Wed, 07 Apr 2010 12:39:27 -0700
6940726: Use BIS instruction for allocation prefetch on Sparc
Summary: Use BIS instruction for allocation prefetch on Sparc
Reviewed-by: twisti
1 /*
2 * Copyright 1997-2010 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
25 #include "incls/_precompiled.incl"
26 #include "incls/_parse1.cpp.incl"
28 // Static array so we can figure out which bytecodes stop us from compiling
29 // the most. Some of the non-static variables are needed in bytecodeInfo.cpp
30 // and eventually should be encapsulated in a proper class (gri 8/18/98).
32 int nodes_created = 0;
33 int methods_parsed = 0;
34 int methods_seen = 0;
35 int blocks_parsed = 0;
36 int blocks_seen = 0;
38 int explicit_null_checks_inserted = 0;
39 int explicit_null_checks_elided = 0;
40 int all_null_checks_found = 0, implicit_null_checks = 0;
41 int implicit_null_throws = 0;
43 int reclaim_idx = 0;
44 int reclaim_in = 0;
45 int reclaim_node = 0;
47 #ifndef PRODUCT
48 bool Parse::BytecodeParseHistogram::_initialized = false;
49 uint Parse::BytecodeParseHistogram::_bytecodes_parsed [Bytecodes::number_of_codes];
50 uint Parse::BytecodeParseHistogram::_nodes_constructed[Bytecodes::number_of_codes];
51 uint Parse::BytecodeParseHistogram::_nodes_transformed[Bytecodes::number_of_codes];
52 uint Parse::BytecodeParseHistogram::_new_values [Bytecodes::number_of_codes];
53 #endif
55 //------------------------------print_statistics-------------------------------
56 #ifndef PRODUCT
57 void Parse::print_statistics() {
58 tty->print_cr("--- Compiler Statistics ---");
59 tty->print("Methods seen: %d Methods parsed: %d", methods_seen, methods_parsed);
60 tty->print(" Nodes created: %d", nodes_created);
61 tty->cr();
62 if (methods_seen != methods_parsed)
63 tty->print_cr("Reasons for parse failures (NOT cumulative):");
64 tty->print_cr("Blocks parsed: %d Blocks seen: %d", blocks_parsed, blocks_seen);
66 if( explicit_null_checks_inserted )
67 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);
68 if( all_null_checks_found )
69 tty->print_cr("%d made implicit (%2d%%)", implicit_null_checks,
70 (100*implicit_null_checks)/all_null_checks_found);
71 if( implicit_null_throws )
72 tty->print_cr("%d implicit null exceptions at runtime",
73 implicit_null_throws);
75 if( PrintParseStatistics && BytecodeParseHistogram::initialized() ) {
76 BytecodeParseHistogram::print();
77 }
78 }
79 #endif
81 //------------------------------ON STACK REPLACEMENT---------------------------
83 // Construct a node which can be used to get incoming state for
84 // on stack replacement.
85 Node *Parse::fetch_interpreter_state(int index,
86 BasicType bt,
87 Node *local_addrs,
88 Node *local_addrs_base) {
89 Node *mem = memory(Compile::AliasIdxRaw);
90 Node *adr = basic_plus_adr( local_addrs_base, local_addrs, -index*wordSize );
92 // Very similar to LoadNode::make, except we handle un-aligned longs and
93 // doubles on Sparc. Intel can handle them just fine directly.
94 Node *l;
95 switch( bt ) { // Signature is flattened
96 case T_INT: l = new (C, 3) LoadINode( 0, mem, adr, TypeRawPtr::BOTTOM ); break;
97 case T_FLOAT: l = new (C, 3) LoadFNode( 0, mem, adr, TypeRawPtr::BOTTOM ); break;
98 case T_ADDRESS: l = new (C, 3) LoadPNode( 0, mem, adr, TypeRawPtr::BOTTOM, TypeRawPtr::BOTTOM ); break;
99 case T_OBJECT: l = new (C, 3) LoadPNode( 0, mem, adr, TypeRawPtr::BOTTOM, TypeInstPtr::BOTTOM ); break;
100 case T_LONG:
101 case T_DOUBLE: {
102 // Since arguments are in reverse order, the argument address 'adr'
103 // refers to the back half of the long/double. Recompute adr.
104 adr = basic_plus_adr( local_addrs_base, local_addrs, -(index+1)*wordSize );
105 if( Matcher::misaligned_doubles_ok ) {
106 l = (bt == T_DOUBLE)
107 ? (Node*)new (C, 3) LoadDNode( 0, mem, adr, TypeRawPtr::BOTTOM )
108 : (Node*)new (C, 3) LoadLNode( 0, mem, adr, TypeRawPtr::BOTTOM );
109 } else {
110 l = (bt == T_DOUBLE)
111 ? (Node*)new (C, 3) LoadD_unalignedNode( 0, mem, adr, TypeRawPtr::BOTTOM )
112 : (Node*)new (C, 3) LoadL_unalignedNode( 0, mem, adr, TypeRawPtr::BOTTOM );
113 }
114 break;
115 }
116 default: ShouldNotReachHere();
117 }
118 return _gvn.transform(l);
119 }
121 // Helper routine to prevent the interpreter from handing
122 // unexpected typestate to an OSR method.
123 // The Node l is a value newly dug out of the interpreter frame.
124 // The type is the type predicted by ciTypeFlow. Note that it is
125 // not a general type, but can only come from Type::get_typeflow_type.
126 // The safepoint is a map which will feed an uncommon trap.
127 Node* Parse::check_interpreter_type(Node* l, const Type* type,
128 SafePointNode* &bad_type_exit) {
130 const TypeOopPtr* tp = type->isa_oopptr();
132 // TypeFlow may assert null-ness if a type appears unloaded.
133 if (type == TypePtr::NULL_PTR ||
134 (tp != NULL && !tp->klass()->is_loaded())) {
135 // Value must be null, not a real oop.
136 Node* chk = _gvn.transform( new (C, 3) CmpPNode(l, null()) );
137 Node* tst = _gvn.transform( new (C, 2) BoolNode(chk, BoolTest::eq) );
138 IfNode* iff = create_and_map_if(control(), tst, PROB_MAX, COUNT_UNKNOWN);
139 set_control(_gvn.transform( new (C, 1) IfTrueNode(iff) ));
140 Node* bad_type = _gvn.transform( new (C, 1) IfFalseNode(iff) );
141 bad_type_exit->control()->add_req(bad_type);
142 l = null();
143 }
145 // Typeflow can also cut off paths from the CFG, based on
146 // types which appear unloaded, or call sites which appear unlinked.
147 // When paths are cut off, values at later merge points can rise
148 // toward more specific classes. Make sure these specific classes
149 // are still in effect.
150 if (tp != NULL && tp->klass() != C->env()->Object_klass()) {
151 // TypeFlow asserted a specific object type. Value must have that type.
152 Node* bad_type_ctrl = NULL;
153 l = gen_checkcast(l, makecon(TypeKlassPtr::make(tp->klass())), &bad_type_ctrl);
154 bad_type_exit->control()->add_req(bad_type_ctrl);
155 }
157 BasicType bt_l = _gvn.type(l)->basic_type();
158 BasicType bt_t = type->basic_type();
159 assert(_gvn.type(l)->higher_equal(type), "must constrain OSR typestate");
160 return l;
161 }
163 // Helper routine which sets up elements of the initial parser map when
164 // performing a parse for on stack replacement. Add values into map.
165 // The only parameter contains the address of a interpreter arguments.
166 void Parse::load_interpreter_state(Node* osr_buf) {
167 int index;
168 int max_locals = jvms()->loc_size();
169 int max_stack = jvms()->stk_size();
172 // Mismatch between method and jvms can occur since map briefly held
173 // an OSR entry state (which takes up one RawPtr word).
174 assert(max_locals == method()->max_locals(), "sanity");
175 assert(max_stack >= method()->max_stack(), "sanity");
176 assert((int)jvms()->endoff() == TypeFunc::Parms + max_locals + max_stack, "sanity");
177 assert((int)jvms()->endoff() == (int)map()->req(), "sanity");
179 // Find the start block.
180 Block* osr_block = start_block();
181 assert(osr_block->start() == osr_bci(), "sanity");
183 // Set initial BCI.
184 set_parse_bci(osr_block->start());
186 // Set initial stack depth.
187 set_sp(osr_block->start_sp());
189 // Check bailouts. We currently do not perform on stack replacement
190 // of loops in catch blocks or loops which branch with a non-empty stack.
191 if (sp() != 0) {
192 C->record_method_not_compilable("OSR starts with non-empty stack");
193 return;
194 }
195 // Do not OSR inside finally clauses:
196 if (osr_block->has_trap_at(osr_block->start())) {
197 C->record_method_not_compilable("OSR starts with an immediate trap");
198 return;
199 }
201 // Commute monitors from interpreter frame to compiler frame.
202 assert(jvms()->monitor_depth() == 0, "should be no active locks at beginning of osr");
203 int mcnt = osr_block->flow()->monitor_count();
204 Node *monitors_addr = basic_plus_adr(osr_buf, osr_buf, (max_locals+mcnt*2-1)*wordSize);
205 for (index = 0; index < mcnt; index++) {
206 // Make a BoxLockNode for the monitor.
207 Node *box = _gvn.transform(new (C, 1) BoxLockNode(next_monitor()));
210 // Displaced headers and locked objects are interleaved in the
211 // temp OSR buffer. We only copy the locked objects out here.
212 // Fetch the locked object from the OSR temp buffer and copy to our fastlock node.
213 Node *lock_object = fetch_interpreter_state(index*2, T_OBJECT, monitors_addr, osr_buf);
214 // Try and copy the displaced header to the BoxNode
215 Node *displaced_hdr = fetch_interpreter_state((index*2) + 1, T_ADDRESS, monitors_addr, osr_buf);
218 store_to_memory(control(), box, displaced_hdr, T_ADDRESS, Compile::AliasIdxRaw);
220 // Build a bogus FastLockNode (no code will be generated) and push the
221 // monitor into our debug info.
222 const FastLockNode *flock = _gvn.transform(new (C, 3) FastLockNode( 0, lock_object, box ))->as_FastLock();
223 map()->push_monitor(flock);
225 // If the lock is our method synchronization lock, tuck it away in
226 // _sync_lock for return and rethrow exit paths.
227 if (index == 0 && method()->is_synchronized()) {
228 _synch_lock = flock;
229 }
230 }
232 // Use the raw liveness computation to make sure that unexpected
233 // values don't propagate into the OSR frame.
234 MethodLivenessResult live_locals = method()->liveness_at_bci(osr_bci());
235 if (!live_locals.is_valid()) {
236 // Degenerate or breakpointed method.
237 C->record_method_not_compilable("OSR in empty or breakpointed method");
238 return;
239 }
241 // Extract the needed locals from the interpreter frame.
242 Node *locals_addr = basic_plus_adr(osr_buf, osr_buf, (max_locals-1)*wordSize);
244 // find all the locals that the interpreter thinks contain live oops
245 const BitMap live_oops = method()->live_local_oops_at_bci(osr_bci());
246 for (index = 0; index < max_locals; index++) {
248 if (!live_locals.at(index)) {
249 continue;
250 }
252 const Type *type = osr_block->local_type_at(index);
254 if (type->isa_oopptr() != NULL) {
256 // 6403625: Verify that the interpreter oopMap thinks that the oop is live
257 // else we might load a stale oop if the MethodLiveness disagrees with the
258 // result of the interpreter. If the interpreter says it is dead we agree
259 // by making the value go to top.
