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 NodeHash;
71 class NodeCloneInfo;
72 class OptoReg;
73 class ParsePredicateNode;
74 class PhaseCFG;
75 class PhaseGVN;
76 class PhaseIterGVN;
77 class PhaseRegAlloc;
78 class PhaseCCP;
79 class PhaseOutput;
80 class RootNode;
81 class relocInfo;
82 class StartNode;
83 class SafePointNode;
84 class JVMState;
85 class Type;
86 class TypeInt;
87 class TypeInteger;
88 class TypeKlassPtr;
89 class TypePtr;
90 class TypeOopPtr;
91 class TypeFunc;
92 class TypeVect;
93 class Type_Array;
94 class Unique_Node_List;
95 class UnstableIfTrap;
96 class InlineTypeNode;
97 class nmethod;
98 class Node_Stack;
99 struct Final_Reshape_Counts;
100 class VerifyMeetResult;
101
102 enum LoopOptsMode {
103 LoopOptsDefault,
104 LoopOptsNone,
105 LoopOptsMaxUnroll,
106 LoopOptsShenandoahExpand,
107 LoopOptsShenandoahPostExpand,
108 LoopOptsSkipSplitIf,
109 LoopOptsVerify
110 };
111
112 // The type of all node counts and indexes.
113 // It must hold at least 16 bits, but must also be fast to load and store.
114 // This type, if less than 32 bits, could limit the number of possible nodes.
115 // (To make this type platform-specific, move to globalDefinitions_xxx.hpp.)
116 typedef unsigned int node_idx_t;
117
118 class NodeCloneInfo {
119 private:
120 uint64_t _idx_clone_orig;
121 public:
122
123 void set_idx(node_idx_t idx) {
124 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx;
125 }
126 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); }
127
128 void set_gen(int generation) {
129 uint64_t g = (uint64_t)generation << 32;
130 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g;
131 }
132 int gen() const { return (int)(_idx_clone_orig >> 32); }
133
134 void set(uint64_t x) { _idx_clone_orig = x; }
135 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); }
136 uint64_t get() const { return _idx_clone_orig; }
137
138 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {}
139 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); }
140
141 void dump_on(outputStream* st) const;
142 };
143
144 class CloneMap {
145 friend class Compile;
146 private:
147 bool _debug;
148 Dict* _dict;
149 int _clone_idx; // current cloning iteration/generation in loop unroll
150 public:
151 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy
152 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; }
153 Dict* dict() const { return _dict; }
154 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == nullptr, "key existed"); _dict->Insert(_2p(key), (void*)val); }
155 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); }
156 void remove(node_idx_t key) { _dict->Delete(_2p(key)); }
157 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); }
158 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); }
159 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); }
160 int gen(const void* k) const { return gen(_2_node_idx_t(k)); }
161 int max_gen() const;
162 void clone(Node* old, Node* nnn, int gen);
163 void verify_insert_and_clone(Node* old, Node* nnn, int gen);
164 void dump(node_idx_t key, outputStream* st) const;
165
166 int clone_idx() const { return _clone_idx; }
167 void set_clone_idx(int x) { _clone_idx = x; }
168 bool is_debug() const { return _debug; }
169 void set_debug(bool debug) { _debug = debug; }
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 _do_reduce_allocation_merges; // Do try to reduce allocation merges.
183 const bool _eliminate_boxing; // Do boxing elimination.
184 const bool _do_locks_coarsening; // Do locks coarsening
185 const bool _do_superword; // Do SuperWord
186 const bool _install_code; // Install the code that was compiled
187 public:
188 Options(bool subsume_loads,
189 bool do_escape_analysis,
190 bool do_iterative_escape_analysis,
191 bool do_reduce_allocation_merges,
192 bool eliminate_boxing,
193 bool do_locks_coarsening,
194 bool do_superword,
195 bool install_code) :
196 _subsume_loads(subsume_loads),
197 _do_escape_analysis(do_escape_analysis),
198 _do_iterative_escape_analysis(do_iterative_escape_analysis),
199 _do_reduce_allocation_merges(do_reduce_allocation_merges),
200 _eliminate_boxing(eliminate_boxing),
201 _do_locks_coarsening(do_locks_coarsening),
202 _do_superword(do_superword),
203 _install_code(install_code) {
204 }
205
206 static Options for_runtime_stub() {
207 return Options(
208 /* subsume_loads = */ true,
209 /* do_escape_analysis = */ false,
210 /* do_iterative_escape_analysis = */ false,
211 /* do_reduce_allocation_merges = */ false,
212 /* eliminate_boxing = */ false,
213 /* do_lock_coarsening = */ false,
214 /* do_superword = */ true,
215 /* install_code = */ true
216 );
217 }
218 };
219
220 //------------------------------Compile----------------------------------------
221 // This class defines a top-level Compiler invocation.
222
223 class Compile : public Phase {
224 friend class VMStructs;
225
226 public:
227 // Fixed alias indexes. (See also MergeMemNode.)
228 enum {
229 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value)
230 AliasIdxBot = 2, // pseudo-index, aliases to everything
231 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM
232 };
233
234 // Variant of TraceTime(nullptr, &_t_accumulator, CITime);
235 // Integrated with logging. If logging is turned on, and CITimeVerbose is true,
236 // then brackets are put into the log, with time stamps and node counts.
237 // (The time collection itself is always conditionalized on CITime.)
238 class TracePhase : public TraceTime {
239 private:
240 Compile* _compile;
241 CompileLog* _log;
242 const char* _phase_name;
243 bool _dolog;
244 public:
245 TracePhase(const char* name, elapsedTimer* accumulator);
246 ~TracePhase();
247 };
248
249 // Information per category of alias (memory slice)
250 class AliasType {
251 private:
252 friend class Compile;
253
254 int _index; // unique index, used with MergeMemNode
255 const TypePtr* _adr_type; // normalized address type
256 ciField* _field; // relevant instance field, or null if none
257 const Type* _element; // relevant array element type, or null if none
258 bool _is_rewritable; // false if the memory is write-once only
259 int _general_index; // if this is type is an instance, the general
260 // type that this is an instance of
261
262 void Init(int i, const TypePtr* at);
263
264 public:
265 int index() const { return _index; }
266 const TypePtr* adr_type() const { return _adr_type; }
267 ciField* field() const { return _field; }
268 const Type* element() const { return _element; }
269 bool is_rewritable() const { return _is_rewritable; }
270 bool is_volatile() const { return (_field ? _field->is_volatile() : false); }
271 int general_index() const { return (_general_index != 0) ? _general_index : _index; }
272
273 void set_rewritable(bool z) { _is_rewritable = z; }
274 void set_field(ciField* f) {
275 assert(!_field,"");
276 _field = f;
277 if (f->is_final() || f->is_stable()) {
278 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
279 _is_rewritable = false;
280 }
281 }
282 void set_element(const Type* e) {
283 assert(_element == nullptr, "");
284 _element = e;
285 }
286
287 BasicType basic_type() const;
288
289 void print_on(outputStream* st) PRODUCT_RETURN;
290 };
291
292 enum {
293 logAliasCacheSize = 6,
294 AliasCacheSize = (1<<logAliasCacheSize)
295 };
296 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type
297 enum {
298 trapHistLength = MethodData::_trap_hist_limit
299 };
300
301 private:
302 // Fixed parameters to this compilation.
