1 /* 2 * Optimizations for Tiny Code Generator for QEMU 3 * 4 * Copyright (c) 2010 Samsung Electronics. 5 * Contributed by Kirill Batuzov <batuzovk@ispras.ru> 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a copy 8 * of this software and associated documentation files (the "Software"), to deal 9 * in the Software without restriction, including without limitation the rights 10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 11 * copies of the Software, and to permit persons to whom the Software is 12 * furnished to do so, subject to the following conditions: 13 * 14 * The above copyright notice and this permission notice shall be included in 15 * all copies or substantial portions of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 23 * THE SOFTWARE. 24 */ 25 26 #include "qemu/osdep.h" 27 #include "qemu/int128.h" 28 #include "qemu/interval-tree.h" 29 #include "tcg/tcg-op-common.h" 30 #include "tcg-internal.h" 31 #include "tcg-has.h" 32 33 #define CASE_OP_32_64(x) \ 34 glue(glue(case INDEX_op_, x), _i32): \ 35 glue(glue(case INDEX_op_, x), _i64) 36 37 #define CASE_OP_32_64_VEC(x) \ 38 glue(glue(case INDEX_op_, x), _i32): \ 39 glue(glue(case INDEX_op_, x), _i64): \ 40 glue(glue(case INDEX_op_, x), _vec) 41 42 typedef struct MemCopyInfo { 43 IntervalTreeNode itree; 44 QSIMPLEQ_ENTRY (MemCopyInfo) next; 45 TCGTemp *ts; 46 TCGType type; 47 } MemCopyInfo; 48 49 typedef struct TempOptInfo { 50 bool is_const; 51 TCGTemp *prev_copy; 52 TCGTemp *next_copy; 53 QSIMPLEQ_HEAD(, MemCopyInfo) mem_copy; 54 uint64_t val; 55 uint64_t z_mask; /* mask bit is 0 if and only if value bit is 0 */ 56 uint64_t s_mask; /* mask bit is 1 if value bit matches msb */ 57 } TempOptInfo; 58 59 typedef struct OptContext { 60 TCGContext *tcg; 61 TCGOp *prev_mb; 62 TCGTempSet temps_used; 63 64 IntervalTreeRoot mem_copy; 65 QSIMPLEQ_HEAD(, MemCopyInfo) mem_free; 66 67 /* In flight values from optimization. */ 68 TCGType type; 69 } OptContext; 70 71 static inline TempOptInfo *ts_info(TCGTemp *ts) 72 { 73 return ts->state_ptr; 74 } 75 76 static inline TempOptInfo *arg_info(TCGArg arg) 77 { 78 return ts_info(arg_temp(arg)); 79 } 80 81 static inline bool ti_is_const(TempOptInfo *ti) 82 { 83 return ti->is_const; 84 } 85 86 static inline uint64_t ti_const_val(TempOptInfo *ti) 87 { 88 return ti->val; 89 } 90 91 static inline bool ti_is_const_val(TempOptInfo *ti, uint64_t val) 92 { 93 return ti_is_const(ti) && ti_const_val(ti) == val; 94 } 95 96 static inline bool ts_is_const(TCGTemp *ts) 97 { 98 return ti_is_const(ts_info(ts)); 99 } 100 101 static inline bool ts_is_const_val(TCGTemp *ts, uint64_t val) 102 { 103 return ti_is_const_val(ts_info(ts), val); 104 } 105 106 static inline bool arg_is_const(TCGArg arg) 107 { 108 return ts_is_const(arg_temp(arg)); 109 } 110 111 static inline bool arg_is_const_val(TCGArg arg, uint64_t val) 112 { 113 return ts_is_const_val(arg_temp(arg), val); 114 } 115 116 static inline bool ts_is_copy(TCGTemp *ts) 117 { 118 return ts_info(ts)->next_copy != ts; 119 } 120 121 static TCGTemp *cmp_better_copy(TCGTemp *a, TCGTemp *b) 122 { 123 return a->kind < b->kind ? b : a; 124 } 125 126 /* Initialize and activate a temporary. */ 127 static void init_ts_info(OptContext *ctx, TCGTemp *ts) 128 { 129 size_t idx = temp_idx(ts); 130 TempOptInfo *ti; 131 132 if (test_bit(idx, ctx->temps_used.l)) { 133 return; 134 } 135 set_bit(idx, ctx->temps_used.l); 136 137 ti = ts->state_ptr; 138 if (ti == NULL) { 139 ti = tcg_malloc(sizeof(TempOptInfo)); 140 ts->state_ptr = ti; 141 } 142 143 ti->next_copy = ts; 144 ti->prev_copy = ts; 145 QSIMPLEQ_INIT(&ti->mem_copy); 146 if (ts->kind == TEMP_CONST) { 147 ti->is_const = true; 148 ti->val = ts->val; 149 ti->z_mask = ts->val; 150 ti->s_mask = INT64_MIN >> clrsb64(ts->val); 151 } else { 152 ti->is_const = false; 153 ti->z_mask = -1; 154 ti->s_mask = 0; 155 } 156 } 157 158 static MemCopyInfo *mem_copy_first(OptContext *ctx, intptr_t s, intptr_t l) 159 { 160 IntervalTreeNode *r = interval_tree_iter_first(&ctx->mem_copy, s, l); 161 return r ? container_of(r, MemCopyInfo, itree) : NULL; 162 } 163 164 static MemCopyInfo *mem_copy_next(MemCopyInfo *mem, intptr_t s, intptr_t l) 165 { 166 IntervalTreeNode *r = interval_tree_iter_next(&mem->itree, s, l); 167 return r ? container_of(r, MemCopyInfo, itree) : NULL; 168 } 169 170 static void remove_mem_copy(OptContext *ctx, MemCopyInfo *mc) 171 { 172 TCGTemp *ts = mc->ts; 173 TempOptInfo *ti = ts_info(ts); 174 175 interval_tree_remove(&mc->itree, &ctx->mem_copy); 176 QSIMPLEQ_REMOVE(&ti->mem_copy, mc, MemCopyInfo, next); 177 QSIMPLEQ_INSERT_TAIL(&ctx->mem_free, mc, next); 178 } 179 180 static void remove_mem_copy_in(OptContext *ctx, intptr_t s, intptr_t l) 181 { 182 while (true) { 183 MemCopyInfo *mc = mem_copy_first(ctx, s, l); 184 if (!mc) { 185 break; 186 } 187 remove_mem_copy(ctx, mc); 188 } 189 } 190 191 static void remove_mem_copy_all(OptContext *ctx) 192 { 193 remove_mem_copy_in(ctx, 0, -1); 194 tcg_debug_assert(interval_tree_is_empty(&ctx->mem_copy)); 195 } 196 197 static TCGTemp *find_better_copy(TCGTemp *ts) 198 { 199 TCGTemp *i, *ret; 200 201 /* If this is already readonly, we can't do better. */ 202 if (temp_readonly(ts)) { 203 return ts; 204 } 205 206 ret = ts; 207 for (i = ts_info(ts)->next_copy; i != ts; i = ts_info(i)->next_copy) { 208 ret = cmp_better_copy(ret, i); 209 } 210 return ret; 211 } 212 213 static void move_mem_copies(TCGTemp *dst_ts, TCGTemp *src_ts) 214 { 215 TempOptInfo *si = ts_info(src_ts); 216 TempOptInfo *di = ts_info(dst_ts); 217 MemCopyInfo *mc; 218 219 QSIMPLEQ_FOREACH(mc, &si->mem_copy, next) { 220 tcg_debug_assert(mc->ts == src_ts); 221 mc->ts = dst_ts; 222 } 223 QSIMPLEQ_CONCAT(&di->mem_copy, &si->mem_copy); 224 } 225 226 /* Reset TEMP's state, possibly removing the temp for the list of copies. */ 227 static void reset_ts(OptContext *ctx, TCGTemp *ts) 228 { 229 TempOptInfo *ti = ts_info(ts); 230 TCGTemp *pts = ti->prev_copy; 231 TCGTemp *nts = ti->next_copy; 232 TempOptInfo *pi = ts_info(pts); 233 TempOptInfo *ni = ts_info(nts); 234 235 ni->prev_copy = ti->prev_copy; 236 pi->next_copy = ti->next_copy; 237 ti->next_copy = ts; 238 ti->prev_copy = ts; 239 ti->is_const = false; 240 ti->z_mask = -1; 241 ti->s_mask = 0; 242 243 if (!QSIMPLEQ_EMPTY(&ti->mem_copy)) { 244 if (ts == nts) { 245 /* Last temp copy being removed, the mem copies die. */ 246 MemCopyInfo *mc; 247 QSIMPLEQ_FOREACH(mc, &ti->mem_copy, next) { 248 interval_tree_remove(&mc->itree, &ctx->mem_copy); 249 } 250 QSIMPLEQ_CONCAT(&ctx->mem_free, &ti->mem_copy); 251 } else { 252 move_mem_copies(find_better_copy(nts), ts); 253 } 254 } 255 } 256 257 static void reset_temp(OptContext *ctx, TCGArg arg) 258 { 259 reset_ts(ctx, arg_temp(arg)); 260 } 261 262 static void record_mem_copy(OptContext *ctx, TCGType type, 263 TCGTemp *ts, intptr_t start, intptr_t last) 264 { 265 MemCopyInfo *mc; 266 TempOptInfo *ti; 267 268 mc = QSIMPLEQ_FIRST(&ctx->mem_free); 269 if (mc) { 270 QSIMPLEQ_REMOVE_HEAD(&ctx->mem_free, next); 271 } else { 272 mc = tcg_malloc(sizeof(*mc)); 273 } 274 275 memset(mc, 0, sizeof(*mc)); 276 mc->itree.start = start; 277 mc->itree.last = last; 278 mc->type = type; 279 interval_tree_insert(&mc->itree, &ctx->mem_copy); 280 281 ts = find_better_copy(ts); 282 ti = ts_info(ts); 283 mc->ts = ts; 284 QSIMPLEQ_INSERT_TAIL(&ti->mem_copy, mc, next); 285 } 286 287 static bool ts_are_copies(TCGTemp *ts1, TCGTemp *ts2) 288 { 289 TCGTemp *i; 290 291 if (ts1 == ts2) { 292 return true; 293 } 294 295 if (!ts_is_copy(ts1) || !ts_is_copy(ts2)) { 296 return false; 297 } 298 299 for (i = ts_info(ts1)->next_copy; i != ts1; i = ts_info(i)->next_copy) { 300 if (i == ts2) { 301 return true; 302 } 303 } 304 305 return false; 306 } 307 308 static bool args_are_copies(TCGArg arg1, TCGArg arg2) 309 { 310 return ts_are_copies(arg_temp(arg1), arg_temp(arg2)); 311 } 312 313 static TCGTemp *find_mem_copy_for(OptContext *ctx, TCGType type, intptr_t s) 314 { 315 MemCopyInfo *mc; 316 317 for (mc = mem_copy_first(ctx, s, s); mc; mc = mem_copy_next(mc, s, s)) { 318 if (mc->itree.start == s && mc->type == type) { 319 return find_better_copy(mc->ts); 320 } 321 } 322 return NULL; 323 } 324 325 static TCGArg arg_new_constant(OptContext *ctx, uint64_t val) 326 { 327 TCGType type = ctx->type; 328 TCGTemp *ts; 329 330 if (type == TCG_TYPE_I32) { 331 val = (int32_t)val; 332 } 333 334 ts = tcg_constant_internal(type, val); 335 init_ts_info(ctx, ts); 336 337 return temp_arg(ts); 338 } 339 340 static TCGArg arg_new_temp(OptContext *ctx) 341 { 342 TCGTemp *ts = tcg_temp_new_internal(ctx->type, TEMP_EBB); 343 init_ts_info(ctx, ts); 344 return temp_arg(ts); 345 } 346 347 static bool tcg_opt_gen_mov(OptContext *ctx, TCGOp *op, TCGArg dst, TCGArg src) 348 { 349 TCGTemp *dst_ts = arg_temp(dst); 350 TCGTemp *src_ts = arg_temp(src); 351 TempOptInfo *di; 352 TempOptInfo *si; 353 TCGOpcode new_op; 354 355 if (ts_are_copies(dst_ts, src_ts)) { 356 tcg_op_remove(ctx->tcg, op); 357 return true; 358 } 359 360 reset_ts(ctx, dst_ts); 361 di = ts_info(dst_ts); 362 si = ts_info(src_ts); 363 364 switch (ctx->type) { 365 case TCG_TYPE_I32: 366 new_op = INDEX_op_mov_i32; 367 break; 368 case TCG_TYPE_I64: 369 new_op = INDEX_op_mov_i64; 370 break; 371 case TCG_TYPE_V64: 372 case TCG_TYPE_V128: 373 case TCG_TYPE_V256: 374 /* TCGOP_TYPE and TCGOP_VECE remain unchanged. */ 375 new_op = INDEX_op_mov_vec; 376 break; 377 default: 378 g_assert_not_reached(); 379 } 380 op->opc = new_op; 381 op->args[0] = dst; 382 op->args[1] = src; 383 384 di->z_mask = si->z_mask; 385 di->s_mask = si->s_mask; 386 387 if (src_ts->type == dst_ts->type) { 388 TempOptInfo *ni = ts_info(si->next_copy); 389 390 di->next_copy = si->next_copy; 391 di->prev_copy = src_ts; 392 ni->prev_copy = dst_ts; 393 si->next_copy = dst_ts; 394 di->is_const = si->is_const; 395 di->val = si->val; 396 397 if (!QSIMPLEQ_EMPTY(&si->mem_copy) 398 && cmp_better_copy(src_ts, dst_ts) == dst_ts) { 399 move_mem_copies(dst_ts, src_ts); 400 } 401 } 402 return true; 403 } 404 405 static bool tcg_opt_gen_movi(OptContext *ctx, TCGOp *op, 406 TCGArg dst, uint64_t val) 407 { 408 /* Convert movi to mov with constant temp. */ 409 return tcg_opt_gen_mov(ctx, op, dst, arg_new_constant(ctx, val)); 410 } 411 412 static uint64_t do_constant_folding_2(TCGOpcode op, uint64_t x, uint64_t y) 413 { 414 uint64_t l64, h64; 415 416 switch (op) { 417 CASE_OP_32_64(add): 418 return x + y; 419 420 CASE_OP_32_64(sub): 421 return x - y; 422 423 CASE_OP_32_64(mul): 424 return x * y; 425 426 CASE_OP_32_64_VEC(and): 427 return x & y; 428 429 CASE_OP_32_64_VEC(or): 430 return x | y; 431 432 CASE_OP_32_64_VEC(xor): 433 return x ^ y; 434 435 case INDEX_op_shl_i32: 436 return (uint32_t)x << (y & 31); 437 438 case INDEX_op_shl_i64: 439 return (uint64_t)x << (y & 63); 440 441 case INDEX_op_shr_i32: 442 return (uint32_t)x >> (y & 31); 443 444 case INDEX_op_shr_i64: 445 return (uint64_t)x >> (y & 63); 446 447 case INDEX_op_sar_i32: 448 return (int32_t)x >> (y & 31); 449 450 case INDEX_op_sar_i64: 451 return (int64_t)x >> (y & 63); 452 453 case INDEX_op_rotr_i32: 454 return ror32(x, y & 31); 455 456 case INDEX_op_rotr_i64: 457 return ror64(x, y & 63); 458 459 case INDEX_op_rotl_i32: 460 return rol32(x, y & 31); 461 462 case INDEX_op_rotl_i64: 463 return rol64(x, y & 63); 464 465 CASE_OP_32_64_VEC(not): 466 return ~x; 467 468 CASE_OP_32_64(neg): 469 return -x; 470 471 CASE_OP_32_64_VEC(andc): 472 return x & ~y; 473 474 CASE_OP_32_64_VEC(orc): 475 return x | ~y; 476 477 CASE_OP_32_64_VEC(eqv): 478 return ~(x ^ y); 479 480 CASE_OP_32_64_VEC(nand): 481 return ~(x & y); 482 483 CASE_OP_32_64_VEC(nor): 484 return ~(x | y); 485 486 case INDEX_op_clz_i32: 487 return (uint32_t)x ? clz32(x) : y; 488 489 case INDEX_op_clz_i64: 490 return x ? clz64(x) : y; 491 492 case INDEX_op_ctz_i32: 493 return (uint32_t)x ? ctz32(x) : y; 494 495 case INDEX_op_ctz_i64: 496 return x ? ctz64(x) : y; 497 498 case INDEX_op_ctpop_i32: 499 return ctpop32(x); 500 501 case INDEX_op_ctpop_i64: 502 return ctpop64(x); 503 504 CASE_OP_32_64(ext8s): 505 return (int8_t)x; 506 507 CASE_OP_32_64(ext16s): 508 return (int16_t)x; 509 510 CASE_OP_32_64(ext8u): 511 return (uint8_t)x; 512 513 CASE_OP_32_64(ext16u): 514 return (uint16_t)x; 515 516 CASE_OP_32_64(bswap16): 517 x = bswap16(x); 518 return y & TCG_BSWAP_OS ? (int16_t)x : x; 519 520 CASE_OP_32_64(bswap32): 521 x = bswap32(x); 522 return y & TCG_BSWAP_OS ? (int32_t)x : x; 523 524 case INDEX_op_bswap64_i64: 525 return bswap64(x); 526 527 case INDEX_op_ext_i32_i64: 528 case INDEX_op_ext32s_i64: 529 return (int32_t)x; 530 531 case INDEX_op_extu_i32_i64: 532 case INDEX_op_extrl_i64_i32: 533 case INDEX_op_ext32u_i64: 534 return (uint32_t)x; 535 536 case INDEX_op_extrh_i64_i32: 537 return (uint64_t)x >> 32; 538 539 case INDEX_op_muluh_i32: 540 return ((uint64_t)(uint32_t)x * (uint32_t)y) >> 32; 541 case INDEX_op_mulsh_i32: 542 return ((int64_t)(int32_t)x * (int32_t)y) >> 32; 543 544 case INDEX_op_muluh_i64: 545 mulu64(&l64, &h64, x, y); 546 return h64; 547 case INDEX_op_mulsh_i64: 548 muls64(&l64, &h64, x, y); 549 return h64; 550 551 case INDEX_op_div_i32: 552 /* Avoid crashing on divide by zero, otherwise undefined. */ 553 return (int32_t)x / ((int32_t)y ? : 1); 554 case INDEX_op_divu_i32: 555 return (uint32_t)x / ((uint32_t)y ? : 1); 556 case INDEX_op_div_i64: 557 return (int64_t)x / ((int64_t)y ? : 1); 558 case INDEX_op_divu_i64: 559 return (uint64_t)x / ((uint64_t)y ? : 1); 560 561 case INDEX_op_rem_i32: 562 return (int32_t)x % ((int32_t)y ? : 1); 563 case INDEX_op_remu_i32: 564 return (uint32_t)x % ((uint32_t)y ? : 1); 565 case INDEX_op_rem_i64: 566 return (int64_t)x % ((int64_t)y ? : 1); 567 case INDEX_op_remu_i64: 568 return (uint64_t)x % ((uint64_t)y ? : 1); 569 570 default: 571 g_assert_not_reached(); 572 } 573 } 574 575 static uint64_t do_constant_folding(TCGOpcode op, TCGType type, 576 uint64_t x, uint64_t y) 577 { 578 uint64_t res = do_constant_folding_2(op, x, y); 579 if (type == TCG_TYPE_I32) { 580 res = (int32_t)res; 581 } 582 return res; 583 } 584 585 static bool do_constant_folding_cond_32(uint32_t x, uint32_t y, TCGCond c) 586 { 587 switch (c) { 588 case TCG_COND_EQ: 589 return x == y; 590 case TCG_COND_NE: 591 return x != y; 592 case TCG_COND_LT: 593 return (int32_t)x < (int32_t)y; 594 case TCG_COND_GE: 595 return (int32_t)x >= (int32_t)y; 596 case TCG_COND_LE: 597 return (int32_t)x <= (int32_t)y; 598 case TCG_COND_GT: 599 return (int32_t)x > (int32_t)y; 600 case TCG_COND_LTU: 601 return x < y; 602 case TCG_COND_GEU: 603 return x >= y; 604 case TCG_COND_LEU: 605 return x <= y; 606 case TCG_COND_GTU: 607 return x > y; 608 case TCG_COND_TSTEQ: 609 return (x & y) == 0; 610 case TCG_COND_TSTNE: 611 return (x & y) != 0; 612 case TCG_COND_ALWAYS: 613 case TCG_COND_NEVER: 614 break; 615 } 616 g_assert_not_reached(); 617 } 618 619 static bool do_constant_folding_cond_64(uint64_t x, uint64_t y, TCGCond c) 620 { 621 switch (c) { 622 case TCG_COND_EQ: 623 return x == y; 624 case TCG_COND_NE: 625 return x != y; 626 case TCG_COND_LT: 627 return (int64_t)x < (int64_t)y; 628 case TCG_COND_GE: 629 return (int64_t)x >= (int64_t)y; 630 case TCG_COND_LE: 631 return (int64_t)x <= (int64_t)y; 632 case TCG_COND_GT: 633 return (int64_t)x > (int64_t)y; 634 case TCG_COND_LTU: 635 return x < y; 636 case TCG_COND_GEU: 637 return x >= y; 638 case TCG_COND_LEU: 639 return x <= y; 640 case TCG_COND_GTU: 641 return x > y; 642 case TCG_COND_TSTEQ: 643 return (x & y) == 0; 644 case TCG_COND_TSTNE: 645 return (x & y) != 0; 646 case TCG_COND_ALWAYS: 647 case TCG_COND_NEVER: 648 break; 649 } 650 g_assert_not_reached(); 651 } 652 653 static int do_constant_folding_cond_eq(TCGCond c) 654 { 655 switch (c) { 656 case TCG_COND_GT: 657 case TCG_COND_LTU: 658 case TCG_COND_LT: 659 case TCG_COND_GTU: 660 case TCG_COND_NE: 661 return 0; 662 case TCG_COND_GE: 663 case TCG_COND_GEU: 664 case TCG_COND_LE: 665 case TCG_COND_LEU: 666 case TCG_COND_EQ: 667 return 1; 668 case TCG_COND_TSTEQ: 669 case TCG_COND_TSTNE: 670 return -1; 671 case TCG_COND_ALWAYS: 672 case TCG_COND_NEVER: 673 break; 674 } 675 g_assert_not_reached(); 676 } 677 678 /* 679 * Return -1 if the condition can't be simplified, 680 * and the result of the condition (0 or 1) if it can. 681 */ 682 static int do_constant_folding_cond(TCGType type, TCGArg x, 683 TCGArg y, TCGCond c) 684 { 685 if (arg_is_const(x) && arg_is_const(y)) { 686 uint64_t xv = arg_info(x)->val; 687 uint64_t yv = arg_info(y)->val; 688 689 switch (type) { 690 case TCG_TYPE_I32: 691 return do_constant_folding_cond_32(xv, yv, c); 692 case TCG_TYPE_I64: 693 return do_constant_folding_cond_64(xv, yv, c); 694 default: 695 /* Only scalar comparisons are optimizable */ 696 return -1; 697 } 698 } else if (args_are_copies(x, y)) { 699 return do_constant_folding_cond_eq(c); 700 } else if (arg_is_const_val(y, 0)) { 701 switch (c) { 702 case TCG_COND_LTU: 703 case TCG_COND_TSTNE: 704 return 0; 705 case TCG_COND_GEU: 706 case TCG_COND_TSTEQ: 707 return 1; 708 default: 709 return -1; 710 } 711 } 712 return -1; 713 } 714 715 /** 716 * swap_commutative: 717 * @dest: TCGArg of the destination argument, or NO_DEST. 718 * @p1: first paired argument 719 * @p2: second paired argument 720 * 721 * If *@p1 is a constant and *@p2 is not, swap. 722 * If *@p2 matches @dest, swap. 723 * Return true if a swap was performed. 724 */ 725 726 #define NO_DEST temp_arg(NULL) 727 728 static bool swap_commutative(TCGArg dest, TCGArg *p1, TCGArg *p2) 729 { 730 TCGArg a1 = *p1, a2 = *p2; 731 int sum = 0; 732 sum += arg_is_const(a1); 733 sum -= arg_is_const(a2); 734 735 /* Prefer the constant in second argument, and then the form 736 op a, a, b, which is better handled on non-RISC hosts. */ 737 if (sum > 0 || (sum == 0 && dest == a2)) { 738 *p1 = a2; 739 *p2 = a1; 740 return true; 741 } 742 return false; 743 } 744 745 static bool swap_commutative2(TCGArg *p1, TCGArg *p2) 746 { 747 int sum = 0; 748 sum += arg_is_const(p1[0]); 749 sum += arg_is_const(p1[1]); 750 sum -= arg_is_const(p2[0]); 751 sum -= arg_is_const(p2[1]); 752 if (sum > 0) { 753 TCGArg t; 754 t = p1[0], p1[0] = p2[0], p2[0] = t; 755 t = p1[1], p1[1] = p2[1], p2[1] = t; 756 return true; 757 } 758 return false; 759 } 760 761 /* 762 * Return -1 if the condition can't be simplified, 763 * and the result of the condition (0 or 1) if it can. 764 */ 765 static int do_constant_folding_cond1(OptContext *ctx, TCGOp *op, TCGArg dest, 766 TCGArg *p1, TCGArg *p2, TCGArg *pcond) 767 { 768 TCGCond cond; 769 TempOptInfo *i1; 770 bool swap; 771 int r; 772 773 swap = swap_commutative(dest, p1, p2); 774 cond = *pcond; 775 if (swap) { 776 *pcond = cond = tcg_swap_cond(cond); 777 } 778 779 r = do_constant_folding_cond(ctx->type, *p1, *p2, cond); 780 if (r >= 0) { 781 return r; 782 } 783 if (!is_tst_cond(cond)) { 784 return -1; 785 } 786 787 i1 = arg_info(*p1); 788 789 /* 790 * TSTNE x,x -> NE x,0 791 * TSTNE x,i -> NE x,0 if i includes all nonzero bits of x 792 */ 793 if (args_are_copies(*p1, *p2) || 794 (arg_is_const(*p2) && (i1->z_mask & ~arg_info(*p2)->val) == 0)) { 795 *p2 = arg_new_constant(ctx, 0); 796 *pcond = tcg_tst_eqne_cond(cond); 797 return -1; 798 } 799 800 /* TSTNE x,i -> LT x,0 if i only includes sign bit copies */ 801 if (arg_is_const(*p2) && (arg_info(*p2)->val & ~i1->s_mask) == 0) { 802 *p2 = arg_new_constant(ctx, 0); 803 *pcond = tcg_tst_ltge_cond(cond); 804 return -1; 805 } 806 807 /* Expand to AND with a temporary if no backend support. */ 808 if (!TCG_TARGET_HAS_tst) { 809 TCGOpcode and_opc = (ctx->type == TCG_TYPE_I32 810 ? INDEX_op_and_i32 : INDEX_op_and_i64); 811 TCGOp *op2 = tcg_op_insert_before(ctx->tcg, op, and_opc, 3); 812 TCGArg tmp = arg_new_temp(ctx); 813 814 op2->args[0] = tmp; 815 op2->args[1] = *p1; 816 op2->args[2] = *p2; 817 818 *p1 = tmp; 819 *p2 = arg_new_constant(ctx, 0); 820 *pcond = tcg_tst_eqne_cond(cond); 821 } 822 return -1; 823 } 824 825 static int do_constant_folding_cond2(OptContext *ctx, TCGOp *op, TCGArg *args) 826 { 827 TCGArg al, ah, bl, bh; 828 TCGCond c; 829 bool swap; 830 int r; 831 832 swap = swap_commutative2(args, args + 2); 833 c = args[4]; 834 if (swap) { 835 args[4] = c = tcg_swap_cond(c); 836 } 837 838 al = args[0]; 839 ah = args[1]; 840 bl = args[2]; 841 bh = args[3]; 842 843 if (arg_is_const(bl) && arg_is_const(bh)) { 844 tcg_target_ulong blv = arg_info(bl)->val; 845 tcg_target_ulong bhv = arg_info(bh)->val; 846 uint64_t b = deposit64(blv, 32, 32, bhv); 847 848 if (arg_is_const(al) && arg_is_const(ah)) { 849 tcg_target_ulong alv = arg_info(al)->val; 850 tcg_target_ulong ahv = arg_info(ah)->val; 851 uint64_t a = deposit64(alv, 32, 32, ahv); 852 853 r = do_constant_folding_cond_64(a, b, c); 854 if (r >= 0) { 855 return r; 856 } 857 } 858 859 if (b == 0) { 860 switch (c) { 861 case TCG_COND_LTU: 862 case TCG_COND_TSTNE: 863 return 0; 864 case TCG_COND_GEU: 865 case TCG_COND_TSTEQ: 866 return 1; 867 default: 868 break; 869 } 870 } 871 872 /* TSTNE x,-1 -> NE x,0 */ 873 if (b == -1 && is_tst_cond(c)) { 874 args[3] = args[2] = arg_new_constant(ctx, 0); 875 args[4] = tcg_tst_eqne_cond(c); 876 return -1; 877 } 878 879 /* TSTNE x,sign -> LT x,0 */ 880 if (b == INT64_MIN && is_tst_cond(c)) { 881 /* bl must be 0, so copy that to bh */ 882 args[3] = bl; 883 args[4] = tcg_tst_ltge_cond(c); 884 return -1; 885 } 886 } 887 888 if (args_are_copies(al, bl) && args_are_copies(ah, bh)) { 889 r = do_constant_folding_cond_eq(c); 890 if (r >= 0) { 891 return r; 892 } 893 894 /* TSTNE x,x -> NE x,0 */ 895 if (is_tst_cond(c)) { 896 args[3] = args[2] = arg_new_constant(ctx, 0); 897 args[4] = tcg_tst_eqne_cond(c); 898 return -1; 899 } 900 } 901 902 /* Expand to AND with a temporary if no backend support. */ 903 if (!TCG_TARGET_HAS_tst && is_tst_cond(c)) { 904 TCGOp *op1 = tcg_op_insert_before(ctx->tcg, op, INDEX_op_and_i32, 3); 905 TCGOp *op2 = tcg_op_insert_before(ctx->tcg, op, INDEX_op_and_i32, 3); 906 TCGArg t1 = arg_new_temp(ctx); 907 TCGArg t2 = arg_new_temp(ctx); 908 909 op1->args[0] = t1; 910 op1->args[1] = al; 911 op1->args[2] = bl; 912 op2->args[0] = t2; 913 op2->args[1] = ah; 914 op2->args[2] = bh; 915 916 args[0] = t1; 917 args[1] = t2; 918 args[3] = args[2] = arg_new_constant(ctx, 0); 919 args[4] = tcg_tst_eqne_cond(c); 920 } 921 return -1; 922 } 923 924 static void init_arguments(OptContext *ctx, TCGOp *op, int nb_args) 925 { 926 for (int i = 0; i < nb_args; i++) { 927 TCGTemp *ts = arg_temp(op->args[i]); 928 init_ts_info(ctx, ts); 929 } 930 } 931 932 static void copy_propagate(OptContext *ctx, TCGOp *op, 933 int nb_oargs, int nb_iargs) 934 { 935 for (int i = nb_oargs; i < nb_oargs + nb_iargs; i++) { 936 TCGTemp *ts = arg_temp(op->args[i]); 937 if (ts_is_copy(ts)) { 938 op->args[i] = temp_arg(find_better_copy(ts)); 939 } 940 } 941 } 942 943 static void finish_bb(OptContext *ctx) 944 { 945 /* We only optimize memory barriers across basic blocks. */ 946 ctx->prev_mb = NULL; 947 } 948 949 static void finish_ebb(OptContext *ctx) 950 { 951 finish_bb(ctx); 952 /* We only optimize across extended basic blocks. */ 953 memset(&ctx->temps_used, 0, sizeof(ctx->temps_used)); 954 remove_mem_copy_all(ctx); 955 } 956 957 static bool finish_folding(OptContext *ctx, TCGOp *op) 958 { 959 const TCGOpDef *def = &tcg_op_defs[op->opc]; 960 int i, nb_oargs; 961 962 nb_oargs = def->nb_oargs; 963 for (i = 0; i < nb_oargs; i++) { 964 TCGTemp *ts = arg_temp(op->args[i]); 965 reset_ts(ctx, ts); 966 } 967 return true; 968 } 969 970 /* 971 * The fold_* functions return true when processing is complete, 972 * usually by folding the operation to a constant or to a copy, 973 * and calling tcg_opt_gen_{mov,movi}. They may do other things, 974 * like collect information about the value produced, for use in 975 * optimizing a subsequent operation. 976 * 977 * These first fold_* functions are all helpers, used by other 978 * folders for more specific operations. 979 */ 980 981 static bool fold_const1(OptContext *ctx, TCGOp *op) 982 { 983 if (arg_is_const(op->args[1])) { 984 uint64_t t; 985 986 t = arg_info(op->args[1])->val; 987 t = do_constant_folding(op->opc, ctx->type, t, 0); 988 return tcg_opt_gen_movi(ctx, op, op->args[0], t); 989 } 990 return false; 991 } 992 993 static bool fold_const2(OptContext *ctx, TCGOp *op) 994 { 995 if (arg_is_const(op->args[1]) && arg_is_const(op->args[2])) { 996 uint64_t t1 = arg_info(op->args[1])->val; 997 uint64_t t2 = arg_info(op->args[2])->val; 998 999 t1 = do_constant_folding(op->opc, ctx->type, t1, t2); 1000 return tcg_opt_gen_movi(ctx, op, op->args[0], t1); 1001 } 1002 return false; 1003 } 1004 1005 static bool fold_commutative(OptContext *ctx, TCGOp *op) 1006 { 1007 swap_commutative(op->args[0], &op->args[1], &op->args[2]); 1008 return false; 1009 } 1010 1011 static bool fold_const2_commutative(OptContext *ctx, TCGOp *op) 1012 { 1013 swap_commutative(op->args[0], &op->args[1], &op->args[2]); 1014 return fold_const2(ctx, op); 1015 } 1016 1017 /* 1018 * Record "zero" and "sign" masks for the single output of @op. 1019 * See TempOptInfo definition of z_mask and s_mask. 1020 * If z_mask allows, fold the output to constant zero. 1021 * The passed s_mask may be augmented by z_mask. 1022 */ 1023 static bool fold_masks_zs(OptContext *ctx, TCGOp *op, 1024 uint64_t z_mask, int64_t s_mask) 1025 { 1026 const TCGOpDef *def = &tcg_op_defs[op->opc]; 1027 TCGTemp *ts; 1028 TempOptInfo *ti; 1029 int rep; 1030 1031 /* Only single-output opcodes are supported here. */ 1032 tcg_debug_assert(def->nb_oargs == 1); 1033 1034 /* 1035 * 32-bit ops generate 32-bit results, which for the purpose of 1036 * simplifying tcg are sign-extended. Certainly that's how we 1037 * represent our constants elsewhere. Note that the bits will 1038 * be reset properly for a 64-bit value when encountering the 1039 * type changing opcodes. 1040 */ 1041 if (ctx->type == TCG_TYPE_I32) { 1042 z_mask = (int32_t)z_mask; 1043 s_mask |= INT32_MIN; 1044 } 1045 1046 if (z_mask == 0) { 1047 return tcg_opt_gen_movi(ctx, op, op->args[0], 0); 1048 } 1049 1050 ts = arg_temp(op->args[0]); 1051 reset_ts(ctx, ts); 1052 1053 ti = ts_info(ts); 1054 ti->z_mask = z_mask; 1055 1056 /* Canonicalize s_mask and incorporate data from z_mask. */ 1057 rep = clz64(~s_mask); 1058 rep = MAX(rep, clz64(z_mask)); 1059 rep = MAX(rep - 1, 0); 1060 ti->s_mask = INT64_MIN >> rep; 1061 1062 return true; 1063 } 1064 1065 static bool fold_masks_z(OptContext *ctx, TCGOp *op, uint64_t z_mask) 1066 { 1067 return fold_masks_zs(ctx, op, z_mask, 0); 1068 } 1069 1070 static bool fold_masks_s(OptContext *ctx, TCGOp *op, uint64_t s_mask) 1071 { 1072 return fold_masks_zs(ctx, op, -1, s_mask); 1073 } 1074 1075 /* 1076 * An "affected" mask bit is 0 if and only if the result is identical 1077 * to the first input. Thus if the entire mask is 0, the operation 1078 * is equivalent to a copy. 1079 */ 1080 static bool fold_affected_mask(OptContext *ctx, TCGOp *op, uint64_t a_mask) 1081 { 1082 if (ctx->type == TCG_TYPE_I32) { 1083 a_mask = (uint32_t)a_mask; 1084 } 1085 if (a_mask == 0) { 1086 return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[1]); 1087 } 1088 return false; 1089 } 1090 1091 /* 1092 * Convert @op to NOT, if NOT is supported by the host. 1093 * Return true f the conversion is successful, which will still 1094 * indicate that the processing is complete. 1095 */ 1096 static bool fold_not(OptContext *ctx, TCGOp *op); 1097 static bool fold_to_not(OptContext *ctx, TCGOp *op, int idx) 1098 { 1099 TCGOpcode not_op; 1100 bool have_not; 1101 1102 switch (ctx->type) { 1103 case TCG_TYPE_I32: 1104 not_op = INDEX_op_not_i32; 1105 have_not = TCG_TARGET_HAS_not_i32; 1106 break; 1107 case TCG_TYPE_I64: 1108 not_op = INDEX_op_not_i64; 1109 have_not = TCG_TARGET_HAS_not_i64; 1110 break; 1111 case TCG_TYPE_V64: 1112 case TCG_TYPE_V128: 1113 case TCG_TYPE_V256: 1114 not_op = INDEX_op_not_vec; 1115 have_not = TCG_TARGET_HAS_not_vec; 1116 break; 1117 default: 1118 g_assert_not_reached(); 1119 } 1120 if (have_not) { 1121 op->opc = not_op; 1122 op->args[1] = op->args[idx]; 1123 return fold_not(ctx, op); 1124 } 1125 return false; 1126 } 1127 1128 /* If the binary operation has first argument @i, fold to @i. */ 1129 static bool fold_ix_to_i(OptContext *ctx, TCGOp *op, uint64_t i) 1130 { 1131 if (arg_is_const_val(op->args[1], i)) { 1132 return tcg_opt_gen_movi(ctx, op, op->args[0], i); 1133 } 1134 return false; 1135 } 1136 1137 /* If the binary operation has first argument @i, fold to NOT. */ 1138 static bool fold_ix_to_not(OptContext *ctx, TCGOp *op, uint64_t i) 1139 { 1140 if (arg_is_const_val(op->args[1], i)) { 1141 return fold_to_not(ctx, op, 2); 1142 } 1143 return false; 1144 } 1145 1146 /* If the binary operation has second argument @i, fold to @i. */ 1147 static bool fold_xi_to_i(OptContext *ctx, TCGOp *op, uint64_t i) 1148 { 1149 if (arg_is_const_val(op->args[2], i)) { 1150 return tcg_opt_gen_movi(ctx, op, op->args[0], i); 1151 } 1152 return false; 1153 } 1154 1155 /* If the binary operation has second argument @i, fold to identity. */ 1156 static bool fold_xi_to_x(OptContext *ctx, TCGOp *op, uint64_t i) 1157 { 1158 if (arg_is_const_val(op->args[2], i)) { 1159 return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[1]); 1160 } 1161 return false; 1162 } 1163 1164 /* If the binary operation has second argument @i, fold to NOT. */ 1165 static bool fold_xi_to_not(OptContext *ctx, TCGOp *op, uint64_t i) 1166 { 1167 if (arg_is_const_val(op->args[2], i)) { 1168 return fold_to_not(ctx, op, 1); 1169 } 1170 return false; 1171 } 1172 1173 /* If the binary operation has both arguments equal, fold to @i. */ 1174 static bool fold_xx_to_i(OptContext *ctx, TCGOp *op, uint64_t i) 1175 { 1176 if (args_are_copies(op->args[1], op->args[2])) { 1177 return tcg_opt_gen_movi(ctx, op, op->args[0], i); 1178 } 1179 return false; 1180 } 1181 1182 /* If the binary operation has both arguments equal, fold to identity. */ 1183 static bool fold_xx_to_x(OptContext *ctx, TCGOp *op) 1184 { 1185 if (args_are_copies(op->args[1], op->args[2])) { 1186 return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[1]); 1187 } 1188 return false; 1189 } 1190 1191 /* 1192 * These outermost fold_<op> functions are sorted alphabetically. 