run barrier callback in BarrierPhyInstrOperand::~BarrierPhyInstrOperand #7702
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -26,6 +26,8 @@ limitations under the License. | |
| #include "oneflow/core/profiler/profiler.h" | ||
| #include "oneflow/core/common/cpp_attribute.h" | ||
| #include "oneflow/core/common/global.h" | ||
| #include "oneflow/core/common/foreign_lock_helper.h" | ||
| #include <typeinfo> | ||
|
|
||
| namespace oneflow { | ||
| namespace vm { | ||
|
|
@@ -189,19 +191,21 @@ void VirtualMachineEngine::ReleaseFinishedInstructions() { | |
| // in stream->DeleteInstruction(...) | ||
| intrusive::shared_ptr<InstructionMsg> instr_msg(instruction_ptr->mut_instr_msg()); | ||
| stream->DeleteInstruction(LivelyInstructionListErase(instruction_ptr)); | ||
| static constexpr int kFlushWindowSize = 32; | ||
| MoveInstructionMsgToGarbageMsgList(kFlushWindowSize, std::move(instr_msg)); | ||
| } | ||
| if (stream->running_instruction_list().empty()) { mut_active_stream_list()->Erase(stream); } | ||
| } | ||
| } | ||
|
|
||
| void VirtualMachineEngine::MoveInstructionMsgToGarbageMsgList( | ||
| int flush_window_size, intrusive::shared_ptr<InstructionMsg>&& instr_msg) { | ||
| local_garbage_msg_list_.EmplaceBack(std::move(instr_msg)); | ||
| // local_garbage_msg_list_ is the cache of garbage_msg_list_. | ||
| // `kWindowSize` controls the frequency of the usage of mutexed list. | ||
| if (unlikely(local_garbage_msg_list_.size() > flush_window_size)) { | ||
| MoveToGarbageMsgListAndNotifyGC(); | ||
| } | ||
| } | ||
|
|
||
| void VirtualMachineEngine::MoveToGarbageMsgListAndNotifyGC() { | ||
|
|
@@ -495,7 +499,11 @@ void VirtualMachineEngine::TryRunBarrierInstruction() { | |
| CHECK(OnSchedulerThread(stream_type)); | ||
| stream_type.Run(sequnential_instruction); | ||
| mut_barrier_instruction_list()->Erase(sequnential_instruction); | ||
| intrusive::shared_ptr<InstructionMsg> instr_msg(sequnential_instruction->mut_instr_msg()); | ||
| LivelyInstructionListErase(sequnential_instruction); | ||
| sequnential_instruction->clear_instr_msg(); | ||
| constexpr int kZeroWindowSize = 0; // flush immediately. | ||
| MoveInstructionMsgToGarbageMsgList(kZeroWindowSize, std::move(instr_msg)); | ||
| OF_PROFILER_RANGE_POP(); | ||
| } | ||
|
|
||
|
|
@@ -534,7 +542,18 @@ void VirtualMachineEngine::Schedule() { | |
| void VirtualMachineEngine::Callback() { | ||
| InstructionMsgList garbage_msg_list; | ||
| mut_garbage_msg_list()->MoveTo(&garbage_msg_list); | ||
| INTRUSIVE_FOR_EACH(garbage, &garbage_msg_list) { | ||
| CHECK_JUST(Global<ForeignLockHelper>::Get()->WithScopedAcquire([&, this]() -> Maybe<void> { | ||
| garbage_msg_list.Erase(garbage.Mutable()); | ||
| while (garbage->ref_cnt() > 1) { | ||
| // Do nothing. wait until all other threads ref_cnts released. | ||
| } | ||
|
There was a problem hiding this comment. call back线程busy wait其它线程释放指令，以确认该指令在其它所有线程使用完成了 There was a problem hiding this comment. main线程的sync会等待这个CallBack执行完成？ There was a problem hiding this comment. 这个部分是需要main线程代码自行用bc来处理。 There was a problem hiding this comment. 这里之前不太熟悉，尝试整理一下。 1、当调用Sync时，被视为是BarrierInstruction Maybe<void> ClusterSync() {
auto bc = std::make_shared<BlockingCounter>(1);
JUST(PhysicalRun([bc](InstructionsBuilder* builder) -> Maybe<void> {
JUST(builder->ComputeGlobalFrontSeqBarrier()); // 是FrontSeq的指令，会被TryRunBarrierInstruction执行
JUST(builder->ComputeRankFrontSeqCallback([bc]() { bc->Decrease(); })); // 指令callback执行时，就bc减1
return Maybe<void>::Ok();
}));
JUST(bc->WaitUntilCntEqualZero(VirtualMachine::GetPredicatorNoMoreInstructionsFinished())); // 主线程等待bc为0，效果是阻塞主线程等待BarrierInstruction的callback执行
return Maybe<void>::Ok();
}2、TryRunBarrierInstruction时，立即做指令的gc，触发BarrierInstruction及之前的所有指令的callback的执行 constexpr int kZeroWindowSize = 0; // flush immediately.
MoveInstructionMsgToGarbageMsgList(kZeroWindowSize, std::move(instr_msg));3、callback线程执行gc得到的所有指令的callback void VirtualMachine::CallbackLoop(const std::function<void()>& Initializer) {
Initializer();
auto* vm = mut_vm();
while (callback_notifier_.WaitAndClearNotifiedCnt() == kNotifierStatusSuccess) { vm->Callback(); }
}执行有两个特点：
这样gc中的callback执行完，代表BarrierInstruction及之前的指令都执行完了，bc置0，main线程继续。 There was a problem hiding this comment. 这里的 busy wait 会有性能问题吗？ cpu 占用率过高等问题 There was a problem hiding this comment.
我更新了第548行的注释。这里只是处理一个非常罕见的情况，那个罕见情况会导致对象不是在callback线程里释放，这违背我们的本意。 |
||
| CHECK_NOTNULL(garbage->phy_instr_operand()); | ||
| CHECK_EQ(garbage->phy_instr_operand().use_count(), 1) << garbage->DebugName(); | ||
| // Destruct garbage. | ||
| return Maybe<void>::Ok(); | ||
| })); | ||
| } | ||
| } | ||
|
|
||
| void VirtualMachineEngine::NotifyCallback() { MoveToGarbageMsgListAndNotifyGC(); } | ||
|
|
||
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
这里支持了在sync vm时，立即发送指令的GC