260 //
262 if (!live_oops.at(index)) {
263 if (C->log() != NULL) {
264 C->log()->elem("OSR_mismatch local_index='%d'",index);
265 }
266 set_local(index, null());
267 // and ignore it for the loads
268 continue;
269 }
270 }
272 // Filter out TOP, HALF, and BOTTOM. (Cf. ensure_phi.)
273 if (type == Type::TOP || type == Type::HALF) {
274 continue;
275 }
276 // If the type falls to bottom, then this must be a local that
277 // is mixing ints and oops or some such. Forcing it to top
278 // makes it go dead.
279 if (type == Type::BOTTOM) {
280 continue;
281 }
282 // Construct code to access the appropriate local.
283 Node *value = fetch_interpreter_state(index, type->basic_type(), locals_addr, osr_buf);
284 set_local(index, value);
285 }
287 // Extract the needed stack entries from the interpreter frame.
288 for (index = 0; index < sp(); index++) {
289 const Type *type = osr_block->stack_type_at(index);
290 if (type != Type::TOP) {
291 // Currently the compiler bails out when attempting to on stack replace
292 // at a bci with a non-empty stack. We should not reach here.
293 ShouldNotReachHere();
294 }
295 }
297 // End the OSR migration
298 make_runtime_call(RC_LEAF, OptoRuntime::osr_end_Type(),
299 CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_end),
300 "OSR_migration_end", TypeRawPtr::BOTTOM,
301 osr_buf);
303 // Now that the interpreter state is loaded, make sure it will match
304 // at execution time what the compiler is expecting now:
305 SafePointNode* bad_type_exit = clone_map();
306 bad_type_exit->set_control(new (C, 1) RegionNode(1));
308 assert(osr_block->flow()->jsrs()->size() == 0, "should be no jsrs live at osr point");
309 for (index = 0; index < max_locals; index++) {
310 if (stopped()) break;
311 Node* l = local(index);
312 if (l->is_top()) continue; // nothing here
313 const Type *type = osr_block->local_type_at(index);
314 if (type->isa_oopptr() != NULL) {
315 if (!live_oops.at(index)) {
316 // skip type check for dead oops
317 continue;
318 }
319 }
320 if (osr_block->flow()->local_type_at(index)->is_return_address()) {
321 // In our current system it's illegal for jsr addresses to be
322 // live into an OSR entry point because the compiler performs
323 // inlining of jsrs. ciTypeFlow has a bailout that detect this
324 // case and aborts the compile if addresses are live into an OSR
325 // entry point. Because of that we can assume that any address
326 // locals at the OSR entry point are dead. Method liveness
327 // isn't precise enought to figure out that they are dead in all
328 // cases so simply skip checking address locals all
329 // together. Any type check is guaranteed to fail since the
330 // interpreter type is the result of a load which might have any
331 // value and the expected type is a constant.
332 continue;
333 }
334 set_local(index, check_interpreter_type(l, type, bad_type_exit));
335 }
337 for (index = 0; index < sp(); index++) {
338 if (stopped()) break;
339 Node* l = stack(index);
340 if (l->is_top()) continue; // nothing here
341 const Type *type = osr_block->stack_type_at(index);
342 set_stack(index, check_interpreter_type(l, type, bad_type_exit));
343 }
345 if (bad_type_exit->control()->req() > 1) {
346 // Build an uncommon trap here, if any inputs can be unexpected.
347 bad_type_exit->set_control(_gvn.transform( bad_type_exit->control() ));
348 record_for_igvn(bad_type_exit->control());
349 SafePointNode* types_are_good = map();
350 set_map(bad_type_exit);
351 // The unexpected type happens because a new edge is active
352 // in the CFG, which typeflow had previously ignored.
353 // E.g., Object x = coldAtFirst() && notReached()? "str": new Integer(123).
354 // This x will be typed as Integer if notReached is not yet linked.
355 uncommon_trap(Deoptimization::Reason_unreached,
356 Deoptimization::Action_reinterpret);
357 set_map(types_are_good);
358 }
359 }
361 //------------------------------Parse------------------------------------------
362 // Main parser constructor.
363 Parse::Parse(JVMState* caller, ciMethod* parse_method, float expected_uses)
364 : _exits(caller)
365 {
366 // Init some variables
367 _caller = caller;
368 _method = parse_method;
369 _expected_uses = expected_uses;
370 _depth = 1 + (caller->has_method() ? caller->depth() : 0);
371 _wrote_final = false;
372 _entry_bci = InvocationEntryBci;
373 _tf = NULL;
374 _block = NULL;
375 debug_only(_block_count = -1);
376 debug_only(_blocks = (Block*)-1);
377 #ifndef PRODUCT
378 if (PrintCompilation || PrintOpto) {
379 // Make sure I have an inline tree, so I can print messages about it.
380 JVMState* ilt_caller = is_osr_parse() ? caller->caller() : caller;
381 InlineTree::find_subtree_from_root(C->ilt(), ilt_caller, parse_method, true);
382 }
383 _max_switch_depth = 0;
384 _est_switch_depth = 0;
385 #endif
387 _tf = TypeFunc::make(method());
388 _iter.reset_to_method(method());
389 _flow = method()->get_flow_analysis();
390 if (_flow->failing()) {
391 C->record_method_not_compilable_all_tiers(_flow->failure_reason());
392 }
394 #ifndef PRODUCT
395 if (_flow->has_irreducible_entry()) {
396 C->set_parsed_irreducible_loop(true);
397 }
398 #endif
400 if (_expected_uses <= 0) {
401 _prof_factor = 1;
402 } else {
403 float prof_total = parse_method->interpreter_invocation_count();
404 if (prof_total <= _expected_uses) {
405 _prof_factor = 1;
406 } else {
407 _prof_factor = _expected_uses / prof_total;
408 }
409 }
411 CompileLog* log = C->log();
412 if (log != NULL) {
413 log->begin_head("parse method='%d' uses='%g'",
414 log->identify(parse_method), expected_uses);
415 if (depth() == 1 && C->is_osr_compilation()) {
416 log->print(" osr_bci='%d'", C->entry_bci());
417 }
418 log->stamp();
419 log->end_head();
420 }
422 // Accumulate deoptimization counts.
423 // (The range_check and store_check counts are checked elsewhere.)
424 ciMethodData* md = method()->method_data();
425 for (uint reason = 0; reason < md->trap_reason_limit(); reason++) {
426 uint md_count = md->trap_count(reason);
427 if (md_count != 0) {
428 if (md_count == md->trap_count_limit())
429 md_count += md->overflow_trap_count();
430 uint total_count = C->trap_count(reason);
431 uint old_count = total_count;
432 total_count += md_count;
433 // Saturate the add if it overflows.
434 if (total_count < old_count || total_count < md_count)
435 total_count = (uint)-1;
436 C->set_trap_count(reason, total_count);
437 if (log != NULL)
438 log->elem("observe trap='%s' count='%d' total='%d'",
439 Deoptimization::trap_reason_name(reason),
440 md_count, total_count);
441 }
442 }
443 // Accumulate total sum of decompilations, also.
444 C->set_decompile_count(C->decompile_count() + md->decompile_count());
446 _count_invocations = C->do_count_invocations();
447 _method_data_update = C->do_method_data_update();
449 if (log != NULL && method()->has_exception_handlers()) {
450 log->elem("observe that='has_exception_handlers'");
451 }
453 assert(method()->can_be_compiled(), "Can not parse this method, cutout earlier");
454 assert(method()->has_balanced_monitors(), "Can not parse unbalanced monitors, cutout earlier");
456 // Always register dependence if JVMTI is enabled, because
457 // either breakpoint setting or hotswapping of methods may
458 // cause deoptimization.
459 if (C->env()->jvmti_can_hotswap_or_post_breakpoint()) {
460 C->dependencies()->assert_evol_method(method());
461 }
463 methods_seen++;
465 // Do some special top-level things.
466 if (depth() == 1 && C->is_osr_compilation()) {
467 _entry_bci = C->entry_bci();
468 _flow = method()->get_osr_flow_analysis(osr_bci());
469 if (_flow->failing()) {
470 C->record_method_not_compilable(_flow->failure_reason());
471 #ifndef PRODUCT
472 if (PrintOpto && (Verbose || WizardMode)) {
473 tty->print_cr("OSR @%d type flow bailout: %s", _entry_bci, _flow->failure_reason());
474 if (Verbose) {
475 method()->print_oop();
476 method()->print_codes();
477 _flow->print();
478 }
479 }
480 #endif
481 }
482 _tf = C->tf(); // the OSR entry type is different
483 }
485 #ifdef ASSERT
486 if (depth() == 1) {
487 assert(C->is_osr_compilation() == this->is_osr_parse(), "OSR in sync");
488 if (C->tf() != tf()) {
489 MutexLockerEx ml(Compile_lock, Mutex::_no_safepoint_check_flag);
490 assert(C->env()->system_dictionary_modification_counter_changed(),
491 "Must invalidate if TypeFuncs differ");
492 }
493 } else {
494 assert(!this->is_osr_parse(), "no recursive OSR");
495 }
496 #endif
498 methods_parsed++;
499 #ifndef PRODUCT
500 // add method size here to guarantee that inlined methods are added too
501 if (TimeCompiler)
502 _total_bytes_compiled += method()->code_size();
504 show_parse_info();
505 #endif
507 if (failing()) {
508 if (log) log->done("parse");
509 return;
510 }
512 gvn().set_type(root(), root()->bottom_type());
513 gvn().transform(top());
515 // Import the results of the ciTypeFlow.
516 init_blocks();
518 // Merge point for all normal exits
519 build_exits();
521 // Setup the initial JVM state map.
522 SafePointNode* entry_map = create_entry_map();
524 // Check for bailouts during map initialization
525 if (failing() || entry_map == NULL) {
526 if (log) log->done("parse");
527 return;
528 }
530 Node_Notes* caller_nn = C->default_node_notes();
531 // Collect debug info for inlined calls unless -XX:-DebugInlinedCalls.
532 if (DebugInlinedCalls || depth() == 1) {
533 C->set_default_node_notes(make_node_notes(caller_nn));
534 }
536 if (is_osr_parse()) {
537 Node* osr_buf = entry_map->in(TypeFunc::Parms+0);
538 entry_map->set_req(TypeFunc::Parms+0, top());
539 set_map(entry_map);
540 load_interpreter_state(osr_buf);
541 } else {
542 set_map(entry_map);
543 do_method_entry();
544 }
546 // Check for bailouts during method entry.
547 if (failing()) {
548 if (log) log->done("parse");
549 C->set_default_node_notes(caller_nn);
550 return;
551 }
553 entry_map = map(); // capture any changes performed by method setup code
554 assert(jvms()->endoff() == map()->req(), "map matches JVMS layout");
556 // We begin parsing as if we have just encountered a jump to the
557 // method entry.
