Thu, 16 Jan 2014 14:25:51 +0100
8029101: PPC64 (part 211): ordering of Independent Reads of Independent Writes
Reviewed-by: dholmes, kvn
Contributed-by: martin.doerr@sap.com
1 /*
2 * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "compiler/compileLog.hpp"
27 #include "interpreter/linkResolver.hpp"
28 #include "oops/method.hpp"
29 #include "opto/addnode.hpp"
30 #include "opto/idealGraphPrinter.hpp"
31 #include "opto/locknode.hpp"
32 #include "opto/memnode.hpp"
33 #include "opto/parse.hpp"
34 #include "opto/rootnode.hpp"
35 #include "opto/runtime.hpp"
36 #include "runtime/arguments.hpp"
37 #include "runtime/handles.inline.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "utilities/copy.hpp"
41 // Static array so we can figure out which bytecodes stop us from compiling
42 // the most. Some of the non-static variables are needed in bytecodeInfo.cpp
43 // and eventually should be encapsulated in a proper class (gri 8/18/98).
45 int nodes_created = 0;
46 int methods_parsed = 0;
47 int methods_seen = 0;
48 int blocks_parsed = 0;
49 int blocks_seen = 0;
51 int explicit_null_checks_inserted = 0;
52 int explicit_null_checks_elided = 0;
53 int all_null_checks_found = 0, implicit_null_checks = 0;
54 int implicit_null_throws = 0;
56 int reclaim_idx = 0;
57 int reclaim_in = 0;
58 int reclaim_node = 0;
60 #ifndef PRODUCT
61 bool Parse::BytecodeParseHistogram::_initialized = false;
62 uint Parse::BytecodeParseHistogram::_bytecodes_parsed [Bytecodes::number_of_codes];
63 uint Parse::BytecodeParseHistogram::_nodes_constructed[Bytecodes::number_of_codes];
64 uint Parse::BytecodeParseHistogram::_nodes_transformed[Bytecodes::number_of_codes];
65 uint Parse::BytecodeParseHistogram::_new_values [Bytecodes::number_of_codes];
66 #endif
68 //------------------------------print_statistics-------------------------------
69 #ifndef PRODUCT
70 void Parse::print_statistics() {
71 tty->print_cr("--- Compiler Statistics ---");
72 tty->print("Methods seen: %d Methods parsed: %d", methods_seen, methods_parsed);
73 tty->print(" Nodes created: %d", nodes_created);
74 tty->cr();
75 if (methods_seen != methods_parsed)
76 tty->print_cr("Reasons for parse failures (NOT cumulative):");
77 tty->print_cr("Blocks parsed: %d Blocks seen: %d", blocks_parsed, blocks_seen);
79 if( explicit_null_checks_inserted )
80 tty->print_cr("%d original NULL checks - %d elided (%2d%%); optimizer leaves %d,", explicit_null_checks_inserted, explicit_null_checks_elided, (100*explicit_null_checks_elided)/explicit_null_checks_inserted, all_null_checks_found);
81 if( all_null_checks_found )
82 tty->print_cr("%d made implicit (%2d%%)", implicit_null_checks,
83 (100*implicit_null_checks)/all_null_checks_found);
84 if( implicit_null_throws )
85 tty->print_cr("%d implicit null exceptions at runtime",
86 implicit_null_throws);
88 if( PrintParseStatistics && BytecodeParseHistogram::initialized() ) {
89 BytecodeParseHistogram::print();
90 }
91 }
92 #endif
94 //------------------------------ON STACK REPLACEMENT---------------------------
96 // Construct a node which can be used to get incoming state for
97 // on stack replacement.
98 Node *Parse::fetch_interpreter_state(int index,
99 BasicType bt,
100 Node *local_addrs,
101 Node *local_addrs_base) {
102 Node *mem = memory(Compile::AliasIdxRaw);
103 Node *adr = basic_plus_adr( local_addrs_base, local_addrs, -index*wordSize );
104 Node *ctl = control();
106 // Very similar to LoadNode::make, except we handle un-aligned longs and
107 // doubles on Sparc. Intel can handle them just fine directly.
108 Node *l;
109 switch (bt) { // Signature is flattened
110 case T_INT: l = new (C) LoadINode(ctl, mem, adr, TypeRawPtr::BOTTOM, TypeInt::INT, MemNode::unordered); break;
111 case T_FLOAT: l = new (C) LoadFNode(ctl, mem, adr, TypeRawPtr::BOTTOM, Type::FLOAT, MemNode::unordered); break;
112 case T_ADDRESS: l = new (C) LoadPNode(ctl, mem, adr, TypeRawPtr::BOTTOM, TypeRawPtr::BOTTOM, MemNode::unordered); break;
113 case T_OBJECT: l = new (C) LoadPNode(ctl, mem, adr, TypeRawPtr::BOTTOM, TypeInstPtr::BOTTOM, MemNode::unordered); break;
114 case T_LONG:
115 case T_DOUBLE: {
116 // Since arguments are in reverse order, the argument address 'adr'
117 // refers to the back half of the long/double. Recompute adr.
118 adr = basic_plus_adr(local_addrs_base, local_addrs, -(index+1)*wordSize);
119 if (Matcher::misaligned_doubles_ok) {
120 l = (bt == T_DOUBLE)
121 ? (Node*)new (C) LoadDNode(ctl, mem, adr, TypeRawPtr::BOTTOM, Type::DOUBLE, MemNode::unordered)
122 : (Node*)new (C) LoadLNode(ctl, mem, adr, TypeRawPtr::BOTTOM, TypeLong::LONG, MemNode::unordered);
123 } else {
124 l = (bt == T_DOUBLE)
125 ? (Node*)new (C) LoadD_unalignedNode(ctl, mem, adr, TypeRawPtr::BOTTOM, MemNode::unordered)
126 : (Node*)new (C) LoadL_unalignedNode(ctl, mem, adr, TypeRawPtr::BOTTOM, MemNode::unordered);
127 }
128 break;
129 }
130 default: ShouldNotReachHere();
131 }
132 return _gvn.transform(l);
133 }
135 // Helper routine to prevent the interpreter from handing
136 // unexpected typestate to an OSR method.
137 // The Node l is a value newly dug out of the interpreter frame.
138 // The type is the type predicted by ciTypeFlow. Note that it is
139 // not a general type, but can only come from Type::get_typeflow_type.
140 // The safepoint is a map which will feed an uncommon trap.
141 Node* Parse::check_interpreter_type(Node* l, const Type* type,
142 SafePointNode* &bad_type_exit) {
144 const TypeOopPtr* tp = type->isa_oopptr();
146 // TypeFlow may assert null-ness if a type appears unloaded.
147 if (type == TypePtr::NULL_PTR ||
148 (tp != NULL && !tp->klass()->is_loaded())) {
149 // Value must be null, not a real oop.
150 Node* chk = _gvn.transform( new (C) CmpPNode(l, null()) );
151 Node* tst = _gvn.transform( new (C) BoolNode(chk, BoolTest::eq) );
152 IfNode* iff = create_and_map_if(control(), tst, PROB_MAX, COUNT_UNKNOWN);
153 set_control(_gvn.transform( new (C) IfTrueNode(iff) ));
154 Node* bad_type = _gvn.transform( new (C) IfFalseNode(iff) );
155 bad_type_exit->control()->add_req(bad_type);
156 l = null();
157 }
159 // Typeflow can also cut off paths from the CFG, based on
160 // types which appear unloaded, or call sites which appear unlinked.
161 // When paths are cut off, values at later merge points can rise
162 // toward more specific classes. Make sure these specific classes
163 // are still in effect.
164 if (tp != NULL && tp->klass() != C->env()->Object_klass()) {
165 // TypeFlow asserted a specific object type. Value must have that type.
166 Node* bad_type_ctrl = NULL;
167 l = gen_checkcast(l, makecon(TypeKlassPtr::make(tp->klass())), &bad_type_ctrl);
168 bad_type_exit->control()->add_req(bad_type_ctrl);
169 }
171 BasicType bt_l = _gvn.type(l)->basic_type();
172 BasicType bt_t = type->basic_type();
173 assert(_gvn.type(l)->higher_equal(type), "must constrain OSR typestate");
174 return l;
175 }
177 // Helper routine which sets up elements of the initial parser map when
178 // performing a parse for on stack replacement. Add values into map.
179 // The only parameter contains the address of a interpreter arguments.
180 void Parse::load_interpreter_state(Node* osr_buf) {
181 int index;
182 int max_locals = jvms()->loc_size();
183 int max_stack = jvms()->stk_size();
186 // Mismatch between method and jvms can occur since map briefly held
187 // an OSR entry state (which takes up one RawPtr word).
188 assert(max_locals == method()->max_locals(), "sanity");
189 assert(max_stack >= method()->max_stack(), "sanity");
190 assert((int)jvms()->endoff() == TypeFunc::Parms + max_locals + max_stack, "sanity");
191 assert((int)jvms()->endoff() == (int)map()->req(), "sanity");
193 // Find the start block.
194 Block* osr_block = start_block();
195 assert(osr_block->start() == osr_bci(), "sanity");
197 // Set initial BCI.
198 set_parse_bci(osr_block->start());
200 // Set initial stack depth.
201 set_sp(osr_block->start_sp());
203 // Check bailouts. We currently do not perform on stack replacement
204 // of loops in catch blocks or loops which branch with a non-empty stack.
205 if (sp() != 0) {
206 C->record_method_not_compilable("OSR starts with non-empty stack");
207 return;
208 }
209 // Do not OSR inside finally clauses:
210 if (osr_block->has_trap_at(osr_block->start())) {
211 C->record_method_not_compilable("OSR starts with an immediate trap");
212 return;
213 }
215 // Commute monitors from interpreter frame to compiler frame.
216 assert(jvms()->monitor_depth() == 0, "should be no active locks at beginning of osr");
217 int mcnt = osr_block->flow()->monitor_count();
218 Node *monitors_addr = basic_plus_adr(osr_buf, osr_buf, (max_locals+mcnt*2-1)*wordSize);
219 for (index = 0; index < mcnt; index++) {
220 // Make a BoxLockNode for the monitor.
221 Node *box = _gvn.transform(new (C) BoxLockNode(next_monitor()));
224 // Displaced headers and locked objects are interleaved in the
225 // temp OSR buffer. We only copy the locked objects out here.
226 // Fetch the locked object from the OSR temp buffer and copy to our fastlock node.
227 Node *lock_object = fetch_interpreter_state(index*2, T_OBJECT, monitors_addr, osr_buf);
228 // Try and copy the displaced header to the BoxNode
229 Node *displaced_hdr = fetch_interpreter_state((index*2) + 1, T_ADDRESS, monitors_addr, osr_buf);
232 store_to_memory(control(), box, displaced_hdr, T_ADDRESS, Compile::AliasIdxRaw, MemNode::unordered);
234 // Build a bogus FastLockNode (no code will be generated) and push the
235 // monitor into our debug info.
236 const FastLockNode *flock = _gvn.transform(new (C) FastLockNode( 0, lock_object, box ))->as_FastLock();
237 map()->push_monitor(flock);
239 // If the lock is our method synchronization lock, tuck it away in
240 // _sync_lock for return and rethrow exit paths.
241 if (index == 0 && method()->is_synchronized()) {
242 _synch_lock = flock;
243 }
244 }
246 // Use the raw liveness computation to make sure that unexpected
247 // values don't propagate into the OSR frame.
248 MethodLivenessResult live_locals = method()->liveness_at_bci(osr_bci());
249 if (!live_locals.is_valid()) {
250 // Degenerate or breakpointed method.
