?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
vga: pass owner to vga_init
[qemu/ar7.git] / target-arm / translate.c
blobaf2aef29e3b52174d732533e74c8ab4dc972b0d3
1 /*
2 * ARM translation
3 *
4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2005-2007 CodeSourcery
6 * Copyright (c) 2007 OpenedHand, Ltd.
7 *
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.
12 *
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.
17 *
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/>.
20 */
21 #include <stdarg.h>
22 #include <stdlib.h>
23 #include <stdio.h>
24 #include <string.h>
25 #include <inttypes.h>
26
27 #include "cpu.h"
28 #include "disas/disas.h"
29 #include "tcg-op.h"
30 #include "qemu/log.h"
31
32 #include "helper.h"
33 #define GEN_HELPER 1
34 #include "helper.h"
35
36 #define ENABLE_ARCH_4T arm_feature(env, ARM_FEATURE_V4T)
37 #define ENABLE_ARCH_5 arm_feature(env, ARM_FEATURE_V5)
38 /* currently all emulated v5 cores are also v5TE, so don't bother */
39 #define ENABLE_ARCH_5TE arm_feature(env, ARM_FEATURE_V5)
40 #define ENABLE_ARCH_5J 0
41 #define ENABLE_ARCH_6 arm_feature(env, ARM_FEATURE_V6)
42 #define ENABLE_ARCH_6K arm_feature(env, ARM_FEATURE_V6K)
43 #define ENABLE_ARCH_6T2 arm_feature(env, ARM_FEATURE_THUMB2)
44 #define ENABLE_ARCH_7 arm_feature(env, ARM_FEATURE_V7)
45
46 #define ARCH(x) do { if (!ENABLE_ARCH_##x) goto illegal_op; } while(0)
47
48 /* internal defines */
49 typedef struct DisasContext {
50 target_ulong pc;
51 int is_jmp;
52 /* Nonzero if this instruction has been conditionally skipped. */
53 int condjmp;
54 /* The label that will be jumped to when the instruction is skipped. */
55 int condlabel;
56 /* Thumb-2 conditional execution bits. */
57 int condexec_mask;
58 int condexec_cond;
59 struct TranslationBlock *tb;
60 int singlestep_enabled;
61 int thumb;
62 int bswap_code;
63 #if !defined(CONFIG_USER_ONLY)
64 int user;
65 #endif
66 int vfp_enabled;
67 int vec_len;
68 int vec_stride;
69 } DisasContext;
70
71 static uint32_t gen_opc_condexec_bits[OPC_BUF_SIZE];
72
73 #if defined(CONFIG_USER_ONLY)
74 #define IS_USER(s) 1
75 #else
76 #define IS_USER(s) (s->user)
77 #endif
78
79 /* These instructions trap after executing, so defer them until after the
80 conditional execution state has been updated. */
81 #define DISAS_WFI 4
82 #define DISAS_SWI 5
83
84 static TCGv_ptr cpu_env;
85 /* We reuse the same 64-bit temporaries for efficiency. */
86 static TCGv_i64 cpu_V0, cpu_V1, cpu_M0;
87 static TCGv_i32 cpu_R[16];
88 static TCGv_i32 cpu_CF, cpu_NF, cpu_VF, cpu_ZF;
89 static TCGv_i32 cpu_exclusive_addr;
90 static TCGv_i32 cpu_exclusive_val;
91 static TCGv_i32 cpu_exclusive_high;
92 #ifdef CONFIG_USER_ONLY
93 static TCGv_i32 cpu_exclusive_test;
94 static TCGv_i32 cpu_exclusive_info;
95 #endif
96
97 /* FIXME: These should be removed. */
98 static TCGv_i32 cpu_F0s, cpu_F1s;
99 static TCGv_i64 cpu_F0d, cpu_F1d;
100
101 #include "exec/gen-icount.h"
102
103 static const char *regnames[] =
104 { "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7",
105 "r8", "r9", "r10", "r11", "r12", "r13", "r14", "pc" };
106
107 /* initialize TCG globals. */
108 void arm_translate_init(void)
109 {
110 int i;
111
112 cpu_env = tcg_global_reg_new_ptr(TCG_AREG0, "env");
113
114 for (i = 0; i < 16; i++) {
115 cpu_R[i] = tcg_global_mem_new_i32(TCG_AREG0,
116 offsetof(CPUARMState, regs[i]),
117 regnames[i]);
118 }
119 cpu_CF = tcg_global_mem_new_i32(TCG_AREG0, offsetof(CPUARMState, CF), "CF");
120 cpu_NF = tcg_global_mem_new_i32(TCG_AREG0, offsetof(CPUARMState, NF), "NF");
121 cpu_VF = tcg_global_mem_new_i32(TCG_AREG0, offsetof(CPUARMState, VF), "VF");
122 cpu_ZF = tcg_global_mem_new_i32(TCG_AREG0, offsetof(CPUARMState, ZF), "ZF");
123
124 cpu_exclusive_addr = tcg_global_mem_new_i32(TCG_AREG0,
125 offsetof(CPUARMState, exclusive_addr), "exclusive_addr");
126 cpu_exclusive_val = tcg_global_mem_new_i32(TCG_AREG0,
127 offsetof(CPUARMState, exclusive_val), "exclusive_val");
128 cpu_exclusive_high = tcg_global_mem_new_i32(TCG_AREG0,
129 offsetof(CPUARMState, exclusive_high), "exclusive_high");
130 #ifdef CONFIG_USER_ONLY
131 cpu_exclusive_test = tcg_global_mem_new_i32(TCG_AREG0,
132 offsetof(CPUARMState, exclusive_test), "exclusive_test");
133 cpu_exclusive_info = tcg_global_mem_new_i32(TCG_AREG0,
134 offsetof(CPUARMState, exclusive_info), "exclusive_info");
135 #endif
136
137 #define GEN_HELPER 2
138 #include "helper.h"
139 }
140
141 static inline TCGv_i32 load_cpu_offset(int offset)
142 {
143 TCGv_i32 tmp = tcg_temp_new_i32();
144 tcg_gen_ld_i32(tmp, cpu_env, offset);
145 return tmp;
146 }
147
148 #define load_cpu_field(name) load_cpu_offset(offsetof(CPUARMState, name))
149
150 static inline void store_cpu_offset(TCGv_i32 var, int offset)
151 {
152 tcg_gen_st_i32(var, cpu_env, offset);
153 tcg_temp_free_i32(var);
154 }
155
156 #define store_cpu_field(var, name) \
157 store_cpu_offset(var, offsetof(CPUARMState, name))
158
159 /* Set a variable to the value of a CPU register. */
160 static void load_reg_var(DisasContext *s, TCGv_i32 var, int reg)
161 {
162 if (reg == 15) {
163 uint32_t addr;
164 /* normally, since we updated PC, we need only to add one insn */
165 if (s->thumb)
166 addr = (long)s->pc + 2;
167 else
168 addr = (long)s->pc + 4;
169 tcg_gen_movi_i32(var, addr);
170 } else {
171 tcg_gen_mov_i32(var, cpu_R[reg]);
172 }
173 }
174
175 /* Create a new temporary and set it to the value of a CPU register. */
176 static inline TCGv_i32 load_reg(DisasContext *s, int reg)
177 {
178 TCGv_i32 tmp = tcg_temp_new_i32();
179 load_reg_var(s, tmp, reg);
180 return tmp;
181 }
182
183 /* Set a CPU register. The source must be a temporary and will be
184 marked as dead. */
185 static void store_reg(DisasContext *s, int reg, TCGv_i32 var)
186 {
187 if (reg == 15) {
188 tcg_gen_andi_i32(var, var, ~1);
189 s->is_jmp = DISAS_JUMP;
190 }
191 tcg_gen_mov_i32(cpu_R[reg], var);
192 tcg_temp_free_i32(var);
193 }
194
195 /* Value extensions. */
196 #define gen_uxtb(var) tcg_gen_ext8u_i32(var, var)
197 #define gen_uxth(var) tcg_gen_ext16u_i32(var, var)
198 #define gen_sxtb(var) tcg_gen_ext8s_i32(var, var)
199 #define gen_sxth(var) tcg_gen_ext16s_i32(var, var)
200
201 #define gen_sxtb16(var) gen_helper_sxtb16(var, var)
202 #define gen_uxtb16(var) gen_helper_uxtb16(var, var)
203
204
205 static inline void gen_set_cpsr(TCGv_i32 var, uint32_t mask)
206 {
207 TCGv_i32 tmp_mask = tcg_const_i32(mask);
208 gen_helper_cpsr_write(cpu_env, var, tmp_mask);
209 tcg_temp_free_i32(tmp_mask);
210 }
211 /* Set NZCV flags from the high 4 bits of var. */
212 #define gen_set_nzcv(var) gen_set_cpsr(var, CPSR_NZCV)
213
214 static void gen_exception(int excp)
215 {
216 TCGv_i32 tmp = tcg_temp_new_i32();
217 tcg_gen_movi_i32(tmp, excp);
218 gen_helper_exception(cpu_env, tmp);
219 tcg_temp_free_i32(tmp);
220 }
221
222 static void gen_smul_dual(TCGv_i32 a, TCGv_i32 b)
223 {
224 TCGv_i32 tmp1 = tcg_temp_new_i32();
225 TCGv_i32 tmp2 = tcg_temp_new_i32();
226 tcg_gen_ext16s_i32(tmp1, a);
227 tcg_gen_ext16s_i32(tmp2, b);
228 tcg_gen_mul_i32(tmp1, tmp1, tmp2);
229 tcg_temp_free_i32(tmp2);
230 tcg_gen_sari_i32(a, a, 16);
231 tcg_gen_sari_i32(b, b, 16);
232 tcg_gen_mul_i32(b, b, a);
233 tcg_gen_mov_i32(a, tmp1);
234 tcg_temp_free_i32(tmp1);
235 }
236
237 /* Byteswap each halfword. */
238 static void gen_rev16(TCGv_i32 var)
239 {
240 TCGv_i32 tmp = tcg_temp_new_i32();
241 tcg_gen_shri_i32(tmp, var, 8);
242 tcg_gen_andi_i32(tmp, tmp, 0x00ff00ff);
243 tcg_gen_shli_i32(var, var, 8);
244 tcg_gen_andi_i32(var, var, 0xff00ff00);
245 tcg_gen_or_i32(var, var, tmp);
246 tcg_temp_free_i32(tmp);
247 }
248
249 /* Byteswap low halfword and sign extend. */
250 static void gen_revsh(TCGv_i32 var)
251 {
252 tcg_gen_ext16u_i32(var, var);
253 tcg_gen_bswap16_i32(var, var);
254 tcg_gen_ext16s_i32(var, var);
255 }
256
257 /* Unsigned bitfield extract. */
258 static void gen_ubfx(TCGv_i32 var, int shift, uint32_t mask)
259 {
260 if (shift)
261 tcg_gen_shri_i32(var, var, shift);
262 tcg_gen_andi_i32(var, var, mask);
263 }
264
265 /* Signed bitfield extract. */
266 static void gen_sbfx(TCGv_i32 var, int shift, int width)
267 {
268 uint32_t signbit;
269
270 if (shift)
271 tcg_gen_sari_i32(var, var, shift);
272 if (shift + width < 32) {
273 signbit = 1u << (width - 1);
274 tcg_gen_andi_i32(var, var, (1u << width) - 1);
275 tcg_gen_xori_i32(var, var, signbit);
276 tcg_gen_subi_i32(var, var, signbit);
277 }
278 }
279
280 /* Return (b << 32) + a. Mark inputs as dead */
281 static TCGv_i64 gen_addq_msw(TCGv_i64 a, TCGv_i32 b)
282 {
283 TCGv_i64 tmp64 = tcg_temp_new_i64();
284
285 tcg_gen_extu_i32_i64(tmp64, b);
286 tcg_temp_free_i32(b);
287 tcg_gen_shli_i64(tmp64, tmp64, 32);
288 tcg_gen_add_i64(a, tmp64, a);
289
290 tcg_temp_free_i64(tmp64);
291 return a;
292 }
293
294 /* Return (b << 32) - a. Mark inputs as dead. */
295 static TCGv_i64 gen_subq_msw(TCGv_i64 a, TCGv_i32 b)
296 {
297 TCGv_i64 tmp64 = tcg_temp_new_i64();
298
299 tcg_gen_extu_i32_i64(tmp64, b);
300 tcg_temp_free_i32(b);
301 tcg_gen_shli_i64(tmp64, tmp64, 32);
302 tcg_gen_sub_i64(a, tmp64, a);
303
304 tcg_temp_free_i64(tmp64);
305 return a;
306 }
307
308 /* 32x32->64 multiply. Marks inputs as dead. */
309 static TCGv_i64 gen_mulu_i64_i32(TCGv_i32 a, TCGv_i32 b)
310 {
311 TCGv_i32 lo = tcg_temp_new_i32();
312 TCGv_i32 hi = tcg_temp_new_i32();
313 TCGv_i64 ret;
314
315 tcg_gen_mulu2_i32(lo, hi, a, b);
316 tcg_temp_free_i32(a);
317 tcg_temp_free_i32(b);
318
319 ret = tcg_temp_new_i64();
320 tcg_gen_concat_i32_i64(ret, lo, hi);
321 tcg_temp_free_i32(lo);
322 tcg_temp_free_i32(hi);
323
324 return ret;
325 }
326
327 static TCGv_i64 gen_muls_i64_i32(TCGv_i32 a, TCGv_i32 b)
328 {
329 TCGv_i32 lo = tcg_temp_new_i32();
330 TCGv_i32 hi = tcg_temp_new_i32();
331 TCGv_i64 ret;
332
333 tcg_gen_muls2_i32(lo, hi, a, b);
334 tcg_temp_free_i32(a);
335 tcg_temp_free_i32(b);
336
337 ret = tcg_temp_new_i64();
338 tcg_gen_concat_i32_i64(ret, lo, hi);
339 tcg_temp_free_i32(lo);
340 tcg_temp_free_i32(hi);
341
342 return ret;
343 }
344
345 /* Swap low and high halfwords. */
346 static void gen_swap_half(TCGv_i32 var)
347 {
348 TCGv_i32 tmp = tcg_temp_new_i32();
349 tcg_gen_shri_i32(tmp, var, 16);
350 tcg_gen_shli_i32(var, var, 16);
351 tcg_gen_or_i32(var, var, tmp);
352 tcg_temp_free_i32(tmp);
353 }
354
355 /* Dual 16-bit add. Result placed in t0 and t1 is marked as dead.
356 tmp = (t0 ^ t1) & 0x8000;
357 t0 &= ~0x8000;
358 t1 &= ~0x8000;
359 t0 = (t0 + t1) ^ tmp;
360 */
361
362 static void gen_add16(TCGv_i32 t0, TCGv_i32 t1)
363 {
364 TCGv_i32 tmp = tcg_temp_new_i32();
365 tcg_gen_xor_i32(tmp, t0, t1);
366 tcg_gen_andi_i32(tmp, tmp, 0x8000);
367 tcg_gen_andi_i32(t0, t0, ~0x8000);
368 tcg_gen_andi_i32(t1, t1, ~0x8000);
369 tcg_gen_add_i32(t0, t0, t1);
370 tcg_gen_xor_i32(t0, t0, tmp);
371 tcg_temp_free_i32(tmp);
372 tcg_temp_free_i32(t1);
373 }
374
375 /* Set CF to the top bit of var. */
376 static void gen_set_CF_bit31(TCGv_i32 var)
377 {
378 tcg_gen_shri_i32(cpu_CF, var, 31);
379 }
380
381 /* Set N and Z flags from var. */
382 static inline void gen_logic_CC(TCGv_i32 var)
383 {
384 tcg_gen_mov_i32(cpu_NF, var);
385 tcg_gen_mov_i32(cpu_ZF, var);
386 }
387
388 /* T0 += T1 + CF. */
389 static void gen_adc(TCGv_i32 t0, TCGv_i32 t1)
390 {
391 tcg_gen_add_i32(t0, t0, t1);
392 tcg_gen_add_i32(t0, t0, cpu_CF);
393 }
394
395 /* dest = T0 + T1 + CF. */
396 static void gen_add_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
397 {
398 tcg_gen_add_i32(dest, t0, t1);
399 tcg_gen_add_i32(dest, dest, cpu_CF);
400 }
401
402 /* dest = T0 - T1 + CF - 1. */
403 static void gen_sub_carry(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
404 {
405 tcg_gen_sub_i32(dest, t0, t1);
406 tcg_gen_add_i32(dest, dest, cpu_CF);
407 tcg_gen_subi_i32(dest, dest, 1);
408 }
409
410 /* dest = T0 + T1. Compute C, N, V and Z flags */
411 static void gen_add_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
412 {
413 TCGv_i32 tmp = tcg_temp_new_i32();
414 tcg_gen_movi_i32(tmp, 0);
415 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, t1, tmp);
416 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
417 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
418 tcg_gen_xor_i32(tmp, t0, t1);
419 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
420 tcg_temp_free_i32(tmp);
421 tcg_gen_mov_i32(dest, cpu_NF);
422 }
423
424 /* dest = T0 + T1 + CF. Compute C, N, V and Z flags */
425 static void gen_adc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
426 {
427 TCGv_i32 tmp = tcg_temp_new_i32();
428 if (TCG_TARGET_HAS_add2_i32) {
429 tcg_gen_movi_i32(tmp, 0);
430 tcg_gen_add2_i32(cpu_NF, cpu_CF, t0, tmp, cpu_CF, tmp);
431 tcg_gen_add2_i32(cpu_NF, cpu_CF, cpu_NF, cpu_CF, t1, tmp);
432 } else {
433 TCGv_i64 q0 = tcg_temp_new_i64();
434 TCGv_i64 q1 = tcg_temp_new_i64();
435 tcg_gen_extu_i32_i64(q0, t0);
436 tcg_gen_extu_i32_i64(q1, t1);
437 tcg_gen_add_i64(q0, q0, q1);
438 tcg_gen_extu_i32_i64(q1, cpu_CF);
439 tcg_gen_add_i64(q0, q0, q1);
440 tcg_gen_extr_i64_i32(cpu_NF, cpu_CF, q0);
441 tcg_temp_free_i64(q0);
442 tcg_temp_free_i64(q1);
443 }
444 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
445 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
446 tcg_gen_xor_i32(tmp, t0, t1);
447 tcg_gen_andc_i32(cpu_VF, cpu_VF, tmp);
448 tcg_temp_free_i32(tmp);
449 tcg_gen_mov_i32(dest, cpu_NF);
450 }
451
452 /* dest = T0 - T1. Compute C, N, V and Z flags */
453 static void gen_sub_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
454 {
455 TCGv_i32 tmp;
456 tcg_gen_sub_i32(cpu_NF, t0, t1);
457 tcg_gen_mov_i32(cpu_ZF, cpu_NF);
458 tcg_gen_setcond_i32(TCG_COND_GEU, cpu_CF, t0, t1);
459 tcg_gen_xor_i32(cpu_VF, cpu_NF, t0);
460 tmp = tcg_temp_new_i32();
461 tcg_gen_xor_i32(tmp, t0, t1);
462 tcg_gen_and_i32(cpu_VF, cpu_VF, tmp);
463 tcg_temp_free_i32(tmp);
464 tcg_gen_mov_i32(dest, cpu_NF);
465 }
466
467 /* dest = T0 + ~T1 + CF. Compute C, N, V and Z flags */
468 static void gen_sbc_CC(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
469 {
470 TCGv_i32 tmp = tcg_temp_new_i32();
471 tcg_gen_not_i32(tmp, t1);
472 gen_adc_CC(dest, t0, tmp);
473 tcg_temp_free_i32(tmp);
474 }
475
476 #define GEN_SHIFT(name) \
477 static void gen_##name(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1) \
478 { \
479 TCGv_i32 tmp1, tmp2, tmp3; \
480 tmp1 = tcg_temp_new_i32(); \
481 tcg_gen_andi_i32(tmp1, t1, 0xff); \
482 tmp2 = tcg_const_i32(0); \
483 tmp3 = tcg_const_i32(0x1f); \
484 tcg_gen_movcond_i32(TCG_COND_GTU, tmp2, tmp1, tmp3, tmp2, t0); \
485 tcg_temp_free_i32(tmp3); \
486 tcg_gen_andi_i32(tmp1, tmp1, 0x1f); \
487 tcg_gen_##name##_i32(dest, tmp2, tmp1); \
488 tcg_temp_free_i32(tmp2); \
489 tcg_temp_free_i32(tmp1); \
490 }
491 GEN_SHIFT(shl)
492 GEN_SHIFT(shr)
493 #undef GEN_SHIFT
494
495 static void gen_sar(TCGv_i32 dest, TCGv_i32 t0, TCGv_i32 t1)
496 {
497 TCGv_i32 tmp1, tmp2;
498 tmp1 = tcg_temp_new_i32();
499 tcg_gen_andi_i32(tmp1, t1, 0xff);
500 tmp2 = tcg_const_i32(0x1f);
501 tcg_gen_movcond_i32(TCG_COND_GTU, tmp1, tmp1, tmp2, tmp2, tmp1);
502 tcg_temp_free_i32(tmp2);
503 tcg_gen_sar_i32(dest, t0, tmp1);
504 tcg_temp_free_i32(tmp1);
505 }
506
507 static void tcg_gen_abs_i32(TCGv_i32 dest, TCGv_i32 src)
508 {
509 TCGv_i32 c0 = tcg_const_i32(0);
510 TCGv_i32 tmp = tcg_temp_new_i32();
511 tcg_gen_neg_i32(tmp, src);
512 tcg_gen_movcond_i32(TCG_COND_GT, dest, src, c0, src, tmp);
513 tcg_temp_free_i32(c0);
514 tcg_temp_free_i32(tmp);
515 }
516
517 static void shifter_out_im(TCGv_i32 var, int shift)
518 {
519 if (shift == 0) {
520 tcg_gen_andi_i32(cpu_CF, var, 1);
521 } else {
522 tcg_gen_shri_i32(cpu_CF, var, shift);
523 if (shift != 31) {
524 tcg_gen_andi_i32(cpu_CF, cpu_CF, 1);
525 }
526 }
527 }
528
529 /* Shift by immediate. Includes special handling for shift == 0. */
530 static inline void gen_arm_shift_im(TCGv_i32 var, int shiftop,
531 int shift, int flags)
532 {
533 switch (shiftop) {
534 case 0: /* LSL */
535 if (shift != 0) {
536 if (flags)
537 shifter_out_im(var, 32 - shift);
538 tcg_gen_shli_i32(var, var, shift);
539 }
540 break;
541 case 1: /* LSR */
542 if (shift == 0) {
543 if (flags) {
544 tcg_gen_shri_i32(cpu_CF, var, 31);
545 }
546 tcg_gen_movi_i32(var, 0);
547 } else {
548 if (flags)
549 shifter_out_im(var, shift - 1);
550 tcg_gen_shri_i32(var, var, shift);
551 }
552 break;
553 case 2: /* ASR */
554 if (shift == 0)
555 shift = 32;
556 if (flags)
557 shifter_out_im(var, shift - 1);
558 if (shift == 32)
559 shift = 31;
560 tcg_gen_sari_i32(var, var, shift);
561 break;
562 case 3: /* ROR/RRX */
563 if (shift != 0) {
564 if (flags)
565 shifter_out_im(var, shift - 1);
566 tcg_gen_rotri_i32(var, var, shift); break;
567 } else {
568 TCGv_i32 tmp = tcg_temp_new_i32();
569 tcg_gen_shli_i32(tmp, cpu_CF, 31);
570 if (flags)
571 shifter_out_im(var, 0);
572 tcg_gen_shri_i32(var, var, 1);
573 tcg_gen_or_i32(var, var, tmp);
574 tcg_temp_free_i32(tmp);
575 }
576 }
577 };
578
579 static inline void gen_arm_shift_reg(TCGv_i32 var, int shiftop,
580 TCGv_i32 shift, int flags)
581 {
582 if (flags) {
583 switch (shiftop) {
584 case 0: gen_helper_shl_cc(var, cpu_env, var, shift); break;
585 case 1: gen_helper_shr_cc(var, cpu_env, var, shift); break;
586 case 2: gen_helper_sar_cc(var, cpu_env, var, shift); break;
587 case 3: gen_helper_ror_cc(var, cpu_env, var, shift); break;
588 }
589 } else {
590 switch (shiftop) {
591 case 0:
592 gen_shl(var, var, shift);
593 break;
594 case 1:
595 gen_shr(var, var, shift);
596 break;
597 case 2:
598 gen_sar(var, var, shift);
599 break;
600 case 3: tcg_gen_andi_i32(shift, shift, 0x1f);
601 tcg_gen_rotr_i32(var, var, shift); break;
602 }
603 }
604 tcg_temp_free_i32(shift);
605 }
606
607 #define PAS_OP(pfx) \
608 switch (op2) { \
609 case 0: gen_pas_helper(glue(pfx,add16)); break; \
610 case 1: gen_pas_helper(glue(pfx,addsubx)); break; \
611 case 2: gen_pas_helper(glue(pfx,subaddx)); break; \
612 case 3: gen_pas_helper(glue(pfx,sub16)); break; \
613 case 4: gen_pas_helper(glue(pfx,add8)); break; \
614 case 7: gen_pas_helper(glue(pfx,sub8)); break; \
615 }
616 static void gen_arm_parallel_addsub(int op1, int op2, TCGv_i32 a, TCGv_i32 b)
617 {
618 TCGv_ptr tmp;
619
620 switch (op1) {
621 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b, tmp)
622 case 1:
623 tmp = tcg_temp_new_ptr();
624 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
625 PAS_OP(s)
626 tcg_temp_free_ptr(tmp);
627 break;
628 case 5:
629 tmp = tcg_temp_new_ptr();
630 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
631 PAS_OP(u)
632 tcg_temp_free_ptr(tmp);
633 break;
634 #undef gen_pas_helper
635 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b)
636 case 2:
637 PAS_OP(q);
638 break;
639 case 3:
640 PAS_OP(sh);
641 break;
642 case 6:
643 PAS_OP(uq);
644 break;
645 case 7:
646 PAS_OP(uh);
647 break;
648 #undef gen_pas_helper
649 }
650 }
651 #undef PAS_OP
652
653 /* For unknown reasons Arm and Thumb-2 use arbitrarily different encodings. */
654 #define PAS_OP(pfx) \
655 switch (op1) { \
656 case 0: gen_pas_helper(glue(pfx,add8)); break; \
657 case 1: gen_pas_helper(glue(pfx,add16)); break; \
658 case 2: gen_pas_helper(glue(pfx,addsubx)); break; \
659 case 4: gen_pas_helper(glue(pfx,sub8)); break; \
660 case 5: gen_pas_helper(glue(pfx,sub16)); break; \
661 case 6: gen_pas_helper(glue(pfx,subaddx)); break; \
662 }
663 static void gen_thumb2_parallel_addsub(int op1, int op2, TCGv_i32 a, TCGv_i32 b)
664 {
665 TCGv_ptr tmp;
666
667 switch (op2) {
668 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b, tmp)
669 case 0:
670 tmp = tcg_temp_new_ptr();
671 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
672 PAS_OP(s)
673 tcg_temp_free_ptr(tmp);
674 break;
675 case 4:
676 tmp = tcg_temp_new_ptr();
677 tcg_gen_addi_ptr(tmp, cpu_env, offsetof(CPUARMState, GE));
678 PAS_OP(u)
679 tcg_temp_free_ptr(tmp);
680 break;
681 #undef gen_pas_helper
682 #define gen_pas_helper(name) glue(gen_helper_,name)(a, a, b)
683 case 1:
684 PAS_OP(q);
685 break;
686 case 2:
687 PAS_OP(sh);
688 break;
689 case 5:
690 PAS_OP(uq);
691 break;
692 case 6:
693 PAS_OP(uh);
694 break;
695 #undef gen_pas_helper
696 }
697 }
698 #undef PAS_OP
699
700 static void gen_test_cc(int cc, int label)
701 {
702 TCGv_i32 tmp;
703 int inv;
704
705 switch (cc) {
706 case 0: /* eq: Z */
707 tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_ZF, 0, label);
708 break;
709 case 1: /* ne: !Z */
710 tcg_gen_brcondi_i32(TCG_COND_NE, cpu_ZF, 0, label);
711 break;
712 case 2: /* cs: C */
713 tcg_gen_brcondi_i32(TCG_COND_NE, cpu_CF, 0, label);
714 break;
715 case 3: /* cc: !C */
716 tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_CF, 0, label);
717 break;
718 case 4: /* mi: N */
719 tcg_gen_brcondi_i32(TCG_COND_LT, cpu_NF, 0, label);
720 break;
721 case 5: /* pl: !N */
722 tcg_gen_brcondi_i32(TCG_COND_GE, cpu_NF, 0, label);
723 break;
724 case 6: /* vs: V */
725 tcg_gen_brcondi_i32(TCG_COND_LT, cpu_VF, 0, label);
726 break;
727 case 7: /* vc: !V */
728 tcg_gen_brcondi_i32(TCG_COND_GE, cpu_VF, 0, label);
729 break;
730 case 8: /* hi: C && !Z */
731 inv = gen_new_label();
732 tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_CF, 0, inv);
733 tcg_gen_brcondi_i32(TCG_COND_NE, cpu_ZF, 0, label);
734 gen_set_label(inv);
735 break;
736 case 9: /* ls: !C || Z */
737 tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_CF, 0, label);
738 tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_ZF, 0, label);
739 break;
740 case 10: /* ge: N == V -> N ^ V == 0 */
741 tmp = tcg_temp_new_i32();
742 tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
743 tcg_gen_brcondi_i32(TCG_COND_GE, tmp, 0, label);
744 tcg_temp_free_i32(tmp);
745 break;
746 case 11: /* lt: N != V -> N ^ V != 0 */
747 tmp = tcg_temp_new_i32();
748 tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
749 tcg_gen_brcondi_i32(TCG_COND_LT, tmp, 0, label);
750 tcg_temp_free_i32(tmp);
751 break;
752 case 12: /* gt: !Z && N == V */
753 inv = gen_new_label();
754 tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_ZF, 0, inv);
755 tmp = tcg_temp_new_i32();
756 tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
757 tcg_gen_brcondi_i32(TCG_COND_GE, tmp, 0, label);
758 tcg_temp_free_i32(tmp);
759 gen_set_label(inv);
760 break;
761 case 13: /* le: Z || N != V */
762 tcg_gen_brcondi_i32(TCG_COND_EQ, cpu_ZF, 0, label);
763 tmp = tcg_temp_new_i32();
764 tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF);
765 tcg_gen_brcondi_i32(TCG_COND_LT, tmp, 0, label);
766 tcg_temp_free_i32(tmp);
767 break;
768 default:
769 fprintf(stderr, "Bad condition code 0x%x\n", cc);
770 abort();
771 }
772 }
773
774 static const uint8_t table_logic_cc[16] = {
775 1, /* and */
776 1, /* xor */
777 0, /* sub */
778 0, /* rsb */
779 0, /* add */
780 0, /* adc */
781 0, /* sbc */
782 0, /* rsc */
783 1, /* andl */
784 1, /* xorl */
785 0, /* cmp */
786 0, /* cmn */
787 1, /* orr */
788 1, /* mov */
789 1, /* bic */
790 1, /* mvn */
791 };
792
793 /* Set PC and Thumb state from an immediate address. */
794 static inline void gen_bx_im(DisasContext *s, uint32_t addr)
795 {
796 TCGv_i32 tmp;
797
798 s->is_jmp = DISAS_UPDATE;
799 if (s->thumb != (addr & 1)) {
800 tmp = tcg_temp_new_i32();
801 tcg_gen_movi_i32(tmp, addr & 1);
802 tcg_gen_st_i32(tmp, cpu_env, offsetof(CPUARMState, thumb));
803 tcg_temp_free_i32(tmp);
804 }
805 tcg_gen_movi_i32(cpu_R[15], addr & ~1);
806 }
807
808 /* Set PC and Thumb state from var. var is marked as dead. */
809 static inline void gen_bx(DisasContext *s, TCGv_i32 var)
810 {
811 s->is_jmp = DISAS_UPDATE;
812 tcg_gen_andi_i32(cpu_R[15], var, ~1);
813 tcg_gen_andi_i32(var, var, 1);
814 store_cpu_field(var, thumb);
815 }
816
817 /* Variant of store_reg which uses branch&exchange logic when storing
818 to r15 in ARM architecture v7 and above. The source must be a temporary
819 and will be marked as dead. */
820 static inline void store_reg_bx(CPUARMState *env, DisasContext *s,
821 int reg, TCGv_i32 var)
822 {
823 if (reg == 15 && ENABLE_ARCH_7) {
824 gen_bx(s, var);
825 } else {
826 store_reg(s, reg, var);
827 }
828 }
829
830 /* Variant of store_reg which uses branch&exchange logic when storing
831 * to r15 in ARM architecture v5T and above. This is used for storing
832 * the results of a LDR/LDM/POP into r15, and corresponds to the cases
833 * in the ARM ARM which use the LoadWritePC() pseudocode function. */
834 static inline void store_reg_from_load(CPUARMState *env, DisasContext *s,
835 int reg, TCGv_i32 var)
836 {
837 if (reg == 15 && ENABLE_ARCH_5) {
838 gen_bx(s, var);
839 } else {
840 store_reg(s, reg, var);
841 }
842 }
843
844 static inline void gen_set_pc_im(uint32_t val)
845 {
846 tcg_gen_movi_i32(cpu_R[15], val);
847 }
848
849 /* Force a TB lookup after an instruction that changes the CPU state. */
850 static inline void gen_lookup_tb(DisasContext *s)
851 {
852 tcg_gen_movi_i32(cpu_R[15], s->pc & ~1);
853 s->is_jmp = DISAS_UPDATE;
854 }
855
856 static inline void gen_add_data_offset(DisasContext *s, unsigned int insn,
857 TCGv_i32 var)
858 {
859 int val, rm, shift, shiftop;
860 TCGv_i32 offset;
861
862 if (!(insn & (1 << 25))) {
863 /* immediate */
864 val = insn & 0xfff;
865 if (!(insn & (1 << 23)))
866 val = -val;
867 if (val != 0)
868 tcg_gen_addi_i32(var, var, val);
869 } else {
870 /* shift/register */
871 rm = (insn) & 0xf;
872 shift = (insn >> 7) & 0x1f;
873 shiftop = (insn >> 5) & 3;
874 offset = load_reg(s, rm);
875 gen_arm_shift_im(offset, shiftop, shift, 0);
876 if (!(insn & (1 << 23)))
877 tcg_gen_sub_i32(var, var, offset);
878 else
879 tcg_gen_add_i32(var, var, offset);
880 tcg_temp_free_i32(offset);
881 }
882 }
883
884 static inline void gen_add_datah_offset(DisasContext *s, unsigned int insn,
885 int extra, TCGv_i32 var)
886 {
887 int val, rm;
888 TCGv_i32 offset;
889
890 if (insn & (1 << 22)) {
891 /* immediate */
892 val = (insn & 0xf) | ((insn >> 4) & 0xf0);
893 if (!(insn & (1 << 23)))
894 val = -val;
895 val += extra;
896 if (val != 0)
897 tcg_gen_addi_i32(var, var, val);
898 } else {
899 /* register */
900 if (extra)
901 tcg_gen_addi_i32(var, var, extra);
902 rm = (insn) & 0xf;
903 offset = load_reg(s, rm);
904 if (!(insn & (1 << 23)))
905 tcg_gen_sub_i32(var, var, offset);
906 else
907 tcg_gen_add_i32(var, var, offset);
908 tcg_temp_free_i32(offset);
909 }
910 }
911
912 static TCGv_ptr get_fpstatus_ptr(int neon)
913 {
914 TCGv_ptr statusptr = tcg_temp_new_ptr();
915 int offset;
916 if (neon) {
917 offset = offsetof(CPUARMState, vfp.standard_fp_status);
918 } else {
919 offset = offsetof(CPUARMState, vfp.fp_status);
920 }
921 tcg_gen_addi_ptr(statusptr, cpu_env, offset);
922 return statusptr;
923 }
924
925 #define VFP_OP2(name) \
926 static inline void gen_vfp_##name(int dp) \
927 { \
928 TCGv_ptr fpst = get_fpstatus_ptr(0); \
929 if (dp) { \
930 gen_helper_vfp_##name##d(cpu_F0d, cpu_F0d, cpu_F1d, fpst); \
931 } else { \
932 gen_helper_vfp_##name##s(cpu_F0s, cpu_F0s, cpu_F1s, fpst); \
933 } \
934 tcg_temp_free_ptr(fpst); \
935 }
936
937 VFP_OP2(add)
938 VFP_OP2(sub)
939 VFP_OP2(mul)
940 VFP_OP2(div)
941
942 #undef VFP_OP2
943
944 static inline void gen_vfp_F1_mul(int dp)
945 {
946 /* Like gen_vfp_mul() but put result in F1 */
947 TCGv_ptr fpst = get_fpstatus_ptr(0);
948 if (dp) {
949 gen_helper_vfp_muld(cpu_F1d, cpu_F0d, cpu_F1d, fpst);
950 } else {
951 gen_helper_vfp_muls(cpu_F1s, cpu_F0s, cpu_F1s, fpst);
952 }
953 tcg_temp_free_ptr(fpst);
954 }
955
956 static inline void gen_vfp_F1_neg(int dp)
957 {
958 /* Like gen_vfp_neg() but put result in F1 */
959 if (dp) {
960 gen_helper_vfp_negd(cpu_F1d, cpu_F0d);
961 } else {
962 gen_helper_vfp_negs(cpu_F1s, cpu_F0s);
963 }
964 }
965
966 static inline void gen_vfp_abs(int dp)
967 {
968 if (dp)
969 gen_helper_vfp_absd(cpu_F0d, cpu_F0d);
970 else
971 gen_helper_vfp_abss(cpu_F0s, cpu_F0s);
972 }
973
974 static inline void gen_vfp_neg(int dp)
975 {
976 if (dp)
977 gen_helper_vfp_negd(cpu_F0d, cpu_F0d);
978 else
979 gen_helper_vfp_negs(cpu_F0s, cpu_F0s);
980 }
981
982 static inline void gen_vfp_sqrt(int dp)
983 {
984 if (dp)
985 gen_helper_vfp_sqrtd(cpu_F0d, cpu_F0d, cpu_env);
986 else
987 gen_helper_vfp_sqrts(cpu_F0s, cpu_F0s, cpu_env);
988 }
989
990 static inline void gen_vfp_cmp(int dp)
991 {
992 if (dp)
993 gen_helper_vfp_cmpd(cpu_F0d, cpu_F1d, cpu_env);
994 else
995 gen_helper_vfp_cmps(cpu_F0s, cpu_F1s, cpu_env);
996 }
997
998 static inline void gen_vfp_cmpe(int dp)
999 {
1000 if (dp)
1001 gen_helper_vfp_cmped(cpu_F0d, cpu_F1d, cpu_env);
1002 else
1003 gen_helper_vfp_cmpes(cpu_F0s, cpu_F1s, cpu_env);
1004 }
1005
1006 static inline void gen_vfp_F1_ld0(int dp)
1007 {
1008 if (dp)
1009 tcg_gen_movi_i64(cpu_F1d, 0);
1010 else
1011 tcg_gen_movi_i32(cpu_F1s, 0);
1012 }
1013
1014 #define VFP_GEN_ITOF(name) \
1015 static inline void gen_vfp_##name(int dp, int neon) \
1016 { \
1017 TCGv_ptr statusptr = get_fpstatus_ptr(neon); \
1018 if (dp) { \
1019 gen_helper_vfp_##name##d(cpu_F0d, cpu_F0s, statusptr); \
1020 } else { \
1021 gen_helper_vfp_##name##s(cpu_F0s, cpu_F0s, statusptr); \
1022 } \
1023 tcg_temp_free_ptr(statusptr); \
1024 }
1025
1026 VFP_GEN_ITOF(uito)
1027 VFP_GEN_ITOF(sito)
1028 #undef VFP_GEN_ITOF
1029
1030 #define VFP_GEN_FTOI(name) \
1031 static inline void gen_vfp_##name(int dp, int neon) \
1032 { \
1033 TCGv_ptr statusptr = get_fpstatus_ptr(neon); \
1034 if (dp) { \
1035 gen_helper_vfp_##name##d(cpu_F0s, cpu_F0d, statusptr); \
1036 } else { \
1037 gen_helper_vfp_##name##s(cpu_F0s, cpu_F0s, statusptr); \
1038 } \
1039 tcg_temp_free_ptr(statusptr); \
1040 }
1041
1042 VFP_GEN_FTOI(toui)
1043 VFP_GEN_FTOI(touiz)
1044 VFP_GEN_FTOI(tosi)
1045 VFP_GEN_FTOI(tosiz)
1046 #undef VFP_GEN_FTOI
1047
1048 #define VFP_GEN_FIX(name) \
1049 static inline void gen_vfp_##name(int dp, int shift, int neon) \
1050 { \
1051 TCGv_i32 tmp_shift = tcg_const_i32(shift); \
1052 TCGv_ptr statusptr = get_fpstatus_ptr(neon); \
1053 if (dp) { \
1054 gen_helper_vfp_##name##d(cpu_F0d, cpu_F0d, tmp_shift, statusptr); \
1055 } else { \
1056 gen_helper_vfp_##name##s(cpu_F0s, cpu_F0s, tmp_shift, statusptr); \
1057 } \
1058 tcg_temp_free_i32(tmp_shift); \
1059 tcg_temp_free_ptr(statusptr); \
1060 }
1061 VFP_GEN_FIX(tosh)
1062 VFP_GEN_FIX(tosl)
1063 VFP_GEN_FIX(touh)
1064 VFP_GEN_FIX(toul)
1065 VFP_GEN_FIX(shto)
1066 VFP_GEN_FIX(slto)
1067 VFP_GEN_FIX(uhto)
1068 VFP_GEN_FIX(ulto)
1069 #undef VFP_GEN_FIX
1070
1071 static inline void gen_vfp_ld(DisasContext *s, int dp, TCGv_i32 addr)
1072 {
1073 if (dp)
1074 tcg_gen_qemu_ld64(cpu_F0d, addr, IS_USER(s));
1075 else
1076 tcg_gen_qemu_ld32u(cpu_F0s, addr, IS_USER(s));
1077 }
1078
1079 static inline void gen_vfp_st(DisasContext *s, int dp, TCGv_i32 addr)
1080 {
1081 if (dp)
1082 tcg_gen_qemu_st64(cpu_F0d, addr, IS_USER(s));
1083 else
1084 tcg_gen_qemu_st32(cpu_F0s, addr, IS_USER(s));
1085 }
1086
1087 static inline long
1088 vfp_reg_offset (int dp, int reg)
1089 {
1090 if (dp)
1091 return offsetof(CPUARMState, vfp.regs[reg]);
1092 else if (reg & 1) {
1093 return offsetof(CPUARMState, vfp.regs[reg >> 1])
1094 + offsetof(CPU_DoubleU, l.upper);
1095 } else {
1096 return offsetof(CPUARMState, vfp.regs[reg >> 1])
1097 + offsetof(CPU_DoubleU, l.lower);
1098 }
1099 }
1100
1101 /* Return the offset of a 32-bit piece of a NEON register.
