1 /*
   2  * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #ifndef SHARE_VM_OPTO_COMPILE_HPP
  26 #define SHARE_VM_OPTO_COMPILE_HPP
  27 
  28 #include "asm/codeBuffer.hpp"
  29 #include "ci/compilerInterface.hpp"
  30 #include "code/debugInfoRec.hpp"
  31 #include "code/exceptionHandlerTable.hpp"
  32 #include "compiler/compilerOracle.hpp"
  33 #include "compiler/compileBroker.hpp"
  34 #include "libadt/dict.hpp"
  35 #include "libadt/vectset.hpp"
  36 #include "jfr/jfrEvents.hpp"
  37 #include "memory/resourceArea.hpp"
  38 #include "oops/methodData.hpp"
  39 #include "opto/idealGraphPrinter.hpp"
  40 #include "opto/phasetype.hpp"
  41 #include "opto/phase.hpp"
  42 #include "opto/regmask.hpp"
  43 #include "runtime/deoptimization.hpp"
  44 #include "runtime/timerTrace.hpp"
  45 #include "runtime/vmThread.hpp"
  46 #include "utilities/ticks.hpp"
  47 
  48 class AddPNode;
  49 class Block;
  50 class Bundle;
  51 class C2Compiler;
  52 class CallGenerator;
  53 class CloneMap;
  54 class ConnectionGraph;
  55 class InlineTree;
  56 class Int_Array;
  57 class LoadBarrierNode;
  58 class Matcher;
  59 class MachConstantNode;
  60 class MachConstantBaseNode;
  61 class MachNode;
  62 class MachOper;
  63 class MachSafePointNode;
  64 class Node;
  65 class Node_Array;
  66 class Node_Notes;
  67 class NodeCloneInfo;
  68 class OptoReg;
  69 class PhaseCFG;
  70 class PhaseGVN;
  71 class PhaseIterGVN;
  72 class PhaseRegAlloc;
  73 class PhaseCCP;
  74 class PhaseCCP_DCE;
  75 class RootNode;
  76 class relocInfo;
  77 class Scope;
  78 class StartNode;
  79 class SafePointNode;
  80 class JVMState;
  81 class Type;
  82 class TypeData;
  83 class TypeInt;
  84 class TypePtr;
  85 class TypeOopPtr;
  86 class TypeFunc;
  87 class Unique_Node_List;
  88 class nmethod;
  89 class WarmCallInfo;
  90 class Node_Stack;
  91 struct Final_Reshape_Counts;
  92 
  93 enum LoopOptsMode {
  94   LoopOptsDefault,
  95   LoopOptsNone,
  96   LoopOptsSkipSplitIf,
  97   LoopOptsVerify,
  98   LoopOptsLastRound
  99 };
 100 
 101 typedef unsigned int node_idx_t;
 102 class NodeCloneInfo {
 103  private:
 104   uint64_t _idx_clone_orig;
 105  public:
 106 
 107   void set_idx(node_idx_t idx) {
 108     _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx;
 109   }
 110   node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); }
 111 
 112   void set_gen(int generation) {
 113     uint64_t g = (uint64_t)generation << 32;
 114     _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g;
 115   }
 116   int gen() const { return (int)(_idx_clone_orig >> 32); }
 117 
 118   void set(uint64_t x) { _idx_clone_orig = x; }
 119   void set(node_idx_t x, int g) { set_idx(x); set_gen(g); }
 120   uint64_t get() const { return _idx_clone_orig; }
 121 
 122   NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {}
 123   NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); }
 124 
 125   void dump() const;
 126 };
 127 
 128 class CloneMap {
 129   friend class Compile;
 130  private:
 131   bool      _debug;
 132   Dict*     _dict;
 133   int       _clone_idx;   // current cloning iteration/generation in loop unroll
 134  public:
 135   void*     _2p(node_idx_t key)   const          { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy
 136   node_idx_t _2_node_idx_t(const void* k) const  { return (node_idx_t)(intptr_t)k; }
 137   Dict*     dict()                const          { return _dict; }
 138   void insert(node_idx_t key, uint64_t val)      { assert(_dict->operator[](_2p(key)) == NULL, "key existed"); _dict->Insert(_2p(key), (void*)val); }
 139   void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); }
 140   void remove(node_idx_t key)                    { _dict->Delete(_2p(key)); }
 141   uint64_t value(node_idx_t key)  const          { return (uint64_t)_dict->operator[](_2p(key)); }
 142   node_idx_t idx(node_idx_t key)  const          { return NodeCloneInfo(value(key)).idx(); }
 143   int gen(node_idx_t key)         const          { return NodeCloneInfo(value(key)).gen(); }
 144   int gen(const void* k)          const          { return gen(_2_node_idx_t(k)); }
 145   int max_gen()                   const;
 146   void clone(Node* old, Node* nnn, int gen);
 147   void verify_insert_and_clone(Node* old, Node* nnn, int gen);
 148   void dump(node_idx_t key)       const;
 149 
 150   int  clone_idx() const                         { return _clone_idx; }
 151   void set_clone_idx(int x)                      { _clone_idx = x; }
 152   bool is_debug()                 const          { return _debug; }
 153   void set_debug(bool debug)                     { _debug = debug; }
 154   static const char* debug_option_name;
 155 
 156   bool same_idx(node_idx_t k1, node_idx_t k2)  const { return idx(k1) == idx(k2); }
 157   bool same_gen(node_idx_t k1, node_idx_t k2)  const { return gen(k1) == gen(k2); }
 158 };
 159 
 160 //------------------------------Compile----------------------------------------
 161 // This class defines a top-level Compiler invocation.
 162 
 163 class Compile : public Phase {
 164   friend class VMStructs;
 165 
 166  public:
 167   // Fixed alias indexes.  (See also MergeMemNode.)
 168   enum {
 169     AliasIdxTop = 1,  // pseudo-index, aliases to nothing (used as sentinel value)
 170     AliasIdxBot = 2,  // pseudo-index, aliases to everything
 171     AliasIdxRaw = 3   // hard-wired index for TypeRawPtr::BOTTOM
 172   };
 173 
 174   // Variant of TraceTime(NULL, &_t_accumulator, CITime);
 175   // Integrated with logging.  If logging is turned on, and CITimeVerbose is true,
 176   // then brackets are put into the log, with time stamps and node counts.
 177   // (The time collection itself is always conditionalized on CITime.)
 178   class TracePhase : public TraceTime {
 179    private:
 180     Compile*    C;
 181     CompileLog* _log;
 182     const char* _phase_name;
 183     bool _dolog;
 184    public:
 185     TracePhase(const char* name, elapsedTimer* accumulator);
 186     ~TracePhase();
 187   };
 188 
 189   // Information per category of alias (memory slice)
 190   class AliasType {
 191    private:
 192     friend class Compile;
 193 
 194     int             _index;         // unique index, used with MergeMemNode
 195     const TypePtr*  _adr_type;      // normalized address type
 196     ciField*        _field;         // relevant instance field, or null if none
 197     const Type*     _element;       // relevant array element type, or null if none
 198     bool            _is_rewritable; // false if the memory is write-once only
 199     int             _general_index; // if this is type is an instance, the general
 200                                     // type that this is an instance of
 201 
 202     void Init(int i, const TypePtr* at);
 203 
 204    public:
 205     int             index()         const { return _index; }
 206     const TypePtr*  adr_type()      const { return _adr_type; }
 207     ciField*        field()         const { return _field; }
 208     const Type*     element()       const { return _element; }
 209     bool            is_rewritable() const { return _is_rewritable; }
 210     bool            is_volatile()   const { return (_field ? _field->is_volatile() : false); }
 211     int             general_index() const { return (_general_index != 0) ? _general_index : _index; }
 212 
 213     void set_rewritable(bool z) { _is_rewritable = z; }
 214     void set_field(ciField* f) {
 215       assert(!_field,"");
 216       _field = f;
 217       if (f->is_final() || f->is_stable()) {
 218         // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
 219         _is_rewritable = false;
 220       }
 221     }
 222     void set_element(const Type* e) {
 223       assert(_element == NULL, "");
 224       _element = e;
 225     }
 226 
 227     BasicType basic_type() const;
 228 
 229     void print_on(outputStream* st) PRODUCT_RETURN;
 230   };
 231 
 232   enum {
 233     logAliasCacheSize = 6,
 234     AliasCacheSize = (1<<logAliasCacheSize)
 235   };
 236   struct AliasCacheEntry { const TypePtr* _adr_type; int _index; };  // simple duple type
 237   enum {
 238     trapHistLength = MethodData::_trap_hist_limit
 239   };
 240 
 241   // Constant entry of the constant table.
