1 /*
2 * Copyright (c) 2000, 2012, 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 //---------------------------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 public:
40 enum {
41 xxxunusedxxx
42 };
43
44 private:
45 ciMethod* _method; // The method being called.
46
47 protected:
48 CallGenerator(ciMethod* method) : _method(method) {}
49
50 public:
51 // Accessors
52 ciMethod* method() const { return _method; }
53
54 // is_inline: At least some code implementing the method is copied here.
55 virtual bool is_inline() const { return false; }
56 // is_intrinsic: There's a method-specific way of generating the inline code.
57 virtual bool is_intrinsic() const { return false; }
58 // is_parse: Bytecodes implementing the specific method are copied here.
59 virtual bool is_parse() const { return false; }
60 // is_virtual: The call uses the receiver type to select or check the method.
61 virtual bool is_virtual() const { return false; }
62 // is_deferred: The decision whether to inline or not is deferred.
63 virtual bool is_deferred() const { return false; }
64 // is_predicted: Uses an explicit check against a predicted type.
65 virtual bool is_predicted() const { return false; }
66 // is_trap: Does not return to the caller. (E.g., uncommon trap.)
67 virtual bool is_trap() const { return false; }
68
69 // is_late_inline: supports conversion of call into an inline
70 virtual bool is_late_inline() const { return false; }
71 // Replace the call with an inline version of the code
72 virtual void do_late_inline() { ShouldNotReachHere(); }
73
74 virtual CallStaticJavaNode* call_node() const { ShouldNotReachHere(); return NULL; }
75
76 // Note: It is possible for a CG to be both inline and virtual.
77 // (The hashCode intrinsic does a vtable check and an inlined fast path.)
78
79 // Utilities:
80 const TypeFunc* tf() const;
81
82 // The given jvms has state and arguments for a call to my method.
83 // Edges after jvms->argoff() carry all (pre-popped) argument values.
84 //
85 // Update the map with state and return values (if any) and return it.
86 // The return values (0, 1, or 2) must be pushed on the map's stack,
87 // and the sp of the jvms incremented accordingly.
88 //
89 // The jvms is returned on success. Alternatively, a copy of the
90 // given jvms, suitably updated, may be returned, in which case the
91 // caller should discard the original jvms.
92 //
93 // The non-Parm edges of the returned map will contain updated global state,
94 // and one or two edges before jvms->sp() will carry any return values.
95 // Other map edges may contain locals or monitors, and should not
96 // be changed in meaning.
97 //
98 // If the call traps, the returned map must have a control edge of top.
99 // If the call can throw, the returned map must report has_exceptions().
100 //
101 // If the result is NULL, it means that this CallGenerator was unable
102 // to handle the given call, and another CallGenerator should be consulted.
103 virtual JVMState* generate(JVMState* jvms) = 0;
104
105 // How to generate a call site that is inlined:
106 static CallGenerator* for_inline(ciMethod* m, float expected_uses = -1);
107 // How to generate code for an on-stack replacement handler.
108 static CallGenerator* for_osr(ciMethod* m, int osr_bci);
109
110 // How to generate vanilla out-of-line call sites:
111 static CallGenerator* for_direct_call(ciMethod* m, bool separate_io_projs = false); // static, special
112 static CallGenerator* for_virtual_call(ciMethod* m, int vtable_index); // virtual, interface
113 static CallGenerator* for_dynamic_call(ciMethod* m); // invokedynamic
114
115 static CallGenerator* for_method_handle_call( JVMState* jvms, ciMethod* caller, ciMethod* callee);
116 static CallGenerator* for_method_handle_inline(JVMState* jvms, ciMethod* caller, ciMethod* callee);
117
118 // How to generate a replace a direct call with an inline version
119 static CallGenerator* for_late_inline(ciMethod* m, CallGenerator* inline_cg);
120
121 // How to make a call but defer the decision whether to inline or not.
