Colin Cross | 7bb052a | 2015-02-03 12:59:37 -0800 | [diff] [blame^] | 1 | // Copyright 2009 The Go Authors. All rights reserved. |
| 2 | // Use of this source code is governed by a BSD-style |
| 3 | // license that can be found in the LICENSE file. |
| 4 | |
| 5 | #include <u.h> |
| 6 | #include <libc.h> |
| 7 | #include "gg.h" |
| 8 | |
| 9 | /* |
| 10 | * attempt to generate 64-bit |
| 11 | * res = n |
| 12 | * return 1 on success, 0 if op not handled. |
| 13 | */ |
| 14 | void |
| 15 | cgen64(Node *n, Node *res) |
| 16 | { |
| 17 | Node t1, t2, *l, *r; |
| 18 | Node lo1, lo2, hi1, hi2; |
| 19 | Node al, ah, bl, bh, cl, ch, s, n1, creg; |
| 20 | Prog *p1, *p2, *p3, *p4, *p5, *p6; |
| 21 | |
| 22 | uint64 v; |
| 23 | |
| 24 | if(res->op != OINDREG && res->op != ONAME) { |
| 25 | dump("n", n); |
| 26 | dump("res", res); |
| 27 | fatal("cgen64 %O of %O", n->op, res->op); |
| 28 | } |
| 29 | |
| 30 | l = n->left; |
| 31 | if(!l->addable) { |
| 32 | tempname(&t1, l->type); |
| 33 | cgen(l, &t1); |
| 34 | l = &t1; |
| 35 | } |
| 36 | |
| 37 | split64(l, &lo1, &hi1); |
| 38 | switch(n->op) { |
| 39 | default: |
| 40 | fatal("cgen64 %O", n->op); |
| 41 | |
| 42 | case OMINUS: |
| 43 | split64(res, &lo2, &hi2); |
| 44 | |
| 45 | regalloc(&t1, lo1.type, N); |
| 46 | regalloc(&al, lo1.type, N); |
| 47 | regalloc(&ah, hi1.type, N); |
| 48 | |
| 49 | gins(AMOVW, &lo1, &al); |
| 50 | gins(AMOVW, &hi1, &ah); |
| 51 | |
| 52 | gmove(ncon(0), &t1); |
| 53 | p1 = gins(ASUB, &al, &t1); |
| 54 | p1->scond |= C_SBIT; |
| 55 | gins(AMOVW, &t1, &lo2); |
| 56 | |
| 57 | gmove(ncon(0), &t1); |
| 58 | gins(ASBC, &ah, &t1); |
| 59 | gins(AMOVW, &t1, &hi2); |
| 60 | |
| 61 | regfree(&t1); |
| 62 | regfree(&al); |
| 63 | regfree(&ah); |
| 64 | splitclean(); |
| 65 | splitclean(); |
| 66 | return; |
| 67 | |
| 68 | case OCOM: |
| 69 | regalloc(&t1, lo1.type, N); |
| 70 | gmove(ncon(-1), &t1); |
| 71 | |
| 72 | split64(res, &lo2, &hi2); |
| 73 | regalloc(&n1, lo1.type, N); |
| 74 | |
| 75 | gins(AMOVW, &lo1, &n1); |
| 76 | gins(AEOR, &t1, &n1); |
| 77 | gins(AMOVW, &n1, &lo2); |
| 78 | |
| 79 | gins(AMOVW, &hi1, &n1); |
| 80 | gins(AEOR, &t1, &n1); |
| 81 | gins(AMOVW, &n1, &hi2); |
| 82 | |
| 83 | regfree(&t1); |
| 84 | regfree(&n1); |
| 85 | splitclean(); |
| 86 | splitclean(); |
| 87 | return; |
| 88 | |
| 89 | case OADD: |
| 90 | case OSUB: |
| 91 | case OMUL: |
| 92 | case OLSH: |
| 93 | case ORSH: |
| 94 | case OAND: |
| 95 | case OOR: |
| 96 | case OXOR: |
| 97 | case OLROT: |
| 98 | // binary operators. |
| 99 | // common setup below. |
| 100 | break; |
| 101 | } |
| 102 | |
| 103 | // setup for binary operators |
| 104 | r = n->right; |
