tcg/arm: add rotation ops
[qemu.git] / tcg / arm / tcg-target.c
blob602fdf233916cff6cba3c634a00468e4bd48e750
1 /*
2 * Tiny Code Generator for QEMU
4 * Copyright (c) 2008 Andrzej Zaborowski
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
25 #if defined(__ARM_ARCH_7__) || \
26 defined(__ARM_ARCH_7A__) || \
27 defined(__ARM_ARCH_7EM__) || \
28 defined(__ARM_ARCH_7M__) || \
29 defined(__ARM_ARCH_7R__)
30 #define USE_ARMV7_INSTRUCTIONS
31 #endif
33 #if defined(USE_ARMV7_INSTRUCTIONS) || \
34 defined(__ARM_ARCH_6J__) || \
35 defined(__ARM_ARCH_6K__) || \
36 defined(__ARM_ARCH_6T2__) || \
37 defined(__ARM_ARCH_6Z__) || \
38 defined(__ARM_ARCH_6ZK__)
39 #define USE_ARMV6_INSTRUCTIONS
40 #endif
42 #if defined(USE_ARMV6_INSTRUCTIONS) || \
43 defined(__ARM_ARCH_5T__) || \
44 defined(__ARM_ARCH_5TE__) || \
45 defined(__ARM_ARCH_5TEJ__)
46 #define USE_ARMV5_INSTRUCTIONS
47 #endif
49 #ifdef USE_ARMV5_INSTRUCTIONS
50 static const int use_armv5_instructions = 1;
51 #else
52 static const int use_armv5_instructions = 0;
53 #endif
54 #undef USE_ARMV5_INSTRUCTIONS
56 #ifdef USE_ARMV6_INSTRUCTIONS
57 static const int use_armv6_instructions = 1;
58 #else
59 static const int use_armv6_instructions = 0;
60 #endif
61 #undef USE_ARMV6_INSTRUCTIONS
63 #ifdef USE_ARMV7_INSTRUCTIONS
64 static const int use_armv7_instructions = 1;
65 #else
66 static const int use_armv7_instructions = 0;
67 #endif
68 #undef USE_ARMV7_INSTRUCTIONS
70 #ifndef NDEBUG
71 static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
72 "%r0",
73 "%r1",
74 "%r2",
75 "%r3",
76 "%r4",
77 "%r5",
78 "%r6",
79 "%r7",
80 "%r8",
81 "%r9",
82 "%r10",
83 "%r11",
84 "%r12",
85 "%r13",
86 "%r14",
87 "%pc",
89 #endif
91 static const int tcg_target_reg_alloc_order[] = {
92 TCG_REG_R0,
93 TCG_REG_R1,
94 TCG_REG_R2,
95 TCG_REG_R3,
96 TCG_REG_R4,
97 TCG_REG_R5,
98 TCG_REG_R6,
99 TCG_REG_R7,
100 TCG_REG_R8,
101 TCG_REG_R9,
102 TCG_REG_R10,
103 TCG_REG_R11,
104 TCG_REG_R12,
105 TCG_REG_R13,
106 TCG_REG_R14,
109 static const int tcg_target_call_iarg_regs[4] = {
110 TCG_REG_R0, TCG_REG_R1, TCG_REG_R2, TCG_REG_R3
112 static const int tcg_target_call_oarg_regs[2] = {
113 TCG_REG_R0, TCG_REG_R1
116 static void patch_reloc(uint8_t *code_ptr, int type,
117 tcg_target_long value, tcg_target_long addend)
119 switch (type) {
120 case R_ARM_ABS32:
121 *(uint32_t *) code_ptr = value;
122 break;
124 case R_ARM_CALL:
125 case R_ARM_JUMP24:
126 default:
127 tcg_abort();
129 case R_ARM_PC24:
130 *(uint32_t *) code_ptr = ((*(uint32_t *) code_ptr) & 0xff000000) |
131 (((value - ((tcg_target_long) code_ptr + 8)) >> 2) & 0xffffff);
132 break;
136 /* maximum number of register used for input function arguments */
137 static inline int tcg_target_get_call_iarg_regs_count(int flags)
139 return 4;
142 /* parse target specific constraints */
143 static int target_parse_constraint(TCGArgConstraint *ct, const char **pct_str)
145 const char *ct_str;
147 ct_str = *pct_str;
148 switch (ct_str[0]) {
149 case 'I':
150 ct->ct |= TCG_CT_CONST_ARM;
151 break;
153 case 'r':
154 #ifndef CONFIG_SOFTMMU
155 case 'd':
156 case 'D':
157 case 'x':
158 case 'X':
159 #endif
160 ct->ct |= TCG_CT_REG;
161 tcg_regset_set32(ct->u.regs, 0, (1 << TCG_TARGET_NB_REGS) - 1);
162 break;
164 #ifdef CONFIG_SOFTMMU
165 /* qemu_ld/st inputs (unless 'X', 'd' or 'D') */
166 case 'x':
167 ct->ct |= TCG_CT_REG;
168 tcg_regset_set32(ct->u.regs, 0, (1 << TCG_TARGET_NB_REGS) - 1);
169 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R0);
170 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R1);
171 break;
173 /* qemu_ld64 data_reg */
174 case 'd':
175 ct->ct |= TCG_CT_REG;
176 tcg_regset_set32(ct->u.regs, 0, (1 << TCG_TARGET_NB_REGS) - 1);
177 /* r1 is still needed to load data_reg2, so don't use it. */
178 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R1);
179 break;
181 /* qemu_ld/st64 data_reg2 */
182 case 'D':
183 ct->ct |= TCG_CT_REG;
184 tcg_regset_set32(ct->u.regs, 0, (1 << TCG_TARGET_NB_REGS) - 1);
185 /* r0, r1 and optionally r2 will be overwritten by the address
186 * and the low word of data, so don't use these. */
187 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R0);
188 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R1);
189 # if TARGET_LONG_BITS == 64
190 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R2);
191 # endif
192 break;
194 # if TARGET_LONG_BITS == 64
195 /* qemu_ld/st addr_reg2 */
196 case 'X':
197 ct->ct |= TCG_CT_REG;
198 tcg_regset_set32(ct->u.regs, 0, (1 << TCG_TARGET_NB_REGS) - 1);
199 /* r0 will be overwritten by the low word of base, so don't use it. */
200 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R0);
201 tcg_regset_reset_reg(ct->u.regs, TCG_REG_R1);
202 break;
203 # endif
204 #endif
206 default:
207 return -1;
209 ct_str++;
210 *pct_str = ct_str;
212 return 0;
215 static inline uint32_t rotl(uint32_t val, int n)
217 return (val << n) | (val >> (32 - n));
220 /* ARM immediates for ALU instructions are made of an unsigned 8-bit
221 right-rotated by an even amount between 0 and 30. */
222 static inline int encode_imm(uint32_t imm)
224 int shift;
226 /* simple case, only lower bits */
227 if ((imm & ~0xff) == 0)
228 return 0;
229 /* then try a simple even shift */
230 shift = ctz32(imm) & ~1;
231 if (((imm >> shift) & ~0xff) == 0)
232 return 32 - shift;
233 /* now try harder with rotations */
234 if ((rotl(imm, 2) & ~0xff) == 0)
235 return 2;
236 if ((rotl(imm, 4) & ~0xff) == 0)
237 return 4;
238 if ((rotl(imm, 6) & ~0xff) == 0)
239 return 6;
240 /* imm can't be encoded */
241 return -1;
244 static inline int check_fit_imm(uint32_t imm)
246 return encode_imm(imm) >= 0;
249 /* Test if a constant matches the constraint.