303 const int _compile_id;
304 const Options _options; // Compilation options
305 ciMethod* _method; // The method being compiled.
306 int _entry_bci; // entry bci for osr methods.
307 const TypeFunc* _tf; // My kind of signature
308 InlineTree* _ilt; // Ditto (temporary).
309 address _stub_function; // VM entry for stub being compiled, or null
310 const char* _stub_name; // Name of stub or adapter being compiled, or null
311 address _stub_entry_point; // Compile code entry for generated stub, or null
312
313 // Control of this compilation.
314 int _max_inline_size; // Max inline size for this compilation
315 int _freq_inline_size; // Max hot method inline size for this compilation
316 int _fixed_slots; // count of frame slots not allocated by the register
317 // allocator i.e. locks, original deopt pc, etc.
318 uintx _max_node_limit; // Max unique node count during a single compilation.
319
320 bool _post_loop_opts_phase; // Loop opts are finished.
321
322 int _major_progress; // Count of something big happening
323 bool _inlining_progress; // progress doing incremental inlining?
324 bool _inlining_incrementally;// Are we doing incremental inlining (post parse)
325 bool _do_cleanup; // Cleanup is needed before proceeding with incremental inlining
326 bool _has_loops; // True if the method _may_ have some loops
327 bool _has_split_ifs; // True if the method _may_ have some split-if
328 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores.
329 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated
330 bool _has_boxed_value; // True if a boxed object is allocated
331 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess
332 bool _has_circular_inline_type; // True if method loads an inline type with a circular, non-flat field
333 uint _max_vector_size; // Maximum size of generated vectors
334 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper
335 uint _trap_hist[trapHistLength]; // Cumulative traps
336 bool _trap_can_recompile; // Have we emitted a recompiling trap?
337 uint _decompile_count; // Cumulative decompilation counts.
338 bool _do_inlining; // True if we intend to do inlining
339 bool _do_scheduling; // True if we intend to do scheduling
340 bool _do_freq_based_layout; // True if we intend to do frequency based block layout
341 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations
342 bool _use_cmove; // True if CMove should be used without profitability analysis
343 bool _do_aliasing; // True if we intend to do aliasing
344 bool _print_assembly; // True if we should dump assembly code for this compilation
345 bool _print_inlining; // True if we should print inlining for this compilation
346 bool _print_intrinsics; // True if we should print intrinsics for this compilation
347 #ifndef PRODUCT
348 uint _igv_idx; // Counter for IGV node identifiers
349 bool _trace_opto_output;
350 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
351 #endif
352 bool _has_irreducible_loop; // Found irreducible loops
353 // JSR 292
354 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes.
355 bool _has_monitors; // Metadata transfered to nmethod to enable Continuations lock-detection fastpath
356 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry
357 RTMState _rtm_state; // State of Restricted Transactional Memory usage
358 int _loop_opts_cnt; // loop opts round
359 bool _has_flat_accesses; // Any known flat array accesses?
360 bool _flat_accesses_share_alias; // Initially all flat array share a single slice
361 bool _scalarize_in_safepoints; // Scalarize inline types in safepoint debug info
362 uint _stress_seed; // Seed for stress testing
363
364 // Compilation environment.
365 Arena _comp_arena; // Arena with lifetime equivalent to Compile
366 void* _barrier_set_state; // Potential GC barrier state for Compile
367 ciEnv* _env; // CI interface
368 DirectiveSet* _directive; // Compiler directive
369 CompileLog* _log; // from CompilerThread
370 const char* _failure_reason; // for record_failure/failing pattern
371 GrowableArray<CallGenerator*> _intrinsics; // List of intrinsics.
372 GrowableArray<Node*> _macro_nodes; // List of nodes which need to be expanded before matching.
373 GrowableArray<ParsePredicateNode*> _parse_predicates; // List of Parse Predicates.
374 GrowableArray<Node*> _template_assertion_predicate_opaqs; // List of Opaque4 nodes for Template Assertion Predicates.
375 GrowableArray<Node*> _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
376 GrowableArray<Node*> _for_post_loop_igvn; // List of nodes for IGVN after loop opts are over
377 GrowableArray<Node*> _inline_type_nodes; // List of InlineType nodes
378 GrowableArray<UnstableIfTrap*> _unstable_if_traps; // List of ifnodes after IGVN
379 GrowableArray<Node_List*> _coarsened_locks; // List of coarsened Lock and Unlock nodes
380 ConnectionGraph* _congraph;
381 #ifndef PRODUCT
382 IdealGraphPrinter* _igv_printer;
383 static IdealGraphPrinter* _debug_file_printer;
384 static IdealGraphPrinter* _debug_network_printer;
385 #endif
386
387
388 // Node management
389 uint _unique; // Counter for unique Node indices
390 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N).
391 // So use this to keep count and make the call O(1).
392 VectorSet _dead_node_list; // Set of dead nodes
393 DEBUG_ONLY(Unique_Node_List* _modified_nodes;) // List of nodes which inputs were modified
394 DEBUG_ONLY(bool _phase_optimize_finished;) // Used for live node verification while creating new nodes
395
396 // Arenas for new-space and old-space nodes.
397 // Swapped between using _node_arena.
398 // The lifetime of the old-space nodes is during xform.
399 Arena _node_arena_one;
400 Arena _node_arena_two;
401 Arena* _node_arena;
402 public:
403 Arena* swap_old_and_new() {
404 Arena* filled_arena_ptr = _node_arena;
405 Arena* old_arena_ptr = old_arena();
406 old_arena_ptr->destruct_contents();
407 _node_arena = old_arena_ptr;
408 return filled_arena_ptr;
409 }
410 private:
411 RootNode* _root; // Unique root of compilation, or null after bail-out.
412 Node* _top; // Unique top node. (Reset by various phases.)
413
414 Node* _immutable_memory; // Initial memory state
415
416 Node* _recent_alloc_obj;
417 Node* _recent_alloc_ctl;
418
419 // Constant table
420 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton.
