Mercurial > hg > graal-compiler
view src/share/vm/memory/specialized_oop_closures.hpp @ 2149:7e37af9d69ef
7011379: G1: overly long concurrent marking cycles
Summary: This changeset introduces filtering of SATB buffers at the point when they are about to be enqueued. If this filtering clears enough entries on each buffer, the buffer can then be re-used and not enqueued. This cuts down the number of SATB buffers that need to be processed by the concurrent marking threads.
Reviewed-by: johnc, ysr
author | tonyp |
---|---|
date | Wed, 19 Jan 2011 09:35:17 -0500 |
parents | f95d63e2154a |
children | da91efe96a93 |
line wrap: on
line source
/* * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License version 2 only, as * published by the Free Software Foundation. * * This code is distributed in the hope that it will be useful, but WITHOUT * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License * version 2 for more details (a copy is included in the LICENSE file that * accompanied this code). * * You should have received a copy of the GNU General Public License version * 2 along with this work; if not, write to the Free Software Foundation, * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. * * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA * or visit www.oracle.com if you need additional information or have any * questions. * */ #ifndef SHARE_VM_MEMORY_SPECIALIZED_OOP_CLOSURES_HPP #define SHARE_VM_MEMORY_SPECIALIZED_OOP_CLOSURES_HPP #include "runtime/atomic.hpp" #ifndef SERIALGC #include "gc_implementation/g1/g1_specialized_oop_closures.hpp" #endif // The following OopClosure types get specialized versions of // "oop_oop_iterate" that invoke the closures' do_oop methods // non-virtually, using a mechanism defined in this file. Extend these // macros in the obvious way to add specializations for new closures. // Forward declarations. class OopClosure; class OopsInGenClosure; // DefNew class ScanClosure; class FastScanClosure; class FilteringClosure; // ParNew class ParScanWithBarrierClosure; class ParScanWithoutBarrierClosure; // CMS class MarkRefsIntoAndScanClosure; class Par_MarkRefsIntoAndScanClosure; class PushAndMarkClosure; class Par_PushAndMarkClosure; class PushOrMarkClosure; class Par_PushOrMarkClosure; class CMSKeepAliveClosure; class CMSInnerParMarkAndPushClosure; // This macro applies an argument macro to all OopClosures for which we // want specialized bodies of "oop_oop_iterate". The arguments to "f" are: // "f(closureType, non_virtual)" // where "closureType" is the name of the particular subclass of OopClosure, // and "non_virtual" will be the string "_nv" if the closure type should // have its "do_oop" method invoked non-virtually, or else the // string "_v". ("OopClosure" itself will be the only class in the latter // category.) // This is split into several because of a Visual C++ 6.0 compiler bug // where very long macros cause the compiler to crash // Some other heap might define further specialized closures. #ifndef FURTHER_SPECIALIZED_OOP_OOP_ITERATE_CLOSURES #define FURTHER_SPECIALIZED_OOP_OOP_ITERATE_CLOSURES(f) \ /* None */ #endif #define SPECIALIZED_OOP_OOP_ITERATE_CLOSURES_S(f) \ f(ScanClosure,_nv) \ f(FastScanClosure,_nv) \ f(FilteringClosure,_nv) #ifndef SERIALGC #define SPECIALIZED_OOP_OOP_ITERATE_CLOSURES_P(f) \ f(ParScanWithBarrierClosure,_nv) \ f(ParScanWithoutBarrierClosure,_nv) #else // SERIALGC #define SPECIALIZED_OOP_OOP_ITERATE_CLOSURES_P(f) #endif // SERIALGC #define SPECIALIZED_OOP_OOP_ITERATE_CLOSURES_1(f) \ SPECIALIZED_OOP_OOP_ITERATE_CLOSURES_S(f) \ SPECIALIZED_OOP_OOP_ITERATE_CLOSURES_P(f) #ifndef SERIALGC #define SPECIALIZED_OOP_OOP_ITERATE_CLOSURES_2(f) \ f(MarkRefsIntoAndScanClosure,_nv) \ f(Par_MarkRefsIntoAndScanClosure,_nv) \ f(PushAndMarkClosure,_nv) \ f(Par_PushAndMarkClosure,_nv) \ f(PushOrMarkClosure,_nv) \ f(Par_PushOrMarkClosure,_nv) \ f(CMSKeepAliveClosure,_nv) \ f(CMSInnerParMarkAndPushClosure,_nv) \ FURTHER_SPECIALIZED_OOP_OOP_ITERATE_CLOSURES(f) #else // SERIALGC #define SPECIALIZED_OOP_OOP_ITERATE_CLOSURES_2(f) #endif // SERIALGC // We separate these out, because sometime the general one has // a different definition from the specialized ones, and sometimes it // doesn't. #define ALL_OOP_OOP_ITERATE_CLOSURES_1(f) \ f(OopClosure,_v) \ SPECIALIZED_OOP_OOP_ITERATE_CLOSURES_1(f) #define ALL_OOP_OOP_ITERATE_CLOSURES_2(f) \ SPECIALIZED_OOP_OOP_ITERATE_CLOSURES_2(f) #ifndef SERIALGC // This macro applies an argument macro to all OopClosures for which we // want specialized bodies of a family of methods related to // "par_oop_iterate". The arguments to f are the same as above. // The "root_class" is the most general