Wed, 07 Sep 2011 09:35:52 +0200
7086394: c2/arm: enable UseFPUForSpilling
Summary: ARM has instructions to move data directly between the fpu and integer registers.
Reviewed-by: kvn, never
1 /*
2 * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #ifndef SHARE_VM_OPTO_PARSE_HPP
26 #define SHARE_VM_OPTO_PARSE_HPP
28 #include "ci/ciMethodData.hpp"
29 #include "ci/ciTypeFlow.hpp"
30 #include "compiler/methodLiveness.hpp"
31 #include "libadt/vectset.hpp"
32 #include "oops/generateOopMap.hpp"
33 #include "opto/graphKit.hpp"
34 #include "opto/subnode.hpp"
36 class BytecodeParseHistogram;
37 class InlineTree;
38 class Parse;
39 class SwitchRange;
42 //------------------------------InlineTree-------------------------------------
43 class InlineTree : public ResourceObj {
44 Compile* C; // cache
45 JVMState* _caller_jvms; // state of caller
46 ciMethod* _method; // method being called by the caller_jvms
47 InlineTree* _caller_tree;
48 uint _count_inline_bcs; // Accumulated count of inlined bytecodes
49 // Call-site count / interpreter invocation count, scaled recursively.
50 // Always between 0.0 and 1.0. Represents the percentage of the method's
51 // total execution time used at this call site.
52 const float _site_invoke_ratio;
53 const int _max_inline_level; // the maximum inline level for this sub-tree (may be adjusted)
54 float compute_callee_frequency( int caller_bci ) const;
56 GrowableArray<InlineTree*> _subtrees;
57 friend class Compile;
59 protected:
60 InlineTree(Compile* C,
61 const InlineTree* caller_tree,
62 ciMethod* callee_method,
63 JVMState* caller_jvms,
64 int caller_bci,
65 float site_invoke_ratio,
66 int max_inline_level);
67 InlineTree *build_inline_tree_for_callee(ciMethod* callee_method,
68 JVMState* caller_jvms,
69 int caller_bci);
70 const char* try_to_inline(ciMethod* callee_method, ciMethod* caller_method, int caller_bci, ciCallProfile& profile, WarmCallInfo* wci_result);
71 const char* should_inline(ciMethod* callee_method, ciMethod* caller_method, int caller_bci, ciCallProfile& profile, WarmCallInfo* wci_result) const;
72 const char* should_not_inline(ciMethod* callee_method, ciMethod* caller_method, WarmCallInfo* wci_result) const;
73 void print_inlining(ciMethod *callee_method, int caller_bci, const char *failure_msg) const;
75 InlineTree *caller_tree() const { return _caller_tree; }
76 InlineTree* callee_at(int bci, ciMethod* m) const;
77 int inline_level() const { return stack_depth(); }
78 int stack_depth() const { return _caller_jvms ? _caller_jvms->depth() : 0; }
80 public:
81 static const char* check_can_parse(ciMethod* callee);
83 static InlineTree* build_inline_tree_root();
84 static InlineTree* find_subtree_from_root(InlineTree* root, JVMState* jvms, ciMethod* callee, bool create_if_not_found = false);
86 // For temporary (stack-allocated, stateless) ilts:
87 InlineTree(Compile* c, ciMethod* callee_method, JVMState* caller_jvms, float site_invoke_ratio, int max_inline_level);
89 // InlineTree enum
90 enum InlineStyle {
91 Inline_do_not_inline = 0, //
92 Inline_cha_is_monomorphic = 1, //
93 Inline_type_profile_monomorphic = 2 //
94 };
96 // See if it is OK to inline.
97 // The receiver is the inline tree for the caller.
98 //
99 // The result is a temperature indication. If it is hot or cold,
100 // inlining is immediate or undesirable. Otherwise, the info block
101 // returned is newly allocated and may be enqueued.
102 //
103 // If the method is inlinable, a new inline subtree is created on the fly,
104 // and may be accessed by find_subtree_from_root.
