root / tcg / i386 / tcg-target.c @ 9363dedb
History | View | Annotate | Download (41.4 kB)
1 |
/*
|
---|---|
2 |
* Tiny Code Generator for QEMU
|
3 |
*
|
4 |
* Copyright (c) 2008 Fabrice Bellard
|
5 |
*
|
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:
|
12 |
*
|
13 |
* The above copyright notice and this permission notice shall be included in
|
14 |
* all copies or substantial portions of the Software.
|
15 |
*
|
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.
|
23 |
*/
|
24 |
|
25 |
#ifndef NDEBUG
|
26 |
static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = { |
27 |
"%eax",
|
28 |
"%ecx",
|
29 |
"%edx",
|
30 |
"%ebx",
|
31 |
"%esp",
|
32 |
"%ebp",
|
33 |
"%esi",
|
34 |
"%edi",
|
35 |
}; |
36 |
#endif
|
37 |
|
38 |
static const int tcg_target_reg_alloc_order[] = { |
39 |
TCG_REG_EBX, |
40 |
TCG_REG_ESI, |
41 |
TCG_REG_EDI, |
42 |
TCG_REG_EBP, |
43 |
TCG_REG_ECX, |
44 |
TCG_REG_EDX, |
45 |
TCG_REG_EAX, |
46 |
}; |
47 |
|
48 |
static const int tcg_target_call_iarg_regs[3] = { TCG_REG_EAX, TCG_REG_EDX, TCG_REG_ECX }; |
49 |
static const int tcg_target_call_oarg_regs[2] = { TCG_REG_EAX, TCG_REG_EDX }; |
50 |
|
51 |
static uint8_t *tb_ret_addr;
|
52 |
|
53 |
static void patch_reloc(uint8_t *code_ptr, int type, |
54 |
tcg_target_long value, tcg_target_long addend) |
55 |
{ |
56 |
value += addend; |
57 |
switch(type) {
|
58 |
case R_386_32:
|
59 |
*(uint32_t *)code_ptr = value; |
60 |
break;
|
61 |
case R_386_PC32:
|
62 |
*(uint32_t *)code_ptr = value - (long)code_ptr;
|
63 |
break;
|
64 |
case R_386_PC8:
|
65 |
value -= (long)code_ptr;
|
66 |
if (value != (int8_t)value) {
|
67 |
tcg_abort(); |
68 |
} |
69 |
*(uint8_t *)code_ptr = value; |
70 |
break;
|
71 |
default:
|
72 |
tcg_abort(); |
73 |
} |
74 |
} |
75 |
|
76 |
/* maximum number of register used for input function arguments */
|
77 |
static inline int tcg_target_get_call_iarg_regs_count(int flags) |
78 |
{ |
79 |
flags &= TCG_CALL_TYPE_MASK; |
80 |
switch(flags) {
|
81 |
case TCG_CALL_TYPE_STD:
|
82 |
return 0; |
83 |
case TCG_CALL_TYPE_REGPARM_1:
|
84 |
case TCG_CALL_TYPE_REGPARM_2:
|
85 |
case TCG_CALL_TYPE_REGPARM:
|
86 |
return flags - TCG_CALL_TYPE_REGPARM_1 + 1; |
87 |
default:
|
88 |
tcg_abort(); |
89 |
} |
90 |
} |
91 |
|
92 |
/* parse target specific constraints */
|
93 |
static int target_parse_constraint(TCGArgConstraint *ct, const char **pct_str) |
94 |
{ |
95 |
const char *ct_str; |
96 |
|
97 |
ct_str = *pct_str; |
98 |
switch(ct_str[0]) { |
99 |
case 'a': |
100 |
ct->ct |= TCG_CT_REG; |
101 |
tcg_regset_set_reg(ct->u.regs, TCG_REG_EAX); |
102 |
break;
|
103 |
case 'b': |
104 |
ct->ct |= TCG_CT_REG; |
105 |
tcg_regset_set_reg(ct->u.regs, TCG_REG_EBX); |
106 |
break;
|
107 |
case 'c': |
108 |
ct->ct |= TCG_CT_REG; |
109 |
tcg_regset_set_reg(ct->u.regs, TCG_REG_ECX); |
110 |
break;
|
111 |
case 'd': |
112 |
ct->ct |= TCG_CT_REG; |
113 |
tcg_regset_set_reg(ct->u.regs, TCG_REG_EDX); |
114 |
break;
|
115 |
case 'S': |
116 |
ct->ct |= TCG_CT_REG; |
117 |
tcg_regset_set_reg(ct->u.regs, TCG_REG_ESI); |
118 |
break;
|
119 |
case 'D': |
120 |
ct->ct |= TCG_CT_REG; |
121 |
tcg_regset_set_reg(ct->u.regs, TCG_REG_EDI); |
122 |
break;
|
123 |
case 'q': |
124 |
ct->ct |= TCG_CT_REG; |
125 |
tcg_regset_set32(ct->u.regs, 0, 0xf); |
126 |
break;
|
127 |
case 'r': |
128 |
ct->ct |= TCG_CT_REG; |
129 |
tcg_regset_set32(ct->u.regs, 0, 0xff); |
130 |
break;
|
131 |
|
132 |
/* qemu_ld/st address constraint */
|
133 |
case 'L': |
134 |
ct->ct |= TCG_CT_REG; |
135 |
tcg_regset_set32(ct->u.regs, 0, 0xff); |
136 |
tcg_regset_reset_reg(ct->u.regs, TCG_REG_EAX); |
137 |
tcg_regset_reset_reg(ct->u.regs, TCG_REG_EDX); |
138 |
break;
|
139 |
default:
|
140 |
return -1; |
141 |
} |
142 |
ct_str++; |
143 |
*pct_str = ct_str; |
144 |
return 0; |
145 |
} |
146 |
|
147 |
/* test if a constant matches the constraint */
|
148 |
static inline int tcg_target_const_match(tcg_target_long val, |
149 |
const TCGArgConstraint *arg_ct)
|
150 |
{ |
151 |
int ct;
|
152 |
ct = arg_ct->ct; |
153 |
if (ct & TCG_CT_CONST)
|
154 |
return 1; |
155 |
else
|
156 |
return 0; |
157 |
} |
158 |
|
159 |
#define P_EXT 0x100 /* 0x0f opcode prefix */ |
160 |
|
161 |
#define OPC_ARITH_EvIz (0x81) |
162 |
#define OPC_ARITH_EvIb (0x83) |
163 |
#define OPC_ARITH_GvEv (0x03) /* ... plus (ARITH_FOO << 3) */ |
164 |
#define OPC_ADD_GvEv (OPC_ARITH_GvEv | (ARITH_ADD << 3)) |
165 |
#define OPC_BSWAP (0xc8 | P_EXT) |
166 |
#define OPC_CALL_Jz (0xe8) |
167 |
#define OPC_CMP_GvEv (OPC_ARITH_GvEv | (ARITH_CMP << 3)) |
168 |
#define OPC_DEC_r32 (0x48) |
169 |
#define OPC_INC_r32 (0x40) |
170 |
#define OPC_JCC_long (0x80 | P_EXT) /* ... plus condition code */ |
171 |
#define OPC_JCC_short (0x70) /* ... plus condition code */ |
172 |
#define OPC_JMP_long (0xe9) |
173 |
#define OPC_JMP_short (0xeb) |
174 |
#define OPC_MOVB_EvGv (0x88) /* stores, more or less */ |
175 |
#define OPC_MOVL_EvGv (0x89) /* stores, more or less */ |
176 |
#define OPC_MOVL_GvEv (0x8b) /* loads, more or less */ |
177 |
#define OPC_MOVL_Iv (0xb8) |
178 |
#define OPC_MOVSBL (0xbe | P_EXT) |
179 |
#define OPC_MOVSWL (0xbf | P_EXT) |
180 |
#define OPC_MOVZBL (0xb6 | P_EXT) |
181 |
#define OPC_MOVZWL (0xb7 | P_EXT) |
182 |
#define OPC_POP_r32 (0x58) |
183 |
#define OPC_PUSH_r32 (0x50) |
184 |
#define OPC_PUSH_Iv (0x68) |
185 |
#define OPC_PUSH_Ib (0x6a) |
186 |
#define OPC_RET (0xc3) |
187 |
#define OPC_SETCC (0x90 | P_EXT) /* ... plus condition code */ |
188 |
#define OPC_SHIFT_1 (0xd1) |
189 |
#define OPC_SHIFT_Ib (0xc1) |
190 |
#define OPC_SHIFT_cl (0xd3) |
191 |
#define OPC_TESTL (0x85) |
192 |
|
193 |
#define OPC_GRP3_Ev (0xf7) |
194 |
#define OPC_GRP5 (0xff) |
195 |
|
196 |
/* Group 1 opcode extensions for 0x80-0x83.
