Kihiro Bando

Here is my baseline version ```cpp using DeviceType = Kokkos::Cuda; using ExecutionSpaceType = DeviceType::execution_space; using RangePolicyType = Kokkos::RangePolicy< ExecutionSpaceType, Kokkos::Schedule >; using Kokkos1DView = Kokkos::View< double *, Kokkos::LayoutRight, DeviceType::memory_space>; using...

For the tests I have been running, ns=5, nb=(p+1)^3 for p=2,3,4 and ne=64000. The way to interpret this is in a finite element like discretization, I have ns solution variables,...

@vqd8a some things I tried. You mentioned the use of `TeamVectorGemm`. I assume you referred to just changing `TeamGemm` into `TeamVectorGemm`. I looked at the examples in the wiki and...

waouh, thanks a lot @vqd8a and @e10harvey! @e10harvey yes data layout is certainly important. The different layout for the view of right matrices comes from the optimization of other kernels...