Tue, 10 Jun 2014 12:28:06 -0700
8035605: Expand functionality of PredictedIntrinsicGenerator
Summary: Allow several predicates and separate intrinsic methods per one intrinsified (virtual) method.
Reviewed-by: roland
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 */
25 #ifndef SHARE_VM_OPTO_CALLGENERATOR_HPP
26 #define SHARE_VM_OPTO_CALLGENERATOR_HPP
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"
34 class Parse;
36 //---------------------------CallGenerator-------------------------------------
37 // The subclasses of this class handle generation of ideal nodes for
38 // call sites and method entry points.
40 class CallGenerator : public ResourceObj {
41 public:
42 enum {
43 xxxunusedxxx
44 };
46 private:
47 ciMethod* _method; // The method being called.
49 protected:
50 CallGenerator(ciMethod* method) : _method(method) {}
52 public:
53 // Accessors
54 ciMethod* method() const { return _method; }
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_predicated: Uses an explicit check (predicate).
67 virtual bool is_predicated() const { return false; }
68 virtual int predicates_count() const { return 0; }
69 // is_trap: Does not return to the caller. (E.g., uncommon trap.)
70 virtual bool is_trap() const { return false; }
71 // does_virtual_dispatch: Should try inlining as normal method first.
72 virtual bool does_virtual_dispatch() const { return false; }
74 // is_late_inline: supports conversion of call into an inline
75 virtual bool is_late_inline() const { return false; }
76 // same but for method handle calls
77 virtual bool is_mh_late_inline() const { return false; }
78 virtual bool is_string_late_inline() const{ return false; }
80 // for method handle calls: have we tried inlinining the call already?
81 virtual bool already_attempted() const { ShouldNotReachHere(); return false; }
83 // Replace the call with an inline version of the code
84 virtual void do_late_inline() { ShouldNotReachHere(); }
86 virtual CallStaticJavaNode* call_node() const { ShouldNotReachHere(); return NULL; }
88 // Note: It is possible for a CG to be both inline and virtual.
89 // (The hashCode intrinsic does a vtable check and an inlined fast path.)
91 // Utilities:
92 const TypeFunc* tf() const;
94 // The given jvms has state and arguments for a call to my method.
95 // Edges after jvms->argoff() carry all (pre-popped) argument values.
96 //
97 // Update the map with state and return values (if any) and return it.
98 // The return values (0, 1, or 2) must be pushed on the map's stack,
99 // and the sp of the jvms incremented accordingly.
100 //
101 // The jvms is returned on success. Alternatively, a copy of the
102 // given jvms, suitably updated, may be returned, in which case the
103 // caller should discard the original jvms.
104 //
105 // The non-Parm edges of the returned map will contain updated global state,
106 // and one or two edges before jvms->sp() will carry any return values.
107 // Other map edges may contain locals or monitors, and should not
108 // be changed in meaning.
109 //
110 // If the call traps, the returned map must have a control edge of top.
111 // If the call can throw, the returned map must report has_exceptions().
112 //
113 // If the result is NULL, it means that this CallGenerator was unable
114 // to handle the given call, and another CallGenerator should be consulted.
115 virtual JVMState* generate(JVMState* jvms, Parse* parent_parser) = 0;
117 // How to generate a call site that is inlined:
118 static CallGenerator* for_inline(ciMethod* m, float expected_uses = -1);
119 // How to generate code for an on-stack replacement handler.
120 static CallGenerator* for_osr(ciMethod* m, int osr_bci);
122 // How to generate vanilla out-of-line call sites:
123 static CallGenerator* for_direct_call(ciMethod* m, bool separate_io_projs = false); // static, special
124 static CallGenerator* for_virtual_call(ciMethod* m, int vtable_index); // virtual, interface
125 static CallGenerator* for_dynamic_call(ciMethod* m); // invokedynamic
127 static CallGenerator* for_method_handle_call( JVMState* jvms, ciMethod* caller, ciMethod* callee, bool delayed_forbidden);
128 static CallGenerator* for_method_handle_inline(JVMState* jvms, ciMethod* caller, ciMethod* callee, bool& input_not_const);
130 // How to generate a replace a direct call with an inline version
131 static CallGenerator* for_late_inline(ciMethod* m, CallGenerator* inline_cg);
132 static CallGenerator* for_mh_late_inline(ciMethod* caller, ciMethod* callee, bool input_not_const);
133 static CallGenerator* for_string_late_inline(ciMethod* m, CallGenerator* inline_cg);
134 static CallGenerator* for_boxing_late_inline(ciMethod* m, CallGenerator* inline_cg);
136 // How to make a call but defer the decision whether to inline or not.
