1 /*
2 * Copyright (c) 1997, 2025, 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/compileBroker.hpp"
33 #include "compiler/compilerEvent.hpp"
34 #include "compiler/cHeapStringHolder.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 #include "utilities/vmEnums.hpp"
49 #include "opto/printinlining.hpp"
50
51 class AbstractLockNode;
52 class AddPNode;
53 class Block;
54 class Bundle;
55 class CallGenerator;
56 class CallStaticJavaNode;
57 class CloneMap;
58 class CompilationFailureInfo;
59 class ConnectionGraph;
60 class IdealGraphPrinter;
61 class InlineTree;
62 class Matcher;
63 class MachConstantNode;
64 class MachConstantBaseNode;
65 class MachNode;
66 class MachOper;
67 class MachSafePointNode;
68 class Node;
69 class Node_Array;
70 class Node_List;
71 class Node_Notes;
72 class NodeHash;
73 class NodeCloneInfo;
74 class OpaqueTemplateAssertionPredicateNode;
75 class OptoReg;
76 class ParsePredicateNode;
77 class PhaseCFG;
78 class PhaseGVN;
79 class PhaseIterGVN;
80 class PhaseRegAlloc;
81 class PhaseCCP;
82 class PhaseOutput;
83 class RootNode;
84 class relocInfo;
85 class StartNode;
86 class SafePointNode;
87 class JVMState;
88 class Type;
89 class TypeInt;
90 class TypeInteger;
91 class TypeKlassPtr;
92 class TypePtr;
93 class TypeOopPtr;
94 class TypeFunc;
95 class TypeVect;
96 class Type_Array;
97 class Unique_Node_List;
98 class UnstableIfTrap;
99 class nmethod;
100 class Node_Stack;
101 struct Final_Reshape_Counts;
102 class VerifyMeetResult;
103
104 enum LoopOptsMode {
105 LoopOptsDefault,
106 LoopOptsNone,
107 LoopOptsMaxUnroll,
108 LoopOptsShenandoahExpand,
109 LoopOptsShenandoahPostExpand,
110 LoopOptsSkipSplitIf,
111 LoopOptsVerify
112 };
113
114 // The type of all node counts and indexes.
115 // It must hold at least 16 bits, but must also be fast to load and store.
116 // This type, if less than 32 bits, could limit the number of possible nodes.
117 // (To make this type platform-specific, move to globalDefinitions_xxx.hpp.)
118 typedef unsigned int node_idx_t;
119
120 class NodeCloneInfo {
121 private:
122 uint64_t _idx_clone_orig;
123 public:
124
125 void set_idx(node_idx_t idx) {
126 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx;
127 }
128 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); }
129
130 void set_gen(int generation) {
131 uint64_t g = (uint64_t)generation << 32;
132 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g;
133 }
134 int gen() const { return (int)(_idx_clone_orig >> 32); }
135
136 void set(uint64_t x) { _idx_clone_orig = x; }
137 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); }
138 uint64_t get() const { return _idx_clone_orig; }
139
140 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {}
141 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); }
142
143 void dump_on(outputStream* st) const;
144 };
145
146 class CloneMap {
147 friend class Compile;
148 private:
149 bool _debug;
150 Dict* _dict;
151 int _clone_idx; // current cloning iteration/generation in loop unroll
152 public:
153 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy
154 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; }
155 Dict* dict() const { return _dict; }
156 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == nullptr, "key existed"); _dict->Insert(_2p(key), (void*)val); }
157 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); }
158 void remove(node_idx_t key) { _dict->Delete(_2p(key)); }
159 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); }
160 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); }
161 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); }
162 int gen(const void* k) const { return gen(_2_node_idx_t(k)); }
163 int max_gen() const;
164 void clone(Node* old, Node* nnn, int gen);
165 void verify_insert_and_clone(Node* old, Node* nnn, int gen);
166 void dump(node_idx_t key, outputStream* st) const;
167
168 int clone_idx() const { return _clone_idx; }
169 void set_clone_idx(int x) { _clone_idx = x; }
170 bool is_debug() const { return _debug; }
171 void set_debug(bool debug) { _debug = debug; }
172
173 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); }
174 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); }
175 };
176
177 class Options {
178 friend class Compile;
179 private:
180 const bool _subsume_loads; // Load can be matched as part of a larger op.
181 const bool _do_escape_analysis; // Do escape analysis.
182 const bool _do_iterative_escape_analysis; // Do iterative escape analysis.
183 const bool _do_reduce_allocation_merges; // Do try to reduce allocation merges.
184 const bool _eliminate_boxing; // Do boxing elimination.
185 const bool _do_locks_coarsening; // Do locks coarsening
186 const bool _for_preload; // Generate code for preload (before Java method execution), do class init barriers
187 const bool _do_superword; // Do SuperWord
188 const bool _install_code; // Install the code that was compiled
189 public:
190 Options(bool subsume_loads,
191 bool do_escape_analysis,
192 bool do_iterative_escape_analysis,
193 bool do_reduce_allocation_merges,
194 bool eliminate_boxing,
195 bool do_locks_coarsening,
196 bool do_superword,
197 bool for_preload,
198 bool install_code) :
199 _subsume_loads(subsume_loads),
200 _do_escape_analysis(do_escape_analysis),
201 _do_iterative_escape_analysis(do_iterative_escape_analysis),
202 _do_reduce_allocation_merges(do_reduce_allocation_merges),
203 _eliminate_boxing(eliminate_boxing),
204 _do_locks_coarsening(do_locks_coarsening),
205 _for_preload(for_preload),
206 _do_superword(do_superword),
207 _install_code(install_code) {
208 }
209
210 static Options for_runtime_stub() {
211 return Options(
212 /* subsume_loads = */ true,
213 /* do_escape_analysis = */ false,
214 /* do_iterative_escape_analysis = */ false,
215 /* do_reduce_allocation_merges = */ false,
216 /* eliminate_boxing = */ false,
217 /* do_lock_coarsening = */ false,
218 /* for_preload = */ false,
219 /* do_superword = */ true,
220 /* install_code = */ true
221 );
222 }
223 };
224
225 //------------------------------Compile----------------------------------------
226 // This class defines a top-level Compiler invocation.
227
228 class Compile : public Phase {
229
230 public:
231 // Fixed alias indexes. (See also MergeMemNode.)
232 enum {
233 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value)
234 AliasIdxBot = 2, // pseudo-index, aliases to everything
235 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM
236 };
237
238 // Variant of TraceTime(nullptr, &_t_accumulator, CITime);
239 // Integrated with logging. If logging is turned on, and CITimeVerbose is true,
240 // then brackets are put into the log, with time stamps and node counts.
241 // (The time collection itself is always conditionalized on CITime.)
