1 /*
2 * Copyright (c) 2005, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "ci/bcEscapeAnalyzer.hpp"
27 #include "compiler/compileLog.hpp"
28 #include "gc/shared/c2/barrierSetC2.hpp"
29 #include "libadt/vectset.hpp"
30 #include "memory/allocation.hpp"
31 #include "memory/resourceArea.hpp"
32 #include "opto/c2compiler.hpp"
33 #include "opto/arraycopynode.hpp"
34 #include "opto/callnode.hpp"
35 #include "opto/cfgnode.hpp"
36 #include "opto/compile.hpp"
37 #include "opto/escape.hpp"
38 #include "opto/phaseX.hpp"
39 #include "opto/movenode.hpp"
40 #include "opto/rootnode.hpp"
41 #include "utilities/macros.hpp"
42 #if INCLUDE_G1GC
43 #include "gc/g1/g1ThreadLocalData.hpp"
44 #endif // INCLUDE_G1GC
45 #if INCLUDE_ZGC
46 #include "gc/z/c2/zBarrierSetC2.hpp"
47 #endif
48 #if INCLUDE_SHENANDOAHGC
49 #include "gc/shenandoah/c2/shenandoahBarrierSetC2.hpp"
50 #endif
51
52 ConnectionGraph::ConnectionGraph(Compile * C, PhaseIterGVN *igvn) :
53 _nodes(C->comp_arena(), C->unique(), C->unique(), NULL),
54 _in_worklist(C->comp_arena()),
55 _next_pidx(0),
56 _collecting(true),
57 _verify(false),
58 _compile(C),
59 _igvn(igvn),
60 _node_map(C->comp_arena()) {
61 // Add unknown java object.
62 add_java_object(C->top(), PointsToNode::GlobalEscape);
63 phantom_obj = ptnode_adr(C->top()->_idx)->as_JavaObject();
64 // Add ConP(#NULL) and ConN(#NULL) nodes.
65 Node* oop_null = igvn->zerocon(T_OBJECT);
66 assert(oop_null->_idx < nodes_size(), "should be created already");
67 add_java_object(oop_null, PointsToNode::NoEscape);
68 null_obj = ptnode_adr(oop_null->_idx)->as_JavaObject();
69 if (UseCompressedOops) {
70 Node* noop_null = igvn->zerocon(T_NARROWOOP);
71 assert(noop_null->_idx < nodes_size(), "should be created already");
72 map_ideal_node(noop_null, null_obj);
73 }
74 _pcmp_neq = NULL; // Should be initialized
75 _pcmp_eq = NULL;
76 }
77
78 bool ConnectionGraph::has_candidates(Compile *C) {
79 // EA brings benefits only when the code has allocations and/or locks which
80 // are represented by ideal Macro nodes.
81 int cnt = C->macro_count();
82 for (int i = 0; i < cnt; i++) {
83 Node *n = C->macro_node(i);
84 if (n->is_Allocate())
85 return true;
86 if (n->is_Lock()) {
87 Node* obj = n->as_Lock()->obj_node()->uncast();
88 if (!(obj->is_Parm() || obj->is_Con()))
89 return true;
90 }
91 if (n->is_CallStaticJava() &&
92 n->as_CallStaticJava()->is_boxing_method()) {
93 return true;
94 }
95 }
96 return false;
97 }
98
99 void ConnectionGraph::do_analysis(Compile *C, PhaseIterGVN *igvn) {
100 Compile::TracePhase tp("escapeAnalysis", &Phase::timers[Phase::_t_escapeAnalysis]);
101 ResourceMark rm;
102
103 // Add ConP#NULL and ConN#NULL nodes before ConnectionGraph construction
104 // to create space for them in ConnectionGraph::_nodes[].
105 Node* oop_null = igvn->zerocon(T_OBJECT);
106 Node* noop_null = igvn->zerocon(T_NARROWOOP);
107 ConnectionGraph* congraph = new(C->comp_arena()) ConnectionGraph(C, igvn);
108 // Perform escape analysis
109 if (congraph->compute_escape()) {
110 // There are non escaping objects.
111 C->set_congraph(congraph);
112 }
113 // Cleanup.
114 if (oop_null->outcnt() == 0)
115 igvn->hash_delete(oop_null);
116 if (noop_null->outcnt() == 0)
117 igvn->hash_delete(noop_null);
118 }
119
120 bool ConnectionGraph::compute_escape() {
121 Compile* C = _compile;
122 PhaseGVN* igvn = _igvn;
123
124 // Worklists used by EA.
125 Unique_Node_List delayed_worklist;
126 GrowableArray<Node*> alloc_worklist;
127 GrowableArray<Node*> ptr_cmp_worklist;
128 GrowableArray<Node*> storestore_worklist;
129 GrowableArray<ArrayCopyNode*> arraycopy_worklist;
130 GrowableArray<PointsToNode*> ptnodes_worklist;
131 GrowableArray<JavaObjectNode*> java_objects_worklist;
132 GrowableArray<JavaObjectNode*> non_escaped_worklist;
133 GrowableArray<FieldNode*> oop_fields_worklist;
134 DEBUG_ONLY( GrowableArray<Node*> addp_worklist; )
135
136 { Compile::TracePhase tp("connectionGraph", &Phase::timers[Phase::_t_connectionGraph]);
137
138 // 1. Populate Connection Graph (CG) with PointsTo nodes.
139 ideal_nodes.map(C->live_nodes(), NULL); // preallocate space
140 // Initialize worklist
141 if (C->root() != NULL) {
142 ideal_nodes.push(C->root());
143 }
144 // Processed ideal nodes are unique on ideal_nodes list
145 // but several ideal nodes are mapped to the phantom_obj.
146 // To avoid duplicated entries on the following worklists
147 // add the phantom_obj only once to them.
148 ptnodes_worklist.append(phantom_obj);
149 java_objects_worklist.append(phantom_obj);
150 for( uint next = 0; next < ideal_nodes.size(); ++next ) {
151 Node* n = ideal_nodes.at(next);
152 // Create PointsTo nodes and add them to Connection Graph. Called
153 // only once per ideal node since ideal_nodes is Unique_Node list.
154 add_node_to_connection_graph(n, &delayed_worklist);
155 PointsToNode* ptn = ptnode_adr(n->_idx);
156 if (ptn != NULL && ptn != phantom_obj) {
157 ptnodes_worklist.append(ptn);
158 if (ptn->is_JavaObject()) {
159 java_objects_worklist.append(ptn->as_JavaObject());
160 if ((n->is_Allocate() || n->is_CallStaticJava()) &&
161 (ptn->escape_state() < PointsToNode::GlobalEscape)) {
162 // Only allocations and java static calls results are interesting.
163 non_escaped_worklist.append(ptn->as_JavaObject());
164 }
165 } else if (ptn->is_Field() && ptn->as_Field()->is_oop()) {
166 oop_fields_worklist.append(ptn->as_Field());
167 }
168 }
169 if (n->is_MergeMem()) {
170 // Collect all MergeMem nodes to add memory slices for
171 // scalar replaceable objects in split_unique_types().
172 _mergemem_worklist.append(n->as_MergeMem());
173 } else if (OptimizePtrCompare && n->is_Cmp() &&
174 (n->Opcode() == Op_CmpP || n->Opcode() == Op_CmpN)) {
175 // Collect compare pointers nodes.
176 ptr_cmp_worklist.append(n);
177 } else if (n->is_MemBarStoreStore()) {
178 // Collect all MemBarStoreStore nodes so that depending on the
179 // escape status of the associated Allocate node some of them
180 // may be eliminated.
181 storestore_worklist.append(n);
182 } else if (n->is_MemBar() && (n->Opcode() == Op_MemBarRelease) &&
183 (n->req() > MemBarNode::Precedent)) {
184 record_for_optimizer(n);
185 #ifdef ASSERT
186 } else if (n->is_AddP()) {
187 // Collect address nodes for graph verification.
188 addp_worklist.append(n);
189 #endif
190 } else if (n->is_ArrayCopy()) {
191 // Keep a list of ArrayCopy nodes so if one of its input is non
192 // escaping, we can record a unique type
193 arraycopy_worklist.append(n->as_ArrayCopy());
194 }
195 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
196 Node* m = n->fast_out(i); // Get user
197 ideal_nodes.push(m);
198 }
199 }
200 if (non_escaped_worklist.length() == 0) {
201 _collecting = false;
202 return false; // Nothing to do.
203 }
204 // Add final simple edges to graph.
205 while(delayed_worklist.size() > 0) {
206 Node* n = delayed_worklist.pop();
207 add_final_edges(n);
208 }
209 int ptnodes_length = ptnodes_worklist.length();
210
211 #ifdef ASSERT
212 if (VerifyConnectionGraph) {
213 // Verify that no new simple edges could be created and all
214 // local vars has edges.
215 _verify = true;
216 for (int next = 0; next < ptnodes_length; ++next) {
217 PointsToNode* ptn = ptnodes_worklist.at(next);
218 add_final_edges(ptn->ideal_node());
219 if (ptn->is_LocalVar() && ptn->edge_count() == 0) {
220 ptn->dump();
221 assert(ptn->as_LocalVar()->edge_count() > 0, "sanity");
222 }
223 }
224 _verify = false;
225 }
226 #endif
227 // Bytecode analyzer BCEscapeAnalyzer, used for Call nodes
228 // processing, calls to CI to resolve symbols (types, fields, methods)
229 // referenced in bytecode. During symbol resolution VM may throw
230 // an exception which CI cleans and converts to compilation failure.
231 if (C->failing()) return false;
232
233 // 2. Finish Graph construction by propagating references to all
234 // java objects through graph.
235 if (!complete_connection_graph(ptnodes_worklist, non_escaped_worklist,
236 java_objects_worklist, oop_fields_worklist)) {
237 // All objects escaped or hit time or iterations limits.
238 _collecting = false;
239 return false;
240 }
241
242 // 3. Adjust scalar_replaceable state of nonescaping objects and push
243 // scalar replaceable allocations on alloc_worklist for processing
244 // in split_unique_types().
245 int non_escaped_length = non_escaped_worklist.length();
246 for (int next = 0; next < non_escaped_length; next++) {
247 JavaObjectNode* ptn = non_escaped_worklist.at(next);
248 bool noescape = (ptn->escape_state() == PointsToNode::NoEscape);
249 Node* n = ptn->ideal_node();
250 if (n->is_Allocate()) {
251 n->as_Allocate()->_is_non_escaping = noescape;
252 }
253 if (n->is_CallStaticJava()) {
254 n->as_CallStaticJava()->_is_non_escaping = noescape;
255 }
256 if (noescape && ptn->scalar_replaceable()) {
257 adjust_scalar_replaceable_state(ptn);
258 if (ptn->scalar_replaceable()) {
259 alloc_worklist.append(ptn->ideal_node());
260 }
261 }
262 }
263
264 #ifdef ASSERT
265 if (VerifyConnectionGraph) {
266 // Verify that graph is complete - no new edges could be added or needed.
267 verify_connection_graph(ptnodes_worklist, non_escaped_worklist,
268 java_objects_worklist, addp_worklist);
269 }
270 assert(C->unique() == nodes_size(), "no new ideal nodes should be added during ConnectionGraph build");
271 assert(null_obj->escape_state() == PointsToNode::NoEscape &&
272 null_obj->edge_count() == 0 &&
273 !null_obj->arraycopy_src() &&
274 !null_obj->arraycopy_dst(), "sanity");
275 #endif
276
277 _collecting = false;
278
279 } // TracePhase t3("connectionGraph")
280
281 // 4. Optimize ideal graph based on EA information.
282 bool has_non_escaping_obj = (non_escaped_worklist.length() > 0);
283 if (has_non_escaping_obj) {
284 optimize_ideal_graph(ptr_cmp_worklist, storestore_worklist);
285 }
286
287 #ifndef PRODUCT
288 if (PrintEscapeAnalysis) {
289 dump(ptnodes_worklist); // Dump ConnectionGraph
290 }
291 #endif
292
293 bool has_scalar_replaceable_candidates = (alloc_worklist.length() > 0);
294 #ifdef ASSERT
295 if (VerifyConnectionGraph) {
296 int alloc_length = alloc_worklist.length();
297 for (int next = 0; next < alloc_length; ++next) {
298 Node* n = alloc_worklist.at(next);
299 PointsToNode* ptn = ptnode_adr(n->_idx);
300 assert(ptn->escape_state() == PointsToNode::NoEscape && ptn->scalar_replaceable(), "sanity");
301 }
302 }
303 #endif
304
305 // 5. Separate memory graph for scalar replaceable allcations.
306 if (has_scalar_replaceable_candidates &&
307 C->AliasLevel() >= 3 && EliminateAllocations) {
308 // Now use the escape information to create unique types for
309 // scalar replaceable objects.
310 split_unique_types(alloc_worklist, arraycopy_worklist);
311 if (C->failing()) return false;
312 C->print_method(PHASE_AFTER_EA, 2);
313
314 #ifdef ASSERT
315 } else if (Verbose && (PrintEscapeAnalysis || PrintEliminateAllocations)) {
316 tty->print("=== No allocations eliminated for ");
317 C->method()->print_short_name();
318 if(!EliminateAllocations) {
319 tty->print(" since EliminateAllocations is off ===");
320 } else if(!has_scalar_replaceable_candidates) {
321 tty->print(" since there are no scalar replaceable candidates ===");
322 } else if(C->AliasLevel() < 3) {
323 tty->print(" since AliasLevel < 3 ===");
324 }
325 tty->cr();
326 #endif
327 }
328 return has_non_escaping_obj;
329 }
330
331 // Utility function for nodes that load an object
332 void ConnectionGraph::add_objload_to_connection_graph(Node *n, Unique_Node_List *delayed_worklist) {
333 // Using isa_ptr() instead of isa_oopptr() for LoadP and Phi because
334 // ThreadLocal has RawPtr type.
335 const Type* t = _igvn->type(n);
336 if (t->make_ptr() != NULL) {
337 Node* adr = n->in(MemNode::Address);
338 #ifdef ASSERT
339 if (!adr->is_AddP()) {
340 assert(_igvn->type(adr)->isa_rawptr(), "sanity");
341 } else {
342 assert((ptnode_adr(adr->_idx) == NULL ||
343 ptnode_adr(adr->_idx)->as_Field()->is_oop()), "sanity");
344 }
345 #endif
346 add_local_var_and_edge(n, PointsToNode::NoEscape,
347 adr, delayed_worklist);
348 }
349 }
350
351 // Populate Connection Graph with PointsTo nodes and create simple
352 // connection graph edges.
353 void ConnectionGraph::add_node_to_connection_graph(Node *n, Unique_Node_List *delayed_worklist) {
354 assert(!_verify, "this method should not be called for verification");
355 PhaseGVN* igvn = _igvn;
356 uint n_idx = n->_idx;
357 PointsToNode* n_ptn = ptnode_adr(n_idx);
358 if (n_ptn != NULL)
359 return; // No need to redefine PointsTo node during first iteration.
360
361 if (n->is_Call()) {
362 // Arguments to allocation and locking don't escape.
363 if (n->is_AbstractLock()) {
364 // Put Lock and Unlock nodes on IGVN worklist to process them during
365 // first IGVN optimization when escape information is still available.
366 record_for_optimizer(n);
367 } else if (n->is_Allocate()) {
368 add_call_node(n->as_Call());
369 record_for_optimizer(n);
370 } else {
371 if (n->is_CallStaticJava()) {
372 const char* name = n->as_CallStaticJava()->_name;
373 if (name != NULL && strcmp(name, "uncommon_trap") == 0)
374 return; // Skip uncommon traps
375 }
376 // Don't mark as processed since call's arguments have to be processed.
377 delayed_worklist->push(n);
378 // Check if a call returns an object.
379 if ((n->as_Call()->returns_pointer() &&
380 n->as_Call()->proj_out_or_null(TypeFunc::Parms) != NULL) ||
381 (n->is_CallStaticJava() &&
382 n->as_CallStaticJava()->is_boxing_method())) {
383 add_call_node(n->as_Call());
384 }
385 }
386 return;
387 }
388 // Put this check here to process call arguments since some call nodes
389 // point to phantom_obj.
390 if (n_ptn == phantom_obj || n_ptn == null_obj)
391 return; // Skip predefined nodes.
392
393 int opcode = n->Opcode();
394 switch (opcode) {
395 case Op_AddP: {
396 Node* base = get_addp_base(n);
397 PointsToNode* ptn_base = ptnode_adr(base->_idx);
398 // Field nodes are created for all field types. They are used in
399 // adjust_scalar_replaceable_state() and split_unique_types().
400 // Note, non-oop fields will have only base edges in Connection
401 // Graph because such fields are not used for oop loads and stores.
402 int offset = address_offset(n, igvn);
403 add_field(n, PointsToNode::NoEscape, offset);
404 if (ptn_base == NULL) {
405 delayed_worklist->push(n); // Process it later.
406 } else {
407 n_ptn = ptnode_adr(n_idx);
408 add_base(n_ptn->as_Field(), ptn_base);
409 }
410 break;
411 }
412 case Op_CastX2P: {
413 map_ideal_node(n, phantom_obj);
414 break;
415 }
416 case Op_CastPP:
417 case Op_CheckCastPP:
418 case Op_EncodeP:
419 case Op_DecodeN:
420 case Op_EncodePKlass:
421 case Op_DecodeNKlass: {
422 add_local_var_and_edge(n, PointsToNode::NoEscape,
423 n->in(1), delayed_worklist);
424 break;
425 }
426 case Op_CMoveP: {
427 add_local_var(n, PointsToNode::NoEscape);
428 // Do not add edges during first iteration because some could be
429 // not defined yet.
430 delayed_worklist->push(n);
431 break;
432 }
433 case Op_ConP:
434 case Op_ConN:
435 case Op_ConNKlass: {
436 // assume all oop constants globally escape except for null
437 PointsToNode::EscapeState es;
438 const Type* t = igvn->type(n);
439 if (t == TypePtr::NULL_PTR || t == TypeNarrowOop::NULL_PTR) {
440 es = PointsToNode::NoEscape;
441 } else {
442 es = PointsToNode::GlobalEscape;
443 }
444 add_java_object(n, es);
445 break;
446 }
447 case Op_CreateEx: {
448 // assume that all exception objects globally escape
449 map_ideal_node(n, phantom_obj);
450 break;
451 }
452 case Op_LoadKlass:
453 case Op_LoadNKlass: {
454 // Unknown class is loaded
455 map_ideal_node(n, phantom_obj);
456 break;
457 }
458 case Op_LoadP:
459 #if INCLUDE_ZGC
460 case Op_LoadBarrierSlowReg:
461 case Op_LoadBarrierWeakSlowReg:
462 #endif
463 case Op_LoadN:
464 case Op_LoadPLocked: {
465 add_objload_to_connection_graph(n, delayed_worklist);
466 break;
467 }
468 case Op_Parm: {
469 map_ideal_node(n, phantom_obj);
470 break;
471 }
472 case Op_PartialSubtypeCheck: {
473 // Produces Null or notNull and is used in only in CmpP so
474 // phantom_obj could be used.
475 map_ideal_node(n, phantom_obj); // Result is unknown
476 break;
477 }
478 case Op_Phi: {
479 // Using isa_ptr() instead of isa_oopptr() for LoadP and Phi because
480 // ThreadLocal has RawPtr type.
481 const Type* t = n->as_Phi()->type();
482 if (t->make_ptr() != NULL) {
483 add_local_var(n, PointsToNode::NoEscape);
484 // Do not add edges during first iteration because some could be
485 // not defined yet.
486 delayed_worklist->push(n);
487 }
488 break;
489 }
490 case Op_Proj: {
491 // we are only interested in the oop result projection from a call
492 if (n->as_Proj()->_con == TypeFunc::Parms && n->in(0)->is_Call() &&
493 n->in(0)->as_Call()->returns_pointer()) {
494 add_local_var_and_edge(n, PointsToNode::NoEscape,
495 n->in(0), delayed_worklist);
496 }
497 #if INCLUDE_ZGC
498 else if (UseZGC) {
499 if (n->as_Proj()->_con == LoadBarrierNode::Oop && n->in(0)->is_LoadBarrier()) {
500 add_local_var_and_edge(n, PointsToNode::NoEscape, n->in(0)->in(LoadBarrierNode::Oop), delayed_worklist);
501 }
502 }
503 #endif
504 break;
505 }
506 case Op_Rethrow: // Exception object escapes
507 case Op_Return: {
508 if (n->req() > TypeFunc::Parms &&
509 igvn->type(n->in(TypeFunc::Parms))->isa_oopptr()) {
510 // Treat Return value as LocalVar with GlobalEscape escape state.