250 * TODO: define constraints for:
252 * ldr/str offset: between -0xfff and 0xfff
253 * ldrh/strh offset: between -0xff and 0xff
254 * mov operand2: values represented with x << (2 * y), x < 0x100
255 * add, sub, eor...: ditto
257 static inline int tcg_target_const_match(tcg_target_long val,
258 const TCGArgConstraint *arg_ct)
260 int ct;
261 ct = arg_ct->ct;
262 if (ct & TCG_CT_CONST)
263 return 1;
264 else if ((ct & TCG_CT_CONST_ARM) && check_fit_imm(val))
265 return 1;
266 else
267 return 0;
270 enum arm_data_opc_e {
271 ARITH_AND = 0x0,
272 ARITH_EOR = 0x1,
273 ARITH_SUB = 0x2,
274 ARITH_RSB = 0x3,
275 ARITH_ADD = 0x4,
276 ARITH_ADC = 0x5,
277 ARITH_SBC = 0x6,
278 ARITH_RSC = 0x7,
279 ARITH_TST = 0x8,
280 ARITH_CMP = 0xa,
281 ARITH_CMN = 0xb,
282 ARITH_ORR = 0xc,
283 ARITH_MOV = 0xd,
284 ARITH_BIC = 0xe,
285 ARITH_MVN = 0xf,
288 #define TO_CPSR(opc) \
289 ((opc == ARITH_CMP || opc == ARITH_CMN || opc == ARITH_TST) << 20)
291 #define SHIFT_IMM_LSL(im) (((im) << 7) | 0x00)
292 #define SHIFT_IMM_LSR(im) (((im) << 7) | 0x20)
293 #define SHIFT_IMM_ASR(im) (((im) << 7) | 0x40)
294 #define SHIFT_IMM_ROR(im) (((im) << 7) | 0x60)
295 #define SHIFT_REG_LSL(rs) (((rs) << 8) | 0x10)
296 #define SHIFT_REG_LSR(rs) (((rs) << 8) | 0x30)
297 #define SHIFT_REG_ASR(rs) (((rs) << 8) | 0x50)
298 #define SHIFT_REG_ROR(rs) (((rs) << 8) | 0x70)
300 enum arm_cond_code_e {
301 COND_EQ = 0x0,
302 COND_NE = 0x1,
303 COND_CS = 0x2, /* Unsigned greater or equal */
304 COND_CC = 0x3, /* Unsigned less than */
305 COND_MI = 0x4, /* Negative */
306 COND_PL = 0x5, /* Zero or greater */
307 COND_VS = 0x6, /* Overflow */
308 COND_VC = 0x7, /* No overflow */
309 COND_HI = 0x8, /* Unsigned greater than */
310 COND_LS = 0x9, /* Unsigned less or equal */
311 COND_GE = 0xa,
312 COND_LT = 0xb,
313 COND_GT = 0xc,
314 COND_LE = 0xd,
315 COND_AL = 0xe,
318 static const uint8_t tcg_cond_to_arm_cond[10] = {
319 [TCG_COND_EQ] = COND_EQ,
320 [TCG_COND_NE] = COND_NE,
321 [TCG_COND_LT] = COND_LT,
322 [TCG_COND_GE] = COND_GE,
323 [TCG_COND_LE] = COND_LE,
324 [TCG_COND_GT] = COND_GT,
325 /* unsigned */
326 [TCG_COND_LTU] = COND_CC,
327 [TCG_COND_GEU] = COND_CS,
328 [TCG_COND_LEU] = COND_LS,
329 [TCG_COND_GTU] = COND_HI,
332 static inline void tcg_out_bx(TCGContext *s, int cond, int rn)
334 tcg_out32(s, (cond << 28) | 0x012fff10 | rn);
337 static inline void tcg_out_b(TCGContext *s, int cond, int32_t offset)
339 tcg_out32(s, (cond << 28) | 0x0a000000 |
340 (((offset - 8) >> 2) & 0x00ffffff));
343 static inline void tcg_out_b_noaddr(TCGContext *s, int cond)
345 #ifdef HOST_WORDS_BIGENDIAN
346 tcg_out8(s, (cond << 4) | 0x0a);
347 s->code_ptr += 3;
348 #else
349 s->code_ptr += 3;
350 tcg_out8(s, (cond << 4) | 0x0a);
351 #endif
354 static inline void tcg_out_bl(TCGContext *s, int cond, int32_t offset)
356 tcg_out32(s, (cond << 28) | 0x0b000000 |
357 (((offset - 8) >> 2) & 0x00ffffff));
360 static inline void tcg_out_blx(TCGContext *s, int cond, int rn)
362 tcg_out32(s, (cond << 28) | 0x012fff30 | rn);
365 static inline void tcg_out_dat_reg(TCGContext *s,
366 int cond, int opc, int rd, int rn, int rm, int shift)
368 tcg_out32(s, (cond << 28) | (0 << 25) | (opc << 21) | TO_CPSR(opc) |
369 (rn << 16) | (rd << 12) | shift | rm);
372 static inline void tcg_out_dat_reg2(TCGContext *s,
373 int cond, int opc0, int opc1, int rd0, int rd1,
374 int rn0, int rn1, int rm0, int rm1, int shift)
376 if (rd0 == rn1 || rd0 == rm1) {
377 tcg_out32(s, (cond << 28) | (0 << 25) | (opc0 << 21) | (1 << 20) |
378 (rn0 << 16) | (8 << 12) | shift | rm0);
379 tcg_out32(s, (cond << 28) | (0 << 25) | (opc1 << 21) |
380 (rn1 << 16) | (rd1 << 12) | shift | rm1);
381 tcg_out_dat_reg(s, cond, ARITH_MOV,
382 rd0, 0, TCG_REG_R8, SHIFT_IMM_LSL(0));
383 } else {
384 tcg_out32(s, (cond << 28) | (0 << 25) | (opc0 << 21) | (1 << 20) |
385 (rn0 << 16) | (rd0 << 12) | shift | rm0);
386 tcg_out32(s, (cond << 28) | (0 << 25) | (opc1 << 21) |
387 (rn1 << 16) | (rd1 << 12) | shift | rm1);
391 static inline void tcg_out_dat_imm(TCGContext *s,
392 int cond, int opc, int rd, int rn, int im)
394 tcg_out32(s, (cond << 28) | (1 << 25) | (opc << 21) | TO_CPSR(opc) |
395 (rn << 16) | (rd << 12) | im);
398 static inline void tcg_out_movi32(TCGContext *s,
399 int cond, int rd, int32_t arg)
401 int offset = (uint32_t) arg - ((uint32_t) s->code_ptr + 8);
403 /* TODO: This is very suboptimal, we can easily have a constant
404 * pool somewhere after all the instructions. */
406 if (arg < 0 && arg > -0x100)
407 return tcg_out_dat_imm(s, cond, ARITH_MVN, rd, 0, (~arg) & 0xff);
409 if (offset < 0x100 && offset > -0x100)
410 return offset >= 0 ?
411 tcg_out_dat_imm(s, cond, ARITH_ADD, rd, 15, offset) :
412 tcg_out_dat_imm(s, cond, ARITH_SUB, rd, 15, -offset);
414 if (use_armv7_instructions) {
415 /* use movw/movt */
416 /* movw */
417 tcg_out32(s, (cond << 28) | 0x03000000 | (rd << 12)
418 | ((arg << 4) & 0x000f0000) | (arg & 0xfff));
419 if (arg & 0xffff0000)
420 /* movt */
421 tcg_out32(s, (cond << 28) | 0x03400000 | (rd << 12)
422 | ((arg >> 12) & 0x000f0000) | ((arg >> 16) & 0xfff));
423 } else {
424 tcg_out_dat_imm(s, cond, ARITH_MOV, rd, 0, arg & 0xff);
425 if (arg & 0x0000ff00)
426 tcg_out_dat_imm(s, cond, ARITH_ORR, rd, rd,
427 ((arg >> 8) & 0xff) | 0xc00);
428 if (arg & 0x00ff0000)
429 tcg_out_dat_imm(s, cond, ARITH_ORR, rd, rd,
430 ((arg >> 16) & 0xff) | 0x800);
431 if (arg & 0xff000000)
432 tcg_out_dat_imm(s, cond, ARITH_ORR, rd, rd,
433 ((arg >> 24) & 0xff) | 0x400);
437 static inline void tcg_out_mul32(TCGContext *s,
438 int cond, int rd, int rs, int rm)
440 if (rd != rm)
441 tcg_out32(s, (cond << 28) | (rd << 16) | (0 << 12) |
442 (rs << 8) | 0x90 | rm);
443 else if (rd != rs)
444 tcg_out32(s, (cond << 28) | (rd << 16) | (0 << 12) |
445 (rm << 8) | 0x90 | rs);
446 else {
447 tcg_out32(s, (cond << 28) | ( 8 << 16) | (0 << 12) |
448 (rs << 8) | 0x90 | rm);
449 tcg_out_dat_reg(s, cond, ARITH_MOV,
450 rd, 0, TCG_REG_R8, SHIFT_IMM_LSL(0));
