Fri, 12 Feb 2010 15:27:36 -0800
Merge
1 /*
2 * Copyright 1997-2009 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 class BytecodeParseHistogram;
26 class InlineTree;
27 class Parse;
28 class SwitchRange;
31 //------------------------------InlineTree-------------------------------------
32 class InlineTree : public ResourceObj {
33 Compile* C; // cache
34 JVMState* _caller_jvms; // state of caller
35 ciMethod* _method; // method being called by the caller_jvms
36 InlineTree* _caller_tree;
37 uint _count_inline_bcs; // Accumulated count of inlined bytecodes
38 // Call-site count / interpreter invocation count, scaled recursively.
39 // Always between 0.0 and 1.0. Represents the percentage of the method's
40 // total execution time used at this call site.
41 const float _site_invoke_ratio;
42 const int _site_depth_adjust;
43 float compute_callee_frequency( int caller_bci ) const;
45 GrowableArray<InlineTree*> _subtrees;
46 friend class Compile;
48 protected:
49 InlineTree(Compile* C,
50 const InlineTree* caller_tree,
51 ciMethod* callee_method,
52 JVMState* caller_jvms,
53 int caller_bci,
54 float site_invoke_ratio,
55 int site_depth_adjust);
56 InlineTree *build_inline_tree_for_callee(ciMethod* callee_method,
57 JVMState* caller_jvms,
58 int caller_bci);
59 const char* try_to_inline(ciMethod* callee_method, ciMethod* caller_method, int caller_bci, ciCallProfile& profile, WarmCallInfo* wci_result);
60 const char* shouldInline(ciMethod* callee_method, ciMethod* caller_method, int caller_bci, ciCallProfile& profile, WarmCallInfo* wci_result) const;
61 const char* shouldNotInline(ciMethod* callee_method, ciMethod* caller_method, WarmCallInfo* wci_result) const;
62 void print_inlining(ciMethod *callee_method, int caller_bci, const char *failure_msg) const PRODUCT_RETURN;
64 InlineTree *caller_tree() const { return _caller_tree; }
65 InlineTree* callee_at(int bci, ciMethod* m) const;
66 int inline_depth() const { return stack_depth() + _site_depth_adjust; }
67 int stack_depth() const { return _caller_jvms ? _caller_jvms->depth() : 0; }
69 public:
70 static InlineTree* build_inline_tree_root();
71 static InlineTree* find_subtree_from_root(InlineTree* root, JVMState* jvms, ciMethod* callee, bool create_if_not_found = false);
73 // For temporary (stack-allocated, stateless) ilts:
74 InlineTree(Compile* c, ciMethod* callee_method, JVMState* caller_jvms, float site_invoke_ratio, int site_depth_adjust);
76 // InlineTree enum
77 enum InlineStyle {
78 Inline_do_not_inline = 0, //
79 Inline_cha_is_monomorphic = 1, //
80 Inline_type_profile_monomorphic = 2 //
81 };
83 // See if it is OK to inline.
84 // The receiver is the inline tree for the caller.
85 //
86 // The result is a temperature indication. If it is hot or cold,
87 // inlining is immediate or undesirable. Otherwise, the info block
88 // returned is newly allocated and may be enqueued.
89 //
90 // If the method is inlinable, a new inline subtree is created on the fly,
91 // and may be accessed by find_subtree_from_root.
92 // The call_method is the dest_method for a special or static invocation.
93 // The call_method is an optimized virtual method candidate otherwise.
94 WarmCallInfo* ok_to_inline(ciMethod *call_method, JVMState* caller_jvms, ciCallProfile& profile, WarmCallInfo* wci);
96 // Information about inlined method
97 JVMState* caller_jvms() const { return _caller_jvms; }
98 ciMethod *method() const { return _method; }
99 int caller_bci() const { return _caller_jvms ? _caller_jvms->bci() : InvocationEntryBci; }
100 uint count_inline_bcs() const { return _count_inline_bcs; }
101 float site_invoke_ratio() const { return _site_invoke_ratio; };
103 #ifndef PRODUCT
104 private:
105 uint _count_inlines; // Count of inlined methods
106 public:
107 // Debug information collected during parse
108 uint count_inlines() const { return _count_inlines; };
109 #endif
110 GrowableArray<InlineTree*> subtrees() { return _subtrees; }
111 };
114 //-----------------------------------------------------------------------------
115 //------------------------------Parse------------------------------------------
116 // Parse bytecodes, build a Graph
117 class Parse : public GraphKit {
118 public:
119 // Per-block information needed by the parser:
120 class Block {
121 private:
122 ciTypeFlow::Block* _flow;
123 int _pred_count; // how many predecessors in CFG?
