image.png

撰文|郑建华、赵露阳

1

Op在虚拟机里的执行

1.1 PhysicalRun和InstructionsBuilder

上一篇文章《OneFlow源码解析:Op、Kernel与解释器》中提到:

PhysicalRun接受一个lambda函数作为参数,这里即InstructionsBuilder->Call方法,该方法接受kernel、input/output的eager blob object、kernel执行的上下文作为参数。Call方法实际会完成OpCall指令的构建,并最终将其派发至vm指令列表中,等待VM实际调度执行。

这个PhysicalRun函数里包裹着一个lambda函数:

JUST(PhysicalRun([&](InstructionsBuilder* builder) -> Maybe<void> {
    return builder->Call(xxx);
}));

其中,lambda函数接受一个InstructionsBuilder指针(builder),并调用builder->Call方法,用于实际完成Op指令在VM中的构建。而PhysicalRun(https://github.com/Oneflow-In...)在 oneflow/core/framework/instructions_builder.h中定义,其接受lambda函数作为模版参数(CallbackT):

// Make VM instructions with instruction builder and run instructions with physical/local view.
template<typename CallbackT>
Maybe<void> PhysicalRun(const CallbackT& Build) {
  vm::InstructionList instruction_list;
  InstructionsBuilder instructions_builder(&instruction_list);
  JUST(Build(&instructions_builder));
  JUST(vm::Run(instructions_builder.mut_instruction_list()));
  return Maybe<void>::Ok();
}

可见,PhysicalRun函数中,首先初始化一个InstructionsBuilder,然后将InstructionsBuilder指针作为参数传给lambda函数,完成实际指令的构建;最后通过vm::Run()方法将该指令发送至VM,等候VM实际调度和执行。Run方法如下:

Maybe<void> Run(vm::InstructionList* instruction_list) {
  auto* virtual_machine = JUST(SingletonMaybe<VirtualMachine>());
  JUST(virtual_machine->Receive(instruction_list));
  return Maybe<void>::Ok();
}

可以看见,Run()方法获取了全局单例的VM对象指针,然后通过vm的Receive()方法,将该条指令发送给虚拟机(所以这里Run其实有点歧义,更贴切的意思,其实是指令发送或传送)。

这个VirtualMachine->Receive方法很重要,会在后面的第2.章节中详细展开。

1.2 InstructionsBuilder

上面PhysicalRun函数中的InstructionsBuilder,类似一个指令构建的helper,InstructionsBuilder的系列方法配合指令策略(InstructionPolicy),可以帮助构建不同类型的vm指令。

从InstructionsBuilder
https://github.com/Oneflow-In...)的定义中,我们可以看到指令的构建方法,其中常用方法如下:

// 用于lazy mode(nn.Graph)
// Build VM execution instructions with NNGraph's inputs/outputs/parameters for NNGraph execution.
Maybe<void> LaunchLazyJob(const vm::EagerBlobObjectListPtr& inputs,
                          const vm::EagerBlobObjectListPtr& outputs,
                          const vm::EagerBlobObjectListPtr& parameters,
                          const std::shared_ptr<NNGraphIf>& nn_graph);


// 用于全局同步,同步等待所有指令调用完成
Maybe<void> GlobalSync();

// 用于Tensor内存释放(归还allocator)
Maybe<void> ReleaseTensor(const std::shared_ptr<vm::EagerBlobObject>& eager_blob_object);

// 操作Tensor实际内存(blob)
template<typename T>
Maybe<void> AccessBlobByCallback(
    const T tensor,
    const std::function<void(ep::Stream*, const std::shared_ptr<vm::EagerBlobObject>&)>& callback,
    const std::string& modifier);

// 最常用的指令构建方法,用于构造op执行所需的OpCall指令
Maybe<void> Call(const std::shared_ptr<one::StatefulOpKernel>& opkernel,
                   vm::EagerBlobObjectList&& input_eager_blob_objects,
                   vm::EagerBlobObjectList&& output_eager_blob_objects,
                   const one::OpExprInterpContext& ctx, Symbol<Stream> stream);

