Replication of the nekRS microbenchmark "weightedInnerProdBenchmark" using CUDA.

The benchmark performs the following steps:

• Weighted dot product kernel reducing three input arrays over thread blocks
• Loop on host to reduce the Nblocks-sized array to a scalar value
• MPI_Allreduce to sum the scalar values across all ranks

Informations on how to run the bechmark are included in the slurm script submit.slm.