Skip to content

Sub in onnx + lean format #226

Closed
Closed
@MirkoCalvi

Description

@MirkoCalvi

Current Problem

The "sub" mathematical operation does not respect the onnx format and it is not present any "lean" version of the function

Standard & lean format

The "standard version" of a method is in the format:

    pub fn methodName (args) !Tensor(T) {
       ...
       outputShape = compute output shape...
       outputTensor = compute output tensor...
       ...
       methodName_lean(input, args..., output)

       return outputTensor;
    }

It is suggested to write a get_methodName_oputput_shape() to compute the shape of the output so that the same method can be called during the generation of the readyGraph.

The "lean version" of a method is in the format:

    pub inline fn methodName_lean (args) void {
       ...
       ... actual computation of the output
    }

"lean" and "standard" are mandatory.
The lean version is called during the codegen and the standard version is used in testings.

Metadata

Metadata

Assignees

No one assigned

    Labels

    MathRelated to TensorMath

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions