1 /*
2 * Copyright (c) 1997, 2023, 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 Matcher;
61 class MachConstantNode;
62 class MachConstantBaseNode;
63 class MachNode;
64 class MachOper;
65 class MachSafePointNode;
66 class Node;
67 class Node_Array;
68 class Node_List;
69 class Node_Notes;
70 class NodeCloneInfo;
71 class OptoReg;
72 class PhaseCFG;
73 class PhaseGVN;
74 class PhaseIterGVN;
75 class PhaseRegAlloc;
76 class PhaseCCP;
77 class PhaseOutput;
78 class RootNode;
79 class relocInfo;
80 class StartNode;
81 class SafePointNode;
82 class JVMState;
83 class Type;
84 class TypeInt;
85 class TypeInteger;
86 class TypeKlassPtr;
87 class TypePtr;
88 class TypeOopPtr;
89 class TypeFunc;
90 class TypeVect;
91 class Unique_Node_List;
92 class UnstableIfTrap;
93 class InlineTypeNode;
94 class nmethod;
95 class Node_Stack;
96 struct Final_Reshape_Counts;
97
98 enum LoopOptsMode {
99 LoopOptsDefault,
100 LoopOptsNone,
101 LoopOptsMaxUnroll,
102 LoopOptsShenandoahExpand,
103 LoopOptsShenandoahPostExpand,
104 LoopOptsSkipSplitIf,
105 LoopOptsVerify
106 };
107
108 // The type of all node counts and indexes.
109 // It must hold at least 16 bits, but must also be fast to load and store.
110 // This type, if less than 32 bits, could limit the number of possible nodes.
111 // (To make this type platform-specific, move to globalDefinitions_xxx.hpp.)
112 typedef unsigned int node_idx_t;
113
114 class NodeCloneInfo {
115 private:
116 uint64_t _idx_clone_orig;
117 public:
118
119 void set_idx(node_idx_t idx) {
120 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx;
121 }
122 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); }
123
124 void set_gen(int generation) {
125 uint64_t g = (uint64_t)generation << 32;
126 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g;
127 }
128 int gen() const { return (int)(_idx_clone_orig >> 32); }
129
130 void set(uint64_t x) { _idx_clone_orig = x; }
131 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); }
132 uint64_t get() const { return _idx_clone_orig; }
133
134 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {}
135 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); }
136
137 void dump() const;
138 };
139
140 class CloneMap {
141 friend class Compile;
142 private:
143 bool _debug;
144 Dict* _dict;
145 int _clone_idx; // current cloning iteration/generation in loop unroll
146 public:
147 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy
148 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; }
149 Dict* dict() const { return _dict; }
150 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == NULL, "key existed"); _dict->Insert(_2p(key), (void*)val); }
151 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); }
152 void remove(node_idx_t key) { _dict->Delete(_2p(key)); }
153 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); }
154 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); }
155 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); }
156 int gen(const void* k) const { return gen(_2_node_idx_t(k)); }
157 int max_gen() const;
158 void clone(Node* old, Node* nnn, int gen);
159 void verify_insert_and_clone(Node* old, Node* nnn, int gen);
160 void dump(node_idx_t key) const;
161
162 int clone_idx() const { return _clone_idx; }
163 void set_clone_idx(int x) { _clone_idx = x; }
164 bool is_debug() const { return _debug; }
165 void set_debug(bool debug) { _debug = debug; }
166
167 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); }
168 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); }
169 };
170
171 class Options {
172 friend class Compile;
173 friend class VMStructs;
174 private:
175 const bool _subsume_loads; // Load can be matched as part of a larger op.
176 const bool _do_escape_analysis; // Do escape analysis.
177 const bool _do_iterative_escape_analysis; // Do iterative escape analysis.
178 const bool _eliminate_boxing; // Do boxing elimination.
179 const bool _do_locks_coarsening; // Do locks coarsening
180 const bool _install_code; // Install the code that was compiled
181 public:
182 Options(bool subsume_loads, bool do_escape_analysis,
183 bool do_iterative_escape_analysis,
184 bool eliminate_boxing, bool do_locks_coarsening,
185 bool install_code) :
186 _subsume_loads(subsume_loads),
187 _do_escape_analysis(do_escape_analysis),
188 _do_iterative_escape_analysis(do_iterative_escape_analysis),
189 _eliminate_boxing(eliminate_boxing),
190 _do_locks_coarsening(do_locks_coarsening),
191 _install_code(install_code) {
192 }
193
194 static Options for_runtime_stub() {
195 return Options(
196 /* subsume_loads = */ true,
197 /* do_escape_analysis = */ false,
198 /* do_iterative_escape_analysis = */ false,
199 /* eliminate_boxing = */ false,
200 /* do_lock_coarsening = */ false,
201 /* install_code = */ true
202 );
203 }
204 };
205
206 //------------------------------Compile----------------------------------------
207 // This class defines a top-level Compiler invocation.
208
209 class Compile : public Phase {
210 friend class VMStructs;
211
212 public:
213 // Fixed alias indexes. (See also MergeMemNode.)
214 enum {
215 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value)
216 AliasIdxBot = 2, // pseudo-index, aliases to everything
217 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM
218 };
219
220 // Variant of TraceTime(NULL, &_t_accumulator, CITime);
221 // Integrated with logging. If logging is turned on, and CITimeVerbose is true,
222 // then brackets are put into the log, with time stamps and node counts.
223 // (The time collection itself is always conditionalized on CITime.)