251 C->record_method_not_compilable("OSR in empty or breakpointed method");
252 return;
253 }
255 // Extract the needed locals from the interpreter frame.
256 Node *locals_addr = basic_plus_adr(osr_buf, osr_buf, (max_locals-1)*wordSize);
258 // find all the locals that the interpreter thinks contain live oops
259 const BitMap live_oops = method()->live_local_oops_at_bci(osr_bci());
260 for (index = 0; index < max_locals; index++) {
262 if (!live_locals.at(index)) {
263 continue;
264 }
266 const Type *type = osr_block->local_type_at(index);
268 if (type->isa_oopptr() != NULL) {
270 // 6403625: Verify that the interpreter oopMap thinks that the oop is live
271 // else we might load a stale oop if the MethodLiveness disagrees with the
272 // result of the interpreter. If the interpreter says it is dead we agree
273 // by making the value go to top.
274 //
276 if (!live_oops.at(index)) {
277 if (C->log() != NULL) {
278 C->log()->elem("OSR_mismatch local_index='%d'",index);
279 }
280 set_local(index, null());
281 // and ignore it for the loads
282 continue;
283 }
284 }
286 // Filter out TOP, HALF, and BOTTOM. (Cf. ensure_phi.)
287 if (type == Type::TOP || type == Type::HALF) {
288 continue;
289 }
290 // If the type falls to bottom, then this must be a local that
291 // is mixing ints and oops or some such. Forcing it to top
292 // makes it go dead.
293 if (type == Type::BOTTOM) {
294 continue;
295 }
296 // Construct code to access the appropriate local.
297 BasicType bt = type->basic_type();
298 if (type == TypePtr::NULL_PTR) {
299 // Ptr types are mixed together with T_ADDRESS but NULL is
300 // really for T_OBJECT types so correct it.
301 bt = T_OBJECT;
302 }
303 Node *value = fetch_interpreter_state(index, bt, locals_addr, osr_buf);
304 set_local(index, value);
305 }
307 // Extract the needed stack entries from the interpreter frame.
308 for (index = 0; index < sp(); index++) {
309 const Type *type = osr_block->stack_type_at(index);
310 if (type != Type::TOP) {
311 // Currently the compiler bails out when attempting to on stack replace
312 // at a bci with a non-empty stack. We should not reach here.
313 ShouldNotReachHere();
314 }
315 }
317 // End the OSR migration
318 make_runtime_call(RC_LEAF, OptoRuntime::osr_end_Type(),
319 CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_end),
320 "OSR_migration_end", TypeRawPtr::BOTTOM,
321 osr_buf);
323 // Now that the interpreter state is loaded, make sure it will match
324 // at execution time what the compiler is expecting now:
325 SafePointNode* bad_type_exit = clone_map();
326 bad_type_exit->set_control(new (C) RegionNode(1));
328 assert(osr_block->flow()->jsrs()->size() == 0, "should be no jsrs live at osr point");
329 for (index = 0; index < max_locals; index++) {
330 if (stopped()) break;
331 Node* l = local(index);
332 if (l->is_top()) continue; // nothing here
333 const Type *type = osr_block->local_type_at(index);
334 if (type->isa_oopptr() != NULL) {
335 if (!live_oops.at(index)) {
336 // skip type check for dead oops
337 continue;
338 }
339 }
340 if (osr_block->flow()->local_type_at(index)->is_return_address()) {
341 // In our current system it's illegal for jsr addresses to be
342 // live into an OSR entry point because the compiler performs
343 // inlining of jsrs. ciTypeFlow has a bailout that detect this
344 // case and aborts the compile if addresses are live into an OSR
345 // entry point. Because of that we can assume that any address
346 // locals at the OSR entry point are dead. Method liveness
347 // isn't precise enought to figure out that they are dead in all
348 // cases so simply skip checking address locals all
349 // together. Any type check is guaranteed to fail since the
350 // interpreter type is the result of a load which might have any
351 // value and the expected type is a constant.
352 continue;
353 }
354 set_local(index, check_interpreter_type(l, type, bad_type_exit));
355 }
357 for (index = 0; index < sp(); index++) {
358 if (stopped()) break;
359 Node* l = stack(index);
360 if (l->is_top()) continue; // nothing here
361 const Type *type = osr_block->stack_type_at(index);
362 set_stack(index, check_interpreter_type(l, type, bad_type_exit));
363 }
365 if (bad_type_exit->control()->req() > 1) {
366 // Build an uncommon trap here, if any inputs can be unexpected.
367 bad_type_exit->set_control(_gvn.transform( bad_type_exit->control() ));
368 record_for_igvn(bad_type_exit->control());
369 SafePointNode* types_are_good = map();
370 set_map(bad_type_exit);
371 // The unexpected type happens because a new edge is active
372 // in the CFG, which typeflow had previously ignored.
373 // E.g., Object x = coldAtFirst() && notReached()? "str": new Integer(123).
374 // This x will be typed as Integer if notReached is not yet linked.
375 // It could also happen due to a problem in ciTypeFlow analysis.
376 uncommon_trap(Deoptimization::Reason_constraint,
377 Deoptimization::Action_reinterpret);
378 set_map(types_are_good);
379 }
380 }
382 //------------------------------Parse------------------------------------------
383 // Main parser constructor.
384 Parse::Parse(JVMState* caller, ciMethod* parse_method, float expected_uses, Parse* parent)
385 : _exits(caller), _parent(parent)
386 {
387 // Init some variables
388 _caller = caller;
389 _method = parse_method;
390 _expected_uses = expected_uses;
391 _depth = 1 + (caller->has_method() ? caller->depth() : 0);
392 _wrote_final = false;
393 _wrote_volatile = false;
394 _alloc_with_final = NULL;
395 _entry_bci = InvocationEntryBci;
396 _tf = NULL;
397 _block = NULL;
398 debug_only(_block_count = -1);
399 debug_only(_blocks = (Block*)-1);
400 #ifndef PRODUCT
401 if (PrintCompilation || PrintOpto) {
402 // Make sure I have an inline tree, so I can print messages about it.
403 JVMState* ilt_caller = is_osr_parse() ? caller->caller() : caller;
404 InlineTree::find_subtree_from_root(C->ilt(), ilt_caller, parse_method);
405 }
406 _max_switch_depth = 0;
407 _est_switch_depth = 0;
408 #endif
410 _tf = TypeFunc::make(method());
411 _iter.reset_to_method(method());
412 _flow = method()->get_flow_analysis();
413 if (_flow->failing()) {
414 C->record_method_not_compilable_all_tiers(_flow->failure_reason());
415 }
417 #ifndef PRODUCT
418 if (_flow->has_irreducible_entry()) {
419 C->set_parsed_irreducible_loop(true);
420 }
421 #endif
423 if (_expected_uses <= 0) {
424 _prof_factor = 1;
425 } else {
426 float prof_total = parse_method->interpreter_invocation_count();
427 if (prof_total <= _expected_uses) {
428 _prof_factor = 1;
429 } else {
430 _prof_factor = _expected_uses / prof_total;
431 }
432 }
434 CompileLog* log = C->log();
435 if (log != NULL) {
436 log->begin_head("parse method='%d' uses='%g'",
437 log->identify(parse_method), expected_uses);
438 if (depth() == 1 && C->is_osr_compilation()) {
439 log->print(" osr_bci='%d'", C->entry_bci());
440 }
441 log->stamp();
442 log->end_head();
443 }
445 // Accumulate deoptimization counts.
446 // (The range_check and store_check counts are checked elsewhere.)
447 ciMethodData* md = method()->method_data();
448 for (uint reason = 0; reason < md->trap_reason_limit(); reason++) {
449 uint md_count = md->trap_count(reason);
450 if (md_count != 0) {
451 if (md_count == md->trap_count_limit())
452 md_count += md->overflow_trap_count();
453 uint total_count = C->trap_count(reason);
454 uint old_count = total_count;
455 total_count += md_count;
456 // Saturate the add if it overflows.
457 if (total_count < old_count || total_count < md_count)
458 total_count = (uint)-1;
459 C->set_trap_count(reason, total_count);
460 if (log != NULL)
461 log->elem("observe trap='%s' count='%d' total='%d'",
462 Deoptimization::trap_reason_name(reason),
463 md_count, total_count);
464 }
465 }
466 // Accumulate total sum of decompilations, also.
467 C->set_decompile_count(C->decompile_count() + md->decompile_count());
469 _count_invocations = C->do_count_invocations();
470 _method_data_update = C->do_method_data_update();
472 if (log != NULL && method()->has_exception_handlers()) {
473 log->elem("observe that='has_exception_handlers'");
474 }
476 assert(method()->can_be_compiled(), "Can not parse this method, cutout earlier");
477 assert(method()->has_balanced_monitors(), "Can not parse unbalanced monitors, cutout earlier");
479 // Always register dependence if JVMTI is enabled, because
480 // either breakpoint setting or hotswapping of methods may
481 // cause deoptimization.
482 if (C->env()->jvmti_can_hotswap_or_post_breakpoint()) {
483 C->dependencies()->assert_evol_method(method());
484 }
486 methods_seen++;
488 // Do some special top-level things.
489 if (depth() == 1 && C->is_osr_compilation()) {
490 _entry_bci = C->entry_bci();
491 _flow = method()->get_osr_flow_analysis(osr_bci());
492 if (_flow->failing()) {
493 C->record_method_not_compilable(_flow->failure_reason());
494 #ifndef PRODUCT
495 if (PrintOpto && (Verbose || WizardMode)) {
496 tty->print_cr("OSR @%d type flow bailout: %s", _entry_bci, _flow->failure_reason());
497 if (Verbose) {
498 method()->print();
499 method()->print_codes();
500 _flow->print();
501 }
502 }
503 #endif
504 }
505 _tf = C->tf(); // the OSR entry type is different
506 }
508 #ifdef ASSERT
509 if (depth() == 1) {
510 assert(C->is_osr_compilation() == this->is_osr_parse(), "OSR in sync");
511 if (C->tf() != tf()) {
512 MutexLockerEx ml(Compile_lock, Mutex::_no_safepoint_check_flag);
513 assert(C->env()->system_dictionary_modification_counter_changed(),
514 "Must invalidate if TypeFuncs differ");
515 }
516 } else {
517 assert(!this->is_osr_parse(), "no recursive OSR");
518 }
519 #endif
521 methods_parsed++;
522 #ifndef PRODUCT
523 // add method size here to guarantee that inlined methods are added too
524 if (TimeCompiler)
525 _total_bytes_compiled += method()->code_size();
527 show_parse_info();
528 #endif
530 if (failing()) {
531 if (log) log->done("parse");
532 return;
533 }
535 gvn().set_type(root(), root()->bottom_type());
536 gvn().transform(top());
538 // Import the results of the ciTypeFlow.
539 init_blocks();
541 // Merge point for all normal exits
542 build_exits();
544 // Setup the initial JVM state map.
545 SafePointNode* entry_map = create_entry_map();
547 // Check for bailouts during map initialization
548 if (failing() || entry_map == NULL) {
549 if (log) log->done("parse");
550 return;
551 }
553 Node_Notes* caller_nn = C->default_node_notes();
554 // Collect debug info for inlined calls unless -XX:-DebugInlinedCalls.
555 if (DebugInlinedCalls || depth() == 1) {
556 C->set_default_node_notes(make_node_notes(caller_nn));
557 }
559 if (is_osr_parse()) {
560 Node* osr_buf = entry_map->in(TypeFunc::Parms+0);
561 entry_map->set_req(TypeFunc::Parms+0, top());
562 set_map(entry_map);
563 load_interpreter_state(osr_buf);
564 } else {
565 set_map(entry_map);
566 do_method_entry();
567 }
569 // Check for bailouts during method entry.