421
422
423 // Blocked array of debugging and profiling information,
424 // tracked per node.
425 enum { _log2_node_notes_block_size = 8,
426 _node_notes_block_size = (1<<_log2_node_notes_block_size)
427 };
428 GrowableArray<Node_Notes*>* _node_note_array;
429 Node_Notes* _default_node_notes; // default notes for new nodes
430
431 // After parsing and every bulk phase we hang onto the Root instruction.
432 // The RootNode instruction is where the whole program begins. It produces
433 // the initial Control and BOTTOM for everybody else.
434
435 // Type management
436 Arena _Compile_types; // Arena for all types
437 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared()
438 Dict* _type_dict; // Intern table
439 CloneMap _clone_map; // used for recording history of cloned nodes
440 size_t _type_last_size; // Last allocation size (see Type::operator new/delete)
441 ciMethod* _last_tf_m; // Cache for
442 const TypeFunc* _last_tf; // TypeFunc::make
443 AliasType** _alias_types; // List of alias types seen so far.
444 int _num_alias_types; // Logical length of _alias_types
445 int _max_alias_types; // Physical length of _alias_types
446 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
447
448 // Parsing, optimization
449 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN
450
451 // Shared worklist for all IGVN rounds. Nodes can be pushed to it at any time.
452 // If pushed outside IGVN, the Node is processed in the next IGVN round.
453 Unique_Node_List* _igvn_worklist;
454
455 // Shared type array for GVN, IGVN and CCP. It maps node idx -> Type*.
456 Type_Array* _types;
457
458 // Shared node hash table for GVN, IGVN and CCP.
459 NodeHash* _node_hash;
460
461 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after main parsing has finished.
462 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
463 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
464
465 GrowableArray<CallGenerator*> _vector_reboxing_late_inlines; // same but for vector reboxing operations
466
467 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
468 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending
469
470 // "MemLimit" directive was specified and the memory limit was hit during compilation
471 bool _oom;
472
473 // Inlining may not happen in parse order which would make
474 // PrintInlining output confusing. Keep track of PrintInlining
475 // pieces in order.
476 class PrintInliningBuffer : public CHeapObj<mtCompiler> {
477 private:
478 CallGenerator* _cg;
479 stringStream _ss;
480 static const size_t default_stream_buffer_size = 128;
481
482 public:
483 PrintInliningBuffer()
484 : _cg(nullptr), _ss(default_stream_buffer_size) {}
485
486 stringStream* ss() { return &_ss; }
487 CallGenerator* cg() { return _cg; }
488 void set_cg(CallGenerator* cg) { _cg = cg; }
489 };
490
491 stringStream* _print_inlining_stream;
492 GrowableArray<PrintInliningBuffer*>* _print_inlining_list;
493 int _print_inlining_idx;
494 char* _print_inlining_output;
495
496 // Only keep nodes in the expensive node list that need to be optimized
497 void cleanup_expensive_nodes(PhaseIterGVN &igvn);
498 // Use for sorting expensive nodes to bring similar nodes together
499 static int cmp_expensive_nodes(Node** n1, Node** n2);
500 // Expensive nodes list already sorted?
501 bool expensive_nodes_sorted() const;
502 // Remove the speculative part of types and clean up the graph
503 void remove_speculative_types(PhaseIterGVN &igvn);
504
505 void* _replay_inline_data; // Pointer to data loaded from file
506
507 void print_inlining_init();
508 void print_inlining_reinit();
509 void print_inlining_commit();
510 void print_inlining_push();
511 PrintInliningBuffer* print_inlining_current();
512
513 void log_late_inline_failure(CallGenerator* cg, const char* msg);
514 DEBUG_ONLY(bool _exception_backedge;)
515
516 void record_method_not_compilable_oom();
517
518 public:
519
520 void* barrier_set_state() const { return _barrier_set_state; }
521
522 stringStream* print_inlining_stream() {
523 assert(print_inlining() || print_intrinsics(), "PrintInlining off?");
524 return _print_inlining_stream;
525 }
526
527 void print_inlining_update(CallGenerator* cg);
528 void print_inlining_update_delayed(CallGenerator* cg);
529 void print_inlining_move_to(CallGenerator* cg);
530 void print_inlining_assert_ready();
531 void print_inlining_reset();
532
533 void print_inlining(ciMethod* method, int inline_level, int bci, InliningResult result, const char* msg = nullptr) {
534 stringStream ss;
535 CompileTask::print_inlining_inner(&ss, method, inline_level, bci, result, msg);
536 print_inlining_stream()->print("%s", ss.freeze());
537 }
538
539 #ifndef PRODUCT
540 IdealGraphPrinter* igv_printer() { return _igv_printer; }
541 #endif
542
543 void log_late_inline(CallGenerator* cg);
544 void log_inline_id(CallGenerator* cg);
545 void log_inline_failure(const char* msg);
546
547 void* replay_inline_data() const { return _replay_inline_data; }
548
549 // Dump inlining replay data to the stream.
550 void dump_inline_data(outputStream* out);
551 void dump_inline_data_reduced(outputStream* out);
552
553 private:
554 // Matching, CFG layout, allocation, code generation
555 PhaseCFG* _cfg; // Results of CFG finding
556 int _java_calls; // Number of java calls in the method
557 int _inner_loops; // Number of inner loops in the method
558 Matcher* _matcher; // Engine to map ideal to machine instructions
559 PhaseRegAlloc* _regalloc; // Results of register allocation.
560 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout)
561 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin
562 void* _indexSet_free_block_list; // free list of IndexSet bit blocks
563 int _interpreter_frame_size;
564
565 PhaseOutput* _output;
566
567 public:
568 // Accessors
569
570 // The Compile instance currently active in this (compiler) thread.
571 static Compile* current() {
572 return (Compile*) ciEnv::current()->compiler_data();
573 }
574
575 int interpreter_frame_size() const { return _interpreter_frame_size; }
576
577 PhaseOutput* output() const { return _output; }
578 void set_output(PhaseOutput* o) { _output = o; }
579
580 // ID for this compilation. Useful for setting breakpoints in the debugger.
