1 /* 2 * Copyright (c) 1997, 2015, 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 "memory/resourceArea.hpp" 37 #include "opto/idealGraphPrinter.hpp" 38 #include "opto/phasetype.hpp" 39 #include "opto/phase.hpp" 40 #include "opto/regmask.hpp" 41 #include "runtime/deoptimization.hpp" 42 #include "runtime/vmThread.hpp" 43 #include "trace/tracing.hpp" 44 #include "utilities/ticks.hpp" 45 46 class Block; 47 class Bundle; 48 class C2Compiler; 49 class CallGenerator; 50 class CloneMap; 51 class ConnectionGraph; 52 class InlineTree; 53 class Int_Array; 54 class Matcher; 55 class MachConstantNode; 56 class MachConstantBaseNode; 57 class MachNode; 58 class MachOper; 59 class MachSafePointNode; 60 class Node; 61 class Node_Array; 62 class Node_Notes; 63 class NodeCloneInfo; 64 class OptoReg; 65 class PhaseCFG; 66 class PhaseGVN; 67 class PhaseIterGVN; 68 class PhaseRegAlloc; 69 class PhaseCCP; 70 class PhaseCCP_DCE; 71 class RootNode; 72 class relocInfo; 73 class Scope; 74 class StartNode; 75 class SafePointNode; 76 class JVMState; 77 class Type; 78 class TypeData; 79 class TypeInt; 80 class TypePtr; 81 class TypeOopPtr; 82 class TypeFunc; 83 class Unique_Node_List; 84 class nmethod; 85 class WarmCallInfo; 86 class Node_Stack; 87 struct Final_Reshape_Counts; 88 89 typedef unsigned int node_idx_t; 90 class NodeCloneInfo { 91 private: 92 uint64_t _idx_clone_orig; 93 public: 94 95 void set_idx(node_idx_t idx) { 96 _idx_clone_orig = _idx_clone_orig & CONST64(0xFFFFFFFF00000000) | idx; 97 } 98 node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); } 99 100 void set_gen(int generation) { 101 uint64_t g = (uint64_t)generation << 32; 102 _idx_clone_orig = _idx_clone_orig & 0xFFFFFFFF | g; 103 } 104 int gen() const { return (int)(_idx_clone_orig >> 32); } 105 106 void set(uint64_t x) { _idx_clone_orig = x; } 107 void set(node_idx_t x, int g) { set_idx(x); set_gen(g); } 108 uint64_t get() const { return _idx_clone_orig; } 109 110 NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {} 111 NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); } 112 113 void dump() const; 114 }; 115 116 class CloneMap { 117 friend class Compile; 118 private: 119 bool _debug; 120 Dict* _dict; 121 int _clone_idx; // current cloning iteration/generation in loop unroll 122 public: 123 void* _2p(node_idx_t key) const { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy 124 node_idx_t _2_node_idx_t(const void* k) const { return (node_idx_t)(intptr_t)k; } 125 Dict* dict() const { return _dict; } 126 void insert(node_idx_t key, uint64_t val) { assert(_dict->operator[](_2p(key)) == NULL, "key existed"); _dict->Insert(_2p(key), (void*)val); } 127 void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); } 128 void remove(node_idx_t key) { _dict->Delete(_2p(key)); } 129 uint64_t value(node_idx_t key) const { return (uint64_t)_dict->operator[](_2p(key)); } 130 node_idx_t idx(node_idx_t key) const { return NodeCloneInfo(value(key)).idx(); } 131 int gen(node_idx_t key) const { return NodeCloneInfo(value(key)).gen(); } 132 int gen(const void* k) const { return gen(_2_node_idx_t(k)); } 133 int max_gen() const; 134 void clone(Node* old, Node* nnn, int gen); 135 void verify_insert_and_clone(Node* old, Node* nnn, int gen); 136 void dump(node_idx_t key) const; 137 138 int clone_idx() const { return _clone_idx; } 139 void set_clone_idx(int x) { _clone_idx = x; } 140 bool is_debug() const { return _debug; } 141 void set_debug(bool debug) { _debug = debug; } 142 static const char* debug_option_name; 143 144 bool same_idx(node_idx_t k1, node_idx_t k2) const { return idx(k1) == idx(k2); } 145 bool same_gen(node_idx_t k1, node_idx_t k2) const { return gen(k1) == gen(k2); } 146 }; 147 148 //------------------------------Compile---------------------------------------- 149 // This class defines a top-level Compiler invocation. 150 151 class Compile : public Phase { 152 friend class VMStructs; 153 154 public: 155 // Fixed alias indexes. (See also MergeMemNode.) 156 enum { 157 AliasIdxTop = 1, // pseudo-index, aliases to nothing (used as sentinel value) 158 AliasIdxBot = 2, // pseudo-index, aliases to everything 159 AliasIdxRaw = 3 // hard-wired index for TypeRawPtr::BOTTOM 160 }; 161 162 // Variant of TraceTime(NULL, &_t_accumulator, CITime); 163 // Integrated with logging. If logging is turned on, and CITimeVerbose is true, 164 // then brackets are put into the log, with time stamps and node counts. 165 // (The time collection itself is always conditionalized on CITime.) 166 class TracePhase : public TraceTime { 167 private: 168 Compile* C; 169 CompileLog* _log; 170 const char* _phase_name; 171 bool _dolog; 172 public: 173 TracePhase(const char* name, elapsedTimer* accumulator); 174 ~TracePhase(); 175 }; 176 177 // Information per category of alias (memory slice) 178 class AliasType { 179 private: 180 friend class Compile; 181 182 int _index; // unique index, used with MergeMemNode 183 const TypePtr* _adr_type; // normalized address type 184 ciField* _field; // relevant instance field, or null if none 185 const Type* _element; // relevant array element type, or null if none 186 bool _is_rewritable; // false if the memory is write-once only 187 int _general_index; // if this is type is an instance, the general 188 // type that this is an instance of 189 190 void Init(int i, const TypePtr* at); 191 192 public: 193 int index() const { return _index; } 194 const TypePtr* adr_type() const { return _adr_type; } 195 ciField* field() const { return _field; } 196 const Type* element() const { return _element; } 197 bool is_rewritable() const { return _is_rewritable; } 198 bool is_volatile() const { return (_field ? _field->is_volatile() : false); } 199 int general_index() const { return (_general_index != 0) ? _general_index : _index; } 200 201 void set_rewritable(bool z) { _is_rewritable = z; } 202 void set_field(ciField* f) { 203 assert(!_field,""); 204 _field = f; 205 if (f->is_final() || f->is_stable()) { 206 // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops. 207 _is_rewritable = false; 208 } 209 } 210 void set_element(const Type* e) { 211 assert(_element == NULL, ""); 212 _element = e; 213 } 214 215 void print_on(outputStream* st) PRODUCT_RETURN; 216 }; 217 218 enum { 219 logAliasCacheSize = 6, 220 AliasCacheSize = (1<<logAliasCacheSize) 221 }; 222 struct AliasCacheEntry { const TypePtr* _adr_type; int _index; }; // simple duple type 223 enum { 224 trapHistLength = MethodData::_trap_hist_limit 225 }; 226 227 // Constant entry of the constant table. 228 class Constant { 229 private: 230 BasicType _type; 231 union { 232 jvalue _value; 233 Metadata* _metadata; 234 } _v; 235 int _offset; // offset of this constant (in bytes) relative to the constant table base. 236 float _freq; 237 bool _can_be_reused; // true (default) if the value can be shared with other users. 