1 /* 2 * Copyright (c) 2000, 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_CALLGENERATOR_HPP 26 #define SHARE_OPTO_CALLGENERATOR_HPP 27 28 #include "compiler/compileBroker.hpp" 29 #include "opto/callnode.hpp" 30 #include "opto/compile.hpp" 31 #include "opto/type.hpp" 32 #include "runtime/deoptimization.hpp" 33 34 //---------------------------CallGenerator------------------------------------- 35 // The subclasses of this class handle generation of ideal nodes for 36 // call sites and method entry points. 37 38 class CallGenerator : public ResourceObj { 39 private: 40 ciMethod* _method; // The method being called. 41 42 protected: CallGenerator(ciMethod * method)43 CallGenerator(ciMethod* method) : _method(method) {} 44 45 void do_late_inline_helper(); 46 do_late_inline_check(Compile * C,JVMState * jvms)47 virtual bool do_late_inline_check(Compile* C, JVMState* jvms) { ShouldNotReachHere(); return false; } inline_cg() const48 virtual CallGenerator* inline_cg() const { ShouldNotReachHere(); return NULL; } is_pure_call() const49 virtual bool is_pure_call() const { ShouldNotReachHere(); return false; } 50 51 public: 52 // Accessors method() const53 ciMethod* method() const { return _method; } 54 55 // is_inline: At least some code implementing the method is copied here. is_inline() const56 virtual bool is_inline() const { return false; } 57 // is_intrinsic: There's a method-specific way of generating the inline code. is_intrinsic() const58 virtual bool is_intrinsic() const { return false; } 59 // is_parse: Bytecodes implementing the specific method are copied here. is_parse() const60 virtual bool is_parse() const { return false; } 61 // is_virtual: The call uses the receiver type to select or check the method. is_virtual() const62 virtual bool is_virtual() const { return false; } 63 // is_deferred: The decision whether to inline or not is deferred. is_deferred() const64 virtual bool is_deferred() const { return false; } 65 // is_predicated: Uses an explicit check (predicate). is_predicated() const66 virtual bool is_predicated() const { return false; } predicates_count() const67 virtual int predicates_count() const { return 0; } 68 // is_trap: Does not return to the caller. (E.g., uncommon trap.) is_trap() const69 virtual bool is_trap() const { return false; } 70 // does_virtual_dispatch: Should try inlining as normal method first. does_virtual_dispatch() const71 virtual bool does_virtual_dispatch() const { return false; } 72 73 // is_late_inline: supports conversion of call into an inline is_late_inline() const74 virtual bool is_late_inline() const { return false; } 75 // same but for method handle calls is_mh_late_inline() const76 virtual bool is_mh_late_inline() const { return false; } is_string_late_inline() const77 virtual bool is_string_late_inline() const { return false; } is_virtual_late_inline() const78 virtual bool is_virtual_late_inline() const { return false; } 79 80 // Replace the call with an inline version of the code do_late_inline()81 virtual void do_late_inline() { ShouldNotReachHere(); } 82 call_node() const83 virtual CallNode* call_node() const { return NULL; } with_call_node(CallNode * call)84 virtual CallGenerator* with_call_node(CallNode* call) { return this; } 85 set_unique_id(jlong id)86 virtual void set_unique_id(jlong id) { fatal("unique id only for late inlines"); }; unique_id() const87 virtual jlong unique_id() const { fatal("unique id only for late inlines"); return 0; }; 88 set_callee_method(ciMethod * callee)89 virtual void set_callee_method(ciMethod* callee) { ShouldNotReachHere(); } 90 91 // Note: It is possible for a CG to be both inline and virtual. 92 // (The hashCode intrinsic does a vtable check and an inlined fast path.) 93 94 // Allocate CallGenerators only in Compile arena since some of them are referenced from CallNodes. operator new(size_t size)95 void* operator new(size_t size) throw() { 96 Compile* C = Compile::current(); 97 return ResourceObj::operator new(size, C->comp_arena()); 98 } 99 100 // Utilities: 101 const TypeFunc* tf() const; 102 103 // The given jvms has state and arguments for a call to my method. 