124 int _preds_parsed; // how many of these have been parsed?
125 uint _count; // how many times executed? Currently only set by _goto's
126 bool _is_parsed; // has this block been parsed yet?
127 bool _is_handler; // is this block an exception handler?
128 SafePointNode* _start_map; // all values flowing into this block
129 MethodLivenessResult _live_locals; // lazily initialized liveness bitmap
131 int _num_successors; // Includes only normal control flow.
132 int _all_successors; // Include exception paths also.
133 Block** _successors;
135 // Use init_node/init_graph to initialize Blocks.
136 // Block() : _live_locals((uintptr_t*)NULL,0) { ShouldNotReachHere(); }
137 Block() : _live_locals(NULL,0) { ShouldNotReachHere(); }
139 public:
141 // Set up the block data structure itself.
142 void init_node(Parse* outer, int po);
143 // Set up the block's relations to other blocks.
144 void init_graph(Parse* outer);
146 ciTypeFlow::Block* flow() const { return _flow; }
147 int pred_count() const { return _pred_count; }
148 int preds_parsed() const { return _preds_parsed; }
149 bool is_parsed() const { return _is_parsed; }
150 bool is_handler() const { return _is_handler; }
151 void set_count( uint x ) { _count = x; }
152 uint count() const { return _count; }
154 SafePointNode* start_map() const { assert(is_merged(),""); return _start_map; }
155 void set_start_map(SafePointNode* m) { assert(!is_merged(), ""); _start_map = m; }
157 // True after any predecessor flows control into this block
158 bool is_merged() const { return _start_map != NULL; }
160 // True when all non-exception predecessors have been parsed.
161 bool is_ready() const { return preds_parsed() == pred_count(); }
163 int num_successors() const { return _num_successors; }
164 int all_successors() const { return _all_successors; }
165 Block* successor_at(int i) const {
166 assert((uint)i < (uint)all_successors(), "");
167 return _successors[i];
168 }
169 Block* successor_for_bci(int bci);
171 int start() const { return flow()->start(); }
172 int limit() const { return flow()->limit(); }
173 int rpo() const { return flow()->rpo(); }
174 int start_sp() const { return flow()->stack_size(); }
176 bool is_loop_head() const { return flow()->is_loop_head(); }
177 bool is_SEL_head() const { return flow()->is_single_entry_loop_head(); }
178 bool is_SEL_backedge(Block* pred) const{ return is_SEL_head() && pred->rpo() >= rpo(); }
179 bool is_invariant_local(uint i) const {
180 const JVMState* jvms = start_map()->jvms();
181 if (!jvms->is_loc(i) || flow()->outer()->has_irreducible_entry()) return false;
182 return flow()->is_invariant_local(i - jvms->locoff());
183 }
184 bool can_elide_SEL_phi(uint i) const { assert(is_SEL_head(),""); return is_invariant_local(i); }
186 const Type* peek(int off=0) const { return stack_type_at(start_sp() - (off+1)); }
188 const Type* stack_type_at(int i) const;
189 const Type* local_type_at(int i) const;
190 static const Type* get_type(ciType* t) { return Type::get_typeflow_type(t); }
192 bool has_trap_at(int bci) const { return flow()->has_trap() && flow()->trap_bci() == bci; }
194 // Call this just before parsing a block.
195 void mark_parsed() {
196 assert(!_is_parsed, "must parse each block exactly once");
197 _is_parsed = true;
198 }
200 // Return the phi/region input index for the "current" pred,
201 // and bump the pred number. For historical reasons these index
202 // numbers are handed out in descending order. The last index is
203 // always PhiNode::Input (i.e., 1). The value returned is known
204 // as a "path number" because it distinguishes by which path we are
205 // entering the block.
206 int next_path_num() {
207 assert(preds_parsed() < pred_count(), "too many preds?");
208 return pred_count() - _preds_parsed++;
209 }
211 // Add a previously unaccounted predecessor to this block.
212 // This operates by increasing the size of the block's region
213 // and all its phi nodes (if any). The value returned is a
214 // path number ("pnum").
215 int add_new_path();
217 // Initialize me by recording the parser's map. My own map must be NULL.
218 void record_state(Parse* outer);
219 };
221 #ifndef PRODUCT
222 // BytecodeParseHistogram collects number of bytecodes parsed, nodes constructed, and transformations.
