1 /* 2 * Copyright (c) 1997, 2023, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #ifndef SHARE_OPTO_COMPILE_HPP 26 #define SHARE_OPTO_COMPILE_HPP 27 28 #include "asm/codeBuffer.hpp" 29 #include "ci/compilerInterface.hpp" 30 #include "code/debugInfoRec.hpp" 31 #include "compiler/compiler_globals.hpp" 32 #include "compiler/compilerOracle.hpp" 33 #include "compiler/compileBroker.hpp" 34 #include "compiler/compilerEvent.hpp" 35 #include "libadt/dict.hpp" 36 #include "libadt/vectset.hpp" 37 #include "memory/resourceArea.hpp" 38 #include "oops/methodData.hpp" 39 #include "opto/idealGraphPrinter.hpp" 40 #include "opto/phasetype.hpp" 41 #include "opto/phase.hpp" 42 #include "opto/regmask.hpp" 43 #include "runtime/deoptimization.hpp" 44 #include "runtime/sharedRuntime.hpp" 45 #include "runtime/timerTrace.hpp" 46 #include "runtime/vmThread.hpp" 47 #include "utilities/ticks.hpp" 48 49 class AbstractLockNode; 50 class AddPNode; 51 class AllocateNode; 52 class Block; 53 class Bundle; 54 class CallGenerator; 55 class CallStaticJavaNode; 56 class CloneMap; 57 class CompilationFailureInfo; 58 class ConnectionGraph; 59 class IdealGraphPrinter; 60 class InlineTree; 61 class Matcher; 62 class MachConstantNode; 63 class MachConstantBaseNode; 64 class MachNode; 65 class MachOper; 66 class MachSafePointNode; 67 class Node; 68 class Node_Array; 69 class Node_List; 70 class Node_Notes; 71 class NodeHash; 72 class NodeCloneInfo; 73 class OptoReg; 74 class ParsePredicateNode; 75 class PartialEscapeAnalysis; 76 class PhaseCFG; 77 class PhaseGVN; 78 class PhaseIterGVN; 79 class PhaseRegAlloc; 80 class PhaseCCP; 81 class PhaseOutput; 82 class RootNode; 83 class relocInfo; 84 class StartNode; 85 class SafePointNode; 86 class JVMState; 87 class Type; 88 class TypeInt; 89 class TypeInteger; 90 class TypeKlassPtr; 91 class TypePtr; 92 class TypeOopPtr; 93 class TypeFunc; 94 class TypeVect; 95 class Type_Array; 96 class Unique_Node_List; 97 class UnstableIfTrap; 98 class nmethod; 99 class Node_Stack; 100 struct Final_Reshape_Counts; 101 class VerifyMeetResult; 102 103 typedef AllocateNode* ObjID; 104 105 enum LoopOptsMode { 106 LoopOptsDefault, 107 LoopOptsNone, 108 LoopOptsMaxUnroll, 109 LoopOptsShenandoahExpand, 110 LoopOptsShenandoahPostExpand, 111 LoopOptsSkipSplitIf, 112 LoopOptsVerify 113 }; 114 115 // The type of all node counts and indexes. 116 // It must hold at least 16 bits, but must also be fast to load and store. 117 // This type, if less than 32 bits, could limit the number of possible nodes. 118 // (To make this type platform-specific, move to globalDefinitions_xxx.hpp.) 119 typedef unsigned int node_idx_t; 120 121 class NodeCloneInfo { 122 private: 123 uint64_t _idx_clone_orig; 124 public: 125 126 void set_idx(node_idx_t idx) { 127 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx; 128 } 129 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); } 130 131 void set_gen(int generation) { 132 uint64_t g = (uint64_t)generation << 32; 133 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g; 134 } 135 int gen() const { return (int)(_idx_clone_orig >> 32); } 136 137 void set(uint64_t x) { _idx_clone_orig = x; } 138 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); } 139 uint64_t get() const { return _idx_clone_orig; } 140 141 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {} 142 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); } 143 144 void dump_on(outputStream* st) const; 145 }; 146 147 class CloneMap { 148 friend class Compile; 149 private: 150 bool _debug; 151 Dict* _dict; 152 int _clone_idx; // current cloning iteration/generation in loop unroll 153 public: 154 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy 155 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; } 156 Dict* dict() const { return _dict; } 157 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == nullptr, "key existed"); _dict->Insert(_2p(key), (void*)val); } 158 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); } 159 void remove(node_idx_t key) { _dict->Delete(_2p(key)); } 160 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); } 161 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); } 162 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); } 163 int gen(const void* k) const { return gen(_2_node_idx_t(k)); } 164 int max_gen() const; 165 void clone(Node* old, Node* nnn, int gen); 166 void verify_insert_and_clone(Node* old, Node* nnn, int gen); 167 void dump(node_idx_t key, outputStream* st) const; 168 169 int clone_idx() const { return _clone_idx; } 170 void set_clone_idx(int x) { _clone_idx = x; } 171 bool is_debug() const { return _debug; } 172 void set_debug(bool debug) { _debug = debug; } 173 174 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); } 175 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); } 176 }; 177 178 class Options { 179 friend class Compile; 180 friend class VMStructs; 181 private: 182 const bool _subsume_loads; // Load can be matched as part of a larger op. 183 const bool _do_escape_analysis; // Do escape analysis. 184 const bool _do_iterative_escape_analysis; // Do iterative escape analysis. 185 const bool _do_reduce_allocation_merges; // Do try to reduce allocation merges. 186 const bool _eliminate_boxing; // Do boxing elimination. 187 const bool _do_locks_coarsening; // Do locks coarsening 188 const bool _do_superword; // Do SuperWord 189 const bool _install_code; // Install the code that was compiled 190 public: 191 Options(bool subsume_loads, 192 bool do_escape_analysis, 193 bool do_iterative_escape_analysis, 194 bool do_reduce_allocation_merges, 195 bool eliminate_boxing, 196 bool do_locks_coarsening, 197 bool do_superword, 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 _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 /* do_superword = */ true, 218 /* install_code = */ true 219 ); 220 } 221 }; 222 223 //------------------------------Compile---------------------------------------- 224 // This class defines a top-level Compiler invocation. 