|
197 |
These are also used as modifiers for OPC_ARITH. */
|
198 |
#define ARITH_ADD 0 |
199 |
#define ARITH_OR 1 |
200 |
#define ARITH_ADC 2 |
201 |
#define ARITH_SBB 3 |
202 |
#define ARITH_AND 4 |
203 |
#define ARITH_SUB 5 |
204 |
#define ARITH_XOR 6 |
205 |
#define ARITH_CMP 7 |
206 |
|
207 |
/* Group 2 opcode extensions for 0xc0, 0xc1, 0xd0-0xd3. */
|
208 |
#define SHIFT_ROL 0 |
209 |
#define SHIFT_ROR 1 |
210 |
#define SHIFT_SHL 4 |
211 |
#define SHIFT_SHR 5 |
212 |
#define SHIFT_SAR 7 |
213 |
|
214 |
/* Group 3 opcode extensions for 0xf6, 0xf7. To be used with OPC_GRP3. */
|
215 |
#define EXT3_NOT 2 |
216 |
#define EXT3_NEG 3 |
217 |
#define EXT3_MUL 4 |
218 |
#define EXT3_IMUL 5 |
219 |
#define EXT3_DIV 6 |
220 |
#define EXT3_IDIV 7 |
221 |
|
222 |
/* Group 5 opcode extensions for 0xff. To be used with OPC_GRP5. */
|
223 |
#define EXT5_CALLN_Ev 2 |
224 |
#define EXT5_JMPN_Ev 4 |
225 |
|
226 |
/* Condition codes to be added to OPC_JCC_{long,short}. */
|
227 |
#define JCC_JMP (-1) |
228 |
#define JCC_JO 0x0 |
229 |
#define JCC_JNO 0x1 |
230 |
#define JCC_JB 0x2 |
231 |
#define JCC_JAE 0x3 |
232 |
#define JCC_JE 0x4 |
233 |
#define JCC_JNE 0x5 |
234 |
#define JCC_JBE 0x6 |
235 |
#define JCC_JA 0x7 |
236 |
#define JCC_JS 0x8 |
237 |
#define JCC_JNS 0x9 |
238 |
#define JCC_JP 0xa |
239 |
#define JCC_JNP 0xb |
240 |
#define JCC_JL 0xc |
241 |
#define JCC_JGE 0xd |
242 |
#define JCC_JLE 0xe |
243 |
#define JCC_JG 0xf |
244 |
|
245 |
static const uint8_t tcg_cond_to_jcc[10] = { |
246 |
[TCG_COND_EQ] = JCC_JE, |
247 |
[TCG_COND_NE] = JCC_JNE, |
248 |
[TCG_COND_LT] = JCC_JL, |
249 |
[TCG_COND_GE] = JCC_JGE, |
250 |
[TCG_COND_LE] = JCC_JLE, |
251 |
[TCG_COND_GT] = JCC_JG, |
252 |
[TCG_COND_LTU] = JCC_JB, |
253 |
[TCG_COND_GEU] = JCC_JAE, |
254 |
[TCG_COND_LEU] = JCC_JBE, |
255 |
[TCG_COND_GTU] = JCC_JA, |
256 |
}; |
257 |
|
258 |
static inline void tcg_out_opc(TCGContext *s, int opc) |
259 |
{ |
260 |
if (opc & P_EXT)
|
261 |
tcg_out8(s, 0x0f);
|
262 |
tcg_out8(s, opc); |
263 |
} |
264 |
|
265 |
static inline void tcg_out_modrm(TCGContext *s, int opc, int r, int rm) |
266 |
{ |
267 |
tcg_out_opc(s, opc); |
268 |
tcg_out8(s, 0xc0 | (r << 3) | rm); |
269 |
} |
270 |
|
271 |
/* rm == -1 means no register index */
|
272 |
static inline void tcg_out_modrm_offset(TCGContext *s, int opc, int r, int rm, |
273 |
int32_t offset) |
274 |
{ |
275 |
tcg_out_opc(s, opc); |
276 |
if (rm == -1) { |
277 |
tcg_out8(s, 0x05 | (r << 3)); |
278 |
tcg_out32(s, offset); |
279 |
} else if (offset == 0 && rm != TCG_REG_EBP) { |
280 |
if (rm == TCG_REG_ESP) {
|
281 |
tcg_out8(s, 0x04 | (r << 3)); |
282 |
tcg_out8(s, 0x24);
|
283 |
} else {
|
284 |
tcg_out8(s, 0x00 | (r << 3) | rm); |
285 |
} |
286 |
} else if ((int8_t)offset == offset) { |
287 |
if (rm == TCG_REG_ESP) {
|
288 |
tcg_out8(s, 0x44 | (r << 3)); |
289 |
tcg_out8(s, 0x24);
|
290 |
} else {
|
291 |
tcg_out8(s, 0x40 | (r << 3) | rm); |
292 |
} |
293 |
tcg_out8(s, offset); |
294 |
} else {
|
295 |
if (rm == TCG_REG_ESP) {
|
296 |
tcg_out8(s, 0x84 | (r << 3)); |
297 |
tcg_out8(s, 0x24);
|
298 |
} else {
|
299 |
tcg_out8(s, 0x80 | (r << 3) | rm); |
300 |
} |
301 |
tcg_out32(s, offset); |
302 |
} |
303 |
} |
304 |
|
305 |
/* Generate dest op= src. Uses the same ARITH_* codes as tgen_arithi. */
|
306 |
static inline void tgen_arithr(TCGContext *s, int subop, int dest, int src) |
307 |
{ |
308 |
tcg_out_modrm(s, OPC_ARITH_GvEv + (subop << 3), dest, src);
|
309 |
} |
310 |
|
311 |
static inline void tcg_out_mov(TCGContext *s, int ret, int arg) |
312 |
{ |
313 |
if (arg != ret) {
|
314 |
tcg_out_modrm(s, OPC_MOVL_GvEv, ret, arg); |
315 |
} |
316 |
} |
317 |
|
318 |
static inline void tcg_out_movi(TCGContext *s, TCGType type, |
319 |
int ret, int32_t arg)
|
320 |
{ |
321 |
if (arg == 0) { |
322 |
tgen_arithr(s, ARITH_XOR, ret, ret); |
323 |
} else {
|
324 |
tcg_out8(s, OPC_MOVL_Iv + ret); |
325 |
tcg_out32(s, arg); |
326 |
} |
327 |
} |
328 |
|
329 |
static inline void tcg_out_pushi(TCGContext *s, tcg_target_long val) |
330 |
{ |
331 |
if (val == (int8_t)val) {
|
332 |
tcg_out_opc(s, OPC_PUSH_Ib); |
333 |
tcg_out8(s, val); |
334 |
} else {
|
335 |
tcg_out_opc(s, OPC_PUSH_Iv); |
336 |
tcg_out32(s, val); |
337 |
} |
338 |
} |
339 |
|
340 |
static inline void tcg_out_push(TCGContext *s, int reg) |
341 |
{ |
342 |
tcg_out_opc(s, OPC_PUSH_r32 + reg); |
343 |
} |
344 |
|
345 |
static inline void tcg_out_pop(TCGContext *s, int reg) |
346 |
{ |
347 |
tcg_out_opc(s, OPC_POP_r32 + reg); |
348 |
} |
349 |
|
350 |
static inline void tcg_out_ld(TCGContext *s, TCGType type, int ret, |
351 |
int arg1, tcg_target_long arg2)
|
352 |
{ |
353 |
tcg_out_modrm_offset(s, OPC_MOVL_GvEv, ret, arg1, arg2); |
354 |
} |
355 |
|
356 |
static inline void tcg_out_st(TCGContext *s, TCGType type, int arg, |
357 |
int arg1, tcg_target_long arg2)
|
358 |
{ |
359 |
tcg_out_modrm_offset(s, OPC_MOVL_EvGv, arg, arg1, arg2); |
360 |
} |
361 |
|
362 |
static void tcg_out_shifti(TCGContext *s, int subopc, int reg, int count) |
363 |
{ |
364 |
if (count == 1) { |
365 |
tcg_out_modrm(s, OPC_SHIFT_1, subopc, reg); |
366 |
} else {
|
367 |
tcg_out_modrm(s, OPC_SHIFT_Ib, subopc, reg); |
368 |
tcg_out8(s, count); |
369 |
} |
370 |
} |
371 |
|
372 |
static inline void tcg_out_bswap32(TCGContext *s, int reg) |
373 |
{ |
374 |
tcg_out_opc(s, OPC_BSWAP + reg); |
375 |
} |
376 |
|
377 |
static inline void tcg_out_rolw_8(TCGContext *s, int reg) |
378 |
{ |
379 |
tcg_out8(s, 0x66);
|
380 |
tcg_out_shifti(s, SHIFT_ROL, reg, 8);
|
381 |
} |
382 |
|
383 |
static inline void tcg_out_ext8u(TCGContext *s, int dest, int src) |
384 |
{ |
385 |
/* movzbl */
|
386 |
assert(src < 4);
|
387 |
tcg_out_modrm(s, OPC_MOVZBL, dest, src); |
388 |
} |
389 |
|
390 |