238 239 public: 240 Constant() : _type(T_ILLEGAL), _offset(-1), _freq(0.0f), _can_be_reused(true) { _v._value.l = 0; } 241 Constant(BasicType type, jvalue value, float freq = 0.0f, bool can_be_reused = true) : 242 _type(type), 243 _offset(-1), 244 _freq(freq), 245 _can_be_reused(can_be_reused) 246 { 247 assert(type != T_METADATA, "wrong constructor"); 248 _v._value = value; 249 } 250 Constant(Metadata* metadata, bool can_be_reused = true) : 251 _type(T_METADATA), 252 _offset(-1), 253 _freq(0.0f), 254 _can_be_reused(can_be_reused) 255 { 256 _v._metadata = metadata; 257 } 258 259 bool operator==(const Constant& other); 260 261 BasicType type() const { return _type; } 262 263 jlong get_jlong() const { return _v._value.j; } 264 jfloat get_jfloat() const { return _v._value.f; } 265 jdouble get_jdouble() const { return _v._value.d; } 266 jobject get_jobject() const { return _v._value.l; } 267 268 Metadata* get_metadata() const { return _v._metadata; } 269 270 int offset() const { return _offset; } 271 void set_offset(int offset) { _offset = offset; } 272 273 float freq() const { return _freq; } 274 void inc_freq(float freq) { _freq += freq; } 275 276 bool can_be_reused() const { return _can_be_reused; } 277 }; 278 279 // Constant table. 280 class ConstantTable { 281 private: 282 GrowableArray<Constant> _constants; // Constants of this table. 283 int _size; // Size in bytes the emitted constant table takes (including padding). 284 int _table_base_offset; // Offset of the table base that gets added to the constant offsets. 285 int _nof_jump_tables; // Number of jump-tables in this constant table. 286 287 static int qsort_comparator(Constant* a, Constant* b); 288 289 // We use negative frequencies to keep the order of the 290 // jump-tables in which they were added. Otherwise we get into 291 // trouble with relocation. 292 float next_jump_table_freq() { return -1.0f * (++_nof_jump_tables); } 293 294 public: 295 ConstantTable() : 296 _size(-1), 297 _table_base_offset(-1), // We can use -1 here since the constant table is always bigger than 2 bytes (-(size / 2), see MachConstantBaseNode::emit). 298 _nof_jump_tables(0) 299 {} 300 301 int size() const { assert(_size != -1, "not calculated yet"); return _size; } 302 303 int calculate_table_base_offset() const; // AD specific 304 void set_table_base_offset(int x) { assert(_table_base_offset == -1 || x == _table_base_offset, "can't change"); _table_base_offset = x; } 305 int table_base_offset() const { assert(_table_base_offset != -1, "not set yet"); return _table_base_offset; } 306 307 void emit(CodeBuffer& cb); 308 309 // Returns the offset of the last entry (the top) of the constant table. 310 int top_offset() const { assert(_constants.top().offset() != -1, "not bound yet"); return _constants.top().offset(); } 311 312 void calculate_offsets_and_size(); 313 int find_offset(Constant& con) const; 314 315 void add(Constant& con); 316 Constant add(MachConstantNode* n, BasicType type, jvalue value); 317 Constant add(Metadata* metadata); 318 Constant add(MachConstantNode* n, MachOper* oper); 319 Constant add(MachConstantNode* n, jfloat f) { 320 jvalue value; value.f = f; 321 return add(n, T_FLOAT, value); 322 } 323 Constant add(MachConstantNode* n, jdouble d) { 324 jvalue value; value.d = d; 325 return add(n, T_DOUBLE, value); 326 } 327 328 // Jump-table 329 Constant add_jump_table(MachConstantNode* n); 330 void fill_jump_table(CodeBuffer& cb, MachConstantNode* n, GrowableArray<Label*> labels) const; 331 }; 332 333 private: 334 // Fixed parameters to this compilation. 335 const int _compile_id; 336 const bool _save_argument_registers; // save/restore arg regs for trampolines 337 const bool _subsume_loads; // Load can be matched as part of a larger op. 338 const bool _do_escape_analysis; // Do escape analysis. 339 const bool _eliminate_boxing; // Do boxing elimination. 340 ciMethod* _method; // The method being compiled. 341 int _entry_bci; // entry bci for osr methods. 342 const TypeFunc* _tf; // My kind of signature 343 InlineTree* _ilt; // Ditto (temporary). 344 address _stub_function; // VM entry for stub being compiled, or NULL 345 const char* _stub_name; // Name of stub or adapter being compiled, or NULL 346 address _stub_entry_point; // Compile code entry for generated stub, or NULL 347 348 // Control of this compilation. 349 int _num_loop_opts; // Number of iterations for doing loop optimiztions 350 int _max_inline_size; // Max inline size for this compilation 351 int _freq_inline_size; // Max hot method inline size for this compilation 352 int _fixed_slots; // count of frame slots not allocated by the register 353 // allocator i.e. locks, original deopt pc, etc. 354 uintx _max_node_limit; // Max unique node count during a single compilation. 355 // For deopt 356 int _orig_pc_slot; 357 int _orig_pc_slot_offset_in_bytes; 358 359 int _major_progress; // Count of something big happening 360 bool _inlining_progress; // progress doing incremental inlining? 361 bool _inlining_incrementally;// Are we doing incremental inlining (post parse) 362 bool _has_loops; // True if the method _may_ have some loops 363 bool _has_split_ifs; // True if the method _may_ have some split-if 364 bool _has_unsafe_access; // True if the method _may_ produce faults in unsafe loads or stores. 365 bool _has_stringbuilder; // True StringBuffers or StringBuilders are allocated 366 bool _has_boxed_value; // True if a boxed object is allocated 367 bool _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess 368 int _max_vector_size; // Maximum size of generated vectors 369 uint _trap_hist[trapHistLength]; // Cumulative traps 370 bool _trap_can_recompile; // Have we emitted a recompiling trap? 371 uint _decompile_count; // Cumulative decompilation counts. 372 bool _do_inlining; // True if we intend to do inlining 373 bool _do_scheduling; // True if we intend to do scheduling 374 bool _do_freq_based_layout; // True if we intend to do frequency based block layout 375 bool _do_count_invocations; // True if we generate code to count invocations 376 bool _do_method_data_update; // True if we generate code to update MethodData*s 377 bool _do_vector_loop; // True if allowed to execute loop in parallel iterations 378 bool _use_cmove; // True if CMove should be used without profitability analysis 379 bool _age_code; // True if we need to profile code age (decrement the aging counter) 380 int _AliasLevel; // Locally-adjusted version of AliasLevel flag. 381 bool _print_assembly; // True if we should dump assembly code for this compilation 382 bool _print_inlining; // True if we should print inlining for this compilation 383 bool _print_intrinsics; // True if we should print intrinsics for this compilation 384 #ifndef PRODUCT 385 bool _trace_opto_output; 386 bool _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing 387 #endif 388 bool _has_irreducible_loop; // Found irreducible loops 389 // JSR 292 390 bool _has_method_handle_invokes; // True if this method has MethodHandle invokes. 391 RTMState _rtm_state; // State of Restricted Transactional Memory usage 392 393 // Compilation environment. 