122 static CallGenerator* for_warm_call(WarmCallInfo* ci,
123 CallGenerator* if_cold,
124 CallGenerator* if_hot);
125
126 // How to make a call that optimistically assumes a receiver type:
127 static CallGenerator* for_predicted_call(ciKlass* predicted_receiver,
128 CallGenerator* if_missed,
129 CallGenerator* if_hit,
130 float hit_prob);
131
132 // How to make a call that optimistically assumes a MethodHandle target:
133 static CallGenerator* for_predicted_dynamic_call(ciMethodHandle* predicted_method_handle,
134 CallGenerator* if_missed,
135 CallGenerator* if_hit,
136 float hit_prob);
137
138 // How to make a call that gives up and goes back to the interpreter:
139 static CallGenerator* for_uncommon_trap(ciMethod* m,
140 Deoptimization::DeoptReason reason,
141 Deoptimization::DeoptAction action);
142
143 // Registry for intrinsics:
144 static CallGenerator* for_intrinsic(ciMethod* m);
145 static void register_intrinsic(ciMethod* m, CallGenerator* cg);
146 static CallGenerator* for_predicted_intrinsic(CallGenerator* intrinsic,
147 CallGenerator* cg);
148 virtual Node* generate_predicate(JVMState* jvms) { return NULL; };
149
150 static void print_inlining(Compile* C, ciMethod* callee, int inline_level, int bci, const char* msg) {
151 if (PrintInlining)
152 C->print_inlining(callee, inline_level, bci, msg);
153 }
154 };
155
156
157 //------------------------InlineCallGenerator----------------------------------
158 class InlineCallGenerator : public CallGenerator {
159 protected:
160 InlineCallGenerator(ciMethod* method) : CallGenerator(method) {}
161
162 public:
163 virtual bool is_inline() const { return true; }
164 };
165
166
167 //---------------------------WarmCallInfo--------------------------------------
168 // A struct to collect information about a given call site.
169 // Helps sort call sites into "hot", "medium", and "cold".
170 // Participates in the queueing of "medium" call sites for possible inlining.
171 class WarmCallInfo : public ResourceObj {
172 private:
173
174 CallNode* _call; // The CallNode which may be inlined.
175 CallGenerator* _hot_cg;// CG for expanding the call node
176
177 // These are the metrics we use to evaluate call sites:
178
179 float _count; // How often do we expect to reach this site?
180 float _profit; // How much time do we expect to save by inlining?
181 float _work; // How long do we expect the average call to take?
182 float _size; // How big do we expect the inlined code to be?
183
184 float _heat; // Combined score inducing total order on call sites.
185 WarmCallInfo* _next; // Next cooler call info in pending queue.
186
187 // Count is the number of times this call site is expected to be executed.
188 // Large count is favorable for inlining, because the extra compilation
189 // work will be amortized more completely.
190
191 // Profit is a rough measure of the amount of time we expect to save
192 // per execution of this site if we inline it. (1.0 == call overhead)
193 // Large profit favors inlining. Negative profit disables inlining.
194
195 // Work is a rough measure of the amount of time a typical out-of-line
196 // call from this site is expected to take. (1.0 == call, no-op, return)
197 // Small work is somewhat favorable for inlining, since methods with
198 // short "hot" traces are more likely to inline smoothly.
199
200 // Size is the number of graph nodes we expect this method to produce,
201 // not counting the inlining of any further warm calls it may include.
202 // Small size favors inlining, since small methods are more likely to
203 // inline smoothly. The size is estimated by examining the native code
204 // if available. The method bytecodes are also examined, assuming
205 // empirically observed node counts for each kind of bytecode.
206
207 // Heat is the combined "goodness" of a site's inlining. If we were
208 // omniscient, it would be the difference of two sums of future execution
209 // times of code emitted for this site (amortized across multiple sites if
210 // sharing applies). The two sums are for versions of this call site with
211 // and without inlining.
