57871462 |
1 | /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * |
c6c3b1b3 |
2 | * Mupen64plus/PCSX - assem_arm.c * |
20d507ba |
3 | * Copyright (C) 2009-2011 Ari64 * |
c6c3b1b3 |
4 | * Copyright (C) 2010-2011 GraÅžvydas "notaz" Ignotas * |
57871462 |
5 | * * |
6 | * This program is free software; you can redistribute it and/or modify * |
7 | * it under the terms of the GNU General Public License as published by * |
8 | * the Free Software Foundation; either version 2 of the License, or * |
9 | * (at your option) any later version. * |
10 | * * |
11 | * This program is distributed in the hope that it will be useful, * |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of * |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * |
14 | * GNU General Public License for more details. * |
15 | * * |
16 | * You should have received a copy of the GNU General Public License * |
17 | * along with this program; if not, write to the * |
18 | * Free Software Foundation, Inc., * |
19 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. * |
20 | * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ |
21 | |
6c0eefaf |
22 | #include "../gte.h" |
23 | #define FLAGLESS |
24 | #include "../gte.h" |
25 | #undef FLAGLESS |
054175e9 |
26 | #include "../gte_arm.h" |
27 | #include "../gte_neon.h" |
28 | #include "pcnt.h" |
665f33e1 |
29 | #include "arm_features.h" |
054175e9 |
30 | |
a327ad27 |
31 | #if !BASE_ADDR_FIXED |
bdeade46 |
32 | char translation_cache[1 << TARGET_SIZE_2] __attribute__((aligned(4096))); |
33 | #endif |
34 | |
4d646738 |
35 | #ifndef __MACH__ |
36 | #define CALLER_SAVE_REGS 0x100f |
37 | #else |
38 | #define CALLER_SAVE_REGS 0x120f |
39 | #endif |
40 | |
e2b5e7aa |
41 | #define unused __attribute__((unused)) |
42 | |
57871462 |
43 | extern int cycle_count; |
44 | extern int last_count; |
45 | extern int pcaddr; |
46 | extern int pending_exception; |
47 | extern int branch_target; |
48 | extern uint64_t readmem_dword; |
57871462 |
49 | extern void *dynarec_local; |
57871462 |
50 | extern u_int mini_ht[32][2]; |
57871462 |
51 | |
52 | void indirect_jump_indexed(); |
53 | void indirect_jump(); |
54 | void do_interrupt(); |
55 | void jump_vaddr_r0(); |
56 | void jump_vaddr_r1(); |
57 | void jump_vaddr_r2(); |
58 | void jump_vaddr_r3(); |
59 | void jump_vaddr_r4(); |
60 | void jump_vaddr_r5(); |
61 | void jump_vaddr_r6(); |
62 | void jump_vaddr_r7(); |
63 | void jump_vaddr_r8(); |
64 | void jump_vaddr_r9(); |
65 | void jump_vaddr_r10(); |
66 | void jump_vaddr_r12(); |
67 | |
68 | const u_int jump_vaddr_reg[16] = { |
69 | (int)jump_vaddr_r0, |
70 | (int)jump_vaddr_r1, |
71 | (int)jump_vaddr_r2, |
72 | (int)jump_vaddr_r3, |
73 | (int)jump_vaddr_r4, |
74 | (int)jump_vaddr_r5, |
75 | (int)jump_vaddr_r6, |
76 | (int)jump_vaddr_r7, |
77 | (int)jump_vaddr_r8, |
78 | (int)jump_vaddr_r9, |
79 | (int)jump_vaddr_r10, |
80 | 0, |
81 | (int)jump_vaddr_r12, |
82 | 0, |
83 | 0, |
84 | 0}; |
85 | |
0bbd1454 |
86 | void invalidate_addr_r0(); |
87 | void invalidate_addr_r1(); |
88 | void invalidate_addr_r2(); |
89 | void invalidate_addr_r3(); |
90 | void invalidate_addr_r4(); |
91 | void invalidate_addr_r5(); |
92 | void invalidate_addr_r6(); |
93 | void invalidate_addr_r7(); |
94 | void invalidate_addr_r8(); |
95 | void invalidate_addr_r9(); |
96 | void invalidate_addr_r10(); |
97 | void invalidate_addr_r12(); |
98 | |
99 | const u_int invalidate_addr_reg[16] = { |
100 | (int)invalidate_addr_r0, |
101 | (int)invalidate_addr_r1, |
102 | (int)invalidate_addr_r2, |
103 | (int)invalidate_addr_r3, |
104 | (int)invalidate_addr_r4, |
105 | (int)invalidate_addr_r5, |
106 | (int)invalidate_addr_r6, |
107 | (int)invalidate_addr_r7, |
108 | (int)invalidate_addr_r8, |
109 | (int)invalidate_addr_r9, |
110 | (int)invalidate_addr_r10, |
111 | 0, |
112 | (int)invalidate_addr_r12, |
113 | 0, |
114 | 0, |
115 | 0}; |
116 | |
e2b5e7aa |
117 | static unsigned int needs_clear_cache[1<<(TARGET_SIZE_2-17)]; |
dd3a91a1 |
118 | |
57871462 |
119 | /* Linker */ |
120 | |
e2b5e7aa |
121 | static void set_jump_target(int addr,u_int target) |
57871462 |
122 | { |
123 | u_char *ptr=(u_char *)addr; |
124 | u_int *ptr2=(u_int *)ptr; |
125 | if(ptr[3]==0xe2) { |
126 | assert((target-(u_int)ptr2-8)<1024); |
127 | assert((addr&3)==0); |
128 | assert((target&3)==0); |
129 | *ptr2=(*ptr2&0xFFFFF000)|((target-(u_int)ptr2-8)>>2)|0xF00; |
130 | //printf("target=%x addr=%x insn=%x\n",target,addr,*ptr2); |
131 | } |
132 | else if(ptr[3]==0x72) { |
133 | // generated by emit_jno_unlikely |
134 | if((target-(u_int)ptr2-8)<1024) { |
135 | assert((addr&3)==0); |
136 | assert((target&3)==0); |
137 | *ptr2=(*ptr2&0xFFFFF000)|((target-(u_int)ptr2-8)>>2)|0xF00; |
138 | } |
139 | else if((target-(u_int)ptr2-8)<4096&&!((target-(u_int)ptr2-8)&15)) { |
140 | assert((addr&3)==0); |
141 | assert((target&3)==0); |
142 | *ptr2=(*ptr2&0xFFFFF000)|((target-(u_int)ptr2-8)>>4)|0xE00; |
143 | } |
144 | else *ptr2=(0x7A000000)|(((target-(u_int)ptr2-8)<<6)>>8); |
145 | } |
146 | else { |
147 | assert((ptr[3]&0x0e)==0xa); |
148 | *ptr2=(*ptr2&0xFF000000)|(((target-(u_int)ptr2-8)<<6)>>8); |
149 | } |
150 | } |
151 | |
152 | // This optionally copies the instruction from the target of the branch into |
153 | // the space before the branch. Works, but the difference in speed is |
154 | // usually insignificant. |
e2b5e7aa |
155 | #if 0 |
156 | static void set_jump_target_fillslot(int addr,u_int target,int copy) |
57871462 |
157 | { |
158 | u_char *ptr=(u_char *)addr; |
159 | u_int *ptr2=(u_int *)ptr; |
160 | assert(!copy||ptr2[-1]==0xe28dd000); |
161 | if(ptr[3]==0xe2) { |
162 | assert(!copy); |
163 | assert((target-(u_int)ptr2-8)<4096); |
164 | *ptr2=(*ptr2&0xFFFFF000)|(target-(u_int)ptr2-8); |
165 | } |
166 | else { |
167 | assert((ptr[3]&0x0e)==0xa); |
168 | u_int target_insn=*(u_int *)target; |
169 | if((target_insn&0x0e100000)==0) { // ALU, no immediate, no flags |
170 | copy=0; |
171 | } |
172 | if((target_insn&0x0c100000)==0x04100000) { // Load |
173 | copy=0; |
174 | } |
175 | if(target_insn&0x08000000) { |
176 | copy=0; |
177 | } |
178 | if(copy) { |
179 | ptr2[-1]=target_insn; |
180 | target+=4; |
181 | } |
182 | *ptr2=(*ptr2&0xFF000000)|(((target-(u_int)ptr2-8)<<6)>>8); |
183 | } |
184 | } |
e2b5e7aa |
185 | #endif |
57871462 |
186 | |
187 | /* Literal pool */ |
e2b5e7aa |
188 | static void add_literal(int addr,int val) |
57871462 |
189 | { |
15776b68 |
190 | assert(literalcount<sizeof(literals)/sizeof(literals[0])); |
57871462 |
191 | literals[literalcount][0]=addr; |
192 | literals[literalcount][1]=val; |
9f51b4b9 |
193 | literalcount++; |
194 | } |
57871462 |
195 | |
e2b5e7aa |
196 | static void *kill_pointer(void *stub) |
57871462 |
197 | { |
198 | int *ptr=(int *)(stub+4); |
199 | assert((*ptr&0x0ff00000)==0x05900000); |
200 | u_int offset=*ptr&0xfff; |
201 | int **l_ptr=(void *)ptr+offset+8; |
202 | int *i_ptr=*l_ptr; |
203 | set_jump_target((int)i_ptr,(int)stub); |
f76eeef9 |
204 | return i_ptr; |
57871462 |
205 | } |
206 | |
f968d35d |
207 | // find where external branch is liked to using addr of it's stub: |
208 | // get address that insn one after stub loads (dyna_linker arg1), |
209 | // treat it as a pointer to branch insn, |
210 | // return addr where that branch jumps to |
e2b5e7aa |
211 | static int get_pointer(void *stub) |
57871462 |
212 | { |
213 | //printf("get_pointer(%x)\n",(int)stub); |
214 | int *ptr=(int *)(stub+4); |
f968d35d |
215 | assert((*ptr&0x0fff0000)==0x059f0000); |
57871462 |
216 | u_int offset=*ptr&0xfff; |
217 | int **l_ptr=(void *)ptr+offset+8; |
218 | int *i_ptr=*l_ptr; |
219 | assert((*i_ptr&0x0f000000)==0x0a000000); |
220 | return (int)i_ptr+((*i_ptr<<8)>>6)+8; |
221 | } |
222 | |
223 | // Find the "clean" entry point from a "dirty" entry point |
224 | // by skipping past the call to verify_code |
e2b5e7aa |
225 | static u_int get_clean_addr(int addr) |
57871462 |
226 | { |
227 | int *ptr=(int *)addr; |
665f33e1 |
228 | #ifndef HAVE_ARMV7 |
57871462 |
229 | ptr+=4; |
230 | #else |
231 | ptr+=6; |
232 | #endif |
233 | if((*ptr&0xFF000000)!=0xeb000000) ptr++; |
234 | assert((*ptr&0xFF000000)==0xeb000000); // bl instruction |
235 | ptr++; |
236 | if((*ptr&0xFF000000)==0xea000000) { |
237 | return (int)ptr+((*ptr<<8)>>6)+8; // follow jump |
238 | } |
239 | return (u_int)ptr; |
240 | } |
241 | |
e2b5e7aa |
242 | static int verify_dirty(u_int *ptr) |
57871462 |
243 | { |
665f33e1 |
244 | #ifndef HAVE_ARMV7 |
57871462 |
245 | // get from literal pool |
15776b68 |
246 | assert((*ptr&0xFFFF0000)==0xe59f0000); |
57871462 |
247 | u_int offset=*ptr&0xfff; |
248 | u_int *l_ptr=(void *)ptr+offset+8; |
249 | u_int source=l_ptr[0]; |
250 | u_int copy=l_ptr[1]; |
251 | u_int len=l_ptr[2]; |
252 | ptr+=4; |
253 | #else |
254 | // ARMv7 movw/movt |
255 | assert((*ptr&0xFFF00000)==0xe3000000); |
256 | u_int source=(ptr[0]&0xFFF)+((ptr[0]>>4)&0xF000)+((ptr[2]<<16)&0xFFF0000)+((ptr[2]<<12)&0xF0000000); |
257 | u_int copy=(ptr[1]&0xFFF)+((ptr[1]>>4)&0xF000)+((ptr[3]<<16)&0xFFF0000)+((ptr[3]<<12)&0xF0000000); |
258 | u_int len=(ptr[4]&0xFFF)+((ptr[4]>>4)&0xF000); |
259 | ptr+=6; |
260 | #endif |
261 | if((*ptr&0xFF000000)!=0xeb000000) ptr++; |
262 | assert((*ptr&0xFF000000)==0xeb000000); // bl instruction |
57871462 |
263 | //printf("verify_dirty: %x %x %x\n",source,copy,len); |
264 | return !memcmp((void *)source,(void *)copy,len); |
265 | } |
266 | |
267 | // This doesn't necessarily find all clean entry points, just |
268 | // guarantees that it's not dirty |
e2b5e7aa |
269 | static int isclean(int addr) |
57871462 |
270 | { |
665f33e1 |
271 | #ifndef HAVE_ARMV7 |
581335b0 |
272 | u_int *ptr=((u_int *)addr)+4; |
57871462 |
273 | #else |
581335b0 |
274 | u_int *ptr=((u_int *)addr)+6; |
57871462 |
275 | #endif |
276 | if((*ptr&0xFF000000)!=0xeb000000) ptr++; |
277 | if((*ptr&0xFF000000)!=0xeb000000) return 1; // bl instruction |
278 | if((int)ptr+((*ptr<<8)>>6)+8==(int)verify_code) return 0; |
279 | if((int)ptr+((*ptr<<8)>>6)+8==(int)verify_code_vm) return 0; |
280 | if((int)ptr+((*ptr<<8)>>6)+8==(int)verify_code_ds) return 0; |
281 | return 1; |
282 | } |
283 | |
4a35de07 |
284 | // get source that block at addr was compiled from (host pointers) |
e2b5e7aa |
285 | static void get_bounds(int addr,u_int *start,u_int *end) |
57871462 |
286 | { |
287 | u_int *ptr=(u_int *)addr; |
665f33e1 |
288 | #ifndef HAVE_ARMV7 |
57871462 |
289 | // get from literal pool |
15776b68 |
290 | assert((*ptr&0xFFFF0000)==0xe59f0000); |
57871462 |
291 | u_int offset=*ptr&0xfff; |
292 | u_int *l_ptr=(void *)ptr+offset+8; |
293 | u_int source=l_ptr[0]; |
294 | //u_int copy=l_ptr[1]; |
295 | u_int len=l_ptr[2]; |
296 | ptr+=4; |
297 | #else |
298 | // ARMv7 movw/movt |
299 | assert((*ptr&0xFFF00000)==0xe3000000); |
300 | u_int source=(ptr[0]&0xFFF)+((ptr[0]>>4)&0xF000)+((ptr[2]<<16)&0xFFF0000)+((ptr[2]<<12)&0xF0000000); |
301 | //u_int copy=(ptr[1]&0xFFF)+((ptr[1]>>4)&0xF000)+((ptr[3]<<16)&0xFFF0000)+((ptr[3]<<12)&0xF0000000); |
302 | u_int len=(ptr[4]&0xFFF)+((ptr[4]>>4)&0xF000); |
303 | ptr+=6; |
304 | #endif |
305 | if((*ptr&0xFF000000)!=0xeb000000) ptr++; |
306 | assert((*ptr&0xFF000000)==0xeb000000); // bl instruction |
57871462 |
307 | *start=source; |
308 | *end=source+len; |
309 | } |
310 | |
311 | /* Register allocation */ |
312 | |
313 | // Note: registers are allocated clean (unmodified state) |
314 | // if you intend to modify the register, you must call dirty_reg(). |
e2b5e7aa |
315 | static void alloc_reg(struct regstat *cur,int i,signed char reg) |
57871462 |
316 | { |
317 | int r,hr; |
318 | int preferred_reg = (reg&7); |
319 | if(reg==CCREG) preferred_reg=HOST_CCREG; |
320 | if(reg==PTEMP||reg==FTEMP) preferred_reg=12; |
9f51b4b9 |
321 | |
57871462 |
322 | // Don't allocate unused registers |
323 | if((cur->u>>reg)&1) return; |
9f51b4b9 |
324 | |
57871462 |
325 | // see if it's already allocated |
326 | for(hr=0;hr<HOST_REGS;hr++) |
327 | { |
328 | if(cur->regmap[hr]==reg) return; |
329 | } |
9f51b4b9 |
330 | |
57871462 |
331 | // Keep the same mapping if the register was already allocated in a loop |
332 | preferred_reg = loop_reg(i,reg,preferred_reg); |
9f51b4b9 |
333 | |
57871462 |
334 | // Try to allocate the preferred register |
335 | if(cur->regmap[preferred_reg]==-1) { |
336 | cur->regmap[preferred_reg]=reg; |
337 | cur->dirty&=~(1<<preferred_reg); |
338 | cur->isconst&=~(1<<preferred_reg); |
339 | return; |
340 | } |
341 | r=cur->regmap[preferred_reg]; |
342 | if(r<64&&((cur->u>>r)&1)) { |
343 | cur->regmap[preferred_reg]=reg; |
344 | cur->dirty&=~(1<<preferred_reg); |
345 | cur->isconst&=~(1<<preferred_reg); |
346 | return; |
347 | } |
348 | if(r>=64&&((cur->uu>>(r&63))&1)) { |
349 | cur->regmap[preferred_reg]=reg; |
350 | cur->dirty&=~(1<<preferred_reg); |
351 | cur->isconst&=~(1<<preferred_reg); |
352 | return; |
353 | } |
9f51b4b9 |
354 | |
57871462 |
355 | // Clear any unneeded registers |
356 | // We try to keep the mapping consistent, if possible, because it |
357 | // makes branches easier (especially loops). So we try to allocate |
358 | // first (see above) before removing old mappings. If this is not |
359 | // possible then go ahead and clear out the registers that are no |
360 | // longer needed. |
361 | for(hr=0;hr<HOST_REGS;hr++) |
362 | { |
363 | r=cur->regmap[hr]; |
364 | if(r>=0) { |
365 | if(r<64) { |
366 | if((cur->u>>r)&1) {cur->regmap[hr]=-1;break;} |
367 | } |
368 | else |
369 | { |
370 | if((cur->uu>>(r&63))&1) {cur->regmap[hr]=-1;break;} |
371 | } |
372 | } |
373 | } |
374 | // Try to allocate any available register, but prefer |
375 | // registers that have not been used recently. |
376 | if(i>0) { |
377 | for(hr=0;hr<HOST_REGS;hr++) { |
378 | if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) { |
379 | if(regs[i-1].regmap[hr]!=rs1[i-1]&®s[i-1].regmap[hr]!=rs2[i-1]&®s[i-1].regmap[hr]!=rt1[i-1]&®s[i-1].regmap[hr]!=rt2[i-1]) { |
380 | cur->regmap[hr]=reg; |
381 | cur->dirty&=~(1<<hr); |
382 | cur->isconst&=~(1<<hr); |
383 | return; |
384 | } |
385 | } |
386 | } |
387 | } |
388 | // Try to allocate any available register |
389 | for(hr=0;hr<HOST_REGS;hr++) { |
390 | if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) { |
391 | cur->regmap[hr]=reg; |
392 | cur->dirty&=~(1<<hr); |
393 | cur->isconst&=~(1<<hr); |
394 | return; |
395 | } |
396 | } |
9f51b4b9 |
397 | |
57871462 |
398 | // Ok, now we have to evict someone |
399 | // Pick a register we hopefully won't need soon |
400 | u_char hsn[MAXREG+1]; |
401 | memset(hsn,10,sizeof(hsn)); |
402 | int j; |
403 | lsn(hsn,i,&preferred_reg); |
404 | //printf("eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",cur->regmap[0],cur->regmap[1],cur->regmap[2],cur->regmap[3],cur->regmap[5],cur->regmap[6],cur->regmap[7]); |
405 | //printf("hsn(%x): %d %d %d %d %d %d %d\n",start+i*4,hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]); |
406 | if(i>0) { |
407 | // Don't evict the cycle count at entry points, otherwise the entry |
408 | // stub will have to write it. |
409 | if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2; |
410 | if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP||itype[i-2]==FJUMP)) hsn[CCREG]=2; |
411 | for(j=10;j>=3;j--) |
412 | { |
413 | // Alloc preferred register if available |
414 | if(hsn[r=cur->regmap[preferred_reg]&63]==j) { |
415 | for(hr=0;hr<HOST_REGS;hr++) { |
416 | // Evict both parts of a 64-bit register |
417 | if((cur->regmap[hr]&63)==r) { |
418 | cur->regmap[hr]=-1; |
419 | cur->dirty&=~(1<<hr); |
420 | cur->isconst&=~(1<<hr); |
421 | } |
422 | } |
423 | cur->regmap[preferred_reg]=reg; |
424 | return; |
425 | } |
426 | for(r=1;r<=MAXREG;r++) |
427 | { |
428 | if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) { |
429 | for(hr=0;hr<HOST_REGS;hr++) { |
430 | if(hr!=HOST_CCREG||j<hsn[CCREG]) { |
431 | if(cur->regmap[hr]==r+64) { |
432 | cur->regmap[hr]=reg; |
433 | cur->dirty&=~(1<<hr); |
434 | cur->isconst&=~(1<<hr); |
435 | return; |
436 | } |
437 | } |
438 | } |
439 | for(hr=0;hr<HOST_REGS;hr++) { |
440 | if(hr!=HOST_CCREG||j<hsn[CCREG]) { |
441 | if(cur->regmap[hr]==r) { |
442 | cur->regmap[hr]=reg; |
443 | cur->dirty&=~(1<<hr); |
444 | cur->isconst&=~(1<<hr); |
445 | return; |
446 | } |
447 | } |
448 | } |
449 | } |
450 | } |
451 | } |
452 | } |
453 | for(j=10;j>=0;j--) |
454 | { |
455 | for(r=1;r<=MAXREG;r++) |
456 | { |
457 | if(hsn[r]==j) { |
458 | for(hr=0;hr<HOST_REGS;hr++) { |
459 | if(cur->regmap[hr]==r+64) { |
460 | cur->regmap[hr]=reg; |
461 | cur->dirty&=~(1<<hr); |
462 | cur->isconst&=~(1<<hr); |
463 | return; |
464 | } |
465 | } |
466 | for(hr=0;hr<HOST_REGS;hr++) { |
467 | if(cur->regmap[hr]==r) { |
468 | cur->regmap[hr]=reg; |
469 | cur->dirty&=~(1<<hr); |
470 | cur->isconst&=~(1<<hr); |
471 | return; |
472 | } |
473 | } |
474 | } |
475 | } |
476 | } |
c43b5311 |
477 | SysPrintf("This shouldn't happen (alloc_reg)");exit(1); |
57871462 |
478 | } |
479 | |
e2b5e7aa |
480 | static void alloc_reg64(struct regstat *cur,int i,signed char reg) |
57871462 |
481 | { |
482 | int preferred_reg = 8+(reg&1); |
483 | int r,hr; |
9f51b4b9 |
484 | |
57871462 |
485 | // allocate the lower 32 bits |
486 | alloc_reg(cur,i,reg); |
9f51b4b9 |
487 | |
57871462 |
488 | // Don't allocate unused registers |
489 | if((cur->uu>>reg)&1) return; |
9f51b4b9 |
490 | |
57871462 |
491 | // see if the upper half is already allocated |
492 | for(hr=0;hr<HOST_REGS;hr++) |
493 | { |
494 | if(cur->regmap[hr]==reg+64) return; |
495 | } |
9f51b4b9 |
496 | |
57871462 |
497 | // Keep the same mapping if the register was already allocated in a loop |
498 | preferred_reg = loop_reg(i,reg,preferred_reg); |
9f51b4b9 |
499 | |
57871462 |
500 | // Try to allocate the preferred register |
501 | if(cur->regmap[preferred_reg]==-1) { |
502 | cur->regmap[preferred_reg]=reg|64; |
503 | cur->dirty&=~(1<<preferred_reg); |
504 | cur->isconst&=~(1<<preferred_reg); |
505 | return; |
506 | } |
507 | r=cur->regmap[preferred_reg]; |
508 | if(r<64&&((cur->u>>r)&1)) { |
509 | cur->regmap[preferred_reg]=reg|64; |
510 | cur->dirty&=~(1<<preferred_reg); |
511 | cur->isconst&=~(1<<preferred_reg); |
512 | return; |
513 | } |
514 | if(r>=64&&((cur->uu>>(r&63))&1)) { |
515 | cur->regmap[preferred_reg]=reg|64; |
516 | cur->dirty&=~(1<<preferred_reg); |
517 | cur->isconst&=~(1<<preferred_reg); |
518 | return; |
519 | } |
9f51b4b9 |
520 | |
57871462 |
521 | // Clear any unneeded registers |
522 | // We try to keep the mapping consistent, if possible, because it |
523 | // makes branches easier (especially loops). So we try to allocate |
524 | // first (see above) before removing old mappings. If this is not |
525 | // possible then go ahead and clear out the registers that are no |
526 | // longer needed. |
527 | for(hr=HOST_REGS-1;hr>=0;hr--) |
528 | { |
