1 /*
   2  * Copyright (c) 2000, 2010, 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_CALLGENERATOR_HPP
  26 #define SHARE_VM_OPTO_CALLGENERATOR_HPP
  27 
  28 #include "opto/callnode.hpp"
  29 #include "opto/compile.hpp"
  30 #include "opto/type.hpp"
  31 #include "runtime/deoptimization.hpp"
  32 
  33 //---------------------------CallGenerator-------------------------------------
  34 // The subclasses of this class handle generation of ideal nodes for
  35 // call sites and method entry points.
  36 
  37 class CallGenerator : public ResourceObj {
  38  public:
  39   enum {
  40     xxxunusedxxx
  41   };
  42 
  43  private:
  44   ciMethod*             _method;                // The method being called.
  45 
  46  protected:
  47   CallGenerator(ciMethod* method);
  48 
  49  public:
  50   // Accessors
  51   ciMethod*         method() const              { return _method; }
  52 
  53   // is_inline: At least some code implementing the method is copied here.
  54   virtual bool      is_inline() const           { return false; }
  55   // is_intrinsic: There's a method-specific way of generating the inline code.
  56   virtual bool      is_intrinsic() const        { return false; }
  57   // is_parse: Bytecodes implementing the specific method are copied here.
  58   virtual bool      is_parse() const            { return false; }
  59   // is_virtual: The call uses the receiver type to select or check the method.
  60   virtual bool      is_virtual() const          { return false; }
  61   // is_deferred: The decision whether to inline or not is deferred.
  62   virtual bool      is_deferred() const         { return false; }
  63   // is_predicted: Uses an explicit check against a predicted type.
  64   virtual bool      is_predicted() const        { return false; }
  65   // is_trap: Does not return to the caller.  (E.g., uncommon trap.)
  66   virtual bool      is_trap() const             { return false; }
  67 
  68   // is_late_inline: supports conversion of call into an inline
  69   virtual bool      is_late_inline() const      { return false; }
  70   // Replace the call with an inline version of the code
  71   virtual void do_late_inline() { ShouldNotReachHere(); }
  72 
  73   virtual CallStaticJavaNode* call_node() const { ShouldNotReachHere(); return NULL; }
  74 
  75   // Note:  It is possible for a CG to be both inline and virtual.
  76   // (The hashCode intrinsic does a vtable check and an inlined fast path.)
  77 
  78   // Utilities:
  79   const TypeFunc*   tf() const;
  80 
  81   // The given jvms has state and arguments for a call to my method.
  82   // Edges after jvms->argoff() carry all (pre-popped) argument values.
  83   //
  84   // Update the map with state and return values (if any) and return it.
  85   // The return values (0, 1, or 2) must be pushed on the map's stack,
  86   // and the sp of the jvms incremented accordingly.
  87   //
  88   // The jvms is returned on success.  Alternatively, a copy of the
  89   // given jvms, suitably updated, may be returned, in which case the
  90   // caller should discard the original jvms.
  91   //
  92   // The non-Parm edges of the returned map will contain updated global state,
  93   // and one or two edges before jvms->sp() will carry any return values.
  94   // Other map edges may contain locals or monitors, and should not
  95   // be changed in meaning.
  96   //
  97   // If the call traps, the returned map must have a control edge of top.
  98   // If the call can throw, the returned map must report has_exceptions().
  99   //
 100   // If the result is NULL, it means that this CallGenerator was unable
 101   // to handle the given call, and another CallGenerator should be consulted.
 102   virtual JVMState* generate(JVMState* jvms) = 0;
 103 
 104   // How to generate a call site that is inlined:
 105   static CallGenerator* for_inline(ciMethod* m, float expected_uses = -1);
 106   // How to generate code for an on-stack replacement handler.