454 static inline void tcg_out_umull32(TCGContext *s,
455 int cond, int rd0, int rd1, int rs, int rm)
457 if (rd0 != rm && rd1 != rm)
458 tcg_out32(s, (cond << 28) | 0x800090 |
459 (rd1 << 16) | (rd0 << 12) | (rs << 8) | rm);
460 else if (rd0 != rs && rd1 != rs)
461 tcg_out32(s, (cond << 28) | 0x800090 |
462 (rd1 << 16) | (rd0 << 12) | (rm << 8) | rs);
463 else {
464 tcg_out_dat_reg(s, cond, ARITH_MOV,
465 TCG_REG_R8, 0, rm, SHIFT_IMM_LSL(0));
466 tcg_out32(s, (cond << 28) | 0x800098 |
467 (rd1 << 16) | (rd0 << 12) | (rs << 8));
471 static inline void tcg_out_smull32(TCGContext *s,
472 int cond, int rd0, int rd1, int rs, int rm)
474 if (rd0 != rm && rd1 != rm)
475 tcg_out32(s, (cond << 28) | 0xc00090 |
476 (rd1 << 16) | (rd0 << 12) | (rs << 8) | rm);
477 else if (rd0 != rs && rd1 != rs)
478 tcg_out32(s, (cond << 28) | 0xc00090 |
479 (rd1 << 16) | (rd0 << 12) | (rm << 8) | rs);
480 else {
481 tcg_out_dat_reg(s, cond, ARITH_MOV,
482 TCG_REG_R8, 0, rm, SHIFT_IMM_LSL(0));
483 tcg_out32(s, (cond << 28) | 0xc00098 |
484 (rd1 << 16) | (rd0 << 12) | (rs << 8));
488 static inline void tcg_out_ld32_12(TCGContext *s, int cond,
489 int rd, int rn, tcg_target_long im)
491 if (im >= 0)
492 tcg_out32(s, (cond << 28) | 0x05900000 |
493 (rn << 16) | (rd << 12) | (im & 0xfff));
494 else
495 tcg_out32(s, (cond << 28) | 0x05100000 |
496 (rn << 16) | (rd << 12) | ((-im) & 0xfff));
499 static inline void tcg_out_st32_12(TCGContext *s, int cond,
500 int rd, int rn, tcg_target_long im)
502 if (im >= 0)
503 tcg_out32(s, (cond << 28) | 0x05800000 |
504 (rn << 16) | (rd << 12) | (im & 0xfff));
505 else
506 tcg_out32(s, (cond << 28) | 0x05000000 |
507 (rn << 16) | (rd << 12) | ((-im) & 0xfff));
510 static inline void tcg_out_ld32_r(TCGContext *s, int cond,
511 int rd, int rn, int rm)
513 tcg_out32(s, (cond << 28) | 0x07900000 |
514 (rn << 16) | (rd << 12) | rm);
517 static inline void tcg_out_st32_r(TCGContext *s, int cond,
518 int rd, int rn, int rm)
520 tcg_out32(s, (cond << 28) | 0x07800000 |
521 (rn << 16) | (rd << 12) | rm);
524 /* Register pre-increment with base writeback. */
525 static inline void tcg_out_ld32_rwb(TCGContext *s, int cond,
526 int rd, int rn, int rm)
528 tcg_out32(s, (cond << 28) | 0x07b00000 |
529 (rn << 16) | (rd << 12) | rm);
532 static inline void tcg_out_st32_rwb(TCGContext *s, int cond,
533 int rd, int rn, int rm)
535 tcg_out32(s, (cond << 28) | 0x07a00000 |
536 (rn << 16) | (rd << 12) | rm);
539 static inline void tcg_out_ld16u_8(TCGContext *s, int cond,
540 int rd, int rn, tcg_target_long im)
542 if (im >= 0)
543 tcg_out32(s, (cond << 28) | 0x01d000b0 |
544 (rn << 16) | (rd << 12) |
545 ((im & 0xf0) << 4) | (im & 0xf));
546 else
547 tcg_out32(s, (cond << 28) | 0x015000b0 |
548 (rn << 16) | (rd << 12) |
549 (((-im) & 0xf0) << 4) | ((-im) & 0xf));
552 static inline void tcg_out_st16_8(TCGContext *s, int cond,
553 int rd, int rn, tcg_target_long im)
555 if (im >= 0)
556 tcg_out32(s, (cond << 28) | 0x01c000b0 |
557 (rn << 16) | (rd << 12) |
558 ((im & 0xf0) << 4) | (im & 0xf));
559 else
560 tcg_out32(s, (cond << 28) | 0x014000b0 |
561 (rn << 16) | (rd << 12) |
562 (((-im) & 0xf0) << 4) | ((-im) & 0xf));
565 static inline void tcg_out_ld16u_r(TCGContext *s, int cond,
566 int rd, int rn, int rm)
568 tcg_out32(s, (cond << 28) | 0x019000b0 |
569 (rn << 16) | (rd << 12) | rm);
572 static inline void tcg_out_st16_r(TCGContext *s, int cond,
573 int rd, int rn, int rm)
575 tcg_out32(s, (cond << 28) | 0x018000b0 |
576 (rn << 16) | (rd << 12) | rm);
579 static inline void tcg_out_ld16s_8(TCGContext *s, int cond,
580 int rd, int rn, tcg_target_long im)
582 if (im >= 0)
583 tcg_out32(s, (cond << 28) | 0x01d000f0 |
584 (rn << 16) | (rd << 12) |
585 ((im & 0xf0) << 4) | (im & 0xf));
586 else
587 tcg_out32(s, (cond << 28) | 0x015000f0 |
588 (rn << 16) | (rd << 12) |
589 (((-im) & 0xf0) << 4) | ((-im) & 0xf));
592 static inline void tcg_out_ld16s_r(TCGContext *s, int cond,
593 int rd, int rn, int rm)
595 tcg_out32(s, (cond << 28) | 0x019000f0 |
596 (rn << 16) | (rd << 12) | rm);
599 static inline void tcg_out_ld8_12(TCGContext *s, int cond,
600 int rd, int rn, tcg_target_long im)
602 if (im >= 0)
603 tcg_out32(s, (cond << 28) | 0x05d00000 |
604 (rn << 16) | (rd << 12) | (im & 0xfff));
605 else
606 tcg_out32(s, (cond << 28) | 0x05500000 |
607 (rn << 16) | (rd << 12) | ((-im) & 0xfff));
610 static inline void tcg_out_st8_12(TCGContext *s, int cond,
611 int rd, int rn, tcg_target_long im)
613 if (im >= 0)
614 tcg_out32(s, (cond << 28) | 0x05c00000 |
615 (rn << 16) | (rd << 12) | (im & 0xfff));
616 else
617 tcg_out32(s, (cond << 28) | 0x05400000 |
618 (rn << 16) | (rd << 12) | ((-im) & 0xfff));
621 static inline void tcg_out_ld8_r(TCGContext *s, int cond,
622 int rd, int rn, int rm)
624 tcg_out32(s, (cond << 28) | 0x07d00000 |
625 (rn << 16) | (rd << 12) | rm);
628 static inline void tcg_out_st8_r(TCGContext *s, int cond,
629 int rd, int rn, int rm)
631 tcg_out32(s, (cond << 28) | 0x07c00000 |
632 (rn << 16) | (rd << 12) | rm);
635 static inline void tcg_out_ld8s_8(TCGContext *s, int cond,
636 int rd, int rn, tcg_target_long im)
638 if (im >= 0)
639 tcg_out32(s, (cond << 28) | 0x01d000d0 |
640 (rn << 16) | (rd << 12) |
641 ((im & 0xf0) << 4) | (im & 0xf));
642 else
643 tcg_out32(s, (cond << 28) | 0x015000d0 |
644 (rn << 16) | (rd << 12) |
645 (((-im) & 0xf0) << 4) | ((-im) & 0xf));
648 static inline void tcg_out_ld8s_r(TCGContext *s, int cond,
649 int rd, int rn, int rm)
651 tcg_out32(s, (cond << 28) | 0x019000d0 |
652 (rn << 16) | (rd << 12) | rm);
655 static inline void tcg_out_ld32u(TCGContext *s, int cond,
656 int rd, int rn, int32_t offset)
658 if (offset > 0xfff || offset < -0xfff) {
659 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
660 tcg_out_ld32_r(s, cond, rd, rn, TCG_REG_R8);
661 } else
662 tcg_out_ld32_12(s, cond, rd, rn, offset);
665 static inline void tcg_out_st32(TCGContext *s, int cond,
666 int rd, int rn, int32_t offset)
