GCC 13 Compiler Finally Adds Support For AMD GFX90A "Aldebaran"

It was over a year ago that AMD initially added the "GFX90A" target to their LLVM AMDGPU compiler back-end while now this week added to the GNU Compiler Collection for the GCC 13 release not due out until next year is its GFX90A support for the GNU toolchain.

GFX90A is what was then enabled under Linux with the "Aldebaran" codename and then formally launched last November as the AMD Instinct MI200 series.

The AMDGPU LLVM support remains AMD's primary focus given that it's used from their ROCm compute stack through to their AMDVLK and RadeonSI graphics drivers and elsewhere. Meanwhile the past few years CodeSourcery / Mentor Graphics has been working on the "AMDGCN" back-end within the GCC compiler for those wanting to use the GNU compiler toolchain and allow OpenMP/OpenACC offloading to AMD GPUs.

CodeSourcery, which is now part of Siemens, has been the principal driver of the AMD GPU offload support for GCC. With AMD's official compute stack and their own engineering work has been focused on their AOMP compiler that is based on LLVM for GPU offloading. AMD for their part is also already enabling "GFX940" as next-gen CDNA within LLVM and the rest of the AMD Linux graphics stack for what will presumably be the AMD Instinct MI300 series launching later this year.

So as of yesterday, GCC 13 Git has GFX90A support for supporting the Aldebaran / MI200 series. The AMDGCN target in GCC now supports Fiji, Vega 10, Vega 20, GFX908, and GFX90A as available targets. RDNA/RDNA2 support is not available with this GCC implementation and hasn't been a focus for CodeSourcery or other GCC contributors.