104 // Edges after jvms->argoff() carry all (pre-popped) argument values. 105 // 106 // Update the map with state and return values (if any) and return it. 107 // The return values (0, 1, or 2) must be pushed on the map's stack, 108 // and the sp of the jvms incremented accordingly. 109 // 110 // The jvms is returned on success. Alternatively, a copy of the 111 // given jvms, suitably updated, may be returned, in which case the 112 // caller should discard the original jvms. 113 // 114 // The non-Parm edges of the returned map will contain updated global state, 115 // and one or two edges before jvms->sp() will carry any return values. 116 // Other map edges may contain locals or monitors, and should not 117 // be changed in meaning. 118 // 119 // If the call traps, the returned map must have a control edge of top. 120 // If the call can throw, the returned map must report has_exceptions(). 121 // 122 // If the result is NULL, it means that this CallGenerator was unable 123 // to handle the given call, and another CallGenerator should be consulted. 124 virtual JVMState* generate(JVMState* jvms) = 0; 125 126 // How to generate a call site that is inlined: 127 static CallGenerator* for_inline(ciMethod* m, float expected_uses = -1); 128 // How to generate code for an on-stack replacement handler. 129 static CallGenerator* for_osr(ciMethod* m, int osr_bci); 130 131 // How to generate vanilla out-of-line call sites: 132 static CallGenerator* for_direct_call(ciMethod* m, bool separate_io_projs = false); // static, special 133 static CallGenerator* for_virtual_call(ciMethod* m, int vtable_index); // virtual, interface 134 135 static CallGenerator* for_method_handle_call( JVMState* jvms, ciMethod* caller, ciMethod* callee, bool allow_inline); 136 static CallGenerator* for_method_handle_inline(JVMState* jvms, ciMethod* caller, ciMethod* callee, bool allow_inline, bool& input_not_const); 137 138 // How to generate a replace a direct call with an inline version 139 static CallGenerator* for_late_inline(ciMethod* m, CallGenerator* inline_cg); 140 static CallGenerator* for_mh_late_inline(ciMethod* caller, ciMethod* callee, bool input_not_const); 141 static CallGenerator* for_string_late_inline(ciMethod* m, CallGenerator* inline_cg); 142 static CallGenerator* for_boxing_late_inline(ciMethod* m, CallGenerator* inline_cg); 143 static CallGenerator* for_vector_reboxing_late_inline(ciMethod* m, CallGenerator* inline_cg); 144 145 // How to make a call but defer the decision whether to inline or not. 146 static CallGenerator* for_warm_call(WarmCallInfo* ci, 147 CallGenerator* if_cold, 148 CallGenerator* if_hot); 149 150 static CallGenerator* for_late_inline_virtual(ciMethod* m, int vtable_index, float expected_uses); 151 152 // How to make a call that optimistically assumes a receiver type: 153 static CallGenerator* for_predicted_call(ciKlass* predicted_receiver, 154 CallGenerator* if_missed, 155 CallGenerator* if_hit, 156 float hit_prob); 157 158 static CallGenerator* for_guarded_call(ciKlass* predicted_receiver, 159 CallGenerator* if_missed, 160 CallGenerator* if_hit); 161 162 // How to make a call that optimistically assumes a MethodHandle target: 163 static CallGenerator* for_predicted_dynamic_call(ciMethodHandle* predicted_method_handle, 164 CallGenerator* if_missed, 165 CallGenerator* if_hit, 166 float hit_prob); 167 168 // How to make a call that gives up and goes back to the interpreter: 169 static CallGenerator* for_uncommon_trap(ciMethod* m, 170 Deoptimization::DeoptReason reason, 171 Deoptimization::DeoptAction action); 172 173 // Registry for intrinsics: 174 static CallGenerator* for_intrinsic(ciMethod* m); 175 static void register_intrinsic(ciMethod* m, CallGenerator* cg); 176 static CallGenerator* for_predicated_intrinsic(CallGenerator* intrinsic, 177 CallGenerator* cg); generate_predicate(JVMState * jvms,int predicate)178 virtual Node* generate_predicate(JVMState* jvms, int predicate) { return NULL; }; 179 print_inlining_late(const