242 class TracePhase : public TraceTime {
243 private:
244 Compile* const _compile;
245 CompileLog* _log;
246 const bool _dolog;
247 public:
248 TracePhase(PhaseTraceId phaseTraceId);
249 TracePhase(const char* name, PhaseTraceId phaseTraceId);
250 ~TracePhase();
251 const char* phase_name() const { return title(); }
252 };
253
254 // Information per category of alias (memory slice)
255 class AliasType {
256 private:
257 friend class Compile;
258
259 int _index; // unique index, used with MergeMemNode
260 const TypePtr* _adr_type; // normalized address type
261 ciField* _field; // relevant instance field, or null if none
262 const Type* _element; // relevant array element type, or null if none
263 bool _is_rewritable; // false if the memory is write-once only
264 int _general_index; // if this is type is an instance, the general
265 // type that this is an instance of
266
267 void Init(int i, const TypePtr* at);
268
269 public:
270 int index() const { return _index; }
271 const TypePtr* adr_type() const { return _adr_type; }
272 ciField* field() const { return _field; }
273 const Type* element() const { return _element; }
274 bool is_rewritable() const { return _is_rewritable; }
275 bool is_volatile() const { return (_field ? _field->is_volatile() : false); }
276 int general_index() const { return (_general_index != 0) ? _general_index : _index; }
277
278 void set_rewritable(bool z) { _is_rewritable = z; }
279 void set_field(ciField* f) {
280 assert(!_field,"");
281 _field = f;
282 if (f->is_final() || f->is_stable()) {
283 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
284 _is_rewritable = false;
285 }
286 }
287 void set_element(const Type* e) {
288 assert(_element == nullptr, "");
289 _element = e;
290 }
291
292 BasicType basic_type() const;
293
294 void print_on(outputStream* st) PRODUCT_RETURN;
295 };
296
297 enum {
298 logAliasCacheSize = 6,
299 AliasCacheSize = (1<<logAliasCacheSize)
300 };
301 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type
302 enum {
303 trapHistLength = MethodData::_trap_hist_limit
304 };
305
306 private:
307 // Fixed parameters to this compilation.
308 const int _compile_id;
309 const Options _options; // Compilation options
310 ciMethod* _method; // The method being compiled.
311 int _entry_bci; // entry bci for osr methods.
312 const TypeFunc* _tf; // My kind of signature
313 InlineTree* _ilt; // Ditto (temporary).
314 address _stub_function; // VM entry for stub being compiled, or null
315 const char* _stub_name; // Name of stub or adapter being compiled, or null
316 address _stub_entry_point; // Compile code entry for generated stub, or null
317
318 // Control of this compilation.
319 int _max_inline_size; // Max inline size for this compilation
320 int _freq_inline_size; // Max hot method inline size for this compilation
321 int _fixed_slots; // count of frame slots not allocated by the register
322 // allocator i.e. locks, original deopt pc, etc.
323 uintx _max_node_limit; // Max unique node count during a single compilation.
324
325 bool _post_loop_opts_phase; // Loop opts are finished.
326 bool _merge_stores_phase; // Phase for merging stores, after post loop opts phase.
327 bool _allow_macro_nodes; // True if we allow creation of macro nodes.
328
329 int _major_progress; // Count of something big happening
330 bool _inlining_progress; // progress doing incremental inlining?
331 bool _inlining_incrementally;// Are we doing incremental inlining (post parse)
332 bool _do_cleanup; // Cleanup is needed before proceeding with incremental inlining
333 bool _has_loops; // True if the method _may_ have some loops
334 bool _has_split_ifs; // True if the method _may_ have some split-if
335 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores.
336 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated
337 bool _has_boxed_value; // True if a boxed object is allocated
338 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess
339 uint _max_vector_size; // Maximum size of generated vectors
340 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper
341 uint _trap_hist[trapHistLength]; // Cumulative traps
342 bool _trap_can_recompile; // Have we emitted a recompiling trap?
343 uint _decompile_count; // Cumulative decompilation counts.
344 bool _do_inlining; // True if we intend to do inlining
345 bool _do_scheduling; // True if we intend to do scheduling
346 bool _do_freq_based_layout; // True if we intend to do frequency based block layout
347 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations
348 bool _use_cmove; // True if CMove should be used without profitability analysis
349 bool _do_aliasing; // True if we intend to do aliasing
350 bool _print_assembly; // True if we should dump assembly code for this compilation
351 bool _print_inlining; // True if we should print inlining for this compilation
352 bool _print_intrinsics; // True if we should print intrinsics for this compilation
353 #ifndef PRODUCT
354 uint _igv_idx; // Counter for IGV node identifiers
355 uint _igv_phase_iter[PHASE_NUM_TYPES]; // Counters for IGV phase iterations
356 bool _trace_opto_output;
357 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
358 #endif
359 bool _has_irreducible_loop; // Found irreducible loops
360 // JSR 292
361 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes.
362 bool _has_monitors; // Metadata transfered to nmethod to enable Continuations lock-detection fastpath
363 bool _has_scoped_access; // For shared scope closure
364 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry
365 bool _has_clinit_barriers; // True if compiled code has clinit barriers
366 int _loop_opts_cnt; // loop opts round
367 uint _stress_seed; // Seed for stress testing
368
369 // Compilation environment.
370 Arena _comp_arena; // Arena with lifetime equivalent to Compile
371 void* _barrier_set_state; // Potential GC barrier state for Compile
372 ciEnv* _env; // CI interface
373 DirectiveSet* _directive; // Compiler directive
374 CompileLog* _log; // from CompilerThread
375 CHeapStringHolder _failure_reason; // for record_failure/failing pattern
376 CompilationFailureInfo* _first_failure_details; // Details for the first failure happening during compilation
377 GrowableArray<CallGenerator*> _intrinsics; // List of intrinsics.
378 GrowableArray<Node*> _macro_nodes; // List of nodes which need to be expanded before matching.
379 GrowableArray<ParsePredicateNode*> _parse_predicates; // List of Parse Predicates.
380 // List of OpaqueTemplateAssertionPredicateNode nodes for Template Assertion Predicates which can be seen as list
381 // of Template Assertion Predicates themselves.
382 GrowableArray<OpaqueTemplateAssertionPredicateNode*> _template_assertion_predicate_opaques;
383 GrowableArray<Node*> _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
384 GrowableArray<Node*> _for_post_loop_igvn; // List of nodes for IGVN after loop opts are over
385 GrowableArray<Node*> _for_merge_stores_igvn; // List of nodes for IGVN merge stores
386 GrowableArray<UnstableIfTrap*> _unstable_if_traps; // List of ifnodes after IGVN
387 GrowableArray<Node_List*> _coarsened_locks; // List of coarsened Lock and Unlock nodes
388 ConnectionGraph* _congraph;
389 #ifndef PRODUCT
390 IdealGraphPrinter* _igv_printer;
391 static IdealGraphPrinter* _debug_file_printer;
392 static IdealGraphPrinter* _debug_network_printer;
393 #endif
394
395
396 // Node management
397 uint _unique; // Counter for unique Node indices
398 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N).