668 if (offset > 0xfff || offset < -0xfff) {
669 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
670 tcg_out_st32_r(s, cond, rd, rn, TCG_REG_R8);
671 } else
672 tcg_out_st32_12(s, cond, rd, rn, offset);
675 static inline void tcg_out_ld16u(TCGContext *s, int cond,
676 int rd, int rn, int32_t offset)
678 if (offset > 0xff || offset < -0xff) {
679 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
680 tcg_out_ld16u_r(s, cond, rd, rn, TCG_REG_R8);
681 } else
682 tcg_out_ld16u_8(s, cond, rd, rn, offset);
685 static inline void tcg_out_ld16s(TCGContext *s, int cond,
686 int rd, int rn, int32_t offset)
688 if (offset > 0xff || offset < -0xff) {
689 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
690 tcg_out_ld16s_r(s, cond, rd, rn, TCG_REG_R8);
691 } else
692 tcg_out_ld16s_8(s, cond, rd, rn, offset);
695 static inline void tcg_out_st16(TCGContext *s, int cond,
696 int rd, int rn, int32_t offset)
698 if (offset > 0xff || offset < -0xff) {
699 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
700 tcg_out_st16_r(s, cond, rd, rn, TCG_REG_R8);
701 } else
702 tcg_out_st16_8(s, cond, rd, rn, offset);
705 static inline void tcg_out_ld8u(TCGContext *s, int cond,
706 int rd, int rn, int32_t offset)
708 if (offset > 0xfff || offset < -0xfff) {
709 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
710 tcg_out_ld8_r(s, cond, rd, rn, TCG_REG_R8);
711 } else
712 tcg_out_ld8_12(s, cond, rd, rn, offset);
715 static inline void tcg_out_ld8s(TCGContext *s, int cond,
716 int rd, int rn, int32_t offset)
718 if (offset > 0xff || offset < -0xff) {
719 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
720 tcg_out_ld8s_r(s, cond, rd, rn, TCG_REG_R8);
721 } else
722 tcg_out_ld8s_8(s, cond, rd, rn, offset);
725 static inline void tcg_out_st8(TCGContext *s, int cond,
726 int rd, int rn, int32_t offset)
728 if (offset > 0xfff || offset < -0xfff) {
729 tcg_out_movi32(s, cond, TCG_REG_R8, offset);
730 tcg_out_st8_r(s, cond, rd, rn, TCG_REG_R8);
731 } else
732 tcg_out_st8_12(s, cond, rd, rn, offset);
735 static inline void tcg_out_goto(TCGContext *s, int cond, uint32_t addr)
737 int32_t val;
739 val = addr - (tcg_target_long) s->code_ptr;
740 if (val - 8 < 0x01fffffd && val - 8 > -0x01fffffd)
741 tcg_out_b(s, cond, val);
742 else {
743 #if 1
744 tcg_abort();
745 #else
746 if (cond == COND_AL) {
747 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, -4);
748 tcg_out32(s, addr); /* XXX: This is l->u.value, can we use it? */
749 } else {
750 tcg_out_movi32(s, cond, TCG_REG_R8, val - 8);
751 tcg_out_dat_reg(s, cond, ARITH_ADD,
752 TCG_REG_PC, TCG_REG_PC,
753 TCG_REG_R8, SHIFT_IMM_LSL(0));
755 #endif
759 static inline void tcg_out_call(TCGContext *s, int cond, uint32_t addr)
761 int32_t val;
763 val = addr - (tcg_target_long) s->code_ptr;
764 if (val < 0x01fffffd && val > -0x01fffffd)
765 tcg_out_bl(s, cond, val);
766 else {
767 #if 1
768 tcg_abort();
769 #else
770 if (cond == COND_AL) {
771 tcg_out_dat_imm(s, cond, ARITH_ADD, TCG_REG_R14, TCG_REG_PC, 4);
772 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, -4);
773 tcg_out32(s, addr); /* XXX: This is l->u.value, can we use it? */
774 } else {
775 tcg_out_movi32(s, cond, TCG_REG_R9, addr);
776 tcg_out_dat_reg(s, cond, ARITH_MOV, TCG_REG_R14, 0,
777 TCG_REG_PC, SHIFT_IMM_LSL(0));
778 tcg_out_bx(s, cond, TCG_REG_R9);
780 #endif
784 static inline void tcg_out_callr(TCGContext *s, int cond, int arg)
786 if (use_armv5_instructions) {
787 tcg_out_blx(s, cond, arg);
788 } else {
789 tcg_out_dat_reg(s, cond, ARITH_MOV, TCG_REG_R14, 0,
790 TCG_REG_PC, SHIFT_IMM_LSL(0));
791 tcg_out_bx(s, cond, arg);
795 static inline void tcg_out_goto_label(TCGContext *s, int cond, int label_index)
797 TCGLabel *l = &s->labels[label_index];
799 if (l->has_value)
800 tcg_out_goto(s, cond, l->u.value);
801 else if (cond == COND_AL) {
802 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, -4);
803 tcg_out_reloc(s, s->code_ptr, R_ARM_ABS32, label_index, 31337);
804 s->code_ptr += 4;
805 } else {
806 /* Probably this should be preferred even for COND_AL... */
807 tcg_out_reloc(s, s->code_ptr, R_ARM_PC24, label_index, 31337);
808 tcg_out_b_noaddr(s, cond);
812 #ifdef CONFIG_SOFTMMU
814 #include "../../softmmu_defs.h"
816 static void *qemu_ld_helpers[4] = {
817 __ldb_mmu,
818 __ldw_mmu,
819 __ldl_mmu,
820 __ldq_mmu,
823 static void *qemu_st_helpers[4] = {
824 __stb_mmu,
825 __stw_mmu,
826 __stl_mmu,
827 __stq_mmu,
829 #endif
831 #define TLB_SHIFT (CPU_TLB_ENTRY_BITS + CPU_TLB_BITS)
833 static inline void tcg_out_qemu_ld(TCGContext *s, int cond,
834 const TCGArg *args, int opc)
836 int addr_reg, data_reg, data_reg2;
837 #ifdef CONFIG_SOFTMMU
838 int mem_index, s_bits;
839 # if TARGET_LONG_BITS == 64
840 int addr_reg2;
841 # endif
842 uint32_t *label_ptr;
843 #endif
845 data_reg = *args++;
846 if (opc == 3)
847 data_reg2 = *args++;
848 else
849 data_reg2 = 0; /* suppress warning */
850 addr_reg = *args++;
851 #ifdef CONFIG_SOFTMMU
852 # if TARGET_LONG_BITS == 64
853 addr_reg2 = *args++;
854 # endif
855 mem_index = *args;
856 s_bits = opc & 3;
858 /* Should generate something like the following:
859 * shr r8, addr_reg, #TARGET_PAGE_BITS
860 * and r0, r8, #(CPU_TLB_SIZE - 1) @ Assumption: CPU_TLB_BITS <= 8
861 * add r0, env, r0 lsl #CPU_TLB_ENTRY_BITS
863 # if CPU_TLB_BITS > 8
864 # error
865 # endif
866 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, TCG_REG_R8,
867 0, addr_reg, SHIFT_IMM_LSR(TARGET_PAGE_BITS));
868 tcg_out_dat_imm(s, COND_AL, ARITH_AND,
869 TCG_REG_R0, TCG_REG_R8, CPU_TLB_SIZE - 1);
870 tcg_out_dat_reg(s, COND_AL, ARITH_ADD, TCG_REG_R0, TCG_AREG0,
871 TCG_REG_R0, SHIFT_IMM_LSL(CPU_TLB_ENTRY_BITS));
872 /* In the
873 * ldr r1 [r0, #(offsetof(CPUState, tlb_table[mem_index][0].addr_read))]
874 * below, the offset is likely to exceed 12 bits if mem_index != 0 and
875 * not exceed otherwise, so use an
876 * add r0, r0, #(mem_index * sizeof *CPUState.tlb_table)
877 * before.
