1 /* 2 * Copyright (c) 2000, 2011, 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_virtual_call(ciMethod* m, int vtable_index); // virtual, interface 112 static CallGenerator* for_dynamic_call(ciMethod* m); // invokedynamic 113 114 static CallGenerator* for_method_handle_call(Node* method_handle, JVMState* jvms, ciMethod* caller, ciMethod* callee, ciCallProfile profile); 115 static CallGenerator* for_invokedynamic_call( JVMState* jvms, ciMethod* caller, ciMethod* callee, ciCallProfile profile); 116 117 static CallGenerator* for_method_handle_inline(Node* method_handle, JVMState* jvms, ciMethod* caller, ciMethod* callee, ciCallProfile profile); 118 static CallGenerator* for_invokedynamic_inline(ciCallSite* call_site, JVMState* jvms, ciMethod* caller, ciMethod* callee, ciCallProfile profile); 119 120 // How to generate a replace a direct call with an inline version 121 static CallGenerator* for_late_inline(ciMethod* m, CallGenerator* inline_cg); 122 123 // How to make a call but defer the decision whether to inline or not. 124 static CallGenerator* for_warm_call(WarmCallInfo* ci, 125 CallGenerator* if_cold, 126 CallGenerator* if_hot); 127 128 // How to make a call that optimistically assumes a receiver type: 129 static CallGenerator* for_predicted_call(ciKlass* predicted_receiver, 130 CallGenerator* if_missed, 131 CallGenerator* if_hit, 132 float hit_prob); 133 134 // How to make a call that optimistically assumes a MethodHandle target: 135 static CallGenerator* for_predicted_dynamic_call(ciMethodHandle* predicted_method_handle, 136 CallGenerator* if_missed, 137 CallGenerator* if_hit, 138 float hit_prob); 139 140 // How to make a call that gives up and goes back to the interpreter: 141 static CallGenerator* for_uncommon_trap(ciMethod* m, 142 Deoptimization::DeoptReason reason, 143 Deoptimization::DeoptAction action); 144 145 // Registry for intrinsics: 146 static CallGenerator* for_intrinsic(ciMethod* m); 147 static void register_intrinsic(ciMethod* m, CallGenerator* cg); 148 }; 149 150 class InlineCallGenerator : public CallGenerator { 151 virtual bool is_inline() const { return true; } 152 153 protected: 154 InlineCallGenerator(ciMethod* method) : CallGenerator(method) { } 155 }; 156 157 158 //---------------------------WarmCallInfo-------------------------------------- 159 // A struct to collect information about a given call site. 160 // Helps sort call sites into "hot", "medium", and "cold". 161 // Participates in the queueing of "medium" call sites for possible inlining. 162 class WarmCallInfo : public ResourceObj { 163 private: 164 165 CallNode* _call; // The CallNode which may be inlined. 166 CallGenerator* _hot_cg;// CG for expanding the call node 167 168 // These are the metrics we use to evaluate call sites: 169 170 float _count; // How often do we expect to reach this site? 171 float _profit; // How much time do we expect to save by inlining? 172 float _work; // How long do we expect the average call to take? 173 float _size; // How big do we expect the inlined code to be? 174 175 float _heat; // Combined score inducing total order on call sites. 176 WarmCallInfo* _next; // Next cooler call info in pending queue. 177 178 // Count is the number of times this call site is expected to be executed. 179 // Large count is favorable for inlining, because the extra compilation 180 // work will be amortized more completely. 181 182 // Profit is a rough measure of the amount of time we expect to save 183 // per execution of this site if we inline it. (1.0 == call overhead) 184 // Large profit favors inlining. Negative profit disables inlining. 185 186 // Work is a rough measure of the amount of time a typical out-of-line 187 // call from this site is expected to take. (1.0 == call, no-op, return) 188 // Small work is somewhat favorable for inlining, since methods with 189 // short "hot" traces are more likely to inline smoothly. 190 191 // Size is the number of graph nodes we expect this method to produce, 192 // not counting the inlining of any further warm calls it may include. 193 // Small size favors inlining, since small methods are more likely to 194 // inline smoothly. The size is estimated by examining the native code 195 // if available. The method bytecodes are also examined, assuming 196 // empirically observed node counts for each kind of bytecode. 