399 // So use this to keep count and make the call O(1).
400 VectorSet _dead_node_list; // Set of dead nodes
401 DEBUG_ONLY(Unique_Node_List* _modified_nodes;) // List of nodes which inputs were modified
402 DEBUG_ONLY(bool _phase_optimize_finished;) // Used for live node verification while creating new nodes
403
404 DEBUG_ONLY(bool _phase_verify_ideal_loop;) // Are we in PhaseIdealLoop verification?
405
406 // Arenas for new-space and old-space nodes.
407 // Swapped between using _node_arena.
408 // The lifetime of the old-space nodes is during xform.
409 Arena _node_arena_one;
410 Arena _node_arena_two;
411 Arena* _node_arena;
412 public:
413 Arena* swap_old_and_new() {
414 Arena* filled_arena_ptr = _node_arena;
415 Arena* old_arena_ptr = old_arena();
416 old_arena_ptr->destruct_contents();
417 _node_arena = old_arena_ptr;
418 return filled_arena_ptr;
419 }
420 private:
421 RootNode* _root; // Unique root of compilation, or null after bail-out.
422 Node* _top; // Unique top node. (Reset by various phases.)
423
424 Node* _immutable_memory; // Initial memory state
425
426 Node* _recent_alloc_obj;
427 Node* _recent_alloc_ctl;
428
429 // Constant table
430 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton.
431
432
433 // Blocked array of debugging and profiling information,
434 // tracked per node.
435 enum { _log2_node_notes_block_size = 8,
436 _node_notes_block_size = (1<<_log2_node_notes_block_size)
437 };
438 GrowableArray<Node_Notes*>* _node_note_array;
439 Node_Notes* _default_node_notes; // default notes for new nodes
440
441 // After parsing and every bulk phase we hang onto the Root instruction.
442 // The RootNode instruction is where the whole program begins. It produces
443 // the initial Control and BOTTOM for everybody else.
444
445 // Type management
446 Arena _Compile_types; // Arena for all types
447 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared()
448 Dict* _type_dict; // Intern table
449 CloneMap _clone_map; // used for recording history of cloned nodes
450 size_t _type_last_size; // Last allocation size (see Type::operator new/delete)
451 ciMethod* _last_tf_m; // Cache for
452 const TypeFunc* _last_tf; // TypeFunc::make
453 AliasType** _alias_types; // List of alias types seen so far.
454 int _num_alias_types; // Logical length of _alias_types
455 int _max_alias_types; // Physical length of _alias_types
456 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
457
458 // Parsing, optimization
459 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN
460
461 // Shared worklist for all IGVN rounds. Nodes can be pushed to it at any time.
462 // If pushed outside IGVN, the Node is processed in the next IGVN round.
463 Unique_Node_List* _igvn_worklist;
464
465 // Shared type array for GVN, IGVN and CCP. It maps node idx -> Type*.
466 Type_Array* _types;
467
468 // Shared node hash table for GVN, IGVN and CCP.
469 NodeHash* _node_hash;
470
471 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after main parsing has finished.
472 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
473 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
474
475 GrowableArray<CallGenerator*> _vector_reboxing_late_inlines; // same but for vector reboxing operations
476
477 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
478 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending
479
480 // "MemLimit" directive was specified and the memory limit was hit during compilation
481 bool _oom;
482
483 // Only keep nodes in the expensive node list that need to be optimized
484 void cleanup_expensive_nodes(PhaseIterGVN &igvn);
485 // Use for sorting expensive nodes to bring similar nodes together
486 static int cmp_expensive_nodes(Node** n1, Node** n2);
487 // Expensive nodes list already sorted?
488 bool expensive_nodes_sorted() const;
489 // Remove the speculative part of types and clean up the graph
490 void remove_speculative_types(PhaseIterGVN &igvn);
491
492 void* _replay_inline_data; // Pointer to data loaded from file
493
494 void log_late_inline_failure(CallGenerator* cg, const char* msg);
495 DEBUG_ONLY(bool _exception_backedge;)
496
497 void record_method_not_compilable_oom();
498
499 InlinePrinter _inline_printer;
500
501 public:
502 void* barrier_set_state() const { return _barrier_set_state; }
503
504 InlinePrinter* inline_printer() { return &_inline_printer; }
505
506 #ifndef PRODUCT
507 IdealGraphPrinter* igv_printer() { return _igv_printer; }
508 void reset_igv_phase_iter(CompilerPhaseType cpt) { _igv_phase_iter[cpt] = 0; }
509 #endif
510
511 void log_late_inline(CallGenerator* cg);
512 void log_inline_id(CallGenerator* cg);
513 void log_inline_failure(const char* msg);
514
515 void* replay_inline_data() const { return _replay_inline_data; }
516
517 // Dump inlining replay data to the stream.
518 void dump_inline_data(outputStream* out);
519 void dump_inline_data_reduced(outputStream* out);
520
521 private:
522 // Matching, CFG layout, allocation, code generation
523 PhaseCFG* _cfg; // Results of CFG finding
524 int _java_calls; // Number of java calls in the method
525 int _inner_loops; // Number of inner loops in the method
526 Matcher* _matcher; // Engine to map ideal to machine instructions
527 PhaseRegAlloc* _regalloc; // Results of register allocation.
528 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout)
529 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin
530 void* _indexSet_free_block_list; // free list of IndexSet bit blocks
531 int _interpreter_frame_size;
532
533 PhaseOutput* _output;
534
535 public:
536 // Accessors
537
538 // The Compile instance currently active in this (compiler) thread.
539 static Compile* current() {
540 return (Compile*) ciEnv::current()->compiler_data();
541 }
542
543 int interpreter_frame_size() const { return _interpreter_frame_size; }
544
545 PhaseOutput* output() const { return _output; }
546 void set_output(PhaseOutput* o) { _output = o; }
547
548 // ID for this compilation. Useful for setting breakpoints in the debugger.
549 int compile_id() const { return _compile_id; }
550 DirectiveSet* directive() const { return _directive; }
551
552 // Does this compilation allow instructions to subsume loads? User
553 // instructions that subsume a load may result in an unschedulable
554 // instruction sequence.
