As per our discussion last Friday, we would like to create a reference CUDA version w/o libraries. I asked whether that should be CUDA C or CUDA Fortran, and the answer was CUDA C. Okay, in this case I'd like to start with creating C++ variant of program, that later could be used to create CUDA version. C++ implementation closely mimics Fortran: modules -> namespaces, (i,j) -> [j][i], array zeroing -> memset, etc. I compared results for "128 128 10000 1" - they are not identical, but very similar.
Hi,
As per our discussion last Friday, we would like to create a reference CUDA version w/o libraries. I asked whether that should be CUDA C or CUDA Fortran, and the answer was CUDA C. Okay, in this case I'd like to start with creating C++ variant of program, that later could be used to create CUDA version. C++ implementation closely mimics Fortran: modules -> namespaces, (i,j) -> [j][i], array zeroing -> memset, etc. I compared results for "128 128 10000 1" - they are not identical, but very similar.