1193 * 1194 * The ordering of the transformations should be: 1195 * 1) those that produce a constant 1196 * 2) those that produce a copy 1197 * 3) those that produce information about the result value. 1198 */ 1199 1200 static bool fold_or(OptContext *ctx, TCGOp *op); 1201 static bool fold_orc(OptContext *ctx, TCGOp *op); 1202 static bool fold_xor(OptContext *ctx, TCGOp *op); 1203 1204 static bool fold_add(OptContext *ctx, TCGOp *op) 1205 { 1206 if (fold_const2_commutative(ctx, op) || 1207 fold_xi_to_x(ctx, op, 0)) { 1208 return true; 1209 } 1210 return finish_folding(ctx, op); 1211 } 1212 1213 /* We cannot as yet do_constant_folding with vectors. */ 1214 static bool fold_add_vec(OptContext *ctx, TCGOp *op) 1215 { 1216 if (fold_commutative(ctx, op) || 1217 fold_xi_to_x(ctx, op, 0)) { 1218 return true; 1219 } 1220 return finish_folding(ctx, op); 1221 } 1222 1223 static bool fold_addsub2(OptContext *ctx, TCGOp *op, bool add) 1224 { 1225 bool a_const = arg_is_const(op->args[2]) && arg_is_const(op->args[3]); 1226 bool b_const = arg_is_const(op->args[4]) && arg_is_const(op->args[5]); 1227 1228 if (a_const && b_const) { 1229 uint64_t al = arg_info(op->args[2])->val; 1230 uint64_t ah = arg_info(op->args[3])->val; 1231 uint64_t bl = arg_info(op->args[4])->val; 1232 uint64_t bh = arg_info(op->args[5])->val; 1233 TCGArg rl, rh; 1234 TCGOp *op2; 1235 1236 if (ctx->type == TCG_TYPE_I32) { 1237 uint64_t a = deposit64(al, 32, 32, ah); 1238 uint64_t b = deposit64(bl, 32, 32, bh); 1239 1240 if (add) { 1241 a += b; 1242 } else { 1243 a -= b; 1244 } 1245 1246 al = sextract64(a, 0, 32); 1247 ah = sextract64(a, 32, 32); 1248 } else { 1249 Int128 a = int128_make128(al, ah); 1250 Int128 b = int128_make128(bl, bh); 1251 1252 if (add) { 1253 a = int128_add(a, b); 1254 } else { 1255 a = int128_sub(a, b); 1256 } 1257 1258 al = int128_getlo(a); 1259 ah = int128_gethi(a); 1260 } 1261 1262 rl = op->args[0]; 1263 rh = op->args[1]; 1264 1265 /* The proper opcode is supplied by tcg_opt_gen_mov. */ 1266 op2 = tcg_op_insert_before(ctx->tcg, op, 0, 2); 1267 1268 tcg_opt_gen_movi(ctx, op, rl, al); 1269 tcg_opt_gen_movi(ctx, op2, rh, ah); 1270 return true; 1271 } 1272 1273 /* Fold sub2 r,x,i to add2 r,x,-i */ 1274 if (!add && b_const) { 1275 uint64_t bl = arg_info(op->args[4])->val; 1276 uint64_t bh = arg_info(op->args[5])->val; 1277 1278 /* Negate the two parts without assembling and disassembling. */ 1279 bl = -bl; 1280 bh = ~bh + !bl; 1281 1282 op->opc = (ctx->type == TCG_TYPE_I32 1283 ? INDEX_op_add2_i32 : INDEX_op_add2_i64); 1284 op->args[4] = arg_new_constant(ctx, bl); 1285 op->args[5] = arg_new_constant(ctx, bh); 1286 } 1287 return finish_folding(ctx, op); 1288 } 1289 1290 static bool fold_add2(OptContext *ctx, TCGOp *op) 1291 { 1292 /* Note that the high and low parts may be independently swapped. */ 1293 swap_commutative(op->args[0], &op->args[2], &op->args[4]); 1294 swap_commutative(op->args[1], &op->args[3], &op->args[5]); 1295 1296 return fold_addsub2(ctx, op, true); 1297 } 1298 1299 static bool fold_and(OptContext *ctx, TCGOp *op) 1300 { 1301 uint64_t z1, z2, z_mask, s_mask; 1302 TempOptInfo *t1, *t2; 1303 1304 if (fold_const2_commutative(ctx, op) || 1305 fold_xi_to_i(ctx, op, 0) || 1306 fold_xi_to_x(ctx, op, -1) || 1307 fold_xx_to_x(ctx, op)) { 1308 return true; 1309 } 1310 1311 t1 = arg_info(op->args[1]); 1312 t2 = arg_info(op->args[2]); 1313 z1 = t1->z_mask; 1314 z2 = t2->z_mask; 1315 1316 /* 1317 * Known-zeros does not imply known-ones. Therefore unless 1318 * arg2 is constant, we can't infer affected bits from it. 1319 */ 1320 if (ti_is_const(t2) && fold_affected_mask(ctx, op, z1 & ~z2)) { 1321 return true; 1322 } 1323 1324 z_mask = z1 & z2; 1325 1326 /* 1327 * Sign repetitions are perforce all identical, whether they are 1 or 0. 1328 * Bitwise operations preserve the relative quantity of the repetitions. 1329 */ 1330 s_mask = t1->s_mask & t2->s_mask; 1331 1332 return fold_masks_zs(ctx, op, z_mask, s_mask); 1333 } 1334 1335 static bool fold_andc(OptContext *ctx, TCGOp *op) 1336 { 1337 uint64_t z_mask, s_mask; 1338 TempOptInfo *t1, *t2; 1339 1340 if (fold_const2(ctx, op) || 1341 fold_xx_to_i(ctx, op, 0) || 1342 fold_xi_to_x(ctx, op, 0) || 1343 fold_ix_to_not(ctx, op, -1)) { 1344 return true; 1345 } 1346 1347 t1 = arg_info(op->args[1]); 1348 t2 = arg_info(op->args[2]); 1349 z_mask = t1->z_mask; 1350 1351 /* 1352 * Known-zeros does not imply known-ones. Therefore unless 1353 * arg2 is constant, we can't infer anything from it. 1354 */ 1355 if (ti_is_const(t2)) { 1356 uint64_t v2 = ti_const_val(t2); 1357 if (fold_affected_mask(ctx, op, z_mask & v2)) { 1358 return true; 1359 } 1360 z_mask &= ~v2; 1361 } 1362 1363 s_mask = t1->s_mask & t2->s_mask; 1364 return fold_masks_zs(ctx, op, z_mask, s_mask); 1365 } 1366 1367 static bool fold_bitsel_vec(OptContext *ctx, TCGOp *op) 1368 { 1369 /* If true and false values are the same, eliminate the cmp. */ 1370 if (args_are_copies(op->args[2], op->args[3])) { 1371 return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[2]); 1372 } 1373 1374 if (arg_is_const(op->args[2]) && arg_is_const(op->args[3])) { 1375 uint64_t tv = arg_info(op->args[2])->val; 1376 uint64_t fv = arg_info(op->args[3])->val; 1377 1378 if (tv == -1 && fv == 0) { 1379 return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[1]); 1380 } 1381 if (tv == 0 && fv == -1) { 1382 if (TCG_TARGET_HAS_not_vec) { 1383 op->opc = INDEX_op_not_vec; 1384 return fold_not(ctx, op); 1385 } else { 1386 op->opc = INDEX_op_xor_vec; 1387 op->args[2] = arg_new_constant(ctx, -1); 1388 return fold_xor(ctx, op); 1389 } 1390 } 1391 } 1392 if (arg_is_const(op->args[2])) { 1393 uint64_t tv = arg_info(op->args[2])->val; 1394 if (tv == -1) { 1395 op->opc = INDEX_op_or_vec; 1396 op->args[2] = op->args[3]; 1397 return fold_or(ctx, op); 1398 } 1399 if (tv == 0 && TCG_TARGET_HAS_andc_vec) { 1400 op->opc = INDEX_op_andc_vec; 1401 op->args[2] = op->args[1]; 1402 op->args[1] = op->args[3]; 1403 return fold_andc(ctx, op); 1404 } 1405 } 1406 if (arg_is_const(op->args[3])) { 1407 uint64_t fv = arg_info(op->args[3])->val; 1408 if (fv == 0) { 1409 op->opc = INDEX_op_and_vec; 1410 return fold_and(ctx, op); 1411 } 1412 if (fv == -1 && TCG_TARGET_HAS_orc_vec) { 1413 op->opc = INDEX_op_orc_vec; 1414 op->args[2] = op->args[1]; 1415 op->args[1] = op->args[3]; 1416 return fold_orc(ctx, op); 1417 } 1418 } 1419 return finish_folding(ctx, op); 1420 } 1421 1422 static bool fold_brcond(OptContext *ctx, TCGOp *op) 1423 { 1424 int i = do_constant_folding_cond1(ctx, op, NO_DEST, &op->args[0], 1425 &op->args[1], &op->args[2]); 1426 if (i == 0) { 1427 tcg_op_remove(ctx->tcg, op); 1428 return true; 1429 } 1430 if (i > 0) { 1431 op->opc = INDEX_op_br; 1432 op->args[0] = op->args[3]; 1433 finish_ebb(ctx); 1434 } else { 1435 finish_bb(ctx); 1436 } 1437 return true; 1438 } 1439 1440 static bool fold_brcond2(OptContext *ctx, TCGOp *op) 1441 { 1442 TCGCond cond; 1443 TCGArg label; 1444 int i, inv = 0; 1445 1446 i = do_constant_folding_cond2(ctx, op, &op->args[0]); 1447 cond = op->args[4]; 1448 label = op->args[5]; 1449 if (i >= 0) { 1450 goto do_brcond_const; 1451 } 1452 1453 switch (cond) { 1454 case TCG_COND_LT: 1455 case TCG_COND_GE: 1456 /* 1457 * Simplify LT/GE comparisons vs zero to a single compare 1458 * vs the high word of the input. 1459 */ 1460 if (arg_is_const_val(op->args[2], 0) && 1461 arg_is_const_val(op->args[3], 0)) { 1462 goto do_brcond_high; 1463 } 1464 break; 1465 1466 case TCG_COND_NE: 1467 inv = 1; 1468 QEMU_FALLTHROUGH; 1469 case TCG_COND_EQ: 1470 /* 1471 * Simplify EQ/NE comparisons where one of the pairs 1472 * can be simplified. 1473 */ 1474 i = do_constant_folding_cond(TCG_TYPE_I32, op->args[0], 1475 op->args[2], cond); 1476 switch (i ^ inv) { 1477 case 0: 1478 goto do_brcond_const; 1479 case 1: 1480 goto do_brcond_high; 1481 } 1482 1483 i = do_constant_folding_cond(TCG_TYPE_I32, op->args[1], 1484 op->args[3], cond); 1485 switch (i ^ inv) { 1486 case 0: 1487 goto do_brcond_const; 1488 case 1: 1489 goto do_brcond_low; 1490 } 1491 break; 1492 1493 case TCG_COND_TSTEQ: 1494 case TCG_COND_TSTNE: 1495 if (arg_is_const_val(op->args[2], 0)) { 1496 goto do_brcond_high; 1497 } 1498 if (arg_is_const_val(op->args[3], 0)) { 1499 goto do_brcond_low; 1500 } 1501 break; 1502 1503 default: 1504 break; 1505 1506 do_brcond_low: 1507 op->opc = INDEX_op_brcond_i32; 1508 op->args[1] = op->args[2]; 1509 op->args[2] = cond; 1510 op->args[3] = label; 1511 return fold_brcond(ctx, op); 1512 1513 do_brcond_high: 1514 op->opc = INDEX_op_brcond_i32; 1515 op->args[0] = op->args[1]; 1516 op->args[1] = op->args[3]; 1517 op->args[2] = cond; 1518 op->args[3] = label; 1519 return fold_brcond(ctx, op); 1520 1521 do_brcond_const: 1522 if (i == 0) { 1523 tcg_op_remove(ctx->tcg, op); 1524 return true; 1525 } 1526 op->opc = INDEX_op_br; 1527 op->args[0] = label; 1528 finish_ebb(ctx); 1529 return true; 1530 } 1531 1532 finish_bb(ctx); 1533 return true; 1534 } 1535 1536 static bool fold_bswap(OptContext *ctx, TCGOp *op) 1537 { 1538 uint64_t z_mask, s_mask, sign; 1539 TempOptInfo *t1 = arg_info(op->args[1]); 1540 1541 if (ti_is_const(t1)) { 1542 return tcg_opt_gen_movi(ctx, op, op->args[0], 1543 do_constant_folding(op->opc, ctx->type, 1544 ti_const_val(t1), 1545 op->args[2])); 1546 } 1547 1548 z_mask = t1->z_mask; 1549 switch (op->opc) { 1550 case INDEX_op_bswap16_i32: 1551 case INDEX_op_bswap16_i64: 1552 z_mask = bswap16(z_mask); 1553 sign = INT16_MIN; 1554 break; 1555 case INDEX_op_bswap32_i32: 1556 case INDEX_op_bswap32_i64: 1557 z_mask = bswap32(z_mask); 1558 sign = INT32_MIN; 1559 break; 1560 case INDEX_op_bswap64_i64: 1561 z_mask = bswap64(z_mask); 1562 sign = INT64_MIN; 1563 break; 1564 default: 1565 g_assert_not_reached(); 1566 } 1567 1568 s_mask = 0; 1569 switch (op->args[2] & (TCG_BSWAP_OZ | TCG_BSWAP_OS)) { 1570 case TCG_BSWAP_OZ: 1571 break; 1572 case TCG_BSWAP_OS: 1573 /* If the sign bit may be 1, force all the bits above to 1. */ 1574 if (z_mask & sign) { 1575 z_mask |= sign; 1576 } 1577 /* The value and therefore s_mask is explicitly sign-extended. */ 1578 s_mask = sign; 1579 break; 1580 default: 1581 /* The high bits are undefined: force all bits above the sign to 1. */ 1582 z_mask |= sign << 1; 1583 break; 1584 } 1585 1586 return fold_masks_zs(ctx, op, z_mask, s_mask); 1587 } 1588 1589 static bool fold_call(OptContext *ctx, TCGOp *op) 1590 { 1591 TCGContext *s = ctx->tcg; 1592 int nb_oargs = TCGOP_CALLO(op); 1593 int nb_iargs = TCGOP_CALLI(op); 1594 int flags, i; 1595 1596 init_arguments(ctx, op, nb_oargs + nb_iargs); 1597 copy_propagate(ctx, op, nb_oargs, nb_iargs); 1598 1599 /* If the function reads or writes globals, reset temp data. */ 1600 flags = tcg_call_flags(op); 1601 if (!