static void tcg_out_ext8s(TCGContext *s, int dest, int src) |
391 |
{ |
392 |
/* movsbl */
|
393 |
assert(src < 4);
|
394 |
tcg_out_modrm(s, OPC_MOVSBL, dest, src); |
395 |
} |
396 |
|
397 |
static inline void tcg_out_ext16u(TCGContext *s, int dest, int src) |
398 |
{ |
399 |
/* movzwl */
|
400 |
tcg_out_modrm(s, OPC_MOVZWL, dest, src); |
401 |
} |
402 |
|
403 |
static inline void tcg_out_ext16s(TCGContext *s, int dest, int src) |
404 |
{ |
405 |
/* movswl */
|
406 |
tcg_out_modrm(s, OPC_MOVSWL, dest, src); |
407 |
} |
408 |
|
409 |
static inline void tgen_arithi(TCGContext *s, int c, int r0, |
410 |
int32_t val, int cf)
|
411 |
{ |
412 |
/* ??? While INC is 2 bytes shorter than ADDL $1, they also induce
|
413 |
partial flags update stalls on Pentium4 and are not recommended
|
414 |
by current Intel optimization manuals. */
|
415 |
if (!cf && (c == ARITH_ADD || c == ARITH_SUB) && (val == 1 || val == -1)) { |
416 |
int opc = ((c == ARITH_ADD) ^ (val < 0) ? OPC_INC_r32 : OPC_DEC_r32); |
417 |
tcg_out_opc(s, opc + r0); |
418 |
} else if (val == (int8_t)val) { |
419 |
tcg_out_modrm(s, OPC_ARITH_EvIb, c, r0); |
420 |
tcg_out8(s, val); |
421 |
} else if (c == ARITH_AND && val == 0xffu && r0 < 4) { |
422 |
tcg_out_ext8u(s, r0, r0); |
423 |
} else if (c == ARITH_AND && val == 0xffffu) { |
424 |
tcg_out_ext16u(s, r0, r0); |
425 |
} else {
|
426 |
tcg_out_modrm(s, OPC_ARITH_EvIz, c, r0); |
427 |
tcg_out32(s, val); |
428 |
} |
429 |
} |
430 |
|
431 |
static void tcg_out_addi(TCGContext *s, int reg, tcg_target_long val) |
432 |
{ |
433 |
if (val != 0) |
434 |
tgen_arithi(s, ARITH_ADD, reg, val, 0);
|
435 |
} |
436 |
|
437 |
/* Use SMALL != 0 to force a short forward branch. */
|
438 |
static void tcg_out_jxx(TCGContext *s, int opc, int label_index, int small) |
439 |
{ |
440 |
int32_t val, val1; |
441 |
TCGLabel *l = &s->labels[label_index]; |
442 |
|
443 |
if (l->has_value) {
|
444 |
val = l->u.value - (tcg_target_long)s->code_ptr; |
445 |
val1 = val - 2;
|
446 |
if ((int8_t)val1 == val1) {
|
447 |
if (opc == -1) { |
448 |
tcg_out8(s, OPC_JMP_short); |
449 |
} else {
|
450 |
tcg_out8(s, OPC_JCC_short + opc); |
451 |
} |
452 |
tcg_out8(s, val1); |
453 |
} else {
|
454 |
if (small) {
|
455 |
tcg_abort(); |
456 |
} |
457 |
if (opc == -1) { |
458 |
tcg_out8(s, OPC_JMP_long); |
459 |
tcg_out32(s, val - 5);
|
460 |
} else {
|
461 |
tcg_out_opc(s, OPC_JCC_long + opc); |
462 |
tcg_out32(s, val - 6);
|
463 |
} |
464 |
} |
465 |
} else if (small) { |
466 |
if (opc == -1) { |
467 |
tcg_out8(s, OPC_JMP_short); |
468 |
} else {
|
469 |
tcg_out8(s, OPC_JCC_short + opc); |
470 |
} |
471 |
tcg_out_reloc(s, s->code_ptr, R_386_PC8, label_index, -1);
|
472 |
s->code_ptr += 1;
|
473 |
} else {
|
474 |
if (opc == -1) { |
475 |
tcg_out8(s, OPC_JMP_long); |
476 |
} else {
|
477 |
tcg_out_opc(s, OPC_JCC_long + opc); |
478 |
} |
479 |
tcg_out_reloc(s, s->code_ptr, R_386_PC32, label_index, -4);
|
480 |
s->code_ptr += 4;
|
481 |
} |
482 |
} |
483 |
|
484 |
static void tcg_out_cmp(TCGContext *s, TCGArg arg1, TCGArg arg2, |
485 |
int const_arg2)
|
486 |
{ |
487 |
if (const_arg2) {
|
488 |
if (arg2 == 0) { |
489 |
/* test r, r */
|
490 |
tcg_out_modrm(s, OPC_TESTL, arg1, arg1); |
491 |
} else {
|
492 |
tgen_arithi(s, ARITH_CMP, arg1, arg2, 0);
|
493 |
} |
494 |
} else {
|
495 |
tgen_arithr(s, ARITH_CMP, arg1, arg2); |
496 |
} |
497 |
} |
498 |
|
499 |
static void tcg_out_brcond(TCGContext *s, TCGCond cond, |
500 |
TCGArg arg1, TCGArg arg2, int const_arg2,
|
501 |
int label_index, int small) |
502 |
{ |
503 |
tcg_out_cmp(s, arg1, arg2, const_arg2); |
504 |
tcg_out_jxx(s, tcg_cond_to_jcc[cond], label_index, small); |
505 |
} |
506 |
|
507 |
/* XXX: we implement it at the target level to avoid having to
|
508 |
handle cross basic blocks temporaries */
|
509 |
static void tcg_out_brcond2(TCGContext *s, const TCGArg *args, |
510 |
const int *const_args, int small) |
511 |
{ |
512 |
int label_next;
|
513 |
label_next = gen_new_label(); |
514 |
switch(args[4]) { |
515 |
case TCG_COND_EQ:
|
516 |
tcg_out_brcond(s, TCG_COND_NE, args[0], args[2], const_args[2], |
517 |
label_next, 1);
|
518 |
tcg_out_brcond(s, TCG_COND_EQ, args[1], args[3], const_args[3], |
519 |
args[5], small);
|
520 |
break;
|
521 |
case TCG_COND_NE:
|
522 |
tcg_out_brcond(s, TCG_COND_NE, args[0], args[2], const_args[2], |
523 |
args[5], small);
|
524 |
tcg_out_brcond(s, TCG_COND_NE, args[1], args[3], const_args[3], |
525 |
args[5], small);
|
526 |
break;
|
527 |
case TCG_COND_LT:
|
528 |
tcg_out_brcond(s, TCG_COND_LT, args[1], args[3], const_args[3], |
529 |
args[5], small);
|
530 |
tcg_out_jxx(s, JCC_JNE, label_next, 1);
|
531 |
tcg_out_brcond(s, TCG_COND_LTU, args[0], args[2], const_args[2], |
532 |
args[5], small);
|
533 |
break;
|
534 |
case TCG_COND_LE:
|
535 |
tcg_out_brcond(s, TCG_COND_LT, args[1], args[3], const_args[3], |
536 |
args[5], small);
|
537 |
tcg_out_jxx(s, JCC_JNE, label_next, 1);
|
538 |
tcg_out_brcond(s, TCG_COND_LEU, args[0], args[2], const_args[2], |
539 |
args[5], small);
|
540 |
break;
|
541 |
case TCG_COND_GT:
|
542 |
tcg_out_brcond(s, TCG_COND_GT, args[1], args[3], const_args[3], |
543 |
args[5], small);
|
544 |
tcg_out_jxx(s, JCC_JNE, label_next, 1);
|
545 |
tcg_out_brcond(s, TCG_COND_GTU, args[0], args[2], const_args[2], |
546 |
args[5], small);
|
547 |
break;
|
548 |
case TCG_COND_GE:
|
549 |
tcg_out_brcond(s, TCG_COND_GT, args[1], args[3], const_args[3], |
550 |
args[5], small);
|
551 |
tcg_out_jxx(s, JCC_JNE, label_next, 1);
|
552 |
tcg_out_brcond(s, TCG_COND_GEU, args[0], args[2], const_args[2], |
553 |
args[5], small);
|
554 |
break;
|
555 |
case TCG_COND_LTU:
|
556 |
tcg_out_brcond(s, TCG_COND_LTU, args[1], args[3], const_args[3], |
557 |
args[5], small);
|
