Mercurial > hg > CbC > CbC_gcc
annotate gcc/postreload-gcse.c @ 144:8f4e72ab4e11
fix segmentation fault caused by nothing next cur_op to end
| author | Takahiro SHIMIZU <anatofuz@cr.ie.u-ryukyu.ac.jp> |
|---|---|
| date | Sun, 23 Dec 2018 21:23:56 +0900 |
| parents | 84e7813d76e9 |
| children | 1830386684a0 |
| rev | line source |
|---|---|
| 0 | 1 /* Post reload partially redundant load elimination |
| 131 | 2 Copyright (C) 2004-2018 Free Software Foundation, Inc. |
| 0 | 3 |
| 4 This file is part of GCC. | |
| 5 | |
| 6 GCC is free software; you can redistribute it and/or modify it under | |
| 7 the terms of the GNU General Public License as published by the Free | |
| 8 Software Foundation; either version 3, or (at your option) any later | |
| 9 version. | |
| 10 | |
| 11 GCC is distributed in the hope that it will be useful, but WITHOUT ANY | |
| 12 WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
| 13 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
| 14 for more details. | |
| 15 | |
| 16 You should have received a copy of the GNU General Public License | |
| 17 along with GCC; see the file COPYING3. If not see | |
| 18 <http://www.gnu.org/licenses/>. */ | |
| 19 | |
| 20 #include "config.h" | |
| 21 #include "system.h" | |
| 22 #include "coretypes.h" | |
| 111 | 23 #include "backend.h" |
| 24 #include "target.h" | |
| 0 | 25 #include "rtl.h" |
| 26 #include "tree.h" | |
| 111 | 27 #include "predict.h" |
| 28 #include "df.h" | |
| 29 #include "memmodel.h" | |
| 0 | 30 #include "tm_p.h" |
| 31 #include "insn-config.h" | |
| 111 | 32 #include "emit-rtl.h" |
| 0 | 33 #include "recog.h" |
| 111 | 34 |
| 35 #include "cfgrtl.h" | |
| 36 #include "profile.h" | |
| 0 | 37 #include "expr.h" |
| 38 #include "params.h" | |
| 39 #include "tree-pass.h" | |
| 40 #include "dbgcnt.h" | |
| 111 | 41 #include "gcse-common.h" |
| 0 | 42 |
| 43 /* The following code implements gcse after reload, the purpose of this | |
| 44 pass is to cleanup redundant loads generated by reload and other | |
| 45 optimizations that come after gcse. It searches for simple inter-block | |
| 46 redundancies and tries to eliminate them by adding moves and loads | |
| 47 in cold places. | |
| 48 | |
| 49 Perform partially redundant load elimination, try to eliminate redundant | |
| 50 loads created by the reload pass. We try to look for full or partial | |
| 51 redundant loads fed by one or more loads/stores in predecessor BBs, | |
| 52 and try adding loads to make them fully redundant. We also check if | |
| 53 it's worth adding loads to be able to delete the redundant load. | |
| 54 | |
| 55 Algorithm: | |
| 56 1. Build available expressions hash table: | |
| 57 For each load/store instruction, if the loaded/stored memory didn't | |
| 58 change until the end of the basic block add this memory expression to | |
| 59 the hash table. | |
| 60 2. Perform Redundancy elimination: | |
| 61 For each load instruction do the following: | |
| 62 perform partial redundancy elimination, check if it's worth adding | |
| 63 loads to make the load fully redundant. If so add loads and | |
| 64 register copies and delete the load. | |
| 65 3. Delete instructions made redundant in step 2. | |
| 66 | |
| 67 Future enhancement: | |
| 68 If the loaded register is used/defined between load and some store, | |
| 69 look for some other free register between load and all its stores, | |
| 70 and replace the load with a copy from this register to the loaded | |
| 71 register. | |
| 72 */ | |
| 73 | |
| 74 | |
| 75 /* Keep statistics of this pass. */ | |
| 76 static struct | |
| 77 { | |
| 78 int moves_inserted; | |
| 79 int copies_inserted; | |
| 80 int insns_deleted; | |
| 81 } stats; | |
| 82 | |
| 83 /* We need to keep a hash table of expressions. The table entries are of | |
| 84 type 'struct expr', and for each expression there is a single linked | |
| 85 list of occurrences. */ | |
| 86 | |
| 87 /* Expression elements in the hash table. */ | |
| 88 struct expr | |
| 89 { | |
| 90 /* The expression (SET_SRC for expressions, PATTERN for assignments). */ | |
| 91 rtx expr; | |
| 92 | |
| 93 /* The same hash for this entry. */ | |
| 94 hashval_t hash; | |
| 95 | |
| 111 | 96 /* Index in the transparent bitmaps. */ |
| 97 unsigned int bitmap_index; | |
| 98 | |
| 0 | 99 /* List of available occurrence in basic blocks in the function. */ |
| 100 struct occr *avail_occr; | |
| 101 }; | |
| 102 | |
| 111 | 103 /* Hashtable helpers. */ |
| 104 | |
| 105 struct expr_hasher : nofree_ptr_hash <expr> | |
| 106 { | |
| 107 static inline hashval_t hash (const expr *); | |
| 108 static inline bool equal (const expr *, const expr *); | |
| 109 }; | |
| 110 | |
| 111 | |
| 112 /* Hash expression X. | |
| 113 DO_NOT_RECORD_P is a boolean indicating if a volatile operand is found | |
| 114 or if the expression contains something we don't want to insert in the | |
| 115 table. */ | |
| 116 | |
| 117 static hashval_t | |
| 118 hash_expr (rtx x, int *do_not_record_p) | |
| 119 { | |
| 120 *do_not_record_p = 0; | |
| 121 return hash_rtx (x, GET_MODE (x), do_not_record_p, | |
| 122 NULL, /*have_reg_qty=*/false); | |
| 123 } | |
| 124 | |
| 125 /* Callback for hashtab. | |
| 126 Return the hash value for expression EXP. We don't actually hash | |
| 127 here, we just return the cached hash value. */ | |
| 128 | |
| 129 inline hashval_t | |
| 130 expr_hasher::hash (const expr *exp) | |
| 131 { | |
| 132 return exp->hash; | |
| 133 } | |
| 134 | |
| 135 /* Callback for hashtab. | |
| 136 Return nonzero if exp1 is equivalent to exp2. */ | |
| 137 | |
| 138 inline bool | |
| 139 expr_hasher::equal (const expr *exp1, const expr *exp2) | |
| 140 { | |
| 141 int equiv_p = exp_equiv_p (exp1->expr, exp2->expr, 0, true); | |
| 142 | |
| 143 gcc_assert (!equiv_p || exp1->hash == exp2->hash); | |
| 144 return equiv_p; | |
| 145 } | |
| 146 | |
| 147 /* The table itself. */ | |
| 148 static hash_table<expr_hasher> *expr_table; | |
| 149 | |
| 150 | |
| 0 | 151 static struct obstack expr_obstack; |
| 152 | |
| 153 /* Occurrence of an expression. | |
| 154 There is at most one occurrence per basic block. If a pattern appears | |
| 155 more than once, the last appearance is used. */ | |
| 156 | |
| 157 struct occr | |
| 158 { | |
| 159 /* Next occurrence of this expression. */ | |
| 160 struct occr *next; | |
| 161 /* The insn that computes the expression. */ | |
| 111 | 162 rtx_insn *insn; |
| 0 | 163 /* Nonzero if this [anticipatable] occurrence has been deleted. */ |
| 164 char deleted_p; | |
| 165 }; | |
| 166 | |
| 167 static struct obstack occr_obstack; | |
| 168 | |
| 169 /* The following structure holds the information about the occurrences of | |
| 170 the redundant instructions. */ | |
| 171 struct unoccr | |
| 172 { | |
| 173 struct unoccr *next; | |
| 174 edge pred; | |
| 111 | 175 rtx_insn *insn; |
| 0 | 176 }; |
| 177 | |
