Post operations. More...

#include <dnnl.hpp>

Inheritance diagram for dnnl::post_ops:

Collaboration diagram for dnnl::post_ops:

Public Member Functions
	post_ops ()
	Creates an empty sequence of post operations.

int	len () const
	Returns the length of post operations.

primitive::kind	kind (int index) const
	Returns the kind of post operation with index `index`.

void	append_sum (float scale=1.)
	Appends accumulation (sum) post operation. More...

void	get_params_sum (int index, float &scale) const
	Gets the parameters of the accumulation (sum) post operation with index `index`. More...

void	append_eltwise (float scale, algorithm alg, float alpha, float beta)
	Appends eltwise post operation. More...

void	get_params_eltwise (int index, float &scale, algorithm &alg, float &alpha, float &beta) const
	Gets the eltwise parameters of the post operation with index `index`.

Public Member Functions inherited from dnnl::handle< dnnl_post_ops_t >
	handle ()=default
	Empty constructor. More...

	handle (dnnl_post_ops_t t, bool weak=false)
	Constructs a C handle wrapper from a C handle. More...

void	reset (dnnl_post_ops_t t, bool weak=false)
	Resets the value of a C handle. More...

dnnl_post_ops_t	get (bool allow_emtpy=false) const
	Returns the value of the underlying C handle.

Detailed Description

Post operations.

See also: Primitive Attributes: Post-ops

Examples:: cnn_inference_int8.cpp, and performance_profiling.cpp.

Member Function Documentation

◆ append_sum()

void dnnl::post_ops::append_sum ( float scale = 1. )

inline

Appends accumulation (sum) post operation.

Prior to accumulating the result, the previous value would be multiplied by scale.

The kind of this post operation is dnnl_sum.

This feature might improve performance for cases like residual learning blocks, where the result of convolution is accumulated to the previously computed activations. The parameter scale might be extreme for the integer-based computations when the result and previous activations have different logical scaling factors.

In the simplest case when the accumulation is the only post operation, the computations would be: dst[] <- scale * dst[] + op(...) // instead of dst[] <- op(...)

Note: This post operation (as well as all the others) disregards the original layout of the destination; that is, the layout of the original destination is expected to be the same as the layout of the stored destination.

◆ get_params_sum()

void dnnl::post_ops::get_params_sum	(	int	index,
		float &	scale
	)		const

inline

Gets the parameters of the accumulation (sum) post operation with index index.

◆ append_eltwise()

void dnnl::post_ops::append_eltwise	(	float	scale,
		algorithm	alg,
		float	alpha,
		float	beta
	)

inline

Appends eltwise post operation.

The kind of this post operation is dnnl_eltwise.

In the simplest case when the eltwise is the only post operation, the computations would be: dst[] <- scale * eltwise_op ( op(...) ) // instead of dst[] <- op(...) where eltwise_op is configured with the given parameters.

Examples:: cnn_inference_int8.cpp, and performance_profiling.cpp.

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

include/dnnl.hpp

Public Member Functions

Detailed Description

Member Function Documentation

◆ append_sum()

◆ get_params_sum()

◆ append_eltwise()