GCC Is Potentially Months From Seeing Radeon OpenMP 4.5 / OpenACC 2.6 GPU Offloading

[phoronix]

Phoronix: GCC Is Potentially Months From Seeing Radeon OpenMP 4.5 / OpenACC 2.6 GPU Offloading

At last month's GNU Tools Cauldron was an update on the Radeon GCN back-end state for the GCC compiler, which is likely to see more code land around year's end...

http://www.phoronix.com/scan.php?page=news_item&px=Radeon-GCC-GCN-Port-2019-State

[Jabberwocky]

I'm guessing latency or real-time applications are not the top priority at the moment. I wonder how latency compares to running OpenMP tasks on the CPU vs GPU offloading. I've always wanted to investigate using GPUs for large scale real-time physics processing that works on Nvidia and AMD without using OpenCL.

Thanks to Thomas Schwinge for presenting Andrew Stubbs' work.

[artivision]

Well i don't believe that we will have any distro builds with offloading bytecode on their core. Probably those who develop those stuff are short minded and they cannot develop something that will automatically build accelerated bytecode from any source. So we will have to wait for someone to support this on his app by spending thousands of man-hours. Waiting for ever that is.

[elldekaa]

What about clang? Anyone from AMD working on supporting OpenMP GPU offloading for Radeon on upstream repository? I saw some experimental implementation but they seemed quite dead... libomptarget and nvidia offloading are quite in good shape (not yet the performance of native cuda implementation but things are getting better).