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/cHeapStringHolder.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/compiler_globals.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/phase.hpp"
41 #include "opto/phasetype.hpp"
42 #include "opto/printinlining.hpp"
43 #include "opto/regmask.hpp"
44 #include "runtime/deoptimization.hpp"
45 #include "runtime/sharedRuntime.hpp"
46 #include "runtime/timerTrace.hpp"
47 #include "runtime/vmThread.hpp"
48 #include "utilities/ticks.hpp"
49 #include "utilities/vmEnums.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 LoopOptsSkipSplitIf,
110 LoopOptsVerify
111 };
112
113 // The type of all node counts and indexes.
114 // It must hold at least 16 bits, but must also be fast to load and store.
115 // This type, if less than 32 bits, could limit the number of possible nodes.
116 // (To make this type platform-specific, move to globalDefinitions_xxx.hpp.)
117 typedef unsigned int node_idx_t;
118
119 class NodeCloneInfo {
120 private:
121 uint64_t _idx_clone_orig;
122 public:
123
124 void set_idx(node_idx_t idx) {
125 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx;
126 }
127 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); }
128
129 void set_gen(int generation) {
130 uint64_t g = (uint64_t)generation << 32;
131 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g;
132 }
133 int gen() const { return (int)(_idx_clone_orig >> 32); }
134
135 void set(uint64_t x) { _idx_clone_orig = x; }
136 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); }
137 uint64_t get() const { return _idx_clone_orig; }
138
139 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {}
140 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); }
141
142 void dump_on(outputStream* st) const;
143 };
144
145 class CloneMap {
146 friend class Compile;
147 private:
148 bool _debug;
149 Dict* _dict;
150 int _clone_idx; // current cloning iteration/generation in loop unroll
151 public:
152 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy
153 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; }
154 Dict* dict() const { return _dict; }
155 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == nullptr, "key existed"); _dict->Insert(_2p(key), (void*)val); }
156 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); }
157 void remove(node_idx_t key) { _dict->Delete(_2p(key)); }
158 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); }
159 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); }
160 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); }
161 int gen(const void* k) const { return gen(_2_node_idx_t(k)); }
162 int max_gen() const;
163 void clone(Node* old, Node* nnn, int gen);
164 void verify_insert_and_clone(Node* old, Node* nnn, int gen);
165 void dump(node_idx_t key, outputStream* st) const;
166
167 int clone_idx() const { return _clone_idx; }
168 void set_clone_idx(int x) { _clone_idx = x; }
169 bool is_debug() const { return _debug; }
170 void set_debug(bool debug) { _debug = debug; }
171
172 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); }
173 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); }
174 };
175
176 class Options {
177 friend class Compile;
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 _for_preload; // Generate code for preload (before Java method execution), do class init barriers
186 const bool _do_superword; // Do SuperWord
187 const bool _install_code; // Install the code that was compiled
188 public:
189 Options(bool subsume_loads,
190 bool do_escape_analysis,
191 bool do_iterative_escape_analysis,
192 bool do_reduce_allocation_merges,
193 bool eliminate_boxing,
194 bool do_locks_coarsening,
195 bool do_superword,
196 bool for_preload,
197 bool install_code) :
198 _subsume_loads(subsume_loads),
199 _do_escape_analysis(do_escape_analysis),
200 _do_iterative_escape_analysis(do_iterative_escape_analysis),
201 _do_reduce_allocation_merges(do_reduce_allocation_merges),
202 _eliminate_boxing(eliminate_boxing),
203 _do_locks_coarsening(do_locks_coarsening),
204 _for_preload(for_preload),
205 _do_superword(do_superword),
206 _install_code(install_code) {
207 }
208
209 static Options for_runtime_stub() {
210 return Options(
211 /* subsume_loads = */ true,
212 /* do_escape_analysis = */ false,
213 /* do_iterative_escape_analysis = */ false,
214 /* do_reduce_allocation_merges = */ false,
215 /* eliminate_boxing = */ false,
216 /* do_lock_coarsening = */ false,
217 /* for_preload = */ false,
218 /* do_superword = */ true,
219 /* install_code = */ true
220 );
221 }
222 };
223
224 //------------------------------Compile----------------------------------------
225 // This class defines a top-level Compiler invocation.
226
227 class Compile : public Phase {
228
229 public:
230 // Fixed alias indexes. (See also MergeMemNode.)
231 enum {
232 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value)
233 AliasIdxBot = 2, // pseudo-index, aliases to everything
234 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM
235 };
236
237 // Variant of TraceTime(nullptr, &_t_accumulator, CITime);
238 // Integrated with logging. If logging is turned on, and CITimeVerbose is true,
239 // then brackets are put into the log, with time stamps and node counts.
240 // (The time collection itself is always conditionalized on CITime.)
241 class TracePhase : public TraceTime {
242 private:
243 Compile* const _compile;
244 CompileLog* _log;
245 const bool _dolog;
246 public:
247 TracePhase(PhaseTraceId phaseTraceId);
248 TracePhase(const char* name, PhaseTraceId phaseTraceId);
249 ~TracePhase();
250 const char* phase_name() const { return title(); }
251 };
252
253 // Information per category of alias (memory slice)
254 class AliasType {
255 private:
256 friend class Compile;
257
258 int _index; // unique index, used with MergeMemNode
259 const TypePtr* _adr_type; // normalized address type
260 ciField* _field; // relevant instance field, or null if none
261 const Type* _element; // relevant array element type, or null if none
262 bool _is_rewritable; // false if the memory is write-once only
263 int _general_index; // if this is type is an instance, the general
264 // type that this is an instance of
265
266 void Init(int i, const TypePtr* at);
267
268 public:
269 int index() const { return _index; }
270 const TypePtr* adr_type() const { return _adr_type; }
271 ciField* field() const { return _field; }
272 const Type* element() const { return _element; }
273 bool is_rewritable() const { return _is_rewritable; }
274 bool is_volatile() const { return (_field ? _field->is_volatile() : false); }
275 int general_index() const { return (_general_index != 0) ? _general_index : _index; }
276
277 void set_rewritable(bool z) { _is_rewritable = z; }
278 void set_field(ciField* f) {
279 assert(!_field,"");
280 _field = f;
281 if (f->is_final() || f->is_stable()) {
282 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
283 _is_rewritable = false;
284 }
285 }
286 void set_element(const Type* e) {
287 assert(_element == nullptr, "");
288 _element = e;
289 }
290
291 BasicType basic_type() const;
292
293 void print_on(outputStream* st) PRODUCT_RETURN;
294 };
295
296 enum {
297 logAliasCacheSize = 6,
298 AliasCacheSize = (1<<logAliasCacheSize)
299 };
300 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type
301 enum {
302 trapHistLength = MethodData::_trap_hist_limit
303 };
304
305 private:
306 // Fixed parameters to this compilation.
