With the upcoming Linux kernel 4.9 release in the arch repos, the flag was
set to enable building AMDGPU support in Southern Islands (SI - GCN 1.0) by
default. The weird thing though is that Sea Islands (CIK - GCN 1.1) support
is still not enabled, even though the upstream support for CIK has existed
since 4.6 and SI has only just been marked as experimental in 4.9.
Why did we flip the flag for SI and still keep it disabled for CIK?
Hopefully it was just an oversight, as I would love to see support for
building the AMDGPU kernel module support for both of these architectures
by default.
It seems weird to push a kernel that has AMDGPU support for GCN 1.0 and
1.2, while leaving GCN 1.1 still in the dark.
Lastly, some may argue that as long as it is still considered experimental
by upstream, it makes good sense to keep from building support by default.
However, since the Radeon kernel driver takes precedence over AMDGPU for
both SI and CIK architectures, in order to use the AMDGPU kernel driver the
user would have to explicitely blacklist the Radeon kernel module. Thus,
there should not really be a scenary where a user is running an
experimental driver without their informed consent.
The only thing garnered by keeping the CIK module disabled at build time is
that anyone with a 2012-2015 AMD graphics card needs to build their own
kernel to test out the arguable "better" driver, thus gaining TONS of
benefits, least of which is Vulkan support.
Thanks for reading!
-Britt