| 178 static struct obstack unoccr_obstack; | |
| 179 | |
| 180 /* Array where each element is the CUID if the insn that last set the hard | |
| 181 register with the number of the element, since the start of the current | |
| 182 basic block. | |
| 183 | |
| 184 This array is used during the building of the hash table (step 1) to | |
| 185 determine if a reg is killed before the end of a basic block. | |
| 186 | |
| 187 It is also used when eliminating partial redundancies (step 2) to see | |
| 188 if a reg was modified since the start of a basic block. */ | |
| 189 static int *reg_avail_info; | |
| 190 | |
| 191 /* A list of insns that may modify memory within the current basic block. */ | |
| 192 struct modifies_mem | |
| 193 { | |
| 111 | 194 rtx_insn *insn; |
| 0 | 195 struct modifies_mem *next; |
| 196 }; | |
| 197 static struct modifies_mem *modifies_mem_list; | |
| 198 | |
| 199 /* The modifies_mem structs also go on an obstack, only this obstack is | |
| 200 freed each time after completing the analysis or transformations on | |
| 201 a basic block. So we allocate a dummy modifies_mem_obstack_bottom | |
| 202 object on the obstack to keep track of the bottom of the obstack. */ | |
| 203 static struct obstack modifies_mem_obstack; | |
| 204 static struct modifies_mem *modifies_mem_obstack_bottom; | |
| 205 | |
| 206 /* Mapping of insn UIDs to CUIDs. | |
| 207 CUIDs are like UIDs except they increase monotonically in each basic | |
| 208 block, have no gaps, and only apply to real insns. */ | |
| 209 static int *uid_cuid; | |
| 210 #define INSN_CUID(INSN) (uid_cuid[INSN_UID (INSN)]) | |
| 111 | 211 |
| 212 /* Bitmap of blocks which have memory stores. */ | |
| 213 static bitmap modify_mem_list_set; | |
| 214 | |
| 215 /* Bitmap of blocks which have calls. */ | |
| 216 static bitmap blocks_with_calls; | |
| 217 | |
| 218 /* Vector indexed by block # with a list of all the insns that | |
| 219 modify memory within the block. */ | |
| 220 static vec<rtx_insn *> *modify_mem_list; | |
| 221 | |
| 222 /* Vector indexed by block # with a canonicalized list of insns | |
| 223 that modify memory in the block. */ | |
| 224 static vec<modify_pair> *canon_modify_mem_list; | |
| 225 | |
| 226 /* Vector of simple bitmaps indexed by block number. Each component sbitmap | |
| 227 indicates which expressions are transparent through the block. */ | |
| 228 static sbitmap *transp; | |
| 0 | 229 |
| 230 | |
| 231 /* Helpers for memory allocation/freeing. */ | |
| 232 static void alloc_mem (void); | |
| 233 static void free_mem (void); | |
| 234 | |
| 235 /* Support for hash table construction and transformations. */ | |
| 111 | 236 static bool oprs_unchanged_p (rtx, rtx_insn *, bool); |
| 237 static void record_last_reg_set_info (rtx_insn *, rtx); | |
| 238 static void record_last_reg_set_info_regno (rtx_insn *, int); | |
| 239 static void record_last_mem_set_info (rtx_insn *); | |
| 0 | 240 static void record_last_set_info (rtx, const_rtx, void *); |
| 111 | 241 static void record_opr_changes (rtx_insn *); |
| 0 | 242 |
| 243 static void find_mem_conflicts (rtx, const_rtx, void *); | |
| 244 static int load_killed_in_block_p (int, rtx, bool); | |
| 245 static void reset_opr_set_tables (void); | |
| 246 | |
| 247 /* Hash table support. */ | |
| 248 static hashval_t hash_expr (rtx, int *); | |
| 111 | 249 static void insert_expr_in_table (rtx, rtx_insn *); |
| 0 | 250 static struct expr *lookup_expr_in_table (rtx); |
| 251 static void dump_hash_table (FILE *); | |
| 252 | |
| 253 /* Helpers for eliminate_partially_redundant_load. */ | |
| 254 static bool reg_killed_on_edge (rtx, edge); | |
| 255 static bool reg_used_on_edge (rtx, edge); | |
| 256 | |
| 111 | 257 static rtx get_avail_load_store_reg (rtx_insn *); |
| 0 | 258 |
| 259 static bool bb_has_well_behaved_predecessors (basic_block); | |
| 111 | 260 static struct occr* get_bb_avail_insn (basic_block, struct occr *, int); |
| 261 static void hash_scan_set (rtx_insn *); | |
| 0 | 262 static void compute_hash_table (void); |
| 263 | |
| 264 /* The work horses of this pass. */ | |
| 265 static void eliminate_partially_redundant_load (basic_block, | |
| 111 | 266 rtx_insn *, |
| 0 | 267 struct expr *); |
| 268 static void eliminate_partially_redundant_loads (void); | |
| 269 | |
| 270 | |
| 271 /* Allocate memory for the CUID mapping array and register/memory | |
| 272 tracking tables. */ | |
| 273 | |
| 274 static void | |
| 275 alloc_mem (void) | |
| 276 { | |
| 277 int i; | |
| 278 basic_block bb; | |
| 111 | 279 rtx_insn *insn; |
| 0 | 280 |
| 281 /* Find the largest UID and create a mapping from UIDs to CUIDs. */ | |
| 282 uid_cuid = XCNEWVEC (int, get_max_uid () + 1); | |
| 283 i = 1; | |
| 111 | 284 FOR_EACH_BB_FN (bb, cfun) |
| 0 | 285 FOR_BB_INSNS (bb, insn) |
| 286 { | |
| 287 if (INSN_P (insn)) | |
| 288 uid_cuid[INSN_UID (insn)] = i++; | |
| 289 else | |
| 290 uid_cuid[INSN_UID (insn)] = i; | |
| 291 } | |
| 292 | |
| 293 /* Allocate the available expressions hash table. We don't want to | |
| 294 make the hash table too small, but unnecessarily making it too large | |
| 295 also doesn't help. The i/4 is a gcse.c relic, and seems like a | |
| 296 reasonable choice. */ | |
| 111 | 297 expr_table = new hash_table<expr_hasher> (MAX (i / 4, 13)); |
| 0 | 298 |
| 299 /* We allocate everything on obstacks because we often can roll back | |
| 300 the whole obstack to some point. Freeing obstacks is very fast. */ | |
| 301 gcc_obstack_init (&expr_obstack); | |
| 302 gcc_obstack_init (&occr_obstack); | |
| 303 gcc_obstack_init (&unoccr_obstack); | |
| 304 gcc_obstack_init (&modifies_mem_obstack); | |
| 305 | |
| 306 /* Working array used to track the last set for each register | |
| 307 in the current block. */ | |
| 308 reg_avail_info = (int *) xmalloc (FIRST_PSEUDO_REGISTER * sizeof (int)); | |
| 309 | |
| 310 /* Put a dummy modifies_mem object on the modifies_mem_obstack, so we | |
| 311 can roll it back in reset_opr_set_tables. */ | |
| 312 modifies_mem_obstack_bottom = | |
| 313 (struct modifies_mem *) obstack_alloc (&modifies_mem_obstack, | |
| 314 sizeof (struct modifies_mem)); | |
| 111 | 315 |
| 316 blocks_with_calls = BITMAP_ALLOC (NULL); | |
| 317 modify_mem_list_set = BITMAP_ALLOC (NULL); | |
| 318 | |
| 319 modify_mem_list = (vec_rtx_heap *) xcalloc (last_basic_block_for_fn (cfun), | |
| 320 sizeof (vec_rtx_heap)); | |
| 321 canon_modify_mem_list | |
| 322 = (vec_modify_pair_heap *) xcalloc (last_basic_block_for_fn (cfun), | |
| 323 sizeof (vec_modify_pair_heap)); | |
| 0 | 324 } |
| 325 | |
| 326 /* Free memory allocated by alloc_mem. */ | |
| 327 | |
| 328 static void | |
| 329 free_mem (void) | |
| 330 { | |
| 331 free (uid_cuid); | |
| 332 | |
| 111 | 333 delete expr_table; |
| 334 expr_table = NULL; | |
| 0 | 335 |
| 336 obstack_free (&expr_obstack, NULL); | |
| 337 obstack_free (&occr_obstack, NULL); | |
