_temp_new_i32();
5011 tcg_gen_sub_i32(t, a, b);
5012 tcg_gen_sub_i32(d, b, a);
5013 tcg_gen_movcond_i32(TCG_COND_LTU, d, a, b, d, t);
5014 tcg_temp_free_i32(t);
5017 static void gen_uabd_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
5019 TCGv_i64 t = tcg_temp_new_i64();
5021 tcg_gen_sub_i64(t, a, b);
5022 tcg_gen_sub_i64(d, b, a);
5023 tcg_gen_movcond_i64(TCG_COND_LTU, d, a, b, d, t);
5024 tcg_temp_free_i64(t);
5027 static void gen_uabd_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
5029 TCGv_vec t = tcg_temp_new_vec_matching(d);
5031 tcg_gen_umin_vec(vece, t, a, b);
5032 tcg_gen_umax_vec(vece, d, a, b);
5033 tcg_gen_sub_vec(vece, d, d, t);
5034 tcg_temp_free_vec(t);
5037 void gen_gvec_uabd(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
5038 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
5040 static const TCGOpcode vecop_list[] = {
5041 INDEX_op_sub_vec, INDEX_op_umin_vec, INDEX_op_umax_vec, 0
5043 static const GVecGen3 ops[4] = {
5044 { .fniv = gen_uabd_vec,
5045 .fno = gen_helper_gvec_uabd_b,
5046 .opt_opc = vecop_list,
5048 { .fniv = gen_uabd_vec,
5049 .fno = gen_helper_gvec_uabd_h,
5050 .opt_opc = vecop_list,
5052 { .fni4 = gen_uabd_i32,
5053 .fniv = gen_uabd_vec,
5054 .fno = gen_helper_gvec_uabd_s,
5055 .opt_opc = vecop_list,
5057 { .fni8 = gen_uabd_i64,
5058 .fniv = gen_uabd_vec,
5059 .fno = gen_helper_gvec_uabd_d,
5060 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
5061 .opt_opc = vecop_list,
5064 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
5067 static void gen_saba_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
5069 TCGv_i32 t = tcg_temp_new_i32();
5070 gen_sabd_i32(t, a, b);
5071 tcg_gen_add_i32(d, d, t);
5072 tcg_temp_free_i32(t);
5075 static void gen_saba_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
5077 TCGv_i64 t = tcg_temp_new_i64();
5078 gen_sabd_i64(t, a, b);
5079 tcg_gen_add_i64(d, d, t);
5080 tcg_temp_free_i64(t);
5083 static void gen_saba_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
5085 TCGv_vec t = tcg_temp_new_vec_matching(d);
5086 gen_sabd_vec(vece, t, a, b);
5087 tcg_gen_add_vec(vece, d, d, t);
5088 tcg_temp_free_vec(t);
5091 void gen_gvec_saba(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
5092 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
5094 static const TCGOpcode vecop_list[] = {
5095 INDEX_op_sub_vec, INDEX_op_add_vec,
5096 INDEX_op_smin_vec, INDEX_op_smax_vec, 0
5098 static const GVecGen3 ops[4] = {
5099 { .fniv = gen_saba_vec,
5100 .fno = gen_helper_gvec_saba_b,
5101 .opt_opc = vecop_list,
5104 { .fniv = gen_saba_vec,
5105 .fno = gen_helper_gvec_saba_h,
5106 .opt_opc = vecop_list,
5109 { .fni4 = gen_saba_i32,
5110 .fniv = gen_saba_vec,
5111 .fno = gen_helper_gvec_saba_s,
5112 .opt_opc = vecop_list,
5115 { .fni8 = gen_saba_i64,
5116 .fniv = gen_saba_vec,
5117 .fno = gen_helper_gvec_saba_d,
5118 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
5119 .opt_opc = vecop_list,
5123 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
5126 static void gen_uaba_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
5128 TCGv_i32 t = tcg_temp_new_i32();
5129 gen_uabd_i32(t, a, b);
5130 tcg_gen_add_i32(d, d, t);
5131 tcg_temp_free_i32(t);
5134 static void gen_uaba_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
5136 TCGv_i64 t = tcg_temp_new_i64();
5137 gen_uabd_i64(t, a, b);
5138 tcg_gen_add_i64(d, d, t);
5139 tcg_temp_free_i64(t);
5142 static void gen_uaba_vec(unsigned vece, TCGv_vec d, TCGv_vec a, TCGv_vec b)
5144 TCGv_vec t = tcg_temp_new_vec_matching(d);
5145 gen_uabd_vec(vece, t, a, b);
5146 tcg_gen_add_vec(vece, d, d, t);
5147 tcg_temp_free_vec(t);
5150 void gen_gvec_uaba(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
5151 uint32_t rm_ofs, uint32_t opr_sz, uint32_t max_sz)
5153 static const TCGOpcode vecop_list[] = {
5154 INDEX_op_sub_vec, INDEX_op_add_vec,
5155 INDEX_op_umin_vec, INDEX_op_umax_vec, 0
5157 static const GVecGen3 ops[4] = {
5158 { .fniv = gen_uaba_vec,
5159 .fno = gen_helper_gvec_uaba_b,
5160 .opt_opc = vecop_list,
5163 { .fniv = gen_uaba_vec,
5164 .fno = gen_helper_gvec_uaba_h,
5165 .opt_opc = vecop_list,
5168 { .fni4 = gen_uaba_i32,
5169 .fniv = gen_uaba_vec,
5170 .fno = gen_helper_gvec_uaba_s,
5171 .opt_opc = vecop_list,
5174 { .fni8 = gen_uaba_i64,
5175 .fniv = gen_uaba_vec,
5176 .fno = gen_helper_gvec_uaba_d,
5177 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
5178 .opt_opc = vecop_list,
5182 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, &ops[vece]);
5185 /* Translate a NEON data processing instruction. Return nonzero if the
5186 instruction is invalid.
5187 We process data in a mixture of 32-bit and 64-bit chunks.
5188 Mostly we use 32-bit chunks so we can use normal scalar instructions. */
5190 static int disas_neon_data_insn(DisasContext *s, uint32_t insn)
5194 int rd, rn, rm, rd_ofs, rn_ofs, rm_ofs;
5200 TCGv_i32 tmp, tmp2, tmp3, tmp4, tmp5;
5204 if (!arm_dc_feature(s, ARM_FEATURE_NEON)) {
5208 /* FIXME: this access check should not take precedence over UNDEF
5209 * for invalid encodings; we will generate incorrect syndrome information
5210 * for attempts to execute invalid vfp/neon encodings with FP disabled.
5212 if (s->fp_excp_el) {
5213 gen_exception_insn(s, s->pc_curr, EXCP_UDEF,
5214 syn_simd_access_trap(1, 0xe, false), s->fp_excp_el);
5218 if (!s->vfp_enabled)
5220 q = (insn & (1 << 6)) != 0;
5221 u = (insn >> 24) & 1;
5222 VFP_DREG_D(rd, insn);
5223 VFP_DREG_N(rn, insn);
5224 VFP_DREG_M(rm, insn);
5225 size = (insn >> 20) & 3;
5226 vec_size = q ? 16 : 8;
5227 rd_ofs = neon_reg_offset(rd, 0);
5228 rn_ofs = neon_reg_offset(rn, 0);
5229 rm_ofs = neon_reg_offset(rm, 0);
5231 if ((insn & (1 << 23)) == 0) {
5232 /* Three register same length: handled by decodetree */
5234 } else if (insn & (1 << 4)) {
5235 /* Two registers and shift or reg and imm: handled by decodetree */
5237 } else { /* (insn & 0x00800010 == 0x00800000) */
5239 op = (insn >> 8) & 0xf;
5240 if ((insn & (1 << 6)) == 0) {
5241 /* Three registers of different lengths. */
5245 /* undefreq: bit 0 : UNDEF if size == 0
5246 * bit 1 : UNDEF if size == 1
5247 * bit 2 : UNDEF if size == 2
5248 * bit 3 : UNDEF if U == 1
5249 * Note that [2:0] set implies 'always UNDEF'
5252 /* prewiden, src1_wide, src2_wide, undefreq */
5253 static const int neon_3reg_wide[16][4] = {
5254 {1, 0, 0, 0}, /* VADDL */
5255 {1, 1, 0, 0}, /* VADDW */
5256 {1, 0, 0, 0}, /* VSUBL */
5257 {1, 1, 0, 0}, /* VSUBW */
5258 {0, 1, 1, 0}, /* VADDHN */
5259 {0, 0, 0, 0}, /* VABAL */
5260 {0, 1, 1, 0}, /* VSUBHN */
5261 {0, 0, 0, 0}, /* VABDL */
5262 {0, 0, 0, 0}, /* VMLAL */
5263 {0, 0, 0, 9}, /* VQDMLAL */
5264 {0, 0, 0, 0}, /* VMLSL */
5265 {0, 0, 0, 9}, /* VQDMLSL */
5266 {0, 0, 0, 0}, /* Integer VMULL */
5267 {0, 0, 0, 9}, /* VQDMULL */
5268 {0, 0, 0, 0xa}, /* Polynomial VMULL */
5269 {0, 0, 0, 7}, /* Reserved: always UNDEF */
5272 prewiden = neon_3reg_wide[op][0];
5273 src1_wide = neon_3reg_wide[op][1];
5274 src2_wide = neon_3reg_wide[op][2];
5275 undefreq = neon_3reg_wide[op][3];
5277 if ((undefreq & (1 << size)) ||
5278 ((undefreq & 8) && u)) {
5281 if ((src1_wide && (rn & 1)) ||
5282 (src2_wide && (rm & 1)) ||
5283 (!src2_wide && (rd & 1))) {
5287 /* Handle polynomial VMULL in a single pass. */
5291 tcg_gen_gvec_3_ool(rd_ofs, rn_ofs, rm_ofs, 16, 16,
5292 0, gen_helper_neon_pmull_h);
5295 if (!dc_isar_feature(aa32_pmull, s)) {
5298 tcg_gen_gvec_3_ool(rd_ofs, rn_ofs, rm_ofs, 16, 16,
5299 0, gen_helper_gvec_pmull_q);
5304 /* Avoid overlapping operands. Wide source operands are
5305 always aligned so will never overlap with wide
5306 destinations in problematic ways. */
5307 if (rd == rm && !src2_wide) {
5308 tmp = neon_load_reg(rm, 1);
5309 neon_store_scratch(2, tmp);
5310 } else if (rd == rn && !src1_wide) {
5311 tmp = neon_load_reg(rn, 1);
5312 neon_store_scratch(2, tmp);
5315 for (pass = 0; pass < 2; pass++) {
5317 neon_load_reg64(cpu_V0, rn + pass);
5320 if (pass == 1 && rd == rn) {
5321 tmp = neon_load_scratch(2);
5323 tmp = neon_load_reg(rn, pass);
5326 gen_neon_widen(cpu_V0, tmp, size, u);
5330 neon_load_reg64(cpu_V1, rm + pass);
5333 if (pass == 1 && rd == rm) {
5334 tmp2 = neon_load_scratch(2);
5336 tmp2 = neon_load_reg(rm, pass);
5339 gen_neon_widen(cpu_V1, tmp2, size, u);
5343 case 0: case 1: case 4: /* VADDL, VADDW, VADDHN, VRADDHN */
5344 gen_neon_addl(size);
5346 case 2: case 3: case 6: /* VSUBL, VSUBW, VSUBHN, VRSUBHN */
5347 gen_neon_subl(size);
5349 case 5: case 7: /* VABAL, VABDL */
5350 switch ((size << 1) | u) {
5352 gen_helper_neon_abdl_s16(cpu_V0, tmp, tmp2);
5355 gen_helper_neon_abdl_u16(cpu_V0, tmp, tmp2);
5358 gen_helper_neon_abdl_s32(cpu_V0, tmp, tmp2);
5361 gen_helper_neon_abdl_u32(cpu_V0, tmp, tmp2);
5364 gen_helper_neon_abdl_s64(cpu_V0, tmp, tmp2);
5367 gen_helper_neon_abdl_u64(cpu_V0, tmp, tmp2);
5371 tcg_temp_free_i32(tmp2);
5372 tcg_temp_free_i32(tmp);
5374 case 8: case 9: case 10: case 11: case 12: case 13:
5375 /* VMLAL, VQDMLAL, VMLSL, VQDMLSL, VMULL, VQDMULL */
5376 gen_neon_mull(cpu_V0, tmp, tmp2, size, u);
5378 default: /* 15 is RESERVED: caught earlier */
5383 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
5384 neon_store_reg64(cpu_V0, rd + pass);
5385 } else if (op == 5 || (op >= 8 && op <= 11)) {
5387 neon_load_reg64(cpu_V1, rd + pass);
5389 case 10: /* VMLSL */
5390 gen_neon_negl(cpu_V0, size);
5392 case 5: case 8: /* VABAL, VMLAL */
5393 gen_neon_addl(size);
5395 case 9: case 11: /* VQDMLAL, VQDMLSL */
5396 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
5398 gen_neon_negl(cpu_V0, size);
5400 gen_neon_addl_saturate(cpu_V0, cpu_V1, size);
5405 neon_store_reg64(cpu_V0, rd + pass);
5406 } else if (op == 4 || op == 6) {
5407 /* Narrowing operation. */
5408 tmp = tcg_temp_new_i32();
5412 gen_helper_neon_narrow_high_u8(tmp, cpu_V0);
5415 gen_helper_neon_narrow_high_u16(tmp, cpu_V0);
5418 tcg_gen_extrh_i64_i32(tmp, cpu_V0);
5425 gen_helper_neon_narrow_round_high_u8(tmp, cpu_V0);
5428 gen_helper_neon_narrow_round_high_u16(tmp, cpu_V0);
5431 tcg_gen_addi_i64(cpu_V0, cpu_V0, 1u << 31);
5432 tcg_gen_extrh_i64_i32(tmp, cpu_V0);
5440 neon_store_reg(rd, 0, tmp3);
5441 neon_store_reg(rd, 1, tmp);
5444 /* Write back the result. */
5445 neon_store_reg64(cpu_V0, rd + pass);
5449 /* Two registers and a scalar. NB that for ops of this form
5450 * the ARM ARM labels bit 24 as Q, but it is in our variable
5457 case 1: /* Float VMLA scalar */
5458 case 5: /* Floating point VMLS scalar */
5459 case 9: /* Floating point VMUL scalar */
5464 case 0: /* Integer VMLA scalar */
5465 case 4: /* Integer VMLS scalar */
5466 case 8: /* Integer VMUL scalar */
5467 case 12: /* VQDMULH scalar */
5468 case 13: /* VQRDMULH scalar */
5469 if (u && ((rd | rn) & 1)) {
5472 tmp = neon_get_scalar(size, rm);
5473 neon_store_scratch(0, tmp);
5474 for (pass = 0; pass < (u ? 4 : 2); pass++) {
5475 tmp = neon_load_scratch(0);
5476 tmp2 = neon_load_reg(rn, pass);
5479 gen_helper_neon_qdmulh_s16(tmp, cpu_env, tmp, tmp2);
5481 gen_helper_neon_qdmulh_s32(tmp, cpu_env, tmp, tmp2);
5483 } else if (op == 13) {
5485 gen_helper_neon_qrdmulh_s16(tmp, cpu_env, tmp, tmp2);
5487 gen_helper_neon_qrdmulh_s32(tmp, cpu_env, tmp, tmp2);
5489 } else if (op & 1) {
5490 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5491 gen_helper_vfp_muls(tmp, tmp, tmp2, fpstatus);
5492 tcg_temp_free_ptr(fpstatus);
5495 case 0: gen_helper_neon_mul_u8(tmp, tmp, tmp2); break;
5496 case 1: gen_helper_neon_mul_u16(tmp, tmp, tmp2); break;
5497 case 2: tcg_gen_mul_i32(tmp, tmp, tmp2); break;
5501 tcg_temp_free_i32(tmp2);
5504 tmp2 = neon_load_reg(rd, pass);
5507 gen_neon_add(size, tmp, tmp2);
5511 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5512 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
5513 tcg_temp_free_ptr(fpstatus);
5517 gen_neon_rsb(size, tmp, tmp2);
5521 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5522 gen_helper_vfp_subs(tmp, tmp2, tmp, fpstatus);
5523 tcg_temp_free_ptr(fpstatus);
5529 tcg_temp_free_i32(tmp2);
5531 neon_store_reg(rd, pass, tmp);
5534 case 3: /* VQDMLAL scalar */
5535 case 7: /* VQDMLSL scalar */
5536 case 11: /* VQDMULL scalar */
5541 case 2: /* VMLAL sclar */
5542 case 6: /* VMLSL scalar */
5543 case 10: /* VMULL scalar */
5547 tmp2 = neon_get_scalar(size, rm);
5548 /* We need a copy of tmp2 because gen_neon_mull
5549 * deletes it during pass 0. */
5550 tmp4 = tcg_temp_new_i32();
5551 tcg_gen_mov_i32(tmp4, tmp2);
5552 tmp3 = neon_load_reg(rn, 1);
5554 for (pass = 0; pass < 2; pass++) {
5556 tmp = neon_load_reg(rn, 0);
5561 gen_neon_mull(cpu_V0, tmp, tmp2, size, u);
5563 neon_load_reg64(cpu_V1, rd + pass);
5567 gen_neon_negl(cpu_V0, size);
5570 gen_neon_addl(size);
5573 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
5575 gen_neon_negl(cpu_V0, size);
5577 gen_neon_addl_saturate(cpu_V0, cpu_V1, size);
5583 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
5588 neon_store_reg64(cpu_V0, rd + pass);
5591 case 14: /* VQRDMLAH scalar */
5592 case 15: /* VQRDMLSH scalar */
5594 NeonGenThreeOpEnvFn *fn;
5596 if (!dc_isar_feature(aa32_rdm, s)) {
5599 if (u && ((rd | rn) & 1)) {
5604 fn = gen_helper_neon_qrdmlah_s16;
5606 fn = gen_helper_neon_qrdmlah_s32;
5610 fn = gen_helper_neon_qrdmlsh_s16;
5612 fn = gen_helper_neon_qrdmlsh_s32;
5616 tmp2 = neon_get_scalar(size, rm);
5617 for (pass = 0; pass < (u ? 4 : 2); pass++) {
5618 tmp = neon_load_reg(rn, pass);
5619 tmp3 = neon_load_reg(rd, pass);
5620 fn(tmp, cpu_env, tmp, tmp2, tmp3);
5621 tcg_temp_free_i32(tmp3);
5622 neon_store_reg(rd, pass, tmp);
5624 tcg_temp_free_i32(tmp2);
5628 g_assert_not_reached();
5631 } else { /* size == 3 */
5634 imm = (insn >> 8) & 0xf;
5639 if (q && ((rd | rn | rm) & 1)) {
5644 neon_load_reg64(cpu_V0, rn);
5646 neon_load_reg64(cpu_V1, rn + 1);
5648 } else if (imm == 8) {
5649 neon_load_reg64(cpu_V0, rn + 1);
5651 neon_load_reg64(cpu_V1, rm);
5654 tmp64 = tcg_temp_new_i64();
5656 neon_load_reg64(cpu_V0, rn);
5657 neon_load_reg64(tmp64, rn + 1);
5659 neon_load_reg64(cpu_V0, rn + 1);
5660 neon_load_reg64(tmp64, rm);
5662 tcg_gen_shri_i64(cpu_V0, cpu_V0, (imm & 7) * 8);
5663 tcg_gen_shli_i64(cpu_V1, tmp64, 64 - ((imm & 7) * 8));
5664 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
5666 neon_load_reg64(cpu_V1, rm);
5668 neon_load_reg64(cpu_V1, rm + 1);
5671 tcg_gen_shli_i64(cpu_V1, cpu_V1, 64 - (imm * 8));
5672 tcg_gen_shri_i64(tmp64, tmp64, imm * 8);
5673 tcg_gen_or_i64(cpu_V1, cpu_V1, tmp64);
5674 tcg_temp_free_i64(tmp64);
5677 neon_load_reg64(cpu_V0, rn);
5678 tcg_gen_shri_i64(cpu_V0, cpu_V0, imm * 8);
5679 neon_load_reg64(cpu_V1, rm);
5680 tcg_gen_shli_i64(cpu_V1, cpu_V1, 64 - (imm * 8));
5681 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
5683 neon_store_reg64(cpu_V0, rd);
5685 neon_store_reg64(cpu_V1, rd + 1);
5687 } else if ((insn & (1 << 11)) == 0) {
5688 /* Two register misc. */
5689 op = ((insn >> 12) & 0x30) | ((insn >> 7) & 0xf);
5690 size = (insn >> 18) & 3;
5691 /* UNDEF for unknown op values and bad op-size combinations */
5692 if ((neon_2rm_sizes[op] & (1 << size)) == 0) {
5695 if (neon_2rm_is_v8_op(op) &&
5696 !arm_dc_feature(s, ARM_FEATURE_V8)) {
5699 if ((op != NEON_2RM_VMOVN && op != NEON_2RM_VQMOVN) &&
5700 q && ((rm | rd) & 1)) {