137 static CallGenerator* for_warm_call(WarmCallInfo* ci,
138 CallGenerator* if_cold,
139 CallGenerator* if_hot);
141 // How to make a call that optimistically assumes a receiver type:
142 static CallGenerator* for_predicted_call(ciKlass* predicted_receiver,
143 CallGenerator* if_missed,
144 CallGenerator* if_hit,
145 float hit_prob);
147 // How to make a call that optimistically assumes a MethodHandle target:
148 static CallGenerator* for_predicted_dynamic_call(ciMethodHandle* predicted_method_handle,
149 CallGenerator* if_missed,
150 CallGenerator* if_hit,
151 float hit_prob);
153 // How to make a call that gives up and goes back to the interpreter:
154 static CallGenerator* for_uncommon_trap(ciMethod* m,
155 Deoptimization::DeoptReason reason,
156 Deoptimization::DeoptAction action);
158 // Registry for intrinsics:
159 static CallGenerator* for_intrinsic(ciMethod* m);
160 static void register_intrinsic(ciMethod* m, CallGenerator* cg);
161 static CallGenerator* for_predicated_intrinsic(CallGenerator* intrinsic,
162 CallGenerator* cg);
163 virtual Node* generate_predicate(JVMState* jvms, int predicate) { return NULL; };
165 virtual void print_inlining_late(const char* msg) { ShouldNotReachHere(); }
167 static void print_inlining(Compile* C, ciMethod* callee, int inline_level, int bci, const char* msg) {
168 if (C->print_inlining()) {
169 C->print_inlining(callee, inline_level, bci, msg);
170 }
171 }
172 };
175 //------------------------InlineCallGenerator----------------------------------
176 class InlineCallGenerator : public CallGenerator {
177 protected:
178 InlineCallGenerator(ciMethod* method) : CallGenerator(method) {}
180 public:
181 virtual bool is_inline() const { return true; }
182 };
185 //---------------------------WarmCallInfo--------------------------------------
186 // A struct to collect information about a given call site.
187 // Helps sort call sites into "hot", "medium", and "cold".
188 // Participates in the queueing of "medium" call sites for possible inlining.
189 class WarmCallInfo : public ResourceObj {
190 private:
192 CallNode* _call; // The CallNode which may be inlined.
193 CallGenerator* _hot_cg;// CG for expanding the call node
195 // These are the metrics we use to evaluate call sites:
197 float _count; // How often do we expect to reach this site?
198 float _profit; // How much time do we expect to save by inlining?
199 float _work; // How long do we expect the average call to take?
200 float _size; // How big do we expect the inlined code to be?
202 float _heat; // Combined score inducing total order on call sites.
203 WarmCallInfo* _next; // Next cooler call info in pending queue.
205 // Count is the number of times this call site is expected to be executed.
206 // Large count is favorable for inlining, because the extra compilation
207 // work will be amortized more completely.
209 // Profit is a rough measure of the amount of time we expect to save
210 // per execution of this site if we inline it. (1.0 == call overhead)
211 // Large profit favors inlining. Negative profit disables inlining.
213 // Work is a rough measure of the amount of time a typical out-of-line
214 // call from this site is expected to take. (1.0 == call, no-op, return)
215 // Small work is somewhat favorable for inlining, since methods with
216 // short "hot" traces are more likely to inline smoothly.
218 // Size is the number of graph nodes we expect this method to produce,
219 // not counting the inlining of any further warm calls it may include.
220 // Small size favors inlining, since small methods are more likely to
221 // inline smoothly. The size is estimated by examining the native code
222 // if available. The method bytecodes are also examined, assuming
223 // empirically observed node counts for each kind of bytecode.