225 226 class Compile : public Phase { 227 friend class VMStructs; 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* _compile; 244 CompileLog* _log; 245 const char* _phase_name; 246 bool _dolog; 247 public: 248 TracePhase(const char* name, elapsedTimer* accumulator); 249 ~TracePhase(); 250 }; 251 252 // Information per category of alias (memory slice) 253 class AliasType { 254 private: 255 friend class Compile; 256 257 int _index; // unique index, used with MergeMemNode 258 const TypePtr* _adr_type; // normalized address type 259 ciField* _field; // relevant instance field, or null if none 260 const Type* _element; // relevant array element type, or null if none 261 bool _is_rewritable; // false if the memory is write-once only 262 int _general_index; // if this is type is an instance, the general 263 // type that this is an instance of 264 265 void Init(int i, const TypePtr* at); 266 267 public: 268 int index() const { return _index; } 269 const TypePtr* adr_type() const { return _adr_type; } 270 ciField* field() const { return _field; } 271 const Type* element() const { return _element; } 272 bool is_rewritable() const { return _is_rewritable; } 273 bool is_volatile() const { return (_field ? _field->is_volatile() : false); } 274 int general_index() const { return (_general_index != 0) ? _general_index : _index; } 275 276 void set_rewritable(bool z) { _is_rewritable = z; } 277 void set_field(ciField* f) { 278 assert(!_field,""); 279 _field = f; 280 if (f->is_final() || f->is_stable()) { 281 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops. 282 _is_rewritable = false; 283 } 284 } 285 void set_element(const Type* e) { 286 assert(_element == nullptr, ""); 287 _element = e; 288 } 289 290 BasicType basic_type() const; 291 292 void print_on(outputStream* st) PRODUCT_RETURN; 293 }; 294 295 enum { 296 logAliasCacheSize = 6, 297 AliasCacheSize = (1<<logAliasCacheSize) 298 }; 299 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type 300 enum { 301 trapHistLength = MethodData::_trap_hist_limit 302 }; 303 304 private: 305 // Fixed parameters to this compilation. 306 const int _compile_id; 307 const Options _options; // Compilation options 308 ciMethod* _method; // The method being compiled. 309 int _entry_bci; // entry bci for osr methods. 310 const TypeFunc* _tf; // My kind of signature 311 InlineTree* _ilt; // Ditto (temporary). 312 address _stub_function; // VM entry for stub being compiled, or null 313 const char* _stub_name; // Name of stub or adapter being compiled, or null 314 address _stub_entry_point; // Compile code entry for generated stub, or null 315 316 // Control of this compilation. 317 int _max_inline_size; // Max inline size for this compilation 318 int _freq_inline_size; // Max hot method inline size for this compilation 319 int _fixed_slots; // count of frame slots not allocated by the register 320 // allocator i.e. locks, original deopt pc, etc. 321 uintx _max_node_limit; // Max unique node count during a single compilation. 322 323 bool _post_loop_opts_phase; // Loop opts are finished. 324 325 int _major_progress; // Count of something big happening 326 bool _inlining_progress; // progress doing incremental inlining? 327 bool _inlining_incrementally;// Are we doing incremental inlining (post parse) 328 bool _do_cleanup; // Cleanup is needed before proceeding with incremental inlining 329 bool _has_loops; // True if the method _may_ have some loops 330 bool _has_split_ifs; // True if the method _may_ have some split-if 331 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores. 332 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated 333 bool _has_boxed_value; // True if a boxed object is allocated 334 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess 335 uint _max_vector_size; // Maximum size of generated vectors 336 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper 337 uint _trap_hist[trapHistLength]; // Cumulative traps 338 bool _trap_can_recompile; // Have we emitted a recompiling trap? 339 uint _decompile_count; // Cumulative decompilation counts. 340 bool _do_inlining; // True if we intend to do inlining 341 bool _do_scheduling; // True if we intend to do scheduling 342 bool _do_freq_based_layout; // True if we intend to do frequency based block layout 343 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations 344 bool _use_cmove; // True if CMove should be used without profitability analysis 345 bool _do_aliasing; // True if we intend to do aliasing 346 bool _print_assembly; // True if we should dump assembly code for this compilation 347 bool _print_inlining; // True if we should print inlining for this compilation 348 bool _print_intrinsics; // True if we should print intrinsics for this compilation 349 #ifndef PRODUCT 350 uint _igv_idx; // Counter for IGV node identifiers 351 uint _igv_phase_iter[PHASE_NUM_TYPES]; // Counters for IGV phase iterations 352 bool _trace_opto_output; 353 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing 354 #endif 355 bool _has_irreducible_loop; // Found irreducible loops 356 // JSR 292 357 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes. 358 bool _has_monitors; // Metadata transfered to nmethod to enable Continuations lock-detection fastpath 359 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry 360 RTMState _rtm_state; // State of Restricted Transactional Memory usage 361 int _loop_opts_cnt; // loop opts round 362 uint _stress_seed; // Seed for stress testing 363 364 // Compilation environment. 365 Arena _comp_arena; // Arena with lifetime equivalent to Compile 366 Arena* _parser_arena; // Arena with lifttime equivalent to the top-level parser. 367 void* _barrier_set_state; // Potential GC barrier state for Compile 368 ciEnv* _env; // CI interface 369 DirectiveSet* _directive; // Compiler directive 370 CompileLog* _log; // from CompilerThread 371 const char* _failure_reason; // for record_failure/failing pattern 372 CompilationFailureInfo* _first_failure_details; // Details for the first failure happening during compilation 373 GrowableArray<CallGenerator*> _intrinsics; // List of intrinsics. 374 GrowableArray<Node*> _macro_nodes; // List of nodes which need to be expanded before matching. 375 GrowableArray<ParsePredicateNode*> _parse_predicates; // List of Parse Predicates. 376 GrowableArray<Node*> _template_assertion_predicate_opaqs; // List of Opaque4 nodes for Template Assertion Predicates. 377 GrowableArray<Node*> _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common 378 GrowableArray<Node*> _for_post_loop_igvn; // List of nodes for IGVN after loop opts are over 379 GrowableArray<UnstableIfTrap*> _unstable_if_traps; // List of ifnodes after IGVN 380 GrowableArray<Node_List*> _coarsened_locks; // List of coarsened Lock and Unlock nodes 381 PartialEscapeAnalysis* _pea; 382 ConnectionGraph* _congraph; 383 #ifndef PRODUCT 384 IdealGraphPrinter* _igv_printer; 385 static IdealGraphPrinter* _debug_file_printer; 386 static IdealGraphPrinter* _debug_network_printer; 387 #endif 388 389 390 // Node management 391 uint _unique; // Counter for unique Node indices 392 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N). 393 // So use this to keep count and make the call O(1). 