879 if (mem_index)
880 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R0, TCG_REG_R0,
881 (mem_index << (TLB_SHIFT & 1)) |
882 ((16 - (TLB_SHIFT >> 1)) << 8));
883 tcg_out_ld32_12(s, COND_AL, TCG_REG_R1, TCG_REG_R0,
884 offsetof(CPUState, tlb_table[0][0].addr_read));
885 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0, TCG_REG_R1,
886 TCG_REG_R8, SHIFT_IMM_LSL(TARGET_PAGE_BITS));
887 /* Check alignment. */
888 if (s_bits)
889 tcg_out_dat_imm(s, COND_EQ, ARITH_TST,
890 0, addr_reg, (1 << s_bits) - 1);
891 # if TARGET_LONG_BITS == 64
892 /* XXX: possibly we could use a block data load or writeback in
893 * the first access. */
894 tcg_out_ld32_12(s, COND_EQ, TCG_REG_R1, TCG_REG_R0,
895 offsetof(CPUState, tlb_table[0][0].addr_read) + 4);
896 tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0,
897 TCG_REG_R1, addr_reg2, SHIFT_IMM_LSL(0));
898 # endif
899 tcg_out_ld32_12(s, COND_EQ, TCG_REG_R1, TCG_REG_R0,
900 offsetof(CPUState, tlb_table[0][0].addend));
902 switch (opc) {
903 case 0:
904 tcg_out_ld8_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
905 break;
906 case 0 | 4:
907 tcg_out_ld8s_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
908 break;
909 case 1:
910 tcg_out_ld16u_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
911 break;
912 case 1 | 4:
913 tcg_out_ld16s_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
914 break;
915 case 2:
916 default:
917 tcg_out_ld32_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
918 break;
919 case 3:
920 tcg_out_ld32_rwb(s, COND_EQ, data_reg, TCG_REG_R1, addr_reg);
921 tcg_out_ld32_12(s, COND_EQ, data_reg2, TCG_REG_R1, 4);
922 break;
925 label_ptr = (void *) s->code_ptr;
926 tcg_out_b(s, COND_EQ, 8);
928 /* TODO: move this code to where the constants pool will be */
929 if (addr_reg != TCG_REG_R0) {
930 tcg_out_dat_reg(s, cond, ARITH_MOV,
931 TCG_REG_R0, 0, addr_reg, SHIFT_IMM_LSL(0));
933 # if TARGET_LONG_BITS == 32
934 tcg_out_dat_imm(s, cond, ARITH_MOV, TCG_REG_R1, 0, mem_index);
935 # else
936 if (addr_reg2 != TCG_REG_R1) {
937 tcg_out_dat_reg(s, cond, ARITH_MOV,
938 TCG_REG_R1, 0, addr_reg2, SHIFT_IMM_LSL(0));
940 tcg_out_dat_imm(s, cond, ARITH_MOV, TCG_REG_R2, 0, mem_index);
941 # endif
942 tcg_out_bl(s, cond, (tcg_target_long) qemu_ld_helpers[s_bits] -
943 (tcg_target_long) s->code_ptr);
945 switch (opc) {
946 case 0 | 4:
947 tcg_out_dat_reg(s, cond, ARITH_MOV,
948 TCG_REG_R0, 0, TCG_REG_R0, SHIFT_IMM_LSL(24));
949 tcg_out_dat_reg(s, cond, ARITH_MOV,
950 data_reg, 0, TCG_REG_R0, SHIFT_IMM_ASR(24));
951 break;
952 case 1 | 4:
953 tcg_out_dat_reg(s, cond, ARITH_MOV,
954 TCG_REG_R0, 0, TCG_REG_R0, SHIFT_IMM_LSL(16));
955 tcg_out_dat_reg(s, cond, ARITH_MOV,
956 data_reg, 0, TCG_REG_R0, SHIFT_IMM_ASR(16));
957 break;
958 case 0:
959 case 1:
960 case 2:
961 default:
962 if (data_reg != TCG_REG_R0) {
963 tcg_out_dat_reg(s, cond, ARITH_MOV,
964 data_reg, 0, TCG_REG_R0, SHIFT_IMM_LSL(0));
966 break;
967 case 3:
968 if (data_reg != TCG_REG_R0) {
969 tcg_out_dat_reg(s, cond, ARITH_MOV,
970 data_reg, 0, TCG_REG_R0, SHIFT_IMM_LSL(0));
972 if (data_reg2 != TCG_REG_R1) {
973 tcg_out_dat_reg(s, cond, ARITH_MOV,
974 data_reg2, 0, TCG_REG_R1, SHIFT_IMM_LSL(0));
976 break;
979 *label_ptr += ((void *) s->code_ptr - (void *) label_ptr - 8) >> 2;
980 #else /* !CONFIG_SOFTMMU */
981 if (GUEST_BASE) {
982 uint32_t offset = GUEST_BASE;
983 int i;
984 int rot;
986 while (offset) {
987 i = ctz32(offset) & ~1;
988 rot = ((32 - i) << 7) & 0xf00;
990 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R8, addr_reg,
991 ((offset >> i) & 0xff) | rot);
992 addr_reg = TCG_REG_R8;
993 offset &= ~(0xff << i);
996 switch (opc) {
997 case 0:
998 tcg_out_ld8_12(s, COND_AL, data_reg, addr_reg, 0);
999 break;
1000 case 0 | 4:
1001 tcg_out_ld8s_8(s, COND_AL, data_reg, addr_reg, 0);
1002 break;
1003 case 1:
1004 tcg_out_ld16u_8(s, COND_AL, data_reg, addr_reg, 0);
1005 break;
1006 case 1 | 4:
1007 tcg_out_ld16s_8(s, COND_AL, data_reg, addr_reg, 0);
1008 break;
1009 case 2:
1010 default:
1011 tcg_out_ld32_12(s, COND_AL, data_reg, addr_reg, 0);
1012 break;
1013 case 3:
1014 /* TODO: use block load -
1015 * check that data_reg2 > data_reg or the other way */
1016 if (data_reg == addr_reg) {
1017 tcg_out_ld32_12(s, COND_AL, data_reg2, addr_reg, 4);
1018 tcg_out_ld32_12(s, COND_AL, data_reg, addr_reg, 0);
1019 } else {
1020 tcg_out_ld32_12(s, COND_AL, data_reg, addr_reg, 0);
1021 tcg_out_ld32_12(s, COND_AL, data_reg2, addr_reg, 4);
1023 break;
1025 #endif
1028 static inline void tcg_out_qemu_st(TCGContext *s, int cond,
1029 const TCGArg *args, int opc)
1031 int addr_reg, data_reg, data_reg2;
1032 #ifdef CONFIG_SOFTMMU
1033 int mem_index, s_bits;
1034 # if TARGET_LONG_BITS == 64
1035 int addr_reg2;
1036 # endif
1037 uint32_t *label_ptr;
1038 #endif
1040 data_reg = *args++;
1041 if (opc == 3)
1042 data_reg2 = *args++;
1043 else
1044 data_reg2 = 0; /* suppress warning */
1045 addr_reg = *args++;
1046 #ifdef CONFIG_SOFTMMU
1047 # if TARGET_LONG_BITS == 64
1048 addr_reg2 = *args++;
1049 # endif
1050 mem_index = *args;
1051 s_bits = opc & 3;
1053 /* Should generate something like the following:
1054 * shr r8, addr_reg, #TARGET_PAGE_BITS
1055 * and r0, r8, #(CPU_TLB_SIZE - 1) @ Assumption: CPU_TLB_BITS <= 8
1056 * add r0, env, r0 lsl #CPU_TLB_ENTRY_BITS
1058 tcg_out_dat_reg(s, COND_AL, ARITH_MOV,
1059 TCG_REG_R8, 0, addr_reg, SHIFT_IMM_LSR(TARGET_PAGE_BITS));
1060 tcg_out_dat_imm(s, COND_AL, ARITH_AND,
1061 TCG_REG_R0, TCG_REG_R8, CPU_TLB_SIZE - 1);
1062 tcg_out_dat_reg(s, COND_AL, ARITH_ADD, TCG_REG_R0,
1063 TCG_AREG0, TCG_REG_R0, SHIFT_IMM_LSL(CPU_TLB_ENTRY_BITS));
1064 /* In the
1065 * ldr r1 [r0, #(offsetof(CPUState, tlb_table[mem_index][0].addr_write))]
1066 * below, the offset is likely to exceed 12 bits if mem_index != 0 and
1067 * not exceed otherwise, so use an
1068 * add r0, r0, #(mem_index * sizeof *CPUState.tlb_table)
1069 * before.