529 | r=cur->regmap[hr]; |
530 | if(r>=0) { |
531 | if(r<64) { |
532 | if((cur->u>>r)&1) {cur->regmap[hr]=-1;break;} |
533 | } |
534 | else |
535 | { |
536 | if((cur->uu>>(r&63))&1) {cur->regmap[hr]=-1;break;} |
537 | } |
538 | } |
539 | } |
540 | // Try to allocate any available register, but prefer |
541 | // registers that have not been used recently. |
542 | if(i>0) { |
543 | for(hr=0;hr<HOST_REGS;hr++) { |
544 | if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) { |
545 | if(regs[i-1].regmap[hr]!=rs1[i-1]&®s[i-1].regmap[hr]!=rs2[i-1]&®s[i-1].regmap[hr]!=rt1[i-1]&®s[i-1].regmap[hr]!=rt2[i-1]) { |
546 | cur->regmap[hr]=reg|64; |
547 | cur->dirty&=~(1<<hr); |
548 | cur->isconst&=~(1<<hr); |
549 | return; |
550 | } |
551 | } |
552 | } |
553 | } |
554 | // Try to allocate any available register |
555 | for(hr=0;hr<HOST_REGS;hr++) { |
556 | if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) { |
557 | cur->regmap[hr]=reg|64; |
558 | cur->dirty&=~(1<<hr); |
559 | cur->isconst&=~(1<<hr); |
560 | return; |
561 | } |
562 | } |
9f51b4b9 |
563 | |
57871462 |
564 | // Ok, now we have to evict someone |
565 | // Pick a register we hopefully won't need soon |
566 | u_char hsn[MAXREG+1]; |
567 | memset(hsn,10,sizeof(hsn)); |
568 | int j; |
569 | lsn(hsn,i,&preferred_reg); |
570 | //printf("eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",cur->regmap[0],cur->regmap[1],cur->regmap[2],cur->regmap[3],cur->regmap[5],cur->regmap[6],cur->regmap[7]); |
571 | //printf("hsn(%x): %d %d %d %d %d %d %d\n",start+i*4,hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]); |
572 | if(i>0) { |
573 | // Don't evict the cycle count at entry points, otherwise the entry |
574 | // stub will have to write it. |
575 | if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2; |
576 | if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP||itype[i-2]==FJUMP)) hsn[CCREG]=2; |
577 | for(j=10;j>=3;j--) |
578 | { |
579 | // Alloc preferred register if available |
580 | if(hsn[r=cur->regmap[preferred_reg]&63]==j) { |
581 | for(hr=0;hr<HOST_REGS;hr++) { |
582 | // Evict both parts of a 64-bit register |
583 | if((cur->regmap[hr]&63)==r) { |
584 | cur->regmap[hr]=-1; |
585 | cur->dirty&=~(1<<hr); |
586 | cur->isconst&=~(1<<hr); |
587 | } |
588 | } |
589 | cur->regmap[preferred_reg]=reg|64; |
590 | return; |
591 | } |
592 | for(r=1;r<=MAXREG;r++) |
593 | { |
594 | if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) { |
595 | for(hr=0;hr<HOST_REGS;hr++) { |
596 | if(hr!=HOST_CCREG||j<hsn[CCREG]) { |
597 | if(cur->regmap[hr]==r+64) { |
598 | cur->regmap[hr]=reg|64; |
599 | cur->dirty&=~(1<<hr); |
600 | cur->isconst&=~(1<<hr); |
601 | return; |
602 | } |
603 | } |
604 | } |
605 | for(hr=0;hr<HOST_REGS;hr++) { |
606 | if(hr!=HOST_CCREG||j<hsn[CCREG]) { |
607 | if(cur->regmap[hr]==r) { |
608 | cur->regmap[hr]=reg|64; |
609 | cur->dirty&=~(1<<hr); |
610 | cur->isconst&=~(1<<hr); |
611 | return; |
612 | } |
613 | } |
614 | } |
615 | } |
616 | } |
617 | } |
618 | } |
619 | for(j=10;j>=0;j--) |
620 | { |
621 | for(r=1;r<=MAXREG;r++) |
622 | { |
623 | if(hsn[r]==j) { |
624 | for(hr=0;hr<HOST_REGS;hr++) { |
625 | if(cur->regmap[hr]==r+64) { |
626 | cur->regmap[hr]=reg|64; |
627 | cur->dirty&=~(1<<hr); |
628 | cur->isconst&=~(1<<hr); |
629 | return; |
630 | } |
631 | } |
632 | for(hr=0;hr<HOST_REGS;hr++) { |
633 | if(cur->regmap[hr]==r) { |
634 | cur->regmap[hr]=reg|64; |
635 | cur->dirty&=~(1<<hr); |
636 | cur->isconst&=~(1<<hr); |
637 | return; |
638 | } |
639 | } |
640 | } |
641 | } |
642 | } |
c43b5311 |
643 | SysPrintf("This shouldn't happen");exit(1); |
57871462 |
644 | } |
645 | |
646 | // Allocate a temporary register. This is done without regard to |
647 | // dirty status or whether the register we request is on the unneeded list |
648 | // Note: This will only allocate one register, even if called multiple times |
e2b5e7aa |
649 | static void alloc_reg_temp(struct regstat *cur,int i,signed char reg) |
57871462 |
650 | { |
651 | int r,hr; |
652 | int preferred_reg = -1; |
9f51b4b9 |
653 | |
57871462 |
654 | // see if it's already allocated |
655 | for(hr=0;hr<HOST_REGS;hr++) |
656 | { |
657 | if(hr!=EXCLUDE_REG&&cur->regmap[hr]==reg) return; |
658 | } |
9f51b4b9 |
659 | |
57871462 |
660 | // Try to allocate any available register |
661 | for(hr=HOST_REGS-1;hr>=0;hr--) { |
662 | if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) { |
663 | cur->regmap[hr]=reg; |
664 | cur->dirty&=~(1<<hr); |
665 | cur->isconst&=~(1<<hr); |
666 | return; |
667 | } |
668 | } |
9f51b4b9 |
669 | |
57871462 |
670 | // Find an unneeded register |
671 | for(hr=HOST_REGS-1;hr>=0;hr--) |
672 | { |
673 | r=cur->regmap[hr]; |
674 | if(r>=0) { |
675 | if(r<64) { |
676 | if((cur->u>>r)&1) { |
677 | if(i==0||((unneeded_reg[i-1]>>r)&1)) { |
678 | cur->regmap[hr]=reg; |
679 | cur->dirty&=~(1<<hr); |
680 | cur->isconst&=~(1<<hr); |
681 | return; |
682 | } |
683 | } |
684 | } |
685 | else |
686 | { |
687 | if((cur->uu>>(r&63))&1) { |
688 | if(i==0||((unneeded_reg_upper[i-1]>>(r&63))&1)) { |
689 | cur->regmap[hr]=reg; |
690 | cur->dirty&=~(1<<hr); |
691 | cur->isconst&=~(1<<hr); |
692 | return; |
693 | } |
694 | } |
695 | } |
696 | } |
697 | } |
9f51b4b9 |
698 | |
57871462 |
699 | // Ok, now we have to evict someone |
700 | // Pick a register we hopefully won't need soon |
701 | // TODO: we might want to follow unconditional jumps here |
702 | // TODO: get rid of dupe code and make this into a function |
703 | u_char hsn[MAXREG+1]; |
704 | memset(hsn,10,sizeof(hsn)); |
705 | int j; |
706 | lsn(hsn,i,&preferred_reg); |
707 | //printf("hsn: %d %d %d %d %d %d %d\n",hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]); |
708 | if(i>0) { |
709 | // Don't evict the cycle count at entry points, otherwise the entry |
710 | // stub will have to write it. |
711 | if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2; |
712 | if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP||itype[i-2]==FJUMP)) hsn[CCREG]=2; |
713 | for(j=10;j>=3;j--) |
714 | { |
715 | for(r=1;r<=MAXREG;r++) |
716 | { |
717 | if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) { |
718 | for(hr=0;hr<HOST_REGS;hr++) { |
719 | if(hr!=HOST_CCREG||hsn[CCREG]>2) { |
720 | if(cur->regmap[hr]==r+64) { |
721 | cur->regmap[hr]=reg; |
722 | cur->dirty&=~(1<<hr); |
723 | cur->isconst&=~(1<<hr); |
724 | return; |
725 | } |
726 | } |
727 | } |
728 | for(hr=0;hr<HOST_REGS;hr++) { |
729 | if(hr!=HOST_CCREG||hsn[CCREG]>2) { |
730 | if(cur->regmap[hr]==r) { |
731 | cur->regmap[hr]=reg; |
732 | cur->dirty&=~(1<<hr); |
733 | cur->isconst&=~(1<<hr); |
734 | return; |
735 | } |
736 | } |
737 | } |
738 | } |
739 | } |
740 | } |
741 | } |
742 | for(j=10;j>=0;j--) |
743 | { |
744 | for(r=1;r<=MAXREG;r++) |
745 | { |
746 | if(hsn[r]==j) { |
747 | for(hr=0;hr<HOST_REGS;hr++) { |
748 | if(cur->regmap[hr]==r+64) { |
749 | cur->regmap[hr]=reg; |
750 | cur->dirty&=~(1<<hr); |
751 | cur->isconst&=~(1<<hr); |
752 | return; |
753 | } |
754 | } |
755 | for(hr=0;hr<HOST_REGS;hr++) { |
756 | if(cur->regmap[hr]==r) { |
757 | cur->regmap[hr]=reg; |
758 | cur->dirty&=~(1<<hr); |
759 | cur->isconst&=~(1<<hr); |
760 | return; |
761 | } |
762 | } |
763 | } |
764 | } |
765 | } |
c43b5311 |
766 | SysPrintf("This shouldn't happen");exit(1); |
57871462 |
767 | } |
e2b5e7aa |
768 | |
57871462 |
769 | // Allocate a specific ARM register. |
e2b5e7aa |
770 | static void alloc_arm_reg(struct regstat *cur,int i,signed char reg,int hr) |
57871462 |
771 | { |
772 | int n; |
f776eb14 |
773 | int dirty=0; |
9f51b4b9 |
774 | |
57871462 |
775 | // see if it's already allocated (and dealloc it) |
776 | for(n=0;n<HOST_REGS;n++) |
777 | { |
f776eb14 |
778 | if(n!=EXCLUDE_REG&&cur->regmap[n]==reg) { |
779 | dirty=(cur->dirty>>n)&1; |
780 | cur->regmap[n]=-1; |
781 | } |
57871462 |
782 | } |
9f51b4b9 |
783 | |
57871462 |
784 | cur->regmap[hr]=reg; |
785 | cur->dirty&=~(1<<hr); |
f776eb14 |
786 | cur->dirty|=dirty<<hr; |
57871462 |
787 | cur->isconst&=~(1<<hr); |
788 | } |
789 | |
790 | // Alloc cycle count into dedicated register |
e2b5e7aa |
791 | static void alloc_cc(struct regstat *cur,int i) |
57871462 |
792 | { |
793 | alloc_arm_reg(cur,i,CCREG,HOST_CCREG); |
794 | } |
795 | |
796 | /* Special alloc */ |
797 | |
798 | |
799 | /* Assembler */ |
800 | |
e2b5e7aa |
801 | static unused char regname[16][4] = { |
57871462 |
802 | "r0", |
803 | "r1", |
804 | "r2", |
805 | "r3", |
806 | "r4", |
807 | "r5", |
808 | "r6", |
809 | "r7", |
810 | "r8", |
811 | "r9", |
812 | "r10", |
813 | "fp", |
814 | "r12", |
815 | "sp", |
816 | "lr", |
817 | "pc"}; |
818 | |
e2b5e7aa |
819 | static void output_w32(u_int word) |
57871462 |
820 | { |
821 | *((u_int *)out)=word; |
822 | out+=4; |
823 | } |
e2b5e7aa |
824 | |
825 | static u_int rd_rn_rm(u_int rd, u_int rn, u_int rm) |
57871462 |
826 | { |
827 | assert(rd<16); |
828 | assert(rn<16); |
829 | assert(rm<16); |
830 | return((rn<<16)|(rd<<12)|rm); |
831 | } |
e2b5e7aa |
832 | |
833 | static u_int rd_rn_imm_shift(u_int rd, u_int rn, u_int imm, u_int shift) |
57871462 |
834 | { |
835 | assert(rd<16); |
836 | assert(rn<16); |
837 | assert(imm<256); |
838 | assert((shift&1)==0); |
839 | return((rn<<16)|(rd<<12)|(((32-shift)&30)<<7)|imm); |
840 | } |
e2b5e7aa |
841 | |
842 | static u_int genimm(u_int imm,u_int *encoded) |
57871462 |
843 | { |
c2e3bd42 |
844 | *encoded=0; |
845 | if(imm==0) return 1; |
57871462 |
846 | int i=32; |
847 | while(i>0) |
848 | { |
849 | if(imm<256) { |
850 | *encoded=((i&30)<<7)|imm; |
851 | return 1; |
852 | } |
853 | imm=(imm>>2)|(imm<<30);i-=2; |
854 | } |
855 | return 0; |
856 | } |
e2b5e7aa |
857 | |
858 | static void genimm_checked(u_int imm,u_int *encoded) |
cfbd3c6e |
859 | { |
860 | u_int ret=genimm(imm,encoded); |
861 | assert(ret); |
581335b0 |
862 | (void)ret; |
cfbd3c6e |
863 | } |
e2b5e7aa |
864 | |
865 | static u_int genjmp(u_int addr) |
57871462 |
866 | { |
867 | int offset=addr-(int)out-8; |
e80343e2 |
868 | if(offset<-33554432||offset>=33554432) { |
869 | if (addr>2) { |
c43b5311 |
870 | SysPrintf("genjmp: out of range: %08x\n", offset); |
e80343e2 |
871 | exit(1); |
872 | } |
873 | return 0; |
874 | } |
57871462 |
875 | return ((u_int)offset>>2)&0xffffff; |
876 | } |
877 | |
e2b5e7aa |
878 | static void emit_mov(int rs,int rt) |
57871462 |
879 | { |
880 | assem_debug("mov %s,%s\n",regname[rt],regname[rs]); |
881 | output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)); |
882 | } |
883 | |
e2b5e7aa |
884 | static void emit_movs(int rs,int rt) |
57871462 |
885 | { |
886 | assem_debug("movs %s,%s\n",regname[rt],regname[rs]); |
887 | output_w32(0xe1b00000|rd_rn_rm(rt,0,rs)); |
888 | } |
889 | |
e2b5e7aa |
890 | static void emit_add(int rs1,int rs2,int rt) |
57871462 |
891 | { |
892 | assem_debug("add %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
893 | output_w32(0xe0800000|rd_rn_rm(rt,rs1,rs2)); |
894 | } |
895 | |
e2b5e7aa |
896 | static void emit_adds(int rs1,int rs2,int rt) |
57871462 |
897 | { |
898 | assem_debug("adds %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
899 | output_w32(0xe0900000|rd_rn_rm(rt,rs1,rs2)); |
900 | } |
901 | |
e2b5e7aa |
902 | static void emit_adcs(int rs1,int rs2,int rt) |
57871462 |
903 | { |
904 | assem_debug("adcs %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
905 | output_w32(0xe0b00000|rd_rn_rm(rt,rs1,rs2)); |
906 | } |
907 | |
e2b5e7aa |
908 | static void emit_sbc(int rs1,int rs2,int rt) |
57871462 |
909 | { |
910 | assem_debug("sbc %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
911 | output_w32(0xe0c00000|rd_rn_rm(rt,rs1,rs2)); |
912 | } |
913 | |
e2b5e7aa |
914 | static void emit_sbcs(int rs1,int rs2,int rt) |
57871462 |
915 | { |
916 | assem_debug("sbcs %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
917 | output_w32(0xe0d00000|rd_rn_rm(rt,rs1,rs2)); |
918 | } |
919 | |
e2b5e7aa |
920 | static void emit_neg(int rs, int rt) |
57871462 |
921 | { |
922 | assem_debug("rsb %s,%s,#0\n",regname[rt],regname[rs]); |
923 | output_w32(0xe2600000|rd_rn_rm(rt,rs,0)); |
924 | } |
925 | |
e2b5e7aa |
926 | static void emit_negs(int rs, int rt) |
57871462 |
927 | { |
928 | assem_debug("rsbs %s,%s,#0\n",regname[rt],regname[rs]); |
929 | output_w32(0xe2700000|rd_rn_rm(rt,rs,0)); |
930 | } |
931 | |
e2b5e7aa |
932 | static void emit_sub(int rs1,int rs2,int rt) |
57871462 |
933 | { |
934 | assem_debug("sub %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
935 | output_w32(0xe0400000|rd_rn_rm(rt,rs1,rs2)); |
936 | } |
937 | |
e2b5e7aa |
938 | static void emit_subs(int rs1,int rs2,int rt) |
57871462 |
939 | { |
940 | assem_debug("subs %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
941 | output_w32(0xe0500000|rd_rn_rm(rt,rs1,rs2)); |
942 | } |
943 | |
e2b5e7aa |
944 | static void emit_zeroreg(int rt) |
57871462 |
945 | { |
946 | assem_debug("mov %s,#0\n",regname[rt]); |
947 | output_w32(0xe3a00000|rd_rn_rm(rt,0,0)); |
948 | } |
949 | |
e2b5e7aa |
950 | static void emit_loadlp(u_int imm,u_int rt) |
790ee18e |
951 | { |
952 | add_literal((int)out,imm); |
953 | assem_debug("ldr %s,pc+? [=%x]\n",regname[rt],imm); |
954 | output_w32(0xe5900000|rd_rn_rm(rt,15,0)); |
955 | } |
e2b5e7aa |
956 | |
957 | static void emit_movw(u_int imm,u_int rt) |
790ee18e |
958 | { |
959 | assert(imm<65536); |
960 | assem_debug("movw %s,#%d (0x%x)\n",regname[rt],imm,imm); |
961 | output_w32(0xe3000000|rd_rn_rm(rt,0,0)|(imm&0xfff)|((imm<<4)&0xf0000)); |
962 | } |
e2b5e7aa |
963 | |
964 | static void emit_movt(u_int imm,u_int rt) |
790ee18e |
965 | { |
966 | assem_debug("movt %s,#%d (0x%x)\n",regname[rt],imm&0xffff0000,imm&0xffff0000); |
967 | output_w32(0xe3400000|rd_rn_rm(rt,0,0)|((imm>>16)&0xfff)|((imm>>12)&0xf0000)); |
968 | } |
e2b5e7aa |
969 | |
970 | static void emit_movimm(u_int imm,u_int rt) |
790ee18e |
971 | { |
972 | u_int armval; |
973 | if(genimm(imm,&armval)) { |
974 | assem_debug("mov %s,#%d\n",regname[rt],imm); |
975 | output_w32(0xe3a00000|rd_rn_rm(rt,0,0)|armval); |
976 | }else if(genimm(~imm,&armval)) { |
977 | assem_debug("mvn %s,#%d\n",regname[rt],imm); |
978 | output_w32(0xe3e00000|rd_rn_rm(rt,0,0)|armval); |
979 | }else if(imm<65536) { |
665f33e1 |
980 | #ifndef HAVE_ARMV7 |
790ee18e |
981 | assem_debug("mov %s,#%d\n",regname[rt],imm&0xFF00); |
982 | output_w32(0xe3a00000|rd_rn_imm_shift(rt,0,imm>>8,8)); |
983 | assem_debug("add %s,%s,#%d\n",regname[rt],regname[rt],imm&0xFF); |
984 | output_w32(0xe2800000|rd_rn_imm_shift(rt,rt,imm&0xff,0)); |
985 | #else |
986 | emit_movw(imm,rt); |
987 | #endif |
988 | }else{ |
665f33e1 |
989 | #ifndef HAVE_ARMV7 |
790ee18e |
990 | emit_loadlp(imm,rt); |
991 | #else |
992 | emit_movw(imm&0x0000FFFF,rt); |
993 | emit_movt(imm&0xFFFF0000,rt); |
994 | #endif |
995 | } |
996 | } |
e2b5e7aa |
997 | |
998 | static void emit_pcreladdr(u_int rt) |
790ee18e |
999 | { |
1000 | assem_debug("add %s,pc,#?\n",regname[rt]); |
1001 | output_w32(0xe2800000|rd_rn_rm(rt,15,0)); |
1002 | } |
1003 | |
e2b5e7aa |
1004 | static void emit_loadreg(int r, int hr) |
57871462 |
1005 | { |
3d624f89 |
1006 | if(r&64) { |
c43b5311 |
1007 | SysPrintf("64bit load in 32bit mode!\n"); |
7f2607ea |
1008 | assert(0); |
1009 | return; |
3d624f89 |
1010 | } |
57871462 |
1011 | if((r&63)==0) |
1012 | emit_zeroreg(hr); |
1013 | else { |
3d624f89 |
1014 | int addr=((int)reg)+((r&63)<<REG_SHIFT)+((r&64)>>4); |
57871462 |
1015 | if((r&63)==HIREG) addr=(int)&hi+((r&64)>>4); |
1016 | if((r&63)==LOREG) addr=(int)&lo+((r&64)>>4); |
1017 | if(r==CCREG) addr=(int)&cycle_count; |
1018 | if(r==CSREG) addr=(int)&Status; |
1019 | if(r==FSREG) addr=(int)&FCR31; |
1020 | if(r==INVCP) addr=(int)&invc_ptr; |
1021 | u_int offset = addr-(u_int)&dynarec_local; |
1022 | assert(offset<4096); |
1023 | assem_debug("ldr %s,fp+%d\n",regname[hr],offset); |
1024 | output_w32(0xe5900000|rd_rn_rm(hr,FP,0)|offset); |
1025 | } |
1026 | } |
e2b5e7aa |
1027 | |
1028 | static void emit_storereg(int r, int hr) |
57871462 |
1029 | { |
3d624f89 |
1030 | if(r&64) { |
c43b5311 |
1031 | SysPrintf("64bit store in 32bit mode!\n"); |
7f2607ea |
1032 | assert(0); |
1033 | return; |
3d624f89 |
1034 | } |
3d624f89 |
1035 | int addr=((int)reg)+((r&63)<<REG_SHIFT)+((r&64)>>4); |
57871462 |
1036 | if((r&63)==HIREG) addr=(int)&hi+((r&64)>>4); |
1037 | if((r&63)==LOREG) addr=(int)&lo+((r&64)>>4); |
1038 | if(r==CCREG) addr=(int)&cycle_count; |
1039 | if(r==FSREG) addr=(int)&FCR31; |
1040 | u_int offset = addr-(u_int)&dynarec_local; |
1041 | assert(offset<4096); |
1042 | assem_debug("str %s,fp+%d\n",regname[hr],offset); |
1043 | output_w32(0xe5800000|rd_rn_rm(hr,FP,0)|offset); |
1044 | } |
1045 | |
e2b5e7aa |
1046 | static void emit_test(int rs, int rt) |
57871462 |
1047 | { |
1048 | assem_debug("tst %s,%s\n",regname[rs],regname[rt]); |
1049 | output_w32(0xe1100000|rd_rn_rm(0,rs,rt)); |
1050 | } |
1051 | |
e2b5e7aa |
1052 | static void emit_testimm(int rs,int imm) |
57871462 |
1053 | { |
1054 | u_int armval; |
5a05d80c |
1055 | assem_debug("tst %s,#%d\n",regname[rs],imm); |
cfbd3c6e |
1056 | genimm_checked(imm,&armval); |
57871462 |
1057 | output_w32(0xe3100000|rd_rn_rm(0,rs,0)|armval); |
1058 | } |
1059 | |
e2b5e7aa |
1060 | static void emit_testeqimm(int rs,int imm) |
b9b61529 |
1061 | { |
1062 | u_int armval; |
1063 | assem_debug("tsteq %s,$%d\n",regname[rs],imm); |
cfbd3c6e |
1064 | genimm_checked(imm,&armval); |
b9b61529 |
1065 | output_w32(0x03100000|rd_rn_rm(0,rs,0)|armval); |
1066 | } |
1067 | |
e2b5e7aa |
1068 | static void emit_not(int rs,int rt) |
57871462 |
1069 | { |
1070 | assem_debug("mvn %s,%s\n",regname[rt],regname[rs]); |
1071 | output_w32(0xe1e00000|rd_rn_rm(rt,0,rs)); |
1072 | } |
1073 | |
e2b5e7aa |
1074 | static void emit_mvnmi(int rs,int rt) |
b9b61529 |
1075 | { |
1076 | assem_debug("mvnmi %s,%s\n",regname[rt],regname[rs]); |
1077 | output_w32(0x41e00000|rd_rn_rm(rt,0,rs)); |
1078 | } |
1079 | |
e2b5e7aa |
1080 | static void emit_and(u_int rs1,u_int rs2,u_int rt) |
57871462 |
1081 | { |
1082 | assem_debug("and %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
1083 | output_w32(0xe0000000|rd_rn_rm(rt,rs1,rs2)); |
1084 | } |
1085 | |
e2b5e7aa |
1086 | static void emit_or(u_int rs1,u_int rs2,u_int rt) |
57871462 |
1087 | { |
1088 | assem_debug("orr %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
1089 | output_w32(0xe1800000|rd_rn_rm(rt,rs1,rs2)); |
1090 | } |
e2b5e7aa |
1091 | |
1092 | static void emit_or_and_set_flags(int rs1,int rs2,int rt) |
57871462 |
1093 | { |
1094 | assem_debug("orrs %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
1095 | output_w32(0xe1900000|rd_rn_rm(rt,rs1,rs2)); |
1096 | } |
1097 | |
e2b5e7aa |
1098 | static void emit_orrshl_imm(u_int rs,u_int imm,u_int rt) |
f70d384d |
1099 | { |
1100 | assert(rs<16); |
1101 | assert(rt<16); |
1102 | assert(imm<32); |
