Calculate filter. More...

Public Member Functions
virtual size_t	workspace ()=0
	return required workspace size in bytes

virtual void	enqueue (Tensor &dy, Tensor &dw, Tensor &ws, float beta, ExecutionContext const &e)=0
	make sure you provide worksapce of size workspace() for operations More...

Static Public Member Functions
static std::unique_ptr< BiasBackwardFilter >	create (Context &ctx, Shape const &dy_shape, DataType dt=float_data)
	Create operator for backward bias calculation. dy_shape is the shape of output tesnor for IP it should be (B,feaures_out), for Conv2d (B,feaures_out,H,W) features is number of output features - size of bias tensor.

Detailed Description

Calculate filter.

Member Function Documentation

virtual void dlprim::core::BiasBackwardFilter::enqueue	(	Tensor &	dy,
		Tensor &	dw,
		Tensor &	ws,
		float	beta,
		ExecutionContext const &	e
	)

pure virtual

make sure you provide worksapce of size workspace() for operations

if workspace() == 0 you can provide non initialized tensor

The documentation for this class was generated from the following file: