16-bit float support for PASCAL


Does anyone know if PGI Community Edition compiler supports the lattest 16-bit float for NVIDIA PASCAL archtecture ?

If so, do you know how can I use it, i.e. which files should I include, which types should I use ?

16-bit float should improve performance a lot by losing a little bit of precision and I would like to explore this fact

Thank you !

Hi tlgimenes,

No, sorry. We don’t yet support half-precision. We have had one other request for this feature (logged as TPR#23044) and I have added your post to this request.

  • Mat