Commit | Line | Data |
---|---|---|
57871462 | 1 | /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * |
2 | * Mupen64plus - new_dynarec.c * | |
20d507ba | 3 | * Copyright (C) 2009-2011 Ari64 * |
57871462 | 4 | * * |
5 | * This program is free software; you can redistribute it and/or modify * | |
6 | * it under the terms of the GNU General Public License as published by * | |
7 | * the Free Software Foundation; either version 2 of the License, or * | |
8 | * (at your option) any later version. * | |
9 | * * | |
10 | * This program is distributed in the hope that it will be useful, * | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of * | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * | |
13 | * GNU General Public License for more details. * | |
14 | * * | |
15 | * You should have received a copy of the GNU General Public License * | |
16 | * along with this program; if not, write to the * | |
17 | * Free Software Foundation, Inc., * | |
18 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. * | |
19 | * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ | |
20 | ||
21 | #include <stdlib.h> | |
22 | #include <stdint.h> //include for uint64_t | |
23 | #include <assert.h> | |
d848b60a | 24 | #include <errno.h> |
4600ba03 | 25 | #include <sys/mman.h> |
57871462 | 26 | |
3d624f89 | 27 | #include "emu_if.h" //emulator interface |
57871462 | 28 | |
4600ba03 | 29 | //#define DISASM |
30 | //#define assem_debug printf | |
31 | //#define inv_debug printf | |
32 | #define assem_debug(...) | |
33 | #define inv_debug(...) | |
57871462 | 34 | |
35 | #ifdef __i386__ | |
36 | #include "assem_x86.h" | |
37 | #endif | |
38 | #ifdef __x86_64__ | |
39 | #include "assem_x64.h" | |
40 | #endif | |
41 | #ifdef __arm__ | |
42 | #include "assem_arm.h" | |
43 | #endif | |
44 | ||
f23d3386 | 45 | #ifdef __BLACKBERRY_QNX__ |
a4874585 C |
46 | #undef __clear_cache |
47 | #define __clear_cache(start,end) msync(start, (size_t)((void*)end - (void*)start), MS_SYNC | MS_CACHE_ONLY | MS_INVALIDATE_ICACHE); | |
c7b746f0 | 48 | #elif defined(__MACH__) |
49 | #include <libkern/OSCacheControl.h> | |
50 | #define __clear_cache mach_clear_cache | |
51 | static void __clear_cache(void *start, void *end) { | |
52 | size_t len = (char *)end - (char *)start; | |
53 | sys_dcache_flush(start, len); | |
54 | sys_icache_invalidate(start, len); | |
55 | } | |
f23d3386 | 56 | #endif |
a4874585 | 57 | |
57871462 | 58 | #define MAXBLOCK 4096 |
59 | #define MAX_OUTPUT_BLOCK_SIZE 262144 | |
2573466a | 60 | |
57871462 | 61 | struct regstat |
62 | { | |
63 | signed char regmap_entry[HOST_REGS]; | |
64 | signed char regmap[HOST_REGS]; | |
65 | uint64_t was32; | |
66 | uint64_t is32; | |
67 | uint64_t wasdirty; | |
68 | uint64_t dirty; | |
69 | uint64_t u; | |
70 | uint64_t uu; | |
71 | u_int wasconst; | |
72 | u_int isconst; | |
8575a877 | 73 | u_int loadedconst; // host regs that have constants loaded |
74 | u_int waswritten; // MIPS regs that were used as store base before | |
57871462 | 75 | }; |
76 | ||
de5a60c3 | 77 | // note: asm depends on this layout |
57871462 | 78 | struct ll_entry |
79 | { | |
80 | u_int vaddr; | |
de5a60c3 | 81 | u_int reg_sv_flags; |
57871462 | 82 | void *addr; |
83 | struct ll_entry *next; | |
84 | }; | |
85 | ||
86 | u_int start; | |
87 | u_int *source; | |
57871462 | 88 | char insn[MAXBLOCK][10]; |
89 | u_char itype[MAXBLOCK]; | |
90 | u_char opcode[MAXBLOCK]; | |
91 | u_char opcode2[MAXBLOCK]; | |
92 | u_char bt[MAXBLOCK]; | |
93 | u_char rs1[MAXBLOCK]; | |
94 | u_char rs2[MAXBLOCK]; | |
95 | u_char rt1[MAXBLOCK]; | |
96 | u_char rt2[MAXBLOCK]; | |
97 | u_char us1[MAXBLOCK]; | |
98 | u_char us2[MAXBLOCK]; | |
99 | u_char dep1[MAXBLOCK]; | |
100 | u_char dep2[MAXBLOCK]; | |
101 | u_char lt1[MAXBLOCK]; | |
bedfea38 | 102 | static uint64_t gte_rs[MAXBLOCK]; // gte: 32 data and 32 ctl regs |
103 | static uint64_t gte_rt[MAXBLOCK]; | |
104 | static uint64_t gte_unneeded[MAXBLOCK]; | |
ffb0b9e0 | 105 | static u_int smrv[32]; // speculated MIPS register values |
106 | static u_int smrv_strong; // mask or regs that are likely to have correct values | |
107 | static u_int smrv_weak; // same, but somewhat less likely | |
108 | static u_int smrv_strong_next; // same, but after current insn executes | |
109 | static u_int smrv_weak_next; | |
57871462 | 110 | int imm[MAXBLOCK]; |
111 | u_int ba[MAXBLOCK]; | |
112 | char likely[MAXBLOCK]; | |
113 | char is_ds[MAXBLOCK]; | |
e1190b87 | 114 | char ooo[MAXBLOCK]; |
57871462 | 115 | uint64_t unneeded_reg[MAXBLOCK]; |
116 | uint64_t unneeded_reg_upper[MAXBLOCK]; | |
117 | uint64_t branch_unneeded_reg[MAXBLOCK]; | |
118 | uint64_t branch_unneeded_reg_upper[MAXBLOCK]; | |
57871462 | 119 | uint64_t pr32[MAXBLOCK]; |
120 | signed char regmap_pre[MAXBLOCK][HOST_REGS]; | |
956f3129 | 121 | static uint64_t current_constmap[HOST_REGS]; |
122 | static uint64_t constmap[MAXBLOCK][HOST_REGS]; | |
123 | static struct regstat regs[MAXBLOCK]; | |
124 | static struct regstat branch_regs[MAXBLOCK]; | |
e1190b87 | 125 | signed char minimum_free_regs[MAXBLOCK]; |
57871462 | 126 | u_int needed_reg[MAXBLOCK]; |
57871462 | 127 | u_int wont_dirty[MAXBLOCK]; |
128 | u_int will_dirty[MAXBLOCK]; | |
129 | int ccadj[MAXBLOCK]; | |
130 | int slen; | |
131 | u_int instr_addr[MAXBLOCK]; | |
132 | u_int link_addr[MAXBLOCK][3]; | |
133 | int linkcount; | |
134 | u_int stubs[MAXBLOCK*3][8]; | |
135 | int stubcount; | |
136 | u_int literals[1024][2]; | |
137 | int literalcount; | |
138 | int is_delayslot; | |
139 | int cop1_usable; | |
140 | u_char *out; | |
de5a60c3 | 141 | struct ll_entry *jump_in[4096] __attribute__((aligned(16))); |
57871462 | 142 | struct ll_entry *jump_out[4096]; |
143 | struct ll_entry *jump_dirty[4096]; | |
144 | u_int hash_table[65536][4] __attribute__((aligned(16))); | |
145 | char shadow[1048576] __attribute__((aligned(16))); | |
146 | void *copy; | |
147 | int expirep; | |
2f546f9a | 148 | int new_dynarec_did_compile; |
0ff8c62c | 149 | int new_dynarec_hacks; |
57871462 | 150 | u_int stop_after_jal; |
a327ad27 | 151 | #ifndef RAM_FIXED |
152 | static u_int ram_offset; | |
153 | #else | |
154 | static const u_int ram_offset=0; | |
155 | #endif | |
57871462 | 156 | extern u_char restore_candidate[512]; |
157 | extern int cycle_count; | |
158 | ||
159 | /* registers that may be allocated */ | |
160 | /* 1-31 gpr */ | |
161 | #define HIREG 32 // hi | |
162 | #define LOREG 33 // lo | |
163 | #define FSREG 34 // FPU status (FCSR) | |
164 | #define CSREG 35 // Coprocessor status | |
165 | #define CCREG 36 // Cycle count | |
166 | #define INVCP 37 // Pointer to invalid_code | |
1edfcc68 | 167 | //#define MMREG 38 // Pointer to memory_map |
619e5ded | 168 | #define ROREG 39 // ram offset (if rdram!=0x80000000) |
169 | #define TEMPREG 40 | |
170 | #define FTEMP 40 // FPU temporary register | |
171 | #define PTEMP 41 // Prefetch temporary register | |
1edfcc68 | 172 | //#define TLREG 42 // TLB mapping offset |
619e5ded | 173 | #define RHASH 43 // Return address hash |
174 | #define RHTBL 44 // Return address hash table address | |
175 | #define RTEMP 45 // JR/JALR address register | |
176 | #define MAXREG 45 | |
177 | #define AGEN1 46 // Address generation temporary register | |
1edfcc68 | 178 | //#define AGEN2 47 // Address generation temporary register |
179 | //#define MGEN1 48 // Maptable address generation temporary register | |
180 | //#define MGEN2 49 // Maptable address generation temporary register | |
619e5ded | 181 | #define BTREG 50 // Branch target temporary register |
57871462 | 182 | |
183 | /* instruction types */ | |
184 | #define NOP 0 // No operation | |
185 | #define LOAD 1 // Load | |
186 | #define STORE 2 // Store | |
187 | #define LOADLR 3 // Unaligned load | |
188 | #define STORELR 4 // Unaligned store | |
9f51b4b9 | 189 | #define MOV 5 // Move |
57871462 | 190 | #define ALU 6 // Arithmetic/logic |
191 | #define MULTDIV 7 // Multiply/divide | |
192 | #define SHIFT 8 // Shift by register | |
193 | #define SHIFTIMM 9// Shift by immediate | |
194 | #define IMM16 10 // 16-bit immediate | |
195 | #define RJUMP 11 // Unconditional jump to register | |
196 | #define UJUMP 12 // Unconditional jump | |
197 | #define CJUMP 13 // Conditional branch (BEQ/BNE/BGTZ/BLEZ) | |
198 | #define SJUMP 14 // Conditional branch (regimm format) | |
199 | #define COP0 15 // Coprocessor 0 | |
200 | #define COP1 16 // Coprocessor 1 | |
201 | #define C1LS 17 // Coprocessor 1 load/store | |
202 | #define FJUMP 18 // Conditional branch (floating point) | |
203 | #define FLOAT 19 // Floating point unit | |
204 | #define FCONV 20 // Convert integer to float | |
205 | #define FCOMP 21 // Floating point compare (sets FSREG) | |
206 | #define SYSCALL 22// SYSCALL | |
207 | #define OTHER 23 // Other | |
208 | #define SPAN 24 // Branch/delay slot spans 2 pages | |
209 | #define NI 25 // Not implemented | |
7139f3c8 | 210 | #define HLECALL 26// PCSX fake opcodes for HLE |
b9b61529 | 211 | #define COP2 27 // Coprocessor 2 move |
212 | #define C2LS 28 // Coprocessor 2 load/store | |
213 | #define C2OP 29 // Coprocessor 2 operation | |
1e973cb0 | 214 | #define INTCALL 30// Call interpreter to handle rare corner cases |
57871462 | 215 | |
216 | /* stubs */ | |
217 | #define CC_STUB 1 | |
218 | #define FP_STUB 2 | |
219 | #define LOADB_STUB 3 | |
220 | #define LOADH_STUB 4 | |
221 | #define LOADW_STUB 5 | |
222 | #define LOADD_STUB 6 | |
223 | #define LOADBU_STUB 7 | |
224 | #define LOADHU_STUB 8 | |
225 | #define STOREB_STUB 9 | |
226 | #define STOREH_STUB 10 | |
227 | #define STOREW_STUB 11 | |
228 | #define STORED_STUB 12 | |
229 | #define STORELR_STUB 13 | |
230 | #define INVCODE_STUB 14 | |
231 | ||
232 | /* branch codes */ | |
233 | #define TAKEN 1 | |
234 | #define NOTTAKEN 2 | |
235 | #define NULLDS 3 | |
236 | ||
237 | // asm linkage | |
238 | int new_recompile_block(int addr); | |
239 | void *get_addr_ht(u_int vaddr); | |
240 | void invalidate_block(u_int block); | |
241 | void invalidate_addr(u_int addr); | |
242 | void remove_hash(int vaddr); | |
57871462 | 243 | void dyna_linker(); |
244 | void dyna_linker_ds(); | |
245 | void verify_code(); | |
246 | void verify_code_vm(); | |
247 | void verify_code_ds(); | |
248 | void cc_interrupt(); | |
249 | void fp_exception(); | |
250 | void fp_exception_ds(); | |
7139f3c8 | 251 | void jump_syscall_hle(); |
7139f3c8 | 252 | void jump_hlecall(); |
1e973cb0 | 253 | void jump_intcall(); |
7139f3c8 | 254 | void new_dyna_leave(); |
57871462 | 255 | |
57871462 | 256 | // Needed by assembler |
257 | void wb_register(signed char r,signed char regmap[],uint64_t dirty,uint64_t is32); | |
258 | void wb_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty); | |
259 | void wb_needed_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr); | |
260 | void load_all_regs(signed char i_regmap[]); | |
261 | void load_needed_regs(signed char i_regmap[],signed char next_regmap[]); | |
262 | void load_regs_entry(int t); | |
263 | void load_all_consts(signed char regmap[],int is32,u_int dirty,int i); | |
264 | ||
265 | int tracedebug=0; | |
266 | ||
267 | //#define DEBUG_CYCLE_COUNT 1 | |
268 | ||
b6e87b2b | 269 | #define NO_CYCLE_PENALTY_THR 12 |
270 | ||
4e9dcd7f | 271 | int cycle_multiplier; // 100 for 1.0 |
272 | ||
273 | static int CLOCK_ADJUST(int x) | |
274 | { | |
275 | int s=(x>>31)|1; | |
276 | return (x * cycle_multiplier + s * 50) / 100; | |
277 | } | |
278 | ||
94d23bb9 | 279 | static u_int get_page(u_int vaddr) |
57871462 | 280 | { |
0ce47d46 | 281 | u_int page=vaddr&~0xe0000000; |
282 | if (page < 0x1000000) | |
283 | page &= ~0x0e00000; // RAM mirrors | |
284 | page>>=12; | |
57871462 | 285 | if(page>2048) page=2048+(page&2047); |
94d23bb9 | 286 | return page; |
287 | } | |
288 | ||
d25604ca | 289 | // no virtual mem in PCSX |
290 | static u_int get_vpage(u_int vaddr) | |
291 | { | |
292 | return get_page(vaddr); | |
293 | } | |
94d23bb9 | 294 | |
295 | // Get address from virtual address | |
296 | // This is called from the recompiled JR/JALR instructions | |
297 | void *get_addr(u_int vaddr) | |
298 | { | |
299 | u_int page=get_page(vaddr); | |
300 | u_int vpage=get_vpage(vaddr); | |
57871462 | 301 | struct ll_entry *head; |
302 | //printf("TRACE: count=%d next=%d (get_addr %x,page %d)\n",Count,next_interupt,vaddr,page); | |
303 | head=jump_in[page]; | |
304 | while(head!=NULL) { | |
de5a60c3 | 305 | if(head->vaddr==vaddr) { |
57871462 | 306 | //printf("TRACE: count=%d next=%d (get_addr match %x: %x)\n",Count,next_interupt,vaddr,(int)head->addr); |
307 | int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF]; | |
308 | ht_bin[3]=ht_bin[1]; | |
309 | ht_bin[2]=ht_bin[0]; | |
310 | ht_bin[1]=(int)head->addr; | |
311 | ht_bin[0]=vaddr; | |
312 | return head->addr; | |
313 | } | |
314 | head=head->next; | |
315 | } | |
316 | head=jump_dirty[vpage]; | |
317 | while(head!=NULL) { | |
de5a60c3 | 318 | if(head->vaddr==vaddr) { |
57871462 | 319 | //printf("TRACE: count=%d next=%d (get_addr match dirty %x: %x)\n",Count,next_interupt,vaddr,(int)head->addr); |
320 | // Don't restore blocks which are about to expire from the cache | |
321 | if((((u_int)head->addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) | |
322 | if(verify_dirty(head->addr)) { | |
323 | //printf("restore candidate: %x (%d) d=%d\n",vaddr,page,invalid_code[vaddr>>12]); | |
324 | invalid_code[vaddr>>12]=0; | |
9be4ba64 | 325 | inv_code_start=inv_code_end=~0; |
57871462 | 326 | if(vpage<2048) { |
57871462 | 327 | restore_candidate[vpage>>3]|=1<<(vpage&7); |
328 | } | |
329 | else restore_candidate[page>>3]|=1<<(page&7); | |
330 | int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF]; | |
331 | if(ht_bin[0]==vaddr) { | |
332 | ht_bin[1]=(int)head->addr; // Replace existing entry | |
333 | } | |
334 | else | |
335 | { | |
336 | ht_bin[3]=ht_bin[1]; | |
337 | ht_bin[2]=ht_bin[0]; | |
338 | ht_bin[1]=(int)head->addr; | |
339 | ht_bin[0]=vaddr; | |
340 | } | |
341 | return head->addr; | |
342 | } | |
343 | } | |
344 | head=head->next; | |
345 | } | |
346 | //printf("TRACE: count=%d next=%d (get_addr no-match %x)\n",Count,next_interupt,vaddr); | |
347 | int r=new_recompile_block(vaddr); | |
348 | if(r==0) return get_addr(vaddr); | |
349 | // Execute in unmapped page, generate pagefault execption | |
350 | Status|=2; | |
351 | Cause=(vaddr<<31)|0x8; | |
352 | EPC=(vaddr&1)?vaddr-5:vaddr; | |