558 |
tcg_out_jxx(s, JCC_JNE, label_next, 1);
|
559 |
tcg_out_brcond(s, TCG_COND_LTU, args[0], args[2], const_args[2], |
560 |
args[5], small);
|
561 |
break;
|
562 |
case TCG_COND_LEU:
|
563 |
tcg_out_brcond(s, TCG_COND_LTU, args[1], args[3], const_args[3], |
564 |
args[5], small);
|
565 |
tcg_out_jxx(s, JCC_JNE, label_next, 1);
|
566 |
tcg_out_brcond(s, TCG_COND_LEU, args[0], args[2], const_args[2], |
567 |
args[5], small);
|
568 |
break;
|
569 |
case TCG_COND_GTU:
|
570 |
tcg_out_brcond(s, TCG_COND_GTU, args[1], args[3], const_args[3], |
571 |
args[5], small);
|
572 |
tcg_out_jxx(s, JCC_JNE, label_next, 1);
|
573 |
tcg_out_brcond(s, TCG_COND_GTU, args[0], args[2], const_args[2], |
574 |
args[5], small);
|
575 |
break;
|
576 |
case TCG_COND_GEU:
|
577 |
tcg_out_brcond(s, TCG_COND_GTU, args[1], args[3], const_args[3], |
578 |
args[5], small);
|
579 |
tcg_out_jxx(s, JCC_JNE, label_next, 1);
|
580 |
tcg_out_brcond(s, TCG_COND_GEU, args[0], args[2], const_args[2], |
581 |
args[5], small);
|
582 |
break;
|
583 |
default:
|
584 |
tcg_abort(); |
585 |
} |
586 |
tcg_out_label(s, label_next, (tcg_target_long)s->code_ptr); |
587 |
} |
588 |
|
589 |
static void tcg_out_setcond(TCGContext *s, TCGCond cond, TCGArg dest, |
590 |
TCGArg arg1, TCGArg arg2, int const_arg2)
|
591 |
{ |
592 |
tcg_out_cmp(s, arg1, arg2, const_arg2); |
593 |
tcg_out_modrm(s, OPC_SETCC | tcg_cond_to_jcc[cond], 0, dest);
|
594 |
tcg_out_ext8u(s, dest, dest); |
595 |
} |
596 |
|
597 |
static void tcg_out_setcond2(TCGContext *s, const TCGArg *args, |
598 |
const int *const_args) |
599 |
{ |
600 |
TCGArg new_args[6];
|
601 |
int label_true, label_over;
|
602 |
|
603 |
memcpy(new_args, args+1, 5*sizeof(TCGArg)); |
604 |
|
605 |
if (args[0] == args[1] || args[0] == args[2] |
606 |
|| (!const_args[3] && args[0] == args[3]) |
607 |
|| (!const_args[4] && args[0] == args[4])) { |
608 |
/* When the destination overlaps with one of the argument
|
609 |
registers, don't do anything tricky. */
|
610 |
label_true = gen_new_label(); |
611 |
label_over = gen_new_label(); |
612 |
|
613 |
new_args[5] = label_true;
|
614 |
tcg_out_brcond2(s, new_args, const_args+1, 1); |
615 |
|
616 |
tcg_out_movi(s, TCG_TYPE_I32, args[0], 0); |
617 |
tcg_out_jxx(s, JCC_JMP, label_over, 1);
|
618 |
tcg_out_label(s, label_true, (tcg_target_long)s->code_ptr); |
619 |
|
620 |
tcg_out_movi(s, TCG_TYPE_I32, args[0], 1); |
621 |
tcg_out_label(s, label_over, (tcg_target_long)s->code_ptr); |
622 |
} else {
|
623 |
/* When the destination does not overlap one of the arguments,
|
624 |
clear the destination first, jump if cond false, and emit an
|
625 |
increment in the true case. This results in smaller code. */
|
626 |
|
627 |
tcg_out_movi(s, TCG_TYPE_I32, args[0], 0); |
628 |
|
629 |
label_over = gen_new_label(); |
630 |
new_args[4] = tcg_invert_cond(new_args[4]); |
631 |
new_args[5] = label_over;
|
632 |
tcg_out_brcond2(s, new_args, const_args+1, 1); |
633 |
|
634 |
tgen_arithi(s, ARITH_ADD, args[0], 1, 0); |
635 |
tcg_out_label(s, label_over, (tcg_target_long)s->code_ptr); |
636 |
} |
637 |
} |
638 |
|
639 |
static void tcg_out_calli(TCGContext *s, tcg_target_long dest) |
640 |
{ |
641 |
tcg_out_opc(s, OPC_CALL_Jz); |
642 |
tcg_out32(s, dest - (tcg_target_long)s->code_ptr - 4);
|
643 |
} |
644 |
|
645 |
#if defined(CONFIG_SOFTMMU)
|
646 |
|
647 |
#include "../../softmmu_defs.h" |
648 |
|
649 |
static void *qemu_ld_helpers[4] = { |
650 |
__ldb_mmu, |
651 |
__ldw_mmu, |
652 |
__ldl_mmu, |
653 |
__ldq_mmu, |
654 |
}; |
655 |
|
656 |
static void *qemu_st_helpers[4] = { |
657 |
__stb_mmu, |
658 |
__stw_mmu, |
659 |
__stl_mmu, |
660 |
__stq_mmu, |
661 |
}; |
662 |
#endif
|
663 |
|
664 |
#ifndef CONFIG_USER_ONLY
|
665 |
#define GUEST_BASE 0 |
666 |
#endif
|
667 |
|
668 |
/* XXX: qemu_ld and qemu_st could be modified to clobber only EDX and
|
669 |
EAX. It will be useful once fixed registers globals are less
|
670 |
common. */
|
671 |
static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, |
672 |
int opc)
|
673 |
{ |
674 |
int addr_reg, data_reg, data_reg2, r0, r1, mem_index, s_bits, bswap;
|
675 |
#if defined(CONFIG_SOFTMMU)
|
676 |
uint8_t *label1_ptr, *label2_ptr; |
677 |
#endif
|
678 |
#if TARGET_LONG_BITS == 64 |
679 |
#if defined(CONFIG_SOFTMMU)
|
680 |
uint8_t *label3_ptr; |
681 |
#endif
|
682 |
int addr_reg2;
|
683 |
#endif
|
684 |
|
685 |
data_reg = *args++; |
686 |
if (opc == 3) |
687 |
data_reg2 = *args++; |
688 |
else
|
689 |
data_reg2 = 0;
|
690 |
addr_reg = *args++; |
691 |
#if TARGET_LONG_BITS == 64 |
692 |
addr_reg2 = *args++; |
693 |
#endif
|
694 |
mem_index = *args; |
695 |
s_bits = opc & 3;
|
696 |
|
697 |
r0 = TCG_REG_EAX; |
698 |
r1 = TCG_REG_EDX; |
699 |
|
700 |
#if defined(CONFIG_SOFTMMU)
|
701 |
tcg_out_mov(s, r1, addr_reg); |
702 |
tcg_out_mov(s, r0, addr_reg); |
703 |
|
704 |
tcg_out_shifti(s, SHIFT_SHR, r1, TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS); |
705 |
|
706 |
tgen_arithi(s, ARITH_AND, r0, TARGET_PAGE_MASK | ((1 << s_bits) - 1), 0); |
707 |
tgen_arithi(s, ARITH_AND, r1, (CPU_TLB_SIZE - 1) << CPU_TLB_ENTRY_BITS, 0); |
708 |
|
709 |
tcg_out_opc(s, 0x8d); /* lea offset(r1, %ebp), r1 */ |
710 |
tcg_out8(s, 0x80 | (r1 << 3) | 0x04); |
711 |
tcg_out8(s, (5 << 3) | r1); |
712 |
tcg_out32(s, offsetof(CPUState, tlb_table[mem_index][0].addr_read));
|
713 |
|
714 |
/* cmp 0(r1), r0 */
|
715 |
tcg_out_modrm_offset(s, OPC_CMP_GvEv, r0, r1, 0);
|
716 |
|
717 |
tcg_out_mov(s, r0, addr_reg); |
718 |
|
719 |
#if TARGET_LONG_BITS == 32 |
720 |
/* je label1 */
|
721 |
tcg_out8(s, OPC_JCC_short + JCC_JE); |
722 |
label1_ptr = s->code_ptr; |
723 |
s->code_ptr++; |
724 |
#else
|
725 |
/* jne label3 */
|
726 |
tcg_out8(s, OPC_JCC_short + JCC_JNE); |
727 |
label3_ptr = s->code_ptr; |