| 105 | if(r != N && !r->addable) { |
| 106 | tempname(&t2, r->type); |
| 107 | cgen(r, &t2); |
| 108 | r = &t2; |
| 109 | } |
| 110 | if(is64(r->type)) |
| 111 | split64(r, &lo2, &hi2); |
| 112 | |
| 113 | regalloc(&al, lo1.type, N); |
| 114 | regalloc(&ah, hi1.type, N); |
| 115 | |
| 116 | // Do op. Leave result in ah:al. |
| 117 | switch(n->op) { |
| 118 | default: |
| 119 | fatal("cgen64: not implemented: %N\n", n); |
| 120 | |
| 121 | case OADD: |
| 122 | // TODO: Constants |
| 123 | regalloc(&bl, types[TPTR32], N); |
| 124 | regalloc(&bh, types[TPTR32], N); |
| 125 | gins(AMOVW, &hi1, &ah); |
| 126 | gins(AMOVW, &lo1, &al); |
| 127 | gins(AMOVW, &hi2, &bh); |
| 128 | gins(AMOVW, &lo2, &bl); |
| 129 | p1 = gins(AADD, &bl, &al); |
| 130 | p1->scond |= C_SBIT; |
| 131 | gins(AADC, &bh, &ah); |
| 132 | regfree(&bl); |
| 133 | regfree(&bh); |
| 134 | break; |
| 135 | |
| 136 | case OSUB: |
| 137 | // TODO: Constants. |
| 138 | regalloc(&bl, types[TPTR32], N); |
| 139 | regalloc(&bh, types[TPTR32], N); |
| 140 | gins(AMOVW, &lo1, &al); |
| 141 | gins(AMOVW, &hi1, &ah); |
| 142 | gins(AMOVW, &lo2, &bl); |
| 143 | gins(AMOVW, &hi2, &bh); |
| 144 | p1 = gins(ASUB, &bl, &al); |
| 145 | p1->scond |= C_SBIT; |
| 146 | gins(ASBC, &bh, &ah); |
| 147 | regfree(&bl); |
| 148 | regfree(&bh); |
| 149 | break; |
| 150 | |
| 151 | case OMUL: |
| 152 | // TODO(kaib): this can be done with 4 regs and does not need 6 |
| 153 | regalloc(&bl, types[TPTR32], N); |
| 154 | regalloc(&bh, types[TPTR32], N); |
| 155 | regalloc(&cl, types[TPTR32], N); |
| 156 | regalloc(&ch, types[TPTR32], N); |
| 157 | |
| 158 | // load args into bh:bl and bh:bl. |
| 159 | gins(AMOVW, &hi1, &bh); |
| 160 | gins(AMOVW, &lo1, &bl); |
| 161 | gins(AMOVW, &hi2, &ch); |
| 162 | gins(AMOVW, &lo2, &cl); |
| 163 | |
| 164 | // bl * cl -> ah al |
| 165 | p1 = gins(AMULLU, N, N); |
| 166 | p1->from.type = D_REG; |
| 167 | p1->from.reg = bl.val.u.reg; |
| 168 | p1->reg = cl.val.u.reg; |
| 169 | p1->to.type = D_REGREG; |
| 170 | p1->to.reg = ah.val.u.reg; |
| 171 | p1->to.offset = al.val.u.reg; |
| 172 | //print("%P\n", p1); |
| 173 | |
| 174 | // bl * ch + ah -> ah |
| 175 | p1 = gins(AMULA, N, N); |
| 176 | p1->from.type = D_REG; |
| 177 | p1->from.reg = bl.val.u.reg; |
| 178 | p1->reg = ch.val.u.reg; |
| 179 | p1->to.type = D_REGREG2; |
| 180 | p1->to.reg = ah.val.u.reg; |
| 181 | p1->to.offset = ah.val.u.reg; |
| 182 | //print("%P\n", p1); |
| 183 | |
| 184 | // bh * cl + ah -> ah |
| 185 | p1 = gins(AMULA, N, N); |
| 186 | p1->from.type = D_REG; |
| 187 | p1->from.reg = bh.val.u.reg; |
| 188 | p1->reg = cl.val.u.reg; |
| 189 | p1->to.type = D_REGREG2; |
| 190 | p1->to.reg = ah.val.u.reg; |