1071 if (mem_index)
1072 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R0, TCG_REG_R0,
1073 (mem_index << (TLB_SHIFT & 1)) |
1074 ((16 - (TLB_SHIFT >> 1)) << 8));
1075 tcg_out_ld32_12(s, COND_AL, TCG_REG_R1, TCG_REG_R0,
1076 offsetof(CPUState, tlb_table[0][0].addr_write));
1077 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0, TCG_REG_R1,
1078 TCG_REG_R8, SHIFT_IMM_LSL(TARGET_PAGE_BITS));
1079 /* Check alignment. */
1080 if (s_bits)
1081 tcg_out_dat_imm(s, COND_EQ, ARITH_TST,
1082 0, addr_reg, (1 << s_bits) - 1);
1083 # if TARGET_LONG_BITS == 64
1084 /* XXX: possibly we could use a block data load or writeback in
1085 * the first access. */
1086 tcg_out_ld32_12(s, COND_EQ, TCG_REG_R1, TCG_REG_R0,
1087 offsetof(CPUState, tlb_table[0][0].addr_write) + 4);
1088 tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0,
1089 TCG_REG_R1, addr_reg2, SHIFT_IMM_LSL(0));
1090 # endif
1091 tcg_out_ld32_12(s, COND_EQ, TCG_REG_R1, TCG_REG_R0,
1092 offsetof(CPUState, tlb_table[0][0].addend));
1094 switch (opc) {
1095 case 0:
1096 tcg_out_st8_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1097 break;
1098 case 1:
1099 tcg_out_st16_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1100 break;
1101 case 2:
1102 default:
1103 tcg_out_st32_r(s, COND_EQ, data_reg, addr_reg, TCG_REG_R1);
1104 break;
1105 case 3:
1106 tcg_out_st32_rwb(s, COND_EQ, data_reg, TCG_REG_R1, addr_reg);
1107 tcg_out_st32_12(s, COND_EQ, data_reg2, TCG_REG_R1, 4);
1108 break;
1111 label_ptr = (void *) s->code_ptr;
1112 tcg_out_b(s, COND_EQ, 8);
1114 /* TODO: move this code to where the constants pool will be */
1115 if (addr_reg != TCG_REG_R0) {
1116 tcg_out_dat_reg(s, cond, ARITH_MOV,
1117 TCG_REG_R0, 0, addr_reg, SHIFT_IMM_LSL(0));
1119 # if TARGET_LONG_BITS == 32
1120 switch (opc) {
1121 case 0:
1122 tcg_out_dat_imm(s, cond, ARITH_AND, TCG_REG_R1, data_reg, 0xff);
1123 tcg_out_dat_imm(s, cond, ARITH_MOV, TCG_REG_R2, 0, mem_index);
1124 break;
1125 case 1:
1126 tcg_out_dat_reg(s, cond, ARITH_MOV,
1127 TCG_REG_R1, 0, data_reg, SHIFT_IMM_LSL(16));
1128 tcg_out_dat_reg(s, cond, ARITH_MOV,
1129 TCG_REG_R1, 0, TCG_REG_R1, SHIFT_IMM_LSR(16));
1130 tcg_out_dat_imm(s, cond, ARITH_MOV, TCG_REG_R2, 0, mem_index);
1131 break;
1132 case 2:
1133 if (data_reg != TCG_REG_R1) {
1134 tcg_out_dat_reg(s, cond, ARITH_MOV,
1135 TCG_REG_R1, 0, data_reg, SHIFT_IMM_LSL(0));
1137 tcg_out_dat_imm(s, cond, ARITH_MOV, TCG_REG_R2, 0, mem_index);
1138 break;
1139 case 3:
1140 if (data_reg != TCG_REG_R1) {
1141 tcg_out_dat_reg(s, cond, ARITH_MOV,
1142 TCG_REG_R1, 0, data_reg, SHIFT_IMM_LSL(0));
1144 if (data_reg2 != TCG_REG_R2) {
1145 tcg_out_dat_reg(s, cond, ARITH_MOV,
1146 TCG_REG_R2, 0, data_reg2, SHIFT_IMM_LSL(0));
1148 tcg_out_dat_imm(s, cond, ARITH_MOV, TCG_REG_R3, 0, mem_index);
1149 break;
1151 # else
1152 if (addr_reg2 != TCG_REG_R1) {
1153 tcg_out_dat_reg(s, cond, ARITH_MOV,
1154 TCG_REG_R1, 0, addr_reg2, SHIFT_IMM_LSL(0));
1156 switch (opc) {
1157 case 0:
1158 tcg_out_dat_imm(s, cond, ARITH_AND, TCG_REG_R2, data_reg, 0xff);
1159 tcg_out_dat_imm(s, cond, ARITH_MOV, TCG_REG_R3, 0, mem_index);
1160 break;
1161 case 1:
1162 tcg_out_dat_reg(s, cond, ARITH_MOV,
1163 TCG_REG_R2, 0, data_reg, SHIFT_IMM_LSL(16));
1164 tcg_out_dat_reg(s, cond, ARITH_MOV,
1165 TCG_REG_R2, 0, TCG_REG_R2, SHIFT_IMM_LSR(16));
1166 tcg_out_dat_imm(s, cond, ARITH_MOV, TCG_REG_R3, 0, mem_index);
1167 break;
1168 case 2:
1169 if (data_reg != TCG_REG_R2) {
1170 tcg_out_dat_reg(s, cond, ARITH_MOV,
1171 TCG_REG_R2, 0, data_reg, SHIFT_IMM_LSL(0));
1173 tcg_out_dat_imm(s, cond, ARITH_MOV, TCG_REG_R3, 0, mem_index);
1174 break;
1175 case 3:
1176 tcg_out_dat_imm(s, cond, ARITH_MOV, TCG_REG_R8, 0, mem_index);
1177 tcg_out32(s, (cond << 28) | 0x052d8010); /* str r8, [sp, #-0x10]! */
1178 if (data_reg != TCG_REG_R2) {
1179 tcg_out_dat_reg(s, cond, ARITH_MOV,
1180 TCG_REG_R2, 0, data_reg, SHIFT_IMM_LSL(0));
1182 if (data_reg2 != TCG_REG_R3) {
1183 tcg_out_dat_reg(s, cond, ARITH_MOV,
1184 TCG_REG_R3, 0, data_reg2, SHIFT_IMM_LSL(0));
1186 break;
1188 # endif
1190 tcg_out_bl(s, cond, (tcg_target_long) qemu_st_helpers[s_bits] -
1191 (tcg_target_long) s->code_ptr);
1192 # if TARGET_LONG_BITS == 64
1193 if (opc == 3)
1194 tcg_out_dat_imm(s, cond, ARITH_ADD, TCG_REG_R13, TCG_REG_R13, 0x10);
1195 # endif
1197 *label_ptr += ((void *) s->code_ptr - (void *) label_ptr - 8) >> 2;
1198 #else /* !CONFIG_SOFTMMU */
1199 if (GUEST_BASE) {
1200 uint32_t offset = GUEST_BASE;
1201 int i;
1202 int rot;
1204 while (offset) {
1205 i = ctz32(offset) & ~1;
1206 rot = ((32 - i) << 7) & 0xf00;
1208 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R8, addr_reg,
1209 ((offset >> i) & 0xff) | rot);
1210 addr_reg = TCG_REG_R8;
1211 offset &= ~(0xff << i);
1214 switch (opc) {
1215 case 0:
1216 tcg_out_st8_12(s, COND_AL, data_reg, addr_reg, 0);
1217 break;
1218 case 1:
1219 tcg_out_st16_8(s, COND_AL, data_reg, addr_reg, 0);
1220 break;
1221 case 2:
1222 default:
1223 tcg_out_st32_12(s, COND_AL, data_reg, addr_reg, 0);
1224 break;
1225 case 3:
1226 /* TODO: use block store -
1227 * check that data_reg2 > data_reg or the other way */
1228 tcg_out_st32_12(s, COND_AL, data_reg, addr_reg, 0);
1229 tcg_out_st32_12(s, COND_AL, data_reg2, addr_reg, 4);
1230 break;
1232 #endif
1235 static uint8_t *tb_ret_addr;
1237 static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
1238 const TCGArg *args, const int *const_args)
1240 int c;
1242 switch (opc) {
1243 case INDEX_op_exit_tb:
1245 uint8_t *ld_ptr = s->code_ptr;
1246 if (args[0] >> 8)
1247 tcg_out_ld32_12(s, COND_AL, TCG_REG_R0, TCG_REG_PC, 0);
1248 else
1249 tcg_out_dat_imm(s, COND_AL, ARITH_MOV, TCG_REG_R0, 0, args[0]);
1250 tcg_out_goto(s, COND_AL, (tcg_target_ulong) tb_ret_addr);
1251 if (args[0] >> 8) {
1252 *ld_ptr = (uint8_t) (s->code_ptr - ld_ptr) - 8;
1253 tcg_out32(s, args[0]);
1256 break;
1257 case INDEX_op_goto_tb:
1258 if (s->tb_jmp_offset) {
1259 /* Direct jump method */
1260 #if defined(USE_DIRECT_JUMP)
1261 s->tb_jmp_offset[args[0]] = s->code_ptr - s->code_buf;
1262 tcg_out_b(s, COND_AL, 8);
1263 #else
1264 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, -4);
1265 s->tb_jmp_offset[args[0]] = s->code_ptr - s->code_buf;
1266 tcg_out32(s, 0);
1267 #endif
1268 } else {
1269 /* Indirect jump method */
1270 #if 1
1271 c = (int) (s->tb_next + args[0]) - ((int) s->code_ptr + 8);
1272 if (c > 0xfff || c < -0xfff) {
1273 tcg_out_movi32(s, COND_AL, TCG_REG_R0,
1274 (tcg_target_long) (s->tb_next + args[0]));
1275 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_R0, 0);
1276 } else