5704 case NEON_2RM_VREV64:
5705 for (pass = 0; pass < (q ? 2 : 1); pass++) {
5706 tmp = neon_load_reg(rm, pass * 2);
5707 tmp2 = neon_load_reg(rm, pass * 2 + 1);
5709 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
5710 case 1: gen_swap_half(tmp); break;
5711 case 2: /* no-op */ break;
5714 neon_store_reg(rd, pass * 2 + 1, tmp);
5716 neon_store_reg(rd, pass * 2, tmp2);
5719 case 0: tcg_gen_bswap32_i32(tmp2, tmp2); break;
5720 case 1: gen_swap_half(tmp2); break;
5723 neon_store_reg(rd, pass * 2, tmp2);
5727 case NEON_2RM_VPADDL: case NEON_2RM_VPADDL_U:
5728 case NEON_2RM_VPADAL: case NEON_2RM_VPADAL_U:
5729 for (pass = 0; pass < q + 1; pass++) {
5730 tmp = neon_load_reg(rm, pass * 2);
5731 gen_neon_widen(cpu_V0, tmp, size, op & 1);
5732 tmp = neon_load_reg(rm, pass * 2 + 1);
5733 gen_neon_widen(cpu_V1, tmp, size, op & 1);
5735 case 0: gen_helper_neon_paddl_u16(CPU_V001); break;
5736 case 1: gen_helper_neon_paddl_u32(CPU_V001); break;
5737 case 2: tcg_gen_add_i64(CPU_V001); break;
5740 if (op >= NEON_2RM_VPADAL) {
5742 neon_load_reg64(cpu_V1, rd + pass);
5743 gen_neon_addl(size);
5745 neon_store_reg64(cpu_V0, rd + pass);
5751 for (n = 0; n < (q ? 4 : 2); n += 2) {
5752 tmp = neon_load_reg(rm, n);
5753 tmp2 = neon_load_reg(rd, n + 1);
5754 neon_store_reg(rm, n, tmp2);
5755 neon_store_reg(rd, n + 1, tmp);
5762 if (gen_neon_unzip(rd, rm, size, q)) {
5767 if (gen_neon_zip(rd, rm, size, q)) {
5771 case NEON_2RM_VMOVN: case NEON_2RM_VQMOVN:
5772 /* also VQMOVUN; op field and mnemonics don't line up */
5777 for (pass = 0; pass < 2; pass++) {
5778 neon_load_reg64(cpu_V0, rm + pass);
5779 tmp = tcg_temp_new_i32();
5780 gen_neon_narrow_op(op == NEON_2RM_VMOVN, q, size,
5785 neon_store_reg(rd, 0, tmp2);
5786 neon_store_reg(rd, 1, tmp);
5790 case NEON_2RM_VSHLL:
5791 if (q || (rd & 1)) {
5794 tmp = neon_load_reg(rm, 0);
5795 tmp2 = neon_load_reg(rm, 1);
5796 for (pass = 0; pass < 2; pass++) {
5799 gen_neon_widen(cpu_V0, tmp, size, 1);
5800 tcg_gen_shli_i64(cpu_V0, cpu_V0, 8 << size);
5801 neon_store_reg64(cpu_V0, rd + pass);
5804 case NEON_2RM_VCVT_F16_F32:
5809 if (!dc_isar_feature(aa32_fp16_spconv, s) ||
5813 fpst = get_fpstatus_ptr(true);
5814 ahp = get_ahp_flag();
5815 tmp = neon_load_reg(rm, 0);
5816 gen_helper_vfp_fcvt_f32_to_f16(tmp, tmp, fpst, ahp);
5817 tmp2 = neon_load_reg(rm, 1);
5818 gen_helper_vfp_fcvt_f32_to_f16(tmp2, tmp2, fpst, ahp);
5819 tcg_gen_shli_i32(tmp2, tmp2, 16);
5820 tcg_gen_or_i32(tmp2, tmp2, tmp);
5821 tcg_temp_free_i32(tmp);
5822 tmp = neon_load_reg(rm, 2);
5823 gen_helper_vfp_fcvt_f32_to_f16(tmp, tmp, fpst, ahp);
5824 tmp3 = neon_load_reg(rm, 3);
5825 neon_store_reg(rd, 0, tmp2);
5826 gen_helper_vfp_fcvt_f32_to_f16(tmp3, tmp3, fpst, ahp);
5827 tcg_gen_shli_i32(tmp3, tmp3, 16);
5828 tcg_gen_or_i32(tmp3, tmp3, tmp);
5829 neon_store_reg(rd, 1, tmp3);
5830 tcg_temp_free_i32(tmp);
5831 tcg_temp_free_i32(ahp);
5832 tcg_temp_free_ptr(fpst);
5835 case NEON_2RM_VCVT_F32_F16:
5839 if (!dc_isar_feature(aa32_fp16_spconv, s) ||
5843 fpst = get_fpstatus_ptr(true);
5844 ahp = get_ahp_flag();
5845 tmp3 = tcg_temp_new_i32();
5846 tmp = neon_load_reg(rm, 0);
5847 tmp2 = neon_load_reg(rm, 1);
5848 tcg_gen_ext16u_i32(tmp3, tmp);
5849 gen_helper_vfp_fcvt_f16_to_f32(tmp3, tmp3, fpst, ahp);
5850 neon_store_reg(rd, 0, tmp3);
5851 tcg_gen_shri_i32(tmp, tmp, 16);
5852 gen_helper_vfp_fcvt_f16_to_f32(tmp, tmp, fpst, ahp);
5853 neon_store_reg(rd, 1, tmp);
5854 tmp3 = tcg_temp_new_i32();
5855 tcg_gen_ext16u_i32(tmp3, tmp2);
5856 gen_helper_vfp_fcvt_f16_to_f32(tmp3, tmp3, fpst, ahp);
5857 neon_store_reg(rd, 2, tmp3);
5858 tcg_gen_shri_i32(tmp2, tmp2, 16);
5859 gen_helper_vfp_fcvt_f16_to_f32(tmp2, tmp2, fpst, ahp);
5860 neon_store_reg(rd, 3, tmp2);
5861 tcg_temp_free_i32(ahp);
5862 tcg_temp_free_ptr(fpst);
5865 case NEON_2RM_AESE: case NEON_2RM_AESMC:
5866 if (!dc_isar_feature(aa32_aes, s) || ((rm | rd) & 1)) {
5870 * Bit 6 is the lowest opcode bit; it distinguishes
5871 * between encryption (AESE/AESMC) and decryption
5874 if (op == NEON_2RM_AESE) {
5875 tcg_gen_gvec_3_ool(vfp_reg_offset(true, rd),
5876 vfp_reg_offset(true, rd),
5877 vfp_reg_offset(true, rm),
5878 16, 16, extract32(insn, 6, 1),
5879 gen_helper_crypto_aese);
5881 tcg_gen_gvec_2_ool(vfp_reg_offset(true, rd),
5882 vfp_reg_offset(true, rm),
5883 16, 16, extract32(insn, 6, 1),
5884 gen_helper_crypto_aesmc);
5887 case NEON_2RM_SHA1H:
5888 if (!dc_isar_feature(aa32_sha1, s) || ((rm | rd) & 1)) {
5891 tcg_gen_gvec_2_ool(rd_ofs, rm_ofs, 16, 16, 0,
5892 gen_helper_crypto_sha1h);
5894 case NEON_2RM_SHA1SU1:
5895 if ((rm | rd) & 1) {
5898 /* bit 6 (q): set -> SHA256SU0, cleared -> SHA1SU1 */
5900 if (!dc_isar_feature(aa32_sha2, s)) {
5903 } else if (!dc_isar_feature(aa32_sha1, s)) {
5906 tcg_gen_gvec_2_ool(rd_ofs, rm_ofs, 16, 16, 0,
5907 q ? gen_helper_crypto_sha256su0
5908 : gen_helper_crypto_sha1su1);
5911 tcg_gen_gvec_not(0, rd_ofs, rm_ofs, vec_size, vec_size);
5914 tcg_gen_gvec_neg(size, rd_ofs, rm_ofs, vec_size, vec_size);
5917 tcg_gen_gvec_abs(size, rd_ofs, rm_ofs, vec_size, vec_size);
5920 case NEON_2RM_VCEQ0:
5921 gen_gvec_ceq0(size, rd_ofs, rm_ofs, vec_size, vec_size);
5923 case NEON_2RM_VCGT0:
5924 gen_gvec_cgt0(size, rd_ofs, rm_ofs, vec_size, vec_size);
5926 case NEON_2RM_VCLE0:
5927 gen_gvec_cle0(size, rd_ofs, rm_ofs, vec_size, vec_size);
5929 case NEON_2RM_VCGE0:
5930 gen_gvec_cge0(size, rd_ofs, rm_ofs, vec_size, vec_size);
5932 case NEON_2RM_VCLT0:
5933 gen_gvec_clt0(size, rd_ofs, rm_ofs, vec_size, vec_size);
5938 for (pass = 0; pass < (q ? 4 : 2); pass++) {
5939 tmp = neon_load_reg(rm, pass);
5941 case NEON_2RM_VREV32:
5943 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
5944 case 1: gen_swap_half(tmp); break;
5948 case NEON_2RM_VREV16:
5949 gen_rev16(tmp, tmp);
5953 case 0: gen_helper_neon_cls_s8(tmp, tmp); break;
5954 case 1: gen_helper_neon_cls_s16(tmp, tmp); break;
5955 case 2: gen_helper_neon_cls_s32(tmp, tmp); break;
5961 case 0: gen_helper_neon_clz_u8(tmp, tmp); break;
5962 case 1: gen_helper_neon_clz_u16(tmp, tmp); break;
5963 case 2: tcg_gen_clzi_i32(tmp, tmp, 32); break;
5968 gen_helper_neon_cnt_u8(tmp, tmp);
5970 case NEON_2RM_VQABS:
5973 gen_helper_neon_qabs_s8(tmp, cpu_env, tmp);
5976 gen_helper_neon_qabs_s16(tmp, cpu_env, tmp);
5979 gen_helper_neon_qabs_s32(tmp, cpu_env, tmp);
5984 case NEON_2RM_VQNEG:
5987 gen_helper_neon_qneg_s8(tmp, cpu_env, tmp);
5990 gen_helper_neon_qneg_s16(tmp, cpu_env, tmp);
5993 gen_helper_neon_qneg_s32(tmp, cpu_env, tmp);
5998 case NEON_2RM_VCGT0_F:
6000 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6001 tmp2 = tcg_const_i32(0);
6002 gen_helper_neon_cgt_f32(tmp, tmp, tmp2, fpstatus);
6003 tcg_temp_free_i32(tmp2);
6004 tcg_temp_free_ptr(fpstatus);
6007 case NEON_2RM_VCGE0_F:
6009 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6010 tmp2 = tcg_const_i32(0);
6011 gen_helper_neon_cge_f32(tmp, tmp, tmp2, fpstatus);
6012 tcg_temp_free_i32(tmp2);
6013 tcg_temp_free_ptr(fpstatus);
6016 case NEON_2RM_VCEQ0_F:
6018 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6019 tmp2 = tcg_const_i32(0);
6020 gen_helper_neon_ceq_f32(tmp, tmp, tmp2, fpstatus);
6021 tcg_temp_free_i32(tmp2);
6022 tcg_temp_free_ptr(fpstatus);
6025 case NEON_2RM_VCLE0_F:
6027 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6028 tmp2 = tcg_const_i32(0);
6029 gen_helper_neon_cge_f32(tmp, tmp2, tmp, fpstatus);
6030 tcg_temp_free_i32(tmp2);
6031 tcg_temp_free_ptr(fpstatus);
6034 case NEON_2RM_VCLT0_F:
6036 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6037 tmp2 = tcg_const_i32(0);
6038 gen_helper_neon_cgt_f32(tmp, tmp2, tmp, fpstatus);
6039 tcg_temp_free_i32(tmp2);
6040 tcg_temp_free_ptr(fpstatus);
6043 case NEON_2RM_VABS_F:
6044 gen_helper_vfp_abss(tmp, tmp);
6046 case NEON_2RM_VNEG_F:
6047 gen_helper_vfp_negs(tmp, tmp);
6050 tmp2 = neon_load_reg(rd, pass);
6051 neon_store_reg(rm, pass, tmp2);
6054 tmp2 = neon_load_reg(rd, pass);
6056 case 0: gen_neon_trn_u8(tmp, tmp2); break;
6057 case 1: gen_neon_trn_u16(tmp, tmp2); break;
6060 neon_store_reg(rm, pass, tmp2);
6062 case NEON_2RM_VRINTN:
6063 case NEON_2RM_VRINTA:
6064 case NEON_2RM_VRINTM:
6065 case NEON_2RM_VRINTP:
6066 case NEON_2RM_VRINTZ:
6069 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6072 if (op == NEON_2RM_VRINTZ) {
6073 rmode = FPROUNDING_ZERO;
6075 rmode = fp_decode_rm[((op & 0x6) >> 1) ^ 1];
6078 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rmode));
6079 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
6081 gen_helper_rints(tmp, tmp, fpstatus);
6082 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
6084 tcg_temp_free_ptr(fpstatus);
6085 tcg_temp_free_i32(tcg_rmode);
6088 case NEON_2RM_VRINTX:
6090 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6091 gen_helper_rints_exact(tmp, tmp, fpstatus);
6092 tcg_temp_free_ptr(fpstatus);
6095 case NEON_2RM_VCVTAU:
6096 case NEON_2RM_VCVTAS:
6097 case NEON_2RM_VCVTNU:
6098 case NEON_2RM_VCVTNS:
6099 case NEON_2RM_VCVTPU:
6100 case NEON_2RM_VCVTPS:
6101 case NEON_2RM_VCVTMU:
6102 case NEON_2RM_VCVTMS:
6104 bool is_signed = !extract32(insn, 7, 1);
6105 TCGv_ptr fpst = get_fpstatus_ptr(1);
6106 TCGv_i32 tcg_rmode, tcg_shift;
6107 int rmode = fp_decode_rm[extract32(insn, 8, 2)];
6109 tcg_shift = tcg_const_i32(0);
6110 tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rmode));
6111 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
6115 gen_helper_vfp_tosls(tmp, tmp,
6118 gen_helper_vfp_touls(tmp, tmp,
6122 gen_helper_set_neon_rmode(tcg_rmode, tcg_rmode,
6124 tcg_temp_free_i32(tcg_rmode);
6125 tcg_temp_free_i32(tcg_shift);
6126 tcg_temp_free_ptr(fpst);
6129 case NEON_2RM_VRECPE:
6130 gen_helper_recpe_u32(tmp, tmp);
6132 case NEON_2RM_VRSQRTE:
6133 gen_helper_rsqrte_u32(tmp, tmp);
6135 case NEON_2RM_VRECPE_F:
6137 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6138 gen_helper_recpe_f32(tmp, tmp, fpstatus);
6139 tcg_temp_free_ptr(fpstatus);
6142 case NEON_2RM_VRSQRTE_F:
6144 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6145 gen_helper_rsqrte_f32(tmp, tmp, fpstatus);
6146 tcg_temp_free_ptr(fpstatus);
6149 case NEON_2RM_VCVT_FS: /* VCVT.F32.S32 */
6151 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6152 gen_helper_vfp_sitos(tmp, tmp, fpstatus);
6153 tcg_temp_free_ptr(fpstatus);
6156 case NEON_2RM_VCVT_FU: /* VCVT.F32.U32 */
6158 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6159 gen_helper_vfp_uitos(tmp, tmp, fpstatus);
6160 tcg_temp_free_ptr(fpstatus);
6163 case NEON_2RM_VCVT_SF: /* VCVT.S32.F32 */
6165 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6166 gen_helper_vfp_tosizs(tmp, tmp, fpstatus);
6167 tcg_temp_free_ptr(fpstatus);
6170 case NEON_2RM_VCVT_UF: /* VCVT.U32.F32 */
6172 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6173 gen_helper_vfp_touizs(tmp, tmp, fpstatus);
6174 tcg_temp_free_ptr(fpstatus);
6178 /* Reserved op values were caught by the
6179 * neon_2rm_sizes[] check earlier.
6183 neon_store_reg(rd, pass, tmp);
6187 } else if ((insn & (1 << 10)) == 0) {
6189 int n = ((insn >> 8) & 3) + 1;
6190 if ((rn + n) > 32) {
6191 /* This is UNPREDICTABLE; we choose to UNDEF to avoid the
6192 * helper function running off the end of the register file.
6197 if (insn & (1 << 6)) {
6198 tmp = neon_load_reg(rd, 0);
6200 tmp = tcg_temp_new_i32();
6201 tcg_gen_movi_i32(tmp, 0);
6203 tmp2 = neon_load_reg(rm, 0);
6204 ptr1 = vfp_reg_ptr(true, rn);
6205 tmp5 = tcg_const_i32(n);
6206 gen_helper_neon_tbl(tmp2, tmp2, tmp, ptr1, tmp5);
6207 tcg_temp_free_i32(tmp);
6208 if (insn & (1 << 6)) {
6209 tmp = neon_load_reg(rd, 1);
6211 tmp = tcg_temp_new_i32();
6212 tcg_gen_movi_i32(tmp, 0);
6214 tmp3 = neon_load_reg(rm, 1);
6215 gen_helper_neon_tbl(tmp3, tmp3, tmp, ptr1, tmp5);
6216 tcg_temp_free_i32(tmp5);
6217 tcg_temp_free_ptr(ptr1);
6218 neon_store_reg(rd, 0, tmp2);
6219 neon_store_reg(rd, 1, tmp3);
6220 tcg_temp_free_i32(tmp);
6221 } else if ((insn & 0x380) == 0) {
6226 if ((insn & (7 << 16)) == 0 || (q && (rd & 1))) {
6229 if (insn & (1 << 16)) {
6231 element = (insn >> 17) & 7;
6232 } else if (insn & (1 << 17)) {
6234 element = (insn >> 18) & 3;
6237 element = (insn >> 19) & 1;
6239 tcg_gen_gvec_dup_mem(size, neon_reg_offset(rd, 0),
6240 neon_element_offset(rm, element, size),
6241 q ? 16 : 8, q ? 16 : 8);
6250 static int disas_coproc_insn(DisasContext *s, uint32_t insn)
6252 int cpnum, is64, crn, crm, opc1, opc2, isread, rt, rt2;
6253 const ARMCPRegInfo *ri;
6255 cpnum = (insn >> 8) & 0xf;
6257 /* First check for coprocessor space used for XScale/iwMMXt insns */
6258 if (arm_dc_feature(s, ARM_FEATURE_XSCALE) && (cpnum < 2)) {
6259 if (extract32(s->c15_cpar, cpnum, 1) == 0) {
6262 if (arm_dc_feature(s, ARM_FEATURE_IWMMXT)) {
6263 return disas_iwmmxt_insn(s, insn);
6264 } else if (arm_dc_feature(s, ARM_FEATURE_XSCALE)) {
6265 return disas_dsp_insn(s, insn);
6270 /* Otherwise treat as a generic register access */
6271 is64 = (insn & (1 << 25)) == 0;
6272 if (!is64 && ((insn & (1 << 4)) == 0)) {
6280 opc1 = (insn >> 4) & 0xf;
6282 rt2 = (insn >> 16) & 0xf;
6284 crn = (insn >> 16) & 0xf;
6285 opc1 = (insn >> 21) & 7;
6286 opc2 = (insn >> 5) & 7;
6289 isread = (insn >> 20) & 1;
6290 rt = (insn >> 12) & 0xf;
6292 ri = get_arm_cp_reginfo(s->cp_regs,
6293 ENCODE_CP_REG(cpnum, is64, s->ns, crn, crm, opc1, opc2));
6297 /* Check access permissions */
6298 if (!cp_access_ok(s->current_el, ri, isread)) {
6302 if (s->hstr_active || ri->accessfn ||
6303 (arm_dc_feature(s, ARM_FEATURE_XSCALE) && cpnum < 14)) {
6304 /* Emit code to perform further access permissions checks at
6305 * runtime; this may result in an exception.
6306 * Note that on XScale all cp0..c13 registers do an access check
6307 * call in order to handle c15_cpar.
6310 TCGv_i32 tcg_syn, tcg_isread;
6313 /* Note that since we are an implementation which takes an
6314 * exception on a trapped conditional instruction only if the
6315 * instruction passes its condition code check, we can take
6316 * advantage of the clause in the ARM ARM that allows us to set
6317 * the COND field in the instruction to 0xE in all cases.
6318 * We could fish the actual condition out of the insn (ARM)
6319 * or the condexec bits (Thumb) but it isn't necessary.
6324 syndrome = syn_cp14_rrt_trap(1, 0xe, opc1, crm, rt, rt2,
6327 syndrome = syn_cp14_rt_trap(1, 0xe, opc1, opc2, crn, crm,
6333 syndrome = syn_cp15_rrt_trap(1, 0xe, opc1, crm, rt, rt2,
6336 syndrome = syn_cp15_rt_trap(1, 0xe, opc1, opc2, crn, crm,
6341 /* ARMv8 defines that only coprocessors 14 and 15 exist,
6342 * so this can only happen if this is an ARMv7 or earlier CPU,
6343 * in which case the syndrome information won't actually be
6346 assert(!arm_dc_feature(s, ARM_FEATURE_V8));
6347 syndrome = syn_uncategorized();
6351 gen_set_condexec(s);
6352 gen_set_pc_im(s, s->pc_curr);
6353 tmpptr = tcg_const_ptr(ri);
6354 tcg_syn = tcg_const_i32(syndrome);
6355 tcg_isread = tcg_const_i32(isread);
6356 gen_helper_access_check_cp_reg(cpu_env, tmpptr, tcg_syn,
6358 tcg_temp_free_ptr(tmpptr);
6359 tcg_temp_free_i32(tcg_syn);
6360 tcg_temp_free_i32(tcg_isread);
6361 } else if (ri->type & ARM_CP_RAISES_EXC) {
6363 * The readfn or writefn might raise an exception;
6364 * synchronize the CPU state in case it does.
