1 /* 2 * Copyright (c) 2005, 2018, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "ci/bcEscapeAnalyzer.hpp" 27 #include "compiler/compileLog.hpp" 28 #include "gc/shared/barrierSet.hpp" 29 #include "gc/shared/c2/barrierSetC2.hpp" 30 #include "libadt/vectset.hpp" 31 #include "memory/allocation.hpp" 32 #include "memory/resourceArea.hpp" 33 #include "opto/c2compiler.hpp" 34 #include "opto/arraycopynode.hpp" 35 #include "opto/callnode.hpp" 36 #include "opto/cfgnode.hpp" 37 #include "opto/compile.hpp" 38 #include "opto/escape.hpp" 39 #include "opto/phaseX.hpp" 40 #include "opto/movenode.hpp" 41 #include "opto/rootnode.hpp" 42 #include "utilities/macros.hpp" 43 44 ConnectionGraph::ConnectionGraph(Compile * C, PhaseIterGVN *igvn) : 45 _nodes(C->comp_arena(), C->unique(), C->unique(), NULL), 46 _in_worklist(C->comp_arena()), 47 _next_pidx(0), 48 _collecting(true), 49 _verify(false), 50 _compile(C), 51 _igvn(igvn), 52 _node_map(C->comp_arena()) { 53 // Add unknown java object. 54 add_java_object(C->top(), PointsToNode::GlobalEscape); 55 phantom_obj = ptnode_adr(C->top()->_idx)->as_JavaObject(); 56 // Add ConP(#NULL) and ConN(#NULL) nodes. 57 Node* oop_null = igvn->zerocon(T_OBJECT); 58 assert(oop_null->_idx < nodes_size(), "should be created already"); 59 add_java_object(oop_null, PointsToNode::NoEscape); 60 null_obj = ptnode_adr(oop_null->_idx)->as_JavaObject(); 61 if (UseCompressedOops) { 62 Node* noop_null = igvn->zerocon(T_NARROWOOP); 63 assert(noop_null->_idx < nodes_size(), "should be created already"); 64 map_ideal_node(noop_null, null_obj); 65 } 66 _pcmp_neq = NULL; // Should be initialized 67 _pcmp_eq = NULL; 68 } 69 70 bool ConnectionGraph::has_candidates(Compile *C) { 71 // EA brings benefits only when the code has allocations and/or locks which 72 // are represented by ideal Macro nodes. 73 int cnt = C->macro_count(); 74 for (int i = 0; i < cnt; i++) { 75 Node *n = C->macro_node(i); 76 if (n->is_Allocate()) 77 return true; 78 if (n->is_Lock()) { 79 Node* obj = n->as_Lock()->obj_node()->uncast(); 80 if (!(obj->is_Parm() || obj->is_Con())) 81 return true; 82 } 83 if (n->is_CallStaticJava() && 84 n->as_CallStaticJava()->is_boxing_method()) { 85 return true; 86 } 87 } 88 return false; 89 } 90 91 void ConnectionGraph::do_analysis(Compile *C, PhaseIterGVN *igvn) { 92 Compile::TracePhase tp("escapeAnalysis", &Phase::timers[Phase::_t_escapeAnalysis]); 93 ResourceMark rm; 94 95 // Add ConP#NULL and ConN#NULL nodes before ConnectionGraph construction 96 // to create space for them in ConnectionGraph::_nodes[]. 97 Node* oop_null = igvn->zerocon(T_OBJECT); 98 Node* noop_null = igvn->zerocon(T_NARROWOOP); 99 ConnectionGraph* congraph = new(C->comp_arena()) ConnectionGraph(C, igvn); 100 // Perform escape analysis 101 if (congraph->compute_escape()) { 102 // There are non escaping objects. 103 C->set_congraph(congraph); 104 } 105 // Cleanup. 106 if (oop_null->outcnt() == 0) 107 igvn->hash_delete(oop_null); 108 if (noop_null->outcnt() == 0) 109 igvn->hash_delete(noop_null); 110 } 111 112 bool ConnectionGraph::compute_escape() { 113 Compile* C = _compile; 114 PhaseGVN* igvn = _igvn; 115 116 // Worklists used by EA. 117 Unique_Node_List delayed_worklist; 118 GrowableArray<Node*> alloc_worklist; 119 GrowableArray<Node*> ptr_cmp_worklist; 120 GrowableArray<Node*> storestore_worklist; 121 GrowableArray<ArrayCopyNode*> arraycopy_worklist; 122 GrowableArray<PointsToNode*> ptnodes_worklist; 123 GrowableArray<JavaObjectNode*> java_objects_worklist; 124 GrowableArray<JavaObjectNode*> non_escaped_worklist; 125 GrowableArray<FieldNode*> oop_fields_worklist; 126 DEBUG_ONLY( GrowableArray<Node*> addp_worklist; ) 127 128 { Compile::TracePhase tp("connectionGraph", &Phase::timers[Phase::_t_connectionGraph]); 129 130 // 1. Populate Connection Graph (CG) with PointsTo nodes. 131 ideal_nodes.map(C->live_nodes(), NULL); // preallocate space 132 // Initialize worklist 133 if (C->root() != NULL) { 134 ideal_nodes.push(C->root()); 135 } 136 // Processed ideal nodes are unique on ideal_nodes list 137 // but several ideal nodes are mapped to the phantom_obj. 138 // To avoid duplicated entries on the following worklists 139 // add the phantom_obj only once to them. 140 ptnodes_worklist.append(phantom_obj); 141 java_objects_worklist.append(phantom_obj); 142 for( uint next = 0; next < ideal_nodes.size(); ++next ) { 143 Node* n = ideal_nodes.at(next); 144 // Create PointsTo nodes and add them to Connection Graph. Called 145 // only once per ideal node since ideal_nodes is Unique_Node list. 146 add_node_to_connection_graph(n, &delayed_worklist); 147 PointsToNode* ptn = ptnode_adr(n->_idx); 148 if (ptn != NULL && ptn != phantom_obj) { 149 ptnodes_worklist.append(ptn); 150 if (ptn->is_JavaObject()) { 151 java_objects_worklist.append(ptn->as_JavaObject()); 152 if ((n->is_Allocate() || n->is_CallStaticJava()) && 153 (ptn->escape_state() < PointsToNode::GlobalEscape)) { 154 // Only allocations and java static calls results are interesting. 155 non_escaped_worklist.append(ptn->as_JavaObject()); 156 } 157 } else if (ptn->is_Field() && ptn->as_Field()->is_oop()) { 158 oop_fields_worklist.append(ptn->as_Field()); 159 } 160 } 161 if (n->is_MergeMem()) { 162 // Collect all MergeMem nodes to add memory slices for 163 // scalar replaceable objects in split_unique_types(). 164 _mergemem_worklist.append(n->as_MergeMem()); 165 } else if (OptimizePtrCompare && n->is_Cmp() && 166 (n->Opcode() == Op_CmpP || n->Opcode() == Op_CmpN)) { 167 // Collect compare pointers nodes. 168 ptr_cmp_worklist.append(n); 169 } else if (n->is_MemBarStoreStore()) { 170 // Collect all MemBarStoreStore nodes so that depending on the 171 // escape status of the associated Allocate node some of them 172 // may be eliminated. 173 storestore_worklist.append(n); 174 } else if (n->is_MemBar() && (n->Opcode() == Op_MemBarRelease) && 175 (n->req() > MemBarNode::Precedent)) { 176 record_for_optimizer(n); 177 #ifdef ASSERT 178 } else if (n->is_AddP()) { 179 // Collect address nodes for graph verification. 180 addp_worklist.append(n); 181 #endif 182 } else if (n->is_ArrayCopy()) { 183 // Keep a list of ArrayCopy nodes so if one of its input is non 184 // escaping, we can record a unique type 185 arraycopy_worklist.append(n->as_ArrayCopy()); 186 } 187 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { 188 Node* m = n->fast_out(i); // Get user 189 ideal_nodes.push(m); 190 } 191 } 192 if (non_escaped_worklist.length() == 0) { 193 _collecting = false; 194 return false; // Nothing to do. 195 } 196 // Add final simple edges to graph. 197 while(delayed_worklist.size() > 0) { 198 Node* n = delayed_worklist.pop(); 199 add_final_edges(n); 200 } 201 int ptnodes_length = ptnodes_worklist.length(); 202 203 #ifdef ASSERT 204 if (VerifyConnectionGraph) { 205 // Verify that no new simple edges could be created and all 206 // local vars has edges. 207 _verify = true; 208 for (int next = 0; next < ptnodes_length; ++next) { 209 PointsToNode* ptn = ptnodes_worklist.at(next); 210 add_final_edges(ptn->ideal_node()); 211 if (ptn->is_LocalVar() && ptn->edge_count() == 0) { 212 ptn->dump(); 213 assert(ptn->as_LocalVar()->edge_count() > 0, "sanity"); 214 } 215 } 216 _verify = false; 217 } 218 #endif 219 // Bytecode analyzer BCEscapeAnalyzer, used for Call nodes 220 // processing, calls to CI to resolve symbols (types, fields, methods) 221 // referenced in bytecode. During symbol resolution VM may throw 222 // an exception which CI cleans and converts to compilation failure. 223 if (C->failing()) return false; 224 225 // 2. Finish Graph construction by propagating references to all 226 // java objects through graph. 227 if (!complete_connection_graph(ptnodes_worklist, non_escaped_worklist, 228 java_objects_worklist, oop_fields_worklist)) { 229 // All objects escaped or hit time or iterations limits. 230 _collecting = false; 231 return false; 232 } 233 234 // 3. Adjust scalar_replaceable state of nonescaping objects and push 235 // scalar replaceable allocations on alloc_worklist for processing 236 // in split_unique_types(). 237 int non_escaped_length = non_escaped_worklist.length(); 238 for (int next = 0; next < non_escaped_length; next++) { 239 JavaObjectNode* ptn = non_escaped_worklist.at(next); 240 bool noescape = (ptn->escape_state() == PointsToNode::NoEscape); 241 Node* n = ptn->ideal_node(); 242 if (n->is_Allocate()) { 243 n->as_Allocate()->_is_non_escaping = noescape; 244 } 245 if (n->is_CallStaticJava()) { 246 n->as_CallStaticJava()->_is_non_escaping = noescape; 247 } 248 if (noescape && ptn->scalar_replaceable()) { 249 adjust_scalar_replaceable_state(ptn); 250 if (ptn->scalar_replaceable()) { 251 alloc_worklist.append(ptn->ideal_node()); 252 } 253 } 254 } 255 256 #ifdef ASSERT 257 if (VerifyConnectionGraph) { 258 // Verify that graph is complete - no new edges could be added or needed. 259 verify_connection_graph(ptnodes_worklist, non_escaped_worklist, 260 java_objects_worklist, addp_worklist); 261 } 262 assert(C->unique() == nodes_size(), "no new ideal nodes should be added during ConnectionGraph build"); 263 assert(null_obj->escape_state() == PointsToNode::NoEscape && 264 null_obj->edge_count() == 0 && 265 !null_obj->arraycopy_src() && 266 !null_obj->arraycopy_dst(), "sanity"); 267 #endif 268 269 _collecting = false; 270 271 } // TracePhase t3("connectionGraph") 272 273 // 4. Optimize ideal graph based on EA information. 274 bool has_non_escaping_obj = (non_escaped_worklist.length() > 0); 275 if (has_non_escaping_obj) { 276 optimize_ideal_graph(ptr_cmp_worklist, storestore_worklist); 277 } 278 279 #ifndef PRODUCT 280 if (PrintEscapeAnalysis) { 281 dump(ptnodes_worklist); // Dump ConnectionGraph 282 } 283 #endif 284 285 bool has_scalar_replaceable_candidates = (alloc_worklist.length() > 0); 286 #ifdef ASSERT 287 if (VerifyConnectionGraph) { 288 int alloc_length = alloc_worklist.length(); 289 for (int next = 0; next < alloc_length; ++next) { 290 Node* n = alloc_worklist.at(next); 291 PointsToNode* ptn = ptnode_adr(n->_idx); 292 assert(ptn->escape_state() == PointsToNode::NoEscape && ptn->scalar_replaceable(), "sanity"); 293 } 294 } 295 #endif 296 297 // 5. Separate memory graph for scalar replaceable allcations. 298 if (has_scalar_replaceable_candidates && 299 C->AliasLevel() >= 3 && EliminateAllocations) { 300 // Now use the escape information to create unique types for 301 // scalar replaceable objects. 302 split_unique_types(alloc_worklist, arraycopy_worklist); 303 if (C->failing()) return false; 304 C->print_method(PHASE_AFTER_EA, 2); 305 306 #ifdef ASSERT 307 } else if (Verbose && (PrintEscapeAnalysis || PrintEliminateAllocations)) { 308 tty->print("=== No allocations eliminated for "); 309 C->method()->print_short_name(); 310 if(!EliminateAllocations) { 311 tty->print(" since EliminateAllocations is off ==="); 312 } else if(!has_scalar_replaceable_candidates) { 313 tty->print(" since there are no scalar replaceable candidates ==="); 314 } else if(C->AliasLevel() < 3) { 315 tty->print(" since AliasLevel < 3 ==="); 316 } 317 tty->cr(); 318 #endif 319 } 320 return has_non_escaping_obj; 321 } 322 323 // Utility function for nodes that load an object 324 void ConnectionGraph::add_objload_to_connection_graph(Node *n, Unique_Node_List *delayed_worklist) { 325 // Using isa_ptr() instead of isa_oopptr() for LoadP and Phi because 326 // ThreadLocal has RawPtr type. 327 const Type* t = _igvn->type(n); 328 if (t->make_ptr() != NULL) { 329 Node* adr = n->in(MemNode::Address); 330 #ifdef ASSERT 331 if (!adr->is_AddP()) { 332 assert(_igvn->type(adr)->isa_rawptr(), "sanity"); 333 } else { 334 assert((ptnode_adr(adr->_idx) == NULL || 335 ptnode_adr(adr->_idx)->as_Field()->is_oop()), "sanity"); 336 } 337 #endif 338 add_local_var_and_edge(n, PointsToNode::NoEscape, 339 adr, delayed_worklist); 340 } 341 } 342 343 // Populate Connection Graph with PointsTo nodes and create simple 344 // connection graph edges. 345 void ConnectionGraph::add_node_to_connection_graph(Node *n, Unique_Node_List *delayed_worklist) { 346 assert(!_verify, "this method should not be called for verification"); 347 PhaseGVN* igvn = _igvn; 348 uint n_idx = n->_idx; 349 PointsToNode* n_ptn = ptnode_adr(n_idx); 350 if (n_ptn != NULL) 351 return; // No need to redefine PointsTo node during first iteration. 352 353 if (n->is_Call()) { 354 // Arguments to allocation and locking don't escape. 355 if (n->is_AbstractLock()) { 356 // Put Lock and Unlock nodes on IGVN worklist to process them during 357 // first IGVN optimization when escape information is still available. 358 record_for_optimizer(n); 359 } else if (n->is_Allocate()) { 360 add_call_node(n->as_Call()); 361 record_for_optimizer(n); 362 } else { 363 if (n->is_CallStaticJava()) { 364 const char* name = n->as_CallStaticJava()->_name; 365 if (name != NULL && strcmp(name, "uncommon_trap") == 0) 366 return; // Skip uncommon traps 367 } 368 // Don't mark as processed since call's arguments have to be processed. 369 delayed_worklist->push(n); 370 // Check if a call returns an object. 371 if ((n->as_Call()->returns_pointer() && 372 n->as_Call()->proj_out_or_null(TypeFunc::Parms) != NULL) || 373 (n->is_CallStaticJava() && 374 n->as_CallStaticJava()->is_boxing_method())) { 375 add_call_node(n->as_Call()); 376 } 377 } 378 return; 379 } 380 // Put this check here to process call arguments since some call nodes 381 // point to phantom_obj. 382 if (n_ptn == phantom_obj || n_ptn == null_obj) 383 return; // Skip predefined nodes. 384 385 int opcode = n->Opcode(); 386 bool gc_handled = BarrierSet::barrier_set()->barrier_set_c2()->escape_add_to_con_graph(this, igvn, delayed_worklist, n, opcode); 387 if (gc_handled) { 388 return; // Ignore node if already handled by GC. 389 } 390 switch (opcode) { 391 case Op_AddP: { 392 Node* base = get_addp_base(n); 393 PointsToNode* ptn_base = ptnode_adr(base->_idx); 394 // Field nodes are created for all field types. They are used in 395 // adjust_scalar_replaceable_state() and split_unique_types(). 396 // Note, non-oop fields will have only base edges in Connection 397 // Graph because such fields are not used for oop loads and stores. 398 int offset = address_offset(n, igvn); 399 add_field(n, PointsToNode::NoEscape, offset); 400 if (ptn_base == NULL) { 401 delayed_worklist->push(n); // Process it later. 402 } else { 403 n_ptn = ptnode_adr(n_idx); 404 add_base(n_ptn->as_Field(), ptn_base); 405 } 406 break; 407 } 408 case Op_CastX2P: { 409 map_ideal_node(n, phantom_obj); 410 break; 411 } 412 case Op_CastPP: 413 case Op_CheckCastPP: 414 case Op_EncodeP: 415 case Op_DecodeN: 416 case Op_EncodePKlass: 417 case Op_DecodeNKlass: { 418 add_local_var_and_edge(n, PointsToNode::NoEscape, 419 n->in(1), delayed_worklist); 420 break; 421 } 422 case Op_CMoveP: { 423 add_local_var(n, PointsToNode::NoEscape); 424 // Do not add edges during first iteration because some could be 425 // not defined yet. 426 delayed_worklist->push(n); 427 break; 428 } 429 case Op_ConP: 430 case Op_ConN: 431 case Op_ConNKlass: { 432 // assume all oop constants globally escape except for null 433 PointsToNode::EscapeState es; 434 const Type* t = igvn->type(n); 435 if (t == TypePtr::NULL_PTR || t == TypeNarrowOop::NULL_PTR) { 436 es = PointsToNode::NoEscape; 437 } else { 438 es = PointsToNode::GlobalEscape; 439 } 440 add_java_object(n, es); 441 break; 442 } 443 case Op_CreateEx: { 444 // assume that all exception objects globally escape 445 map_ideal_node(n, phantom_obj); 446 break; 447 } 448 case Op_LoadKlass: 449 case Op_LoadNKlass: { 450 // Unknown class is loaded 451 map_ideal_node(n, phantom_obj); 452 break; 453 } 454 case Op_LoadP: 455 case Op_LoadN: 456 case Op_LoadPLocked: { 457 add_objload_to_connection_graph(n, delayed_worklist); 458 break; 459 } 460 case Op_Parm: { 461 map_ideal_node(n, phantom_obj); 462 break; 463 } 464 case Op_PartialSubtypeCheck: { 465 // Produces Null or notNull and is used in only in CmpP so 466 // phantom_obj could be used. 467 map_ideal_node(n, phantom_obj); // Result is unknown 468 break; 469 } 470 case Op_Phi: { 471 // Using isa_ptr() instead of isa_oopptr() for LoadP and Phi because 472 // ThreadLocal has RawPtr type. 473 const Type* t = n->as_Phi()->type(); 474 if (t->make_ptr() != NULL) { 475 add_local_var(n, PointsToNode::NoEscape); 476 // Do not add edges during first iteration because some could be 477 // not defined yet. 478 delayed_worklist->push(n); 479 } 480 break; 481 } 482 case Op_Proj: { 483 // we are only interested in the oop result projection from a call 484 if (n->as_Proj()->_con == TypeFunc::Parms && n->in(0)->is_Call() && 485 n->in(0)->as_Call()->returns_pointer()) { 486 add_local_var_and_edge(n, PointsToNode::NoEscape, 487 n->in(0), delayed_worklist); 488 } 489 break; 490 } 491 case Op_Rethrow: // Exception object escapes 492 case Op_Return: { 493 if (n->req() > TypeFunc::Parms && 494 igvn->type(n->in(TypeFunc::Parms))->isa_oopptr()) { 495 // Treat Return value as LocalVar with GlobalEscape escape state. 