Mercurial > hg > CbC > CbC_gcc
annotate gcc/cfgexpand.c @ 5:a4c410aa4714
modifing gimple.[ch], cfgexpand.c
to make gimple to support cbc goto flags.
| author | kent <kent@cr.ie.u-ryukyu.ac.jp> |
|---|---|
| date | Thu, 30 Jul 2009 19:22:01 +0900 |
| parents | a06113de4d67 |
| children | 9de9dad105d4 |
| rev | line source |
|---|---|
| 0 | 1 /* A pass for lowering trees to RTL. |
| 2 Copyright (C) 2004, 2005, 2006, 2007, 2008, 2009 | |
| 3 Free Software Foundation, Inc. | |
| 4 | |
| 5 This file is part of GCC. | |
| 6 | |
| 7 GCC is free software; you can redistribute it and/or modify | |
| 8 it under the terms of the GNU General Public License as published by | |
| 9 the Free Software Foundation; either version 3, or (at your option) | |
| 10 any later version. | |
| 11 | |
| 12 GCC is distributed in the hope that it will be useful, | |
| 13 but WITHOUT ANY WARRANTY; without even the implied warranty of | |
| 14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
| 15 GNU General Public License for more details. | |
| 16 | |
| 17 You should have received a copy of the GNU General Public License | |
| 18 along with GCC; see the file COPYING3. If not see | |
| 19 <http://www.gnu.org/licenses/>. */ | |
| 20 | |
| 21 #include "config.h" | |
| 22 #include "system.h" | |
| 23 #include "coretypes.h" | |
| 24 #include "tm.h" | |
| 25 #include "tree.h" | |
| 26 #include "rtl.h" | |
| 27 #include "tm_p.h" | |
| 28 #include "basic-block.h" | |
| 29 #include "function.h" | |
| 30 #include "expr.h" | |
| 31 #include "langhooks.h" | |
| 32 #include "tree-flow.h" | |
| 33 #include "timevar.h" | |
| 34 #include "tree-dump.h" | |
| 35 #include "tree-pass.h" | |
| 36 #include "except.h" | |
| 37 #include "flags.h" | |
| 38 #include "diagnostic.h" | |
| 39 #include "toplev.h" | |
| 40 #include "debug.h" | |
| 41 #include "params.h" | |
| 42 #include "tree-inline.h" | |
| 43 #include "value-prof.h" | |
| 44 #include "target.h" | |
|
5
a4c410aa4714
modifing gimple.[ch], cfgexpand.c
kent <kent@cr.ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
45 #ifndef noCbC |
|
a4c410aa4714
modifing gimple.[ch], cfgexpand.c
kent <kent@cr.ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
46 #include "cbc-tree.h" |
|
a4c410aa4714
modifing gimple.[ch], cfgexpand.c
kent <kent@cr.ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
47 #endif |
| 0 | 48 |
| 49 | |
| 50 /* Return an expression tree corresponding to the RHS of GIMPLE | |
| 51 statement STMT. */ | |
| 52 | |
| 53 tree | |
| 54 gimple_assign_rhs_to_tree (gimple stmt) | |
| 55 { | |
| 56 tree t; | |
| 57 enum gimple_rhs_class grhs_class; | |
| 58 | |
| 59 grhs_class = get_gimple_rhs_class (gimple_expr_code (stmt)); | |
| 60 | |
| 61 if (grhs_class == GIMPLE_BINARY_RHS) | |
| 62 t = build2 (gimple_assign_rhs_code (stmt), | |
| 63 TREE_TYPE (gimple_assign_lhs (stmt)), | |
| 64 gimple_assign_rhs1 (stmt), | |
| 65 gimple_assign_rhs2 (stmt)); | |
| 66 else if (grhs_class == GIMPLE_UNARY_RHS) | |
| 67 t = build1 (gimple_assign_rhs_code (stmt), | |
| 68 TREE_TYPE (gimple_assign_lhs (stmt)), | |
| 69 gimple_assign_rhs1 (stmt)); | |
| 70 else if (grhs_class == GIMPLE_SINGLE_RHS) | |
| 71 t = gimple_assign_rhs1 (stmt); | |
| 72 else | |
| 73 gcc_unreachable (); | |
| 74 | |
| 75 return t; | |
| 76 } | |
| 77 | |
| 78 /* Return an expression tree corresponding to the PREDICATE of GIMPLE_COND | |
| 79 statement STMT. */ | |
| 80 | |
| 81 static tree | |
| 82 gimple_cond_pred_to_tree (gimple stmt) | |
| 83 { | |
| 84 return build2 (gimple_cond_code (stmt), boolean_type_node, | |
| 85 gimple_cond_lhs (stmt), gimple_cond_rhs (stmt)); | |
| 86 } | |
| 87 | |
| 88 /* Helper for gimple_to_tree. Set EXPR_LOCATION for every expression | |
| 89 inside *TP. DATA is the location to set. */ | |
| 90 | |
| 91 static tree | |
| 92 set_expr_location_r (tree *tp, int *ws ATTRIBUTE_UNUSED, void *data) | |
| 93 { | |
| 94 location_t *loc = (location_t *) data; | |
| 95 if (EXPR_P (*tp)) | |
| 96 SET_EXPR_LOCATION (*tp, *loc); | |
| 97 | |
| 98 return NULL_TREE; | |
| 99 } | |
| 100 | |
| 101 | |
| 102 /* RTL expansion has traditionally been done on trees, so the | |
| 103 transition to doing it on GIMPLE tuples is very invasive to the RTL | |
| 104 expander. To facilitate the transition, this function takes a | |
| 105 GIMPLE tuple STMT and returns the same statement in the form of a | |
| 106 tree. */ | |
| 107 | |
| 108 static tree | |
| 109 gimple_to_tree (gimple stmt) | |
| 110 { | |
| 111 tree t; | |
| 112 int rn; | |
| 113 tree_ann_common_t ann; | |
| 114 location_t loc; | |
| 115 | |
| 116 switch (gimple_code (stmt)) | |
| 117 { | |
| 118 case GIMPLE_ASSIGN: | |
| 119 { | |
| 120 tree lhs = gimple_assign_lhs (stmt); | |
| 121 | |
| 122 t = gimple_assign_rhs_to_tree (stmt); | |
| 123 t = build2 (MODIFY_EXPR, TREE_TYPE (lhs), lhs, t); | |
| 124 if (gimple_assign_nontemporal_move_p (stmt)) | |
| 125 MOVE_NONTEMPORAL (t) = true; | |
| 126 } | |
| 127 break; | |
| 128 | |
| 129 case GIMPLE_COND: | |
| 130 t = gimple_cond_pred_to_tree (stmt); | |
| 131 t = build3 (COND_EXPR, void_type_node, t, NULL_TREE, NULL_TREE); | |
| 132 break; | |
| 133 | |
| 134 case GIMPLE_GOTO: | |
| 135 t = build1 (GOTO_EXPR, void_type_node, gimple_goto_dest (stmt)); | |
| 136 break; | |
| 137 | |
| 138 case GIMPLE_LABEL: | |
| 139 t = build1 (LABEL_EXPR, void_type_node, gimple_label_label (stmt)); | |
| 140 break; | |
| 141 | |
| 142 case GIMPLE_RETURN: | |
| 143 { | |
| 144 tree retval = gimple_return_retval (stmt); | |
| 145 | |
| 146 if (retval && retval != error_mark_node) | |
| 147 { | |
| 148 tree result = DECL_RESULT (current_function_decl); | |
| 149 | |
| 150 /* If we are not returning the current function's RESULT_DECL, | |
| 151 build an assignment to it. */ | |
| 152 if (retval != result) | |
| 153 { | |
| 154 /* I believe that a function's RESULT_DECL is unique. */ | |
| 155 gcc_assert (TREE_CODE (retval) != RESULT_DECL); | |
| 156 | |
| 157 retval = build2 (MODIFY_EXPR, TREE_TYPE (result), | |
| 158 result, retval); | |
| 159 } | |
| 160 } | |
| 161 t = build1 (RETURN_EXPR, void_type_node, retval); | |
| 162 } | |
| 163 break; | |
| 164 | |
| 165 case GIMPLE_ASM: | |
| 166 { | |
| 167 size_t i, n; | |
| 168 tree out, in, cl; | |
| 169 const char *s; | |
| 170 | |
| 171 out = NULL_TREE; | |
| 172 n = gimple_asm_noutputs (stmt); | |
| 173 if (n > 0) | |
| 174 { | |
| 175 t = out = gimple_asm_output_op (stmt, 0); | |
| 176 for (i = 1; i < n; i++) | |
| 177 { | |
| 178 TREE_CHAIN (t) = gimple_asm_output_op (stmt, i); | |
| 179 t = gimple_asm_output_op (stmt, i); | |
| 180 } | |
| 181 } | |
| 182 | |
| 183 in = NULL_TREE; | |
| 184 n = gimple_asm_ninputs (stmt); | |
| 185 if (n > 0) | |
| 186 { | |
| 187 t = in = gimple_asm_input_op (stmt, 0); | |
| 188 for (i = 1; i < n; i++) | |
| 189 { | |
| 190 TREE_CHAIN (t) = gimple_asm_input_op (stmt, i); | |
| 191 t = gimple_asm_input_op (stmt, i); | |
| 192 } | |
| 193 } | |
| 194 | |
| 195 cl = NULL_TREE; | |
| 196 n = gimple_asm_nclobbers (stmt); | |
| 197 if (n > 0) | |
| 198 { | |
| 199 t = cl = gimple_asm_clobber_op (stmt, 0); | |
| 200 for (i = 1; i < n; i++) | |
| 201 { | |
| 202 TREE_CHAIN (t) = gimple_asm_clobber_op (stmt, i); | |
| 203 t = gimple_asm_clobber_op (stmt, i); | |
| 204 } | |
| 205 } | |
| 206 | |
| 207 s = gimple_asm_string (stmt); | |
| 208 t = build4 (ASM_EXPR, void_type_node, build_string (strlen (s), s), | |
| 209 out, in, cl); | |
| 210 ASM_VOLATILE_P (t) = gimple_asm_volatile_p (stmt); | |
| 211 ASM_INPUT_P (t) = gimple_asm_input_p (stmt); | |
| 212 } | |
| 213 break; | |
| 214 | |
| 215 case GIMPLE_CALL: | |
| 216 { | |
| 217 size_t i; | |
| 218 tree fn; | |
| 219 tree_ann_common_t ann; | |
| 220 | |
| 221 t = build_vl_exp (CALL_EXPR, gimple_call_num_args (stmt) + 3); | |
| 222 | |
| 223 CALL_EXPR_FN (t) = gimple_call_fn (stmt); | |
| 224 TREE_TYPE (t) = gimple_call_return_type (stmt); | |
| 225 CALL_EXPR_STATIC_CHAIN (t) = gimple_call_chain (stmt); | |
| 226 | |
| 227 for (i = 0; i < gimple_call_num_args (stmt); i++) | |
| 228 CALL_EXPR_ARG (t, i) = gimple_call_arg (stmt, i); | |
| 229 | |
| 230 if (!(gimple_call_flags (stmt) & (ECF_CONST | ECF_PURE))) | |
| 231 TREE_SIDE_EFFECTS (t) = 1; | |
| 232 | |
| 233 if (gimple_call_flags (stmt) & ECF_NOTHROW) | |
| 234 TREE_NOTHROW (t) = 1; | |
| 235 | |
| 236 CALL_EXPR_TAILCALL (t) = gimple_call_tail_p (stmt); | |
| 237 CALL_EXPR_RETURN_SLOT_OPT (t) = gimple_call_return_slot_opt_p (stmt); | |
| 238 CALL_FROM_THUNK_P (t) = gimple_call_from_thunk_p (stmt); | |
| 239 CALL_CANNOT_INLINE_P (t) = gimple_call_cannot_inline_p (stmt); | |
| 240 CALL_EXPR_VA_ARG_PACK (t) = gimple_call_va_arg_pack_p (stmt); | |
|
5
a4c410aa4714
modifing gimple.[ch], cfgexpand.c
kent <kent@cr.ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
241 #ifndef noCbC |
|
a4c410aa4714
modifing gimple.[ch], cfgexpand.c
kent <kent@cr.ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
242 CALL_EXPR_CBC_GOTO (t) = gimple_call_cbc_goto_p (stmt); |
|
a4c410aa4714
modifing gimple.[ch], cfgexpand.c
kent <kent@cr.ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
243 #endif |
| 0 | 244 |
| 245 /* If the call has a LHS then create a MODIFY_EXPR to hold it. */ | |
| 246 { | |
| 247 tree lhs = gimple_call_lhs (stmt); | |
| 248 | |
| 249 if (lhs) | |
| 250 t = build2 (MODIFY_EXPR, TREE_TYPE (lhs), lhs, t); | |
| 251 } | |
| 252 | |
| 253 /* Record the original call statement, as it may be used | |
| 254 to retrieve profile information during expansion. */ | |
| 255 | |
| 256 if ((fn = gimple_call_fndecl (stmt)) != NULL_TREE | |
| 257 && DECL_BUILT_IN (fn)) | |
| 258 { | |
| 259 ann = get_tree_common_ann (t); | |
| 260 ann->stmt = stmt; | |
| 261 } | |
| 262 } | |
| 263 break; | |
| 264 | |
| 265 case GIMPLE_SWITCH: | |
| 266 { | |
| 267 tree label_vec; | |
| 268 size_t i; | |
| 269 tree elt = gimple_switch_label (stmt, 0); | |
| 270 | |
| 271 label_vec = make_tree_vec (gimple_switch_num_labels (stmt)); | |
| 272 | |
| 273 if (!CASE_LOW (elt) && !CASE_HIGH (elt)) | |
| 274 { | |
| 275 for (i = 1; i < gimple_switch_num_labels (stmt); i++) | |
| 276 TREE_VEC_ELT (label_vec, i - 1) = gimple_switch_label (stmt, i); | |
| 277 | |
| 278 /* The default case in a SWITCH_EXPR must be at the end of | |
| 279 the label vector. */ | |
| 280 TREE_VEC_ELT (label_vec, i - 1) = gimple_switch_label (stmt, 0); | |
| 281 } | |
| 282 else | |
| 283 { | |
| 284 for (i = 0; i < gimple_switch_num_labels (stmt); i++) | |
| 285 TREE_VEC_ELT (label_vec, i) = gimple_switch_label (stmt, i); | |
| 286 } | |
| 287 | |
| 288 t = build3 (SWITCH_EXPR, void_type_node, gimple_switch_index (stmt), | |
| 289 NULL, label_vec); | |
| 290 } | |
| 291 break; | |
| 292 | |
| 293 case GIMPLE_NOP: | |
| 294 case GIMPLE_PREDICT: | |
| 295 t = build1 (NOP_EXPR, void_type_node, size_zero_node); | |
| 296 break; | |
| 297 | |
| 298 case GIMPLE_RESX: | |
| 299 t = build_resx (gimple_resx_region (stmt)); | |
| 300 break; | |
| 301 | |
| 302 default: | |
| 303 if (errorcount == 0) | |
| 304 { | |
| 305 error ("Unrecognized GIMPLE statement during RTL expansion"); | |
| 306 print_gimple_stmt (stderr, stmt, 4, 0); | |
| 307 gcc_unreachable (); | |
| 308 } | |
| 309 else | |
| 310 { | |
| 311 /* Ignore any bad gimple codes if we're going to die anyhow, | |
| 312 so we can at least set TREE_ASM_WRITTEN and have the rest | |
| 313 of compilation advance without sudden ICE death. */ | |
| 314 t = build1 (NOP_EXPR, void_type_node, size_zero_node); | |
| 315 break; | |
| 316 } | |
| 317 } | |
| 318 | |