1103 | assem_debug("orr %s,%s,%s,lsl #%d\n",regname[rt],regname[rt],regname[rs],imm); |
1104 | output_w32(0xe1800000|rd_rn_rm(rt,rt,rs)|(imm<<7)); |
1105 | } |
1106 | |
e2b5e7aa |
1107 | static void emit_orrshr_imm(u_int rs,u_int imm,u_int rt) |
576bbd8f |
1108 | { |
1109 | assert(rs<16); |
1110 | assert(rt<16); |
1111 | assert(imm<32); |
1112 | assem_debug("orr %s,%s,%s,lsr #%d\n",regname[rt],regname[rt],regname[rs],imm); |
1113 | output_w32(0xe1800020|rd_rn_rm(rt,rt,rs)|(imm<<7)); |
1114 | } |
1115 | |
e2b5e7aa |
1116 | static void emit_xor(u_int rs1,u_int rs2,u_int rt) |
57871462 |
1117 | { |
1118 | assem_debug("eor %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
1119 | output_w32(0xe0200000|rd_rn_rm(rt,rs1,rs2)); |
1120 | } |
1121 | |
e2b5e7aa |
1122 | static void emit_addimm(u_int rs,int imm,u_int rt) |
57871462 |
1123 | { |
1124 | assert(rs<16); |
1125 | assert(rt<16); |
1126 | if(imm!=0) { |
57871462 |
1127 | u_int armval; |
1128 | if(genimm(imm,&armval)) { |
1129 | assem_debug("add %s,%s,#%d\n",regname[rt],regname[rs],imm); |
1130 | output_w32(0xe2800000|rd_rn_rm(rt,rs,0)|armval); |
1131 | }else if(genimm(-imm,&armval)) { |
8a0a8423 |
1132 | assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rs],-imm); |
57871462 |
1133 | output_w32(0xe2400000|rd_rn_rm(rt,rs,0)|armval); |
397614d0 |
1134 | #ifdef HAVE_ARMV7 |
1135 | }else if(rt!=rs&&(u_int)imm<65536) { |
1136 | emit_movw(imm&0x0000ffff,rt); |
1137 | emit_add(rs,rt,rt); |
1138 | }else if(rt!=rs&&(u_int)-imm<65536) { |
1139 | emit_movw(-imm&0x0000ffff,rt); |
1140 | emit_sub(rs,rt,rt); |
1141 | #endif |
1142 | }else if((u_int)-imm<65536) { |
57871462 |
1143 | assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rs],(-imm)&0xFF00); |
1144 | assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rt],(-imm)&0xFF); |
1145 | output_w32(0xe2400000|rd_rn_imm_shift(rt,rs,(-imm)>>8,8)); |
1146 | output_w32(0xe2400000|rd_rn_imm_shift(rt,rt,(-imm)&0xff,0)); |
397614d0 |
1147 | }else { |
1148 | do { |
1149 | int shift = (ffs(imm) - 1) & ~1; |
1150 | int imm8 = imm & (0xff << shift); |
1151 | genimm_checked(imm8,&armval); |
1152 | assem_debug("add %s,%s,#0x%x\n",regname[rt],regname[rs],imm8); |
1153 | output_w32(0xe2800000|rd_rn_rm(rt,rs,0)|armval); |
1154 | rs = rt; |
1155 | imm &= ~imm8; |
1156 | } |
1157 | while (imm != 0); |
57871462 |
1158 | } |
1159 | } |
1160 | else if(rs!=rt) emit_mov(rs,rt); |
1161 | } |
1162 | |
e2b5e7aa |
1163 | static void emit_addimm_and_set_flags(int imm,int rt) |
57871462 |
1164 | { |
1165 | assert(imm>-65536&&imm<65536); |
1166 | u_int armval; |
1167 | if(genimm(imm,&armval)) { |
1168 | assem_debug("adds %s,%s,#%d\n",regname[rt],regname[rt],imm); |
1169 | output_w32(0xe2900000|rd_rn_rm(rt,rt,0)|armval); |
1170 | }else if(genimm(-imm,&armval)) { |
1171 | assem_debug("subs %s,%s,#%d\n",regname[rt],regname[rt],imm); |
1172 | output_w32(0xe2500000|rd_rn_rm(rt,rt,0)|armval); |
1173 | }else if(imm<0) { |
1174 | assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rt],(-imm)&0xFF00); |
1175 | assem_debug("subs %s,%s,#%d\n",regname[rt],regname[rt],(-imm)&0xFF); |
1176 | output_w32(0xe2400000|rd_rn_imm_shift(rt,rt,(-imm)>>8,8)); |
1177 | output_w32(0xe2500000|rd_rn_imm_shift(rt,rt,(-imm)&0xff,0)); |
1178 | }else{ |
1179 | assem_debug("add %s,%s,#%d\n",regname[rt],regname[rt],imm&0xFF00); |
1180 | assem_debug("adds %s,%s,#%d\n",regname[rt],regname[rt],imm&0xFF); |
1181 | output_w32(0xe2800000|rd_rn_imm_shift(rt,rt,imm>>8,8)); |
1182 | output_w32(0xe2900000|rd_rn_imm_shift(rt,rt,imm&0xff,0)); |
1183 | } |
1184 | } |
e2b5e7aa |
1185 | |
1186 | static void emit_addimm_no_flags(u_int imm,u_int rt) |
57871462 |
1187 | { |
1188 | emit_addimm(rt,imm,rt); |
1189 | } |
1190 | |
e2b5e7aa |
1191 | static void emit_addnop(u_int r) |
57871462 |
1192 | { |
1193 | assert(r<16); |
1194 | assem_debug("add %s,%s,#0 (nop)\n",regname[r],regname[r]); |
1195 | output_w32(0xe2800000|rd_rn_rm(r,r,0)); |
1196 | } |
1197 | |
e2b5e7aa |
1198 | static void emit_adcimm(u_int rs,int imm,u_int rt) |
57871462 |
1199 | { |
1200 | u_int armval; |
cfbd3c6e |
1201 | genimm_checked(imm,&armval); |
57871462 |
1202 | assem_debug("adc %s,%s,#%d\n",regname[rt],regname[rs],imm); |
1203 | output_w32(0xe2a00000|rd_rn_rm(rt,rs,0)|armval); |
1204 | } |
1edfcc68 |
1205 | |
e2b5e7aa |
1206 | static void emit_rscimm(int rs,int imm,u_int rt) |
57871462 |
1207 | { |
1208 | assert(0); |
1209 | u_int armval; |
cfbd3c6e |
1210 | genimm_checked(imm,&armval); |
57871462 |
1211 | assem_debug("rsc %s,%s,#%d\n",regname[rt],regname[rs],imm); |
1212 | output_w32(0xe2e00000|rd_rn_rm(rt,rs,0)|armval); |
1213 | } |
1214 | |
e2b5e7aa |
1215 | static void emit_addimm64_32(int rsh,int rsl,int imm,int rth,int rtl) |
57871462 |
1216 | { |
1217 | // TODO: if(genimm(imm,&armval)) ... |
1218 | // else |
1219 | emit_movimm(imm,HOST_TEMPREG); |
1220 | emit_adds(HOST_TEMPREG,rsl,rtl); |
1221 | emit_adcimm(rsh,0,rth); |
1222 | } |
1223 | |
e2b5e7aa |
1224 | static void emit_andimm(int rs,int imm,int rt) |
57871462 |
1225 | { |
1226 | u_int armval; |
790ee18e |
1227 | if(imm==0) { |
1228 | emit_zeroreg(rt); |
1229 | }else if(genimm(imm,&armval)) { |
57871462 |
1230 | assem_debug("and %s,%s,#%d\n",regname[rt],regname[rs],imm); |
1231 | output_w32(0xe2000000|rd_rn_rm(rt,rs,0)|armval); |
1232 | }else if(genimm(~imm,&armval)) { |
1233 | assem_debug("bic %s,%s,#%d\n",regname[rt],regname[rs],imm); |
1234 | output_w32(0xe3c00000|rd_rn_rm(rt,rs,0)|armval); |
1235 | }else if(imm==65535) { |
332a4533 |
1236 | #ifndef HAVE_ARMV6 |
57871462 |
1237 | assem_debug("bic %s,%s,#FF000000\n",regname[rt],regname[rs]); |
1238 | output_w32(0xe3c00000|rd_rn_rm(rt,rs,0)|0x4FF); |
1239 | assem_debug("bic %s,%s,#00FF0000\n",regname[rt],regname[rt]); |
1240 | output_w32(0xe3c00000|rd_rn_rm(rt,rt,0)|0x8FF); |
1241 | #else |
1242 | assem_debug("uxth %s,%s\n",regname[rt],regname[rs]); |
1243 | output_w32(0xe6ff0070|rd_rn_rm(rt,0,rs)); |
1244 | #endif |
1245 | }else{ |
1246 | assert(imm>0&&imm<65535); |
665f33e1 |
1247 | #ifndef HAVE_ARMV7 |
57871462 |
1248 | assem_debug("mov r14,#%d\n",imm&0xFF00); |
1249 | output_w32(0xe3a00000|rd_rn_imm_shift(HOST_TEMPREG,0,imm>>8,8)); |
1250 | assem_debug("add r14,r14,#%d\n",imm&0xFF); |
1251 | output_w32(0xe2800000|rd_rn_imm_shift(HOST_TEMPREG,HOST_TEMPREG,imm&0xff,0)); |
1252 | #else |
1253 | emit_movw(imm,HOST_TEMPREG); |
1254 | #endif |
1255 | assem_debug("and %s,%s,r14\n",regname[rt],regname[rs]); |
1256 | output_w32(0xe0000000|rd_rn_rm(rt,rs,HOST_TEMPREG)); |
1257 | } |
1258 | } |
1259 | |
e2b5e7aa |
1260 | static void emit_orimm(int rs,int imm,int rt) |
57871462 |
1261 | { |
1262 | u_int armval; |
790ee18e |
1263 | if(imm==0) { |
1264 | if(rs!=rt) emit_mov(rs,rt); |
1265 | }else if(genimm(imm,&armval)) { |
57871462 |
1266 | assem_debug("orr %s,%s,#%d\n",regname[rt],regname[rs],imm); |
1267 | output_w32(0xe3800000|rd_rn_rm(rt,rs,0)|armval); |
1268 | }else{ |
1269 | assert(imm>0&&imm<65536); |
1270 | assem_debug("orr %s,%s,#%d\n",regname[rt],regname[rs],imm&0xFF00); |
1271 | assem_debug("orr %s,%s,#%d\n",regname[rt],regname[rs],imm&0xFF); |
1272 | output_w32(0xe3800000|rd_rn_imm_shift(rt,rs,imm>>8,8)); |
1273 | output_w32(0xe3800000|rd_rn_imm_shift(rt,rt,imm&0xff,0)); |
1274 | } |
1275 | } |
1276 | |
e2b5e7aa |
1277 | static void emit_xorimm(int rs,int imm,int rt) |
57871462 |
1278 | { |
57871462 |
1279 | u_int armval; |
790ee18e |
1280 | if(imm==0) { |
1281 | if(rs!=rt) emit_mov(rs,rt); |
1282 | }else if(genimm(imm,&armval)) { |
57871462 |
1283 | assem_debug("eor %s,%s,#%d\n",regname[rt],regname[rs],imm); |
1284 | output_w32(0xe2200000|rd_rn_rm(rt,rs,0)|armval); |
1285 | }else{ |
514ed0d9 |
1286 | assert(imm>0&&imm<65536); |
57871462 |
1287 | assem_debug("eor %s,%s,#%d\n",regname[rt],regname[rs],imm&0xFF00); |
1288 | assem_debug("eor %s,%s,#%d\n",regname[rt],regname[rs],imm&0xFF); |
1289 | output_w32(0xe2200000|rd_rn_imm_shift(rt,rs,imm>>8,8)); |
1290 | output_w32(0xe2200000|rd_rn_imm_shift(rt,rt,imm&0xff,0)); |
1291 | } |
1292 | } |
1293 | |
e2b5e7aa |
1294 | static void emit_shlimm(int rs,u_int imm,int rt) |
57871462 |
1295 | { |
1296 | assert(imm>0); |
1297 | assert(imm<32); |
1298 | //if(imm==1) ... |
1299 | assem_debug("lsl %s,%s,#%d\n",regname[rt],regname[rs],imm); |
1300 | output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|(imm<<7)); |
1301 | } |
1302 | |
e2b5e7aa |
1303 | static void emit_lsls_imm(int rs,int imm,int rt) |
c6c3b1b3 |
1304 | { |
1305 | assert(imm>0); |
1306 | assert(imm<32); |
1307 | assem_debug("lsls %s,%s,#%d\n",regname[rt],regname[rs],imm); |
1308 | output_w32(0xe1b00000|rd_rn_rm(rt,0,rs)|(imm<<7)); |
1309 | } |
1310 | |
e2b5e7aa |
1311 | static unused void emit_lslpls_imm(int rs,int imm,int rt) |
665f33e1 |
1312 | { |
1313 | assert(imm>0); |
1314 | assert(imm<32); |
1315 | assem_debug("lslpls %s,%s,#%d\n",regname[rt],regname[rs],imm); |
1316 | output_w32(0x51b00000|rd_rn_rm(rt,0,rs)|(imm<<7)); |
1317 | } |
1318 | |
e2b5e7aa |
1319 | static void emit_shrimm(int rs,u_int imm,int rt) |
57871462 |
1320 | { |
1321 | assert(imm>0); |
1322 | assert(imm<32); |
1323 | assem_debug("lsr %s,%s,#%d\n",regname[rt],regname[rs],imm); |
1324 | output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|0x20|(imm<<7)); |
1325 | } |
1326 | |
e2b5e7aa |
1327 | static void emit_sarimm(int rs,u_int imm,int rt) |
57871462 |
1328 | { |
1329 | assert(imm>0); |
1330 | assert(imm<32); |
1331 | assem_debug("asr %s,%s,#%d\n",regname[rt],regname[rs],imm); |
1332 | output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|0x40|(imm<<7)); |
1333 | } |
1334 | |
e2b5e7aa |
1335 | static void emit_rorimm(int rs,u_int imm,int rt) |
57871462 |
1336 | { |
1337 | assert(imm>0); |
1338 | assert(imm<32); |
1339 | assem_debug("ror %s,%s,#%d\n",regname[rt],regname[rs],imm); |
1340 | output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|0x60|(imm<<7)); |
1341 | } |
1342 | |
e2b5e7aa |
1343 | static void emit_shldimm(int rs,int rs2,u_int imm,int rt) |
57871462 |
1344 | { |
1345 | assem_debug("shld %%%s,%%%s,%d\n",regname[rt],regname[rs2],imm); |
1346 | assert(imm>0); |
1347 | assert(imm<32); |
1348 | //if(imm==1) ... |
1349 | assem_debug("lsl %s,%s,#%d\n",regname[rt],regname[rs],imm); |
1350 | output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|(imm<<7)); |
1351 | assem_debug("orr %s,%s,%s,lsr #%d\n",regname[rt],regname[rt],regname[rs2],32-imm); |
1352 | output_w32(0xe1800020|rd_rn_rm(rt,rt,rs2)|((32-imm)<<7)); |
1353 | } |
1354 | |
e2b5e7aa |
1355 | static void emit_shrdimm(int rs,int rs2,u_int imm,int rt) |
57871462 |
1356 | { |
1357 | assem_debug("shrd %%%s,%%%s,%d\n",regname[rt],regname[rs2],imm); |
1358 | assert(imm>0); |
1359 | assert(imm<32); |
1360 | //if(imm==1) ... |
1361 | assem_debug("lsr %s,%s,#%d\n",regname[rt],regname[rs],imm); |
1362 | output_w32(0xe1a00020|rd_rn_rm(rt,0,rs)|(imm<<7)); |
1363 | assem_debug("orr %s,%s,%s,lsl #%d\n",regname[rt],regname[rt],regname[rs2],32-imm); |
1364 | output_w32(0xe1800000|rd_rn_rm(rt,rt,rs2)|((32-imm)<<7)); |
1365 | } |
1366 | |
e2b5e7aa |
1367 | static void emit_signextend16(int rs,int rt) |
b9b61529 |
1368 | { |
332a4533 |
1369 | #ifndef HAVE_ARMV6 |
b9b61529 |
1370 | emit_shlimm(rs,16,rt); |
1371 | emit_sarimm(rt,16,rt); |
1372 | #else |
1373 | assem_debug("sxth %s,%s\n",regname[rt],regname[rs]); |
1374 | output_w32(0xe6bf0070|rd_rn_rm(rt,0,rs)); |
1375 | #endif |
1376 | } |
1377 | |
e2b5e7aa |
1378 | static void emit_signextend8(int rs,int rt) |
c6c3b1b3 |
1379 | { |
332a4533 |
1380 | #ifndef HAVE_ARMV6 |
c6c3b1b3 |
1381 | emit_shlimm(rs,24,rt); |
1382 | emit_sarimm(rt,24,rt); |
1383 | #else |
1384 | assem_debug("sxtb %s,%s\n",regname[rt],regname[rs]); |
1385 | output_w32(0xe6af0070|rd_rn_rm(rt,0,rs)); |
1386 | #endif |
1387 | } |
1388 | |
e2b5e7aa |
1389 | static void emit_shl(u_int rs,u_int shift,u_int rt) |
57871462 |
1390 | { |
1391 | assert(rs<16); |
1392 | assert(rt<16); |
1393 | assert(shift<16); |
1394 | //if(imm==1) ... |
1395 | assem_debug("lsl %s,%s,%s\n",regname[rt],regname[rs],regname[shift]); |
1396 | output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|0x10|(shift<<8)); |
1397 | } |
e2b5e7aa |
1398 | |
1399 | static void emit_shr(u_int rs,u_int shift,u_int rt) |
57871462 |
1400 | { |
1401 | assert(rs<16); |
1402 | assert(rt<16); |
1403 | assert(shift<16); |
1404 | assem_debug("lsr %s,%s,%s\n",regname[rt],regname[rs],regname[shift]); |
1405 | output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|0x30|(shift<<8)); |
1406 | } |
e2b5e7aa |
1407 | |
1408 | static void emit_sar(u_int rs,u_int shift,u_int rt) |
57871462 |
1409 | { |
1410 | assert(rs<16); |
1411 | assert(rt<16); |
1412 | assert(shift<16); |
1413 | assem_debug("asr %s,%s,%s\n",regname[rt],regname[rs],regname[shift]); |
1414 | output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|0x50|(shift<<8)); |
1415 | } |
57871462 |
1416 | |
e2b5e7aa |
1417 | static void emit_orrshl(u_int rs,u_int shift,u_int rt) |
57871462 |
1418 | { |
1419 | assert(rs<16); |
1420 | assert(rt<16); |
1421 | assert(shift<16); |
1422 | assem_debug("orr %s,%s,%s,lsl %s\n",regname[rt],regname[rt],regname[rs],regname[shift]); |
1423 | output_w32(0xe1800000|rd_rn_rm(rt,rt,rs)|0x10|(shift<<8)); |
1424 | } |
e2b5e7aa |
1425 | |
1426 | static void emit_orrshr(u_int rs,u_int shift,u_int rt) |
57871462 |
1427 | { |
1428 | assert(rs<16); |
1429 | assert(rt<16); |
1430 | assert(shift<16); |
1431 | assem_debug("orr %s,%s,%s,lsr %s\n",regname[rt],regname[rt],regname[rs],regname[shift]); |
1432 | output_w32(0xe1800000|rd_rn_rm(rt,rt,rs)|0x30|(shift<<8)); |
1433 | } |
1434 | |
e2b5e7aa |
1435 | static void emit_cmpimm(int rs,int imm) |
57871462 |
1436 | { |
1437 | u_int armval; |
1438 | if(genimm(imm,&armval)) { |
5a05d80c |
1439 | assem_debug("cmp %s,#%d\n",regname[rs],imm); |
57871462 |
1440 | output_w32(0xe3500000|rd_rn_rm(0,rs,0)|armval); |
1441 | }else if(genimm(-imm,&armval)) { |
5a05d80c |
1442 | assem_debug("cmn %s,#%d\n",regname[rs],imm); |
57871462 |
1443 | output_w32(0xe3700000|rd_rn_rm(0,rs,0)|armval); |
1444 | }else if(imm>0) { |
1445 | assert(imm<65536); |
57871462 |
1446 | emit_movimm(imm,HOST_TEMPREG); |
57871462 |
1447 | assem_debug("cmp %s,r14\n",regname[rs]); |
1448 | output_w32(0xe1500000|rd_rn_rm(0,rs,HOST_TEMPREG)); |
1449 | }else{ |
1450 | assert(imm>-65536); |
57871462 |
1451 | emit_movimm(-imm,HOST_TEMPREG); |
57871462 |
1452 | assem_debug("cmn %s,r14\n",regname[rs]); |
1453 | output_w32(0xe1700000|rd_rn_rm(0,rs,HOST_TEMPREG)); |
1454 | } |
1455 | } |
1456 | |
e2b5e7aa |
1457 | static void emit_cmovne_imm(int imm,int rt) |
57871462 |
1458 | { |
1459 | assem_debug("movne %s,#%d\n",regname[rt],imm); |
1460 | u_int armval; |
cfbd3c6e |
1461 | genimm_checked(imm,&armval); |
57871462 |
1462 | output_w32(0x13a00000|rd_rn_rm(rt,0,0)|armval); |
1463 | } |
e2b5e7aa |
1464 | |
1465 | static void emit_cmovl_imm(int imm,int rt) |
57871462 |
1466 | { |
1467 | assem_debug("movlt %s,#%d\n",regname[rt],imm); |
1468 | u_int armval; |
cfbd3c6e |
1469 | genimm_checked(imm,&armval); |
57871462 |
1470 | output_w32(0xb3a00000|rd_rn_rm(rt,0,0)|armval); |
1471 | } |
e2b5e7aa |
1472 | |
1473 | static void emit_cmovb_imm(int imm,int rt) |
57871462 |
1474 | { |
1475 | assem_debug("movcc %s,#%d\n",regname[rt],imm); |
1476 | u_int armval; |
cfbd3c6e |
1477 | genimm_checked(imm,&armval); |
57871462 |
1478 | output_w32(0x33a00000|rd_rn_rm(rt,0,0)|armval); |
1479 | } |
e2b5e7aa |
1480 | |
1481 | static void emit_cmovs_imm(int imm,int rt) |
57871462 |
1482 | { |
1483 | assem_debug("movmi %s,#%d\n",regname[rt],imm); |
1484 | u_int armval; |
cfbd3c6e |
1485 | genimm_checked(imm,&armval); |
57871462 |
1486 | output_w32(0x43a00000|rd_rn_rm(rt,0,0)|armval); |
1487 | } |
e2b5e7aa |
1488 | |
1489 | static void emit_cmove_reg(int rs,int rt) |
57871462 |
1490 | { |
1491 | assem_debug("moveq %s,%s\n",regname[rt],regname[rs]); |
1492 | output_w32(0x01a00000|rd_rn_rm(rt,0,rs)); |
1493 | } |
e2b5e7aa |
1494 | |
1495 | static void emit_cmovne_reg(int rs,int rt) |
57871462 |
1496 | { |
1497 | assem_debug("movne %s,%s\n",regname[rt],regname[rs]); |
1498 | output_w32(0x11a00000|rd_rn_rm(rt,0,rs)); |
1499 | } |
e2b5e7aa |
1500 | |
1501 | static void emit_cmovl_reg(int rs,int rt) |
57871462 |
1502 | { |
1503 | assem_debug("movlt %s,%s\n",regname[rt],regname[rs]); |
1504 | output_w32(0xb1a00000|rd_rn_rm(rt,0,rs)); |
1505 | } |
e2b5e7aa |
1506 | |
1507 | static void emit_cmovs_reg(int rs,int rt) |
57871462 |
1508 | { |
1509 | assem_debug("movmi %s,%s\n",regname[rt],regname[rs]); |
1510 | output_w32(0x41a00000|rd_rn_rm(rt,0,rs)); |
1511 | } |
1512 | |
e2b5e7aa |
1513 | static void emit_slti32(int rs,int imm,int rt) |
57871462 |
1514 | { |
1515 | if(rs!=rt) emit_zeroreg(rt); |
1516 | emit_cmpimm(rs,imm); |
1517 | if(rs==rt) emit_movimm(0,rt); |
1518 | emit_cmovl_imm(1,rt); |
1519 | } |
e2b5e7aa |
1520 | |
1521 | static void emit_sltiu32(int rs,int imm,int rt) |
57871462 |
1522 | { |
1523 | if(rs!=rt) emit_zeroreg(rt); |
1524 | emit_cmpimm(rs,imm); |
1525 | if(rs==rt) emit_movimm(0,rt); |
1526 | emit_cmovb_imm(1,rt); |
1527 | } |
e2b5e7aa |
1528 | |
1529 | static void emit_slti64_32(int rsh,int rsl,int imm,int rt) |
57871462 |
1530 | { |
1531 | assert(rsh!=rt); |
1532 | emit_slti32(rsl,imm,rt); |
1533 | if(imm>=0) |
1534 | { |
1535 | emit_test(rsh,rsh); |
1536 | emit_cmovne_imm(0,rt); |
1537 | emit_cmovs_imm(1,rt); |
1538 | } |
1539 | else |
1540 | { |
1541 | emit_cmpimm(rsh,-1); |
1542 | emit_cmovne_imm(0,rt); |
1543 | emit_cmovl_imm(1,rt); |
1544 | } |
1545 | } |
e2b5e7aa |
1546 | |
1547 | static void emit_sltiu64_32(int rsh,int rsl,int imm,int rt) |
57871462 |
1548 | { |
1549 | assert(rsh!=rt); |
1550 | emit_sltiu32(rsl,imm,rt); |
1551 | if(imm>=0) |
1552 | { |
1553 | emit_test(rsh,rsh); |
1554 | emit_cmovne_imm(0,rt); |
1555 | } |
1556 | else |
1557 | { |
1558 | emit_cmpimm(rsh,-1); |
1559 | emit_cmovne_imm(1,rt); |
1560 | } |
1561 | } |
1562 | |
e2b5e7aa |
1563 | static void emit_cmp(int rs,int rt) |
57871462 |
1564 | { |
1565 | assem_debug("cmp %s,%s\n",regname[rs],regname[rt]); |
1566 | output_w32(0xe1500000|rd_rn_rm(0,rs,rt)); |
1567 | } |
e2b5e7aa |
1568 | |
1569 | static void emit_set_gz32(int rs, int rt) |
57871462 |
1570 | { |
1571 | //assem_debug("set_gz32\n"); |
1572 | emit_cmpimm(rs,1); |
1573 | emit_movimm(1,rt); |
1574 | emit_cmovl_imm(0,rt); |
1575 | } |
e2b5e7aa |
1576 | |
1577 | static void emit_set_nz32(int rs, int rt) |
57871462 |
1578 | { |
1579 | //assem_debug("set_nz32\n"); |
1580 | if(rs!=rt) emit_movs(rs,rt); |
1581 | else emit_test(rs,rs); |
1582 | emit_cmovne_imm(1,rt); |
1583 | } |
e2b5e7aa |
1584 | |
1585 | static void emit_set_gz64_32(int rsh, int rsl, int rt) |
57871462 |
1586 | { |
1587 | //assem_debug("set_gz64\n"); |
1588 | emit_set_gz32(rsl,rt); |
1589 | emit_test(rsh,rsh); |
