1 /* 2 * Copyright (c) 1997, 2023, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #ifndef SHARE_OPTO_COMPILE_HPP 26 #define SHARE_OPTO_COMPILE_HPP 27 28 #include "asm/codeBuffer.hpp" 29 #include "ci/compilerInterface.hpp" 30 #include "code/debugInfoRec.hpp" 31 #include "compiler/compiler_globals.hpp" 32 #include "compiler/compilerOracle.hpp" 33 #include "compiler/compileBroker.hpp" 34 #include "compiler/compilerEvent.hpp" 35 #include "libadt/dict.hpp" 36 #include "libadt/vectset.hpp" 37 #include "memory/resourceArea.hpp" 38 #include "oops/methodData.hpp" 39 #include "opto/idealGraphPrinter.hpp" 40 #include "opto/phasetype.hpp" 41 #include "opto/phase.hpp" 42 #include "opto/regmask.hpp" 43 #include "runtime/deoptimization.hpp" 44 #include "runtime/sharedRuntime.hpp" 45 #include "runtime/timerTrace.hpp" 46 #include "runtime/vmThread.hpp" 47 #include "utilities/ticks.hpp" 48 49 class AbstractLockNode; 50 class AddPNode; 51 class Block; 52 class Bundle; 53 class CallGenerator; 54 class CallStaticJavaNode; 55 class CloneMap; 56 class CompilationFailureInfo; 57 class ConnectionGraph; 58 class IdealGraphPrinter; 59 class InlineTree; 60 class Matcher; 61 class MachConstantNode; 62 class MachConstantBaseNode; 63 class MachNode; 64 class MachOper; 65 class MachSafePointNode; 66 class Node; 67 class Node_Array; 68 class Node_List; 69 class Node_Notes; 70 class NodeHash; 71 class NodeCloneInfo; 72 class OptoReg; 73 class ParsePredicateNode; 74 class PhaseCFG; 75 class PhaseGVN; 76 class PhaseIterGVN; 77 class PhaseRegAlloc; 78 class PhaseCCP; 79 class PhaseOutput; 80 class RootNode; 81 class relocInfo; 82 class StartNode; 83 class SafePointNode; 84 class JVMState; 85 class Type; 86 class TypeInt; 87 class TypeInteger; 88 class TypeKlassPtr; 89 class TypePtr; 90 class TypeOopPtr; 91 class TypeFunc; 92 class TypeVect; 93 class Type_Array; 94 class Unique_Node_List; 95 class UnstableIfTrap; 96 class nmethod; 97 class Node_Stack; 98 struct Final_Reshape_Counts; 99 class VerifyMeetResult; 100 101 enum LoopOptsMode { 102 LoopOptsDefault, 103 LoopOptsNone, 104 LoopOptsMaxUnroll, 105 LoopOptsShenandoahExpand, 106 LoopOptsShenandoahPostExpand, 107 LoopOptsSkipSplitIf, 108 LoopOptsVerify 109 }; 110 111 // The type of all node counts and indexes. 112 // It must hold at least 16 bits, but must also be fast to load and store. 113 // This type, if less than 32 bits, could limit the number of possible nodes. 114 // (To make this type platform-specific, move to globalDefinitions_xxx.hpp.) 115 typedef unsigned int node_idx_t; 116 117 class NodeCloneInfo { 118 private: 119 uint64_t _idx_clone_orig; 120 public: 121 122 void set_idx(node_idx_t idx) { 123 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx; 124 } 125 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); } 126 127 void set_gen(int generation) { 128 uint64_t g = (uint64_t)generation << 32; 129 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g; 130 } 131 int gen() const { return (int)(_idx_clone_orig >> 32); } 132 133 void set(uint64_t x) { _idx_clone_orig = x; } 134 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); } 135 uint64_t get() const { return _idx_clone_orig; } 136 137 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {} 138 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); } 139 140 void dump_on(outputStream* st) const; 141 }; 142 143 class CloneMap { 144 friend class Compile; 145 private: 146 bool _debug; 147 Dict* _dict; 148 int _clone_idx; // current cloning iteration/generation in loop unroll 149 public: 150 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy 151 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; } 152 Dict* dict() const { return _dict; } 153 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == nullptr, "key existed"); _dict->Insert(_2p(key), (void*)val); } 154 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); } 155 void remove(node_idx_t key) { _dict->Delete(_2p(key)); } 156 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); } 157 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); } 158 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); } 159 int gen(const void* k) const { return gen(_2_node_idx_t(k)); } 160 int max_gen() const; 161 void clone(Node* old, Node* nnn, int gen); 162 void verify_insert_and_clone(Node* old, Node* nnn, int gen); 163 void dump(node_idx_t key, outputStream* st) const; 164 165 int clone_idx() const { return _clone_idx; } 166 void set_clone_idx(int x) { _clone_idx = x; } 167 bool is_debug() const { return _debug; } 168 void set_debug(bool debug) { _debug = debug; } 169 170 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); } 171 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); } 172 }; 173 174 class Options { 175 friend class Compile; 176 friend class VMStructs; 177 private: 178 const bool _subsume_loads; // Load can be matched as part of a larger op. 179 const bool _do_escape_analysis; // Do escape analysis. 180 const bool _do_iterative_escape_analysis; // Do iterative escape analysis. 181 const bool _do_reduce_allocation_merges; // Do try to reduce allocation merges. 182 const bool _eliminate_boxing; // Do boxing elimination. 183 const bool _do_locks_coarsening; // Do locks coarsening 184 const bool _do_superword; // Do SuperWord 185 const bool _install_code; // Install the code that was compiled 186 public: 187 Options(bool subsume_loads, 188 bool do_escape_analysis, 189 bool do_iterative_escape_analysis, 190 bool do_reduce_allocation_merges, 191 bool eliminate_boxing, 192 bool do_locks_coarsening, 193 bool do_superword, 194 bool install_code) : 195 _subsume_loads(subsume_loads), 196 _do_escape_analysis(do_escape_analysis), 197 _do_iterative_escape_analysis(do_iterative_escape_analysis), 198 _do_reduce_allocation_merges(do_reduce_allocation_merges), 199 _eliminate_boxing(eliminate_boxing), 200 _do_locks_coarsening(do_locks_coarsening), 201 _do_superword(do_superword), 202 _install_code(install_code) { 203 } 204 205 static Options for_runtime_stub() { 206 return Options( 207 /* subsume_loads = */ true, 208 /* do_escape_analysis = */ false, 209 /* do_iterative_escape_analysis = */ false, 210 /* do_reduce_allocation_merges = */ false, 211 /* eliminate_boxing = */ false, 212 /* do_lock_coarsening = */ false, 213 /* do_superword = */ true, 214 /* install_code = */ true 215 ); 216 } 217 }; 218 219 //------------------------------Compile---------------------------------------- 220 // This class defines a top-level Compiler invocation. 221 222 class Compile : public Phase { 223 friend class VMStructs; 224 225 public: 226 // Fixed alias indexes. (See also MergeMemNode.) 