1102 zero is the least significant end of the register. */
1103 static inline long
1104 neon_reg_offset (int reg, int n)
1105 {
1106 int sreg;
1107 sreg = reg * 2 + n;
1108 return vfp_reg_offset(0, sreg);
1109 }
1110
1111 static TCGv_i32 neon_load_reg(int reg, int pass)
1112 {
1113 TCGv_i32 tmp = tcg_temp_new_i32();
1114 tcg_gen_ld_i32(tmp, cpu_env, neon_reg_offset(reg, pass));
1115 return tmp;
1116 }
1117
1118 static void neon_store_reg(int reg, int pass, TCGv_i32 var)
1119 {
1120 tcg_gen_st_i32(var, cpu_env, neon_reg_offset(reg, pass));
1121 tcg_temp_free_i32(var);
1122 }
1123
1124 static inline void neon_load_reg64(TCGv_i64 var, int reg)
1125 {
1126 tcg_gen_ld_i64(var, cpu_env, vfp_reg_offset(1, reg));
1127 }
1128
1129 static inline void neon_store_reg64(TCGv_i64 var, int reg)
1130 {
1131 tcg_gen_st_i64(var, cpu_env, vfp_reg_offset(1, reg));
1132 }
1133
1134 #define tcg_gen_ld_f32 tcg_gen_ld_i32
1135 #define tcg_gen_ld_f64 tcg_gen_ld_i64
1136 #define tcg_gen_st_f32 tcg_gen_st_i32
1137 #define tcg_gen_st_f64 tcg_gen_st_i64
1138
1139 static inline void gen_mov_F0_vreg(int dp, int reg)
1140 {
1141 if (dp)
1142 tcg_gen_ld_f64(cpu_F0d, cpu_env, vfp_reg_offset(dp, reg));
1143 else
1144 tcg_gen_ld_f32(cpu_F0s, cpu_env, vfp_reg_offset(dp, reg));
1145 }
1146
1147 static inline void gen_mov_F1_vreg(int dp, int reg)
1148 {
1149 if (dp)
1150 tcg_gen_ld_f64(cpu_F1d, cpu_env, vfp_reg_offset(dp, reg));
1151 else
1152 tcg_gen_ld_f32(cpu_F1s, cpu_env, vfp_reg_offset(dp, reg));
1153 }
1154
1155 static inline void gen_mov_vreg_F0(int dp, int reg)
1156 {
1157 if (dp)
1158 tcg_gen_st_f64(cpu_F0d, cpu_env, vfp_reg_offset(dp, reg));
1159 else
1160 tcg_gen_st_f32(cpu_F0s, cpu_env, vfp_reg_offset(dp, reg));
1161 }
1162
1163 #define ARM_CP_RW_BIT (1 << 20)
1164
1165 static inline void iwmmxt_load_reg(TCGv_i64 var, int reg)
1166 {
1167 tcg_gen_ld_i64(var, cpu_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1168 }
1169
1170 static inline void iwmmxt_store_reg(TCGv_i64 var, int reg)
1171 {
1172 tcg_gen_st_i64(var, cpu_env, offsetof(CPUARMState, iwmmxt.regs[reg]));
1173 }
1174
1175 static inline TCGv_i32 iwmmxt_load_creg(int reg)
1176 {
1177 TCGv_i32 var = tcg_temp_new_i32();
1178 tcg_gen_ld_i32(var, cpu_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1179 return var;
1180 }
1181
1182 static inline void iwmmxt_store_creg(int reg, TCGv_i32 var)
1183 {
1184 tcg_gen_st_i32(var, cpu_env, offsetof(CPUARMState, iwmmxt.cregs[reg]));
1185 tcg_temp_free_i32(var);
1186 }
1187
1188 static inline void gen_op_iwmmxt_movq_wRn_M0(int rn)
1189 {
1190 iwmmxt_store_reg(cpu_M0, rn);
1191 }
1192
1193 static inline void gen_op_iwmmxt_movq_M0_wRn(int rn)
1194 {
1195 iwmmxt_load_reg(cpu_M0, rn);
1196 }
1197
1198 static inline void gen_op_iwmmxt_orq_M0_wRn(int rn)
1199 {
1200 iwmmxt_load_reg(cpu_V1, rn);
1201 tcg_gen_or_i64(cpu_M0, cpu_M0, cpu_V1);
1202 }
1203
1204 static inline void gen_op_iwmmxt_andq_M0_wRn(int rn)
1205 {
1206 iwmmxt_load_reg(cpu_V1, rn);
1207 tcg_gen_and_i64(cpu_M0, cpu_M0, cpu_V1);
1208 }
1209
1210 static inline void gen_op_iwmmxt_xorq_M0_wRn(int rn)
1211 {
1212 iwmmxt_load_reg(cpu_V1, rn);
1213 tcg_gen_xor_i64(cpu_M0, cpu_M0, cpu_V1);
1214 }
1215
1216 #define IWMMXT_OP(name) \
1217 static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1218 { \
1219 iwmmxt_load_reg(cpu_V1, rn); \
1220 gen_helper_iwmmxt_##name(cpu_M0, cpu_M0, cpu_V1); \
1221 }
1222
1223 #define IWMMXT_OP_ENV(name) \
1224 static inline void gen_op_iwmmxt_##name##_M0_wRn(int rn) \
1225 { \
1226 iwmmxt_load_reg(cpu_V1, rn); \
1227 gen_helper_iwmmxt_##name(cpu_M0, cpu_env, cpu_M0, cpu_V1); \
1228 }
1229
1230 #define IWMMXT_OP_ENV_SIZE(name) \
1231 IWMMXT_OP_ENV(name##b) \
1232 IWMMXT_OP_ENV(name##w) \
1233 IWMMXT_OP_ENV(name##l)
1234
1235 #define IWMMXT_OP_ENV1(name) \
1236 static inline void gen_op_iwmmxt_##name##_M0(void) \
1237 { \
1238 gen_helper_iwmmxt_##name(cpu_M0, cpu_env, cpu_M0); \
1239 }
1240
1241 IWMMXT_OP(maddsq)
1242 IWMMXT_OP(madduq)
1243 IWMMXT_OP(sadb)
1244 IWMMXT_OP(sadw)
1245 IWMMXT_OP(mulslw)
1246 IWMMXT_OP(mulshw)
1247 IWMMXT_OP(mululw)
1248 IWMMXT_OP(muluhw)
1249 IWMMXT_OP(macsw)
1250 IWMMXT_OP(macuw)
1251
1252 IWMMXT_OP_ENV_SIZE(unpackl)
1253 IWMMXT_OP_ENV_SIZE(unpackh)
1254
1255 IWMMXT_OP_ENV1(unpacklub)
1256 IWMMXT_OP_ENV1(unpackluw)
1257 IWMMXT_OP_ENV1(unpacklul)
1258 IWMMXT_OP_ENV1(unpackhub)
1259 IWMMXT_OP_ENV1(unpackhuw)
1260 IWMMXT_OP_ENV1(unpackhul)
1261 IWMMXT_OP_ENV1(unpacklsb)
1262 IWMMXT_OP_ENV1(unpacklsw)
1263 IWMMXT_OP_ENV1(unpacklsl)
1264 IWMMXT_OP_ENV1(unpackhsb)
1265 IWMMXT_OP_ENV1(unpackhsw)
1266 IWMMXT_OP_ENV1(unpackhsl)
1267
1268 IWMMXT_OP_ENV_SIZE(cmpeq)
1269 IWMMXT_OP_ENV_SIZE(cmpgtu)
1270 IWMMXT_OP_ENV_SIZE(cmpgts)
1271
1272 IWMMXT_OP_ENV_SIZE(mins)
1273 IWMMXT_OP_ENV_SIZE(minu)
1274 IWMMXT_OP_ENV_SIZE(maxs)
1275 IWMMXT_OP_ENV_SIZE(maxu)
1276
1277 IWMMXT_OP_ENV_SIZE(subn)
1278 IWMMXT_OP_ENV_SIZE(addn)
1279 IWMMXT_OP_ENV_SIZE(subu)
1280 IWMMXT_OP_ENV_SIZE(addu)
1281 IWMMXT_OP_ENV_SIZE(subs)
1282 IWMMXT_OP_ENV_SIZE(adds)
1283
1284 IWMMXT_OP_ENV(avgb0)
1285 IWMMXT_OP_ENV(avgb1)
1286 IWMMXT_OP_ENV(avgw0)
1287 IWMMXT_OP_ENV(avgw1)
1288
1289 IWMMXT_OP(msadb)
1290
1291 IWMMXT_OP_ENV(packuw)
1292 IWMMXT_OP_ENV(packul)
1293 IWMMXT_OP_ENV(packuq)
1294 IWMMXT_OP_ENV(packsw)
1295 IWMMXT_OP_ENV(packsl)
1296 IWMMXT_OP_ENV(packsq)
1297
1298 static void gen_op_iwmmxt_set_mup(void)
1299 {
1300 TCGv_i32 tmp;
1301 tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1302 tcg_gen_ori_i32(tmp, tmp, 2);
1303 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1304 }
1305
1306 static void gen_op_iwmmxt_set_cup(void)
1307 {
1308 TCGv_i32 tmp;
1309 tmp = load_cpu_field(iwmmxt.cregs[ARM_IWMMXT_wCon]);
1310 tcg_gen_ori_i32(tmp, tmp, 1);
1311 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCon]);
1312 }
1313
1314 static void gen_op_iwmmxt_setpsr_nz(void)
1315 {
1316 TCGv_i32 tmp = tcg_temp_new_i32();
1317 gen_helper_iwmmxt_setpsr_nz(tmp, cpu_M0);
1318 store_cpu_field(tmp, iwmmxt.cregs[ARM_IWMMXT_wCASF]);
1319 }
1320
1321 static inline void gen_op_iwmmxt_addl_M0_wRn(int rn)
1322 {
1323 iwmmxt_load_reg(cpu_V1, rn);
1324 tcg_gen_ext32u_i64(cpu_V1, cpu_V1);
1325 tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
1326 }
1327
1328 static inline int gen_iwmmxt_address(DisasContext *s, uint32_t insn,
1329 TCGv_i32 dest)
1330 {
1331 int rd;
1332 uint32_t offset;
1333 TCGv_i32 tmp;
1334
1335 rd = (insn >> 16) & 0xf;
1336 tmp = load_reg(s, rd);
1337
1338 offset = (insn & 0xff) << ((insn >> 7) & 2);
1339 if (insn & (1 << 24)) {
1340 /* Pre indexed */
1341 if (insn & (1 << 23))
1342 tcg_gen_addi_i32(tmp, tmp, offset);
1343 else
1344 tcg_gen_addi_i32(tmp, tmp, -offset);
1345 tcg_gen_mov_i32(dest, tmp);
1346 if (insn & (1 << 21))
1347 store_reg(s, rd, tmp);
1348 else
1349 tcg_temp_free_i32(tmp);
1350 } else if (insn & (1 << 21)) {
1351 /* Post indexed */
1352 tcg_gen_mov_i32(dest, tmp);
1353 if (insn & (1 << 23))
1354 tcg_gen_addi_i32(tmp, tmp, offset);
1355 else
1356 tcg_gen_addi_i32(tmp, tmp, -offset);
1357 store_reg(s, rd, tmp);
1358 } else if (!(insn & (1 << 23)))
1359 return 1;
1360 return 0;
1361 }
1362
1363 static inline int gen_iwmmxt_shift(uint32_t insn, uint32_t mask, TCGv_i32 dest)
1364 {
1365 int rd = (insn >> 0) & 0xf;
1366 TCGv_i32 tmp;
1367
1368 if (insn & (1 << 8)) {
1369 if (rd < ARM_IWMMXT_wCGR0 || rd > ARM_IWMMXT_wCGR3) {
1370 return 1;
1371 } else {
1372 tmp = iwmmxt_load_creg(rd);
1373 }
1374 } else {
1375 tmp = tcg_temp_new_i32();
1376 iwmmxt_load_reg(cpu_V0, rd);
1377 tcg_gen_trunc_i64_i32(tmp, cpu_V0);
1378 }
1379 tcg_gen_andi_i32(tmp, tmp, mask);
1380 tcg_gen_mov_i32(dest, tmp);
1381 tcg_temp_free_i32(tmp);
1382 return 0;
1383 }
1384
1385 /* Disassemble an iwMMXt instruction. Returns nonzero if an error occurred
1386 (ie. an undefined instruction). */
1387 static int disas_iwmmxt_insn(CPUARMState *env, DisasContext *s, uint32_t insn)
1388 {
1389 int rd, wrd;
1390 int rdhi, rdlo, rd0, rd1, i;
1391 TCGv_i32 addr;
1392 TCGv_i32 tmp, tmp2, tmp3;
1393
1394 if ((insn & 0x0e000e00) == 0x0c000000) {
1395 if ((insn & 0x0fe00ff0) == 0x0c400000) {
1396 wrd = insn & 0xf;
1397 rdlo = (insn >> 12) & 0xf;
1398 rdhi = (insn >> 16) & 0xf;
1399 if (insn & ARM_CP_RW_BIT) { /* TMRRC */
1400 iwmmxt_load_reg(cpu_V0, wrd);
1401 tcg_gen_trunc_i64_i32(cpu_R[rdlo], cpu_V0);
1402 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
1403 tcg_gen_trunc_i64_i32(cpu_R[rdhi], cpu_V0);
1404 } else { /* TMCRR */
1405 tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
1406 iwmmxt_store_reg(cpu_V0, wrd);
1407 gen_op_iwmmxt_set_mup();
1408 }
1409 return 0;
1410 }
1411
1412 wrd = (insn >> 12) & 0xf;
1413 addr = tcg_temp_new_i32();
1414 if (gen_iwmmxt_address(s, insn, addr)) {
1415 tcg_temp_free_i32(addr);
1416 return 1;
1417 }
1418 if (insn & ARM_CP_RW_BIT) {
1419 if ((insn >> 28) == 0xf) { /* WLDRW wCx */
1420 tmp = tcg_temp_new_i32();
1421 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
1422 iwmmxt_store_creg(wrd, tmp);
1423 } else {
1424 i = 1;
1425 if (insn & (1 << 8)) {
1426 if (insn & (1 << 22)) { /* WLDRD */
1427 tcg_gen_qemu_ld64(cpu_M0, addr, IS_USER(s));
1428 i = 0;
1429 } else { /* WLDRW wRd */
1430 tmp = tcg_temp_new_i32();
1431 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
1432 }
1433 } else {
1434 tmp = tcg_temp_new_i32();
1435 if (insn & (1 << 22)) { /* WLDRH */
1436 tcg_gen_qemu_ld16u(tmp, addr, IS_USER(s));
1437 } else { /* WLDRB */
1438 tcg_gen_qemu_ld8u(tmp, addr, IS_USER(s));
1439 }
1440 }
1441 if (i) {
1442 tcg_gen_extu_i32_i64(cpu_M0, tmp);
1443 tcg_temp_free_i32(tmp);
1444 }
1445 gen_op_iwmmxt_movq_wRn_M0(wrd);
1446 }
1447 } else {
1448 if ((insn >> 28) == 0xf) { /* WSTRW wCx */
1449 tmp = iwmmxt_load_creg(wrd);
1450 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
1451 } else {
1452 gen_op_iwmmxt_movq_M0_wRn(wrd);
1453 tmp = tcg_temp_new_i32();
1454 if (insn & (1 << 8)) {
1455 if (insn & (1 << 22)) { /* WSTRD */
1456 tcg_gen_qemu_st64(cpu_M0, addr, IS_USER(s));
1457 } else { /* WSTRW wRd */
1458 tcg_gen_trunc_i64_i32(tmp, cpu_M0);
1459 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
1460 }
1461 } else {
1462 if (insn & (1 << 22)) { /* WSTRH */
1463 tcg_gen_trunc_i64_i32(tmp, cpu_M0);
1464 tcg_gen_qemu_st16(tmp, addr, IS_USER(s));
1465 } else { /* WSTRB */
1466 tcg_gen_trunc_i64_i32(tmp, cpu_M0);
1467 tcg_gen_qemu_st8(tmp, addr, IS_USER(s));
1468 }
1469 }
1470 }
1471 tcg_temp_free_i32(tmp);
1472 }
1473 tcg_temp_free_i32(addr);
1474 return 0;
1475 }
1476
1477 if ((insn & 0x0f000000) != 0x0e000000)
1478 return 1;
1479
1480 switch (((insn >> 12) & 0xf00) | ((insn >> 4) & 0xff)) {
1481 case 0x000: /* WOR */
1482 wrd = (insn >> 12) & 0xf;
1483 rd0 = (insn >> 0) & 0xf;
1484 rd1 = (insn >> 16) & 0xf;
1485 gen_op_iwmmxt_movq_M0_wRn(rd0);
1486 gen_op_iwmmxt_orq_M0_wRn(rd1);
1487 gen_op_iwmmxt_setpsr_nz();
1488 gen_op_iwmmxt_movq_wRn_M0(wrd);
1489 gen_op_iwmmxt_set_mup();
1490 gen_op_iwmmxt_set_cup();
1491 break;
1492 case 0x011: /* TMCR */
1493 if (insn & 0xf)
1494 return 1;
1495 rd = (insn >> 12) & 0xf;
1496 wrd = (insn >> 16) & 0xf;
1497 switch (wrd) {
1498 case ARM_IWMMXT_wCID:
1499 case ARM_IWMMXT_wCASF:
1500 break;
1501 case ARM_IWMMXT_wCon:
1502 gen_op_iwmmxt_set_cup();
1503 /* Fall through. */
1504 case ARM_IWMMXT_wCSSF:
1505 tmp = iwmmxt_load_creg(wrd);
1506 tmp2 = load_reg(s, rd);
1507 tcg_gen_andc_i32(tmp, tmp, tmp2);
1508 tcg_temp_free_i32(tmp2);
1509 iwmmxt_store_creg(wrd, tmp);
1510 break;
1511 case ARM_IWMMXT_wCGR0:
1512 case ARM_IWMMXT_wCGR1:
1513 case ARM_IWMMXT_wCGR2:
1514 case ARM_IWMMXT_wCGR3:
1515 gen_op_iwmmxt_set_cup();
1516 tmp = load_reg(s, rd);
1517 iwmmxt_store_creg(wrd, tmp);
1518 break;
1519 default:
1520 return 1;
1521 }
1522 break;
1523 case 0x100: /* WXOR */
1524 wrd = (insn >> 12) & 0xf;
1525 rd0 = (insn >> 0) & 0xf;
1526 rd1 = (insn >> 16) & 0xf;
1527 gen_op_iwmmxt_movq_M0_wRn(rd0);
1528 gen_op_iwmmxt_xorq_M0_wRn(rd1);
1529 gen_op_iwmmxt_setpsr_nz();
1530 gen_op_iwmmxt_movq_wRn_M0(wrd);
1531 gen_op_iwmmxt_set_mup();
1532 gen_op_iwmmxt_set_cup();
1533 break;
1534 case 0x111: /* TMRC */
1535 if (insn & 0xf)
1536 return 1;
1537 rd = (insn >> 12) & 0xf;
1538 wrd = (insn >> 16) & 0xf;
1539 tmp = iwmmxt_load_creg(wrd);
1540 store_reg(s, rd, tmp);
1541 break;
1542 case 0x300: /* WANDN */
1543 wrd = (insn >> 12) & 0xf;
1544 rd0 = (insn >> 0) & 0xf;
1545 rd1 = (insn >> 16) & 0xf;
1546 gen_op_iwmmxt_movq_M0_wRn(rd0);
1547 tcg_gen_neg_i64(cpu_M0, cpu_M0);
1548 gen_op_iwmmxt_andq_M0_wRn(rd1);
1549 gen_op_iwmmxt_setpsr_nz();
1550 gen_op_iwmmxt_movq_wRn_M0(wrd);
1551 gen_op_iwmmxt_set_mup();
1552 gen_op_iwmmxt_set_cup();
1553 break;
1554 case 0x200: /* WAND */
1555 wrd = (insn >> 12) & 0xf;
1556 rd0 = (insn >> 0) & 0xf;
1557 rd1 = (insn >> 16) & 0xf;
1558 gen_op_iwmmxt_movq_M0_wRn(rd0);
1559 gen_op_iwmmxt_andq_M0_wRn(rd1);
1560 gen_op_iwmmxt_setpsr_nz();
1561 gen_op_iwmmxt_movq_wRn_M0(wrd);
1562 gen_op_iwmmxt_set_mup();
1563 gen_op_iwmmxt_set_cup();
1564 break;
1565 case 0x810: case 0xa10: /* WMADD */
1566 wrd = (insn >> 12) & 0xf;
1567 rd0 = (insn >> 0) & 0xf;
1568 rd1 = (insn >> 16) & 0xf;
1569 gen_op_iwmmxt_movq_M0_wRn(rd0);
1570 if (insn & (1 << 21))
1571 gen_op_iwmmxt_maddsq_M0_wRn(rd1);
1572 else
1573 gen_op_iwmmxt_madduq_M0_wRn(rd1);
1574 gen_op_iwmmxt_movq_wRn_M0(wrd);
1575 gen_op_iwmmxt_set_mup();
1576 break;
1577 case 0x10e: case 0x50e: case 0x90e: case 0xd0e: /* WUNPCKIL */
1578 wrd = (insn >> 12) & 0xf;
1579 rd0 = (insn >> 16) & 0xf;
1580 rd1 = (insn >> 0) & 0xf;
1581 gen_op_iwmmxt_movq_M0_wRn(rd0);
1582 switch ((insn >> 22) & 3) {
1583 case 0:
1584 gen_op_iwmmxt_unpacklb_M0_wRn(rd1);
1585 break;
1586 case 1:
1587 gen_op_iwmmxt_unpacklw_M0_wRn(rd1);
1588 break;
1589 case 2:
1590 gen_op_iwmmxt_unpackll_M0_wRn(rd1);
1591 break;
1592 case 3:
1593 return 1;
1594 }
1595 gen_op_iwmmxt_movq_wRn_M0(wrd);
1596 gen_op_iwmmxt_set_mup();
1597 gen_op_iwmmxt_set_cup();
1598 break;
1599 case 0x10c: case 0x50c: case 0x90c: case 0xd0c: /* WUNPCKIH */
1600 wrd = (insn >> 12) & 0xf;
1601 rd0 = (insn >> 16) & 0xf;
1602 rd1 = (insn >> 0) & 0xf;
1603 gen_op_iwmmxt_movq_M0_wRn(rd0);
1604 switch ((insn >> 22) & 3) {
1605 case 0:
1606 gen_op_iwmmxt_unpackhb_M0_wRn(rd1);
1607 break;
1608 case 1:
1609 gen_op_iwmmxt_unpackhw_M0_wRn(rd1);
1610 break;
1611 case 2:
1612 gen_op_iwmmxt_unpackhl_M0_wRn(rd1);
1613 break;
1614 case 3:
1615 return 1;
1616 }
1617 gen_op_iwmmxt_movq_wRn_M0(wrd);
1618 gen_op_iwmmxt_set_mup();
1619 gen_op_iwmmxt_set_cup();
1620 break;
1621 case 0x012: case 0x112: case 0x412: case 0x512: /* WSAD */
1622 wrd = (insn >> 12) & 0xf;
1623 rd0 = (insn >> 16) & 0xf;
1624 rd1 = (insn >> 0) & 0xf;
1625 gen_op_iwmmxt_movq_M0_wRn(rd0);
1626 if (insn & (1 << 22))
1627 gen_op_iwmmxt_sadw_M0_wRn(rd1);
1628 else
1629 gen_op_iwmmxt_sadb_M0_wRn(rd1);
1630 if (!(insn & (1 << 20)))
1631 gen_op_iwmmxt_addl_M0_wRn(wrd);
1632 gen_op_iwmmxt_movq_wRn_M0(wrd);
1633 gen_op_iwmmxt_set_mup();
1634 break;
1635 case 0x010: case 0x110: case 0x210: case 0x310: /* WMUL */
1636 wrd = (insn >> 12) & 0xf;
1637 rd0 = (insn >> 16) & 0xf;
1638 rd1 = (insn >> 0) & 0xf;
1639 gen_op_iwmmxt_movq_M0_wRn(rd0);
1640 if (insn & (1 << 21)) {
1641 if (insn & (1 << 20))
1642 gen_op_iwmmxt_mulshw_M0_wRn(rd1);
1643 else
1644 gen_op_iwmmxt_mulslw_M0_wRn(rd1);
1645 } else {
1646 if (insn & (1 << 20))
1647 gen_op_iwmmxt_muluhw_M0_wRn(rd1);
1648 else
1649 gen_op_iwmmxt_mululw_M0_wRn(rd1);
1650 }
1651 gen_op_iwmmxt_movq_wRn_M0(wrd);
1652 gen_op_iwmmxt_set_mup();
1653 break;
1654 case 0x410: case 0x510: case 0x610: case 0x710: /* WMAC */
1655 wrd = (insn >> 12) & 0xf;
1656 rd0 = (insn >> 16) & 0xf;
1657 rd1 = (insn >> 0) & 0xf;
1658 gen_op_iwmmxt_movq_M0_wRn(rd0);
1659 if (insn & (1 << 21))
1660 gen_op_iwmmxt_macsw_M0_wRn(rd1);
1661 else
1662 gen_op_iwmmxt_macuw_M0_wRn(rd1);
1663 if (!(insn & (1 << 20))) {
1664 iwmmxt_load_reg(cpu_V1, wrd);
1665 tcg_gen_add_i64(cpu_M0, cpu_M0, cpu_V1);
1666 }
1667 gen_op_iwmmxt_movq_wRn_M0(wrd);
1668 gen_op_iwmmxt_set_mup();
1669 break;
1670 case 0x006: case 0x406: case 0x806: case 0xc06: /* WCMPEQ */
1671 wrd = (insn >> 12) & 0xf;
1672 rd0 = (insn >> 16) & 0xf;
1673 rd1 = (insn >> 0) & 0xf;
1674 gen_op_iwmmxt_movq_M0_wRn(rd0);
1675 switch ((insn >> 22) & 3) {
1676 case 0:
1677 gen_op_iwmmxt_cmpeqb_M0_wRn(rd1);
1678 break;
1679 case 1:
1680 gen_op_iwmmxt_cmpeqw_M0_wRn(rd1);
1681 break;
1682 case 2:
1683 gen_op_iwmmxt_cmpeql_M0_wRn(rd1);
1684 break;
1685 case 3:
1686 return 1;
1687 }
1688 gen_op_iwmmxt_movq_wRn_M0(wrd);
1689 gen_op_iwmmxt_set_mup();
1690 gen_op_iwmmxt_set_cup();
1691 break;
1692 case 0x800: case 0x900: case 0xc00: case 0xd00: /* WAVG2 */
1693 wrd = (insn >> 12) & 0xf;
1694 rd0 = (insn >> 16) & 0xf;
1695 rd1 = (insn >> 0) & 0xf;
1696 gen_op_iwmmxt_movq_M0_wRn(rd0);
1697 if (insn & (1 << 22)) {
1698 if (insn & (1 << 20))
1699 gen_op_iwmmxt_avgw1_M0_wRn(rd1);
1700 else
1701 gen_op_iwmmxt_avgw0_M0_wRn(rd1);
1702 } else {
1703 if (insn & (1 << 20))
1704 gen_op_iwmmxt_avgb1_M0_wRn(rd1);
1705 else
1706 gen_op_iwmmxt_avgb0_M0_wRn(rd1);
1707 }
1708 gen_op_iwmmxt_movq_wRn_M0(wrd);
1709 gen_op_iwmmxt_set_mup();
1710 gen_op_iwmmxt_set_cup();
1711 break;
1712 case 0x802: case 0x902: case 0xa02: case 0xb02: /* WALIGNR */
1713 wrd = (insn >> 12) & 0xf;
1714 rd0 = (insn >> 16) & 0xf;
1715 rd1 = (insn >> 0) & 0xf;
1716 gen_op_iwmmxt_movq_M0_wRn(rd0);
1717 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCGR0 + ((insn >> 20) & 3));
1718 tcg_gen_andi_i32(tmp, tmp, 7);
1719 iwmmxt_load_reg(cpu_V1, rd1);
1720 gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1, tmp);
1721 tcg_temp_free_i32(tmp);
1722 gen_op_iwmmxt_movq_wRn_M0(wrd);
1723 gen_op_iwmmxt_set_mup();
1724 break;
1725 case 0x601: case 0x605: case 0x609: case 0x60d: /* TINSR */
1726 if (((insn >> 6) & 3) == 3)
1727 return 1;
1728 rd = (insn >> 12) & 0xf;
1729 wrd = (insn >> 16) & 0xf;
1730 tmp = load_reg(s, rd);
1731 gen_op_iwmmxt_movq_M0_wRn(wrd);
1732 switch ((insn >> 6) & 3) {
1733 case 0:
1734 tmp2 = tcg_const_i32(0xff);
1735 tmp3 = tcg_const_i32((insn & 7) << 3);
1736 break;
1737 case 1:
1738 tmp2 = tcg_const_i32(0xffff);
1739 tmp3 = tcg_const_i32((insn & 3) << 4);
1740 break;
1741 case 2:
1742 tmp2 = tcg_const_i32(0xffffffff);
1743 tmp3 = tcg_const_i32((insn & 1) << 5);
1744 break;
1745 default:
1746 TCGV_UNUSED_I32(tmp2);
1747 TCGV_UNUSED_I32(tmp3);
1748 }
1749 gen_helper_iwmmxt_insr(cpu_M0, cpu_M0, tmp, tmp2, tmp3);
1750 tcg_temp_free_i32(tmp3);
1751 tcg_temp_free_i32(tmp2);
1752 tcg_temp_free_i32(tmp);
1753 gen_op_iwmmxt_movq_wRn_M0(wrd);
1754 gen_op_iwmmxt_set_mup();
1755 break;
1756 case 0x107: case 0x507: case 0x907: case 0xd07: /* TEXTRM */
1757 rd = (insn >> 12) & 0xf;
1758 wrd = (insn >> 16) & 0xf;
1759 if (rd == 15 || ((insn >> 22) & 3) == 3)
1760 return 1;
1761 gen_op_iwmmxt_movq_M0_wRn(wrd);
1762 tmp = tcg_temp_new_i32();
1763 switch ((insn >> 22) & 3) {
1764 case 0:
1765 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 7) << 3);
1766 tcg_gen_trunc_i64_i32(tmp, cpu_M0);
1767 if (insn & 8) {
1768 tcg_gen_ext8s_i32(tmp, tmp);
1769 } else {
1770 tcg_gen_andi_i32(tmp, tmp, 0xff);
1771 }
1772 break;
1773 case 1:
1774 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 3) << 4);
1775 tcg_gen_trunc_i64_i32(tmp, cpu_M0);
1776 if (insn & 8) {
1777 tcg_gen_ext16s_i32(tmp, tmp);
1778 } else {
1779 tcg_gen_andi_i32(tmp, tmp, 0xffff);
1780 }
1781 break;
1782 case 2:
1783 tcg_gen_shri_i64(cpu_M0, cpu_M0, (insn & 1) << 5);
1784 tcg_gen_trunc_i64_i32(tmp, cpu_M0);
1785 break;
1786 }
1787 store_reg(s, rd, tmp);
1788 break;
1789 case 0x117: case 0x517: case 0x917: case 0xd17: /* TEXTRC */
1790 if ((insn & 0x000ff008) != 0x0003f000 || ((insn >> 22) & 3) == 3)
1791 return 1;
1792 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
1793 switch ((insn >> 22) & 3) {
1794 case 0:
1795 tcg_gen_shri_i32(tmp, tmp, ((insn & 7) << 2) + 0);
1796 break;
1797 case 1:
1798 tcg_gen_shri_i32(tmp, tmp, ((insn & 3) << 3) + 4);
1799 break;
1800 case 2:
1801 tcg_gen_shri_i32(tmp, tmp, ((insn & 1) << 4) + 12);
1802 break;
1803 }
1804 tcg_gen_shli_i32(tmp, tmp, 28);
1805 gen_set_nzcv(tmp);
1806 tcg_temp_free_i32(tmp);
1807 break;
1808 case 0x401: case 0x405: case 0x409: case 0x40d: /* TBCST */
1809 if (((insn >> 6) & 3) == 3)
1810 return 1;
1811 rd = (insn >> 12) & 0xf;
1812 wrd = (insn >> 16) & 0xf;
1813 tmp = load_reg(s, rd);
1814 switch ((insn >> 6) & 3) {
1815 case 0:
1816 gen_helper_iwmmxt_bcstb(cpu_M0, tmp);
1817 break;
1818 case 1:
1819 gen_helper_iwmmxt_bcstw(cpu_M0, tmp);
1820 break;
1821 case 2:
1822 gen_helper_iwmmxt_bcstl(cpu_M0, tmp);
1823 break;
1824 }
1825 tcg_temp_free_i32(tmp);
1826 gen_op_iwmmxt_movq_wRn_M0(wrd);
1827 gen_op_iwmmxt_set_mup();
1828 break;
1829 case 0x113: case 0x513: case 0x913: case 0xd13: /* TANDC */
1830 if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
1831 return 1;
1832 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
1833 tmp2 = tcg_temp_new_i32();
1834 tcg_gen_mov_i32(tmp2, tmp);
1835 switch ((insn >> 22) & 3) {
1836 case 0:
1837 for (i = 0; i < 7; i ++) {
1838 tcg_gen_shli_i32(tmp2, tmp2, 4);
1839 tcg_gen_and_i32(tmp, tmp, tmp2);
1840 }
1841 break;
1842 case 1:
1843 for (i = 0; i < 3; i ++) {
1844 tcg_gen_shli_i32(tmp2, tmp2, 8);
1845 tcg_gen_and_i32(tmp, tmp, tmp2);
1846 }
1847 break;
1848 case 2:
1849 tcg_gen_shli_i32(tmp2, tmp2, 16);
1850 tcg_gen_and_i32(tmp, tmp, tmp2);
1851 break;
1852 }
1853 gen_set_nzcv(tmp);
1854 tcg_temp_free_i32(tmp2);
1855 tcg_temp_free_i32(tmp);
1856 break;
1857 case 0x01c: case 0x41c: case 0x81c: case 0xc1c: /* WACC */
1858 wrd = (insn >> 12) & 0xf;
1859 rd0 = (insn >> 16) & 0xf;
1860 gen_op_iwmmxt_movq_M0_wRn(rd0);
1861 switch ((insn >> 22) & 3) {
1862 case 0:
1863 gen_helper_iwmmxt_addcb(cpu_M0, cpu_M0);
1864 break;
1865 case 1:
1866 gen_helper_iwmmxt_addcw(cpu_M0, cpu_M0);
1867 break;
1868 case 2:
1869 gen_helper_iwmmxt_addcl(cpu_M0, cpu_M0);
1870 break;
1871 case 3:
1872 return 1;
1873 }
1874 gen_op_iwmmxt_movq_wRn_M0(wrd);
1875 gen_op_iwmmxt_set_mup();
1876 break;
1877 case 0x115: case 0x515: case 0x915: case 0xd15: /* TORC */
1878 if ((insn & 0x000ff00f) != 0x0003f000 || ((insn >> 22) & 3) == 3)
1879 return 1;
1880 tmp = iwmmxt_load_creg(ARM_IWMMXT_wCASF);
1881 tmp2 = tcg_temp_new_i32();
1882 tcg_gen_mov_i32(tmp2, tmp);
1883 switch ((insn >> 22) & 3) {
1884 case 0:
1885 for (i = 0; i < 7; i ++) {
1886 tcg_gen_shli_i32(tmp2, tmp2, 4);
1887 tcg_gen_or_i32(tmp, tmp, tmp2);
1888 }
1889 break;
1890 case 1:
1891 for (i = 0; i < 3; i ++) {
1892 tcg_gen_shli_i32(tmp2, tmp2, 8);
1893 tcg_gen_or_i32(tmp, tmp, tmp2);
1894 }
1895 break;
1896 case 2:
1897 tcg_gen_shli_i32(tmp2, tmp2, 16);
1898 tcg_gen_or_i32(tmp, tmp, tmp2);
1899 break;
1900 }
1901 gen_set_nzcv(tmp);
1902 tcg_temp_free_i32(tmp2);
1903 tcg_temp_free_i32(tmp);
1904 break;
1905 case 0x103: case 0x503: case 0x903: case 0xd03: /* TMOVMSK */
1906 rd = (insn >> 12) & 0xf;
1907 rd0 = (insn >> 16) & 0xf;
1908 if ((insn & 0xf) != 0 || ((insn >> 22) & 3) == 3)
1909 return 1;
1910 gen_op_iwmmxt_movq_M0_wRn(rd0);
1911 tmp = tcg_temp_new_i32();
1912 switch ((insn >> 22) & 3) {
1913 case 0:
1914 gen_helper_iwmmxt_msbb(tmp, cpu_M0);
1915 break;
1916 case 1:
1917 gen_helper_iwmmxt_msbw(tmp, cpu_M0);
1918 break;
1919 case 2:
1920 gen_helper_iwmmxt_msbl(tmp, cpu_M0);
1921 break;
1922 }
1923 store_reg(s, rd, tmp);
1924 break;
1925 case 0x106: case 0x306: case 0x506: case 0x706: /* WCMPGT */
1926 case 0x906: case 0xb06: case 0xd06: case 0xf06:
1927 wrd = (insn >> 12) & 0xf;
1928 rd0 = (insn >> 16) & 0xf;
1929 rd1 = (insn >> 0) & 0xf;
1930 gen_op_iwmmxt_movq_M0_wRn(rd0);
1931 switch ((insn >> 22) & 3) {
1932 case 0:
1933 if (insn & (1 << 21))
1934 gen_op_iwmmxt_cmpgtsb_M0_wRn(rd1);
1935 else
1936 gen_op_iwmmxt_cmpgtub_M0_wRn(rd1);
1937 break;
1938 case 1:
1939 if (insn & (1 << 21))
1940 gen_op_iwmmxt_cmpgtsw_M0_wRn(rd1);
1941 else
1942 gen_op_iwmmxt_cmpgtuw_M0_wRn(rd1);
1943 break;
1944 case 2:
1945 if (insn & (1 << 21))
1946 gen_op_iwmmxt_cmpgtsl_M0_wRn(rd1);
1947 else
1948 gen_op_iwmmxt_cmpgtul_M0_wRn(rd1);
1949 break;
1950 case 3:
1951 return 1;
1952 }
1953 gen_op_iwmmxt_movq_wRn_M0(wrd);
1954 gen_op_iwmmxt_set_mup();
1955 gen_op_iwmmxt_set_cup();
1956 break;
1957 case 0x00e: case 0x20e: case 0x40e: case 0x60e: /* WUNPCKEL */
1958 case 0x80e: case 0xa0e: case 0xc0e: case 0xe0e:
1959 wrd = (insn >> 12) & 0xf;
1960 rd0 = (insn >> 16) & 0xf;
1961 gen_op_iwmmxt_movq_M0_wRn(rd0);
1962 switch ((insn >> 22) & 3) {
1963 case 0:
1964 if (insn & (1 << 21))
1965 gen_op_iwmmxt_unpacklsb_M0();
1966 else
1967 gen_op_iwmmxt_unpacklub_M0();
1968 break;
1969 case 1:
1970 if (insn & (1 << 21))
1971 gen_op_iwmmxt_unpacklsw_M0();
1972 else
1973 gen_op_iwmmxt_unpackluw_M0();
1974 break;
1975 case 2:
1976 if (insn & (1 << 21))
1977 gen_op_iwmmxt_unpacklsl_M0();
1978 else
1979 gen_op_iwmmxt_unpacklul_M0();
1980 break;
1981 case 3:
1982 return 1;
1983 }
1984 gen_op_iwmmxt_movq_wRn_M0(wrd);
1985 gen_op_iwmmxt_set_mup();
1986 gen_op_iwmmxt_set_cup();
1987 break;
1988 case 0x00c: case 0x20c: case 0x40c: case 0x60c: /* WUNPCKEH */
1989 case 0x80c: case 0xa0c: case 0xc0c: case 0xe0c:
1990 wrd = (insn >> 12) & 0xf;
1991 rd0 = (insn >> 16) & 0xf;
1992 gen_op_iwmmxt_movq_M0_wRn(rd0);
1993 switch ((insn >> 22) & 3) {
1994 case 0:
1995 if (insn & (1 << 21))
1996 gen_op_iwmmxt_unpackhsb_M0();
1997 else
1998 gen_op_iwmmxt_unpackhub_M0();
1999 break;
2000 case 1:
2001 if (insn & (1 << 21))
2002 gen_op_iwmmxt_unpackhsw_M0();
2003 else
2004 gen_op_iwmmxt_unpackhuw_M0();
2005 break;
2006 case 2:
2007 if (insn & (1 << 21))
2008 gen_op_iwmmxt_unpackhsl_M0();
2009 else
2010 gen_op_iwmmxt_unpackhul_M0();
2011 break;
2012 case 3:
2013 return 1;
2014 }
2015 gen_op_iwmmxt_movq_wRn_M0(wrd);
2016 gen_op_iwmmxt_set_mup();
2017 gen_op_iwmmxt_set_cup();
2018 break;
2019 case 0x204: case 0x604: case 0xa04: case 0xe04: /* WSRL */
2020 case 0x214: case 0x614: case 0xa14: case 0xe14:
2021 if (((insn >> 22) & 3) == 0)
2022 return 1;
2023 wrd = (insn >> 12) & 0xf;
2024 rd0 = (insn >> 16) & 0xf;
2025 gen_op_iwmmxt_movq_M0_wRn(rd0);
2026 tmp = tcg_temp_new_i32();
2027 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2028 tcg_temp_free_i32(tmp);
2029 return 1;
2030 }
2031 switch ((insn >> 22) & 3) {
2032 case 1:
2033 gen_helper_iwmmxt_srlw(cpu_M0, cpu_env, cpu_M0, tmp);
2034 break;
2035 case 2:
2036 gen_helper_iwmmxt_srll(cpu_M0, cpu_env, cpu_M0, tmp);
2037 break;
2038 case 3:
2039 gen_helper_iwmmxt_srlq(cpu_M0, cpu_env, cpu_M0, tmp);
2040 break;
2041 }
2042 tcg_temp_free_i32(tmp);
2043 gen_op_iwmmxt_movq_wRn_M0(wrd);
2044 gen_op_iwmmxt_set_mup();
2045 gen_op_iwmmxt_set_cup();
2046 break;
2047 case 0x004: case 0x404: case 0x804: case 0xc04: /* WSRA */
2048 case 0x014: case 0x414: case 0x814: case 0xc14:
2049 if (((insn >> 22) & 3) == 0)
2050 return 1;
2051 wrd = (insn >> 12) & 0xf;
2052 rd0 = (insn >> 16) & 0xf;
2053 gen_op_iwmmxt_movq_M0_wRn(rd0);
2054 tmp = tcg_temp_new_i32();
2055 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2056 tcg_temp_free_i32(tmp);
2057 return 1;
2058 }
2059 switch ((insn >> 22) & 3) {
2060 case 1:
2061 gen_helper_iwmmxt_sraw(cpu_M0, cpu_env, cpu_M0, tmp);
2062 break;
2063 case 2:
2064 gen_helper_iwmmxt_sral(cpu_M0, cpu_env, cpu_M0, tmp);
2065 break;
2066 case 3:
2067 gen_helper_iwmmxt_sraq(cpu_M0, cpu_env, cpu_M0, tmp);
2068 break;
2069 }
2070 tcg_temp_free_i32(tmp);
2071 gen_op_iwmmxt_movq_wRn_M0(wrd);
2072 gen_op_iwmmxt_set_mup();
2073 gen_op_iwmmxt_set_cup();
2074 break;
2075 case 0x104: case 0x504: case 0x904: case 0xd04: /* WSLL */
2076 case 0x114: case 0x514: case 0x914: case 0xd14:
2077 if (((insn >> 22) & 3) == 0)
2078 return 1;
2079 wrd = (insn >> 12) & 0xf;
2080 rd0 = (insn >> 16) & 0xf;
2081 gen_op_iwmmxt_movq_M0_wRn(rd0);
2082 tmp = tcg_temp_new_i32();
2083 if (gen_iwmmxt_shift(insn, 0xff, tmp)) {
2084 tcg_temp_free_i32(tmp);
2085 return 1;
2086 }
2087 switch ((insn >> 22) & 3) {
2088 case 1:
2089 gen_helper_iwmmxt_sllw(cpu_M0, cpu_env, cpu_M0, tmp);
2090 break;
2091 case 2:
2092 gen_helper_iwmmxt_slll(cpu_M0, cpu_env, cpu_M0, tmp);
2093 break;
2094 case 3:
2095 gen_helper_iwmmxt_sllq(cpu_M0, cpu_env, cpu_M0, tmp);
2096 break;
2097 }
2098 tcg_temp_free_i32(tmp);
2099 gen_op_iwmmxt_movq_wRn_M0(wrd);
2100 gen_op_iwmmxt_set_mup();
2101 gen_op_iwmmxt_set_cup();
2102 break;
2103 case 0x304: case 0x704: case 0xb04: case 0xf04: /* WROR */
2104 case 0x314: case 0x714: case 0xb14: case 0xf14:
2105 if (((insn >> 22) & 3) == 0)
2106 return 1;
2107 wrd = (insn >> 12) & 0xf;
2108 rd0 = (insn >> 16) & 0xf;
2109 gen_op_iwmmxt_movq_M0_wRn(rd0);
2110 tmp = tcg_temp_new_i32();
2111 switch ((insn >> 22) & 3) {
2112 case 1:
2113 if (gen_iwmmxt_shift(insn, 0xf, tmp)) {
2114 tcg_temp_free_i32(tmp);
2115 return 1;
2116 }
2117 gen_helper_iwmmxt_rorw(cpu_M0, cpu_env, cpu_M0, tmp);
2118 break;
2119 case 2:
2120 if (gen_iwmmxt_shift(insn, 0x1f, tmp)) {
2121 tcg_temp_free_i32(tmp);
2122 return 1;
2123 }
2124 gen_helper_iwmmxt_rorl(cpu_M0, cpu_env, cpu_M0, tmp);
2125 break;
2126 case 3:
2127 if (gen_iwmmxt_shift(insn, 0x3f, tmp)) {
2128 tcg_temp_free_i32(tmp);
2129 return 1;
2130 }
2131 gen_helper_iwmmxt_rorq(cpu_M0, cpu_env, cpu_M0, tmp);
2132 break;
2133 }
2134 tcg_temp_free_i32(tmp);
2135 gen_op_iwmmxt_movq_wRn_M0(wrd);
2136 gen_op_iwmmxt_set_mup();
2137 gen_op_iwmmxt_set_cup();
2138 break;
2139 case 0x116: case 0x316: case 0x516: case 0x716: /* WMIN */
2140 case 0x916: case 0xb16: case 0xd16: case 0xf16:
2141 wrd = (insn >> 12) & 0xf;
2142 rd0 = (insn >> 16) & 0xf;
2143 rd1 = (insn >> 0) & 0xf;
2144 gen_op_iwmmxt_movq_M0_wRn(rd0);
2145 switch ((insn >> 22) & 3) {
2146 case 0:
2147 if (insn & (1 << 21))
2148 gen_op_iwmmxt_minsb_M0_wRn(rd1);
2149 else
2150 gen_op_iwmmxt_minub_M0_wRn(rd1);
2151 break;
2152 case 1:
2153 if (insn & (1 << 21))
2154 gen_op_iwmmxt_minsw_M0_wRn(rd1);
2155 else
2156 gen_op_iwmmxt_minuw_M0_wRn(rd1);
2157 break;
2158 case 2:
2159 if (insn & (1 << 21))
2160 gen_op_iwmmxt_minsl_M0_wRn(rd1);
2161 else
2162 gen_op_iwmmxt_minul_M0_wRn(rd1);
2163 break;
2164 case 3:
2165 return 1;
2166 }
2167 gen_op_iwmmxt_movq_wRn_M0(wrd);
2168 gen_op_iwmmxt_set_mup();
2169 break;
2170 case 0x016: case 0x216: case 0x416: case 0x616: /* WMAX */
2171 case 0x816: case 0xa16: case 0xc16: case 0xe16:
2172 wrd = (insn >> 12) & 0xf;
2173 rd0 = (insn >> 16) & 0xf;
2174 rd1 = (insn >> 0) & 0xf;
2175 gen_op_iwmmxt_movq_M0_wRn(rd0);
2176 switch ((insn >> 22) & 3) {
2177 case 0:
2178 if (insn & (1 << 21))
2179 gen_op_iwmmxt_maxsb_M0_wRn(rd1);
2180 else
2181 gen_op_iwmmxt_maxub_M0_wRn(rd1);
2182 break;
2183 case 1:
2184 if (insn & (1 << 21))
2185 gen_op_iwmmxt_maxsw_M0_wRn(rd1);
2186 else
2187 gen_op_iwmmxt_maxuw_M0_wRn(rd1);
2188 break;
2189 case 2:
2190 if (insn & (1 << 21))
2191 gen_op_iwmmxt_maxsl_M0_wRn(rd1);
2192 else
2193 gen_op_iwmmxt_maxul_M0_wRn(rd1);
2194 break;
2195 case 3:
2196 return 1;
2197 }
2198 gen_op_iwmmxt_movq_wRn_M0(wrd);
2199 gen_op_iwmmxt_set_mup();
2200 break;
2201 case 0x002: case 0x102: case 0x202: case 0x302: /* WALIGNI */
2202 case 0x402: case 0x502: case 0x602: case 0x702:
2203 wrd = (insn >> 12) & 0xf;
2204 rd0 = (insn >> 16) & 0xf;
2205 rd1 = (insn >> 0) & 0xf;
2206 gen_op_iwmmxt_movq_M0_wRn(rd0);
2207 tmp = tcg_const_i32((insn >> 20) & 3);
2208 iwmmxt_load_reg(cpu_V1, rd1);
2209 gen_helper_iwmmxt_align(cpu_M0, cpu_M0, cpu_V1, tmp);
2210 tcg_temp_free_i32(tmp);
2211 gen_op_iwmmxt_movq_wRn_M0(wrd);
2212 gen_op_iwmmxt_set_mup();
2213 break;
2214 case 0x01a: case 0x11a: case 0x21a: case 0x31a: /* WSUB */
2215 case 0x41a: case 0x51a: case 0x61a: case 0x71a:
2216 case 0x81a: case 0x91a: case 0xa1a: case 0xb1a:
2217 case 0xc1a: case 0xd1a: case 0xe1a: case 0xf1a:
2218 wrd = (insn >> 12) & 0xf;
2219 rd0 = (insn >> 16) & 0xf;
2220 rd1 = (insn >> 0) & 0xf;
2221 gen_op_iwmmxt_movq_M0_wRn(rd0);
2222 switch ((insn >> 20) & 0xf) {
2223 case 0x0:
2224 gen_op_iwmmxt_subnb_M0_wRn(rd1);
2225 break;
2226 case 0x1:
2227 gen_op_iwmmxt_subub_M0_wRn(rd1);
2228 break;
2229 case 0x3:
2230 gen_op_iwmmxt_subsb_M0_wRn(rd1);
2231 break;
2232 case 0x4:
2233 gen_op_iwmmxt_subnw_M0_wRn(rd1);
2234 break;
2235 case 0x5:
2236 gen_op_iwmmxt_subuw_M0_wRn(rd1);
2237 break;
2238 case 0x7:
2239 gen_op_iwmmxt_subsw_M0_wRn(rd1);
2240 break;
2241 case 0x8:
2242 gen_op_iwmmxt_subnl_M0_wRn(rd1);
2243 break;
2244 case 0x9:
2245 gen_op_iwmmxt_subul_M0_wRn(rd1);
2246 break;
2247 case 0xb:
2248 gen_op_iwmmxt_subsl_M0_wRn(rd1);
2249 break;
2250 default:
2251 return 1;
2252 }
2253 gen_op_iwmmxt_movq_wRn_M0(wrd);
2254 gen_op_iwmmxt_set_mup();
2255 gen_op_iwmmxt_set_cup();
2256 break;
2257 case 0x01e: case 0x11e: case 0x21e: case 0x31e: /* WSHUFH */
2258 case 0x41e: case 0x51e: case 0x61e: case 0x71e:
2259 case 0x81e: case 0x91e: case 0xa1e: case 0xb1e:
2260 case 0xc1e: case 0xd1e: case 0xe1e: case 0xf1e:
2261 wrd = (insn >> 12) & 0xf;
2262 rd0 = (insn >> 16) & 0xf;
2263 gen_op_iwmmxt_movq_M0_wRn(rd0);
2264 tmp = tcg_const_i32(((insn >> 16) & 0xf0) | (insn & 0x0f));
2265 gen_helper_iwmmxt_shufh(cpu_M0, cpu_env, cpu_M0, tmp);
2266 tcg_temp_free_i32(tmp);
2267 gen_op_iwmmxt_movq_wRn_M0(wrd);
2268 gen_op_iwmmxt_set_mup();
2269 gen_op_iwmmxt_set_cup();
2270 break;
2271 case 0x018: case 0x118: case 0x218: case 0x318: /* WADD */
2272 case 0x418: case 0x518: case 0x618: case 0x718:
2273 case 0x818: case 0x918: case 0xa18: case 0xb18:
2274 case 0xc18: case 0xd18: case 0xe18: case 0xf18:
2275 wrd = (insn >> 12) & 0xf;
2276 rd0 = (insn >> 16) & 0xf;
2277 rd1 = (insn >> 0) & 0xf;
2278 gen_op_iwmmxt_movq_M0_wRn(rd0);
2279 switch ((insn >> 20) & 0xf) {
2280 case 0x0:
2281 gen_op_iwmmxt_addnb_M0_wRn(rd1);
2282 break;
2283 case 0x1:
2284 gen_op_iwmmxt_addub_M0_wRn(rd1);
2285 break;
2286 case 0x3:
2287 gen_op_iwmmxt_addsb_M0_wRn(rd1);
2288 break;
2289 case 0x4:
2290 gen_op_iwmmxt_addnw_M0_wRn(rd1);
2291 break;
2292 case 0x5:
2293 gen_op_iwmmxt_adduw_M0_wRn(rd1);
2294 break;
2295 case 0x7:
2296 gen_op_iwmmxt_addsw_M0_wRn(rd1);
2297 break;
2298 case 0x8:
2299 gen_op_iwmmxt_addnl_M0_wRn(rd1);
2300 break;
2301 case 0x9:
2302 gen_op_iwmmxt_addul_M0_wRn(rd1);
2303 break;
2304 case 0xb:
2305 gen_op_iwmmxt_addsl_M0_wRn(rd1);
2306 break;
2307 default:
2308 return 1;
2309 }
2310 gen_op_iwmmxt_movq_wRn_M0(wrd);
2311 gen_op_iwmmxt_set_mup();
2312 gen_op_iwmmxt_set_cup();
2313 break;
2314 case 0x008: case 0x108: case 0x208: case 0x308: /* WPACK */
2315 case 0x408: case 0x508: case 0x608: case 0x708:
2316 case 0x808: case 0x908: case 0xa08: case 0xb08:
2317 case 0xc08: case 0xd08: case 0xe08: case 0xf08:
2318 if (!(insn & (1 << 20)) || ((insn >> 22) & 3) == 0)
2319 return 1;
2320 wrd = (insn >> 12) & 0xf;
2321 rd0 = (insn >> 16) & 0xf;
2322 rd1 = (insn >> 0) & 0xf;
2323 gen_op_iwmmxt_movq_M0_wRn(rd0);
2324 switch ((insn >> 22) & 3) {
2325 case 1:
2326 if (insn & (1 << 21))
2327 gen_op_iwmmxt_packsw_M0_wRn(rd1);
2328 else
2329 gen_op_iwmmxt_packuw_M0_wRn(rd1);
2330 break;
2331 case 2:
2332 if (insn & (1 << 21))
2333 gen_op_iwmmxt_packsl_M0_wRn(rd1);
2334 else
2335 gen_op_iwmmxt_packul_M0_wRn(rd1);
2336 break;
2337 case 3:
2338 if (insn & (1 << 21))
2339 gen_op_iwmmxt_packsq_M0_wRn(rd1);
2340 else
2341 gen_op_iwmmxt_packuq_M0_wRn(rd1);
2342 break;
2343 }
2344 gen_op_iwmmxt_movq_wRn_M0(wrd);
2345 gen_op_iwmmxt_set_mup();
2346 gen_op_iwmmxt_set_cup();
2347 break;
2348 case 0x201: case 0x203: case 0x205: case 0x207:
2349 case 0x209: case 0x20b: case 0x20d: case 0x20f:
2350 case 0x211: case 0x213: case 0x215: case 0x217:
2351 case 0x219: case 0x21b: case 0x21d: case 0x21f:
2352 wrd = (insn >> 5) & 0xf;
2353 rd0 = (insn >> 12) & 0xf;
2354 rd1 = (insn >> 0) & 0xf;
2355 if (rd0 == 0xf || rd1 == 0xf)
2356 return 1;
2357 gen_op_iwmmxt_movq_M0_wRn(wrd);
2358 tmp = load_reg(s, rd0);
2359 tmp2 = load_reg(s, rd1);
2360 switch ((insn >> 16) & 0xf) {
2361 case 0x0: /* TMIA */
2362 gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2363 break;
2364 case 0x8: /* TMIAPH */
2365 gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2366 break;
2367 case 0xc: case 0xd: case 0xe: case 0xf: /* TMIAxy */
2368 if (insn & (1 << 16))
2369 tcg_gen_shri_i32(tmp, tmp, 16);
2370 if (insn & (1 << 17))
2371 tcg_gen_shri_i32(tmp2, tmp2, 16);
2372 gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2373 break;
2374 default:
2375 tcg_temp_free_i32(tmp2);
2376 tcg_temp_free_i32(tmp);
2377 return 1;
2378 }
2379 tcg_temp_free_i32(tmp2);
2380 tcg_temp_free_i32(tmp);
2381 gen_op_iwmmxt_movq_wRn_M0(wrd);
2382 gen_op_iwmmxt_set_mup();
2383 break;
2384 default:
2385 return 1;
2386 }
2387
2388 return 0;
2389 }
2390
2391 /* Disassemble an XScale DSP instruction. Returns nonzero if an error occurred
2392 (ie. an undefined instruction). */
2393 static int disas_dsp_insn(CPUARMState *env, DisasContext *s, uint32_t insn)
2394 {
2395 int acc, rd0, rd1, rdhi, rdlo;
2396 TCGv_i32 tmp, tmp2;
2397
2398 if ((insn & 0x0ff00f10) == 0x0e200010) {
2399 /* Multiply with Internal Accumulate Format */
2400 rd0 = (insn >> 12) & 0xf;
2401 rd1 = insn & 0xf;
2402 acc = (insn >> 5) & 7;
2403
2404 if (acc != 0)
2405 return 1;
2406
2407 tmp = load_reg(s, rd0);
2408 tmp2 = load_reg(s, rd1);
2409 switch ((insn >> 16) & 0xf) {
2410 case 0x0: /* MIA */
2411 gen_helper_iwmmxt_muladdsl(cpu_M0, cpu_M0, tmp, tmp2);
2412 break;
2413 case 0x8: /* MIAPH */
2414 gen_helper_iwmmxt_muladdsw(cpu_M0, cpu_M0, tmp, tmp2);
2415 break;
2416 case 0xc: /* MIABB */
2417 case 0xd: /* MIABT */
2418 case 0xe: /* MIATB */
2419 case 0xf: /* MIATT */
2420 if (insn & (1 << 16))
2421 tcg_gen_shri_i32(tmp, tmp, 16);
2422 if (insn & (1 << 17))
2423 tcg_gen_shri_i32(tmp2, tmp2, 16);
2424 gen_helper_iwmmxt_muladdswl(cpu_M0, cpu_M0, tmp, tmp2);
2425 break;
2426 default:
2427 return 1;
2428 }
2429 tcg_temp_free_i32(tmp2);
2430 tcg_temp_free_i32(tmp);
2431
2432 gen_op_iwmmxt_movq_wRn_M0(acc);
2433 return 0;
2434 }
2435
2436 if ((insn & 0x0fe00ff8) == 0x0c400000) {
2437 /* Internal Accumulator Access Format */
2438 rdhi = (insn >> 16) & 0xf;
2439 rdlo = (insn >> 12) & 0xf;
2440 acc = insn & 7;
2441
2442 if (acc != 0)
2443 return 1;
2444
2445 if (insn & ARM_CP_RW_BIT) { /* MRA */
2446 iwmmxt_load_reg(cpu_V0, acc);
2447 tcg_gen_trunc_i64_i32(cpu_R[rdlo], cpu_V0);
2448 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
2449 tcg_gen_trunc_i64_i32(cpu_R[rdhi], cpu_V0);
2450 tcg_gen_andi_i32(cpu_R[rdhi], cpu_R[rdhi], (1 << (40 - 32)) - 1);
2451 } else { /* MAR */
2452 tcg_gen_concat_i32_i64(cpu_V0, cpu_R[rdlo], cpu_R[rdhi]);
2453 iwmmxt_store_reg(cpu_V0, acc);
2454 }
2455 return 0;
2456 }
2457
2458 return 1;
2459 }
2460
2461 #define VFP_REG_SHR(x, n) (((n) > 0) ? (x) >> (n) : (x) << -(n))
2462 #define VFP_SREG(insn, bigbit, smallbit) \
2463 ((VFP_REG_SHR(insn, bigbit - 1) & 0x1e) | (((insn) >> (smallbit)) & 1))
2464 #define VFP_DREG(reg, insn, bigbit, smallbit) do { \
2465 if (arm_feature(env, ARM_FEATURE_VFP3)) { \
2466 reg = (((insn) >> (bigbit)) & 0x0f) \
2467 | (((insn) >> ((smallbit) - 4)) & 0x10); \
2468 } else { \
2469 if (insn & (1 << (smallbit))) \
2470 return 1; \
2471 reg = ((insn) >> (bigbit)) & 0x0f; \
2472 }} while (0)
2473
2474 #define VFP_SREG_D(insn) VFP_SREG(insn, 12, 22)
2475 #define VFP_DREG_D(reg, insn) VFP_DREG(reg, insn, 12, 22)
2476 #define VFP_SREG_N(insn) VFP_SREG(insn, 16, 7)
2477 #define VFP_DREG_N(reg, insn) VFP_DREG(reg, insn, 16, 7)
2478 #define VFP_SREG_M(insn) VFP_SREG(insn, 0, 5)
2479 #define VFP_DREG_M(reg, insn) VFP_DREG(reg, insn, 0, 5)
2480
2481 /* Move between integer and VFP cores. */
2482 static TCGv_i32 gen_vfp_mrs(void)
2483 {
2484 TCGv_i32 tmp = tcg_temp_new_i32();
2485 tcg_gen_mov_i32(tmp, cpu_F0s);
2486 return tmp;
2487 }
2488
2489 static void gen_vfp_msr(TCGv_i32 tmp)
2490 {
2491 tcg_gen_mov_i32(cpu_F0s, tmp);
2492 tcg_temp_free_i32(tmp);
2493 }
2494
2495 static void gen_neon_dup_u8(TCGv_i32 var, int shift)
2496 {
2497 TCGv_i32 tmp = tcg_temp_new_i32();
2498 if (shift)
2499 tcg_gen_shri_i32(var, var, shift);
2500 tcg_gen_ext8u_i32(var, var);
2501 tcg_gen_shli_i32(tmp, var, 8);
2502 tcg_gen_or_i32(var, var, tmp);
2503 tcg_gen_shli_i32(tmp, var, 16);
2504 tcg_gen_or_i32(var, var, tmp);
2505 tcg_temp_free_i32(tmp);
2506 }
2507
2508 static void gen_neon_dup_low16(TCGv_i32 var)
2509 {
2510 TCGv_i32 tmp = tcg_temp_new_i32();
2511 tcg_gen_ext16u_i32(var, var);
2512 tcg_gen_shli_i32(tmp, var, 16);
2513 tcg_gen_or_i32(var, var, tmp);
2514 tcg_temp_free_i32(tmp);
2515 }
2516
2517 static void gen_neon_dup_high16(TCGv_i32 var)
2518 {
2519 TCGv_i32 tmp = tcg_temp_new_i32();
2520 tcg_gen_andi_i32(var, var, 0xffff0000);
2521 tcg_gen_shri_i32(tmp, var, 16);
2522 tcg_gen_or_i32(var, var, tmp);
2523 tcg_temp_free_i32(tmp);
2524 }
2525
2526 static TCGv_i32 gen_load_and_replicate(DisasContext *s, TCGv_i32 addr, int size)
2527 {
2528 /* Load a single Neon element and replicate into a 32 bit TCG reg */
2529 TCGv_i32 tmp = tcg_temp_new_i32();
2530 switch (size) {
2531 case 0:
2532 tcg_gen_qemu_ld8u(tmp, addr, IS_USER(s));
2533 gen_neon_dup_u8(tmp, 0);
2534 break;
2535 case 1:
2536 tcg_gen_qemu_ld16u(tmp, addr, IS_USER(s));
2537 gen_neon_dup_low16(tmp);
2538 break;
2539 case 2:
2540 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
2541 break;
2542 default: /* Avoid compiler warnings. */
2543 abort();
2544 }
2545 return tmp;
2546 }
2547
2548 /* Disassemble a VFP instruction. Returns nonzero if an error occurred
2549 (ie. an undefined instruction). */
2550 static int disas_vfp_insn(CPUARMState * env, DisasContext *s, uint32_t insn)
2551 {
2552 uint32_t rd, rn, rm, op, i, n, offset, delta_d, delta_m, bank_mask;
2553 int dp, veclen;
2554 TCGv_i32 addr;
2555 TCGv_i32 tmp;
2556 TCGv_i32 tmp2;
2557
2558 if (!arm_feature(env, ARM_FEATURE_VFP))
2559 return 1;
2560
2561 if (!s->vfp_enabled) {
2562 /* VFP disabled. Only allow fmxr/fmrx to/from some control regs. */
2563 if ((insn & 0x0fe00fff) != 0x0ee00a10)
2564 return 1;
2565 rn = (insn >> 16) & 0xf;
2566 if (rn != ARM_VFP_FPSID && rn != ARM_VFP_FPEXC
2567 && rn != ARM_VFP_MVFR1 && rn != ARM_VFP_MVFR0)
2568 return 1;
2569 }
2570 dp = ((insn & 0xf00) == 0xb00);
2571 switch ((insn >> 24) & 0xf) {
2572 case 0xe:
2573 if (insn & (1 << 4)) {
2574 /* single register transfer */
2575 rd = (insn >> 12) & 0xf;
2576 if (dp) {
2577 int size;
2578 int pass;
2579
2580 VFP_DREG_N(rn, insn);
2581 if (insn & 0xf)
2582 return 1;
2583 if (insn & 0x00c00060
2584 && !arm_feature(env, ARM_FEATURE_NEON))
2585 return 1;
2586
2587 pass = (insn >> 21) & 1;
2588 if (insn & (1 << 22)) {
2589 size = 0;
2590 offset = ((insn >> 5) & 3) * 8;
2591 } else if (insn & (1 << 5)) {
2592 size = 1;
2593 offset = (insn & (1 << 6)) ? 16 : 0;
2594 } else {
2595 size = 2;
2596 offset = 0;
2597 }
2598 if (insn & ARM_CP_RW_BIT) {
2599 /* vfp->arm */
2600 tmp = neon_load_reg(rn, pass);
2601 switch (size) {
2602 case 0:
2603 if (offset)
2604 tcg_gen_shri_i32(tmp, tmp, offset);
2605 if (insn & (1 << 23))
2606 gen_uxtb(tmp);
2607 else
2608 gen_sxtb(tmp);
2609 break;
2610 case 1:
2611 if (insn & (1 << 23)) {
2612 if (offset) {
2613 tcg_gen_shri_i32(tmp, tmp, 16);
2614 } else {
2615 gen_uxth(tmp);
2616 }
2617 } else {
2618 if (offset) {
2619 tcg_gen_sari_i32(tmp, tmp, 16);
2620 } else {
2621 gen_sxth(tmp);
2622 }
2623 }
2624 break;
2625 case 2:
2626 break;
2627 }
2628 store_reg(s, rd, tmp);
2629 } else {
2630 /* arm->vfp */
2631 tmp = load_reg(s, rd);
2632 if (insn & (1 << 23)) {
2633 /* VDUP */
2634 if (size == 0) {
2635 gen_neon_dup_u8(tmp, 0);
2636 } else if (size == 1) {
2637 gen_neon_dup_low16(tmp);
2638 }
2639 for (n = 0; n <= pass * 2; n++) {
2640 tmp2 = tcg_temp_new_i32();
2641 tcg_gen_mov_i32(tmp2, tmp);
2642 neon_store_reg(rn, n, tmp2);
2643 }
2644 neon_store_reg(rn, n, tmp);
2645 } else {
2646 /* VMOV */
2647 switch (size) {
2648 case 0:
2649 tmp2 = neon_load_reg(rn, pass);
2650 tcg_gen_deposit_i32(tmp, tmp2, tmp, offset, 8);
2651 tcg_temp_free_i32(tmp2);
2652 break;
2653 case 1:
2654 tmp2 = neon_load_reg(rn, pass);
2655 tcg_gen_deposit_i32(tmp, tmp2, tmp, offset, 16);
2656 tcg_temp_free_i32(tmp2);
2657 break;
2658 case 2:
2659 break;
2660 }
2661 neon_store_reg(rn, pass, tmp);
2662 }
2663 }
2664 } else { /* !dp */
2665 if ((insn & 0x6f) != 0x00)
2666 return 1;
2667 rn = VFP_SREG_N(insn);
2668 if (insn & ARM_CP_RW_BIT) {
2669 /* vfp->arm */
2670 if (insn & (1 << 21)) {
2671 /* system register */
2672 rn >>= 1;
2673
2674 switch (rn) {
2675 case ARM_VFP_FPSID:
2676 /* VFP2 allows access to FSID from userspace.
