Implementation in Collections library of CUDA operators for key operators.

Issue addressed

Nektar++ cannot yet be used with GPU co-processors. This MR enables the evaluation of these operators on NVIDIA GPUs.

Proposed solution

This MR incorporates a CUDA implementation of the key operator kernels: Helmholtz, BwdTrans, PhysDeriv and IProduct into the Collections library.

Detail on implementation

The kernels are integrated with the existing factory-pattern in Collections which decouples the code from the rest of the library kernels and other Nektar++ libraries. This enables it to be easily enabled using a CMake option, compiling the CUDA modules only if requested, and avoiding any direct dependency on them elsewhere in the code.

The kernels support two CUDA vector-widths: 1 and 32.

Notes

• In order to install CUDA operators at this stage, the Scotch library should be installed with double-precision support by including the flag -IDXSIZE64 in the file /cmake/ThirdPartyScotch.cmake. Therefore, it is recommended to request that Nektar++ compile Scotch, rather than using a system-installed version.
• If METIS is used, the #define REALTYPEWIDTH 32 should be manually changed to #define REALTYPEWIDTH 64 in the file /build/ThirdParty/metis-5.1.0/include/metis.h for double precision.