OpenACC parallelism with Fortran array operations

Accelerated Computing HPC Compilers nvc, nvc++ and nvfortran
1

Hello, I was reading the OpenACC Programming and Best Practices guide, and the guide includes a comment that says this:

Fortran programmers should also note that an OpenACC compiler will parallelize Fortran array syntax that is contained in a kernels construct. When using parallel instead, it will be necessary to explicitly introduce loops over the elements of the arrays.

Is this true for the nvfortran compiler as well?

2

Yes, we follow the standard here.

With “kernels”, the compiler can discover parallelism so the implied do loop of array syntax can be implicitly parallelized.

With “parallel”, the programmer is telling the compiler where to parallelism, so can only be used on explicit loops.