227 enum { 228 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value) 229 AliasIdxBot = 2, // pseudo-index, aliases to everything 230 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM 231 }; 232 233 // Variant of TraceTime(nullptr, &_t_accumulator, CITime); 234 // Integrated with logging. If logging is turned on, and CITimeVerbose is true, 235 // then brackets are put into the log, with time stamps and node counts. 236 // (The time collection itself is always conditionalized on CITime.) 237 class TracePhase : public TraceTime { 238 private: 239 Compile* _compile; 240 CompileLog* _log; 241 const char* _phase_name; 242 bool _dolog; 243 public: 244 TracePhase(const char* name, elapsedTimer* accumulator); 245 ~TracePhase(); 246 }; 247 248 // Information per category of alias (memory slice) 249 class AliasType { 250 private: 251 friend class Compile; 252 253 int _index; // unique index, used with MergeMemNode 254 const TypePtr* _adr_type; // normalized address type 255 ciField* _field; // relevant instance field, or null if none 256 const Type* _element; // relevant array element type, or null if none 257 bool _is_rewritable; // false if the memory is write-once only 258 int _general_index; // if this is type is an instance, the general 259 // type that this is an instance of 260 261 void Init(int i, const TypePtr* at); 262 263 public: 264 int index() const { return _index; } 265 const TypePtr* adr_type() const { return _adr_type; } 266 ciField* field() const { return _field; } 267 const Type* element() const { return _element; } 268 bool is_rewritable() const { return _is_rewritable; } 269 bool is_volatile() const { return (_field ? _field->is_volatile() : false); } 270 int general_index() const { return (_general_index != 0) ? _general_index : _index; } 271 272 void set_rewritable(bool z) { _is_rewritable = z; } 273 void set_field(ciField* f) { 274 assert(!_field,""); 275 _field = f; 276 if (f->is_final() || f->is_stable()) { 277 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops. 278 _is_rewritable = false; 279 } 280 } 281 void set_element(const Type* e) { 282 assert(_element == nullptr, ""); 283 _element = e; 284 } 285 286 BasicType basic_type() const; 287 288 void print_on(outputStream* st) PRODUCT_RETURN; 289 }; 290 291 enum { 292 logAliasCacheSize = 6, 293 AliasCacheSize = (1<<logAliasCacheSize) 294 }; 295 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type 296 enum { 297 trapHistLength = MethodData::_trap_hist_limit 298 }; 299 300 private: 301 // Fixed parameters to this compilation. 302 const int _compile_id; 303 const Options _options; // Compilation options 304 ciMethod* _method; // The method being compiled. 305 int _entry_bci; // entry bci for osr methods. 306 const TypeFunc* _tf; // My kind of signature 307 InlineTree* _ilt; // Ditto (temporary). 308 address _stub_function; // VM entry for stub being compiled, or null 309 const char* _stub_name; // Name of stub or adapter being compiled, or null 310 address _stub_entry_point; // Compile code entry for generated stub, or null 311 312 // Control of this compilation. 313 int _max_inline_size; // Max inline size for this compilation 314 int _freq_inline_size; // Max hot method inline size for this compilation 315 int _fixed_slots; // count of frame slots not allocated by the register 316 // allocator i.e. locks, original deopt pc, etc. 317 uintx _max_node_limit; // Max unique node count during a single compilation. 318 319 bool _post_loop_opts_phase; // Loop opts are finished. 320 321 int _major_progress; // Count of something big happening 322 bool _inlining_progress; // progress doing incremental inlining? 323 bool _inlining_incrementally;// Are we doing incremental inlining (post parse) 324 bool _do_cleanup; // Cleanup is needed before proceeding with incremental inlining 325 bool _has_loops; // True if the method _may_ have some loops 326 bool _has_split_ifs; // True if the method _may_ have some split-if 327 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores. 328 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated 329 bool _has_boxed_value; // True if a boxed object is allocated 330 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess 331 uint _max_vector_size; // Maximum size of generated vectors 332 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper 333 uint _trap_hist[trapHistLength]; // Cumulative traps 334 bool _trap_can_recompile; // Have we emitted a recompiling trap? 335 uint _decompile_count; // Cumulative decompilation counts. 336 bool _do_inlining; // True if we intend to do inlining 337 bool _do_scheduling; // True if we intend to do scheduling 338 bool _do_freq_based_layout; // True if we intend to do frequency based block layout 339 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations 340 bool _use_cmove; // True if CMove should be used without profitability analysis 341 bool _do_aliasing; // True if we intend to do aliasing 342 bool _print_assembly; // True if we should dump assembly code for this compilation 343 bool _print_inlining; // True if we should print inlining for this compilation 344 bool _print_intrinsics; // True if we should print intrinsics for this compilation 345 #ifndef PRODUCT 346 uint _igv_idx; // Counter for IGV node identifiers 347 uint _igv_phase_iter[PHASE_NUM_TYPES]; // Counters for IGV phase iterations 348 bool _trace_opto_output; 349 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing 350 #endif 351 bool _has_irreducible_loop; // Found irreducible loops 352 // JSR 292 353 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes. 354 bool _has_monitors; // Metadata transfered to nmethod to enable Continuations lock-detection fastpath 355 bool _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry 356 RTMState _rtm_state; // State of Restricted Transactional Memory usage 357 int _loop_opts_cnt; // loop opts round 358 uint _stress_seed; // Seed for stress testing 359 360 // Compilation environment. 361 Arena _comp_arena; // Arena with lifetime equivalent to Compile 362 void* _barrier_set_state; // Potential GC barrier state for Compile 363 ciEnv* _env; // CI interface 364 DirectiveSet* _directive; // Compiler directive 365 CompileLog* _log; // from CompilerThread 366 const char* _failure_reason; // for record_failure/failing pattern 367 CompilationFailureInfo* _first_failure_details; // Details for the first failure happening during compilation 368 GrowableArray<CallGenerator*> _intrinsics; // List of intrinsics. 369 GrowableArray<Node*> _macro_nodes; // List of nodes which need to be expanded before matching. 370 GrowableArray<ParsePredicateNode*> _parse_predicates; // List of Parse Predicates. 