1 /* 2 * Copyright (c) 1997, 2018, 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_VM_OPTO_COMPILE_HPP 26 #define SHARE_VM_OPTO_COMPILE_HPP 27 28 #include "asm/codeBuffer.hpp" 29 #include "ci/compilerInterface.hpp" 30 #include "code/debugInfoRec.hpp" 31 #include "code/exceptionHandlerTable.hpp" 32 #include "compiler/compilerOracle.hpp" 33 #include "compiler/compileBroker.hpp" 34 #include "libadt/dict.hpp" 35 #include "libadt/vectset.hpp" 36 #include "jfr/jfrEvents.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/timerTrace.hpp" 45 #include "runtime/vmThread.hpp" 46 #include "utilities/ticks.hpp" 47 48 class AddPNode; 49 class Block; 50 class Bundle; 51 class C2Compiler; 52 class CallGenerator; 53 class CloneMap; 54 class ConnectionGraph; 55 class InlineTree; 56 class Int_Array; 57 class LoadBarrierNode; 58 class Matcher; 59 class MachConstantNode; 60 class MachConstantBaseNode; 61 class MachNode; 62 class MachOper; 63 class MachSafePointNode; 64 class Node; 65 class Node_Array; 66 class Node_Notes; 67 class NodeCloneInfo; 68 class OptoReg; 69 class PhaseCFG; 70 class PhaseGVN; 71 class PhaseIterGVN; 72 class PhaseRegAlloc; 73 class PhaseCCP; 74 class PhaseCCP_DCE; 75 class RootNode; 76 class relocInfo; 77 class Scope; 78 class StartNode; 79 class SafePointNode; 80 class JVMState; 81 class Type; 82 class TypeData; 83 class TypeInt; 84 class TypePtr; 85 class TypeOopPtr; 86 class TypeFunc; 87 class Unique_Node_List; 88 class nmethod; 89 class WarmCallInfo; 90 class Node_Stack; 91 struct Final_Reshape_Counts; 92 93 enum LoopOptsMode { 94 LoopOptsDefault, 95 LoopOptsNone, 96 LoopOptsSkipSplitIf, 97 LoopOptsVerify, 98 LoopOptsLastRound 99 }; 100 101 typedef unsigned int node_idx_t; 102 class NodeCloneInfo { 103 private: 104 uint64_t _idx_clone_orig; 105 public: 106 107 void set_idx(node_idx_t idx) { 108 _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx; 109 } 110 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); } 111 112 void set_gen(int generation) { 113 uint64_t g = (uint64_t)generation << 32; 114 _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g; 115 } 116 int gen() const { return (int)(_idx_clone_orig >> 32); } 117 118 void set(uint64_t x) { _idx_clone_orig = x; } 119 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); } 120 uint64_t get() const { return _idx_clone_orig; } 121 122 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {} 123 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); } 124 125 void dump() const; 126 }; 127 128 class CloneMap { 129 friend class Compile; 130 private: 131 bool _debug; 132 Dict* _dict; 133 int _clone_idx; // current cloning iteration/generation in loop unroll 134 public: 135 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy 136 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; } 137 Dict* dict() const { return _dict; } 138 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == NULL, "key existed"); _dict->Insert(_2p(key), (void*)val); } 139 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); } 140 void remove(node_idx_t key) { _dict->Delete(_2p(key)); } 141 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); } 142 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); } 143 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); } 144 int gen(const void* k) const { return gen(_2_node_idx_t(k)); } 145 int max_gen() const; 146 void clone(Node* old, Node* nnn, int gen); 147 void verify_insert_and_clone(Node* old, Node* nnn, int gen); 148 void dump(node_idx_t key) const; 149 150 int clone_idx() const { return _clone_idx; } 151 void set_clone_idx(int x) { _clone_idx = x; } 152 bool is_debug() const { return _debug; } 153 void set_debug(bool debug) { _debug = debug; } 154 static const char* debug_option_name; 155 156 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); } 157 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); } 158 }; 159 160 //------------------------------Compile---------------------------------------- 161 // This class defines a top-level Compiler invocation. 162 163 class Compile : public Phase { 164 friend class VMStructs; 165 166 public: 167 // Fixed alias indexes. (See also MergeMemNode.) 168 enum { 169 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value) 170 AliasIdxBot = 2, // pseudo-index, aliases to everything 171 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM 172 }; 173 174 // Variant of TraceTime(NULL, &_t_accumulator, CITime); 175 // Integrated with logging. If logging is turned on, and CITimeVerbose is true, 176 // then brackets are put into the log, with time stamps and node counts. 177 // (The time collection itself is always conditionalized on CITime.) 178 class TracePhase : public TraceTime { 179 private: 180 Compile* C; 181 CompileLog* _log; 182 const char* _phase_name; 183 bool _dolog; 184 public: 185 TracePhase(const char* name, elapsedTimer* accumulator); 186 ~TracePhase(); 187 }; 188 189 // Information per category of alias (memory slice) 190 class AliasType { 191 private: 192 friend class Compile; 193 194 int _index; // unique index, used with MergeMemNode 195 const TypePtr* _adr_type; // normalized address type 196 ciField* _field; // relevant instance field, or null if none 197 const Type* _element; // relevant array element type, or null if none 198 bool _is_rewritable; // false if the memory is write-once only 199 int _general_index; // if this is type is an instance, the general 200 // type that this is an instance of 201 202 void Init(int i, const TypePtr* at); 203 204 public: 205 int index() const { return _index; } 206 const TypePtr* adr_type() const { return _adr_type; } 207 ciField* field() const { return _field; } 208 const Type* element() const { return _element; } 209 bool is_rewritable() const { return _is_rewritable; } 210 bool is_volatile() const { return (_field ? _field->is_volatile() : false); } 211 int general_index() const { return (_general_index != 0) ? _general_index : _index; } 212 213 void set_rewritable(bool z) { _is_rewritable = z; } 214 void set_field(ciField* f) { 215 assert(!_field,""); 216 _field = f; 217 if (f->is_final() || f->is_stable()) { 218 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops. 219 _is_rewritable = false; 220 } 221 } 222 void set_element(const Type* e) { 223 assert(_element == NULL, ""); 224 _element = e; 225 } 226 227 BasicType basic_type() const; 228 229 void print_on(outputStream* st) PRODUCT_RETURN; 230 }; 231 232 enum { 233 logAliasCacheSize = 6, 234 AliasCacheSize = (1<<logAliasCacheSize) 235 }; 236 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type 237 enum { 238 trapHistLength = MethodData::_trap_hist_limit 239 }; 240 241 // Constant entry of the constant table. 242 class Constant { 243 private: 244 BasicType _type; 245 union { 246 jvalue _value; 247 Metadata* _metadata; 248 } _v; 249 int _offset; // offset of this constant (in bytes) relative to the constant table base. 250 float _freq; 251 bool _can_be_reused; // true (default) if the value can be shared with other users. 