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