558 Block* entry_block = start_block();
559 assert(entry_block->start() == (is_osr_parse() ? osr_bci() : 0), "");
560 set_map_clone(entry_map);
561 merge_common(entry_block, entry_block->next_path_num());
563 #ifndef PRODUCT
564 BytecodeParseHistogram *parse_histogram_obj = new (C->env()->arena()) BytecodeParseHistogram(this, C);
565 set_parse_histogram( parse_histogram_obj );
566 #endif
568 // Parse all the basic blocks.
569 do_all_blocks();
571 C->set_default_node_notes(caller_nn);
573 // Check for bailouts during conversion to graph
574 if (failing()) {
575 if (log) log->done("parse");
576 return;
577 }
579 // Fix up all exiting control flow.
580 set_map(entry_map);
581 do_exits();
583 if (log) log->done("parse nodes='%d' memory='%d'",
584 C->unique(), C->node_arena()->used());
585 }
587 //---------------------------do_all_blocks-------------------------------------
588 void Parse::do_all_blocks() {
589 bool has_irreducible = flow()->has_irreducible_entry();
591 // Walk over all blocks in Reverse Post-Order.
592 while (true) {
593 bool progress = false;
594 for (int rpo = 0; rpo < block_count(); rpo++) {
595 Block* block = rpo_at(rpo);
597 if (block->is_parsed()) continue;
599 if (!block->is_merged()) {
600 // Dead block, no state reaches this block
601 continue;
602 }
604 // Prepare to parse this block.
605 load_state_from(block);
607 if (stopped()) {
608 // Block is dead.
609 continue;
610 }
612 blocks_parsed++;
614 progress = true;
615 if (block->is_loop_head() || block->is_handler() || has_irreducible && !block->is_ready()) {
616 // Not all preds have been parsed. We must build phis everywhere.
617 // (Note that dead locals do not get phis built, ever.)
618 ensure_phis_everywhere();
620 // Leave behind an undisturbed copy of the map, for future merges.
621 set_map(clone_map());
622 }
624 if (control()->is_Region() && !block->is_loop_head() && !has_irreducible && !block->is_handler()) {
625 // In the absence of irreducible loops, the Region and Phis
626 // associated with a merge that doesn't involve a backedge can
627 // be simplified now since the RPO parsing order guarantees
628 // that any path which was supposed to reach here has already
629 // been parsed or must be dead.
630 Node* c = control();
631 Node* result = _gvn.transform_no_reclaim(control());
632 if (c != result && TraceOptoParse) {
633 tty->print_cr("Block #%d replace %d with %d", block->rpo(), c->_idx, result->_idx);
634 }
635 if (result != top()) {
636 record_for_igvn(result);
637 }
638 }
640 // Parse the block.
641 do_one_block();
643 // Check for bailouts.
644 if (failing()) return;
645 }
647 // with irreducible loops multiple passes might be necessary to parse everything
648 if (!has_irreducible || !progress) {
649 break;
650 }
651 }
653 blocks_seen += block_count();
655 #ifndef PRODUCT
656 // Make sure there are no half-processed blocks remaining.
657 // Every remaining unprocessed block is dead and may be ignored now.
658 for (int rpo = 0; rpo < block_count(); rpo++) {
659 Block* block = rpo_at(rpo);
660 if (!block->is_parsed()) {
661 if (TraceOptoParse) {
662 tty->print_cr("Skipped dead block %d at bci:%d", rpo, block->start());
663 }
664 assert(!block->is_merged(), "no half-processed blocks");
665 }
666 }
667 #endif
668 }
670 //-------------------------------build_exits----------------------------------
671 // Build normal and exceptional exit merge points.
672 void Parse::build_exits() {
673 // make a clone of caller to prevent sharing of side-effects
674 _exits.set_map(_exits.clone_map());
675 _exits.clean_stack(_exits.sp());
676 _exits.sync_jvms();
678 RegionNode* region = new (C, 1) RegionNode(1);
679 record_for_igvn(region);
680 gvn().set_type_bottom(region);
681 _exits.set_control(region);
683 // Note: iophi and memphi are not transformed until do_exits.
684 Node* iophi = new (C, region->req()) PhiNode(region, Type::ABIO);
685 Node* memphi = new (C, region->req()) PhiNode(region, Type::MEMORY, TypePtr::BOTTOM);
686 _exits.set_i_o(iophi);
687 _exits.set_all_memory(memphi);
689 // Add a return value to the exit state. (Do not push it yet.)
690 if (tf()->range()->cnt() > TypeFunc::Parms) {
691 const Type* ret_type = tf()->range()->field_at(TypeFunc::Parms);
692 // Don't "bind" an unloaded return klass to the ret_phi. If the klass
693 // becomes loaded during the subsequent parsing, the loaded and unloaded
694 // types will not join when we transform and push in do_exits().
695 const TypeOopPtr* ret_oop_type = ret_type->isa_oopptr();
696 if (ret_oop_type && !ret_oop_type->klass()->is_loaded()) {
697 ret_type = TypeOopPtr::BOTTOM;
698 }
699 int ret_size = type2size[ret_type->basic_type()];
700 Node* ret_phi = new (C, region->req()) PhiNode(region, ret_type);
701 _exits.ensure_stack(ret_size);
702 assert((int)(tf()->range()->cnt() - TypeFunc::Parms) == ret_size, "good tf range");
703 assert(method()->return_type()->size() == ret_size, "tf agrees w/ method");
704 _exits.set_argument(0, ret_phi); // here is where the parser finds it
705 // Note: ret_phi is not yet pushed, until do_exits.
706 }
707 }
710 //----------------------------build_start_state-------------------------------
711 // Construct a state which contains only the incoming arguments from an
712 // unknown caller. The method & bci will be NULL & InvocationEntryBci.
713 JVMState* Compile::build_start_state(StartNode* start, const TypeFunc* tf) {
714 int arg_size = tf->domain()->cnt();
715 int max_size = MAX2(arg_size, (int)tf->range()->cnt());
716 JVMState* jvms = new (this) JVMState(max_size - TypeFunc::Parms);
717 SafePointNode* map = new (this, max_size) SafePointNode(max_size, NULL);
718 record_for_igvn(map);
719 assert(arg_size == TypeFunc::Parms + (is_osr_compilation() ? 1 : method()->arg_size()), "correct arg_size");
720 Node_Notes* old_nn = default_node_notes();
721 if (old_nn != NULL && has_method()) {
722 Node_Notes* entry_nn = old_nn->clone(this);
723 JVMState* entry_jvms = new(this) JVMState(method(), old_nn->jvms());
724 entry_jvms->set_offsets(0);
725 entry_jvms->set_bci(entry_bci());
726 entry_nn->set_jvms(entry_jvms);
727 set_default_node_notes(entry_nn);
728 }
729 uint i;
730 for (i = 0; i < (uint)arg_size; i++) {
731 Node* parm = initial_gvn()->transform(new (this, 1) ParmNode(start, i));
732 map->init_req(i, parm);
733 // Record all these guys for later GVN.
734 record_for_igvn(parm);
735 }
736 for (; i < map->req(); i++) {
737 map->init_req(i, top());
738 }
739 assert(jvms->argoff() == TypeFunc::Parms, "parser gets arguments here");
740 set_default_node_notes(old_nn);
741 map->set_jvms(jvms);
742 jvms->set_map(map);
743 return jvms;
744 }
746 //-----------------------------make_node_notes---------------------------------
747 Node_Notes* Parse::make_node_notes(Node_Notes* caller_nn) {
748 if (caller_nn == NULL) return NULL;
749 Node_Notes* nn = caller_nn->clone(C);
750 JVMState* caller_jvms = nn->jvms();
751 JVMState* jvms = new (C) JVMState(method(), caller_jvms);
752 jvms->set_offsets(0);
753 jvms->set_bci(_entry_bci);
754 nn->set_jvms(jvms);
755 return nn;
756 }
759 //--------------------------return_values--------------------------------------
760 void Compile::return_values(JVMState* jvms) {
761 GraphKit kit(jvms);
762 Node* ret = new (this, TypeFunc::Parms) ReturnNode(TypeFunc::Parms,
763 kit.control(),
764 kit.i_o(),
765 kit.reset_memory(),
766 kit.frameptr(),
767 kit.returnadr());
768 // Add zero or 1 return values
769 int ret_size = tf()->range()->cnt() - TypeFunc::Parms;
770 if (ret_size > 0) {
771 kit.inc_sp(-ret_size); // pop the return value(s)
772 kit.sync_jvms();
773 ret->add_req(kit.argument(0));
774 // Note: The second dummy edge is not needed by a ReturnNode.
775 }
776 // bind it to root
777 root()->add_req(ret);
778 record_for_igvn(ret);
779 initial_gvn()->transform_no_reclaim(ret);
780 }
782 //------------------------rethrow_exceptions-----------------------------------
783 // Bind all exception states in the list into a single RethrowNode.
784 void Compile::rethrow_exceptions(JVMState* jvms) {
785 GraphKit kit(jvms);
786 if (!kit.has_exceptions()) return; // nothing to generate
787 // Load my combined exception state into the kit, with all phis transformed:
788 SafePointNode* ex_map = kit.combine_and_pop_all_exception_states();
789 Node* ex_oop = kit.use_exception_state(ex_map);
790 RethrowNode* exit = new (this, TypeFunc::Parms + 1) RethrowNode(kit.control(),
791 kit.i_o(), kit.reset_memory(),
792 kit.frameptr(), kit.returnadr(),
793 // like a return but with exception input
794 ex_oop);
795 // bind to root
796 root()->add_req(exit);
797 record_for_igvn(exit);
798 initial_gvn()->transform_no_reclaim(exit);
799 }
801 //---------------------------do_exceptions-------------------------------------
802 // Process exceptions arising from the current bytecode.
803 // Send caught exceptions to the proper handler within this method.
804 // Unhandled exceptions feed into _exit.
805 void Parse::do_exceptions() {
806 if (!has_exceptions()) return;
808 if (failing()) {
809 // Pop them all off and throw them away.
810 while (pop_exception_state() != NULL) ;
811 return;
812 }
814 PreserveJVMState pjvms(this, false);
816 SafePointNode* ex_map;
817 while ((ex_map = pop_exception_state()) != NULL) {
818 if (!method()->has_exception_handlers()) {
819 // Common case: Transfer control outward.
820 // Doing it this early allows the exceptions to common up
821 // even between adjacent method calls.
822 throw_to_exit(ex_map);
823 } else {
824 // Have to look at the exception first.
825 assert(stopped(), "catch_inline_exceptions trashes the map");
826 catch_inline_exceptions(ex_map);
827 stop_and_kill_map(); // we used up this exception state; kill it
828 }
829 }
831 // We now return to our regularly scheduled program:
832 }
834 //---------------------------throw_to_exit-------------------------------------
835 // Merge the given map into an exception exit from this method.
836 // The exception exit will handle any unlocking of receiver.
837 // The ex_oop must be saved within the ex_map, unlike merge_exception.
838 void Parse::throw_to_exit(SafePointNode* ex_map) {
839 // Pop the JVMS to (a copy of) the caller.
