1 /*
2 * Copyright (c) 2000, 2013, 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 "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 class Parse;
35
36 //---------------------------CallGenerator-------------------------------------
37 // The subclasses of this class handle generation of ideal nodes for
38 // call sites and method entry points.
39
40 class CallGenerator : public ResourceObj {
41 public:
42 enum {
43 xxxunusedxxx
44 };
45
46 private:
47 ciMethod* _method; // The method being called.
48
49 protected:
50 CallGenerator(ciMethod* method) : _method(method) {}
51
52 public:
53 // Accessors
54 ciMethod* method() const { return _method; }
55
56 // is_inline: At least some code implementing the method is copied here.
57 virtual bool is_inline() const { return false; }
58 // is_intrinsic: There's a method-specific way of generating the inline code.
59 virtual bool is_intrinsic() const { return false; }
60 // is_parse: Bytecodes implementing the specific method are copied here.
61 virtual bool is_parse() const { return false; }
62 // is_virtual: The call uses the receiver type to select or check the method.
63 virtual bool is_virtual() const { return false; }
64 // is_deferred: The decision whether to inline or not is deferred.
65 virtual bool is_deferred() const { return false; }
66 // is_predicted: Uses an explicit check against a predicted type.
67 virtual bool is_predicted() const { return false; }
68 // is_trap: Does not return to the caller. (E.g., uncommon trap.)
69 virtual bool is_trap() const { return false; }
70 // does_virtual_dispatch: Should try inlining as normal method first.
71 virtual bool does_virtual_dispatch() const { return false; }
72
73 // is_late_inline: supports conversion of call into an inline
74 virtual bool is_late_inline() const { return false; }
75 // same but for method handle calls
76 virtual bool is_mh_late_inline() const { return false; }
77 virtual bool is_string_late_inline() const{ return false; }
78
79 // for method handle calls: have we tried inlinining the call already?
80 virtual bool already_attempted() const { ShouldNotReachHere(); return false; }
81
82 // Replace the call with an inline version of the code
83 virtual void do_late_inline() { ShouldNotReachHere(); }
84
85 virtual CallStaticJavaNode* call_node() const { ShouldNotReachHere(); return NULL; }
86
87 virtual void set_unique_id(jlong id) { fatal("unique id only for late inlines"); };
88 virtual jlong unique_id() const { fatal("unique id only for late inlines"); return 0; };
89
90 // Note: It is possible for a CG to be both inline and virtual.
91 // (The hashCode intrinsic does a vtable check and an inlined fast path.)
92
93 // Utilities:
94 const TypeFunc* tf() const;
95
96 // The given jvms has state and arguments for a call to my method.
97 // Edges after jvms->argoff() carry all (pre-popped) argument values.
98 //
99 // Update the map with state and return values (if any) and return it.
100 // The return values (0, 1, or 2) must be pushed on the map's stack,
101 // and the sp of the jvms incremented accordingly.
102 //
103 // The jvms is returned on success. Alternatively, a copy of the
104 // given jvms, suitably updated, may be returned, in which case the
105 // caller should discard the original jvms.
106 //
107 // The non-Parm edges of the returned map will contain updated global state,
108 // and one or two edges before jvms->sp() will carry any return values.
109 // Other map edges may contain locals or monitors, and should not
110 // be changed in meaning.
111 //
112 // If the call traps, the returned map must have a control edge of top.
113 // If the call can throw, the returned map must report has_exceptions().
114 //
115 // If the result is NULL, it means that this CallGenerator was unable
116 // to handle the given call, and another CallGenerator should be consulted.
117 virtual JVMState* generate(JVMState* jvms, Parse* parent_parser) = 0;
118
119 // How to generate a call site that is inlined:
120 static CallGenerator* for_inline(ciMethod* m, float expected_uses = -1);
121 // How to generate code for an on-stack replacement handler.
