[Trilinos-Users] [EXTERNAL] Trilinos-Users Digest, Vol 137, Issue 4

[Hoemmen, Mark]

On 1/23/17, 4:00 AM, "Trilinos-Users on behalf of trilinos-users-request at trilinos.org" <trilinos-users-bounces at trilinos.org on behalf of trilinos-users-request at trilinos.org> wrote:
>Date: Sat, 21 Jan 2017 02:33:55 -0700
>From: Ashesh Chattopadhyay <ashesh6810 at gmail.com>
>To: Trilinos-Users at trilinos.org
>Subject: [Trilinos-Users] Trilinos-Kokkos with CUDA 7.0 ,	issues with C++ lambdas
>
>Dear All,
>
>I know that after and including CUDA 7.5 an experimental feature to pass
>lambdas from host to device was provided through the options
>--expt-extended-lambda
>
>and I can use it with -DKOKKOS_CUDA_USE_LAMBDA to compile Kokkos code with
>execution space as CUDA, however, I am being forced to use CUDA 7.0
>currently and I see that Kokkos can be used with CUDA 7.0 but the nvcc
>compiler does not recognize the --expt-extended-lambda option and without
>that I get a compiler error at the lambda. Can you please advise me as to
>how I can use CUDA 7.0 to compile and run Kokkos code.

If you don’t have CUDA >= 7.5, you cannot use lambdas in Kokkos::parallel_*.  Instead, you must write functors.  These are classes with state and an operator() that implements what the lambda would implement.  Please refer to the Kokkos tutorial for examples.

mfh