223 class BytecodeParseHistogram : public ResourceObj {
224 private:
225 enum BPHType {
226 BPH_transforms,
227 BPH_values
228 };
229 static bool _initialized;
230 static uint _bytecodes_parsed [Bytecodes::number_of_codes];
231 static uint _nodes_constructed[Bytecodes::number_of_codes];
232 static uint _nodes_transformed[Bytecodes::number_of_codes];
233 static uint _new_values [Bytecodes::number_of_codes];
235 Bytecodes::Code _initial_bytecode;
236 int _initial_node_count;
237 int _initial_transforms;
238 int _initial_values;
240 Parse *_parser;
241 Compile *_compiler;
243 // Initialization
244 static void reset();
246 // Return info being collected, select with global flag 'BytecodeParseInfo'
247 int current_count(BPHType info_selector);
249 public:
250 BytecodeParseHistogram(Parse *p, Compile *c);
251 static bool initialized();
253 // Record info when starting to parse one bytecode
254 void set_initial_state( Bytecodes::Code bc );
255 // Record results of parsing one bytecode
256 void record_change();
258 // Profile printing
259 static void print(float cutoff = 0.01F); // cutoff in percent
260 };
262 public:
263 // Record work done during parsing
264 BytecodeParseHistogram* _parse_histogram;
265 void set_parse_histogram(BytecodeParseHistogram *bph) { _parse_histogram = bph; }
266 BytecodeParseHistogram* parse_histogram() { return _parse_histogram; }
267 #endif
269 private:
270 friend class Block;
272 // Variables which characterize this compilation as a whole:
274 JVMState* _caller; // JVMS which carries incoming args & state.
275 float _expected_uses; // expected number of calls to this code
276 float _prof_factor; // discount applied to my profile counts
277 int _depth; // Inline tree depth, for debug printouts
278 const TypeFunc*_tf; // My kind of function type
279 int _entry_bci; // the osr bci or InvocationEntryBci
281 ciTypeFlow* _flow; // Results of previous flow pass.
282 Block* _blocks; // Array of basic-block structs.
283 int _block_count; // Number of elements in _blocks.
285 GraphKit _exits; // Record all normal returns and throws here.
286 bool _wrote_final; // Did we write a final field?
287 bool _count_invocations; // update and test invocation counter
288 bool _method_data_update; // update method data oop
290 // Variables which track Java semantics during bytecode parsing:
292 Block* _block; // block currently getting parsed
293 ciBytecodeStream _iter; // stream of this method's bytecodes
295 int _blocks_merged; // Progress meter: state merges from BB preds
296 int _blocks_parsed; // Progress meter: BBs actually parsed
298 const FastLockNode* _synch_lock; // FastLockNode for synchronized method
300 #ifndef PRODUCT
301 int _max_switch_depth; // Debugging SwitchRanges.
302 int _est_switch_depth; // Debugging SwitchRanges.
303 #endif
305 public:
306 // Constructor
307 Parse(JVMState* caller, ciMethod* parse_method, float expected_uses);
309 virtual Parse* is_Parse() const { return (Parse*)this; }
311 public:
312 // Accessors.
313 JVMState* caller() const { return _caller; }
314 float expected_uses() const { return _expected_uses; }
315 float prof_factor() const { return _prof_factor; }
316 int depth() const { return _depth; }
317 const TypeFunc* tf() const { return _tf; }
318 // entry_bci() -- see osr_bci, etc.
320 ciTypeFlow* flow() const { return _flow; }
321 // blocks() -- see rpo_at, start_block, etc.
322 int block_count() const { return _block_count; }
324 GraphKit& exits() { return _exits; }
325 bool wrote_final() const { return _wrote_final; }
326 void set_wrote_final(bool z) { _wrote_final = z; }
327 bool count_invocations() const { return _count_invocations; }
328 bool method_data_update() const { return _method_data_update; }
330 Block* block() const { return _block; }
331 ciBytecodeStream& iter() { return _iter; }
332 Bytecodes::Code bc() const { return _iter.cur_bc(); }
334 void set_block(Block* b) { _block = b; }
336 // Derived accessors:
337 bool is_normal_parse() const { return _entry_bci == InvocationEntryBci; }
338 bool is_osr_parse() const { return _entry_bci != InvocationEntryBci; }
339 int osr_bci() const { assert(is_osr_parse(),""); return _entry_bci; }
341 void set_parse_bci(int bci);
343 // Must this parse be aborted?