371 GrowableArray<Node*> _template_assertion_predicate_opaqs; // List of Opaque4 nodes for Template Assertion Predicates. 372 GrowableArray<Node*> _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common 373 GrowableArray<Node*> _for_post_loop_igvn; // List of nodes for IGVN after loop opts are over 374 GrowableArray<UnstableIfTrap*> _unstable_if_traps; // List of ifnodes after IGVN 375 GrowableArray<Node_List*> _coarsened_locks; // List of coarsened Lock and Unlock nodes 376 ConnectionGraph* _congraph; 377 #ifndef PRODUCT 378 IdealGraphPrinter* _igv_printer; 379 static IdealGraphPrinter* _debug_file_printer; 380 static IdealGraphPrinter* _debug_network_printer; 381 #endif 382 383 384 // Node management 385 uint _unique; // Counter for unique Node indices 386 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N). 387 // So use this to keep count and make the call O(1). 388 VectorSet _dead_node_list; // Set of dead nodes 389 DEBUG_ONLY(Unique_Node_List* _modified_nodes;) // List of nodes which inputs were modified 390 DEBUG_ONLY(bool _phase_optimize_finished;) // Used for live node verification while creating new nodes 391 392 // Arenas for new-space and old-space nodes. 393 // Swapped between using _node_arena. 394 // The lifetime of the old-space nodes is during xform. 395 Arena _node_arena_one; 396 Arena _node_arena_two; 397 Arena* _node_arena; 398 public: 399 Arena* swap_old_and_new() { 400 Arena* filled_arena_ptr = _node_arena; 401 Arena* old_arena_ptr = old_arena(); 402 old_arena_ptr->destruct_contents(); 403 _node_arena = old_arena_ptr; 404 return filled_arena_ptr; 405 } 406 private: 407 RootNode* _root; // Unique root of compilation, or null after bail-out. 408 Node* _top; // Unique top node. (Reset by various phases.) 409 410 Node* _immutable_memory; // Initial memory state 411 412 Node* _recent_alloc_obj; 413 Node* _recent_alloc_ctl; 414 415 // Constant table 416 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton. 417 418 419 // Blocked array of debugging and profiling information, 420 // tracked per node. 421 enum { _log2_node_notes_block_size = 8, 422 _node_notes_block_size = (1<<_log2_node_notes_block_size) 423 }; 424 GrowableArray<Node_Notes*>* _node_note_array; 425 Node_Notes* _default_node_notes; // default notes for new nodes 426 427 // After parsing and every bulk phase we hang onto the Root instruction. 428 // The RootNode instruction is where the whole program begins. It produces 429 // the initial Control and BOTTOM for everybody else. 430 431 // Type management 432 Arena _Compile_types; // Arena for all types 433 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared() 434 Dict* _type_dict; // Intern table 435 CloneMap _clone_map; // used for recording history of cloned nodes 436 size_t _type_last_size; // Last allocation size (see Type::operator new/delete) 437 ciMethod* _last_tf_m; // Cache for 438 const TypeFunc* _last_tf; // TypeFunc::make 439 AliasType** _alias_types; // List of alias types seen so far. 440 int _num_alias_types; // Logical length of _alias_types 441 int _max_alias_types; // Physical length of _alias_types 442 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking 443 444 // Parsing, optimization 445 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN 446 447 // Shared worklist for all IGVN rounds. Nodes can be pushed to it at any time. 448 // If pushed outside IGVN, the Node is processed in the next IGVN round. 449 Unique_Node_List* _igvn_worklist; 450 451 // Shared type array for GVN, IGVN and CCP. It maps node idx -> Type*. 452 Type_Array* _types; 453 454 // Shared node hash table for GVN, IGVN and CCP. 455 NodeHash* _node_hash; 456 457 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after main parsing has finished. 458 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations 459 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations 460 461 GrowableArray<CallGenerator*> _vector_reboxing_late_inlines; // same but for vector reboxing operations 462 463 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining) 464 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending 465 466 // "MemLimit" directive was specified and the memory limit was hit during compilation 467 bool _oom; 468 469 // Inlining may not happen in parse order which would make 470 // PrintInlining output confusing. Keep track of PrintInlining 471 // pieces in order. 472 class PrintInliningBuffer : public CHeapObj<mtCompiler> { 473 private: 474 CallGenerator* _cg; 475 stringStream _ss; 476 static const size_t default_stream_buffer_size = 128; 477 478 public: 479 PrintInliningBuffer() 480 : _cg(nullptr), _ss(default_stream_buffer_size) {} 481 482 stringStream* ss() { return &_ss; } 483 CallGenerator* cg() { return _cg; } 484 void set_cg(CallGenerator* cg) { _cg = cg; } 485 }; 486 487 stringStream* _print_inlining_stream; 488 GrowableArray<PrintInliningBuffer*>* _print_inlining_list; 489 int _print_inlining_idx; 490 char* _print_inlining_output; 491 492 // Only keep nodes in the expensive node list that need to be optimized 493 void cleanup_expensive_nodes(PhaseIterGVN &igvn); 494 // Use for sorting expensive nodes to bring similar nodes together 495 static int cmp_expensive_nodes(Node** n1, Node** n2); 496 // Expensive nodes list already sorted? 497 bool expensive_nodes_sorted() const; 498 // Remove the speculative part of types and clean up the graph 499 void remove_speculative_types(PhaseIterGVN &igvn); 500 501 void* _replay_inline_data; // Pointer to data loaded from file 502 503 void print_inlining_init(); 504 void print_inlining_reinit(); 505 void print_inlining_commit(); 506 void print_inlining_push(); 507 PrintInliningBuffer* print_inlining_current(); 508 509 void log_late_inline_failure(CallGenerator* cg, const char* msg); 510 DEBUG_ONLY(bool _exception_backedge;) 511 512 void record_method_not_compilable_oom(); 513 514 public: 515 516 void* barrier_set_state() const { return _barrier_set_state; } 517 518 stringStream* print_inlining_stream() { 519 assert(print_inlining() || print_intrinsics(), "PrintInlining off?"); 520 return _print_inlining_stream; 521 } 522 523 void print_inlining_update(CallGenerator* cg); 524 void print_inlining_update_delayed(CallGenerator* cg); 525 void print_inlining_move_to(CallGenerator* cg); 526 void print_inlining_assert_ready(); 527 void print_inlining_reset(); 528 529 void print_inlining(ciMethod* method, int inline_level, int bci, InliningResult result, const char* msg = nullptr) { 530 stringStream ss; 531 CompileTask::print_inlining_inner(&ss, method, inline_level, bci, result, msg); 532 print_inlining_stream()->print("%s", ss.freeze()); 533 } 534 535 #ifndef PRODUCT 536 IdealGraphPrinter* igv_printer() { return _igv_printer; } 537 void reset_igv_phase_iter(CompilerPhaseType cpt) { _igv_phase_iter[cpt] = 0; } 538 #endif 539 540 void log_late_inline(CallGenerator* cg); 541 void log_inline_id(CallGenerator* cg); 542 void log_inline_failure(const char* msg); 543 544 void* replay_inline_data() const { return _replay_inline_data; } 545 546 // Dump inlining replay data to the stream. 