| 338 obstack_free (&unoccr_obstack, NULL); | |
| 339 obstack_free (&modifies_mem_obstack, NULL); | |
| 340 | |
| 111 | 341 unsigned i; |
| 342 bitmap_iterator bi; | |
| 343 EXECUTE_IF_SET_IN_BITMAP (modify_mem_list_set, 0, i, bi) | |
| 344 { | |
| 345 modify_mem_list[i].release (); | |
| 346 canon_modify_mem_list[i].release (); | |
| 347 } | |
| 0 | 348 |
| 111 | 349 BITMAP_FREE (blocks_with_calls); |
| 350 BITMAP_FREE (modify_mem_list_set); | |
| 351 free (reg_avail_info); | |
| 352 free (modify_mem_list); | |
| 353 free (canon_modify_mem_list); | |
| 0 | 354 } |
| 355 | |
| 356 | |
| 357 /* Insert expression X in INSN in the hash TABLE. | |
| 358 If it is already present, record it as the last occurrence in INSN's | |
| 359 basic block. */ | |
| 360 | |
| 361 static void | |
| 111 | 362 insert_expr_in_table (rtx x, rtx_insn *insn) |
| 0 | 363 { |
| 364 int do_not_record_p; | |
| 365 hashval_t hash; | |
| 366 struct expr *cur_expr, **slot; | |
| 367 struct occr *avail_occr, *last_occr = NULL; | |
| 368 | |
| 369 hash = hash_expr (x, &do_not_record_p); | |
| 370 | |
| 371 /* Do not insert expression in the table if it contains volatile operands, | |
| 372 or if hash_expr determines the expression is something we don't want | |
| 373 to or can't handle. */ | |
| 374 if (do_not_record_p) | |
| 375 return; | |
| 376 | |
| 377 /* We anticipate that redundant expressions are rare, so for convenience | |
| 378 allocate a new hash table element here already and set its fields. | |
| 379 If we don't do this, we need a hack with a static struct expr. Anyway, | |
| 380 obstack_free is really fast and one more obstack_alloc doesn't hurt if | |
| 381 we're going to see more expressions later on. */ | |
| 382 cur_expr = (struct expr *) obstack_alloc (&expr_obstack, | |
| 383 sizeof (struct expr)); | |
| 384 cur_expr->expr = x; | |
| 385 cur_expr->hash = hash; | |
| 386 cur_expr->avail_occr = NULL; | |
| 387 | |
| 111 | 388 slot = expr_table->find_slot_with_hash (cur_expr, hash, INSERT); |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
389 |
| 0 | 390 if (! (*slot)) |
| 111 | 391 { |
| 392 /* The expression isn't found, so insert it. */ | |
| 393 *slot = cur_expr; | |
| 394 | |
| 395 /* Anytime we add an entry to the table, record the index | |
| 396 of the new entry. The bitmap index starts counting | |
| 397 at zero. */ | |
| 398 cur_expr->bitmap_index = expr_table->elements () - 1; | |
| 399 } | |
| 0 | 400 else |
| 401 { | |
| 402 /* The expression is already in the table, so roll back the | |
| 403 obstack and use the existing table entry. */ | |
| 404 obstack_free (&expr_obstack, cur_expr); | |
| 405 cur_expr = *slot; | |
| 406 } | |
| 407 | |
| 408 /* Search for another occurrence in the same basic block. */ | |
| 409 avail_occr = cur_expr->avail_occr; | |
|
63
b7f97abdc517
update gcc from gcc-4.5.0 to gcc-4.6
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
55
diff
changeset
|
410 while (avail_occr |
|
b7f97abdc517
update gcc from gcc-4.5.0 to gcc-4.6
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
55
diff
changeset
|
411 && BLOCK_FOR_INSN (avail_occr->insn) != BLOCK_FOR_INSN (insn)) |
| 0 | 412 { |
| 413 /* If an occurrence isn't found, save a pointer to the end of | |
| 414 the list. */ | |
| 415 last_occr = avail_occr; | |
| 416 avail_occr = avail_occr->next; | |
| 417 } | |
| 418 | |
| 419 if (avail_occr) | |
| 420 /* Found another instance of the expression in the same basic block. | |
| 421 Prefer this occurrence to the currently recorded one. We want | |
| 422 the last one in the block and the block is scanned from start | |
| 423 to end. */ | |
| 424 avail_occr->insn = insn; | |
| 425 else | |
| 426 { | |
| 427 /* First occurrence of this expression in this basic block. */ | |
| 428 avail_occr = (struct occr *) obstack_alloc (&occr_obstack, | |
| 429 sizeof (struct occr)); | |
| 430 | |
| 431 /* First occurrence of this expression in any block? */ | |
| 432 if (cur_expr->avail_occr == NULL) | |
| 433 cur_expr->avail_occr = avail_occr; | |
| 434 else | |
| 435 last_occr->next = avail_occr; | |
| 436 | |
| 437 avail_occr->insn = insn; | |
| 438 avail_occr->next = NULL; | |
| 439 avail_occr->deleted_p = 0; | |
| 440 } | |
| 441 } | |
| 442 | |
| 443 | |
| 444 /* Lookup pattern PAT in the expression hash table. | |
| 445 The result is a pointer to the table entry, or NULL if not found. */ | |
| 446 | |
| 447 static struct expr * | |
| 448 lookup_expr_in_table (rtx pat) | |
| 449 { | |
| 450 int do_not_record_p; | |
| 451 struct expr **slot, *tmp_expr; | |
| 452 hashval_t hash = hash_expr (pat, &do_not_record_p); | |
| 453 | |
| 454 if (do_not_record_p) | |
| 455 return NULL; | |
| 456 | |
| 457 tmp_expr = (struct expr *) obstack_alloc (&expr_obstack, | |
| 458 sizeof (struct expr)); | |
| 459 tmp_expr->expr = pat; | |
| 460 tmp_expr->hash = hash; | |
| 461 tmp_expr->avail_occr = NULL; | |
| 462 | |
| 111 | 463 slot = expr_table->find_slot_with_hash (tmp_expr, hash, INSERT); |
| 0 | 464 obstack_free (&expr_obstack, tmp_expr); |
| 465 | |
| 466 if (!slot) | |
| 467 return NULL; | |
| 468 else | |
| 469 return (*slot); | |
| 470 } | |
| 471 | |
| 472 | |
| 473 /* Dump all expressions and occurrences that are currently in the | |
| 474 expression hash table to FILE. */ | |
| 475 | |
| 476 /* This helper is called via htab_traverse. */ | |
| 111 | 477 int |
| 478 dump_expr_hash_table_entry (expr **slot, FILE *file) | |
| 0 | 479 { |
| 111 | 480 struct expr *exprs = *slot; |
| 0 | 481 struct occr *occr; |
| 482 | |
| 483 fprintf (file, "expr: "); | |
| 111 | 484 print_rtl (file, exprs->expr); |
| 485 fprintf (file,"\nhashcode: %u\n", exprs->hash); | |
| 0 | 486 fprintf (file,"list of occurrences:\n"); |
| 111 | 487 occr = exprs->avail_occr; |
| 0 | 488 while (occr) |
| 489 { | |
| 111 | 490 rtx_insn *insn = occr->insn; |
| 0 | 491 print_rtl_single (file, insn); |
| 492 fprintf (file, "\n"); | |
| 493 occr = occr->next; | |
| 494 } | |
| 495 fprintf (file, "\n"); | |
| 496 return 1; | |
| 497 } | |
| 498 | |
| 499 static void | |
| 500 dump_hash_table (FILE *file) | |
| 501 { | |
| 502 fprintf (file, "\n\nexpression hash table\n"); | |
| 503 fprintf (file, "size %ld, %ld elements, %f collision/search ratio\n", | |
| 111 | 504 (long) expr_table->size (), |
| 505 (long) expr_table->elements (), | |
| 506 expr_table->collisions ()); | |
| 507 if (expr_table->elements () > 0) | |
| 0 | 508 { |
| 509 fprintf (file, "\n\ntable entries:\n"); | |
| 111 | 510 expr_table->traverse <FILE *, dump_expr_hash_table_entry> (file); |
| 0 | 511 } |
| 512 fprintf (file, "\n"); | |
| 513 } | |
| 514 | |
| 515 /* Return true if register X is recorded as being set by an instruction | |
| 516 whose CUID is greater than the one given. */ | |
| 517 | |
| 518 static bool | |
| 519 reg_changed_after_insn_p (rtx x, int cuid) | |
| 520 { | |
| 521 unsigned int regno, end_regno; | |