1277 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, c);
1278 #else
1279 tcg_out_ld32_12(s, COND_AL, TCG_REG_R0, TCG_REG_PC, 0);
1280 tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_R0, 0);
1281 tcg_out32(s, (tcg_target_long) (s->tb_next + args[0]));
1282 #endif
1284 s->tb_next_offset[args[0]] = s->code_ptr - s->code_buf;
1285 break;
1286 case INDEX_op_call:
1287 if (const_args[0])
1288 tcg_out_call(s, COND_AL, args[0]);
1289 else
1290 tcg_out_callr(s, COND_AL, args[0]);
1291 break;
1292 case INDEX_op_jmp:
1293 if (const_args[0])
1294 tcg_out_goto(s, COND_AL, args[0]);
1295 else
1296 tcg_out_bx(s, COND_AL, args[0]);
1297 break;
1298 case INDEX_op_br:
1299 tcg_out_goto_label(s, COND_AL, args[0]);
1300 break;
1302 case INDEX_op_ld8u_i32:
1303 tcg_out_ld8u(s, COND_AL, args[0], args[1], args[2]);
1304 break;
1305 case INDEX_op_ld8s_i32:
1306 tcg_out_ld8s(s, COND_AL, args[0], args[1], args[2]);
1307 break;
1308 case INDEX_op_ld16u_i32:
1309 tcg_out_ld16u(s, COND_AL, args[0], args[1], args[2]);
1310 break;
1311 case INDEX_op_ld16s_i32:
1312 tcg_out_ld16s(s, COND_AL, args[0], args[1], args[2]);
1313 break;
1314 case INDEX_op_ld_i32:
1315 tcg_out_ld32u(s, COND_AL, args[0], args[1], args[2]);
1316 break;
1317 case INDEX_op_st8_i32:
1318 tcg_out_st8(s, COND_AL, args[0], args[1], args[2]);
1319 break;
1320 case INDEX_op_st16_i32:
1321 tcg_out_st16(s, COND_AL, args[0], args[1], args[2]);
1322 break;
1323 case INDEX_op_st_i32:
1324 tcg_out_st32(s, COND_AL, args[0], args[1], args[2]);
1325 break;
1327 case INDEX_op_mov_i32:
1328 tcg_out_dat_reg(s, COND_AL, ARITH_MOV,
1329 args[0], 0, args[1], SHIFT_IMM_LSL(0));
1330 break;
1331 case INDEX_op_movi_i32:
1332 tcg_out_movi32(s, COND_AL, args[0], args[1]);
1333 break;
1334 case INDEX_op_add_i32:
1335 c = ARITH_ADD;
1336 goto gen_arith;
1337 case INDEX_op_sub_i32:
1338 c = ARITH_SUB;
1339 goto gen_arith;
1340 case INDEX_op_and_i32:
1341 c = ARITH_AND;
1342 goto gen_arith;
1343 case INDEX_op_andc_i32:
1344 c = ARITH_BIC;
1345 goto gen_arith;
1346 case INDEX_op_or_i32:
1347 c = ARITH_ORR;
1348 goto gen_arith;
1349 case INDEX_op_xor_i32:
1350 c = ARITH_EOR;
1351 /* Fall through. */
1352 gen_arith:
1353 if (const_args[2]) {
1354 int rot;
1355 rot = encode_imm(args[2]);
1356 tcg_out_dat_imm(s, COND_AL, c,
1357 args[0], args[1], rotl(args[2], rot) | (rot << 7));
1358 } else
1359 tcg_out_dat_reg(s, COND_AL, c,
1360 args[0], args[1], args[2], SHIFT_IMM_LSL(0));
1361 break;
1362 case INDEX_op_add2_i32:
1363 tcg_out_dat_reg2(s, COND_AL, ARITH_ADD, ARITH_ADC,
1364 args[0], args[1], args[2], args[3],
1365 args[4], args[5], SHIFT_IMM_LSL(0));
1366 break;
1367 case INDEX_op_sub2_i32:
1368 tcg_out_dat_reg2(s, COND_AL, ARITH_SUB, ARITH_SBC,
1369 args[0], args[1], args[2], args[3],
1370 args[4], args[5], SHIFT_IMM_LSL(0));
1371 break;
1372 case INDEX_op_neg_i32:
1373 tcg_out_dat_imm(s, COND_AL, ARITH_RSB, args[0], args[1], 0);
1374 break;
1375 case INDEX_op_not_i32:
1376 tcg_out_dat_reg(s, COND_AL,
1377 ARITH_MVN, args[0], 0, args[1], SHIFT_IMM_LSL(0));
1378 break;
1379 case INDEX_op_mul_i32:
1380 tcg_out_mul32(s, COND_AL, args[0], args[1], args[2]);
1381 break;
1382 case INDEX_op_mulu2_i32:
1383 tcg_out_umull32(s, COND_AL, args[0], args[1], args[2], args[3]);
1384 break;
1385 /* XXX: Perhaps args[2] & 0x1f is wrong */
1386 case INDEX_op_shl_i32:
1387 c = const_args[2] ?
1388 SHIFT_IMM_LSL(args[2] & 0x1f) : SHIFT_REG_LSL(args[2]);
1389 goto gen_shift32;
1390 case INDEX_op_shr_i32:
1391 c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_LSR(args[2] & 0x1f) :
1392 SHIFT_IMM_LSL(0) : SHIFT_REG_LSR(args[2]);
1393 goto gen_shift32;
1394 case INDEX_op_sar_i32:
1395 c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_ASR(args[2] & 0x1f) :
1396 SHIFT_IMM_LSL(0) : SHIFT_REG_ASR(args[2]);
1397 goto gen_shift32;
1398 case INDEX_op_rotr_i32:
1399 c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_ROR(args[2] & 0x1f) :
1400 SHIFT_IMM_LSL(0) : SHIFT_REG_ROR(args[2]);
1401 /* Fall through. */
1402 gen_shift32:
1403 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1], c);
1404 break;
1406 case INDEX_op_rotl_i32:
1407 if (const_args[2]) {
1408 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1],
1409 ((0x20 - args[2]) & 0x1f) ?
1410 SHIFT_IMM_ROR((0x20 - args[2]) & 0x1f) :
1411 SHIFT_IMM_LSL(0));
1412 } else {
1413 tcg_out_dat_imm(s, COND_AL, ARITH_RSB, TCG_REG_R8, args[1], 0x20);
1414 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1],
1415 SHIFT_REG_ROR(TCG_REG_R8));
1417 break;
1419 case INDEX_op_brcond_i32:
1420 if (const_args[1]) {
1421 int rot;
1422 rot = encode_imm(args[1]);
1423 tcg_out_dat_imm(s, COND_AL, ARITH_CMP, 0,
1424 args[0], rotl(args[1], rot) | (rot << 7));
1425 } else {
1426 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0,
1427 args[0], args[1], SHIFT_IMM_LSL(0));
1429 tcg_out_goto_label(s, tcg_cond_to_arm_cond[args[2]], args[3]);
1430 break;
1431 case INDEX_op_brcond2_i32:
1432 /* The resulting conditions are:
1433 * TCG_COND_EQ --> a0 == a2 && a1 == a3,
1434 * TCG_COND_NE --> (a0 != a2 && a1 == a3) || a1 != a3,
1435 * TCG_COND_LT(U) --> (a0 < a2 && a1 == a3) || a1 < a3,
1436 * TCG_COND_GE(U) --> (a0 >= a2 && a1 == a3) || (a1 >= a3 && a1 != a3),
1437 * TCG_COND_LE(U) --> (a0 <= a2 && a1 == a3) || (a1 <= a3 && a1 != a3),
1438 * TCG_COND_GT(U) --> (a0 > a2 && a1 == a3) || a1 > a3,
1440 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0,
1441 args[1], args[3], SHIFT_IMM_LSL(0));
1442 tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0,
1443 args[0], args[2], SHIFT_IMM_LSL(0));
1444 tcg_out_goto_label(s, tcg_cond_to_arm_cond[args[4]], args[5]);
1445 break;
1446 case INDEX_op_setcond_i32:
1447 if (const_args[2]) {
1448 int rot;
1449 rot = encode_imm(args[2]);
1450 tcg_out_dat_imm(s, COND_AL, ARITH_CMP, 0,
1451 args[1], rotl(args[2], rot) | (rot << 7));
1452 } else {
1453 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0,
1454 args[1], args[2], SHIFT_IMM_LSL(0));
1456 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[args[3]],
1457 ARITH_MOV, args[0], 0, 1);
1458 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[tcg_invert_cond(args[3])],
1459 ARITH_MOV, args[0], 0, 0);
1460 break;
1461 case INDEX_op_setcond2_i32:
1462 /* See brcond2_i32 comment */
1463 tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0,
1464 args[2], args[4], SHIFT_IMM_LSL(0));
1465 tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0,
1466 args[1], args[3], SHIFT_IMM_LSL(0));
1467 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[args[5]],
1468 ARITH_MOV, args[0], 0, 1);
1469 tcg_out_dat_imm(s, tcg_cond_to_arm_cond[tcg_invert_cond(args[5])],