char * msg)180 virtual void print_inlining_late(const char* msg) { ShouldNotReachHere(); } 181 print_inlining(Compile * C,ciMethod * callee,int inline_level,int bci,const char * msg)182 static void print_inlining(Compile* C, ciMethod* callee, int inline_level, int bci, const char* msg) { 183 if (C->print_inlining()) { 184 C->print_inlining(callee, inline_level, bci, msg); 185 } 186 } 187 print_inlining_failure(Compile * C,ciMethod * callee,int inline_level,int bci,const char * msg)188 static void print_inlining_failure(Compile* C, ciMethod* callee, int inline_level, int bci, const char* msg) { 189 print_inlining(C, callee, inline_level, bci, msg); 190 C->log_inline_failure(msg); 191 } 192 193 static bool is_inlined_method_handle_intrinsic(JVMState* jvms, ciMethod* m); 194 static bool is_inlined_method_handle_intrinsic(ciMethod* caller, int bci, ciMethod* m); 195 static bool is_inlined_method_handle_intrinsic(ciMethod* symbolic_info, ciMethod* m); 196 }; 197 198 199 //------------------------InlineCallGenerator---------------------------------- 200 class InlineCallGenerator : public CallGenerator { 201 protected: InlineCallGenerator(ciMethod * method)202 InlineCallGenerator(ciMethod* method) : CallGenerator(method) {} 203 204 public: is_inline() const205 virtual bool is_inline() const { return true; } 206 }; 207 208 209 //---------------------------WarmCallInfo-------------------------------------- 210 // A struct to collect information about a given call site. 211 // Helps sort call sites into "hot", "medium", and "cold". 212 // Participates in the queueing of "medium" call sites for possible inlining. 213 class WarmCallInfo : public ResourceObj { 214 private: 215 216 CallNode* _call; // The CallNode which may be inlined. 217 CallGenerator* _hot_cg;// CG for expanding the call node 218 219 // These are the metrics we use to evaluate call sites: 220 221 float _count; // How often do we expect to reach this site? 222 float _profit; // How much time do we expect to save by inlining? 223 float _work; // How long do we expect the average call to take? 224 float _size; // How big do we expect the inlined code to be? 225 226 float _heat; // Combined score inducing total order on call sites. 227 WarmCallInfo* _next; // Next cooler call info in pending queue. 228 229 // Count is the number of times this call site is expected to be executed. 230 // Large count is favorable for inlining, because the extra compilation 231 // work will be amortized more completely. 232 233 // Profit is a rough measure of the amount of time we expect to save 234 // per execution of this site if we inline it. (1.0 == call overhead) 235 // Large profit favors inlining. Negative profit disables inlining. 236 237 // Work is a rough measure of the amount of time a typical out-of-line 238 // call from this site is expected to take. (1.0 == call, no-op, return) 239 // Small work is somewhat favorable for inlining, since methods with 240 // short "hot" traces are more likely to inline smoothly. 241 242 // Size is the number of graph nodes we expect this method to produce, 243 // not counting the inlining of any further warm calls it may include. 244 // Small size favors inlining, since small methods are more likely to 245 // inline smoothly. The size is estimated by examining the native code 246 // if available. The method bytecodes are also examined, assuming 247 // empirically observed node counts for each kind of bytecode. 248 249 // Heat is the combined "goodness" of a site's inlining. If we were 250 // omniscient, it would be the difference of two sums of future execution 251 // times of code emitted for this site (amortized across multiple sites if 252 // sharing applies). The two sums are for versions of this call site with 253 // and without inlining. 254 255 // We approximate this mythical quantity by playing with averages, 256 // rough estimates, and assumptions that history repeats itself. 