496 add_local_var_and_edge(n, PointsToNode::GlobalEscape, 497 n->in(TypeFunc::Parms), delayed_worklist); 498 } 499 break; 500 } 501 case Op_CompareAndExchangeP: 502 case Op_CompareAndExchangeN: 503 case Op_GetAndSetP: 504 case Op_GetAndSetN: { 505 add_objload_to_connection_graph(n, delayed_worklist); 506 // fallthrough 507 } 508 case Op_StoreP: 509 case Op_StoreN: 510 case Op_StoreNKlass: 511 case Op_StorePConditional: 512 case Op_WeakCompareAndSwapP: 513 case Op_WeakCompareAndSwapN: 514 case Op_CompareAndSwapP: 515 case Op_CompareAndSwapN: { 516 add_to_congraph_unsafe_access(n, opcode, delayed_worklist); 517 break; 518 } 519 case Op_AryEq: 520 case Op_HasNegatives: 521 case Op_StrComp: 522 case Op_StrEquals: 523 case Op_StrIndexOf: 524 case Op_StrIndexOfChar: 525 case Op_StrInflatedCopy: 526 case Op_StrCompressedCopy: 527 case Op_EncodeISOArray: { 528 add_local_var(n, PointsToNode::ArgEscape); 529 delayed_worklist->push(n); // Process it later. 530 break; 531 } 532 case Op_ThreadLocal: { 533 add_java_object(n, PointsToNode::ArgEscape); 534 break; 535 } 536 default: 537 ; // Do nothing for nodes not related to EA. 538 } 539 return; 540 } 541 542 #ifdef ASSERT 543 #define ELSE_FAIL(name) \ 544 /* Should not be called for not pointer type. */ \ 545 n->dump(1); \ 546 assert(false, name); \ 547 break; 548 #else 549 #define ELSE_FAIL(name) \ 550 break; 551 #endif 552 553 // Add final simple edges to graph. 554 void ConnectionGraph::add_final_edges(Node *n) { 555 PointsToNode* n_ptn = ptnode_adr(n->_idx); 556 #ifdef ASSERT 557 if (_verify && n_ptn->is_JavaObject()) 558 return; // This method does not change graph for JavaObject. 559 #endif 560 561 if (n->is_Call()) { 562 process_call_arguments(n->as_Call()); 563 return; 564 } 565 assert(n->is_Store() || n->is_LoadStore() || 566 (n_ptn != NULL) && (n_ptn->ideal_node() != NULL), 567 "node should be registered already"); 568 int opcode = n->Opcode(); 569 bool gc_handled = BarrierSet::barrier_set()->barrier_set_c2()->escape_add_final_edges(this, _igvn, n, opcode); 570 if (gc_handled) { 571 return; // Ignore node if already handled by GC. 572 } 573 switch (opcode) { 574 case Op_AddP: { 575 Node* base = get_addp_base(n); 576 PointsToNode* ptn_base = ptnode_adr(base->_idx); 577 assert(ptn_base != NULL, "field's base should be registered"); 578 add_base(n_ptn->as_Field(), ptn_base); 579 break; 580 } 581 case Op_CastPP: 582 case Op_CheckCastPP: 583 case Op_EncodeP: 584 case Op_DecodeN: 585 case Op_EncodePKlass: 586 case Op_DecodeNKlass: { 587 add_local_var_and_edge(n, PointsToNode::NoEscape, 588 n->in(1), NULL); 589 break; 590 } 591 case Op_CMoveP: { 592 for (uint i = CMoveNode::IfFalse; i < n->req(); i++) { 593 Node* in = n->in(i); 594 if (in == NULL) 595 continue; // ignore NULL 596 Node* uncast_in = in->uncast(); 597 if (uncast_in->is_top() || uncast_in == n) 598 continue; // ignore top or inputs which go back this node 599 PointsToNode* ptn = ptnode_adr(in->_idx); 600 assert(ptn != NULL, "node should be registered"); 601 add_edge(n_ptn, ptn); 602 } 603 break; 604 } 605 case Op_LoadP: 606 case Op_LoadN: 607 case Op_LoadPLocked: { 608 // Using isa_ptr() instead of isa_oopptr() for LoadP and Phi because 609 // ThreadLocal has RawPtr type. 610 const Type* t = _igvn->type(n); 611 if (t->make_ptr() != NULL) { 612 Node* adr = n->in(MemNode::Address); 613 add_local_var_and_edge(n, PointsToNode::NoEscape, adr, NULL); 614 break; 615 } 616 ELSE_FAIL("Op_LoadP"); 617 } 618 case Op_Phi: { 619 // Using isa_ptr() instead of isa_oopptr() for LoadP and Phi because 620 // ThreadLocal has RawPtr type. 621 const Type* t = n->as_Phi()->type(); 622 if (t->make_ptr() != NULL) { 623 for (uint i = 1; i < n->req(); i++) { 624 Node* in = n->in(i); 625 if (in == NULL) 626 continue; // ignore NULL 627 Node* uncast_in = in->uncast(); 628 if (uncast_in->is_top() || uncast_in == n) 629 continue; // ignore top or inputs which go back this node 630 PointsToNode* ptn = ptnode_adr(in->_idx); 631 assert(ptn != NULL, "node should be registered"); 632 add_edge(n_ptn, ptn); 633 } 634 break; 635 } 636 ELSE_FAIL("Op_Phi"); 637 } 638 case Op_Proj: { 639 // we are only interested in the oop result projection from a call 640 if (n->as_Proj()->_con == TypeFunc::Parms && n->in(0)->is_Call() && 641 n->in(0)->as_Call()->returns_pointer()) { 642 add_local_var_and_edge(n, PointsToNode::NoEscape, n->in(0), NULL); 643 break; 644 } 645 ELSE_FAIL("Op_Proj"); 646 } 647 case Op_Rethrow: // Exception object escapes 648 case Op_Return: { 649 if (n->req() > TypeFunc::Parms && 650 _igvn->type(n->in(TypeFunc::Parms))->isa_oopptr()) { 651 // Treat Return value as LocalVar with GlobalEscape escape state. 652 add_local_var_and_edge(n, PointsToNode::GlobalEscape, 653 n->in(TypeFunc::Parms), NULL); 654 break; 655 } 656 ELSE_FAIL("Op_Return"); 657 } 658 case Op_StoreP: 659 case Op_StoreN: 660 case Op_StoreNKlass: 661 case Op_StorePConditional: 662 case Op_CompareAndExchangeP: 663 case Op_CompareAndExchangeN: 664 case Op_CompareAndSwapP: 665 case Op_CompareAndSwapN: 666 case Op_WeakCompareAndSwapP: 667 case Op_WeakCompareAndSwapN: 668 case Op_GetAndSetP: 669 case Op_GetAndSetN: { 670 if (add_final_edges_unsafe_access(n, opcode)) { 671 break; 672 } 673 ELSE_FAIL("Op_StoreP"); 674 } 675 case Op_AryEq: 676 case Op_HasNegatives: 677 case Op_StrComp: 678 case Op_StrEquals: 679 case Op_StrIndexOf: 680 case Op_StrIndexOfChar: 681 case Op_StrInflatedCopy: 682 case Op_StrCompressedCopy: 683 case Op_EncodeISOArray: { 684 // char[]/byte[] arrays passed to string intrinsic do not escape but 685 // they are not scalar replaceable. Adjust escape state for them. 686 // Start from in(2) edge since in(1) is memory edge. 687 for (uint i = 2; i < n->req(); i++) { 688 Node* adr = n->in(i); 689 const Type* at = _igvn->type(adr); 690 if (!adr->is_top() && at->isa_ptr()) { 691 assert(at == Type::TOP || at == TypePtr::NULL_PTR || 692 at->isa_ptr() != NULL, "expecting a pointer"); 693 if (adr->is_AddP()) { 694 adr = get_addp_base(adr); 695 } 696 PointsToNode* ptn = ptnode_adr(adr->_idx); 697 assert(ptn != NULL, "node should be registered"); 698 add_edge(n_ptn, ptn); 699 } 700 } 701 break; 702 } 703 default: { 704 // This method should be called only for EA specific nodes which may 705 // miss some edges when they were created. 706 #ifdef ASSERT 707 n->dump(1); 708 #endif 709 guarantee(false, "unknown node"); 710 } 711 } 712 return; 713 } 714 715 void ConnectionGraph::add_to_congraph_unsafe_access(Node* n, uint opcode, Unique_Node_List* delayed_worklist) { 716 Node* adr = n->in(MemNode::Address); 717 const Type* adr_type = _igvn->type(adr); 718 adr_type = adr_type->make_ptr(); 719 if (adr_type == NULL) { 720 return; // skip dead nodes 721 } 722 if (adr_type->isa_oopptr() 723 || ((opcode == Op_StoreP || opcode == Op_StoreN || opcode == Op_StoreNKlass) 724 && adr_type == TypeRawPtr::NOTNULL 725 && adr->in(AddPNode::Address)->is_Proj() 726 && adr->in(AddPNode::Address)->in(0)->is_Allocate())) { 727 delayed_worklist->push(n); // Process it later. 728 #ifdef ASSERT 729 assert (adr->is_AddP(), "expecting an AddP"); 730 if (adr_type == TypeRawPtr::NOTNULL) { 731 // Verify a raw address for a store captured by Initialize node. 732 int offs = (int) _igvn->find_intptr_t_con(adr->in(AddPNode::Offset), Type::OffsetBot); 733 assert(offs != Type::OffsetBot, "offset must be a constant"); 734 } 735 #endif 736 } else { 737 // Ignore copy the displaced header to the BoxNode (OSR compilation). 738 if (adr->is_BoxLock()) { 739 return; 740 } 741 // Stored value escapes in unsafe access. 742 if ((opcode == Op_StoreP) && adr_type->isa_rawptr()) { 743 delayed_worklist->push(n); // Process unsafe access later. 744 return; 745 } 746 #ifdef ASSERT 747 n->dump(1); 748 assert(false, "not unsafe"); 749 #endif 750 } 751 } 752 753 bool ConnectionGraph::add_final_edges_unsafe_access(Node* n, uint opcode) { 754 Node* adr = n->in(MemNode::Address); 755 const Type *adr_type = _igvn->type(adr); 756 adr_type = adr_type->make_ptr(); 757 #ifdef ASSERT 758 if (adr_type == NULL) { 759 n->dump(1); 760 assert(adr_type != NULL, "dead node should not be on list"); 761 return true; 762 } 763 #endif 764 765 if (opcode == Op_GetAndSetP || opcode == Op_GetAndSetN || 766 opcode == Op_CompareAndExchangeN || opcode == Op_CompareAndExchangeP) { 767 add_local_var_and_edge(n, PointsToNode::NoEscape, adr, NULL); 768 } 769 770 if (adr_type->isa_oopptr() 771 || ((opcode == Op_StoreP || opcode == Op_StoreN || opcode == Op_StoreNKlass) 772 && adr_type == TypeRawPtr::NOTNULL 773 && adr->in(AddPNode::Address)->is_Proj() 774 && adr->in(AddPNode::Address)->in(0)->is_Allocate())) { 775 // Point Address to Value 776 PointsToNode* adr_ptn = ptnode_adr(adr->_idx); 777 assert(adr_ptn != NULL && 778 adr_ptn->as_Field()->is_oop(), "node should be registered"); 779 Node* val = n->in(MemNode::ValueIn); 780 PointsToNode* ptn = ptnode_adr(val->_idx); 781 assert(ptn != NULL, "node should be registered"); 782 add_edge(adr_ptn, ptn); 783 return true; 784 } else if ((opcode == Op_StoreP) && adr_type->isa_rawptr()) { 785 // Stored value escapes in unsafe access. 786 Node* val = n->in(MemNode::ValueIn); 787 PointsToNode* ptn = ptnode_adr(val->_idx); 788 assert(ptn != NULL, "node should be registered"); 789 set_escape_state(ptn, PointsToNode::GlobalEscape); 790 // Add edge to object for unsafe access with offset. 791 PointsToNode* adr_ptn = ptnode_adr(adr->_idx); 792 assert(adr_ptn != NULL, "node should be registered"); 793 if (adr_ptn->is_Field()) { 794 assert(adr_ptn->as_Field()->is_oop(), "should be oop field"); 795 add_edge(adr_ptn, ptn); 796 } 797 return true; 798 } 799 return false; 800 } 801 802 void ConnectionGraph::add_call_node(CallNode* call) { 803 assert(call->returns_pointer(), "only for call which returns pointer"); 804 uint call_idx = call->_idx; 805 if (call->is_Allocate()) { 806 Node* k = call->in(AllocateNode::KlassNode); 807 const TypeKlassPtr* kt = k->bottom_type()->isa_klassptr(); 808 assert(kt != NULL, "TypeKlassPtr required."); 809 ciKlass* cik = kt->klass(); 810 PointsToNode::EscapeState es = PointsToNode::NoEscape; 811 bool scalar_replaceable = true; 812 if (call->is_AllocateArray()) { 813 if (!cik->is_array_klass()) { // StressReflectiveCode 814 es = PointsToNode::GlobalEscape; 815 } else { 816 int length = call->in(AllocateNode::ALength)->find_int_con(-1); 817 if (length < 0 || length > EliminateAllocationArraySizeLimit) { 818 // Not scalar replaceable if the length is not constant or too big. 819 scalar_replaceable = false; 820 } 821 } 822 } else { // Allocate instance 823 if (cik->is_subclass_of(_compile->env()->Thread_klass()) || 824 cik->is_subclass_of(_compile->env()->Reference_klass()) || 825 !cik->is_instance_klass() || // StressReflectiveCode 826 !cik->as_instance_klass()->can_be_instantiated() || 827 cik->as_instance_klass()->has_finalizer()) { 828 es = PointsToNode::GlobalEscape; 829 } 830 } 831 add_java_object(call, es); 832 PointsToNode* ptn = ptnode_adr(call_idx); 833 if (!scalar_replaceable && ptn->scalar_replaceable()) { 834 ptn->set_scalar_replaceable(false); 835 } 836 } else if (call->is_CallStaticJava()) { 837 // Call nodes could be different types: 838 // 839 // 1. CallDynamicJavaNode (what happened during call is unknown): 840 // 841 // - mapped to GlobalEscape JavaObject node if oop is returned; 842 // 843 // - all oop arguments are escaping globally; 844 // 845 // 2. CallStaticJavaNode (execute bytecode analysis if possible): 846 // 847 // - the same as CallDynamicJavaNode if can't do bytecode analysis; 848 // 849 // - mapped to GlobalEscape JavaObject node if unknown oop is returned; 850 // - mapped to NoEscape JavaObject node if non-escaping object allocated 851 // during call is returned; 852 // - mapped to ArgEscape LocalVar node pointed to object arguments 853 // which are returned and does not escape during call; 854 // 855 // - oop arguments escaping status is defined by bytecode analysis; 856 // 857 // For a static call, we know exactly what method is being called. 858 // Use bytecode estimator to record whether the call's return value escapes. 859 ciMethod* meth = call->as_CallJava()->method(); 860 if (meth == NULL) { 861 const char* name = call->as_CallStaticJava()->_name; 862 assert(strncmp(name, "_multianewarray", 15) == 0, "TODO: add failed case check"); 863 // Returns a newly allocated unescaped object. 864 add_java_object(call, PointsToNode::NoEscape); 865 ptnode_adr(call_idx)->set_scalar_replaceable(false); 866 } else if (meth->is_boxing_method()) { 867 // Returns boxing object 868 PointsToNode::EscapeState es; 869 vmIntrinsics::ID intr = meth->intrinsic_id(); 870 if (intr == vmIntrinsics::_floatValue || intr == vmIntrinsics::_doubleValue) { 871 // It does not escape if object is always allocated. 872 es = PointsToNode::NoEscape; 873 } else { 874 // It escapes globally if object could be loaded from cache. 875 es = PointsToNode::GlobalEscape; 876 } 877 add_java_object(call, es); 878 } else { 879 BCEscapeAnalyzer* call_analyzer = meth->get_bcea(); 880 call_analyzer->copy_dependencies(_compile->dependencies()); 881 if (call_analyzer->is_return_allocated()) { 882 // Returns a newly allocated unescaped object, simply 883 // update dependency information. 884 // Mark it as NoEscape so that objects referenced by 885 // it's fields will be marked as NoEscape at least. 886 add_java_object(call, PointsToNode::NoEscape); 887 ptnode_adr(call_idx)->set_scalar_replaceable(false); 888 } else { 889 // Determine whether any arguments are returned. 890 const TypeTuple* d = call->tf()->domain(); 891 bool ret_arg = false; 892 for (uint i = TypeFunc::Parms; i < d->cnt(); i++) { 893 if (d->field_at(i)->isa_ptr() != NULL && 894 call_analyzer->is_arg_returned(i - TypeFunc::Parms)) { 895 ret_arg = true; 896 break; 897 } 898 } 899 if (ret_arg) { 900 add_local_var(call, PointsToNode::ArgEscape); 901 } else { 902 // Returns unknown object. 903 map_ideal_node(call, phantom_obj); 904 } 905 } 906 } 907 } else { 908 // An other type of call, assume the worst case: 909 // returned value is unknown and globally escapes. 910 assert(call->Opcode() == Op_CallDynamicJava, "add failed case check"); 911 map_ideal_node(call, phantom_obj); 912 } 913 } 914 915 void ConnectionGraph::process_call_arguments(CallNode *call) { 916 bool is_arraycopy = false; 917 switch (call->Opcode()) { 918 #ifdef ASSERT 919 case Op_Allocate: 920 case Op_AllocateArray: 921 case Op_Lock: 922 case Op_Unlock: 923 assert(false, "should be done already"); 924 break; 925 #endif 926 case Op_ArrayCopy: 927 case Op_CallLeafNoFP: 928 // Most array copies are ArrayCopy nodes at this point but there 929 // are still a few direct calls to the copy subroutines (See 930 // PhaseStringOpts::copy_string()) 931 is_arraycopy = (call->Opcode() == Op_ArrayCopy) || 932 call->as_CallLeaf()->is_call_to_arraycopystub(); 933 // fall through 934 case Op_CallLeaf: { 935 // Stub calls, objects do not escape but they are not scale replaceable. 936 // Adjust escape state for outgoing arguments. 937 const TypeTuple * d = call->tf()->domain(); 938 bool src_has_oops = false; 939 for (uint i = TypeFunc::Parms; i < d->cnt(); i++) { 940 const Type* at = d->field_at(i); 941 Node *arg = call->in(i); 942 if (arg == NULL) { 943 continue; 944 } 945 const Type *aat = _igvn->type(arg); 946 if (arg->is_top() || !at->isa_ptr() || !aat->isa_ptr()) 947 continue; 948 if (arg->is_AddP()) { 949 // 950 // The inline_native_clone() case when the arraycopy stub is called 951 // after the allocation before Initialize and CheckCastPP nodes. 952 // Or normal arraycopy for object arrays case. 953 // 954 // Set AddP's base (Allocate) as not scalar replaceable since 955 // pointer to the base (with offset) is passed as argument. 956 // 957 arg = get_addp_base(arg); 958 } 959 PointsToNode* arg_ptn = ptnode_adr(arg->_idx); 960 assert(arg_ptn != NULL, "should be registered"); 961 PointsToNode::EscapeState arg_esc = arg_ptn->escape_state(); 962 if (is_arraycopy || arg_esc < PointsToNode::ArgEscape) { 963 assert(aat == Type::TOP || aat == TypePtr::NULL_PTR || 964 aat->isa_ptr() != NULL, "expecting an Ptr"); 965 bool arg_has_oops = aat->isa_oopptr() && 966 (aat->isa_oopptr()->klass() == NULL || aat->isa_instptr() || 967 (aat->isa_aryptr() && aat->isa_aryptr()->klass()->is_obj_array_klass())); 968 if (i == TypeFunc::Parms) { 969 src_has_oops = arg_has_oops; 970 } 971 // 972 // src or dst could be j.l.Object when other is basic type array: 973 // 974 // arraycopy(char[],0,Object*,0,size); 975 // arraycopy(Object*,0,char[],0,size); 976 // 977 // Don't add edges in such cases. 