| 319 /* If STMT is inside an exception region, record it in the generated | |
| 320 expression. */ | |
| 321 rn = lookup_stmt_eh_region (stmt); | |
| 322 if (rn >= 0) | |
| 323 { | |
| 324 tree call = get_call_expr_in (t); | |
| 325 | |
| 326 ann = get_tree_common_ann (t); | |
| 327 ann->rn = rn; | |
| 328 | |
| 329 /* For a CALL_EXPR on the RHS of an assignment, calls.c looks up | |
| 330 the CALL_EXPR not the assignment statment for EH region number. */ | |
| 331 if (call && call != t) | |
| 332 { | |
| 333 ann = get_tree_common_ann (call); | |
| 334 ann->rn = rn; | |
| 335 } | |
| 336 } | |
| 337 | |
| 338 /* Set EXPR_LOCATION in all the embedded expressions. */ | |
| 339 loc = gimple_location (stmt); | |
| 340 walk_tree (&t, set_expr_location_r, (void *) &loc, NULL); | |
| 341 | |
| 342 TREE_BLOCK (t) = gimple_block (stmt); | |
| 343 | |
| 344 return t; | |
| 345 } | |
| 346 | |
| 347 | |
| 348 /* Release back to GC memory allocated by gimple_to_tree. */ | |
| 349 | |
| 350 static void | |
| 351 release_stmt_tree (gimple stmt, tree stmt_tree) | |
| 352 { | |
| 353 tree_ann_common_t ann; | |
| 354 | |
| 355 switch (gimple_code (stmt)) | |
| 356 { | |
| 357 case GIMPLE_ASSIGN: | |
| 358 if (get_gimple_rhs_class (gimple_expr_code (stmt)) != GIMPLE_SINGLE_RHS) | |
| 359 ggc_free (TREE_OPERAND (stmt_tree, 1)); | |
| 360 break; | |
| 361 case GIMPLE_COND: | |
| 362 ggc_free (COND_EXPR_COND (stmt_tree)); | |
| 363 break; | |
| 364 case GIMPLE_RETURN: | |
| 365 if (TREE_OPERAND (stmt_tree, 0) | |
| 366 && TREE_CODE (TREE_OPERAND (stmt_tree, 0)) == MODIFY_EXPR) | |
| 367 ggc_free (TREE_OPERAND (stmt_tree, 0)); | |
| 368 break; | |
| 369 case GIMPLE_CALL: | |
| 370 if (gimple_call_lhs (stmt)) | |
| 371 { | |
| 372 ann = tree_common_ann (TREE_OPERAND (stmt_tree, 1)); | |
| 373 if (ann) | |
| 374 ggc_free (ann); | |
| 375 ggc_free (TREE_OPERAND (stmt_tree, 1)); | |
| 376 } | |
| 377 break; | |
| 378 default: | |
| 379 break; | |
| 380 } | |
| 381 ann = tree_common_ann (stmt_tree); | |
| 382 if (ann) | |
| 383 ggc_free (ann); | |
| 384 ggc_free (stmt_tree); | |
| 385 } | |
| 386 | |
| 387 | |
| 388 /* Verify that there is exactly single jump instruction since last and attach | |
| 389 REG_BR_PROB note specifying probability. | |
| 390 ??? We really ought to pass the probability down to RTL expanders and let it | |
| 391 re-distribute it when the conditional expands into multiple conditionals. | |
| 392 This is however difficult to do. */ | |
| 393 void | |
| 394 add_reg_br_prob_note (rtx last, int probability) | |
| 395 { | |
| 396 if (profile_status == PROFILE_ABSENT) | |
| 397 return; | |
| 398 for (last = NEXT_INSN (last); last && NEXT_INSN (last); last = NEXT_INSN (last)) | |
| 399 if (JUMP_P (last)) | |
| 400 { | |
| 401 /* It is common to emit condjump-around-jump sequence when we don't know | |
| 402 how to reverse the conditional. Special case this. */ | |
| 403 if (!any_condjump_p (last) | |
| 404 || !JUMP_P (NEXT_INSN (last)) | |
| 405 || !simplejump_p (NEXT_INSN (last)) | |
| 406 || !NEXT_INSN (NEXT_INSN (last)) | |
| 407 || !BARRIER_P (NEXT_INSN (NEXT_INSN (last))) | |
| 408 || !NEXT_INSN (NEXT_INSN (NEXT_INSN (last))) | |
| 409 || !LABEL_P (NEXT_INSN (NEXT_INSN (NEXT_INSN (last)))) | |
| 410 || NEXT_INSN (NEXT_INSN (NEXT_INSN (NEXT_INSN (last))))) | |
| 411 goto failed; | |
| 412 gcc_assert (!find_reg_note (last, REG_BR_PROB, 0)); | |
| 413 add_reg_note (last, REG_BR_PROB, | |
| 414 GEN_INT (REG_BR_PROB_BASE - probability)); | |
| 415 return; | |
| 416 } | |
| 417 if (!last || !JUMP_P (last) || !any_condjump_p (last)) | |
| 418 goto failed; | |
| 419 gcc_assert (!find_reg_note (last, REG_BR_PROB, 0)); | |
| 420 add_reg_note (last, REG_BR_PROB, GEN_INT (probability)); | |
| 421 return; | |
| 422 failed: | |
| 423 if (dump_file) | |
| 424 fprintf (dump_file, "Failed to add probability note\n"); | |
| 425 } | |
| 426 | |
| 427 | |
| 428 #ifndef STACK_ALIGNMENT_NEEDED | |
| 429 #define STACK_ALIGNMENT_NEEDED 1 | |
| 430 #endif | |
| 431 | |
| 432 | |
| 433 /* This structure holds data relevant to one variable that will be | |
| 434 placed in a stack slot. */ | |
| 435 struct stack_var | |
| 436 { | |
| 437 /* The Variable. */ | |
| 438 tree decl; | |
| 439 | |
| 440 /* The offset of the variable. During partitioning, this is the | |
| 441 offset relative to the partition. After partitioning, this | |
| 442 is relative to the stack frame. */ | |
| 443 HOST_WIDE_INT offset; | |
| 444 | |
| 445 /* Initially, the size of the variable. Later, the size of the partition, | |
| 446 if this variable becomes it's partition's representative. */ | |
| 447 HOST_WIDE_INT size; | |
| 448 | |
| 449 /* The *byte* alignment required for this variable. Or as, with the | |
| 450 size, the alignment for this partition. */ | |
| 451 unsigned int alignb; | |
| 452 | |
| 453 /* The partition representative. */ | |
| 454 size_t representative; | |
| 455 | |
| 456 /* The next stack variable in the partition, or EOC. */ | |
| 457 size_t next; | |
| 458 }; | |
| 459 | |
| 460 #define EOC ((size_t)-1) | |
| 461 | |
| 462 /* We have an array of such objects while deciding allocation. */ | |
| 463 static struct stack_var *stack_vars; | |
| 464 static size_t stack_vars_alloc; | |
| 465 static size_t stack_vars_num; | |
| 466 | |
| 467 /* An array of indices such that stack_vars[stack_vars_sorted[i]].size | |
| 468 is non-decreasing. */ | |
| 469 static size_t *stack_vars_sorted; | |
| 470 | |
| 471 /* We have an interference graph between such objects. This graph | |
| 472 is lower triangular. */ | |
| 473 static bool *stack_vars_conflict; | |
| 474 static size_t stack_vars_conflict_alloc; | |
| 475 | |
| 476 /* The phase of the stack frame. This is the known misalignment of | |
| 477 virtual_stack_vars_rtx from PREFERRED_STACK_BOUNDARY. That is, | |
| 478 (frame_offset+frame_phase) % PREFERRED_STACK_BOUNDARY == 0. */ | |
| 479 static int frame_phase; | |
| 480 | |
| 481 /* Used during expand_used_vars to remember if we saw any decls for | |
| 482 which we'd like to enable stack smashing protection. */ | |
| 483 static bool has_protected_decls; | |
| 484 | |
| 485 /* Used during expand_used_vars. Remember if we say a character buffer | |
| 486 smaller than our cutoff threshold. Used for -Wstack-protector. */ | |
| 487 static bool has_short_buffer; | |
| 488 | |
| 489 /* Discover the byte alignment to use for DECL. Ignore alignment | |
| 490 we can't do with expected alignment of the stack boundary. */ | |
| 491 | |
| 492 static unsigned int | |
| 493 get_decl_align_unit (tree decl) | |
| 494 { | |
| 495 unsigned int align; | |
| 496 | |
| 497 align = LOCAL_DECL_ALIGNMENT (decl); | |
| 498 | |
| 499 if (align > MAX_SUPPORTED_STACK_ALIGNMENT) | |
| 500 align = MAX_SUPPORTED_STACK_ALIGNMENT; | |
| 501 | |
| 502 if (SUPPORTS_STACK_ALIGNMENT) | |
| 503 { | |
| 504 if (crtl->stack_alignment_estimated < align) | |
| 505 { | |
| 506 gcc_assert(!crtl->stack_realign_processed); | |
| 507 crtl->stack_alignment_estimated = align; | |
| 508 } | |
| 509 } | |
| 510 | |
| 511 /* stack_alignment_needed > PREFERRED_STACK_BOUNDARY is permitted. | |
| 512 So here we only make sure stack_alignment_needed >= align. */ | |
| 513 if (crtl->stack_alignment_needed < align) | |
| 514 crtl->stack_alignment_needed = align; | |
| 515 if (crtl->max_used_stack_slot_alignment < crtl->stack_alignment_needed) | |
| 516 crtl->max_used_stack_slot_alignment = crtl->stack_alignment_needed; | |
| 517 | |
| 518 return align / BITS_PER_UNIT; | |
| 519 } | |
| 520 | |
| 521 /* Allocate SIZE bytes at byte alignment ALIGN from the stack frame. | |
| 522 Return the frame offset. */ | |
| 523 | |
| 524 static HOST_WIDE_INT | |
| 525 alloc_stack_frame_space (HOST_WIDE_INT size, HOST_WIDE_INT align) | |
| 526 { | |
| 527 HOST_WIDE_INT offset, new_frame_offset; | |
| 528 | |
| 529 new_frame_offset = frame_offset; | |
| 530 if (FRAME_GROWS_DOWNWARD) | |
| 531 { | |
| 532 new_frame_offset -= size + frame_phase; | |
| 533 new_frame_offset &= -align; | |
| 534 new_frame_offset += frame_phase; | |
| 535 offset = new_frame_offset; | |
| 536 } | |
| 537 else | |
| 538 { | |
| 539 new_frame_offset -= frame_phase; | |
| 540 new_frame_offset += align - 1; | |
| 541 new_frame_offset &= -align; | |
| 542 new_frame_offset += frame_phase; | |
| 543 offset = new_frame_offset; | |
| 544 new_frame_offset += size; | |
| 545 } | |
| 546 frame_offset = new_frame_offset; | |
| 547 | |
| 548 if (frame_offset_overflow (frame_offset, cfun->decl)) | |
| 549 frame_offset = offset = 0; | |
| 550 | |
| 551 return offset; | |
| 552 } | |
| 553 | |
| 554 /* Accumulate DECL into STACK_VARS. */ | |
| 555 | |
| 556 static void | |
| 557 add_stack_var (tree decl) | |
| 558 { | |
| 559 if (stack_vars_num >= stack_vars_alloc) | |
| 560 { | |
| 561 if (stack_vars_alloc) | |
| 562 stack_vars_alloc = stack_vars_alloc * 3 / 2; | |
| 563 else | |
| 564 stack_vars_alloc = 32; | |
| 565 stack_vars | |
| 566 = XRESIZEVEC (struct stack_var, stack_vars, stack_vars_alloc); | |
| 567 } | |
| 568 stack_vars[stack_vars_num].decl = decl; | |
| 569 stack_vars[stack_vars_num].offset = 0; | |
| 570 stack_vars[stack_vars_num].size = tree_low_cst (DECL_SIZE_UNIT (decl), 1); | |
| 571 stack_vars[stack_vars_num].alignb = get_decl_align_unit (decl); | |
| 572 | |
| 573 /* All variables are initially in their own partition. */ | |
| 574 stack_vars[stack_vars_num].representative = stack_vars_num; | |
| 575 stack_vars[stack_vars_num].next = EOC; | |
| 576 | |
| 577 /* Ensure that this decl doesn't get put onto the list twice. */ | |
| 578 SET_DECL_RTL (decl, pc_rtx); | |
| 579 | |
| 580 stack_vars_num++; | |
| 581 } | |
| 582 | |
| 583 /* Compute the linear index of a lower-triangular coordinate (I, J). */ | |
| 584 | |
| 585 static size_t | |
| 586 triangular_index (size_t i, size_t j) | |
| 587 { | |
| 588 if (i < j) | |
| 589 { | |
| 590 size_t t; | |
| 591 t = i, i = j, j = t; | |
| 592 } | |
| 593 return (i * (i + 1)) / 2 + j; | |
| 594 } | |
| 595 | |
| 596 /* Ensure that STACK_VARS_CONFLICT is large enough for N objects. */ | |
| 597 | |
| 598 static void | |
| 599 resize_stack_vars_conflict (size_t n) | |
| 600 { | |
| 601 size_t size = triangular_index (n-1, n-1) + 1; | |
| 602 | |
| 603 if (size <= stack_vars_conflict_alloc) | |
| 604 return; | |
| 605 | |
| 606 stack_vars_conflict = XRESIZEVEC (bool, stack_vars_conflict, size); | |
| 607 memset (stack_vars_conflict + stack_vars_conflict_alloc, 0, | |
| 608 (size - stack_vars_conflict_alloc) * sizeof (bool)); | |
| 609 stack_vars_conflict_alloc = size; | |
| 610 } | |
| 611 | |
| 612 /* Make the decls associated with luid's X and Y conflict. */ | |
| 613 | |
| 614 static void | |
| 615 add_stack_var_conflict (size_t x, size_t y) | |
| 616 { | |
| 617 size_t index = triangular_index (x, y); | |
| 618 gcc_assert (index < stack_vars_conflict_alloc); | |
| 619 stack_vars_conflict[index] = true; | |
| 620 } | |
| 621 | |
| 622 /* Check whether the decls associated with luid's X and Y conflict. */ | |
| 623 | |
| 624 static bool | |
| 625 stack_var_conflict_p (size_t x, size_t y) | |
| 626 { | |
| 627 size_t index = triangular_index (x, y); | |
| 628 gcc_assert (index < stack_vars_conflict_alloc); | |
| 629 return stack_vars_conflict[index]; | |
| 630 } | |
| 631 | |
| 632 /* Returns true if TYPE is or contains a union type. */ | |
| 633 | |
| 634 static bool | |
| 635 aggregate_contains_union_type (tree type) | |
| 636 { | |
| 637 tree field; | |
| 638 | |
| 639 if (TREE_CODE (type) == UNION_TYPE | |
