RADV Gets Patches For Float64 Support, Now In Mesa Git

[Sevard]

Pretty sure they've said in the past they are aiming to get the open drivers support on launch, but at this point it's pretty clear it won't be upstream in time.

Why not? Vega is Q2 product, so they still have some time to merge code.

[bridgman]

when vega comes do you plan to have both open and AMDGPU PRO options available at launch day?

Yes, we plan to have both options available at launch day, plus a ROCM stack.

[ariel]

Yes, we plan to have both options available at launch day, plus a ROCM stack.

Will the ROCM stack have a working openCL stack by then? how long do we need to wait...

I recently added an RX360 to my fanless skylake workstation to improve the photography workflow - hoping that it would speed up darktable's raw processing. amdgpu-pro --compute installed nicely in one try... only for the DKMS build (and amdgpu-pro opencl) to break after the first kernel update. This is on Fedora 25.
While it worked, it was awesome (500% faster than CPU / core i5) but using the binary driver reminds of the old days of linux, with DKMS being so flaky (or the amdgpu-pro driver recent kernel support being flaky, not sure which). Choice being going back to an older, less-secure kernel OR being able to use amdgpu-pro opencl.
We sorely need a true open source driver properly maintained so that it works with the most popular distros/kernels out there. No more DKMS please. Compatibility with only some ancient RHEL or ubuntu kernel helps some but doesn't really cut it.

[smitty3268]

Why not? Vega is Q2 product, so they still have some time to merge code.

It will likely launch while linux 4.12 is out, and they've already missed 4.11.

I'd love to be wrong, but in looking at the DAL situation I really don't see it being accepted upstream in just 1 more version. They're almost certainly going to have to submit at least 1 more time, then fix up issues that the kernel maintainers bring up, and that kind of turnaround isn't quick.

Also, I think bridgman pretty much confirmed in another thread that they're now mostly targeting just having the code out there on launch, and not targeting it for upstream inclusion yet, which pretty much validates my opinion.