| 191 | p1->to.offset = ah.val.u.reg; |
| 192 | //print("%P\n", p1); |
| 193 | |
| 194 | regfree(&bh); |
| 195 | regfree(&bl); |
| 196 | regfree(&ch); |
| 197 | regfree(&cl); |
| 198 | |
| 199 | break; |
| 200 | |
| 201 | case OLROT: |
| 202 | // We only rotate by a constant c in [0,64). |
| 203 | // if c >= 32: |
| 204 | // lo, hi = hi, lo |
| 205 | // c -= 32 |
| 206 | // if c == 0: |
| 207 | // no-op |
| 208 | // else: |
| 209 | // t = hi |
| 210 | // shld hi:lo, c |
| 211 | // shld lo:t, c |
| 212 | v = mpgetfix(r->val.u.xval); |
| 213 | regalloc(&bl, lo1.type, N); |
| 214 | regalloc(&bh, hi1.type, N); |
| 215 | if(v >= 32) { |
| 216 | // reverse during load to do the first 32 bits of rotate |
| 217 | v -= 32; |
| 218 | gins(AMOVW, &hi1, &bl); |
| 219 | gins(AMOVW, &lo1, &bh); |
| 220 | } else { |
| 221 | gins(AMOVW, &hi1, &bh); |
| 222 | gins(AMOVW, &lo1, &bl); |
| 223 | } |
| 224 | if(v == 0) { |
| 225 | gins(AMOVW, &bh, &ah); |
| 226 | gins(AMOVW, &bl, &al); |
| 227 | } else { |
| 228 | // rotate by 1 <= v <= 31 |
| 229 | // MOVW bl<<v, al |
| 230 | // MOVW bh<<v, ah |
| 231 | // OR bl>>(32-v), ah |
| 232 | // OR bh>>(32-v), al |
| 233 | gshift(AMOVW, &bl, SHIFT_LL, v, &al); |
| 234 | gshift(AMOVW, &bh, SHIFT_LL, v, &ah); |
| 235 | gshift(AORR, &bl, SHIFT_LR, 32-v, &ah); |
| 236 | gshift(AORR, &bh, SHIFT_LR, 32-v, &al); |
| 237 | } |
| 238 | regfree(&bl); |
| 239 | regfree(&bh); |
| 240 | break; |
| 241 | |
| 242 | case OLSH: |
| 243 | regalloc(&bl, lo1.type, N); |
| 244 | regalloc(&bh, hi1.type, N); |
| 245 | gins(AMOVW, &hi1, &bh); |
| 246 | gins(AMOVW, &lo1, &bl); |
| 247 | |
| 248 | if(r->op == OLITERAL) { |
| 249 | v = mpgetfix(r->val.u.xval); |
| 250 | if(v >= 64) { |
| 251 | // TODO(kaib): replace with gins(AMOVW, nodintconst(0), &al) |
| 252 | // here and below (verify it optimizes to EOR) |
| 253 | gins(AEOR, &al, &al); |
| 254 | gins(AEOR, &ah, &ah); |
| 255 | } else |
| 256 | if(v > 32) { |
| 257 | gins(AEOR, &al, &al); |
| 258 | // MOVW bl<<(v-32), ah |
| 259 | gshift(AMOVW, &bl, SHIFT_LL, (v-32), &ah); |
| 260 | } else |
| 261 | if(v == 32) { |
| 262 | gins(AEOR, &al, &al); |
| 263 | gins(AMOVW, &bl, &ah); |
| 264 | } else |
| 265 | if(v > 0) { |
| 266 | // MOVW bl<<v, al |
| 267 | gshift(AMOVW, &bl, SHIFT_LL, v, &al); |
| 268 | |
| 269 | // MOVW bh<<v, ah |
| 270 | gshift(AMOVW, &bh, SHIFT_LL, v, &ah); |
| 271 | |
| 272 | // OR bl>>(32-v), ah |
| 273 | gshift(AORR, &bl, SHIFT_LR, 32-v, &ah); |
| 274 | } else { |
| 275 | gins(AMOVW, &bl, &al); |
| 276 | gins(AMOVW, &bh, &ah); |
| 277 | } |
| 278 | goto olsh_break; |
| 279 | } |
| 280 | |
| 281 | regalloc(&s, types[TUINT32], N); |