344 bool failing() { return C->failing(); }
346 Block* rpo_at(int rpo) {
347 assert(0 <= rpo && rpo < _block_count, "oob");
348 return &_blocks[rpo];
349 }
350 Block* start_block() {
351 return rpo_at(flow()->start_block()->rpo());
352 }
353 // Can return NULL if the flow pass did not complete a block.
354 Block* successor_for_bci(int bci) {
355 return block()->successor_for_bci(bci);
356 }
358 private:
359 // Create a JVMS & map for the initial state of this method.
360 SafePointNode* create_entry_map();
362 // OSR helpers
363 Node *fetch_interpreter_state(int index, BasicType bt, Node *local_addrs, Node *local_addrs_base);
364 Node* check_interpreter_type(Node* l, const Type* type, SafePointNode* &bad_type_exit);
365 void load_interpreter_state(Node* osr_buf);
367 // Functions for managing basic blocks:
368 void init_blocks();
369 void load_state_from(Block* b);
370 void store_state_to(Block* b) { b->record_state(this); }
372 // Parse all the basic blocks.
373 void do_all_blocks();
375 // Parse the current basic block
376 void do_one_block();
378 // Raise an error if we get a bad ciTypeFlow CFG.
379 void handle_missing_successor(int bci);
381 // first actions (before BCI 0)
382 void do_method_entry();
384 // implementation of monitorenter/monitorexit
385 void do_monitor_enter();
386 void do_monitor_exit();
388 // Eagerly create phie throughout the state, to cope with back edges.
389 void ensure_phis_everywhere();
391 // Merge the current mapping into the basic block starting at bci
392 void merge( int target_bci);
393 // Same as plain merge, except that it allocates a new path number.
394 void merge_new_path( int target_bci);
395 // Merge the current mapping into an exception handler.
396 void merge_exception(int target_bci);
397 // Helper: Merge the current mapping into the given basic block
398 void merge_common(Block* target, int pnum);
399 // Helper functions for merging individual cells.
400 PhiNode *ensure_phi( int idx, bool nocreate = false);
401 PhiNode *ensure_memory_phi(int idx, bool nocreate = false);
402 // Helper to merge the current memory state into the given basic block
403 void merge_memory_edges(MergeMemNode* n, int pnum, bool nophi);
405 // Parse this bytecode, and alter the Parsers JVM->Node mapping
406 void do_one_bytecode();
408 // helper function to generate array store check
409 void array_store_check();
410 // Helper function to generate array load
411 void array_load(BasicType etype);
412 // Helper function to generate array store
413 void array_store(BasicType etype);
414 // Helper function to compute array addressing
415 Node* array_addressing(BasicType type, int vals, const Type* *result2=NULL);
417 // Pass current map to exits
418 void return_current(Node* value);
420 // Register finalizers on return from Object.<init>
421 void call_register_finalizer();
423 // Insert a compiler safepoint into the graph
424 void add_safepoint();
426 // Insert a compiler safepoint into the graph, if there is a back-branch.
427 void maybe_add_safepoint(int target_bci) {
428 if (UseLoopSafepoints && target_bci <= bci()) {
429 add_safepoint();
430 }
431 }
433 // Return true if the parser should add a loop predicate
434 bool should_add_predicate(int target_bci);
435 // Insert a loop predicate into the graph
436 void add_predicate();
438 // Note: Intrinsic generation routines may be found in library_call.cpp.