570 if (failing()) {
571 if (log) log->done("parse");
572 C->set_default_node_notes(caller_nn);
573 return;
574 }
576 entry_map = map(); // capture any changes performed by method setup code
577 assert(jvms()->endoff() == map()->req(), "map matches JVMS layout");
579 // We begin parsing as if we have just encountered a jump to the
580 // method entry.
581 Block* entry_block = start_block();
582 assert(entry_block->start() == (is_osr_parse() ? osr_bci() : 0), "");
583 set_map_clone(entry_map);
584 merge_common(entry_block, entry_block->next_path_num());
586 #ifndef PRODUCT
587 BytecodeParseHistogram *parse_histogram_obj = new (C->env()->arena()) BytecodeParseHistogram(this, C);
588 set_parse_histogram( parse_histogram_obj );
589 #endif
591 // Parse all the basic blocks.
592 do_all_blocks();
594 C->set_default_node_notes(caller_nn);
596 // Check for bailouts during conversion to graph
597 if (failing()) {
598 if (log) log->done("parse");
599 return;
600 }
602 // Fix up all exiting control flow.
603 set_map(entry_map);
604 do_exits();
606 if (log) log->done("parse nodes='%d' live='%d' memory='%d'",
607 C->unique(), C->live_nodes(), C->node_arena()->used());
608 }
610 //---------------------------do_all_blocks-------------------------------------
611 void Parse::do_all_blocks() {
612 bool has_irreducible = flow()->has_irreducible_entry();
614 // Walk over all blocks in Reverse Post-Order.
615 while (true) {
616 bool progress = false;
617 for (int rpo = 0; rpo < block_count(); rpo++) {
618 Block* block = rpo_at(rpo);
620 if (block->is_parsed()) continue;
622 if (!block->is_merged()) {
623 // Dead block, no state reaches this block
624 continue;
625 }
627 // Prepare to parse this block.
628 load_state_from(block);
630 if (stopped()) {
631 // Block is dead.
632 continue;
633 }
635 blocks_parsed++;
637 progress = true;
638 if (block->is_loop_head() || block->is_handler() || has_irreducible && !block->is_ready()) {
639 // Not all preds have been parsed. We must build phis everywhere.
640 // (Note that dead locals do not get phis built, ever.)
641 ensure_phis_everywhere();
643 if (block->is_SEL_head() &&
644 (UseLoopPredicate || LoopLimitCheck)) {
645 // Add predicate to single entry (not irreducible) loop head.
646 assert(!block->has_merged_backedge(), "only entry paths should be merged for now");
647 // Need correct bci for predicate.
648 // It is fine to set it here since do_one_block() will set it anyway.
649 set_parse_bci(block->start());
650 add_predicate();
651 // Add new region for back branches.
652 int edges = block->pred_count() - block->preds_parsed() + 1; // +1 for original region
653 RegionNode *r = new (C) RegionNode(edges+1);
654 _gvn.set_type(r, Type::CONTROL);
655 record_for_igvn(r);
656 r->init_req(edges, control());
657 set_control(r);
658 // Add new phis.
659 ensure_phis_everywhere();
660 }
662 // Leave behind an undisturbed copy of the map, for future merges.
663 set_map(clone_map());
664 }
666 if (control()->is_Region() && !block->is_loop_head() && !has_irreducible && !block->is_handler()) {
667 // In the absence of irreducible loops, the Region and Phis
668 // associated with a merge that doesn't involve a backedge can
669 // be simplified now since the RPO parsing order guarantees
670 // that any path which was supposed to reach here has already
671 // been parsed or must be dead.
672 Node* c = control();
673 Node* result = _gvn.transform_no_reclaim(control());
674 if (c != result && TraceOptoParse) {
675 tty->print_cr("Block #%d replace %d with %d", block->rpo(), c->_idx, result->_idx);
676 }
677 if (result != top()) {
678 record_for_igvn(result);
679 }
680 }
682 // Parse the block.
683 do_one_block();
685 // Check for bailouts.
686 if (failing()) return;
687 }
689 // with irreducible loops multiple passes might be necessary to parse everything
690 if (!has_irreducible || !progress) {
691 break;
692 }
693 }
695 blocks_seen += block_count();
697 #ifndef PRODUCT
698 // Make sure there are no half-processed blocks remaining.
699 // Every remaining unprocessed block is dead and may be ignored now.
700 for (int rpo = 0; rpo < block_count(); rpo++) {
701 Block* block = rpo_at(rpo);
702 if (!block->is_parsed()) {
703 if (TraceOptoParse) {
704 tty->print_cr("Skipped dead block %d at bci:%d", rpo, block->start());
705 }
706 assert(!block->is_merged(), "no half-processed blocks");
707 }
708 }
709 #endif
710 }
712 //-------------------------------build_exits----------------------------------
713 // Build normal and exceptional exit merge points.
714 void Parse::build_exits() {
715 // make a clone of caller to prevent sharing of side-effects
716 _exits.set_map(_exits.clone_map());
717 _exits.clean_stack(_exits.sp());
718 _exits.sync_jvms();
720 RegionNode* region = new (C) RegionNode(1);
721 record_for_igvn(region);
722 gvn().set_type_bottom(region);
723 _exits.set_control(region);
725 // Note: iophi and memphi are not transformed until do_exits.
726 Node* iophi = new (C) PhiNode(region, Type::ABIO);
727 Node* memphi = new (C) PhiNode(region, Type::MEMORY, TypePtr::BOTTOM);
728 gvn().set_type_bottom(iophi);
729 gvn().set_type_bottom(memphi);
730 _exits.set_i_o(iophi);
731 _exits.set_all_memory(memphi);
733 // Add a return value to the exit state. (Do not push it yet.)
734 if (tf()->range()->cnt() > TypeFunc::Parms) {
735 const Type* ret_type = tf()->range()->field_at(TypeFunc::Parms);
736 // Don't "bind" an unloaded return klass to the ret_phi. If the klass
737 // becomes loaded during the subsequent parsing, the loaded and unloaded
738 // types will not join when we transform and push in do_exits().
739 const TypeOopPtr* ret_oop_type = ret_type->isa_oopptr();
740 if (ret_oop_type && !ret_oop_type->klass()->is_loaded()) {
741 ret_type = TypeOopPtr::BOTTOM;
742 }
743 int ret_size = type2size[ret_type->basic_type()];
744 Node* ret_phi = new (C) PhiNode(region, ret_type);
745 gvn().set_type_bottom(ret_phi);
746 _exits.ensure_stack(ret_size);
747 assert((int)(tf()->range()->cnt() - TypeFunc::Parms) == ret_size, "good tf range");
748 assert(method()->return_type()->size() == ret_size, "tf agrees w/ method");
749 _exits.set_argument(0, ret_phi); // here is where the parser finds it
750 // Note: ret_phi is not yet pushed, until do_exits.
751 }
752 }
755 //----------------------------build_start_state-------------------------------
756 // Construct a state which contains only the incoming arguments from an
757 // unknown caller. The method & bci will be NULL & InvocationEntryBci.
758 JVMState* Compile::build_start_state(StartNode* start, const TypeFunc* tf) {
759 int arg_size = tf->domain()->cnt();
760 int max_size = MAX2(arg_size, (int)tf->range()->cnt());
761 JVMState* jvms = new (this) JVMState(max_size - TypeFunc::Parms);
762 SafePointNode* map = new (this) SafePointNode(max_size, NULL);
763 record_for_igvn(map);
764 assert(arg_size == TypeFunc::Parms + (is_osr_compilation() ? 1 : method()->arg_size()), "correct arg_size");
765 Node_Notes* old_nn = default_node_notes();
766 if (old_nn != NULL && has_method()) {
767 Node_Notes* entry_nn = old_nn->clone(this);
768 JVMState* entry_jvms = new(this) JVMState(method(), old_nn->jvms());
769 entry_jvms->set_offsets(0);
770 entry_jvms->set_bci(entry_bci());
771 entry_nn->set_jvms(entry_jvms);
772 set_default_node_notes(entry_nn);
773 }
774 uint i;
775 for (i = 0; i < (uint)arg_size; i++) {
776 Node* parm = initial_gvn()->transform(new (this) ParmNode(start, i));
777 map->init_req(i, parm);
778 // Record all these guys for later GVN.
779 record_for_igvn(parm);
780 }
781 for (; i < map->req(); i++) {
782 map->init_req(i, top());
783 }
784 assert(jvms->argoff() == TypeFunc::Parms, "parser gets arguments here");
785 set_default_node_notes(old_nn);
786 map->set_jvms(jvms);
787 jvms->set_map(map);
788 return jvms;
789 }
791 //-----------------------------make_node_notes---------------------------------
792 Node_Notes* Parse::make_node_notes(Node_Notes* caller_nn) {
793 if (caller_nn == NULL) return NULL;
794 Node_Notes* nn = caller_nn->clone(C);
795 JVMState* caller_jvms = nn->jvms();
796 JVMState* jvms = new (C) JVMState(method(), caller_jvms);
797 jvms->set_offsets(0);
798 jvms->set_bci(_entry_bci);
799 nn->set_jvms(jvms);
800 return nn;
801 }
804 //--------------------------return_values--------------------------------------
805 void Compile::return_values(JVMState* jvms) {
806 GraphKit kit(jvms);
807 Node* ret = new (this) ReturnNode(TypeFunc::Parms,
808 kit.control(),
809 kit.i_o(),
810 kit.reset_memory(),
811 kit.frameptr(),
812 kit.returnadr());
813 // Add zero or 1 return values
814 int ret_size = tf()->range()->cnt() - TypeFunc::Parms;
815 if (ret_size > 0) {
816 kit.inc_sp(-ret_size); // pop the return value(s)
817 kit.sync_jvms();
818 ret->add_req(kit.argument(0));
819 // Note: The second dummy edge is not needed by a ReturnNode.
820 }
821 // bind it to root
822 root()->add_req(ret);
823 record_for_igvn(ret);
824 initial_gvn()->transform_no_reclaim(ret);
825 }
827 //------------------------rethrow_exceptions-----------------------------------
828 // Bind all exception states in the list into a single RethrowNode.
829 void Compile::rethrow_exceptions(JVMState* jvms) {
830 GraphKit kit(jvms);
831 if (!kit.has_exceptions()) return; // nothing to generate
832 // Load my combined exception state into the kit, with all phis transformed:
833 SafePointNode* ex_map = kit.combine_and_pop_all_exception_states();
834 Node* ex_oop = kit.use_exception_state(ex_map);
835 RethrowNode* exit = new (this) RethrowNode(kit.control(),
836 kit.i_o(), kit.reset_memory(),
837 kit.frameptr(), kit.returnadr(),
838 // like a return but with exception input
839 ex_oop);
840 // bind to root
841 root()->add_req(exit);
842 record_for_igvn(exit);
843 initial_gvn()->transform_no_reclaim(exit);
844 }
846 //---------------------------do_exceptions-------------------------------------
847 // Process exceptions arising from the current bytecode.
848 // Send caught exceptions to the proper handler within this method.
849 // Unhandled exceptions feed into _exit.
850 void Parse::do_exceptions() {
851 if (!has_exceptions()) return;
853 if (failing()) {
854 // Pop them all off and throw them away.
855 while (pop_exception_state() != NULL) ;
856 return;
857 }
859 PreserveJVMState pjvms(this, false);
861 SafePointNode* ex_map;
862 while ((ex_map = pop_exception_state()) != NULL) {
863 if (!method()->has_exception_handlers()) {
864 // Common case: Transfer control outward.