307 const int _compile_id;
308 const Options _options; // Compilation options
309 ciMethod* _method; // The method being compiled.
310 int _entry_bci; // entry bci for osr methods.
311 const TypeFunc* _tf; // My kind of signature
312 InlineTree* _ilt; // Ditto (temporary).
313 address _stub_function; // VM entry for stub being compiled, or null
314 const char* _stub_name; // Name of stub or adapter being compiled, or null
315 StubId _stub_id; // unique id for stub or NO_STUBID
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 _phase_counter; // Counter for the number of already printed phases
355 uint _igv_idx; // Counter for IGV node identifiers
356 uint _igv_phase_iter[PHASE_NUM_TYPES]; // Counters for IGV phase iterations
357 bool _trace_opto_output;
358 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
359 #endif
360 bool _has_irreducible_loop; // Found irreducible loops
361 bool _has_monitors; // Metadata transfered to nmethod to enable Continuations lock-detection fastpath
362 bool _has_scoped_access; // For shared scope closure
363 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry
364 bool _has_clinit_barriers; // True if compiled code has clinit barriers
365 int _loop_opts_cnt; // loop opts round
366 uint _stress_seed; // Seed for stress testing
367
368 // Compilation environment.
369 Arena _comp_arena; // Arena with lifetime equivalent to Compile
370 void* _barrier_set_state; // Potential GC barrier state for Compile
371 ciEnv* _env; // CI interface
372 DirectiveSet* _directive; // Compiler directive
373 CompileLog* _log; // from CompilerThread
374 CHeapStringHolder _failure_reason; // for record_failure/failing pattern
375 CompilationFailureInfo* _first_failure_details; // Details for the first failure happening during compilation
376 GrowableArray<CallGenerator*> _intrinsics; // List of intrinsics.
377 GrowableArray<Node*> _macro_nodes; // List of nodes which need to be expanded before matching.
378 GrowableArray<ParsePredicateNode*> _parse_predicates; // List of Parse Predicates.
379 // List of OpaqueTemplateAssertionPredicateNode nodes for Template Assertion Predicates which can be seen as list
380 // of Template Assertion Predicates themselves.
381 GrowableArray<OpaqueTemplateAssertionPredicateNode*> _template_assertion_predicate_opaques;
382 GrowableArray<Node*> _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
383 GrowableArray<Node*> _for_post_loop_igvn; // List of nodes for IGVN after loop opts are over
384 GrowableArray<Node*> _for_merge_stores_igvn; // List of nodes for IGVN merge stores
385 GrowableArray<UnstableIfTrap*> _unstable_if_traps; // List of ifnodes after IGVN
386 GrowableArray<Node_List*> _coarsened_locks; // List of coarsened Lock and Unlock nodes
387 ConnectionGraph* _congraph;
388 #ifndef PRODUCT
389 IdealGraphPrinter* _igv_printer;
390 static IdealGraphPrinter* _debug_file_printer;
391 static IdealGraphPrinter* _debug_network_printer;
392 #endif
393
394
395 // Node management
396 uint _unique; // Counter for unique Node indices
397 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N).
398 // So use this to keep count and make the call O(1).
399 VectorSet _dead_node_list; // Set of dead nodes
400 DEBUG_ONLY(Unique_Node_List* _modified_nodes;) // List of nodes which inputs were modified
401 DEBUG_ONLY(bool _phase_optimize_finished;) // Used for live node verification while creating new nodes
402
403 DEBUG_ONLY(bool _phase_verify_ideal_loop;) // Are we in PhaseIdealLoop verification?
404
405 // Arenas for new-space and old-space nodes.
406 // Swapped between using _node_arena.
407 // The lifetime of the old-space nodes is during xform.
408 Arena _node_arena_one;
409 Arena _node_arena_two;
410 Arena* _node_arena;
411 public:
412 Arena* swap_old_and_new() {
413 Arena* filled_arena_ptr = _node_arena;
414 Arena* old_arena_ptr = old_arena();
415 old_arena_ptr->destruct_contents();
416 _node_arena = old_arena_ptr;
417 return filled_arena_ptr;
418 }
419 private:
420 RootNode* _root; // Unique root of compilation, or null after bail-out.
421 Node* _top; // Unique top node. (Reset by various phases.)
422
423 Node* _immutable_memory; // Initial memory state
424
425 Node* _recent_alloc_obj;
426 Node* _recent_alloc_ctl;
427
428 // Constant table
429 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton.
430
431
432 // Blocked array of debugging and profiling information,
433 // tracked per node.
434 enum { _log2_node_notes_block_size = 8,
435 _node_notes_block_size = (1<<_log2_node_notes_block_size)
436 };
437 GrowableArray<Node_Notes*>* _node_note_array;
438 Node_Notes* _default_node_notes; // default notes for new nodes
439
440 // After parsing and every bulk phase we hang onto the Root instruction.
441 // The RootNode instruction is where the whole program begins. It produces
442 // the initial Control and BOTTOM for everybody else.
443
444 // Type management
445 Arena _Compile_types; // Arena for all types
446 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared()
447 Dict* _type_dict; // Intern table
448 CloneMap _clone_map; // used for recording history of cloned nodes
449 size_t _type_last_size; // Last allocation size (see Type::operator new/delete)
450 ciMethod* _last_tf_m; // Cache for
451 const TypeFunc* _last_tf; // TypeFunc::make
452 AliasType** _alias_types; // List of alias types seen so far.
453 int _num_alias_types; // Logical length of _alias_types
454 int _max_alias_types; // Physical length of _alias_types
455 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
456
457 // Parsing, optimization
458 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN
459
460 // Shared worklist for all IGVN rounds. Nodes can be pushed to it at any time.
461 // If pushed outside IGVN, the Node is processed in the next IGVN round.
462 Unique_Node_List* _igvn_worklist;
463
464 // Shared type array for GVN, IGVN and CCP. It maps node idx -> Type*.
465 Type_Array* _types;
466
467 // Shared node hash table for GVN, IGVN and CCP.
