Tpetra::MultiVector not using cuBLAS for GEMM, making Belos slow
Created by: mhoemmen
@trilinos/tpetra @trilinos/belos @trilinos/kokkos-kernels @vbrunini
This is a Trilinos mirror of the following kokkos-kernels issue: https://github.com/kokkos/kokkos-kernels/issues/397 . Once we verify the proposed fix, be sure to patch Trilinos and submit a fix to kokkos-kernels.