1 /*
2 * Copyright (c) 1997, 2022, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #ifndef SHARE_OPTO_COMPILE_HPP
26 #define SHARE_OPTO_COMPILE_HPP
27
28 #include "asm/codeBuffer.hpp"
29 #include "ci/compilerInterface.hpp"
30 #include "code/debugInfoRec.hpp"
31 #include "compiler/compiler_globals.hpp"
32 #include "compiler/compilerOracle.hpp"
33 #include "compiler/compileBroker.hpp"
34 #include "compiler/compilerEvent.hpp"
35 #include "libadt/dict.hpp"
36 #include "libadt/vectset.hpp"
37 #include "memory/resourceArea.hpp"
38 #include "oops/methodData.hpp"
39 #include "opto/idealGraphPrinter.hpp"
40 #include "opto/phasetype.hpp"
41 #include "opto/phase.hpp"
42 #include "opto/regmask.hpp"
43 #include "runtime/deoptimization.hpp"
44 #include "runtime/sharedRuntime.hpp"
45 #include "runtime/timerTrace.hpp"
46 #include "runtime/vmThread.hpp"
47 #include "utilities/ticks.hpp"
48
49 class AbstractLockNode;
50 class AddPNode;
51 class Block;
52 class Bundle;
53 class CallGenerator;
54 class CallNode;
55 class CallStaticJavaNode;
56 class CloneMap;
57 class ConnectionGraph;
58 class IdealGraphPrinter;
59 class InlineTree;
60 class Int_Array;
61 class Matcher;
62 class MachConstantNode;
63 class MachConstantBaseNode;
64 class MachNode;
65 class MachOper;
66 class MachSafePointNode;
67 class Node;
68 class Node_Array;
69 class Node_List;
70 class Node_Notes;
71 class NodeCloneInfo;
72 class OptoReg;
73 class PhaseCFG;
74 class PhaseGVN;
75 class PhaseIterGVN;
76 class PhaseRegAlloc;
77 class PhaseCCP;
78 class PhaseOutput;
79 class RootNode;
80 class relocInfo;
81 class Scope;
82 class StartNode;
83 class SafePointNode;
84 class JVMState;
85 class Type;
86 class TypeData;
87 class TypeInt;
88 class TypeInteger;
89 class TypeKlassPtr;
90 class TypePtr;
91 class TypeOopPtr;
92 class TypeFunc;
93 class TypeVect;
94 class Unique_Node_List;
95 class UnstableIfTrap;
96 class InlineTypeBaseNode;
97 class nmethod;
98 class Node_Stack;
99 struct Final_Reshape_Counts;
100
101 enum LoopOptsMode {
102 LoopOptsDefault,
103 LoopOptsNone,
104 LoopOptsMaxUnroll,
105 LoopOptsShenandoahExpand,
106 LoopOptsShenandoahPostExpand,
107 LoopOptsSkipSplitIf,
108 LoopOptsVerify
109 };
110
111 // The type of all node counts and indexes.
112 // It must hold at least 16 bits, but must also be fast to load and store.
113 // This type, if less than 32 bits, could limit the number of possible nodes.
114 // (To make this type platform-specific, move to globalDefinitions_xxx.hpp.)
115 typedef unsigned int node_idx_t;
116
117 class NodeCloneInfo {
118 private:
119 uint64_t _idx_clone_orig;
120 public:
121
122 void set_idx(node_idx_t idx) {
123 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx;
124 }
125 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); }
126
127 void set_gen(int generation) {
128 uint64_t g = (uint64_t)generation << 32;
129 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g;
130 }
131 int gen() const { return (int)(_idx_clone_orig >> 32); }
132
133 void set(uint64_t x) { _idx_clone_orig = x; }
134 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); }
135 uint64_t get() const { return _idx_clone_orig; }
136
137 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {}
138 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); }
139
140 void dump() const;
141 };
142
143 class CloneMap {
144 friend class Compile;
145 private:
146 bool _debug;
147 Dict* _dict;
148 int _clone_idx; // current cloning iteration/generation in loop unroll
149 public:
150 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy
151 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; }
152 Dict* dict() const { return _dict; }
153 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == NULL, "key existed"); _dict->Insert(_2p(key), (void*)val); }
154 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); }
155 void remove(node_idx_t key) { _dict->Delete(_2p(key)); }
156 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); }
157 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); }
158 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); }
159 int gen(const void* k) const { return gen(_2_node_idx_t(k)); }
160 int max_gen() const;
161 void clone(Node* old, Node* nnn, int gen);
162 void verify_insert_and_clone(Node* old, Node* nnn, int gen);
163 void dump(node_idx_t key) const;
164
165 int clone_idx() const { return _clone_idx; }
166 void set_clone_idx(int x) { _clone_idx = x; }
167 bool is_debug() const { return _debug; }
168 void set_debug(bool debug) { _debug = debug; }
169 static const char* debug_option_name;
170
171 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); }
172 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); }
173 };
174
175 class Options {
176 friend class Compile;
177 friend class VMStructs;
178 private:
179 const bool _subsume_loads; // Load can be matched as part of a larger op.
180 const bool _do_escape_analysis; // Do escape analysis.
181 const bool _do_iterative_escape_analysis; // Do iterative escape analysis.
182 const bool _eliminate_boxing; // Do boxing elimination.
183 const bool _do_locks_coarsening; // Do locks coarsening
184 const bool _install_code; // Install the code that was compiled
185 public:
186 Options(bool subsume_loads, bool do_escape_analysis,
187 bool do_iterative_escape_analysis,
188 bool eliminate_boxing, bool do_locks_coarsening,
189 bool install_code) :
190 _subsume_loads(subsume_loads),
191 _do_escape_analysis(do_escape_analysis),
192 _do_iterative_escape_analysis(do_iterative_escape_analysis),
193 _eliminate_boxing(eliminate_boxing),
194 _do_locks_coarsening(do_locks_coarsening),
195 _install_code(install_code) {
196 }
197
198 static Options for_runtime_stub() {
199 return Options(
200 /* subsume_loads = */ true,
201 /* do_escape_analysis = */ false,
202 /* do_iterative_escape_analysis = */ false,
203 /* eliminate_boxing = */ false,
204 /* do_lock_coarsening = */ false,
205 /* install_code = */ true
206 );
207 }
208 };
209
210 //------------------------------Compile----------------------------------------
211 // This class defines a top-level Compiler invocation.
