1 /*
   2  * Copyright (c) 1997, 2019, 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_OPTO_COMPILE_HPP
  26 #define SHARE_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 IdealGraphPrinter;
  56 class InlineTree;
  57 class Int_Array;
  58 class LoadBarrierNode;
  59 class Matcher;
  60 class MachConstantNode;
  61 class MachConstantBaseNode;
  62 class MachNode;
  63 class MachOper;
  64 class MachSafePointNode;
  65 class Node;
  66 class Node_Array;
  67 class Node_Notes;
  68 class NodeCloneInfo;
  69 class OptoReg;
  70 class PhaseCFG;
  71 class PhaseGVN;
  72 class PhaseIterGVN;
  73 class PhaseRegAlloc;
  74 class PhaseCCP;
  75 class PhaseCCP_DCE;
  76 class RootNode;
  77 class relocInfo;
  78 class Scope;
  79 class StartNode;
  80 class SafePointNode;
  81 class JVMState;
  82 class Type;
  83 class TypeData;
  84 class TypeInt;
  85 class TypePtr;
  86 class TypeOopPtr;
  87 class TypeFunc;
  88 class Unique_Node_List;
  89 class nmethod;
  90 class WarmCallInfo;
  91 class Node_Stack;
  92 struct Final_Reshape_Counts;
  93 
  94 enum LoopOptsMode {
  95   LoopOptsDefault,
  96   LoopOptsNone,
  97   LoopOptsShenandoahExpand,
  98   LoopOptsShenandoahPostExpand,
  99   LoopOptsZBarrierInsertion,
 100   LoopOptsSkipSplitIf,
 101   LoopOptsVerify
 102 };
 103 
 104 typedef unsigned int node_idx_t;
 105 class NodeCloneInfo {
 106  private:
 107   uint64_t _idx_clone_orig;
 108  public:
 109 
 110   void set_idx(node_idx_t idx) {
 111     _idx_clone_orig = (_idx_clone_orig & CONST64(0xFFFFFFFF00000000)) | idx;
 112   }
 113   node_idx_t idx() const { return (node_idx_t)(_idx_clone_orig & 0xFFFFFFFF); }
 114 
 115   void set_gen(int generation) {
 116     uint64_t g = (uint64_t)generation << 32;
 117     _idx_clone_orig = (_idx_clone_orig & 0xFFFFFFFF) | g;
 118   }
 119   int gen() const { return (int)(_idx_clone_orig >> 32); }
 120 
 121   void set(uint64_t x) { _idx_clone_orig = x; }
 122   void set(node_idx_t x, int g) { set_idx(x); set_gen(g); }
 123   uint64_t get() const { return _idx_clone_orig; }
 124 
 125   NodeCloneInfo(uint64_t idx_clone_orig) : _idx_clone_orig(idx_clone_orig) {}
 126   NodeCloneInfo(node_idx_t x, int g) : _idx_clone_orig(0) { set(x, g); }
 127 
 128   void dump() const;
 129 };
 130 
 131 class CloneMap {
 132   friend class Compile;
 133  private:
 134   bool      _debug;
 135   Dict*     _dict;
 136   int       _clone_idx;   // current cloning iteration/generation in loop unroll
 137  public:
 138   void*     _2p(node_idx_t key)   const          { return (void*)(intptr_t)key; } // 2 conversion functions to make gcc happy
 139   node_idx_t _2_node_idx_t(const void* k) const  { return (node_idx_t)(intptr_t)k; }
 140   Dict*     dict()                const          { return _dict; }
 141   void insert(node_idx_t key, uint64_t val)      { assert(_dict->operator[](_2p(key)) == NULL, "key existed"); _dict->Insert(_2p(key), (void*)val); }
 142   void insert(node_idx_t key, NodeCloneInfo& ci) { insert(key, ci.get()); }
 143   void remove(node_idx_t key)                    { _dict->Delete(_2p(key)); }
 144   uint64_t value(node_idx_t key)  const          { return (uint64_t)_dict->operator[](_2p(key)); }
 145   node_idx_t idx(node_idx_t key)  const          { return NodeCloneInfo(value(key)).idx(); }
 146   int gen(node_idx_t key)         const          { return NodeCloneInfo(value(key)).gen(); }
 147   int gen(const void* k)          const          { return gen(_2_node_idx_t(k)); }
 148   int max_gen()                   const;
 149   void clone(Node* old, Node* nnn, int gen);
 150   void verify_insert_and_clone(Node* old, Node* nnn, int gen);
 151   void dump(node_idx_t key)       const;
 152 
 153   int  clone_idx() const                         { return _clone_idx; }
 154   void set_clone_idx(int x)                      { _clone_idx = x; }
 155   bool is_debug()                 const          { return _debug; }
 156   void set_debug(bool debug)                     { _debug = debug; }
 157   static const char* debug_option_name;
 158 
 159   bool same_idx(node_idx_t k1, node_idx_t k2)  const { return idx(k1) == idx(k2); }
 160   bool same_gen(node_idx_t k1, node_idx_t k2)  const { return gen(k1) == gen(k2); }
 161 };
 162 
 163 //------------------------------Compile----------------------------------------
 164 // This class defines a top-level Compiler invocation.
 165 
 166 class Compile : public Phase {
 167   friend class VMStructs;
 168 
 169  public:
 170   // Fixed alias indexes.  (See also MergeMemNode.)
 171   enum {
 172     AliasIdxTop = 1,  // pseudo-index, aliases to nothing (used as sentinel value)
 173     AliasIdxBot = 2,  // pseudo-index, aliases to everything
 174     AliasIdxRaw = 3   // hard-wired index for TypeRawPtr::BOTTOM
 175   };
 176 
 177   // Variant of TraceTime(NULL, &_t_accumulator, CITime);
 178   // Integrated with logging.  If logging is turned on, and CITimeVerbose is true,
 179   // then brackets are put into the log, with time stamps and node counts.
 180   // (The time collection itself is always conditionalized on CITime.)
 181   class TracePhase : public TraceTime {
 182    private:
 183     Compile*    C;
 184     CompileLog* _log;
 185     const char* _phase_name;
 186     bool _dolog;
 187    public:
 188     TracePhase(const char* name, elapsedTimer* accumulator);
 189     ~TracePhase();
 190   };
 191 
 192   // Information per category of alias (memory slice)
 193   class AliasType {
 194    private:
 195     friend class Compile;
 196 
 197     int             _index;         // unique index, used with MergeMemNode
 198     const TypePtr*  _adr_type;      // normalized address type
 199     ciField*        _field;         // relevant instance field, or null if none
 200     const Type*     _element;       // relevant array element type, or null if none
 201     bool            _is_rewritable; // false if the memory is write-once only
 202     int             _general_index; // if this is type is an instance, the general
 203                                     // type that this is an instance of
 204 
 205     void Init(int i, const TypePtr* at);
 206 
 207    public:
 208     int             index()         const { return _index; }
 209     const TypePtr*  adr_type()      const { return _adr_type; }
 210     ciField*        field()         const { return _field; }
 211     const Type*     element()       const { return _element; }
 212     bool            is_rewritable() const { return _is_rewritable; }
 213     bool            is_volatile()   const { return (_field ? _field->is_volatile() : false); }
 214     int             general_index() const { return (_general_index != 0) ? _general_index : _index; }
 215 
 216     void set_rewritable(bool z) { _is_rewritable = z; }
 217     void set_field(ciField* f) {
 218       assert(!_field,"");
 219       _field = f;
 220       if (f->is_final() || f->is_stable()) {
 221         // In the case of @Stable, multiple writes are possible but may be assumed to be no-ops.
 222         _is_rewritable = false;
 223       }
 224     }
 225     void set_element(const Type* e) {
 226       assert(_element == NULL, "");
 227       _element = e;
 228     }
 229 
 230     BasicType basic_type() const;
 231 
 232     void print_on(outputStream* st) PRODUCT_RETURN;
 233   };
 234 
 235   enum {
 236     logAliasCacheSize = 6,
 237     AliasCacheSize = (1<<logAliasCacheSize)
 238   };
 239   struct AliasCacheEntry { const TypePtr* _adr_type; int _index; };  // simple duple type
 240   enum {
 241     trapHistLength = MethodData::_trap_hist_limit
 242   };
 243 
 244   // Constant entry of the constant table.
 245   class Constant {
 246   private:
 247     BasicType _type;
 248     union {
 249       jvalue    _value;
 250       Metadata* _metadata;
 251     } _v;
 252     int       _offset;         // offset of this constant (in bytes) relative to the constant table base.
 253     float     _freq;
 254     bool      _can_be_reused;  // true (default) if the value can be shared with other users.
