Mon, 26 Sep 2011 10:24:05 -0700
7081933: Use zeroing elimination optimization for large array
Summary: Don't zero new typeArray during runtime call if the allocation is followed by arraycopy into it.
Reviewed-by: twisti
1 /*
2 * Copyright (c) 2000, 2011, 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 "opto/callnode.hpp"
29 #include "opto/compile.hpp"
30 #include "opto/type.hpp"
31 #include "runtime/deoptimization.hpp"
33 //---------------------------CallGenerator-------------------------------------
34 // The subclasses of this class handle generation of ideal nodes for
35 // call sites and method entry points.
37 class CallGenerator : public ResourceObj {
38 public:
39 enum {
40 xxxunusedxxx
41 };
43 private:
44 ciMethod* _method; // The method being called.
46 protected:
47 CallGenerator(ciMethod* method);
49 public:
50 // Accessors
51 ciMethod* method() const { return _method; }
53 // is_inline: At least some code implementing the method is copied here.
54 virtual bool is_inline() const { return false; }
55 // is_intrinsic: There's a method-specific way of generating the inline code.
56 virtual bool is_intrinsic() const { return false; }
57 // is_parse: Bytecodes implementing the specific method are copied here.
58 virtual bool is_parse() const { return false; }
59 // is_virtual: The call uses the receiver type to select or check the method.
60 virtual bool is_virtual() const { return false; }
61 // is_deferred: The decision whether to inline or not is deferred.
62 virtual bool is_deferred() const { return false; }
63 // is_predicted: Uses an explicit check against a predicted type.
64 virtual bool is_predicted() const { return false; }
65 // is_trap: Does not return to the caller. (E.g., uncommon trap.)
66 virtual bool is_trap() const { return false; }
68 // is_late_inline: supports conversion of call into an inline
69 virtual bool is_late_inline() const { return false; }
70 // Replace the call with an inline version of the code
71 virtual void do_late_inline() { ShouldNotReachHere(); }
73 virtual CallStaticJavaNode* call_node() const { ShouldNotReachHere(); return NULL; }
75 // Note: It is possible for a CG to be both inline and virtual.
76 // (The hashCode intrinsic does a vtable check and an inlined fast path.)
78 // Utilities:
79 const TypeFunc* tf() const;
81 // The given jvms has state and arguments for a call to my method.
82 // Edges after jvms->argoff() carry all (pre-popped) argument values.
83 //
84 // Update the map with state and return values (if any) and return it.
85 // The return values (0, 1, or 2) must be pushed on the map's stack,
86 // and the sp of the jvms incremented accordingly.
87 //
88 // The jvms is returned on success. Alternatively, a copy of the
89 // given jvms, suitably updated, may be returned, in which case the
90 // caller should discard the original jvms.
91 //
92 // The non-Parm edges of the returned map will contain updated global state,
93 // and one or two edges before jvms->sp() will carry any return values.
94 // Other map edges may contain locals or monitors, and should not
95 // be changed in meaning.
96 //
97 // If the call traps, the returned map must have a control edge of top.
98 // If the call can throw, the returned map must report has_exceptions().
99 //
100 // If the result is NULL, it means that this CallGenerator was unable
101 // to handle the given call, and another CallGenerator should be consulted.
102 virtual JVMState* generate(JVMState* jvms) = 0;
104 // How to generate a call site that is inlined:
105 static CallGenerator* for_inline(ciMethod* m, float expected_uses = -1);
106 // How to generate code for an on-stack replacement handler.