2677 VFP3 restricts all id registers to privileged
2678 accesses. */
2679 if (IS_USER(s)
2680 && arm_feature(env, ARM_FEATURE_VFP3))
2681 return 1;
2682 tmp = load_cpu_field(vfp.xregs[rn]);
2683 break;
2684 case ARM_VFP_FPEXC:
2685 if (IS_USER(s))
2686 return 1;
2687 tmp = load_cpu_field(vfp.xregs[rn]);
2688 break;
2689 case ARM_VFP_FPINST:
2690 case ARM_VFP_FPINST2:
2691 /* Not present in VFP3. */
2692 if (IS_USER(s)
2693 || arm_feature(env, ARM_FEATURE_VFP3))
2694 return 1;
2695 tmp = load_cpu_field(vfp.xregs[rn]);
2696 break;
2697 case ARM_VFP_FPSCR:
2698 if (rd == 15) {
2699 tmp = load_cpu_field(vfp.xregs[ARM_VFP_FPSCR]);
2700 tcg_gen_andi_i32(tmp, tmp, 0xf0000000);
2701 } else {
2702 tmp = tcg_temp_new_i32();
2703 gen_helper_vfp_get_fpscr(tmp, cpu_env);
2704 }
2705 break;
2706 case ARM_VFP_MVFR0:
2707 case ARM_VFP_MVFR1:
2708 if (IS_USER(s)
2709 || !arm_feature(env, ARM_FEATURE_MVFR))
2710 return 1;
2711 tmp = load_cpu_field(vfp.xregs[rn]);
2712 break;
2713 default:
2714 return 1;
2715 }
2716 } else {
2717 gen_mov_F0_vreg(0, rn);
2718 tmp = gen_vfp_mrs();
2719 }
2720 if (rd == 15) {
2721 /* Set the 4 flag bits in the CPSR. */
2722 gen_set_nzcv(tmp);
2723 tcg_temp_free_i32(tmp);
2724 } else {
2725 store_reg(s, rd, tmp);
2726 }
2727 } else {
2728 /* arm->vfp */
2729 if (insn & (1 << 21)) {
2730 rn >>= 1;
2731 /* system register */
2732 switch (rn) {
2733 case ARM_VFP_FPSID:
2734 case ARM_VFP_MVFR0:
2735 case ARM_VFP_MVFR1:
2736 /* Writes are ignored. */
2737 break;
2738 case ARM_VFP_FPSCR:
2739 tmp = load_reg(s, rd);
2740 gen_helper_vfp_set_fpscr(cpu_env, tmp);
2741 tcg_temp_free_i32(tmp);
2742 gen_lookup_tb(s);
2743 break;
2744 case ARM_VFP_FPEXC:
2745 if (IS_USER(s))
2746 return 1;
2747 /* TODO: VFP subarchitecture support.
2748 * For now, keep the EN bit only */
2749 tmp = load_reg(s, rd);
2750 tcg_gen_andi_i32(tmp, tmp, 1 << 30);
2751 store_cpu_field(tmp, vfp.xregs[rn]);
2752 gen_lookup_tb(s);
2753 break;
2754 case ARM_VFP_FPINST:
2755 case ARM_VFP_FPINST2:
2756 tmp = load_reg(s, rd);
2757 store_cpu_field(tmp, vfp.xregs[rn]);
2758 break;
2759 default:
2760 return 1;
2761 }
2762 } else {
2763 tmp = load_reg(s, rd);
2764 gen_vfp_msr(tmp);
2765 gen_mov_vreg_F0(0, rn);
2766 }
2767 }
2768 }
2769 } else {
2770 /* data processing */
2771 /* The opcode is in bits 23, 21, 20 and 6. */
2772 op = ((insn >> 20) & 8) | ((insn >> 19) & 6) | ((insn >> 6) & 1);
2773 if (dp) {
2774 if (op == 15) {
2775 /* rn is opcode */
2776 rn = ((insn >> 15) & 0x1e) | ((insn >> 7) & 1);
2777 } else {
2778 /* rn is register number */
2779 VFP_DREG_N(rn, insn);
2780 }
2781
2782 if (op == 15 && (rn == 15 || ((rn & 0x1c) == 0x18))) {
2783 /* Integer or single precision destination. */
2784 rd = VFP_SREG_D(insn);
2785 } else {
2786 VFP_DREG_D(rd, insn);
2787 }
2788 if (op == 15 &&
2789 (((rn & 0x1c) == 0x10) || ((rn & 0x14) == 0x14))) {
2790 /* VCVT from int is always from S reg regardless of dp bit.
2791 * VCVT with immediate frac_bits has same format as SREG_M
2792 */
2793 rm = VFP_SREG_M(insn);
2794 } else {
2795 VFP_DREG_M(rm, insn);
2796 }
2797 } else {
2798 rn = VFP_SREG_N(insn);
2799 if (op == 15 && rn == 15) {
2800 /* Double precision destination. */
2801 VFP_DREG_D(rd, insn);
2802 } else {
2803 rd = VFP_SREG_D(insn);
2804 }
2805 /* NB that we implicitly rely on the encoding for the frac_bits
2806 * in VCVT of fixed to float being the same as that of an SREG_M
2807 */
2808 rm = VFP_SREG_M(insn);
2809 }
2810
2811 veclen = s->vec_len;
2812 if (op == 15 && rn > 3)
2813 veclen = 0;
2814
2815 /* Shut up compiler warnings. */
2816 delta_m = 0;
2817 delta_d = 0;
2818 bank_mask = 0;
2819
2820 if (veclen > 0) {
2821 if (dp)
2822 bank_mask = 0xc;
2823 else
2824 bank_mask = 0x18;
2825
2826 /* Figure out what type of vector operation this is. */
2827 if ((rd & bank_mask) == 0) {
2828 /* scalar */
2829 veclen = 0;
2830 } else {
2831 if (dp)
2832 delta_d = (s->vec_stride >> 1) + 1;
2833 else
2834 delta_d = s->vec_stride + 1;
2835
2836 if ((rm & bank_mask) == 0) {
2837 /* mixed scalar/vector */
2838 delta_m = 0;
2839 } else {
2840 /* vector */
2841 delta_m = delta_d;
2842 }
2843 }
2844 }
2845
2846 /* Load the initial operands. */
2847 if (op == 15) {
2848 switch (rn) {
2849 case 16:
2850 case 17:
2851 /* Integer source */
2852 gen_mov_F0_vreg(0, rm);
2853 break;
2854 case 8:
2855 case 9:
2856 /* Compare */
2857 gen_mov_F0_vreg(dp, rd);
2858 gen_mov_F1_vreg(dp, rm);
2859 break;
2860 case 10:
2861 case 11:
2862 /* Compare with zero */
2863 gen_mov_F0_vreg(dp, rd);
2864 gen_vfp_F1_ld0(dp);
2865 break;
2866 case 20:
2867 case 21:
2868 case 22:
2869 case 23:
2870 case 28:
2871 case 29:
2872 case 30:
2873 case 31:
2874 /* Source and destination the same. */
2875 gen_mov_F0_vreg(dp, rd);
2876 break;
2877 case 4:
2878 case 5:
2879 case 6:
2880 case 7:
2881 /* VCVTB, VCVTT: only present with the halfprec extension,
2882 * UNPREDICTABLE if bit 8 is set (we choose to UNDEF)
2883 */
2884 if (dp || !arm_feature(env, ARM_FEATURE_VFP_FP16)) {
2885 return 1;
2886 }
2887 /* Otherwise fall through */
2888 default:
2889 /* One source operand. */
2890 gen_mov_F0_vreg(dp, rm);
2891 break;
2892 }
2893 } else {
2894 /* Two source operands. */
2895 gen_mov_F0_vreg(dp, rn);
2896 gen_mov_F1_vreg(dp, rm);
2897 }
2898
2899 for (;;) {
2900 /* Perform the calculation. */
2901 switch (op) {
2902 case 0: /* VMLA: fd + (fn * fm) */
2903 /* Note that order of inputs to the add matters for NaNs */
2904 gen_vfp_F1_mul(dp);
2905 gen_mov_F0_vreg(dp, rd);
2906 gen_vfp_add(dp);
2907 break;
2908 case 1: /* VMLS: fd + -(fn * fm) */
2909 gen_vfp_mul(dp);
2910 gen_vfp_F1_neg(dp);
2911 gen_mov_F0_vreg(dp, rd);
2912 gen_vfp_add(dp);
2913 break;
2914 case 2: /* VNMLS: -fd + (fn * fm) */
2915 /* Note that it isn't valid to replace (-A + B) with (B - A)
2916 * or similar plausible looking simplifications
2917 * because this will give wrong results for NaNs.
2918 */
2919 gen_vfp_F1_mul(dp);
2920 gen_mov_F0_vreg(dp, rd);
2921 gen_vfp_neg(dp);
2922 gen_vfp_add(dp);
2923 break;
2924 case 3: /* VNMLA: -fd + -(fn * fm) */
2925 gen_vfp_mul(dp);
2926 gen_vfp_F1_neg(dp);
2927 gen_mov_F0_vreg(dp, rd);
2928 gen_vfp_neg(dp);
2929 gen_vfp_add(dp);
2930 break;
2931 case 4: /* mul: fn * fm */
2932 gen_vfp_mul(dp);
2933 break;
2934 case 5: /* nmul: -(fn * fm) */
2935 gen_vfp_mul(dp);
2936 gen_vfp_neg(dp);
2937 break;
2938 case 6: /* add: fn + fm */
2939 gen_vfp_add(dp);
2940 break;
2941 case 7: /* sub: fn - fm */
2942 gen_vfp_sub(dp);
2943 break;
2944 case 8: /* div: fn / fm */
2945 gen_vfp_div(dp);
2946 break;
2947 case 10: /* VFNMA : fd = muladd(-fd, fn, fm) */
2948 case 11: /* VFNMS : fd = muladd(-fd, -fn, fm) */
2949 case 12: /* VFMA : fd = muladd( fd, fn, fm) */
2950 case 13: /* VFMS : fd = muladd( fd, -fn, fm) */
2951 /* These are fused multiply-add, and must be done as one
2952 * floating point operation with no rounding between the
2953 * multiplication and addition steps.
2954 * NB that doing the negations here as separate steps is
2955 * correct : an input NaN should come out with its sign bit
2956 * flipped if it is a negated-input.
2957 */
2958 if (!arm_feature(env, ARM_FEATURE_VFP4)) {
2959 return 1;
2960 }
2961 if (dp) {
2962 TCGv_ptr fpst;
2963 TCGv_i64 frd;
2964 if (op & 1) {
2965 /* VFNMS, VFMS */
2966 gen_helper_vfp_negd(cpu_F0d, cpu_F0d);
2967 }
2968 frd = tcg_temp_new_i64();
2969 tcg_gen_ld_f64(frd, cpu_env, vfp_reg_offset(dp, rd));
2970 if (op & 2) {
2971 /* VFNMA, VFNMS */
2972 gen_helper_vfp_negd(frd, frd);
2973 }
2974 fpst = get_fpstatus_ptr(0);
2975 gen_helper_vfp_muladdd(cpu_F0d, cpu_F0d,
2976 cpu_F1d, frd, fpst);
2977 tcg_temp_free_ptr(fpst);
2978 tcg_temp_free_i64(frd);
2979 } else {
2980 TCGv_ptr fpst;
2981 TCGv_i32 frd;
2982 if (op & 1) {
2983 /* VFNMS, VFMS */
2984 gen_helper_vfp_negs(cpu_F0s, cpu_F0s);
2985 }
2986 frd = tcg_temp_new_i32();
2987 tcg_gen_ld_f32(frd, cpu_env, vfp_reg_offset(dp, rd));
2988 if (op & 2) {
2989 gen_helper_vfp_negs(frd, frd);
2990 }
2991 fpst = get_fpstatus_ptr(0);
2992 gen_helper_vfp_muladds(cpu_F0s, cpu_F0s,
2993 cpu_F1s, frd, fpst);
2994 tcg_temp_free_ptr(fpst);
2995 tcg_temp_free_i32(frd);
2996 }
2997 break;
2998 case 14: /* fconst */
2999 if (!arm_feature(env, ARM_FEATURE_VFP3))
3000 return 1;
3001
3002 n = (insn << 12) & 0x80000000;
3003 i = ((insn >> 12) & 0x70) | (insn & 0xf);
3004 if (dp) {
3005 if (i & 0x40)
3006 i |= 0x3f80;
3007 else
3008 i |= 0x4000;
3009 n |= i << 16;
3010 tcg_gen_movi_i64(cpu_F0d, ((uint64_t)n) << 32);
3011 } else {
3012 if (i & 0x40)
3013 i |= 0x780;
3014 else
3015 i |= 0x800;
3016 n |= i << 19;
3017 tcg_gen_movi_i32(cpu_F0s, n);
3018 }
3019 break;
3020 case 15: /* extension space */
3021 switch (rn) {
3022 case 0: /* cpy */
3023 /* no-op */
3024 break;
3025 case 1: /* abs */
3026 gen_vfp_abs(dp);
3027 break;
3028 case 2: /* neg */
3029 gen_vfp_neg(dp);
3030 break;
3031 case 3: /* sqrt */
3032 gen_vfp_sqrt(dp);
3033 break;
3034 case 4: /* vcvtb.f32.f16 */
3035 tmp = gen_vfp_mrs();
3036 tcg_gen_ext16u_i32(tmp, tmp);
3037 gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp, cpu_env);
3038 tcg_temp_free_i32(tmp);
3039 break;
3040 case 5: /* vcvtt.f32.f16 */
3041 tmp = gen_vfp_mrs();
3042 tcg_gen_shri_i32(tmp, tmp, 16);
3043 gen_helper_vfp_fcvt_f16_to_f32(cpu_F0s, tmp, cpu_env);
3044 tcg_temp_free_i32(tmp);
3045 break;
3046 case 6: /* vcvtb.f16.f32 */
3047 tmp = tcg_temp_new_i32();
3048 gen_helper_vfp_fcvt_f32_to_f16(tmp, cpu_F0s, cpu_env);
3049 gen_mov_F0_vreg(0, rd);
3050 tmp2 = gen_vfp_mrs();
3051 tcg_gen_andi_i32(tmp2, tmp2, 0xffff0000);
3052 tcg_gen_or_i32(tmp, tmp, tmp2);
3053 tcg_temp_free_i32(tmp2);
3054 gen_vfp_msr(tmp);
3055 break;
3056 case 7: /* vcvtt.f16.f32 */
3057 tmp = tcg_temp_new_i32();
3058 gen_helper_vfp_fcvt_f32_to_f16(tmp, cpu_F0s, cpu_env);
3059 tcg_gen_shli_i32(tmp, tmp, 16);
3060 gen_mov_F0_vreg(0, rd);
3061 tmp2 = gen_vfp_mrs();
3062 tcg_gen_ext16u_i32(tmp2, tmp2);
3063 tcg_gen_or_i32(tmp, tmp, tmp2);
3064 tcg_temp_free_i32(tmp2);
3065 gen_vfp_msr(tmp);
3066 break;
3067 case 8: /* cmp */
3068 gen_vfp_cmp(dp);
3069 break;
3070 case 9: /* cmpe */
3071 gen_vfp_cmpe(dp);
3072 break;
3073 case 10: /* cmpz */
3074 gen_vfp_cmp(dp);
3075 break;
3076 case 11: /* cmpez */
3077 gen_vfp_F1_ld0(dp);
3078 gen_vfp_cmpe(dp);
3079 break;
3080 case 15: /* single<->double conversion */
3081 if (dp)
3082 gen_helper_vfp_fcvtsd(cpu_F0s, cpu_F0d, cpu_env);
3083 else
3084 gen_helper_vfp_fcvtds(cpu_F0d, cpu_F0s, cpu_env);
3085 break;
3086 case 16: /* fuito */
3087 gen_vfp_uito(dp, 0);
3088 break;
3089 case 17: /* fsito */
3090 gen_vfp_sito(dp, 0);
3091 break;
3092 case 20: /* fshto */
3093 if (!arm_feature(env, ARM_FEATURE_VFP3))
3094 return 1;
3095 gen_vfp_shto(dp, 16 - rm, 0);
3096 break;
3097 case 21: /* fslto */
3098 if (!arm_feature(env, ARM_FEATURE_VFP3))
3099 return 1;
3100 gen_vfp_slto(dp, 32 - rm, 0);
3101 break;
3102 case 22: /* fuhto */
3103 if (!arm_feature(env, ARM_FEATURE_VFP3))
3104 return 1;
3105 gen_vfp_uhto(dp, 16 - rm, 0);
3106 break;
3107 case 23: /* fulto */
3108 if (!arm_feature(env, ARM_FEATURE_VFP3))
3109 return 1;
3110 gen_vfp_ulto(dp, 32 - rm, 0);
3111 break;
3112 case 24: /* ftoui */
3113 gen_vfp_toui(dp, 0);
3114 break;
3115 case 25: /* ftouiz */
3116 gen_vfp_touiz(dp, 0);
3117 break;
3118 case 26: /* ftosi */
3119 gen_vfp_tosi(dp, 0);
3120 break;
3121 case 27: /* ftosiz */
3122 gen_vfp_tosiz(dp, 0);
3123 break;
3124 case 28: /* ftosh */
3125 if (!arm_feature(env, ARM_FEATURE_VFP3))
3126 return 1;
3127 gen_vfp_tosh(dp, 16 - rm, 0);
3128 break;
3129 case 29: /* ftosl */
3130 if (!arm_feature(env, ARM_FEATURE_VFP3))
3131 return 1;
3132 gen_vfp_tosl(dp, 32 - rm, 0);
3133 break;
3134 case 30: /* ftouh */
3135 if (!arm_feature(env, ARM_FEATURE_VFP3))
3136 return 1;
3137 gen_vfp_touh(dp, 16 - rm, 0);
3138 break;
3139 case 31: /* ftoul */
3140 if (!arm_feature(env, ARM_FEATURE_VFP3))
3141 return 1;
3142 gen_vfp_toul(dp, 32 - rm, 0);
3143 break;
3144 default: /* undefined */
3145 return 1;
3146 }
3147 break;
3148 default: /* undefined */
3149 return 1;
3150 }
3151
3152 /* Write back the result. */
3153 if (op == 15 && (rn >= 8 && rn <= 11))
3154 ; /* Comparison, do nothing. */
3155 else if (op == 15 && dp && ((rn & 0x1c) == 0x18))
3156 /* VCVT double to int: always integer result. */
3157 gen_mov_vreg_F0(0, rd);
3158 else if (op == 15 && rn == 15)
3159 /* conversion */
3160 gen_mov_vreg_F0(!dp, rd);
3161 else
3162 gen_mov_vreg_F0(dp, rd);
3163
3164 /* break out of the loop if we have finished */
3165 if (veclen == 0)
3166 break;
3167
3168 if (op == 15 && delta_m == 0) {
3169 /* single source one-many */
3170 while (veclen--) {
3171 rd = ((rd + delta_d) & (bank_mask - 1))
3172 | (rd & bank_mask);
3173 gen_mov_vreg_F0(dp, rd);
3174 }
3175 break;
3176 }
3177 /* Setup the next operands. */
3178 veclen--;
3179 rd = ((rd + delta_d) & (bank_mask - 1))
3180 | (rd & bank_mask);
3181
3182 if (op == 15) {
3183 /* One source operand. */
3184 rm = ((rm + delta_m) & (bank_mask - 1))
3185 | (rm & bank_mask);
3186 gen_mov_F0_vreg(dp, rm);
3187 } else {
3188 /* Two source operands. */
3189 rn = ((rn + delta_d) & (bank_mask - 1))
3190 | (rn & bank_mask);
3191 gen_mov_F0_vreg(dp, rn);
3192 if (delta_m) {
3193 rm = ((rm + delta_m) & (bank_mask - 1))
3194 | (rm & bank_mask);
3195 gen_mov_F1_vreg(dp, rm);
3196 }
3197 }
3198 }
3199 }
3200 break;
3201 case 0xc:
3202 case 0xd:
3203 if ((insn & 0x03e00000) == 0x00400000) {
3204 /* two-register transfer */
3205 rn = (insn >> 16) & 0xf;
3206 rd = (insn >> 12) & 0xf;
3207 if (dp) {
3208 VFP_DREG_M(rm, insn);
3209 } else {
3210 rm = VFP_SREG_M(insn);
3211 }
3212
3213 if (insn & ARM_CP_RW_BIT) {
3214 /* vfp->arm */
3215 if (dp) {
3216 gen_mov_F0_vreg(0, rm * 2);
3217 tmp = gen_vfp_mrs();
3218 store_reg(s, rd, tmp);
3219 gen_mov_F0_vreg(0, rm * 2 + 1);
3220 tmp = gen_vfp_mrs();
3221 store_reg(s, rn, tmp);
3222 } else {
3223 gen_mov_F0_vreg(0, rm);
3224 tmp = gen_vfp_mrs();
3225 store_reg(s, rd, tmp);
3226 gen_mov_F0_vreg(0, rm + 1);
3227 tmp = gen_vfp_mrs();
3228 store_reg(s, rn, tmp);
3229 }
3230 } else {
3231 /* arm->vfp */
3232 if (dp) {
3233 tmp = load_reg(s, rd);
3234 gen_vfp_msr(tmp);
3235 gen_mov_vreg_F0(0, rm * 2);
3236 tmp = load_reg(s, rn);
3237 gen_vfp_msr(tmp);
3238 gen_mov_vreg_F0(0, rm * 2 + 1);
3239 } else {
3240 tmp = load_reg(s, rd);
3241 gen_vfp_msr(tmp);
3242 gen_mov_vreg_F0(0, rm);
3243 tmp = load_reg(s, rn);
3244 gen_vfp_msr(tmp);
3245 gen_mov_vreg_F0(0, rm + 1);
3246 }
3247 }
3248 } else {
3249 /* Load/store */
3250 rn = (insn >> 16) & 0xf;
3251 if (dp)
3252 VFP_DREG_D(rd, insn);
3253 else
3254 rd = VFP_SREG_D(insn);
3255 if ((insn & 0x01200000) == 0x01000000) {
3256 /* Single load/store */
3257 offset = (insn & 0xff) << 2;
3258 if ((insn & (1 << 23)) == 0)
3259 offset = -offset;
3260 if (s->thumb && rn == 15) {
3261 /* This is actually UNPREDICTABLE */
3262 addr = tcg_temp_new_i32();
3263 tcg_gen_movi_i32(addr, s->pc & ~2);
3264 } else {
3265 addr = load_reg(s, rn);
3266 }
3267 tcg_gen_addi_i32(addr, addr, offset);
3268 if (insn & (1 << 20)) {
3269 gen_vfp_ld(s, dp, addr);
3270 gen_mov_vreg_F0(dp, rd);
3271 } else {
3272 gen_mov_F0_vreg(dp, rd);
3273 gen_vfp_st(s, dp, addr);
3274 }
3275 tcg_temp_free_i32(addr);
3276 } else {
3277 /* load/store multiple */
3278 int w = insn & (1 << 21);
3279 if (dp)
3280 n = (insn >> 1) & 0x7f;
3281 else
3282 n = insn & 0xff;
3283
3284 if (w && !(((insn >> 23) ^ (insn >> 24)) & 1)) {
3285 /* P == U , W == 1 => UNDEF */
3286 return 1;
3287 }
3288 if (n == 0 || (rd + n) > 32 || (dp && n > 16)) {
3289 /* UNPREDICTABLE cases for bad immediates: we choose to
3290 * UNDEF to avoid generating huge numbers of TCG ops
3291 */
3292 return 1;
3293 }
3294 if (rn == 15 && w) {
3295 /* writeback to PC is UNPREDICTABLE, we choose to UNDEF */
3296 return 1;
3297 }
3298
3299 if (s->thumb && rn == 15) {
3300 /* This is actually UNPREDICTABLE */
3301 addr = tcg_temp_new_i32();
3302 tcg_gen_movi_i32(addr, s->pc & ~2);
3303 } else {
3304 addr = load_reg(s, rn);
3305 }
3306 if (insn & (1 << 24)) /* pre-decrement */
3307 tcg_gen_addi_i32(addr, addr, -((insn & 0xff) << 2));
3308
3309 if (dp)
3310 offset = 8;
3311 else
3312 offset = 4;
3313 for (i = 0; i < n; i++) {
3314 if (insn & ARM_CP_RW_BIT) {
3315 /* load */
3316 gen_vfp_ld(s, dp, addr);
3317 gen_mov_vreg_F0(dp, rd + i);
3318 } else {
3319 /* store */
3320 gen_mov_F0_vreg(dp, rd + i);
3321 gen_vfp_st(s, dp, addr);
3322 }
3323 tcg_gen_addi_i32(addr, addr, offset);
3324 }
3325 if (w) {
3326 /* writeback */
3327 if (insn & (1 << 24))
3328 offset = -offset * n;
3329 else if (dp && (insn & 1))
3330 offset = 4;
3331 else
3332 offset = 0;
3333
3334 if (offset != 0)
3335 tcg_gen_addi_i32(addr, addr, offset);
3336 store_reg(s, rn, addr);
3337 } else {
3338 tcg_temp_free_i32(addr);
3339 }
3340 }
3341 }
3342 break;
3343 default:
3344 /* Should never happen. */
3345 return 1;
3346 }
3347 return 0;
3348 }
3349
3350 static inline void gen_goto_tb(DisasContext *s, int n, uint32_t dest)
3351 {
3352 TranslationBlock *tb;
3353
3354 tb = s->tb;
3355 if ((tb->pc & TARGET_PAGE_MASK) == (dest & TARGET_PAGE_MASK)) {
3356 tcg_gen_goto_tb(n);
3357 gen_set_pc_im(dest);
3358 tcg_gen_exit_tb((tcg_target_long)tb + n);
3359 } else {
3360 gen_set_pc_im(dest);
3361 tcg_gen_exit_tb(0);
3362 }
3363 }
3364
3365 static inline void gen_jmp (DisasContext *s, uint32_t dest)
3366 {
3367 if (unlikely(s->singlestep_enabled)) {
3368 /* An indirect jump so that we still trigger the debug exception. */
3369 if (s->thumb)
3370 dest |= 1;
3371 gen_bx_im(s, dest);
3372 } else {
3373 gen_goto_tb(s, 0, dest);
3374 s->is_jmp = DISAS_TB_JUMP;
3375 }
3376 }
3377
3378 static inline void gen_mulxy(TCGv_i32 t0, TCGv_i32 t1, int x, int y)
3379 {
3380 if (x)
3381 tcg_gen_sari_i32(t0, t0, 16);
3382 else
3383 gen_sxth(t0);
3384 if (y)
3385 tcg_gen_sari_i32(t1, t1, 16);
3386 else
3387 gen_sxth(t1);
3388 tcg_gen_mul_i32(t0, t0, t1);
3389 }
3390
3391 /* Return the mask of PSR bits set by a MSR instruction. */
3392 static uint32_t msr_mask(CPUARMState *env, DisasContext *s, int flags, int spsr) {
3393 uint32_t mask;
3394
3395 mask = 0;
3396 if (flags & (1 << 0))
3397 mask |= 0xff;
3398 if (flags & (1 << 1))
3399 mask |= 0xff00;
3400 if (flags & (1 << 2))
3401 mask |= 0xff0000;
3402 if (flags & (1 << 3))
3403 mask |= 0xff000000;
3404
3405 /* Mask out undefined bits. */
3406 mask &= ~CPSR_RESERVED;
3407 if (!arm_feature(env, ARM_FEATURE_V4T))
3408 mask &= ~CPSR_T;
3409 if (!arm_feature(env, ARM_FEATURE_V5))
3410 mask &= ~CPSR_Q; /* V5TE in reality*/
3411 if (!arm_feature(env, ARM_FEATURE_V6))
3412 mask &= ~(CPSR_E | CPSR_GE);
3413 if (!arm_feature(env, ARM_FEATURE_THUMB2))
3414 mask &= ~CPSR_IT;
3415 /* Mask out execution state bits. */
3416 if (!spsr)
3417 mask &= ~CPSR_EXEC;
3418 /* Mask out privileged bits. */
3419 if (IS_USER(s))
3420 mask &= CPSR_USER;
3421 return mask;
3422 }
3423
3424 /* Returns nonzero if access to the PSR is not permitted. Marks t0 as dead. */
3425 static int gen_set_psr(DisasContext *s, uint32_t mask, int spsr, TCGv_i32 t0)
3426 {
3427 TCGv_i32 tmp;
3428 if (spsr) {
3429 /* ??? This is also undefined in system mode. */
3430 if (IS_USER(s))
3431 return 1;
3432
3433 tmp = load_cpu_field(spsr);
3434 tcg_gen_andi_i32(tmp, tmp, ~mask);
3435 tcg_gen_andi_i32(t0, t0, mask);
3436 tcg_gen_or_i32(tmp, tmp, t0);
3437 store_cpu_field(tmp, spsr);
3438 } else {
3439 gen_set_cpsr(t0, mask);
3440 }
3441 tcg_temp_free_i32(t0);
3442 gen_lookup_tb(s);
3443 return 0;
3444 }
3445
3446 /* Returns nonzero if access to the PSR is not permitted. */
3447 static int gen_set_psr_im(DisasContext *s, uint32_t mask, int spsr, uint32_t val)
3448 {
3449 TCGv_i32 tmp;
3450 tmp = tcg_temp_new_i32();
3451 tcg_gen_movi_i32(tmp, val);
3452 return gen_set_psr(s, mask, spsr, tmp);
3453 }
3454
3455 /* Generate an old-style exception return. Marks pc as dead. */
3456 static void gen_exception_return(DisasContext *s, TCGv_i32 pc)
3457 {
3458 TCGv_i32 tmp;
3459 store_reg(s, 15, pc);
3460 tmp = load_cpu_field(spsr);
3461 gen_set_cpsr(tmp, 0xffffffff);
3462 tcg_temp_free_i32(tmp);
3463 s->is_jmp = DISAS_UPDATE;
3464 }
3465
3466 /* Generate a v6 exception return. Marks both values as dead. */
3467 static void gen_rfe(DisasContext *s, TCGv_i32 pc, TCGv_i32 cpsr)
3468 {
3469 gen_set_cpsr(cpsr, 0xffffffff);
3470 tcg_temp_free_i32(cpsr);
3471 store_reg(s, 15, pc);
3472 s->is_jmp = DISAS_UPDATE;
3473 }
3474
3475 static inline void
3476 gen_set_condexec (DisasContext *s)
3477 {
3478 if (s->condexec_mask) {
3479 uint32_t val = (s->condexec_cond << 4) | (s->condexec_mask >> 1);
3480 TCGv_i32 tmp = tcg_temp_new_i32();
3481 tcg_gen_movi_i32(tmp, val);
3482 store_cpu_field(tmp, condexec_bits);
3483 }
3484 }
3485
3486 static void gen_exception_insn(DisasContext *s, int offset, int excp)
3487 {
3488 gen_set_condexec(s);
3489 gen_set_pc_im(s->pc - offset);
3490 gen_exception(excp);
3491 s->is_jmp = DISAS_JUMP;
3492 }
3493
3494 static void gen_nop_hint(DisasContext *s, int val)
3495 {
3496 switch (val) {
3497 case 3: /* wfi */
3498 gen_set_pc_im(s->pc);
3499 s->is_jmp = DISAS_WFI;
3500 break;
3501 case 2: /* wfe */
3502 case 4: /* sev */
3503 /* TODO: Implement SEV and WFE. May help SMP performance. */
3504 default: /* nop */
3505 break;
3506 }
3507 }
3508
3509 #define CPU_V001 cpu_V0, cpu_V0, cpu_V1
3510
3511 static inline void gen_neon_add(int size, TCGv_i32 t0, TCGv_i32 t1)
3512 {
3513 switch (size) {
3514 case 0: gen_helper_neon_add_u8(t0, t0, t1); break;
3515 case 1: gen_helper_neon_add_u16(t0, t0, t1); break;
3516 case 2: tcg_gen_add_i32(t0, t0, t1); break;
3517 default: abort();
3518 }
3519 }
3520
3521 static inline void gen_neon_rsb(int size, TCGv_i32 t0, TCGv_i32 t1)
3522 {
3523 switch (size) {
3524 case 0: gen_helper_neon_sub_u8(t0, t1, t0); break;
3525 case 1: gen_helper_neon_sub_u16(t0, t1, t0); break;
3526 case 2: tcg_gen_sub_i32(t0, t1, t0); break;
3527 default: return;
3528 }
3529 }
3530
3531 /* 32-bit pairwise ops end up the same as the elementwise versions. */
3532 #define gen_helper_neon_pmax_s32 gen_helper_neon_max_s32
3533 #define gen_helper_neon_pmax_u32 gen_helper_neon_max_u32
3534 #define gen_helper_neon_pmin_s32 gen_helper_neon_min_s32
3535 #define gen_helper_neon_pmin_u32 gen_helper_neon_min_u32
3536
3537 #define GEN_NEON_INTEGER_OP_ENV(name) do { \
3538 switch ((size << 1) | u) { \
3539 case 0: \
3540 gen_helper_neon_##name##_s8(tmp, cpu_env, tmp, tmp2); \
3541 break; \
3542 case 1: \
3543 gen_helper_neon_##name##_u8(tmp, cpu_env, tmp, tmp2); \
3544 break; \
3545 case 2: \
3546 gen_helper_neon_##name##_s16(tmp, cpu_env, tmp, tmp2); \
3547 break; \
3548 case 3: \
3549 gen_helper_neon_##name##_u16(tmp, cpu_env, tmp, tmp2); \
3550 break; \
3551 case 4: \
3552 gen_helper_neon_##name##_s32(tmp, cpu_env, tmp, tmp2); \
3553 break; \
3554 case 5: \
3555 gen_helper_neon_##name##_u32(tmp, cpu_env, tmp, tmp2); \
3556 break; \
3557 default: return 1; \
3558 }} while (0)
3559
3560 #define GEN_NEON_INTEGER_OP(name) do { \
3561 switch ((size << 1) | u) { \
3562 case 0: \
3563 gen_helper_neon_##name##_s8(tmp, tmp, tmp2); \
3564 break; \
3565 case 1: \
3566 gen_helper_neon_##name##_u8(tmp, tmp, tmp2); \
3567 break; \
3568 case 2: \
3569 gen_helper_neon_##name##_s16(tmp, tmp, tmp2); \
3570 break; \
3571 case 3: \
3572 gen_helper_neon_##name##_u16(tmp, tmp, tmp2); \
3573 break; \
3574 case 4: \
3575 gen_helper_neon_##name##_s32(tmp, tmp, tmp2); \
3576 break; \
3577 case 5: \
3578 gen_helper_neon_##name##_u32(tmp, tmp, tmp2); \
3579 break; \
3580 default: return 1; \
3581 }} while (0)
3582
3583 static TCGv_i32 neon_load_scratch(int scratch)
3584 {
3585 TCGv_i32 tmp = tcg_temp_new_i32();
3586 tcg_gen_ld_i32(tmp, cpu_env, offsetof(CPUARMState, vfp.scratch[scratch]));
3587 return tmp;
3588 }
3589
3590 static void neon_store_scratch(int scratch, TCGv_i32 var)
3591 {
3592 tcg_gen_st_i32(var, cpu_env, offsetof(CPUARMState, vfp.scratch[scratch]));
3593 tcg_temp_free_i32(var);
3594 }
3595
3596 static inline TCGv_i32 neon_get_scalar(int size, int reg)
3597 {
3598 TCGv_i32 tmp;
3599 if (size == 1) {
3600 tmp = neon_load_reg(reg & 7, reg >> 4);
3601 if (reg & 8) {
3602 gen_neon_dup_high16(tmp);
3603 } else {
3604 gen_neon_dup_low16(tmp);
3605 }
3606 } else {
3607 tmp = neon_load_reg(reg & 15, reg >> 4);
3608 }
3609 return tmp;
3610 }
3611
3612 static int gen_neon_unzip(int rd, int rm, int size, int q)
3613 {
3614 TCGv_i32 tmp, tmp2;
3615 if (!q && size == 2) {
3616 return 1;
3617 }
3618 tmp = tcg_const_i32(rd);
3619 tmp2 = tcg_const_i32(rm);
3620 if (q) {
3621 switch (size) {
3622 case 0:
3623 gen_helper_neon_qunzip8(cpu_env, tmp, tmp2);
3624 break;
3625 case 1:
3626 gen_helper_neon_qunzip16(cpu_env, tmp, tmp2);
3627 break;
3628 case 2:
3629 gen_helper_neon_qunzip32(cpu_env, tmp, tmp2);
3630 break;
3631 default:
3632 abort();
3633 }
3634 } else {
3635 switch (size) {
3636 case 0:
3637 gen_helper_neon_unzip8(cpu_env, tmp, tmp2);
3638 break;
3639 case 1:
3640 gen_helper_neon_unzip16(cpu_env, tmp, tmp2);
3641 break;
3642 default:
3643 abort();
3644 }
3645 }
3646 tcg_temp_free_i32(tmp);
3647 tcg_temp_free_i32(tmp2);
3648 return 0;
3649 }
3650
3651 static int gen_neon_zip(int rd, int rm, int size, int q)
3652 {
3653 TCGv_i32 tmp, tmp2;
3654 if (!q && size == 2) {
3655 return 1;
3656 }
3657 tmp = tcg_const_i32(rd);
3658 tmp2 = tcg_const_i32(rm);
3659 if (q) {
3660 switch (size) {
3661 case 0:
3662 gen_helper_neon_qzip8(cpu_env, tmp, tmp2);
3663 break;
3664 case 1:
3665 gen_helper_neon_qzip16(cpu_env, tmp, tmp2);
3666 break;
3667 case 2:
3668 gen_helper_neon_qzip32(cpu_env, tmp, tmp2);
3669 break;
3670 default:
3671 abort();
3672 }
3673 } else {
3674 switch (size) {
3675 case 0:
3676 gen_helper_neon_zip8(cpu_env, tmp, tmp2);
3677 break;
3678 case 1:
3679 gen_helper_neon_zip16(cpu_env, tmp, tmp2);
3680 break;
3681 default:
3682 abort();
3683 }
3684 }
3685 tcg_temp_free_i32(tmp);
3686 tcg_temp_free_i32(tmp2);
3687 return 0;
3688 }
3689
3690 static void gen_neon_trn_u8(TCGv_i32 t0, TCGv_i32 t1)
3691 {
3692 TCGv_i32 rd, tmp;
3693
3694 rd = tcg_temp_new_i32();
3695 tmp = tcg_temp_new_i32();
3696
3697 tcg_gen_shli_i32(rd, t0, 8);
3698 tcg_gen_andi_i32(rd, rd, 0xff00ff00);
3699 tcg_gen_andi_i32(tmp, t1, 0x00ff00ff);
3700 tcg_gen_or_i32(rd, rd, tmp);
3701
3702 tcg_gen_shri_i32(t1, t1, 8);
3703 tcg_gen_andi_i32(t1, t1, 0x00ff00ff);
3704 tcg_gen_andi_i32(tmp, t0, 0xff00ff00);
3705 tcg_gen_or_i32(t1, t1, tmp);
3706 tcg_gen_mov_i32(t0, rd);
3707
3708 tcg_temp_free_i32(tmp);
3709 tcg_temp_free_i32(rd);
3710 }
3711
3712 static void gen_neon_trn_u16(TCGv_i32 t0, TCGv_i32 t1)
3713 {
3714 TCGv_i32 rd, tmp;
3715
3716 rd = tcg_temp_new_i32();
3717 tmp = tcg_temp_new_i32();
3718
3719 tcg_gen_shli_i32(rd, t0, 16);
3720 tcg_gen_andi_i32(tmp, t1, 0xffff);
3721 tcg_gen_or_i32(rd, rd, tmp);
3722 tcg_gen_shri_i32(t1, t1, 16);
3723 tcg_gen_andi_i32(tmp, t0, 0xffff0000);
3724 tcg_gen_or_i32(t1, t1, tmp);
3725 tcg_gen_mov_i32(t0, rd);
3726
3727 tcg_temp_free_i32(tmp);
3728 tcg_temp_free_i32(rd);
3729 }
3730
3731
3732 static struct {
3733 int nregs;
3734 int interleave;
3735 int spacing;
3736 } neon_ls_element_type[11] = {
3737 {4, 4, 1},
3738 {4, 4, 2},
3739 {4, 1, 1},
3740 {4, 2, 1},
3741 {3, 3, 1},
3742 {3, 3, 2},
3743 {3, 1, 1},
3744 {1, 1, 1},
3745 {2, 2, 1},
3746 {2, 2, 2},
3747 {2, 1, 1}
3748 };
3749
3750 /* Translate a NEON load/store element instruction. Return nonzero if the
3751 instruction is invalid. */
3752 static int disas_neon_ls_insn(CPUARMState * env, DisasContext *s, uint32_t insn)
3753 {
3754 int rd, rn, rm;
3755 int op;
3756 int nregs;
3757 int interleave;
3758 int spacing;
3759 int stride;
3760 int size;
3761 int reg;
3762 int pass;
3763 int load;
3764 int shift;
3765 int n;
3766 TCGv_i32 addr;
3767 TCGv_i32 tmp;
3768 TCGv_i32 tmp2;
3769 TCGv_i64 tmp64;
3770
3771 if (!s->vfp_enabled)
3772 return 1;
3773 VFP_DREG_D(rd, insn);
3774 rn = (insn >> 16) & 0xf;
3775 rm = insn & 0xf;
3776 load = (insn & (1 << 21)) != 0;
3777 if ((insn & (1 << 23)) == 0) {
3778 /* Load store all elements. */
3779 op = (insn >> 8) & 0xf;
3780 size = (insn >> 6) & 3;
3781 if (op > 10)
3782 return 1;
3783 /* Catch UNDEF cases for bad values of align field */
3784 switch (op & 0xc) {
3785 case 4:
3786 if (((insn >> 5) & 1) == 1) {
3787 return 1;
3788 }
3789 break;
3790 case 8:
3791 if (((insn >> 4) & 3) == 3) {
3792 return 1;
3793 }
3794 break;
3795 default:
3796 break;
3797 }
3798 nregs = neon_ls_element_type[op].nregs;
3799 interleave = neon_ls_element_type[op].interleave;
3800 spacing = neon_ls_element_type[op].spacing;
3801 if (size == 3 && (interleave | spacing) != 1)
3802 return 1;
3803 addr = tcg_temp_new_i32();
3804 load_reg_var(s, addr, rn);
3805 stride = (1 << size) * interleave;
3806 for (reg = 0; reg < nregs; reg++) {
3807 if (interleave > 2 || (interleave == 2 && nregs == 2)) {
3808 load_reg_var(s, addr, rn);
3809 tcg_gen_addi_i32(addr, addr, (1 << size) * reg);
3810 } else if (interleave == 2 && nregs == 4 && reg == 2) {
3811 load_reg_var(s, addr, rn);
3812 tcg_gen_addi_i32(addr, addr, 1 << size);
3813 }
3814 if (size == 3) {
3815 tmp64 = tcg_temp_new_i64();
3816 if (load) {
3817 tcg_gen_qemu_ld64(tmp64, addr, IS_USER(s));
3818 neon_store_reg64(tmp64, rd);
3819 } else {
3820 neon_load_reg64(tmp64, rd);
3821 tcg_gen_qemu_st64(tmp64, addr, IS_USER(s));
3822 }
3823 tcg_temp_free_i64(tmp64);
3824 tcg_gen_addi_i32(addr, addr, stride);
3825 } else {
3826 for (pass = 0; pass < 2; pass++) {
3827 if (size == 2) {
3828 if (load) {
3829 tmp = tcg_temp_new_i32();
3830 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
3831 neon_store_reg(rd, pass, tmp);
3832 } else {
3833 tmp = neon_load_reg(rd, pass);
3834 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
3835 tcg_temp_free_i32(tmp);
3836 }
3837 tcg_gen_addi_i32(addr, addr, stride);
3838 } else if (size == 1) {
3839 if (load) {
3840 tmp = tcg_temp_new_i32();
3841 tcg_gen_qemu_ld16u(tmp, addr, IS_USER(s));
3842 tcg_gen_addi_i32(addr, addr, stride);
3843 tmp2 = tcg_temp_new_i32();
3844 tcg_gen_qemu_ld16u(tmp2, addr, IS_USER(s));
3845 tcg_gen_addi_i32(addr, addr, stride);
3846 tcg_gen_shli_i32(tmp2, tmp2, 16);
3847 tcg_gen_or_i32(tmp, tmp, tmp2);
3848 tcg_temp_free_i32(tmp2);
3849 neon_store_reg(rd, pass, tmp);
3850 } else {
3851 tmp = neon_load_reg(rd, pass);
3852 tmp2 = tcg_temp_new_i32();
3853 tcg_gen_shri_i32(tmp2, tmp, 16);
3854 tcg_gen_qemu_st16(tmp, addr, IS_USER(s));
3855 tcg_temp_free_i32(tmp);
3856 tcg_gen_addi_i32(addr, addr, stride);
3857 tcg_gen_qemu_st16(tmp2, addr, IS_USER(s));
3858 tcg_temp_free_i32(tmp2);
3859 tcg_gen_addi_i32(addr, addr, stride);
3860 }
3861 } else /* size == 0 */ {
3862 if (load) {
3863 TCGV_UNUSED_I32(tmp2);
3864 for (n = 0; n < 4; n++) {
3865 tmp = tcg_temp_new_i32();
3866 tcg_gen_qemu_ld8u(tmp, addr, IS_USER(s));
3867 tcg_gen_addi_i32(addr, addr, stride);
3868 if (n == 0) {
3869 tmp2 = tmp;
3870 } else {
3871 tcg_gen_shli_i32(tmp, tmp, n * 8);
3872 tcg_gen_or_i32(tmp2, tmp2, tmp);
3873 tcg_temp_free_i32(tmp);
3874 }
3875 }
3876 neon_store_reg(rd, pass, tmp2);
3877 } else {
3878 tmp2 = neon_load_reg(rd, pass);
3879 for (n = 0; n < 4; n++) {
3880 tmp = tcg_temp_new_i32();
3881 if (n == 0) {
3882 tcg_gen_mov_i32(tmp, tmp2);
3883 } else {
3884 tcg_gen_shri_i32(tmp, tmp2, n * 8);
3885 }
3886 tcg_gen_qemu_st8(tmp, addr, IS_USER(s));
3887 tcg_temp_free_i32(tmp);
3888 tcg_gen_addi_i32(addr, addr, stride);
3889 }
3890 tcg_temp_free_i32(tmp2);
3891 }
3892 }
3893 }
3894 }
3895 rd += spacing;
3896 }
3897 tcg_temp_free_i32(addr);
3898 stride = nregs * 8;
3899 } else {
3900 size = (insn >> 10) & 3;
3901 if (size == 3) {
3902 /* Load single element to all lanes. */
3903 int a = (insn >> 4) & 1;
3904 if (!load) {
3905 return 1;
3906 }
3907 size = (insn >> 6) & 3;
3908 nregs = ((insn >> 8) & 3) + 1;
3909
3910 if (size == 3) {
3911 if (nregs != 4 || a == 0) {
3912 return 1;
3913 }
3914 /* For VLD4 size==3 a == 1 means 32 bits at 16 byte alignment */
3915 size = 2;
3916 }
3917 if (nregs == 1 && a == 1 && size == 0) {
3918 return 1;
3919 }
3920 if (nregs == 3 && a == 1) {
3921 return 1;
3922 }
3923 addr = tcg_temp_new_i32();
3924 load_reg_var(s, addr, rn);
3925 if (nregs == 1) {
3926 /* VLD1 to all lanes: bit 5 indicates how many Dregs to write */
3927 tmp = gen_load_and_replicate(s, addr, size);
3928 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 0));
3929 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 1));
3930 if (insn & (1 << 5)) {
3931 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd + 1, 0));
3932 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd + 1, 1));
3933 }
3934 tcg_temp_free_i32(tmp);
3935 } else {
3936 /* VLD2/3/4 to all lanes: bit 5 indicates register stride */
3937 stride = (insn & (1 << 5)) ? 2 : 1;
3938 for (reg = 0; reg < nregs; reg++) {
3939 tmp = gen_load_and_replicate(s, addr, size);
3940 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 0));
3941 tcg_gen_st_i32(tmp, cpu_env, neon_reg_offset(rd, 1));
3942 tcg_temp_free_i32(tmp);
3943 tcg_gen_addi_i32(addr, addr, 1 << size);
3944 rd += stride;
3945 }
3946 }
3947 tcg_temp_free_i32(addr);
3948 stride = (1 << size) * nregs;
3949 } else {
3950 /* Single element. */
3951 int idx = (insn >> 4) & 0xf;
3952 pass = (insn >> 7) & 1;
3953 switch (size) {
3954 case 0:
3955 shift = ((insn >> 5) & 3) * 8;
3956 stride = 1;
3957 break;
3958 case 1:
3959 shift = ((insn >> 6) & 1) * 16;
3960 stride = (insn & (1 << 5)) ? 2 : 1;
3961 break;
3962 case 2:
3963 shift = 0;
3964 stride = (insn & (1 << 6)) ? 2 : 1;
3965 break;
3966 default:
3967 abort();
3968 }
3969 nregs = ((insn >> 8) & 3) + 1;
3970 /* Catch the UNDEF cases. This is unavoidably a bit messy. */
3971 switch (nregs) {
3972 case 1:
3973 if (((idx & (1 << size)) != 0) ||
3974 (size == 2 && ((idx & 3) == 1 || (idx & 3) == 2))) {
3975 return 1;
3976 }
3977 break;
3978 case 3:
3979 if ((idx & 1) != 0) {
3980 return 1;
3981 }
3982 /* fall through */
3983 case 2:
3984 if (size == 2 && (idx & 2) != 0) {
3985 return 1;
3986 }
3987 break;
3988 case 4:
3989 if ((size == 2) && ((idx & 3) == 3)) {
3990 return 1;
3991 }
3992 break;
3993 default:
3994 abort();
3995 }
3996 if ((rd + stride * (nregs - 1)) > 31) {
3997 /* Attempts to write off the end of the register file
3998 * are UNPREDICTABLE; we choose to UNDEF because otherwise
3999 * the neon_load_reg() would write off the end of the array.