865 // Doing it this early allows the exceptions to common up
866 // even between adjacent method calls.
867 throw_to_exit(ex_map);
868 } else {
869 // Have to look at the exception first.
870 assert(stopped(), "catch_inline_exceptions trashes the map");
871 catch_inline_exceptions(ex_map);
872 stop_and_kill_map(); // we used up this exception state; kill it
873 }
874 }
876 // We now return to our regularly scheduled program:
877 }
879 //---------------------------throw_to_exit-------------------------------------
880 // Merge the given map into an exception exit from this method.
881 // The exception exit will handle any unlocking of receiver.
882 // The ex_oop must be saved within the ex_map, unlike merge_exception.
883 void Parse::throw_to_exit(SafePointNode* ex_map) {
884 // Pop the JVMS to (a copy of) the caller.
885 GraphKit caller;
886 caller.set_map_clone(_caller->map());
887 caller.set_bci(_caller->bci());
888 caller.set_sp(_caller->sp());
889 // Copy out the standard machine state:
890 for (uint i = 0; i < TypeFunc::Parms; i++) {
891 caller.map()->set_req(i, ex_map->in(i));
892 }
893 // ...and the exception:
894 Node* ex_oop = saved_ex_oop(ex_map);
895 SafePointNode* caller_ex_map = caller.make_exception_state(ex_oop);
896 // Finally, collect the new exception state in my exits:
897 _exits.add_exception_state(caller_ex_map);
898 }
900 //------------------------------do_exits---------------------------------------
901 void Parse::do_exits() {
902 set_parse_bci(InvocationEntryBci);
904 // Now peephole on the return bits
905 Node* region = _exits.control();
906 _exits.set_control(gvn().transform(region));
908 Node* iophi = _exits.i_o();
909 _exits.set_i_o(gvn().transform(iophi));
911 // On PPC64, also add MemBarRelease for constructors which write
912 // volatile fields. As support_IRIW_for_not_multiple_copy_atomic_cpu
913 // is set on PPC64, no sync instruction is issued after volatile
914 // stores. We want to quarantee the same behaviour as on platforms
915 // with total store order, although this is not required by the Java
916 // memory model. So as with finals, we add a barrier here.
917 if (wrote_final() PPC64_ONLY(|| (wrote_volatile() && method()->is_initializer()))) {
918 // This method (which must be a constructor by the rules of Java)
919 // wrote a final. The effects of all initializations must be
920 // committed to memory before any code after the constructor
921 // publishes the reference to the newly constructor object.
922 // Rather than wait for the publication, we simply block the
923 // writes here. Rather than put a barrier on only those writes
924 // which are required to complete, we force all writes to complete.
925 //
926 // "All bets are off" unless the first publication occurs after a
927 // normal return from the constructor. We do not attempt to detect
928 // such unusual early publications. But no barrier is needed on
929 // exceptional returns, since they cannot publish normally.
930 //
931 _exits.insert_mem_bar(Op_MemBarRelease, alloc_with_final());
932 #ifndef PRODUCT
933 if (PrintOpto && (Verbose || WizardMode)) {
934 method()->print_name();
935 tty->print_cr(" writes finals and needs a memory barrier");
936 }
937 #endif
938 }
940 for (MergeMemStream mms(_exits.merged_memory()); mms.next_non_empty(); ) {
941 // transform each slice of the original memphi:
942 mms.set_memory(_gvn.transform(mms.memory()));
943 }
945 if (tf()->range()->cnt() > TypeFunc::Parms) {
946 const Type* ret_type = tf()->range()->field_at(TypeFunc::Parms);
947 Node* ret_phi = _gvn.transform( _exits.argument(0) );
948 assert(_exits.control()->is_top() || !_gvn.type(ret_phi)->empty(), "return value must be well defined");
949 _exits.push_node(ret_type->basic_type(), ret_phi);
950 }
952 // Note: Logic for creating and optimizing the ReturnNode is in Compile.
954 // Unlock along the exceptional paths.
955 // This is done late so that we can common up equivalent exceptions
956 // (e.g., null checks) arising from multiple points within this method.
957 // See GraphKit::add_exception_state, which performs the commoning.
958 bool do_synch = method()->is_synchronized() && GenerateSynchronizationCode;
960 // record exit from a method if compiled while Dtrace is turned on.
961 if (do_synch || C->env()->dtrace_method_probes()) {
962 // First move the exception list out of _exits:
963 GraphKit kit(_exits.transfer_exceptions_into_jvms());
964 SafePointNode* normal_map = kit.map(); // keep this guy safe
965 // Now re-collect the exceptions into _exits:
966 SafePointNode* ex_map;
967 while ((ex_map = kit.pop_exception_state()) != NULL) {
968 Node* ex_oop = kit.use_exception_state(ex_map);
969 // Force the exiting JVM state to have this method at InvocationEntryBci.
970 // The exiting JVM state is otherwise a copy of the calling JVMS.
971 JVMState* caller = kit.jvms();
972 JVMState* ex_jvms = caller->clone_shallow(C);
973 ex_jvms->set_map(kit.clone_map());
974 ex_jvms->map()->set_jvms(ex_jvms);
975 ex_jvms->set_bci( InvocationEntryBci);
976 kit.set_jvms(ex_jvms);
977 if (do_synch) {
978 // Add on the synchronized-method box/object combo
979 kit.map()->push_monitor(_synch_lock);
980 // Unlock!
981 kit.shared_unlock(_synch_lock->box_node(), _synch_lock->obj_node());
982 }
983 if (C->env()->dtrace_method_probes()) {
984 kit.make_dtrace_method_exit(method());
985 }
986 // Done with exception-path processing.
987 ex_map = kit.make_exception_state(ex_oop);
988 assert(ex_jvms->same_calls_as(ex_map->jvms()), "sanity");
989 // Pop the last vestige of this method:
990 ex_map->set_jvms(caller->clone_shallow(C));
991 ex_map->jvms()->set_map(ex_map);
992 _exits.push_exception_state(ex_map);
993 }
994 assert(_exits.map() == normal_map, "keep the same return state");
995 }
997 {
998 // Capture very early exceptions (receiver null checks) from caller JVMS
999 GraphKit caller(_caller);
1000 SafePointNode* ex_map;
1001 while ((ex_map = caller.pop_exception_state()) != NULL) {
1002 _exits.add_exception_state(ex_map);
1003 }
1004 }
1005 }
1007 //-----------------------------create_entry_map-------------------------------
1008 // Initialize our parser map to contain the types at method entry.
1009 // For OSR, the map contains a single RawPtr parameter.
1010 // Initial monitor locking for sync. methods is performed by do_method_entry.
1011 SafePointNode* Parse::create_entry_map() {
1012 // Check for really stupid bail-out cases.
1013 uint len = TypeFunc::Parms + method()->max_locals() + method()->max_stack();
1014 if (len >= 32760) {
1015 C->record_method_not_compilable_all_tiers("too many local variables");
1016 return NULL;
1017 }
1019 // If this is an inlined method, we may have to do a receiver null check.
1020 if (_caller->has_method() && is_normal_parse() && !method()->is_static()) {
1021 GraphKit kit(_caller);
1022 kit.null_check_receiver_before_call(method());
1023 _caller = kit.transfer_exceptions_into_jvms();
1024 if (kit.stopped()) {
1025 _exits.add_exception_states_from(_caller);
1026 _exits.set_jvms(_caller);
1027 return NULL;
1028 }
1029 }
1031 assert(method() != NULL, "parser must have a method");
1033 // Create an initial safepoint to hold JVM state during parsing
1034 JVMState* jvms = new (C) JVMState(method(), _caller->has_method() ? _caller : NULL);
1035 set_map(new (C) SafePointNode(len, jvms));
1036 jvms->set_map(map());
1037 record_for_igvn(map());
1038 assert(jvms->endoff() == len, "correct jvms sizing");
1040 SafePointNode* inmap = _caller->map();
1041 assert(inmap != NULL, "must have inmap");
1043 uint i;
1045 // Pass thru the predefined input parameters.
1046 for (i = 0; i < TypeFunc::Parms; i++) {
1047 map()->init_req(i, inmap->in(i));
1048 }
1050 if (depth() == 1) {
1051 assert(map()->memory()->Opcode() == Op_Parm, "");
1052 // Insert the memory aliasing node
1053 set_all_memory(reset_memory());
1054 }
1055 assert(merged_memory(), "");
1057 // Now add the locals which are initially bound to arguments:
1058 uint arg_size = tf()->domain()->cnt();
1059 ensure_stack(arg_size - TypeFunc::Parms); // OSR methods have funny args
1060 for (i = TypeFunc::Parms; i < arg_size; i++) {
1061 map()->init_req(i, inmap->argument(_caller, i - TypeFunc::Parms));
1062 }
1064 // Clear out the rest of the map (locals and stack)
1065 for (i = arg_size; i < len; i++) {
1066 map()->init_req(i, top());
1067 }
1069 SafePointNode* entry_map = stop();
1070 return entry_map;
1071 }
1073 //-----------------------------do_method_entry--------------------------------
1074 // Emit any code needed in the pseudo-block before BCI zero.
1075 // The main thing to do is lock the receiver of a synchronized method.
1076 void Parse::do_method_entry() {
1077 set_parse_bci(InvocationEntryBci); // Pseudo-BCP
1078 set_sp(0); // Java Stack Pointer
1080 NOT_PRODUCT( count_compiled_calls(true/*at_method_entry*/, false/*is_inline*/); )
1082 if (C->env()->dtrace_method_probes()) {
1083 make_dtrace_method_entry(method());
1084 }
1086 // If the method is synchronized, we need to construct a lock node, attach
1087 // it to the Start node, and pin it there.
1088 if (method()->is_synchronized()) {
1089 // Insert a FastLockNode right after the Start which takes as arguments
1090 // the current thread pointer, the "this" pointer & the address of the
1091 // stack slot pair used for the lock. The "this" pointer is a projection
1092 // off the start node, but the locking spot has to be constructed by
1093 // creating a ConLNode of 0, and boxing it with a BoxLockNode. The BoxLockNode
1094 // becomes the second argument to the FastLockNode call. The
1095 // FastLockNode becomes the new control parent to pin it to the start.
1097 // Setup Object Pointer
1098 Node *lock_obj = NULL;
1099 if(method()->is_static()) {
1100 ciInstance* mirror = _method->holder()->java_mirror();
1101 const TypeInstPtr *t_lock = TypeInstPtr::make(mirror);
1102 lock_obj = makecon(t_lock);
1103 } else { // Else pass the "this" pointer,
1104 lock_obj = local(0); // which is Parm0 from StartNode
1105 }
1106 // Clear out dead values from the debug info.
1107 kill_dead_locals();
1108 // Build the FastLockNode
1109 _synch_lock = shared_lock(lock_obj);
1110 }
1112 // Feed profiling data for parameters to the type system so it can
1113 // propagate it as speculative types
1114 record_profiled_parameters_for_speculation();
1116 if (depth() == 1) {
1117 increment_and_test_invocation_counter(Tier2CompileThreshold);
1118 }
1119 }
1121 //------------------------------init_blocks------------------------------------
1122 // Initialize our parser map to contain the types/monitors at method entry.
1123 void Parse::init_blocks() {
1124 // Create the blocks.
1125 _block_count = flow()->block_count();
1126 _blocks = NEW_RESOURCE_ARRAY(Block, _block_count);
1127 Copy::zero_to_bytes(_blocks, sizeof(Block)*_block_count);
1129 int rpo;
1131 // Initialize the structs.