| 522 | |
| 523 regno = REGNO (x); | |
| 111 | 524 end_regno = END_REGNO (x); |
| 0 | 525 do |
| 526 if (reg_avail_info[regno] > cuid) | |
| 527 return true; | |
| 528 while (++regno < end_regno); | |
| 529 return false; | |
| 530 } | |
| 531 | |
| 532 /* Return nonzero if the operands of expression X are unchanged | |
| 533 1) from the start of INSN's basic block up to but not including INSN | |
| 534 if AFTER_INSN is false, or | |
| 535 2) from INSN to the end of INSN's basic block if AFTER_INSN is true. */ | |
| 536 | |
| 537 static bool | |
| 111 | 538 oprs_unchanged_p (rtx x, rtx_insn *insn, bool after_insn) |
| 0 | 539 { |
| 540 int i, j; | |
| 541 enum rtx_code code; | |
| 542 const char *fmt; | |
| 543 | |
| 544 if (x == 0) | |
| 545 return 1; | |
| 546 | |
| 547 code = GET_CODE (x); | |
| 548 switch (code) | |
| 549 { | |
| 550 case REG: | |
| 551 /* We are called after register allocation. */ | |
| 552 gcc_assert (REGNO (x) < FIRST_PSEUDO_REGISTER); | |
| 553 if (after_insn) | |
| 554 return !reg_changed_after_insn_p (x, INSN_CUID (insn) - 1); | |
| 555 else | |
| 556 return !reg_changed_after_insn_p (x, 0); | |
| 557 | |
| 558 case MEM: | |
| 559 if (load_killed_in_block_p (INSN_CUID (insn), x, after_insn)) | |
| 560 return 0; | |
| 561 else | |
| 562 return oprs_unchanged_p (XEXP (x, 0), insn, after_insn); | |
| 563 | |
| 564 case PC: | |
| 565 case CC0: /*FIXME*/ | |
| 566 case CONST: | |
| 111 | 567 CASE_CONST_ANY: |
| 0 | 568 case SYMBOL_REF: |
| 569 case LABEL_REF: | |
| 570 case ADDR_VEC: | |
| 571 case ADDR_DIFF_VEC: | |
| 572 return 1; | |
| 573 | |
| 574 case PRE_DEC: | |
| 575 case PRE_INC: | |
| 576 case POST_DEC: | |
| 577 case POST_INC: | |
| 578 case PRE_MODIFY: | |
| 579 case POST_MODIFY: | |
| 580 if (after_insn) | |
| 581 return 0; | |
| 582 break; | |
| 583 | |
| 584 default: | |
| 585 break; | |
| 586 } | |
| 587 | |
| 588 for (i = GET_RTX_LENGTH (code) - 1, fmt = GET_RTX_FORMAT (code); i >= 0; i--) | |
| 589 { | |
| 590 if (fmt[i] == 'e') | |
| 591 { | |
| 592 if (! oprs_unchanged_p (XEXP (x, i), insn, after_insn)) | |
| 593 return 0; | |
| 594 } | |
| 595 else if (fmt[i] == 'E') | |
| 596 for (j = 0; j < XVECLEN (x, i); j++) | |
| 597 if (! oprs_unchanged_p (XVECEXP (x, i, j), insn, after_insn)) | |
| 598 return 0; | |
| 599 } | |
| 600 | |
| 601 return 1; | |
| 602 } | |
| 603 | |
| 604 | |
| 605 /* Used for communication between find_mem_conflicts and | |
| 606 load_killed_in_block_p. Nonzero if find_mem_conflicts finds a | |
| 607 conflict between two memory references. | |
| 608 This is a bit of a hack to work around the limitations of note_stores. */ | |
| 609 static int mems_conflict_p; | |
| 610 | |
| 611 /* DEST is the output of an instruction. If it is a memory reference, and | |
| 612 possibly conflicts with the load found in DATA, then set mems_conflict_p | |
| 613 to a nonzero value. */ | |
| 614 | |
| 615 static void | |
| 616 find_mem_conflicts (rtx dest, const_rtx setter ATTRIBUTE_UNUSED, | |
| 617 void *data) | |
| 618 { | |
| 619 rtx mem_op = (rtx) data; | |
| 620 | |
| 621 while (GET_CODE (dest) == SUBREG | |
| 622 || GET_CODE (dest) == ZERO_EXTRACT | |
| 623 || GET_CODE (dest) == STRICT_LOW_PART) | |
| 624 dest = XEXP (dest, 0); | |
| 625 | |
| 626 /* If DEST is not a MEM, then it will not conflict with the load. Note | |
| 627 that function calls are assumed to clobber memory, but are handled | |
| 628 elsewhere. */ | |
| 629 if (! MEM_P (dest)) | |
| 630 return; | |
| 631 | |
| 111 | 632 if (true_dependence (dest, GET_MODE (dest), mem_op)) |
| 0 | 633 mems_conflict_p = 1; |
| 634 } | |
| 635 | |
| 636 | |
| 637 /* Return nonzero if the expression in X (a memory reference) is killed | |
| 638 in the current basic block before (if AFTER_INSN is false) or after | |
| 639 (if AFTER_INSN is true) the insn with the CUID in UID_LIMIT. | |
| 640 | |
| 641 This function assumes that the modifies_mem table is flushed when | |
| 642 the hash table construction or redundancy elimination phases start | |
| 643 processing a new basic block. */ | |
| 644 | |
| 645 static int | |
| 646 load_killed_in_block_p (int uid_limit, rtx x, bool after_insn) | |
| 647 { | |
| 648 struct modifies_mem *list_entry = modifies_mem_list; | |
| 649 | |
| 650 while (list_entry) | |
| 651 { | |
| 111 | 652 rtx_insn *setter = list_entry->insn; |
| 0 | 653 |
| 654 /* Ignore entries in the list that do not apply. */ | |
| 655 if ((after_insn | |
| 656 && INSN_CUID (setter) < uid_limit) | |
| 657 || (! after_insn | |
| 658 && INSN_CUID (setter) > uid_limit)) | |
| 659 { | |
| 660 list_entry = list_entry->next; | |
| 661 continue; | |
| 662 } | |
| 663 | |
| 664 /* If SETTER is a call everything is clobbered. Note that calls | |
| 665 to pure functions are never put on the list, so we need not | |
| 666 worry about them. */ | |
| 667 if (CALL_P (setter)) | |
| 668 return 1; | |
| 669 | |
| 670 /* SETTER must be an insn of some kind that sets memory. Call | |
| 671 note_stores to examine each hunk of memory that is modified. | |
| 672 It will set mems_conflict_p to nonzero if there may be a | |
| 673 conflict between X and SETTER. */ | |
| 674 mems_conflict_p = 0; | |
| 675 note_stores (PATTERN (setter), find_mem_conflicts, x); | |
| 676 if (mems_conflict_p) | |
| 677 return 1; | |
| 678 | |
| 679 list_entry = list_entry->next; | |
| 680 } | |
| 681 return 0; | |
| 682 } | |
| 683 | |
| 684 | |
| 685 /* Record register first/last/block set information for REGNO in INSN. */ | |
| 686 | |
| 687 static inline void | |
| 111 | 688 record_last_reg_set_info (rtx_insn *insn, rtx reg) |
| 0 | 689 { |
| 690 unsigned int regno, end_regno; | |
| 691 | |
| 692 regno = REGNO (reg); | |
| 111 | 693 end_regno = END_REGNO (reg); |
| 0 | 694 do |
| 695 reg_avail_info[regno] = INSN_CUID (insn); | |
| 696 while (++regno < end_regno); | |
| 697 } | |
| 698 | |
| 699 static inline void | |
| 111 | 700 record_last_reg_set_info_regno (rtx_insn *insn, int regno) |
| 0 | 701 { |
| 702 reg_avail_info[regno] = INSN_CUID (insn); | |
| 703 } | |
| 704 | |
| 705 | |
| 706 /* Record memory modification information for INSN. We do not actually care | |
| 707 about the memory location(s) that are set, or even how they are set (consider | |
| 708 a CALL_INSN). We merely need to record which insns modify memory. */ | |
| 709 | |
| 710 static void | |
| 111 | 711 record_last_mem_set_info (rtx_insn *insn) |
| 0 | 712 { |
| 713 struct modifies_mem *list_entry; | |
| 714 | |
| 715 list_entry = (struct modifies_mem *) obstack_alloc (&modifies_mem_obstack, | |
| 716 sizeof (struct modifies_mem)); | |
| 717 list_entry->insn = insn; | |
| 718 list_entry->next = modifies_mem_list; | |
| 719 modifies_mem_list = list_entry; | |
| 111 | 720 |
| 721 record_last_mem_set_info_common (insn, modify_mem_list, | |
| 722 canon_modify_mem_list, | |
| 723 modify_mem_list_set, | |
| 724 blocks_with_calls); | |
