Class PassStrategy¶
Defined in File paddle_pass_builder.h
Inheritance Relationships¶
Base Type¶
public paddle::PaddlePassBuilder(Class PaddlePassBuilder)
Derived Types¶
public paddle::CpuPassStrategy(Class CpuPassStrategy)public paddle::GpuPassStrategy(Class GpuPassStrategy)
Class Documentation¶
-
class paddle::PassStrategy : public paddle::PaddlePassBuilder¶
This class defines the pass strategies like whether to use gpu/cuDNN kernel/MKLDNN.
Subclassed by paddle::CpuPassStrategy, paddle::GpuPassStrategy
Public Functions
-
inline explicit PassStrategy(const std::vector<std::string> &passes)¶
Constructor of PassStrategy class. It works the same as PaddlePassBuilder class.
- Parameters
passes – [in] passes’ types.
-
inline virtual void EnableCUDNN()¶
Enable the use of cuDNN kernel.
-
inline virtual void EnableMKLDNN()¶
Enable the use of MKLDNN. The MKLDNN control exists in both CPU and GPU mode, because there can still be some CPU kernels running in GPU mode.
-
inline virtual void EnableMkldnnQuantizer()¶
Enable MKLDNN quantize optimization.
-
inline bool use_gpu() const¶
Check if we are using gpu.
- Returns
A bool variable implying whether we are in gpu mode.
-
virtual ~PassStrategy() = default¶
Default destructor.
-
inline explicit PassStrategy(const std::vector<std::string> &passes)¶