 107   static CallGenerator* for_osr(ciMethod* m, int osr_bci);
 108 
 109   // How to generate vanilla out-of-line call sites:
 110   static CallGenerator* for_direct_call(ciMethod* m, bool separate_io_projs = false);   // static, special
 111   static CallGenerator* for_dynamic_call(ciMethod* m);   // invokedynamic
 112   static CallGenerator* for_virtual_call(ciMethod* m, int vtable_index);  // virtual, interface
 113 
 114   // How to generate a replace a direct call with an inline version
 115   static CallGenerator* for_late_inline(ciMethod* m, CallGenerator* inline_cg);
 116 
 117   // How to make a call but defer the decision whether to inline or not.
 118   static CallGenerator* for_warm_call(WarmCallInfo* ci,
 119                                       CallGenerator* if_cold,
 120                                       CallGenerator* if_hot);
 121 
 122   // How to make a call that optimistically assumes a receiver type:
 123   static CallGenerator* for_predicted_call(ciKlass* predicted_receiver,
 124                                            CallGenerator* if_missed,
 125                                            CallGenerator* if_hit,
 126                                            float hit_prob);
 127 
 128   // How to make a call that optimistically assumes a MethodHandle target:
 129   static CallGenerator* for_predicted_dynamic_call(ciMethodHandle* predicted_method_handle,
 130                                                    CallGenerator* if_missed,
 131                                                    CallGenerator* if_hit,
 132                                                    float hit_prob);
 133 
 134   // How to make a call that gives up and goes back to the interpreter:
 135   static CallGenerator* for_uncommon_trap(ciMethod* m,
 136                                           Deoptimization::DeoptReason reason,
 137                                           Deoptimization::DeoptAction action);
 138 
 139   // Registry for intrinsics:
 140   static CallGenerator* for_intrinsic(ciMethod* m);
 141   static void register_intrinsic(ciMethod* m, CallGenerator* cg);
 142 };
 143 
 144 class InlineCallGenerator : public CallGenerator {
 145   virtual bool      is_inline() const           { return true; }
 146 
 147  protected:
 148   InlineCallGenerator(ciMethod* method) : CallGenerator(method) { }
 149 };
 150 
 151 
 152 //---------------------------WarmCallInfo--------------------------------------
 153 // A struct to collect information about a given call site.
 154 // Helps sort call sites into "hot", "medium", and "cold".
 155 // Participates in the queueing of "medium" call sites for possible inlining.
 156 class WarmCallInfo : public ResourceObj {
 157  private:
 158 
 159   CallNode*     _call;   // The CallNode which may be inlined.
 160   CallGenerator* _hot_cg;// CG for expanding the call node
 161 
 162   // These are the metrics we use to evaluate call sites:
 163 
 164   float         _count;  // How often do we expect to reach this site?
 165   float         _profit; // How much time do we expect to save by inlining?
 166   float         _work;   // How long do we expect the average call to take?
 167   float         _size;   // How big do we expect the inlined code to be?
 168 
 169   float         _heat;   // Combined score inducing total order on call sites.
 170   WarmCallInfo* _next;   // Next cooler call info in pending queue.
 171 
 172   // Count is the number of times this call site is expected to be executed.
 173   // Large count is favorable for inlining, because the extra compilation
 174   // work will be amortized more completely.
 175 
 176   // Profit is a rough measure of the amount of time we expect to save
 177   // per execution of this site if we inline it.  (1.0 == call overhead)
 178   // Large profit favors inlining.  Negative profit disables inlining.
 179 
 180   // Work is a rough measure of the amount of time a typical out-of-line
 181   // call from this site is expected to take.  (1.0 == call, no-op, return)
 182   // Small work is somewhat favorable for inlining, since methods with
 183   // short "hot" traces are more likely to inline smoothly.
 184 
 185   // Size is the number of graph nodes we expect this method to produce,
 186   // not counting the inlining of any further warm calls it may include.
 187   // Small size favors inlining, since small methods are more likely to
 188   // inline smoothly.  The size is estimated by examining the native code
 189   // if available.  The method bytecodes are also examined, assuming
 190   // empirically observed node counts for each kind of bytecode.
 191 
 192   // Heat is the combined "goodness" of a site's inlining.  If we were
 193   // omniscient, it would be the difference of two sums of future execution
 194   // times of code emitted for this site (amortized across multiple sites if
 195   // sharing applies).  The two sums are for versions of this call site with
 196   // and without inlining.