 255 
 256   public:
 257     Constant() : _type(T_ILLEGAL), _offset(-1), _freq(0.0f), _can_be_reused(true) { _v._value.l = 0; }
 258     Constant(BasicType type, jvalue value, float freq = 0.0f, bool can_be_reused = true) :
 259       _type(type),
 260       _offset(-1),
 261       _freq(freq),
 262       _can_be_reused(can_be_reused)
 263     {
 264       assert(type != T_METADATA, "wrong constructor");
 265       _v._value = value;
 266     }
 267     Constant(Metadata* metadata, bool can_be_reused = true) :
 268       _type(T_METADATA),
 269       _offset(-1),
 270       _freq(0.0f),
 271       _can_be_reused(can_be_reused)
 272     {
 273       _v._metadata = metadata;
 274     }
 275 
 276     bool operator==(const Constant& other);
 277 
 278     BasicType type()      const    { return _type; }
 279 
 280     jint    get_jint()    const    { return _v._value.i; }
 281     jlong   get_jlong()   const    { return _v._value.j; }
 282     jfloat  get_jfloat()  const    { return _v._value.f; }
 283     jdouble get_jdouble() const    { return _v._value.d; }
 284     jobject get_jobject() const    { return _v._value.l; }
 285 
 286     Metadata* get_metadata() const { return _v._metadata; }
 287 
 288     int         offset()  const    { return _offset; }
 289     void    set_offset(int offset) {        _offset = offset; }
 290 
 291     float       freq()    const    { return _freq;         }
 292     void    inc_freq(float freq)   {        _freq += freq; }
 293 
 294     bool    can_be_reused() const  { return _can_be_reused; }
 295   };
 296 
 297   // Constant table.
 298   class ConstantTable {
 299   private:
 300     GrowableArray<Constant> _constants;          // Constants of this table.
 301     int                     _size;               // Size in bytes the emitted constant table takes (including padding).
 302     int                     _table_base_offset;  // Offset of the table base that gets added to the constant offsets.
 303     int                     _nof_jump_tables;    // Number of jump-tables in this constant table.
 304 
 305     static int qsort_comparator(Constant* a, Constant* b);
 306 
 307     // We use negative frequencies to keep the order of the
 308     // jump-tables in which they were added.  Otherwise we get into
 309     // trouble with relocation.
 310     float next_jump_table_freq() { return -1.0f * (++_nof_jump_tables); }
 311 
 312   public:
 313     ConstantTable() :
 314       _size(-1),
 315       _table_base_offset(-1),  // We can use -1 here since the constant table is always bigger than 2 bytes (-(size / 2), see MachConstantBaseNode::emit).
 316       _nof_jump_tables(0)
 317     {}
 318 
 319     int size() const { assert(_size != -1, "not calculated yet"); return _size; }
 320 
 321     int calculate_table_base_offset() const;  // AD specific
 322     void set_table_base_offset(int x)  { assert(_table_base_offset == -1 || x == _table_base_offset, "can't change"); _table_base_offset = x; }
 323     int      table_base_offset() const { assert(_table_base_offset != -1, "not set yet");                      return _table_base_offset; }
 324 
 325     void emit(CodeBuffer& cb);
 326 
 327     // Returns the offset of the last entry (the top) of the constant table.
 328     int  top_offset() const { assert(_constants.top().offset() != -1, "not bound yet"); return _constants.top().offset(); }
 329 
 330     void calculate_offsets_and_size();
 331     int  find_offset(Constant& con) const;
 332 
 333     void     add(Constant& con);
 334     Constant add(MachConstantNode* n, BasicType type, jvalue value);
 335     Constant add(Metadata* metadata);
 336     Constant add(MachConstantNode* n, MachOper* oper);
 337     Constant add(MachConstantNode* n, jint i) {
 338       jvalue value; value.i = i;
 339       return add(n, T_INT, value);
 340     }
 341     Constant add(MachConstantNode* n, jlong j) {
 342       jvalue value; value.j = j;
 343       return add(n, T_LONG, value);
 344     }
 345     Constant add(MachConstantNode* n, jfloat f) {
 346       jvalue value; value.f = f;
 347       return add(n, T_FLOAT, value);
 348     }
 349     Constant add(MachConstantNode* n, jdouble d) {
 350       jvalue value; value.d = d;
 351       return add(n, T_DOUBLE, value);
 352     }
 353 
 354     // Jump-table
 355     Constant  add_jump_table(MachConstantNode* n);
 356     void     fill_jump_table(CodeBuffer& cb, MachConstantNode* n, GrowableArray<Label*> labels) const;
 357   };
 358 
 359  private:
 360   // Fixed parameters to this compilation.
 361   const int             _compile_id;
 362   const bool            _save_argument_registers; // save/restore arg regs for trampolines
 363   const bool            _subsume_loads;         // Load can be matched as part of a larger op.
 364   const bool            _do_escape_analysis;    // Do escape analysis.
 365   const bool            _eliminate_boxing;      // Do boxing elimination.
 366   ciMethod*             _method;                // The method being compiled.
 367   int                   _entry_bci;             // entry bci for osr methods.
 368   const TypeFunc*       _tf;                    // My kind of signature
 369   InlineTree*           _ilt;                   // Ditto (temporary).
 370   address               _stub_function;         // VM entry for stub being compiled, or NULL
 371   const char*           _stub_name;             // Name of stub or adapter being compiled, or NULL
 372   address               _stub_entry_point;      // Compile code entry for generated stub, or NULL
 373 
 374   // Control of this compilation.
 375   int                   _max_inline_size;       // Max inline size for this compilation
 376   int                   _freq_inline_size;      // Max hot method inline size for this compilation
 377   int                   _fixed_slots;           // count of frame slots not allocated by the register
 378                                                 // allocator i.e. locks, original deopt pc, etc.
 379   uintx                 _max_node_limit;        // Max unique node count during a single compilation.
 380   // For deopt
 381   int                   _orig_pc_slot;
 382   int                   _orig_pc_slot_offset_in_bytes;
 383 
 384   int                   _major_progress;        // Count of something big happening
 385   bool                  _inlining_progress;     // progress doing incremental inlining?
 386   bool                  _inlining_incrementally;// Are we doing incremental inlining (post parse)
 387   bool                  _do_cleanup;            // Cleanup is needed before proceeding with incremental inlining
 388   bool                  _has_loops;             // True if the method _may_ have some loops
 389   bool                  _has_split_ifs;         // True if the method _may_ have some split-if
 390   bool                  _has_unsafe_access;     // True if the method _may_ produce faults in unsafe loads or stores.
 391   bool                  _has_stringbuilder;     // True StringBuffers or StringBuilders are allocated
 392   bool                  _has_boxed_value;       // True if a boxed object is allocated
 393   bool                  _has_reserved_stack_access; // True if the method or an inlined method is annotated with ReservedStackAccess
 394   uint                  _max_vector_size;       // Maximum size of generated vectors
 395   bool                  _clear_upper_avx;       // Clear upper bits of ymm registers using vzeroupper
 396   uint                  _trap_hist[trapHistLength];  // Cumulative traps
 397   bool                  _trap_can_recompile;    // Have we emitted a recompiling trap?
 398   uint                  _decompile_count;       // Cumulative decompilation counts.
 399   bool                  _do_inlining;           // True if we intend to do inlining
 400   bool                  _do_scheduling;         // True if we intend to do scheduling
 401   bool                  _do_freq_based_layout;  // True if we intend to do frequency based block layout
 402   bool                  _do_count_invocations;  // True if we generate code to count invocations
 403   bool                  _do_method_data_update; // True if we generate code to update MethodData*s
 404   bool                  _do_vector_loop;        // True if allowed to execute loop in parallel iterations
 405   bool                  _use_cmove;             // True if CMove should be used without profitability analysis
 406   bool                  _age_code;              // True if we need to profile code age (decrement the aging counter)
 407   int                   _AliasLevel;            // Locally-adjusted version of AliasLevel flag.
 408   bool                  _print_assembly;        // True if we should dump assembly code for this compilation
 409   bool                  _print_inlining;        // True if we should print inlining for this compilation
 410   bool                  _print_intrinsics;      // True if we should print intrinsics for this compilation
 411 #ifndef PRODUCT
 412   bool                  _trace_opto_output;
 413   bool                  _parsed_irreducible_loop; // True if ciTypeFlow detected irreducible loops during parsing
 414 #endif
 415   bool                  _has_irreducible_loop;  // Found irreducible loops
 416   // JSR 292
 417   bool                  _has_method_handle_invokes; // True if this method has MethodHandle invokes.