511 add_local_var_and_edge(n, PointsToNode::GlobalEscape,
512 n->in(TypeFunc::Parms), delayed_worklist);
513 }
514 break;
515 }
516 case Op_CompareAndExchangeP:
517 case Op_CompareAndExchangeN:
518 #if INCLUDE_SHENANDOAHGC
519 case Op_ShenandoahCompareAndExchangeP:
520 case Op_ShenandoahCompareAndExchangeN:
521 #endif
522 case Op_GetAndSetP:
523 case Op_GetAndSetN: {
524 add_objload_to_connection_graph(n, delayed_worklist);
525 // fallthrough
526 }
527 case Op_StoreP:
528 case Op_StoreN:
529 case Op_StoreNKlass:
530 case Op_StorePConditional:
531 #if INCLUDE_SHENANDOAHGC
532 case Op_ShenandoahWeakCompareAndSwapP:
533 case Op_ShenandoahWeakCompareAndSwapN:
534 case Op_ShenandoahCompareAndSwapP:
535 case Op_ShenandoahCompareAndSwapN:
536 #endif
537 case Op_WeakCompareAndSwapP:
538 case Op_WeakCompareAndSwapN:
539 case Op_CompareAndSwapP:
540 case Op_CompareAndSwapN: {
541 Node* adr = n->in(MemNode::Address);
542 const Type *adr_type = igvn->type(adr);
543 adr_type = adr_type->make_ptr();
544 if (adr_type == NULL) {
545 break; // skip dead nodes
546 }
547 if ( adr_type->isa_oopptr()
548 || ( (opcode == Op_StoreP || opcode == Op_StoreN || opcode == Op_StoreNKlass)
549 && adr_type == TypeRawPtr::NOTNULL
550 && adr->in(AddPNode::Address)->is_Proj()
551 && adr->in(AddPNode::Address)->in(0)->is_Allocate())) {
552 delayed_worklist->push(n); // Process it later.
553 #ifdef ASSERT
554 assert(adr->is_AddP(), "expecting an AddP");
555 if (adr_type == TypeRawPtr::NOTNULL) {
556 // Verify a raw address for a store captured by Initialize node.
557 int offs = (int)igvn->find_intptr_t_con(adr->in(AddPNode::Offset), Type::OffsetBot);
558 assert(offs != Type::OffsetBot, "offset must be a constant");
559 }
560 #endif
561 } else {
562 // Ignore copy the displaced header to the BoxNode (OSR compilation).
563 if (adr->is_BoxLock())
564 break;
565 // Stored value escapes in unsafe access.
566 if ((opcode == Op_StoreP) && adr_type->isa_rawptr()) {
567 // Pointer stores in G1 barriers looks like unsafe access.
568 // Ignore such stores to be able scalar replace non-escaping
569 // allocations.
570 #if INCLUDE_G1GC || INCLUDE_SHENANDOAHGC
571 if ((UseG1GC || UseShenandoahGC) && adr->is_AddP()) {
572 Node* base = get_addp_base(adr);
573 if (base->Opcode() == Op_LoadP &&
574 base->in(MemNode::Address)->is_AddP()) {
575 adr = base->in(MemNode::Address);
576 Node* tls = get_addp_base(adr);
577 if (tls->Opcode() == Op_ThreadLocal) {
578 int offs = (int)igvn->find_intptr_t_con(adr->in(AddPNode::Offset), Type::OffsetBot);
579 #if INCLUDE_G1GC && INCLUDE_SHENANDOAHGC
580 const int buf_offset = in_bytes(UseG1GC ? G1ThreadLocalData::satb_mark_queue_buffer_offset()
581 : ShenandoahThreadLocalData::satb_mark_queue_buffer_offset());
582 #elif INCLUDE_G1GC
583 const int buf_offset = in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset());
584 #else
585 const int buf_offset = in_bytes(ShenandoahThreadLocalData::satb_mark_queue_buffer_offset());
586 #endif
587 if (offs == buf_offset) {
588 break; // G1 pre barrier previous oop value store.
589 }
590 if (offs == in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset())) {
591 break; // G1 post barrier card address store.
592 }
593 }
594 }
595 }
596 #endif
597 delayed_worklist->push(n); // Process unsafe access later.
598 break;
599 }
600 #ifdef ASSERT
601 n->dump(1);
602 assert(false, "not unsafe or G1 barrier raw StoreP");
603 #endif
604 }
605 break;
606 }
607 case Op_AryEq:
608 case Op_HasNegatives:
609 case Op_StrComp:
610 case Op_StrEquals:
611 case Op_StrIndexOf:
612 case Op_StrIndexOfChar:
613 case Op_StrInflatedCopy:
614 case Op_StrCompressedCopy:
615 case Op_EncodeISOArray: {
616 add_local_var(n, PointsToNode::ArgEscape);
617 delayed_worklist->push(n); // Process it later.
618 break;
619 }
620 case Op_ThreadLocal: {
621 add_java_object(n, PointsToNode::ArgEscape);
622 break;
623 }
624 #if INCLUDE_SHENANDOAHGC
625 case Op_ShenandoahEnqueueBarrier:
626 add_local_var_and_edge(n, PointsToNode::NoEscape, n->in(1), delayed_worklist);
627 break;
628 case Op_ShenandoahLoadReferenceBarrier:
629 add_local_var_and_edge(n, PointsToNode::NoEscape, n->in(ShenandoahLoadReferenceBarrierNode::ValueIn), delayed_worklist);
630 #endif
631 default:
632 ; // Do nothing for nodes not related to EA.
633 }
634 return;
635 }
636
637 #ifdef ASSERT
638 #define ELSE_FAIL(name) \
639 /* Should not be called for not pointer type. */ \
640 n->dump(1); \
641 assert(false, name); \
642 break;
643 #else
644 #define ELSE_FAIL(name) \
645 break;
646 #endif
647
648 // Add final simple edges to graph.
649 void ConnectionGraph::add_final_edges(Node *n) {
650 PointsToNode* n_ptn = ptnode_adr(n->_idx);
651 #ifdef ASSERT
652 if (_verify && n_ptn->is_JavaObject())
653 return; // This method does not change graph for JavaObject.
654 #endif
655
656 if (n->is_Call()) {
657 process_call_arguments(n->as_Call());
658 return;
659 }
660 assert(n->is_Store() || n->is_LoadStore() ||
661 (n_ptn != NULL) && (n_ptn->ideal_node() != NULL),
662 "node should be registered already");
663 int opcode = n->Opcode();
664 switch (opcode) {
665 case Op_AddP: {
666 Node* base = get_addp_base(n);
667 PointsToNode* ptn_base = ptnode_adr(base->_idx);
668 assert(ptn_base != NULL, "field's base should be registered");
669 add_base(n_ptn->as_Field(), ptn_base);
670 break;
671 }
672 case Op_CastPP:
673 case Op_CheckCastPP:
674 case Op_EncodeP:
675 case Op_DecodeN:
676 case Op_EncodePKlass:
677 case Op_DecodeNKlass: {
678 add_local_var_and_edge(n, PointsToNode::NoEscape,
679 n->in(1), NULL);
680 break;
681 }
682 case Op_CMoveP: {
683 for (uint i = CMoveNode::IfFalse; i < n->req(); i++) {
684 Node* in = n->in(i);
685 if (in == NULL)
686 continue; // ignore NULL
687 Node* uncast_in = in->uncast();
688 if (uncast_in->is_top() || uncast_in == n)
689 continue; // ignore top or inputs which go back this node
690 PointsToNode* ptn = ptnode_adr(in->_idx);
691 assert(ptn != NULL, "node should be registered");
692 add_edge(n_ptn, ptn);
693 }
694 break;
695 }
696 case Op_LoadP:
697 #if INCLUDE_ZGC
698 case Op_LoadBarrierSlowReg:
699 case Op_LoadBarrierWeakSlowReg:
700 #endif
701 case Op_LoadN:
702 case Op_LoadPLocked: {
703 // Using isa_ptr() instead of isa_oopptr() for LoadP and Phi because
704 // ThreadLocal has RawPtr type.
705 const Type* t = _igvn->type(n);
706 if (t->make_ptr() != NULL) {
707 Node* adr = n->in(MemNode::Address);
708 add_local_var_and_edge(n, PointsToNode::NoEscape, adr, NULL);
709 break;
710 }
711 ELSE_FAIL("Op_LoadP");
712 }
713 case Op_Phi: {
714 // Using isa_ptr() instead of isa_oopptr() for LoadP and Phi because
715 // ThreadLocal has RawPtr type.
716 const Type* t = n->as_Phi()->type();
717 if (t->make_ptr() != NULL) {
718 for (uint i = 1; i < n->req(); i++) {
719 Node* in = n->in(i);
720 if (in == NULL)
721 continue; // ignore NULL
722 Node* uncast_in = in->uncast();
723 if (uncast_in->is_top() || uncast_in == n)
724 continue; // ignore top or inputs which go back this node
725 PointsToNode* ptn = ptnode_adr(in->_idx);
726 assert(ptn != NULL, "node should be registered");
727 add_edge(n_ptn, ptn);
728 }
729 break;
730 }
731 ELSE_FAIL("Op_Phi");
732 }
733 case Op_Proj: {
734 // we are only interested in the oop result projection from a call
735 if (n->as_Proj()->_con == TypeFunc::Parms && n->in(0)->is_Call() &&
736 n->in(0)->as_Call()->returns_pointer()) {
737 add_local_var_and_edge(n, PointsToNode::NoEscape, n->in(0), NULL);
738 break;
739 }
740 #if INCLUDE_ZGC
741 else if (UseZGC) {
742 if (n->as_Proj()->_con == LoadBarrierNode::Oop && n->in(0)->is_LoadBarrier()) {
743 add_local_var_and_edge(n, PointsToNode::NoEscape, n->in(0)->in(LoadBarrierNode::Oop), NULL);
744 break;
745 }
746 }
747 #endif
748 ELSE_FAIL("Op_Proj");
749 }
750 case Op_Rethrow: // Exception object escapes
751 case Op_Return: {
752 if (n->req() > TypeFunc::Parms &&
753 _igvn->type(n->in(TypeFunc::Parms))->isa_oopptr()) {
754 // Treat Return value as LocalVar with GlobalEscape escape state.
755 add_local_var_and_edge(n, PointsToNode::GlobalEscape,
756 n->in(TypeFunc::Parms), NULL);
757 break;
758 }
759 ELSE_FAIL("Op_Return");
760 }
761 case Op_StoreP:
762 case Op_StoreN:
763 case Op_StoreNKlass:
764 case Op_StorePConditional:
765 case Op_CompareAndExchangeP:
766 case Op_CompareAndExchangeN:
767 case Op_CompareAndSwapP:
768 case Op_CompareAndSwapN:
769 case Op_WeakCompareAndSwapP:
770 case Op_WeakCompareAndSwapN:
771 #if INCLUDE_SHENANDOAHGC
772 case Op_ShenandoahCompareAndExchangeP:
773 case Op_ShenandoahCompareAndExchangeN:
774 case Op_ShenandoahCompareAndSwapP:
775 case Op_ShenandoahCompareAndSwapN:
776 case Op_ShenandoahWeakCompareAndSwapP:
777 case Op_ShenandoahWeakCompareAndSwapN:
778 #endif
779 case Op_GetAndSetP:
780 case Op_GetAndSetN: {
781 Node* adr = n->in(MemNode::Address);
782 const Type *adr_type = _igvn->type(adr);
783 adr_type = adr_type->make_ptr();
784 #ifdef ASSERT
785 if (adr_type == NULL) {
786 n->dump(1);
787 assert(adr_type != NULL, "dead node should not be on list");
788 break;
789 }
790 #endif
791 if (opcode == Op_GetAndSetP || opcode == Op_GetAndSetN ||
792 #if INCLUDE_SHENANDOAHGC
793 opcode == Op_ShenandoahCompareAndExchangeN || opcode == Op_ShenandoahCompareAndExchangeP ||
794 #endif
795 opcode == Op_CompareAndExchangeN || opcode == Op_CompareAndExchangeP) {
796 add_local_var_and_edge(n, PointsToNode::NoEscape, adr, NULL);
797 }
798 if ( adr_type->isa_oopptr()
799 || ( (opcode == Op_StoreP || opcode == Op_StoreN || opcode == Op_StoreNKlass)
800 && adr_type == TypeRawPtr::NOTNULL
801 && adr->in(AddPNode::Address)->is_Proj()
802 && adr->in(AddPNode::Address)->in(0)->is_Allocate())) {
803 // Point Address to Value
804 PointsToNode* adr_ptn = ptnode_adr(adr->_idx);
805 assert(adr_ptn != NULL &&
806 adr_ptn->as_Field()->is_oop(), "node should be registered");
807 Node *val = n->in(MemNode::ValueIn);
808 PointsToNode* ptn = ptnode_adr(val->_idx);
809 assert(ptn != NULL, "node should be registered");
810 add_edge(adr_ptn, ptn);
811 break;
812 } else if ((opcode == Op_StoreP) && adr_type->isa_rawptr()) {
813 // Stored value escapes in unsafe access.
814 Node *val = n->in(MemNode::ValueIn);
815 PointsToNode* ptn = ptnode_adr(val->_idx);
816 assert(ptn != NULL, "node should be registered");
817 set_escape_state(ptn, PointsToNode::GlobalEscape);
818 // Add edge to object for unsafe access with offset.
819 PointsToNode* adr_ptn = ptnode_adr(adr->_idx);
820 assert(adr_ptn != NULL, "node should be registered");
821 if (adr_ptn->is_Field()) {
822 assert(adr_ptn->as_Field()->is_oop(), "should be oop field");
823 add_edge(adr_ptn, ptn);
824 }
825 break;
826 }
827 ELSE_FAIL("Op_StoreP");
828 }
829 case Op_AryEq:
830 case Op_HasNegatives:
831 case Op_StrComp:
832 case Op_StrEquals:
833 case Op_StrIndexOf:
834 case Op_StrIndexOfChar:
835 case Op_StrInflatedCopy:
836 case Op_StrCompressedCopy:
837 case Op_EncodeISOArray: {
838 // char[]/byte[] arrays passed to string intrinsic do not escape but
839 // they are not scalar replaceable. Adjust escape state for them.
840 // Start from in(2) edge since in(1) is memory edge.
841 for (uint i = 2; i < n->req(); i++) {
842 Node* adr = n->in(i);
843 const Type* at = _igvn->type(adr);
844 if (!adr->is_top() && at->isa_ptr()) {
845 assert(at == Type::TOP || at == TypePtr::NULL_PTR ||
846 at->isa_ptr() != NULL, "expecting a pointer");
847 if (adr->is_AddP()) {
848 adr = get_addp_base(adr);
849 }
850 PointsToNode* ptn = ptnode_adr(adr->_idx);
851 assert(ptn != NULL, "node should be registered");
852 add_edge(n_ptn, ptn);
853 }
854 }
855 break;
856 }
857 #if INCLUDE_SHENANDOAHGC
858 case Op_ShenandoahEnqueueBarrier:
859 add_local_var_and_edge(n, PointsToNode::NoEscape, n->in(1), NULL);
860 break;
861 case Op_ShenandoahLoadReferenceBarrier:
862 add_local_var_and_edge(n, PointsToNode::NoEscape, n->in(ShenandoahLoadReferenceBarrierNode::ValueIn), NULL);
863 break;
864 #endif
865 default: {
866 // This method should be called only for EA specific nodes which may
867 // miss some edges when they were created.
868 #ifdef ASSERT
869 n->dump(1);
870 #endif
871 guarantee(false, "unknown node");
872 }
873 }
874 return;
875 }
876
877 void ConnectionGraph::add_call_node(CallNode* call) {
878 assert(call->returns_pointer(), "only for call which returns pointer");
879 uint call_idx = call->_idx;
880 if (call->is_Allocate()) {
881 Node* k = call->in(AllocateNode::KlassNode);
882 const TypeKlassPtr* kt = k->bottom_type()->isa_klassptr();
883 assert(kt != NULL, "TypeKlassPtr required.");
884 ciKlass* cik = kt->klass();
885 PointsToNode::EscapeState es = PointsToNode::NoEscape;
886 bool scalar_replaceable = true;
887 if (call->is_AllocateArray()) {
888 if (!cik->is_array_klass()) { // StressReflectiveCode
889 es = PointsToNode::GlobalEscape;
890 } else {
891 int length = call->in(AllocateNode::ALength)->find_int_con(-1);
892 if (length < 0 || length > EliminateAllocationArraySizeLimit) {
893 // Not scalar replaceable if the length is not constant or too big.
894 scalar_replaceable = false;
895 }
896 }
897 } else { // Allocate instance
898 if (cik->is_subclass_of(_compile->env()->Thread_klass()) ||
899 cik->is_subclass_of(_compile->env()->Reference_klass()) ||
900 !cik->is_instance_klass() || // StressReflectiveCode
901 !cik->as_instance_klass()->can_be_instantiated() ||
902 cik->as_instance_klass()->has_finalizer()) {
903 es = PointsToNode::GlobalEscape;
904 }
905 }
906 add_java_object(call, es);
907 PointsToNode* ptn = ptnode_adr(call_idx);
908 if (!scalar_replaceable && ptn->scalar_replaceable()) {
909 ptn->set_scalar_replaceable(false);
910 }
911 } else if (call->is_CallStaticJava()) {
912 // Call nodes could be different types:
913 //
914 // 1. CallDynamicJavaNode (what happened during call is unknown):
915 //
916 // - mapped to GlobalEscape JavaObject node if oop is returned;
917 //
918 // - all oop arguments are escaping globally;
919 //
920 // 2. CallStaticJavaNode (execute bytecode analysis if possible):
921 //
922 // - the same as CallDynamicJavaNode if can't do bytecode analysis;
923 //
924 // - mapped to GlobalEscape JavaObject node if unknown oop is returned;
925 // - mapped to NoEscape JavaObject node if non-escaping object allocated
926 // during call is returned;
927 // - mapped to ArgEscape LocalVar node pointed to object arguments
928 // which are returned and does not escape during call;
929 //
930 // - oop arguments escaping status is defined by bytecode analysis;
931 //
932 // For a static call, we know exactly what method is being called.
933 // Use bytecode estimator to record whether the call's return value escapes.
934 ciMethod* meth = call->as_CallJava()->method();
935 if (meth == NULL) {
936 const char* name = call->as_CallStaticJava()->_name;
937 assert(strncmp(name, "_multianewarray", 15) == 0, "TODO: add failed case check");
938 // Returns a newly allocated unescaped object.
939 add_java_object(call, PointsToNode::NoEscape);
940 ptnode_adr(call_idx)->set_scalar_replaceable(false);
941 } else if (meth->is_boxing_method()) {
942 // Returns boxing object
943 PointsToNode::EscapeState es;
944 vmIntrinsics::ID intr = meth->intrinsic_id();
945 if (intr == vmIntrinsics::_floatValue || intr == vmIntrinsics::_doubleValue) {
946 // It does not escape if object is always allocated.
947 es = PointsToNode::NoEscape;
948 } else {
949 // It escapes globally if object could be loaded from cache.
950 es = PointsToNode::GlobalEscape;
951 }
952 add_java_object(call, es);
953 } else {
954 BCEscapeAnalyzer* call_analyzer = meth->get_bcea();
955 call_analyzer->copy_dependencies(_compile->dependencies());
956 if (call_analyzer->is_return_allocated()) {
957 // Returns a newly allocated unescaped object, simply
958 // update dependency information.
959 // Mark it as NoEscape so that objects referenced by
960 // it's fields will be marked as NoEscape at least.