| 640 || TREE_CODE (type) == QUAL_UNION_TYPE) | |
| 641 return true; | |
| 642 if (TREE_CODE (type) == ARRAY_TYPE) | |
| 643 return aggregate_contains_union_type (TREE_TYPE (type)); | |
| 644 if (TREE_CODE (type) != RECORD_TYPE) | |
| 645 return false; | |
| 646 | |
| 647 for (field = TYPE_FIELDS (type); field; field = TREE_CHAIN (field)) | |
| 648 if (TREE_CODE (field) == FIELD_DECL) | |
| 649 if (aggregate_contains_union_type (TREE_TYPE (field))) | |
| 650 return true; | |
| 651 | |
| 652 return false; | |
| 653 } | |
| 654 | |
| 655 /* A subroutine of expand_used_vars. If two variables X and Y have alias | |
| 656 sets that do not conflict, then do add a conflict for these variables | |
| 657 in the interference graph. We also need to make sure to add conflicts | |
| 658 for union containing structures. Else RTL alias analysis comes along | |
| 659 and due to type based aliasing rules decides that for two overlapping | |
| 660 union temporaries { short s; int i; } accesses to the same mem through | |
| 661 different types may not alias and happily reorders stores across | |
| 662 life-time boundaries of the temporaries (See PR25654). | |
| 663 We also have to mind MEM_IN_STRUCT_P and MEM_SCALAR_P. */ | |
| 664 | |
| 665 static void | |
| 666 add_alias_set_conflicts (void) | |
| 667 { | |
| 668 size_t i, j, n = stack_vars_num; | |
| 669 | |
| 670 for (i = 0; i < n; ++i) | |
| 671 { | |
| 672 tree type_i = TREE_TYPE (stack_vars[i].decl); | |
| 673 bool aggr_i = AGGREGATE_TYPE_P (type_i); | |
| 674 bool contains_union; | |
| 675 | |
| 676 contains_union = aggregate_contains_union_type (type_i); | |
| 677 for (j = 0; j < i; ++j) | |
| 678 { | |
| 679 tree type_j = TREE_TYPE (stack_vars[j].decl); | |
| 680 bool aggr_j = AGGREGATE_TYPE_P (type_j); | |
| 681 if (aggr_i != aggr_j | |
| 682 /* Either the objects conflict by means of type based | |
| 683 aliasing rules, or we need to add a conflict. */ | |
| 684 || !objects_must_conflict_p (type_i, type_j) | |
| 685 /* In case the types do not conflict ensure that access | |
| 686 to elements will conflict. In case of unions we have | |
| 687 to be careful as type based aliasing rules may say | |
| 688 access to the same memory does not conflict. So play | |
| 689 safe and add a conflict in this case. */ | |
| 690 || contains_union) | |
| 691 add_stack_var_conflict (i, j); | |
| 692 } | |
| 693 } | |
| 694 } | |
| 695 | |
| 696 /* A subroutine of partition_stack_vars. A comparison function for qsort, | |
| 697 sorting an array of indices by the size of the object. */ | |
| 698 | |
| 699 static int | |
| 700 stack_var_size_cmp (const void *a, const void *b) | |
| 701 { | |
| 702 HOST_WIDE_INT sa = stack_vars[*(const size_t *)a].size; | |
| 703 HOST_WIDE_INT sb = stack_vars[*(const size_t *)b].size; | |
| 704 unsigned int uida = DECL_UID (stack_vars[*(const size_t *)a].decl); | |
| 705 unsigned int uidb = DECL_UID (stack_vars[*(const size_t *)b].decl); | |
| 706 | |
| 707 if (sa < sb) | |
| 708 return -1; | |
| 709 if (sa > sb) | |
| 710 return 1; | |
| 711 /* For stack variables of the same size use the uid of the decl | |
| 712 to make the sort stable. */ | |
| 713 if (uida < uidb) | |
| 714 return -1; | |
| 715 if (uida > uidb) | |
| 716 return 1; | |
| 717 return 0; | |
| 718 } | |
| 719 | |
| 720 /* A subroutine of partition_stack_vars. The UNION portion of a UNION/FIND | |
| 721 partitioning algorithm. Partitions A and B are known to be non-conflicting. | |
| 722 Merge them into a single partition A. | |
| 723 | |
| 724 At the same time, add OFFSET to all variables in partition B. At the end | |
| 725 of the partitioning process we've have a nice block easy to lay out within | |
| 726 the stack frame. */ | |
| 727 | |
| 728 static void | |
| 729 union_stack_vars (size_t a, size_t b, HOST_WIDE_INT offset) | |
| 730 { | |
| 731 size_t i, last; | |
| 732 | |
| 733 /* Update each element of partition B with the given offset, | |
| 734 and merge them into partition A. */ | |
| 735 for (last = i = b; i != EOC; last = i, i = stack_vars[i].next) | |
| 736 { | |
| 737 stack_vars[i].offset += offset; | |
| 738 stack_vars[i].representative = a; | |
| 739 } | |
| 740 stack_vars[last].next = stack_vars[a].next; | |
| 741 stack_vars[a].next = b; | |
| 742 | |
| 743 /* Update the required alignment of partition A to account for B. */ | |
| 744 if (stack_vars[a].alignb < stack_vars[b].alignb) | |
| 745 stack_vars[a].alignb = stack_vars[b].alignb; | |
| 746 | |
| 747 /* Update the interference graph and merge the conflicts. */ | |
| 748 for (last = stack_vars_num, i = 0; i < last; ++i) | |
| 749 if (stack_var_conflict_p (b, i)) | |
| 750 add_stack_var_conflict (a, i); | |
| 751 } | |
| 752 | |
| 753 /* A subroutine of expand_used_vars. Binpack the variables into | |
| 754 partitions constrained by the interference graph. The overall | |
| 755 algorithm used is as follows: | |
| 756 | |
| 757 Sort the objects by size. | |
| 758 For each object A { | |
| 759 S = size(A) | |
| 760 O = 0 | |
| 761 loop { | |
| 762 Look for the largest non-conflicting object B with size <= S. | |
| 763 UNION (A, B) | |
| 764 offset(B) = O | |
| 765 O += size(B) | |
| 766 S -= size(B) | |
| 767 } | |
| 768 } | |
| 769 */ | |
| 770 | |
| 771 static void | |
| 772 partition_stack_vars (void) | |
| 773 { | |
| 774 size_t si, sj, n = stack_vars_num; | |
| 775 | |
| 776 stack_vars_sorted = XNEWVEC (size_t, stack_vars_num); | |
| 777 for (si = 0; si < n; ++si) | |
| 778 stack_vars_sorted[si] = si; | |
| 779 | |
| 780 if (n == 1) | |
| 781 return; | |
| 782 | |
| 783 qsort (stack_vars_sorted, n, sizeof (size_t), stack_var_size_cmp); | |
| 784 | |
| 785 /* Special case: detect when all variables conflict, and thus we can't | |
| 786 do anything during the partitioning loop. It isn't uncommon (with | |
| 787 C code at least) to declare all variables at the top of the function, | |
| 788 and if we're not inlining, then all variables will be in the same scope. | |
| 789 Take advantage of very fast libc routines for this scan. */ | |
| 790 gcc_assert (sizeof(bool) == sizeof(char)); | |
| 791 if (memchr (stack_vars_conflict, false, stack_vars_conflict_alloc) == NULL) | |
| 792 return; | |
| 793 | |
| 794 for (si = 0; si < n; ++si) | |
| 795 { | |
| 796 size_t i = stack_vars_sorted[si]; | |
| 797 HOST_WIDE_INT isize = stack_vars[i].size; | |
| 798 HOST_WIDE_INT offset = 0; | |
| 799 | |
| 800 for (sj = si; sj-- > 0; ) | |
| 801 { | |
| 802 size_t j = stack_vars_sorted[sj]; | |
| 803 HOST_WIDE_INT jsize = stack_vars[j].size; | |
| 804 unsigned int jalign = stack_vars[j].alignb; | |
| 805 | |
| 806 /* Ignore objects that aren't partition representatives. */ | |
| 807 if (stack_vars[j].representative != j) | |
| 808 continue; | |
| 809 | |
| 810 /* Ignore objects too large for the remaining space. */ | |
| 811 if (isize < jsize) | |
| 812 continue; | |
| 813 | |
| 814 /* Ignore conflicting objects. */ | |
| 815 if (stack_var_conflict_p (i, j)) | |
| 816 continue; | |
| 817 | |
| 818 /* Refine the remaining space check to include alignment. */ | |
| 819 if (offset & (jalign - 1)) | |
| 820 { | |
| 821 HOST_WIDE_INT toff = offset; | |
| 822 toff += jalign - 1; | |
| 823 toff &= -(HOST_WIDE_INT)jalign; | |
| 824 if (isize - (toff - offset) < jsize) | |
| 825 continue; | |
| 826 | |
| 827 isize -= toff - offset; | |
| 828 offset = toff; | |
| 829 } | |
| 830 | |
| 831 /* UNION the objects, placing J at OFFSET. */ | |
| 832 union_stack_vars (i, j, offset); | |
| 833 | |
| 834 isize -= jsize; | |
| 835 if (isize == 0) | |
| 836 break; | |
| 837 } | |
| 838 } | |
| 839 } | |
| 840 | |
| 841 /* A debugging aid for expand_used_vars. Dump the generated partitions. */ | |
| 842 | |
| 843 static void | |
| 844 dump_stack_var_partition (void) | |
| 845 { | |
| 846 size_t si, i, j, n = stack_vars_num; | |
| 847 | |
| 848 for (si = 0; si < n; ++si) | |
| 849 { | |
| 850 i = stack_vars_sorted[si]; | |
| 851 | |
| 852 /* Skip variables that aren't partition representatives, for now. */ | |
| 853 if (stack_vars[i].representative != i) | |
| 854 continue; | |
| 855 | |
| 856 fprintf (dump_file, "Partition %lu: size " HOST_WIDE_INT_PRINT_DEC | |
| 857 " align %u\n", (unsigned long) i, stack_vars[i].size, | |
| 858 stack_vars[i].alignb); | |
| 859 | |
| 860 for (j = i; j != EOC; j = stack_vars[j].next) | |
| 861 { | |
| 862 fputc ('\t', dump_file); | |
| 863 print_generic_expr (dump_file, stack_vars[j].decl, dump_flags); | |
| 864 fprintf (dump_file, ", offset " HOST_WIDE_INT_PRINT_DEC "\n", | |
| 865 stack_vars[j].offset); | |
| 866 } | |
| 867 } | |
| 868 } | |
| 869 | |
| 870 /* Assign rtl to DECL at frame offset OFFSET. */ | |
| 871 | |
| 872 static void | |
| 873 expand_one_stack_var_at (tree decl, HOST_WIDE_INT offset) | |
| 874 { | |
| 875 HOST_WIDE_INT align; | |
| 876 rtx x; | |
| 877 | |
| 878 /* If this fails, we've overflowed the stack frame. Error nicely? */ | |
| 879 gcc_assert (offset == trunc_int_for_mode (offset, Pmode)); | |
| 880 | |
| 881 x = plus_constant (virtual_stack_vars_rtx, offset); | |
| 882 x = gen_rtx_MEM (DECL_MODE (decl), x); | |
| 883 | |
| 884 /* Set alignment we actually gave this decl. */ | |
| 885 offset -= frame_phase; | |
| 886 align = offset & -offset; | |
| 887 align *= BITS_PER_UNIT; | |
| 888 if (align > STACK_BOUNDARY || align == 0) | |
| 889 align = STACK_BOUNDARY; | |
| 890 DECL_ALIGN (decl) = align; | |
| 891 DECL_USER_ALIGN (decl) = 0; | |
| 892 | |
| 893 set_mem_attributes (x, decl, true); | |
| 894 SET_DECL_RTL (decl, x); | |
| 895 } | |
| 896 | |
| 897 /* A subroutine of expand_used_vars. Give each partition representative | |
| 898 a unique location within the stack frame. Update each partition member | |
| 899 with that location. */ | |
| 900 | |
| 901 static void | |
| 902 expand_stack_vars (bool (*pred) (tree)) | |
| 903 { | |
| 904 size_t si, i, j, n = stack_vars_num; | |
| 905 | |
| 906 for (si = 0; si < n; ++si) | |
| 907 { | |
| 908 HOST_WIDE_INT offset; | |
| 909 | |
| 910 i = stack_vars_sorted[si]; | |
| 911 | |
| 912 /* Skip variables that aren't partition representatives, for now. */ | |
| 913 if (stack_vars[i].representative != i) | |
| 914 continue; | |
| 915 | |
| 916 /* Skip variables that have already had rtl assigned. See also | |
| 917 add_stack_var where we perpetrate this pc_rtx hack. */ | |
| 918 if (DECL_RTL (stack_vars[i].decl) != pc_rtx) | |
| 919 continue; | |
| 920 | |
| 921 /* Check the predicate to see whether this variable should be | |
| 922 allocated in this pass. */ | |
| 923 if (pred && !pred (stack_vars[i].decl)) | |
| 924 continue; | |
| 925 | |
| 926 offset = alloc_stack_frame_space (stack_vars[i].size, | |
| 927 stack_vars[i].alignb); | |
| 928 | |
| 929 /* Create rtl for each variable based on their location within the | |
| 930 partition. */ | |
| 931 for (j = i; j != EOC; j = stack_vars[j].next) | |
| 932 { | |
| 933 gcc_assert (stack_vars[j].offset <= stack_vars[i].size); | |
| 934 expand_one_stack_var_at (stack_vars[j].decl, | |
| 935 stack_vars[j].offset + offset); | |
| 936 } | |
| 937 } | |
| 938 } | |
| 939 | |
| 940 /* Take into account all sizes of partitions and reset DECL_RTLs. */ | |
| 941 static HOST_WIDE_INT | |
| 942 account_stack_vars (void) | |
| 943 { | |
| 944 size_t si, j, i, n = stack_vars_num; | |
| 945 HOST_WIDE_INT size = 0; | |
| 946 | |
| 947 for (si = 0; si < n; ++si) | |
| 948 { | |
| 949 i = stack_vars_sorted[si]; | |
| 950 | |
| 951 /* Skip variables that aren't partition representatives, for now. */ | |