 418   RTMState              _rtm_state;             // State of Restricted Transactional Memory usage
 419   int                   _loop_opts_cnt;         // loop opts round
 420   bool                  _clinit_barrier_on_entry; // True if clinit barrier is needed on nmethod entry
 421 
 422   // Compilation environment.
 423   Arena                 _comp_arena;            // Arena with lifetime equivalent to Compile
 424   void*                 _barrier_set_state;     // Potential GC barrier state for Compile
 425   ciEnv*                _env;                   // CI interface
 426   DirectiveSet*         _directive;             // Compiler directive
 427   CompileLog*           _log;                   // from CompilerThread
 428   const char*           _failure_reason;        // for record_failure/failing pattern
 429   GrowableArray<CallGenerator*>* _intrinsics;   // List of intrinsics.
 430   GrowableArray<Node*>* _macro_nodes;           // List of nodes which need to be expanded before matching.
 431   GrowableArray<Node*>* _predicate_opaqs;       // List of Opaque1 nodes for the loop predicates.
 432   GrowableArray<Node*>* _expensive_nodes;       // List of nodes that are expensive to compute and that we'd better not let the GVN freely common
 433   GrowableArray<Node*>* _range_check_casts;     // List of CastII nodes with a range check dependency
 434   GrowableArray<Node*>* _opaque4_nodes;         // List of Opaque4 nodes that have a default value
 435   ConnectionGraph*      _congraph;
 436 #ifndef PRODUCT
 437   IdealGraphPrinter*    _printer;
 438 #endif
 439 
 440 
 441   // Node management
 442   uint                  _unique;                // Counter for unique Node indices
 443   VectorSet             _dead_node_list;        // Set of dead nodes
 444   uint                  _dead_node_count;       // Number of dead nodes; VectorSet::Size() is O(N).
 445                                                 // So use this to keep count and make the call O(1).
 446   DEBUG_ONLY( Unique_Node_List* _modified_nodes; )  // List of nodes which inputs were modified
 447 
 448   debug_only(static int _debug_idx;)            // Monotonic counter (not reset), use -XX:BreakAtNode=<idx>
 449   Arena                 _node_arena;            // Arena for new-space Nodes
 450   Arena                 _old_arena;             // Arena for old-space Nodes, lifetime during xform
 451   RootNode*             _root;                  // Unique root of compilation, or NULL after bail-out.
 452   Node*                 _top;                   // Unique top node.  (Reset by various phases.)
 453 
 454   Node*                 _immutable_memory;      // Initial memory state
 455 
 456   Node*                 _recent_alloc_obj;
 457   Node*                 _recent_alloc_ctl;
 458 
 459   // Constant table
 460   ConstantTable         _constant_table;        // The constant table for this compile.
 461   MachConstantBaseNode* _mach_constant_base_node;  // Constant table base node singleton.
 462 
 463 
 464   // Blocked array of debugging and profiling information,
 465   // tracked per node.
 466   enum { _log2_node_notes_block_size = 8,
 467          _node_notes_block_size = (1<<_log2_node_notes_block_size)
 468   };
 469   GrowableArray<Node_Notes*>* _node_note_array;
 470   Node_Notes*           _default_node_notes;  // default notes for new nodes
 471 
 472   // After parsing and every bulk phase we hang onto the Root instruction.
 473   // The RootNode instruction is where the whole program begins.  It produces
 474   // the initial Control and BOTTOM for everybody else.
 475 
 476   // Type management
 477   Arena                 _Compile_types;         // Arena for all types
 478   Arena*                _type_arena;            // Alias for _Compile_types except in Initialize_shared()
 479   Dict*                 _type_dict;             // Intern table
 480   CloneMap              _clone_map;             // used for recording history of cloned nodes
 481   size_t                _type_last_size;        // Last allocation size (see Type::operator new/delete)
 482   ciMethod*             _last_tf_m;             // Cache for
 483   const TypeFunc*       _last_tf;               //  TypeFunc::make
 484   AliasType**           _alias_types;           // List of alias types seen so far.
 485   int                   _num_alias_types;       // Logical length of _alias_types
 486   int                   _max_alias_types;       // Physical length of _alias_types
 487   AliasCacheEntry       _alias_cache[AliasCacheSize]; // Gets aliases w/o data structure walking
 488 
 489   // Parsing, optimization
 490   PhaseGVN*             _initial_gvn;           // Results of parse-time PhaseGVN
 491   Unique_Node_List*     _for_igvn;              // Initial work-list for next round of Iterative GVN
 492   WarmCallInfo*         _warm_calls;            // Sorted work-list for heat-based inlining.
 493 
 494   GrowableArray<CallGenerator*> _late_inlines;        // List of CallGenerators to be revisited after
 495                                                       // main parsing has finished.
 496   GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
 497 
 498   GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
 499 
 500   int                           _late_inlines_pos;    // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
 501   uint                          _number_of_mh_late_inlines; // number of method handle late inlining still pending
 502 
 503 
 504   // Inlining may not happen in parse order which would make
 505   // PrintInlining output confusing. Keep track of PrintInlining
 506   // pieces in order.
 507   class PrintInliningBuffer : public ResourceObj {
 508    private:
 509     CallGenerator* _cg;
 510     stringStream* _ss;
 511 
 512    public:
 513     PrintInliningBuffer()
 514       : _cg(NULL) { _ss = new stringStream(); }
 515 
 516     stringStream* ss() const { return _ss; }
 517     CallGenerator* cg() const { return _cg; }
 518     void set_cg(CallGenerator* cg) { _cg = cg; }
 519   };
 520 
 521   stringStream* _print_inlining_stream;
 522   GrowableArray<PrintInliningBuffer>* _print_inlining_list;
 523   int _print_inlining_idx;
 524   char* _print_inlining_output;
 525 
 526   // Only keep nodes in the expensive node list that need to be optimized
 527   void cleanup_expensive_nodes(PhaseIterGVN &igvn);
 528   // Use for sorting expensive nodes to bring similar nodes together
 529   static int cmp_expensive_nodes(Node** n1, Node** n2);
 530   // Expensive nodes list already sorted?
 531   bool expensive_nodes_sorted() const;
 532   // Remove the speculative part of types and clean up the graph
 533   void remove_speculative_types(PhaseIterGVN &igvn);
 534 
 535   void* _replay_inline_data; // Pointer to data loaded from file
 536 
 537   void print_inlining_init();
 538   void print_inlining_reinit();
 539   void print_inlining_commit();
 540   void print_inlining_push();
 541   PrintInliningBuffer& print_inlining_current();
 542 
 543   void log_late_inline_failure(CallGenerator* cg, const char* msg);
 544 
 545  public:
 546 
 547   void* barrier_set_state() const { return _barrier_set_state; }
 548 
 549   outputStream* print_inlining_stream() const {
 550     assert(print_inlining() || print_intrinsics(), "PrintInlining off?");
 551     return _print_inlining_stream;
 552   }
 553 
 554   void print_inlining_update(CallGenerator* cg);
 555   void print_inlining_update_delayed(CallGenerator* cg);
 556   void print_inlining_move_to(CallGenerator* cg);
 557   void print_inlining_assert_ready();
 558   void print_inlining_reset();
 559 
 560   void print_inlining(ciMethod* method, int inline_level, int bci, const char* msg = NULL) {
 561     stringStream ss;
 562     CompileTask::print_inlining_inner(&ss, method, inline_level, bci, msg);
 563     print_inlining_stream()->print("%s", ss.as_string());
 564   }
 565 
 566 #ifndef PRODUCT
 567   IdealGraphPrinter* printer() { return _printer; }
 568 #endif
 569 
 570   void log_late_inline(CallGenerator* cg);
 571   void log_inline_id(CallGenerator* cg);
 572   void log_inline_failure(const char* msg);
 573 
 574   void* replay_inline_data() const { return _replay_inline_data; }
 575 
 576   // Dump inlining replay data to the stream.