581 int compile_id() const { return _compile_id; }
582 DirectiveSet* directive() const { return _directive; }
583
584 // Does this compilation allow instructions to subsume loads? User
585 // instructions that subsume a load may result in an unschedulable
586 // instruction sequence.
587 bool subsume_loads() const { return _options._subsume_loads; }
588 /** Do escape analysis. */
589 bool do_escape_analysis() const { return _options._do_escape_analysis; }
590 bool do_iterative_escape_analysis() const { return _options._do_iterative_escape_analysis; }
591 bool do_reduce_allocation_merges() const { return _options._do_reduce_allocation_merges; }
592 /** Do boxing elimination. */
593 bool eliminate_boxing() const { return _options._eliminate_boxing; }
594 /** Do aggressive boxing elimination. */
595 bool aggressive_unboxing() const { return _options._eliminate_boxing && AggressiveUnboxing; }
596 bool should_install_code() const { return _options._install_code; }
597 /** Do locks coarsening. */
598 bool do_locks_coarsening() const { return _options._do_locks_coarsening; }
599 bool do_superword() const { return _options._do_superword; }
600
601 // Other fixed compilation parameters.
602 ciMethod* method() const { return _method; }
603 int entry_bci() const { return _entry_bci; }
604 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; }
605 bool is_method_compilation() const { return (_method != nullptr && !_method->flags().is_native()); }
606 const TypeFunc* tf() const { assert(_tf!=nullptr, ""); return _tf; }
607 void init_tf(const TypeFunc* tf) { assert(_tf==nullptr, ""); _tf = tf; }
608 InlineTree* ilt() const { return _ilt; }
609 address stub_function() const { return _stub_function; }
610 const char* stub_name() const { return _stub_name; }
611 address stub_entry_point() const { return _stub_entry_point; }
612 void set_stub_entry_point(address z) { _stub_entry_point = z; }
613
614 // Control of this compilation.
615 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; }
616 void set_fixed_slots(int n) { _fixed_slots = n; }
617 int major_progress() const { return _major_progress; }
618 void set_inlining_progress(bool z) { _inlining_progress = z; }
619 int inlining_progress() const { return _inlining_progress; }
620 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
621 int inlining_incrementally() const { return _inlining_incrementally; }
622 void set_do_cleanup(bool z) { _do_cleanup = z; }
623 int do_cleanup() const { return _do_cleanup; }
624 void set_major_progress() { _major_progress++; }
625 void restore_major_progress(int progress) { _major_progress += progress; }
626 void clear_major_progress() { _major_progress = 0; }
627 int max_inline_size() const { return _max_inline_size; }
628 void set_freq_inline_size(int n) { _freq_inline_size = n; }
629 int freq_inline_size() const { return _freq_inline_size; }
630 void set_max_inline_size(int n) { _max_inline_size = n; }
631 bool has_loops() const { return _has_loops; }
632 void set_has_loops(bool z) { _has_loops = z; }
633 bool has_split_ifs() const { return _has_split_ifs; }
634 void set_has_split_ifs(bool z) { _has_split_ifs = z; }
635 bool has_unsafe_access() const { return _has_unsafe_access; }
636 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; }
637 bool has_stringbuilder() const { return _has_stringbuilder; }
638 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; }
639 bool has_boxed_value() const { return _has_boxed_value; }
640 void set_has_boxed_value(bool z) { _has_boxed_value = z; }
641 bool has_reserved_stack_access() const { return _has_reserved_stack_access; }
642 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
643 bool has_circular_inline_type() const { return _has_circular_inline_type; }
644 void set_has_circular_inline_type(bool z) { _has_circular_inline_type = z; }
645 uint max_vector_size() const { return _max_vector_size; }
646 void set_max_vector_size(uint s) { _max_vector_size = s; }
647 bool clear_upper_avx() const { return _clear_upper_avx; }
648 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; }
649 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; }
650 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
651 bool trap_can_recompile() const { return _trap_can_recompile; }
652 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; }
653 uint decompile_count() const { return _decompile_count; }
654 void set_decompile_count(uint c) { _decompile_count = c; }
655 bool allow_range_check_smearing() const;
656 bool do_inlining() const { return _do_inlining; }
657 void set_do_inlining(bool z) { _do_inlining = z; }
658 bool do_scheduling() const { return _do_scheduling; }
659 void set_do_scheduling(bool z) { _do_scheduling = z; }
660 bool do_freq_based_layout() const{ return _do_freq_based_layout; }
661 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
662 bool do_vector_loop() const { return _do_vector_loop; }
663 void set_do_vector_loop(bool z) { _do_vector_loop = z; }
664 bool use_cmove() const { return _use_cmove; }
665 void set_use_cmove(bool z) { _use_cmove = z; }
666 bool do_aliasing() const { return _do_aliasing; }
667 bool print_assembly() const { return _print_assembly; }
668 void set_print_assembly(bool z) { _print_assembly = z; }
669 bool print_inlining() const { return _print_inlining; }
670 void set_print_inlining(bool z) { _print_inlining = z; }
671 bool print_intrinsics() const { return _print_intrinsics; }
672 void set_print_intrinsics(bool z) { _print_intrinsics = z; }
673 RTMState rtm_state() const { return _rtm_state; }
674 void set_rtm_state(RTMState s) { _rtm_state = s; }
675 bool use_rtm() const { return (_rtm_state & NoRTM) == 0; }
676 bool profile_rtm() const { return _rtm_state == ProfileRTM; }
677 uint max_node_limit() const { return (uint)_max_node_limit; }
678 void set_max_node_limit(uint n) { _max_node_limit = n; }
679 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; }
680 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; }
681 void set_flat_accesses() { _has_flat_accesses = true; }
682 bool flat_accesses_share_alias() const { return _flat_accesses_share_alias; }
683 void set_flat_accesses_share_alias(bool z) { _flat_accesses_share_alias = z; }
684 bool scalarize_in_safepoints() const { return _scalarize_in_safepoints; }
685 void set_scalarize_in_safepoints(bool z) { _scalarize_in_safepoints = z; }
686
687 // Support for scalarized inline type calling convention
688 bool has_scalarized_args() const { return _method != nullptr && _method->has_scalarized_args(); }
689 bool needs_stack_repair() const { return _method != nullptr && _method->get_Method()->c2_needs_stack_repair(); }
690
691 bool has_monitors() const { return _has_monitors; }
692 void set_has_monitors(bool v) { _has_monitors = v; }
693
694 // check the CompilerOracle for special behaviours for this compile
695 bool method_has_option(enum CompileCommand option) {
696 return method() != nullptr && method()->has_option(option);
697 }
698
699 #ifndef PRODUCT
700 uint next_igv_idx() { return _igv_idx++; }
701 bool trace_opto_output() const { return _trace_opto_output; }
702 void print_ideal_ir(const char* phase_name);
703 bool should_print_ideal() const { return _directive->PrintIdealOption; }
704 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
705 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
706 int _in_dump_cnt; // Required for dumping ir nodes.