(flags & (TCG_CALL_NO_READ_GLOBALS | TCG_CALL_NO_WRITE_GLOBALS))) { 1602 int nb_globals = s->nb_globals; 1603 1604 for (i = 0; i < nb_globals; i++) { 1605 if (test_bit(i, ctx->temps_used.l)) { 1606 reset_ts(ctx, &ctx->tcg->temps[i]); 1607 } 1608 } 1609 } 1610 1611 /* If the function has side effects, reset mem data. */ 1612 if (!(flags & TCG_CALL_NO_SIDE_EFFECTS)) { 1613 remove_mem_copy_all(ctx); 1614 } 1615 1616 /* Reset temp data for outputs. */ 1617 for (i = 0; i < nb_oargs; i++) { 1618 reset_temp(ctx, op->args[i]); 1619 } 1620 1621 /* Stop optimizing MB across calls. */ 1622 ctx->prev_mb = NULL; 1623 return true; 1624 } 1625 1626 static bool fold_cmp_vec(OptContext *ctx, TCGOp *op) 1627 { 1628 /* Canonicalize the comparison to put immediate second. */ 1629 if (swap_commutative(NO_DEST, &op->args[1], &op->args[2])) { 1630 op->args[3] = tcg_swap_cond(op->args[3]); 1631 } 1632 return finish_folding(ctx, op); 1633 } 1634 1635 static bool fold_cmpsel_vec(OptContext *ctx, TCGOp *op) 1636 { 1637 /* If true and false values are the same, eliminate the cmp. */ 1638 if (args_are_copies(op->args[3], op->args[4])) { 1639 return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[3]); 1640 } 1641 1642 /* Canonicalize the comparison to put immediate second. */ 1643 if (swap_commutative(NO_DEST, &op->args[1], &op->args[2])) { 1644 op->args[5] = tcg_swap_cond(op->args[5]); 1645 } 1646 /* 1647 * Canonicalize the "false" input reg to match the destination, 1648 * so that the tcg backend can implement "move if true". 1649 */ 1650 if (swap_commutative(op->args[0], &op->args[4], &op->args[3])) { 1651 op->args[5] = tcg_invert_cond(op->args[5]); 1652 } 1653 return finish_folding(ctx, op); 1654 } 1655 1656 static bool fold_count_zeros(OptContext *ctx, TCGOp *op) 1657 { 1658 uint64_t z_mask, s_mask; 1659 TempOptInfo *t1 = arg_info(op->args[1]); 1660 TempOptInfo *t2 = arg_info(op->args[2]); 1661 1662 if (ti_is_const(t1)) { 1663 uint64_t t = ti_const_val(t1); 1664 1665 if (t != 0) { 1666 t = do_constant_folding(op->opc, ctx->type, t, 0); 1667 return tcg_opt_gen_movi(ctx, op, op->args[0], t); 1668 } 1669 return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[2]); 1670 } 1671 1672 switch (ctx->type) { 1673 case TCG_TYPE_I32: 1674 z_mask = 31; 1675 break; 1676 case TCG_TYPE_I64: 1677 z_mask = 63; 1678 break; 1679 default: 1680 g_assert_not_reached(); 1681 } 1682 s_mask = ~z_mask; 1683 z_mask |= t2->z_mask; 1684 s_mask &= t2->s_mask; 1685 1686 return fold_masks_zs(ctx, op, z_mask, s_mask); 1687 } 1688 1689 static bool fold_ctpop(OptContext *ctx, TCGOp *op) 1690 { 1691 uint64_t z_mask; 1692 1693 if (fold_const1(ctx, op)) { 1694 return true; 1695 } 1696 1697 switch (ctx->type) { 1698 case TCG_TYPE_I32: 1699 z_mask = 32 | 31; 1700 break; 1701 case TCG_TYPE_I64: 1702 z_mask = 64 | 63; 1703 break; 1704 default: 1705 g_assert_not_reached(); 1706 } 1707 return fold_masks_z(ctx, op, z_mask); 1708 } 1709 1710 static bool fold_deposit(OptContext *ctx, TCGOp *op) 1711 { 1712 TempOptInfo *t1 = arg_info(op->args[1]); 1713 TempOptInfo *t2 = arg_info(op->args[2]); 1714 int ofs = op->args[3]; 1715 int len = op->args[4]; 1716 int width; 1717 TCGOpcode and_opc; 1718 uint64_t z_mask, s_mask; 1719 1720 if (ti_is_const(t1) && ti_is_const(t2)) { 1721 return tcg_opt_gen_movi(ctx, op, op->args[0], 1722 deposit64(ti_const_val(t1), ofs, len, 1723 ti_const_val(t2))); 1724 } 1725 1726 switch (ctx->type) { 1727 case TCG_TYPE_I32: 1728 and_opc = INDEX_op_and_i32; 1729 width = 32; 1730 break; 1731 case TCG_TYPE_I64: 1732 and_opc = INDEX_op_and_i64; 1733 width = 64; 1734 break; 1735 default: 1736 g_assert_not_reached(); 1737 } 1738 1739 /* Inserting a value into zero at offset 0. */ 1740 if (ti_is_const_val(t1, 0) && ofs == 0) { 1741 uint64_t mask = MAKE_64BIT_MASK(0, len); 1742 1743 op->opc = and_opc; 1744 op->args[1] = op->args[2]; 1745 op->args[2] = arg_new_constant(ctx, mask); 1746 return fold_and(ctx, op); 1747 } 1748 1749 /* Inserting zero into a value. */ 1750 if (ti_is_const_val(t2, 0)) { 1751 uint64_t mask = deposit64(-1, ofs, len, 0); 1752 1753 op->opc = and_opc; 1754 op->args[2] = arg_new_constant(ctx, mask); 1755 return fold_and(ctx, op); 1756 } 1757 1758 /* The s_mask from the top portion of the deposit is still valid. */ 1759 if (ofs + len == width) { 1760 s_mask = t2->s_mask << ofs; 1761 } else { 1762 s_mask = t1->s_mask & ~MAKE_64BIT_MASK(0, ofs + len); 1763 } 1764 1765 z_mask = deposit64(t1->z_mask, ofs, len, t2->z_mask); 1766 return fold_masks_zs(ctx, op, z_mask, s_mask); 1767 } 1768 1769 static bool fold_divide(OptContext *ctx, TCGOp *op) 1770 { 1771 if (fold_const2(ctx, op) || 1772 fold_xi_to_x(ctx, op, 1)) { 1773 return true; 1774 } 1775 return finish_folding(ctx, op); 1776 } 1777 1778 static bool fold_dup(OptContext *ctx, TCGOp *op) 1779 { 1780 if (arg_is_const(op->args[1])) { 1781 uint64_t t = arg_info(op->args[1])->val; 1782 t = dup_const(TCGOP_VECE(op), t); 1783 return tcg_opt_gen_movi(ctx, op, op->args[0], t); 1784 } 1785 return finish_folding(ctx, op); 1786 } 1787 1788 static bool fold_dup2(OptContext *ctx, TCGOp *op) 1789 { 1790 if (arg_is_const(op->args[1]) && arg_is_const(op->args[2])) { 1791 uint64_t t = deposit64(arg_info(op->args[1])->val, 32, 32, 1792 arg_info(op->args[2])->val); 1793 return tcg_opt_gen_movi(ctx, op, op->args[0], t); 1794 } 1795 1796 if (args_are_copies(op->args[1], op->args[2])) { 1797 op->opc = INDEX_op_dup_vec; 1798 TCGOP_VECE(op) = MO_32; 1799 } 1800 return finish_folding(ctx, op); 1801 } 1802 1803 static bool fold_eqv(OptContext *ctx, TCGOp *op) 1804 { 1805 uint64_t s_mask; 1806 1807 if (fold_const2_commutative(ctx, op) || 1808 fold_xi_to_x(ctx, op, -1) || 1809 fold_xi_to_not(ctx, op, 0)) { 1810 return true; 1811 } 1812 1813 s_mask = arg_info(op->args[1])->s_mask 1814 & arg_info(op->args[2])->s_mask; 1815 return fold_masks_s(ctx, op, s_mask); 1816 } 1817 1818 static bool fold_extract(OptContext *ctx, TCGOp *op) 1819 { 1820 uint64_t z_mask_old, z_mask; 1821 TempOptInfo *t1 = arg_info(op->args[1]); 1822 int pos = op->args[2]; 1823 int len = op->args[3]; 1824 1825 if (ti_is_const(t1)) { 1826 return tcg_opt_gen_movi(ctx, op, op->args[0], 1827 extract64(ti_const_val(t1), pos, len)); 1828 } 1829 1830 z_mask_old = t1->z_mask; 1831 z_mask = extract64(z_mask_old, pos, len); 1832 if (pos == 0 && fold_affected_mask(ctx, op, z_mask_old ^ z_mask)) { 1833 return true; 1834 } 1835 1836 return fold_masks_z(ctx, op, z_mask); 1837 } 1838 1839 static bool fold_extract2(OptContext *ctx, TCGOp *op) 1840 { 1841 if (arg_is_const(op->args[1]) && arg_is_const(op->args[2])) { 1842 uint64_t v1 = arg_info(op->args[1])->val; 1843 uint64_t v2 = arg_info(op->args[2])->val; 1844 int shr = op->args[3]; 1845 1846 if (op->opc == INDEX_op_extract2_i64) { 1847 v1 >>= shr; 1848 v2 <<= 64 - shr; 1849 } else { 1850 v1 = (uint32_t)v1 >> shr; 1851 v2 = (uint64_t)((int32_t)v2 << (32 - shr)); 1852 } 1853 return tcg_opt_gen_movi(ctx, op, op->args[0], v1 | v2); 1854 } 1855 return finish_folding(ctx, op); 1856 } 1857 1858 static bool fold_exts(OptContext *ctx, TCGOp *op) 1859 { 1860 uint64_t s_mask_old, s_mask, z_mask; 1861 bool type_change = false; 1862 TempOptInfo *t1; 1863 1864 if (fold_const1(ctx, op)) { 1865 return true; 1866 } 1867 1868 t1 = arg_info(op->args[1]); 1869 z_mask = t1->z_mask; 1870 s_mask = t1->s_mask; 1871 s_mask_old = s_mask; 1872 1873 switch (op->opc) { 1874 CASE_OP_32_64(ext8s): 1875 s_mask |= INT8_MIN; 1876 z_mask = (int8_t)z_mask; 1877 break; 1878 CASE_OP_32_64(ext16s): 1879 s_mask |= INT16_MIN; 1880 z_mask = (int16_t)z_mask; 1881 break; 1882 case INDEX_op_ext_i32_i64: 1883 type_change = true; 1884 QEMU_FALLTHROUGH; 1885 case INDEX_op_ext32s_i64: 1886 s_mask |= INT32_MIN; 1887 z_mask = (int32_t)z_mask; 1888 break; 1889 default: 1890 g_assert_not_reached(); 1891 } 1892 1893 if (!type_change && fold_affected_mask(ctx, op, s_mask & ~s_mask_old)) { 1894 return true; 1895 } 1896 1897 return fold_masks_zs(ctx, op, z_mask, s_mask); 1898 } 1899 1900 static bool fold_extu(OptContext *ctx, TCGOp *op) 1901 { 1902 uint64_t z_mask_old, z_mask; 1903 bool type_change = false; 1904 1905 if (fold_const1(ctx, op)) { 1906 return true; 1907 } 1908 1909 z_mask_old = z_mask = arg_info(op->args[1])->z_mask; 1910 1911 switch (op->opc) { 1912 CASE_OP_32_64(ext8u): 1913 z_mask = (uint8_t)z_mask; 1914 break; 1915 CASE_OP_32_64(ext16u): 1916 z_mask = (uint16_t)z_mask; 1917 break; 1918 case INDEX_op_extrl_i64_i32: 1919 case INDEX_op_extu_i32_i64: 1920 type_change = true; 1921 QEMU_FALLTHROUGH; 1922 case INDEX_op_ext32u_i64: 1923 z_mask = (uint32_t)z_mask; 1924 break; 1925 case INDEX_op_extrh_i64_i32: 1926 type_change = true; 1927 z_mask >>= 32; 1928 break; 1929 default: 1930 g_assert_not_reached(); 1931 } 1932 1933 if (!type_change && fold_affected_mask(ctx, op, z_mask_old ^ z_mask)) { 1934 return true; 1935 } 1936 1937 return fold_masks_z(ctx, op, z_mask); 1938 } 1939 1940 static bool fold_mb(OptContext *ctx, TCGOp *op) 1941 { 1942 /* Eliminate duplicate and redundant fence instructions. */ 1943 if (ctx->prev_mb) { 1944 /* 1945 * Merge two barriers of the same type into one, 1946 * or a weaker barrier into a stronger one, 1947 * or two weaker barriers into a stronger one. 1948 * mb X; mb Y => mb X|Y 1949 * mb; strl => mb; st 1950 * ldaq; mb => ld; mb 1951 * ldaq; strl => ld; mb; st 1952 * Other combinations are also merged into a strong 1953 * barrier. This is stricter than specified but for 1954 * the purposes of TCG is better than not optimizing. 