224 class TracePhase : public TraceTime {
225 private:
226 Compile* C;
227 CompileLog* _log;
228 const char* _phase_name;
229 bool _dolog;
230 public:
231 TracePhase(const char* name, elapsedTimer* accumulator);
232 ~TracePhase();
233 };
234
235 // Information per category of alias (memory slice)
236 class AliasType {
237 private:
238 friend class Compile;
239
240 int _index; // unique index, used with MergeMemNode
241 const TypePtr* _adr_type; // normalized address type
242 ciField* _field; // relevant instance field, or null if none
243 const Type* _element; // relevant array element type, or null if none
244 bool _is_rewritable; // false if the memory is write-once only
245 int _general_index; // if this is type is an instance, the general
246 // type that this is an instance of
247
248 void Init(int i, const TypePtr* at);
249
250 public:
251 int index() const { return _index; }
252 const TypePtr* adr_type() const { return _adr_type; }
253 ciField* field() const { return _field; }
254 const Type* element() const { return _element; }
255 bool is_rewritable() const { return _is_rewritable; }
256 bool is_volatile() const { return (_field ? _field->is_volatile() : false); }
257 int general_index() const { return (_general_index != 0) ? _general_index : _index; }
258
259 void set_rewritable(bool z) { _is_rewritable = z; }
260 void set_field(ciField* f) {
261 assert(!_field,"");
262 _field = f;
263 if (f->is_final() || f->is_stable()) {
264 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
265 _is_rewritable = false;
266 }
267 }
268 void set_element(const Type* e) {
269 assert(_element == NULL, "");
270 _element = e;
271 }
272
273 BasicType basic_type() const;
274
275 void print_on(outputStream* st) PRODUCT_RETURN;
276 };
277
278 enum {
279 logAliasCacheSize = 6,
280 AliasCacheSize = (1<<logAliasCacheSize)
281 };
282 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type
283 enum {
284 trapHistLength = MethodData::_trap_hist_limit
285 };
286
287 private:
288 // Fixed parameters to this compilation.
289 const int _compile_id;
290 const Options _options; // Compilation options
291 ciMethod* _method; // The method being compiled.
292 int _entry_bci; // entry bci for osr methods.
293 const TypeFunc* _tf; // My kind of signature
294 InlineTree* _ilt; // Ditto (temporary).
295 address _stub_function; // VM entry for stub being compiled, or NULL
296 const char* _stub_name; // Name of stub or adapter being compiled, or NULL
297 address _stub_entry_point; // Compile code entry for generated stub, or NULL
298
299 // Control of this compilation.
300 int _max_inline_size; // Max inline size for this compilation
301 int _freq_inline_size; // Max hot method inline size for this compilation
302 int _fixed_slots; // count of frame slots not allocated by the register
303 // allocator i.e. locks, original deopt pc, etc.
304 uintx _max_node_limit; // Max unique node count during a single compilation.
305
306 bool _post_loop_opts_phase; // Loop opts are finished.
307
308 int _major_progress; // Count of something big happening
309 bool _inlining_progress; // progress doing incremental inlining?
310 bool _inlining_incrementally;// Are we doing incremental inlining (post parse)
311 bool _do_cleanup; // Cleanup is needed before proceeding with incremental inlining
312 bool _has_loops; // True if the method _may_ have some loops
313 bool _has_split_ifs; // True if the method _may_ have some split-if
314 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores.
315 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated
316 bool _has_boxed_value; // True if a boxed object is allocated
317 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess
318 uint _max_vector_size; // Maximum size of generated vectors
319 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper
320 uint _trap_hist[trapHistLength]; // Cumulative traps
321 bool _trap_can_recompile; // Have we emitted a recompiling trap?
322 uint _decompile_count; // Cumulative decompilation counts.
323 bool _do_inlining; // True if we intend to do inlining
324 bool _do_scheduling; // True if we intend to do scheduling
325 bool _do_freq_based_layout; // True if we intend to do frequency based block layout
326 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations
327 bool _use_cmove; // True if CMove should be used without profitability analysis
328 bool _do_aliasing; // True if we intend to do aliasing
329 bool _print_assembly; // True if we should dump assembly code for this compilation
330 bool _print_inlining; // True if we should print inlining for this compilation
331 bool _print_intrinsics; // True if we should print intrinsics for this compilation
332 #ifndef PRODUCT
333 uint _igv_idx; // Counter for IGV node identifiers
334 bool _trace_opto_output;
335 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
336 #endif
337 bool _has_irreducible_loop; // Found irreducible loops
338 // JSR 292
339 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes.
340 bool _has_monitors; // Metadata transfered to nmethod to enable Continuations lock-detection fastpath
341 RTMState _rtm_state; // State of Restricted Transactional Memory usage
342 int _loop_opts_cnt; // loop opts round
343 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry
344 bool _has_flattened_accesses; // Any known flattened array accesses?
345 bool _flattened_accesses_share_alias; // Initially all flattened array share a single slice
346 bool _scalarize_in_safepoints; // Scalarize inline types in safepoint debug info
347 uint _stress_seed; // Seed for stress testing
348
349 // Compilation environment.
350 Arena _comp_arena; // Arena with lifetime equivalent to Compile
351 void* _barrier_set_state; // Potential GC barrier state for Compile
352 ciEnv* _env; // CI interface
353 DirectiveSet* _directive; // Compiler directive
354 CompileLog* _log; // from CompilerThread
355 const char* _failure_reason; // for record_failure/failing pattern
356 GrowableArray<CallGenerator*> _intrinsics; // List of intrinsics.
357 GrowableArray<Node*> _macro_nodes; // List of nodes which need to be expanded before matching.
358 GrowableArray<Node*> _predicate_opaqs; // List of Opaque1 nodes for the loop predicates.
359 GrowableArray<Node*> _skeleton_predicate_opaqs; // List of Opaque4 nodes for the loop skeleton predicates.
360 GrowableArray<Node*> _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
361 GrowableArray<Node*> _for_post_loop_igvn; // List of nodes for IGVN after loop opts are over
362 GrowableArray<Node*> _inline_type_nodes; // List of InlineType nodes
363 GrowableArray<UnstableIfTrap*> _unstable_if_traps; // List of ifnodes after IGVN
364 GrowableArray<Node_List*> _coarsened_locks; // List of coarsened Lock and Unlock nodes
365 ConnectionGraph* _congraph;
366 #ifndef PRODUCT
367 IdealGraphPrinter* _igv_printer;
368 static IdealGraphPrinter* _debug_file_printer;
369 static IdealGraphPrinter* _debug_network_printer;
370 #endif
371
372
373 // Node management
374 uint _unique; // Counter for unique Node indices
375 VectorSet _dead_node_list; // Set of dead nodes
376 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N).