 242   class Constant {
 243   private:
 244     BasicType _type;
 245     union {
 246       jvalue    _value;
 247       Metadata* _metadata;
 248     } _v;
 249     int       _offset;         // offset of this constant (in bytes) relative to the constant table base.
 250     float     _freq;
 251     bool      _can_be_reused;  // true (default) if the value can be shared with other users.
 252 
 253   public:
 254     Constant() : _type(T_ILLEGAL), _offset(-1), _freq(0.0f), _can_be_reused(true) { _v._value.l = 0; }
 255     Constant(BasicType type, jvalue value, float freq = 0.0f, bool can_be_reused = true) :
 256       _type(type),
 257       _offset(-1),
 258       _freq(freq),
 259       _can_be_reused(can_be_reused)
 260     {
 261       assert(type != T_METADATA, "wrong constructor");
 262       _v._value = value;
 263     }
 264     Constant(Metadata* metadata, bool can_be_reused = true) :
 265       _type(T_METADATA),
 266       _offset(-1),
 267       _freq(0.0f),
 268       _can_be_reused(can_be_reused)
 269     {
 270       _v._metadata = metadata;
 271     }
 272 
 273     bool operator==(const Constant& other);
 274 
 275     BasicType type()      const    { return _type; }
 276 
 277     jint    get_jint()    const    { return _v._value.i; }
 278     jlong   get_jlong()   const    { return _v._value.j; }
 279     jfloat  get_jfloat()  const    { return _v._value.f; }
 280     jdouble get_jdouble() const    { return _v._value.d; }
 281     jobject get_jobject() const    { return _v._value.l; }
 282 
 283     Metadata* get_metadata() const { return _v._metadata; }
 284 
 285     int         offset()  const    { return _offset; }
 286     void    set_offset(int offset) {        _offset = offset; }
 287 
 288     float       freq()    const    { return _freq;         }
 289     void    inc_freq(float freq)   {        _freq += freq; }
 290 
 291     bool    can_be_reused() const  { return _can_be_reused; }
 292   };
 293 
 294   // Constant table.
 295   class ConstantTable {
 296   private:
 297     GrowableArray<Constant> _constants;          // Constants of this table.
 298     int                     _size;               // Size in bytes the emitted constant table takes (including padding).
 299     int                     _table_base_offset;  // Offset of the table base that gets added to the constant offsets.
 300     int                     _nof_jump_tables;    // Number of jump-tables in this constant table.
 301 
 302     static int qsort_comparator(Constant* a, Constant* b);
 303 
 304     // We use negative frequencies to keep the order of the
 305     // jump-tables in which they were added.  Otherwise we get into
 306     // trouble with relocation.
 307     float next_jump_table_freq() { return -1.0f * (++_nof_jump_tables); }
 308 
 309   public:
 310     ConstantTable() :
 311       _size(-1),
 312       _table_base_offset(-1),  // We can use -1 here since the constant table is always bigger than 2 bytes (-(size / 2), see MachConstantBaseNode::emit).
 313       _nof_jump_tables(0)
 314     {}
 315 
 316     int size() const { assert(_size != -1, "not calculated yet"); return _size; }
 317 
 318     int calculate_table_base_offset() const;  // AD specific
 319     void set_table_base_offset(int x)  { assert(_table_base_offset == -1 || x == _table_base_offset, "can't change"); _table_base_offset = x; }
 320     int      table_base_offset() const { assert(_table_base_offset != -1, "not set yet");                      return _table_base_offset; }
 321 
 322     void emit(CodeBuffer& cb);
 323 
 324     // Returns the offset of the last entry (the top) of the constant table.
 325     int  top_offset() const { assert(_constants.top().offset() != -1, "not bound yet"); return _constants.top().offset(); }
 326 
 327     void calculate_offsets_and_size();
 328     int  find_offset(Constant& con) const;
 329 
 330     void     add(Constant& con);
 331     Constant add(MachConstantNode* n, BasicType type, jvalue value);
 332     Constant add(Metadata* metadata);
 333     Constant add(MachConstantNode* n, MachOper* oper);
 334     Constant add(MachConstantNode* n, jint i) {
 335       jvalue value; value.i = i;
 336       return add(n, T_INT, value);
 337     }
 338     Constant add(MachConstantNode* n, jlong j) {
 339       jvalue value; value.j = j;
 340       return add(n, T_LONG, value);
 341     }
 342     Constant add(MachConstantNode* n, jfloat f) {
 343       jvalue value; value.f = f;
 344       return add(n, T_FLOAT, value);
 345     }
 346     Constant add(MachConstantNode* n, jdouble d) {
 347       jvalue value; value.d = d;
 348       return add(n, T_DOUBLE, value);
 349     }
 350 
 351     // Jump-table
 352     Constant  add_jump_table(MachConstantNode* n);
 353     void     fill_jump_table(CodeBuffer& cb, MachConstantNode* n, GrowableArray<Label*> labels) const;
 354   };
 355 
 356  private:
 357   // Fixed parameters to this compilation.
 358   const int             _compile_id;
 359   const bool            _save_argument_registers; // save/restore arg regs for trampolines
 360   const bool            _subsume_loads;         // Load can be matched as part of a larger op.
 361   const bool            _do_escape_analysis;    // Do escape analysis.
 362   const bool            _eliminate_boxing;      // Do boxing elimination.
 363   ciMethod*             _method;                // The method being compiled.
 364   int                   _entry_bci;             // entry bci for osr methods.
 365   const TypeFunc*       _tf;                    // My kind of signature
 366   InlineTree*           _ilt;                   // Ditto (temporary).
 367   address               _stub_function;         // VM entry for stub being compiled, or NULL
 368   const char*           _stub_name;             // Name of stub or adapter being compiled, or NULL
 369   address               _stub_entry_point;      // Compile code entry for generated stub, or NULL
 370 
 371   // Control of this compilation.
 372   int                   _max_inline_size;       // Max inline size for this compilation
 373   int                   _freq_inline_size;      // Max hot method inline size for this compilation
 374   int                   _fixed_slots;           // count of frame slots not allocated by the register
 375                                                 // allocator i.e. locks, original deopt pc, etc.
 376   uintx                 _max_node_limit;        // Max unique node count during a single compilation.
 377   // For deopt
 378   int                   _orig_pc_slot;
 379   int                   _orig_pc_slot_offset_in_bytes;
 380 
 381   int                   _major_progress;        // Count of something big happening
 382   bool                  _inlining_progress;     // progress doing incremental inlining?
 383   bool                  _inlining_incrementally;// Are we doing incremental inlining (post parse)
 384   bool                  _has_loops;             // True if the method _may_ have some loops
 385   bool                  _has_split_ifs;         // True if the method _may_ have some split-if
 386   bool                  _has_unsafe_access;     // True if the method _may_ produce faults in unsafe loads or stores.
 387   bool                  _has_stringbuilder;     // True StringBuffers or StringBuilders are allocated
 388   bool                  _has_boxed_value;       // True if a boxed object is allocated
 389   bool                  _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess
 390   uint                  _max_vector_size;       // Maximum size of generated vectors
 391   bool                  _clear_upper_avx;       // Clear upper bits of ymm registers using vzeroupper
 392   uint                  _trap_hist[trapHistLength];  // Cumulative traps
 393   bool                  _trap_can_recompile;    // Have we emitted a recompiling trap?
 394   uint                  _decompile_count;       // Cumulative decompilation counts.