212
213 class Compile : public Phase {
214 friend class VMStructs;
215
216 public:
217 // Fixed alias indexes. (See also MergeMemNode.)
218 enum {
219 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value)
220 AliasIdxBot = 2, // pseudo-index, aliases to everything
221 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM
222 };
223
224 // Variant of TraceTime(NULL, &_t_accumulator, CITime);
225 // Integrated with logging. If logging is turned on, and CITimeVerbose is true,
226 // then brackets are put into the log, with time stamps and node counts.
227 // (The time collection itself is always conditionalized on CITime.)
228 class TracePhase : public TraceTime {
229 private:
230 Compile* C;
231 CompileLog* _log;
232 const char* _phase_name;
233 bool _dolog;
234 public:
235 TracePhase(const char* name, elapsedTimer* accumulator);
236 ~TracePhase();
237 };
238
239 // Information per category of alias (memory slice)
240 class AliasType {
241 private:
242 friend class Compile;
243
244 int _index; // unique index, used with MergeMemNode
245 const TypePtr* _adr_type; // normalized address type
246 ciField* _field; // relevant instance field, or null if none
247 const Type* _element; // relevant array element type, or null if none
248 bool _is_rewritable; // false if the memory is write-once only
249 int _general_index; // if this is type is an instance, the general
250 // type that this is an instance of
251
252 void Init(int i, const TypePtr* at);
253
254 public:
255 int index() const { return _index; }
256 const TypePtr* adr_type() const { return _adr_type; }
257 ciField* field() const { return _field; }
258 const Type* element() const { return _element; }
259 bool is_rewritable() const { return _is_rewritable; }
260 bool is_volatile() const { return (_field ? _field->is_volatile() : false); }
261 int general_index() const { return (_general_index != 0) ? _general_index : _index; }
262
263 void set_rewritable(bool z) { _is_rewritable = z; }
264 void set_field(ciField* f) {
265 assert(!_field,"");
266 _field = f;
267 if (f->is_final() || f->is_stable()) {
268 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
269 _is_rewritable = false;
270 }
271 }
272 void set_element(const Type* e) {
273 assert(_element == NULL, "");
274 _element = e;
275 }
276
277 BasicType basic_type() const;
278
279 void print_on(outputStream* st) PRODUCT_RETURN;
280 };
281
282 enum {
283 logAliasCacheSize = 6,
284 AliasCacheSize = (1<<logAliasCacheSize)
285 };
286 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type
287 enum {
288 trapHistLength = MethodData::_trap_hist_limit
289 };
290
291 private:
292 // Fixed parameters to this compilation.
293 const int _compile_id;
294 const Options _options; // Compilation options
295 ciMethod* _method; // The method being compiled.
296 int _entry_bci; // entry bci for osr methods.
297 const TypeFunc* _tf; // My kind of signature
298 InlineTree* _ilt; // Ditto (temporary).
299 address _stub_function; // VM entry for stub being compiled, or NULL
300 const char* _stub_name; // Name of stub or adapter being compiled, or NULL
301 address _stub_entry_point; // Compile code entry for generated stub, or NULL
302
303 // Control of this compilation.
304 int _max_inline_size; // Max inline size for this compilation
305 int _freq_inline_size; // Max hot method inline size for this compilation
306 int _fixed_slots; // count of frame slots not allocated by the register
307 // allocator i.e. locks, original deopt pc, etc.
308 uintx _max_node_limit; // Max unique node count during a single compilation.
309
310 bool _post_loop_opts_phase; // Loop opts are finished.
311
312 int _major_progress; // Count of something big happening
313 bool _inlining_progress; // progress doing incremental inlining?
314 bool _inlining_incrementally;// Are we doing incremental inlining (post parse)
315 bool _do_cleanup; // Cleanup is needed before proceeding with incremental inlining
316 bool _has_loops; // True if the method _may_ have some loops
317 bool _has_split_ifs; // True if the method _may_ have some split-if
318 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores.
319 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated
320 bool _has_boxed_value; // True if a boxed object is allocated
321 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess
322 uint _max_vector_size; // Maximum size of generated vectors
323 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper
324 uint _trap_hist[trapHistLength]; // Cumulative traps
325 bool _trap_can_recompile; // Have we emitted a recompiling trap?
326 uint _decompile_count; // Cumulative decompilation counts.
327 bool _do_inlining; // True if we intend to do inlining
328 bool _do_scheduling; // True if we intend to do scheduling
329 bool _do_freq_based_layout; // True if we intend to do frequency based block layout
330 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations
331 bool _use_cmove; // True if CMove should be used without profitability analysis
332 int _AliasLevel; // Locally-adjusted version of AliasLevel flag.
333 bool _print_assembly; // True if we should dump assembly code for this compilation
334 bool _print_inlining; // True if we should print inlining for this compilation
335 bool _print_intrinsics; // True if we should print intrinsics for this compilation
336 #ifndef PRODUCT
337 uint _igv_idx; // Counter for IGV node identifiers
338 bool _trace_opto_output;
339 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
340 #endif
341 bool _has_irreducible_loop; // Found irreducible loops
342 // JSR 292
343 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes.
344 bool _has_monitors; // Metadata transfered to nmethod to enable Continuations lock-detection fastpath
345 RTMState _rtm_state; // State of Restricted Transactional Memory usage
346 int _loop_opts_cnt; // loop opts round
347 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry
348 bool _has_flattened_accesses; // Any known flattened array accesses?
349 bool _flattened_accesses_share_alias; // Initially all flattened array share a single slice
350 bool _scalarize_in_safepoints; // Scalarize inline types in safepoint debug info
351 uint _stress_seed; // Seed for stress testing
352
353 // Compilation environment.