6366 gen_set_condexec(s);
6367 gen_set_pc_im(s, s->pc_curr);
6370 /* Handle special cases first */
6371 switch (ri->type & ~(ARM_CP_FLAG_MASK & ~ARM_CP_SPECIAL)) {
6378 gen_set_pc_im(s, s->base.pc_next);
6379 s->base.is_jmp = DISAS_WFI;
6385 if ((tb_cflags(s->base.tb) & CF_USE_ICOUNT) && (ri->type & ARM_CP_IO)) {
6394 if (ri->type & ARM_CP_CONST) {
6395 tmp64 = tcg_const_i64(ri->resetvalue);
6396 } else if (ri->readfn) {
6398 tmp64 = tcg_temp_new_i64();
6399 tmpptr = tcg_const_ptr(ri);
6400 gen_helper_get_cp_reg64(tmp64, cpu_env, tmpptr);
6401 tcg_temp_free_ptr(tmpptr);
6403 tmp64 = tcg_temp_new_i64();
6404 tcg_gen_ld_i64(tmp64, cpu_env, ri->fieldoffset);
6406 tmp = tcg_temp_new_i32();
6407 tcg_gen_extrl_i64_i32(tmp, tmp64);
6408 store_reg(s, rt, tmp);
6409 tmp = tcg_temp_new_i32();
6410 tcg_gen_extrh_i64_i32(tmp, tmp64);
6411 tcg_temp_free_i64(tmp64);
6412 store_reg(s, rt2, tmp);
6415 if (ri->type & ARM_CP_CONST) {
6416 tmp = tcg_const_i32(ri->resetvalue);
6417 } else if (ri->readfn) {
6419 tmp = tcg_temp_new_i32();
6420 tmpptr = tcg_const_ptr(ri);
6421 gen_helper_get_cp_reg(tmp, cpu_env, tmpptr);
6422 tcg_temp_free_ptr(tmpptr);
6424 tmp = load_cpu_offset(ri->fieldoffset);
6427 /* Destination register of r15 for 32 bit loads sets
6428 * the condition codes from the high 4 bits of the value
6431 tcg_temp_free_i32(tmp);
6433 store_reg(s, rt, tmp);
6438 if (ri->type & ARM_CP_CONST) {
6439 /* If not forbidden by access permissions, treat as WI */
6444 TCGv_i32 tmplo, tmphi;
6445 TCGv_i64 tmp64 = tcg_temp_new_i64();
6446 tmplo = load_reg(s, rt);
6447 tmphi = load_reg(s, rt2);
6448 tcg_gen_concat_i32_i64(tmp64, tmplo, tmphi);
6449 tcg_temp_free_i32(tmplo);
6450 tcg_temp_free_i32(tmphi);
6452 TCGv_ptr tmpptr = tcg_const_ptr(ri);
6453 gen_helper_set_cp_reg64(cpu_env, tmpptr, tmp64);
6454 tcg_temp_free_ptr(tmpptr);
6456 tcg_gen_st_i64(tmp64, cpu_env, ri->fieldoffset);
6458 tcg_temp_free_i64(tmp64);
6463 tmp = load_reg(s, rt);
6464 tmpptr = tcg_const_ptr(ri);
6465 gen_helper_set_cp_reg(cpu_env, tmpptr, tmp);
6466 tcg_temp_free_ptr(tmpptr);
6467 tcg_temp_free_i32(tmp);
6469 TCGv_i32 tmp = load_reg(s, rt);
6470 store_cpu_offset(tmp, ri->fieldoffset);
6475 /* I/O operations must end the TB here (whether read or write) */
6476 need_exit_tb = ((tb_cflags(s->base.tb) & CF_USE_ICOUNT) &&
6477 (ri->type & ARM_CP_IO));
6479 if (!isread && !(ri->type & ARM_CP_SUPPRESS_TB_END)) {
6481 * A write to any coprocessor register that ends a TB
6482 * must rebuild the hflags for the next TB.
6484 TCGv_i32 tcg_el = tcg_const_i32(s->current_el);
6485 if (arm_dc_feature(s, ARM_FEATURE_M)) {
6486 gen_helper_rebuild_hflags_m32(cpu_env, tcg_el);
6488 if (ri->type & ARM_CP_NEWEL) {
6489 gen_helper_rebuild_hflags_a32_newel(cpu_env);
6491 gen_helper_rebuild_hflags_a32(cpu_env, tcg_el);
6494 tcg_temp_free_i32(tcg_el);
6496 * We default to ending the TB on a coprocessor register write,
6497 * but allow this to be suppressed by the register definition
6498 * (usually only necessary to work around guest bugs).
6500 need_exit_tb = true;
6509 /* Unknown register; this might be a guest error or a QEMU
6510 * unimplemented feature.
6513 qemu_log_mask(LOG_UNIMP, "%s access to unsupported AArch32 "
6514 "64 bit system register cp:%d opc1: %d crm:%d "
6516 isread ? "read" : "write", cpnum, opc1, crm,
6517 s->ns ? "non-secure" : "secure");
6519 qemu_log_mask(LOG_UNIMP, "%s access to unsupported AArch32 "
6520 "system register cp:%d opc1:%d crn:%d crm:%d opc2:%d "
6522 isread ? "read" : "write", cpnum, opc1, crn, crm, opc2,
6523 s->ns ? "non-secure" : "secure");
6530 /* Store a 64-bit value to a register pair. Clobbers val. */
6531 static void gen_storeq_reg(DisasContext *s, int rlow, int rhigh, TCGv_i64 val)
6534 tmp = tcg_temp_new_i32();
6535 tcg_gen_extrl_i64_i32(tmp, val);
6536 store_reg(s, rlow, tmp);
6537 tmp = tcg_temp_new_i32();
6538 tcg_gen_extrh_i64_i32(tmp, val);
6539 store_reg(s, rhigh, tmp);
6542 /* load and add a 64-bit value from a register pair. */
6543 static void gen_addq(DisasContext *s, TCGv_i64 val, int rlow, int rhigh)
6549 /* Load 64-bit value rd:rn. */
6550 tmpl = load_reg(s, rlow);
6551 tmph = load_reg(s, rhigh);
6552 tmp = tcg_temp_new_i64();
6553 tcg_gen_concat_i32_i64(tmp, tmpl, tmph);
6554 tcg_temp_free_i32(tmpl);
6555 tcg_temp_free_i32(tmph);
6556 tcg_gen_add_i64(val, val, tmp);
6557 tcg_temp_free_i64(tmp);
6560 /* Set N and Z flags from hi|lo. */
6561 static void gen_logicq_cc(TCGv_i32 lo, TCGv_i32 hi)
6563 tcg_gen_mov_i32(cpu_NF, hi);
6564 tcg_gen_or_i32(cpu_ZF, lo, hi);
6567 /* Load/Store exclusive instructions are implemented by remembering
6568 the value/address loaded, and seeing if these are the same
6569 when the store is performed. This should be sufficient to implement
6570 the architecturally mandated semantics, and avoids having to monitor
6571 regular stores. The compare vs the remembered value is done during
6572 the cmpxchg operation, but we must compare the addresses manually. */
6573 static void gen_load_exclusive(DisasContext *s, int rt, int rt2,
6574 TCGv_i32 addr, int size)
6576 TCGv_i32 tmp = tcg_temp_new_i32();
6577 MemOp opc = size | MO_ALIGN | s->be_data;
6582 TCGv_i32 tmp2 = tcg_temp_new_i32();
6583 TCGv_i64 t64 = tcg_temp_new_i64();
6585 /* For AArch32, architecturally the 32-bit word at the lowest
6586 * address is always Rt and the one at addr+4 is Rt2, even if
6587 * the CPU is big-endian. That means we don't want to do a
6588 * gen_aa32_ld_i64(), which invokes gen_aa32_frob64() as if
6589 * for an architecturally 64-bit access, but instead do a
6590 * 64-bit access using MO_BE if appropriate and then split
6592 * This only makes a difference for BE32 user-mode, where
6593 * frob64() must not flip the two halves of the 64-bit data
6594 * but this code must treat BE32 user-mode like BE32 system.
6596 TCGv taddr = gen_aa32_addr(s, addr, opc);
6598 tcg_gen_qemu_ld_i64(t64, taddr, get_mem_index(s), opc);
6599 tcg_temp_free(taddr);
6600 tcg_gen_mov_i64(cpu_exclusive_val, t64);
6601 if (s->be_data == MO_BE) {
6602 tcg_gen_extr_i64_i32(tmp2, tmp, t64);
6604 tcg_gen_extr_i64_i32(tmp, tmp2, t64);
6606 tcg_temp_free_i64(t64);
6608 store_reg(s, rt2, tmp2);
6610 gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), opc);
6611 tcg_gen_extu_i32_i64(cpu_exclusive_val, tmp);
6614 store_reg(s, rt, tmp);
6615 tcg_gen_extu_i32_i64(cpu_exclusive_addr, addr);
6618 static void gen_clrex(DisasContext *s)
6620 tcg_gen_movi_i64(cpu_exclusive_addr, -1);
6623 static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
6624 TCGv_i32 addr, int size)
6626 TCGv_i32 t0, t1, t2;
6629 TCGLabel *done_label;
6630 TCGLabel *fail_label;
6631 MemOp opc = size | MO_ALIGN | s->be_data;
6633 /* if (env->exclusive_addr == addr && env->exclusive_val == [addr]) {
6639 fail_label = gen_new_label();
6640 done_label = gen_new_label();
6641 extaddr = tcg_temp_new_i64();
6642 tcg_gen_extu_i32_i64(extaddr, addr);
6643 tcg_gen_brcond_i64(TCG_COND_NE, extaddr, cpu_exclusive_addr, fail_label);
6644 tcg_temp_free_i64(extaddr);
6646 taddr = gen_aa32_addr(s, addr, opc);
6647 t0 = tcg_temp_new_i32();
6648 t1 = load_reg(s, rt);
6650 TCGv_i64 o64 = tcg_temp_new_i64();
6651 TCGv_i64 n64 = tcg_temp_new_i64();
6653 t2 = load_reg(s, rt2);
6654 /* For AArch32, architecturally the 32-bit word at the lowest
6655 * address is always Rt and the one at addr+4 is Rt2, even if
6656 * the CPU is big-endian. Since we're going to treat this as a
6657 * single 64-bit BE store, we need to put the two halves in the
6658 * opposite order for BE to LE, so that they end up in the right
6660 * We don't want gen_aa32_frob64() because that does the wrong
6661 * thing for BE32 usermode.
6663 if (s->be_data == MO_BE) {
6664 tcg_gen_concat_i32_i64(n64, t2, t1);
6666 tcg_gen_concat_i32_i64(n64, t1, t2);
6668 tcg_temp_free_i32(t2);
6670 tcg_gen_atomic_cmpxchg_i64(o64, taddr, cpu_exclusive_val, n64,
6671 get_mem_index(s), opc);
6672 tcg_temp_free_i64(n64);
6674 tcg_gen_setcond_i64(TCG_COND_NE, o64, o64, cpu_exclusive_val);
6675 tcg_gen_extrl_i64_i32(t0, o64);
6677 tcg_temp_free_i64(o64);
6679 t2 = tcg_temp_new_i32();
6680 tcg_gen_extrl_i64_i32(t2, cpu_exclusive_val);
6681 tcg_gen_atomic_cmpxchg_i32(t0, taddr, t2, t1, get_mem_index(s), opc);
6682 tcg_gen_setcond_i32(TCG_COND_NE, t0, t0, t2);
6683 tcg_temp_free_i32(t2);
6685 tcg_temp_free_i32(t1);
6686 tcg_temp_free(taddr);
6687 tcg_gen_mov_i32(cpu_R[rd], t0);
6688 tcg_temp_free_i32(t0);
6689 tcg_gen_br(done_label);
6691 gen_set_label(fail_label);
6692 tcg_gen_movi_i32(cpu_R[rd], 1);
6693 gen_set_label(done_label);
6694 tcg_gen_movi_i64(cpu_exclusive_addr, -1);
6700 * @mode: mode field from insn (which stack to store to)
6701 * @amode: addressing mode (DA/IA/DB/IB), encoded as per P,U bits in ARM insn
6702 * @writeback: true if writeback bit set
6704 * Generate code for the SRS (Store Return State) insn.
6706 static void gen_srs(DisasContext *s,
6707 uint32_t mode, uint32_t amode, bool writeback)
6714 * - trapped to EL3 if EL3 is AArch64 and we are at Secure EL1
6715 * and specified mode is monitor mode
6716 * - UNDEFINED in Hyp mode
6717 * - UNPREDICTABLE in User or System mode
6718 * - UNPREDICTABLE if the specified mode is:
6719 * -- not implemented
6720 * -- not a valid mode number
6721 * -- a mode that's at a higher exception level
6722 * -- Monitor, if we are Non-secure
6723 * For the UNPREDICTABLE cases we choose to UNDEF.
6725 if (s->current_el == 1 && !s->ns && mode == ARM_CPU_MODE_MON) {
6726 gen_exception_insn(s, s->pc_curr, EXCP_UDEF, syn_uncategorized(), 3);
6730 if (s->current_el == 0 || s->current_el == 2) {
6735 case ARM_CPU_MODE_USR:
6736 case ARM_CPU_MODE_FIQ:
6737 case ARM_CPU_MODE_IRQ:
6738 case ARM_CPU_MODE_SVC:
6739 case ARM_CPU_MODE_ABT:
6740 case ARM_CPU_MODE_UND:
6741 case ARM_CPU_MODE_SYS:
6743 case ARM_CPU_MODE_HYP:
6744 if (s->current_el == 1 || !arm_dc_feature(s, ARM_FEATURE_EL2)) {
6748 case ARM_CPU_MODE_MON:
6749 /* No need to check specifically for "are we non-secure" because
6750 * we've already made EL0 UNDEF and handled the trap for S-EL1;
6751 * so if this isn't EL3 then we must be non-secure.
6753 if (s->current_el != 3) {
6762 unallocated_encoding(s);
6766 addr = tcg_temp_new_i32();
6767 tmp = tcg_const_i32(mode);
6768 /* get_r13_banked() will raise an exception if called from System mode */
6769 gen_set_condexec(s);
6770 gen_set_pc_im(s, s->pc_curr);
6771 gen_helper_get_r13_banked(addr, cpu_env, tmp);
6772 tcg_temp_free_i32(tmp);
6789 tcg_gen_addi_i32(addr, addr, offset);
6790 tmp = load_reg(s, 14);
6791 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
6792 tcg_temp_free_i32(tmp);
6793 tmp = load_cpu_field(spsr);
6794 tcg_gen_addi_i32(addr, addr, 4);
6795 gen_aa32_st32(s, tmp, addr, get_mem_index(s));
6796 tcg_temp_free_i32(tmp);
6814 tcg_gen_addi_i32(addr, addr, offset);
6815 tmp = tcg_const_i32(mode);
6816 gen_helper_set_r13_banked(cpu_env, tmp, addr);
6817 tcg_temp_free_i32(tmp);
6819 tcg_temp_free_i32(addr);
6820 s->base.is_jmp = DISAS_UPDATE;
6823 /* Generate a label used for skipping this instruction */
6824 static void arm_gen_condlabel(DisasContext *s)
6827 s->condlabel = gen_new_label();
6832 /* Skip this instruction if the ARM condition is false */
6833 static void arm_skip_unless(DisasContext *s, uint32_t cond)
6835 arm_gen_condlabel(s);
6836 arm_gen_test_cc(cond ^ 1, s->condlabel);
6841 * Constant expanders for the decoders.
6844 static int negate(DisasContext *s, int x)
6849 static int plus_2(DisasContext *s, int x)
6854 static int times_2(DisasContext *s, int x)
6859 static int times_4(DisasContext *s, int x)
6864 /* Return only the rotation part of T32ExpandImm. */
6865 static int t32_expandimm_rot(DisasContext *s, int x)
6867 return x & 0xc00 ? extract32(x, 7, 5) : 0;
6870 /* Return the unrotated immediate from T32ExpandImm. */
6871 static int t32_expandimm_imm(DisasContext *s, int x)
6873 int imm = extract32(x, 0, 8);
6875 switch (extract32(x, 8, 4)) {
6877 /* Nothing to do. */
6879 case 1: /* 00XY00XY */
6882 case 2: /* XY00XY00 */
6885 case 3: /* XYXYXYXY */
6889 /* Rotated constant. */
6896 static int t32_branch24(DisasContext *s, int x)
6898 /* Convert J1:J2 at x[22:21] to I2:I1, which involves I=J^~S. */
6899 x ^= !(x < 0) * (3 << 21);
6900 /* Append the final zero. */
6904 static int t16_setflags(DisasContext *s)
6906 return s->condexec_mask == 0;
6909 static int t16_push_list(DisasContext *s, int x)
6911 return (x & 0xff) | (x & 0x100) << (14 - 8);
6914 static int t16_pop_list(DisasContext *s, int x)
6916 return (x & 0xff) | (x & 0x100) << (15 - 8);
6920 * Include the generated decoders.
6923 #include "decode-a32.inc.c"
6924 #include "decode-a32-uncond.inc.c"
6925 #include "decode-t32.inc.c"
6926 #include "decode-t16.inc.c"
6928 /* Helpers to swap operands for reverse-subtract. */
6929 static void gen_rsb(TCGv_i32 dst, TCGv_i32 a, TCGv_i32 b)
6931 tcg_gen_sub_i32(dst, b, a);
6934 static void gen_rsb_CC(TCGv_i32 dst, TCGv_i32 a, TCGv_i32 b)
6936 gen_sub_CC(dst, b, a);
6939 static void gen_rsc(TCGv_i32 dest, TCGv_i32 a, TCGv_i32 b)
6941 gen_sub_carry(dest, b, a);
6944 static void gen_rsc_CC(TCGv_i32 dest, TCGv_i32 a, TCGv_i32 b)
6946 gen_sbc_CC(dest, b, a);
6950 * Helpers for the data processing routines.
6952 * After the computation store the results back.
6953 * This may be suppressed altogether (STREG_NONE), require a runtime
6954 * check against the stack limits (STREG_SP_CHECK), or generate an
6955 * exception return. Oh, or store into a register.
6957 * Always return true, indicating success for a trans_* function.
6966 static bool store_reg_kind(DisasContext *s, int rd,
6967 TCGv_i32 val, StoreRegKind kind)
6971 tcg_temp_free_i32(val);
6974 /* See ALUWritePC: Interworking only from a32 mode. */
6976 store_reg(s, rd, val);
6978 store_reg_bx(s, rd, val);
6981 case STREG_SP_CHECK:
6982 store_sp_checked(s, val);
6985 gen_exception_return(s, val);
6988 g_assert_not_reached();
6992 * Data Processing (register)
6994 * Operate, with set flags, one register source,
6995 * one immediate shifted register source, and a destination.
6997 static bool op_s_rrr_shi(DisasContext *s, arg_s_rrr_shi *a,
6998 void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32),
6999 int logic_cc, StoreRegKind kind)
7001 TCGv_i32 tmp1, tmp2;
7003 tmp2 = load_reg(s, a->rm);
7004 gen_arm_shift_im(tmp2, a->shty, a->shim, logic_cc);
7005 tmp1 = load_reg(s, a->rn);
7007 gen(tmp1, tmp1, tmp2);
7008 tcg_temp_free_i32(tmp2);
7013 return store_reg_kind(s, a->rd, tmp1, kind);
7016 static bool op_s_rxr_shi(DisasContext *s, arg_s_rrr_shi *a,
7017 void (*gen)(TCGv_i32, TCGv_i32),
7018 int logic_cc, StoreRegKind kind)
7022 tmp = load_reg(s, a->rm);
7023 gen_arm_shift_im(tmp, a->shty, a->shim, logic_cc);
7029 return store_reg_kind(s, a->rd, tmp, kind);
7033 * Data-processing (register-shifted register)
7035 * Operate, with set flags, one register source,
7036 * one register shifted register source, and a destination.
7038 static bool op_s_rrr_shr(DisasContext *s, arg_s_rrr_shr *a,
7039 void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32),
7040 int logic_cc, StoreRegKind kind)
7042 TCGv_i32 tmp1, tmp2;
7044 tmp1 = load_reg(s, a->rs);
7045 tmp2 = load_reg(s, a->rm);
7046 gen_arm_shift_reg(tmp2, a->shty, tmp1, logic_cc);
7047 tmp1 = load_reg(s, a->rn);
7049 gen(tmp1, tmp1, tmp2);
7050 tcg_temp_free_i32(tmp2);
7055 return store_reg_kind(s, a->rd, tmp1, kind);
7058 static bool op_s_rxr_shr(DisasContext *s, arg_s_rrr_shr *a,
7059 void (*gen)(TCGv_i32, TCGv_i32),
7060 int logic_cc, StoreRegKind kind)
7062 TCGv_i32 tmp1, tmp2;
7064 tmp1 = load_reg(s, a->rs);
7065 tmp2 = load_reg(s, a->rm);
7066 gen_arm_shift_reg(tmp2, a->shty, tmp1, logic_cc);
7072 return store_reg_kind(s, a->rd, tmp2, kind);
7076 * Data-processing (immediate)
7078 * Operate, with set flags, one register source,
7079 * one rotated immediate, and a destination.
