1 /*
2 * Copyright (c) 2000, 2005, 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 //---------------------------CallGenerator-------------------------------------
26 // The subclasses of this class handle generation of ideal nodes for
27 // call sites and method entry points.
28
29 class CallGenerator : public ResourceObj {
30 public:
31 enum {
32 xxxunusedxxx
33 };
34
35 private:
36 ciMethod* _method; // The method being called.
37
38 protected:
39 CallGenerator(ciMethod* method);
40
41 public:
42 // Accessors
43 ciMethod* method() const { return _method; }
44
45 // is_inline: At least some code implementing the method is copied here.
46 virtual bool is_inline() const { return false; }
47 // is_intrinsic: There's a method-specific way of generating the inline code.
48 virtual bool is_intrinsic() const { return false; }
49 // is_parse: Bytecodes implementing the specific method are copied here.
50 virtual bool is_parse() const { return false; }
51 // is_virtual: The call uses the receiver type to select or check the method.
52 virtual bool is_virtual() const { return false; }
53 // is_deferred: The decision whether to inline or not is deferred.
54 virtual bool is_deferred() const { return false; }
55 // is_predicted: Uses an explicit check against a predicted type.
56 virtual bool is_predicted() const { return false; }
57 // is_trap: Does not return to the caller. (E.g., uncommon trap.)
58 virtual bool is_trap() const { return false; }
59
60 // is_late_inline: supports conversion of call into an inline
61 virtual bool is_late_inline() const { return false; }
62 // Replace the call with an inline version of the code
63 virtual void do_late_inline() { ShouldNotReachHere(); }
64
65 virtual CallStaticJavaNode* call_node() const { ShouldNotReachHere(); return NULL; }
66
67 // Note: It is possible for a CG to be both inline and virtual.
68 // (The hashCode intrinsic does a vtable check and an inlined fast path.)
69
70 // Utilities:
71 const TypeFunc* tf() const;
72
73 // The given jvms has state and arguments for a call to my method.
74 // Edges after jvms->argoff() carry all (pre-popped) argument values.
75 //
76 // Update the map with state and return values (if any) and return it.
77 // The return values (0, 1, or 2) must be pushed on the map's stack,
78 // and the sp of the jvms incremented accordingly.
79 //
80 // The jvms is returned on success. Alternatively, a copy of the
81 // given jvms, suitably updated, may be returned, in which case the
82 // caller should discard the original jvms.
83 //
84 // The non-Parm edges of the returned map will contain updated global state,
85 // and one or two edges before jvms->sp() will carry any return values.
86 // Other map edges may contain locals or monitors, and should not
87 // be changed in meaning.
88 //
89 // If the call traps, the returned map must have a control edge of top.
90 // If the call can throw, the returned map must report has_exceptions().
91 //
92 // If the result is NULL, it means that this CallGenerator was unable
93 // to handle the given call, and another CallGenerator should be consulted.
94 virtual JVMState* generate(JVMState* jvms) = 0;
95
96 // How to generate a call site that is inlined:
97 static CallGenerator* for_inline(ciMethod* m, float expected_uses = -1);
98 // How to generate code for an on-stack replacement handler.
99 static CallGenerator* for_osr(ciMethod* m, int osr_bci);
100
101 // How to generate vanilla out-of-line call sites:
102 static CallGenerator* for_direct_call(ciMethod* m, bool separate_io_projs = false); // static, special
103 static CallGenerator* for_dynamic_call(ciMethod* m); // invokedynamic
104 static CallGenerator* for_virtual_call(ciMethod* m, int vtable_index); // virtual, interface
105
106 // How to generate a replace a direct call with an inline version
107 static CallGenerator* for_late_inline(ciMethod* m, CallGenerator* inline_cg);
108
109 // How to make a call but defer the decision whether to inline or not.
110 static CallGenerator* for_warm_call(WarmCallInfo* ci,
111 CallGenerator* if_cold,
112 CallGenerator* if_hot);
113
114 // How to make a call that optimistically assumes a receiver type:
115 static CallGenerator* for_predicted_call(ciKlass* predicted_receiver,
116 CallGenerator* if_missed,
117 CallGenerator* if_hit,
118 float hit_prob);
119
120 // How to make a call that optimistically assumes a MethodHandle target:
121 static CallGenerator* for_predicted_dynamic_call(ciMethodHandle* predicted_method_handle,
122 CallGenerator* if_missed,
123 CallGenerator* if_hit,
124 float hit_prob);
125
126 // How to make a call that gives up and goes back to the interpreter:
127 static CallGenerator* for_uncommon_trap(ciMethod* m,
128 Deoptimization::DeoptReason reason,
129 Deoptimization::DeoptAction action);
130
131 // Registry for intrinsics:
132 static CallGenerator* for_intrinsic(ciMethod* m);
133 static void register_intrinsic(ciMethod* m, CallGenerator* cg);
134 };
135
136 class InlineCallGenerator : public CallGenerator {
137 virtual bool is_inline() const { return true; }
138
139 protected:
140 InlineCallGenerator(ciMethod* method) : CallGenerator(method) { }
141 };
142
143
144 //---------------------------WarmCallInfo--------------------------------------
145 // A struct to collect information about a given call site.
146 // Helps sort call sites into "hot", "medium", and "cold".
147 // Participates in the queueing of "medium" call sites for possible inlining.