840 GraphKit caller;
841 caller.set_map_clone(_caller->map());
842 caller.set_bci(_caller->bci());
843 caller.set_sp(_caller->sp());
844 // Copy out the standard machine state:
845 for (uint i = 0; i < TypeFunc::Parms; i++) {
846 caller.map()->set_req(i, ex_map->in(i));
847 }
848 // ...and the exception:
849 Node* ex_oop = saved_ex_oop(ex_map);
850 SafePointNode* caller_ex_map = caller.make_exception_state(ex_oop);
851 // Finally, collect the new exception state in my exits:
852 _exits.add_exception_state(caller_ex_map);
853 }
855 //------------------------------do_exits---------------------------------------
856 void Parse::do_exits() {
857 set_parse_bci(InvocationEntryBci);
859 // Now peephole on the return bits
860 Node* region = _exits.control();
861 _exits.set_control(gvn().transform(region));
863 Node* iophi = _exits.i_o();
864 _exits.set_i_o(gvn().transform(iophi));
866 if (wrote_final()) {
867 // This method (which must be a constructor by the rules of Java)
868 // wrote a final. The effects of all initializations must be
869 // committed to memory before any code after the constructor
870 // publishes the reference to the newly constructor object.
871 // Rather than wait for the publication, we simply block the
872 // writes here. Rather than put a barrier on only those writes
873 // which are required to complete, we force all writes to complete.
874 //
875 // "All bets are off" unless the first publication occurs after a
876 // normal return from the constructor. We do not attempt to detect
877 // such unusual early publications. But no barrier is needed on
878 // exceptional returns, since they cannot publish normally.
879 //
880 _exits.insert_mem_bar(Op_MemBarRelease);
881 #ifndef PRODUCT
882 if (PrintOpto && (Verbose || WizardMode)) {
883 method()->print_name();
884 tty->print_cr(" writes finals and needs a memory barrier");
885 }
886 #endif
887 }
889 for (MergeMemStream mms(_exits.merged_memory()); mms.next_non_empty(); ) {
890 // transform each slice of the original memphi:
891 mms.set_memory(_gvn.transform(mms.memory()));
892 }
894 if (tf()->range()->cnt() > TypeFunc::Parms) {
895 const Type* ret_type = tf()->range()->field_at(TypeFunc::Parms);
896 Node* ret_phi = _gvn.transform( _exits.argument(0) );
897 assert(_exits.control()->is_top() || !_gvn.type(ret_phi)->empty(), "return value must be well defined");
898 _exits.push_node(ret_type->basic_type(), ret_phi);
899 }
901 // Note: Logic for creating and optimizing the ReturnNode is in Compile.
903 // Unlock along the exceptional paths.
904 // This is done late so that we can common up equivalent exceptions
905 // (e.g., null checks) arising from multiple points within this method.
906 // See GraphKit::add_exception_state, which performs the commoning.
907 bool do_synch = method()->is_synchronized() && GenerateSynchronizationCode;
909 // record exit from a method if compiled while Dtrace is turned on.
910 if (do_synch || C->env()->dtrace_method_probes()) {
911 // First move the exception list out of _exits:
912 GraphKit kit(_exits.transfer_exceptions_into_jvms());
913 SafePointNode* normal_map = kit.map(); // keep this guy safe
914 // Now re-collect the exceptions into _exits:
915 SafePointNode* ex_map;
916 while ((ex_map = kit.pop_exception_state()) != NULL) {
917 Node* ex_oop = kit.use_exception_state(ex_map);
918 // Force the exiting JVM state to have this method at InvocationEntryBci.
919 // The exiting JVM state is otherwise a copy of the calling JVMS.
920 JVMState* caller = kit.jvms();
921 JVMState* ex_jvms = caller->clone_shallow(C);
922 ex_jvms->set_map(kit.clone_map());
923 ex_jvms->map()->set_jvms(ex_jvms);
924 ex_jvms->set_bci( InvocationEntryBci);
925 kit.set_jvms(ex_jvms);
926 if (do_synch) {
927 // Add on the synchronized-method box/object combo
928 kit.map()->push_monitor(_synch_lock);
929 // Unlock!
930 kit.shared_unlock(_synch_lock->box_node(), _synch_lock->obj_node());
931 }
932 if (C->env()->dtrace_method_probes()) {
933 kit.make_dtrace_method_exit(method());
934 }
935 // Done with exception-path processing.
936 ex_map = kit.make_exception_state(ex_oop);
937 assert(ex_jvms->same_calls_as(ex_map->jvms()), "sanity");
938 // Pop the last vestige of this method:
939 ex_map->set_jvms(caller->clone_shallow(C));
940 ex_map->jvms()->set_map(ex_map);
941 _exits.push_exception_state(ex_map);
942 }
943 assert(_exits.map() == normal_map, "keep the same return state");
944 }
946 {
947 // Capture very early exceptions (receiver null checks) from caller JVMS
948 GraphKit caller(_caller);
949 SafePointNode* ex_map;
950 while ((ex_map = caller.pop_exception_state()) != NULL) {
951 _exits.add_exception_state(ex_map);
952 }
953 }
954 }
956 //-----------------------------create_entry_map-------------------------------
957 // Initialize our parser map to contain the types at method entry.
958 // For OSR, the map contains a single RawPtr parameter.
959 // Initial monitor locking for sync. methods is performed by do_method_entry.
960 SafePointNode* Parse::create_entry_map() {
961 // Check for really stupid bail-out cases.
962 uint len = TypeFunc::Parms + method()->max_locals() + method()->max_stack();
963 if (len >= 32760) {
964 C->record_method_not_compilable_all_tiers("too many local variables");
965 return NULL;
966 }
968 // If this is an inlined method, we may have to do a receiver null check.
969 if (_caller->has_method() && is_normal_parse() && !method()->is_static()) {
970 GraphKit kit(_caller);
971 kit.null_check_receiver(method());
972 _caller = kit.transfer_exceptions_into_jvms();
973 if (kit.stopped()) {
974 _exits.add_exception_states_from(_caller);
975 _exits.set_jvms(_caller);
976 return NULL;
977 }
978 }
980 assert(method() != NULL, "parser must have a method");
982 // Create an initial safepoint to hold JVM state during parsing
983 JVMState* jvms = new (C) JVMState(method(), _caller->has_method() ? _caller : NULL);
984 set_map(new (C, len) SafePointNode(len, jvms));
985 jvms->set_map(map());
986 record_for_igvn(map());
987 assert(jvms->endoff() == len, "correct jvms sizing");
989 SafePointNode* inmap = _caller->map();
990 assert(inmap != NULL, "must have inmap");
992 uint i;
994 // Pass thru the predefined input parameters.
995 for (i = 0; i < TypeFunc::Parms; i++) {
996 map()->init_req(i, inmap->in(i));
997 }
999 if (depth() == 1) {
1000 assert(map()->memory()->Opcode() == Op_Parm, "");
1001 // Insert the memory aliasing node
1002 set_all_memory(reset_memory());
1003 }
1004 assert(merged_memory(), "");
1006 // Now add the locals which are initially bound to arguments:
1007 uint arg_size = tf()->domain()->cnt();
1008 ensure_stack(arg_size - TypeFunc::Parms); // OSR methods have funny args
1009 for (i = TypeFunc::Parms; i < arg_size; i++) {
1010 map()->init_req(i, inmap->argument(_caller, i - TypeFunc::Parms));
1011 }
1013 // Clear out the rest of the map (locals and stack)
1014 for (i = arg_size; i < len; i++) {
1015 map()->init_req(i, top());
1016 }
1018 SafePointNode* entry_map = stop();
1019 return entry_map;
1020 }
1022 //-----------------------------do_method_entry--------------------------------
1023 // Emit any code needed in the pseudo-block before BCI zero.
1024 // The main thing to do is lock the receiver of a synchronized method.
1025 void Parse::do_method_entry() {
1026 set_parse_bci(InvocationEntryBci); // Pseudo-BCP
1027 set_sp(0); // Java Stack Pointer
1029 NOT_PRODUCT( count_compiled_calls(true/*at_method_entry*/, false/*is_inline*/); )
1031 if (C->env()->dtrace_method_probes()) {
1032 make_dtrace_method_entry(method());
1033 }
1035 // If the method is synchronized, we need to construct a lock node, attach
1036 // it to the Start node, and pin it there.
1037 if (method()->is_synchronized()) {
1038 // Insert a FastLockNode right after the Start which takes as arguments
1039 // the current thread pointer, the "this" pointer & the address of the
1040 // stack slot pair used for the lock. The "this" pointer is a projection
1041 // off the start node, but the locking spot has to be constructed by
1042 // creating a ConLNode of 0, and boxing it with a BoxLockNode. The BoxLockNode
1043 // becomes the second argument to the FastLockNode call. The
1044 // FastLockNode becomes the new control parent to pin it to the start.
1046 // Setup Object Pointer
1047 Node *lock_obj = NULL;
1048 if(method()->is_static()) {
1049 ciInstance* mirror = _method->holder()->java_mirror();
1050 const TypeInstPtr *t_lock = TypeInstPtr::make(mirror);
1051 lock_obj = makecon(t_lock);
1052 } else { // Else pass the "this" pointer,
1053 lock_obj = local(0); // which is Parm0 from StartNode
1054 }
1055 // Clear out dead values from the debug info.
1056 kill_dead_locals();
1057 // Build the FastLockNode
1058 _synch_lock = shared_lock(lock_obj);
1059 }
1061 if (depth() == 1) {
1062 increment_and_test_invocation_counter(Tier2CompileThreshold);
1063 }
1064 }
1066 //------------------------------init_blocks------------------------------------
1067 // Initialize our parser map to contain the types/monitors at method entry.
1068 void Parse::init_blocks() {
1069 // Create the blocks.
1070 _block_count = flow()->block_count();
1071 _blocks = NEW_RESOURCE_ARRAY(Block, _block_count);
1072 Copy::zero_to_bytes(_blocks, sizeof(Block)*_block_count);
1074 int rpo;
1076 // Initialize the structs.
1077 for (rpo = 0; rpo < block_count(); rpo++) {
1078 Block* block = rpo_at(rpo);
1079 block->init_node(this, rpo);
1080 }
1082 // Collect predecessor and successor information.
1083 for (rpo = 0; rpo < block_count(); rpo++) {
1084 Block* block = rpo_at(rpo);
1085 block->init_graph(this);
1086 }
1087 }
1089 //-------------------------------init_node-------------------------------------
1090 void Parse::Block::init_node(Parse* outer, int rpo) {
1091 _flow = outer->flow()->rpo_at(rpo);
1092 _pred_count = 0;
1093 _preds_parsed = 0;
1094 _count = 0;
1095 assert(pred_count() == 0 && preds_parsed() == 0, "sanity");
1096 assert(!(is_merged() || is_parsed() || is_handler()), "sanity");
1097 assert(_live_locals.size() == 0, "sanity");
1099 // entry point has additional predecessor
1100 if (flow()->is_start()) _pred_count++;
1101 assert(flow()->is_start() == (this == outer->start_block()), "");
1102 }
1104 //-------------------------------init_graph------------------------------------
1105 void Parse::Block::init_graph(Parse* outer) {
1106 // Create the successor list for this parser block.