377 // So use this to keep count and make the call O(1).
378 DEBUG_ONLY(Unique_Node_List* _modified_nodes;) // List of nodes which inputs were modified
379 DEBUG_ONLY(bool _phase_optimize_finished;) // Used for live node verification while creating new nodes
380
381 debug_only(static int _debug_idx;) // Monotonic counter (not reset), use -XX:BreakAtNode=<idx>
382 Arena _node_arena; // Arena for new-space Nodes
383 Arena _old_arena; // Arena for old-space Nodes, lifetime during xform
384 RootNode* _root; // Unique root of compilation, or NULL after bail-out.
385 Node* _top; // Unique top node. (Reset by various phases.)
386
387 Node* _immutable_memory; // Initial memory state
388
389 Node* _recent_alloc_obj;
390 Node* _recent_alloc_ctl;
391
392 // Constant table
393 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton.
394
395
396 // Blocked array of debugging and profiling information,
397 // tracked per node.
398 enum { _log2_node_notes_block_size = 8,
399 _node_notes_block_size = (1<<_log2_node_notes_block_size)
400 };
401 GrowableArray<Node_Notes*>* _node_note_array;
402 Node_Notes* _default_node_notes; // default notes for new nodes
403
404 // After parsing and every bulk phase we hang onto the Root instruction.
405 // The RootNode instruction is where the whole program begins. It produces
406 // the initial Control and BOTTOM for everybody else.
407
408 // Type management
409 Arena _Compile_types; // Arena for all types
410 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared()
411 Dict* _type_dict; // Intern table
412 CloneMap _clone_map; // used for recording history of cloned nodes
413 size_t _type_last_size; // Last allocation size (see Type::operator new/delete)
414 ciMethod* _last_tf_m; // Cache for
415 const TypeFunc* _last_tf; // TypeFunc::make
416 AliasType** _alias_types; // List of alias types seen so far.
417 int _num_alias_types; // Logical length of _alias_types
418 int _max_alias_types; // Physical length of _alias_types
419 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
420
421 // Parsing, optimization
422 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN
423 Unique_Node_List* _for_igvn; // Initial work-list for next round of Iterative GVN
424
425 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after main parsing has finished.
426 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
427 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
428
429 GrowableArray<CallGenerator*> _vector_reboxing_late_inlines; // same but for vector reboxing operations
430
431 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
432 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending
433
434 // Inlining may not happen in parse order which would make
435 // PrintInlining output confusing. Keep track of PrintInlining
436 // pieces in order.
437 class PrintInliningBuffer : public CHeapObj<mtCompiler> {
438 private:
439 CallGenerator* _cg;
440 stringStream _ss;
441 static const size_t default_stream_buffer_size = 128;
442
443 public:
444 PrintInliningBuffer()
445 : _cg(NULL), _ss(default_stream_buffer_size) {}
446
447 stringStream* ss() { return &_ss; }
448 CallGenerator* cg() { return _cg; }
449 void set_cg(CallGenerator* cg) { _cg = cg; }
450 };
451
452 stringStream* _print_inlining_stream;
453 GrowableArray<PrintInliningBuffer*>* _print_inlining_list;
454 int _print_inlining_idx;
455 char* _print_inlining_output;
456
457 // Only keep nodes in the expensive node list that need to be optimized
458 void cleanup_expensive_nodes(PhaseIterGVN &igvn);
459 // Use for sorting expensive nodes to bring similar nodes together
460 static int cmp_expensive_nodes(Node** n1, Node** n2);
461 // Expensive nodes list already sorted?
462 bool expensive_nodes_sorted() const;
463 // Remove the speculative part of types and clean up the graph
464 void remove_speculative_types(PhaseIterGVN &igvn);
465
466 void* _replay_inline_data; // Pointer to data loaded from file
467
468 void print_inlining_init();
469 void print_inlining_reinit();
470 void print_inlining_commit();
471 void print_inlining_push();
472 PrintInliningBuffer* print_inlining_current();
473
474 void log_late_inline_failure(CallGenerator* cg, const char* msg);
475 DEBUG_ONLY(bool _exception_backedge;)
476
477 public:
478
479 void* barrier_set_state() const { return _barrier_set_state; }
480
481 stringStream* print_inlining_stream() {
482 assert(print_inlining() || print_intrinsics(), "PrintInlining off?");
483 return _print_inlining_stream;
484 }
485
486 void print_inlining_update(CallGenerator* cg);
487 void print_inlining_update_delayed(CallGenerator* cg);
488 void print_inlining_move_to(CallGenerator* cg);
489 void print_inlining_assert_ready();
490 void print_inlining_reset();
491
492 void print_inlining(ciMethod* method, int inline_level, int bci, const char* msg = NULL) {
493 stringStream ss;
494 CompileTask::print_inlining_inner(&ss, method, inline_level, bci, msg);
495 print_inlining_stream()->print("%s", ss.freeze());
496 }
497
498 #ifndef PRODUCT
499 IdealGraphPrinter* igv_printer() { return _igv_printer; }
500 #endif
501
502 void log_late_inline(CallGenerator* cg);
503 void log_inline_id(CallGenerator* cg);
504 void log_inline_failure(const char* msg);
505
506 void* replay_inline_data() const { return _replay_inline_data; }
507
508 // Dump inlining replay data to the stream.