107 static CallGenerator* for_osr(ciMethod* m, int osr_bci);
109 // How to generate vanilla out-of-line call sites:
110 static CallGenerator* for_direct_call(ciMethod* m, bool separate_io_projs = false); // static, special
111 static CallGenerator* for_dynamic_call(ciMethod* m); // invokedynamic
112 static CallGenerator* for_virtual_call(ciMethod* m, int vtable_index); // virtual, interface
114 static CallGenerator* for_method_handle_inline(Node* method_handle, JVMState* jvms, ciMethod* caller, ciMethod* callee, ciCallProfile profile);
115 static CallGenerator* for_invokedynamic_inline(ciCallSite* call_site, JVMState* jvms, ciMethod* caller, ciMethod* callee, ciCallProfile profile);
117 // How to generate a replace a direct call with an inline version
118 static CallGenerator* for_late_inline(ciMethod* m, CallGenerator* inline_cg);
120 // How to make a call but defer the decision whether to inline or not.
121 static CallGenerator* for_warm_call(WarmCallInfo* ci,
122 CallGenerator* if_cold,
123 CallGenerator* if_hot);
125 // How to make a call that optimistically assumes a receiver type:
126 static CallGenerator* for_predicted_call(ciKlass* predicted_receiver,
127 CallGenerator* if_missed,
128 CallGenerator* if_hit,
129 float hit_prob);
131 // How to make a call that optimistically assumes a MethodHandle target:
132 static CallGenerator* for_predicted_dynamic_call(ciMethodHandle* predicted_method_handle,
133 CallGenerator* if_missed,
134 CallGenerator* if_hit,
135 float hit_prob);
137 // How to make a call that gives up and goes back to the interpreter:
138 static CallGenerator* for_uncommon_trap(ciMethod* m,
139 Deoptimization::DeoptReason reason,
140 Deoptimization::DeoptAction action);
142 // Registry for intrinsics:
143 static CallGenerator* for_intrinsic(ciMethod* m);
144 static void register_intrinsic(ciMethod* m, CallGenerator* cg);
145 };
147 class InlineCallGenerator : public CallGenerator {
148 virtual bool is_inline() const { return true; }
150 protected:
151 InlineCallGenerator(ciMethod* method) : CallGenerator(method) { }
152 };
155 //---------------------------WarmCallInfo--------------------------------------
156 // A struct to collect information about a given call site.
157 // Helps sort call sites into "hot", "medium", and "cold".
158 // Participates in the queueing of "medium" call sites for possible inlining.
159 class WarmCallInfo : public ResourceObj {
160 private:
162 CallNode* _call; // The CallNode which may be inlined.
163 CallGenerator* _hot_cg;// CG for expanding the call node
165 // These are the metrics we use to evaluate call sites:
167 float _count; // How often do we expect to reach this site?
168 float _profit; // How much time do we expect to save by inlining?
169 float _work; // How long do we expect the average call to take?
170 float _size; // How big do we expect the inlined code to be?
172 float _heat; // Combined score inducing total order on call sites.
173 WarmCallInfo* _next; // Next cooler call info in pending queue.
175 // Count is the number of times this call site is expected to be executed.
176 // Large count is favorable for inlining, because the extra compilation
177 // work will be amortized more completely.
179 // Profit is a rough measure of the amount of time we expect to save
180 // per execution of this site if we inline it. (1.0 == call overhead)
181 // Large profit favors inlining. Negative profit disables inlining.
183 // Work is a rough measure of the amount of time a typical out-of-line
184 // call from this site is expected to take. (1.0 == call, no-op, return)
185 // Small work is somewhat favorable for inlining, since methods with
186 // short "hot" traces are more likely to inline smoothly.
188 // Size is the number of graph nodes we expect this method to produce,
189 // not counting the inlining of any further warm calls it may include.
190 // Small size favors inlining, since small methods are more likely to
191 // inline smoothly. The size is estimated by examining the native code
192 // if available. The method bytecodes are also examined, assuming
193 // empirically observed node counts for each kind of bytecode.
195 // Heat is the combined "goodness" of a site's inlining. If we were
196 // omniscient, it would be the difference of two sums of future execution
197 // times of code emitted for this site (amortized across multiple sites if
198 // sharing applies). The two sums are for versions of this call site with
199 // and without inlining.