353 | BadVAddr=(vaddr&~1); | |
354 | Context=(Context&0xFF80000F)|((BadVAddr>>9)&0x007FFFF0); | |
355 | EntryHi=BadVAddr&0xFFFFE000; | |
356 | return get_addr_ht(0x80000000); | |
357 | } | |
358 | // Look up address in hash table first | |
359 | void *get_addr_ht(u_int vaddr) | |
360 | { | |
361 | //printf("TRACE: count=%d next=%d (get_addr_ht %x)\n",Count,next_interupt,vaddr); | |
362 | int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF]; | |
363 | if(ht_bin[0]==vaddr) return (void *)ht_bin[1]; | |
364 | if(ht_bin[2]==vaddr) return (void *)ht_bin[3]; | |
365 | return get_addr(vaddr); | |
366 | } | |
367 | ||
57871462 | 368 | void clear_all_regs(signed char regmap[]) |
369 | { | |
370 | int hr; | |
371 | for (hr=0;hr<HOST_REGS;hr++) regmap[hr]=-1; | |
372 | } | |
373 | ||
374 | signed char get_reg(signed char regmap[],int r) | |
375 | { | |
376 | int hr; | |
377 | for (hr=0;hr<HOST_REGS;hr++) if(hr!=EXCLUDE_REG&®map[hr]==r) return hr; | |
378 | return -1; | |
379 | } | |
380 | ||
381 | // Find a register that is available for two consecutive cycles | |
382 | signed char get_reg2(signed char regmap1[],signed char regmap2[],int r) | |
383 | { | |
384 | int hr; | |
385 | for (hr=0;hr<HOST_REGS;hr++) if(hr!=EXCLUDE_REG&®map1[hr]==r&®map2[hr]==r) return hr; | |
386 | return -1; | |
387 | } | |
388 | ||
389 | int count_free_regs(signed char regmap[]) | |
390 | { | |
391 | int count=0; | |
392 | int hr; | |
393 | for(hr=0;hr<HOST_REGS;hr++) | |
394 | { | |
395 | if(hr!=EXCLUDE_REG) { | |
396 | if(regmap[hr]<0) count++; | |
397 | } | |
398 | } | |
399 | return count; | |
400 | } | |
401 | ||
402 | void dirty_reg(struct regstat *cur,signed char reg) | |
403 | { | |
404 | int hr; | |
405 | if(!reg) return; | |
406 | for (hr=0;hr<HOST_REGS;hr++) { | |
407 | if((cur->regmap[hr]&63)==reg) { | |
408 | cur->dirty|=1<<hr; | |
409 | } | |
410 | } | |
411 | } | |
412 | ||
413 | // If we dirty the lower half of a 64 bit register which is now being | |
414 | // sign-extended, we need to dump the upper half. | |
415 | // Note: Do this only after completion of the instruction, because | |
416 | // some instructions may need to read the full 64-bit value even if | |
417 | // overwriting it (eg SLTI, DSRA32). | |
418 | static void flush_dirty_uppers(struct regstat *cur) | |
419 | { | |
420 | int hr,reg; | |
421 | for (hr=0;hr<HOST_REGS;hr++) { | |
422 | if((cur->dirty>>hr)&1) { | |
423 | reg=cur->regmap[hr]; | |
9f51b4b9 | 424 | if(reg>=64) |
57871462 | 425 | if((cur->is32>>(reg&63))&1) cur->regmap[hr]=-1; |
426 | } | |
427 | } | |
428 | } | |
429 | ||
430 | void set_const(struct regstat *cur,signed char reg,uint64_t value) | |
431 | { | |
432 | int hr; | |
433 | if(!reg) return; | |
434 | for (hr=0;hr<HOST_REGS;hr++) { | |
435 | if(cur->regmap[hr]==reg) { | |
436 | cur->isconst|=1<<hr; | |
956f3129 | 437 | current_constmap[hr]=value; |
57871462 | 438 | } |
439 | else if((cur->regmap[hr]^64)==reg) { | |
440 | cur->isconst|=1<<hr; | |
956f3129 | 441 | current_constmap[hr]=value>>32; |
57871462 | 442 | } |
443 | } | |
444 | } | |
445 | ||
446 | void clear_const(struct regstat *cur,signed char reg) | |
447 | { | |
448 | int hr; | |
449 | if(!reg) return; | |
450 | for (hr=0;hr<HOST_REGS;hr++) { | |
451 | if((cur->regmap[hr]&63)==reg) { | |
452 | cur->isconst&=~(1<<hr); | |
453 | } | |
454 | } | |
455 | } | |
456 | ||
457 | int is_const(struct regstat *cur,signed char reg) | |
458 | { | |
459 | int hr; | |
79c75f1b | 460 | if(reg<0) return 0; |
57871462 | 461 | if(!reg) return 1; |
462 | for (hr=0;hr<HOST_REGS;hr++) { | |
463 | if((cur->regmap[hr]&63)==reg) { | |
464 | return (cur->isconst>>hr)&1; | |
465 | } | |
466 | } | |
467 | return 0; | |
468 | } | |
469 | uint64_t get_const(struct regstat *cur,signed char reg) | |
470 | { | |
471 | int hr; | |
472 | if(!reg) return 0; | |
473 | for (hr=0;hr<HOST_REGS;hr++) { | |
474 | if(cur->regmap[hr]==reg) { | |
956f3129 | 475 | return current_constmap[hr]; |
57871462 | 476 | } |
477 | } | |
c43b5311 | 478 | SysPrintf("Unknown constant in r%d\n",reg); |
57871462 | 479 | exit(1); |
480 | } | |
481 | ||
482 | // Least soon needed registers | |
483 | // Look at the next ten instructions and see which registers | |
484 | // will be used. Try not to reallocate these. | |
485 | void lsn(u_char hsn[], int i, int *preferred_reg) | |
486 | { | |
487 | int j; | |
488 | int b=-1; | |
489 | for(j=0;j<9;j++) | |
490 | { | |
491 | if(i+j>=slen) { | |
492 | j=slen-i-1; | |
493 | break; | |
494 | } | |
495 | if(itype[i+j]==UJUMP||itype[i+j]==RJUMP||(source[i+j]>>16)==0x1000) | |
496 | { | |
497 | // Don't go past an unconditonal jump | |
498 | j++; | |
499 | break; | |
500 | } | |
501 | } | |
502 | for(;j>=0;j--) | |
503 | { | |
504 | if(rs1[i+j]) hsn[rs1[i+j]]=j; | |
505 | if(rs2[i+j]) hsn[rs2[i+j]]=j; | |
506 | if(rt1[i+j]) hsn[rt1[i+j]]=j; | |
507 | if(rt2[i+j]) hsn[rt2[i+j]]=j; | |
508 | if(itype[i+j]==STORE || itype[i+j]==STORELR) { | |
509 | // Stores can allocate zero | |
510 | hsn[rs1[i+j]]=j; | |
511 | hsn[rs2[i+j]]=j; | |
512 | } | |
513 | // On some architectures stores need invc_ptr | |
514 | #if defined(HOST_IMM8) | |
b9b61529 | 515 | if(itype[i+j]==STORE || itype[i+j]==STORELR || (opcode[i+j]&0x3b)==0x39 || (opcode[i+j]&0x3b)==0x3a) { |
57871462 | 516 | hsn[INVCP]=j; |
517 | } | |
518 | #endif | |
519 | if(i+j>=0&&(itype[i+j]==UJUMP||itype[i+j]==CJUMP||itype[i+j]==SJUMP||itype[i+j]==FJUMP)) | |
520 | { | |
521 | hsn[CCREG]=j; | |
522 | b=j; | |
523 | } | |
524 | } | |
525 | if(b>=0) | |
526 | { | |
527 | if(ba[i+b]>=start && ba[i+b]<(start+slen*4)) | |
528 | { | |
529 | // Follow first branch | |
530 | int t=(ba[i+b]-start)>>2; | |
531 | j=7-b;if(t+j>=slen) j=slen-t-1; | |
532 | for(;j>=0;j--) | |
533 | { | |
534 | if(rs1[t+j]) if(hsn[rs1[t+j]]>j+b+2) hsn[rs1[t+j]]=j+b+2; | |
535 | if(rs2[t+j]) if(hsn[rs2[t+j]]>j+b+2) hsn[rs2[t+j]]=j+b+2; | |
536 | //if(rt1[t+j]) if(hsn[rt1[t+j]]>j+b+2) hsn[rt1[t+j]]=j+b+2; | |
537 | //if(rt2[t+j]) if(hsn[rt2[t+j]]>j+b+2) hsn[rt2[t+j]]=j+b+2; | |
538 | } | |
539 | } | |
540 | // TODO: preferred register based on backward branch | |
541 | } | |
542 | // Delay slot should preferably not overwrite branch conditions or cycle count | |
543 | if(i>0&&(itype[i-1]==RJUMP||itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==FJUMP)) { | |
544 | if(rs1[i-1]) if(hsn[rs1[i-1]]>1) hsn[rs1[i-1]]=1; | |
545 | if(rs2[i-1]) if(hsn[rs2[i-1]]>1) hsn[rs2[i-1]]=1; | |
546 | hsn[CCREG]=1; | |
547 | // ...or hash tables | |
548 | hsn[RHASH]=1; | |
549 | hsn[RHTBL]=1; | |
550 | } | |
551 | // Coprocessor load/store needs FTEMP, even if not declared | |
b9b61529 | 552 | if(itype[i]==C1LS||itype[i]==C2LS) { |
57871462 | 553 | hsn[FTEMP]=0; |
554 | } | |
555 | // Load L/R also uses FTEMP as a temporary register | |
556 | if(itype[i]==LOADLR) { | |
557 | hsn[FTEMP]=0; | |
558 | } | |
b7918751 | 559 | // Also SWL/SWR/SDL/SDR |
560 | if(opcode[i]==0x2a||opcode[i]==0x2e||opcode[i]==0x2c||opcode[i]==0x2d) { | |
57871462 | 561 | hsn[FTEMP]=0; |
562 | } | |
57871462 | 563 | // Don't remove the miniht registers |
564 | if(itype[i]==UJUMP||itype[i]==RJUMP) | |
565 | { | |
566 | hsn[RHASH]=0; | |
567 | hsn[RHTBL]=0; | |
568 | } | |
569 | } | |
570 | ||
571 | // We only want to allocate registers if we're going to use them again soon | |
572 | int needed_again(int r, int i) | |
573 | { | |
574 | int j; | |
575 | int b=-1; | |
576 | int rn=10; | |
9f51b4b9 | 577 | |
57871462 | 578 | if(i>0&&(itype[i-1]==UJUMP||itype[i-1]==RJUMP||(source[i-1]>>16)==0x1000)) |
579 | { | |
580 | if(ba[i-1]<start || ba[i-1]>start+slen*4-4) | |
581 | return 0; // Don't need any registers if exiting the block | |
582 | } | |
583 | for(j=0;j<9;j++) | |
584 | { | |
585 | if(i+j>=slen) { | |
586 | j=slen-i-1; | |
587 | break; | |
588 | } | |
589 | if(itype[i+j]==UJUMP||itype[i+j]==RJUMP||(source[i+j]>>16)==0x1000) | |
590 | { | |
591 | // Don't go past an unconditonal jump | |
592 | j++; | |
593 | break; | |
594 | } | |
1e973cb0 | 595 | if(itype[i+j]==SYSCALL||itype[i+j]==HLECALL||itype[i+j]==INTCALL||((source[i+j]&0xfc00003f)==0x0d)) |
57871462 | 596 | { |
597 | break; | |
598 | } | |
599 | } | |
600 | for(;j>=1;j--) | |
601 | { | |
602 | if(rs1[i+j]==r) rn=j; | |
603 | if(rs2[i+j]==r) rn=j; | |
604 | if((unneeded_reg[i+j]>>r)&1) rn=10; | |
605 | if(i+j>=0&&(itype[i+j]==UJUMP||itype[i+j]==CJUMP||itype[i+j]==SJUMP||itype[i+j]==FJUMP)) | |
606 | { | |
607 | b=j; | |
608 | } | |
609 | } | |
610 | /* | |
611 | if(b>=0) | |
612 | { | |
613 | if(ba[i+b]>=start && ba[i+b]<(start+slen*4)) | |
614 | { | |
615 | // Follow first branch | |
616 | int o=rn; | |
617 | int t=(ba[i+b]-start)>>2; | |
618 | j=7-b;if(t+j>=slen) j=slen-t-1; | |
619 | for(;j>=0;j--) | |
620 | { | |
621 | if(!((unneeded_reg[t+j]>>r)&1)) { | |
622 | if(rs1[t+j]==r) if(rn>j+b+2) rn=j+b+2; | |
623 | if(rs2[t+j]==r) if(rn>j+b+2) rn=j+b+2; | |
624 | } | |
625 | else rn=o; | |
626 | } | |
627 | } | |
628 | }*/ | |
b7217e13 | 629 | if(rn<10) return 1; |
57871462 | 630 | return 0; |
631 | } | |
632 | ||
633 | // Try to match register allocations at the end of a loop with those | |
634 | // at the beginning | |
635 | int loop_reg(int i, int r, int hr) | |
636 | { | |
637 | int j,k; | |
638 | for(j=0;j<9;j++) | |
639 | { | |
640 | if(i+j>=slen) { | |
641 | j=slen-i-1; | |
642 | break; | |
643 | } | |
644 | if(itype[i+j]==UJUMP||itype[i+j]==RJUMP||(source[i+j]>>16)==0x1000) | |
645 | { | |
646 | // Don't go past an unconditonal jump | |
647 | j++; | |
648 | break; | |
649 | } | |
650 | } | |
651 | k=0; | |
652 | if(i>0){ | |
653 | if(itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==FJUMP) | |
654 | k--; | |
655 | } | |
656 | for(;k<j;k++) | |
657 | { | |
658 | if(r<64&&((unneeded_reg[i+k]>>r)&1)) return hr; | |
659 | if(r>64&&((unneeded_reg_upper[i+k]>>r)&1)) return hr; | |
660 | if(i+k>=0&&(itype[i+k]==UJUMP||itype[i+k]==CJUMP||itype[i+k]==SJUMP||itype[i+k]==FJUMP)) | |
661 | { | |
662 | if(ba[i+k]>=start && ba[i+k]<(start+i*4)) | |
663 | { | |
664 | int t=(ba[i+k]-start)>>2; | |
665 | int reg=get_reg(regs[t].regmap_entry,r); | |
666 | if(reg>=0) return reg; | |
667 | //reg=get_reg(regs[t+1].regmap_entry,r); | |
668 | //if(reg>=0) return reg; | |
669 | } | |
670 | } | |
671 | } | |
672 | return hr; | |
673 | } | |
674 | ||
675 | ||
676 | // Allocate every register, preserving source/target regs | |
677 | void alloc_all(struct regstat *cur,int i) | |
678 | { | |
679 | int hr; | |
9f51b4b9 | 680 | |
57871462 | 681 | for(hr=0;hr<HOST_REGS;hr++) { |
682 | if(hr!=EXCLUDE_REG) { | |
683 | if(((cur->regmap[hr]&63)!=rs1[i])&&((cur->regmap[hr]&63)!=rs2[i])&& | |
684 | ((cur->regmap[hr]&63)!=rt1[i])&&((cur->regmap[hr]&63)!=rt2[i])) | |
685 | { | |
686 | cur->regmap[hr]=-1; | |
687 | cur->dirty&=~(1<<hr); | |
688 | } | |
689 | // Don't need zeros | |
690 | if((cur->regmap[hr]&63)==0) | |
691 | { | |
692 | cur->regmap[hr]=-1; | |
693 | cur->dirty&=~(1<<hr); | |
694 | } | |
695 | } | |
696 | } | |
697 | } | |
698 | ||
57871462 | 699 | #ifdef __i386__ |
700 | #include "assem_x86.c" | |
701 | #endif | |
702 | #ifdef __x86_64__ | |
703 | #include "assem_x64.c" | |
704 | #endif | |
705 | #ifdef __arm__ | |
706 | #include "assem_arm.c" | |
707 | #endif | |
708 | ||
709 | // Add virtual address mapping to linked list | |
710 | void ll_add(struct ll_entry **head,int vaddr,void *addr) | |
711 | { | |
712 | struct ll_entry *new_entry; | |
713 | new_entry=malloc(sizeof(struct ll_entry)); | |
714 | assert(new_entry!=NULL); | |
715 | new_entry->vaddr=vaddr; | |
de5a60c3 | 716 | new_entry->reg_sv_flags=0; |
57871462 | 717 | new_entry->addr=addr; |
718 | new_entry->next=*head; | |
719 | *head=new_entry; | |
720 | } | |
721 | ||
de5a60c3 | 722 | void ll_add_flags(struct ll_entry **head,int vaddr,u_int reg_sv_flags,void *addr) |
57871462 | 723 | { |
7139f3c8 | 724 | ll_add(head,vaddr,addr); |
de5a60c3 | 725 | (*head)->reg_sv_flags=reg_sv_flags; |
57871462 | 726 | } |
727 | ||
728 | // Check if an address is already compiled | |
729 | // but don't return addresses which are about to expire from the cache | |
730 | void *check_addr(u_int vaddr) | |
731 | { | |
732 | u_int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF]; | |
733 | if(ht_bin[0]==vaddr) { | |
734 | if(((ht_bin[1]-MAX_OUTPUT_BLOCK_SIZE-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) | |
735 | if(isclean(ht_bin[1])) return (void *)ht_bin[1]; | |
736 | } | |
737 | if(ht_bin[2]==vaddr) { | |
738 | if(((ht_bin[3]-MAX_OUTPUT_BLOCK_SIZE-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) | |
739 | if(isclean(ht_bin[3])) return (void *)ht_bin[3]; | |
740 | } | |
94d23bb9 | 741 | u_int page=get_page(vaddr); |
57871462 | 742 | struct ll_entry *head; |
743 | head=jump_in[page]; | |
744 | while(head!=NULL) { | |
de5a60c3 | 745 | if(head->vaddr==vaddr) { |
57871462 | 746 | if((((u_int)head->addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) { |
747 | // Update existing entry with current address | |
748 | if(ht_bin[0]==vaddr) { | |
749 | ht_bin[1]=(int)head->addr; | |
750 | return head->addr; | |
751 | } | |
752 | if(ht_bin[2]==vaddr) { | |
753 | ht_bin[3]=(int)head->addr; | |
754 | return head->addr; | |
755 | } | |
756 | // Insert into hash table with low priority. | |
757 | // Don't evict existing entries, as they are probably | |
758 | // addresses that are being accessed frequently. | |
759 | if(ht_bin[0]==-1) { | |
760 | ht_bin[1]=(int)head->addr; | |
761 | ht_bin[0]=vaddr; | |
762 | }else if(ht_bin[2]==-1) { | |
763 | ht_bin[3]=(int)head->addr; | |
764 | ht_bin[2]=vaddr; | |
765 | } | |
766 | return head->addr; | |
767 | } | |
768 | } | |
769 | head=head->next; | |
770 | } | |
771 | return 0; | |
772 | } | |
773 | ||
774 | void remove_hash(int vaddr) | |
775 | { | |
776 | //printf("remove hash: %x\n",vaddr); | |
777 | int *ht_bin=hash_table[(((vaddr)>>16)^vaddr)&0xFFFF]; | |
778 | if(ht_bin[2]==vaddr) { | |
779 | ht_bin[2]=ht_bin[3]=-1; | |
780 | } | |
781 | if(ht_bin[0]==vaddr) { | |
782 | ht_bin[0]=ht_bin[2]; | |
783 | ht_bin[1]=ht_bin[3]; | |
784 | ht_bin[2]=ht_bin[3]=-1; | |
785 | } | |
786 | } | |
787 | ||
788 | void ll_remove_matching_addrs(struct ll_entry **head,int addr,int shift) | |
789 | { | |
790 | struct ll_entry *next; | |
791 | while(*head) { | |
9f51b4b9 | 792 | if(((u_int)((*head)->addr)>>shift)==(addr>>shift) || |
57871462 | 793 | ((u_int)((*head)->addr-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(addr>>shift)) |
794 | { | |
795 | inv_debug("EXP: Remove pointer to %x (%x)\n",(int)(*head)->addr,(*head)->vaddr); | |
796 | remove_hash((*head)->vaddr); | |
797 | next=(*head)->next; | |
798 | free(*head); | |
799 | *head=next; | |
800 | } | |
801 | else | |
802 | { | |
803 | head=&((*head)->next); | |
804 | } | |
805 | } | |
806 | } | |
807 | ||
808 | // Remove all entries from linked list | |
809 | void ll_clear(struct ll_entry **head) | |
810 | { | |
811 | struct ll_entry *cur; | |
812 | struct ll_entry *next; | |
813 | if(cur=*head) { | |
814 | *head=0; | |
815 | while(cur) { | |
816 | next=cur->next; | |
817 | free(cur); | |
818 | cur=next; | |
819 | } | |
820 | } | |
821 | } | |
822 | ||
823 | // Dereference the pointers and remove if it matches | |
824 | void ll_kill_pointers(struct ll_entry *head,int addr,int shift) | |
825 | { | |
826 | while(head) { | |
827 | int ptr=get_pointer(head->addr); | |
828 | inv_debug("EXP: Lookup pointer to %x at %x (%x)\n",(int)ptr,(int)head->addr,head->vaddr); | |
829 | if(((ptr>>shift)==(addr>>shift)) || | |
830 | (((ptr-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(addr>>shift))) | |
831 | { | |
5088bb70 | 832 | inv_debug("EXP: Kill pointer at %x (%x)\n",(int)head->addr,head->vaddr); |
f76eeef9 | 833 | u_int host_addr=(u_int)kill_pointer(head->addr); |
dd3a91a1 | 834 | #ifdef __arm__ |
835 | needs_clear_cache[(host_addr-(u_int)BASE_ADDR)>>17]|=1<<(((host_addr-(u_int)BASE_ADDR)>>12)&31); | |
836 | #endif | |
57871462 | 837 | } |
838 | head=head->next; | |
839 | } | |
840 | } | |
841 | ||
842 | // This is called when we write to a compiled block (see do_invstub) | |
f76eeef9 | 843 | void invalidate_page(u_int page) |
57871462 | 844 | { |
57871462 | 845 | struct ll_entry *head; |
846 | struct ll_entry *next; | |
847 | head=jump_in[page]; | |
848 | jump_in[page]=0; | |
849 | while(head!=NULL) { | |
850 | inv_debug("INVALIDATE: %x\n",head->vaddr); | |
851 | remove_hash(head->vaddr); | |
852 | next=head->next; | |
853 | free(head); | |
854 | head=next; | |
855 | } | |
856 | head=jump_out[page]; | |
857 | jump_out[page]=0; | |
858 | while(head!=NULL) { | |
859 | inv_debug("INVALIDATE: kill pointer to %x (%x)\n",head->vaddr,(int)head->addr); | |
f76eeef9 | 860 | u_int host_addr=(u_int)kill_pointer(head->addr); |
dd3a91a1 | 861 | #ifdef __arm__ |
862 | needs_clear_cache[(host_addr-(u_int)BASE_ADDR)>>17]|=1<<(((host_addr-(u_int)BASE_ADDR)>>12)&31); | |
863 | #endif | |
57871462 | 864 | next=head->next; |
865 | free(head); | |
866 | head=next; | |
867 | } | |
57871462 | 868 | } |
9be4ba64 | 869 | |
870 | static void invalidate_block_range(u_int block, u_int first, u_int last) | |
57871462 | 871 | { |
94d23bb9 | 872 | u_int page=get_page(block<<12); |
57871462 | 873 | //printf("first=%d last=%d\n",first,last); |
f76eeef9 | 874 | invalidate_page(page); |
57871462 | 875 | assert(first+5>page); // NB: this assumes MAXBLOCK<=4096 (4 pages) |
876 | assert(last<page+5); | |
877 | // Invalidate the adjacent pages if a block crosses a 4K boundary | |
878 | while(first<page) { | |
879 | invalidate_page(first); | |
880 | first++; | |
881 | } | |
882 | for(first=page+1;first<last;first++) { | |
883 | invalidate_page(first); | |
884 | } | |
dd3a91a1 | 885 | #ifdef __arm__ |
886 | do_clear_cache(); | |
887 | #endif | |
9f51b4b9 | 888 | |
57871462 | 889 | // Don't trap writes |
890 | invalid_code[block]=1; | |
f76eeef9 | 891 | |
57871462 | 892 | #ifdef USE_MINI_HT |
893 | memset(mini_ht,-1,sizeof(mini_ht)); | |
894 | #endif | |
895 | } | |
9be4ba64 | 896 | |
897 | void invalidate_block(u_int block) | |
898 | { | |
899 | u_int page=get_page(block<<12); | |
900 | u_int vpage=get_vpage(block<<12); | |
901 | inv_debug("INVALIDATE: %x (%d)\n",block<<12,page); | |
902 | //inv_debug("invalid_code[block]=%d\n",invalid_code[block]); | |
903 | u_int first,last; | |
904 | first=last=page; | |
905 | struct ll_entry *head; | |
906 | head=jump_dirty[vpage]; | |
907 | //printf("page=%d vpage=%d\n",page,vpage); | |
908 | while(head!=NULL) { | |
909 | u_int start,end; | |
910 | if(vpage>2047||(head->vaddr>>12)==block) { // Ignore vaddr hash collision | |
911 | get_bounds((int)head->addr,&start,&end); | |
912 | //printf("start: %x end: %x\n",start,end); | |
4a35de07 | 913 | if(page<2048&&start>=(u_int)rdram&&end<(u_int)rdram+RAM_SIZE) { |
9be4ba64 | 914 | if(((start-(u_int)rdram)>>12)<=page&&((end-1-(u_int)rdram)>>12)>=page) { |
915 | if((((start-(u_int)rdram)>>12)&2047)<first) first=((start-(u_int)rdram)>>12)&2047; | |
916 | if((((end-1-(u_int)rdram)>>12)&2047)>last) last=((end-1-(u_int)rdram)>>12)&2047; | |
917 | } | |
918 | } | |
9be4ba64 | 919 | } |
920 | head=head->next; | |
921 | } | |
922 | invalidate_block_range(block,first,last); | |
923 | } | |
924 | ||
57871462 | 925 | void invalidate_addr(u_int addr) |
926 | { | |
9be4ba64 | 927 | //static int rhits; |
928 | // this check is done by the caller | |
929 | //if (inv_code_start<=addr&&addr<=inv_code_end) { rhits++; return; } | |
d25604ca | 930 | u_int page=get_vpage(addr); |
9be4ba64 | 931 | if(page<2048) { // RAM |
932 | struct ll_entry *head; | |
933 | u_int addr_min=~0, addr_max=0; | |
4a35de07 | 934 | u_int mask=RAM_SIZE-1; |
935 | u_int addr_main=0x80000000|(addr&mask); | |
9be4ba64 | 936 | int pg1; |
4a35de07 | 937 | inv_code_start=addr_main&~0xfff; |
938 | inv_code_end=addr_main|0xfff; | |
9be4ba64 | 939 | pg1=page; |
940 | if (pg1>0) { | |
941 | // must check previous page too because of spans.. | |
942 | pg1--; | |
943 | inv_code_start-=0x1000; | |
944 | } | |
945 | for(;pg1<=page;pg1++) { | |
946 | for(head=jump_dirty[pg1];head!=NULL;head=head->next) { | |
947 | u_int start,end; | |
948 | get_bounds((int)head->addr,&start,&end); | |
4a35de07 | 949 | if(ram_offset) { |
950 | start-=ram_offset; | |
951 | end-=ram_offset; | |
952 | } | |
953 | if(start<=addr_main&&addr_main<end) { | |
9be4ba64 | 954 | if(start<addr_min) addr_min=start; |
955 | if(end>addr_max) addr_max=end; | |
956 | } | |
4a35de07 | 957 | else if(addr_main<start) { |
9be4ba64 | 958 | if(start<inv_code_end) |
959 | inv_code_end=start-1; | |
960 | } | |
961 | else { | |
962 | if(end>inv_code_start) | |
963 | inv_code_start=end; | |
964 | } | |
965 | } | |
966 | } | |
967 | if (addr_min!=~0) { | |
968 | inv_debug("INV ADDR: %08x hit %08x-%08x\n", addr, addr_min, addr_max); | |
969 | inv_code_start=inv_code_end=~0; | |
970 | invalidate_block_range(addr>>12,(addr_min&mask)>>12,(addr_max&mask)>>12); | |
971 | return; | |
972 | } | |
973 | else { | |
4a35de07 | 974 | inv_code_start=(addr&~mask)|(inv_code_start&mask); |
975 | inv_code_end=(addr&~mask)|(inv_code_end&mask); | |
d25604ca | 976 | inv_debug("INV ADDR: %08x miss, inv %08x-%08x, sk %d\n", addr, inv_code_start, inv_code_end, 0); |
9be4ba64 | 977 | return; |
d25604ca | 978 | } |
9be4ba64 | 979 | } |
57871462 | 980 | invalidate_block(addr>>12); |
981 | } | |
9be4ba64 | 982 | |
dd3a91a1 | 983 | // This is called when loading a save state. |
984 | // Anything could have changed, so invalidate everything. | |
57871462 | 985 | void invalidate_all_pages() |
986 | { | |
987 | u_int page,n; | |
988 | for(page=0;page<4096;page++) | |
989 | invalidate_page(page); | |
990 | for(page=0;page<1048576;page++) | |
991 | if(!invalid_code[page]) { | |
992 | restore_candidate[(page&2047)>>3]|=1<<(page&7); | |
993 | restore_candidate[((page&2047)>>3)+256]|=1<<(page&7); | |
994 | } | |
995 | #ifdef __arm__ | |
996 | __clear_cache((void *)BASE_ADDR,(void *)BASE_ADDR+(1<<TARGET_SIZE_2)); | |
997 | #endif | |
998 | #ifdef USE_MINI_HT | |
999 | memset(mini_ht,-1,sizeof(mini_ht)); | |
1000 | #endif | |
57871462 | 1001 | } |
1002 | ||
1003 | // Add an entry to jump_out after making a link | |
1004 | void add_link(u_int vaddr,void *src) | |
1005 | { | |
94d23bb9 | 1006 | u_int page=get_page(vaddr); |
57871462 | 1007 | inv_debug("add_link: %x -> %x (%d)\n",(int)src,vaddr,page); |
76f71c27 | 1008 | int *ptr=(int *)(src+4); |
1009 | assert((*ptr&0x0fff0000)==0x059f0000); | |
57871462 | 1010 | ll_add(jump_out+page,vaddr,src); |
1011 | //int ptr=get_pointer(src); | |
1012 | //inv_debug("add_link: Pointer is to %x\n",(int)ptr); | |
1013 | } | |
1014 | ||
1015 | // If a code block was found to be unmodified (bit was set in | |
1016 | // restore_candidate) and it remains unmodified (bit is clear | |
1017 | // in invalid_code) then move the entries for that 4K page from | |
1018 | // the dirty list to the clean list. | |
1019 | void clean_blocks(u_int page) | |
1020 | { | |
1021 | struct ll_entry *head; | |
1022 | inv_debug("INV: clean_blocks page=%d\n",page); | |
1023 | head=jump_dirty[page]; | |
1024 | while(head!=NULL) { | |
1025 | if(!invalid_code[head->vaddr>>12]) { | |
1026 | // Don't restore blocks which are about to expire from the cache | |
1027 | if((((u_int)head->addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) { | |
1028 | u_int start,end; | |
1029 | if(verify_dirty((int)head->addr)) { | |
1030 | //printf("Possibly Restore %x (%x)\n",head->vaddr, (int)head->addr); | |
1031 | u_int i; | |
1032 | u_int inv=0; | |
1033 | get_bounds((int)head->addr,&start,&end); | |
4cb76aa4 | 1034 | if(start-(u_int)rdram<RAM_SIZE) { |
57871462 | 1035 | for(i=(start-(u_int)rdram+0x80000000)>>12;i<=(end-1-(u_int)rdram+0x80000000)>>12;i++) { |
1036 | inv|=invalid_code[i]; | |
1037 | } | |
1038 | } | |
4cb76aa4 | 1039 | else if((signed int)head->vaddr>=(signed int)0x80000000+RAM_SIZE) { |
57871462 | 1040 | inv=1; |
1041 | } | |
1042 | if(!inv) { | |
1043 | void * clean_addr=(void *)get_clean_addr((int)head->addr); | |
1044 | if((((u_int)clean_addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) { | |
1045 | u_int ppage=page; | |
57871462 | 1046 | inv_debug("INV: Restored %x (%x/%x)\n",head->vaddr, (int)head->addr, (int)clean_addr); |
1047 | //printf("page=%x, addr=%x\n",page,head->vaddr); | |
1048 | //assert(head->vaddr>>12==(page|0x80000)); | |
de5a60c3 | 1049 | ll_add_flags(jump_in+ppage,head->vaddr,head->reg_sv_flags,clean_addr); |
57871462 | 1050 | int *ht_bin=hash_table[((head->vaddr>>16)^head->vaddr)&0xFFFF]; |
de5a60c3 | 1051 | if(ht_bin[0]==head->vaddr) { |
1052 | ht_bin[1]=(int)clean_addr; // Replace existing entry | |
1053 | } | |
1054 | if(ht_bin[2]==head->vaddr) { | |
1055 | ht_bin[3]=(int)clean_addr; // Replace existing entry | |
57871462 | 1056 | } |
1057 | } | |
1058 | } | |
1059 | } | |
1060 | } | |
1061 | } | |
1062 | head=head->next; | |
1063 | } | |
1064 | } | |
1065 | ||
1066 | ||
1067 | void mov_alloc(struct regstat *current,int i) | |
1068 | { | |
1069 | // Note: Don't need to actually alloc the source registers | |
1070 | if((~current->is32>>rs1[i])&1) { | |
1071 | //alloc_reg64(current,i,rs1[i]); | |
1072 | alloc_reg64(current,i,rt1[i]); | |
1073 | current->is32&=~(1LL<<rt1[i]); | |
1074 | } else { | |
1075 | //alloc_reg(current,i,rs1[i]); | |
1076 | alloc_reg(current,i,rt1[i]); | |
1077 | current->is32|=(1LL<<rt1[i]); | |
1078 | } | |
1079 | clear_const(current,rs1[i]); | |
1080 | clear_const(current,rt1[i]); | |
1081 | dirty_reg(current,rt1[i]); | |
1082 | } | |
1083 | ||
1084 | void shiftimm_alloc(struct regstat *current,int i) | |
1085 | { | |
57871462 | 1086 | if(opcode2[i]<=0x3) // SLL/SRL/SRA |
1087 | { | |
1088 | if(rt1[i]) { | |
1089 | if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); | |
1090 | else lt1[i]=rs1[i]; | |
1091 | alloc_reg(current,i,rt1[i]); | |
1092 | current->is32|=1LL<<rt1[i]; | |
1093 | dirty_reg(current,rt1[i]); | |
dc49e339 | 1094 | if(is_const(current,rs1[i])) { |
1095 | int v=get_const(current,rs1[i]); | |
1096 | if(opcode2[i]==0x00) set_const(current,rt1[i],v<<imm[i]); | |
1097 | if(opcode2[i]==0x02) set_const(current,rt1[i],(u_int)v>>imm[i]); | |
1098 | if(opcode2[i]==0x03) set_const(current,rt1[i],v>>imm[i]); | |
1099 | } | |
1100 | else clear_const(current,rt1[i]); | |
57871462 | 1101 | } |
1102 | } | |
dc49e339 | 1103 | else |
1104 | { | |
1105 | clear_const(current,rs1[i]); | |
1106 | clear_const(current,rt1[i]); | |
1107 | } | |
1108 | ||
57871462 | 1109 | if(opcode2[i]>=0x38&&opcode2[i]<=0x3b) // DSLL/DSRL/DSRA |
1110 | { | |
1111 | if(rt1[i]) { | |
1112 | if(rs1[i]) alloc_reg64(current,i,rs1[i]); | |
1113 | alloc_reg64(current,i,rt1[i]); | |
1114 | current->is32&=~(1LL<<rt1[i]); | |
1115 | dirty_reg(current,rt1[i]); | |
1116 | } | |
1117 | } | |
1118 | if(opcode2[i]==0x3c) // DSLL32 | |
1119 | { | |
1120 | if(rt1[i]) { | |
1121 | if(rs1[i]) alloc_reg(current,i,rs1[i]); | |
1122 | alloc_reg64(current,i,rt1[i]); | |
1123 | current->is32&=~(1LL<<rt1[i]); | |
1124 | dirty_reg(current,rt1[i]); | |
1125 | } | |
1126 | } | |
1127 | if(opcode2[i]==0x3e) // DSRL32 | |
1128 | { | |
1129 | if(rt1[i]) { | |
1130 | alloc_reg64(current,i,rs1[i]); | |
1131 | if(imm[i]==32) { | |
1132 | alloc_reg64(current,i,rt1[i]); | |
1133 | current->is32&=~(1LL<<rt1[i]); | |
1134 | } else { | |
1135 | alloc_reg(current,i,rt1[i]); | |
1136 | current->is32|=1LL<<rt1[i]; | |
1137 | } | |
1138 | dirty_reg(current,rt1[i]); | |
1139 | } | |
1140 | } | |
1141 | if(opcode2[i]==0x3f) // DSRA32 | |
1142 | { | |
1143 | if(rt1[i]) { | |
1144 | alloc_reg64(current,i,rs1[i]); | |
1145 | alloc_reg(current,i,rt1[i]); | |
1146 | current->is32|=1LL<<rt1[i]; | |
1147 | dirty_reg(current,rt1[i]); | |
1148 | } | |
1149 | } | |
1150 | } | |
1151 | ||
1152 | void shift_alloc(struct regstat *current,int i) | |
1153 | { | |
1154 | if(rt1[i]) { | |
1155 | if(opcode2[i]<=0x07) // SLLV/SRLV/SRAV | |
1156 | { | |
1157 | if(rs1[i]) alloc_reg(current,i,rs1[i]); | |
1158 | if(rs2[i]) alloc_reg(current,i,rs2[i]); | |
1159 | alloc_reg(current,i,rt1[i]); | |
e1190b87 | 1160 | if(rt1[i]==rs2[i]) { |
1161 | alloc_reg_temp(current,i,-1); | |
1162 | minimum_free_regs[i]=1; | |
1163 | } | |
57871462 | 1164 | current->is32|=1LL<<rt1[i]; |
1165 | } else { // DSLLV/DSRLV/DSRAV | |
1166 | if(rs1[i]) alloc_reg64(current,i,rs1[i]); | |
1167 | if(rs2[i]) alloc_reg(current,i,rs2[i]); | |
1168 | alloc_reg64(current,i,rt1[i]); | |
1169 | current->is32&=~(1LL<<rt1[i]); | |
1170 | if(opcode2[i]==0x16||opcode2[i]==0x17) // DSRLV and DSRAV need a temporary register | |
e1190b87 | 1171 | { |
57871462 | 1172 | alloc_reg_temp(current,i,-1); |
e1190b87 | 1173 | minimum_free_regs[i]=1; |
1174 | } | |
57871462 | 1175 | } |
1176 | clear_const(current,rs1[i]); | |
1177 | clear_const(current,rs2[i]); | |
