The GCC 10 Compiler Lands OpenMP / OpenACC Offloading To AMD Radeon GPUs

A few days ago I wrote about the OpenMP / OpenACC offloading patches for Radeon "GCN" GPUs being posted and seeking inclusion in the GCC 10 compiler that will be released in a few months. Those patches were successfully merged meaning this next annual update to the GNU Compiler Collection will feature initial OpenMP/OpenACC code offloading support to supported AMD GPU targets.

After GCC 9 only had the initial AMD Radeon GCN target in place, GCC 10 in early 2020 will feature the initial offloading support using the modern OpenMP and OpenACC APIs, thanks to the merges this week. The libgomp port and associated bits for the AMD GCN back-end have landed thanks to the work done by Code Sourcery under contract with AMD.

This opens up the offloading support to modern GPUs with the GCC compiler though is primarily for Vega and Polaris with no support in mainline GCC yet for Navi. The GCC 10.1 stable compiler release should be out in Q2'2020.

More details on the current offloading capabilities of the GCC compiler can be found via this Wiki page.

For those looking at GCC-based Radeon GPU offloading support right now, Code Sourcery / Mentor Graphics recently released a new Sourcery CodeBench Lite Edition with their very latest AMD GPU patches.