1132 for (rpo = 0; rpo < block_count(); rpo++) {
1133 Block* block = rpo_at(rpo);
1134 block->init_node(this, rpo);
1135 }
1137 // Collect predecessor and successor information.
1138 for (rpo = 0; rpo < block_count(); rpo++) {
1139 Block* block = rpo_at(rpo);
1140 block->init_graph(this);
1141 }
1142 }
1144 //-------------------------------init_node-------------------------------------
1145 void Parse::Block::init_node(Parse* outer, int rpo) {
1146 _flow = outer->flow()->rpo_at(rpo);
1147 _pred_count = 0;
1148 _preds_parsed = 0;
1149 _count = 0;
1150 assert(pred_count() == 0 && preds_parsed() == 0, "sanity");
1151 assert(!(is_merged() || is_parsed() || is_handler() || has_merged_backedge()), "sanity");
1152 assert(_live_locals.size() == 0, "sanity");
1154 // entry point has additional predecessor
1155 if (flow()->is_start()) _pred_count++;
1156 assert(flow()->is_start() == (this == outer->start_block()), "");
1157 }
1159 //-------------------------------init_graph------------------------------------
1160 void Parse::Block::init_graph(Parse* outer) {
1161 // Create the successor list for this parser block.
1162 GrowableArray<ciTypeFlow::Block*>* tfs = flow()->successors();
1163 GrowableArray<ciTypeFlow::Block*>* tfe = flow()->exceptions();
1164 int ns = tfs->length();
1165 int ne = tfe->length();
1166 _num_successors = ns;
1167 _all_successors = ns+ne;
1168 _successors = (ns+ne == 0) ? NULL : NEW_RESOURCE_ARRAY(Block*, ns+ne);
1169 int p = 0;
1170 for (int i = 0; i < ns+ne; i++) {
1171 ciTypeFlow::Block* tf2 = (i < ns) ? tfs->at(i) : tfe->at(i-ns);
1172 Block* block2 = outer->rpo_at(tf2->rpo());
1173 _successors[i] = block2;
1175 // Accumulate pred info for the other block, too.
1176 if (i < ns) {
1177 block2->_pred_count++;
1178 } else {
1179 block2->_is_handler = true;
1180 }
1182 #ifdef ASSERT
1183 // A block's successors must be distinguishable by BCI.
1184 // That is, no bytecode is allowed to branch to two different
1185 // clones of the same code location.
1186 for (int j = 0; j < i; j++) {
1187 Block* block1 = _successors[j];
1188 if (block1 == block2) continue; // duplicates are OK
1189 assert(block1->start() != block2->start(), "successors have unique bcis");
1190 }
1191 #endif
1192 }
1194 // Note: We never call next_path_num along exception paths, so they
1195 // never get processed as "ready". Also, the input phis of exception
1196 // handlers get specially processed, so that
1197 }
1199 //---------------------------successor_for_bci---------------------------------
1200 Parse::Block* Parse::Block::successor_for_bci(int bci) {
1201 for (int i = 0; i < all_successors(); i++) {
1202 Block* block2 = successor_at(i);
1203 if (block2->start() == bci) return block2;
1204 }
1205 // We can actually reach here if ciTypeFlow traps out a block
1206 // due to an unloaded class, and concurrently with compilation the
1207 // class is then loaded, so that a later phase of the parser is
1208 // able to see more of the bytecode CFG. Or, the flow pass and
1209 // the parser can have a minor difference of opinion about executability
1210 // of bytecodes. For example, "obj.field = null" is executable even
1211 // if the field's type is an unloaded class; the flow pass used to
1212 // make a trap for such code.
1213 return NULL;
1214 }
1217 //-----------------------------stack_type_at-----------------------------------
1218 const Type* Parse::Block::stack_type_at(int i) const {
1219 return get_type(flow()->stack_type_at(i));
1220 }
1223 //-----------------------------local_type_at-----------------------------------
1224 const Type* Parse::Block::local_type_at(int i) const {
1225 // Make dead locals fall to bottom.
1226 if (_live_locals.size() == 0) {
1227 MethodLivenessResult live_locals = flow()->outer()->method()->liveness_at_bci(start());
1228 // This bitmap can be zero length if we saw a breakpoint.
1229 // In such cases, pretend they are all live.
1230 ((Block*)this)->_live_locals = live_locals;
1231 }
1232 if (_live_locals.size() > 0 && !_live_locals.at(i))
1233 return Type::BOTTOM;
1235 return get_type(flow()->local_type_at(i));
1236 }
1239 #ifndef PRODUCT
1241 //----------------------------name_for_bc--------------------------------------
1242 // helper method for BytecodeParseHistogram
1243 static const char* name_for_bc(int i) {
1244 return Bytecodes::is_defined(i) ? Bytecodes::name(Bytecodes::cast(i)) : "xxxunusedxxx";
1245 }
1247 //----------------------------BytecodeParseHistogram------------------------------------
1248 Parse::BytecodeParseHistogram::BytecodeParseHistogram(Parse *p, Compile *c) {
1249 _parser = p;
1250 _compiler = c;
1251 if( ! _initialized ) { _initialized = true; reset(); }
1252 }
1254 //----------------------------current_count------------------------------------
1255 int Parse::BytecodeParseHistogram::current_count(BPHType bph_type) {
1256 switch( bph_type ) {
1257 case BPH_transforms: { return _parser->gvn().made_progress(); }
1258 case BPH_values: { return _parser->gvn().made_new_values(); }
1259 default: { ShouldNotReachHere(); return 0; }
1260 }
1261 }
1263 //----------------------------initialized--------------------------------------
1264 bool Parse::BytecodeParseHistogram::initialized() { return _initialized; }
1266 //----------------------------reset--------------------------------------------
1267 void Parse::BytecodeParseHistogram::reset() {
1268 int i = Bytecodes::number_of_codes;
1269 while (i-- > 0) { _bytecodes_parsed[i] = 0; _nodes_constructed[i] = 0; _nodes_transformed[i] = 0; _new_values[i] = 0; }
1270 }
1272 //----------------------------set_initial_state--------------------------------
1273 // Record info when starting to parse one bytecode
1274 void Parse::BytecodeParseHistogram::set_initial_state( Bytecodes::Code bc ) {
1275 if( PrintParseStatistics && !_parser->is_osr_parse() ) {
1276 _initial_bytecode = bc;
1277 _initial_node_count = _compiler->unique();
1278 _initial_transforms = current_count(BPH_transforms);
1279 _initial_values = current_count(BPH_values);
1280 }
1281 }
1283 //----------------------------record_change--------------------------------
1284 // Record results of parsing one bytecode
1285 void Parse::BytecodeParseHistogram::record_change() {
1286 if( PrintParseStatistics && !_parser->is_osr_parse() ) {
1287 ++_bytecodes_parsed[_initial_bytecode];
1288 _nodes_constructed [_initial_bytecode] += (_compiler->unique() - _initial_node_count);
1289 _nodes_transformed [_initial_bytecode] += (current_count(BPH_transforms) - _initial_transforms);
1290 _new_values [_initial_bytecode] += (current_count(BPH_values) - _initial_values);
1291 }
1292 }
1295 //----------------------------print--------------------------------------------
1296 void Parse::BytecodeParseHistogram::print(float cutoff) {
1297 ResourceMark rm;
1298 // print profile
1299 int total = 0;
1300 int i = 0;
1301 for( i = 0; i < Bytecodes::number_of_codes; ++i ) { total += _bytecodes_parsed[i]; }
1302 int abs_sum = 0;
1303 tty->cr(); //0123456789012345678901234567890123456789012345678901234567890123456789
1304 tty->print_cr("Histogram of %d parsed bytecodes:", total);
1305 if( total == 0 ) { return; }
1306 tty->cr();
1307 tty->print_cr("absolute: count of compiled bytecodes of this type");
1308 tty->print_cr("relative: percentage contribution to compiled nodes");
1309 tty->print_cr("nodes : Average number of nodes constructed per bytecode");
1310 tty->print_cr("rnodes : Significance towards total nodes constructed, (nodes*relative)");
1311 tty->print_cr("transforms: Average amount of tranform progress per bytecode compiled");
1312 tty->print_cr("values : Average number of node values improved per bytecode");
1313 tty->print_cr("name : Bytecode name");
1314 tty->cr();
1315 tty->print_cr(" absolute relative nodes rnodes transforms values name");
1316 tty->print_cr("----------------------------------------------------------------------");
1317 while (--i > 0) {
1318 int abs = _bytecodes_parsed[i];
1319 float rel = abs * 100.0F / total;
1320 float nodes = _bytecodes_parsed[i] == 0 ? 0 : (1.0F * _nodes_constructed[i])/_bytecodes_parsed[i];
1321 float rnodes = _bytecodes_parsed[i] == 0 ? 0 : rel * nodes;
1322 float xforms = _bytecodes_parsed[i] == 0 ? 0 : (1.0F * _nodes_transformed[i])/_bytecodes_parsed[i];
1323 float values = _bytecodes_parsed[i] == 0 ? 0 : (1.0F * _new_values [i])/_bytecodes_parsed[i];
1324 if (cutoff <= rel) {
1325 tty->print_cr("%10d %7.2f%% %6.1f %6.2f %6.1f %6.1f %s", abs, rel, nodes, rnodes, xforms, values, name_for_bc(i));
1326 abs_sum += abs;
1327 }
1328 }
1329 tty->print_cr("----------------------------------------------------------------------");
1330 float rel_sum = abs_sum * 100.0F / total;
1331 tty->print_cr("%10d %7.2f%% (cutoff = %.2f%%)", abs_sum, rel_sum, cutoff);
1332 tty->print_cr("----------------------------------------------------------------------");
1333 tty->cr();
1334 }
1335 #endif
1337 //----------------------------load_state_from----------------------------------
1338 // Load block/map/sp. But not do not touch iter/bci.
1339 void Parse::load_state_from(Block* block) {
1340 set_block(block);
1341 // load the block's JVM state:
1342 set_map(block->start_map());
1343 set_sp( block->start_sp());
1344 }
1347 //-----------------------------record_state------------------------------------
1348 void Parse::Block::record_state(Parse* p) {
1349 assert(!is_merged(), "can only record state once, on 1st inflow");
1350 assert(start_sp() == p->sp(), "stack pointer must agree with ciTypeFlow");
1351 set_start_map(p->stop());
1352 }
1355 //------------------------------do_one_block-----------------------------------
1356 void Parse::do_one_block() {
1357 if (TraceOptoParse) {
1358 Block *b = block();
1359 int ns = b->num_successors();
1360 int nt = b->all_successors();
1362 tty->print("Parsing block #%d at bci [%d,%d), successors: ",
1363 block()->rpo(), block()->start(), block()->limit());
1364 for (int i = 0; i < nt; i++) {
1365 tty->print((( i < ns) ? " %d" : " %d(e)"), b->successor_at(i)->rpo());
1366 }
1367 if (b->is_loop_head()) tty->print(" lphd");
1368 tty->print_cr("");
1369 }
1371 assert(block()->is_merged(), "must be merged before being parsed");
1372 block()->mark_parsed();
1373 ++_blocks_parsed;
1375 // Set iterator to start of block.
1376 iter().reset_to_bci(block()->start());
1378 CompileLog* log = C->log();
1380 // Parse bytecodes
1381 while (!stopped() && !failing()) {
1382 iter().next();
1384 // Learn the current bci from the iterator:
1385 set_parse_bci(iter().cur_bci());
1387 if (bci() == block()->limit()) {
1388 // Do not walk into the next block until directed by do_all_blocks.