1178 | clear_const(current,rt1[i]); | |
1179 | dirty_reg(current,rt1[i]); | |
1180 | } | |
1181 | } | |
1182 | ||
1183 | void alu_alloc(struct regstat *current,int i) | |
1184 | { | |
1185 | if(opcode2[i]>=0x20&&opcode2[i]<=0x23) { // ADD/ADDU/SUB/SUBU | |
1186 | if(rt1[i]) { | |
1187 | if(rs1[i]&&rs2[i]) { | |
1188 | alloc_reg(current,i,rs1[i]); | |
1189 | alloc_reg(current,i,rs2[i]); | |
1190 | } | |
1191 | else { | |
1192 | if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); | |
1193 | if(rs2[i]&&needed_again(rs2[i],i)) alloc_reg(current,i,rs2[i]); | |
1194 | } | |
1195 | alloc_reg(current,i,rt1[i]); | |
1196 | } | |
1197 | current->is32|=1LL<<rt1[i]; | |
1198 | } | |
1199 | if(opcode2[i]==0x2a||opcode2[i]==0x2b) { // SLT/SLTU | |
1200 | if(rt1[i]) { | |
1201 | if(!((current->is32>>rs1[i])&(current->is32>>rs2[i])&1)) | |
1202 | { | |
1203 | alloc_reg64(current,i,rs1[i]); | |
1204 | alloc_reg64(current,i,rs2[i]); | |
1205 | alloc_reg(current,i,rt1[i]); | |
1206 | } else { | |
1207 | alloc_reg(current,i,rs1[i]); | |
1208 | alloc_reg(current,i,rs2[i]); | |
1209 | alloc_reg(current,i,rt1[i]); | |
1210 | } | |
1211 | } | |
1212 | current->is32|=1LL<<rt1[i]; | |
1213 | } | |
1214 | if(opcode2[i]>=0x24&&opcode2[i]<=0x27) { // AND/OR/XOR/NOR | |
1215 | if(rt1[i]) { | |
1216 | if(rs1[i]&&rs2[i]) { | |
1217 | alloc_reg(current,i,rs1[i]); | |
1218 | alloc_reg(current,i,rs2[i]); | |
1219 | } | |
1220 | else | |
1221 | { | |
1222 | if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); | |
1223 | if(rs2[i]&&needed_again(rs2[i],i)) alloc_reg(current,i,rs2[i]); | |
1224 | } | |
1225 | alloc_reg(current,i,rt1[i]); | |
1226 | if(!((current->is32>>rs1[i])&(current->is32>>rs2[i])&1)) | |
1227 | { | |
1228 | if(!((current->uu>>rt1[i])&1)) { | |
1229 | alloc_reg64(current,i,rt1[i]); | |
1230 | } | |
1231 | if(get_reg(current->regmap,rt1[i]|64)>=0) { | |
1232 | if(rs1[i]&&rs2[i]) { | |
1233 | alloc_reg64(current,i,rs1[i]); | |
1234 | alloc_reg64(current,i,rs2[i]); | |
1235 | } | |
1236 | else | |
1237 | { | |
1238 | // Is is really worth it to keep 64-bit values in registers? | |
1239 | #ifdef NATIVE_64BIT | |
1240 | if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg64(current,i,rs1[i]); | |
1241 | if(rs2[i]&&needed_again(rs2[i],i)) alloc_reg64(current,i,rs2[i]); | |
1242 | #endif | |
1243 | } | |
1244 | } | |
1245 | current->is32&=~(1LL<<rt1[i]); | |
1246 | } else { | |
1247 | current->is32|=1LL<<rt1[i]; | |
1248 | } | |
1249 | } | |
1250 | } | |
1251 | if(opcode2[i]>=0x2c&&opcode2[i]<=0x2f) { // DADD/DADDU/DSUB/DSUBU | |
1252 | if(rt1[i]) { | |
1253 | if(rs1[i]&&rs2[i]) { | |
1254 | if(!((current->uu>>rt1[i])&1)||get_reg(current->regmap,rt1[i]|64)>=0) { | |
1255 | alloc_reg64(current,i,rs1[i]); | |
1256 | alloc_reg64(current,i,rs2[i]); | |
1257 | alloc_reg64(current,i,rt1[i]); | |
1258 | } else { | |
1259 | alloc_reg(current,i,rs1[i]); | |
1260 | alloc_reg(current,i,rs2[i]); | |
1261 | alloc_reg(current,i,rt1[i]); | |
1262 | } | |
1263 | } | |
1264 | else { | |
1265 | alloc_reg(current,i,rt1[i]); | |
1266 | if(!((current->uu>>rt1[i])&1)||get_reg(current->regmap,rt1[i]|64)>=0) { | |
1267 | // DADD used as move, or zeroing | |
1268 | // If we have a 64-bit source, then make the target 64 bits too | |
1269 | if(rs1[i]&&!((current->is32>>rs1[i])&1)) { | |
1270 | if(get_reg(current->regmap,rs1[i])>=0) alloc_reg64(current,i,rs1[i]); | |
1271 | alloc_reg64(current,i,rt1[i]); | |
1272 | } else if(rs2[i]&&!((current->is32>>rs2[i])&1)) { | |
1273 | if(get_reg(current->regmap,rs2[i])>=0) alloc_reg64(current,i,rs2[i]); | |
1274 | alloc_reg64(current,i,rt1[i]); | |
1275 | } | |
1276 | if(opcode2[i]>=0x2e&&rs2[i]) { | |
1277 | // DSUB used as negation - 64-bit result | |
1278 | // If we have a 32-bit register, extend it to 64 bits | |
1279 | if(get_reg(current->regmap,rs2[i])>=0) alloc_reg64(current,i,rs2[i]); | |
1280 | alloc_reg64(current,i,rt1[i]); | |
1281 | } | |
1282 | } | |
1283 | } | |
1284 | if(rs1[i]&&rs2[i]) { | |
1285 | current->is32&=~(1LL<<rt1[i]); | |
1286 | } else if(rs1[i]) { | |
1287 | current->is32&=~(1LL<<rt1[i]); | |
1288 | if((current->is32>>rs1[i])&1) | |
1289 | current->is32|=1LL<<rt1[i]; | |
1290 | } else if(rs2[i]) { | |
1291 | current->is32&=~(1LL<<rt1[i]); | |
1292 | if((current->is32>>rs2[i])&1) | |
1293 | current->is32|=1LL<<rt1[i]; | |
1294 | } else { | |
1295 | current->is32|=1LL<<rt1[i]; | |
1296 | } | |
1297 | } | |
1298 | } | |
1299 | clear_const(current,rs1[i]); | |
1300 | clear_const(current,rs2[i]); | |
1301 | clear_const(current,rt1[i]); | |
1302 | dirty_reg(current,rt1[i]); | |
1303 | } | |
1304 | ||
1305 | void imm16_alloc(struct regstat *current,int i) | |
1306 | { | |
1307 | if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); | |
1308 | else lt1[i]=rs1[i]; | |
1309 | if(rt1[i]) alloc_reg(current,i,rt1[i]); | |
1310 | if(opcode[i]==0x18||opcode[i]==0x19) { // DADDI/DADDIU | |
1311 | current->is32&=~(1LL<<rt1[i]); | |
1312 | if(!((current->uu>>rt1[i])&1)||get_reg(current->regmap,rt1[i]|64)>=0) { | |
1313 | // TODO: Could preserve the 32-bit flag if the immediate is zero | |
1314 | alloc_reg64(current,i,rt1[i]); | |
1315 | alloc_reg64(current,i,rs1[i]); | |
1316 | } | |
1317 | clear_const(current,rs1[i]); | |
1318 | clear_const(current,rt1[i]); | |
1319 | } | |
1320 | else if(opcode[i]==0x0a||opcode[i]==0x0b) { // SLTI/SLTIU | |
1321 | if((~current->is32>>rs1[i])&1) alloc_reg64(current,i,rs1[i]); | |
1322 | current->is32|=1LL<<rt1[i]; | |
1323 | clear_const(current,rs1[i]); | |
1324 | clear_const(current,rt1[i]); | |
1325 | } | |
1326 | else if(opcode[i]>=0x0c&&opcode[i]<=0x0e) { // ANDI/ORI/XORI | |
1327 | if(((~current->is32>>rs1[i])&1)&&opcode[i]>0x0c) { | |
1328 | if(rs1[i]!=rt1[i]) { | |
1329 | if(needed_again(rs1[i],i)) alloc_reg64(current,i,rs1[i]); | |
1330 | alloc_reg64(current,i,rt1[i]); | |
1331 | current->is32&=~(1LL<<rt1[i]); | |
1332 | } | |
1333 | } | |
1334 | else current->is32|=1LL<<rt1[i]; // ANDI clears upper bits | |
1335 | if(is_const(current,rs1[i])) { | |
1336 | int v=get_const(current,rs1[i]); | |
1337 | if(opcode[i]==0x0c) set_const(current,rt1[i],v&imm[i]); | |
1338 | if(opcode[i]==0x0d) set_const(current,rt1[i],v|imm[i]); | |
1339 | if(opcode[i]==0x0e) set_const(current,rt1[i],v^imm[i]); | |
1340 | } | |
1341 | else clear_const(current,rt1[i]); | |
1342 | } | |
1343 | else if(opcode[i]==0x08||opcode[i]==0x09) { // ADDI/ADDIU | |
1344 | if(is_const(current,rs1[i])) { | |
1345 | int v=get_const(current,rs1[i]); | |
1346 | set_const(current,rt1[i],v+imm[i]); | |
1347 | } | |
1348 | else clear_const(current,rt1[i]); | |
1349 | current->is32|=1LL<<rt1[i]; | |
1350 | } | |
1351 | else { | |
1352 | set_const(current,rt1[i],((long long)((short)imm[i]))<<16); // LUI | |
1353 | current->is32|=1LL<<rt1[i]; | |
1354 | } | |
1355 | dirty_reg(current,rt1[i]); | |
1356 | } | |
1357 | ||
1358 | void load_alloc(struct regstat *current,int i) | |
1359 | { | |
1360 | clear_const(current,rt1[i]); | |
1361 | //if(rs1[i]!=rt1[i]&&needed_again(rs1[i],i)) clear_const(current,rs1[i]); // Does this help or hurt? | |
1362 | if(!rs1[i]) current->u&=~1LL; // Allow allocating r0 if it's the source register | |
1363 | if(needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); | |
373d1d07 | 1364 | if(rt1[i]&&!((current->u>>rt1[i])&1)) { |
57871462 | 1365 | alloc_reg(current,i,rt1[i]); |
373d1d07 | 1366 | assert(get_reg(current->regmap,rt1[i])>=0); |
57871462 | 1367 | if(opcode[i]==0x27||opcode[i]==0x37) // LWU/LD |
1368 | { | |
1369 | current->is32&=~(1LL<<rt1[i]); | |
1370 | alloc_reg64(current,i,rt1[i]); | |
1371 | } | |
1372 | else if(opcode[i]==0x1A||opcode[i]==0x1B) // LDL/LDR | |
1373 | { | |
1374 | current->is32&=~(1LL<<rt1[i]); | |
1375 | alloc_reg64(current,i,rt1[i]); | |
1376 | alloc_all(current,i); | |
1377 | alloc_reg64(current,i,FTEMP); | |
e1190b87 | 1378 | minimum_free_regs[i]=HOST_REGS; |
57871462 | 1379 | } |
1380 | else current->is32|=1LL<<rt1[i]; | |
1381 | dirty_reg(current,rt1[i]); | |
57871462 | 1382 | // LWL/LWR need a temporary register for the old value |
1383 | if(opcode[i]==0x22||opcode[i]==0x26) | |
1384 | { | |
1385 | alloc_reg(current,i,FTEMP); | |
1386 | alloc_reg_temp(current,i,-1); | |
e1190b87 | 1387 | minimum_free_regs[i]=1; |
57871462 | 1388 | } |
1389 | } | |
1390 | else | |
1391 | { | |
373d1d07 | 1392 | // Load to r0 or unneeded register (dummy load) |
57871462 | 1393 | // but we still need a register to calculate the address |
535d208a | 1394 | if(opcode[i]==0x22||opcode[i]==0x26) |
1395 | { | |
1396 | alloc_reg(current,i,FTEMP); // LWL/LWR need another temporary | |
1397 | } | |
57871462 | 1398 | alloc_reg_temp(current,i,-1); |
e1190b87 | 1399 | minimum_free_regs[i]=1; |
535d208a | 1400 | if(opcode[i]==0x1A||opcode[i]==0x1B) // LDL/LDR |
1401 | { | |
1402 | alloc_all(current,i); | |
1403 | alloc_reg64(current,i,FTEMP); | |
e1190b87 | 1404 | minimum_free_regs[i]=HOST_REGS; |
535d208a | 1405 | } |
57871462 | 1406 | } |
1407 | } | |
1408 | ||
1409 | void store_alloc(struct regstat *current,int i) | |
1410 | { | |
1411 | clear_const(current,rs2[i]); | |
1412 | if(!(rs2[i])) current->u&=~1LL; // Allow allocating r0 if necessary | |
1413 | if(needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); | |
1414 | alloc_reg(current,i,rs2[i]); | |
1415 | if(opcode[i]==0x2c||opcode[i]==0x2d||opcode[i]==0x3f) { // 64-bit SDL/SDR/SD | |
1416 | alloc_reg64(current,i,rs2[i]); | |
1417 | if(rs2[i]) alloc_reg(current,i,FTEMP); | |
1418 | } | |
57871462 | 1419 | #if defined(HOST_IMM8) |
1420 | // On CPUs without 32-bit immediates we need a pointer to invalid_code | |
1421 | else alloc_reg(current,i,INVCP); | |
1422 | #endif | |
b7918751 | 1423 | if(opcode[i]==0x2a||opcode[i]==0x2e||opcode[i]==0x2c||opcode[i]==0x2d) { // SWL/SWL/SDL/SDR |
57871462 | 1424 | alloc_reg(current,i,FTEMP); |
1425 | } | |
1426 | // We need a temporary register for address generation | |
1427 | alloc_reg_temp(current,i,-1); | |
e1190b87 | 1428 | minimum_free_regs[i]=1; |
57871462 | 1429 | } |
1430 | ||
1431 | void c1ls_alloc(struct regstat *current,int i) | |
1432 | { | |
1433 | //clear_const(current,rs1[i]); // FIXME | |
1434 | clear_const(current,rt1[i]); | |
1435 | if(needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); | |
1436 | alloc_reg(current,i,CSREG); // Status | |
1437 | alloc_reg(current,i,FTEMP); | |
1438 | if(opcode[i]==0x35||opcode[i]==0x3d) { // 64-bit LDC1/SDC1 | |
1439 | alloc_reg64(current,i,FTEMP); | |
1440 | } | |
57871462 | 1441 | #if defined(HOST_IMM8) |
1442 | // On CPUs without 32-bit immediates we need a pointer to invalid_code | |
1443 | else if((opcode[i]&0x3b)==0x39) // SWC1/SDC1 | |
1444 | alloc_reg(current,i,INVCP); | |
1445 | #endif | |
1446 | // We need a temporary register for address generation | |
1447 | alloc_reg_temp(current,i,-1); | |
1448 | } | |
1449 | ||
b9b61529 | 1450 | void c2ls_alloc(struct regstat *current,int i) |
1451 | { | |
1452 | clear_const(current,rt1[i]); | |
1453 | if(needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]); | |
1454 | alloc_reg(current,i,FTEMP); | |
b9b61529 | 1455 | #if defined(HOST_IMM8) |
1456 | // On CPUs without 32-bit immediates we need a pointer to invalid_code | |
1edfcc68 | 1457 | if((opcode[i]&0x3b)==0x3a) // SWC2/SDC2 |
b9b61529 | 1458 | alloc_reg(current,i,INVCP); |
1459 | #endif | |
1460 | // We need a temporary register for address generation | |
1461 | alloc_reg_temp(current,i,-1); | |
e1190b87 | 1462 | minimum_free_regs[i]=1; |
b9b61529 | 1463 | } |
1464 | ||
57871462 | 1465 | #ifndef multdiv_alloc |
1466 | void multdiv_alloc(struct regstat *current,int i) | |
1467 | { | |
1468 | // case 0x18: MULT | |
1469 | // case 0x19: MULTU | |
1470 | // case 0x1A: DIV | |
1471 | // case 0x1B: DIVU | |
1472 | // case 0x1C: DMULT | |
1473 | // case 0x1D: DMULTU | |
1474 | // case 0x1E: DDIV | |
1475 | // case 0x1F: DDIVU | |
1476 | clear_const(current,rs1[i]); | |
1477 | clear_const(current,rs2[i]); | |
1478 | if(rs1[i]&&rs2[i]) | |
1479 | { | |
1480 | if((opcode2[i]&4)==0) // 32-bit | |
1481 | { | |
1482 | current->u&=~(1LL<<HIREG); | |
1483 | current->u&=~(1LL<<LOREG); | |
1484 | alloc_reg(current,i,HIREG); | |
1485 | alloc_reg(current,i,LOREG); | |
1486 | alloc_reg(current,i,rs1[i]); | |
1487 | alloc_reg(current,i,rs2[i]); | |
1488 | current->is32|=1LL<<HIREG; | |
1489 | current->is32|=1LL<<LOREG; | |
1490 | dirty_reg(current,HIREG); | |
1491 | dirty_reg(current,LOREG); | |
1492 | } | |
1493 | else // 64-bit | |
1494 | { | |
1495 | current->u&=~(1LL<<HIREG); | |
1496 | current->u&=~(1LL<<LOREG); | |
1497 | current->uu&=~(1LL<<HIREG); | |
1498 | current->uu&=~(1LL<<LOREG); | |
1499 | alloc_reg64(current,i,HIREG); | |
1500 | //if(HOST_REGS>10) alloc_reg64(current,i,LOREG); | |
1501 | alloc_reg64(current,i,rs1[i]); | |
1502 | alloc_reg64(current,i,rs2[i]); | |
1503 | alloc_all(current,i); | |
1504 | current->is32&=~(1LL<<HIREG); | |
1505 | current->is32&=~(1LL<<LOREG); | |
1506 | dirty_reg(current,HIREG); | |
1507 | dirty_reg(current,LOREG); | |
e1190b87 | 1508 | minimum_free_regs[i]=HOST_REGS; |
57871462 | 1509 | } |
1510 | } | |
1511 | else | |
1512 | { | |
1513 | // Multiply by zero is zero. | |
1514 | // MIPS does not have a divide by zero exception. | |
1515 | // The result is undefined, we return zero. | |
1516 | alloc_reg(current,i,HIREG); | |
1517 | alloc_reg(current,i,LOREG); | |
1518 | current->is32|=1LL<<HIREG; | |
1519 | current->is32|=1LL<<LOREG; | |
1520 | dirty_reg(current,HIREG); | |
1521 | dirty_reg(current,LOREG); | |
1522 | } | |
1523 | } | |
1524 | #endif | |
1525 | ||
1526 | void cop0_alloc(struct regstat *current,int i) | |
1527 | { | |
1528 | if(opcode2[i]==0) // MFC0 | |
1529 | { | |
1530 | if(rt1[i]) { | |
1531 | clear_const(current,rt1[i]); | |
1532 | alloc_all(current,i); | |
1533 | alloc_reg(current,i,rt1[i]); | |
1534 | current->is32|=1LL<<rt1[i]; | |
1535 | dirty_reg(current,rt1[i]); | |
1536 | } | |
1537 | } | |
1538 | else if(opcode2[i]==4) // MTC0 | |
1539 | { | |
1540 | if(rs1[i]){ | |
1541 | clear_const(current,rs1[i]); | |
1542 | alloc_reg(current,i,rs1[i]); | |
1543 | alloc_all(current,i); | |
1544 | } | |
1545 | else { | |
1546 | alloc_all(current,i); // FIXME: Keep r0 | |
1547 | current->u&=~1LL; | |
1548 | alloc_reg(current,i,0); | |
1549 | } | |
1550 | } | |
1551 | else | |
1552 | { | |
1553 | // TLBR/TLBWI/TLBWR/TLBP/ERET | |
1554 | assert(opcode2[i]==0x10); | |
1555 | alloc_all(current,i); | |
1556 | } | |
e1190b87 | 1557 | minimum_free_regs[i]=HOST_REGS; |
57871462 | 1558 | } |
1559 | ||
1560 | void cop1_alloc(struct regstat *current,int i) | |
1561 | { | |
1562 | alloc_reg(current,i,CSREG); // Load status | |
1563 | if(opcode2[i]<3) // MFC1/DMFC1/CFC1 | |
1564 | { | |
7de557a6 | 1565 | if(rt1[i]){ |
1566 | clear_const(current,rt1[i]); | |
1567 | if(opcode2[i]==1) { | |
1568 | alloc_reg64(current,i,rt1[i]); // DMFC1 | |
1569 | current->is32&=~(1LL<<rt1[i]); | |
1570 | }else{ | |
1571 | alloc_reg(current,i,rt1[i]); // MFC1/CFC1 | |
1572 | current->is32|=1LL<<rt1[i]; | |
1573 | } | |
1574 | dirty_reg(current,rt1[i]); | |
57871462 | 1575 | } |
57871462 | 1576 | alloc_reg_temp(current,i,-1); |
1577 | } | |
1578 | else if(opcode2[i]>3) // MTC1/DMTC1/CTC1 | |
1579 | { | |
1580 | if(rs1[i]){ | |
1581 | clear_const(current,rs1[i]); | |
1582 | if(opcode2[i]==5) | |
1583 | alloc_reg64(current,i,rs1[i]); // DMTC1 | |
1584 | else | |
1585 | alloc_reg(current,i,rs1[i]); // MTC1/CTC1 | |
1586 | alloc_reg_temp(current,i,-1); | |
1587 | } | |
1588 | else { | |
1589 | current->u&=~1LL; | |
1590 | alloc_reg(current,i,0); | |
1591 | alloc_reg_temp(current,i,-1); | |
1592 | } | |
1593 | } | |
e1190b87 | 1594 | minimum_free_regs[i]=1; |
57871462 | 1595 | } |
1596 | void fconv_alloc(struct regstat *current,int i) | |
1597 | { | |
1598 | alloc_reg(current,i,CSREG); // Load status | |
1599 | alloc_reg_temp(current,i,-1); | |
e1190b87 | 1600 | minimum_free_regs[i]=1; |
57871462 | 1601 | } |
1602 | void float_alloc(struct regstat *current,int i) | |
1603 | { | |
1604 | alloc_reg(current,i,CSREG); // Load status | |
1605 | alloc_reg_temp(current,i,-1); | |
e1190b87 | 1606 | minimum_free_regs[i]=1; |
57871462 | 1607 | } |
b9b61529 | 1608 | void c2op_alloc(struct regstat *current,int i) |
1609 | { | |
1610 | alloc_reg_temp(current,i,-1); | |
1611 | } | |
57871462 | 1612 | void fcomp_alloc(struct regstat *current,int i) |
1613 | { | |
1614 | alloc_reg(current,i,CSREG); // Load status | |
1615 | alloc_reg(current,i,FSREG); // Load flags | |
1616 | dirty_reg(current,FSREG); // Flag will be modified | |
1617 | alloc_reg_temp(current,i,-1); | |
e1190b87 | 1618 | minimum_free_regs[i]=1; |
57871462 | 1619 | } |
1620 | ||
1621 | void syscall_alloc(struct regstat *current,int i) | |
1622 | { | |
1623 | alloc_cc(current,i); | |
1624 | dirty_reg(current,CCREG); | |
1625 | alloc_all(current,i); | |
e1190b87 | 1626 | minimum_free_regs[i]=HOST_REGS; |
57871462 | 1627 | current->isconst=0; |
1628 | } | |
1629 | ||
1630 | void delayslot_alloc(struct regstat *current,int i) | |
1631 | { | |
1632 | switch(itype[i]) { | |
1633 | case UJUMP: | |
1634 | case CJUMP: | |
1635 | case SJUMP: | |
1636 | case RJUMP: | |
1637 | case FJUMP: | |
1638 | case SYSCALL: | |
7139f3c8 | 1639 | case HLECALL: |
57871462 | 1640 | case SPAN: |
1641 | assem_debug("jump in the delay slot. this shouldn't happen.\n");//exit(1); | |
c43b5311 | 1642 | SysPrintf("Disabled speculative precompilation\n"); |
57871462 | 1643 | stop_after_jal=1; |
1644 | break; | |
1645 | case IMM16: | |
1646 | imm16_alloc(current,i); | |
1647 | break; | |
1648 | case LOAD: | |
1649 | case LOADLR: | |
1650 | load_alloc(current,i); | |
1651 | break; | |
1652 | case STORE: | |
1653 | case STORELR: | |
1654 | store_alloc(current,i); | |
1655 | break; | |
1656 | case ALU: | |
1657 | alu_alloc(current,i); | |
1658 | break; | |
1659 | case SHIFT: | |
1660 | shift_alloc(current,i); | |
1661 | break; | |
1662 | case MULTDIV: | |
1663 | multdiv_alloc(current,i); | |
1664 | break; | |
1665 | case SHIFTIMM: | |
1666 | shiftimm_alloc(current,i); | |
1667 | break; | |
1668 | case MOV: | |
1669 | mov_alloc(current,i); | |
1670 | break; | |
1671 | case COP0: | |
1672 | cop0_alloc(current,i); | |
1673 | break; | |
1674 | case COP1: | |
b9b61529 | 1675 | case COP2: |
57871462 | 1676 | cop1_alloc(current,i); |
1677 | break; | |
1678 | case C1LS: | |
1679 | c1ls_alloc(current,i); | |
1680 | break; | |
b9b61529 | 1681 | case C2LS: |
1682 | c2ls_alloc(current,i); | |
1683 | break; | |
57871462 | 1684 | case FCONV: |
1685 | fconv_alloc(current,i); | |
1686 | break; | |
1687 | case FLOAT: | |
1688 | float_alloc(current,i); | |
1689 | break; | |
1690 | case FCOMP: | |
1691 | fcomp_alloc(current,i); | |
1692 | break; | |
b9b61529 | 1693 | case C2OP: |
1694 | c2op_alloc(current,i); | |
1695 | break; | |
57871462 | 1696 | } |
1697 | } | |
1698 | ||
1699 | // Special case where a branch and delay slot span two pages in virtual memory | |
1700 | static void pagespan_alloc(struct regstat *current,int i) | |
1701 | { | |
1702 | current->isconst=0; | |
1703 | current->wasconst=0; | |
1704 | regs[i].wasconst=0; | |
e1190b87 | 1705 | minimum_free_regs[i]=HOST_REGS; |
57871462 | 1706 | alloc_all(current,i); |
1707 | alloc_cc(current,i); | |
1708 | dirty_reg(current,CCREG); | |
1709 | if(opcode[i]==3) // JAL | |
1710 | { | |
1711 | alloc_reg(current,i,31); | |
1712 | dirty_reg(current,31); | |
1713 | } | |
1714 | if(opcode[i]==0&&(opcode2[i]&0x3E)==8) // JR/JALR | |
1715 | { | |
1716 | alloc_reg(current,i,rs1[i]); | |
5067f341 | 1717 | if (rt1[i]!=0) { |
1718 | alloc_reg(current,i,rt1[i]); | |
1719 | dirty_reg(current,rt1[i]); | |
57871462 | 1720 | } |
1721 | } | |
1722 | if((opcode[i]&0x2E)==4) // BEQ/BNE/BEQL/BNEL | |
1723 | { | |
1724 | if(rs1[i]) alloc_reg(current,i,rs1[i]); | |
1725 | if(rs2[i]) alloc_reg(current,i,rs2[i]); | |
1726 | if(!((current->is32>>rs1[i])&(current->is32>>rs2[i])&1)) | |
1727 | { | |
1728 | if(rs1[i]) alloc_reg64(current,i,rs1[i]); | |
1729 | if(rs2[i]) alloc_reg64(current,i,rs2[i]); | |
1730 | } | |
1731 | } | |
1732 | else | |
1733 | if((opcode[i]&0x2E)==6) // BLEZ/BGTZ/BLEZL/BGTZL | |
1734 | { | |
1735 | if(rs1[i]) alloc_reg(current,i,rs1[i]); | |
1736 | if(!((current->is32>>rs1[i])&1)) | |
1737 | { | |
1738 | if(rs1[i]) alloc_reg64(current,i,rs1[i]); | |
1739 | } | |
1740 | } | |
1741 | else | |
1742 | if(opcode[i]==0x11) // BC1 | |
1743 | { | |
1744 | alloc_reg(current,i,FSREG); | |
1745 | alloc_reg(current,i,CSREG); | |
1746 | } | |
1747 | //else ... | |
1748 | } | |
1749 | ||
1750 | add_stub(int type,int addr,int retaddr,int a,int b,int c,int d,int e) | |
1751 | { | |
1752 | stubs[stubcount][0]=type; | |
1753 | stubs[stubcount][1]=addr; | |
1754 | stubs[stubcount][2]=retaddr; | |
1755 | stubs[stubcount][3]=a; | |
1756 | stubs[stubcount][4]=b; | |
1757 | stubs[stubcount][5]=c; | |
1758 | stubs[stubcount][6]=d; | |
1759 | stubs[stubcount][7]=e; | |
1760 | stubcount++; | |
1761 | } | |
1762 | ||
1763 | // Write out a single register | |
1764 | void wb_register(signed char r,signed char regmap[],uint64_t dirty,uint64_t is32) | |
1765 | { | |
1766 | int hr; | |
1767 | for(hr=0;hr<HOST_REGS;hr++) { | |
1768 | if(hr!=EXCLUDE_REG) { | |
1769 | if((regmap[hr]&63)==r) { | |
1770 | if((dirty>>hr)&1) { | |
1771 | if(regmap[hr]<64) { | |
1772 | emit_storereg(r,hr); | |
57871462 | 1773 | }else{ |
1774 | emit_storereg(r|64,hr); | |
1775 | } | |
1776 | } | |
1777 | } | |
1778 | } | |
1779 | } | |
1780 | } | |
1781 | ||
1782 | int mchecksum() | |
1783 | { | |
1784 | //if(!tracedebug) return 0; | |
1785 | int i; | |
1786 | int sum=0; | |
1787 | for(i=0;i<2097152;i++) { | |
1788 | unsigned int temp=sum; | |
1789 | sum<<=1; | |
1790 | sum|=(~temp)>>31; | |
1791 | sum^=((u_int *)rdram)[i]; | |
1792 | } | |
1793 | return sum; | |
1794 | } | |
1795 | int rchecksum() | |
1796 | { | |
1797 | int i; | |
1798 | int sum=0; | |
1799 | for(i=0;i<64;i++) | |
1800 | sum^=((u_int *)reg)[i]; | |
1801 | return sum; | |
1802 | } | |
57871462 | 1803 | void rlist() |
1804 | { | |
1805 | int i; | |
1806 | printf("TRACE: "); | |
1807 | for(i=0;i<32;i++) | |
1808 | printf("r%d:%8x%8x ",i,((int *)(reg+i))[1],((int *)(reg+i))[0]); | |
1809 | printf("\n"); | |
57871462 | 1810 | } |
1811 | ||
1812 | void enabletrace() | |
1813 | { | |
1814 | tracedebug=1; | |
1815 | } | |
1816 | ||
1817 | void memdebug(int i) | |
1818 | { | |
1819 | //printf("TRACE: count=%d next=%d (checksum %x) lo=%8x%8x\n",Count,next_interupt,mchecksum(),(int)(reg[LOREG]>>32),(int)reg[LOREG]); | |
1820 | //printf("TRACE: count=%d next=%d (rchecksum %x)\n",Count,next_interupt,rchecksum()); | |
1821 | //rlist(); | |
1822 | //if(tracedebug) { | |
1823 | //if(Count>=-2084597794) { | |
1824 | if((signed int)Count>=-2084597794&&(signed int)Count<0) { | |
1825 | //if(0) { | |
1826 | printf("TRACE: count=%d next=%d (checksum %x)\n",Count,next_interupt,mchecksum()); | |
1827 | //printf("TRACE: count=%d next=%d (checksum %x) Status=%x\n",Count,next_interupt,mchecksum(),Status); | |
1828 | //printf("TRACE: count=%d next=%d (checksum %x) hi=%8x%8x\n",Count,next_interupt,mchecksum(),(int)(reg[HIREG]>>32),(int)reg[HIREG]); | |
1829 | rlist(); | |
1830 | #ifdef __i386__ | |
1831 | printf("TRACE: %x\n",(&i)[-1]); | |
1832 | #endif | |
1833 | #ifdef __arm__ | |
1834 | int j; | |
1835 | printf("TRACE: %x \n",(&j)[10]); | |
1836 | printf("TRACE: %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x\n",(&j)[1],(&j)[2],(&j)[3],(&j)[4],(&j)[5],(&j)[6],(&j)[7],(&j)[8],(&j)[9],(&j)[10],(&j)[11],(&j)[12],(&j)[13],(&j)[14],(&j)[15],(&j)[16],(&j)[17],(&j)[18],(&j)[19],(&j)[20]); | |
1837 | #endif | |
1838 | //fflush(stdout); | |
1839 | } | |
1840 | //printf("TRACE: %x\n",(&i)[-1]); | |
1841 | } | |
1842 | ||
57871462 | 1843 | void alu_assemble(int i,struct regstat *i_regs) |
1844 | { | |
1845 | if(opcode2[i]>=0x20&&opcode2[i]<=0x23) { // ADD/ADDU/SUB/SUBU | |
1846 | if(rt1[i]) { | |
1847 | signed char s1,s2,t; | |
1848 | t=get_reg(i_regs->regmap,rt1[i]); | |
1849 | if(t>=0) { | |
1850 | s1=get_reg(i_regs->regmap,rs1[i]); | |
1851 | s2=get_reg(i_regs->regmap,rs2[i]); | |
1852 | if(rs1[i]&&rs2[i]) { | |
1853 | assert(s1>=0); | |
1854 | assert(s2>=0); | |
1855 | if(opcode2[i]&2) emit_sub(s1,s2,t); | |
1856 | else emit_add(s1,s2,t); | |
1857 | } | |
1858 | else if(rs1[i]) { | |
1859 | if(s1>=0) emit_mov(s1,t); | |
1860 | else emit_loadreg(rs1[i],t); | |
1861 | } | |
1862 | else if(rs2[i]) { | |
1863 | if(s2>=0) { | |
1864 | if(opcode2[i]&2) emit_neg(s2,t); | |
1865 | else emit_mov(s2,t); | |
1866 | } | |
1867 | else { | |
1868 | emit_loadreg(rs2[i],t); | |
1869 | if(opcode2[i]&2) emit_neg(t,t); | |
1870 | } | |
1871 | } | |
1872 | else emit_zeroreg(t); | |
1873 | } | |
1874 | } | |
1875 | } | |
1876 | if(opcode2[i]>=0x2c&&opcode2[i]<=0x2f) { // DADD/DADDU/DSUB/DSUBU | |
1877 | if(rt1[i]) { | |
1878 | signed char s1l,s2l,s1h,s2h,tl,th; | |
1879 | tl=get_reg(i_regs->regmap,rt1[i]); | |
1880 | th=get_reg(i_regs->regmap,rt1[i]|64); | |
1881 | if(tl>=0) { | |
1882 | s1l=get_reg(i_regs->regmap,rs1[i]); | |
1883 | s2l=get_reg(i_regs->regmap,rs2[i]); | |
1884 | s1h=get_reg(i_regs->regmap,rs1[i]|64); | |
1885 | s2h=get_reg(i_regs->regmap,rs2[i]|64); | |
1886 | if(rs1[i]&&rs2[i]) { | |
1887 | assert(s1l>=0); | |
1888 | assert(s2l>=0); | |
1889 | if(opcode2[i]&2) emit_subs(s1l,s2l,tl); | |
1890 | else emit_adds(s1l,s2l,tl); | |
1891 | if(th>=0) { | |
1892 | #ifdef INVERTED_CARRY | |
1893 | if(opcode2[i]&2) {if(s1h!=th) emit_mov(s1h,th);emit_sbb(th,s2h);} | |
1894 | #else | |
1895 | if(opcode2[i]&2) emit_sbc(s1h,s2h,th); | |
1896 | #endif | |
1897 | else emit_add(s1h,s2h,th); | |
1898 | } | |
1899 | } | |
1900 | else if(rs1[i]) { | |
1901 | if(s1l>=0) emit_mov(s1l,tl); | |
1902 | else emit_loadreg(rs1[i],tl); | |
1903 | if(th>=0) { | |
1904 | if(s1h>=0) emit_mov(s1h,th); | |
1905 | else emit_loadreg(rs1[i]|64,th); | |
1906 | } | |
1907 | } | |
1908 | else if(rs2[i]) { | |
1909 | if(s2l>=0) { | |
1910 | if(opcode2[i]&2) emit_negs(s2l,tl); | |
1911 | else emit_mov(s2l,tl); | |
1912 | } | |
1913 | else { | |
1914 | emit_loadreg(rs2[i],tl); | |
1915 | if(opcode2[i]&2) emit_negs(tl,tl); | |
1916 | } | |
1917 | if(th>=0) { | |
1918 | #ifdef INVERTED_CARRY | |
1919 | if(s2h>=0) emit_mov(s2h,th); | |
1920 | else emit_loadreg(rs2[i]|64,th); | |
1921 | if(opcode2[i]&2) { | |
1922 | emit_adcimm(-1,th); // x86 has inverted carry flag | |
1923 | emit_not(th,th); | |
1924 | } | |
1925 | #else | |
1926 | if(opcode2[i]&2) { | |
1927 | if(s2h>=0) emit_rscimm(s2h,0,th); | |
1928 | else { | |
1929 | emit_loadreg(rs2[i]|64,th); | |
1930 | emit_rscimm(th,0,th); | |
1931 | } | |
1932 | }else{ | |
1933 | if(s2h>=0) emit_mov(s2h,th); | |
1934 | else emit_loadreg(rs2[i]|64,th); | |
1935 | } | |
1936 | #endif | |
1937 | } | |
1938 | } | |
1939 | else { | |
1940 | emit_zeroreg(tl); | |
1941 | if(th>=0) emit_zeroreg(th); | |
1942 | } | |
1943 | } | |
1944 | } | |
1945 | } | |
1946 | if(opcode2[i]==0x2a||opcode2[i]==0x2b) { // SLT/SLTU | |
1947 | if(rt1[i]) { | |
1948 | signed char s1l,s1h,s2l,s2h,t; | |
1949 | if(!((i_regs->was32>>rs1[i])&(i_regs->was32>>rs2[i])&1)) | |
1950 | { | |
1951 | t=get_reg(i_regs->regmap,rt1[i]); | |
1952 | //assert(t>=0); | |
1953 | if(t>=0) { | |
1954 | s1l=get_reg(i_regs->regmap,rs1[i]); | |
1955 | s1h=get_reg(i_regs->regmap,rs1[i]|64); | |
1956 | s2l=get_reg(i_regs->regmap,rs2[i]); | |
1957 | s2h=get_reg(i_regs->regmap,rs2[i]|64); | |
1958 | if(rs2[i]==0) // rx<r0 | |
1959 | { | |
1960 | assert(s1h>=0); | |
1961 | if(opcode2[i]==0x2a) // SLT | |
1962 | emit_shrimm(s1h,31,t); | |
1963 | else // SLTU (unsigned can not be less than zero) | |
1964 | emit_zeroreg(t); | |
1965 | } | |
1966 | else if(rs1[i]==0) // r0<rx | |
1967 | { | |
1968 | assert(s2h>=0); | |
1969 | if(opcode2[i]==0x2a) // SLT | |
1970 | emit_set_gz64_32(s2h,s2l,t); | |
1971 | else // SLTU (set if not zero) | |
1972 | emit_set_nz64_32(s2h,s2l,t); | |
1973 | } | |
1974 | else { | |
1975 | assert(s1l>=0);assert(s1h>=0); | |
1976 | assert(s2l>=0);assert(s2h>=0); | |
1977 | if(opcode2[i]==0x2a) // SLT | |
1978 | emit_set_if_less64_32(s1h,s1l,s2h,s2l,t); | |
1979 | else // SLTU | |
1980 | emit_set_if_carry64_32(s1h,s1l,s2h,s2l,t); | |
1981 | } | |
1982 | } | |
1983 | } else { | |
1984 | t=get_reg(i_regs->regmap,rt1[i]); | |
1985 | //assert(t>=0); | |
1986 | if(t>=0) { | |
1987 | s1l=get_reg(i_regs->regmap,rs1[i]); | |
1988 | s2l=get_reg(i_regs->regmap,rs2[i]); | |
1989 | if(rs2[i]==0) // rx<r0 | |
1990 | { | |
1991 | assert(s1l>=0); | |
1992 | if(opcode2[i]==0x2a) // SLT | |
1993 | emit_shrimm(s1l,31,t); | |
1994 | else // SLTU (unsigned can not be less than zero) | |
1995 | emit_zeroreg(t); | |
1996 | } | |
1997 | else if(rs1[i]==0) // r0<rx | |