961 add_java_object(call, PointsToNode::NoEscape);
962 ptnode_adr(call_idx)->set_scalar_replaceable(false);
963 } else {
964 // Determine whether any arguments are returned.
965 const TypeTuple* d = call->tf()->domain();
966 bool ret_arg = false;
967 for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
968 if (d->field_at(i)->isa_ptr() != NULL &&
969 call_analyzer->is_arg_returned(i - TypeFunc::Parms)) {
970 ret_arg = true;
971 break;
972 }
973 }
974 if (ret_arg) {
975 add_local_var(call, PointsToNode::ArgEscape);
976 } else {
977 // Returns unknown object.
978 map_ideal_node(call, phantom_obj);
979 }
980 }
981 }
982 } else {
983 // An other type of call, assume the worst case:
984 // returned value is unknown and globally escapes.
985 assert(call->Opcode() == Op_CallDynamicJava, "add failed case check");
986 map_ideal_node(call, phantom_obj);
987 }
988 }
989
990 void ConnectionGraph::process_call_arguments(CallNode *call) {
991 bool is_arraycopy = false;
992 switch (call->Opcode()) {
993 #ifdef ASSERT
994 case Op_Allocate:
995 case Op_AllocateArray:
996 case Op_Lock:
997 case Op_Unlock:
998 assert(false, "should be done already");
999 break;
1000 #endif
1001 case Op_ArrayCopy:
1002 case Op_CallLeafNoFP:
1003 // Most array copies are ArrayCopy nodes at this point but there
1004 // are still a few direct calls to the copy subroutines (See
1005 // PhaseStringOpts::copy_string())
1006 is_arraycopy = (call->Opcode() == Op_ArrayCopy) ||
1007 call->as_CallLeaf()->is_call_to_arraycopystub();
1008 // fall through
1009 case Op_CallLeaf: {
1010 // Stub calls, objects do not escape but they are not scale replaceable.
1011 // Adjust escape state for outgoing arguments.
1012 const TypeTuple * d = call->tf()->domain();
1013 bool src_has_oops = false;
1014 for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
1015 const Type* at = d->field_at(i);
1016 Node *arg = call->in(i);
1017 if (arg == NULL) {
1018 continue;
1019 }
1020 const Type *aat = _igvn->type(arg);
1021 if (arg->is_top() || !at->isa_ptr() || !aat->isa_ptr())
1022 continue;
1023 if (arg->is_AddP()) {
1024 //
1025 // The inline_native_clone() case when the arraycopy stub is called
1026 // after the allocation before Initialize and CheckCastPP nodes.
1027 // Or normal arraycopy for object arrays case.
1028 //
1029 // Set AddP's base (Allocate) as not scalar replaceable since
1030 // pointer to the base (with offset) is passed as argument.
1031 //
1032 arg = get_addp_base(arg);
1033 }
1034 PointsToNode* arg_ptn = ptnode_adr(arg->_idx);
1035 assert(arg_ptn != NULL, "should be registered");
1036 PointsToNode::EscapeState arg_esc = arg_ptn->escape_state();
1037 if (is_arraycopy || arg_esc < PointsToNode::ArgEscape) {
1038 assert(aat == Type::TOP || aat == TypePtr::NULL_PTR ||
1039 aat->isa_ptr() != NULL, "expecting an Ptr");
1040 bool arg_has_oops = aat->isa_oopptr() &&
1041 (aat->isa_oopptr()->klass() == NULL || aat->isa_instptr() ||
1042 (aat->isa_aryptr() && aat->isa_aryptr()->klass()->is_obj_array_klass()));
1043 if (i == TypeFunc::Parms) {
1044 src_has_oops = arg_has_oops;
1045 }
1046 //
1047 // src or dst could be j.l.Object when other is basic type array:
1048 //
1049 // arraycopy(char[],0,Object*,0,size);
1050 // arraycopy(Object*,0,char[],0,size);
1051 //
1052 // Don't add edges in such cases.
1053 //
1054 bool arg_is_arraycopy_dest = src_has_oops && is_arraycopy &&
1055 arg_has_oops && (i > TypeFunc::Parms);
1056 #ifdef ASSERT
1057 if (!(is_arraycopy ||
1058 BarrierSet::barrier_set()->barrier_set_c2()->is_gc_barrier_node(call) ||
1059 (call->as_CallLeaf()->_name != NULL &&
1060 (strcmp(call->as_CallLeaf()->_name, "updateBytesCRC32") == 0 ||
1061 strcmp(call->as_CallLeaf()->_name, "updateBytesCRC32C") == 0 ||
1062 strcmp(call->as_CallLeaf()->_name, "updateBytesAdler32") == 0 ||
1063 strcmp(call->as_CallLeaf()->_name, "aescrypt_encryptBlock") == 0 ||
1064 strcmp(call->as_CallLeaf()->_name, "aescrypt_decryptBlock") == 0 ||
1065 strcmp(call->as_CallLeaf()->_name, "cipherBlockChaining_encryptAESCrypt") == 0 ||
1066 strcmp(call->as_CallLeaf()->_name, "cipherBlockChaining_decryptAESCrypt") == 0 ||
1067 strcmp(call->as_CallLeaf()->_name, "electronicCodeBook_encryptAESCrypt") == 0 ||
1068 strcmp(call->as_CallLeaf()->_name, "electronicCodeBook_decryptAESCrypt") == 0 ||
1069 strcmp(call->as_CallLeaf()->_name, "counterMode_AESCrypt") == 0 ||
1070 strcmp(call->as_CallLeaf()->_name, "ghash_processBlocks") == 0 ||
1071 strcmp(call->as_CallLeaf()->_name, "encodeBlock") == 0 ||
1072 strcmp(call->as_CallLeaf()->_name, "sha1_implCompress") == 0 ||
1073 strcmp(call->as_CallLeaf()->_name, "sha1_implCompressMB") == 0 ||
1074 strcmp(call->as_CallLeaf()->_name, "sha256_implCompress") == 0 ||
1075 strcmp(call->as_CallLeaf()->_name, "sha256_implCompressMB") == 0 ||
1076 strcmp(call->as_CallLeaf()->_name, "sha512_implCompress") == 0 ||
1077 strcmp(call->as_CallLeaf()->_name, "sha512_implCompressMB") == 0 ||
1078 strcmp(call->as_CallLeaf()->_name, "multiplyToLen") == 0 ||
1079 strcmp(call->as_CallLeaf()->_name, "squareToLen") == 0 ||
1080 strcmp(call->as_CallLeaf()->_name, "mulAdd") == 0 ||
1081 strcmp(call->as_CallLeaf()->_name, "montgomery_multiply") == 0 ||
1082 strcmp(call->as_CallLeaf()->_name, "montgomery_square") == 0 ||
1083 strcmp(call->as_CallLeaf()->_name, "vectorizedMismatch") == 0)
1084 ))) {
1085 call->dump();
1086 fatal("EA unexpected CallLeaf %s", call->as_CallLeaf()->_name);
1087 }
1088 #endif
1089 // Always process arraycopy's destination object since
1090 // we need to add all possible edges to references in
1091 // source object.
1092 if (arg_esc >= PointsToNode::ArgEscape &&
1093 !arg_is_arraycopy_dest) {
1094 continue;
1095 }
1096 PointsToNode::EscapeState es = PointsToNode::ArgEscape;
1097 if (call->is_ArrayCopy()) {
1098 ArrayCopyNode* ac = call->as_ArrayCopy();
1099 if (ac->is_clonebasic() ||
1100 ac->is_arraycopy_validated() ||
1101 ac->is_copyof_validated() ||
1102 ac->is_copyofrange_validated()) {
1103 es = PointsToNode::NoEscape;
1104 }
1105 }
1106 set_escape_state(arg_ptn, es);
1107 if (arg_is_arraycopy_dest) {
1108 Node* src = call->in(TypeFunc::Parms);
1109 if (src->is_AddP()) {
1110 src = get_addp_base(src);
1111 }
1112 PointsToNode* src_ptn = ptnode_adr(src->_idx);
1113 assert(src_ptn != NULL, "should be registered");
1114 if (arg_ptn != src_ptn) {
1115 // Special arraycopy edge:
1116 // A destination object's field can't have the source object
1117 // as base since objects escape states are not related.
1118 // Only escape state of destination object's fields affects
1119 // escape state of fields in source object.
1120 add_arraycopy(call, es, src_ptn, arg_ptn);
1121 }
1122 }
1123 }
1124 }
1125 break;
1126 }
1127 case Op_CallStaticJava: {
1128 // For a static call, we know exactly what method is being called.
1129 // Use bytecode estimator to record the call's escape affects
1130 #ifdef ASSERT
1131 const char* name = call->as_CallStaticJava()->_name;
1132 assert((name == NULL || strcmp(name, "uncommon_trap") != 0), "normal calls only");
1133 #endif
1134 ciMethod* meth = call->as_CallJava()->method();
1135 if ((meth != NULL) && meth->is_boxing_method()) {
1136 break; // Boxing methods do not modify any oops.
1137 }
1138 BCEscapeAnalyzer* call_analyzer = (meth !=NULL) ? meth->get_bcea() : NULL;
1139 // fall-through if not a Java method or no analyzer information
1140 if (call_analyzer != NULL) {
1141 PointsToNode* call_ptn = ptnode_adr(call->_idx);
1142 const TypeTuple* d = call->tf()->domain();
1143 for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
1144 const Type* at = d->field_at(i);
1145 int k = i - TypeFunc::Parms;
1146 Node* arg = call->in(i);
1147 PointsToNode* arg_ptn = ptnode_adr(arg->_idx);
1148 if (at->isa_ptr() != NULL &&
1149 call_analyzer->is_arg_returned(k)) {
1150 // The call returns arguments.
1151 if (call_ptn != NULL) { // Is call's result used?
1152 assert(call_ptn->is_LocalVar(), "node should be registered");
1153 assert(arg_ptn != NULL, "node should be registered");
1154 add_edge(call_ptn, arg_ptn);
1155 }
1156 }
1157 if (at->isa_oopptr() != NULL &&
1158 arg_ptn->escape_state() < PointsToNode::GlobalEscape) {
1159 if (!call_analyzer->is_arg_stack(k)) {
1160 // The argument global escapes
1161 set_escape_state(arg_ptn, PointsToNode::GlobalEscape);
1162 } else {
1163 set_escape_state(arg_ptn, PointsToNode::ArgEscape);
1164 if (!call_analyzer->is_arg_local(k)) {
1165 // The argument itself doesn't escape, but any fields might
1166 set_fields_escape_state(arg_ptn, PointsToNode::GlobalEscape);
1167 }
1168 }
1169 }
1170 }
1171 if (call_ptn != NULL && call_ptn->is_LocalVar()) {
1172 // The call returns arguments.
1173 assert(call_ptn->edge_count() > 0, "sanity");
1174 if (!call_analyzer->is_return_local()) {
1175 // Returns also unknown object.
1176 add_edge(call_ptn, phantom_obj);
1177 }
1178 }
1179 break;
1180 }
1181 }
1182 default: {
1183 // Fall-through here if not a Java method or no analyzer information
1184 // or some other type of call, assume the worst case: all arguments
1185 // globally escape.
1186 const TypeTuple* d = call->tf()->domain();
1187 for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
1188 const Type* at = d->field_at(i);
1189 if (at->isa_oopptr() != NULL) {
1190 Node* arg = call->in(i);
1191 if (arg->is_AddP()) {
1192 arg = get_addp_base(arg);
1193 }
1194 assert(ptnode_adr(arg->_idx) != NULL, "should be defined already");
1195 set_escape_state(ptnode_adr(arg->_idx), PointsToNode::GlobalEscape);
1196 }
1197 }
1198 }
1199 }
1200 }
1201
1202
1203 // Finish Graph construction.
1204 bool ConnectionGraph::complete_connection_graph(
1205 GrowableArray<PointsToNode*>& ptnodes_worklist,
1206 GrowableArray<JavaObjectNode*>& non_escaped_worklist,
1207 GrowableArray<JavaObjectNode*>& java_objects_worklist,
1208 GrowableArray<FieldNode*>& oop_fields_worklist) {
1209 // Normally only 1-3 passes needed to build Connection Graph depending
1210 // on graph complexity. Observed 8 passes in jvm2008 compiler.compiler.
1211 // Set limit to 20 to catch situation when something did go wrong and
1212 // bailout Escape Analysis.
1213 // Also limit build time to 20 sec (60 in debug VM), EscapeAnalysisTimeout flag.
1214 #define CG_BUILD_ITER_LIMIT 20
1215
1216 // Propagate GlobalEscape and ArgEscape escape states and check that
1217 // we still have non-escaping objects. The method pushs on _worklist
1218 // Field nodes which reference phantom_object.
1219 if (!find_non_escaped_objects(ptnodes_worklist, non_escaped_worklist)) {
1220 return false; // Nothing to do.
1221 }
1222 // Now propagate references to all JavaObject nodes.
1223 int java_objects_length = java_objects_worklist.length();
1224 elapsedTimer time;
1225 bool timeout = false;
1226 int new_edges = 1;
1227 int iterations = 0;
1228 do {
1229 while ((new_edges > 0) &&
1230 (iterations++ < CG_BUILD_ITER_LIMIT)) {
1231 double start_time = time.seconds();
1232 time.start();
1233 new_edges = 0;
1234 // Propagate references to phantom_object for nodes pushed on _worklist
1235 // by find_non_escaped_objects() and find_field_value().
1236 new_edges += add_java_object_edges(phantom_obj, false);
1237 for (int next = 0; next < java_objects_length; ++next) {
1238 JavaObjectNode* ptn = java_objects_worklist.at(next);
1239 new_edges += add_java_object_edges(ptn, true);
1240
1241 #define SAMPLE_SIZE 4
1242 if ((next % SAMPLE_SIZE) == 0) {
1243 // Each 4 iterations calculate how much time it will take
1244 // to complete graph construction.
1245 time.stop();
1246 // Poll for requests from shutdown mechanism to quiesce compiler
1247 // because Connection graph construction may take long time.
1248 CompileBroker::maybe_block();
1249 double stop_time = time.seconds();
1250 double time_per_iter = (stop_time - start_time) / (double)SAMPLE_SIZE;
1251 double time_until_end = time_per_iter * (double)(java_objects_length - next);
1252 if ((start_time + time_until_end) >= EscapeAnalysisTimeout) {
1253 timeout = true;
1254 break; // Timeout
1255 }
1256 start_time = stop_time;
1257 time.start();
1258 }
1259 #undef SAMPLE_SIZE
1260
1261 }
1262 if (timeout) break;
1263 if (new_edges > 0) {
1264 // Update escape states on each iteration if graph was updated.
1265 if (!find_non_escaped_objects(ptnodes_worklist, non_escaped_worklist)) {
1266 return false; // Nothing to do.
1267 }
1268 }
1269 time.stop();
1270 if (time.seconds() >= EscapeAnalysisTimeout) {
1271 timeout = true;
1272 break;
1273 }
1274 }
1275 if ((iterations < CG_BUILD_ITER_LIMIT) && !timeout) {
1276 time.start();
1277 // Find fields which have unknown value.
1278 int fields_length = oop_fields_worklist.length();
1279 for (int next = 0; next < fields_length; next++) {
1280 FieldNode* field = oop_fields_worklist.at(next);
1281 if (field->edge_count() == 0) {
1282 new_edges += find_field_value(field);
1283 // This code may added new edges to phantom_object.
1284 // Need an other cycle to propagate references to phantom_object.
1285 }
1286 }
1287 time.stop();
1288 if (time.seconds() >= EscapeAnalysisTimeout) {
1289 timeout = true;
1290 break;
1291 }
1292 } else {
1293 new_edges = 0; // Bailout
1294 }
1295 } while (new_edges > 0);
1296
1297 // Bailout if passed limits.
1298 if ((iterations >= CG_BUILD_ITER_LIMIT) || timeout) {
1299 Compile* C = _compile;
1300 if (C->log() != NULL) {
1301 C->log()->begin_elem("connectionGraph_bailout reason='reached ");
1302 C->log()->text("%s", timeout ? "time" : "iterations");
1303 C->log()->end_elem(" limit'");
1304 }
1305 assert(ExitEscapeAnalysisOnTimeout, "infinite EA connection graph build (%f sec, %d iterations) with %d nodes and worklist size %d",
1306 time.seconds(), iterations, nodes_size(), ptnodes_worklist.length());
1307 // Possible infinite build_connection_graph loop,
1308 // bailout (no changes to ideal graph were made).
1309 return false;
1310 }
1311 #ifdef ASSERT
1312 if (Verbose && PrintEscapeAnalysis) {
1313 tty->print_cr("EA: %d iterations to build connection graph with %d nodes and worklist size %d",
1314 iterations, nodes_size(), ptnodes_worklist.length());
1315 }
1316 #endif
1317
1318 #undef CG_BUILD_ITER_LIMIT
1319
1320 // Find fields initialized by NULL for non-escaping Allocations.
1321 int non_escaped_length = non_escaped_worklist.length();
1322 for (int next = 0; next < non_escaped_length; next++) {
1323 JavaObjectNode* ptn = non_escaped_worklist.at(next);
1324 PointsToNode::EscapeState es = ptn->escape_state();
1325 assert(es <= PointsToNode::ArgEscape, "sanity");
1326 if (es == PointsToNode::NoEscape) {
1327 if (find_init_values(ptn, null_obj, _igvn) > 0) {
1328 // Adding references to NULL object does not change escape states
1329 // since it does not escape. Also no fields are added to NULL object.
1330 add_java_object_edges(null_obj, false);
1331 }
1332 }
1333 Node* n = ptn->ideal_node();
1334 if (n->is_Allocate()) {
1335 // The object allocated by this Allocate node will never be
1336 // seen by an other thread. Mark it so that when it is
1337 // expanded no MemBarStoreStore is added.
1338 InitializeNode* ini = n->as_Allocate()->initialization();
1339 if (ini != NULL)
1340 ini->set_does_not_escape();
1341 }
1342 }
1343 return true; // Finished graph construction.
1344 }
1345
1346 // Propagate GlobalEscape and ArgEscape escape states to all nodes
1347 // and check that we still have non-escaping java objects.
1348 bool ConnectionGraph::find_non_escaped_objects(GrowableArray<PointsToNode*>& ptnodes_worklist,
1349 GrowableArray<JavaObjectNode*>& non_escaped_worklist) {
1350 GrowableArray<PointsToNode*> escape_worklist;
1351 // First, put all nodes with GlobalEscape and ArgEscape states on worklist.
1352 int ptnodes_length = ptnodes_worklist.length();
1353 for (int next = 0; next < ptnodes_length; ++next) {
1354 PointsToNode* ptn = ptnodes_worklist.at(next);
1355 if (ptn->escape_state() >= PointsToNode::ArgEscape ||
1356 ptn->fields_escape_state() >= PointsToNode::ArgEscape) {
1357 escape_worklist.push(ptn);
1358 }
1359 }
1360 // Set escape states to referenced nodes (edges list).
1361 while (escape_worklist.length() > 0) {
1362 PointsToNode* ptn = escape_worklist.pop();
1363 PointsToNode::EscapeState es = ptn->escape_state();
1364 PointsToNode::EscapeState field_es = ptn->fields_escape_state();
1365 if (ptn->is_Field() && ptn->as_Field()->is_oop() &&
1366 es >= PointsToNode::ArgEscape) {
1367 // GlobalEscape or ArgEscape state of field means it has unknown value.
1368 if (add_edge(ptn, phantom_obj)) {
1369 // New edge was added
1370 add_field_uses_to_worklist(ptn->as_Field());
1371 }
1372 }
1373 for (EdgeIterator i(ptn); i.has_next(); i.next()) {
1374 PointsToNode* e = i.get();
1375 if (e->is_Arraycopy()) {
1376 assert(ptn->arraycopy_dst(), "sanity");
1377 // Propagate only fields escape state through arraycopy edge.
1378 if (e->fields_escape_state() < field_es) {
1379 set_fields_escape_state(e, field_es);
1380 escape_worklist.push(e);
1381 }
1382 } else if (es >= field_es) {
1383 // fields_escape_state is also set to 'es' if it is less than 'es'.