7081 * Note that logic_cc && a->rot setting CF based on the msb of the
7082 * immediate is the reason why we must pass in the unrotated form
7085 static bool op_s_rri_rot(DisasContext *s, arg_s_rri_rot *a,
7086 void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32),
7087 int logic_cc, StoreRegKind kind)
7089 TCGv_i32 tmp1, tmp2;
7092 imm = ror32(a->imm, a->rot);
7093 if (logic_cc && a->rot) {
7094 tcg_gen_movi_i32(cpu_CF, imm >> 31);
7096 tmp2 = tcg_const_i32(imm);
7097 tmp1 = load_reg(s, a->rn);
7099 gen(tmp1, tmp1, tmp2);
7100 tcg_temp_free_i32(tmp2);
7105 return store_reg_kind(s, a->rd, tmp1, kind);
7108 static bool op_s_rxi_rot(DisasContext *s, arg_s_rri_rot *a,
7109 void (*gen)(TCGv_i32, TCGv_i32),
7110 int logic_cc, StoreRegKind kind)
7115 imm = ror32(a->imm, a->rot);
7116 if (logic_cc && a->rot) {
7117 tcg_gen_movi_i32(cpu_CF, imm >> 31);
7119 tmp = tcg_const_i32(imm);
7125 return store_reg_kind(s, a->rd, tmp, kind);
7128 #define DO_ANY3(NAME, OP, L, K) \
7129 static bool trans_##NAME##_rrri(DisasContext *s, arg_s_rrr_shi *a) \
7130 { StoreRegKind k = (K); return op_s_rrr_shi(s, a, OP, L, k); } \
7131 static bool trans_##NAME##_rrrr(DisasContext *s, arg_s_rrr_shr *a) \
7132 { StoreRegKind k = (K); return op_s_rrr_shr(s, a, OP, L, k); } \
7133 static bool trans_##NAME##_rri(DisasContext *s, arg_s_rri_rot *a) \
7134 { StoreRegKind k = (K); return op_s_rri_rot(s, a, OP, L, k); }
7136 #define DO_ANY2(NAME, OP, L, K) \
7137 static bool trans_##NAME##_rxri(DisasContext *s, arg_s_rrr_shi *a) \
7138 { StoreRegKind k = (K); return op_s_rxr_shi(s, a, OP, L, k); } \
7139 static bool trans_##NAME##_rxrr(DisasContext *s, arg_s_rrr_shr *a) \
7140 { StoreRegKind k = (K); return op_s_rxr_shr(s, a, OP, L, k); } \
7141 static bool trans_##NAME##_rxi(DisasContext *s, arg_s_rri_rot *a) \
7142 { StoreRegKind k = (K); return op_s_rxi_rot(s, a, OP, L, k); }
7144 #define DO_CMP2(NAME, OP, L) \
7145 static bool trans_##NAME##_xrri(DisasContext *s, arg_s_rrr_shi *a) \
7146 { return op_s_rrr_shi(s, a, OP, L, STREG_NONE); } \
7147 static bool trans_##NAME##_xrrr(DisasContext *s, arg_s_rrr_shr *a) \
7148 { return op_s_rrr_shr(s, a, OP, L, STREG_NONE); } \
7149 static bool trans_##NAME##_xri(DisasContext *s, arg_s_rri_rot *a) \
7150 { return op_s_rri_rot(s, a, OP, L, STREG_NONE); }
7152 DO_ANY3(AND, tcg_gen_and_i32, a->s, STREG_NORMAL)
7153 DO_ANY3(EOR, tcg_gen_xor_i32, a->s, STREG_NORMAL)
7154 DO_ANY3(ORR, tcg_gen_or_i32, a->s, STREG_NORMAL)
7155 DO_ANY3(BIC, tcg_gen_andc_i32, a->s, STREG_NORMAL)
7157 DO_ANY3(RSB, a->s ? gen_rsb_CC : gen_rsb, false, STREG_NORMAL)
7158 DO_ANY3(ADC, a->s ? gen_adc_CC : gen_add_carry, false, STREG_NORMAL)
7159 DO_ANY3(SBC, a->s ? gen_sbc_CC : gen_sub_carry, false, STREG_NORMAL)
7160 DO_ANY3(RSC, a->s ? gen_rsc_CC : gen_rsc, false, STREG_NORMAL)
7162 DO_CMP2(TST, tcg_gen_and_i32, true)
7163 DO_CMP2(TEQ, tcg_gen_xor_i32, true)
7164 DO_CMP2(CMN, gen_add_CC, false)
7165 DO_CMP2(CMP, gen_sub_CC, false)
7167 DO_ANY3(ADD, a->s ? gen_add_CC : tcg_gen_add_i32, false,
7168 a->rd == 13 && a->rn == 13 ? STREG_SP_CHECK : STREG_NORMAL)
7171 * Note for the computation of StoreRegKind we return out of the
7172 * middle of the functions that are expanded by DO_ANY3, and that
7173 * we modify a->s via that parameter before it is used by OP.
7175 DO_ANY3(SUB, a->s ? gen_sub_CC : tcg_gen_sub_i32, false,
7177 StoreRegKind ret = STREG_NORMAL;
7178 if (a->rd == 15 && a->s) {
7180 * See ALUExceptionReturn:
7181 * In User mode, UNPREDICTABLE; we choose UNDEF.
7182 * In Hyp mode, UNDEFINED.
7184 if (IS_USER(s) || s->current_el == 2) {
7185 unallocated_encoding(s);
7188 /* There is no writeback of nzcv to PSTATE. */
7190 ret = STREG_EXC_RET;
7191 } else if (a->rd == 13 && a->rn == 13) {
7192 ret = STREG_SP_CHECK;
7197 DO_ANY2(MOV, tcg_gen_mov_i32, a->s,
7199 StoreRegKind ret = STREG_NORMAL;
7200 if (a->rd == 15 && a->s) {
7202 * See ALUExceptionReturn:
7203 * In User mode, UNPREDICTABLE; we choose UNDEF.
7204 * In Hyp mode, UNDEFINED.
7206 if (IS_USER(s) || s->current_el == 2) {
7207 unallocated_encoding(s);
7210 /* There is no writeback of nzcv to PSTATE. */
7212 ret = STREG_EXC_RET;
7213 } else if (a->rd == 13) {
7214 ret = STREG_SP_CHECK;
7219 DO_ANY2(MVN, tcg_gen_not_i32, a->s, STREG_NORMAL)
7222 * ORN is only available with T32, so there is no register-shifted-register
7223 * form of the insn. Using the DO_ANY3 macro would create an unused function.
7225 static bool trans_ORN_rrri(DisasContext *s, arg_s_rrr_shi *a)
7227 return op_s_rrr_shi(s, a, tcg_gen_orc_i32, a->s, STREG_NORMAL);
7230 static bool trans_ORN_rri(DisasContext *s, arg_s_rri_rot *a)
7232 return op_s_rri_rot(s, a, tcg_gen_orc_i32, a->s, STREG_NORMAL);
7239 static bool trans_ADR(DisasContext *s, arg_ri *a)
7241 store_reg_bx(s, a->rd, add_reg_for_lit(s, 15, a->imm));
7245 static bool trans_MOVW(DisasContext *s, arg_MOVW *a)
7249 if (!ENABLE_ARCH_6T2) {
7253 tmp = tcg_const_i32(a->imm);
7254 store_reg(s, a->rd, tmp);
7258 static bool trans_MOVT(DisasContext *s, arg_MOVW *a)
7262 if (!ENABLE_ARCH_6T2) {
7266 tmp = load_reg(s, a->rd);
7267 tcg_gen_ext16u_i32(tmp, tmp);
7268 tcg_gen_ori_i32(tmp, tmp, a->imm << 16);
7269 store_reg(s, a->rd, tmp);
7274 * Multiply and multiply accumulate
7277 static bool op_mla(DisasContext *s, arg_s_rrrr *a, bool add)
7281 t1 = load_reg(s, a->rn);
7282 t2 = load_reg(s, a->rm);
7283 tcg_gen_mul_i32(t1, t1, t2);
7284 tcg_temp_free_i32(t2);
7286 t2 = load_reg(s, a->ra);
7287 tcg_gen_add_i32(t1, t1, t2);
7288 tcg_temp_free_i32(t2);
7293 store_reg(s, a->rd, t1);
7297 static bool trans_MUL(DisasContext *s, arg_MUL *a)
7299 return op_mla(s, a, false);
7302 static bool trans_MLA(DisasContext *s, arg_MLA *a)
7304 return op_mla(s, a, true);
7307 static bool trans_MLS(DisasContext *s, arg_MLS *a)
7311 if (!ENABLE_ARCH_6T2) {
7314 t1 = load_reg(s, a->rn);
7315 t2 = load_reg(s, a->rm);
7316 tcg_gen_mul_i32(t1, t1, t2);
7317 tcg_temp_free_i32(t2);
7318 t2 = load_reg(s, a->ra);
7319 tcg_gen_sub_i32(t1, t2, t1);
7320 tcg_temp_free_i32(t2);
7321 store_reg(s, a->rd, t1);
7325 static bool op_mlal(DisasContext *s, arg_s_rrrr *a, bool uns, bool add)
7327 TCGv_i32 t0, t1, t2, t3;
7329 t0 = load_reg(s, a->rm);
7330 t1 = load_reg(s, a->rn);
7332 tcg_gen_mulu2_i32(t0, t1, t0, t1);
7334 tcg_gen_muls2_i32(t0, t1, t0, t1);
7337 t2 = load_reg(s, a->ra);
7338 t3 = load_reg(s, a->rd);
7339 tcg_gen_add2_i32(t0, t1, t0, t1, t2, t3);
7340 tcg_temp_free_i32(t2);
7341 tcg_temp_free_i32(t3);
7344 gen_logicq_cc(t0, t1);
7346 store_reg(s, a->ra, t0);
7347 store_reg(s, a->rd, t1);
7351 static bool trans_UMULL(DisasContext *s, arg_UMULL *a)
7353 return op_mlal(s, a, true, false);
7356 static bool trans_SMULL(DisasContext *s, arg_SMULL *a)
7358 return op_mlal(s, a, false, false);
7361 static bool trans_UMLAL(DisasContext *s, arg_UMLAL *a)
7363 return op_mlal(s, a, true, true);
7366 static bool trans_SMLAL(DisasContext *s, arg_SMLAL *a)
7368 return op_mlal(s, a, false, true);
7371 static bool trans_UMAAL(DisasContext *s, arg_UMAAL *a)
7373 TCGv_i32 t0, t1, t2, zero;
7376 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
7381 t0 = load_reg(s, a->rm);
7382 t1 = load_reg(s, a->rn);
7383 tcg_gen_mulu2_i32(t0, t1, t0, t1);
7384 zero = tcg_const_i32(0);
7385 t2 = load_reg(s, a->ra);
7386 tcg_gen_add2_i32(t0, t1, t0, t1, t2, zero);
7387 tcg_temp_free_i32(t2);
7388 t2 = load_reg(s, a->rd);
7389 tcg_gen_add2_i32(t0, t1, t0, t1, t2, zero);
7390 tcg_temp_free_i32(t2);
7391 tcg_temp_free_i32(zero);
7392 store_reg(s, a->ra, t0);
7393 store_reg(s, a->rd, t1);
7398 * Saturating addition and subtraction
7401 static bool op_qaddsub(DisasContext *s, arg_rrr *a, bool add, bool doub)
7406 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
7407 : !ENABLE_ARCH_5TE) {
7411 t0 = load_reg(s, a->rm);
7412 t1 = load_reg(s, a->rn);
7414 gen_helper_add_saturate(t1, cpu_env, t1, t1);
7417 gen_helper_add_saturate(t0, cpu_env, t0, t1);
7419 gen_helper_sub_saturate(t0, cpu_env, t0, t1);
7421 tcg_temp_free_i32(t1);
7422 store_reg(s, a->rd, t0);
7426 #define DO_QADDSUB(NAME, ADD, DOUB) \
7427 static bool trans_##NAME(DisasContext *s, arg_rrr *a) \
7429 return op_qaddsub(s, a, ADD, DOUB); \
7432 DO_QADDSUB(QADD, true, false)
7433 DO_QADDSUB(QSUB, false, false)
7434 DO_QADDSUB(QDADD, true, true)
7435 DO_QADDSUB(QDSUB, false, true)
7440 * Halfword multiply and multiply accumulate
7443 static bool op_smlaxxx(DisasContext *s, arg_rrrr *a,
7444 int add_long, bool nt, bool mt)
7446 TCGv_i32 t0, t1, tl, th;
7449 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
7450 : !ENABLE_ARCH_5TE) {
7454 t0 = load_reg(s, a->rn);
7455 t1 = load_reg(s, a->rm);
7456 gen_mulxy(t0, t1, nt, mt);
7457 tcg_temp_free_i32(t1);
7461 store_reg(s, a->rd, t0);
7464 t1 = load_reg(s, a->ra);
7465 gen_helper_add_setq(t0, cpu_env, t0, t1);
7466 tcg_temp_free_i32(t1);
7467 store_reg(s, a->rd, t0);
7470 tl = load_reg(s, a->ra);
7471 th = load_reg(s, a->rd);
7472 /* Sign-extend the 32-bit product to 64 bits. */
7473 t1 = tcg_temp_new_i32();
7474 tcg_gen_sari_i32(t1, t0, 31);
7475 tcg_gen_add2_i32(tl, th, tl, th, t0, t1);
7476 tcg_temp_free_i32(t0);
7477 tcg_temp_free_i32(t1);
7478 store_reg(s, a->ra, tl);
7479 store_reg(s, a->rd, th);
7482 g_assert_not_reached();
7487 #define DO_SMLAX(NAME, add, nt, mt) \
7488 static bool trans_##NAME(DisasContext *s, arg_rrrr *a) \
7490 return op_smlaxxx(s, a, add, nt, mt); \
7493 DO_SMLAX(SMULBB, 0, 0, 0)
7494 DO_SMLAX(SMULBT, 0, 0, 1)
7495 DO_SMLAX(SMULTB, 0, 1, 0)
7496 DO_SMLAX(SMULTT, 0, 1, 1)
7498 DO_SMLAX(SMLABB, 1, 0, 0)
7499 DO_SMLAX(SMLABT, 1, 0, 1)
7500 DO_SMLAX(SMLATB, 1, 1, 0)
7501 DO_SMLAX(SMLATT, 1, 1, 1)
7503 DO_SMLAX(SMLALBB, 2, 0, 0)
7504 DO_SMLAX(SMLALBT, 2, 0, 1)
7505 DO_SMLAX(SMLALTB, 2, 1, 0)
7506 DO_SMLAX(SMLALTT, 2, 1, 1)
7510 static bool op_smlawx(DisasContext *s, arg_rrrr *a, bool add, bool mt)
7514 if (!ENABLE_ARCH_5TE) {
7518 t0 = load_reg(s, a->rn);
7519 t1 = load_reg(s, a->rm);
7521 * Since the nominal result is product<47:16>, shift the 16-bit
7522 * input up by 16 bits, so that the result is at product<63:32>.
7525 tcg_gen_andi_i32(t1, t1, 0xffff0000);
7527 tcg_gen_shli_i32(t1, t1, 16);
7529 tcg_gen_muls2_i32(t0, t1, t0, t1);
7530 tcg_temp_free_i32(t0);
7532 t0 = load_reg(s, a->ra);
7533 gen_helper_add_setq(t1, cpu_env, t1, t0);
7534 tcg_temp_free_i32(t0);
7536 store_reg(s, a->rd, t1);
7540 #define DO_SMLAWX(NAME, add, mt) \
7541 static bool trans_##NAME(DisasContext *s, arg_rrrr *a) \
7543 return op_smlawx(s, a, add, mt); \
7546 DO_SMLAWX(SMULWB, 0, 0)
7547 DO_SMLAWX(SMULWT, 0, 1)
7548 DO_SMLAWX(SMLAWB, 1, 0)
7549 DO_SMLAWX(SMLAWT, 1, 1)
7554 * MSR (immediate) and hints
7557 static bool trans_YIELD(DisasContext *s, arg_YIELD *a)
7560 * When running single-threaded TCG code, use the helper to ensure that
7561 * the next round-robin scheduled vCPU gets a crack. When running in
7562 * MTTCG we don't generate jumps to the helper as it won't affect the
7563 * scheduling of other vCPUs.
7565 if (!(tb_cflags(s->base.tb) & CF_PARALLEL)) {
7566 gen_set_pc_im(s, s->base.pc_next);
7567 s->base.is_jmp = DISAS_YIELD;
7572 static bool trans_WFE(DisasContext *s, arg_WFE *a)
7575 * When running single-threaded TCG code, use the helper to ensure that
7576 * the next round-robin scheduled vCPU gets a crack. In MTTCG mode we
7577 * just skip this instruction. Currently the SEV/SEVL instructions,
7578 * which are *one* of many ways to wake the CPU from WFE, are not
7579 * implemented so we can't sleep like WFI does.
7581 if (!(tb_cflags(s->base.tb) & CF_PARALLEL)) {
7582 gen_set_pc_im(s, s->base.pc_next);
7583 s->base.is_jmp = DISAS_WFE;
7588 static bool trans_WFI(DisasContext *s, arg_WFI *a)
7590 /* For WFI, halt the vCPU until an IRQ. */
7591 gen_set_pc_im(s, s->base.pc_next);
7592 s->base.is_jmp = DISAS_WFI;
7596 static bool trans_NOP(DisasContext *s, arg_NOP *a)
7601 static bool trans_MSR_imm(DisasContext *s, arg_MSR_imm *a)
7603 uint32_t val = ror32(a->imm, a->rot * 2);
7604 uint32_t mask = msr_mask(s, a->mask, a->r);
7606 if (gen_set_psr_im(s, mask, a->r, val)) {
7607 unallocated_encoding(s);
7613 * Cyclic Redundancy Check
7616 static bool op_crc32(DisasContext *s, arg_rrr *a, bool c, MemOp sz)
7618 TCGv_i32 t1, t2, t3;
7620 if (!dc_isar_feature(aa32_crc32, s)) {
7624 t1 = load_reg(s, a->rn);
7625 t2 = load_reg(s, a->rm);
7636 g_assert_not_reached();
7638 t3 = tcg_const_i32(1 << sz);
7640 gen_helper_crc32c(t1, t1, t2, t3);
7642 gen_helper_crc32(t1, t1, t2, t3);
7644 tcg_temp_free_i32(t2);
7645 tcg_temp_free_i32(t3);
7646 store_reg(s, a->rd, t1);
7650 #define DO_CRC32(NAME, c, sz) \
7651 static bool trans_##NAME(DisasContext *s, arg_rrr *a) \
7652 { return op_crc32(s, a, c, sz); }
7654 DO_CRC32(CRC32B, false, MO_8)
7655 DO_CRC32(CRC32H, false, MO_16)
7656 DO_CRC32(CRC32W, false, MO_32)
7657 DO_CRC32(CRC32CB, true, MO_8)
7658 DO_CRC32(CRC32CH, true, MO_16)
7659 DO_CRC32(CRC32CW, true, MO_32)
7664 * Miscellaneous instructions
7667 static bool trans_MRS_bank(DisasContext *s, arg_MRS_bank *a)
7669 if (arm_dc_feature(s, ARM_FEATURE_M)) {
7672 gen_mrs_banked(s, a->r, a->sysm, a->rd);
7676 static bool trans_MSR_bank(DisasContext *s, arg_MSR_bank *a)
7678 if (arm_dc_feature(s, ARM_FEATURE_M)) {
7681 gen_msr_banked(s, a->r, a->sysm, a->rn);
7685 static bool trans_MRS_reg(DisasContext *s, arg_MRS_reg *a)
7689 if (arm_dc_feature(s, ARM_FEATURE_M)) {
7694 unallocated_encoding(s);
7697 tmp = load_cpu_field(spsr);
7699 tmp = tcg_temp_new_i32();
7700 gen_helper_cpsr_read(tmp, cpu_env);
7702 store_reg(s, a->rd, tmp);
7706 static bool trans_MSR_reg(DisasContext *s, arg_MSR_reg *a)
7709 uint32_t mask = msr_mask(s, a->mask, a->r);
7711 if (arm_dc_feature(s, ARM_FEATURE_M)) {
7714 tmp = load_reg(s, a->rn);
7715 if (gen_set_psr(s, mask, a->r, tmp)) {
7716 unallocated_encoding(s);
7721 static bool trans_MRS_v7m(DisasContext *s, arg_MRS_v7m *a)
7725 if (!arm_dc_feature(s, ARM_FEATURE_M)) {
7728 tmp = tcg_const_i32(a->sysm);
7729 gen_helper_v7m_mrs(tmp, cpu_env, tmp);
7730 store_reg(s, a->rd, tmp);
7734 static bool trans_MSR_v7m(DisasContext *s, arg_MSR_v7m *a)
7738 if (!arm_dc_feature(s, ARM_FEATURE_M)) {
7741 addr = tcg_const_i32((a->mask << 10) | a->sysm);
7742 reg = load_reg(s, a->rn);
7743 gen_helper_v7m_msr(cpu_env, addr, reg);
7744 tcg_temp_free_i32(addr);
7745 tcg_temp_free_i32(reg);
7746 /* If we wrote to CONTROL, the EL might have changed */
7747 gen_helper_rebuild_hflags_m32_newel(cpu_env);
7752 static bool trans_BX(DisasContext *s, arg_BX *a)
7754 if (!ENABLE_ARCH_4T) {
7757 gen_bx_excret(s, load_reg(s, a->rm));
7761 static bool trans_BXJ(DisasContext *s, arg_BXJ *a)
7763 if (!ENABLE_ARCH_5J || arm_dc_feature(s, ARM_FEATURE_M)) {
7766 /* Trivial implementation equivalent to bx. */
7767 gen_bx(s, load_reg(s, a->rm));
7771 static bool trans_BLX_r(DisasContext *s, arg_BLX_r *a)
7775 if (!ENABLE_ARCH_5) {
7778 tmp = load_reg(s, a->rm);
7779 tcg_gen_movi_i32(cpu_R[14], s->base.pc_next | s->thumb);
7785 * BXNS/BLXNS: only exist for v8M with the security extensions,
7786 * and always UNDEF if NonSecure. We don't implement these in
7787 * the user-only mode either (in theory you can use them from
7788 * Secure User mode but they are too tied in to system emulation).