| 282 | regalloc(&creg, types[TUINT32], N); |
| 283 | if (is64(r->type)) { |
| 284 | // shift is >= 1<<32 |
| 285 | split64(r, &cl, &ch); |
| 286 | gmove(&ch, &s); |
| 287 | gins(ATST, &s, N); |
| 288 | p6 = gbranch(ABNE, T, 0); |
| 289 | gmove(&cl, &s); |
| 290 | splitclean(); |
| 291 | } else { |
| 292 | gmove(r, &s); |
| 293 | p6 = P; |
| 294 | } |
| 295 | gins(ATST, &s, N); |
| 296 | |
| 297 | // shift == 0 |
| 298 | p1 = gins(AMOVW, &bl, &al); |
| 299 | p1->scond = C_SCOND_EQ; |
| 300 | p1 = gins(AMOVW, &bh, &ah); |
| 301 | p1->scond = C_SCOND_EQ; |
| 302 | p2 = gbranch(ABEQ, T, 0); |
| 303 | |
| 304 | // shift is < 32 |
| 305 | nodconst(&n1, types[TUINT32], 32); |
| 306 | gmove(&n1, &creg); |
| 307 | gcmp(ACMP, &s, &creg); |
| 308 | |
| 309 | // MOVW.LO bl<<s, al |
| 310 | p1 = gregshift(AMOVW, &bl, SHIFT_LL, &s, &al); |
| 311 | p1->scond = C_SCOND_LO; |
| 312 | |
| 313 | // MOVW.LO bh<<s, ah |
| 314 | p1 = gregshift(AMOVW, &bh, SHIFT_LL, &s, &ah); |
| 315 | p1->scond = C_SCOND_LO; |
| 316 | |
| 317 | // SUB.LO s, creg |
| 318 | p1 = gins(ASUB, &s, &creg); |
| 319 | p1->scond = C_SCOND_LO; |
| 320 | |
| 321 | // OR.LO bl>>creg, ah |
| 322 | p1 = gregshift(AORR, &bl, SHIFT_LR, &creg, &ah); |
| 323 | p1->scond = C_SCOND_LO; |
| 324 | |
| 325 | // BLO end |
| 326 | p3 = gbranch(ABLO, T, 0); |
| 327 | |
| 328 | // shift == 32 |
| 329 | p1 = gins(AEOR, &al, &al); |
| 330 | p1->scond = C_SCOND_EQ; |
| 331 | p1 = gins(AMOVW, &bl, &ah); |
| 332 | p1->scond = C_SCOND_EQ; |
| 333 | p4 = gbranch(ABEQ, T, 0); |
| 334 | |
| 335 | // shift is < 64 |
| 336 | nodconst(&n1, types[TUINT32], 64); |
| 337 | gmove(&n1, &creg); |
| 338 | gcmp(ACMP, &s, &creg); |
| 339 | |
| 340 | // EOR.LO al, al |
| 341 | p1 = gins(AEOR, &al, &al); |
| 342 | p1->scond = C_SCOND_LO; |
| 343 | |
| 344 | // MOVW.LO creg>>1, creg |
| 345 | p1 = gshift(AMOVW, &creg, SHIFT_LR, 1, &creg); |
| 346 | p1->scond = C_SCOND_LO; |
| 347 | |
| 348 | // SUB.LO creg, s |
| 349 | p1 = gins(ASUB, &creg, &s); |
| 350 | p1->scond = C_SCOND_LO; |
| 351 | |
| 352 | // MOVW bl<<s, ah |
| 353 | p1 = gregshift(AMOVW, &bl, SHIFT_LL, &s, &ah); |
| 354 | p1->scond = C_SCOND_LO; |
| 355 | |
| 356 | p5 = gbranch(ABLO, T, 0); |
| 357 | |
| 358 | // shift >= 64 |
| 359 | if (p6 != P) patch(p6, pc); |
| 360 | gins(AEOR, &al, &al); |
| 361 | gins(AEOR, &ah, &ah); |
| 362 | |
| 363 | patch(p2, pc); |
| 364 | patch(p3, pc); |
| 365 | patch(p4, pc); |
| 366 | patch(p5, pc); |
| 367 | regfree(&s); |
| 368 | regfree(&creg); |
| 369 | |
| 370 | olsh_break: |
| 371 | regfree(&bl); |
| 372 | regfree(&bh); |
| 373 | break; |
| 374 | |
| 375 | |
| 376 | case ORSH: |
| 377 | regalloc(&bl, lo1.type, N); |