105 // The call_method is the dest_method for a special or static invocation.
106 // The call_method is an optimized virtual method candidate otherwise.
107 WarmCallInfo* ok_to_inline(ciMethod *call_method, JVMState* caller_jvms, ciCallProfile& profile, WarmCallInfo* wci);
109 // Information about inlined method
110 JVMState* caller_jvms() const { return _caller_jvms; }
111 ciMethod *method() const { return _method; }
112 int caller_bci() const { return _caller_jvms ? _caller_jvms->bci() : InvocationEntryBci; }
113 uint count_inline_bcs() const { return _count_inline_bcs; }
114 float site_invoke_ratio() const { return _site_invoke_ratio; };
116 #ifndef PRODUCT
117 private:
118 uint _count_inlines; // Count of inlined methods
119 public:
120 // Debug information collected during parse
121 uint count_inlines() const { return _count_inlines; };
122 #endif
123 GrowableArray<InlineTree*> subtrees() { return _subtrees; }
124 };
127 //-----------------------------------------------------------------------------
128 //------------------------------Parse------------------------------------------
129 // Parse bytecodes, build a Graph
130 class Parse : public GraphKit {
131 public:
132 // Per-block information needed by the parser:
133 class Block {
134 private:
135 ciTypeFlow::Block* _flow;
136 int _pred_count; // how many predecessors in CFG?
137 int _preds_parsed; // how many of these have been parsed?
138 uint _count; // how many times executed? Currently only set by _goto's
139 bool _is_parsed; // has this block been parsed yet?
140 bool _is_handler; // is this block an exception handler?
141 bool _has_merged_backedge; // does this block have merged backedge?
142 SafePointNode* _start_map; // all values flowing into this block
143 MethodLivenessResult _live_locals; // lazily initialized liveness bitmap
145 int _num_successors; // Includes only normal control flow.
146 int _all_successors; // Include exception paths also.
147 Block** _successors;
149 // Use init_node/init_graph to initialize Blocks.
150 // Block() : _live_locals((uintptr_t*)NULL,0) { ShouldNotReachHere(); }
151 Block() : _live_locals(NULL,0) { ShouldNotReachHere(); }
153 public:
155 // Set up the block data structure itself.
156 void init_node(Parse* outer, int po);
157 // Set up the block's relations to other blocks.
158 void init_graph(Parse* outer);
160 ciTypeFlow::Block* flow() const { return _flow; }
161 int pred_count() const { return _pred_count; }
162 int preds_parsed() const { return _preds_parsed; }
163 bool is_parsed() const { return _is_parsed; }
164 bool is_handler() const { return _is_handler; }
165 void set_count( uint x ) { _count = x; }
166 uint count() const { return _count; }
168 SafePointNode* start_map() const { assert(is_merged(),""); return _start_map; }
169 void set_start_map(SafePointNode* m) { assert(!is_merged(), ""); _start_map = m; }
171 // True after any predecessor flows control into this block
172 bool is_merged() const { return _start_map != NULL; }
174 #ifdef ASSERT
175 // True after backedge predecessor flows control into this block
176 bool has_merged_backedge() const { return _has_merged_backedge; }
177 void mark_merged_backedge(Block* pred) {
178 assert(is_SEL_head(), "should be loop head");
179 if (pred != NULL && is_SEL_backedge(pred)) {
180 assert(is_parsed(), "block should be parsed before merging backedges");
181 _has_merged_backedge = true;
182 }
183 }
184 #endif
186 // True when all non-exception predecessors have been parsed.