| 0 | 725 } |
| 726 | |
| 727 /* Called from compute_hash_table via note_stores to handle one | |
| 728 SET or CLOBBER in an insn. DATA is really the instruction in which | |
| 729 the SET is taking place. */ | |
| 730 | |
| 731 static void | |
| 732 record_last_set_info (rtx dest, const_rtx setter ATTRIBUTE_UNUSED, void *data) | |
| 733 { | |
| 111 | 734 rtx_insn *last_set_insn = (rtx_insn *) data; |
| 0 | 735 |
| 736 if (GET_CODE (dest) == SUBREG) | |
| 737 dest = SUBREG_REG (dest); | |
| 738 | |
| 739 if (REG_P (dest)) | |
| 740 record_last_reg_set_info (last_set_insn, dest); | |
| 741 else if (MEM_P (dest)) | |
| 742 { | |
| 743 /* Ignore pushes, they don't clobber memory. They may still | |
| 744 clobber the stack pointer though. Some targets do argument | |
| 745 pushes without adding REG_INC notes. See e.g. PR25196, | |
| 746 where a pushsi2 on i386 doesn't have REG_INC notes. Note | |
| 747 such changes here too. */ | |
| 748 if (! push_operand (dest, GET_MODE (dest))) | |
| 749 record_last_mem_set_info (last_set_insn); | |
| 750 else | |
| 751 record_last_reg_set_info_regno (last_set_insn, STACK_POINTER_REGNUM); | |
| 752 } | |
| 753 } | |
| 754 | |
| 755 | |
| 756 /* Reset tables used to keep track of what's still available since the | |
| 757 start of the block. */ | |
| 758 | |
| 759 static void | |
| 760 reset_opr_set_tables (void) | |
| 761 { | |
| 762 memset (reg_avail_info, 0, FIRST_PSEUDO_REGISTER * sizeof (int)); | |
| 763 obstack_free (&modifies_mem_obstack, modifies_mem_obstack_bottom); | |
| 764 modifies_mem_list = NULL; | |
| 765 } | |
| 766 | |
| 767 | |
| 768 /* Record things set by INSN. | |
| 769 This data is used by oprs_unchanged_p. */ | |
| 770 | |
| 771 static void | |
| 111 | 772 record_opr_changes (rtx_insn *insn) |
| 0 | 773 { |
| 774 rtx note; | |
| 775 | |
| 776 /* Find all stores and record them. */ | |
| 777 note_stores (PATTERN (insn), record_last_set_info, insn); | |
| 778 | |
| 779 /* Also record autoincremented REGs for this insn as changed. */ | |
| 780 for (note = REG_NOTES (insn); note; note = XEXP (note, 1)) | |
| 781 if (REG_NOTE_KIND (note) == REG_INC) | |
| 782 record_last_reg_set_info (insn, XEXP (note, 0)); | |
| 783 | |
| 784 /* Finally, if this is a call, record all call clobbers. */ | |
| 785 if (CALL_P (insn)) | |
| 786 { | |
| 787 unsigned int regno; | |
| 788 rtx link, x; | |
| 111 | 789 hard_reg_set_iterator hrsi; |
| 790 EXECUTE_IF_SET_IN_HARD_REG_SET (regs_invalidated_by_call, 0, regno, hrsi) | |
| 791 record_last_reg_set_info_regno (insn, regno); | |
| 0 | 792 |
| 793 for (link = CALL_INSN_FUNCTION_USAGE (insn); link; link = XEXP (link, 1)) | |
| 131 | 794 { |
| 795 gcc_assert (GET_CODE (XEXP (link, 0)) != CLOBBER_HIGH); | |
| 796 if (GET_CODE (XEXP (link, 0)) == CLOBBER) | |
| 797 { | |
| 798 x = XEXP (XEXP (link, 0), 0); | |
| 799 if (REG_P (x)) | |
| 800 { | |
| 801 gcc_assert (HARD_REGISTER_P (x)); | |
| 802 record_last_reg_set_info (insn, x); | |
| 803 } | |
| 804 } | |
| 805 } | |
| 0 | 806 |
| 807 if (! RTL_CONST_OR_PURE_CALL_P (insn)) | |
| 808 record_last_mem_set_info (insn); | |
| 809 } | |
| 810 } | |
| 811 | |
| 812 | |
| 813 /* Scan the pattern of INSN and add an entry to the hash TABLE. | |
| 814 After reload we are interested in loads/stores only. */ | |
| 815 | |
| 816 static void | |
| 111 | 817 hash_scan_set (rtx_insn *insn) |
| 0 | 818 { |
| 819 rtx pat = PATTERN (insn); | |
| 820 rtx src = SET_SRC (pat); | |
| 821 rtx dest = SET_DEST (pat); | |
| 822 | |
| 823 /* We are only interested in loads and stores. */ | |
| 824 if (! MEM_P (src) && ! MEM_P (dest)) | |
| 825 return; | |
| 826 | |
| 827 /* Don't mess with jumps and nops. */ | |
| 828 if (JUMP_P (insn) || set_noop_p (pat)) | |
| 829 return; | |
| 830 | |
| 831 if (REG_P (dest)) | |
| 832 { | |
| 833 if (/* Don't CSE something if we can't do a reg/reg copy. */ | |
| 834 can_copy_p (GET_MODE (dest)) | |
| 835 /* Is SET_SRC something we want to gcse? */ | |
| 836 && general_operand (src, GET_MODE (src)) | |
| 837 #ifdef STACK_REGS | |
| 838 /* Never consider insns touching the register stack. It may | |
| 839 create situations that reg-stack cannot handle (e.g. a stack | |
| 840 register live across an abnormal edge). */ | |
| 841 && (REGNO (dest) < FIRST_STACK_REG || REGNO (dest) > LAST_STACK_REG) | |
| 842 #endif | |
| 843 /* An expression is not available if its operands are | |
| 844 subsequently modified, including this insn. */ | |
| 845 && oprs_unchanged_p (src, insn, true)) | |
| 846 { | |
| 847 insert_expr_in_table (src, insn); | |
| 848 } | |
| 849 } | |
| 850 else if (REG_P (src)) | |
| 851 { | |
| 852 /* Only record sets of pseudo-regs in the hash table. */ | |
| 853 if (/* Don't CSE something if we can't do a reg/reg copy. */ | |
| 854 can_copy_p (GET_MODE (src)) | |
| 855 /* Is SET_DEST something we want to gcse? */ | |
| 856 && general_operand (dest, GET_MODE (dest)) | |
| 857 #ifdef STACK_REGS | |
| 858 /* As above for STACK_REGS. */ | |
| 859 && (REGNO (src) < FIRST_STACK_REG || REGNO (src) > LAST_STACK_REG) | |
| 860 #endif | |
| 861 && ! (flag_float_store && FLOAT_MODE_P (GET_MODE (dest))) | |
| 862 /* Check if the memory expression is killed after insn. */ | |
| 863 && ! load_killed_in_block_p (INSN_CUID (insn) + 1, dest, true) | |
| 864 && oprs_unchanged_p (XEXP (dest, 0), insn, true)) | |
| 865 { | |
| 866 insert_expr_in_table (dest, insn); | |
| 867 } | |
| 868 } | |
| 869 } | |
| 870 | |
| 871 | |
| 872 /* Create hash table of memory expressions available at end of basic | |
| 873 blocks. Basically you should think of this hash table as the | |
| 874 representation of AVAIL_OUT. This is the set of expressions that | |
| 875 is generated in a basic block and not killed before the end of the | |
| 876 same basic block. Notice that this is really a local computation. */ | |
| 877 | |
| 878 static void | |
| 879 compute_hash_table (void) | |
| 880 { | |
| 881 basic_block bb; | |
| 882 | |
| 111 | 883 FOR_EACH_BB_FN (bb, cfun) |
| 0 | 884 { |
| 111 | 885 rtx_insn *insn; |
| 0 | 886 |
| 887 /* First pass over the instructions records information used to | |
| 888 determine when registers and memory are last set. | |
| 889 Since we compute a "local" AVAIL_OUT, reset the tables that | |
| 890 help us keep track of what has been modified since the start | |
| 891 of the block. */ | |
| 892 reset_opr_set_tables (); | |
| 893 FOR_BB_INSNS (bb, insn) | |
| 894 { | |
| 895 if (INSN_P (insn)) | |
| 896 record_opr_changes (insn); | |
| 897 } | |
| 898 | |
| 899 /* The next pass actually builds the hash table. */ | |
| 900 FOR_BB_INSNS (bb, insn) | |
| 901 if (INSN_P (insn) && GET_CODE (PATTERN (insn)) == SET) | |
| 902 hash_scan_set (insn); | |
| 903 } | |
| 904 } | |
| 905 | |
| 906 | |
| 907 /* Check if register REG is killed in any insn waiting to be inserted on | |