978 // 979 bool arg_is_arraycopy_dest = src_has_oops && is_arraycopy && 980 arg_has_oops && (i > TypeFunc::Parms); 981 #ifdef ASSERT 982 if (!(is_arraycopy || 983 BarrierSet::barrier_set()->barrier_set_c2()->is_gc_barrier_node(call) || 984 (call->as_CallLeaf()->_name != NULL && 985 (strcmp(call->as_CallLeaf()->_name, "updateBytesCRC32") == 0 || 986 strcmp(call->as_CallLeaf()->_name, "updateBytesCRC32C") == 0 || 987 strcmp(call->as_CallLeaf()->_name, "updateBytesAdler32") == 0 || 988 strcmp(call->as_CallLeaf()->_name, "aescrypt_encryptBlock") == 0 || 989 strcmp(call->as_CallLeaf()->_name, "aescrypt_decryptBlock") == 0 || 990 strcmp(call->as_CallLeaf()->_name, "cipherBlockChaining_encryptAESCrypt") == 0 || 991 strcmp(call->as_CallLeaf()->_name, "cipherBlockChaining_decryptAESCrypt") == 0 || 992 strcmp(call->as_CallLeaf()->_name, "counterMode_AESCrypt") == 0 || 993 strcmp(call->as_CallLeaf()->_name, "ghash_processBlocks") == 0 || 994 strcmp(call->as_CallLeaf()->_name, "encodeBlock") == 0 || 995 strcmp(call->as_CallLeaf()->_name, "sha1_implCompress") == 0 || 996 strcmp(call->as_CallLeaf()->_name, "sha1_implCompressMB") == 0 || 997 strcmp(call->as_CallLeaf()->_name, "sha256_implCompress") == 0 || 998 strcmp(call->as_CallLeaf()->_name, "sha256_implCompressMB") == 0 || 999 strcmp(call->as_CallLeaf()->_name, "sha512_implCompress") == 0 || 1000 strcmp(call->as_CallLeaf()->_name, "sha512_implCompressMB") == 0 || 1001 strcmp(call->as_CallLeaf()->_name, "multiplyToLen") == 0 || 1002 strcmp(call->as_CallLeaf()->_name, "squareToLen") == 0 || 1003 strcmp(call->as_CallLeaf()->_name, "mulAdd") == 0 || 1004 strcmp(call->as_CallLeaf()->_name, "montgomery_multiply") == 0 || 1005 strcmp(call->as_CallLeaf()->_name, "montgomery_square") == 0 || 1006 strcmp(call->as_CallLeaf()->_name, "vectorizedMismatch") == 0) 1007 ))) { 1008 call->dump(); 1009 fatal("EA unexpected CallLeaf %s", call->as_CallLeaf()->_name); 1010 } 1011 #endif 1012 // Always process arraycopy's destination object since 1013 // we need to add all possible edges to references in 1014 // source object. 1015 if (arg_esc >= PointsToNode::ArgEscape && 1016 !arg_is_arraycopy_dest) { 1017 continue; 1018 } 1019 PointsToNode::EscapeState es = PointsToNode::ArgEscape; 1020 if (call->is_ArrayCopy()) { 1021 ArrayCopyNode* ac = call->as_ArrayCopy(); 1022 if (ac->is_clonebasic() || 1023 ac->is_arraycopy_validated() || 1024 ac->is_copyof_validated() || 1025 ac->is_copyofrange_validated()) { 1026 es = PointsToNode::NoEscape; 1027 } 1028 } 1029 set_escape_state(arg_ptn, es); 1030 if (arg_is_arraycopy_dest) { 1031 Node* src = call->in(TypeFunc::Parms); 1032 if (src->is_AddP()) { 1033 src = get_addp_base(src); 1034 } 1035 PointsToNode* src_ptn = ptnode_adr(src->_idx); 1036 assert(src_ptn != NULL, "should be registered"); 1037 if (arg_ptn != src_ptn) { 1038 // Special arraycopy edge: 1039 // A destination object's field can't have the source object 1040 // as base since objects escape states are not related. 1041 // Only escape state of destination object's fields affects 1042 // escape state of fields in source object. 1043 add_arraycopy(call, es, src_ptn, arg_ptn); 1044 } 1045 } 1046 } 1047 } 1048 break; 1049 } 1050 case Op_CallStaticJava: { 1051 // For a static call, we know exactly what method is being called. 1052 // Use bytecode estimator to record the call's escape affects 1053 #ifdef ASSERT 1054 const char* name = call->as_CallStaticJava()->_name; 1055 assert((name == NULL || strcmp(name, "uncommon_trap") != 0), "normal calls only"); 1056 #endif 1057 ciMethod* meth = call->as_CallJava()->method(); 1058 if ((meth != NULL) && meth->is_boxing_method()) { 1059 break; // Boxing methods do not modify any oops. 1060 } 1061 BCEscapeAnalyzer* call_analyzer = (meth !=NULL) ? meth->get_bcea() : NULL; 1062 // fall-through if not a Java method or no analyzer information 1063 if (call_analyzer != NULL) { 1064 PointsToNode* call_ptn = ptnode_adr(call->_idx); 1065 const TypeTuple* d = call->tf()->domain(); 1066 for (uint i = TypeFunc::Parms; i < d->cnt(); i++) { 1067 const Type* at = d->field_at(i); 1068 int k = i - TypeFunc::Parms; 1069 Node* arg = call->in(i); 1070 PointsToNode* arg_ptn = ptnode_adr(arg->_idx); 1071 if (at->isa_ptr() != NULL && 1072 call_analyzer->is_arg_returned(k)) { 1073 // The call returns arguments. 1074 if (call_ptn != NULL) { // Is call's result used? 1075 assert(call_ptn->is_LocalVar(), "node should be registered"); 1076 assert(arg_ptn != NULL, "node should be registered"); 1077 add_edge(call_ptn, arg_ptn); 1078 } 1079 } 1080 if (at->isa_oopptr() != NULL && 1081 arg_ptn->escape_state() < PointsToNode::GlobalEscape) { 1082 if (!call_analyzer->is_arg_stack(k)) { 1083 // The argument global escapes 1084 set_escape_state(arg_ptn, PointsToNode::GlobalEscape); 1085 } else { 1086 set_escape_state(arg_ptn, PointsToNode::ArgEscape); 1087 if (!call_analyzer->is_arg_local(k)) { 1088 // The argument itself doesn't escape, but any fields might 1089 set_fields_escape_state(arg_ptn, PointsToNode::GlobalEscape); 1090 } 1091 } 1092 } 1093 } 1094 if (call_ptn != NULL && call_ptn->is_LocalVar()) { 1095 // The call returns arguments. 1096 assert(call_ptn->edge_count() > 0, "sanity"); 1097 if (!call_analyzer->is_return_local()) { 1098 // Returns also unknown object. 1099 add_edge(call_ptn, phantom_obj); 1100 } 1101 } 1102 break; 1103 } 1104 } 1105 default: { 1106 // Fall-through here if not a Java method or no analyzer information 1107 // or some other type of call, assume the worst case: all arguments 1108 // globally escape. 1109 const TypeTuple* d = call->tf()->domain(); 1110 for (uint i = TypeFunc::Parms; i < d->cnt(); i++) { 1111 const Type* at = d->field_at(i); 1112 if (at->isa_oopptr() != NULL) { 1113 Node* arg = call->in(i); 1114 if (arg->is_AddP()) { 1115 arg = get_addp_base(arg); 1116 } 1117 assert(ptnode_adr(arg->_idx) != NULL, "should be defined already"); 1118 set_escape_state(ptnode_adr(arg->_idx), PointsToNode::GlobalEscape); 1119 } 1120 } 1121 } 1122 } 1123 } 1124 1125 1126 // Finish Graph construction. 1127 bool ConnectionGraph::complete_connection_graph( 1128 GrowableArray<PointsToNode*>& ptnodes_worklist, 1129 GrowableArray<JavaObjectNode*>& non_escaped_worklist, 1130 GrowableArray<JavaObjectNode*>& java_objects_worklist, 1131 GrowableArray<FieldNode*>& oop_fields_worklist) { 1132 // Normally only 1-3 passes needed to build Connection Graph depending 1133 // on graph complexity. Observed 8 passes in jvm2008 compiler.compiler. 1134 // Set limit to 20 to catch situation when something did go wrong and 1135 // bailout Escape Analysis. 1136 // Also limit build time to 20 sec (60 in debug VM), EscapeAnalysisTimeout flag. 1137 #define CG_BUILD_ITER_LIMIT 20 1138 1139 // Propagate GlobalEscape and ArgEscape escape states and check that 1140 // we still have non-escaping objects. The method pushs on _worklist 1141 // Field nodes which reference phantom_object. 1142 if (!find_non_escaped_objects(ptnodes_worklist, non_escaped_worklist)) { 1143 return false; // Nothing to do. 1144 } 1145 // Now propagate references to all JavaObject nodes. 1146 int java_objects_length = java_objects_worklist.length(); 1147 elapsedTimer time; 1148 bool timeout = false; 1149 int new_edges = 1; 1150 int iterations = 0; 1151 do { 1152 while ((new_edges > 0) && 1153 (iterations++ < CG_BUILD_ITER_LIMIT)) { 1154 double start_time = time.seconds(); 1155 time.start(); 1156 new_edges = 0; 1157 // Propagate references to phantom_object for nodes pushed on _worklist 1158 // by find_non_escaped_objects() and find_field_value(). 1159 new_edges += add_java_object_edges(phantom_obj, false); 1160 for (int next = 0; next < java_objects_length; ++next) { 1161 JavaObjectNode* ptn = java_objects_worklist.at(next); 1162 new_edges += add_java_object_edges(ptn, true); 1163 1164 #define SAMPLE_SIZE 4 1165 if ((next % SAMPLE_SIZE) == 0) { 1166 // Each 4 iterations calculate how much time it will take 1167 // to complete graph construction. 1168 time.stop(); 1169 // Poll for requests from shutdown mechanism to quiesce compiler 1170 // because Connection graph construction may take long time. 1171 CompileBroker::maybe_block(); 1172 double stop_time = time.seconds(); 1173 double time_per_iter = (stop_time - start_time) / (double)SAMPLE_SIZE; 1174 double time_until_end = time_per_iter * (double)(java_objects_length - next); 1175 if ((start_time + time_until_end) >= EscapeAnalysisTimeout) { 1176 timeout = true; 1177 break; // Timeout 1178 } 1179 start_time = stop_time; 1180 time.start(); 1181 } 1182 #undef SAMPLE_SIZE 1183 1184 } 1185 if (timeout) break; 1186 if (new_edges > 0) { 1187 // Update escape states on each iteration if graph was updated. 1188 if (!find_non_escaped_objects(ptnodes_worklist, non_escaped_worklist)) { 1189 return false; // Nothing to do. 1190 } 1191 } 1192 time.stop(); 1193 if (time.seconds() >= EscapeAnalysisTimeout) { 1194 timeout = true; 1195 break; 1196 } 1197 } 1198 if ((iterations < CG_BUILD_ITER_LIMIT) && !timeout) { 1199 time.start(); 1200 // Find fields which have unknown value. 1201 int fields_length = oop_fields_worklist.length(); 1202 for (int next = 0; next < fields_length; next++) { 1203 FieldNode* field = oop_fields_worklist.at(next); 1204 if (field->edge_count() == 0) { 1205 new_edges += find_field_value(field); 1206 // This code may added new edges to phantom_object. 1207 // Need an other cycle to propagate references to phantom_object. 1208 } 1209 } 1210 time.stop(); 1211 if (time.seconds() >= EscapeAnalysisTimeout) { 1212 timeout = true; 1213 break; 1214 } 1215 } else { 1216 new_edges = 0; // Bailout 1217 } 1218 } while (new_edges > 0); 1219 1220 // Bailout if passed limits. 1221 if ((iterations >= CG_BUILD_ITER_LIMIT) || timeout) { 1222 Compile* C = _compile; 1223 if (C->log() != NULL) { 1224 C->log()->begin_elem("connectionGraph_bailout reason='reached "); 1225 C->log()->text("%s", timeout ? "time" : "iterations"); 1226 C->log()->end_elem(" limit'"); 1227 } 1228 assert(ExitEscapeAnalysisOnTimeout, "infinite EA connection graph build (%f sec, %d iterations) with %d nodes and worklist size %d", 1229 time.seconds(), iterations, nodes_size(), ptnodes_worklist.length()); 1230 // Possible infinite build_connection_graph loop, 1231 // bailout (no changes to ideal graph were made). 1232 return false; 1233 } 1234 #ifdef ASSERT 1235 if (Verbose && PrintEscapeAnalysis) { 1236 tty->print_cr("EA: %d iterations to build connection graph with %d nodes and worklist size %d", 1237 iterations, nodes_size(), ptnodes_worklist.length()); 1238 } 1239 #endif 1240 1241 #undef CG_BUILD_ITER_LIMIT 1242 1243 // Find fields initialized by NULL for non-escaping Allocations. 1244 int non_escaped_length = non_escaped_worklist.length(); 1245 for (int next = 0; next < non_escaped_length; next++) { 1246 JavaObjectNode* ptn = non_escaped_worklist.at(next); 1247 PointsToNode::EscapeState es = ptn->escape_state(); 1248 assert(es <= PointsToNode::ArgEscape, "sanity"); 1249 if (es == PointsToNode::NoEscape) { 1250 if (find_init_values(ptn, null_obj, _igvn) > 0) { 1251 // Adding references to NULL object does not change escape states 1252 // since it does not escape. Also no fields are added to NULL object. 1253 add_java_object_edges(null_obj, false); 1254 } 1255 } 1256 Node* n = ptn->ideal_node(); 1257 if (n->is_Allocate()) { 1258 // The object allocated by this Allocate node will never be 1259 // seen by an other thread. Mark it so that when it is 1260 // expanded no MemBarStoreStore is added. 1261 InitializeNode* ini = n->as_Allocate()->initialization(); 1262 if (ini != NULL) 1263 ini->set_does_not_escape(); 1264 } 1265 } 1266 return true; // Finished graph construction. 1267 } 1268 1269 // Propagate GlobalEscape and ArgEscape escape states to all nodes 1270 // and check that we still have non-escaping java objects. 1271 bool ConnectionGraph::find_non_escaped_objects(GrowableArray<PointsToNode*>& ptnodes_worklist, 1272 GrowableArray<JavaObjectNode*>& non_escaped_worklist) { 1273 GrowableArray<PointsToNode*> escape_worklist; 1274 // First, put all nodes with GlobalEscape and ArgEscape states on worklist. 1275 int ptnodes_length = ptnodes_worklist.length(); 1276 for (int next = 0; next < ptnodes_length; ++next) { 1277 PointsToNode* ptn = ptnodes_worklist.at(next); 1278 if (ptn->escape_state() >= PointsToNode::ArgEscape || 1279 ptn->fields_escape_state() >= PointsToNode::ArgEscape) { 1280 escape_worklist.push(ptn); 1281 } 1282 } 1283 // Set escape states to referenced nodes (edges list). 1284 while (escape_worklist.length() > 0) { 1285 PointsToNode* ptn = escape_worklist.pop(); 1286 PointsToNode::EscapeState es = ptn->escape_state(); 1287 PointsToNode::EscapeState field_es = ptn->fields_escape_state(); 1288 if (ptn->is_Field() && ptn->as_Field()->is_oop() && 1289 es >= PointsToNode::ArgEscape) { 1290 // GlobalEscape or ArgEscape state of field means it has unknown value. 1291 if (add_edge(ptn, phantom_obj)) { 1292 // New edge was added 1293 add_field_uses_to_worklist(ptn->as_Field()); 1294 } 1295 } 1296 for (EdgeIterator i(ptn); i.has_next(); i.next()) { 1297 PointsToNode* e = i.get(); 1298 if (e->is_Arraycopy()) { 1299 assert(ptn->arraycopy_dst(), "sanity"); 1300 // Propagate only fields escape state through arraycopy edge. 1301 if (e->fields_escape_state() < field_es) { 1302 set_fields_escape_state(e, field_es); 1303 escape_worklist.push(e); 1304 } 1305 } else if (es >= field_es) { 1306 // fields_escape_state is also set to 'es' if it is less than 'es'. 1307 if (e->escape_state() < es) { 1308 set_escape_state(e, es); 1309 escape_worklist.push(e); 1310 } 1311 } else { 1312 // Propagate field escape state. 1313 bool es_changed = false; 1314 if (e->fields_escape_state() < field_es) { 1315 set_fields_escape_state(e, field_es); 1316 es_changed = true; 1317 } 1318 if ((e->escape_state() < field_es) && 1319 e->is_Field() && ptn->is_JavaObject() && 1320 e->as_Field()->is_oop()) { 1321 // Change escape state of referenced fields. 1322 set_escape_state(e, field_es); 1323 es_changed = true; 1324 } else if (e->escape_state() < es) { 1325 set_escape_state(e, es); 1326 es_changed = true; 1327 } 1328 if (es_changed) { 1329 escape_worklist.push(e); 1330 } 1331 } 1332 } 1333 } 1334 // Remove escaped objects from non_escaped list. 1335 for (int next = non_escaped_worklist.length()-1; next >= 0 ; --next) { 1336 JavaObjectNode* ptn = non_escaped_worklist.at(next); 1337 if (ptn->escape_state() >= PointsToNode::GlobalEscape) { 1338 non_escaped_worklist.delete_at(next); 1339 } 1340 if (ptn->escape_state() == PointsToNode::NoEscape) { 1341 // Find fields in non-escaped allocations which have unknown value. 1342 find_init_values(ptn, phantom_obj, NULL); 1343 } 1344 } 1345 return (non_escaped_worklist.length() > 0); 1346 } 1347 1348 // Add all references to JavaObject node by walking over all uses. 1349 int ConnectionGraph::add_java_object_edges(JavaObjectNode* jobj, bool populate_worklist) { 1350 int new_edges = 0; 1351 if (populate_worklist) { 1352 // Populate _worklist by uses of jobj's uses. 1353 for (UseIterator i(jobj); i.has_next(); i.next()) { 1354 PointsToNode* use = i.get(); 1355 if (use->is_Arraycopy()) 1356 continue; 1357 add_uses_to_worklist(use); 1358 if (use->is_Field() && use->as_Field()->is_oop()) { 1359 // Put on worklist all field's uses (loads) and 1360 // related field nodes (same base and offset). 1361 add_field_uses_to_worklist(use->as_Field()); 1362 } 1363 } 1364 } 1365 for (int l = 0; l < _worklist.length(); l++) { 1366 PointsToNode* use = _worklist.at(l); 1367 if (PointsToNode::is_base_use(use)) { 1368 // Add reference from jobj to field and from field to jobj (field's base). 1369 use = PointsToNode::get_use_node(use)->as_Field(); 1370 if (add_base(use->as_Field(), jobj)) { 1371 new_edges++; 1372 } 1373 continue; 1374 } 1375 assert(!use->is_JavaObject(), "sanity"); 1376 if (use->is_Arraycopy()) { 1377 if (jobj == null_obj) // NULL object does not have field edges 1378 continue; 1379 // Added edge from Arraycopy node to arraycopy's source java object 1380 if (add_edge(use, jobj)) { 1381 jobj->set_arraycopy_src(); 1382 new_edges++; 1383 } 1384 // and stop here. 1385 continue; 1386 } 1387 if (!add_edge(use, jobj)) 1388 continue; // No new edge added, there was such edge already. 