1470 ARITH_MOV, args[0], 0, 0);
1471 break;
1473 case INDEX_op_qemu_ld8u:
1474 tcg_out_qemu_ld(s, COND_AL, args, 0);
1475 break;
1476 case INDEX_op_qemu_ld8s:
1477 tcg_out_qemu_ld(s, COND_AL, args, 0 | 4);
1478 break;
1479 case INDEX_op_qemu_ld16u:
1480 tcg_out_qemu_ld(s, COND_AL, args, 1);
1481 break;
1482 case INDEX_op_qemu_ld16s:
1483 tcg_out_qemu_ld(s, COND_AL, args, 1 | 4);
1484 break;
1485 case INDEX_op_qemu_ld32:
1486 tcg_out_qemu_ld(s, COND_AL, args, 2);
1487 break;
1488 case INDEX_op_qemu_ld64:
1489 tcg_out_qemu_ld(s, COND_AL, args, 3);
1490 break;
1492 case INDEX_op_qemu_st8:
1493 tcg_out_qemu_st(s, COND_AL, args, 0);
1494 break;
1495 case INDEX_op_qemu_st16:
1496 tcg_out_qemu_st(s, COND_AL, args, 1);
1497 break;
1498 case INDEX_op_qemu_st32:
1499 tcg_out_qemu_st(s, COND_AL, args, 2);
1500 break;
1501 case INDEX_op_qemu_st64:
1502 tcg_out_qemu_st(s, COND_AL, args, 3);
1503 break;
1505 case INDEX_op_ext8s_i32:
1506 if (use_armv6_instructions) {
1507 /* sxtb */
1508 tcg_out32(s, 0xe6af0070 | (args[0] << 12) | args[1]);
1509 } else {
1510 tcg_out_dat_reg(s, COND_AL, ARITH_MOV,
1511 args[0], 0, args[1], SHIFT_IMM_LSL(24));
1512 tcg_out_dat_reg(s, COND_AL, ARITH_MOV,
1513 args[0], 0, args[0], SHIFT_IMM_ASR(24));
1515 break;
1516 case INDEX_op_ext16s_i32:
1517 if (use_armv6_instructions) {
1518 /* sxth */
1519 tcg_out32(s, 0xe6bf0070 | (args[0] << 12) | args[1]);
1520 } else {
1521 tcg_out_dat_reg(s, COND_AL, ARITH_MOV,
1522 args[0], 0, args[1], SHIFT_IMM_LSL(16));
1523 tcg_out_dat_reg(s, COND_AL, ARITH_MOV,
1524 args[0], 0, args[0], SHIFT_IMM_ASR(16));
1526 break;
1528 default:
1529 tcg_abort();
1533 static const TCGTargetOpDef arm_op_defs[] = {
1534 { INDEX_op_exit_tb, { } },
1535 { INDEX_op_goto_tb, { } },
1536 { INDEX_op_call, { "ri" } },
1537 { INDEX_op_jmp, { "ri" } },
1538 { INDEX_op_br, { } },
1540 { INDEX_op_mov_i32, { "r", "r" } },
1541 { INDEX_op_movi_i32, { "r" } },
1543 { INDEX_op_ld8u_i32, { "r", "r" } },
1544 { INDEX_op_ld8s_i32, { "r", "r" } },
1545 { INDEX_op_ld16u_i32, { "r", "r" } },
1546 { INDEX_op_ld16s_i32, { "r", "r" } },
1547 { INDEX_op_ld_i32, { "r", "r" } },
1548 { INDEX_op_st8_i32, { "r", "r" } },
1549 { INDEX_op_st16_i32, { "r", "r" } },
1550 { INDEX_op_st_i32, { "r", "r" } },
1552 /* TODO: "r", "r", "ri" */
1553 { INDEX_op_add_i32, { "r", "r", "rI" } },
1554 { INDEX_op_sub_i32, { "r", "r", "rI" } },
1555 { INDEX_op_mul_i32, { "r", "r", "r" } },
1556 { INDEX_op_mulu2_i32, { "r", "r", "r", "r" } },
1557 { INDEX_op_and_i32, { "r", "r", "rI" } },
1558 { INDEX_op_andc_i32, { "r", "r", "rI" } },
1559 { INDEX_op_or_i32, { "r", "r", "rI" } },
1560 { INDEX_op_xor_i32, { "r", "r", "rI" } },
1561 { INDEX_op_neg_i32, { "r", "r" } },
1562 { INDEX_op_not_i32, { "r", "r" } },
1564 { INDEX_op_shl_i32, { "r", "r", "ri" } },
1565 { INDEX_op_shr_i32, { "r", "r", "ri" } },
1566 { INDEX_op_sar_i32, { "r", "r", "ri" } },
1567 { INDEX_op_rotl_i32, { "r", "r", "ri" } },
1568 { INDEX_op_rotr_i32, { "r", "r", "ri" } },
1570 { INDEX_op_brcond_i32, { "r", "rI" } },
1571 { INDEX_op_setcond_i32, { "r", "r", "rI" } },
1573 /* TODO: "r", "r", "r", "r", "ri", "ri" */
1574 { INDEX_op_add2_i32, { "r", "r", "r", "r", "r", "r" } },
1575 { INDEX_op_sub2_i32, { "r", "r", "r", "r", "r", "r" } },
1576 { INDEX_op_brcond2_i32, { "r", "r", "r", "r" } },
1577 { INDEX_op_setcond2_i32, { "r", "r", "r", "r", "r" } },
1579 #if TARGET_LONG_BITS == 32
1580 { INDEX_op_qemu_ld8u, { "r", "x" } },
1581 { INDEX_op_qemu_ld8s, { "r", "x" } },
1582 { INDEX_op_qemu_ld16u, { "r", "x" } },
1583 { INDEX_op_qemu_ld16s, { "r", "x" } },
1584 { INDEX_op_qemu_ld32, { "r", "x" } },
1585 { INDEX_op_qemu_ld64, { "d", "r", "x" } },
1587 { INDEX_op_qemu_st8, { "x", "x" } },
1588 { INDEX_op_qemu_st16, { "x", "x" } },
1589 { INDEX_op_qemu_st32, { "x", "x" } },
1590 { INDEX_op_qemu_st64, { "x", "D", "x" } },
1591 #else
1592 { INDEX_op_qemu_ld8u, { "r", "x", "X" } },
1593 { INDEX_op_qemu_ld8s, { "r", "x", "X" } },
1594 { INDEX_op_qemu_ld16u, { "r", "x", "X" } },
1595 { INDEX_op_qemu_ld16s, { "r", "x", "X" } },
1596 { INDEX_op_qemu_ld32, { "r", "x", "X" } },
1597 { INDEX_op_qemu_ld64, { "d", "r", "x", "X" } },
1599 { INDEX_op_qemu_st8, { "x", "x", "X" } },
1600 { INDEX_op_qemu_st16, { "x", "x", "X" } },
1601 { INDEX_op_qemu_st32, { "x", "x", "X" } },
1602 { INDEX_op_qemu_st64, { "x", "D", "x", "X" } },
1603 #endif
1605 { INDEX_op_ext8s_i32, { "r", "r" } },
1606 { INDEX_op_ext16s_i32, { "r", "r" } },
1608 { -1 },
1611 void tcg_target_init(TCGContext *s)
1613 #if !defined(CONFIG_USER_ONLY)
1614 /* fail safe */
1615 if ((1 << CPU_TLB_ENTRY_BITS) != sizeof(CPUTLBEntry))
1616 tcg_abort();
1617 #endif
1619 tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I32], 0, 0xffff);
1620 tcg_regset_set32(tcg_target_call_clobber_regs, 0,
1621 (1 << TCG_REG_R0) |
1622 (1 << TCG_REG_R1) |
1623 (1 << TCG_REG_R2) |
1624 (1 << TCG_REG_R3) |
1625 (1 << TCG_REG_R12) |
1626 (1 << TCG_REG_R14));
1628 tcg_regset_clear(s->reserved_regs);
1629 tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
1630 tcg_regset_set_reg(s->reserved_regs, TCG_REG_R8);
1631 tcg_regset_set_reg(s->reserved_regs, TCG_REG_PC);
1633 tcg_add_target_add_op_defs(arm_op_defs);
1636 static inline void tcg_out_ld(TCGContext *s, TCGType type, int arg,
1637 int arg1, tcg_target_long arg2)
1639 tcg_out_ld32u(s, COND_AL, arg, arg1, arg2);
1642 static inline void tcg_out_st(TCGContext *s, TCGType type, int arg,
1643 int arg1, tcg_target_long arg2)
1645 tcg_out_st32(s, COND_AL, arg, arg1, arg2);
1648 static void tcg_out_addi(TCGContext *s, int reg, tcg_target_long val)
1650 if (val > 0)
1651 if (val < 0x100)
1652 tcg_out_dat_imm(s, COND_AL, ARITH_ADD, reg, reg, val);
1653 else
1654 tcg_abort();
1655 else if (val < 0) {
1656 if (val > -0x100)
1657 tcg_out_dat_imm(s, COND_AL, ARITH_SUB, reg, reg, -val);
1658 else
1659 tcg_abort();
1663 static inline void tcg_out_mov(TCGContext *s, int ret, int arg)
1665 tcg_out_dat_reg(s, COND_AL, ARITH_MOV, ret, 0, arg, SHIFT_IMM_LSL(0));
1668 static inline void tcg_out_movi(TCGContext *s, TCGType type,
1669 int ret, tcg_target_long arg)
1671 tcg_out_movi32(s, COND_AL, ret, arg);
1674 void tcg_target_qemu_prologue(TCGContext *s)
1676 /* There is no need to save r7, it is used to store the address
1677 of the env structure and is not modified by GCC. */
1679 /* stmdb sp!, { r4 - r6, r8 - r11, lr } */
1680 tcg_out32(s, (COND_AL << 28) | 0x092d4f70);
1682 tcg_out_bx(s, COND_AL, TCG_REG_R0);
1683 tb_ret_addr = s->code_ptr;
1685 /* ldmia sp!, { r4 - r6, r8 - r11, pc } */
1686 tcg_out32(s, (COND_AL << 28) | 0x08bd8f70);