Skip to content

Latest commit

 

History

History
26 lines (18 loc) · 1.14 KB

CONTRIBUTING.md

File metadata and controls

26 lines (18 loc) · 1.14 KB
Raw

Project structure

Rules and notes for how to structure the project

Undecided

Should matrix deref to tensor-2d or the other way around?

Lazy code execution

Lazy code execution should be handled by the users of slas.

Exceptions

There is a more performent way of performing an operation on a lazily computet type.

Example of when slas should NOT handle lazy code execution

Normalization. There is no operation defined in slas that benefits from knowing that a type has been normalized. This would mean that all special cases that justify lazy normalization would have to be implemented by the user anyway.

Example of when slas should handle lazy code execution

Matrix transpose. Matrix multiplication as defined in slas can be done faster if it is known that a matrix has been lazily transposed. This means that operations can be performed faster without the user needing to redefine the matrix multiplication operation for lazily transposed matricies.

How should lazy code execution be handled by the user, when it is not handled by slas?

I don't know yet, but some possible ideas are:

  • wrapper types + flagstack + deref to slas types.