| 378 | regalloc(&bh, hi1.type, N); |
| 379 | gins(AMOVW, &hi1, &bh); |
| 380 | gins(AMOVW, &lo1, &bl); |
| 381 | |
| 382 | if(r->op == OLITERAL) { |
| 383 | v = mpgetfix(r->val.u.xval); |
| 384 | if(v >= 64) { |
| 385 | if(bh.type->etype == TINT32) { |
| 386 | // MOVW bh->31, al |
| 387 | gshift(AMOVW, &bh, SHIFT_AR, 31, &al); |
| 388 | |
| 389 | // MOVW bh->31, ah |
| 390 | gshift(AMOVW, &bh, SHIFT_AR, 31, &ah); |
| 391 | } else { |
| 392 | gins(AEOR, &al, &al); |
| 393 | gins(AEOR, &ah, &ah); |
| 394 | } |
| 395 | } else |
| 396 | if(v > 32) { |
| 397 | if(bh.type->etype == TINT32) { |
| 398 | // MOVW bh->(v-32), al |
| 399 | gshift(AMOVW, &bh, SHIFT_AR, v-32, &al); |
| 400 | |
| 401 | // MOVW bh->31, ah |
| 402 | gshift(AMOVW, &bh, SHIFT_AR, 31, &ah); |
| 403 | } else { |
| 404 | // MOVW bh>>(v-32), al |
| 405 | gshift(AMOVW, &bh, SHIFT_LR, v-32, &al); |
| 406 | gins(AEOR, &ah, &ah); |
| 407 | } |
| 408 | } else |
| 409 | if(v == 32) { |
| 410 | gins(AMOVW, &bh, &al); |
| 411 | if(bh.type->etype == TINT32) { |
| 412 | // MOVW bh->31, ah |
| 413 | gshift(AMOVW, &bh, SHIFT_AR, 31, &ah); |
| 414 | } else { |
| 415 | gins(AEOR, &ah, &ah); |
| 416 | } |
| 417 | } else |
| 418 | if( v > 0) { |
| 419 | // MOVW bl>>v, al |
| 420 | gshift(AMOVW, &bl, SHIFT_LR, v, &al); |
| 421 | |
| 422 | // OR bh<<(32-v), al |
| 423 | gshift(AORR, &bh, SHIFT_LL, 32-v, &al); |
| 424 | |
| 425 | if(bh.type->etype == TINT32) { |
| 426 | // MOVW bh->v, ah |
| 427 | gshift(AMOVW, &bh, SHIFT_AR, v, &ah); |
| 428 | } else { |
| 429 | // MOVW bh>>v, ah |
| 430 | gshift(AMOVW, &bh, SHIFT_LR, v, &ah); |
| 431 | } |
| 432 | } else { |
| 433 | gins(AMOVW, &bl, &al); |
| 434 | gins(AMOVW, &bh, &ah); |
| 435 | } |
| 436 | goto orsh_break; |
| 437 | } |
| 438 | |
| 439 | regalloc(&s, types[TUINT32], N); |
| 440 | regalloc(&creg, types[TUINT32], N); |
| 441 | if(is64(r->type)) { |
| 442 | // shift is >= 1<<32 |
| 443 | split64(r, &cl, &ch); |
| 444 | gmove(&ch, &s); |
| 445 | gins(ATST, &s, N); |
| 446 | if(bh.type->etype == TINT32) |
| 447 | p1 = gshift(AMOVW, &bh, SHIFT_AR, 31, &ah); |
| 448 | else |
| 449 | p1 = gins(AEOR, &ah, &ah); |
| 450 | p1->scond = C_SCOND_NE; |
| 451 | p6 = gbranch(ABNE, T, 0); |
| 452 | gmove(&cl, &s); |
| 453 | splitclean(); |
| 454 | } else { |
| 455 | gmove(r, &s); |
| 456 | p6 = P; |
| 457 | } |
| 458 | gins(ATST, &s, N); |
| 459 | |
| 460 | // shift == 0 |
| 461 | p1 = gins(AMOVW, &bl, &al); |
| 462 | p1->scond = C_SCOND_EQ; |
| 463 | p1 = gins(AMOVW, &bh, &ah); |
| 464 | p1->scond = C_SCOND_EQ; |
| 465 | p2 = gbranch(ABEQ, T, 0); |
| 466 | |
| 467 | // check if shift is < 32 |
| 468 | nodconst(&n1, types[TUINT32], 32); |