1384 if (e->escape_state() < es) {
1385 set_escape_state(e, es);
1386 escape_worklist.push(e);
1387 }
1388 } else {
1389 // Propagate field escape state.
1390 bool es_changed = false;
1391 if (e->fields_escape_state() < field_es) {
1392 set_fields_escape_state(e, field_es);
1393 es_changed = true;
1394 }
1395 if ((e->escape_state() < field_es) &&
1396 e->is_Field() && ptn->is_JavaObject() &&
1397 e->as_Field()->is_oop()) {
1398 // Change escape state of referenced fields.
1399 set_escape_state(e, field_es);
1400 es_changed = true;
1401 } else if (e->escape_state() < es) {
1402 set_escape_state(e, es);
1403 es_changed = true;
1404 }
1405 if (es_changed) {
1406 escape_worklist.push(e);
1407 }
1408 }
1409 }
1410 }
1411 // Remove escaped objects from non_escaped list.
1412 for (int next = non_escaped_worklist.length()-1; next >= 0 ; --next) {
1413 JavaObjectNode* ptn = non_escaped_worklist.at(next);
1414 if (ptn->escape_state() >= PointsToNode::GlobalEscape) {
1415 non_escaped_worklist.delete_at(next);
1416 }
1417 if (ptn->escape_state() == PointsToNode::NoEscape) {
1418 // Find fields in non-escaped allocations which have unknown value.
1419 find_init_values(ptn, phantom_obj, NULL);
1420 }
1421 }
1422 return (non_escaped_worklist.length() > 0);
1423 }
1424
1425 // Add all references to JavaObject node by walking over all uses.
1426 int ConnectionGraph::add_java_object_edges(JavaObjectNode* jobj, bool populate_worklist) {
1427 int new_edges = 0;
1428 if (populate_worklist) {
1429 // Populate _worklist by uses of jobj's uses.
1430 for (UseIterator i(jobj); i.has_next(); i.next()) {
1431 PointsToNode* use = i.get();
1432 if (use->is_Arraycopy())
1433 continue;
1434 add_uses_to_worklist(use);
1435 if (use->is_Field() && use->as_Field()->is_oop()) {
1436 // Put on worklist all field's uses (loads) and
1437 // related field nodes (same base and offset).
1438 add_field_uses_to_worklist(use->as_Field());
1439 }
1440 }
1441 }
1442 for (int l = 0; l < _worklist.length(); l++) {
1443 PointsToNode* use = _worklist.at(l);
1444 if (PointsToNode::is_base_use(use)) {
1445 // Add reference from jobj to field and from field to jobj (field's base).
1446 use = PointsToNode::get_use_node(use)->as_Field();
1447 if (add_base(use->as_Field(), jobj)) {
1448 new_edges++;
1449 }
1450 continue;
1451 }
1452 assert(!use->is_JavaObject(), "sanity");
1453 if (use->is_Arraycopy()) {
1454 if (jobj == null_obj) // NULL object does not have field edges
1455 continue;
1456 // Added edge from Arraycopy node to arraycopy's source java object
1457 if (add_edge(use, jobj)) {
1458 jobj->set_arraycopy_src();
1459 new_edges++;
1460 }
1461 // and stop here.
1462 continue;
1463 }
1464 if (!add_edge(use, jobj))
1465 continue; // No new edge added, there was such edge already.
1466 new_edges++;
1467 if (use->is_LocalVar()) {
1468 add_uses_to_worklist(use);
1469 if (use->arraycopy_dst()) {
1470 for (EdgeIterator i(use); i.has_next(); i.next()) {
1471 PointsToNode* e = i.get();
1472 if (e->is_Arraycopy()) {
1473 if (jobj == null_obj) // NULL object does not have field edges
1474 continue;
1475 // Add edge from arraycopy's destination java object to Arraycopy node.
1476 if (add_edge(jobj, e)) {
1477 new_edges++;
1478 jobj->set_arraycopy_dst();
1479 }
1480 }
1481 }
1482 }
1483 } else {
1484 // Added new edge to stored in field values.
1485 // Put on worklist all field's uses (loads) and
1486 // related field nodes (same base and offset).
1487 add_field_uses_to_worklist(use->as_Field());
1488 }
1489 }
1490 _worklist.clear();
1491 _in_worklist.Reset();
1492 return new_edges;
1493 }
1494
1495 // Put on worklist all related field nodes.
1496 void ConnectionGraph::add_field_uses_to_worklist(FieldNode* field) {
1497 assert(field->is_oop(), "sanity");
1498 int offset = field->offset();
1499 add_uses_to_worklist(field);
1500 // Loop over all bases of this field and push on worklist Field nodes
1501 // with the same offset and base (since they may reference the same field).
1502 for (BaseIterator i(field); i.has_next(); i.next()) {
1503 PointsToNode* base = i.get();
1504 add_fields_to_worklist(field, base);
1505 // Check if the base was source object of arraycopy and go over arraycopy's
1506 // destination objects since values stored to a field of source object are
1507 // accessable by uses (loads) of fields of destination objects.
1508 if (base->arraycopy_src()) {
1509 for (UseIterator j(base); j.has_next(); j.next()) {
1510 PointsToNode* arycp = j.get();
1511 if (arycp->is_Arraycopy()) {
1512 for (UseIterator k(arycp); k.has_next(); k.next()) {
1513 PointsToNode* abase = k.get();
1514 if (abase->arraycopy_dst() && abase != base) {
1515 // Look for the same arraycopy reference.
1516 add_fields_to_worklist(field, abase);
1517 }
1518 }
1519 }
1520 }
1521 }
1522 }
1523 }
1524
1525 // Put on worklist all related field nodes.
1526 void ConnectionGraph::add_fields_to_worklist(FieldNode* field, PointsToNode* base) {
1527 int offset = field->offset();
1528 if (base->is_LocalVar()) {
1529 for (UseIterator j(base); j.has_next(); j.next()) {
1530 PointsToNode* f = j.get();
1531 if (PointsToNode::is_base_use(f)) { // Field
1532 f = PointsToNode::get_use_node(f);
1533 if (f == field || !f->as_Field()->is_oop())
1534 continue;
1535 int offs = f->as_Field()->offset();
1536 if (offs == offset || offset == Type::OffsetBot || offs == Type::OffsetBot) {
1537 add_to_worklist(f);
1538 }
1539 }
1540 }
1541 } else {
1542 assert(base->is_JavaObject(), "sanity");
1543 if (// Skip phantom_object since it is only used to indicate that
1544 // this field's content globally escapes.
1545 (base != phantom_obj) &&
1546 // NULL object node does not have fields.
1547 (base != null_obj)) {
1548 for (EdgeIterator i(base); i.has_next(); i.next()) {
1549 PointsToNode* f = i.get();
1550 // Skip arraycopy edge since store to destination object field
1551 // does not update value in source object field.
1552 if (f->is_Arraycopy()) {
1553 assert(base->arraycopy_dst(), "sanity");
1554 continue;
1555 }
1556 if (f == field || !f->as_Field()->is_oop())
1557 continue;
1558 int offs = f->as_Field()->offset();
1559 if (offs == offset || offset == Type::OffsetBot || offs == Type::OffsetBot) {
1560 add_to_worklist(f);
1561 }
1562 }
1563 }
1564 }
1565 }
1566
1567 // Find fields which have unknown value.
1568 int ConnectionGraph::find_field_value(FieldNode* field) {
1569 // Escaped fields should have init value already.
1570 assert(field->escape_state() == PointsToNode::NoEscape, "sanity");
1571 int new_edges = 0;
1572 for (BaseIterator i(field); i.has_next(); i.next()) {
1573 PointsToNode* base = i.get();
1574 if (base->is_JavaObject()) {
1575 // Skip Allocate's fields which will be processed later.
1576 if (base->ideal_node()->is_Allocate())
1577 return 0;
1578 assert(base == null_obj, "only NULL ptr base expected here");
1579 }
1580 }
1581 if (add_edge(field, phantom_obj)) {
1582 // New edge was added
1583 new_edges++;
1584 add_field_uses_to_worklist(field);
1585 }
1586 return new_edges;
1587 }
1588
1589 // Find fields initializing values for allocations.
1590 int ConnectionGraph::find_init_values(JavaObjectNode* pta, PointsToNode* init_val, PhaseTransform* phase) {
1591 assert(pta->escape_state() == PointsToNode::NoEscape, "Not escaped Allocate nodes only");
1592 int new_edges = 0;
1593 Node* alloc = pta->ideal_node();
1594 if (init_val == phantom_obj) {
1595 // Do nothing for Allocate nodes since its fields values are
1596 // "known" unless they are initialized by arraycopy/clone.
1597 if (alloc->is_Allocate() && !pta->arraycopy_dst())
1598 return 0;
1599 assert(pta->arraycopy_dst() || alloc->as_CallStaticJava(), "sanity");
1600 #ifdef ASSERT
1601 if (!pta->arraycopy_dst() && alloc->as_CallStaticJava()->method() == NULL) {
1602 const char* name = alloc->as_CallStaticJava()->_name;
1603 assert(strncmp(name, "_multianewarray", 15) == 0, "sanity");
1604 }
1605 #endif
1606 // Non-escaped allocation returned from Java or runtime call have
1607 // unknown values in fields.
1608 for (EdgeIterator i(pta); i.has_next(); i.next()) {
1609 PointsToNode* field = i.get();
1610 if (field->is_Field() && field->as_Field()->is_oop()) {
1611 if (add_edge(field, phantom_obj)) {
1612 // New edge was added
1613 new_edges++;
1614 add_field_uses_to_worklist(field->as_Field());
1615 }
1616 }
1617 }
1618 return new_edges;
1619 }
1620 assert(init_val == null_obj, "sanity");
1621 // Do nothing for Call nodes since its fields values are unknown.
1622 if (!alloc->is_Allocate())
1623 return 0;
1624
1625 InitializeNode* ini = alloc->as_Allocate()->initialization();
1626 bool visited_bottom_offset = false;
1627 GrowableArray<int> offsets_worklist;
1628
1629 // Check if an oop field's initializing value is recorded and add
1630 // a corresponding NULL if field's value if it is not recorded.
1631 // Connection Graph does not record a default initialization by NULL
1632 // captured by Initialize node.
1633 //
1634 for (EdgeIterator i(pta); i.has_next(); i.next()) {
1635 PointsToNode* field = i.get(); // Field (AddP)
1636 if (!field->is_Field() || !field->as_Field()->is_oop())
1637 continue; // Not oop field
1638 int offset = field->as_Field()->offset();
1639 if (offset == Type::OffsetBot) {
1640 if (!visited_bottom_offset) {
1641 // OffsetBot is used to reference array's element,
1642 // always add reference to NULL to all Field nodes since we don't
1643 // known which element is referenced.
1644 if (add_edge(field, null_obj)) {
1645 // New edge was added
1646 new_edges++;
1647 add_field_uses_to_worklist(field->as_Field());
1648 visited_bottom_offset = true;
1649 }
1650 }
1651 } else {
1652 // Check only oop fields.
1653 const Type* adr_type = field->ideal_node()->as_AddP()->bottom_type();
1654 if (adr_type->isa_rawptr()) {
1655 #ifdef ASSERT
1656 // Raw pointers are used for initializing stores so skip it
1657 // since it should be recorded already
1658 Node* base = get_addp_base(field->ideal_node());
1659 assert(adr_type->isa_rawptr() && base->is_Proj() &&
1660 (base->in(0) == alloc),"unexpected pointer type");
1661 #endif
1662 continue;
1663 }
1664 if (!offsets_worklist.contains(offset)) {
1665 offsets_worklist.append(offset);
1666 Node* value = NULL;
1667 if (ini != NULL) {
1668 // StoreP::memory_type() == T_ADDRESS
1669 BasicType ft = UseCompressedOops ? T_NARROWOOP : T_ADDRESS;
1670 Node* store = ini->find_captured_store(offset, type2aelembytes(ft, true), phase);
1671 // Make sure initializing store has the same type as this AddP.
1672 // This AddP may reference non existing field because it is on a
1673 // dead branch of bimorphic call which is not eliminated yet.
1674 if (store != NULL && store->is_Store() &&
1675 store->as_Store()->memory_type() == ft) {
1676 value = store->in(MemNode::ValueIn);
1677 #ifdef ASSERT
1678 if (VerifyConnectionGraph) {
1679 // Verify that AddP already points to all objects the value points to.
1680 PointsToNode* val = ptnode_adr(value->_idx);
1681 assert((val != NULL), "should be processed already");
1682 PointsToNode* missed_obj = NULL;
1683 if (val->is_JavaObject()) {
1684 if (!field->points_to(val->as_JavaObject())) {
1685 missed_obj = val;
1686 }
1687 } else {
1688 if (!val->is_LocalVar() || (val->edge_count() == 0)) {
1689 tty->print_cr("----------init store has invalid value -----");
1690 store->dump();
1691 val->dump();
1692 assert(val->is_LocalVar() && (val->edge_count() > 0), "should be processed already");
1693 }
1694 for (EdgeIterator j(val); j.has_next(); j.next()) {
1695 PointsToNode* obj = j.get();
1696 if (obj->is_JavaObject()) {
1697 if (!field->points_to(obj->as_JavaObject())) {
1698 missed_obj = obj;
1699 break;
1700 }
1701 }
1702 }
1703 }
1704 if (missed_obj != NULL) {
1705 tty->print_cr("----------field---------------------------------");
1706 field->dump();
1707 tty->print_cr("----------missed referernce to object-----------");
1708 missed_obj->dump();
1709 tty->print_cr("----------object referernced by init store -----");
1710 store->dump();
1711 val->dump();
1712 assert(!field->points_to(missed_obj->as_JavaObject()), "missed JavaObject reference");
1713 }
1714 }
1715 #endif
1716 } else {
1717 // There could be initializing stores which follow allocation.
1718 // For example, a volatile field store is not collected
1719 // by Initialize node.
1720 //
1721 // Need to check for dependent loads to separate such stores from
1722 // stores which follow loads. For now, add initial value NULL so
1723 // that compare pointers optimization works correctly.
1724 }
1725 }
1726 if (value == NULL) {
1727 // A field's initializing value was not recorded. Add NULL.
1728 if (add_edge(field, null_obj)) {
1729 // New edge was added
1730 new_edges++;
1731 add_field_uses_to_worklist(field->as_Field());
1732 }
1733 }
1734 }
1735 }
1736 }
1737 return new_edges;
1738 }
1739
1740 // Adjust scalar_replaceable state after Connection Graph is built.
1741 void ConnectionGraph::adjust_scalar_replaceable_state(JavaObjectNode* jobj) {
1742 // Search for non-escaping objects which are not scalar replaceable
1743 // and mark them to propagate the state to referenced objects.
1744
1745 // 1. An object is not scalar replaceable if the field into which it is
1746 // stored has unknown offset (stored into unknown element of an array).
1747 //
1748 for (UseIterator i(jobj); i.has_next(); i.next()) {
1749 PointsToNode* use = i.get();
1750 if (use->is_Arraycopy()) {
1751 continue;
1752 }
1753 if (use->is_Field()) {
1754 FieldNode* field = use->as_Field();
1755 assert(field->is_oop() && field->scalar_replaceable(), "sanity");
1756 if (field->offset() == Type::OffsetBot) {
1757 jobj->set_scalar_replaceable(false);
1758 return;
1759 }
1760 // 2. An object is not scalar replaceable if the field into which it is
1761 // stored has multiple bases one of which is null.
1762 if (field->base_count() > 1) {
1763 for (BaseIterator i(field); i.has_next(); i.next()) {
1764 PointsToNode* base = i.get();
1765 if (base == null_obj) {
1766 jobj->set_scalar_replaceable(false);
1767 return;
1768 }
1769 }
1770 }
1771 }
1772 assert(use->is_Field() || use->is_LocalVar(), "sanity");
1773 // 3. An object is not scalar replaceable if it is merged with other objects.
1774 for (EdgeIterator j(use); j.has_next(); j.next()) {
1775 PointsToNode* ptn = j.get();
1776 if (ptn->is_JavaObject() && ptn != jobj) {
1777 // Mark all objects.
1778 jobj->set_scalar_replaceable(false);
1779 ptn->set_scalar_replaceable(false);
1780 }
1781 }
1782 if (!jobj->scalar_replaceable()) {
1783 return;
1784 }
1785 }
1786
1787 for (EdgeIterator j(jobj); j.has_next(); j.next()) {
1788 if (j.get()->is_Arraycopy()) {
1789 continue;
1790 }
1791
1792 // Non-escaping object node should point only to field nodes.
1793 FieldNode* field = j.get()->as_Field();
1794 int offset = field->as_Field()->offset();
1795
1796 // 4. An object is not scalar replaceable if it has a field with unknown
1797 // offset (array's element is accessed in loop).
1798 if (offset == Type::OffsetBot) {
1799 jobj->set_scalar_replaceable(false);
1800 return;
1801 }
1802 // 5. Currently an object is not scalar replaceable if a LoadStore node
1803 // access its field since the field value is unknown after it.
1804 //
1805 Node* n = field->ideal_node();
1806
1807 // Test for an unsafe access that was parsed as maybe off heap
1808 // (with a CheckCastPP to raw memory).
1809 assert(n->is_AddP(), "expect an address computation");
1810 if (n->in(AddPNode::Base)->is_top() &&
1811 n->in(AddPNode::Address)->Opcode() == Op_CheckCastPP) {
1812 assert(n->in(AddPNode::Address)->bottom_type()->isa_rawptr(), "raw address so raw cast expected");
1813 assert(_igvn->type(n->in(AddPNode::Address)->in(1))->isa_oopptr(), "cast pattern at unsafe access expected");
1814 jobj->set_scalar_replaceable(false);
1815 return;
1816 }
1817
1818 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1819 Node* u = n->fast_out(i);
1820 if (u->is_LoadStore() || (u->is_Mem() && u->as_Mem()->is_mismatched_access())) {
1821 jobj->set_scalar_replaceable(false);
1822 return;
1823 }
1824 }
1825
1826 // 6. Or the address may point to more then one object. This may produce
1827 // the false positive result (set not scalar replaceable)
1828 // since the flow-insensitive escape analysis can't separate
1829 // the case when stores overwrite the field's value from the case
1830 // when stores happened on different control branches.
1831 //
1832 // Note: it will disable scalar replacement in some cases:
1833 //
1834 // Point p[] = new Point[1];
1835 // p[0] = new Point(); // Will be not scalar replaced
1836 //
1837 // but it will save us from incorrect optimizations in next cases:
1838 //
1839 // Point p[] = new Point[1];
1840 // if ( x ) p[0] = new Point(); // Will be not scalar replaced
1841 //
1842 if (field->base_count() > 1) {
1843 for (BaseIterator i(field); i.has_next(); i.next()) {
1844 PointsToNode* base = i.get();
1845 // Don't take into account LocalVar nodes which
1846 // may point to only one object which should be also
1847 // this field's base by now.
1848 if (base->is_JavaObject() && base != jobj) {
1849 // Mark all bases.
1850 jobj->set_scalar_replaceable(false);
1851 base->set_scalar_replaceable(false);
1852 }
1853 }
1854 }
1855 }
1856 }
1857
1858 #ifdef ASSERT
1859 void ConnectionGraph::verify_connection_graph(
1860 GrowableArray<PointsToNode*>& ptnodes_worklist,
1861 GrowableArray<JavaObjectNode*>& non_escaped_worklist,
1862 GrowableArray<JavaObjectNode*>& java_objects_worklist,
1863 GrowableArray<Node*>& addp_worklist) {
1864 // Verify that graph is complete - no new edges could be added.
1865 int java_objects_length = java_objects_worklist.length();
1866 int non_escaped_length = non_escaped_worklist.length();
1867 int new_edges = 0;
1868 for (int next = 0; next < java_objects_length; ++next) {
1869 JavaObjectNode* ptn = java_objects_worklist.at(next);
1870 new_edges += add_java_object_edges(ptn, true);
1871 }
1872 assert(new_edges == 0, "graph was not complete");
1873 // Verify that escape state is final.
