Skip to content

Gpu support#4

Open
dev10110 wants to merge 5 commits intomainfrom
gpu_support
Open

Gpu support#4
dev10110 wants to merge 5 commits intomainfrom
gpu_support

Conversation

@dev10110
Copy link
Copy Markdown
Owner

Allows constructing a STGPKFProblem where all the large matrices live on an Nvidia GPU.

mostly a pass through of CuArray instead of Array , but because of fairly extensive use of Kronecker products in specific matrices, it requires some special implementations of Kronecker products to work on the GPU.

The CUDA requirement is added in as a weak dependency (i.e., as a package extension) so it should only be loaded if the user has loaded the CUDA package.

The main entry point is CudaSTGPKFProblem instead of STGPKFProblem - the rest of it should work as before.

There is a breaking change on the type heirarchy, where before STGPKFProblem{P, F, ...} has been replaced with STGPKFProblem{F, P} where F refers to the float-type, and P refers to the state-type of the problem domain.

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.

1 participant