 395   bool                  _do_inlining;           // True if we intend to do inlining
 396   bool                  _do_scheduling;         // True if we intend to do scheduling
 397   bool                  _do_freq_based_layout;  // True if we intend to do frequency based block layout
 398   bool                  _do_count_invocations;  // True if we generate code to count invocations
 399   bool                  _do_method_data_update; // True if we generate code to update MethodData*s
 400   bool                  _do_vector_loop;        // True if allowed to execute loop in parallel iterations
 401   bool                  _use_cmove;             // True if CMove should be used without profitability analysis
 402   bool                  _age_code;              // True if we need to profile code age (decrement the aging counter)
 403   int                   _AliasLevel;            // Locally-adjusted version of AliasLevel flag.
 404   bool                  _print_assembly;        // True if we should dump assembly code for this compilation
 405   bool                  _print_inlining;        // True if we should print inlining for this compilation
 406   bool                  _print_intrinsics;      // True if we should print intrinsics for this compilation
 407 #ifndef PRODUCT
 408   bool                  _trace_opto_output;
 409   bool                  _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
 410 #endif
 411   bool                  _has_irreducible_loop;  // Found irreducible loops
 412   // JSR 292
 413   bool                  _has_method_handle_invokes; // True if this method has MethodHandle invokes.
 414   RTMState              _rtm_state;             // State of Restricted Transactional Memory usage
 415   int                   _loop_opts_cnt;         // loop opts round
 416 
 417   // Compilation environment.
 418   Arena                 _comp_arena;            // Arena with lifetime equivalent to Compile
 419   void*                 _barrier_set_state;     // Potential GC barrier state for Compile
 420   ciEnv*                _env;                   // CI interface
 421   DirectiveSet*         _directive;             // Compiler directive
 422   CompileLog*           _log;                   // from CompilerThread
 423   const char*           _failure_reason;        // for record_failure/failing pattern
 424   GrowableArray<CallGenerator*>* _intrinsics;   // List of intrinsics.
 425   GrowableArray<Node*>* _macro_nodes;           // List of nodes which need to be expanded before matching.
 426   GrowableArray<Node*>* _predicate_opaqs;       // List of Opaque1 nodes for the loop predicates.
 427   GrowableArray<Node*>* _expensive_nodes;       // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
 428   GrowableArray<Node*>* _range_check_casts;     // List of CastII nodes with a range check dependency
 429   GrowableArray<Node*>* _opaque4_nodes;         // List of Opaque4 nodes that have a default value
 430   ConnectionGraph*      _congraph;
 431 #ifndef PRODUCT
 432   IdealGraphPrinter*    _printer;
 433 #endif
 434 
 435 
 436   // Node management
 437   uint                  _unique;                // Counter for unique Node indices
 438   VectorSet             _dead_node_list;        // Set of dead nodes
 439   uint                  _dead_node_count;       // Number of dead nodes; VectorSet::Size() is O(N).
 440                                                 // So use this to keep count and make the call O(1).
 441   DEBUG_ONLY( Unique_Node_List* _modified_nodes; )  // List of nodes which inputs were modified
 442 
 443   debug_only(static int _debug_idx;)            // Monotonic counter (not reset), use -XX:BreakAtNode=<idx>
 444   Arena                 _node_arena;            // Arena for new-space Nodes
 445   Arena                 _old_arena;             // Arena for old-space Nodes, lifetime during xform
 446   RootNode*             _root;                  // Unique root of compilation, or NULL after bail-out.
 447   Node*                 _top;                   // Unique top node.  (Reset by various phases.)
 448 
 449   Node*                 _immutable_memory;      // Initial memory state
 450 
 451   Node*                 _recent_alloc_obj;
 452   Node*                 _recent_alloc_ctl;
 453 
 454   // Constant table
 455   ConstantTable         _constant_table;        // The constant table for this compile.
 456   MachConstantBaseNode* _mach_constant_base_node;  // Constant table base node singleton.
 457 
 458 
 459   // Blocked array of debugging and profiling information,
 460   // tracked per node.
 461   enum { _log2_node_notes_block_size = 8,
 462          _node_notes_block_size = (1<<_log2_node_notes_block_size)
 463   };
 464   GrowableArray<Node_Notes*>* _node_note_array;
 465   Node_Notes*           _default_node_notes;  // default notes for new nodes
 466 
 467   // After parsing and every bulk phase we hang onto the Root instruction.
 468   // The RootNode instruction is where the whole program begins.  It produces
 469   // the initial Control and BOTTOM for everybody else.
 470 
 471   // Type management
 472   Arena                 _Compile_types;         // Arena for all types
 473   Arena*                _type_arena;            // Alias for _Compile_types except in Initialize_shared()
 474   Dict*                 _type_dict;             // Intern table
 475   CloneMap              _clone_map;             // used for recording history of cloned nodes
 476   void*                 _type_hwm;              // Last allocation (see Type::operator new/delete)
 477   size_t                _type_last_size;        // Last allocation size (see Type::operator new/delete)
 478   ciMethod*             _last_tf_m;             // Cache for
 479   const TypeFunc*       _last_tf;               //  TypeFunc::make
 480   AliasType**           _alias_types;           // List of alias types seen so far.
 481   int                   _num_alias_types;       // Logical length of _alias_types
 482   int                   _max_alias_types;       // Physical length of _alias_types
 483   AliasCacheEntry       _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
 484 
 485   // Parsing, optimization
 486   PhaseGVN*             _initial_gvn;           // Results of parse-time PhaseGVN
 487   Unique_Node_List*     _for_igvn;              // Initial work-list for next round of Iterative GVN
 488   WarmCallInfo*         _warm_calls;            // Sorted work-list for heat-based inlining.
 489 
 490   GrowableArray<CallGenerator*> _late_inlines;        // List of CallGenerators to be revisited after
 491                                                       // main parsing has finished.
 492   GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
 493 
 494   GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
 495 
 496   int                           _late_inlines_pos;    // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
 497   uint                          _number_of_mh_late_inlines; // number of method handle late inlining still pending
 498 
 499 
 500   // Inlining may not happen in parse order which would make
 501   // PrintInlining output confusing. Keep track of PrintInlining
 502   // pieces in order.
 503   class PrintInliningBuffer : public ResourceObj {
 504    private:
 505     CallGenerator* _cg;
 506     stringStream* _ss;
 507 
 508    public:
 509     PrintInliningBuffer()
 510       : _cg(NULL) { _ss = new stringStream(); }
 511 
 512     stringStream* ss() const { return _ss; }
 513     CallGenerator* cg() const { return _cg; }
 514     void set_cg(CallGenerator* cg) { _cg = cg; }
 515   };
 516 
 517   stringStream* _print_inlining_stream;
 518   GrowableArray<PrintInliningBuffer>* _print_inlining_list;
 519   int _print_inlining_idx;
 520   char* _print_inlining_output;
 521 
 522   // Only keep nodes in the expensive node list that need to be optimized
 523   void cleanup_expensive_nodes(PhaseIterGVN &igvn);
 524   // Use for sorting expensive nodes to bring similar nodes together
 525   static int cmp_expensive_nodes(Node** n1, Node** n2);
 526   // Expensive nodes list already sorted?