7790 static bool trans_BXNS(DisasContext *s, arg_BXNS *a)
7792 if (!s->v8m_secure || IS_USER_ONLY) {
7793 unallocated_encoding(s);
7800 static bool trans_BLXNS(DisasContext *s, arg_BLXNS *a)
7802 if (!s->v8m_secure || IS_USER_ONLY) {
7803 unallocated_encoding(s);
7805 gen_blxns(s, a->rm);
7810 static bool trans_CLZ(DisasContext *s, arg_CLZ *a)
7814 if (!ENABLE_ARCH_5) {
7817 tmp = load_reg(s, a->rm);
7818 tcg_gen_clzi_i32(tmp, tmp, 32);
7819 store_reg(s, a->rd, tmp);
7823 static bool trans_ERET(DisasContext *s, arg_ERET *a)
7827 if (!arm_dc_feature(s, ARM_FEATURE_V7VE)) {
7831 unallocated_encoding(s);
7834 if (s->current_el == 2) {
7835 /* ERET from Hyp uses ELR_Hyp, not LR */
7836 tmp = load_cpu_field(elr_el[2]);
7838 tmp = load_reg(s, 14);
7840 gen_exception_return(s, tmp);
7844 static bool trans_HLT(DisasContext *s, arg_HLT *a)
7850 static bool trans_BKPT(DisasContext *s, arg_BKPT *a)
7852 if (!ENABLE_ARCH_5) {
7855 if (arm_dc_feature(s, ARM_FEATURE_M) &&
7856 semihosting_enabled() &&
7857 #ifndef CONFIG_USER_ONLY
7861 gen_exception_internal_insn(s, s->pc_curr, EXCP_SEMIHOST);
7863 gen_exception_bkpt_insn(s, syn_aa32_bkpt(a->imm, false));
7868 static bool trans_HVC(DisasContext *s, arg_HVC *a)
7870 if (!ENABLE_ARCH_7 || arm_dc_feature(s, ARM_FEATURE_M)) {
7874 unallocated_encoding(s);
7881 static bool trans_SMC(DisasContext *s, arg_SMC *a)
7883 if (!ENABLE_ARCH_6K || arm_dc_feature(s, ARM_FEATURE_M)) {
7887 unallocated_encoding(s);
7894 static bool trans_SG(DisasContext *s, arg_SG *a)
7896 if (!arm_dc_feature(s, ARM_FEATURE_M) ||
7897 !arm_dc_feature(s, ARM_FEATURE_V8)) {
7902 * The bulk of the behaviour for this instruction is implemented
7903 * in v7m_handle_execute_nsc(), which deals with the insn when
7904 * it is executed by a CPU in non-secure state from memory
7905 * which is Secure & NonSecure-Callable.
7906 * Here we only need to handle the remaining cases:
7907 * * in NS memory (including the "security extension not
7908 * implemented" case) : NOP
7909 * * in S memory but CPU already secure (clear IT bits)
7910 * We know that the attribute for the memory this insn is
7911 * in must match the current CPU state, because otherwise
7912 * get_phys_addr_pmsav8 would have generated an exception.
7914 if (s->v8m_secure) {
7915 /* Like the IT insn, we don't need to generate any code */
7916 s->condexec_cond = 0;
7917 s->condexec_mask = 0;
7922 static bool trans_TT(DisasContext *s, arg_TT *a)
7926 if (!arm_dc_feature(s, ARM_FEATURE_M) ||
7927 !arm_dc_feature(s, ARM_FEATURE_V8)) {
7930 if (a->rd == 13 || a->rd == 15 || a->rn == 15) {
7931 /* We UNDEF for these UNPREDICTABLE cases */
7932 unallocated_encoding(s);
7935 if (a->A && !s->v8m_secure) {
7936 /* This case is UNDEFINED. */
7937 unallocated_encoding(s);
7941 addr = load_reg(s, a->rn);
7942 tmp = tcg_const_i32((a->A << 1) | a->T);
7943 gen_helper_v7m_tt(tmp, cpu_env, addr, tmp);
7944 tcg_temp_free_i32(addr);
7945 store_reg(s, a->rd, tmp);
7950 * Load/store register index
7953 static ISSInfo make_issinfo(DisasContext *s, int rd, bool p, bool w)
7957 /* ISS not valid if writeback */
7960 if (s->base.pc_next - s->pc_curr == 2) {
7969 static TCGv_i32 op_addr_rr_pre(DisasContext *s, arg_ldst_rr *a)
7971 TCGv_i32 addr = load_reg(s, a->rn);
7973 if (s->v8m_stackcheck && a->rn == 13 && a->w) {
7974 gen_helper_v8m_stackcheck(cpu_env, addr);
7978 TCGv_i32 ofs = load_reg(s, a->rm);
7979 gen_arm_shift_im(ofs, a->shtype, a->shimm, 0);
7981 tcg_gen_add_i32(addr, addr, ofs);
7983 tcg_gen_sub_i32(addr, addr, ofs);
7985 tcg_temp_free_i32(ofs);
7990 static void op_addr_rr_post(DisasContext *s, arg_ldst_rr *a,
7991 TCGv_i32 addr, int address_offset)
7994 TCGv_i32 ofs = load_reg(s, a->rm);
7995 gen_arm_shift_im(ofs, a->shtype, a->shimm, 0);
7997 tcg_gen_add_i32(addr, addr, ofs);
7999 tcg_gen_sub_i32(addr, addr, ofs);
8001 tcg_temp_free_i32(ofs);
8003 tcg_temp_free_i32(addr);
8006 tcg_gen_addi_i32(addr, addr, address_offset);
8007 store_reg(s, a->rn, addr);
8010 static bool op_load_rr(DisasContext *s, arg_ldst_rr *a,
8011 MemOp mop, int mem_idx)
8013 ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w);
8016 addr = op_addr_rr_pre(s, a);
8018 tmp = tcg_temp_new_i32();
8019 gen_aa32_ld_i32(s, tmp, addr, mem_idx, mop | s->be_data);
8020 disas_set_da_iss(s, mop, issinfo);
8023 * Perform base writeback before the loaded value to
8024 * ensure correct behavior with overlapping index registers.
8026 op_addr_rr_post(s, a, addr, 0);
8027 store_reg_from_load(s, a->rt, tmp);
8031 static bool op_store_rr(DisasContext *s, arg_ldst_rr *a,
8032 MemOp mop, int mem_idx)
8034 ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w) | ISSIsWrite;
8037 addr = op_addr_rr_pre(s, a);
8039 tmp = load_reg(s, a->rt);
8040 gen_aa32_st_i32(s, tmp, addr, mem_idx, mop | s->be_data);
8041 disas_set_da_iss(s, mop, issinfo);
8042 tcg_temp_free_i32(tmp);
8044 op_addr_rr_post(s, a, addr, 0);
8048 static bool trans_LDRD_rr(DisasContext *s, arg_ldst_rr *a)
8050 int mem_idx = get_mem_index(s);
8053 if (!ENABLE_ARCH_5TE) {
8057 unallocated_encoding(s);
8060 addr = op_addr_rr_pre(s, a);
8062 tmp = tcg_temp_new_i32();
8063 gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8064 store_reg(s, a->rt, tmp);
8066 tcg_gen_addi_i32(addr, addr, 4);
8068 tmp = tcg_temp_new_i32();
8069 gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8070 store_reg(s, a->rt + 1, tmp);
8072 /* LDRD w/ base writeback is undefined if the registers overlap. */
8073 op_addr_rr_post(s, a, addr, -4);
8077 static bool trans_STRD_rr(DisasContext *s, arg_ldst_rr *a)
8079 int mem_idx = get_mem_index(s);
8082 if (!ENABLE_ARCH_5TE) {
8086 unallocated_encoding(s);
8089 addr = op_addr_rr_pre(s, a);
8091 tmp = load_reg(s, a->rt);
8092 gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8093 tcg_temp_free_i32(tmp);
8095 tcg_gen_addi_i32(addr, addr, 4);
8097 tmp = load_reg(s, a->rt + 1);
8098 gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8099 tcg_temp_free_i32(tmp);
8101 op_addr_rr_post(s, a, addr, -4);
8106 * Load/store immediate index
8109 static TCGv_i32 op_addr_ri_pre(DisasContext *s, arg_ldst_ri *a)
8117 if (s->v8m_stackcheck && a->rn == 13 && a->w) {
8119 * Stackcheck. Here we know 'addr' is the current SP;
8120 * U is set if we're moving SP up, else down. It is
8121 * UNKNOWN whether the limit check triggers when SP starts
8122 * below the limit and ends up above it; we chose to do so.
8125 TCGv_i32 newsp = tcg_temp_new_i32();
8126 tcg_gen_addi_i32(newsp, cpu_R[13], ofs);
8127 gen_helper_v8m_stackcheck(cpu_env, newsp);
8128 tcg_temp_free_i32(newsp);
8130 gen_helper_v8m_stackcheck(cpu_env, cpu_R[13]);
8134 return add_reg_for_lit(s, a->rn, a->p ? ofs : 0);
8137 static void op_addr_ri_post(DisasContext *s, arg_ldst_ri *a,
8138 TCGv_i32 addr, int address_offset)
8142 address_offset += a->imm;
8144 address_offset -= a->imm;
8147 tcg_temp_free_i32(addr);
8150 tcg_gen_addi_i32(addr, addr, address_offset);
8151 store_reg(s, a->rn, addr);
8154 static bool op_load_ri(DisasContext *s, arg_ldst_ri *a,
8155 MemOp mop, int mem_idx)
8157 ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w);
8160 addr = op_addr_ri_pre(s, a);
8162 tmp = tcg_temp_new_i32();
8163 gen_aa32_ld_i32(s, tmp, addr, mem_idx, mop | s->be_data);
8164 disas_set_da_iss(s, mop, issinfo);
8167 * Perform base writeback before the loaded value to
8168 * ensure correct behavior with overlapping index registers.
8170 op_addr_ri_post(s, a, addr, 0);
8171 store_reg_from_load(s, a->rt, tmp);
8175 static bool op_store_ri(DisasContext *s, arg_ldst_ri *a,
8176 MemOp mop, int mem_idx)
8178 ISSInfo issinfo = make_issinfo(s, a->rt, a->p, a->w) | ISSIsWrite;
8181 addr = op_addr_ri_pre(s, a);
8183 tmp = load_reg(s, a->rt);
8184 gen_aa32_st_i32(s, tmp, addr, mem_idx, mop | s->be_data);
8185 disas_set_da_iss(s, mop, issinfo);
8186 tcg_temp_free_i32(tmp);
8188 op_addr_ri_post(s, a, addr, 0);
8192 static bool op_ldrd_ri(DisasContext *s, arg_ldst_ri *a, int rt2)
8194 int mem_idx = get_mem_index(s);
8197 addr = op_addr_ri_pre(s, a);
8199 tmp = tcg_temp_new_i32();
8200 gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8201 store_reg(s, a->rt, tmp);
8203 tcg_gen_addi_i32(addr, addr, 4);
8205 tmp = tcg_temp_new_i32();
8206 gen_aa32_ld_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8207 store_reg(s, rt2, tmp);
8209 /* LDRD w/ base writeback is undefined if the registers overlap. */
8210 op_addr_ri_post(s, a, addr, -4);
8214 static bool trans_LDRD_ri_a32(DisasContext *s, arg_ldst_ri *a)
8216 if (!ENABLE_ARCH_5TE || (a->rt & 1)) {
8219 return op_ldrd_ri(s, a, a->rt + 1);
8222 static bool trans_LDRD_ri_t32(DisasContext *s, arg_ldst_ri2 *a)
8225 .u = a->u, .w = a->w, .p = a->p,
8226 .rn = a->rn, .rt = a->rt, .imm = a->imm
8228 return op_ldrd_ri(s, &b, a->rt2);
8231 static bool op_strd_ri(DisasContext *s, arg_ldst_ri *a, int rt2)
8233 int mem_idx = get_mem_index(s);
8236 addr = op_addr_ri_pre(s, a);
8238 tmp = load_reg(s, a->rt);
8239 gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8240 tcg_temp_free_i32(tmp);
8242 tcg_gen_addi_i32(addr, addr, 4);
8244 tmp = load_reg(s, rt2);
8245 gen_aa32_st_i32(s, tmp, addr, mem_idx, MO_UL | s->be_data);
8246 tcg_temp_free_i32(tmp);
8248 op_addr_ri_post(s, a, addr, -4);
8252 static bool trans_STRD_ri_a32(DisasContext *s, arg_ldst_ri *a)
8254 if (!ENABLE_ARCH_5TE || (a->rt & 1)) {
8257 return op_strd_ri(s, a, a->rt + 1);
8260 static bool trans_STRD_ri_t32(DisasContext *s, arg_ldst_ri2 *a)
8263 .u = a->u, .w = a->w, .p = a->p,
8264 .rn = a->rn, .rt = a->rt, .imm = a->imm
8266 return op_strd_ri(s, &b, a->rt2);
8269 #define DO_LDST(NAME, WHICH, MEMOP) \
8270 static bool trans_##NAME##_ri(DisasContext *s, arg_ldst_ri *a) \
8272 return op_##WHICH##_ri(s, a, MEMOP, get_mem_index(s)); \
8274 static bool trans_##NAME##T_ri(DisasContext *s, arg_ldst_ri *a) \
8276 return op_##WHICH##_ri(s, a, MEMOP, get_a32_user_mem_index(s)); \
8278 static bool trans_##NAME##_rr(DisasContext *s, arg_ldst_rr *a) \
8280 return op_##WHICH##_rr(s, a, MEMOP, get_mem_index(s)); \
8282 static bool trans_##NAME##T_rr(DisasContext *s, arg_ldst_rr *a) \
8284 return op_##WHICH##_rr(s, a, MEMOP, get_a32_user_mem_index(s)); \
8287 DO_LDST(LDR, load, MO_UL)
8288 DO_LDST(LDRB, load, MO_UB)
8289 DO_LDST(LDRH, load, MO_UW)
8290 DO_LDST(LDRSB, load, MO_SB)
8291 DO_LDST(LDRSH, load, MO_SW)
8293 DO_LDST(STR, store, MO_UL)
8294 DO_LDST(STRB, store, MO_UB)
8295 DO_LDST(STRH, store, MO_UW)
8300 * Synchronization primitives
8303 static bool op_swp(DisasContext *s, arg_SWP *a, MemOp opc)
8309 addr = load_reg(s, a->rn);
8310 taddr = gen_aa32_addr(s, addr, opc);
8311 tcg_temp_free_i32(addr);
8313 tmp = load_reg(s, a->rt2);
8314 tcg_gen_atomic_xchg_i32(tmp, taddr, tmp, get_mem_index(s), opc);
8315 tcg_temp_free(taddr);
8317 store_reg(s, a->rt, tmp);
8321 static bool trans_SWP(DisasContext *s, arg_SWP *a)
8323 return op_swp(s, a, MO_UL | MO_ALIGN);
8326 static bool trans_SWPB(DisasContext *s, arg_SWP *a)
8328 return op_swp(s, a, MO_UB);
8332 * Load/Store Exclusive and Load-Acquire/Store-Release
8335 static bool op_strex(DisasContext *s, arg_STREX *a, MemOp mop, bool rel)
8338 /* Some cases stopped being UNPREDICTABLE in v8A (but not v8M) */
8339 bool v8a = ENABLE_ARCH_8 && !arm_dc_feature(s, ARM_FEATURE_M);
8341 /* We UNDEF for these UNPREDICTABLE cases. */
8342 if (a->rd == 15 || a->rn == 15 || a->rt == 15
8343 || a->rd == a->rn || a->rd == a->rt
8344 || (!v8a && s->thumb && (a->rd == 13 || a->rt == 13))
8348 || (!v8a && s->thumb && a->rt2 == 13)))) {
8349 unallocated_encoding(s);
8354 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_STRL);
8357 addr = tcg_temp_local_new_i32();
8358 load_reg_var(s, addr, a->rn);
8359 tcg_gen_addi_i32(addr, addr, a->imm);
8361 gen_store_exclusive(s, a->rd, a->rt, a->rt2, addr, mop);
8362 tcg_temp_free_i32(addr);
8366 static bool trans_STREX(DisasContext *s, arg_STREX *a)
8368 if (!ENABLE_ARCH_6) {
8371 return op_strex(s, a, MO_32, false);
8374 static bool trans_STREXD_a32(DisasContext *s, arg_STREX *a)
8376 if (!ENABLE_ARCH_6K) {
8379 /* We UNDEF for these UNPREDICTABLE cases. */
8381 unallocated_encoding(s);
8385 return op_strex(s, a, MO_64, false);
8388 static bool trans_STREXD_t32(DisasContext *s, arg_STREX *a)
8390 return op_strex(s, a, MO_64, false);
8393 static bool trans_STREXB(DisasContext *s, arg_STREX *a)
8395 if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
8398 return op_strex(s, a, MO_8, false);
8401 static bool trans_STREXH(DisasContext *s, arg_STREX *a)
8403 if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
8406 return op_strex(s, a, MO_16, false);
8409 static bool trans_STLEX(DisasContext *s, arg_STREX *a)
8411 if (!ENABLE_ARCH_8) {
8414 return op_strex(s, a, MO_32, true);
8417 static bool trans_STLEXD_a32(DisasContext *s, arg_STREX *a)
8419 if (!ENABLE_ARCH_8) {
8422 /* We UNDEF for these UNPREDICTABLE cases. */
8424 unallocated_encoding(s);
8428 return op_strex(s, a, MO_64, true);
8431 static bool trans_STLEXD_t32(DisasContext *s, arg_STREX *a)
8433 if (!ENABLE_ARCH_8) {
8436 return op_strex(s, a, MO_64, true);
8439 static bool trans_STLEXB(DisasContext *s, arg_STREX *a)
8441 if (!ENABLE_ARCH_8) {
8444 return op_strex(s, a, MO_8, true);
8447 static bool trans_STLEXH(DisasContext *s, arg_STREX *a)
8449 if (!ENABLE_ARCH_8) {
8452 return op_strex(s, a, MO_16, true);
8455 static bool op_stl(DisasContext *s, arg_STL *a, MemOp mop)
8459 if (!ENABLE_ARCH_8) {
8462 /* We UNDEF for these UNPREDICTABLE cases. */
8463 if (a->rn == 15 || a->rt == 15) {
8464 unallocated_encoding(s);
8468 addr = load_reg(s, a->rn);
8469 tmp = load_reg(s, a->rt);
8470 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_STRL);
8471 gen_aa32_st_i32(s, tmp, addr, get_mem_index(s), mop | s->be_data);
8472 disas_set_da_iss(s, mop, a->rt | ISSIsAcqRel | ISSIsWrite);
8474 tcg_temp_free_i32(tmp);
8475 tcg_temp_free_i32(addr);
8479 static bool trans_STL(DisasContext *s, arg_STL *a)
8481 return op_stl(s, a, MO_UL);
8484 static bool trans_STLB(DisasContext *s, arg_STL *a)
8486 return op_stl(s, a, MO_UB);
8489 static bool trans_STLH(DisasContext *s, arg_STL *a)
8491 return op_stl(s, a, MO_UW);
8494 static bool op_ldrex(DisasContext *s, arg_LDREX *a, MemOp mop, bool acq)
8497 /* Some cases stopped being UNPREDICTABLE in v8A (but not v8M) */
8498 bool v8a = ENABLE_ARCH_8 && !arm_dc_feature(s, ARM_FEATURE_M);
8500 /* We UNDEF for these UNPREDICTABLE cases. */
8501 if (a->rn == 15 || a->rt == 15
8502 || (!v8a && s->thumb && a->rt == 13)
8504 && (a->rt2 == 15 || a->rt == a->rt2
8505 || (!v8a && s->thumb && a->rt2 == 13)))) {
8506 unallocated_encoding(s);
8510 addr = tcg_temp_local_new_i32();
8511 load_reg_var(s, addr, a->rn);
8512 tcg_gen_addi_i32(addr, addr, a->imm);
8514 gen_load_exclusive(s, a->rt, a->rt2, addr, mop);
8515 tcg_temp_free_i32(addr);
8518 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_LDAQ);
8523 static bool trans_LDREX(DisasContext *s, arg_LDREX *a)
8525 if (!ENABLE_ARCH_6) {
8528 return op_ldrex(s, a, MO_32, false);
8531 static bool trans_LDREXD_a32(DisasContext *s, arg_LDREX *a)
8533 if (!ENABLE_ARCH_6K) {