1955 */ 1956 ctx->prev_mb->args[0] |= op->args[0]; 1957 tcg_op_remove(ctx->tcg, op); 1958 } else { 1959 ctx->prev_mb = op; 1960 } 1961 return true; 1962 } 1963 1964 static bool fold_mov(OptContext *ctx, TCGOp *op) 1965 { 1966 return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[1]); 1967 } 1968 1969 static bool fold_movcond(OptContext *ctx, TCGOp *op) 1970 { 1971 uint64_t z_mask, s_mask; 1972 TempOptInfo *tt, *ft; 1973 int i; 1974 1975 /* If true and false values are the same, eliminate the cmp. */ 1976 if (args_are_copies(op->args[3], op->args[4])) { 1977 return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[3]); 1978 } 1979 1980 /* 1981 * Canonicalize the "false" input reg to match the destination reg so 1982 * that the tcg backend can implement a "move if true" operation. 1983 */ 1984 if (swap_commutative(op->args[0], &op->args[4], &op->args[3])) { 1985 op->args[5] = tcg_invert_cond(op->args[5]); 1986 } 1987 1988 i = do_constant_folding_cond1(ctx, op, NO_DEST, &op->args[1], 1989 &op->args[2], &op->args[5]); 1990 if (i >= 0) { 1991 return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[4 - i]); 1992 } 1993 1994 tt = arg_info(op->args[3]); 1995 ft = arg_info(op->args[4]); 1996 z_mask = tt->z_mask | ft->z_mask; 1997 s_mask = tt->s_mask & ft->s_mask; 1998 1999 if (ti_is_const(tt) && ti_is_const(ft)) { 2000 uint64_t tv = ti_const_val(tt); 2001 uint64_t fv = ti_const_val(ft); 2002 TCGOpcode opc, negopc = 0; 2003 TCGCond cond = op->args[5]; 2004 2005 switch (ctx->type) { 2006 case TCG_TYPE_I32: 2007 opc = INDEX_op_setcond_i32; 2008 if (TCG_TARGET_HAS_negsetcond_i32) { 2009 negopc = INDEX_op_negsetcond_i32; 2010 } 2011 tv = (int32_t)tv; 2012 fv = (int32_t)fv; 2013 break; 2014 case TCG_TYPE_I64: 2015 opc = INDEX_op_setcond_i64; 2016 if (TCG_TARGET_HAS_negsetcond_i64) { 2017 negopc = INDEX_op_negsetcond_i64; 2018 } 2019 break; 2020 default: 2021 g_assert_not_reached(); 2022 } 2023 2024 if (tv == 1 && fv == 0) { 2025 op->opc = opc; 2026 op->args[3] = cond; 2027 } else if (fv == 1 && tv == 0) { 2028 op->opc = opc; 2029 op->args[3] = tcg_invert_cond(cond); 2030 } else if (negopc) { 2031 if (tv == -1 && fv == 0) { 2032 op->opc = negopc; 2033 op->args[3] = cond; 2034 } else if (fv == -1 && tv == 0) { 2035 op->opc = negopc; 2036 op->args[3] = tcg_invert_cond(cond); 2037 } 2038 } 2039 } 2040 2041 return fold_masks_zs(ctx, op, z_mask, s_mask); 2042 } 2043 2044 static bool fold_mul(OptContext *ctx, TCGOp *op) 2045 { 2046 if (fold_const2(ctx, op) || 2047 fold_xi_to_i(ctx, op, 0) || 2048 fold_xi_to_x(ctx, op, 1)) { 2049 return true; 2050 } 2051 return finish_folding(ctx, op); 2052 } 2053 2054 static bool fold_mul_highpart(OptContext *ctx, TCGOp *op) 2055 { 2056 if (fold_const2_commutative(ctx, op) || 2057 fold_xi_to_i(ctx, op, 0)) { 2058 return true; 2059 } 2060 return finish_folding(ctx, op); 2061 } 2062 2063 static bool fold_multiply2(OptContext *ctx, TCGOp *op) 2064 { 2065 swap_commutative(op->args[0], &op->args[2], &op->args[3]); 2066 2067 if (arg_is_const(op->args[2]) && arg_is_const(op->args[3])) { 2068 uint64_t a = arg_info(op->args[2])->val; 2069 uint64_t b = arg_info(op->args[3])->val; 2070 uint64_t h, l; 2071 TCGArg rl, rh; 2072 TCGOp *op2; 2073 2074 switch (op->opc) { 2075 case INDEX_op_mulu2_i32: 2076 l = (uint64_t)(uint32_t)a * (uint32_t)b; 2077 h = (int32_t)(l >> 32); 2078 l = (int32_t)l; 2079 break; 2080 case INDEX_op_muls2_i32: 2081 l = (int64_t)(int32_t)a * (int32_t)b; 2082 h = l >> 32; 2083 l = (int32_t)l; 2084 break; 2085 case INDEX_op_mulu2_i64: 2086 mulu64(&l, &h, a, b); 2087 break; 2088 case INDEX_op_muls2_i64: 2089 muls64(&l, &h, a, b); 2090 break; 2091 default: 2092 g_assert_not_reached(); 2093 } 2094 2095 rl = op->args[0]; 2096 rh = op->args[1]; 2097 2098 /* The proper opcode is supplied by tcg_opt_gen_mov. */ 2099 op2 = tcg_op_insert_before(ctx->tcg, op, 0, 2); 2100 2101 tcg_opt_gen_movi(ctx, op, rl, l); 2102 tcg_opt_gen_movi(ctx, op2, rh, h); 2103 return true; 2104 } 2105 return finish_folding(ctx, op); 2106 } 2107 2108 static bool fold_nand(OptContext *ctx, TCGOp *op) 2109 { 2110 uint64_t s_mask; 2111 2112 if (fold_const2_commutative(ctx, op) || 2113 fold_xi_to_not(ctx, op, -1)) { 2114 return true; 2115 } 2116 2117 s_mask = arg_info(op->args[1])->s_mask 2118 & arg_info(op->args[2])->s_mask; 2119 return fold_masks_s(ctx, op, s_mask); 2120 } 2121 2122 static bool fold_neg_no_const(OptContext *ctx, TCGOp *op) 2123 { 2124 /* Set to 1 all bits to the left of the rightmost. */ 2125 uint64_t z_mask = arg_info(op->args[1])->z_mask; 2126 z_mask = -(z_mask & -z_mask); 2127 2128 return fold_masks_z(ctx, op, z_mask); 2129 } 2130 2131 static bool fold_neg(OptContext *ctx, TCGOp *op) 2132 { 2133 return fold_const1(ctx, op) || fold_neg_no_const(ctx, op); 2134 } 2135 2136 static bool fold_nor(OptContext *ctx, TCGOp *op) 2137 { 2138 uint64_t s_mask; 2139 2140 if (fold_const2_commutative(ctx, op) || 2141 fold_xi_to_not(ctx, op, 0)) { 2142 return true; 2143 } 2144 2145 s_mask = arg_info(op->args[1])->s_mask 2146 & arg_info(op->args[2])->s_mask; 2147 return fold_masks_s(ctx, op, s_mask); 2148 } 2149 2150 static bool fold_not(OptContext *ctx, TCGOp *op) 2151 { 2152 if (fold_const1(ctx, op)) { 2153 return true; 2154 } 2155 return fold_masks_s(ctx, op, arg_info(op->args[1])->s_mask); 2156 } 2157 2158 static bool fold_or(OptContext *ctx, TCGOp *op) 2159 { 2160 uint64_t z_mask, s_mask; 2161 TempOptInfo *t1, *t2; 2162 2163 if (fold_const2_commutative(ctx, op) || 2164 fold_xi_to_x(ctx, op, 0) || 2165 fold_xx_to_x(ctx, op)) { 2166 return true; 2167 } 2168 2169 t1 = arg_info(op->args[1]); 2170 t2 = arg_info(op->args[2]); 2171 z_mask = t1->z_mask | t2->z_mask; 2172 s_mask = t1->s_mask & t2->s_mask; 2173 return fold_masks_zs(ctx, op, z_mask, s_mask); 2174 } 2175 2176 static bool fold_orc(OptContext *ctx, TCGOp *op) 2177 { 2178 uint64_t s_mask; 2179 2180 if (fold_const2(ctx, op) || 2181 fold_xx_to_i(ctx, op, -1) || 2182 fold_xi_to_x(ctx, op, -1) || 2183 fold_ix_to_not(ctx, op, 0)) { 2184 return true; 2185 } 2186 2187 s_mask = arg_info(op->args[1])->s_mask 2188 & arg_info(op->args[2])->s_mask; 2189 return fold_masks_s(ctx, op, s_mask); 2190 } 2191 2192 static bool fold_qemu_ld_1reg(OptContext *ctx, TCGOp *op) 2193 { 2194 const TCGOpDef *def = &tcg_op_defs[op->opc]; 2195 MemOpIdx oi = op->args[def->nb_oargs + def->nb_iargs]; 2196 MemOp mop = get_memop(oi); 2197 int width = 8 * memop_size(mop); 2198 uint64_t z_mask = -1, s_mask = 0; 2199 2200 if (width < 64) { 2201 if (mop & MO_SIGN) { 2202 s_mask = MAKE_64BIT_MASK(width - 1, 64 - (width - 1)); 2203 } else { 2204 z_mask = MAKE_64BIT_MASK(0, width); 2205 } 2206 } 2207 2208 /* Opcodes that touch guest memory stop the mb optimization. */ 2209 ctx->prev_mb = NULL; 2210 2211 return fold_masks_zs(ctx, op, z_mask, s_mask); 2212 } 2213 2214 static bool fold_qemu_ld_2reg(OptContext *ctx, TCGOp *op) 2215 { 2216 /* Opcodes that touch guest memory stop the mb optimization. */ 2217 ctx->prev_mb = NULL; 2218 return finish_folding(ctx, op); 2219 } 2220 2221 static bool fold_qemu_st(OptContext *ctx, TCGOp *op) 2222 { 2223 /* Opcodes that touch guest memory stop the mb optimization. */ 2224 ctx->prev_mb = NULL; 2225 return true; 2226 } 2227 2228 static bool fold_remainder(OptContext *ctx, TCGOp *op) 2229 { 2230 if (fold_const2(ctx, op) || 2231 fold_xx_to_i(ctx, op, 0)) { 2232 return true; 2233 } 2234 return finish_folding(ctx, op); 2235 } 2236 2237 /* Return 1 if finished, -1 if simplified, 0 if unchanged. */ 2238 static int fold_setcond_zmask(OptContext *ctx, TCGOp *op, bool neg) 2239 { 2240 uint64_t a_zmask, b_val; 2241 TCGCond cond; 2242 2243 if (!arg_is_const(op->args[2])) { 2244 return false; 2245 } 2246 2247 a_zmask = arg_info(op->args[1])->z_mask; 2248 b_val = arg_info(op->args[2])->val; 2249 cond = op->args[3]; 2250 2251 if (ctx->type == TCG_TYPE_I32) { 2252 a_zmask = (uint32_t)a_zmask; 2253 b_val = (uint32_t)b_val; 2254 } 2255 2256 /* 2257 * A with only low bits set vs B with high bits set means that A < B. 2258 */ 2259 if (a_zmask < b_val) { 2260 bool inv = false; 2261 2262 switch (cond) { 2263 case TCG_COND_NE: 2264 case TCG_COND_LEU: 2265 case TCG_COND_LTU: 2266 inv = true; 2267 /* fall through */ 2268 case TCG_COND_GTU: 2269 case TCG_COND_GEU: 2270 case TCG_COND_EQ: 2271 return tcg_opt_gen_movi(ctx, op, op->args[0], neg ? -inv : inv); 2272 default: 2273 break; 2274 } 2275 } 2276 2277 /* 2278 * A with only lsb set is already boolean. 