4000 */
4001 return 1;
4002 }
4003 addr = tcg_temp_new_i32();
4004 load_reg_var(s, addr, rn);
4005 for (reg = 0; reg < nregs; reg++) {
4006 if (load) {
4007 tmp = tcg_temp_new_i32();
4008 switch (size) {
4009 case 0:
4010 tcg_gen_qemu_ld8u(tmp, addr, IS_USER(s));
4011 break;
4012 case 1:
4013 tcg_gen_qemu_ld16u(tmp, addr, IS_USER(s));
4014 break;
4015 case 2:
4016 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
4017 break;
4018 default: /* Avoid compiler warnings. */
4019 abort();
4020 }
4021 if (size != 2) {
4022 tmp2 = neon_load_reg(rd, pass);
4023 tcg_gen_deposit_i32(tmp, tmp2, tmp,
4024 shift, size ? 16 : 8);
4025 tcg_temp_free_i32(tmp2);
4026 }
4027 neon_store_reg(rd, pass, tmp);
4028 } else { /* Store */
4029 tmp = neon_load_reg(rd, pass);
4030 if (shift)
4031 tcg_gen_shri_i32(tmp, tmp, shift);
4032 switch (size) {
4033 case 0:
4034 tcg_gen_qemu_st8(tmp, addr, IS_USER(s));
4035 break;
4036 case 1:
4037 tcg_gen_qemu_st16(tmp, addr, IS_USER(s));
4038 break;
4039 case 2:
4040 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
4041 break;
4042 }
4043 tcg_temp_free_i32(tmp);
4044 }
4045 rd += stride;
4046 tcg_gen_addi_i32(addr, addr, 1 << size);
4047 }
4048 tcg_temp_free_i32(addr);
4049 stride = nregs * (1 << size);
4050 }
4051 }
4052 if (rm != 15) {
4053 TCGv_i32 base;
4054
4055 base = load_reg(s, rn);
4056 if (rm == 13) {
4057 tcg_gen_addi_i32(base, base, stride);
4058 } else {
4059 TCGv_i32 index;
4060 index = load_reg(s, rm);
4061 tcg_gen_add_i32(base, base, index);
4062 tcg_temp_free_i32(index);
4063 }
4064 store_reg(s, rn, base);
4065 }
4066 return 0;
4067 }
4068
4069 /* Bitwise select. dest = c ? t : f. Clobbers T and F. */
4070 static void gen_neon_bsl(TCGv_i32 dest, TCGv_i32 t, TCGv_i32 f, TCGv_i32 c)
4071 {
4072 tcg_gen_and_i32(t, t, c);
4073 tcg_gen_andc_i32(f, f, c);
4074 tcg_gen_or_i32(dest, t, f);
4075 }
4076
4077 static inline void gen_neon_narrow(int size, TCGv_i32 dest, TCGv_i64 src)
4078 {
4079 switch (size) {
4080 case 0: gen_helper_neon_narrow_u8(dest, src); break;
4081 case 1: gen_helper_neon_narrow_u16(dest, src); break;
4082 case 2: tcg_gen_trunc_i64_i32(dest, src); break;
4083 default: abort();
4084 }
4085 }
4086
4087 static inline void gen_neon_narrow_sats(int size, TCGv_i32 dest, TCGv_i64 src)
4088 {
4089 switch (size) {
4090 case 0: gen_helper_neon_narrow_sat_s8(dest, cpu_env, src); break;
4091 case 1: gen_helper_neon_narrow_sat_s16(dest, cpu_env, src); break;
4092 case 2: gen_helper_neon_narrow_sat_s32(dest, cpu_env, src); break;
4093 default: abort();
4094 }
4095 }
4096
4097 static inline void gen_neon_narrow_satu(int size, TCGv_i32 dest, TCGv_i64 src)
4098 {
4099 switch (size) {
4100 case 0: gen_helper_neon_narrow_sat_u8(dest, cpu_env, src); break;
4101 case 1: gen_helper_neon_narrow_sat_u16(dest, cpu_env, src); break;
4102 case 2: gen_helper_neon_narrow_sat_u32(dest, cpu_env, src); break;
4103 default: abort();
4104 }
4105 }
4106
4107 static inline void gen_neon_unarrow_sats(int size, TCGv_i32 dest, TCGv_i64 src)
4108 {
4109 switch (size) {
4110 case 0: gen_helper_neon_unarrow_sat8(dest, cpu_env, src); break;
4111 case 1: gen_helper_neon_unarrow_sat16(dest, cpu_env, src); break;
4112 case 2: gen_helper_neon_unarrow_sat32(dest, cpu_env, src); break;
4113 default: abort();
4114 }
4115 }
4116
4117 static inline void gen_neon_shift_narrow(int size, TCGv_i32 var, TCGv_i32 shift,
4118 int q, int u)
4119 {
4120 if (q) {
4121 if (u) {
4122 switch (size) {
4123 case 1: gen_helper_neon_rshl_u16(var, var, shift); break;
4124 case 2: gen_helper_neon_rshl_u32(var, var, shift); break;
4125 default: abort();
4126 }
4127 } else {
4128 switch (size) {
4129 case 1: gen_helper_neon_rshl_s16(var, var, shift); break;
4130 case 2: gen_helper_neon_rshl_s32(var, var, shift); break;
4131 default: abort();
4132 }
4133 }
4134 } else {
4135 if (u) {
4136 switch (size) {
4137 case 1: gen_helper_neon_shl_u16(var, var, shift); break;
4138 case 2: gen_helper_neon_shl_u32(var, var, shift); break;
4139 default: abort();
4140 }
4141 } else {
4142 switch (size) {
4143 case 1: gen_helper_neon_shl_s16(var, var, shift); break;
4144 case 2: gen_helper_neon_shl_s32(var, var, shift); break;
4145 default: abort();
4146 }
4147 }
4148 }
4149 }
4150
4151 static inline void gen_neon_widen(TCGv_i64 dest, TCGv_i32 src, int size, int u)
4152 {
4153 if (u) {
4154 switch (size) {
4155 case 0: gen_helper_neon_widen_u8(dest, src); break;
4156 case 1: gen_helper_neon_widen_u16(dest, src); break;
4157 case 2: tcg_gen_extu_i32_i64(dest, src); break;
4158 default: abort();
4159 }
4160 } else {
4161 switch (size) {
4162 case 0: gen_helper_neon_widen_s8(dest, src); break;
4163 case 1: gen_helper_neon_widen_s16(dest, src); break;
4164 case 2: tcg_gen_ext_i32_i64(dest, src); break;
4165 default: abort();
4166 }
4167 }
4168 tcg_temp_free_i32(src);
4169 }
4170
4171 static inline void gen_neon_addl(int size)
4172 {
4173 switch (size) {
4174 case 0: gen_helper_neon_addl_u16(CPU_V001); break;
4175 case 1: gen_helper_neon_addl_u32(CPU_V001); break;
4176 case 2: tcg_gen_add_i64(CPU_V001); break;
4177 default: abort();
4178 }
4179 }
4180
4181 static inline void gen_neon_subl(int size)
4182 {
4183 switch (size) {
4184 case 0: gen_helper_neon_subl_u16(CPU_V001); break;
4185 case 1: gen_helper_neon_subl_u32(CPU_V001); break;
4186 case 2: tcg_gen_sub_i64(CPU_V001); break;
4187 default: abort();
4188 }
4189 }
4190
4191 static inline void gen_neon_negl(TCGv_i64 var, int size)
4192 {
4193 switch (size) {
4194 case 0: gen_helper_neon_negl_u16(var, var); break;
4195 case 1: gen_helper_neon_negl_u32(var, var); break;
4196 case 2:
4197 tcg_gen_neg_i64(var, var);
4198 break;
4199 default: abort();
4200 }
4201 }
4202
4203 static inline void gen_neon_addl_saturate(TCGv_i64 op0, TCGv_i64 op1, int size)
4204 {
4205 switch (size) {
4206 case 1: gen_helper_neon_addl_saturate_s32(op0, cpu_env, op0, op1); break;
4207 case 2: gen_helper_neon_addl_saturate_s64(op0, cpu_env, op0, op1); break;
4208 default: abort();
4209 }
4210 }
4211
4212 static inline void gen_neon_mull(TCGv_i64 dest, TCGv_i32 a, TCGv_i32 b,
4213 int size, int u)
4214 {
4215 TCGv_i64 tmp;
4216
4217 switch ((size << 1) | u) {
4218 case 0: gen_helper_neon_mull_s8(dest, a, b); break;
4219 case 1: gen_helper_neon_mull_u8(dest, a, b); break;
4220 case 2: gen_helper_neon_mull_s16(dest, a, b); break;
4221 case 3: gen_helper_neon_mull_u16(dest, a, b); break;
4222 case 4:
4223 tmp = gen_muls_i64_i32(a, b);
4224 tcg_gen_mov_i64(dest, tmp);
4225 tcg_temp_free_i64(tmp);
4226 break;
4227 case 5:
4228 tmp = gen_mulu_i64_i32(a, b);
4229 tcg_gen_mov_i64(dest, tmp);
4230 tcg_temp_free_i64(tmp);
4231 break;
4232 default: abort();
4233 }
4234
4235 /* gen_helper_neon_mull_[su]{8|16} do not free their parameters.
4236 Don't forget to clean them now. */
4237 if (size < 2) {
4238 tcg_temp_free_i32(a);
4239 tcg_temp_free_i32(b);
4240 }
4241 }
4242
4243 static void gen_neon_narrow_op(int op, int u, int size,
4244 TCGv_i32 dest, TCGv_i64 src)
4245 {
4246 if (op) {
4247 if (u) {
4248 gen_neon_unarrow_sats(size, dest, src);
4249 } else {
4250 gen_neon_narrow(size, dest, src);
4251 }
4252 } else {
4253 if (u) {
4254 gen_neon_narrow_satu(size, dest, src);
4255 } else {
4256 gen_neon_narrow_sats(size, dest, src);
4257 }
4258 }
4259 }
4260
4261 /* Symbolic constants for op fields for Neon 3-register same-length.
4262 * The values correspond to bits [11:8,4]; see the ARM ARM DDI0406B
4263 * table A7-9.
4264 */
4265 #define NEON_3R_VHADD 0
4266 #define NEON_3R_VQADD 1
4267 #define NEON_3R_VRHADD 2
4268 #define NEON_3R_LOGIC 3 /* VAND,VBIC,VORR,VMOV,VORN,VEOR,VBIF,VBIT,VBSL */
4269 #define NEON_3R_VHSUB 4
4270 #define NEON_3R_VQSUB 5
4271 #define NEON_3R_VCGT 6
4272 #define NEON_3R_VCGE 7
4273 #define NEON_3R_VSHL 8
4274 #define NEON_3R_VQSHL 9
4275 #define NEON_3R_VRSHL 10
4276 #define NEON_3R_VQRSHL 11
4277 #define NEON_3R_VMAX 12
4278 #define NEON_3R_VMIN 13
4279 #define NEON_3R_VABD 14
4280 #define NEON_3R_VABA 15
4281 #define NEON_3R_VADD_VSUB 16
4282 #define NEON_3R_VTST_VCEQ 17
4283 #define NEON_3R_VML 18 /* VMLA, VMLAL, VMLS, VMLSL */
4284 #define NEON_3R_VMUL 19
4285 #define NEON_3R_VPMAX 20
4286 #define NEON_3R_VPMIN 21
4287 #define NEON_3R_VQDMULH_VQRDMULH 22
4288 #define NEON_3R_VPADD 23
4289 #define NEON_3R_VFM 25 /* VFMA, VFMS : float fused multiply-add */
4290 #define NEON_3R_FLOAT_ARITH 26 /* float VADD, VSUB, VPADD, VABD */
4291 #define NEON_3R_FLOAT_MULTIPLY 27 /* float VMLA, VMLS, VMUL */
4292 #define NEON_3R_FLOAT_CMP 28 /* float VCEQ, VCGE, VCGT */
4293 #define NEON_3R_FLOAT_ACMP 29 /* float VACGE, VACGT, VACLE, VACLT */
4294 #define NEON_3R_FLOAT_MINMAX 30 /* float VMIN, VMAX */
4295 #define NEON_3R_VRECPS_VRSQRTS 31 /* float VRECPS, VRSQRTS */
4296
4297 static const uint8_t neon_3r_sizes[] = {
4298 [NEON_3R_VHADD] = 0x7,
4299 [NEON_3R_VQADD] = 0xf,
4300 [NEON_3R_VRHADD] = 0x7,
4301 [NEON_3R_LOGIC] = 0xf, /* size field encodes op type */
4302 [NEON_3R_VHSUB] = 0x7,
4303 [NEON_3R_VQSUB] = 0xf,
4304 [NEON_3R_VCGT] = 0x7,
4305 [NEON_3R_VCGE] = 0x7,
4306 [NEON_3R_VSHL] = 0xf,
4307 [NEON_3R_VQSHL] = 0xf,
4308 [NEON_3R_VRSHL] = 0xf,
4309 [NEON_3R_VQRSHL] = 0xf,
4310 [NEON_3R_VMAX] = 0x7,
4311 [NEON_3R_VMIN] = 0x7,
4312 [NEON_3R_VABD] = 0x7,
4313 [NEON_3R_VABA] = 0x7,
4314 [NEON_3R_VADD_VSUB] = 0xf,
4315 [NEON_3R_VTST_VCEQ] = 0x7,
4316 [NEON_3R_VML] = 0x7,
4317 [NEON_3R_VMUL] = 0x7,
4318 [NEON_3R_VPMAX] = 0x7,
4319 [NEON_3R_VPMIN] = 0x7,
4320 [NEON_3R_VQDMULH_VQRDMULH] = 0x6,
4321 [NEON_3R_VPADD] = 0x7,
4322 [NEON_3R_VFM] = 0x5, /* size bit 1 encodes op */
4323 [NEON_3R_FLOAT_ARITH] = 0x5, /* size bit 1 encodes op */
4324 [NEON_3R_FLOAT_MULTIPLY] = 0x5, /* size bit 1 encodes op */
4325 [NEON_3R_FLOAT_CMP] = 0x5, /* size bit 1 encodes op */
4326 [NEON_3R_FLOAT_ACMP] = 0x5, /* size bit 1 encodes op */
4327 [NEON_3R_FLOAT_MINMAX] = 0x5, /* size bit 1 encodes op */
4328 [NEON_3R_VRECPS_VRSQRTS] = 0x5, /* size bit 1 encodes op */
4329 };
4330
4331 /* Symbolic constants for op fields for Neon 2-register miscellaneous.
4332 * The values correspond to bits [17:16,10:7]; see the ARM ARM DDI0406B
4333 * table A7-13.
4334 */
4335 #define NEON_2RM_VREV64 0
4336 #define NEON_2RM_VREV32 1
4337 #define NEON_2RM_VREV16 2
4338 #define NEON_2RM_VPADDL 4
4339 #define NEON_2RM_VPADDL_U 5
4340 #define NEON_2RM_VCLS 8
4341 #define NEON_2RM_VCLZ 9
4342 #define NEON_2RM_VCNT 10
4343 #define NEON_2RM_VMVN 11
4344 #define NEON_2RM_VPADAL 12
4345 #define NEON_2RM_VPADAL_U 13
4346 #define NEON_2RM_VQABS 14
4347 #define NEON_2RM_VQNEG 15
4348 #define NEON_2RM_VCGT0 16
4349 #define NEON_2RM_VCGE0 17
4350 #define NEON_2RM_VCEQ0 18
4351 #define NEON_2RM_VCLE0 19
4352 #define NEON_2RM_VCLT0 20
4353 #define NEON_2RM_VABS 22
4354 #define NEON_2RM_VNEG 23
4355 #define NEON_2RM_VCGT0_F 24
4356 #define NEON_2RM_VCGE0_F 25
4357 #define NEON_2RM_VCEQ0_F 26
4358 #define NEON_2RM_VCLE0_F 27
4359 #define NEON_2RM_VCLT0_F 28
4360 #define NEON_2RM_VABS_F 30
4361 #define NEON_2RM_VNEG_F 31
4362 #define NEON_2RM_VSWP 32
4363 #define NEON_2RM_VTRN 33
4364 #define NEON_2RM_VUZP 34
4365 #define NEON_2RM_VZIP 35
4366 #define NEON_2RM_VMOVN 36 /* Includes VQMOVN, VQMOVUN */
4367 #define NEON_2RM_VQMOVN 37 /* Includes VQMOVUN */
4368 #define NEON_2RM_VSHLL 38
4369 #define NEON_2RM_VCVT_F16_F32 44
4370 #define NEON_2RM_VCVT_F32_F16 46
4371 #define NEON_2RM_VRECPE 56
4372 #define NEON_2RM_VRSQRTE 57
4373 #define NEON_2RM_VRECPE_F 58
4374 #define NEON_2RM_VRSQRTE_F 59
4375 #define NEON_2RM_VCVT_FS 60
4376 #define NEON_2RM_VCVT_FU 61
4377 #define NEON_2RM_VCVT_SF 62
4378 #define NEON_2RM_VCVT_UF 63
4379
4380 static int neon_2rm_is_float_op(int op)
4381 {
4382 /* Return true if this neon 2reg-misc op is float-to-float */
4383 return (op == NEON_2RM_VABS_F || op == NEON_2RM_VNEG_F ||
4384 op >= NEON_2RM_VRECPE_F);
4385 }
4386
4387 /* Each entry in this array has bit n set if the insn allows
4388 * size value n (otherwise it will UNDEF). Since unallocated
4389 * op values will have no bits set they always UNDEF.
4390 */
4391 static const uint8_t neon_2rm_sizes[] = {
4392 [NEON_2RM_VREV64] = 0x7,
4393 [NEON_2RM_VREV32] = 0x3,
4394 [NEON_2RM_VREV16] = 0x1,
4395 [NEON_2RM_VPADDL] = 0x7,
4396 [NEON_2RM_VPADDL_U] = 0x7,
4397 [NEON_2RM_VCLS] = 0x7,
4398 [NEON_2RM_VCLZ] = 0x7,
4399 [NEON_2RM_VCNT] = 0x1,
4400 [NEON_2RM_VMVN] = 0x1,
4401 [NEON_2RM_VPADAL] = 0x7,
4402 [NEON_2RM_VPADAL_U] = 0x7,
4403 [NEON_2RM_VQABS] = 0x7,
4404 [NEON_2RM_VQNEG] = 0x7,
4405 [NEON_2RM_VCGT0] = 0x7,
4406 [NEON_2RM_VCGE0] = 0x7,
4407 [NEON_2RM_VCEQ0] = 0x7,
4408 [NEON_2RM_VCLE0] = 0x7,
4409 [NEON_2RM_VCLT0] = 0x7,
4410 [NEON_2RM_VABS] = 0x7,
4411 [NEON_2RM_VNEG] = 0x7,
4412 [NEON_2RM_VCGT0_F] = 0x4,
4413 [NEON_2RM_VCGE0_F] = 0x4,
4414 [NEON_2RM_VCEQ0_F] = 0x4,
4415 [NEON_2RM_VCLE0_F] = 0x4,
4416 [NEON_2RM_VCLT0_F] = 0x4,
4417 [NEON_2RM_VABS_F] = 0x4,
4418 [NEON_2RM_VNEG_F] = 0x4,
4419 [NEON_2RM_VSWP] = 0x1,
4420 [NEON_2RM_VTRN] = 0x7,
4421 [NEON_2RM_VUZP] = 0x7,
4422 [NEON_2RM_VZIP] = 0x7,
4423 [NEON_2RM_VMOVN] = 0x7,
4424 [NEON_2RM_VQMOVN] = 0x7,
4425 [NEON_2RM_VSHLL] = 0x7,
4426 [NEON_2RM_VCVT_F16_F32] = 0x2,
4427 [NEON_2RM_VCVT_F32_F16] = 0x2,
4428 [NEON_2RM_VRECPE] = 0x4,
4429 [NEON_2RM_VRSQRTE] = 0x4,
4430 [NEON_2RM_VRECPE_F] = 0x4,
4431 [NEON_2RM_VRSQRTE_F] = 0x4,
4432 [NEON_2RM_VCVT_FS] = 0x4,
4433 [NEON_2RM_VCVT_FU] = 0x4,
4434 [NEON_2RM_VCVT_SF] = 0x4,
4435 [NEON_2RM_VCVT_UF] = 0x4,
4436 };
4437
4438 /* Translate a NEON data processing instruction. Return nonzero if the
4439 instruction is invalid.
4440 We process data in a mixture of 32-bit and 64-bit chunks.
4441 Mostly we use 32-bit chunks so we can use normal scalar instructions. */
4442
4443 static int disas_neon_data_insn(CPUARMState * env, DisasContext *s, uint32_t insn)
4444 {
4445 int op;
4446 int q;
4447 int rd, rn, rm;
4448 int size;
4449 int shift;
4450 int pass;
4451 int count;
4452 int pairwise;
4453 int u;
4454 uint32_t imm, mask;
4455 TCGv_i32 tmp, tmp2, tmp3, tmp4, tmp5;
4456 TCGv_i64 tmp64;
4457
4458 if (!s->vfp_enabled)
4459 return 1;
4460 q = (insn & (1 << 6)) != 0;
4461 u = (insn >> 24) & 1;
4462 VFP_DREG_D(rd, insn);
4463 VFP_DREG_N(rn, insn);
4464 VFP_DREG_M(rm, insn);
4465 size = (insn >> 20) & 3;
4466 if ((insn & (1 << 23)) == 0) {
4467 /* Three register same length. */
4468 op = ((insn >> 7) & 0x1e) | ((insn >> 4) & 1);
4469 /* Catch invalid op and bad size combinations: UNDEF */
4470 if ((neon_3r_sizes[op] & (1 << size)) == 0) {
4471 return 1;
4472 }
4473 /* All insns of this form UNDEF for either this condition or the
4474 * superset of cases "Q==1"; we catch the latter later.
4475 */
4476 if (q && ((rd | rn | rm) & 1)) {
4477 return 1;
4478 }
4479 if (size == 3 && op != NEON_3R_LOGIC) {
4480 /* 64-bit element instructions. */
4481 for (pass = 0; pass < (q ? 2 : 1); pass++) {
4482 neon_load_reg64(cpu_V0, rn + pass);
4483 neon_load_reg64(cpu_V1, rm + pass);
4484 switch (op) {
4485 case NEON_3R_VQADD:
4486 if (u) {
4487 gen_helper_neon_qadd_u64(cpu_V0, cpu_env,
4488 cpu_V0, cpu_V1);
4489 } else {
4490 gen_helper_neon_qadd_s64(cpu_V0, cpu_env,
4491 cpu_V0, cpu_V1);
4492 }
4493 break;
4494 case NEON_3R_VQSUB:
4495 if (u) {
4496 gen_helper_neon_qsub_u64(cpu_V0, cpu_env,
4497 cpu_V0, cpu_V1);
4498 } else {
4499 gen_helper_neon_qsub_s64(cpu_V0, cpu_env,
4500 cpu_V0, cpu_V1);
4501 }
4502 break;
4503 case NEON_3R_VSHL:
4504 if (u) {
4505 gen_helper_neon_shl_u64(cpu_V0, cpu_V1, cpu_V0);
4506 } else {
4507 gen_helper_neon_shl_s64(cpu_V0, cpu_V1, cpu_V0);
4508 }
4509 break;
4510 case NEON_3R_VQSHL:
4511 if (u) {
4512 gen_helper_neon_qshl_u64(cpu_V0, cpu_env,
4513 cpu_V1, cpu_V0);
4514 } else {
4515 gen_helper_neon_qshl_s64(cpu_V0, cpu_env,
4516 cpu_V1, cpu_V0);
4517 }
4518 break;
4519 case NEON_3R_VRSHL:
4520 if (u) {
4521 gen_helper_neon_rshl_u64(cpu_V0, cpu_V1, cpu_V0);
4522 } else {
4523 gen_helper_neon_rshl_s64(cpu_V0, cpu_V1, cpu_V0);
4524 }
4525 break;
4526 case NEON_3R_VQRSHL:
4527 if (u) {
4528 gen_helper_neon_qrshl_u64(cpu_V0, cpu_env,
4529 cpu_V1, cpu_V0);
4530 } else {
4531 gen_helper_neon_qrshl_s64(cpu_V0, cpu_env,
4532 cpu_V1, cpu_V0);
4533 }
4534 break;
4535 case NEON_3R_VADD_VSUB:
4536 if (u) {
4537 tcg_gen_sub_i64(CPU_V001);
4538 } else {
4539 tcg_gen_add_i64(CPU_V001);
4540 }
4541 break;
4542 default:
4543 abort();
4544 }
4545 neon_store_reg64(cpu_V0, rd + pass);
4546 }
4547 return 0;
4548 }
4549 pairwise = 0;
4550 switch (op) {
4551 case NEON_3R_VSHL:
4552 case NEON_3R_VQSHL:
4553 case NEON_3R_VRSHL:
4554 case NEON_3R_VQRSHL:
4555 {
4556 int rtmp;
4557 /* Shift instruction operands are reversed. */
4558 rtmp = rn;
4559 rn = rm;
4560 rm = rtmp;
4561 }
4562 break;
4563 case NEON_3R_VPADD:
4564 if (u) {
4565 return 1;
4566 }
4567 /* Fall through */
4568 case NEON_3R_VPMAX:
4569 case NEON_3R_VPMIN:
4570 pairwise = 1;
4571 break;
4572 case NEON_3R_FLOAT_ARITH:
4573 pairwise = (u && size < 2); /* if VPADD (float) */
4574 break;
4575 case NEON_3R_FLOAT_MINMAX:
4576 pairwise = u; /* if VPMIN/VPMAX (float) */
4577 break;
4578 case NEON_3R_FLOAT_CMP:
4579 if (!u && size) {
4580 /* no encoding for U=0 C=1x */
4581 return 1;
4582 }
4583 break;
4584 case NEON_3R_FLOAT_ACMP:
4585 if (!u) {
4586 return 1;
4587 }
4588 break;
4589 case NEON_3R_VRECPS_VRSQRTS:
4590 if (u) {
4591 return 1;
4592 }
4593 break;
4594 case NEON_3R_VMUL:
4595 if (u && (size != 0)) {
4596 /* UNDEF on invalid size for polynomial subcase */
4597 return 1;
4598 }
4599 break;
4600 case NEON_3R_VFM:
4601 if (!arm_feature(env, ARM_FEATURE_VFP4) || u) {
4602 return 1;
4603 }
4604 break;
4605 default:
4606 break;
4607 }
4608
4609 if (pairwise && q) {
4610 /* All the pairwise insns UNDEF if Q is set */
4611 return 1;
4612 }
4613
4614 for (pass = 0; pass < (q ? 4 : 2); pass++) {
4615
4616 if (pairwise) {
4617 /* Pairwise. */
4618 if (pass < 1) {
4619 tmp = neon_load_reg(rn, 0);
4620 tmp2 = neon_load_reg(rn, 1);
4621 } else {
4622 tmp = neon_load_reg(rm, 0);
4623 tmp2 = neon_load_reg(rm, 1);
4624 }
4625 } else {
4626 /* Elementwise. */
4627 tmp = neon_load_reg(rn, pass);
4628 tmp2 = neon_load_reg(rm, pass);
4629 }
4630 switch (op) {
4631 case NEON_3R_VHADD:
4632 GEN_NEON_INTEGER_OP(hadd);
4633 break;
4634 case NEON_3R_VQADD:
4635 GEN_NEON_INTEGER_OP_ENV(qadd);
4636 break;
4637 case NEON_3R_VRHADD:
4638 GEN_NEON_INTEGER_OP(rhadd);
4639 break;
4640 case NEON_3R_LOGIC: /* Logic ops. */
4641 switch ((u << 2) | size) {
4642 case 0: /* VAND */
4643 tcg_gen_and_i32(tmp, tmp, tmp2);
4644 break;
4645 case 1: /* BIC */
4646 tcg_gen_andc_i32(tmp, tmp, tmp2);
4647 break;
4648 case 2: /* VORR */
4649 tcg_gen_or_i32(tmp, tmp, tmp2);
4650 break;
4651 case 3: /* VORN */
4652 tcg_gen_orc_i32(tmp, tmp, tmp2);
4653 break;
4654 case 4: /* VEOR */
4655 tcg_gen_xor_i32(tmp, tmp, tmp2);
4656 break;
4657 case 5: /* VBSL */
4658 tmp3 = neon_load_reg(rd, pass);
4659 gen_neon_bsl(tmp, tmp, tmp2, tmp3);
4660 tcg_temp_free_i32(tmp3);
4661 break;
4662 case 6: /* VBIT */
4663 tmp3 = neon_load_reg(rd, pass);
4664 gen_neon_bsl(tmp, tmp, tmp3, tmp2);
4665 tcg_temp_free_i32(tmp3);
4666 break;
4667 case 7: /* VBIF */
4668 tmp3 = neon_load_reg(rd, pass);
4669 gen_neon_bsl(tmp, tmp3, tmp, tmp2);
4670 tcg_temp_free_i32(tmp3);
4671 break;
4672 }
4673 break;
4674 case NEON_3R_VHSUB:
4675 GEN_NEON_INTEGER_OP(hsub);
4676 break;
4677 case NEON_3R_VQSUB:
4678 GEN_NEON_INTEGER_OP_ENV(qsub);
4679 break;
4680 case NEON_3R_VCGT:
4681 GEN_NEON_INTEGER_OP(cgt);
4682 break;
4683 case NEON_3R_VCGE:
4684 GEN_NEON_INTEGER_OP(cge);
4685 break;
4686 case NEON_3R_VSHL:
4687 GEN_NEON_INTEGER_OP(shl);
4688 break;
4689 case NEON_3R_VQSHL:
4690 GEN_NEON_INTEGER_OP_ENV(qshl);
4691 break;
4692 case NEON_3R_VRSHL:
4693 GEN_NEON_INTEGER_OP(rshl);
4694 break;
4695 case NEON_3R_VQRSHL:
4696 GEN_NEON_INTEGER_OP_ENV(qrshl);
4697 break;
4698 case NEON_3R_VMAX:
4699 GEN_NEON_INTEGER_OP(max);
4700 break;
4701 case NEON_3R_VMIN:
4702 GEN_NEON_INTEGER_OP(min);
4703 break;
4704 case NEON_3R_VABD:
4705 GEN_NEON_INTEGER_OP(abd);
4706 break;
4707 case NEON_3R_VABA:
4708 GEN_NEON_INTEGER_OP(abd);
4709 tcg_temp_free_i32(tmp2);
4710 tmp2 = neon_load_reg(rd, pass);
4711 gen_neon_add(size, tmp, tmp2);
4712 break;
4713 case NEON_3R_VADD_VSUB:
4714 if (!u) { /* VADD */
4715 gen_neon_add(size, tmp, tmp2);
4716 } else { /* VSUB */
4717 switch (size) {
4718 case 0: gen_helper_neon_sub_u8(tmp, tmp, tmp2); break;
4719 case 1: gen_helper_neon_sub_u16(tmp, tmp, tmp2); break;
4720 case 2: tcg_gen_sub_i32(tmp, tmp, tmp2); break;
4721 default: abort();
4722 }
4723 }
4724 break;
4725 case NEON_3R_VTST_VCEQ:
4726 if (!u) { /* VTST */
4727 switch (size) {
4728 case 0: gen_helper_neon_tst_u8(tmp, tmp, tmp2); break;
4729 case 1: gen_helper_neon_tst_u16(tmp, tmp, tmp2); break;
4730 case 2: gen_helper_neon_tst_u32(tmp, tmp, tmp2); break;
4731 default: abort();
4732 }
4733 } else { /* VCEQ */
4734 switch (size) {
4735 case 0: gen_helper_neon_ceq_u8(tmp, tmp, tmp2); break;
4736 case 1: gen_helper_neon_ceq_u16(tmp, tmp, tmp2); break;
4737 case 2: gen_helper_neon_ceq_u32(tmp, tmp, tmp2); break;
4738 default: abort();
4739 }
4740 }
4741 break;
4742 case NEON_3R_VML: /* VMLA, VMLAL, VMLS,VMLSL */
4743 switch (size) {
4744 case 0: gen_helper_neon_mul_u8(tmp, tmp, tmp2); break;
4745 case 1: gen_helper_neon_mul_u16(tmp, tmp, tmp2); break;
4746 case 2: tcg_gen_mul_i32(tmp, tmp, tmp2); break;
4747 default: abort();
4748 }
4749 tcg_temp_free_i32(tmp2);
4750 tmp2 = neon_load_reg(rd, pass);
4751 if (u) { /* VMLS */
4752 gen_neon_rsb(size, tmp, tmp2);
4753 } else { /* VMLA */
4754 gen_neon_add(size, tmp, tmp2);
4755 }
4756 break;
4757 case NEON_3R_VMUL:
4758 if (u) { /* polynomial */
4759 gen_helper_neon_mul_p8(tmp, tmp, tmp2);
4760 } else { /* Integer */
4761 switch (size) {
4762 case 0: gen_helper_neon_mul_u8(tmp, tmp, tmp2); break;
4763 case 1: gen_helper_neon_mul_u16(tmp, tmp, tmp2); break;
4764 case 2: tcg_gen_mul_i32(tmp, tmp, tmp2); break;
4765 default: abort();
4766 }
4767 }
4768 break;
4769 case NEON_3R_VPMAX:
4770 GEN_NEON_INTEGER_OP(pmax);
4771 break;
4772 case NEON_3R_VPMIN:
4773 GEN_NEON_INTEGER_OP(pmin);
4774 break;
4775 case NEON_3R_VQDMULH_VQRDMULH: /* Multiply high. */
4776 if (!u) { /* VQDMULH */
4777 switch (size) {
4778 case 1:
4779 gen_helper_neon_qdmulh_s16(tmp, cpu_env, tmp, tmp2);
4780 break;
4781 case 2:
4782 gen_helper_neon_qdmulh_s32(tmp, cpu_env, tmp, tmp2);
4783 break;
4784 default: abort();
4785 }
4786 } else { /* VQRDMULH */
4787 switch (size) {
4788 case 1:
4789 gen_helper_neon_qrdmulh_s16(tmp, cpu_env, tmp, tmp2);
4790 break;
4791 case 2:
4792 gen_helper_neon_qrdmulh_s32(tmp, cpu_env, tmp, tmp2);
4793 break;
4794 default: abort();
4795 }
4796 }
4797 break;
4798 case NEON_3R_VPADD:
4799 switch (size) {
4800 case 0: gen_helper_neon_padd_u8(tmp, tmp, tmp2); break;
4801 case 1: gen_helper_neon_padd_u16(tmp, tmp, tmp2); break;
4802 case 2: tcg_gen_add_i32(tmp, tmp, tmp2); break;
4803 default: abort();
4804 }
4805 break;
4806 case NEON_3R_FLOAT_ARITH: /* Floating point arithmetic. */
4807 {
4808 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
4809 switch ((u << 2) | size) {
4810 case 0: /* VADD */
4811 case 4: /* VPADD */
4812 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
4813 break;
4814 case 2: /* VSUB */
4815 gen_helper_vfp_subs(tmp, tmp, tmp2, fpstatus);
4816 break;
4817 case 6: /* VABD */
4818 gen_helper_neon_abd_f32(tmp, tmp, tmp2, fpstatus);
4819 break;
4820 default:
4821 abort();
4822 }
4823 tcg_temp_free_ptr(fpstatus);
4824 break;
4825 }
4826 case NEON_3R_FLOAT_MULTIPLY:
4827 {
4828 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
4829 gen_helper_vfp_muls(tmp, tmp, tmp2, fpstatus);
4830 if (!u) {
4831 tcg_temp_free_i32(tmp2);
4832 tmp2 = neon_load_reg(rd, pass);
4833 if (size == 0) {
4834 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
4835 } else {
4836 gen_helper_vfp_subs(tmp, tmp2, tmp, fpstatus);
4837 }
4838 }
4839 tcg_temp_free_ptr(fpstatus);
4840 break;
4841 }
4842 case NEON_3R_FLOAT_CMP:
4843 {
4844 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
4845 if (!u) {
4846 gen_helper_neon_ceq_f32(tmp, tmp, tmp2, fpstatus);
4847 } else {
4848 if (size == 0) {
4849 gen_helper_neon_cge_f32(tmp, tmp, tmp2, fpstatus);
4850 } else {
4851 gen_helper_neon_cgt_f32(tmp, tmp, tmp2, fpstatus);
4852 }
4853 }
4854 tcg_temp_free_ptr(fpstatus);
4855 break;
4856 }
4857 case NEON_3R_FLOAT_ACMP:
4858 {
4859 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
4860 if (size == 0) {
4861 gen_helper_neon_acge_f32(tmp, tmp, tmp2, fpstatus);
4862 } else {
4863 gen_helper_neon_acgt_f32(tmp, tmp, tmp2, fpstatus);
4864 }
4865 tcg_temp_free_ptr(fpstatus);
4866 break;
4867 }
4868 case NEON_3R_FLOAT_MINMAX:
4869 {
4870 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
4871 if (size == 0) {
4872 gen_helper_neon_max_f32(tmp, tmp, tmp2, fpstatus);
4873 } else {
4874 gen_helper_neon_min_f32(tmp, tmp, tmp2, fpstatus);
4875 }
4876 tcg_temp_free_ptr(fpstatus);
4877 break;
4878 }
4879 case NEON_3R_VRECPS_VRSQRTS:
4880 if (size == 0)
4881 gen_helper_recps_f32(tmp, tmp, tmp2, cpu_env);
4882 else
4883 gen_helper_rsqrts_f32(tmp, tmp, tmp2, cpu_env);
4884 break;
4885 case NEON_3R_VFM:
4886 {
4887 /* VFMA, VFMS: fused multiply-add */
4888 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
4889 TCGv_i32 tmp3 = neon_load_reg(rd, pass);
4890 if (size) {
4891 /* VFMS */
4892 gen_helper_vfp_negs(tmp, tmp);
4893 }
4894 gen_helper_vfp_muladds(tmp, tmp, tmp2, tmp3, fpstatus);
4895 tcg_temp_free_i32(tmp3);
4896 tcg_temp_free_ptr(fpstatus);
4897 break;
4898 }
4899 default:
4900 abort();
4901 }
4902 tcg_temp_free_i32(tmp2);
4903
4904 /* Save the result. For elementwise operations we can put it
4905 straight into the destination register. For pairwise operations
4906 we have to be careful to avoid clobbering the source operands. */
4907 if (pairwise && rd == rm) {
4908 neon_store_scratch(pass, tmp);
4909 } else {
4910 neon_store_reg(rd, pass, tmp);
4911 }
4912
4913 } /* for pass */
4914 if (pairwise && rd == rm) {
4915 for (pass = 0; pass < (q ? 4 : 2); pass++) {
4916 tmp = neon_load_scratch(pass);
4917 neon_store_reg(rd, pass, tmp);
4918 }
4919 }
4920 /* End of 3 register same size operations. */
4921 } else if (insn & (1 << 4)) {
4922 if ((insn & 0x00380080) != 0) {
4923 /* Two registers and shift. */
4924 op = (insn >> 8) & 0xf;
4925 if (insn & (1 << 7)) {
4926 /* 64-bit shift. */
4927 if (op > 7) {
4928 return 1;
4929 }
4930 size = 3;
4931 } else {
4932 size = 2;
4933 while ((insn & (1 << (size + 19))) == 0)
4934 size--;
4935 }
4936 shift = (insn >> 16) & ((1 << (3 + size)) - 1);
4937 /* To avoid excessive duplication of ops we implement shift
4938 by immediate using the variable shift operations. */
4939 if (op < 8) {
4940 /* Shift by immediate:
4941 VSHR, VSRA, VRSHR, VRSRA, VSRI, VSHL, VQSHL, VQSHLU. */
4942 if (q && ((rd | rm) & 1)) {
4943 return 1;
4944 }
4945 if (!u && (op == 4 || op == 6)) {
4946 return 1;
4947 }
4948 /* Right shifts are encoded as N - shift, where N is the
4949 element size in bits. */
4950 if (op <= 4)
4951 shift = shift - (1 << (size + 3));
4952 if (size == 3) {
4953 count = q + 1;
4954 } else {
4955 count = q ? 4: 2;
4956 }
4957 switch (size) {
4958 case 0:
4959 imm = (uint8_t) shift;
4960 imm |= imm << 8;
4961 imm |= imm << 16;
4962 break;
4963 case 1:
4964 imm = (uint16_t) shift;
4965 imm |= imm << 16;
4966 break;
4967 case 2:
4968 case 3:
4969 imm = shift;
4970 break;
4971 default:
4972 abort();
4973 }
4974
4975 for (pass = 0; pass < count; pass++) {
4976 if (size == 3) {
4977 neon_load_reg64(cpu_V0, rm + pass);
4978 tcg_gen_movi_i64(cpu_V1, imm);
4979 switch (op) {
4980 case 0: /* VSHR */
4981 case 1: /* VSRA */
4982 if (u)
4983 gen_helper_neon_shl_u64(cpu_V0, cpu_V0, cpu_V1);
4984 else
4985 gen_helper_neon_shl_s64(cpu_V0, cpu_V0, cpu_V1);
4986 break;
4987 case 2: /* VRSHR */
4988 case 3: /* VRSRA */
4989 if (u)
4990 gen_helper_neon_rshl_u64(cpu_V0, cpu_V0, cpu_V1);
4991 else
4992 gen_helper_neon_rshl_s64(cpu_V0, cpu_V0, cpu_V1);
4993 break;
4994 case 4: /* VSRI */
4995 case 5: /* VSHL, VSLI */
4996 gen_helper_neon_shl_u64(cpu_V0, cpu_V0, cpu_V1);
4997 break;
4998 case 6: /* VQSHLU */
4999 gen_helper_neon_qshlu_s64(cpu_V0, cpu_env,
5000 cpu_V0, cpu_V1);
5001 break;
5002 case 7: /* VQSHL */
5003 if (u) {
5004 gen_helper_neon_qshl_u64(cpu_V0, cpu_env,
5005 cpu_V0, cpu_V1);
5006 } else {
5007 gen_helper_neon_qshl_s64(cpu_V0, cpu_env,
5008 cpu_V0, cpu_V1);
5009 }
5010 break;
5011 }
5012 if (op == 1 || op == 3) {
5013 /* Accumulate. */
5014 neon_load_reg64(cpu_V1, rd + pass);
5015 tcg_gen_add_i64(cpu_V0, cpu_V0, cpu_V1);
5016 } else if (op == 4 || (op == 5 && u)) {
5017 /* Insert */
5018 neon_load_reg64(cpu_V1, rd + pass);
5019 uint64_t mask;
5020 if (shift < -63 || shift > 63) {
5021 mask = 0;
5022 } else {
5023 if (op == 4) {
5024 mask = 0xffffffffffffffffull >> -shift;
5025 } else {
5026 mask = 0xffffffffffffffffull << shift;
5027 }
5028 }
5029 tcg_gen_andi_i64(cpu_V1, cpu_V1, ~mask);
5030 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
5031 }
5032 neon_store_reg64(cpu_V0, rd + pass);
5033 } else { /* size < 3 */
5034 /* Operands in T0 and T1. */
5035 tmp = neon_load_reg(rm, pass);
5036 tmp2 = tcg_temp_new_i32();
5037 tcg_gen_movi_i32(tmp2, imm);
5038 switch (op) {
5039 case 0: /* VSHR */
5040 case 1: /* VSRA */
5041 GEN_NEON_INTEGER_OP(shl);
5042 break;
5043 case 2: /* VRSHR */
5044 case 3: /* VRSRA */
5045 GEN_NEON_INTEGER_OP(rshl);
5046 break;
5047 case 4: /* VSRI */
5048 case 5: /* VSHL, VSLI */
5049 switch (size) {
5050 case 0: gen_helper_neon_shl_u8(tmp, tmp, tmp2); break;
5051 case 1: gen_helper_neon_shl_u16(tmp, tmp, tmp2); break;
5052 case 2: gen_helper_neon_shl_u32(tmp, tmp, tmp2); break;
5053 default: abort();
5054 }
5055 break;
5056 case 6: /* VQSHLU */
5057 switch (size) {
5058 case 0:
5059 gen_helper_neon_qshlu_s8(tmp, cpu_env,
5060 tmp, tmp2);
5061 break;
5062 case 1:
5063 gen_helper_neon_qshlu_s16(tmp, cpu_env,
5064 tmp, tmp2);
5065 break;
5066 case 2:
5067 gen_helper_neon_qshlu_s32(tmp, cpu_env,
5068 tmp, tmp2);
5069 break;
5070 default:
5071 abort();
5072 }
5073 break;
5074 case 7: /* VQSHL */
5075 GEN_NEON_INTEGER_OP_ENV(qshl);
5076 break;
5077 }
5078 tcg_temp_free_i32(tmp2);
5079
5080 if (op == 1 || op == 3) {
5081 /* Accumulate. */
5082 tmp2 = neon_load_reg(rd, pass);
5083 gen_neon_add(size, tmp, tmp2);
5084 tcg_temp_free_i32(tmp2);
5085 } else if (op == 4 || (op == 5 && u)) {
5086 /* Insert */
5087 switch (size) {
5088 case 0:
5089 if (op == 4)
5090 mask = 0xff >> -shift;
5091 else
5092 mask = (uint8_t)(0xff << shift);
5093 mask |= mask << 8;
5094 mask |= mask << 16;
5095 break;
5096 case 1:
5097 if (op == 4)
5098 mask = 0xffff >> -shift;
5099 else
5100 mask = (uint16_t)(0xffff << shift);
5101 mask |= mask << 16;
5102 break;
5103 case 2:
5104 if (shift < -31 || shift > 31) {
5105 mask = 0;
5106 } else {
5107 if (op == 4)
5108 mask = 0xffffffffu >> -shift;
5109 else
5110 mask = 0xffffffffu << shift;
5111 }
5112 break;
5113 default:
5114 abort();
5115 }
5116 tmp2 = neon_load_reg(rd, pass);
5117 tcg_gen_andi_i32(tmp, tmp, mask);
5118 tcg_gen_andi_i32(tmp2, tmp2, ~mask);
5119 tcg_gen_or_i32(tmp, tmp, tmp2);
5120 tcg_temp_free_i32(tmp2);
5121 }
5122 neon_store_reg(rd, pass, tmp);
5123 }
5124 } /* for pass */
5125 } else if (op < 10) {
5126 /* Shift by immediate and narrow:
5127 VSHRN, VRSHRN, VQSHRN, VQRSHRN. */
5128 int input_unsigned = (op == 8) ? !u : u;
5129 if (rm & 1) {
5130 return 1;
5131 }
5132 shift = shift - (1 << (size + 3));
5133 size++;
5134 if (size == 3) {
5135 tmp64 = tcg_const_i64(shift);
5136 neon_load_reg64(cpu_V0, rm);
5137 neon_load_reg64(cpu_V1, rm + 1);
5138 for (pass = 0; pass < 2; pass++) {
5139 TCGv_i64 in;
5140 if (pass == 0) {
5141 in = cpu_V0;
5142 } else {
5143 in = cpu_V1;
5144 }
5145 if (q) {
5146 if (input_unsigned) {
5147 gen_helper_neon_rshl_u64(cpu_V0, in, tmp64);
5148 } else {
5149 gen_helper_neon_rshl_s64(cpu_V0, in, tmp64);
5150 }
5151 } else {
5152 if (input_unsigned) {
5153 gen_helper_neon_shl_u64(cpu_V0, in, tmp64);
5154 } else {
5155 gen_helper_neon_shl_s64(cpu_V0, in, tmp64);
5156 }
5157 }
5158 tmp = tcg_temp_new_i32();
5159 gen_neon_narrow_op(op == 8, u, size - 1, tmp, cpu_V0);
5160 neon_store_reg(rd, pass, tmp);
5161 } /* for pass */
5162 tcg_temp_free_i64(tmp64);
5163 } else {
5164 if (size == 1) {
5165 imm = (uint16_t)shift;
5166 imm |= imm << 16;
5167 } else {
5168 /* size == 2 */
5169 imm = (uint32_t)shift;
5170 }
5171 tmp2 = tcg_const_i32(imm);
5172 tmp4 = neon_load_reg(rm + 1, 0);
5173 tmp5 = neon_load_reg(rm + 1, 1);
5174 for (pass = 0; pass < 2; pass++) {
5175 if (pass == 0) {
5176 tmp = neon_load_reg(rm, 0);
5177 } else {
5178 tmp = tmp4;
5179 }
5180 gen_neon_shift_narrow(size, tmp, tmp2, q,
5181 input_unsigned);
5182 if (pass == 0) {
5183 tmp3 = neon_load_reg(rm, 1);
5184 } else {
5185 tmp3 = tmp5;
5186 }
5187 gen_neon_shift_narrow(size, tmp3, tmp2, q,
5188 input_unsigned);
5189 tcg_gen_concat_i32_i64(cpu_V0, tmp, tmp3);
5190 tcg_temp_free_i32(tmp);
5191 tcg_temp_free_i32(tmp3);
5192 tmp = tcg_temp_new_i32();
5193 gen_neon_narrow_op(op == 8, u, size - 1, tmp, cpu_V0);
5194 neon_store_reg(rd, pass, tmp);
5195 } /* for pass */
5196 tcg_temp_free_i32(tmp2);
5197 }
5198 } else if (op == 10) {
5199 /* VSHLL, VMOVL */
5200 if (q || (rd & 1)) {
5201 return 1;
5202 }
5203 tmp = neon_load_reg(rm, 0);
5204 tmp2 = neon_load_reg(rm, 1);
5205 for (pass = 0; pass < 2; pass++) {
5206 if (pass == 1)
5207 tmp = tmp2;
5208
5209 gen_neon_widen(cpu_V0, tmp, size, u);
5210
5211 if (shift != 0) {
5212 /* The shift is less than the width of the source
5213 type, so we can just shift the whole register. */
5214 tcg_gen_shli_i64(cpu_V0, cpu_V0, shift);
5215 /* Widen the result of shift: we need to clear
5216 * the potential overflow bits resulting from
5217 * left bits of the narrow input appearing as
5218 * right bits of left the neighbour narrow
5219 * input. */
5220 if (size < 2 || !u) {
5221 uint64_t imm64;
5222 if (size == 0) {
5223 imm = (0xffu >> (8 - shift));
5224 imm |= imm << 16;
5225 } else if (size == 1) {
5226 imm = 0xffff >> (16 - shift);
5227 } else {
5228 /* size == 2 */
5229 imm = 0xffffffff >> (32 - shift);
5230 }
5231 if (size < 2) {
5232 imm64 = imm | (((uint64_t)imm) << 32);
5233 } else {
5234 imm64 = imm;
5235 }
5236 tcg_gen_andi_i64(cpu_V0, cpu_V0, ~imm64);
5237 }
5238 }
5239 neon_store_reg64(cpu_V0, rd + pass);
5240 }
5241 } else if (op >= 14) {
5242 /* VCVT fixed-point. */
5243 if (!(insn & (1 << 21)) || (q && ((rd | rm) & 1))) {
5244 return 1;
5245 }
5246 /* We have already masked out the must-be-1 top bit of imm6,
5247 * hence this 32-shift where the ARM ARM has 64-imm6.