728 |
s->code_ptr++; |
729 |
|
730 |
/* cmp 4(r1), addr_reg2 */
|
731 |
tcg_out_modrm_offset(s, OPC_CMP_GvEv, addr_reg2, r1, 4);
|
732 |
|
733 |
/* je label1 */
|
734 |
tcg_out8(s, OPC_JCC_short + JCC_JE); |
735 |
label1_ptr = s->code_ptr; |
736 |
s->code_ptr++; |
737 |
|
738 |
/* label3: */
|
739 |
*label3_ptr = s->code_ptr - label3_ptr - 1;
|
740 |
#endif
|
741 |
|
742 |
/* XXX: move that code at the end of the TB */
|
743 |
#if TARGET_LONG_BITS == 32 |
744 |
tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_EDX, mem_index); |
745 |
#else
|
746 |
tcg_out_mov(s, TCG_REG_EDX, addr_reg2); |
747 |
tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_ECX, mem_index); |
748 |
#endif
|
749 |
tcg_out_calli(s, (tcg_target_long)qemu_ld_helpers[s_bits]); |
750 |
|
751 |
switch(opc) {
|
752 |
case 0 | 4: |
753 |
tcg_out_ext8s(s, data_reg, TCG_REG_EAX); |
754 |
break;
|
755 |
case 1 | 4: |
756 |
tcg_out_ext16s(s, data_reg, TCG_REG_EAX); |
757 |
break;
|
758 |
case 0: |
759 |
tcg_out_ext8u(s, data_reg, TCG_REG_EAX); |
760 |
break;
|
761 |
case 1: |
762 |
tcg_out_ext16u(s, data_reg, TCG_REG_EAX); |
763 |
break;
|
764 |
case 2: |
765 |
default:
|
766 |
tcg_out_mov(s, data_reg, TCG_REG_EAX); |
767 |
break;
|
768 |
case 3: |
769 |
if (data_reg == TCG_REG_EDX) {
|
770 |
tcg_out_opc(s, 0x90 + TCG_REG_EDX); /* xchg %edx, %eax */ |
771 |
tcg_out_mov(s, data_reg2, TCG_REG_EAX); |
772 |
} else {
|
773 |
tcg_out_mov(s, data_reg, TCG_REG_EAX); |
774 |
tcg_out_mov(s, data_reg2, TCG_REG_EDX); |
775 |
} |
776 |
break;
|
777 |
} |
778 |
|
779 |
/* jmp label2 */
|
780 |
tcg_out8(s, OPC_JMP_short); |
781 |
label2_ptr = s->code_ptr; |
782 |
s->code_ptr++; |
783 |
|
784 |
/* label1: */
|
785 |
*label1_ptr = s->code_ptr - label1_ptr - 1;
|
786 |
|
787 |
/* add x(r1), r0 */
|
788 |
tcg_out_modrm_offset(s, OPC_ADD_GvEv, r0, r1, |
789 |
offsetof(CPUTLBEntry, addend) - |
790 |
offsetof(CPUTLBEntry, addr_read)); |
791 |
#else
|
792 |
r0 = addr_reg; |
793 |
#endif
|
794 |
|
795 |
#ifdef TARGET_WORDS_BIGENDIAN
|
796 |
bswap = 1;
|
797 |
#else
|
798 |
bswap = 0;
|
799 |
#endif
|
800 |
switch(opc) {
|
801 |
case 0: |
802 |
/* movzbl */
|
803 |
tcg_out_modrm_offset(s, OPC_MOVZBL, data_reg, r0, GUEST_BASE); |
804 |
break;
|
805 |
case 0 | 4: |
806 |
/* movsbl */
|
807 |
tcg_out_modrm_offset(s, OPC_MOVSBL, data_reg, r0, GUEST_BASE); |
808 |
break;
|
809 |
case 1: |
810 |
/* movzwl */
|
811 |
tcg_out_modrm_offset(s, OPC_MOVZWL, data_reg, r0, GUEST_BASE); |
812 |
if (bswap) {
|
813 |
tcg_out_rolw_8(s, data_reg); |
814 |
} |
815 |
break;
|
816 |
case 1 | 4: |
817 |
/* movswl */
|
818 |
tcg_out_modrm_offset(s, OPC_MOVSWL, data_reg, r0, GUEST_BASE); |
819 |
if (bswap) {
|
820 |
tcg_out_rolw_8(s, data_reg); |
821 |
|
822 |
/* movswl data_reg, data_reg */
|
823 |
tcg_out_modrm(s, OPC_MOVSWL, data_reg, data_reg); |
824 |
} |
825 |
break;
|
826 |
case 2: |
827 |
tcg_out_ld(s, TCG_TYPE_I32, data_reg, r0, GUEST_BASE); |
828 |
if (bswap) {
|
829 |
tcg_out_bswap32(s, data_reg); |
830 |
} |
831 |
break;
|
832 |
case 3: |
833 |
if (bswap) {
|
834 |
int t = data_reg;
|
835 |
data_reg = data_reg2; |
836 |
data_reg2 = t; |
837 |
} |
838 |
if (r0 != data_reg) {
|
839 |
tcg_out_ld(s, TCG_TYPE_I32, data_reg, r0, GUEST_BASE); |
840 |
tcg_out_ld(s, TCG_TYPE_I32, data_reg2, r0, GUEST_BASE + 4);
|
841 |
} else {
|
842 |
tcg_out_ld(s, TCG_TYPE_I32, data_reg2, r0, GUEST_BASE + 4);
|
843 |
tcg_out_ld(s, TCG_TYPE_I32, data_reg, r0, GUEST_BASE); |
844 |
} |
845 |
if (bswap) {
|
846 |
tcg_out_bswap32(s, data_reg); |
847 |
tcg_out_bswap32(s, data_reg2); |
848 |
} |
849 |
break;
|
850 |
default:
|
851 |
tcg_abort(); |
852 |
} |
853 |
|
854 |
#if defined(CONFIG_SOFTMMU)
|
855 |
/* label2: */
|
856 |
*label2_ptr = s->code_ptr - label2_ptr - 1;
|
857 |
#endif
|
858 |
} |
859 |
|
860 |
|
861 |
static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, |
862 |
int opc)
|
863 |
{ |
864 |
int addr_reg, data_reg, data_reg2, r0, r1, mem_index, s_bits, bswap;
|
865 |
#if defined(CONFIG_SOFTMMU)
|
866 |
int stack_adjust;
|
867 |
uint8_t *label1_ptr, *label2_ptr; |
868 |
#endif
|
869 |
#if TARGET_LONG_BITS == 64 |
870 |
#if defined(CONFIG_SOFTMMU)
|
871 |
uint8_t *label3_ptr; |
872 |
#endif
|
873 |
int addr_reg2;
|
874 |
#endif
|
875 |
|
876 |
data_reg = *args++; |
877 |
if (opc == 3) |
878 |
data_reg2 = *args++; |
879 |
else
|
880 |
data_reg2 = 0;
|
881 |
addr_reg = *args++; |
882 |
#if TARGET_LONG_BITS == 64 |
883 |
addr_reg2 = *args++; |
884 |
#endif
|
885 |
mem_index = *args; |
886 |
|
887 |
s_bits = opc; |
888 |
|
889 |
r0 = TCG_REG_EAX; |
890 |
r1 = TCG_REG_EDX; |
891 |
|
892 |
#if defined(CONFIG_SOFTMMU)
|
893 |
tcg_out_mov(s, r1, addr_reg); |
894 |
tcg_out_mov(s, r0, addr_reg); |
895 |
|
896 |
tcg_out_shifti(s, SHIFT_SHR, r1, TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS); |
897 |
|
898 |
tgen_arithi(s, ARITH_AND, r0, TARGET_PAGE_MASK | ((1 << s_bits) - 1), 0); |
899 |
tgen_arithi(s, ARITH_AND, r1, (CPU_TLB_SIZE - 1) << CPU_TLB_ENTRY_BITS, 0); |
900 |
|
901 |
tcg_out_opc(s, 0x8d); /* lea offset(r1, %ebp), r1 */ |
902 |
tcg_out8(s, 0x80 | (r1 << 3) | 0x04); |
903 |
tcg_out8(s, (5 << 3) | r1); |
904 |
tcg_out32(s, offsetof(CPUState, tlb_table[mem_index][0].addr_write));
|
905 |
|
906 |
/* cmp 0(r1), r0 */
|
907 |
tcg_out_modrm_offset(s, OPC_CMP_GvEv, r0, r1, 0);
|
908 |
|
909 |
tcg_out_mov(s, r0, addr_reg); |
910 |
|
911 |
#if TARGET_LONG_BITS == 32 |
912 |
/* je label1 */
|
913 |
tcg_out8(s, OPC_JCC_short + JCC_JE); |
914 |
label1_ptr = s->code_ptr; |
915 |
s->code_ptr++; |
916 |
#else
|
917 |
/* jne label3 */
|
918 |
tcg_out8(s, OPC_JCC_short + JCC_JNE); |
919 |
label3_ptr = s->code_ptr; |
920 |
s->code_ptr++; |
921 |
|
922 |
/* cmp 4(r1), addr_reg2 */
|
923 |
