1 /*
2 * Copyright (c) 2005, 2012, 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 "libadt/vectset.hpp"
29 #include "memory/allocation.hpp"
30 #include "opto/c2compiler.hpp"
31 #include "opto/callnode.hpp"
32 #include "opto/cfgnode.hpp"
33 #include "opto/compile.hpp"
34 #include "opto/escape.hpp"
35 #include "opto/phaseX.hpp"
36 #include "opto/rootnode.hpp"
37
38 ConnectionGraph::ConnectionGraph(Compile * C, PhaseIterGVN *igvn) :
39 _nodes(C->comp_arena(), C->unique(), C->unique(), NULL),
40 _collecting(true),
41 _verify(false),
42 _compile(C),
43 _igvn(igvn),
44 _node_map(C->comp_arena()) {
45 // Add unknown java object.
46 add_java_object(C->top(), PointsToNode::GlobalEscape);
47 phantom_obj = ptnode_adr(C->top()->_idx)->as_JavaObject();
48 // Add ConP(#NULL) and ConN(#NULL) nodes.
49 Node* oop_null = igvn->zerocon(T_OBJECT);
50 assert(oop_null->_idx < nodes_size(), "should be created already");
51 add_java_object(oop_null, PointsToNode::NoEscape);
52 null_obj = ptnode_adr(oop_null->_idx)->as_JavaObject();
53 if (UseCompressedOops) {
54 Node* noop_null = igvn->zerocon(T_NARROWOOP);
55 assert(noop_null->_idx < nodes_size(), "should be created already");
56 map_ideal_node(noop_null, null_obj);
57 }
58 _pcmp_neq = NULL; // Should be initialized
59 _pcmp_eq = NULL;
60 }
61
62 bool ConnectionGraph::has_candidates(Compile *C) {
63 // EA brings benefits only when the code has allocations and/or locks which
64 // are represented by ideal Macro nodes.
65 int cnt = C->macro_count();
66 for( int i=0; i < cnt; i++ ) {
67 Node *n = C->macro_node(i);
68 if ( n->is_Allocate() )
69 return true;
70 if( n->is_Lock() ) {
71 Node* obj = n->as_Lock()->obj_node()->uncast();
72 if( !(obj->is_Parm() || obj->is_Con()) )
73 return true;
74 }
75 }
76 return false;
77 }
78
79 void ConnectionGraph::do_analysis(Compile *C, PhaseIterGVN *igvn) {
80 Compile::TracePhase t2("escapeAnalysis", &Phase::_t_escapeAnalysis, true);
81 ResourceMark rm;
82
83 // Add ConP#NULL and ConN#NULL nodes before ConnectionGraph construction
84 // to create space for them in ConnectionGraph::_nodes[].
85 Node* oop_null = igvn->zerocon(T_OBJECT);
86 Node* noop_null = igvn->zerocon(T_NARROWOOP);
87 ConnectionGraph* congraph = new(C->comp_arena()) ConnectionGraph(C, igvn);
88 // Perform escape analysis
89 if (congraph->compute_escape()) {
90 // There are non escaping objects.
91 C->set_congraph(congraph);
92 }
93 // Cleanup.
94 if (oop_null->outcnt() == 0)
95 igvn->hash_delete(oop_null);
96 if (noop_null->outcnt() == 0)
97 igvn->hash_delete(noop_null);
98 }
99
100 bool ConnectionGraph::compute_escape() {
101 Compile* C = _compile;
102 PhaseGVN* igvn = _igvn;
103
104 // Worklists used by EA.
105 Unique_Node_List delayed_worklist;
106 GrowableArray<Node*> alloc_worklist;
107 GrowableArray<Node*> ptr_cmp_worklist;
108 GrowableArray<Node*> storestore_worklist;
109 GrowableArray<PointsToNode*> ptnodes_worklist;
110 GrowableArray<JavaObjectNode*> java_objects_worklist;
111 GrowableArray<JavaObjectNode*> non_escaped_worklist;
112 GrowableArray<FieldNode*> oop_fields_worklist;
113 DEBUG_ONLY( GrowableArray<Node*> addp_worklist; )
114
115 { Compile::TracePhase t3("connectionGraph", &Phase::_t_connectionGraph, true);
116
117 // 1. Populate Connection Graph (CG) with PointsTo nodes.
118 ideal_nodes.map(C->unique(), NULL); // preallocate space
119 // Initialize worklist
120 if (C->root() != NULL) {
121 ideal_nodes.push(C->root());
122 }
123 for( uint next = 0; next < ideal_nodes.size(); ++next ) {
124 Node* n = ideal_nodes.at(next);
125 // Create PointsTo nodes and add them to Connection Graph. Called
126 // only once per ideal node since ideal_nodes is Unique_Node list.
127 add_node_to_connection_graph(n, &delayed_worklist);
128 PointsToNode* ptn = ptnode_adr(n->_idx);
129 if (ptn != NULL) {
130 ptnodes_worklist.append(ptn);
131 if (ptn->is_JavaObject()) {
132 java_objects_worklist.append(ptn->as_JavaObject());
133 if ((n->is_Allocate() || n->is_CallStaticJava()) &&
134 (ptn->escape_state() < PointsToNode::GlobalEscape)) {
135 // Only allocations and java static calls results are interesting.
136 non_escaped_worklist.append(ptn->as_JavaObject());
137 }
138 } else if (ptn->is_Field() && ptn->as_Field()->is_oop()) {
139 oop_fields_worklist.append(ptn->as_Field());
140 }
141 }
142 if (n->is_MergeMem()) {
143 // Collect all MergeMem nodes to add memory slices for
144 // scalar replaceable objects in split_unique_types().
145 _mergemem_worklist.append(n->as_MergeMem());
146 } else if (OptimizePtrCompare && n->is_Cmp() &&
147 (n->Opcode() == Op_CmpP || n->Opcode() == Op_CmpN)) {
148 // Collect compare pointers nodes.
149 ptr_cmp_worklist.append(n);
150 } else if (n->is_MemBarStoreStore()) {
151 // Collect all MemBarStoreStore nodes so that depending on the
152 // escape status of the associated Allocate node some of them
153 // may be eliminated.
154 storestore_worklist.append(n);
155 #ifdef ASSERT
156 } else if(n->is_AddP()) {
157 // Collect address nodes for graph verification.
158 addp_worklist.append(n);
159 #endif
160 }
161 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
162 Node* m = n->fast_out(i); // Get user
163 ideal_nodes.push(m);
164 }
165 }
166 if (non_escaped_worklist.length() == 0) {
167 _collecting = false;
168 return false; // Nothing to do.
169 }
170 // Add final simple edges to graph.
171 while(delayed_worklist.size() > 0) {
172 Node* n = delayed_worklist.pop();
173 add_final_edges(n);
174 }
175 int ptnodes_length = ptnodes_worklist.length();
176
177 #ifdef ASSERT
178 if (VerifyConnectionGraph) {
179 // Verify that no new simple edges could be created and all
180 // local vars has edges.
181 _verify = true;
182 for (int next = 0; next < ptnodes_length; ++next) {
183 PointsToNode* ptn = ptnodes_worklist.at(next);
184 add_final_edges(ptn->ideal_node());
185 if (ptn->is_LocalVar() && ptn->edge_count() == 0) {
186 ptn->dump();
187 assert(ptn->as_LocalVar()->edge_count() > 0, "sanity");
188 }
189 }
190 _verify = false;
191 }
192 #endif
193
194 // 2. Finish Graph construction by propagating references to all
195 // java objects through graph.
196 if (!complete_connection_graph(ptnodes_worklist, non_escaped_worklist,
197 java_objects_worklist, oop_fields_worklist)) {
198 // All objects escaped or hit time or iterations limits.
199 _collecting = false;
200 return false;
201 }
202
203 // 3. Adjust scalar_replaceable state of nonescaping objects and push
204 // scalar replaceable allocations on alloc_worklist for processing
205 // in split_unique_types().
206 int non_escaped_length = non_escaped_worklist.length();
207 for (int next = 0; next < non_escaped_length; next++) {
208 JavaObjectNode* ptn = non_escaped_worklist.at(next);
209 if (ptn->escape_state() == PointsToNode::NoEscape &&
210 ptn->scalar_replaceable()) {
211 adjust_scalar_replaceable_state(ptn);
212 if (ptn->scalar_replaceable()) {
213 alloc_worklist.append(ptn->ideal_node());
214 }
215 }
216 }
217
218 #ifdef ASSERT
219 if (VerifyConnectionGraph) {
220 // Verify that graph is complete - no new edges could be added or needed.
221 verify_connection_graph(ptnodes_worklist, non_escaped_worklist,
222 java_objects_worklist, addp_worklist);
223 }
224 assert(C->unique() == nodes_size(), "no new ideal nodes should be added during ConnectionGraph build");
225 assert(null_obj->escape_state() == PointsToNode::NoEscape &&
226 null_obj->edge_count() == 0 &&
227 !null_obj->arraycopy_src() &&
228 !null_obj->arraycopy_dst(), "sanity");
229 #endif
230
231 _collecting = false;
232
233 } // TracePhase t3("connectionGraph")
234
235 // 4. Optimize ideal graph based on EA information.
236 bool has_non_escaping_obj = (non_escaped_worklist.length() > 0);
237 if (has_non_escaping_obj) {
238 optimize_ideal_graph(ptr_cmp_worklist, storestore_worklist);
239 }
240
241 #ifndef PRODUCT
242 if (PrintEscapeAnalysis) {
243 dump(ptnodes_worklist); // Dump ConnectionGraph
244 }
245 #endif
246
247 bool has_scalar_replaceable_candidates = (alloc_worklist.length() > 0);
248 #ifdef ASSERT
249 if (VerifyConnectionGraph) {
250 int alloc_length = alloc_worklist.length();
251 for (int next = 0; next < alloc_length; ++next) {
252 Node* n = alloc_worklist.at(next);
253 PointsToNode* ptn = ptnode_adr(n->_idx);
254 assert(ptn->escape_state() == PointsToNode::NoEscape && ptn->scalar_replaceable(), "sanity");
255 }
256 }
257 #endif
258
259 // 5. Separate memory graph for scalar replaceable allcations.
260 if (has_scalar_replaceable_candidates &&
261 C->AliasLevel() >= 3 && EliminateAllocations) {
262 // Now use the escape information to create unique types for
263 // scalar replaceable objects.
264 split_unique_types(alloc_worklist);
265 if (C->failing()) return false;
266 C->print_method("After Escape Analysis", 2);
267
268 #ifdef ASSERT
269 } else if (Verbose && (PrintEscapeAnalysis || PrintEliminateAllocations)) {
270 tty->print("=== No allocations eliminated for ");
271 C->method()->print_short_name();
272 if(!EliminateAllocations) {
273 tty->print(" since EliminateAllocations is off ===");
274 } else if(!has_scalar_replaceable_candidates) {
275 tty->print(" since there are no scalar replaceable candidates ===");
276 } else if(C->AliasLevel() < 3) {
277 tty->print(" since AliasLevel < 3 ===");
278 }
279 tty->cr();
280 #endif
281 }
282 return has_non_escaping_obj;
283 }
284
285 // Utility function for nodes that load an object
286 void ConnectionGraph::add_objload_to_connection_graph(Node *n, Unique_Node_List *delayed_worklist) {
287 // Using isa_ptr() instead of isa_oopptr() for LoadP and Phi because
288 // ThreadLocal has RawPtr type.
289 const Type* t = _igvn->type(n);
290 if (t->make_ptr() != NULL) {
291 Node* adr = n->in(MemNode::Address);
292 #ifdef ASSERT
293 if (!adr->is_AddP()) {
294 assert(_igvn->type(adr)->isa_rawptr(), "sanity");
295 } else {
296 assert((ptnode_adr(adr->_idx) == NULL ||
297 ptnode_adr(adr->_idx)->as_Field()->is_oop()), "sanity");
298 }
299 #endif
300 add_local_var_and_edge(n, PointsToNode::NoEscape,
301 adr, delayed_worklist);
302 }
303 }
304
305 // Populate Connection Graph with PointsTo nodes and create simple
306 // connection graph edges.
307 void ConnectionGraph::add_node_to_connection_graph(Node *n, Unique_Node_List *delayed_worklist) {
308 assert(!_verify, "this method sould not be called for verification");
309 PhaseGVN* igvn = _igvn;
310 uint n_idx = n->_idx;
311 PointsToNode* n_ptn = ptnode_adr(n_idx);
312 if (n_ptn != NULL)
313 return; // No need to redefine PointsTo node during first iteration.
314
315 if (n->is_Call()) {
316 // Arguments to allocation and locking don't escape.
317 if (n->is_AbstractLock()) {
318 // Put Lock and Unlock nodes on IGVN worklist to process them during
319 // first IGVN optimization when escape information is still available.
320 record_for_optimizer(n);
321 } else if (n->is_Allocate()) {
322 add_call_node(n->as_Call());
323 record_for_optimizer(n);
324 } else {
325 if (n->is_CallStaticJava()) {
326 const char* name = n->as_CallStaticJava()->_name;
327 if (name != NULL && strcmp(name, "uncommon_trap") == 0)
328 return; // Skip uncommon traps
329 }
330 // Don't mark as processed since call's arguments have to be processed.
331 delayed_worklist->push(n);
332 // Check if a call returns an object.
333 if (n->as_Call()->returns_pointer() &&
334 n->as_Call()->proj_out(TypeFunc::Parms) != NULL) {
335 add_call_node(n->as_Call());
336 }
337 }
338 return;
339 }
340 // Put this check here to process call arguments since some call nodes
341 // point to phantom_obj.
342 if (n_ptn == phantom_obj || n_ptn == null_obj)
343 return; // Skip predefined nodes.
344
345 int opcode = n->Opcode();
346 switch (opcode) {
347 case Op_AddP: {
348 Node* base = get_addp_base(n);
349 PointsToNode* ptn_base = ptnode_adr(base->_idx);
350 // Field nodes are created for all field types. They are used in
351 // adjust_scalar_replaceable_state() and split_unique_types().
352 // Note, non-oop fields will have only base edges in Connection
353 // Graph because such fields are not used for oop loads and stores.
354 int offset = address_offset(n, igvn);
355 add_field(n, PointsToNode::NoEscape, offset);
356 if (ptn_base == NULL) {
357 delayed_worklist->push(n); // Process it later.
358 } else {
359 n_ptn = ptnode_adr(n_idx);
360 add_base(n_ptn->as_Field(), ptn_base);
361 }
362 break;
363 }
364 case Op_CastX2P: {
365 map_ideal_node(n, phantom_obj);
366 break;
367 }
368 case Op_CastPP:
369 case Op_CheckCastPP:
370 case Op_EncodeP:
371 case Op_DecodeN: {
372 add_local_var_and_edge(n, PointsToNode::NoEscape,
373 n->in(1), delayed_worklist);
374 break;
375 }
376 case Op_CMoveP: {
377 add_local_var(n, PointsToNode::NoEscape);
378 // Do not add edges during first iteration because some could be
379 // not defined yet.
380 delayed_worklist->push(n);
381 break;
382 }
383 case Op_ConP:
384 case Op_ConN: {
385 // assume all oop constants globally escape except for null
386 PointsToNode::EscapeState es;
387 if (igvn->type(n) == TypePtr::NULL_PTR ||
388 igvn->type(n) == TypeNarrowOop::NULL_PTR) {
389 es = PointsToNode::NoEscape;
390 } else {
391 es = PointsToNode::GlobalEscape;
392 }
393 add_java_object(n, es);
394 break;
395 }
396 case Op_CreateEx: {
397 // assume that all exception objects globally escape
398 add_java_object(n, PointsToNode::GlobalEscape);
399 break;
400 }
401 case Op_LoadKlass:
402 case Op_LoadNKlass: {
403 // Unknown class is loaded
404 map_ideal_node(n, phantom_obj);
405 break;
406 }
407 case Op_LoadP:
408 case Op_LoadN:
409 case Op_LoadPLocked: {
410 add_objload_to_connection_graph(n, delayed_worklist);
411 break;
412 }
413 case Op_Parm: {
414 map_ideal_node(n, phantom_obj);
415 break;
416 }
417 case Op_PartialSubtypeCheck: {
418 // Produces Null or notNull and is used in only in CmpP so
419 // phantom_obj could be used.
420 map_ideal_node(n, phantom_obj); // Result is unknown
421 break;
422 }
423 case Op_Phi: {
424 // Using isa_ptr() instead of isa_oopptr() for LoadP and Phi because
425 // ThreadLocal has RawPtr type.
426 const Type* t = n->as_Phi()->type();
427 if (t->make_ptr() != NULL) {
428 add_local_var(n, PointsToNode::NoEscape);
429 // Do not add edges during first iteration because some could be
430 // not defined yet.
431 delayed_worklist->push(n);
432 }
433 break;
434 }
435 case Op_Proj: {
436 // we are only interested in the oop result projection from a call
437 if (n->as_Proj()->_con == TypeFunc::Parms && n->in(0)->is_Call() &&
438 n->in(0)->as_Call()->returns_pointer()) {
439 add_local_var_and_edge(n, PointsToNode::NoEscape,
440 n->in(0), delayed_worklist);
441 }
442 break;
443 }
444 case Op_Rethrow: // Exception object escapes
445 case Op_Return: {
446 if (n->req() > TypeFunc::Parms &&
447 igvn->type(n->in(TypeFunc::Parms))->isa_oopptr()) {
448 // Treat Return value as LocalVar with GlobalEscape escape state.