| 469 | gmove(&n1, &creg); |
| 470 | gcmp(ACMP, &s, &creg); |
| 471 | |
| 472 | // MOVW.LO bl>>s, al |
| 473 | p1 = gregshift(AMOVW, &bl, SHIFT_LR, &s, &al); |
| 474 | p1->scond = C_SCOND_LO; |
| 475 | |
| 476 | // SUB.LO s,creg |
| 477 | p1 = gins(ASUB, &s, &creg); |
| 478 | p1->scond = C_SCOND_LO; |
| 479 | |
| 480 | // OR.LO bh<<(32-s), al |
| 481 | p1 = gregshift(AORR, &bh, SHIFT_LL, &creg, &al); |
| 482 | p1->scond = C_SCOND_LO; |
| 483 | |
| 484 | if(bh.type->etype == TINT32) { |
| 485 | // MOVW bh->s, ah |
| 486 | p1 = gregshift(AMOVW, &bh, SHIFT_AR, &s, &ah); |
| 487 | } else { |
| 488 | // MOVW bh>>s, ah |
| 489 | p1 = gregshift(AMOVW, &bh, SHIFT_LR, &s, &ah); |
| 490 | } |
| 491 | p1->scond = C_SCOND_LO; |
| 492 | |
| 493 | // BLO end |
| 494 | p3 = gbranch(ABLO, T, 0); |
| 495 | |
| 496 | // shift == 32 |
| 497 | p1 = gins(AMOVW, &bh, &al); |
| 498 | p1->scond = C_SCOND_EQ; |
| 499 | if(bh.type->etype == TINT32) |
| 500 | gshift(AMOVW, &bh, SHIFT_AR, 31, &ah); |
| 501 | else |
| 502 | gins(AEOR, &ah, &ah); |
| 503 | p4 = gbranch(ABEQ, T, 0); |
| 504 | |
| 505 | // check if shift is < 64 |
| 506 | nodconst(&n1, types[TUINT32], 64); |
| 507 | gmove(&n1, &creg); |
| 508 | gcmp(ACMP, &s, &creg); |
| 509 | |
| 510 | // MOVW.LO creg>>1, creg |
| 511 | p1 = gshift(AMOVW, &creg, SHIFT_LR, 1, &creg); |
| 512 | p1->scond = C_SCOND_LO; |
| 513 | |
| 514 | // SUB.LO creg, s |
| 515 | p1 = gins(ASUB, &creg, &s); |
| 516 | p1->scond = C_SCOND_LO; |
| 517 | |
| 518 | if(bh.type->etype == TINT32) { |
| 519 | // MOVW bh->(s-32), al |
| 520 | p1 = gregshift(AMOVW, &bh, SHIFT_AR, &s, &al); |
| 521 | p1->scond = C_SCOND_LO; |
| 522 | } else { |
| 523 | // MOVW bh>>(v-32), al |
| 524 | p1 = gregshift(AMOVW, &bh, SHIFT_LR, &s, &al); |
| 525 | p1->scond = C_SCOND_LO; |
| 526 | } |
| 527 | |
| 528 | // BLO end |
| 529 | p5 = gbranch(ABLO, T, 0); |
| 530 | |
| 531 | // s >= 64 |
| 532 | if(p6 != P) |
| 533 | patch(p6, pc); |
| 534 | if(bh.type->etype == TINT32) { |
| 535 | // MOVW bh->31, al |
| 536 | gshift(AMOVW, &bh, SHIFT_AR, 31, &al); |
| 537 | } else { |
| 538 | gins(AEOR, &al, &al); |
| 539 | } |
| 540 | |
| 541 | patch(p2, pc); |
| 542 | patch(p3, pc); |
| 543 | patch(p4, pc); |
| 544 | patch(p5, pc); |
| 545 | regfree(&s); |
| 546 | regfree(&creg); |
| 547 | |
| 548 | |
| 549 | orsh_break: |
| 550 | regfree(&bl); |
| 551 | regfree(&bh); |
| 552 | break; |
| 553 | |
| 554 | case OXOR: |
| 555 | case OAND: |
| 556 | case OOR: |
| 557 | // TODO(kaib): literal optimizations |
| 558 | // make constant the right side (it usually is anyway). |
| 559 | // if(lo1.op == OLITERAL) { |
| 560 | // nswap(&lo1, &lo2); |
| 561 | // nswap(&hi1, &hi2); |