201 // We approximate this mythical quantity by playing with averages,
202 // rough estimates, and assumptions that history repeats itself.
203 // The basic formula count * profit is heuristically adjusted
204 // by looking at the expected compilation and execution times of
205 // of the inlined call.
207 // Note: Some of these metrics may not be present in the final product,
208 // but exist in development builds to experiment with inline policy tuning.
210 // This heuristic framework does not model well the very significant
211 // effects of multiple-level inlining. It is possible to see no immediate
212 // profit from inlining X->Y, but to get great profit from a subsequent
213 // inlining X->Y->Z.
215 // This framework does not take well into account the problem of N**2 code
216 // size in a clique of mutually inlinable methods.
218 WarmCallInfo* next() const { return _next; }
219 void set_next(WarmCallInfo* n) { _next = n; }
221 static WarmCallInfo _always_hot;
222 static WarmCallInfo _always_cold;
224 // Constructor intitialization of always_hot and always_cold
225 WarmCallInfo(float c, float p, float w, float s) {
226 _call = NULL;
227 _hot_cg = NULL;
228 _next = NULL;
229 _count = c;
230 _profit = p;
231 _work = w;
232 _size = s;
233 _heat = 0;
234 }
236 public:
237 // Because WarmInfo objects live over the entire lifetime of the
238 // Compile object, they are allocated into the comp_arena, which
239 // does not get resource marked or reset during the compile process
240 void *operator new( size_t x, Compile* C ) { return C->comp_arena()->Amalloc(x); }
241 void operator delete( void * ) { } // fast deallocation
243 static WarmCallInfo* always_hot();
244 static WarmCallInfo* always_cold();
246 WarmCallInfo() {
247 _call = NULL;
248 _hot_cg = NULL;
249 _next = NULL;
250 _count = _profit = _work = _size = _heat = 0;
251 }
253 CallNode* call() const { return _call; }
254 float count() const { return _count; }
255 float size() const { return _size; }
256 float work() const { return _work; }
257 float profit() const { return _profit; }
258 float heat() const { return _heat; }
260 void set_count(float x) { _count = x; }
261 void set_size(float x) { _size = x; }
262 void set_work(float x) { _work = x; }
263 void set_profit(float x) { _profit = x; }
264 void set_heat(float x) { _heat = x; }
266 // Load initial heuristics from profiles, etc.
267 // The heuristics can be tweaked further by the caller.
268 void init(JVMState* call_site, ciMethod* call_method, ciCallProfile& profile, float prof_factor);
270 static float MAX_VALUE() { return +1.0e10; }
271 static float MIN_VALUE() { return -1.0e10; }
273 float compute_heat() const;
275 void set_call(CallNode* call) { _call = call; }
276 void set_hot_cg(CallGenerator* cg) { _hot_cg = cg; }
278 // Do not queue very hot or very cold calls.
279 // Make very cold ones out of line immediately.
280 // Inline very hot ones immediately.
281 // These queries apply various tunable limits
282 // to the above metrics in a systematic way.
283 // Test for coldness before testing for hotness.
284 bool is_cold() const;
285 bool is_hot() const;
287 // Force a warm call to be hot. This worklists the call node for inlining.
288 void make_hot();
290 // Force a warm call to be cold. This worklists the call node for out-of-lining.
291 void make_cold();
293 // A reproducible total ordering, in which heat is the major key.
294 bool warmer_than(WarmCallInfo* that);
296 // List management. These methods are called with the list head,
297 // and return the new list head, inserting or removing the receiver.
298 WarmCallInfo* insert_into(WarmCallInfo* head);
299 WarmCallInfo* remove_from(WarmCallInfo* head);
301 #ifndef PRODUCT
302 void print() const;
303 void print_all() const;
304 int count_all() const;
305 #endif
306 };
308 #endif // SHARE_VM_OPTO_CALLGENERATOR_HPP