5248 */
5249 shift = 32 - shift;
5250 for (pass = 0; pass < (q ? 4 : 2); pass++) {
5251 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, pass));
5252 if (!(op & 1)) {
5253 if (u)
5254 gen_vfp_ulto(0, shift, 1);
5255 else
5256 gen_vfp_slto(0, shift, 1);
5257 } else {
5258 if (u)
5259 gen_vfp_toul(0, shift, 1);
5260 else
5261 gen_vfp_tosl(0, shift, 1);
5262 }
5263 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, pass));
5264 }
5265 } else {
5266 return 1;
5267 }
5268 } else { /* (insn & 0x00380080) == 0 */
5269 int invert;
5270 if (q && (rd & 1)) {
5271 return 1;
5272 }
5273
5274 op = (insn >> 8) & 0xf;
5275 /* One register and immediate. */
5276 imm = (u << 7) | ((insn >> 12) & 0x70) | (insn & 0xf);
5277 invert = (insn & (1 << 5)) != 0;
5278 /* Note that op = 2,3,4,5,6,7,10,11,12,13 imm=0 is UNPREDICTABLE.
5279 * We choose to not special-case this and will behave as if a
5280 * valid constant encoding of 0 had been given.
5281 */
5282 switch (op) {
5283 case 0: case 1:
5284 /* no-op */
5285 break;
5286 case 2: case 3:
5287 imm <<= 8;
5288 break;
5289 case 4: case 5:
5290 imm <<= 16;
5291 break;
5292 case 6: case 7:
5293 imm <<= 24;
5294 break;
5295 case 8: case 9:
5296 imm |= imm << 16;
5297 break;
5298 case 10: case 11:
5299 imm = (imm << 8) | (imm << 24);
5300 break;
5301 case 12:
5302 imm = (imm << 8) | 0xff;
5303 break;
5304 case 13:
5305 imm = (imm << 16) | 0xffff;
5306 break;
5307 case 14:
5308 imm |= (imm << 8) | (imm << 16) | (imm << 24);
5309 if (invert)
5310 imm = ~imm;
5311 break;
5312 case 15:
5313 if (invert) {
5314 return 1;
5315 }
5316 imm = ((imm & 0x80) << 24) | ((imm & 0x3f) << 19)
5317 | ((imm & 0x40) ? (0x1f << 25) : (1 << 30));
5318 break;
5319 }
5320 if (invert)
5321 imm = ~imm;
5322
5323 for (pass = 0; pass < (q ? 4 : 2); pass++) {
5324 if (op & 1 && op < 12) {
5325 tmp = neon_load_reg(rd, pass);
5326 if (invert) {
5327 /* The immediate value has already been inverted, so
5328 BIC becomes AND. */
5329 tcg_gen_andi_i32(tmp, tmp, imm);
5330 } else {
5331 tcg_gen_ori_i32(tmp, tmp, imm);
5332 }
5333 } else {
5334 /* VMOV, VMVN. */
5335 tmp = tcg_temp_new_i32();
5336 if (op == 14 && invert) {
5337 int n;
5338 uint32_t val;
5339 val = 0;
5340 for (n = 0; n < 4; n++) {
5341 if (imm & (1 << (n + (pass & 1) * 4)))
5342 val |= 0xff << (n * 8);
5343 }
5344 tcg_gen_movi_i32(tmp, val);
5345 } else {
5346 tcg_gen_movi_i32(tmp, imm);
5347 }
5348 }
5349 neon_store_reg(rd, pass, tmp);
5350 }
5351 }
5352 } else { /* (insn & 0x00800010 == 0x00800000) */
5353 if (size != 3) {
5354 op = (insn >> 8) & 0xf;
5355 if ((insn & (1 << 6)) == 0) {
5356 /* Three registers of different lengths. */
5357 int src1_wide;
5358 int src2_wide;
5359 int prewiden;
5360 /* undefreq: bit 0 : UNDEF if size != 0
5361 * bit 1 : UNDEF if size == 0
5362 * bit 2 : UNDEF if U == 1
5363 * Note that [1:0] set implies 'always UNDEF'
5364 */
5365 int undefreq;
5366 /* prewiden, src1_wide, src2_wide, undefreq */
5367 static const int neon_3reg_wide[16][4] = {
5368 {1, 0, 0, 0}, /* VADDL */
5369 {1, 1, 0, 0}, /* VADDW */
5370 {1, 0, 0, 0}, /* VSUBL */
5371 {1, 1, 0, 0}, /* VSUBW */
5372 {0, 1, 1, 0}, /* VADDHN */
5373 {0, 0, 0, 0}, /* VABAL */
5374 {0, 1, 1, 0}, /* VSUBHN */
5375 {0, 0, 0, 0}, /* VABDL */
5376 {0, 0, 0, 0}, /* VMLAL */
5377 {0, 0, 0, 6}, /* VQDMLAL */
5378 {0, 0, 0, 0}, /* VMLSL */
5379 {0, 0, 0, 6}, /* VQDMLSL */
5380 {0, 0, 0, 0}, /* Integer VMULL */
5381 {0, 0, 0, 2}, /* VQDMULL */
5382 {0, 0, 0, 5}, /* Polynomial VMULL */
5383 {0, 0, 0, 3}, /* Reserved: always UNDEF */
5384 };
5385
5386 prewiden = neon_3reg_wide[op][0];
5387 src1_wide = neon_3reg_wide[op][1];
5388 src2_wide = neon_3reg_wide[op][2];
5389 undefreq = neon_3reg_wide[op][3];
5390
5391 if (((undefreq & 1) && (size != 0)) ||
5392 ((undefreq & 2) && (size == 0)) ||
5393 ((undefreq & 4) && u)) {
5394 return 1;
5395 }
5396 if ((src1_wide && (rn & 1)) ||
5397 (src2_wide && (rm & 1)) ||
5398 (!src2_wide && (rd & 1))) {
5399 return 1;
5400 }
5401
5402 /* Avoid overlapping operands. Wide source operands are
5403 always aligned so will never overlap with wide
5404 destinations in problematic ways. */
5405 if (rd == rm && !src2_wide) {
5406 tmp = neon_load_reg(rm, 1);
5407 neon_store_scratch(2, tmp);
5408 } else if (rd == rn && !src1_wide) {
5409 tmp = neon_load_reg(rn, 1);
5410 neon_store_scratch(2, tmp);
5411 }
5412 TCGV_UNUSED_I32(tmp3);
5413 for (pass = 0; pass < 2; pass++) {
5414 if (src1_wide) {
5415 neon_load_reg64(cpu_V0, rn + pass);
5416 TCGV_UNUSED_I32(tmp);
5417 } else {
5418 if (pass == 1 && rd == rn) {
5419 tmp = neon_load_scratch(2);
5420 } else {
5421 tmp = neon_load_reg(rn, pass);
5422 }
5423 if (prewiden) {
5424 gen_neon_widen(cpu_V0, tmp, size, u);
5425 }
5426 }
5427 if (src2_wide) {
5428 neon_load_reg64(cpu_V1, rm + pass);
5429 TCGV_UNUSED_I32(tmp2);
5430 } else {
5431 if (pass == 1 && rd == rm) {
5432 tmp2 = neon_load_scratch(2);
5433 } else {
5434 tmp2 = neon_load_reg(rm, pass);
5435 }
5436 if (prewiden) {
5437 gen_neon_widen(cpu_V1, tmp2, size, u);
5438 }
5439 }
5440 switch (op) {
5441 case 0: case 1: case 4: /* VADDL, VADDW, VADDHN, VRADDHN */
5442 gen_neon_addl(size);
5443 break;
5444 case 2: case 3: case 6: /* VSUBL, VSUBW, VSUBHN, VRSUBHN */
5445 gen_neon_subl(size);
5446 break;
5447 case 5: case 7: /* VABAL, VABDL */
5448 switch ((size << 1) | u) {
5449 case 0:
5450 gen_helper_neon_abdl_s16(cpu_V0, tmp, tmp2);
5451 break;
5452 case 1:
5453 gen_helper_neon_abdl_u16(cpu_V0, tmp, tmp2);
5454 break;
5455 case 2:
5456 gen_helper_neon_abdl_s32(cpu_V0, tmp, tmp2);
5457 break;
5458 case 3:
5459 gen_helper_neon_abdl_u32(cpu_V0, tmp, tmp2);
5460 break;
5461 case 4:
5462 gen_helper_neon_abdl_s64(cpu_V0, tmp, tmp2);
5463 break;
5464 case 5:
5465 gen_helper_neon_abdl_u64(cpu_V0, tmp, tmp2);
5466 break;
5467 default: abort();
5468 }
5469 tcg_temp_free_i32(tmp2);
5470 tcg_temp_free_i32(tmp);
5471 break;
5472 case 8: case 9: case 10: case 11: case 12: case 13:
5473 /* VMLAL, VQDMLAL, VMLSL, VQDMLSL, VMULL, VQDMULL */
5474 gen_neon_mull(cpu_V0, tmp, tmp2, size, u);
5475 break;
5476 case 14: /* Polynomial VMULL */
5477 gen_helper_neon_mull_p8(cpu_V0, tmp, tmp2);
5478 tcg_temp_free_i32(tmp2);
5479 tcg_temp_free_i32(tmp);
5480 break;
5481 default: /* 15 is RESERVED: caught earlier */
5482 abort();
5483 }
5484 if (op == 13) {
5485 /* VQDMULL */
5486 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
5487 neon_store_reg64(cpu_V0, rd + pass);
5488 } else if (op == 5 || (op >= 8 && op <= 11)) {
5489 /* Accumulate. */
5490 neon_load_reg64(cpu_V1, rd + pass);
5491 switch (op) {
5492 case 10: /* VMLSL */
5493 gen_neon_negl(cpu_V0, size);
5494 /* Fall through */
5495 case 5: case 8: /* VABAL, VMLAL */
5496 gen_neon_addl(size);
5497 break;
5498 case 9: case 11: /* VQDMLAL, VQDMLSL */
5499 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
5500 if (op == 11) {
5501 gen_neon_negl(cpu_V0, size);
5502 }
5503 gen_neon_addl_saturate(cpu_V0, cpu_V1, size);
5504 break;
5505 default:
5506 abort();
5507 }
5508 neon_store_reg64(cpu_V0, rd + pass);
5509 } else if (op == 4 || op == 6) {
5510 /* Narrowing operation. */
5511 tmp = tcg_temp_new_i32();
5512 if (!u) {
5513 switch (size) {
5514 case 0:
5515 gen_helper_neon_narrow_high_u8(tmp, cpu_V0);
5516 break;
5517 case 1:
5518 gen_helper_neon_narrow_high_u16(tmp, cpu_V0);
5519 break;
5520 case 2:
5521 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
5522 tcg_gen_trunc_i64_i32(tmp, cpu_V0);
5523 break;
5524 default: abort();
5525 }
5526 } else {
5527 switch (size) {
5528 case 0:
5529 gen_helper_neon_narrow_round_high_u8(tmp, cpu_V0);
5530 break;
5531 case 1:
5532 gen_helper_neon_narrow_round_high_u16(tmp, cpu_V0);
5533 break;
5534 case 2:
5535 tcg_gen_addi_i64(cpu_V0, cpu_V0, 1u << 31);
5536 tcg_gen_shri_i64(cpu_V0, cpu_V0, 32);
5537 tcg_gen_trunc_i64_i32(tmp, cpu_V0);
5538 break;
5539 default: abort();
5540 }
5541 }
5542 if (pass == 0) {
5543 tmp3 = tmp;
5544 } else {
5545 neon_store_reg(rd, 0, tmp3);
5546 neon_store_reg(rd, 1, tmp);
5547 }
5548 } else {
5549 /* Write back the result. */
5550 neon_store_reg64(cpu_V0, rd + pass);
5551 }
5552 }
5553 } else {
5554 /* Two registers and a scalar. NB that for ops of this form
5555 * the ARM ARM labels bit 24 as Q, but it is in our variable
5556 * 'u', not 'q'.
5557 */
5558 if (size == 0) {
5559 return 1;
5560 }
5561 switch (op) {
5562 case 1: /* Float VMLA scalar */
5563 case 5: /* Floating point VMLS scalar */
5564 case 9: /* Floating point VMUL scalar */
5565 if (size == 1) {
5566 return 1;
5567 }
5568 /* fall through */
5569 case 0: /* Integer VMLA scalar */
5570 case 4: /* Integer VMLS scalar */
5571 case 8: /* Integer VMUL scalar */
5572 case 12: /* VQDMULH scalar */
5573 case 13: /* VQRDMULH scalar */
5574 if (u && ((rd | rn) & 1)) {
5575 return 1;
5576 }
5577 tmp = neon_get_scalar(size, rm);
5578 neon_store_scratch(0, tmp);
5579 for (pass = 0; pass < (u ? 4 : 2); pass++) {
5580 tmp = neon_load_scratch(0);
5581 tmp2 = neon_load_reg(rn, pass);
5582 if (op == 12) {
5583 if (size == 1) {
5584 gen_helper_neon_qdmulh_s16(tmp, cpu_env, tmp, tmp2);
5585 } else {
5586 gen_helper_neon_qdmulh_s32(tmp, cpu_env, tmp, tmp2);
5587 }
5588 } else if (op == 13) {
5589 if (size == 1) {
5590 gen_helper_neon_qrdmulh_s16(tmp, cpu_env, tmp, tmp2);
5591 } else {
5592 gen_helper_neon_qrdmulh_s32(tmp, cpu_env, tmp, tmp2);
5593 }
5594 } else if (op & 1) {
5595 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5596 gen_helper_vfp_muls(tmp, tmp, tmp2, fpstatus);
5597 tcg_temp_free_ptr(fpstatus);
5598 } else {
5599 switch (size) {
5600 case 0: gen_helper_neon_mul_u8(tmp, tmp, tmp2); break;
5601 case 1: gen_helper_neon_mul_u16(tmp, tmp, tmp2); break;
5602 case 2: tcg_gen_mul_i32(tmp, tmp, tmp2); break;
5603 default: abort();
5604 }
5605 }
5606 tcg_temp_free_i32(tmp2);
5607 if (op < 8) {
5608 /* Accumulate. */
5609 tmp2 = neon_load_reg(rd, pass);
5610 switch (op) {
5611 case 0:
5612 gen_neon_add(size, tmp, tmp2);
5613 break;
5614 case 1:
5615 {
5616 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5617 gen_helper_vfp_adds(tmp, tmp, tmp2, fpstatus);
5618 tcg_temp_free_ptr(fpstatus);
5619 break;
5620 }
5621 case 4:
5622 gen_neon_rsb(size, tmp, tmp2);
5623 break;
5624 case 5:
5625 {
5626 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
5627 gen_helper_vfp_subs(tmp, tmp2, tmp, fpstatus);
5628 tcg_temp_free_ptr(fpstatus);
5629 break;
5630 }
5631 default:
5632 abort();
5633 }
5634 tcg_temp_free_i32(tmp2);
5635 }
5636 neon_store_reg(rd, pass, tmp);
5637 }
5638 break;
5639 case 3: /* VQDMLAL scalar */
5640 case 7: /* VQDMLSL scalar */
5641 case 11: /* VQDMULL scalar */
5642 if (u == 1) {
5643 return 1;
5644 }
5645 /* fall through */
5646 case 2: /* VMLAL sclar */
5647 case 6: /* VMLSL scalar */
5648 case 10: /* VMULL scalar */
5649 if (rd & 1) {
5650 return 1;
5651 }
5652 tmp2 = neon_get_scalar(size, rm);
5653 /* We need a copy of tmp2 because gen_neon_mull
5654 * deletes it during pass 0. */
5655 tmp4 = tcg_temp_new_i32();
5656 tcg_gen_mov_i32(tmp4, tmp2);
5657 tmp3 = neon_load_reg(rn, 1);
5658
5659 for (pass = 0; pass < 2; pass++) {
5660 if (pass == 0) {
5661 tmp = neon_load_reg(rn, 0);
5662 } else {
5663 tmp = tmp3;
5664 tmp2 = tmp4;
5665 }
5666 gen_neon_mull(cpu_V0, tmp, tmp2, size, u);
5667 if (op != 11) {
5668 neon_load_reg64(cpu_V1, rd + pass);
5669 }
5670 switch (op) {
5671 case 6:
5672 gen_neon_negl(cpu_V0, size);
5673 /* Fall through */
5674 case 2:
5675 gen_neon_addl(size);
5676 break;
5677 case 3: case 7:
5678 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
5679 if (op == 7) {
5680 gen_neon_negl(cpu_V0, size);
5681 }
5682 gen_neon_addl_saturate(cpu_V0, cpu_V1, size);
5683 break;
5684 case 10:
5685 /* no-op */
5686 break;
5687 case 11:
5688 gen_neon_addl_saturate(cpu_V0, cpu_V0, size);
5689 break;
5690 default:
5691 abort();
5692 }
5693 neon_store_reg64(cpu_V0, rd + pass);
5694 }
5695
5696
5697 break;
5698 default: /* 14 and 15 are RESERVED */
5699 return 1;
5700 }
5701 }
5702 } else { /* size == 3 */
5703 if (!u) {
5704 /* Extract. */
5705 imm = (insn >> 8) & 0xf;
5706
5707 if (imm > 7 && !q)
5708 return 1;
5709
5710 if (q && ((rd | rn | rm) & 1)) {
5711 return 1;
5712 }
5713
5714 if (imm == 0) {
5715 neon_load_reg64(cpu_V0, rn);
5716 if (q) {
5717 neon_load_reg64(cpu_V1, rn + 1);
5718 }
5719 } else if (imm == 8) {
5720 neon_load_reg64(cpu_V0, rn + 1);
5721 if (q) {
5722 neon_load_reg64(cpu_V1, rm);
5723 }
5724 } else if (q) {
5725 tmp64 = tcg_temp_new_i64();
5726 if (imm < 8) {
5727 neon_load_reg64(cpu_V0, rn);
5728 neon_load_reg64(tmp64, rn + 1);
5729 } else {
5730 neon_load_reg64(cpu_V0, rn + 1);
5731 neon_load_reg64(tmp64, rm);
5732 }
5733 tcg_gen_shri_i64(cpu_V0, cpu_V0, (imm & 7) * 8);
5734 tcg_gen_shli_i64(cpu_V1, tmp64, 64 - ((imm & 7) * 8));
5735 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
5736 if (imm < 8) {
5737 neon_load_reg64(cpu_V1, rm);
5738 } else {
5739 neon_load_reg64(cpu_V1, rm + 1);
5740 imm -= 8;
5741 }
5742 tcg_gen_shli_i64(cpu_V1, cpu_V1, 64 - (imm * 8));
5743 tcg_gen_shri_i64(tmp64, tmp64, imm * 8);
5744 tcg_gen_or_i64(cpu_V1, cpu_V1, tmp64);
5745 tcg_temp_free_i64(tmp64);
5746 } else {
5747 /* BUGFIX */
5748 neon_load_reg64(cpu_V0, rn);
5749 tcg_gen_shri_i64(cpu_V0, cpu_V0, imm * 8);
5750 neon_load_reg64(cpu_V1, rm);
5751 tcg_gen_shli_i64(cpu_V1, cpu_V1, 64 - (imm * 8));
5752 tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
5753 }
5754 neon_store_reg64(cpu_V0, rd);
5755 if (q) {
5756 neon_store_reg64(cpu_V1, rd + 1);
5757 }
5758 } else if ((insn & (1 << 11)) == 0) {
5759 /* Two register misc. */
5760 op = ((insn >> 12) & 0x30) | ((insn >> 7) & 0xf);
5761 size = (insn >> 18) & 3;
5762 /* UNDEF for unknown op values and bad op-size combinations */
5763 if ((neon_2rm_sizes[op] & (1 << size)) == 0) {
5764 return 1;
5765 }
5766 if ((op != NEON_2RM_VMOVN && op != NEON_2RM_VQMOVN) &&
5767 q && ((rm | rd) & 1)) {
5768 return 1;
5769 }
5770 switch (op) {
5771 case NEON_2RM_VREV64:
5772 for (pass = 0; pass < (q ? 2 : 1); pass++) {
5773 tmp = neon_load_reg(rm, pass * 2);
5774 tmp2 = neon_load_reg(rm, pass * 2 + 1);
5775 switch (size) {
5776 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
5777 case 1: gen_swap_half(tmp); break;
5778 case 2: /* no-op */ break;
5779 default: abort();
5780 }
5781 neon_store_reg(rd, pass * 2 + 1, tmp);
5782 if (size == 2) {
5783 neon_store_reg(rd, pass * 2, tmp2);
5784 } else {
5785 switch (size) {
5786 case 0: tcg_gen_bswap32_i32(tmp2, tmp2); break;
5787 case 1: gen_swap_half(tmp2); break;
5788 default: abort();
5789 }
5790 neon_store_reg(rd, pass * 2, tmp2);
5791 }
5792 }
5793 break;
5794 case NEON_2RM_VPADDL: case NEON_2RM_VPADDL_U:
5795 case NEON_2RM_VPADAL: case NEON_2RM_VPADAL_U:
5796 for (pass = 0; pass < q + 1; pass++) {
5797 tmp = neon_load_reg(rm, pass * 2);
5798 gen_neon_widen(cpu_V0, tmp, size, op & 1);
5799 tmp = neon_load_reg(rm, pass * 2 + 1);
5800 gen_neon_widen(cpu_V1, tmp, size, op & 1);
5801 switch (size) {
5802 case 0: gen_helper_neon_paddl_u16(CPU_V001); break;
5803 case 1: gen_helper_neon_paddl_u32(CPU_V001); break;
5804 case 2: tcg_gen_add_i64(CPU_V001); break;
5805 default: abort();
5806 }
5807 if (op >= NEON_2RM_VPADAL) {
5808 /* Accumulate. */
5809 neon_load_reg64(cpu_V1, rd + pass);
5810 gen_neon_addl(size);
5811 }
5812 neon_store_reg64(cpu_V0, rd + pass);
5813 }
5814 break;
5815 case NEON_2RM_VTRN:
5816 if (size == 2) {
5817 int n;
5818 for (n = 0; n < (q ? 4 : 2); n += 2) {
5819 tmp = neon_load_reg(rm, n);
5820 tmp2 = neon_load_reg(rd, n + 1);
5821 neon_store_reg(rm, n, tmp2);
5822 neon_store_reg(rd, n + 1, tmp);
5823 }
5824 } else {
5825 goto elementwise;
5826 }
5827 break;
5828 case NEON_2RM_VUZP:
5829 if (gen_neon_unzip(rd, rm, size, q)) {
5830 return 1;
5831 }
5832 break;
5833 case NEON_2RM_VZIP:
5834 if (gen_neon_zip(rd, rm, size, q)) {
5835 return 1;
5836 }
5837 break;
5838 case NEON_2RM_VMOVN: case NEON_2RM_VQMOVN:
5839 /* also VQMOVUN; op field and mnemonics don't line up */
5840 if (rm & 1) {
5841 return 1;
5842 }
5843 TCGV_UNUSED_I32(tmp2);
5844 for (pass = 0; pass < 2; pass++) {
5845 neon_load_reg64(cpu_V0, rm + pass);
5846 tmp = tcg_temp_new_i32();
5847 gen_neon_narrow_op(op == NEON_2RM_VMOVN, q, size,
5848 tmp, cpu_V0);
5849 if (pass == 0) {
5850 tmp2 = tmp;
5851 } else {
5852 neon_store_reg(rd, 0, tmp2);
5853 neon_store_reg(rd, 1, tmp);
5854 }
5855 }
5856 break;
5857 case NEON_2RM_VSHLL:
5858 if (q || (rd & 1)) {
5859 return 1;
5860 }
5861 tmp = neon_load_reg(rm, 0);
5862 tmp2 = neon_load_reg(rm, 1);
5863 for (pass = 0; pass < 2; pass++) {
5864 if (pass == 1)
5865 tmp = tmp2;
5866 gen_neon_widen(cpu_V0, tmp, size, 1);
5867 tcg_gen_shli_i64(cpu_V0, cpu_V0, 8 << size);
5868 neon_store_reg64(cpu_V0, rd + pass);
5869 }
5870 break;
5871 case NEON_2RM_VCVT_F16_F32:
5872 if (!arm_feature(env, ARM_FEATURE_VFP_FP16) ||
5873 q || (rm & 1)) {
5874 return 1;
5875 }
5876 tmp = tcg_temp_new_i32();
5877 tmp2 = tcg_temp_new_i32();
5878 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 0));
5879 gen_helper_neon_fcvt_f32_to_f16(tmp, cpu_F0s, cpu_env);
5880 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 1));
5881 gen_helper_neon_fcvt_f32_to_f16(tmp2, cpu_F0s, cpu_env);
5882 tcg_gen_shli_i32(tmp2, tmp2, 16);
5883 tcg_gen_or_i32(tmp2, tmp2, tmp);
5884 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 2));
5885 gen_helper_neon_fcvt_f32_to_f16(tmp, cpu_F0s, cpu_env);
5886 tcg_gen_ld_f32(cpu_F0s, cpu_env, neon_reg_offset(rm, 3));
5887 neon_store_reg(rd, 0, tmp2);
5888 tmp2 = tcg_temp_new_i32();
5889 gen_helper_neon_fcvt_f32_to_f16(tmp2, cpu_F0s, cpu_env);
5890 tcg_gen_shli_i32(tmp2, tmp2, 16);
5891 tcg_gen_or_i32(tmp2, tmp2, tmp);
5892 neon_store_reg(rd, 1, tmp2);
5893 tcg_temp_free_i32(tmp);
5894 break;
5895 case NEON_2RM_VCVT_F32_F16:
5896 if (!arm_feature(env, ARM_FEATURE_VFP_FP16) ||
5897 q || (rd & 1)) {
5898 return 1;
5899 }
5900 tmp3 = tcg_temp_new_i32();
5901 tmp = neon_load_reg(rm, 0);
5902 tmp2 = neon_load_reg(rm, 1);
5903 tcg_gen_ext16u_i32(tmp3, tmp);
5904 gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
5905 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 0));
5906 tcg_gen_shri_i32(tmp3, tmp, 16);
5907 gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
5908 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 1));
5909 tcg_temp_free_i32(tmp);
5910 tcg_gen_ext16u_i32(tmp3, tmp2);
5911 gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
5912 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 2));
5913 tcg_gen_shri_i32(tmp3, tmp2, 16);
5914 gen_helper_neon_fcvt_f16_to_f32(cpu_F0s, tmp3, cpu_env);
5915 tcg_gen_st_f32(cpu_F0s, cpu_env, neon_reg_offset(rd, 3));
5916 tcg_temp_free_i32(tmp2);
5917 tcg_temp_free_i32(tmp3);
5918 break;
5919 default:
5920 elementwise:
5921 for (pass = 0; pass < (q ? 4 : 2); pass++) {
5922 if (neon_2rm_is_float_op(op)) {
5923 tcg_gen_ld_f32(cpu_F0s, cpu_env,
5924 neon_reg_offset(rm, pass));
5925 TCGV_UNUSED_I32(tmp);
5926 } else {
5927 tmp = neon_load_reg(rm, pass);
5928 }
5929 switch (op) {
5930 case NEON_2RM_VREV32:
5931 switch (size) {
5932 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
5933 case 1: gen_swap_half(tmp); break;
5934 default: abort();
5935 }
5936 break;
5937 case NEON_2RM_VREV16:
5938 gen_rev16(tmp);
5939 break;
5940 case NEON_2RM_VCLS:
5941 switch (size) {
5942 case 0: gen_helper_neon_cls_s8(tmp, tmp); break;
5943 case 1: gen_helper_neon_cls_s16(tmp, tmp); break;
5944 case 2: gen_helper_neon_cls_s32(tmp, tmp); break;
5945 default: abort();
5946 }
5947 break;
5948 case NEON_2RM_VCLZ:
5949 switch (size) {
5950 case 0: gen_helper_neon_clz_u8(tmp, tmp); break;
5951 case 1: gen_helper_neon_clz_u16(tmp, tmp); break;
5952 case 2: gen_helper_clz(tmp, tmp); break;
5953 default: abort();
5954 }
5955 break;
5956 case NEON_2RM_VCNT:
5957 gen_helper_neon_cnt_u8(tmp, tmp);
5958 break;
5959 case NEON_2RM_VMVN:
5960 tcg_gen_not_i32(tmp, tmp);
5961 break;
5962 case NEON_2RM_VQABS:
5963 switch (size) {
5964 case 0:
5965 gen_helper_neon_qabs_s8(tmp, cpu_env, tmp);
5966 break;
5967 case 1:
5968 gen_helper_neon_qabs_s16(tmp, cpu_env, tmp);
5969 break;
5970 case 2:
5971 gen_helper_neon_qabs_s32(tmp, cpu_env, tmp);
5972 break;
5973 default: abort();
5974 }
5975 break;
5976 case NEON_2RM_VQNEG:
5977 switch (size) {
5978 case 0:
5979 gen_helper_neon_qneg_s8(tmp, cpu_env, tmp);
5980 break;
5981 case 1:
5982 gen_helper_neon_qneg_s16(tmp, cpu_env, tmp);
5983 break;
5984 case 2:
5985 gen_helper_neon_qneg_s32(tmp, cpu_env, tmp);
5986 break;
5987 default: abort();
5988 }
5989 break;
5990 case NEON_2RM_VCGT0: case NEON_2RM_VCLE0:
5991 tmp2 = tcg_const_i32(0);
5992 switch(size) {
5993 case 0: gen_helper_neon_cgt_s8(tmp, tmp, tmp2); break;
5994 case 1: gen_helper_neon_cgt_s16(tmp, tmp, tmp2); break;
5995 case 2: gen_helper_neon_cgt_s32(tmp, tmp, tmp2); break;
5996 default: abort();
5997 }
5998 tcg_temp_free_i32(tmp2);
5999 if (op == NEON_2RM_VCLE0) {
6000 tcg_gen_not_i32(tmp, tmp);
6001 }
6002 break;
6003 case NEON_2RM_VCGE0: case NEON_2RM_VCLT0:
6004 tmp2 = tcg_const_i32(0);
6005 switch(size) {
6006 case 0: gen_helper_neon_cge_s8(tmp, tmp, tmp2); break;
6007 case 1: gen_helper_neon_cge_s16(tmp, tmp, tmp2); break;
6008 case 2: gen_helper_neon_cge_s32(tmp, tmp, tmp2); break;
6009 default: abort();
6010 }
6011 tcg_temp_free_i32(tmp2);
6012 if (op == NEON_2RM_VCLT0) {
6013 tcg_gen_not_i32(tmp, tmp);
6014 }
6015 break;
6016 case NEON_2RM_VCEQ0:
6017 tmp2 = tcg_const_i32(0);
6018 switch(size) {
6019 case 0: gen_helper_neon_ceq_u8(tmp, tmp, tmp2); break;
6020 case 1: gen_helper_neon_ceq_u16(tmp, tmp, tmp2); break;
6021 case 2: gen_helper_neon_ceq_u32(tmp, tmp, tmp2); break;
6022 default: abort();
6023 }
6024 tcg_temp_free_i32(tmp2);
6025 break;
6026 case NEON_2RM_VABS:
6027 switch(size) {
6028 case 0: gen_helper_neon_abs_s8(tmp, tmp); break;
6029 case 1: gen_helper_neon_abs_s16(tmp, tmp); break;
6030 case 2: tcg_gen_abs_i32(tmp, tmp); break;
6031 default: abort();
6032 }
6033 break;
6034 case NEON_2RM_VNEG:
6035 tmp2 = tcg_const_i32(0);
6036 gen_neon_rsb(size, tmp, tmp2);
6037 tcg_temp_free_i32(tmp2);
6038 break;
6039 case NEON_2RM_VCGT0_F:
6040 {
6041 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6042 tmp2 = tcg_const_i32(0);
6043 gen_helper_neon_cgt_f32(tmp, tmp, tmp2, fpstatus);
6044 tcg_temp_free_i32(tmp2);
6045 tcg_temp_free_ptr(fpstatus);
6046 break;
6047 }
6048 case NEON_2RM_VCGE0_F:
6049 {
6050 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6051 tmp2 = tcg_const_i32(0);
6052 gen_helper_neon_cge_f32(tmp, tmp, tmp2, fpstatus);
6053 tcg_temp_free_i32(tmp2);
6054 tcg_temp_free_ptr(fpstatus);
6055 break;
6056 }
6057 case NEON_2RM_VCEQ0_F:
6058 {
6059 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6060 tmp2 = tcg_const_i32(0);
6061 gen_helper_neon_ceq_f32(tmp, tmp, tmp2, fpstatus);
6062 tcg_temp_free_i32(tmp2);
6063 tcg_temp_free_ptr(fpstatus);
6064 break;
6065 }
6066 case NEON_2RM_VCLE0_F:
6067 {
6068 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6069 tmp2 = tcg_const_i32(0);
6070 gen_helper_neon_cge_f32(tmp, tmp2, tmp, fpstatus);
6071 tcg_temp_free_i32(tmp2);
6072 tcg_temp_free_ptr(fpstatus);
6073 break;
6074 }
6075 case NEON_2RM_VCLT0_F:
6076 {
6077 TCGv_ptr fpstatus = get_fpstatus_ptr(1);
6078 tmp2 = tcg_const_i32(0);
6079 gen_helper_neon_cgt_f32(tmp, tmp2, tmp, fpstatus);
6080 tcg_temp_free_i32(tmp2);
6081 tcg_temp_free_ptr(fpstatus);
6082 break;
6083 }
6084 case NEON_2RM_VABS_F:
6085 gen_vfp_abs(0);
6086 break;
6087 case NEON_2RM_VNEG_F:
6088 gen_vfp_neg(0);
6089 break;
6090 case NEON_2RM_VSWP:
6091 tmp2 = neon_load_reg(rd, pass);
6092 neon_store_reg(rm, pass, tmp2);
6093 break;
6094 case NEON_2RM_VTRN:
6095 tmp2 = neon_load_reg(rd, pass);
6096 switch (size) {
6097 case 0: gen_neon_trn_u8(tmp, tmp2); break;
6098 case 1: gen_neon_trn_u16(tmp, tmp2); break;
6099 default: abort();
6100 }
6101 neon_store_reg(rm, pass, tmp2);
6102 break;
6103 case NEON_2RM_VRECPE:
6104 gen_helper_recpe_u32(tmp, tmp, cpu_env);
6105 break;
6106 case NEON_2RM_VRSQRTE:
6107 gen_helper_rsqrte_u32(tmp, tmp, cpu_env);
6108 break;
6109 case NEON_2RM_VRECPE_F:
6110 gen_helper_recpe_f32(cpu_F0s, cpu_F0s, cpu_env);
6111 break;
6112 case NEON_2RM_VRSQRTE_F:
6113 gen_helper_rsqrte_f32(cpu_F0s, cpu_F0s, cpu_env);
6114 break;
6115 case NEON_2RM_VCVT_FS: /* VCVT.F32.S32 */
6116 gen_vfp_sito(0, 1);
6117 break;
6118 case NEON_2RM_VCVT_FU: /* VCVT.F32.U32 */
6119 gen_vfp_uito(0, 1);
6120 break;
6121 case NEON_2RM_VCVT_SF: /* VCVT.S32.F32 */
6122 gen_vfp_tosiz(0, 1);
6123 break;
6124 case NEON_2RM_VCVT_UF: /* VCVT.U32.F32 */
6125 gen_vfp_touiz(0, 1);
6126 break;
6127 default:
6128 /* Reserved op values were caught by the
6129 * neon_2rm_sizes[] check earlier.
6130 */
6131 abort();
6132 }
6133 if (neon_2rm_is_float_op(op)) {
6134 tcg_gen_st_f32(cpu_F0s, cpu_env,
6135 neon_reg_offset(rd, pass));
6136 } else {
6137 neon_store_reg(rd, pass, tmp);
6138 }
6139 }
6140 break;
6141 }
6142 } else if ((insn & (1 << 10)) == 0) {
6143 /* VTBL, VTBX. */
6144 int n = ((insn >> 8) & 3) + 1;
6145 if ((rn + n) > 32) {
6146 /* This is UNPREDICTABLE; we choose to UNDEF to avoid the
6147 * helper function running off the end of the register file.