2279 */ 2280 if (a_zmask <= 1) { 2281 bool convert = false; 2282 bool inv = false; 2283 2284 switch (cond) { 2285 case TCG_COND_EQ: 2286 inv = true; 2287 /* fall through */ 2288 case TCG_COND_NE: 2289 convert = (b_val == 0); 2290 break; 2291 case TCG_COND_LTU: 2292 case TCG_COND_TSTEQ: 2293 inv = true; 2294 /* fall through */ 2295 case TCG_COND_GEU: 2296 case TCG_COND_TSTNE: 2297 convert = (b_val == 1); 2298 break; 2299 default: 2300 break; 2301 } 2302 if (convert) { 2303 TCGOpcode add_opc, xor_opc, neg_opc; 2304 2305 if (!inv && !neg) { 2306 return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[1]); 2307 } 2308 2309 switch (ctx->type) { 2310 case TCG_TYPE_I32: 2311 add_opc = INDEX_op_add_i32; 2312 neg_opc = INDEX_op_neg_i32; 2313 xor_opc = INDEX_op_xor_i32; 2314 break; 2315 case TCG_TYPE_I64: 2316 add_opc = INDEX_op_add_i64; 2317 neg_opc = INDEX_op_neg_i64; 2318 xor_opc = INDEX_op_xor_i64; 2319 break; 2320 default: 2321 g_assert_not_reached(); 2322 } 2323 2324 if (!inv) { 2325 op->opc = neg_opc; 2326 } else if (neg) { 2327 op->opc = add_opc; 2328 op->args[2] = arg_new_constant(ctx, -1); 2329 } else { 2330 op->opc = xor_opc; 2331 op->args[2] = arg_new_constant(ctx, 1); 2332 } 2333 return -1; 2334 } 2335 } 2336 return 0; 2337 } 2338 2339 static void fold_setcond_tst_pow2(OptContext *ctx, TCGOp *op, bool neg) 2340 { 2341 TCGOpcode and_opc, sub_opc, xor_opc, neg_opc, shr_opc; 2342 TCGOpcode uext_opc = 0, sext_opc = 0; 2343 TCGCond cond = op->args[3]; 2344 TCGArg ret, src1, src2; 2345 TCGOp *op2; 2346 uint64_t val; 2347 int sh; 2348 bool inv; 2349 2350 if (!is_tst_cond(cond) || !arg_is_const(op->args[2])) { 2351 return; 2352 } 2353 2354 src2 = op->args[2]; 2355 val = arg_info(src2)->val; 2356 if (!is_power_of_2(val)) { 2357 return; 2358 } 2359 sh = ctz64(val); 2360 2361 switch (ctx->type) { 2362 case TCG_TYPE_I32: 2363 and_opc = INDEX_op_and_i32; 2364 sub_opc = INDEX_op_sub_i32; 2365 xor_opc = INDEX_op_xor_i32; 2366 shr_opc = INDEX_op_shr_i32; 2367 neg_opc = INDEX_op_neg_i32; 2368 if (TCG_TARGET_extract_valid(TCG_TYPE_I32, sh, 1)) { 2369 uext_opc = INDEX_op_extract_i32; 2370 } 2371 if (TCG_TARGET_sextract_valid(TCG_TYPE_I32, sh, 1)) { 2372 sext_opc = INDEX_op_sextract_i32; 2373 } 2374 break; 2375 case TCG_TYPE_I64: 2376 and_opc = INDEX_op_and_i64; 2377 sub_opc = INDEX_op_sub_i64; 2378 xor_opc = INDEX_op_xor_i64; 2379 shr_opc = INDEX_op_shr_i64; 2380 neg_opc = INDEX_op_neg_i64; 2381 if (TCG_TARGET_extract_valid(TCG_TYPE_I64, sh, 1)) { 2382 uext_opc = INDEX_op_extract_i64; 2383 } 2384 if (TCG_TARGET_sextract_valid(TCG_TYPE_I64, sh, 1)) { 2385 sext_opc = INDEX_op_sextract_i64; 2386 } 2387 break; 2388 default: 2389 g_assert_not_reached(); 2390 } 2391 2392 ret = op->args[0]; 2393 src1 = op->args[1]; 2394 inv = cond == TCG_COND_TSTEQ; 2395 2396 if (sh && sext_opc && neg && !inv) { 2397 op->opc = sext_opc; 2398 op->args[1] = src1; 2399 op->args[2] = sh; 2400 op->args[3] = 1; 2401 return; 2402 } else if (sh && uext_opc) { 2403 op->opc = uext_opc; 2404 op->args[1] = src1; 2405 op->args[2] = sh; 2406 op->args[3] = 1; 2407 } else { 2408 if (sh) { 2409 op2 = tcg_op_insert_before(ctx->tcg, op, shr_opc, 3); 2410 op2->args[0] = ret; 2411 op2->args[1] = src1; 2412 op2->args[2] = arg_new_constant(ctx, sh); 2413 src1 = ret; 2414 } 2415 op->opc = and_opc; 2416 op->args[1] = src1; 2417 op->args[2] = arg_new_constant(ctx, 1); 2418 } 2419 2420 if (neg && inv) { 2421 op2 = tcg_op_insert_after(ctx->tcg, op, sub_opc, 3); 2422 op2->args[0] = ret; 2423 op2->args[1] = ret; 2424 op2->args[2] = arg_new_constant(ctx, 1); 2425 } else if (inv) { 2426 op2 = tcg_op_insert_after(ctx->tcg, op, xor_opc, 3); 2427 op2->args[0] = ret; 2428 op2->args[1] = ret; 2429 op2->args[2] = arg_new_constant(ctx, 1); 2430 } else if (neg) { 2431 op2 = tcg_op_insert_after(ctx->tcg, op, neg_opc, 2); 2432 op2->args[0] = ret; 2433 op2->args[1] = ret; 2434 } 2435 } 2436 2437 static bool fold_setcond(OptContext *ctx, TCGOp *op) 2438 { 2439 int i = do_constant_folding_cond1(ctx, op, op->args[0], &op->args[1], 2440 &op->args[2], &op->args[3]); 2441 if (i >= 0) { 2442 return tcg_opt_gen_movi(ctx, op, op->args[0], i); 2443 } 2444 2445 i = fold_setcond_zmask(ctx, op, false); 2446 if (i > 0) { 2447 return true; 2448 } 2449 if (i == 0) { 2450 fold_setcond_tst_pow2(ctx, op, false); 2451 } 2452 2453 return fold_masks_z(ctx, op, 1); 2454 } 2455 2456 static bool fold_negsetcond(OptContext *ctx, TCGOp *op) 2457 { 2458 int i = do_constant_folding_cond1(ctx, op, op->args[0], &op->args[1], 2459 &op->args[2], &op->args[3]); 2460 if (i >= 0) { 2461 return tcg_opt_gen_movi(ctx, op, op->args[0], -i); 2462 } 2463 2464 i = fold_setcond_zmask(ctx, op, true); 2465 if (i > 0) { 2466 return true; 2467 } 2468 if (i == 0) { 2469 fold_setcond_tst_pow2(ctx, op, true); 2470 } 2471 2472 /* Value is {0,-1} so all bits are repetitions of the sign. */ 2473 return fold_masks_s(ctx, op, -1); 2474 } 2475 2476 static bool fold_setcond2(OptContext *ctx, TCGOp *op) 2477 { 2478 TCGCond cond; 2479 int i, inv = 0; 2480 2481 i = do_constant_folding_cond2(ctx, op, &op->args[1]); 2482 cond = op->args[5]; 2483 if (i >= 0) { 2484 goto do_setcond_const; 2485 } 2486 2487 switch (cond) { 2488 case TCG_COND_LT: 2489 case TCG_COND_GE: 2490 /* 2491 * Simplify LT/GE comparisons vs zero to a single compare 2492 * vs the high word of the input. 2493 */ 2494 if (arg_is_const_val(op->args[3], 0) && 2495 arg_is_const_val(op->args[4], 0)) { 2496 goto do_setcond_high; 2497 } 2498 break; 2499 2500 case TCG_COND_NE: 2501 inv = 1; 2502 QEMU_FALLTHROUGH; 2503 case TCG_COND_EQ: 2504 /* 2505 * Simplify EQ/NE comparisons where one of the pairs 2506 * can be simplified. 2507 */ 2508 i = do_constant_folding_cond(TCG_TYPE_I32, op->args[1], 2509 op->args[3], cond); 2510 switch (i ^ inv) { 2511 case 0: 2512 goto do_setcond_const; 2513 case 1: 2514 goto do_setcond_high; 2515 } 2516 2517 i = do_constant_folding_cond(TCG_TYPE_I32, op->args[2], 2518 op->args[4], cond); 2519 switch (i ^ inv) { 2520 case 0: 2521 goto do_setcond_const; 2522 case 1: 2523 goto do_setcond_low; 2524 } 2525 break; 2526 2527 case TCG_COND_TSTEQ: 2528 case TCG_COND_TSTNE: 2529 if (arg_is_const_val(op->args[3], 0)) { 2530 goto do_setcond_high; 2531 } 2532 if (arg_is_const_val(op->args[4], 0)) { 2533 goto do_setcond_low; 2534 } 2535 break; 2536 2537 default: 2538 break; 2539 2540 do_setcond_low: 2541 op->args[2] = op->args[3]; 2542 op->args[3] = cond; 2543 op->opc = INDEX_op_setcond_i32; 2544 return fold_setcond(ctx, op); 2545 2546 do_setcond_high: 2547 op->args[1] = op->args[2]; 2548 op->args[2] = op->args[4]; 2549 op->args[3] = cond; 2550 op->opc = INDEX_op_setcond_i32; 2551 return fold_setcond(ctx, op); 2552 } 2553 2554 return fold_masks_z(ctx, op, 1); 2555 2556 do_setcond_const: 2557 return tcg_opt_gen_movi(ctx, op, op->args[0], i); 2558 } 2559 2560 static bool fold_sextract(OptContext *ctx, TCGOp *op) 2561 { 2562 uint64_t z_mask, s_mask, s_mask_old; 2563 TempOptInfo *t1 = arg_info(op->args[1]); 2564 int pos = op->args[2]; 2565 int len = op->args[3]; 2566 2567 if (ti_is_const(t1)) { 2568 return tcg_opt_gen_movi(ctx, op, op->args[0], 2569 sextract64(ti_const_val(t1), pos, len)); 2570 } 2571 2572 s_mask_old = t1->s_mask; 2573 s_mask = s_mask_old >> pos; 2574 s_mask |= -1ull << (len - 1); 2575 2576 if (pos == 0 && fold_affected_mask(ctx, op, s_mask & ~s_mask_old)) { 2577 return true; 2578 } 2579 2580 z_mask = sextract64(t1->z_mask, pos, len); 2581 return fold_masks_zs(ctx, op, z_mask, s_mask); 2582 } 2583 2584 static bool fold_shift(OptContext *ctx, TCGOp *op) 2585 { 2586 uint64_t s_mask, z_mask; 2587 TempOptInfo *t1, *t2; 2588 2589 if (fold_const2(ctx, op) || 2590 fold_ix_to_i(ctx, op, 0) || 2591 fold_xi_to_x(ctx, op, 0)) { 2592 return true; 2593 } 2594 2595 t1 = arg_info(op->args[1]); 2596 t2 = arg_info(op->args[2]); 2597 s_mask = t1->s_mask; 2598 z_mask = t1->z_mask; 2599 2600 if (ti_is_const(t2)) { 2601 int sh = ti_const_val(t2); 2602 2603 z_mask = do_constant_folding(op->opc, ctx->type, z_mask, sh); 2604 s_mask = do_constant_folding(op->opc, ctx->type, s_mask, sh); 2605 2606 return fold_masks_zs(ctx, op, z_mask, s_mask); 2607 } 2608 2609 switch (op->opc) { 2610 CASE_OP_32_64(sar): 2611 /* 2612 * Arithmetic right shift will not reduce the number of 2613 * input sign repetitions. 2614 */ 2615 return fold_masks_s(ctx, op, s_mask); 2616 CASE_OP_32_64(shr): 2617 /* 2618 * If the sign bit is known zero, then logical right shift 2619 * will not reduce the number of input sign repetitions. 2620 */ 2621 if (~z_mask & -s_mask) { 2622 return fold_masks_s(ctx, op, s_mask); 2623 } 2624 break; 2625 default: 2626 break; 2627 } 2628 2629 return finish_folding(ctx, op); 2630 } 2631 2632 static bool fold_sub_to_neg(OptContext *ctx, TCGOp *op) 2633 { 2634 TCGOpcode neg_op; 2635 bool have_neg; 2636 2637 if (!arg_is_const(op->args[1]) || arg_info(op->args[1])->val != 0) { 2638 return false; 2639 } 2640 2641 switch (ctx->type) { 2642 case TCG_TYPE_I32: 2643 neg_op = INDEX_op_neg_i32; 2644 have_neg = true; 2645 break; 2646 case TCG_TYPE_I64: 2647 neg_op = INDEX_op_neg_i64; 2648 have_neg = true; 2649 break; 2650 case TCG_TYPE_V64: 2651 case TCG_TYPE_V128: 2652 case TCG_TYPE_V256: 2653 neg_op = INDEX_op_neg_vec; 2654 have_neg = (TCG_TARGET_HAS_neg_vec && 2655 tcg_can_emit_vec_op(neg_op, ctx->type, TCGOP_VECE(op)) > 0); 2656 break; 2657 default: 2658 g_assert_not_reached(); 2659 } 2660 if (have_neg) { 2661 op->opc = neg_op; 2662 op->args[1] = op->args[2]; 2663 return fold_neg_no_const(ctx, op); 2664 } 2665 return false; 2666 } 2667 2668 /* We cannot as yet do_constant_folding with vectors. */ 2669 static bool fold_sub_vec(OptContext *ctx, TCGOp *op) 2670 { 2671 if (fold_xx_to_i(ctx, op, 0) || 2672 fold_xi_to_x(ctx, op, 0) || 2673 fold_sub_to_neg(ctx, op)) { 2674 return true; 2675 } 2676 return finish_folding(ctx, op); 2677 } 2678 2679 static bool fold_sub(OptContext *ctx, TCGOp *op) 2680 { 2681 if (fold_const2(ctx, op) || 2682 fold_xx_to_i(ctx, op, 0) || 2683 fold_xi_to_x(ctx, op, 0) || 2684 fold_sub_to_neg(ctx, op)) { 2685 return true; 2686 } 2687 2688 /* Fold sub r,x,i to add r,x,-i */ 2689 if (arg_is_const(op->args[2])) { 2690 uint64_t val = arg_info(op->args[2])->val; 2691 2692 op->opc = (ctx->type == TCG_TYPE_I32 2693 ? INDEX_op_add_i32 : INDEX_op_add_i64); 2694 op->args[2] = arg_new_constant(ctx, -val); 2695 } 2696 return finish_folding(ctx, op); 2697 } 2698 2699 static bool fold_sub2(OptContext *ctx, TCGOp *op) 2700 { 2701 return