394 VectorSet _dead_node_list; // Set of dead nodes 395 DEBUG_ONLY(Unique_Node_List* _modified_nodes;) // List of nodes which inputs were modified 396 DEBUG_ONLY(bool _phase_optimize_finished;) // Used for live node verification while creating new nodes 397 398 // Arenas for new-space and old-space nodes. 399 // Swapped between using _node_arena. 400 // The lifetime of the old-space nodes is during xform. 401 Arena _node_arena_one; 402 Arena _node_arena_two; 403 Arena* _node_arena; 404 public: 405 Arena* swap_old_and_new() { 406 Arena* filled_arena_ptr = _node_arena; 407 Arena* old_arena_ptr = old_arena(); 408 old_arena_ptr->destruct_contents(); 409 _node_arena = old_arena_ptr; 410 return filled_arena_ptr; 411 } 412 private: 413 RootNode* _root; // Unique root of compilation, or null after bail-out. 414 Node* _top; // Unique top node. (Reset by various phases.) 415 416 Node* _immutable_memory; // Initial memory state 417 418 Node* _recent_alloc_obj; 419 Node* _recent_alloc_ctl; 420 421 // Constant table 422 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton. 423 424 425 // Blocked array of debugging and profiling information, 426 // tracked per node. 427 enum { _log2_node_notes_block_size = 8, 428 _node_notes_block_size = (1<<_log2_node_notes_block_size) 429 }; 430 GrowableArray<Node_Notes*>* _node_note_array; 431 Node_Notes* _default_node_notes; // default notes for new nodes 432 433 // After parsing and every bulk phase we hang onto the Root instruction. 434 // The RootNode instruction is where the whole program begins. It produces 435 // the initial Control and BOTTOM for everybody else. 436 437 // Type management 438 Arena _Compile_types; // Arena for all types 439 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared() 440 Dict* _type_dict; // Intern table 441 CloneMap _clone_map; // used for recording history of cloned nodes 442 size_t _type_last_size; // Last allocation size (see Type::operator new/delete) 443 ciMethod* _last_tf_m; // Cache for 444 const TypeFunc* _last_tf; // TypeFunc::make 445 AliasType** _alias_types; // List of alias types seen so far. 446 int _num_alias_types; // Logical length of _alias_types 447 int _max_alias_types; // Physical length of _alias_types 448 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking 449 450 // Parsing, optimization 451 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN 452 453 // Shared worklist for all IGVN rounds. Nodes can be pushed to it at any time. 454 // If pushed outside IGVN, the Node is processed in the next IGVN round. 455 Unique_Node_List* _igvn_worklist; 456 457 // Shared type array for GVN, IGVN and CCP. It maps node idx -> Type*. 458 Type_Array* _types; 459 460 // Shared node hash table for GVN, IGVN and CCP. 461 NodeHash* _node_hash; 462 463 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after main parsing has finished. 464 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations 465 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations 466 467 GrowableArray<CallGenerator*> _vector_reboxing_late_inlines; // same but for vector reboxing operations 468 469 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining) 470 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending 471 472 // "MemLimit" directive was specified and the memory limit was hit during compilation 473 bool _oom; 474 475 // Inlining may not happen in parse order which would make 476 // PrintInlining output confusing. Keep track of PrintInlining 477 // pieces in order. 478 class PrintInliningBuffer : public CHeapObj<mtCompiler> { 479 private: 480 CallGenerator* _cg; 481 stringStream _ss; 482 static const size_t default_stream_buffer_size = 128; 483 484 public: 485 PrintInliningBuffer() 486 : _cg(nullptr), _ss(default_stream_buffer_size) {} 487 488 stringStream* ss() { return &_ss; } 489 CallGenerator* cg() { return _cg; } 490 void set_cg(CallGenerator* cg) { _cg = cg; } 491 }; 492 493 stringStream* _print_inlining_stream; 494 GrowableArray<PrintInliningBuffer*>* _print_inlining_list; 495 int _print_inlining_idx; 496 char* _print_inlining_output; 497 498 // Only keep nodes in the expensive node list that need to be optimized 499 void cleanup_expensive_nodes(PhaseIterGVN &igvn); 500 // Use for sorting expensive nodes to bring similar nodes together 501 static int cmp_expensive_nodes(Node** n1, Node** n2); 502 // Expensive nodes list already sorted? 503 bool expensive_nodes_sorted() const; 504 // Remove the speculative part of types and clean up the graph 505 void remove_speculative_types(PhaseIterGVN &igvn); 506 507 void* _replay_inline_data; // Pointer to data loaded from file 508 509 void print_inlining_init(); 510 void print_inlining_reinit(); 511 void print_inlining_commit(); 512 void print_inlining_push(); 513 PrintInliningBuffer* print_inlining_current(); 514 515 void log_late_inline_failure(CallGenerator* cg, const char* msg); 516 DEBUG_ONLY(bool _exception_backedge;) 517 518 void record_method_not_compilable_oom(); 519 520 public: 521 522 void* barrier_set_state() const { return _barrier_set_state; } 523 524 stringStream* print_inlining_stream() { 525 assert(print_inlining() || print_intrinsics(), "PrintInlining off?"); 526 return _print_inlining_stream; 527 } 528 529 void print_inlining_update(CallGenerator* cg); 530 void print_inlining_update_delayed(CallGenerator* cg); 531 void print_inlining_move_to(CallGenerator* cg); 532 void print_inlining_assert_ready(); 533 void print_inlining_reset(); 534 535 void print_inlining(ciMethod* method, int inline_level, int bci, InliningResult result, const char* msg = nullptr) { 536 stringStream ss; 537 CompileTask::print_inlining_inner(&ss, method, inline_level, bci, result, msg); 538 print_inlining_stream()->print("%s", ss.freeze()); 539 } 540 541 #ifndef PRODUCT 542 IdealGraphPrinter* igv_printer() { return _igv_printer; } 543 void reset_igv_phase_iter(CompilerPhaseType cpt) { _igv_phase_iter[cpt] = 0; } 544 #endif 545 546 void log_late_inline(CallGenerator* cg); 547 void log_inline_id(CallGenerator* cg); 548 void log_inline_failure(const char* msg); 549 550 void* replay_inline_data() const { return _replay_inline_data; } 551 552 // Dump inlining replay data to the stream. 553 void dump_inline_data(outputStream* out); 554 void dump_inline_data_reduced(outputStream* out); 555 556 private: 557 // Matching, CFG layout, allocation, code generation 558 PhaseCFG* _cfg; // Results of CFG finding 559 int _java_calls; // Number of java calls in the method 560 int _inner_loops; // Number of inner loops in the method 561 Matcher* _matcher; // Engine to map ideal to machine instructions 562 PhaseRegAlloc* _regalloc; // Results of register allocation. 