Skip to content

[TransferEngine] Add cudaDeviceSynchronize after first address mapping #725

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 5 commits into
base: main
Choose a base branch
from
Open

[TransferEngine] Add cudaDeviceSynchronize after first address mapping #725

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
52d9e20
Add cudaDeviceSynchronize after first address mapping
alogfans Aug 8, 2025
8d8a02c
Move position
alogfans Aug 8, 2025
f106477
New attempt to fix incorrect transfer
alogfans Aug 12, 2025
b43ca23
More pessimistic
alogfans Aug 12, 2025
cf6cfd9
Fix format issue
alogfans Aug 12, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 16 additions & 0 deletions mooncake-integration/transfer_engine/transfer_engine_py.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,16 +21,28 @@
#include <pybind11/stl.h>

#ifdef USE_MNNVL
#include <ATen/cuda/CUDAContext.h> // pytorch dependencies
#include <cuda_runtime.h>

#include "transport/nvlink_transport/nvlink_transport.h"
static void *allocateMemory(size_t size) {
return mooncake::NvlinkTransport::allocatePinnedLocalMemory(size);
}
static void freeMemory(void *ptr) {
mooncake::NvlinkTransport::freePinnedLocalMemory(ptr);
}
static void synchronizePyTorchEvents() {
auto stream = at::cuda::getCurrentCUDAStream().stream();
cudaEvent_t ev;
cudaEventCreate(&ev);
cudaEventRecord(ev, stream);
cudaEventSynchronize(ev);
cudaEventDestroy(ev);
}
#else
static void *allocateMemory(size_t size) { return malloc(size); }
static void freeMemory(void *ptr) { free(ptr); }
static void synchronizePyTorchEvents() {}
#endif

TransferEnginePy::TransferEnginePy() {
Expand Down Expand Up @@ -258,6 +270,7 @@ int TransferEnginePy::transferSync(const char *target_hostname,
uintptr_t peer_buffer_address, size_t length,
TransferOpcode opcode) {
pybind11::gil_scoped_release release;
synchronizePyTorchEvents();
Transport::SegmentHandle handle;
{
std::lock_guard<std::mutex> guard(mutex_);
Expand Down Expand Up @@ -331,6 +344,7 @@ int TransferEnginePy::batchTransferSync(
std::vector<uintptr_t> peer_buffer_addresses, std::vector<size_t> lengths,
TransferOpcode opcode) {
pybind11::gil_scoped_release release;
synchronizePyTorchEvents();
Transport::SegmentHandle handle;
{
std::lock_guard<std::mutex> guard(mutex_);
Expand Down Expand Up @@ -419,6 +433,7 @@ batch_id_t TransferEnginePy::batchTransferAsync(
const std::vector<uintptr_t> &peer_buffer_addresses,
const std::vector<size_t> &lengths, TransferOpcode opcode) {
pybind11::gil_scoped_release release;
synchronizePyTorchEvents();
Transport::SegmentHandle handle;
{
std::lock_guard<std::mutex> guard(mutex_);
Expand Down Expand Up @@ -541,6 +556,7 @@ batch_id_t TransferEnginePy::transferSubmitWrite(const char *target_hostname,
uintptr_t peer_buffer_address,
size_t length) {
pybind11::gil_scoped_release release;
synchronizePyTorchEvents();
Transport::SegmentHandle handle;
{
std::lock_guard<std::mutex> guard(mutex_);
Expand Down
Loading