394 Arena _comp_arena; // Arena with lifetime equivalent to Compile 395 ciEnv* _env; // CI interface 396 DirectiveSet* _directive; // Compiler directive 397 CompileLog* _log; // from CompilerThread 398 const char* _failure_reason; // for record_failure/failing pattern 399 GrowableArray<CallGenerator*>* _intrinsics; // List of intrinsics. 400 GrowableArray<Node*>* _macro_nodes; // List of nodes which need to be expanded before matching. 401 GrowableArray<Node*>* _predicate_opaqs; // List of Opaque1 nodes for the loop predicates. 402 GrowableArray<Node*>* _expensive_nodes; // List of nodes that are expensive to compute and that we'd better not let the GVN freely common 403 #ifdef _LP64 404 GrowableArray<Node*>* _range_check_casts; // List of CastII nodes with a range check dependency 405 #endif 406 ConnectionGraph* _congraph; 407 #ifndef PRODUCT 408 IdealGraphPrinter* _printer; 409 #endif 410 411 412 // Node management 413 uint _unique; // Counter for unique Node indices 414 VectorSet _dead_node_list; // Set of dead nodes 415 uint _dead_node_count; // Number of dead nodes; VectorSet::Size() is O(N). 416 // So use this to keep count and make the call O(1). 417 DEBUG_ONLY( Unique_Node_List* _modified_nodes; ) // List of nodes which inputs were modified 418 419 debug_only(static int _debug_idx;) // Monotonic counter (not reset), use -XX:BreakAtNode=<idx> 420 Arena _node_arena; // Arena for new-space Nodes 421 Arena _old_arena; // Arena for old-space Nodes, lifetime during xform 422 RootNode* _root; // Unique root of compilation, or NULL after bail-out. 423 Node* _top; // Unique top node. (Reset by various phases.) 424 425 Node* _immutable_memory; // Initial memory state 426 427 Node* _recent_alloc_obj; 428 Node* _recent_alloc_ctl; 429 430 // Constant table 431 ConstantTable _constant_table; // The constant table for this compile. 432 MachConstantBaseNode* _mach_constant_base_node; // Constant table base node singleton. 433 434 435 // Blocked array of debugging and profiling information, 436 // tracked per node. 437 enum { _log2_node_notes_block_size = 8, 438 _node_notes_block_size = (1<<_log2_node_notes_block_size) 439 }; 440 GrowableArray<Node_Notes*>* _node_note_array; 441 Node_Notes* _default_node_notes; // default notes for new nodes 442 443 // After parsing and every bulk phase we hang onto the Root instruction. 444 // The RootNode instruction is where the whole program begins. It produces 445 // the initial Control and BOTTOM for everybody else. 446 447 // Type management 448 Arena _Compile_types; // Arena for all types 449 Arena* _type_arena; // Alias for _Compile_types except in Initialize_shared() 450 Dict* _type_dict; // Intern table 451 CloneMap _clone_map; // used for recording history of cloned nodes 452 void* _type_hwm; // Last allocation (see Type::operator new/delete) 453 size_t _type_last_size; // Last allocation size (see Type::operator new/delete) 454 ciMethod* _last_tf_m; // Cache for 455 const TypeFunc* _last_tf; // TypeFunc::make 456 AliasType** _alias_types; // List of alias types seen so far. 457 int _num_alias_types; // Logical length of _alias_types 458 int _max_alias_types; // Physical length of _alias_types 459 AliasCacheEntry _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking 460 461 // Parsing, optimization 462 PhaseGVN* _initial_gvn; // Results of parse-time PhaseGVN 463 Unique_Node_List* _for_igvn; // Initial work-list for next round of Iterative GVN 464 WarmCallInfo* _warm_calls; // Sorted work-list for heat-based inlining. 465 466 GrowableArray<CallGenerator*> _late_inlines; // List of CallGenerators to be revisited after 467 // main parsing has finished. 468 GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations 469 470 GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations 471 472 int _late_inlines_pos; // Where in the queue should the next late inlining candidate go (emulate depth first inlining) 473 uint _number_of_mh_late_inlines; // number of method handle late inlining still pending 474 475 476 // Inlining may not happen in parse order which would make 477 // PrintInlining output confusing. Keep track of PrintInlining 478 // pieces in order. 479 class PrintInliningBuffer : public ResourceObj { 480 private: 481 CallGenerator* _cg; 482 stringStream* _ss; 483 484 public: 485 PrintInliningBuffer() 486 : _cg(NULL) { _ss = new stringStream(); } 487 488 stringStream* ss() const { return _ss; } 489 CallGenerator* cg() const { return _cg; } 490 void set_cg(CallGenerator* cg) { _cg = cg; } 491 }; 492 493 stringStream* _print_inlining_stream; 494 GrowableArray<PrintInliningBuffer>* _print_inlining_list; 495 int _print_inlining_idx; 496 char* _print_inlining_output; 497 498 // Only keep nodes in the expensive node list that need to be optimized 499 void cleanup_expensive_nodes(PhaseIterGVN &igvn); 500 // Use for sorting expensive nodes to bring similar nodes together 501 static int cmp_expensive_nodes(Node** n1, Node** n2); 502 // Expensive nodes list already sorted? 503 bool expensive_nodes_sorted() const; 504 // Remove the speculative part of types and clean up the graph 505 void remove_speculative_types(PhaseIterGVN &igvn); 506 507 void* _replay_inline_data; // Pointer to data loaded from file 508 509 void print_inlining_init(); 510 void print_inlining_reinit(); 511 void print_inlining_commit(); 512 void print_inlining_push(); 513 PrintInliningBuffer& print_inlining_current(); 514 515 void log_late_inline_failure(CallGenerator* cg, const char* msg); 516 517 public: 518 519 outputStream* print_inlining_stream() const { 520 assert(print_inlining() || print_intrinsics(), "PrintInlining off?"); 521 return _print_inlining_stream; 522 } 523 524 void print_inlining_update(CallGenerator* cg); 525 void print_inlining_update_delayed(CallGenerator* cg); 526 void print_inlining_move_to(CallGenerator* cg); 527 void print_inlining_assert_ready(); 528 void print_inlining_reset(); 529 530 void print_inlining(ciMethod* method, int inline_level, int bci, const char* msg = NULL) { 531 stringStream ss; 532 CompileTask::print_inlining_inner(&ss, method, inline_level, bci, msg); 533 print_inlining_stream()->print("%s", ss.as_string()); 534 } 535 536 #ifndef PRODUCT 537 IdealGraphPrinter* printer() { return _printer; } 538 #endif 539 540 void log_late_inline(CallGenerator* cg); 541 void log_inline_id(CallGenerator* cg); 542 void log_inline_failure(const char* msg); 543 544 void* replay_inline_data() const { return _replay_inline_data; } 545 546 // Dump inlining replay data to the stream. 547 void dump_inline_data(outputStream* out); 548 549 private: 550 // Matching, CFG layout, allocation, code generation 551 PhaseCFG* _cfg; // Results of CFG finding 552 bool _select_24_bit_instr; // We selected an instruction with a 24-bit result 553 bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results 554 int _java_calls; // Number of java calls in the method 555 int _inner_loops; // Number of inner loops in the method 556 Matcher* _matcher; // Engine to map ideal to machine instructions 557 PhaseRegAlloc* _regalloc; // Results of register allocation. 