1107 GrowableArray<ciTypeFlow::Block*>* tfs = flow()->successors();
1108 GrowableArray<ciTypeFlow::Block*>* tfe = flow()->exceptions();
1109 int ns = tfs->length();
1110 int ne = tfe->length();
1111 _num_successors = ns;
1112 _all_successors = ns+ne;
1113 _successors = (ns+ne == 0) ? NULL : NEW_RESOURCE_ARRAY(Block*, ns+ne);
1114 int p = 0;
1115 for (int i = 0; i < ns+ne; i++) {
1116 ciTypeFlow::Block* tf2 = (i < ns) ? tfs->at(i) : tfe->at(i-ns);
1117 Block* block2 = outer->rpo_at(tf2->rpo());
1118 _successors[i] = block2;
1120 // Accumulate pred info for the other block, too.
1121 if (i < ns) {
1122 block2->_pred_count++;
1123 } else {
1124 block2->_is_handler = true;
1125 }
1127 #ifdef ASSERT
1128 // A block's successors must be distinguishable by BCI.
1129 // That is, no bytecode is allowed to branch to two different
1130 // clones of the same code location.
1131 for (int j = 0; j < i; j++) {
1132 Block* block1 = _successors[j];
1133 if (block1 == block2) continue; // duplicates are OK
1134 assert(block1->start() != block2->start(), "successors have unique bcis");
1135 }
1136 #endif
1137 }
1139 // Note: We never call next_path_num along exception paths, so they
1140 // never get processed as "ready". Also, the input phis of exception
1141 // handlers get specially processed, so that
1142 }
1144 //---------------------------successor_for_bci---------------------------------
1145 Parse::Block* Parse::Block::successor_for_bci(int bci) {
1146 for (int i = 0; i < all_successors(); i++) {
1147 Block* block2 = successor_at(i);
1148 if (block2->start() == bci) return block2;
1149 }
1150 // We can actually reach here if ciTypeFlow traps out a block
1151 // due to an unloaded class, and concurrently with compilation the
1152 // class is then loaded, so that a later phase of the parser is
1153 // able to see more of the bytecode CFG. Or, the flow pass and
1154 // the parser can have a minor difference of opinion about executability
1155 // of bytecodes. For example, "obj.field = null" is executable even
1156 // if the field's type is an unloaded class; the flow pass used to
1157 // make a trap for such code.
1158 return NULL;
1159 }
1162 //-----------------------------stack_type_at-----------------------------------
1163 const Type* Parse::Block::stack_type_at(int i) const {
1164 return get_type(flow()->stack_type_at(i));
1165 }
1168 //-----------------------------local_type_at-----------------------------------
1169 const Type* Parse::Block::local_type_at(int i) const {
1170 // Make dead locals fall to bottom.
1171 if (_live_locals.size() == 0) {
1172 MethodLivenessResult live_locals = flow()->outer()->method()->liveness_at_bci(start());
1173 // This bitmap can be zero length if we saw a breakpoint.
1174 // In such cases, pretend they are all live.
1175 ((Block*)this)->_live_locals = live_locals;
1176 }
1177 if (_live_locals.size() > 0 && !_live_locals.at(i))
1178 return Type::BOTTOM;
1180 return get_type(flow()->local_type_at(i));
1181 }
1184 #ifndef PRODUCT
1186 //----------------------------name_for_bc--------------------------------------
1187 // helper method for BytecodeParseHistogram
1188 static const char* name_for_bc(int i) {
1189 return Bytecodes::is_defined(i) ? Bytecodes::name(Bytecodes::cast(i)) : "xxxunusedxxx";
1190 }
1192 //----------------------------BytecodeParseHistogram------------------------------------
1193 Parse::BytecodeParseHistogram::BytecodeParseHistogram(Parse *p, Compile *c) {
1194 _parser = p;
1195 _compiler = c;
1196 if( ! _initialized ) { _initialized = true; reset(); }
1197 }
1199 //----------------------------current_count------------------------------------
1200 int Parse::BytecodeParseHistogram::current_count(BPHType bph_type) {
1201 switch( bph_type ) {
1202 case BPH_transforms: { return _parser->gvn().made_progress(); }
1203 case BPH_values: { return _parser->gvn().made_new_values(); }
1204 default: { ShouldNotReachHere(); return 0; }
1205 }
1206 }
1208 //----------------------------initialized--------------------------------------
1209 bool Parse::BytecodeParseHistogram::initialized() { return _initialized; }
1211 //----------------------------reset--------------------------------------------
1212 void Parse::BytecodeParseHistogram::reset() {
1213 int i = Bytecodes::number_of_codes;
1214 while (i-- > 0) { _bytecodes_parsed[i] = 0; _nodes_constructed[i] = 0; _nodes_transformed[i] = 0; _new_values[i] = 0; }
1215 }
1217 //----------------------------set_initial_state--------------------------------
1218 // Record info when starting to parse one bytecode
1219 void Parse::BytecodeParseHistogram::set_initial_state( Bytecodes::Code bc ) {
1220 if( PrintParseStatistics && !_parser->is_osr_parse() ) {
1221 _initial_bytecode = bc;
1222 _initial_node_count = _compiler->unique();
1223 _initial_transforms = current_count(BPH_transforms);
1224 _initial_values = current_count(BPH_values);
1225 }
1226 }
1228 //----------------------------record_change--------------------------------
1229 // Record results of parsing one bytecode
1230 void Parse::BytecodeParseHistogram::record_change() {
1231 if( PrintParseStatistics && !_parser->is_osr_parse() ) {
1232 ++_bytecodes_parsed[_initial_bytecode];
1233 _nodes_constructed [_initial_bytecode] += (_compiler->unique() - _initial_node_count);
1234 _nodes_transformed [_initial_bytecode] += (current_count(BPH_transforms) - _initial_transforms);
1235 _new_values [_initial_bytecode] += (current_count(BPH_values) - _initial_values);
1236 }
1237 }
1240 //----------------------------print--------------------------------------------
1241 void Parse::BytecodeParseHistogram::print(float cutoff) {
1242 ResourceMark rm;
1243 // print profile
1244 int total = 0;
1245 int i = 0;
1246 for( i = 0; i < Bytecodes::number_of_codes; ++i ) { total += _bytecodes_parsed[i]; }
1247 int abs_sum = 0;
1248 tty->cr(); //0123456789012345678901234567890123456789012345678901234567890123456789
1249 tty->print_cr("Histogram of %d parsed bytecodes:", total);
1250 if( total == 0 ) { return; }
1251 tty->cr();
1252 tty->print_cr("absolute: count of compiled bytecodes of this type");
1253 tty->print_cr("relative: percentage contribution to compiled nodes");
1254 tty->print_cr("nodes : Average number of nodes constructed per bytecode");
1255 tty->print_cr("rnodes : Significance towards total nodes constructed, (nodes*relative)");
1256 tty->print_cr("transforms: Average amount of tranform progress per bytecode compiled");
1257 tty->print_cr("values : Average number of node values improved per bytecode");
1258 tty->print_cr("name : Bytecode name");
1259 tty->cr();
1260 tty->print_cr(" absolute relative nodes rnodes transforms values name");
1261 tty->print_cr("----------------------------------------------------------------------");
1262 while (--i > 0) {
1263 int abs = _bytecodes_parsed[i];
1264 float rel = abs * 100.0F / total;
1265 float nodes = _bytecodes_parsed[i] == 0 ? 0 : (1.0F * _nodes_constructed[i])/_bytecodes_parsed[i];
1266 float rnodes = _bytecodes_parsed[i] == 0 ? 0 : rel * nodes;
1267 float xforms = _bytecodes_parsed[i] == 0 ? 0 : (1.0F * _nodes_transformed[i])/_bytecodes_parsed[i];
1268 float values = _bytecodes_parsed[i] == 0 ? 0 : (1.0F * _new_values [i])/_bytecodes_parsed[i];
1269 if (cutoff <= rel) {
1270 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));
1271 abs_sum += abs;
1272 }
1273 }
1274 tty->print_cr("----------------------------------------------------------------------");
1275 float rel_sum = abs_sum * 100.0F / total;
1276 tty->print_cr("%10d %7.2f%% (cutoff = %.2f%%)", abs_sum, rel_sum, cutoff);
1277 tty->print_cr("----------------------------------------------------------------------");
1278 tty->cr();
1279 }
1280 #endif
1282 //----------------------------load_state_from----------------------------------
1283 // Load block/map/sp. But not do not touch iter/bci.
1284 void Parse::load_state_from(Block* block) {
1285 set_block(block);
1286 // load the block's JVM state:
1287 set_map(block->start_map());
1288 set_sp( block->start_sp());
1289 }
1292 //-----------------------------record_state------------------------------------
1293 void Parse::Block::record_state(Parse* p) {
1294 assert(!is_merged(), "can only record state once, on 1st inflow");
1295 assert(start_sp() == p->sp(), "stack pointer must agree with ciTypeFlow");
1296 set_start_map(p->stop());
1297 }
1300 //------------------------------do_one_block-----------------------------------
1301 void Parse::do_one_block() {
1302 if (TraceOptoParse) {
1303 Block *b = block();
1304 int ns = b->num_successors();
1305 int nt = b->all_successors();
1307 tty->print("Parsing block #%d at bci [%d,%d), successors: ",
1308 block()->rpo(), block()->start(), block()->limit());
1309 for (int i = 0; i < nt; i++) {
1310 tty->print((( i < ns) ? " %d" : " %d(e)"), b->successor_at(i)->rpo());
1311 }
1312 if (b->is_loop_head()) tty->print(" lphd");
1313 tty->print_cr("");
1314 }
1316 assert(block()->is_merged(), "must be merged before being parsed");
1317 block()->mark_parsed();
1318 ++_blocks_parsed;
1320 // Set iterator to start of block.
1321 iter().reset_to_bci(block()->start());
1323 CompileLog* log = C->log();
1325 // Parse bytecodes
1326 while (!stopped() && !failing()) {
1327 iter().next();
1329 // Learn the current bci from the iterator:
1330 set_parse_bci(iter().cur_bci());
1332 if (bci() == block()->limit()) {
1333 // insert a predicate if it falls through to a loop head block
1334 if (should_add_predicate(bci())){
1335 add_predicate();
1336 }
1337 // Do not walk into the next block until directed by do_all_blocks.
1338 merge(bci());
1339 break;
1340 }
1341 assert(bci() < block()->limit(), "bci still in block");
1343 if (log != NULL) {
1344 // Output an optional context marker, to help place actions
1345 // that occur during parsing of this BC. If there is no log
1346 // output until the next context string, this context string
1347 // will be silently ignored.
1348 log->context()->reset();
1349 log->context()->print_cr("<bc code='%d' bci='%d'/>", (int)bc(), bci());
1350 }
1352 if (block()->has_trap_at(bci())) {
1353 // We must respect the flow pass's traps, because it will refuse
1354 // to produce successors for trapping blocks.
1355 int trap_index = block()->flow()->trap_index();
1356 assert(trap_index != 0, "trap index must be valid");
1357 uncommon_trap(trap_index);
1358 break;
1359 }
1361 NOT_PRODUCT( parse_histogram()->set_initial_state(bc()); );
1363 #ifdef ASSERT
1364 int pre_bc_sp = sp();
1365 int inputs, depth;
1366 bool have_se = !stopped() && compute_stack_effects(inputs, depth);
1367 assert(!have_se || pre_bc_sp >= inputs, "have enough stack to execute this BC");
1368 #endif //ASSERT
1370 do_one_bytecode();
1372 assert(!have_se || stopped() || failing() || (sp() - pre_bc_sp) == depth, "correct depth prediction");
1374 do_exceptions();
1376 NOT_PRODUCT( parse_histogram()->record_change(); );
1378 if (log != NULL) log->context()->reset(); // done w/ this one
1380 // Fall into next bytecode. Each bytecode normally has 1 sequential
1381 // successor which is typically made ready by visiting this bytecode.