509 void dump_inline_data(outputStream* out);
510 void dump_inline_data_reduced(outputStream* out);
511
512 private:
513 // Matching, CFG layout, allocation, code generation
514 PhaseCFG* _cfg; // Results of CFG finding
515 int _java_calls; // Number of java calls in the method
516 int _inner_loops; // Number of inner loops in the method
517 Matcher* _matcher; // Engine to map ideal to machine instructions
518 PhaseRegAlloc* _regalloc; // Results of register allocation.
519 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout)
520 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin
521 void* _indexSet_free_block_list; // free list of IndexSet bit blocks
522 int _interpreter_frame_size;
523
524 PhaseOutput* _output;
525
526 public:
527 // Accessors
528
529 // The Compile instance currently active in this (compiler) thread.
530 static Compile* current() {
531 return (Compile*) ciEnv::current()->compiler_data();
532 }
533
534 int interpreter_frame_size() const { return _interpreter_frame_size; }
535
536 PhaseOutput* output() const { return _output; }
537 void set_output(PhaseOutput* o) { _output = o; }
538
539 // ID for this compilation. Useful for setting breakpoints in the debugger.
540 int compile_id() const { return _compile_id; }
541 DirectiveSet* directive() const { return _directive; }
542
543 // Does this compilation allow instructions to subsume loads? User
544 // instructions that subsume a load may result in an unschedulable
545 // instruction sequence.
546 bool subsume_loads() const { return _options._subsume_loads; }
547 /** Do escape analysis. */
548 bool do_escape_analysis() const { return _options._do_escape_analysis; }
549 bool do_iterative_escape_analysis() const { return _options._do_iterative_escape_analysis; }
550 /** Do boxing elimination. */
551 bool eliminate_boxing() const { return _options._eliminate_boxing; }
552 /** Do aggressive boxing elimination. */
553 bool aggressive_unboxing() const { return _options._eliminate_boxing && AggressiveUnboxing; }
554 bool should_install_code() const { return _options._install_code; }
555 /** Do locks coarsening. */
556 bool do_locks_coarsening() const { return _options._do_locks_coarsening; }
557
558 // Other fixed compilation parameters.
559 ciMethod* method() const { return _method; }
560 int entry_bci() const { return _entry_bci; }
561 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; }
562 bool is_method_compilation() const { return (_method != NULL && !_method->flags().is_native()); }
563 const TypeFunc* tf() const { assert(_tf!=NULL, ""); return _tf; }
564 void init_tf(const TypeFunc* tf) { assert(_tf==NULL, ""); _tf = tf; }
565 InlineTree* ilt() const { return _ilt; }
566 address stub_function() const { return _stub_function; }
567 const char* stub_name() const { return _stub_name; }
568 address stub_entry_point() const { return _stub_entry_point; }
569 void set_stub_entry_point(address z) { _stub_entry_point = z; }
570
571 // Control of this compilation.
572 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; }
573 void set_fixed_slots(int n) { _fixed_slots = n; }
574 int major_progress() const { return _major_progress; }
575 void set_inlining_progress(bool z) { _inlining_progress = z; }
576 int inlining_progress() const { return _inlining_progress; }
577 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
578 int inlining_incrementally() const { return _inlining_incrementally; }
579 void set_do_cleanup(bool z) { _do_cleanup = z; }
580 int do_cleanup() const { return _do_cleanup; }
581 void set_major_progress() { _major_progress++; }
582 void restore_major_progress(int progress) { _major_progress += progress; }
583 void clear_major_progress() { _major_progress = 0; }
584 int max_inline_size() const { return _max_inline_size; }
585 void set_freq_inline_size(int n) { _freq_inline_size = n; }
586 int freq_inline_size() const { return _freq_inline_size; }
587 void set_max_inline_size(int n) { _max_inline_size = n; }
588 bool has_loops() const { return _has_loops; }
589 void set_has_loops(bool z) { _has_loops = z; }
590 bool has_split_ifs() const { return _has_split_ifs; }
591 void set_has_split_ifs(bool z) { _has_split_ifs = z; }
592 bool has_unsafe_access() const { return _has_unsafe_access; }
593 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; }
594 bool has_stringbuilder() const { return _has_stringbuilder; }
595 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; }
596 bool has_boxed_value() const { return _has_boxed_value; }
597 void set_has_boxed_value(bool z) { _has_boxed_value = z; }
598 bool has_reserved_stack_access() const { return _has_reserved_stack_access; }
599 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
600 uint max_vector_size() const { return _max_vector_size; }
601 void set_max_vector_size(uint s) { _max_vector_size = s; }
602 bool clear_upper_avx() const { return _clear_upper_avx; }
603 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; }
604 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; }
605 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
606 bool trap_can_recompile() const { return _trap_can_recompile; }
607 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; }
608 uint decompile_count() const { return _decompile_count; }
609 void set_decompile_count(uint c) { _decompile_count = c; }
610 bool allow_range_check_smearing() const;
611 bool do_inlining() const { return _do_inlining; }
612 void set_do_inlining(bool z) { _do_inlining = z; }
613 bool do_scheduling() const { return _do_scheduling; }
614 void set_do_scheduling(bool z) { _do_scheduling = z; }
615 bool do_freq_based_layout() const{ return _do_freq_based_layout; }
616 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
617 bool do_vector_loop() const { return _do_vector_loop; }
618 void set_do_vector_loop(bool z) { _do_vector_loop = z; }
619 bool use_cmove() const { return _use_cmove; }
620 void set_use_cmove(bool z) { _use_cmove = z; }
621 bool do_aliasing() const { return _do_aliasing; }
622 bool print_assembly() const { return _print_assembly; }
623 void set_print_assembly(bool z) { _print_assembly = z; }
624 bool print_inlining() const { return _print_inlining; }
625 void set_print_inlining(bool z) { _print_inlining = z; }
626 bool print_intrinsics() const { return _print_intrinsics; }
627 void set_print_intrinsics(bool z) { _print_intrinsics = z; }
628 RTMState rtm_state() const { return _rtm_state; }
629 void set_rtm_state(RTMState s) { _rtm_state = s; }
630 bool use_rtm() const { return (_rtm_state & NoRTM) == 0; }
631 bool profile_rtm() const { return _rtm_state == ProfileRTM; }
632 uint max_node_limit() const { return (uint)_max_node_limit; }
633 void set_max_node_limit(uint n) { _max_node_limit = n; }
634 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; }
635 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; }
636 void set_flattened_accesses() { _has_flattened_accesses = true; }
637 bool flattened_accesses_share_alias() const { return _flattened_accesses_share_alias; }
638 void set_flattened_accesses_share_alias(bool z) { _flattened_accesses_share_alias = z; }
639 bool scalarize_in_safepoints() const { return _scalarize_in_safepoints; }
640 void set_scalarize_in_safepoints(bool z) { _scalarize_in_safepoints = z; }
641
642 // Support for scalarized inline type calling convention
643 bool has_scalarized_args() const { return _method != NULL && _method->has_scalarized_args(); }
644 bool needs_stack_repair() const { return _method != NULL && _method->get_Method()->c2_needs_stack_repair(); }
645
646 bool has_monitors() const { return _has_monitors; }
647 void set_has_monitors(bool v) { _has_monitors = v; }
648
649 // check the CompilerOracle for special behaviours for this compile
650 bool method_has_option(enum CompileCommand option) {
651 return method() != NULL && method()->has_option(option);
652 }
653
654 #ifndef PRODUCT
655 uint next_igv_idx() { return _igv_idx++; }
656 bool trace_opto_output() const { return _trace_opto_output; }
657 void print_ideal_ir(const char* phase_name);
658 bool should_print_ideal() const { return _directive->PrintIdealOption; }
659 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
660 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
661 int _in_dump_cnt; // Required for dumping ir nodes.