558 int _frame_slots; // Size of total frame in stack slots 559 CodeOffsets _code_offsets; // Offsets into the code for various interesting entries 560 RegMask _FIRST_STACK_mask; // All stack slots usable for spills (depends on frame layout) 561 Arena* _indexSet_arena; // control IndexSet allocation within PhaseChaitin 562 void* _indexSet_free_block_list; // free list of IndexSet bit blocks 563 int _interpreter_frame_size; 564 565 uint _node_bundling_limit; 566 Bundle* _node_bundling_base; // Information for instruction bundling 567 568 // Instruction bits passed off to the VM 569 int _method_size; // Size of nmethod code segment in bytes 570 CodeBuffer _code_buffer; // Where the code is assembled 571 int _first_block_size; // Size of unvalidated entry point code / OSR poison code 572 ExceptionHandlerTable _handler_table; // Table of native-code exception handlers 573 ImplicitExceptionTable _inc_table; // Table of implicit null checks in native code 574 OopMapSet* _oop_map_set; // Table of oop maps (one for each safepoint location) 575 static int _CompiledZap_count; // counter compared against CompileZap[First/Last] 576 BufferBlob* _scratch_buffer_blob; // For temporary code buffers. 577 relocInfo* _scratch_locs_memory; // For temporary code buffers. 578 int _scratch_const_size; // For temporary code buffers. 579 bool _in_scratch_emit_size; // true when in scratch_emit_size. 580 581 public: 582 // Accessors 583 584 // The Compile instance currently active in this (compiler) thread. 585 static Compile* current() { 586 return (Compile*) ciEnv::current()->compiler_data(); 587 } 588 589 // ID for this compilation. Useful for setting breakpoints in the debugger. 590 int compile_id() const { return _compile_id; } 591 DirectiveSet* directive() const { return _directive; } 592 593 // Does this compilation allow instructions to subsume loads? User 594 // instructions that subsume a load may result in an unschedulable 595 // instruction sequence. 596 bool subsume_loads() const { return _subsume_loads; } 597 /** Do escape analysis. */ 598 bool do_escape_analysis() const { return _do_escape_analysis; } 599 /** Do boxing elimination. */ 600 bool eliminate_boxing() const { return _eliminate_boxing; } 601 /** Do aggressive boxing elimination. */ 602 bool aggressive_unboxing() const { return _eliminate_boxing && AggressiveUnboxing; } 603 bool save_argument_registers() const { return _save_argument_registers; } 604 605 606 // Other fixed compilation parameters. 607 ciMethod* method() const { return _method; } 608 int entry_bci() const { return _entry_bci; } 609 bool is_osr_compilation() const { return _entry_bci != InvocationEntryBci; } 610 bool is_method_compilation() const { return (_method != NULL && !_method->flags().is_native()); } 611 const TypeFunc* tf() const { assert(_tf!=NULL, ""); return _tf; } 612 void init_tf(const TypeFunc* tf) { assert(_tf==NULL, ""); _tf = tf; } 613 InlineTree* ilt() const { return _ilt; } 614 address stub_function() const { return _stub_function; } 615 const char* stub_name() const { return _stub_name; } 616 address stub_entry_point() const { return _stub_entry_point; } 617 618 // Control of this compilation. 619 int fixed_slots() const { assert(_fixed_slots >= 0, ""); return _fixed_slots; } 620 void set_fixed_slots(int n) { _fixed_slots = n; } 621 int major_progress() const { return _major_progress; } 622 void set_inlining_progress(bool z) { _inlining_progress = z; } 623 int inlining_progress() const { return _inlining_progress; } 624 void set_inlining_incrementally(bool z) { _inlining_incrementally = z; } 625 int inlining_incrementally() const { return _inlining_incrementally; } 626 void set_major_progress() { _major_progress++; } 627 void clear_major_progress() { _major_progress = 0; } 628 int num_loop_opts() const { return _num_loop_opts; } 629 void set_num_loop_opts(int n) { _num_loop_opts = n; } 630 int max_inline_size() const { return _max_inline_size; } 631 void set_freq_inline_size(int n) { _freq_inline_size = n; } 632 int freq_inline_size() const { return _freq_inline_size; } 633 void set_max_inline_size(int n) { _max_inline_size = n; } 634 bool has_loops() const { return _has_loops; } 635 void set_has_loops(bool z) { _has_loops = z; } 636 bool has_split_ifs() const { return _has_split_ifs; } 637 void set_has_split_ifs(bool z) { _has_split_ifs = z; } 638 bool has_unsafe_access() const { return _has_unsafe_access; } 639 void set_has_unsafe_access(bool z) { _has_unsafe_access = z; } 640 bool has_stringbuilder() const { return _has_stringbuilder; } 641 void set_has_stringbuilder(bool z) { _has_stringbuilder = z; } 642 bool has_boxed_value() const { return _has_boxed_value; } 643 void set_has_boxed_value(bool z) { _has_boxed_value = z; } 644 bool has_reserved_stack_access() const { return _has_reserved_stack_access; } 645 void set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; } 646 int max_vector_size() const { return _max_vector_size; } 647 void set_max_vector_size(int s) { _max_vector_size = s; } 648 void set_trap_count(uint r, uint c) { assert(r < trapHistLength, "oob"); _trap_hist[r] = c; } 649 uint trap_count(uint r) const { assert(r < trapHistLength, "oob"); return _trap_hist[r]; } 650 bool trap_can_recompile() const { return _trap_can_recompile; } 651 void set_trap_can_recompile(bool z) { _trap_can_recompile = z; } 652 uint decompile_count() const { return _decompile_count; } 653 void set_decompile_count(uint c) { _decompile_count = c; } 654 bool allow_range_check_smearing() const; 655 bool do_inlining() const { return _do_inlining; } 656 void set_do_inlining(bool z) { _do_inlining = z; } 657 bool do_scheduling() const { return _do_scheduling; } 658 void set_do_scheduling(bool z) { _do_scheduling = z; } 659 bool do_freq_based_layout() const{ return _do_freq_based_layout; } 660 void set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; } 661 bool do_count_invocations() const{ return _do_count_invocations; } 662 void set_do_count_invocations(bool z){ _do_count_invocations = z; } 663 bool do_method_data_update() const { return _do_method_data_update; } 664 void set_do_method_data_update(bool z) { _do_method_data_update = z; } 665 bool do_vector_loop() const { return _do_vector_loop; } 666 void set_do_vector_loop(bool z) { _do_vector_loop = z; } 667 bool use_cmove() const { return _use_cmove; } 668 void set_use_cmove(bool z) { _use_cmove = z; } 669 bool age_code() const { return _age_code; } 670 void set_age_code(bool z) { _age_code = z; } 671 int AliasLevel() const { return _AliasLevel; } 672 bool print_assembly() const { return _print_assembly; } 673 void set_print_assembly(bool z) { _print_assembly = z; } 674 bool print_inlining() const { return _print_inlining; } 675 void set_print_inlining(bool z) { _print_inlining = z; } 676 bool print_intrinsics() const { return _print_intrinsics; } 677 void set_print_intrinsics(bool z) { _print_intrinsics = z; } 678 RTMState rtm_state() const { return _rtm_state; } 679 void set_rtm_state(RTMState s) { _rtm_state = s; } 680 bool use_rtm() const { return (_rtm_state & NoRTM) == 0; } 681 bool profile_rtm() const { return _rtm_state == ProfileRTM; } 682 uint max_node_limit() const { return (uint)_max_node_limit; } 683 void set_max_node_limit(uint n) { _max_node_limit = n; } 684 685 // check the CompilerOracle for special behaviours for this compile 686 bool method_has_option(const char * option) { 687 return method() != NULL && method()->has_option(option); 688 } 689 690 #ifndef PRODUCT 691 bool trace_opto_output() const { return _trace_opto_output; } 692 bool parsed_irreducible_loop() const { return _parsed_irreducible_loop; } 693 void set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; } 694 int _in_dump_cnt; // Required for dumping ir nodes. 