tcg_out_modrm_offset(s, OPC_CMP_GvEv, addr_reg2, r1, 4);
|
924 |
|
925 |
/* je label1 */
|
926 |
tcg_out8(s, OPC_JCC_short + JCC_JE); |
927 |
label1_ptr = s->code_ptr; |
928 |
s->code_ptr++; |
929 |
|
930 |
/* label3: */
|
931 |
*label3_ptr = s->code_ptr - label3_ptr - 1;
|
932 |
#endif
|
933 |
|
934 |
/* XXX: move that code at the end of the TB */
|
935 |
#if TARGET_LONG_BITS == 32 |
936 |
if (opc == 3) { |
937 |
tcg_out_mov(s, TCG_REG_EDX, data_reg); |
938 |
tcg_out_mov(s, TCG_REG_ECX, data_reg2); |
939 |
tcg_out_pushi(s, mem_index); |
940 |
stack_adjust = 4;
|
941 |
} else {
|
942 |
switch(opc) {
|
943 |
case 0: |
944 |
tcg_out_ext8u(s, TCG_REG_EDX, data_reg); |
945 |
break;
|
946 |
case 1: |
947 |
tcg_out_ext16u(s, TCG_REG_EDX, data_reg); |
948 |
break;
|
949 |
case 2: |
950 |
tcg_out_mov(s, TCG_REG_EDX, data_reg); |
951 |
break;
|
952 |
} |
953 |
tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_ECX, mem_index); |
954 |
stack_adjust = 0;
|
955 |
} |
956 |
#else
|
957 |
if (opc == 3) { |
958 |
tcg_out_mov(s, TCG_REG_EDX, addr_reg2); |
959 |
tcg_out_pushi(s, mem_index); |
960 |
tcg_out_push(s, data_reg2); |
961 |
tcg_out_push(s, data_reg); |
962 |
stack_adjust = 12;
|
963 |
} else {
|
964 |
tcg_out_mov(s, TCG_REG_EDX, addr_reg2); |
965 |
switch(opc) {
|
966 |
case 0: |
967 |
tcg_out_ext8u(s, TCG_REG_ECX, data_reg); |
968 |
break;
|
969 |
case 1: |
970 |
tcg_out_ext16u(s, TCG_REG_ECX, data_reg); |
971 |
break;
|
972 |
case 2: |
973 |
tcg_out_mov(s, TCG_REG_ECX, data_reg); |
974 |
break;
|
975 |
} |
976 |
tcg_out_pushi(s, mem_index); |
977 |
stack_adjust = 4;
|
978 |
} |
979 |
#endif
|
980 |
|
981 |
tcg_out_calli(s, (tcg_target_long)qemu_st_helpers[s_bits]); |
982 |
|
983 |
if (stack_adjust == 4) { |
984 |
/* Pop and discard. This is 2 bytes smaller than the add. */
|
985 |
tcg_out_pop(s, TCG_REG_ECX); |
986 |
} else if (stack_adjust != 0) { |
987 |
tcg_out_addi(s, TCG_REG_ESP, stack_adjust); |
988 |
} |
989 |
|
990 |
/* jmp label2 */
|
991 |
tcg_out8(s, OPC_JMP_short); |
992 |
label2_ptr = s->code_ptr; |
993 |
s->code_ptr++; |
994 |
|
995 |
/* label1: */
|
996 |
*label1_ptr = s->code_ptr - label1_ptr - 1;
|
997 |
|
998 |
/* add x(r1), r0 */
|
999 |
tcg_out_modrm_offset(s, OPC_ADD_GvEv, r0, r1, |
1000 |
offsetof(CPUTLBEntry, addend) - |
1001 |
offsetof(CPUTLBEntry, addr_write)); |
1002 |
#else
|
1003 |
r0 = addr_reg; |
1004 |
#endif
|
1005 |
|
1006 |
#ifdef TARGET_WORDS_BIGENDIAN
|
1007 |
bswap = 1;
|
1008 |
#else
|
1009 |
bswap = 0;
|
1010 |
#endif
|
1011 |
switch(opc) {
|
1012 |
case 0: |
1013 |
tcg_out_modrm_offset(s, OPC_MOVB_EvGv, data_reg, r0, GUEST_BASE); |
1014 |
break;
|
1015 |
case 1: |
1016 |
if (bswap) {
|
1017 |
tcg_out_mov(s, r1, data_reg); |
1018 |
tcg_out_rolw_8(s, r1); |
1019 |
data_reg = r1; |
1020 |
} |
1021 |
/* movw */
|
1022 |
tcg_out8(s, 0x66);
|
1023 |
tcg_out_modrm_offset(s, OPC_MOVL_EvGv, data_reg, r0, GUEST_BASE); |
1024 |
break;
|
1025 |
case 2: |
1026 |
if (bswap) {
|
1027 |
tcg_out_mov(s, r1, data_reg); |
1028 |
tcg_out_bswap32(s, r1); |
1029 |
data_reg = r1; |
1030 |
} |
1031 |
tcg_out_st(s, TCG_TYPE_I32, data_reg, r0, GUEST_BASE); |
1032 |
break;
|
1033 |
case 3: |
1034 |
if (bswap) {
|
1035 |
tcg_out_mov(s, r1, data_reg2); |
1036 |
tcg_out_bswap32(s, r1); |
1037 |
tcg_out_st(s, TCG_TYPE_I32, r1, r0, GUEST_BASE); |
1038 |
tcg_out_mov(s, r1, data_reg); |
1039 |
tcg_out_bswap32(s, r1); |
1040 |
tcg_out_st(s, TCG_TYPE_I32, r1, r0, GUEST_BASE + 4);
|
1041 |
} else {
|
1042 |
tcg_out_st(s, TCG_TYPE_I32, data_reg, r0, GUEST_BASE); |
1043 |
tcg_out_st(s, TCG_TYPE_I32, data_reg2, r0, GUEST_BASE + 4);
|
1044 |
} |
1045 |
break;
|
1046 |
default:
|
1047 |
tcg_abort(); |
1048 |
} |
1049 |
|
1050 |
#if defined(CONFIG_SOFTMMU)
|
1051 |
/* label2: */
|
1052 |
*label2_ptr = s->code_ptr - label2_ptr - 1;
|
1053 |
#endif
|
1054 |
} |
1055 |
|
1056 |
static inline void tcg_out_op(TCGContext *s, TCGOpcode opc, |
1057 |
const TCGArg *args, const int *const_args) |
1058 |
{ |
1059 |
int c;
|
1060 |
|
1061 |
switch(opc) {
|
1062 |
case INDEX_op_exit_tb:
|
1063 |
tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_EAX, args[0]);
|
1064 |
tcg_out8(s, OPC_JMP_long); /* jmp tb_ret_addr */
|
1065 |
tcg_out32(s, tb_ret_addr - s->code_ptr - 4);
|
1066 |
break;
|
1067 |
case INDEX_op_goto_tb:
|
1068 |
if (s->tb_jmp_offset) {
|
1069 |
/* direct jump method */
|
1070 |
tcg_out8(s, OPC_JMP_long); /* jmp im */
|
1071 |
s->tb_jmp_offset[args[0]] = s->code_ptr - s->code_buf;
|
1072 |
tcg_out32(s, 0);
|
1073 |
} else {
|
1074 |
/* indirect jump method */
|
1075 |
tcg_out_modrm_offset(s, OPC_GRP5, EXT5_JMPN_Ev, -1,
|
1076 |
(tcg_target_long)(s->tb_next + args[0]));
|
1077 |
} |
1078 |
s->tb_next_offset[args[0]] = s->code_ptr - s->code_buf;
|
1079 |
break;
|
1080 |
case INDEX_op_call:
|
1081 |
if (const_args[0]) { |
1082 |
tcg_out_calli(s, args[0]);
|
1083 |
} else {
|
1084 |
/* call *reg */
|
1085 |
tcg_out_modrm(s, OPC_GRP5, EXT5_CALLN_Ev, args[0]);
|
1086 |
} |
1087 |
break;
|
1088 |
case INDEX_op_jmp:
|
1089 |
if (const_args[0]) { |
1090 |
tcg_out8(s, OPC_JMP_long); |
1091 |
tcg_out32(s, args[0] - (tcg_target_long)s->code_ptr - 4); |
1092 |
} else {
|
1093 |
/* jmp *reg */
|
1094 |
tcg_out_modrm(s, OPC_GRP5, EXT5_JMPN_Ev, args[0]);
|
1095 |
} |
1096 |
break;
|
1097 |
case INDEX_op_br:
|
1098 |
tcg_out_jxx(s, JCC_JMP, args[0], 0); |
1099 |
break;
|
1100 |
case INDEX_op_movi_i32:
|
1101 |
tcg_out_movi(s, TCG_TYPE_I32, args[0], args[1]); |
1102 |
break;
|
1103 |
case INDEX_op_ld8u_i32:
|
1104 |
/* movzbl */
|
1105 |
tcg_out_modrm_offset(s, OPC_MOVZBL, args[0], args[1], args[2]); |
1106 |
break;
|
1107 |
case INDEX_op_ld8s_i32:
|
1108 |
/* movsbl */
|
1109 |
tcg_out_modrm_offset(s, OPC_MOVSBL, args[0], args[1], args[2]); |