252 253 public: 254 Constant() : _type(T_ILLEGAL), _offset(-1), _freq(0.0f), _can_be_reused(true) { _v._value.l = 0; } 255 Constant(BasicType type, jvalue value, float freq = 0.0f, bool can_be_reused = true) : 256 _type(type), 257 _offset(-1), 258 _freq(freq), 259 _can_be_reused(can_be_reused) 260 { 261 assert(type != T_METADATA, "wrong constructor"); 262 _v._value = value; 263 } 264 Constant(Metadata* metadata, bool can_be_reused = true) : 265 _type(T_METADATA), 266 _offset(-1), 267 _freq(0.0f), 268 _can_be_reused(can_be_reused) 269 { 270 _v._metadata = metadata; 271 } 272 273 bool operator==(const Constant& other); 274 275 BasicType type() const { return _type; } 276 277 jint get_jint() const { return _v._value.i; } 278 jlong get_jlong() const { return _v._value.j; } 279 jfloat get_jfloat() const { return _v._value.f; } 280 jdouble get_jdouble() const { return _v._value.d; } 281 jobject get_jobject() const { return _v._value.l; } 282 283 Metadata* get_metadata() const { return _v._metadata; } 284 285 int offset() const { return _offset; } 286 void set_offset(int offset) { _offset = offset; } 287 288 float freq() const { return _freq; } 289 void inc_freq(float freq) { _freq += freq; } 290 291 bool can_be_reused() const { return _can_be_reused; } 292 }; 293 294 // Constant table. 295 class ConstantTable { 296 private: 297 GrowableArray<Constant> _constants; // Constants of this table. 298 int _size; // Size in bytes the emitted constant table takes (including padding). 299 int _table_base_offset; // Offset of the table base that gets added to the constant offsets. 300 int _nof_jump_tables; // Number of jump-tables in this constant table. 301 302 static int qsort_comparator(Constant* a, Constant* b); 303 304 // We use negative frequencies to keep the order of the 305 // jump-tables in which they were added. Otherwise we get into 306 // trouble with relocation. 307 float next_jump_table_freq() { return -1.0f * (++_nof_jump_tables); } 308 309 public: 310 ConstantTable() : 311 _size(-1), 312 _table_base_offset(-1), // We can use -1 here since the constant table is always bigger than 2 bytes (-(size / 2), see MachConstantBaseNode::emit). 313 _nof_jump_tables(0) 314 {} 315 316 int size() const { assert(_size != -1, "not calculated yet"); return _size; } 317 318 int calculate_table_base_offset() const; // AD specific 319 void set_table_base_offset(int x) { assert(_table_base_offset == -1 || x == _table_base_offset, "can't change"); _table_base_offset = x; } 320 int table_base_offset() const { assert(_table_base_offset != -1, "not set yet"); return _table_base_offset; } 321 322 void emit(CodeBuffer& cb); 323 324 // Returns the offset of the last entry (the top) of the constant table. 325 int top_offset() const { assert(_constants.top().offset() != -1, "not bound yet"); return _constants.top().offset(); } 326 327 void calculate_offsets_and_size(); 328 int find_offset(Constant& con) const; 329 330 void add(Constant& con); 331 Constant add(MachConstantNode* n, BasicType type, jvalue value); 332 Constant add(Metadata* metadata); 333 Constant add(MachConstantNode* n, MachOper* oper); 334 Constant add(MachConstantNode* n, jint i) { 335 jvalue value; value.i = i; 336 return add(n, T_INT, value); 337 } 338 Constant add(MachConstantNode* n, jlong j) { 339 jvalue value; value.j = j; 340 return add(n, T_LONG, value); 341 } 342 Constant add(MachConstantNode* n, jfloat f) { 343 jvalue value; value.f = f; 344 return add(n, T_FLOAT, value); 345 } 346 Constant add(MachConstantNode* n, jdouble d) { 347 jvalue value; value.d = d; 348 return add(n, T_DOUBLE, value); 349 } 350 351 // Jump-table 352 Constant add_jump_table(MachConstantNode* n); 353 void fill_jump_table(CodeBuffer& cb, MachConstantNode* n, GrowableArray<Label*> labels) const; 354 }; 355 356 private: 357 // Fixed parameters to this compilation. 358 const int _compile_id; 359 const bool _save_argument_registers; // save/restore arg regs for trampolines 360 const bool _subsume_loads; // Load can be matched as part of a larger op. 361 const bool _do_escape_analysis; // Do escape analysis. 362 const bool _eliminate_boxing; // Do boxing elimination. 363 ciMethod* _method; // The method being compiled. 364 int _entry_bci; // entry bci for osr methods. 365 const TypeFunc* _tf; // My kind of signature 366 InlineTree* _ilt; // Ditto (temporary). 367 address _stub_function; // VM entry for stub being compiled, or NULL 368 const char* _stub_name; // Name of stub or adapter being compiled, or NULL 369 address _stub_entry_point; // Compile code entry for generated stub, or NULL 370 371 // Control of this compilation. 372 int _num_loop_opts; // Number of iterations for doing loop optimiztions 373 int _max_inline_size; // Max inline size for this compilation 374 int _freq_inline_size; // Max hot method inline size for this compilation 375 int _fixed_slots; // count of frame slots not allocated by the register 376 // allocator i.e. locks, original deopt pc, etc. 377 uintx _max_node_limit; // Max unique node count during a single compilation. 378 // For deopt 379 int _orig_pc_slot; 380 int _orig_pc_slot_offset_in_bytes; 381 382 int _major_progress; // Count of something big happening 383 bool _inlining_progress; // progress doing incremental inlining? 384 bool _inlining_incrementally;// Are we doing incremental inlining (post parse) 385 bool _has_loops; // True if the method _may_ have some loops 386 bool _has_split_ifs; // True if the method _may_ have some split-if 387 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores. 388 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated 389 bool _has_boxed_value; // True if a boxed object is allocated 390 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess 391 uint _max_vector_size; // Maximum size of generated vectors 392 bool _clear_upper_avx; // Clear upper bits of ymm registers using vzeroupper 393 uint _trap_hist[trapHistLength]; // Cumulative traps 394 bool _trap_can_recompile; // Have we emitted a recompiling trap? 395 uint _decompile_count; // Cumulative decompilation counts. 396 bool _do_inlining; // True if we intend to do inlining 397 bool _do_scheduling; // True if we intend to do scheduling 398 bool _do_freq_based_layout; // True if we intend to do frequency based block layout 399 bool _do_count_invocations; // True if we generate code to count invocations 400 bool _do_method_data_update; // True if we generate code to update MethodData*s 401 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations 402 bool _use_cmove; // True if CMove should be used without profitability analysis 403 bool _age_code; // True if we need to profile code age (decrement the aging counter) 404 int _AliasLevel; // Locally-adjusted version of AliasLevel flag. 405 bool _print_assembly; // True if we should dump assembly code for this compilation 406 bool _print_inlining; // True if we should print inlining for this compilation 407 bool _print_intrinsics; // True if we should print intrinsics for this compilation 408 #ifndef PRODUCT 409 bool _trace_opto_output; 410 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing 411 #endif 412 bool _has_irreducible_loop; // Found irreducible loops 413 // JSR 292 414 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes. 415 RTMState _rtm_state; // State of Restricted Transactional Memory usage 416 417 // Compilation environment. 