 527   bool expensive_nodes_sorted() const;
 528   // Remove the speculative part of types and clean up the graph
 529   void remove_speculative_types(PhaseIterGVN &igvn);
 530 
 531   void* _replay_inline_data; // Pointer to data loaded from file
 532 
 533   void print_inlining_init();
 534   void print_inlining_reinit();
 535   void print_inlining_commit();
 536   void print_inlining_push();
 537   PrintInliningBuffer& print_inlining_current();
 538 
 539   void log_late_inline_failure(CallGenerator* cg, const char* msg);
 540 
 541  public:
 542 
 543   void* barrier_set_state() const { return _barrier_set_state; }
 544 
 545   outputStream* print_inlining_stream() const {
 546     assert(print_inlining() || print_intrinsics(), "PrintInlining off?");
 547     return _print_inlining_stream;
 548   }
 549 
 550   void print_inlining_update(CallGenerator* cg);
 551   void print_inlining_update_delayed(CallGenerator* cg);
 552   void print_inlining_move_to(CallGenerator* cg);
 553   void print_inlining_assert_ready();
 554   void print_inlining_reset();
 555 
 556   void print_inlining(ciMethod* method, int inline_level, int bci, const char* msg = NULL) {
 557     stringStream ss;
 558     CompileTask::print_inlining_inner(&ss, method, inline_level, bci, msg);
 559     print_inlining_stream()->print("%s", ss.as_string());
 560   }
 561 
 562 #ifndef PRODUCT
 563   IdealGraphPrinter* printer() { return _printer; }
 564 #endif
 565 
 566   void log_late_inline(CallGenerator* cg);
 567   void log_inline_id(CallGenerator* cg);
 568   void log_inline_failure(const char* msg);
 569 
 570   void* replay_inline_data() const { return _replay_inline_data; }
 571 
 572   // Dump inlining replay data to the stream.
 573   void dump_inline_data(outputStream* out);
 574 
 575  private:
 576   // Matching, CFG layout, allocation, code generation
 577   PhaseCFG*             _cfg;                   // Results of CFG finding
 578   bool                  _select_24_bit_instr;   // We selected an instruction with a 24-bit result
 579   bool                  _in_24_bit_fp_mode;     // We are emitting instructions with 24-bit results
 580   int                   _java_calls;            // Number of java calls in the method
 581   int                   _inner_loops;           // Number of inner loops in the method
 582   Matcher*              _matcher;               // Engine to map ideal to machine instructions
 583   PhaseRegAlloc*        _regalloc;              // Results of register allocation.
 584   int                   _frame_slots;           // Size of total frame in stack slots
 585   CodeOffsets           _code_offsets;          // Offsets into the code for various interesting entries
 586   RegMask               _FIRST_STACK_mask;      // All stack slots usable for spills (depends on frame layout)
 587   Arena*                _indexSet_arena;        // control IndexSet allocation within PhaseChaitin
 588   void*                 _indexSet_free_block_list; // free list of IndexSet bit blocks
 589   int                   _interpreter_frame_size;
 590 
 591   uint                  _node_bundling_limit;
 592   Bundle*               _node_bundling_base;    // Information for instruction bundling
 593 
 594   // Instruction bits passed off to the VM
 595   int                   _method_size;           // Size of nmethod code segment in bytes
 596   CodeBuffer            _code_buffer;           // Where the code is assembled
 597   int                   _first_block_size;      // Size of unvalidated entry point code / OSR poison code
 598   ExceptionHandlerTable _handler_table;         // Table of native-code exception handlers
 599   ImplicitExceptionTable _inc_table;            // Table of implicit null checks in native code
 600   OopMapSet*            _oop_map_set;           // Table of oop maps (one for each safepoint location)
 601   static int            _CompiledZap_count;     // counter compared against CompileZap[First/Last]
 602   BufferBlob*           _scratch_buffer_blob;   // For temporary code buffers.
 603   relocInfo*            _scratch_locs_memory;   // For temporary code buffers.
 604   int                   _scratch_const_size;    // For temporary code buffers.
 605   bool                  _in_scratch_emit_size;  // true when in scratch_emit_size.
 606 
 607   void reshape_address(AddPNode* n);
 608 
 609  public:
 610   // Accessors
 611 
 612   // The Compile instance currently active in this (compiler) thread.
 613   static Compile* current() {
 614     return (Compile*) ciEnv::current()->compiler_data();
 615   }
 616 
 617   // ID for this compilation.  Useful for setting breakpoints in the debugger.
 618   int               compile_id() const          { return _compile_id; }
 619   DirectiveSet*     directive() const           { return _directive; }
 620 
 621   // Does this compilation allow instructions to subsume loads?  User
 622   // instructions that subsume a load may result in an unschedulable
 623   // instruction sequence.
 624   bool              subsume_loads() const       { return _subsume_loads; }
 625   /** Do escape analysis. */
 626   bool              do_escape_analysis() const  { return _do_escape_analysis; }
 627   /** Do boxing elimination. */
 628   bool              eliminate_boxing() const    { return _eliminate_boxing; }
 629   /** Do aggressive boxing elimination. */
 630   bool              aggressive_unboxing() const { return _eliminate_boxing && AggressiveUnboxing; }
 631   bool              save_argument_registers() const { return _save_argument_registers; }
 632 
 633 
 634   // Other fixed compilation parameters.
 635   ciMethod*         method() const              { return _method; }
 636   int               entry_bci() const           { return _entry_bci; }
 637   bool              is_osr_compilation() const  { return _entry_bci != InvocationEntryBci; }
 638   bool              is_method_compilation() const { return (_method != NULL && !_method->flags().is_native()); }
 639   const TypeFunc*   tf() const                  { assert(_tf!=NULL, ""); return _tf; }
 640   void         init_tf(const TypeFunc* tf)      { assert(_tf==NULL, ""); _tf = tf; }
 641   InlineTree*       ilt() const                 { return _ilt; }
 642   address           stub_function() const       { return _stub_function; }
 643   const char*       stub_name() const           { return _stub_name; }
 644   address           stub_entry_point() const    { return _stub_entry_point; }
 645 
 646   // Control of this compilation.
 647   int               fixed_slots() const         { assert(_fixed_slots >= 0, "");         return _fixed_slots; }
 648   void          set_fixed_slots(int n)          { _fixed_slots = n; }
 649   int               major_progress() const      { return _major_progress; }
 650   void          set_inlining_progress(bool z)   { _inlining_progress = z; }
 651   int               inlining_progress() const   { return _inlining_progress; }
 652   void          set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
 653   int               inlining_incrementally() const { return _inlining_incrementally; }
 654   void          set_major_progress()            { _major_progress++; }
 655   void        clear_major_progress()            { _major_progress = 0; }
 656   int               max_inline_size() const     { return _max_inline_size; }
 657   void          set_freq_inline_size(int n)     { _freq_inline_size = n; }
 658   int               freq_inline_size() const    { return _freq_inline_size; }
 659   void          set_max_inline_size(int n)      { _max_inline_size = n; }
 660   bool              has_loops() const           { return _has_loops; }
 661   void          set_has_loops(bool z)           { _has_loops = z; }
 662   bool              has_split_ifs() const       { return _has_split_ifs; }
 663   void          set_has_split_ifs(bool z)       { _has_split_ifs = z; }
 664   bool              has_unsafe_access() const   { return _has_unsafe_access; }
 665   void          set_has_unsafe_access(bool z)   { _has_unsafe_access = z; }
 666   bool              has_stringbuilder() const   { return _has_stringbuilder; }
 667   void          set_has_stringbuilder(bool z)   { _has_stringbuilder = z; }
 668   bool              has_boxed_value() const     { return _has_boxed_value; }
 669   void          set_has_boxed_value(bool z)     { _has_boxed_value = z; }
 670   bool              has_reserved_stack_access() const { return _has_reserved_stack_access; }
 671   void          set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
 672   uint              max_vector_size() const     { return _max_vector_size; }
 673   void          set_max_vector_size(uint s)     { _max_vector_size = s; }
 674   bool              clear_upper_avx() const     { return _clear_upper_avx; }
 675   void          set_clear_upper_avx(bool s)     { _clear_upper_avx = s; }
 676   void          set_trap_count(uint r, uint c)  { assert(r < trapHistLength, "oob");        _trap_hist[r] = c; }
 677   uint              trap_count(uint r) const    { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
 678   bool              trap_can_recompile() const  { return _trap_can_recompile; }
 679   void          set_trap_can_recompile(bool z)  { _trap_can_recompile = z; }
 680   uint              decompile_count() const     { return _decompile_count; }
 681   void          set_decompile_count(uint c)     { _decompile_count = c; }
 682   bool              allow_range_check_smearing() const;
 683   bool              do_inlining() const         { return _do_inlining; }
 684   void          set_do_inlining(bool z)         { _do_inlining = z; }
 685   bool              do_scheduling() const       { return _do_scheduling; }
 686   void          set_do_scheduling(bool z)       { _do_scheduling = z; }
 687   bool              do_freq_based_layout() const{ return _do_freq_based_layout; }
 688   void          set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
 689   bool              do_count_invocations() const{ return _do_count_invocations; }
 690   void          set_do_count_invocations(bool z){ _do_count_invocations = z; }
 691   bool              do_method_data_update() const { return _do_method_data_update; }
 692   void          set_do_method_data_update(bool z) { _do_method_data_update = z; }
 693   bool              do_vector_loop() const      { return _do_vector_loop; }
 694   void          set_do_vector_loop(bool z)      { _do_vector_loop = z; }
 695   bool              use_cmove() const           { return _use_cmove; }
 696   void          set_use_cmove(bool z)           { _use_cmove = z; }
 697   bool              age_code() const             { return _age_code; }
 698   void          set_age_code(bool z)             { _age_code = z; }
 699   int               AliasLevel() const           { return _AliasLevel; }
 700   bool              print_assembly() const       { return _print_assembly; }
 701   void          set_print_assembly(bool z)       { _print_assembly = z; }
 702   bool              print_inlining() const       { return _print_inlining; }
 703   void          set_print_inlining(bool z)       { _print_inlining = z; }
 704   bool              print_intrinsics() const     { return _print_intrinsics; }
 705   void          set_print_intrinsics(bool z)     { _print_intrinsics = z; }
 706   RTMState          rtm_state()  const           { return _rtm_state; }
 707   void          set_rtm_state(RTMState s)        { _rtm_state = s; }
 708   bool              use_rtm() const              { return (_rtm_state & NoRTM) == 0; }
 709   bool          profile_rtm() const              { return _rtm_state == ProfileRTM; }
 710   uint              max_node_limit() const       { return (uint)_max_node_limit; }
 711   void          set_max_node_limit(uint n)       { _max_node_limit = n; }
 712 
 713   // check the CompilerOracle for special behaviours for this compile
 714   bool          method_has_option(const char * option) {
 715     return method() != NULL && method()->has_option(option);
 716   }
 717 
 718 #ifndef PRODUCT
 719   bool          trace_opto_output() const       { return _trace_opto_output; }
 720   bool              parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
 721   void          set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
 722   int _in_dump_cnt;  // Required for dumping ir nodes.