197 198 // Heat is the combined "goodness" of a site's inlining. If we were 199 // omniscient, it would be the difference of two sums of future execution 200 // times of code emitted for this site (amortized across multiple sites if 201 // sharing applies). The two sums are for versions of this call site with 202 // and without inlining. 203 204 // We approximate this mythical quantity by playing with averages, 205 // rough estimates, and assumptions that history repeats itself. 206 // The basic formula count * profit is heuristically adjusted 207 // by looking at the expected compilation and execution times of 208 // of the inlined call. 209 210 // Note: Some of these metrics may not be present in the final product, 211 // but exist in development builds to experiment with inline policy tuning. 212 213 // This heuristic framework does not model well the very significant 214 // effects of multiple-level inlining. It is possible to see no immediate 215 // profit from inlining X->Y, but to get great profit from a subsequent 216 // inlining X->Y->Z. 217 218 // This framework does not take well into account the problem of N**2 code 219 // size in a clique of mutually inlinable methods. 220 221 WarmCallInfo* next() const { return _next; } 222 void set_next(WarmCallInfo* n) { _next = n; } 223 224 static WarmCallInfo _always_hot; 225 static WarmCallInfo _always_cold; 226 227 // Constructor intitialization of always_hot and always_cold 228 WarmCallInfo(float c, float p, float w, float s) { 229 _call = NULL; 230 _hot_cg = NULL; 231 _next = NULL; 232 _count = c; 233 _profit = p; 234 _work = w; 235 _size = s; 236 _heat = 0; 237 } 238 239 public: 240 // Because WarmInfo objects live over the entire lifetime of the 241 // Compile object, they are allocated into the comp_arena, which 242 // does not get resource marked or reset during the compile process 243 void *operator new( size_t x, Compile* C ) { return C->comp_arena()->Amalloc(x); } 244 void operator delete( void * ) { } // fast deallocation 245 246 static WarmCallInfo* always_hot(); 247 static WarmCallInfo* always_cold(); 248 249 WarmCallInfo() { 250 _call = NULL; 251 _hot_cg = NULL; 252 _next = NULL; 253 _count = _profit = _work = _size = _heat = 0; 254 } 255 256 CallNode* call() const { return _call; } 257 float count() const { return _count; } 258 float size() const { return _size; } 259 float work() const { return _work; } 260 float profit() const { return _profit; } 261 float heat() const { return _heat; } 262 263 void set_count(float x) { _count = x; } 264 void set_size(float x) { _size = x; } 265 void set_work(float x) { _work = x; } 266 void set_profit(float x) { _profit = x; } 267 void set_heat(float x) { _heat = x; } 268 269 // Load initial heuristics from profiles, etc. 270 // The heuristics can be tweaked further by the caller. 271 void init(JVMState* call_site, ciMethod* call_method, ciCallProfile& profile, float prof_factor); 272 273 static float MAX_VALUE() { return +1.0e10; } 274 static float MIN_VALUE() { return -1.0e10; } 275 276 float compute_heat() const; 277 278 void set_call(CallNode* call) { _call = call; } 279 void set_hot_cg(CallGenerator* cg) { _hot_cg = cg; } 280 281 // Do not queue very hot or very cold calls. 282 // Make very cold ones out of line immediately. 283 // Inline very hot ones immediately. 284 // These queries apply various tunable limits 285 // to the above metrics in a systematic way. 286 // Test for coldness before testing for hotness. 287 bool is_cold() const; 288 bool is_hot() const; 289 290 // Force a warm call to be hot. This worklists the call node for inlining. 291 void make_hot(); 292 293 // Force a warm call to be cold. This worklists the call node for out-of-lining. 294 void make_cold(); 295 296 // A reproducible total ordering, in which heat is the major key. 297 bool warmer_than(WarmCallInfo* that); 298 299 // List management. These methods are called with the list head, 300 // and return the new list head, inserting or removing the receiver. 301 WarmCallInfo* insert_into(WarmCallInfo* head); 302 WarmCallInfo* remove_from(WarmCallInfo* head); 303 304 #ifndef PRODUCT 305 void print() const; 306 void print_all() const; 307 int count_all() const; 308 #endif 309 }; 310 311 #endif // SHARE_VM_OPTO_CALLGENERATOR_HPP