1389 new_edges++; 1390 if (use->is_LocalVar()) { 1391 add_uses_to_worklist(use); 1392 if (use->arraycopy_dst()) { 1393 for (EdgeIterator i(use); i.has_next(); i.next()) { 1394 PointsToNode* e = i.get(); 1395 if (e->is_Arraycopy()) { 1396 if (jobj == null_obj) // NULL object does not have field edges 1397 continue; 1398 // Add edge from arraycopy's destination java object to Arraycopy node. 1399 if (add_edge(jobj, e)) { 1400 new_edges++; 1401 jobj->set_arraycopy_dst(); 1402 } 1403 } 1404 } 1405 } 1406 } else { 1407 // Added new edge to stored in field values. 1408 // Put on worklist all field's uses (loads) and 1409 // related field nodes (same base and offset). 1410 add_field_uses_to_worklist(use->as_Field()); 1411 } 1412 } 1413 _worklist.clear(); 1414 _in_worklist.Reset(); 1415 return new_edges; 1416 } 1417 1418 // Put on worklist all related field nodes. 1419 void ConnectionGraph::add_field_uses_to_worklist(FieldNode* field) { 1420 assert(field->is_oop(), "sanity"); 1421 int offset = field->offset(); 1422 add_uses_to_worklist(field); 1423 // Loop over all bases of this field and push on worklist Field nodes 1424 // with the same offset and base (since they may reference the same field). 1425 for (BaseIterator i(field); i.has_next(); i.next()) { 1426 PointsToNode* base = i.get(); 1427 add_fields_to_worklist(field, base); 1428 // Check if the base was source object of arraycopy and go over arraycopy's 1429 // destination objects since values stored to a field of source object are 1430 // accessable by uses (loads) of fields of destination objects. 1431 if (base->arraycopy_src()) { 1432 for (UseIterator j(base); j.has_next(); j.next()) { 1433 PointsToNode* arycp = j.get(); 1434 if (arycp->is_Arraycopy()) { 1435 for (UseIterator k(arycp); k.has_next(); k.next()) { 1436 PointsToNode* abase = k.get(); 1437 if (abase->arraycopy_dst() && abase != base) { 1438 // Look for the same arraycopy reference. 1439 add_fields_to_worklist(field, abase); 1440 } 1441 } 1442 } 1443 } 1444 } 1445 } 1446 } 1447 1448 // Put on worklist all related field nodes. 1449 void ConnectionGraph::add_fields_to_worklist(FieldNode* field, PointsToNode* base) { 1450 int offset = field->offset(); 1451 if (base->is_LocalVar()) { 1452 for (UseIterator j(base); j.has_next(); j.next()) { 1453 PointsToNode* f = j.get(); 1454 if (PointsToNode::is_base_use(f)) { // Field 1455 f = PointsToNode::get_use_node(f); 1456 if (f == field || !f->as_Field()->is_oop()) 1457 continue; 1458 int offs = f->as_Field()->offset(); 1459 if (offs == offset || offset == Type::OffsetBot || offs == Type::OffsetBot) { 1460 add_to_worklist(f); 1461 } 1462 } 1463 } 1464 } else { 1465 assert(base->is_JavaObject(), "sanity"); 1466 if (// Skip phantom_object since it is only used to indicate that 1467 // this field's content globally escapes. 1468 (base != phantom_obj) && 1469 // NULL object node does not have fields. 1470 (base != null_obj)) { 1471 for (EdgeIterator i(base); i.has_next(); i.next()) { 1472 PointsToNode* f = i.get(); 1473 // Skip arraycopy edge since store to destination object field 1474 // does not update value in source object field. 1475 if (f->is_Arraycopy()) { 1476 assert(base->arraycopy_dst(), "sanity"); 1477 continue; 1478 } 1479 if (f == field || !f->as_Field()->is_oop()) 1480 continue; 1481 int offs = f->as_Field()->offset(); 1482 if (offs == offset || offset == Type::OffsetBot || offs == Type::OffsetBot) { 1483 add_to_worklist(f); 1484 } 1485 } 1486 } 1487 } 1488 } 1489 1490 // Find fields which have unknown value. 1491 int ConnectionGraph::find_field_value(FieldNode* field) { 1492 // Escaped fields should have init value already. 1493 assert(field->escape_state() == PointsToNode::NoEscape, "sanity"); 1494 int new_edges = 0; 1495 for (BaseIterator i(field); i.has_next(); i.next()) { 1496 PointsToNode* base = i.get(); 1497 if (base->is_JavaObject()) { 1498 // Skip Allocate's fields which will be processed later. 1499 if (base->ideal_node()->is_Allocate()) 1500 return 0; 1501 assert(base == null_obj, "only NULL ptr base expected here"); 1502 } 1503 } 1504 if (add_edge(field, phantom_obj)) { 1505 // New edge was added 1506 new_edges++; 1507 add_field_uses_to_worklist(field); 1508 } 1509 return new_edges; 1510 } 1511 1512 // Find fields initializing values for allocations. 1513 int ConnectionGraph::find_init_values(JavaObjectNode* pta, PointsToNode* init_val, PhaseTransform* phase) { 1514 assert(pta->escape_state() == PointsToNode::NoEscape, "Not escaped Allocate nodes only"); 1515 int new_edges = 0; 1516 Node* alloc = pta->ideal_node(); 1517 if (init_val == phantom_obj) { 1518 // Do nothing for Allocate nodes since its fields values are 1519 // "known" unless they are initialized by arraycopy/clone. 1520 if (alloc->is_Allocate() && !pta->arraycopy_dst()) 1521 return 0; 1522 assert(pta->arraycopy_dst() || alloc->as_CallStaticJava(), "sanity"); 1523 #ifdef ASSERT 1524 if (!pta->arraycopy_dst() && alloc->as_CallStaticJava()->method() == NULL) { 1525 const char* name = alloc->as_CallStaticJava()->_name; 1526 assert(strncmp(name, "_multianewarray", 15) == 0, "sanity"); 1527 } 1528 #endif 1529 // Non-escaped allocation returned from Java or runtime call have 1530 // unknown values in fields. 1531 for (EdgeIterator i(pta); i.has_next(); i.next()) { 1532 PointsToNode* field = i.get(); 1533 if (field->is_Field() && field->as_Field()->is_oop()) { 1534 if (add_edge(field, phantom_obj)) { 1535 // New edge was added 1536 new_edges++; 1537 add_field_uses_to_worklist(field->as_Field()); 1538 } 1539 } 1540 } 1541 return new_edges; 1542 } 1543 assert(init_val == null_obj, "sanity"); 1544 // Do nothing for Call nodes since its fields values are unknown. 1545 if (!alloc->is_Allocate()) 1546 return 0; 1547 1548 InitializeNode* ini = alloc->as_Allocate()->initialization(); 1549 bool visited_bottom_offset = false; 1550 GrowableArray<int> offsets_worklist; 1551 1552 // Check if an oop field's initializing value is recorded and add 1553 // a corresponding NULL if field's value if it is not recorded. 1554 // Connection Graph does not record a default initialization by NULL 1555 // captured by Initialize node. 1556 // 1557 for (EdgeIterator i(pta); i.has_next(); i.next()) { 1558 PointsToNode* field = i.get(); // Field (AddP) 1559 if (!field->is_Field() || !field->as_Field()->is_oop()) 1560 continue; // Not oop field 1561 int offset = field->as_Field()->offset(); 1562 if (offset == Type::OffsetBot) { 1563 if (!visited_bottom_offset) { 1564 // OffsetBot is used to reference array's element, 1565 // always add reference to NULL to all Field nodes since we don't 1566 // known which element is referenced. 1567 if (add_edge(field, null_obj)) { 1568 // New edge was added 1569 new_edges++; 1570 add_field_uses_to_worklist(field->as_Field()); 1571 visited_bottom_offset = true; 1572 } 1573 } 1574 } else { 1575 // Check only oop fields. 1576 const Type* adr_type = field->ideal_node()->as_AddP()->bottom_type(); 1577 if (adr_type->isa_rawptr()) { 1578 #ifdef ASSERT 1579 // Raw pointers are used for initializing stores so skip it 1580 // since it should be recorded already 1581 Node* base = get_addp_base(field->ideal_node()); 1582 assert(adr_type->isa_rawptr() && base->is_Proj() && 1583 (base->in(0) == alloc),"unexpected pointer type"); 1584 #endif 1585 continue; 1586 } 1587 if (!offsets_worklist.contains(offset)) { 1588 offsets_worklist.append(offset); 1589 Node* value = NULL; 1590 if (ini != NULL) { 1591 // StoreP::memory_type() == T_ADDRESS 1592 BasicType ft = UseCompressedOops ? T_NARROWOOP : T_ADDRESS; 1593 Node* store = ini->find_captured_store(offset, type2aelembytes(ft, true), phase); 1594 // Make sure initializing store has the same type as this AddP. 1595 // This AddP may reference non existing field because it is on a 1596 // dead branch of bimorphic call which is not eliminated yet. 1597 if (store != NULL && store->is_Store() && 1598 store->as_Store()->memory_type() == ft) { 1599 value = store->in(MemNode::ValueIn); 1600 #ifdef ASSERT 1601 if (VerifyConnectionGraph) { 1602 // Verify that AddP already points to all objects the value points to. 1603 PointsToNode* val = ptnode_adr(value->_idx); 1604 assert((val != NULL), "should be processed already"); 1605 PointsToNode* missed_obj = NULL; 1606 if (val->is_JavaObject()) { 1607 if (!field->points_to(val->as_JavaObject())) { 1608 missed_obj = val; 1609 } 1610 } else { 1611 if (!val->is_LocalVar() || (val->edge_count() == 0)) { 1612 tty->print_cr("----------init store has invalid value -----"); 1613 store->dump(); 1614 val->dump(); 1615 assert(val->is_LocalVar() && (val->edge_count() > 0), "should be processed already"); 1616 } 1617 for (EdgeIterator j(val); j.has_next(); j.next()) { 1618 PointsToNode* obj = j.get(); 1619 if (obj->is_JavaObject()) { 1620 if (!field->points_to(obj->as_JavaObject())) { 1621 missed_obj = obj; 1622 break; 1623 } 1624 } 1625 } 1626 } 1627 if (missed_obj != NULL) { 1628 tty->print_cr("----------field---------------------------------"); 1629 field->dump(); 1630 tty->print_cr("----------missed referernce to object-----------"); 1631 missed_obj->dump(); 1632 tty->print_cr("----------object referernced by init store -----"); 1633 store->dump(); 1634 val->dump(); 1635 assert(!field->points_to(missed_obj->as_JavaObject()), "missed JavaObject reference"); 1636 } 1637 } 1638 #endif 1639 } else { 1640 // There could be initializing stores which follow allocation. 1641 // For example, a volatile field store is not collected 1642 // by Initialize node. 1643 // 1644 // Need to check for dependent loads to separate such stores from 1645 // stores which follow loads. For now, add initial value NULL so 1646 // that compare pointers optimization works correctly. 1647 } 1648 } 1649 if (value == NULL) { 1650 // A field's initializing value was not recorded. Add NULL. 1651 if (add_edge(field, null_obj)) { 1652 // New edge was added 1653 new_edges++; 1654 add_field_uses_to_worklist(field->as_Field()); 1655 } 1656 } 1657 } 1658 } 1659 } 1660 return new_edges; 1661 } 1662 1663 // Adjust scalar_replaceable state after Connection Graph is built. 1664 void ConnectionGraph::adjust_scalar_replaceable_state(JavaObjectNode* jobj) { 1665 // Search for non-escaping objects which are not scalar replaceable 1666 // and mark them to propagate the state to referenced objects. 1667 1668 // 1. An object is not scalar replaceable if the field into which it is 1669 // stored has unknown offset (stored into unknown element of an array). 1670 // 1671 for (UseIterator i(jobj); i.has_next(); i.next()) { 1672 PointsToNode* use = i.get(); 1673 if (use->is_Arraycopy()) { 1674 continue; 1675 } 1676 if (use->is_Field()) { 1677 FieldNode* field = use->as_Field(); 1678 assert(field->is_oop() && field->scalar_replaceable(), "sanity"); 1679 if (field->offset() == Type::OffsetBot) { 1680 jobj->set_scalar_replaceable(false); 1681 return; 1682 } 1683 // 2. An object is not scalar replaceable if the field into which it is 1684 // stored has multiple bases one of which is null. 1685 if (field->base_count() > 1) { 1686 for (BaseIterator i(field); i.has_next(); i.next()) { 1687 PointsToNode* base = i.get(); 1688 if (base == null_obj) { 1689 jobj->set_scalar_replaceable(false); 1690 return; 1691 } 1692 } 1693 } 1694 } 1695 assert(use->is_Field() || use->is_LocalVar(), "sanity"); 1696 // 3. An object is not scalar replaceable if it is merged with other objects. 1697 for (EdgeIterator j(use); j.has_next(); j.next()) { 1698 PointsToNode* ptn = j.get(); 1699 if (ptn->is_JavaObject() && ptn != jobj) { 1700 // Mark all objects. 1701 jobj->set_scalar_replaceable(false); 1702 ptn->set_scalar_replaceable(false); 1703 } 1704 } 1705 if (!jobj->scalar_replaceable()) { 1706 return; 1707 } 1708 } 1709 1710 for (EdgeIterator j(jobj); j.has_next(); j.next()) { 1711 if (j.get()->is_Arraycopy()) { 1712 continue; 1713 } 1714 1715 // Non-escaping object node should point only to field nodes. 1716 FieldNode* field = j.get()->as_Field(); 1717 int offset = field->as_Field()->offset(); 1718 1719 // 4. An object is not scalar replaceable if it has a field with unknown 1720 // offset (array's element is accessed in loop). 1721 if (offset == Type::OffsetBot) { 1722 jobj->set_scalar_replaceable(false); 1723 return; 1724 } 1725 // 5. Currently an object is not scalar replaceable if a LoadStore node 1726 // access its field since the field value is unknown after it. 1727 // 1728 Node* n = field->ideal_node(); 1729 1730 // Test for an unsafe access that was parsed as maybe off heap 1731 // (with a CheckCastPP to raw memory). 1732 assert(n->is_AddP(), "expect an address computation"); 1733 if (n->in(AddPNode::Base)->is_top() && 1734 n->in(AddPNode::Address)->Opcode() == Op_CheckCastPP) { 1735 assert(n->in(AddPNode::Address)->bottom_type()->isa_rawptr(), "raw address so raw cast expected"); 1736 assert(_igvn->type(n->in(AddPNode::Address)->in(1))->isa_oopptr(), "cast pattern at unsafe access expected"); 1737 jobj->set_scalar_replaceable(false); 1738 return; 1739 } 1740 1741 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { 1742 Node* u = n->fast_out(i); 1743 if (u->is_LoadStore() || (u->is_Mem() && u->as_Mem()->is_mismatched_access())) { 1744 jobj->set_scalar_replaceable(false); 1745 return; 1746 } 1747 } 1748 1749 // 6. Or the address may point to more then one object. This may produce 1750 // the false positive result (set not scalar replaceable) 1751 // since the flow-insensitive escape analysis can't separate 1752 // the case when stores overwrite the field's value from the case 1753 // when stores happened on different control branches. 1754 // 1755 // Note: it will disable scalar replacement in some cases: 1756 // 1757 // Point p[] = new Point[1]; 1758 // p[0] = new Point(); // Will be not scalar replaced 1759 // 1760 // but it will save us from incorrect optimizations in next cases: 1761 // 1762 // Point p[] = new Point[1]; 1763 // if ( x ) p[0] = new Point(); // Will be not scalar replaced 1764 // 1765 if (field->base_count() > 1) { 1766 for (BaseIterator i(field); i.has_next(); i.next()) { 1767 PointsToNode* base = i.get(); 1768 // Don't take into account LocalVar nodes which 1769 // may point to only one object which should be also 1770 // this field's base by now. 1771 if (base->is_JavaObject() && base != jobj) { 1772 // Mark all bases. 1773 jobj->set_scalar_replaceable(false); 1774 base->set_scalar_replaceable(false); 1775 } 1776 } 1777 } 1778 } 1779 } 1780 1781 #ifdef ASSERT 1782 void ConnectionGraph::verify_connection_graph( 1783 GrowableArray<PointsToNode*>& ptnodes_worklist, 1784 GrowableArray<JavaObjectNode*>& non_escaped_worklist, 1785 GrowableArray<JavaObjectNode*>& java_objects_worklist, 1786 GrowableArray<Node*>& addp_worklist) { 1787 // Verify that graph is complete - no new edges could be added. 1788 int java_objects_length = java_objects_worklist.length(); 1789 int non_escaped_length = non_escaped_worklist.length(); 1790 int new_edges = 0; 1791 for (int next = 0; next < java_objects_length; ++next) { 1792 JavaObjectNode* ptn = java_objects_worklist.at(next); 1793 new_edges += add_java_object_edges(ptn, true); 1794 } 1795 assert(new_edges == 0, "graph was not complete"); 1796 // Verify that escape state is final. 1797 int length = non_escaped_worklist.length(); 1798 find_non_escaped_objects(ptnodes_worklist, non_escaped_worklist); 1799 assert((non_escaped_length == non_escaped_worklist.length()) && 1800 (non_escaped_length == length) && 1801 (_worklist.length() == 0), "escape state was not final"); 1802 1803 // Verify fields information. 1804 int addp_length = addp_worklist.length(); 1805 for (int next = 0; next < addp_length; ++next ) { 1806 Node* n = addp_worklist.at(next); 1807 FieldNode* field = ptnode_adr(n->_idx)->as_Field(); 1808 if (field->is_oop()) { 1809 // Verify that field has all bases 1810 Node* base = get_addp_base(n); 1811 PointsToNode* ptn = ptnode_adr(base->_idx); 1812 if (ptn->is_JavaObject()) { 1813 assert(field->has_base(ptn->as_JavaObject()), "sanity"); 1814 } else { 1815 assert(ptn->is_LocalVar(), "sanity"); 1816 for (EdgeIterator i(ptn); i.has_next(); i.next()) { 1817 PointsToNode* e = i.get(); 1818 if (e->is_JavaObject()) { 1819 assert(field->has_base(e->as_JavaObject()), "sanity"); 1820 } 1821 } 1822 } 1823 // Verify that all fields have initializing values. 