1382 // If the successor has several predecessors, then it is a merge
1383 // point, starts a new basic block, and is handled like other basic blocks.
1384 }
1385 }
1388 //------------------------------merge------------------------------------------
1389 void Parse::set_parse_bci(int bci) {
1390 set_bci(bci);
1391 Node_Notes* nn = C->default_node_notes();
1392 if (nn == NULL) return;
1394 // Collect debug info for inlined calls unless -XX:-DebugInlinedCalls.
1395 if (!DebugInlinedCalls && depth() > 1) {
1396 return;
1397 }
1399 // Update the JVMS annotation, if present.
1400 JVMState* jvms = nn->jvms();
1401 if (jvms != NULL && jvms->bci() != bci) {
1402 // Update the JVMS.
1403 jvms = jvms->clone_shallow(C);
1404 jvms->set_bci(bci);
1405 nn->set_jvms(jvms);
1406 }
1407 }
1409 //------------------------------merge------------------------------------------
1410 // Merge the current mapping into the basic block starting at bci
1411 void Parse::merge(int target_bci) {
1412 Block* target = successor_for_bci(target_bci);
1413 if (target == NULL) { handle_missing_successor(target_bci); return; }
1414 assert(!target->is_ready(), "our arrival must be expected");
1415 int pnum = target->next_path_num();
1416 merge_common(target, pnum);
1417 }
1419 //-------------------------merge_new_path--------------------------------------
1420 // Merge the current mapping into the basic block, using a new path
1421 void Parse::merge_new_path(int target_bci) {
1422 Block* target = successor_for_bci(target_bci);
1423 if (target == NULL) { handle_missing_successor(target_bci); return; }
1424 assert(!target->is_ready(), "new path into frozen graph");
1425 int pnum = target->add_new_path();
1426 merge_common(target, pnum);
1427 }
1429 //-------------------------merge_exception-------------------------------------
1430 // Merge the current mapping into the basic block starting at bci
1431 // The ex_oop must be pushed on the stack, unlike throw_to_exit.
1432 void Parse::merge_exception(int target_bci) {
1433 assert(sp() == 1, "must have only the throw exception on the stack");
1434 Block* target = successor_for_bci(target_bci);
1435 if (target == NULL) { handle_missing_successor(target_bci); return; }
1436 assert(target->is_handler(), "exceptions are handled by special blocks");
1437 int pnum = target->add_new_path();
1438 merge_common(target, pnum);
1439 }
1441 //--------------------handle_missing_successor---------------------------------
1442 void Parse::handle_missing_successor(int target_bci) {
1443 #ifndef PRODUCT
1444 Block* b = block();
1445 int trap_bci = b->flow()->has_trap()? b->flow()->trap_bci(): -1;
1446 tty->print_cr("### Missing successor at bci:%d for block #%d (trap_bci:%d)", target_bci, b->rpo(), trap_bci);
1447 #endif
1448 ShouldNotReachHere();
1449 }
1451 //--------------------------merge_common---------------------------------------
1452 void Parse::merge_common(Parse::Block* target, int pnum) {
1453 if (TraceOptoParse) {
1454 tty->print("Merging state at block #%d bci:%d", target->rpo(), target->start());
1455 }
1457 // Zap extra stack slots to top
1458 assert(sp() == target->start_sp(), "");
1459 clean_stack(sp());
1461 if (!target->is_merged()) { // No prior mapping at this bci
1462 if (TraceOptoParse) { tty->print(" with empty state"); }
1464 // If this path is dead, do not bother capturing it as a merge.
1465 // It is "as if" we had 1 fewer predecessors from the beginning.
1466 if (stopped()) {
1467 if (TraceOptoParse) tty->print_cr(", but path is dead and doesn't count");
1468 return;
1469 }
1471 // Record that a new block has been merged.
1472 ++_blocks_merged;
1474 // Make a region if we know there are multiple or unpredictable inputs.
1475 // (Also, if this is a plain fall-through, we might see another region,
1476 // which must not be allowed into this block's map.)
1477 if (pnum > PhiNode::Input // Known multiple inputs.
1478 || target->is_handler() // These have unpredictable inputs.
1479 || target->is_loop_head() // Known multiple inputs
1480 || control()->is_Region()) { // We must hide this guy.
1481 // Add a Region to start the new basic block. Phis will be added
1482 // later lazily.
1483 int edges = target->pred_count();
1484 if (edges < pnum) edges = pnum; // might be a new path!
1485 Node *r = new (C, edges+1) RegionNode(edges+1);
1486 gvn().set_type(r, Type::CONTROL);
1487 record_for_igvn(r);
1488 // zap all inputs to NULL for debugging (done in Node(uint) constructor)
1489 // for (int j = 1; j < edges+1; j++) { r->init_req(j, NULL); }
1490 r->init_req(pnum, control());
1491 set_control(r);
1492 }
1494 // Convert the existing Parser mapping into a mapping at this bci.
1495 store_state_to(target);
1496 assert(target->is_merged(), "do not come here twice");
1498 } else { // Prior mapping at this bci
1499 if (TraceOptoParse) { tty->print(" with previous state"); }
1501 // We must not manufacture more phis if the target is already parsed.
1502 bool nophi = target->is_parsed();
1504 SafePointNode* newin = map();// Hang on to incoming mapping
1505 Block* save_block = block(); // Hang on to incoming block;
1506 load_state_from(target); // Get prior mapping
1508 assert(newin->jvms()->locoff() == jvms()->locoff(), "JVMS layouts agree");
1509 assert(newin->jvms()->stkoff() == jvms()->stkoff(), "JVMS layouts agree");
1510 assert(newin->jvms()->monoff() == jvms()->monoff(), "JVMS layouts agree");
1511 assert(newin->jvms()->endoff() == jvms()->endoff(), "JVMS layouts agree");
1513 // Iterate over my current mapping and the old mapping.
1514 // Where different, insert Phi functions.
1515 // Use any existing Phi functions.
1516 assert(control()->is_Region(), "must be merging to a region");
1517 RegionNode* r = control()->as_Region();
1519 // Compute where to merge into
1520 // Merge incoming control path
1521 r->init_req(pnum, newin->control());
1523 if (pnum == 1) { // Last merge for this Region?
1524 if (!block()->flow()->is_irreducible_entry()) {
1525 Node* result = _gvn.transform_no_reclaim(r);
1526 if (r != result && TraceOptoParse) {
1527 tty->print_cr("Block #%d replace %d with %d", block()->rpo(), r->_idx, result->_idx);
1528 }
1529 }
1530 record_for_igvn(r);
1531 }
1533 // Update all the non-control inputs to map:
1534 assert(TypeFunc::Parms == newin->jvms()->locoff(), "parser map should contain only youngest jvms");
1535 bool check_elide_phi = target->is_SEL_backedge(save_block);
1536 for (uint j = 1; j < newin->req(); j++) {
1537 Node* m = map()->in(j); // Current state of target.
1538 Node* n = newin->in(j); // Incoming change to target state.
1539 PhiNode* phi;
1540 if (m->is_Phi() && m->as_Phi()->region() == r)
1541 phi = m->as_Phi();
1542 else
1543 phi = NULL;
1544 if (m != n) { // Different; must merge
1545 switch (j) {
1546 // Frame pointer and Return Address never changes
1547 case TypeFunc::FramePtr:// Drop m, use the original value
1548 case TypeFunc::ReturnAdr:
1549 break;
1550 case TypeFunc::Memory: // Merge inputs to the MergeMem node
1551 assert(phi == NULL, "the merge contains phis, not vice versa");
1552 merge_memory_edges(n->as_MergeMem(), pnum, nophi);
1553 continue;
1554 default: // All normal stuff
1555 if (phi == NULL) {
1556 if (!check_elide_phi || !target->can_elide_SEL_phi(j)) {
1557 phi = ensure_phi(j, nophi);
1558 }
1559 }
1560 break;
1561 }
1562 }
1563 // At this point, n might be top if:
1564 // - there is no phi (because TypeFlow detected a conflict), or
1565 // - the corresponding control edges is top (a dead incoming path)
1566 // It is a bug if we create a phi which sees a garbage value on a live path.
1568 if (phi != NULL) {
1569 assert(n != top() || r->in(pnum) == top(), "live value must not be garbage");
1570 assert(phi->region() == r, "");
1571 phi->set_req(pnum, n); // Then add 'n' to the merge
1572 if (pnum == PhiNode::Input) {
1573 // Last merge for this Phi.
1574 // So far, Phis have had a reasonable type from ciTypeFlow.
1575 // Now _gvn will join that with the meet of current inputs.
1576 // BOTTOM is never permissible here, 'cause pessimistically
1577 // Phis of pointers cannot lose the basic pointer type.
1578 debug_only(const Type* bt1 = phi->bottom_type());
1579 assert(bt1 != Type::BOTTOM, "should not be building conflict phis");
1580 map()->set_req(j, _gvn.transform_no_reclaim(phi));
1581 debug_only(const Type* bt2 = phi->bottom_type());
1582 assert(bt2->higher_equal(bt1), "must be consistent with type-flow");
1583 record_for_igvn(phi);
1584 }
1585 }
1586 } // End of for all values to be merged
1588 if (pnum == PhiNode::Input &&
1589 !r->in(0)) { // The occasional useless Region
1590 assert(control() == r, "");
1591 set_control(r->nonnull_req());
1592 }
1594 // newin has been subsumed into the lazy merge, and is now dead.
1595 set_block(save_block);
1597 stop(); // done with this guy, for now
1598 }
1600 if (TraceOptoParse) {
1601 tty->print_cr(" on path %d", pnum);
1602 }
1604 // Done with this parser state.
1605 assert(stopped(), "");
1606 }
1609 //--------------------------merge_memory_edges---------------------------------
1610 void Parse::merge_memory_edges(MergeMemNode* n, int pnum, bool nophi) {
1611 // (nophi means we must not create phis, because we already parsed here)
1612 assert(n != NULL, "");
1613 // Merge the inputs to the MergeMems
1614 MergeMemNode* m = merged_memory();
1616 assert(control()->is_Region(), "must be merging to a region");
1617 RegionNode* r = control()->as_Region();
1619 PhiNode* base = NULL;
1620 MergeMemNode* remerge = NULL;
1621 for (MergeMemStream mms(m, n); mms.next_non_empty2(); ) {
1622 Node *p = mms.force_memory();
1623 Node *q = mms.memory2();
1624 if (mms.is_empty() && nophi) {
1625 // Trouble: No new splits allowed after a loop body is parsed.
1626 // Instead, wire the new split into a MergeMem on the backedge.
1627 // The optimizer will sort it out, slicing the phi.
