Version 0.4.0

Offload the Eigen3 matrix-matrix multiplication to an Nvidia GPU using CUBLAS.

Changed

• Split the memory management (CudaMatrix) from the CUBLAS invocation (CudaPipeline)
• Moved all the allocation to the smart pointers inside CudaMatrix
• Removed unused headers

Introduced smart pointer

[0.3.0] 26/09/2019

Added

• Smart pointers to handle cuda resources
• New CudaMatrix class
• Check available memory in the GPU before computing

Removed

• Template class, implementation only available for double
• Triple tensor product
• Shapes struct

Tensor matrix multiplication

[0.2.0] 27/08/2019

Added

• Tensor matrix multiplacation using gemmbatched
• Async calls to memory copies.
• Properly free memory after the tensor operation is done.