 723 #endif
 724   bool              has_irreducible_loop() const { return _has_irreducible_loop; }
 725   void          set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
 726 
 727   // JSR 292
 728   bool              has_method_handle_invokes() const { return _has_method_handle_invokes;     }
 729   void          set_has_method_handle_invokes(bool z) {        _has_method_handle_invokes = z; }
 730 
 731   Ticks _latest_stage_start_counter;
 732 
 733   void begin_method() {
 734 #ifndef PRODUCT
 735     if (_printer && _printer->should_print(1)) {
 736       _printer->begin_method();
 737     }
 738 #endif
 739     C->_latest_stage_start_counter.stamp();
 740   }
 741 
 742   void print_method(CompilerPhaseType cpt, int level = 1) {
 743     EventCompilerPhase event;
 744     if (event.should_commit()) {
 745       event.set_starttime(C->_latest_stage_start_counter);
 746       event.set_phase((u1) cpt);
 747       event.set_compileId(C->_compile_id);
 748       event.set_phaseLevel(level);
 749       event.commit();
 750     }
 751 
 752 
 753 #ifndef PRODUCT
 754     if (_printer && _printer->should_print(level)) {
 755       _printer->print_method(CompilerPhaseTypeHelper::to_string(cpt), level);
 756     }
 757 #endif
 758     C->_latest_stage_start_counter.stamp();
 759   }
 760 
 761   void end_method(int level = 1) {
 762     EventCompilerPhase event;
 763     if (event.should_commit()) {
 764       event.set_starttime(C->_latest_stage_start_counter);
 765       event.set_phase((u1) PHASE_END);
 766       event.set_compileId(C->_compile_id);
 767       event.set_phaseLevel(level);
 768       event.commit();
 769     }
 770 #ifndef PRODUCT
 771     if (_printer && _printer->should_print(level)) {
 772       _printer->end_method();
 773     }
 774 #endif
 775   }
 776 
 777   int           macro_count()             const { return _macro_nodes->length(); }
 778   int           predicate_count()         const { return _predicate_opaqs->length();}
 779   int           expensive_count()         const { return _expensive_nodes->length(); }
 780   Node*         macro_node(int idx)       const { return _macro_nodes->at(idx); }
 781   Node*         predicate_opaque1_node(int idx) const { return _predicate_opaqs->at(idx);}
 782   Node*         expensive_node(int idx)   const { return _expensive_nodes->at(idx); }
 783   ConnectionGraph* congraph()                   { return _congraph;}
 784   void set_congraph(ConnectionGraph* congraph)  { _congraph = congraph;}
 785   void add_macro_node(Node * n) {
 786     //assert(n->is_macro(), "must be a macro node");
 787     assert(!_macro_nodes->contains(n), "duplicate entry in expand list");
 788     _macro_nodes->append(n);
 789   }
 790   void remove_macro_node(Node * n) {
 791     // this function may be called twice for a node so check
 792     // that the node is in the array before attempting to remove it
 793     if (_macro_nodes->contains(n))
 794       _macro_nodes->remove(n);
 795     // remove from _predicate_opaqs list also if it is there
 796     if (predicate_count() > 0 && _predicate_opaqs->contains(n)){
 797       _predicate_opaqs->remove(n);
 798     }
 799   }
 800   void add_expensive_node(Node * n);
 801   void remove_expensive_node(Node * n) {
 802     if (_expensive_nodes->contains(n)) {
 803       _expensive_nodes->remove(n);
 804     }
 805   }
 806   void add_predicate_opaq(Node * n) {
 807     assert(!_predicate_opaqs->contains(n), "duplicate entry in predicate opaque1");
 808     assert(_macro_nodes->contains(n), "should have already been in macro list");
 809     _predicate_opaqs->append(n);
 810   }
 811 
 812   // Range check dependent CastII nodes that can be removed after loop optimizations
 813   void add_range_check_cast(Node* n);
 814   void remove_range_check_cast(Node* n) {
 815     if (_range_check_casts->contains(n)) {
 816       _range_check_casts->remove(n);
 817     }
 818   }
 819   Node* range_check_cast_node(int idx) const { return _range_check_casts->at(idx);  }
 820   int   range_check_cast_count()       const { return _range_check_casts->length(); }
 821   // Remove all range check dependent CastIINodes.
 822   void  remove_range_check_casts(PhaseIterGVN &igvn);
 823 
 824   void add_opaque4_node(Node* n);
 825   void remove_opaque4_node(Node* n) {
 826     if (_opaque4_nodes->contains(n)) {
 827       _opaque4_nodes->remove(n);
 828     }
 829   }
 830   Node* opaque4_node(int idx) const { return _opaque4_nodes->at(idx);  }
 831   int   opaque4_count()       const { return _opaque4_nodes->length(); }
 832   void  remove_opaque4_nodes(PhaseIterGVN &igvn);
 833 
 834   // remove the opaque nodes that protect the predicates so that the unused checks and
 835   // uncommon traps will be eliminated from the graph.
 836   void cleanup_loop_predicates(PhaseIterGVN &igvn);
 837   bool is_predicate_opaq(Node * n) {
 838     return _predicate_opaqs->contains(n);
 839   }
 840 
 841   // Are there candidate expensive nodes for optimization?
 842   bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
 843   // Check whether n1 and n2 are similar
 844   static int cmp_expensive_nodes(Node* n1, Node* n2);
 845   // Sort expensive nodes to locate similar expensive nodes
 846   void sort_expensive_nodes();
 847 
 848   // Compilation environment.