468 NodeHash* _node_hash;
469
470 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after main parsing has finished.
471 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
472 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
473
474 GrowableArray<CallGenerator*> _vector_reboxing_late_inlines; // same but for vector reboxing operations
475
476 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
477 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending
478
479 // "MemLimit" directive was specified and the memory limit was hit during compilation
480 bool _oom;
481
482 // Only keep nodes in the expensive node list that need to be optimized
483 void cleanup_expensive_nodes(PhaseIterGVN &igvn);
484 // Use for sorting expensive nodes to bring similar nodes together
485 static int cmp_expensive_nodes(Node** n1, Node** n2);
486 // Expensive nodes list already sorted?
487 bool expensive_nodes_sorted() const;
488 // Remove the speculative part of types and clean up the graph
489 void remove_speculative_types(PhaseIterGVN &igvn);
490
491 void* _replay_inline_data; // Pointer to data loaded from file
492
493 void log_late_inline_failure(CallGenerator* cg, const char* msg);
494 DEBUG_ONLY(bool _exception_backedge;)
495
496 void record_method_not_compilable_oom();
497
498 InlinePrinter _inline_printer;
499
500 public:
501 void* barrier_set_state() const { return _barrier_set_state; }
502
503 InlinePrinter* inline_printer() { return &_inline_printer; }
504
505 #ifndef PRODUCT
506 IdealGraphPrinter* igv_printer() { return _igv_printer; }
507 void reset_igv_phase_iter(CompilerPhaseType cpt) { _igv_phase_iter[cpt] = 0; }
508 #endif
509
510 void log_late_inline(CallGenerator* cg);
511 void log_inline_id(CallGenerator* cg);
512 void log_inline_failure(const char* msg);
513
514 void* replay_inline_data() const { return _replay_inline_data; }
515
516 // Dump inlining replay data to the stream.
517 void dump_inline_data(outputStream* out);
518 void dump_inline_data_reduced(outputStream* out);
519
520 private:
521 // Matching, CFG layout, allocation, code generation
522 PhaseCFG* _cfg; // Results of CFG finding
523 int _java_calls; // Number of java calls in the method
524 int _inner_loops; // Number of inner loops in the method
525 Matcher* _matcher; // Engine to map ideal to machine instructions
526 PhaseRegAlloc* _regalloc; // Results of register allocation.
527 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout)
528 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin
529 void* _indexSet_free_block_list; // free list of IndexSet bit blocks
530 int _interpreter_frame_size;
531
532 // Holds dynamically allocated extensions of short-lived register masks. Such
533 // extensions are potentially quite large and need tight resource marks which
534 // may conflict with other allocations in the default resource area.
535 // Therefore, we use a dedicated resource area for register masks.
536 ResourceArea _regmask_arena;
537
538 PhaseOutput* _output;
539
540 public:
541 // Accessors
542
543 // The Compile instance currently active in this (compiler) thread.
544 static Compile* current() {
545 return (Compile*) ciEnv::current()->compiler_data();
546 }
547
548 int interpreter_frame_size() const { return _interpreter_frame_size; }
549
550 PhaseOutput* output() const { return _output; }
551 void set_output(PhaseOutput* o) { _output = o; }
552
553 // ID for this compilation. Useful for setting breakpoints in the debugger.
554 int compile_id() const { return _compile_id; }
555 DirectiveSet* directive() const { return _directive; }
556
557 // Does this compilation allow instructions to subsume loads? User
558 // instructions that subsume a load may result in an unschedulable
559 // instruction sequence.
560 bool subsume_loads() const { return _options._subsume_loads; }
561 /** Do escape analysis. */
562 bool do_escape_analysis() const { return _options._do_escape_analysis; }
563 bool do_iterative_escape_analysis() const { return _options._do_iterative_escape_analysis; }
564 bool do_reduce_allocation_merges() const { return _options._do_reduce_allocation_merges; }
565 /** Do boxing elimination. */
566 bool eliminate_boxing() const { return _options._eliminate_boxing; }
567 /** Do aggressive boxing elimination. */
568 bool aggressive_unboxing() const { return _options._eliminate_boxing && AggressiveUnboxing; }
569 bool should_install_code() const { return _options._install_code; }
570 /** Do locks coarsening. */
571 bool do_locks_coarsening() const { return _options._do_locks_coarsening; }
572 bool do_superword() const { return _options._do_superword; }
573
574 bool do_clinit_barriers() const { return _options._for_preload; }
575 bool for_preload() const { return _options._for_preload; }
576
577 // Other fixed compilation parameters.
578 ciMethod* method() const { return _method; }
579 int entry_bci() const { return _entry_bci; }
580 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; }
581 bool is_method_compilation() const { return (_method != nullptr && !_method->flags().is_native()); }
582 const TypeFunc* tf() const { assert(_tf!=nullptr, ""); return _tf; }
583 void init_tf(const TypeFunc* tf) { assert(_tf==nullptr, ""); _tf = tf; }
584 InlineTree* ilt() const { return _ilt; }
585 address stub_function() const { return _stub_function; }
586 const char* stub_name() const { return _stub_name; }
587 StubId stub_id() const { return _stub_id; }
588 address stub_entry_point() const { return _stub_entry_point; }
589 void set_stub_entry_point(address z) { _stub_entry_point = z; }
590
591 // Control of this compilation.