1110 |
break;
|
1111 |
case INDEX_op_ld16u_i32:
|
1112 |
/* movzwl */
|
1113 |
tcg_out_modrm_offset(s, OPC_MOVZWL, args[0], args[1], args[2]); |
1114 |
break;
|
1115 |
case INDEX_op_ld16s_i32:
|
1116 |
/* movswl */
|
1117 |
tcg_out_modrm_offset(s, OPC_MOVSWL, args[0], args[1], args[2]); |
1118 |
break;
|
1119 |
case INDEX_op_ld_i32:
|
1120 |
tcg_out_ld(s, TCG_TYPE_I32, args[0], args[1], args[2]); |
1121 |
break;
|
1122 |
case INDEX_op_st8_i32:
|
1123 |
/* movb */
|
1124 |
tcg_out_modrm_offset(s, OPC_MOVB_EvGv, args[0], args[1], args[2]); |
1125 |
break;
|
1126 |
case INDEX_op_st16_i32:
|
1127 |
/* movw */
|
1128 |
tcg_out8(s, 0x66);
|
1129 |
tcg_out_modrm_offset(s, OPC_MOVL_EvGv, args[0], args[1], args[2]); |
1130 |
break;
|
1131 |
case INDEX_op_st_i32:
|
1132 |
tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]); |
1133 |
break;
|
1134 |
case INDEX_op_sub_i32:
|
1135 |
c = ARITH_SUB; |
1136 |
goto gen_arith;
|
1137 |
case INDEX_op_and_i32:
|
1138 |
c = ARITH_AND; |
1139 |
goto gen_arith;
|
1140 |
case INDEX_op_or_i32:
|
1141 |
c = ARITH_OR; |
1142 |
goto gen_arith;
|
1143 |
case INDEX_op_xor_i32:
|
1144 |
c = ARITH_XOR; |
1145 |
goto gen_arith;
|
1146 |
case INDEX_op_add_i32:
|
1147 |
c = ARITH_ADD; |
1148 |
gen_arith:
|
1149 |
if (const_args[2]) { |
1150 |
tgen_arithi(s, c, args[0], args[2], 0); |
1151 |
} else {
|
1152 |
tgen_arithr(s, c, args[0], args[2]); |
1153 |
} |
1154 |
break;
|
1155 |
case INDEX_op_mul_i32:
|
1156 |
if (const_args[2]) { |
1157 |
int32_t val; |
1158 |
val = args[2];
|
1159 |
if (val == (int8_t)val) {
|
1160 |
tcg_out_modrm(s, 0x6b, args[0], args[0]); |
1161 |
tcg_out8(s, val); |
1162 |
} else {
|
1163 |
tcg_out_modrm(s, 0x69, args[0], args[0]); |
1164 |
tcg_out32(s, val); |
1165 |
} |
1166 |
} else {
|
1167 |
tcg_out_modrm(s, 0xaf | P_EXT, args[0], args[2]); |
1168 |
} |
1169 |
break;
|
1170 |
case INDEX_op_mulu2_i32:
|
1171 |
tcg_out_modrm(s, OPC_GRP3_Ev, EXT3_MUL, args[3]);
|
1172 |
break;
|
1173 |
case INDEX_op_div2_i32:
|
1174 |
tcg_out_modrm(s, OPC_GRP3_Ev, EXT3_IDIV, args[4]);
|
1175 |
break;
|
1176 |
case INDEX_op_divu2_i32:
|
1177 |
tcg_out_modrm(s, OPC_GRP3_Ev, EXT3_DIV, args[4]);
|
1178 |
break;
|
1179 |
case INDEX_op_shl_i32:
|
1180 |
c = SHIFT_SHL; |
1181 |
gen_shift32:
|
1182 |
if (const_args[2]) { |
1183 |
tcg_out_shifti(s, c, args[0], args[2]); |
1184 |
} else {
|
1185 |
tcg_out_modrm(s, OPC_SHIFT_cl, c, args[0]);
|
1186 |
} |
1187 |
break;
|
1188 |
case INDEX_op_shr_i32:
|
1189 |
c = SHIFT_SHR; |
1190 |
goto gen_shift32;
|
1191 |
case INDEX_op_sar_i32:
|
1192 |
c = SHIFT_SAR; |
1193 |
goto gen_shift32;
|
1194 |
case INDEX_op_rotl_i32:
|
1195 |
c = SHIFT_ROL; |
1196 |
goto gen_shift32;
|
1197 |
case INDEX_op_rotr_i32:
|
1198 |
c = SHIFT_ROR; |
1199 |
goto gen_shift32;
|
1200 |
|
1201 |
case INDEX_op_add2_i32:
|
1202 |
if (const_args[4]) { |
1203 |
tgen_arithi(s, ARITH_ADD, args[0], args[4], 1); |
1204 |
} else {
|
1205 |
tgen_arithr(s, ARITH_ADD, args[0], args[4]); |
1206 |
} |
1207 |
if (const_args[5]) { |
1208 |
tgen_arithi(s, ARITH_ADC, args[1], args[5], 1); |
1209 |
} else {
|
1210 |
tgen_arithr(s, ARITH_ADC, args[1], args[5]); |
1211 |
} |
1212 |
break;
|
1213 |
case INDEX_op_sub2_i32:
|
1214 |
if (const_args[4]) { |
1215 |
tgen_arithi(s, ARITH_SUB, args[0], args[4], 1); |
1216 |
} else {
|
1217 |
tgen_arithr(s, ARITH_SUB, args[0], args[4]); |
1218 |
} |
1219 |
if (const_args[5]) { |
1220 |
tgen_arithi(s, ARITH_SBB, args[1], args[5], 1); |
1221 |
} else {
|
1222 |
tgen_arithr(s, ARITH_SBB, args[1], args[5]); |
1223 |
} |
1224 |
break;
|
1225 |
case INDEX_op_brcond_i32:
|
1226 |
tcg_out_brcond(s, args[2], args[0], args[1], const_args[1], |
1227 |
args[3], 0); |
1228 |
break;
|
1229 |
case INDEX_op_brcond2_i32:
|
1230 |
tcg_out_brcond2(s, args, const_args, 0);
|
1231 |
break;
|
1232 |
|
1233 |
case INDEX_op_bswap16_i32:
|
1234 |
tcg_out_rolw_8(s, args[0]);
|
1235 |
break;
|
1236 |
case INDEX_op_bswap32_i32:
|
1237 |
tcg_out_bswap32(s, args[0]);
|
1238 |
break;
|
1239 |
|
1240 |
case INDEX_op_neg_i32:
|
1241 |
tcg_out_modrm(s, OPC_GRP3_Ev, EXT3_NEG, args[0]);
|
1242 |
break;
|
1243 |
|
1244 |
case INDEX_op_not_i32:
|
1245 |
tcg_out_modrm(s, OPC_GRP3_Ev, EXT3_NOT, args[0]);
|
1246 |
break;
|
1247 |
|
1248 |
case INDEX_op_ext8s_i32:
|
1249 |
tcg_out_ext8s(s, args[0], args[1]); |
1250 |
break;
|
1251 |
case INDEX_op_ext16s_i32:
|
1252 |
tcg_out_ext16s(s, args[0], args[1]); |
1253 |
break;
|
1254 |
case INDEX_op_ext8u_i32:
|
1255 |
tcg_out_ext8u(s, args[0], args[1]); |
1256 |
break;
|
1257 |
case INDEX_op_ext16u_i32:
|
1258 |
tcg_out_ext16u(s, args[0], args[1]); |
1259 |
break;
|
1260 |
|
1261 |
case INDEX_op_setcond_i32:
|
1262 |
tcg_out_setcond(s, args[3], args[0], args[1], args[2], const_args[2]); |
1263 |
break;
|
1264 |
case INDEX_op_setcond2_i32:
|
1265 |
tcg_out_setcond2(s, args, const_args); |
1266 |
break;
|
1267 |
|
1268 |
case INDEX_op_qemu_ld8u:
|
1269 |
tcg_out_qemu_ld(s, args, 0);
|
1270 |
break;
|
1271 |
case INDEX_op_qemu_ld8s:
|
1272 |
tcg_out_qemu_ld(s, args, 0 | 4); |
1273 |
break;
|
1274 |
case INDEX_op_qemu_ld16u:
|
1275 |
tcg_out_qemu_ld(s, args, 1);
|
1276 |
break;
|
1277 |
case INDEX_op_qemu_ld16s:
|
1278 |
tcg_out_qemu_ld(s, args, 1 | 4); |
1279 |
break;
|
1280 |
case INDEX_op_qemu_ld32:
|
1281 |
tcg_out_qemu_ld(s, args, 2);
|
1282 |
break;
|
1283 |
case INDEX_op_qemu_ld64:
|
1284 |
tcg_out_qemu_ld(s, args, 3);
|
1285 |
break;
|
1286 |
|
1287 |
case INDEX_op_qemu_st8:
|
1288 |
tcg_out_qemu_st(s, args, 0);
|
1289 |
break;
|
1290 |
case INDEX_op_qemu_st16:
|
1291 |
tcg_out_qemu_st(s, args, 1);
|
1292 |
break;
|
1293 |
case INDEX_op_qemu_st32:
|
1294 |
tcg_out_qemu_st(s, args, 2);
|
1295 |
break;
|
1296 |
case INDEX_op_qemu_st64:
|
1297 |