449 add_local_var_and_edge(n, PointsToNode::GlobalEscape,
450 n->in(TypeFunc::Parms), delayed_worklist);
451 }
452 break;
453 }
454 case Op_GetAndSetP:
455 case Op_GetAndSetN: {
456 add_objload_to_connection_graph(n, delayed_worklist);
457 // fallthrough
458 }
459 case Op_StoreP:
460 case Op_StoreN:
461 case Op_StorePConditional:
462 case Op_CompareAndSwapP:
463 case Op_CompareAndSwapN: {
464 Node* adr = n->in(MemNode::Address);
465 const Type *adr_type = igvn->type(adr);
466 adr_type = adr_type->make_ptr();
467 if (adr_type->isa_oopptr() ||
468 (opcode == Op_StoreP || opcode == Op_StoreN) &&
469 (adr_type == TypeRawPtr::NOTNULL &&
470 adr->in(AddPNode::Address)->is_Proj() &&
471 adr->in(AddPNode::Address)->in(0)->is_Allocate())) {
472 delayed_worklist->push(n); // Process it later.
473 #ifdef ASSERT
474 assert(adr->is_AddP(), "expecting an AddP");
475 if (adr_type == TypeRawPtr::NOTNULL) {
476 // Verify a raw address for a store captured by Initialize node.
477 int offs = (int)igvn->find_intptr_t_con(adr->in(AddPNode::Offset), Type::OffsetBot);
478 assert(offs != Type::OffsetBot, "offset must be a constant");
479 }
480 #endif
481 } else {
482 // Ignore copy the displaced header to the BoxNode (OSR compilation).
483 if (adr->is_BoxLock())
484 break;
485 // Stored value escapes in unsafe access.
486 if ((opcode == Op_StoreP) && (adr_type == TypeRawPtr::BOTTOM)) {
487 // Pointer stores in G1 barriers looks like unsafe access.
488 // Ignore such stores to be able scalar replace non-escaping
489 // allocations.
490 if (UseG1GC && adr->is_AddP()) {
491 Node* base = get_addp_base(adr);
492 if (base->Opcode() == Op_LoadP &&
493 base->in(MemNode::Address)->is_AddP()) {
494 adr = base->in(MemNode::Address);
495 Node* tls = get_addp_base(adr);
496 if (tls->Opcode() == Op_ThreadLocal) {
497 int offs = (int)igvn->find_intptr_t_con(adr->in(AddPNode::Offset), Type::OffsetBot);
498 if (offs == in_bytes(JavaThread::satb_mark_queue_offset() +
499 PtrQueue::byte_offset_of_buf())) {
500 break; // G1 pre barier previous oop value store.
501 }
502 if (offs == in_bytes(JavaThread::dirty_card_queue_offset() +
503 PtrQueue::byte_offset_of_buf())) {
504 break; // G1 post barier card address store.
505 }
506 }
507 }
508 }
509 delayed_worklist->push(n); // Process unsafe access later.
510 break;
511 }
512 #ifdef ASSERT
513 n->dump(1);
514 assert(false, "not unsafe or G1 barrier raw StoreP");
515 #endif
516 }
517 break;
518 }
519 case Op_AryEq:
520 case Op_StrComp:
521 case Op_StrEquals:
522 case Op_StrIndexOf: {
523 add_local_var(n, PointsToNode::ArgEscape);
524 delayed_worklist->push(n); // Process it later.
525 break;
526 }
527 case Op_ThreadLocal: {
528 add_java_object(n, PointsToNode::ArgEscape);
529 break;
530 }
531 default:
532 ; // Do nothing for nodes not related to EA.
533 }
534 return;
535 }
536
537 #ifdef ASSERT
538 #define ELSE_FAIL(name) \
539 /* Should not be called for not pointer type. */ \
540 n->dump(1); \
541 assert(false, name); \
542 break;
543 #else
544 #define ELSE_FAIL(name) \
545 break;
546 #endif
547
548 // Add final simple edges to graph.
549 void ConnectionGraph::add_final_edges(Node *n) {
550 PointsToNode* n_ptn = ptnode_adr(n->_idx);
551 #ifdef ASSERT
552 if (_verify && n_ptn->is_JavaObject())
553 return; // This method does not change graph for JavaObject.
554 #endif
555
556 if (n->is_Call()) {
557 process_call_arguments(n->as_Call());
558 return;
559 }
560 assert(n->is_Store() || n->is_LoadStore() ||
561 (n_ptn != NULL) && (n_ptn->ideal_node() != NULL),
562 "node should be registered already");
563 int opcode = n->Opcode();
564 switch (opcode) {
565 case Op_AddP: {
566 Node* base = get_addp_base(n);
567 PointsToNode* ptn_base = ptnode_adr(base->_idx);
568 assert(ptn_base != NULL, "field's base should be registered");
569 add_base(n_ptn->as_Field(), ptn_base);
570 break;
571 }
572 case Op_CastPP:
573 case Op_CheckCastPP:
574 case Op_EncodeP:
575 case Op_DecodeN: {
576 add_local_var_and_edge(n, PointsToNode::NoEscape,
577 n->in(1), NULL);
578 break;
579 }
580 case Op_CMoveP: {
581 for (uint i = CMoveNode::IfFalse; i < n->req(); i++) {
582 Node* in = n->in(i);
583 if (in == NULL)
584 continue; // ignore NULL
585 Node* uncast_in = in->uncast();
586 if (uncast_in->is_top() || uncast_in == n)
587 continue; // ignore top or inputs which go back this node
588 PointsToNode* ptn = ptnode_adr(in->_idx);
589 assert(ptn != NULL, "node should be registered");
590 add_edge(n_ptn, ptn);
591 }
592 break;
593 }
594 case Op_LoadP:
595 case Op_LoadN:
596 case Op_LoadPLocked: {
597 // Using isa_ptr() instead of isa_oopptr() for LoadP and Phi because
598 // ThreadLocal has RawPtr type.
599 const Type* t = _igvn->type(n);
600 if (t->make_ptr() != NULL) {
601 Node* adr = n->in(MemNode::Address);
602 add_local_var_and_edge(n, PointsToNode::NoEscape, adr, NULL);
603 break;
604 }
605 ELSE_FAIL("Op_LoadP");
606 }
607 case Op_Phi: {
608 // Using isa_ptr() instead of isa_oopptr() for LoadP and Phi because
609 // ThreadLocal has RawPtr type.
610 const Type* t = n->as_Phi()->type();
611 if (t->make_ptr() != NULL) {
612 for (uint i = 1; i < n->req(); i++) {
613 Node* in = n->in(i);
614 if (in == NULL)
615 continue; // ignore NULL
616 Node* uncast_in = in->uncast();
617 if (uncast_in->is_top() || uncast_in == n)
618 continue; // ignore top or inputs which go back this node
619 PointsToNode* ptn = ptnode_adr(in->_idx);
620 assert(ptn != NULL, "node should be registered");
621 add_edge(n_ptn, ptn);
622 }
623 break;
624 }
625 ELSE_FAIL("Op_Phi");
626 }
627 case Op_Proj: {
628 // we are only interested in the oop result projection from a call
629 if (n->as_Proj()->_con == TypeFunc::Parms && n->in(0)->is_Call() &&
630 n->in(0)->as_Call()->returns_pointer()) {
631 add_local_var_and_edge(n, PointsToNode::NoEscape, n->in(0), NULL);
632 break;
633 }
634 ELSE_FAIL("Op_Proj");
635 }
636 case Op_Rethrow: // Exception object escapes
637 case Op_Return: {
638 if (n->req() > TypeFunc::Parms &&
639 _igvn->type(n->in(TypeFunc::Parms))->isa_oopptr()) {
640 // Treat Return value as LocalVar with GlobalEscape escape state.
641 add_local_var_and_edge(n, PointsToNode::GlobalEscape,
642 n->in(TypeFunc::Parms), NULL);
643 break;
644 }
645 ELSE_FAIL("Op_Return");
646 }
647 case Op_StoreP:
648 case Op_StoreN:
649 case Op_StorePConditional:
650 case Op_CompareAndSwapP:
651 case Op_CompareAndSwapN:
652 case Op_GetAndSetP:
653 case Op_GetAndSetN: {
654 Node* adr = n->in(MemNode::Address);
655 if (opcode == Op_GetAndSetP || opcode == Op_GetAndSetN) {
656 const Type* t = _igvn->type(n);
657 if (t->make_ptr() != NULL) {
658 add_local_var_and_edge(n, PointsToNode::NoEscape, adr, NULL);
659 }
660 }
661 const Type *adr_type = _igvn->type(adr);
662 adr_type = adr_type->make_ptr();
663 if (adr_type->isa_oopptr() ||
664 (opcode == Op_StoreP || opcode == Op_StoreN) &&
665 (adr_type == TypeRawPtr::NOTNULL &&
666 adr->in(AddPNode::Address)->is_Proj() &&
667 adr->in(AddPNode::Address)->in(0)->is_Allocate())) {
668 // Point Address to Value
669 PointsToNode* adr_ptn = ptnode_adr(adr->_idx);
670 assert(adr_ptn != NULL &&
671 adr_ptn->as_Field()->is_oop(), "node should be registered");
672 Node *val = n->in(MemNode::ValueIn);
673 PointsToNode* ptn = ptnode_adr(val->_idx);
674 assert(ptn != NULL, "node should be registered");
675 add_edge(adr_ptn, ptn);
676 break;
677 } else if ((opcode == Op_StoreP) && (adr_type == TypeRawPtr::BOTTOM)) {
678 // Stored value escapes in unsafe access.
679 Node *val = n->in(MemNode::ValueIn);
680 PointsToNode* ptn = ptnode_adr(val->_idx);
681 assert(ptn != NULL, "node should be registered");
682 ptn->set_escape_state(PointsToNode::GlobalEscape);
683 // Add edge to object for unsafe access with offset.
684 PointsToNode* adr_ptn = ptnode_adr(adr->_idx);
685 assert(adr_ptn != NULL, "node should be registered");
686 if (adr_ptn->is_Field()) {
687 assert(adr_ptn->as_Field()->is_oop(), "should be oop field");
688 add_edge(adr_ptn, ptn);
689 }
690 break;
691 }
692 ELSE_FAIL("Op_StoreP");
693 }
694 case Op_AryEq:
695 case Op_StrComp:
696 case Op_StrEquals:
697 case Op_StrIndexOf: {
698 // char[] arrays passed to string intrinsic do not escape but
699 // they are not scalar replaceable. Adjust escape state for them.
700 // Start from in(2) edge since in(1) is memory edge.
701 for (uint i = 2; i < n->req(); i++) {
702 Node* adr = n->in(i);
703 const Type* at = _igvn->type(adr);
704 if (!adr->is_top() && at->isa_ptr()) {
705 assert(at == Type::TOP || at == TypePtr::NULL_PTR ||
706 at->isa_ptr() != NULL, "expecting a pointer");
707 if (adr->is_AddP()) {
708 adr = get_addp_base(adr);
709 }
710 PointsToNode* ptn = ptnode_adr(adr->_idx);
711 assert(ptn != NULL, "node should be registered");
712 add_edge(n_ptn, ptn);
713 }
714 }
715 break;
716 }
717 default: {
718 // This method should be called only for EA specific nodes which may
719 // miss some edges when they were created.
720 #ifdef ASSERT
721 n->dump(1);
722 #endif
723 guarantee(false, "unknown node");
724 }
725 }
726 return;
727 }
728
729 void ConnectionGraph::add_call_node(CallNode* call) {
730 assert(call->returns_pointer(), "only for call which returns pointer");
731 uint call_idx = call->_idx;
732 if (call->is_Allocate()) {
733 Node* k = call->in(AllocateNode::KlassNode);
734 const TypeKlassPtr* kt = k->bottom_type()->isa_klassptr();
735 assert(kt != NULL, "TypeKlassPtr required.");
736 ciKlass* cik = kt->klass();
737 PointsToNode::EscapeState es = PointsToNode::NoEscape;
738 bool scalar_replaceable = true;
739 if (call->is_AllocateArray()) {
740 if (!cik->is_array_klass()) { // StressReflectiveCode
741 es = PointsToNode::GlobalEscape;
742 } else {
743 int length = call->in(AllocateNode::ALength)->find_int_con(-1);
744 if (length < 0 || length > EliminateAllocationArraySizeLimit) {
745 // Not scalar replaceable if the length is not constant or too big.
746 scalar_replaceable = false;
747 }
748 }
749 } else { // Allocate instance
750 if (cik->is_subclass_of(_compile->env()->Thread_klass()) ||
751 !cik->is_instance_klass() || // StressReflectiveCode
752 cik->as_instance_klass()->has_finalizer()) {
753 es = PointsToNode::GlobalEscape;
754 }
755 }
756 add_java_object(call, es);
757 PointsToNode* ptn = ptnode_adr(call_idx);
758 if (!scalar_replaceable && ptn->scalar_replaceable()) {
759 ptn->set_scalar_replaceable(false);
760 }
761 } else if (call->is_CallStaticJava()) {
762 // Call nodes could be different types:
763 //
764 // 1. CallDynamicJavaNode (what happened during call is unknown):
765 //
766 // - mapped to GlobalEscape JavaObject node if oop is returned;
767 //
768 // - all oop arguments are escaping globally;
769 //
770 // 2. CallStaticJavaNode (execute bytecode analysis if possible):
771 //
772 // - the same as CallDynamicJavaNode if can't do bytecode analysis;
773 //
774 // - mapped to GlobalEscape JavaObject node if unknown oop is returned;
775 // - mapped to NoEscape JavaObject node if non-escaping object allocated
776 // during call is returned;
777 // - mapped to ArgEscape LocalVar node pointed to object arguments
778 // which are returned and does not escape during call;
779 //
780 // - oop arguments escaping status is defined by bytecode analysis;
781 //
782 // For a static call, we know exactly what method is being called.
783 // Use bytecode estimator to record whether the call's return value escapes.
784 ciMethod* meth = call->as_CallJava()->method();
785 if (meth == NULL) {
786 const char* name = call->as_CallStaticJava()->_name;
787 assert(strncmp(name, "_multianewarray", 15) == 0, "TODO: add failed case check");
788 // Returns a newly allocated unescaped object.
789 add_java_object(call, PointsToNode::NoEscape);
790 ptnode_adr(call_idx)->set_scalar_replaceable(false);
791 } else {
792 BCEscapeAnalyzer* call_analyzer = meth->get_bcea();
793 call_analyzer->copy_dependencies(_compile->dependencies());
794 if (call_analyzer->is_return_allocated()) {
795 // Returns a newly allocated unescaped object, simply
796 // update dependency information.
797 // Mark it as NoEscape so that objects referenced by
798 // it's fields will be marked as NoEscape at least.
799 add_java_object(call, PointsToNode::NoEscape);
800 ptnode_adr(call_idx)->set_scalar_replaceable(false);
801 } else {
802 // Determine whether any arguments are returned.
803 const TypeTuple* d = call->tf()->domain();
804 bool ret_arg = false;
805 for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
806 if (d->field_at(i)->isa_ptr() != NULL &&
807 call_analyzer->is_arg_returned(i - TypeFunc::Parms)) {
808 ret_arg = true;
809 break;
810 }
811 }
812 if (ret_arg) {
813 add_local_var(call, PointsToNode::ArgEscape);
814 } else {
815 // Returns unknown object.
816 map_ideal_node(call, phantom_obj);
817 }
818 }
819 }
820 } else {
821 // An other type of call, assume the worst case:
822 // returned value is unknown and globally escapes.
823 assert(call->Opcode() == Op_CallDynamicJava, "add failed case check");
824 map_ideal_node(call, phantom_obj);
825 }
826 }
827
828 void ConnectionGraph::process_call_arguments(CallNode *call) {
829 bool is_arraycopy = false;
830 switch (call->Opcode()) {
831 #ifdef ASSERT
832 case Op_Allocate:
833 case Op_AllocateArray:
834 case Op_Lock:
835 case Op_Unlock:
836 assert(false, "should be done already");
837 break;
838 #endif
839 case Op_CallLeafNoFP:
840 is_arraycopy = (call->as_CallLeaf()->_name != NULL &&
841 strstr(call->as_CallLeaf()->_name, "arraycopy") != 0);
842 // fall through
843 case Op_CallLeaf: {
844 // Stub calls, objects do not escape but they are not scale replaceable.
845 // Adjust escape state for outgoing arguments.
846 const TypeTuple * d = call->tf()->domain();
847 bool src_has_oops = false;
848 for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
849 const Type* at = d->field_at(i);
850 Node *arg = call->in(i);
851 const Type *aat = _igvn->type(arg);
852 if (arg->is_top() || !at->isa_ptr() || !aat->isa_ptr())
853 continue;
854 if (arg->is_AddP()) {
855 //
856 // The inline_native_clone() case when the arraycopy stub is called
857 // after the allocation before Initialize and CheckCastPP nodes.