1590 | emit_cmovne_imm(1,rt); |
1591 | emit_cmovs_imm(0,rt); |
1592 | } |
e2b5e7aa |
1593 | |
1594 | static void emit_set_nz64_32(int rsh, int rsl, int rt) |
57871462 |
1595 | { |
1596 | //assem_debug("set_nz64\n"); |
1597 | emit_or_and_set_flags(rsh,rsl,rt); |
1598 | emit_cmovne_imm(1,rt); |
1599 | } |
e2b5e7aa |
1600 | |
1601 | static void emit_set_if_less32(int rs1, int rs2, int rt) |
57871462 |
1602 | { |
1603 | //assem_debug("set if less (%%%s,%%%s),%%%s\n",regname[rs1],regname[rs2],regname[rt]); |
1604 | if(rs1!=rt&&rs2!=rt) emit_zeroreg(rt); |
1605 | emit_cmp(rs1,rs2); |
1606 | if(rs1==rt||rs2==rt) emit_movimm(0,rt); |
1607 | emit_cmovl_imm(1,rt); |
1608 | } |
e2b5e7aa |
1609 | |
1610 | static void emit_set_if_carry32(int rs1, int rs2, int rt) |
57871462 |
1611 | { |
1612 | //assem_debug("set if carry (%%%s,%%%s),%%%s\n",regname[rs1],regname[rs2],regname[rt]); |
1613 | if(rs1!=rt&&rs2!=rt) emit_zeroreg(rt); |
1614 | emit_cmp(rs1,rs2); |
1615 | if(rs1==rt||rs2==rt) emit_movimm(0,rt); |
1616 | emit_cmovb_imm(1,rt); |
1617 | } |
e2b5e7aa |
1618 | |
1619 | static void emit_set_if_less64_32(int u1, int l1, int u2, int l2, int rt) |
57871462 |
1620 | { |
1621 | //assem_debug("set if less64 (%%%s,%%%s,%%%s,%%%s),%%%s\n",regname[u1],regname[l1],regname[u2],regname[l2],regname[rt]); |
1622 | assert(u1!=rt); |
1623 | assert(u2!=rt); |
1624 | emit_cmp(l1,l2); |
1625 | emit_movimm(0,rt); |
1626 | emit_sbcs(u1,u2,HOST_TEMPREG); |
1627 | emit_cmovl_imm(1,rt); |
1628 | } |
e2b5e7aa |
1629 | |
1630 | static void emit_set_if_carry64_32(int u1, int l1, int u2, int l2, int rt) |
57871462 |
1631 | { |
1632 | //assem_debug("set if carry64 (%%%s,%%%s,%%%s,%%%s),%%%s\n",regname[u1],regname[l1],regname[u2],regname[l2],regname[rt]); |
1633 | assert(u1!=rt); |
1634 | assert(u2!=rt); |
1635 | emit_cmp(l1,l2); |
1636 | emit_movimm(0,rt); |
1637 | emit_sbcs(u1,u2,HOST_TEMPREG); |
1638 | emit_cmovb_imm(1,rt); |
1639 | } |
1640 | |
e2b5e7aa |
1641 | static void emit_call(int a) |
57871462 |
1642 | { |
1643 | assem_debug("bl %x (%x+%x)\n",a,(int)out,a-(int)out-8); |
1644 | u_int offset=genjmp(a); |
1645 | output_w32(0xeb000000|offset); |
1646 | } |
e2b5e7aa |
1647 | |
1648 | static void emit_jmp(int a) |
57871462 |
1649 | { |
1650 | assem_debug("b %x (%x+%x)\n",a,(int)out,a-(int)out-8); |
1651 | u_int offset=genjmp(a); |
1652 | output_w32(0xea000000|offset); |
1653 | } |
e2b5e7aa |
1654 | |
1655 | static void emit_jne(int a) |
57871462 |
1656 | { |
1657 | assem_debug("bne %x\n",a); |
1658 | u_int offset=genjmp(a); |
1659 | output_w32(0x1a000000|offset); |
1660 | } |
e2b5e7aa |
1661 | |
1662 | static void emit_jeq(int a) |
57871462 |
1663 | { |
1664 | assem_debug("beq %x\n",a); |
1665 | u_int offset=genjmp(a); |
1666 | output_w32(0x0a000000|offset); |
1667 | } |
e2b5e7aa |
1668 | |
1669 | static void emit_js(int a) |
57871462 |
1670 | { |
1671 | assem_debug("bmi %x\n",a); |
1672 | u_int offset=genjmp(a); |
1673 | output_w32(0x4a000000|offset); |
1674 | } |
e2b5e7aa |
1675 | |
1676 | static void emit_jns(int a) |
57871462 |
1677 | { |
1678 | assem_debug("bpl %x\n",a); |
1679 | u_int offset=genjmp(a); |
1680 | output_w32(0x5a000000|offset); |
1681 | } |
e2b5e7aa |
1682 | |
1683 | static void emit_jl(int a) |
57871462 |
1684 | { |
1685 | assem_debug("blt %x\n",a); |
1686 | u_int offset=genjmp(a); |
1687 | output_w32(0xba000000|offset); |
1688 | } |
e2b5e7aa |
1689 | |
1690 | static void emit_jge(int a) |
57871462 |
1691 | { |
1692 | assem_debug("bge %x\n",a); |
1693 | u_int offset=genjmp(a); |
1694 | output_w32(0xaa000000|offset); |
1695 | } |
e2b5e7aa |
1696 | |
1697 | static void emit_jno(int a) |
57871462 |
1698 | { |
1699 | assem_debug("bvc %x\n",a); |
1700 | u_int offset=genjmp(a); |
1701 | output_w32(0x7a000000|offset); |
1702 | } |
e2b5e7aa |
1703 | |
1704 | static void emit_jc(int a) |
57871462 |
1705 | { |
1706 | assem_debug("bcs %x\n",a); |
1707 | u_int offset=genjmp(a); |
1708 | output_w32(0x2a000000|offset); |
1709 | } |
e2b5e7aa |
1710 | |
1711 | static void emit_jcc(int a) |
57871462 |
1712 | { |
1713 | assem_debug("bcc %x\n",a); |
1714 | u_int offset=genjmp(a); |
1715 | output_w32(0x3a000000|offset); |
1716 | } |
1717 | |
e2b5e7aa |
1718 | static void emit_callreg(u_int r) |
57871462 |
1719 | { |
c6c3b1b3 |
1720 | assert(r<15); |
1721 | assem_debug("blx %s\n",regname[r]); |
1722 | output_w32(0xe12fff30|r); |
57871462 |
1723 | } |
e2b5e7aa |
1724 | |
1725 | static void emit_jmpreg(u_int r) |
57871462 |
1726 | { |
1727 | assem_debug("mov pc,%s\n",regname[r]); |
1728 | output_w32(0xe1a00000|rd_rn_rm(15,0,r)); |
1729 | } |
1730 | |
e2b5e7aa |
1731 | static void emit_readword_indexed(int offset, int rs, int rt) |
57871462 |
1732 | { |
1733 | assert(offset>-4096&&offset<4096); |
1734 | assem_debug("ldr %s,%s+%d\n",regname[rt],regname[rs],offset); |
1735 | if(offset>=0) { |
1736 | output_w32(0xe5900000|rd_rn_rm(rt,rs,0)|offset); |
1737 | }else{ |
1738 | output_w32(0xe5100000|rd_rn_rm(rt,rs,0)|(-offset)); |
1739 | } |
1740 | } |
e2b5e7aa |
1741 | |
1742 | static void emit_readword_dualindexedx4(int rs1, int rs2, int rt) |
57871462 |
1743 | { |
1744 | assem_debug("ldr %s,%s,%s lsl #2\n",regname[rt],regname[rs1],regname[rs2]); |
1745 | output_w32(0xe7900000|rd_rn_rm(rt,rs1,rs2)|0x100); |
1746 | } |
e2b5e7aa |
1747 | |
1748 | static void emit_ldrcc_dualindexed(int rs1, int rs2, int rt) |
c6c3b1b3 |
1749 | { |
1750 | assem_debug("ldrcc %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
1751 | output_w32(0x37900000|rd_rn_rm(rt,rs1,rs2)); |
1752 | } |
e2b5e7aa |
1753 | |
1754 | static void emit_ldrccb_dualindexed(int rs1, int rs2, int rt) |
c6c3b1b3 |
1755 | { |
1756 | assem_debug("ldrccb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
1757 | output_w32(0x37d00000|rd_rn_rm(rt,rs1,rs2)); |
1758 | } |
e2b5e7aa |
1759 | |
1760 | static void emit_ldrccsb_dualindexed(int rs1, int rs2, int rt) |
c6c3b1b3 |
1761 | { |
1762 | assem_debug("ldrccsb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
1763 | output_w32(0x319000d0|rd_rn_rm(rt,rs1,rs2)); |
1764 | } |
e2b5e7aa |
1765 | |
1766 | static void emit_ldrcch_dualindexed(int rs1, int rs2, int rt) |
c6c3b1b3 |
1767 | { |
1768 | assem_debug("ldrcch %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
1769 | output_w32(0x319000b0|rd_rn_rm(rt,rs1,rs2)); |
1770 | } |
e2b5e7aa |
1771 | |
1772 | static void emit_ldrccsh_dualindexed(int rs1, int rs2, int rt) |
c6c3b1b3 |
1773 | { |
1774 | assem_debug("ldrccsh %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
1775 | output_w32(0x319000f0|rd_rn_rm(rt,rs1,rs2)); |
1776 | } |
e2b5e7aa |
1777 | |
1778 | static void emit_readword_indexed_tlb(int addr, int rs, int map, int rt) |
57871462 |
1779 | { |
1780 | if(map<0) emit_readword_indexed(addr, rs, rt); |
1781 | else { |
1782 | assert(addr==0); |
1783 | emit_readword_dualindexedx4(rs, map, rt); |
1784 | } |
1785 | } |
e2b5e7aa |
1786 | |
1787 | static void emit_readdword_indexed_tlb(int addr, int rs, int map, int rh, int rl) |
57871462 |
1788 | { |
1789 | if(map<0) { |
1790 | if(rh>=0) emit_readword_indexed(addr, rs, rh); |
1791 | emit_readword_indexed(addr+4, rs, rl); |
1792 | }else{ |
1793 | assert(rh!=rs); |
1794 | if(rh>=0) emit_readword_indexed_tlb(addr, rs, map, rh); |
1795 | emit_addimm(map,1,map); |
1796 | emit_readword_indexed_tlb(addr, rs, map, rl); |
1797 | } |
1798 | } |
e2b5e7aa |
1799 | |
1800 | static void emit_movsbl_indexed(int offset, int rs, int rt) |
57871462 |
1801 | { |
1802 | assert(offset>-256&&offset<256); |
1803 | assem_debug("ldrsb %s,%s+%d\n",regname[rt],regname[rs],offset); |
1804 | if(offset>=0) { |
1805 | output_w32(0xe1d000d0|rd_rn_rm(rt,rs,0)|((offset<<4)&0xf00)|(offset&0xf)); |
1806 | }else{ |
1807 | output_w32(0xe15000d0|rd_rn_rm(rt,rs,0)|(((-offset)<<4)&0xf00)|((-offset)&0xf)); |
1808 | } |
1809 | } |
e2b5e7aa |
1810 | |
1811 | static void emit_movsbl_indexed_tlb(int addr, int rs, int map, int rt) |
57871462 |
1812 | { |
1813 | if(map<0) emit_movsbl_indexed(addr, rs, rt); |
1814 | else { |
1815 | if(addr==0) { |
1816 | emit_shlimm(map,2,map); |
1817 | assem_debug("ldrsb %s,%s+%s\n",regname[rt],regname[rs],regname[map]); |
1818 | output_w32(0xe19000d0|rd_rn_rm(rt,rs,map)); |
1819 | }else{ |
1820 | assert(addr>-256&&addr<256); |
1821 | assem_debug("add %s,%s,%s,lsl #2\n",regname[rt],regname[rs],regname[map]); |
1822 | output_w32(0xe0800000|rd_rn_rm(rt,rs,map)|(2<<7)); |
1823 | emit_movsbl_indexed(addr, rt, rt); |
1824 | } |
1825 | } |
1826 | } |
e2b5e7aa |
1827 | |
1828 | static void emit_movswl_indexed(int offset, int rs, int rt) |
57871462 |
1829 | { |
1830 | assert(offset>-256&&offset<256); |
1831 | assem_debug("ldrsh %s,%s+%d\n",regname[rt],regname[rs],offset); |
1832 | if(offset>=0) { |
1833 | output_w32(0xe1d000f0|rd_rn_rm(rt,rs,0)|((offset<<4)&0xf00)|(offset&0xf)); |
1834 | }else{ |
1835 | output_w32(0xe15000f0|rd_rn_rm(rt,rs,0)|(((-offset)<<4)&0xf00)|((-offset)&0xf)); |
1836 | } |
1837 | } |
e2b5e7aa |
1838 | |
1839 | static void emit_movzbl_indexed(int offset, int rs, int rt) |
57871462 |
1840 | { |
1841 | assert(offset>-4096&&offset<4096); |
1842 | assem_debug("ldrb %s,%s+%d\n",regname[rt],regname[rs],offset); |
1843 | if(offset>=0) { |
1844 | output_w32(0xe5d00000|rd_rn_rm(rt,rs,0)|offset); |
1845 | }else{ |
1846 | output_w32(0xe5500000|rd_rn_rm(rt,rs,0)|(-offset)); |
1847 | } |
1848 | } |
e2b5e7aa |
1849 | |
1850 | static void emit_movzbl_dualindexedx4(int rs1, int rs2, int rt) |
57871462 |
1851 | { |
1852 | assem_debug("ldrb %s,%s,%s lsl #2\n",regname[rt],regname[rs1],regname[rs2]); |
1853 | output_w32(0xe7d00000|rd_rn_rm(rt,rs1,rs2)|0x100); |
1854 | } |
e2b5e7aa |
1855 | |
1856 | static void emit_movzbl_indexed_tlb(int addr, int rs, int map, int rt) |
57871462 |
1857 | { |
1858 | if(map<0) emit_movzbl_indexed(addr, rs, rt); |
1859 | else { |
1860 | if(addr==0) { |
1861 | emit_movzbl_dualindexedx4(rs, map, rt); |
1862 | }else{ |
1863 | emit_addimm(rs,addr,rt); |
1864 | emit_movzbl_dualindexedx4(rt, map, rt); |
1865 | } |
1866 | } |
1867 | } |
e2b5e7aa |
1868 | |
1869 | static void emit_movzwl_indexed(int offset, int rs, int rt) |
57871462 |
1870 | { |
1871 | assert(offset>-256&&offset<256); |
1872 | assem_debug("ldrh %s,%s+%d\n",regname[rt],regname[rs],offset); |
1873 | if(offset>=0) { |
1874 | output_w32(0xe1d000b0|rd_rn_rm(rt,rs,0)|((offset<<4)&0xf00)|(offset&0xf)); |
1875 | }else{ |
1876 | output_w32(0xe15000b0|rd_rn_rm(rt,rs,0)|(((-offset)<<4)&0xf00)|((-offset)&0xf)); |
1877 | } |
1878 | } |
e2b5e7aa |
1879 | |
054175e9 |
1880 | static void emit_ldrd(int offset, int rs, int rt) |
1881 | { |
1882 | assert(offset>-256&&offset<256); |
1883 | assem_debug("ldrd %s,%s+%d\n",regname[rt],regname[rs],offset); |
1884 | if(offset>=0) { |
1885 | output_w32(0xe1c000d0|rd_rn_rm(rt,rs,0)|((offset<<4)&0xf00)|(offset&0xf)); |
1886 | }else{ |
1887 | output_w32(0xe14000d0|rd_rn_rm(rt,rs,0)|(((-offset)<<4)&0xf00)|((-offset)&0xf)); |
1888 | } |
1889 | } |
e2b5e7aa |
1890 | |
1891 | static void emit_readword(int addr, int rt) |
57871462 |
1892 | { |
1893 | u_int offset = addr-(u_int)&dynarec_local; |
1894 | assert(offset<4096); |
1895 | assem_debug("ldr %s,fp+%d\n",regname[rt],offset); |
1896 | output_w32(0xe5900000|rd_rn_rm(rt,FP,0)|offset); |
1897 | } |
e2b5e7aa |
1898 | |
1899 | static unused void emit_movsbl(int addr, int rt) |
57871462 |
1900 | { |
1901 | u_int offset = addr-(u_int)&dynarec_local; |
1902 | assert(offset<256); |
1903 | assem_debug("ldrsb %s,fp+%d\n",regname[rt],offset); |
1904 | output_w32(0xe1d000d0|rd_rn_rm(rt,FP,0)|((offset<<4)&0xf00)|(offset&0xf)); |
1905 | } |
e2b5e7aa |
1906 | |
1907 | static unused void emit_movswl(int addr, int rt) |
57871462 |
1908 | { |
1909 | u_int offset = addr-(u_int)&dynarec_local; |
1910 | assert(offset<256); |
1911 | assem_debug("ldrsh %s,fp+%d\n",regname[rt],offset); |
1912 | output_w32(0xe1d000f0|rd_rn_rm(rt,FP,0)|((offset<<4)&0xf00)|(offset&0xf)); |
1913 | } |
e2b5e7aa |
1914 | |
1915 | static unused void emit_movzbl(int addr, int rt) |
57871462 |
1916 | { |
1917 | u_int offset = addr-(u_int)&dynarec_local; |
1918 | assert(offset<4096); |
1919 | assem_debug("ldrb %s,fp+%d\n",regname[rt],offset); |
1920 | output_w32(0xe5d00000|rd_rn_rm(rt,FP,0)|offset); |
1921 | } |
e2b5e7aa |
1922 | |
1923 | static unused void emit_movzwl(int addr, int rt) |
57871462 |
1924 | { |
1925 | u_int offset = addr-(u_int)&dynarec_local; |
1926 | assert(offset<256); |
1927 | assem_debug("ldrh %s,fp+%d\n",regname[rt],offset); |
1928 | output_w32(0xe1d000b0|rd_rn_rm(rt,FP,0)|((offset<<4)&0xf00)|(offset&0xf)); |
1929 | } |
57871462 |
1930 | |
e2b5e7aa |
1931 | static void emit_writeword_indexed(int rt, int offset, int rs) |
57871462 |
1932 | { |
1933 | assert(offset>-4096&&offset<4096); |
1934 | assem_debug("str %s,%s+%d\n",regname[rt],regname[rs],offset); |
1935 | if(offset>=0) { |
1936 | output_w32(0xe5800000|rd_rn_rm(rt,rs,0)|offset); |
1937 | }else{ |
1938 | output_w32(0xe5000000|rd_rn_rm(rt,rs,0)|(-offset)); |
1939 | } |
1940 | } |
e2b5e7aa |
1941 | |
1942 | static void emit_writeword_dualindexedx4(int rt, int rs1, int rs2) |
57871462 |
1943 | { |
1944 | assem_debug("str %s,%s,%s lsl #2\n",regname[rt],regname[rs1],regname[rs2]); |
1945 | output_w32(0xe7800000|rd_rn_rm(rt,rs1,rs2)|0x100); |
1946 | } |
e2b5e7aa |
1947 | |
1948 | static void emit_writeword_indexed_tlb(int rt, int addr, int rs, int map, int temp) |
57871462 |
1949 | { |
1950 | if(map<0) emit_writeword_indexed(rt, addr, rs); |
1951 | else { |
1952 | assert(addr==0); |
1953 | emit_writeword_dualindexedx4(rt, rs, map); |
1954 | } |
1955 | } |
e2b5e7aa |
1956 | |
1957 | static void emit_writedword_indexed_tlb(int rh, int rl, int addr, int rs, int map, int temp) |
57871462 |
1958 | { |
1959 | if(map<0) { |
1960 | if(rh>=0) emit_writeword_indexed(rh, addr, rs); |
1961 | emit_writeword_indexed(rl, addr+4, rs); |
1962 | }else{ |
1963 | assert(rh>=0); |
1964 | if(temp!=rs) emit_addimm(map,1,temp); |
1965 | emit_writeword_indexed_tlb(rh, addr, rs, map, temp); |
1966 | if(temp!=rs) emit_writeword_indexed_tlb(rl, addr, rs, temp, temp); |
1967 | else { |
1968 | emit_addimm(rs,4,rs); |
1969 | emit_writeword_indexed_tlb(rl, addr, rs, map, temp); |
1970 | } |
1971 | } |
1972 | } |
e2b5e7aa |
1973 | |
1974 | static void emit_writehword_indexed(int rt, int offset, int rs) |
57871462 |
1975 | { |
1976 | assert(offset>-256&&offset<256); |
1977 | assem_debug("strh %s,%s+%d\n",regname[rt],regname[rs],offset); |
1978 | if(offset>=0) { |
1979 | output_w32(0xe1c000b0|rd_rn_rm(rt,rs,0)|((offset<<4)&0xf00)|(offset&0xf)); |
1980 | }else{ |
1981 | output_w32(0xe14000b0|rd_rn_rm(rt,rs,0)|(((-offset)<<4)&0xf00)|((-offset)&0xf)); |
1982 | } |
1983 | } |
e2b5e7aa |
1984 | |
1985 | static void emit_writebyte_indexed(int rt, int offset, int rs) |
57871462 |
1986 | { |
1987 | assert(offset>-4096&&offset<4096); |
1988 | assem_debug("strb %s,%s+%d\n",regname[rt],regname[rs],offset); |
1989 | if(offset>=0) { |
1990 | output_w32(0xe5c00000|rd_rn_rm(rt,rs,0)|offset); |
1991 | }else{ |
1992 | output_w32(0xe5400000|rd_rn_rm(rt,rs,0)|(-offset)); |
1993 | } |
1994 | } |
e2b5e7aa |
1995 | |
1996 | static void emit_writebyte_dualindexedx4(int rt, int rs1, int rs2) |
57871462 |
1997 | { |
1998 | assem_debug("strb %s,%s,%s lsl #2\n",regname[rt],regname[rs1],regname[rs2]); |
1999 | output_w32(0xe7c00000|rd_rn_rm(rt,rs1,rs2)|0x100); |
2000 | } |
e2b5e7aa |
2001 | |
2002 | static void emit_writebyte_indexed_tlb(int rt, int addr, int rs, int map, int temp) |
57871462 |
2003 | { |
2004 | if(map<0) emit_writebyte_indexed(rt, addr, rs); |
2005 | else { |
2006 | if(addr==0) { |
2007 | emit_writebyte_dualindexedx4(rt, rs, map); |
2008 | }else{ |
2009 | emit_addimm(rs,addr,temp); |
2010 | emit_writebyte_dualindexedx4(rt, temp, map); |
2011 | } |
2012 | } |
2013 | } |
e2b5e7aa |
2014 | |
2015 | static void emit_strcc_dualindexed(int rs1, int rs2, int rt) |
b96d3df7 |
2016 | { |
2017 | assem_debug("strcc %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
2018 | output_w32(0x37800000|rd_rn_rm(rt,rs1,rs2)); |
2019 | } |
e2b5e7aa |
2020 | |
2021 | static void emit_strccb_dualindexed(int rs1, int rs2, int rt) |
b96d3df7 |
2022 | { |
2023 | assem_debug("strccb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
2024 | output_w32(0x37c00000|rd_rn_rm(rt,rs1,rs2)); |
2025 | } |
e2b5e7aa |
2026 | |
2027 | static void emit_strcch_dualindexed(int rs1, int rs2, int rt) |
b96d3df7 |
2028 | { |
2029 | assem_debug("strcch %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
2030 | output_w32(0x318000b0|rd_rn_rm(rt,rs1,rs2)); |
2031 | } |
e2b5e7aa |
2032 | |
2033 | static void emit_writeword(int rt, int addr) |
57871462 |
2034 | { |
2035 | u_int offset = addr-(u_int)&dynarec_local; |
2036 | assert(offset<4096); |
2037 | assem_debug("str %s,fp+%d\n",regname[rt],offset); |
2038 | output_w32(0xe5800000|rd_rn_rm(rt,FP,0)|offset); |
2039 | } |
e2b5e7aa |
2040 | |
2041 | static unused void emit_writehword(int rt, int addr) |
57871462 |
2042 | { |
2043 | u_int offset = addr-(u_int)&dynarec_local; |
2044 | assert(offset<256); |
2045 | assem_debug("strh %s,fp+%d\n",regname[rt],offset); |
2046 | output_w32(0xe1c000b0|rd_rn_rm(rt,FP,0)|((offset<<4)&0xf00)|(offset&0xf)); |
2047 | } |
e2b5e7aa |
2048 | |
2049 | static unused void emit_writebyte(int rt, int addr) |
57871462 |
2050 | { |
2051 | u_int offset = addr-(u_int)&dynarec_local; |
2052 | assert(offset<4096); |
74426039 |
2053 | assem_debug("strb %s,fp+%d\n",regname[rt],offset); |
57871462 |
2054 | output_w32(0xe5c00000|rd_rn_rm(rt,FP,0)|offset); |
2055 | } |
57871462 |
2056 | |
e2b5e7aa |
2057 | static void emit_umull(u_int rs1, u_int rs2, u_int hi, u_int lo) |
57871462 |
2058 | { |
2059 | assem_debug("umull %s, %s, %s, %s\n",regname[lo],regname[hi],regname[rs1],regname[rs2]); |
2060 | assert(rs1<16); |
2061 | assert(rs2<16); |
2062 | assert(hi<16); |
2063 | assert(lo<16); |
2064 | output_w32(0xe0800090|(hi<<16)|(lo<<12)|(rs2<<8)|rs1); |
2065 | } |
e2b5e7aa |
2066 | |
2067 | static void emit_smull(u_int rs1, u_int rs2, u_int hi, u_int lo) |
57871462 |
2068 | { |
2069 | assem_debug("smull %s, %s, %s, %s\n",regname[lo],regname[hi],regname[rs1],regname[rs2]); |
2070 | assert(rs1<16); |
2071 | assert(rs2<16); |
2072 | assert(hi<16); |
2073 | assert(lo<16); |
2074 | output_w32(0xe0c00090|(hi<<16)|(lo<<12)|(rs2<<8)|rs1); |
2075 | } |
2076 | |
e2b5e7aa |
2077 | static void emit_clz(int rs,int rt) |
57871462 |
2078 | { |
2079 | assem_debug("clz %s,%s\n",regname[rt],regname[rs]); |
2080 | output_w32(0xe16f0f10|rd_rn_rm(rt,0,rs)); |
2081 | } |
2082 | |
e2b5e7aa |
2083 | static void emit_subcs(int rs1,int rs2,int rt) |
57871462 |
2084 | { |
2085 | assem_debug("subcs %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
2086 | output_w32(0x20400000|rd_rn_rm(rt,rs1,rs2)); |