1874 int length = non_escaped_worklist.length();
1875 find_non_escaped_objects(ptnodes_worklist, non_escaped_worklist);
1876 assert((non_escaped_length == non_escaped_worklist.length()) &&
1877 (non_escaped_length == length) &&
1878 (_worklist.length() == 0), "escape state was not final");
1879
1880 // Verify fields information.
1881 int addp_length = addp_worklist.length();
1882 for (int next = 0; next < addp_length; ++next ) {
1883 Node* n = addp_worklist.at(next);
1884 FieldNode* field = ptnode_adr(n->_idx)->as_Field();
1885 if (field->is_oop()) {
1886 // Verify that field has all bases
1887 Node* base = get_addp_base(n);
1888 PointsToNode* ptn = ptnode_adr(base->_idx);
1889 if (ptn->is_JavaObject()) {
1890 assert(field->has_base(ptn->as_JavaObject()), "sanity");
1891 } else {
1892 assert(ptn->is_LocalVar(), "sanity");
1893 for (EdgeIterator i(ptn); i.has_next(); i.next()) {
1894 PointsToNode* e = i.get();
1895 if (e->is_JavaObject()) {
1896 assert(field->has_base(e->as_JavaObject()), "sanity");
1897 }
1898 }
1899 }
1900 // Verify that all fields have initializing values.
1901 if (field->edge_count() == 0) {
1902 tty->print_cr("----------field does not have references----------");
1903 field->dump();
1904 for (BaseIterator i(field); i.has_next(); i.next()) {
1905 PointsToNode* base = i.get();
1906 tty->print_cr("----------field has next base---------------------");
1907 base->dump();
1908 if (base->is_JavaObject() && (base != phantom_obj) && (base != null_obj)) {
1909 tty->print_cr("----------base has fields-------------------------");
1910 for (EdgeIterator j(base); j.has_next(); j.next()) {
1911 j.get()->dump();
1912 }
1913 tty->print_cr("----------base has references---------------------");
1914 for (UseIterator j(base); j.has_next(); j.next()) {
1915 j.get()->dump();
1916 }
1917 }
1918 }
1919 for (UseIterator i(field); i.has_next(); i.next()) {
1920 i.get()->dump();
1921 }
1922 assert(field->edge_count() > 0, "sanity");
1923 }
1924 }
1925 }
1926 }
1927 #endif
1928
1929 // Optimize ideal graph.
1930 void ConnectionGraph::optimize_ideal_graph(GrowableArray<Node*>& ptr_cmp_worklist,
1931 GrowableArray<Node*>& storestore_worklist) {
1932 Compile* C = _compile;
1933 PhaseIterGVN* igvn = _igvn;
1934 if (EliminateLocks) {
1935 // Mark locks before changing ideal graph.
1936 int cnt = C->macro_count();
1937 for( int i=0; i < cnt; i++ ) {
1938 Node *n = C->macro_node(i);
1939 if (n->is_AbstractLock()) { // Lock and Unlock nodes
1940 AbstractLockNode* alock = n->as_AbstractLock();
1941 if (!alock->is_non_esc_obj()) {
1942 if (not_global_escape(alock->obj_node())) {
1943 assert(!alock->is_eliminated() || alock->is_coarsened(), "sanity");
1944 // The lock could be marked eliminated by lock coarsening
1945 // code during first IGVN before EA. Replace coarsened flag
1946 // to eliminate all associated locks/unlocks.
1947 #ifdef ASSERT
1948 alock->log_lock_optimization(C, "eliminate_lock_set_non_esc3");
1949 #endif
1950 alock->set_non_esc_obj();
1951 }
1952 }
1953 }
1954 }
1955 }
1956
1957 if (OptimizePtrCompare) {
1958 // Add ConI(#CC_GT) and ConI(#CC_EQ).
1959 _pcmp_neq = igvn->makecon(TypeInt::CC_GT);
1960 _pcmp_eq = igvn->makecon(TypeInt::CC_EQ);
1961 // Optimize objects compare.
1962 while (ptr_cmp_worklist.length() != 0) {
1963 Node *n = ptr_cmp_worklist.pop();
1964 Node *res = optimize_ptr_compare(n);
1965 if (res != NULL) {
1966 #ifndef PRODUCT
1967 if (PrintOptimizePtrCompare) {
1968 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"));
1969 if (Verbose) {
1970 n->dump(1);
1971 }
1972 }
1973 #endif
1974 igvn->replace_node(n, res);
1975 }
1976 }
1977 // cleanup
1978 if (_pcmp_neq->outcnt() == 0)
1979 igvn->hash_delete(_pcmp_neq);
1980 if (_pcmp_eq->outcnt() == 0)
1981 igvn->hash_delete(_pcmp_eq);
1982 }
1983
1984 // For MemBarStoreStore nodes added in library_call.cpp, check
1985 // escape status of associated AllocateNode and optimize out
1986 // MemBarStoreStore node if the allocated object never escapes.
1987 while (storestore_worklist.length() != 0) {
1988 Node *n = storestore_worklist.pop();
1989 MemBarStoreStoreNode *storestore = n ->as_MemBarStoreStore();
1990 Node *alloc = storestore->in(MemBarNode::Precedent)->in(0);
1991 assert (alloc->is_Allocate(), "storestore should point to AllocateNode");
1992 if (not_global_escape(alloc)) {
1993 MemBarNode* mb = MemBarNode::make(C, Op_MemBarCPUOrder, Compile::AliasIdxBot);
1994 mb->init_req(TypeFunc::Memory, storestore->in(TypeFunc::Memory));
1995 mb->init_req(TypeFunc::Control, storestore->in(TypeFunc::Control));
1996 igvn->register_new_node_with_optimizer(mb);
1997 igvn->replace_node(storestore, mb);
1998 }
1999 }
2000 }
2001
2002 // Optimize objects compare.
2003 Node* ConnectionGraph::optimize_ptr_compare(Node* n) {
2004 assert(OptimizePtrCompare, "sanity");
2005 PointsToNode* ptn1 = ptnode_adr(n->in(1)->_idx);
2006 PointsToNode* ptn2 = ptnode_adr(n->in(2)->_idx);
2007 JavaObjectNode* jobj1 = unique_java_object(n->in(1));
2008 JavaObjectNode* jobj2 = unique_java_object(n->in(2));
2009 assert(ptn1->is_JavaObject() || ptn1->is_LocalVar(), "sanity");
2010 assert(ptn2->is_JavaObject() || ptn2->is_LocalVar(), "sanity");
2011
2012 // Check simple cases first.
2013 if (jobj1 != NULL) {
2014 if (jobj1->escape_state() == PointsToNode::NoEscape) {
2015 if (jobj1 == jobj2) {
2016 // Comparing the same not escaping object.
2017 return _pcmp_eq;
2018 }
2019 Node* obj = jobj1->ideal_node();
2020 // Comparing not escaping allocation.
2021 if ((obj->is_Allocate() || obj->is_CallStaticJava()) &&
2022 !ptn2->points_to(jobj1)) {
2023 return _pcmp_neq; // This includes nullness check.
2024 }
2025 }
2026 }
2027 if (jobj2 != NULL) {
2028 if (jobj2->escape_state() == PointsToNode::NoEscape) {
2029 Node* obj = jobj2->ideal_node();
2030 // Comparing not escaping allocation.
2031 if ((obj->is_Allocate() || obj->is_CallStaticJava()) &&
2032 !ptn1->points_to(jobj2)) {
2033 return _pcmp_neq; // This includes nullness check.
2034 }
2035 }
2036 }
2037 if (jobj1 != NULL && jobj1 != phantom_obj &&
2038 jobj2 != NULL && jobj2 != phantom_obj &&
2039 jobj1->ideal_node()->is_Con() &&
2040 jobj2->ideal_node()->is_Con()) {
2041 // Klass or String constants compare. Need to be careful with
2042 // compressed pointers - compare types of ConN and ConP instead of nodes.
2043 const Type* t1 = jobj1->ideal_node()->get_ptr_type();
2044 const Type* t2 = jobj2->ideal_node()->get_ptr_type();
2045 if (t1->make_ptr() == t2->make_ptr()) {
2046 return _pcmp_eq;
2047 } else {
2048 return _pcmp_neq;
2049 }
2050 }
2051 if (ptn1->meet(ptn2)) {
2052 return NULL; // Sets are not disjoint
2053 }
2054
2055 // Sets are disjoint.
2056 bool set1_has_unknown_ptr = ptn1->points_to(phantom_obj);
2057 bool set2_has_unknown_ptr = ptn2->points_to(phantom_obj);
2058 bool set1_has_null_ptr = ptn1->points_to(null_obj);
2059 bool set2_has_null_ptr = ptn2->points_to(null_obj);
2060 if ((set1_has_unknown_ptr && set2_has_null_ptr) ||
2061 (set2_has_unknown_ptr && set1_has_null_ptr)) {
2062 // Check nullness of unknown object.
2063 return NULL;
2064 }
2065
2066 // Disjointness by itself is not sufficient since
2067 // alias analysis is not complete for escaped objects.
2068 // Disjoint sets are definitely unrelated only when
2069 // at least one set has only not escaping allocations.
2070 if (!set1_has_unknown_ptr && !set1_has_null_ptr) {
2071 if (ptn1->non_escaping_allocation()) {
2072 return _pcmp_neq;
2073 }
2074 }
2075 if (!set2_has_unknown_ptr && !set2_has_null_ptr) {
2076 if (ptn2->non_escaping_allocation()) {
2077 return _pcmp_neq;
2078 }
2079 }
2080 return NULL;
2081 }
2082
2083 // Connection Graph constuction functions.
2084
2085 void ConnectionGraph::add_local_var(Node *n, PointsToNode::EscapeState es) {
2086 PointsToNode* ptadr = _nodes.at(n->_idx);
2087 if (ptadr != NULL) {
2088 assert(ptadr->is_LocalVar() && ptadr->ideal_node() == n, "sanity");
2089 return;
2090 }
2091 Compile* C = _compile;
2092 ptadr = new (C->comp_arena()) LocalVarNode(this, n, es);
2093 _nodes.at_put(n->_idx, ptadr);
2094 }
2095
2096 void ConnectionGraph::add_java_object(Node *n, PointsToNode::EscapeState es) {
2097 PointsToNode* ptadr = _nodes.at(n->_idx);
2098 if (ptadr != NULL) {
2099 assert(ptadr->is_JavaObject() && ptadr->ideal_node() == n, "sanity");
2100 return;
2101 }
2102 Compile* C = _compile;
2103 ptadr = new (C->comp_arena()) JavaObjectNode(this, n, es);
2104 _nodes.at_put(n->_idx, ptadr);
2105 }
2106
2107 void ConnectionGraph::add_field(Node *n, PointsToNode::EscapeState es, int offset) {
2108 PointsToNode* ptadr = _nodes.at(n->_idx);
2109 if (ptadr != NULL) {
2110 assert(ptadr->is_Field() && ptadr->ideal_node() == n, "sanity");
2111 return;
2112 }
2113 bool unsafe = false;
2114 bool is_oop = is_oop_field(n, offset, &unsafe);
2115 if (unsafe) {
2116 es = PointsToNode::GlobalEscape;
2117 }
2118 Compile* C = _compile;
2119 FieldNode* field = new (C->comp_arena()) FieldNode(this, n, es, offset, is_oop);
2120 _nodes.at_put(n->_idx, field);
2121 }
2122
2123 void ConnectionGraph::add_arraycopy(Node *n, PointsToNode::EscapeState es,
2124 PointsToNode* src, PointsToNode* dst) {
2125 assert(!src->is_Field() && !dst->is_Field(), "only for JavaObject and LocalVar");
2126 assert((src != null_obj) && (dst != null_obj), "not for ConP NULL");
2127 PointsToNode* ptadr = _nodes.at(n->_idx);
2128 if (ptadr != NULL) {
2129 assert(ptadr->is_Arraycopy() && ptadr->ideal_node() == n, "sanity");
2130 return;
2131 }
2132 Compile* C = _compile;
2133 ptadr = new (C->comp_arena()) ArraycopyNode(this, n, es);
2134 _nodes.at_put(n->_idx, ptadr);
2135 // Add edge from arraycopy node to source object.
2136 (void)add_edge(ptadr, src);
2137 src->set_arraycopy_src();
2138 // Add edge from destination object to arraycopy node.
2139 (void)add_edge(dst, ptadr);
2140 dst->set_arraycopy_dst();
2141 }
2142
2143 bool ConnectionGraph::is_oop_field(Node* n, int offset, bool* unsafe) {
2144 const Type* adr_type = n->as_AddP()->bottom_type();
2145 BasicType bt = T_INT;
2146 if (offset == Type::OffsetBot) {
2147 // Check only oop fields.
2148 if (!adr_type->isa_aryptr() ||
2149 (adr_type->isa_aryptr()->klass() == NULL) ||
2150 adr_type->isa_aryptr()->klass()->is_obj_array_klass()) {
2151 // OffsetBot is used to reference array's element. Ignore first AddP.
2152 if (find_second_addp(n, n->in(AddPNode::Base)) == NULL) {
2153 bt = T_OBJECT;
2154 }
2155 }
2156 } else if (offset != oopDesc::klass_offset_in_bytes()) {
2157 if (adr_type->isa_instptr()) {
2158 ciField* field = _compile->alias_type(adr_type->isa_instptr())->field();
2159 if (field != NULL) {
2160 bt = field->layout_type();
2161 } else {
2162 // Check for unsafe oop field access
2163 if (n->has_out_with(Op_StoreP, Op_LoadP, Op_StoreN, Op_LoadN) ||
2164 n->has_out_with(Op_GetAndSetP, Op_GetAndSetN, Op_CompareAndExchangeP, Op_CompareAndExchangeN) ||
2165 #if INCLUDE_SHENANDOAHGC
2166 n->has_out_with(Op_ShenandoahCompareAndExchangeP) || n->has_out_with(Op_ShenandoahCompareAndExchangeN) ||
2167 n->has_out_with(Op_ShenandoahCompareAndSwapP, Op_ShenandoahCompareAndSwapN, Op_ShenandoahWeakCompareAndSwapP, Op_ShenandoahWeakCompareAndSwapN) ||
2168 #endif
2169 n->has_out_with(Op_CompareAndSwapP, Op_CompareAndSwapN, Op_WeakCompareAndSwapP, Op_WeakCompareAndSwapN)) {
2170 bt = T_OBJECT;
2171 (*unsafe) = true;
2172 }
2173 }
2174 } else if (adr_type->isa_aryptr()) {
2175 if (offset == arrayOopDesc::length_offset_in_bytes()) {
2176 // Ignore array length load.
2177 } else if (find_second_addp(n, n->in(AddPNode::Base)) != NULL) {
2178 // Ignore first AddP.
2179 } else {
2180 const Type* elemtype = adr_type->isa_aryptr()->elem();
2181 bt = elemtype->array_element_basic_type();
2182 }
2183 } else if (adr_type->isa_rawptr() || adr_type->isa_klassptr()) {
2184 // Allocation initialization, ThreadLocal field access, unsafe access
2185 if (n->has_out_with(Op_StoreP, Op_LoadP, Op_StoreN, Op_LoadN) ||
2186 n->has_out_with(Op_GetAndSetP, Op_GetAndSetN, Op_CompareAndExchangeP, Op_CompareAndExchangeN) ||
2187 n->has_out_with(Op_CompareAndSwapP, Op_CompareAndSwapN, Op_WeakCompareAndSwapP, Op_WeakCompareAndSwapN)) {
2188 bt = T_OBJECT;
2189 }
2190 }
2191 }
2192 return (bt == T_OBJECT || bt == T_NARROWOOP || bt == T_ARRAY);
2193 }
2194
2195 // Returns unique pointed java object or NULL.
2196 JavaObjectNode* ConnectionGraph::unique_java_object(Node *n) {
2197 assert(!_collecting, "should not call when contructed graph");
2198 // If the node was created after the escape computation we can't answer.
2199 uint idx = n->_idx;
2200 if (idx >= nodes_size()) {
2201 return NULL;
2202 }
2203 PointsToNode* ptn = ptnode_adr(idx);
2204 if (ptn == NULL) {
2205 return NULL;
2206 }
2207 if (ptn->is_JavaObject()) {
2208 return ptn->as_JavaObject();
2209 }
2210 assert(ptn->is_LocalVar(), "sanity");
2211 // Check all java objects it points to.
2212 JavaObjectNode* jobj = NULL;
2213 for (EdgeIterator i(ptn); i.has_next(); i.next()) {
2214 PointsToNode* e = i.get();
2215 if (e->is_JavaObject()) {
2216 if (jobj == NULL) {
2217 jobj = e->as_JavaObject();
2218 } else if (jobj != e) {
2219 return NULL;
2220 }
2221 }
2222 }
2223 return jobj;
2224 }
2225
2226 // Return true if this node points only to non-escaping allocations.
2227 bool PointsToNode::non_escaping_allocation() {
2228 if (is_JavaObject()) {
2229 Node* n = ideal_node();
2230 if (n->is_Allocate() || n->is_CallStaticJava()) {
2231 return (escape_state() == PointsToNode::NoEscape);
2232 } else {
2233 return false;
2234 }
2235 }
2236 assert(is_LocalVar(), "sanity");
2237 // Check all java objects it points to.
2238 for (EdgeIterator i(this); i.has_next(); i.next()) {
2239 PointsToNode* e = i.get();
2240 if (e->is_JavaObject()) {
2241 Node* n = e->ideal_node();
2242 if ((e->escape_state() != PointsToNode::NoEscape) ||
2243 !(n->is_Allocate() || n->is_CallStaticJava())) {
2244 return false;
2245 }
2246 }
2247 }
2248 return true;
2249 }
2250
2251 // Return true if we know the node does not escape globally.
2252 bool ConnectionGraph::not_global_escape(Node *n) {
2253 assert(!_collecting, "should not call during graph construction");
2254 // If the node was created after the escape computation we can't answer.
2255 uint idx = n->_idx;
2256 if (idx >= nodes_size()) {
2257 return false;
2258 }
2259 PointsToNode* ptn = ptnode_adr(idx);
2260 if (ptn == NULL) {
2261 return false; // not in congraph (e.g. ConI)
2262 }
2263 PointsToNode::EscapeState es = ptn->escape_state();
2264 // If we have already computed a value, return it.
2265 if (es >= PointsToNode::GlobalEscape)
2266 return false;
2267 if (ptn->is_JavaObject()) {
2268 return true; // (es < PointsToNode::GlobalEscape);
2269 }
2270 assert(ptn->is_LocalVar(), "sanity");
2271 // Check all java objects it points to.
2272 for (EdgeIterator i(ptn); i.has_next(); i.next()) {
2273 if (i.get()->escape_state() >= PointsToNode::GlobalEscape)
2274 return false;
2275 }
2276 return true;
2277 }
2278
2279
2280 // Helper functions
2281
2282 // Return true if this node points to specified node or nodes it points to.
2283 bool PointsToNode::points_to(JavaObjectNode* ptn) const {
2284 if (is_JavaObject()) {
2285 return (this == ptn);
2286 }
2287 assert(is_LocalVar() || is_Field(), "sanity");
2288 for (EdgeIterator i(this); i.has_next(); i.next()) {
2289 if (i.get() == ptn)
2290 return true;
2291 }
2292 return false;
2293 }
2294
2295 // Return true if one node points to an other.
2296 bool PointsToNode::meet(PointsToNode* ptn) {
2297 if (this == ptn) {
2298 return true;
2299 } else if (ptn->is_JavaObject()) {
2300 return this->points_to(ptn->as_JavaObject());
2301 } else if (this->is_JavaObject()) {
2302 return ptn->points_to(this->as_JavaObject());
2303 }
2304 assert(this->is_LocalVar() && ptn->is_LocalVar(), "sanity");
2305 int ptn_count = ptn->edge_count();
2306 for (EdgeIterator i(this); i.has_next(); i.next()) {
2307 PointsToNode* this_e = i.get();
2308 for (int j = 0; j < ptn_count; j++) {
2309 if (this_e == ptn->edge(j))
2310 return true;
2311 }
2312 }
2313 return false;
2314 }
2315
2316 #ifdef ASSERT
2317 // Return true if bases point to this java object.