1389 merge(bci());
1390 break;
1391 }
1392 assert(bci() < block()->limit(), "bci still in block");
1394 if (log != NULL) {
1395 // Output an optional context marker, to help place actions
1396 // that occur during parsing of this BC. If there is no log
1397 // output until the next context string, this context string
1398 // will be silently ignored.
1399 log->set_context("bc code='%d' bci='%d'", (int)bc(), bci());
1400 }
1402 if (block()->has_trap_at(bci())) {
1403 // We must respect the flow pass's traps, because it will refuse
1404 // to produce successors for trapping blocks.
1405 int trap_index = block()->flow()->trap_index();
1406 assert(trap_index != 0, "trap index must be valid");
1407 uncommon_trap(trap_index);
1408 break;
1409 }
1411 NOT_PRODUCT( parse_histogram()->set_initial_state(bc()); );
1413 #ifdef ASSERT
1414 int pre_bc_sp = sp();
1415 int inputs, depth;
1416 bool have_se = !stopped() && compute_stack_effects(inputs, depth);
1417 assert(!have_se || pre_bc_sp >= inputs, err_msg_res("have enough stack to execute this BC: pre_bc_sp=%d, inputs=%d", pre_bc_sp, inputs));
1418 #endif //ASSERT
1420 do_one_bytecode();
1422 assert(!have_se || stopped() || failing() || (sp() - pre_bc_sp) == depth,
1423 err_msg_res("incorrect depth prediction: sp=%d, pre_bc_sp=%d, depth=%d", sp(), pre_bc_sp, depth));
1425 do_exceptions();
1427 NOT_PRODUCT( parse_histogram()->record_change(); );
1429 if (log != NULL)
1430 log->clear_context(); // skip marker if nothing was printed
1432 // Fall into next bytecode. Each bytecode normally has 1 sequential
1433 // successor which is typically made ready by visiting this bytecode.
1434 // If the successor has several predecessors, then it is a merge
1435 // point, starts a new basic block, and is handled like other basic blocks.
1436 }
1437 }
1440 //------------------------------merge------------------------------------------
1441 void Parse::set_parse_bci(int bci) {
1442 set_bci(bci);
1443 Node_Notes* nn = C->default_node_notes();
1444 if (nn == NULL) return;
1446 // Collect debug info for inlined calls unless -XX:-DebugInlinedCalls.
1447 if (!DebugInlinedCalls && depth() > 1) {
1448 return;
1449 }
1451 // Update the JVMS annotation, if present.
1452 JVMState* jvms = nn->jvms();
1453 if (jvms != NULL && jvms->bci() != bci) {
1454 // Update the JVMS.
1455 jvms = jvms->clone_shallow(C);
1456 jvms->set_bci(bci);
1457 nn->set_jvms(jvms);
1458 }
1459 }
1461 //------------------------------merge------------------------------------------
1462 // Merge the current mapping into the basic block starting at bci
1463 void Parse::merge(int target_bci) {
1464 Block* target = successor_for_bci(target_bci);
1465 if (target == NULL) { handle_missing_successor(target_bci); return; }
1466 assert(!target->is_ready(), "our arrival must be expected");
1467 int pnum = target->next_path_num();
1468 merge_common(target, pnum);
1469 }
1471 //-------------------------merge_new_path--------------------------------------
1472 // Merge the current mapping into the basic block, using a new path
1473 void Parse::merge_new_path(int target_bci) {
1474 Block* target = successor_for_bci(target_bci);
1475 if (target == NULL) { handle_missing_successor(target_bci); return; }
1476 assert(!target->is_ready(), "new path into frozen graph");
1477 int pnum = target->add_new_path();
1478 merge_common(target, pnum);
1479 }
1481 //-------------------------merge_exception-------------------------------------
1482 // Merge the current mapping into the basic block starting at bci
1483 // The ex_oop must be pushed on the stack, unlike throw_to_exit.
1484 void Parse::merge_exception(int target_bci) {
1485 assert(sp() == 1, "must have only the throw exception on the stack");
1486 Block* target = successor_for_bci(target_bci);
1487 if (target == NULL) { handle_missing_successor(target_bci); return; }
1488 assert(target->is_handler(), "exceptions are handled by special blocks");
1489 int pnum = target->add_new_path();
1490 merge_common(target, pnum);
1491 }
1493 //--------------------handle_missing_successor---------------------------------
1494 void Parse::handle_missing_successor(int target_bci) {
1495 #ifndef PRODUCT
1496 Block* b = block();
1497 int trap_bci = b->flow()->has_trap()? b->flow()->trap_bci(): -1;
1498 tty->print_cr("### Missing successor at bci:%d for block #%d (trap_bci:%d)", target_bci, b->rpo(), trap_bci);
1499 #endif
1500 ShouldNotReachHere();
1501 }
1503 //--------------------------merge_common---------------------------------------
1504 void Parse::merge_common(Parse::Block* target, int pnum) {
1505 if (TraceOptoParse) {
1506 tty->print("Merging state at block #%d bci:%d", target->rpo(), target->start());
1507 }
1509 // Zap extra stack slots to top
1510 assert(sp() == target->start_sp(), "");
1511 clean_stack(sp());
1513 if (!target->is_merged()) { // No prior mapping at this bci
1514 if (TraceOptoParse) { tty->print(" with empty state"); }
1516 // If this path is dead, do not bother capturing it as a merge.
1517 // It is "as if" we had 1 fewer predecessors from the beginning.
1518 if (stopped()) {
1519 if (TraceOptoParse) tty->print_cr(", but path is dead and doesn't count");
1520 return;
1521 }
1523 // Record that a new block has been merged.
1524 ++_blocks_merged;
1526 // Make a region if we know there are multiple or unpredictable inputs.
1527 // (Also, if this is a plain fall-through, we might see another region,
1528 // which must not be allowed into this block's map.)
1529 if (pnum > PhiNode::Input // Known multiple inputs.
1530 || target->is_handler() // These have unpredictable inputs.
1531 || target->is_loop_head() // Known multiple inputs
1532 || control()->is_Region()) { // We must hide this guy.
1534 int current_bci = bci();
1535 set_parse_bci(target->start()); // Set target bci
1536 if (target->is_SEL_head()) {
1537 DEBUG_ONLY( target->mark_merged_backedge(block()); )
1538 if (target->start() == 0) {
1539 // Add loop predicate for the special case when
1540 // there are backbranches to the method entry.
1541 add_predicate();
1542 }
1543 }
1544 // Add a Region to start the new basic block. Phis will be added
1545 // later lazily.
1546 int edges = target->pred_count();
1547 if (edges < pnum) edges = pnum; // might be a new path!
1548 RegionNode *r = new (C) RegionNode(edges+1);
1549 gvn().set_type(r, Type::CONTROL);
1550 record_for_igvn(r);
1551 // zap all inputs to NULL for debugging (done in Node(uint) constructor)
1552 // for (int j = 1; j < edges+1; j++) { r->init_req(j, NULL); }
1553 r->init_req(pnum, control());
1554 set_control(r);
1555 set_parse_bci(current_bci); // Restore bci
1556 }
1558 // Convert the existing Parser mapping into a mapping at this bci.
1559 store_state_to(target);
1560 assert(target->is_merged(), "do not come here twice");
1562 } else { // Prior mapping at this bci
1563 if (TraceOptoParse) { tty->print(" with previous state"); }
1564 #ifdef ASSERT
1565 if (target->is_SEL_head()) {
1566 target->mark_merged_backedge(block());
1567 }
1568 #endif
1569 // We must not manufacture more phis if the target is already parsed.
1570 bool nophi = target->is_parsed();
1572 SafePointNode* newin = map();// Hang on to incoming mapping
1573 Block* save_block = block(); // Hang on to incoming block;
1574 load_state_from(target); // Get prior mapping
1576 assert(newin->jvms()->locoff() == jvms()->locoff(), "JVMS layouts agree");
1577 assert(newin->jvms()->stkoff() == jvms()->stkoff(), "JVMS layouts agree");
1578 assert(newin->jvms()->monoff() == jvms()->monoff(), "JVMS layouts agree");
1579 assert(newin->jvms()->endoff() == jvms()->endoff(), "JVMS layouts agree");
1581 // Iterate over my current mapping and the old mapping.
1582 // Where different, insert Phi functions.
1583 // Use any existing Phi functions.
1584 assert(control()->is_Region(), "must be merging to a region");
1585 RegionNode* r = control()->as_Region();
1587 // Compute where to merge into
1588 // Merge incoming control path
1589 r->init_req(pnum, newin->control());
1591 if (pnum == 1) { // Last merge for this Region?
1592 if (!block()->flow()->is_irreducible_entry()) {
1593 Node* result = _gvn.transform_no_reclaim(r);
1594 if (r != result && TraceOptoParse) {
1595 tty->print_cr("Block #%d replace %d with %d", block()->rpo(), r->_idx, result->_idx);
1596 }
1597 }
1598 record_for_igvn(r);
1599 }
1601 // Update all the non-control inputs to map:
1602 assert(TypeFunc::Parms == newin->jvms()->locoff(), "parser map should contain only youngest jvms");
1603 bool check_elide_phi = target->is_SEL_backedge(save_block);
1604 for (uint j = 1; j < newin->req(); j++) {
1605 Node* m = map()->in(j); // Current state of target.
1606 Node* n = newin->in(j); // Incoming change to target state.
1607 PhiNode* phi;
1608 if (m->is_Phi() && m->as_Phi()->region() == r)
1609 phi = m->as_Phi();
1610 else
1611 phi = NULL;
1612 if (m != n) { // Different; must merge
1613 switch (j) {
1614 // Frame pointer and Return Address never changes
1615 case TypeFunc::FramePtr:// Drop m, use the original value
1616 case TypeFunc::ReturnAdr:
1617 break;
1618 case TypeFunc::Memory: // Merge inputs to the MergeMem node
1619 assert(phi == NULL, "the merge contains phis, not vice versa");
1620 merge_memory_edges(n->as_MergeMem(), pnum, nophi);
1621 continue;
1622 default: // All normal stuff
1623 if (phi == NULL) {
1624 const JVMState* jvms = map()->jvms();
1625 if (EliminateNestedLocks &&
1626 jvms->is_mon(j) && jvms->is_monitor_box(j)) {
1627 // BoxLock nodes are not commoning.
1628 // Use old BoxLock node as merged box.
1629 assert(newin->jvms()->is_monitor_box(j), "sanity");
1630 // This assert also tests that nodes are BoxLock.
1631 assert(BoxLockNode::same_slot(n, m), "sanity");
1632 C->gvn_replace_by(n, m);
1633 } else if (!check_elide_phi || !target->can_elide_SEL_phi(j)) {
1634 phi = ensure_phi(j, nophi);
1635 }
1636 }
1637 break;
1638 }
1639 }
1640 // At this point, n might be top if:
1641 // - there is no phi (because TypeFlow detected a conflict), or
1642 // - the corresponding control edges is top (a dead incoming path)
1643 // It is a bug if we create a phi which sees a garbage value on a live path.
1645 if (phi != NULL) {
1646 assert(n != top() || r->in(pnum) == top(), "live value must not be garbage");
1647 assert(phi->region() == r, "");
1648 phi->set_req(pnum, n); // Then add 'n' to the merge
1649 if (pnum == PhiNode::Input) {
1650 // Last merge for this Phi.
1651 // So far, Phis have had a reasonable type from ciTypeFlow.
1652 // Now _gvn will join that with the meet of current inputs.
1653 // BOTTOM is never permissible here, 'cause pessimistically
1654 // Phis of pointers cannot lose the basic pointer type.