1824 if (field->edge_count() == 0) { 1825 tty->print_cr("----------field does not have references----------"); 1826 field->dump(); 1827 for (BaseIterator i(field); i.has_next(); i.next()) { 1828 PointsToNode* base = i.get(); 1829 tty->print_cr("----------field has next base---------------------"); 1830 base->dump(); 1831 if (base->is_JavaObject() && (base != phantom_obj) && (base != null_obj)) { 1832 tty->print_cr("----------base has fields-------------------------"); 1833 for (EdgeIterator j(base); j.has_next(); j.next()) { 1834 j.get()->dump(); 1835 } 1836 tty->print_cr("----------base has references---------------------"); 1837 for (UseIterator j(base); j.has_next(); j.next()) { 1838 j.get()->dump(); 1839 } 1840 } 1841 } 1842 for (UseIterator i(field); i.has_next(); i.next()) { 1843 i.get()->dump(); 1844 } 1845 assert(field->edge_count() > 0, "sanity"); 1846 } 1847 } 1848 } 1849 } 1850 #endif 1851 1852 // Optimize ideal graph. 1853 void ConnectionGraph::optimize_ideal_graph(GrowableArray<Node*>& ptr_cmp_worklist, 1854 GrowableArray<Node*>& storestore_worklist) { 1855 Compile* C = _compile; 1856 PhaseIterGVN* igvn = _igvn; 1857 if (EliminateLocks) { 1858 // Mark locks before changing ideal graph. 1859 int cnt = C->macro_count(); 1860 for( int i=0; i < cnt; i++ ) { 1861 Node *n = C->macro_node(i); 1862 if (n->is_AbstractLock()) { // Lock and Unlock nodes 1863 AbstractLockNode* alock = n->as_AbstractLock(); 1864 if (!alock->is_non_esc_obj()) { 1865 if (not_global_escape(alock->obj_node())) { 1866 assert(!alock->is_eliminated() || alock->is_coarsened(), "sanity"); 1867 // The lock could be marked eliminated by lock coarsening 1868 // code during first IGVN before EA. Replace coarsened flag 1869 // to eliminate all associated locks/unlocks. 1870 #ifdef ASSERT 1871 alock->log_lock_optimization(C, "eliminate_lock_set_non_esc3"); 1872 #endif 1873 alock->set_non_esc_obj(); 1874 } 1875 } 1876 } 1877 } 1878 } 1879 1880 if (OptimizePtrCompare) { 1881 // Add ConI(#CC_GT) and ConI(#CC_EQ). 1882 _pcmp_neq = igvn->makecon(TypeInt::CC_GT); 1883 _pcmp_eq = igvn->makecon(TypeInt::CC_EQ); 1884 // Optimize objects compare. 1885 while (ptr_cmp_worklist.length() != 0) { 1886 Node *n = ptr_cmp_worklist.pop(); 1887 Node *res = optimize_ptr_compare(n); 1888 if (res != NULL) { 1889 #ifndef PRODUCT 1890 if (PrintOptimizePtrCompare) { 1891 tty->print_cr("++++ Replaced: %d %s(%d,%d) --> %s", n->_idx, (n->Opcode() == Op_CmpP ? "CmpP" : "CmpN"), n->in(1)->_idx, n->in(2)->_idx, (res == _pcmp_eq ? "EQ" : "NotEQ")); 1892 if (Verbose) { 1893 n->dump(1); 1894 } 1895 } 1896 #endif 1897 igvn->replace_node(n, res); 1898 } 1899 } 1900 // cleanup 1901 if (_pcmp_neq->outcnt() == 0) 1902 igvn->hash_delete(_pcmp_neq); 1903 if (_pcmp_eq->outcnt() == 0) 1904 igvn->hash_delete(_pcmp_eq); 1905 } 1906 1907 // For MemBarStoreStore nodes added in library_call.cpp, check 1908 // escape status of associated AllocateNode and optimize out 1909 // MemBarStoreStore node if the allocated object never escapes. 1910 while (storestore_worklist.length() != 0) { 1911 Node *n = storestore_worklist.pop(); 1912 MemBarStoreStoreNode *storestore = n ->as_MemBarStoreStore(); 1913 Node *alloc = storestore->in(MemBarNode::Precedent)->in(0); 1914 assert (alloc->is_Allocate(), "storestore should point to AllocateNode"); 1915 if (not_global_escape(alloc)) { 1916 MemBarNode* mb = MemBarNode::make(C, Op_MemBarCPUOrder, Compile::AliasIdxBot); 1917 mb->init_req(TypeFunc::Memory, storestore->in(TypeFunc::Memory)); 1918 mb->init_req(TypeFunc::Control, storestore->in(TypeFunc::Control)); 1919 igvn->register_new_node_with_optimizer(mb); 1920 igvn->replace_node(storestore, mb); 1921 } 1922 } 1923 } 1924 1925 // Optimize objects compare. 1926 Node* ConnectionGraph::optimize_ptr_compare(Node* n) { 1927 assert(OptimizePtrCompare, "sanity"); 1928 PointsToNode* ptn1 = ptnode_adr(n->in(1)->_idx); 1929 PointsToNode* ptn2 = ptnode_adr(n->in(2)->_idx); 1930 JavaObjectNode* jobj1 = unique_java_object(n->in(1)); 1931 JavaObjectNode* jobj2 = unique_java_object(n->in(2)); 1932 assert(ptn1->is_JavaObject() || ptn1->is_LocalVar(), "sanity"); 1933 assert(ptn2->is_JavaObject() || ptn2->is_LocalVar(), "sanity"); 1934 1935 // Check simple cases first. 1936 if (jobj1 != NULL) { 1937 if (jobj1->escape_state() == PointsToNode::NoEscape) { 1938 if (jobj1 == jobj2) { 1939 // Comparing the same not escaping object. 1940 return _pcmp_eq; 1941 } 1942 Node* obj = jobj1->ideal_node(); 1943 // Comparing not escaping allocation. 1944 if ((obj->is_Allocate() || obj->is_CallStaticJava()) && 1945 !ptn2->points_to(jobj1)) { 1946 return _pcmp_neq; // This includes nullness check. 1947 } 1948 } 1949 } 1950 if (jobj2 != NULL) { 1951 if (jobj2->escape_state() == PointsToNode::NoEscape) { 1952 Node* obj = jobj2->ideal_node(); 1953 // Comparing not escaping allocation. 1954 if ((obj->is_Allocate() || obj->is_CallStaticJava()) && 1955 !ptn1->points_to(jobj2)) { 1956 return _pcmp_neq; // This includes nullness check. 1957 } 1958 } 1959 } 1960 if (jobj1 != NULL && jobj1 != phantom_obj && 1961 jobj2 != NULL && jobj2 != phantom_obj && 1962 jobj1->ideal_node()->is_Con() && 1963 jobj2->ideal_node()->is_Con()) { 1964 // Klass or String constants compare. Need to be careful with 1965 // compressed pointers - compare types of ConN and ConP instead of nodes. 1966 const Type* t1 = jobj1->ideal_node()->get_ptr_type(); 1967 const Type* t2 = jobj2->ideal_node()->get_ptr_type(); 1968 if (t1->make_ptr() == t2->make_ptr()) { 1969 return _pcmp_eq; 1970 } else { 1971 return _pcmp_neq; 1972 } 1973 } 1974 if (ptn1->meet(ptn2)) { 1975 return NULL; // Sets are not disjoint 1976 } 1977 1978 // Sets are disjoint. 1979 bool set1_has_unknown_ptr = ptn1->points_to(phantom_obj); 1980 bool set2_has_unknown_ptr = ptn2->points_to(phantom_obj); 1981 bool set1_has_null_ptr = ptn1->points_to(null_obj); 1982 bool set2_has_null_ptr = ptn2->points_to(null_obj); 1983 if ((set1_has_unknown_ptr && set2_has_null_ptr) || 1984 (set2_has_unknown_ptr && set1_has_null_ptr)) { 1985 // Check nullness of unknown object. 1986 return NULL; 1987 } 1988 1989 // Disjointness by itself is not sufficient since 1990 // alias analysis is not complete for escaped objects. 1991 // Disjoint sets are definitely unrelated only when 1992 // at least one set has only not escaping allocations. 1993 if (!set1_has_unknown_ptr && !set1_has_null_ptr) { 1994 if (ptn1->non_escaping_allocation()) { 1995 return _pcmp_neq; 1996 } 1997 } 1998 if (!set2_has_unknown_ptr && !set2_has_null_ptr) { 1999 if (ptn2->non_escaping_allocation()) { 2000 return _pcmp_neq; 2001 } 2002 } 2003 return NULL; 2004 } 2005 2006 // Connection Graph constuction functions. 2007 2008 void ConnectionGraph::add_local_var(Node *n, PointsToNode::EscapeState es) { 2009 PointsToNode* ptadr = _nodes.at(n->_idx); 2010 if (ptadr != NULL) { 2011 assert(ptadr->is_LocalVar() && ptadr->ideal_node() == n, "sanity"); 2012 return; 2013 } 2014 Compile* C = _compile; 2015 ptadr = new (C->comp_arena()) LocalVarNode(this, n, es); 2016 _nodes.at_put(n->_idx, ptadr); 2017 } 2018 2019 void ConnectionGraph::add_java_object(Node *n, PointsToNode::EscapeState es) { 2020 PointsToNode* ptadr = _nodes.at(n->_idx); 2021 if (ptadr != NULL) { 2022 assert(ptadr->is_JavaObject() && ptadr->ideal_node() == n, "sanity"); 2023 return; 2024 } 2025 Compile* C = _compile; 2026 ptadr = new (C->comp_arena()) JavaObjectNode(this, n, es); 2027 _nodes.at_put(n->_idx, ptadr); 2028 } 2029 2030 void ConnectionGraph::add_field(Node *n, PointsToNode::EscapeState es, int offset) { 2031 PointsToNode* ptadr = _nodes.at(n->_idx); 2032 if (ptadr != NULL) { 2033 assert(ptadr->is_Field() && ptadr->ideal_node() == n, "sanity"); 2034 return; 2035 } 2036 bool unsafe = false; 2037 bool is_oop = is_oop_field(n, offset, &unsafe); 2038 if (unsafe) { 2039 es = PointsToNode::GlobalEscape; 2040 } 2041 Compile* C = _compile; 2042 FieldNode* field = new (C->comp_arena()) FieldNode(this, n, es, offset, is_oop); 2043 _nodes.at_put(n->_idx, field); 2044 } 2045 2046 void ConnectionGraph::add_arraycopy(Node *n, PointsToNode::EscapeState es, 2047 PointsToNode* src, PointsToNode* dst) { 2048 assert(!src->is_Field() && !dst->is_Field(), "only for JavaObject and LocalVar"); 2049 assert((src != null_obj) && (dst != null_obj), "not for ConP NULL"); 2050 PointsToNode* ptadr = _nodes.at(n->_idx); 2051 if (ptadr != NULL) { 2052 assert(ptadr->is_Arraycopy() && ptadr->ideal_node() == n, "sanity"); 2053 return; 2054 } 2055 Compile* C = _compile; 2056 ptadr = new (C->comp_arena()) ArraycopyNode(this, n, es); 2057 _nodes.at_put(n->_idx, ptadr); 2058 // Add edge from arraycopy node to source object. 2059 (void)add_edge(ptadr, src); 2060 src->set_arraycopy_src(); 2061 // Add edge from destination object to arraycopy node. 2062 (void)add_edge(dst, ptadr); 2063 dst->set_arraycopy_dst(); 2064 } 2065 2066 bool ConnectionGraph::is_oop_field(Node* n, int offset, bool* unsafe) { 2067 const Type* adr_type = n->as_AddP()->bottom_type(); 2068 BasicType bt = T_INT; 2069 if (offset == Type::OffsetBot) { 2070 // Check only oop fields. 2071 if (!adr_type->isa_aryptr() || 2072 (adr_type->isa_aryptr()->klass() == NULL) || 2073 adr_type->isa_aryptr()->klass()->is_obj_array_klass()) { 2074 // OffsetBot is used to reference array's element. Ignore first AddP. 2075 if (find_second_addp(n, n->in(AddPNode::Base)) == NULL) { 2076 bt = T_OBJECT; 2077 } 2078 } 2079 } else if (offset != oopDesc::klass_offset_in_bytes()) { 2080 if (adr_type->isa_instptr()) { 2081 ciField* field = _compile->alias_type(adr_type->isa_instptr())->field(); 2082 if (field != NULL) { 2083 bt = field->layout_type(); 2084 } else { 2085 // Check for unsafe oop field access 2086 if (n->has_out_with(Op_StoreP, Op_LoadP, Op_StoreN, Op_LoadN) || 2087 n->has_out_with(Op_GetAndSetP, Op_GetAndSetN, Op_CompareAndExchangeP, Op_CompareAndExchangeN) || 2088 n->has_out_with(Op_CompareAndSwapP, Op_CompareAndSwapN, Op_WeakCompareAndSwapP, Op_WeakCompareAndSwapN) || 2089 BarrierSet::barrier_set()->barrier_set_c2()->escape_has_out_with_unsafe_object(n)) { 2090 bt = T_OBJECT; 2091 (*unsafe) = true; 2092 } 2093 } 2094 } else if (adr_type->isa_aryptr()) { 2095 if (offset == arrayOopDesc::length_offset_in_bytes()) { 2096 // Ignore array length load. 2097 } else if (find_second_addp(n, n->in(AddPNode::Base)) != NULL) { 2098 // Ignore first AddP. 2099 } else { 2100 const Type* elemtype = adr_type->isa_aryptr()->elem(); 2101 bt = elemtype->array_element_basic_type(); 2102 } 2103 } else if (adr_type->isa_rawptr() || adr_type->isa_klassptr()) { 2104 // Allocation initialization, ThreadLocal field access, unsafe access 2105 if (n->has_out_with(Op_StoreP, Op_LoadP, Op_StoreN, Op_LoadN) || 2106 n->has_out_with(Op_GetAndSetP, Op_GetAndSetN, Op_CompareAndExchangeP, Op_CompareAndExchangeN) || 2107 n->has_out_with(Op_CompareAndSwapP, Op_CompareAndSwapN, Op_WeakCompareAndSwapP, Op_WeakCompareAndSwapN) || 2108 BarrierSet::barrier_set()->barrier_set_c2()->escape_has_out_with_unsafe_object(n)) { 2109 bt = T_OBJECT; 2110 } 2111 } 2112 } 2113 return (bt == T_OBJECT || bt == T_NARROWOOP || bt == T_ARRAY); 2114 } 2115 2116 // Returns unique pointed java object or NULL. 2117 JavaObjectNode* ConnectionGraph::unique_java_object(Node *n) { 2118 assert(!_collecting, "should not call when contructed graph"); 2119 // If the node was created after the escape computation we can't answer. 2120 uint idx = n->_idx; 2121 if (idx >= nodes_size()) { 2122 return NULL; 2123 } 2124 PointsToNode* ptn = ptnode_adr(idx); 2125 if (ptn->is_JavaObject()) { 2126 return ptn->as_JavaObject(); 2127 } 2128 assert(ptn->is_LocalVar(), "sanity"); 2129 // Check all java objects it points to. 2130 JavaObjectNode* jobj = NULL; 2131 for (EdgeIterator i(ptn); i.has_next(); i.next()) { 2132 PointsToNode* e = i.get(); 2133 if (e->is_JavaObject()) { 2134 if (jobj == NULL) { 2135 jobj = e->as_JavaObject(); 2136 } else if (jobj != e) { 2137 return NULL; 2138 } 2139 } 2140 } 2141 return jobj; 2142 } 2143 2144 // Return true if this node points only to non-escaping allocations. 2145 bool PointsToNode::non_escaping_allocation() { 2146 if (is_JavaObject()) { 2147 Node* n = ideal_node(); 2148 if (n->is_Allocate() || n->is_CallStaticJava()) { 2149 return (escape_state() == PointsToNode::NoEscape); 2150 } else { 2151 return false; 2152 } 2153 } 2154 assert(is_LocalVar(), "sanity"); 2155 // Check all java objects it points to. 2156 for (EdgeIterator i(this); i.has_next(); i.next()) { 2157 PointsToNode* e = i.get(); 2158 if (e->is_JavaObject()) { 2159 Node* n = e->ideal_node(); 2160 if ((e->escape_state() != PointsToNode::NoEscape) || 2161 !(n->is_Allocate() || n->is_CallStaticJava())) { 2162 return false; 2163 } 2164 } 2165 } 2166 return true; 2167 } 2168 2169 // Return true if we know the node does not escape globally. 2170 bool ConnectionGraph::not_global_escape(Node *n) { 2171 assert(!_collecting, "should not call during graph construction"); 2172 // If the node was created after the escape computation we can't answer. 2173 uint idx = n->_idx; 2174 if (idx >= nodes_size()) { 2175 return false; 2176 } 2177 PointsToNode* ptn = ptnode_adr(idx); 2178 PointsToNode::EscapeState es = ptn->escape_state(); 2179 // If we have already computed a value, return it. 2180 if (es >= PointsToNode::GlobalEscape) 2181 return false; 2182 if (ptn->is_JavaObject()) { 2183 return true; // (es < PointsToNode::GlobalEscape); 2184 } 2185 assert(ptn->is_LocalVar(), "sanity"); 2186 // Check all java objects it points to. 2187 for (EdgeIterator i(ptn); i.has_next(); i.next()) { 2188 if (i.get()->escape_state() >= PointsToNode::GlobalEscape) 2189 return false; 2190 } 2191 return true; 2192 } 2193 2194 2195 // Helper functions 2196 2197 // Return true if this node points to specified node or nodes it points to. 2198 bool PointsToNode::points_to(JavaObjectNode* ptn) const { 2199 if (is_JavaObject()) { 2200 return (this == ptn); 2201 } 2202 assert(is_LocalVar() || is_Field(), "sanity"); 2203 for (EdgeIterator i(this); i.has_next(); i.next()) { 2204 if (i.get() == ptn) 2205 return true; 2206 } 2207 return false; 2208 } 2209 2210 // Return true if one node points to an other. 2211 bool PointsToNode::meet(PointsToNode* ptn) { 2212 if (this == ptn) { 2213 return true; 2214 } else if (ptn->is_JavaObject()) { 2215 return this->points_to(ptn->as_JavaObject()); 2216 } else if (this->is_JavaObject()) { 2217 return ptn->points_to(this->as_JavaObject()); 2218 } 2219 assert(this->is_LocalVar() && ptn->is_LocalVar(), "sanity"); 2220 int ptn_count = ptn->edge_count(); 2221 for (EdgeIterator i(this); i.has_next(); i.next()) { 2222 PointsToNode* this_e = i.get(); 2223 for (int j = 0; j < ptn_count; j++) { 2224 if (this_e == ptn->edge(j)) 2225 return true; 2226 } 2227 } 2228 return false; 2229 } 2230 2231 #ifdef ASSERT 2232 // Return true if bases point to this java object. 2233 bool FieldNode::has_base(JavaObjectNode* jobj) const { 2234 for (BaseIterator i(this); i.has_next(); i.next()) { 2235 if (i.get() == jobj) 2236 return true; 2237 } 2238 return false; 2239 } 2240 #endif 2241 2242 int ConnectionGraph::address_offset(Node* adr, PhaseTransform *phase) { 2243 const Type *adr_type = phase->type(adr); 2244 if (adr->is_AddP() && adr_type->isa_oopptr() == NULL && 2245 adr->in(AddPNode::Address)->is_Proj() && 2246 adr->in(AddPNode::Address)->in(0)->is_Allocate()) { 2247 // We are computing a raw address for a store captured by an Initialize 2248 // compute an appropriate address type. AddP cases #3 and #5 (see below). 2249 int offs = (int)phase->find_intptr_t_con(adr->in(AddPNode::Offset), Type::OffsetBot); 2250 assert(offs != Type::OffsetBot || 2251 adr->in(AddPNode::Address)->in(0)->is_AllocateArray(), 2252 "offset must be a constant or it is initialization of array"); 2253 return offs; 2254 } 2255 const TypePtr *t_ptr = adr_type->isa_ptr(); 2256 assert(t_ptr != NULL, "must be a pointer type"); 2257 return t_ptr->offset(); 2258 } 2259 2260 Node* ConnectionGraph::get_addp_base(Node *addp) { 2261 assert(addp->is_AddP(), "must be AddP"); 2262 // 2263 // AddP cases for Base and Address inputs: 2264 // case #1. Direct object's field reference: 2265 // Allocate 2266 // | 2267 // Proj #5 ( oop result ) 2268 // | 2269 // CheckCastPP (cast to instance type) 2270 // | | 2271 // AddP ( base == address ) 2272 // 2273 // case #2. Indirect object's field reference: 2274 // Phi 2275 // | 2276 // CastPP (cast to instance type) 2277 // | | 2278 // AddP ( base == address ) 2279 // 2280 // case #3. Raw object's field reference for Initialize node: 2281 // Allocate 2282 // | 2283 // Proj #5 ( oop result ) 2284 // top | 2285 // \ | 2286 // AddP ( base == top ) 2287 // 2288 // case #4. Array's element reference: 2289 // {CheckCastPP | CastPP} 2290 // | | | 2291 // | AddP ( array's element offset ) 2292 // | | 2293 // AddP ( array's offset ) 2294 // 2295 // case #5. Raw object's field reference for arraycopy stub call: 2296 // The inline_native_clone() case when the arraycopy stub is called 2297 // after the allocation before Initialize and CheckCastPP nodes. 