Page 1 of 2
1.7.426 mask select stars not working
Posted: Sat Oct 10, 2020 12:24 pm
by alacant
Hi Ivo, everyone
StarTools-Linux64-GPU opencl via beignet on lubuntu 20.04
mask - auto - stars produces the attached image.
similar result (only part of the image selected) with turning off tracking Track - grain removal
beignet error:
Beignet: "Exec event 0xfc0e50 error, type is 4592 (repeated hundreds of times).
-
- ss3.jpg (240.79 KiB) Viewed 4551 times
Re: 1.7.426 mask select stars not working
Posted: Sun Oct 11, 2020 1:30 pm
by happy-kat
auto star mask working ok for me using windows flavours.
Re: 1.7.426 mask select stars not working
Posted: Mon Oct 12, 2020 2:44 am
by admin
Hmmm...
If this is a case of the load being to much for the iGPU to handle before the driver triggers a timeout, I just release a new version 1.7.427 that may help with this in some edge cases. I'm worried Beignet still isn't playing ball.
How big was the image (X * Y resolution)?
Thank you,
Re: 1.7.426 mask select stars not working
Posted: Mon Oct 12, 2020 11:30 am
by alacant
Hi
Same mask error with 427 but no beignet messages.
Image size 2571 x 1804
Cheers
Re: 1.7.426 mask select stars not working
Posted: Mon Oct 12, 2020 12:14 pm
by admin
alacant wrote: ↑Mon Oct 12, 2020 11:30 am
Hi
Same mask error with 427 but no beignet messages.
Image size 2571 x 1804
Cheers
I was just able to replicate this on an i5 5300U, also running Beignet. The kernel ring buffer (dmesg) showed this;
[ 3402.936844] [drm] stuck on render ring
[ 3402.937147] [drm] GPU HANG: ecode 8:0:0x8ed1fff2, in StarTools-Linux [8232], reason: Ring hung, action: reset
[ 3402.937149] [drm] GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.
[ 3402.937150] [drm] Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel
[ 3402.937151] [drm] drm/i915 developers can then reassign to the right component if it's not a kernel issue.
[ 3402.937152] [drm] The gpu crash dump is required to analyze gpu hangs, so please always attach it.
[ 3402.937153] [drm] GPU crash dump saved to /sys/class/drm/card0/error
[ 3402.939322] drm/i915: Resetting chip after gpu hang
[ 3414.937655] [drm] stuck on render ring
[ 3414.938016] [drm] GPU HANG: ecode 8:0:0x8ed9fff2, in StarTools-Linux [8250], reason: Ring hung, action: reset
[ 3414.940179] drm/i915: Resetting chip after gpu hang
I can replicate this if I really stress the system, playing a YouTube video in tandem with processing in StarTools.
Googling this problem seem to point to issues with the Intel drivers in newer Linux kernels.
Does your dmesg show something similar?
Re: 1.7.426 mask select stars not working
Posted: Mon Oct 12, 2020 12:20 pm
by admin
Also, I have been reading that changing acceleration method from SNA to UXA is supposed to help....
Re: 1.7.426 mask select stars not working
Posted: Mon Oct 12, 2020 3:00 pm
by alacant
changing acceleration method from SNA to UXA
How would I do that?
I've had to go back to non GPU for the current session but I'll get the dmesg from the GPU version.
Cheers
Re: 1.7.426 mask select stars not working
Posted: Tue Oct 13, 2020 12:13 am
by admin
It's a Xorg.conf setting
This should be achievable by creating a file named
20-intel.conf in /usr/share/X11/xorg.conf.d/ (for us Linux Mint / Ubuntu users)
In it, you will want to put;
Code: Select all
Section "Device"
Identifier "Intel Graphics"
Driver "intel"
Option "AccelMethod" "uxa"
EndSection
Afterwards you should
no longer see any mention of SNA being used when doing;
Code: Select all
cat /var/log/Xorg.0.log | grep 'intel'
Re: 1.7.426 mask select stars not working
Posted: Tue Oct 13, 2020 12:26 am
by admin
I just processed a 2283x1685 image successfully with those changes on the i5-5300U's iGPU, but as it seems to be an issue with flaky drivers, I can't guarantee that switching to UXA stabilises them without more rigorous testing...
Do let me know if this works for you!
Re: 1.7.426 mask select stars not working
Posted: Tue Oct 13, 2020 12:35 pm
by admin
Another update; after some more testing on the i5-5300U, I'm definitely seeing a marked improvement in iGPU stability. This is looking promising!