Mercurial > hg > graal-compiler
view src/share/vm/shark/sharkNativeWrapper.cpp @ 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 | c7f3d0b4570f |
line wrap: on
line source
/* * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright 2009, 2010 Red Hat, Inc. * 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. * */ #include "precompiled.hpp" #include "shark/llvmHeaders.hpp" #include "shark/sharkNativeWrapper.hpp" #include "shark/sharkType.hpp" using namespace llvm; void SharkNativeWrapper::initialize(const char *name) { // Create the function _function = Function::Create( SharkType::entry_point_type(), GlobalVariable::InternalLinkage, name); // Get our arguments Function::arg_iterator ai = function()->arg_begin(); Argument *method = ai++; method->setName("method"); Argument *base_pc = ai++; base_pc->setName("base_pc"); code_buffer()->set_base_pc(base_pc); Argument *thread = ai++; thread->setName("thread"); set_thread(thread); // Create and push our stack frame builder()->SetInsertPoint(CreateBlock()); _stack = SharkStack::CreateBuildAndPushFrame(this, method); NOT_PRODUCT(method = NULL); // Create the oopmap. We use the one oopmap for every call site in // the wrapper, which results in the odd mild inefficiency but is a // damn sight easier to code. OopMap *oopmap = new OopMap( SharkStack::oopmap_slot_munge(stack()->oopmap_frame_size()), SharkStack::oopmap_slot_munge(arg_size())); oopmap->set_oop(SharkStack::slot2reg(stack()->method_slot_offset())); // Set up the oop_tmp slot if required: // - For static methods we use it to handlize the class argument // for the call, and to protect the same during slow path locks // (if synchronized). // - For methods returning oops, we use it to protect the return // value across safepoints or slow path unlocking. if (is_static() || is_returning_oop()) { _oop_tmp_slot = stack()->slot_addr( stack()->oop_tmp_slot_offset(), SharkType::oop_type(), "oop_tmp_slot"); oopmap->set_oop(SharkStack::slot2reg(stack()->oop_tmp_slot_offset())); } // Set up the monitor slot, for synchronized methods if (is_synchronized()) { Unimplemented(); _lock_slot_offset = 23; } // Start building the argument list std::vector<const Type*> param_types; std::vector<Value*> param_values; const PointerType *box_type = PointerType::getUnqual(SharkType::oop_type()); // First argument is the JNIEnv param_types.push_back(SharkType::jniEnv_type()); param_values.push_back( builder()->CreateAddressOfStructEntry( thread, JavaThread::jni_environment_offset(), SharkType::jniEnv_type(), "jni_environment")); // For static methods, the second argument is the class if (is_static()) { builder()->CreateStore( builder()->CreateInlineOop( JNIHandles::make_local( target()->method_holder()->klass_part()->java_mirror())), oop_tmp_slot()); param_types.push_back(box_type); param_values.push_back(oop_tmp_slot()); _receiver_slot_offset = stack()->oop_tmp_slot_offset(); } else if (is_returning_oop()) { // The oop_tmp slot is registered in the oopmap, // so we need to clear it. This is one of the // mild inefficiencies I mentioned earlier. builder()->CreateStore(LLVMValue::null(), oop_tmp_slot()); } // Parse the arguments for (int i = 0; i < arg_size(); i++) { int slot_offset = stack()->locals_slots_offset() + arg_size() - 1 - i; int adjusted_offset = slot_offset; BasicBlock *null, *not_null, *merge; Value *box; PHINode *phi; switch (arg_type(i)) { case T_VOID: break; case T_OBJECT: case T_ARRAY: null = CreateBlock("null"); not_null = CreateBlock("not_null"); merge = CreateBlock("merge"); box = stack()->slot_addr(slot_offset, SharkType::oop_type()); builder()->CreateCondBr( builder()->CreateICmp( ICmpInst::ICMP_EQ, builder()->CreateLoad(box), LLVMValue::null()), null, not_null); builder()->SetInsertPoint(null); builder()->CreateBr(merge); builder()->SetInsertPoint(not_null); builder()->CreateBr(merge); builder()->SetInsertPoint(merge); phi = builder()->CreatePHI(box_type, "boxed_object"); phi->addIncoming(ConstantPointerNull::get(box_type), null); phi->addIncoming(box, not_null); box = phi; param_types.push_back(box_type); param_values.push_back(box); oopmap->set_oop(SharkStack::slot2reg(slot_offset)); if (i == 0 && !is_static()) _receiver_slot_offset = slot_offset; break; case T_LONG: case T_DOUBLE: adjusted_offset--; // fall through default: const Type *param_type = SharkType::to_stackType(arg_type(i)); param_types.push_back(param_type); param_values.push_back( builder()->CreateLoad(stack()->slot_addr(adjusted_offset, param_type))); } } // The oopmap is now complete, and everything is written // into the frame except the PC. int pc_offset = code_buffer()->create_unique_offset(); _oop_maps = new