2298 // Allocate 2299 // | 2300 // Proj #5 ( oop result ) 2301 // | | 2302 // AddP ( base == address ) 2303 // 2304 // case #6. Constant Pool, ThreadLocal, CastX2P or 2305 // Raw object's field reference: 2306 // {ConP, ThreadLocal, CastX2P, raw Load} 2307 // top | 2308 // \ | 2309 // AddP ( base == top ) 2310 // 2311 // case #7. Klass's field reference. 2312 // LoadKlass 2313 // | | 2314 // AddP ( base == address ) 2315 // 2316 // case #8. narrow Klass's field reference. 2317 // LoadNKlass 2318 // | 2319 // DecodeN 2320 // | | 2321 // AddP ( base == address ) 2322 // 2323 // case #9. Mixed unsafe access 2324 // {instance} 2325 // | 2326 // CheckCastPP (raw) 2327 // top | 2328 // \ | 2329 // AddP ( base == top ) 2330 // 2331 Node *base = addp->in(AddPNode::Base); 2332 if (base->uncast()->is_top()) { // The AddP case #3 and #6 and #9. 2333 base = addp->in(AddPNode::Address); 2334 while (base->is_AddP()) { 2335 // Case #6 (unsafe access) may have several chained AddP nodes. 2336 assert(base->in(AddPNode::Base)->uncast()->is_top(), "expected unsafe access address only"); 2337 base = base->in(AddPNode::Address); 2338 } 2339 if (base->Opcode() == Op_CheckCastPP && 2340 base->bottom_type()->isa_rawptr() && 2341 _igvn->type(base->in(1))->isa_oopptr()) { 2342 base = base->in(1); // Case #9 2343 } else { 2344 Node* uncast_base = base->uncast(); 2345 int opcode = uncast_base->Opcode(); 2346 assert(opcode == Op_ConP || opcode == Op_ThreadLocal || 2347 opcode == Op_CastX2P || uncast_base->is_DecodeNarrowPtr() || 2348 (uncast_base->is_Mem() && (uncast_base->bottom_type()->isa_rawptr() != NULL)) || 2349 (uncast_base->is_Proj() && uncast_base->in(0)->is_Allocate()) || 2350 BarrierSet::barrier_set()->barrier_set_c2()->escape_is_barrier_node(uncast_base), "sanity"); 2351 } 2352 } 2353 return base; 2354 } 2355 2356 Node* ConnectionGraph::find_second_addp(Node* addp, Node* n) { 2357 assert(addp->is_AddP() && addp->outcnt() > 0, "Don't process dead nodes"); 2358 Node* addp2 = addp->raw_out(0); 2359 if (addp->outcnt() == 1 && addp2->is_AddP() && 2360 addp2->in(AddPNode::Base) == n && 2361 addp2->in(AddPNode::Address) == addp) { 2362 assert(addp->in(AddPNode::Base) == n, "expecting the same base"); 2363 // 2364 // Find array's offset to push it on worklist first and 2365 // as result process an array's element offset first (pushed second) 2366 // to avoid CastPP for the array's offset. 2367 // Otherwise the inserted CastPP (LocalVar) will point to what 2368 // the AddP (Field) points to. Which would be wrong since 2369 // the algorithm expects the CastPP has the same point as 2370 // as AddP's base CheckCastPP (LocalVar). 2371 // 2372 // ArrayAllocation 2373 // | 2374 // CheckCastPP 2375 // | 2376 // memProj (from ArrayAllocation CheckCastPP) 2377 // | || 2378 // | || Int (element index) 2379 // | || | ConI (log(element size)) 2380 // | || | / 2381 // | || LShift 2382 // | || / 2383 // | AddP (array's element offset) 2384 // | | 2385 // | | ConI (array's offset: #12(32-bits) or #24(64-bits)) 2386 // | / / 2387 // AddP (array's offset) 2388 // | 2389 // Load/Store (memory operation on array's element) 2390 // 2391 return addp2; 2392 } 2393 return NULL; 2394 } 2395 2396 // 2397 // Adjust the type and inputs of an AddP which computes the 2398 // address of a field of an instance 2399 // 2400 bool ConnectionGraph::split_AddP(Node *addp, Node *base) { 2401 PhaseGVN* igvn = _igvn; 2402 const TypeOopPtr *base_t = igvn->type(base)->isa_oopptr(); 2403 assert(base_t != NULL && base_t->is_known_instance(), "expecting instance oopptr"); 2404 const TypeOopPtr *t = igvn->type(addp)->isa_oopptr(); 2405 if (t == NULL) { 2406 // We are computing a raw address for a store captured by an Initialize 2407 // compute an appropriate address type (cases #3 and #5). 2408 assert(igvn->type(addp) == TypeRawPtr::NOTNULL, "must be raw pointer"); 2409 assert(addp->in(AddPNode::Address)->is_Proj(), "base of raw address must be result projection from allocation"); 2410 intptr_t offs = (int)igvn->find_intptr_t_con(addp->in(AddPNode::Offset), Type::OffsetBot); 2411 assert(offs != Type::OffsetBot, "offset must be a constant"); 2412 t = base_t->add_offset(offs)->is_oopptr(); 2413 } 2414 int inst_id = base_t->instance_id(); 2415 assert(!t->is_known_instance() || t->instance_id() == inst_id, 2416 "old type must be non-instance or match new type"); 2417 2418 // The type 't' could be subclass of 'base_t'. 2419 // As result t->offset() could be large then base_t's size and it will 2420 // cause the failure in add_offset() with narrow oops since TypeOopPtr() 2421 // constructor verifies correctness of the offset. 2422 // 2423 // It could happened on subclass's branch (from the type profiling 2424 // inlining) which was not eliminated during parsing since the exactness 2425 // of the allocation type was not propagated to the subclass type check. 2426 // 2427 // Or the type 't' could be not related to 'base_t' at all. 2428 // It could happened when CHA type is different from MDO type on a dead path 2429 // (for example, from instanceof check) which is not collapsed during parsing. 2430 // 2431 // Do nothing for such AddP node and don't process its users since 2432 // this code branch will go away. 2433 // 2434 if (!t->is_known_instance() && 2435 !base_t->klass()->is_subtype_of(t->klass())) { 2436 return false; // bail out 2437 } 2438 const TypeOopPtr *tinst = base_t->add_offset(t->offset())->is_oopptr(); 2439 // Do NOT remove the next line: ensure a new alias index is allocated 2440 // for the instance type. Note: C++ will not remove it since the call 2441 // has side effect. 2442 int alias_idx = _compile->get_alias_index(tinst); 2443 igvn->set_type(addp, tinst); 2444 // record the allocation in the node map 2445 set_map(addp, get_map(base->_idx)); 2446 // Set addp's Base and Address to 'base'. 2447 Node *abase = addp->in(AddPNode::Base); 2448 Node *adr = addp->in(AddPNode::Address); 2449 if (adr->is_Proj() && adr->in(0)->is_Allocate() && 2450 adr->in(0)->_idx == (uint)inst_id) { 2451 // Skip AddP cases #3 and #5. 2452 } else { 2453 assert(!abase->is_top(), "sanity"); // AddP case #3 2454 if (abase != base) { 2455 igvn->hash_delete(addp); 2456 addp->set_req(AddPNode::Base, base); 2457 if (abase == adr) { 2458 addp->set_req(AddPNode::Address, base); 2459 } else { 2460 // AddP case #4 (adr is array's element offset AddP node) 2461 #ifdef ASSERT 2462 const TypeOopPtr *atype = igvn->type(adr)->isa_oopptr(); 2463 assert(adr->is_AddP() && atype != NULL && 2464 atype->instance_id() == inst_id, "array's element offset should be processed first"); 2465 #endif 2466 } 2467 igvn->hash_insert(addp); 2468 } 2469 } 2470 // Put on IGVN worklist since at least addp's type was changed above. 2471 record_for_optimizer(addp); 2472 return true; 2473 } 2474 2475 // 2476 // Create a new version of orig_phi if necessary. Returns either the newly 2477 // created phi or an existing phi. Sets create_new to indicate whether a new 2478 // phi was created. Cache the last newly created phi in the node map. 2479 // 2480 PhiNode *ConnectionGraph::create_split_phi(PhiNode *orig_phi, int alias_idx, GrowableArray<PhiNode *> &orig_phi_worklist, bool &new_created) { 2481 Compile *C = _compile; 2482 PhaseGVN* igvn = _igvn; 2483 new_created = false; 2484 int phi_alias_idx = C->get_alias_index(orig_phi->adr_type()); 2485 // nothing to do if orig_phi is bottom memory or matches alias_idx 2486 if (phi_alias_idx == alias_idx) { 2487 return orig_phi; 2488 } 2489 // Have we recently created a Phi for this alias index? 2490 PhiNode *result = get_map_phi(orig_phi->_idx); 2491 if (result != NULL && C->get_alias_index(result->adr_type()) == alias_idx) { 2492 return result; 2493 } 2494 // Previous check may fail when the same wide memory Phi was split into Phis 2495 // for different memory slices. Search all Phis for this region. 2496 if (result != NULL) { 2497 Node* region = orig_phi->in(0); 2498 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) { 2499 Node* phi = region->fast_out(i); 2500 if (phi->is_Phi() && 2501 C->get_alias_index(phi->as_Phi()->adr_type()) == alias_idx) { 2502 assert(phi->_idx >= nodes_size(), "only new Phi per instance memory slice"); 2503 return phi->as_Phi(); 2504 } 2505 } 2506 } 2507 if (C->live_nodes() + 2*NodeLimitFudgeFactor > C->max_node_limit()) { 2508 if (C->do_escape_analysis() == true && !C->failing()) { 2509 // Retry compilation without escape analysis. 2510 // If this is the first failure, the sentinel string will "stick" 2511 // to the Compile object, and the C2Compiler will see it and retry. 2512 C->record_failure(C2Compiler::retry_no_escape_analysis()); 2513 } 2514 return NULL; 2515 } 2516 orig_phi_worklist.append_if_missing(orig_phi); 2517 const TypePtr *atype = C->get_adr_type(alias_idx); 2518 result = PhiNode::make(orig_phi->in(0), NULL, Type::MEMORY, atype); 2519 C->copy_node_notes_to(result, orig_phi); 2520 igvn->set_type(result, result->bottom_type()); 2521 record_for_optimizer(result); 2522 set_map(orig_phi, result); 2523 new_created = true; 2524 return result; 2525 } 2526 2527 // 2528 // Return a new version of Memory Phi "orig_phi" with the inputs having the 2529 // specified alias index. 2530 // 2531 PhiNode *ConnectionGraph::split_memory_phi(PhiNode *orig_phi, int alias_idx, GrowableArray<PhiNode *> &orig_phi_worklist) { 2532 assert(alias_idx != Compile::AliasIdxBot, "can't split out bottom memory"); 2533 Compile *C = _compile; 2534 PhaseGVN* igvn = _igvn; 2535 bool new_phi_created; 2536 PhiNode *result = create_split_phi(orig_phi, alias_idx, orig_phi_worklist, new_phi_created); 2537 if (!new_phi_created) { 2538 return result; 2539 } 2540 GrowableArray<PhiNode *> phi_list; 2541 GrowableArray<uint> cur_input; 2542 PhiNode *phi = orig_phi; 2543 uint idx = 1; 2544 bool finished = false; 2545 while(!finished) { 2546 while (idx < phi->req()) { 2547 Node *mem = find_inst_mem(phi->in(idx), alias_idx, orig_phi_worklist); 2548 if (mem != NULL && mem->is_Phi()) { 2549 PhiNode *newphi = create_split_phi(mem->as_Phi(), alias_idx, orig_phi_worklist, new_phi_created); 2550 if (new_phi_created) { 2551 // found an phi for which we created a new split, push current one on worklist and begin 2552 // processing new one 2553 phi_list.push(phi); 2554 cur_input.push(idx); 2555 phi = mem->as_Phi(); 2556 result = newphi; 2557 idx = 1; 2558 continue; 2559 } else { 2560 mem = newphi; 2561 } 2562 } 2563 if (C->failing()) { 2564 return NULL; 2565 } 2566 result->set_req(idx++, mem); 2567 } 2568 #ifdef ASSERT 2569 // verify that the new Phi has an input for each input of the original 2570 assert( phi->req() == result->req(), "must have same number of inputs."); 2571 assert( result->in(0) != NULL && result->in(0) == phi->in(0), "regions must match"); 2572 #endif 2573 // Check if all new phi's inputs have specified alias index. 2574 // Otherwise use old phi. 2575 for (uint i = 1; i < phi->req(); i++) { 2576 Node* in = result->in(i); 2577 assert((phi->in(i) == NULL) == (in == NULL), "inputs must correspond."); 2578 } 2579 // we have finished processing a Phi, see if there are any more to do 2580 finished = (phi_list.length() == 0 ); 2581 if (!finished) { 2582 phi = phi_list.pop(); 2583 idx = cur_input.pop(); 2584 PhiNode *prev_result = get_map_phi(phi->_idx); 2585 prev_result->set_req(idx++, result); 2586 result = prev_result; 2587 } 2588 } 2589 return result; 2590 } 2591 2592 // 2593 // The next methods are derived from methods in MemNode. 2594 // 2595 Node* ConnectionGraph::step_through_mergemem(MergeMemNode *mmem, int alias_idx, const TypeOopPtr *toop) { 2596 Node *mem = mmem; 2597 // TypeOopPtr::NOTNULL+any is an OOP with unknown offset - generally 2598 // means an array I have not precisely typed yet. Do not do any 2599 // alias stuff with it any time soon. 2600 if (toop->base() != Type::AnyPtr && 2601 !(toop->klass() != NULL && 2602 toop->klass()->is_java_lang_Object() && 2603 toop->offset() == Type::OffsetBot)) { 2604 mem = mmem->memory_at(alias_idx); 2605 // Update input if it is progress over what we have now 2606 } 2607 return mem; 2608 } 2609 2610 // 2611 // Move memory users to their memory slices. 2612 // 2613 void ConnectionGraph::move_inst_mem(Node* n, GrowableArray<PhiNode *> &orig_phis) { 2614 Compile* C = _compile; 2615 PhaseGVN* igvn = _igvn; 2616 const TypePtr* tp = igvn->type(n->in(MemNode::Address))->isa_ptr(); 2617 assert(tp != NULL, "ptr type"); 2618 int alias_idx = C->get_alias_index(tp); 2619 int general_idx = C->get_general_index(alias_idx); 2620 2621 // Move users first 2622 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { 2623 Node* use = n->fast_out(i); 2624 if (use->is_MergeMem()) { 2625 MergeMemNode* mmem = use->as_MergeMem(); 2626 assert(n == mmem->memory_at(alias_idx), "should be on instance memory slice"); 2627 if (n != mmem->memory_at(general_idx) || alias_idx == general_idx) { 2628 continue; // Nothing to do 2629 } 2630 // Replace previous general reference to mem node. 2631 uint orig_uniq = C->unique(); 2632 Node* m = find_inst_mem(n, general_idx, orig_phis); 2633 assert(orig_uniq == C->unique(), "no new nodes"); 2634 mmem->set_memory_at(general_idx, m); 2635 --imax; 2636 --i; 2637 } else if (use->is_MemBar()) { 2638 assert(!use->is_Initialize(), "initializing stores should not be moved"); 2639 if (use->req() > MemBarNode::Precedent && 2640 use->in(MemBarNode::Precedent) == n) { 2641 // Don't move related membars. 2642 record_for_optimizer(use); 2643 continue; 2644 } 2645 tp = use->as_MemBar()->adr_type()->isa_ptr(); 2646 if ((tp != NULL && C->get_alias_index(tp) == alias_idx) || 2647 alias_idx == general_idx) { 2648 continue; // Nothing to do 2649 } 2650 // Move to general memory slice. 2651 uint orig_uniq = C->unique(); 2652 Node* m = find_inst_mem(n, general_idx, orig_phis); 2653 assert(orig_uniq == C->unique(), "no new nodes"); 2654 igvn->hash_delete(use); 2655 imax -= use->replace_edge(n, m); 2656 igvn->hash_insert(use); 2657 record_for_optimizer(use); 2658 --i; 2659 #ifdef ASSERT 2660 } else if (use->is_Mem()) { 2661 if (use->Opcode() == Op_StoreCM && use->in(MemNode::OopStore) == n) { 2662 // Don't move related cardmark. 2663 continue; 2664 } 2665 // Memory nodes should have new memory input. 2666 tp = igvn->type(use->in(MemNode::Address))->isa_ptr(); 2667 assert(tp != NULL, "ptr type"); 2668 int idx = C->get_alias_index(tp); 2669 assert(get_map(use->_idx) != NULL || idx == alias_idx, 2670 "Following memory nodes should have new memory input or be on the same memory slice"); 2671 } else if (use->is_Phi()) { 2672 // Phi nodes should be split and moved already. 2673 tp = use->as_Phi()->adr_type()->isa_ptr(); 2674 assert(tp != NULL, "ptr type"); 2675 int idx = C->get_alias_index(tp); 2676 assert(idx == alias_idx, "Following Phi nodes should be on the same memory slice"); 2677 } else { 2678 use->dump(); 2679 assert(false, "should not be here"); 2680 #endif 2681 } 2682 } 2683 } 2684 2685 // 2686 // Search memory chain of "mem" to find a MemNode whose address 2687 // is the specified alias index. 2688 // 2689 Node* ConnectionGraph::find_inst_mem(Node *orig_mem, int alias_idx, GrowableArray<PhiNode *> &orig_phis) { 2690 if (orig_mem == NULL) 2691 return orig_mem; 2692 Compile* C = _compile; 2693 PhaseGVN* igvn = _igvn; 2694 const TypeOopPtr *toop = C->get_adr_type(alias_idx)->isa_oopptr(); 2695 bool is_instance = (toop != NULL) && toop->is_known_instance(); 2696 Node *start_mem = C->start()->proj_out_or_null(TypeFunc::Memory); 2697 Node *prev = NULL; 2698 Node *result = orig_mem; 2699 while (prev != result) { 2700 prev = result; 2701 if (result == start_mem) 2702 break; // hit one of our sentinels 2703 if (result->is_Mem()) { 2704 const Type *at = igvn->type(result->in(MemNode::Address)); 2705 if (at == Type::TOP) 2706 break; // Dead 2707 assert (at->isa_ptr() != NULL, "pointer type required."); 2708 int idx = C->get_alias_index(at->is_ptr()); 2709 if (idx == alias_idx) 2710 break; // Found 2711 if (!is_instance && (at->isa_oopptr() == NULL || 2712 !at->is_oopptr()->is_known_instance())) { 2713 break; // Do not skip store to general memory slice. 