tcg_out_qemu_st(s, args, 3);
|
1298 |
break;
|
1299 |
|
1300 |
default:
|
1301 |
tcg_abort(); |
1302 |
} |
1303 |
} |
1304 |
|
1305 |
static const TCGTargetOpDef x86_op_defs[] = { |
1306 |
{ INDEX_op_exit_tb, { } }, |
1307 |
{ INDEX_op_goto_tb, { } }, |
1308 |
{ INDEX_op_call, { "ri" } },
|
1309 |
{ INDEX_op_jmp, { "ri" } },
|
1310 |
{ INDEX_op_br, { } }, |
1311 |
{ INDEX_op_mov_i32, { "r", "r" } }, |
1312 |
{ INDEX_op_movi_i32, { "r" } },
|
1313 |
{ INDEX_op_ld8u_i32, { "r", "r" } }, |
1314 |
{ INDEX_op_ld8s_i32, { "r", "r" } }, |
1315 |
{ INDEX_op_ld16u_i32, { "r", "r" } }, |
1316 |
{ INDEX_op_ld16s_i32, { "r", "r" } }, |
1317 |
{ INDEX_op_ld_i32, { "r", "r" } }, |
1318 |
{ INDEX_op_st8_i32, { "q", "r" } }, |
1319 |
{ INDEX_op_st16_i32, { "r", "r" } }, |
1320 |
{ INDEX_op_st_i32, { "r", "r" } }, |
1321 |
|
1322 |
{ INDEX_op_add_i32, { "r", "0", "ri" } }, |
1323 |
{ INDEX_op_sub_i32, { "r", "0", "ri" } }, |
1324 |
{ INDEX_op_mul_i32, { "r", "0", "ri" } }, |
1325 |
{ INDEX_op_mulu2_i32, { "a", "d", "a", "r" } }, |
1326 |
{ INDEX_op_div2_i32, { "a", "d", "0", "1", "r" } }, |
1327 |
{ INDEX_op_divu2_i32, { "a", "d", "0", "1", "r" } }, |
1328 |
{ INDEX_op_and_i32, { "r", "0", "ri" } }, |
1329 |
{ INDEX_op_or_i32, { "r", "0", "ri" } }, |
1330 |
{ INDEX_op_xor_i32, { "r", "0", "ri" } }, |
1331 |
|
1332 |
{ INDEX_op_shl_i32, { "r", "0", "ci" } }, |
1333 |
{ INDEX_op_shr_i32, { "r", "0", "ci" } }, |
1334 |
{ INDEX_op_sar_i32, { "r", "0", "ci" } }, |
1335 |
{ INDEX_op_rotl_i32, { "r", "0", "ci" } }, |
1336 |
{ INDEX_op_rotr_i32, { "r", "0", "ci" } }, |
1337 |
|
1338 |
{ INDEX_op_brcond_i32, { "r", "ri" } }, |
1339 |
|
1340 |
{ INDEX_op_add2_i32, { "r", "r", "0", "1", "ri", "ri" } }, |
1341 |
{ INDEX_op_sub2_i32, { "r", "r", "0", "1", "ri", "ri" } }, |
1342 |
{ INDEX_op_brcond2_i32, { "r", "r", "ri", "ri" } }, |
1343 |
|
1344 |
{ INDEX_op_bswap16_i32, { "r", "0" } }, |
1345 |
{ INDEX_op_bswap32_i32, { "r", "0" } }, |
1346 |
|
1347 |
{ INDEX_op_neg_i32, { "r", "0" } }, |
1348 |
|
1349 |
{ INDEX_op_not_i32, { "r", "0" } }, |
1350 |
|
1351 |
{ INDEX_op_ext8s_i32, { "r", "q" } }, |
1352 |
{ INDEX_op_ext16s_i32, { "r", "r" } }, |
1353 |
{ INDEX_op_ext8u_i32, { "r", "q" } }, |
1354 |
{ INDEX_op_ext16u_i32, { "r", "r" } }, |
1355 |
|
1356 |
{ INDEX_op_setcond_i32, { "q", "r", "ri" } }, |
1357 |
{ INDEX_op_setcond2_i32, { "r", "r", "r", "ri", "ri" } }, |
1358 |
|
1359 |
#if TARGET_LONG_BITS == 32 |
1360 |
{ INDEX_op_qemu_ld8u, { "r", "L" } }, |
1361 |
{ INDEX_op_qemu_ld8s, { "r", "L" } }, |
1362 |
{ INDEX_op_qemu_ld16u, { "r", "L" } }, |
1363 |
{ INDEX_op_qemu_ld16s, { "r", "L" } }, |
1364 |
{ INDEX_op_qemu_ld32, { "r", "L" } }, |
1365 |
{ INDEX_op_qemu_ld64, { "r", "r", "L" } }, |
1366 |
|
1367 |
{ INDEX_op_qemu_st8, { "cb", "L" } }, |
1368 |
{ INDEX_op_qemu_st16, { "L", "L" } }, |
1369 |
{ INDEX_op_qemu_st32, { "L", "L" } }, |
1370 |
{ INDEX_op_qemu_st64, { "L", "L", "L" } }, |
1371 |
#else
|
1372 |
{ INDEX_op_qemu_ld8u, { "r", "L", "L" } }, |
1373 |
{ INDEX_op_qemu_ld8s, { "r", "L", "L" } }, |
1374 |
{ INDEX_op_qemu_ld16u, { "r", "L", "L" } }, |
1375 |
{ INDEX_op_qemu_ld16s, { "r", "L", "L" } }, |
1376 |
{ INDEX_op_qemu_ld32, { "r", "L", "L" } }, |
1377 |
{ INDEX_op_qemu_ld64, { "r", "r", "L", "L" } }, |
1378 |
|
1379 |
{ INDEX_op_qemu_st8, { "cb", "L", "L" } }, |
1380 |
{ INDEX_op_qemu_st16, { "L", "L", "L" } }, |
1381 |
{ INDEX_op_qemu_st32, { "L", "L", "L" } }, |
1382 |
{ INDEX_op_qemu_st64, { "L", "L", "L", "L" } }, |
1383 |
#endif
|
1384 |
{ -1 },
|
1385 |
}; |
1386 |
|
1387 |
static int tcg_target_callee_save_regs[] = { |
1388 |
/* TCG_REG_EBP, */ /* currently used for the global env, so no |
1389 |
need to save */
|
1390 |
TCG_REG_EBX, |
1391 |
TCG_REG_ESI, |
1392 |
TCG_REG_EDI, |
1393 |
}; |
1394 |
|
1395 |
/* Generate global QEMU prologue and epilogue code */
|
1396 |
void tcg_target_qemu_prologue(TCGContext *s)
|
1397 |
{ |
1398 |
int i, frame_size, push_size, stack_addend;
|
1399 |
|
1400 |
/* TB prologue */
|
1401 |
/* save all callee saved registers */
|
1402 |
for(i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); i++) { |
1403 |
tcg_out_push(s, tcg_target_callee_save_regs[i]); |
1404 |
} |
1405 |
/* reserve some stack space */
|
1406 |
push_size = 4 + ARRAY_SIZE(tcg_target_callee_save_regs) * 4; |
1407 |
frame_size = push_size + TCG_STATIC_CALL_ARGS_SIZE; |
1408 |
frame_size = (frame_size + TCG_TARGET_STACK_ALIGN - 1) &
|
1409 |
~(TCG_TARGET_STACK_ALIGN - 1);
|
1410 |
stack_addend = frame_size - push_size; |
1411 |
tcg_out_addi(s, TCG_REG_ESP, -stack_addend); |
1412 |
|
1413 |
tcg_out_modrm(s, OPC_GRP5, EXT5_JMPN_Ev, TCG_REG_EAX); /* jmp *%eax */
|
1414 |
|
1415 |
/* TB epilogue */
|
1416 |
tb_ret_addr = s->code_ptr; |
1417 |
tcg_out_addi(s, TCG_REG_ESP, stack_addend); |
1418 |
for(i = ARRAY_SIZE(tcg_target_callee_save_regs) - 1; i >= 0; i--) { |
1419 |
tcg_out_pop(s, tcg_target_callee_save_regs[i]); |
1420 |
} |
1421 |
tcg_out_opc(s, OPC_RET); |
1422 |
} |
1423 |
|
1424 |
void tcg_target_init(TCGContext *s)
|
1425 |
{ |
1426 |
#if !defined(CONFIG_USER_ONLY)
|
1427 |
/* fail safe */
|
1428 |
if ((1 << CPU_TLB_ENTRY_BITS) != sizeof(CPUTLBEntry)) |
1429 |
tcg_abort(); |
1430 |
#endif
|
1431 |
|
1432 |
tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I32], 0, 0xff); |
1433 |
|
1434 |
tcg_regset_clear(tcg_target_call_clobber_regs); |
1435 |
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_EAX); |
1436 |
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_EDX); |
1437 |
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_ECX); |
1438 |
|
1439 |
tcg_regset_clear(s->reserved_regs); |
1440 |
tcg_regset_set_reg(s->reserved_regs, TCG_REG_ESP); |
1441 |
|
1442 |
tcg_add_target_add_op_defs(x86_op_defs); |
1443 |
} |