| 952 if (stack_vars[i].representative != i) | |
| 953 continue; | |
| 954 | |
| 955 size += stack_vars[i].size; | |
| 956 for (j = i; j != EOC; j = stack_vars[j].next) | |
| 957 SET_DECL_RTL (stack_vars[j].decl, NULL); | |
| 958 } | |
| 959 return size; | |
| 960 } | |
| 961 | |
| 962 /* A subroutine of expand_one_var. Called to immediately assign rtl | |
| 963 to a variable to be allocated in the stack frame. */ | |
| 964 | |
| 965 static void | |
| 966 expand_one_stack_var (tree var) | |
| 967 { | |
| 968 HOST_WIDE_INT size, offset, align; | |
| 969 | |
| 970 size = tree_low_cst (DECL_SIZE_UNIT (var), 1); | |
| 971 align = get_decl_align_unit (var); | |
| 972 offset = alloc_stack_frame_space (size, align); | |
| 973 | |
| 974 expand_one_stack_var_at (var, offset); | |
| 975 } | |
| 976 | |
| 977 /* A subroutine of expand_one_var. Called to assign rtl to a VAR_DECL | |
| 978 that will reside in a hard register. */ | |
| 979 | |
| 980 static void | |
| 981 expand_one_hard_reg_var (tree var) | |
| 982 { | |
| 983 rest_of_decl_compilation (var, 0, 0); | |
| 984 } | |
| 985 | |
| 986 /* A subroutine of expand_one_var. Called to assign rtl to a VAR_DECL | |
| 987 that will reside in a pseudo register. */ | |
| 988 | |
| 989 static void | |
| 990 expand_one_register_var (tree var) | |
| 991 { | |
| 992 tree type = TREE_TYPE (var); | |
| 993 int unsignedp = TYPE_UNSIGNED (type); | |
| 994 enum machine_mode reg_mode | |
| 995 = promote_mode (type, DECL_MODE (var), &unsignedp, 0); | |
| 996 rtx x = gen_reg_rtx (reg_mode); | |
| 997 | |
| 998 SET_DECL_RTL (var, x); | |
| 999 | |
| 1000 /* Note if the object is a user variable. */ | |
| 1001 if (!DECL_ARTIFICIAL (var)) | |
| 1002 mark_user_reg (x); | |
| 1003 | |
| 1004 if (POINTER_TYPE_P (type)) | |
| 1005 mark_reg_pointer (x, TYPE_ALIGN (TREE_TYPE (TREE_TYPE (var)))); | |
| 1006 } | |
| 1007 | |
| 1008 /* A subroutine of expand_one_var. Called to assign rtl to a VAR_DECL that | |
| 1009 has some associated error, e.g. its type is error-mark. We just need | |
| 1010 to pick something that won't crash the rest of the compiler. */ | |
| 1011 | |
| 1012 static void | |
| 1013 expand_one_error_var (tree var) | |
| 1014 { | |
| 1015 enum machine_mode mode = DECL_MODE (var); | |
| 1016 rtx x; | |
| 1017 | |
| 1018 if (mode == BLKmode) | |
| 1019 x = gen_rtx_MEM (BLKmode, const0_rtx); | |
| 1020 else if (mode == VOIDmode) | |
| 1021 x = const0_rtx; | |
| 1022 else | |
| 1023 x = gen_reg_rtx (mode); | |
| 1024 | |
| 1025 SET_DECL_RTL (var, x); | |
| 1026 } | |
| 1027 | |
| 1028 /* A subroutine of expand_one_var. VAR is a variable that will be | |
| 1029 allocated to the local stack frame. Return true if we wish to | |
| 1030 add VAR to STACK_VARS so that it will be coalesced with other | |
| 1031 variables. Return false to allocate VAR immediately. | |
| 1032 | |
| 1033 This function is used to reduce the number of variables considered | |
| 1034 for coalescing, which reduces the size of the quadratic problem. */ | |
| 1035 | |
| 1036 static bool | |
| 1037 defer_stack_allocation (tree var, bool toplevel) | |
| 1038 { | |
| 1039 /* If stack protection is enabled, *all* stack variables must be deferred, | |
| 1040 so that we can re-order the strings to the top of the frame. */ | |
| 1041 if (flag_stack_protect) | |
| 1042 return true; | |
| 1043 | |
| 1044 /* Variables in the outermost scope automatically conflict with | |
| 1045 every other variable. The only reason to want to defer them | |
| 1046 at all is that, after sorting, we can more efficiently pack | |
| 1047 small variables in the stack frame. Continue to defer at -O2. */ | |
| 1048 if (toplevel && optimize < 2) | |
| 1049 return false; | |
| 1050 | |
| 1051 /* Without optimization, *most* variables are allocated from the | |
| 1052 stack, which makes the quadratic problem large exactly when we | |
| 1053 want compilation to proceed as quickly as possible. On the | |
| 1054 other hand, we don't want the function's stack frame size to | |
| 1055 get completely out of hand. So we avoid adding scalars and | |
| 1056 "small" aggregates to the list at all. */ | |
| 1057 if (optimize == 0 && tree_low_cst (DECL_SIZE_UNIT (var), 1) < 32) | |
| 1058 return false; | |
| 1059 | |
| 1060 return true; | |
| 1061 } | |
| 1062 | |
| 1063 /* A subroutine of expand_used_vars. Expand one variable according to | |
| 1064 its flavor. Variables to be placed on the stack are not actually | |
| 1065 expanded yet, merely recorded. | |
| 1066 When REALLY_EXPAND is false, only add stack values to be allocated. | |
| 1067 Return stack usage this variable is supposed to take. | |
| 1068 */ | |
| 1069 | |
| 1070 static HOST_WIDE_INT | |
| 1071 expand_one_var (tree var, bool toplevel, bool really_expand) | |
| 1072 { | |
| 1073 if (SUPPORTS_STACK_ALIGNMENT | |
| 1074 && TREE_TYPE (var) != error_mark_node | |
| 1075 && TREE_CODE (var) == VAR_DECL) | |
| 1076 { | |
| 1077 unsigned int align; | |
| 1078 | |
| 1079 /* Because we don't know if VAR will be in register or on stack, | |
| 1080 we conservatively assume it will be on stack even if VAR is | |
| 1081 eventually put into register after RA pass. For non-automatic | |
| 1082 variables, which won't be on stack, we collect alignment of | |
| 1083 type and ignore user specified alignment. */ | |
| 1084 if (TREE_STATIC (var) || DECL_EXTERNAL (var)) | |
| 1085 align = TYPE_ALIGN (TREE_TYPE (var)); | |
| 1086 else | |
| 1087 align = DECL_ALIGN (var); | |
| 1088 | |
| 1089 if (crtl->stack_alignment_estimated < align) | |
| 1090 { | |
| 1091 /* stack_alignment_estimated shouldn't change after stack | |
| 1092 realign decision made */ | |
| 1093 gcc_assert(!crtl->stack_realign_processed); | |
| 1094 crtl->stack_alignment_estimated = align; | |
| 1095 } | |
| 1096 } | |
| 1097 | |
| 1098 if (TREE_CODE (var) != VAR_DECL) | |
| 1099 ; | |
| 1100 else if (DECL_EXTERNAL (var)) | |
| 1101 ; | |
| 1102 else if (DECL_HAS_VALUE_EXPR_P (var)) | |
| 1103 ; | |
| 1104 else if (TREE_STATIC (var)) | |
| 1105 ; | |
| 1106 else if (DECL_RTL_SET_P (var)) | |
| 1107 ; | |
| 1108 else if (TREE_TYPE (var) == error_mark_node) | |
| 1109 { | |
| 1110 if (really_expand) | |
| 1111 expand_one_error_var (var); | |
| 1112 } | |
| 1113 else if (DECL_HARD_REGISTER (var)) | |
| 1114 { | |
| 1115 if (really_expand) | |
| 1116 expand_one_hard_reg_var (var); | |
| 1117 } | |
| 1118 else if (use_register_for_decl (var)) | |
| 1119 { | |
| 1120 if (really_expand) | |
| 1121 expand_one_register_var (var); | |
| 1122 } | |
| 1123 else if (defer_stack_allocation (var, toplevel)) | |
| 1124 add_stack_var (var); | |
| 1125 else | |
| 1126 { | |
| 1127 if (really_expand) | |
| 1128 expand_one_stack_var (var); | |
| 1129 return tree_low_cst (DECL_SIZE_UNIT (var), 1); | |
| 1130 } | |
| 1131 return 0; | |
| 1132 } | |
| 1133 | |
| 1134 /* A subroutine of expand_used_vars. Walk down through the BLOCK tree | |
| 1135 expanding variables. Those variables that can be put into registers | |
| 1136 are allocated pseudos; those that can't are put on the stack. | |
| 1137 | |
| 1138 TOPLEVEL is true if this is the outermost BLOCK. */ | |
| 1139 | |
| 1140 static void | |
| 1141 expand_used_vars_for_block (tree block, bool toplevel) | |
| 1142 { | |
| 1143 size_t i, j, old_sv_num, this_sv_num, new_sv_num; | |
| 1144 tree t; | |
| 1145 | |
| 1146 old_sv_num = toplevel ? 0 : stack_vars_num; | |
| 1147 | |
| 1148 /* Expand all variables at this level. */ | |
| 1149 for (t = BLOCK_VARS (block); t ; t = TREE_CHAIN (t)) | |
| 1150 if (TREE_USED (t)) | |
| 1151 expand_one_var (t, toplevel, true); | |
| 1152 | |
| 1153 this_sv_num = stack_vars_num; | |
| 1154 | |
| 1155 /* Expand all variables at containing levels. */ | |
| 1156 for (t = BLOCK_SUBBLOCKS (block); t ; t = BLOCK_CHAIN (t)) | |
| 1157 expand_used_vars_for_block (t, false); | |
| 1158 | |
| 1159 /* Since we do not track exact variable lifetimes (which is not even | |
| 1160 possible for variables whose address escapes), we mirror the block | |
| 1161 tree in the interference graph. Here we cause all variables at this | |
| 1162 level, and all sublevels, to conflict. Do make certain that a | |
| 1163 variable conflicts with itself. */ | |
| 1164 if (old_sv_num < this_sv_num) | |
| 1165 { | |
| 1166 new_sv_num = stack_vars_num; | |
| 1167 resize_stack_vars_conflict (new_sv_num); | |
| 1168 | |
| 1169 for (i = old_sv_num; i < new_sv_num; ++i) | |
| 1170 for (j = i < this_sv_num ? i+1 : this_sv_num; j-- > old_sv_num ;) | |
| 1171 add_stack_var_conflict (i, j); | |
| 1172 } | |
| 1173 } | |
| 1174 | |
| 1175 /* A subroutine of expand_used_vars. Walk down through the BLOCK tree | |
| 1176 and clear TREE_USED on all local variables. */ | |
| 1177 | |
| 1178 static void | |
| 1179 clear_tree_used (tree block) | |
| 1180 { | |
| 1181 tree t; | |
| 1182 | |
| 1183 for (t = BLOCK_VARS (block); t ; t = TREE_CHAIN (t)) | |
| 1184 /* if (!TREE_STATIC (t) && !DECL_EXTERNAL (t)) */ | |
| 1185 TREE_USED (t) = 0; | |
| 1186 | |
| 1187 for (t = BLOCK_SUBBLOCKS (block); t ; t = BLOCK_CHAIN (t)) | |
| 1188 clear_tree_used (t); | |
| 1189 } | |
| 1190 | |
| 1191 /* Examine TYPE and determine a bit mask of the following features. */ | |
| 1192 | |
| 1193 #define SPCT_HAS_LARGE_CHAR_ARRAY 1 | |
| 1194 #define SPCT_HAS_SMALL_CHAR_ARRAY 2 | |
| 1195 #define SPCT_HAS_ARRAY 4 | |
| 1196 #define SPCT_HAS_AGGREGATE 8 | |
| 1197 | |
| 1198 static unsigned int | |
| 1199 stack_protect_classify_type (tree type) | |
| 1200 { | |
| 1201 unsigned int ret = 0; | |
| 1202 tree t; | |
| 1203 | |
| 1204 switch (TREE_CODE (type)) | |
| 1205 { | |
| 1206 case ARRAY_TYPE: | |
| 1207 t = TYPE_MAIN_VARIANT (TREE_TYPE (type)); | |
| 1208 if (t == char_type_node | |
| 1209 || t == signed_char_type_node | |
| 1210 || t == unsigned_char_type_node) | |
| 1211 { | |
| 1212 unsigned HOST_WIDE_INT max = PARAM_VALUE (PARAM_SSP_BUFFER_SIZE); | |
| 1213 unsigned HOST_WIDE_INT len; | |
| 1214 | |
| 1215 if (!TYPE_SIZE_UNIT (type) | |
| 1216 || !host_integerp (TYPE_SIZE_UNIT (type), 1)) | |
| 1217 len = max; | |
| 1218 else | |
| 1219 len = tree_low_cst (TYPE_SIZE_UNIT (type), 1); | |
| 1220 | |
| 1221 if (len < max) | |
| 1222 ret = SPCT_HAS_SMALL_CHAR_ARRAY | SPCT_HAS_ARRAY; | |
| 1223 else | |
| 1224 ret = SPCT_HAS_LARGE_CHAR_ARRAY | SPCT_HAS_ARRAY; | |
| 1225 } | |
| 1226 else | |
| 1227 ret = SPCT_HAS_ARRAY; | |
| 1228 break; | |
| 1229 | |
| 1230 case UNION_TYPE: | |
| 1231 case QUAL_UNION_TYPE: | |
| 1232 case RECORD_TYPE: | |
| 1233 ret = SPCT_HAS_AGGREGATE; | |
| 1234 for (t = TYPE_FIELDS (type); t ; t = TREE_CHAIN (t)) | |
| 1235 if (TREE_CODE (t) == FIELD_DECL) | |
| 1236 ret |= stack_protect_classify_type (TREE_TYPE (t)); | |
| 1237 break; | |
| 1238 | |
| 1239 default: | |
| 1240 break; | |
| 1241 } | |
| 1242 | |
| 1243 return ret; | |
| 1244 } | |
| 1245 | |
| 1246 /* Return nonzero if DECL should be segregated into the "vulnerable" upper | |
| 1247 part of the local stack frame. Remember if we ever return nonzero for | |
| 1248 any variable in this function. The return value is the phase number in | |
| 1249 which the variable should be allocated. */ | |
| 1250 | |
| 1251 static int | |
| 1252 stack_protect_decl_phase (tree decl) | |
| 1253 { | |
| 1254 unsigned int bits = stack_protect_classify_type (TREE_TYPE (decl)); | |
| 1255 int ret = 0; | |
| 1256 | |
| 1257 if (bits & SPCT_HAS_SMALL_CHAR_ARRAY) | |
| 1258 has_short_buffer = true; | |
| 1259 | |
| 1260 if (flag_stack_protect == 2) | |
| 1261 { | |