2714 } 2715 result = result->in(MemNode::Memory); 2716 } 2717 if (!is_instance) 2718 continue; // don't search further for non-instance types 2719 // skip over a call which does not affect this memory slice 2720 if (result->is_Proj() && result->as_Proj()->_con == TypeFunc::Memory) { 2721 Node *proj_in = result->in(0); 2722 if (proj_in->is_Allocate() && proj_in->_idx == (uint)toop->instance_id()) { 2723 break; // hit one of our sentinels 2724 } else if (proj_in->is_Call()) { 2725 // ArrayCopy node processed here as well 2726 CallNode *call = proj_in->as_Call(); 2727 if (!call->may_modify(toop, igvn)) { 2728 result = call->in(TypeFunc::Memory); 2729 } 2730 } else if (proj_in->is_Initialize()) { 2731 AllocateNode* alloc = proj_in->as_Initialize()->allocation(); 2732 // Stop if this is the initialization for the object instance which 2733 // which contains this memory slice, otherwise skip over it. 2734 if (alloc == NULL || alloc->_idx != (uint)toop->instance_id()) { 2735 result = proj_in->in(TypeFunc::Memory); 2736 } 2737 } else if (proj_in->is_MemBar()) { 2738 if (proj_in->in(TypeFunc::Memory)->is_MergeMem() && 2739 proj_in->in(TypeFunc::Memory)->as_MergeMem()->in(Compile::AliasIdxRaw)->is_Proj() && 2740 proj_in->in(TypeFunc::Memory)->as_MergeMem()->in(Compile::AliasIdxRaw)->in(0)->is_ArrayCopy()) { 2741 // clone 2742 ArrayCopyNode* ac = proj_in->in(TypeFunc::Memory)->as_MergeMem()->in(Compile::AliasIdxRaw)->in(0)->as_ArrayCopy(); 2743 if (ac->may_modify(toop, igvn)) { 2744 break; 2745 } 2746 } 2747 result = proj_in->in(TypeFunc::Memory); 2748 } 2749 } else if (result->is_MergeMem()) { 2750 MergeMemNode *mmem = result->as_MergeMem(); 2751 result = step_through_mergemem(mmem, alias_idx, toop); 2752 if (result == mmem->base_memory()) { 2753 // Didn't find instance memory, search through general slice recursively. 2754 result = mmem->memory_at(C->get_general_index(alias_idx)); 2755 result = find_inst_mem(result, alias_idx, orig_phis); 2756 if (C->failing()) { 2757 return NULL; 2758 } 2759 mmem->set_memory_at(alias_idx, result); 2760 } 2761 } else if (result->is_Phi() && 2762 C->get_alias_index(result->as_Phi()->adr_type()) != alias_idx) { 2763 Node *un = result->as_Phi()->unique_input(igvn); 2764 if (un != NULL) { 2765 orig_phis.append_if_missing(result->as_Phi()); 2766 result = un; 2767 } else { 2768 break; 2769 } 2770 } else if (result->is_ClearArray()) { 2771 if (!ClearArrayNode::step_through(&result, (uint)toop->instance_id(), igvn)) { 2772 // Can not bypass initialization of the instance 2773 // we are looking for. 2774 break; 2775 } 2776 // Otherwise skip it (the call updated 'result' value). 2777 } else if (result->Opcode() == Op_SCMemProj) { 2778 Node* mem = result->in(0); 2779 Node* adr = NULL; 2780 if (mem->is_LoadStore()) { 2781 adr = mem->in(MemNode::Address); 2782 } else { 2783 assert(mem->Opcode() == Op_EncodeISOArray || 2784 mem->Opcode() == Op_StrCompressedCopy, "sanity"); 2785 adr = mem->in(3); // Memory edge corresponds to destination array 2786 } 2787 const Type *at = igvn->type(adr); 2788 if (at != Type::TOP) { 2789 assert(at->isa_ptr() != NULL, "pointer type required."); 2790 int idx = C->get_alias_index(at->is_ptr()); 2791 if (idx == alias_idx) { 2792 // Assert in debug mode 2793 assert(false, "Object is not scalar replaceable if a LoadStore node accesses its field"); 2794 break; // In product mode return SCMemProj node 2795 } 2796 } 2797 result = mem->in(MemNode::Memory); 2798 } else if (result->Opcode() == Op_StrInflatedCopy) { 2799 Node* adr = result->in(3); // Memory edge corresponds to destination array 2800 const Type *at = igvn->type(adr); 2801 if (at != Type::TOP) { 2802 assert(at->isa_ptr() != NULL, "pointer type required."); 2803 int idx = C->get_alias_index(at->is_ptr()); 2804 if (idx == alias_idx) { 2805 // Assert in debug mode 2806 assert(false, "Object is not scalar replaceable if a StrInflatedCopy node accesses its field"); 2807 break; // In product mode return SCMemProj node 2808 } 2809 } 2810 result = result->in(MemNode::Memory); 2811 } 2812 } 2813 if (result->is_Phi()) { 2814 PhiNode *mphi = result->as_Phi(); 2815 assert(mphi->bottom_type() == Type::MEMORY, "memory phi required"); 2816 const TypePtr *t = mphi->adr_type(); 2817 if (!is_instance) { 2818 // Push all non-instance Phis on the orig_phis worklist to update inputs 2819 // during Phase 4 if needed. 2820 orig_phis.append_if_missing(mphi); 2821 } else if (C->get_alias_index(t) != alias_idx) { 2822 // Create a new Phi with the specified alias index type. 2823 result = split_memory_phi(mphi, alias_idx, orig_phis); 2824 } 2825 } 2826 // the result is either MemNode, PhiNode, InitializeNode. 2827 return result; 2828 } 2829 2830 // 2831 // Convert the types of unescaped object to instance types where possible, 2832 // propagate the new type information through the graph, and update memory 2833 // edges and MergeMem inputs to reflect the new type. 2834 // 2835 // We start with allocations (and calls which may be allocations) on alloc_worklist. 2836 // The processing is done in 4 phases: 2837 // 2838 // Phase 1: Process possible allocations from alloc_worklist. Create instance 2839 // types for the CheckCastPP for allocations where possible. 2840 // Propagate the new types through users as follows: 2841 // casts and Phi: push users on alloc_worklist 2842 // AddP: cast Base and Address inputs to the instance type 2843 // push any AddP users on alloc_worklist and push any memnode 2844 // users onto memnode_worklist. 2845 // Phase 2: Process MemNode's from memnode_worklist. compute new address type and 2846 // search the Memory chain for a store with the appropriate type 2847 // address type. If a Phi is found, create a new version with 2848 // the appropriate memory slices from each of the Phi inputs. 2849 // For stores, process the users as follows: 2850 // MemNode: push on memnode_worklist 2851 // MergeMem: push on mergemem_worklist 2852 // Phase 3: Process MergeMem nodes from mergemem_worklist. Walk each memory slice 2853 // moving the first node encountered of each instance type to the 2854 // the input corresponding to its alias index. 2855 // appropriate memory slice. 2856 // Phase 4: Update the inputs of non-instance memory Phis and the Memory input of memnodes. 2857 // 2858 // In the following example, the CheckCastPP nodes are the cast of allocation 2859 // results and the allocation of node 29 is unescaped and eligible to be an 2860 // instance type. 2861 // 2862 // We start with: 2863 // 2864 // 7 Parm #memory 2865 // 10 ConI "12" 2866 // 19 CheckCastPP "Foo" 2867 // 20 AddP _ 19 19 10 Foo+12 alias_index=4 2868 // 29 CheckCastPP "Foo" 2869 // 30 AddP _ 29 29 10 Foo+12 alias_index=4 2870 // 2871 // 40 StoreP 25 7 20 ... alias_index=4 2872 // 50 StoreP 35 40 30 ... alias_index=4 2873 // 60 StoreP 45 50 20 ... alias_index=4 2874 // 70 LoadP _ 60 30 ... alias_index=4 2875 // 80 Phi 75 50 60 Memory alias_index=4 2876 // 90 LoadP _ 80 30 ... alias_index=4 2877 // 100 LoadP _ 80 20 ... alias_index=4 2878 // 2879 // 2880 // Phase 1 creates an instance type for node 29 assigning it an instance id of 24 2881 // and creating a new alias index for node 30. This gives: 2882 // 2883 // 7 Parm #memory 2884 // 10 ConI "12" 2885 // 19 CheckCastPP "Foo" 2886 // 20 AddP _ 19 19 10 Foo+12 alias_index=4 2887 // 29 CheckCastPP "Foo" iid=24 2888 // 30 AddP _ 29 29 10 Foo+12 alias_index=6 iid=24 2889 // 2890 // 40 StoreP 25 7 20 ... alias_index=4 2891 // 50 StoreP 35 40 30 ... alias_index=6 2892 // 60 StoreP 45 50 20 ... alias_index=4 2893 // 70 LoadP _ 60 30 ... alias_index=6 2894 // 80 Phi 75 50 60 Memory alias_index=4 2895 // 90 LoadP _ 80 30 ... alias_index=6 2896 // 100 LoadP _ 80 20 ... alias_index=4 2897 // 2898 // In phase 2, new memory inputs are computed for the loads and stores, 2899 // And a new version of the phi is created. In phase 4, the inputs to 2900 // node 80 are updated and then the memory nodes are updated with the 2901 // values computed in phase 2. This results in: 2902 // 2903 // 7 Parm #memory 2904 // 10 ConI "12" 2905 // 19 CheckCastPP "Foo" 2906 // 20 AddP _ 19 19 10 Foo+12 alias_index=4 2907 // 29 CheckCastPP "Foo" iid=24 2908 // 30 AddP _ 29 29 10 Foo+12 alias_index=6 iid=24 2909 // 2910 // 40 StoreP 25 7 20 ... alias_index=4 2911 // 50 StoreP 35 7 30 ... alias_index=6 2912 // 60 StoreP 45 40 20 ... alias_index=4 2913 // 70 LoadP _ 50 30 ... alias_index=6 2914 // 80 Phi 75 40 60 Memory alias_index=4 2915 // 120 Phi 75 50 50 Memory alias_index=6 2916 // 90 LoadP _ 120 30 ... alias_index=6 2917 // 100 LoadP _ 80 20 ... alias_index=4 2918 // 2919 void ConnectionGraph::split_unique_types(GrowableArray<Node *> &alloc_worklist, GrowableArray<ArrayCopyNode*> &arraycopy_worklist) { 2920 GrowableArray<Node *> memnode_worklist; 2921 GrowableArray<PhiNode *> orig_phis; 2922 PhaseIterGVN *igvn = _igvn; 2923 uint new_index_start = (uint) _compile->num_alias_types(); 2924 Arena* arena = Thread::current()->resource_area(); 2925 VectorSet visited(arena); 2926 ideal_nodes.clear(); // Reset for use with set_map/get_map. 2927 uint unique_old = _compile->unique(); 2928 2929 // Phase 1: Process possible allocations from alloc_worklist. 2930 // Create instance types for the CheckCastPP for allocations where possible. 2931 // 2932 // (Note: don't forget to change the order of the second AddP node on 2933 // the alloc_worklist if the order of the worklist processing is changed, 2934 // see the comment in find_second_addp().) 2935 // 2936 while (alloc_worklist.length() != 0) { 2937 Node *n = alloc_worklist.pop(); 2938 uint ni = n->_idx; 2939 if (n->is_Call()) { 2940 CallNode *alloc = n->as_Call(); 2941 // copy escape information to call node 2942 PointsToNode* ptn = ptnode_adr(alloc->_idx); 2943 PointsToNode::EscapeState es = ptn->escape_state(); 2944 // We have an allocation or call which returns a Java object, 2945 // see if it is unescaped. 2946 if (es != PointsToNode::NoEscape || !ptn->scalar_replaceable()) 2947 continue; 2948 // Find CheckCastPP for the allocate or for the return value of a call 2949 n = alloc->result_cast(); 2950 if (n == NULL) { // No uses except Initialize node 2951 if (alloc->is_Allocate()) { 2952 // Set the scalar_replaceable flag for allocation 2953 // so it could be eliminated if it has no uses. 2954 alloc->as_Allocate()->_is_scalar_replaceable = true; 2955 } 2956 if (alloc->is_CallStaticJava()) { 2957 // Set the scalar_replaceable flag for boxing method 2958 // so it could be eliminated if it has no uses. 2959 alloc->as_CallStaticJava()->_is_scalar_replaceable = true; 2960 } 2961 continue; 2962 } 2963 if (!n->is_CheckCastPP()) { // not unique CheckCastPP. 2964 assert(!alloc->is_Allocate(), "allocation should have unique type"); 2965 continue; 2966 } 2967 2968 // The inline code for Object.clone() casts the allocation result to 2969 // java.lang.Object and then to the actual type of the allocated 2970 // object. Detect this case and use the second cast. 2971 // Also detect j.l.reflect.Array.newInstance(jobject, jint) case when 2972 // the allocation result is cast to java.lang.Object and then 2973 // to the actual Array type. 2974 if (alloc->is_Allocate() && n->as_Type()->type() == TypeInstPtr::NOTNULL 2975 && (alloc->is_AllocateArray() || 2976 igvn->type(alloc->in(AllocateNode::KlassNode)) != TypeKlassPtr::OBJECT)) { 2977 Node *cast2 = NULL; 2978 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { 2979 Node *use = n->fast_out(i); 2980 if (use->is_CheckCastPP()) { 2981 cast2 = use; 2982 break; 2983 } 2984 } 2985 if (cast2 != NULL) { 2986 n = cast2; 2987 } else { 2988 // Non-scalar replaceable if the allocation type is unknown statically 2989 // (reflection allocation), the object can't be restored during 2990 // deoptimization without precise type. 2991 continue; 2992 } 2993 } 2994 2995 const TypeOopPtr *t = igvn->type(n)->isa_oopptr(); 2996 if (t == NULL) 2997 continue; // not a TypeOopPtr 2998 if (!t->klass_is_exact()) 2999 continue; // not an unique type 3000 3001 if (alloc->is_Allocate()) { 3002 // Set the scalar_replaceable flag for allocation 3003 // so it could be eliminated. 3004 alloc->as_Allocate()->_is_scalar_replaceable = true; 3005 } 3006 if (alloc->is_CallStaticJava()) { 3007 // Set the scalar_replaceable flag for boxing method 3008 // so it could be eliminated. 3009 alloc->as_CallStaticJava()->_is_scalar_replaceable = true; 3010 } 3011 set_escape_state(ptnode_adr(n->_idx), es); // CheckCastPP escape state 3012 // in order for an object to be scalar-replaceable, it must be: 3013 // - a direct allocation (not a call returning an object) 3014 // - non-escaping 3015 // - eligible to be a unique type 3016 // - not determined to be ineligible by escape analysis 3017 set_map(alloc, n); 3018 set_map(n, alloc); 3019 const TypeOopPtr* tinst = t->cast_to_instance_id(ni); 3020 igvn->hash_delete(n); 3021 igvn->set_type(n, tinst); 3022 n->raise_bottom_type(tinst); 3023 igvn->hash_insert(n); 3024 record_for_optimizer(n); 3025 // Allocate an alias index for the header fields. Accesses to 3026 // the header emitted during macro expansion wouldn't have 3027 // correct memory state otherwise. 3028 _compile->get_alias_index(tinst->add_offset(oopDesc::mark_offset_in_bytes())); 3029 _compile->get_alias_index(tinst->add_offset(oopDesc::klass_offset_in_bytes())); 3030 if (alloc->is_Allocate() && (t->isa_instptr() || t->isa_aryptr())) { 3031 3032 // First, put on the worklist all Field edges from Connection Graph 3033 // which is more accurate than putting immediate users from Ideal Graph. 3034 for (EdgeIterator e(ptn); e.has_next(); e.next()) { 3035 PointsToNode* tgt = e.get(); 3036 if (tgt->is_Arraycopy()) { 3037 continue; 3038 } 3039 Node* use = tgt->ideal_node(); 3040 assert(tgt->is_Field() && use->is_AddP(), 3041 "only AddP nodes are Field edges in CG"); 3042 if (use->outcnt() > 0) { // Don't process dead nodes 3043 Node* addp2 = find_second_addp(use, use->in(AddPNode::Base)); 3044 if (addp2 != NULL) { 3045 assert(alloc->is_AllocateArray(),"array allocation was expected"); 3046 alloc_worklist.append_if_missing(addp2); 3047 } 3048 alloc_worklist.append_if_missing(use); 3049 } 3050 } 3051 3052 // An allocation may have an Initialize which has raw stores. Scan 3053 // the users of the raw allocation result and push AddP users 3054 // on alloc_worklist. 3055 Node *raw_result = alloc->proj_out_or_null(TypeFunc::Parms); 3056 assert (raw_result != NULL, "must have an allocation result"); 3057 for (DUIterator_Fast imax, i = raw_result->fast_outs(imax); i < imax; i++) { 3058 Node *use = raw_result->fast_out(i); 3059 if (use->is_AddP() && use->outcnt() > 0) { // Don't process dead nodes 3060 Node* addp2 = find_second_addp(use, raw_result); 3061 if (addp2 != NULL) { 3062 assert(alloc->is_AllocateArray(),"array allocation was expected"); 3063 alloc_worklist.append_if_missing(addp2); 3064 } 3065 alloc_worklist.append_if_missing(use); 3066 } else if (use->is_MemBar()) { 3067 memnode_worklist.append_if_missing(use); 3068 } 3069 } 3070 } 3071 } else if (n->is_AddP()) { 3072 JavaObjectNode* jobj = unique_java_object(get_addp_base(n)); 3073 if (jobj == NULL || jobj == phantom_obj) { 3074 #ifdef ASSERT 3075 ptnode_adr(get_addp_base(n)->_idx)->dump(); 3076 ptnode_adr(n->_idx)->dump(); 3077 assert(jobj != NULL && jobj != phantom_obj, "escaped allocation"); 3078 #endif 3079 _compile->record_failure(C2Compiler::retry_no_escape_analysis()); 3080 return; 3081 } 3082 Node *base = get_map(jobj->idx()); // CheckCastPP node 3083 if (!split_AddP(n, base)) continue; // wrong type from dead path 3084 } else if (n->is_Phi() || 3085 n->is_CheckCastPP() || 3086 n->is_EncodeP() || 3087 n->is_DecodeN() || 3088 BarrierSet::barrier_set()->barrier_set_c2()->escape_is_barrier_node(n) || 3089 (n->is_ConstraintCast() && n->Opcode() == Op_CastPP)) { 3090 if (visited.test_set(n->_idx)) { 3091 assert(n->is_Phi(), "loops only through Phi's"); 3092 continue; // already processed 3093 } 3094 JavaObjectNode* jobj = unique_java_object(n); 3095 if (jobj == NULL || jobj == phantom_obj) { 3096 #ifdef ASSERT 3097 ptnode_adr(n->_idx)->dump(); 3098 assert(jobj != NULL && jobj != phantom_obj, "escaped allocation"); 3099 #endif 3100 _compile->record_failure(C2Compiler::retry_no_escape_analysis()); 3101 return; 3102 } else { 3103 Node *val = get_map(jobj->idx()); // CheckCastPP node 3104 TypeNode *tn = n->as_Type(); 3105 const TypeOopPtr* tinst = igvn->type(val)->isa_oopptr(); 3106 assert(tinst != NULL && tinst->is_known_instance() && 3107 tinst->instance_id() == jobj->idx() , "instance type expected."); 3108 3109 const Type *tn_type = igvn->type(tn); 3110 const TypeOopPtr *tn_t; 3111 if (tn_type->isa_narrowoop()) { 3112 tn_t = tn_type->make_ptr()->isa_oopptr(); 3113 } else { 3114 tn_t = tn_type->isa_oopptr(); 3115 } 3116 if (tn_t != NULL && tinst->klass()->is_subtype_of(tn_t->klass())) { 3117 if (tn_type->isa_narrowoop()) { 3118 tn_type = tinst->make_narrowoop(); 3119 } else { 3120 tn_type = tinst; 3121 } 3122 