1628 if (remerge == NULL) {
1629 assert(base != NULL, "");
1630 assert(base->in(0) != NULL, "should not be xformed away");
1631 remerge = MergeMemNode::make(C, base->in(pnum));
1632 gvn().set_type(remerge, Type::MEMORY);
1633 base->set_req(pnum, remerge);
1634 }
1635 remerge->set_memory_at(mms.alias_idx(), q);
1636 continue;
1637 }
1638 assert(!q->is_MergeMem(), "");
1639 PhiNode* phi;
1640 if (p != q) {
1641 phi = ensure_memory_phi(mms.alias_idx(), nophi);
1642 } else {
1643 if (p->is_Phi() && p->as_Phi()->region() == r)
1644 phi = p->as_Phi();
1645 else
1646 phi = NULL;
1647 }
1648 // Insert q into local phi
1649 if (phi != NULL) {
1650 assert(phi->region() == r, "");
1651 p = phi;
1652 phi->set_req(pnum, q);
1653 if (mms.at_base_memory()) {
1654 base = phi; // delay transforming it
1655 } else if (pnum == 1) {
1656 record_for_igvn(phi);
1657 p = _gvn.transform_no_reclaim(phi);
1658 }
1659 mms.set_memory(p);// store back through the iterator
1660 }
1661 }
1662 // Transform base last, in case we must fiddle with remerging.
1663 if (base != NULL && pnum == 1) {
1664 record_for_igvn(base);
1665 m->set_base_memory( _gvn.transform_no_reclaim(base) );
1666 }
1667 }
1670 //------------------------ensure_phis_everywhere-------------------------------
1671 void Parse::ensure_phis_everywhere() {
1672 ensure_phi(TypeFunc::I_O);
1674 // Ensure a phi on all currently known memories.
1675 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
1676 ensure_memory_phi(mms.alias_idx());
1677 debug_only(mms.set_memory()); // keep the iterator happy
1678 }
1680 // Note: This is our only chance to create phis for memory slices.
1681 // If we miss a slice that crops up later, it will have to be
1682 // merged into the base-memory phi that we are building here.
1683 // Later, the optimizer will comb out the knot, and build separate
1684 // phi-loops for each memory slice that matters.
1686 // Monitors must nest nicely and not get confused amongst themselves.
1687 // Phi-ify everything up to the monitors, though.
1688 uint monoff = map()->jvms()->monoff();
1689 uint nof_monitors = map()->jvms()->nof_monitors();
1691 assert(TypeFunc::Parms == map()->jvms()->locoff(), "parser map should contain only youngest jvms");
1692 bool check_elide_phi = block()->is_SEL_head();
1693 for (uint i = TypeFunc::Parms; i < monoff; i++) {
1694 if (!check_elide_phi || !block()->can_elide_SEL_phi(i)) {
1695 ensure_phi(i);
1696 }
1697 }
1699 // Even monitors need Phis, though they are well-structured.
1700 // This is true for OSR methods, and also for the rare cases where
1701 // a monitor object is the subject of a replace_in_map operation.
1702 // See bugs 4426707 and 5043395.
1703 for (uint m = 0; m < nof_monitors; m++) {
1704 ensure_phi(map()->jvms()->monitor_obj_offset(m));
1705 }
1706 }
1709 //-----------------------------add_new_path------------------------------------
1710 // Add a previously unaccounted predecessor to this block.
1711 int Parse::Block::add_new_path() {
1712 // If there is no map, return the lowest unused path number.
1713 if (!is_merged()) return pred_count()+1; // there will be a map shortly
1715 SafePointNode* map = start_map();
1716 if (!map->control()->is_Region())
1717 return pred_count()+1; // there may be a region some day
1718 RegionNode* r = map->control()->as_Region();
1720 // Add new path to the region.
1721 uint pnum = r->req();
1722 r->add_req(NULL);
1724 for (uint i = 1; i < map->req(); i++) {
1725 Node* n = map->in(i);
1726 if (i == TypeFunc::Memory) {
1727 // Ensure a phi on all currently known memories.
1728 for (MergeMemStream mms(n->as_MergeMem()); mms.next_non_empty(); ) {
1729 Node* phi = mms.memory();
1730 if (phi->is_Phi() && phi->as_Phi()->region() == r) {
1731 assert(phi->req() == pnum, "must be same size as region");
1732 phi->add_req(NULL);
1733 }
1734 }
1735 } else {
1736 if (n->is_Phi() && n->as_Phi()->region() == r) {
1737 assert(n->req() == pnum, "must be same size as region");
1738 n->add_req(NULL);
1739 }
1740 }
1741 }
1743 return pnum;
1744 }
1746 //------------------------------ensure_phi-------------------------------------
1747 // Turn the idx'th entry of the current map into a Phi
1748 PhiNode *Parse::ensure_phi(int idx, bool nocreate) {
1749 SafePointNode* map = this->map();
1750 Node* region = map->control();
1751 assert(region->is_Region(), "");
1753 Node* o = map->in(idx);
1754 assert(o != NULL, "");
1756 if (o == top()) return NULL; // TOP always merges into TOP
1758 if (o->is_Phi() && o->as_Phi()->region() == region) {
1759 return o->as_Phi();
1760 }
1762 // Now use a Phi here for merging
1763 assert(!nocreate, "Cannot build a phi for a block already parsed.");
1764 const JVMState* jvms = map->jvms();
1765 const Type* t;
1766 if (jvms->is_loc(idx)) {
1767 t = block()->local_type_at(idx - jvms->locoff());
1768 } else if (jvms->is_stk(idx)) {
1769 t = block()->stack_type_at(idx - jvms->stkoff());
1770 } else if (jvms->is_mon(idx)) {
1771 assert(!jvms->is_monitor_box(idx), "no phis for boxes");
1772 t = TypeInstPtr::BOTTOM; // this is sufficient for a lock object
1773 } else if ((uint)idx < TypeFunc::Parms) {
1774 t = o->bottom_type(); // Type::RETURN_ADDRESS or such-like.
1775 } else {
1776 assert(false, "no type information for this phi");
1777 }
1779 // If the type falls to bottom, then this must be a local that
1780 // is mixing ints and oops or some such. Forcing it to top
1781 // makes it go dead.
1782 if (t == Type::BOTTOM) {
1783 map->set_req(idx, top());
1784 return NULL;
1785 }
1787 // Do not create phis for top either.
1788 // A top on a non-null control flow must be an unused even after the.phi.
1789 if (t == Type::TOP || t == Type::HALF) {
1790 map->set_req(idx, top());
1791 return NULL;
1792 }
1794 PhiNode* phi = PhiNode::make(region, o, t);
1795 gvn().set_type(phi, t);
1796 if (C->do_escape_analysis()) record_for_igvn(phi);
1797 map->set_req(idx, phi);
1798 return phi;
1799 }
1801 //--------------------------ensure_memory_phi----------------------------------
1802 // Turn the idx'th slice of the current memory into a Phi
1803 PhiNode *Parse::ensure_memory_phi(int idx, bool nocreate) {
1804 MergeMemNode* mem = merged_memory();
1805 Node* region = control();
1806 assert(region->is_Region(), "");
1808 Node *o = (idx == Compile::AliasIdxBot)? mem->base_memory(): mem->memory_at(idx);
1809 assert(o != NULL && o != top(), "");
1811 PhiNode* phi;
1812 if (o->is_Phi() && o->as_Phi()->region() == region) {
1813 phi = o->as_Phi();
1814 if (phi == mem->base_memory() && idx >= Compile::AliasIdxRaw) {
1815 // clone the shared base memory phi to make a new memory split
1816 assert(!nocreate, "Cannot build a phi for a block already parsed.");
1817 const Type* t = phi->bottom_type();
1818 const TypePtr* adr_type = C->get_adr_type(idx);
1819 phi = phi->slice_memory(adr_type);
1820 gvn().set_type(phi, t);
1821 }
1822 return phi;
1823 }
1825 // Now use a Phi here for merging
1826 assert(!nocreate, "Cannot build a phi for a block already parsed.");
1827 const Type* t = o->bottom_type();
1828 const TypePtr* adr_type = C->get_adr_type(idx);
1829 phi = PhiNode::make(region, o, t, adr_type);
1830 gvn().set_type(phi, t);
1831 if (idx == Compile::AliasIdxBot)
1832 mem->set_base_memory(phi);
1833 else
1834 mem->set_memory_at(idx, phi);
1835 return phi;
1836 }
1838 //------------------------------call_register_finalizer-----------------------
1839 // Check the klass of the receiver and call register_finalizer if the
1840 // class need finalization.
1841 void Parse::call_register_finalizer() {
1842 Node* receiver = local(0);
1843 assert(receiver != NULL && receiver->bottom_type()->isa_instptr() != NULL,
1844 "must have non-null instance type");
1846 const TypeInstPtr *tinst = receiver->bottom_type()->isa_instptr();
1847 if (tinst != NULL && tinst->klass()->is_loaded() && !tinst->klass_is_exact()) {
1848 // The type isn't known exactly so see if CHA tells us anything.
1849 ciInstanceKlass* ik = tinst->klass()->as_instance_klass();
1850 if (!Dependencies::has_finalizable_subclass(ik)) {
1851 // No finalizable subclasses so skip the dynamic check.
1852 C->dependencies()->assert_has_no_finalizable_subclasses(ik);
1853 return;
1854 }
1855 }
1857 // Insert a dynamic test for whether the instance needs
1858 // finalization. In general this will fold up since the concrete
1859 // class is often visible so the access flags are constant.
1860 Node* klass_addr = basic_plus_adr( receiver, receiver, oopDesc::klass_offset_in_bytes() );
1861 Node* klass = _gvn.transform( LoadKlassNode::make(_gvn, immutable_memory(), klass_addr, TypeInstPtr::KLASS) );
1863 Node* access_flags_addr = basic_plus_adr(klass, klass, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc));
1864 Node* access_flags = make_load(NULL, access_flags_addr, TypeInt::INT, T_INT);
1866 Node* mask = _gvn.transform(new (C, 3) AndINode(access_flags, intcon(JVM_ACC_HAS_FINALIZER)));
1867 Node* check = _gvn.transform(new (C, 3) CmpINode(mask, intcon(0)));
1868 Node* test = _gvn.transform(new (C, 2) BoolNode(check, BoolTest::ne));
1870 IfNode* iff = create_and_map_if(control(), test, PROB_MAX, COUNT_UNKNOWN);
1872 RegionNode* result_rgn = new (C, 3) RegionNode(3);
1873 record_for_igvn(result_rgn);
1875 Node *skip_register = _gvn.transform(new (C, 1) IfFalseNode(iff));
1876 result_rgn->init_req(1, skip_register);
1878 Node *needs_register = _gvn.transform(new (C, 1) IfTrueNode(iff));
1879 set_control(needs_register);
1880 if (stopped()) {
1881 // There is no slow path.