| 1262 if ((bits & (SPCT_HAS_SMALL_CHAR_ARRAY | SPCT_HAS_LARGE_CHAR_ARRAY)) | |
| 1263 && !(bits & SPCT_HAS_AGGREGATE)) | |
| 1264 ret = 1; | |
| 1265 else if (bits & SPCT_HAS_ARRAY) | |
| 1266 ret = 2; | |
| 1267 } | |
| 1268 else | |
| 1269 ret = (bits & SPCT_HAS_LARGE_CHAR_ARRAY) != 0; | |
| 1270 | |
| 1271 if (ret) | |
| 1272 has_protected_decls = true; | |
| 1273 | |
| 1274 return ret; | |
| 1275 } | |
| 1276 | |
| 1277 /* Two helper routines that check for phase 1 and phase 2. These are used | |
| 1278 as callbacks for expand_stack_vars. */ | |
| 1279 | |
| 1280 static bool | |
| 1281 stack_protect_decl_phase_1 (tree decl) | |
| 1282 { | |
| 1283 return stack_protect_decl_phase (decl) == 1; | |
| 1284 } | |
| 1285 | |
| 1286 static bool | |
| 1287 stack_protect_decl_phase_2 (tree decl) | |
| 1288 { | |
| 1289 return stack_protect_decl_phase (decl) == 2; | |
| 1290 } | |
| 1291 | |
| 1292 /* Ensure that variables in different stack protection phases conflict | |
| 1293 so that they are not merged and share the same stack slot. */ | |
| 1294 | |
| 1295 static void | |
| 1296 add_stack_protection_conflicts (void) | |
| 1297 { | |
| 1298 size_t i, j, n = stack_vars_num; | |
| 1299 unsigned char *phase; | |
| 1300 | |
| 1301 phase = XNEWVEC (unsigned char, n); | |
| 1302 for (i = 0; i < n; ++i) | |
| 1303 phase[i] = stack_protect_decl_phase (stack_vars[i].decl); | |
| 1304 | |
| 1305 for (i = 0; i < n; ++i) | |
| 1306 { | |
| 1307 unsigned char ph_i = phase[i]; | |
| 1308 for (j = 0; j < i; ++j) | |
| 1309 if (ph_i != phase[j]) | |
| 1310 add_stack_var_conflict (i, j); | |
| 1311 } | |
| 1312 | |
| 1313 XDELETEVEC (phase); | |
| 1314 } | |
| 1315 | |
| 1316 /* Create a decl for the guard at the top of the stack frame. */ | |
| 1317 | |
| 1318 static void | |
| 1319 create_stack_guard (void) | |
| 1320 { | |
| 1321 tree guard = build_decl (VAR_DECL, NULL, ptr_type_node); | |
| 1322 TREE_THIS_VOLATILE (guard) = 1; | |
| 1323 TREE_USED (guard) = 1; | |
| 1324 expand_one_stack_var (guard); | |
| 1325 crtl->stack_protect_guard = guard; | |
| 1326 } | |
| 1327 | |
| 1328 /* A subroutine of expand_used_vars. Walk down through the BLOCK tree | |
| 1329 expanding variables. Those variables that can be put into registers | |
| 1330 are allocated pseudos; those that can't are put on the stack. | |
| 1331 | |
| 1332 TOPLEVEL is true if this is the outermost BLOCK. */ | |
| 1333 | |
| 1334 static HOST_WIDE_INT | |
| 1335 account_used_vars_for_block (tree block, bool toplevel) | |
| 1336 { | |
| 1337 size_t i, j, old_sv_num, this_sv_num, new_sv_num; | |
| 1338 tree t; | |
| 1339 HOST_WIDE_INT size = 0; | |
| 1340 | |
| 1341 old_sv_num = toplevel ? 0 : stack_vars_num; | |
| 1342 | |
| 1343 /* Expand all variables at this level. */ | |
| 1344 for (t = BLOCK_VARS (block); t ; t = TREE_CHAIN (t)) | |
| 1345 if (TREE_USED (t)) | |
| 1346 size += expand_one_var (t, toplevel, false); | |
| 1347 | |
| 1348 this_sv_num = stack_vars_num; | |
| 1349 | |
| 1350 /* Expand all variables at containing levels. */ | |
| 1351 for (t = BLOCK_SUBBLOCKS (block); t ; t = BLOCK_CHAIN (t)) | |
| 1352 size += account_used_vars_for_block (t, false); | |
| 1353 | |
| 1354 /* Since we do not track exact variable lifetimes (which is not even | |
| 1355 possible for variables whose address escapes), we mirror the block | |
| 1356 tree in the interference graph. Here we cause all variables at this | |
| 1357 level, and all sublevels, to conflict. Do make certain that a | |
| 1358 variable conflicts with itself. */ | |
| 1359 if (old_sv_num < this_sv_num) | |
| 1360 { | |
| 1361 new_sv_num = stack_vars_num; | |
| 1362 resize_stack_vars_conflict (new_sv_num); | |
| 1363 | |
| 1364 for (i = old_sv_num; i < new_sv_num; ++i) | |
| 1365 for (j = i < this_sv_num ? i+1 : this_sv_num; j-- > old_sv_num ;) | |
| 1366 add_stack_var_conflict (i, j); | |
| 1367 } | |
| 1368 return size; | |
| 1369 } | |
| 1370 | |
| 1371 /* Prepare for expanding variables. */ | |
| 1372 static void | |
| 1373 init_vars_expansion (void) | |
| 1374 { | |
| 1375 tree t; | |
| 1376 /* Set TREE_USED on all variables in the local_decls. */ | |
| 1377 for (t = cfun->local_decls; t; t = TREE_CHAIN (t)) | |
| 1378 TREE_USED (TREE_VALUE (t)) = 1; | |
| 1379 | |
| 1380 /* Clear TREE_USED on all variables associated with a block scope. */ | |
| 1381 clear_tree_used (DECL_INITIAL (current_function_decl)); | |
| 1382 | |
| 1383 /* Initialize local stack smashing state. */ | |
| 1384 has_protected_decls = false; | |
| 1385 has_short_buffer = false; | |
| 1386 } | |
| 1387 | |
| 1388 /* Free up stack variable graph data. */ | |
| 1389 static void | |
| 1390 fini_vars_expansion (void) | |
| 1391 { | |
| 1392 XDELETEVEC (stack_vars); | |
| 1393 XDELETEVEC (stack_vars_sorted); | |
| 1394 XDELETEVEC (stack_vars_conflict); | |
| 1395 stack_vars = NULL; | |
| 1396 stack_vars_alloc = stack_vars_num = 0; | |
| 1397 stack_vars_conflict = NULL; | |
| 1398 stack_vars_conflict_alloc = 0; | |
| 1399 } | |
| 1400 | |
| 1401 /* Make a fair guess for the size of the stack frame of the current | |
| 1402 function. This doesn't have to be exact, the result is only used | |
| 1403 in the inline heuristics. So we don't want to run the full stack | |
| 1404 var packing algorithm (which is quadratic in the number of stack | |
| 1405 vars). Instead, we calculate the total size of all stack vars. | |
| 1406 This turns out to be a pretty fair estimate -- packing of stack | |
| 1407 vars doesn't happen very often. */ | |
| 1408 | |
| 1409 HOST_WIDE_INT | |
| 1410 estimated_stack_frame_size (void) | |
| 1411 { | |
| 1412 HOST_WIDE_INT size = 0; | |
| 1413 size_t i; | |
| 1414 tree t, outer_block = DECL_INITIAL (current_function_decl); | |
| 1415 | |
| 1416 init_vars_expansion (); | |
| 1417 | |
| 1418 for (t = cfun->local_decls; t; t = TREE_CHAIN (t)) | |
| 1419 { | |
| 1420 tree var = TREE_VALUE (t); | |
| 1421 | |
| 1422 if (TREE_USED (var)) | |
| 1423 size += expand_one_var (var, true, false); | |
| 1424 TREE_USED (var) = 1; | |
| 1425 } | |
| 1426 size += account_used_vars_for_block (outer_block, true); | |
| 1427 | |
| 1428 if (stack_vars_num > 0) | |
| 1429 { | |
| 1430 /* Fake sorting the stack vars for account_stack_vars (). */ | |
| 1431 stack_vars_sorted = XNEWVEC (size_t, stack_vars_num); | |
| 1432 for (i = 0; i < stack_vars_num; ++i) | |
| 1433 stack_vars_sorted[i] = i; | |
| 1434 size += account_stack_vars (); | |
| 1435 fini_vars_expansion (); | |
| 1436 } | |
| 1437 | |
| 1438 return size; | |
| 1439 } | |
| 1440 | |
| 1441 /* Expand all variables used in the function. */ | |
| 1442 | |
| 1443 static void | |
| 1444 expand_used_vars (void) | |
| 1445 { | |
| 1446 tree t, next, outer_block = DECL_INITIAL (current_function_decl); | |
| 1447 | |
| 1448 /* Compute the phase of the stack frame for this function. */ | |
| 1449 { | |
| 1450 int align = PREFERRED_STACK_BOUNDARY / BITS_PER_UNIT; | |
| 1451 int off = STARTING_FRAME_OFFSET % align; | |
| 1452 frame_phase = off ? align - off : 0; | |
| 1453 } | |
| 1454 | |
| 1455 init_vars_expansion (); | |
| 1456 | |
| 1457 /* At this point all variables on the local_decls with TREE_USED | |
| 1458 set are not associated with any block scope. Lay them out. */ | |
| 1459 t = cfun->local_decls; | |
| 1460 cfun->local_decls = NULL_TREE; | |
| 1461 for (; t; t = next) | |
| 1462 { | |
| 1463 tree var = TREE_VALUE (t); | |
| 1464 bool expand_now = false; | |
| 1465 | |
| 1466 next = TREE_CHAIN (t); | |
| 1467 | |
| 1468 /* We didn't set a block for static or extern because it's hard | |
| 1469 to tell the difference between a global variable (re)declared | |
| 1470 in a local scope, and one that's really declared there to | |
| 1471 begin with. And it doesn't really matter much, since we're | |
| 1472 not giving them stack space. Expand them now. */ | |
| 1473 if (TREE_STATIC (var) || DECL_EXTERNAL (var)) | |
| 1474 expand_now = true; | |
| 1475 | |
| 1476 /* Any variable that could have been hoisted into an SSA_NAME | |
| 1477 will have been propagated anywhere the optimizers chose, | |
| 1478 i.e. not confined to their original block. Allocate them | |
| 1479 as if they were defined in the outermost scope. */ | |
| 1480 else if (is_gimple_reg (var)) | |
| 1481 expand_now = true; | |
| 1482 | |
| 1483 /* If the variable is not associated with any block, then it | |
| 1484 was created by the optimizers, and could be live anywhere | |
| 1485 in the function. */ | |
| 1486 else if (TREE_USED (var)) | |
| 1487 expand_now = true; | |
| 1488 | |
| 1489 /* Finally, mark all variables on the list as used. We'll use | |
| 1490 this in a moment when we expand those associated with scopes. */ | |
| 1491 TREE_USED (var) = 1; | |
| 1492 | |
| 1493 if (expand_now) | |
| 1494 { | |
| 1495 expand_one_var (var, true, true); | |
| 1496 if (DECL_ARTIFICIAL (var) && !DECL_IGNORED_P (var)) | |
| 1497 { | |
| 1498 rtx rtl = DECL_RTL_IF_SET (var); | |
| 1499 | |
| 1500 /* Keep artificial non-ignored vars in cfun->local_decls | |
| 1501 chain until instantiate_decls. */ | |
| 1502 if (rtl && (MEM_P (rtl) || GET_CODE (rtl) == CONCAT)) | |
| 1503 { | |
| 1504 TREE_CHAIN (t) = cfun->local_decls; | |
| 1505 cfun->local_decls = t; | |
| 1506 continue; | |
| 1507 } | |
| 1508 } | |
| 1509 } | |
| 1510 | |
| 1511 ggc_free (t); | |
| 1512 } | |
| 1513 | |
| 1514 /* At this point, all variables within the block tree with TREE_USED | |
| 1515 set are actually used by the optimized function. Lay them out. */ | |
| 1516 expand_used_vars_for_block (outer_block, true); | |
| 1517 | |
| 1518 if (stack_vars_num > 0) | |
| 1519 { | |
| 1520 /* Due to the way alias sets work, no variables with non-conflicting | |
| 1521 alias sets may be assigned the same address. Add conflicts to | |
| 1522 reflect this. */ | |
| 1523 add_alias_set_conflicts (); | |
| 1524 | |
| 1525 /* If stack protection is enabled, we don't share space between | |
| 1526 vulnerable data and non-vulnerable data. */ | |
| 1527 if (flag_stack_protect) | |
| 1528 add_stack_protection_conflicts (); | |
| 1529 | |
| 1530 /* Now that we have collected all stack variables, and have computed a | |
| 1531 minimal interference graph, attempt to save some stack space. */ | |
| 1532 partition_stack_vars (); | |
| 1533 if (dump_file) | |
| 1534 dump_stack_var_partition (); | |
| 1535 } | |
| 1536 | |
| 1537 /* There are several conditions under which we should create a | |
| 1538 stack guard: protect-all, alloca used, protected decls present. */ | |
| 1539 if (flag_stack_protect == 2 | |
| 1540 || (flag_stack_protect | |
| 1541 && (cfun->calls_alloca || has_protected_decls))) | |
| 1542 create_stack_guard (); | |
| 1543 | |
| 1544 /* Assign rtl to each variable based on these partitions. */ | |
| 1545 if (stack_vars_num > 0) | |
| 1546 { | |
| 1547 /* Reorder decls to be protected by iterating over the variables | |
| 1548 array multiple times, and allocating out of each phase in turn. */ | |
| 1549 /* ??? We could probably integrate this into the qsort we did | |
| 1550 earlier, such that we naturally see these variables first, | |
| 1551 and thus naturally allocate things in the right order. */ | |
| 1552 if (has_protected_decls) | |
| 1553 { | |
| 1554 /* Phase 1 contains only character arrays. */ | |
| 1555 expand_stack_vars (stack_protect_decl_phase_1); | |
| 1556 | |
| 1557 /* Phase 2 contains other kinds of arrays. */ | |
| 1558 if (flag_stack_protect == 2) | |
| 1559 expand_stack_vars (stack_protect_decl_phase_2); | |
| 1560 } | |
| 1561 | |
| 1562 expand_stack_vars (NULL); | |