592 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; }
593 void set_fixed_slots(int n) { _fixed_slots = n; }
594 int major_progress() const { return _major_progress; }
595 void set_inlining_progress(bool z) { _inlining_progress = z; }
596 int inlining_progress() const { return _inlining_progress; }
597 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
598 int inlining_incrementally() const { return _inlining_incrementally; }
599 void set_do_cleanup(bool z) { _do_cleanup = z; }
600 int do_cleanup() const { return _do_cleanup; }
601 void set_major_progress() { _major_progress++; }
602 void restore_major_progress(int progress) { _major_progress += progress; }
603 void clear_major_progress() { _major_progress = 0; }
604 int max_inline_size() const { return _max_inline_size; }
605 void set_freq_inline_size(int n) { _freq_inline_size = n; }
606 int freq_inline_size() const { return _freq_inline_size; }
607 void set_max_inline_size(int n) { _max_inline_size = n; }
608 bool has_loops() const { return _has_loops; }
609 void set_has_loops(bool z) { _has_loops = z; }
610 bool has_split_ifs() const { return _has_split_ifs; }
611 void set_has_split_ifs(bool z) { _has_split_ifs = z; }
612 bool has_unsafe_access() const { return _has_unsafe_access; }
613 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; }
614 bool has_stringbuilder() const { return _has_stringbuilder; }
615 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; }
616 bool has_boxed_value() const { return _has_boxed_value; }
617 void set_has_boxed_value(bool z) { _has_boxed_value = z; }
618 bool has_reserved_stack_access() const { return _has_reserved_stack_access; }
619 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
620 uint max_vector_size() const { return _max_vector_size; }
621 void set_max_vector_size(uint s) { _max_vector_size = s; }
622 bool clear_upper_avx() const { return _clear_upper_avx; }
623 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; }
624 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; }
625 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
626 bool trap_can_recompile() const { return _trap_can_recompile; }
627 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; }
628 uint decompile_count() const { return _decompile_count; }
629 void set_decompile_count(uint c) { _decompile_count = c; }
630 bool allow_range_check_smearing() const;
631 bool do_inlining() const { return _do_inlining; }
632 void set_do_inlining(bool z) { _do_inlining = z; }
633 bool do_scheduling() const { return _do_scheduling; }
634 void set_do_scheduling(bool z) { _do_scheduling = z; }
635 bool do_freq_based_layout() const{ return _do_freq_based_layout; }
636 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
637 bool do_vector_loop() const { return _do_vector_loop; }
638 void set_do_vector_loop(bool z) { _do_vector_loop = z; }
639 bool use_cmove() const { return _use_cmove; }
640 void set_use_cmove(bool z) { _use_cmove = z; }
641 bool do_aliasing() const { return _do_aliasing; }
642 bool print_assembly() const { return _print_assembly; }
643 void set_print_assembly(bool z) { _print_assembly = z; }
644 bool print_inlining() const { return _print_inlining; }
645 void set_print_inlining(bool z) { _print_inlining = z; }
646 bool print_intrinsics() const { return _print_intrinsics; }
647 void set_print_intrinsics(bool z) { _print_intrinsics = z; }
648 uint max_node_limit() const { return (uint)_max_node_limit; }
649 void set_max_node_limit(uint n) { _max_node_limit = n; }
650 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; }
651 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; }
652 bool has_monitors() const { return _has_monitors; }
653 void set_has_monitors(bool v) { _has_monitors = v; }
654 bool has_scoped_access() const { return _has_scoped_access; }
655 void set_has_scoped_access(bool v) { _has_scoped_access = v; }
656 bool has_clinit_barriers() { return _has_clinit_barriers; }
657 void set_has_clinit_barriers(bool z) { _has_clinit_barriers = z; }
658
659 // check the CompilerOracle for special behaviours for this compile
660 bool method_has_option(CompileCommandEnum option) const {
661 return method() != nullptr && method()->has_option(option);
662 }
663
664 #ifndef PRODUCT
665 uint next_igv_idx() { return _igv_idx++; }
666 bool trace_opto_output() const { return _trace_opto_output; }
667 void print_phase(const char* phase_name);
668 void print_ideal_ir(const char* phase_name);
669 bool should_print_ideal() const { return _directive->PrintIdealOption; }
670 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
671 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
672 int _in_dump_cnt; // Required for dumping ir nodes.
673 #endif
674 bool has_irreducible_loop() const { return _has_irreducible_loop; }
675 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
676
677 Ticks _latest_stage_start_counter;
678
679 void begin_method();
680 void end_method();
681
682 void print_method(CompilerPhaseType cpt, int level, Node* n = nullptr);
683
684 #ifndef PRODUCT
685 bool should_print_igv(int level);
686 bool should_print_phase(int level) const;
687 bool should_print_ideal_phase(CompilerPhaseType cpt) const;
688 void init_igv();
689 void dump_igv(const char* graph_name, int level = 3) {
690 if (should_print_igv(level)) {
691 _igv_printer->print_graph(graph_name, nullptr);
692 }
693 }
694
695 void igv_print_method_to_file(const char* phase_name = nullptr, bool append = false, const frame* fr = nullptr);
696 void igv_print_method_to_network(const char* phase_name = nullptr, const frame* fr = nullptr);
697 void igv_print_graph_to_network(const char* name, GrowableArray<const Node*>& visible_nodes, const frame* fr);
698 static IdealGraphPrinter* debug_file_printer() { return _debug_file_printer; }
699 static IdealGraphPrinter* debug_network_printer() { return _debug_network_printer; }
700 #endif
701
702 const GrowableArray<ParsePredicateNode*>& parse_predicates() const {
703 return _parse_predicates;
704 }
705
706 const GrowableArray<OpaqueTemplateAssertionPredicateNode*>& template_assertion_predicate_opaques() const {
707 return _template_assertion_predicate_opaques;
708 }
709
710 int macro_count() const { return _macro_nodes.length(); }
711 int parse_predicate_count() const { return _parse_predicates.length(); }
712 int template_assertion_predicate_count() const { return _template_assertion_predicate_opaques.length(); }
713 int expensive_count() const { return _expensive_nodes.length(); }
714 int coarsened_count() const { return _coarsened_locks.length(); }
715
716 Node* macro_node(int idx) const { return _macro_nodes.at(idx); }
717
718 Node* expensive_node(int idx) const { return _expensive_nodes.at(idx); }
719
720 ConnectionGraph* congraph() { return _congraph;}
721 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;}
722 void add_macro_node(Node * n) {
723 //assert(n->is_macro(), "must be a macro node");
724 assert(!_macro_nodes.contains(n), "duplicate entry in expand list");
725 _macro_nodes.append(n);
726 }
727 void remove_macro_node(Node* n) {
728 // this function may be called twice for a node so we can only remove it
729 // if it's still existing.