 577   void dump_inline_data(outputStream* out);
 578 
 579  private:
 580   // Matching, CFG layout, allocation, code generation
 581   PhaseCFG*             _cfg;                   // Results of CFG finding
 582   bool                  _select_24_bit_instr;   // We selected an instruction with a 24-bit result
 583   bool                  _in_24_bit_fp_mode;     // We are emitting instructions with 24-bit results
 584   int                   _java_calls;            // Number of java calls in the method
 585   int                   _inner_loops;           // Number of inner loops in the method
 586   Matcher*              _matcher;               // Engine to map ideal to machine instructions
 587   PhaseRegAlloc*        _regalloc;              // Results of register allocation.
 588   int                   _frame_slots;           // Size of total frame in stack slots
 589   CodeOffsets           _code_offsets;          // Offsets into the code for various interesting entries
 590   RegMask               _FIRST_STACK_mask;      // All stack slots usable for spills (depends on frame layout)
 591   Arena*                _indexSet_arena;        // control IndexSet allocation within PhaseChaitin
 592   void*                 _indexSet_free_block_list; // free list of IndexSet bit blocks
 593   int                   _interpreter_frame_size;
 594 
 595   uint                  _node_bundling_limit;
 596   Bundle*               _node_bundling_base;    // Information for instruction bundling
 597 
 598   // Instruction bits passed off to the VM
 599   int                   _method_size;           // Size of nmethod code segment in bytes
 600   CodeBuffer            _code_buffer;           // Where the code is assembled
 601   int                   _first_block_size;      // Size of unvalidated entry point code / OSR poison code
 602   ExceptionHandlerTable _handler_table;         // Table of native-code exception handlers
 603   ImplicitExceptionTable _inc_table;            // Table of implicit null checks in native code
 604   OopMapSet*            _oop_map_set;           // Table of oop maps (one for each safepoint location)
 605   static int            _CompiledZap_count;     // counter compared against CompileZap[First/Last]
 606   BufferBlob*           _scratch_buffer_blob;   // For temporary code buffers.
 607   relocInfo*            _scratch_locs_memory;   // For temporary code buffers.
 608   int                   _scratch_const_size;    // For temporary code buffers.
 609   bool                  _in_scratch_emit_size;  // true when in scratch_emit_size.
 610 
 611   void reshape_address(AddPNode* n);
 612 
 613  public:
 614   // Accessors
 615 
 616   // The Compile instance currently active in this (compiler) thread.
 617   static Compile* current() {
 618     return (Compile*) ciEnv::current()->compiler_data();
 619   }
 620 
 621   // ID for this compilation.  Useful for setting breakpoints in the debugger.
 622   int               compile_id() const          { return _compile_id; }
 623   DirectiveSet*     directive() const           { return _directive; }
 624 
 625   // Does this compilation allow instructions to subsume loads?  User
 626   // instructions that subsume a load may result in an unschedulable
 627   // instruction sequence.
 628   bool              subsume_loads() const       { return _subsume_loads; }
 629   /** Do escape analysis. */
 630   bool              do_escape_analysis() const  { return _do_escape_analysis; }
 631   /** Do boxing elimination. */
 632   bool              eliminate_boxing() const    { return _eliminate_boxing; }
 633   /** Do aggressive boxing elimination. */
 634   bool              aggressive_unboxing() const { return _eliminate_boxing && AggressiveUnboxing; }
 635   bool              save_argument_registers() const { return _save_argument_registers; }
 636 
 637 
 638   // Other fixed compilation parameters.
 639   ciMethod*         method() const              { return _method; }
 640   int               entry_bci() const           { return _entry_bci; }
 641   bool              is_osr_compilation() const  { return _entry_bci != InvocationEntryBci; }
 642   bool              is_method_compilation() const { return (_method != NULL && !_method->flags().is_native()); }
 643   const TypeFunc*   tf() const                  { assert(_tf!=NULL, ""); return _tf; }
 644   void         init_tf(const TypeFunc* tf)      { assert(_tf==NULL, ""); _tf = tf; }
 645   InlineTree*       ilt() const                 { return _ilt; }
 646   address           stub_function() const       { return _stub_function; }
 647   const char*       stub_name() const           { return _stub_name; }
 648   address           stub_entry_point() const    { return _stub_entry_point; }
 649 
 650   // Control of this compilation.
 651   int               fixed_slots() const         { assert(_fixed_slots >= 0, "");         return _fixed_slots; }
 652   void          set_fixed_slots(int n)          { _fixed_slots = n; }
 653   int               major_progress() const      { return _major_progress; }
 654   void          set_inlining_progress(bool z)   { _inlining_progress = z; }
 655   int               inlining_progress() const   { return _inlining_progress; }
 656   void          set_inlining_incrementally(bool z) { _inlining_incrementally = z; }
 657   int               inlining_incrementally() const { return _inlining_incrementally; }
 658   void          set_do_cleanup(bool z)          { _do_cleanup = z; }
 659   int               do_cleanup() const          { return _do_cleanup; }
 660   void          set_major_progress()            { _major_progress++; }
 661   void          restore_major_progress(int progress) { _major_progress += progress; }
 662   void        clear_major_progress()            { _major_progress = 0; }
 663   int               max_inline_size() const     { return _max_inline_size; }
 664   void          set_freq_inline_size(int n)     { _freq_inline_size = n; }
 665   int               freq_inline_size() const    { return _freq_inline_size; }
 666   void          set_max_inline_size(int n)      { _max_inline_size = n; }
 667   bool              has_loops() const           { return _has_loops; }
 668   void          set_has_loops(bool z)           { _has_loops = z; }
 669   bool              has_split_ifs() const       { return _has_split_ifs; }
 670   void          set_has_split_ifs(bool z)       { _has_split_ifs = z; }
 671   bool              has_unsafe_access() const   { return _has_unsafe_access; }
 672   void          set_has_unsafe_access(bool z)   { _has_unsafe_access = z; }
 673   bool              has_stringbuilder() const   { return _has_stringbuilder; }
 674   void          set_has_stringbuilder(bool z)   { _has_stringbuilder = z; }
 675   bool              has_boxed_value() const     { return _has_boxed_value; }
 676   void          set_has_boxed_value(bool z)     { _has_boxed_value = z; }
 677   bool              has_reserved_stack_access() const { return _has_reserved_stack_access; }
 678   void          set_has_reserved_stack_access(bool z) { _has_reserved_stack_access = z; }
 679   uint              max_vector_size() const     { return _max_vector_size; }
 680   void          set_max_vector_size(uint s)     { _max_vector_size = s; }
 681   bool              clear_upper_avx() const     { return _clear_upper_avx; }
 682   void          set_clear_upper_avx(bool s)     { _clear_upper_avx = s; }
 683   void          set_trap_count(uint r, uint c)  { assert(r < trapHistLength, "oob");        _trap_hist[r] = c; }
 684   uint              trap_count(uint r) const    { assert(r < trapHistLength, "oob"); return _trap_hist[r]; }
 685   bool              trap_can_recompile() const  { return _trap_can_recompile; }
 686   void          set_trap_can_recompile(bool z)  { _trap_can_recompile = z; }
 687   uint              decompile_count() const     { return _decompile_count; }
 688   void          set_decompile_count(uint c)     { _decompile_count = c; }
 689   bool              allow_range_check_smearing() const;
 690   bool              do_inlining() const         { return _do_inlining; }
 691   void          set_do_inlining(bool z)         { _do_inlining = z; }
 692   bool              do_scheduling() const       { return _do_scheduling; }
 693   void          set_do_scheduling(bool z)       { _do_scheduling = z; }
 694   bool              do_freq_based_layout() const{ return _do_freq_based_layout; }
 695   void          set_do_freq_based_layout(bool z){ _do_freq_based_layout = z; }
 696   bool              do_count_invocations() const{ return _do_count_invocations; }
 697   void          set_do_count_invocations(bool z){ _do_count_invocations = z; }
 698   bool              do_method_data_update() const { return _do_method_data_update; }
 699   void          set_do_method_data_update(bool z) { _do_method_data_update = z; }
 700   bool              do_vector_loop() const      { return _do_vector_loop; }
 701   void          set_do_vector_loop(bool z)      { _do_vector_loop = z; }
 702   bool              use_cmove() const           { return _use_cmove; }
 703   void          set_use_cmove(bool z)           { _use_cmove = z; }
 704   bool              age_code() const             { return _age_code; }
 705   void          set_age_code(bool z)             { _age_code = z; }
 706   int               AliasLevel() const           { return _AliasLevel; }
 707   bool              print_assembly() const       { return _print_assembly; }
 708   void          set_print_assembly(bool z)       { _print_assembly = z; }
 709   bool              print_inlining() const       { return _print_inlining; }
 710   void          set_print_inlining(bool z)       { _print_inlining = z; }
 711   bool              print_intrinsics() const     { return _print_intrinsics; }
 712   void          set_print_intrinsics(bool z)     { _print_intrinsics = z; }
 713   RTMState          rtm_state()  const           { return _rtm_state; }
 714   void          set_rtm_state(RTMState s)        { _rtm_state = s; }
 715   bool              use_rtm() const              { return (_rtm_state & NoRTM) == 0; }
 716   bool          profile_rtm() const              { return _rtm_state == ProfileRTM; }
 717   uint              max_node_limit() const       { return (uint)_max_node_limit; }
 718   void          set_max_node_limit(uint n)       { _max_node_limit = n; }
 719   bool              clinit_barrier_on_entry()       { return _clinit_barrier_on_entry; }
 720   void          set_clinit_barrier_on_entry(bool z) { _clinit_barrier_on_entry = z; }
 721 
 722   // check the CompilerOracle for special behaviours for this compile
 723   bool          method_has_option(const char * option) {
 724     return method() != NULL && method()->has_option(option);
 725   }
 726 
 727 #ifndef PRODUCT
 728   bool          trace_opto_output() const       { return _trace_opto_output; }
 729   bool              parsed_irreducible_loop() const { return _parsed_irreducible_loop; }
 730   void          set_parsed_irreducible_loop(bool z) { _parsed_irreducible_loop = z; }
 731   int _in_dump_cnt;  // Required for dumping ir nodes.