225 // Heat is the combined "goodness" of a site's inlining. If we were
226 // omniscient, it would be the difference of two sums of future execution
227 // times of code emitted for this site (amortized across multiple sites if
228 // sharing applies). The two sums are for versions of this call site with
229 // and without inlining.
231 // We approximate this mythical quantity by playing with averages,
232 // rough estimates, and assumptions that history repeats itself.
233 // The basic formula count * profit is heuristically adjusted
234 // by looking at the expected compilation and execution times of
235 // of the inlined call.
237 // Note: Some of these metrics may not be present in the final product,
238 // but exist in development builds to experiment with inline policy tuning.
240 // This heuristic framework does not model well the very significant
241 // effects of multiple-level inlining. It is possible to see no immediate
242 // profit from inlining X->Y, but to get great profit from a subsequent
243 // inlining X->Y->Z.
245 // This framework does not take well into account the problem of N**2 code
246 // size in a clique of mutually inlinable methods.
248 WarmCallInfo* next() const { return _next; }
249 void set_next(WarmCallInfo* n) { _next = n; }
251 static WarmCallInfo _always_hot;
252 static WarmCallInfo _always_cold;
254 // Constructor intitialization of always_hot and always_cold
255 WarmCallInfo(float c, float p, float w, float s) {
256 _call = NULL;
257 _hot_cg = NULL;
258 _next = NULL;
259 _count = c;
260 _profit = p;
261 _work = w;
262 _size = s;
263 _heat = 0;
264 }
266 public:
267 // Because WarmInfo objects live over the entire lifetime of the
268 // Compile object, they are allocated into the comp_arena, which
269 // does not get resource marked or reset during the compile process
270 void *operator new( size_t x, Compile* C ) throw() { return C->comp_arena()->Amalloc(x); }
271 void operator delete( void * ) { } // fast deallocation
273 static WarmCallInfo* always_hot();
274 static WarmCallInfo* always_cold();
276 WarmCallInfo() {
277 _call = NULL;
278 _hot_cg = NULL;
279 _next = NULL;
280 _count = _profit = _work = _size = _heat = 0;
281 }
283 CallNode* call() const { return _call; }
284 float count() const { return _count; }
285 float size() const { return _size; }
286 float work() const { return _work; }
287 float profit() const { return _profit; }
288 float heat() const { return _heat; }
290 void set_count(float x) { _count = x; }
291 void set_size(float x) { _size = x; }
292 void set_work(float x) { _work = x; }
293 void set_profit(float x) { _profit = x; }
294 void set_heat(float x) { _heat = x; }
296 // Load initial heuristics from profiles, etc.
297 // The heuristics can be tweaked further by the caller.
298 void init(JVMState* call_site, ciMethod* call_method, ciCallProfile& profile, float prof_factor);
300 static float MAX_VALUE() { return +1.0e10; }
301 static float MIN_VALUE() { return -1.0e10; }
303 float compute_heat() const;
305 void set_call(CallNode* call) { _call = call; }
306 void set_hot_cg(CallGenerator* cg) { _hot_cg = cg; }
308 // Do not queue very hot or very cold calls.
309 // Make very cold ones out of line immediately.
310 // Inline very hot ones immediately.
311 // These queries apply various tunable limits
312 // to the above metrics in a systematic way.
313 // Test for coldness before testing for hotness.
314 bool is_cold() const;
315 bool is_hot() const;
317 // Force a warm call to be hot. This worklists the call node for inlining.
318 void make_hot();
320 // Force a warm call to be cold. This worklists the call node for out-of-lining.
321 void make_cold();
323 // A reproducible total ordering, in which heat is the major key.
324 bool warmer_than(WarmCallInfo* that);
326 // List management. These methods are called with the list head,
327 // and return the new list head, inserting or removing the receiver.
328 WarmCallInfo* insert_into(WarmCallInfo* head);
329 WarmCallInfo* remove_from(WarmCallInfo* head);
331 #ifndef PRODUCT
332 void print() const;
333 void print_all() const;
334 int count_all() const;
335 #endif
336 };
338 #endif // SHARE_VM_OPTO_CALLGENERATOR_HPP