858 // Or normal arraycopy for object arrays case.
859 //
860 // Set AddP's base (Allocate) as not scalar replaceable since
861 // pointer to the base (with offset) is passed as argument.
862 //
863 arg = get_addp_base(arg);
864 }
865 PointsToNode* arg_ptn = ptnode_adr(arg->_idx);
866 assert(arg_ptn != NULL, "should be registered");
867 PointsToNode::EscapeState arg_esc = arg_ptn->escape_state();
868 if (is_arraycopy || arg_esc < PointsToNode::ArgEscape) {
869 assert(aat == Type::TOP || aat == TypePtr::NULL_PTR ||
870 aat->isa_ptr() != NULL, "expecting an Ptr");
871 bool arg_has_oops = aat->isa_oopptr() &&
872 (aat->isa_oopptr()->klass() == NULL || aat->isa_instptr() ||
873 (aat->isa_aryptr() && aat->isa_aryptr()->klass()->is_obj_array_klass()));
874 if (i == TypeFunc::Parms) {
875 src_has_oops = arg_has_oops;
876 }
877 //
878 // src or dst could be j.l.Object when other is basic type array:
879 //
880 // arraycopy(char[],0,Object*,0,size);
881 // arraycopy(Object*,0,char[],0,size);
882 //
883 // Don't add edges in such cases.
884 //
885 bool arg_is_arraycopy_dest = src_has_oops && is_arraycopy &&
886 arg_has_oops && (i > TypeFunc::Parms);
887 #ifdef ASSERT
888 if (!(is_arraycopy ||
889 call->as_CallLeaf()->_name != NULL &&
890 (strcmp(call->as_CallLeaf()->_name, "g1_wb_pre") == 0 ||
891 strcmp(call->as_CallLeaf()->_name, "g1_wb_post") == 0 ))
892 ) {
893 call->dump();
894 assert(false, "EA: unexpected CallLeaf");
895 }
896 #endif
897 // Always process arraycopy's destination object since
898 // we need to add all possible edges to references in
899 // source object.
900 if (arg_esc >= PointsToNode::ArgEscape &&
901 !arg_is_arraycopy_dest) {
902 continue;
903 }
904 set_escape_state(arg_ptn, PointsToNode::ArgEscape);
905 if (arg_is_arraycopy_dest) {
906 Node* src = call->in(TypeFunc::Parms);
907 if (src->is_AddP()) {
908 src = get_addp_base(src);
909 }
910 PointsToNode* src_ptn = ptnode_adr(src->_idx);
911 assert(src_ptn != NULL, "should be registered");
912 if (arg_ptn != src_ptn) {
913 // Special arraycopy edge:
914 // A destination object's field can't have the source object
915 // as base since objects escape states are not related.
916 // Only escape state of destination object's fields affects
917 // escape state of fields in source object.
918 add_arraycopy(call, PointsToNode::ArgEscape, src_ptn, arg_ptn);
919 }
920 }
921 }
922 }
923 break;
924 }
925 case Op_CallStaticJava: {
926 // For a static call, we know exactly what method is being called.
927 // Use bytecode estimator to record the call's escape affects
928 #ifdef ASSERT
929 const char* name = call->as_CallStaticJava()->_name;
930 assert((name == NULL || strcmp(name, "uncommon_trap") != 0), "normal calls only");
931 #endif
932 ciMethod* meth = call->as_CallJava()->method();
933 BCEscapeAnalyzer* call_analyzer = (meth !=NULL) ? meth->get_bcea() : NULL;
934 // fall-through if not a Java method or no analyzer information
935 if (call_analyzer != NULL) {
936 PointsToNode* call_ptn = ptnode_adr(call->_idx);
937 const TypeTuple* d = call->tf()->domain();
938 for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
939 const Type* at = d->field_at(i);
940 int k = i - TypeFunc::Parms;
941 Node* arg = call->in(i);
942 PointsToNode* arg_ptn = ptnode_adr(arg->_idx);
943 if (at->isa_ptr() != NULL &&
944 call_analyzer->is_arg_returned(k)) {
945 // The call returns arguments.
946 if (call_ptn != NULL) { // Is call's result used?
947 assert(call_ptn->is_LocalVar(), "node should be registered");
948 assert(arg_ptn != NULL, "node should be registered");
949 add_edge(call_ptn, arg_ptn);
950 }
951 }
952 if (at->isa_oopptr() != NULL &&
953 arg_ptn->escape_state() < PointsToNode::GlobalEscape) {
954 if (!call_analyzer->is_arg_stack(k)) {
955 // The argument global escapes
956 set_escape_state(arg_ptn, PointsToNode::GlobalEscape);
957 } else {
958 set_escape_state(arg_ptn, PointsToNode::ArgEscape);
959 if (!call_analyzer->is_arg_local(k)) {
960 // The argument itself doesn't escape, but any fields might
961 set_fields_escape_state(arg_ptn, PointsToNode::GlobalEscape);
962 }
963 }
964 }
965 }
966 if (call_ptn != NULL && call_ptn->is_LocalVar()) {
967 // The call returns arguments.
968 assert(call_ptn->edge_count() > 0, "sanity");
969 if (!call_analyzer->is_return_local()) {
970 // Returns also unknown object.
971 add_edge(call_ptn, phantom_obj);
972 }
973 }
974 break;
975 }
976 }
977 default: {
978 // Fall-through here if not a Java method or no analyzer information
979 // or some other type of call, assume the worst case: all arguments
980 // globally escape.
981 const TypeTuple* d = call->tf()->domain();
982 for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
983 const Type* at = d->field_at(i);
984 if (at->isa_oopptr() != NULL) {
985 Node* arg = call->in(i);
986 if (arg->is_AddP()) {
987 arg = get_addp_base(arg);
988 }
989 assert(ptnode_adr(arg->_idx) != NULL, "should be defined already");
990 set_escape_state(ptnode_adr(arg->_idx), PointsToNode::GlobalEscape);
991 }
992 }
993 }
994 }
995 }
996
997
998 // Finish Graph construction.
999 bool ConnectionGraph::complete_connection_graph(
1000 GrowableArray<PointsToNode*>& ptnodes_worklist,
1001 GrowableArray<JavaObjectNode*>& non_escaped_worklist,
1002 GrowableArray<JavaObjectNode*>& java_objects_worklist,
1003 GrowableArray<FieldNode*>& oop_fields_worklist) {
1004 // Normally only 1-3 passes needed to build Connection Graph depending
1005 // on graph complexity. Observed 8 passes in jvm2008 compiler.compiler.
1006 // Set limit to 20 to catch situation when something did go wrong and
1007 // bailout Escape Analysis.
1008 // Also limit build time to 30 sec (60 in debug VM).
1009 #define CG_BUILD_ITER_LIMIT 20
1010 #ifdef ASSERT
1011 #define CG_BUILD_TIME_LIMIT 60.0
1012 #else
1013 #define CG_BUILD_TIME_LIMIT 30.0
1014 #endif
1015
1016 // Propagate GlobalEscape and ArgEscape escape states and check that
1017 // we still have non-escaping objects. The method pushs on _worklist
1018 // Field nodes which reference phantom_object.
1019 if (!find_non_escaped_objects(ptnodes_worklist, non_escaped_worklist)) {
1020 return false; // Nothing to do.
1021 }
1022 // Now propagate references to all JavaObject nodes.
1023 int java_objects_length = java_objects_worklist.length();
1024 elapsedTimer time;
1025 int new_edges = 1;
1026 int iterations = 0;
1027 do {
1028 while ((new_edges > 0) &&
1029 (iterations++ < CG_BUILD_ITER_LIMIT) &&
1030 (time.seconds() < CG_BUILD_TIME_LIMIT)) {
1031 time.start();
1032 new_edges = 0;
1033 // Propagate references to phantom_object for nodes pushed on _worklist
1034 // by find_non_escaped_objects() and find_field_value().
1035 new_edges += add_java_object_edges(phantom_obj, false);
1036 for (int next = 0; next < java_objects_length; ++next) {
1037 JavaObjectNode* ptn = java_objects_worklist.at(next);
1038 new_edges += add_java_object_edges(ptn, true);
1039 }
1040 if (new_edges > 0) {
1041 // Update escape states on each iteration if graph was updated.
1042 if (!find_non_escaped_objects(ptnodes_worklist, non_escaped_worklist)) {
1043 return false; // Nothing to do.
1044 }
1045 }
1046 time.stop();
1047 }
1048 if ((iterations < CG_BUILD_ITER_LIMIT) &&
1049 (time.seconds() < CG_BUILD_TIME_LIMIT)) {
1050 time.start();
1051 // Find fields which have unknown value.
1052 int fields_length = oop_fields_worklist.length();
1053 for (int next = 0; next < fields_length; next++) {
1054 FieldNode* field = oop_fields_worklist.at(next);
1055 if (field->edge_count() == 0) {
1056 new_edges += find_field_value(field);
1057 // This code may added new edges to phantom_object.
1058 // Need an other cycle to propagate references to phantom_object.
1059 }
1060 }
1061 time.stop();
1062 } else {
1063 new_edges = 0; // Bailout
1064 }
1065 } while (new_edges > 0);
1066
1067 // Bailout if passed limits.
1068 if ((iterations >= CG_BUILD_ITER_LIMIT) ||
1069 (time.seconds() >= CG_BUILD_TIME_LIMIT)) {
1070 Compile* C = _compile;
1071 if (C->log() != NULL) {
1072 C->log()->begin_elem("connectionGraph_bailout reason='reached ");
1073 C->log()->text("%s", (iterations >= CG_BUILD_ITER_LIMIT) ? "iterations" : "time");
1074 C->log()->end_elem(" limit'");
1075 }
1076 assert(false, err_msg_res("infinite EA connection graph build (%f sec, %d iterations) with %d nodes and worklist size %d",
1077 time.seconds(), iterations, nodes_size(), ptnodes_worklist.length()));
1078 // Possible infinite build_connection_graph loop,
1079 // bailout (no changes to ideal graph were made).
1080 return false;
1081 }
1082 #ifdef ASSERT
1083 if (Verbose && PrintEscapeAnalysis) {
1084 tty->print_cr("EA: %d iterations to build connection graph with %d nodes and worklist size %d",
1085 iterations, nodes_size(), ptnodes_worklist.length());
1086 }
1087 #endif
1088
1089 #undef CG_BUILD_ITER_LIMIT
1090 #undef CG_BUILD_TIME_LIMIT
1091
1092 // Find fields initialized by NULL for non-escaping Allocations.
1093 int non_escaped_length = non_escaped_worklist.length();
1094 for (int next = 0; next < non_escaped_length; next++) {
1095 JavaObjectNode* ptn = non_escaped_worklist.at(next);
1096 PointsToNode::EscapeState es = ptn->escape_state();
1097 assert(es <= PointsToNode::ArgEscape, "sanity");
1098 if (es == PointsToNode::NoEscape) {
1099 if (find_init_values(ptn, null_obj, _igvn) > 0) {
1100 // Adding references to NULL object does not change escape states
1101 // since it does not escape. Also no fields are added to NULL object.
1102 add_java_object_edges(null_obj, false);
1103 }
1104 }
1105 Node* n = ptn->ideal_node();
1106 if (n->is_Allocate()) {
1107 // The object allocated by this Allocate node will never be
1108 // seen by an other thread. Mark it so that when it is
1109 // expanded no MemBarStoreStore is added.
1110 InitializeNode* ini = n->as_Allocate()->initialization();
1111 if (ini != NULL)
1112 ini->set_does_not_escape();
1113 }
1114 }
1115 return true; // Finished graph construction.
1116 }
1117
1118 // Propagate GlobalEscape and ArgEscape escape states to all nodes
1119 // and check that we still have non-escaping java objects.
1120 bool ConnectionGraph::find_non_escaped_objects(GrowableArray<PointsToNode*>& ptnodes_worklist,
1121 GrowableArray<JavaObjectNode*>& non_escaped_worklist) {
1122 GrowableArray<PointsToNode*> escape_worklist;
1123 // First, put all nodes with GlobalEscape and ArgEscape states on worklist.
1124 int ptnodes_length = ptnodes_worklist.length();
1125 for (int next = 0; next < ptnodes_length; ++next) {
1126 PointsToNode* ptn = ptnodes_worklist.at(next);
1127 if (ptn->escape_state() >= PointsToNode::ArgEscape ||
1128 ptn->fields_escape_state() >= PointsToNode::ArgEscape) {
1129 escape_worklist.push(ptn);
1130 }
1131 }
1132 // Set escape states to referenced nodes (edges list).
1133 while (escape_worklist.length() > 0) {
1134 PointsToNode* ptn = escape_worklist.pop();
1135 PointsToNode::EscapeState es = ptn->escape_state();
1136 PointsToNode::EscapeState field_es = ptn->fields_escape_state();
1137 if (ptn->is_Field() && ptn->as_Field()->is_oop() &&
1138 es >= PointsToNode::ArgEscape) {
1139 // GlobalEscape or ArgEscape state of field means it has unknown value.
1140 if (add_edge(ptn, phantom_obj)) {
1141 // New edge was added
1142 add_field_uses_to_worklist(ptn->as_Field());
1143 }
1144 }
1145 for (EdgeIterator i(ptn); i.has_next(); i.next()) {
1146 PointsToNode* e = i.get();
1147 if (e->is_Arraycopy()) {
1148 assert(ptn->arraycopy_dst(), "sanity");
1149 // Propagate only fields escape state through arraycopy edge.
1150 if (e->fields_escape_state() < field_es) {
1151 set_fields_escape_state(e, field_es);
1152 escape_worklist.push(e);
1153 }
1154 } else if (es >= field_es) {
1155 // fields_escape_state is also set to 'es' if it is less than 'es'.
1156 if (e->escape_state() < es) {
1157 set_escape_state(e, es);
1158 escape_worklist.push(e);
1159 }
1160 } else {
1161 // Propagate field escape state.
1162 bool es_changed = false;
1163 if (e->fields_escape_state() < field_es) {
1164 set_fields_escape_state(e, field_es);
1165 es_changed = true;
1166 }
1167 if ((e->escape_state() < field_es) &&
1168 e->is_Field() && ptn->is_JavaObject() &&
1169 e->as_Field()->is_oop()) {
1170 // Change escape state of referenced fileds.
1171 set_escape_state(e, field_es);
1172 es_changed = true;;
1173 } else if (e->escape_state() < es) {
1174 set_escape_state(e, es);
1175 es_changed = true;;
1176 }
1177 if (es_changed) {
1178 escape_worklist.push(e);
1179 }
1180 }
1181 }
1182 }
1183 // Remove escaped objects from non_escaped list.
1184 for (int next = non_escaped_worklist.length()-1; next >= 0 ; --next) {
1185 JavaObjectNode* ptn = non_escaped_worklist.at(next);
1186 if (ptn->escape_state() >= PointsToNode::GlobalEscape) {
1187 non_escaped_worklist.delete_at(next);
1188 }
1189 if (ptn->escape_state() == PointsToNode::NoEscape) {
1190 // Find fields in non-escaped allocations which have unknown value.
1191 find_init_values(ptn, phantom_obj, NULL);
1192 }
1193 }
1194 return (non_escaped_worklist.length() > 0);
1195 }
1196
1197 // Add all references to JavaObject node by walking over all uses.
1198 int ConnectionGraph::add_java_object_edges(JavaObjectNode* jobj, bool populate_worklist) {
1199 int new_edges = 0;
1200 if (populate_worklist) {
1201 // Populate _worklist by uses of jobj's uses.
1202 for (UseIterator i(jobj); i.has_next(); i.next()) {
1203 PointsToNode* use = i.get();
1204 if (use->is_Arraycopy())
1205 continue;
1206 add_uses_to_worklist(use);
1207 if (use->is_Field() && use->as_Field()->is_oop()) {
1208 // Put on worklist all field's uses (loads) and
1209 // related field nodes (same base and offset).
1210 add_field_uses_to_worklist(use->as_Field());
1211 }
1212 }
1213 }
1214 while(_worklist.length() > 0) {
1215 PointsToNode* use = _worklist.pop();
1216 if (PointsToNode::is_base_use(use)) {
1217 // Add reference from jobj to field and from field to jobj (field's base).
1218 use = PointsToNode::get_use_node(use)->as_Field();
1219 if (add_base(use->as_Field(), jobj)) {
1220 new_edges++;
1221 }
1222 continue;
1223 }
1224 assert(!use->is_JavaObject(), "sanity");
1225 if (use->is_Arraycopy()) {
1226 if (jobj == null_obj) // NULL object does not have field edges
1227 continue;
1228 // Added edge from Arraycopy node to arraycopy's source java object
1229 if (add_edge(use, jobj)) {
1230 jobj->set_arraycopy_src();
1231 new_edges++;
1232 }
1233 // and stop here.
1234 continue;
1235 }
1236 if (!add_edge(use, jobj))
1237 continue; // No new edge added, there was such edge already.