 732 #endif
 733   bool              has_irreducible_loop() const { return _has_irreducible_loop; }
 734   void          set_has_irreducible_loop(bool z) { _has_irreducible_loop = z; }
 735 
 736   // JSR 292
 737   bool              has_method_handle_invokes() const { return _has_method_handle_invokes;     }
 738   void          set_has_method_handle_invokes(bool z) {        _has_method_handle_invokes = z; }
 739 
 740   Ticks _latest_stage_start_counter;
 741 
 742   void begin_method() {
 743 #ifndef PRODUCT
 744     if (_printer && _printer->should_print(1)) {
 745       _printer->begin_method();
 746     }
 747 #endif
 748     C->_latest_stage_start_counter.stamp();
 749   }
 750 
 751   bool should_print(int level = 1) {
 752 #ifndef PRODUCT
 753     return (_printer && _printer->should_print(level));
 754 #else
 755     return false;
 756 #endif
 757   }
 758 
 759   void print_method(CompilerPhaseType cpt, int level = 1, int idx = 0) {
 760     EventCompilerPhase event;
 761     if (event.should_commit()) {
 762       event.set_starttime(C->_latest_stage_start_counter);
 763       event.set_phase((u1) cpt);
 764       event.set_compileId(C->_compile_id);
 765       event.set_phaseLevel(level);
 766       event.commit();
 767     }
 768 
 769 #ifndef PRODUCT
 770     if (should_print(level)) {
 771       char output[1024];
 772       if (idx != 0) {
 773         sprintf(output, "%s:%d", CompilerPhaseTypeHelper::to_string(cpt), idx);
 774       } else {
 775         sprintf(output, "%s", CompilerPhaseTypeHelper::to_string(cpt));
 776       }
 777       _printer->print_method(output, level);
 778     }
 779 #endif
 780     C->_latest_stage_start_counter.stamp();
 781   }
 782 
 783   void end_method(int level = 1) {
 784     EventCompilerPhase event;
 785     if (event.should_commit()) {
 786       event.set_starttime(C->_latest_stage_start_counter);
 787       event.set_phase((u1) PHASE_END);
 788       event.set_compileId(C->_compile_id);
 789       event.set_phaseLevel(level);
 790       event.commit();
 791     }
 792 #ifndef PRODUCT
 793     if (_printer && _printer->should_print(level)) {
 794       _printer->end_method();
 795     }
 796 #endif
 797   }
 798 
 799   int           macro_count()             const { return _macro_nodes->length(); }
 800   int           predicate_count()         const { return _predicate_opaqs->length();}
 801   int           expensive_count()         const { return _expensive_nodes->length(); }
 802   Node*         macro_node(int idx)       const { return _macro_nodes->at(idx); }
 803   Node*         predicate_opaque1_node(int idx) const { return _predicate_opaqs->at(idx);}
 804   Node*         expensive_node(int idx)   const { return _expensive_nodes->at(idx); }
 805   ConnectionGraph* congraph()                   { return _congraph;}
 806   void set_congraph(ConnectionGraph* congraph)  { _congraph = congraph;}
 807   void add_macro_node(Node * n) {
 808     //assert(n->is_macro(), "must be a macro node");
 809     assert(!_macro_nodes->contains(n), "duplicate entry in expand list");
 810     _macro_nodes->append(n);
 811   }
 812   void remove_macro_node(Node * n) {
 813     // this function may be called twice for a node so check
 814     // that the node is in the array before attempting to remove it
 815     if (_macro_nodes->contains(n))
 816       _macro_nodes->remove(n);
 817     // remove from _predicate_opaqs list also if it is there
 818     if (predicate_count() > 0 && _predicate_opaqs->contains(n)){
 819       _predicate_opaqs->remove(n);
 820     }
 821   }
 822   void add_expensive_node(Node * n);
 823   void remove_expensive_node(Node * n) {
 824     if (_expensive_nodes->contains(n)) {
 825       _expensive_nodes->remove(n);
 826     }
 827   }
 828   void add_predicate_opaq(Node * n) {
 829     assert(!_predicate_opaqs->contains(n), "duplicate entry in predicate opaque1");
 830     assert(_macro_nodes->contains(n), "should have already been in macro list");
 831     _predicate_opaqs->append(n);
 832   }
 833 
 834   // Range check dependent CastII nodes that can be removed after loop optimizations
 835   void add_range_check_cast(Node* n);
 836   void remove_range_check_cast(Node* n) {
 837     if (_range_check_casts->contains(n)) {
 838       _range_check_casts->remove(n);
 839     }
 840   }
 841   Node* range_check_cast_node(int idx) const { return _range_check_casts->at(idx);  }
 842   int   range_check_cast_count()       const { return _range_check_casts->length(); }
 843   // Remove all range check dependent CastIINodes.
 844   void  remove_range_check_casts(PhaseIterGVN &igvn);
 845 
 846   void add_opaque4_node(Node* n);
 847   void remove_opaque4_node(Node* n) {
 848     if (_opaque4_nodes->contains(n)) {
 849       _opaque4_nodes->remove(n);
 850     }
 851   }
 852   Node* opaque4_node(int idx) const { return _opaque4_nodes->at(idx);  }
 853   int   opaque4_count()       const { return _opaque4_nodes->length(); }
 854   void  remove_opaque4_nodes(PhaseIterGVN &igvn);
 855 
 856   // remove the opaque nodes that protect the predicates so that the unused checks and
 857   // uncommon traps will be eliminated from the graph.
 858   void cleanup_loop_predicates(PhaseIterGVN &igvn);
 859   bool is_predicate_opaq(Node * n) {
 860     return _predicate_opaqs->contains(n);
 861   }
 862 
 863   // Are there candidate expensive nodes for optimization?
 864   bool should_optimize_expensive_nodes(PhaseIterGVN &igvn);
 865   // Check whether n1 and n2 are similar
 866   static int cmp_expensive_nodes(Node* n1, Node* n2);
 867   // Sort expensive nodes to locate similar expensive nodes
 868   void sort_expensive_nodes();
 869 
 870   // Compilation environment.
 871   Arena*      comp_arena()           { return &_comp_arena; }
 872   ciEnv*      env() const            { return _env; }
 873   CompileLog* log() const            { return _log; }
 874   bool        failing() const        { return _env->failing() || _failure_reason != NULL; }
 875   const char* failure_reason() const { return (_env->failing()) ? _env->failure_reason() : _failure_reason; }
 876 
 877   bool failure_reason_is(const char* r) const {
 878     return (r == _failure_reason) || (r != NULL && _failure_reason != NULL && strcmp(r, _failure_reason) == 0);
 879   }
 880 
 881   void record_failure(const char* reason);
 882   void record_method_not_compilable(const char* reason) {
 883     // Bailouts cover "all_tiers" when TieredCompilation is off.
 884     env()->record_method_not_compilable(reason, !TieredCompilation);
 885     // Record failure reason.