2318 bool FieldNode::has_base(JavaObjectNode* jobj) const {
2319 for (BaseIterator i(this); i.has_next(); i.next()) {
2320 if (i.get() == jobj)
2321 return true;
2322 }
2323 return false;
2324 }
2325 #endif
2326
2327 int ConnectionGraph::address_offset(Node* adr, PhaseTransform *phase) {
2328 const Type *adr_type = phase->type(adr);
2329 if (adr->is_AddP() && adr_type->isa_oopptr() == NULL &&
2330 adr->in(AddPNode::Address)->is_Proj() &&
2331 adr->in(AddPNode::Address)->in(0)->is_Allocate()) {
2332 // We are computing a raw address for a store captured by an Initialize
2333 // compute an appropriate address type. AddP cases #3 and #5 (see below).
2334 int offs = (int)phase->find_intptr_t_con(adr->in(AddPNode::Offset), Type::OffsetBot);
2335 assert(offs != Type::OffsetBot ||
2336 adr->in(AddPNode::Address)->in(0)->is_AllocateArray(),
2337 "offset must be a constant or it is initialization of array");
2338 return offs;
2339 }
2340 const TypePtr *t_ptr = adr_type->isa_ptr();
2341 assert(t_ptr != NULL, "must be a pointer type");
2342 return t_ptr->offset();
2343 }
2344
2345 Node* ConnectionGraph::get_addp_base(Node *addp) {
2346 assert(addp->is_AddP(), "must be AddP");
2347 //
2348 // AddP cases for Base and Address inputs:
2349 // case #1. Direct object's field reference:
2350 // Allocate
2351 // |
2352 // Proj #5 ( oop result )
2353 // |
2354 // CheckCastPP (cast to instance type)
2355 // | |
2356 // AddP ( base == address )
2357 //
2358 // case #2. Indirect object's field reference:
2359 // Phi
2360 // |
2361 // CastPP (cast to instance type)
2362 // | |
2363 // AddP ( base == address )
2364 //
2365 // case #3. Raw object's field reference for Initialize node:
2366 // Allocate
2367 // |
2368 // Proj #5 ( oop result )
2369 // top |
2370 // \ |
2371 // AddP ( base == top )
2372 //
2373 // case #4. Array's element reference:
2374 // {CheckCastPP | CastPP}
2375 // | | |
2376 // | AddP ( array's element offset )
2377 // | |
2378 // AddP ( array's offset )
2379 //
2380 // case #5. Raw object's field reference for arraycopy stub call:
2381 // The inline_native_clone() case when the arraycopy stub is called
2382 // after the allocation before Initialize and CheckCastPP nodes.
2383 // Allocate
2384 // |
2385 // Proj #5 ( oop result )
2386 // | |
2387 // AddP ( base == address )
2388 //
2389 // case #6. Constant Pool, ThreadLocal, CastX2P or
2390 // Raw object's field reference:
2391 // {ConP, ThreadLocal, CastX2P, raw Load}
2392 // top |
2393 // \ |
2394 // AddP ( base == top )
2395 //
2396 // case #7. Klass's field reference.
2397 // LoadKlass
2398 // | |
2399 // AddP ( base == address )
2400 //
2401 // case #8. narrow Klass's field reference.
2402 // LoadNKlass
2403 // |
2404 // DecodeN
2405 // | |
2406 // AddP ( base == address )
2407 //
2408 // case #9. Mixed unsafe access
2409 // {instance}
2410 // |
2411 // CheckCastPP (raw)
2412 // top |
2413 // \ |
2414 // AddP ( base == top )
2415 //
2416 Node *base = addp->in(AddPNode::Base);
2417 if (base->uncast()->is_top()) { // The AddP case #3 and #6 and #9.
2418 base = addp->in(AddPNode::Address);
2419 while (base->is_AddP()) {
2420 // Case #6 (unsafe access) may have several chained AddP nodes.
2421 assert(base->in(AddPNode::Base)->uncast()->is_top(), "expected unsafe access address only");
2422 base = base->in(AddPNode::Address);
2423 }
2424 if (base->Opcode() == Op_CheckCastPP &&
2425 base->bottom_type()->isa_rawptr() &&
2426 _igvn->type(base->in(1))->isa_oopptr()) {
2427 base = base->in(1); // Case #9
2428 } else {
2429 Node* uncast_base = base->uncast();
2430 int opcode = uncast_base->Opcode();
2431 assert(opcode == Op_ConP || opcode == Op_ThreadLocal ||
2432 opcode == Op_CastX2P || uncast_base->is_DecodeNarrowPtr() ||
2433 (uncast_base->is_Mem() && (uncast_base->bottom_type()->isa_rawptr() != NULL)) ||
2434 (uncast_base->is_Proj() && uncast_base->in(0)->is_Allocate()) ||
2435 uncast_base->Opcode() == Op_ShenandoahLoadReferenceBarrier, "sanity");
2436 }
2437 }
2438 return base;
2439 }
2440
2441 Node* ConnectionGraph::find_second_addp(Node* addp, Node* n) {
2442 assert(addp->is_AddP() && addp->outcnt() > 0, "Don't process dead nodes");
2443 Node* addp2 = addp->raw_out(0);
2444 if (addp->outcnt() == 1 && addp2->is_AddP() &&
2445 addp2->in(AddPNode::Base) == n &&
2446 addp2->in(AddPNode::Address) == addp) {
2447 assert(addp->in(AddPNode::Base) == n, "expecting the same base");
2448 //
2449 // Find array's offset to push it on worklist first and
2450 // as result process an array's element offset first (pushed second)
2451 // to avoid CastPP for the array's offset.
2452 // Otherwise the inserted CastPP (LocalVar) will point to what
2453 // the AddP (Field) points to. Which would be wrong since
2454 // the algorithm expects the CastPP has the same point as
2455 // as AddP's base CheckCastPP (LocalVar).
2456 //
2457 // ArrayAllocation
2458 // |
2459 // CheckCastPP
2460 // |
2461 // memProj (from ArrayAllocation CheckCastPP)
2462 // | ||
2463 // | || Int (element index)
2464 // | || | ConI (log(element size))
2465 // | || | /
2466 // | || LShift
2467 // | || /
2468 // | AddP (array's element offset)
2469 // | |
2470 // | | ConI (array's offset: #12(32-bits) or #24(64-bits))
2471 // | / /
2472 // AddP (array's offset)
2473 // |
2474 // Load/Store (memory operation on array's element)
2475 //
2476 return addp2;
2477 }
2478 return NULL;
2479 }
2480
2481 //
2482 // Adjust the type and inputs of an AddP which computes the
2483 // address of a field of an instance
2484 //
2485 bool ConnectionGraph::split_AddP(Node *addp, Node *base) {
2486 PhaseGVN* igvn = _igvn;
2487 const TypeOopPtr *base_t = igvn->type(base)->isa_oopptr();
2488 assert(base_t != NULL && base_t->is_known_instance(), "expecting instance oopptr");
2489 const TypeOopPtr *t = igvn->type(addp)->isa_oopptr();
2490 if (t == NULL) {
2491 // We are computing a raw address for a store captured by an Initialize
2492 // compute an appropriate address type (cases #3 and #5).
2493 assert(igvn->type(addp) == TypeRawPtr::NOTNULL, "must be raw pointer");
2494 assert(addp->in(AddPNode::Address)->is_Proj(), "base of raw address must be result projection from allocation");
2495 intptr_t offs = (int)igvn->find_intptr_t_con(addp->in(AddPNode::Offset), Type::OffsetBot);
2496 assert(offs != Type::OffsetBot, "offset must be a constant");
2497 t = base_t->add_offset(offs)->is_oopptr();
2498 }
2499 int inst_id = base_t->instance_id();
2500 assert(!t->is_known_instance() || t->instance_id() == inst_id,
2501 "old type must be non-instance or match new type");
2502
2503 // The type 't' could be subclass of 'base_t'.
2504 // As result t->offset() could be large then base_t's size and it will
2505 // cause the failure in add_offset() with narrow oops since TypeOopPtr()
2506 // constructor verifies correctness of the offset.
2507 //
2508 // It could happened on subclass's branch (from the type profiling
2509 // inlining) which was not eliminated during parsing since the exactness
2510 // of the allocation type was not propagated to the subclass type check.
2511 //
2512 // Or the type 't' could be not related to 'base_t' at all.
2513 // It could happened when CHA type is different from MDO type on a dead path
2514 // (for example, from instanceof check) which is not collapsed during parsing.
2515 //
2516 // Do nothing for such AddP node and don't process its users since
2517 // this code branch will go away.
2518 //
2519 if (!t->is_known_instance() &&
2520 !base_t->klass()->is_subtype_of(t->klass())) {
2521 return false; // bail out
2522 }
2523 const TypeOopPtr *tinst = base_t->add_offset(t->offset())->is_oopptr();
2524 // Do NOT remove the next line: ensure a new alias index is allocated
2525 // for the instance type. Note: C++ will not remove it since the call
2526 // has side effect.
2527 int alias_idx = _compile->get_alias_index(tinst);
2528 igvn->set_type(addp, tinst);
2529 // record the allocation in the node map
2530 set_map(addp, get_map(base->_idx));
2531 // Set addp's Base and Address to 'base'.
2532 Node *abase = addp->in(AddPNode::Base);
2533 Node *adr = addp->in(AddPNode::Address);
2534 if (adr->is_Proj() && adr->in(0)->is_Allocate() &&
2535 adr->in(0)->_idx == (uint)inst_id) {
2536 // Skip AddP cases #3 and #5.
2537 } else {
2538 assert(!abase->is_top(), "sanity"); // AddP case #3
2539 if (abase != base) {
2540 igvn->hash_delete(addp);
2541 addp->set_req(AddPNode::Base, base);
2542 if (abase == adr) {
2543 addp->set_req(AddPNode::Address, base);
2544 } else {
2545 // AddP case #4 (adr is array's element offset AddP node)
2546 #ifdef ASSERT
2547 const TypeOopPtr *atype = igvn->type(adr)->isa_oopptr();
2548 assert(adr->is_AddP() && atype != NULL &&
2549 atype->instance_id() == inst_id, "array's element offset should be processed first");
2550 #endif
2551 }
2552 igvn->hash_insert(addp);
2553 }
2554 }
2555 // Put on IGVN worklist since at least addp's type was changed above.
2556 record_for_optimizer(addp);
2557 return true;
2558 }
2559
2560 //
2561 // Create a new version of orig_phi if necessary. Returns either the newly
2562 // created phi or an existing phi. Sets create_new to indicate whether a new
2563 // phi was created. Cache the last newly created phi in the node map.
2564 //
2565 PhiNode *ConnectionGraph::create_split_phi(PhiNode *orig_phi, int alias_idx, GrowableArray<PhiNode *> &orig_phi_worklist, bool &new_created) {
2566 Compile *C = _compile;
2567 PhaseGVN* igvn = _igvn;
2568 new_created = false;
2569 int phi_alias_idx = C->get_alias_index(orig_phi->adr_type());
2570 // nothing to do if orig_phi is bottom memory or matches alias_idx
2571 if (phi_alias_idx == alias_idx) {
2572 return orig_phi;
2573 }
2574 // Have we recently created a Phi for this alias index?
2575 PhiNode *result = get_map_phi(orig_phi->_idx);
2576 if (result != NULL && C->get_alias_index(result->adr_type()) == alias_idx) {
2577 return result;
2578 }
2579 // Previous check may fail when the same wide memory Phi was split into Phis
2580 // for different memory slices. Search all Phis for this region.
2581 if (result != NULL) {
2582 Node* region = orig_phi->in(0);
2583 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
2584 Node* phi = region->fast_out(i);
2585 if (phi->is_Phi() &&
2586 C->get_alias_index(phi->as_Phi()->adr_type()) == alias_idx) {
2587 assert(phi->_idx >= nodes_size(), "only new Phi per instance memory slice");
2588 return phi->as_Phi();
2589 }
2590 }
2591 }
2592 if (C->live_nodes() + 2*NodeLimitFudgeFactor > C->max_node_limit()) {
2593 if (C->do_escape_analysis() == true && !C->failing()) {
2594 // Retry compilation without escape analysis.
2595 // If this is the first failure, the sentinel string will "stick"
2596 // to the Compile object, and the C2Compiler will see it and retry.
2597 C->record_failure(C2Compiler::retry_no_escape_analysis());
2598 }
2599 return NULL;
2600 }
2601 orig_phi_worklist.append_if_missing(orig_phi);
2602 const TypePtr *atype = C->get_adr_type(alias_idx);
2603 result = PhiNode::make(orig_phi->in(0), NULL, Type::MEMORY, atype);
2604 C->copy_node_notes_to(result, orig_phi);
2605 igvn->set_type(result, result->bottom_type());
2606 record_for_optimizer(result);
2607 set_map(orig_phi, result);
2608 new_created = true;
2609 return result;
2610 }
2611
2612 //
2613 // Return a new version of Memory Phi "orig_phi" with the inputs having the
2614 // specified alias index.
2615 //
2616 PhiNode *ConnectionGraph::split_memory_phi(PhiNode *orig_phi, int alias_idx, GrowableArray<PhiNode *> &orig_phi_worklist) {
2617 assert(alias_idx != Compile::AliasIdxBot, "can't split out bottom memory");
2618 Compile *C = _compile;
2619 PhaseGVN* igvn = _igvn;
2620 bool new_phi_created;
2621 PhiNode *result = create_split_phi(orig_phi, alias_idx, orig_phi_worklist, new_phi_created);
2622 if (!new_phi_created) {
2623 return result;
2624 }
2625 GrowableArray<PhiNode *> phi_list;
2626 GrowableArray<uint> cur_input;
2627 PhiNode *phi = orig_phi;
2628 uint idx = 1;
2629 bool finished = false;
2630 while(!finished) {
2631 while (idx < phi->req()) {
2632 Node *mem = find_inst_mem(phi->in(idx), alias_idx, orig_phi_worklist);
2633 if (mem != NULL && mem->is_Phi()) {
2634 PhiNode *newphi = create_split_phi(mem->as_Phi(), alias_idx, orig_phi_worklist, new_phi_created);
2635 if (new_phi_created) {
2636 // found an phi for which we created a new split, push current one on worklist and begin
2637 // processing new one
2638 phi_list.push(phi);
2639 cur_input.push(idx);
2640 phi = mem->as_Phi();
2641 result = newphi;
2642 idx = 1;
2643 continue;
2644 } else {
2645 mem = newphi;
2646 }
2647 }
2648 if (C->failing()) {
2649 return NULL;
2650 }
2651 result->set_req(idx++, mem);
2652 }
2653 #ifdef ASSERT
2654 // verify that the new Phi has an input for each input of the original
2655 assert( phi->req() == result->req(), "must have same number of inputs.");
2656 assert( result->in(0) != NULL && result->in(0) == phi->in(0), "regions must match");
2657 #endif
2658 // Check if all new phi's inputs have specified alias index.
2659 // Otherwise use old phi.
2660 for (uint i = 1; i < phi->req(); i++) {
2661 Node* in = result->in(i);
2662 assert((phi->in(i) == NULL) == (in == NULL), "inputs must correspond.");
2663 }
2664 // we have finished processing a Phi, see if there are any more to do
2665 finished = (phi_list.length() == 0 );
2666 if (!finished) {
2667 phi = phi_list.pop();
2668 idx = cur_input.pop();
2669 PhiNode *prev_result = get_map_phi(phi->_idx);
2670 prev_result->set_req(idx++, result);
2671 result = prev_result;
2672 }
2673 }
2674 return result;
2675 }
2676
2677 //
2678 // The next methods are derived from methods in MemNode.
2679 //
2680 Node* ConnectionGraph::step_through_mergemem(MergeMemNode *mmem, int alias_idx, const TypeOopPtr *toop) {
2681 Node *mem = mmem;
2682 // TypeOopPtr::NOTNULL+any is an OOP with unknown offset - generally
2683 // means an array I have not precisely typed yet. Do not do any
2684 // alias stuff with it any time soon.
2685 if (toop->base() != Type::AnyPtr &&
2686 !(toop->klass() != NULL &&
2687 toop->klass()->is_java_lang_Object() &&
2688 toop->offset() == Type::OffsetBot)) {
2689 mem = mmem->memory_at(alias_idx);
2690 // Update input if it is progress over what we have now
2691 }
2692 return mem;
2693 }
2694
2695 //
2696 // Move memory users to their memory slices.
2697 //
2698 void ConnectionGraph::move_inst_mem(Node* n, GrowableArray<PhiNode *> &orig_phis) {
2699 Compile* C = _compile;
2700 PhaseGVN* igvn = _igvn;
2701 const TypePtr* tp = igvn->type(n->in(MemNode::Address))->isa_ptr();
2702 assert(tp != NULL, "ptr type");
2703 int alias_idx = C->get_alias_index(tp);
2704 int general_idx = C->get_general_index(alias_idx);
2705
2706 // Move users first
2707 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
2708 Node* use = n->fast_out(i);
2709 if (use->is_MergeMem()) {
2710 MergeMemNode* mmem = use->as_MergeMem();
2711 assert(n == mmem->memory_at(alias_idx), "should be on instance memory slice");
2712 if (n != mmem->memory_at(general_idx) || alias_idx == general_idx) {
2713 continue; // Nothing to do
2714 }
2715 // Replace previous general reference to mem node.
2716 uint orig_uniq = C->unique();
2717 Node* m = find_inst_mem(n, general_idx, orig_phis);
2718 assert(orig_uniq == C->unique(), "no new nodes");
2719 mmem->set_memory_at(general_idx, m);
2720 --imax;
2721 --i;
2722 } else if (use->is_MemBar()) {
2723 assert(!use->is_Initialize(), "initializing stores should not be moved");
2724 if (use->req() > MemBarNode::Precedent &&
2725 use->in(MemBarNode::Precedent) == n) {
2726 // Don't move related membars.
2727 record_for_optimizer(use);
2728 continue;
2729 }
2730 tp = use->as_MemBar()->adr_type()->isa_ptr();
2731 if ((tp != NULL && C->get_alias_index(tp) == alias_idx) ||
2732 alias_idx == general_idx) {
2733 continue; // Nothing to do
2734 }
2735 // Move to general memory slice.
2736 uint orig_uniq = C->unique();
2737 Node* m = find_inst_mem(n, general_idx, orig_phis);
2738 assert(orig_uniq == C->unique(), "no new nodes");
2739 igvn->hash_delete(use);
2740 imax -= use->replace_edge(n, m);
2741 igvn->hash_insert(use);
2742 record_for_optimizer(use);
2743 --i;
2744 #ifdef ASSERT
2745 } else if (use->is_Mem()) {
2746 if (use->Opcode() == Op_StoreCM && use->in(MemNode::OopStore) == n) {
2747 // Don't move related cardmark.
2748 continue;
2749 }
2750 // Memory nodes should have new memory input.
2751 tp = igvn->type(use->in(MemNode::Address))->isa_ptr();
2752 assert(tp != NULL, "ptr type");
2753 int idx = C->get_alias_index(tp);
2754 assert(get_map(use->_idx) != NULL || idx == alias_idx,
2755 "Following memory nodes should have new memory input or be on the same memory slice");
2756 } else if (use->is_Phi()) {
2757 // Phi nodes should be split and moved already.
2758 tp = use->as_Phi()->adr_type()->isa_ptr();
2759 assert(tp != NULL, "ptr type");
2760 int idx = C->get_alias_index(tp);
2761 assert(idx == alias_idx, "Following Phi nodes should be on the same memory slice");
2762 } else {
2763 use->dump();
2764 assert(false, "should not be here");
2765 #endif
2766 }
2767 }
2768 }
2769
2770 //
2771 // Search memory chain of "mem" to find a MemNode whose address
2772 // is the specified alias index.