418 Arena _comp_arena; // Arena with lifetime equivalent to Compile 419 void* _barrier_set_state; // Potential GC barrier state for Compile 420 ciEnv* _env; // CI interface 421 DirectiveSet* _directive; // Compiler directive 422 CompileLog* _log; // from CompilerThread 423 const char* _failure_reason; // for record_failure/failing pattern 424 GrowableArray<CallGenerator*>* _intrinsics; // List of intrinsics. 425 GrowableArray<Node*>* _macro_nodes; // List of nodes which need to be expanded before matching. 426 GrowableArray<Node*>* _predicate_opaqs; // List of Opaque1 nodes for the loop predicates. 427 GrowableArray<Node*>* _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common 428 GrowableArray<Node*>* _range_check_casts; // List of CastII nodes with a range check dependency 429 GrowableArray<Node*>* _opaque4_nodes; // List of Opaque4 nodes that have a default value 430 ConnectionGraph* _congraph; 431 #ifndef PRODUCT 432 IdealGraphPrinter* _printer; 433 #endif 434 435 436 // Node management 437 uint _unique; // Counter for unique Node indices 438 VectorSet _dead_node_list; // Set of dead nodes 439 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N). 440 // So use this to keep count and make the call O(1). 441 DEBUG_ONLY( Unique_Node_List* _modified_nodes; ) // List of nodes which inputs were modified 442 443 debug_only(static int _debug_idx;) // Monotonic counter (not reset), use -XX:BreakAtNode=<idx> 444 Arena _node_arena; // Arena for new-space Nodes 445 Arena _old_arena; // Arena for old-space Nodes, lifetime during xform 446 RootNode* _root; // Unique root of compilation, or NULL after bail-out. 447 Node* _top; // Unique top node. (Reset by various phases.) 448 449 Node* _immutable_memory; // Initial memory state 450 451 Node* _recent_alloc_obj; 452 Node* _recent_alloc_ctl; 453 454 // Constant table 455 ConstantTable _constant_table; // The constant table for this compile. 456 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton. 457 458 459 // Blocked array of debugging and profiling information, 460 // tracked per node. 461 enum { _log2_node_notes_block_size = 8, 462 _node_notes_block_size = (1<<_log2_node_notes_block_size) 463 }; 464 GrowableArray<Node_Notes*>* _node_note_array; 465 Node_Notes* _default_node_notes; // default notes for new nodes 466 467 // After parsing and every bulk phase we hang onto the Root instruction. 468 // The RootNode instruction is where the whole program begins. It produces 469 // the initial Control and BOTTOM for everybody else. 470 471 // Type management 472 Arena _Compile_types; // Arena for all types 473 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared() 474 Dict* _type_dict; // Intern table 475 CloneMap _clone_map; // used for recording history of cloned nodes 476 void* _type_hwm; // Last allocation (see Type::operator new/delete) 477 size_t _type_last_size; // Last allocation size (see Type::operator new/delete) 478 ciMethod* _last_tf_m; // Cache for 479 const TypeFunc* _last_tf; // TypeFunc::make 480 AliasType** _alias_types; // List of alias types seen so far. 481 int _num_alias_types; // Logical length of _alias_types 482 int _max_alias_types; // Physical length of _alias_types 483 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking 484 485 // Parsing, optimization 486 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN 487 Unique_Node_List* _for_igvn; // Initial work-list for next round of Iterative GVN 488 WarmCallInfo* _warm_calls; // Sorted work-list for heat-based inlining. 489 490 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after 491 // main parsing has finished. 492 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations 493 494 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations 495 496 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining) 497 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending 498 499 500 // Inlining may not happen in parse order which would make 501 // PrintInlining output confusing. Keep track of PrintInlining 502 // pieces in order. 503 class PrintInliningBuffer : public ResourceObj { 504 private: 505 CallGenerator* _cg; 506 stringStream* _ss; 507 508 public: 509 PrintInliningBuffer() 510 : _cg(NULL) { _ss = new stringStream(); } 511 512 void freeStream() { _ss->~stringStream(); _ss = NULL; } 513 514 stringStream* ss() const { return _ss; } 515 CallGenerator* cg() const { return _cg; } 516 void set_cg(CallGenerator* cg) { _cg = cg; } 517 }; 518 519 stringStream* _print_inlining_stream; 520 GrowableArray<PrintInliningBuffer>* _print_inlining_list; 521 int _print_inlining_idx; 522 char* _print_inlining_output; 523 524 // Only keep nodes in the expensive node list that need to be optimized 525 void cleanup_expensive_nodes(PhaseIterGVN &igvn); 526 // Use for sorting expensive nodes to bring similar nodes together 527 static int cmp_expensive_nodes(Node** n1, Node** n2); 528 // Expensive nodes list already sorted? 529 bool expensive_nodes_sorted() const; 530 // Remove the speculative part of types and clean up the graph 531 void remove_speculative_types(PhaseIterGVN &igvn); 532 533 void* _replay_inline_data; // Pointer to data loaded from file 534 535 void print_inlining_stream_free(); 536 void print_inlining_init(); 537 void print_inlining_reinit(); 538 void print_inlining_commit(); 539 void print_inlining_push(); 540 PrintInliningBuffer& print_inlining_current(); 541 542 void log_late_inline_failure(CallGenerator* cg, const char* msg); 543 544 public: 545 546 void* barrier_set_state() const { return _barrier_set_state; } 547 548 outputStream* print_inlining_stream() const { 549 assert(print_inlining() || print_intrinsics(), "PrintInlining off?"); 550 return _print_inlining_stream; 551 } 552 553 void print_inlining_update(CallGenerator* cg); 554 void print_inlining_update_delayed(CallGenerator* cg); 555 void print_inlining_move_to(CallGenerator* cg); 556 void print_inlining_assert_ready(); 557 void print_inlining_reset(); 558 559 void print_inlining(ciMethod* method, int inline_level, int bci, const char* msg = NULL) { 560 stringStream ss; 561 CompileTask::print_inlining_inner(&ss, method, inline_level, bci, msg); 562 print_inlining_stream()->print("%s", ss.as_string()); 563 } 564 565 #ifndef PRODUCT 566 IdealGraphPrinter* printer() { return _printer; } 567 #endif 568 569 void log_late_inline(CallGenerator* cg); 570 void log_inline_id(CallGenerator* cg); 571 void log_inline_failure(const char* msg); 572 573 void* replay_inline_data() const { return _replay_inline_data; } 574 575 // Dump inlining replay data to the stream. 