1.3 InstructionPolicy

InstructionPolicy
https://github.com/Oneflow-In...)——指令策略,通常用于配合InstructionsBuilder实际构建出不同的vm指令。InstructionPolicy的子类实现如下:

image.png

这些子类的InstructionPolicy可近似认为是指令类型。如,用于Op执行的OpCallInstructionPolicy、用于Tensor内存释放的ReleaseTensorInstructionPolicy、用于屏障阻塞的BarrierInstructionPolicy等。

以Op执行为例:

JUST(PhysicalRun([&](InstructionsBuilder* builder) -> Maybe<void> {
    return builder->Call(xxx);
}));

实际上是通过InstructionsBuilder的Call方法
https://github.com/Oneflow-In...),配合OpCall的指令策略(OpCallInstructionPolicy),构造了OpCall指令:

Maybe<void> InstructionsBuilder::Call(
    const std::shared_ptr<one::StatefulOpKernel>& opkernel,
    vm::EagerBlobObjectList&& input_eager_blob_objects,
    vm::EagerBlobObjectList&& output_eager_blob_objects,
    const std::shared_ptr<const one::GlobalTensorInferResult>& global_tensor_infer_result,
    const one::OpExprInterpContext& ctx, Symbol<Stream> stream) {
  ...
  ...
  // 获取当前vm stream
  auto* vm_stream = JUST(Singleton<VirtualMachine>::Get()->GetVmStream(stream));
  // 通过OpCallInstructionPolicy初始化OpCall指令
  auto instruction = intrusive::make_shared<vm::Instruction>(
      vm_stream, std::make_shared<vm::OpCallInstructionPolicy>(
                     vm_stream, opkernel, std::move(input_eager_blob_objects),
                     std::move(output_eager_blob_objects), global_tensor_infer_result, ctx,
                     *one::CurrentDevVmDepObjectConsumeMode()));
  // 指令入列表
  instruction_list_->EmplaceBack(std::move(instruction));
  return Maybe<void>::Ok();
}

并将构建好的指令塞入指令列表,待后续VM调度并实际执行。

2

虚拟机的运行原理

2.1 VM初始化

OneFlow环境初始化时,会触发VirtualMachineScope
https://github.com/Oneflow-In...)的初始化:

VirtualMachineScope::VirtualMachineScope(const Resource& resource) {
  Singleton<VirtualMachine>::New();
}

进而触发VM对象——VirtualMachine
https://github.com/Oneflow-In...)的初始化。VM作为一个Singleton对象,全局唯一。

VirtualMachine::VirtualMachine() : disable_vm_threads_(false), scheduler_stopped_(false) {
  // Class VirtualMachineEngine only cares the basic logical of vm, while class VirtualMachine
  // manages threads and condition variables.
  // In order to notify threads in VirtualMachineEngine, a notify callback lambda should be take as
  // an argument for VirtualMachineEngine's constructor.
  engine_ = intrusive::make_shared<vm::VirtualMachineEngine>();
  OF_PROFILER_NAME_THIS_HOST_THREAD("_Main");
  std::function<void()> SchedulerInitializer;
  GetSchedulerThreadInitializer(&SchedulerInitializer);
  schedule_thread_ = std::thread(&VirtualMachine::ScheduleLoop, this, SchedulerInitializer);
  transport_local_dep_object_.Reset();
}

VM初始化中最重要的内容,便是:

1.初始化了一个VM的执行引擎——VirtualMachineEngine
2.通过VirtualMachine::ScheduleLoop启动了VM的调度线程

VirtualMachine::ScheduleLoop

VM对象只负责条件变量和线程管理;而主要业务逻辑处理(包括指令构建、调度、派发和执行等),则由VirtualMachineEngine
https://github.com/Oneflow-In...)对象负责;VM初始化时还开辟了单独的schedule线程用于VM引擎处理调度逻辑,在VirtualMachine::ScheduleLoop
https://github.com/Oneflow-In...)中:

void VirtualMachine::ScheduleLoop(const std::function<void()>& Initializer) {
  SyncVmModeGuard guard(SyncVmMode::kEnable);
  Initializer();
  MultiThreadScheduleCtx schedule_ctx{};
  while (pending_notifier_.WaitAndClearNotifiedCnt() == kNotifierStatusSuccess) {
    OF_PROFILER_RANGE_GUARD("VirtualMachine::ScheduleLoop");
    auto start = std::chrono::steady_clock::now();
    static constexpr int kWorkingMicroseconds = 1000;
    // Every time this thread wakes up, engine_ is scheduled for about `kWorkingMicroseconds`.
    // The cost of os thread switching is about 5-10 microseconds. Doing more scheduling in
    // a single waiting up can reach higher performance.
    do {
      do {
        const size_t total_inserted = engine_->total_inserted_instruction_cnt();
        const size_t total_erased = engine_->total_erased_instruction_cnt();
        engine_->Schedule(schedule_ctx);
        if (ThreadLocalEnvBool<ONEFLOW_VM_ENABLE_SCHEDULE_YIELD>()
            && total_inserted == engine_->total_inserted_instruction_cnt()
            && total_erased == engine_->total_erased_instruction_cnt()) {  // nothing handled.
          std::this_thread::yield();
        }
      } while (!engine_->SchedulerThreadUnsafeEmpty());
    } while (MicrosecondsFrom(start) < kWorkingMicroseconds);
  }
  ScheduleUntilVMEmpty(engine_.Mutable(), schedule_ctx);
  CHECK_JUST(ForEachThreadCtx(engine_.Mutable(), [&](vm::ThreadCtx* thread_ctx) -> Maybe<void> {
    thread_ctx->mut_notifier()->Close();
    return Maybe<void>::Ok();
  }));
  {
    std::unique_lock<std::mutex> lock(worker_threads_mutex_);
    for (const auto& worker_thread : worker_threads_) { worker_thread->join(); }
  }
  scheduler_stopped_ = true;
}

ScheduleLoop是一个近似于busy loop的while循环,pending_notifier_是VM内部维护的成员,实际上是ScheduleLoop线程的通知/唤醒者,其定义位于oneflow/oneflow/core/common/notifier.h

class Notifier final {
 public:
  OF_DISALLOW_COPY_AND_MOVE(Notifier);
  Notifier() : notified_cnt_(0), is_closed_(false) {}
  ~Notifier() = default;

  NotifierStatus Notify();
  NotifierStatus WaitAndClearNotifiedCnt();
  void Close();

 private:
  size_t notified_cnt_;
  std::mutex mutex_;
  bool is_closed_;
  std::condition_variable cond_;
};
class Notifier final {
 public:
  OF_DISALLOW_COPY_AND_MOVE(Notifier);
  Notifier() : notified_cnt_(0), is_closed_(false) {}
  ~Notifier() = default;

  NotifierStatus Notify();
  NotifierStatus WaitAndClearNotifiedCnt();
  void Close();

 private:
  size_t notified_cnt_;
  std::mutex mutex_;
  bool is_closed_;
  std::condition_variable cond_;
};

其主要维护了互斥锁mutex_、线程是否关闭的flag is_closed_、条件变量cond_。忽略线程唤醒、超时相关逻辑,ScheduleLoop中最重要的事情是engine_->Schedule(schedule_ctx);

while (pending_notifier_.WaitAndClearNotifiedCnt() == kNotifierStatusSuccess) {
    auto start = std::chrono::steady_clock::now();
    ...
    do {
      do {
        ...
        engine_->Schedule(schedule_ctx);
        ...
      } while (!engine_->SchedulerThreadUnsafeEmpty());
    } while (MicrosecondsFrom(start) < kWorkingMicroseconds);
  }

当VM维护的指令队列不为空时,便不断唤醒VM引擎执行指令调度逻辑——engine->Schedule()

