The AMDGPU Additions For Linux 4.7 Are Enormous
Last week was when AMD's Alex Deucher aligned the first PR of Radeon/AMDGPU updates for Linux 4.7. That pull request from last week has yet to be honored in DRM-Next, but will represent one of the biggest deltas of all the kernel driver updates for Linux 4.7. That pull request adds in around 60,000 lines of code and it still doesn't add the big DAL display abstraction layer.
The work sitting around since last week is for providing next-gen Polaris GPU support for AMDGPU, clean-ups for the VCE and UVD video encode/decode support, GPUVM improvements, scheduler improvements, clockgating and PowerPlay power management improvements, asynchronous page-flipping support, and many other changes. Polaris is the most dominating part of the Radeon/AMDGPU updates set for Linux 4.7, but aside from that soon-to-launch GPU support there are also many other improvements. The current display support of Polaris was described by Alex as "on par with other asics, for advanced features DAL is required."
This current Polaris support for Linux 4.7 includes power management, graphics, compute, SDMA, UVD, and VCE support so it's nice and well rounded ahead of the launch. Of course, it won't be until after Polaris ships that Linux 4.7 will be declared stable and then begin working its way into Linux distributions, but at least this Linux 4.7 DRM code should be available as a DKMS module via future AMDGPU PRO hybrid driver stack updates for employing on existing Linux kernel releases.
This AMDGPU support in Linux 4.7 doesn't include the imminent, experimental GCN 1.0 support and the GCN 1.1 / Sea Islands support remains experimental and disabled-by-default for this kernel version.
Today marks more AMDGPU fixes being queued up for Linux 4.7. The latest work to report on today is just a few hundred lines worth of changes and has the async page-flips for AMDGPU, UVD fixes for Polaris, many GPUVM fixes, code clean-ups, and other bug fixes.
Once this code has landed in DRM-Next or the mainline Linux kernel, I'll certainly be around with a plethora of our usual graphics benchmarks of this new code using a variety of GCN GPUs.