| 562 | // } |
| 563 | // if(lo2.op == OLITERAL) { |
| 564 | // // special cases for constants. |
| 565 | // lv = mpgetfix(lo2.val.u.xval); |
| 566 | // hv = mpgetfix(hi2.val.u.xval); |
| 567 | // splitclean(); // right side |
| 568 | // split64(res, &lo2, &hi2); |
| 569 | // switch(n->op) { |
| 570 | // case OXOR: |
| 571 | // gmove(&lo1, &lo2); |
| 572 | // gmove(&hi1, &hi2); |
| 573 | // switch(lv) { |
| 574 | // case 0: |
| 575 | // break; |
| 576 | // case 0xffffffffu: |
| 577 | // gins(ANOTL, N, &lo2); |
| 578 | // break; |
| 579 | // default: |
| 580 | // gins(AXORL, ncon(lv), &lo2); |
| 581 | // break; |
| 582 | // } |
| 583 | // switch(hv) { |
| 584 | // case 0: |
| 585 | // break; |
| 586 | // case 0xffffffffu: |
| 587 | // gins(ANOTL, N, &hi2); |
| 588 | // break; |
| 589 | // default: |
| 590 | // gins(AXORL, ncon(hv), &hi2); |
| 591 | // break; |
| 592 | // } |
| 593 | // break; |
| 594 | |
| 595 | // case OAND: |
| 596 | // switch(lv) { |
| 597 | // case 0: |
| 598 | // gins(AMOVL, ncon(0), &lo2); |
| 599 | // break; |
| 600 | // default: |
| 601 | // gmove(&lo1, &lo2); |
| 602 | // if(lv != 0xffffffffu) |
| 603 | // gins(AANDL, ncon(lv), &lo2); |
| 604 | // break; |
| 605 | // } |
| 606 | // switch(hv) { |
| 607 | // case 0: |
| 608 | // gins(AMOVL, ncon(0), &hi2); |
| 609 | // break; |
| 610 | // default: |
| 611 | // gmove(&hi1, &hi2); |
| 612 | // if(hv != 0xffffffffu) |
| 613 | // gins(AANDL, ncon(hv), &hi2); |
| 614 | // break; |
| 615 | // } |
| 616 | // break; |
| 617 | |
| 618 | // case OOR: |
| 619 | // switch(lv) { |
| 620 | // case 0: |
| 621 | // gmove(&lo1, &lo2); |
| 622 | // break; |
| 623 | // case 0xffffffffu: |
| 624 | // gins(AMOVL, ncon(0xffffffffu), &lo2); |
| 625 | // break; |
| 626 | // default: |
| 627 | // gmove(&lo1, &lo2); |
| 628 | // gins(AORL, ncon(lv), &lo2); |
| 629 | // break; |
| 630 | // } |
| 631 | // switch(hv) { |
| 632 | // case 0: |
| 633 | // gmove(&hi1, &hi2); |
| 634 | // break; |
| 635 | // case 0xffffffffu: |
| 636 | // gins(AMOVL, ncon(0xffffffffu), &hi2); |
| 637 | // break; |
| 638 | // default: |
| 639 | // gmove(&hi1, &hi2); |
| 640 | // gins(AORL, ncon(hv), &hi2); |
| 641 | // break; |
| 642 | // } |
| 643 | // break; |
| 644 | // } |
| 645 | // splitclean(); |
| 646 | // splitclean(); |
| 647 | // goto out; |
| 648 | // } |
| 649 | regalloc(&n1, lo1.type, N); |
| 650 | gins(AMOVW, &lo1, &al); |
| 651 | gins(AMOVW, &hi1, &ah); |
| 652 | gins(AMOVW, &lo2, &n1); |
| 653 | gins(optoas(n->op, lo1.type), &n1, &al); |
| 654 | gins(AMOVW, &hi2, &n1); |
| 655 | gins(optoas(n->op, lo1.type), &n1, &ah); |
| 656 | regfree(&n1); |
| 657 | break; |
| 658 | } |