class to define; this may be // "OopClosure" in some applications and "OopsInGenClosure" in others. #define SPECIALIZED_PAR_OOP_ITERATE_CLOSURES(f) \ f(MarkRefsIntoAndScanClosure,_nv) \ f(PushAndMarkClosure,_nv) \ f(Par_MarkRefsIntoAndScanClosure,_nv) \ f(Par_PushAndMarkClosure,_nv) #define ALL_PAR_OOP_ITERATE_CLOSURES(f) \ f(OopClosure,_v) \ SPECIALIZED_PAR_OOP_ITERATE_CLOSURES(f) #endif // SERIALGC // This macro applies an argument macro to all OopClosures for which we // want specialized bodies of a family of methods related to // "oops_since_save_marks_do". The arguments to f are the same as above. // The "root_class" is the most general class to define; this may be // "OopClosure" in some applications and "OopsInGenClosure" in others. // Some other heap might define further specialized closures. #ifndef FURTHER_SPECIALIZED_SINCE_SAVE_MARKS_CLOSURES #define FURTHER_SPECIALIZED_SINCE_SAVE_MARKS_CLOSURES(f) \ /* None */ #endif #define SPECIALIZED_SINCE_SAVE_MARKS_CLOSURES_YOUNG_S(f) \ f(ScanClosure,_nv) \ f(FastScanClosure,_nv) #ifndef SERIALGC #define SPECIALIZED_SINCE_SAVE_MARKS_CLOSURES_YOUNG_P(f) \ f(ParScanWithBarrierClosure,_nv) \ f(ParScanWithoutBarrierClosure,_nv) \ FURTHER_SPECIALIZED_SINCE_SAVE_MARKS_CLOSURES(f) #else // SERIALGC #define SPECIALIZED_SINCE_SAVE_MARKS_CLOSURES_YOUNG_P(f) #endif // SERIALGC #define SPECIALIZED_SINCE_SAVE_MARKS_CLOSURES_YOUNG(f) \ SPECIALIZED_SINCE_SAVE_MARKS_CLOSURES_YOUNG_S(f) \ SPECIALIZED_SINCE_SAVE_MARKS_CLOSURES_YOUNG_P(f) #define SPECIALIZED_SINCE_SAVE_MARKS_CLOSURES(f) \ SPECIALIZED_SINCE_SAVE_MARKS_CLOSURES_YOUNG(f) // We separate these out, because sometime the general one has // a different definition from the specialized ones, and sometimes it // doesn't. // NOTE: One of the valid criticisms of this // specialize-oop_oop_iterate-for-specific-closures idiom is that it is // easy to have a silent performance bug: if you fail to de-virtualize, // things still work, just slower. The "SpecializationStats" mode is // intended to at least make such a failure easy to detect. // *Not* using the ALL_SINCE_SAVE_MARKS_CLOSURES(f) macro defined // below means that *only* closures for which oop_oop_iterate specializations // exist above may be applied to "oops_since_save_marks". That is, // this form of the performance bug is caught statically. When you add // a definition for the general type, this property goes away. // Make sure you test with SpecializationStats to find such bugs // when introducing a new closure where you don't want virtual dispatch. #define ALL_SINCE_SAVE_MARKS_CLOSURES(f) \ f(OopsInGenClosure,_v) \ SPECIALIZED_SINCE_SAVE_MARKS_CLOSURES(f) // For keeping stats on effectiveness. #define ENABLE_SPECIALIZATION_STATS 0 class SpecializationStats { public: enum Kind { ik, // instanceKlass irk, // instanceRefKlass oa, // objArrayKlass NUM_Kinds }; #if ENABLE_SPECIALIZATION_STATS private: static bool _init; static bool _wrapped; static jint _numCallsAll; static jint _numCallsTotal[NUM_Kinds]; static jint _numCalls_nv[NUM_Kinds]; static jint _numDoOopCallsTotal[NUM_Kinds]; static jint _numDoOopCalls_nv[NUM_Kinds]; public: #endif static void clear() PRODUCT_RETURN; static inline void record_call() PRODUCT_RETURN; static inline void record_iterate_call_v(Kind k) PRODUCT_RETURN; static inline void record_iterate_call_nv(Kind k) PRODUCT_RETURN; static inline void record_do_oop_call_v(Kind k) PRODUCT_RETURN; static inline void record_do_oop_call_nv(Kind k) PRODUCT_RETURN; static void print() PRODUCT_RETURN; }; #ifndef PRODUCT #if ENABLE_SPECIALIZATION_STATS inline void SpecializationStats::record_call() { Atomic::inc(&_numCallsAll); } inline void SpecializationStats::record_iterate_call_v(Kind k) { Atomic::inc(&_numCallsTotal[k]); } inline void SpecializationStats::record_iterate_call_nv(Kind k) { Atomic::inc(&_numCallsTotal[k]); Atomic::inc(&_numCalls_nv[k]); } inline void SpecializationStats::record_do_oop_call_v(Kind k) { Atomic::inc(&_numDoOopCallsTotal[k]); } inline void SpecializationStats::record_do_oop_call_nv(Kind k) { Atomic::inc(&_numDoOopCallsTotal[k]); Atomic::inc(&_numDoOopCalls_nv[k]); } #else // !ENABLE_SPECIALIZATION_STATS inline void SpecializationStats::record_call() {} inline void SpecializationStats::record_iterate_call_v(Kind k) {} inline void SpecializationStats::record_iterate_call_nv(Kind k) {} inline void SpecializationStats::record_do_oop_call_v(Kind k) {} inline void SpecializationStats::record_do_oop_call_nv(Kind k) {} inline void SpecializationStats::clear() {} inline void SpecializationStats::print() {} #endif // ENABLE_SPECIALIZATION_STATS #endif // !PRODUCT #endif // SHARE_VM_MEMORY_SPECIALIZED_OOP_CLOSURES_HPP