555 bool subsume_loads() const { return _options._subsume_loads; }
556 /** Do escape analysis. */
557 bool do_escape_analysis() const { return _options._do_escape_analysis; }
558 bool do_iterative_escape_analysis() const { return _options._do_iterative_escape_analysis; }
559 bool do_reduce_allocation_merges() const { return _options._do_reduce_allocation_merges; }
560 /** Do boxing elimination. */
561 bool eliminate_boxing() const { return _options._eliminate_boxing; }
562 /** Do aggressive boxing elimination. */
563 bool aggressive_unboxing() const { return _options._eliminate_boxing && AggressiveUnboxing; }
564 bool should_install_code() const { return _options._install_code; }
565 /** Do locks coarsening. */
566 bool do_locks_coarsening() const { return _options._do_locks_coarsening; }
567 bool do_superword() const { return _options._do_superword; }
568
569 bool do_clinit_barriers() const { return _options._for_preload; }
570 bool for_preload() const { return _options._for_preload; }
571
572 // Other fixed compilation parameters.
573 ciMethod* method() const { return _method; }
574 int entry_bci() const { return _entry_bci; }
575 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; }
576 bool is_method_compilation() const { return (_method != nullptr && !_method->flags().is_native()); }
577 const TypeFunc* tf() const { assert(_tf!=nullptr, ""); return _tf; }
578 void init_tf(const TypeFunc* tf) { assert(_tf==nullptr, ""); _tf = tf; }
579 InlineTree* ilt() const { return _ilt; }
580 address stub_function() const { return _stub_function; }
581 const char* stub_name() const { return _stub_name; }
582 address stub_entry_point() const { return _stub_entry_point; }
583 void set_stub_entry_point(address z) { _stub_entry_point = z; }
584
585 // Control of this compilation.
586 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; }
587 void set_fixed_slots(int n) { _fixed_slots = n; }
588 int major_progress() const { return _major_progress; }
589 void set_inlining_progress(bool z) { _inlining_progress = z; }
590 int inlining_progress() const { return _inlining_progress; }
591 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
592 int inlining_incrementally() const { return _inlining_incrementally; }
593 void set_do_cleanup(bool z) { _do_cleanup = z; }
594 int do_cleanup() const { return _do_cleanup; }
595 void set_major_progress() { _major_progress++; }
596 void restore_major_progress(int progress) { _major_progress += progress; }
597 void clear_major_progress() { _major_progress = 0; }
598 int max_inline_size() const { return _max_inline_size; }
599 void set_freq_inline_size(int n) { _freq_inline_size = n; }
600 int freq_inline_size() const { return _freq_inline_size; }
601 void set_max_inline_size(int n) { _max_inline_size = n; }
602 bool has_loops() const { return _has_loops; }
603 void set_has_loops(bool z) { _has_loops = z; }
604 bool has_split_ifs() const { return _has_split_ifs; }
605 void set_has_split_ifs(bool z) { _has_split_ifs = z; }
606 bool has_unsafe_access() const { return _has_unsafe_access; }
607 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; }
608 bool has_stringbuilder() const { return _has_stringbuilder; }
609 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; }
610 bool has_boxed_value() const { return _has_boxed_value; }
611 void set_has_boxed_value(bool z) { _has_boxed_value = z; }
612 bool has_reserved_stack_access() const { return _has_reserved_stack_access; }
613 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
614 uint max_vector_size() const { return _max_vector_size; }
615 void set_max_vector_size(uint s) { _max_vector_size = s; }
616 bool clear_upper_avx() const { return _clear_upper_avx; }
617 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; }
618 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; }
619 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
620 bool trap_can_recompile() const { return _trap_can_recompile; }
621 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; }
622 uint decompile_count() const { return _decompile_count; }
623 void set_decompile_count(uint c) { _decompile_count = c; }
624 bool allow_range_check_smearing() const;
625 bool do_inlining() const { return _do_inlining; }
626 void set_do_inlining(bool z) { _do_inlining = z; }
627 bool do_scheduling() const { return _do_scheduling; }
628 void set_do_scheduling(bool z) { _do_scheduling = z; }
629 bool do_freq_based_layout() const{ return _do_freq_based_layout; }
630 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
631 bool do_vector_loop() const { return _do_vector_loop; }
632 void set_do_vector_loop(bool z) { _do_vector_loop = z; }
633 bool use_cmove() const { return _use_cmove; }
634 void set_use_cmove(bool z) { _use_cmove = z; }
635 bool do_aliasing() const { return _do_aliasing; }
636 bool print_assembly() const { return _print_assembly; }
637 void set_print_assembly(bool z) { _print_assembly = z; }
638 bool print_inlining() const { return _print_inlining; }
639 void set_print_inlining(bool z) { _print_inlining = z; }
640 bool print_intrinsics() const { return _print_intrinsics; }
641 void set_print_intrinsics(bool z) { _print_intrinsics = z; }
642 uint max_node_limit() const { return (uint)_max_node_limit; }
643 void set_max_node_limit(uint n) { _max_node_limit = n; }
644 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; }
645 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; }
646 bool has_monitors() const { return _has_monitors; }
647 void set_has_monitors(bool v) { _has_monitors = v; }
648 bool has_scoped_access() const { return _has_scoped_access; }
649 void set_has_scoped_access(bool v) { _has_scoped_access = v; }
650 bool has_clinit_barriers() { return _has_clinit_barriers; }
651 void set_has_clinit_barriers(bool z) { _has_clinit_barriers = z; }
652
653 // check the CompilerOracle for special behaviours for this compile
654 bool method_has_option(CompileCommandEnum option) {
655 return method() != nullptr && method()->has_option(option);
656 }
657
658 #ifndef PRODUCT
659 uint next_igv_idx() { return _igv_idx++; }
660 bool trace_opto_output() const { return _trace_opto_output; }
661 void print_ideal_ir(const char* phase_name);
662 bool should_print_ideal() const { return _directive->PrintIdealOption; }
663 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
664 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
665 int _in_dump_cnt; // Required for dumping ir nodes.