1238 new_edges++;
1239 if (use->is_LocalVar()) {
1240 add_uses_to_worklist(use);
1241 if (use->arraycopy_dst()) {
1242 for (EdgeIterator i(use); i.has_next(); i.next()) {
1243 PointsToNode* e = i.get();
1244 if (e->is_Arraycopy()) {
1245 if (jobj == null_obj) // NULL object does not have field edges
1246 continue;
1247 // Add edge from arraycopy's destination java object to Arraycopy node.
1248 if (add_edge(jobj, e)) {
1249 new_edges++;
1250 jobj->set_arraycopy_dst();
1251 }
1252 }
1253 }
1254 }
1255 } else {
1256 // Added new edge to stored in field values.
1257 // Put on worklist all field's uses (loads) and
1258 // related field nodes (same base and offset).
1259 add_field_uses_to_worklist(use->as_Field());
1260 }
1261 }
1262 return new_edges;
1263 }
1264
1265 // Put on worklist all related field nodes.
1266 void ConnectionGraph::add_field_uses_to_worklist(FieldNode* field) {
1267 assert(field->is_oop(), "sanity");
1268 int offset = field->offset();
1269 add_uses_to_worklist(field);
1270 // Loop over all bases of this field and push on worklist Field nodes
1271 // with the same offset and base (since they may reference the same field).
1272 for (BaseIterator i(field); i.has_next(); i.next()) {
1273 PointsToNode* base = i.get();
1274 add_fields_to_worklist(field, base);
1275 // Check if the base was source object of arraycopy and go over arraycopy's
1276 // destination objects since values stored to a field of source object are
1277 // accessable by uses (loads) of fields of destination objects.
1278 if (base->arraycopy_src()) {
1279 for (UseIterator j(base); j.has_next(); j.next()) {
1280 PointsToNode* arycp = j.get();
1281 if (arycp->is_Arraycopy()) {
1282 for (UseIterator k(arycp); k.has_next(); k.next()) {
1283 PointsToNode* abase = k.get();
1284 if (abase->arraycopy_dst() && abase != base) {
1285 // Look for the same arracopy reference.
1286 add_fields_to_worklist(field, abase);
1287 }
1288 }
1289 }
1290 }
1291 }
1292 }
1293 }
1294
1295 // Put on worklist all related field nodes.
1296 void ConnectionGraph::add_fields_to_worklist(FieldNode* field, PointsToNode* base) {
1297 int offset = field->offset();
1298 if (base->is_LocalVar()) {
1299 for (UseIterator j(base); j.has_next(); j.next()) {
1300 PointsToNode* f = j.get();
1301 if (PointsToNode::is_base_use(f)) { // Field
1302 f = PointsToNode::get_use_node(f);
1303 if (f == field || !f->as_Field()->is_oop())
1304 continue;
1305 int offs = f->as_Field()->offset();
1306 if (offs == offset || offset == Type::OffsetBot || offs == Type::OffsetBot) {
1307 add_to_worklist(f);
1308 }
1309 }
1310 }
1311 } else {
1312 assert(base->is_JavaObject(), "sanity");
1313 if (// Skip phantom_object since it is only used to indicate that
1314 // this field's content globally escapes.
1315 (base != phantom_obj) &&
1316 // NULL object node does not have fields.
1317 (base != null_obj)) {
1318 for (EdgeIterator i(base); i.has_next(); i.next()) {
1319 PointsToNode* f = i.get();
1320 // Skip arraycopy edge since store to destination object field
1321 // does not update value in source object field.
1322 if (f->is_Arraycopy()) {
1323 assert(base->arraycopy_dst(), "sanity");
1324 continue;
1325 }
1326 if (f == field || !f->as_Field()->is_oop())
1327 continue;
1328 int offs = f->as_Field()->offset();
1329 if (offs == offset || offset == Type::OffsetBot || offs == Type::OffsetBot) {
1330 add_to_worklist(f);
1331 }
1332 }
1333 }
1334 }
1335 }
1336
1337 // Find fields which have unknown value.
1338 int ConnectionGraph::find_field_value(FieldNode* field) {
1339 // Escaped fields should have init value already.
1340 assert(field->escape_state() == PointsToNode::NoEscape, "sanity");
1341 int new_edges = 0;
1342 for (BaseIterator i(field); i.has_next(); i.next()) {
1343 PointsToNode* base = i.get();
1344 if (base->is_JavaObject()) {
1345 // Skip Allocate's fields which will be processed later.
1346 if (base->ideal_node()->is_Allocate())
1347 return 0;
1348 assert(base == null_obj, "only NULL ptr base expected here");
1349 }
1350 }
1351 if (add_edge(field, phantom_obj)) {
1352 // New edge was added
1353 new_edges++;
1354 add_field_uses_to_worklist(field);
1355 }
1356 return new_edges;
1357 }
1358
1359 // Find fields initializing values for allocations.
1360 int ConnectionGraph::find_init_values(JavaObjectNode* pta, PointsToNode* init_val, PhaseTransform* phase) {
1361 assert(pta->escape_state() == PointsToNode::NoEscape, "Not escaped Allocate nodes only");
1362 int new_edges = 0;
1363 Node* alloc = pta->ideal_node();
1364 if (init_val == phantom_obj) {
1365 // Do nothing for Allocate nodes since its fields values are "known".
1366 if (alloc->is_Allocate())
1367 return 0;
1368 assert(alloc->as_CallStaticJava(), "sanity");
1369 #ifdef ASSERT
1370 if (alloc->as_CallStaticJava()->method() == NULL) {
1371 const char* name = alloc->as_CallStaticJava()->_name;
1372 assert(strncmp(name, "_multianewarray", 15) == 0, "sanity");
1373 }
1374 #endif
1375 // Non-escaped allocation returned from Java or runtime call have
1376 // unknown values in fields.
1377 for (EdgeIterator i(pta); i.has_next(); i.next()) {
1378 PointsToNode* ptn = i.get();
1379 if (ptn->is_Field() && ptn->as_Field()->is_oop()) {
1380 if (add_edge(ptn, phantom_obj)) {
1381 // New edge was added
1382 new_edges++;
1383 add_field_uses_to_worklist(ptn->as_Field());
1384 }
1385 }
1386 }
1387 return new_edges;
1388 }
1389 assert(init_val == null_obj, "sanity");
1390 // Do nothing for Call nodes since its fields values are unknown.
1391 if (!alloc->is_Allocate())
1392 return 0;
1393
1394 InitializeNode* ini = alloc->as_Allocate()->initialization();
1395 Compile* C = _compile;
1396 bool visited_bottom_offset = false;
1397 GrowableArray<int> offsets_worklist;
1398
1399 // Check if an oop field's initializing value is recorded and add
1400 // a corresponding NULL if field's value if it is not recorded.
1401 // Connection Graph does not record a default initialization by NULL
1402 // captured by Initialize node.
1403 //
1404 for (EdgeIterator i(pta); i.has_next(); i.next()) {
1405 PointsToNode* ptn = i.get(); // Field (AddP)
1406 if (!ptn->is_Field() || !ptn->as_Field()->is_oop())
1407 continue; // Not oop field
1408 int offset = ptn->as_Field()->offset();
1409 if (offset == Type::OffsetBot) {
1410 if (!visited_bottom_offset) {
1411 // OffsetBot is used to reference array's element,
1412 // always add reference to NULL to all Field nodes since we don't
1413 // known which element is referenced.
1414 if (add_edge(ptn, null_obj)) {
1415 // New edge was added
1416 new_edges++;
1417 add_field_uses_to_worklist(ptn->as_Field());
1418 visited_bottom_offset = true;
1419 }
1420 }
1421 } else {
1422 // Check only oop fields.
1423 const Type* adr_type = ptn->ideal_node()->as_AddP()->bottom_type();
1424 if (adr_type->isa_rawptr()) {
1425 #ifdef ASSERT
1426 // Raw pointers are used for initializing stores so skip it
1427 // since it should be recorded already
1428 Node* base = get_addp_base(ptn->ideal_node());
1429 assert(adr_type->isa_rawptr() && base->is_Proj() &&
1430 (base->in(0) == alloc),"unexpected pointer type");
1431 #endif
1432 continue;
1433 }
1434 if (!offsets_worklist.contains(offset)) {
1435 offsets_worklist.append(offset);
1436 Node* value = NULL;
1437 if (ini != NULL) {
1438 BasicType ft = UseCompressedOops ? T_NARROWOOP : T_OBJECT;
1439 Node* store = ini->find_captured_store(offset, type2aelembytes(ft), phase);
1440 if (store != NULL && store->is_Store()) {
1441 value = store->in(MemNode::ValueIn);
1442 } else {
1443 // There could be initializing stores which follow allocation.
1444 // For example, a volatile field store is not collected
1445 // by Initialize node.
1446 //
1447 // Need to check for dependent loads to separate such stores from
1448 // stores which follow loads. For now, add initial value NULL so
1449 // that compare pointers optimization works correctly.
1450 }
1451 }
1452 if (value == NULL) {
1453 // A field's initializing value was not recorded. Add NULL.
1454 if (add_edge(ptn, null_obj)) {
1455 // New edge was added
1456 new_edges++;
1457 add_field_uses_to_worklist(ptn->as_Field());
1458 }
1459 }
1460 }
1461 }
1462 }
1463 return new_edges;
1464 }
1465
1466 // Adjust scalar_replaceable state after Connection Graph is built.
1467 void ConnectionGraph::adjust_scalar_replaceable_state(JavaObjectNode* jobj) {
1468 // Search for non-escaping objects which are not scalar replaceable
1469 // and mark them to propagate the state to referenced objects.
1470
1471 // 1. An object is not scalar replaceable if the field into which it is
1472 // stored has unknown offset (stored into unknown element of an array).
1473 //
1474 for (UseIterator i(jobj); i.has_next(); i.next()) {
1475 PointsToNode* use = i.get();
1476 assert(!use->is_Arraycopy(), "sanity");
1477 if (use->is_Field()) {
1478 FieldNode* field = use->as_Field();
1479 assert(field->is_oop() && field->scalar_replaceable() &&
1480 field->fields_escape_state() == PointsToNode::NoEscape, "sanity");
1481 if (field->offset() == Type::OffsetBot) {
1482 jobj->set_scalar_replaceable(false);
1483 return;
1484 }
1485 }
1486 assert(use->is_Field() || use->is_LocalVar(), "sanity");
1487 // 2. An object is not scalar replaceable if it is merged with other objects.
1488 for (EdgeIterator j(use); j.has_next(); j.next()) {
1489 PointsToNode* ptn = j.get();
1490 if (ptn->is_JavaObject() && ptn != jobj) {
1491 // Mark all objects.
1492 jobj->set_scalar_replaceable(false);
1493 ptn->set_scalar_replaceable(false);
1494 }
1495 }
1496 if (!jobj->scalar_replaceable()) {
1497 return;
1498 }
1499 }
1500
1501 for (EdgeIterator j(jobj); j.has_next(); j.next()) {
1502 // Non-escaping object node should point only to field nodes.
1503 FieldNode* field = j.get()->as_Field();
1504 int offset = field->as_Field()->offset();
1505
1506 // 3. An object is not scalar replaceable if it has a field with unknown
1507 // offset (array's element is accessed in loop).
1508 if (offset == Type::OffsetBot) {
1509 jobj->set_scalar_replaceable(false);
1510 return;
1511 }
1512 // 4. Currently an object is not scalar replaceable if a LoadStore node
1513 // access its field since the field value is unknown after it.
1514 //
1515 Node* n = field->ideal_node();
1516 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1517 if (n->fast_out(i)->is_LoadStore()) {
1518 jobj->set_scalar_replaceable(false);
1519 return;
1520 }
1521 }
1522
1523 // 5. Or the address may point to more then one object. This may produce
1524 // the false positive result (set not scalar replaceable)
1525 // since the flow-insensitive escape analysis can't separate
1526 // the case when stores overwrite the field's value from the case
1527 // when stores happened on different control branches.
1528 //
1529 // Note: it will disable scalar replacement in some cases:
1530 //
1531 // Point p[] = new Point[1];
1532 // p[0] = new Point(); // Will be not scalar replaced
1533 //
1534 // but it will save us from incorrect optimizations in next cases:
1535 //
1536 // Point p[] = new Point[1];
1537 // if ( x ) p[0] = new Point(); // Will be not scalar replaced
1538 //
1539 if (field->base_count() > 1) {
1540 for (BaseIterator i(field); i.has_next(); i.next()) {
1541 PointsToNode* base = i.get();
1542 // Don't take into account LocalVar nodes which
1543 // may point to only one object which should be also
1544 // this field's base by now.
1545 if (base->is_JavaObject() && base != jobj) {
1546 // Mark all bases.
1547 jobj->set_scalar_replaceable(false);
1548 base->set_scalar_replaceable(false);
1549 }
1550 }
1551 }
1552 }
1553 }
1554
1555 #ifdef ASSERT
1556 void ConnectionGraph::verify_connection_graph(
1557 GrowableArray<PointsToNode*>& ptnodes_worklist,
1558 GrowableArray<JavaObjectNode*>& non_escaped_worklist,
1559 GrowableArray<JavaObjectNode*>& java_objects_worklist,
1560 GrowableArray<Node*>& addp_worklist) {
1561 // Verify that graph is complete - no new edges could be added.
1562 int java_objects_length = java_objects_worklist.length();
1563 int non_escaped_length = non_escaped_worklist.length();
1564 int new_edges = 0;
1565 for (int next = 0; next < java_objects_length; ++next) {
1566 JavaObjectNode* ptn = java_objects_worklist.at(next);
1567 new_edges += add_java_object_edges(ptn, true);
1568 }
1569 assert(new_edges == 0, "graph was not complete");
1570 // Verify that escape state is final.
1571 int length = non_escaped_worklist.length();
1572 find_non_escaped_objects(ptnodes_worklist, non_escaped_worklist);
1573 assert((non_escaped_length == non_escaped_worklist.length()) &&
1574 (non_escaped_length == length) &&
1575 (_worklist.length() == 0), "escape state was not final");
1576
1577 // Verify fields information.
1578 int addp_length = addp_worklist.length();
1579 for (int next = 0; next < addp_length; ++next ) {
1580 Node* n = addp_worklist.at(next);
1581 FieldNode* field = ptnode_adr(n->_idx)->as_Field();
1582 if (field->is_oop()) {
1583 // Verify that field has all bases
1584 Node* base = get_addp_base(n);
1585 PointsToNode* ptn = ptnode_adr(base->_idx);
1586 if (ptn->is_JavaObject()) {
1587 assert(field->has_base(ptn->as_JavaObject()), "sanity");
1588 } else {
1589 assert(ptn->is_LocalVar(), "sanity");
1590 for (EdgeIterator i(ptn); i.has_next(); i.next()) {
1591 PointsToNode* e = i.get();
1592 if (e->is_JavaObject()) {
1593 assert(field->has_base(e->as_JavaObject()), "sanity");
1594 }
1595 }
1596 }
1597 // Verify that all fields have initializing values.
1598 if (field->edge_count() == 0) {
1599 field->dump();
1600 assert(field->edge_count() > 0, "sanity");
1601 }
1602 }
1603 }
1604 }
1605 #endif
1606
1607 // Optimize ideal graph.
1608 void ConnectionGraph::optimize_ideal_graph(GrowableArray<Node*>& ptr_cmp_worklist,
1609 GrowableArray<Node*>& storestore_worklist) {
1610 Compile* C = _compile;
1611 PhaseIterGVN* igvn = _igvn;
1612 if (EliminateLocks) {
1613 // Mark locks before changing ideal graph.
1614 int cnt = C->macro_count();
1615 for( int i=0; i < cnt; i++ ) {
1616 Node *n = C->macro_node(i);
1617 if (n->is_AbstractLock()) { // Lock and Unlock nodes
1618 AbstractLockNode* alock = n->as_AbstractLock();
1619 if (!alock->is_non_esc_obj()) {
1620 if (not_global_escape(alock->obj_node())) {
1621 assert(!alock->is_eliminated() || alock->is_coarsened(), "sanity");
1622 // The lock could be marked eliminated by lock coarsening
1623 // code during first IGVN before EA. Replace coarsened flag
1624 // to eliminate all associated locks/unlocks.
1625 alock->set_non_esc_obj();
1626 }
1627 }
1628 }
1629 }
1630 }
1631
1632 if (OptimizePtrCompare) {
1633 // Add ConI(#CC_GT) and ConI(#CC_EQ).
1634 _pcmp_neq = igvn->makecon(TypeInt::CC_GT);
1635 _pcmp_eq = igvn->makecon(TypeInt::CC_EQ);
1636 // Optimize objects compare.
1637 while (ptr_cmp_worklist.length() != 0) {
1638 Node *n = ptr_cmp_worklist.pop();
1639 Node *res = optimize_ptr_compare(n);
1640 if (res != NULL) {
1641 #ifndef PRODUCT
1642 if (PrintOptimizePtrCompare) {
1643 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"));
1644 if (Verbose) {
1645 n->dump(1);
1646 }
1647 }
1648 #endif
1649 igvn->replace_node(n, res);
1650 }
1651 }
1652 // cleanup
1653 if (_pcmp_neq->outcnt() == 0)
1654 igvn->hash_delete(_pcmp_neq);
1655 if (_pcmp_eq->outcnt() == 0)
1656 igvn->hash_delete(_pcmp_eq);
1657 }
1658
1659 // For MemBarStoreStore nodes added in library_call.cpp, check
1660 // escape status of associated AllocateNode and optimize out
1661 // MemBarStoreStore node if the allocated object never escapes.