 849   Arena*      comp_arena()           { return &_comp_arena; }
 850   ciEnv*      env() const            { return _env; }
 851   CompileLog* log() const            { return _log; }
 852   bool        failing() const        { return _env->failing() || _failure_reason != NULL; }
 853   const char* failure_reason() const { return (_env->failing()) ? _env->failure_reason() : _failure_reason; }
 854 
 855   bool failure_reason_is(const char* r) const {
 856     return (r == _failure_reason) || (r != NULL && _failure_reason != NULL && strcmp(r, _failure_reason) == 0);
 857   }
 858 
 859   void record_failure(const char* reason);
 860   void record_method_not_compilable(const char* reason) {
 861     // Bailouts cover "all_tiers" when TieredCompilation is off.
 862     env()->record_method_not_compilable(reason, !TieredCompilation);
 863     // Record failure reason.
 864     record_failure(reason);
 865   }
 866   bool check_node_count(uint margin, const char* reason) {
 867     if (live_nodes() + margin > max_node_limit()) {
 868       record_method_not_compilable(reason);
 869       return true;
 870     } else {
 871       return false;
 872     }
 873   }
 874 
 875   // Node management
 876   uint         unique() const              { return _unique; }
 877   uint         next_unique()               { return _unique++; }
 878   void         set_unique(uint i)          { _unique = i; }
 879   static int   debug_idx()                 { return debug_only(_debug_idx)+0; }
 880   static void  set_debug_idx(int i)        { debug_only(_debug_idx = i); }
 881   Arena*       node_arena()                { return &_node_arena; }
 882   Arena*       old_arena()                 { return &_old_arena; }
 883   RootNode*    root() const                { return _root; }
 884   void         set_root(RootNode* r)       { _root = r; }
 885   StartNode*   start() const;              // (Derived from root.)
 886   void         init_start(StartNode* s);
 887   Node*        immutable_memory();
 888 
 889   Node*        recent_alloc_ctl() const    { return _recent_alloc_ctl; }
 890   Node*        recent_alloc_obj() const    { return _recent_alloc_obj; }
 891   void         set_recent_alloc(Node* ctl, Node* obj) {
 892                                                   _recent_alloc_ctl = ctl;
 893                                                   _recent_alloc_obj = obj;
 894                                            }
 895   void         record_dead_node(uint idx)  { if (_dead_node_list.test_set(idx)) return;
 896                                              _dead_node_count++;
 897                                            }
 898   bool         is_dead_node(uint idx)      { return _dead_node_list.test(idx) != 0; }
 899   uint         dead_node_count()           { return _dead_node_count; }
 900   void         reset_dead_node_list()      { _dead_node_list.Reset();
 901                                              _dead_node_count = 0;
 902                                            }
 903   uint          live_nodes() const         {
 904     int  val = _unique - _dead_node_count;
 905     assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
 906             return (uint) val;
 907                                            }
 908 #ifdef ASSERT
 909   uint         count_live_nodes_by_graph_walk();
 910   void         print_missing_nodes();
 911 #endif
 912 
 913   // Record modified nodes to check that they are put on IGVN worklist
 914   void         record_modified_node(Node* n) NOT_DEBUG_RETURN;
 915   void         remove_modified_node(Node* n) NOT_DEBUG_RETURN;
 916   DEBUG_ONLY( Unique_Node_List*   modified_nodes() const { return _modified_nodes; } )
 917 
 918   // Constant table
 919   ConstantTable&   constant_table() { return _constant_table; }
 920 
 921   MachConstantBaseNode*     mach_constant_base_node();
 922   bool                  has_mach_constant_base_node() const { return _mach_constant_base_node != NULL; }
 923   // Generated by adlc, true if CallNode requires MachConstantBase.
 924   bool                      needs_clone_jvms();
 925 
 926   // Handy undefined Node
 927   Node*             top() const                 { return _top; }
 928 
 929   // these are used by guys who need to know about creation and transformation of top:
 930   Node*             cached_top_node()           { return _top; }
 931   void          set_cached_top_node(Node* tn);
 932 
 933   GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
 934   void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
 935   Node_Notes* default_node_notes() const        { return _default_node_notes; }
 936   void    set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
 937 
 938   Node_Notes*       node_notes_at(int idx) {
 939     return locate_node_notes(_node_note_array, idx, false);
 940   }
 941   inline bool   set_node_notes_at(int idx, Node_Notes* value);
 942 
 943   // Copy notes from source to dest, if they exist.
 944   // Overwrite dest only if source provides something.
 945   // Return true if information was moved.
 946   bool copy_node_notes_to(Node* dest, Node* source);
 947 
 948   // Workhorse function to sort out the blocked Node_Notes array:
 949   inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
 950                                        int idx, bool can_grow = false);
 951 
 952   void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
 953 
 954   // Type management
 955   Arena*            type_arena()                { return _type_arena; }
 956   Dict*             type_dict()                 { return _type_dict; }
 957   void*             type_hwm()                  { return _type_hwm; }
 958   size_t            type_last_size()            { return _type_last_size; }
 959   int               num_alias_types()           { return _num_alias_types; }
 960 
 961   void          init_type_arena()                       { _type_arena = &_Compile_types; }
 962   void          set_type_arena(Arena* a)                { _type_arena = a; }
 963   void          set_type_dict(Dict* d)                  { _type_dict = d; }
 964   void          set_type_hwm(void* p)                   { _type_hwm = p; }
 965   void          set_type_last_size(size_t sz)           { _type_last_size = sz; }
 966 
 967   const TypeFunc* last_tf(ciMethod* m) {
 968     return (m == _last_tf_m) ? _last_tf : NULL;
 969   }
 970   void set_last_tf(ciMethod* m, const TypeFunc* tf) {
 971     assert(m != NULL || tf == NULL, "");
 972     _last_tf_m = m;
 973     _last_tf = tf;
 974   }
 975 
 976   AliasType*        alias_type(int                idx)  { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
 977   AliasType*        alias_type(const TypePtr* adr_type, ciField* field = NULL) { return find_alias_type(adr_type, false, field); }
 978   bool         have_alias_type(const TypePtr* adr_type);
 979   AliasType*        alias_type(ciField*         field);
 980 
 981   int               get_alias_index(const TypePtr* at)  { return alias_type(at)->index(); }
 982   const TypePtr*    get_adr_type(uint aidx)             { return alias_type(aidx)->adr_type(); }
 983   int               get_general_index(uint aidx)        { return alias_type(aidx)->general_index(); }
 984 
 985   // Building nodes
 986   void              rethrow_exceptions(JVMState* jvms);
 987   void              return_values(JVMState* jvms);
 988   JVMState*         build_start_state(StartNode* start, const TypeFunc* tf);
 989 
 990   // Decide how to build a call.
 991   // The profile factor is a discount to apply to this site's interp. profile.
 992   CallGenerator*    call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
 993                                    JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = NULL,
 994                                    bool allow_intrinsics = true, bool delayed_forbidden = false);
 995   bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
 996     return should_delay_string_inlining(call_method, jvms) ||
 997            should_delay_boxing_inlining(call_method, jvms);
 998   }
 999   bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
1000   bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
1001 
1002   // Helper functions to identify inlining potential at call-site
1003   ciMethod* optimize_virtual_call(ciMethod* caller, int bci, ciInstanceKlass* klass,
1004                                   ciKlass* holder, ciMethod* callee,
1005                                   const TypeOopPtr* receiver_type, bool is_virtual,
1006                                   bool &call_does_dispatch, int &vtable_index,
1007                                   bool check_access = true);
1008   ciMethod* optimize_inlining(ciMethod* caller, int bci, ciInstanceKlass* klass,
1009                               ciMethod* callee, const TypeOopPtr* receiver_type,
1010                               bool check_access = true);
1011 
1012   // Report if there were too many traps at a current method and bci.
1013   // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
1014   // If there is no MDO at all, report no trap unless told to assume it.
1015   bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1016   // This version, unspecific to a particular bci, asks if
1017   // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
1018   bool too_many_traps(Deoptimization::DeoptReason reason,
1019                       // Privately used parameter for logging:
1020                       ciMethodData* logmd = NULL);
1021   // Report if there were too many recompiles at a method and bci.
1022   bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1023   // Return a bitset with the reasons where deoptimization is allowed,
1024   // i.e., where there were not too many uncommon traps.