122 static CallGenerator* for_osr(ciMethod* m, int osr_bci);
123
124 // How to generate vanilla out-of-line call sites:
125 static CallGenerator* for_direct_call(ciMethod* m, bool separate_io_projs = false); // static, special
126 static CallGenerator* for_virtual_call(ciMethod* m, int vtable_index); // virtual, interface
127 static CallGenerator* for_dynamic_call(ciMethod* m); // invokedynamic
128
129 static CallGenerator* for_method_handle_call( JVMState* jvms, ciMethod* caller, ciMethod* callee, bool delayed_forbidden);
130 static CallGenerator* for_method_handle_inline(JVMState* jvms, ciMethod* caller, ciMethod* callee, bool& input_not_const);
131
132 // How to generate a replace a direct call with an inline version
133 static CallGenerator* for_late_inline(ciMethod* m, CallGenerator* inline_cg);
134 static CallGenerator* for_mh_late_inline(ciMethod* caller, ciMethod* callee, bool input_not_const);
135 static CallGenerator* for_string_late_inline(ciMethod* m, CallGenerator* inline_cg);
136 static CallGenerator* for_boxing_late_inline(ciMethod* m, CallGenerator* inline_cg);
137
138 // How to make a call but defer the decision whether to inline or not.
139 static CallGenerator* for_warm_call(WarmCallInfo* ci,
140 CallGenerator* if_cold,
141 CallGenerator* if_hot);
142
143 // How to make a call that optimistically assumes a receiver type:
144 static CallGenerator* for_predicted_call(ciKlass* predicted_receiver,
145 CallGenerator* if_missed,
146 CallGenerator* if_hit,
147 float hit_prob);
148
149 // How to make a call that optimistically assumes a MethodHandle target:
150 static CallGenerator* for_predicted_dynamic_call(ciMethodHandle* predicted_method_handle,
151 CallGenerator* if_missed,
152 CallGenerator* if_hit,
153 float hit_prob);
154
155 // How to make a call that gives up and goes back to the interpreter:
156 static CallGenerator* for_uncommon_trap(ciMethod* m,
157 Deoptimization::DeoptReason reason,
158 Deoptimization::DeoptAction action);
159
160 // Registry for intrinsics:
161 static CallGenerator* for_intrinsic(ciMethod* m);
162 static void register_intrinsic(ciMethod* m, CallGenerator* cg);
163 static CallGenerator* for_predicted_intrinsic(CallGenerator* intrinsic,
164 CallGenerator* cg);
165 virtual Node* generate_predicate(JVMState* jvms) { return NULL; };
166
167 virtual void print_inlining_late(const char* msg) { ShouldNotReachHere(); }
168
169 static void print_inlining(Compile* C, ciMethod* callee, int inline_level, int bci, const char* msg) {
170 if (C->print_inlining()) {
171 C->print_inlining(callee, inline_level, bci, msg);
172 }
173 }
174 };
175
176
177 //------------------------InlineCallGenerator----------------------------------
178 class InlineCallGenerator : public CallGenerator {
179 protected:
180 InlineCallGenerator(ciMethod* method) : CallGenerator(method) {}
181
182 public:
183 virtual bool is_inline() const { return true; }
184 };
185
186
187 //---------------------------WarmCallInfo--------------------------------------
188 // A struct to collect information about a given call site.
189 // Helps sort call sites into "hot", "medium", and "cold".
190 // Participates in the queueing of "medium" call sites for possible inlining.
191 class WarmCallInfo : public ResourceObj {
192 private:
193
194 CallNode* _call; // The CallNode which may be inlined.
195 CallGenerator* _hot_cg;// CG for expanding the call node
196
197 // These are the metrics we use to evaluate call sites:
198
199 float _count; // How often do we expect to reach this site?
200 float _profit; // How much time do we expect to save by inlining?
201 float _work; // How long do we expect the average call to take?
202 float _size; // How big do we expect the inlined code to be?
203
204 float _heat; // Combined score inducing total order on call sites.
205 WarmCallInfo* _next; // Next cooler call info in pending queue.
206
207 // Count is the number of times this call site is expected to be executed.
208 // Large count is favorable for inlining, because the extra compilation
209 // work will be amortized more completely.
210
211 // Profit is a rough measure of the amount of time we expect to save
212 // per execution of this site if we inline it. (1.0 == call overhead)
213 // Large profit favors inlining. Negative profit disables inlining.
214
215 // Work is a rough measure of the amount of time a typical out-of-line
216 // call from this site is expected to take. (1.0 == call, no-op, return)
217 // Small work is somewhat favorable for inlining, since methods with
218 // short "hot" traces are more likely to inline smoothly.
219
220 // Size is the number of graph nodes we expect this method to produce,
221 // not counting the inlining of any further warm calls it may include.
222 // Small size favors inlining, since small methods are more likely to
223 // inline smoothly. The size is estimated by examining the native code
224 // if available. The method bytecodes are also examined, assuming
225 // empirically observed node counts for each kind of bytecode.
