Skip to content

[Runtime] Share resource between GPUCompatibleCPUDevice. #285

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
merged 1 commit into from
Jun 30, 2022
Merged

[Runtime] Share resource between GPUCompatibleCPUDevice. #285

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
10 changes: 4 additions & 6 deletions tensorflow/core/common_runtime/direct_session.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -274,12 +274,10 @@ class DirectSessionFactory : public SessionFactory {
ResourceMgr* shared_rmgr = new ResourceMgr("localhost");
DeviceResourceMgrMap dev_rmgr_map;
std::string dev_prefix("/job:localhost/replica:0/task:0");
for (int i = 0; i < session_num; ++i) {
std::string dev_name = dev_prefix + "/device:CPU:" + std::to_string(i);
dev_rmgr_map.device_rmgr_map[dev_name] = shared_rmgr;
dev_name = dev_prefix + "/device:cpu:" + std::to_string(i);
dev_rmgr_map.device_rmgr_map[dev_name] = shared_rmgr;
}
dev_rmgr_map.device_rmgr_map[dev_prefix+"/device:CPU:0"] = shared_rmgr;
dev_rmgr_map.device_rmgr_map[dev_prefix+"/device:cpu:0"] = shared_rmgr;
dev_rmgr_map.device_rmgr_map["/device:CPU:0"] = shared_rmgr;
dev_rmgr_map.device_rmgr_map["/device:cpu:0"] = shared_rmgr;

std::vector<std::unique_ptr<Device>> devices;
TF_RETURN_IF_ERROR(DeviceFactory::AddDevices(
Expand Down
22 changes: 21 additions & 1 deletion tensorflow/core/common_runtime/gpu/gpu_device_factory.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -90,6 +90,19 @@ class GPUCompatibleCPUDevice : public ThreadPoolDevice {
options.config.gpu_options().force_gpu_compatible();
}
}
GPUCompatibleCPUDevice(const SessionOptions& options, const string& name,
Bytes memory_limit, const DeviceLocality& locality,
Allocator* allocator,
const DeviceResourceMgrMap* dev_rmgr_map)
: ThreadPoolDevice(options, name, memory_limit,
locality, allocator, dev_rmgr_map),
numa_node_(locality.numa_node()) {
if (options.config.has_gpu_options()) {
force_gpu_compatible_ =
options.config.gpu_options().force_gpu_compatible();
}
}

~GPUCompatibleCPUDevice() override {}

Allocator* GetAllocator(AllocatorAttributes attr) override {
Expand Down Expand Up @@ -118,6 +131,12 @@ class GPUCompatibleCPUDeviceFactory : public DeviceFactory {

Status CreateDevices(const SessionOptions& options, const string& name_prefix,
std::vector<std::unique_ptr<Device>>* devices) override {
return CreateDevices(options, name_prefix, devices, nullptr);
}

Status CreateDevices(const SessionOptions& options, const string& name_prefix,
std::vector<std::unique_ptr<Device>>* devices,
const DeviceResourceMgrMap* dev_rmgr_map) override {
int n = 1;
auto iter = options.config.device_count().find("CPU");
if (iter != options.config.device_count().end()) {
Expand All @@ -133,7 +152,8 @@ class GPUCompatibleCPUDeviceFactory : public DeviceFactory {
locality.set_numa_node(numa_node);
devices->push_back(absl::make_unique<GPUCompatibleCPUDevice>(
options, name, Bytes(256 << 20), DeviceLocality(),
ProcessState::singleton()->GetCPUAllocator(numa_node)));
ProcessState::singleton()->GetCPUAllocator(numa_node),
dev_rmgr_map));
}

return Status::OK();
Expand Down