8536 /* We UNDEF for these UNPREDICTABLE cases. */
8538 unallocated_encoding(s);
8542 return op_ldrex(s, a, MO_64, false);
8545 static bool trans_LDREXD_t32(DisasContext *s, arg_LDREX *a)
8547 return op_ldrex(s, a, MO_64, false);
8550 static bool trans_LDREXB(DisasContext *s, arg_LDREX *a)
8552 if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
8555 return op_ldrex(s, a, MO_8, false);
8558 static bool trans_LDREXH(DisasContext *s, arg_LDREX *a)
8560 if (s->thumb ? !ENABLE_ARCH_7 : !ENABLE_ARCH_6K) {
8563 return op_ldrex(s, a, MO_16, false);
8566 static bool trans_LDAEX(DisasContext *s, arg_LDREX *a)
8568 if (!ENABLE_ARCH_8) {
8571 return op_ldrex(s, a, MO_32, true);
8574 static bool trans_LDAEXD_a32(DisasContext *s, arg_LDREX *a)
8576 if (!ENABLE_ARCH_8) {
8579 /* We UNDEF for these UNPREDICTABLE cases. */
8581 unallocated_encoding(s);
8585 return op_ldrex(s, a, MO_64, true);
8588 static bool trans_LDAEXD_t32(DisasContext *s, arg_LDREX *a)
8590 if (!ENABLE_ARCH_8) {
8593 return op_ldrex(s, a, MO_64, true);
8596 static bool trans_LDAEXB(DisasContext *s, arg_LDREX *a)
8598 if (!ENABLE_ARCH_8) {
8601 return op_ldrex(s, a, MO_8, true);
8604 static bool trans_LDAEXH(DisasContext *s, arg_LDREX *a)
8606 if (!ENABLE_ARCH_8) {
8609 return op_ldrex(s, a, MO_16, true);
8612 static bool op_lda(DisasContext *s, arg_LDA *a, MemOp mop)
8616 if (!ENABLE_ARCH_8) {
8619 /* We UNDEF for these UNPREDICTABLE cases. */
8620 if (a->rn == 15 || a->rt == 15) {
8621 unallocated_encoding(s);
8625 addr = load_reg(s, a->rn);
8626 tmp = tcg_temp_new_i32();
8627 gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s), mop | s->be_data);
8628 disas_set_da_iss(s, mop, a->rt | ISSIsAcqRel);
8629 tcg_temp_free_i32(addr);
8631 store_reg(s, a->rt, tmp);
8632 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_STRL);
8636 static bool trans_LDA(DisasContext *s, arg_LDA *a)
8638 return op_lda(s, a, MO_UL);
8641 static bool trans_LDAB(DisasContext *s, arg_LDA *a)
8643 return op_lda(s, a, MO_UB);
8646 static bool trans_LDAH(DisasContext *s, arg_LDA *a)
8648 return op_lda(s, a, MO_UW);
8652 * Media instructions
8655 static bool trans_USADA8(DisasContext *s, arg_USADA8 *a)
8659 if (!ENABLE_ARCH_6) {
8663 t1 = load_reg(s, a->rn);
8664 t2 = load_reg(s, a->rm);
8665 gen_helper_usad8(t1, t1, t2);
8666 tcg_temp_free_i32(t2);
8668 t2 = load_reg(s, a->ra);
8669 tcg_gen_add_i32(t1, t1, t2);
8670 tcg_temp_free_i32(t2);
8672 store_reg(s, a->rd, t1);
8676 static bool op_bfx(DisasContext *s, arg_UBFX *a, bool u)
8679 int width = a->widthm1 + 1;
8682 if (!ENABLE_ARCH_6T2) {
8685 if (shift + width > 32) {
8686 /* UNPREDICTABLE; we choose to UNDEF */
8687 unallocated_encoding(s);
8691 tmp = load_reg(s, a->rn);
8693 tcg_gen_extract_i32(tmp, tmp, shift, width);
8695 tcg_gen_sextract_i32(tmp, tmp, shift, width);
8697 store_reg(s, a->rd, tmp);
8701 static bool trans_SBFX(DisasContext *s, arg_SBFX *a)
8703 return op_bfx(s, a, false);
8706 static bool trans_UBFX(DisasContext *s, arg_UBFX *a)
8708 return op_bfx(s, a, true);
8711 static bool trans_BFCI(DisasContext *s, arg_BFCI *a)
8714 int msb = a->msb, lsb = a->lsb;
8717 if (!ENABLE_ARCH_6T2) {
8721 /* UNPREDICTABLE; we choose to UNDEF */
8722 unallocated_encoding(s);
8726 width = msb + 1 - lsb;
8729 tmp = tcg_const_i32(0);
8732 tmp = load_reg(s, a->rn);
8735 TCGv_i32 tmp2 = load_reg(s, a->rd);
8736 tcg_gen_deposit_i32(tmp, tmp2, tmp, lsb, width);
8737 tcg_temp_free_i32(tmp2);
8739 store_reg(s, a->rd, tmp);
8743 static bool trans_UDF(DisasContext *s, arg_UDF *a)
8745 unallocated_encoding(s);
8750 * Parallel addition and subtraction
8753 static bool op_par_addsub(DisasContext *s, arg_rrr *a,
8754 void (*gen)(TCGv_i32, TCGv_i32, TCGv_i32))
8759 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
8764 t0 = load_reg(s, a->rn);
8765 t1 = load_reg(s, a->rm);
8769 tcg_temp_free_i32(t1);
8770 store_reg(s, a->rd, t0);
8774 static bool op_par_addsub_ge(DisasContext *s, arg_rrr *a,
8775 void (*gen)(TCGv_i32, TCGv_i32,
8776 TCGv_i32, TCGv_ptr))
8782 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
8787 t0 = load_reg(s, a->rn);
8788 t1 = load_reg(s, a->rm);
8790 ge = tcg_temp_new_ptr();
8791 tcg_gen_addi_ptr(ge, cpu_env, offsetof(CPUARMState, GE));
8792 gen(t0, t0, t1, ge);
8794 tcg_temp_free_ptr(ge);
8795 tcg_temp_free_i32(t1);
8796 store_reg(s, a->rd, t0);
8800 #define DO_PAR_ADDSUB(NAME, helper) \
8801 static bool trans_##NAME(DisasContext *s, arg_rrr *a) \
8803 return op_par_addsub(s, a, helper); \
8806 #define DO_PAR_ADDSUB_GE(NAME, helper) \
8807 static bool trans_##NAME(DisasContext *s, arg_rrr *a) \
8809 return op_par_addsub_ge(s, a, helper); \
8812 DO_PAR_ADDSUB_GE(SADD16, gen_helper_sadd16)
8813 DO_PAR_ADDSUB_GE(SASX, gen_helper_saddsubx)
8814 DO_PAR_ADDSUB_GE(SSAX, gen_helper_ssubaddx)
8815 DO_PAR_ADDSUB_GE(SSUB16, gen_helper_ssub16)
8816 DO_PAR_ADDSUB_GE(SADD8, gen_helper_sadd8)
8817 DO_PAR_ADDSUB_GE(SSUB8, gen_helper_ssub8)
8819 DO_PAR_ADDSUB_GE(UADD16, gen_helper_uadd16)
8820 DO_PAR_ADDSUB_GE(UASX, gen_helper_uaddsubx)
8821 DO_PAR_ADDSUB_GE(USAX, gen_helper_usubaddx)
8822 DO_PAR_ADDSUB_GE(USUB16, gen_helper_usub16)
8823 DO_PAR_ADDSUB_GE(UADD8, gen_helper_uadd8)
8824 DO_PAR_ADDSUB_GE(USUB8, gen_helper_usub8)
8826 DO_PAR_ADDSUB(QADD16, gen_helper_qadd16)
8827 DO_PAR_ADDSUB(QASX, gen_helper_qaddsubx)
8828 DO_PAR_ADDSUB(QSAX, gen_helper_qsubaddx)
8829 DO_PAR_ADDSUB(QSUB16, gen_helper_qsub16)
8830 DO_PAR_ADDSUB(QADD8, gen_helper_qadd8)
8831 DO_PAR_ADDSUB(QSUB8, gen_helper_qsub8)
8833 DO_PAR_ADDSUB(UQADD16, gen_helper_uqadd16)
8834 DO_PAR_ADDSUB(UQASX, gen_helper_uqaddsubx)
8835 DO_PAR_ADDSUB(UQSAX, gen_helper_uqsubaddx)
8836 DO_PAR_ADDSUB(UQSUB16, gen_helper_uqsub16)
8837 DO_PAR_ADDSUB(UQADD8, gen_helper_uqadd8)
8838 DO_PAR_ADDSUB(UQSUB8, gen_helper_uqsub8)
8840 DO_PAR_ADDSUB(SHADD16, gen_helper_shadd16)
8841 DO_PAR_ADDSUB(SHASX, gen_helper_shaddsubx)
8842 DO_PAR_ADDSUB(SHSAX, gen_helper_shsubaddx)
8843 DO_PAR_ADDSUB(SHSUB16, gen_helper_shsub16)
8844 DO_PAR_ADDSUB(SHADD8, gen_helper_shadd8)
8845 DO_PAR_ADDSUB(SHSUB8, gen_helper_shsub8)
8847 DO_PAR_ADDSUB(UHADD16, gen_helper_uhadd16)
8848 DO_PAR_ADDSUB(UHASX, gen_helper_uhaddsubx)
8849 DO_PAR_ADDSUB(UHSAX, gen_helper_uhsubaddx)
8850 DO_PAR_ADDSUB(UHSUB16, gen_helper_uhsub16)
8851 DO_PAR_ADDSUB(UHADD8, gen_helper_uhadd8)
8852 DO_PAR_ADDSUB(UHSUB8, gen_helper_uhsub8)
8854 #undef DO_PAR_ADDSUB
8855 #undef DO_PAR_ADDSUB_GE
8858 * Packing, unpacking, saturation, and reversal
8861 static bool trans_PKH(DisasContext *s, arg_PKH *a)
8867 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
8872 tn = load_reg(s, a->rn);
8873 tm = load_reg(s, a->rm);
8879 tcg_gen_sari_i32(tm, tm, shift);
8880 tcg_gen_deposit_i32(tn, tn, tm, 0, 16);
8883 tcg_gen_shli_i32(tm, tm, shift);
8884 tcg_gen_deposit_i32(tn, tm, tn, 0, 16);
8886 tcg_temp_free_i32(tm);
8887 store_reg(s, a->rd, tn);
8891 static bool op_sat(DisasContext *s, arg_sat *a,
8892 void (*gen)(TCGv_i32, TCGv_env, TCGv_i32, TCGv_i32))
8894 TCGv_i32 tmp, satimm;
8897 if (!ENABLE_ARCH_6) {
8901 tmp = load_reg(s, a->rn);
8903 tcg_gen_sari_i32(tmp, tmp, shift ? shift : 31);
8905 tcg_gen_shli_i32(tmp, tmp, shift);
8908 satimm = tcg_const_i32(a->satimm);
8909 gen(tmp, cpu_env, tmp, satimm);
8910 tcg_temp_free_i32(satimm);
8912 store_reg(s, a->rd, tmp);
8916 static bool trans_SSAT(DisasContext *s, arg_sat *a)
8918 return op_sat(s, a, gen_helper_ssat);
8921 static bool trans_USAT(DisasContext *s, arg_sat *a)
8923 return op_sat(s, a, gen_helper_usat);
8926 static bool trans_SSAT16(DisasContext *s, arg_sat *a)
8928 if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
8931 return op_sat(s, a, gen_helper_ssat16);
8934 static bool trans_USAT16(DisasContext *s, arg_sat *a)
8936 if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
8939 return op_sat(s, a, gen_helper_usat16);
8942 static bool op_xta(DisasContext *s, arg_rrr_rot *a,
8943 void (*gen_extract)(TCGv_i32, TCGv_i32),
8944 void (*gen_add)(TCGv_i32, TCGv_i32, TCGv_i32))
8948 if (!ENABLE_ARCH_6) {
8952 tmp = load_reg(s, a->rm);
8954 * TODO: In many cases we could do a shift instead of a rotate.
8955 * Combined with a simple extend, that becomes an extract.
8957 tcg_gen_rotri_i32(tmp, tmp, a->rot * 8);
8958 gen_extract(tmp, tmp);
8961 TCGv_i32 tmp2 = load_reg(s, a->rn);
8962 gen_add(tmp, tmp, tmp2);
8963 tcg_temp_free_i32(tmp2);
8965 store_reg(s, a->rd, tmp);
8969 static bool trans_SXTAB(DisasContext *s, arg_rrr_rot *a)
8971 return op_xta(s, a, tcg_gen_ext8s_i32, tcg_gen_add_i32);
8974 static bool trans_SXTAH(DisasContext *s, arg_rrr_rot *a)
8976 return op_xta(s, a, tcg_gen_ext16s_i32, tcg_gen_add_i32);
8979 static bool trans_SXTAB16(DisasContext *s, arg_rrr_rot *a)
8981 if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
8984 return op_xta(s, a, gen_helper_sxtb16, gen_add16);
8987 static bool trans_UXTAB(DisasContext *s, arg_rrr_rot *a)
8989 return op_xta(s, a, tcg_gen_ext8u_i32, tcg_gen_add_i32);
8992 static bool trans_UXTAH(DisasContext *s, arg_rrr_rot *a)
8994 return op_xta(s, a, tcg_gen_ext16u_i32, tcg_gen_add_i32);
8997 static bool trans_UXTAB16(DisasContext *s, arg_rrr_rot *a)
8999 if (s->thumb && !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
9002 return op_xta(s, a, gen_helper_uxtb16, gen_add16);
9005 static bool trans_SEL(DisasContext *s, arg_rrr *a)
9007 TCGv_i32 t1, t2, t3;
9010 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
9015 t1 = load_reg(s, a->rn);
9016 t2 = load_reg(s, a->rm);
9017 t3 = tcg_temp_new_i32();
9018 tcg_gen_ld_i32(t3, cpu_env, offsetof(CPUARMState, GE));
9019 gen_helper_sel_flags(t1, t3, t1, t2);
9020 tcg_temp_free_i32(t3);
9021 tcg_temp_free_i32(t2);
9022 store_reg(s, a->rd, t1);
9026 static bool op_rr(DisasContext *s, arg_rr *a,
9027 void (*gen)(TCGv_i32, TCGv_i32))
9031 tmp = load_reg(s, a->rm);
9033 store_reg(s, a->rd, tmp);
9037 static bool trans_REV(DisasContext *s, arg_rr *a)
9039 if (!ENABLE_ARCH_6) {
9042 return op_rr(s, a, tcg_gen_bswap32_i32);
9045 static bool trans_REV16(DisasContext *s, arg_rr *a)
9047 if (!ENABLE_ARCH_6) {
9050 return op_rr(s, a, gen_rev16);
9053 static bool trans_REVSH(DisasContext *s, arg_rr *a)
9055 if (!ENABLE_ARCH_6) {
9058 return op_rr(s, a, gen_revsh);
9061 static bool trans_RBIT(DisasContext *s, arg_rr *a)
9063 if (!ENABLE_ARCH_6T2) {
9066 return op_rr(s, a, gen_helper_rbit);
9070 * Signed multiply, signed and unsigned divide
9073 static bool op_smlad(DisasContext *s, arg_rrrr *a, bool m_swap, bool sub)
9077 if (!ENABLE_ARCH_6) {
9081 t1 = load_reg(s, a->rn);
9082 t2 = load_reg(s, a->rm);
9086 gen_smul_dual(t1, t2);
9089 /* This subtraction cannot overflow. */
9090 tcg_gen_sub_i32(t1, t1, t2);
9093 * This addition cannot overflow 32 bits; however it may
9094 * overflow considered as a signed operation, in which case
9095 * we must set the Q flag.
9097 gen_helper_add_setq(t1, cpu_env, t1, t2);
9099 tcg_temp_free_i32(t2);
9102 t2 = load_reg(s, a->ra);
9103 gen_helper_add_setq(t1, cpu_env, t1, t2);
9104 tcg_temp_free_i32(t2);
9106 store_reg(s, a->rd, t1);
9110 static bool trans_SMLAD(DisasContext *s, arg_rrrr *a)
9112 return op_smlad(s, a, false, false);
9115 static bool trans_SMLADX(DisasContext *s, arg_rrrr *a)
9117 return op_smlad(s, a, true, false);
9120 static bool trans_SMLSD(DisasContext *s, arg_rrrr *a)
9122 return op_smlad(s, a, false, true);
9125 static bool trans_SMLSDX(DisasContext *s, arg_rrrr *a)
9127 return op_smlad(s, a, true, true);
9130 static bool op_smlald(DisasContext *s, arg_rrrr *a, bool m_swap, bool sub)
9135 if (!ENABLE_ARCH_6) {
9139 t1 = load_reg(s, a->rn);
9140 t2 = load_reg(s, a->rm);
9144 gen_smul_dual(t1, t2);
9146 l1 = tcg_temp_new_i64();
9147 l2 = tcg_temp_new_i64();
9148 tcg_gen_ext_i32_i64(l1, t1);
9149 tcg_gen_ext_i32_i64(l2, t2);
9150 tcg_temp_free_i32(t1);
9151 tcg_temp_free_i32(t2);
9154 tcg_gen_sub_i64(l1, l1, l2);
9156 tcg_gen_add_i64(l1, l1, l2);
9158 tcg_temp_free_i64(l2);
9160 gen_addq(s, l1, a->ra, a->rd);
9161 gen_storeq_reg(s, a->ra, a->rd, l1);
9162 tcg_temp_free_i64(l1);
9166 static bool trans_SMLALD(DisasContext *s, arg_rrrr *a)
9168 return op_smlald(s, a, false, false);
9171 static bool trans_SMLALDX(DisasContext *s, arg_rrrr *a)
9173 return op_smlald(s, a, true, false);
9176 static bool trans_SMLSLD(DisasContext *s, arg_rrrr *a)
9178 return op_smlald(s, a, false, true);
9181 static bool trans_SMLSLDX(DisasContext *s, arg_rrrr *a)
9183 return op_smlald(s, a, true, true);
9186 static bool op_smmla(DisasContext *s, arg_rrrr *a, bool round, bool sub)
9191 ? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
9196 t1 = load_reg(s, a->rn);
9197 t2 = load_reg(s, a->rm);
9198 tcg_gen_muls2_i32(t2, t1, t1, t2);
9201 TCGv_i32 t3 = load_reg(s, a->ra);
9204 * For SMMLS, we need a 64-bit subtract. Borrow caused by
9205 * a non-zero multiplicand lowpart, and the correct result
9206 * lowpart for rounding.
9208 TCGv_i32 zero = tcg_const_i32(0);
9209 tcg_gen_sub2_i32(t2, t1, zero, t3, t2, t1);
9210 tcg_temp_free_i32(zero);
9212 tcg_gen_add_i32(t1, t1, t3);
9214 tcg_temp_free_i32(t3);
9218 * Adding 0x80000000 to the 64-bit quantity means that we have
9219 * carry in to the high word when the low word has the msb set.
9221 tcg_gen_shri_i32(t2, t2, 31);
9222 tcg_gen_add_i32(t1, t1, t2);
9224 tcg_temp_free_i32(t2);
9225 store_reg(s, a->rd, t1);
9229 static bool trans_SMMLA(DisasContext *s, arg_rrrr *a)
9231 return op_smmla(s, a, false, false);
9234 static bool trans_SMMLAR(DisasContext *s, arg_rrrr *a)
9236 return op_smmla(s, a, true, false);
9239 static bool trans_SMMLS(DisasContext *s, arg_rrrr *a)
9241 return op_smmla(s, a, false, true);
9244 static bool trans_SMMLSR(DisasContext *s, arg_rrrr *a)
9246 return op_smmla(s, a, true, true);
9249 static bool op_div(DisasContext *s, arg_rrr *a, bool u)
9254 ? !dc_isar_feature(aa32_thumb_div, s)
9255 : !dc_isar_feature(aa32_arm_div, s)) {
9259 t1 = load_reg(s, a->rn);
9260 t2 = load_reg(s, a->rm);
9262 gen_helper_udiv(t1, t1, t2);
9264 gen_helper_sdiv(t1, t1, t2);
9266 tcg_temp_free_i32(t2);
9267 store_reg(s, a->rd, t1);
9271 static bool trans_SDIV(DisasContext *s, arg_rrr *a)
9273 return op_div(s, a, false);
9276 static bool trans_UDIV(DisasContext *s, arg_rrr *a)
9278 return op_div(s, a, true);
9282 * Block data transfer
9285 static TCGv_i32 op_addr_block_pre(DisasContext *s, arg_ldst_block *a, int n)
9287 TCGv_i32 addr = load_reg(s, a->rn);
9292 tcg_gen_addi_i32(addr, addr, 4);
9295 tcg_gen_addi_i32(addr, addr, -(n * 4));
9297 } else if (!a->i && n != 1) {
9298 /* post decrement */
9299 tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
9302 if (s->v8m_stackcheck && a->rn == 13 && a->w) {
9304 * If the writeback is incrementing SP rather than
9305 * decrementing it, and the initial SP is below the
9306 * stack limit but the final written-back SP would
9307 * be above, then then we must not perform any memory
9308 * accesses, but it is IMPDEF whether we generate
9309 * an exception. We choose to do so in this case.
9310 * At this point 'addr' is the lowest address, so
9311 * either the original SP (if incrementing) or our
9312 * final SP (if decrementing), so that's what we check.