1662 while (storestore_worklist.length() != 0) {
1663 Node *n = storestore_worklist.pop();
1664 MemBarStoreStoreNode *storestore = n ->as_MemBarStoreStore();
1665 Node *alloc = storestore->in(MemBarNode::Precedent)->in(0);
1666 assert (alloc->is_Allocate(), "storestore should point to AllocateNode");
1667 if (not_global_escape(alloc)) {
1668 MemBarNode* mb = MemBarNode::make(C, Op_MemBarCPUOrder, Compile::AliasIdxBot);
1669 mb->init_req(TypeFunc::Memory, storestore->in(TypeFunc::Memory));
1670 mb->init_req(TypeFunc::Control, storestore->in(TypeFunc::Control));
1671 igvn->register_new_node_with_optimizer(mb);
1672 igvn->replace_node(storestore, mb);
1673 }
1674 }
1675 }
1676
1677 // Optimize objects compare.
1678 Node* ConnectionGraph::optimize_ptr_compare(Node* n) {
1679 assert(OptimizePtrCompare, "sanity");
1680 PointsToNode* ptn1 = ptnode_adr(n->in(1)->_idx);
1681 PointsToNode* ptn2 = ptnode_adr(n->in(2)->_idx);
1682 JavaObjectNode* jobj1 = unique_java_object(n->in(1));
1683 JavaObjectNode* jobj2 = unique_java_object(n->in(2));
1684 assert(ptn1->is_JavaObject() || ptn1->is_LocalVar(), "sanity");
1685 assert(ptn2->is_JavaObject() || ptn2->is_LocalVar(), "sanity");
1686
1687 // Check simple cases first.
1688 if (jobj1 != NULL) {
1689 if (jobj1->escape_state() == PointsToNode::NoEscape) {
1690 if (jobj1 == jobj2) {
1691 // Comparing the same not escaping object.
1692 return _pcmp_eq;
1693 }
1694 Node* obj = jobj1->ideal_node();
1695 // Comparing not escaping allocation.
1696 if ((obj->is_Allocate() || obj->is_CallStaticJava()) &&
1697 !ptn2->points_to(jobj1)) {
1698 return _pcmp_neq; // This includes nullness check.
1699 }
1700 }
1701 }
1702 if (jobj2 != NULL) {
1703 if (jobj2->escape_state() == PointsToNode::NoEscape) {
1704 Node* obj = jobj2->ideal_node();
1705 // Comparing not escaping allocation.
1706 if ((obj->is_Allocate() || obj->is_CallStaticJava()) &&
1707 !ptn1->points_to(jobj2)) {
1708 return _pcmp_neq; // This includes nullness check.
1709 }
1710 }
1711 }
1712 if (jobj1 != NULL && jobj1 != phantom_obj &&
1713 jobj2 != NULL && jobj2 != phantom_obj &&
1714 jobj1->ideal_node()->is_Con() &&
1715 jobj2->ideal_node()->is_Con()) {
1716 // Klass or String constants compare. Need to be careful with
1717 // compressed pointers - compare types of ConN and ConP instead of nodes.
1718 const Type* t1 = jobj1->ideal_node()->bottom_type()->make_ptr();
1719 const Type* t2 = jobj2->ideal_node()->bottom_type()->make_ptr();
1720 assert(t1 != NULL && t2 != NULL, "sanity");
1721 if (t1->make_ptr() == t2->make_ptr()) {
1722 return _pcmp_eq;
1723 } else {
1724 return _pcmp_neq;
1725 }
1726 }
1727 if (ptn1->meet(ptn2)) {
1728 return NULL; // Sets are not disjoint
1729 }
1730
1731 // Sets are disjoint.
1732 bool set1_has_unknown_ptr = ptn1->points_to(phantom_obj);
1733 bool set2_has_unknown_ptr = ptn2->points_to(phantom_obj);
1734 bool set1_has_null_ptr = ptn1->points_to(null_obj);
1735 bool set2_has_null_ptr = ptn2->points_to(null_obj);
1736 if (set1_has_unknown_ptr && set2_has_null_ptr ||
1737 set2_has_unknown_ptr && set1_has_null_ptr) {
1738 // Check nullness of unknown object.
1739 return NULL;
1740 }
1741
1742 // Disjointness by itself is not sufficient since
1743 // alias analysis is not complete for escaped objects.
1744 // Disjoint sets are definitely unrelated only when
1745 // at least one set has only not escaping allocations.
1746 if (!set1_has_unknown_ptr && !set1_has_null_ptr) {
1747 if (ptn1->non_escaping_allocation()) {
1748 return _pcmp_neq;
1749 }
1750 }
1751 if (!set2_has_unknown_ptr && !set2_has_null_ptr) {
1752 if (ptn2->non_escaping_allocation()) {
1753 return _pcmp_neq;
1754 }
1755 }
1756 return NULL;
1757 }
1758
1759 // Connection Graph constuction functions.
1760
1761 void ConnectionGraph::add_local_var(Node *n, PointsToNode::EscapeState es) {
1762 PointsToNode* ptadr = _nodes.at(n->_idx);
1763 if (ptadr != NULL) {
1764 assert(ptadr->is_LocalVar() && ptadr->ideal_node() == n, "sanity");
1765 return;
1766 }
1767 Compile* C = _compile;
1768 ptadr = new (C->comp_arena()) LocalVarNode(C, n, es);
1769 _nodes.at_put(n->_idx, ptadr);
1770 }
1771
1772 void ConnectionGraph::add_java_object(Node *n, PointsToNode::EscapeState es) {
1773 PointsToNode* ptadr = _nodes.at(n->_idx);
1774 if (ptadr != NULL) {
1775 assert(ptadr->is_JavaObject() && ptadr->ideal_node() == n, "sanity");
1776 return;
1777 }
1778 Compile* C = _compile;
1779 ptadr = new (C->comp_arena()) JavaObjectNode(C, n, es);
1780 _nodes.at_put(n->_idx, ptadr);
1781 }
1782
1783 void ConnectionGraph::add_field(Node *n, PointsToNode::EscapeState es, int offset) {
1784 PointsToNode* ptadr = _nodes.at(n->_idx);
1785 if (ptadr != NULL) {
1786 assert(ptadr->is_Field() && ptadr->ideal_node() == n, "sanity");
1787 return;
1788 }
1789 bool unsafe = false;
1790 bool is_oop = is_oop_field(n, offset, &unsafe);
1791 if (unsafe) {
1792 es = PointsToNode::GlobalEscape;
1793 }
1794 Compile* C = _compile;
1795 FieldNode* field = new (C->comp_arena()) FieldNode(C, n, es, offset, is_oop);
1796 _nodes.at_put(n->_idx, field);
1797 }
1798
1799 void ConnectionGraph::add_arraycopy(Node *n, PointsToNode::EscapeState es,
1800 PointsToNode* src, PointsToNode* dst) {
1801 assert(!src->is_Field() && !dst->is_Field(), "only for JavaObject and LocalVar");
1802 assert((src != null_obj) && (dst != null_obj), "not for ConP NULL");
1803 PointsToNode* ptadr = _nodes.at(n->_idx);
1804 if (ptadr != NULL) {
1805 assert(ptadr->is_Arraycopy() && ptadr->ideal_node() == n, "sanity");
1806 return;
1807 }
1808 Compile* C = _compile;
1809 ptadr = new (C->comp_arena()) ArraycopyNode(C, n, es);
1810 _nodes.at_put(n->_idx, ptadr);
1811 // Add edge from arraycopy node to source object.
1812 (void)add_edge(ptadr, src);
1813 src->set_arraycopy_src();
1814 // Add edge from destination object to arraycopy node.
1815 (void)add_edge(dst, ptadr);
1816 dst->set_arraycopy_dst();
1817 }
1818
1819 bool ConnectionGraph::is_oop_field(Node* n, int offset, bool* unsafe) {
1820 const Type* adr_type = n->as_AddP()->bottom_type();
1821 BasicType bt = T_INT;
1822 if (offset == Type::OffsetBot) {
1823 // Check only oop fields.
1824 if (!adr_type->isa_aryptr() ||
1825 (adr_type->isa_aryptr()->klass() == NULL) ||
1826 adr_type->isa_aryptr()->klass()->is_obj_array_klass()) {
1827 // OffsetBot is used to reference array's element. Ignore first AddP.
1828 if (find_second_addp(n, n->in(AddPNode::Base)) == NULL) {
1829 bt = T_OBJECT;
1830 }
1831 }
1832 } else if (offset != oopDesc::klass_offset_in_bytes()) {
1833 if (adr_type->isa_instptr()) {
1834 ciField* field = _compile->alias_type(adr_type->isa_instptr())->field();
1835 if (field != NULL) {
1836 bt = field->layout_type();
1837 } else {
1838 // Check for unsafe oop field access
1839 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1840 int opcode = n->fast_out(i)->Opcode();
1841 if (opcode == Op_StoreP || opcode == Op_LoadP ||
1842 opcode == Op_StoreN || opcode == Op_LoadN) {
1843 bt = T_OBJECT;
1844 (*unsafe) = true;
1845 break;
1846 }
1847 }
1848 }
1849 } else if (adr_type->isa_aryptr()) {
1850 if (offset == arrayOopDesc::length_offset_in_bytes()) {
1851 // Ignore array length load.
1852 } else if (find_second_addp(n, n->in(AddPNode::Base)) != NULL) {
1853 // Ignore first AddP.
1854 } else {
1855 const Type* elemtype = adr_type->isa_aryptr()->elem();
1856 bt = elemtype->array_element_basic_type();
1857 }
1858 } else if (adr_type->isa_rawptr() || adr_type->isa_klassptr()) {
1859 // Allocation initialization, ThreadLocal field access, unsafe access
1860 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1861 int opcode = n->fast_out(i)->Opcode();
1862 if (opcode == Op_StoreP || opcode == Op_LoadP ||
1863 opcode == Op_StoreN || opcode == Op_LoadN) {
1864 bt = T_OBJECT;
1865 break;
1866 }
1867 }
1868 }
1869 }
1870 return (bt == T_OBJECT || bt == T_NARROWOOP || bt == T_ARRAY);
1871 }
1872
1873 // Returns unique pointed java object or NULL.
1874 JavaObjectNode* ConnectionGraph::unique_java_object(Node *n) {
1875 assert(!_collecting, "should not call when contructed graph");
1876 // If the node was created after the escape computation we can't answer.
1877 uint idx = n->_idx;
1878 if (idx >= nodes_size()) {
1879 return NULL;
1880 }
1881 PointsToNode* ptn = ptnode_adr(idx);
1882 if (ptn->is_JavaObject()) {
1883 return ptn->as_JavaObject();
1884 }
1885 assert(ptn->is_LocalVar(), "sanity");
1886 // Check all java objects it points to.
1887 JavaObjectNode* jobj = NULL;
1888 for (EdgeIterator i(ptn); i.has_next(); i.next()) {
1889 PointsToNode* e = i.get();
1890 if (e->is_JavaObject()) {
1891 if (jobj == NULL) {
1892 jobj = e->as_JavaObject();
1893 } else if (jobj != e) {
1894 return NULL;
1895 }
1896 }
1897 }
1898 return jobj;
1899 }
1900
1901 // Return true if this node points only to non-escaping allocations.
1902 bool PointsToNode::non_escaping_allocation() {
1903 if (is_JavaObject()) {
1904 Node* n = ideal_node();
1905 if (n->is_Allocate() || n->is_CallStaticJava()) {
1906 return (escape_state() == PointsToNode::NoEscape);
1907 } else {
1908 return false;
1909 }
1910 }
1911 assert(is_LocalVar(), "sanity");
1912 // Check all java objects it points to.
1913 for (EdgeIterator i(this); i.has_next(); i.next()) {
1914 PointsToNode* e = i.get();
1915 if (e->is_JavaObject()) {
1916 Node* n = e->ideal_node();
1917 if ((e->escape_state() != PointsToNode::NoEscape) ||
1918 !(n->is_Allocate() || n->is_CallStaticJava())) {
1919 return false;
1920 }
1921 }
1922 }
1923 return true;
1924 }
1925
1926 // Return true if we know the node does not escape globally.
1927 bool ConnectionGraph::not_global_escape(Node *n) {
1928 assert(!_collecting, "should not call during graph construction");
1929 // If the node was created after the escape computation we can't answer.
1930 uint idx = n->_idx;
1931 if (idx >= nodes_size()) {
1932 return false;
1933 }
1934 PointsToNode* ptn = ptnode_adr(idx);
1935 PointsToNode::EscapeState es = ptn->escape_state();
1936 // If we have already computed a value, return it.
1937 if (es >= PointsToNode::GlobalEscape)
1938 return false;
1939 if (ptn->is_JavaObject()) {
1940 return true; // (es < PointsToNode::GlobalEscape);
1941 }
1942 assert(ptn->is_LocalVar(), "sanity");
1943 // Check all java objects it points to.
1944 for (EdgeIterator i(ptn); i.has_next(); i.next()) {
1945 if (i.get()->escape_state() >= PointsToNode::GlobalEscape)
1946 return false;
1947 }
1948 return true;
1949 }
1950
1951
1952 // Helper functions
1953
1954 // Return true if this node points to specified node or nodes it points to.
1955 bool PointsToNode::points_to(JavaObjectNode* ptn) const {
1956 if (is_JavaObject()) {
1957 return (this == ptn);
1958 }
1959 assert(is_LocalVar(), "sanity");
1960 for (EdgeIterator i(this); i.has_next(); i.next()) {
1961 if (i.get() == ptn)
1962 return true;
1963 }
1964 return false;
1965 }
1966
1967 // Return true if one node points to an other.
1968 bool PointsToNode::meet(PointsToNode* ptn) {
1969 if (this == ptn) {
1970 return true;
1971 } else if (ptn->is_JavaObject()) {
1972 return this->points_to(ptn->as_JavaObject());
1973 } else if (this->is_JavaObject()) {
1974 return ptn->points_to(this->as_JavaObject());
1975 }
1976 assert(this->is_LocalVar() && ptn->is_LocalVar(), "sanity");
1977 int ptn_count = ptn->edge_count();
1978 for (EdgeIterator i(this); i.has_next(); i.next()) {
1979 PointsToNode* this_e = i.get();
1980 for (int j = 0; j < ptn_count; j++) {
1981 if (this_e == ptn->edge(j))
1982 return true;
1983 }
1984 }
1985 return false;
1986 }
1987
1988 #ifdef ASSERT
1989 // Return true if bases point to this java object.
1990 bool FieldNode::has_base(JavaObjectNode* jobj) const {
1991 for (BaseIterator i(this); i.has_next(); i.next()) {
1992 if (i.get() == jobj)
1993 return true;
1994 }
1995 return false;
1996 }
1997 #endif
1998
1999 int ConnectionGraph::address_offset(Node* adr, PhaseTransform *phase) {
2000 const Type *adr_type = phase->type(adr);
2001 if (adr->is_AddP() && adr_type->isa_oopptr() == NULL &&
2002 adr->in(AddPNode::Address)->is_Proj() &&
2003 adr->in(AddPNode::Address)->in(0)->is_Allocate()) {
2004 // We are computing a raw address for a store captured by an Initialize
2005 // compute an appropriate address type. AddP cases #3 and #5 (see below).
2006 int offs = (int)phase->find_intptr_t_con(adr->in(AddPNode::Offset), Type::OffsetBot);
2007 assert(offs != Type::OffsetBot ||
2008 adr->in(AddPNode::Address)->in(0)->is_AllocateArray(),
2009 "offset must be a constant or it is initialization of array");
2010 return offs;
2011 }
2012 const TypePtr *t_ptr = adr_type->isa_ptr();
2013 assert(t_ptr != NULL, "must be a pointer type");
2014 return t_ptr->offset();
2015 }
2016
2017 Node* ConnectionGraph::get_addp_base(Node *addp) {
2018 assert(addp->is_AddP(), "must be AddP");
2019 //
2020 // AddP cases for Base and Address inputs:
2021 // case #1. Direct object's field reference:
2022 // Allocate
2023 // |
2024 // Proj #5 ( oop result )
2025 // |
2026 // CheckCastPP (cast to instance type)
2027 // | |
2028 // AddP ( base == address )
2029 //
2030 // case #2. Indirect object's field reference:
2031 // Phi
2032 // |
2033 // CastPP (cast to instance type)
2034 // | |
2035 // AddP ( base == address )
2036 //
2037 // case #3. Raw object's field reference for Initialize node:
2038 // Allocate
2039 // |
2040 // Proj #5 ( oop result )
2041 // top |
2042 // \ |
2043 // AddP ( base == top )
2044 //
2045 // case #4. Array's element reference:
2046 // {CheckCastPP | CastPP}
2047 // | | |
2048 // | AddP ( array's element offset )
2049 // | |
2050 // AddP ( array's offset )
2051 //
2052 // case #5. Raw object's field reference for arraycopy stub call:
2053 // The inline_native_clone() case when the arraycopy stub is called
2054 // after the allocation before Initialize and CheckCastPP nodes.
