AMD GCN OpenMP/OpenACC Offloading Patches For The GCC 10 Compiler

Over the past year Code Sourcery / Mentor Graphics has been working extensively on the new AMD Radeon "GCN" back-end for the GCC code compiler. With the code that is found in GCC 9 and up to now in GCC 10 hasn't supported OpenMP/OpenACC parallel programming interfaces but that could soon change with patches under review.

The Radeon GPU support in GCC up to now hasn't supported OpenMP or OpenACC for offloading to the graphics processor and thus its practicality has been limited.

We've seen the Radeon back-end for GCC improved this year with various additions/improvements while Code Sourcery's goal has been to deliver OpenMP/OpenACC AMD GCN support for GCC 10.

Time is running out on getting this functionality into GCC 10 with feature development soon ending, but the latest patches were sent out on Tuesday. Seven patches amounting to over six thousand lines of new code were sent out for providing the basic OpenMP/OpenACC offloading support to AMD GCN GPUs.

Andrew Stubbs of Code Sourcery mentioned, "The test results are not yet perfect, but there are many more passes than failures, so this is a good starting point. The rest of the issues can be addressed as bugs during [later stage development of GCC 10]."

We'll see in the coming days if this Radeon OpenMP/OpenACC GPU offloading will make it for GCC 10, which will be out as stable in Q2'2020.