695 #endif 696 bool has_irreducible_loop() const { return _has_irreducible_loop; } 697 void set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; } 698 699 // JSR 292 700 bool has_method_handle_invokes() const { return _has_method_handle_invokes; } 701 void set_has_method_handle_invokes(bool z) { _has_method_handle_invokes = z; } 702 703 Ticks _latest_stage_start_counter; 704 705 void begin_method() { 706 #ifndef PRODUCT 707 if (_printer && _printer->should_print(1)) { 708 _printer->begin_method(); 709 } 710 #endif 711 C->_latest_stage_start_counter.stamp(); 712 } 713 714 void print_method(CompilerPhaseType cpt, int level = 1) { 715 EventCompilerPhase event; 716 if (event.should_commit()) { 717 event.set_starttime(C->_latest_stage_start_counter); 718 event.set_phase((u1) cpt); 719 event.set_compileID(C->_compile_id); 720 event.set_phaseLevel(level); 721 event.commit(); 722 } 723 724 725 #ifndef PRODUCT 726 if (_printer && _printer->should_print(level)) { 727 _printer->print_method(CompilerPhaseTypeHelper::to_string(cpt), level); 728 } 729 #endif 730 C->_latest_stage_start_counter.stamp(); 731 } 732 733 void end_method(int level = 1) { 734 EventCompilerPhase event; 735 if (event.should_commit()) { 736 event.set_starttime(C->_latest_stage_start_counter); 737 event.set_phase((u1) PHASE_END); 738 event.set_compileID(C->_compile_id); 739 event.set_phaseLevel(level); 740 event.commit(); 741 } 742 #ifndef PRODUCT 743 if (_printer && _printer->should_print(level)) { 744 _printer->end_method(); 745 } 746 #endif 747 } 748 749 int macro_count() const { return _macro_nodes->length(); } 750 int predicate_count() const { return _predicate_opaqs->length();} 751 int expensive_count() const { return _expensive_nodes->length(); } 752 Node* macro_node(int idx) const { return _macro_nodes->at(idx); } 753 Node* predicate_opaque1_node(int idx) const { return _predicate_opaqs->at(idx);} 754 Node* expensive_node(int idx) const { return _expensive_nodes->at(idx); } 755 ConnectionGraph* congraph() { return _congraph;} 756 void set_congraph(ConnectionGraph* congraph) { _congraph = congraph;} 757 void add_macro_node(Node * n) { 758 //assert(n->is_macro(), "must be a macro node"); 759 assert(!_macro_nodes->contains(n), "duplicate entry in expand list"); 760 _macro_nodes->append(n); 761 } 762 void remove_macro_node(Node * n) { 763 // this function may be called twice for a node so check 764 // that the node is in the array before attempting to remove it 765 if (_macro_nodes->contains(n)) 766 _macro_nodes->remove(n); 767 // remove from _predicate_opaqs list also if it is there 768 if (predicate_count() > 0 && _predicate_opaqs->contains(n)){ 769 _predicate_opaqs->remove(n); 770 } 771 } 772 void add_expensive_node(Node * n); 773 void remove_expensive_node(Node * n) { 774 if (_expensive_nodes->contains(n)) { 775 _expensive_nodes->remove(n); 776 } 777 } 778 void add_predicate_opaq(Node * n) { 779 assert(!_predicate_opaqs->contains(n), "duplicate entry in predicate opaque1"); 780 assert(_macro_nodes->contains(n), "should have already been in macro list"); 781 _predicate_opaqs->append(n); 782 } 783 784 #ifdef _LP64 785 // Range check dependent CastII nodes that can be removed after loop optimizations 786 void add_range_check_cast(Node* n); 787 void remove_range_check_cast(Node* n) { 788 if (_range_check_casts->contains(n)) { 789 _range_check_casts->remove(n); 790 } 791 } 792 Node* range_check_cast_node(int idx) const { return _range_check_casts->at(idx); } 793 int range_check_cast_count() const { return _range_check_casts->length(); } 794 // Remove all range check dependent CastIINodes. 795 void remove_range_check_casts(PhaseIterGVN &igvn); 796 #endif 797 798 // remove the opaque nodes that protect the predicates so that the unused checks and 799 // uncommon traps will be eliminated from the graph. 800 void cleanup_loop_predicates(PhaseIterGVN &igvn); 801 bool is_predicate_opaq(Node * n) { 802 return _predicate_opaqs->contains(n); 803 } 804 805 // Are there candidate expensive nodes for optimization? 806 bool should_optimize_expensive_nodes(PhaseIterGVN &igvn); 807 // Check whether n1 and n2 are similar 808 static int cmp_expensive_nodes(Node* n1, Node* n2); 809 // Sort expensive nodes to locate similar expensive nodes 810 void sort_expensive_nodes(); 811 812 // Compilation environment. 813 Arena* comp_arena() { return &_comp_arena; } 814 ciEnv* env() const { return _env; } 815 CompileLog* log() const { return _log; } 816 bool failing() const { return _env->failing() || _failure_reason != NULL; } 817 const char* failure_reason() const { return (_env->failing()) ? _env->failure_reason() : _failure_reason; } 818 819 bool failure_reason_is(const char* r) const { 820 return (r == _failure_reason) || (r != NULL && _failure_reason != NULL && strcmp(r, _failure_reason) == 0); 821 } 822 823 void record_failure(const char* reason); 824 void record_method_not_compilable(const char* reason, bool all_tiers = false) { 825 // All bailouts cover "all_tiers" when TieredCompilation is off. 826 if (!TieredCompilation) all_tiers = true; 827 env()->record_method_not_compilable(reason, all_tiers); 828 // Record failure reason. 