1655 debug_only(const Type* bt1 = phi->bottom_type());
1656 assert(bt1 != Type::BOTTOM, "should not be building conflict phis");
1657 map()->set_req(j, _gvn.transform_no_reclaim(phi));
1658 debug_only(const Type* bt2 = phi->bottom_type());
1659 assert(bt2->higher_equal(bt1), "must be consistent with type-flow");
1660 record_for_igvn(phi);
1661 }
1662 }
1663 } // End of for all values to be merged
1665 if (pnum == PhiNode::Input &&
1666 !r->in(0)) { // The occasional useless Region
1667 assert(control() == r, "");
1668 set_control(r->nonnull_req());
1669 }
1671 // newin has been subsumed into the lazy merge, and is now dead.
1672 set_block(save_block);
1674 stop(); // done with this guy, for now
1675 }
1677 if (TraceOptoParse) {
1678 tty->print_cr(" on path %d", pnum);
1679 }
1681 // Done with this parser state.
1682 assert(stopped(), "");
1683 }
1686 //--------------------------merge_memory_edges---------------------------------
1687 void Parse::merge_memory_edges(MergeMemNode* n, int pnum, bool nophi) {
1688 // (nophi means we must not create phis, because we already parsed here)
1689 assert(n != NULL, "");
1690 // Merge the inputs to the MergeMems
1691 MergeMemNode* m = merged_memory();
1693 assert(control()->is_Region(), "must be merging to a region");
1694 RegionNode* r = control()->as_Region();
1696 PhiNode* base = NULL;
1697 MergeMemNode* remerge = NULL;
1698 for (MergeMemStream mms(m, n); mms.next_non_empty2(); ) {
1699 Node *p = mms.force_memory();
1700 Node *q = mms.memory2();
1701 if (mms.is_empty() && nophi) {
1702 // Trouble: No new splits allowed after a loop body is parsed.
1703 // Instead, wire the new split into a MergeMem on the backedge.
1704 // The optimizer will sort it out, slicing the phi.
1705 if (remerge == NULL) {
1706 assert(base != NULL, "");
1707 assert(base->in(0) != NULL, "should not be xformed away");
1708 remerge = MergeMemNode::make(C, base->in(pnum));
1709 gvn().set_type(remerge, Type::MEMORY);
1710 base->set_req(pnum, remerge);
1711 }
1712 remerge->set_memory_at(mms.alias_idx(), q);
1713 continue;
1714 }
1715 assert(!q->is_MergeMem(), "");
1716 PhiNode* phi;
1717 if (p != q) {
1718 phi = ensure_memory_phi(mms.alias_idx(), nophi);
1719 } else {
1720 if (p->is_Phi() && p->as_Phi()->region() == r)
1721 phi = p->as_Phi();
1722 else
1723 phi = NULL;
1724 }
1725 // Insert q into local phi
1726 if (phi != NULL) {
1727 assert(phi->region() == r, "");
1728 p = phi;
1729 phi->set_req(pnum, q);
1730 if (mms.at_base_memory()) {
1731 base = phi; // delay transforming it
1732 } else if (pnum == 1) {
1733 record_for_igvn(phi);
1734 p = _gvn.transform_no_reclaim(phi);
1735 }
1736 mms.set_memory(p);// store back through the iterator
1737 }
1738 }
1739 // Transform base last, in case we must fiddle with remerging.
1740 if (base != NULL && pnum == 1) {
1741 record_for_igvn(base);
1742 m->set_base_memory( _gvn.transform_no_reclaim(base) );
1743 }
1744 }
1747 //------------------------ensure_phis_everywhere-------------------------------
1748 void Parse::ensure_phis_everywhere() {
1749 ensure_phi(TypeFunc::I_O);
1751 // Ensure a phi on all currently known memories.
1752 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
1753 ensure_memory_phi(mms.alias_idx());
1754 debug_only(mms.set_memory()); // keep the iterator happy
1755 }
1757 // Note: This is our only chance to create phis for memory slices.
1758 // If we miss a slice that crops up later, it will have to be
1759 // merged into the base-memory phi that we are building here.
1760 // Later, the optimizer will comb out the knot, and build separate
1761 // phi-loops for each memory slice that matters.
1763 // Monitors must nest nicely and not get confused amongst themselves.
1764 // Phi-ify everything up to the monitors, though.
1765 uint monoff = map()->jvms()->monoff();
1766 uint nof_monitors = map()->jvms()->nof_monitors();
1768 assert(TypeFunc::Parms == map()->jvms()->locoff(), "parser map should contain only youngest jvms");
1769 bool check_elide_phi = block()->is_SEL_head();
1770 for (uint i = TypeFunc::Parms; i < monoff; i++) {
1771 if (!check_elide_phi || !block()->can_elide_SEL_phi(i)) {
1772 ensure_phi(i);
1773 }
1774 }
1776 // Even monitors need Phis, though they are well-structured.
1777 // This is true for OSR methods, and also for the rare cases where
1778 // a monitor object is the subject of a replace_in_map operation.
1779 // See bugs 4426707 and 5043395.
1780 for (uint m = 0; m < nof_monitors; m++) {
1781 ensure_phi(map()->jvms()->monitor_obj_offset(m));
1782 }
1783 }
1786 //-----------------------------add_new_path------------------------------------
1787 // Add a previously unaccounted predecessor to this block.
1788 int Parse::Block::add_new_path() {
1789 // If there is no map, return the lowest unused path number.
1790 if (!is_merged()) return pred_count()+1; // there will be a map shortly
1792 SafePointNode* map = start_map();
1793 if (!map->control()->is_Region())
1794 return pred_count()+1; // there may be a region some day
1795 RegionNode* r = map->control()->as_Region();
1797 // Add new path to the region.
1798 uint pnum = r->req();
1799 r->add_req(NULL);
1801 for (uint i = 1; i < map->req(); i++) {
1802 Node* n = map->in(i);
1803 if (i == TypeFunc::Memory) {
1804 // Ensure a phi on all currently known memories.
1805 for (MergeMemStream mms(n->as_MergeMem()); mms.next_non_empty(); ) {
1806 Node* phi = mms.memory();
1807 if (phi->is_Phi() && phi->as_Phi()->region() == r) {
1808 assert(phi->req() == pnum, "must be same size as region");
1809 phi->add_req(NULL);
1810 }
1811 }
1812 } else {
1813 if (n->is_Phi() && n->as_Phi()->region() == r) {
1814 assert(n->req() == pnum, "must be same size as region");
1815 n->add_req(NULL);
1816 }
1817 }
1818 }
1820 return pnum;
1821 }
1823 //------------------------------ensure_phi-------------------------------------
1824 // Turn the idx'th entry of the current map into a Phi
1825 PhiNode *Parse::ensure_phi(int idx, bool nocreate) {
1826 SafePointNode* map = this->map();
1827 Node* region = map->control();
1828 assert(region->is_Region(), "");
1830 Node* o = map->in(idx);
1831 assert(o != NULL, "");
1833 if (o == top()) return NULL; // TOP always merges into TOP
1835 if (o->is_Phi() && o->as_Phi()->region() == region) {
1836 return o->as_Phi();
1837 }
1839 // Now use a Phi here for merging
1840 assert(!nocreate, "Cannot build a phi for a block already parsed.");
1841 const JVMState* jvms = map->jvms();
1842 const Type* t;
1843 if (jvms->is_loc(idx)) {
1844 t = block()->local_type_at(idx - jvms->locoff());
1845 } else if (jvms->is_stk(idx)) {
1846 t = block()->stack_type_at(idx - jvms->stkoff());
1847 } else if (jvms->is_mon(idx)) {
1848 assert(!jvms->is_monitor_box(idx), "no phis for boxes");
1849 t = TypeInstPtr::BOTTOM; // this is sufficient for a lock object
1850 } else if ((uint)idx < TypeFunc::Parms) {
1851 t = o->bottom_type(); // Type::RETURN_ADDRESS or such-like.
1852 } else {
1853 assert(false, "no type information for this phi");
1854 }
1856 // If the type falls to bottom, then this must be a local that
1857 // is mixing ints and oops or some such. Forcing it to top
1858 // makes it go dead.
1859 if (t == Type::BOTTOM) {
1860 map->set_req(idx, top());
1861 return NULL;
1862 }
1864 // Do not create phis for top either.
1865 // A top on a non-null control flow must be an unused even after the.phi.
1866 if (t == Type::TOP || t == Type::HALF) {
1867 map->set_req(idx, top());
1868 return NULL;
1869 }
1871 PhiNode* phi = PhiNode::make(region, o, t);
1872 gvn().set_type(phi, t);
1873 if (C->do_escape_analysis()) record_for_igvn(phi);
1874 map->set_req(idx, phi);
1875 return phi;
1876 }
1878 //--------------------------ensure_memory_phi----------------------------------
1879 // Turn the idx'th slice of the current memory into a Phi
1880 PhiNode *Parse::ensure_memory_phi(int idx, bool nocreate) {
1881 MergeMemNode* mem = merged_memory();
1882 Node* region = control();
1883 assert(region->is_Region(), "");
1885 Node *o = (idx == Compile::AliasIdxBot)? mem->base_memory(): mem->memory_at(idx);
1886 assert(o != NULL && o != top(), "");
1888 PhiNode* phi;
1889 if (o->is_Phi() && o->as_Phi()->region() == region) {
1890 phi = o->as_Phi();
1891 if (phi == mem->base_memory() && idx >= Compile::AliasIdxRaw) {
1892 // clone the shared base memory phi to make a new memory split
1893 assert(!nocreate, "Cannot build a phi for a block already parsed.");
1894 const Type* t = phi->bottom_type();
1895 const TypePtr* adr_type = C->get_adr_type(idx);
1896 phi = phi->slice_memory(adr_type);
1897 gvn().set_type(phi, t);
1898 }
1899 return phi;
1900 }
1902 // Now use a Phi here for merging
1903 assert(!nocreate, "Cannot build a phi for a block already parsed.");
1904 const Type* t = o->bottom_type();
1905 const TypePtr* adr_type = C->get_adr_type(idx);
1906 phi = PhiNode::make(region, o, t, adr_type);
1907 gvn().set_type(phi, t);
1908 if (idx == Compile::AliasIdxBot)
1909 mem->set_base_memory(phi);
1910 else
1911 mem->set_memory_at(idx, phi);
1912 return phi;
1913 }
1915 //------------------------------call_register_finalizer-----------------------
1916 // Check the klass of the receiver and call register_finalizer if the
1917 // class need finalization.
1918 void Parse::call_register_finalizer() {
1919 Node* receiver = local(0);
1920 assert(receiver != NULL && receiver->bottom_type()->isa_instptr() != NULL,
1921 "must have non-null instance type");
1923 const TypeInstPtr *tinst = receiver->bottom_type()->isa_instptr();
1924 if (tinst != NULL && tinst->klass()->is_loaded() && !tinst->klass_is_exact()) {
1925 // The type isn't known exactly so see if CHA tells us anything.
1926 ciInstanceKlass* ik = tinst->klass()->as_instance_klass();
1927 if (!Dependencies::has_finalizable_subclass(ik)) {
1928 // No finalizable subclasses so skip the dynamic check.
1929 C->dependencies()->assert_has_no_finalizable_subclasses(ik);
1930 return;
1931 }
1932 }
1934 // Insert a dynamic test for whether the instance needs
1935 // finalization. In general this will fold up since the concrete
1936 // class is often visible so the access flags are constant.