730 _macro_nodes.remove_if_existing(n);
731 // Remove from coarsened locks list if present
732 if (coarsened_count() > 0) {
733 remove_coarsened_lock(n);
734 }
735 }
736 void add_expensive_node(Node* n);
737 void remove_expensive_node(Node* n) {
738 _expensive_nodes.remove_if_existing(n);
739 }
740
741 void add_parse_predicate(ParsePredicateNode* n) {
742 assert(!_parse_predicates.contains(n), "duplicate entry in Parse Predicate list");
743 _parse_predicates.append(n);
744 }
745
746 void remove_parse_predicate(ParsePredicateNode* n) {
747 if (parse_predicate_count() > 0) {
748 _parse_predicates.remove_if_existing(n);
749 }
750 }
751
752 void add_template_assertion_predicate_opaque(OpaqueTemplateAssertionPredicateNode* n) {
753 assert(!_template_assertion_predicate_opaques.contains(n),
754 "Duplicate entry in Template Assertion Predicate OpaqueTemplateAssertionPredicate list");
755 _template_assertion_predicate_opaques.append(n);
756 }
757
758 void remove_template_assertion_predicate_opaque(OpaqueTemplateAssertionPredicateNode* n) {
759 if (template_assertion_predicate_count() > 0) {
760 _template_assertion_predicate_opaques.remove_if_existing(n);
761 }
762 }
763 void add_coarsened_locks(GrowableArray<AbstractLockNode*>& locks);
764 void remove_coarsened_lock(Node* n);
765 bool coarsened_locks_consistent();
766 void mark_unbalanced_boxes() const;
767
768 bool post_loop_opts_phase() { return _post_loop_opts_phase; }
769 void set_post_loop_opts_phase() { _post_loop_opts_phase = true; }
770 void reset_post_loop_opts_phase() { _post_loop_opts_phase = false; }
771
772 #ifdef ASSERT
773 bool phase_verify_ideal_loop() const { return _phase_verify_ideal_loop; }
774 void set_phase_verify_ideal_loop() { _phase_verify_ideal_loop = true; }
775 void reset_phase_verify_ideal_loop() { _phase_verify_ideal_loop = false; }
776 #endif
777
778 bool allow_macro_nodes() { return _allow_macro_nodes; }
779 void reset_allow_macro_nodes() { _allow_macro_nodes = false; }
780
781 void record_for_post_loop_opts_igvn(Node* n);
782 void remove_from_post_loop_opts_igvn(Node* n);
783 void process_for_post_loop_opts_igvn(PhaseIterGVN& igvn);
784
785 void record_unstable_if_trap(UnstableIfTrap* trap);
786 bool remove_unstable_if_trap(CallStaticJavaNode* unc, bool yield);
787 void remove_useless_unstable_if_traps(Unique_Node_List &useful);
788 void process_for_unstable_if_traps(PhaseIterGVN& igvn);
789
790 bool merge_stores_phase() { return _merge_stores_phase; }
791 void set_merge_stores_phase() { _merge_stores_phase = true; }
792 void record_for_merge_stores_igvn(Node* n);
793 void remove_from_merge_stores_igvn(Node* n);
794 void process_for_merge_stores_igvn(PhaseIterGVN& igvn);
795
796 void shuffle_macro_nodes();
797 void sort_macro_nodes();
798
799 void mark_parse_predicate_nodes_useless(PhaseIterGVN& igvn);
800
801 // Are there candidate expensive nodes for optimization?
802 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
803 // Check whether n1 and n2 are similar
804 static int cmp_expensive_nodes(Node* n1, Node* n2);
805 // Sort expensive nodes to locate similar expensive nodes
806 void sort_expensive_nodes();
807
808 // Compilation environment.
809 Arena* comp_arena() { return &_comp_arena; }
810 ciEnv* env() const { return _env; }
811 CompileLog* log() const { return _log; }
812
813 bool failing_internal() const {
814 return _env->failing() ||
815 _failure_reason.get() != nullptr;
816 }
817
818 const char* failure_reason() const {
819 return _env->failing() ? _env->failure_reason()
820 : _failure_reason.get();
821 }
822
823 const CompilationFailureInfo* first_failure_details() const { return _first_failure_details; }
824
825 bool failing() {
826 if (failing_internal()) {
827 return true;
828 }
829 #ifdef ASSERT
830 // Disable stress code for PhaseIdealLoop verification (would have cascading effects).
831 if (phase_verify_ideal_loop()) {
832 return false;
833 }
834 if (StressBailout) {
835 return fail_randomly();
836 }
837 #endif
838 return false;
839 }
840
841 #ifdef ASSERT
842 bool fail_randomly();
843 bool failure_is_artificial();
844 #endif
845
846 bool failure_reason_is(const char* r) const {
847 return (r == _failure_reason.get()) ||
848 (r != nullptr &&
849 _failure_reason.get() != nullptr &&
850 strcmp(r, _failure_reason.get()) == 0);
851 }
852
853 void record_failure(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false));
854 void record_method_not_compilable(const char* reason DEBUG_ONLY(COMMA bool allow_multiple_failures = false)) {
855 env()->record_method_not_compilable(reason);
856 // Record failure reason.
857 record_failure(reason DEBUG_ONLY(COMMA allow_multiple_failures));
858 }
859 bool check_node_count(uint margin, const char* reason) {
860 if (oom()) {
861 record_method_not_compilable_oom();
862 return true;
863 }
864 if (live_nodes() + margin > max_node_limit()) {
865 record_method_not_compilable(reason);
866 return true;
867 } else {
868 return false;
869 }
870 }
871 bool oom() const { return _oom; }
872 void set_oom() { _oom = true; }
873
874 // Node management
875 uint unique() const { return _unique; }
876 uint next_unique() { return _unique++; }
877 void set_unique(uint i) { _unique = i; }
878 Arena* node_arena() { return _node_arena; }
879 Arena* old_arena() { return (&_node_arena_one == _node_arena) ? &_node_arena_two : &_node_arena_one; }
880 RootNode* root() const { return _root; }
881 void set_root(RootNode* r) { _root = r; }
882 StartNode* start() const; // (Derived from root.)
883 void verify_start(StartNode* s) const NOT_DEBUG_RETURN;
884 Node* immutable_memory();
885
886 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; }
887 Node* recent_alloc_obj() const { return _recent_alloc_obj; }
888 void set_recent_alloc(Node* ctl, Node* obj) {
889 _recent_alloc_ctl = ctl;
890 _recent_alloc_obj = obj;
891 }
892 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return;
893 _dead_node_count++;
894 }
895 void reset_dead_node_list() { _dead_node_list.reset();
896 _dead_node_count = 0;
897 }
898 uint live_nodes() const {
899 int val = _unique - _dead_node_count;
900 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
901 return (uint) val;
902 }
903 #ifdef ASSERT
904 void set_phase_optimize_finished() { _phase_optimize_finished = true; }
905 bool phase_optimize_finished() const { return _phase_optimize_finished; }
906 uint count_live_nodes_by_graph_walk();
907 void print_missing_nodes();
908 #endif
909
910 // Record modified nodes to check that they are put on IGVN worklist
911 void record_modified_node(Node* n) NOT_DEBUG_RETURN;
912 void remove_modified_node(Node* n) NOT_DEBUG_RETURN;
913 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } )
914
915 MachConstantBaseNode* mach_constant_base_node();
916 bool has_mach_constant_base_node() const { return _mach_constant_base_node != nullptr; }
917 // Generated by adlc, true if CallNode requires MachConstantBase.