547 void dump_inline_data(outputStream* out); 548 void dump_inline_data_reduced(outputStream* out); 549 550 private: 551 // Matching, CFG layout, allocation, code generation 552 PhaseCFG* _cfg; // Results of CFG finding 553 int _java_calls; // Number of java calls in the method 554 int _inner_loops; // Number of inner loops in the method 555 Matcher* _matcher; // Engine to map ideal to machine instructions 556 PhaseRegAlloc* _regalloc; // Results of register allocation. 557 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout) 558 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin 559 void* _indexSet_free_block_list; // free list of IndexSet bit blocks 560 int _interpreter_frame_size; 561 562 PhaseOutput* _output; 563 564 public: 565 // Accessors 566 567 // The Compile instance currently active in this (compiler) thread. 568 static Compile* current() { 569 return (Compile*) ciEnv::current()->compiler_data(); 570 } 571 572 int interpreter_frame_size() const { return _interpreter_frame_size; } 573 574 PhaseOutput* output() const { return _output; } 575 void set_output(PhaseOutput* o) { _output = o; } 576 577 // ID for this compilation. Useful for setting breakpoints in the debugger. 578 int compile_id() const { return _compile_id; } 579 DirectiveSet* directive() const { return _directive; } 580 581 // Does this compilation allow instructions to subsume loads? User 582 // instructions that subsume a load may result in an unschedulable 583 // instruction sequence. 584 bool subsume_loads() const { return _options._subsume_loads; } 585 /** Do escape analysis. */ 586 bool do_escape_analysis() const { return _options._do_escape_analysis; } 587 bool do_iterative_escape_analysis() const { return _options._do_iterative_escape_analysis; } 588 bool do_reduce_allocation_merges() const { return _options._do_reduce_allocation_merges; } 589 /** Do boxing elimination. */ 590 bool eliminate_boxing() const { return _options._eliminate_boxing; } 591 /** Do aggressive boxing elimination. */ 592 bool aggressive_unboxing() const { return _options._eliminate_boxing && AggressiveUnboxing; } 593 bool should_install_code() const { return _options._install_code; } 594 /** Do locks coarsening. */ 595 bool do_locks_coarsening() const { return _options._do_locks_coarsening; } 596 bool do_superword() const { return _options._do_superword; } 597 598 // Other fixed compilation parameters. 599 ciMethod* method() const { return _method; } 600 int entry_bci() const { return _entry_bci; } 601 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; } 602 bool is_method_compilation() const { return (_method != nullptr && !_method->flags().is_native()); } 603 const TypeFunc* tf() const { assert(_tf!=nullptr, ""); return _tf; } 604 void init_tf(const TypeFunc* tf) { assert(_tf==nullptr, ""); _tf = tf; } 605 InlineTree* ilt() const { return _ilt; } 606 address stub_function() const { return _stub_function; } 607 const char* stub_name() const { return _stub_name; } 608 address stub_entry_point() const { return _stub_entry_point; } 609 void set_stub_entry_point(address z) { _stub_entry_point = z; } 610 611 // Control of this compilation. 612 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; } 613 void set_fixed_slots(int n) { _fixed_slots = n; } 614 int major_progress() const { return _major_progress; } 615 void set_inlining_progress(bool z) { _inlining_progress = z; } 616 int inlining_progress() const { return _inlining_progress; } 617 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; } 618 int inlining_incrementally() const { return _inlining_incrementally; } 619 void set_do_cleanup(bool z) { _do_cleanup = z; } 620 int do_cleanup() const { return _do_cleanup; } 621 void set_major_progress() { _major_progress++; } 622 void restore_major_progress(int progress) { _major_progress += progress; } 623 void clear_major_progress() { _major_progress = 0; } 624 int max_inline_size() const { return _max_inline_size; } 625 void set_freq_inline_size(int n) { _freq_inline_size = n; } 626 int freq_inline_size() const { return _freq_inline_size; } 627 void set_max_inline_size(int n) { _max_inline_size = n; } 628 bool has_loops() const { return _has_loops; } 629 void set_has_loops(bool z) { _has_loops = z; } 630 bool has_split_ifs() const { return _has_split_ifs; } 631 void set_has_split_ifs(bool z) { _has_split_ifs = z; } 632 bool has_unsafe_access() const { return _has_unsafe_access; } 633 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; } 634 bool has_stringbuilder() const { return _has_stringbuilder; } 635 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; } 636 bool has_boxed_value() const { return _has_boxed_value; } 637 void set_has_boxed_value(bool z) { _has_boxed_value = z; } 638 bool has_reserved_stack_access() const { return _has_reserved_stack_access; } 639 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; } 640 uint max_vector_size() const { return _max_vector_size; } 641 void set_max_vector_size(uint s) { _max_vector_size = s; } 642 bool clear_upper_avx() const { return _clear_upper_avx; } 643 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; } 644 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; } 645 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; } 646 bool trap_can_recompile() const { return _trap_can_recompile; } 647 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; } 648 uint decompile_count() const { return _decompile_count; } 649 void set_decompile_count(uint c) { _decompile_count = c; } 650 bool allow_range_check_smearing() const; 651 bool do_inlining() const { return _do_inlining; } 652 void set_do_inlining(bool z) { _do_inlining = z; } 653 bool do_scheduling() const { return _do_scheduling; } 654 void set_do_scheduling(bool z) { _do_scheduling = z; } 655 bool do_freq_based_layout() const{ return _do_freq_based_layout; } 656 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; } 657 bool do_vector_loop() const { return _do_vector_loop; } 658 void set_do_vector_loop(bool z) { _do_vector_loop = z; } 659 bool use_cmove() const { return _use_cmove; } 660 void set_use_cmove(bool z) { _use_cmove = z; } 661 bool do_aliasing() const { return _do_aliasing; } 662 bool print_assembly() const { return _print_assembly; } 663 void set_print_assembly(bool z) { _print_assembly = z; } 664 bool print_inlining() const { return _print_inlining; } 665 void set_print_inlining(bool z) { _print_inlining = z; } 666 bool print_intrinsics() const { return _print_intrinsics; } 667 void set_print_intrinsics(bool z) { _print_intrinsics = z; } 668 RTMState rtm_state() const { return _rtm_state; } 669 void set_rtm_state(RTMState s) { _rtm_state = s; } 670 bool use_rtm() const { return (_rtm_state & NoRTM) == 0; } 671 bool profile_rtm() const { return _rtm_state == ProfileRTM; } 672 uint max_node_limit() const { return (uint)_max_node_limit; } 673 void set_max_node_limit(uint n) { _max_node_limit = n; } 674 bool clinit_barrier_on_entry() { return _clinit_barrier_on_entry; } 675 void set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; } 676 bool has_monitors() const { return _has_monitors; } 677 void set_has_monitors(bool v) { _has_monitors = v; } 678 679 // check the CompilerOracle for special behaviours for this compile 680 bool method_has_option(enum CompileCommand option) { 681 return method() != nullptr && method()->has_option(option); 682 } 683 684 #ifndef PRODUCT 685 uint next_igv_idx() { return _igv_idx++; } 686 bool trace_opto_output() const { return _trace_opto_output; } 687 void print_ideal_ir(const char* phase_name); 688 bool should_print_ideal() const { return _directive->PrintIdealOption; } 689 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; } 690 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; } 691 int _in_dump_cnt; // Required for dumping ir nodes. 692 #endif 693 bool has_irreducible_loop() const { return _has_irreducible_loop; } 694 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; } 695 696 // JSR 292 697 bool has_method_handle_invokes() const { return _has_method_handle_invokes; } 698 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; } 699 700 Ticks _latest_stage_start_counter; 701 702 void begin_method(); 703 void end_method(); 704 bool should_print_igv(int level); 705 bool should_print_phase(CompilerPhaseType cpt); 706 707 void print_method(CompilerPhaseType cpt, int level, Node* n = nullptr); 708 709 #ifndef PRODUCT 710 void dump_igv(const char* graph_name, int level = 3) { 711 if (should_print_igv(level)) { 712 _igv_printer->print_method(graph_name, level); 713 } 714 } 715 716 void igv_print_method_to_file(const char* phase_name = "Debug", bool append = false); 717 void igv_print_method_to_network(const char* phase_name = "Debug"); 718 static IdealGraphPrinter* debug_file_printer() { return _debug_file_printer; } 719 static IdealGraphPrinter* debug_network_printer() { return _debug_network_printer; } 720 #endif 721 722 int macro_count() const { return _macro_nodes.length(); } 723 int parse_predicate_count() const { return _parse_predicates.length(); } 724 int template_assertion_predicate_count() const { return _template_assertion_predicate_opaqs.length(); } 725 int expensive_count() const { return _expensive_nodes.length(); } 726 int coarsened_count() const { return _coarsened_locks.length(); } 727 728 Node* macro_node(int idx) const { return _macro_nodes.at(idx); } 729 ParsePredicateNode* parse_predicate(int idx) const { return _parse_predicates.at(idx); } 730 731 Node* template_assertion_predicate_opaq_node(int idx) const { 732 return _template_assertion_predicate_opaqs.at(idx); 733 } 734 735 Node* expensive_node(int idx) const { return _expensive_nodes.at(idx); } 736 737 ConnectionGraph* congraph() { return _congraph;} 738 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;} 739 void add_macro_node(Node * n) { 740 //assert(n->is_macro(), "must be a macro node"); 741 assert(!_macro_nodes.contains(n), "duplicate entry in expand list"); 742 _macro_nodes.append(n); 743 } 744 void remove_macro_node(Node* n) { 745 // this function may be called twice for a node so we can only remove it 746 // if it's still existing. 747 _macro_nodes.remove_if_existing(n); 748 // Remove from coarsened locks list if present 749 if (coarsened_count() > 0) { 750 remove_coarsened_lock(n); 751 } 752 } 753 void add_expensive_node(Node* n); 754 void remove_expensive_node(Node* n) { 755 _expensive_nodes.remove_if_existing(n); 756 } 757 758 void add_parse_predicate(ParsePredicateNode* n) { 759 assert(!_parse_predicates.contains(n), "duplicate entry in Parse Predicate list"); 760 _parse_predicates.append(n); 761 } 762 763 void remove_parse_predicate(ParsePredicateNode* n) { 764 if (parse_predicate_count() > 0) { 765 _parse_predicates.remove_if_existing(n); 766 } 767 } 768 769 void add_template_assertion_predicate_opaq(Node* n) { 770 assert(!_template_assertion_predicate_opaqs.contains(n), 771 "duplicate entry in template assertion predicate opaque4 list"); 772 _template_assertion_predicate_opaqs.append(n); 773 } 774 775 void remove_template_assertion_predicate_opaq(Node* n) { 776 if (template_assertion_predicate_count() > 0) { 777 _template_assertion_predicate_opaqs.remove_if_existing(n); 778 } 779 } 780 void add_coarsened_locks(GrowableArray<AbstractLockNode*>& locks); 781 void remove_coarsened_lock(Node* n); 782 bool coarsened_locks_consistent(); 783 784 bool post_loop_opts_phase() { return _post_loop_opts_phase; } 785 void set_post_loop_opts_phase() { _post_loop_opts_phase = true; } 786 void reset_post_loop_opts_phase() { _post_loop_opts_phase = false; } 787 788 void record_for_post_loop_opts_igvn(Node* n); 789 void remove_from_post_loop_opts_igvn(Node* n); 790 void process_for_post_loop_opts_igvn(PhaseIterGVN& igvn); 791 792 void record_unstable_if_trap(UnstableIfTrap* trap); 793 bool remove_unstable_if_trap(CallStaticJavaNode* unc, bool yield); 794 void remove_useless_unstable_if_traps(Unique_Node_List &useful); 795 void process_for_unstable_if_traps(PhaseIterGVN& igvn); 796 797 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 bool failing() const { return _env->failing() || _failure_reason != nullptr; } 813 const char* failure_reason() const { return (_env->failing()) ? _env->failure_reason() : _failure_reason; } 814 const CompilationFailureInfo* first_failure_details() const { return _first_failure_details; } 815 816 bool failure_reason_is(const char* r) const { 817 return (r == _failure_reason) || (r != nullptr && _failure_reason != nullptr && strcmp(r, _failure_reason) == 0); 818 } 819 820 void record_failure(const char* reason); 821 void record_method_not_compilable(const char* reason) { 822 env()->record_method_not_compilable(reason); 823 // Record failure reason. 