662 #endif
663 bool has_irreducible_loop() const { return _has_irreducible_loop; }
664 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
665
666 // JSR 292
667 bool has_method_handle_invokes() const { return _has_method_handle_invokes; }
668 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; }
669
670 Ticks _latest_stage_start_counter;
671
672 void begin_method();
673 void end_method();
674 bool should_print_igv(int level);
675 bool should_print_phase(CompilerPhaseType cpt);
676
677 void print_method(CompilerPhaseType cpt, int level, Node* n = nullptr);
678
679 #ifndef PRODUCT
680 void igv_print_method_to_file(const char* phase_name = "Debug", bool append = false);
681 void igv_print_method_to_network(const char* phase_name = "Debug");
682 static IdealGraphPrinter* debug_file_printer() { return _debug_file_printer; }
683 static IdealGraphPrinter* debug_network_printer() { return _debug_network_printer; }
684 #endif
685
686 int macro_count() const { return _macro_nodes.length(); }
687 int predicate_count() const { return _predicate_opaqs.length(); }
688 int skeleton_predicate_count() const { return _skeleton_predicate_opaqs.length(); }
689 int expensive_count() const { return _expensive_nodes.length(); }
690 int coarsened_count() const { return _coarsened_locks.length(); }
691
692 Node* macro_node(int idx) const { return _macro_nodes.at(idx); }
693 Node* predicate_opaque1_node(int idx) const { return _predicate_opaqs.at(idx); }
694 Node* skeleton_predicate_opaque4_node(int idx) const { return _skeleton_predicate_opaqs.at(idx); }
695 Node* expensive_node(int idx) const { return _expensive_nodes.at(idx); }
696
697 ConnectionGraph* congraph() { return _congraph;}
698 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;}
699 void add_macro_node(Node * n) {
700 //assert(n->is_macro(), "must be a macro node");
701 assert(!_macro_nodes.contains(n), "duplicate entry in expand list");
702 _macro_nodes.append(n);
703 }
704 void remove_macro_node(Node* n) {
705 // this function may be called twice for a node so we can only remove it
706 // if it's still existing.
707 _macro_nodes.remove_if_existing(n);
708 // remove from _predicate_opaqs list also if it is there
709 if (predicate_count() > 0) {
710 _predicate_opaqs.remove_if_existing(n);
711 }
712 // Remove from coarsened locks list if present
713 if (coarsened_count() > 0) {
714 remove_coarsened_lock(n);
715 }
716 }
717 void add_expensive_node(Node* n);
718 void remove_expensive_node(Node* n) {
719 _expensive_nodes.remove_if_existing(n);
720 }
721 void add_predicate_opaq(Node* n) {
722 assert(!_predicate_opaqs.contains(n), "duplicate entry in predicate opaque1");
723 assert(_macro_nodes.contains(n), "should have already been in macro list");
724 _predicate_opaqs.append(n);
725 }
726 void add_skeleton_predicate_opaq(Node* n) {
727 assert(!_skeleton_predicate_opaqs.contains(n), "duplicate entry in skeleton predicate opaque4 list");
728 _skeleton_predicate_opaqs.append(n);
729 }
730 void remove_skeleton_predicate_opaq(Node* n) {
731 if (skeleton_predicate_count() > 0) {
732 _skeleton_predicate_opaqs.remove_if_existing(n);
733 }
734 }
735 void add_coarsened_locks(GrowableArray<AbstractLockNode*>& locks);
736 void remove_coarsened_lock(Node* n);
737 bool coarsened_locks_consistent();
738
739 bool post_loop_opts_phase() { return _post_loop_opts_phase; }
740 void set_post_loop_opts_phase() { _post_loop_opts_phase = true; }
741 void reset_post_loop_opts_phase() { _post_loop_opts_phase = false; }
742
743 void record_for_post_loop_opts_igvn(Node* n);
744 void remove_from_post_loop_opts_igvn(Node* n);
745 void process_for_post_loop_opts_igvn(PhaseIterGVN& igvn);
746
747 // Keep track of inline type nodes for later processing
748 void add_inline_type(Node* n);
749 void remove_inline_type(Node* n);
750 void process_inline_types(PhaseIterGVN &igvn, bool remove = false);
751
752 void adjust_flattened_array_access_aliases(PhaseIterGVN& igvn);
753
754 void record_unstable_if_trap(UnstableIfTrap* trap);
755 bool remove_unstable_if_trap(CallStaticJavaNode* unc, bool yield);
756 void remove_useless_unstable_if_traps(Unique_Node_List &useful);
757 void process_for_unstable_if_traps(PhaseIterGVN& igvn);
758
759 void sort_macro_nodes();
760
761 // remove the opaque nodes that protect the predicates so that the unused checks and
762 // uncommon traps will be eliminated from the graph.