563 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout) 564 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin 565 void* _indexSet_free_block_list; // free list of IndexSet bit blocks 566 int _interpreter_frame_size; 567 568 PhaseOutput* _output; 569 570 public: 571 // Accessors 572 573 // The Compile instance currently active in this (compiler) thread. 574 static Compile* current() { 575 return (Compile*) ciEnv::current()->compiler_data(); 576 } 577 578 int interpreter_frame_size() const { return _interpreter_frame_size; } 579 580 PhaseOutput* output() const { return _output; } 581 void set_output(PhaseOutput* o) { _output = o; } 582 583 // ID for this compilation. Useful for setting breakpoints in the debugger. 584 int compile_id() const { return _compile_id; } 585 DirectiveSet* directive() const { return _directive; } 586 587 // Does this compilation allow instructions to subsume loads? User 588 // instructions that subsume a load may result in an unschedulable 589 // instruction sequence. 590 bool subsume_loads() const { return _options._subsume_loads; } 591 /** Do escape analysis. */ 592 bool do_escape_analysis() const { return _options._do_escape_analysis; } 593 bool do_iterative_escape_analysis() const { return _options._do_iterative_escape_analysis; } 594 bool do_reduce_allocation_merges() const { return _options._do_reduce_allocation_merges; } 595 /** Do boxing elimination. */ 596 bool eliminate_boxing() const { return _options._eliminate_boxing; } 597 /** Do aggressive boxing elimination. */ 598 bool aggressive_unboxing() const { return _options._eliminate_boxing && AggressiveUnboxing; } 599 bool should_install_code() const { return _options._install_code; } 600 /** Do locks coarsening. */ 601 bool do_locks_coarsening() const { return _options._do_locks_coarsening; } 602 bool do_superword() const { return _options._do_superword; } 603 604 // Other fixed compilation parameters. 605 ciMethod* method() const { return _method; } 606 int entry_bci() const { return _entry_bci; } 607 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; } 608 bool is_method_compilation() const { return (_method != nullptr && !_method->flags().is_native()); } 609 const TypeFunc* tf() const { assert(_tf!=nullptr, ""); return _tf; } 610 void init_tf(const TypeFunc* tf) { assert(_tf==nullptr, ""); _tf = tf; } 611 InlineTree* ilt() const { return _ilt; } 612 address stub_function() const { return _stub_function; } 613 const char* stub_name() const { return _stub_name; } 614 address stub_entry_point() const { return _stub_entry_point; } 615 void set_stub_entry_point(address z) { _stub_entry_point = z; } 616 617 // Control of this compilation. 618 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; } 619 void set_fixed_slots(int n) { _fixed_slots = n; } 620 int major_progress() const { return _major_progress; } 621 void set_inlining_progress(bool z) { _inlining_progress = z; } 622 int inlining_progress() const { return _inlining_progress; } 623 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; } 624 int inlining_incrementally() const { return _inlining_incrementally; } 625 void set_do_cleanup(bool z) { _do_cleanup = z; } 626 int do_cleanup() const { return _do_cleanup; } 627 void set_major_progress() { _major_progress++; } 628 void restore_major_progress(int progress) { _major_progress += progress; } 629 void clear_major_progress() { _major_progress = 0; } 630 int max_inline_size() const { return _max_inline_size; } 631 void set_freq_inline_size(int n) { _freq_inline_size = n; } 632 int freq_inline_size() const { return _freq_inline_size; } 633 void set_max_inline_size(int n) { _max_inline_size = n; } 634 bool has_loops() const { return _has_loops; } 635 void set_has_loops(bool z) { _has_loops = z; } 636 bool has_split_ifs() const { return _has_split_ifs; } 637 void set_has_split_ifs(bool z) { _has_split_ifs = z; } 638 bool has_unsafe_access() const { return _has_unsafe_access; } 639 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; } 640 bool has_stringbuilder() const { return _has_stringbuilder; } 641 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; } 642 bool has_boxed_value() const { return _has_boxed_value; } 643 void set_has_boxed_value(bool z) { _has_boxed_value = z; } 644 bool has_reserved_stack_access() const { return _has_reserved_stack_access; } 645 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; } 646 uint max_vector_size() const { return _max_vector_size; } 647 void set_max_vector_size(uint s) { _max_vector_size = s; } 648 bool clear_upper_avx() const { return _clear_upper_avx; } 649 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; } 650 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; } 651 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; } 652 bool trap_can_recompile() const { return _trap_can_recompile; } 653 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; } 654 uint decompile_count() const { return _decompile_count; } 655 void set_decompile_count(uint c) { _decompile_count = c; } 656 bool allow_range_check_smearing() const; 657 bool do_inlining() const { return _do_inlining; } 658 void set_do_inlining(bool z) { _do_inlining = z; } 659 bool do_scheduling() const { return _do_scheduling; } 660 void set_do_scheduling(bool z) { _do_scheduling = z; } 661 bool do_freq_based_layout() const{ return _do_freq_based_layout; } 662 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; } 663 bool do_vector_loop() const { return _do_vector_loop; } 664 void set_do_vector_loop(bool z) { _do_vector_loop = z; } 665 bool use_cmove() const { return _use_cmove; } 666 void set_use_cmove(bool z) { _use_cmove = z; } 667 bool do_aliasing() const { return _do_aliasing; } 668 bool print_assembly() const { return _print_assembly; } 669 void set_print_assembly(bool z) { _print_assembly = z; } 670 bool print_inlining() const { return _print_inlining; } 671 void set_print_inlining(bool z) { _print_inlining = z; } 672 bool print_intrinsics() const { return _print_intrinsics; } 673 void set_print_intrinsics(bool z) { _print_intrinsics = z; } 674 RTMState rtm_state() const { return _rtm_state; } 675 void set_rtm_state(RTMState s) { _rtm_state = s; } 676 bool use_rtm() const { return (_rtm_state & NoRTM) == 0; } 677 bool profile_rtm() const { return _rtm_state == ProfileRTM; } 678 uint max_node_limit() const { return (uint)_max_node_limit; } 679 void set_max_node_limit(uint n) { _max_node_limit = n; } 680 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; } 681 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; } 682 bool has_monitors() const { return _has_monitors; } 683 void set_has_monitors(bool v) { _has_monitors = v; } 684 685 // check the CompilerOracle for special behaviours for this compile 686 bool method_has_option(enum CompileCommand option) { 687 return method() != nullptr && method()->has_option(option); 688 } 689 690 #ifndef PRODUCT 691 uint next_igv_idx() { return _igv_idx++; } 692 bool trace_opto_output() const { return _trace_opto_output; } 693 void print_ideal_ir(const char* phase_name); 694 bool should_print_ideal() const { return _directive->PrintIdealOption; } 695 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; } 696 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; } 697 int _in_dump_cnt; // Required for dumping ir nodes. 