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