2087 | } |
2088 | |
e2b5e7aa |
2089 | static void emit_shrcc_imm(int rs,u_int imm,int rt) |
57871462 |
2090 | { |
2091 | assert(imm>0); |
2092 | assert(imm<32); |
2093 | assem_debug("lsrcc %s,%s,#%d\n",regname[rt],regname[rs],imm); |
2094 | output_w32(0x31a00000|rd_rn_rm(rt,0,rs)|0x20|(imm<<7)); |
2095 | } |
2096 | |
e2b5e7aa |
2097 | static void emit_shrne_imm(int rs,u_int imm,int rt) |
b1be1eee |
2098 | { |
2099 | assert(imm>0); |
2100 | assert(imm<32); |
2101 | assem_debug("lsrne %s,%s,#%d\n",regname[rt],regname[rs],imm); |
2102 | output_w32(0x11a00000|rd_rn_rm(rt,0,rs)|0x20|(imm<<7)); |
2103 | } |
2104 | |
e2b5e7aa |
2105 | static void emit_negmi(int rs, int rt) |
57871462 |
2106 | { |
2107 | assem_debug("rsbmi %s,%s,#0\n",regname[rt],regname[rs]); |
2108 | output_w32(0x42600000|rd_rn_rm(rt,rs,0)); |
2109 | } |
2110 | |
e2b5e7aa |
2111 | static void emit_negsmi(int rs, int rt) |
57871462 |
2112 | { |
2113 | assem_debug("rsbsmi %s,%s,#0\n",regname[rt],regname[rs]); |
2114 | output_w32(0x42700000|rd_rn_rm(rt,rs,0)); |
2115 | } |
2116 | |
e2b5e7aa |
2117 | static void emit_orreq(u_int rs1,u_int rs2,u_int rt) |
57871462 |
2118 | { |
2119 | assem_debug("orreq %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
2120 | output_w32(0x01800000|rd_rn_rm(rt,rs1,rs2)); |
2121 | } |
2122 | |
e2b5e7aa |
2123 | static void emit_orrne(u_int rs1,u_int rs2,u_int rt) |
57871462 |
2124 | { |
2125 | assem_debug("orrne %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); |
2126 | output_w32(0x11800000|rd_rn_rm(rt,rs1,rs2)); |
2127 | } |
2128 | |
e2b5e7aa |
2129 | static void emit_bic_lsl(u_int rs1,u_int rs2,u_int shift,u_int rt) |
57871462 |
2130 | { |
2131 | assem_debug("bic %s,%s,%s lsl %s\n",regname[rt],regname[rs1],regname[rs2],regname[shift]); |
2132 | output_w32(0xe1C00000|rd_rn_rm(rt,rs1,rs2)|0x10|(shift<<8)); |
2133 | } |
2134 | |
e2b5e7aa |
2135 | static void emit_biceq_lsl(u_int rs1,u_int rs2,u_int shift,u_int rt) |
57871462 |
2136 | { |
2137 | assem_debug("biceq %s,%s,%s lsl %s\n",regname[rt],regname[rs1],regname[rs2],regname[shift]); |
2138 | output_w32(0x01C00000|rd_rn_rm(rt,rs1,rs2)|0x10|(shift<<8)); |
2139 | } |
2140 | |
e2b5e7aa |
2141 | static void emit_bicne_lsl(u_int rs1,u_int rs2,u_int shift,u_int rt) |
57871462 |
2142 | { |
2143 | assem_debug("bicne %s,%s,%s lsl %s\n",regname[rt],regname[rs1],regname[rs2],regname[shift]); |
2144 | output_w32(0x11C00000|rd_rn_rm(rt,rs1,rs2)|0x10|(shift<<8)); |
2145 | } |
2146 | |
e2b5e7aa |
2147 | static void emit_bic_lsr(u_int rs1,u_int rs2,u_int shift,u_int rt) |
57871462 |
2148 | { |
2149 | assem_debug("bic %s,%s,%s lsr %s\n",regname[rt],regname[rs1],regname[rs2],regname[shift]); |
2150 | output_w32(0xe1C00000|rd_rn_rm(rt,rs1,rs2)|0x30|(shift<<8)); |
2151 | } |
2152 | |
e2b5e7aa |
2153 | static void emit_biceq_lsr(u_int rs1,u_int rs2,u_int shift,u_int rt) |
57871462 |
2154 | { |
2155 | assem_debug("biceq %s,%s,%s lsr %s\n",regname[rt],regname[rs1],regname[rs2],regname[shift]); |
2156 | output_w32(0x01C00000|rd_rn_rm(rt,rs1,rs2)|0x30|(shift<<8)); |
2157 | } |
2158 | |
e2b5e7aa |
2159 | static void emit_bicne_lsr(u_int rs1,u_int rs2,u_int shift,u_int rt) |
57871462 |
2160 | { |
2161 | assem_debug("bicne %s,%s,%s lsr %s\n",regname[rt],regname[rs1],regname[rs2],regname[shift]); |
2162 | output_w32(0x11C00000|rd_rn_rm(rt,rs1,rs2)|0x30|(shift<<8)); |
2163 | } |
2164 | |
e2b5e7aa |
2165 | static void emit_teq(int rs, int rt) |
57871462 |
2166 | { |
2167 | assem_debug("teq %s,%s\n",regname[rs],regname[rt]); |
2168 | output_w32(0xe1300000|rd_rn_rm(0,rs,rt)); |
2169 | } |
2170 | |
e2b5e7aa |
2171 | static void emit_rsbimm(int rs, int imm, int rt) |
57871462 |
2172 | { |
2173 | u_int armval; |
cfbd3c6e |
2174 | genimm_checked(imm,&armval); |
57871462 |
2175 | assem_debug("rsb %s,%s,#%d\n",regname[rt],regname[rs],imm); |
2176 | output_w32(0xe2600000|rd_rn_rm(rt,rs,0)|armval); |
2177 | } |
2178 | |
2179 | // Load 2 immediates optimizing for small code size |
e2b5e7aa |
2180 | static void emit_mov2imm_compact(int imm1,u_int rt1,int imm2,u_int rt2) |
57871462 |
2181 | { |
2182 | emit_movimm(imm1,rt1); |
2183 | u_int armval; |
2184 | if(genimm(imm2-imm1,&armval)) { |
2185 | assem_debug("add %s,%s,#%d\n",regname[rt2],regname[rt1],imm2-imm1); |
2186 | output_w32(0xe2800000|rd_rn_rm(rt2,rt1,0)|armval); |
2187 | }else if(genimm(imm1-imm2,&armval)) { |
2188 | assem_debug("sub %s,%s,#%d\n",regname[rt2],regname[rt1],imm1-imm2); |
2189 | output_w32(0xe2400000|rd_rn_rm(rt2,rt1,0)|armval); |
2190 | } |
2191 | else emit_movimm(imm2,rt2); |
2192 | } |
2193 | |
2194 | // Conditionally select one of two immediates, optimizing for small code size |
2195 | // This will only be called if HAVE_CMOV_IMM is defined |
e2b5e7aa |
2196 | static void emit_cmov2imm_e_ne_compact(int imm1,int imm2,u_int rt) |
57871462 |
2197 | { |
2198 | u_int armval; |
2199 | if(genimm(imm2-imm1,&armval)) { |
2200 | emit_movimm(imm1,rt); |
2201 | assem_debug("addne %s,%s,#%d\n",regname[rt],regname[rt],imm2-imm1); |
2202 | output_w32(0x12800000|rd_rn_rm(rt,rt,0)|armval); |
2203 | }else if(genimm(imm1-imm2,&armval)) { |
2204 | emit_movimm(imm1,rt); |
2205 | assem_debug("subne %s,%s,#%d\n",regname[rt],regname[rt],imm1-imm2); |
2206 | output_w32(0x12400000|rd_rn_rm(rt,rt,0)|armval); |
2207 | } |
2208 | else { |
665f33e1 |
2209 | #ifndef HAVE_ARMV7 |
57871462 |
2210 | emit_movimm(imm1,rt); |
2211 | add_literal((int)out,imm2); |
2212 | assem_debug("ldrne %s,pc+? [=%x]\n",regname[rt],imm2); |
2213 | output_w32(0x15900000|rd_rn_rm(rt,15,0)); |
2214 | #else |
2215 | emit_movw(imm1&0x0000FFFF,rt); |
2216 | if((imm1&0xFFFF)!=(imm2&0xFFFF)) { |
2217 | assem_debug("movwne %s,#%d (0x%x)\n",regname[rt],imm2&0xFFFF,imm2&0xFFFF); |
2218 | output_w32(0x13000000|rd_rn_rm(rt,0,0)|(imm2&0xfff)|((imm2<<4)&0xf0000)); |
2219 | } |
2220 | emit_movt(imm1&0xFFFF0000,rt); |
2221 | if((imm1&0xFFFF0000)!=(imm2&0xFFFF0000)) { |
2222 | assem_debug("movtne %s,#%d (0x%x)\n",regname[rt],imm2&0xffff0000,imm2&0xffff0000); |
2223 | output_w32(0x13400000|rd_rn_rm(rt,0,0)|((imm2>>16)&0xfff)|((imm2>>12)&0xf0000)); |
2224 | } |
2225 | #endif |
2226 | } |
2227 | } |
2228 | |
57871462 |
2229 | // special case for checking invalid_code |
e2b5e7aa |
2230 | static void emit_cmpmem_indexedsr12_reg(int base,int r,int imm) |
57871462 |
2231 | { |
2232 | assert(imm<128&&imm>=0); |
2233 | assert(r>=0&&r<16); |
2234 | assem_debug("ldrb lr,%s,%s lsr #12\n",regname[base],regname[r]); |
2235 | output_w32(0xe7d00000|rd_rn_rm(HOST_TEMPREG,base,r)|0x620); |
2236 | emit_cmpimm(HOST_TEMPREG,imm); |
2237 | } |
2238 | |
e2b5e7aa |
2239 | static void emit_callne(int a) |
0bbd1454 |
2240 | { |
2241 | assem_debug("blne %x\n",a); |
2242 | u_int offset=genjmp(a); |
2243 | output_w32(0x1b000000|offset); |
2244 | } |
2245 | |
57871462 |
2246 | // Used to preload hash table entries |
e2b5e7aa |
2247 | static unused void emit_prefetchreg(int r) |
57871462 |
2248 | { |
2249 | assem_debug("pld %s\n",regname[r]); |
2250 | output_w32(0xf5d0f000|rd_rn_rm(0,r,0)); |
2251 | } |
2252 | |
2253 | // Special case for mini_ht |
e2b5e7aa |
2254 | static void emit_ldreq_indexed(int rs, u_int offset, int rt) |
57871462 |
2255 | { |
2256 | assert(offset<4096); |
2257 | assem_debug("ldreq %s,[%s, #%d]\n",regname[rt],regname[rs],offset); |
2258 | output_w32(0x05900000|rd_rn_rm(rt,rs,0)|offset); |
2259 | } |
2260 | |
e2b5e7aa |
2261 | static unused void emit_bicne_imm(int rs,int imm,int rt) |
57871462 |
2262 | { |
2263 | u_int armval; |
cfbd3c6e |
2264 | genimm_checked(imm,&armval); |
57871462 |
2265 | assem_debug("bicne %s,%s,#%d\n",regname[rt],regname[rs],imm); |
2266 | output_w32(0x13c00000|rd_rn_rm(rt,rs,0)|armval); |
2267 | } |
2268 | |
e2b5e7aa |
2269 | static unused void emit_biccs_imm(int rs,int imm,int rt) |
57871462 |
2270 | { |
2271 | u_int armval; |
cfbd3c6e |
2272 | genimm_checked(imm,&armval); |
57871462 |
2273 | assem_debug("biccs %s,%s,#%d\n",regname[rt],regname[rs],imm); |
2274 | output_w32(0x23c00000|rd_rn_rm(rt,rs,0)|armval); |
2275 | } |
2276 | |
e2b5e7aa |
2277 | static unused void emit_bicvc_imm(int rs,int imm,int rt) |
57871462 |
2278 | { |
2279 | u_int armval; |
cfbd3c6e |
2280 | genimm_checked(imm,&armval); |
57871462 |
2281 | assem_debug("bicvc %s,%s,#%d\n",regname[rt],regname[rs],imm); |
2282 | output_w32(0x73c00000|rd_rn_rm(rt,rs,0)|armval); |
2283 | } |
2284 | |
e2b5e7aa |
2285 | static unused void emit_bichi_imm(int rs,int imm,int rt) |
57871462 |
2286 | { |
2287 | u_int armval; |
cfbd3c6e |
2288 | genimm_checked(imm,&armval); |
57871462 |
2289 | assem_debug("bichi %s,%s,#%d\n",regname[rt],regname[rs],imm); |
2290 | output_w32(0x83c00000|rd_rn_rm(rt,rs,0)|armval); |
2291 | } |
2292 | |
e2b5e7aa |
2293 | static unused void emit_orrvs_imm(int rs,int imm,int rt) |
57871462 |
2294 | { |
2295 | u_int armval; |
cfbd3c6e |
2296 | genimm_checked(imm,&armval); |
57871462 |
2297 | assem_debug("orrvs %s,%s,#%d\n",regname[rt],regname[rs],imm); |
2298 | output_w32(0x63800000|rd_rn_rm(rt,rs,0)|armval); |
2299 | } |
2300 | |
e2b5e7aa |
2301 | static void emit_orrne_imm(int rs,int imm,int rt) |
b9b61529 |
2302 | { |
2303 | u_int armval; |
cfbd3c6e |
2304 | genimm_checked(imm,&armval); |
b9b61529 |
2305 | assem_debug("orrne %s,%s,#%d\n",regname[rt],regname[rs],imm); |
2306 | output_w32(0x13800000|rd_rn_rm(rt,rs,0)|armval); |
2307 | } |
2308 | |
e2b5e7aa |
2309 | static void emit_andne_imm(int rs,int imm,int rt) |
b9b61529 |
2310 | { |
2311 | u_int armval; |
cfbd3c6e |
2312 | genimm_checked(imm,&armval); |
b9b61529 |
2313 | assem_debug("andne %s,%s,#%d\n",regname[rt],regname[rs],imm); |
2314 | output_w32(0x12000000|rd_rn_rm(rt,rs,0)|armval); |
2315 | } |
2316 | |
e2b5e7aa |
2317 | static unused void emit_addpl_imm(int rs,int imm,int rt) |
665f33e1 |
2318 | { |
2319 | u_int armval; |
2320 | genimm_checked(imm,&armval); |
2321 | assem_debug("addpl %s,%s,#%d\n",regname[rt],regname[rs],imm); |
2322 | output_w32(0x52800000|rd_rn_rm(rt,rs,0)|armval); |
2323 | } |
2324 | |
e2b5e7aa |
2325 | static void emit_jno_unlikely(int a) |
57871462 |
2326 | { |
2327 | //emit_jno(a); |
2328 | assem_debug("addvc pc,pc,#? (%x)\n",/*a-(int)out-8,*/a); |
2329 | output_w32(0x72800000|rd_rn_rm(15,15,0)); |
2330 | } |
2331 | |
054175e9 |
2332 | static void save_regs_all(u_int reglist) |
57871462 |
2333 | { |
054175e9 |
2334 | int i; |
57871462 |
2335 | if(!reglist) return; |
2336 | assem_debug("stmia fp,{"); |
054175e9 |
2337 | for(i=0;i<16;i++) |
2338 | if(reglist&(1<<i)) |
2339 | assem_debug("r%d,",i); |
57871462 |
2340 | assem_debug("}\n"); |
2341 | output_w32(0xe88b0000|reglist); |
2342 | } |
e2b5e7aa |
2343 | |
054175e9 |
2344 | static void restore_regs_all(u_int reglist) |
57871462 |
2345 | { |
054175e9 |
2346 | int i; |
57871462 |
2347 | if(!reglist) return; |
2348 | assem_debug("ldmia fp,{"); |
054175e9 |
2349 | for(i=0;i<16;i++) |
2350 | if(reglist&(1<<i)) |
2351 | assem_debug("r%d,",i); |
57871462 |
2352 | assem_debug("}\n"); |
2353 | output_w32(0xe89b0000|reglist); |
2354 | } |
e2b5e7aa |
2355 | |
054175e9 |
2356 | // Save registers before function call |
2357 | static void save_regs(u_int reglist) |
2358 | { |
4d646738 |
2359 | reglist&=CALLER_SAVE_REGS; // only save the caller-save registers, r0-r3, r12 |
054175e9 |
2360 | save_regs_all(reglist); |
2361 | } |
e2b5e7aa |
2362 | |
054175e9 |
2363 | // Restore registers after function call |
2364 | static void restore_regs(u_int reglist) |
2365 | { |
4d646738 |
2366 | reglist&=CALLER_SAVE_REGS; |
054175e9 |
2367 | restore_regs_all(reglist); |
2368 | } |
57871462 |
2369 | |
57871462 |
2370 | /* Stubs/epilogue */ |
2371 | |
e2b5e7aa |
2372 | static void literal_pool(int n) |
57871462 |
2373 | { |
2374 | if(!literalcount) return; |
2375 | if(n) { |
2376 | if((int)out-literals[0][0]<4096-n) return; |
2377 | } |
2378 | u_int *ptr; |
2379 | int i; |
2380 | for(i=0;i<literalcount;i++) |
2381 | { |
77750690 |
2382 | u_int l_addr=(u_int)out; |
2383 | int j; |
2384 | for(j=0;j<i;j++) { |
2385 | if(literals[j][1]==literals[i][1]) { |
2386 | //printf("dup %08x\n",literals[i][1]); |
2387 | l_addr=literals[j][0]; |
2388 | break; |
2389 | } |
2390 | } |
57871462 |
2391 | ptr=(u_int *)literals[i][0]; |
77750690 |
2392 | u_int offset=l_addr-(u_int)ptr-8; |
57871462 |
2393 | assert(offset<4096); |
2394 | assert(!(offset&3)); |
2395 | *ptr|=offset; |
77750690 |
2396 | if(l_addr==(u_int)out) { |
2397 | literals[i][0]=l_addr; // remember for dupes |
2398 | output_w32(literals[i][1]); |
2399 | } |
57871462 |
2400 | } |
2401 | literalcount=0; |
2402 | } |
2403 | |
e2b5e7aa |
2404 | static void literal_pool_jumpover(int n) |
57871462 |
2405 | { |
2406 | if(!literalcount) return; |
2407 | if(n) { |
2408 | if((int)out-literals[0][0]<4096-n) return; |
2409 | } |
2410 | int jaddr=(int)out; |
2411 | emit_jmp(0); |
2412 | literal_pool(0); |
2413 | set_jump_target(jaddr,(int)out); |
2414 | } |
2415 | |
e2b5e7aa |
2416 | static void emit_extjump2(u_int addr, int target, int linker) |
57871462 |
2417 | { |
2418 | u_char *ptr=(u_char *)addr; |
2419 | assert((ptr[3]&0x0e)==0xa); |
e2b5e7aa |
2420 | (void)ptr; |
2421 | |
57871462 |
2422 | emit_loadlp(target,0); |
2423 | emit_loadlp(addr,1); |
24385cae |
2424 | assert(addr>=BASE_ADDR&&addr<(BASE_ADDR+(1<<TARGET_SIZE_2))); |
57871462 |
2425 | //assert((target>=0x80000000&&target<0x80800000)||(target>0xA4000000&&target<0xA4001000)); |
2426 | //DEBUG > |
2427 | #ifdef DEBUG_CYCLE_COUNT |
2428 | emit_readword((int)&last_count,ECX); |
2429 | emit_add(HOST_CCREG,ECX,HOST_CCREG); |
2430 | emit_readword((int)&next_interupt,ECX); |
2431 | emit_writeword(HOST_CCREG,(int)&Count); |
2432 | emit_sub(HOST_CCREG,ECX,HOST_CCREG); |
2433 | emit_writeword(ECX,(int)&last_count); |
2434 | #endif |
2435 | //DEBUG < |
2436 | emit_jmp(linker); |
2437 | } |
2438 | |
e2b5e7aa |
2439 | static void emit_extjump(int addr, int target) |
57871462 |
2440 | { |
2441 | emit_extjump2(addr, target, (int)dyna_linker); |
2442 | } |
e2b5e7aa |
2443 | |
2444 | static void emit_extjump_ds(int addr, int target) |
57871462 |
2445 | { |
2446 | emit_extjump2(addr, target, (int)dyna_linker_ds); |
2447 | } |
2448 | |
13e35c04 |
2449 | // put rt_val into rt, potentially making use of rs with value rs_val |
2450 | static void emit_movimm_from(u_int rs_val,int rs,u_int rt_val,int rt) |
2451 | { |
8575a877 |
2452 | u_int armval; |
2453 | int diff; |
2454 | if(genimm(rt_val,&armval)) { |
2455 | assem_debug("mov %s,#%d\n",regname[rt],rt_val); |
2456 | output_w32(0xe3a00000|rd_rn_rm(rt,0,0)|armval); |
2457 | return; |
2458 | } |
2459 | if(genimm(~rt_val,&armval)) { |
2460 | assem_debug("mvn %s,#%d\n",regname[rt],rt_val); |
2461 | output_w32(0xe3e00000|rd_rn_rm(rt,0,0)|armval); |
2462 | return; |
2463 | } |
2464 | diff=rt_val-rs_val; |
2465 | if(genimm(diff,&armval)) { |
2466 | assem_debug("add %s,%s,#%d\n",regname[rt],regname[rs],diff); |
2467 | output_w32(0xe2800000|rd_rn_rm(rt,rs,0)|armval); |
2468 | return; |
2469 | }else if(genimm(-diff,&armval)) { |
2470 | assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rs],-diff); |
2471 | output_w32(0xe2400000|rd_rn_rm(rt,rs,0)|armval); |
2472 | return; |
2473 | } |
2474 | emit_movimm(rt_val,rt); |
2475 | } |
2476 | |
2477 | // return 1 if above function can do it's job cheaply |
2478 | static int is_similar_value(u_int v1,u_int v2) |
2479 | { |
13e35c04 |
2480 | u_int xs; |
8575a877 |
2481 | int diff; |
2482 | if(v1==v2) return 1; |
2483 | diff=v2-v1; |
2484 | for(xs=diff;xs!=0&&(xs&3)==0;xs>>=2) |
13e35c04 |
2485 | ; |
8575a877 |
2486 | if(xs<0x100) return 1; |
2487 | for(xs=-diff;xs!=0&&(xs&3)==0;xs>>=2) |
2488 | ; |
2489 | if(xs<0x100) return 1; |
2490 | return 0; |
13e35c04 |
2491 | } |
cbbab9cd |
2492 | |
b96d3df7 |
2493 | // trashes r2 |
2494 | static void pass_args(int a0, int a1) |
2495 | { |
2496 | if(a0==1&&a1==0) { |
2497 | // must swap |
2498 | emit_mov(a0,2); emit_mov(a1,1); emit_mov(2,0); |
2499 | } |
2500 | else if(a0!=0&&a1==0) { |
2501 | emit_mov(a1,1); |
2502 | if (a0>=0) emit_mov(a0,0); |
2503 | } |
2504 | else { |
2505 | if(a0>=0&&a0!=0) emit_mov(a0,0); |
2506 | if(a1>=0&&a1!=1) emit_mov(a1,1); |
2507 | } |
2508 | } |
2509 | |
b1be1eee |
2510 | static void mov_loadtype_adj(int type,int rs,int rt) |
2511 | { |
2512 | switch(type) { |
2513 | case LOADB_STUB: emit_signextend8(rs,rt); break; |
2514 | case LOADBU_STUB: emit_andimm(rs,0xff,rt); break; |
2515 | case LOADH_STUB: emit_signextend16(rs,rt); break; |
2516 | case LOADHU_STUB: emit_andimm(rs,0xffff,rt); break; |
2517 | case LOADW_STUB: if(rs!=rt) emit_mov(rs,rt); break; |
2518 | default: assert(0); |
2519 | } |
2520 | } |
2521 | |
b1be1eee |
2522 | #include "pcsxmem.h" |
2523 | #include "pcsxmem_inline.c" |
b1be1eee |
2524 | |
e2b5e7aa |
2525 | static void do_readstub(int n) |
57871462 |
2526 | { |
2527 | assem_debug("do_readstub %x\n",start+stubs[n][3]*4); |
2528 | literal_pool(256); |
2529 | set_jump_target(stubs[n][1],(int)out); |
2530 | int type=stubs[n][0]; |
2531 | int i=stubs[n][3]; |
2532 | int rs=stubs[n][4]; |
2533 | struct regstat *i_regs=(struct regstat *)stubs[n][5]; |
2534 | u_int reglist=stubs[n][7]; |
2535 | signed char *i_regmap=i_regs->regmap; |
581335b0 |
2536 | int rt; |
b9b61529 |
2537 | if(itype[i]==C1LS||itype[i]==C2LS||itype[i]==LOADLR) { |
57871462 |
2538 | rt=get_reg(i_regmap,FTEMP); |
2539 | }else{ |
57871462 |
2540 | rt=get_reg(i_regmap,rt1[i]); |
2541 | } |
2542 | assert(rs>=0); |
c6c3b1b3 |
2543 | int r,temp=-1,temp2=HOST_TEMPREG,regs_saved=0,restore_jump=0; |
2544 | reglist|=(1<<rs); |
2545 | for(r=0;r<=12;r++) { |
2546 | if(((1<<r)&0x13ff)&&((1<<r)®list)==0) { |
2547 | temp=r; break; |
2548 | } |
2549 | } |
db829eeb |
2550 | if(rt>=0&&rt1[i]!=0) |
c6c3b1b3 |
2551 | reglist&=~(1<<rt); |
2552 | if(temp==-1) { |
2553 | save_regs(reglist); |
2554 | regs_saved=1; |
2555 | temp=(rs==0)?2:0; |
2556 | } |
2557 | if((regs_saved||(reglist&2)==0)&&temp!=1&&rs!=1) |
2558 | temp2=1; |
2559 | emit_readword((int)&mem_rtab,temp); |
2560 | emit_shrimm(rs,12,temp2); |
2561 | emit_readword_dualindexedx4(temp,temp2,temp2); |
2562 | emit_lsls_imm(temp2,1,temp2); |
2563 | if(itype[i]==C1LS||itype[i]==C2LS||(rt>=0&&rt1[i]!=0)) { |
2564 | switch(type) { |
2565 | case LOADB_STUB: emit_ldrccsb_dualindexed(temp2,rs,rt); break; |
2566 | case LOADBU_STUB: emit_ldrccb_dualindexed(temp2,rs,rt); break; |
2567 | case LOADH_STUB: emit_ldrccsh_dualindexed(temp2,rs,rt); break; |
2568 | case LOADHU_STUB: emit_ldrcch_dualindexed(temp2,rs,rt); break; |
2569 | case LOADW_STUB: emit_ldrcc_dualindexed(temp2,rs,rt); break; |
2570 | } |
2571 | } |
2572 | if(regs_saved) { |
2573 | restore_jump=(int)out; |
2574 | emit_jcc(0); // jump to reg restore |
2575 | } |
2576 | else |
2577 | emit_jcc(stubs[n][2]); // return address |
2578 | |
2579 | if(!regs_saved) |
2580 | save_regs(reglist); |