226
227 // Heat is the combined "goodness" of a site's inlining. If we were
228 // omniscient, it would be the difference of two sums of future execution
229 // times of code emitted for this site (amortized across multiple sites if
230 // sharing applies). The two sums are for versions of this call site with
231 // and without inlining.
232
233 // We approximate this mythical quantity by playing with averages,
234 // rough estimates, and assumptions that history repeats itself.
235 // The basic formula count * profit is heuristically adjusted
236 // by looking at the expected compilation and execution times of
237 // of the inlined call.
238
239 // Note: Some of these metrics may not be present in the final product,
240 // but exist in development builds to experiment with inline policy tuning.
241
242 // This heuristic framework does not model well the very significant
243 // effects of multiple-level inlining. It is possible to see no immediate
244 // profit from inlining X->Y, but to get great profit from a subsequent
245 // inlining X->Y->Z.
246
247 // This framework does not take well into account the problem of N**2 code
248 // size in a clique of mutually inlinable methods.
249
250 WarmCallInfo* next() const { return _next; }
251 void set_next(WarmCallInfo* n) { _next = n; }
252
253 static WarmCallInfo _always_hot;
254 static WarmCallInfo _always_cold;
255
256 // Constructor intitialization of always_hot and always_cold
257 WarmCallInfo(float c, float p, float w, float s) {
258 _call = NULL;
259 _hot_cg = NULL;
260 _next = NULL;
261 _count = c;
262 _profit = p;
263 _work = w;
264 _size = s;
265 _heat = 0;
266 }
267
268 public:
269 // Because WarmInfo objects live over the entire lifetime of the
270 // Compile object, they are allocated into the comp_arena, which
271 // does not get resource marked or reset during the compile process
272 void *operator new( size_t x, Compile* C ) throw() { return C->comp_arena()->Amalloc(x); }
273 void operator delete( void * ) { } // fast deallocation
274
275 static WarmCallInfo* always_hot();
276 static WarmCallInfo* always_cold();
277
278 WarmCallInfo() {
279 _call = NULL;
280 _hot_cg = NULL;
281 _next = NULL;
282 _count = _profit = _work = _size = _heat = 0;
283 }
284
285 CallNode* call() const { return _call; }
286 float count() const { return _count; }
287 float size() const { return _size; }
288 float work() const { return _work; }
289 float profit() const { return _profit; }
290 float heat() const { return _heat; }
291
292 void set_count(float x) { _count = x; }
293 void set_size(float x) { _size = x; }
294 void set_work(float x) { _work = x; }
295 void set_profit(float x) { _profit = x; }
296 void set_heat(float x) { _heat = x; }
297
298 // Load initial heuristics from profiles, etc.
299 // The heuristics can be tweaked further by the caller.
300 void init(JVMState* call_site, ciMethod* call_method, ciCallProfile& profile, float prof_factor);
301
302 static float MAX_VALUE() { return +1.0e10; }
303 static float MIN_VALUE() { return -1.0e10; }
304
305 float compute_heat() const;
306
307 void set_call(CallNode* call) { _call = call; }
308 void set_hot_cg(CallGenerator* cg) { _hot_cg = cg; }
309
310 // Do not queue very hot or very cold calls.
311 // Make very cold ones out of line immediately.
312 // Inline very hot ones immediately.
313 // These queries apply various tunable limits
314 // to the above metrics in a systematic way.
315 // Test for coldness before testing for hotness.
316 bool is_cold() const;
317 bool is_hot() const;
318
319 // Force a warm call to be hot. This worklists the call node for inlining.
320 void make_hot();
321
322 // Force a warm call to be cold. This worklists the call node for out-of-lining.
323 void make_cold();
324
325 // A reproducible total ordering, in which heat is the major key.
326 bool warmer_than(WarmCallInfo* that);
327
328 // List management. These methods are called with the list head,
329 // and return the new list head, inserting or removing the receiver.
330 WarmCallInfo* insert_into(WarmCallInfo* head);
331 WarmCallInfo* remove_from(WarmCallInfo* head);
332
333 #ifndef PRODUCT
334 void print() const;
335 void print_all() const;
336 int count_all() const;
337 #endif
338 };
339
340 #endif // SHARE_VM_OPTO_CALLGENERATOR_HPP