2.2 VM指令调度

void VirtualMachineEngine::Schedule(const ScheduleCtx& schedule_ctx) {
  // Release finished instructions and try to schedule out instructions in DAG onto ready list.
  if (unlikely(mut_active_stream_list()->size())) { ReleaseFinishedInstructions(schedule_ctx); }
  // Try run the first barrier instruction.
  if (unlikely(mut_barrier_instruction_list()->size())) { TryRunBarrierInstruction(schedule_ctx); }
  // Handle pending instructions, and try schedule them to ready list.
  // Use thread_unsafe_size to avoid acquiring mutex lock.
  // The inconsistency between pending_instruction_list.list_head_.list_head_.container_ and
  // pending_instruction_list.list_head_.list_head_.size_ is not a fatal error because
  // VirtualMachineEngine::Schedule is always in a buzy loop. All instructions will get handled
  // eventually.
  //  VirtualMachineEngine::Receive may be less effiencient if the thread safe version
  //  `pending_instruction_list().size()` used here, because VirtualMachineEngine::Schedule is more
  //  likely to get the mutex lock.
  if (unlikely(local_pending_instruction_list().size())) {
    HandleLocalPending();
  } else if (unlikely(pending_instruction_list().thread_unsafe_size())) {
    // MoveTo is under a lock.
    mut_pending_instruction_list()->MoveTo(mut_local_pending_instruction_list());
    if (local_pending_instruction_list().size()) { HandleLocalPending(); }
  }
  // dispatch ready instructions and try to schedule out instructions in DAG onto ready list.
  if (unlikely(mut_ready_instruction_list()->size())) {
    DispatchAndPrescheduleInstructions(schedule_ctx);
  }
  // handle scheduler probes
  if (unlikely(local_probe_list_.size())) {
    HandleLocalProbe();
  } else if (unlikely(probe_list_.thread_unsafe_size())) {
    probe_list_.MoveTo(&local_probe_list_);
    if (local_probe_list_.size()) { HandleLocalProbe(); }
  }
}

VM引擎维护了一系列指令列表的成员:

InstructionMutexedList pending_instruction_list_;
// local_pending_instruction_list_ should be consider as the cache of pending_instruction_list_.
InstructionList local_pending_instruction_list_;
ReadyInstructionList ready_instruction_list_;
LivelyInstructionList lively_instruction_list_;
BarrierInstructionList barrier_instruction_list_;
  • pending相关的instruction_list是悬挂/待处理的指令列表;
  • lively相关的instruction_list是活跃的正在执行中的指令列表;
  • ready相关的instruction_list则是已完成准备工作(指令融合、指令DAG构建等)待执行的指令列表;

VM引擎Schedule时,会对指令队列做相应处理,包括:

  • 将已完成准备工作的指令放入ready_instruction_list_中维护;
  • 尝试运行barrier指令列表(barrier_instruction_list_)中的第一条指令;
  • 如果本地pending指令列表(local_pending_instruction_list_)非空,则通过HandleLocalPending方法处理这些悬挂指令(指令融合、指令执行DAG图构建、插入ready列表)
  • 如果ready指令列表非空,则通过DispatchAndPrescheduleInstructions方法进行指令派发和预调度处理。

这里重点介绍指令派发相关的DispatchAndPrescheduleInstructions方法,其中DispatchAndPrescheduleInstructions中最主要的是就是DispatchInstruction指令派发方法,这里的指令派发可以认为实际上就是指令执行

2.3 VM指令派发

VirtualMachineEngine::DispatchInstruction
https://github.com/Oneflow-In...)方法是vm引擎中的核心,其实际完成了指令的派发和实际执行,代码如下:

template<void (VirtualMachineEngine::*OOMHandler)(vm::Stream*, const ScheduleCtx&)>
void VirtualMachineEngine::DispatchInstruction(Instruction* instruction,
                                               const ScheduleCtx& schedule_ctx) {
  auto* stream = instruction->mut_stream();
  // Prepare
  {
    // 指令的Prepare
    const auto& ret = TRY(instruction->Prepare());
    if (unlikely(!ret.IsOk())) {
      // 处理指令Prepare过程中的OOM的逻辑
      if (ret.error()->has_out_of_memory_error()) {
        // 让allocator释放不必要的cacahe,再重新执行指令的Prepare
        (this->*OOMHandler)(stream, schedule_ctx);
        ...
      }
    }
  }
  // 将当前指令放入running_instruction_list
  stream->mut_running_instruction_list()->PushBack(instruction);
  if (stream->active_stream_hook().empty()) { mut_active_stream_list()->PushBack(stream); }
  // Compute
  if (OnSchedulerThread(*stream)) {
    // StreamPolicy的Run方法触发指令的实际执行——Compute
    stream->stream_policy().Run(instruction);
  } else {
    stream->mut_thread_ctx()->mut_worker_pending_instruction_list()->PushBack(instruction);
    schedule_ctx.OnWorkerLoadPending(stream->mut_thread_ctx());
  }
}