9314 gen_helper_v8m_stackcheck(cpu_env, addr);
9320 static void op_addr_block_post(DisasContext *s, arg_ldst_block *a,
9321 TCGv_i32 addr, int n)
9327 /* post increment */
9328 tcg_gen_addi_i32(addr, addr, 4);
9330 /* post decrement */
9331 tcg_gen_addi_i32(addr, addr, -(n * 4));
9333 } else if (!a->i && n != 1) {
9335 tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
9337 store_reg(s, a->rn, addr);
9339 tcg_temp_free_i32(addr);
9343 static bool op_stm(DisasContext *s, arg_ldst_block *a, int min_n)
9345 int i, j, n, list, mem_idx;
9347 TCGv_i32 addr, tmp, tmp2;
9352 /* Only usable in supervisor mode. */
9353 unallocated_encoding(s);
9360 if (n < min_n || a->rn == 15) {
9361 unallocated_encoding(s);
9365 addr = op_addr_block_pre(s, a, n);
9366 mem_idx = get_mem_index(s);
9368 for (i = j = 0; i < 16; i++) {
9369 if (!(list & (1 << i))) {
9373 if (user && i != 15) {
9374 tmp = tcg_temp_new_i32();
9375 tmp2 = tcg_const_i32(i);
9376 gen_helper_get_user_reg(tmp, cpu_env, tmp2);
9377 tcg_temp_free_i32(tmp2);
9379 tmp = load_reg(s, i);
9381 gen_aa32_st32(s, tmp, addr, mem_idx);
9382 tcg_temp_free_i32(tmp);
9384 /* No need to add after the last transfer. */
9386 tcg_gen_addi_i32(addr, addr, 4);
9390 op_addr_block_post(s, a, addr, n);
9394 static bool trans_STM(DisasContext *s, arg_ldst_block *a)
9396 /* BitCount(list) < 1 is UNPREDICTABLE */
9397 return op_stm(s, a, 1);
9400 static bool trans_STM_t32(DisasContext *s, arg_ldst_block *a)
9402 /* Writeback register in register list is UNPREDICTABLE for T32. */
9403 if (a->w && (a->list & (1 << a->rn))) {
9404 unallocated_encoding(s);
9407 /* BitCount(list) < 2 is UNPREDICTABLE */
9408 return op_stm(s, a, 2);
9411 static bool do_ldm(DisasContext *s, arg_ldst_block *a, int min_n)
9413 int i, j, n, list, mem_idx;
9416 bool exc_return = false;
9417 TCGv_i32 addr, tmp, tmp2, loaded_var;
9420 /* LDM (user), LDM (exception return) */
9422 /* Only usable in supervisor mode. */
9423 unallocated_encoding(s);
9426 if (extract32(a->list, 15, 1)) {
9430 /* LDM (user) does not allow writeback. */
9432 unallocated_encoding(s);
9440 if (n < min_n || a->rn == 15) {
9441 unallocated_encoding(s);
9445 addr = op_addr_block_pre(s, a, n);
9446 mem_idx = get_mem_index(s);
9447 loaded_base = false;
9450 for (i = j = 0; i < 16; i++) {
9451 if (!(list & (1 << i))) {
9455 tmp = tcg_temp_new_i32();
9456 gen_aa32_ld32u(s, tmp, addr, mem_idx);
9458 tmp2 = tcg_const_i32(i);
9459 gen_helper_set_user_reg(cpu_env, tmp2, tmp);
9460 tcg_temp_free_i32(tmp2);
9461 tcg_temp_free_i32(tmp);
9462 } else if (i == a->rn) {
9465 } else if (i == 15 && exc_return) {
9466 store_pc_exc_ret(s, tmp);
9468 store_reg_from_load(s, i, tmp);
9471 /* No need to add after the last transfer. */
9473 tcg_gen_addi_i32(addr, addr, 4);
9477 op_addr_block_post(s, a, addr, n);
9480 /* Note that we reject base == pc above. */
9481 store_reg(s, a->rn, loaded_var);
9485 /* Restore CPSR from SPSR. */
9486 tmp = load_cpu_field(spsr);
9487 if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
9490 gen_helper_cpsr_write_eret(cpu_env, tmp);
9491 if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
9494 tcg_temp_free_i32(tmp);
9495 /* Must exit loop to check un-masked IRQs */
9496 s->base.is_jmp = DISAS_EXIT;
9501 static bool trans_LDM_a32(DisasContext *s, arg_ldst_block *a)
9504 * Writeback register in register list is UNPREDICTABLE
9505 * for ArchVersion() >= 7. Prior to v7, A32 would write
9506 * an UNKNOWN value to the base register.
9508 if (ENABLE_ARCH_7 && a->w && (a->list & (1 << a->rn))) {
9509 unallocated_encoding(s);
9512 /* BitCount(list) < 1 is UNPREDICTABLE */
9513 return do_ldm(s, a, 1);
9516 static bool trans_LDM_t32(DisasContext *s, arg_ldst_block *a)
9518 /* Writeback register in register list is UNPREDICTABLE for T32. */
9519 if (a->w && (a->list & (1 << a->rn))) {
9520 unallocated_encoding(s);
9523 /* BitCount(list) < 2 is UNPREDICTABLE */
9524 return do_ldm(s, a, 2);
9527 static bool trans_LDM_t16(DisasContext *s, arg_ldst_block *a)
9529 /* Writeback is conditional on the base register not being loaded. */
9530 a->w = !(a->list & (1 << a->rn));
9531 /* BitCount(list) < 1 is UNPREDICTABLE */
9532 return do_ldm(s, a, 1);
9536 * Branch, branch with link
9539 static bool trans_B(DisasContext *s, arg_i *a)
9541 gen_jmp(s, read_pc(s) + a->imm);
9545 static bool trans_B_cond_thumb(DisasContext *s, arg_ci *a)
9547 /* This has cond from encoding, required to be outside IT block. */
9548 if (a->cond >= 0xe) {
9551 if (s->condexec_mask) {
9552 unallocated_encoding(s);
9555 arm_skip_unless(s, a->cond);
9556 gen_jmp(s, read_pc(s) + a->imm);
9560 static bool trans_BL(DisasContext *s, arg_i *a)
9562 tcg_gen_movi_i32(cpu_R[14], s->base.pc_next | s->thumb);
9563 gen_jmp(s, read_pc(s) + a->imm);
9567 static bool trans_BLX_i(DisasContext *s, arg_BLX_i *a)
9571 /* For A32, ARCH(5) is checked near the start of the uncond block. */
9572 if (s->thumb && (a->imm & 2)) {
9575 tcg_gen_movi_i32(cpu_R[14], s->base.pc_next | s->thumb);
9576 tmp = tcg_const_i32(!s->thumb);
9577 store_cpu_field(tmp, thumb);
9578 gen_jmp(s, (read_pc(s) & ~3) + a->imm);
9582 static bool trans_BL_BLX_prefix(DisasContext *s, arg_BL_BLX_prefix *a)
9584 assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
9585 tcg_gen_movi_i32(cpu_R[14], read_pc(s) + (a->imm << 12));
9589 static bool trans_BL_suffix(DisasContext *s, arg_BL_suffix *a)
9591 TCGv_i32 tmp = tcg_temp_new_i32();
9593 assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
9594 tcg_gen_addi_i32(tmp, cpu_R[14], (a->imm << 1) | 1);
9595 tcg_gen_movi_i32(cpu_R[14], s->base.pc_next | 1);
9600 static bool trans_BLX_suffix(DisasContext *s, arg_BLX_suffix *a)
9604 assert(!arm_dc_feature(s, ARM_FEATURE_THUMB2));
9605 if (!ENABLE_ARCH_5) {
9608 tmp = tcg_temp_new_i32();
9609 tcg_gen_addi_i32(tmp, cpu_R[14], a->imm << 1);
9610 tcg_gen_andi_i32(tmp, tmp, 0xfffffffc);
9611 tcg_gen_movi_i32(cpu_R[14], s->base.pc_next | 1);
9616 static bool op_tbranch(DisasContext *s, arg_tbranch *a, bool half)
9620 tmp = load_reg(s, a->rm);
9622 tcg_gen_add_i32(tmp, tmp, tmp);
9624 addr = load_reg(s, a->rn);
9625 tcg_gen_add_i32(addr, addr, tmp);
9627 gen_aa32_ld_i32(s, tmp, addr, get_mem_index(s),
9628 half ? MO_UW | s->be_data : MO_UB);
9629 tcg_temp_free_i32(addr);
9631 tcg_gen_add_i32(tmp, tmp, tmp);
9632 tcg_gen_addi_i32(tmp, tmp, read_pc(s));
9633 store_reg(s, 15, tmp);
9637 static bool trans_TBB(DisasContext *s, arg_tbranch *a)
9639 return op_tbranch(s, a, false);
9642 static bool trans_TBH(DisasContext *s, arg_tbranch *a)
9644 return op_tbranch(s, a, true);
9647 static bool trans_CBZ(DisasContext *s, arg_CBZ *a)
9649 TCGv_i32 tmp = load_reg(s, a->rn);
9651 arm_gen_condlabel(s);
9652 tcg_gen_brcondi_i32(a->nz ? TCG_COND_EQ : TCG_COND_NE,
9653 tmp, 0, s->condlabel);
9654 tcg_temp_free_i32(tmp);
9655 gen_jmp(s, read_pc(s) + a->imm);
9660 * Supervisor call - both T32 & A32 come here so we need to check
9661 * which mode we are in when checking for semihosting.
9664 static bool trans_SVC(DisasContext *s, arg_SVC *a)
9666 const uint32_t semihost_imm = s->thumb ? 0xab : 0x123456;
9668 if (!arm_dc_feature(s, ARM_FEATURE_M) && semihosting_enabled() &&
9669 #ifndef CONFIG_USER_ONLY
9672 (a->imm == semihost_imm)) {
9673 gen_exception_internal_insn(s, s->pc_curr, EXCP_SEMIHOST);
9675 gen_set_pc_im(s, s->base.pc_next);
9676 s->svc_imm = a->imm;
9677 s->base.is_jmp = DISAS_SWI;
9683 * Unconditional system instructions
9686 static bool trans_RFE(DisasContext *s, arg_RFE *a)
9688 static const int8_t pre_offset[4] = {
9689 /* DA */ -4, /* IA */ 0, /* DB */ -8, /* IB */ 4
9691 static const int8_t post_offset[4] = {
9692 /* DA */ -8, /* IA */ 4, /* DB */ -4, /* IB */ 0
9694 TCGv_i32 addr, t1, t2;
9696 if (!ENABLE_ARCH_6 || arm_dc_feature(s, ARM_FEATURE_M)) {
9700 unallocated_encoding(s);
9704 addr = load_reg(s, a->rn);
9705 tcg_gen_addi_i32(addr, addr, pre_offset[a->pu]);
9707 /* Load PC into tmp and CPSR into tmp2. */
9708 t1 = tcg_temp_new_i32();
9709 gen_aa32_ld32u(s, t1, addr, get_mem_index(s));
9710 tcg_gen_addi_i32(addr, addr, 4);
9711 t2 = tcg_temp_new_i32();
9712 gen_aa32_ld32u(s, t2, addr, get_mem_index(s));
9715 /* Base writeback. */
9716 tcg_gen_addi_i32(addr, addr, post_offset[a->pu]);
9717 store_reg(s, a->rn, addr);
9719 tcg_temp_free_i32(addr);
9725 static bool trans_SRS(DisasContext *s, arg_SRS *a)
9727 if (!ENABLE_ARCH_6 || arm_dc_feature(s, ARM_FEATURE_M)) {
9730 gen_srs(s, a->mode, a->pu, a->w);
9734 static bool trans_CPS(DisasContext *s, arg_CPS *a)
9738 if (!ENABLE_ARCH_6 || arm_dc_feature(s, ARM_FEATURE_M)) {
9742 /* Implemented as NOP in user mode. */
9745 /* TODO: There are quite a lot of UNPREDICTABLE argument combinations. */
9767 gen_set_psr_im(s, mask, 0, val);
9772 static bool trans_CPS_v7m(DisasContext *s, arg_CPS_v7m *a)
9774 TCGv_i32 tmp, addr, el;
9776 if (!arm_dc_feature(s, ARM_FEATURE_M)) {
9780 /* Implemented as NOP in user mode. */
9784 tmp = tcg_const_i32(a->im);
9787 addr = tcg_const_i32(19);
9788 gen_helper_v7m_msr(cpu_env, addr, tmp);
9789 tcg_temp_free_i32(addr);
9793 addr = tcg_const_i32(16);
9794 gen_helper_v7m_msr(cpu_env, addr, tmp);
9795 tcg_temp_free_i32(addr);
9797 el = tcg_const_i32(s->current_el);
9798 gen_helper_rebuild_hflags_m32(cpu_env, el);
9799 tcg_temp_free_i32(el);
9800 tcg_temp_free_i32(tmp);
9806 * Clear-Exclusive, Barriers
9809 static bool trans_CLREX(DisasContext *s, arg_CLREX *a)
9812 ? !ENABLE_ARCH_7 && !arm_dc_feature(s, ARM_FEATURE_M)
9813 : !ENABLE_ARCH_6K) {
9820 static bool trans_DSB(DisasContext *s, arg_DSB *a)
9822 if (!ENABLE_ARCH_7 && !arm_dc_feature(s, ARM_FEATURE_M)) {
9825 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
9829 static bool trans_DMB(DisasContext *s, arg_DMB *a)
9831 return trans_DSB(s, NULL);
9834 static bool trans_ISB(DisasContext *s, arg_ISB *a)
9836 if (!ENABLE_ARCH_7 && !arm_dc_feature(s, ARM_FEATURE_M)) {
9840 * We need to break the TB after this insn to execute
9841 * self-modifying code correctly and also to take
9842 * any pending interrupts immediately.
9844 gen_goto_tb(s, 0, s->base.pc_next);
9848 static bool trans_SB(DisasContext *s, arg_SB *a)
9850 if (!dc_isar_feature(aa32_sb, s)) {
9854 * TODO: There is no speculation barrier opcode
9855 * for TCG; MB and end the TB instead.
9857 tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
9858 gen_goto_tb(s, 0, s->base.pc_next);
9862 static bool trans_SETEND(DisasContext *s, arg_SETEND *a)
9864 if (!ENABLE_ARCH_6) {
9867 if (a->E != (s->be_data == MO_BE)) {
9868 gen_helper_setend(cpu_env);
9869 s->base.is_jmp = DISAS_UPDATE;
9875 * Preload instructions
9876 * All are nops, contingent on the appropriate arch level.
9879 static bool trans_PLD(DisasContext *s, arg_PLD *a)
9881 return ENABLE_ARCH_5TE;
9884 static bool trans_PLDW(DisasContext *s, arg_PLD *a)
9886 return arm_dc_feature(s, ARM_FEATURE_V7MP);
9889 static bool trans_PLI(DisasContext *s, arg_PLD *a)
9891 return ENABLE_ARCH_7;
9898 static bool trans_IT(DisasContext *s, arg_IT *a)
9900 int cond_mask = a->cond_mask;
9903 * No actual code generated for this insn, just setup state.
9905 * Combinations of firstcond and mask which set up an 0b1111
9906 * condition are UNPREDICTABLE; we take the CONSTRAINED
9907 * UNPREDICTABLE choice to treat 0b1111 the same as 0b1110,
9908 * i.e. both meaning "execute always".
9910 s->condexec_cond = (cond_mask >> 4) & 0xe;
9911 s->condexec_mask = cond_mask & 0x1f;
9919 static void disas_arm_insn(DisasContext *s, unsigned int insn)
9921 unsigned int cond = insn >> 28;
9923 /* M variants do not implement ARM mode; this must raise the INVSTATE
9924 * UsageFault exception.
9926 if (arm_dc_feature(s, ARM_FEATURE_M)) {
9927 gen_exception_insn(s, s->pc_curr, EXCP_INVSTATE, syn_uncategorized(),
9928 default_exception_el(s));
9933 /* In ARMv3 and v4 the NV condition is UNPREDICTABLE; we
9934 * choose to UNDEF. In ARMv5 and above the space is used
9935 * for miscellaneous unconditional instructions.
9939 /* Unconditional instructions. */
9940 /* TODO: Perhaps merge these into one decodetree output file. */
9941 if (disas_a32_uncond(s, insn) ||
9942 disas_vfp_uncond(s, insn) ||
9943 disas_neon_dp(s, insn) ||
9944 disas_neon_ls(s, insn) ||
9945 disas_neon_shared(s, insn)) {
9948 /* fall back to legacy decoder */
9950 if (((insn >> 25) & 7) == 1) {
9951 /* NEON Data processing. */
9952 if (disas_neon_data_insn(s, insn)) {
9957 if ((insn & 0x0e000f00) == 0x0c000100) {
9958 if (arm_dc_feature(s, ARM_FEATURE_IWMMXT)) {
9959 /* iWMMXt register transfer. */
9960 if (extract32(s->c15_cpar, 1, 1)) {
9961 if (!disas_iwmmxt_insn(s, insn)) {
9970 /* if not always execute, we generate a conditional jump to
9972 arm_skip_unless(s, cond);
9975 /* TODO: Perhaps merge these into one decodetree output file. */
9976 if (disas_a32(s, insn) ||
9977 disas_vfp(s, insn)) {
9980 /* fall back to legacy decoder */
9982 switch ((insn >> 24) & 0xf) {
9986 if (((insn >> 8) & 0xe) == 10) {
9987 /* VFP, but failed disas_vfp. */
9990 if (disas_coproc_insn(s, insn)) {
9997 unallocated_encoding(s);
10002 static bool thumb_insn_is_16bit(DisasContext *s, uint32_t pc, uint32_t insn)
10005 * Return true if this is a 16 bit instruction. We must be precise
10006 * about this (matching the decode).
10008 if ((insn >> 11) < 0x1d) {
10009 /* Definitely a 16-bit instruction */
10013 /* Top five bits 0b11101 / 0b11110 / 0b11111 : this is the
10014 * first half of a 32-bit Thumb insn. Thumb-1 cores might
10015 * end up actually treating this as two 16-bit insns, though,
10016 * if it's half of a bl/blx pair that might span a page boundary.
10018 if (arm_dc_feature(s, ARM_FEATURE_THUMB2) ||
10019 arm_dc_feature(s, ARM_FEATURE_M)) {
10020 /* Thumb2 cores (including all M profile ones) always treat
10021 * 32-bit insns as 32-bit.
10026 if ((insn >> 11) == 0x1e && pc - s->page_start < TARGET_PAGE_SIZE - 3) {
10027 /* 0b1111_0xxx_xxxx_xxxx : BL/BLX prefix, and the suffix
10028 * is not on the next page; we merge this into a 32-bit
10033 /* 0b1110_1xxx_xxxx_xxxx : BLX suffix (or UNDEF);
10034 * 0b1111_1xxx_xxxx_xxxx : BL suffix;
10035 * 0b1111_0xxx_xxxx_xxxx : BL/BLX prefix on the end of a page
10036 * -- handle as single 16 bit insn
10041 /* Translate a 32-bit thumb instruction. */
10042 static void disas_thumb2_insn(DisasContext *s, uint32_t insn)
10045 * ARMv6-M supports a limited subset of Thumb2 instructions.
10046 * Other Thumb1 architectures allow only 32-bit
10047 * combined BL/BLX prefix and suffix.
10049 if (arm_dc_feature(s, ARM_FEATURE_M) &&
10050 !arm_dc_feature(s, ARM_FEATURE_V7)) {
10052 bool found = false;
10053 static const uint32_t armv6m_insn[] = {0xf3808000 /* msr */,
10054 0xf3b08040 /* dsb */,
10055 0xf3b08050 /* dmb */,
10056 0xf3b08060 /* isb */,
10057 0xf3e08000 /* mrs */,
10058 0xf000d000 /* bl */};
10059 static const uint32_t armv6m_mask[] = {0xffe0d000,
10066 for (i = 0; i < ARRAY_SIZE(armv6m_insn); i++) {
10067 if ((insn & armv6m_mask[i]) == armv6m_insn[i]) {
10075 } else if ((insn & 0xf800e800) != 0xf000e800) {
10079 if ((insn & 0xef000000) == 0xef000000) {
10081 * T32 encodings 0b111p_1111_qqqq_qqqq_qqqq_qqqq_qqqq_qqqq
10083 * A32 encodings 0b1111_001p_qqqq_qqqq_qqqq_qqqq_qqqq_qqqq
10085 uint32_t a32_insn = (insn & 0xe2ffffff) |
10086 ((insn & (1 << 28)) >> 4) | (1 << 28);
10088 if (disas_neon_dp(s, a32_insn)) {
10093 if ((insn & 0xff100000) == 0xf9000000) {
10095 * T32 encodings 0b1111_1001_ppp0_qqqq_qqqq_qqqq_qqqq_qqqq
10097 * A32 encodings 0b1111_0100_ppp0_qqqq_qqqq_qqqq_qqqq_qqqq
10099 uint32_t a32_insn = (insn & 0x00ffffff) | 0xf4000000;
10101 if (disas_neon_ls(s, a32_insn)) {
10107 * TODO: Perhaps merge these into one decodetree output file.
10108 * Note disas_vfp is written for a32 with cond field in the
10109 * top nibble. The t32 encoding requires 0xe in the top nibble.
10111 if (disas_t32(s, insn) ||
10112 disas_vfp_uncond(s, insn) ||
10113 disas_neon_shared(s, insn) ||
10114 ((insn >> 28) == 0xe && disas_vfp(s, insn))) {
10117 /* fall back to legacy decoder */
10119 switch ((insn >> 25) & 0xf) {
10120 case 0: case 1: case 2: case 3:
10121 /* 16-bit instructions. Should never happen. */
10123 case 6: case 7: case 14: case 15:
10125 if (arm_dc_feature(s, ARM_FEATURE_M)) {
10126 /* 0b111x_11xx_xxxx_xxxx_xxxx_xxxx_xxxx_xxxx */
10127 if (extract32(insn, 24, 2) == 3) {
10128 goto illegal_op; /* op0 = 0b11 : unallocated */
10131 if (((insn >> 8) & 0xe) == 10 &&
10132 dc_isar_feature(aa32_fpsp_v2, s)) {
10133 /* FP, and the CPU supports it */
10136 /* All other insns: NOCP */
10137 gen_exception_insn(s, s->pc_curr, EXCP_NOCP,
10138 syn_uncategorized(),
10139 default_exception_el(s));
10143 if (((insn >> 24) & 3) == 3) {
10144 /* Translate into the equivalent ARM encoding. */
10145 insn = (insn & 0xe2ffffff) | ((insn & (1 << 28)) >> 4) | (1 << 28);
10146 if (disas_neon_data_insn(s, insn)) {
10149 } else if (((insn >> 8) & 0xe) == 10) {
10150 /* VFP, but failed disas_vfp. */
10153 if (insn & (1 << 28))
10155 if (disas_coproc_insn(s, insn)) {
10164 unallocated_encoding(s);
10168 static void disas_thumb_insn(DisasContext *s, uint32_t insn)
10170 if (!disas_t16(s, insn)) {
10171 unallocated_encoding(s);
10175 static bool insn_crosses_page(CPUARMState *env, DisasContext *s)
10177 /* Return true if the insn at dc->base.pc_next might cross a page boundary.
10178 * (False positives are OK, false negatives are not.)
10179 * We know this is a Thumb insn, and our caller ensures we are
10180 * only called if dc->base.pc_next is less than 4 bytes from the page
10181 * boundary, so we cross the page if the first 16 bits indicate
10182 * that this is a 32 bit insn.