OopMapSet(); oop_maps()->add_gc_map(pc_offset, oopmap); builder()->CreateStore( builder()->code_buffer_address(pc_offset), stack()->slot_addr(stack()->pc_slot_offset())); // Set up the Java frame anchor stack()->CreateSetLastJavaFrame(); // Lock if necessary if (is_synchronized()) Unimplemented(); // Change the thread state to _thread_in_native CreateSetThreadState(_thread_in_native); // Make the call BasicType result_type = target()->result_type(); const Type* return_type; if (result_type == T_VOID) return_type = SharkType::void_type(); else if (is_returning_oop()) return_type = box_type; else return_type = SharkType::to_arrayType(result_type); Value* native_function = builder()->CreateIntToPtr( LLVMValue::intptr_constant((intptr_t) target()->native_function()), PointerType::getUnqual( FunctionType::get(return_type, param_types, false))); Value *result = builder()->CreateCall( native_function, param_values.begin(), param_values.end()); // Start the transition back to _thread_in_Java CreateSetThreadState(_thread_in_native_trans); // Make sure new state is visible in the GC thread if (os::is_MP()) { if (UseMembar) builder()->CreateMemoryBarrier(SharkBuilder::BARRIER_STORELOAD); else CreateWriteMemorySerializePage(); } // Handle safepoint operations, pending suspend requests, // and pending asynchronous exceptions. BasicBlock *check_thread = CreateBlock("check_thread"); BasicBlock *do_safepoint = CreateBlock("do_safepoint"); BasicBlock *safepointed = CreateBlock("safepointed"); Value *global_state = builder()->CreateLoad( builder()->CreateIntToPtr( LLVMValue::intptr_constant( (intptr_t) SafepointSynchronize::address_of_state()), PointerType::getUnqual(SharkType::jint_type())), "global_state"); builder()->CreateCondBr( builder()->CreateICmpNE( global_state, LLVMValue::jint_constant(SafepointSynchronize::_not_synchronized)), do_safepoint, check_thread); builder()->SetInsertPoint(check_thread); Value *thread_state = builder()->CreateValueOfStructEntry( thread, JavaThread::suspend_flags_offset(), SharkType::jint_type(), "thread_state"); builder()->CreateCondBr( builder()->CreateICmpNE( thread_state, LLVMValue::jint_constant(0)), do_safepoint, safepointed); builder()->SetInsertPoint(do_safepoint); builder()->CreateCall( builder()->check_special_condition_for_native_trans(), thread); builder()->CreateBr(safepointed); // Finally we can change the thread state to _thread_in_Java builder()->SetInsertPoint(safepointed); CreateSetThreadState(_thread_in_Java); // Clear the frame anchor stack()->CreateResetLastJavaFrame(); // If there is a pending exception then we can just unwind and // return. It seems totally wrong that unlocking is skipped here // but apparently the template interpreter does this so we do too. BasicBlock *exception = CreateBlock("exception"); BasicBlock *no_exception = CreateBlock("no_exception"); builder()->CreateCondBr( builder()->CreateICmpEQ( CreateLoadPendingException(), LLVMValue::null()), no_exception, exception); builder()->SetInsertPoint(exception); CreateResetHandleBlock(); stack()->CreatePopFrame(0); builder()->CreateRet(LLVMValue::jint_constant(0)); builder()->SetInsertPoint(no_exception); // If the result was an oop then unbox it before // releasing the handle it might be protected by if (is_returning_oop()) { BasicBlock *null = builder()->GetInsertBlock(); BasicBlock *not_null = CreateBlock("not_null"); BasicBlock *merge = CreateBlock("merge"); builder()->CreateCondBr( builder()->CreateICmpNE(result, ConstantPointerNull::get(box_type)), not_null, merge); builder()->SetInsertPoint(not_null); Value *unboxed_result = builder()->CreateLoad(result); builder()->CreateBr(merge); builder()->SetInsertPoint(merge); PHINode *phi = builder()->CreatePHI(SharkType::oop_type(), "result"); phi->addIncoming(LLVMValue::null(), null); phi->addIncoming(unboxed_result, not_null); result = phi; } // Reset handle block CreateResetHandleBlock(); // Unlock if necessary. if (is_synchronized()) Unimplemented(); // Unwind and return Value *result_addr = stack()->CreatePopFrame(type2size[result_type]); if (result_type != T_VOID) { bool needs_cast = false; bool is_signed = false; switch (result_type) { case T_BOOLEAN: result = builder()->CreateICmpNE(result, LLVMValue::jbyte_constant(0)); needs_cast = true; break; case T_CHAR: needs_cast = true; break; case T_BYTE: case T_SHORT: needs_cast = true; is_signed = true; break; } if (needs_cast) { result = builder()->CreateIntCast( result, SharkType::to_stackType(result_type), is_signed); } builder()->CreateStore( result, builder()->CreateIntToPtr( result_addr, PointerType::getUnqual(SharkType::to_stackType(result_type)))); } builder()->CreateRet(LLVMValue::jint_constant(0)); }