354 Arena _comp_arena; // Arena with lifetime equivalent to Compile
355 void* _barrier_set_state; // Potential GC barrier state for Compile
356 ciEnv* _env; // CI interface
357 DirectiveSet* _directive; // Compiler directive
358 CompileLog* _log; // from CompilerThread
359 const char* _failure_reason; // for record_failure/failing pattern
360 GrowableArray<CallGenerator*> _intrinsics; // List of intrinsics.
361 GrowableArray<Node*> _macro_nodes; // List of nodes which need to be expanded before matching.
362 GrowableArray<Node*> _predicate_opaqs; // List of Opaque1 nodes for the loop predicates.
363 GrowableArray<Node*> _skeleton_predicate_opaqs; // List of Opaque4 nodes for the loop skeleton predicates.
364 GrowableArray<Node*> _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
365 GrowableArray<Node*> _for_post_loop_igvn; // List of nodes for IGVN after loop opts are over
366 GrowableArray<Node*> _inline_type_nodes; // List of InlineType nodes
367 GrowableArray<UnstableIfTrap*> _unstable_if_traps; // List of ifnodes after IGVN
368 GrowableArray<Node_List*> _coarsened_locks; // List of coarsened Lock and Unlock nodes
369 ConnectionGraph* _congraph;
370 #ifndef PRODUCT
371 IdealGraphPrinter* _igv_printer;
372 static IdealGraphPrinter* _debug_file_printer;
373 static IdealGraphPrinter* _debug_network_printer;
374 #endif
375
376
377 // Node management
378 uint _unique; // Counter for unique Node indices
379 VectorSet _dead_node_list; // Set of dead nodes
380 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N).
381 // So use this to keep count and make the call O(1).
382 DEBUG_ONLY(Unique_Node_List* _modified_nodes;) // List of nodes which inputs were modified
383 DEBUG_ONLY(bool _phase_optimize_finished;) // Used for live node verification while creating new nodes
384
385 debug_only(static int _debug_idx;) // Monotonic counter (not reset), use -XX:BreakAtNode=<idx>
386 Arena _node_arena; // Arena for new-space Nodes
387 Arena _old_arena; // Arena for old-space Nodes, lifetime during xform
388 RootNode* _root; // Unique root of compilation, or NULL after bail-out.
389 Node* _top; // Unique top node. (Reset by various phases.)
390
391 Node* _immutable_memory; // Initial memory state
392
393 Node* _recent_alloc_obj;
394 Node* _recent_alloc_ctl;
395
396 // Constant table
397 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton.
398
399
400 // Blocked array of debugging and profiling information,
401 // tracked per node.
402 enum { _log2_node_notes_block_size = 8,
403 _node_notes_block_size = (1<<_log2_node_notes_block_size)
404 };
405 GrowableArray<Node_Notes*>* _node_note_array;
406 Node_Notes* _default_node_notes; // default notes for new nodes
407
408 // After parsing and every bulk phase we hang onto the Root instruction.
409 // The RootNode instruction is where the whole program begins. It produces
410 // the initial Control and BOTTOM for everybody else.
411
412 // Type management
413 Arena _Compile_types; // Arena for all types
414 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared()
415 Dict* _type_dict; // Intern table
416 CloneMap _clone_map; // used for recording history of cloned nodes
417 size_t _type_last_size; // Last allocation size (see Type::operator new/delete)
418 ciMethod* _last_tf_m; // Cache for
419 const TypeFunc* _last_tf; // TypeFunc::make
420 AliasType** _alias_types; // List of alias types seen so far.
421 int _num_alias_types; // Logical length of _alias_types
422 int _max_alias_types; // Physical length of _alias_types
423 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
424
425 // Parsing, optimization
426 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN
427 Unique_Node_List* _for_igvn; // Initial work-list for next round of Iterative GVN
428
429 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after main parsing has finished.
430 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
431 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
432
433 GrowableArray<CallGenerator*> _vector_reboxing_late_inlines; // same but for vector reboxing operations
434
435 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
436 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending
437
438 // Inlining may not happen in parse order which would make
439 // PrintInlining output confusing. Keep track of PrintInlining
440 // pieces in order.
441 class PrintInliningBuffer : public CHeapObj<mtCompiler> {
442 private:
443 CallGenerator* _cg;
444 stringStream _ss;
445 static const size_t default_stream_buffer_size = 128;
446
447 public:
448 PrintInliningBuffer()
449 : _cg(NULL), _ss(default_stream_buffer_size) {}
450
451 stringStream* ss() { return &_ss; }
452 CallGenerator* cg() { return _cg; }
453 void set_cg(CallGenerator* cg) { _cg = cg; }
454 };
455
456 stringStream* _print_inlining_stream;
457 GrowableArray<PrintInliningBuffer*>* _print_inlining_list;
458 int _print_inlining_idx;
459 char* _print_inlining_output;
460
461 // Only keep nodes in the expensive node list that need to be optimized
462 void cleanup_expensive_nodes(PhaseIterGVN &igvn);
463 // Use for sorting expensive nodes to bring similar nodes together
464 static int cmp_expensive_nodes(Node** n1, Node** n2);
465 // Expensive nodes list already sorted?
466 bool expensive_nodes_sorted() const;
467 // Remove the speculative part of types and clean up the graph
468 void remove_speculative_types(PhaseIterGVN &igvn);
469
470 void* _replay_inline_data; // Pointer to data loaded from file
471
472 void print_inlining_stream_free();
473 void print_inlining_init();
474 void print_inlining_reinit();
475 void print_inlining_commit();
476 void print_inlining_push();
477 PrintInliningBuffer* print_inlining_current();
478
479 void log_late_inline_failure(CallGenerator* cg, const char* msg);
480 DEBUG_ONLY(bool _exception_backedge;)
481
482 public:
483
484 void* barrier_set_state() const { return _barrier_set_state; }
485
486 outputStream* print_inlining_stream() const {
487 assert(print_inlining() || print_intrinsics(), "PrintInlining off?");
488 return _print_inlining_stream;
489 }
490
491 void print_inlining_update(CallGenerator* cg);
492 void print_inlining_update_delayed(CallGenerator* cg);
493 void print_inlining_move_to(CallGenerator* cg);
494 void print_inlining_assert_ready();
495 void print_inlining_reset();
496
497 void print_inlining(ciMethod* method, int inline_level, int bci, const char* msg = NULL) {
498 stringStream ss;
499 CompileTask::print_inlining_inner(&ss, method, inline_level, bci, msg);
500 print_inlining_stream()->print("%s", ss.freeze());
501 }
502
503 #ifndef PRODUCT
504 IdealGraphPrinter* igv_printer() { return _igv_printer; }
505 #endif
506
507 void log_late_inline(CallGenerator* cg);
508 void log_inline_id(CallGenerator* cg);
509 void log_inline_failure(const char* msg);
510
511 void* replay_inline_data() const { return _replay_inline_data; }
512
513 // Dump inlining replay data to the stream.