187 bool is_ready() const { return preds_parsed() == pred_count(); }
189 int num_successors() const { return _num_successors; }
190 int all_successors() const { return _all_successors; }
191 Block* successor_at(int i) const {
192 assert((uint)i < (uint)all_successors(), "");
193 return _successors[i];
194 }
195 Block* successor_for_bci(int bci);
197 int start() const { return flow()->start(); }
198 int limit() const { return flow()->limit(); }
199 int rpo() const { return flow()->rpo(); }
200 int start_sp() const { return flow()->stack_size(); }
202 bool is_loop_head() const { return flow()->is_loop_head(); }
203 bool is_SEL_head() const { return flow()->is_single_entry_loop_head(); }
204 bool is_SEL_backedge(Block* pred) const{ return is_SEL_head() && pred->rpo() >= rpo(); }
205 bool is_invariant_local(uint i) const {
206 const JVMState* jvms = start_map()->jvms();
207 if (!jvms->is_loc(i) || flow()->outer()->has_irreducible_entry()) return false;
208 return flow()->is_invariant_local(i - jvms->locoff());
209 }
210 bool can_elide_SEL_phi(uint i) const { assert(is_SEL_head(),""); return is_invariant_local(i); }
212 const Type* peek(int off=0) const { return stack_type_at(start_sp() - (off+1)); }
214 const Type* stack_type_at(int i) const;
215 const Type* local_type_at(int i) const;
216 static const Type* get_type(ciType* t) { return Type::get_typeflow_type(t); }
218 bool has_trap_at(int bci) const { return flow()->has_trap() && flow()->trap_bci() == bci; }
220 // Call this just before parsing a block.
221 void mark_parsed() {
222 assert(!_is_parsed, "must parse each block exactly once");
223 _is_parsed = true;
224 }
226 // Return the phi/region input index for the "current" pred,
227 // and bump the pred number. For historical reasons these index
228 // numbers are handed out in descending order. The last index is
229 // always PhiNode::Input (i.e., 1). The value returned is known
230 // as a "path number" because it distinguishes by which path we are
231 // entering the block.
232 int next_path_num() {
233 assert(preds_parsed() < pred_count(), "too many preds?");
234 return pred_count() - _preds_parsed++;
235 }
237 // Add a previously unaccounted predecessor to this block.
238 // This operates by increasing the size of the block's region
239 // and all its phi nodes (if any). The value returned is a
240 // path number ("pnum").
241 int add_new_path();
243 // Initialize me by recording the parser's map. My own map must be NULL.
244 void record_state(Parse* outer);
245 };
247 #ifndef PRODUCT
248 // BytecodeParseHistogram collects number of bytecodes parsed, nodes constructed, and transformations.
249 class BytecodeParseHistogram : public ResourceObj {
250 private:
251 enum BPHType {
252 BPH_transforms,
253 BPH_values
254 };
255 static bool _initialized;
256 static uint _bytecodes_parsed [Bytecodes::number_of_codes];
257 static uint _nodes_constructed[Bytecodes::number_of_codes];
258 static uint _nodes_transformed[Bytecodes::number_of_codes];
259 static uint _new_values [Bytecodes::number_of_codes];
261 Bytecodes::Code _initial_bytecode;
262 int _initial_node_count;
263 int _initial_transforms;
264 int _initial_values;
266 Parse *_parser;
267 Compile *_compiler;
269 // Initialization
270 static void reset();
272 // Return info being collected, select with global flag 'BytecodeParseInfo'
273 int current_count(BPHType info_selector);
275 public:
276 BytecodeParseHistogram(Parse *p, Compile *c);
277 static bool initialized();
279 // Record info when starting to parse one bytecode
280 void set_initial_state( Bytecodes::Code bc );
281 // Record results of parsing one bytecode
282 void record_change();
284 // Profile printing
285 static void print(float cutoff = 0.01F); // cutoff in percent
286 };
288 public:
289 // Record work done during parsing
290 BytecodeParseHistogram* _parse_histogram;
291 void set_parse_histogram(BytecodeParseHistogram *bph) { _parse_histogram = bph; }
292 BytecodeParseHistogram* parse_histogram() { return _parse_histogram; }
293 #endif
295 private:
296 friend class Block;
298 // Variables which characterize this compilation as a whole:
300 JVMState* _caller; // JVMS which carries incoming args & state.
301 float _expected_uses; // expected number of calls to this code
302 float _prof_factor; // discount applied to my profile counts
303 int _depth; // Inline tree depth, for debug printouts
304 const TypeFunc*_tf; // My kind of function type
305 int _entry_bci; // the osr bci or InvocationEntryBci
307 ciTypeFlow* _flow; // Results of previous flow pass.