| 908 edge E. This function is required to check that our data flow analysis | |
| 909 is still valid prior to commit_edge_insertions. */ | |
| 910 | |
| 911 static bool | |
| 912 reg_killed_on_edge (rtx reg, edge e) | |
| 913 { | |
| 111 | 914 rtx_insn *insn; |
| 0 | 915 |
| 916 for (insn = e->insns.r; insn; insn = NEXT_INSN (insn)) | |
| 917 if (INSN_P (insn) && reg_set_p (reg, insn)) | |
| 918 return true; | |
| 919 | |
| 920 return false; | |
| 921 } | |
| 922 | |
| 923 /* Similar to above - check if register REG is used in any insn waiting | |
| 924 to be inserted on edge E. | |
| 925 Assumes no such insn can be a CALL_INSN; if so call reg_used_between_p | |
| 926 with PREV(insn),NEXT(insn) instead of calling reg_overlap_mentioned_p. */ | |
| 927 | |
| 928 static bool | |
| 929 reg_used_on_edge (rtx reg, edge e) | |
| 930 { | |
| 111 | 931 rtx_insn *insn; |
| 0 | 932 |
| 933 for (insn = e->insns.r; insn; insn = NEXT_INSN (insn)) | |
| 934 if (INSN_P (insn) && reg_overlap_mentioned_p (reg, PATTERN (insn))) | |
| 935 return true; | |
| 936 | |
| 937 return false; | |
| 938 } | |
| 939 | |
| 940 /* Return the loaded/stored register of a load/store instruction. */ | |
| 941 | |
| 942 static rtx | |
| 111 | 943 get_avail_load_store_reg (rtx_insn *insn) |
| 0 | 944 { |
| 945 if (REG_P (SET_DEST (PATTERN (insn)))) | |
| 946 /* A load. */ | |
| 111 | 947 return SET_DEST (PATTERN (insn)); |
| 0 | 948 else |
| 949 { | |
| 950 /* A store. */ | |
| 951 gcc_assert (REG_P (SET_SRC (PATTERN (insn)))); | |
| 952 return SET_SRC (PATTERN (insn)); | |
| 953 } | |
| 954 } | |
| 955 | |
| 956 /* Return nonzero if the predecessors of BB are "well behaved". */ | |
| 957 | |
| 958 static bool | |
| 959 bb_has_well_behaved_predecessors (basic_block bb) | |
| 960 { | |
| 961 edge pred; | |
| 962 edge_iterator ei; | |
| 963 | |
| 964 if (EDGE_COUNT (bb->preds) == 0) | |
| 965 return false; | |
| 966 | |
| 967 FOR_EACH_EDGE (pred, ei, bb->preds) | |
| 968 { | |
| 111 | 969 /* commit_one_edge_insertion refuses to insert on abnormal edges even if |
| 970 the source has only one successor so EDGE_CRITICAL_P is too weak. */ | |
| 971 if ((pred->flags & EDGE_ABNORMAL) && !single_pred_p (pred->dest)) | |
| 0 | 972 return false; |
| 973 | |
| 111 | 974 if ((pred->flags & EDGE_ABNORMAL_CALL) && cfun->has_nonlocal_label) |
| 975 return false; | |
| 976 | |
| 977 if (tablejump_p (BB_END (pred->src), NULL, NULL)) | |
| 0 | 978 return false; |
| 979 } | |
| 980 return true; | |
| 981 } | |
| 982 | |
| 983 | |
| 984 /* Search for the occurrences of expression in BB. */ | |
| 985 | |
| 986 static struct occr* | |
| 111 | 987 get_bb_avail_insn (basic_block bb, struct occr *orig_occr, int bitmap_index) |
| 0 | 988 { |
| 111 | 989 struct occr *occr = orig_occr; |
| 990 | |
| 0 | 991 for (; occr != NULL; occr = occr->next) |
| 992 if (BLOCK_FOR_INSN (occr->insn) == bb) | |
| 993 return occr; | |
| 111 | 994 |
| 995 /* If we could not find an occurrence in BB, see if BB | |
| 996 has a single predecessor with an occurrence that is | |
| 997 transparent through BB. */ | |
| 998 if (single_pred_p (bb) | |
| 999 && bitmap_bit_p (transp[bb->index], bitmap_index) | |
| 1000 && (occr = get_bb_avail_insn (single_pred (bb), orig_occr, bitmap_index))) | |
| 1001 { | |
| 1002 rtx avail_reg = get_avail_load_store_reg (occr->insn); | |
| 1003 if (!reg_set_between_p (avail_reg, | |
| 1004 PREV_INSN (BB_HEAD (bb)), | |
| 1005 NEXT_INSN (BB_END (bb))) | |
| 1006 && !reg_killed_on_edge (avail_reg, single_pred_edge (bb))) | |
| 1007 return occr; | |
| 1008 } | |
| 1009 | |
| 0 | 1010 return NULL; |
| 1011 } | |
| 1012 | |
| 1013 | |
| 111 | 1014 /* This helper is called via htab_traverse. */ |
| 1015 int | |
| 1016 compute_expr_transp (expr **slot, FILE *dump_file ATTRIBUTE_UNUSED) | |
| 1017 { | |
| 1018 struct expr *expr = *slot; | |
| 1019 | |
| 1020 compute_transp (expr->expr, expr->bitmap_index, transp, | |
| 1021 blocks_with_calls, modify_mem_list_set, | |
| 1022 canon_modify_mem_list); | |
| 1023 return 1; | |
| 1024 } | |
| 1025 | |
| 0 | 1026 /* This handles the case where several stores feed a partially redundant |
| 1027 load. It checks if the redundancy elimination is possible and if it's | |
| 1028 worth it. | |
| 1029 | |
| 1030 Redundancy elimination is possible if, | |
| 1031 1) None of the operands of an insn have been modified since the start | |
| 1032 of the current basic block. | |
| 1033 2) In any predecessor of the current basic block, the same expression | |
| 1034 is generated. | |
| 1035 | |
| 1036 See the function body for the heuristics that determine if eliminating | |
| 1037 a redundancy is also worth doing, assuming it is possible. */ | |
| 1038 | |
| 1039 static void | |
| 111 | 1040 eliminate_partially_redundant_load (basic_block bb, rtx_insn *insn, |
| 0 | 1041 struct expr *expr) |
| 1042 { | |
| 1043 edge pred; | |
| 111 | 1044 rtx_insn *avail_insn = NULL; |
| 0 | 1045 rtx avail_reg; |
| 1046 rtx dest, pat; | |
| 1047 struct occr *a_occr; | |
| 1048 struct unoccr *occr, *avail_occrs = NULL; | |
| 1049 struct unoccr *unoccr, *unavail_occrs = NULL, *rollback_unoccr = NULL; | |
| 1050 int npred_ok = 0; | |
| 111 | 1051 profile_count ok_count = profile_count::zero (); |
| 1052 /* Redundant load execution count. */ | |
| 1053 profile_count critical_count = profile_count::zero (); | |
| 1054 /* Execution count of critical edges. */ | |
| 0 | 1055 edge_iterator ei; |
| 1056 bool critical_edge_split = false; | |
| 1057 | |
| 1058 /* The execution count of the loads to be added to make the | |
| 1059 load fully redundant. */ | |
| 111 | 1060 profile_count not_ok_count = profile_count::zero (); |
| 0 | 1061 basic_block pred_bb; |
| 1062 | |
| 1063 pat = PATTERN (insn); | |
| 1064 dest = SET_DEST (pat); | |
| 1065 | |
| 1066 /* Check that the loaded register is not used, set, or killed from the | |
| 1067 beginning of the block. */ | |
| 1068 if (reg_changed_after_insn_p (dest, 0) | |
| 1069 || reg_used_between_p (dest, PREV_INSN (BB_HEAD (bb)), insn)) | |
| 1070 return; | |
| 1071 | |
| 1072 /* Check potential for replacing load with copy for predecessors. */ | |
| 1073 FOR_EACH_EDGE (pred, ei, bb->preds) | |
| 1074 { | |
| 111 | 1075 rtx_insn *next_pred_bb_end; |
| 0 | 1076 |
| 111 | 1077 avail_insn = NULL; |
| 0 | 1078 avail_reg = NULL_RTX; |
| 1079 pred_bb = pred->src; | |
| 111 | 1080 for (a_occr = get_bb_avail_insn (pred_bb, |
| 1081 expr->avail_occr, | |
| 1082 expr->bitmap_index); | |
| 1083 a_occr; | |
| 1084 a_occr = get_bb_avail_insn (pred_bb, | |
| 1085 a_occr->next, | |
| 1086 expr->bitmap_index)) | |
| 0 | 1087 { |
| 1088 /* Check if the loaded register is not used. */ | |
| 1089 avail_insn = a_occr->insn; | |
| 1090 avail_reg = get_avail_load_store_reg (avail_insn); | |
| 1091 gcc_assert (avail_reg); | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
1092 |
| 0 | 1093 /* Make sure we can generate a move from register avail_reg to |