| 1563 | |
| 1564 fini_vars_expansion (); | |
| 1565 } | |
| 1566 | |
| 1567 /* If the target requires that FRAME_OFFSET be aligned, do it. */ | |
| 1568 if (STACK_ALIGNMENT_NEEDED) | |
| 1569 { | |
| 1570 HOST_WIDE_INT align = PREFERRED_STACK_BOUNDARY / BITS_PER_UNIT; | |
| 1571 if (!FRAME_GROWS_DOWNWARD) | |
| 1572 frame_offset += align - 1; | |
| 1573 frame_offset &= -align; | |
| 1574 } | |
| 1575 } | |
| 1576 | |
| 1577 | |
| 1578 /* If we need to produce a detailed dump, print the tree representation | |
| 1579 for STMT to the dump file. SINCE is the last RTX after which the RTL | |
| 1580 generated for STMT should have been appended. */ | |
| 1581 | |
| 1582 static void | |
| 1583 maybe_dump_rtl_for_gimple_stmt (gimple stmt, rtx since) | |
| 1584 { | |
| 1585 if (dump_file && (dump_flags & TDF_DETAILS)) | |
| 1586 { | |
| 1587 fprintf (dump_file, "\n;; "); | |
| 1588 print_gimple_stmt (dump_file, stmt, 0, TDF_SLIM); | |
| 1589 fprintf (dump_file, "\n"); | |
| 1590 | |
| 1591 print_rtl (dump_file, since ? NEXT_INSN (since) : since); | |
| 1592 } | |
| 1593 } | |
| 1594 | |
| 1595 /* Maps the blocks that do not contain tree labels to rtx labels. */ | |
| 1596 | |
| 1597 static struct pointer_map_t *lab_rtx_for_bb; | |
| 1598 | |
| 1599 /* Returns the label_rtx expression for a label starting basic block BB. */ | |
| 1600 | |
| 1601 static rtx | |
| 1602 label_rtx_for_bb (basic_block bb ATTRIBUTE_UNUSED) | |
| 1603 { | |
| 1604 gimple_stmt_iterator gsi; | |
| 1605 tree lab; | |
| 1606 gimple lab_stmt; | |
| 1607 void **elt; | |
| 1608 | |
| 1609 if (bb->flags & BB_RTL) | |
| 1610 return block_label (bb); | |
| 1611 | |
| 1612 elt = pointer_map_contains (lab_rtx_for_bb, bb); | |
| 1613 if (elt) | |
| 1614 return (rtx) *elt; | |
| 1615 | |
| 1616 /* Find the tree label if it is present. */ | |
| 1617 | |
| 1618 for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi)) | |
| 1619 { | |
| 1620 lab_stmt = gsi_stmt (gsi); | |
| 1621 if (gimple_code (lab_stmt) != GIMPLE_LABEL) | |
| 1622 break; | |
| 1623 | |
| 1624 lab = gimple_label_label (lab_stmt); | |
| 1625 if (DECL_NONLOCAL (lab)) | |
| 1626 break; | |
| 1627 | |
| 1628 return label_rtx (lab); | |
| 1629 } | |
| 1630 | |
| 1631 elt = pointer_map_insert (lab_rtx_for_bb, bb); | |
| 1632 *elt = gen_label_rtx (); | |
| 1633 return (rtx) *elt; | |
| 1634 } | |
| 1635 | |
| 1636 | |
| 1637 /* A subroutine of expand_gimple_basic_block. Expand one GIMPLE_COND. | |
| 1638 Returns a new basic block if we've terminated the current basic | |
| 1639 block and created a new one. */ | |
| 1640 | |
| 1641 static basic_block | |
| 1642 expand_gimple_cond (basic_block bb, gimple stmt) | |
| 1643 { | |
| 1644 basic_block new_bb, dest; | |
| 1645 edge new_edge; | |
| 1646 edge true_edge; | |
| 1647 edge false_edge; | |
| 1648 tree pred = gimple_cond_pred_to_tree (stmt); | |
| 1649 rtx last2, last; | |
| 1650 | |
| 1651 last2 = last = get_last_insn (); | |
| 1652 | |
| 1653 extract_true_false_edges_from_block (bb, &true_edge, &false_edge); | |
| 1654 if (gimple_has_location (stmt)) | |
| 1655 { | |
| 1656 set_curr_insn_source_location (gimple_location (stmt)); | |
| 1657 set_curr_insn_block (gimple_block (stmt)); | |
| 1658 } | |
| 1659 | |
| 1660 /* These flags have no purpose in RTL land. */ | |
| 1661 true_edge->flags &= ~EDGE_TRUE_VALUE; | |
| 1662 false_edge->flags &= ~EDGE_FALSE_VALUE; | |
| 1663 | |
| 1664 /* We can either have a pure conditional jump with one fallthru edge or | |
| 1665 two-way jump that needs to be decomposed into two basic blocks. */ | |
| 1666 if (false_edge->dest == bb->next_bb) | |
| 1667 { | |
| 1668 jumpif (pred, label_rtx_for_bb (true_edge->dest)); | |
| 1669 add_reg_br_prob_note (last, true_edge->probability); | |
| 1670 maybe_dump_rtl_for_gimple_stmt (stmt, last); | |
| 1671 if (true_edge->goto_locus) | |
| 1672 { | |
| 1673 set_curr_insn_source_location (true_edge->goto_locus); | |
| 1674 set_curr_insn_block (true_edge->goto_block); | |
| 1675 true_edge->goto_locus = curr_insn_locator (); | |
| 1676 } | |
| 1677 true_edge->goto_block = NULL; | |
| 1678 false_edge->flags |= EDGE_FALLTHRU; | |
| 1679 ggc_free (pred); | |
| 1680 return NULL; | |
| 1681 } | |
| 1682 if (true_edge->dest == bb->next_bb) | |
| 1683 { | |
| 1684 jumpifnot (pred, label_rtx_for_bb (false_edge->dest)); | |
| 1685 add_reg_br_prob_note (last, false_edge->probability); | |
| 1686 maybe_dump_rtl_for_gimple_stmt (stmt, last); | |
| 1687 if (false_edge->goto_locus) | |
| 1688 { | |
| 1689 set_curr_insn_source_location (false_edge->goto_locus); | |
| 1690 set_curr_insn_block (false_edge->goto_block); | |
| 1691 false_edge->goto_locus = curr_insn_locator (); | |
| 1692 } | |
| 1693 false_edge->goto_block = NULL; | |
| 1694 true_edge->flags |= EDGE_FALLTHRU; | |
| 1695 ggc_free (pred); | |
| 1696 return NULL; | |
| 1697 } | |
| 1698 | |
| 1699 jumpif (pred, label_rtx_for_bb (true_edge->dest)); | |
| 1700 add_reg_br_prob_note (last, true_edge->probability); | |
| 1701 last = get_last_insn (); | |
| 1702 if (false_edge->goto_locus) | |
| 1703 { | |
| 1704 set_curr_insn_source_location (false_edge->goto_locus); | |
| 1705 set_curr_insn_block (false_edge->goto_block); | |
| 1706 false_edge->goto_locus = curr_insn_locator (); | |
| 1707 } | |
| 1708 false_edge->goto_block = NULL; | |
| 1709 emit_jump (label_rtx_for_bb (false_edge->dest)); | |
| 1710 | |
| 1711 BB_END (bb) = last; | |
| 1712 if (BARRIER_P (BB_END (bb))) | |
| 1713 BB_END (bb) = PREV_INSN (BB_END (bb)); | |
| 1714 update_bb_for_insn (bb); | |
| 1715 | |
| 1716 new_bb = create_basic_block (NEXT_INSN (last), get_last_insn (), bb); | |
| 1717 dest = false_edge->dest; | |
| 1718 redirect_edge_succ (false_edge, new_bb); | |
| 1719 false_edge->flags |= EDGE_FALLTHRU; | |
| 1720 new_bb->count = false_edge->count; | |
| 1721 new_bb->frequency = EDGE_FREQUENCY (false_edge); | |
| 1722 new_edge = make_edge (new_bb, dest, 0); | |
| 1723 new_edge->probability = REG_BR_PROB_BASE; | |
| 1724 new_edge->count = new_bb->count; | |
| 1725 if (BARRIER_P (BB_END (new_bb))) | |
| 1726 BB_END (new_bb) = PREV_INSN (BB_END (new_bb)); | |
| 1727 update_bb_for_insn (new_bb); | |
| 1728 | |
| 1729 maybe_dump_rtl_for_gimple_stmt (stmt, last2); | |
| 1730 | |
| 1731 if (true_edge->goto_locus) | |
| 1732 { | |
| 1733 set_curr_insn_source_location (true_edge->goto_locus); | |
| 1734 set_curr_insn_block (true_edge->goto_block); | |
| 1735 true_edge->goto_locus = curr_insn_locator (); | |
| 1736 } | |
| 1737 true_edge->goto_block = NULL; | |
| 1738 | |
| 1739 ggc_free (pred); | |
| 1740 return new_bb; | |
| 1741 } | |
| 1742 | |
| 1743 /* A subroutine of expand_gimple_basic_block. Expand one GIMPLE_CALL | |
| 1744 that has CALL_EXPR_TAILCALL set. Returns non-null if we actually | |
| 1745 generated a tail call (something that might be denied by the ABI | |
| 1746 rules governing the call; see calls.c). | |
| 1747 | |
| 1748 Sets CAN_FALLTHRU if we generated a *conditional* tail call, and | |
| 1749 can still reach the rest of BB. The case here is __builtin_sqrt, | |
| 1750 where the NaN result goes through the external function (with a | |
| 1751 tailcall) and the normal result happens via a sqrt instruction. */ | |
| 1752 | |
| 1753 static basic_block | |
| 1754 expand_gimple_tailcall (basic_block bb, gimple stmt, bool *can_fallthru) | |
| 1755 { | |
| 1756 rtx last2, last; | |
| 1757 edge e; | |
| 1758 edge_iterator ei; | |
| 1759 int probability; | |
| 1760 gcov_type count; | |
| 1761 tree stmt_tree = gimple_to_tree (stmt); | |
| 1762 | |
| 1763 last2 = last = get_last_insn (); | |
| 1764 | |
| 1765 expand_expr_stmt (stmt_tree); | |
| 1766 | |
| 1767 release_stmt_tree (stmt, stmt_tree); | |
| 1768 | |
| 1769 for (last = NEXT_INSN (last); last; last = NEXT_INSN (last)) | |
| 1770 if (CALL_P (last) && SIBLING_CALL_P (last)) | |
| 1771 goto found; | |
| 1772 | |
| 1773 maybe_dump_rtl_for_gimple_stmt (stmt, last2); | |
| 1774 | |
| 1775 *can_fallthru = true; | |
| 1776 return NULL; | |
| 1777 | |
| 1778 found: | |
| 1779 /* ??? Wouldn't it be better to just reset any pending stack adjust? | |
| 1780 Any instructions emitted here are about to be deleted. */ | |
| 1781 do_pending_stack_adjust (); | |
| 1782 | |
| 1783 /* Remove any non-eh, non-abnormal edges that don't go to exit. */ | |
| 1784 /* ??? I.e. the fallthrough edge. HOWEVER! If there were to be | |
| 1785 EH or abnormal edges, we shouldn't have created a tail call in | |
| 1786 the first place. So it seems to me we should just be removing | |
| 1787 all edges here, or redirecting the existing fallthru edge to | |
| 1788 the exit block. */ | |
| 1789 | |
| 1790 probability = 0; | |
| 1791 count = 0; | |
| 1792 | |
| 1793 for (ei = ei_start (bb->succs); (e = ei_safe_edge (ei)); ) | |
| 1794 { | |
| 1795 if (!(e->flags & (EDGE_ABNORMAL | EDGE_EH))) | |
| 1796 { | |
| 1797 if (e->dest != EXIT_BLOCK_PTR) | |
| 1798 { | |
| 1799 e->dest->count -= e->count; | |
| 1800 e->dest->frequency -= EDGE_FREQUENCY (e); | |
| 1801 if (e->dest->count < 0) | |
| 1802 e->dest->count = 0; | |
| 1803 if (e->dest->frequency < 0) | |
| 1804 e->dest->frequency = 0; | |
| 1805 } | |
| 1806 count += e->count; | |
| 1807 probability += e->probability; | |
| 1808 remove_edge (e); | |
| 1809 } | |
| 1810 else | |
| 1811 ei_next (&ei); | |
| 1812 } | |
| 1813 | |
| 1814 /* This is somewhat ugly: the call_expr expander often emits instructions | |
| 1815 after the sibcall (to perform the function return). These confuse the | |
| 1816 find_many_sub_basic_blocks code, so we need to get rid of these. */ | |
| 1817 last = NEXT_INSN (last); | |
| 1818 gcc_assert (BARRIER_P (last)); | |
| 1819 | |
| 1820 *can_fallthru = false; | |
| 1821 while (NEXT_INSN (last)) | |
| 1822 { | |
| 1823 /* For instance an sqrt builtin expander expands if with | |
| 1824 sibcall in the then and label for `else`. */ | |
| 1825 if (LABEL_P (NEXT_INSN (last))) | |
| 1826 { | |
| 1827 *can_fallthru = true; | |
| 1828 break; | |
| 1829 } | |
| 1830 delete_insn (NEXT_INSN (last)); | |
| 1831 } | |
| 1832 | |
| 1833 e = make_edge (bb, EXIT_BLOCK_PTR, EDGE_ABNORMAL | EDGE_SIBCALL); | |
| 1834 e->probability += probability; | |
| 1835 e->count += count; | |
| 1836 BB_END (bb) = last; | |
| 1837 update_bb_for_insn (bb); | |
| 1838 | |
| 1839 if (NEXT_INSN (last)) | |
| 1840 { | |
| 1841 bb = create_basic_block (NEXT_INSN (last), get_last_insn (), bb); | |
| 1842 | |
| 1843 last = BB_END (bb); | |
| 1844 if (BARRIER_P (last)) | |
| 1845 BB_END (bb) = PREV_INSN (last); | |
| 1846 } | |
| 1847 | |
| 1848 maybe_dump_rtl_for_gimple_stmt (stmt, last2); | |
| 1849 | |
| 1850 return bb; | |
| 1851 } | |
| 1852 | |
| 1853 /* Expand basic block BB from GIMPLE trees to RTL. */ | |
| 1854 | |
| 1855 static basic_block | |
| 1856 expand_gimple_basic_block (basic_block bb) | |
| 1857 { | |
| 1858 gimple_stmt_iterator gsi; | |
| 1859 gimple_seq stmts; | |
| 1860 gimple stmt = NULL; | |
| 1861 rtx note, last; | |
| 1862 edge e; | |
| 1863 edge_iterator ei; | |
| 1864 void **elt; | |
| 1865 | |
| 1866 if (dump_file) | |
| 1867 fprintf (dump_file, "\n;; Generating RTL for gimple basic block %d\n", | |
| 1868 bb->index); | |
| 1869 | |
| 1870 /* Note that since we are now transitioning from GIMPLE to RTL, we | |
| 1871 cannot use the gsi_*_bb() routines because they expect the basic | |
| 1872 block to be in GIMPLE, instead of RTL. Therefore, we need to | |
| 1873 access the BB sequence directly. */ | |
| 1874 stmts = bb_seq (bb); | |
| 1875 bb->il.gimple = NULL; | |