igvn->hash_delete(tn); 3123 igvn->set_type(tn, tn_type); 3124 tn->set_type(tn_type); 3125 igvn->hash_insert(tn); 3126 record_for_optimizer(n); 3127 } else { 3128 assert(tn_type == TypePtr::NULL_PTR || 3129 tn_t != NULL && !tinst->klass()->is_subtype_of(tn_t->klass()), 3130 "unexpected type"); 3131 continue; // Skip dead path with different type 3132 } 3133 } 3134 } else { 3135 debug_only(n->dump();) 3136 assert(false, "EA: unexpected node"); 3137 continue; 3138 } 3139 // push allocation's users on appropriate worklist 3140 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { 3141 Node *use = n->fast_out(i); 3142 if(use->is_Mem() && use->in(MemNode::Address) == n) { 3143 // Load/store to instance's field 3144 memnode_worklist.append_if_missing(use); 3145 } else if (use->is_MemBar()) { 3146 if (use->in(TypeFunc::Memory) == n) { // Ignore precedent edge 3147 memnode_worklist.append_if_missing(use); 3148 } 3149 } else if (use->is_AddP() && use->outcnt() > 0) { // No dead nodes 3150 Node* addp2 = find_second_addp(use, n); 3151 if (addp2 != NULL) { 3152 alloc_worklist.append_if_missing(addp2); 3153 } 3154 alloc_worklist.append_if_missing(use); 3155 } else if (use->is_Phi() || 3156 use->is_CheckCastPP() || 3157 use->is_EncodeNarrowPtr() || 3158 use->is_DecodeNarrowPtr() || 3159 BarrierSet::barrier_set()->barrier_set_c2()->escape_is_barrier_node(use) || 3160 (use->is_ConstraintCast() && use->Opcode() == Op_CastPP)) { 3161 alloc_worklist.append_if_missing(use); 3162 #ifdef ASSERT 3163 } else if (use->is_Mem()) { 3164 assert(use->in(MemNode::Address) != n, "EA: missing allocation reference path"); 3165 } else if (use->is_MergeMem()) { 3166 assert(_mergemem_worklist.contains(use->as_MergeMem()), "EA: missing MergeMem node in the worklist"); 3167 } else if (use->is_SafePoint()) { 3168 // Look for MergeMem nodes for calls which reference unique allocation 3169 // (through CheckCastPP nodes) even for debug info. 3170 Node* m = use->in(TypeFunc::Memory); 3171 if (m->is_MergeMem()) { 3172 assert(_mergemem_worklist.contains(m->as_MergeMem()), "EA: missing MergeMem node in the worklist"); 3173 } 3174 } else if (use->Opcode() == Op_EncodeISOArray) { 3175 if (use->in(MemNode::Memory) == n || use->in(3) == n) { 3176 // EncodeISOArray overwrites destination array 3177 memnode_worklist.append_if_missing(use); 3178 } 3179 } else { 3180 uint op = use->Opcode(); 3181 if ((op == Op_StrCompressedCopy || op == Op_StrInflatedCopy) && 3182 (use->in(MemNode::Memory) == n)) { 3183 // They overwrite memory edge corresponding to destination array, 3184 memnode_worklist.append_if_missing(use); 3185 } else if (!(op == Op_CmpP || op == Op_Conv2B || 3186 op == Op_CastP2X || op == Op_StoreCM || 3187 op == Op_FastLock || op == Op_AryEq || op == Op_StrComp || op == Op_HasNegatives || 3188 op == Op_StrCompressedCopy || op == Op_StrInflatedCopy || 3189 op == Op_StrEquals || op == Op_StrIndexOf || op == Op_StrIndexOfChar || 3190 BarrierSet::barrier_set()->barrier_set_c2()->is_gc_barrier_node(use))) { 3191 n->dump(); 3192 use->dump(); 3193 assert(false, "EA: missing allocation reference path"); 3194 } 3195 #endif 3196 } 3197 } 3198 3199 } 3200 3201 // Go over all ArrayCopy nodes and if one of the inputs has a unique 3202 // type, record it in the ArrayCopy node so we know what memory this 3203 // node uses/modified. 3204 for (int next = 0; next < arraycopy_worklist.length(); next++) { 3205 ArrayCopyNode* ac = arraycopy_worklist.at(next); 3206 Node* dest = ac->in(ArrayCopyNode::Dest); 3207 if (dest->is_AddP()) { 3208 dest = get_addp_base(dest); 3209 } 3210 JavaObjectNode* jobj = unique_java_object(dest); 3211 if (jobj != NULL) { 3212 Node *base = get_map(jobj->idx()); 3213 if (base != NULL) { 3214 const TypeOopPtr *base_t = _igvn->type(base)->isa_oopptr(); 3215 ac->_dest_type = base_t; 3216 } 3217 } 3218 Node* src = ac->in(ArrayCopyNode::Src); 3219 if (src->is_AddP()) { 3220 src = get_addp_base(src); 3221 } 3222 jobj = unique_java_object(src); 3223 if (jobj != NULL) { 3224 Node* base = get_map(jobj->idx()); 3225 if (base != NULL) { 3226 const TypeOopPtr *base_t = _igvn->type(base)->isa_oopptr(); 3227 ac->_src_type = base_t; 3228 } 3229 } 3230 } 3231 3232 // New alias types were created in split_AddP(). 3233 uint new_index_end = (uint) _compile->num_alias_types(); 3234 assert(unique_old == _compile->unique(), "there should be no new ideal nodes after Phase 1"); 3235 3236 // Phase 2: Process MemNode's from memnode_worklist. compute new address type and 3237 // compute new values for Memory inputs (the Memory inputs are not 3238 // actually updated until phase 4.) 3239 if (memnode_worklist.length() == 0) 3240 return; // nothing to do 3241 while (memnode_worklist.length() != 0) { 3242 Node *n = memnode_worklist.pop(); 3243 if (visited.test_set(n->_idx)) 3244 continue; 3245 if (n->is_Phi() || n->is_ClearArray()) { 3246 // we don't need to do anything, but the users must be pushed 3247 } else if (n->is_MemBar()) { // Initialize, MemBar nodes 3248 // we don't need to do anything, but the users must be pushed 3249 n = n->as_MemBar()->proj_out_or_null(TypeFunc::Memory); 3250 if (n == NULL) 3251 continue; 3252 } else if (n->Opcode() == Op_StrCompressedCopy || 3253 n->Opcode() == Op_EncodeISOArray) { 3254 // get the memory projection 3255 n = n->find_out_with(Op_SCMemProj); 3256 assert(n != NULL && n->Opcode() == Op_SCMemProj, "memory projection required"); 3257 } else { 3258 assert(n->is_Mem(), "memory node required."); 3259 Node *addr = n->in(MemNode::Address); 3260 const Type *addr_t = igvn->type(addr); 3261 if (addr_t == Type::TOP) 3262 continue; 3263 assert (addr_t->isa_ptr() != NULL, "pointer type required."); 3264 int alias_idx = _compile->get_alias_index(addr_t->is_ptr()); 3265 assert ((uint)alias_idx < new_index_end, "wrong alias index"); 3266 Node *mem = find_inst_mem(n->in(MemNode::Memory), alias_idx, orig_phis); 3267 if (_compile->failing()) { 3268 return; 3269 } 3270 if (mem != n->in(MemNode::Memory)) { 3271 // We delay the memory edge update since we need old one in 3272 // MergeMem code below when instances memory slices are separated. 3273 set_map(n, mem); 3274 } 3275 if (n->is_Load()) { 3276 continue; // don't push users 3277 } else if (n->is_LoadStore()) { 3278 // get the memory projection 3279 n = n->find_out_with(Op_SCMemProj); 3280 assert(n != NULL && n->Opcode() == Op_SCMemProj, "memory projection required"); 3281 } 3282 } 3283 // push user on appropriate worklist 3284 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { 3285 Node *use = n->fast_out(i); 3286 if (use->is_Phi() || use->is_ClearArray()) { 3287 memnode_worklist.append_if_missing(use); 3288 } else if (use->is_Mem() && use->in(MemNode::Memory) == n) { 3289 if (use->Opcode() == Op_StoreCM) // Ignore cardmark stores 3290 continue; 3291 memnode_worklist.append_if_missing(use); 3292 } else if (use->is_MemBar()) { 3293 if (use->in(TypeFunc::Memory) == n) { // Ignore precedent edge 3294 memnode_worklist.append_if_missing(use); 3295 } 3296 #ifdef ASSERT 3297 } else if(use->is_Mem()) { 3298 assert(use->in(MemNode::Memory) != n, "EA: missing memory path"); 3299 } else if (use->is_MergeMem()) { 3300 assert(_mergemem_worklist.contains(use->as_MergeMem()), "EA: missing MergeMem node in the worklist"); 3301 } else if (use->Opcode() == Op_EncodeISOArray) { 3302 if (use->in(MemNode::Memory) == n || use->in(3) == n) { 3303 // EncodeISOArray overwrites destination array 3304 memnode_worklist.append_if_missing(use); 3305 } 3306 } else { 3307 uint op = use->Opcode(); 3308 if ((use->in(MemNode::Memory) == n) && 3309 (op == Op_StrCompressedCopy || op == Op_StrInflatedCopy)) { 3310 // They overwrite memory edge corresponding to destination array, 3311 memnode_worklist.append_if_missing(use); 3312 } else if (!(BarrierSet::barrier_set()->barrier_set_c2()->is_gc_barrier_node(use) || 3313 op == Op_AryEq || op == Op_StrComp || op == Op_HasNegatives || 3314 op == Op_StrCompressedCopy || op == Op_StrInflatedCopy || 3315 op == Op_StrEquals || op == Op_StrIndexOf || op == Op_StrIndexOfChar)) { 3316 n->dump(); 3317 use->dump(); 3318 assert(false, "EA: missing memory path"); 3319 } 3320 #endif 3321 } 3322 } 3323 } 3324 3325 // Phase 3: Process MergeMem nodes from mergemem_worklist. 3326 // Walk each memory slice moving the first node encountered of each 3327 // instance type to the the input corresponding to its alias index. 3328 uint length = _mergemem_worklist.length(); 3329 for( uint next = 0; next < length; ++next ) { 3330 MergeMemNode* nmm = _mergemem_worklist.at(next); 3331 assert(!visited.test_set(nmm->_idx), "should not be visited before"); 3332 // Note: we don't want to use MergeMemStream here because we only want to 3333 // scan inputs which exist at the start, not ones we add during processing. 3334 // Note 2: MergeMem may already contains instance memory slices added 3335 // during find_inst_mem() call when memory nodes were processed above. 3336 igvn->hash_delete(nmm); 3337 uint nslices = MIN2(nmm->req(), new_index_start); 3338 for (uint i = Compile::AliasIdxRaw+1; i < nslices; i++) { 3339 Node* mem = nmm->in(i); 3340 Node* cur = NULL; 3341 if (mem == NULL || mem->is_top()) 3342 continue; 3343 // First, update mergemem by moving memory nodes to corresponding slices 3344 // if their type became more precise since this mergemem was created. 3345 while (mem->is_Mem()) { 3346 const Type *at = igvn->type(mem->in(MemNode::Address)); 3347 if (at != Type::TOP) { 3348 assert (at->isa_ptr() != NULL, "pointer type required."); 3349 uint idx = (uint)_compile->get_alias_index(at->is_ptr()); 3350 if (idx == i) { 3351 if (cur == NULL) 3352 cur = mem; 3353 } else { 3354 if (idx >= nmm->req() || nmm->is_empty_memory(nmm->in(idx))) { 3355 nmm->set_memory_at(idx, mem); 3356 } 3357 } 3358 } 3359 mem = mem->in(MemNode::Memory); 3360 } 3361 nmm->set_memory_at(i, (cur != NULL) ? cur : mem); 3362 // Find any instance of the current type if we haven't encountered 3363 // already a memory slice of the instance along the memory chain. 3364 for (uint ni = new_index_start; ni < new_index_end; ni++) { 3365 if((uint)_compile->get_general_index(ni) == i) { 3366 Node *m = (ni >= nmm->req()) ? nmm->empty_memory() : nmm->in(ni); 3367 if (nmm->is_empty_memory(m)) { 3368 Node* result = find_inst_mem(mem, ni, orig_phis); 3369 if (_compile->failing()) { 3370 return; 3371 } 3372 nmm->set_memory_at(ni, result); 3373 } 3374 } 3375 } 3376 } 3377 // Find the rest of instances values 3378 for (uint ni = new_index_start; ni < new_index_end; ni++) { 3379 const TypeOopPtr *tinst = _compile->get_adr_type(ni)->isa_oopptr(); 3380 Node* result = step_through_mergemem(nmm, ni, tinst); 3381 if (result == nmm->base_memory()) { 3382 // Didn't find instance memory, search through general slice recursively. 3383 result = nmm->memory_at(_compile->get_general_index(ni)); 3384 result = find_inst_mem(result, ni, orig_phis); 3385 if (_compile->failing()) { 3386 return; 3387 } 3388 nmm->set_memory_at(ni, result); 3389 } 3390 } 3391 igvn->hash_insert(nmm); 3392 record_for_optimizer(nmm); 3393 } 3394 3395 // Phase 4: Update the inputs of non-instance memory Phis and 3396 // the Memory input of memnodes 3397 // First update the inputs of any non-instance Phi's from 3398 // which we split out an instance Phi. Note we don't have 3399 // to recursively process Phi's encounted on the input memory 3400 // chains as is done in split_memory_phi() since they will 3401 // also be processed here. 3402 for (int j = 0; j < orig_phis.length(); j++) { 3403 PhiNode *phi = orig_phis.at(j); 3404 int alias_idx = _compile->get_alias_index(phi->adr_type()); 3405 igvn->hash_delete(phi); 3406 for (uint i = 1; i < phi->req(); i++) { 3407 Node *mem = phi->in(i); 3408 Node *new_mem = find_inst_mem(mem, alias_idx, orig_phis); 3409 if (_compile->failing()) { 3410 return; 3411 } 3412 if (mem != new_mem) { 3413 phi->set_req(i, new_mem); 3414 } 3415 } 3416 igvn->hash_insert(phi); 3417 record_for_optimizer(phi); 3418 } 3419 3420 // Update the memory inputs of MemNodes with the value we computed 3421 // in Phase 2 and move stores memory users to corresponding memory slices. 3422 // Disable memory split verification code until the fix for 6984348. 3423 // Currently it produces false negative results since it does not cover all cases. 3424 #if 0 // ifdef ASSERT 3425 visited.Reset(); 3426 Node_Stack old_mems(arena, _compile->unique() >> 2); 3427 #endif 3428 for (uint i = 0; i < ideal_nodes.size(); i++) { 3429 Node* n = ideal_nodes.at(i); 3430 Node* nmem = get_map(n->_idx); 3431 assert(nmem != NULL, "sanity"); 3432 if (n->is_Mem()) { 3433 #if 0 // ifdef ASSERT 3434 Node* old_mem = n->in(MemNode::Memory); 3435 if (!visited.test_set(old_mem->_idx)) { 3436 old_mems.push(old_mem, old_mem->outcnt()); 3437 } 3438 #endif 3439 assert(n->in(MemNode::Memory) != nmem, "sanity"); 3440 if (!n->is_Load()) { 3441 // Move memory users of a store first. 3442 move_inst_mem(n, orig_phis); 3443 } 3444 // Now update memory input 3445 igvn->hash_delete(n); 3446 n->set_req(MemNode::Memory, nmem); 3447 igvn->hash_insert(n); 3448 record_for_optimizer(n); 3449 } else { 3450 assert(n->is_Allocate() || n->is_CheckCastPP() || 3451 n->is_AddP() || n->is_Phi(), "unknown node used for set_map()"); 3452 } 3453 } 3454 #if 0 // ifdef ASSERT 3455 // Verify that memory was split correctly 3456 while (old_mems.is_nonempty()) { 3457 Node* old_mem = old_mems.node(); 3458 uint old_cnt = old_mems.index(); 3459 old_mems.pop(); 3460 assert(old_cnt == old_mem->outcnt(), "old mem could be lost"); 3461 } 3462 #endif 3463 } 3464 3465 #ifndef PRODUCT 3466 static const char *node_type_names[] = { 3467 "UnknownType", 3468 "JavaObject", 3469 "LocalVar", 3470 "Field", 3471 "Arraycopy" 3472 }; 3473 3474 static const char *esc_names[] = { 3475 "UnknownEscape", 3476 "NoEscape", 3477 "ArgEscape", 3478 "GlobalEscape" 3479 }; 3480 3481 void PointsToNode::dump(bool print_state) const { 3482 NodeType nt = node_type(); 3483 tty->print("%s ", node_type_names[(int) nt]); 3484 if (print_state) { 3485 EscapeState es = escape_state(); 3486 EscapeState fields_es = fields_escape_state(); 3487 tty->print("%s(%s) ", esc_names[(int)es], esc_names[(int)fields_es]); 3488 if (nt == PointsToNode::JavaObject && !this->scalar_replaceable()) 3489 tty->print("NSR "); 3490 } 3491 if (is_Field()) { 3492 FieldNode* f = (FieldNode*)this; 3493 if (f->is_oop()) 3494 tty->print("oop "); 3495 if (f->offset() > 0) 3496 tty->print("+%d ", f->offset()); 3497 tty->print("("); 3498 for (BaseIterator i(f); i.has_next(); i.next()) { 3499 PointsToNode* b = i.get(); 3500 tty->print(" %d%s", b->idx(),(b->is_JavaObject() ? "P" : "")); 3501 } 3502 tty->print(" )"); 3503 } 3504 tty->print("["); 3505 for (EdgeIterator i(this); i.has_next(); i.next()) { 3506 PointsToNode* e = i.get(); 3507 tty->print(" %d%s%s", e->idx(),(e->is_JavaObject() ? "P" : (e->is_Field() ? "F" : "")), e->is_Arraycopy() ? "cp" : ""); 3508 } 3509 tty->print(" ["); 3510 for (UseIterator i(this); i.has_next(); i.next()) { 3511 PointsToNode* u = i.get(); 3512 bool is_base = false; 3513 if (PointsToNode::is_base_use(u)) { 3514 is_base = true; 3515 u = PointsToNode::get_use_node(u)->as_Field(); 3516 } 3517 tty->print(" %d%s%s", u->idx(), is_base ? "b" : "", u->is_Arraycopy() ? "cp" : ""); 3518 } 3519 tty->print(" ]] "); 3520 if (_node == NULL) 3521 tty->print_cr("<null>"); 3522 else 3523 _node->dump(); 3524 } 3525 3526 void ConnectionGraph::dump(GrowableArray<PointsToNode*>& ptnodes_worklist) { 3527 bool first = true; 3528 int ptnodes_length = ptnodes_worklist.length(); 3529 for (int i = 0; i < ptnodes_length; i++) { 3530 PointsToNode *ptn = ptnodes_worklist.at(i); 3531 if (ptn == NULL || !ptn->is_JavaObject()) 3532 continue; 3533 PointsToNode::EscapeState es = ptn->escape_state(); 3534 if ((es != PointsToNode::NoEscape) && !Verbose) { 3535 continue; 3536 } 3537 Node* n = ptn->ideal_node(); 3538 if (n->is_Allocate() || (n->is_CallStaticJava() && 3539 n->as_CallStaticJava()->is_boxing_method())) { 3540 if (first) { 3541 tty->cr(); 3542 tty->print("======== Connection graph for "); 3543 _compile->method()->print_short_name(); 3544 tty->cr(); 3545 first = false; 3546 } 3547 ptn->dump(); 3548 // Print all locals and fields which reference this allocation 3549 for (UseIterator j(ptn); j.has_next(); j.next()) { 3550 PointsToNode* use = j.get(); 3551 if (use->is_LocalVar()) { 3552 use->dump(Verbose); 3553 } else if (Verbose) { 3554 use->dump(); 3555 } 3556 } 3557 tty->cr(); 3558 } 3559 } 3560 } 3561 #endif 3562 3563 void ConnectionGraph::record_for_optimizer(Node *n) { 3564 _igvn->_worklist.push(n); 3565 _igvn->add_users_to_worklist(n); 3566 }