1998 | { | |
1999 | assert(s2l>=0); | |
2000 | if(opcode2[i]==0x2a) // SLT | |
2001 | emit_set_gz32(s2l,t); | |
2002 | else // SLTU (set if not zero) | |
2003 | emit_set_nz32(s2l,t); | |
2004 | } | |
2005 | else{ | |
2006 | assert(s1l>=0);assert(s2l>=0); | |
2007 | if(opcode2[i]==0x2a) // SLT | |
2008 | emit_set_if_less32(s1l,s2l,t); | |
2009 | else // SLTU | |
2010 | emit_set_if_carry32(s1l,s2l,t); | |
2011 | } | |
2012 | } | |
2013 | } | |
2014 | } | |
2015 | } | |
2016 | if(opcode2[i]>=0x24&&opcode2[i]<=0x27) { // AND/OR/XOR/NOR | |
2017 | if(rt1[i]) { | |
2018 | signed char s1l,s1h,s2l,s2h,th,tl; | |
2019 | tl=get_reg(i_regs->regmap,rt1[i]); | |
2020 | th=get_reg(i_regs->regmap,rt1[i]|64); | |
2021 | if(!((i_regs->was32>>rs1[i])&(i_regs->was32>>rs2[i])&1)&&th>=0) | |
2022 | { | |
2023 | assert(tl>=0); | |
2024 | if(tl>=0) { | |
2025 | s1l=get_reg(i_regs->regmap,rs1[i]); | |
2026 | s1h=get_reg(i_regs->regmap,rs1[i]|64); | |
2027 | s2l=get_reg(i_regs->regmap,rs2[i]); | |
2028 | s2h=get_reg(i_regs->regmap,rs2[i]|64); | |
2029 | if(rs1[i]&&rs2[i]) { | |
2030 | assert(s1l>=0);assert(s1h>=0); | |
2031 | assert(s2l>=0);assert(s2h>=0); | |
2032 | if(opcode2[i]==0x24) { // AND | |
2033 | emit_and(s1l,s2l,tl); | |
2034 | emit_and(s1h,s2h,th); | |
2035 | } else | |
2036 | if(opcode2[i]==0x25) { // OR | |
2037 | emit_or(s1l,s2l,tl); | |
2038 | emit_or(s1h,s2h,th); | |
2039 | } else | |
2040 | if(opcode2[i]==0x26) { // XOR | |
2041 | emit_xor(s1l,s2l,tl); | |
2042 | emit_xor(s1h,s2h,th); | |
2043 | } else | |
2044 | if(opcode2[i]==0x27) { // NOR | |
2045 | emit_or(s1l,s2l,tl); | |
2046 | emit_or(s1h,s2h,th); | |
2047 | emit_not(tl,tl); | |
2048 | emit_not(th,th); | |
2049 | } | |
2050 | } | |
2051 | else | |
2052 | { | |
2053 | if(opcode2[i]==0x24) { // AND | |
2054 | emit_zeroreg(tl); | |
2055 | emit_zeroreg(th); | |
2056 | } else | |
2057 | if(opcode2[i]==0x25||opcode2[i]==0x26) { // OR/XOR | |
2058 | if(rs1[i]){ | |
2059 | if(s1l>=0) emit_mov(s1l,tl); | |
2060 | else emit_loadreg(rs1[i],tl); | |
2061 | if(s1h>=0) emit_mov(s1h,th); | |
2062 | else emit_loadreg(rs1[i]|64,th); | |
2063 | } | |
2064 | else | |
2065 | if(rs2[i]){ | |
2066 | if(s2l>=0) emit_mov(s2l,tl); | |
2067 | else emit_loadreg(rs2[i],tl); | |
2068 | if(s2h>=0) emit_mov(s2h,th); | |
2069 | else emit_loadreg(rs2[i]|64,th); | |
2070 | } | |
2071 | else{ | |
2072 | emit_zeroreg(tl); | |
2073 | emit_zeroreg(th); | |
2074 | } | |
2075 | } else | |
2076 | if(opcode2[i]==0x27) { // NOR | |
2077 | if(rs1[i]){ | |
2078 | if(s1l>=0) emit_not(s1l,tl); | |
2079 | else{ | |
2080 | emit_loadreg(rs1[i],tl); | |
2081 | emit_not(tl,tl); | |
2082 | } | |
2083 | if(s1h>=0) emit_not(s1h,th); | |
2084 | else{ | |
2085 | emit_loadreg(rs1[i]|64,th); | |
2086 | emit_not(th,th); | |
2087 | } | |
2088 | } | |
2089 | else | |
2090 | if(rs2[i]){ | |
2091 | if(s2l>=0) emit_not(s2l,tl); | |
2092 | else{ | |
2093 | emit_loadreg(rs2[i],tl); | |
2094 | emit_not(tl,tl); | |
2095 | } | |
2096 | if(s2h>=0) emit_not(s2h,th); | |
2097 | else{ | |
2098 | emit_loadreg(rs2[i]|64,th); | |
2099 | emit_not(th,th); | |
2100 | } | |
2101 | } | |
2102 | else { | |
2103 | emit_movimm(-1,tl); | |
2104 | emit_movimm(-1,th); | |
2105 | } | |
2106 | } | |
2107 | } | |
2108 | } | |
2109 | } | |
2110 | else | |
2111 | { | |
2112 | // 32 bit | |
2113 | if(tl>=0) { | |
2114 | s1l=get_reg(i_regs->regmap,rs1[i]); | |
2115 | s2l=get_reg(i_regs->regmap,rs2[i]); | |
2116 | if(rs1[i]&&rs2[i]) { | |
2117 | assert(s1l>=0); | |
2118 | assert(s2l>=0); | |
2119 | if(opcode2[i]==0x24) { // AND | |
2120 | emit_and(s1l,s2l,tl); | |
2121 | } else | |
2122 | if(opcode2[i]==0x25) { // OR | |
2123 | emit_or(s1l,s2l,tl); | |
2124 | } else | |
2125 | if(opcode2[i]==0x26) { // XOR | |
2126 | emit_xor(s1l,s2l,tl); | |
2127 | } else | |
2128 | if(opcode2[i]==0x27) { // NOR | |
2129 | emit_or(s1l,s2l,tl); | |
2130 | emit_not(tl,tl); | |
2131 | } | |
2132 | } | |
2133 | else | |
2134 | { | |
2135 | if(opcode2[i]==0x24) { // AND | |
2136 | emit_zeroreg(tl); | |
2137 | } else | |
2138 | if(opcode2[i]==0x25||opcode2[i]==0x26) { // OR/XOR | |
2139 | if(rs1[i]){ | |
2140 | if(s1l>=0) emit_mov(s1l,tl); | |
2141 | else emit_loadreg(rs1[i],tl); // CHECK: regmap_entry? | |
2142 | } | |
2143 | else | |
2144 | if(rs2[i]){ | |
2145 | if(s2l>=0) emit_mov(s2l,tl); | |
2146 | else emit_loadreg(rs2[i],tl); // CHECK: regmap_entry? | |
2147 | } | |
2148 | else emit_zeroreg(tl); | |
2149 | } else | |
2150 | if(opcode2[i]==0x27) { // NOR | |
2151 | if(rs1[i]){ | |
2152 | if(s1l>=0) emit_not(s1l,tl); | |
2153 | else { | |
2154 | emit_loadreg(rs1[i],tl); | |
2155 | emit_not(tl,tl); | |
2156 | } | |
2157 | } | |
2158 | else | |
2159 | if(rs2[i]){ | |
2160 | if(s2l>=0) emit_not(s2l,tl); | |
2161 | else { | |
2162 | emit_loadreg(rs2[i],tl); | |
2163 | emit_not(tl,tl); | |
2164 | } | |
2165 | } | |
2166 | else emit_movimm(-1,tl); | |
2167 | } | |
2168 | } | |
2169 | } | |
2170 | } | |
2171 | } | |
2172 | } | |
2173 | } | |
2174 | ||
2175 | void imm16_assemble(int i,struct regstat *i_regs) | |
2176 | { | |
2177 | if (opcode[i]==0x0f) { // LUI | |
2178 | if(rt1[i]) { | |
2179 | signed char t; | |
2180 | t=get_reg(i_regs->regmap,rt1[i]); | |
2181 | //assert(t>=0); | |
2182 | if(t>=0) { | |
2183 | if(!((i_regs->isconst>>t)&1)) | |
2184 | emit_movimm(imm[i]<<16,t); | |
2185 | } | |
2186 | } | |
2187 | } | |
2188 | if(opcode[i]==0x08||opcode[i]==0x09) { // ADDI/ADDIU | |
2189 | if(rt1[i]) { | |
2190 | signed char s,t; | |
2191 | t=get_reg(i_regs->regmap,rt1[i]); | |
2192 | s=get_reg(i_regs->regmap,rs1[i]); | |
2193 | if(rs1[i]) { | |
2194 | //assert(t>=0); | |
2195 | //assert(s>=0); | |
2196 | if(t>=0) { | |
2197 | if(!((i_regs->isconst>>t)&1)) { | |
2198 | if(s<0) { | |
2199 | if(i_regs->regmap_entry[t]!=rs1[i]) emit_loadreg(rs1[i],t); | |
2200 | emit_addimm(t,imm[i],t); | |
2201 | }else{ | |
2202 | if(!((i_regs->wasconst>>s)&1)) | |
2203 | emit_addimm(s,imm[i],t); | |
2204 | else | |
2205 | emit_movimm(constmap[i][s]+imm[i],t); | |
2206 | } | |
2207 | } | |
2208 | } | |
2209 | } else { | |
2210 | if(t>=0) { | |
2211 | if(!((i_regs->isconst>>t)&1)) | |
2212 | emit_movimm(imm[i],t); | |
2213 | } | |
2214 | } | |
2215 | } | |
2216 | } | |
2217 | if(opcode[i]==0x18||opcode[i]==0x19) { // DADDI/DADDIU | |
2218 | if(rt1[i]) { | |
2219 | signed char sh,sl,th,tl; | |
2220 | th=get_reg(i_regs->regmap,rt1[i]|64); | |
2221 | tl=get_reg(i_regs->regmap,rt1[i]); | |
2222 | sh=get_reg(i_regs->regmap,rs1[i]|64); | |
2223 | sl=get_reg(i_regs->regmap,rs1[i]); | |
2224 | if(tl>=0) { | |
2225 | if(rs1[i]) { | |
2226 | assert(sh>=0); | |
2227 | assert(sl>=0); | |
2228 | if(th>=0) { | |
2229 | emit_addimm64_32(sh,sl,imm[i],th,tl); | |
2230 | } | |
2231 | else { | |
2232 | emit_addimm(sl,imm[i],tl); | |
2233 | } | |
2234 | } else { | |
2235 | emit_movimm(imm[i],tl); | |
2236 | if(th>=0) emit_movimm(((signed int)imm[i])>>31,th); | |
2237 | } | |
2238 | } | |
2239 | } | |
2240 | } | |
2241 | else if(opcode[i]==0x0a||opcode[i]==0x0b) { // SLTI/SLTIU | |
2242 | if(rt1[i]) { | |
2243 | //assert(rs1[i]!=0); // r0 might be valid, but it's probably a bug | |
2244 | signed char sh,sl,t; | |
2245 | t=get_reg(i_regs->regmap,rt1[i]); | |
2246 | sh=get_reg(i_regs->regmap,rs1[i]|64); | |
2247 | sl=get_reg(i_regs->regmap,rs1[i]); | |
2248 | //assert(t>=0); | |
2249 | if(t>=0) { | |
2250 | if(rs1[i]>0) { | |
2251 | if(sh<0) assert((i_regs->was32>>rs1[i])&1); | |
2252 | if(sh<0||((i_regs->was32>>rs1[i])&1)) { | |
2253 | if(opcode[i]==0x0a) { // SLTI | |
2254 | if(sl<0) { | |
2255 | if(i_regs->regmap_entry[t]!=rs1[i]) emit_loadreg(rs1[i],t); | |
2256 | emit_slti32(t,imm[i],t); | |
2257 | }else{ | |
2258 | emit_slti32(sl,imm[i],t); | |
2259 | } | |
2260 | } | |
2261 | else { // SLTIU | |
2262 | if(sl<0) { | |
2263 | if(i_regs->regmap_entry[t]!=rs1[i]) emit_loadreg(rs1[i],t); | |
2264 | emit_sltiu32(t,imm[i],t); | |
2265 | }else{ | |
2266 | emit_sltiu32(sl,imm[i],t); | |
2267 | } | |
2268 | } | |
2269 | }else{ // 64-bit | |
2270 | assert(sl>=0); | |
2271 | if(opcode[i]==0x0a) // SLTI | |
2272 | emit_slti64_32(sh,sl,imm[i],t); | |
2273 | else // SLTIU | |
2274 | emit_sltiu64_32(sh,sl,imm[i],t); | |
2275 | } | |
2276 | }else{ | |
2277 | // SLTI(U) with r0 is just stupid, | |
2278 | // nonetheless examples can be found | |
2279 | if(opcode[i]==0x0a) // SLTI | |
2280 | if(0<imm[i]) emit_movimm(1,t); | |
2281 | else emit_zeroreg(t); | |
2282 | else // SLTIU | |
2283 | { | |
2284 | if(imm[i]) emit_movimm(1,t); | |
2285 | else emit_zeroreg(t); | |
2286 | } | |
2287 | } | |
2288 | } | |
2289 | } | |
2290 | } | |
2291 | else if(opcode[i]>=0x0c&&opcode[i]<=0x0e) { // ANDI/ORI/XORI | |
2292 | if(rt1[i]) { | |
2293 | signed char sh,sl,th,tl; | |
2294 | th=get_reg(i_regs->regmap,rt1[i]|64); | |
2295 | tl=get_reg(i_regs->regmap,rt1[i]); | |
2296 | sh=get_reg(i_regs->regmap,rs1[i]|64); | |
2297 | sl=get_reg(i_regs->regmap,rs1[i]); | |
2298 | if(tl>=0 && !((i_regs->isconst>>tl)&1)) { | |
2299 | if(opcode[i]==0x0c) //ANDI | |
2300 | { | |
2301 | if(rs1[i]) { | |
2302 | if(sl<0) { | |
2303 | if(i_regs->regmap_entry[tl]!=rs1[i]) emit_loadreg(rs1[i],tl); | |
2304 | emit_andimm(tl,imm[i],tl); | |
2305 | }else{ | |
2306 | if(!((i_regs->wasconst>>sl)&1)) | |
2307 | emit_andimm(sl,imm[i],tl); | |
2308 | else | |
2309 | emit_movimm(constmap[i][sl]&imm[i],tl); | |
2310 | } | |
2311 | } | |
2312 | else | |
2313 | emit_zeroreg(tl); | |
2314 | if(th>=0) emit_zeroreg(th); | |
2315 | } | |
2316 | else | |
2317 | { | |
2318 | if(rs1[i]) { | |
2319 | if(sl<0) { | |
2320 | if(i_regs->regmap_entry[tl]!=rs1[i]) emit_loadreg(rs1[i],tl); | |
2321 | } | |
2322 | if(th>=0) { | |
2323 | if(sh<0) { | |
2324 | emit_loadreg(rs1[i]|64,th); | |
2325 | }else{ | |
2326 | emit_mov(sh,th); | |
2327 | } | |
2328 | } | |
2329 | if(opcode[i]==0x0d) //ORI | |
2330 | if(sl<0) { | |
2331 | emit_orimm(tl,imm[i],tl); | |
2332 | }else{ | |
2333 | if(!((i_regs->wasconst>>sl)&1)) | |
2334 | emit_orimm(sl,imm[i],tl); | |
2335 | else | |
2336 | emit_movimm(constmap[i][sl]|imm[i],tl); | |
2337 | } | |
2338 | if(opcode[i]==0x0e) //XORI | |
2339 | if(sl<0) { | |
2340 | emit_xorimm(tl,imm[i],tl); | |
2341 | }else{ | |
2342 | if(!((i_regs->wasconst>>sl)&1)) | |
2343 | emit_xorimm(sl,imm[i],tl); | |
2344 | else | |
2345 | emit_movimm(constmap[i][sl]^imm[i],tl); | |
2346 | } | |
2347 | } | |
2348 | else { | |
2349 | emit_movimm(imm[i],tl); | |
2350 | if(th>=0) emit_zeroreg(th); | |
2351 | } | |
2352 | } | |
2353 | } | |
2354 | } | |
2355 | } | |
2356 | } | |
2357 | ||
2358 | void shiftimm_assemble(int i,struct regstat *i_regs) | |
2359 | { | |
2360 | if(opcode2[i]<=0x3) // SLL/SRL/SRA | |
2361 | { | |
2362 | if(rt1[i]) { | |
2363 | signed char s,t; | |
2364 | t=get_reg(i_regs->regmap,rt1[i]); | |
2365 | s=get_reg(i_regs->regmap,rs1[i]); | |
2366 | //assert(t>=0); | |
dc49e339 | 2367 | if(t>=0&&!((i_regs->isconst>>t)&1)){ |
57871462 | 2368 | if(rs1[i]==0) |
2369 | { | |
2370 | emit_zeroreg(t); | |
2371 | } | |
2372 | else | |
2373 | { | |
2374 | if(s<0&&i_regs->regmap_entry[t]!=rs1[i]) emit_loadreg(rs1[i],t); | |
2375 | if(imm[i]) { | |
2376 | if(opcode2[i]==0) // SLL | |
2377 | { | |
2378 | emit_shlimm(s<0?t:s,imm[i],t); | |
2379 | } | |
2380 | if(opcode2[i]==2) // SRL | |
2381 | { | |
2382 | emit_shrimm(s<0?t:s,imm[i],t); | |
2383 | } | |
2384 | if(opcode2[i]==3) // SRA | |
2385 | { | |
2386 | emit_sarimm(s<0?t:s,imm[i],t); | |
2387 | } | |
2388 | }else{ | |
2389 | // Shift by zero | |
2390 | if(s>=0 && s!=t) emit_mov(s,t); | |
2391 | } | |
2392 | } | |
2393 | } | |
2394 | //emit_storereg(rt1[i],t); //DEBUG | |
2395 | } | |
2396 | } | |
2397 | if(opcode2[i]>=0x38&&opcode2[i]<=0x3b) // DSLL/DSRL/DSRA | |
2398 | { | |
2399 | if(rt1[i]) { | |
2400 | signed char sh,sl,th,tl; | |
2401 | th=get_reg(i_regs->regmap,rt1[i]|64); | |
2402 | tl=get_reg(i_regs->regmap,rt1[i]); | |
2403 | sh=get_reg(i_regs->regmap,rs1[i]|64); | |
2404 | sl=get_reg(i_regs->regmap,rs1[i]); | |
2405 | if(tl>=0) { | |
2406 | if(rs1[i]==0) | |
2407 | { | |
2408 | emit_zeroreg(tl); | |
2409 | if(th>=0) emit_zeroreg(th); | |
2410 | } | |
2411 | else | |
2412 | { | |
2413 | assert(sl>=0); | |
2414 | assert(sh>=0); | |
2415 | if(imm[i]) { | |
2416 | if(opcode2[i]==0x38) // DSLL | |
2417 | { | |
2418 | if(th>=0) emit_shldimm(sh,sl,imm[i],th); | |
2419 | emit_shlimm(sl,imm[i],tl); | |
2420 | } | |
2421 | if(opcode2[i]==0x3a) // DSRL | |
2422 | { | |
2423 | emit_shrdimm(sl,sh,imm[i],tl); | |
2424 | if(th>=0) emit_shrimm(sh,imm[i],th); | |
2425 | } | |
2426 | if(opcode2[i]==0x3b) // DSRA | |
2427 | { | |
2428 | emit_shrdimm(sl,sh,imm[i],tl); | |
2429 | if(th>=0) emit_sarimm(sh,imm[i],th); | |
2430 | } | |
2431 | }else{ | |
2432 | // Shift by zero | |
2433 | if(sl!=tl) emit_mov(sl,tl); | |
2434 | if(th>=0&&sh!=th) emit_mov(sh,th); | |
2435 | } | |
2436 | } | |
2437 | } | |
2438 | } | |
2439 | } | |
2440 | if(opcode2[i]==0x3c) // DSLL32 | |
2441 | { | |
2442 | if(rt1[i]) { | |
2443 | signed char sl,tl,th; | |
2444 | tl=get_reg(i_regs->regmap,rt1[i]); | |
2445 | th=get_reg(i_regs->regmap,rt1[i]|64); | |
2446 | sl=get_reg(i_regs->regmap,rs1[i]); | |
2447 | if(th>=0||tl>=0){ | |
2448 | assert(tl>=0); | |
2449 | assert(th>=0); | |
2450 | assert(sl>=0); | |
2451 | emit_mov(sl,th); | |
2452 | emit_zeroreg(tl); | |
2453 | if(imm[i]>32) | |
2454 | { | |
2455 | emit_shlimm(th,imm[i]&31,th); | |
2456 | } | |
2457 | } | |
2458 | } | |
2459 | } | |
2460 | if(opcode2[i]==0x3e) // DSRL32 | |
2461 | { | |
2462 | if(rt1[i]) { | |
2463 | signed char sh,tl,th; | |
2464 | tl=get_reg(i_regs->regmap,rt1[i]); | |
2465 | th=get_reg(i_regs->regmap,rt1[i]|64); | |
2466 | sh=get_reg(i_regs->regmap,rs1[i]|64); | |
2467 | if(tl>=0){ | |
2468 | assert(sh>=0); | |
2469 | emit_mov(sh,tl); | |
2470 | if(th>=0) emit_zeroreg(th); | |
2471 | if(imm[i]>32) | |
2472 | { | |
2473 | emit_shrimm(tl,imm[i]&31,tl); | |
2474 | } | |
2475 | } | |
2476 | } | |
2477 | } | |
2478 | if(opcode2[i]==0x3f) // DSRA32 | |
2479 | { | |
2480 | if(rt1[i]) { | |
2481 | signed char sh,tl; | |
2482 | tl=get_reg(i_regs->regmap,rt1[i]); | |
2483 | sh=get_reg(i_regs->regmap,rs1[i]|64); | |
2484 | if(tl>=0){ | |
2485 | assert(sh>=0); | |
2486 | emit_mov(sh,tl); | |
2487 | if(imm[i]>32) | |
2488 | { | |
2489 | emit_sarimm(tl,imm[i]&31,tl); | |
2490 | } | |
2491 | } | |
2492 | } | |
2493 | } | |
2494 | } | |
2495 | ||
2496 | #ifndef shift_assemble | |
2497 | void shift_assemble(int i,struct regstat *i_regs) | |
2498 | { | |
2499 | printf("Need shift_assemble for this architecture.\n"); | |
2500 | exit(1); | |
2501 | } | |
2502 | #endif | |
2503 | ||
2504 | void load_assemble(int i,struct regstat *i_regs) | |
2505 | { | |
2506 | int s,th,tl,addr,map=-1; | |
2507 | int offset; | |
2508 | int jaddr=0; | |
5bf843dc | 2509 | int memtarget=0,c=0; |
b1570849 | 2510 | int fastload_reg_override=0; |
57871462 | 2511 | u_int hr,reglist=0; |
2512 | th=get_reg(i_regs->regmap,rt1[i]|64); | |
2513 | tl=get_reg(i_regs->regmap,rt1[i]); | |
2514 | s=get_reg(i_regs->regmap,rs1[i]); | |
2515 | offset=imm[i]; | |
2516 | for(hr=0;hr<HOST_REGS;hr++) { | |
2517 | if(i_regs->regmap[hr]>=0) reglist|=1<<hr; | |
2518 | } | |
2519 | if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<<HOST_CCREG); | |
2520 | if(s>=0) { | |
2521 | c=(i_regs->wasconst>>s)&1; | |
af4ee1fe | 2522 | if (c) { |
2523 | memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE; | |
af4ee1fe | 2524 | } |
57871462 | 2525 | } |
57871462 | 2526 | //printf("load_assemble: c=%d\n",c); |
2527 | //if(c) printf("load_assemble: const=%x\n",(int)constmap[i][s]+offset); | |
2528 | // FIXME: Even if the load is a NOP, we should check for pagefaults... | |
f18c0f46 | 2529 | if(tl<0&&(!c||(((u_int)constmap[i][s]+offset)>>16)==0x1f80) |
2530 | ||rt1[i]==0) { | |
5bf843dc | 2531 | // could be FIFO, must perform the read |
f18c0f46 | 2532 | // ||dummy read |
5bf843dc | 2533 | assem_debug("(forced read)\n"); |
2534 | tl=get_reg(i_regs->regmap,-1); | |
2535 | assert(tl>=0); | |
5bf843dc | 2536 | } |
2537 | if(offset||s<0||c) addr=tl; | |
2538 | else addr=s; | |
535d208a | 2539 | //if(tl<0) tl=get_reg(i_regs->regmap,-1); |
2540 | if(tl>=0) { | |
2541 | //printf("load_assemble: c=%d\n",c); | |
2542 | //if(c) printf("load_assemble: const=%x\n",(int)constmap[i][s]+offset); | |
2543 | assert(tl>=0); // Even if the load is a NOP, we must check for pagefaults and I/O | |
2544 | reglist&=~(1<<tl); | |
2545 | if(th>=0) reglist&=~(1<<th); | |
1edfcc68 | 2546 | if(!c) { |
2547 | #ifdef RAM_OFFSET | |
2548 | map=get_reg(i_regs->regmap,ROREG); | |
2549 | if(map<0) emit_loadreg(ROREG,map=HOST_TEMPREG); | |
2550 | #endif | |
2551 | #ifdef R29_HACK | |
2552 | // Strmnnrmn's speed hack | |
2553 | if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE) | |
2554 | #endif | |
2555 | { | |
2556 | jaddr=emit_fastpath_cmp_jump(i,addr,&fastload_reg_override); | |
535d208a | 2557 | } |
1edfcc68 | 2558 | } |
2559 | else if(ram_offset&&memtarget) { | |
2560 | emit_addimm(addr,ram_offset,HOST_TEMPREG); | |
2561 | fastload_reg_override=HOST_TEMPREG; | |
535d208a | 2562 | } |
2563 | int dummy=(rt1[i]==0)||(tl!=get_reg(i_regs->regmap,rt1[i])); // ignore loads to r0 and unneeded reg | |
2564 | if (opcode[i]==0x20) { // LB | |
2565 | if(!c||memtarget) { | |
2566 | if(!dummy) { | |
57871462 | 2567 | #ifdef HOST_IMM_ADDR32 |
2568 | if(c) | |
2569 | emit_movsbl_tlb((constmap[i][s]+offset)^3,map,tl); | |
2570 | else | |
2571 | #endif | |
2572 | { | |
2573 | //emit_xorimm(addr,3,tl); | |
57871462 | 2574 | //emit_movsbl_indexed((int)rdram-0x80000000,tl,tl); |
535d208a | 2575 | int x=0,a=tl; |
2002a1db | 2576 | #ifdef BIG_ENDIAN_MIPS |
57871462 | 2577 | if(!c) emit_xorimm(addr,3,tl); |
2578 | else x=((constmap[i][s]+offset)^3)-(constmap[i][s]+offset); | |
2002a1db | 2579 | #else |
535d208a | 2580 | if(!c) a=addr; |
dadf55f2 | 2581 | #endif |
b1570849 | 2582 | if(fastload_reg_override) a=fastload_reg_override; |
2583 | ||
535d208a | 2584 | emit_movsbl_indexed_tlb(x,a,map,tl); |
57871462 | 2585 | } |
57871462 | 2586 | } |
535d208a | 2587 | if(jaddr) |
2588 | add_stub(LOADB_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist); | |
57871462 | 2589 | } |
535d208a | 2590 | else |
2591 | inline_readstub(LOADB_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist); | |
2592 | } | |
2593 | if (opcode[i]==0x21) { // LH | |
2594 | if(!c||memtarget) { | |
2595 | if(!dummy) { | |
57871462 | 2596 | #ifdef HOST_IMM_ADDR32 |
2597 | if(c) | |
2598 | emit_movswl_tlb((constmap[i][s]+offset)^2,map,tl); | |
2599 | else | |
2600 | #endif | |
2601 | { | |
535d208a | 2602 | int x=0,a=tl; |
2002a1db | 2603 | #ifdef BIG_ENDIAN_MIPS |
57871462 | 2604 | if(!c) emit_xorimm(addr,2,tl); |
2605 | else x=((constmap[i][s]+offset)^2)-(constmap[i][s]+offset); | |
2002a1db | 2606 | #else |
535d208a | 2607 | if(!c) a=addr; |
dadf55f2 | 2608 | #endif |
b1570849 | 2609 | if(fastload_reg_override) a=fastload_reg_override; |
57871462 | 2610 | //#ifdef |
2611 | //emit_movswl_indexed_tlb(x,tl,map,tl); | |
2612 | //else | |
2613 | if(map>=0) { | |
535d208a | 2614 | emit_movswl_indexed(x,a,tl); |
2615 | }else{ | |
a327ad27 | 2616 | #if 1 //def RAM_OFFSET |
535d208a | 2617 | emit_movswl_indexed(x,a,tl); |
2618 | #else | |
2619 | emit_movswl_indexed((int)rdram-0x80000000+x,a,tl); | |
2620 | #endif | |
2621 | } | |
57871462 | 2622 | } |
57871462 | 2623 | } |
535d208a | 2624 | if(jaddr) |
2625 | add_stub(LOADH_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist); | |
57871462 | 2626 | } |
535d208a | 2627 | else |
2628 | inline_readstub(LOADH_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist); | |
2629 | } | |
2630 | if (opcode[i]==0x23) { // LW | |
2631 | if(!c||memtarget) { | |
2632 | if(!dummy) { | |
dadf55f2 | 2633 | int a=addr; |
b1570849 | 2634 | if(fastload_reg_override) a=fastload_reg_override; |
57871462 | 2635 | //emit_readword_indexed((int)rdram-0x80000000,addr,tl); |
2636 | #ifdef HOST_IMM_ADDR32 | |
2637 | if(c) | |
2638 | emit_readword_tlb(constmap[i][s]+offset,map,tl); | |
2639 | else | |
2640 | #endif | |
dadf55f2 | 2641 | emit_readword_indexed_tlb(0,a,map,tl); |
57871462 | 2642 | } |
535d208a | 2643 | if(jaddr) |
2644 | add_stub(LOADW_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist); | |
57871462 | 2645 | } |
535d208a | 2646 | else |
2647 | inline_readstub(LOADW_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist); | |
2648 | } | |
2649 | if (opcode[i]==0x24) { // LBU | |
2650 | if(!c||memtarget) { | |
2651 | if(!dummy) { | |
57871462 | 2652 | #ifdef HOST_IMM_ADDR32 |
2653 | if(c) | |
2654 | emit_movzbl_tlb((constmap[i][s]+offset)^3,map,tl); | |
2655 | else | |
2656 | #endif | |
2657 | { | |
2658 | //emit_xorimm(addr,3,tl); | |
57871462 | 2659 | //emit_movzbl_indexed((int)rdram-0x80000000,tl,tl); |
535d208a | 2660 | int x=0,a=tl; |
2002a1db | 2661 | #ifdef BIG_ENDIAN_MIPS |
57871462 | 2662 | if(!c) emit_xorimm(addr,3,tl); |
2663 | else x=((constmap[i][s]+offset)^3)-(constmap[i][s]+offset); | |
2002a1db | 2664 | #else |
535d208a | 2665 | if(!c) a=addr; |
dadf55f2 | 2666 | #endif |
b1570849 | 2667 | if(fastload_reg_override) a=fastload_reg_override; |
2668 | ||
535d208a | 2669 | emit_movzbl_indexed_tlb(x,a,map,tl); |
57871462 | 2670 | } |
57871462 | 2671 | } |
535d208a | 2672 | if(jaddr) |
2673 | add_stub(LOADBU_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist); | |
57871462 | 2674 | } |
535d208a | 2675 | else |
2676 | inline_readstub(LOADBU_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist); | |
2677 | } | |
2678 | if (opcode[i]==0x25) { // LHU | |
2679 | if(!c||memtarget) { | |
2680 | if(!dummy) { | |
57871462 | 2681 | #ifdef HOST_IMM_ADDR32 |
2682 | if(c) | |
2683 | emit_movzwl_tlb((constmap[i][s]+offset)^2,map,tl); | |
2684 | else | |
2685 | #endif | |
2686 | { | |
535d208a | 2687 | int x=0,a=tl; |
2002a1db | 2688 | #ifdef BIG_ENDIAN_MIPS |
57871462 | 2689 | if(!c) emit_xorimm(addr,2,tl); |
2690 | else x=((constmap[i][s]+offset)^2)-(constmap[i][s]+offset); | |
2002a1db | 2691 | #else |
535d208a | 2692 | if(!c) a=addr; |
dadf55f2 | 2693 | #endif |
b1570849 | 2694 | if(fastload_reg_override) a=fastload_reg_override; |
57871462 | 2695 | //#ifdef |
2696 | //emit_movzwl_indexed_tlb(x,tl,map,tl); | |
2697 | //#else | |
2698 | if(map>=0) { | |
535d208a | 2699 | emit_movzwl_indexed(x,a,tl); |
2700 | }else{ | |
a327ad27 | 2701 | #if 1 //def RAM_OFFSET |
535d208a | 2702 | emit_movzwl_indexed(x,a,tl); |
2703 | #else | |
2704 | emit_movzwl_indexed((int)rdram-0x80000000+x,a,tl); | |
2705 | #endif | |
2706 | } | |
57871462 | 2707 | } |
2708 | } | |
535d208a | 2709 | if(jaddr) |
2710 | add_stub(LOADHU_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist); | |
57871462 | 2711 | } |
535d208a | 2712 | else |
2713 | inline_readstub(LOADHU_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist); | |
2714 | } | |
2715 | if (opcode[i]==0x27) { // LWU | |
2716 | assert(th>=0); | |
2717 | if(!c||memtarget) { | |
2718 | if(!dummy) { | |
dadf55f2 | 2719 | int a=addr; |
b1570849 | 2720 | if(fastload_reg_override) a=fastload_reg_override; |
57871462 | 2721 | //emit_readword_indexed((int)rdram-0x80000000,addr,tl); |
2722 | #ifdef HOST_IMM_ADDR32 | |
2723 | if(c) | |
2724 | emit_readword_tlb(constmap[i][s]+offset,map,tl); | |
2725 | else | |
2726 | #endif | |
dadf55f2 | 2727 | emit_readword_indexed_tlb(0,a,map,tl); |
57871462 | 2728 | } |
535d208a | 2729 | if(jaddr) |
2730 | add_stub(LOADW_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist); | |
2731 | } | |
2732 | else { | |
2733 | inline_readstub(LOADW_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist); | |
57871462 | 2734 | } |
535d208a | 2735 | emit_zeroreg(th); |
2736 | } | |
2737 | if (opcode[i]==0x37) { // LD | |
2738 | if(!c||memtarget) { | |
2739 | if(!dummy) { | |
dadf55f2 | 2740 | int a=addr; |
b1570849 | 2741 | if(fastload_reg_override) a=fastload_reg_override; |
57871462 | 2742 | //if(th>=0) emit_readword_indexed((int)rdram-0x80000000,addr,th); |
2743 | //emit_readword_indexed((int)rdram-0x7FFFFFFC,addr,tl); | |
2744 | #ifdef HOST_IMM_ADDR32 | |
2745 | if(c) | |
2746 | emit_readdword_tlb(constmap[i][s]+offset,map,th,tl); | |
2747 | else | |
2748 | #endif | |
dadf55f2 | 2749 | emit_readdword_indexed_tlb(0,a,map,th,tl); |
57871462 | 2750 | } |
535d208a | 2751 | if(jaddr) |
2752 | add_stub(LOADD_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist); | |
57871462 | 2753 | } |
535d208a | 2754 | else |
2755 | inline_readstub(LOADD_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist); | |
57871462 | 2756 | } |
535d208a | 2757 | } |
2758 | //emit_storereg(rt1[i],tl); // DEBUG | |
57871462 | 2759 | //if(opcode[i]==0x23) |
2760 | //if(opcode[i]==0x24) | |
2761 | //if(opcode[i]==0x23||opcode[i]==0x24) | |
2762 | /*if(opcode[i]==0x21||opcode[i]==0x23||opcode[i]==0x24) | |
2763 | { | |
2764 | //emit_pusha(); | |
2765 | save_regs(0x100f); | |
2766 | emit_readword((int)&last_count,ECX); | |
2767 | #ifdef __i386__ | |
2768 | if(get_reg(i_regs->regmap,CCREG)<0) | |
2769 | emit_loadreg(CCREG,HOST_CCREG); | |
2770 | emit_add(HOST_CCREG,ECX,HOST_CCREG); | |
2771 | emit_addimm(HOST_CCREG,2*ccadj[i],HOST_CCREG); | |
2772 | emit_writeword(HOST_CCREG,(int)&Count); | |
2773 | #endif | |
2774 | #ifdef __arm__ | |
2775 | if(get_reg(i_regs->regmap,CCREG)<0) | |
2776 | emit_loadreg(CCREG,0); | |
2777 | else | |
2778 | emit_mov(HOST_CCREG,0); | |
2779 | emit_add(0,ECX,0); | |
2780 | emit_addimm(0,2*ccadj[i],0); | |
2781 | emit_writeword(0,(int)&Count); | |
2782 | #endif | |
2783 | emit_call((int)memdebug); | |
2784 | //emit_popa(); | |
2785 | restore_regs(0x100f); | |
2786 | }/**/ | |
2787 | } | |
2788 | ||
2789 | #ifndef loadlr_assemble | |
2790 | void loadlr_assemble(int i,struct regstat *i_regs) | |
2791 | { | |
2792 | printf("Need loadlr_assemble for this architecture.\n"); | |
2793 | exit(1); | |
2794 | } | |
2795 | #endif | |
2796 | ||
2797 | void store_assemble(int i,struct regstat *i_regs) | |
2798 | { | |
2799 | int s,th,tl,map=-1; | |
2800 | int addr,temp; | |
2801 | int offset; | |
2802 | int jaddr=0,jaddr2,type; | |
666a299d | 2803 | int memtarget=0,c=0; |
57871462 | 2804 | int agr=AGEN1+(i&1); |
b1570849 | 2805 | int faststore_reg_override=0; |
57871462 | 2806 | u_int hr,reglist=0; |
2807 | th=get_reg(i_regs->regmap,rs2[i]|64); | |
2808 | tl=get_reg(i_regs->regmap,rs2[i]); | |
2809 | s=get_reg(i_regs->regmap,rs1[i]); | |
2810 | temp=get_reg(i_regs->regmap,agr); | |
2811 | if(temp<0) temp=get_reg(i_regs->regmap,-1); | |
2812 | offset=imm[i]; | |
2813 | if(s>=0) { | |
2814 | c=(i_regs->wasconst>>s)&1; | |
af4ee1fe | 2815 | if(c) { |
2816 | memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE; | |
af4ee1fe | 2817 | } |
57871462 | 2818 | } |
2819 | assert(tl>=0); | |
2820 | assert(temp>=0); | |
2821 | for(hr=0;hr<HOST_REGS;hr++) { | |
2822 | if(i_regs->regmap[hr]>=0) reglist|=1<<hr; | |
2823 | } | |
2824 | if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<<HOST_CCREG); | |
2825 | if(offset||s<0||c) addr=temp; | |
2826 | else addr=s; | |
1edfcc68 | 2827 | if(!c) { |
2828 | jaddr=emit_fastpath_cmp_jump(i,addr,&faststore_reg_override); | |
2829 | } | |
2830 | else if(ram_offset&&memtarget) { | |
2831 | emit_addimm(addr,ram_offset,HOST_TEMPREG); | |
2832 | faststore_reg_override=HOST_TEMPREG; | |
57871462 | 2833 | } |
2834 | ||
2835 | if (opcode[i]==0x28) { // SB | |
2836 | if(!c||memtarget) { | |
97a238a6 | 2837 | int x=0,a=temp; |
2002a1db | 2838 | #ifdef BIG_ENDIAN_MIPS |
57871462 | 2839 | if(!c) emit_xorimm(addr,3,temp); |
2840 | else x=((constmap[i][s]+offset)^3)-(constmap[i][s]+offset); | |
2002a1db | 2841 | #else |
97a238a6 | 2842 | if(!c) a=addr; |
dadf55f2 | 2843 | #endif |
b1570849 | 2844 | if(faststore_reg_override) a=faststore_reg_override; |
57871462 | 2845 | //emit_writebyte_indexed(tl,(int)rdram-0x80000000,temp); |
97a238a6 | 2846 | emit_writebyte_indexed_tlb(tl,x,a,map,a); |
57871462 | 2847 | } |
2848 | type=STOREB_STUB; | |
2849 | } | |
2850 | if (opcode[i]==0x29) { // SH | |
2851 | if(!c||memtarget) { | |
97a238a6 | 2852 | int x=0,a=temp; |
2002a1db | 2853 | #ifdef BIG_ENDIAN_MIPS |
57871462 | 2854 | if(!c) emit_xorimm(addr,2,temp); |
2855 | else x=((constmap[i][s]+offset)^2)-(constmap[i][s]+offset); | |
2002a1db | 2856 | #else |
97a238a6 | 2857 | if(!c) a=addr; |
dadf55f2 | 2858 | #endif |
b1570849 | 2859 | if(faststore_reg_override) a=faststore_reg_override; |
57871462 | 2860 | //#ifdef |
2861 | //emit_writehword_indexed_tlb(tl,x,temp,map,temp); | |
2862 | //#else | |
2863 | if(map>=0) { | |
97a238a6 | 2864 | emit_writehword_indexed(tl,x,a); |
57871462 | 2865 | }else |
a327ad27 | 2866 | //emit_writehword_indexed(tl,(int)rdram-0x80000000+x,a); |
2867 | emit_writehword_indexed(tl,x,a); | |
57871462 | 2868 | } |
2869 | type=STOREH_STUB; | |
2870 | } | |