440 // Helper function to setup Ideal Call nodes
441 void do_call();
443 // Helper function to uncommon-trap or bailout for non-compilable call-sites
444 bool can_not_compile_call_site(ciMethod *dest_method, ciInstanceKlass *klass);
446 // Helper function to identify inlining potential at call-site
447 ciMethod* optimize_inlining(ciMethod* caller, int bci, ciInstanceKlass* klass,
448 ciMethod *dest_method, const TypeOopPtr* receiver_type);
450 // Helper function to setup for type-profile based inlining
451 bool prepare_type_profile_inline(ciInstanceKlass* prof_klass, ciMethod* prof_method);
453 // Helper functions for type checking bytecodes:
454 void do_checkcast();
455 void do_instanceof();
457 // Helper functions for shifting & arithmetic
458 void modf();
459 void modd();
460 void l2f();
462 void do_irem();
464 // implementation of _get* and _put* bytecodes
465 void do_getstatic() { do_field_access(true, false); }
466 void do_getfield () { do_field_access(true, true); }
467 void do_putstatic() { do_field_access(false, false); }
468 void do_putfield () { do_field_access(false, true); }
470 // common code for making initial checks and forming addresses
471 void do_field_access(bool is_get, bool is_field);
472 bool static_field_ok_in_clinit(ciField *field, ciMethod *method);
474 // common code for actually performing the load or store
475 void do_get_xxx(const TypePtr* obj_type, Node* obj, ciField* field, bool is_field);
476 void do_put_xxx(const TypePtr* obj_type, Node* obj, ciField* field, bool is_field);
478 // loading from a constant field or the constant pool
479 // returns false if push failed (non-perm field constants only, not ldcs)
480 bool push_constant(ciConstant con, bool require_constant = false);
482 // implementation of object creation bytecodes
483 void do_new();
484 void do_newarray(BasicType elemtype);
485 void do_anewarray();
486 void do_multianewarray();
487 Node* expand_multianewarray(ciArrayKlass* array_klass, Node* *lengths, int ndimensions, int nargs);
489 // implementation of jsr/ret
490 void do_jsr();
491 void do_ret();
493 float dynamic_branch_prediction(float &cnt);
494 float branch_prediction(float &cnt, BoolTest::mask btest, int target_bci);
495 bool seems_never_taken(float prob);
497 void do_ifnull(BoolTest::mask btest, Node* c);
498 void do_if(BoolTest::mask btest, Node* c);
499 int repush_if_args();
500 void adjust_map_after_if(BoolTest::mask btest, Node* c, float prob,
501 Block* path, Block* other_path);
502 IfNode* jump_if_fork_int(Node* a, Node* b, BoolTest::mask mask);
503 Node* jump_if_join(Node* iffalse, Node* iftrue);
504 void jump_if_true_fork(IfNode *ifNode, int dest_bci_if_true, int prof_table_index);
505 void jump_if_false_fork(IfNode *ifNode, int dest_bci_if_false, int prof_table_index);
506 void jump_if_always_fork(int dest_bci_if_true, int prof_table_index);
508 friend class SwitchRange;
509 void do_tableswitch();
510 void do_lookupswitch();
511 void jump_switch_ranges(Node* a, SwitchRange* lo, SwitchRange* hi, int depth = 0);
512 bool create_jump_tables(Node* a, SwitchRange* lo, SwitchRange* hi);
514 // helper functions for methodData style profiling
515 void test_counter_against_threshold(Node* cnt, int limit);
516 void increment_and_test_invocation_counter(int limit);
517 void test_for_osr_md_counter_at(ciMethodData* md, ciProfileData* data, ByteSize offset, int limit);
518 Node* method_data_addressing(ciMethodData* md, ciProfileData* data, ByteSize offset, Node* idx = NULL, uint stride = 0);
519 void increment_md_counter_at(ciMethodData* md, ciProfileData* data, ByteSize offset, Node* idx = NULL, uint stride = 0);
520 void set_md_flag_at(ciMethodData* md, ciProfileData* data, int flag_constant);
522 void profile_method_entry();
523 void profile_taken_branch(int target_bci, bool force_update = false);
524 void profile_not_taken_branch(bool force_update = false);
525 void profile_call(Node* receiver);
526 void profile_generic_call();
527 void profile_receiver_type(Node* receiver);
528 void profile_ret(int target_bci);
529 void profile_null_checkcast();
530 void profile_switch_case(int table_index);
532 // helper function for call statistics
533 void count_compiled_calls(bool at_method_entry, bool is_inline) PRODUCT_RETURN;
535 Node_Notes* make_node_notes(Node_Notes* caller_nn);
537 // Helper functions for handling normal and abnormal exits.
538 void build_exits();
540 // Fix up all exceptional control flow exiting a single bytecode.
541 void do_exceptions();
543 // Fix up all exiting control flow at the end of the parse.
544 void do_exits();
546 // Add Catch/CatchProjs
547 // The call is either a Java call or the VM's rethrow stub
548 void catch_call_exceptions(ciExceptionHandlerStream&);
550 // Handle all exceptions thrown by the inlined method.
551 // Also handles exceptions for individual bytecodes.
552 void catch_inline_exceptions(SafePointNode* ex_map);
554 // Bytecode classifier, helps decide to use uncommon_trap vs. rethrow_C.
555 bool can_rerun_bytecode();
557 // Merge the given map into correct exceptional exit state.
558 // Assumes that there is no applicable local handler.
559 void throw_to_exit(SafePointNode* ex_map);
561 public:
562 #ifndef PRODUCT
563 // Handle PrintOpto, etc.
564 void show_parse_info();
565 void dump_map_adr_mem() const;
566 static void print_statistics(); // Print some performance counters
567 void dump();
568 void dump_bci(int bci);
569 #endif
570 };