666 #endif
667 bool has_irreducible_loop() const { return _has_irreducible_loop; }
668 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
669
670 // JSR 292
671 bool has_method_handle_invokes() const { return _has_method_handle_invokes; }
672 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; }
673
674 Ticks _latest_stage_start_counter;
675
676 void begin_method();
677 void end_method();
678 bool should_print_igv(int level);
679 bool should_print_phase(CompilerPhaseType cpt);
680
681 void print_method(CompilerPhaseType cpt, int level, Node* n = nullptr);
682
683 #ifndef PRODUCT
684 void init_igv();
685 void dump_igv(const char* graph_name, int level = 3) {
686 if (should_print_igv(level)) {
687 _igv_printer->print_graph(graph_name);
688 }
689 }
690
691 void igv_print_method_to_file(const char* phase_name = "Debug", bool append = false);
692 void igv_print_method_to_network(const char* phase_name = "Debug");
693 void igv_print_graph_to_network(const char* name, Node* node, GrowableArray<const Node*>& visible_nodes);
694 static IdealGraphPrinter* debug_file_printer() { return _debug_file_printer; }
695 static IdealGraphPrinter* debug_network_printer() { return _debug_network_printer; }
696 #endif
697
698 const GrowableArray<ParsePredicateNode*>& parse_predicates() const {
699 return _parse_predicates;
700 }
701
702 const GrowableArray<OpaqueTemplateAssertionPredicateNode*>& template_assertion_predicate_opaques() const {
703 return _template_assertion_predicate_opaques;
704 }
705
706 int macro_count() const { return _macro_nodes.length(); }
707 int parse_predicate_count() const { return _parse_predicates.length(); }
708 int template_assertion_predicate_count() const { return _template_assertion_predicate_opaques.length(); }
709 int expensive_count() const { return _expensive_nodes.length(); }
710 int coarsened_count() const { return _coarsened_locks.length(); }
711
712 Node* macro_node(int idx) const { return _macro_nodes.at(idx); }
713
714 Node* expensive_node(int idx) const { return _expensive_nodes.at(idx); }
715
716 ConnectionGraph* congraph() { return _congraph;}
717 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;}
718 void add_macro_node(Node * n) {
719 //assert(n->is_macro(), "must be a macro node");
720 assert(!_macro_nodes.contains(n), "duplicate entry in expand list");
721 _macro_nodes.append(n);
722 }
723 void remove_macro_node(Node* n) {
724 // this function may be called twice for a node so we can only remove it
725 // if it's still existing.
726 _macro_nodes.remove_if_existing(n);
727 // Remove from coarsened locks list if present
728 if (coarsened_count() > 0) {
729 remove_coarsened_lock(n);
730 }
731 }
732 void add_expensive_node(Node* n);
733 void remove_expensive_node(Node* n) {
734 _expensive_nodes.remove_if_existing(n);
735 }
736
737 void add_parse_predicate(ParsePredicateNode* n) {
738 assert(!_parse_predicates.contains(n), "duplicate entry in Parse Predicate list");
739 _parse_predicates.append(n);
740 }
741
742 void remove_parse_predicate(ParsePredicateNode* n) {
743 if (parse_predicate_count() > 0) {
744 _parse_predicates.remove_if_existing(n);
745 }
746 }
747
748 void add_template_assertion_predicate_opaque(OpaqueTemplateAssertionPredicateNode* n) {
749 assert(!_template_assertion_predicate_opaques.contains(n),
750 "Duplicate entry in Template Assertion Predicate OpaqueTemplateAssertionPredicate list");
751 _template_assertion_predicate_opaques.append(n);
752 }
753
754 void remove_template_assertion_predicate_opaque(OpaqueTemplateAssertionPredicateNode* n) {
755 if (template_assertion_predicate_count() > 0) {
756 _template_assertion_predicate_opaques.remove_if_existing(n);
757 }
758 }
759 void add_coarsened_locks(GrowableArray<AbstractLockNode*>& locks);
760 void remove_coarsened_lock(Node* n);
761 bool coarsened_locks_consistent();
762 void mark_unbalanced_boxes() const;
763
764 bool post_loop_opts_phase() { return _post_loop_opts_phase; }
765 void set_post_loop_opts_phase() { _post_loop_opts_phase = true; }
766 void reset_post_loop_opts_phase() { _post_loop_opts_phase = false; }
767
768 #ifdef ASSERT
769 bool phase_verify_ideal_loop() const { return _phase_verify_ideal_loop; }
770 void set_phase_verify_ideal_loop() { _phase_verify_ideal_loop = true; }
771 void reset_phase_verify_ideal_loop() { _phase_verify_ideal_loop = false; }
772 #endif
773
774 bool allow_macro_nodes() { return _allow_macro_nodes; }
775 void reset_allow_macro_nodes() { _allow_macro_nodes = false; }
776
777 void record_for_post_loop_opts_igvn(Node* n);
778 void remove_from_post_loop_opts_igvn(Node* n);
779 void process_for_post_loop_opts_igvn(PhaseIterGVN& igvn);
780
781 void record_unstable_if_trap(UnstableIfTrap* trap);
782 bool remove_unstable_if_trap(CallStaticJavaNode* unc, bool yield);
783 void remove_useless_unstable_if_traps(Unique_Node_List &useful);
784 void process_for_unstable_if_traps(PhaseIterGVN& igvn);
785
786 bool merge_stores_phase() { return _merge_stores_phase; }
787 void set_merge_stores_phase() { _merge_stores_phase = true; }
788 void record_for_merge_stores_igvn(Node* n);
789 void remove_from_merge_stores_igvn(Node* n);
790 void process_for_merge_stores_igvn(PhaseIterGVN& igvn);
791
792 void shuffle_macro_nodes();
793 void sort_macro_nodes();
794
795 void mark_parse_predicate_nodes_useless(PhaseIterGVN& igvn);
796
797 // Are there candidate expensive nodes for optimization?
798 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
799 // Check whether n1 and n2 are similar
800 static int cmp_expensive_nodes(Node* n1, Node* n2);
801 // Sort expensive nodes to locate similar expensive nodes
802 void sort_expensive_nodes();
803
804 // Compilation environment.
805 Arena* comp_arena() { return &_comp_arena; }
806 ciEnv* env() const { return _env; }
807 CompileLog* log() const { return _log; }
808
809 bool failing_internal() const {
810 return _env->failing() ||
811 _failure_reason.get() != nullptr;
812 }
813
814 const char* failure_reason() const {
815 return _env->failing() ? _env->failure_reason()
816 : _failure_reason.get();
817 }
818
819 const CompilationFailureInfo* first_failure_details() const { return _first_failure_details; }
820
821 bool failing() {
822 if (failing_internal()) {
823 return true;
824 }
825 #ifdef ASSERT
826 // Disable stress code for PhaseIdealLoop verification (would have cascading effects).
827 if (phase_verify_ideal_loop()) {
828 return false;
829 }
830 if (StressBailout) {
831 return fail_randomly();
832 }
833 #endif
834 return false;
835 }
836
837 #ifdef ASSERT
838 bool fail_randomly();
839 bool failure_is_artificial();
840 #endif
841
842 bool failure_reason_is(const char* r) const {
843 return (r == _failure_reason.get()) ||
844 (r != nullptr &&
845 _failure_reason.get() != nullptr &&
846 strcmp(r, _failure_reason.get()) == 0);
847 }
848
849 void record_failure(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false));
850 void record_method_not_compilable(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false)) {
851 env()->record_method_not_compilable(reason);
852 // Record failure reason.