6148 */
6149 return 1;
6150 }
6151 n <<= 3;
6152 if (insn & (1 << 6)) {
6153 tmp = neon_load_reg(rd, 0);
6154 } else {
6155 tmp = tcg_temp_new_i32();
6156 tcg_gen_movi_i32(tmp, 0);
6157 }
6158 tmp2 = neon_load_reg(rm, 0);
6159 tmp4 = tcg_const_i32(rn);
6160 tmp5 = tcg_const_i32(n);
6161 gen_helper_neon_tbl(tmp2, cpu_env, tmp2, tmp, tmp4, tmp5);
6162 tcg_temp_free_i32(tmp);
6163 if (insn & (1 << 6)) {
6164 tmp = neon_load_reg(rd, 1);
6165 } else {
6166 tmp = tcg_temp_new_i32();
6167 tcg_gen_movi_i32(tmp, 0);
6168 }
6169 tmp3 = neon_load_reg(rm, 1);
6170 gen_helper_neon_tbl(tmp3, cpu_env, tmp3, tmp, tmp4, tmp5);
6171 tcg_temp_free_i32(tmp5);
6172 tcg_temp_free_i32(tmp4);
6173 neon_store_reg(rd, 0, tmp2);
6174 neon_store_reg(rd, 1, tmp3);
6175 tcg_temp_free_i32(tmp);
6176 } else if ((insn & 0x380) == 0) {
6177 /* VDUP */
6178 if ((insn & (7 << 16)) == 0 || (q && (rd & 1))) {
6179 return 1;
6180 }
6181 if (insn & (1 << 19)) {
6182 tmp = neon_load_reg(rm, 1);
6183 } else {
6184 tmp = neon_load_reg(rm, 0);
6185 }
6186 if (insn & (1 << 16)) {
6187 gen_neon_dup_u8(tmp, ((insn >> 17) & 3) * 8);
6188 } else if (insn & (1 << 17)) {
6189 if ((insn >> 18) & 1)
6190 gen_neon_dup_high16(tmp);
6191 else
6192 gen_neon_dup_low16(tmp);
6193 }
6194 for (pass = 0; pass < (q ? 4 : 2); pass++) {
6195 tmp2 = tcg_temp_new_i32();
6196 tcg_gen_mov_i32(tmp2, tmp);
6197 neon_store_reg(rd, pass, tmp2);
6198 }
6199 tcg_temp_free_i32(tmp);
6200 } else {
6201 return 1;
6202 }
6203 }
6204 }
6205 return 0;
6206 }
6207
6208 static int disas_coproc_insn(CPUARMState * env, DisasContext *s, uint32_t insn)
6209 {
6210 int cpnum, is64, crn, crm, opc1, opc2, isread, rt, rt2;
6211 const ARMCPRegInfo *ri;
6212 ARMCPU *cpu = arm_env_get_cpu(env);
6213
6214 cpnum = (insn >> 8) & 0xf;
6215 if (arm_feature(env, ARM_FEATURE_XSCALE)
6216 && ((env->cp15.c15_cpar ^ 0x3fff) & (1 << cpnum)))
6217 return 1;
6218
6219 /* First check for coprocessor space used for actual instructions */
6220 switch (cpnum) {
6221 case 0:
6222 case 1:
6223 if (arm_feature(env, ARM_FEATURE_IWMMXT)) {
6224 return disas_iwmmxt_insn(env, s, insn);
6225 } else if (arm_feature(env, ARM_FEATURE_XSCALE)) {
6226 return disas_dsp_insn(env, s, insn);
6227 }
6228 return 1;
6229 case 10:
6230 case 11:
6231 return disas_vfp_insn (env, s, insn);
6232 default:
6233 break;
6234 }
6235
6236 /* Otherwise treat as a generic register access */
6237 is64 = (insn & (1 << 25)) == 0;
6238 if (!is64 && ((insn & (1 << 4)) == 0)) {
6239 /* cdp */
6240 return 1;
6241 }
6242
6243 crm = insn & 0xf;
6244 if (is64) {
6245 crn = 0;
6246 opc1 = (insn >> 4) & 0xf;
6247 opc2 = 0;
6248 rt2 = (insn >> 16) & 0xf;
6249 } else {
6250 crn = (insn >> 16) & 0xf;
6251 opc1 = (insn >> 21) & 7;
6252 opc2 = (insn >> 5) & 7;
6253 rt2 = 0;
6254 }
6255 isread = (insn >> 20) & 1;
6256 rt = (insn >> 12) & 0xf;
6257
6258 ri = get_arm_cp_reginfo(cpu,
6259 ENCODE_CP_REG(cpnum, is64, crn, crm, opc1, opc2));
6260 if (ri) {
6261 /* Check access permissions */
6262 if (!cp_access_ok(env, ri, isread)) {
6263 return 1;
6264 }
6265
6266 /* Handle special cases first */
6267 switch (ri->type & ~(ARM_CP_FLAG_MASK & ~ARM_CP_SPECIAL)) {
6268 case ARM_CP_NOP:
6269 return 0;
6270 case ARM_CP_WFI:
6271 if (isread) {
6272 return 1;
6273 }
6274 gen_set_pc_im(s->pc);
6275 s->is_jmp = DISAS_WFI;
6276 return 0;
6277 default:
6278 break;
6279 }
6280
6281 if (isread) {
6282 /* Read */
6283 if (is64) {
6284 TCGv_i64 tmp64;
6285 TCGv_i32 tmp;
6286 if (ri->type & ARM_CP_CONST) {
6287 tmp64 = tcg_const_i64(ri->resetvalue);
6288 } else if (ri->readfn) {
6289 TCGv_ptr tmpptr;
6290 gen_set_pc_im(s->pc);
6291 tmp64 = tcg_temp_new_i64();
6292 tmpptr = tcg_const_ptr(ri);
6293 gen_helper_get_cp_reg64(tmp64, cpu_env, tmpptr);
6294 tcg_temp_free_ptr(tmpptr);
6295 } else {
6296 tmp64 = tcg_temp_new_i64();
6297 tcg_gen_ld_i64(tmp64, cpu_env, ri->fieldoffset);
6298 }
6299 tmp = tcg_temp_new_i32();
6300 tcg_gen_trunc_i64_i32(tmp, tmp64);
6301 store_reg(s, rt, tmp);
6302 tcg_gen_shri_i64(tmp64, tmp64, 32);
6303 tmp = tcg_temp_new_i32();
6304 tcg_gen_trunc_i64_i32(tmp, tmp64);
6305 tcg_temp_free_i64(tmp64);
6306 store_reg(s, rt2, tmp);
6307 } else {
6308 TCGv_i32 tmp;
6309 if (ri->type & ARM_CP_CONST) {
6310 tmp = tcg_const_i32(ri->resetvalue);
6311 } else if (ri->readfn) {
6312 TCGv_ptr tmpptr;
6313 gen_set_pc_im(s->pc);
6314 tmp = tcg_temp_new_i32();
6315 tmpptr = tcg_const_ptr(ri);
6316 gen_helper_get_cp_reg(tmp, cpu_env, tmpptr);
6317 tcg_temp_free_ptr(tmpptr);
6318 } else {
6319 tmp = load_cpu_offset(ri->fieldoffset);
6320 }
6321 if (rt == 15) {
6322 /* Destination register of r15 for 32 bit loads sets
6323 * the condition codes from the high 4 bits of the value
6324 */
6325 gen_set_nzcv(tmp);
6326 tcg_temp_free_i32(tmp);
6327 } else {
6328 store_reg(s, rt, tmp);
6329 }
6330 }
6331 } else {
6332 /* Write */
6333 if (ri->type & ARM_CP_CONST) {
6334 /* If not forbidden by access permissions, treat as WI */
6335 return 0;
6336 }
6337
6338 if (is64) {
6339 TCGv_i32 tmplo, tmphi;
6340 TCGv_i64 tmp64 = tcg_temp_new_i64();
6341 tmplo = load_reg(s, rt);
6342 tmphi = load_reg(s, rt2);
6343 tcg_gen_concat_i32_i64(tmp64, tmplo, tmphi);
6344 tcg_temp_free_i32(tmplo);
6345 tcg_temp_free_i32(tmphi);
6346 if (ri->writefn) {
6347 TCGv_ptr tmpptr = tcg_const_ptr(ri);
6348 gen_set_pc_im(s->pc);
6349 gen_helper_set_cp_reg64(cpu_env, tmpptr, tmp64);
6350 tcg_temp_free_ptr(tmpptr);
6351 } else {
6352 tcg_gen_st_i64(tmp64, cpu_env, ri->fieldoffset);
6353 }
6354 tcg_temp_free_i64(tmp64);
6355 } else {
6356 if (ri->writefn) {
6357 TCGv_i32 tmp;
6358 TCGv_ptr tmpptr;
6359 gen_set_pc_im(s->pc);
6360 tmp = load_reg(s, rt);
6361 tmpptr = tcg_const_ptr(ri);
6362 gen_helper_set_cp_reg(cpu_env, tmpptr, tmp);
6363 tcg_temp_free_ptr(tmpptr);
6364 tcg_temp_free_i32(tmp);
6365 } else {
6366 TCGv_i32 tmp = load_reg(s, rt);
6367 store_cpu_offset(tmp, ri->fieldoffset);
6368 }
6369 }
6370 /* We default to ending the TB on a coprocessor register write,
6371 * but allow this to be suppressed by the register definition
6372 * (usually only necessary to work around guest bugs).
6373 */
6374 if (!(ri->type & ARM_CP_SUPPRESS_TB_END)) {
6375 gen_lookup_tb(s);
6376 }
6377 }
6378 return 0;
6379 }
6380
6381 return 1;
6382 }
6383
6384
6385 /* Store a 64-bit value to a register pair. Clobbers val. */
6386 static void gen_storeq_reg(DisasContext *s, int rlow, int rhigh, TCGv_i64 val)
6387 {
6388 TCGv_i32 tmp;
6389 tmp = tcg_temp_new_i32();
6390 tcg_gen_trunc_i64_i32(tmp, val);
6391 store_reg(s, rlow, tmp);
6392 tmp = tcg_temp_new_i32();
6393 tcg_gen_shri_i64(val, val, 32);
6394 tcg_gen_trunc_i64_i32(tmp, val);
6395 store_reg(s, rhigh, tmp);
6396 }
6397
6398 /* load a 32-bit value from a register and perform a 64-bit accumulate. */
6399 static void gen_addq_lo(DisasContext *s, TCGv_i64 val, int rlow)
6400 {
6401 TCGv_i64 tmp;
6402 TCGv_i32 tmp2;
6403
6404 /* Load value and extend to 64 bits. */
6405 tmp = tcg_temp_new_i64();
6406 tmp2 = load_reg(s, rlow);
6407 tcg_gen_extu_i32_i64(tmp, tmp2);
6408 tcg_temp_free_i32(tmp2);
6409 tcg_gen_add_i64(val, val, tmp);
6410 tcg_temp_free_i64(tmp);
6411 }
6412
6413 /* load and add a 64-bit value from a register pair. */
6414 static void gen_addq(DisasContext *s, TCGv_i64 val, int rlow, int rhigh)
6415 {
6416 TCGv_i64 tmp;
6417 TCGv_i32 tmpl;
6418 TCGv_i32 tmph;
6419
6420 /* Load 64-bit value rd:rn. */
6421 tmpl = load_reg(s, rlow);
6422 tmph = load_reg(s, rhigh);
6423 tmp = tcg_temp_new_i64();
6424 tcg_gen_concat_i32_i64(tmp, tmpl, tmph);
6425 tcg_temp_free_i32(tmpl);
6426 tcg_temp_free_i32(tmph);
6427 tcg_gen_add_i64(val, val, tmp);
6428 tcg_temp_free_i64(tmp);
6429 }
6430
6431 /* Set N and Z flags from hi|lo. */
6432 static void gen_logicq_cc(TCGv_i32 lo, TCGv_i32 hi)
6433 {
6434 tcg_gen_mov_i32(cpu_NF, hi);
6435 tcg_gen_or_i32(cpu_ZF, lo, hi);
6436 }
6437
6438 /* Load/Store exclusive instructions are implemented by remembering
6439 the value/address loaded, and seeing if these are the same
6440 when the store is performed. This should be sufficient to implement
6441 the architecturally mandated semantics, and avoids having to monitor
6442 regular stores.
6443
6444 In system emulation mode only one CPU will be running at once, so
6445 this sequence is effectively atomic. In user emulation mode we
6446 throw an exception and handle the atomic operation elsewhere. */
6447 static void gen_load_exclusive(DisasContext *s, int rt, int rt2,
6448 TCGv_i32 addr, int size)
6449 {
6450 TCGv_i32 tmp = tcg_temp_new_i32();
6451
6452 switch (size) {
6453 case 0:
6454 tcg_gen_qemu_ld8u(tmp, addr, IS_USER(s));
6455 break;
6456 case 1:
6457 tcg_gen_qemu_ld16u(tmp, addr, IS_USER(s));
6458 break;
6459 case 2:
6460 case 3:
6461 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
6462 break;
6463 default:
6464 abort();
6465 }
6466 tcg_gen_mov_i32(cpu_exclusive_val, tmp);
6467 store_reg(s, rt, tmp);
6468 if (size == 3) {
6469 TCGv_i32 tmp2 = tcg_temp_new_i32();
6470 tcg_gen_addi_i32(tmp2, addr, 4);
6471 tmp = tcg_temp_new_i32();
6472 tcg_gen_qemu_ld32u(tmp, tmp2, IS_USER(s));
6473 tcg_temp_free_i32(tmp2);
6474 tcg_gen_mov_i32(cpu_exclusive_high, tmp);
6475 store_reg(s, rt2, tmp);
6476 }
6477 tcg_gen_mov_i32(cpu_exclusive_addr, addr);
6478 }
6479
6480 static void gen_clrex(DisasContext *s)
6481 {
6482 tcg_gen_movi_i32(cpu_exclusive_addr, -1);
6483 }
6484
6485 #ifdef CONFIG_USER_ONLY
6486 static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
6487 TCGv_i32 addr, int size)
6488 {
6489 tcg_gen_mov_i32(cpu_exclusive_test, addr);
6490 tcg_gen_movi_i32(cpu_exclusive_info,
6491 size | (rd << 4) | (rt << 8) | (rt2 << 12));
6492 gen_exception_insn(s, 4, EXCP_STREX);
6493 }
6494 #else
6495 static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
6496 TCGv_i32 addr, int size)
6497 {
6498 TCGv_i32 tmp;
6499 int done_label;
6500 int fail_label;
6501
6502 /* if (env->exclusive_addr == addr && env->exclusive_val == [addr]) {
6503 [addr] = {Rt};
6504 {Rd} = 0;
6505 } else {
6506 {Rd} = 1;
6507 } */
6508 fail_label = gen_new_label();
6509 done_label = gen_new_label();
6510 tcg_gen_brcond_i32(TCG_COND_NE, addr, cpu_exclusive_addr, fail_label);
6511 tmp = tcg_temp_new_i32();
6512 switch (size) {
6513 case 0:
6514 tcg_gen_qemu_ld8u(tmp, addr, IS_USER(s));
6515 break;
6516 case 1:
6517 tcg_gen_qemu_ld16u(tmp, addr, IS_USER(s));
6518 break;
6519 case 2:
6520 case 3:
6521 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
6522 break;
6523 default:
6524 abort();
6525 }
6526 tcg_gen_brcond_i32(TCG_COND_NE, tmp, cpu_exclusive_val, fail_label);
6527 tcg_temp_free_i32(tmp);
6528 if (size == 3) {
6529 TCGv_i32 tmp2 = tcg_temp_new_i32();
6530 tcg_gen_addi_i32(tmp2, addr, 4);
6531 tmp = tcg_temp_new_i32();
6532 tcg_gen_qemu_ld32u(tmp, tmp2, IS_USER(s));
6533 tcg_temp_free_i32(tmp2);
6534 tcg_gen_brcond_i32(TCG_COND_NE, tmp, cpu_exclusive_high, fail_label);
6535 tcg_temp_free_i32(tmp);
6536 }
6537 tmp = load_reg(s, rt);
6538 switch (size) {
6539 case 0:
6540 tcg_gen_qemu_st8(tmp, addr, IS_USER(s));
6541 break;
6542 case 1:
6543 tcg_gen_qemu_st16(tmp, addr, IS_USER(s));
6544 break;
6545 case 2:
6546 case 3:
6547 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
6548 break;
6549 default:
6550 abort();
6551 }
6552 tcg_temp_free_i32(tmp);
6553 if (size == 3) {
6554 tcg_gen_addi_i32(addr, addr, 4);
6555 tmp = load_reg(s, rt2);
6556 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
6557 tcg_temp_free_i32(tmp);
6558 }
6559 tcg_gen_movi_i32(cpu_R[rd], 0);
6560 tcg_gen_br(done_label);
6561 gen_set_label(fail_label);
6562 tcg_gen_movi_i32(cpu_R[rd], 1);
6563 gen_set_label(done_label);
6564 tcg_gen_movi_i32(cpu_exclusive_addr, -1);
6565 }
6566 #endif
6567
6568 /* gen_srs:
6569 * @env: CPUARMState
6570 * @s: DisasContext
6571 * @mode: mode field from insn (which stack to store to)
6572 * @amode: addressing mode (DA/IA/DB/IB), encoded as per P,U bits in ARM insn
6573 * @writeback: true if writeback bit set
6574 *
6575 * Generate code for the SRS (Store Return State) insn.
6576 */
6577 static void gen_srs(DisasContext *s,
6578 uint32_t mode, uint32_t amode, bool writeback)
6579 {
6580 int32_t offset;
6581 TCGv_i32 addr = tcg_temp_new_i32();
6582 TCGv_i32 tmp = tcg_const_i32(mode);
6583 gen_helper_get_r13_banked(addr, cpu_env, tmp);
6584 tcg_temp_free_i32(tmp);
6585 switch (amode) {
6586 case 0: /* DA */
6587 offset = -4;
6588 break;
6589 case 1: /* IA */
6590 offset = 0;
6591 break;
6592 case 2: /* DB */
6593 offset = -8;
6594 break;
6595 case 3: /* IB */
6596 offset = 4;
6597 break;
6598 default:
6599 abort();
6600 }
6601 tcg_gen_addi_i32(addr, addr, offset);
6602 tmp = load_reg(s, 14);
6603 tcg_gen_qemu_st32(tmp, addr, 0);
6604 tcg_temp_free_i32(tmp);
6605 tmp = load_cpu_field(spsr);
6606 tcg_gen_addi_i32(addr, addr, 4);
6607 tcg_gen_qemu_st32(tmp, addr, 0);
6608 tcg_temp_free_i32(tmp);
6609 if (writeback) {
6610 switch (amode) {
6611 case 0:
6612 offset = -8;
6613 break;
6614 case 1:
6615 offset = 4;
6616 break;
6617 case 2:
6618 offset = -4;
6619 break;
6620 case 3:
6621 offset = 0;
6622 break;
6623 default:
6624 abort();
6625 }
6626 tcg_gen_addi_i32(addr, addr, offset);
6627 tmp = tcg_const_i32(mode);
6628 gen_helper_set_r13_banked(cpu_env, tmp, addr);
6629 tcg_temp_free_i32(tmp);
6630 }
6631 tcg_temp_free_i32(addr);
6632 }
6633
6634 static void disas_arm_insn(CPUARMState * env, DisasContext *s)
6635 {
6636 unsigned int cond, insn, val, op1, i, shift, rm, rs, rn, rd, sh;
6637 TCGv_i32 tmp;
6638 TCGv_i32 tmp2;
6639 TCGv_i32 tmp3;
6640 TCGv_i32 addr;
6641 TCGv_i64 tmp64;
6642
6643 insn = arm_ldl_code(env, s->pc, s->bswap_code);
6644 s->pc += 4;
6645
6646 /* M variants do not implement ARM mode. */
6647 if (IS_M(env))
6648 goto illegal_op;
6649 cond = insn >> 28;
6650 if (cond == 0xf){
6651 /* In ARMv3 and v4 the NV condition is UNPREDICTABLE; we
6652 * choose to UNDEF. In ARMv5 and above the space is used
6653 * for miscellaneous unconditional instructions.
6654 */
6655 ARCH(5);
6656
6657 /* Unconditional instructions. */
6658 if (((insn >> 25) & 7) == 1) {
6659 /* NEON Data processing. */
6660 if (!arm_feature(env, ARM_FEATURE_NEON))
6661 goto illegal_op;
6662
6663 if (disas_neon_data_insn(env, s, insn))
6664 goto illegal_op;
6665 return;
6666 }
6667 if ((insn & 0x0f100000) == 0x04000000) {
6668 /* NEON load/store. */
6669 if (!arm_feature(env, ARM_FEATURE_NEON))
6670 goto illegal_op;
6671
6672 if (disas_neon_ls_insn(env, s, insn))
6673 goto illegal_op;
6674 return;
6675 }
6676 if (((insn & 0x0f30f000) == 0x0510f000) ||
6677 ((insn & 0x0f30f010) == 0x0710f000)) {
6678 if ((insn & (1 << 22)) == 0) {
6679 /* PLDW; v7MP */
6680 if (!arm_feature(env, ARM_FEATURE_V7MP)) {
6681 goto illegal_op;
6682 }
6683 }
6684 /* Otherwise PLD; v5TE+ */
6685 ARCH(5TE);
6686 return;
6687 }
6688 if (((insn & 0x0f70f000) == 0x0450f000) ||
6689 ((insn & 0x0f70f010) == 0x0650f000)) {
6690 ARCH(7);
6691 return; /* PLI; V7 */
6692 }
6693 if (((insn & 0x0f700000) == 0x04100000) ||
6694 ((insn & 0x0f700010) == 0x06100000)) {
6695 if (!arm_feature(env, ARM_FEATURE_V7MP)) {
6696 goto illegal_op;
6697 }
6698 return; /* v7MP: Unallocated memory hint: must NOP */
6699 }
6700
6701 if ((insn & 0x0ffffdff) == 0x01010000) {
6702 ARCH(6);
6703 /* setend */
6704 if (((insn >> 9) & 1) != s->bswap_code) {
6705 /* Dynamic endianness switching not implemented. */
6706 goto illegal_op;
6707 }
6708 return;
6709 } else if ((insn & 0x0fffff00) == 0x057ff000) {
6710 switch ((insn >> 4) & 0xf) {
6711 case 1: /* clrex */
6712 ARCH(6K);
6713 gen_clrex(s);
6714 return;
6715 case 4: /* dsb */
6716 case 5: /* dmb */
6717 case 6: /* isb */
6718 ARCH(7);
6719 /* We don't emulate caches so these are a no-op. */
6720 return;
6721 default:
6722 goto illegal_op;
6723 }
6724 } else if ((insn & 0x0e5fffe0) == 0x084d0500) {
6725 /* srs */
6726 if (IS_USER(s)) {
6727 goto illegal_op;
6728 }
6729 ARCH(6);
6730 gen_srs(s, (insn & 0x1f), (insn >> 23) & 3, insn & (1 << 21));
6731 return;
6732 } else if ((insn & 0x0e50ffe0) == 0x08100a00) {
6733 /* rfe */
6734 int32_t offset;
6735 if (IS_USER(s))
6736 goto illegal_op;
6737 ARCH(6);
6738 rn = (insn >> 16) & 0xf;
6739 addr = load_reg(s, rn);
6740 i = (insn >> 23) & 3;
6741 switch (i) {
6742 case 0: offset = -4; break; /* DA */
6743 case 1: offset = 0; break; /* IA */
6744 case 2: offset = -8; break; /* DB */
6745 case 3: offset = 4; break; /* IB */
6746 default: abort();
6747 }
6748 if (offset)
6749 tcg_gen_addi_i32(addr, addr, offset);
6750 /* Load PC into tmp and CPSR into tmp2. */
6751 tmp = tcg_temp_new_i32();
6752 tcg_gen_qemu_ld32u(tmp, addr, 0);
6753 tcg_gen_addi_i32(addr, addr, 4);
6754 tmp2 = tcg_temp_new_i32();
6755 tcg_gen_qemu_ld32u(tmp2, addr, 0);
6756 if (insn & (1 << 21)) {
6757 /* Base writeback. */
6758 switch (i) {
6759 case 0: offset = -8; break;
6760 case 1: offset = 4; break;
6761 case 2: offset = -4; break;
6762 case 3: offset = 0; break;
6763 default: abort();
6764 }
6765 if (offset)
6766 tcg_gen_addi_i32(addr, addr, offset);
6767 store_reg(s, rn, addr);
6768 } else {
6769 tcg_temp_free_i32(addr);
6770 }
6771 gen_rfe(s, tmp, tmp2);
6772 return;
6773 } else if ((insn & 0x0e000000) == 0x0a000000) {
6774 /* branch link and change to thumb (blx <offset>) */
6775 int32_t offset;
6776
6777 val = (uint32_t)s->pc;
6778 tmp = tcg_temp_new_i32();
6779 tcg_gen_movi_i32(tmp, val);
6780 store_reg(s, 14, tmp);
6781 /* Sign-extend the 24-bit offset */
6782 offset = (((int32_t)insn) << 8) >> 8;
6783 /* offset * 4 + bit24 * 2 + (thumb bit) */
6784 val += (offset << 2) | ((insn >> 23) & 2) | 1;
6785 /* pipeline offset */
6786 val += 4;
6787 /* protected by ARCH(5); above, near the start of uncond block */
6788 gen_bx_im(s, val);
6789 return;
6790 } else if ((insn & 0x0e000f00) == 0x0c000100) {
6791 if (arm_feature(env, ARM_FEATURE_IWMMXT)) {
6792 /* iWMMXt register transfer. */
6793 if (env->cp15.c15_cpar & (1 << 1))
6794 if (!disas_iwmmxt_insn(env, s, insn))
6795 return;
6796 }
6797 } else if ((insn & 0x0fe00000) == 0x0c400000) {
6798 /* Coprocessor double register transfer. */
6799 ARCH(5TE);
6800 } else if ((insn & 0x0f000010) == 0x0e000010) {
6801 /* Additional coprocessor register transfer. */
6802 } else if ((insn & 0x0ff10020) == 0x01000000) {
6803 uint32_t mask;
6804 uint32_t val;
6805 /* cps (privileged) */
6806 if (IS_USER(s))
6807 return;
6808 mask = val = 0;
6809 if (insn & (1 << 19)) {
6810 if (insn & (1 << 8))
6811 mask |= CPSR_A;
6812 if (insn & (1 << 7))
6813 mask |= CPSR_I;
6814 if (insn & (1 << 6))
6815 mask |= CPSR_F;
6816 if (insn & (1 << 18))
6817 val |= mask;
6818 }
6819 if (insn & (1 << 17)) {
6820 mask |= CPSR_M;
6821 val |= (insn & 0x1f);
6822 }
6823 if (mask) {
6824 gen_set_psr_im(s, mask, 0, val);
6825 }
6826 return;
6827 }
6828 goto illegal_op;
6829 }
6830 if (cond != 0xe) {
6831 /* if not always execute, we generate a conditional jump to
6832 next instruction */
6833 s->condlabel = gen_new_label();
6834 gen_test_cc(cond ^ 1, s->condlabel);
6835 s->condjmp = 1;
6836 }
6837 if ((insn & 0x0f900000) == 0x03000000) {
6838 if ((insn & (1 << 21)) == 0) {
6839 ARCH(6T2);
6840 rd = (insn >> 12) & 0xf;
6841 val = ((insn >> 4) & 0xf000) | (insn & 0xfff);
6842 if ((insn & (1 << 22)) == 0) {
6843 /* MOVW */
6844 tmp = tcg_temp_new_i32();
6845 tcg_gen_movi_i32(tmp, val);
6846 } else {
6847 /* MOVT */
6848 tmp = load_reg(s, rd);
6849 tcg_gen_ext16u_i32(tmp, tmp);
6850 tcg_gen_ori_i32(tmp, tmp, val << 16);
6851 }
6852 store_reg(s, rd, tmp);
6853 } else {
6854 if (((insn >> 12) & 0xf) != 0xf)
6855 goto illegal_op;
6856 if (((insn >> 16) & 0xf) == 0) {
6857 gen_nop_hint(s, insn & 0xff);
6858 } else {
6859 /* CPSR = immediate */
6860 val = insn & 0xff;
6861 shift = ((insn >> 8) & 0xf) * 2;
6862 if (shift)
6863 val = (val >> shift) | (val << (32 - shift));
6864 i = ((insn & (1 << 22)) != 0);
6865 if (gen_set_psr_im(s, msr_mask(env, s, (insn >> 16) & 0xf, i), i, val))
6866 goto illegal_op;
6867 }
6868 }
6869 } else if ((insn & 0x0f900000) == 0x01000000
6870 && (insn & 0x00000090) != 0x00000090) {
6871 /* miscellaneous instructions */
6872 op1 = (insn >> 21) & 3;
6873 sh = (insn >> 4) & 0xf;
6874 rm = insn & 0xf;
6875 switch (sh) {
6876 case 0x0: /* move program status register */
6877 if (op1 & 1) {
6878 /* PSR = reg */
6879 tmp = load_reg(s, rm);
6880 i = ((op1 & 2) != 0);
6881 if (gen_set_psr(s, msr_mask(env, s, (insn >> 16) & 0xf, i), i, tmp))
6882 goto illegal_op;
6883 } else {
6884 /* reg = PSR */
6885 rd = (insn >> 12) & 0xf;
6886 if (op1 & 2) {
6887 if (IS_USER(s))
6888 goto illegal_op;
6889 tmp = load_cpu_field(spsr);
6890 } else {
6891 tmp = tcg_temp_new_i32();
6892 gen_helper_cpsr_read(tmp, cpu_env);
6893 }
6894 store_reg(s, rd, tmp);
6895 }
6896 break;
6897 case 0x1:
6898 if (op1 == 1) {
6899 /* branch/exchange thumb (bx). */
6900 ARCH(4T);
6901 tmp = load_reg(s, rm);
6902 gen_bx(s, tmp);
6903 } else if (op1 == 3) {
6904 /* clz */
6905 ARCH(5);
6906 rd = (insn >> 12) & 0xf;
6907 tmp = load_reg(s, rm);
6908 gen_helper_clz(tmp, tmp);
6909 store_reg(s, rd, tmp);
6910 } else {
6911 goto illegal_op;
6912 }
6913 break;
6914 case 0x2:
6915 if (op1 == 1) {
6916 ARCH(5J); /* bxj */
6917 /* Trivial implementation equivalent to bx. */
6918 tmp = load_reg(s, rm);
6919 gen_bx(s, tmp);
6920 } else {
6921 goto illegal_op;
6922 }
6923 break;
6924 case 0x3:
6925 if (op1 != 1)
6926 goto illegal_op;
6927
6928 ARCH(5);
6929 /* branch link/exchange thumb (blx) */
6930 tmp = load_reg(s, rm);
6931 tmp2 = tcg_temp_new_i32();
6932 tcg_gen_movi_i32(tmp2, s->pc);
6933 store_reg(s, 14, tmp2);
6934 gen_bx(s, tmp);
6935 break;
6936 case 0x5: /* saturating add/subtract */
6937 ARCH(5TE);
6938 rd = (insn >> 12) & 0xf;
6939 rn = (insn >> 16) & 0xf;
6940 tmp = load_reg(s, rm);
6941 tmp2 = load_reg(s, rn);
6942 if (op1 & 2)
6943 gen_helper_double_saturate(tmp2, cpu_env, tmp2);
6944 if (op1 & 1)
6945 gen_helper_sub_saturate(tmp, cpu_env, tmp, tmp2);
6946 else
6947 gen_helper_add_saturate(tmp, cpu_env, tmp, tmp2);
6948 tcg_temp_free_i32(tmp2);
6949 store_reg(s, rd, tmp);
6950 break;
6951 case 7:
6952 /* SMC instruction (op1 == 3)
6953 and undefined instructions (op1 == 0 || op1 == 2)
6954 will trap */
6955 if (op1 != 1) {
6956 goto illegal_op;
6957 }
6958 /* bkpt */
6959 ARCH(5);
6960 gen_exception_insn(s, 4, EXCP_BKPT);
6961 break;
6962 case 0x8: /* signed multiply */
6963 case 0xa:
6964 case 0xc:
6965 case 0xe:
6966 ARCH(5TE);
6967 rs = (insn >> 8) & 0xf;
6968 rn = (insn >> 12) & 0xf;
6969 rd = (insn >> 16) & 0xf;
6970 if (op1 == 1) {
6971 /* (32 * 16) >> 16 */
6972 tmp = load_reg(s, rm);
6973 tmp2 = load_reg(s, rs);
6974 if (sh & 4)
6975 tcg_gen_sari_i32(tmp2, tmp2, 16);
6976 else
6977 gen_sxth(tmp2);
6978 tmp64 = gen_muls_i64_i32(tmp, tmp2);
6979 tcg_gen_shri_i64(tmp64, tmp64, 16);
6980 tmp = tcg_temp_new_i32();
6981 tcg_gen_trunc_i64_i32(tmp, tmp64);
6982 tcg_temp_free_i64(tmp64);
6983 if ((sh & 2) == 0) {
6984 tmp2 = load_reg(s, rn);
6985 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
6986 tcg_temp_free_i32(tmp2);
6987 }
6988 store_reg(s, rd, tmp);
6989 } else {
6990 /* 16 * 16 */
6991 tmp = load_reg(s, rm);
6992 tmp2 = load_reg(s, rs);
6993 gen_mulxy(tmp, tmp2, sh & 2, sh & 4);
6994 tcg_temp_free_i32(tmp2);
6995 if (op1 == 2) {
6996 tmp64 = tcg_temp_new_i64();
6997 tcg_gen_ext_i32_i64(tmp64, tmp);
6998 tcg_temp_free_i32(tmp);
6999 gen_addq(s, tmp64, rn, rd);
7000 gen_storeq_reg(s, rn, rd, tmp64);
7001 tcg_temp_free_i64(tmp64);
7002 } else {
7003 if (op1 == 0) {
7004 tmp2 = load_reg(s, rn);
7005 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
7006 tcg_temp_free_i32(tmp2);
7007 }
7008 store_reg(s, rd, tmp);
7009 }
7010 }
7011 break;
7012 default:
7013 goto illegal_op;
7014 }
7015 } else if (((insn & 0x0e000000) == 0 &&
7016 (insn & 0x00000090) != 0x90) ||
7017 ((insn & 0x0e000000) == (1 << 25))) {
7018 int set_cc, logic_cc, shiftop;
7019
7020 op1 = (insn >> 21) & 0xf;
7021 set_cc = (insn >> 20) & 1;
7022 logic_cc = table_logic_cc[op1] & set_cc;
7023
7024 /* data processing instruction */
7025 if (insn & (1 << 25)) {
7026 /* immediate operand */
7027 val = insn & 0xff;
7028 shift = ((insn >> 8) & 0xf) * 2;
7029 if (shift) {
7030 val = (val >> shift) | (val << (32 - shift));
7031 }
7032 tmp2 = tcg_temp_new_i32();
7033 tcg_gen_movi_i32(tmp2, val);
7034 if (logic_cc && shift) {
7035 gen_set_CF_bit31(tmp2);
7036 }
7037 } else {
7038 /* register */
7039 rm = (insn) & 0xf;
7040 tmp2 = load_reg(s, rm);
7041 shiftop = (insn >> 5) & 3;
7042 if (!(insn & (1 << 4))) {
7043 shift = (insn >> 7) & 0x1f;
7044 gen_arm_shift_im(tmp2, shiftop, shift, logic_cc);
7045 } else {
7046 rs = (insn >> 8) & 0xf;
7047 tmp = load_reg(s, rs);
7048 gen_arm_shift_reg(tmp2, shiftop, tmp, logic_cc);
7049 }
7050 }
7051 if (op1 != 0x0f && op1 != 0x0d) {
7052 rn = (insn >> 16) & 0xf;
7053 tmp = load_reg(s, rn);
7054 } else {
7055 TCGV_UNUSED_I32(tmp);
7056 }
7057 rd = (insn >> 12) & 0xf;
7058 switch(op1) {
7059 case 0x00:
7060 tcg_gen_and_i32(tmp, tmp, tmp2);
7061 if (logic_cc) {
7062 gen_logic_CC(tmp);
7063 }
7064 store_reg_bx(env, s, rd, tmp);
7065 break;
7066 case 0x01:
7067 tcg_gen_xor_i32(tmp, tmp, tmp2);
7068 if (logic_cc) {
7069 gen_logic_CC(tmp);
7070 }
7071 store_reg_bx(env, s, rd, tmp);
7072 break;
7073 case 0x02:
7074 if (set_cc && rd == 15) {
7075 /* SUBS r15, ... is used for exception return. */
7076 if (IS_USER(s)) {
7077 goto illegal_op;
7078 }
7079 gen_sub_CC(tmp, tmp, tmp2);
7080 gen_exception_return(s, tmp);
7081 } else {
7082 if (set_cc) {
7083 gen_sub_CC(tmp, tmp, tmp2);
7084 } else {
7085 tcg_gen_sub_i32(tmp, tmp, tmp2);
7086 }
7087 store_reg_bx(env, s, rd, tmp);
7088 }
7089 break;
7090 case 0x03:
7091 if (set_cc) {
7092 gen_sub_CC(tmp, tmp2, tmp);
7093 } else {
7094 tcg_gen_sub_i32(tmp, tmp2, tmp);
7095 }
7096 store_reg_bx(env, s, rd, tmp);
7097 break;
7098 case 0x04:
7099 if (set_cc) {
7100 gen_add_CC(tmp, tmp, tmp2);
7101 } else {
7102 tcg_gen_add_i32(tmp, tmp, tmp2);
7103 }
7104 store_reg_bx(env, s, rd, tmp);
7105 break;
7106 case 0x05:
7107 if (set_cc) {
7108 gen_adc_CC(tmp, tmp, tmp2);
7109 } else {
7110 gen_add_carry(tmp, tmp, tmp2);
7111 }
7112 store_reg_bx(env, s, rd, tmp);
7113 break;
7114 case 0x06:
7115 if (set_cc) {
7116 gen_sbc_CC(tmp, tmp, tmp2);
7117 } else {
7118 gen_sub_carry(tmp, tmp, tmp2);
7119 }
7120 store_reg_bx(env, s, rd, tmp);
7121 break;
7122 case 0x07:
7123 if (set_cc) {
7124 gen_sbc_CC(tmp, tmp2, tmp);
7125 } else {
7126 gen_sub_carry(tmp, tmp2, tmp);
7127 }
7128 store_reg_bx(env, s, rd, tmp);
7129 break;
7130 case 0x08:
7131 if (set_cc) {
7132 tcg_gen_and_i32(tmp, tmp, tmp2);
7133 gen_logic_CC(tmp);
7134 }
7135 tcg_temp_free_i32(tmp);
7136 break;
7137 case 0x09:
7138 if (set_cc) {
7139 tcg_gen_xor_i32(tmp, tmp, tmp2);
7140 gen_logic_CC(tmp);
7141 }
7142 tcg_temp_free_i32(tmp);
7143 break;
7144 case 0x0a:
7145 if (set_cc) {
7146 gen_sub_CC(tmp, tmp, tmp2);
7147 }
7148 tcg_temp_free_i32(tmp);
7149 break;
7150 case 0x0b:
7151 if (set_cc) {
7152 gen_add_CC(tmp, tmp, tmp2);
7153 }
7154 tcg_temp_free_i32(tmp);
7155 break;
7156 case 0x0c:
7157 tcg_gen_or_i32(tmp, tmp, tmp2);
7158 if (logic_cc) {
7159 gen_logic_CC(tmp);
7160 }
7161 store_reg_bx(env, s, rd, tmp);
7162 break;
7163 case 0x0d:
7164 if (logic_cc && rd == 15) {
7165 /* MOVS r15, ... is used for exception return. */
7166 if (IS_USER(s)) {
7167 goto illegal_op;
7168 }
7169 gen_exception_return(s, tmp2);
7170 } else {
7171 if (logic_cc) {
7172 gen_logic_CC(tmp2);
7173 }
7174 store_reg_bx(env, s, rd, tmp2);
7175 }
7176 break;
7177 case 0x0e:
7178 tcg_gen_andc_i32(tmp, tmp, tmp2);
7179 if (logic_cc) {
7180 gen_logic_CC(tmp);
7181 }
7182 store_reg_bx(env, s, rd, tmp);
7183 break;
7184 default:
7185 case 0x0f:
7186 tcg_gen_not_i32(tmp2, tmp2);
7187 if (logic_cc) {
7188 gen_logic_CC(tmp2);
7189 }
7190 store_reg_bx(env, s, rd, tmp2);
7191 break;
7192 }
7193 if (op1 != 0x0f && op1 != 0x0d) {
7194 tcg_temp_free_i32(tmp2);
7195 }
7196 } else {
7197 /* other instructions */
7198 op1 = (insn >> 24) & 0xf;
7199 switch(op1) {
7200 case 0x0:
7201 case 0x1:
7202 /* multiplies, extra load/stores */
7203 sh = (insn >> 5) & 3;
7204 if (sh == 0) {
7205 if (op1 == 0x0) {
7206 rd = (insn >> 16) & 0xf;
7207 rn = (insn >> 12) & 0xf;
7208 rs = (insn >> 8) & 0xf;
7209 rm = (insn) & 0xf;
7210 op1 = (insn >> 20) & 0xf;
7211 switch (op1) {
7212 case 0: case 1: case 2: case 3: case 6:
7213 /* 32 bit mul */
7214 tmp = load_reg(s, rs);
7215 tmp2 = load_reg(s, rm);
7216 tcg_gen_mul_i32(tmp, tmp, tmp2);
7217 tcg_temp_free_i32(tmp2);
7218 if (insn & (1 << 22)) {
7219 /* Subtract (mls) */
7220 ARCH(6T2);
7221 tmp2 = load_reg(s, rn);
7222 tcg_gen_sub_i32(tmp, tmp2, tmp);
7223 tcg_temp_free_i32(tmp2);
7224 } else if (insn & (1 << 21)) {
7225 /* Add */
7226 tmp2 = load_reg(s, rn);
7227 tcg_gen_add_i32(tmp, tmp, tmp2);
7228 tcg_temp_free_i32(tmp2);
7229 }
7230 if (insn & (1 << 20))
7231 gen_logic_CC(tmp);
7232 store_reg(s, rd, tmp);
7233 break;
7234 case 4:
7235 /* 64 bit mul double accumulate (UMAAL) */
7236 ARCH(6);
7237 tmp = load_reg(s, rs);
7238 tmp2 = load_reg(s, rm);
7239 tmp64 = gen_mulu_i64_i32(tmp, tmp2);
7240 gen_addq_lo(s, tmp64, rn);
7241 gen_addq_lo(s, tmp64, rd);
7242 gen_storeq_reg(s, rn, rd, tmp64);
7243 tcg_temp_free_i64(tmp64);
7244 break;
7245 case 8: case 9: case 10: case 11:
7246 case 12: case 13: case 14: case 15:
7247 /* 64 bit mul: UMULL, UMLAL, SMULL, SMLAL. */
7248 tmp = load_reg(s, rs);
7249 tmp2 = load_reg(s, rm);
7250 if (insn & (1 << 22)) {
7251 tcg_gen_muls2_i32(tmp, tmp2, tmp, tmp2);
7252 } else {
7253 tcg_gen_mulu2_i32(tmp, tmp2, tmp, tmp2);
7254 }
7255 if (insn & (1 << 21)) { /* mult accumulate */
7256 TCGv_i32 al = load_reg(s, rn);
7257 TCGv_i32 ah = load_reg(s, rd);
7258 tcg_gen_add2_i32(tmp, tmp2, tmp, tmp2, al, ah);
7259 tcg_temp_free_i32(al);
7260 tcg_temp_free_i32(ah);
7261 }
7262 if (insn & (1 << 20)) {
7263 gen_logicq_cc(tmp, tmp2);
7264 }
7265 store_reg(s, rn, tmp);
7266 store_reg(s, rd, tmp2);
7267 break;
7268 default:
7269 goto illegal_op;
7270 }
7271 } else {
7272 rn = (insn >> 16) & 0xf;
7273 rd = (insn >> 12) & 0xf;
7274 if (insn & (1 << 23)) {
7275 /* load/store exclusive */
7276 op1 = (insn >> 21) & 0x3;
7277 if (op1)
7278 ARCH(6K);
7279 else
7280 ARCH(6);
7281 addr = tcg_temp_local_new_i32();
7282 load_reg_var(s, addr, rn);
7283 if (insn & (1 << 20)) {
7284 switch (op1) {
7285 case 0: /* ldrex */
7286 gen_load_exclusive(s, rd, 15, addr, 2);
7287 break;
7288 case 1: /* ldrexd */
7289 gen_load_exclusive(s, rd, rd + 1, addr, 3);
7290 break;
7291 case 2: /* ldrexb */
7292 gen_load_exclusive(s, rd, 15, addr, 0);
7293 break;
7294 case 3: /* ldrexh */
7295 gen_load_exclusive(s, rd, 15, addr, 1);
7296 break;
7297 default:
7298 abort();
7299 }
7300 } else {
7301 rm = insn & 0xf;
7302 switch (op1) {
7303 case 0: /* strex */
7304 gen_store_exclusive(s, rd, rm, 15, addr, 2);
7305 break;
7306 case 1: /* strexd */
7307 gen_store_exclusive(s, rd, rm, rm + 1, addr, 3);
7308 break;
7309 case 2: /* strexb */
7310 gen_store_exclusive(s, rd, rm, 15, addr, 0);
7311 break;
7312 case 3: /* strexh */
7313 gen_store_exclusive(s, rd, rm, 15, addr, 1);
7314 break;
7315 default:
7316 abort();
7317 }
7318 }
7319 tcg_temp_free_i32(addr);
7320 } else {
7321 /* SWP instruction */
7322 rm = (insn) & 0xf;
7323
7324 /* ??? This is not really atomic. However we know
7325 we never have multiple CPUs running in parallel,
7326 so it is good enough. */
7327 addr = load_reg(s, rn);
7328 tmp = load_reg(s, rm);
7329 tmp2 = tcg_temp_new_i32();
7330 if (insn & (1 << 22)) {
7331 tcg_gen_qemu_ld8u(tmp2, addr, IS_USER(s));
7332 tcg_gen_qemu_st8(tmp, addr, IS_USER(s));
7333 } else {
7334 tcg_gen_qemu_ld32u(tmp2, addr, IS_USER(s));
7335 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
7336 }
7337 tcg_temp_free_i32(tmp);
7338 tcg_temp_free_i32(addr);
7339 store_reg(s, rd, tmp2);
7340 }
7341 }
7342 } else {
7343 int address_offset;
7344 int load;
7345 /* Misc load/store */
7346 rn = (insn >> 16) & 0xf;
7347 rd = (insn >> 12) & 0xf;
7348 addr = load_reg(s, rn);
7349 if (insn & (1 << 24))
7350 gen_add_datah_offset(s, insn, 0, addr);
7351 address_offset = 0;
7352 if (insn & (1 << 20)) {
7353 /* load */
7354 tmp = tcg_temp_new_i32();
7355 switch(sh) {
7356 case 1:
7357 tcg_gen_qemu_ld16u(tmp, addr, IS_USER(s));
7358 break;
7359 case 2:
7360 tcg_gen_qemu_ld8s(tmp, addr, IS_USER(s));
7361 break;
7362 default:
7363 case 3:
7364 tcg_gen_qemu_ld16s(tmp, addr, IS_USER(s));
7365 break;
7366 }
7367 load = 1;
7368 } else if (sh & 2) {
7369 ARCH(5TE);
7370 /* doubleword */
7371 if (sh & 1) {
7372 /* store */
7373 tmp = load_reg(s, rd);
7374 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
7375 tcg_temp_free_i32(tmp);
7376 tcg_gen_addi_i32(addr, addr, 4);
7377 tmp = load_reg(s, rd + 1);
7378 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
7379 tcg_temp_free_i32(tmp);
7380 load = 0;
7381 } else {
7382 /* load */
7383 tmp = tcg_temp_new_i32();
7384 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
7385 store_reg(s, rd, tmp);
7386 tcg_gen_addi_i32(addr, addr, 4);
7387 tmp = tcg_temp_new_i32();
7388 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
7389 rd++;
7390 load = 1;
7391 }
7392 address_offset = -4;
7393 } else {
7394 /* store */
7395 tmp = load_reg(s, rd);
7396 tcg_gen_qemu_st16(tmp, addr, IS_USER(s));
7397 tcg_temp_free_i32(tmp);
7398 load = 0;
7399 }
7400 /* Perform base writeback before the loaded value to
7401 ensure correct behavior with overlapping index registers.
7402 ldrd with base writeback is is undefined if the
7403 destination and index registers overlap. */
7404 if (!(insn & (1 << 24))) {
7405 gen_add_datah_offset(s, insn, address_offset, addr);
7406 store_reg(s, rn, addr);
7407 } else if (insn & (1 << 21)) {
7408 if (address_offset)
7409 tcg_gen_addi_i32(addr, addr, address_offset);
7410 store_reg(s, rn, addr);
7411 } else {
7412 tcg_temp_free_i32(addr);
7413 }
7414 if (load) {
7415 /* Complete the load. */
7416 store_reg(s, rd, tmp);
7417 }
7418 }
7419 break;
7420 case 0x4:
7421 case 0x5:
7422 goto do_ldst;
7423 case 0x6:
7424 case 0x7:
7425 if (insn & (1 << 4)) {
7426 ARCH(6);
7427 /* Armv6 Media instructions. */
7428 rm = insn & 0xf;
7429 rn = (insn >> 16) & 0xf;
7430 rd = (insn >> 12) & 0xf;
7431 rs = (insn >> 8) & 0xf;
7432 switch ((insn >> 23) & 3) {
7433 case 0: /* Parallel add/subtract. */
7434 op1 = (insn >> 20) & 7;
7435 tmp = load_reg(s, rn);
7436 tmp2 = load_reg(s, rm);
7437 sh = (insn >> 5) & 7;
7438 if ((op1 & 3) == 0 || sh == 5 || sh == 6)
7439 goto illegal_op;
7440 gen_arm_parallel_addsub(op1, sh, tmp, tmp2);
7441 tcg_temp_free_i32(tmp2);
7442 store_reg(s, rd, tmp);
7443 break;
7444 case 1:
7445 if ((insn & 0x00700020) == 0) {
7446 /* Halfword pack. */
7447 tmp = load_reg(s, rn);
7448 tmp2 = load_reg(s, rm);
7449 shift = (insn >> 7) & 0x1f;
7450 if (insn & (1 << 6)) {
7451 /* pkhtb */
7452 if (shift == 0)
7453 shift = 31;
7454 tcg_gen_sari_i32(tmp2, tmp2, shift);
7455 tcg_gen_andi_i32(tmp, tmp, 0xffff0000);
7456 tcg_gen_ext16u_i32(tmp2, tmp2);
7457 } else {
7458 /* pkhbt */
7459 if (shift)
7460 tcg_gen_shli_i32(tmp2, tmp2, shift);
7461 tcg_gen_ext16u_i32(tmp, tmp);
7462 tcg_gen_andi_i32(tmp2, tmp2, 0xffff0000);
7463 }
7464 tcg_gen_or_i32(tmp, tmp, tmp2);
7465 tcg_temp_free_i32(tmp2);
7466 store_reg(s, rd, tmp);
7467 } else if ((insn & 0x00200020) == 0x00200000) {
7468 /* [us]sat */
7469 tmp = load_reg(s, rm);
7470 shift = (insn >> 7) & 0x1f;
7471 if (insn & (1 << 6)) {
7472 if (shift == 0)
7473 shift = 31;
7474 tcg_gen_sari_i32(tmp, tmp, shift);
7475 } else {
7476 tcg_gen_shli_i32(tmp, tmp, shift);
7477 }
7478 sh = (insn >> 16) & 0x1f;
7479 tmp2 = tcg_const_i32(sh);
7480 if (insn & (1 << 22))
7481 gen_helper_usat(tmp, cpu_env, tmp, tmp2);
7482 else
7483 gen_helper_ssat(tmp, cpu_env, tmp, tmp2);
7484 tcg_temp_free_i32(tmp2);
7485 store_reg(s, rd, tmp);
7486 } else if ((insn & 0x00300fe0) == 0x00200f20) {
7487 /* [us]sat16 */
7488 tmp = load_reg(s, rm);
7489 sh = (insn >> 16) & 0x1f;
7490 tmp2 = tcg_const_i32(sh);
7491 if (insn & (1 << 22))
7492 gen_helper_usat16(tmp, cpu_env, tmp, tmp2);
7493 else
7494 gen_helper_ssat16(tmp, cpu_env, tmp, tmp2);
7495 tcg_temp_free_i32(tmp2);
7496 store_reg(s, rd, tmp);
7497 } else if ((insn & 0x00700fe0) == 0x00000fa0) {
7498 /* Select bytes. */
7499 tmp = load_reg(s, rn);
7500 tmp2 = load_reg(s, rm);
7501 tmp3 = tcg_temp_new_i32();
7502 tcg_gen_ld_i32(tmp3, cpu_env, offsetof(CPUARMState, GE));
7503 gen_helper_sel_flags(tmp, tmp3, tmp, tmp2);
7504 tcg_temp_free_i32(tmp3);
7505 tcg_temp_free_i32(tmp2);
7506 store_reg(s, rd, tmp);
7507 } else if ((insn & 0x000003e0) == 0x00000060) {
7508 tmp = load_reg(s, rm);
7509 shift = (insn >> 10) & 3;
7510 /* ??? In many cases it's not necessary to do a
7511 rotate, a shift is sufficient. */
7512 if (shift != 0)
7513 tcg_gen_rotri_i32(tmp, tmp, shift * 8);
7514 op1 = (insn >> 20) & 7;
7515 switch (op1) {
7516 case 0: gen_sxtb16(tmp); break;
7517 case 2: gen_sxtb(tmp); break;
7518 case 3: gen_sxth(tmp); break;
7519 case 4: gen_uxtb16(tmp); break;
7520 case 6: gen_uxtb(tmp); break;
7521 case 7: gen_uxth(tmp); break;
7522 default: goto illegal_op;
7523 }
7524 if (rn != 15) {
7525 tmp2 = load_reg(s, rn);
7526 if ((op1 & 3) == 0) {
7527 gen_add16(tmp, tmp2);
7528 } else {
7529 tcg_gen_add_i32(tmp, tmp, tmp2);
7530 tcg_temp_free_i32(tmp2);
7531 }
7532 }
7533 store_reg(s, rd, tmp);
7534 } else if ((insn & 0x003f0f60) == 0x003f0f20) {
7535 /* rev */
7536 tmp = load_reg(s, rm);
7537 if (insn & (1 << 22)) {
7538 if (insn & (1 << 7)) {
7539 gen_revsh(tmp);
7540 } else {
7541 ARCH(6T2);
7542 gen_helper_rbit(tmp, tmp);
7543 }
7544 } else {
7545 if (insn & (1 << 7))
7546 gen_rev16(tmp);
7547 else
7548 tcg_gen_bswap32_i32(tmp, tmp);
7549 }
7550 store_reg(s, rd, tmp);
7551 } else {
7552 goto illegal_op;
7553 }
7554 break;
7555 case 2: /* Multiplies (Type 3). */
7556 switch ((insn >> 20) & 0x7) {
7557 case 5:
7558 if (((insn >> 6) ^ (insn >> 7)) & 1) {
7559 /* op2 not 00x or 11x : UNDEF */
7560 goto illegal_op;
7561 }
7562 /* Signed multiply most significant [accumulate].