829 record_failure(reason); 830 } 831 void record_method_not_compilable_all_tiers(const char* reason) { 832 record_method_not_compilable(reason, true); 833 } 834 bool check_node_count(uint margin, const char* reason) { 835 if (live_nodes() + margin > max_node_limit()) { 836 record_method_not_compilable(reason); 837 return true; 838 } else { 839 return false; 840 } 841 } 842 843 // Node management 844 uint unique() const { return _unique; } 845 uint next_unique() { return _unique++; } 846 void set_unique(uint i) { _unique = i; } 847 static int debug_idx() { return debug_only(_debug_idx)+0; } 848 static void set_debug_idx(int i) { debug_only(_debug_idx = i); } 849 Arena* node_arena() { return &_node_arena; } 850 Arena* old_arena() { return &_old_arena; } 851 RootNode* root() const { return _root; } 852 void set_root(RootNode* r) { _root = r; } 853 StartNode* start() const; // (Derived from root.) 854 void init_start(StartNode* s); 855 Node* immutable_memory(); 856 857 Node* recent_alloc_ctl() const { return _recent_alloc_ctl; } 858 Node* recent_alloc_obj() const { return _recent_alloc_obj; } 859 void set_recent_alloc(Node* ctl, Node* obj) { 860 _recent_alloc_ctl = ctl; 861 _recent_alloc_obj = obj; 862 } 863 void record_dead_node(uint idx) { if (_dead_node_list.test_set(idx)) return; 864 _dead_node_count++; 865 } 866 bool is_dead_node(uint idx) { return _dead_node_list.test(idx) != 0; } 867 uint dead_node_count() { return _dead_node_count; } 868 void reset_dead_node_list() { _dead_node_list.Reset(); 869 _dead_node_count = 0; 870 } 871 uint live_nodes() const { 872 int val = _unique - _dead_node_count; 873 assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count); 874 return (uint) val; 875 } 876 #ifdef ASSERT 877 uint count_live_nodes_by_graph_walk(); 878 void print_missing_nodes(); 879 #endif 880 881 // Record modified nodes to check that they are put on IGVN worklist 882 void record_modified_node(Node* n) NOT_DEBUG_RETURN; 883 void remove_modified_node(Node* n) NOT_DEBUG_RETURN; 884 DEBUG_ONLY( Unique_Node_List* modified_nodes() const { return _modified_nodes; } ) 885 886 // Constant table 887 ConstantTable& constant_table() { return _constant_table; } 888 889 MachConstantBaseNode* mach_constant_base_node(); 890 bool has_mach_constant_base_node() const { return _mach_constant_base_node != NULL; } 891 // Generated by adlc, true if CallNode requires MachConstantBase. 892 bool needs_clone_jvms(); 893 894 // Handy undefined Node 895 Node* top() const { return _top; } 896 897 // these are used by guys who need to know about creation and transformation of top: 898 Node* cached_top_node() { return _top; } 899 void set_cached_top_node(Node* tn); 900 901 GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; } 902 void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; } 903 Node_Notes* default_node_notes() const { return _default_node_notes; } 904 void set_default_node_notes(Node_Notes* n) { _default_node_notes = n; } 905 906 Node_Notes* node_notes_at(int idx) { 907 return locate_node_notes(_node_note_array, idx, false); 908 } 909 inline bool set_node_notes_at(int idx, Node_Notes* value); 910 911 // Copy notes from source to dest, if they exist. 912 // Overwrite dest only if source provides something. 913 // Return true if information was moved. 914 bool copy_node_notes_to(Node* dest, Node* source); 915 916 // Workhorse function to sort out the blocked Node_Notes array: 917 inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr, 918 int idx, bool can_grow = false); 919 920 void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by); 921 922 // Type management 923 Arena* type_arena() { return _type_arena; } 924 Dict* type_dict() { return _type_dict; } 925 void* type_hwm() { return _type_hwm; } 926 size_t type_last_size() { return _type_last_size; } 927 int num_alias_types() { return _num_alias_types; } 928 929 void init_type_arena() { _type_arena = &_Compile_types; } 930 void set_type_arena(Arena* a) { _type_arena = a; } 931 void set_type_dict(Dict* d) { _type_dict = d; } 932 void set_type_hwm(void* p) { _type_hwm = p; } 933 void set_type_last_size(size_t sz) { _type_last_size = sz; } 934 935 const TypeFunc* last_tf(ciMethod* m) { 936 return (m == _last_tf_m) ? _last_tf : NULL; 937 } 938 void set_last_tf(ciMethod* m, const TypeFunc* tf) { 939 assert(m != NULL || tf == NULL, ""); 940 _last_tf_m = m; 941 _last_tf = tf; 942 } 943 944 AliasType* alias_type(int idx) { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; } 945 AliasType* alias_type(const TypePtr* adr_type, ciField* field = NULL) { return find_alias_type(adr_type, false, field); } 946 bool have_alias_type(const TypePtr* adr_type); 947 AliasType* alias_type(ciField* field); 948 949 int get_alias_index(const TypePtr* at) { return alias_type(at)->index(); } 950 const TypePtr* get_adr_type(uint aidx) { return alias_type(aidx)->adr_type(); } 951 int get_general_index(uint aidx) { return alias_type(aidx)->general_index(); } 952 953 // Building nodes 954 void rethrow_exceptions(JVMState* jvms); 955 void return_values(JVMState* jvms); 956 JVMState* build_start_state(StartNode* start, const TypeFunc* tf); 957 958 // Decide how to build a call. 959 // The profile factor is a discount to apply to this site's interp. profile. 960 CallGenerator* call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch, 961 JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = NULL, 962 bool allow_intrinsics = true, bool delayed_forbidden = false); 963 bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) { 964 return should_delay_string_inlining(call_method, jvms) || 965 should_delay_boxing_inlining(call_method, jvms); 966 } 967 bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms); 968 bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms); 969 970 // Helper functions to identify inlining potential at call-site 971 ciMethod* optimize_virtual_call(ciMethod* caller, int bci, ciInstanceKlass* klass, 972 ciKlass* holder, ciMethod* callee, 973 const TypeOopPtr* receiver_type, bool is_virtual, 974 bool &call_does_dispatch, int &vtable_index, 975 bool check_access = true); 976 ciMethod* optimize_inlining(ciMethod* caller, int bci, ciInstanceKlass* klass, 977 ciMethod* callee, const TypeOopPtr* receiver_type, 978 bool check_access = true); 979 980 // Report if there were too many traps at a current method and bci. 981 // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded. 982 // If there is no MDO at all, report no trap unless told to assume it. 983 bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason); 984 // This version, unspecific to a particular bci, asks if 985 // PerMethodTrapLimit was exceeded for all inlined methods seen so far. 986 bool too_many_traps(Deoptimization::DeoptReason reason, 987 // Privately used parameter for logging: 988 ciMethodData* logmd = NULL); 989 // Report if there were too many recompiles at a method and bci. 990 bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason); 991 // Return a bitset with the reasons where deoptimization is allowed, 992 // i.e., where there were not too many uncommon traps. 993 int _allowed_reasons; 994 int allowed_deopt_reasons() { return _allowed_reasons; } 995 void set_allowed_deopt_reasons(); 996 997 // Parsing, optimization 998 PhaseGVN* initial_gvn() { return _initial_gvn; } 999 Unique_Node_List* for_igvn() { return _for_igvn; } 1000 inline void record_for_igvn(Node* n); // Body is after class Unique_Node_List. 1001 void set_initial_gvn(PhaseGVN *gvn) { _initial_gvn = gvn; } 1002 void set_for_igvn(Unique_Node_List *for_igvn) { _for_igvn = for_igvn; } 1003 1004 // Replace n by nn using initial_gvn, calling hash_delete and 1005 // record_for_igvn as needed. 