918 bool needs_deep_clone_jvms();
919
920 // Handy undefined Node
921 Node* top() const { return _top; }
922
923 // these are used by guys who need to know about creation and transformation of top:
924 Node* cached_top_node() { return _top; }
925 void set_cached_top_node(Node* tn);
926
927 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
928 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
929 Node_Notes* default_node_notes() const { return _default_node_notes; }
930 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
931
932 Node_Notes* node_notes_at(int idx);
933
934 inline bool set_node_notes_at(int idx, Node_Notes* value);
935 // Copy notes from source to dest, if they exist.
936 // Overwrite dest only if source provides something.
937 // Return true if information was moved.
938 bool copy_node_notes_to(Node* dest, Node* source);
939
940 // Workhorse function to sort out the blocked Node_Notes array:
941 Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
942 int idx, bool can_grow = false);
943
944 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
945
946 // Type management
947 Arena* type_arena() { return _type_arena; }
948 Dict* type_dict() { return _type_dict; }
949 size_t type_last_size() { return _type_last_size; }
950 int num_alias_types() { return _num_alias_types; }
951
952 void init_type_arena() { _type_arena = &_Compile_types; }
953 void set_type_arena(Arena* a) { _type_arena = a; }
954 void set_type_dict(Dict* d) { _type_dict = d; }
955 void set_type_last_size(size_t sz) { _type_last_size = sz; }
956
957 const TypeFunc* last_tf(ciMethod* m) {
958 return (m == _last_tf_m) ? _last_tf : nullptr;
959 }
960 void set_last_tf(ciMethod* m, const TypeFunc* tf) {
961 assert(m != nullptr || tf == nullptr, "");
962 _last_tf_m = m;
963 _last_tf = tf;
964 }
965
966 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
967 AliasType* alias_type(const TypePtr* adr_type, ciField* field = nullptr) { return find_alias_type(adr_type, false, field); }
968 bool have_alias_type(const TypePtr* adr_type);
969 AliasType* alias_type(ciField* field);
970
971 int get_alias_index(const TypePtr* at) { return alias_type(at)->index(); }
972 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); }
973 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); }
974
975 // Building nodes
976 void rethrow_exceptions(JVMState* jvms);
977 void return_values(JVMState* jvms);
978 JVMState* build_start_state(StartNode* start, const TypeFunc* tf);
979
980 // Decide how to build a call.
981 // The profile factor is a discount to apply to this site's interp. profile.
982 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
983 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = nullptr,
984 bool allow_intrinsics = true);
985 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
986 return should_delay_string_inlining(call_method, jvms) ||
987 should_delay_boxing_inlining(call_method, jvms) ||
988 should_delay_vector_inlining(call_method, jvms);
989 }
990 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
991 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
992 bool should_delay_vector_inlining(ciMethod* call_method, JVMState* jvms);
993 bool should_delay_vector_reboxing_inlining(ciMethod* call_method, JVMState* jvms);
994
995 // Helper functions to identify inlining potential at call-site
996 ciMethod* optimize_virtual_call(ciMethod* caller, ciInstanceKlass* klass,
997 ciKlass* holder, ciMethod* callee,
998 const TypeOopPtr* receiver_type, bool is_virtual,
999 bool &call_does_dispatch, int &vtable_index,
1000 bool check_access = true);
1001 ciMethod* optimize_inlining(ciMethod* caller, ciInstanceKlass* klass, ciKlass* holder,
1002 ciMethod* callee, const TypeOopPtr* receiver_type,
1003 bool check_access = true);
1004
1005 // Report if there were too many traps at a current method and bci.
1006 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
1007 // If there is no MDO at all, report no trap unless told to assume it.
1008 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1009 // This version, unspecific to a particular bci, asks if
1010 // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
1011 bool too_many_traps(Deoptimization::DeoptReason reason,
1012 // Privately used parameter for logging:
1013 ciMethodData* logmd = nullptr);
1014 // Report if there were too many recompiles at a method and bci.
1015 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1016 // Report if there were too many traps or recompiles at a method and bci.
1017 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
1018 return too_many_traps(method, bci, reason) ||
1019 too_many_recompiles(method, bci, reason);
1020 }
1021 // Return a bitset with the reasons where deoptimization is allowed,
1022 // i.e., where there were not too many uncommon traps.
1023 int _allowed_reasons;
1024 int allowed_deopt_reasons() { return _allowed_reasons; }
1025 void set_allowed_deopt_reasons();
1026
1027 // Parsing, optimization
1028 PhaseGVN* initial_gvn() { return _initial_gvn; }
1029 Unique_Node_List* igvn_worklist() {
1030 assert(_igvn_worklist != nullptr, "must be created in Compile::Compile");
1031 return _igvn_worklist;
1032 }
1033 Type_Array* types() {
1034 assert(_types != nullptr, "must be created in Compile::Compile");
1035 return _types;
1036 }
1037 NodeHash* node_hash() {
1038 assert(_node_hash != nullptr, "must be created in Compile::Compile");
1039 return _node_hash;
1040 }
1041 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1042 inline void remove_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp.
1043 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; }
1044
1045 // Replace n by nn using initial_gvn, calling hash_delete and
1046 // record_for_igvn as needed.
1047 void gvn_replace_by(Node* n, Node* nn);
1048
1049
1050 void identify_useful_nodes(Unique_Node_List &useful);
1051 void update_dead_node_list(Unique_Node_List &useful);
1052 void disconnect_useless_nodes(Unique_Node_List& useful, Unique_Node_List& worklist, const Unique_Node_List* root_and_safepoints = nullptr);
1053
1054 void remove_useless_node(Node* dead);
1055
1056 // Record this CallGenerator for inlining at the end of parsing.