7563 (SMMUL, SMMLA, SMMLS) */
7564 tmp = load_reg(s, rm);
7565 tmp2 = load_reg(s, rs);
7566 tmp64 = gen_muls_i64_i32(tmp, tmp2);
7567
7568 if (rd != 15) {
7569 tmp = load_reg(s, rd);
7570 if (insn & (1 << 6)) {
7571 tmp64 = gen_subq_msw(tmp64, tmp);
7572 } else {
7573 tmp64 = gen_addq_msw(tmp64, tmp);
7574 }
7575 }
7576 if (insn & (1 << 5)) {
7577 tcg_gen_addi_i64(tmp64, tmp64, 0x80000000u);
7578 }
7579 tcg_gen_shri_i64(tmp64, tmp64, 32);
7580 tmp = tcg_temp_new_i32();
7581 tcg_gen_trunc_i64_i32(tmp, tmp64);
7582 tcg_temp_free_i64(tmp64);
7583 store_reg(s, rn, tmp);
7584 break;
7585 case 0:
7586 case 4:
7587 /* SMLAD, SMUAD, SMLSD, SMUSD, SMLALD, SMLSLD */
7588 if (insn & (1 << 7)) {
7589 goto illegal_op;
7590 }
7591 tmp = load_reg(s, rm);
7592 tmp2 = load_reg(s, rs);
7593 if (insn & (1 << 5))
7594 gen_swap_half(tmp2);
7595 gen_smul_dual(tmp, tmp2);
7596 if (insn & (1 << 6)) {
7597 /* This subtraction cannot overflow. */
7598 tcg_gen_sub_i32(tmp, tmp, tmp2);
7599 } else {
7600 /* This addition cannot overflow 32 bits;
7601 * however it may overflow considered as a signed
7602 * operation, in which case we must set the Q flag.
7603 */
7604 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
7605 }
7606 tcg_temp_free_i32(tmp2);
7607 if (insn & (1 << 22)) {
7608 /* smlald, smlsld */
7609 tmp64 = tcg_temp_new_i64();
7610 tcg_gen_ext_i32_i64(tmp64, tmp);
7611 tcg_temp_free_i32(tmp);
7612 gen_addq(s, tmp64, rd, rn);
7613 gen_storeq_reg(s, rd, rn, tmp64);
7614 tcg_temp_free_i64(tmp64);
7615 } else {
7616 /* smuad, smusd, smlad, smlsd */
7617 if (rd != 15)
7618 {
7619 tmp2 = load_reg(s, rd);
7620 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
7621 tcg_temp_free_i32(tmp2);
7622 }
7623 store_reg(s, rn, tmp);
7624 }
7625 break;
7626 case 1:
7627 case 3:
7628 /* SDIV, UDIV */
7629 if (!arm_feature(env, ARM_FEATURE_ARM_DIV)) {
7630 goto illegal_op;
7631 }
7632 if (((insn >> 5) & 7) || (rd != 15)) {
7633 goto illegal_op;
7634 }
7635 tmp = load_reg(s, rm);
7636 tmp2 = load_reg(s, rs);
7637 if (insn & (1 << 21)) {
7638 gen_helper_udiv(tmp, tmp, tmp2);
7639 } else {
7640 gen_helper_sdiv(tmp, tmp, tmp2);
7641 }
7642 tcg_temp_free_i32(tmp2);
7643 store_reg(s, rn, tmp);
7644 break;
7645 default:
7646 goto illegal_op;
7647 }
7648 break;
7649 case 3:
7650 op1 = ((insn >> 17) & 0x38) | ((insn >> 5) & 7);
7651 switch (op1) {
7652 case 0: /* Unsigned sum of absolute differences. */
7653 ARCH(6);
7654 tmp = load_reg(s, rm);
7655 tmp2 = load_reg(s, rs);
7656 gen_helper_usad8(tmp, tmp, tmp2);
7657 tcg_temp_free_i32(tmp2);
7658 if (rd != 15) {
7659 tmp2 = load_reg(s, rd);
7660 tcg_gen_add_i32(tmp, tmp, tmp2);
7661 tcg_temp_free_i32(tmp2);
7662 }
7663 store_reg(s, rn, tmp);
7664 break;
7665 case 0x20: case 0x24: case 0x28: case 0x2c:
7666 /* Bitfield insert/clear. */
7667 ARCH(6T2);
7668 shift = (insn >> 7) & 0x1f;
7669 i = (insn >> 16) & 0x1f;
7670 i = i + 1 - shift;
7671 if (rm == 15) {
7672 tmp = tcg_temp_new_i32();
7673 tcg_gen_movi_i32(tmp, 0);
7674 } else {
7675 tmp = load_reg(s, rm);
7676 }
7677 if (i != 32) {
7678 tmp2 = load_reg(s, rd);
7679 tcg_gen_deposit_i32(tmp, tmp2, tmp, shift, i);
7680 tcg_temp_free_i32(tmp2);
7681 }
7682 store_reg(s, rd, tmp);
7683 break;
7684 case 0x12: case 0x16: case 0x1a: case 0x1e: /* sbfx */
7685 case 0x32: case 0x36: case 0x3a: case 0x3e: /* ubfx */
7686 ARCH(6T2);
7687 tmp = load_reg(s, rm);
7688 shift = (insn >> 7) & 0x1f;
7689 i = ((insn >> 16) & 0x1f) + 1;
7690 if (shift + i > 32)
7691 goto illegal_op;
7692 if (i < 32) {
7693 if (op1 & 0x20) {
7694 gen_ubfx(tmp, shift, (1u << i) - 1);
7695 } else {
7696 gen_sbfx(tmp, shift, i);
7697 }
7698 }
7699 store_reg(s, rd, tmp);
7700 break;
7701 default:
7702 goto illegal_op;
7703 }
7704 break;
7705 }
7706 break;
7707 }
7708 do_ldst:
7709 /* Check for undefined extension instructions
7710 * per the ARM Bible IE:
7711 * xxxx 0111 1111 xxxx xxxx xxxx 1111 xxxx
7712 */
7713 sh = (0xf << 20) | (0xf << 4);
7714 if (op1 == 0x7 && ((insn & sh) == sh))
7715 {
7716 goto illegal_op;
7717 }
7718 /* load/store byte/word */
7719 rn = (insn >> 16) & 0xf;
7720 rd = (insn >> 12) & 0xf;
7721 tmp2 = load_reg(s, rn);
7722 i = (IS_USER(s) || (insn & 0x01200000) == 0x00200000);
7723 if (insn & (1 << 24))
7724 gen_add_data_offset(s, insn, tmp2);
7725 if (insn & (1 << 20)) {
7726 /* load */
7727 tmp = tcg_temp_new_i32();
7728 if (insn & (1 << 22)) {
7729 tcg_gen_qemu_ld8u(tmp, tmp2, i);
7730 } else {
7731 tcg_gen_qemu_ld32u(tmp, tmp2, i);
7732 }
7733 } else {
7734 /* store */
7735 tmp = load_reg(s, rd);
7736 if (insn & (1 << 22)) {
7737 tcg_gen_qemu_st8(tmp, tmp2, i);
7738 } else {
7739 tcg_gen_qemu_st32(tmp, tmp2, i);
7740 }
7741 tcg_temp_free_i32(tmp);
7742 }
7743 if (!(insn & (1 << 24))) {
7744 gen_add_data_offset(s, insn, tmp2);
7745 store_reg(s, rn, tmp2);
7746 } else if (insn & (1 << 21)) {
7747 store_reg(s, rn, tmp2);
7748 } else {
7749 tcg_temp_free_i32(tmp2);
7750 }
7751 if (insn & (1 << 20)) {
7752 /* Complete the load. */
7753 store_reg_from_load(env, s, rd, tmp);
7754 }
7755 break;
7756 case 0x08:
7757 case 0x09:
7758 {
7759 int j, n, user, loaded_base;
7760 TCGv_i32 loaded_var;
7761 /* load/store multiple words */
7762 /* XXX: store correct base if write back */
7763 user = 0;
7764 if (insn & (1 << 22)) {
7765 if (IS_USER(s))
7766 goto illegal_op; /* only usable in supervisor mode */
7767
7768 if ((insn & (1 << 15)) == 0)
7769 user = 1;
7770 }
7771 rn = (insn >> 16) & 0xf;
7772 addr = load_reg(s, rn);
7773
7774 /* compute total size */
7775 loaded_base = 0;
7776 TCGV_UNUSED_I32(loaded_var);
7777 n = 0;
7778 for(i=0;i<16;i++) {
7779 if (insn & (1 << i))
7780 n++;
7781 }
7782 /* XXX: test invalid n == 0 case ? */
7783 if (insn & (1 << 23)) {
7784 if (insn & (1 << 24)) {
7785 /* pre increment */
7786 tcg_gen_addi_i32(addr, addr, 4);
7787 } else {
7788 /* post increment */
7789 }
7790 } else {
7791 if (insn & (1 << 24)) {
7792 /* pre decrement */
7793 tcg_gen_addi_i32(addr, addr, -(n * 4));
7794 } else {
7795 /* post decrement */
7796 if (n != 1)
7797 tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
7798 }
7799 }
7800 j = 0;
7801 for(i=0;i<16;i++) {
7802 if (insn & (1 << i)) {
7803 if (insn & (1 << 20)) {
7804 /* load */
7805 tmp = tcg_temp_new_i32();
7806 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
7807 if (user) {
7808 tmp2 = tcg_const_i32(i);
7809 gen_helper_set_user_reg(cpu_env, tmp2, tmp);
7810 tcg_temp_free_i32(tmp2);
7811 tcg_temp_free_i32(tmp);
7812 } else if (i == rn) {
7813 loaded_var = tmp;
7814 loaded_base = 1;
7815 } else {
7816 store_reg_from_load(env, s, i, tmp);
7817 }
7818 } else {
7819 /* store */
7820 if (i == 15) {
7821 /* special case: r15 = PC + 8 */
7822 val = (long)s->pc + 4;
7823 tmp = tcg_temp_new_i32();
7824 tcg_gen_movi_i32(tmp, val);
7825 } else if (user) {
7826 tmp = tcg_temp_new_i32();
7827 tmp2 = tcg_const_i32(i);
7828 gen_helper_get_user_reg(tmp, cpu_env, tmp2);
7829 tcg_temp_free_i32(tmp2);
7830 } else {
7831 tmp = load_reg(s, i);
7832 }
7833 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
7834 tcg_temp_free_i32(tmp);
7835 }
7836 j++;
7837 /* no need to add after the last transfer */
7838 if (j != n)
7839 tcg_gen_addi_i32(addr, addr, 4);
7840 }
7841 }
7842 if (insn & (1 << 21)) {
7843 /* write back */
7844 if (insn & (1 << 23)) {
7845 if (insn & (1 << 24)) {
7846 /* pre increment */
7847 } else {
7848 /* post increment */
7849 tcg_gen_addi_i32(addr, addr, 4);
7850 }
7851 } else {
7852 if (insn & (1 << 24)) {
7853 /* pre decrement */
7854 if (n != 1)
7855 tcg_gen_addi_i32(addr, addr, -((n - 1) * 4));
7856 } else {
7857 /* post decrement */
7858 tcg_gen_addi_i32(addr, addr, -(n * 4));
7859 }
7860 }
7861 store_reg(s, rn, addr);
7862 } else {
7863 tcg_temp_free_i32(addr);
7864 }
7865 if (loaded_base) {
7866 store_reg(s, rn, loaded_var);
7867 }
7868 if ((insn & (1 << 22)) && !user) {
7869 /* Restore CPSR from SPSR. */
7870 tmp = load_cpu_field(spsr);
7871 gen_set_cpsr(tmp, 0xffffffff);
7872 tcg_temp_free_i32(tmp);
7873 s->is_jmp = DISAS_UPDATE;
7874 }
7875 }
7876 break;
7877 case 0xa:
7878 case 0xb:
7879 {
7880 int32_t offset;
7881
7882 /* branch (and link) */
7883 val = (int32_t)s->pc;
7884 if (insn & (1 << 24)) {
7885 tmp = tcg_temp_new_i32();
7886 tcg_gen_movi_i32(tmp, val);
7887 store_reg(s, 14, tmp);
7888 }
7889 offset = (((int32_t)insn << 8) >> 8);
7890 val += (offset << 2) + 4;
7891 gen_jmp(s, val);
7892 }
7893 break;
7894 case 0xc:
7895 case 0xd:
7896 case 0xe:
7897 /* Coprocessor. */
7898 if (disas_coproc_insn(env, s, insn))
7899 goto illegal_op;
7900 break;
7901 case 0xf:
7902 /* swi */
7903 gen_set_pc_im(s->pc);
7904 s->is_jmp = DISAS_SWI;
7905 break;
7906 default:
7907 illegal_op:
7908 gen_exception_insn(s, 4, EXCP_UDEF);
7909 break;
7910 }
7911 }
7912 }
7913
7914 /* Return true if this is a Thumb-2 logical op. */
7915 static int
7916 thumb2_logic_op(int op)
7917 {
7918 return (op < 8);
7919 }
7920
7921 /* Generate code for a Thumb-2 data processing operation. If CONDS is nonzero
7922 then set condition code flags based on the result of the operation.
7923 If SHIFTER_OUT is nonzero then set the carry flag for logical operations
7924 to the high bit of T1.
7925 Returns zero if the opcode is valid. */
7926
7927 static int
7928 gen_thumb2_data_op(DisasContext *s, int op, int conds, uint32_t shifter_out,
7929 TCGv_i32 t0, TCGv_i32 t1)
7930 {
7931 int logic_cc;
7932
7933 logic_cc = 0;
7934 switch (op) {
7935 case 0: /* and */
7936 tcg_gen_and_i32(t0, t0, t1);
7937 logic_cc = conds;
7938 break;
7939 case 1: /* bic */
7940 tcg_gen_andc_i32(t0, t0, t1);
7941 logic_cc = conds;
7942 break;
7943 case 2: /* orr */
7944 tcg_gen_or_i32(t0, t0, t1);
7945 logic_cc = conds;
7946 break;
7947 case 3: /* orn */
7948 tcg_gen_orc_i32(t0, t0, t1);
7949 logic_cc = conds;
7950 break;
7951 case 4: /* eor */
7952 tcg_gen_xor_i32(t0, t0, t1);
7953 logic_cc = conds;
7954 break;
7955 case 8: /* add */
7956 if (conds)
7957 gen_add_CC(t0, t0, t1);
7958 else
7959 tcg_gen_add_i32(t0, t0, t1);
7960 break;
7961 case 10: /* adc */
7962 if (conds)
7963 gen_adc_CC(t0, t0, t1);
7964 else
7965 gen_adc(t0, t1);
7966 break;
7967 case 11: /* sbc */
7968 if (conds) {
7969 gen_sbc_CC(t0, t0, t1);
7970 } else {
7971 gen_sub_carry(t0, t0, t1);
7972 }
7973 break;
7974 case 13: /* sub */
7975 if (conds)
7976 gen_sub_CC(t0, t0, t1);
7977 else
7978 tcg_gen_sub_i32(t0, t0, t1);
7979 break;
7980 case 14: /* rsb */
7981 if (conds)
7982 gen_sub_CC(t0, t1, t0);
7983 else
7984 tcg_gen_sub_i32(t0, t1, t0);
7985 break;
7986 default: /* 5, 6, 7, 9, 12, 15. */
7987 return 1;
7988 }
7989 if (logic_cc) {
7990 gen_logic_CC(t0);
7991 if (shifter_out)
7992 gen_set_CF_bit31(t1);
7993 }
7994 return 0;
7995 }
7996
7997 /* Translate a 32-bit thumb instruction. Returns nonzero if the instruction
7998 is not legal. */
7999 static int disas_thumb2_insn(CPUARMState *env, DisasContext *s, uint16_t insn_hw1)
8000 {
8001 uint32_t insn, imm, shift, offset;
8002 uint32_t rd, rn, rm, rs;
8003 TCGv_i32 tmp;
8004 TCGv_i32 tmp2;
8005 TCGv_i32 tmp3;
8006 TCGv_i32 addr;
8007 TCGv_i64 tmp64;
8008 int op;
8009 int shiftop;
8010 int conds;
8011 int logic_cc;
8012
8013 if (!(arm_feature(env, ARM_FEATURE_THUMB2)
8014 || arm_feature (env, ARM_FEATURE_M))) {
8015 /* Thumb-1 cores may need to treat bl and blx as a pair of
8016 16-bit instructions to get correct prefetch abort behavior. */
8017 insn = insn_hw1;
8018 if ((insn & (1 << 12)) == 0) {
8019 ARCH(5);
8020 /* Second half of blx. */
8021 offset = ((insn & 0x7ff) << 1);
8022 tmp = load_reg(s, 14);
8023 tcg_gen_addi_i32(tmp, tmp, offset);
8024 tcg_gen_andi_i32(tmp, tmp, 0xfffffffc);
8025
8026 tmp2 = tcg_temp_new_i32();
8027 tcg_gen_movi_i32(tmp2, s->pc | 1);
8028 store_reg(s, 14, tmp2);
8029 gen_bx(s, tmp);
8030 return 0;
8031 }
8032 if (insn & (1 << 11)) {
8033 /* Second half of bl. */
8034 offset = ((insn & 0x7ff) << 1) | 1;
8035 tmp = load_reg(s, 14);
8036 tcg_gen_addi_i32(tmp, tmp, offset);
8037
8038 tmp2 = tcg_temp_new_i32();
8039 tcg_gen_movi_i32(tmp2, s->pc | 1);
8040 store_reg(s, 14, tmp2);
8041 gen_bx(s, tmp);
8042 return 0;
8043 }
8044 if ((s->pc & ~TARGET_PAGE_MASK) == 0) {
8045 /* Instruction spans a page boundary. Implement it as two
8046 16-bit instructions in case the second half causes an
8047 prefetch abort. */
8048 offset = ((int32_t)insn << 21) >> 9;
8049 tcg_gen_movi_i32(cpu_R[14], s->pc + 2 + offset);
8050 return 0;
8051 }
8052 /* Fall through to 32-bit decode. */
8053 }
8054
8055 insn = arm_lduw_code(env, s->pc, s->bswap_code);
8056 s->pc += 2;
8057 insn |= (uint32_t)insn_hw1 << 16;
8058
8059 if ((insn & 0xf800e800) != 0xf000e800) {
8060 ARCH(6T2);
8061 }
8062
8063 rn = (insn >> 16) & 0xf;
8064 rs = (insn >> 12) & 0xf;
8065 rd = (insn >> 8) & 0xf;
8066 rm = insn & 0xf;
8067 switch ((insn >> 25) & 0xf) {
8068 case 0: case 1: case 2: case 3:
8069 /* 16-bit instructions. Should never happen. */
8070 abort();
8071 case 4:
8072 if (insn & (1 << 22)) {
8073 /* Other load/store, table branch. */
8074 if (insn & 0x01200000) {
8075 /* Load/store doubleword. */
8076 if (rn == 15) {
8077 addr = tcg_temp_new_i32();
8078 tcg_gen_movi_i32(addr, s->pc & ~3);
8079 } else {
8080 addr = load_reg(s, rn);
8081 }
8082 offset = (insn & 0xff) * 4;
8083 if ((insn & (1 << 23)) == 0)
8084 offset = -offset;
8085 if (insn & (1 << 24)) {
8086 tcg_gen_addi_i32(addr, addr, offset);
8087 offset = 0;
8088 }
8089 if (insn & (1 << 20)) {
8090 /* ldrd */
8091 tmp = tcg_temp_new_i32();
8092 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
8093 store_reg(s, rs, tmp);
8094 tcg_gen_addi_i32(addr, addr, 4);
8095 tmp = tcg_temp_new_i32();
8096 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
8097 store_reg(s, rd, tmp);
8098 } else {
8099 /* strd */
8100 tmp = load_reg(s, rs);
8101 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
8102 tcg_temp_free_i32(tmp);
8103 tcg_gen_addi_i32(addr, addr, 4);
8104 tmp = load_reg(s, rd);
8105 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
8106 tcg_temp_free_i32(tmp);
8107 }
8108 if (insn & (1 << 21)) {
8109 /* Base writeback. */
8110 if (rn == 15)
8111 goto illegal_op;
8112 tcg_gen_addi_i32(addr, addr, offset - 4);
8113 store_reg(s, rn, addr);
8114 } else {
8115 tcg_temp_free_i32(addr);
8116 }
8117 } else if ((insn & (1 << 23)) == 0) {
8118 /* Load/store exclusive word. */
8119 addr = tcg_temp_local_new_i32();
8120 load_reg_var(s, addr, rn);
8121 tcg_gen_addi_i32(addr, addr, (insn & 0xff) << 2);
8122 if (insn & (1 << 20)) {
8123 gen_load_exclusive(s, rs, 15, addr, 2);
8124 } else {
8125 gen_store_exclusive(s, rd, rs, 15, addr, 2);
8126 }
8127 tcg_temp_free_i32(addr);
8128 } else if ((insn & (1 << 6)) == 0) {
8129 /* Table Branch. */
8130 if (rn == 15) {
8131 addr = tcg_temp_new_i32();
8132 tcg_gen_movi_i32(addr, s->pc);
8133 } else {
8134 addr = load_reg(s, rn);
8135 }
8136 tmp = load_reg(s, rm);
8137 tcg_gen_add_i32(addr, addr, tmp);
8138 if (insn & (1 << 4)) {
8139 /* tbh */
8140 tcg_gen_add_i32(addr, addr, tmp);
8141 tcg_temp_free_i32(tmp);
8142 tmp = tcg_temp_new_i32();
8143 tcg_gen_qemu_ld16u(tmp, addr, IS_USER(s));
8144 } else { /* tbb */
8145 tcg_temp_free_i32(tmp);
8146 tmp = tcg_temp_new_i32();
8147 tcg_gen_qemu_ld8u(tmp, addr, IS_USER(s));
8148 }
8149 tcg_temp_free_i32(addr);
8150 tcg_gen_shli_i32(tmp, tmp, 1);
8151 tcg_gen_addi_i32(tmp, tmp, s->pc);
8152 store_reg(s, 15, tmp);
8153 } else {
8154 /* Load/store exclusive byte/halfword/doubleword. */
8155 ARCH(7);
8156 op = (insn >> 4) & 0x3;
8157 if (op == 2) {
8158 goto illegal_op;
8159 }
8160 addr = tcg_temp_local_new_i32();
8161 load_reg_var(s, addr, rn);
8162 if (insn & (1 << 20)) {
8163 gen_load_exclusive(s, rs, rd, addr, op);
8164 } else {
8165 gen_store_exclusive(s, rm, rs, rd, addr, op);
8166 }
8167 tcg_temp_free_i32(addr);
8168 }
8169 } else {
8170 /* Load/store multiple, RFE, SRS. */
8171 if (((insn >> 23) & 1) == ((insn >> 24) & 1)) {
8172 /* RFE, SRS: not available in user mode or on M profile */
8173 if (IS_USER(s) || IS_M(env)) {
8174 goto illegal_op;
8175 }
8176 if (insn & (1 << 20)) {
8177 /* rfe */
8178 addr = load_reg(s, rn);
8179 if ((insn & (1 << 24)) == 0)
8180 tcg_gen_addi_i32(addr, addr, -8);
8181 /* Load PC into tmp and CPSR into tmp2. */
8182 tmp = tcg_temp_new_i32();
8183 tcg_gen_qemu_ld32u(tmp, addr, 0);
8184 tcg_gen_addi_i32(addr, addr, 4);
8185 tmp2 = tcg_temp_new_i32();
8186 tcg_gen_qemu_ld32u(tmp2, addr, 0);
8187 if (insn & (1 << 21)) {
8188 /* Base writeback. */
8189 if (insn & (1 << 24)) {
8190 tcg_gen_addi_i32(addr, addr, 4);
8191 } else {
8192 tcg_gen_addi_i32(addr, addr, -4);
8193 }
8194 store_reg(s, rn, addr);
8195 } else {
8196 tcg_temp_free_i32(addr);
8197 }
8198 gen_rfe(s, tmp, tmp2);
8199 } else {
8200 /* srs */
8201 gen_srs(s, (insn & 0x1f), (insn & (1 << 24)) ? 1 : 2,
8202 insn & (1 << 21));
8203 }
8204 } else {
8205 int i, loaded_base = 0;
8206 TCGv_i32 loaded_var;
8207 /* Load/store multiple. */
8208 addr = load_reg(s, rn);
8209 offset = 0;
8210 for (i = 0; i < 16; i++) {
8211 if (insn & (1 << i))
8212 offset += 4;
8213 }
8214 if (insn & (1 << 24)) {
8215 tcg_gen_addi_i32(addr, addr, -offset);
8216 }
8217
8218 TCGV_UNUSED_I32(loaded_var);
8219 for (i = 0; i < 16; i++) {
8220 if ((insn & (1 << i)) == 0)
8221 continue;
8222 if (insn & (1 << 20)) {
8223 /* Load. */
8224 tmp = tcg_temp_new_i32();
8225 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
8226 if (i == 15) {
8227 gen_bx(s, tmp);
8228 } else if (i == rn) {
8229 loaded_var = tmp;
8230 loaded_base = 1;
8231 } else {
8232 store_reg(s, i, tmp);
8233 }
8234 } else {
8235 /* Store. */
8236 tmp = load_reg(s, i);
8237 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
8238 tcg_temp_free_i32(tmp);
8239 }
8240 tcg_gen_addi_i32(addr, addr, 4);
8241 }
8242 if (loaded_base) {
8243 store_reg(s, rn, loaded_var);
8244 }
8245 if (insn & (1 << 21)) {
8246 /* Base register writeback. */
8247 if (insn & (1 << 24)) {
8248 tcg_gen_addi_i32(addr, addr, -offset);
8249 }
8250 /* Fault if writeback register is in register list. */
8251 if (insn & (1 << rn))
8252 goto illegal_op;
8253 store_reg(s, rn, addr);
8254 } else {
8255 tcg_temp_free_i32(addr);
8256 }
8257 }
8258 }
8259 break;
8260 case 5:
8261
8262 op = (insn >> 21) & 0xf;
8263 if (op == 6) {
8264 /* Halfword pack. */
8265 tmp = load_reg(s, rn);
8266 tmp2 = load_reg(s, rm);
8267 shift = ((insn >> 10) & 0x1c) | ((insn >> 6) & 0x3);
8268 if (insn & (1 << 5)) {
8269 /* pkhtb */
8270 if (shift == 0)
8271 shift = 31;
8272 tcg_gen_sari_i32(tmp2, tmp2, shift);
8273 tcg_gen_andi_i32(tmp, tmp, 0xffff0000);
8274 tcg_gen_ext16u_i32(tmp2, tmp2);
8275 } else {
8276 /* pkhbt */
8277 if (shift)
8278 tcg_gen_shli_i32(tmp2, tmp2, shift);
8279 tcg_gen_ext16u_i32(tmp, tmp);
8280 tcg_gen_andi_i32(tmp2, tmp2, 0xffff0000);
8281 }
8282 tcg_gen_or_i32(tmp, tmp, tmp2);
8283 tcg_temp_free_i32(tmp2);
8284 store_reg(s, rd, tmp);
8285 } else {
8286 /* Data processing register constant shift. */
8287 if (rn == 15) {
8288 tmp = tcg_temp_new_i32();
8289 tcg_gen_movi_i32(tmp, 0);
8290 } else {
8291 tmp = load_reg(s, rn);
8292 }
8293 tmp2 = load_reg(s, rm);
8294
8295 shiftop = (insn >> 4) & 3;
8296 shift = ((insn >> 6) & 3) | ((insn >> 10) & 0x1c);
8297 conds = (insn & (1 << 20)) != 0;
8298 logic_cc = (conds && thumb2_logic_op(op));
8299 gen_arm_shift_im(tmp2, shiftop, shift, logic_cc);
8300 if (gen_thumb2_data_op(s, op, conds, 0, tmp, tmp2))
8301 goto illegal_op;
8302 tcg_temp_free_i32(tmp2);
8303 if (rd != 15) {
8304 store_reg(s, rd, tmp);
8305 } else {
8306 tcg_temp_free_i32(tmp);
8307 }
8308 }
8309 break;
8310 case 13: /* Misc data processing. */
8311 op = ((insn >> 22) & 6) | ((insn >> 7) & 1);
8312 if (op < 4 && (insn & 0xf000) != 0xf000)
8313 goto illegal_op;
8314 switch (op) {
8315 case 0: /* Register controlled shift. */
8316 tmp = load_reg(s, rn);
8317 tmp2 = load_reg(s, rm);
8318 if ((insn & 0x70) != 0)
8319 goto illegal_op;
8320 op = (insn >> 21) & 3;
8321 logic_cc = (insn & (1 << 20)) != 0;
8322 gen_arm_shift_reg(tmp, op, tmp2, logic_cc);
8323 if (logic_cc)
8324 gen_logic_CC(tmp);
8325 store_reg_bx(env, s, rd, tmp);
8326 break;
8327 case 1: /* Sign/zero extend. */
8328 tmp = load_reg(s, rm);
8329 shift = (insn >> 4) & 3;
8330 /* ??? In many cases it's not necessary to do a
8331 rotate, a shift is sufficient. */
8332 if (shift != 0)
8333 tcg_gen_rotri_i32(tmp, tmp, shift * 8);
8334 op = (insn >> 20) & 7;
8335 switch (op) {
8336 case 0: gen_sxth(tmp); break;
8337 case 1: gen_uxth(tmp); break;
8338 case 2: gen_sxtb16(tmp); break;
8339 case 3: gen_uxtb16(tmp); break;
8340 case 4: gen_sxtb(tmp); break;
8341 case 5: gen_uxtb(tmp); break;
8342 default: goto illegal_op;
8343 }
8344 if (rn != 15) {
8345 tmp2 = load_reg(s, rn);
8346 if ((op >> 1) == 1) {
8347 gen_add16(tmp, tmp2);
8348 } else {
8349 tcg_gen_add_i32(tmp, tmp, tmp2);
8350 tcg_temp_free_i32(tmp2);
8351 }
8352 }
8353 store_reg(s, rd, tmp);
8354 break;
8355 case 2: /* SIMD add/subtract. */
8356 op = (insn >> 20) & 7;
8357 shift = (insn >> 4) & 7;
8358 if ((op & 3) == 3 || (shift & 3) == 3)
8359 goto illegal_op;
8360 tmp = load_reg(s, rn);
8361 tmp2 = load_reg(s, rm);
8362 gen_thumb2_parallel_addsub(op, shift, tmp, tmp2);
8363 tcg_temp_free_i32(tmp2);
8364 store_reg(s, rd, tmp);
8365 break;
8366 case 3: /* Other data processing. */
8367 op = ((insn >> 17) & 0x38) | ((insn >> 4) & 7);
8368 if (op < 4) {
8369 /* Saturating add/subtract. */
8370 tmp = load_reg(s, rn);
8371 tmp2 = load_reg(s, rm);
8372 if (op & 1)
8373 gen_helper_double_saturate(tmp, cpu_env, tmp);
8374 if (op & 2)
8375 gen_helper_sub_saturate(tmp, cpu_env, tmp2, tmp);
8376 else
8377 gen_helper_add_saturate(tmp, cpu_env, tmp, tmp2);
8378 tcg_temp_free_i32(tmp2);
8379 } else {
8380 tmp = load_reg(s, rn);
8381 switch (op) {
8382 case 0x0a: /* rbit */
8383 gen_helper_rbit(tmp, tmp);
8384 break;
8385 case 0x08: /* rev */
8386 tcg_gen_bswap32_i32(tmp, tmp);
8387 break;
8388 case 0x09: /* rev16 */
8389 gen_rev16(tmp);
8390 break;
8391 case 0x0b: /* revsh */
8392 gen_revsh(tmp);
8393 break;
8394 case 0x10: /* sel */
8395 tmp2 = load_reg(s, rm);
8396 tmp3 = tcg_temp_new_i32();
8397 tcg_gen_ld_i32(tmp3, cpu_env, offsetof(CPUARMState, GE));
8398 gen_helper_sel_flags(tmp, tmp3, tmp, tmp2);
8399 tcg_temp_free_i32(tmp3);
8400 tcg_temp_free_i32(tmp2);
8401 break;
8402 case 0x18: /* clz */
8403 gen_helper_clz(tmp, tmp);
8404 break;
8405 default:
8406 goto illegal_op;
8407 }
8408 }
8409 store_reg(s, rd, tmp);
8410 break;
8411 case 4: case 5: /* 32-bit multiply. Sum of absolute differences. */
8412 op = (insn >> 4) & 0xf;
8413 tmp = load_reg(s, rn);
8414 tmp2 = load_reg(s, rm);
8415 switch ((insn >> 20) & 7) {
8416 case 0: /* 32 x 32 -> 32 */
8417 tcg_gen_mul_i32(tmp, tmp, tmp2);
8418 tcg_temp_free_i32(tmp2);
8419 if (rs != 15) {
8420 tmp2 = load_reg(s, rs);
8421 if (op)
8422 tcg_gen_sub_i32(tmp, tmp2, tmp);
8423 else
8424 tcg_gen_add_i32(tmp, tmp, tmp2);
8425 tcg_temp_free_i32(tmp2);
8426 }
8427 break;
8428 case 1: /* 16 x 16 -> 32 */
8429 gen_mulxy(tmp, tmp2, op & 2, op & 1);
8430 tcg_temp_free_i32(tmp2);
8431 if (rs != 15) {
8432 tmp2 = load_reg(s, rs);
8433 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
8434 tcg_temp_free_i32(tmp2);
8435 }
8436 break;
8437 case 2: /* Dual multiply add. */
8438 case 4: /* Dual multiply subtract. */
8439 if (op)
8440 gen_swap_half(tmp2);
8441 gen_smul_dual(tmp, tmp2);
8442 if (insn & (1 << 22)) {
8443 /* This subtraction cannot overflow. */
8444 tcg_gen_sub_i32(tmp, tmp, tmp2);
8445 } else {
8446 /* This addition cannot overflow 32 bits;
8447 * however it may overflow considered as a signed
8448 * operation, in which case we must set the Q flag.
8449 */
8450 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
8451 }
8452 tcg_temp_free_i32(tmp2);
8453 if (rs != 15)
8454 {
8455 tmp2 = load_reg(s, rs);
8456 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
8457 tcg_temp_free_i32(tmp2);
8458 }
8459 break;
8460 case 3: /* 32 * 16 -> 32msb */
8461 if (op)
8462 tcg_gen_sari_i32(tmp2, tmp2, 16);
8463 else
8464 gen_sxth(tmp2);
8465 tmp64 = gen_muls_i64_i32(tmp, tmp2);
8466 tcg_gen_shri_i64(tmp64, tmp64, 16);
8467 tmp = tcg_temp_new_i32();
8468 tcg_gen_trunc_i64_i32(tmp, tmp64);
8469 tcg_temp_free_i64(tmp64);
8470 if (rs != 15)
8471 {
8472 tmp2 = load_reg(s, rs);
8473 gen_helper_add_setq(tmp, cpu_env, tmp, tmp2);
8474 tcg_temp_free_i32(tmp2);
8475 }
8476 break;
8477 case 5: case 6: /* 32 * 32 -> 32msb (SMMUL, SMMLA, SMMLS) */
8478 tmp64 = gen_muls_i64_i32(tmp, tmp2);
8479 if (rs != 15) {
8480 tmp = load_reg(s, rs);
8481 if (insn & (1 << 20)) {
8482 tmp64 = gen_addq_msw(tmp64, tmp);
8483 } else {
8484 tmp64 = gen_subq_msw(tmp64, tmp);
8485 }
8486 }
8487 if (insn & (1 << 4)) {
8488 tcg_gen_addi_i64(tmp64, tmp64, 0x80000000u);
8489 }
8490 tcg_gen_shri_i64(tmp64, tmp64, 32);
8491 tmp = tcg_temp_new_i32();
8492 tcg_gen_trunc_i64_i32(tmp, tmp64);
8493 tcg_temp_free_i64(tmp64);
8494 break;
8495 case 7: /* Unsigned sum of absolute differences. */
8496 gen_helper_usad8(tmp, tmp, tmp2);
8497 tcg_temp_free_i32(tmp2);
8498 if (rs != 15) {
8499 tmp2 = load_reg(s, rs);
8500 tcg_gen_add_i32(tmp, tmp, tmp2);
8501 tcg_temp_free_i32(tmp2);
8502 }
8503 break;
8504 }
8505 store_reg(s, rd, tmp);
8506 break;
8507 case 6: case 7: /* 64-bit multiply, Divide. */
8508 op = ((insn >> 4) & 0xf) | ((insn >> 16) & 0x70);
8509 tmp = load_reg(s, rn);
8510 tmp2 = load_reg(s, rm);
8511 if ((op & 0x50) == 0x10) {
8512 /* sdiv, udiv */
8513 if (!arm_feature(env, ARM_FEATURE_THUMB_DIV)) {
8514 goto illegal_op;
8515 }
8516 if (op & 0x20)
8517 gen_helper_udiv(tmp, tmp, tmp2);
8518 else
8519 gen_helper_sdiv(tmp, tmp, tmp2);
8520 tcg_temp_free_i32(tmp2);
8521 store_reg(s, rd, tmp);
8522 } else if ((op & 0xe) == 0xc) {
8523 /* Dual multiply accumulate long. */
8524 if (op & 1)
8525 gen_swap_half(tmp2);
8526 gen_smul_dual(tmp, tmp2);
8527 if (op & 0x10) {
8528 tcg_gen_sub_i32(tmp, tmp, tmp2);
8529 } else {
8530 tcg_gen_add_i32(tmp, tmp, tmp2);
8531 }
8532 tcg_temp_free_i32(tmp2);
8533 /* BUGFIX */
8534 tmp64 = tcg_temp_new_i64();
8535 tcg_gen_ext_i32_i64(tmp64, tmp);
8536 tcg_temp_free_i32(tmp);
8537 gen_addq(s, tmp64, rs, rd);
8538 gen_storeq_reg(s, rs, rd, tmp64);
8539 tcg_temp_free_i64(tmp64);
8540 } else {
8541 if (op & 0x20) {
8542 /* Unsigned 64-bit multiply */
8543 tmp64 = gen_mulu_i64_i32(tmp, tmp2);
8544 } else {
8545 if (op & 8) {
8546 /* smlalxy */
8547 gen_mulxy(tmp, tmp2, op & 2, op & 1);
8548 tcg_temp_free_i32(tmp2);
8549 tmp64 = tcg_temp_new_i64();
8550 tcg_gen_ext_i32_i64(tmp64, tmp);
8551 tcg_temp_free_i32(tmp);
8552 } else {
8553 /* Signed 64-bit multiply */
8554 tmp64 = gen_muls_i64_i32(tmp, tmp2);
8555 }
8556 }
8557 if (op & 4) {
8558 /* umaal */
8559 gen_addq_lo(s, tmp64, rs);
8560 gen_addq_lo(s, tmp64, rd);
8561 } else if (op & 0x40) {
8562 /* 64-bit accumulate. */
8563 gen_addq(s, tmp64, rs, rd);
8564 }
8565 gen_storeq_reg(s, rs, rd, tmp64);
8566 tcg_temp_free_i64(tmp64);
8567 }
8568 break;
8569 }
8570 break;
8571 case 6: case 7: case 14: case 15:
8572 /* Coprocessor. */
8573 if (((insn >> 24) & 3) == 3) {
8574 /* Translate into the equivalent ARM encoding. */
8575 insn = (insn & 0xe2ffffff) | ((insn & (1 << 28)) >> 4) | (1 << 28);
8576 if (disas_neon_data_insn(env, s, insn))
8577 goto illegal_op;
8578 } else {
8579 if (insn & (1 << 28))
8580 goto illegal_op;
8581 if (disas_coproc_insn (env, s, insn))
8582 goto illegal_op;
8583 }
8584 break;
8585 case 8: case 9: case 10: case 11:
8586 if (insn & (1 << 15)) {
8587 /* Branches, misc control. */
8588 if (insn & 0x5000) {
8589 /* Unconditional branch. */
8590 /* signextend(hw1[10:0]) -> offset[:12]. */
8591 offset = ((int32_t)insn << 5) >> 9 & ~(int32_t)0xfff;
8592 /* hw1[10:0] -> offset[11:1]. */
8593 offset |= (insn & 0x7ff) << 1;
8594 /* (~hw2[13, 11] ^ offset[24]) -> offset[23,22]
8595 offset[24:22] already have the same value because of the
8596 sign extension above. */
8597 offset ^= ((~insn) & (1 << 13)) << 10;
8598 offset ^= ((~insn) & (1 << 11)) << 11;
8599
8600 if (insn & (1 << 14)) {
8601 /* Branch and link. */
8602 tcg_gen_movi_i32(cpu_R[14], s->pc | 1);
8603 }
8604
8605 offset += s->pc;
8606 if (insn & (1 << 12)) {
8607 /* b/bl */
8608 gen_jmp(s, offset);
8609 } else {
8610 /* blx */
8611 offset &= ~(uint32_t)2;
8612 /* thumb2 bx, no need to check */
8613 gen_bx_im(s, offset);
8614 }
8615 } else if (((insn >> 23) & 7) == 7) {
8616 /* Misc control */
8617 if (insn & (1 << 13))
8618 goto illegal_op;
8619
8620 if (insn & (1 << 26)) {
8621 /* Secure monitor call (v6Z) */
8622 goto illegal_op; /* not implemented. */
8623 } else {
8624 op = (insn >> 20) & 7;
8625 switch (op) {
8626 case 0: /* msr cpsr. */
8627 if (IS_M(env)) {
8628 tmp = load_reg(s, rn);
8629 addr = tcg_const_i32(insn & 0xff);
8630 gen_helper_v7m_msr(cpu_env, addr, tmp);
8631 tcg_temp_free_i32(addr);
8632 tcg_temp_free_i32(tmp);
8633 gen_lookup_tb(s);
8634 break;
8635 }
8636 /* fall through */
8637 case 1: /* msr spsr. */
8638 if (IS_M(env))
8639 goto illegal_op;
8640 tmp = load_reg(s, rn);
8641 if (gen_set_psr(s,
8642 msr_mask(env, s, (insn >> 8) & 0xf, op == 1),
8643 op == 1, tmp))
8644 goto illegal_op;
8645 break;
8646 case 2: /* cps, nop-hint. */
8647 if (((insn >> 8) & 7) == 0) {
8648 gen_nop_hint(s, insn & 0xff);
8649 }
8650 /* Implemented as NOP in user mode. */
8651 if (IS_USER(s))
8652 break;
8653 offset = 0;
8654 imm = 0;
8655 if (insn & (1 << 10)) {
8656 if (insn & (1 << 7))
8657 offset |= CPSR_A;
8658 if (insn & (1 << 6))
8659 offset |= CPSR_I;
8660 if (insn & (1 << 5))
8661 offset |= CPSR_F;
8662 if (insn & (1 << 9))
8663 imm = CPSR_A | CPSR_I | CPSR_F;
8664 }
8665 if (insn & (1 << 8)) {
8666 offset |= 0x1f;
8667 imm |= (insn & 0x1f);
8668 }
8669 if (offset) {
8670 gen_set_psr_im(s, offset, 0, imm);
8671 }
8672 break;
8673 case 3: /* Special control operations. */
8674 ARCH(7);
8675 op = (insn >> 4) & 0xf;
8676 switch (op) {
8677 case 2: /* clrex */
8678 gen_clrex(s);
8679 break;
8680 case 4: /* dsb */
8681 case 5: /* dmb */
8682 case 6: /* isb */
8683 /* These execute as NOPs. */
8684 break;
8685 default:
8686 goto illegal_op;
8687 }
8688 break;
8689 case 4: /* bxj */
8690 /* Trivial implementation equivalent to bx. */
8691 tmp = load_reg(s, rn);
8692 gen_bx(s, tmp);
8693 break;
8694 case 5: /* Exception return. */
8695 if (IS_USER(s)) {
8696 goto illegal_op;
8697 }
8698 if (rn != 14 || rd != 15) {
8699 goto illegal_op;
8700 }
8701 tmp = load_reg(s, rn);
8702 tcg_gen_subi_i32(tmp, tmp, insn & 0xff);
8703 gen_exception_return(s, tmp);
8704 break;
8705 case 6: /* mrs cpsr. */
8706 tmp = tcg_temp_new_i32();
8707 if (IS_M(env)) {
8708 addr = tcg_const_i32(insn & 0xff);
8709 gen_helper_v7m_mrs(tmp, cpu_env, addr);
8710 tcg_temp_free_i32(addr);
8711 } else {
8712 gen_helper_cpsr_read(tmp, cpu_env);
8713 }
8714 store_reg(s, rd, tmp);
8715 break;
8716 case 7: /* mrs spsr. */
8717 /* Not accessible in user mode. */
8718 if (IS_USER(s) || IS_M(env))
8719 goto illegal_op;
8720 tmp = load_cpu_field(spsr);
8721 store_reg(s, rd, tmp);
8722 break;
8723 }
8724 }
8725 } else {
8726 /* Conditional branch. */
8727 op = (insn >> 22) & 0xf;
8728 /* Generate a conditional jump to next instruction. */
8729 s->condlabel = gen_new_label();
8730 gen_test_cc(op ^ 1, s->condlabel);
8731 s->condjmp = 1;
8732
8733 /* offset[11:1] = insn[10:0] */
8734 offset = (insn & 0x7ff) << 1;
8735 /* offset[17:12] = insn[21:16]. */
8736 offset |= (insn & 0x003f0000) >> 4;
8737 /* offset[31:20] = insn[26]. */
8738 offset |= ((int32_t)((insn << 5) & 0x80000000)) >> 11;
8739 /* offset[18] = insn[13]. */
8740 offset |= (insn & (1 << 13)) << 5;
8741 /* offset[19] = insn[11]. */
8742 offset |= (insn & (1 << 11)) << 8;
8743
8744 /* jump to the offset */
8745 gen_jmp(s, s->pc + offset);
8746 }
8747 } else {
8748 /* Data processing immediate. */
8749 if (insn & (1 << 25)) {
8750 if (insn & (1 << 24)) {
8751 if (insn & (1 << 20))
8752 goto illegal_op;
8753 /* Bitfield/Saturate. */
8754 op = (insn >> 21) & 7;
8755 imm = insn & 0x1f;
8756 shift = ((insn >> 6) & 3) | ((insn >> 10) & 0x1c);
8757 if (rn == 15) {
8758 tmp = tcg_temp_new_i32();
8759 tcg_gen_movi_i32(tmp, 0);
8760 } else {
8761 tmp = load_reg(s, rn);
8762 }
8763 switch (op) {
8764 case 2: /* Signed bitfield extract. */
8765 imm++;
8766 if (shift + imm > 32)
8767 goto illegal_op;
8768 if (imm < 32)
8769 gen_sbfx(tmp, shift, imm);
8770 break;
8771 case 6: /* Unsigned bitfield extract. */
8772 imm++;
8773 if (shift + imm > 32)
8774 goto illegal_op;
8775 if (imm < 32)
8776 gen_ubfx(tmp, shift, (1u << imm) - 1);
8777 break;
8778 case 3: /* Bitfield insert/clear. */
8779 if (imm < shift)
8780 goto illegal_op;
8781 imm = imm + 1 - shift;
8782 if (imm != 32) {
8783 tmp2 = load_reg(s, rd);
8784 tcg_gen_deposit_i32(tmp, tmp2, tmp, shift, imm);
8785 tcg_temp_free_i32(tmp2);
8786 }
8787 break;
8788 case 7:
8789 goto illegal_op;
8790 default: /* Saturate. */
8791 if (shift) {
8792 if (op & 1)
8793 tcg_gen_sari_i32(tmp, tmp, shift);
8794 else
8795 tcg_gen_shli_i32(tmp, tmp, shift);
8796 }
8797 tmp2 = tcg_const_i32(imm);
8798 if (op & 4) {
8799 /* Unsigned. */
8800 if ((op & 1) && shift == 0)
8801 gen_helper_usat16(tmp, cpu_env, tmp, tmp2);
8802 else
8803 gen_helper_usat(tmp, cpu_env, tmp, tmp2);
8804 } else {
8805 /* Signed. */
8806 if ((op & 1) && shift == 0)
8807 gen_helper_ssat16(tmp, cpu_env, tmp, tmp2);
8808 else
8809 gen_helper_ssat(tmp, cpu_env, tmp, tmp2);
8810 }
8811 tcg_temp_free_i32(tmp2);
8812 break;
8813 }
8814 store_reg(s, rd, tmp);
8815 } else {
8816 imm = ((insn & 0x04000000) >> 15)
8817 | ((insn & 0x7000) >> 4) | (insn & 0xff);
8818 if (insn & (1 << 22)) {
8819 /* 16-bit immediate. */
8820 imm |= (insn >> 4) & 0xf000;
8821 if (insn & (1 << 23)) {
8822 /* movt */
8823 tmp = load_reg(s, rd);
8824 tcg_gen_ext16u_i32(tmp, tmp);
8825 tcg_gen_ori_i32(tmp, tmp, imm << 16);
8826 } else {
8827 /* movw */
8828 tmp = tcg_temp_new_i32();
8829 tcg_gen_movi_i32(tmp, imm);
8830 }
8831 } else {
8832 /* Add/sub 12-bit immediate. */
8833 if (rn == 15) {
8834 offset = s->pc & ~(uint32_t)3;
8835 if (insn & (1 << 23))
8836 offset -= imm;
8837 else
8838 offset += imm;
8839 tmp = tcg_temp_new_i32();
8840 tcg_gen_movi_i32(tmp, offset);
8841 } else {
8842 tmp = load_reg(s, rn);
8843 if (insn & (1 << 23))
8844 tcg_gen_subi_i32(tmp, tmp, imm);
8845 else
8846 tcg_gen_addi_i32(tmp, tmp, imm);
8847 }
8848 }
8849 store_reg(s, rd, tmp);
8850 }
8851 } else {
8852 int shifter_out = 0;
8853 /* modified 12-bit immediate. */
8854 shift = ((insn & 0x04000000) >> 23) | ((insn & 0x7000) >> 12);
8855 imm = (insn & 0xff);
8856 switch (shift) {
8857 case 0: /* XY */
8858 /* Nothing to do. */
8859 break;
8860 case 1: /* 00XY00XY */
8861 imm |= imm << 16;
8862 break;
8863 case 2: /* XY00XY00 */
8864 imm |= imm << 16;
8865 imm <<= 8;
8866 break;
8867 case 3: /* XYXYXYXY */
8868 imm |= imm << 16;
8869 imm |= imm << 8;
8870 break;
8871 default: /* Rotated constant. */
8872 shift = (shift << 1) | (imm >> 7);
8873 imm |= 0x80;
8874 imm = imm << (32 - shift);
8875 shifter_out = 1;
8876 break;
8877 }
8878 tmp2 = tcg_temp_new_i32();
8879 tcg_gen_movi_i32(tmp2, imm);
8880 rn = (insn >> 16) & 0xf;
8881 if (rn == 15) {
8882 tmp = tcg_temp_new_i32();
8883 tcg_gen_movi_i32(tmp, 0);
8884 } else {
8885 tmp = load_reg(s, rn);
8886 }
8887 op = (insn >> 21) & 0xf;
8888 if (gen_thumb2_data_op(s, op, (insn & (1 << 20)) != 0,
8889 shifter_out, tmp, tmp2))
8890 goto illegal_op;
8891 tcg_temp_free_i32(tmp2);
8892 rd = (insn >> 8) & 0xf;
8893 if (rd != 15) {
8894 store_reg(s, rd, tmp);
8895 } else {
8896 tcg_temp_free_i32(tmp);
8897 }
8898 }
8899 }
8900 break;
8901 case 12: /* Load/store single data item. */
8902 {
8903 int postinc = 0;
8904 int writeback = 0;
8905 int user;
8906 if ((insn & 0x01100000) == 0x01000000) {
8907 if (disas_neon_ls_insn(env, s, insn))
8908 goto illegal_op;
8909 break;
8910 }
8911 op = ((insn >> 21) & 3) | ((insn >> 22) & 4);
8912 if (rs == 15) {
8913 if (!(insn & (1 << 20))) {
8914 goto illegal_op;
8915 }
8916 if (op != 2) {
8917 /* Byte or halfword load space with dest == r15 : memory hints.