1006 void gvn_replace_by(Node* n, Node* nn); 1007 1008 1009 void identify_useful_nodes(Unique_Node_List &useful); 1010 void update_dead_node_list(Unique_Node_List &useful); 1011 void remove_useless_nodes (Unique_Node_List &useful); 1012 1013 WarmCallInfo* warm_calls() const { return _warm_calls; } 1014 void set_warm_calls(WarmCallInfo* l) { _warm_calls = l; } 1015 WarmCallInfo* pop_warm_call(); 1016 1017 // Record this CallGenerator for inlining at the end of parsing. 1018 void add_late_inline(CallGenerator* cg) { 1019 _late_inlines.insert_before(_late_inlines_pos, cg); 1020 _late_inlines_pos++; 1021 } 1022 1023 void prepend_late_inline(CallGenerator* cg) { 1024 _late_inlines.insert_before(0, cg); 1025 } 1026 1027 void add_string_late_inline(CallGenerator* cg) { 1028 _string_late_inlines.push(cg); 1029 } 1030 1031 void add_boxing_late_inline(CallGenerator* cg) { 1032 _boxing_late_inlines.push(cg); 1033 } 1034 1035 void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful); 1036 1037 void process_print_inlining(); 1038 void dump_print_inlining(); 1039 1040 bool over_inlining_cutoff() const { 1041 if (!inlining_incrementally()) { 1042 return unique() > (uint)NodeCountInliningCutoff; 1043 } else { 1044 return live_nodes() > (uint)LiveNodeCountInliningCutoff; 1045 } 1046 } 1047 1048 void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; } 1049 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--; } 1050 bool has_mh_late_inlines() const { return _number_of_mh_late_inlines > 0; } 1051 1052 void inline_incrementally_one(PhaseIterGVN& igvn); 1053 void inline_incrementally(PhaseIterGVN& igvn); 1054 void inline_string_calls(bool parse_time); 1055 void inline_boxing_calls(PhaseIterGVN& igvn); 1056 1057 // Matching, CFG layout, allocation, code generation 1058 PhaseCFG* cfg() { return _cfg; } 1059 bool select_24_bit_instr() const { return _select_24_bit_instr; } 1060 bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; } 1061 bool has_java_calls() const { return _java_calls > 0; } 1062 int java_calls() const { return _java_calls; } 1063 int inner_loops() const { return _inner_loops; } 1064 Matcher* matcher() { return _matcher; } 1065 PhaseRegAlloc* regalloc() { return _regalloc; } 1066 int frame_slots() const { return _frame_slots; } 1067 int frame_size_in_words() const; // frame_slots in units of the polymorphic 'words' 1068 int frame_size_in_bytes() const { return _frame_slots << LogBytesPerInt; } 1069 RegMask& FIRST_STACK_mask() { return _FIRST_STACK_mask; } 1070 Arena* indexSet_arena() { return _indexSet_arena; } 1071 void* indexSet_free_block_list() { return _indexSet_free_block_list; } 1072 uint node_bundling_limit() { return _node_bundling_limit; } 1073 Bundle* node_bundling_base() { return _node_bundling_base; } 1074 void set_node_bundling_limit(uint n) { _node_bundling_limit = n; } 1075 void set_node_bundling_base(Bundle* b) { _node_bundling_base = b; } 1076 bool starts_bundle(const Node *n) const; 1077 bool need_stack_bang(int frame_size_in_bytes) const; 1078 bool need_register_stack_bang() const; 1079 1080 void update_interpreter_frame_size(int size) { 1081 if (_interpreter_frame_size < size) { 1082 _interpreter_frame_size = size; 1083 } 1084 } 1085 int bang_size_in_bytes() const; 1086 1087 void set_matcher(Matcher* m) { _matcher = m; } 1088 //void set_regalloc(PhaseRegAlloc* ra) { _regalloc = ra; } 1089 void set_indexSet_arena(Arena* a) { _indexSet_arena = a; } 1090 void set_indexSet_free_block_list(void* p) { _indexSet_free_block_list = p; } 1091 1092 // Remember if this compilation changes hardware mode to 24-bit precision 1093 void set_24_bit_selection_and_mode(bool selection, bool mode) { 1094 _select_24_bit_instr = selection; 1095 _in_24_bit_fp_mode = mode; 1096 } 1097 1098 void set_java_calls(int z) { _java_calls = z; } 1099 void set_inner_loops(int z) { _inner_loops = z; } 1100 1101 // Instruction bits passed off to the VM 1102 int code_size() { return _method_size; } 1103 CodeBuffer* code_buffer() { return &_code_buffer; } 1104 int first_block_size() { return _first_block_size; } 1105 void set_frame_complete(int off) { _code_offsets.set_value(CodeOffsets::Frame_Complete, off); } 1106 ExceptionHandlerTable* handler_table() { return &_handler_table; } 1107 ImplicitExceptionTable* inc_table() { return &_inc_table; } 1108 OopMapSet* oop_map_set() { return _oop_map_set; } 1109 DebugInformationRecorder* debug_info() { return env()->debug_info(); } 1110 Dependencies* dependencies() { return env()->dependencies(); } 1111 static int CompiledZap_count() { return _CompiledZap_count; } 1112 BufferBlob* scratch_buffer_blob() { return _scratch_buffer_blob; } 1113 void init_scratch_buffer_blob(int const_size); 1114 void clear_scratch_buffer_blob(); 1115 void set_scratch_buffer_blob(BufferBlob* b) { _scratch_buffer_blob = b; } 1116 relocInfo* scratch_locs_memory() { return _scratch_locs_memory; } 1117 void set_scratch_locs_memory(relocInfo* b) { _scratch_locs_memory = b; } 1118 1119 // emit to scratch blob, report resulting size 1120 uint scratch_emit_size(const Node* n); 1121 void set_in_scratch_emit_size(bool x) { _in_scratch_emit_size = x; } 1122 bool in_scratch_emit_size() const { return _in_scratch_emit_size; } 1123 1124 enum ScratchBufferBlob { 1125 MAX_inst_size = 1024, 1126 MAX_locs_size = 128, // number of relocInfo elements 1127 MAX_const_size = 128, 1128 MAX_stubs_size = 128 1129 }; 1130 1131 // Major entry point. Given a Scope, compile the associated method. 1132 // For normal compilations, entry_bci is InvocationEntryBci. For on stack 1133 // replacement, entry_bci indicates the bytecode for which to compile a 1134 // continuation. 1135 Compile(ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, 1136 int entry_bci, bool subsume_loads, bool do_escape_analysis, 1137 bool eliminate_boxing, DirectiveSet* directive); 1138 1139 // Second major entry point. From the TypeFunc signature, generate code 1140 // to pass arguments from the Java calling convention to the C calling 1141 // convention. 1142 Compile(ciEnv* ci_env, const TypeFunc *(*gen)(), 1143 address stub_function, const char *stub_name, 1144 int is_fancy_jump, bool pass_tls, 1145 bool save_arg_registers, bool return_pc, DirectiveSet* directive); 1146 1147 // From the TypeFunc signature, generate code to pass arguments 1148 // from Compiled calling convention to Interpreter's calling convention 1149 void Generate_Compiled_To_Interpreter_Graph(const TypeFunc *tf, address interpreter_entry); 1150 1151 // From the TypeFunc signature, generate code to pass arguments 1152 // from Interpreter's calling convention to Compiler's calling convention 1153 void Generate_Interpreter_To_Compiled_Graph(const TypeFunc *tf); 1154 1155 // Are we compiling a method? 1156 bool has_method() { return method() != NULL; } 1157 1158 // Maybe print some information about this compile. 