 197 
 198   // We approximate this mythical quantity by playing with averages,
 199   // rough estimates, and assumptions that history repeats itself.
 200   // The basic formula count * profit is heuristically adjusted
 201   // by looking at the expected compilation and execution times of
 202   // of the inlined call.
 203 
 204   // Note:  Some of these metrics may not be present in the final product,
 205   // but exist in development builds to experiment with inline policy tuning.
 206 
 207   // This heuristic framework does not model well the very significant
 208   // effects of multiple-level inlining.  It is possible to see no immediate
 209   // profit from inlining X->Y, but to get great profit from a subsequent
 210   // inlining X->Y->Z.
 211 
 212   // This framework does not take well into account the problem of N**2 code
 213   // size in a clique of mutually inlinable methods.
 214 
 215   WarmCallInfo*  next() const          { return _next; }
 216   void       set_next(WarmCallInfo* n) { _next = n; }
 217 
 218   static WarmCallInfo* _always_hot;
 219   static WarmCallInfo* _always_cold;
 220 
 221  public:
 222   // Because WarmInfo objects live over the entire lifetime of the
 223   // Compile object, they are allocated into the comp_arena, which
 224   // does not get resource marked or reset during the compile process
 225   void *operator new( size_t x, Compile* C ) { return C->comp_arena()->Amalloc(x); }
 226   void operator delete( void * ) { } // fast deallocation
 227 
 228   static WarmCallInfo* always_hot();
 229   static WarmCallInfo* always_cold();
 230 
 231   WarmCallInfo() {
 232     _call = NULL;
 233     _hot_cg = NULL;
 234     _next = NULL;
 235     _count = _profit = _work = _size = _heat = 0;
 236   }
 237 
 238   CallNode* call() const { return _call; }
 239   float count()    const { return _count; }
 240   float size()     const { return _size; }
 241   float work()     const { return _work; }
 242   float profit()   const { return _profit; }
 243   float heat()     const { return _heat; }
 244 
 245   void set_count(float x)     { _count = x; }
 246   void set_size(float x)      { _size = x; }
 247   void set_work(float x)      { _work = x; }
 248   void set_profit(float x)    { _profit = x; }
 249   void set_heat(float x)      { _heat = x; }
 250 
 251   // Load initial heuristics from profiles, etc.
 252   // The heuristics can be tweaked further by the caller.
 253   void init(JVMState* call_site, ciMethod* call_method, ciCallProfile& profile, float prof_factor);
 254 
 255   static float MAX_VALUE() { return +1.0e10; }
 256   static float MIN_VALUE() { return -1.0e10; }
 257 
 258   float compute_heat() const;
 259 
 260   void set_call(CallNode* call)      { _call = call; }
 261   void set_hot_cg(CallGenerator* cg) { _hot_cg = cg; }
 262 
 263   // Do not queue very hot or very cold calls.
 264   // Make very cold ones out of line immediately.
 265   // Inline very hot ones immediately.
 266   // These queries apply various tunable limits
 267   // to the above metrics in a systematic way.
 268   // Test for coldness before testing for hotness.
 269   bool is_cold() const;
 270   bool is_hot() const;
 271 
 272   // Force a warm call to be hot.  This worklists the call node for inlining.
 273   void make_hot();
 274 
 275   // Force a warm call to be cold.  This worklists the call node for out-of-lining.
 276   void make_cold();
 277 
 278   // A reproducible total ordering, in which heat is the major key.
 279   bool warmer_than(WarmCallInfo* that);
 280 
 281   // List management.  These methods are called with the list head,
 282   // and return the new list head, inserting or removing the receiver.
 283   WarmCallInfo* insert_into(WarmCallInfo* head);
 284   WarmCallInfo* remove_from(WarmCallInfo* head);
 285 
 286 #ifndef PRODUCT
 287   void print() const;
 288   void print_all() const;
 289   int count_all() const;
 290 #endif
 291 };
 292 
 293 #endif // SHARE_VM_OPTO_CALLGENERATOR_HPP