308 Block* _blocks; // Array of basic-block structs.
309 int _block_count; // Number of elements in _blocks.
311 GraphKit _exits; // Record all normal returns and throws here.
312 bool _wrote_final; // Did we write a final field?
313 bool _count_invocations; // update and test invocation counter
314 bool _method_data_update; // update method data oop
316 // Variables which track Java semantics during bytecode parsing:
318 Block* _block; // block currently getting parsed
319 ciBytecodeStream _iter; // stream of this method's bytecodes
321 int _blocks_merged; // Progress meter: state merges from BB preds
322 int _blocks_parsed; // Progress meter: BBs actually parsed
324 const FastLockNode* _synch_lock; // FastLockNode for synchronized method
326 #ifndef PRODUCT
327 int _max_switch_depth; // Debugging SwitchRanges.
328 int _est_switch_depth; // Debugging SwitchRanges.
329 #endif
331 public:
332 // Constructor
333 Parse(JVMState* caller, ciMethod* parse_method, float expected_uses);
335 virtual Parse* is_Parse() const { return (Parse*)this; }
337 public:
338 // Accessors.
339 JVMState* caller() const { return _caller; }
340 float expected_uses() const { return _expected_uses; }
341 float prof_factor() const { return _prof_factor; }
342 int depth() const { return _depth; }
343 const TypeFunc* tf() const { return _tf; }
344 // entry_bci() -- see osr_bci, etc.
346 ciTypeFlow* flow() const { return _flow; }
347 // blocks() -- see rpo_at, start_block, etc.
348 int block_count() const { return _block_count; }
350 GraphKit& exits() { return _exits; }
351 bool wrote_final() const { return _wrote_final; }
352 void set_wrote_final(bool z) { _wrote_final = z; }
353 bool count_invocations() const { return _count_invocations; }
354 bool method_data_update() const { return _method_data_update; }
356 Block* block() const { return _block; }
357 ciBytecodeStream& iter() { return _iter; }
358 Bytecodes::Code bc() const { return _iter.cur_bc(); }
360 void set_block(Block* b) { _block = b; }
362 // Derived accessors:
363 bool is_normal_parse() const { return _entry_bci == InvocationEntryBci; }
364 bool is_osr_parse() const { return _entry_bci != InvocationEntryBci; }
365 int osr_bci() const { assert(is_osr_parse(),""); return _entry_bci; }
367 void set_parse_bci(int bci);
369 // Must this parse be aborted?
370 bool failing() { return C->failing(); }
372 Block* rpo_at(int rpo) {
373 assert(0 <= rpo && rpo < _block_count, "oob");
374 return &_blocks[rpo];
375 }
376 Block* start_block() {
377 return rpo_at(flow()->start_block()->rpo());
378 }
379 // Can return NULL if the flow pass did not complete a block.
380 Block* successor_for_bci(int bci) {
381 return block()->successor_for_bci(bci);
382 }
384 private:
385 // Create a JVMS & map for the initial state of this method.
386 SafePointNode* create_entry_map();
388 // OSR helpers
389 Node *fetch_interpreter_state(int index, BasicType bt, Node *local_addrs, Node *local_addrs_base);
390 Node* check_interpreter_type(Node* l, const Type* type, SafePointNode* &bad_type_exit);
391 void load_interpreter_state(Node* osr_buf);
393 // Functions for managing basic blocks:
394 void init_blocks();
395 void load_state_from(Block* b);
396 void store_state_to(Block* b) { b->record_state(this); }
398 // Parse all the basic blocks.
399 void do_all_blocks();
401 // Parse the current basic block
402 void do_one_block();
404 // Raise an error if we get a bad ciTypeFlow CFG.