824 record_failure(reason); 825 } 826 bool check_node_count(uint margin, const char* reason) { 827 if (oom()) { 828 record_method_not_compilable_oom(); 829 return true; 830 } 831 if (live_nodes() + margin > max_node_limit()) { 832 record_method_not_compilable(reason); 833 return true; 834 } else { 835 return false; 836 } 837 } 838 bool oom() const { return _oom; } 839 void set_oom() { _oom = true; } 840 841 // Node management 842 uint unique() const { return _unique; } 843 uint next_unique() { return _unique++; } 844 void set_unique(uint i) { _unique = i; } 845 Arena* node_arena() { return _node_arena; } 846 Arena* old_arena() { return (&_node_arena_one == _node_arena) ? &_node_arena_two : &_node_arena_one; } 847 RootNode* root() const { return _root; } 848 void set_root(RootNode* r) { _root = r; } 849 StartNode* start() const; // (Derived from root.) 850 void init_start(StartNode* s); 851 Node* immutable_memory(); 852 853 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; } 854 Node* recent_alloc_obj() const { return _recent_alloc_obj; } 855 void set_recent_alloc(Node* ctl, Node* obj) { 856 _recent_alloc_ctl = ctl; 857 _recent_alloc_obj = obj; 858 } 859 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return; 860 _dead_node_count++; 861 } 862 void reset_dead_node_list() { _dead_node_list.reset(); 863 _dead_node_count = 0; 864 } 865 uint live_nodes() const { 866 int val = _unique - _dead_node_count; 867 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count); 868 return (uint) val; 869 } 870 #ifdef ASSERT 871 void set_phase_optimize_finished() { _phase_optimize_finished = true; } 872 bool phase_optimize_finished() const { return _phase_optimize_finished; } 873 uint count_live_nodes_by_graph_walk(); 874 void print_missing_nodes(); 875 #endif 876 877 // Record modified nodes to check that they are put on IGVN worklist 878 void record_modified_node(Node* n) NOT_DEBUG_RETURN; 879 void remove_modified_node(Node* n) NOT_DEBUG_RETURN; 880 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } ) 881 882 MachConstantBaseNode* mach_constant_base_node(); 883 bool has_mach_constant_base_node() const { return _mach_constant_base_node != nullptr; } 884 // Generated by adlc, true if CallNode requires MachConstantBase. 885 bool needs_deep_clone_jvms(); 886 887 // Handy undefined Node 888 Node* top() const { return _top; } 889 890 // these are used by guys who need to know about creation and transformation of top: 891 Node* cached_top_node() { return _top; } 892 void set_cached_top_node(Node* tn); 893 894 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; } 895 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; } 896 Node_Notes* default_node_notes() const { return _default_node_notes; } 897 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; } 898 899 Node_Notes* node_notes_at(int idx) { 900 return locate_node_notes(_node_note_array, idx, false); 901 } 902 inline bool set_node_notes_at(int idx, Node_Notes* value); 903 904 // Copy notes from source to dest, if they exist. 905 // Overwrite dest only if source provides something. 906 // Return true if information was moved. 907 bool copy_node_notes_to(Node* dest, Node* source); 908 909 // Workhorse function to sort out the blocked Node_Notes array: 910 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr, 911 int idx, bool can_grow = false); 912 913 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by); 914 915 // Type management 916 Arena* type_arena() { return _type_arena; } 917 Dict* type_dict() { return _type_dict; } 918 size_t type_last_size() { return _type_last_size; } 919 int num_alias_types() { return _num_alias_types; } 920 921 void init_type_arena() { _type_arena = &_Compile_types; } 922 void set_type_arena(Arena* a) { _type_arena = a; } 923 void set_type_dict(Dict* d) { _type_dict = d; } 924 void set_type_last_size(size_t sz) { _type_last_size = sz; } 925 926 const TypeFunc* last_tf(ciMethod* m) { 927 return (m == _last_tf_m) ? _last_tf : nullptr; 928 } 929 void set_last_tf(ciMethod* m, const TypeFunc* tf) { 930 assert(m != nullptr || tf == nullptr, ""); 931 _last_tf_m = m; 932 _last_tf = tf; 933 } 934 935 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; } 936 AliasType* alias_type(const TypePtr* adr_type, ciField* field = nullptr) { return find_alias_type(adr_type, false, field); } 937 bool have_alias_type(const TypePtr* adr_type); 938 AliasType* alias_type(ciField* field); 939 940 int get_alias_index(const TypePtr* at) { return alias_type(at)->index(); } 941 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); } 942 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); } 943 944 // Building nodes 945 void rethrow_exceptions(JVMState* jvms); 946 void return_values(JVMState* jvms); 947 JVMState* build_start_state(StartNode* start, const TypeFunc* tf); 948 949 // Decide how to build a call. 950 // The profile factor is a discount to apply to this site's interp. profile. 951 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch, 952 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = nullptr, 953 bool allow_intrinsics = true); 954 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) { 955 return should_delay_string_inlining(call_method, jvms) || 956 should_delay_boxing_inlining(call_method, jvms) || 957 should_delay_vector_inlining(call_method, jvms); 958 } 959 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms); 960 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms); 961 bool should_delay_vector_inlining(ciMethod* call_method, JVMState* jvms); 962 bool should_delay_vector_reboxing_inlining(ciMethod* call_method, JVMState* jvms); 963 964 // Helper functions to identify inlining potential at call-site 965 ciMethod* optimize_virtual_call(ciMethod* caller, ciInstanceKlass* klass, 966 ciKlass* holder, ciMethod* callee, 967 const TypeOopPtr* receiver_type, bool is_virtual, 968 bool &call_does_dispatch, int &vtable_index, 969 bool check_access = true); 970 ciMethod* optimize_inlining(ciMethod* caller, ciInstanceKlass* klass, ciKlass* holder, 971 ciMethod* callee, const TypeOopPtr* receiver_type, 972 bool check_access = true); 973 974 // Report if there were too many traps at a current method and bci. 975 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded. 976 // If there is no MDO at all, report no trap unless told to assume it. 977 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason); 978 // This version, unspecific to a particular bci, asks if 979 // PerMethodTrapLimit was exceeded for all inlined methods seen so far. 980 bool too_many_traps(Deoptimization::DeoptReason reason, 981 // Privately used parameter for logging: 982 ciMethodData* logmd = nullptr); 983 // Report if there were too many recompiles at a method and bci. 984 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason); 985 // Report if there were too many traps or recompiles at a method and bci. 986 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) { 987 return too_many_traps(method, bci, reason) || 988 too_many_recompiles(method, bci, reason); 989 } 990 // Return a bitset with the reasons where deoptimization is allowed, 991 // i.e., where there were not too many uncommon traps. 