1159 void print_compile_messages(); 1160 1161 // Final graph reshaping, a post-pass after the regular optimizer is done. 1162 bool final_graph_reshaping(); 1163 1164 // returns true if adr is completely contained in the given alias category 1165 bool must_alias(const TypePtr* adr, int alias_idx); 1166 1167 // returns true if adr overlaps with the given alias category 1168 bool can_alias(const TypePtr* adr, int alias_idx); 1169 1170 // Driver for converting compiler's IR into machine code bits 1171 void Output(); 1172 1173 // Accessors for node bundling info. 1174 Bundle* node_bundling(const Node *n); 1175 bool valid_bundle_info(const Node *n); 1176 1177 // Schedule and Bundle the instructions 1178 void ScheduleAndBundle(); 1179 1180 // Build OopMaps for each GC point 1181 void BuildOopMaps(); 1182 1183 // Append debug info for the node "local" at safepoint node "sfpt" to the 1184 // "array", May also consult and add to "objs", which describes the 1185 // scalar-replaced objects. 1186 void FillLocArray( int idx, MachSafePointNode* sfpt, 1187 Node *local, GrowableArray<ScopeValue*> *array, 1188 GrowableArray<ScopeValue*> *objs ); 1189 1190 // If "objs" contains an ObjectValue whose id is "id", returns it, else NULL. 1191 static ObjectValue* sv_for_node_id(GrowableArray<ScopeValue*> *objs, int id); 1192 // Requres that "objs" does not contains an ObjectValue whose id matches 1193 // that of "sv. Appends "sv". 1194 static void set_sv_for_object_node(GrowableArray<ScopeValue*> *objs, 1195 ObjectValue* sv ); 1196 1197 // Process an OopMap Element while emitting nodes 1198 void Process_OopMap_Node(MachNode *mach, int code_offset); 1199 1200 // Initialize code buffer 1201 CodeBuffer* init_buffer(uint* blk_starts); 1202 1203 // Write out basic block data to code buffer 1204 void fill_buffer(CodeBuffer* cb, uint* blk_starts); 1205 1206 // Determine which variable sized branches can be shortened 1207 void shorten_branches(uint* blk_starts, int& code_size, int& reloc_size, int& stub_size); 1208 1209 // Compute the size of first NumberOfLoopInstrToAlign instructions 1210 // at the head of a loop. 1211 void compute_loop_first_inst_sizes(); 1212 1213 // Compute the information for the exception tables 1214 void FillExceptionTables(uint cnt, uint *call_returns, uint *inct_starts, Label *blk_labels); 1215 1216 // Stack slots that may be unused by the calling convention but must 1217 // otherwise be preserved. On Intel this includes the return address. 1218 // On PowerPC it includes the 4 words holding the old TOC & LR glue. 1219 uint in_preserve_stack_slots(); 1220 1221 // "Top of Stack" slots that may be unused by the calling convention but must 1222 // otherwise be preserved. 1223 // On Intel these are not necessary and the value can be zero. 1224 // On Sparc this describes the words reserved for storing a register window 1225 // when an interrupt occurs. 1226 static uint out_preserve_stack_slots(); 1227 1228 // Number of outgoing stack slots killed above the out_preserve_stack_slots 1229 // for calls to C. Supports the var-args backing area for register parms. 1230 uint varargs_C_out_slots_killed() const; 1231 1232 // Number of Stack Slots consumed by a synchronization entry 1233 int sync_stack_slots() const; 1234 1235 // Compute the name of old_SP. See <arch>.ad for frame layout. 1236 OptoReg::Name compute_old_SP(); 1237 1238 private: 1239 // Phase control: 1240 void Init(int aliaslevel); // Prepare for a single compilation 1241 int Inline_Warm(); // Find more inlining work. 1242 void Finish_Warm(); // Give up on further inlines. 1243 void Optimize(); // Given a graph, optimize it 1244 void Code_Gen(); // Generate code from a graph 1245 1246 // Management of the AliasType table. 1247 void grow_alias_types(); 1248 AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type); 1249 const TypePtr *flatten_alias_type(const TypePtr* adr_type) const; 1250 AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field); 1251 1252 void verify_top(Node*) const PRODUCT_RETURN; 1253 1254 // Intrinsic setup. 1255 void register_library_intrinsics(); // initializer 1256 CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual); // constructor 1257 int intrinsic_insertion_index(ciMethod* m, bool is_virtual); // helper 1258 CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual); // query fn 1259 void register_intrinsic(CallGenerator* cg); // update fn 1260 1261 #ifndef PRODUCT 1262 static juint _intrinsic_hist_count[vmIntrinsics::ID_LIMIT]; 1263 static jubyte _intrinsic_hist_flags[vmIntrinsics::ID_LIMIT]; 1264 #endif 1265 // Function calls made by the public function final_graph_reshaping. 1266 // No need to be made public as they are not called elsewhere. 1267 void final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &frc); 1268 void final_graph_reshaping_walk( Node_Stack &nstack, Node *root, Final_Reshape_Counts &frc ); 1269 void eliminate_redundant_card_marks(Node* n); 1270 1271 public: 1272 1273 // Note: Histogram array size is about 1 Kb. 1274 enum { // flag bits: 1275 _intrinsic_worked = 1, // succeeded at least once 1276 _intrinsic_failed = 2, // tried it but it failed 1277 _intrinsic_disabled = 4, // was requested but disabled (e.g., -XX:-InlineUnsafeOps) 1278 _intrinsic_virtual = 8, // was seen in the virtual form (rare) 1279 _intrinsic_both = 16 // was seen in the non-virtual form (usual) 1280 }; 1281 // Update histogram. Return boolean if this is a first-time occurrence. 1282 static bool gather_intrinsic_statistics(vmIntrinsics::ID id, 1283 bool is_virtual, int flags) PRODUCT_RETURN0; 1284 static void print_intrinsic_statistics() PRODUCT_RETURN; 1285 1286 // Graph verification code 1287 // Walk the node list, verifying that there is a one-to-one 1288 // correspondence between Use-Def edges and Def-Use edges 1289 // The option no_dead_code enables stronger checks that the 1290 // graph is strongly connected from root in both directions. 1291 void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN; 1292 1293 // Verify GC barrier patterns 1294 void verify_barriers() PRODUCT_RETURN; 1295 1296 // End-of-run dumps. 1297 static void print_statistics() PRODUCT_RETURN; 1298 1299 // Dump formatted assembly 1300 void dump_asm(int *pcs = NULL, uint pc_limit = 0) PRODUCT_RETURN; 1301 void dump_pc(int *pcs, int pc_limit, Node *n); 1302 1303 // Verify ADLC assumptions during startup 1304 static void adlc_verification() PRODUCT_RETURN; 1305 1306 // Definitions of pd methods 1307 static void pd_compiler2_init(); 1308 1309 // Static parse-time type checking logic for gen_subtype_check: 1310 enum { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test }; 1311 int static_subtype_check(ciKlass* superk, ciKlass* subk); 1312 1313 static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype, 1314 // Optional control dependency (for example, on range check) 1315 Node* ctrl = NULL); 1316 1317 #ifdef _LP64 1318 // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check) 1319 static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl); 1320 #endif 1321 1322 // Auxiliary method for randomized fuzzing/stressing 1323 static bool randomized_select(int count); 1324 1325 // supporting clone_map 1326 CloneMap& clone_map(); 1327 void set_clone_map(Dict* d); 1328 1329 }; 1330 1331 #endif // SHARE_VM_OPTO_COMPILE_HPP