| 659 | if(is64(r->type)) |
| 660 | splitclean(); |
| 661 | splitclean(); |
| 662 | |
| 663 | split64(res, &lo1, &hi1); |
| 664 | gins(AMOVW, &al, &lo1); |
| 665 | gins(AMOVW, &ah, &hi1); |
| 666 | splitclean(); |
| 667 | |
| 668 | //out: |
| 669 | regfree(&al); |
| 670 | regfree(&ah); |
| 671 | } |
| 672 | |
| 673 | /* |
| 674 | * generate comparison of nl, nr, both 64-bit. |
| 675 | * nl is memory; nr is constant or memory. |
| 676 | */ |
| 677 | void |
| 678 | cmp64(Node *nl, Node *nr, int op, int likely, Prog *to) |
| 679 | { |
| 680 | Node lo1, hi1, lo2, hi2, r1, r2; |
| 681 | Prog *br; |
| 682 | Type *t; |
| 683 | |
| 684 | split64(nl, &lo1, &hi1); |
| 685 | split64(nr, &lo2, &hi2); |
| 686 | |
| 687 | // compare most significant word; |
| 688 | // if they differ, we're done. |
| 689 | t = hi1.type; |
| 690 | regalloc(&r1, types[TINT32], N); |
| 691 | regalloc(&r2, types[TINT32], N); |
| 692 | gins(AMOVW, &hi1, &r1); |
| 693 | gins(AMOVW, &hi2, &r2); |
| 694 | gcmp(ACMP, &r1, &r2); |
| 695 | regfree(&r1); |
| 696 | regfree(&r2); |
| 697 | |
| 698 | br = P; |
| 699 | switch(op) { |
| 700 | default: |
| 701 | fatal("cmp64 %O %T", op, t); |
| 702 | case OEQ: |
| 703 | // cmp hi |
| 704 | // bne L |
| 705 | // cmp lo |
| 706 | // beq to |
| 707 | // L: |
| 708 | br = gbranch(ABNE, T, -likely); |
| 709 | break; |
| 710 | case ONE: |
| 711 | // cmp hi |
| 712 | // bne to |
| 713 | // cmp lo |
| 714 | // bne to |
| 715 | patch(gbranch(ABNE, T, likely), to); |
| 716 | break; |
| 717 | case OGE: |
| 718 | case OGT: |
| 719 | // cmp hi |
| 720 | // bgt to |
| 721 | // blt L |
| 722 | // cmp lo |
| 723 | // bge to (or bgt to) |
| 724 | // L: |
| 725 | patch(gbranch(optoas(OGT, t), T, likely), to); |
| 726 | br = gbranch(optoas(OLT, t), T, -likely); |
| 727 | break; |
| 728 | case OLE: |
| 729 | case OLT: |
| 730 | // cmp hi |
| 731 | // blt to |
| 732 | // bgt L |
| 733 | // cmp lo |
| 734 | // ble to (or jlt to) |
| 735 | // L: |
| 736 | patch(gbranch(optoas(OLT, t), T, likely), to); |
| 737 | br = gbranch(optoas(OGT, t), T, -likely); |
| 738 | break; |
| 739 | } |
| 740 | |
| 741 | // compare least significant word |
| 742 | t = lo1.type; |
| 743 | regalloc(&r1, types[TINT32], N); |
| 744 | regalloc(&r2, types[TINT32], N); |
| 745 | gins(AMOVW, &lo1, &r1); |
| 746 | gins(AMOVW, &lo2, &r2); |
| 747 | gcmp(ACMP, &r1, &r2); |
| 748 | regfree(&r1); |
| 749 | regfree(&r2); |
| 750 | |
| 751 | // jump again |
| 752 | patch(gbranch(optoas(op, t), T, likely), to); |
| 753 | |
| 754 | // point first branch down here if appropriate |
| 755 | if(br != P) |
| 756 | patch(br, pc); |
| 757 | |
| 758 | splitclean(); |
| 759 | splitclean(); |
| 760 | } |