2581 | int handler=0; |
2582 | if(type==LOADB_STUB||type==LOADBU_STUB) |
2583 | handler=(int)jump_handler_read8; |
2584 | if(type==LOADH_STUB||type==LOADHU_STUB) |
2585 | handler=(int)jump_handler_read16; |
2586 | if(type==LOADW_STUB) |
2587 | handler=(int)jump_handler_read32; |
2588 | assert(handler!=0); |
b96d3df7 |
2589 | pass_args(rs,temp2); |
c6c3b1b3 |
2590 | int cc=get_reg(i_regmap,CCREG); |
2591 | if(cc<0) |
2592 | emit_loadreg(CCREG,2); |
2573466a |
2593 | emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n][6]+1),2); |
c6c3b1b3 |
2594 | emit_call(handler); |
2595 | if(itype[i]==C1LS||itype[i]==C2LS||(rt>=0&&rt1[i]!=0)) { |
b1be1eee |
2596 | mov_loadtype_adj(type,0,rt); |
c6c3b1b3 |
2597 | } |
2598 | if(restore_jump) |
2599 | set_jump_target(restore_jump,(int)out); |
2600 | restore_regs(reglist); |
2601 | emit_jmp(stubs[n][2]); // return address |
57871462 |
2602 | } |
2603 | |
c6c3b1b3 |
2604 | // return memhandler, or get directly accessable address and return 0 |
e2b5e7aa |
2605 | static u_int get_direct_memhandler(void *table,u_int addr,int type,u_int *addr_host) |
c6c3b1b3 |
2606 | { |
2607 | u_int l1,l2=0; |
2608 | l1=((u_int *)table)[addr>>12]; |
2609 | if((l1&(1<<31))==0) { |
2610 | u_int v=l1<<1; |
2611 | *addr_host=v+addr; |
2612 | return 0; |
2613 | } |
2614 | else { |
2615 | l1<<=1; |
2616 | if(type==LOADB_STUB||type==LOADBU_STUB||type==STOREB_STUB) |
2617 | l2=((u_int *)l1)[0x1000/4 + 0x1000/2 + (addr&0xfff)]; |
b96d3df7 |
2618 | else if(type==LOADH_STUB||type==LOADHU_STUB||type==STOREH_STUB) |
c6c3b1b3 |
2619 | l2=((u_int *)l1)[0x1000/4 + (addr&0xfff)/2]; |
2620 | else |
2621 | l2=((u_int *)l1)[(addr&0xfff)/4]; |
2622 | if((l2&(1<<31))==0) { |
2623 | u_int v=l2<<1; |
2624 | *addr_host=v+(addr&0xfff); |
2625 | return 0; |
2626 | } |
2627 | return l2<<1; |
2628 | } |
2629 | } |
c6c3b1b3 |
2630 | |
e2b5e7aa |
2631 | static void inline_readstub(int type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist) |
57871462 |
2632 | { |
2633 | int rs=get_reg(regmap,target); |
57871462 |
2634 | int rt=get_reg(regmap,target); |
535d208a |
2635 | if(rs<0) rs=get_reg(regmap,-1); |
57871462 |
2636 | assert(rs>=0); |
b1be1eee |
2637 | u_int handler,host_addr=0,is_dynamic,far_call=0; |
2638 | int cc=get_reg(regmap,CCREG); |
2639 | if(pcsx_direct_read(type,addr,CLOCK_ADJUST(adj+1),cc,target?rs:-1,rt)) |
2640 | return; |
c6c3b1b3 |
2641 | handler=get_direct_memhandler(mem_rtab,addr,type,&host_addr); |
2642 | if (handler==0) { |
db829eeb |
2643 | if(rt<0||rt1[i]==0) |
c6c3b1b3 |
2644 | return; |
13e35c04 |
2645 | if(addr!=host_addr) |
2646 | emit_movimm_from(addr,rs,host_addr,rs); |
c6c3b1b3 |
2647 | switch(type) { |
2648 | case LOADB_STUB: emit_movsbl_indexed(0,rs,rt); break; |
2649 | case LOADBU_STUB: emit_movzbl_indexed(0,rs,rt); break; |
2650 | case LOADH_STUB: emit_movswl_indexed(0,rs,rt); break; |
2651 | case LOADHU_STUB: emit_movzwl_indexed(0,rs,rt); break; |
2652 | case LOADW_STUB: emit_readword_indexed(0,rs,rt); break; |
2653 | default: assert(0); |
2654 | } |
2655 | return; |
2656 | } |
b1be1eee |
2657 | is_dynamic=pcsxmem_is_handler_dynamic(addr); |
2658 | if(is_dynamic) { |
2659 | if(type==LOADB_STUB||type==LOADBU_STUB) |
2660 | handler=(int)jump_handler_read8; |
2661 | if(type==LOADH_STUB||type==LOADHU_STUB) |
2662 | handler=(int)jump_handler_read16; |
2663 | if(type==LOADW_STUB) |
2664 | handler=(int)jump_handler_read32; |
2665 | } |
c6c3b1b3 |
2666 | |
2667 | // call a memhandler |
db829eeb |
2668 | if(rt>=0&&rt1[i]!=0) |
c6c3b1b3 |
2669 | reglist&=~(1<<rt); |
2670 | save_regs(reglist); |
2671 | if(target==0) |
2672 | emit_movimm(addr,0); |
2673 | else if(rs!=0) |
2674 | emit_mov(rs,0); |
c6c3b1b3 |
2675 | int offset=(int)handler-(int)out-8; |
2676 | if(offset<-33554432||offset>=33554432) { |
2677 | // unreachable memhandler, a plugin func perhaps |
b1be1eee |
2678 | emit_movimm(handler,12); |
2679 | far_call=1; |
2680 | } |
2681 | if(cc<0) |
2682 | emit_loadreg(CCREG,2); |
2683 | if(is_dynamic) { |
2684 | emit_movimm(((u_int *)mem_rtab)[addr>>12]<<1,1); |
2685 | emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj+1),2); |
c6c3b1b3 |
2686 | } |
b1be1eee |
2687 | else { |
2688 | emit_readword((int)&last_count,3); |
2689 | emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj+1),2); |
2690 | emit_add(2,3,2); |
2691 | emit_writeword(2,(int)&Count); |
2692 | } |
2693 | |
2694 | if(far_call) |
2695 | emit_callreg(12); |
c6c3b1b3 |
2696 | else |
2697 | emit_call(handler); |
b1be1eee |
2698 | |
db829eeb |
2699 | if(rt>=0&&rt1[i]!=0) { |
c6c3b1b3 |
2700 | switch(type) { |
2701 | case LOADB_STUB: emit_signextend8(0,rt); break; |
2702 | case LOADBU_STUB: emit_andimm(0,0xff,rt); break; |
2703 | case LOADH_STUB: emit_signextend16(0,rt); break; |
2704 | case LOADHU_STUB: emit_andimm(0,0xffff,rt); break; |
2705 | case LOADW_STUB: if(rt!=0) emit_mov(0,rt); break; |
2706 | default: assert(0); |
2707 | } |
2708 | } |
2709 | restore_regs(reglist); |
57871462 |
2710 | } |
2711 | |
e2b5e7aa |
2712 | static void do_writestub(int n) |
57871462 |
2713 | { |
2714 | assem_debug("do_writestub %x\n",start+stubs[n][3]*4); |
2715 | literal_pool(256); |
2716 | set_jump_target(stubs[n][1],(int)out); |
2717 | int type=stubs[n][0]; |
2718 | int i=stubs[n][3]; |
2719 | int rs=stubs[n][4]; |
2720 | struct regstat *i_regs=(struct regstat *)stubs[n][5]; |
2721 | u_int reglist=stubs[n][7]; |
2722 | signed char *i_regmap=i_regs->regmap; |
581335b0 |
2723 | int rt,r; |
b9b61529 |
2724 | if(itype[i]==C1LS||itype[i]==C2LS) { |
57871462 |
2725 | rt=get_reg(i_regmap,r=FTEMP); |
2726 | }else{ |
57871462 |
2727 | rt=get_reg(i_regmap,r=rs2[i]); |
2728 | } |
2729 | assert(rs>=0); |
2730 | assert(rt>=0); |
b96d3df7 |
2731 | int rtmp,temp=-1,temp2=HOST_TEMPREG,regs_saved=0,restore_jump=0,ra; |
2732 | int reglist2=reglist|(1<<rs)|(1<<rt); |
2733 | for(rtmp=0;rtmp<=12;rtmp++) { |
2734 | if(((1<<rtmp)&0x13ff)&&((1<<rtmp)®list2)==0) { |
2735 | temp=rtmp; break; |
2736 | } |
2737 | } |
2738 | if(temp==-1) { |
2739 | save_regs(reglist); |
2740 | regs_saved=1; |
2741 | for(rtmp=0;rtmp<=3;rtmp++) |
2742 | if(rtmp!=rs&&rtmp!=rt) |
2743 | {temp=rtmp;break;} |
2744 | } |
2745 | if((regs_saved||(reglist2&8)==0)&&temp!=3&&rs!=3&&rt!=3) |
2746 | temp2=3; |
2747 | emit_readword((int)&mem_wtab,temp); |
2748 | emit_shrimm(rs,12,temp2); |
2749 | emit_readword_dualindexedx4(temp,temp2,temp2); |
2750 | emit_lsls_imm(temp2,1,temp2); |
2751 | switch(type) { |
2752 | case STOREB_STUB: emit_strccb_dualindexed(temp2,rs,rt); break; |
2753 | case STOREH_STUB: emit_strcch_dualindexed(temp2,rs,rt); break; |
2754 | case STOREW_STUB: emit_strcc_dualindexed(temp2,rs,rt); break; |
2755 | default: assert(0); |
2756 | } |
2757 | if(regs_saved) { |
2758 | restore_jump=(int)out; |
2759 | emit_jcc(0); // jump to reg restore |
2760 | } |
2761 | else |
2762 | emit_jcc(stubs[n][2]); // return address (invcode check) |
2763 | |
2764 | if(!regs_saved) |
2765 | save_regs(reglist); |
2766 | int handler=0; |
2767 | switch(type) { |
2768 | case STOREB_STUB: handler=(int)jump_handler_write8; break; |
2769 | case STOREH_STUB: handler=(int)jump_handler_write16; break; |
2770 | case STOREW_STUB: handler=(int)jump_handler_write32; break; |
2771 | } |
2772 | assert(handler!=0); |
2773 | pass_args(rs,rt); |
2774 | if(temp2!=3) |
2775 | emit_mov(temp2,3); |
2776 | int cc=get_reg(i_regmap,CCREG); |
2777 | if(cc<0) |
2778 | emit_loadreg(CCREG,2); |
2573466a |
2779 | emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n][6]+1),2); |
b96d3df7 |
2780 | // returns new cycle_count |
2781 | emit_call(handler); |
2573466a |
2782 | emit_addimm(0,-CLOCK_ADJUST((int)stubs[n][6]+1),cc<0?2:cc); |
b96d3df7 |
2783 | if(cc<0) |
2784 | emit_storereg(CCREG,2); |
2785 | if(restore_jump) |
2786 | set_jump_target(restore_jump,(int)out); |
2787 | restore_regs(reglist); |
2788 | ra=stubs[n][2]; |
b96d3df7 |
2789 | emit_jmp(ra); |
57871462 |
2790 | } |
2791 | |
e2b5e7aa |
2792 | static void inline_writestub(int type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist) |
57871462 |
2793 | { |
2794 | int rs=get_reg(regmap,-1); |
57871462 |
2795 | int rt=get_reg(regmap,target); |
2796 | assert(rs>=0); |
2797 | assert(rt>=0); |
b96d3df7 |
2798 | u_int handler,host_addr=0; |
b96d3df7 |
2799 | handler=get_direct_memhandler(mem_wtab,addr,type,&host_addr); |
2800 | if (handler==0) { |
13e35c04 |
2801 | if(addr!=host_addr) |
2802 | emit_movimm_from(addr,rs,host_addr,rs); |
b96d3df7 |
2803 | switch(type) { |
2804 | case STOREB_STUB: emit_writebyte_indexed(rt,0,rs); break; |
2805 | case STOREH_STUB: emit_writehword_indexed(rt,0,rs); break; |
2806 | case STOREW_STUB: emit_writeword_indexed(rt,0,rs); break; |
2807 | default: assert(0); |
2808 | } |
2809 | return; |
2810 | } |
2811 | |
2812 | // call a memhandler |
2813 | save_regs(reglist); |
13e35c04 |
2814 | pass_args(rs,rt); |
b96d3df7 |
2815 | int cc=get_reg(regmap,CCREG); |
2816 | if(cc<0) |
2817 | emit_loadreg(CCREG,2); |
2573466a |
2818 | emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj+1),2); |
b96d3df7 |
2819 | emit_movimm(handler,3); |
2820 | // returns new cycle_count |
2821 | emit_call((int)jump_handler_write_h); |
2573466a |
2822 | emit_addimm(0,-CLOCK_ADJUST(adj+1),cc<0?2:cc); |
b96d3df7 |
2823 | if(cc<0) |
2824 | emit_storereg(CCREG,2); |
2825 | restore_regs(reglist); |
57871462 |
2826 | } |
2827 | |
e2b5e7aa |
2828 | static void do_unalignedwritestub(int n) |
57871462 |
2829 | { |
b7918751 |
2830 | assem_debug("do_unalignedwritestub %x\n",start+stubs[n][3]*4); |
2831 | literal_pool(256); |
57871462 |
2832 | set_jump_target(stubs[n][1],(int)out); |
b7918751 |
2833 | |
2834 | int i=stubs[n][3]; |
2835 | struct regstat *i_regs=(struct regstat *)stubs[n][4]; |
2836 | int addr=stubs[n][5]; |
2837 | u_int reglist=stubs[n][7]; |
2838 | signed char *i_regmap=i_regs->regmap; |
2839 | int temp2=get_reg(i_regmap,FTEMP); |
2840 | int rt; |
b7918751 |
2841 | rt=get_reg(i_regmap,rs2[i]); |
2842 | assert(rt>=0); |
2843 | assert(addr>=0); |
2844 | assert(opcode[i]==0x2a||opcode[i]==0x2e); // SWL/SWR only implemented |
2845 | reglist|=(1<<addr); |
2846 | reglist&=~(1<<temp2); |
2847 | |
b96d3df7 |
2848 | #if 1 |
2849 | // don't bother with it and call write handler |
2850 | save_regs(reglist); |
2851 | pass_args(addr,rt); |
2852 | int cc=get_reg(i_regmap,CCREG); |
2853 | if(cc<0) |
2854 | emit_loadreg(CCREG,2); |
2573466a |
2855 | emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n][6]+1),2); |
b96d3df7 |
2856 | emit_call((int)(opcode[i]==0x2a?jump_handle_swl:jump_handle_swr)); |
2573466a |
2857 | emit_addimm(0,-CLOCK_ADJUST((int)stubs[n][6]+1),cc<0?2:cc); |
b96d3df7 |
2858 | if(cc<0) |
2859 | emit_storereg(CCREG,2); |
2860 | restore_regs(reglist); |
2861 | emit_jmp(stubs[n][2]); // return address |
2862 | #else |
b7918751 |
2863 | emit_andimm(addr,0xfffffffc,temp2); |
2864 | emit_writeword(temp2,(int)&address); |
2865 | |
2866 | save_regs(reglist); |
b7918751 |
2867 | emit_shrimm(addr,16,1); |
2868 | int cc=get_reg(i_regmap,CCREG); |
2869 | if(cc<0) { |
2870 | emit_loadreg(CCREG,2); |
2871 | } |
2872 | emit_movimm((u_int)readmem,0); |
2873 | emit_addimm(cc<0?2:cc,2*stubs[n][6]+2,2); |
b7918751 |
2874 | emit_call((int)&indirect_jump_indexed); |
2875 | restore_regs(reglist); |
2876 | |
2877 | emit_readword((int)&readmem_dword,temp2); |
2878 | int temp=addr; //hmh |
2879 | emit_shlimm(addr,3,temp); |
2880 | emit_andimm(temp,24,temp); |
2881 | #ifdef BIG_ENDIAN_MIPS |
2882 | if (opcode[i]==0x2e) // SWR |
2883 | #else |
2884 | if (opcode[i]==0x2a) // SWL |
2885 | #endif |
2886 | emit_xorimm(temp,24,temp); |
2887 | emit_movimm(-1,HOST_TEMPREG); |
55439448 |
2888 | if (opcode[i]==0x2a) { // SWL |
b7918751 |
2889 | emit_bic_lsr(temp2,HOST_TEMPREG,temp,temp2); |
2890 | emit_orrshr(rt,temp,temp2); |
2891 | }else{ |
2892 | emit_bic_lsl(temp2,HOST_TEMPREG,temp,temp2); |
2893 | emit_orrshl(rt,temp,temp2); |
2894 | } |
2895 | emit_readword((int)&address,addr); |
2896 | emit_writeword(temp2,(int)&word); |
2897 | //save_regs(reglist); // don't need to, no state changes |
2898 | emit_shrimm(addr,16,1); |
2899 | emit_movimm((u_int)writemem,0); |
2900 | //emit_call((int)&indirect_jump_indexed); |
2901 | emit_mov(15,14); |
2902 | emit_readword_dualindexedx4(0,1,15); |
2903 | emit_readword((int)&Count,HOST_TEMPREG); |
2904 | emit_readword((int)&next_interupt,2); |
2905 | emit_addimm(HOST_TEMPREG,-2*stubs[n][6]-2,HOST_TEMPREG); |
2906 | emit_writeword(2,(int)&last_count); |
2907 | emit_sub(HOST_TEMPREG,2,cc<0?HOST_TEMPREG:cc); |
2908 | if(cc<0) { |
2909 | emit_storereg(CCREG,HOST_TEMPREG); |
2910 | } |
2911 | restore_regs(reglist); |
57871462 |
2912 | emit_jmp(stubs[n][2]); // return address |
b96d3df7 |
2913 | #endif |
57871462 |
2914 | } |
2915 | |
e2b5e7aa |
2916 | static void do_invstub(int n) |
57871462 |
2917 | { |
2918 | literal_pool(20); |
2919 | u_int reglist=stubs[n][3]; |
2920 | set_jump_target(stubs[n][1],(int)out); |
2921 | save_regs(reglist); |
2922 | if(stubs[n][4]!=0) emit_mov(stubs[n][4],0); |
2923 | emit_call((int)&invalidate_addr); |
2924 | restore_regs(reglist); |
2925 | emit_jmp(stubs[n][2]); // return address |
2926 | } |
2927 | |
2928 | int do_dirty_stub(int i) |
2929 | { |
2930 | assem_debug("do_dirty_stub %x\n",start+i*4); |
71e490c5 |
2931 | u_int addr=(u_int)source; |
57871462 |
2932 | // Careful about the code output here, verify_dirty needs to parse it. |
665f33e1 |
2933 | #ifndef HAVE_ARMV7 |
ac545b3a |
2934 | emit_loadlp(addr,1); |
57871462 |
2935 | emit_loadlp((int)copy,2); |
2936 | emit_loadlp(slen*4,3); |
2937 | #else |
ac545b3a |
2938 | emit_movw(addr&0x0000FFFF,1); |
57871462 |
2939 | emit_movw(((u_int)copy)&0x0000FFFF,2); |
ac545b3a |
2940 | emit_movt(addr&0xFFFF0000,1); |
57871462 |
2941 | emit_movt(((u_int)copy)&0xFFFF0000,2); |
2942 | emit_movw(slen*4,3); |
2943 | #endif |
2944 | emit_movimm(start+i*4,0); |
2945 | emit_call((int)start<(int)0xC0000000?(int)&verify_code:(int)&verify_code_vm); |
2946 | int entry=(int)out; |
2947 | load_regs_entry(i); |
2948 | if(entry==(int)out) entry=instr_addr[i]; |
2949 | emit_jmp(instr_addr[i]); |
2950 | return entry; |
2951 | } |
2952 | |
e2b5e7aa |
2953 | static void do_dirty_stub_ds() |
57871462 |
2954 | { |
2955 | // Careful about the code output here, verify_dirty needs to parse it. |
665f33e1 |
2956 | #ifndef HAVE_ARMV7 |
57871462 |
2957 | emit_loadlp((int)start<(int)0xC0000000?(int)source:(int)start,1); |
2958 | emit_loadlp((int)copy,2); |
2959 | emit_loadlp(slen*4,3); |
2960 | #else |
2961 | emit_movw(((int)start<(int)0xC0000000?(u_int)source:(u_int)start)&0x0000FFFF,1); |
2962 | emit_movw(((u_int)copy)&0x0000FFFF,2); |
2963 | emit_movt(((int)start<(int)0xC0000000?(u_int)source:(u_int)start)&0xFFFF0000,1); |
2964 | emit_movt(((u_int)copy)&0xFFFF0000,2); |
2965 | emit_movw(slen*4,3); |
2966 | #endif |
2967 | emit_movimm(start+1,0); |
2968 | emit_call((int)&verify_code_ds); |
2969 | } |
2970 | |
e2b5e7aa |
2971 | static void do_cop1stub(int n) |
57871462 |
2972 | { |
2973 | literal_pool(256); |
2974 | assem_debug("do_cop1stub %x\n",start+stubs[n][3]*4); |
2975 | set_jump_target(stubs[n][1],(int)out); |
2976 | int i=stubs[n][3]; |
3d624f89 |
2977 | // int rs=stubs[n][4]; |
57871462 |
2978 | struct regstat *i_regs=(struct regstat *)stubs[n][5]; |
2979 | int ds=stubs[n][6]; |
2980 | if(!ds) { |
2981 | load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty,i); |
2982 | //if(i_regs!=®s[i]) printf("oops: regs[i]=%x i_regs=%x",(int)®s[i],(int)i_regs); |
2983 | } |
2984 | //else {printf("fp exception in delay slot\n");} |
2985 | wb_dirtys(i_regs->regmap_entry,i_regs->was32,i_regs->wasdirty); |
2986 | if(regs[i].regmap_entry[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG); |
2987 | emit_movimm(start+(i-ds)*4,EAX); // Get PC |
2573466a |
2988 | emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG); // CHECK: is this right? There should probably be an extra cycle... |
57871462 |
2989 | emit_jmp(ds?(int)fp_exception_ds:(int)fp_exception); |
2990 | } |
2991 | |
57871462 |
2992 | /* Special assem */ |
2993 | |
e2b5e7aa |
2994 | static void shift_assemble_arm(int i,struct regstat *i_regs) |
57871462 |
2995 | { |
2996 | if(rt1[i]) { |
2997 | if(opcode2[i]<=0x07) // SLLV/SRLV/SRAV |
2998 | { |
2999 | signed char s,t,shift; |
3000 | t=get_reg(i_regs->regmap,rt1[i]); |
3001 | s=get_reg(i_regs->regmap,rs1[i]); |
3002 | shift=get_reg(i_regs->regmap,rs2[i]); |
3003 | if(t>=0){ |
3004 | if(rs1[i]==0) |
3005 | { |
3006 | emit_zeroreg(t); |
3007 | } |
3008 | else if(rs2[i]==0) |
3009 | { |
3010 | assert(s>=0); |
3011 | if(s!=t) emit_mov(s,t); |
3012 | } |
3013 | else |
3014 | { |
3015 | emit_andimm(shift,31,HOST_TEMPREG); |
3016 | if(opcode2[i]==4) // SLLV |
3017 | { |
3018 | emit_shl(s,HOST_TEMPREG,t); |
3019 | } |
3020 | if(opcode2[i]==6) // SRLV |
3021 | { |
3022 | emit_shr(s,HOST_TEMPREG,t); |
3023 | } |
3024 | if(opcode2[i]==7) // SRAV |
3025 | { |
3026 | emit_sar(s,HOST_TEMPREG,t); |
3027 | } |
3028 | } |
3029 | } |
3030 | } else { // DSLLV/DSRLV/DSRAV |
3031 | signed char sh,sl,th,tl,shift; |
3032 | th=get_reg(i_regs->regmap,rt1[i]|64); |
3033 | tl=get_reg(i_regs->regmap,rt1[i]); |
3034 | sh=get_reg(i_regs->regmap,rs1[i]|64); |
3035 | sl=get_reg(i_regs->regmap,rs1[i]); |
3036 | shift=get_reg(i_regs->regmap,rs2[i]); |
3037 | if(tl>=0){ |
3038 | if(rs1[i]==0) |
3039 | { |
3040 | emit_zeroreg(tl); |
3041 | if(th>=0) emit_zeroreg(th); |
3042 | } |
3043 | else if(rs2[i]==0) |
3044 | { |
3045 | assert(sl>=0); |
3046 | if(sl!=tl) emit_mov(sl,tl); |
3047 | if(th>=0&&sh!=th) emit_mov(sh,th); |
3048 | } |
3049 | else |
3050 | { |
3051 | // FIXME: What if shift==tl ? |
3052 | assert(shift!=tl); |
3053 | int temp=get_reg(i_regs->regmap,-1); |
3054 | int real_th=th; |
3055 | if(th<0&&opcode2[i]!=0x14) {th=temp;} // DSLLV doesn't need a temporary register |
3056 | assert(sl>=0); |
3057 | assert(sh>=0); |
3058 | emit_andimm(shift,31,HOST_TEMPREG); |
3059 | if(opcode2[i]==0x14) // DSLLV |
3060 | { |
3061 | if(th>=0) emit_shl(sh,HOST_TEMPREG,th); |
3062 | emit_rsbimm(HOST_TEMPREG,32,HOST_TEMPREG); |
3063 | emit_orrshr(sl,HOST_TEMPREG,th); |
3064 | emit_andimm(shift,31,HOST_TEMPREG); |
3065 | emit_testimm(shift,32); |
3066 | emit_shl(sl,HOST_TEMPREG,tl); |
3067 | if(th>=0) emit_cmovne_reg(tl,th); |
3068 | emit_cmovne_imm(0,tl); |
3069 | } |
3070 | if(opcode2[i]==0x16) // DSRLV |
3071 | { |
3072 | assert(th>=0); |
3073 | emit_shr(sl,HOST_TEMPREG,tl); |
3074 | emit_rsbimm(HOST_TEMPREG,32,HOST_TEMPREG); |
3075 | emit_orrshl(sh,HOST_TEMPREG,tl); |
3076 | emit_andimm(shift,31,HOST_TEMPREG); |
3077 | emit_testimm(shift,32); |
3078 | emit_shr(sh,HOST_TEMPREG,th); |
3079 | emit_cmovne_reg(th,tl); |
3080 | if(real_th>=0) emit_cmovne_imm(0,th); |
3081 | } |
3082 | if(opcode2[i]==0x17) // DSRAV |