1882 result_rgn->init_req(2, top());
1883 } else {
1884 Node *call = make_runtime_call(RC_NO_LEAF,
1885 OptoRuntime::register_finalizer_Type(),
1886 OptoRuntime::register_finalizer_Java(),
1887 NULL, TypePtr::BOTTOM,
1888 receiver);
1889 make_slow_call_ex(call, env()->Throwable_klass(), true);
1891 Node* fast_io = call->in(TypeFunc::I_O);
1892 Node* fast_mem = call->in(TypeFunc::Memory);
1893 // These two phis are pre-filled with copies of of the fast IO and Memory
1894 Node* io_phi = PhiNode::make(result_rgn, fast_io, Type::ABIO);
1895 Node* mem_phi = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
1897 result_rgn->init_req(2, control());
1898 io_phi ->init_req(2, i_o());
1899 mem_phi ->init_req(2, reset_memory());
1901 set_all_memory( _gvn.transform(mem_phi) );
1902 set_i_o( _gvn.transform(io_phi) );
1903 }
1905 set_control( _gvn.transform(result_rgn) );
1906 }
1908 //------------------------------return_current---------------------------------
1909 // Append current _map to _exit_return
1910 void Parse::return_current(Node* value) {
1911 if (RegisterFinalizersAtInit &&
1912 method()->intrinsic_id() == vmIntrinsics::_Object_init) {
1913 call_register_finalizer();
1914 }
1916 // Do not set_parse_bci, so that return goo is credited to the return insn.
1917 set_bci(InvocationEntryBci);
1918 if (method()->is_synchronized() && GenerateSynchronizationCode) {
1919 shared_unlock(_synch_lock->box_node(), _synch_lock->obj_node());
1920 }
1921 if (C->env()->dtrace_method_probes()) {
1922 make_dtrace_method_exit(method());
1923 }
1924 SafePointNode* exit_return = _exits.map();
1925 exit_return->in( TypeFunc::Control )->add_req( control() );
1926 exit_return->in( TypeFunc::I_O )->add_req( i_o () );
1927 Node *mem = exit_return->in( TypeFunc::Memory );
1928 for (MergeMemStream mms(mem->as_MergeMem(), merged_memory()); mms.next_non_empty2(); ) {
1929 if (mms.is_empty()) {
1930 // get a copy of the base memory, and patch just this one input
1931 const TypePtr* adr_type = mms.adr_type(C);
1932 Node* phi = mms.force_memory()->as_Phi()->slice_memory(adr_type);
1933 assert(phi->as_Phi()->region() == mms.base_memory()->in(0), "");
1934 gvn().set_type_bottom(phi);
1935 phi->del_req(phi->req()-1); // prepare to re-patch
1936 mms.set_memory(phi);
1937 }
1938 mms.memory()->add_req(mms.memory2());
1939 }
1941 // frame pointer is always same, already captured
1942 if (value != NULL) {
1943 // If returning oops to an interface-return, there is a silent free
1944 // cast from oop to interface allowed by the Verifier. Make it explicit
1945 // here.
1946 Node* phi = _exits.argument(0);
1947 const TypeInstPtr *tr = phi->bottom_type()->isa_instptr();
1948 if( tr && tr->klass()->is_loaded() &&
1949 tr->klass()->is_interface() ) {
1950 const TypeInstPtr *tp = value->bottom_type()->isa_instptr();
1951 if (tp && tp->klass()->is_loaded() &&
1952 !tp->klass()->is_interface()) {
1953 // sharpen the type eagerly; this eases certain assert checking
1954 if (tp->higher_equal(TypeInstPtr::NOTNULL))
1955 tr = tr->join(TypeInstPtr::NOTNULL)->is_instptr();
1956 value = _gvn.transform(new (C, 2) CheckCastPPNode(0,value,tr));
1957 }
1958 }
1959 phi->add_req(value);
1960 }
1962 stop_and_kill_map(); // This CFG path dies here
1963 }
1966 //------------------------------add_safepoint----------------------------------
1967 void Parse::add_safepoint() {
1968 // See if we can avoid this safepoint. No need for a SafePoint immediately
1969 // after a Call (except Leaf Call) or another SafePoint.
1970 Node *proj = control();
1971 bool add_poll_param = SafePointNode::needs_polling_address_input();
1972 uint parms = add_poll_param ? TypeFunc::Parms+1 : TypeFunc::Parms;
1973 if( proj->is_Proj() ) {
1974 Node *n0 = proj->in(0);
1975 if( n0->is_Catch() ) {
1976 n0 = n0->in(0)->in(0);
1977 assert( n0->is_Call(), "expect a call here" );
1978 }
1979 if( n0->is_Call() ) {
1980 if( n0->as_Call()->guaranteed_safepoint() )
1981 return;
1982 } else if( n0->is_SafePoint() && n0->req() >= parms ) {
1983 return;
1984 }
1985 }
1987 // Clear out dead values from the debug info.
1988 kill_dead_locals();
1990 // Clone the JVM State
1991 SafePointNode *sfpnt = new (C, parms) SafePointNode(parms, NULL);
1993 // Capture memory state BEFORE a SafePoint. Since we can block at a
1994 // SafePoint we need our GC state to be safe; i.e. we need all our current
1995 // write barriers (card marks) to not float down after the SafePoint so we
1996 // must read raw memory. Likewise we need all oop stores to match the card
1997 // marks. If deopt can happen, we need ALL stores (we need the correct JVM
1998 // state on a deopt).
2000 // We do not need to WRITE the memory state after a SafePoint. The control
2001 // edge will keep card-marks and oop-stores from floating up from below a
2002 // SafePoint and our true dependency added here will keep them from floating
2003 // down below a SafePoint.
2005 // Clone the current memory state
2006 Node* mem = MergeMemNode::make(C, map()->memory());
2008 mem = _gvn.transform(mem);
2010 // Pass control through the safepoint
2011 sfpnt->init_req(TypeFunc::Control , control());
2012 // Fix edges normally used by a call
2013 sfpnt->init_req(TypeFunc::I_O , top() );
2014 sfpnt->init_req(TypeFunc::Memory , mem );
2015 sfpnt->init_req(TypeFunc::ReturnAdr, top() );
2016 sfpnt->init_req(TypeFunc::FramePtr , top() );
2018 // Create a node for the polling address
2019 if( add_poll_param ) {
2020 Node *polladr = ConPNode::make(C, (address)os::get_polling_page());
2021 sfpnt->init_req(TypeFunc::Parms+0, _gvn.transform(polladr));
2022 }
2024 // Fix up the JVM State edges
2025 add_safepoint_edges(sfpnt);
2026 Node *transformed_sfpnt = _gvn.transform(sfpnt);
2027 set_control(transformed_sfpnt);
2029 // Provide an edge from root to safepoint. This makes the safepoint
2030 // appear useful until the parse has completed.
2031 if( OptoRemoveUseless && transformed_sfpnt->is_SafePoint() ) {
2032 assert(C->root() != NULL, "Expect parse is still valid");
2033 C->root()->add_prec(transformed_sfpnt);
2034 }
2035 }
2037 //------------------------------should_add_predicate--------------------------
2038 bool Parse::should_add_predicate(int target_bci) {
2039 if (!UseLoopPredicate) return false;
2040 Block* target = successor_for_bci(target_bci);
2041 if (target != NULL &&
2042 target->is_loop_head() &&
2043 block()->rpo() < target->rpo()) {
2044 return true;
2045 }
2046 return false;
2047 }
2049 //------------------------------add_predicate---------------------------------
2050 void Parse::add_predicate() {
2051 assert(UseLoopPredicate,"use only for loop predicate");
2052 Node *cont = _gvn.intcon(1);
2053 Node* opq = _gvn.transform(new (C, 2) Opaque1Node(C, cont));
2054 Node *bol = _gvn.transform(new (C, 2) Conv2BNode(opq));
2055 IfNode* iff = create_and_map_if(control(), bol, PROB_MAX, COUNT_UNKNOWN);
2056 Node* iffalse = _gvn.transform(new (C, 1) IfFalseNode(iff));
2057 C->add_predicate_opaq(opq);
2058 {
2059 PreserveJVMState pjvms(this);
2060 set_control(iffalse);
2061 uncommon_trap(Deoptimization::Reason_predicate,
2062 Deoptimization::Action_maybe_recompile);
2063 }
2064 Node* iftrue = _gvn.transform(new (C, 1) IfTrueNode(iff));
2065 set_control(iftrue);
2066 }
2068 #ifndef PRODUCT
2069 //------------------------show_parse_info--------------------------------------
2070 void Parse::show_parse_info() {
2071 InlineTree* ilt = NULL;
2072 if (C->ilt() != NULL) {
2073 JVMState* caller_jvms = is_osr_parse() ? caller()->caller() : caller();
2074 ilt = InlineTree::find_subtree_from_root(C->ilt(), caller_jvms, method());
2075 }
2076 if (PrintCompilation && Verbose) {
2077 if (depth() == 1) {
2078 if( ilt->count_inlines() ) {
2079 tty->print(" __inlined %d (%d bytes)", ilt->count_inlines(),
2080 ilt->count_inline_bcs());
2081 tty->cr();
2082 }
2083 } else {
2084 if (method()->is_synchronized()) tty->print("s");
2085 if (method()->has_exception_handlers()) tty->print("!");
2086 // Check this is not the final compiled version
2087 if (C->trap_can_recompile()) {
2088 tty->print("-");
2089 } else {
2090 tty->print(" ");
2091 }
2092 method()->print_short_name();
2093 if (is_osr_parse()) {
2094 tty->print(" @ %d", osr_bci());
2095 }
2096 tty->print(" (%d bytes)",method()->code_size());
2097 if (ilt->count_inlines()) {
2098 tty->print(" __inlined %d (%d bytes)", ilt->count_inlines(),
2099 ilt->count_inline_bcs());
2100 }
2101 tty->cr();
2102 }
2103 }
2104 if (PrintOpto && (depth() == 1 || PrintOptoInlining)) {
2105 // Print that we succeeded; suppress this message on the first osr parse.
2107 if (method()->is_synchronized()) tty->print("s");
2108 if (method()->has_exception_handlers()) tty->print("!");
2109 // Check this is not the final compiled version
2110 if (C->trap_can_recompile() && depth() == 1) {
2111 tty->print("-");
2112 } else {
2113 tty->print(" ");
2114 }
2115 if( depth() != 1 ) { tty->print(" "); } // missing compile count
2116 for (int i = 1; i < depth(); ++i) { tty->print(" "); }
2117 method()->print_short_name();
2118 if (is_osr_parse()) {
2119 tty->print(" @ %d", osr_bci());
2120 }
2121 if (ilt->caller_bci() != -1) {
2122 tty->print(" @ %d", ilt->caller_bci());
2123 }
2124 tty->print(" (%d bytes)",method()->code_size());
2125 if (ilt->count_inlines()) {
2126 tty->print(" __inlined %d (%d bytes)", ilt->count_inlines(),
2127 ilt->count_inline_bcs());
2128 }
2129 tty->cr();
2130 }
2131 }
2134 //------------------------------dump-------------------------------------------
2135 // Dump information associated with the bytecodes of current _method
2136 void Parse::dump() {
2137 if( method() != NULL ) {
2138 // Iterate over bytecodes
2139 ciBytecodeStream iter(method());
2140 for( Bytecodes::Code bc = iter.next(); bc != ciBytecodeStream::EOBC() ; bc = iter.next() ) {
2141 dump_bci( iter.cur_bci() );
2142 tty->cr();
2143 }
2144 }
2145 }
2147 // Dump information associated with a byte code index, 'bci'
2148 void Parse::dump_bci(int bci) {
2149 // Output info on merge-points, cloning, and within _jsr..._ret
2150 // NYI
2151 tty->print(" bci:%d", bci);
2152 }
2154 #endif