763 void cleanup_loop_predicates(PhaseIterGVN &igvn);
764 bool is_predicate_opaq(Node* n) {
765 return _predicate_opaqs.contains(n);
766 }
767
768 // Are there candidate expensive nodes for optimization?
769 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
770 // Check whether n1 and n2 are similar
771 static int cmp_expensive_nodes(Node* n1, Node* n2);
772 // Sort expensive nodes to locate similar expensive nodes
773 void sort_expensive_nodes();
774
775 // Compilation environment.
776 Arena* comp_arena() { return &_comp_arena; }
777 ciEnv* env() const { return _env; }
778 CompileLog* log() const { return _log; }
779 bool failing() const { return _env->failing() || _failure_reason != NULL; }
780 const char* failure_reason() const { return (_env->failing()) ? _env->failure_reason() : _failure_reason; }
781
782 bool failure_reason_is(const char* r) const {
783 return (r == _failure_reason) || (r != NULL && _failure_reason != NULL && strcmp(r, _failure_reason) == 0);
784 }
785
786 void record_failure(const char* reason);
787 void record_method_not_compilable(const char* reason) {
788 env()->record_method_not_compilable(reason);
789 // Record failure reason.
790 record_failure(reason);
791 }
792 bool check_node_count(uint margin, const char* reason) {
793 if (live_nodes() + margin > max_node_limit()) {
794 record_method_not_compilable(reason);
795 return true;
796 } else {
797 return false;
798 }
799 }
800
801 // Node management
802 uint unique() const { return _unique; }
803 uint next_unique() { return _unique++; }
804 void set_unique(uint i) { _unique = i; }
805 static int debug_idx() { return debug_only(_debug_idx)+0; }
806 static void set_debug_idx(int i) { debug_only(_debug_idx = i); }
807 Arena* node_arena() { return &_node_arena; }
808 Arena* old_arena() { return &_old_arena; }
809 RootNode* root() const { return _root; }
810 void set_root(RootNode* r) { _root = r; }
811 StartNode* start() const; // (Derived from root.)
812 void init_start(StartNode* s);
813 Node* immutable_memory();
814
815 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; }
816 Node* recent_alloc_obj() const { return _recent_alloc_obj; }
817 void set_recent_alloc(Node* ctl, Node* obj) {
818 _recent_alloc_ctl = ctl;
819 _recent_alloc_obj = obj;
820 }
821 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return;
822 _dead_node_count++;
823 }
824 void reset_dead_node_list() { _dead_node_list.reset();
825 _dead_node_count = 0;
826 }
827 uint live_nodes() const {
828 int val = _unique - _dead_node_count;
829 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
830 return (uint) val;
831 }
832 #ifdef ASSERT
833 void set_phase_optimize_finished() { _phase_optimize_finished = true; }
834 bool phase_optimize_finished() const { return _phase_optimize_finished; }
835 uint count_live_nodes_by_graph_walk();
836 void print_missing_nodes();
837 #endif
838
839 // Record modified nodes to check that they are put on IGVN worklist
840 void record_modified_node(Node* n) NOT_DEBUG_RETURN;
841 void remove_modified_node(Node* n) NOT_DEBUG_RETURN;
842 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } )
843
844 MachConstantBaseNode* mach_constant_base_node();
845 bool has_mach_constant_base_node() const { return _mach_constant_base_node != NULL; }
846 // Generated by adlc, true if CallNode requires MachConstantBase.
847 bool needs_deep_clone_jvms();
848
849 // Handy undefined Node
850 Node* top() const { return _top; }
851
852 // these are used by guys who need to know about creation and transformation of top:
853 Node* cached_top_node() { return _top; }
854 void set_cached_top_node(Node* tn);
855
856 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
857 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
858 Node_Notes* default_node_notes() const { return _default_node_notes; }
859 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
860
861 Node_Notes* node_notes_at(int idx) {
862 return locate_node_notes(_node_note_array, idx, false);
863 }
864 inline bool set_node_notes_at(int idx, Node_Notes* value);
865
866 // Copy notes from source to dest, if they exist.
867 // Overwrite dest only if source provides something.
868 // Return true if information was moved.
869 bool copy_node_notes_to(Node* dest, Node* source);
870
871 // Workhorse function to sort out the blocked Node_Notes array:
872 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
873 int idx, bool can_grow = false);
874
875 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
876
877 // Type management
878 Arena* type_arena() { return _type_arena; }
879 Dict* type_dict() { return _type_dict; }
880 size_t type_last_size() { return _type_last_size; }
881 int num_alias_types() { return _num_alias_types; }
882
883 void init_type_arena() { _type_arena = &_Compile_types; }
884 void set_type_arena(Arena* a) { _type_arena = a; }
885 void set_type_dict(Dict* d) { _type_dict = d; }
886 void set_type_last_size(size_t sz) { _type_last_size = sz; }
887
888 const TypeFunc* last_tf(ciMethod* m) {
889 return (m == _last_tf_m) ? _last_tf : NULL;
890 }
891 void set_last_tf(ciMethod* m, const TypeFunc* tf) {
892 assert(m != NULL || tf == NULL, "");
893 _last_tf_m = m;
894 _last_tf = tf;
895 }
896
897 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
898 AliasType* alias_type(const TypePtr* adr_type, ciField* field = NULL, bool uncached = false) { return find_alias_type(adr_type, false, field, uncached); }
899 bool have_alias_type(const TypePtr* adr_type);
900 AliasType* alias_type(ciField* field);
901
902 int get_alias_index(const TypePtr* at, bool uncached = false) { return alias_type(at, NULL, uncached)->index(); }
903 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); }
904 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); }
905
906 // Building nodes
907 void rethrow_exceptions(JVMState* jvms);
908 void return_values(JVMState* jvms);
909 JVMState* build_start_state(StartNode* start, const TypeFunc* tf);
910
911 // Decide how to build a call.