 886     record_failure(reason);
 887   }
 888   bool check_node_count(uint margin, const char* reason) {
 889     if (live_nodes() + margin > max_node_limit()) {
 890       record_method_not_compilable(reason);
 891       return true;
 892     } else {
 893       return false;
 894     }
 895   }
 896 
 897   // Node management
 898   uint         unique() const              { return _unique; }
 899   uint         next_unique()               { return _unique++; }
 900   void         set_unique(uint i)          { _unique = i; }
 901   static int   debug_idx()                 { return debug_only(_debug_idx)+0; }
 902   static void  set_debug_idx(int i)        { debug_only(_debug_idx = i); }
 903   Arena*       node_arena()                { return &_node_arena; }
 904   Arena*       old_arena()                 { return &_old_arena; }
 905   RootNode*    root() const                { return _root; }
 906   void         set_root(RootNode* r)       { _root = r; }
 907   StartNode*   start() const;              // (Derived from root.)
 908   void         init_start(StartNode* s);
 909   Node*        immutable_memory();
 910 
 911   Node*        recent_alloc_ctl() const    { return _recent_alloc_ctl; }
 912   Node*        recent_alloc_obj() const    { return _recent_alloc_obj; }
 913   void         set_recent_alloc(Node* ctl, Node* obj) {
 914                                                   _recent_alloc_ctl = ctl;
 915                                                   _recent_alloc_obj = obj;
 916                                            }
 917   void         record_dead_node(uint idx)  { if (_dead_node_list.test_set(idx)) return;
 918                                              _dead_node_count++;
 919                                            }
 920   bool         is_dead_node(uint idx)      { return _dead_node_list.test(idx) != 0; }
 921   uint         dead_node_count()           { return _dead_node_count; }
 922   void         reset_dead_node_list()      { _dead_node_list.Reset();
 923                                              _dead_node_count = 0;
 924                                            }
 925   uint          live_nodes() const         {
 926     int  val = _unique - _dead_node_count;
 927     assert (val >= 0, "number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count);
 928             return (uint) val;
 929                                            }
 930 #ifdef ASSERT
 931   uint         count_live_nodes_by_graph_walk();
 932   void         print_missing_nodes();
 933 #endif
 934 
 935   // Record modified nodes to check that they are put on IGVN worklist
 936   void         record_modified_node(Node* n) NOT_DEBUG_RETURN;
 937   void         remove_modified_node(Node* n) NOT_DEBUG_RETURN;
 938   DEBUG_ONLY( Unique_Node_List*   modified_nodes() const { return _modified_nodes; } )
 939 
 940   // Constant table
 941   ConstantTable&   constant_table() { return _constant_table; }
 942 
 943   MachConstantBaseNode*     mach_constant_base_node();
 944   bool                  has_mach_constant_base_node() const { return _mach_constant_base_node != NULL; }
 945   // Generated by adlc, true if CallNode requires MachConstantBase.
 946   bool                      needs_clone_jvms();
 947 
 948   // Handy undefined Node
 949   Node*             top() const                 { return _top; }
 950 
 951   // these are used by guys who need to know about creation and transformation of top:
 952   Node*             cached_top_node()           { return _top; }
 953   void          set_cached_top_node(Node* tn);
 954 
 955   GrowableArray<Node_Notes*>* node_note_array() const { return _node_note_array; }
 956   void set_node_note_array(GrowableArray<Node_Notes*>* arr) { _node_note_array = arr; }
 957   Node_Notes* default_node_notes() const        { return _default_node_notes; }
 958   void    set_default_node_notes(Node_Notes* n) { _default_node_notes = n; }
 959 
 960   Node_Notes*       node_notes_at(int idx) {
 961     return locate_node_notes(_node_note_array, idx, false);
 962   }
 963   inline bool   set_node_notes_at(int idx, Node_Notes* value);
 964 
 965   // Copy notes from source to dest, if they exist.
 966   // Overwrite dest only if source provides something.
 967   // Return true if information was moved.
 968   bool copy_node_notes_to(Node* dest, Node* source);
 969 
 970   // Workhorse function to sort out the blocked Node_Notes array:
 971   inline Node_Notes* locate_node_notes(GrowableArray<Node_Notes*>* arr,
 972                                        int idx, bool can_grow = false);
 973 
 974   void grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by);
 975 
 976   // Type management
 977   Arena*            type_arena()                { return _type_arena; }
 978   Dict*             type_dict()                 { return _type_dict; }
 979   size_t            type_last_size()            { return _type_last_size; }
 980   int               num_alias_types()           { return _num_alias_types; }
 981 
 982   void          init_type_arena()                       { _type_arena = &_Compile_types; }
 983   void          set_type_arena(Arena* a)                { _type_arena = a; }
 984   void          set_type_dict(Dict* d)                  { _type_dict = d; }
 985   void          set_type_last_size(size_t sz)           { _type_last_size = sz; }
 986 
 987   const TypeFunc* last_tf(ciMethod* m) {
 988     return (m == _last_tf_m) ? _last_tf : NULL;
 989   }
 990   void set_last_tf(ciMethod* m, const TypeFunc* tf) {
 991     assert(m != NULL || tf == NULL, "");
 992     _last_tf_m = m;
 993     _last_tf = tf;
 994   }
 995 
 996   AliasType*        alias_type(int                idx)  { assert(idx < num_alias_types(), "oob"); return _alias_types[idx]; }
 997   AliasType*        alias_type(const TypePtr* adr_type, ciField* field = NULL) { return find_alias_type(adr_type, false, field); }
 998   bool         have_alias_type(const TypePtr* adr_type);
 999   AliasType*        alias_type(ciField*         field);
1000 
1001   int               get_alias_index(const TypePtr* at)  { return alias_type(at)->index(); }
1002   const TypePtr*    get_adr_type(uint aidx)             { return alias_type(aidx)->adr_type(); }
1003   int               get_general_index(uint aidx)        { return alias_type(aidx)->general_index(); }
1004 
1005   // Building nodes
1006   void              rethrow_exceptions(JVMState* jvms);
1007   void              return_values(JVMState* jvms);
1008   JVMState*         build_start_state(StartNode* start, const TypeFunc* tf);
1009 
1010   // Decide how to build a call.
1011   // The profile factor is a discount to apply to this site's interp. profile.
1012   CallGenerator*    call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch,
1013                                    JVMState* jvms, bool allow_inline, float profile_factor, ciKlass* speculative_receiver_type = NULL,
1014                                    bool allow_intrinsics = true, bool delayed_forbidden = false);
1015   bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
1016     return should_delay_string_inlining(call_method, jvms) ||
1017            should_delay_boxing_inlining(call_method, jvms);
1018   }
1019   bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
1020   bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
1021 
1022   // Helper functions to identify inlining potential at call-site
1023   ciMethod* optimize_virtual_call(ciMethod* caller, int bci, ciInstanceKlass* klass,
1024                                   ciKlass* holder, ciMethod* callee,
1025                                   const TypeOopPtr* receiver_type, bool is_virtual,
1026                                   bool &call_does_dispatch, int &vtable_index,
1027                                   bool check_access = true);
1028   ciMethod* optimize_inlining(ciMethod* caller, int bci, ciInstanceKlass* klass,
1029                               ciMethod* callee, const TypeOopPtr* receiver_type,
1030                               bool check_access = true);
1031 
1032   // Report if there were too many traps at a current method and bci.
1033   // Report if a trap was recorded, and/or PerMethodTrapLimit was exceeded.
1034   // If there is no MDO at all, report no trap unless told to assume it.
1035   bool too_many_traps(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1036   // This version, unspecific to a particular bci, asks if
1037   // PerMethodTrapLimit was exceeded for all inlined methods seen so far.
1038   bool too_many_traps(Deoptimization::DeoptReason reason,
1039                       // Privately used parameter for logging:
1040                       ciMethodData* logmd = NULL);
1041   // Report if there were too many recompiles at a method and bci.
1042   bool too_many_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason);
1043   // Report if there were too many traps or recompiles at a method and bci.
1044   bool too_many_traps_or_recompiles(ciMethod* method, int bci, Deoptimization::DeoptReason reason) {
1045     return too_many_traps(method, bci, reason) ||
1046            too_many_recompiles(method, bci, reason);
1047   }
1048   // Return a bitset with the reasons where deoptimization is allowed,
1049   // i.e., where there were not too many uncommon traps.
1050   int _allowed_reasons;
1051   int      allowed_deopt_reasons() { return _allowed_reasons; }
1052   void set_allowed_deopt_reasons();
1053 
1054   // Parsing, optimization
1055   PhaseGVN*         initial_gvn()               { return _initial_gvn; }
1056   Unique_Node_List* for_igvn()                  { return _for_igvn; }
1057   inline void       record_for_igvn(Node* n);   // Body is after class Unique_Node_List.
1058   void          set_initial_gvn(PhaseGVN *gvn)           { _initial_gvn = gvn; }
1059   void          set_for_igvn(Unique_Node_List *for_igvn) { _for_igvn = for_igvn; }
1060 
1061   // Replace n by nn using initial_gvn, calling hash_delete and
1062   // record_for_igvn as needed.
1063   void gvn_replace_by(Node* n, Node* nn);
1064 
1065 
1066   void              identify_useful_nodes(Unique_Node_List &useful);
1067   void              update_dead_node_list(Unique_Node_List &useful);
1068   void              remove_useless_nodes (Unique_Node_List &useful);
1069 
1070   WarmCallInfo*     warm_calls() const          { return _warm_calls; }
1071   void          set_warm_calls(WarmCallInfo* l) { _warm_calls = l; }
1072   WarmCallInfo* pop_warm_call();
1073 
1074   // Record this CallGenerator for inlining at the end of parsing.
