OpenACC & XEON-PHI

Hello

We have a Fortran-Program with OpenACC-statements. Currently we create NVIDIA-Code.

Question:
Can we create XEON-PHI code with this or do we have to add another parallelization technique?

(Sorry if this question is a recurrence. But as far as I can see there a some contradictory statements on this issue… :-)

Thank you
Benedikt

Hi Benedikt,

Can we create XEON-PHI code with this or do we have to add another parallelization technique?

OpenACC can be used to target Xeon-Phi, however we (PGI) don’t support it as of yet. We had a prototype version working awhile ago for Knights Corner but Intel told us that they we’re redesigning the Phi so we decided not to productize it. Our current plan is to target x86 multi-core, then add support for AVX 512 which should then allow us to support the future Knights Landing product. No firm timeline however.

Hope this helps,
Mat

Thanks for clarification, Mat! Seems to be a complex subject. :-)
Benedikt