A primitive to compute element-wise operations such as parametric rectifier linear unit (ReLU). More...

Functions
mkldnn_status_t MKLDNN_API	mkldnn_eltwise_forward_desc_init (mkldnn_eltwise_desc_t eltwise_desc, mkldnn_prop_kind_t prop_kind, mkldnn_alg_kind_t alg_kind, const mkldnn_memory_desc_t data_desc, float alpha, float beta)
	Initializes an `eltwise_desc` for forward propagation using `prop_kind` (possible values are mkldnn_forward_training and mkldnn_forward_inference), `alg_kind` algorithm, memory descriptor `data_desc`, `alpha`, and `beta` parameters. More...

mkldnn_status_t MKLDNN_API	mkldnn_eltwise_backward_desc_init (mkldnn_eltwise_desc_t eltwise_desc, mkldnn_alg_kind_t alg_kind, const mkldnn_memory_desc_t diff_data_desc, const mkldnn_memory_desc_t *data_desc, float alpha, float beta)
	Initializes an `eltwise_desc` for backward propagation using `alg_kind` algorithm memory descriptors `diff_data_desc` and `data_desc`, and the `alpha` and `beta` parameters. More...

Detailed Description

A primitive to compute element-wise operations such as parametric rectifier linear unit (ReLU).

Both forward and backward passes support in-place operation; that is, src and dst point to the same memory for forward pass, and diff_dst and diff_src point to the same memory for backward pass.

Warning: Because the original src is required for backward pass, in-place forward pass in general cannot be applied during training. However, for some kinds of element-wise operations (namely ReLU with alpha parameter equals 0), dst and src can be interchangeable for the backward pass, which enables performance of in-place forward even for training.

See also: Eltwise in developer guide; Eltwise in C++ API

Function Documentation

◆ mkldnn_eltwise_forward_desc_init()

mkldnn_status_t MKLDNN_API mkldnn_eltwise_forward_desc_init	(	mkldnn_eltwise_desc_t *	eltwise_desc,
		mkldnn_prop_kind_t	prop_kind,
		mkldnn_alg_kind_t	alg_kind,
		const mkldnn_memory_desc_t *	data_desc,
		float	alpha,
		float	beta
	)

Initializes an eltwise_desc for forward propagation using prop_kind (possible values are mkldnn_forward_training and mkldnn_forward_inference), alg_kind algorithm, memory descriptor data_desc, alpha, and beta parameters.

See also: mkldnn_eltwise_desc_t for details.

Inputs:

src (mkldnn_query_src_md, 0)

Outputs:

dst (mkldnn_query_dst_md, 0)

Examples:: cpu_cnn_inference_f32.c, cpu_cnn_training_f32.c, and gpu_getting_started.c.

◆ mkldnn_eltwise_backward_desc_init()

mkldnn_status_t MKLDNN_API mkldnn_eltwise_backward_desc_init	(	mkldnn_eltwise_desc_t *	eltwise_desc,
		mkldnn_alg_kind_t	alg_kind,
		const mkldnn_memory_desc_t *	diff_data_desc,
		const mkldnn_memory_desc_t *	data_desc,
		float	alpha,
		float	beta
	)

Initializes an eltwise_desc for backward propagation using alg_kind algorithm memory descriptors diff_data_desc and data_desc, and the alpha and beta parameters.

See also: mkldnn_eltwise_desc_t for details.