698 #endif 699 bool has_irreducible_loop() const { return _has_irreducible_loop; } 700 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; } 701 702 // JSR 292 703 bool has_method_handle_invokes() const { return _has_method_handle_invokes; } 704 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; } 705 706 Ticks _latest_stage_start_counter; 707 708 void begin_method(); 709 void end_method(); 710 bool should_print_igv(int level); 711 bool should_print_phase(CompilerPhaseType cpt); 712 713 void print_method(CompilerPhaseType cpt, int level, Node* n = nullptr); 714 715 #ifndef PRODUCT 716 void dump_igv(const char* graph_name, int level = 3) { 717 if (should_print_igv(level)) { 718 _igv_printer->print_method(graph_name, level); 719 } 720 } 721 722 void igv_print_method_to_file(const char* phase_name = "Debug", bool append = false); 723 void igv_print_method_to_network(const char* phase_name = "Debug"); 724 static IdealGraphPrinter* debug_file_printer() { return _debug_file_printer; } 725 static IdealGraphPrinter* debug_network_printer() { return _debug_network_printer; } 726 #endif 727 728 int macro_count() const { return _macro_nodes.length(); } 729 int parse_predicate_count() const { return _parse_predicates.length(); } 730 int template_assertion_predicate_count() const { return _template_assertion_predicate_opaqs.length(); } 731 int expensive_count() const { return _expensive_nodes.length(); } 732 int coarsened_count() const { return _coarsened_locks.length(); } 733 734 Node* macro_node(int idx) const { return _macro_nodes.at(idx); } 735 ParsePredicateNode* parse_predicate(int idx) const { return _parse_predicates.at(idx); } 736 737 Node* template_assertion_predicate_opaq_node(int idx) const { 738 return _template_assertion_predicate_opaqs.at(idx); 739 } 740 741 Node* expensive_node(int idx) const { return _expensive_nodes.at(idx); } 742 743 ConnectionGraph* congraph() { return _congraph;} 744 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;} 745 void add_macro_node(Node * n) { 746 //assert(n->is_macro(), "must be a macro node"); 747 assert(!_macro_nodes.contains(n), "duplicate entry in expand list"); 748 _macro_nodes.append(n); 749 } 750 void remove_macro_node(Node* n) { 751 // this function may be called twice for a node so we can only remove it 752 // if it's still existing. 753 _macro_nodes.remove_if_existing(n); 754 // Remove from coarsened locks list if present 755 if (coarsened_count() > 0) { 756 remove_coarsened_lock(n); 757 } 758 } 759 void add_expensive_node(Node* n); 760 void remove_expensive_node(Node* n) { 761 _expensive_nodes.remove_if_existing(n); 762 } 763 764 void add_parse_predicate(ParsePredicateNode* n) { 765 assert(!_parse_predicates.contains(n), "duplicate entry in Parse Predicate list"); 766 _parse_predicates.append(n); 767 } 768 769 void remove_parse_predicate(ParsePredicateNode* n) { 770 if (parse_predicate_count() > 0) { 771 _parse_predicates.remove_if_existing(n); 772 } 773 } 774 775 void add_template_assertion_predicate_opaq(Node* n) { 776 assert(!_template_assertion_predicate_opaqs.contains(n), 777 "duplicate entry in template assertion predicate opaque4 list"); 778 _template_assertion_predicate_opaqs.append(n); 779 } 780 781 void remove_template_assertion_predicate_opaq(Node* n) { 782 if (template_assertion_predicate_count() > 0) { 783 _template_assertion_predicate_opaqs.remove_if_existing(n); 784 } 785 } 786 void add_coarsened_locks(GrowableArray<AbstractLockNode*>& locks); 787 void remove_coarsened_lock(Node* n); 788 bool coarsened_locks_consistent(); 789 790 PartialEscapeAnalysis* PEA() const { 791 assert(!DoPartialEscapeAnalysis || _pea != nullptr, "sanity check"); 792 return _pea; 793 } 794 795 bool post_loop_opts_phase() { return _post_loop_opts_phase; } 796 void set_post_loop_opts_phase() { _post_loop_opts_phase = true; } 797 void reset_post_loop_opts_phase() { _post_loop_opts_phase = false; } 798 799 void record_for_post_loop_opts_igvn(Node* n); 800 void remove_from_post_loop_opts_igvn(Node* n); 801 void process_for_post_loop_opts_igvn(PhaseIterGVN& igvn); 802 803 void record_unstable_if_trap(UnstableIfTrap* trap); 804 bool remove_unstable_if_trap(CallStaticJavaNode* unc, bool yield); 805 void remove_useless_unstable_if_traps(Unique_Node_List &useful); 806 void process_for_unstable_if_traps(PhaseIterGVN& igvn); 807 808 void sort_macro_nodes(); 809 810 void mark_parse_predicate_nodes_useless(PhaseIterGVN& igvn); 811 812 // Are there candidate expensive nodes for optimization? 813 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn); 814 // Check whether n1 and n2 are similar 815 static int cmp_expensive_nodes(Node* n1, Node* n2); 816 // Sort expensive nodes to locate similar expensive nodes 817 void sort_expensive_nodes(); 818 819 // Compilation environment. 820 Arena* comp_arena() { return &_comp_arena; } 821 Arena* parser_arena() { return comp_arena(); } // for the time being, just return comp_arena. 822 ciEnv* env() const { return _env; } 823 CompileLog* log() const { return _log; } 824 bool failing() const { return _env->failing() || _failure_reason != nullptr; } 825 const char* failure_reason() const { return (_env->failing()) ? _env->failure_reason() : _failure_reason; } 826 const CompilationFailureInfo* first_failure_details() const { return _first_failure_details; } 827 828 bool failure_reason_is(const char* r) const { 829 return (r == _failure_reason) || (r != nullptr && _failure_reason != nullptr && strcmp(r, _failure_reason) == 0); 830 } 831 832 void record_failure(const char* reason); 833 void record_method_not_compilable(const char* reason) { 834 env()->record_method_not_compilable(reason); 835 // Record failure reason. 