212
213 // We approximate this mythical quantity by playing with averages,
214 // rough estimates, and assumptions that history repeats itself.
215 // The basic formula count * profit is heuristically adjusted
216 // by looking at the expected compilation and execution times of
217 // of the inlined call.
218
219 // Note: Some of these metrics may not be present in the final product,
220 // but exist in development builds to experiment with inline policy tuning.
221
222 // This heuristic framework does not model well the very significant
223 // effects of multiple-level inlining. It is possible to see no immediate
224 // profit from inlining X->Y, but to get great profit from a subsequent
225 // inlining X->Y->Z.
226
227 // This framework does not take well into account the problem of N**2 code
228 // size in a clique of mutually inlinable methods.
229
230 WarmCallInfo* next() const { return _next; }
231 void set_next(WarmCallInfo* n) { _next = n; }
232
233 static WarmCallInfo _always_hot;
234 static WarmCallInfo _always_cold;
235
236 // Constructor intitialization of always_hot and always_cold
237 WarmCallInfo(float c, float p, float w, float s) {
238 _call = NULL;
239 _hot_cg = NULL;
240 _next = NULL;
241 _count = c;
242 _profit = p;
243 _work = w;
244 _size = s;
245 _heat = 0;
246 }
247
248 public:
249 // Because WarmInfo objects live over the entire lifetime of the
250 // Compile object, they are allocated into the comp_arena, which
251 // does not get resource marked or reset during the compile process
252 void *operator new( size_t x, Compile* C ) { return C->comp_arena()->Amalloc(x); }
253 void operator delete( void * ) { } // fast deallocation
254
255 static WarmCallInfo* always_hot();
256 static WarmCallInfo* always_cold();
257
258 WarmCallInfo() {
259 _call = NULL;
260 _hot_cg = NULL;
261 _next = NULL;
262 _count = _profit = _work = _size = _heat = 0;
263 }
264
265 CallNode* call() const { return _call; }
266 float count() const { return _count; }
267 float size() const { return _size; }
268 float work() const { return _work; }
269 float profit() const { return _profit; }
270 float heat() const { return _heat; }
271
272 void set_count(float x) { _count = x; }
273 void set_size(float x) { _size = x; }
274 void set_work(float x) { _work = x; }
275 void set_profit(float x) { _profit = x; }
276 void set_heat(float x) { _heat = x; }
277
278 // Load initial heuristics from profiles, etc.
279 // The heuristics can be tweaked further by the caller.
280 void init(JVMState* call_site, ciMethod* call_method, ciCallProfile& profile, float prof_factor);
281
282 static float MAX_VALUE() { return +1.0e10; }
283 static float MIN_VALUE() { return -1.0e10; }
284
285 float compute_heat() const;
286
287 void set_call(CallNode* call) { _call = call; }
288 void set_hot_cg(CallGenerator* cg) { _hot_cg = cg; }
289
290 // Do not queue very hot or very cold calls.
291 // Make very cold ones out of line immediately.
292 // Inline very hot ones immediately.
293 // These queries apply various tunable limits
294 // to the above metrics in a systematic way.
295 // Test for coldness before testing for hotness.
296 bool is_cold() const;
297 bool is_hot() const;
298
299 // Force a warm call to be hot. This worklists the call node for inlining.
300 void make_hot();
301
302 // Force a warm call to be cold. This worklists the call node for out-of-lining.
303 void make_cold();
304
305 // A reproducible total ordering, in which heat is the major key.
306 bool warmer_than(WarmCallInfo* that);
307
308 // List management. These methods are called with the list head,
309 // and return the new list head, inserting or removing the receiver.
310 WarmCallInfo* insert_into(WarmCallInfo* head);
311 WarmCallInfo* remove_from(WarmCallInfo* head);
312
313 #ifndef PRODUCT
314 void print() const;
315 void print_all() const;
316 int count_all() const;
317 #endif
318 };
319
320 #endif // SHARE_VM_OPTO_CALLGENERATOR_HPP