1937 Node* klass_addr = basic_plus_adr( receiver, receiver, oopDesc::klass_offset_in_bytes() );
1938 Node* klass = _gvn.transform( LoadKlassNode::make(_gvn, immutable_memory(), klass_addr, TypeInstPtr::KLASS) );
1940 Node* access_flags_addr = basic_plus_adr(klass, klass, in_bytes(Klass::access_flags_offset()));
1941 Node* access_flags = make_load(NULL, access_flags_addr, TypeInt::INT, T_INT, MemNode::unordered);
1943 Node* mask = _gvn.transform(new (C) AndINode(access_flags, intcon(JVM_ACC_HAS_FINALIZER)));
1944 Node* check = _gvn.transform(new (C) CmpINode(mask, intcon(0)));
1945 Node* test = _gvn.transform(new (C) BoolNode(check, BoolTest::ne));
1947 IfNode* iff = create_and_map_if(control(), test, PROB_MAX, COUNT_UNKNOWN);
1949 RegionNode* result_rgn = new (C) RegionNode(3);
1950 record_for_igvn(result_rgn);
1952 Node *skip_register = _gvn.transform(new (C) IfFalseNode(iff));
1953 result_rgn->init_req(1, skip_register);
1955 Node *needs_register = _gvn.transform(new (C) IfTrueNode(iff));
1956 set_control(needs_register);
1957 if (stopped()) {
1958 // There is no slow path.
1959 result_rgn->init_req(2, top());
1960 } else {
1961 Node *call = make_runtime_call(RC_NO_LEAF,
1962 OptoRuntime::register_finalizer_Type(),
1963 OptoRuntime::register_finalizer_Java(),
1964 NULL, TypePtr::BOTTOM,
1965 receiver);
1966 make_slow_call_ex(call, env()->Throwable_klass(), true);
1968 Node* fast_io = call->in(TypeFunc::I_O);
1969 Node* fast_mem = call->in(TypeFunc::Memory);
1970 // These two phis are pre-filled with copies of of the fast IO and Memory
1971 Node* io_phi = PhiNode::make(result_rgn, fast_io, Type::ABIO);
1972 Node* mem_phi = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
1974 result_rgn->init_req(2, control());
1975 io_phi ->init_req(2, i_o());
1976 mem_phi ->init_req(2, reset_memory());
1978 set_all_memory( _gvn.transform(mem_phi) );
1979 set_i_o( _gvn.transform(io_phi) );
1980 }
1982 set_control( _gvn.transform(result_rgn) );
1983 }
1985 //------------------------------return_current---------------------------------
1986 // Append current _map to _exit_return
1987 void Parse::return_current(Node* value) {
1988 if (RegisterFinalizersAtInit &&
1989 method()->intrinsic_id() == vmIntrinsics::_Object_init) {
1990 call_register_finalizer();
1991 }
1993 // Do not set_parse_bci, so that return goo is credited to the return insn.
1994 set_bci(InvocationEntryBci);
1995 if (method()->is_synchronized() && GenerateSynchronizationCode) {
1996 shared_unlock(_synch_lock->box_node(), _synch_lock->obj_node());
1997 }
1998 if (C->env()->dtrace_method_probes()) {
1999 make_dtrace_method_exit(method());
2000 }
2001 SafePointNode* exit_return = _exits.map();
2002 exit_return->in( TypeFunc::Control )->add_req( control() );
2003 exit_return->in( TypeFunc::I_O )->add_req( i_o () );
2004 Node *mem = exit_return->in( TypeFunc::Memory );
2005 for (MergeMemStream mms(mem->as_MergeMem(), merged_memory()); mms.next_non_empty2(); ) {
2006 if (mms.is_empty()) {
2007 // get a copy of the base memory, and patch just this one input
2008 const TypePtr* adr_type = mms.adr_type(C);
2009 Node* phi = mms.force_memory()->as_Phi()->slice_memory(adr_type);
2010 assert(phi->as_Phi()->region() == mms.base_memory()->in(0), "");
2011 gvn().set_type_bottom(phi);
2012 phi->del_req(phi->req()-1); // prepare to re-patch
2013 mms.set_memory(phi);
2014 }
2015 mms.memory()->add_req(mms.memory2());
2016 }
2018 // frame pointer is always same, already captured
2019 if (value != NULL) {
2020 // If returning oops to an interface-return, there is a silent free
2021 // cast from oop to interface allowed by the Verifier. Make it explicit
2022 // here.
2023 Node* phi = _exits.argument(0);
2024 const TypeInstPtr *tr = phi->bottom_type()->isa_instptr();
2025 if( tr && tr->klass()->is_loaded() &&
2026 tr->klass()->is_interface() ) {
2027 const TypeInstPtr *tp = value->bottom_type()->isa_instptr();
2028 if (tp && tp->klass()->is_loaded() &&
2029 !tp->klass()->is_interface()) {
2030 // sharpen the type eagerly; this eases certain assert checking
2031 if (tp->higher_equal(TypeInstPtr::NOTNULL))
2032 tr = tr->join(TypeInstPtr::NOTNULL)->is_instptr();
2033 value = _gvn.transform(new (C) CheckCastPPNode(0,value,tr));
2034 }
2035 }
2036 phi->add_req(value);
2037 }
2039 stop_and_kill_map(); // This CFG path dies here
2040 }
2043 //------------------------------add_safepoint----------------------------------
2044 void Parse::add_safepoint() {
2045 // See if we can avoid this safepoint. No need for a SafePoint immediately
2046 // after a Call (except Leaf Call) or another SafePoint.
2047 Node *proj = control();
2048 bool add_poll_param = SafePointNode::needs_polling_address_input();
2049 uint parms = add_poll_param ? TypeFunc::Parms+1 : TypeFunc::Parms;
2050 if( proj->is_Proj() ) {
2051 Node *n0 = proj->in(0);
2052 if( n0->is_Catch() ) {
2053 n0 = n0->in(0)->in(0);
2054 assert( n0->is_Call(), "expect a call here" );
2055 }
2056 if( n0->is_Call() ) {
2057 if( n0->as_Call()->guaranteed_safepoint() )
2058 return;
2059 } else if( n0->is_SafePoint() && n0->req() >= parms ) {
2060 return;
2061 }
2062 }
2064 // Clear out dead values from the debug info.
2065 kill_dead_locals();
2067 // Clone the JVM State
2068 SafePointNode *sfpnt = new (C) SafePointNode(parms, NULL);
2070 // Capture memory state BEFORE a SafePoint. Since we can block at a
2071 // SafePoint we need our GC state to be safe; i.e. we need all our current
2072 // write barriers (card marks) to not float down after the SafePoint so we
2073 // must read raw memory. Likewise we need all oop stores to match the card
2074 // marks. If deopt can happen, we need ALL stores (we need the correct JVM
2075 // state on a deopt).
2077 // We do not need to WRITE the memory state after a SafePoint. The control
2078 // edge will keep card-marks and oop-stores from floating up from below a
2079 // SafePoint and our true dependency added here will keep them from floating
2080 // down below a SafePoint.
2082 // Clone the current memory state
2083 Node* mem = MergeMemNode::make(C, map()->memory());
2085 mem = _gvn.transform(mem);
2087 // Pass control through the safepoint
2088 sfpnt->init_req(TypeFunc::Control , control());
2089 // Fix edges normally used by a call
2090 sfpnt->init_req(TypeFunc::I_O , top() );
2091 sfpnt->init_req(TypeFunc::Memory , mem );
2092 sfpnt->init_req(TypeFunc::ReturnAdr, top() );
2093 sfpnt->init_req(TypeFunc::FramePtr , top() );
2095 // Create a node for the polling address
2096 if( add_poll_param ) {
2097 Node *polladr = ConPNode::make(C, (address)os::get_polling_page());
2098 sfpnt->init_req(TypeFunc::Parms+0, _gvn.transform(polladr));
2099 }
2101 // Fix up the JVM State edges
2102 add_safepoint_edges(sfpnt);
2103 Node *transformed_sfpnt = _gvn.transform(sfpnt);
2104 set_control(transformed_sfpnt);
2106 // Provide an edge from root to safepoint. This makes the safepoint
2107 // appear useful until the parse has completed.
2108 if( OptoRemoveUseless && transformed_sfpnt->is_SafePoint() ) {
2109 assert(C->root() != NULL, "Expect parse is still valid");
2110 C->root()->add_prec(transformed_sfpnt);
2111 }
2112 }
2114 #ifndef PRODUCT
2115 //------------------------show_parse_info--------------------------------------
2116 void Parse::show_parse_info() {
2117 InlineTree* ilt = NULL;
2118 if (C->ilt() != NULL) {
2119 JVMState* caller_jvms = is_osr_parse() ? caller()->caller() : caller();
2120 ilt = InlineTree::find_subtree_from_root(C->ilt(), caller_jvms, method());
2121 }
2122 if (PrintCompilation && Verbose) {
2123 if (depth() == 1) {
2124 if( ilt->count_inlines() ) {
2125 tty->print(" __inlined %d (%d bytes)", ilt->count_inlines(),
2126 ilt->count_inline_bcs());
2127 tty->cr();
2128 }
2129 } else {
2130 if (method()->is_synchronized()) tty->print("s");
2131 if (method()->has_exception_handlers()) tty->print("!");
2132 // Check this is not the final compiled version
2133 if (C->trap_can_recompile()) {
2134 tty->print("-");
2135 } else {
2136 tty->print(" ");
2137 }
2138 method()->print_short_name();
2139 if (is_osr_parse()) {
2140 tty->print(" @ %d", osr_bci());
2141 }
2142 tty->print(" (%d bytes)",method()->code_size());
2143 if (ilt->count_inlines()) {
2144 tty->print(" __inlined %d (%d bytes)", ilt->count_inlines(),
2145 ilt->count_inline_bcs());
2146 }
2147 tty->cr();
2148 }
2149 }
2150 if (PrintOpto && (depth() == 1 || PrintOptoInlining)) {
2151 // Print that we succeeded; suppress this message on the first osr parse.
2153 if (method()->is_synchronized()) tty->print("s");
2154 if (method()->has_exception_handlers()) tty->print("!");
2155 // Check this is not the final compiled version
2156 if (C->trap_can_recompile() && depth() == 1) {
2157 tty->print("-");
2158 } else {
2159 tty->print(" ");
2160 }
2161 if( depth() != 1 ) { tty->print(" "); } // missing compile count
2162 for (int i = 1; i < depth(); ++i) { tty->print(" "); }
2163 method()->print_short_name();
2164 if (is_osr_parse()) {
2165 tty->print(" @ %d", osr_bci());
2166 }
2167 if (ilt->caller_bci() != -1) {
2168 tty->print(" @ %d", ilt->caller_bci());
2169 }
2170 tty->print(" (%d bytes)",method()->code_size());
2171 if (ilt->count_inlines()) {
2172 tty->print(" __inlined %d (%d bytes)", ilt->count_inlines(),
2173 ilt->count_inline_bcs());
2174 }
2175 tty->cr();
2176 }
2177 }
2180 //------------------------------dump-------------------------------------------
2181 // Dump information associated with the bytecodes of current _method
2182 void Parse::dump() {
2183 if( method() != NULL ) {
2184 // Iterate over bytecodes
2185 ciBytecodeStream iter(method());
2186 for( Bytecodes::Code bc = iter.next(); bc != ciBytecodeStream::EOBC() ; bc = iter.next() ) {
2187 dump_bci( iter.cur_bci() );
2188 tty->cr();
2189 }
2190 }
2191 }
2193 // Dump information associated with a byte code index, 'bci'
2194 void Parse::dump_bci(int bci) {
2195 // Output info on merge-points, cloning, and within _jsr..._ret
2196 // NYI
2197 tty->print(" bci:%d", bci);
2198 }
2200 #endif