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