| 1876 rtl_profile_for_bb (bb); | |
| 1877 init_rtl_bb_info (bb); | |
| 1878 bb->flags |= BB_RTL; | |
| 1879 | |
| 1880 /* Remove the RETURN_EXPR if we may fall though to the exit | |
| 1881 instead. */ | |
| 1882 gsi = gsi_last (stmts); | |
| 1883 if (!gsi_end_p (gsi) | |
| 1884 && gimple_code (gsi_stmt (gsi)) == GIMPLE_RETURN) | |
| 1885 { | |
| 1886 gimple ret_stmt = gsi_stmt (gsi); | |
| 1887 | |
| 1888 gcc_assert (single_succ_p (bb)); | |
| 1889 gcc_assert (single_succ (bb) == EXIT_BLOCK_PTR); | |
| 1890 | |
| 1891 if (bb->next_bb == EXIT_BLOCK_PTR | |
| 1892 && !gimple_return_retval (ret_stmt)) | |
| 1893 { | |
| 1894 gsi_remove (&gsi, false); | |
| 1895 single_succ_edge (bb)->flags |= EDGE_FALLTHRU; | |
| 1896 } | |
| 1897 } | |
| 1898 | |
| 1899 gsi = gsi_start (stmts); | |
| 1900 if (!gsi_end_p (gsi)) | |
| 1901 { | |
| 1902 stmt = gsi_stmt (gsi); | |
| 1903 if (gimple_code (stmt) != GIMPLE_LABEL) | |
| 1904 stmt = NULL; | |
| 1905 } | |
| 1906 | |
| 1907 elt = pointer_map_contains (lab_rtx_for_bb, bb); | |
| 1908 | |
| 1909 if (stmt || elt) | |
| 1910 { | |
| 1911 last = get_last_insn (); | |
| 1912 | |
| 1913 if (stmt) | |
| 1914 { | |
| 1915 tree stmt_tree = gimple_to_tree (stmt); | |
| 1916 expand_expr_stmt (stmt_tree); | |
| 1917 release_stmt_tree (stmt, stmt_tree); | |
| 1918 gsi_next (&gsi); | |
| 1919 } | |
| 1920 | |
| 1921 if (elt) | |
| 1922 emit_label ((rtx) *elt); | |
| 1923 | |
| 1924 /* Java emits line number notes in the top of labels. | |
| 1925 ??? Make this go away once line number notes are obsoleted. */ | |
| 1926 BB_HEAD (bb) = NEXT_INSN (last); | |
| 1927 if (NOTE_P (BB_HEAD (bb))) | |
| 1928 BB_HEAD (bb) = NEXT_INSN (BB_HEAD (bb)); | |
| 1929 note = emit_note_after (NOTE_INSN_BASIC_BLOCK, BB_HEAD (bb)); | |
| 1930 | |
| 1931 maybe_dump_rtl_for_gimple_stmt (stmt, last); | |
| 1932 } | |
| 1933 else | |
| 1934 note = BB_HEAD (bb) = emit_note (NOTE_INSN_BASIC_BLOCK); | |
| 1935 | |
| 1936 NOTE_BASIC_BLOCK (note) = bb; | |
| 1937 | |
| 1938 for (ei = ei_start (bb->succs); (e = ei_safe_edge (ei)); ) | |
| 1939 { | |
| 1940 /* Clear EDGE_EXECUTABLE. This flag is never used in the backend. */ | |
| 1941 e->flags &= ~EDGE_EXECUTABLE; | |
| 1942 | |
| 1943 /* At the moment not all abnormal edges match the RTL representation. | |
| 1944 It is safe to remove them here as find_many_sub_basic_blocks will | |
| 1945 rediscover them. In the future we should get this fixed properly. */ | |
| 1946 if (e->flags & EDGE_ABNORMAL) | |
| 1947 remove_edge (e); | |
| 1948 else | |
| 1949 ei_next (&ei); | |
| 1950 } | |
| 1951 | |
| 1952 for (; !gsi_end_p (gsi); gsi_next (&gsi)) | |
| 1953 { | |
| 1954 gimple stmt = gsi_stmt (gsi); | |
| 1955 basic_block new_bb; | |
| 1956 | |
| 1957 /* Expand this statement, then evaluate the resulting RTL and | |
| 1958 fixup the CFG accordingly. */ | |
| 1959 if (gimple_code (stmt) == GIMPLE_COND) | |
| 1960 { | |
| 1961 new_bb = expand_gimple_cond (bb, stmt); | |
| 1962 if (new_bb) | |
| 1963 return new_bb; | |
| 1964 } | |
| 1965 else | |
| 1966 { | |
| 1967 if (is_gimple_call (stmt) && gimple_call_tail_p (stmt)) | |
| 1968 { | |
| 1969 bool can_fallthru; | |
| 1970 new_bb = expand_gimple_tailcall (bb, stmt, &can_fallthru); | |
| 1971 if (new_bb) | |
| 1972 { | |
| 1973 if (can_fallthru) | |
| 1974 bb = new_bb; | |
| 1975 else | |
| 1976 return new_bb; | |
| 1977 } | |
| 1978 } | |
| 1979 else if (gimple_code (stmt) != GIMPLE_CHANGE_DYNAMIC_TYPE) | |
| 1980 { | |
| 1981 tree stmt_tree = gimple_to_tree (stmt); | |
| 1982 last = get_last_insn (); | |
| 1983 expand_expr_stmt (stmt_tree); | |
| 1984 maybe_dump_rtl_for_gimple_stmt (stmt, last); | |
| 1985 release_stmt_tree (stmt, stmt_tree); | |
| 1986 } | |
| 1987 } | |
| 1988 } | |
| 1989 | |
| 1990 /* Expand implicit goto and convert goto_locus. */ | |
| 1991 FOR_EACH_EDGE (e, ei, bb->succs) | |
| 1992 { | |
| 1993 if (e->goto_locus && e->goto_block) | |
| 1994 { | |
| 1995 set_curr_insn_source_location (e->goto_locus); | |
| 1996 set_curr_insn_block (e->goto_block); | |
| 1997 e->goto_locus = curr_insn_locator (); | |
| 1998 } | |
| 1999 e->goto_block = NULL; | |
| 2000 if ((e->flags & EDGE_FALLTHRU) && e->dest != bb->next_bb) | |
| 2001 { | |
| 2002 emit_jump (label_rtx_for_bb (e->dest)); | |
| 2003 e->flags &= ~EDGE_FALLTHRU; | |
| 2004 } | |
| 2005 } | |
| 2006 | |
| 2007 do_pending_stack_adjust (); | |
| 2008 | |
| 2009 /* Find the block tail. The last insn in the block is the insn | |
| 2010 before a barrier and/or table jump insn. */ | |
| 2011 last = get_last_insn (); | |
| 2012 if (BARRIER_P (last)) | |
| 2013 last = PREV_INSN (last); | |
| 2014 if (JUMP_TABLE_DATA_P (last)) | |
| 2015 last = PREV_INSN (PREV_INSN (last)); | |
| 2016 BB_END (bb) = last; | |
| 2017 | |
| 2018 update_bb_for_insn (bb); | |
| 2019 | |
| 2020 return bb; | |
| 2021 } | |
| 2022 | |
| 2023 | |
| 2024 /* Create a basic block for initialization code. */ | |
| 2025 | |
| 2026 static basic_block | |
| 2027 construct_init_block (void) | |
| 2028 { | |
| 2029 basic_block init_block, first_block; | |
| 2030 edge e = NULL; | |
| 2031 int flags; | |
| 2032 | |
| 2033 /* Multiple entry points not supported yet. */ | |
| 2034 gcc_assert (EDGE_COUNT (ENTRY_BLOCK_PTR->succs) == 1); | |
| 2035 init_rtl_bb_info (ENTRY_BLOCK_PTR); | |
| 2036 init_rtl_bb_info (EXIT_BLOCK_PTR); | |
| 2037 ENTRY_BLOCK_PTR->flags |= BB_RTL; | |
| 2038 EXIT_BLOCK_PTR->flags |= BB_RTL; | |
| 2039 | |
| 2040 e = EDGE_SUCC (ENTRY_BLOCK_PTR, 0); | |
| 2041 | |
| 2042 /* When entry edge points to first basic block, we don't need jump, | |
| 2043 otherwise we have to jump into proper target. */ | |
| 2044 if (e && e->dest != ENTRY_BLOCK_PTR->next_bb) | |
| 2045 { | |
| 2046 tree label = gimple_block_label (e->dest); | |
| 2047 | |
| 2048 emit_jump (label_rtx (label)); | |
| 2049 flags = 0; | |
| 2050 } | |
| 2051 else | |
| 2052 flags = EDGE_FALLTHRU; | |
| 2053 | |
| 2054 init_block = create_basic_block (NEXT_INSN (get_insns ()), | |
| 2055 get_last_insn (), | |
| 2056 ENTRY_BLOCK_PTR); | |
| 2057 init_block->frequency = ENTRY_BLOCK_PTR->frequency; | |
| 2058 init_block->count = ENTRY_BLOCK_PTR->count; | |
| 2059 if (e) | |
| 2060 { | |
| 2061 first_block = e->dest; | |
| 2062 redirect_edge_succ (e, init_block); | |
| 2063 e = make_edge (init_block, first_block, flags); | |
| 2064 } | |
| 2065 else | |
| 2066 e = make_edge (init_block, EXIT_BLOCK_PTR, EDGE_FALLTHRU); | |
| 2067 e->probability = REG_BR_PROB_BASE; | |
| 2068 e->count = ENTRY_BLOCK_PTR->count; | |
| 2069 | |
| 2070 update_bb_for_insn (init_block); | |
| 2071 return init_block; | |
| 2072 } | |
| 2073 | |
| 2074 /* For each lexical block, set BLOCK_NUMBER to the depth at which it is | |
| 2075 found in the block tree. */ | |
| 2076 | |
| 2077 static void | |
| 2078 set_block_levels (tree block, int level) | |
| 2079 { | |
| 2080 while (block) | |
| 2081 { | |
| 2082 BLOCK_NUMBER (block) = level; | |
| 2083 set_block_levels (BLOCK_SUBBLOCKS (block), level + 1); | |
| 2084 block = BLOCK_CHAIN (block); | |
| 2085 } | |
| 2086 } | |
| 2087 | |
| 2088 /* Create a block containing landing pads and similar stuff. */ | |
| 2089 | |
| 2090 static void | |
| 2091 construct_exit_block (void) | |
| 2092 { | |
| 2093 rtx head = get_last_insn (); | |
| 2094 rtx end; | |
| 2095 basic_block exit_block; | |
| 2096 edge e, e2; | |
| 2097 unsigned ix; | |
| 2098 edge_iterator ei; | |
| 2099 rtx orig_end = BB_END (EXIT_BLOCK_PTR->prev_bb); | |
| 2100 | |
| 2101 rtl_profile_for_bb (EXIT_BLOCK_PTR); | |
| 2102 | |
| 2103 /* Make sure the locus is set to the end of the function, so that | |
| 2104 epilogue line numbers and warnings are set properly. */ | |
| 2105 if (cfun->function_end_locus != UNKNOWN_LOCATION) | |
| 2106 input_location = cfun->function_end_locus; | |
| 2107 | |
| 2108 /* The following insns belong to the top scope. */ | |
| 2109 set_curr_insn_block (DECL_INITIAL (current_function_decl)); | |
| 2110 | |
| 2111 /* Generate rtl for function exit. */ | |
| 2112 expand_function_end (); | |
| 2113 | |
| 2114 end = get_last_insn (); | |
| 2115 if (head == end) | |
| 2116 return; | |
| 2117 /* While emitting the function end we could move end of the last basic block. | |
| 2118 */ | |
| 2119 BB_END (EXIT_BLOCK_PTR->prev_bb) = orig_end; | |
| 2120 while (NEXT_INSN (head) && NOTE_P (NEXT_INSN (head))) | |
| 2121 head = NEXT_INSN (head); | |
| 2122 exit_block = create_basic_block (NEXT_INSN (head), end, | |
| 2123 EXIT_BLOCK_PTR->prev_bb); | |
| 2124 exit_block->frequency = EXIT_BLOCK_PTR->frequency; | |
| 2125 exit_block->count = EXIT_BLOCK_PTR->count; | |
| 2126 | |
| 2127 ix = 0; | |
| 2128 while (ix < EDGE_COUNT (EXIT_BLOCK_PTR->preds)) | |
| 2129 { | |
| 2130 e = EDGE_PRED (EXIT_BLOCK_PTR, ix); | |
| 2131 if (!(e->flags & EDGE_ABNORMAL)) | |
| 2132 redirect_edge_succ (e, exit_block); | |
| 2133 else | |
| 2134 ix++; | |
| 2135 } | |
| 2136 | |
| 2137 e = make_edge (exit_block, EXIT_BLOCK_PTR, EDGE_FALLTHRU); | |
| 2138 e->probability = REG_BR_PROB_BASE; | |
| 2139 e->count = EXIT_BLOCK_PTR->count; | |
| 2140 FOR_EACH_EDGE (e2, ei, EXIT_BLOCK_PTR->preds) | |
| 2141 if (e2 != e) | |
| 2142 { | |
| 2143 e->count -= e2->count; | |
| 2144 exit_block->count -= e2->count; | |
| 2145 exit_block->frequency -= EDGE_FREQUENCY (e2); | |
| 2146 } | |
| 2147 if (e->count < 0) | |
| 2148 e->count = 0; | |
| 2149 if (exit_block->count < 0) | |
| 2150 exit_block->count = 0; | |
| 2151 if (exit_block->frequency < 0) | |
| 2152 exit_block->frequency = 0; | |
| 2153 update_bb_for_insn (exit_block); | |
| 2154 } | |
| 2155 | |
| 2156 /* Helper function for discover_nonconstant_array_refs. | |
| 2157 Look for ARRAY_REF nodes with non-constant indexes and mark them | |
| 2158 addressable. */ | |
| 2159 | |
| 2160 static tree | |
| 2161 discover_nonconstant_array_refs_r (tree * tp, int *walk_subtrees, | |
| 2162 void *data ATTRIBUTE_UNUSED) | |
| 2163 { | |
| 2164 tree t = *tp; | |
| 2165 | |
| 2166 if (IS_TYPE_OR_DECL_P (t)) | |
| 2167 *walk_subtrees = 0; | |
| 2168 else if (TREE_CODE (t) == ARRAY_REF || TREE_CODE (t) == ARRAY_RANGE_REF) | |
| 2169 { | |
| 2170 while (((TREE_CODE (t) == ARRAY_REF || TREE_CODE (t) == ARRAY_RANGE_REF) | |
| 2171 && is_gimple_min_invariant (TREE_OPERAND (t, 1)) | |
| 2172 && (!TREE_OPERAND (t, 2) | |
| 2173 || is_gimple_min_invariant (TREE_OPERAND (t, 2)))) | |
| 2174 || (TREE_CODE (t) == COMPONENT_REF | |
| 2175 && (!TREE_OPERAND (t,2) | |
| 2176 || is_gimple_min_invariant (TREE_OPERAND (t, 2)))) | |
| 2177 || TREE_CODE (t) == BIT_FIELD_REF | |
| 2178 || TREE_CODE (t) == REALPART_EXPR | |
| 2179 || TREE_CODE (t) == IMAGPART_EXPR | |
| 2180 || TREE_CODE (t) == VIEW_CONVERT_EXPR | |
| 2181 || CONVERT_EXPR_P (t)) | |
| 2182 t = TREE_OPERAND (t, 0); | |
| 2183 | |
| 2184 if (TREE_CODE (t) == ARRAY_REF || TREE_CODE (t) == ARRAY_RANGE_REF) | |
| 2185 { | |
| 2186 t = get_base_address (t); | |
| 2187 if (t && DECL_P (t)) | |
| 2188 TREE_ADDRESSABLE (t) = 1; | |
| 2189 } | |
| 2190 | |
| 2191 *walk_subtrees = 0; | |
| 2192 } | |
| 2193 | |
| 2194 return NULL_TREE; | |
| 2195 } | |
| 2196 | |
| 2197 /* RTL expansion is not able to compile array references with variable | |
| 2198 offsets for arrays stored in single register. Discover such | |
| 2199 expressions and mark variables as addressable to avoid this | |
| 2200 scenario. */ | |
| 2201 | |
| 2202 static void | |
| 2203 discover_nonconstant_array_refs (void) | |
| 2204 { | |
| 2205 basic_block bb; | |
| 2206 gimple_stmt_iterator gsi; | |
| 2207 | |
| 2208 FOR_EACH_BB (bb) | |
| 2209 for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi)) | |
| 2210 { | |
| 2211 gimple stmt = gsi_stmt (gsi); | |
| 2212 walk_gimple_op (stmt, discover_nonconstant_array_refs_r, NULL); | |
| 2213 } | |
| 2214 } | |
| 2215 | |
| 2216 /* This function sets crtl->args.internal_arg_pointer to a virtual | |
| 2217 register if DRAP is needed. Local register allocator will replace | |
| 2218 virtual_incoming_args_rtx with the virtual register. */ | |
| 2219 | |
| 2220 static void | |
| 2221 expand_stack_alignment (void) | |
| 2222 { | |
| 2223 rtx drap_rtx; | |
| 2224 unsigned int preferred_stack_boundary; | |
| 2225 | |
| 2226 if (! SUPPORTS_STACK_ALIGNMENT) | |
| 2227 return; | |
| 2228 | |
| 2229 if (cfun->calls_alloca | |
| 2230 || cfun->has_nonlocal_label | |
| 2231 || crtl->has_nonlocal_goto) | |
| 2232 crtl->need_drap = true; | |
| 2233 | |
| 2234 gcc_assert (crtl->stack_alignment_needed | |
| 2235 <= crtl->stack_alignment_estimated); | |
| 2236 | |
| 2237 /* Update crtl->stack_alignment_estimated and use it later to align | |
| 2238 stack. We check PREFERRED_STACK_BOUNDARY if there may be non-call | |
| 2239 exceptions since callgraph doesn't collect incoming stack alignment | |
| 2240 in this case. */ | |
| 2241 if (flag_non_call_exceptions | |
| 2242 && PREFERRED_STACK_BOUNDARY > crtl->preferred_stack_boundary) | |
| 2243 preferred_stack_boundary = PREFERRED_STACK_BOUNDARY; | |
| 2244 else | |
| 2245 preferred_stack_boundary = crtl->preferred_stack_boundary; | |
| 2246 if (preferred_stack_boundary > crtl->stack_alignment_estimated) | |
| 2247 crtl->stack_alignment_estimated = preferred_stack_boundary; | |
| 2248 if (preferred_stack_boundary > crtl->stack_alignment_needed) | |
| 2249 crtl->stack_alignment_needed = preferred_stack_boundary; | |
| 2250 | |
| 2251 crtl->stack_realign_needed | |
| 2252 = INCOMING_STACK_BOUNDARY < crtl->stack_alignment_estimated; | |
| 2253 crtl->stack_realign_tried = crtl->stack_realign_needed; | |
| 2254 | |
| 2255 crtl->stack_realign_processed = true; | |
| 2256 | |
| 2257 /* Target has to redefine TARGET_GET_DRAP_RTX to support stack | |
| 2258 alignment. */ | |
| 2259 gcc_assert (targetm.calls.get_drap_rtx != NULL); | |
| 2260 drap_rtx = targetm.calls.get_drap_rtx (); | |
| 2261 | |
| 2262 /* stack_realign_drap and drap_rtx must match. */ | |
| 2263 gcc_assert ((stack_realign_drap != 0) == (drap_rtx != NULL)); | |
| 2264 | |
| 2265 /* Do nothing if NULL is returned, which means DRAP is not needed. */ | |
| 2266 if (NULL != drap_rtx) | |
| 2267 { | |
| 2268 crtl->args.internal_arg_pointer = drap_rtx; | |
| 2269 | |
| 2270 /* Call fixup_tail_calls to clean up REG_EQUIV note if DRAP is | |
| 2271 needed. */ | |
| 2272 fixup_tail_calls (); | |
| 2273 } | |
| 2274 } | |
| 2275 | |
| 2276 /* Translate the intermediate representation contained in the CFG | |
| 2277 from GIMPLE trees to RTL. | |
| 2278 | |
| 2279 We do conversion per basic block and preserve/update the tree CFG. | |
| 2280 This implies we have to do some magic as the CFG can simultaneously | |
| 2281 consist of basic blocks containing RTL and GIMPLE trees. This can | |
| 2282 confuse the CFG hooks, so be careful to not manipulate CFG during | |
| 2283 the expansion. */ | |
| 2284 | |
| 2285 static unsigned int | |
| 2286 gimple_expand_cfg (void) | |
| 2287 { | |
| 2288 basic_block bb, init_block; | |
| 2289 sbitmap blocks; | |
| 2290 edge_iterator ei; | |
| 2291 edge e; | |
| 2292 | |
| 2293 /* Some backends want to know that we are expanding to RTL. */ | |
| 2294 currently_expanding_to_rtl = 1; | |
| 2295 | |
| 2296 rtl_profile_for_bb (ENTRY_BLOCK_PTR); | |
| 2297 | |
| 2298 insn_locators_alloc (); | |
| 2299 if (!DECL_BUILT_IN (current_function_decl)) | |
| 2300 { | |
| 2301 /* Eventually, all FEs should explicitly set function_start_locus. */ | |
| 2302 if (cfun->function_start_locus == UNKNOWN_LOCATION) | |
| 2303 set_curr_insn_source_location | |
| 2304 (DECL_SOURCE_LOCATION (current_function_decl)); | |
| 2305 else | |
| 2306 set_curr_insn_source_location (cfun->function_start_locus); | |
| 2307 } | |
| 2308 set_curr_insn_block (DECL_INITIAL (current_function_decl)); | |
| 2309 prologue_locator = curr_insn_locator (); | |
| 2310 | |
| 2311 /* Make sure first insn is a note even if we don't want linenums. | |
| 2312 This makes sure the first insn will never be deleted. | |
| 2313 Also, final expects a note to appear there. */ | |
| 2314 emit_note (NOTE_INSN_DELETED); | |
| 2315 | |
| 2316 /* Mark arrays indexed with non-constant indices with TREE_ADDRESSABLE. */ | |
| 2317 discover_nonconstant_array_refs (); | |
| 2318 | |
| 2319 targetm.expand_to_rtl_hook (); | |
| 2320 crtl->stack_alignment_needed = STACK_BOUNDARY; | |
| 2321 crtl->max_used_stack_slot_alignment = STACK_BOUNDARY; | |
| 2322 crtl->stack_alignment_estimated = STACK_BOUNDARY; | |
| 2323 crtl->preferred_stack_boundary = STACK_BOUNDARY; | |
| 2324 cfun->cfg->max_jumptable_ents = 0; | |
| 2325 | |
| 2326 | |
| 2327 /* Expand the variables recorded during gimple lowering. */ | |
| 2328 expand_used_vars (); | |
| 2329 | |
| 2330 /* Honor stack protection warnings. */ | |
| 2331 if (warn_stack_protect) | |
| 2332 { | |
| 2333 if (cfun->calls_alloca) | |
| 2334 warning (OPT_Wstack_protector, | |
| 2335 "not protecting local variables: variable length buffer"); | |
| 2336 if (has_short_buffer && !crtl->stack_protect_guard) | |
| 2337 warning (OPT_Wstack_protector, | |
| 2338 "not protecting function: no buffer at least %d bytes long", | |
| 2339 (int) PARAM_VALUE (PARAM_SSP_BUFFER_SIZE)); | |
| 2340 } | |
| 2341 | |
| 2342 /* Set up parameters and prepare for return, for the function. */ | |
| 2343 expand_function_start (current_function_decl); | |
| 2344 | |
| 2345 /* If this function is `main', emit a call to `__main' | |
| 2346 to run global initializers, etc. */ | |
| 2347 if (DECL_NAME (current_function_decl) | |
| 2348 && MAIN_NAME_P (DECL_NAME (current_function_decl)) | |
| 2349 && DECL_FILE_SCOPE_P (current_function_decl)) | |
| 2350 expand_main_function (); | |
| 2351 | |
| 2352 /* Initialize the stack_protect_guard field. This must happen after the | |
| 2353 call to __main (if any) so that the external decl is initialized. */ | |
| 2354 if (crtl->stack_protect_guard) | |
| 2355 stack_protect_prologue (); | |
| 2356 | |
| 2357 /* Update stack boundary if needed. */ | |
| 2358 if (SUPPORTS_STACK_ALIGNMENT) | |
| 2359 { | |
| 2360 /* Call update_stack_boundary here to update incoming stack | |
| 2361 boundary before TARGET_FUNCTION_OK_FOR_SIBCALL is called. | |
| 2362 TARGET_FUNCTION_OK_FOR_SIBCALL needs to know the accurate | |
| 2363 incoming stack alignment to check if it is OK to perform | |
| 2364 sibcall optimization since sibcall optimization will only | |
| 2365 align the outgoing stack to incoming stack boundary. */ | |
| 2366 if (targetm.calls.update_stack_boundary) | |
| 2367 targetm.calls.update_stack_boundary (); | |
| 2368 | |
| 2369 /* The incoming stack frame has to be aligned at least at | |
| 2370 parm_stack_boundary. */ | |
| 2371 gcc_assert (crtl->parm_stack_boundary <= INCOMING_STACK_BOUNDARY); | |
| 2372 } | |
| 2373 | |
| 2374 /* Register rtl specific functions for cfg. */ | |
| 2375 rtl_register_cfg_hooks (); | |
| 2376 | |
| 2377 init_block = construct_init_block (); | |
| 2378 | |
| 2379 /* Clear EDGE_EXECUTABLE on the entry edge(s). It is cleaned from the | |
| 2380 remaining edges in expand_gimple_basic_block. */ | |
| 2381 FOR_EACH_EDGE (e, ei, ENTRY_BLOCK_PTR->succs) | |
| 2382 e->flags &= ~EDGE_EXECUTABLE; | |
| 2383 | |
| 2384 lab_rtx_for_bb = pointer_map_create (); | |
| 2385 FOR_BB_BETWEEN (bb, init_block->next_bb, EXIT_BLOCK_PTR, next_bb) | |
| 2386 bb = expand_gimple_basic_block (bb); | |
| 2387 | |
| 2388 /* Expansion is used by optimization passes too, set maybe_hot_insn_p | |
| 2389 conservatively to true until they are all profile aware. */ | |
| 2390 pointer_map_destroy (lab_rtx_for_bb); | |
| 2391 free_histograms (); | |
| 2392 | |
| 2393 construct_exit_block (); | |
| 2394 set_curr_insn_block (DECL_INITIAL (current_function_decl)); | |
| 2395 insn_locators_finalize (); | |
| 2396 | |
| 2397 /* We're done expanding trees to RTL. */ | |
| 2398 currently_expanding_to_rtl = 0; | |
| 2399 | |
| 2400 /* Convert tree EH labels to RTL EH labels and zap the tree EH table. */ | |
| 2401 convert_from_eh_region_ranges (); | |
| 2402 set_eh_throw_stmt_table (cfun, NULL); | |
| 2403 | |
| 2404 rebuild_jump_labels (get_insns ()); | |
| 2405 find_exception_handler_labels (); | |
| 2406 | |
| 2407 blocks = sbitmap_alloc (last_basic_block); | |
| 2408 sbitmap_ones (blocks); | |
| 2409 find_many_sub_basic_blocks (blocks); | |
| 2410 purge_all_dead_edges (); | |
| 2411 sbitmap_free (blocks); | |
| 2412 | |
| 2413 compact_blocks (); | |
| 2414 | |
| 2415 expand_stack_alignment (); | |
| 2416 | |
| 2417 #ifdef ENABLE_CHECKING | |
| 2418 verify_flow_info (); | |
| 2419 #endif | |
| 2420 | |
| 2421 /* There's no need to defer outputting this function any more; we | |
| 2422 know we want to output it. */ | |
| 2423 DECL_DEFER_OUTPUT (current_function_decl) = 0; | |
| 2424 | |
| 2425 /* Now that we're done expanding trees to RTL, we shouldn't have any | |
| 2426 more CONCATs anywhere. */ | |
| 2427 generating_concat_p = 0; | |
| 2428 | |
| 2429 if (dump_file) | |
| 2430 { | |
| 2431 fprintf (dump_file, | |
| 2432 "\n\n;;\n;; Full RTL generated for this function:\n;;\n"); | |
| 2433 /* And the pass manager will dump RTL for us. */ | |
| 2434 } | |
| 2435 | |
| 2436 /* If we're emitting a nested function, make sure its parent gets | |
| 2437 emitted as well. Doing otherwise confuses debug info. */ | |
| 2438 { | |
| 2439 tree parent; | |
| 2440 for (parent = DECL_CONTEXT (current_function_decl); | |
| 2441 parent != NULL_TREE; | |
| 2442 parent = get_containing_scope (parent)) | |
| 2443 if (TREE_CODE (parent) == FUNCTION_DECL) | |
| 2444 TREE_SYMBOL_REFERENCED (DECL_ASSEMBLER_NAME (parent)) = 1; | |
| 2445 } | |
| 2446 | |
| 2447 /* We are now committed to emitting code for this function. Do any | |
| 2448 preparation, such as emitting abstract debug info for the inline | |
| 2449 before it gets mangled by optimization. */ | |
| 2450 if (cgraph_function_possibly_inlined_p (current_function_decl)) | |
| 2451 (*debug_hooks->outlining_inline_function) (current_function_decl); | |
| 2452 | |
| 2453 TREE_ASM_WRITTEN (current_function_decl) = 1; | |
| 2454 | |
| 2455 /* After expanding, the return labels are no longer needed. */ | |
| 2456 return_label = NULL; | |
| 2457 naked_return_label = NULL; | |
| 2458 /* Tag the blocks with a depth number so that change_scope can find | |
| 2459 the common parent easily. */ | |
| 2460 set_block_levels (DECL_INITIAL (cfun->decl), 0); | |
| 2461 default_rtl_profile (); | |
| 2462 return 0; | |
| 2463 } | |
| 2464 | |
| 2465 struct rtl_opt_pass pass_expand = | |
| 2466 { | |
| 2467 { | |
| 2468 RTL_PASS, | |
| 2469 "expand", /* name */ | |
| 2470 NULL, /* gate */ | |
| 2471 gimple_expand_cfg, /* execute */ | |
| 2472 NULL, /* sub */ | |
| 2473 NULL, /* next */ | |
| 2474 0, /* static_pass_number */ | |
| 2475 TV_EXPAND, /* tv_id */ | |
| 2476 /* ??? If TER is enabled, we actually receive GENERIC. */ | |
| 2477 PROP_gimple_leh | PROP_cfg, /* properties_required */ | |
| 2478 PROP_rtl, /* properties_provided */ | |
| 2479 PROP_trees, /* properties_destroyed */ | |
| 2480 0, /* todo_flags_start */ | |
| 2481 TODO_dump_func, /* todo_flags_finish */ | |
| 2482 } | |
| 2483 }; |