514 void dump_inline_data(outputStream* out);
515 void dump_inline_data_reduced(outputStream* out);
516
517 private:
518 // Matching, CFG layout, allocation, code generation
519 PhaseCFG* _cfg; // Results of CFG finding
520 int _java_calls; // Number of java calls in the method
521 int _inner_loops; // Number of inner loops in the method
522 Matcher* _matcher; // Engine to map ideal to machine instructions
523 PhaseRegAlloc* _regalloc; // Results of register allocation.
524 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout)
525 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin
526 void* _indexSet_free_block_list; // free list of IndexSet bit blocks
527 int _interpreter_frame_size;
528
529 PhaseOutput* _output;
530
531 public:
532 // Accessors
533
534 // The Compile instance currently active in this (compiler) thread.
535 static Compile* current() {
536 return (Compile*) ciEnv::current()->compiler_data();
537 }
538
539 int interpreter_frame_size() const { return _interpreter_frame_size; }
540
541 PhaseOutput* output() const { return _output; }
542 void set_output(PhaseOutput* o) { _output = o; }
543
544 // ID for this compilation. Useful for setting breakpoints in the debugger.
545 int compile_id() const { return _compile_id; }
546 DirectiveSet* directive() const { return _directive; }
547
548 // Does this compilation allow instructions to subsume loads? User
549 // instructions that subsume a load may result in an unschedulable
550 // instruction sequence.
551 bool subsume_loads() const { return _options._subsume_loads; }
552 /** Do escape analysis. */
553 bool do_escape_analysis() const { return _options._do_escape_analysis; }
554 bool do_iterative_escape_analysis() const { return _options._do_iterative_escape_analysis; }
555 /** Do boxing elimination. */
556 bool eliminate_boxing() const { return _options._eliminate_boxing; }
557 /** Do aggressive boxing elimination. */
558 bool aggressive_unboxing() const { return _options._eliminate_boxing && AggressiveUnboxing; }
559 bool should_install_code() const { return _options._install_code; }
560 /** Do locks coarsening. */
561 bool do_locks_coarsening() const { return _options._do_locks_coarsening; }
562
563 // Other fixed compilation parameters.
564 ciMethod* method() const { return _method; }
565 int entry_bci() const { return _entry_bci; }
566 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; }
567 bool is_method_compilation() const { return (_method != NULL && !_method->flags().is_native()); }
568 const TypeFunc* tf() const { assert(_tf!=NULL, ""); return _tf; }
569 void init_tf(const TypeFunc* tf) { assert(_tf==NULL, ""); _tf = tf; }
570 InlineTree* ilt() const { return _ilt; }
571 address stub_function() const { return _stub_function; }
572 const char* stub_name() const { return _stub_name; }
573 address stub_entry_point() const { return _stub_entry_point; }
574 void set_stub_entry_point(address z) { _stub_entry_point = z; }
575
576 // Control of this compilation.
577 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; }
578 void set_fixed_slots(int n) { _fixed_slots = n; }
579 int major_progress() const { return _major_progress; }
580 void set_inlining_progress(bool z) { _inlining_progress = z; }
581 int inlining_progress() const { return _inlining_progress; }
582 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
583 int inlining_incrementally() const { return _inlining_incrementally; }
584 void set_do_cleanup(bool z) { _do_cleanup = z; }
585 int do_cleanup() const { return _do_cleanup; }
586 void set_major_progress() { _major_progress++; }
587 void restore_major_progress(int progress) { _major_progress += progress; }
588 void clear_major_progress() { _major_progress = 0; }
589 int max_inline_size() const { return _max_inline_size; }
590 void set_freq_inline_size(int n) { _freq_inline_size = n; }
591 int freq_inline_size() const { return _freq_inline_size; }
592 void set_max_inline_size(int n) { _max_inline_size = n; }
593 bool has_loops() const { return _has_loops; }
594 void set_has_loops(bool z) { _has_loops = z; }
595 bool has_split_ifs() const { return _has_split_ifs; }
596 void set_has_split_ifs(bool z) { _has_split_ifs = z; }
597 bool has_unsafe_access() const { return _has_unsafe_access; }
598 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; }
599 bool has_stringbuilder() const { return _has_stringbuilder; }
600 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; }
601 bool has_boxed_value() const { return _has_boxed_value; }
602 void set_has_boxed_value(bool z) { _has_boxed_value = z; }
603 bool has_reserved_stack_access() const { return _has_reserved_stack_access; }
604 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
605 uint max_vector_size() const { return _max_vector_size; }
606 void set_max_vector_size(uint s) { _max_vector_size = s; }
607 bool clear_upper_avx() const { return _clear_upper_avx; }
608 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; }
609 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; }
610 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
611 bool trap_can_recompile() const { return _trap_can_recompile; }
612 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; }
613 uint decompile_count() const { return _decompile_count; }
614 void set_decompile_count(uint c) { _decompile_count = c; }
615 bool allow_range_check_smearing() const;
616 bool do_inlining() const { return _do_inlining; }
617 void set_do_inlining(bool z) { _do_inlining = z; }
618 bool do_scheduling() const { return _do_scheduling; }
619 void set_do_scheduling(bool z) { _do_scheduling = z; }
620 bool do_freq_based_layout() const{ return _do_freq_based_layout; }
621 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
622 bool do_vector_loop() const { return _do_vector_loop; }
623 void set_do_vector_loop(bool z) { _do_vector_loop = z; }
624 bool use_cmove() const { return _use_cmove; }
625 void set_use_cmove(bool z) { _use_cmove = z; }
626 int AliasLevel() const { return _AliasLevel; }
627 bool print_assembly() const { return _print_assembly; }
628 void set_print_assembly(bool z) { _print_assembly = z; }
629 bool print_inlining() const { return _print_inlining; }
630 void set_print_inlining(bool z) { _print_inlining = z; }
631 bool print_intrinsics() const { return _print_intrinsics; }
632 void set_print_intrinsics(bool z) { _print_intrinsics = z; }
633 RTMState rtm_state() const { return _rtm_state; }
634 void set_rtm_state(RTMState s) { _rtm_state = s; }
635 bool use_rtm() const { return (_rtm_state & NoRTM) == 0; }
636 bool profile_rtm() const { return _rtm_state == ProfileRTM; }
637 uint max_node_limit() const { return (uint)_max_node_limit; }
638 void set_max_node_limit(uint n) { _max_node_limit = n; }
639 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; }
640 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; }
641 void set_flattened_accesses() { _has_flattened_accesses = true; }
642 bool flattened_accesses_share_alias() const { return _flattened_accesses_share_alias; }
643 void set_flattened_accesses_share_alias(bool z) { _flattened_accesses_share_alias = z; }
644 bool scalarize_in_safepoints() const { return _scalarize_in_safepoints; }
645 void set_scalarize_in_safepoints(bool z) { _scalarize_in_safepoints = z; }
646
647 // Support for scalarized inline type calling convention
648 bool has_scalarized_args() const { return _method != NULL && _method->has_scalarized_args(); }
649 bool needs_stack_repair() const { return _method != NULL && _method->get_Method()->c2_needs_stack_repair(); }
650
651 bool has_monitors() const { return _has_monitors; }
652 void set_has_monitors(bool v) { _has_monitors = v; }
653
654 // check the CompilerOracle for special behaviours for this compile
655 bool method_has_option(enum CompileCommand option) {
656 return method() != NULL && method()->has_option(option);
657 }
658
659 #ifndef PRODUCT
660 uint next_igv_idx() { return _igv_idx++; }
661 bool trace_opto_output() const { return _trace_opto_output; }
662 void print_ideal_ir(const char* phase_name);
663 bool should_print_ideal() const { return _directive->PrintIdealOption; }
664 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
665 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
666 int _in_dump_cnt; // Required for dumping ir nodes.