707 #endif
708 bool has_irreducible_loop() const { return _has_irreducible_loop; }
709 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
710
711 // JSR 292
712 bool has_method_handle_invokes() const { return _has_method_handle_invokes; }
713 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; }
714
715 Ticks _latest_stage_start_counter;
716
717 void begin_method();
718 void end_method();
719 bool should_print_igv(int level);
720 bool should_print_phase(CompilerPhaseType cpt);
721
722 void print_method(CompilerPhaseType cpt, int level, Node* n = nullptr);
723
724 #ifndef PRODUCT
725 void dump_igv(const char* graph_name, int level = 3) {
726 if (should_print_igv(level)) {
727 _igv_printer->print_method(graph_name, level);
728 }
729 }
730
731 void igv_print_method_to_file(const char* phase_name = "Debug", bool append = false);
732 void igv_print_method_to_network(const char* phase_name = "Debug");
733 static IdealGraphPrinter* debug_file_printer() { return _debug_file_printer; }
734 static IdealGraphPrinter* debug_network_printer() { return _debug_network_printer; }
735 #endif
736
737 int macro_count() const { return _macro_nodes.length(); }
738 int parse_predicate_count() const { return _parse_predicates.length(); }
739 int template_assertion_predicate_count() const { return _template_assertion_predicate_opaqs.length(); }
740 int expensive_count() const { return _expensive_nodes.length(); }
741 int coarsened_count() const { return _coarsened_locks.length(); }
742
743 Node* macro_node(int idx) const { return _macro_nodes.at(idx); }
744 ParsePredicateNode* parse_predicate(int idx) const { return _parse_predicates.at(idx); }
745
746 Node* template_assertion_predicate_opaq_node(int idx) const {
747 return _template_assertion_predicate_opaqs.at(idx);
748 }
749
750 Node* expensive_node(int idx) const { return _expensive_nodes.at(idx); }
751
752 ConnectionGraph* congraph() { return _congraph;}
753 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;}
754 void add_macro_node(Node * n) {
755 //assert(n->is_macro(), "must be a macro node");
756 assert(!_macro_nodes.contains(n), "duplicate entry in expand list");
757 _macro_nodes.append(n);
758 }
759 void remove_macro_node(Node* n) {
760 // this function may be called twice for a node so we can only remove it
761 // if it's still existing.
762 _macro_nodes.remove_if_existing(n);
763 // Remove from coarsened locks list if present
764 if (coarsened_count() > 0) {
765 remove_coarsened_lock(n);
766 }
767 }
768 void add_expensive_node(Node* n);
769 void remove_expensive_node(Node* n) {
770 _expensive_nodes.remove_if_existing(n);
771 }
772
773 void add_parse_predicate(ParsePredicateNode* n) {
774 assert(!_parse_predicates.contains(n), "duplicate entry in Parse Predicate list");
775 _parse_predicates.append(n);
776 }
777
778 void remove_parse_predicate(ParsePredicateNode* n) {
779 if (parse_predicate_count() > 0) {
780 _parse_predicates.remove_if_existing(n);
781 }
782 }
783
784 void add_template_assertion_predicate_opaq(Node* n) {
785 assert(!_template_assertion_predicate_opaqs.contains(n),
786 "duplicate entry in template assertion predicate opaque4 list");
787 _template_assertion_predicate_opaqs.append(n);
788 }
789
790 void remove_template_assertion_predicate_opaq(Node* n) {
791 if (template_assertion_predicate_count() > 0) {
792 _template_assertion_predicate_opaqs.remove_if_existing(n);
793 }
794 }
795 void add_coarsened_locks(GrowableArray<AbstractLockNode*>& locks);
796 void remove_coarsened_lock(Node* n);
797 bool coarsened_locks_consistent();
798
799 bool post_loop_opts_phase() { return _post_loop_opts_phase; }
800 void set_post_loop_opts_phase() { _post_loop_opts_phase = true; }
801 void reset_post_loop_opts_phase() { _post_loop_opts_phase = false; }
802
803 void record_for_post_loop_opts_igvn(Node* n);
804 void remove_from_post_loop_opts_igvn(Node* n);
805 void process_for_post_loop_opts_igvn(PhaseIterGVN& igvn);
806
807 // Keep track of inline type nodes for later processing
808 void add_inline_type(Node* n);
809 void remove_inline_type(Node* n);
810 void process_inline_types(PhaseIterGVN &igvn, bool remove = false);
811
812 void adjust_flat_array_access_aliases(PhaseIterGVN& igvn);
813
814 void record_unstable_if_trap(UnstableIfTrap* trap);
815 bool remove_unstable_if_trap(CallStaticJavaNode* unc, bool yield);
816 void remove_useless_unstable_if_traps(Unique_Node_List &useful);
817 void process_for_unstable_if_traps(PhaseIterGVN& igvn);
818
819 void sort_macro_nodes();
820
821 void mark_parse_predicate_nodes_useless(PhaseIterGVN& igvn);
822
823 // Are there candidate expensive nodes for optimization?
824 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
825 // Check whether n1 and n2 are similar
826 static int cmp_expensive_nodes(Node* n1, Node* n2);
827 // Sort expensive nodes to locate similar expensive nodes
828 void sort_expensive_nodes();
829
830 // Compilation environment.
831 Arena* comp_arena() { return &_comp_arena; }
832 ciEnv* env() const { return _env; }
833 CompileLog* log() const { return _log; }
834 bool failing() const { return _env->failing() || _failure_reason != nullptr; }
835 const char* failure_reason() const { return (_env->failing()) ? _env->failure_reason() : _failure_reason; }
836
837 bool failure_reason_is(const char* r) const {
838 return (r == _failure_reason) || (r != nullptr && _failure_reason != nullptr && strcmp(r, _failure_reason) == 0);
839 }
840
841 void record_failure(const char* reason);
842 void record_method_not_compilable(const char* reason) {
843 env()->record_method_not_compilable(reason);
844 // Record failure reason.
