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