8918 * Catch them early so we don't emit pointless addressing code.
8919 * This space is a mix of:
8920 * PLD/PLDW/PLI, which we implement as NOPs (note that unlike
8921 * the ARM encodings, PLDW space doesn't UNDEF for non-v7MP
8922 * cores)
8923 * unallocated hints, which must be treated as NOPs
8924 * UNPREDICTABLE space, which we NOP or UNDEF depending on
8925 * which is easiest for the decoding logic
8926 * Some space which must UNDEF
8927 */
8928 int op1 = (insn >> 23) & 3;
8929 int op2 = (insn >> 6) & 0x3f;
8930 if (op & 2) {
8931 goto illegal_op;
8932 }
8933 if (rn == 15) {
8934 /* UNPREDICTABLE, unallocated hint or
8935 * PLD/PLDW/PLI (literal)
8936 */
8937 return 0;
8938 }
8939 if (op1 & 1) {
8940 return 0; /* PLD/PLDW/PLI or unallocated hint */
8941 }
8942 if ((op2 == 0) || ((op2 & 0x3c) == 0x30)) {
8943 return 0; /* PLD/PLDW/PLI or unallocated hint */
8944 }
8945 /* UNDEF space, or an UNPREDICTABLE */
8946 return 1;
8947 }
8948 }
8949 user = IS_USER(s);
8950 if (rn == 15) {
8951 addr = tcg_temp_new_i32();
8952 /* PC relative. */
8953 /* s->pc has already been incremented by 4. */
8954 imm = s->pc & 0xfffffffc;
8955 if (insn & (1 << 23))
8956 imm += insn & 0xfff;
8957 else
8958 imm -= insn & 0xfff;
8959 tcg_gen_movi_i32(addr, imm);
8960 } else {
8961 addr = load_reg(s, rn);
8962 if (insn & (1 << 23)) {
8963 /* Positive offset. */
8964 imm = insn & 0xfff;
8965 tcg_gen_addi_i32(addr, addr, imm);
8966 } else {
8967 imm = insn & 0xff;
8968 switch ((insn >> 8) & 0xf) {
8969 case 0x0: /* Shifted Register. */
8970 shift = (insn >> 4) & 0xf;
8971 if (shift > 3) {
8972 tcg_temp_free_i32(addr);
8973 goto illegal_op;
8974 }
8975 tmp = load_reg(s, rm);
8976 if (shift)
8977 tcg_gen_shli_i32(tmp, tmp, shift);
8978 tcg_gen_add_i32(addr, addr, tmp);
8979 tcg_temp_free_i32(tmp);
8980 break;
8981 case 0xc: /* Negative offset. */
8982 tcg_gen_addi_i32(addr, addr, -imm);
8983 break;
8984 case 0xe: /* User privilege. */
8985 tcg_gen_addi_i32(addr, addr, imm);
8986 user = 1;
8987 break;
8988 case 0x9: /* Post-decrement. */
8989 imm = -imm;
8990 /* Fall through. */
8991 case 0xb: /* Post-increment. */
8992 postinc = 1;
8993 writeback = 1;
8994 break;
8995 case 0xd: /* Pre-decrement. */
8996 imm = -imm;
8997 /* Fall through. */
8998 case 0xf: /* Pre-increment. */
8999 tcg_gen_addi_i32(addr, addr, imm);
9000 writeback = 1;
9001 break;
9002 default:
9003 tcg_temp_free_i32(addr);
9004 goto illegal_op;
9005 }
9006 }
9007 }
9008 if (insn & (1 << 20)) {
9009 /* Load. */
9010 tmp = tcg_temp_new_i32();
9011 switch (op) {
9012 case 0:
9013 tcg_gen_qemu_ld8u(tmp, addr, user);
9014 break;
9015 case 4:
9016 tcg_gen_qemu_ld8s(tmp, addr, user);
9017 break;
9018 case 1:
9019 tcg_gen_qemu_ld16u(tmp, addr, user);
9020 break;
9021 case 5:
9022 tcg_gen_qemu_ld16s(tmp, addr, user);
9023 break;
9024 case 2:
9025 tcg_gen_qemu_ld32u(tmp, addr, user);
9026 break;
9027 default:
9028 tcg_temp_free_i32(tmp);
9029 tcg_temp_free_i32(addr);
9030 goto illegal_op;
9031 }
9032 if (rs == 15) {
9033 gen_bx(s, tmp);
9034 } else {
9035 store_reg(s, rs, tmp);
9036 }
9037 } else {
9038 /* Store. */
9039 tmp = load_reg(s, rs);
9040 switch (op) {
9041 case 0:
9042 tcg_gen_qemu_st8(tmp, addr, user);
9043 break;
9044 case 1:
9045 tcg_gen_qemu_st16(tmp, addr, user);
9046 break;
9047 case 2:
9048 tcg_gen_qemu_st32(tmp, addr, user);
9049 break;
9050 default:
9051 tcg_temp_free_i32(tmp);
9052 tcg_temp_free_i32(addr);
9053 goto illegal_op;
9054 }
9055 tcg_temp_free_i32(tmp);
9056 }
9057 if (postinc)
9058 tcg_gen_addi_i32(addr, addr, imm);
9059 if (writeback) {
9060 store_reg(s, rn, addr);
9061 } else {
9062 tcg_temp_free_i32(addr);
9063 }
9064 }
9065 break;
9066 default:
9067 goto illegal_op;
9068 }
9069 return 0;
9070 illegal_op:
9071 return 1;
9072 }
9073
9074 static void disas_thumb_insn(CPUARMState *env, DisasContext *s)
9075 {
9076 uint32_t val, insn, op, rm, rn, rd, shift, cond;
9077 int32_t offset;
9078 int i;
9079 TCGv_i32 tmp;
9080 TCGv_i32 tmp2;
9081 TCGv_i32 addr;
9082
9083 if (s->condexec_mask) {
9084 cond = s->condexec_cond;
9085 if (cond != 0x0e) { /* Skip conditional when condition is AL. */
9086 s->condlabel = gen_new_label();
9087 gen_test_cc(cond ^ 1, s->condlabel);
9088 s->condjmp = 1;
9089 }
9090 }
9091
9092 insn = arm_lduw_code(env, s->pc, s->bswap_code);
9093 s->pc += 2;
9094
9095 switch (insn >> 12) {
9096 case 0: case 1:
9097
9098 rd = insn & 7;
9099 op = (insn >> 11) & 3;
9100 if (op == 3) {
9101 /* add/subtract */
9102 rn = (insn >> 3) & 7;
9103 tmp = load_reg(s, rn);
9104 if (insn & (1 << 10)) {
9105 /* immediate */
9106 tmp2 = tcg_temp_new_i32();
9107 tcg_gen_movi_i32(tmp2, (insn >> 6) & 7);
9108 } else {
9109 /* reg */
9110 rm = (insn >> 6) & 7;
9111 tmp2 = load_reg(s, rm);
9112 }
9113 if (insn & (1 << 9)) {
9114 if (s->condexec_mask)
9115 tcg_gen_sub_i32(tmp, tmp, tmp2);
9116 else
9117 gen_sub_CC(tmp, tmp, tmp2);
9118 } else {
9119 if (s->condexec_mask)
9120 tcg_gen_add_i32(tmp, tmp, tmp2);
9121 else
9122 gen_add_CC(tmp, tmp, tmp2);
9123 }
9124 tcg_temp_free_i32(tmp2);
9125 store_reg(s, rd, tmp);
9126 } else {
9127 /* shift immediate */
9128 rm = (insn >> 3) & 7;
9129 shift = (insn >> 6) & 0x1f;
9130 tmp = load_reg(s, rm);
9131 gen_arm_shift_im(tmp, op, shift, s->condexec_mask == 0);
9132 if (!s->condexec_mask)
9133 gen_logic_CC(tmp);
9134 store_reg(s, rd, tmp);
9135 }
9136 break;
9137 case 2: case 3:
9138 /* arithmetic large immediate */
9139 op = (insn >> 11) & 3;
9140 rd = (insn >> 8) & 0x7;
9141 if (op == 0) { /* mov */
9142 tmp = tcg_temp_new_i32();
9143 tcg_gen_movi_i32(tmp, insn & 0xff);
9144 if (!s->condexec_mask)
9145 gen_logic_CC(tmp);
9146 store_reg(s, rd, tmp);
9147 } else {
9148 tmp = load_reg(s, rd);
9149 tmp2 = tcg_temp_new_i32();
9150 tcg_gen_movi_i32(tmp2, insn & 0xff);
9151 switch (op) {
9152 case 1: /* cmp */
9153 gen_sub_CC(tmp, tmp, tmp2);
9154 tcg_temp_free_i32(tmp);
9155 tcg_temp_free_i32(tmp2);
9156 break;
9157 case 2: /* add */
9158 if (s->condexec_mask)
9159 tcg_gen_add_i32(tmp, tmp, tmp2);
9160 else
9161 gen_add_CC(tmp, tmp, tmp2);
9162 tcg_temp_free_i32(tmp2);
9163 store_reg(s, rd, tmp);
9164 break;
9165 case 3: /* sub */
9166 if (s->condexec_mask)
9167 tcg_gen_sub_i32(tmp, tmp, tmp2);
9168 else
9169 gen_sub_CC(tmp, tmp, tmp2);
9170 tcg_temp_free_i32(tmp2);
9171 store_reg(s, rd, tmp);
9172 break;
9173 }
9174 }
9175 break;
9176 case 4:
9177 if (insn & (1 << 11)) {
9178 rd = (insn >> 8) & 7;
9179 /* load pc-relative. Bit 1 of PC is ignored. */
9180 val = s->pc + 2 + ((insn & 0xff) * 4);
9181 val &= ~(uint32_t)2;
9182 addr = tcg_temp_new_i32();
9183 tcg_gen_movi_i32(addr, val);
9184 tmp = tcg_temp_new_i32();
9185 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
9186 tcg_temp_free_i32(addr);
9187 store_reg(s, rd, tmp);
9188 break;
9189 }
9190 if (insn & (1 << 10)) {
9191 /* data processing extended or blx */
9192 rd = (insn & 7) | ((insn >> 4) & 8);
9193 rm = (insn >> 3) & 0xf;
9194 op = (insn >> 8) & 3;
9195 switch (op) {
9196 case 0: /* add */
9197 tmp = load_reg(s, rd);
9198 tmp2 = load_reg(s, rm);
9199 tcg_gen_add_i32(tmp, tmp, tmp2);
9200 tcg_temp_free_i32(tmp2);
9201 store_reg(s, rd, tmp);
9202 break;
9203 case 1: /* cmp */
9204 tmp = load_reg(s, rd);
9205 tmp2 = load_reg(s, rm);
9206 gen_sub_CC(tmp, tmp, tmp2);
9207 tcg_temp_free_i32(tmp2);
9208 tcg_temp_free_i32(tmp);
9209 break;
9210 case 2: /* mov/cpy */
9211 tmp = load_reg(s, rm);
9212 store_reg(s, rd, tmp);
9213 break;
9214 case 3:/* branch [and link] exchange thumb register */
9215 tmp = load_reg(s, rm);
9216 if (insn & (1 << 7)) {
9217 ARCH(5);
9218 val = (uint32_t)s->pc | 1;
9219 tmp2 = tcg_temp_new_i32();
9220 tcg_gen_movi_i32(tmp2, val);
9221 store_reg(s, 14, tmp2);
9222 }
9223 /* already thumb, no need to check */
9224 gen_bx(s, tmp);
9225 break;
9226 }
9227 break;
9228 }
9229
9230 /* data processing register */
9231 rd = insn & 7;
9232 rm = (insn >> 3) & 7;
9233 op = (insn >> 6) & 0xf;
9234 if (op == 2 || op == 3 || op == 4 || op == 7) {
9235 /* the shift/rotate ops want the operands backwards */
9236 val = rm;
9237 rm = rd;
9238 rd = val;
9239 val = 1;
9240 } else {
9241 val = 0;
9242 }
9243
9244 if (op == 9) { /* neg */
9245 tmp = tcg_temp_new_i32();
9246 tcg_gen_movi_i32(tmp, 0);
9247 } else if (op != 0xf) { /* mvn doesn't read its first operand */
9248 tmp = load_reg(s, rd);
9249 } else {
9250 TCGV_UNUSED_I32(tmp);
9251 }
9252
9253 tmp2 = load_reg(s, rm);
9254 switch (op) {
9255 case 0x0: /* and */
9256 tcg_gen_and_i32(tmp, tmp, tmp2);
9257 if (!s->condexec_mask)
9258 gen_logic_CC(tmp);
9259 break;
9260 case 0x1: /* eor */
9261 tcg_gen_xor_i32(tmp, tmp, tmp2);
9262 if (!s->condexec_mask)
9263 gen_logic_CC(tmp);
9264 break;
9265 case 0x2: /* lsl */
9266 if (s->condexec_mask) {
9267 gen_shl(tmp2, tmp2, tmp);
9268 } else {
9269 gen_helper_shl_cc(tmp2, cpu_env, tmp2, tmp);
9270 gen_logic_CC(tmp2);
9271 }
9272 break;
9273 case 0x3: /* lsr */
9274 if (s->condexec_mask) {
9275 gen_shr(tmp2, tmp2, tmp);
9276 } else {
9277 gen_helper_shr_cc(tmp2, cpu_env, tmp2, tmp);
9278 gen_logic_CC(tmp2);
9279 }
9280 break;
9281 case 0x4: /* asr */
9282 if (s->condexec_mask) {
9283 gen_sar(tmp2, tmp2, tmp);
9284 } else {
9285 gen_helper_sar_cc(tmp2, cpu_env, tmp2, tmp);
9286 gen_logic_CC(tmp2);
9287 }
9288 break;
9289 case 0x5: /* adc */
9290 if (s->condexec_mask) {
9291 gen_adc(tmp, tmp2);
9292 } else {
9293 gen_adc_CC(tmp, tmp, tmp2);
9294 }
9295 break;
9296 case 0x6: /* sbc */
9297 if (s->condexec_mask) {
9298 gen_sub_carry(tmp, tmp, tmp2);
9299 } else {
9300 gen_sbc_CC(tmp, tmp, tmp2);
9301 }
9302 break;
9303 case 0x7: /* ror */
9304 if (s->condexec_mask) {
9305 tcg_gen_andi_i32(tmp, tmp, 0x1f);
9306 tcg_gen_rotr_i32(tmp2, tmp2, tmp);
9307 } else {
9308 gen_helper_ror_cc(tmp2, cpu_env, tmp2, tmp);
9309 gen_logic_CC(tmp2);
9310 }
9311 break;
9312 case 0x8: /* tst */
9313 tcg_gen_and_i32(tmp, tmp, tmp2);
9314 gen_logic_CC(tmp);
9315 rd = 16;
9316 break;
9317 case 0x9: /* neg */
9318 if (s->condexec_mask)
9319 tcg_gen_neg_i32(tmp, tmp2);
9320 else
9321 gen_sub_CC(tmp, tmp, tmp2);
9322 break;
9323 case 0xa: /* cmp */
9324 gen_sub_CC(tmp, tmp, tmp2);
9325 rd = 16;
9326 break;
9327 case 0xb: /* cmn */
9328 gen_add_CC(tmp, tmp, tmp2);
9329 rd = 16;
9330 break;
9331 case 0xc: /* orr */
9332 tcg_gen_or_i32(tmp, tmp, tmp2);
9333 if (!s->condexec_mask)
9334 gen_logic_CC(tmp);
9335 break;
9336 case 0xd: /* mul */
9337 tcg_gen_mul_i32(tmp, tmp, tmp2);
9338 if (!s->condexec_mask)
9339 gen_logic_CC(tmp);
9340 break;
9341 case 0xe: /* bic */
9342 tcg_gen_andc_i32(tmp, tmp, tmp2);
9343 if (!s->condexec_mask)
9344 gen_logic_CC(tmp);
9345 break;
9346 case 0xf: /* mvn */
9347 tcg_gen_not_i32(tmp2, tmp2);
9348 if (!s->condexec_mask)
9349 gen_logic_CC(tmp2);
9350 val = 1;
9351 rm = rd;
9352 break;
9353 }
9354 if (rd != 16) {
9355 if (val) {
9356 store_reg(s, rm, tmp2);
9357 if (op != 0xf)
9358 tcg_temp_free_i32(tmp);
9359 } else {
9360 store_reg(s, rd, tmp);
9361 tcg_temp_free_i32(tmp2);
9362 }
9363 } else {
9364 tcg_temp_free_i32(tmp);
9365 tcg_temp_free_i32(tmp2);
9366 }
9367 break;
9368
9369 case 5:
9370 /* load/store register offset. */
9371 rd = insn & 7;
9372 rn = (insn >> 3) & 7;
9373 rm = (insn >> 6) & 7;
9374 op = (insn >> 9) & 7;
9375 addr = load_reg(s, rn);
9376 tmp = load_reg(s, rm);
9377 tcg_gen_add_i32(addr, addr, tmp);
9378 tcg_temp_free_i32(tmp);
9379
9380 if (op < 3) { /* store */
9381 tmp = load_reg(s, rd);
9382 } else {
9383 tmp = tcg_temp_new_i32();
9384 }
9385
9386 switch (op) {
9387 case 0: /* str */
9388 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
9389 break;
9390 case 1: /* strh */
9391 tcg_gen_qemu_st16(tmp, addr, IS_USER(s));
9392 break;
9393 case 2: /* strb */
9394 tcg_gen_qemu_st8(tmp, addr, IS_USER(s));
9395 break;
9396 case 3: /* ldrsb */
9397 tcg_gen_qemu_ld8s(tmp, addr, IS_USER(s));
9398 break;
9399 case 4: /* ldr */
9400 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
9401 break;
9402 case 5: /* ldrh */
9403 tcg_gen_qemu_ld16u(tmp, addr, IS_USER(s));
9404 break;
9405 case 6: /* ldrb */
9406 tcg_gen_qemu_ld8u(tmp, addr, IS_USER(s));
9407 break;
9408 case 7: /* ldrsh */
9409 tcg_gen_qemu_ld16s(tmp, addr, IS_USER(s));
9410 break;
9411 }
9412 if (op >= 3) { /* load */
9413 store_reg(s, rd, tmp);
9414 } else {
9415 tcg_temp_free_i32(tmp);
9416 }
9417 tcg_temp_free_i32(addr);
9418 break;
9419
9420 case 6:
9421 /* load/store word immediate offset */
9422 rd = insn & 7;
9423 rn = (insn >> 3) & 7;
9424 addr = load_reg(s, rn);
9425 val = (insn >> 4) & 0x7c;
9426 tcg_gen_addi_i32(addr, addr, val);
9427
9428 if (insn & (1 << 11)) {
9429 /* load */
9430 tmp = tcg_temp_new_i32();
9431 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
9432 store_reg(s, rd, tmp);
9433 } else {
9434 /* store */
9435 tmp = load_reg(s, rd);
9436 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
9437 tcg_temp_free_i32(tmp);
9438 }
9439 tcg_temp_free_i32(addr);
9440 break;
9441
9442 case 7:
9443 /* load/store byte immediate offset */
9444 rd = insn & 7;
9445 rn = (insn >> 3) & 7;
9446 addr = load_reg(s, rn);
9447 val = (insn >> 6) & 0x1f;
9448 tcg_gen_addi_i32(addr, addr, val);
9449
9450 if (insn & (1 << 11)) {
9451 /* load */
9452 tmp = tcg_temp_new_i32();
9453 tcg_gen_qemu_ld8u(tmp, addr, IS_USER(s));
9454 store_reg(s, rd, tmp);
9455 } else {
9456 /* store */
9457 tmp = load_reg(s, rd);
9458 tcg_gen_qemu_st8(tmp, addr, IS_USER(s));
9459 tcg_temp_free_i32(tmp);
9460 }
9461 tcg_temp_free_i32(addr);
9462 break;
9463
9464 case 8:
9465 /* load/store halfword immediate offset */
9466 rd = insn & 7;
9467 rn = (insn >> 3) & 7;
9468 addr = load_reg(s, rn);
9469 val = (insn >> 5) & 0x3e;
9470 tcg_gen_addi_i32(addr, addr, val);
9471
9472 if (insn & (1 << 11)) {
9473 /* load */
9474 tmp = tcg_temp_new_i32();
9475 tcg_gen_qemu_ld16u(tmp, addr, IS_USER(s));
9476 store_reg(s, rd, tmp);
9477 } else {
9478 /* store */
9479 tmp = load_reg(s, rd);
9480 tcg_gen_qemu_st16(tmp, addr, IS_USER(s));
9481 tcg_temp_free_i32(tmp);
9482 }
9483 tcg_temp_free_i32(addr);
9484 break;
9485
9486 case 9:
9487 /* load/store from stack */
9488 rd = (insn >> 8) & 7;
9489 addr = load_reg(s, 13);
9490 val = (insn & 0xff) * 4;
9491 tcg_gen_addi_i32(addr, addr, val);
9492
9493 if (insn & (1 << 11)) {
9494 /* load */
9495 tmp = tcg_temp_new_i32();
9496 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
9497 store_reg(s, rd, tmp);
9498 } else {
9499 /* store */
9500 tmp = load_reg(s, rd);
9501 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
9502 tcg_temp_free_i32(tmp);
9503 }
9504 tcg_temp_free_i32(addr);
9505 break;
9506
9507 case 10:
9508 /* add to high reg */
9509 rd = (insn >> 8) & 7;
9510 if (insn & (1 << 11)) {
9511 /* SP */
9512 tmp = load_reg(s, 13);
9513 } else {
9514 /* PC. bit 1 is ignored. */
9515 tmp = tcg_temp_new_i32();
9516 tcg_gen_movi_i32(tmp, (s->pc + 2) & ~(uint32_t)2);
9517 }
9518 val = (insn & 0xff) * 4;
9519 tcg_gen_addi_i32(tmp, tmp, val);
9520 store_reg(s, rd, tmp);
9521 break;
9522
9523 case 11:
9524 /* misc */
9525 op = (insn >> 8) & 0xf;
9526 switch (op) {
9527 case 0:
9528 /* adjust stack pointer */
9529 tmp = load_reg(s, 13);
9530 val = (insn & 0x7f) * 4;
9531 if (insn & (1 << 7))
9532 val = -(int32_t)val;
9533 tcg_gen_addi_i32(tmp, tmp, val);
9534 store_reg(s, 13, tmp);
9535 break;
9536
9537 case 2: /* sign/zero extend. */
9538 ARCH(6);
9539 rd = insn & 7;
9540 rm = (insn >> 3) & 7;
9541 tmp = load_reg(s, rm);
9542 switch ((insn >> 6) & 3) {
9543 case 0: gen_sxth(tmp); break;
9544 case 1: gen_sxtb(tmp); break;
9545 case 2: gen_uxth(tmp); break;
9546 case 3: gen_uxtb(tmp); break;
9547 }
9548 store_reg(s, rd, tmp);
9549 break;
9550 case 4: case 5: case 0xc: case 0xd:
9551 /* push/pop */
9552 addr = load_reg(s, 13);
9553 if (insn & (1 << 8))
9554 offset = 4;
9555 else
9556 offset = 0;
9557 for (i = 0; i < 8; i++) {
9558 if (insn & (1 << i))
9559 offset += 4;
9560 }
9561 if ((insn & (1 << 11)) == 0) {
9562 tcg_gen_addi_i32(addr, addr, -offset);
9563 }
9564 for (i = 0; i < 8; i++) {
9565 if (insn & (1 << i)) {
9566 if (insn & (1 << 11)) {
9567 /* pop */
9568 tmp = tcg_temp_new_i32();
9569 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
9570 store_reg(s, i, tmp);
9571 } else {
9572 /* push */
9573 tmp = load_reg(s, i);
9574 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
9575 tcg_temp_free_i32(tmp);
9576 }
9577 /* advance to the next address. */
9578 tcg_gen_addi_i32(addr, addr, 4);
9579 }
9580 }
9581 TCGV_UNUSED_I32(tmp);
9582 if (insn & (1 << 8)) {
9583 if (insn & (1 << 11)) {
9584 /* pop pc */
9585 tmp = tcg_temp_new_i32();
9586 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
9587 /* don't set the pc until the rest of the instruction
9588 has completed */
9589 } else {
9590 /* push lr */
9591 tmp = load_reg(s, 14);
9592 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
9593 tcg_temp_free_i32(tmp);
9594 }
9595 tcg_gen_addi_i32(addr, addr, 4);
9596 }
9597 if ((insn & (1 << 11)) == 0) {
9598 tcg_gen_addi_i32(addr, addr, -offset);
9599 }
9600 /* write back the new stack pointer */
9601 store_reg(s, 13, addr);
9602 /* set the new PC value */
9603 if ((insn & 0x0900) == 0x0900) {
9604 store_reg_from_load(env, s, 15, tmp);
9605 }
9606 break;
9607
9608 case 1: case 3: case 9: case 11: /* czb */
9609 rm = insn & 7;
9610 tmp = load_reg(s, rm);
9611 s->condlabel = gen_new_label();
9612 s->condjmp = 1;
9613 if (insn & (1 << 11))
9614 tcg_gen_brcondi_i32(TCG_COND_EQ, tmp, 0, s->condlabel);
9615 else
9616 tcg_gen_brcondi_i32(TCG_COND_NE, tmp, 0, s->condlabel);
9617 tcg_temp_free_i32(tmp);
9618 offset = ((insn & 0xf8) >> 2) | (insn & 0x200) >> 3;
9619 val = (uint32_t)s->pc + 2;
9620 val += offset;
9621 gen_jmp(s, val);
9622 break;
9623
9624 case 15: /* IT, nop-hint. */
9625 if ((insn & 0xf) == 0) {
9626 gen_nop_hint(s, (insn >> 4) & 0xf);
9627 break;
9628 }
9629 /* If Then. */
9630 s->condexec_cond = (insn >> 4) & 0xe;
9631 s->condexec_mask = insn & 0x1f;
9632 /* No actual code generated for this insn, just setup state. */
9633 break;
9634
9635 case 0xe: /* bkpt */
9636 ARCH(5);
9637 gen_exception_insn(s, 2, EXCP_BKPT);
9638 break;
9639
9640 case 0xa: /* rev */
9641 ARCH(6);
9642 rn = (insn >> 3) & 0x7;
9643 rd = insn & 0x7;
9644 tmp = load_reg(s, rn);
9645 switch ((insn >> 6) & 3) {
9646 case 0: tcg_gen_bswap32_i32(tmp, tmp); break;
9647 case 1: gen_rev16(tmp); break;
9648 case 3: gen_revsh(tmp); break;
9649 default: goto illegal_op;
9650 }
9651 store_reg(s, rd, tmp);
9652 break;
9653
9654 case 6:
9655 switch ((insn >> 5) & 7) {
9656 case 2:
9657 /* setend */
9658 ARCH(6);
9659 if (((insn >> 3) & 1) != s->bswap_code) {
9660 /* Dynamic endianness switching not implemented. */
9661 goto illegal_op;
9662 }
9663 break;
9664 case 3:
9665 /* cps */
9666 ARCH(6);
9667 if (IS_USER(s)) {
9668 break;
9669 }
9670 if (IS_M(env)) {
9671 tmp = tcg_const_i32((insn & (1 << 4)) != 0);
9672 /* FAULTMASK */
9673 if (insn & 1) {
9674 addr = tcg_const_i32(19);
9675 gen_helper_v7m_msr(cpu_env, addr, tmp);
9676 tcg_temp_free_i32(addr);
9677 }
9678 /* PRIMASK */
9679 if (insn & 2) {
9680 addr = tcg_const_i32(16);
9681 gen_helper_v7m_msr(cpu_env, addr, tmp);
9682 tcg_temp_free_i32(addr);
9683 }
9684 tcg_temp_free_i32(tmp);
9685 gen_lookup_tb(s);
9686 } else {
9687 if (insn & (1 << 4)) {
9688 shift = CPSR_A | CPSR_I | CPSR_F;
9689 } else {
9690 shift = 0;
9691 }
9692 gen_set_psr_im(s, ((insn & 7) << 6), 0, shift);
9693 }
9694 break;
9695 default:
9696 goto undef;
9697 }
9698 break;
9699
9700 default:
9701 goto undef;
9702 }
9703 break;
9704
9705 case 12:
9706 {
9707 /* load/store multiple */
9708 TCGv_i32 loaded_var;
9709 TCGV_UNUSED_I32(loaded_var);
9710 rn = (insn >> 8) & 0x7;
9711 addr = load_reg(s, rn);
9712 for (i = 0; i < 8; i++) {
9713 if (insn & (1 << i)) {
9714 if (insn & (1 << 11)) {
9715 /* load */
9716 tmp = tcg_temp_new_i32();
9717 tcg_gen_qemu_ld32u(tmp, addr, IS_USER(s));
9718 if (i == rn) {
9719 loaded_var = tmp;
9720 } else {
9721 store_reg(s, i, tmp);
9722 }
9723 } else {
9724 /* store */
9725 tmp = load_reg(s, i);
9726 tcg_gen_qemu_st32(tmp, addr, IS_USER(s));
9727 tcg_temp_free_i32(tmp);
9728 }
9729 /* advance to the next address */
9730 tcg_gen_addi_i32(addr, addr, 4);
9731 }
9732 }
9733 if ((insn & (1 << rn)) == 0) {
9734 /* base reg not in list: base register writeback */
9735 store_reg(s, rn, addr);
9736 } else {
9737 /* base reg in list: if load, complete it now */
9738 if (insn & (1 << 11)) {
9739 store_reg(s, rn, loaded_var);
9740 }
9741 tcg_temp_free_i32(addr);
9742 }
9743 break;
9744 }
9745 case 13:
9746 /* conditional branch or swi */
9747 cond = (insn >> 8) & 0xf;
9748 if (cond == 0xe)
9749 goto undef;
9750
9751 if (cond == 0xf) {
9752 /* swi */
9753 gen_set_pc_im(s->pc);
9754 s->is_jmp = DISAS_SWI;
9755 break;
9756 }
9757 /* generate a conditional jump to next instruction */
9758 s->condlabel = gen_new_label();
9759 gen_test_cc(cond ^ 1, s->condlabel);
9760 s->condjmp = 1;
9761
9762 /* jump to the offset */
9763 val = (uint32_t)s->pc + 2;
9764 offset = ((int32_t)insn << 24) >> 24;
9765 val += offset << 1;
9766 gen_jmp(s, val);
9767 break;
9768
9769 case 14:
9770 if (insn & (1 << 11)) {
9771 if (disas_thumb2_insn(env, s, insn))
9772 goto undef32;
9773 break;
9774 }
9775 /* unconditional branch */
9776 val = (uint32_t)s->pc;
9777 offset = ((int32_t)insn << 21) >> 21;
9778 val += (offset << 1) + 2;
9779 gen_jmp(s, val);
9780 break;
9781
9782 case 15:
9783 if (disas_thumb2_insn(env, s, insn))
9784 goto undef32;
9785 break;
9786 }
9787 return;
9788 undef32:
9789 gen_exception_insn(s, 4, EXCP_UDEF);
9790 return;
9791 illegal_op:
9792 undef:
9793 gen_exception_insn(s, 2, EXCP_UDEF);
9794 }
9795
9796 /* generate intermediate code in gen_opc_buf and gen_opparam_buf for
9797 basic block 'tb'. If search_pc is TRUE, also generate PC
9798 information for each intermediate instruction. */
9799 static inline void gen_intermediate_code_internal(CPUARMState *env,
9800 TranslationBlock *tb,
9801 int search_pc)
9802 {
9803 DisasContext dc1, *dc = &dc1;
9804 CPUBreakpoint *bp;
9805 uint16_t *gen_opc_end;
9806 int j, lj;
9807 target_ulong pc_start;
9808 uint32_t next_page_start;
9809 int num_insns;
9810 int max_insns;
9811
9812 /* generate intermediate code */
9813 pc_start = tb->pc;
9814
9815 dc->tb = tb;
9816
9817 gen_opc_end = tcg_ctx.gen_opc_buf + OPC_MAX_SIZE;
9818
9819 dc->is_jmp = DISAS_NEXT;
9820 dc->pc = pc_start;
9821 dc->singlestep_enabled = env->singlestep_enabled;
9822 dc->condjmp = 0;
9823 dc->thumb = ARM_TBFLAG_THUMB(tb->flags);
9824 dc->bswap_code = ARM_TBFLAG_BSWAP_CODE(tb->flags);
9825 dc->condexec_mask = (ARM_TBFLAG_CONDEXEC(tb->flags) & 0xf) << 1;
9826 dc->condexec_cond = ARM_TBFLAG_CONDEXEC(tb->flags) >> 4;
9827 #if !defined(CONFIG_USER_ONLY)
9828 dc->user = (ARM_TBFLAG_PRIV(tb->flags) == 0);
9829 #endif
9830 dc->vfp_enabled = ARM_TBFLAG_VFPEN(tb->flags);
9831 dc->vec_len = ARM_TBFLAG_VECLEN(tb->flags);
9832 dc->vec_stride = ARM_TBFLAG_VECSTRIDE(tb->flags);
9833 cpu_F0s = tcg_temp_new_i32();
9834 cpu_F1s = tcg_temp_new_i32();
9835 cpu_F0d = tcg_temp_new_i64();
9836 cpu_F1d = tcg_temp_new_i64();
9837 cpu_V0 = cpu_F0d;
9838 cpu_V1 = cpu_F1d;
9839 /* FIXME: cpu_M0 can probably be the same as cpu_V0. */
9840 cpu_M0 = tcg_temp_new_i64();
9841 next_page_start = (pc_start & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE;
9842 lj = -1;
9843 num_insns = 0;
9844 max_insns = tb->cflags & CF_COUNT_MASK;
9845 if (max_insns == 0)
9846 max_insns = CF_COUNT_MASK;
9847
9848 gen_tb_start();
9849
9850 tcg_clear_temp_count();
9851
9852 /* A note on handling of the condexec (IT) bits:
9853 *
9854 * We want to avoid the overhead of having to write the updated condexec
9855 * bits back to the CPUARMState for every instruction in an IT block. So:
9856 * (1) if the condexec bits are not already zero then we write
9857 * zero back into the CPUARMState now. This avoids complications trying
9858 * to do it at the end of the block. (For example if we don't do this
9859 * it's hard to identify whether we can safely skip writing condexec
9860 * at the end of the TB, which we definitely want to do for the case
9861 * where a TB doesn't do anything with the IT state at all.)
9862 * (2) if we are going to leave the TB then we call gen_set_condexec()
9863 * which will write the correct value into CPUARMState if zero is wrong.
9864 * This is done both for leaving the TB at the end, and for leaving
9865 * it because of an exception we know will happen, which is done in
9866 * gen_exception_insn(). The latter is necessary because we need to
9867 * leave the TB with the PC/IT state just prior to execution of the
9868 * instruction which caused the exception.
9869 * (3) if we leave the TB unexpectedly (eg a data abort on a load)
9870 * then the CPUARMState will be wrong and we need to reset it.
9871 * This is handled in the same way as restoration of the
9872 * PC in these situations: we will be called again with search_pc=1
9873 * and generate a mapping of the condexec bits for each PC in
9874 * gen_opc_condexec_bits[]. restore_state_to_opc() then uses
9875 * this to restore the condexec bits.
9876 *
9877 * Note that there are no instructions which can read the condexec
9878 * bits, and none which can write non-static values to them, so
9879 * we don't need to care about whether CPUARMState is correct in the
9880 * middle of a TB.
9881 */
9882
9883 /* Reset the conditional execution bits immediately. This avoids
9884 complications trying to do it at the end of the block. */
9885 if (dc->condexec_mask || dc->condexec_cond)
9886 {
9887 TCGv_i32 tmp = tcg_temp_new_i32();
9888 tcg_gen_movi_i32(tmp, 0);
9889 store_cpu_field(tmp, condexec_bits);
9890 }
9891 do {
9892 #ifdef CONFIG_USER_ONLY
9893 /* Intercept jump to the magic kernel page. */
9894 if (dc->pc >= 0xffff0000) {
9895 /* We always get here via a jump, so know we are not in a
9896 conditional execution block. */
9897 gen_exception(EXCP_KERNEL_TRAP);
9898 dc->is_jmp = DISAS_UPDATE;
9899 break;
9900 }
9901 #else
9902 if (dc->pc >= 0xfffffff0 && IS_M(env)) {
9903 /* We always get here via a jump, so know we are not in a
9904 conditional execution block. */
9905 gen_exception(EXCP_EXCEPTION_EXIT);
9906 dc->is_jmp = DISAS_UPDATE;
9907 break;
9908 }
9909 #endif
9910
9911 if (unlikely(!QTAILQ_EMPTY(&env->breakpoints))) {
9912 QTAILQ_FOREACH(bp, &env->breakpoints, entry) {
9913 if (bp->pc == dc->pc) {
9914 gen_exception_insn(dc, 0, EXCP_DEBUG);
9915 /* Advance PC so that clearing the breakpoint will
9916 invalidate this TB. */
9917 dc->pc += 2;
9918 goto done_generating;
9919 }
9920 }
9921 }
9922 if (search_pc) {
9923 j = tcg_ctx.gen_opc_ptr - tcg_ctx.gen_opc_buf;
9924 if (lj < j) {
9925 lj++;
9926 while (lj < j)
9927 tcg_ctx.gen_opc_instr_start[lj++] = 0;
9928 }
9929 tcg_ctx.gen_opc_pc[lj] = dc->pc;
9930 gen_opc_condexec_bits[lj] = (dc->condexec_cond << 4) | (dc->condexec_mask >> 1);
9931 tcg_ctx.gen_opc_instr_start[lj] = 1;
9932 tcg_ctx.gen_opc_icount[lj] = num_insns;
9933 }
9934
9935 if (num_insns + 1 == max_insns && (tb->cflags & CF_LAST_IO))
9936 gen_io_start();
9937
9938 if (unlikely(qemu_loglevel_mask(CPU_LOG_TB_OP | CPU_LOG_TB_OP_OPT))) {
9939 tcg_gen_debug_insn_start(dc->pc);
9940 }
9941
9942 if (dc->thumb) {
9943 disas_thumb_insn(env, dc);
9944 if (dc->condexec_mask) {
9945 dc->condexec_cond = (dc->condexec_cond & 0xe)
9946 | ((dc->condexec_mask >> 4) & 1);
9947 dc->condexec_mask = (dc->condexec_mask << 1) & 0x1f;
9948 if (dc->condexec_mask == 0) {
9949 dc->condexec_cond = 0;
9950 }
9951 }
9952 } else {
9953 disas_arm_insn(env, dc);
9954 }
9955
9956 if (dc->condjmp && !dc->is_jmp) {
9957 gen_set_label(dc->condlabel);
9958 dc->condjmp = 0;
9959 }
9960
9961 if (tcg_check_temp_count()) {
9962 fprintf(stderr, "TCG temporary leak before %08x\n", dc->pc);
9963 }
9964
9965 /* Translation stops when a conditional branch is encountered.
9966 * Otherwise the subsequent code could get translated several times.
9967 * Also stop translation when a page boundary is reached. This
9968 * ensures prefetch aborts occur at the right place. */
9969 num_insns ++;
9970 } while (!dc->is_jmp && tcg_ctx.gen_opc_ptr < gen_opc_end &&
9971 !env->singlestep_enabled &&
9972 !singlestep &&
9973 dc->pc < next_page_start &&
9974 num_insns < max_insns);
9975
9976 if (tb->cflags & CF_LAST_IO) {
9977 if (dc->condjmp) {
9978 /* FIXME: This can theoretically happen with self-modifying
9979 code. */
9980 cpu_abort(env, "IO on conditional branch instruction");
9981 }
9982 gen_io_end();
9983 }
9984
9985 /* At this stage dc->condjmp will only be set when the skipped
9986 instruction was a conditional branch or trap, and the PC has
9987 already been written. */
9988 if (unlikely(env->singlestep_enabled)) {
9989 /* Make sure the pc is updated, and raise a debug exception. */
9990 if (dc->condjmp) {
9991 gen_set_condexec(dc);
9992 if (dc->is_jmp == DISAS_SWI) {
9993 gen_exception(EXCP_SWI);
9994 } else {
9995 gen_exception(EXCP_DEBUG);
9996 }
9997 gen_set_label(dc->condlabel);
9998 }
9999 if (dc->condjmp || !dc->is_jmp) {
10000 gen_set_pc_im(dc->pc);
10001 dc->condjmp = 0;
10002 }
10003 gen_set_condexec(dc);
10004 if (dc->is_jmp == DISAS_SWI && !dc->condjmp) {
10005 gen_exception(EXCP_SWI);
10006 } else {
10007 /* FIXME: Single stepping a WFI insn will not halt
10008 the CPU. */
10009 gen_exception(EXCP_DEBUG);
10010 }
10011 } else {
10012 /* While branches must always occur at the end of an IT block,
10013 there are a few other things that can cause us to terminate
10014 the TB in the middle of an IT block:
10015 - Exception generating instructions (bkpt, swi, undefined).
10016 - Page boundaries.
10017 - Hardware watchpoints.
10018 Hardware breakpoints have already been handled and skip this code.
10019 */
10020 gen_set_condexec(dc);
10021 switch(dc->is_jmp) {
10022 case DISAS_NEXT:
10023 gen_goto_tb(dc, 1, dc->pc);
10024 break;
10025 default:
10026 case DISAS_JUMP:
10027 case DISAS_UPDATE:
10028 /* indicate that the hash table must be used to find the next TB */
10029 tcg_gen_exit_tb(0);
10030 break;
10031 case DISAS_TB_JUMP:
10032 /* nothing more to generate */
10033 break;
10034 case DISAS_WFI:
10035 gen_helper_wfi(cpu_env);
10036 break;
10037 case DISAS_SWI:
10038 gen_exception(EXCP_SWI);
10039 break;
10040 }
10041 if (dc->condjmp) {
10042 gen_set_label(dc->condlabel);
10043 gen_set_condexec(dc);
10044 gen_goto_tb(dc, 1, dc->pc);
10045 dc->condjmp = 0;
10046 }
10047 }
10048
10049 done_generating:
10050 gen_tb_end(tb, num_insns);
10051 *tcg_ctx.gen_opc_ptr = INDEX_op_end;
10052
10053 #ifdef DEBUG_DISAS
10054 if (qemu_loglevel_mask(CPU_LOG_TB_IN_ASM)) {
10055 qemu_log("----------------\n");
10056 qemu_log("IN: %s\n", lookup_symbol(pc_start));
10057 log_target_disas(env, pc_start, dc->pc - pc_start,
10058 dc->thumb | (dc->bswap_code << 1));
10059 qemu_log("\n");
10060 }
10061 #endif
10062 if (search_pc) {
10063 j = tcg_ctx.gen_opc_ptr - tcg_ctx.gen_opc_buf;
10064 lj++;
10065 while (lj <= j)
10066 tcg_ctx.gen_opc_instr_start[lj++] = 0;
10067 } else {
10068 tb->size = dc->pc - pc_start;
10069 tb->icount = num_insns;
10070 }
10071 }
10072
10073 void gen_intermediate_code(CPUARMState *env, TranslationBlock *tb)
10074 {
10075 gen_intermediate_code_internal(env, tb, 0);
10076 }
10077
10078 void gen_intermediate_code_pc(CPUARMState *env, TranslationBlock *tb)
10079 {
10080 gen_intermediate_code_internal(env, tb, 1);
10081 }
10082
10083 static const char *cpu_mode_names[16] = {
10084 "usr", "fiq", "irq", "svc", "???", "???", "???", "abt",
10085 "???", "???", "???", "und", "???", "???", "???", "sys"
10086 };
10087
10088 void arm_cpu_dump_state(CPUState *cs, FILE *f, fprintf_function cpu_fprintf,
10089 int flags)
10090 {
10091 ARMCPU *cpu = ARM_CPU(cs);
10092 CPUARMState *env = &cpu->env;
10093 int i;
10094 uint32_t psr;
10095
10096 for(i=0;i<16;i++) {
10097 cpu_fprintf(f, "R%02d=%08x", i, env->regs[i]);
10098 if ((i % 4) == 3)
10099 cpu_fprintf(f, "\n");
10100 else
10101 cpu_fprintf(f, " ");
10102 }
10103 psr = cpsr_read(env);
10104 cpu_fprintf(f, "PSR=%08x %c%c%c%c %c %s%d\n",
10105 psr,
10106 psr & (1 << 31) ? 'N' : '-',
10107 psr & (1 << 30) ? 'Z' : '-',
10108 psr & (1 << 29) ? 'C' : '-',
10109 psr & (1 << 28) ? 'V' : '-',
10110 psr & CPSR_T ? 'T' : 'A',
10111 cpu_mode_names[psr & 0xf], (psr & 0x10) ? 32 : 26);
10112
10113 if (flags & CPU_DUMP_FPU) {
10114 int numvfpregs = 0;
10115 if (arm_feature(env, ARM_FEATURE_VFP)) {
10116 numvfpregs += 16;
10117 }
10118 if (arm_feature(env, ARM_FEATURE_VFP3)) {
10119 numvfpregs += 16;
10120 }
10121 for (i = 0; i < numvfpregs; i++) {
10122 uint64_t v = float64_val(env->vfp.regs[i]);
10123 cpu_fprintf(f, "s%02d=%08x s%02d=%08x d%02d=%016" PRIx64 "\n",
10124 i * 2, (uint32_t)v,
10125 i * 2 + 1, (uint32_t)(v >> 32),
10126 i, v);
10127 }
10128 cpu_fprintf(f, "FPSCR: %08x\n", (int)env->vfp.xregs[ARM_VFP_FPSCR]);
10129 }
10130 }
10131
10132 void restore_state_to_opc(CPUARMState *env, TranslationBlock *tb, int pc_pos)
10133 {
10134 env->regs[15] = tcg_ctx.gen_opc_pc[pc_pos];
10135 env->condexec_bits = gen_opc_condexec_bits[pc_pos];
10136 }
AR7 emulation for QEMU