845 record_failure(reason);
846 }
847 bool check_node_count(uint margin, const char* reason) {
848 if (oom()) {
849 record_method_not_compilable_oom();
850 return true;
851 }
852 if (live_nodes() + margin > max_node_limit()) {
853 record_method_not_compilable(reason);
854 return true;
855 } else {
856 return false;
857 }
858 }
859 bool oom() const { return _oom; }
860 void set_oom() { _oom = true; }
861
862 // Node management
863 uint unique() const { return _unique; }
864 uint next_unique() { return _unique++; }
865 void set_unique(uint i) { _unique = i; }
866 Arena* node_arena() { return _node_arena; }
867 Arena* old_arena() { return (&_node_arena_one == _node_arena) ? &_node_arena_two : &_node_arena_one; }
868 RootNode* root() const { return _root; }
869 void set_root(RootNode* r) { _root = r; }
870 StartNode* start() const; // (Derived from root.)
871 void init_start(StartNode* s);
872 Node* immutable_memory();
873
874 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; }
875 Node* recent_alloc_obj() const { return _recent_alloc_obj; }
876 void set_recent_alloc(Node* ctl, Node* obj) {
877 _recent_alloc_ctl = ctl;
878 _recent_alloc_obj = obj;
879 }
880 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return;
881 _dead_node_count++;
882 }
883 void reset_dead_node_list() { _dead_node_list.reset();
884 _dead_node_count = 0;
885 }
886 uint live_nodes() const {
887 int val = _unique - _dead_node_count;
888 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
889 return (uint) val;
890 }
891 #ifdef ASSERT
892 void set_phase_optimize_finished() { _phase_optimize_finished = true; }
893 bool phase_optimize_finished() const { return _phase_optimize_finished; }
894 uint count_live_nodes_by_graph_walk();
895 void print_missing_nodes();
896 #endif
897
898 // Record modified nodes to check that they are put on IGVN worklist
899 void record_modified_node(Node* n) NOT_DEBUG_RETURN;
900 void remove_modified_node(Node* n) NOT_DEBUG_RETURN;
901 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } )
902
903 MachConstantBaseNode* mach_constant_base_node();
904 bool has_mach_constant_base_node() const { return _mach_constant_base_node != nullptr; }
905 // Generated by adlc, true if CallNode requires MachConstantBase.
906 bool needs_deep_clone_jvms();
907
908 // Handy undefined Node
909 Node* top() const { return _top; }
910
911 // these are used by guys who need to know about creation and transformation of top:
912 Node* cached_top_node() { return _top; }
913 void set_cached_top_node(Node* tn);
914
915 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
916 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
917 Node_Notes* default_node_notes() const { return _default_node_notes; }
918 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
919
920 Node_Notes* node_notes_at(int idx) {
921 return locate_node_notes(_node_note_array, idx, false);
922 }
923 inline bool set_node_notes_at(int idx, Node_Notes* value);
924
925 // Copy notes from source to dest, if they exist.
926 // Overwrite dest only if source provides something.
927 // Return true if information was moved.
928 bool copy_node_notes_to(Node* dest, Node* source);
929
930 // Workhorse function to sort out the blocked Node_Notes array:
931 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
932 int idx, bool can_grow = false);
933
934 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
935
936 // Type management
937 Arena* type_arena() { return _type_arena; }
938 Dict* type_dict() { return _type_dict; }
939 size_t type_last_size() { return _type_last_size; }
940 int num_alias_types() { return _num_alias_types; }
941
942 void init_type_arena() { _type_arena = &_Compile_types; }
943 void set_type_arena(Arena* a) { _type_arena = a; }
944 void set_type_dict(Dict* d) { _type_dict = d; }
945 void set_type_last_size(size_t sz) { _type_last_size = sz; }
946
947 const TypeFunc* last_tf(ciMethod* m) {
948 return (m == _last_tf_m) ? _last_tf : nullptr;
949 }
950 void set_last_tf(ciMethod* m, const TypeFunc* tf) {
951 assert(m != nullptr || tf == nullptr, "");
952 _last_tf_m = m;
953 _last_tf = tf;
954 }
955
956 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
957 AliasType* alias_type(const TypePtr* adr_type, ciField* field = nullptr, bool uncached = false) { return find_alias_type(adr_type, false, field, uncached); }
958 bool have_alias_type(const TypePtr* adr_type);
959 AliasType* alias_type(ciField* field);
960
961 int get_alias_index(const TypePtr* at, bool uncached = false) { return alias_type(at, nullptr, uncached)->index(); }
962 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); }
963 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); }
964
965 // Building nodes
966 void rethrow_exceptions(JVMState* jvms);
967 void return_values(JVMState* jvms);
968 JVMState* build_start_state(StartNode* start, const TypeFunc* tf);
969
970 // Decide how to build a call.
971 // The profile factor is a discount to apply to this site's interp. profile.
972 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
973 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = nullptr,
974 bool allow_intrinsics = true);
975 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
976 return should_delay_string_inlining(call_method, jvms) ||
977 should_delay_boxing_inlining(call_method, jvms) ||
978 should_delay_vector_inlining(call_method, jvms);
979 }
980 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
981 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
982 bool should_delay_vector_inlining(ciMethod* call_method, JVMState* jvms);
983 bool should_delay_vector_reboxing_inlining(ciMethod* call_method, JVMState* jvms);
984
985 // Helper functions to identify inlining potential at call-site
986 ciMethod* optimize_virtual_call(ciMethod* caller, ciInstanceKlass* klass,
987 ciKlass* holder, ciMethod* callee,
988 const TypeOopPtr* receiver_type, bool is_virtual,
989 bool &call_does_dispatch, int &vtable_index,
990 bool check_access = true);
991 ciMethod* optimize_inlining(ciMethod* caller, ciInstanceKlass* klass, ciKlass* holder,
992 ciMethod* callee, const TypeOopPtr* receiver_type,
993 bool check_access = true);
994
995 // Report if there were too many traps at a current method and bci.
996 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
997 // If there is no MDO at all, report no trap unless told to assume it.
998 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
999 // This version, unspecific to a particular bci, asks if
1000 // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
1001 bool too_many_traps(Deoptimization::DeoptReason reason,
1002 // Privately used parameter for logging:
1003 ciMethodData* logmd = nullptr);
1004 // Report if there were too many recompiles at a method and bci.
1005 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1006 // Report if there were too many traps or recompiles at a method and bci.