2055 // Allocate
2056 // |
2057 // Proj #5 ( oop result )
2058 // | |
2059 // AddP ( base == address )
2060 //
2061 // case #6. Constant Pool, ThreadLocal, CastX2P or
2062 // Raw object's field reference:
2063 // {ConP, ThreadLocal, CastX2P, raw Load}
2064 // top |
2065 // \ |
2066 // AddP ( base == top )
2067 //
2068 // case #7. Klass's field reference.
2069 // LoadKlass
2070 // | |
2071 // AddP ( base == address )
2072 //
2073 // case #8. narrow Klass's field reference.
2074 // LoadNKlass
2075 // |
2076 // DecodeN
2077 // | |
2078 // AddP ( base == address )
2079 //
2080 Node *base = addp->in(AddPNode::Base);
2081 if (base->uncast()->is_top()) { // The AddP case #3 and #6.
2082 base = addp->in(AddPNode::Address);
2083 while (base->is_AddP()) {
2084 // Case #6 (unsafe access) may have several chained AddP nodes.
2085 assert(base->in(AddPNode::Base)->uncast()->is_top(), "expected unsafe access address only");
2086 base = base->in(AddPNode::Address);
2087 }
2088 Node* uncast_base = base->uncast();
2089 int opcode = uncast_base->Opcode();
2090 assert(opcode == Op_ConP || opcode == Op_ThreadLocal ||
2091 opcode == Op_CastX2P || uncast_base->is_DecodeN() ||
2092 (uncast_base->is_Mem() && uncast_base->bottom_type() == TypeRawPtr::NOTNULL) ||
2093 (uncast_base->is_Proj() && uncast_base->in(0)->is_Allocate()), "sanity");
2094 }
2095 return base;
2096 }
2097
2098 Node* ConnectionGraph::find_second_addp(Node* addp, Node* n) {
2099 assert(addp->is_AddP() && addp->outcnt() > 0, "Don't process dead nodes");
2100 Node* addp2 = addp->raw_out(0);
2101 if (addp->outcnt() == 1 && addp2->is_AddP() &&
2102 addp2->in(AddPNode::Base) == n &&
2103 addp2->in(AddPNode::Address) == addp) {
2104 assert(addp->in(AddPNode::Base) == n, "expecting the same base");
2105 //
2106 // Find array's offset to push it on worklist first and
2107 // as result process an array's element offset first (pushed second)
2108 // to avoid CastPP for the array's offset.
2109 // Otherwise the inserted CastPP (LocalVar) will point to what
2110 // the AddP (Field) points to. Which would be wrong since
2111 // the algorithm expects the CastPP has the same point as
2112 // as AddP's base CheckCastPP (LocalVar).
2113 //
2114 // ArrayAllocation
2115 // |
2116 // CheckCastPP
2117 // |
2118 // memProj (from ArrayAllocation CheckCastPP)
2119 // | ||
2120 // | || Int (element index)
2121 // | || | ConI (log(element size))
2122 // | || | /
2123 // | || LShift
2124 // | || /
2125 // | AddP (array's element offset)
2126 // | |
2127 // | | ConI (array's offset: #12(32-bits) or #24(64-bits))
2128 // | / /
2129 // AddP (array's offset)
2130 // |
2131 // Load/Store (memory operation on array's element)
2132 //
2133 return addp2;
2134 }
2135 return NULL;
2136 }
2137
2138 //
2139 // Adjust the type and inputs of an AddP which computes the
2140 // address of a field of an instance
2141 //
2142 bool ConnectionGraph::split_AddP(Node *addp, Node *base) {
2143 PhaseGVN* igvn = _igvn;
2144 const TypeOopPtr *base_t = igvn->type(base)->isa_oopptr();
2145 assert(base_t != NULL && base_t->is_known_instance(), "expecting instance oopptr");
2146 const TypeOopPtr *t = igvn->type(addp)->isa_oopptr();
2147 if (t == NULL) {
2148 // We are computing a raw address for a store captured by an Initialize
2149 // compute an appropriate address type (cases #3 and #5).
2150 assert(igvn->type(addp) == TypeRawPtr::NOTNULL, "must be raw pointer");
2151 assert(addp->in(AddPNode::Address)->is_Proj(), "base of raw address must be result projection from allocation");
2152 intptr_t offs = (int)igvn->find_intptr_t_con(addp->in(AddPNode::Offset), Type::OffsetBot);
2153 assert(offs != Type::OffsetBot, "offset must be a constant");
2154 t = base_t->add_offset(offs)->is_oopptr();
2155 }
2156 int inst_id = base_t->instance_id();
2157 assert(!t->is_known_instance() || t->instance_id() == inst_id,
2158 "old type must be non-instance or match new type");
2159
2160 // The type 't' could be subclass of 'base_t'.
2161 // As result t->offset() could be large then base_t's size and it will
2162 // cause the failure in add_offset() with narrow oops since TypeOopPtr()
2163 // constructor verifies correctness of the offset.
2164 //
2165 // It could happened on subclass's branch (from the type profiling
2166 // inlining) which was not eliminated during parsing since the exactness
2167 // of the allocation type was not propagated to the subclass type check.
2168 //
2169 // Or the type 't' could be not related to 'base_t' at all.
2170 // It could happened when CHA type is different from MDO type on a dead path
2171 // (for example, from instanceof check) which is not collapsed during parsing.
2172 //
2173 // Do nothing for such AddP node and don't process its users since
2174 // this code branch will go away.
2175 //
2176 if (!t->is_known_instance() &&
2177 !base_t->klass()->is_subtype_of(t->klass())) {
2178 return false; // bail out
2179 }
2180 const TypeOopPtr *tinst = base_t->add_offset(t->offset())->is_oopptr();
2181 // Do NOT remove the next line: ensure a new alias index is allocated
2182 // for the instance type. Note: C++ will not remove it since the call
2183 // has side effect.
2184 int alias_idx = _compile->get_alias_index(tinst);
2185 igvn->set_type(addp, tinst);
2186 // record the allocation in the node map
2187 set_map(addp, get_map(base->_idx));
2188 // Set addp's Base and Address to 'base'.
2189 Node *abase = addp->in(AddPNode::Base);
2190 Node *adr = addp->in(AddPNode::Address);
2191 if (adr->is_Proj() && adr->in(0)->is_Allocate() &&
2192 adr->in(0)->_idx == (uint)inst_id) {
2193 // Skip AddP cases #3 and #5.
2194 } else {
2195 assert(!abase->is_top(), "sanity"); // AddP case #3
2196 if (abase != base) {
2197 igvn->hash_delete(addp);
2198 addp->set_req(AddPNode::Base, base);
2199 if (abase == adr) {
2200 addp->set_req(AddPNode::Address, base);
2201 } else {
2202 // AddP case #4 (adr is array's element offset AddP node)
2203 #ifdef ASSERT
2204 const TypeOopPtr *atype = igvn->type(adr)->isa_oopptr();
2205 assert(adr->is_AddP() && atype != NULL &&
2206 atype->instance_id() == inst_id, "array's element offset should be processed first");
2207 #endif
2208 }
2209 igvn->hash_insert(addp);
2210 }
2211 }
2212 // Put on IGVN worklist since at least addp's type was changed above.
2213 record_for_optimizer(addp);
2214 return true;
2215 }
2216
2217 //
2218 // Create a new version of orig_phi if necessary. Returns either the newly
2219 // created phi or an existing phi. Sets create_new to indicate whether a new
2220 // phi was created. Cache the last newly created phi in the node map.
2221 //
2222 PhiNode *ConnectionGraph::create_split_phi(PhiNode *orig_phi, int alias_idx, GrowableArray<PhiNode *> &orig_phi_worklist, bool &new_created) {
2223 Compile *C = _compile;
2224 PhaseGVN* igvn = _igvn;
2225 new_created = false;
2226 int phi_alias_idx = C->get_alias_index(orig_phi->adr_type());
2227 // nothing to do if orig_phi is bottom memory or matches alias_idx
2228 if (phi_alias_idx == alias_idx) {
2229 return orig_phi;
2230 }
2231 // Have we recently created a Phi for this alias index?
2232 PhiNode *result = get_map_phi(orig_phi->_idx);
2233 if (result != NULL && C->get_alias_index(result->adr_type()) == alias_idx) {
2234 return result;
2235 }
2236 // Previous check may fail when the same wide memory Phi was split into Phis
2237 // for different memory slices. Search all Phis for this region.
2238 if (result != NULL) {
2239 Node* region = orig_phi->in(0);
2240 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
2241 Node* phi = region->fast_out(i);
2242 if (phi->is_Phi() &&
2243 C->get_alias_index(phi->as_Phi()->adr_type()) == alias_idx) {
2244 assert(phi->_idx >= nodes_size(), "only new Phi per instance memory slice");
2245 return phi->as_Phi();
2246 }
2247 }
2248 }
2249 if ((int)C->unique() + 2*NodeLimitFudgeFactor > MaxNodeLimit) {
2250 if (C->do_escape_analysis() == true && !C->failing()) {
2251 // Retry compilation without escape analysis.
2252 // If this is the first failure, the sentinel string will "stick"
2253 // to the Compile object, and the C2Compiler will see it and retry.
2254 C->record_failure(C2Compiler::retry_no_escape_analysis());
2255 }
2256 return NULL;
2257 }
2258 orig_phi_worklist.append_if_missing(orig_phi);
2259 const TypePtr *atype = C->get_adr_type(alias_idx);
2260 result = PhiNode::make(orig_phi->in(0), NULL, Type::MEMORY, atype);
2261 C->copy_node_notes_to(result, orig_phi);
2262 igvn->set_type(result, result->bottom_type());
2263 record_for_optimizer(result);
2264 set_map(orig_phi, result);
2265 new_created = true;
2266 return result;
2267 }
2268
2269 //
2270 // Return a new version of Memory Phi "orig_phi" with the inputs having the
2271 // specified alias index.
2272 //
2273 PhiNode *ConnectionGraph::split_memory_phi(PhiNode *orig_phi, int alias_idx, GrowableArray<PhiNode *> &orig_phi_worklist) {
2274 assert(alias_idx != Compile::AliasIdxBot, "can't split out bottom memory");
2275 Compile *C = _compile;
2276 PhaseGVN* igvn = _igvn;
2277 bool new_phi_created;
2278 PhiNode *result = create_split_phi(orig_phi, alias_idx, orig_phi_worklist, new_phi_created);
2279 if (!new_phi_created) {
2280 return result;
2281 }
2282 GrowableArray<PhiNode *> phi_list;
2283 GrowableArray<uint> cur_input;
2284 PhiNode *phi = orig_phi;
2285 uint idx = 1;
2286 bool finished = false;
2287 while(!finished) {
2288 while (idx < phi->req()) {
2289 Node *mem = find_inst_mem(phi->in(idx), alias_idx, orig_phi_worklist);
2290 if (mem != NULL && mem->is_Phi()) {
2291 PhiNode *newphi = create_split_phi(mem->as_Phi(), alias_idx, orig_phi_worklist, new_phi_created);
2292 if (new_phi_created) {
2293 // found an phi for which we created a new split, push current one on worklist and begin
2294 // processing new one
2295 phi_list.push(phi);
2296 cur_input.push(idx);
2297 phi = mem->as_Phi();
2298 result = newphi;
2299 idx = 1;
2300 continue;
2301 } else {
2302 mem = newphi;
2303 }
2304 }
2305 if (C->failing()) {
2306 return NULL;
2307 }
2308 result->set_req(idx++, mem);
2309 }
2310 #ifdef ASSERT
2311 // verify that the new Phi has an input for each input of the original
2312 assert( phi->req() == result->req(), "must have same number of inputs.");
2313 assert( result->in(0) != NULL && result->in(0) == phi->in(0), "regions must match");
2314 #endif
2315 // Check if all new phi's inputs have specified alias index.
2316 // Otherwise use old phi.
2317 for (uint i = 1; i < phi->req(); i++) {
2318 Node* in = result->in(i);
2319 assert((phi->in(i) == NULL) == (in == NULL), "inputs must correspond.");
2320 }
2321 // we have finished processing a Phi, see if there are any more to do
2322 finished = (phi_list.length() == 0 );
2323 if (!finished) {
2324 phi = phi_list.pop();
2325 idx = cur_input.pop();
2326 PhiNode *prev_result = get_map_phi(phi->_idx);
2327 prev_result->set_req(idx++, result);
2328 result = prev_result;
2329 }
2330 }
2331 return result;
2332 }
2333
2334 //
2335 // The next methods are derived from methods in MemNode.
2336 //
2337 Node* ConnectionGraph::step_through_mergemem(MergeMemNode *mmem, int alias_idx, const TypeOopPtr *toop) {
2338 Node *mem = mmem;
2339 // TypeOopPtr::NOTNULL+any is an OOP with unknown offset - generally
2340 // means an array I have not precisely typed yet. Do not do any
2341 // alias stuff with it any time soon.
2342 if (toop->base() != Type::AnyPtr &&
2343 !(toop->klass() != NULL &&
2344 toop->klass()->is_java_lang_Object() &&
2345 toop->offset() == Type::OffsetBot)) {
2346 mem = mmem->memory_at(alias_idx);
2347 // Update input if it is progress over what we have now
2348 }
2349 return mem;
2350 }
2351
2352 //
2353 // Move memory users to their memory slices.
2354 //
2355 void ConnectionGraph::move_inst_mem(Node* n, GrowableArray<PhiNode *> &orig_phis) {
2356 Compile* C = _compile;
2357 PhaseGVN* igvn = _igvn;
2358 const TypePtr* tp = igvn->type(n->in(MemNode::Address))->isa_ptr();
2359 assert(tp != NULL, "ptr type");
2360 int alias_idx = C->get_alias_index(tp);
2361 int general_idx = C->get_general_index(alias_idx);
2362
2363 // Move users first
2364 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
2365 Node* use = n->fast_out(i);
2366 if (use->is_MergeMem()) {
2367 MergeMemNode* mmem = use->as_MergeMem();
2368 assert(n == mmem->memory_at(alias_idx), "should be on instance memory slice");
2369 if (n != mmem->memory_at(general_idx) || alias_idx == general_idx) {
2370 continue; // Nothing to do
2371 }
2372 // Replace previous general reference to mem node.
2373 uint orig_uniq = C->unique();
2374 Node* m = find_inst_mem(n, general_idx, orig_phis);
2375 assert(orig_uniq == C->unique(), "no new nodes");
2376 mmem->set_memory_at(general_idx, m);
2377 --imax;
2378 --i;
2379 } else if (use->is_MemBar()) {
2380 assert(!use->is_Initialize(), "initializing stores should not be moved");
2381 if (use->req() > MemBarNode::Precedent &&
2382 use->in(MemBarNode::Precedent) == n) {
2383 // Don't move related membars.
2384 record_for_optimizer(use);
2385 continue;
2386 }
2387 tp = use->as_MemBar()->adr_type()->isa_ptr();
2388 if (tp != NULL && C->get_alias_index(tp) == alias_idx ||
2389 alias_idx == general_idx) {
2390 continue; // Nothing to do
2391 }
2392 // Move to general memory slice.
2393 uint orig_uniq = C->unique();
2394 Node* m = find_inst_mem(n, general_idx, orig_phis);
2395 assert(orig_uniq == C->unique(), "no new nodes");
2396 igvn->hash_delete(use);
2397 imax -= use->replace_edge(n, m);
2398 igvn->hash_insert(use);
2399 record_for_optimizer(use);
2400 --i;
2401 #ifdef ASSERT
2402 } else if (use->is_Mem()) {
2403 if (use->Opcode() == Op_StoreCM && use->in(MemNode::OopStore) == n) {
2404 // Don't move related cardmark.
2405 continue;
2406 }
2407 // Memory nodes should have new memory input.
2408 tp = igvn->type(use->in(MemNode::Address))->isa_ptr();
2409 assert(tp != NULL, "ptr type");
2410 int idx = C->get_alias_index(tp);
2411 assert(get_map(use->_idx) != NULL || idx == alias_idx,
2412 "Following memory nodes should have new memory input or be on the same memory slice");
2413 } else if (use->is_Phi()) {
2414 // Phi nodes should be split and moved already.
2415 tp = use->as_Phi()->adr_type()->isa_ptr();
2416 assert(tp != NULL, "ptr type");
2417 int idx = C->get_alias_index(tp);
2418 assert(idx == alias_idx, "Following Phi nodes should be on the same memory slice");
2419 } else {
2420 use->dump();
2421 assert(false, "should not be here");
2422 #endif
2423 }
2424 }
2425 }
2426
2427 //
2428 // Search memory chain of "mem" to find a MemNode whose address
2429 // is the specified alias index.
