I am trying to write an application that will have multiple linear algebra drivers - I would like to use Eigen for Matrices, but a gpu tensor library for Tensors.
My approach would be to write a wrapper for both libraries that inherits from an abstract wrapper. The key question is can I do it without templates? Has anybody already done it for me? Eigen uses curious template pattern extensively - is there any simple way to translate one into the other?
Note: I anticipate that some people might say that using eigen gives me no benefit if I do that. This would be true, except in my driver I intend to have a few high level operations written entirely in Eigen - this is why it's useful.
Aucun commentaire:
Enregistrer un commentaire