DispatchInstruction的核心主要有2块:

  • 执行指令的Prepare
  • 执行指令的Compute

Prepare负责一些指令执行前的准备;Compute则是实际的指令执行,指令执行并不是直接通过instruction->Run而是在StreamPolicy的Run方法中完成的,这里又涉及到一个StreamPolicy对象。

StreamPolicy::Run

StreamPolicy
https://github.com/Oneflow-In...)是个虚基类:

class StreamPolicy {
 public:
  virtual ~StreamPolicy() = default;

  virtual ep::Stream* stream() = 0;
  virtual vm::Allocator* mut_allocator() = 0;
  virtual DeviceType device_type() const = 0;

  virtual void InitInstructionStatus(const Stream& stream,
                                     InstructionStatusBuffer* status_buffer) const = 0;
  virtual void DeleteInstructionStatus(const Stream& stream,
                                       InstructionStatusBuffer* status_buffer) const = 0;
  virtual bool QueryInstructionStatusDone(const Stream& stream,
                                          const InstructionStatusBuffer& status_buffer) const = 0;
  virtual void Run(Instruction* instruction) const = 0;

  virtual bool OnSchedulerThread(StreamType stream_type) const;
  virtual bool SupportingTransportInstructions() const = 0;

 protected:
  StreamPolicy() = default;
};
  • stream()方法返回ep::Stream指针,指向的是针对不同平台的ep::stream对象。
  • mut_allocator()方法返回一个vm的Allocator指针,用于内存分配/释放。
  • InitInstructionStatus/QueryInstructionStatusDone/DeleteInstructionStatus用于创建/查询/销毁指令执行状态
  • Run方法则是核心,定义了该Stream具体运行时的逻辑。
这里的ep在oneflow中是execution provider的缩写,ep从本质上来讲就是一个针对不同硬件平台的executor抽象。

StreamPolicy相关的继承和子类如下:

image.png

看一下EpStreamPolicyBase的Run方法(https://github.com/Oneflow-In...):

void EpStreamPolicyBase::Run(Instruction* instruction) const {
  ...
  auto* stream = instruction->mut_stream();
  EpStreamPolicyBase* ep_stream_policy_base =
      dynamic_cast<EpStreamPolicyBase*>(stream->mut_stream_policy());
  ...
  auto* ep_device = ep_stream_policy_base->GetOrCreateEpDevice();
  ep_device->SetAsActiveDevice();
  instruction->Compute();
  ...
}

首先获取了该stream对应的ep device,然后执行了instruction的Compute方法,即指令的实际执行

2.4 VM执行执行

以OpCall指令为例,看一下op指令的Compute
https://github.com/Oneflow-In...):

void OpCallInstructionPolicy::Compute(vm::Instruction* instruction) {
  OpCallInstructionUtil::Compute(this, instruction);
}

OpCallInstructionPolicy方法调用了OpCallInstructionUtil的Compute方法:

image.png

上面我们可以看到,在指令Prepare时,做了output tensor内存分配;而指令Compute中最重要的方法是:

  • TryInitOpKernelStateAndCache——初始化一些kernel计算需要的状态或缓存
  • OpKernelCompute——执行该op对应的kernel,kernel内主要是实际的op计算逻辑

image.png

user kernel统一位于:oneflow/user/kernels目录下,.cpp通常对应cpu kernel逻辑;.cu为cuda kernel逻辑。到这里,就会触发user_kernel的Compute方法,不同op的kernel计算逻辑不同,以rele op为例,实际Compute过程可参考文章《算子在深度学习框架中的执行及interpreter》的第5小节。