1057 void add_late_inline(CallGenerator* cg) {
1058 _late_inlines.insert_before(_late_inlines_pos, cg);
1059 _late_inlines_pos++;
1060 }
1061
1062 void prepend_late_inline(CallGenerator* cg) {
1063 _late_inlines.insert_before(0, cg);
1064 }
1065
1066 void add_string_late_inline(CallGenerator* cg) {
1067 _string_late_inlines.push(cg);
1068 }
1069
1070 void add_boxing_late_inline(CallGenerator* cg) {
1071 _boxing_late_inlines.push(cg);
1072 }
1073
1074 void add_vector_reboxing_late_inline(CallGenerator* cg) {
1075 _vector_reboxing_late_inlines.push(cg);
1076 }
1077
1078 template<typename N, ENABLE_IF(std::is_base_of<Node, N>::value)>
1079 void remove_useless_nodes(GrowableArray<N*>& node_list, Unique_Node_List& useful);
1080
1081 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1082 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Node* dead);
1083
1084 void remove_useless_coarsened_locks(Unique_Node_List& useful);
1085
1086 void dump_print_inlining();
1087
1088 bool over_inlining_cutoff() const {
1089 if (!inlining_incrementally()) {
1090 return unique() > (uint)NodeCountInliningCutoff;
1091 } else {
1092 // Give some room for incremental inlining algorithm to "breathe"
1093 // and avoid thrashing when live node count is close to the limit.
1094 // Keep in mind that live_nodes() isn't accurate during inlining until
1095 // dead node elimination step happens (see Compile::inline_incrementally).
1096 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10;
1097 }
1098 }
1099
1100 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; }
1101 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--; }
1102 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; }
1103
1104 bool inline_incrementally_one();
1105 void inline_incrementally_cleanup(PhaseIterGVN& igvn);
1106 void inline_incrementally(PhaseIterGVN& igvn);
1107 bool should_stress_inlining() { return StressIncrementalInlining && (random() % 2) == 0; }
1108 bool should_delay_inlining() { return AlwaysIncrementalInline || should_stress_inlining(); }
1109 void inline_string_calls(bool parse_time);
1110 void inline_boxing_calls(PhaseIterGVN& igvn);
1111 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1112 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn);
1113
1114 void inline_vector_reboxing_calls();
1115 bool has_vbox_nodes();
1116
1117 void process_late_inline_calls_no_inline(PhaseIterGVN& igvn);
1118
1119 // Matching, CFG layout, allocation, code generation
1120 PhaseCFG* cfg() { return _cfg; }
1121 bool has_java_calls() const { return _java_calls > 0; }
1122 int java_calls() const { return _java_calls; }
1123 int inner_loops() const { return _inner_loops; }
1124 Matcher* matcher() { return _matcher; }
1125 PhaseRegAlloc* regalloc() { return _regalloc; }
1126 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; }
1127 ResourceArea* regmask_arena() { return &_regmask_arena; }
1128 Arena* indexSet_arena() { return _indexSet_arena; }
1129 void* indexSet_free_block_list() { return _indexSet_free_block_list; }
1130 DebugInformationRecorder* debug_info() { return env()->debug_info(); }
1131
1132 void update_interpreter_frame_size(int size) {
1133 if (_interpreter_frame_size < size) {
1134 _interpreter_frame_size = size;
1135 }
1136 }
1137
1138 void set_matcher(Matcher* m) { _matcher = m; }
1139 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; }
1140 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; }
1141 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; }
1142
1143 void set_java_calls(int z) { _java_calls = z; }
1144 void set_inner_loops(int z) { _inner_loops = z; }
1145
1146 Dependencies* dependencies() { return env()->dependencies(); }
1147
1148 // Major entry point. Given a Scope, compile the associated method.
1149 // For normal compilations, entry_bci is InvocationEntryBci. For on stack
1150 // replacement, entry_bci indicates the bytecode for which to compile a
1151 // continuation.
1152 Compile(ciEnv* ci_env, ciMethod* target,
1153 int entry_bci, Options options, DirectiveSet* directive);
1154
1155 // Second major entry point. From the TypeFunc signature, generate code
1156 // to pass arguments from the Java calling convention to the C calling
1157 // convention.
1158 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1159 address stub_function, const char *stub_name,
1160 StubId stub_id, int is_fancy_jump, bool pass_tls,
1161 bool return_pc, DirectiveSet* directive);
1162
1163 ~Compile();
1164
1165 // Are we compiling a method?
1166 bool has_method() { return method() != nullptr; }
1167
1168 // Maybe print some information about this compile.
1169 void print_compile_messages();
1170
1171 // Final graph reshaping, a post-pass after the regular optimizer is done.
1172 bool final_graph_reshaping();
1173
1174 // returns true if adr is completely contained in the given alias category
1175 bool must_alias(const TypePtr* adr, int alias_idx);
1176
1177 // returns true if adr overlaps with the given alias category
1178 bool can_alias(const TypePtr* adr, int alias_idx);
1179
1180 // Stack slots that may be unused by the calling convention but must
1181 // otherwise be preserved. On Intel this includes the return address.
1182 // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1183 uint in_preserve_stack_slots() {
1184 return SharedRuntime::in_preserve_stack_slots();
1185 }
1186
1187 // "Top of Stack" slots that may be unused by the calling convention but must
1188 // otherwise be preserved.
1189 // On Intel these are not necessary and the value can be zero.
1190 static uint out_preserve_stack_slots() {
1191 return SharedRuntime::out_preserve_stack_slots();
1192 }
1193
1194 // Number of outgoing stack slots killed above the out_preserve_stack_slots
1195 // for calls to C. Supports the var-args backing area for register parms.
1196 uint varargs_C_out_slots_killed() const;
1197
1198 // Number of Stack Slots consumed by a synchronization entry
1199 int sync_stack_slots() const;
1200
1201 // Compute the name of old_SP. See <arch>.ad for frame layout.
1202 OptoReg::Name compute_old_SP();
1203
1204 private:
1205 // Phase control:
1206 void Init(bool aliasing); // Prepare for a single compilation
1207 void Optimize(); // Given a graph, optimize it
1208 void Code_Gen(); // Generate code from a graph
1209
1210 // Management of the AliasType table.
1211 void grow_alias_types();
1212 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1213 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1214 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field);
1215
1216 void verify_top(Node*) const PRODUCT_RETURN;
1217
1218 // Intrinsic setup.
1219 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor
1220 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper
1221 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn
1222 void register_intrinsic(CallGenerator* cg); // update fn
1223
1224 #ifndef PRODUCT
1225 static juint _intrinsic_hist_count[];
1226 static jubyte _intrinsic_hist_flags[];
1227 #endif
1228 // Function calls made by the public function final_graph_reshaping.