1075   void              add_late_inline(CallGenerator* cg)        {
1076     _late_inlines.insert_before(_late_inlines_pos, cg);
1077     _late_inlines_pos++;
1078   }
1079 
1080   void              prepend_late_inline(CallGenerator* cg)    {
1081     _late_inlines.insert_before(0, cg);
1082   }
1083 
1084   void              add_string_late_inline(CallGenerator* cg) {
1085     _string_late_inlines.push(cg);
1086   }
1087 
1088   void              add_boxing_late_inline(CallGenerator* cg) {
1089     _boxing_late_inlines.push(cg);
1090   }
1091 
1092   void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
1093 
1094   void process_print_inlining();
1095   void dump_print_inlining();
1096 
1097   bool over_inlining_cutoff() const {
1098     if (!inlining_incrementally()) {
1099       return unique() > (uint)NodeCountInliningCutoff;
1100     } else {
1101       // Give some room for incremental inlining algorithm to "breathe"
1102       // and avoid thrashing when live node count is close to the limit.
1103       // Keep in mind that live_nodes() isn't accurate during inlining until
1104       // dead node elimination step happens (see Compile::inline_incrementally).
1105       return live_nodes() > (uint)LiveNodeCountInliningCutoff * 11 / 10;
1106     }
1107   }
1108 
1109   void inc_number_of_mh_late_inlines() { _number_of_mh_late_inlines++; }
1110   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--; }
1111   bool has_mh_late_inlines() const     { return _number_of_mh_late_inlines > 0; }
1112 
1113   bool inline_incrementally_one();
1114   void inline_incrementally_cleanup(PhaseIterGVN& igvn);
1115   void inline_incrementally(PhaseIterGVN& igvn);
1116   void inline_string_calls(bool parse_time);
1117   void inline_boxing_calls(PhaseIterGVN& igvn);
1118   bool optimize_loops(PhaseIterGVN& igvn, LoopOptsMode mode);
1119   void remove_root_to_sfpts_edges(PhaseIterGVN& igvn);
1120 
1121   // Matching, CFG layout, allocation, code generation
1122   PhaseCFG*         cfg()                       { return _cfg; }
1123   bool              select_24_bit_instr() const { return _select_24_bit_instr; }
1124   bool              in_24_bit_fp_mode() const   { return _in_24_bit_fp_mode; }
1125   bool              has_java_calls() const      { return _java_calls > 0; }
1126   int               java_calls() const          { return _java_calls; }
1127   int               inner_loops() const         { return _inner_loops; }
1128   Matcher*          matcher()                   { return _matcher; }
1129   PhaseRegAlloc*    regalloc()                  { return _regalloc; }
1130   int               frame_slots() const         { return _frame_slots; }
1131   int               frame_size_in_words() const; // frame_slots in units of the polymorphic 'words'
1132   int               frame_size_in_bytes() const { return _frame_slots << LogBytesPerInt; }
1133   RegMask&          FIRST_STACK_mask()          { return _FIRST_STACK_mask; }
1134   Arena*            indexSet_arena()            { return _indexSet_arena; }
1135   void*             indexSet_free_block_list()  { return _indexSet_free_block_list; }
1136   uint              node_bundling_limit()       { return _node_bundling_limit; }
1137   Bundle*           node_bundling_base()        { return _node_bundling_base; }
1138   void          set_node_bundling_limit(uint n) { _node_bundling_limit = n; }
1139   void          set_node_bundling_base(Bundle* b) { _node_bundling_base = b; }
1140   bool          starts_bundle(const Node *n) const;
1141   bool          need_stack_bang(int frame_size_in_bytes) const;
1142   bool          need_register_stack_bang() const;
1143 
1144   void  update_interpreter_frame_size(int size) {
1145     if (_interpreter_frame_size < size) {
1146       _interpreter_frame_size = size;
1147     }
1148   }
1149   int           bang_size_in_bytes() const;
1150 
1151   void          set_matcher(Matcher* m)                 { _matcher = m; }
1152 //void          set_regalloc(PhaseRegAlloc* ra)           { _regalloc = ra; }
1153   void          set_indexSet_arena(Arena* a)            { _indexSet_arena = a; }
1154   void          set_indexSet_free_block_list(void* p)   { _indexSet_free_block_list = p; }
1155 
1156   // Remember if this compilation changes hardware mode to 24-bit precision
1157   void set_24_bit_selection_and_mode(bool selection, bool mode) {
1158     _select_24_bit_instr = selection;
1159     _in_24_bit_fp_mode   = mode;
1160   }
1161 
1162   void  set_java_calls(int z) { _java_calls  = z; }
1163   void set_inner_loops(int z) { _inner_loops = z; }
1164 
1165   // Instruction bits passed off to the VM
1166   int               code_size()                 { return _method_size; }
1167   CodeBuffer*       code_buffer()               { return &_code_buffer; }
1168   int               first_block_size()          { return _first_block_size; }
1169   void              set_frame_complete(int off) { if (!in_scratch_emit_size()) { _code_offsets.set_value(CodeOffsets::Frame_Complete, off); } }
1170   ExceptionHandlerTable*  handler_table()       { return &_handler_table; }
1171   ImplicitExceptionTable* inc_table()           { return &_inc_table; }
1172   OopMapSet*        oop_map_set()               { return _oop_map_set; }
1173   DebugInformationRecorder* debug_info()        { return env()->debug_info(); }
1174   Dependencies*     dependencies()              { return env()->dependencies(); }
1175   static int        CompiledZap_count()         { return _CompiledZap_count; }
1176   BufferBlob*       scratch_buffer_blob()       { return _scratch_buffer_blob; }
1177   void         init_scratch_buffer_blob(int const_size);
1178   void        clear_scratch_buffer_blob();
1179   void          set_scratch_buffer_blob(BufferBlob* b) { _scratch_buffer_blob = b; }
1180   relocInfo*        scratch_locs_memory()       { return _scratch_locs_memory; }
1181   void          set_scratch_locs_memory(relocInfo* b)  { _scratch_locs_memory = b; }
1182 
1183   // emit to scratch blob, report resulting size
1184   uint              scratch_emit_size(const Node* n);
1185   void       set_in_scratch_emit_size(bool x)   {        _in_scratch_emit_size = x; }
1186   bool           in_scratch_emit_size() const   { return _in_scratch_emit_size;     }
1187 
1188   enum ScratchBufferBlob {
1189 #if defined(PPC64)
1190     MAX_inst_size       = 2048,
1191 #else
1192     MAX_inst_size       = 1024,
1193 #endif
1194     MAX_locs_size       = 128, // number of relocInfo elements
1195     MAX_const_size      = 128,
1196     MAX_stubs_size      = 128
1197   };
1198 
1199   // Major entry point.  Given a Scope, compile the associated method.
1200   // For normal compilations, entry_bci is InvocationEntryBci.  For on stack
1201   // replacement, entry_bci indicates the bytecode for which to compile a
1202   // continuation.
1203   Compile(ciEnv* ci_env, C2Compiler* compiler, ciMethod* target,
1204           int entry_bci, bool subsume_loads, bool do_escape_analysis,
1205           bool eliminate_boxing, DirectiveSet* directive);
1206 
1207   // Second major entry point.  From the TypeFunc signature, generate code
1208   // to pass arguments from the Java calling convention to the C calling
1209   // convention.
1210   Compile(ciEnv* ci_env, const TypeFunc *(*gen)(),
1211           address stub_function, const char *stub_name,
1212           int is_fancy_jump, bool pass_tls,
1213           bool save_arg_registers, bool return_pc, DirectiveSet* directive);
1214 
1215   // From the TypeFunc signature, generate code to pass arguments
1216   // from Compiled calling convention to Interpreter's calling convention
1217   void Generate_Compiled_To_Interpreter_Graph(const TypeFunc *tf, address interpreter_entry);
1218 
1219   // From the TypeFunc signature, generate code to pass arguments
1220   // from Interpreter's calling convention to Compiler's calling convention
1221   void Generate_Interpreter_To_Compiled_Graph(const TypeFunc *tf);
1222 
1223   // Are we compiling a method?
1224   bool has_method() { return method() != NULL; }
1225 
1226   // Maybe print some information about this compile.
1227   void print_compile_messages();
1228 
1229   // Final graph reshaping, a post-pass after the regular optimizer is done.
1230   bool final_graph_reshaping();
1231 
1232   // returns true if adr is completely contained in the given alias category
1233   bool must_alias(const TypePtr* adr, int alias_idx);
1234 
1235   // returns true if adr overlaps with the given alias category
1236   bool can_alias(const TypePtr* adr, int alias_idx);
1237 
1238   // Driver for converting compiler's IR into machine code bits
1239   void Output();
1240 
1241   // Accessors for node bundling info.
1242   Bundle* node_bundling(const Node *n);
1243   bool valid_bundle_info(const Node *n);
1244 
1245   // Schedule and Bundle the instructions
1246   void ScheduleAndBundle();
1247 
1248   // Build OopMaps for each GC point
1249   void BuildOopMaps();
1250 
1251   // Append debug info for the node "local" at safepoint node "sfpt" to the
1252   // "array",   May also consult and add to "objs", which describes the
1253   // scalar-replaced objects.
1254   void FillLocArray( int idx, MachSafePointNode* sfpt,
1255                      Node *local, GrowableArray<ScopeValue*> *array,
1256                      GrowableArray<ScopeValue*> *objs );
1257 
1258   // If "objs" contains an ObjectValue whose id is "id", returns it, else NULL.
1259   static ObjectValue* sv_for_node_id(GrowableArray<ScopeValue*> *objs, int id);
1260   // Requres that "objs" does not contains an ObjectValue whose id matches
1261   // that of "sv.  Appends "sv".
1262   static void set_sv_for_object_node(GrowableArray<ScopeValue*> *objs,
1263                                      ObjectValue* sv );
1264 
1265   // Process an OopMap Element while emitting nodes
1266   void Process_OopMap_Node(MachNode *mach, int code_offset);
1267 
1268   // Initialize code buffer
1269   CodeBuffer* init_buffer(uint* blk_starts);
1270 
1271   // Write out basic block data to code buffer
1272   void fill_buffer(CodeBuffer* cb, uint* blk_starts);
1273 
1274   // Determine which variable sized branches can be shortened
1275   void shorten_branches(uint* blk_starts, int& code_size, int& reloc_size, int& stub_size);
1276 
1277   // Compute the size of first NumberOfLoopInstrToAlign instructions
1278   // at the head of a loop.
1279   void compute_loop_first_inst_sizes();
1280 
1281   // Compute the information for the exception tables
1282   void FillExceptionTables(uint cnt, uint *call_returns, uint *inct_starts, Label *blk_labels);
1283 
1284   // Stack slots that may be unused by the calling convention but must
1285   // otherwise be preserved.  On Intel this includes the return address.
1286   // On PowerPC it includes the 4 words holding the old TOC & LR glue.
1287   uint in_preserve_stack_slots();
1288 
1289   // "Top of Stack" slots that may be unused by the calling convention but must
1290   // otherwise be preserved.
1291   // On Intel these are not necessary and the value can be zero.
1292   // On Sparc this describes the words reserved for storing a register window
1293   // when an interrupt occurs.
1294   static uint out_preserve_stack_slots();
1295 
1296   // Number of outgoing stack slots killed above the out_preserve_stack_slots
1297   // for calls to C.  Supports the var-args backing area for register parms.
1298   uint varargs_C_out_slots_killed() const;
1299 
1300   // Number of Stack Slots consumed by a synchronization entry
1301   int sync_stack_slots() const;
1302 
1303   // Compute the name of old_SP.  See <arch>.ad for frame layout.
1304   OptoReg::Name compute_old_SP();
1305 
1306  private:
1307   // Phase control:
1308   void Init(int aliaslevel);                     // Prepare for a single compilation
1309   int  Inline_Warm();                            // Find more inlining work.
1310   void Finish_Warm();                            // Give up on further inlines.
1311   void Optimize();                               // Given a graph, optimize it
1312   void Code_Gen();                               // Generate code from a graph
1313 
1314   // Management of the AliasType table.
1315   void grow_alias_types();
1316   AliasCacheEntry* probe_alias_cache(const TypePtr* adr_type);
1317   const TypePtr *flatten_alias_type(const TypePtr* adr_type) const;
1318   AliasType* find_alias_type(const TypePtr* adr_type, bool no_create, ciField* field);
1319 
1320   void verify_top(Node*) const PRODUCT_RETURN;
1321 
1322   // Intrinsic setup.
1323   void           register_library_intrinsics();                            // initializer
1324   CallGenerator* make_vm_intrinsic(ciMethod* m, bool is_virtual);          // constructor
1325   int            intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found);  // helper
1326   CallGenerator* find_intrinsic(ciMethod* m, bool is_virtual);             // query fn
1327   void           register_intrinsic(CallGenerator* cg);                    // update fn
1328 
1329 #ifndef PRODUCT
1330   static juint  _intrinsic_hist_count[vmIntrinsics::ID_LIMIT];
1331   static jubyte _intrinsic_hist_flags[vmIntrinsics::ID_LIMIT];
1332 #endif
1333   // Function calls made by the public function final_graph_reshaping.
1334   // No need to be made public as they are not called elsewhere.
1335   void final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &frc);
1336   void final_graph_reshaping_main_switch(Node* n, Final_Reshape_Counts& frc, uint nop);
1337   void final_graph_reshaping_walk( Node_Stack &nstack, Node *root, Final_Reshape_Counts &frc );
1338   void eliminate_redundant_card_marks(Node* n);
1339 
1340  public:
1341 
1342   // Note:  Histogram array size is about 1 Kb.
1343   enum {                        // flag bits:
1344     _intrinsic_worked = 1,      // succeeded at least once
1345     _intrinsic_failed = 2,      // tried it but it failed
1346     _intrinsic_disabled = 4,    // was requested but disabled (e.g., -XX:-InlineUnsafeOps)
1347     _intrinsic_virtual = 8,     // was seen in the virtual form (rare)
1348     _intrinsic_both = 16        // was seen in the non-virtual form (usual)
1349   };
1350   // Update histogram.  Return boolean if this is a first-time occurrence.
1351   static bool gather_intrinsic_statistics(vmIntrinsics::ID id,
1352                                           bool is_virtual, int flags) PRODUCT_RETURN0;
1353   static void print_intrinsic_statistics() PRODUCT_RETURN;
1354 
1355   // Graph verification code
1356   // Walk the node list, verifying that there is a one-to-one
1357   // correspondence between Use-Def edges and Def-Use edges
1358   // The option no_dead_code enables stronger checks that the
1359   // graph is strongly connected from root in both directions.
1360   void verify_graph_edges(bool no_dead_code = false) PRODUCT_RETURN;
1361 
1362   // End-of-run dumps.
1363   static void print_statistics() PRODUCT_RETURN;
1364 
1365   // Dump formatted assembly
1366 #if defined(SUPPORT_OPTO_ASSEMBLY)
1367   void dump_asm_on(outputStream* ost, int* pcs, uint pc_limit);
1368   void dump_asm(int* pcs = NULL, uint pc_limit = 0) { dump_asm_on(tty, pcs, pc_limit); }
1369 #else
1370   void dump_asm_on(outputStream* ost, int* pcs, uint pc_limit) { return; }
1371   void dump_asm(int* pcs = NULL, uint pc_limit = 0) { return; }
1372 #endif
1373   void dump_pc(int *pcs, int pc_limit, Node *n);
1374 
1375   // Verify ADLC assumptions during startup
1376   static void adlc_verification() PRODUCT_RETURN;
1377 
1378   // Definitions of pd methods
1379   static void pd_compiler2_init();
1380 
1381   // Static parse-time type checking logic for gen_subtype_check:
1382   enum { SSC_always_false, SSC_always_true, SSC_easy_test, SSC_full_test };
1383   int static_subtype_check(ciKlass* superk, ciKlass* subk);
1384 
1385   static Node* conv_I2X_index(PhaseGVN* phase, Node* offset, const TypeInt* sizetype,
1386                               // Optional control dependency (for example, on range check)
1387                               Node* ctrl = NULL);
1388 
1389   // Convert integer value to a narrowed long type dependent on ctrl (for example, a range check)
1390   static Node* constrained_convI2L(PhaseGVN* phase, Node* value, const TypeInt* itype, Node* ctrl);
1391 
1392   // Auxiliary method for randomized fuzzing/stressing
1393   static bool randomized_select(int count);
1394 
1395   // supporting clone_map
1396   CloneMap&     clone_map();
1397   void          set_clone_map(Dict* d);
1398 
1399   bool needs_clinit_barrier(ciField* ik,         ciMethod* accessing_method);
1400   bool needs_clinit_barrier(ciMethod* ik,        ciMethod* accessing_method);
1401   bool needs_clinit_barrier(ciInstanceKlass* ik, ciMethod* accessing_method);
1402 };
1403 
1404 #endif // SHARE_OPTO_COMPILE_HPP