fold_addsub2(ctx, op, false); 2702 } 2703 2704 static bool fold_tcg_ld(OptContext *ctx, TCGOp *op) 2705 { 2706 uint64_t z_mask = -1, s_mask = 0; 2707 2708 /* We can't do any folding with a load, but we can record bits. */ 2709 switch (op->opc) { 2710 CASE_OP_32_64(ld8s): 2711 s_mask = INT8_MIN; 2712 break; 2713 CASE_OP_32_64(ld8u): 2714 z_mask = MAKE_64BIT_MASK(0, 8); 2715 break; 2716 CASE_OP_32_64(ld16s): 2717 s_mask = INT16_MIN; 2718 break; 2719 CASE_OP_32_64(ld16u): 2720 z_mask = MAKE_64BIT_MASK(0, 16); 2721 break; 2722 case INDEX_op_ld32s_i64: 2723 s_mask = INT32_MIN; 2724 break; 2725 case INDEX_op_ld32u_i64: 2726 z_mask = MAKE_64BIT_MASK(0, 32); 2727 break; 2728 default: 2729 g_assert_not_reached(); 2730 } 2731 return fold_masks_zs(ctx, op, z_mask, s_mask); 2732 } 2733 2734 static bool fold_tcg_ld_memcopy(OptContext *ctx, TCGOp *op) 2735 { 2736 TCGTemp *dst, *src; 2737 intptr_t ofs; 2738 TCGType type; 2739 2740 if (op->args[1] != tcgv_ptr_arg(tcg_env)) { 2741 return finish_folding(ctx, op); 2742 } 2743 2744 type = ctx->type; 2745 ofs = op->args[2]; 2746 dst = arg_temp(op->args[0]); 2747 src = find_mem_copy_for(ctx, type, ofs); 2748 if (src && src->base_type == type) { 2749 return tcg_opt_gen_mov(ctx, op, temp_arg(dst), temp_arg(src)); 2750 } 2751 2752 reset_ts(ctx, dst); 2753 record_mem_copy(ctx, type, dst, ofs, ofs + tcg_type_size(type) - 1); 2754 return true; 2755 } 2756 2757 static bool fold_tcg_st(OptContext *ctx, TCGOp *op) 2758 { 2759 intptr_t ofs = op->args[2]; 2760 intptr_t lm1; 2761 2762 if (op->args[1] != tcgv_ptr_arg(tcg_env)) { 2763 remove_mem_copy_all(ctx); 2764 return true; 2765 } 2766 2767 switch (op->opc) { 2768 CASE_OP_32_64(st8): 2769 lm1 = 0; 2770 break; 2771 CASE_OP_32_64(st16): 2772 lm1 = 1; 2773 break; 2774 case INDEX_op_st32_i64: 2775 case INDEX_op_st_i32: 2776 lm1 = 3; 2777 break; 2778 case INDEX_op_st_i64: 2779 lm1 = 7; 2780 break; 2781 case INDEX_op_st_vec: 2782 lm1 = tcg_type_size(ctx->type) - 1; 2783 break; 2784 default: 2785 g_assert_not_reached(); 2786 } 2787 remove_mem_copy_in(ctx, ofs, ofs + lm1); 2788 return true; 2789 } 2790 2791 static bool fold_tcg_st_memcopy(OptContext *ctx, TCGOp *op) 2792 { 2793 TCGTemp *src; 2794 intptr_t ofs, last; 2795 TCGType type; 2796 2797 if (op->args[1] != tcgv_ptr_arg(tcg_env)) { 2798 return fold_tcg_st(ctx, op); 2799 } 2800 2801 src = arg_temp(op->args[0]); 2802 ofs = op->args[2]; 2803 type = ctx->type; 2804 2805 /* 2806 * Eliminate duplicate stores of a constant. 2807 * This happens frequently when the target ISA zero-extends. 2808 */ 2809 if (ts_is_const(src)) { 2810 TCGTemp *prev = find_mem_copy_for(ctx, type, ofs); 2811 if (src == prev) { 2812 tcg_op_remove(ctx->tcg, op); 2813 return true; 2814 } 2815 } 2816 2817 last = ofs + tcg_type_size(type) - 1; 2818 remove_mem_copy_in(ctx, ofs, last); 2819 record_mem_copy(ctx, type, src, ofs, last); 2820 return true; 2821 } 2822 2823 static bool fold_xor(OptContext *ctx, TCGOp *op) 2824 { 2825 uint64_t z_mask, s_mask; 2826 TempOptInfo *t1, *t2; 2827 2828 if (fold_const2_commutative(ctx, op) || 2829 fold_xx_to_i(ctx, op, 0) || 2830 fold_xi_to_x(ctx, op, 0) || 2831 fold_xi_to_not(ctx, op, -1)) { 2832 return true; 2833 } 2834 2835 t1 = arg_info(op->args[1]); 2836 t2 = arg_info(op->args[2]); 2837 z_mask = t1->z_mask | t2->z_mask; 2838 s_mask = t1->s_mask & t2->s_mask; 2839 return fold_masks_zs(ctx, op, z_mask, s_mask); 2840 } 2841 2842 /* Propagate constants and copies, fold constant expressions. */ 2843 void tcg_optimize(TCGContext *s) 2844 { 2845 int nb_temps, i; 2846 TCGOp *op, *op_next; 2847 OptContext ctx = { .tcg = s }; 2848 2849 QSIMPLEQ_INIT(&ctx.mem_free); 2850 2851 /* Array VALS has an element for each temp. 2852 If this temp holds a constant then its value is kept in VALS' element. 2853 If this temp is a copy of other ones then the other copies are 2854 available through the doubly linked circular list. */ 2855 2856 nb_temps = s->nb_temps; 2857 for (i = 0; i < nb_temps; ++i) { 2858 s->temps[i].state_ptr = NULL; 2859 } 2860 2861 QTAILQ_FOREACH_SAFE(op, &s->ops, link, op_next) { 2862 TCGOpcode opc = op->opc; 2863 const TCGOpDef *def; 2864 bool done = false; 2865 2866 /* Calls are special. */ 2867 if (opc == INDEX_op_call) { 2868 fold_call(&ctx, op); 2869 continue; 2870 } 2871 2872 def = &tcg_op_defs[opc]; 2873 init_arguments(&ctx, op, def->nb_oargs + def->nb_iargs); 2874 copy_propagate(&ctx, op, def->nb_oargs, def->nb_iargs); 2875 2876 /* Pre-compute the type of the operation. */ 2877 ctx.type = TCGOP_TYPE(op); 2878 2879 /* 2880 * Process each opcode. 2881 * Sorted alphabetically by opcode as much as possible. 2882 */ 2883 switch (opc) { 2884 CASE_OP_32_64(add): 2885 done = fold_add(&ctx, op); 2886 break; 2887 case INDEX_op_add_vec: 2888 done = fold_add_vec(&ctx, op); 2889 break; 2890 CASE_OP_32_64(add2): 2891 done = fold_add2(&ctx, op); 2892 break; 2893 CASE_OP_32_64_VEC(and): 2894 done = fold_and(&ctx, op); 2895 break; 2896 CASE_OP_32_64_VEC(andc): 2897 done = fold_andc(&ctx, op); 2898 break; 2899 CASE_OP_32_64(brcond): 2900 done = fold_brcond(&ctx, op); 2901 break; 2902 case INDEX_op_brcond2_i32: 2903 done = fold_brcond2(&ctx, op); 2904 break; 2905 CASE_OP_32_64(bswap16): 2906 CASE_OP_32_64(bswap32): 2907 case INDEX_op_bswap64_i64: 2908 done = fold_bswap(&ctx, op); 2909 break; 2910 CASE_OP_32_64(clz): 2911 CASE_OP_32_64(ctz): 2912 done = fold_count_zeros(&ctx, op); 2913 break; 2914 CASE_OP_32_64(ctpop): 2915 done = fold_ctpop(&ctx, op); 2916 break; 2917 CASE_OP_32_64(deposit): 2918 done = fold_deposit(&ctx, op); 2919 break; 2920 CASE_OP_32_64(div): 2921 CASE_OP_32_64(divu): 2922 done = fold_divide(&ctx, op); 2923 break; 2924 case INDEX_op_dup_vec: 2925 done = fold_dup(&ctx, op); 2926 break; 2927 case INDEX_op_dup2_vec: 2928 done = fold_dup2(&ctx, op); 2929 break; 2930 CASE_OP_32_64_VEC(eqv): 2931 done = fold_eqv(&ctx, op); 2932 break; 2933 CASE_OP_32_64(extract): 2934 done = fold_extract(&ctx, op); 2935 break; 2936 CASE_OP_32_64(extract2): 2937 done = fold_extract2(&ctx, op); 2938 break; 2939 CASE_OP_32_64(ext8s): 2940 CASE_OP_32_64(ext16s): 2941 case INDEX_op_ext32s_i64: 2942 case INDEX_op_ext_i32_i64: 2943 done = fold_exts(&ctx, op); 2944 break; 2945 CASE_OP_32_64(ext8u): 2946 CASE_OP_32_64(ext16u): 2947 case INDEX_op_ext32u_i64: 2948 case INDEX_op_extu_i32_i64: 2949 case INDEX_op_extrl_i64_i32: 2950 case INDEX_op_extrh_i64_i32: 2951 done = fold_extu(&ctx, op); 2952 break; 2953 CASE_OP_32_64(ld8s): 2954 CASE_OP_32_64(ld8u): 2955 CASE_OP_32_64(ld16s): 2956 CASE_OP_32_64(ld16u): 2957 case INDEX_op_ld32s_i64: 2958 case INDEX_op_ld32u_i64: 2959 done = fold_tcg_ld(&ctx, op); 2960 break; 2961 case INDEX_op_ld_i32: 2962 case INDEX_op_ld_i64: 2963 case INDEX_op_ld_vec: 2964 done = fold_tcg_ld_memcopy(&ctx, op); 2965 break; 2966 CASE_OP_32_64(st8): 2967 CASE_OP_32_64(st16): 2968 case INDEX_op_st32_i64: 2969 done = fold_tcg_st(&ctx, op); 2970 break; 2971 case INDEX_op_st_i32: 2972 case INDEX_op_st_i64: 2973 case INDEX_op_st_vec: 2974 done = fold_tcg_st_memcopy(&ctx, op); 2975 break; 2976 case INDEX_op_mb: 2977 done = fold_mb(&ctx, op); 2978 break; 2979 CASE_OP_32_64_VEC(mov): 2980 done = fold_mov(&ctx, op); 2981 break; 2982 CASE_OP_32_64(movcond): 2983 done = fold_movcond(&ctx, op); 2984 break; 2985 CASE_OP_32_64(mul): 2986 done = fold_mul(&ctx, op); 2987 break; 2988 CASE_OP_32_64(mulsh): 2989 CASE_OP_32_64(muluh): 2990 done = fold_mul_highpart(&ctx, op); 2991 break; 2992 CASE_OP_32_64(muls2): 2993 CASE_OP_32_64(mulu2): 2994 done = fold_multiply2(&ctx, op); 2995 break; 2996 CASE_OP_32_64_VEC(nand): 2997 done = fold_nand(&ctx, op); 2998 break; 2999 CASE_OP_32_64(neg): 3000 done = fold_neg(&ctx, op); 3001 break; 3002 CASE_OP_32_64_VEC(nor): 3003 done = fold_nor(&ctx, op); 3004 break; 3005 CASE_OP_32_64_VEC(not): 3006 done = fold_not(&ctx, op); 3007 break; 3008 CASE_OP_32_64_VEC(or): 3009 done = fold_or(&ctx, op); 3010 break; 3011 CASE_OP_32_64_VEC(orc): 3012 done = fold_orc(&ctx, op); 3013 break; 3014 case INDEX_op_qemu_ld_i32: 3015 done = fold_qemu_ld_1reg(&ctx, op); 3016 break; 3017 case INDEX_op_qemu_ld_i64: 3018 if (TCG_TARGET_REG_BITS == 64) { 3019 done = fold_qemu_ld_1reg(&ctx, op); 3020 break; 3021 } 3022 QEMU_FALLTHROUGH; 3023 case INDEX_op_qemu_ld_i128: 3024 done = fold_qemu_ld_2reg(&ctx, op); 3025 break; 3026 case INDEX_op_qemu_st8_i32: 3027 case INDEX_op_qemu_st_i32: 3028 case INDEX_op_qemu_st_i64: 3029 case INDEX_op_qemu_st_i128: 3030 done = fold_qemu_st(&ctx, op); 3031 break; 3032 CASE_OP_32_64(rem): 3033 CASE_OP_32_64(remu): 3034 done = fold_remainder(&ctx, op); 3035 break; 3036 CASE_OP_32_64(rotl): 3037 CASE_OP_32_64(rotr): 3038 CASE_OP_32_64(sar): 3039 CASE_OP_32_64(shl): 3040 CASE_OP_32_64(shr): 3041 done = fold_shift(&ctx, op); 3042 break; 3043 CASE_OP_32_64(setcond): 3044 done = fold_setcond(&ctx, op); 3045 break; 3046 CASE_OP_32_64(negsetcond): 3047 done = fold_negsetcond(&ctx, op); 3048 break; 3049 case INDEX_op_setcond2_i32: 3050 done = fold_setcond2(&ctx, op); 3051 break; 3052 case INDEX_op_cmp_vec: 3053 done = fold_cmp_vec(&ctx, op); 3054 break; 3055 case INDEX_op_cmpsel_vec: 3056 done = fold_cmpsel_vec(&ctx, op); 3057 break; 3058 case INDEX_op_bitsel_vec: 3059 done = fold_bitsel_vec(&ctx, op); 3060 break; 3061 CASE_OP_32_64(sextract): 3062 done = fold_sextract(&ctx, op); 3063 break; 3064 CASE_OP_32_64(sub): 3065 done = fold_sub(&ctx, op); 3066 break; 3067 case INDEX_op_sub_vec: 3068 done = fold_sub_vec(&ctx, op); 3069 break; 3070 CASE_OP_32_64(sub2): 3071 done = fold_sub2(&ctx, op); 3072 break; 3073 CASE_OP_32_64_VEC(xor): 3074 done = fold_xor(&ctx, op); 3075 break; 3076 case INDEX_op_set_label: 3077 case INDEX_op_br: 3078 case INDEX_op_exit_tb: 3079 case INDEX_op_goto_tb: 3080 case INDEX_op_goto_ptr: 3081 finish_ebb(&ctx); 3082 done = true; 3083 break; 3084 default: 3085 done = finish_folding(&ctx, op); 3086 break; 3087 } 3088 tcg_debug_assert(done); 3089 } 3090 } 3091