apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-07 09:31:35 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[Backend] Support Intel GPU with heterogeneous mode (#701)

Browse Source 
* Add some comments for python api

* support openvino gpu

* Add cpu operators

* add interface to specify hetero operators

* remove useless dir

* format code

* remove debug code

* Support GPU for ONNX
This commit is contained in:
Jason
2022-11-25 19:40:56 +08:00
committed by GitHub
parent ad5c9c08b2
commit 5b3fd9dd88
7 changed files with 156 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
fastdeploy/backends/openvino/ov_backend.h
View File

@@ -28,8 +28,9 @@ namespace fastdeploy {
struct OpenVINOBackendOption {
std::string device = "CPU";
int cpu_thread_num = -1;
int ov_num_streams = 1;
int num_streams = 0;
std::map<std::string, std::vector<int64_t>> shape_infos;
std::set<std::string> cpu_operators{"MulticlassNms"};
};
class OpenVINOBackend : public BaseBackend {
@@ -38,13 +39,13 @@ class OpenVINOBackend : public BaseBackend {
OpenVINOBackend() {}
virtual ~OpenVINOBackend() = default;
bool InitFromPaddle(
const std::string& model_file, const std::string& params_file,
const OpenVINOBackendOption& option = OpenVINOBackendOption());
bool
InitFromPaddle(const std::string& model_file, const std::string& params_file,
const OpenVINOBackendOption& option = OpenVINOBackendOption());
bool InitFromOnnx(
const std::string& model_file,
const OpenVINOBackendOption& option = OpenVINOBackendOption());
bool
InitFromOnnx(const std::string& model_file,
const OpenVINOBackendOption& option = OpenVINOBackendOption());
bool Infer(std::vector<FDTensor>& inputs,
std::vector<FDTensor>* outputs) override;
@@ -58,7 +59,7 @@ class OpenVINOBackend : public BaseBackend {
std::vector<TensorInfo> GetInputInfos() override;
std::vector<TensorInfo> GetOutputInfos() override;
std::unique_ptr<BaseBackend> Clone(void *stream = nullptr,
std::unique_ptr<BaseBackend> Clone(void* stream = nullptr,
int device_id = -1) override;
private:
@@ -71,4 +72,5 @@ class OpenVINOBackend : public BaseBackend {
std::vector<TensorInfo> input_infos_;
std::vector<TensorInfo> output_infos_;
};
} // namespace fastdeploy