3083 | { |
3084 | assert(th>=0); |
3085 | emit_shr(sl,HOST_TEMPREG,tl); |
3086 | emit_rsbimm(HOST_TEMPREG,32,HOST_TEMPREG); |
3087 | if(real_th>=0) { |
3088 | assert(temp>=0); |
3089 | emit_sarimm(th,31,temp); |
3090 | } |
3091 | emit_orrshl(sh,HOST_TEMPREG,tl); |
3092 | emit_andimm(shift,31,HOST_TEMPREG); |
3093 | emit_testimm(shift,32); |
3094 | emit_sar(sh,HOST_TEMPREG,th); |
3095 | emit_cmovne_reg(th,tl); |
3096 | if(real_th>=0) emit_cmovne_reg(temp,th); |
3097 | } |
3098 | } |
3099 | } |
3100 | } |
3101 | } |
3102 | } |
ffb0b9e0 |
3103 | |
ffb0b9e0 |
3104 | static void speculate_mov(int rs,int rt) |
3105 | { |
3106 | if(rt!=0) { |
3107 | smrv_strong_next|=1<<rt; |
3108 | smrv[rt]=smrv[rs]; |
3109 | } |
3110 | } |
3111 | |
3112 | static void speculate_mov_weak(int rs,int rt) |
3113 | { |
3114 | if(rt!=0) { |
3115 | smrv_weak_next|=1<<rt; |
3116 | smrv[rt]=smrv[rs]; |
3117 | } |
3118 | } |
3119 | |
3120 | static void speculate_register_values(int i) |
3121 | { |
3122 | if(i==0) { |
3123 | memcpy(smrv,psxRegs.GPR.r,sizeof(smrv)); |
3124 | // gp,sp are likely to stay the same throughout the block |
3125 | smrv_strong_next=(1<<28)|(1<<29)|(1<<30); |
3126 | smrv_weak_next=~smrv_strong_next; |
3127 | //printf(" llr %08x\n", smrv[4]); |
3128 | } |
3129 | smrv_strong=smrv_strong_next; |
3130 | smrv_weak=smrv_weak_next; |
3131 | switch(itype[i]) { |
3132 | case ALU: |
3133 | if ((smrv_strong>>rs1[i])&1) speculate_mov(rs1[i],rt1[i]); |
3134 | else if((smrv_strong>>rs2[i])&1) speculate_mov(rs2[i],rt1[i]); |
3135 | else if((smrv_weak>>rs1[i])&1) speculate_mov_weak(rs1[i],rt1[i]); |
3136 | else if((smrv_weak>>rs2[i])&1) speculate_mov_weak(rs2[i],rt1[i]); |
3137 | else { |
3138 | smrv_strong_next&=~(1<<rt1[i]); |
3139 | smrv_weak_next&=~(1<<rt1[i]); |
3140 | } |
3141 | break; |
3142 | case SHIFTIMM: |
3143 | smrv_strong_next&=~(1<<rt1[i]); |
3144 | smrv_weak_next&=~(1<<rt1[i]); |
3145 | // fallthrough |
3146 | case IMM16: |
3147 | if(rt1[i]&&is_const(®s[i],rt1[i])) { |
3148 | int value,hr=get_reg(regs[i].regmap,rt1[i]); |
3149 | if(hr>=0) { |
3150 | if(get_final_value(hr,i,&value)) |
3151 | smrv[rt1[i]]=value; |
3152 | else smrv[rt1[i]]=constmap[i][hr]; |
3153 | smrv_strong_next|=1<<rt1[i]; |
3154 | } |
3155 | } |
3156 | else { |
3157 | if ((smrv_strong>>rs1[i])&1) speculate_mov(rs1[i],rt1[i]); |
3158 | else if((smrv_weak>>rs1[i])&1) speculate_mov_weak(rs1[i],rt1[i]); |
3159 | } |
3160 | break; |
3161 | case LOAD: |
3162 | if(start<0x2000&&(rt1[i]==26||(smrv[rt1[i]]>>24)==0xa0)) { |
3163 | // special case for BIOS |
3164 | smrv[rt1[i]]=0xa0000000; |
3165 | smrv_strong_next|=1<<rt1[i]; |
3166 | break; |
3167 | } |
3168 | // fallthrough |
3169 | case SHIFT: |
3170 | case LOADLR: |
3171 | case MOV: |
3172 | smrv_strong_next&=~(1<<rt1[i]); |
3173 | smrv_weak_next&=~(1<<rt1[i]); |
3174 | break; |
3175 | case COP0: |
3176 | case COP2: |
3177 | if(opcode2[i]==0||opcode2[i]==2) { // MFC/CFC |
3178 | smrv_strong_next&=~(1<<rt1[i]); |
3179 | smrv_weak_next&=~(1<<rt1[i]); |
3180 | } |
3181 | break; |
3182 | case C2LS: |
3183 | if (opcode[i]==0x32) { // LWC2 |
3184 | smrv_strong_next&=~(1<<rt1[i]); |
3185 | smrv_weak_next&=~(1<<rt1[i]); |
3186 | } |
3187 | break; |
3188 | } |
3189 | #if 0 |
3190 | int r=4; |
3191 | printf("x %08x %08x %d %d c %08x %08x\n",smrv[r],start+i*4, |
3192 | ((smrv_strong>>r)&1),(smrv_weak>>r)&1,regs[i].isconst,regs[i].wasconst); |
3193 | #endif |
3194 | } |
3195 | |
3196 | enum { |
3197 | MTYPE_8000 = 0, |
3198 | MTYPE_8020, |
3199 | MTYPE_0000, |
3200 | MTYPE_A000, |
3201 | MTYPE_1F80, |
3202 | }; |
3203 | |
3204 | static int get_ptr_mem_type(u_int a) |
3205 | { |
3206 | if(a < 0x00200000) { |
3207 | if(a<0x1000&&((start>>20)==0xbfc||(start>>24)==0xa0)) |
3208 | // return wrong, must use memhandler for BIOS self-test to pass |
3209 | // 007 does similar stuff from a00 mirror, weird stuff |
3210 | return MTYPE_8000; |
3211 | return MTYPE_0000; |
3212 | } |
3213 | if(0x1f800000 <= a && a < 0x1f801000) |
3214 | return MTYPE_1F80; |
3215 | if(0x80200000 <= a && a < 0x80800000) |
3216 | return MTYPE_8020; |
3217 | if(0xa0000000 <= a && a < 0xa0200000) |
3218 | return MTYPE_A000; |
3219 | return MTYPE_8000; |
3220 | } |
ffb0b9e0 |
3221 | |
3222 | static int emit_fastpath_cmp_jump(int i,int addr,int *addr_reg_override) |
3223 | { |
581335b0 |
3224 | int jaddr=0,type=0; |
ffb0b9e0 |
3225 | int mr=rs1[i]; |
3226 | if(((smrv_strong|smrv_weak)>>mr)&1) { |
3227 | type=get_ptr_mem_type(smrv[mr]); |
3228 | //printf("set %08x @%08x r%d %d\n", smrv[mr], start+i*4, mr, type); |
3229 | } |
3230 | else { |
3231 | // use the mirror we are running on |
3232 | type=get_ptr_mem_type(start); |
3233 | //printf("set nospec @%08x r%d %d\n", start+i*4, mr, type); |
3234 | } |
3235 | |
3236 | if(type==MTYPE_8020) { // RAM 80200000+ mirror |
3237 | emit_andimm(addr,~0x00e00000,HOST_TEMPREG); |
3238 | addr=*addr_reg_override=HOST_TEMPREG; |
3239 | type=0; |
3240 | } |
3241 | else if(type==MTYPE_0000) { // RAM 0 mirror |
3242 | emit_orimm(addr,0x80000000,HOST_TEMPREG); |
3243 | addr=*addr_reg_override=HOST_TEMPREG; |
3244 | type=0; |
3245 | } |
3246 | else if(type==MTYPE_A000) { // RAM A mirror |
3247 | emit_andimm(addr,~0x20000000,HOST_TEMPREG); |
3248 | addr=*addr_reg_override=HOST_TEMPREG; |
3249 | type=0; |
3250 | } |
3251 | else if(type==MTYPE_1F80) { // scratchpad |
6d760c92 |
3252 | if (psxH == (void *)0x1f800000) { |
3253 | emit_addimm(addr,-0x1f800000,HOST_TEMPREG); |
3254 | emit_cmpimm(HOST_TEMPREG,0x1000); |
3255 | jaddr=(int)out; |
3256 | emit_jc(0); |
3257 | } |
3258 | else { |
3259 | // do usual RAM check, jump will go to the right handler |
3260 | type=0; |
3261 | } |
ffb0b9e0 |
3262 | } |
ffb0b9e0 |
3263 | |
3264 | if(type==0) |
3265 | { |
3266 | emit_cmpimm(addr,RAM_SIZE); |
3267 | jaddr=(int)out; |
3268 | #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK |
3269 | // Hint to branch predictor that the branch is unlikely to be taken |
3270 | if(rs1[i]>=28) |
3271 | emit_jno_unlikely(0); |
3272 | else |
3273 | #endif |
3274 | emit_jno(0); |
a327ad27 |
3275 | if(ram_offset!=0) { |
3276 | emit_addimm(addr,ram_offset,HOST_TEMPREG); |
3277 | addr=*addr_reg_override=HOST_TEMPREG; |
3278 | } |
ffb0b9e0 |
3279 | } |
3280 | |
3281 | return jaddr; |
3282 | } |
3283 | |
57871462 |
3284 | #define shift_assemble shift_assemble_arm |
3285 | |
e2b5e7aa |
3286 | static void loadlr_assemble_arm(int i,struct regstat *i_regs) |
57871462 |
3287 | { |
3288 | int s,th,tl,temp,temp2,addr,map=-1; |
3289 | int offset; |
3290 | int jaddr=0; |
af4ee1fe |
3291 | int memtarget=0,c=0; |
ffb0b9e0 |
3292 | int fastload_reg_override=0; |
57871462 |
3293 | u_int hr,reglist=0; |
3294 | th=get_reg(i_regs->regmap,rt1[i]|64); |
3295 | tl=get_reg(i_regs->regmap,rt1[i]); |
3296 | s=get_reg(i_regs->regmap,rs1[i]); |
3297 | temp=get_reg(i_regs->regmap,-1); |
3298 | temp2=get_reg(i_regs->regmap,FTEMP); |
3299 | addr=get_reg(i_regs->regmap,AGEN1+(i&1)); |
3300 | assert(addr<0); |
3301 | offset=imm[i]; |
3302 | for(hr=0;hr<HOST_REGS;hr++) { |
3303 | if(i_regs->regmap[hr]>=0) reglist|=1<<hr; |
3304 | } |
3305 | reglist|=1<<temp; |
3306 | if(offset||s<0||c) addr=temp2; |
3307 | else addr=s; |
3308 | if(s>=0) { |
3309 | c=(i_regs->wasconst>>s)&1; |
af4ee1fe |
3310 | if(c) { |
3311 | memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE; |
af4ee1fe |
3312 | } |
57871462 |
3313 | } |
1edfcc68 |
3314 | if(!c) { |
3315 | #ifdef RAM_OFFSET |
3316 | map=get_reg(i_regs->regmap,ROREG); |
3317 | if(map<0) emit_loadreg(ROREG,map=HOST_TEMPREG); |
3318 | #endif |
3319 | emit_shlimm(addr,3,temp); |
3320 | if (opcode[i]==0x22||opcode[i]==0x26) { |
3321 | emit_andimm(addr,0xFFFFFFFC,temp2); // LWL/LWR |
3322 | }else{ |
3323 | emit_andimm(addr,0xFFFFFFF8,temp2); // LDL/LDR |
535d208a |
3324 | } |
1edfcc68 |
3325 | jaddr=emit_fastpath_cmp_jump(i,temp2,&fastload_reg_override); |
3326 | } |
3327 | else { |
3328 | if(ram_offset&&memtarget) { |
3329 | emit_addimm(temp2,ram_offset,HOST_TEMPREG); |
3330 | fastload_reg_override=HOST_TEMPREG; |
57871462 |
3331 | } |
1edfcc68 |
3332 | if (opcode[i]==0x22||opcode[i]==0x26) { |
3333 | emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR |
535d208a |
3334 | }else{ |
1edfcc68 |
3335 | emit_movimm(((constmap[i][s]+offset)<<3)&56,temp); // LDL/LDR |
535d208a |
3336 | } |
535d208a |
3337 | } |
3338 | if (opcode[i]==0x22||opcode[i]==0x26) { // LWL/LWR |
3339 | if(!c||memtarget) { |
ffb0b9e0 |
3340 | int a=temp2; |
3341 | if(fastload_reg_override) a=fastload_reg_override; |
535d208a |
3342 | //emit_readword_indexed((int)rdram-0x80000000,temp2,temp2); |
ffb0b9e0 |
3343 | emit_readword_indexed_tlb(0,a,map,temp2); |
535d208a |
3344 | if(jaddr) add_stub(LOADW_STUB,jaddr,(int)out,i,temp2,(int)i_regs,ccadj[i],reglist); |
3345 | } |
3346 | else |
3347 | inline_readstub(LOADW_STUB,i,(constmap[i][s]+offset)&0xFFFFFFFC,i_regs->regmap,FTEMP,ccadj[i],reglist); |
3348 | if(rt1[i]) { |
3349 | assert(tl>=0); |
57871462 |
3350 | emit_andimm(temp,24,temp); |
2002a1db |
3351 | #ifdef BIG_ENDIAN_MIPS |
3352 | if (opcode[i]==0x26) // LWR |
3353 | #else |
3354 | if (opcode[i]==0x22) // LWL |
3355 | #endif |
3356 | emit_xorimm(temp,24,temp); |
57871462 |
3357 | emit_movimm(-1,HOST_TEMPREG); |
3358 | if (opcode[i]==0x26) { |
3359 | emit_shr(temp2,temp,temp2); |
3360 | emit_bic_lsr(tl,HOST_TEMPREG,temp,tl); |
3361 | }else{ |
3362 | emit_shl(temp2,temp,temp2); |
3363 | emit_bic_lsl(tl,HOST_TEMPREG,temp,tl); |
3364 | } |
3365 | emit_or(temp2,tl,tl); |
57871462 |
3366 | } |
535d208a |
3367 | //emit_storereg(rt1[i],tl); // DEBUG |
3368 | } |
3369 | if (opcode[i]==0x1A||opcode[i]==0x1B) { // LDL/LDR |
ffb0b9e0 |
3370 | // FIXME: little endian, fastload_reg_override |
535d208a |
3371 | int temp2h=get_reg(i_regs->regmap,FTEMP|64); |
3372 | if(!c||memtarget) { |
3373 | //if(th>=0) emit_readword_indexed((int)rdram-0x80000000,temp2,temp2h); |
3374 | //emit_readword_indexed((int)rdram-0x7FFFFFFC,temp2,temp2); |
3375 | emit_readdword_indexed_tlb(0,temp2,map,temp2h,temp2); |
3376 | if(jaddr) add_stub(LOADD_STUB,jaddr,(int)out,i,temp2,(int)i_regs,ccadj[i],reglist); |
3377 | } |
3378 | else |
3379 | inline_readstub(LOADD_STUB,i,(constmap[i][s]+offset)&0xFFFFFFF8,i_regs->regmap,FTEMP,ccadj[i],reglist); |
3380 | if(rt1[i]) { |
3381 | assert(th>=0); |
3382 | assert(tl>=0); |
57871462 |
3383 | emit_testimm(temp,32); |
3384 | emit_andimm(temp,24,temp); |
3385 | if (opcode[i]==0x1A) { // LDL |
3386 | emit_rsbimm(temp,32,HOST_TEMPREG); |
3387 | emit_shl(temp2h,temp,temp2h); |
3388 | emit_orrshr(temp2,HOST_TEMPREG,temp2h); |
3389 | emit_movimm(-1,HOST_TEMPREG); |
3390 | emit_shl(temp2,temp,temp2); |
3391 | emit_cmove_reg(temp2h,th); |
3392 | emit_biceq_lsl(tl,HOST_TEMPREG,temp,tl); |
3393 | emit_bicne_lsl(th,HOST_TEMPREG,temp,th); |
3394 | emit_orreq(temp2,tl,tl); |
3395 | emit_orrne(temp2,th,th); |
3396 | } |
3397 | if (opcode[i]==0x1B) { // LDR |
3398 | emit_xorimm(temp,24,temp); |
3399 | emit_rsbimm(temp,32,HOST_TEMPREG); |
3400 | emit_shr(temp2,temp,temp2); |
3401 | emit_orrshl(temp2h,HOST_TEMPREG,temp2); |
3402 | emit_movimm(-1,HOST_TEMPREG); |
3403 | emit_shr(temp2h,temp,temp2h); |
3404 | emit_cmovne_reg(temp2,tl); |
3405 | emit_bicne_lsr(th,HOST_TEMPREG,temp,th); |
3406 | emit_biceq_lsr(tl,HOST_TEMPREG,temp,tl); |
3407 | emit_orrne(temp2h,th,th); |
3408 | emit_orreq(temp2h,tl,tl); |
3409 | } |
3410 | } |
3411 | } |
3412 | } |
3413 | #define loadlr_assemble loadlr_assemble_arm |
3414 | |
e2b5e7aa |
3415 | static void cop0_assemble(int i,struct regstat *i_regs) |
57871462 |
3416 | { |
3417 | if(opcode2[i]==0) // MFC0 |
3418 | { |
3419 | signed char t=get_reg(i_regs->regmap,rt1[i]); |
3420 | char copr=(source[i]>>11)&0x1f; |
3421 | //assert(t>=0); // Why does this happen? OOT is weird |
f1b3b369 |
3422 | if(t>=0&&rt1[i]!=0) { |
7139f3c8 |
3423 | emit_readword((int)®_cop0+copr*4,t); |
57871462 |
3424 | } |
3425 | } |
3426 | else if(opcode2[i]==4) // MTC0 |
3427 | { |
3428 | signed char s=get_reg(i_regs->regmap,rs1[i]); |
3429 | char copr=(source[i]>>11)&0x1f; |
3430 | assert(s>=0); |
63cb0298 |
3431 | wb_register(rs1[i],i_regs->regmap,i_regs->dirty,i_regs->is32); |
7139f3c8 |
3432 | if(copr==9||copr==11||copr==12||copr==13) { |
63cb0298 |
3433 | emit_readword((int)&last_count,HOST_TEMPREG); |
57871462 |
3434 | emit_loadreg(CCREG,HOST_CCREG); // TODO: do proper reg alloc |
63cb0298 |
3435 | emit_add(HOST_CCREG,HOST_TEMPREG,HOST_CCREG); |
2573466a |
3436 | emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG); |
57871462 |
3437 | emit_writeword(HOST_CCREG,(int)&Count); |
3438 | } |
3439 | // What a mess. The status register (12) can enable interrupts, |
3440 | // so needs a special case to handle a pending interrupt. |
3441 | // The interrupt must be taken immediately, because a subsequent |
3442 | // instruction might disable interrupts again. |
7139f3c8 |
3443 | if(copr==12||copr==13) { |
fca1aef2 |
3444 | if (is_delayslot) { |
3445 | // burn cycles to cause cc_interrupt, which will |
3446 | // reschedule next_interupt. Relies on CCREG from above. |
3447 | assem_debug("MTC0 DS %d\n", copr); |
3448 | emit_writeword(HOST_CCREG,(int)&last_count); |
3449 | emit_movimm(0,HOST_CCREG); |
3450 | emit_storereg(CCREG,HOST_CCREG); |
caeefe31 |
3451 | emit_loadreg(rs1[i],1); |
fca1aef2 |
3452 | emit_movimm(copr,0); |
3453 | emit_call((int)pcsx_mtc0_ds); |
042c7287 |
3454 | emit_loadreg(rs1[i],s); |
fca1aef2 |
3455 | return; |
3456 | } |
63cb0298 |
3457 | emit_movimm(start+i*4+4,HOST_TEMPREG); |
3458 | emit_writeword(HOST_TEMPREG,(int)&pcaddr); |
3459 | emit_movimm(0,HOST_TEMPREG); |
3460 | emit_writeword(HOST_TEMPREG,(int)&pending_exception); |
57871462 |
3461 | } |
3462 | //else if(copr==12&&is_delayslot) emit_call((int)MTC0_R12); |
3463 | //else |
caeefe31 |
3464 | if(s==HOST_CCREG) |
3465 | emit_loadreg(rs1[i],1); |
3466 | else if(s!=1) |
63cb0298 |
3467 | emit_mov(s,1); |
fca1aef2 |
3468 | emit_movimm(copr,0); |
3469 | emit_call((int)pcsx_mtc0); |
7139f3c8 |
3470 | if(copr==9||copr==11||copr==12||copr==13) { |
57871462 |
3471 | emit_readword((int)&Count,HOST_CCREG); |
042c7287 |
3472 | emit_readword((int)&next_interupt,HOST_TEMPREG); |
2573466a |
3473 | emit_addimm(HOST_CCREG,-CLOCK_ADJUST(ccadj[i]),HOST_CCREG); |
042c7287 |
3474 | emit_sub(HOST_CCREG,HOST_TEMPREG,HOST_CCREG); |
3475 | emit_writeword(HOST_TEMPREG,(int)&last_count); |
57871462 |
3476 | emit_storereg(CCREG,HOST_CCREG); |
3477 | } |
7139f3c8 |
3478 | if(copr==12||copr==13) { |
57871462 |
3479 | assert(!is_delayslot); |
3480 | emit_readword((int)&pending_exception,14); |
042c7287 |
3481 | emit_test(14,14); |
3482 | emit_jne((int)&do_interrupt); |
57871462 |
3483 | } |
3484 | emit_loadreg(rs1[i],s); |
3485 | if(get_reg(i_regs->regmap,rs1[i]|64)>=0) |
3486 | emit_loadreg(rs1[i]|64,get_reg(i_regs->regmap,rs1[i]|64)); |
57871462 |
3487 | cop1_usable=0; |
3488 | } |
3489 | else |
3490 | { |
3491 | assert(opcode2[i]==0x10); |
576bbd8f |
3492 | if((source[i]&0x3f)==0x10) // RFE |
3493 | { |
3494 | emit_readword((int)&Status,0); |
3495 | emit_andimm(0,0x3c,1); |
3496 | emit_andimm(0,~0xf,0); |
3497 | emit_orrshr_imm(1,2,0); |
3498 | emit_writeword(0,(int)&Status); |
3499 | } |
57871462 |
3500 | } |
3501 | } |
3502 | |
b9b61529 |
3503 | static void cop2_get_dreg(u_int copr,signed char tl,signed char temp) |
3504 | { |
3505 | switch (copr) { |
3506 | case 1: |
3507 | case 3: |
3508 | case 5: |
3509 | case 8: |
3510 | case 9: |
3511 | case 10: |
3512 | case 11: |
3513 | emit_readword((int)®_cop2d[copr],tl); |
3514 | emit_signextend16(tl,tl); |
3515 | emit_writeword(tl,(int)®_cop2d[copr]); // hmh |
3516 | break; |
3517 | case 7: |
3518 | case 16: |
3519 | case 17: |
3520 | case 18: |
3521 | case 19: |
3522 | emit_readword((int)®_cop2d[copr],tl); |
3523 | emit_andimm(tl,0xffff,tl); |
3524 | emit_writeword(tl,(int)®_cop2d[copr]); |
3525 | break; |
3526 | case 15: |
3527 | emit_readword((int)®_cop2d[14],tl); // SXY2 |
3528 | emit_writeword(tl,(int)®_cop2d[copr]); |
3529 | break; |
3530 | case 28: |
b9b61529 |
3531 | case 29: |
3532 | emit_readword((int)®_cop2d[9],temp); |
3533 | emit_testimm(temp,0x8000); // do we need this? |
3534 | emit_andimm(temp,0xf80,temp); |
3535 | emit_andne_imm(temp,0,temp); |
f70d384d |
3536 | emit_shrimm(temp,7,tl); |
b9b61529 |
3537 | emit_readword((int)®_cop2d[10],temp); |
3538 | emit_testimm(temp,0x8000); |
3539 | emit_andimm(temp,0xf80,temp); |
3540 | emit_andne_imm(temp,0,temp); |
f70d384d |
3541 | emit_orrshr_imm(temp,2,tl); |
b9b61529 |
3542 | emit_readword((int)®_cop2d[11],temp); |
3543 | emit_testimm(temp,0x8000); |
3544 | emit_andimm(temp,0xf80,temp); |
3545 | emit_andne_imm(temp,0,temp); |
f70d384d |
3546 | emit_orrshl_imm(temp,3,tl); |
b9b61529 |
3547 | emit_writeword(tl,(int)®_cop2d[copr]); |
3548 | break; |
3549 | default: |
3550 | emit_readword((int)®_cop2d[copr],tl); |
3551 | break; |
3552 | } |
3553 | } |
3554 | |
3555 | static void cop2_put_dreg(u_int copr,signed char sl,signed char temp) |
3556 | { |
3557 | switch (copr) { |
3558 | case 15: |
3559 | emit_readword((int)®_cop2d[13],temp); // SXY1 |
3560 | emit_writeword(sl,(int)®_cop2d[copr]); |
3561 | emit_writeword(temp,(int)®_cop2d[12]); // SXY0 |
3562 | emit_readword((int)®_cop2d[14],temp); // SXY2 |
3563 | emit_writeword(sl,(int)®_cop2d[14]); |
3564 | emit_writeword(temp,(int)®_cop2d[13]); // SXY1 |
3565 | break; |
3566 | case 28: |
3567 | emit_andimm(sl,0x001f,temp); |
f70d384d |
3568 | emit_shlimm(temp,7,temp); |
b9b61529 |
3569 | emit_writeword(temp,(int)®_cop2d[9]); |
3570 | emit_andimm(sl,0x03e0,temp); |
f70d384d |
3571 | emit_shlimm(temp,2,temp); |
b9b61529 |
3572 | emit_writeword(temp,(int)®_cop2d[10]); |
3573 | emit_andimm(sl,0x7c00,temp); |
f70d384d |
3574 | emit_shrimm(temp,3,temp); |
b9b61529 |
3575 | emit_writeword(temp,(int)®_cop2d[11]); |
3576 | emit_writeword(sl,(int)®_cop2d[28]); |
3577 | break; |
3578 | case 30: |
3579 | emit_movs(sl,temp); |
3580 | emit_mvnmi(temp,temp); |
665f33e1 |
3581 | #ifdef HAVE_ARMV5 |
b9b61529 |
3582 | emit_clz(temp,temp); |
665f33e1 |
3583 | #else |
3584 | emit_movs(temp,HOST_TEMPREG); |
3585 | emit_movimm(0,temp); |
3586 | emit_jeq((int)out+4*4); |
3587 | emit_addpl_imm(temp,1,temp); |
3588 | emit_lslpls_imm(HOST_TEMPREG,1,HOST_TEMPREG); |
3589 | emit_jns((int)out-2*4); |
3590 | #endif |
b9b61529 |
3591 | emit_writeword(sl,(int)®_cop2d[30]); |
3592 | emit_writeword(temp,(int)®_cop2d[31]); |
3593 | break; |
b9b61529 |
3594 | case 31: |
3595 | break; |
3596 | default: |
3597 | emit_writeword(sl,(int)®_cop2d[copr]); |
3598 | break; |
3599 | } |
3600 | } |
3601 | |
e2b5e7aa |
3602 | static void cop2_assemble(int i,struct regstat *i_regs) |
b9b61529 |
3603 | { |
3604 | u_int copr=(source[i]>>11)&0x1f; |
3605 | signed char temp=get_reg(i_regs->regmap,-1); |