1025   int _allowed_reasons;
1026   int      allowed_deopt_reasons() { return _allowed_reasons; }
1027   void set_allowed_deopt_reasons();
1028 
1029   // Parsing, optimization
1030   PhaseGVN*         initial_gvn()               { return _initial_gvn; }
1031   Unique_Node_List* for_igvn()                  { return _for_igvn; }
1032   inline void       record_for_igvn(Node* n);   // Body is after class Unique_Node_List.
1033   void          set_initial_gvn(PhaseGVN *gvn)           { _initial_gvn = gvn; }
1034   void          set_for_igvn(Unique_Node_List *for_igvn) { _for_igvn = for_igvn; }
1035 
1036   // Replace n by nn using initial_gvn, calling hash_delete and
1037   // record_for_igvn as needed.
1038   void gvn_replace_by(Node* n, Node* nn);
1039 
1040 
1041   void              identify_useful_nodes(Unique_Node_List &useful);
1042   void              update_dead_node_list(Unique_Node_List &useful);
1043   void              remove_useless_nodes (Unique_Node_List &useful);
1044 
1045   WarmCallInfo*     warm_calls() const          { return _warm_calls; }
1046   void          set_warm_calls(WarmCallInfo* l) { _warm_calls = l; }
1047   WarmCallInfo* pop_warm_call();
1048 
1049   // Record this CallGenerator for inlining at the end of parsing.
1050   void              add_late_inline(CallGenerator* cg)        {
1051     _late_inlines.insert_before(_late_inlines_pos, cg);
1052     _late_inlines_pos++;
1053   }
1054 
1055   void              prepend_late_inline(CallGenerator* cg)    {
1056     _late_inlines.insert_before(0, cg);
1057   }
1058 
1059   void              add_string_late_inline(CallGenerator* cg) {
1060     _string_late_inlines.push(cg);
1061   }
1062 
1063   void              add_boxing_late_inline(CallGenerator* cg) {
1064     _boxing_late_inlines.push(cg);
1065   }
1066 
1067   void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1068 
1069   void process_print_inlining();
1070   void dump_print_inlining();
1071 
1072   bool over_inlining_cutoff() const {
1073     if (!inlining_incrementally()) {
1074       return unique() > (uint)NodeCountInliningCutoff;
1075     } else {
1076       return live_nodes() > (uint)LiveNodeCountInliningCutoff;
1077     }
1078   }
1079 
1080   void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; }
1081   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--; }
1082   bool has_mh_late_inlines() const     { return _number_of_mh_late_inlines > 0; }
1083 
1084   void inline_incrementally_one(PhaseIterGVN& igvn);
1085   void inline_incrementally(PhaseIterGVN& igvn);
1086   void inline_string_calls(bool parse_time);
1087   void inline_boxing_calls(PhaseIterGVN& igvn);
1088   bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1089 
1090   // Matching, CFG layout, allocation, code generation
1091   PhaseCFG*         cfg()                       { return _cfg; }
1092   bool              select_24_bit_instr() const { return _select_24_bit_instr; }
1093   bool              in_24_bit_fp_mode() const   { return _in_24_bit_fp_mode; }
1094   bool              has_java_calls() const      { return _java_calls > 0; }
1095   int               java_calls() const          { return _java_calls; }
1096   int               inner_loops() const         { return _inner_loops; }
1097   Matcher*          matcher()                   { return _matcher; }
1098   PhaseRegAlloc*    regalloc()                  { return _regalloc; }
1099   int               frame_slots() const         { return _frame_slots; }
1100   int               frame_size_in_words() const; // frame_slots in units of the polymorphic 'words'
1101   int               frame_size_in_bytes() const { return _frame_slots << LogBytesPerInt; }
1102   RegMask&          FIRST_STACK_mask()          { return _FIRST_STACK_mask; }
1103   Arena*            indexSet_arena()            { return _indexSet_arena; }
1104   void*             indexSet_free_block_list()  { return _indexSet_free_block_list; }
1105   uint              node_bundling_limit()       { return _node_bundling_limit; }
1106   Bundle*           node_bundling_base()        { return _node_bundling_base; }
1107   void          set_node_bundling_limit(uint n) { _node_bundling_limit = n; }
1108   void          set_node_bundling_base(Bundle* b) { _node_bundling_base = b; }
1109   bool          starts_bundle(const Node *n) const;
1110   bool          need_stack_bang(int frame_size_in_bytes) const;
1111   bool          need_register_stack_bang() const;
1112 
1113   void  update_interpreter_frame_size(int size) {
1114     if (_interpreter_frame_size < size) {
1115       _interpreter_frame_size = size;
1116     }
1117   }
1118   int           bang_size_in_bytes() const;
1119 
1120   void          set_matcher(Matcher* m)                 { _matcher = m; }
1121 //void          set_regalloc(PhaseRegAlloc* ra)           { _regalloc = ra; }
1122   void          set_indexSet_arena(Arena* a)            { _indexSet_arena = a; }
1123   void          set_indexSet_free_block_list(void* p)   { _indexSet_free_block_list = p; }
1124 
1125   // Remember if this compilation changes hardware mode to 24-bit precision
1126   void set_24_bit_selection_and_mode(bool selection, bool mode) {
1127     _select_24_bit_instr = selection;
1128     _in_24_bit_fp_mode   = mode;
1129   }
1130 
1131   void  set_java_calls(int z) { _java_calls  = z; }
1132   void set_inner_loops(int z) { _inner_loops = z; }
1133 
1134   // Instruction bits passed off to the VM
1135   int               code_size()                 { return _method_size; }
1136   CodeBuffer*       code_buffer()               { return &_code_buffer; }
1137   int               first_block_size()          { return _first_block_size; }
1138   void              set_frame_complete(int off) { if (!in_scratch_emit_size()) { _code_offsets.set_value(CodeOffsets::Frame_Complete, off); } }
1139   ExceptionHandlerTable*  handler_table()       { return &_handler_table; }
1140   ImplicitExceptionTable* inc_table()           { return &_inc_table; }
1141   OopMapSet*        oop_map_set()               { return _oop_map_set; }
1142   DebugInformationRecorder* debug_info()        { return env()->debug_info(); }
1143   Dependencies*     dependencies()              { return env()->dependencies(); }
1144   static int        CompiledZap_count()         { return _CompiledZap_count; }
1145   BufferBlob*       scratch_buffer_blob()       { return _scratch_buffer_blob; }
1146   void         init_scratch_buffer_blob(int const_size);
1147   void        clear_scratch_buffer_blob();
1148   void          set_scratch_buffer_blob(BufferBlob* b) { _scratch_buffer_blob = b; }
1149   relocInfo*        scratch_locs_memory()       { return _scratch_locs_memory; }
1150   void          set_scratch_locs_memory(relocInfo* b)  { _scratch_locs_memory = b; }
1151 
1152   // emit to scratch blob, report resulting size
1153   uint              scratch_emit_size(const Node* n);
1154   void       set_in_scratch_emit_size(bool x)   {        _in_scratch_emit_size = x; }
1155   bool           in_scratch_emit_size() const   { return _in_scratch_emit_size;     }
1156 
1157   enum ScratchBufferBlob {
1158 #if defined(PPC64)
1159     MAX_inst_size       = 2048,
1160 #else
1161     MAX_inst_size       = 1024,
1162 #endif
1163     MAX_locs_size       = 128, // number of relocInfo elements
1164     MAX_const_size      = 128,
1165     MAX_stubs_size      = 128
1166   };
1167 
1168   // Major entry point.  Given a Scope, compile the associated method.
1169   // For normal compilations, entry_bci is InvocationEntryBci.  For on stack
1170   // replacement, entry_bci indicates the bytecode for which to compile a
1171   // continuation.
1172   Compile(ciEnv* ci_env, C2Compiler* compiler, ciMethod* target,
1173           int entry_bci, bool subsume_loads, bool do_escape_analysis,
1174           bool eliminate_boxing, DirectiveSet* directive);
1175 
1176   // Second major entry point.  From the TypeFunc signature, generate code
1177   // to pass arguments from the Java calling convention to the C calling
1178   // convention.