2871 | if (opcode[i]==0x2B) { // SW | |
dadf55f2 | 2872 | if(!c||memtarget) { |
2873 | int a=addr; | |
b1570849 | 2874 | if(faststore_reg_override) a=faststore_reg_override; |
57871462 | 2875 | //emit_writeword_indexed(tl,(int)rdram-0x80000000,addr); |
dadf55f2 | 2876 | emit_writeword_indexed_tlb(tl,0,a,map,temp); |
2877 | } | |
57871462 | 2878 | type=STOREW_STUB; |
2879 | } | |
2880 | if (opcode[i]==0x3F) { // SD | |
2881 | if(!c||memtarget) { | |
dadf55f2 | 2882 | int a=addr; |
b1570849 | 2883 | if(faststore_reg_override) a=faststore_reg_override; |
57871462 | 2884 | if(rs2[i]) { |
2885 | assert(th>=0); | |
2886 | //emit_writeword_indexed(th,(int)rdram-0x80000000,addr); | |
2887 | //emit_writeword_indexed(tl,(int)rdram-0x7FFFFFFC,addr); | |
dadf55f2 | 2888 | emit_writedword_indexed_tlb(th,tl,0,a,map,temp); |
57871462 | 2889 | }else{ |
2890 | // Store zero | |
2891 | //emit_writeword_indexed(tl,(int)rdram-0x80000000,temp); | |
2892 | //emit_writeword_indexed(tl,(int)rdram-0x7FFFFFFC,temp); | |
dadf55f2 | 2893 | emit_writedword_indexed_tlb(tl,tl,0,a,map,temp); |
57871462 | 2894 | } |
2895 | } | |
2896 | type=STORED_STUB; | |
2897 | } | |
b96d3df7 | 2898 | if(jaddr) { |
2899 | // PCSX store handlers don't check invcode again | |
2900 | reglist|=1<<addr; | |
2901 | add_stub(type,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist); | |
2902 | jaddr=0; | |
2903 | } | |
1edfcc68 | 2904 | if(!(i_regs->waswritten&(1<<rs1[i]))&&!(new_dynarec_hacks&NDHACK_NO_SMC_CHECK)) { |
57871462 | 2905 | if(!c||memtarget) { |
2906 | #ifdef DESTRUCTIVE_SHIFT | |
2907 | // The x86 shift operation is 'destructive'; it overwrites the | |
2908 | // source register, so we need to make a copy first and use that. | |
2909 | addr=temp; | |
2910 | #endif | |
2911 | #if defined(HOST_IMM8) | |
2912 | int ir=get_reg(i_regs->regmap,INVCP); | |
2913 | assert(ir>=0); | |
2914 | emit_cmpmem_indexedsr12_reg(ir,addr,1); | |
2915 | #else | |
2916 | emit_cmpmem_indexedsr12_imm((int)invalid_code,addr,1); | |
2917 | #endif | |
0bbd1454 | 2918 | #if defined(HAVE_CONDITIONAL_CALL) && !defined(DESTRUCTIVE_SHIFT) |
2919 | emit_callne(invalidate_addr_reg[addr]); | |
2920 | #else | |
57871462 | 2921 | jaddr2=(int)out; |
2922 | emit_jne(0); | |
2923 | add_stub(INVCODE_STUB,jaddr2,(int)out,reglist|(1<<HOST_CCREG),addr,0,0,0); | |
0bbd1454 | 2924 | #endif |
57871462 | 2925 | } |
2926 | } | |
7a518516 | 2927 | u_int addr_val=constmap[i][s]+offset; |
3eaa7048 | 2928 | if(jaddr) { |
2929 | add_stub(type,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist); | |
2930 | } else if(c&&!memtarget) { | |
7a518516 | 2931 | inline_writestub(type,i,addr_val,i_regs->regmap,rs2[i],ccadj[i],reglist); |
2932 | } | |
2933 | // basic current block modification detection.. | |
2934 | // not looking back as that should be in mips cache already | |
2935 | if(c&&start+i*4<addr_val&&addr_val<start+slen*4) { | |
c43b5311 | 2936 | SysPrintf("write to %08x hits block %08x, pc=%08x\n",addr_val,start,start+i*4); |
7a518516 | 2937 | assert(i_regs->regmap==regs[i].regmap); // not delay slot |
2938 | if(i_regs->regmap==regs[i].regmap) { | |
2939 | load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty,i); | |
2940 | wb_dirtys(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty); | |
2941 | emit_movimm(start+i*4+4,0); | |
2942 | emit_writeword(0,(int)&pcaddr); | |
2943 | emit_jmp((int)do_interrupt); | |
2944 | } | |
3eaa7048 | 2945 | } |
57871462 | 2946 | //if(opcode[i]==0x2B || opcode[i]==0x3F) |
2947 | //if(opcode[i]==0x2B || opcode[i]==0x28) | |
2948 | //if(opcode[i]==0x2B || opcode[i]==0x29) | |
2949 | //if(opcode[i]==0x2B) | |
2950 | /*if(opcode[i]==0x2B || opcode[i]==0x28 || opcode[i]==0x29 || opcode[i]==0x3F) | |
2951 | { | |
28d74ee8 | 2952 | #ifdef __i386__ |
2953 | emit_pusha(); | |
2954 | #endif | |
2955 | #ifdef __arm__ | |
57871462 | 2956 | save_regs(0x100f); |
28d74ee8 | 2957 | #endif |
57871462 | 2958 | emit_readword((int)&last_count,ECX); |
2959 | #ifdef __i386__ | |
2960 | if(get_reg(i_regs->regmap,CCREG)<0) | |
2961 | emit_loadreg(CCREG,HOST_CCREG); | |
2962 | emit_add(HOST_CCREG,ECX,HOST_CCREG); | |
2963 | emit_addimm(HOST_CCREG,2*ccadj[i],HOST_CCREG); | |
2964 | emit_writeword(HOST_CCREG,(int)&Count); | |
2965 | #endif | |
2966 | #ifdef __arm__ | |
2967 | if(get_reg(i_regs->regmap,CCREG)<0) | |
2968 | emit_loadreg(CCREG,0); | |
2969 | else | |
2970 | emit_mov(HOST_CCREG,0); | |
2971 | emit_add(0,ECX,0); | |
2972 | emit_addimm(0,2*ccadj[i],0); | |
2973 | emit_writeword(0,(int)&Count); | |
2974 | #endif | |
2975 | emit_call((int)memdebug); | |
28d74ee8 | 2976 | #ifdef __i386__ |
2977 | emit_popa(); | |
2978 | #endif | |
2979 | #ifdef __arm__ | |
57871462 | 2980 | restore_regs(0x100f); |
28d74ee8 | 2981 | #endif |
57871462 | 2982 | }/**/ |
2983 | } | |
2984 | ||
2985 | void storelr_assemble(int i,struct regstat *i_regs) | |
2986 | { | |
2987 | int s,th,tl; | |
2988 | int temp; | |
2989 | int temp2; | |
2990 | int offset; | |
2991 | int jaddr=0,jaddr2; | |
2992 | int case1,case2,case3; | |
2993 | int done0,done1,done2; | |
af4ee1fe | 2994 | int memtarget=0,c=0; |
fab5d06d | 2995 | int agr=AGEN1+(i&1); |
57871462 | 2996 | u_int hr,reglist=0; |
2997 | th=get_reg(i_regs->regmap,rs2[i]|64); | |
2998 | tl=get_reg(i_regs->regmap,rs2[i]); | |
2999 | s=get_reg(i_regs->regmap,rs1[i]); | |
fab5d06d | 3000 | temp=get_reg(i_regs->regmap,agr); |
3001 | if(temp<0) temp=get_reg(i_regs->regmap,-1); | |
57871462 | 3002 | offset=imm[i]; |
3003 | if(s>=0) { | |
3004 | c=(i_regs->isconst>>s)&1; | |
af4ee1fe | 3005 | if(c) { |
3006 | memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE; | |
af4ee1fe | 3007 | } |
57871462 | 3008 | } |
3009 | assert(tl>=0); | |
3010 | for(hr=0;hr<HOST_REGS;hr++) { | |
3011 | if(i_regs->regmap[hr]>=0) reglist|=1<<hr; | |
3012 | } | |
535d208a | 3013 | assert(temp>=0); |
1edfcc68 | 3014 | if(!c) { |
3015 | emit_cmpimm(s<0||offset?temp:s,RAM_SIZE); | |
3016 | if(!offset&&s!=temp) emit_mov(s,temp); | |
3017 | jaddr=(int)out; | |
3018 | emit_jno(0); | |
3019 | } | |
3020 | else | |
3021 | { | |
3022 | if(!memtarget||!rs1[i]) { | |
535d208a | 3023 | jaddr=(int)out; |
3024 | emit_jmp(0); | |
57871462 | 3025 | } |
535d208a | 3026 | } |
1edfcc68 | 3027 | #ifdef RAM_OFFSET |
3028 | int map=get_reg(i_regs->regmap,ROREG); | |
3029 | if(map<0) emit_loadreg(ROREG,map=HOST_TEMPREG); | |
3030 | #else | |
9f51b4b9 | 3031 | if((u_int)rdram!=0x80000000) |
1edfcc68 | 3032 | emit_addimm_no_flags((u_int)rdram-(u_int)0x80000000,temp); |
3033 | #endif | |
535d208a | 3034 | |
3035 | if (opcode[i]==0x2C||opcode[i]==0x2D) { // SDL/SDR | |
3036 | temp2=get_reg(i_regs->regmap,FTEMP); | |
3037 | if(!rs2[i]) temp2=th=tl; | |
3038 | } | |
57871462 | 3039 | |
2002a1db | 3040 | #ifndef BIG_ENDIAN_MIPS |
3041 | emit_xorimm(temp,3,temp); | |
3042 | #endif | |
535d208a | 3043 | emit_testimm(temp,2); |
3044 | case2=(int)out; | |
3045 | emit_jne(0); | |
3046 | emit_testimm(temp,1); | |
3047 | case1=(int)out; | |
3048 | emit_jne(0); | |
3049 | // 0 | |
3050 | if (opcode[i]==0x2A) { // SWL | |
3051 | emit_writeword_indexed(tl,0,temp); | |
3052 | } | |
3053 | if (opcode[i]==0x2E) { // SWR | |
3054 | emit_writebyte_indexed(tl,3,temp); | |
3055 | } | |
3056 | if (opcode[i]==0x2C) { // SDL | |
3057 | emit_writeword_indexed(th,0,temp); | |
3058 | if(rs2[i]) emit_mov(tl,temp2); | |
3059 | } | |
3060 | if (opcode[i]==0x2D) { // SDR | |
3061 | emit_writebyte_indexed(tl,3,temp); | |
3062 | if(rs2[i]) emit_shldimm(th,tl,24,temp2); | |
3063 | } | |
3064 | done0=(int)out; | |
3065 | emit_jmp(0); | |
3066 | // 1 | |
3067 | set_jump_target(case1,(int)out); | |
3068 | if (opcode[i]==0x2A) { // SWL | |
3069 | // Write 3 msb into three least significant bytes | |
3070 | if(rs2[i]) emit_rorimm(tl,8,tl); | |
3071 | emit_writehword_indexed(tl,-1,temp); | |
3072 | if(rs2[i]) emit_rorimm(tl,16,tl); | |
3073 | emit_writebyte_indexed(tl,1,temp); | |
3074 | if(rs2[i]) emit_rorimm(tl,8,tl); | |
3075 | } | |
3076 | if (opcode[i]==0x2E) { // SWR | |
3077 | // Write two lsb into two most significant bytes | |
3078 | emit_writehword_indexed(tl,1,temp); | |
3079 | } | |
3080 | if (opcode[i]==0x2C) { // SDL | |
3081 | if(rs2[i]) emit_shrdimm(tl,th,8,temp2); | |
3082 | // Write 3 msb into three least significant bytes | |
3083 | if(rs2[i]) emit_rorimm(th,8,th); | |
3084 | emit_writehword_indexed(th,-1,temp); | |
3085 | if(rs2[i]) emit_rorimm(th,16,th); | |
3086 | emit_writebyte_indexed(th,1,temp); | |
3087 | if(rs2[i]) emit_rorimm(th,8,th); | |
3088 | } | |
3089 | if (opcode[i]==0x2D) { // SDR | |
3090 | if(rs2[i]) emit_shldimm(th,tl,16,temp2); | |
3091 | // Write two lsb into two most significant bytes | |
3092 | emit_writehword_indexed(tl,1,temp); | |
3093 | } | |
3094 | done1=(int)out; | |
3095 | emit_jmp(0); | |
3096 | // 2 | |
3097 | set_jump_target(case2,(int)out); | |
3098 | emit_testimm(temp,1); | |
3099 | case3=(int)out; | |
3100 | emit_jne(0); | |
3101 | if (opcode[i]==0x2A) { // SWL | |
3102 | // Write two msb into two least significant bytes | |
3103 | if(rs2[i]) emit_rorimm(tl,16,tl); | |
3104 | emit_writehword_indexed(tl,-2,temp); | |
3105 | if(rs2[i]) emit_rorimm(tl,16,tl); | |
3106 | } | |
3107 | if (opcode[i]==0x2E) { // SWR | |
3108 | // Write 3 lsb into three most significant bytes | |
3109 | emit_writebyte_indexed(tl,-1,temp); | |
3110 | if(rs2[i]) emit_rorimm(tl,8,tl); | |
3111 | emit_writehword_indexed(tl,0,temp); | |
3112 | if(rs2[i]) emit_rorimm(tl,24,tl); | |
3113 | } | |
3114 | if (opcode[i]==0x2C) { // SDL | |
3115 | if(rs2[i]) emit_shrdimm(tl,th,16,temp2); | |
3116 | // Write two msb into two least significant bytes | |
3117 | if(rs2[i]) emit_rorimm(th,16,th); | |
3118 | emit_writehword_indexed(th,-2,temp); | |
3119 | if(rs2[i]) emit_rorimm(th,16,th); | |
3120 | } | |
3121 | if (opcode[i]==0x2D) { // SDR | |
3122 | if(rs2[i]) emit_shldimm(th,tl,8,temp2); | |
3123 | // Write 3 lsb into three most significant bytes | |
3124 | emit_writebyte_indexed(tl,-1,temp); | |
3125 | if(rs2[i]) emit_rorimm(tl,8,tl); | |
3126 | emit_writehword_indexed(tl,0,temp); | |
3127 | if(rs2[i]) emit_rorimm(tl,24,tl); | |
3128 | } | |
3129 | done2=(int)out; | |
3130 | emit_jmp(0); | |
3131 | // 3 | |
3132 | set_jump_target(case3,(int)out); | |
3133 | if (opcode[i]==0x2A) { // SWL | |
3134 | // Write msb into least significant byte | |
3135 | if(rs2[i]) emit_rorimm(tl,24,tl); | |
3136 | emit_writebyte_indexed(tl,-3,temp); | |
3137 | if(rs2[i]) emit_rorimm(tl,8,tl); | |
3138 | } | |
3139 | if (opcode[i]==0x2E) { // SWR | |
3140 | // Write entire word | |
3141 | emit_writeword_indexed(tl,-3,temp); | |
3142 | } | |
3143 | if (opcode[i]==0x2C) { // SDL | |
3144 | if(rs2[i]) emit_shrdimm(tl,th,24,temp2); | |
3145 | // Write msb into least significant byte | |
3146 | if(rs2[i]) emit_rorimm(th,24,th); | |
3147 | emit_writebyte_indexed(th,-3,temp); | |
3148 | if(rs2[i]) emit_rorimm(th,8,th); | |
3149 | } | |
3150 | if (opcode[i]==0x2D) { // SDR | |
3151 | if(rs2[i]) emit_mov(th,temp2); | |
3152 | // Write entire word | |
3153 | emit_writeword_indexed(tl,-3,temp); | |
3154 | } | |
3155 | set_jump_target(done0,(int)out); | |
3156 | set_jump_target(done1,(int)out); | |
3157 | set_jump_target(done2,(int)out); | |
3158 | if (opcode[i]==0x2C) { // SDL | |
3159 | emit_testimm(temp,4); | |
57871462 | 3160 | done0=(int)out; |
57871462 | 3161 | emit_jne(0); |
535d208a | 3162 | emit_andimm(temp,~3,temp); |
3163 | emit_writeword_indexed(temp2,4,temp); | |
3164 | set_jump_target(done0,(int)out); | |
3165 | } | |
3166 | if (opcode[i]==0x2D) { // SDR | |
3167 | emit_testimm(temp,4); | |
3168 | done0=(int)out; | |
3169 | emit_jeq(0); | |
3170 | emit_andimm(temp,~3,temp); | |
3171 | emit_writeword_indexed(temp2,-4,temp); | |
57871462 | 3172 | set_jump_target(done0,(int)out); |
57871462 | 3173 | } |
535d208a | 3174 | if(!c||!memtarget) |
3175 | add_stub(STORELR_STUB,jaddr,(int)out,i,(int)i_regs,temp,ccadj[i],reglist); | |
1edfcc68 | 3176 | if(!(i_regs->waswritten&(1<<rs1[i]))&&!(new_dynarec_hacks&NDHACK_NO_SMC_CHECK)) { |
535d208a | 3177 | #ifdef RAM_OFFSET |
3178 | int map=get_reg(i_regs->regmap,ROREG); | |
3179 | if(map<0) map=HOST_TEMPREG; | |
3180 | gen_orig_addr_w(temp,map); | |
3181 | #else | |
57871462 | 3182 | emit_addimm_no_flags((u_int)0x80000000-(u_int)rdram,temp); |
535d208a | 3183 | #endif |
57871462 | 3184 | #if defined(HOST_IMM8) |
3185 | int ir=get_reg(i_regs->regmap,INVCP); | |
3186 | assert(ir>=0); | |
3187 | emit_cmpmem_indexedsr12_reg(ir,temp,1); | |
3188 | #else | |
3189 | emit_cmpmem_indexedsr12_imm((int)invalid_code,temp,1); | |
3190 | #endif | |
535d208a | 3191 | #if defined(HAVE_CONDITIONAL_CALL) && !defined(DESTRUCTIVE_SHIFT) |
3192 | emit_callne(invalidate_addr_reg[temp]); | |
3193 | #else | |
57871462 | 3194 | jaddr2=(int)out; |
3195 | emit_jne(0); | |
3196 | add_stub(INVCODE_STUB,jaddr2,(int)out,reglist|(1<<HOST_CCREG),temp,0,0,0); | |
535d208a | 3197 | #endif |
57871462 | 3198 | } |
3199 | /* | |
3200 | emit_pusha(); | |
3201 | //save_regs(0x100f); | |
3202 | emit_readword((int)&last_count,ECX); | |
3203 | if(get_reg(i_regs->regmap,CCREG)<0) | |
3204 | emit_loadreg(CCREG,HOST_CCREG); | |
3205 | emit_add(HOST_CCREG,ECX,HOST_CCREG); | |
3206 | emit_addimm(HOST_CCREG,2*ccadj[i],HOST_CCREG); | |
3207 | emit_writeword(HOST_CCREG,(int)&Count); | |
3208 | emit_call((int)memdebug); | |
3209 | emit_popa(); | |
3210 | //restore_regs(0x100f); | |
3211 | /**/ | |
3212 | } | |
3213 | ||
3214 | void c1ls_assemble(int i,struct regstat *i_regs) | |
3215 | { | |
3d624f89 | 3216 | cop1_unusable(i, i_regs); |
57871462 | 3217 | } |
3218 | ||
b9b61529 | 3219 | void c2ls_assemble(int i,struct regstat *i_regs) |
3220 | { | |
3221 | int s,tl; | |
3222 | int ar; | |
3223 | int offset; | |
1fd1aceb | 3224 | int memtarget=0,c=0; |
c2e3bd42 | 3225 | int jaddr2=0,jaddr3,type; |
b9b61529 | 3226 | int agr=AGEN1+(i&1); |
ffb0b9e0 | 3227 | int fastio_reg_override=0; |
b9b61529 | 3228 | u_int hr,reglist=0; |
3229 | u_int copr=(source[i]>>16)&0x1f; | |
3230 | s=get_reg(i_regs->regmap,rs1[i]); | |
3231 | tl=get_reg(i_regs->regmap,FTEMP); | |
3232 | offset=imm[i]; | |
3233 | assert(rs1[i]>0); | |
3234 | assert(tl>=0); | |
b9b61529 | 3235 | |
3236 | for(hr=0;hr<HOST_REGS;hr++) { | |
3237 | if(i_regs->regmap[hr]>=0) reglist|=1<<hr; | |
3238 | } | |
3239 | if(i_regs->regmap[HOST_CCREG]==CCREG) | |
3240 | reglist&=~(1<<HOST_CCREG); | |
3241 | ||