992 int _allowed_reasons; 993 int allowed_deopt_reasons() { return _allowed_reasons; } 994 void set_allowed_deopt_reasons(); 995 996 // Parsing, optimization 997 PhaseGVN* initial_gvn() { return _initial_gvn; } 998 Unique_Node_List* igvn_worklist() { 999 assert(_igvn_worklist != nullptr, "must be created in Compile::Compile"); 1000 return _igvn_worklist; 1001 } 1002 Type_Array* types() { 1003 assert(_types != nullptr, "must be created in Compile::Compile"); 1004 return _types; 1005 } 1006 NodeHash* node_hash() { 1007 assert(_node_hash != nullptr, "must be created in Compile::Compile"); 1008 return _node_hash; 1009 } 1010 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp. 1011 inline void remove_for_igvn(Node* n); // Body is after class Unique_Node_List in node.hpp. 1012 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; } 1013 1014 // Replace n by nn using initial_gvn, calling hash_delete and 1015 // record_for_igvn as needed. 1016 void gvn_replace_by(Node* n, Node* nn); 1017 1018 1019 void identify_useful_nodes(Unique_Node_List &useful); 1020 void update_dead_node_list(Unique_Node_List &useful); 1021 void disconnect_useless_nodes(Unique_Node_List& useful, Unique_Node_List& worklist); 1022 1023 void remove_useless_node(Node* dead); 1024 1025 // Record this CallGenerator for inlining at the end of parsing. 1026 void add_late_inline(CallGenerator* cg) { 1027 _late_inlines.insert_before(_late_inlines_pos, cg); 1028 _late_inlines_pos++; 1029 } 1030 1031 void prepend_late_inline(CallGenerator* cg) { 1032 _late_inlines.insert_before(0, cg); 1033 } 1034 1035 void add_string_late_inline(CallGenerator* cg) { 1036 _string_late_inlines.push(cg); 1037 } 1038 1039 void add_boxing_late_inline(CallGenerator* cg) { 1040 _boxing_late_inlines.push(cg); 1041 } 1042 1043 void add_vector_reboxing_late_inline(CallGenerator* cg) { 1044 _vector_reboxing_late_inlines.push(cg); 1045 } 1046 1047 template<typename N, ENABLE_IF(std::is_base_of<Node, N>::value)> 1048 void remove_useless_nodes(GrowableArray<N*>& node_list, Unique_Node_List& useful); 1049 1050 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful); 1051 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Node* dead); 1052 1053 void remove_useless_coarsened_locks(Unique_Node_List& useful); 1054 1055 void process_print_inlining(); 1056 void dump_print_inlining(); 1057 1058 bool over_inlining_cutoff() const { 1059 if (!inlining_incrementally()) { 1060 return unique() > (uint)NodeCountInliningCutoff; 1061 } else { 1062 // Give some room for incremental inlining algorithm to "breathe" 1063 // and avoid thrashing when live node count is close to the limit. 1064 // Keep in mind that live_nodes() isn't accurate during inlining until 1065 // dead node elimination step happens (see Compile::inline_incrementally). 1066 return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10; 1067 } 1068 } 1069 1070 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; } 1071 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--; } 1072 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; } 1073 1074 bool inline_incrementally_one(); 1075 void inline_incrementally_cleanup(PhaseIterGVN& igvn); 1076 void inline_incrementally(PhaseIterGVN& igvn); 1077 bool should_delay_inlining() { return AlwaysIncrementalInline || (StressIncrementalInlining && (random() % 2) == 0); } 1078 void inline_string_calls(bool parse_time); 1079 void inline_boxing_calls(PhaseIterGVN& igvn); 1080 bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode); 1081 void remove_root_to_sfpts_edges(PhaseIterGVN& igvn); 1082 1083 void inline_vector_reboxing_calls(); 1084 bool has_vbox_nodes(); 1085 1086 void process_late_inline_calls_no_inline(PhaseIterGVN& igvn); 1087 1088 // Matching, CFG layout, allocation, code generation 1089 PhaseCFG* cfg() { return _cfg; } 1090 bool has_java_calls() const { return _java_calls > 0; } 1091 int java_calls() const { return _java_calls; } 1092 int inner_loops() const { return _inner_loops; } 1093 Matcher* matcher() { return _matcher; } 1094 PhaseRegAlloc* regalloc() { return _regalloc; } 1095 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; } 1096 Arena* indexSet_arena() { return _indexSet_arena; } 1097 void* indexSet_free_block_list() { return _indexSet_free_block_list; } 1098 DebugInformationRecorder* debug_info() { return env()->debug_info(); } 1099 1100 void update_interpreter_frame_size(int size) { 1101 if (_interpreter_frame_size < size) { 1102 _interpreter_frame_size = size; 1103 } 1104 } 1105 1106 void set_matcher(Matcher* m) { _matcher = m; } 1107 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; } 1108 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; } 1109 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; } 1110 1111 void set_java_calls(int z) { _java_calls = z; } 1112 void set_inner_loops(int z) { _inner_loops = z; } 1113 1114 Dependencies* dependencies() { return env()->dependencies(); } 1115 1116 // Major entry point. Given a Scope, compile the associated method. 1117 // For normal compilations, entry_bci is InvocationEntryBci. For on stack 1118 // replacement, entry_bci indicates the bytecode for which to compile a 1119 // continuation. 1120 Compile(ciEnv* ci_env, ciMethod* target, 1121 int entry_bci, Options options, DirectiveSet* directive); 1122 1123 // Second major entry point. From the TypeFunc signature, generate code 1124 // to pass arguments from the Java calling convention to the C calling 1125 // convention. 1126 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(), 1127 address stub_function, const char *stub_name, 1128 int is_fancy_jump, bool pass_tls, 1129 bool return_pc, DirectiveSet* directive); 1130 1131 ~Compile(); 1132 1133 // Are we compiling a method? 1134 bool has_method() { return method() != nullptr; } 1135 1136 // Maybe print some information about this compile. 1137 void print_compile_messages(); 1138 1139 // Final graph reshaping, a post-pass after the regular optimizer is done. 1140 bool final_graph_reshaping(); 1141 1142 // returns true if adr is completely contained in the given alias category 1143 bool must_alias(const TypePtr* adr, int alias_idx); 1144 1145 // returns true if adr overlaps with the given alias category 1146 bool can_alias(const TypePtr* adr, int alias_idx); 1147 1148 // Stack slots that may be unused by the calling convention but must 1149 // otherwise be preserved. On Intel this includes the return address. 