1229 // No need to be made public as they are not called elsewhere.
1230 void final_graph_reshaping_impl(Node *n, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1231 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop, Unique_Node_List& dead_nodes);
1232 void final_graph_reshaping_walk(Node_Stack& nstack, Node* root, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes);
1233 void handle_div_mod_op(Node* n, BasicType bt, bool is_unsigned);
1234
1235 // Logic cone optimization.
1236 void optimize_logic_cones(PhaseIterGVN &igvn);
1237 void collect_logic_cone_roots(Unique_Node_List& list);
1238 void process_logic_cone_root(PhaseIterGVN &igvn, Node* n, VectorSet& visited);
1239 bool compute_logic_cone(Node* n, Unique_Node_List& partition, Unique_Node_List& inputs);
1240 uint compute_truth_table(Unique_Node_List& partition, Unique_Node_List& inputs);
1241 uint eval_macro_logic_op(uint func, uint op1, uint op2, uint op3);
1242 Node* xform_to_MacroLogicV(PhaseIterGVN &igvn, const TypeVect* vt, Unique_Node_List& partitions, Unique_Node_List& inputs);
1243 void check_no_dead_use() const NOT_DEBUG_RETURN;
1244
1245 public:
1246
1247 // Note: Histogram array size is about 1 Kb.
1248 enum { // flag bits:
1249 _intrinsic_worked = 1, // succeeded at least once
1250 _intrinsic_failed = 2, // tried it but it failed
1251 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1252 _intrinsic_virtual = 8, // was seen in the virtual form (rare)
1253 _intrinsic_both = 16 // was seen in the non-virtual form (usual)
1254 };
1255 // Update histogram. Return boolean if this is a first-time occurrence.
1256 static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1257 bool is_virtual, int flags) PRODUCT_RETURN0;
1258 static void print_intrinsic_statistics() PRODUCT_RETURN;
1259
1260 // Graph verification code
1261 // Walk the node list, verifying that there is a one-to-one correspondence
1262 // between Use-Def edges and Def-Use edges. The option no_dead_code enables
1263 // stronger checks that the graph is strongly connected from starting points
1264 // in both directions.
1265 // root_and_safepoints is used to give the starting points for the traversal.
1266 // If not supplied, only root is used. When this check is called after CCP,
1267 // we need to start traversal from Root and safepoints, just like CCP does its
1268 // own traversal (see PhaseCCP::transform for reasons).
1269 //
1270 // To call this function, there are 2 ways to go:
1271 // - give root_and_safepoints to start traversal everywhere needed (like after CCP)
1272 // - if the whole graph is assumed to be reachable from Root's input,
1273 // root_and_safepoints is not needed (like in PhaseRemoveUseless).
1274 //
1275 // Failure to specify root_and_safepoints in case the graph is not fully
1276 // reachable from Root's input make this check unsound (can miss inconsistencies)
1277 // and even incomplete (can make up non-existing problems) if no_dead_code is
1278 // true.
1279 void verify_graph_edges(bool no_dead_code = false, const Unique_Node_List* root_and_safepoints = nullptr) const PRODUCT_RETURN;
1280
1281 // Verify bi-directional correspondence of edges
1282 void verify_bidirectional_edges(Unique_Node_List& visited, const Unique_Node_List* root_and_safepoints = nullptr) const;
1283
1284 // End-of-run dumps.
1285 static void print_statistics() PRODUCT_RETURN;
1286
1287 // Verify ADLC assumptions during startup
1288 static void adlc_verification() PRODUCT_RETURN;
1289
1290 // Definitions of pd methods
1291 static void pd_compiler2_init();
1292
1293 // Static parse-time type checking logic for gen_subtype_check:
1294 enum SubTypeCheckResult { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1295 SubTypeCheckResult static_subtype_check(const TypeKlassPtr* superk, const TypeKlassPtr* subk, bool skip = StressReflectiveCode);
1296
1297 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1298 // Optional control dependency (for example, on range check)
1299 Node* ctrl = nullptr);
1300
1301 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1302 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl, bool carry_dependency = false);
1303
1304 // Auxiliary methods for randomized fuzzing/stressing
1305 int random();
1306 bool randomized_select(int count);
1307
1308 // seed random number generation and log the seed for repeatability.
1309 void initialize_stress_seed(const DirectiveSet* directive);
1310
1311 // supporting clone_map
1312 CloneMap& clone_map();
1313 void set_clone_map(Dict* d);
1314
1315 bool needs_clinit_barrier(ciField* ik, ciMethod* accessing_method);
1316 bool needs_clinit_barrier(ciMethod* ik, ciMethod* accessing_method);
1317 bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method);
1318
1319 #ifdef ASSERT
1320 VerifyMeetResult* _type_verify;
1321 void set_exception_backedge() { _exception_backedge = true; }
1322 bool has_exception_backedge() const { return _exception_backedge; }
1323 #endif
1324
1325 static bool push_thru_add(PhaseGVN* phase, Node* z, const TypeInteger* tz, const TypeInteger*& rx, const TypeInteger*& ry,
1326 BasicType out_bt, BasicType in_bt);
1327
1328 static Node* narrow_value(BasicType bt, Node* value, const Type* type, PhaseGVN* phase, bool transform_res);
1329
1330 #ifndef PRODUCT
1331 private:
1332 // getting rid of the template makes things easier
1333 Node* make_debug_print_call(const char* str, address call_addr, PhaseGVN* gvn,
1334 Node* parm0 = nullptr, Node* parm1 = nullptr,
1335 Node* parm2 = nullptr, Node* parm3 = nullptr,
1336 Node* parm4 = nullptr, Node* parm5 = nullptr,
1337 Node* parm6 = nullptr) const;
1338
1339 public:
1340 // Creates a CallLeafNode for a runtime call that prints a static string and the values of the
1341 // nodes passed as arguments.
1342 // This function also takes care of doing the necessary wiring, including finding a suitable control
1343 // based on the nodes that need to be printed. Note that passing nodes that have incompatible controls
1344 // is undefined behavior.
1345 template <typename... TT, typename... NN>
1346 Node* make_debug_print(const char* str, PhaseGVN* gvn, NN... in) {
1347 address call_addr = CAST_FROM_FN_PTR(address, SharedRuntime::debug_print<TT...>);
1348 return make_debug_print_call(str, call_addr, gvn, in...);
1349 }
1350 #endif
1351 };
1352
1353 #endif // SHARE_OPTO_COMPILE_HPP