836 record_failure(reason); 837 } 838 bool check_node_count(uint margin, const char* reason) { 839 if (oom()) { 840 record_method_not_compilable_oom(); 841 return true; 842 } 843 if (live_nodes() + margin > max_node_limit()) { 844 record_method_not_compilable(reason); 845 return true; 846 } else { 847 return false; 848 } 849 } 850 bool oom() const { return _oom; } 851 void set_oom() { _oom = true; } 852 853 // Node management 854 uint unique() const { return _unique; } 855 uint next_unique() { return _unique++; } 856 void set_unique(uint i) { _unique = i; } 857 Arena* node_arena() { return _node_arena; } 858 Arena* old_arena() { return (&_node_arena_one == _node_arena) ? &_node_arena_two : &_node_arena_one; } 859 RootNode* root() const { return _root; } 860 void set_root(RootNode* r) { _root = r; } 861 StartNode* start() const; // (Derived from root.) 862 void init_start(StartNode* s); 863 Node* immutable_memory(); 864 865 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; } 866 Node* recent_alloc_obj() const { return _recent_alloc_obj; } 867 void set_recent_alloc(Node* ctl, Node* obj) { 868 _recent_alloc_ctl = ctl; 869 _recent_alloc_obj = obj; 870 } 871 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return; 872 _dead_node_count++; 873 } 874 void reset_dead_node_list() { _dead_node_list.reset(); 875 _dead_node_count = 0; 876 } 877 uint live_nodes() const { 878 int val = _unique - _dead_node_count; 879 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count); 880 return (uint) val; 881 } 882 #ifdef ASSERT 883 void set_phase_optimize_finished() { _phase_optimize_finished = true; } 884 bool phase_optimize_finished() const { return _phase_optimize_finished; } 885 uint count_live_nodes_by_graph_walk(); 886 void print_missing_nodes(); 887 #endif 888 889 // Record modified nodes to check that they are put on IGVN worklist 890 void record_modified_node(Node* n) NOT_DEBUG_RETURN; 891 void remove_modified_node(Node* n) NOT_DEBUG_RETURN; 892 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } ) 893 894 MachConstantBaseNode* mach_constant_base_node(); 895 bool has_mach_constant_base_node() const { return _mach_constant_base_node != nullptr; } 896 // Generated by adlc, true if CallNode requires MachConstantBase. 897 bool needs_deep_clone_jvms(); 898 899 // Handy undefined Node 900 Node* top() const { return _top; } 901 902 // these are used by guys who need to know about creation and transformation of top: 903 Node* cached_top_node() { return _top; } 904 void set_cached_top_node(Node* tn); 905 906 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; } 907 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; } 908 Node_Notes* default_node_notes() const { return _default_node_notes; } 909 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; } 910 911 Node_Notes* node_notes_at(int idx) { 912 return locate_node_notes(_node_note_array, idx, false); 913 } 914 inline bool set_node_notes_at(int idx, Node_Notes* value); 915 916 // Copy notes from source to dest, if they exist. 917 // Overwrite dest only if source provides something. 918 // Return true if information was moved. 919 bool copy_node_notes_to(Node* dest, Node* source); 920 921 // Workhorse function to sort out the blocked Node_Notes array: 922 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr, 923 int idx, bool can_grow = false); 924 925 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by); 926 927 // Type management 928 Arena* type_arena() { return _type_arena; } 929 Dict* type_dict() { return _type_dict; } 930 size_t type_last_size() { return _type_last_size; } 931 int num_alias_types() { return _num_alias_types; } 932 933 void init_type_arena() { _type_arena = &_Compile_types; } 934 void set_type_arena(Arena* a) { _type_arena = a; } 935 void set_type_dict(Dict* d) { _type_dict = d; } 936 void set_type_last_size(size_t sz) { _type_last_size = sz; } 937 938 const TypeFunc* last_tf(ciMethod* m) { 939 return (m == _last_tf_m) ? _last_tf : nullptr; 940 } 941 void set_last_tf(ciMethod* m, const TypeFunc* tf) { 942 assert(m != nullptr || tf == nullptr, ""); 943 _last_tf_m = m; 944 _last_tf = tf; 945 } 946 947 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; } 948 AliasType* alias_type(const TypePtr* adr_type, ciField* field = nullptr) { return find_alias_type(adr_type, false, field); } 949 bool have_alias_type(const TypePtr* adr_type); 950 AliasType* alias_type(ciField* field); 951 952 int get_alias_index(const TypePtr* at) { return alias_type(at)->index(); } 953 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); } 954 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); } 955 956 // Building nodes 957 void rethrow_exceptions(JVMState* jvms); 958 void return_values(JVMState* jvms); 959 JVMState* build_start_state(StartNode* start, const TypeFunc* tf); 960 961 // Decide how to build a call. 962 // The profile factor is a discount to apply to this site's interp. profile. 963 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch, 964 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = nullptr, 965 bool allow_intrinsics = true); 966 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) { 967 return should_delay_string_inlining(call_method, jvms) || 968 should_delay_boxing_inlining(call_method, jvms) || 969 should_delay_vector_inlining(call_method, jvms); 970 } 971 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms); 972 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms); 973 bool should_delay_vector_inlining(ciMethod* call_method, JVMState* jvms); 974 bool should_delay_vector_reboxing_inlining(ciMethod* call_method, JVMState* jvms); 975 976 // Helper functions to identify inlining potential at call-site 977 ciMethod* optimize_virtual_call(ciMethod* caller, ciInstanceKlass* klass, 978 ciKlass* holder, ciMethod* callee, 979 const TypeOopPtr* receiver_type, bool is_virtual, 980 bool &call_does_dispatch, int &vtable_index, 981 bool check_access = true); 982 ciMethod* optimize_inlining(ciMethod* caller, ciInstanceKlass* klass, ciKlass* holder, 983 ciMethod* callee, const TypeOopPtr* receiver_type, 984 bool check_access = true); 985 986 // Report if there were too many traps at a current method and bci. 987 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded. 988 // If there is no MDO at all, report no trap unless told to assume it. 989 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason); 990 // This version, unspecific to a particular bci, asks if 991 // PerMethodTrapLimit was exceeded for all inlined methods seen so far. 992 bool too_many_traps(Deoptimization::DeoptReason reason, 993 // Privately used parameter for logging: 994 ciMethodData* logmd = nullptr); 995 // Report if there were too many recompiles at a method and bci. 996 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason); 997 // Report if there were too many traps or recompiles at a method and bci. 998 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) { 999 return too_many_traps(method, bci, reason) || 1000 too_many_recompiles(method, bci, reason); 1001 } 1002 // Return a bitset with the reasons where deoptimization is allowed, 1003 // i.e., where there were not too many uncommon traps. 