Half-precision in OpenACC

Hi,
I could not find any information on this topic:

is it possible to use half-precision floating point in OpenACC? And if it is the case, how?


Thanks,

Enrico

is it possible to use half-precision floating point in OpenACC?

No, sorry. You’re the first person to request it. Would it be something you want me to add as an RFE (request for enhancement)?

  • Mat

If it is possible: yes… it would be a very interesting feature for us.

Thanks and Best Regards,

Enrico

I added TPR#23044 “OpenACC: support half-precision in device code” with your request.

  • Mat