405 void handle_missing_successor(int bci);
407 // first actions (before BCI 0)
408 void do_method_entry();
410 // implementation of monitorenter/monitorexit
411 void do_monitor_enter();
412 void do_monitor_exit();
414 // Eagerly create phie throughout the state, to cope with back edges.
415 void ensure_phis_everywhere();
417 // Merge the current mapping into the basic block starting at bci
418 void merge( int target_bci);
419 // Same as plain merge, except that it allocates a new path number.
420 void merge_new_path( int target_bci);
421 // Merge the current mapping into an exception handler.
422 void merge_exception(int target_bci);
423 // Helper: Merge the current mapping into the given basic block
424 void merge_common(Block* target, int pnum);
425 // Helper functions for merging individual cells.
426 PhiNode *ensure_phi( int idx, bool nocreate = false);
427 PhiNode *ensure_memory_phi(int idx, bool nocreate = false);
428 // Helper to merge the current memory state into the given basic block
429 void merge_memory_edges(MergeMemNode* n, int pnum, bool nophi);
431 // Parse this bytecode, and alter the Parsers JVM->Node mapping
432 void do_one_bytecode();
434 // helper function to generate array store check
435 void array_store_check();
436 // Helper function to generate array load
437 void array_load(BasicType etype);
438 // Helper function to generate array store
439 void array_store(BasicType etype);
440 // Helper function to compute array addressing
441 Node* array_addressing(BasicType type, int vals, const Type* *result2=NULL);
443 // Pass current map to exits
444 void return_current(Node* value);
446 // Register finalizers on return from Object.<init>
447 void call_register_finalizer();
449 // Insert a compiler safepoint into the graph
450 void add_safepoint();
452 // Insert a compiler safepoint into the graph, if there is a back-branch.
453 void maybe_add_safepoint(int target_bci) {
454 if (UseLoopSafepoints && target_bci <= bci()) {
455 add_safepoint();
456 }
457 }
459 // Note: Intrinsic generation routines may be found in library_call.cpp.
461 // Helper function to setup Ideal Call nodes
462 void do_call();
464 // Helper function to uncommon-trap or bailout for non-compilable call-sites
465 bool can_not_compile_call_site(ciMethod *dest_method, ciInstanceKlass *klass);
467 // Helper function to identify inlining potential at call-site
468 ciMethod* optimize_inlining(ciMethod* caller, int bci, ciInstanceKlass* klass,
469 ciMethod *dest_method, const TypeOopPtr* receiver_type);
471 // Helper function to setup for type-profile based inlining
472 bool prepare_type_profile_inline(ciInstanceKlass* prof_klass, ciMethod* prof_method);
474 // Helper functions for type checking bytecodes:
475 void do_checkcast();
476 void do_instanceof();
478 // Helper functions for shifting & arithmetic
479 void modf();
480 void modd();
481 void l2f();
483 void do_irem();
485 // implementation of _get* and _put* bytecodes
486 void do_getstatic() { do_field_access(true, false); }
487 void do_getfield () { do_field_access(true, true); }
488 void do_putstatic() { do_field_access(false, false); }
489 void do_putfield () { do_field_access(false, true); }
491 // common code for making initial checks and forming addresses
492 void do_field_access(bool is_get, bool is_field);
493 bool static_field_ok_in_clinit(ciField *field, ciMethod *method);
495 // common code for actually performing the load or store
496 void do_get_xxx(Node* obj, ciField* field, bool is_field);
497 void do_put_xxx(Node* obj, ciField* field, bool is_field);
499 // loading from a constant field or the constant pool
500 // returns false if push failed (non-perm field constants only, not ldcs)
501 bool push_constant(ciConstant con, bool require_constant = false);
