Inquiry about processor suboptions for -tp

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

Hello,
I’ve built nvhpc(20.11) on a CascadeLake system. Checking some systems parameters, it’s still using “-tp skylake” as the default setup. the reference guide (pp 55-56) doesn’t list neither CascadeLake nor any Aarch64 processor. This brings 2 questions:

  1. Is there any specific option (or flag) for CascadeLake or should CascadeLake systems be treated as Skylake?
  2. Which -tp option would work for Aarch64 processors? Obviously, this doesn’t apply to the system I’m working on but it’s related.
    Thanks.
2

No. The core architecture of Cascade Lake is largely the same as Skylake so from the compiler perspective, there’s no difference. Cascade Lake did introduce the AVX512 VNNI instruction set but is largely for use with AI, so not something we support as of yet.

Which -tp option would work for Aarch64 processors?

Right now, we don’t differentiate between different ARM architectures so the only option to “-tp” is “host”. (There’s “native” as well but it’s just an alias to “host”).

1 Like
3

Thank you for the info.