912 // The profile factor is a discount to apply to this site's interp. profile.
913 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
914 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = NULL,
915 bool allow_intrinsics = true);
916 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
917 return should_delay_string_inlining(call_method, jvms) ||
918 should_delay_boxing_inlining(call_method, jvms) ||
919 should_delay_vector_inlining(call_method, jvms);
920 }
921 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
922 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
923 bool should_delay_vector_inlining(ciMethod* call_method, JVMState* jvms);
924 bool should_delay_vector_reboxing_inlining(ciMethod* call_method, JVMState* jvms);
925
926 // Helper functions to identify inlining potential at call-site
927 ciMethod* optimize_virtual_call(ciMethod* caller, ciInstanceKlass* klass,
928 ciKlass* holder, ciMethod* callee,
929 const TypeOopPtr* receiver_type, bool is_virtual,
930 bool &call_does_dispatch, int &vtable_index,
931 bool check_access = true);
932 ciMethod* optimize_inlining(ciMethod* caller, ciInstanceKlass* klass, ciKlass* holder,
933 ciMethod* callee, const TypeOopPtr* receiver_type,
934 bool check_access = true);
935
936 // Report if there were too many traps at a current method and bci.
937 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
938 // If there is no MDO at all, report no trap unless told to assume it.
939 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
940 // This version, unspecific to a particular bci, asks if
941 // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
942 bool too_many_traps(Deoptimization::DeoptReason reason,
943 // Privately used parameter for logging:
944 ciMethodData* logmd = NULL);
945 // Report if there were too many recompiles at a method and bci.
946 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
947 // Report if there were too many traps or recompiles at a method and bci.
948 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
949 return too_many_traps(method, bci, reason) ||
950 too_many_recompiles(method, bci, reason);
951 }
952 // Return a bitset with the reasons where deoptimization is allowed,
953 // i.e., where there were not too many uncommon traps.
954 int _allowed_reasons;
955 int allowed_deopt_reasons() { return _allowed_reasons; }
956 void set_allowed_deopt_reasons();
957
958 // Parsing, optimization
959 PhaseGVN* initial_gvn() { return _initial_gvn; }
960 Unique_Node_List* for_igvn() { return _for_igvn; }
961 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List.
962 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; }
963 void set_for_igvn(Unique_Node_List *for_igvn) { _for_igvn = for_igvn; }
964
965 // Replace n by nn using initial_gvn, calling hash_delete and
966 // record_for_igvn as needed.
967 void gvn_replace_by(Node* n, Node* nn);
968
969
970 void identify_useful_nodes(Unique_Node_List &useful);
971 void update_dead_node_list(Unique_Node_List &useful);
972 void disconnect_useless_nodes(Unique_Node_List &useful, Unique_Node_List* worklist);
973
974 void remove_useless_node(Node* dead);
975
976 // Record this CallGenerator for inlining at the end of parsing.
977 void add_late_inline(CallGenerator* cg) {
978 _late_inlines.insert_before(_late_inlines_pos, cg);
979 _late_inlines_pos++;
980 }
981
982 void prepend_late_inline(CallGenerator* cg) {
983 _late_inlines.insert_before(0, cg);
984 }
985
986 void add_string_late_inline(CallGenerator* cg) {
987 _string_late_inlines.push(cg);
988 }
989
990 void add_boxing_late_inline(CallGenerator* cg) {
991 _boxing_late_inlines.push(cg);
992 }
993
994 void add_vector_reboxing_late_inline(CallGenerator* cg) {
995 _vector_reboxing_late_inlines.push(cg);
996 }
997
998 void remove_useless_nodes (GrowableArray<Node*>& node_list, Unique_Node_List &useful);
999
1000 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1001 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Node* dead);
1002
1003 void remove_useless_coarsened_locks(Unique_Node_List& useful);
1004
1005 void process_print_inlining();
1006 void dump_print_inlining();
1007
1008 bool over_inlining_cutoff() const {
1009 if (!inlining_incrementally()) {
1010 return unique() > (uint)NodeCountInliningCutoff;
1011 } else {
1012 // Give some room for incremental inlining algorithm to "breathe"
1013 // and avoid thrashing when live node count is close to the limit.
1014 // Keep in mind that live_nodes() isn't accurate during inlining until
1015 // dead node elimination step happens (see Compile::inline_incrementally).
1016 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10;
1017 }
1018 }
1019
1020 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; }
1021 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--; }
1022 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; }
1023
1024 bool inline_incrementally_one();
1025 void inline_incrementally_cleanup(PhaseIterGVN& igvn);
1026 void inline_incrementally(PhaseIterGVN& igvn);
1027 void inline_string_calls(bool parse_time);
1028 void inline_boxing_calls(PhaseIterGVN& igvn);
1029 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1030 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn);
1031
1032 void inline_vector_reboxing_calls();
1033 bool has_vbox_nodes();
1034
1035 void process_late_inline_calls_no_inline(PhaseIterGVN& igvn);
1036
1037 // Matching, CFG layout, allocation, code generation
1038 PhaseCFG* cfg() { return _cfg; }
1039 bool has_java_calls() const { return _java_calls > 0; }
1040 int java_calls() const { return _java_calls; }
1041 int inner_loops() const { return _inner_loops; }
1042 Matcher* matcher() { return _matcher; }
1043 PhaseRegAlloc* regalloc() { return _regalloc; }
1044 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; }
1045 Arena* indexSet_arena() { return _indexSet_arena; }
1046 void* indexSet_free_block_list() { return _indexSet_free_block_list; }
1047 DebugInformationRecorder* debug_info() { return env()->debug_info(); }
1048
1049 void update_interpreter_frame_size(int size) {
1050 if (_interpreter_frame_size < size) {
1051 _interpreter_frame_size = size;
1052 }
1053 }
1054
1055 void set_matcher(Matcher* m) { _matcher = m; }
1056 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; }
1057 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; }
1058 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; }
1059
1060 void set_java_calls(int z) { _java_calls = z; }
1061 void set_inner_loops(int z) { _inner_loops = z; }
1062
1063 Dependencies* dependencies() { return env()->dependencies(); }
1064
1065 // Major entry point. Given a Scope, compile the associated method.
1066 // For normal compilations, entry_bci is InvocationEntryBci. For on stack
1067 // replacement, entry_bci indicates the bytecode for which to compile a
1068 // continuation.
1069 Compile(ciEnv* ci_env, ciMethod* target,
1070 int entry_bci, Options options, DirectiveSet* directive);
1071
1072 // Second major entry point. From the TypeFunc signature, generate code
1073 // to pass arguments from the Java calling convention to the C calling
1074 // convention.
1075 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1076 address stub_function, const char *stub_name,
1077 int is_fancy_jump, bool pass_tls,
1078 bool return_pc, DirectiveSet* directive);
1079
1080 ~Compile() {
1081 delete _print_inlining_stream;
1082 };
1083
1084 // Are we compiling a method?
1085 bool has_method() { return method() != NULL; }
1086
1087 // Maybe print some information about this compile.
1088 void print_compile_messages();
1089
1090 // Final graph reshaping, a post-pass after the regular optimizer is done.
1091 bool final_graph_reshaping();
1092
1093 // returns true if adr is completely contained in the given alias category
1094 bool must_alias(const TypePtr* adr, int alias_idx);
1095
1096 // returns true if adr overlaps with the given alias category
1097 bool can_alias(const TypePtr* adr, int alias_idx);
1098
1099 // Stack slots that may be unused by the calling convention but must
1100 // otherwise be preserved. On Intel this includes the return address.
1101 // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1102 uint in_preserve_stack_slots() {
1103 return SharedRuntime::in_preserve_stack_slots();
1104 }
1105
1106 // "Top of Stack" slots that may be unused by the calling convention but must
1107 // otherwise be preserved.
1108 // On Intel these are not necessary and the value can be zero.
1109 static uint out_preserve_stack_slots() {
1110 return SharedRuntime::out_preserve_stack_slots();
1111 }
1112
1113 // Number of outgoing stack slots killed above the out_preserve_stack_slots
1114 // for calls to C. Supports the var-args backing area for register parms.
1115 uint varargs_C_out_slots_killed() const;
1116
1117 // Number of Stack Slots consumed by a synchronization entry
1118 int sync_stack_slots() const;
1119
1120 // Compute the name of old_SP. See <arch>.ad for frame layout.
1121 OptoReg::Name compute_old_SP();
1122
1123 private:
1124 // Phase control:
1125 void Init(bool aliasing); // Prepare for a single compilation
1126 void Optimize(); // Given a graph, optimize it
1127 void Code_Gen(); // Generate code from a graph
1128
1129 // Management of the AliasType table.
1130 void grow_alias_types();
1131 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1132 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1133 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field, bool uncached = false);
1134
1135 void verify_top(Node*) const PRODUCT_RETURN;
1136
1137 // Intrinsic setup.
1138 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor
1139 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper
1140 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn
1141 void register_intrinsic(CallGenerator* cg); // update fn
1142
1143 #ifndef PRODUCT
1144 static juint _intrinsic_hist_count[];
1145 static jubyte _intrinsic_hist_flags[];
1146 #endif
1147 // Function calls made by the public function final_graph_reshaping.
1148 // No need to be made public as they are not called elsewhere.
1149 void final_graph_reshaping_impl(Node *n, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1150 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop, Unique_Node_List& dead_nodes);
1151 void final_graph_reshaping_walk(Node_Stack& nstack, Node* root, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1152 void eliminate_redundant_card_marks(Node* n);
1153
1154 // Logic cone optimization.
1155 void optimize_logic_cones(PhaseIterGVN &igvn);
1156 void collect_logic_cone_roots(Unique_Node_List& list);
1157 void process_logic_cone_root(PhaseIterGVN &igvn, Node* n, VectorSet& visited);
1158 bool compute_logic_cone(Node* n, Unique_Node_List& partition, Unique_Node_List& inputs);
1159 uint compute_truth_table(Unique_Node_List& partition, Unique_Node_List& inputs);
1160 uint eval_macro_logic_op(uint func, uint op1, uint op2, uint op3);
1161 Node* xform_to_MacroLogicV(PhaseIterGVN &igvn, const TypeVect* vt, Unique_Node_List& partitions, Unique_Node_List& inputs);
1162 void check_no_dead_use() const NOT_DEBUG_RETURN;
1163
1164 public:
1165
1166 // Note: Histogram array size is about 1 Kb.
1167 enum { // flag bits:
1168 _intrinsic_worked = 1, // succeeded at least once
1169 _intrinsic_failed = 2, // tried it but it failed
1170 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1171 _intrinsic_virtual = 8, // was seen in the virtual form (rare)
1172 _intrinsic_both = 16 // was seen in the non-virtual form (usual)
1173 };
1174 // Update histogram. Return boolean if this is a first-time occurrence.
1175 static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1176 bool is_virtual, int flags) PRODUCT_RETURN0;
1177 static void print_intrinsic_statistics() PRODUCT_RETURN;
1178
1179 // Graph verification code
1180 // Walk the node list, verifying that there is a one-to-one
1181 // correspondence between Use-Def edges and Def-Use edges
1182 // The option no_dead_code enables stronger checks that the
1183 // graph is strongly connected from root in both directions.
1184 void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN;
1185
1186 // Verify bi-directional correspondence of edges
1187 void verify_bidirectional_edges(Unique_Node_List &visited);
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, bool skip = StressReflectiveCode);
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