Sparse CSR CUDA: Add block torch.addmv when mat is sparse (#68708)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/68708
This PR adds block CSR matrix times dense vector multiplication.
cc nikitaved pearu cpuhrsch IvanYashchuk ngimel
Test Plan: Imported from OSS
Reviewed By: pbelevich
Differential Revision: D32647694
Pulled By: cpuhrsch
fbshipit-source-id: a1c120691c4350284b156fe4259eda684b734b66