667 #endif
668 bool has_irreducible_loop() const { return _has_irreducible_loop; }
669 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
670
671 // JSR 292
672 bool has_method_handle_invokes() const { return _has_method_handle_invokes; }
673 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; }
674
675 Ticks _latest_stage_start_counter;
676
677 void begin_method();
678 void end_method();
679 bool should_print_igv(int level);
680 bool should_print_phase(CompilerPhaseType cpt);
681
682 void print_method(CompilerPhaseType cpt, int level, Node* n = nullptr);
683
684 #ifndef PRODUCT
685 void igv_print_method_to_file(const char* phase_name = "Debug", bool append = false);
686 void igv_print_method_to_network(const char* phase_name = "Debug");
687 static IdealGraphPrinter* debug_file_printer() { return _debug_file_printer; }
688 static IdealGraphPrinter* debug_network_printer() { return _debug_network_printer; }
689 #endif
690
691 int macro_count() const { return _macro_nodes.length(); }
692 int predicate_count() const { return _predicate_opaqs.length(); }
693 int skeleton_predicate_count() const { return _skeleton_predicate_opaqs.length(); }
694 int expensive_count() const { return _expensive_nodes.length(); }
695 int coarsened_count() const { return _coarsened_locks.length(); }
696
697 Node* macro_node(int idx) const { return _macro_nodes.at(idx); }
698 Node* predicate_opaque1_node(int idx) const { return _predicate_opaqs.at(idx); }
699 Node* skeleton_predicate_opaque4_node(int idx) const { return _skeleton_predicate_opaqs.at(idx); }
700 Node* expensive_node(int idx) const { return _expensive_nodes.at(idx); }
701
702 ConnectionGraph* congraph() { return _congraph;}
703 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;}
704 void add_macro_node(Node * n) {
705 //assert(n->is_macro(), "must be a macro node");
706 assert(!_macro_nodes.contains(n), "duplicate entry in expand list");
707 _macro_nodes.append(n);
708 }
709 void remove_macro_node(Node* n) {
710 // this function may be called twice for a node so we can only remove it
711 // if it's still existing.
712 _macro_nodes.remove_if_existing(n);
713 // remove from _predicate_opaqs list also if it is there
714 if (predicate_count() > 0) {
715 _predicate_opaqs.remove_if_existing(n);
716 }
717 // Remove from coarsened locks list if present
718 if (coarsened_count() > 0) {
719 remove_coarsened_lock(n);
720 }
721 }
722 void add_expensive_node(Node* n);
723 void remove_expensive_node(Node* n) {
724 _expensive_nodes.remove_if_existing(n);
725 }
726 void add_predicate_opaq(Node* n) {
727 assert(!_predicate_opaqs.contains(n), "duplicate entry in predicate opaque1");
728 assert(_macro_nodes.contains(n), "should have already been in macro list");
729 _predicate_opaqs.append(n);
730 }
731 void add_skeleton_predicate_opaq(Node* n) {
732 assert(!_skeleton_predicate_opaqs.contains(n), "duplicate entry in skeleton predicate opaque4 list");
733 _skeleton_predicate_opaqs.append(n);
734 }
735 void remove_skeleton_predicate_opaq(Node* n) {
736 if (skeleton_predicate_count() > 0) {
737 _skeleton_predicate_opaqs.remove_if_existing(n);
738 }
739 }
740 void add_coarsened_locks(GrowableArray<AbstractLockNode*>& locks);
741 void remove_coarsened_lock(Node* n);
742 bool coarsened_locks_consistent();
743
744 bool post_loop_opts_phase() { return _post_loop_opts_phase; }
745 void set_post_loop_opts_phase() { _post_loop_opts_phase = true; }
746 void reset_post_loop_opts_phase() { _post_loop_opts_phase = false; }
747
748 void record_for_post_loop_opts_igvn(Node* n);
749 void remove_from_post_loop_opts_igvn(Node* n);
750 void process_for_post_loop_opts_igvn(PhaseIterGVN& igvn);
751
752 // Keep track of inline type nodes for later processing
753 void add_inline_type(Node* n);
754 void remove_inline_type(Node* n);
755 void process_inline_types(PhaseIterGVN &igvn, bool remove = false);
756
757 void adjust_flattened_array_access_aliases(PhaseIterGVN& igvn);
758
759 void record_unstable_if_trap(UnstableIfTrap* trap);
760 bool remove_unstable_if_trap(CallStaticJavaNode* unc, bool yield);
761 void remove_useless_unstable_if_traps(Unique_Node_List &useful);
762 void process_for_unstable_if_traps(PhaseIterGVN& igvn);
763
764 void sort_macro_nodes();
765
766 // remove the opaque nodes that protect the predicates so that the unused checks and
767 // uncommon traps will be eliminated from the graph.
768 void cleanup_loop_predicates(PhaseIterGVN &igvn);
769 bool is_predicate_opaq(Node* n) {
770 return _predicate_opaqs.contains(n);
771 }
772
773 // Are there candidate expensive nodes for optimization?
774 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
775 // Check whether n1 and n2 are similar
776 static int cmp_expensive_nodes(Node* n1, Node* n2);
777 // Sort expensive nodes to locate similar expensive nodes
778 void sort_expensive_nodes();
779
780 // Compilation environment.
