Skip to content

perf: use async H2D copy for contiguous shm forward input on npu. #1101

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

Merged
JimHsiung merged 1 commit into from
Mar 25, 2026
Merged

perf: use async H2D copy for contiguous shm forward input on npu. #1101

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
52 changes: 44 additions & 8 deletions xllm/core/runtime/forward_shared_memory_manager.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,9 +16,15 @@ limitations under the License.
#include <gflags/gflags.h>

#include <cstring>
#include <mutex>
#include <optional>
#include <stdexcept>

#include "core/common/global_flags.h"
#include "platform/stream.h"
#if defined(USE_NPU)
#include "platform/npu/device_capture_lock.h"
#endif
#include "core/util/net.h"
#include "core/util/tensor_helper.h"
#include "util/utils.h"
Expand Down Expand Up @@ -1029,16 +1035,36 @@ inline void read_mm_batch_data(const char*& buffer,
inline void deserialize_raw_forward_input(const char*& buffer,
const uint64_t buffer_size,
ForwardInput& forward_input,
const torch::Device& device) {
const torch::Device& device,
Stream* stream) {
const char* device_buffer = nullptr;
#if defined(USE_NPU)
std::optional<std::unique_lock<std::mutex>> capture_lock_guard;
torch::Tensor host_input_buffer;
if (FLAGS_use_contiguous_input_buffer) {
// h to d
auto host_input_buffer =
torch::from_blob(const_cast<char*>(buffer),
{static_cast<int64_t>(buffer_size)},
torch::dtype(torch::kUInt8));
forward_input.device_input_buffer = host_input_buffer.to(device);
host_input_buffer = torch::from_blob(const_cast<char*>(buffer),
{static_cast<int64_t>(buffer_size)},
torch::TensorOptions()
.dtype(torch::kUInt8)
.device(torch::kCPU)
.pinned_memory(true));

auto device_options =
torch::TensorOptions().dtype(torch::kUInt8).device(device);

if (stream != nullptr) {
auto& capture_lock =
::xllm::npu::DeviceCaptureLock::get_instance().get_lock(
device.index());
capture_lock_guard.emplace(capture_lock);
c10::StreamGuard stream_guard = stream->set_stream_guard();
forward_input.device_input_buffer =
safe_to(host_input_buffer, device_options, true);
} else {
forward_input.device_input_buffer =
safe_to(host_input_buffer, device_options);
}

device_buffer = (char*)forward_input.device_input_buffer.data_ptr();
}
#endif
Expand Down Expand Up @@ -1127,6 +1153,12 @@ inline void deserialize_raw_forward_input(const char*& buffer,
// root cause is identified and the error is resolved.
read_tensor(buffer, input_params.new_cache_slots);
read_tensor(buffer, input_params.block_tables);

#if defined(USE_NPU)
if (device_buffer != nullptr && stream != nullptr) {
stream->synchronize();
}
#endif
}

inline void serialize_raw_forward_input(const RawForwardInput& input,
Expand Down Expand Up @@ -1543,6 +1575,9 @@ ForwardSharedMemoryManager::ForwardSharedMemoryManager(const std::string& name,
: SharedMemoryManager(name, size, is_creator), forward_type_(type) {
control_ptr_ = static_cast<ControlMetadata*>(base_address());
metadata_addr_ = static_cast<char*>(base_address()) + sizeof(ControlMetadata);
if (FLAGS_use_contiguous_input_buffer) {
stream_ = std::make_unique<Stream>();
}
}

ForwardSharedMemoryManager::~ForwardSharedMemoryManager() = default;
Expand Down Expand Up @@ -1606,7 +1641,8 @@ void ForwardSharedMemoryManager::raw_input_read(ForwardInput& input,
static_cast<char*>(base_address()) + sizeof(ControlMetadata);
uint64_t total_size;
read_data(data_ptr, total_size);
deserialize_raw_forward_input(data_ptr, total_size, input, device);
deserialize_raw_forward_input(
data_ptr, total_size, input, device, stream_.get());

return;
}
Expand Down
5 changes: 5 additions & 0 deletions xllm/core/runtime/forward_shared_memory_manager.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,12 +15,16 @@ limitations under the License.

#include <stddef.h>

#include <memory>

#include "forward_params.h"
#include "params_utils.h"
#include "util/shared_memory_manager.h"

namespace xllm {

class Stream;

constexpr int64_t kNumWaitNanoseconds = 1000; // 1us

struct ControlMetadata {
Expand Down Expand Up @@ -120,5 +124,6 @@ class ForwardSharedMemoryManager : public SharedMemoryManager {
uint64_t last_version_ = 0;
void* metadata_addr_ = nullptr;
ControlMetadata* control_ptr_ = nullptr;
std::unique_ptr<Stream> stream_ = nullptr;
};
} // namespace xllm
Loading