257 // The basic formula count * profit is heuristically adjusted 258 // by looking at the expected compilation and execution times of 259 // of the inlined call. 260 261 // Note: Some of these metrics may not be present in the final product, 262 // but exist in development builds to experiment with inline policy tuning. 263 264 // This heuristic framework does not model well the very significant 265 // effects of multiple-level inlining. It is possible to see no immediate 266 // profit from inlining X->Y, but to get great profit from a subsequent 267 // inlining X->Y->Z. 268 269 // This framework does not take well into account the problem of N**2 code 270 // size in a clique of mutually inlinable methods. 271 next() const272 WarmCallInfo* next() const { return _next; } set_next(WarmCallInfo * n)273 void set_next(WarmCallInfo* n) { _next = n; } 274 275 static WarmCallInfo _always_hot; 276 static WarmCallInfo _always_cold; 277 278 // Constructor intitialization of always_hot and always_cold WarmCallInfo(float c,float p,float w,float s)279 WarmCallInfo(float c, float p, float w, float s) { 280 _call = NULL; 281 _hot_cg = NULL; 282 _next = NULL; 283 _count = c; 284 _profit = p; 285 _work = w; 286 _size = s; 287 _heat = 0; 288 } 289 290 public: 291 // Because WarmInfo objects live over the entire lifetime of the 292 // Compile object, they are allocated into the comp_arena, which 293 // does not get resource marked or reset during the compile process operator new(size_t x,Compile * C)294 void *operator new( size_t x, Compile* C ) throw() { return C->comp_arena()->Amalloc(x); } operator delete(void *)295 void operator delete( void * ) { } // fast deallocation 296 297 static WarmCallInfo* always_hot(); 298 static WarmCallInfo* always_cold(); 299 WarmCallInfo()300 WarmCallInfo() { 301 _call = NULL; 302 _hot_cg = NULL; 303 _next = NULL; 304 _count = _profit = _work = _size = _heat = 0; 305 } 306 call() const307 CallNode* call() const { return _call; } count() const308 float count() const { return _count; } size() const309 float size() const { return _size; } work() const310 float work() const { return _work; } profit() const311 float profit() const { return _profit; } heat() const312 float heat() const { return _heat; } 313 set_count(float x)314 void set_count(float x) { _count = x; } set_size(float x)315 void set_size(float x) { _size = x; } set_work(float x)316 void set_work(float x) { _work = x; } set_profit(float x)317 void set_profit(float x) { _profit = x; } set_heat(float x)318 void set_heat(float x) { _heat = x; } 319 320 // Load initial heuristics from profiles, etc. 321 // The heuristics can be tweaked further by the caller. 322 void init(JVMState* call_site, ciMethod* call_method, ciCallProfile& profile, float prof_factor); 323 MAX_VALUE()324 static float MAX_VALUE() { return +1.0e10; } MIN_VALUE()325 static float MIN_VALUE() { return -1.0e10; } 326 327 float compute_heat() const; 328 set_call(CallNode * call)329 void set_call(CallNode* call) { _call = call; } set_hot_cg(CallGenerator * cg)330 void set_hot_cg(CallGenerator* cg) { _hot_cg = cg; } 331 332 // Do not queue very hot or very cold calls. 333 // Make very cold ones out of line immediately. 334 // Inline very hot ones immediately. 335 // These queries apply various tunable limits 336 // to the above metrics in a systematic way. 337 // Test for coldness before testing for hotness. 338 bool is_cold() const; 339 bool is_hot() const; 340 341 // Force a warm call to be hot. This worklists the call node for inlining. 342 void make_hot(); 343 344 // Force a warm call to be cold. This worklists the call node for out-of-lining. 345 void make_cold(); 346 347 // A reproducible total ordering, in which heat is the major key. 348 bool warmer_than(WarmCallInfo* that); 349 350 // List management. These methods are called with the list head, 351 // and return the new list head, inserting or removing the receiver. 352 WarmCallInfo* insert_into(WarmCallInfo* head); 353 WarmCallInfo* remove_from(WarmCallInfo* head); 354 355 #ifndef PRODUCT 356 void print() const; 357 void print_all() const; 358 int count_all() const; 359 #endif 360 }; 361 362 #endif // SHARE_OPTO_CALLGENERATOR_HPP 363