2.5 VM指令发送

这里的VM指令发送,指的是VM外部的指令发送过程(不是VM内部的指令派发)。上面2.1~2.3小节介绍了VM以及VM引擎的初始化、VM内部指令的调度、派发和实际执行的过程,那么这些指令是如何发送到VM的呢?答案是:在1.1小节中提到的PhysicalRun

PhysicalRun最终会触发VirtualMachine->Receive方法,并通过VirtualMachineEngine的Receive方法完成外部指令 -> VM内部的发送。

VirtualMachineEngine的Receive方法(https://github.com/Oneflow-In...)主要将该指令通过MoveFrom方法push back到指令悬挂列表(pending_instruction_list_)的末尾,从而完成指令的发送。

// Returns true if old scheduler_pending_instruction_list is empty
Maybe<bool> VirtualMachineEngine::Receive(InstructionList* compute_instruction_list) {
  OF_PROFILER_RANGE_GUARD("vm:Receive");
#ifdef OF_ENABLE_PROFILER
  INTRUSIVE_UNSAFE_FOR_EACH_PTR(compute_instruction, compute_instruction_list) {
    OF_PROFILER_RANGE_GUARD(compute_instruction->DebugName());
  }
#endif

  bool old_list_empty = mut_pending_instruction_list()->MoveFrom(compute_instruction_list);
  return old_list_empty;
}

3

小结

至此,Op执行相关的流程算是大体串了一遍。一句flow.relu()后面会涉及这么多内容。但这里其实也只关注了主干逻辑,忽略了中间大量的细节。

流程的梳理只是第一步,还需要从中归纳总结一些概念和概念之间的关系,再结合公开资料反推印证设计理念的落地实现。

不过目前对代码和设计的了解还很肤浅,下面的内容纯属大胆猜测。

3.1 UserOpExpr

UserOpExpr表示UserOp执行时所需的上下文,其实UserOp只是Op中的一种。下图展示了不同Op的继承关系。可以看到tensor从local/global之间的转换等也都涉及不同的OpExpr。

image.png

3.2 Op执行的宏观脉络

从上面的类关系图出发,以核心类为节点,也能看出Op执行流程的宏观脉络。整个流程大体在下面这些角色之间流转:

ReluFunctor
UserOpExpr
Interpreter
PhysicalRun
VirtualMachine->Receive
VirtualMachine->ScheduleLoop ...

3.3 虚拟机运行和调度总结

VM -> ScheduleLoop

   VM引擎Schedule
           处理悬挂指令(HandleLocalPending)
           指令派发(DispatchInstruction)
                  准备(instruction->Prepare)
                  执行(StreamPolicy.Run -> instruction->Compute)
           指令预调度

VM -> Receive

     VM引擎 -> Receive
           指令入悬挂列表

通常,我们习惯在动态图模式下训练深度学习网络,使用Python搭建网络,并通过各种op进行前向、反向、loss计算、调试debug等过程,这些Python代码可以看作是动态的op的执行序列。

OneFlow虚拟机将op执行序列抽象成了各种VM指令序列。OneFlow的虚拟机会对这些op执行序列进行动态翻译并生成VM指令序列,通过PhysicalRun构造完毕后,动态地将指令发送至VM的悬挂列表中维护。这些指令或在时间上存在先后顺序,或在数据上存在依赖关系,所以悬挂列表中的指令后续会被虚拟机进行一些指令融合、指令连边、动态构建指令DAG图的过程,然后移入就绪列表中维护,等待虚拟机调度并实际执行。虚拟机负责维护若干个指令队列,以及指令在这些队列之间的状态转换。

OneFlow虚拟机还统一了动态图模式(Eager Mode)和静态图模式(Lazy Mode)。静态图模式下,通过nn.Graph编译出深度学习网络的Job,这个Job同样被虚拟机抽象成了VM指令并接受虚拟机的调度和执行。大胆猜测一下,这也为日后动静转换、更极致的性能优化埋下了伏笔。

参考资料

欢迎下载体验 OneFlow v0.8.0 最新版本:https://github.com/Oneflow-In...


OneFlow
10 声望22 粉丝

★ OneFlow深度学习框架:github.com/Oneflow-Inc/oneflow ★ OF云平台:oneflow.cloud