Tue, 05 Jan 2010 13:05:58 +0100
6829187: compiler optimizations required for JSR 292
Summary: C2 implementation for invokedynamic support.
Reviewed-by: kvn, never
1 /*
2 * Copyright 2000-2005 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
25 //---------------------------CallGenerator-------------------------------------
26 // The subclasses of this class handle generation of ideal nodes for
27 // call sites and method entry points.
29 class CallGenerator : public ResourceObj {
30 public:
31 enum {
32 xxxunusedxxx
33 };
35 private:
36 ciMethod* _method; // The method being called.
38 protected:
39 CallGenerator(ciMethod* method);
41 public:
42 // Accessors
43 ciMethod* method() const { return _method; }
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; }
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(); }
65 virtual CallStaticJavaNode* call_node() const { ShouldNotReachHere(); return NULL; }
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.)
70 // Utilities:
71 const TypeFunc* tf() const;
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;
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);
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
106 // How to generate a replace a direct call with an inline version
107 static CallGenerator* for_late_inline(ciMethod* m, CallGenerator* inline_cg);
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);
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);
120 // How to make a call that gives up and goes back to the interpreter:
121 static CallGenerator* for_uncommon_trap(ciMethod* m,
122 Deoptimization::DeoptReason reason,
123 Deoptimization::DeoptAction action);
125 // Registry for intrinsics:
126 static CallGenerator* for_intrinsic(ciMethod* m);
127 static void register_intrinsic(ciMethod* m, CallGenerator* cg);
128 };
130 class InlineCallGenerator : public CallGenerator {
131 virtual bool is_inline() const { return true; }
133 protected:
134 InlineCallGenerator(ciMethod* method) : CallGenerator(method) { }
135 };
138 //---------------------------WarmCallInfo--------------------------------------
139 // A struct to collect information about a given call site.
140 // Helps sort call sites into "hot", "medium", and "cold".
141 // Participates in the queueing of "medium" call sites for possible inlining.
142 class WarmCallInfo : public ResourceObj {
143 private:
145 CallNode* _call; // The CallNode which may be inlined.
146 CallGenerator* _hot_cg;// CG for expanding the call node
148 // These are the metrics we use to evaluate call sites:
150 float _count; // How often do we expect to reach this site?
151 float _profit; // How much time do we expect to save by inlining?
152 float _work; // How long do we expect the average call to take?
153 float _size; // How big do we expect the inlined code to be?
155 float _heat; // Combined score inducing total order on call sites.
156 WarmCallInfo* _next; // Next cooler call info in pending queue.
158 // Count is the number of times this call site is expected to be executed.
159 // Large count is favorable for inlining, because the extra compilation
160 // work will be amortized more completely.
162 // Profit is a rough measure of the amount of time we expect to save
163 // per execution of this site if we inline it. (1.0 == call overhead)
164 // Large profit favors inlining. Negative profit disables inlining.
166 // Work is a rough measure of the amount of time a typical out-of-line
167 // call from this site is expected to take. (1.0 == call, no-op, return)
168 // Small work is somewhat favorable for inlining, since methods with
169 // short "hot" traces are more likely to inline smoothly.
171 // Size is the number of graph nodes we expect this method to produce,
172 // not counting the inlining of any further warm calls it may include.
173 // Small size favors inlining, since small methods are more likely to
174 // inline smoothly. The size is estimated by examining the native code
175 // if available. The method bytecodes are also examined, assuming
176 // empirically observed node counts for each kind of bytecode.
178 // Heat is the combined "goodness" of a site's inlining. If we were
179 // omniscient, it would be the difference of two sums of future execution
180 // times of code emitted for this site (amortized across multiple sites if
181 // sharing applies). The two sums are for versions of this call site with
182 // and without inlining.
184 // We approximate this mythical quantity by playing with averages,
185 // rough estimates, and assumptions that history repeats itself.
186 // The basic formula count * profit is heuristically adjusted
187 // by looking at the expected compilation and execution times of
188 // of the inlined call.
190 // Note: Some of these metrics may not be present in the final product,
191 // but exist in development builds to experiment with inline policy tuning.
193 // This heuristic framework does not model well the very significant
194 // effects of multiple-level inlining. It is possible to see no immediate
195 // profit from inlining X->Y, but to get great profit from a subsequent
196 // inlining X->Y->Z.
198 // This framework does not take well into account the problem of N**2 code
199 // size in a clique of mutually inlinable methods.
201 WarmCallInfo* next() const { return _next; }
202 void set_next(WarmCallInfo* n) { _next = n; }
204 static WarmCallInfo* _always_hot;
205 static WarmCallInfo* _always_cold;
207 public:
208 // Because WarmInfo objects live over the entire lifetime of the
209 // Compile object, they are allocated into the comp_arena, which
210 // does not get resource marked or reset during the compile process
211 void *operator new( size_t x, Compile* C ) { return C->comp_arena()->Amalloc(x); }
212 void operator delete( void * ) { } // fast deallocation
214 static WarmCallInfo* always_hot();
215 static WarmCallInfo* always_cold();
217 WarmCallInfo() {
218 _call = NULL;
219 _hot_cg = NULL;
220 _next = NULL;
221 _count = _profit = _work = _size = _heat = 0;
222 }
224 CallNode* call() const { return _call; }
225 float count() const { return _count; }
226 float size() const { return _size; }
227 float work() const { return _work; }
228 float profit() const { return _profit; }
229 float heat() const { return _heat; }
231 void set_count(float x) { _count = x; }
232 void set_size(float x) { _size = x; }
233 void set_work(float x) { _work = x; }
234 void set_profit(float x) { _profit = x; }
235 void set_heat(float x) { _heat = x; }
237 // Load initial heuristics from profiles, etc.
238 // The heuristics can be tweaked further by the caller.
239 void init(JVMState* call_site, ciMethod* call_method, ciCallProfile& profile, float prof_factor);
241 static float MAX_VALUE() { return +1.0e10; }
242 static float MIN_VALUE() { return -1.0e10; }
244 float compute_heat() const;
246 void set_call(CallNode* call) { _call = call; }
247 void set_hot_cg(CallGenerator* cg) { _hot_cg = cg; }
249 // Do not queue very hot or very cold calls.
250 // Make very cold ones out of line immediately.
251 // Inline very hot ones immediately.
252 // These queries apply various tunable limits
253 // to the above metrics in a systematic way.
254 // Test for coldness before testing for hotness.
255 bool is_cold() const;
256 bool is_hot() const;
258 // Force a warm call to be hot. This worklists the call node for inlining.
259 void make_hot();
261 // Force a warm call to be cold. This worklists the call node for out-of-lining.
262 void make_cold();
264 // A reproducible total ordering, in which heat is the major key.
265 bool warmer_than(WarmCallInfo* that);
267 // List management. These methods are called with the list head,
268 // and return the new list head, inserting or removing the receiver.
269 WarmCallInfo* insert_into(WarmCallInfo* head);
270 WarmCallInfo* remove_from(WarmCallInfo* head);
272 #ifndef PRODUCT
273 void print() const;
274 void print_all() const;
275 int count_all() const;
276 #endif
277 };