853 record_failure(reason DEBUG_ONLY(COMMA allow_multiple_failures));
854 }
855 bool check_node_count(uint margin, const char* reason) {
856 if (oom()) {
857 record_method_not_compilable_oom();
858 return true;
859 }
860 if (live_nodes() + margin > max_node_limit()) {
861 record_method_not_compilable(reason);
862 return true;
863 } else {
864 return false;
865 }
866 }
867 bool oom() const { return _oom; }
868 void set_oom() { _oom = true; }
869
870 // Node management
871 uint unique() const { return _unique; }
872 uint next_unique() { return _unique++; }
873 void set_unique(uint i) { _unique = i; }
874 Arena* node_arena() { return _node_arena; }
875 Arena* old_arena() { return (&_node_arena_one == _node_arena) ? &_node_arena_two : &_node_arena_one; }
876 RootNode* root() const { return _root; }
877 void set_root(RootNode* r) { _root = r; }
878 StartNode* start() const; // (Derived from root.)
879 void verify_start(StartNode* s) const NOT_DEBUG_RETURN;
880 Node* immutable_memory();
881
882 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; }
883 Node* recent_alloc_obj() const { return _recent_alloc_obj; }
884 void set_recent_alloc(Node* ctl, Node* obj) {
885 _recent_alloc_ctl = ctl;
886 _recent_alloc_obj = obj;
887 }
888 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return;
889 _dead_node_count++;
890 }
891 void reset_dead_node_list() { _dead_node_list.reset();
892 _dead_node_count = 0;
893 }
894 uint live_nodes() const {
895 int val = _unique - _dead_node_count;
896 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
897 return (uint) val;
898 }
899 #ifdef ASSERT
900 void set_phase_optimize_finished() { _phase_optimize_finished = true; }
901 bool phase_optimize_finished() const { return _phase_optimize_finished; }
902 uint count_live_nodes_by_graph_walk();
903 void print_missing_nodes();
904 #endif
905
906 // Record modified nodes to check that they are put on IGVN worklist
907 void record_modified_node(Node* n) NOT_DEBUG_RETURN;
908 void remove_modified_node(Node* n) NOT_DEBUG_RETURN;
909 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } )
910
911 MachConstantBaseNode* mach_constant_base_node();
912 bool has_mach_constant_base_node() const { return _mach_constant_base_node != nullptr; }
913 // Generated by adlc, true if CallNode requires MachConstantBase.
914 bool needs_deep_clone_jvms();
915
916 // Handy undefined Node
917 Node* top() const { return _top; }
918
919 // these are used by guys who need to know about creation and transformation of top:
920 Node* cached_top_node() { return _top; }
921 void set_cached_top_node(Node* tn);
922
923 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
924 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
925 Node_Notes* default_node_notes() const { return _default_node_notes; }
926 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
927
928 Node_Notes* node_notes_at(int idx);
929
930 inline bool set_node_notes_at(int idx, Node_Notes* value);
931 // Copy notes from source to dest, if they exist.
932 // Overwrite dest only if source provides something.
933 // Return true if information was moved.
934 bool copy_node_notes_to(Node* dest, Node* source);
935
936 // Workhorse function to sort out the blocked Node_Notes array:
937 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
938 int idx, bool can_grow = false);
939
940 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
941
942 // Type management
943 Arena* type_arena() { return _type_arena; }
944 Dict* type_dict() { return _type_dict; }
945 size_t type_last_size() { return _type_last_size; }
946 int num_alias_types() { return _num_alias_types; }
947
948 void init_type_arena() { _type_arena = &_Compile_types; }
949 void set_type_arena(Arena* a) { _type_arena = a; }
950 void set_type_dict(Dict* d) { _type_dict = d; }
951 void set_type_last_size(size_t sz) { _type_last_size = sz; }
952
953 const TypeFunc* last_tf(ciMethod* m) {
954 return (m == _last_tf_m) ? _last_tf : nullptr;
955 }
956 void set_last_tf(ciMethod* m, const TypeFunc* tf) {
957 assert(m != nullptr || tf == nullptr, "");
958 _last_tf_m = m;
959 _last_tf = tf;
960 }
961
962 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
963 AliasType* alias_type(const TypePtr* adr_type, ciField* field = nullptr) { return find_alias_type(adr_type, false, field); }
964 bool have_alias_type(const TypePtr* adr_type);
965 AliasType* alias_type(ciField* field);
966
967 int get_alias_index(const TypePtr* at) { return alias_type(at)->index(); }
968 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); }
969 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); }
970
971 // Building nodes
972 void rethrow_exceptions(JVMState* jvms);
973 void return_values(JVMState* jvms);
974 JVMState* build_start_state(StartNode* start, const TypeFunc* tf);
975
976 // Decide how to build a call.
977 // The profile factor is a discount to apply to this site's interp. profile.
978 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
979 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = nullptr,
980 bool allow_intrinsics = true);
981 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
982 return should_delay_string_inlining(call_method, jvms) ||
983 should_delay_boxing_inlining(call_method, jvms) ||
984 should_delay_vector_inlining(call_method, jvms);
985 }
986 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
987 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
988 bool should_delay_vector_inlining(ciMethod* call_method, JVMState* jvms);
989 bool should_delay_vector_reboxing_inlining(ciMethod* call_method, JVMState* jvms);
990
991 // Helper functions to identify inlining potential at call-site
992 ciMethod* optimize_virtual_call(ciMethod* caller, ciInstanceKlass* klass,
993 ciKlass* holder, ciMethod* callee,
994 const TypeOopPtr* receiver_type, bool is_virtual,
995 bool &call_does_dispatch, int &vtable_index,
996 bool check_access = true);
997 ciMethod* optimize_inlining(ciMethod* caller, ciInstanceKlass* klass, ciKlass* holder,
998 ciMethod* callee, const TypeOopPtr* receiver_type,
999 bool check_access = true);
1000
1001 // Report if there were too many traps at a current method and bci.
1002 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
1003 // If there is no MDO at all, report no trap unless told to assume it.
1004 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1005 // This version, unspecific to a particular bci, asks if
1006 // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
1007 bool too_many_traps(Deoptimization::DeoptReason reason,
1008 // Privately used parameter for logging:
1009 ciMethodData* logmd = nullptr);
1010 // Report if there were too many recompiles at a method and bci.
1011 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1012 // Report if there were too many traps or recompiles at a method and bci.
1013 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
1014 return too_many_traps(method, bci, reason) ||
1015 too_many_recompiles(method, bci, reason);
1016 }
1017 // Return a bitset with the reasons where deoptimization is allowed,
1018 // i.e., where there were not too many uncommon traps.
1019 int _allowed_reasons;
1020 int allowed_deopt_reasons() { return _allowed_reasons; }
1021 void set_allowed_deopt_reasons();
1022
1023 // Parsing, optimization
1024 PhaseGVN* initial_gvn() { return _initial_gvn; }
1025 Unique_Node_List* igvn_worklist() {
1026 assert(_igvn_worklist != nullptr, "must be created in Compile::Compile");
1027 return _igvn_worklist;
1028 }
1029 Type_Array* types() {
1030 assert(_types != nullptr, "must be created in Compile::Compile");
1031 return _types;
1032 }
1033 NodeHash* node_hash() {
1034 assert(_node_hash != nullptr, "must be created in Compile::Compile");
1035 return _node_hash;
1036 }
1037 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1038 inline void remove_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1039 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; }
1040
1041 // Replace n by nn using initial_gvn, calling hash_delete and
1042 // record_for_igvn as needed.
1043 void gvn_replace_by(Node* n, Node* nn);
1044
1045
1046 void identify_useful_nodes(Unique_Node_List &useful);
1047 void update_dead_node_list(Unique_Node_List &useful);
1048 void disconnect_useless_nodes(Unique_Node_List& useful, Unique_Node_List& worklist, const Unique_Node_List* root_and_safepoints = nullptr);
1049
1050 void remove_useless_node(Node* dead);
1051
1052 // Record this CallGenerator for inlining at the end of parsing.
1053 void add_late_inline(CallGenerator* cg) {
1054 _late_inlines.insert_before(_late_inlines_pos, cg);
1055 _late_inlines_pos++;
1056 }
1057
1058 void prepend_late_inline(CallGenerator* cg) {
1059 _late_inlines.insert_before(0, cg);
1060 }
1061
1062 void add_string_late_inline(CallGenerator* cg) {
1063 _string_late_inlines.push(cg);
1064 }
1065
1066 void add_boxing_late_inline(CallGenerator* cg) {
1067 _boxing_late_inlines.push(cg);
1068 }
1069
1070 void add_vector_reboxing_late_inline(CallGenerator* cg) {
1071 _vector_reboxing_late_inlines.push(cg);
1072 }
1073
1074 template<typename N, ENABLE_IF(std::is_base_of<Node, N>::value)>
1075 void remove_useless_nodes(GrowableArray<N*>& node_list, Unique_Node_List& useful);
1076
1077 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1078 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Node* dead);
1079
1080 void remove_useless_coarsened_locks(Unique_Node_List& useful);
1081
1082 void dump_print_inlining();
1083
1084 bool over_inlining_cutoff() const {
1085 if (!inlining_incrementally()) {
1086 return unique() > (uint)NodeCountInliningCutoff;
1087 } else {
1088 // Give some room for incremental inlining algorithm to "breathe"
1089 // and avoid thrashing when live node count is close to the limit.
1090 // Keep in mind that live_nodes() isn't accurate during inlining until
1091 // dead node elimination step happens (see Compile::inline_incrementally).
1092 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10;
1093 }
1094 }
1095
1096 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; }
1097 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--; }
1098 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; }
1099
1100 bool inline_incrementally_one();
1101 void inline_incrementally_cleanup(PhaseIterGVN& igvn);
1102 void inline_incrementally(PhaseIterGVN& igvn);
1103 bool should_delay_inlining() { return AlwaysIncrementalInline || (StressIncrementalInlining && (random() % 2) == 0); }
1104 void inline_string_calls(bool parse_time);
1105 void inline_boxing_calls(PhaseIterGVN& igvn);
1106 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1107 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn);
1108
1109 void inline_vector_reboxing_calls();
1110 bool has_vbox_nodes();
1111
1112 void process_late_inline_calls_no_inline(PhaseIterGVN& igvn);
1113
1114 // Matching, CFG layout, allocation, code generation
1115 PhaseCFG* cfg() { return _cfg; }
1116 bool has_java_calls() const { return _java_calls > 0; }
1117 int java_calls() const { return _java_calls; }
1118 int inner_loops() const { return _inner_loops; }
1119 Matcher* matcher() { return _matcher; }
1120 PhaseRegAlloc* regalloc() { return _regalloc; }
1121 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; }
1122 Arena* indexSet_arena() { return _indexSet_arena; }
1123 void* indexSet_free_block_list() { return _indexSet_free_block_list; }
1124 DebugInformationRecorder* debug_info() { return env()->debug_info(); }
1125
1126 void update_interpreter_frame_size(int size) {
1127 if (_interpreter_frame_size < size) {
1128 _interpreter_frame_size = size;
1129 }
1130 }
1131
1132 void set_matcher(Matcher* m) { _matcher = m; }
1133 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; }
1134 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; }
1135 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; }
1136
1137 void set_java_calls(int z) { _java_calls = z; }
1138 void set_inner_loops(int z) { _inner_loops = z; }
1139
1140 Dependencies* dependencies() { return env()->dependencies(); }
1141
1142 // Major entry point. Given a Scope, compile the associated method.
1143 // For normal compilations, entry_bci is InvocationEntryBci. For on stack
1144 // replacement, entry_bci indicates the bytecode for which to compile a
1145 // continuation.
1146 Compile(ciEnv* ci_env, ciMethod* target,
1147 int entry_bci, Options options, DirectiveSet* directive);
1148
1149 // Second major entry point. From the TypeFunc signature, generate code
1150 // to pass arguments from the Java calling convention to the C calling
1151 // convention.
1152 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1153 address stub_function, const char *stub_name,
1154 int is_fancy_jump, bool pass_tls,
1155 bool return_pc, DirectiveSet* directive);
1156
1157 ~Compile();
1158
1159 // Are we compiling a method?
1160 bool has_method() { return method() != nullptr; }
1161
1162 // Maybe print some information about this compile.
1163 void print_compile_messages();
1164
1165 // Final graph reshaping, a post-pass after the regular optimizer is done.
1166 bool final_graph_reshaping();
1167
1168 // returns true if adr is completely contained in the given alias category
1169 bool must_alias(const TypePtr* adr, int alias_idx);
1170
1171 // returns true if adr overlaps with the given alias category
1172 bool can_alias(const TypePtr* adr, int alias_idx);
1173
1174 // Stack slots that may be unused by the calling convention but must
1175 // otherwise be preserved. On Intel this includes the return address.
1176 // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1177 uint in_preserve_stack_slots() {
1178 return SharedRuntime::in_preserve_stack_slots();
1179 }
1180
1181 // "Top of Stack" slots that may be unused by the calling convention but must
1182 // otherwise be preserved.
1183 // On Intel these are not necessary and the value can be zero.
1184 static uint out_preserve_stack_slots() {
1185 return SharedRuntime::out_preserve_stack_slots();
1186 }
1187
1188 // Number of outgoing stack slots killed above the out_preserve_stack_slots
1189 // for calls to C. Supports the var-args backing area for register parms.
1190 uint varargs_C_out_slots_killed() const;
1191
1192 // Number of Stack Slots consumed by a synchronization entry
1193 int sync_stack_slots() const;
1194
1195 // Compute the name of old_SP. See <arch>.ad for frame layout.