3606 | if (opcode2[i]==0) { // MFC2 |
3607 | signed char tl=get_reg(i_regs->regmap,rt1[i]); |
f1b3b369 |
3608 | if(tl>=0&&rt1[i]!=0) |
b9b61529 |
3609 | cop2_get_dreg(copr,tl,temp); |
3610 | } |
3611 | else if (opcode2[i]==4) { // MTC2 |
3612 | signed char sl=get_reg(i_regs->regmap,rs1[i]); |
3613 | cop2_put_dreg(copr,sl,temp); |
3614 | } |
3615 | else if (opcode2[i]==2) // CFC2 |
3616 | { |
3617 | signed char tl=get_reg(i_regs->regmap,rt1[i]); |
f1b3b369 |
3618 | if(tl>=0&&rt1[i]!=0) |
b9b61529 |
3619 | emit_readword((int)®_cop2c[copr],tl); |
3620 | } |
3621 | else if (opcode2[i]==6) // CTC2 |
3622 | { |
3623 | signed char sl=get_reg(i_regs->regmap,rs1[i]); |
3624 | switch(copr) { |
3625 | case 4: |
3626 | case 12: |
3627 | case 20: |
3628 | case 26: |
3629 | case 27: |
3630 | case 29: |
3631 | case 30: |
3632 | emit_signextend16(sl,temp); |
3633 | break; |
3634 | case 31: |
3635 | //value = value & 0x7ffff000; |
3636 | //if (value & 0x7f87e000) value |= 0x80000000; |
3637 | emit_shrimm(sl,12,temp); |
3638 | emit_shlimm(temp,12,temp); |
3639 | emit_testimm(temp,0x7f000000); |
3640 | emit_testeqimm(temp,0x00870000); |
3641 | emit_testeqimm(temp,0x0000e000); |
3642 | emit_orrne_imm(temp,0x80000000,temp); |
3643 | break; |
3644 | default: |
3645 | temp=sl; |
3646 | break; |
3647 | } |
3648 | emit_writeword(temp,(int)®_cop2c[copr]); |
3649 | assert(sl>=0); |
3650 | } |
3651 | } |
3652 | |
054175e9 |
3653 | static void c2op_prologue(u_int op,u_int reglist) |
3654 | { |
3655 | save_regs_all(reglist); |
82ed88eb |
3656 | #ifdef PCNT |
3657 | emit_movimm(op,0); |
3658 | emit_call((int)pcnt_gte_start); |
3659 | #endif |
054175e9 |
3660 | emit_addimm(FP,(int)&psxRegs.CP2D.r[0]-(int)&dynarec_local,0); // cop2 regs |
3661 | } |
3662 | |
3663 | static void c2op_epilogue(u_int op,u_int reglist) |
3664 | { |
82ed88eb |
3665 | #ifdef PCNT |
3666 | emit_movimm(op,0); |
3667 | emit_call((int)pcnt_gte_end); |
3668 | #endif |
054175e9 |
3669 | restore_regs_all(reglist); |
3670 | } |
3671 | |
6c0eefaf |
3672 | static void c2op_call_MACtoIR(int lm,int need_flags) |
3673 | { |
3674 | if(need_flags) |
3675 | emit_call((int)(lm?gteMACtoIR_lm1:gteMACtoIR_lm0)); |
3676 | else |
3677 | emit_call((int)(lm?gteMACtoIR_lm1_nf:gteMACtoIR_lm0_nf)); |
3678 | } |
3679 | |
3680 | static void c2op_call_rgb_func(void *func,int lm,int need_ir,int need_flags) |
3681 | { |
3682 | emit_call((int)func); |
3683 | // func is C code and trashes r0 |
3684 | emit_addimm(FP,(int)&psxRegs.CP2D.r[0]-(int)&dynarec_local,0); |
3685 | if(need_flags||need_ir) |
3686 | c2op_call_MACtoIR(lm,need_flags); |
3687 | emit_call((int)(need_flags?gteMACtoRGB:gteMACtoRGB_nf)); |
3688 | } |
3689 | |
054175e9 |
3690 | static void c2op_assemble(int i,struct regstat *i_regs) |
b9b61529 |
3691 | { |
b9b61529 |
3692 | u_int c2op=source[i]&0x3f; |
6c0eefaf |
3693 | u_int hr,reglist_full=0,reglist; |
054175e9 |
3694 | int need_flags,need_ir; |
b9b61529 |
3695 | for(hr=0;hr<HOST_REGS;hr++) { |
6c0eefaf |
3696 | if(i_regs->regmap[hr]>=0) reglist_full|=1<<hr; |
b9b61529 |
3697 | } |
4d646738 |
3698 | reglist=reglist_full&CALLER_SAVE_REGS; |
b9b61529 |
3699 | |
3700 | if (gte_handlers[c2op]!=NULL) { |
bedfea38 |
3701 | need_flags=!(gte_unneeded[i+1]>>63); // +1 because of how liveness detection works |
054175e9 |
3702 | need_ir=(gte_unneeded[i+1]&0xe00)!=0xe00; |
cbbd8dd7 |
3703 | assem_debug("gte op %08x, unneeded %016llx, need_flags %d, need_ir %d\n", |
3704 | source[i],gte_unneeded[i+1],need_flags,need_ir); |
0ff8c62c |
3705 | if(new_dynarec_hacks&NDHACK_GTE_NO_FLAGS) |
3706 | need_flags=0; |
6c0eefaf |
3707 | int shift = (source[i] >> 19) & 1; |
3708 | int lm = (source[i] >> 10) & 1; |
054175e9 |
3709 | switch(c2op) { |
19776aef |
3710 | #ifndef DRC_DBG |
054175e9 |
3711 | case GTE_MVMVA: { |
82336ba3 |
3712 | #ifdef HAVE_ARMV5 |
054175e9 |
3713 | int v = (source[i] >> 15) & 3; |
3714 | int cv = (source[i] >> 13) & 3; |
3715 | int mx = (source[i] >> 17) & 3; |
4d646738 |
3716 | reglist=reglist_full&(CALLER_SAVE_REGS|0xf0); // +{r4-r7} |
054175e9 |
3717 | c2op_prologue(c2op,reglist); |
3718 | /* r4,r5 = VXYZ(v) packed; r6 = &MX11(mx); r7 = &CV1(cv) */ |
3719 | if(v<3) |
3720 | emit_ldrd(v*8,0,4); |
3721 | else { |
3722 | emit_movzwl_indexed(9*4,0,4); // gteIR |
3723 | emit_movzwl_indexed(10*4,0,6); |
3724 | emit_movzwl_indexed(11*4,0,5); |
3725 | emit_orrshl_imm(6,16,4); |
3726 | } |
3727 | if(mx<3) |
3728 | emit_addimm(0,32*4+mx*8*4,6); |
3729 | else |
3730 | emit_readword((int)&zeromem_ptr,6); |
3731 | if(cv<3) |
3732 | emit_addimm(0,32*4+(cv*8+5)*4,7); |
3733 | else |
3734 | emit_readword((int)&zeromem_ptr,7); |
3735 | #ifdef __ARM_NEON__ |
3736 | emit_movimm(source[i],1); // opcode |
3737 | emit_call((int)gteMVMVA_part_neon); |
3738 | if(need_flags) { |
3739 | emit_movimm(lm,1); |
3740 | emit_call((int)gteMACtoIR_flags_neon); |
3741 | } |
3742 | #else |
3743 | if(cv==3&&shift) |
3744 | emit_call((int)gteMVMVA_part_cv3sh12_arm); |
3745 | else { |
3746 | emit_movimm(shift,1); |
3747 | emit_call((int)(need_flags?gteMVMVA_part_arm:gteMVMVA_part_nf_arm)); |
3748 | } |
6c0eefaf |
3749 | if(need_flags||need_ir) |
3750 | c2op_call_MACtoIR(lm,need_flags); |
82336ba3 |
3751 | #endif |
3752 | #else /* if not HAVE_ARMV5 */ |
3753 | c2op_prologue(c2op,reglist); |
3754 | emit_movimm(source[i],1); // opcode |
3755 | emit_writeword(1,(int)&psxRegs.code); |
3756 | emit_call((int)(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op])); |
054175e9 |
3757 | #endif |
3758 | break; |
3759 | } |
6c0eefaf |
3760 | case GTE_OP: |
3761 | c2op_prologue(c2op,reglist); |
3762 | emit_call((int)(shift?gteOP_part_shift:gteOP_part_noshift)); |
3763 | if(need_flags||need_ir) { |
3764 | emit_addimm(FP,(int)&psxRegs.CP2D.r[0]-(int)&dynarec_local,0); |
3765 | c2op_call_MACtoIR(lm,need_flags); |
3766 | } |
3767 | break; |
3768 | case GTE_DPCS: |
3769 | c2op_prologue(c2op,reglist); |
3770 | c2op_call_rgb_func(shift?gteDPCS_part_shift:gteDPCS_part_noshift,lm,need_ir,need_flags); |
3771 | break; |
3772 | case GTE_INTPL: |
3773 | c2op_prologue(c2op,reglist); |
3774 | c2op_call_rgb_func(shift?gteINTPL_part_shift:gteINTPL_part_noshift,lm,need_ir,need_flags); |
3775 | break; |
3776 | case GTE_SQR: |
3777 | c2op_prologue(c2op,reglist); |
3778 | emit_call((int)(shift?gteSQR_part_shift:gteSQR_part_noshift)); |
3779 | if(need_flags||need_ir) { |
3780 | emit_addimm(FP,(int)&psxRegs.CP2D.r[0]-(int)&dynarec_local,0); |
3781 | c2op_call_MACtoIR(lm,need_flags); |
3782 | } |
3783 | break; |
3784 | case GTE_DCPL: |
3785 | c2op_prologue(c2op,reglist); |
3786 | c2op_call_rgb_func(gteDCPL_part,lm,need_ir,need_flags); |
3787 | break; |
3788 | case GTE_GPF: |
3789 | c2op_prologue(c2op,reglist); |
3790 | c2op_call_rgb_func(shift?gteGPF_part_shift:gteGPF_part_noshift,lm,need_ir,need_flags); |
3791 | break; |
3792 | case GTE_GPL: |
3793 | c2op_prologue(c2op,reglist); |
3794 | c2op_call_rgb_func(shift?gteGPL_part_shift:gteGPL_part_noshift,lm,need_ir,need_flags); |
3795 | break; |
19776aef |
3796 | #endif |
054175e9 |
3797 | default: |
054175e9 |
3798 | c2op_prologue(c2op,reglist); |
19776aef |
3799 | #ifdef DRC_DBG |
3800 | emit_movimm(source[i],1); // opcode |
3801 | emit_writeword(1,(int)&psxRegs.code); |
3802 | #endif |
054175e9 |
3803 | emit_call((int)(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op])); |
3804 | break; |
3805 | } |
3806 | c2op_epilogue(c2op,reglist); |
3807 | } |
b9b61529 |
3808 | } |
3809 | |
e2b5e7aa |
3810 | static void cop1_unusable(int i,struct regstat *i_regs) |
3d624f89 |
3811 | { |
3812 | // XXX: should just just do the exception instead |
3813 | if(!cop1_usable) { |
3814 | int jaddr=(int)out; |
3815 | emit_jmp(0); |
3816 | add_stub(FP_STUB,jaddr,(int)out,i,0,(int)i_regs,is_delayslot,0); |
3817 | cop1_usable=1; |
3818 | } |
3819 | } |
3820 | |
e2b5e7aa |
3821 | static void cop1_assemble(int i,struct regstat *i_regs) |
57871462 |
3822 | { |
3d624f89 |
3823 | cop1_unusable(i, i_regs); |
57871462 |
3824 | } |
3825 | |
e2b5e7aa |
3826 | static void fconv_assemble_arm(int i,struct regstat *i_regs) |
57871462 |
3827 | { |
3d624f89 |
3828 | cop1_unusable(i, i_regs); |
57871462 |
3829 | } |
3830 | #define fconv_assemble fconv_assemble_arm |
3831 | |
e2b5e7aa |
3832 | static void fcomp_assemble(int i,struct regstat *i_regs) |
57871462 |
3833 | { |
3d624f89 |
3834 | cop1_unusable(i, i_regs); |
57871462 |
3835 | } |
3836 | |
e2b5e7aa |
3837 | static void float_assemble(int i,struct regstat *i_regs) |
57871462 |
3838 | { |
3d624f89 |
3839 | cop1_unusable(i, i_regs); |
57871462 |
3840 | } |
3841 | |
e2b5e7aa |
3842 | static void multdiv_assemble_arm(int i,struct regstat *i_regs) |
57871462 |
3843 | { |
3844 | // case 0x18: MULT |
3845 | // case 0x19: MULTU |
3846 | // case 0x1A: DIV |
3847 | // case 0x1B: DIVU |
3848 | // case 0x1C: DMULT |
3849 | // case 0x1D: DMULTU |
3850 | // case 0x1E: DDIV |
3851 | // case 0x1F: DDIVU |
3852 | if(rs1[i]&&rs2[i]) |
3853 | { |
3854 | if((opcode2[i]&4)==0) // 32-bit |
3855 | { |
3856 | if(opcode2[i]==0x18) // MULT |
3857 | { |
3858 | signed char m1=get_reg(i_regs->regmap,rs1[i]); |
3859 | signed char m2=get_reg(i_regs->regmap,rs2[i]); |
3860 | signed char hi=get_reg(i_regs->regmap,HIREG); |
3861 | signed char lo=get_reg(i_regs->regmap,LOREG); |
3862 | assert(m1>=0); |
3863 | assert(m2>=0); |
3864 | assert(hi>=0); |
3865 | assert(lo>=0); |
3866 | emit_smull(m1,m2,hi,lo); |
3867 | } |
3868 | if(opcode2[i]==0x19) // MULTU |
3869 | { |
3870 | signed char m1=get_reg(i_regs->regmap,rs1[i]); |
3871 | signed char m2=get_reg(i_regs->regmap,rs2[i]); |
3872 | signed char hi=get_reg(i_regs->regmap,HIREG); |
3873 | signed char lo=get_reg(i_regs->regmap,LOREG); |
3874 | assert(m1>=0); |
3875 | assert(m2>=0); |
3876 | assert(hi>=0); |
3877 | assert(lo>=0); |
3878 | emit_umull(m1,m2,hi,lo); |
3879 | } |
3880 | if(opcode2[i]==0x1A) // DIV |
3881 | { |
3882 | signed char d1=get_reg(i_regs->regmap,rs1[i]); |
3883 | signed char d2=get_reg(i_regs->regmap,rs2[i]); |
3884 | assert(d1>=0); |
3885 | assert(d2>=0); |
3886 | signed char quotient=get_reg(i_regs->regmap,LOREG); |
3887 | signed char remainder=get_reg(i_regs->regmap,HIREG); |
3888 | assert(quotient>=0); |
3889 | assert(remainder>=0); |
3890 | emit_movs(d1,remainder); |
44a80f6a |
3891 | emit_movimm(0xffffffff,quotient); |
3892 | emit_negmi(quotient,quotient); // .. quotient and .. |
3893 | emit_negmi(remainder,remainder); // .. remainder for div0 case (will be negated back after jump) |
57871462 |
3894 | emit_movs(d2,HOST_TEMPREG); |
3895 | emit_jeq((int)out+52); // Division by zero |
82336ba3 |
3896 | emit_negsmi(HOST_TEMPREG,HOST_TEMPREG); |
665f33e1 |
3897 | #ifdef HAVE_ARMV5 |
57871462 |
3898 | emit_clz(HOST_TEMPREG,quotient); |
3899 | emit_shl(HOST_TEMPREG,quotient,HOST_TEMPREG); |
665f33e1 |
3900 | #else |
3901 | emit_movimm(0,quotient); |
3902 | emit_addpl_imm(quotient,1,quotient); |
3903 | emit_lslpls_imm(HOST_TEMPREG,1,HOST_TEMPREG); |
3904 | emit_jns((int)out-2*4); |
3905 | #endif |
57871462 |
3906 | emit_orimm(quotient,1<<31,quotient); |
3907 | emit_shr(quotient,quotient,quotient); |
3908 | emit_cmp(remainder,HOST_TEMPREG); |
3909 | emit_subcs(remainder,HOST_TEMPREG,remainder); |
3910 | emit_adcs(quotient,quotient,quotient); |
3911 | emit_shrimm(HOST_TEMPREG,1,HOST_TEMPREG); |
3912 | emit_jcc((int)out-16); // -4 |
3913 | emit_teq(d1,d2); |
3914 | emit_negmi(quotient,quotient); |
3915 | emit_test(d1,d1); |
3916 | emit_negmi(remainder,remainder); |
3917 | } |
3918 | if(opcode2[i]==0x1B) // DIVU |
3919 | { |
3920 | signed char d1=get_reg(i_regs->regmap,rs1[i]); // dividend |
3921 | signed char d2=get_reg(i_regs->regmap,rs2[i]); // divisor |
3922 | assert(d1>=0); |
3923 | assert(d2>=0); |
3924 | signed char quotient=get_reg(i_regs->regmap,LOREG); |
3925 | signed char remainder=get_reg(i_regs->regmap,HIREG); |
3926 | assert(quotient>=0); |
3927 | assert(remainder>=0); |
44a80f6a |
3928 | emit_mov(d1,remainder); |
3929 | emit_movimm(0xffffffff,quotient); // div0 case |
57871462 |
3930 | emit_test(d2,d2); |
44a80f6a |
3931 | emit_jeq((int)out+40); // Division by zero |
665f33e1 |
3932 | #ifdef HAVE_ARMV5 |
57871462 |
3933 | emit_clz(d2,HOST_TEMPREG); |
3934 | emit_movimm(1<<31,quotient); |
3935 | emit_shl(d2,HOST_TEMPREG,d2); |
665f33e1 |
3936 | #else |
3937 | emit_movimm(0,HOST_TEMPREG); |
82336ba3 |
3938 | emit_addpl_imm(HOST_TEMPREG,1,HOST_TEMPREG); |
3939 | emit_lslpls_imm(d2,1,d2); |
665f33e1 |
3940 | emit_jns((int)out-2*4); |
3941 | emit_movimm(1<<31,quotient); |
3942 | #endif |
57871462 |
3943 | emit_shr(quotient,HOST_TEMPREG,quotient); |
3944 | emit_cmp(remainder,d2); |
3945 | emit_subcs(remainder,d2,remainder); |
3946 | emit_adcs(quotient,quotient,quotient); |
3947 | emit_shrcc_imm(d2,1,d2); |
3948 | emit_jcc((int)out-16); // -4 |
3949 | } |
3950 | } |
3951 | else // 64-bit |
71e490c5 |
3952 | assert(0); |
57871462 |
3953 | } |
3954 | else |
3955 | { |
3956 | // Multiply by zero is zero. |
3957 | // MIPS does not have a divide by zero exception. |
3958 | // The result is undefined, we return zero. |
3959 | signed char hr=get_reg(i_regs->regmap,HIREG); |
3960 | signed char lr=get_reg(i_regs->regmap,LOREG); |
3961 | if(hr>=0) emit_zeroreg(hr); |
3962 | if(lr>=0) emit_zeroreg(lr); |
3963 | } |
3964 | } |
3965 | #define multdiv_assemble multdiv_assemble_arm |
3966 | |
e2b5e7aa |
3967 | static void do_preload_rhash(int r) { |
57871462 |
3968 | // Don't need this for ARM. On x86, this puts the value 0xf8 into the |
3969 | // register. On ARM the hash can be done with a single instruction (below) |
3970 | } |
3971 | |
e2b5e7aa |
3972 | static void do_preload_rhtbl(int ht) { |
57871462 |
3973 | emit_addimm(FP,(int)&mini_ht-(int)&dynarec_local,ht); |
3974 | } |
3975 | |
e2b5e7aa |
3976 | static void do_rhash(int rs,int rh) { |
57871462 |
3977 | emit_andimm(rs,0xf8,rh); |
3978 | } |
3979 | |
e2b5e7aa |
3980 | static void do_miniht_load(int ht,int rh) { |
57871462 |
3981 | assem_debug("ldr %s,[%s,%s]!\n",regname[rh],regname[ht],regname[rh]); |
3982 | output_w32(0xe7b00000|rd_rn_rm(rh,ht,rh)); |
3983 | } |
3984 | |
e2b5e7aa |
3985 | static void do_miniht_jump(int rs,int rh,int ht) { |
57871462 |
3986 | emit_cmp(rh,rs); |
3987 | emit_ldreq_indexed(ht,4,15); |
3988 | #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK |
3989 | emit_mov(rs,7); |
3990 | emit_jmp(jump_vaddr_reg[7]); |
3991 | #else |
3992 | emit_jmp(jump_vaddr_reg[rs]); |
3993 | #endif |
3994 | } |
3995 | |
e2b5e7aa |
3996 | static void do_miniht_insert(u_int return_address,int rt,int temp) { |
665f33e1 |
3997 | #ifndef HAVE_ARMV7 |
57871462 |
3998 | emit_movimm(return_address,rt); // PC into link register |
3999 | add_to_linker((int)out,return_address,1); |
4000 | emit_pcreladdr(temp); |
4001 | emit_writeword(rt,(int)&mini_ht[(return_address&0xFF)>>3][0]); |
4002 | emit_writeword(temp,(int)&mini_ht[(return_address&0xFF)>>3][1]); |
4003 | #else |
4004 | emit_movw(return_address&0x0000FFFF,rt); |
4005 | add_to_linker((int)out,return_address,1); |
4006 | emit_pcreladdr(temp); |
4007 | emit_writeword(temp,(int)&mini_ht[(return_address&0xFF)>>3][1]); |
4008 | emit_movt(return_address&0xFFFF0000,rt); |
4009 | emit_writeword(rt,(int)&mini_ht[(return_address&0xFF)>>3][0]); |
4010 | #endif |
4011 | } |
4012 | |
e2b5e7aa |
4013 | static void wb_valid(signed char pre[],signed char entry[],u_int dirty_pre,u_int dirty,uint64_t is32_pre,uint64_t u,uint64_t uu) |
57871462 |
4014 | { |
4015 | //if(dirty_pre==dirty) return; |
581335b0 |
4016 | int hr,reg; |
57871462 |
4017 | for(hr=0;hr<HOST_REGS;hr++) { |
4018 | if(hr!=EXCLUDE_REG) { |
4019 | reg=pre[hr]; |
4020 | if(((~u)>>(reg&63))&1) { |
f776eb14 |
4021 | if(reg>0) { |
57871462 |
4022 | if(((dirty_pre&~dirty)>>hr)&1) { |
4023 | if(reg>0&®<34) { |
4024 | emit_storereg(reg,hr); |
4025 | if( ((is32_pre&~uu)>>reg)&1 ) { |
4026 | emit_sarimm(hr,31,HOST_TEMPREG); |
4027 | emit_storereg(reg|64,HOST_TEMPREG); |
4028 | } |
4029 | } |
4030 | else if(reg>=64) { |
4031 | emit_storereg(reg,hr); |
4032 | } |
4033 | } |
4034 | } |
57871462 |
4035 | } |
4036 | } |
4037 | } |
4038 | } |
4039 | |
4040 | |
4041 | /* using strd could possibly help but you'd have to allocate registers in pairs |
e2b5e7aa |
4042 | static void wb_invalidate_arm(signed char pre[],signed char entry[],uint64_t dirty,uint64_t is32,uint64_t u,uint64_t uu) |
57871462 |
4043 | { |
4044 | int hr; |
4045 | int wrote=-1; |
4046 | for(hr=HOST_REGS-1;hr>=0;hr--) { |
4047 | if(hr!=EXCLUDE_REG) { |
4048 | if(pre[hr]!=entry[hr]) { |
4049 | if(pre[hr]>=0) { |
4050 | if((dirty>>hr)&1) { |
4051 | if(get_reg(entry,pre[hr])<0) { |
4052 | if(pre[hr]<64) { |
4053 | if(!((u>>pre[hr])&1)) { |
4054 | if(hr<10&&(~hr&1)&&(pre[hr+1]<0||wrote==hr+1)) { |
4055 | if( ((is32>>pre[hr])&1) && !((uu>>pre[hr])&1) ) { |
4056 | emit_sarimm(hr,31,hr+1); |
4057 | emit_strdreg(pre[hr],hr); |
4058 | } |
4059 | else |
4060 | emit_storereg(pre[hr],hr); |
4061 | }else{ |
4062 | emit_storereg(pre[hr],hr); |
4063 | if( ((is32>>pre[hr])&1) && !((uu>>pre[hr])&1) ) { |
4064 | emit_sarimm(hr,31,hr); |
4065 | emit_storereg(pre[hr]|64,hr); |
4066 | } |
4067 | } |
4068 | } |
4069 | }else{ |
4070 | if(!((uu>>(pre[hr]&63))&1) && !((is32>>(pre[hr]&63))&1)) { |
4071 | emit_storereg(pre[hr],hr); |
4072 | } |
4073 | } |
4074 | wrote=hr; |
4075 | } |
4076 | } |
4077 | } |
4078 | } |
4079 | } |
4080 | } |
4081 | for(hr=0;hr<HOST_REGS;hr++) { |
4082 | if(hr!=EXCLUDE_REG) { |
4083 | if(pre[hr]!=entry[hr]) { |
4084 | if(pre[hr]>=0) { |
4085 | int nr; |
4086 | if((nr=get_reg(entry,pre[hr]))>=0) { |
4087 | emit_mov(hr,nr); |
4088 | } |
4089 | } |
4090 | } |
4091 | } |
4092 | } |
4093 | } |
4094 | #define wb_invalidate wb_invalidate_arm |
4095 | */ |
4096 | |
dd3a91a1 |
4097 | // Clearing the cache is rather slow on ARM Linux, so mark the areas |
4098 | // that need to be cleared, and then only clear these areas once. |
e2b5e7aa |
4099 | static void do_clear_cache() |
dd3a91a1 |
4100 | { |
4101 | int i,j; |
4102 | for (i=0;i<(1<<(TARGET_SIZE_2-17));i++) |
4103 | { |
4104 | u_int bitmap=needs_clear_cache[i]; |
4105 | if(bitmap) { |
4106 | u_int start,end; |
9f51b4b9 |
4107 | for(j=0;j<32;j++) |
dd3a91a1 |
4108 | { |
4109 | if(bitmap&(1<<j)) { |
bdeade46 |
4110 | start=(u_int)BASE_ADDR+i*131072+j*4096; |
dd3a91a1 |
4111 | end=start+4095; |
4112 | j++; |
4113 | while(j<32) { |
4114 | if(bitmap&(1<<j)) { |
4115 | end+=4096; |
4116 | j++; |
4117 | }else{ |
4118 | __clear_cache((void *)start,(void *)end); |
4119 | break; |
4120 | } |
4121 | } |
4122 | } |
4123 | } |
4124 | needs_clear_cache[i]=0; |
4125 | } |
4126 | } |
4127 | } |
4128 | |
57871462 |
4129 | // CPU-architecture-specific initialization |
71e490c5 |
4130 | static void arch_init() { |
57871462 |
4131 | } |
b9b61529 |
4132 | |
4133 | // vim:shiftwidth=2:expandtab |