Sfdwhf

I need someone's help in getting my amdgpu driver working properly.

Long time Ubuntu user, but I'm not technical. I have been wrecking my brain for the past 2 days, searching through an endless web.

Long story short: I recently swapped my video card from an AMD HD-6950 (radeon) to an AMD Vega 64 (amdgpu). I really should have been better about backups...

Ubuntu 18.04 + HWE (low-latency kernel 14.18.06) + padoka-stable.

This machine is more of a media workstation, (music production, image editing, video editing, etc.), but sometimes gaming.

I first physically installed the new card (with no software changes), and it seemed to perform worse. Due to the nature of the work, I decided to install amdgpu-pro drivers. They initially failed, and the system would not boot (black screen after grub). Reinstalled under kernel 4.15, and it worked.

However, the performance was fairly poor, and some applications failed (eg. Kdenlive's video preview was all gray). So I decided to uninstall amdgpu-pro, using AMD's uninstall utility.

But 4.18 was still borked. I have no idea what I did. Eventually, I had lots of accidental uninstalls and reinstalls; but at some point, I got it working with great gaming performance, using the amdgpu drivers (confirmed by inxi). I think this was the result of a dpkg-reconfigure xserver-xorg.

However, because I wanted (in particular) openCL, and openCL was not included in amdgpu (confirmed in darktable), I decided to try Rocm, outlined here:
https://rocm.github.io/ROCmInstall.html

Broke the system again--and seemingly worse.

Now after a lot of fiddling, I can boot fine in any kernel (without requiring nomodeset); but inxi reports that it is using the fbdev & ati driver rather than amdgpu (as it was showing before).

I've poured through logs, made config changes, alternated packages, etc. and I cannot figure this out.

Here's where I am today:
-I can run & boot any kernel (4.15, 4.18, or 4.20)

GRUB_CMDLINE_LINUX="radeon.si_support=0 radeon.cik_support=0 amdgpu.si_support=1 amdgpu.cik_support=1"

(^ this worked before. I've also tried amdgpu.dc=0, but that results in blackscreen hang).

/etc/X11/xorg.conf.d/20-amdgpu.conf:

Section "Device"

    Identifier "AMDGPU"

    Driver "amdgpu"

    Option "AccelMethod" "glamor"

    Option "DRI" "3" 

    Option "TearFree" "true"

Endsection

^ This also worked before, and it appears to be picked up fine, according to the xorg logs.

lspci shows:

Subsystem: Sapphire Technology Limited Vega 10 XT [Radeon RX Vega 64]

Kernel driver in use: amdgpu

Kernel modules: amdgpu

But...the results of inxi -GC:

Display Server: x11 (X.Org 1.20.1 )

drivers: fbdev,ati (unloaded: modesetting,vesa,radeon,amdgpu)

OpenGL: renderer: Radeon RX Vega (VEGA10, DRM 3.27.0, 4.20.17-042017-lowlatency, LLVM 8.0.0)

version: 4.5 Mesa 19.0.1 - padoka PPA

Previously, this ^ showed the driver as being amdgpu, rather than fbdev,ati. And performance was noticeably different.

Also worth noting, dmesg doesn't appear to show any amdgpu unloading any more (though it previously did), and my xorg logs in /var/log appear not to be updating. The ones in my ~ do appear to be updating though.

Also, I think I'm now on the hwe-18.04 packages (example for xserver-xorg, etc.)

Can someone help me get to a stable system, that's completely using amdgpu? And ideally some hardware acceleration, including OpenCL? And ideally, open source / repository so that future software update will seamlessly take care of further hardware acceleration?

I hope I am just an idiot and there is a simple fix or reinstall.

What information can I provide?

I am very, very thankful for any help you can give me! This has really been stressing me out!