2773 //
2774 Node* ConnectionGraph::find_inst_mem(Node *orig_mem, int alias_idx, GrowableArray<PhiNode *> &orig_phis) {
2775 if (orig_mem == NULL)
2776 return orig_mem;
2777 Compile* C = _compile;
2778 PhaseGVN* igvn = _igvn;
2779 const TypeOopPtr *toop = C->get_adr_type(alias_idx)->isa_oopptr();
2780 bool is_instance = (toop != NULL) && toop->is_known_instance();
2781 Node *start_mem = C->start()->proj_out_or_null(TypeFunc::Memory);
2782 Node *prev = NULL;
2783 Node *result = orig_mem;
2784 while (prev != result) {
2785 prev = result;
2786 if (result == start_mem)
2787 break; // hit one of our sentinels
2788 if (result->is_Mem()) {
2789 const Type *at = igvn->type(result->in(MemNode::Address));
2790 if (at == Type::TOP)
2791 break; // Dead
2792 assert (at->isa_ptr() != NULL, "pointer type required.");
2793 int idx = C->get_alias_index(at->is_ptr());
2794 if (idx == alias_idx)
2795 break; // Found
2796 if (!is_instance && (at->isa_oopptr() == NULL ||
2797 !at->is_oopptr()->is_known_instance())) {
2798 break; // Do not skip store to general memory slice.
2799 }
2800 result = result->in(MemNode::Memory);
2801 }
2802 if (!is_instance)
2803 continue; // don't search further for non-instance types
2804 // skip over a call which does not affect this memory slice
2805 if (result->is_Proj() && result->as_Proj()->_con == TypeFunc::Memory) {
2806 Node *proj_in = result->in(0);
2807 if (proj_in->is_Allocate() && proj_in->_idx == (uint)toop->instance_id()) {
2808 break; // hit one of our sentinels
2809 } else if (proj_in->is_Call()) {
2810 // ArrayCopy node processed here as well
2811 CallNode *call = proj_in->as_Call();
2812 if (!call->may_modify(toop, igvn)) {
2813 result = call->in(TypeFunc::Memory);
2814 }
2815 } else if (proj_in->is_Initialize()) {
2816 AllocateNode* alloc = proj_in->as_Initialize()->allocation();
2817 // Stop if this is the initialization for the object instance which
2818 // which contains this memory slice, otherwise skip over it.
2819 if (alloc == NULL || alloc->_idx != (uint)toop->instance_id()) {
2820 result = proj_in->in(TypeFunc::Memory);
2821 }
2822 } else if (proj_in->is_MemBar()) {
2823 // Check if there is an array copy for a clone
2824 // Step over GC barrier when ReduceInitialCardMarks is disabled
2825 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
2826 Node* control_proj_ac = bs->step_over_gc_barrier(proj_in->in(0));
2827
2828 if (control_proj_ac->is_Proj() && control_proj_ac->in(0)->is_ArrayCopy()) {
2829 // Stop if it is a clone
2830 ArrayCopyNode* ac = control_proj_ac->in(0)->as_ArrayCopy();
2831 if (ac->may_modify(toop, igvn)) {
2832 break;
2833 }
2834 }
2835 result = proj_in->in(TypeFunc::Memory);
2836 }
2837 } else if (result->is_MergeMem()) {
2838 MergeMemNode *mmem = result->as_MergeMem();
2839 result = step_through_mergemem(mmem, alias_idx, toop);
2840 if (result == mmem->base_memory()) {
2841 // Didn't find instance memory, search through general slice recursively.
2842 result = mmem->memory_at(C->get_general_index(alias_idx));
2843 result = find_inst_mem(result, alias_idx, orig_phis);
2844 if (C->failing()) {
2845 return NULL;
2846 }
2847 mmem->set_memory_at(alias_idx, result);
2848 }
2849 } else if (result->is_Phi() &&
2850 C->get_alias_index(result->as_Phi()->adr_type()) != alias_idx) {
2851 Node *un = result->as_Phi()->unique_input(igvn);
2852 if (un != NULL) {
2853 orig_phis.append_if_missing(result->as_Phi());
2854 result = un;
2855 } else {
2856 break;
2857 }
2858 } else if (result->is_ClearArray()) {
2859 if (!ClearArrayNode::step_through(&result, (uint)toop->instance_id(), igvn)) {
2860 // Can not bypass initialization of the instance
2861 // we are looking for.
2862 break;
2863 }
2864 // Otherwise skip it (the call updated 'result' value).
2865 } else if (result->Opcode() == Op_SCMemProj) {
2866 Node* mem = result->in(0);
2867 Node* adr = NULL;
2868 if (mem->is_LoadStore()) {
2869 adr = mem->in(MemNode::Address);
2870 } else {
2871 assert(mem->Opcode() == Op_EncodeISOArray ||
2872 mem->Opcode() == Op_StrCompressedCopy, "sanity");
2873 adr = mem->in(3); // Memory edge corresponds to destination array
2874 }
2875 const Type *at = igvn->type(adr);
2876 if (at != Type::TOP) {
2877 assert(at->isa_ptr() != NULL, "pointer type required.");
2878 int idx = C->get_alias_index(at->is_ptr());
2879 if (idx == alias_idx) {
2880 // Assert in debug mode
2881 assert(false, "Object is not scalar replaceable if a LoadStore node accesses its field");
2882 break; // In product mode return SCMemProj node
2883 }
2884 }
2885 result = mem->in(MemNode::Memory);
2886 } else if (result->Opcode() == Op_StrInflatedCopy) {
2887 Node* adr = result->in(3); // Memory edge corresponds to destination array
2888 const Type *at = igvn->type(adr);
2889 if (at != Type::TOP) {
2890 assert(at->isa_ptr() != NULL, "pointer type required.");
2891 int idx = C->get_alias_index(at->is_ptr());
2892 if (idx == alias_idx) {
2893 // Assert in debug mode
2894 assert(false, "Object is not scalar replaceable if a StrInflatedCopy node accesses its field");
2895 break; // In product mode return SCMemProj node
2896 }
2897 }
2898 result = result->in(MemNode::Memory);
2899 }
2900 }
2901 if (result->is_Phi()) {
2902 PhiNode *mphi = result->as_Phi();
2903 assert(mphi->bottom_type() == Type::MEMORY, "memory phi required");
2904 const TypePtr *t = mphi->adr_type();
2905 if (!is_instance) {
2906 // Push all non-instance Phis on the orig_phis worklist to update inputs
2907 // during Phase 4 if needed.
2908 orig_phis.append_if_missing(mphi);
2909 } else if (C->get_alias_index(t) != alias_idx) {
2910 // Create a new Phi with the specified alias index type.
2911 result = split_memory_phi(mphi, alias_idx, orig_phis);
2912 }
2913 }
2914 // the result is either MemNode, PhiNode, InitializeNode.
2915 return result;
2916 }
2917
2918 //
2919 // Convert the types of unescaped object to instance types where possible,
2920 // propagate the new type information through the graph, and update memory
2921 // edges and MergeMem inputs to reflect the new type.
2922 //
2923 // We start with allocations (and calls which may be allocations) on alloc_worklist.
2924 // The processing is done in 4 phases:
2925 //
2926 // Phase 1: Process possible allocations from alloc_worklist. Create instance
2927 // types for the CheckCastPP for allocations where possible.
2928 // Propagate the new types through users as follows:
2929 // casts and Phi: push users on alloc_worklist
2930 // AddP: cast Base and Address inputs to the instance type
2931 // push any AddP users on alloc_worklist and push any memnode
2932 // users onto memnode_worklist.
2933 // Phase 2: Process MemNode's from memnode_worklist. compute new address type and
2934 // search the Memory chain for a store with the appropriate type
2935 // address type. If a Phi is found, create a new version with
2936 // the appropriate memory slices from each of the Phi inputs.
2937 // For stores, process the users as follows:
2938 // MemNode: push on memnode_worklist
2939 // MergeMem: push on mergemem_worklist
2940 // Phase 3: Process MergeMem nodes from mergemem_worklist. Walk each memory slice
2941 // moving the first node encountered of each instance type to the
2942 // the input corresponding to its alias index.
2943 // appropriate memory slice.
2944 // Phase 4: Update the inputs of non-instance memory Phis and the Memory input of memnodes.
2945 //
2946 // In the following example, the CheckCastPP nodes are the cast of allocation
2947 // results and the allocation of node 29 is unescaped and eligible to be an
2948 // instance type.
2949 //
2950 // We start with:
2951 //
2952 // 7 Parm #memory
2953 // 10 ConI "12"
2954 // 19 CheckCastPP "Foo"
2955 // 20 AddP _ 19 19 10 Foo+12 alias_index=4
2956 // 29 CheckCastPP "Foo"
2957 // 30 AddP _ 29 29 10 Foo+12 alias_index=4
2958 //
2959 // 40 StoreP 25 7 20 ... alias_index=4
2960 // 50 StoreP 35 40 30 ... alias_index=4
2961 // 60 StoreP 45 50 20 ... alias_index=4
2962 // 70 LoadP _ 60 30 ... alias_index=4
2963 // 80 Phi 75 50 60 Memory alias_index=4
2964 // 90 LoadP _ 80 30 ... alias_index=4
2965 // 100 LoadP _ 80 20 ... alias_index=4
2966 //
2967 //
2968 // Phase 1 creates an instance type for node 29 assigning it an instance id of 24
2969 // and creating a new alias index for node 30. This gives:
2970 //
2971 // 7 Parm #memory
2972 // 10 ConI "12"
2973 // 19 CheckCastPP "Foo"
2974 // 20 AddP _ 19 19 10 Foo+12 alias_index=4
2975 // 29 CheckCastPP "Foo" iid=24
2976 // 30 AddP _ 29 29 10 Foo+12 alias_index=6 iid=24
2977 //
2978 // 40 StoreP 25 7 20 ... alias_index=4
2979 // 50 StoreP 35 40 30 ... alias_index=6
2980 // 60 StoreP 45 50 20 ... alias_index=4
2981 // 70 LoadP _ 60 30 ... alias_index=6
2982 // 80 Phi 75 50 60 Memory alias_index=4
2983 // 90 LoadP _ 80 30 ... alias_index=6
2984 // 100 LoadP _ 80 20 ... alias_index=4
2985 //
2986 // In phase 2, new memory inputs are computed for the loads and stores,
2987 // And a new version of the phi is created. In phase 4, the inputs to
2988 // node 80 are updated and then the memory nodes are updated with the
2989 // values computed in phase 2. This results in:
2990 //
2991 // 7 Parm #memory
2992 // 10 ConI "12"
2993 // 19 CheckCastPP "Foo"
2994 // 20 AddP _ 19 19 10 Foo+12 alias_index=4
2995 // 29 CheckCastPP "Foo" iid=24
2996 // 30 AddP _ 29 29 10 Foo+12 alias_index=6 iid=24
2997 //
2998 // 40 StoreP 25 7 20 ... alias_index=4
2999 // 50 StoreP 35 7 30 ... alias_index=6
3000 // 60 StoreP 45 40 20 ... alias_index=4
3001 // 70 LoadP _ 50 30 ... alias_index=6
3002 // 80 Phi 75 40 60 Memory alias_index=4
3003 // 120 Phi 75 50 50 Memory alias_index=6
3004 // 90 LoadP _ 120 30 ... alias_index=6
3005 // 100 LoadP _ 80 20 ... alias_index=4
3006 //
3007 void ConnectionGraph::split_unique_types(GrowableArray<Node *> &alloc_worklist, GrowableArray<ArrayCopyNode*> &arraycopy_worklist) {
3008 GrowableArray<Node *> memnode_worklist;
3009 GrowableArray<PhiNode *> orig_phis;
3010 PhaseIterGVN *igvn = _igvn;
3011 uint new_index_start = (uint) _compile->num_alias_types();
3012 Arena* arena = Thread::current()->resource_area();
3013 VectorSet visited(arena);
3014 ideal_nodes.clear(); // Reset for use with set_map/get_map.
3015 uint unique_old = _compile->unique();
3016
3017 // Phase 1: Process possible allocations from alloc_worklist.
3018 // Create instance types for the CheckCastPP for allocations where possible.
3019 //
3020 // (Note: don't forget to change the order of the second AddP node on
3021 // the alloc_worklist if the order of the worklist processing is changed,
3022 // see the comment in find_second_addp().)
3023 //
3024 while (alloc_worklist.length() != 0) {
3025 Node *n = alloc_worklist.pop();
3026 uint ni = n->_idx;
3027 if (n->is_Call()) {
3028 CallNode *alloc = n->as_Call();
3029 // copy escape information to call node
3030 PointsToNode* ptn = ptnode_adr(alloc->_idx);
3031 PointsToNode::EscapeState es = ptn->escape_state();
3032 // We have an allocation or call which returns a Java object,
3033 // see if it is unescaped.
3034 if (es != PointsToNode::NoEscape || !ptn->scalar_replaceable())
3035 continue;
3036 // Find CheckCastPP for the allocate or for the return value of a call
3037 n = alloc->result_cast();
3038 if (n == NULL) { // No uses except Initialize node
3039 if (alloc->is_Allocate()) {
3040 // Set the scalar_replaceable flag for allocation
3041 // so it could be eliminated if it has no uses.
3042 alloc->as_Allocate()->_is_scalar_replaceable = true;
3043 }
3044 if (alloc->is_CallStaticJava()) {
3045 // Set the scalar_replaceable flag for boxing method
3046 // so it could be eliminated if it has no uses.
3047 alloc->as_CallStaticJava()->_is_scalar_replaceable = true;
3048 }
3049 continue;
3050 }
3051 if (!n->is_CheckCastPP()) { // not unique CheckCastPP.
3052 assert(!alloc->is_Allocate(), "allocation should have unique type");
3053 continue;
3054 }
3055
3056 // The inline code for Object.clone() casts the allocation result to
3057 // java.lang.Object and then to the actual type of the allocated
3058 // object. Detect this case and use the second cast.
3059 // Also detect j.l.reflect.Array.newInstance(jobject, jint) case when
3060 // the allocation result is cast to java.lang.Object and then
3061 // to the actual Array type.
3062 if (alloc->is_Allocate() && n->as_Type()->type() == TypeInstPtr::NOTNULL
3063 && (alloc->is_AllocateArray() ||
3064 igvn->type(alloc->in(AllocateNode::KlassNode)) != TypeKlassPtr::OBJECT)) {
3065 Node *cast2 = NULL;
3066 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
3067 Node *use = n->fast_out(i);
3068 if (use->is_CheckCastPP()) {
3069 cast2 = use;
3070 break;
3071 }
3072 }
3073 if (cast2 != NULL) {
3074 n = cast2;
3075 } else {
3076 // Non-scalar replaceable if the allocation type is unknown statically
3077 // (reflection allocation), the object can't be restored during
3078 // deoptimization without precise type.
3079 continue;
3080 }
3081 }
3082
3083 const TypeOopPtr *t = igvn->type(n)->isa_oopptr();
3084 if (t == NULL)
3085 continue; // not a TypeOopPtr
3086 if (!t->klass_is_exact())
3087 continue; // not an unique type
3088
3089 if (alloc->is_Allocate()) {
3090 // Set the scalar_replaceable flag for allocation
3091 // so it could be eliminated.
3092 alloc->as_Allocate()->_is_scalar_replaceable = true;
3093 }
3094 if (alloc->is_CallStaticJava()) {
3095 // Set the scalar_replaceable flag for boxing method
3096 // so it could be eliminated.
3097 alloc->as_CallStaticJava()->_is_scalar_replaceable = true;
3098 }
3099 set_escape_state(ptnode_adr(n->_idx), es); // CheckCastPP escape state
3100 // in order for an object to be scalar-replaceable, it must be:
3101 // - a direct allocation (not a call returning an object)
3102 // - non-escaping
3103 // - eligible to be a unique type
3104 // - not determined to be ineligible by escape analysis
3105 set_map(alloc, n);
3106 set_map(n, alloc);
3107 const TypeOopPtr* tinst = t->cast_to_instance_id(ni);
3108 igvn->hash_delete(n);
3109 igvn->set_type(n, tinst);
3110 n->raise_bottom_type(tinst);
3111 igvn->hash_insert(n);
3112 record_for_optimizer(n);
3113 // Allocate an alias index for the header fields. Accesses to
3114 // the header emitted during macro expansion wouldn't have
3115 // correct memory state otherwise.
3116 _compile->get_alias_index(tinst->add_offset(oopDesc::mark_offset_in_bytes()));
3117 _compile->get_alias_index(tinst->add_offset(oopDesc::klass_offset_in_bytes()));
3118 if (alloc->is_Allocate() && (t->isa_instptr() || t->isa_aryptr())) {
3119
3120 // First, put on the worklist all Field edges from Connection Graph
3121 // which is more accurate than putting immediate users from Ideal Graph.
3122 for (EdgeIterator e(ptn); e.has_next(); e.next()) {
3123 PointsToNode* tgt = e.get();
3124 if (tgt->is_Arraycopy()) {
3125 continue;
3126 }
3127 Node* use = tgt->ideal_node();
3128 assert(tgt->is_Field() && use->is_AddP(),
3129 "only AddP nodes are Field edges in CG");
3130 if (use->outcnt() > 0) { // Don't process dead nodes
3131 Node* addp2 = find_second_addp(use, use->in(AddPNode::Base));
3132 if (addp2 != NULL) {
3133 assert(alloc->is_AllocateArray(),"array allocation was expected");
3134 alloc_worklist.append_if_missing(addp2);
3135 }
3136 alloc_worklist.append_if_missing(use);
3137 }
3138 }
3139
3140 // An allocation may have an Initialize which has raw stores. Scan
3141 // the users of the raw allocation result and push AddP users
3142 // on alloc_worklist.
3143 Node *raw_result = alloc->proj_out_or_null(TypeFunc::Parms);
3144 assert (raw_result != NULL, "must have an allocation result");
3145 for (DUIterator_Fast imax, i = raw_result->fast_outs(imax); i < imax; i++) {
3146 Node *use = raw_result->fast_out(i);
3147 if (use->is_AddP() && use->outcnt() > 0) { // Don't process dead nodes
3148 Node* addp2 = find_second_addp(use, raw_result);
3149 if (addp2 != NULL) {
3150 assert(alloc->is_AllocateArray(),"array allocation was expected");
3151 alloc_worklist.append_if_missing(addp2);
3152 }
3153 alloc_worklist.append_if_missing(use);
3154 } else if (use->is_MemBar()) {
3155 memnode_worklist.append_if_missing(use);
3156 }
3157 }
3158 }
3159 } else if (n->is_AddP()) {
3160 JavaObjectNode* jobj = unique_java_object(get_addp_base(n));
3161 if (jobj == NULL || jobj == phantom_obj) {
3162 #ifdef ASSERT
3163 ptnode_adr(get_addp_base(n)->_idx)->dump();
3164 ptnode_adr(n->_idx)->dump();
3165 assert(jobj != NULL && jobj != phantom_obj, "escaped allocation");
3166 #endif
3167 _compile->record_failure(C2Compiler::retry_no_escape_analysis());
3168 return;
3169 }
3170 Node *base = get_map(jobj->idx()); // CheckCastPP node
3171 if (!split_AddP(n, base)) continue; // wrong type from dead path
3172 } else if (n->is_Phi() ||
3173 n->is_CheckCastPP() ||
3174 n->is_EncodeP() ||
3175 n->is_DecodeN() ||
3176 (n->is_ConstraintCast() && n->Opcode() == Op_CastPP)) {
3177 if (visited.test_set(n->_idx)) {
3178 assert(n->is_Phi(), "loops only through Phi's");
3179 continue; // already processed
3180 }
3181 JavaObjectNode* jobj = unique_java_object(n);
3182 if (jobj == NULL || jobj == phantom_obj) {
3183 #ifdef ASSERT
3184 ptnode_adr(n->_idx)->dump();
3185 assert(jobj != NULL && jobj != phantom_obj, "escaped allocation");
3186 #endif
3187 _compile->record_failure(C2Compiler::retry_no_escape_analysis());
3188 return;
3189 } else {
3190 Node *val = get_map(jobj->idx()); // CheckCastPP node
3191 TypeNode *tn = n->as_Type();
3192 const TypeOopPtr* tinst = igvn->type(val)->isa_oopptr();
3193 assert(tinst != NULL && tinst->is_known_instance() &&
3194 tinst->instance_id() == jobj->idx() , "instance type expected.");
3195
3196 const Type *tn_type = igvn->type(tn);
3197 const TypeOopPtr *tn_t;
3198 if (tn_type->isa_narrowoop()) {
3199 tn_t = tn_type->make_ptr()->isa_oopptr();
3200 } else {
3201 tn_t = tn_type->isa_oopptr();
3202 }
3203 if (tn_t != NULL && tinst->klass()->is_subtype_of(tn_t->klass())) {
3204 if (tn_type->isa_narrowoop()) {
3205 tn_type = tinst->make_narrowoop();
3206 } else {
3207 tn_type = tinst;
3208 }
3209 igvn->hash_delete(tn);
3210 igvn->set_type(tn, tn_type);
3211 tn->set_type(tn_type);
3212 igvn->hash_insert(tn);
3213 record_for_optimizer(n);
3214 } else {
3215 assert(tn_type == TypePtr::NULL_PTR ||
3216 tn_t != NULL && !tinst->klass()->is_subtype_of(tn_t->klass()),
3217 "unexpected type");
3218 continue; // Skip dead path with different type
3219 }
3220 }
3221 } else {
3222 debug_only(n->dump();)
3223 assert(false, "EA: unexpected node");
3224 continue;
3225 }
3226 // push allocation's users on appropriate worklist
3227 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
3228 Node *use = n->fast_out(i);
3229 if(use->is_Mem() && use->in(MemNode::Address) == n) {
3230 // Load/store to instance's field
3231 memnode_worklist.append_if_missing(use);
3232 } else if (use->is_MemBar()) {
3233 if (use->in(TypeFunc::Memory) == n) { // Ignore precedent edge
3234 memnode_worklist.append_if_missing(use);
3235 }
3236 } else if (use->is_AddP() && use->outcnt() > 0) { // No dead nodes
3237 Node* addp2 = find_second_addp(use, n);
3238 if (addp2 != NULL) {
3239 alloc_worklist.append_if_missing(addp2);
3240 }
3241 alloc_worklist.append_if_missing(use);
3242 } else if (use->is_Phi() ||
3243 use->is_CheckCastPP() ||
3244 use->is_EncodeNarrowPtr() ||
3245 use->is_DecodeNarrowPtr() ||
3246 (use->is_ConstraintCast() && use->Opcode() == Op_CastPP)) {
3247 alloc_worklist.append_if_missing(use);
3248 #ifdef ASSERT
3249 } else if (use->is_Mem()) {
3250 assert(use->in(MemNode::Address) != n, "EA: missing allocation reference path");
3251 } else if (use->is_MergeMem()) {
3252 assert(_mergemem_worklist.contains(use->as_MergeMem()), "EA: missing MergeMem node in the worklist");
3253 } else if (use->is_SafePoint()) {
3254 // Look for MergeMem nodes for calls which reference unique allocation
3255 // (through CheckCastPP nodes) even for debug info.