| 1094 dest. */ | |
| 111 | 1095 rtx_insn *move = gen_move_insn (copy_rtx (dest), |
| 1096 copy_rtx (avail_reg)); | |
| 1097 extract_insn (move); | |
| 1098 if (! constrain_operands (1, get_preferred_alternatives (insn, | |
| 1099 pred_bb)) | |
| 0 | 1100 || reg_killed_on_edge (avail_reg, pred) |
| 1101 || reg_used_on_edge (dest, pred)) | |
| 1102 { | |
| 1103 avail_insn = NULL; | |
| 1104 continue; | |
| 1105 } | |
| 111 | 1106 next_pred_bb_end = NEXT_INSN (BB_END (BLOCK_FOR_INSN (avail_insn))); |
| 0 | 1107 if (!reg_set_between_p (avail_reg, avail_insn, next_pred_bb_end)) |
| 1108 /* AVAIL_INSN remains non-null. */ | |
| 1109 break; | |
| 1110 else | |
| 1111 avail_insn = NULL; | |
| 1112 } | |
| 1113 | |
| 111 | 1114 if (EDGE_CRITICAL_P (pred) && pred->count ().initialized_p ()) |
| 1115 critical_count += pred->count (); | |
| 0 | 1116 |
| 1117 if (avail_insn != NULL_RTX) | |
| 1118 { | |
| 1119 npred_ok++; | |
| 111 | 1120 if (pred->count ().initialized_p ()) |
| 1121 ok_count = ok_count + pred->count (); | |
| 0 | 1122 if (! set_noop_p (PATTERN (gen_move_insn (copy_rtx (dest), |
| 1123 copy_rtx (avail_reg))))) | |
| 1124 { | |
| 1125 /* Check if there is going to be a split. */ | |
| 1126 if (EDGE_CRITICAL_P (pred)) | |
| 1127 critical_edge_split = true; | |
| 1128 } | |
| 1129 else /* Its a dead move no need to generate. */ | |
| 1130 continue; | |
| 1131 occr = (struct unoccr *) obstack_alloc (&unoccr_obstack, | |
| 1132 sizeof (struct unoccr)); | |
| 1133 occr->insn = avail_insn; | |
| 1134 occr->pred = pred; | |
| 1135 occr->next = avail_occrs; | |
| 1136 avail_occrs = occr; | |
| 1137 if (! rollback_unoccr) | |
| 1138 rollback_unoccr = occr; | |
| 1139 } | |
| 1140 else | |
| 1141 { | |
| 1142 /* Adding a load on a critical edge will cause a split. */ | |
| 1143 if (EDGE_CRITICAL_P (pred)) | |
| 1144 critical_edge_split = true; | |
| 111 | 1145 if (pred->count ().initialized_p ()) |
| 1146 not_ok_count = not_ok_count + pred->count (); | |
| 0 | 1147 unoccr = (struct unoccr *) obstack_alloc (&unoccr_obstack, |
| 1148 sizeof (struct unoccr)); | |
| 111 | 1149 unoccr->insn = NULL; |
| 0 | 1150 unoccr->pred = pred; |
| 1151 unoccr->next = unavail_occrs; | |
| 1152 unavail_occrs = unoccr; | |
| 1153 if (! rollback_unoccr) | |
| 1154 rollback_unoccr = unoccr; | |
| 1155 } | |
| 1156 } | |
| 1157 | |
| 1158 if (/* No load can be replaced by copy. */ | |
| 1159 npred_ok == 0 | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
1160 /* Prevent exploding the code. */ |
| 0 | 1161 || (optimize_bb_for_size_p (bb) && npred_ok > 1) |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
1162 /* If we don't have profile information we cannot tell if splitting |
| 0 | 1163 a critical edge is profitable or not so don't do it. */ |
| 131 | 1164 || ((!profile_info || profile_status_for_fn (cfun) != PROFILE_READ |
| 0 | 1165 || targetm.cannot_modify_jumps_p ()) |
| 1166 && critical_edge_split)) | |
| 1167 goto cleanup; | |
| 1168 | |
| 1169 /* Check if it's worth applying the partial redundancy elimination. */ | |
| 111 | 1170 if (ok_count.to_gcov_type () |
| 1171 < GCSE_AFTER_RELOAD_PARTIAL_FRACTION * not_ok_count.to_gcov_type ()) | |
| 0 | 1172 goto cleanup; |
| 111 | 1173 if (ok_count.to_gcov_type () |
| 1174 < GCSE_AFTER_RELOAD_CRITICAL_FRACTION * critical_count.to_gcov_type ()) | |
| 0 | 1175 goto cleanup; |
| 1176 | |
| 1177 /* Generate moves to the loaded register from where | |
| 1178 the memory is available. */ | |
| 1179 for (occr = avail_occrs; occr; occr = occr->next) | |
| 1180 { | |
| 1181 avail_insn = occr->insn; | |
| 1182 pred = occr->pred; | |
| 1183 /* Set avail_reg to be the register having the value of the | |
| 1184 memory. */ | |
| 1185 avail_reg = get_avail_load_store_reg (avail_insn); | |
| 1186 gcc_assert (avail_reg); | |
| 1187 | |
| 1188 insert_insn_on_edge (gen_move_insn (copy_rtx (dest), | |
| 1189 copy_rtx (avail_reg)), | |
| 1190 pred); | |
| 1191 stats.moves_inserted++; | |
| 1192 | |
| 1193 if (dump_file) | |
| 1194 fprintf (dump_file, | |
| 1195 "generating move from %d to %d on edge from %d to %d\n", | |
| 1196 REGNO (avail_reg), | |
| 1197 REGNO (dest), | |
| 1198 pred->src->index, | |
| 1199 pred->dest->index); | |
| 1200 } | |
| 1201 | |
| 1202 /* Regenerate loads where the memory is unavailable. */ | |
| 1203 for (unoccr = unavail_occrs; unoccr; unoccr = unoccr->next) | |
| 1204 { | |
| 1205 pred = unoccr->pred; | |
| 1206 insert_insn_on_edge (copy_insn (PATTERN (insn)), pred); | |
| 1207 stats.copies_inserted++; | |
| 1208 | |
| 1209 if (dump_file) | |
| 1210 { | |
| 1211 fprintf (dump_file, | |
| 1212 "generating on edge from %d to %d a copy of load: ", | |
| 1213 pred->src->index, | |
| 1214 pred->dest->index); | |
| 1215 print_rtl (dump_file, PATTERN (insn)); | |
| 1216 fprintf (dump_file, "\n"); | |
| 1217 } | |
| 1218 } | |
| 1219 | |
| 1220 /* Delete the insn if it is not available in this block and mark it | |
| 1221 for deletion if it is available. If insn is available it may help | |
| 1222 discover additional redundancies, so mark it for later deletion. */ | |
| 111 | 1223 for (a_occr = get_bb_avail_insn (bb, expr->avail_occr, expr->bitmap_index); |
| 0 | 1224 a_occr && (a_occr->insn != insn); |
| 111 | 1225 a_occr = get_bb_avail_insn (bb, a_occr->next, expr->bitmap_index)) |
| 1226 ; | |
| 0 | 1227 |
| 1228 if (!a_occr) | |
| 1229 { | |
| 1230 stats.insns_deleted++; | |
| 1231 | |
| 1232 if (dump_file) | |
| 1233 { | |
| 1234 fprintf (dump_file, "deleting insn:\n"); | |
| 1235 print_rtl_single (dump_file, insn); | |
| 1236 fprintf (dump_file, "\n"); | |
| 1237 } | |
| 1238 delete_insn (insn); | |
| 1239 } | |
| 1240 else | |
| 1241 a_occr->deleted_p = 1; | |
| 1242 | |
| 1243 cleanup: | |
| 1244 if (rollback_unoccr) | |
| 1245 obstack_free (&unoccr_obstack, rollback_unoccr); | |
| 1246 } | |
| 1247 | |
| 1248 /* Performing the redundancy elimination as described before. */ | |
| 1249 | |
| 1250 static void | |
| 1251 eliminate_partially_redundant_loads (void) | |
| 1252 { | |
| 111 | 1253 rtx_insn *insn; |
| 0 | 1254 basic_block bb; |
| 1255 | |
| 1256 /* Note we start at block 1. */ | |
| 1257 | |
| 111 | 1258 if (ENTRY_BLOCK_PTR_FOR_FN (cfun)->next_bb == EXIT_BLOCK_PTR_FOR_FN (cfun)) |
| 0 | 1259 return; |
| 1260 | |
| 1261 FOR_BB_BETWEEN (bb, | |
| 111 | 1262 ENTRY_BLOCK_PTR_FOR_FN (cfun)->next_bb->next_bb, |
| 1263 EXIT_BLOCK_PTR_FOR_FN (cfun), | |
| 0 | 1264 next_bb) |
| 1265 { | |
| 1266 /* Don't try anything on basic blocks with strange predecessors. */ | |
| 1267 if (! bb_has_well_behaved_predecessors (bb)) | |
| 1268 continue; | |
| 1269 | |
| 1270 /* Do not try anything on cold basic blocks. */ | |
| 1271 if (optimize_bb_for_size_p (bb)) | |
| 1272 continue; | |
| 1273 | |
| 1274 /* Reset the table of things changed since the start of the current | |
| 1275 basic block. */ | |
| 1276 reset_opr_set_tables (); | |
| 1277 | |
| 1278 /* Look at all insns in the current basic block and see if there are | |
| 1279 any loads in it that we can record. */ | |
| 1280 FOR_BB_INSNS (bb, insn) | |
| 1281 { | |
| 1282 /* Is it a load - of the form (set (reg) (mem))? */ | |
| 1283 if (NONJUMP_INSN_P (insn) | |
| 1284 && GET_CODE (PATTERN (insn)) == SET | |
| 1285 && REG_P (SET_DEST (PATTERN (insn))) | |
| 1286 && MEM_P (SET_SRC (PATTERN (insn)))) | |
| 1287 { | |
| 1288 rtx pat = PATTERN (insn); | |
| 1289 rtx src = SET_SRC (pat); | |
| 1290 struct expr *expr; | |
| 1291 | |
| 1292 if (!MEM_VOLATILE_P (src) | |
| 1293 && GET_MODE (src) != BLKmode | |
| 1294 && general_operand (src, GET_MODE (src)) | |
| 1295 /* Are the operands unchanged since the start of the | |
| 1296 block? */ | |
| 1297 && oprs_unchanged_p (src, insn, false) | |
|
67
f6334be47118
update gcc from gcc-4.6-20100522 to gcc-4.6-20110318
nobuyasu <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
63
diff
changeset
|
1298 && !(cfun->can_throw_non_call_exceptions && may_trap_p (src)) |
| 0 | 1299 && !side_effects_p (src) |
| 1300 /* Is the expression recorded? */ | |
| 1301 && (expr = lookup_expr_in_table (src)) != NULL) | |
| 1302 { | |
| 1303 /* We now have a load (insn) and an available memory at | |
| 1304 its BB start (expr). Try to remove the loads if it is | |
| 1305 redundant. */ | |
| 1306 eliminate_partially_redundant_load (bb, insn, expr); | |
| 1307 } | |
| 1308 } | |
| 1309 | |
| 1310 /* Keep track of everything modified by this insn, so that we | |
| 1311 know what has been modified since the start of the current | |
| 1312 basic block. */ | |
| 1313 if (INSN_P (insn)) | |
| 1314 record_opr_changes (insn); | |
| 1315 } | |
| 1316 } | |
| 1317 | |
| 1318 commit_edge_insertions (); | |
| 1319 } | |
| 1320 | |
| 1321 /* Go over the expression hash table and delete insns that were | |
| 1322 marked for later deletion. */ | |
| 1323 | |
| 1324 /* This helper is called via htab_traverse. */ | |
| 111 | 1325 int |
| 1326 delete_redundant_insns_1 (expr **slot, void *data ATTRIBUTE_UNUSED) | |
| 0 | 1327 { |
| 111 | 1328 struct expr *exprs = *slot; |
| 0 | 1329 struct occr *occr; |
| 1330 | |
| 111 | 1331 for (occr = exprs->avail_occr; occr != NULL; occr = occr->next) |
| 0 | 1332 { |
| 1333 if (occr->deleted_p && dbg_cnt (gcse2_delete)) | |
| 1334 { | |
| 1335 delete_insn (occr->insn); | |
| 1336 stats.insns_deleted++; | |
| 1337 | |
| 1338 if (dump_file) | |
| 1339 { | |
| 1340 fprintf (dump_file, "deleting insn:\n"); | |
| 1341 print_rtl_single (dump_file, occr->insn); | |
| 1342 fprintf (dump_file, "\n"); | |
| 1343 } | |
| 1344 } | |
| 1345 } | |
| 1346 | |
| 1347 return 1; | |
| 1348 } | |
| 1349 | |
| 1350 static void | |
| 1351 delete_redundant_insns (void) | |
| 1352 { | |
| 111 | 1353 expr_table->traverse <void *, delete_redundant_insns_1> (NULL); |
| 0 | 1354 if (dump_file) |
| 1355 fprintf (dump_file, "\n"); | |
| 1356 } | |
| 1357 | |
| 1358 /* Main entry point of the GCSE after reload - clean some redundant loads | |
| 1359 due to spilling. */ | |
| 1360 | |
| 1361 static void | |
| 1362 gcse_after_reload_main (rtx f ATTRIBUTE_UNUSED) | |
| 1363 { | |
| 1364 | |
| 1365 memset (&stats, 0, sizeof (stats)); | |
| 1366 | |
| 1367 /* Allocate memory for this pass. | |
| 1368 Also computes and initializes the insns' CUIDs. */ | |
| 1369 alloc_mem (); | |
| 1370 | |
| 1371 /* We need alias analysis. */ | |
| 1372 init_alias_analysis (); | |
| 1373 | |
| 1374 compute_hash_table (); | |
| 1375 | |
| 1376 if (dump_file) | |
| 1377 dump_hash_table (dump_file); | |
| 1378 | |
| 111 | 1379 if (expr_table->elements () > 0) |
| 0 | 1380 { |
| 111 | 1381 /* Knowing which MEMs are transparent through a block can signifiantly |
| 1382 increase the number of redundant loads found. So compute transparency | |
| 1383 information for each memory expression in the hash table. */ | |
| 1384 df_analyze (); | |
| 1385 /* This can not be part of the normal allocation routine because | |
| 1386 we have to know the number of elements in the hash table. */ | |
| 1387 transp = sbitmap_vector_alloc (last_basic_block_for_fn (cfun), | |
| 1388 expr_table->elements ()); | |
| 1389 bitmap_vector_ones (transp, last_basic_block_for_fn (cfun)); | |
| 1390 expr_table->traverse <FILE *, compute_expr_transp> (dump_file); | |
| 0 | 1391 eliminate_partially_redundant_loads (); |
| 1392 delete_redundant_insns (); | |
| 111 | 1393 sbitmap_vector_free (transp); |
| 0 | 1394 |
| 1395 if (dump_file) | |
| 1396 { | |
| 1397 fprintf (dump_file, "GCSE AFTER RELOAD stats:\n"); | |
| 1398 fprintf (dump_file, "copies inserted: %d\n", stats.copies_inserted); | |
| 1399 fprintf (dump_file, "moves inserted: %d\n", stats.moves_inserted); | |
| 1400 fprintf (dump_file, "insns deleted: %d\n", stats.insns_deleted); | |
| 1401 fprintf (dump_file, "\n\n"); | |
| 1402 } | |
| 111 | 1403 |
| 1404 statistics_counter_event (cfun, "copies inserted", | |
| 1405 stats.copies_inserted); | |
| 1406 statistics_counter_event (cfun, "moves inserted", | |
| 1407 stats.moves_inserted); | |
| 1408 statistics_counter_event (cfun, "insns deleted", | |
| 1409 stats.insns_deleted); | |
| 0 | 1410 } |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
1411 |
| 0 | 1412 /* We are finished with alias. */ |
| 1413 end_alias_analysis (); | |
| 1414 | |
| 1415 free_mem (); | |
| 1416 } | |
| 1417 | |
| 1418 | |
| 1419 | |
| 1420 static unsigned int | |
| 1421 rest_of_handle_gcse2 (void) | |
| 1422 { | |
| 1423 gcse_after_reload_main (get_insns ()); | |
| 1424 rebuild_jump_labels (get_insns ()); | |
| 1425 return 0; | |
| 1426 } | |
| 1427 | |
| 111 | 1428 namespace { |
| 1429 | |
| 1430 const pass_data pass_data_gcse2 = | |
| 0 | 1431 { |
| 111 | 1432 RTL_PASS, /* type */ |
| 1433 "gcse2", /* name */ | |
| 1434 OPTGROUP_NONE, /* optinfo_flags */ | |
| 1435 TV_GCSE_AFTER_RELOAD, /* tv_id */ | |
| 1436 0, /* properties_required */ | |
| 1437 0, /* properties_provided */ | |
| 1438 0, /* properties_destroyed */ | |
| 1439 0, /* todo_flags_start */ | |
| 1440 0, /* todo_flags_finish */ | |
| 0 | 1441 }; |
| 1442 | |
| 111 | 1443 class pass_gcse2 : public rtl_opt_pass |
| 1444 { | |
| 1445 public: | |
| 1446 pass_gcse2 (gcc::context *ctxt) | |
| 1447 : rtl_opt_pass (pass_data_gcse2, ctxt) | |
| 1448 {} | |
| 1449 | |
| 1450 /* opt_pass methods: */ | |
| 1451 virtual bool gate (function *fun) | |
| 1452 { | |
| 1453 return (optimize > 0 && flag_gcse_after_reload | |
| 1454 && optimize_function_for_speed_p (fun)); | |
| 1455 } | |
| 1456 | |
| 1457 virtual unsigned int execute (function *) { return rest_of_handle_gcse2 (); } | |
| 1458 | |
| 1459 }; // class pass_gcse2 | |
| 1460 | |
| 1461 } // anon namespace | |
| 1462 | |
| 1463 rtl_opt_pass * | |
| 1464 make_pass_gcse2 (gcc::context *ctxt) | |
| 1465 { | |
| 1466 return new pass_gcse2 (ctxt); | |
| 1467 } |