781 Arena* comp_arena() { return &_comp_arena; }
782 ciEnv* env() const { return _env; }
783 CompileLog* log() const { return _log; }
784 bool failing() const { return _env->failing() || _failure_reason != NULL; }
785 const char* failure_reason() const { return (_env->failing()) ? _env->failure_reason() : _failure_reason; }
786
787 bool failure_reason_is(const char* r) const {
788 return (r == _failure_reason) || (r != NULL && _failure_reason != NULL && strcmp(r, _failure_reason) == 0);
789 }
790
791 void record_failure(const char* reason);
792 void record_method_not_compilable(const char* reason) {
793 env()->record_method_not_compilable(reason);
794 // Record failure reason.
795 record_failure(reason);
796 }
797 bool check_node_count(uint margin, const char* reason) {
798 if (live_nodes() + margin > max_node_limit()) {
799 record_method_not_compilable(reason);
800 return true;
801 } else {
802 return false;
803 }
804 }
805
806 // Node management
807 uint unique() const { return _unique; }
808 uint next_unique() { return _unique++; }
809 void set_unique(uint i) { _unique = i; }
810 static int debug_idx() { return debug_only(_debug_idx)+0; }
811 static void set_debug_idx(int i) { debug_only(_debug_idx = i); }
812 Arena* node_arena() { return &_node_arena; }
813 Arena* old_arena() { return &_old_arena; }
814 RootNode* root() const { return _root; }
815 void set_root(RootNode* r) { _root = r; }
816 StartNode* start() const; // (Derived from root.)
817 void init_start(StartNode* s);
818 Node* immutable_memory();
819
820 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; }
821 Node* recent_alloc_obj() const { return _recent_alloc_obj; }
822 void set_recent_alloc(Node* ctl, Node* obj) {
823 _recent_alloc_ctl = ctl;
824 _recent_alloc_obj = obj;
825 }
826 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return;
827 _dead_node_count++;
828 }
829 void reset_dead_node_list() { _dead_node_list.reset();
830 _dead_node_count = 0;
831 }
832 uint live_nodes() const {
833 int val = _unique - _dead_node_count;
834 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
835 return (uint) val;
836 }
837 #ifdef ASSERT
838 void set_phase_optimize_finished() { _phase_optimize_finished = true; }
839 bool phase_optimize_finished() const { return _phase_optimize_finished; }
840 uint count_live_nodes_by_graph_walk();
841 void print_missing_nodes();
842 #endif
843
844 // Record modified nodes to check that they are put on IGVN worklist
845 void record_modified_node(Node* n) NOT_DEBUG_RETURN;
846 void remove_modified_node(Node* n) NOT_DEBUG_RETURN;
847 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } )
848
849 MachConstantBaseNode* mach_constant_base_node();
850 bool has_mach_constant_base_node() const { return _mach_constant_base_node != NULL; }
851 // Generated by adlc, true if CallNode requires MachConstantBase.
852 bool needs_deep_clone_jvms();
853
854 // Handy undefined Node
855 Node* top() const { return _top; }
856
857 // these are used by guys who need to know about creation and transformation of top:
858 Node* cached_top_node() { return _top; }
859 void set_cached_top_node(Node* tn);
860
861 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
862 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
863 Node_Notes* default_node_notes() const { return _default_node_notes; }
864 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
865
866 Node_Notes* node_notes_at(int idx) {
867 return locate_node_notes(_node_note_array, idx, false);
868 }
869 inline bool set_node_notes_at(int idx, Node_Notes* value);
870
871 // Copy notes from source to dest, if they exist.
872 // Overwrite dest only if source provides something.
873 // Return true if information was moved.
874 bool copy_node_notes_to(Node* dest, Node* source);
875
876 // Workhorse function to sort out the blocked Node_Notes array:
877 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
878 int idx, bool can_grow = false);
879
880 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
881
882 // Type management
883 Arena* type_arena() { return _type_arena; }
884 Dict* type_dict() { return _type_dict; }
885 size_t type_last_size() { return _type_last_size; }
886 int num_alias_types() { return _num_alias_types; }
887
888 void init_type_arena() { _type_arena = &_Compile_types; }
889 void set_type_arena(Arena* a) { _type_arena = a; }
890 void set_type_dict(Dict* d) { _type_dict = d; }
891 void set_type_last_size(size_t sz) { _type_last_size = sz; }
892
893 const TypeFunc* last_tf(ciMethod* m) {
894 return (m == _last_tf_m) ? _last_tf : NULL;
895 }
896 void set_last_tf(ciMethod* m, const TypeFunc* tf) {
897 assert(m != NULL || tf == NULL, "");
898 _last_tf_m = m;
899 _last_tf = tf;
900 }
901
902 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
903 AliasType* alias_type(const TypePtr* adr_type, ciField* field = NULL, bool uncached = false) { return find_alias_type(adr_type, false, field, uncached); }
904 bool have_alias_type(const TypePtr* adr_type);
905 AliasType* alias_type(ciField* field);
906
907 int get_alias_index(const TypePtr* at, bool uncached = false) { return alias_type(at, NULL, uncached)->index(); }
908 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); }
909 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); }
910
911 // Building nodes
912 void rethrow_exceptions(JVMState* jvms);
913 void return_values(JVMState* jvms);
914 JVMState* build_start_state(StartNode* start, const TypeFunc* tf);
915
916 // Decide how to build a call.
917 // The profile factor is a discount to apply to this site's interp. profile.
918 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
919 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = NULL,
920 bool allow_intrinsics = true);
921 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
922 return should_delay_string_inlining(call_method, jvms) ||
923 should_delay_boxing_inlining(call_method, jvms) ||
924 should_delay_vector_inlining(call_method, jvms);
925 }
926 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
927 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
928 bool should_delay_vector_inlining(ciMethod* call_method, JVMState* jvms);
929 bool should_delay_vector_reboxing_inlining(ciMethod* call_method, JVMState* jvms);
930
931 // Helper functions to identify inlining potential at call-site
932 ciMethod* optimize_virtual_call(ciMethod* caller, ciInstanceKlass* klass,
933 ciKlass* holder, ciMethod* callee,
934 const TypeOopPtr* receiver_type, bool is_virtual,
935 bool &call_does_dispatch, int &vtable_index,
936 bool check_access = true);
937 ciMethod* optimize_inlining(ciMethod* caller, ciInstanceKlass* klass, ciKlass* holder,
938 ciMethod* callee, const TypeOopPtr* receiver_type,
939 bool check_access = true);
940
941 // Report if there were too many traps at a current method and bci.