3256 Node* m = use->in(TypeFunc::Memory);
3257 if (m->is_MergeMem()) {
3258 assert(_mergemem_worklist.contains(m->as_MergeMem()), "EA: missing MergeMem node in the worklist");
3259 }
3260 } else if (use->Opcode() == Op_EncodeISOArray) {
3261 if (use->in(MemNode::Memory) == n || use->in(3) == n) {
3262 // EncodeISOArray overwrites destination array
3263 memnode_worklist.append_if_missing(use);
3264 }
3265 } else {
3266 uint op = use->Opcode();
3267 if ((op == Op_StrCompressedCopy || op == Op_StrInflatedCopy) &&
3268 (use->in(MemNode::Memory) == n)) {
3269 // They overwrite memory edge corresponding to destination array,
3270 memnode_worklist.append_if_missing(use);
3271 } else if (!(op == Op_CmpP || op == Op_Conv2B ||
3272 op == Op_CastP2X || op == Op_StoreCM ||
3273 op == Op_FastLock || op == Op_AryEq || op == Op_StrComp || op == Op_HasNegatives ||
3274 op == Op_StrCompressedCopy || op == Op_StrInflatedCopy ||
3275 op == Op_StrEquals || op == Op_StrIndexOf || op == Op_StrIndexOfChar ||
3276 BarrierSet::barrier_set()->barrier_set_c2()->is_gc_barrier_node(use))) {
3277 n->dump();
3278 use->dump();
3279 assert(false, "EA: missing allocation reference path");
3280 }
3281 #endif
3282 }
3283 }
3284
3285 }
3286
3287 // Go over all ArrayCopy nodes and if one of the inputs has a unique
3288 // type, record it in the ArrayCopy node so we know what memory this
3289 // node uses/modified.
3290 for (int next = 0; next < arraycopy_worklist.length(); next++) {
3291 ArrayCopyNode* ac = arraycopy_worklist.at(next);
3292 Node* dest = ac->in(ArrayCopyNode::Dest);
3293 if (dest->is_AddP()) {
3294 dest = get_addp_base(dest);
3295 }
3296 JavaObjectNode* jobj = unique_java_object(dest);
3297 if (jobj != NULL) {
3298 Node *base = get_map(jobj->idx());
3299 if (base != NULL) {
3300 const TypeOopPtr *base_t = _igvn->type(base)->isa_oopptr();
3301 ac->_dest_type = base_t;
3302 }
3303 }
3304 Node* src = ac->in(ArrayCopyNode::Src);
3305 if (src->is_AddP()) {
3306 src = get_addp_base(src);
3307 }
3308 jobj = unique_java_object(src);
3309 if (jobj != NULL) {
3310 Node* base = get_map(jobj->idx());
3311 if (base != NULL) {
3312 const TypeOopPtr *base_t = _igvn->type(base)->isa_oopptr();
3313 ac->_src_type = base_t;
3314 }
3315 }
3316 }
3317
3318 // New alias types were created in split_AddP().
3319 uint new_index_end = (uint) _compile->num_alias_types();
3320 assert(unique_old == _compile->unique(), "there should be no new ideal nodes after Phase 1");
3321
3322 // Phase 2: Process MemNode's from memnode_worklist. compute new address type and
3323 // compute new values for Memory inputs (the Memory inputs are not
3324 // actually updated until phase 4.)
3325 if (memnode_worklist.length() == 0)
3326 return; // nothing to do
3327 while (memnode_worklist.length() != 0) {
3328 Node *n = memnode_worklist.pop();
3329 if (visited.test_set(n->_idx))
3330 continue;
3331 if (n->is_Phi() || n->is_ClearArray()) {
3332 // we don't need to do anything, but the users must be pushed
3333 } else if (n->is_MemBar()) { // Initialize, MemBar nodes
3334 // we don't need to do anything, but the users must be pushed
3335 n = n->as_MemBar()->proj_out_or_null(TypeFunc::Memory);
3336 if (n == NULL)
3337 continue;
3338 } else if (n->Opcode() == Op_StrCompressedCopy ||
3339 n->Opcode() == Op_EncodeISOArray) {
3340 // get the memory projection
3341 n = n->find_out_with(Op_SCMemProj);
3342 assert(n != NULL && n->Opcode() == Op_SCMemProj, "memory projection required");
3343 } else {
3344 assert(n->is_Mem(), "memory node required.");
3345 Node *addr = n->in(MemNode::Address);
3346 const Type *addr_t = igvn->type(addr);
3347 if (addr_t == Type::TOP)
3348 continue;
3349 assert (addr_t->isa_ptr() != NULL, "pointer type required.");
3350 int alias_idx = _compile->get_alias_index(addr_t->is_ptr());
3351 assert ((uint)alias_idx < new_index_end, "wrong alias index");
3352 Node *mem = find_inst_mem(n->in(MemNode::Memory), alias_idx, orig_phis);
3353 if (_compile->failing()) {
3354 return;
3355 }
3356 if (mem != n->in(MemNode::Memory)) {
3357 // We delay the memory edge update since we need old one in
3358 // MergeMem code below when instances memory slices are separated.
3359 set_map(n, mem);
3360 }
3361 if (n->is_Load()) {
3362 continue; // don't push users
3363 } else if (n->is_LoadStore()) {
3364 // get the memory projection
3365 n = n->find_out_with(Op_SCMemProj);
3366 assert(n != NULL && n->Opcode() == Op_SCMemProj, "memory projection required");
3367 }
3368 }
3369 // push user on appropriate worklist
3370 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
3371 Node *use = n->fast_out(i);
3372 if (use->is_Phi() || use->is_ClearArray()) {
3373 memnode_worklist.append_if_missing(use);
3374 } else if (use->is_Mem() && use->in(MemNode::Memory) == n) {
3375 if (use->Opcode() == Op_StoreCM) // Ignore cardmark stores
3376 continue;
3377 memnode_worklist.append_if_missing(use);
3378 } else if (use->is_MemBar()) {
3379 if (use->in(TypeFunc::Memory) == n) { // Ignore precedent edge
3380 memnode_worklist.append_if_missing(use);
3381 }
3382 #ifdef ASSERT
3383 } else if(use->is_Mem()) {
3384 assert(use->in(MemNode::Memory) != n, "EA: missing memory path");
3385 } else if (use->is_MergeMem()) {
3386 assert(_mergemem_worklist.contains(use->as_MergeMem()), "EA: missing MergeMem node in the worklist");
3387 } else if (use->Opcode() == Op_EncodeISOArray) {
3388 if (use->in(MemNode::Memory) == n || use->in(3) == n) {
3389 // EncodeISOArray overwrites destination array
3390 memnode_worklist.append_if_missing(use);
3391 }
3392 } else {
3393 uint op = use->Opcode();
3394 if ((use->in(MemNode::Memory) == n) &&
3395 (op == Op_StrCompressedCopy || op == Op_StrInflatedCopy)) {
3396 // They overwrite memory edge corresponding to destination array,
3397 memnode_worklist.append_if_missing(use);
3398 } else if (!(BarrierSet::barrier_set()->barrier_set_c2()->is_gc_barrier_node(use) ||
3399 op == Op_AryEq || op == Op_StrComp || op == Op_HasNegatives ||
3400 op == Op_StrCompressedCopy || op == Op_StrInflatedCopy ||
3401 op == Op_StrEquals || op == Op_StrIndexOf || op == Op_StrIndexOfChar)) {
3402 n->dump();
3403 use->dump();
3404 assert(false, "EA: missing memory path");
3405 }
3406 #endif
3407 }
3408 }
3409 }
3410
3411 // Phase 3: Process MergeMem nodes from mergemem_worklist.
3412 // Walk each memory slice moving the first node encountered of each
3413 // instance type to the the input corresponding to its alias index.
3414 uint length = _mergemem_worklist.length();
3415 for( uint next = 0; next < length; ++next ) {
3416 MergeMemNode* nmm = _mergemem_worklist.at(next);
3417 assert(!visited.test_set(nmm->_idx), "should not be visited before");
3418 // Note: we don't want to use MergeMemStream here because we only want to
3419 // scan inputs which exist at the start, not ones we add during processing.
3420 // Note 2: MergeMem may already contains instance memory slices added
3421 // during find_inst_mem() call when memory nodes were processed above.
3422 igvn->hash_delete(nmm);
3423 uint nslices = MIN2(nmm->req(), new_index_start);
3424 for (uint i = Compile::AliasIdxRaw+1; i < nslices; i++) {
3425 Node* mem = nmm->in(i);
3426 Node* cur = NULL;
3427 if (mem == NULL || mem->is_top())
3428 continue;
3429 // First, update mergemem by moving memory nodes to corresponding slices
3430 // if their type became more precise since this mergemem was created.
3431 while (mem->is_Mem()) {
3432 const Type *at = igvn->type(mem->in(MemNode::Address));
3433 if (at != Type::TOP) {
3434 assert (at->isa_ptr() != NULL, "pointer type required.");
3435 uint idx = (uint)_compile->get_alias_index(at->is_ptr());
3436 if (idx == i) {
3437 if (cur == NULL)
3438 cur = mem;
3439 } else {
3440 if (idx >= nmm->req() || nmm->is_empty_memory(nmm->in(idx))) {
3441 nmm->set_memory_at(idx, mem);
3442 }
3443 }
3444 }
3445 mem = mem->in(MemNode::Memory);
3446 }
3447 nmm->set_memory_at(i, (cur != NULL) ? cur : mem);
3448 // Find any instance of the current type if we haven't encountered
3449 // already a memory slice of the instance along the memory chain.
3450 for (uint ni = new_index_start; ni < new_index_end; ni++) {
3451 if((uint)_compile->get_general_index(ni) == i) {
3452 Node *m = (ni >= nmm->req()) ? nmm->empty_memory() : nmm->in(ni);
3453 if (nmm->is_empty_memory(m)) {
3454 Node* result = find_inst_mem(mem, ni, orig_phis);
3455 if (_compile->failing()) {
3456 return;
3457 }
3458 nmm->set_memory_at(ni, result);
3459 }
3460 }
3461 }
3462 }
3463 // Find the rest of instances values
3464 for (uint ni = new_index_start; ni < new_index_end; ni++) {
3465 const TypeOopPtr *tinst = _compile->get_adr_type(ni)->isa_oopptr();
3466 Node* result = step_through_mergemem(nmm, ni, tinst);
3467 if (result == nmm->base_memory()) {
3468 // Didn't find instance memory, search through general slice recursively.
3469 result = nmm->memory_at(_compile->get_general_index(ni));
3470 result = find_inst_mem(result, ni, orig_phis);
3471 if (_compile->failing()) {
3472 return;
3473 }
3474 nmm->set_memory_at(ni, result);
3475 }
3476 }
3477 igvn->hash_insert(nmm);
3478 record_for_optimizer(nmm);
3479 }
3480
3481 // Phase 4: Update the inputs of non-instance memory Phis and
3482 // the Memory input of memnodes
3483 // First update the inputs of any non-instance Phi's from
3484 // which we split out an instance Phi. Note we don't have
3485 // to recursively process Phi's encounted on the input memory
3486 // chains as is done in split_memory_phi() since they will
3487 // also be processed here.
3488 for (int j = 0; j < orig_phis.length(); j++) {
3489 PhiNode *phi = orig_phis.at(j);
3490 int alias_idx = _compile->get_alias_index(phi->adr_type());
3491 igvn->hash_delete(phi);
3492 for (uint i = 1; i < phi->req(); i++) {
3493 Node *mem = phi->in(i);
3494 Node *new_mem = find_inst_mem(mem, alias_idx, orig_phis);
3495 if (_compile->failing()) {
3496 return;
3497 }
3498 if (mem != new_mem) {
3499 phi->set_req(i, new_mem);
3500 }
3501 }
3502 igvn->hash_insert(phi);
3503 record_for_optimizer(phi);
3504 }
3505
3506 // Update the memory inputs of MemNodes with the value we computed
3507 // in Phase 2 and move stores memory users to corresponding memory slices.
3508 // Disable memory split verification code until the fix for 6984348.
3509 // Currently it produces false negative results since it does not cover all cases.
3510 #if 0 // ifdef ASSERT
3511 visited.Reset();
3512 Node_Stack old_mems(arena, _compile->unique() >> 2);
3513 #endif
3514 for (uint i = 0; i < ideal_nodes.size(); i++) {
3515 Node* n = ideal_nodes.at(i);
3516 Node* nmem = get_map(n->_idx);
3517 assert(nmem != NULL, "sanity");
3518 if (n->is_Mem()) {
3519 #if 0 // ifdef ASSERT
3520 Node* old_mem = n->in(MemNode::Memory);
3521 if (!visited.test_set(old_mem->_idx)) {
3522 old_mems.push(old_mem, old_mem->outcnt());
3523 }
3524 #endif
3525 assert(n->in(MemNode::Memory) != nmem, "sanity");
3526 if (!n->is_Load()) {
3527 // Move memory users of a store first.
3528 move_inst_mem(n, orig_phis);
3529 }
3530 // Now update memory input
3531 igvn->hash_delete(n);
3532 n->set_req(MemNode::Memory, nmem);
3533 igvn->hash_insert(n);
3534 record_for_optimizer(n);
3535 } else {
3536 assert(n->is_Allocate() || n->is_CheckCastPP() ||
3537 n->is_AddP() || n->is_Phi(), "unknown node used for set_map()");
3538 }
3539 }
3540 #if 0 // ifdef ASSERT
3541 // Verify that memory was split correctly
3542 while (old_mems.is_nonempty()) {
3543 Node* old_mem = old_mems.node();
3544 uint old_cnt = old_mems.index();
3545 old_mems.pop();
3546 assert(old_cnt == old_mem->outcnt(), "old mem could be lost");
3547 }
3548 #endif
3549 }
3550
3551 #ifndef PRODUCT
3552 static const char *node_type_names[] = {
3553 "UnknownType",
3554 "JavaObject",
3555 "LocalVar",
3556 "Field",
3557 "Arraycopy"
3558 };
3559
3560 static const char *esc_names[] = {
3561 "UnknownEscape",
3562 "NoEscape",
3563 "ArgEscape",
3564 "GlobalEscape"
3565 };
3566
3567 void PointsToNode::dump(bool print_state) const {
3568 NodeType nt = node_type();
3569 tty->print("%s ", node_type_names[(int) nt]);
3570 if (print_state) {
3571 EscapeState es = escape_state();
3572 EscapeState fields_es = fields_escape_state();
3573 tty->print("%s(%s) ", esc_names[(int)es], esc_names[(int)fields_es]);
3574 if (nt == PointsToNode::JavaObject && !this->scalar_replaceable())
3575 tty->print("NSR ");
3576 }
3577 if (is_Field()) {
3578 FieldNode* f = (FieldNode*)this;
3579 if (f->is_oop())
3580 tty->print("oop ");
3581 if (f->offset() > 0)
3582 tty->print("+%d ", f->offset());
3583 tty->print("(");
3584 for (BaseIterator i(f); i.has_next(); i.next()) {
3585 PointsToNode* b = i.get();
3586 tty->print(" %d%s", b->idx(),(b->is_JavaObject() ? "P" : ""));
3587 }
3588 tty->print(" )");
3589 }
3590 tty->print("[");
3591 for (EdgeIterator i(this); i.has_next(); i.next()) {
3592 PointsToNode* e = i.get();
3593 tty->print(" %d%s%s", e->idx(),(e->is_JavaObject() ? "P" : (e->is_Field() ? "F" : "")), e->is_Arraycopy() ? "cp" : "");
3594 }
3595 tty->print(" [");
3596 for (UseIterator i(this); i.has_next(); i.next()) {
3597 PointsToNode* u = i.get();
3598 bool is_base = false;
3599 if (PointsToNode::is_base_use(u)) {
3600 is_base = true;
3601 u = PointsToNode::get_use_node(u)->as_Field();
3602 }
3603 tty->print(" %d%s%s", u->idx(), is_base ? "b" : "", u->is_Arraycopy() ? "cp" : "");
3604 }
3605 tty->print(" ]] ");
3606 if (_node == NULL)
3607 tty->print_cr("<null>");
3608 else
3609 _node->dump();
3610 }
3611
3612 void ConnectionGraph::dump(GrowableArray<PointsToNode*>& ptnodes_worklist) {
3613 bool first = true;
3614 int ptnodes_length = ptnodes_worklist.length();
3615 for (int i = 0; i < ptnodes_length; i++) {
3616 PointsToNode *ptn = ptnodes_worklist.at(i);
3617 if (ptn == NULL || !ptn->is_JavaObject())
3618 continue;
3619 PointsToNode::EscapeState es = ptn->escape_state();
3620 if ((es != PointsToNode::NoEscape) && !Verbose) {
3621 continue;
3622 }
3623 Node* n = ptn->ideal_node();
3624 if (n->is_Allocate() || (n->is_CallStaticJava() &&
3625 n->as_CallStaticJava()->is_boxing_method())) {
3626 if (first) {
3627 tty->cr();
3628 tty->print("======== Connection graph for ");
3629 _compile->method()->print_short_name();
3630 tty->cr();
3631 first = false;
3632 }
3633 ptn->dump();
3634 // Print all locals and fields which reference this allocation
3635 for (UseIterator j(ptn); j.has_next(); j.next()) {
3636 PointsToNode* use = j.get();
3637 if (use->is_LocalVar()) {
3638 use->dump(Verbose);
3639 } else if (Verbose) {
3640 use->dump();
3641 }
3642 }
3643 tty->cr();
3644 }
3645 }
3646 }
3647 #endif