1004 int _allowed_reasons; 1005 int allowed_deopt_reasons() { return _allowed_reasons; } 1006 void set_allowed_deopt_reasons(); 1007 1008 // Parsing, optimization 1009 PhaseGVN* initial_gvn() { return _initial_gvn; } 1010 Unique_Node_List* igvn_worklist() { 1011 assert(_igvn_worklist != nullptr, "must be created in Compile::Compile"); 1012 return _igvn_worklist; 1013 } 1014 Type_Array* types() { 1015 assert(_types != nullptr, "must be created in Compile::Compile"); 1016 return _types; 1017 } 1018 NodeHash* node_hash() { 1019 assert(_node_hash != nullptr, "must be created in Compile::Compile"); 1020 return _node_hash; 1021 } 1022 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp. 1023 inline void remove_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp. 1024 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; } 1025 1026 // Replace n by nn using initial_gvn, calling hash_delete and 1027 // record_for_igvn as needed. 1028 void gvn_replace_by(Node* n, Node* nn); 1029 1030 1031 void identify_useful_nodes(Unique_Node_List &useful); 1032 void update_dead_node_list(Unique_Node_List &useful); 1033 void disconnect_useless_nodes(Unique_Node_List& useful, Unique_Node_List& worklist); 1034 1035 void remove_useless_node(Node* dead); 1036 1037 // Record this CallGenerator for inlining at the end of parsing. 1038 void add_late_inline(CallGenerator* cg) { 1039 _late_inlines.insert_before(_late_inlines_pos, cg); 1040 _late_inlines_pos++; 1041 } 1042 1043 void prepend_late_inline(CallGenerator* cg) { 1044 _late_inlines.insert_before(0, cg); 1045 } 1046 1047 void add_string_late_inline(CallGenerator* cg) { 1048 _string_late_inlines.push(cg); 1049 } 1050 1051 void add_boxing_late_inline(CallGenerator* cg) { 1052 _boxing_late_inlines.push(cg); 1053 } 1054 1055 void add_vector_reboxing_late_inline(CallGenerator* cg) { 1056 _vector_reboxing_late_inlines.push(cg); 1057 } 1058 1059 template<typename N, ENABLE_IF(std::is_base_of<Node, N>::value)> 1060 void remove_useless_nodes(GrowableArray<N*>& node_list, Unique_Node_List& useful); 1061 1062 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful); 1063 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Node* dead); 1064 1065 void remove_useless_coarsened_locks(Unique_Node_List& useful); 1066 1067 void process_print_inlining(); 1068 void dump_print_inlining(); 1069 1070 bool over_inlining_cutoff() const { 1071 if (!inlining_incrementally()) { 1072 return unique() > (uint)NodeCountInliningCutoff; 1073 } else { 1074 // Give some room for incremental inlining algorithm to "breathe" 1075 // and avoid thrashing when live node count is close to the limit. 1076 // Keep in mind that live_nodes() isn't accurate during inlining until 1077 // dead node elimination step happens (see Compile::inline_incrementally). 1078 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10; 1079 } 1080 } 1081 1082 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; } 1083 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--; } 1084 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; } 1085 1086 bool inline_incrementally_one(); 1087 void inline_incrementally_cleanup(PhaseIterGVN& igvn); 1088 void inline_incrementally(PhaseIterGVN& igvn); 1089 bool should_delay_inlining() { return AlwaysIncrementalInline || (StressIncrementalInlining && (random() % 2) == 0); } 1090 void inline_string_calls(bool parse_time); 1091 void inline_boxing_calls(PhaseIterGVN& igvn); 1092 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode); 1093 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn); 1094 1095 void inline_vector_reboxing_calls(); 1096 bool has_vbox_nodes(); 1097 1098 void process_late_inline_calls_no_inline(PhaseIterGVN& igvn); 1099 1100 // Matching, CFG layout, allocation, code generation 1101 PhaseCFG* cfg() { return _cfg; } 1102 bool has_java_calls() const { return _java_calls > 0; } 1103 int java_calls() const { return _java_calls; } 1104 int inner_loops() const { return _inner_loops; } 1105 Matcher* matcher() { return _matcher; } 1106 PhaseRegAlloc* regalloc() { return _regalloc; } 1107 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; } 1108 Arena* indexSet_arena() { return _indexSet_arena; } 1109 void* indexSet_free_block_list() { return _indexSet_free_block_list; } 1110 DebugInformationRecorder* debug_info() { return env()->debug_info(); } 1111 1112 void update_interpreter_frame_size(int size) { 1113 if (_interpreter_frame_size < size) { 1114 _interpreter_frame_size = size; 1115 } 1116 } 1117 1118 void set_matcher(Matcher* m) { _matcher = m; } 1119 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; } 1120 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; } 1121 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; } 1122 1123 void set_java_calls(int z) { _java_calls = z; } 1124 void set_inner_loops(int z) { _inner_loops = z; } 1125 1126 Dependencies* dependencies() { return env()->dependencies(); } 1127 1128 // Major entry point. Given a Scope, compile the associated method. 1129 // For normal compilations, entry_bci is InvocationEntryBci. For on stack 1130 // replacement, entry_bci indicates the bytecode for which to compile a 1131 // continuation. 1132 Compile(ciEnv* ci_env, ciMethod* target, 1133 int entry_bci, Options options, DirectiveSet* directive); 1134 1135 // Second major entry point. From the TypeFunc signature, generate code 1136 // to pass arguments from the Java calling convention to the C calling 1137 // convention. 1138 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(), 1139 address stub_function, const char *stub_name, 1140 int is_fancy_jump, bool pass_tls, 1141 bool return_pc, DirectiveSet* directive); 1142 1143 ~Compile(); 1144 1145 // Are we compiling a method? 1146 bool has_method() { return method() != nullptr; } 1147 1148 // Maybe print some information about this compile. 1149 void print_compile_messages(); 1150 1151 // Final graph reshaping, a post-pass after the regular optimizer is done. 1152 bool final_graph_reshaping(); 1153 1154 // returns true if adr is completely contained in the given alias category 1155 bool must_alias(const TypePtr* adr, int alias_idx); 1156 1157 // returns true if adr overlaps with the given alias category 1158 bool can_alias(const TypePtr* adr, int alias_idx); 1159 1160 // Stack slots that may be unused by the calling convention but must 1161 // otherwise be preserved. On Intel this includes the return address. 