1179   Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1180           address stub_function, const char *stub_name,
1181           int is_fancy_jump, bool pass_tls,
1182           bool save_arg_registers, bool return_pc, DirectiveSet* directive);
1183 
1184   // From the TypeFunc signature, generate code to pass arguments
1185   // from Compiled calling convention to Interpreter's calling convention
1186   void Generate_Compiled_To_Interpreter_Graph(const TypeFunc *tf, address interpreter_entry);
1187 
1188   // From the TypeFunc signature, generate code to pass arguments
1189   // from Interpreter's calling convention to Compiler's calling convention
1190   void Generate_Interpreter_To_Compiled_Graph(const TypeFunc *tf);
1191 
1192   // Are we compiling a method?
1193   bool has_method() { return method() != NULL; }
1194 
1195   // Maybe print some information about this compile.
1196   void print_compile_messages();
1197 
1198   // Final graph reshaping, a post-pass after the regular optimizer is done.
1199   bool final_graph_reshaping();
1200 
1201   // returns true if adr is completely contained in the given alias category
1202   bool must_alias(const TypePtr* adr, int alias_idx);
1203 
1204   // returns true if adr overlaps with the given alias category
1205   bool can_alias(const TypePtr* adr, int alias_idx);
1206 
1207   // Driver for converting compiler's IR into machine code bits
1208   void Output();
1209 
1210   // Accessors for node bundling info.
1211   Bundle* node_bundling(const Node *n);
1212   bool valid_bundle_info(const Node *n);
1213 
1214   // Schedule and Bundle the instructions
1215   void ScheduleAndBundle();
1216 
1217   // Build OopMaps for each GC point
1218   void BuildOopMaps();
1219 
1220   // Append debug info for the node "local" at safepoint node "sfpt" to the
1221   // "array",   May also consult and add to "objs", which describes the
1222   // scalar-replaced objects.
1223   void FillLocArray( int idx, MachSafePointNode* sfpt,
1224                      Node *local, GrowableArray<ScopeValue*> *array,
1225                      GrowableArray<ScopeValue*> *objs );
1226 
1227   // If "objs" contains an ObjectValue whose id is "id", returns it, else NULL.
1228   static ObjectValue* sv_for_node_id(GrowableArray<ScopeValue*> *objs, int id);
1229   // Requres that "objs" does not contains an ObjectValue whose id matches
1230   // that of "sv.  Appends "sv".
1231   static void set_sv_for_object_node(GrowableArray<ScopeValue*> *objs,
1232                                      ObjectValue* sv );
1233 
1234   // Process an OopMap Element while emitting nodes
1235   void Process_OopMap_Node(MachNode *mach, int code_offset);
1236 
1237   // Initialize code buffer
1238   CodeBuffer* init_buffer(uint* blk_starts);
1239 
1240   // Write out basic block data to code buffer
1241   void fill_buffer(CodeBuffer* cb, uint* blk_starts);
1242 
1243   // Determine which variable sized branches can be shortened
1244   void shorten_branches(uint* blk_starts, int& code_size, int& reloc_size, int& stub_size);
1245 
1246   // Compute the size of first NumberOfLoopInstrToAlign instructions
1247   // at the head of a loop.
1248   void compute_loop_first_inst_sizes();
1249 
1250   // Compute the information for the exception tables
1251   void FillExceptionTables(uint cnt, uint *call_returns, uint *inct_starts, Label *blk_labels);
1252 
1253   // Stack slots that may be unused by the calling convention but must
1254   // otherwise be preserved.  On Intel this includes the return address.
1255   // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1256   uint in_preserve_stack_slots();
1257 
1258   // "Top of Stack" slots that may be unused by the calling convention but must
1259   // otherwise be preserved.
1260   // On Intel these are not necessary and the value can be zero.
1261   // On Sparc this describes the words reserved for storing a register window
1262   // when an interrupt occurs.
1263   static uint out_preserve_stack_slots();
1264 
1265   // Number of outgoing stack slots killed above the out_preserve_stack_slots
1266   // for calls to C.  Supports the var-args backing area for register parms.
1267   uint varargs_C_out_slots_killed() const;
1268 
1269   // Number of Stack Slots consumed by a synchronization entry
1270   int sync_stack_slots() const;
1271 
1272   // Compute the name of old_SP.  See <arch>.ad for frame layout.
1273   OptoReg::Name compute_old_SP();
1274 
1275  private:
1276   // Phase control:
1277   void Init(int aliaslevel);                     // Prepare for a single compilation
1278   int  Inline_Warm();                            // Find more inlining work.
1279   void Finish_Warm();                            // Give up on further inlines.
1280   void Optimize();                               // Given a graph, optimize it
1281   void Code_Gen();                               // Generate code from a graph
1282 
1283   // Management of the AliasType table.
1284   void grow_alias_types();
1285   AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1286   const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1287   AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field);
1288 
1289   void verify_top(Node*) const PRODUCT_RETURN;
1290 
1291   // Intrinsic setup.
1292   void           register_library_intrinsics();                            // initializer
1293   CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual);          // constructor
1294   int            intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found);  // helper
1295   CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual);             // query fn
1296   void           register_intrinsic(CallGenerator* cg);                    // update fn
1297 
1298 #ifndef PRODUCT
1299   static juint  _intrinsic_hist_count[vmIntrinsics::ID_LIMIT];
1300   static jubyte _intrinsic_hist_flags[vmIntrinsics::ID_LIMIT];
1301 #endif
1302   // Function calls made by the public function final_graph_reshaping.
1303   // No need to be made public as they are not called elsewhere.
1304   void final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &frc);
1305   void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop);
1306   void final_graph_reshaping_walk( Node_Stack &nstack, Node *root, Final_Reshape_Counts &frc );
1307   void eliminate_redundant_card_marks(Node* n);
1308 
1309  public:
1310 
1311   // Note:  Histogram array size is about 1 Kb.
1312   enum {                        // flag bits:
1313     _intrinsic_worked = 1,      // succeeded at least once
1314     _intrinsic_failed = 2,      // tried it but it failed
1315     _intrinsic_disabled = 4,    // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1316     _intrinsic_virtual = 8,     // was seen in the virtual form (rare)
1317     _intrinsic_both = 16        // was seen in the non-virtual form (usual)
1318   };
1319   // Update histogram.  Return boolean if this is a first-time occurrence.
1320   static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1321                                           bool is_virtual, int flags) PRODUCT_RETURN0;
1322   static void print_intrinsic_statistics() PRODUCT_RETURN;
1323 
1324   // Graph verification code
1325   // Walk the node list, verifying that there is a one-to-one
1326   // correspondence between Use-Def edges and Def-Use edges
1327   // The option no_dead_code enables stronger checks that the
1328   // graph is strongly connected from root in both directions.
1329   void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN;
1330 
1331   // End-of-run dumps.
1332   static void print_statistics() PRODUCT_RETURN;
1333 
1334   // Dump formatted assembly
1335   void dump_asm(int *pcs = NULL, uint pc_limit = 0) PRODUCT_RETURN;
1336   void dump_pc(int *pcs, int pc_limit, Node *n);
1337 
1338   // Verify ADLC assumptions during startup
1339   static void adlc_verification() PRODUCT_RETURN;
1340 
1341   // Definitions of pd methods
1342   static void pd_compiler2_init();
1343 
1344   // Static parse-time type checking logic for gen_subtype_check:
1345   enum { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1346   int static_subtype_check(ciKlass* superk, ciKlass* subk);
1347 
1348   static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1349                               // Optional control dependency (for example, on range check)
1350                               Node* ctrl = NULL);
1351 
1352   // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1353   static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl);
1354 
1355   // Auxiliary method for randomized fuzzing/stressing
1356   static bool randomized_select(int count);
1357 
1358   // supporting clone_map
1359   CloneMap&     clone_map();
1360   void          set_clone_map(Dict* d);
1361 };
1362 
1363 #endif // SHARE_VM_OPTO_COMPILE_HPP