2430 //
2431 Node* ConnectionGraph::find_inst_mem(Node *orig_mem, int alias_idx, GrowableArray<PhiNode *> &orig_phis) {
2432 if (orig_mem == NULL)
2433 return orig_mem;
2434 Compile* C = _compile;
2435 PhaseGVN* igvn = _igvn;
2436 const TypeOopPtr *toop = C->get_adr_type(alias_idx)->isa_oopptr();
2437 bool is_instance = (toop != NULL) && toop->is_known_instance();
2438 Node *start_mem = C->start()->proj_out(TypeFunc::Memory);
2439 Node *prev = NULL;
2440 Node *result = orig_mem;
2441 while (prev != result) {
2442 prev = result;
2443 if (result == start_mem)
2444 break; // hit one of our sentinels
2445 if (result->is_Mem()) {
2446 const Type *at = igvn->type(result->in(MemNode::Address));
2447 if (at == Type::TOP)
2448 break; // Dead
2449 assert (at->isa_ptr() != NULL, "pointer type required.");
2450 int idx = C->get_alias_index(at->is_ptr());
2451 if (idx == alias_idx)
2452 break; // Found
2453 if (!is_instance && (at->isa_oopptr() == NULL ||
2454 !at->is_oopptr()->is_known_instance())) {
2455 break; // Do not skip store to general memory slice.
2456 }
2457 result = result->in(MemNode::Memory);
2458 }
2459 if (!is_instance)
2460 continue; // don't search further for non-instance types
2461 // skip over a call which does not affect this memory slice
2462 if (result->is_Proj() && result->as_Proj()->_con == TypeFunc::Memory) {
2463 Node *proj_in = result->in(0);
2464 if (proj_in->is_Allocate() && proj_in->_idx == (uint)toop->instance_id()) {
2465 break; // hit one of our sentinels
2466 } else if (proj_in->is_Call()) {
2467 CallNode *call = proj_in->as_Call();
2468 if (!call->may_modify(toop, igvn)) {
2469 result = call->in(TypeFunc::Memory);
2470 }
2471 } else if (proj_in->is_Initialize()) {
2472 AllocateNode* alloc = proj_in->as_Initialize()->allocation();
2473 // Stop if this is the initialization for the object instance which
2474 // which contains this memory slice, otherwise skip over it.
2475 if (alloc == NULL || alloc->_idx != (uint)toop->instance_id()) {
2476 result = proj_in->in(TypeFunc::Memory);
2477 }
2478 } else if (proj_in->is_MemBar()) {
2479 result = proj_in->in(TypeFunc::Memory);
2480 }
2481 } else if (result->is_MergeMem()) {
2482 MergeMemNode *mmem = result->as_MergeMem();
2483 result = step_through_mergemem(mmem, alias_idx, toop);
2484 if (result == mmem->base_memory()) {
2485 // Didn't find instance memory, search through general slice recursively.
2486 result = mmem->memory_at(C->get_general_index(alias_idx));
2487 result = find_inst_mem(result, alias_idx, orig_phis);
2488 if (C->failing()) {
2489 return NULL;
2490 }
2491 mmem->set_memory_at(alias_idx, result);
2492 }
2493 } else if (result->is_Phi() &&
2494 C->get_alias_index(result->as_Phi()->adr_type()) != alias_idx) {
2495 Node *un = result->as_Phi()->unique_input(igvn);
2496 if (un != NULL) {
2497 orig_phis.append_if_missing(result->as_Phi());
2498 result = un;
2499 } else {
2500 break;
2501 }
2502 } else if (result->is_ClearArray()) {
2503 if (!ClearArrayNode::step_through(&result, (uint)toop->instance_id(), igvn)) {
2504 // Can not bypass initialization of the instance
2505 // we are looking for.
2506 break;
2507 }
2508 // Otherwise skip it (the call updated 'result' value).
2509 } else if (result->Opcode() == Op_SCMemProj) {
2510 assert(result->in(0)->is_LoadStore(), "sanity");
2511 const Type *at = igvn->type(result->in(0)->in(MemNode::Address));
2512 if (at != Type::TOP) {
2513 assert (at->isa_ptr() != NULL, "pointer type required.");
2514 int idx = C->get_alias_index(at->is_ptr());
2515 assert(idx != alias_idx, "Object is not scalar replaceable if a LoadStore node access its field");
2516 break;
2517 }
2518 result = result->in(0)->in(MemNode::Memory);
2519 }
2520 }
2521 if (result->is_Phi()) {
2522 PhiNode *mphi = result->as_Phi();
2523 assert(mphi->bottom_type() == Type::MEMORY, "memory phi required");
2524 const TypePtr *t = mphi->adr_type();
2525 if (!is_instance) {
2526 // Push all non-instance Phis on the orig_phis worklist to update inputs
2527 // during Phase 4 if needed.
2528 orig_phis.append_if_missing(mphi);
2529 } else if (C->get_alias_index(t) != alias_idx) {
2530 // Create a new Phi with the specified alias index type.
2531 result = split_memory_phi(mphi, alias_idx, orig_phis);
2532 }
2533 }
2534 // the result is either MemNode, PhiNode, InitializeNode.
2535 return result;
2536 }
2537
2538 //
2539 // Convert the types of unescaped object to instance types where possible,
2540 // propagate the new type information through the graph, and update memory
2541 // edges and MergeMem inputs to reflect the new type.
2542 //
2543 // We start with allocations (and calls which may be allocations) on alloc_worklist.
2544 // The processing is done in 4 phases:
2545 //
2546 // Phase 1: Process possible allocations from alloc_worklist. Create instance
2547 // types for the CheckCastPP for allocations where possible.
2548 // Propagate the the new types through users as follows:
2549 // casts and Phi: push users on alloc_worklist
2550 // AddP: cast Base and Address inputs to the instance type
2551 // push any AddP users on alloc_worklist and push any memnode
2552 // users onto memnode_worklist.
2553 // Phase 2: Process MemNode's from memnode_worklist. compute new address type and
2554 // search the Memory chain for a store with the appropriate type
2555 // address type. If a Phi is found, create a new version with
2556 // the appropriate memory slices from each of the Phi inputs.
2557 // For stores, process the users as follows:
2558 // MemNode: push on memnode_worklist
2559 // MergeMem: push on mergemem_worklist
2560 // Phase 3: Process MergeMem nodes from mergemem_worklist. Walk each memory slice
2561 // moving the first node encountered of each instance type to the
2562 // the input corresponding to its alias index.
2563 // appropriate memory slice.
2564 // Phase 4: Update the inputs of non-instance memory Phis and the Memory input of memnodes.
2565 //
2566 // In the following example, the CheckCastPP nodes are the cast of allocation
2567 // results and the allocation of node 29 is unescaped and eligible to be an
2568 // instance type.
2569 //
2570 // We start with:
2571 //
2572 // 7 Parm #memory
2573 // 10 ConI "12"
2574 // 19 CheckCastPP "Foo"
2575 // 20 AddP _ 19 19 10 Foo+12 alias_index=4
2576 // 29 CheckCastPP "Foo"
2577 // 30 AddP _ 29 29 10 Foo+12 alias_index=4
2578 //
2579 // 40 StoreP 25 7 20 ... alias_index=4
2580 // 50 StoreP 35 40 30 ... alias_index=4
2581 // 60 StoreP 45 50 20 ... alias_index=4
2582 // 70 LoadP _ 60 30 ... alias_index=4
2583 // 80 Phi 75 50 60 Memory alias_index=4
2584 // 90 LoadP _ 80 30 ... alias_index=4
2585 // 100 LoadP _ 80 20 ... alias_index=4
2586 //
2587 //
2588 // Phase 1 creates an instance type for node 29 assigning it an instance id of 24
2589 // and creating a new alias index for node 30. This gives:
2590 //
2591 // 7 Parm #memory
2592 // 10 ConI "12"
2593 // 19 CheckCastPP "Foo"
2594 // 20 AddP _ 19 19 10 Foo+12 alias_index=4
2595 // 29 CheckCastPP "Foo" iid=24
2596 // 30 AddP _ 29 29 10 Foo+12 alias_index=6 iid=24
2597 //
2598 // 40 StoreP 25 7 20 ... alias_index=4
2599 // 50 StoreP 35 40 30 ... alias_index=6
2600 // 60 StoreP 45 50 20 ... alias_index=4
2601 // 70 LoadP _ 60 30 ... alias_index=6
2602 // 80 Phi 75 50 60 Memory alias_index=4
2603 // 90 LoadP _ 80 30 ... alias_index=6
2604 // 100 LoadP _ 80 20 ... alias_index=4
2605 //
2606 // In phase 2, new memory inputs are computed for the loads and stores,
2607 // And a new version of the phi is created. In phase 4, the inputs to
2608 // node 80 are updated and then the memory nodes are updated with the
2609 // values computed in phase 2. This results in:
2610 //
2611 // 7 Parm #memory
2612 // 10 ConI "12"
2613 // 19 CheckCastPP "Foo"
2614 // 20 AddP _ 19 19 10 Foo+12 alias_index=4
2615 // 29 CheckCastPP "Foo" iid=24
2616 // 30 AddP _ 29 29 10 Foo+12 alias_index=6 iid=24
2617 //
2618 // 40 StoreP 25 7 20 ... alias_index=4
2619 // 50 StoreP 35 7 30 ... alias_index=6
2620 // 60 StoreP 45 40 20 ... alias_index=4
2621 // 70 LoadP _ 50 30 ... alias_index=6
2622 // 80 Phi 75 40 60 Memory alias_index=4
2623 // 120 Phi 75 50 50 Memory alias_index=6
2624 // 90 LoadP _ 120 30 ... alias_index=6
2625 // 100 LoadP _ 80 20 ... alias_index=4
2626 //
2627 void ConnectionGraph::split_unique_types(GrowableArray<Node *> &alloc_worklist) {
2628 GrowableArray<Node *> memnode_worklist;
2629 GrowableArray<PhiNode *> orig_phis;
2630 PhaseIterGVN *igvn = _igvn;
2631 uint new_index_start = (uint) _compile->num_alias_types();
2632 Arena* arena = Thread::current()->resource_area();
2633 VectorSet visited(arena);
2634 ideal_nodes.clear(); // Reset for use with set_map/get_map.
2635 uint unique_old = _compile->unique();
2636
2637 // Phase 1: Process possible allocations from alloc_worklist.
2638 // Create instance types for the CheckCastPP for allocations where possible.
2639 //
2640 // (Note: don't forget to change the order of the second AddP node on
2641 // the alloc_worklist if the order of the worklist processing is changed,
2642 // see the comment in find_second_addp().)
2643 //
2644 while (alloc_worklist.length() != 0) {
2645 Node *n = alloc_worklist.pop();
2646 uint ni = n->_idx;
2647 if (n->is_Call()) {
2648 CallNode *alloc = n->as_Call();
2649 // copy escape information to call node
2650 PointsToNode* ptn = ptnode_adr(alloc->_idx);
2651 PointsToNode::EscapeState es = ptn->escape_state();
2652 // We have an allocation or call which returns a Java object,
2653 // see if it is unescaped.
2654 if (es != PointsToNode::NoEscape || !ptn->scalar_replaceable())
2655 continue;
2656 // Find CheckCastPP for the allocate or for the return value of a call
2657 n = alloc->result_cast();
2658 if (n == NULL) { // No uses except Initialize node
2659 if (alloc->is_Allocate()) {
2660 // Set the scalar_replaceable flag for allocation
2661 // so it could be eliminated if it has no uses.
2662 alloc->as_Allocate()->_is_scalar_replaceable = true;
2663 }
2664 continue;
2665 }
2666 if (!n->is_CheckCastPP()) { // not unique CheckCastPP.
2667 assert(!alloc->is_Allocate(), "allocation should have unique type");
2668 continue;
2669 }
2670
2671 // The inline code for Object.clone() casts the allocation result to
2672 // java.lang.Object and then to the actual type of the allocated
2673 // object. Detect this case and use the second cast.
2674 // Also detect j.l.reflect.Array.newInstance(jobject, jint) case when
2675 // the allocation result is cast to java.lang.Object and then
2676 // to the actual Array type.
2677 if (alloc->is_Allocate() && n->as_Type()->type() == TypeInstPtr::NOTNULL
2678 && (alloc->is_AllocateArray() ||
2679 igvn->type(alloc->in(AllocateNode::KlassNode)) != TypeKlassPtr::OBJECT)) {
2680 Node *cast2 = NULL;
2681 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
2682 Node *use = n->fast_out(i);
2683 if (use->is_CheckCastPP()) {
2684 cast2 = use;
2685 break;
2686 }
2687 }
2688 if (cast2 != NULL) {
2689 n = cast2;
2690 } else {
2691 // Non-scalar replaceable if the allocation type is unknown statically
2692 // (reflection allocation), the object can't be restored during
2693 // deoptimization without precise type.
2694 continue;
2695 }
2696 }
2697 if (alloc->is_Allocate()) {
2698 // Set the scalar_replaceable flag for allocation
2699 // so it could be eliminated.
2700 alloc->as_Allocate()->_is_scalar_replaceable = true;
2701 }
2702 set_escape_state(ptnode_adr(n->_idx), es); // CheckCastPP escape state
2703 // in order for an object to be scalar-replaceable, it must be:
2704 // - a direct allocation (not a call returning an object)
2705 // - non-escaping
2706 // - eligible to be a unique type
2707 // - not determined to be ineligible by escape analysis
2708 set_map(alloc, n);
2709 set_map(n, alloc);
2710 const TypeOopPtr *t = igvn->type(n)->isa_oopptr();
2711 if (t == NULL)
2712 continue; // not a TypeOopPtr
2713 const TypeOopPtr* tinst = t->cast_to_exactness(true)->is_oopptr()->cast_to_instance_id(ni);
2714 igvn->hash_delete(n);
2715 igvn->set_type(n, tinst);
2716 n->raise_bottom_type(tinst);
2717 igvn->hash_insert(n);
2718 record_for_optimizer(n);
2719 if (alloc->is_Allocate() && (t->isa_instptr() || t->isa_aryptr())) {
2720
2721 // First, put on the worklist all Field edges from Connection Graph
2722 // which is more accurate then putting immediate users from Ideal Graph.
2723 for (EdgeIterator e(ptn); e.has_next(); e.next()) {
2724 PointsToNode* tgt = e.get();
2725 Node* use = tgt->ideal_node();
2726 assert(tgt->is_Field() && use->is_AddP(),
2727 "only AddP nodes are Field edges in CG");
2728 if (use->outcnt() > 0) { // Don't process dead nodes
2729 Node* addp2 = find_second_addp(use, use->in(AddPNode::Base));
2730 if (addp2 != NULL) {
2731 assert(alloc->is_AllocateArray(),"array allocation was expected");
2732 alloc_worklist.append_if_missing(addp2);
2733 }
2734 alloc_worklist.append_if_missing(use);
2735 }
2736 }
2737
2738 // An allocation may have an Initialize which has raw stores. Scan
2739 // the users of the raw allocation result and push AddP users
2740 // on alloc_worklist.