1162 // On PowerPC it includes the 4 words holding the old TOC & LR glue. 1163 uint in_preserve_stack_slots() { 1164 return SharedRuntime::in_preserve_stack_slots(); 1165 } 1166 1167 // "Top of Stack" slots that may be unused by the calling convention but must 1168 // otherwise be preserved. 1169 // On Intel these are not necessary and the value can be zero. 1170 static uint out_preserve_stack_slots() { 1171 return SharedRuntime::out_preserve_stack_slots(); 1172 } 1173 1174 // Number of outgoing stack slots killed above the out_preserve_stack_slots 1175 // for calls to C. Supports the var-args backing area for register parms. 1176 uint varargs_C_out_slots_killed() const; 1177 1178 // Number of Stack Slots consumed by a synchronization entry 1179 int sync_stack_slots() const; 1180 1181 // Compute the name of old_SP. See <arch>.ad for frame layout. 1182 OptoReg::Name compute_old_SP(); 1183 1184 private: 1185 // Phase control: 1186 void Init(bool aliasing); // Prepare for a single compilation 1187 void Optimize(); // Given a graph, optimize it 1188 void Code_Gen(); // Generate code from a graph 1189 1190 // Management of the AliasType table. 1191 void grow_alias_types(); 1192 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type); 1193 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const; 1194 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field); 1195 1196 void verify_top(Node*) const PRODUCT_RETURN; 1197 1198 // Intrinsic setup. 1199 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor 1200 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper 1201 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn 1202 void register_intrinsic(CallGenerator* cg); // update fn 1203 1204 #ifndef PRODUCT 1205 static juint _intrinsic_hist_count[]; 1206 static jubyte _intrinsic_hist_flags[]; 1207 #endif 1208 // Function calls made by the public function final_graph_reshaping. 1209 // No need to be made public as they are not called elsewhere. 1210 void final_graph_reshaping_impl(Node *n, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes); 1211 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop, Unique_Node_List& dead_nodes); 1212 void final_graph_reshaping_walk(Node_Stack& nstack, Node* root, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes); 1213 void eliminate_redundant_card_marks(Node* n); 1214 1215 // Logic cone optimization. 1216 void optimize_logic_cones(PhaseIterGVN &igvn); 1217 void collect_logic_cone_roots(Unique_Node_List& list); 1218 void process_logic_cone_root(PhaseIterGVN &igvn, Node* n, VectorSet& visited); 1219 bool compute_logic_cone(Node* n, Unique_Node_List& partition, Unique_Node_List& inputs); 1220 uint compute_truth_table(Unique_Node_List& partition, Unique_Node_List& inputs); 1221 uint eval_macro_logic_op(uint func, uint op1, uint op2, uint op3); 1222 Node* xform_to_MacroLogicV(PhaseIterGVN &igvn, const TypeVect* vt, Unique_Node_List& partitions, Unique_Node_List& inputs); 1223 void check_no_dead_use() const NOT_DEBUG_RETURN; 1224 1225 public: 1226 1227 // Note: Histogram array size is about 1 Kb. 1228 enum { // flag bits: 1229 _intrinsic_worked = 1, // succeeded at least once 1230 _intrinsic_failed = 2, // tried it but it failed 1231 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps) 1232 _intrinsic_virtual = 8, // was seen in the virtual form (rare) 1233 _intrinsic_both = 16 // was seen in the non-virtual form (usual) 1234 }; 1235 // Update histogram. Return boolean if this is a first-time occurrence. 1236 static bool gather_intrinsic_statistics(vmIntrinsics::ID id, 1237 bool is_virtual, int flags) PRODUCT_RETURN0; 1238 static void print_intrinsic_statistics() PRODUCT_RETURN; 1239 1240 // Graph verification code 1241 // Walk the node list, verifying that there is a one-to-one 1242 // correspondence between Use-Def edges and Def-Use edges 1243 // The option no_dead_code enables stronger checks that the 1244 // graph is strongly connected from root in both directions. 1245 void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN; 1246 1247 // Verify bi-directional correspondence of edges 1248 void verify_bidirectional_edges(Unique_Node_List &visited); 1249 1250 // End-of-run dumps. 1251 static void print_statistics() PRODUCT_RETURN; 1252 1253 // Verify ADLC assumptions during startup 1254 static void adlc_verification() PRODUCT_RETURN; 1255 1256 // Definitions of pd methods 1257 static void pd_compiler2_init(); 1258 1259 // Static parse-time type checking logic for gen_subtype_check: 1260 enum SubTypeCheckResult { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test }; 1261 SubTypeCheckResult static_subtype_check(const TypeKlassPtr* superk, const TypeKlassPtr* subk, bool skip = StressReflectiveCode); 1262 1263 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype, 1264 // Optional control dependency (for example, on range check) 1265 Node* ctrl = nullptr); 1266 1267 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check) 1268 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl, bool carry_dependency = false); 1269 1270 // Auxiliary methods for randomized fuzzing/stressing 1271 int random(); 1272 bool randomized_select(int count); 1273 1274 // supporting clone_map 1275 CloneMap& clone_map(); 1276 void set_clone_map(Dict* d); 1277 1278 bool needs_clinit_barrier(ciField* ik, ciMethod* accessing_method); 1279 bool needs_clinit_barrier(ciMethod* ik, ciMethod* accessing_method); 1280 bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method); 1281 1282 #ifdef IA32 1283 private: 1284 bool _select_24_bit_instr; // We selected an instruction with a 24-bit result 1285 bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results 1286 1287 // Remember if this compilation changes hardware mode to 24-bit precision. 1288 void set_24_bit_selection_and_mode(bool selection, bool mode) { 1289 _select_24_bit_instr = selection; 1290 _in_24_bit_fp_mode = mode; 1291 } 1292 1293 public: 1294 bool select_24_bit_instr() const { return _select_24_bit_instr; } 1295 bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; } 1296 #endif // IA32 1297 #ifdef ASSERT 1298 VerifyMeetResult* _type_verify; 1299 void set_exception_backedge() { _exception_backedge = true; } 1300 bool has_exception_backedge() const { return _exception_backedge; } 1301 #endif 1302 1303 static bool push_thru_add(PhaseGVN* phase, Node* z, const TypeInteger* tz, const TypeInteger*& rx, const TypeInteger*& ry, 1304 BasicType out_bt, BasicType in_bt); 1305 1306 static Node* narrow_value(BasicType bt, Node* value, const Type* type, PhaseGVN* phase, bool transform_res); 1307 }; 1308 1309 #endif // SHARE_OPTO_COMPILE_HPP