942 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
943 // If there is no MDO at all, report no trap unless told to assume it.
944 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
945 // This version, unspecific to a particular bci, asks if
946 // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
947 bool too_many_traps(Deoptimization::DeoptReason reason,
948 // Privately used parameter for logging:
949 ciMethodData* logmd = NULL);
950 // Report if there were too many recompiles at a method and bci.
951 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
952 // Report if there were too many traps or recompiles at a method and bci.
953 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
954 return too_many_traps(method, bci, reason) ||
955 too_many_recompiles(method, bci, reason);
956 }
957 // Return a bitset with the reasons where deoptimization is allowed,
958 // i.e., where there were not too many uncommon traps.
959 int _allowed_reasons;
960 int allowed_deopt_reasons() { return _allowed_reasons; }
961 void set_allowed_deopt_reasons();
962
963 // Parsing, optimization
964 PhaseGVN* initial_gvn() { return _initial_gvn; }
965 Unique_Node_List* for_igvn() { return _for_igvn; }
966 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List.
967 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; }
968 void set_for_igvn(Unique_Node_List *for_igvn) { _for_igvn = for_igvn; }
969
970 // Replace n by nn using initial_gvn, calling hash_delete and
971 // record_for_igvn as needed.
972 void gvn_replace_by(Node* n, Node* nn);
973
974
975 void identify_useful_nodes(Unique_Node_List &useful);
976 void update_dead_node_list(Unique_Node_List &useful);
977 void disconnect_useless_nodes(Unique_Node_List &useful, Unique_Node_List* worklist);
978
979 void remove_useless_node(Node* dead);
980
981 // Record this CallGenerator for inlining at the end of parsing.
982 void add_late_inline(CallGenerator* cg) {
983 _late_inlines.insert_before(_late_inlines_pos, cg);
984 _late_inlines_pos++;
985 }
986
987 void prepend_late_inline(CallGenerator* cg) {
988 _late_inlines.insert_before(0, cg);
989 }
990
991 void add_string_late_inline(CallGenerator* cg) {
992 _string_late_inlines.push(cg);
993 }
994
995 void add_boxing_late_inline(CallGenerator* cg) {
996 _boxing_late_inlines.push(cg);
997 }
998
999 void add_vector_reboxing_late_inline(CallGenerator* cg) {
1000 _vector_reboxing_late_inlines.push(cg);
1001 }
1002
1003 void remove_useless_nodes (GrowableArray<Node*>& node_list, Unique_Node_List &useful);
1004
1005 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1006 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Node* dead);
1007
1008 void remove_useless_coarsened_locks(Unique_Node_List& useful);
1009
1010 void process_print_inlining();
1011 void dump_print_inlining();
1012
1013 bool over_inlining_cutoff() const {
1014 if (!inlining_incrementally()) {
1015 return unique() > (uint)NodeCountInliningCutoff;
1016 } else {
1017 // Give some room for incremental inlining algorithm to "breathe"
1018 // and avoid thrashing when live node count is close to the limit.
1019 // Keep in mind that live_nodes() isn't accurate during inlining until
1020 // dead node elimination step happens (see Compile::inline_incrementally).
1021 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10;
1022 }
1023 }
1024
1025 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; }
1026 void dec_number_of_mh_late_inlines() { assert(_number_of_mh_late_inlines > 0, "_number_of_mh_late_inlines < 0 !"); _number_of_mh_late_inlines--; }
1027 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; }
1028
1029 bool inline_incrementally_one();
1030 void inline_incrementally_cleanup(PhaseIterGVN& igvn);
1031 void inline_incrementally(PhaseIterGVN& igvn);
1032 void inline_string_calls(bool parse_time);
1033 void inline_boxing_calls(PhaseIterGVN& igvn);
1034 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1035 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn);
1036
1037 void inline_vector_reboxing_calls();
1038 bool has_vbox_nodes();
1039
1040 void process_late_inline_calls_no_inline(PhaseIterGVN& igvn);
1041
1042 // Matching, CFG layout, allocation, code generation
1043 PhaseCFG* cfg() { return _cfg; }
1044 bool has_java_calls() const { return _java_calls > 0; }
1045 int java_calls() const { return _java_calls; }
1046 int inner_loops() const { return _inner_loops; }
1047 Matcher* matcher() { return _matcher; }
1048 PhaseRegAlloc* regalloc() { return _regalloc; }
1049 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; }
1050 Arena* indexSet_arena() { return _indexSet_arena; }
1051 void* indexSet_free_block_list() { return _indexSet_free_block_list; }
1052 DebugInformationRecorder* debug_info() { return env()->debug_info(); }
1053
1054 void update_interpreter_frame_size(int size) {
1055 if (_interpreter_frame_size < size) {
1056 _interpreter_frame_size = size;
1057 }
1058 }
1059
1060 void set_matcher(Matcher* m) { _matcher = m; }
1061 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; }
1062 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; }
1063 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; }
1064
1065 void set_java_calls(int z) { _java_calls = z; }
1066 void set_inner_loops(int z) { _inner_loops = z; }
1067
1068 Dependencies* dependencies() { return env()->dependencies(); }
1069
1070 // Major entry point. Given a Scope, compile the associated method.
1071 // For normal compilations, entry_bci is InvocationEntryBci. For on stack
1072 // replacement, entry_bci indicates the bytecode for which to compile a
1073 // continuation.
1074 Compile(ciEnv* ci_env, ciMethod* target,
1075 int entry_bci, Options options, DirectiveSet* directive);
1076
1077 // Second major entry point. From the TypeFunc signature, generate code
1078 // to pass arguments from the Java calling convention to the C calling
1079 // convention.
1080 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1081 address stub_function, const char *stub_name,
1082 int is_fancy_jump, bool pass_tls,
1083 bool return_pc, DirectiveSet* directive);
1084
1085 // Are we compiling a method?
1086 bool has_method() { return method() != NULL; }
1087
1088 // Maybe print some information about this compile.
1089 void print_compile_messages();
1090
1091 // Final graph reshaping, a post-pass after the regular optimizer is done.
1092 bool final_graph_reshaping();
1093
1094 // returns true if adr is completely contained in the given alias category
1095 bool must_alias(const TypePtr* adr, int alias_idx);
1096
1097 // returns true if adr overlaps with the given alias category
1098 bool can_alias(const TypePtr* adr, int alias_idx);
1099
1100 // Stack slots that may be unused by the calling convention but must
1101 // otherwise be preserved. On Intel this includes the return address.