2741 Node *raw_result = alloc->proj_out(TypeFunc::Parms);
2742 assert (raw_result != NULL, "must have an allocation result");
2743 for (DUIterator_Fast imax, i = raw_result->fast_outs(imax); i < imax; i++) {
2744 Node *use = raw_result->fast_out(i);
2745 if (use->is_AddP() && use->outcnt() > 0) { // Don't process dead nodes
2746 Node* addp2 = find_second_addp(use, raw_result);
2747 if (addp2 != NULL) {
2748 assert(alloc->is_AllocateArray(),"array allocation was expected");
2749 alloc_worklist.append_if_missing(addp2);
2750 }
2751 alloc_worklist.append_if_missing(use);
2752 } else if (use->is_MemBar()) {
2753 memnode_worklist.append_if_missing(use);
2754 }
2755 }
2756 }
2757 } else if (n->is_AddP()) {
2758 JavaObjectNode* jobj = unique_java_object(get_addp_base(n));
2759 if (jobj == NULL || jobj == phantom_obj) {
2760 #ifdef ASSERT
2761 ptnode_adr(get_addp_base(n)->_idx)->dump();
2762 ptnode_adr(n->_idx)->dump();
2763 assert(jobj != NULL && jobj != phantom_obj, "escaped allocation");
2764 #endif
2765 _compile->record_failure(C2Compiler::retry_no_escape_analysis());
2766 return;
2767 }
2768 Node *base = get_map(jobj->idx()); // CheckCastPP node
2769 if (!split_AddP(n, base)) continue; // wrong type from dead path
2770 } else if (n->is_Phi() ||
2771 n->is_CheckCastPP() ||
2772 n->is_EncodeP() ||
2773 n->is_DecodeN() ||
2774 (n->is_ConstraintCast() && n->Opcode() == Op_CastPP)) {
2775 if (visited.test_set(n->_idx)) {
2776 assert(n->is_Phi(), "loops only through Phi's");
2777 continue; // already processed
2778 }
2779 JavaObjectNode* jobj = unique_java_object(n);
2780 if (jobj == NULL || jobj == phantom_obj) {
2781 #ifdef ASSERT
2782 ptnode_adr(n->_idx)->dump();
2783 assert(jobj != NULL && jobj != phantom_obj, "escaped allocation");
2784 #endif
2785 _compile->record_failure(C2Compiler::retry_no_escape_analysis());
2786 return;
2787 } else {
2788 Node *val = get_map(jobj->idx()); // CheckCastPP node
2789 TypeNode *tn = n->as_Type();
2790 const TypeOopPtr* tinst = igvn->type(val)->isa_oopptr();
2791 assert(tinst != NULL && tinst->is_known_instance() &&
2792 tinst->instance_id() == jobj->idx() , "instance type expected.");
2793
2794 const Type *tn_type = igvn->type(tn);
2795 const TypeOopPtr *tn_t;
2796 if (tn_type->isa_narrowoop()) {
2797 tn_t = tn_type->make_ptr()->isa_oopptr();
2798 } else {
2799 tn_t = tn_type->isa_oopptr();
2800 }
2801 if (tn_t != NULL && tinst->klass()->is_subtype_of(tn_t->klass())) {
2802 if (tn_type->isa_narrowoop()) {
2803 tn_type = tinst->make_narrowoop();
2804 } else {
2805 tn_type = tinst;
2806 }
2807 igvn->hash_delete(tn);
2808 igvn->set_type(tn, tn_type);
2809 tn->set_type(tn_type);
2810 igvn->hash_insert(tn);
2811 record_for_optimizer(n);
2812 } else {
2813 assert(tn_type == TypePtr::NULL_PTR ||
2814 tn_t != NULL && !tinst->klass()->is_subtype_of(tn_t->klass()),
2815 "unexpected type");
2816 continue; // Skip dead path with different type
2817 }
2818 }
2819 } else {
2820 debug_only(n->dump();)
2821 assert(false, "EA: unexpected node");
2822 continue;
2823 }
2824 // push allocation's users on appropriate worklist
2825 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
2826 Node *use = n->fast_out(i);
2827 if(use->is_Mem() && use->in(MemNode::Address) == n) {
2828 // Load/store to instance's field
2829 memnode_worklist.append_if_missing(use);
2830 } else if (use->is_MemBar()) {
2831 memnode_worklist.append_if_missing(use);
2832 } else if (use->is_AddP() && use->outcnt() > 0) { // No dead nodes
2833 Node* addp2 = find_second_addp(use, n);
2834 if (addp2 != NULL) {
2835 alloc_worklist.append_if_missing(addp2);
2836 }
2837 alloc_worklist.append_if_missing(use);
2838 } else if (use->is_Phi() ||
2839 use->is_CheckCastPP() ||
2840 use->is_EncodeP() ||
2841 use->is_DecodeN() ||
2842 (use->is_ConstraintCast() && use->Opcode() == Op_CastPP)) {
2843 alloc_worklist.append_if_missing(use);
2844 #ifdef ASSERT
2845 } else if (use->is_Mem()) {
2846 assert(use->in(MemNode::Address) != n, "EA: missing allocation reference path");
2847 } else if (use->is_MergeMem()) {
2848 assert(_mergemem_worklist.contains(use->as_MergeMem()), "EA: missing MergeMem node in the worklist");
2849 } else if (use->is_SafePoint()) {
2850 // Look for MergeMem nodes for calls which reference unique allocation
2851 // (through CheckCastPP nodes) even for debug info.
2852 Node* m = use->in(TypeFunc::Memory);
2853 if (m->is_MergeMem()) {
2854 assert(_mergemem_worklist.contains(m->as_MergeMem()), "EA: missing MergeMem node in the worklist");
2855 }
2856 } else {
2857 uint op = use->Opcode();
2858 if (!(op == Op_CmpP || op == Op_Conv2B ||
2859 op == Op_CastP2X || op == Op_StoreCM ||
2860 op == Op_FastLock || op == Op_AryEq || op == Op_StrComp ||
2861 op == Op_StrEquals || op == Op_StrIndexOf)) {
2862 n->dump();
2863 use->dump();
2864 assert(false, "EA: missing allocation reference path");
2865 }
2866 #endif
2867 }
2868 }
2869
2870 }
2871 // New alias types were created in split_AddP().
2872 uint new_index_end = (uint) _compile->num_alias_types();
2873 assert(unique_old == _compile->unique(), "there should be no new ideal nodes after Phase 1");
2874
2875 // Phase 2: Process MemNode's from memnode_worklist. compute new address type and
2876 // compute new values for Memory inputs (the Memory inputs are not
2877 // actually updated until phase 4.)
2878 if (memnode_worklist.length() == 0)
2879 return; // nothing to do
2880 while (memnode_worklist.length() != 0) {
2881 Node *n = memnode_worklist.pop();
2882 if (visited.test_set(n->_idx))
2883 continue;
2884 if (n->is_Phi() || n->is_ClearArray()) {
2885 // we don't need to do anything, but the users must be pushed
2886 } else if (n->is_MemBar()) { // Initialize, MemBar nodes
2887 // we don't need to do anything, but the users must be pushed
2888 n = n->as_MemBar()->proj_out(TypeFunc::Memory);
2889 if (n == NULL)
2890 continue;
2891 } else {
2892 assert(n->is_Mem(), "memory node required.");
2893 Node *addr = n->in(MemNode::Address);
2894 const Type *addr_t = igvn->type(addr);
2895 if (addr_t == Type::TOP)
2896 continue;
2897 assert (addr_t->isa_ptr() != NULL, "pointer type required.");
2898 int alias_idx = _compile->get_alias_index(addr_t->is_ptr());
2899 assert ((uint)alias_idx < new_index_end, "wrong alias index");
2900 Node *mem = find_inst_mem(n->in(MemNode::Memory), alias_idx, orig_phis);
2901 if (_compile->failing()) {
2902 return;
2903 }
2904 if (mem != n->in(MemNode::Memory)) {
2905 // We delay the memory edge update since we need old one in
2906 // MergeMem code below when instances memory slices are separated.
2907 set_map(n, mem);
2908 }
2909 if (n->is_Load()) {
2910 continue; // don't push users
2911 } else if (n->is_LoadStore()) {
2912 // get the memory projection
2913 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
2914 Node *use = n->fast_out(i);
2915 if (use->Opcode() == Op_SCMemProj) {
2916 n = use;
2917 break;
2918 }
2919 }
2920 assert(n->Opcode() == Op_SCMemProj, "memory projection required");
2921 }
2922 }
2923 // push user on appropriate worklist
2924 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
2925 Node *use = n->fast_out(i);
2926 if (use->is_Phi() || use->is_ClearArray()) {
2927 memnode_worklist.append_if_missing(use);
2928 } else if(use->is_Mem() && use->in(MemNode::Memory) == n) {
2929 if (use->Opcode() == Op_StoreCM) // Ignore cardmark stores
2930 continue;
2931 memnode_worklist.append_if_missing(use);
2932 } else if (use->is_MemBar()) {
2933 memnode_worklist.append_if_missing(use);
2934 #ifdef ASSERT
2935 } else if(use->is_Mem()) {
2936 assert(use->in(MemNode::Memory) != n, "EA: missing memory path");
2937 } else if (use->is_MergeMem()) {
2938 assert(_mergemem_worklist.contains(use->as_MergeMem()), "EA: missing MergeMem node in the worklist");
2939 } else {
2940 uint op = use->Opcode();
2941 if (!(op == Op_StoreCM ||
2942 (op == Op_CallLeaf && use->as_CallLeaf()->_name != NULL &&
2943 strcmp(use->as_CallLeaf()->_name, "g1_wb_pre") == 0) ||
2944 op == Op_AryEq || op == Op_StrComp ||
2945 op == Op_StrEquals || op == Op_StrIndexOf)) {
2946 n->dump();
2947 use->dump();
2948 assert(false, "EA: missing memory path");
2949 }
2950 #endif
2951 }
2952 }
2953 }
2954
2955 // Phase 3: Process MergeMem nodes from mergemem_worklist.
2956 // Walk each memory slice moving the first node encountered of each
2957 // instance type to the the input corresponding to its alias index.
2958 uint length = _mergemem_worklist.length();
2959 for( uint next = 0; next < length; ++next ) {
2960 MergeMemNode* nmm = _mergemem_worklist.at(next);
2961 assert(!visited.test_set(nmm->_idx), "should not be visited before");
2962 // Note: we don't want to use MergeMemStream here because we only want to
2963 // scan inputs which exist at the start, not ones we add during processing.
2964 // Note 2: MergeMem may already contains instance memory slices added
2965 // during find_inst_mem() call when memory nodes were processed above.
2966 igvn->hash_delete(nmm);
2967 uint nslices = nmm->req();
2968 for (uint i = Compile::AliasIdxRaw+1; i < nslices; i++) {
2969 Node* mem = nmm->in(i);
2970 Node* cur = NULL;
2971 if (mem == NULL || mem->is_top())
2972 continue;
2973 // First, update mergemem by moving memory nodes to corresponding slices
2974 // if their type became more precise since this mergemem was created.
2975 while (mem->is_Mem()) {
2976 const Type *at = igvn->type(mem->in(MemNode::Address));
2977 if (at != Type::TOP) {
2978 assert (at->isa_ptr() != NULL, "pointer type required.");
2979 uint idx = (uint)_compile->get_alias_index(at->is_ptr());
2980 if (idx == i) {
2981 if (cur == NULL)
2982 cur = mem;
2983 } else {
2984 if (idx >= nmm->req() || nmm->is_empty_memory(nmm->in(idx))) {
2985 nmm->set_memory_at(idx, mem);
2986 }
2987 }
2988 }
2989 mem = mem->in(MemNode::Memory);
2990 }
2991 nmm->set_memory_at(i, (cur != NULL) ? cur : mem);
2992 // Find any instance of the current type if we haven't encountered
2993 // already a memory slice of the instance along the memory chain.
2994 for (uint ni = new_index_start; ni < new_index_end; ni++) {
2995 if((uint)_compile->get_general_index(ni) == i) {
2996 Node *m = (ni >= nmm->req()) ? nmm->empty_memory() : nmm->in(ni);
2997 if (nmm->is_empty_memory(m)) {
2998 Node* result = find_inst_mem(mem, ni, orig_phis);
2999 if (_compile->failing()) {
3000 return;
3001 }
3002 nmm->set_memory_at(ni, result);
3003 }
3004 }
3005 }
3006 }
3007 // Find the rest of instances values
3008 for (uint ni = new_index_start; ni < new_index_end; ni++) {
3009 const TypeOopPtr *tinst = _compile->get_adr_type(ni)->isa_oopptr();
3010 Node* result = step_through_mergemem(nmm, ni, tinst);
3011 if (result == nmm->base_memory()) {
3012 // Didn't find instance memory, search through general slice recursively.
3013 result = nmm->memory_at(_compile->get_general_index(ni));
3014 result = find_inst_mem(result, ni, orig_phis);
3015 if (_compile->failing()) {
3016 return;
3017 }
3018 nmm->set_memory_at(ni, result);
3019 }
3020 }
3021 igvn->hash_insert(nmm);
3022 record_for_optimizer(nmm);
3023 }
3024
3025 // Phase 4: Update the inputs of non-instance memory Phis and
3026 // the Memory input of memnodes
3027 // First update the inputs of any non-instance Phi's from
3028 // which we split out an instance Phi. Note we don't have
3029 // to recursively process Phi's encounted on the input memory
3030 // chains as is done in split_memory_phi() since they will
3031 // also be processed here.
3032 for (int j = 0; j < orig_phis.length(); j++) {
3033 PhiNode *phi = orig_phis.at(j);
3034 int alias_idx = _compile->get_alias_index(phi->adr_type());
3035 igvn->hash_delete(phi);
3036 for (uint i = 1; i < phi->req(); i++) {
3037 Node *mem = phi->in(i);
3038 Node *new_mem = find_inst_mem(mem, alias_idx, orig_phis);
3039 if (_compile->failing()) {
3040 return;
3041 }
3042 if (mem != new_mem) {
3043 phi->set_req(i, new_mem);
3044 }
3045 }
3046 igvn->hash_insert(phi);
3047 record_for_optimizer(phi);
3048 }
3049
3050 // Update the memory inputs of MemNodes with the value we computed
3051 // in Phase 2 and move stores memory users to corresponding memory slices.
3052 // Disable memory split verification code until the fix for 6984348.
3053 // Currently it produces false negative results since it does not cover all cases.
3054 #if 0 // ifdef ASSERT
3055 visited.Reset();
3056 Node_Stack old_mems(arena, _compile->unique() >> 2);
3057 #endif
3058 for (uint i = 0; i < ideal_nodes.size(); i++) {
3059 Node* n = ideal_nodes.at(i);
3060 Node* nmem = get_map(n->_idx);
3061 assert(nmem != NULL, "sanity");
3062 if (n->is_Mem()) {
3063 #if 0 // ifdef ASSERT
3064 Node* old_mem = n->in(MemNode::Memory);
3065 if (!visited.test_set(old_mem->_idx)) {
3066 old_mems.push(old_mem, old_mem->outcnt());
3067 }
3068 #endif
3069 assert(n->in(MemNode::Memory) != nmem, "sanity");
3070 if (!n->is_Load()) {
3071 // Move memory users of a store first.
3072 move_inst_mem(n, orig_phis);
3073 }
3074 // Now update memory input
3075 igvn->hash_delete(n);
3076 n->set_req(MemNode::Memory, nmem);
3077 igvn->hash_insert(n);
3078 record_for_optimizer(n);
3079 } else {
3080 assert(n->is_Allocate() || n->is_CheckCastPP() ||
3081 n->is_AddP() || n->is_Phi(), "unknown node used for set_map()");
3082 }
3083 }
3084 #if 0 // ifdef ASSERT
3085 // Verify that memory was split correctly
3086 while (old_mems.is_nonempty()) {
3087 Node* old_mem = old_mems.node();
3088 uint old_cnt = old_mems.index();
3089 old_mems.pop();
3090 assert(old_cnt == old_mem->outcnt(), "old mem could be lost");
3091 }
3092 #endif
3093 }
3094
3095 #ifndef PRODUCT
3096 static const char *node_type_names[] = {
3097 "UnknownType",
3098 "JavaObject",
3099 "LocalVar",
3100 "Field",
3101 "Arraycopy"
3102 };
3103
3104 static const char *esc_names[] = {
3105 "UnknownEscape",
3106 "NoEscape",
3107 "ArgEscape",
3108 "GlobalEscape"
3109 };
3110
3111 void PointsToNode::dump(bool print_state) const {
3112 NodeType nt = node_type();
3113 tty->print("%s ", node_type_names[(int) nt]);
3114 if (print_state) {
3115 EscapeState es = escape_state();
3116 EscapeState fields_es = fields_escape_state();
3117 tty->print("%s(%s) ", esc_names[(int)es], esc_names[(int)fields_es]);
3118 if (nt == PointsToNode::JavaObject && !this->scalar_replaceable())
3119 tty->print("NSR");
3120 }
3121 if (is_Field()) {
3122 FieldNode* f = (FieldNode*)this;
3123 tty->print("(");
3124 for (BaseIterator i(f); i.has_next(); i.next()) {
3125 PointsToNode* b = i.get();
3126 tty->print(" %d%s", b->idx(),(b->is_JavaObject() ? "P" : ""));
3127 }
3128 tty->print(" )");
3129 }
3130 tty->print("[");
3131 for (EdgeIterator i(this); i.has_next(); i.next()) {
3132 PointsToNode* e = i.get();
3133 tty->print(" %d%s%s", e->idx(),(e->is_JavaObject() ? "P" : (e->is_Field() ? "F" : "")), e->is_Arraycopy() ? "cp" : "");
3134 }
3135 tty->print(" [");
3136 for (UseIterator i(this); i.has_next(); i.next()) {
3137 PointsToNode* u = i.get();
3138 bool is_base = false;
3139 if (PointsToNode::is_base_use(u)) {
3140 is_base = true;
3141 u = PointsToNode::get_use_node(u)->as_Field();
3142 }
3143 tty->print(" %d%s%s", u->idx(), is_base ? "b" : "", u->is_Arraycopy() ? "cp" : "");
3144 }
3145 tty->print(" ]] ");
3146 if (_node == NULL)
3147 tty->print_cr("<null>");
3148 else
3149 _node->dump();
3150 }
3151
3152 void ConnectionGraph::dump(GrowableArray<PointsToNode*>& ptnodes_worklist) {
3153 bool first = true;
3154 int ptnodes_length = ptnodes_worklist.length();
3155 for (int i = 0; i < ptnodes_length; i++) {
3156 PointsToNode *ptn = ptnodes_worklist.at(i);
3157 if (ptn == NULL || !ptn->is_JavaObject())
3158 continue;
3159 PointsToNode::EscapeState es = ptn->escape_state();
3160 if (ptn->ideal_node()->is_Allocate() && (es == PointsToNode::NoEscape || Verbose)) {
3161 if (first) {
3162 tty->cr();
3163 tty->print("======== Connection graph for ");
3164 _compile->method()->print_short_name();
3165 tty->cr();
3166 first = false;
3167 }
3168 ptn->dump();
3169 // Print all locals and fields which reference this allocation
3170 for (UseIterator j(ptn); j.has_next(); j.next()) {
3171 PointsToNode* use = j.get();
3172 if (use->is_LocalVar()) {
3173 use->dump(Verbose);
3174 } else if (Verbose) {
3175 use->dump();
3176 }
3177 }
3178 tty->cr();
3179 }
3180 }
3181 }
3182 #endif