576 void dump_inline_data(outputStream* out); 577 578 private: 579 // Matching, CFG layout, allocation, code generation 580 PhaseCFG* _cfg; // Results of CFG finding 581 bool _select_24_bit_instr; // We selected an instruction with a 24-bit result 582 bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results 583 int _java_calls; // Number of java calls in the method 584 int _inner_loops; // Number of inner loops in the method 585 Matcher* _matcher; // Engine to map ideal to machine instructions 586 PhaseRegAlloc* _regalloc; // Results of register allocation. 587 int _frame_slots; // Size of total frame in stack slots 588 CodeOffsets _code_offsets; // Offsets into the code for various interesting entries 589 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout) 590 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin 591 void* _indexSet_free_block_list; // free list of IndexSet bit blocks 592 int _interpreter_frame_size; 593 594 uint _node_bundling_limit; 595 Bundle* _node_bundling_base; // Information for instruction bundling 596 597 // Instruction bits passed off to the VM 598 int _method_size; // Size of nmethod code segment in bytes 599 CodeBuffer _code_buffer; // Where the code is assembled 600 int _first_block_size; // Size of unvalidated entry point code / OSR poison code 601 ExceptionHandlerTable _handler_table; // Table of native-code exception handlers 602 ImplicitExceptionTable _inc_table; // Table of implicit null checks in native code 603 OopMapSet* _oop_map_set; // Table of oop maps (one for each safepoint location) 604 static int _CompiledZap_count; // counter compared against CompileZap[First/Last] 605 BufferBlob* _scratch_buffer_blob; // For temporary code buffers. 606 relocInfo* _scratch_locs_memory; // For temporary code buffers. 607 int _scratch_const_size; // For temporary code buffers. 608 bool _in_scratch_emit_size; // true when in scratch_emit_size. 609 610 void reshape_address(AddPNode* n); 611 612 public: 613 // Accessors 614 615 // The Compile instance currently active in this (compiler) thread. 616 static Compile* current() { 617 return (Compile*) ciEnv::current()->compiler_data(); 618 } 619 620 // ID for this compilation. Useful for setting breakpoints in the debugger. 621 int compile_id() const { return _compile_id; } 622 DirectiveSet* directive() const { return _directive; } 623 624 // Does this compilation allow instructions to subsume loads? User 625 // instructions that subsume a load may result in an unschedulable 626 // instruction sequence. 627 bool subsume_loads() const { return _subsume_loads; } 628 /** Do escape analysis. */ 629 bool do_escape_analysis() const { return _do_escape_analysis; } 630 /** Do boxing elimination. */ 631 bool eliminate_boxing() const { return _eliminate_boxing; } 632 /** Do aggressive boxing elimination. */ 633 bool aggressive_unboxing() const { return _eliminate_boxing && AggressiveUnboxing; } 634 bool save_argument_registers() const { return _save_argument_registers; } 635 636 637 // Other fixed compilation parameters. 638 ciMethod* method() const { return _method; } 639 int entry_bci() const { return _entry_bci; } 640 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; } 641 bool is_method_compilation() const { return (_method != NULL && !_method->flags().is_native()); } 642 const TypeFunc* tf() const { assert(_tf!=NULL, ""); return _tf; } 643 void init_tf(const TypeFunc* tf) { assert(_tf==NULL, ""); _tf = tf; } 644 InlineTree* ilt() const { return _ilt; } 645 address stub_function() const { return _stub_function; } 646 const char* stub_name() const { return _stub_name; } 647 address stub_entry_point() const { return _stub_entry_point; } 648 649 // Control of this compilation. 650 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; } 651 void set_fixed_slots(int n) { _fixed_slots = n; } 652 int major_progress() const { return _major_progress; } 653 void set_inlining_progress(bool z) { _inlining_progress = z; } 654 int inlining_progress() const { return _inlining_progress; } 655 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; } 656 int inlining_incrementally() const { return _inlining_incrementally; } 657 void set_major_progress() { _major_progress++; } 658 void clear_major_progress() { _major_progress = 0; } 659 int num_loop_opts() const { return _num_loop_opts; } 660 void set_num_loop_opts(int n) { _num_loop_opts = n; } 661 int max_inline_size() const { return _max_inline_size; } 662 void set_freq_inline_size(int n) { _freq_inline_size = n; } 663 int freq_inline_size() const { return _freq_inline_size; } 664 void set_max_inline_size(int n) { _max_inline_size = n; } 665 bool has_loops() const { return _has_loops; } 666 void set_has_loops(bool z) { _has_loops = z; } 667 bool has_split_ifs() const { return _has_split_ifs; } 668 void set_has_split_ifs(bool z) { _has_split_ifs = z; } 669 bool has_unsafe_access() const { return _has_unsafe_access; } 670 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; } 671 bool has_stringbuilder() const { return _has_stringbuilder; } 672 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; } 673 bool has_boxed_value() const { return _has_boxed_value; } 674 void set_has_boxed_value(bool z) { _has_boxed_value = z; } 675 bool has_reserved_stack_access() const { return _has_reserved_stack_access; } 676 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; } 677 uint max_vector_size() const { return _max_vector_size; } 678 void set_max_vector_size(uint s) { _max_vector_size = s; } 679 bool clear_upper_avx() const { return _clear_upper_avx; } 680 void set_clear_upper_avx(bool s) { _clear_upper_avx = s; } 681 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; } 682 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; } 683 bool trap_can_recompile() const { return _trap_can_recompile; } 684 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; } 685 uint decompile_count() const { return _decompile_count; } 686 void set_decompile_count(uint c) { _decompile_count = c; } 687 bool allow_range_check_smearing() const; 688 bool do_inlining() const { return _do_inlining; } 689 void set_do_inlining(bool z) { _do_inlining = z; } 690 bool do_scheduling() const { return _do_scheduling; } 691 void set_do_scheduling(bool z) { _do_scheduling = z; } 692 bool do_freq_based_layout() const{ return _do_freq_based_layout; } 693 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; } 694 bool do_count_invocations() const{ return _do_count_invocations; } 695 void set_do_count_invocations(bool z){ _do_count_invocations = z; } 696 bool do_method_data_update() const { return _do_method_data_update; } 697 void set_do_method_data_update(bool z) { _do_method_data_update = z; } 698 bool do_vector_loop() const { return _do_vector_loop; } 699 void set_do_vector_loop(bool z) { _do_vector_loop = z; } 700 bool use_cmove() const { return _use_cmove; } 701 void set_use_cmove(bool z) { _use_cmove = z; } 702 bool age_code() const { return _age_code; } 703 void set_age_code(bool z) { _age_code = z; } 704 int AliasLevel() const { return _AliasLevel; } 705 bool print_assembly() const { return _print_assembly; } 706 void set_print_assembly(bool z) { _print_assembly = z; } 707 bool print_inlining() const { return _print_inlining; } 708 void set_print_inlining(bool z) { _print_inlining = z; } 709 bool print_intrinsics() const { return _print_intrinsics; } 710 void set_print_intrinsics(bool z) { _print_intrinsics = z; } 711 RTMState rtm_state() const { return _rtm_state; } 712 void set_rtm_state(RTMState s) { _rtm_state = s; } 713 bool use_rtm() const { return (_rtm_state & NoRTM) == 0; } 714 bool profile_rtm() const { return _rtm_state == ProfileRTM; } 715 uint max_node_limit() const { return (uint)_max_node_limit; } 716 void set_max_node_limit(uint n) { _max_node_limit = n; } 717 718 // check the CompilerOracle for special behaviours for this compile 719 bool method_has_option(const char * option) { 720 return method() != NULL && method()->has_option(option); 721 } 722 723 #ifndef PRODUCT 724 bool trace_opto_output() const { return _trace_opto_output; } 725 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; } 726 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; } 727 int _in_dump_cnt; // Required for dumping ir nodes. 728 #endif 729 bool has_irreducible_loop() const { return _has_irreducible_loop; } 730 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; } 731 732 // JSR 292 733 bool has_method_handle_invokes() const { return _has_method_handle_invokes; } 734 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; } 735 736 Ticks _latest_stage_start_counter; 737 738 void begin_method() { 739 #ifndef PRODUCT 740 if (_printer && _printer->should_print(1)) { 741 _printer->begin_method(); 742 } 743 #endif 744 C->_latest_stage_start_counter.stamp(); 745 } 746 747 void print_method(CompilerPhaseType cpt, int level = 1) { 748 EventCompilerPhase event; 749 if (event.should_commit()) { 750 event.set_starttime(C->_latest_stage_start_counter); 751 event.set_phase((u1) cpt); 752 event.set_compileId(C->_compile_id); 753 event.set_phaseLevel(level); 754 event.commit(); 755 } 756 757 758 #ifndef PRODUCT 759 if (_printer && _printer->should_print(level)) { 760 _printer->print_method(CompilerPhaseTypeHelper::to_string(cpt), level); 761 } 762 #endif 763 C->_latest_stage_start_counter.stamp(); 764 } 765 766 void end_method(int level = 1) { 767 EventCompilerPhase event; 768 if (event.should_commit()) { 769 event.set_starttime(C->_latest_stage_start_counter); 770 event.set_phase((u1) PHASE_END); 771 event.set_compileId(C->_compile_id); 772 event.set_phaseLevel(level); 773 event.commit(); 774 } 775 #ifndef PRODUCT 776 if (_printer && _printer->should_print(level)) { 777 _printer->end_method(); 778 } 779 #endif 780 } 781 782 int macro_count() const { return _macro_nodes->length(); } 783 int predicate_count() const { return _predicate_opaqs->length();} 784 int expensive_count() const { return _expensive_nodes->length(); } 785 Node* macro_node(int idx) const { return _macro_nodes->at(idx); } 786 Node* predicate_opaque1_node(int idx) const { return _predicate_opaqs->at(idx);} 787 Node* expensive_node(int idx) const { return _expensive_nodes->at(idx); } 788 ConnectionGraph* congraph() { return _congraph;} 789 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;} 790 void add_macro_node(Node * n) { 791 //assert(n->is_macro(), "must be a macro node"); 792 assert(!_macro_nodes->contains(n), "duplicate entry in expand list"); 793 _macro_nodes->append(n); 794 } 795 void remove_macro_node(Node * n) { 796 // this function may be called twice for a node so check 797 // that the node is in the array before attempting to remove it 798 if (_macro_nodes->contains(n)) 799 _macro_nodes->remove(n); 800 // remove from _predicate_opaqs list also if it is there 801 if (predicate_count() > 0 && _predicate_opaqs->contains(n)){ 802 _predicate_opaqs->remove(n); 803 } 804 } 805 void add_expensive_node(Node * n); 806 void remove_expensive_node(Node * n) { 807 if (_expensive_nodes->contains(n)) { 808 _expensive_nodes->remove(n); 809 } 810 } 811 void add_predicate_opaq(Node * n) { 812 assert(!_predicate_opaqs->contains(n), "duplicate entry in predicate opaque1"); 813 assert(_macro_nodes->contains(n), "should have already been in macro list"); 814 _predicate_opaqs->append(n); 815 } 816 817 // Range check dependent CastII nodes that can be removed after loop optimizations 818 void add_range_check_cast(Node* n); 819 void remove_range_check_cast(Node* n) { 820 if (_range_check_casts->contains(n)) { 821 _range_check_casts->remove(n); 822 } 823 } 824 Node* range_check_cast_node(int idx) const { return _range_check_casts->at(idx); } 825 int range_check_cast_count() const { return _range_check_casts->length(); } 826 // Remove all range check dependent CastIINodes. 827 void remove_range_check_casts(PhaseIterGVN &igvn); 828 829 void add_opaque4_node(Node* n); 830 void remove_opaque4_node(Node* n) { 831 if (_opaque4_nodes->contains(n)) { 832 _opaque4_nodes->remove(n); 833 } 834 } 835 Node* opaque4_node(int idx) const { return _opaque4_nodes->at(idx); } 836 int opaque4_count() const { return _opaque4_nodes->length(); } 837 void remove_opaque4_nodes(PhaseIterGVN &igvn); 838 839 // remove the opaque nodes that protect the predicates so that the unused checks and 840 // uncommon traps will be eliminated from the graph. 841 void cleanup_loop_predicates(PhaseIterGVN &igvn); 842 bool is_predicate_opaq(Node * n) { 843 return _predicate_opaqs->contains(n); 844 } 845 846 // Are there candidate expensive nodes for optimization? 847 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn); 848 // Check whether n1 and n2 are similar 849 static int cmp_expensive_nodes(Node* n1, Node* n2); 850 // Sort expensive nodes to locate similar expensive nodes 851 void sort_expensive_nodes(); 852 853 // Compilation environment. 854 Arena* comp_arena() { return &_comp_arena; } 855 ciEnv* env() const { return _env; } 856 CompileLog* log() const { return _log; } 857 bool failing() const { return _env->failing() || _failure_reason != NULL; } 858 const char* failure_reason() const { return (_env->failing()) ? _env->failure_reason() : _failure_reason; } 859 860 bool failure_reason_is(const char* r) const { 861 return (r == _failure_reason) || (r != NULL && _failure_reason != NULL && strcmp(r, _failure_reason) == 0); 862 } 863 864 void record_failure(const char* reason); 865 void record_method_not_compilable(const char* reason) { 866 // Bailouts cover "all_tiers" when TieredCompilation is off. 867 env()->record_method_not_compilable(reason, !TieredCompilation); 868 // Record failure reason. 869 record_failure(reason); 870 } 871 bool check_node_count(uint margin, const char* reason) { 872 if (live_nodes() + margin > max_node_limit()) { 873 record_method_not_compilable(reason); 874 return true; 875 } else { 876 return false; 877 } 878 } 879 880 // Node management 881 uint unique() const { return _unique; } 882 uint next_unique() { return _unique++; } 883 void set_unique(uint i) { _unique = i; } 884 static int debug_idx() { return debug_only(_debug_idx)+0; } 885 static void set_debug_idx(int i) { debug_only(_debug_idx = i); } 886 Arena* node_arena() { return &_node_arena; } 887 Arena* old_arena() { return &_old_arena; } 888 RootNode* root() const { return _root; } 889 void set_root(RootNode* r) { _root = r; } 890 StartNode* start() const; // (Derived from root.) 891 void init_start(StartNode* s); 892 Node* immutable_memory(); 893 894 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; } 895 Node* recent_alloc_obj() const { return _recent_alloc_obj; } 896 void set_recent_alloc(Node* ctl, Node* obj) { 897 _recent_alloc_ctl = ctl; 898 _recent_alloc_obj = obj; 899 } 900 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return; 901 _dead_node_count++; 902 } 903 bool is_dead_node(uint idx) { return _dead_node_list.test(idx) != 0; } 904 uint dead_node_count() { return _dead_node_count; } 905 void reset_dead_node_list() { _dead_node_list.Reset(); 906 _dead_node_count = 0; 907 } 908 uint live_nodes() const { 909 int val = _unique - _dead_node_count; 910 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count); 911 return (uint) val; 912 } 913 #ifdef ASSERT 914 uint count_live_nodes_by_graph_walk(); 915 void print_missing_nodes(); 916 #endif 917 918 // Record modified nodes to check that they are put on IGVN worklist 919 void record_modified_node(Node* n) NOT_DEBUG_RETURN; 920 void remove_modified_node(Node* n) NOT_DEBUG_RETURN; 921 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } ) 922 923 // Constant table 924 ConstantTable& constant_table() { return _constant_table; } 925 926 MachConstantBaseNode* mach_constant_base_node(); 927 bool has_mach_constant_base_node() const { return _mach_constant_base_node != NULL; } 928 // Generated by adlc, true if CallNode requires MachConstantBase. 929 bool needs_clone_jvms(); 930 931 // Handy undefined Node 932 Node* top() const { return _top; } 933 934 // these are used by guys who need to know about creation and transformation of top: 935 Node* cached_top_node() { return _top; } 936 void set_cached_top_node(Node* tn); 937 938 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; } 939 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; } 940 Node_Notes* default_node_notes() const { return _default_node_notes; } 941 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; } 942 943 Node_Notes* node_notes_at(int idx) { 944 return locate_node_notes(_node_note_array, idx, false); 945 } 946 inline bool set_node_notes_at(int idx, Node_Notes* value); 947 948 // Copy notes from source to dest, if they exist. 949 // Overwrite dest only if source provides something. 950 // Return true if information was moved. 951 bool copy_node_notes_to(Node* dest, Node* source); 952 953 // Workhorse function to sort out the blocked Node_Notes array: 954 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr, 955 int idx, bool can_grow = false); 956 957 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by); 958 959 // Type management 960 Arena* type_arena() { return _type_arena; } 961 Dict* type_dict() { return _type_dict; } 962 void* type_hwm() { return _type_hwm; } 963 size_t type_last_size() { return _type_last_size; } 964 int num_alias_types() { return _num_alias_types; } 965 966 void init_type_arena() { _type_arena = &_Compile_types; } 967 void set_type_arena(Arena* a) { _type_arena = a; } 968 void set_type_dict(Dict* d) { _type_dict = d; } 969 void set_type_hwm(void* p) { _type_hwm = p; } 970 void set_type_last_size(size_t sz) { _type_last_size = sz; } 971 972 const TypeFunc* last_tf(ciMethod* m) { 973 return (m == _last_tf_m) ? _last_tf : NULL; 974 } 975 void set_last_tf(ciMethod* m, const TypeFunc* tf) { 976 assert(m != NULL || tf == NULL, ""); 977 _last_tf_m = m; 978 _last_tf = tf; 979 } 980 981 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; } 982 AliasType* alias_type(const TypePtr* adr_type, ciField* field = NULL) { return find_alias_type(adr_type, false, field); } 983 bool have_alias_type(const TypePtr* adr_type); 984 AliasType* alias_type(ciField* field); 985 986 int get_alias_index(const TypePtr* at) { return alias_type(at)->index(); } 987 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); } 988 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); } 989 990 // Building nodes 991 void rethrow_exceptions(JVMState* jvms); 992 void return_values(JVMState* jvms); 993 JVMState* build_start_state(StartNode* start, const TypeFunc* tf); 994 995 // Decide how to build a call. 996 // The profile factor is a discount to apply to this site's interp. profile. 997 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch, 998 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = NULL, 999 bool allow_intrinsics = true, bool delayed_forbidden = false); 1000 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) { 1001 return should_delay_string_inlining(call_method, jvms) || 1002 should_delay_boxing_inlining(call_method, jvms); 1003 } 1004 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms); 1005 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms); 1006 1007 // Helper functions to identify inlining potential at call-site 1008 ciMethod* optimize_virtual_call(ciMethod* caller, int bci, ciInstanceKlass* klass, 1009 ciKlass* holder, ciMethod* callee, 1010 const TypeOopPtr* receiver_type, bool is_virtual, 1011 bool &call_does_dispatch, int &vtable_index, 1012 bool check_access = true); 1013 ciMethod* optimize_inlining(ciMethod* caller, int bci, ciInstanceKlass* klass, 1014 ciMethod* callee, const TypeOopPtr* receiver_type, 1015 bool check_access = true); 1016 1017 // Report if there were too many traps at a current method and bci. 1018 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded. 1019 // If there is no MDO at all, report no trap unless told to assume it. 1020 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason); 1021 // This version, unspecific to a particular bci, asks if 1022 // PerMethodTrapLimit was exceeded for all inlined methods seen so far. 1023 bool too_many_traps(Deoptimization::DeoptReason reason, 1024 // Privately used parameter for logging: 1025 ciMethodData* logmd = NULL); 1026 // Report if there were too many recompiles at a method and bci. 1027 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason); 1028 // Report if there were too many traps or recompiles at a method and bci. 1029 bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) { 1030 return too_many_traps(method, bci, reason) || 1031 too_many_recompiles(method, bci, reason); 1032 } 1033 // Return a bitset with the reasons where deoptimization is allowed, 1034 // i.e., where there were not too many uncommon traps. 1035 int _allowed_reasons; 1036 int allowed_deopt_reasons() { return _allowed_reasons; } 1037 void set_allowed_deopt_reasons(); 1038 1039 // Parsing, optimization 1040 PhaseGVN* initial_gvn() { return _initial_gvn; } 1041 Unique_Node_List* for_igvn() { return _for_igvn; } 1042 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List. 1043 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; } 1044 void set_for_igvn(Unique_Node_List *for_igvn) { _for_igvn = for_igvn; } 1045 1046 // Replace n by nn using initial_gvn, calling hash_delete and 1047 // record_for_igvn as needed. 1048 void gvn_replace_by(Node* n, Node* nn); 1049 1050 1051 void identify_useful_nodes(Unique_Node_List &useful); 1052 void update_dead_node_list(Unique_Node_List &useful); 1053 void remove_useless_nodes (Unique_Node_List &useful); 1054 1055 WarmCallInfo* warm_calls() const { return _warm_calls; } 1056 void set_warm_calls(WarmCallInfo* l) { _warm_calls = l; } 1057 WarmCallInfo* pop_warm_call(); 1058 1059 // Record this CallGenerator for inlining at the end of parsing. 1060 void add_late_inline(CallGenerator* cg) { 1061 _late_inlines.insert_before(_late_inlines_pos, cg); 1062 _late_inlines_pos++; 1063 } 1064 1065 void prepend_late_inline(CallGenerator* cg) { 1066 _late_inlines.insert_before(0, cg); 1067 } 1068 1069 void add_string_late_inline(CallGenerator* cg) { 1070 _string_late_inlines.push(cg); 1071 } 1072 1073 void add_boxing_late_inline(CallGenerator* cg) { 1074 _boxing_late_inlines.push(cg); 1075 } 1076 1077 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful); 1078 1079 void process_print_inlining(); 1080 void dump_print_inlining(); 1081 1082 bool over_inlining_cutoff() const { 1083 if (!inlining_incrementally()) { 1084 return unique() > (uint)NodeCountInliningCutoff; 1085 } else { 1086 return live_nodes() > (uint)LiveNodeCountInliningCutoff; 1087 } 1088 } 1089 1090 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; } 1091 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--; } 1092 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; } 1093 1094 void inline_incrementally_one(PhaseIterGVN& igvn); 1095 void inline_incrementally(PhaseIterGVN& igvn); 1096 void inline_string_calls(bool parse_time); 1097 void inline_boxing_calls(PhaseIterGVN& igvn); 1098 bool optimize_loops(int& loop_opts_cnt, PhaseIterGVN& igvn, LoopOptsMode mode); 1099 1100 // Matching, CFG layout, allocation, code generation 1101 PhaseCFG* cfg() { return _cfg; } 1102 bool select_24_bit_instr() const { return _select_24_bit_instr; } 1103 bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; } 1104 bool has_java_calls() const { return _java_calls > 0; } 1105 int java_calls() const { return _java_calls; } 1106 int inner_loops() const { return _inner_loops; } 1107 Matcher* matcher() { return _matcher; } 1108 PhaseRegAlloc* regalloc() { return _regalloc; } 1109 int frame_slots() const { return _frame_slots; } 1110 int frame_size_in_words() const; // frame_slots in units of the polymorphic 'words' 1111 int frame_size_in_bytes() const { return _frame_slots << LogBytesPerInt; } 1112 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; } 1113 Arena* indexSet_arena() { return _indexSet_arena; } 1114 void* indexSet_free_block_list() { return _indexSet_free_block_list; } 1115 uint node_bundling_limit() { return _node_bundling_limit; } 1116 Bundle* node_bundling_base() { return _node_bundling_base; } 1117 void set_node_bundling_limit(uint n) { _node_bundling_limit = n; } 1118 void set_node_bundling_base(Bundle* b) { _node_bundling_base = b; } 1119 bool starts_bundle(const Node *n) const; 1120 bool need_stack_bang(int frame_size_in_bytes) const; 1121 bool need_register_stack_bang() const; 1122 1123 void update_interpreter_frame_size(int size) { 1124 if (_interpreter_frame_size < size) { 1125 _interpreter_frame_size = size; 1126 } 1127 } 1128 int bang_size_in_bytes() const; 1129 1130 void set_matcher(Matcher* m) { _matcher = m; } 1131 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; } 1132 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; } 1133 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; } 1134 1135 // Remember if this compilation changes hardware mode to 24-bit precision 1136 void set_24_bit_selection_and_mode(bool selection, bool mode) { 1137 _select_24_bit_instr = selection; 1138 _in_24_bit_fp_mode = mode; 1139 } 1140 1141 void set_java_calls(int z) { _java_calls = z; } 1142 void set_inner_loops(int z) { _inner_loops = z; } 1143 1144 // Instruction bits passed off to the VM 1145 int code_size() { return _method_size; } 1146 CodeBuffer* code_buffer() { return &_code_buffer; } 1147 int first_block_size() { return _first_block_size; } 1148 void set_frame_complete(int off) { if (!in_scratch_emit_size()) { _code_offsets.set_value(CodeOffsets::Frame_Complete, off); } } 1149 ExceptionHandlerTable* handler_table() { return &_handler_table; } 1150 ImplicitExceptionTable* inc_table() { return &_inc_table; } 1151 OopMapSet* oop_map_set() { return _oop_map_set; } 1152 DebugInformationRecorder* debug_info() { return env()->debug_info(); } 1153 Dependencies* dependencies() { return env()->dependencies(); } 1154 static int CompiledZap_count() { return _CompiledZap_count; } 1155 BufferBlob* scratch_buffer_blob() { return _scratch_buffer_blob; } 1156 void init_scratch_buffer_blob(int const_size); 1157 void clear_scratch_buffer_blob(); 1158 void set_scratch_buffer_blob(BufferBlob* b) { _scratch_buffer_blob = b; } 1159 relocInfo* scratch_locs_memory() { return _scratch_locs_memory; } 1160 void set_scratch_locs_memory(relocInfo* b) { _scratch_locs_memory = b; } 1161 1162 // emit to scratch blob, report resulting size 1163 uint scratch_emit_size(const Node* n); 1164 void set_in_scratch_emit_size(bool x) { _in_scratch_emit_size = x; } 1165 bool in_scratch_emit_size() const { return _in_scratch_emit_size; } 1166 1167 enum ScratchBufferBlob { 1168 #if defined(PPC64) 1169 MAX_inst_size = 2048, 1170 #else 1171 MAX_inst_size = 1024, 1172 #endif 1173 MAX_locs_size = 128, // number of relocInfo elements 1174 MAX_const_size = 128, 1175 MAX_stubs_size = 128 1176 }; 1177 1178 // Major entry point. Given a Scope, compile the associated method. 1179 // For normal compilations, entry_bci is InvocationEntryBci. For on stack 1180 // replacement, entry_bci indicates the bytecode for which to compile a 1181 // continuation. 1182 Compile(ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, 1183 int entry_bci, bool subsume_loads, bool do_escape_analysis, 1184 bool eliminate_boxing, DirectiveSet* directive); 1185 1186 // Second major entry point. From the TypeFunc signature, generate code 1187 // to pass arguments from the Java calling convention to the C calling 1188 // convention. 1189 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(), 1190 address stub_function, const char *stub_name, 1191 int is_fancy_jump, bool pass_tls, 1192 bool save_arg_registers, bool return_pc, DirectiveSet* directive); 1193 1194 // From the TypeFunc signature, generate code to pass arguments 1195 // from Compiled calling convention to Interpreter's calling convention 1196 void Generate_Compiled_To_Interpreter_Graph(const TypeFunc *tf, address interpreter_entry); 1197 1198 // From the TypeFunc signature, generate code to pass arguments 1199 // from Interpreter's calling convention to Compiler's calling convention 1200 void Generate_Interpreter_To_Compiled_Graph(const TypeFunc *tf); 1201 1202 // Are we compiling a method? 1203 bool has_method() { return method() != NULL; } 1204 1205 // Maybe print some information about this compile. 1206 void print_compile_messages(); 1207 1208 // Final graph reshaping, a post-pass after the regular optimizer is done. 1209 bool final_graph_reshaping(); 1210 1211 // returns true if adr is completely contained in the given alias category 1212 bool must_alias(const TypePtr* adr, int alias_idx); 1213 1214 // returns true if adr overlaps with the given alias category 1215 bool can_alias(const TypePtr* adr, int alias_idx); 1216 1217 // Driver for converting compiler's IR into machine code bits 1218 void Output(); 1219 1220 // Accessors for node bundling info. 1221 Bundle* node_bundling(const Node *n); 1222 bool valid_bundle_info(const Node *n); 1223 1224 // Schedule and Bundle the instructions 1225 void ScheduleAndBundle(); 1226 1227 // Build OopMaps for each GC point 1228 void BuildOopMaps(); 1229 1230 // Append debug info for the node "local" at safepoint node "sfpt" to the 1231 // "array", May also consult and add to "objs", which describes the 1232 // scalar-replaced objects. 1233 void FillLocArray( int idx, MachSafePointNode* sfpt, 1234 Node *local, GrowableArray<ScopeValue*> *array, 1235 GrowableArray<ScopeValue*> *objs ); 1236 1237 // If "objs" contains an ObjectValue whose id is "id", returns it, else NULL. 1238 static ObjectValue* sv_for_node_id(GrowableArray<ScopeValue*> *objs, int id); 1239 // Requres that "objs" does not contains an ObjectValue whose id matches 1240 // that of "sv. Appends "sv". 1241 static void set_sv_for_object_node(GrowableArray<ScopeValue*> *objs, 1242 ObjectValue* sv ); 1243 1244 // Process an OopMap Element while emitting nodes 1245 void Process_OopMap_Node(MachNode *mach, int code_offset); 1246 1247 // Initialize code buffer 1248 CodeBuffer* init_buffer(uint* blk_starts); 1249 1250 // Write out basic block data to code buffer 1251 void fill_buffer(CodeBuffer* cb, uint* blk_starts); 1252 1253 // Determine which variable sized branches can be shortened 1254 void shorten_branches(uint* blk_starts, int& code_size, int& reloc_size, int& stub_size); 1255 1256 // Compute the size of first NumberOfLoopInstrToAlign instructions 1257 // at the head of a loop. 1258 void compute_loop_first_inst_sizes(); 1259 1260 // Compute the information for the exception tables 1261 void FillExceptionTables(uint cnt, uint *call_returns, uint *inct_starts, Label *blk_labels); 1262 1263 // Stack slots that may be unused by the calling convention but must 1264 // otherwise be preserved. On Intel this includes the return address. 1265 // On PowerPC it includes the 4 words holding the old TOC & LR glue. 1266 uint in_preserve_stack_slots(); 1267 1268 // "Top of Stack" slots that may be unused by the calling convention but must 1269 // otherwise be preserved. 1270 // On Intel these are not necessary and the value can be zero. 1271 // On Sparc this describes the words reserved for storing a register window 1272 // when an interrupt occurs. 1273 static uint out_preserve_stack_slots(); 1274 1275 // Number of outgoing stack slots killed above the out_preserve_stack_slots 1276 // for calls to C. Supports the var-args backing area for register parms. 1277 uint varargs_C_out_slots_killed() const; 1278 1279 // Number of Stack Slots consumed by a synchronization entry 1280 int sync_stack_slots() const; 1281 1282 // Compute the name of old_SP. See <arch>.ad for frame layout. 1283 OptoReg::Name compute_old_SP(); 1284 1285 private: 1286 // Phase control: 1287 void Init(int aliaslevel); // Prepare for a single compilation 1288 int Inline_Warm(); // Find more inlining work. 1289 void Finish_Warm(); // Give up on further inlines. 1290 void Optimize(); // Given a graph, optimize it 1291 void Code_Gen(); // Generate code from a graph 1292 1293 // Management of the AliasType table. 1294 void grow_alias_types(); 1295 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type); 1296 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const; 1297 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field); 1298 1299 void verify_top(Node*) const PRODUCT_RETURN; 1300 1301 // Intrinsic setup. 1302 void register_library_intrinsics(); // initializer 1303 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor 1304 int intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found); // helper 1305 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn 1306 void register_intrinsic(CallGenerator* cg); // update fn 1307 1308 #ifndef PRODUCT 1309 static juint _intrinsic_hist_count[vmIntrinsics::ID_LIMIT]; 1310 static jubyte _intrinsic_hist_flags[vmIntrinsics::ID_LIMIT]; 1311 #endif 1312 // Function calls made by the public function final_graph_reshaping. 1313 // No need to be made public as they are not called elsewhere. 1314 void final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &frc); 1315 void final_graph_reshaping_walk( Node_Stack &nstack, Node *root, Final_Reshape_Counts &frc ); 1316 void eliminate_redundant_card_marks(Node* n); 1317 1318 public: 1319 1320 // Note: Histogram array size is about 1 Kb. 1321 enum { // flag bits: 1322 _intrinsic_worked = 1, // succeeded at least once 1323 _intrinsic_failed = 2, // tried it but it failed 1324 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps) 1325 _intrinsic_virtual = 8, // was seen in the virtual form (rare) 1326 _intrinsic_both = 16 // was seen in the non-virtual form (usual) 1327 }; 1328 // Update histogram. Return boolean if this is a first-time occurrence. 1329 static bool gather_intrinsic_statistics(vmIntrinsics::ID id, 1330 bool is_virtual, int flags) PRODUCT_RETURN0; 1331 static void print_intrinsic_statistics() PRODUCT_RETURN; 1332 1333 // Graph verification code 1334 // Walk the node list, verifying that there is a one-to-one 1335 // correspondence between Use-Def edges and Def-Use edges 1336 // The option no_dead_code enables stronger checks that the 1337 // graph is strongly connected from root in both directions. 1338 void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN; 1339 1340 // Verify GC barrier patterns 1341 void verify_barriers() PRODUCT_RETURN; 1342 1343 // End-of-run dumps. 1344 static void print_statistics() PRODUCT_RETURN; 1345 1346 // Dump formatted assembly 1347 void dump_asm(int *pcs = NULL, uint pc_limit = 0) PRODUCT_RETURN; 1348 void dump_pc(int *pcs, int pc_limit, Node *n); 1349 1350 // Verify ADLC assumptions during startup 1351 static void adlc_verification() PRODUCT_RETURN; 1352 1353 // Definitions of pd methods 1354 static void pd_compiler2_init(); 1355 1356 // Static parse-time type checking logic for gen_subtype_check: 1357 enum { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test }; 1358 int static_subtype_check(ciKlass* superk, ciKlass* subk); 1359 1360 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype, 1361 // Optional control dependency (for example, on range check) 1362 Node* ctrl = NULL); 1363 1364 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check) 1365 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl); 1366 1367 // Auxiliary method for randomized fuzzing/stressing 1368 static bool randomized_select(int count); 1369 1370 // supporting clone_map 1371 CloneMap& clone_map(); 1372 void set_clone_map(Dict* d); 1373 1374 bool is_compiling_clinit_for(ciKlass* k); 1375 }; 1376 1377 #endif // SHARE_VM_OPTO_COMPILE_HPP