1102 // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1103 uint in_preserve_stack_slots() {
1104 return SharedRuntime::in_preserve_stack_slots();
1105 }
1106
1107 // "Top of Stack" slots that may be unused by the calling convention but must
1108 // otherwise be preserved.
1109 // On Intel these are not necessary and the value can be zero.
1110 static uint out_preserve_stack_slots() {
1111 return SharedRuntime::out_preserve_stack_slots();
1112 }
1113
1114 // Number of outgoing stack slots killed above the out_preserve_stack_slots
1115 // for calls to C. Supports the var-args backing area for register parms.
1116 uint varargs_C_out_slots_killed() const;
1117
1118 // Number of Stack Slots consumed by a synchronization entry
1119 int sync_stack_slots() const;
1120
1121 // Compute the name of old_SP. See <arch>.ad for frame layout.
1122 OptoReg::Name compute_old_SP();
1123
1124 private:
1125 // Phase control:
1126 void Init(int aliaslevel); // Prepare for a single compilation
1127 int Inline_Warm(); // Find more inlining work.
1128 void Finish_Warm(); // Give up on further inlines.
1129 void Optimize(); // Given a graph, optimize it
1130 void Code_Gen(); // Generate code from a graph
1131
1132 // Management of the AliasType table.
1133 void grow_alias_types();
1134 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1135 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1136 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field, bool uncached = false);
1137
1138 void verify_top(Node*) const PRODUCT_RETURN;
1139
1140 // Intrinsic setup.
1141 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor
1142 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper
1143 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn
1144 void register_intrinsic(CallGenerator* cg); // update fn
1145
1146 #ifndef PRODUCT
1147 static juint _intrinsic_hist_count[];
1148 static jubyte _intrinsic_hist_flags[];
1149 #endif
1150 // Function calls made by the public function final_graph_reshaping.
1151 // No need to be made public as they are not called elsewhere.
1152 void final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &frc);
1153 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop);
1154 void final_graph_reshaping_walk( Node_Stack &nstack, Node *root, Final_Reshape_Counts &frc );
1155 void eliminate_redundant_card_marks(Node* n);
1156
1157 // Logic cone optimization.
1158 void optimize_logic_cones(PhaseIterGVN &igvn);
1159 void collect_logic_cone_roots(Unique_Node_List& list);
1160 void process_logic_cone_root(PhaseIterGVN &igvn, Node* n, VectorSet& visited);
1161 bool compute_logic_cone(Node* n, Unique_Node_List& partition, Unique_Node_List& inputs);
1162 uint compute_truth_table(Unique_Node_List& partition, Unique_Node_List& inputs);
1163 uint eval_macro_logic_op(uint func, uint op1, uint op2, uint op3);
1164 Node* xform_to_MacroLogicV(PhaseIterGVN &igvn, const TypeVect* vt, Unique_Node_List& partitions, Unique_Node_List& inputs);
1165 void check_no_dead_use() const NOT_DEBUG_RETURN;
1166
1167 public:
1168
1169 // Note: Histogram array size is about 1 Kb.
1170 enum { // flag bits:
1171 _intrinsic_worked = 1, // succeeded at least once
1172 _intrinsic_failed = 2, // tried it but it failed
1173 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1174 _intrinsic_virtual = 8, // was seen in the virtual form (rare)
1175 _intrinsic_both = 16 // was seen in the non-virtual form (usual)
1176 };
1177 // Update histogram. Return boolean if this is a first-time occurrence.
1178 static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1179 bool is_virtual, int flags) PRODUCT_RETURN0;
1180 static void print_intrinsic_statistics() PRODUCT_RETURN;
1181
1182 // Graph verification code
1183 // Walk the node list, verifying that there is a one-to-one
1184 // correspondence between Use-Def edges and Def-Use edges
1185 // The option no_dead_code enables stronger checks that the
1186 // graph is strongly connected from root in both directions.
1187 void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN;
1188
1189 // End-of-run dumps.
1190 static void print_statistics() PRODUCT_RETURN;
1191
1192 // Verify ADLC assumptions during startup
1193 static void adlc_verification() PRODUCT_RETURN;
1194
1195 // Definitions of pd methods
1196 static void pd_compiler2_init();
1197
1198 // Static parse-time type checking logic for gen_subtype_check:
1199 enum SubTypeCheckResult { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1200 SubTypeCheckResult static_subtype_check(const TypeKlassPtr* superk, const TypeKlassPtr* subk);
1201
1202 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1203 // Optional control dependency (for example, on range check)
1204 Node* ctrl = NULL);
1205
1206 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1207 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl, bool carry_dependency = false);
1208
1209 Node* optimize_acmp(PhaseGVN* phase, Node* a, Node* b);
1210
1211 // Auxiliary method for randomized fuzzing/stressing
1212 int random();
1213 bool randomized_select(int count);
1214
1215 // supporting clone_map
1216 CloneMap& clone_map();
1217 void set_clone_map(Dict* d);
1218
1219 bool needs_clinit_barrier(ciField* ik, ciMethod* accessing_method);
1220 bool needs_clinit_barrier(ciMethod* ik, ciMethod* accessing_method);
1221 bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method);
1222
1223 #ifdef IA32
1224 private:
1225 bool _select_24_bit_instr; // We selected an instruction with a 24-bit result
1226 bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results
1227
1228 // Remember if this compilation changes hardware mode to 24-bit precision.
1229 void set_24_bit_selection_and_mode(bool selection, bool mode) {
1230 _select_24_bit_instr = selection;
1231 _in_24_bit_fp_mode = mode;
1232 }
1233
1234 public:
1235 bool select_24_bit_instr() const { return _select_24_bit_instr; }
1236 bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; }
1237 #endif // IA32
1238 #ifdef ASSERT
1239 bool _type_verify_symmetry;
1240 void set_exception_backedge() { _exception_backedge = true; }
1241 bool has_exception_backedge() const { return _exception_backedge; }
1242 #endif
1243
1244 static bool push_thru_add(PhaseGVN* phase, Node* z, const TypeInteger* tz, const TypeInteger*& rx, const TypeInteger*& ry,
1245 BasicType out_bt, BasicType in_bt);
1246
1247 static Node* narrow_value(BasicType bt, Node* value, const Type* type, PhaseGVN* phase, bool transform_res);
1248 };
1249
1250 #endif // SHARE_OPTO_COMPILE_HPP