two-level parallel version of transpose GEMV

[iyamazaki]

No description provided.

[mhoemmen]

Why do we need to write an optimized GEMV? Why can't we just call cuBLAS for CUDA and the BLAS for non-CUDA, and have a slow fall-back for unsupported types?

[mhoemmen]

Prefer using new_type = old_type; C++11 type alias syntax to typedef old_type new_type;.

[ndellingwood]

Cross-referencing #443
@mhoemmen I believe the intention of the PR is to resolve issues with the fallback impl in transpose mode, the 2-level parallelism approach seemed a natural way to implement.

[mhoemmen]

@ndellingwood wrote:

I believe the intention of the PR is to resolve issues with the fallback impl in transpose mode, the 2-level parallelism approach seemed a natural way to implement.

Cool, I'm OK with it as long as it's fixing a bug.

[mhoemmen]

Looks OK. The usual kokkos-kernels idiom that Christian has been promoting, is to reduce the number of instantiations of the functor, by having the function that calls the functor assign the input Views to "canonical" View types. For example, the (outer) function might be templated on ViewType, but if the functor really only needs View<const T*, AnonymousSpace>, it would be best to have the functor be templated only on T.

[iyamazaki]

Thank you for the comments @mhoemmen. I am running the spot-check on white, but it is taking a long time @ndellingwood.

[ndellingwood]

@iyamazaki spot-check on kokkos-dev-2 passed, merging in. Thanks!

Running on machine: kokkos-dev-2
Going to test compilers:  gcc/7.3.0 gcc/9.1 intel/18.0.5 clang/8.0 cuda/10.1
Testing compiler gcc/7.3.0
Testing compiler gcc/9.1
  Starting job gcc-7.3.0-OpenMP-release
  Starting job gcc-7.3.0-Pthread-release
  PASSED gcc-7.3.0-OpenMP-release
  Starting job gcc-9.1-OpenMP-release
  PASSED gcc-7.3.0-Pthread-release
Testing compiler intel/18.0.5
  Starting job gcc-9.1-Serial-release
  PASSED gcc-9.1-OpenMP-release
Testing compiler clang/8.0
  Starting job intel-18.0.5-OpenMP-release
  PASSED gcc-9.1-Serial-release
  Starting job clang-8.0-Cuda_OpenMP-release
  PASSED intel-18.0.5-OpenMP-release
Testing compiler cuda/10.1
  Starting job clang-8.0-Pthread_Serial-release
  PASSED clang-8.0-Cuda_OpenMP-release
  PASSED clang-8.0-Pthread_Serial-release
  Starting job cuda-10.1-Cuda_OpenMP-release
  PASSED cuda-10.1-Cuda_OpenMP-release
#######################################################
PASSED TESTS
#######################################################
clang-8.0-Cuda_OpenMP-release build_time=567 run_time=712
clang-8.0-Pthread_Serial-release build_time=356 run_time=883
cuda-10.1-Cuda_OpenMP-release build_time=626 run_time=660
gcc-7.3.0-OpenMP-release build_time=218 run_time=264
gcc-7.3.0-Pthread-release build_time=203 run_time=501
gcc-9.1-OpenMP-release build_time=186 run_time=235
gcc-9.1-Serial-release build_time=157 run_time=486
intel-18.0.5-OpenMP-release build_time=301 run_time=292
#######################################################