148 class WarmCallInfo : public ResourceObj {
149 private:
150
151 CallNode* _call; // The CallNode which may be inlined.
152 CallGenerator* _hot_cg;// CG for expanding the call node
153
154 // These are the metrics we use to evaluate call sites:
155
156 float _count; // How often do we expect to reach this site?
157 float _profit; // How much time do we expect to save by inlining?
158 float _work; // How long do we expect the average call to take?
159 float _size; // How big do we expect the inlined code to be?
160
161 float _heat; // Combined score inducing total order on call sites.
162 WarmCallInfo* _next; // Next cooler call info in pending queue.
163
164 // Count is the number of times this call site is expected to be executed.
165 // Large count is favorable for inlining, because the extra compilation
166 // work will be amortized more completely.
167
168 // Profit is a rough measure of the amount of time we expect to save
169 // per execution of this site if we inline it. (1.0 == call overhead)
170 // Large profit favors inlining. Negative profit disables inlining.
171
172 // Work is a rough measure of the amount of time a typical out-of-line
173 // call from this site is expected to take. (1.0 == call, no-op, return)
174 // Small work is somewhat favorable for inlining, since methods with
175 // short "hot" traces are more likely to inline smoothly.
176
177 // Size is the number of graph nodes we expect this method to produce,
178 // not counting the inlining of any further warm calls it may include.
179 // Small size favors inlining, since small methods are more likely to
180 // inline smoothly. The size is estimated by examining the native code
181 // if available. The method bytecodes are also examined, assuming
182 // empirically observed node counts for each kind of bytecode.
183
184 // Heat is the combined "goodness" of a site's inlining. If we were
185 // omniscient, it would be the difference of two sums of future execution
186 // times of code emitted for this site (amortized across multiple sites if
187 // sharing applies). The two sums are for versions of this call site with
188 // and without inlining.
189
190 // We approximate this mythical quantity by playing with averages,
191 // rough estimates, and assumptions that history repeats itself.
192 // The basic formula count * profit is heuristically adjusted
193 // by looking at the expected compilation and execution times of
194 // of the inlined call.
195
196 // Note: Some of these metrics may not be present in the final product,
197 // but exist in development builds to experiment with inline policy tuning.
198
199 // This heuristic framework does not model well the very significant
200 // effects of multiple-level inlining. It is possible to see no immediate
201 // profit from inlining X->Y, but to get great profit from a subsequent
202 // inlining X->Y->Z.
203
204 // This framework does not take well into account the problem of N**2 code
205 // size in a clique of mutually inlinable methods.
206
207 WarmCallInfo* next() const { return _next; }
208 void set_next(WarmCallInfo* n) { _next = n; }
209
210 static WarmCallInfo* _always_hot;
211 static WarmCallInfo* _always_cold;
212
213 public:
214 // Because WarmInfo objects live over the entire lifetime of the
215 // Compile object, they are allocated into the comp_arena, which
216 // does not get resource marked or reset during the compile process
217 void *operator new( size_t x, Compile* C ) { return C->comp_arena()->Amalloc(x); }
218 void operator delete( void * ) { } // fast deallocation
219
220 static WarmCallInfo* always_hot();
221 static WarmCallInfo* always_cold();
222
223 WarmCallInfo() {
224 _call = NULL;
225 _hot_cg = NULL;
226 _next = NULL;
227 _count = _profit = _work = _size = _heat = 0;
228 }
229
230 CallNode* call() const { return _call; }
231 float count() const { return _count; }
232 float size() const { return _size; }
233 float work() const { return _work; }
234 float profit() const { return _profit; }
235 float heat() const { return _heat; }
236
237 void set_count(float x) { _count = x; }
238 void set_size(float x) { _size = x; }
239 void set_work(float x) { _work = x; }
240 void set_profit(float x) { _profit = x; }
241 void set_heat(float x) { _heat = x; }
242
243 // Load initial heuristics from profiles, etc.
244 // The heuristics can be tweaked further by the caller.
245 void init(JVMState* call_site, ciMethod* call_method, ciCallProfile& profile, float prof_factor);
246
247 static float MAX_VALUE() { return +1.0e10; }
248 static float MIN_VALUE() { return -1.0e10; }
249
250 float compute_heat() const;
251
252 void set_call(CallNode* call) { _call = call; }
253 void set_hot_cg(CallGenerator* cg) { _hot_cg = cg; }
254
255 // Do not queue very hot or very cold calls.
256 // Make very cold ones out of line immediately.
257 // Inline very hot ones immediately.
258 // These queries apply various tunable limits
259 // to the above metrics in a systematic way.
260 // Test for coldness before testing for hotness.
261 bool is_cold() const;
262 bool is_hot() const;
263
264 // Force a warm call to be hot. This worklists the call node for inlining.
265 void make_hot();
266
267 // Force a warm call to be cold. This worklists the call node for out-of-lining.
268 void make_cold();
269
270 // A reproducible total ordering, in which heat is the major key.
271 bool warmer_than(WarmCallInfo* that);
272
273 // List management. These methods are called with the list head,
274 // and return the new list head, inserting or removing the receiver.
275 WarmCallInfo* insert_into(WarmCallInfo* head);
276 WarmCallInfo* remove_from(WarmCallInfo* head);
277
278 #ifndef PRODUCT
279 void print() const;
280 void print_all() const;
281 int count_all() const;
282 #endif
283 };