503 // implementation of object creation bytecodes
504 void emit_guard_for_new(ciInstanceKlass* klass);
505 void do_new();
506 void do_newarray(BasicType elemtype);
507 void do_anewarray();
508 void do_multianewarray();
509 Node* expand_multianewarray(ciArrayKlass* array_klass, Node* *lengths, int ndimensions, int nargs);
511 // implementation of jsr/ret
512 void do_jsr();
513 void do_ret();
515 float dynamic_branch_prediction(float &cnt);
516 float branch_prediction(float &cnt, BoolTest::mask btest, int target_bci);
517 bool seems_never_taken(float prob);
518 bool seems_stable_comparison(BoolTest::mask btest, Node* c);
520 void do_ifnull(BoolTest::mask btest, Node* c);
521 void do_if(BoolTest::mask btest, Node* c);
522 int repush_if_args();
523 void adjust_map_after_if(BoolTest::mask btest, Node* c, float prob,
524 Block* path, Block* other_path);
525 IfNode* jump_if_fork_int(Node* a, Node* b, BoolTest::mask mask);
526 Node* jump_if_join(Node* iffalse, Node* iftrue);
527 void jump_if_true_fork(IfNode *ifNode, int dest_bci_if_true, int prof_table_index);
528 void jump_if_false_fork(IfNode *ifNode, int dest_bci_if_false, int prof_table_index);
529 void jump_if_always_fork(int dest_bci_if_true, int prof_table_index);
531 friend class SwitchRange;
532 void do_tableswitch();
533 void do_lookupswitch();
534 void jump_switch_ranges(Node* a, SwitchRange* lo, SwitchRange* hi, int depth = 0);
535 bool create_jump_tables(Node* a, SwitchRange* lo, SwitchRange* hi);
537 // helper functions for methodData style profiling
538 void test_counter_against_threshold(Node* cnt, int limit);
539 void increment_and_test_invocation_counter(int limit);
540 void test_for_osr_md_counter_at(ciMethodData* md, ciProfileData* data, ByteSize offset, int limit);
541 Node* method_data_addressing(ciMethodData* md, ciProfileData* data, ByteSize offset, Node* idx = NULL, uint stride = 0);
542 void increment_md_counter_at(ciMethodData* md, ciProfileData* data, ByteSize offset, Node* idx = NULL, uint stride = 0);
543 void set_md_flag_at(ciMethodData* md, ciProfileData* data, int flag_constant);
545 void profile_method_entry();
546 void profile_taken_branch(int target_bci, bool force_update = false);
547 void profile_not_taken_branch(bool force_update = false);
548 void profile_call(Node* receiver);
549 void profile_generic_call();
550 void profile_receiver_type(Node* receiver);
551 void profile_ret(int target_bci);
552 void profile_null_checkcast();
553 void profile_switch_case(int table_index);
555 // helper function for call statistics
556 void count_compiled_calls(bool at_method_entry, bool is_inline) PRODUCT_RETURN;
558 Node_Notes* make_node_notes(Node_Notes* caller_nn);
560 // Helper functions for handling normal and abnormal exits.
561 void build_exits();
563 // Fix up all exceptional control flow exiting a single bytecode.
564 void do_exceptions();
566 // Fix up all exiting control flow at the end of the parse.
567 void do_exits();
569 // Add Catch/CatchProjs
570 // The call is either a Java call or the VM's rethrow stub
571 void catch_call_exceptions(ciExceptionHandlerStream&);
573 // Handle all exceptions thrown by the inlined method.
574 // Also handles exceptions for individual bytecodes.
575 void catch_inline_exceptions(SafePointNode* ex_map);
577 // Merge the given map into correct exceptional exit state.
578 // Assumes that there is no applicable local handler.
579 void throw_to_exit(SafePointNode* ex_map);
581 public:
582 #ifndef PRODUCT
583 // Handle PrintOpto, etc.
584 void show_parse_info();
585 void dump_map_adr_mem() const;
586 static void print_statistics(); // Print some performance counters
587 void dump();
588 void dump_bci(int bci);
589 #endif
590 };
592 #endif // SHARE_VM_OPTO_PARSE_HPP