Vega 12/20 Added To AMDGPU LLVM, Confirms New GCN Deep Learning Instructions For Vega 20
The support was added to LLVM and Clang though don't shed too much light on these yet-to-be-launched GPUs, but does confirm deep learning instructions present for Vega 20. In fact, it's the addition of these instructions that are making the commit rather larger.
Over the past month the Vega 12 GPU support has appeared for Mesa 18.1 RadeonSI/RADV and the AMDGPU DRM driver will support it with Linux 4.17. That support is now being rounded off with the necessary AMDGPU compiler back-end support with mainline LLVM for the LLVM 7.0 release coming later this year.
The original Vega 10. The Vega Linux open-source driver support continues to improve.
Not much is unfortunately confirmed for Vega 12 besides being that it's a new part, not any Intel Kabylake G part, and the LLVM patch does confirm it is indeed a discrete GPU. Among the speculation is that Vega 12 could be a lower-end Vega part for succeeding the Radeon RX 500 "Polaris" graphics cards.
Vega 20 meanwhile is a product AMD has teased and mentioned is a 7nm GPU with 32GB of HBM2 memory. The Vega 20 is aimed for machine learning / artificial intelligence workloads albeit not yet launched. So it's not too surprising these patches do confirm new deep learning GPU instructions being present for Vega 20. The deep learning intrinsics added are fdot2, sdot2, udot2, sdot4, udot4, sdot8, and udot8.
But that's about all these new Vega 12/20 patches for LLVM/Clang reveal today that are now mainlined.