1007 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
1008 return too_many_traps(method, bci, reason) ||
1009 too_many_recompiles(method, bci, reason);
1010 }
1011 // Return a bitset with the reasons where deoptimization is allowed,
1012 // i.e., where there were not too many uncommon traps.
1013 int _allowed_reasons;
1014 int allowed_deopt_reasons() { return _allowed_reasons; }
1015 void set_allowed_deopt_reasons();
1016
1017 // Parsing, optimization
1018 PhaseGVN* initial_gvn() { return _initial_gvn; }
1019 Unique_Node_List* igvn_worklist() {
1020 assert(_igvn_worklist != nullptr, "must be created in Compile::Compile");
1021 return _igvn_worklist;
1022 }
1023 Type_Array* types() {
1024 assert(_types != nullptr, "must be created in Compile::Compile");
1025 return _types;
1026 }
1027 NodeHash* node_hash() {
1028 assert(_node_hash != nullptr, "must be created in Compile::Compile");
1029 return _node_hash;
1030 }
1031 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1032 inline void remove_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1033 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; }
1034
1035 // Replace n by nn using initial_gvn, calling hash_delete and
1036 // record_for_igvn as needed.
1037 void gvn_replace_by(Node* n, Node* nn);
1038
1039
1040 void identify_useful_nodes(Unique_Node_List &useful);
1041 void update_dead_node_list(Unique_Node_List &useful);
1042 void disconnect_useless_nodes(Unique_Node_List& useful, Unique_Node_List& worklist);
1043
1044 void remove_useless_node(Node* dead);
1045
1046 // Record this CallGenerator for inlining at the end of parsing.
1047 void add_late_inline(CallGenerator* cg) {
1048 _late_inlines.insert_before(_late_inlines_pos, cg);
1049 _late_inlines_pos++;
1050 }
1051
1052 void prepend_late_inline(CallGenerator* cg) {
1053 _late_inlines.insert_before(0, cg);
1054 }
1055
1056 void add_string_late_inline(CallGenerator* cg) {
1057 _string_late_inlines.push(cg);
1058 }
1059
1060 void add_boxing_late_inline(CallGenerator* cg) {
1061 _boxing_late_inlines.push(cg);
1062 }
1063
1064 void add_vector_reboxing_late_inline(CallGenerator* cg) {
1065 _vector_reboxing_late_inlines.push(cg);
1066 }
1067
1068 template<typename N, ENABLE_IF(std::is_base_of<Node, N>::value)>
1069 void remove_useless_nodes(GrowableArray<N*>& node_list, Unique_Node_List& useful);
1070
1071 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1072 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Node* dead);
1073
1074 void remove_useless_coarsened_locks(Unique_Node_List& useful);
1075
1076 void process_print_inlining();
1077 void dump_print_inlining();
1078
1079 bool over_inlining_cutoff() const {
1080 if (!inlining_incrementally()) {
1081 return unique() > (uint)NodeCountInliningCutoff;
1082 } else {
1083 // Give some room for incremental inlining algorithm to "breathe"
1084 // and avoid thrashing when live node count is close to the limit.
1085 // Keep in mind that live_nodes() isn't accurate during inlining until
1086 // dead node elimination step happens (see Compile::inline_incrementally).
1087 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10;
1088 }
1089 }
1090
1091 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; }
1092 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--; }
1093 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; }
1094
1095 bool inline_incrementally_one();
1096 void inline_incrementally_cleanup(PhaseIterGVN& igvn);
1097 void inline_incrementally(PhaseIterGVN& igvn);
1098 bool should_delay_inlining() { return AlwaysIncrementalInline || (StressIncrementalInlining && (random() % 2) == 0); }
1099 void inline_string_calls(bool parse_time);
1100 void inline_boxing_calls(PhaseIterGVN& igvn);
1101 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1102 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn);
1103
1104 void inline_vector_reboxing_calls();
1105 bool has_vbox_nodes();
1106
1107 void process_late_inline_calls_no_inline(PhaseIterGVN& igvn);
1108
1109 // Matching, CFG layout, allocation, code generation
1110 PhaseCFG* cfg() { return _cfg; }
1111 bool has_java_calls() const { return _java_calls > 0; }
1112 int java_calls() const { return _java_calls; }
1113 int inner_loops() const { return _inner_loops; }
1114 Matcher* matcher() { return _matcher; }
1115 PhaseRegAlloc* regalloc() { return _regalloc; }
1116 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; }
1117 Arena* indexSet_arena() { return _indexSet_arena; }
1118 void* indexSet_free_block_list() { return _indexSet_free_block_list; }
1119 DebugInformationRecorder* debug_info() { return env()->debug_info(); }
1120
1121 void update_interpreter_frame_size(int size) {
1122 if (_interpreter_frame_size < size) {
1123 _interpreter_frame_size = size;
1124 }
1125 }
1126
1127 void set_matcher(Matcher* m) { _matcher = m; }
1128 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; }
1129 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; }
1130 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; }
1131
1132 void set_java_calls(int z) { _java_calls = z; }
1133 void set_inner_loops(int z) { _inner_loops = z; }
1134
1135 Dependencies* dependencies() { return env()->dependencies(); }
1136
1137 // Major entry point. Given a Scope, compile the associated method.
1138 // For normal compilations, entry_bci is InvocationEntryBci. For on stack
1139 // replacement, entry_bci indicates the bytecode for which to compile a
1140 // continuation.
1141 Compile(ciEnv* ci_env, ciMethod* target,
1142 int entry_bci, Options options, DirectiveSet* directive);
1143
1144 // Second major entry point. From the TypeFunc signature, generate code
1145 // to pass arguments from the Java calling convention to the C calling
1146 // convention.
1147 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1148 address stub_function, const char *stub_name,
1149 int is_fancy_jump, bool pass_tls,
1150 bool return_pc, DirectiveSet* directive);
1151
1152 ~Compile() {
1153 delete _print_inlining_stream;
1154 };
1155
1156 // Are we compiling a method?
1157 bool has_method() { return method() != nullptr; }
1158
1159 // Maybe print some information about this compile.
1160 void print_compile_messages();
1161
1162 // Final graph reshaping, a post-pass after the regular optimizer is done.
1163 bool final_graph_reshaping();
1164
1165 // returns true if adr is completely contained in the given alias category
1166 bool must_alias(const TypePtr* adr, int alias_idx);
1167
1168 // returns true if adr overlaps with the given alias category
1169 bool can_alias(const TypePtr* adr, int alias_idx);
1170
1171 // Stack slots that may be unused by the calling convention but must
1172 // otherwise be preserved. On Intel this includes the return address.
1173 // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1174 uint in_preserve_stack_slots() {
1175 return SharedRuntime::in_preserve_stack_slots();
1176 }
1177
1178 // "Top of Stack" slots that may be unused by the calling convention but must
1179 // otherwise be preserved.
1180 // On Intel these are not necessary and the value can be zero.
1181 static uint out_preserve_stack_slots() {
1182 return SharedRuntime::out_preserve_stack_slots();
1183 }
1184
1185 // Number of outgoing stack slots killed above the out_preserve_stack_slots
1186 // for calls to C. Supports the var-args backing area for register parms.
1187 uint varargs_C_out_slots_killed() const;
1188
1189 // Number of Stack Slots consumed by a synchronization entry
1190 int sync_stack_slots() const;
1191
1192 // Compute the name of old_SP. See <arch>.ad for frame layout.
1193 OptoReg::Name compute_old_SP();
1194
1195 private:
1196 // Phase control:
1197 void Init(bool aliasing); // Prepare for a single compilation
1198 void Optimize(); // Given a graph, optimize it
1199 void Code_Gen(); // Generate code from a graph
1200
1201 // Management of the AliasType table.
1202 void grow_alias_types();
1203 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1204 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1205 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field, bool uncached = false);
1206
1207 void verify_top(Node*) const PRODUCT_RETURN;
1208
1209 // Intrinsic setup.
1210 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor
1211 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper
1212 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn
1213 void register_intrinsic(CallGenerator* cg); // update fn
1214
1215 #ifndef PRODUCT
1216 static juint _intrinsic_hist_count[];
1217 static jubyte _intrinsic_hist_flags[];
1218 #endif
1219 // Function calls made by the public function final_graph_reshaping.
1220 // No need to be made public as they are not called elsewhere.
1221 void final_graph_reshaping_impl(Node *n, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1222 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop, Unique_Node_List& dead_nodes);
1223 void final_graph_reshaping_walk(Node_Stack& nstack, Node* root, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1224 void eliminate_redundant_card_marks(Node* n);
1225
1226 // Logic cone optimization.
1227 void optimize_logic_cones(PhaseIterGVN &igvn);
1228 void collect_logic_cone_roots(Unique_Node_List& list);
1229 void process_logic_cone_root(PhaseIterGVN &igvn, Node* n, VectorSet& visited);
1230 bool compute_logic_cone(Node* n, Unique_Node_List& partition, Unique_Node_List& inputs);
1231 uint compute_truth_table(Unique_Node_List& partition, Unique_Node_List& inputs);
1232 uint eval_macro_logic_op(uint func, uint op1, uint op2, uint op3);
1233 Node* xform_to_MacroLogicV(PhaseIterGVN &igvn, const TypeVect* vt, Unique_Node_List& partitions, Unique_Node_List& inputs);
1234 void check_no_dead_use() const NOT_DEBUG_RETURN;
1235
1236 public:
1237
1238 // Note: Histogram array size is about 1 Kb.
1239 enum { // flag bits:
1240 _intrinsic_worked = 1, // succeeded at least once
1241 _intrinsic_failed = 2, // tried it but it failed
1242 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1243 _intrinsic_virtual = 8, // was seen in the virtual form (rare)
1244 _intrinsic_both = 16 // was seen in the non-virtual form (usual)
1245 };
1246 // Update histogram. Return boolean if this is a first-time occurrence.
1247 static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1248 bool is_virtual, int flags) PRODUCT_RETURN0;
1249 static void print_intrinsic_statistics() PRODUCT_RETURN;
1250
1251 // Graph verification code
1252 // Walk the node list, verifying that there is a one-to-one
1253 // correspondence between Use-Def edges and Def-Use edges
1254 // The option no_dead_code enables stronger checks that the
1255 // graph is strongly connected from root in both directions.
1256 void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN;
1257
1258 // Verify bi-directional correspondence of edges
1259 void verify_bidirectional_edges(Unique_Node_List &visited);
1260
1261 // End-of-run dumps.
1262 static void print_statistics() PRODUCT_RETURN;
1263
1264 // Verify ADLC assumptions during startup
1265 static void adlc_verification() PRODUCT_RETURN;
1266
1267 // Definitions of pd methods
1268 static void pd_compiler2_init();
1269
1270 // Static parse-time type checking logic for gen_subtype_check:
1271 enum SubTypeCheckResult { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1272 SubTypeCheckResult static_subtype_check(const TypeKlassPtr* superk, const TypeKlassPtr* subk, bool skip = StressReflectiveCode);
1273
1274 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1275 // Optional control dependency (for example, on range check)
1276 Node* ctrl = nullptr);
1277
1278 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1279 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl, bool carry_dependency = false);
1280
1281 Node* optimize_acmp(PhaseGVN* phase, Node* a, Node* b);
1282
1283 // Auxiliary method for randomized fuzzing/stressing
1284 int random();
1285 bool randomized_select(int count);
1286
1287 // supporting clone_map
1288 CloneMap& clone_map();
1289 void set_clone_map(Dict* d);
1290
1291 bool needs_clinit_barrier(ciField* ik, ciMethod* accessing_method);
1292 bool needs_clinit_barrier(ciMethod* ik, ciMethod* accessing_method);
1293 bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method);
1294
1295 #ifdef IA32
1296 private:
1297 bool _select_24_bit_instr; // We selected an instruction with a 24-bit result
1298 bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results
1299
1300 // Remember if this compilation changes hardware mode to 24-bit precision.
1301 void set_24_bit_selection_and_mode(bool selection, bool mode) {
1302 _select_24_bit_instr = selection;
1303 _in_24_bit_fp_mode = mode;
1304 }
1305
1306 public:
1307 bool select_24_bit_instr() const { return _select_24_bit_instr; }
1308 bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; }
1309 #endif // IA32
1310 #ifdef ASSERT
1311 VerifyMeetResult* _type_verify;
1312 void set_exception_backedge() { _exception_backedge = true; }
1313 bool has_exception_backedge() const { return _exception_backedge; }
1314 #endif
1315
1316 static bool push_thru_add(PhaseGVN* phase, Node* z, const TypeInteger* tz, const TypeInteger*& rx, const TypeInteger*& ry,
1317 BasicType out_bt, BasicType in_bt);
1318
1319 static Node* narrow_value(BasicType bt, Node* value, const Type* type, PhaseGVN* phase, bool transform_res);
1320 };
1321
1322 #endif // SHARE_OPTO_COMPILE_HPP