Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Implement sensitivities for BLAS.gemm #25

Merged
merged 1 commit into from
May 11, 2019
Merged

Implement sensitivities for BLAS.gemm #25

merged 1 commit into from
May 11, 2019

Conversation

ararslan
Copy link
Member

These are ported from Nabla.

Copy link
Member

@willtebbutt willtebbutt left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I've left a comment on the weird failing sensitivity -- I suspect that it's just a minor thing. LGTM other than that, so please do merge once finite-differencing tests pass for alpha.

src/rules/blas.jl Outdated Show resolved Hide resolved
@willtebbutt
Copy link
Member

willtebbutt commented Apr 29, 2019

I've also just realised that the type constraints are probably a bit loose. I think that BLAS routines are usually restricted to strided arrays with particular types of floating-point elements. Perhaps we should do the same here?

@ararslan
Copy link
Member Author

gemm is a case where the methods in LinearAlgebra are, for whatever reason, specified as AbstractMatrix instead of StridedMatrix. That is a good observation about the element type; it's restricted to BlasFloat, and with alpha's type matching the element type of the matrices.

@ararslan
Copy link
Member Author

To do:

  • Add tests for non-square matrices

These are ported from Nabla.
@ararslan
Copy link
Member Author

Rectangular tests added. Good to go?

@willtebbutt
Copy link
Member

LGTM

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

2 participants