Contents

Sum

The sum primitive sums \(N\) tensors. The variable names follow the standard Conventions.

\[\dst(\overline{x}) = \sum\limits_{i = 1}^{N} scales(i) \cdot \src_i(\overline{x})\]

The sum primitive does not have a notion of forward or backward propagations. The backward propagation for the sum operation is simply an identity operation.

Execution Arguments

When executed, the inputs and outputs should be mapped to an execution argument index as specified by the following table.

primitive input/output

execution argument index

\(\src\)

DNNL_ARG_MULTIPLE_SRC

\(\dst\)

DNNL_ARG_DST

Operation Details

  • The \(\dst\) memory format can be either specified by a user or derived the most appropriate one by the primitive. The recommended way is to allow the primitive to choose the appropriate format.

  • The sum primitive requires all source and destination tensors to have the same shape. Implicit broadcasting is not supported.

Post-ops and Attributes

The sum primitive does not support any post-ops or attributes.

Data Types Support

The sum primitive supports arbitrary data types for source and destination tensors.

Data Representation

Sources, Destination

The sum primitive works with arbitrary data tensors. There is no special meaning associated with any logical dimensions.

API

struct dnnl::sum : public dnnl::primitive

Out-of-place summation (sum) primitive.

Public Functions

sum()

Default constructor. Produces an empty object.

sum(const primitive_desc &pd)

Constructs a sum primitive.

Parameters

  • pd: Primitive descriptor for sum primitive.

struct primitive_desc : public dnnl::primitive_desc_base

Primitive descriptor for a sum primitive.

Public Functions

primitive_desc()

Default constructor. Produces an empty object.

primitive_desc(const memory::desc &dst, const std::vector<float> &scales, const std::vector<memory::desc> &srcs, const engine &aengine, const primitive_attr &attr = primitive_attr())

Constructs a primitive descriptor for a sum primitive.

Parameters

  • dst: Destination memory descriptor.

  • scales: Vector of scales to multiply data in each source memory by.

  • srcs: Vector of source memory descriptors.

  • aengine: Engine to perform the operation on.

  • attr: Primitive attributes to use (optional).

primitive_desc(const std::vector<float> &scales, const std::vector<memory::desc> &srcs, const engine &aengine, const primitive_attr &attr = primitive_attr())

Constructs a primitive descriptor for a sum primitive.

This version derives the destination memory descriptor automatically.

Parameters

  • scales: Vector of scales by which to multiply data in each source memory object.

  • srcs: Vector of source memory descriptors.

  • aengine: Engine on which to perform the operation.

  • attr: Primitive attributes to use (optional).

memory::desc src_desc(int idx = 0) const

Returns a source memory descriptor.

Return

Source memory descriptor.

Return

A zero memory descriptor if the primitive does not have a source parameter with index pdx.

Parameters

  • idx: Source index.

memory::desc dst_desc() const

Returns a destination memory descriptor.

Return

Destination memory descriptor.

Return

A zero memory descriptor if the primitive does not have a destination parameter.

Softmax oneCCL