1196 OptoReg::Name compute_old_SP();
1197
1198 private:
1199 // Phase control:
1200 void Init(bool aliasing); // Prepare for a single compilation
1201 void Optimize(); // Given a graph, optimize it
1202 void Code_Gen(); // Generate code from a graph
1203
1204 // Management of the AliasType table.
1205 void grow_alias_types();
1206 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1207 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1208 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field);
1209
1210 void verify_top(Node*) const PRODUCT_RETURN;
1211
1212 // Intrinsic setup.
1213 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor
1214 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper
1215 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn
1216 void register_intrinsic(CallGenerator* cg); // update fn
1217
1218 #ifndef PRODUCT
1219 static juint _intrinsic_hist_count[];
1220 static jubyte _intrinsic_hist_flags[];
1221 #endif
1222 // Function calls made by the public function final_graph_reshaping.
1223 // No need to be made public as they are not called elsewhere.
1224 void final_graph_reshaping_impl(Node *n, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1225 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop, Unique_Node_List& dead_nodes);
1226 void final_graph_reshaping_walk(Node_Stack& nstack, Node* root, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1227 void handle_div_mod_op(Node* n, BasicType bt, bool is_unsigned);
1228
1229 // Logic cone optimization.
1230 void optimize_logic_cones(PhaseIterGVN &igvn);
1231 void collect_logic_cone_roots(Unique_Node_List& list);
1232 void process_logic_cone_root(PhaseIterGVN &igvn, Node* n, VectorSet& visited);
1233 bool compute_logic_cone(Node* n, Unique_Node_List& partition, Unique_Node_List& inputs);
1234 uint compute_truth_table(Unique_Node_List& partition, Unique_Node_List& inputs);
1235 uint eval_macro_logic_op(uint func, uint op1, uint op2, uint op3);
1236 Node* xform_to_MacroLogicV(PhaseIterGVN &igvn, const TypeVect* vt, Unique_Node_List& partitions, Unique_Node_List& inputs);
1237 void check_no_dead_use() const NOT_DEBUG_RETURN;
1238
1239 public:
1240
1241 // Note: Histogram array size is about 1 Kb.
1242 enum { // flag bits:
1243 _intrinsic_worked = 1, // succeeded at least once
1244 _intrinsic_failed = 2, // tried it but it failed
1245 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1246 _intrinsic_virtual = 8, // was seen in the virtual form (rare)
1247 _intrinsic_both = 16 // was seen in the non-virtual form (usual)
1248 };
1249 // Update histogram. Return boolean if this is a first-time occurrence.
1250 static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1251 bool is_virtual, int flags) PRODUCT_RETURN0;
1252 static void print_intrinsic_statistics() PRODUCT_RETURN;
1253
1254 // Graph verification code
1255 // Walk the node list, verifying that there is a one-to-one correspondence
1256 // between Use-Def edges and Def-Use edges. The option no_dead_code enables
1257 // stronger checks that the graph is strongly connected from starting points
1258 // in both directions.
1259 // root_and_safepoints is used to give the starting points for the traversal.
1260 // If not supplied, only root is used. When this check is called after CCP,
1261 // we need to start traversal from Root and safepoints, just like CCP does its
1262 // own traversal (see PhaseCCP::transform for reasons).
1263 //
1264 // To call this function, there are 2 ways to go:
1265 // - give root_and_safepoints to start traversal everywhere needed (like after CCP)
1266 // - if the whole graph is assumed to be reachable from Root's input,
1267 // root_and_safepoints is not needed (like in PhaseRemoveUseless).
1268 //
1269 // Failure to specify root_and_safepoints in case the graph is not fully
1270 // reachable from Root's input make this check unsound (can miss inconsistencies)
1271 // and even incomplete (can make up non-existing problems) if no_dead_code is
1272 // true.
1273 void verify_graph_edges(bool no_dead_code = false, const Unique_Node_List* root_and_safepoints = nullptr) const PRODUCT_RETURN;
1274
1275 // Verify bi-directional correspondence of edges
1276 void verify_bidirectional_edges(Unique_Node_List& visited, const Unique_Node_List* root_and_safepoints = nullptr) const;
1277
1278 // End-of-run dumps.
1279 static void print_statistics() PRODUCT_RETURN;
1280
1281 // Verify ADLC assumptions during startup
1282 static void adlc_verification() PRODUCT_RETURN;
1283
1284 // Definitions of pd methods
1285 static void pd_compiler2_init();
1286
1287 // Static parse-time type checking logic for gen_subtype_check:
1288 enum SubTypeCheckResult { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1289 SubTypeCheckResult static_subtype_check(const TypeKlassPtr* superk, const TypeKlassPtr* subk, bool skip = StressReflectiveCode);
1290
1291 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1292 // Optional control dependency (for example, on range check)
1293 Node* ctrl = nullptr);
1294
1295 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1296 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl, bool carry_dependency = false);
1297
1298 // Auxiliary methods for randomized fuzzing/stressing
1299 int random();
1300 bool randomized_select(int count);
1301
1302 // seed random number generation and log the seed for repeatability.
1303 void initialize_stress_seed(const DirectiveSet* directive);
1304
1305 // supporting clone_map
1306 CloneMap& clone_map();
1307 void set_clone_map(Dict* d);
1308
1309 bool needs_clinit_barrier(ciField* ik, ciMethod* accessing_method);
1310 bool needs_clinit_barrier(ciMethod* ik, ciMethod* accessing_method);
1311 bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method);
1312
1313 #ifdef IA32
1314 private:
1315 bool _select_24_bit_instr; // We selected an instruction with a 24-bit result
1316 bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results
1317
1318 // Remember if this compilation changes hardware mode to 24-bit precision.
1319 void set_24_bit_selection_and_mode(bool selection, bool mode) {
1320 _select_24_bit_instr = selection;
1321 _in_24_bit_fp_mode = mode;
1322 }
1323
1324 public:
1325 bool select_24_bit_instr() const { return _select_24_bit_instr; }
1326 bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; }
1327 #endif // IA32
1328 #ifdef ASSERT
1329 VerifyMeetResult* _type_verify;
1330 void set_exception_backedge() { _exception_backedge = true; }
1331 bool has_exception_backedge() const { return _exception_backedge; }
1332 #endif
1333
1334 static bool push_thru_add(PhaseGVN* phase, Node* z, const TypeInteger* tz, const TypeInteger*& rx, const TypeInteger*& ry,
1335 BasicType out_bt, BasicType in_bt);
1336
1337 static Node* narrow_value(BasicType bt, Node* value, const Type* type, PhaseGVN* phase, bool transform_res);
1338 };
1339
1340 #endif // SHARE_OPTO_COMPILE_HPP