10184 uint16_t insn = arm_lduw_code(env, s->base.pc_next, s->sctlr_b);
10186 return !thumb_insn_is_16bit(s, s->base.pc_next, insn);
10189 static void arm_tr_init_disas_context(DisasContextBase *dcbase, CPUState *cs)
10191 DisasContext *dc = container_of(dcbase, DisasContext, base);
10192 CPUARMState *env = cs->env_ptr;
10193 ARMCPU *cpu = env_archcpu(env);
10194 uint32_t tb_flags = dc->base.tb->flags;
10195 uint32_t condexec, core_mmu_idx;
10197 dc->isar = &cpu->isar;
10201 /* If we are coming from secure EL0 in a system with a 32-bit EL3, then
10202 * there is no secure EL1, so we route exceptions to EL3.
10204 dc->secure_routed_to_el3 = arm_feature(env, ARM_FEATURE_EL3) &&
10205 !arm_el_is_aa64(env, 3);
10206 dc->thumb = FIELD_EX32(tb_flags, TBFLAG_AM32, THUMB);
10207 dc->be_data = FIELD_EX32(tb_flags, TBFLAG_ANY, BE_DATA) ? MO_BE : MO_LE;
10208 condexec = FIELD_EX32(tb_flags, TBFLAG_AM32, CONDEXEC);
10209 dc->condexec_mask = (condexec & 0xf) << 1;
10210 dc->condexec_cond = condexec >> 4;
10212 core_mmu_idx = FIELD_EX32(tb_flags, TBFLAG_ANY, MMUIDX);
10213 dc->mmu_idx = core_to_arm_mmu_idx(env, core_mmu_idx);
10214 dc->current_el = arm_mmu_idx_to_el(dc->mmu_idx);
10215 #if !defined(CONFIG_USER_ONLY)
10216 dc->user = (dc->current_el == 0);
10218 dc->fp_excp_el = FIELD_EX32(tb_flags, TBFLAG_ANY, FPEXC_EL);
10220 if (arm_feature(env, ARM_FEATURE_M)) {
10221 dc->vfp_enabled = 1;
10222 dc->be_data = MO_TE;
10223 dc->v7m_handler_mode = FIELD_EX32(tb_flags, TBFLAG_M32, HANDLER);
10224 dc->v8m_secure = arm_feature(env, ARM_FEATURE_M_SECURITY) &&
10225 regime_is_secure(env, dc->mmu_idx);
10226 dc->v8m_stackcheck = FIELD_EX32(tb_flags, TBFLAG_M32, STACKCHECK);
10227 dc->v8m_fpccr_s_wrong =
10228 FIELD_EX32(tb_flags, TBFLAG_M32, FPCCR_S_WRONG);
10229 dc->v7m_new_fp_ctxt_needed =
10230 FIELD_EX32(tb_flags, TBFLAG_M32, NEW_FP_CTXT_NEEDED);
10231 dc->v7m_lspact = FIELD_EX32(tb_flags, TBFLAG_M32, LSPACT);
10234 FIELD_EX32(tb_flags, TBFLAG_ANY, BE_DATA) ? MO_BE : MO_LE;
10235 dc->debug_target_el =
10236 FIELD_EX32(tb_flags, TBFLAG_ANY, DEBUG_TARGET_EL);
10237 dc->sctlr_b = FIELD_EX32(tb_flags, TBFLAG_A32, SCTLR_B);
10238 dc->hstr_active = FIELD_EX32(tb_flags, TBFLAG_A32, HSTR_ACTIVE);
10239 dc->ns = FIELD_EX32(tb_flags, TBFLAG_A32, NS);
10240 dc->vfp_enabled = FIELD_EX32(tb_flags, TBFLAG_A32, VFPEN);
10241 if (arm_feature(env, ARM_FEATURE_XSCALE)) {
10242 dc->c15_cpar = FIELD_EX32(tb_flags, TBFLAG_A32, XSCALE_CPAR);
10244 dc->vec_len = FIELD_EX32(tb_flags, TBFLAG_A32, VECLEN);
10245 dc->vec_stride = FIELD_EX32(tb_flags, TBFLAG_A32, VECSTRIDE);
10248 dc->cp_regs = cpu->cp_regs;
10249 dc->features = env->features;
10251 /* Single step state. The code-generation logic here is:
10253 * generate code with no special handling for single-stepping (except
10254 * that anything that can make us go to SS_ACTIVE == 1 must end the TB;
10255 * this happens anyway because those changes are all system register or
10257 * SS_ACTIVE == 1, PSTATE.SS == 1: (active-not-pending)
10258 * emit code for one insn
10259 * emit code to clear PSTATE.SS
10260 * emit code to generate software step exception for completed step
10261 * end TB (as usual for having generated an exception)
10262 * SS_ACTIVE == 1, PSTATE.SS == 0: (active-pending)
10263 * emit code to generate a software step exception
10266 dc->ss_active = FIELD_EX32(tb_flags, TBFLAG_ANY, SS_ACTIVE);
10267 dc->pstate_ss = FIELD_EX32(tb_flags, TBFLAG_ANY, PSTATE_SS);
10268 dc->is_ldex = false;
10270 dc->page_start = dc->base.pc_first & TARGET_PAGE_MASK;
10272 /* If architectural single step active, limit to 1. */
10273 if (is_singlestepping(dc)) {
10274 dc->base.max_insns = 1;
10277 /* ARM is a fixed-length ISA. Bound the number of insns to execute
10278 to those left on the page. */
10280 int bound = -(dc->base.pc_first | TARGET_PAGE_MASK) / 4;
10281 dc->base.max_insns = MIN(dc->base.max_insns, bound);
10284 cpu_V0 = tcg_temp_new_i64();
10285 cpu_V1 = tcg_temp_new_i64();
10286 /* FIXME: cpu_M0 can probably be the same as cpu_V0. */
10287 cpu_M0 = tcg_temp_new_i64();
10290 static void arm_tr_tb_start(DisasContextBase *dcbase, CPUState *cpu)
10292 DisasContext *dc = container_of(dcbase, DisasContext, base);
10294 /* A note on handling of the condexec (IT) bits:
10296 * We want to avoid the overhead of having to write the updated condexec
10297 * bits back to the CPUARMState for every instruction in an IT block. So:
10298 * (1) if the condexec bits are not already zero then we write
10299 * zero back into the CPUARMState now. This avoids complications trying
10300 * to do it at the end of the block. (For example if we don't do this
10301 * it's hard to identify whether we can safely skip writing condexec
10302 * at the end of the TB, which we definitely want to do for the case
10303 * where a TB doesn't do anything with the IT state at all.)
10304 * (2) if we are going to leave the TB then we call gen_set_condexec()
10305 * which will write the correct value into CPUARMState if zero is wrong.
10306 * This is done both for leaving the TB at the end, and for leaving
10307 * it because of an exception we know will happen, which is done in
10308 * gen_exception_insn(). The latter is necessary because we need to
10309 * leave the TB with the PC/IT state just prior to execution of the
10310 * instruction which caused the exception.
10311 * (3) if we leave the TB unexpectedly (eg a data abort on a load)
10312 * then the CPUARMState will be wrong and we need to reset it.
10313 * This is handled in the same way as restoration of the
10314 * PC in these situations; we save the value of the condexec bits
10315 * for each PC via tcg_gen_insn_start(), and restore_state_to_opc()
10316 * then uses this to restore them after an exception.
10318 * Note that there are no instructions which can read the condexec
10319 * bits, and none which can write non-static values to them, so
10320 * we don't need to care about whether CPUARMState is correct in the
10324 /* Reset the conditional execution bits immediately. This avoids
10325 complications trying to do it at the end of the block. */
10326 if (dc->condexec_mask || dc->condexec_cond) {
10327 TCGv_i32 tmp = tcg_temp_new_i32();
10328 tcg_gen_movi_i32(tmp, 0);
10329 store_cpu_field(tmp, condexec_bits);
10333 static void arm_tr_insn_start(DisasContextBase *dcbase, CPUState *cpu)
10335 DisasContext *dc = container_of(dcbase, DisasContext, base);
10337 tcg_gen_insn_start(dc->base.pc_next,
10338 (dc->condexec_cond << 4) | (dc->condexec_mask >> 1),
10340 dc->insn_start = tcg_last_op();
10343 static bool arm_tr_breakpoint_check(DisasContextBase *dcbase, CPUState *cpu,
10344 const CPUBreakpoint *bp)
10346 DisasContext *dc = container_of(dcbase, DisasContext, base);
10348 if (bp->flags & BP_CPU) {
10349 gen_set_condexec(dc);
10350 gen_set_pc_im(dc, dc->base.pc_next);
10351 gen_helper_check_breakpoints(cpu_env);
10352 /* End the TB early; it's likely not going to be executed */
10353 dc->base.is_jmp = DISAS_TOO_MANY;
10355 gen_exception_internal_insn(dc, dc->base.pc_next, EXCP_DEBUG);
10356 /* The address covered by the breakpoint must be
10357 included in [tb->pc, tb->pc + tb->size) in order
10358 to for it to be properly cleared -- thus we
10359 increment the PC here so that the logic setting
10360 tb->size below does the right thing. */
10361 /* TODO: Advance PC by correct instruction length to
10362 * avoid disassembler error messages */
10363 dc->base.pc_next += 2;
10364 dc->base.is_jmp = DISAS_NORETURN;
10370 static bool arm_pre_translate_insn(DisasContext *dc)
10372 #ifdef CONFIG_USER_ONLY
10373 /* Intercept jump to the magic kernel page. */
10374 if (dc->base.pc_next >= 0xffff0000) {
10375 /* We always get here via a jump, so know we are not in a
10376 conditional execution block. */
10377 gen_exception_internal(EXCP_KERNEL_TRAP);
10378 dc->base.is_jmp = DISAS_NORETURN;
10383 if (dc->ss_active && !dc->pstate_ss) {
10384 /* Singlestep state is Active-pending.
10385 * If we're in this state at the start of a TB then either
10386 * a) we just took an exception to an EL which is being debugged
10387 * and this is the first insn in the exception handler
10388 * b) debug exceptions were masked and we just unmasked them
10389 * without changing EL (eg by clearing PSTATE.D)
10390 * In either case we're going to take a swstep exception in the
10391 * "did not step an insn" case, and so the syndrome ISV and EX
10392 * bits should be zero.
10394 assert(dc->base.num_insns == 1);
10395 gen_swstep_exception(dc, 0, 0);
10396 dc->base.is_jmp = DISAS_NORETURN;
10403 static void arm_post_translate_insn(DisasContext *dc)
10405 if (dc->condjmp && !dc->base.is_jmp) {
10406 gen_set_label(dc->condlabel);
10409 translator_loop_temp_check(&dc->base);
10412 static void arm_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu)
10414 DisasContext *dc = container_of(dcbase, DisasContext, base);
10415 CPUARMState *env = cpu->env_ptr;
10418 if (arm_pre_translate_insn(dc)) {
10422 dc->pc_curr = dc->base.pc_next;
10423 insn = arm_ldl_code(env, dc->base.pc_next, dc->sctlr_b);
10425 dc->base.pc_next += 4;
10426 disas_arm_insn(dc, insn);
10428 arm_post_translate_insn(dc);
10430 /* ARM is a fixed-length ISA. We performed the cross-page check
10431 in init_disas_context by adjusting max_insns. */
10434 static bool thumb_insn_is_unconditional(DisasContext *s, uint32_t insn)
10436 /* Return true if this Thumb insn is always unconditional,
10437 * even inside an IT block. This is true of only a very few
10438 * instructions: BKPT, HLT, and SG.
10440 * A larger class of instructions are UNPREDICTABLE if used
10441 * inside an IT block; we do not need to detect those here, because
10442 * what we do by default (perform the cc check and update the IT
10443 * bits state machine) is a permitted CONSTRAINED UNPREDICTABLE
10444 * choice for those situations.
10446 * insn is either a 16-bit or a 32-bit instruction; the two are
10447 * distinguishable because for the 16-bit case the top 16 bits
10448 * are zeroes, and that isn't a valid 32-bit encoding.
10450 if ((insn & 0xffffff00) == 0xbe00) {
10455 if ((insn & 0xffffffc0) == 0xba80 && arm_dc_feature(s, ARM_FEATURE_V8) &&
10456 !arm_dc_feature(s, ARM_FEATURE_M)) {
10457 /* HLT: v8A only. This is unconditional even when it is going to
10458 * UNDEF; see the v8A ARM ARM DDI0487B.a H3.3.
10459 * For v7 cores this was a plain old undefined encoding and so
10460 * honours its cc check. (We might be using the encoding as
10461 * a semihosting trap, but we don't change the cc check behaviour
10462 * on that account, because a debugger connected to a real v7A
10463 * core and emulating semihosting traps by catching the UNDEF
10464 * exception would also only see cases where the cc check passed.
10465 * No guest code should be trying to do a HLT semihosting trap
10466 * in an IT block anyway.
10471 if (insn == 0xe97fe97f && arm_dc_feature(s, ARM_FEATURE_V8) &&
10472 arm_dc_feature(s, ARM_FEATURE_M)) {
10480 static void thumb_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu)
10482 DisasContext *dc = container_of(dcbase, DisasContext, base);
10483 CPUARMState *env = cpu->env_ptr;
10487 if (arm_pre_translate_insn(dc)) {
10491 dc->pc_curr = dc->base.pc_next;
10492 insn = arm_lduw_code(env, dc->base.pc_next, dc->sctlr_b);
10493 is_16bit = thumb_insn_is_16bit(dc, dc->base.pc_next, insn);
10494 dc->base.pc_next += 2;
10496 uint32_t insn2 = arm_lduw_code(env, dc->base.pc_next, dc->sctlr_b);
10498 insn = insn << 16 | insn2;
10499 dc->base.pc_next += 2;
10503 if (dc->condexec_mask && !thumb_insn_is_unconditional(dc, insn)) {
10504 uint32_t cond = dc->condexec_cond;
10507 * Conditionally skip the insn. Note that both 0xe and 0xf mean
10508 * "always"; 0xf is not "never".
10511 arm_skip_unless(dc, cond);
10516 disas_thumb_insn(dc, insn);
10518 disas_thumb2_insn(dc, insn);
10521 /* Advance the Thumb condexec condition. */
10522 if (dc->condexec_mask) {
10523 dc->condexec_cond = ((dc->condexec_cond & 0xe) |
10524 ((dc->condexec_mask >> 4) & 1));
10525 dc->condexec_mask = (dc->condexec_mask << 1) & 0x1f;
10526 if (dc->condexec_mask == 0) {
10527 dc->condexec_cond = 0;
10531 arm_post_translate_insn(dc);
10533 /* Thumb is a variable-length ISA. Stop translation when the next insn
10534 * will touch a new page. This ensures that prefetch aborts occur at
10537 * We want to stop the TB if the next insn starts in a new page,
10538 * or if it spans between this page and the next. This means that
10539 * if we're looking at the last halfword in the page we need to
10540 * see if it's a 16-bit Thumb insn (which will fit in this TB)
10541 * or a 32-bit Thumb insn (which won't).
10542 * This is to avoid generating a silly TB with a single 16-bit insn
10543 * in it at the end of this page (which would execute correctly
10544 * but isn't very efficient).
10546 if (dc->base.is_jmp == DISAS_NEXT
10547 && (dc->base.pc_next - dc->page_start >= TARGET_PAGE_SIZE
10548 || (dc->base.pc_next - dc->page_start >= TARGET_PAGE_SIZE - 3
10549 && insn_crosses_page(env, dc)))) {
10550 dc->base.is_jmp = DISAS_TOO_MANY;
10554 static void arm_tr_tb_stop(DisasContextBase *dcbase, CPUState *cpu)
10556 DisasContext *dc = container_of(dcbase, DisasContext, base);
10558 if (tb_cflags(dc->base.tb) & CF_LAST_IO && dc->condjmp) {
10559 /* FIXME: This can theoretically happen with self-modifying code. */
10560 cpu_abort(cpu, "IO on conditional branch instruction");
10563 /* At this stage dc->condjmp will only be set when the skipped
10564 instruction was a conditional branch or trap, and the PC has
10565 already been written. */
10566 gen_set_condexec(dc);
10567 if (dc->base.is_jmp == DISAS_BX_EXCRET) {
10568 /* Exception return branches need some special case code at the
10569 * end of the TB, which is complex enough that it has to
10570 * handle the single-step vs not and the condition-failed
10571 * insn codepath itself.
10573 gen_bx_excret_final_code(dc);
10574 } else if (unlikely(is_singlestepping(dc))) {
10575 /* Unconditional and "condition passed" instruction codepath. */
10576 switch (dc->base.is_jmp) {
10578 gen_ss_advance(dc);
10579 gen_exception(EXCP_SWI, syn_aa32_svc(dc->svc_imm, dc->thumb),
10580 default_exception_el(dc));
10583 gen_ss_advance(dc);
10584 gen_exception(EXCP_HVC, syn_aa32_hvc(dc->svc_imm), 2);
10587 gen_ss_advance(dc);
10588 gen_exception(EXCP_SMC, syn_aa32_smc(), 3);
10591 case DISAS_TOO_MANY:
10593 gen_set_pc_im(dc, dc->base.pc_next);
10596 /* FIXME: Single stepping a WFI insn will not halt the CPU. */
10597 gen_singlestep_exception(dc);
10599 case DISAS_NORETURN:
10603 /* While branches must always occur at the end of an IT block,
10604 there are a few other things that can cause us to terminate
10605 the TB in the middle of an IT block:
10606 - Exception generating instructions (bkpt, swi, undefined).
10608 - Hardware watchpoints.
10609 Hardware breakpoints have already been handled and skip this code.
10611 switch(dc->base.is_jmp) {
10613 case DISAS_TOO_MANY:
10614 gen_goto_tb(dc, 1, dc->base.pc_next);
10620 gen_set_pc_im(dc, dc->base.pc_next);
10623 /* indicate that the hash table must be used to find the next TB */
10624 tcg_gen_exit_tb(NULL, 0);
10626 case DISAS_NORETURN:
10627 /* nothing more to generate */
10631 TCGv_i32 tmp = tcg_const_i32((dc->thumb &&
10632 !(dc->insn & (1U << 31))) ? 2 : 4);
10634 gen_helper_wfi(cpu_env, tmp);
10635 tcg_temp_free_i32(tmp);
10636 /* The helper doesn't necessarily throw an exception, but we
10637 * must go back to the main loop to check for interrupts anyway.
10639 tcg_gen_exit_tb(NULL, 0);
10643 gen_helper_wfe(cpu_env);
10646 gen_helper_yield(cpu_env);
10649 gen_exception(EXCP_SWI, syn_aa32_svc(dc->svc_imm, dc->thumb),
10650 default_exception_el(dc));
10653 gen_exception(EXCP_HVC, syn_aa32_hvc(dc->svc_imm), 2);
10656 gen_exception(EXCP_SMC, syn_aa32_smc(), 3);
10662 /* "Condition failed" instruction codepath for the branch/trap insn */
10663 gen_set_label(dc->condlabel);
10664 gen_set_condexec(dc);
10665 if (unlikely(is_singlestepping(dc))) {
10666 gen_set_pc_im(dc, dc->base.pc_next);
10667 gen_singlestep_exception(dc);
10669 gen_goto_tb(dc, 1, dc->base.pc_next);
10674 static void arm_tr_disas_log(const DisasContextBase *dcbase, CPUState *cpu)
10676 DisasContext *dc = container_of(dcbase, DisasContext, base);
10678 qemu_log("IN: %s\n", lookup_symbol(dc->base.pc_first));
10679 log_target_disas(cpu, dc->base.pc_first, dc->base.tb->size);
10682 static const TranslatorOps arm_translator_ops = {
10683 .init_disas_context = arm_tr_init_disas_context,
10684 .tb_start = arm_tr_tb_start,
10685 .insn_start = arm_tr_insn_start,
10686 .breakpoint_check = arm_tr_breakpoint_check,
10687 .translate_insn = arm_tr_translate_insn,
10688 .tb_stop = arm_tr_tb_stop,
10689 .disas_log = arm_tr_disas_log,
10692 static const TranslatorOps thumb_translator_ops = {
10693 .init_disas_context = arm_tr_init_disas_context,
10694 .tb_start = arm_tr_tb_start,
10695 .insn_start = arm_tr_insn_start,
10696 .breakpoint_check = arm_tr_breakpoint_check,
10697 .translate_insn = thumb_tr_translate_insn,
10698 .tb_stop = arm_tr_tb_stop,
10699 .disas_log = arm_tr_disas_log,
10702 /* generate intermediate code for basic block 'tb'. */
10703 void gen_intermediate_code(CPUState *cpu, TranslationBlock *tb, int max_insns)
10705 DisasContext dc = { };
10706 const TranslatorOps *ops = &arm_translator_ops;
10708 if (FIELD_EX32(tb->flags, TBFLAG_AM32, THUMB)) {
10709 ops = &thumb_translator_ops;
10711 #ifdef TARGET_AARCH64
10712 if (FIELD_EX32(tb->flags, TBFLAG_ANY, AARCH64_STATE)) {
10713 ops = &aarch64_translator_ops;
10717 translator_loop(ops, &dc.base, cpu, tb, max_insns);
10720 void restore_state_to_opc(CPUARMState *env, TranslationBlock *tb,
10721 target_ulong *data)
10725 env->condexec_bits = 0;
10726 env->exception.syndrome = data[2] << ARM_INSN_START_WORD2_SHIFT;
10728 env->regs[15] = data[0];
10729 env->condexec_bits = data[1];
10730 env->exception.syndrome = data[2] << ARM_INSN_START_WORD2_SHIFT;