onnxruntime/onnxruntime/core/framework/execution_steps.cc
Tang, Cheng a81faee41e
Multi-stream execution support (#13495)
**Description**: This PR including following works:
1. provide stream and related synchronization abstractions in
onnxruntime.
2. enhance onnxruntime's execution planner / executor / memory arena to
support execute multiple streams in parallel.
3. deprecate the parallel executor for cpu.
4. deprecate the Fence mechanism. 
5. update the cuda / tensorrt EP to support the stream mechanism,
support running different request in different cuda stream.

**Motivation and Context**
- Why is this change required? 
currently, the execution plan is just a linear list of those primitives,
ort will execute them step by step. For any given graph, ORT will
serialize it to a fixed execution order. This sequential execution
design simplifies most scenarios, but it has the following limitations:
1. it is difficult to enable inter-node parallelization, we have a
half-baked parallel executor but it is very difficult to make it work
with GPU.
2. The fence mechanism can work with single gpu stream + cpu thread
case, but when extend to multiple stream, it is difficult to manage the
cross GPU stream synchronizations.
3. our cuda EP rely on the BFCArena to make the memory management work
with the GPU async kernels, but current BFCArena is not aware of the
streams, so it doesn't behavior correctly when run with multiple
streams.

This PR enhance our existing execution plan and executor to support
multiple stream execution. we use an unified algorithm to mange both
single stream and multiple stream scenarios.
This PR mainly focus on the infrastructure support for multiple stream
execution, that is said, given a valid stream assignment, onnxruntime
can execute it correctly. How to generate a good stream assignment for a
given model will be in the future PR.

Co-authored-by: Cheng Tang <chenta@microsoft.com@orttrainingdev9.d32nl1ml4oruzj4qz3bqlggovf.px.internal.cloudapp.net>
Co-authored-by: Cheng Tang <chenta@microsoft.com>
Co-authored-by: RandySheriffH <48490400+RandySheriffH@users.noreply.github.com>
Co-authored-by: Randy Shuai <rashuai@microsoft.com>
Co-authored-by: cao lei <jslhcl@gmail.com>
Co-authored-by: Lei Cao <leca@microsoft.com>
2022-12-15 07:39:29 -08:00

116 lines
5.2 KiB
C++

// Copyright (c) Microsoft Corporation. All rights reserved.
// Licensed under the MIT License.
#include "core/framework/execution_steps.h"
#include "core/framework/sequential_executor.h"
namespace onnxruntime {
BarrierStep::BarrierStep(size_t id) : SequentialExecutionPlan::ExecutionStep(),
barrier_id_{id} {}
Status BarrierStep::Execute(StreamExecutionContext& ctx,
size_t /*stream_idx*/,
SessionScope& /*session_scope*/,
const bool& /*terminate_flag*/,
bool& continue_flag) {
continue_flag = ctx.DecCountDownBarrier(barrier_id_);
return Status::OK();
}
std::string BarrierStep::ToString() const {
return ::onnxruntime::MakeString("Set a barrier with id: ",
barrier_id_, ", count: ", 2, ".");
}
#ifdef ENABLE_TRAINING
bool BarrierStep::IsBarrier() const { return true; }
#endif
WaitOnEPStep::WaitOnEPStep(WaitNotificationFn handle,
NotificationIndex idx) : SequentialExecutionPlan::ExecutionStep(),
wait_handle_(handle),
notification_idx_(idx) {}
Status WaitOnEPStep::Execute(StreamExecutionContext& ctx,
size_t stream_idx,
SessionScope& /*session_scope*/,
const bool& /*terminate_flag*/,
bool& continue_flag) {
ORT_ENFORCE(wait_handle_, "WaitOnEPStep.wait_handle is null");
wait_handle_(*ctx.GetDeviceStream(stream_idx), *ctx.GetNotification(notification_idx_));
// update streams clock status
if (ctx.GetDeviceStream(stream_idx)) {
ctx.GetDeviceStream(stream_idx)->UpdateStreamClock(ctx.GetNotification(notification_idx_)->GetStreamSyncTable());
}
LOGS(ctx.GetLogger(), INFO) << "stream " << stream_idx << " wait on Notification with id: " << notification_idx_;
continue_flag = true;
return Status::OK();
}
std::string WaitOnEPStep::ToString() const {
return ::onnxruntime::MakeString("WaitOnEPStep: wait on notification with id: ",
notification_idx_, ". ");
}
LaunchKernelStep::LaunchKernelStep(NodeIndex index) : SequentialExecutionPlan::ExecutionStep(),
node_index_{index} {}
Status LaunchKernelStep::Execute(StreamExecutionContext& ctx,
size_t stream_idx,
SessionScope& session_scope,
const bool& terminate_flag,
bool& continue_flag) {
#ifdef ENABLE_TRAINING
auto* node_to_execute = ctx.GetNodeToExecute();
if (node_to_execute && node_to_execute->count(node_index_) == 0) {
continue_flag = true;
return Status::OK();
}
#endif
onnxruntime::Status status = ExecuteKernel(ctx, node_index_, stream_idx, terminate_flag, session_scope);
continue_flag = status.IsOK();
return status;
}
std::string LaunchKernelStep::ToString() const {
return ::onnxruntime::MakeString("Launch kernel with node id: ", node_index_, ". ");
}
ActivateNotificationStep::ActivateNotificationStep(
NotificationIndex notification_index) : SequentialExecutionPlan::ExecutionStep(),
notification_idx_(notification_index) {}
Status ActivateNotificationStep::Execute(StreamExecutionContext& ctx,
size_t stream_idx,
SessionScope& /*session_scope*/,
const bool& /*terminate_flag*/,
bool& continue_flag) {
if (ctx.GetNotification(notification_idx_)) {
ctx.GetNotification(notification_idx_)->ActivateAndUpdate();
}
LOGS(ctx.GetLogger(), INFO) << "stream " << stream_idx
<< " activate notification with index " << notification_idx_;
continue_flag = true;
return Status::OK();
}
std::string ActivateNotificationStep::ToString() const {
return ::onnxruntime::MakeString("ActivateNotificationStep: activate notification with id: ",
notification_idx_, ". ");
}
TriggerDownstreamStep::TriggerDownstreamStep(size_t trigger_point_index) : SequentialExecutionPlan::ExecutionStep(),
trigger_point_index_(trigger_point_index) {}
Status TriggerDownstreamStep::Execute(StreamExecutionContext& ctx,
size_t /*stream_idx*/,
SessionScope& session_scope,
const bool& terminate_flag,
bool& continue_flag) {
ScheduleDownstream(ctx, trigger_point_index_, ctx.SingleThreadMode(), terminate_flag, session_scope);
continue_flag = true;
return Status::OK();
}
std::string TriggerDownstreamStep::ToString() const {
return ::onnxruntime::MakeString("TriggerDownstreamStep: trigger downstream of trigger point: ",
trigger_point_index_, ".");
}
} // namespace onnxruntime