1150 // On PowerPC it includes the 4 words holding the old TOC & LR glue. 1151 uint in_preserve_stack_slots() { 1152 return SharedRuntime::in_preserve_stack_slots(); 1153 } 1154 1155 // "Top of Stack" slots that may be unused by the calling convention but must 1156 // otherwise be preserved. 1157 // On Intel these are not necessary and the value can be zero. 1158 static uint out_preserve_stack_slots() { 1159 return SharedRuntime::out_preserve_stack_slots(); 1160 } 1161 1162 // Number of outgoing stack slots killed above the out_preserve_stack_slots 1163 // for calls to C. Supports the var-args backing area for register parms. 1164 uint varargs_C_out_slots_killed() const; 1165 1166 // Number of Stack Slots consumed by a synchronization entry 1167 int sync_stack_slots() const; 1168 1169 // Compute the name of old_SP. See <arch>.ad for frame layout. 1170 OptoReg::Name compute_old_SP(); 1171 1172 private: 1173 // Phase control: 1174 void Init(bool aliasing); // Prepare for a single compilation 1175 void Optimize(); // Given a graph, optimize it 1176 void Code_Gen(); // Generate code from a graph 1177 1178 // Management of the AliasType table. 1179 void grow_alias_types(); 1180 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type); 1181 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const; 1182 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field); 1183 1184 void verify_top(Node*) const PRODUCT_RETURN; 1185 1186 // Intrinsic setup. 1187 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor 1188 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper 1189 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn 1190 void register_intrinsic(CallGenerator* cg); // update fn 1191 1192 #ifndef PRODUCT 1193 static juint _intrinsic_hist_count[]; 1194 static jubyte _intrinsic_hist_flags[]; 1195 #endif 1196 // Function calls made by the public function final_graph_reshaping. 1197 // No need to be made public as they are not called elsewhere. 1198 void final_graph_reshaping_impl(Node *n, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes); 1199 void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop, Unique_Node_List& dead_nodes); 1200 void final_graph_reshaping_walk(Node_Stack& nstack, Node* root, Final_Reshape_Counts& frc, Unique_Node_List& dead_nodes); 1201 void eliminate_redundant_card_marks(Node* n); 1202 1203 // Logic cone optimization. 1204 void optimize_logic_cones(PhaseIterGVN &igvn); 1205 void collect_logic_cone_roots(Unique_Node_List& list); 1206 void process_logic_cone_root(PhaseIterGVN &igvn, Node* n, VectorSet& visited); 1207 bool compute_logic_cone(Node* n, Unique_Node_List& partition, Unique_Node_List& inputs); 1208 uint compute_truth_table(Unique_Node_List& partition, Unique_Node_List& inputs); 1209 uint eval_macro_logic_op(uint func, uint op1, uint op2, uint op3); 1210 Node* xform_to_MacroLogicV(PhaseIterGVN &igvn, const TypeVect* vt, Unique_Node_List& partitions, Unique_Node_List& inputs); 1211 void check_no_dead_use() const NOT_DEBUG_RETURN; 1212 1213 public: 1214 1215 // Note: Histogram array size is about 1 Kb. 1216 enum { // flag bits: 1217 _intrinsic_worked = 1, // succeeded at least once 1218 _intrinsic_failed = 2, // tried it but it failed 1219 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps) 1220 _intrinsic_virtual = 8, // was seen in the virtual form (rare) 1221 _intrinsic_both = 16 // was seen in the non-virtual form (usual) 1222 }; 1223 // Update histogram. Return boolean if this is a first-time occurrence. 1224 static bool gather_intrinsic_statistics(vmIntrinsics::ID id, 1225 bool is_virtual, int flags) PRODUCT_RETURN0; 1226 static void print_intrinsic_statistics() PRODUCT_RETURN; 1227 1228 // Graph verification code 1229 // Walk the node list, verifying that there is a one-to-one 1230 // correspondence between Use-Def edges and Def-Use edges 1231 // The option no_dead_code enables stronger checks that the 1232 // graph is strongly connected from root in both directions. 1233 void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN; 1234 1235 // Verify bi-directional correspondence of edges 1236 void verify_bidirectional_edges(Unique_Node_List &visited); 1237 1238 // End-of-run dumps. 1239 static void print_statistics() PRODUCT_RETURN; 1240 1241 // Verify ADLC assumptions during startup 1242 static void adlc_verification() PRODUCT_RETURN; 1243 1244 // Definitions of pd methods 1245 static void pd_compiler2_init(); 1246 1247 // Static parse-time type checking logic for gen_subtype_check: 1248 enum SubTypeCheckResult { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test }; 1249 SubTypeCheckResult static_subtype_check(const TypeKlassPtr* superk, const TypeKlassPtr* subk, bool skip = StressReflectiveCode); 1250 1251 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype, 1252 // Optional control dependency (for example, on range check) 1253 Node* ctrl = nullptr); 1254 1255 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check) 1256 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl, bool carry_dependency = false); 1257 1258 // Auxiliary methods for randomized fuzzing/stressing 1259 int random(); 1260 bool randomized_select(int count); 1261 1262 // supporting clone_map 1263 CloneMap& clone_map(); 1264 void set_clone_map(Dict* d); 1265 1266 bool needs_clinit_barrier(ciField* ik, ciMethod* accessing_method); 1267 bool needs_clinit_barrier(ciMethod* ik, ciMethod* accessing_method); 1268 bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method); 1269 1270 #ifdef IA32 1271 private: 1272 bool _select_24_bit_instr; // We selected an instruction with a 24-bit result 1273 bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results 1274 1275 // Remember if this compilation changes hardware mode to 24-bit precision. 1276 void set_24_bit_selection_and_mode(bool selection, bool mode) { 1277 _select_24_bit_instr = selection; 1278 _in_24_bit_fp_mode = mode; 1279 } 1280 1281 public: 1282 bool select_24_bit_instr() const { return _select_24_bit_instr; } 1283 bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; } 1284 #endif // IA32 1285 #ifdef ASSERT 1286 VerifyMeetResult* _type_verify; 1287 void set_exception_backedge() { _exception_backedge = true; } 1288 bool has_exception_backedge() const { return _exception_backedge; } 1289 #endif 1290 1291 static bool push_thru_add(PhaseGVN* phase, Node* z, const TypeInteger* tz, const TypeInteger*& rx, const TypeInteger*& ry, 1292 BasicType out_bt, BasicType in_bt); 1293 1294 static Node* narrow_value(BasicType bt, Node* value, const Type* type, PhaseGVN* phase, bool transform_res); 1295 }; 1296 1297 #endif // SHARE_OPTO_COMPILE_HPP