Freezes GNOME shell and then restarts back to GDM

[arnxxau]

first it displays

Took 28.561646ms to draw frame. 
Completed first meaningful paint

then it freezes

I have tried single image mode and zip mode and they freeze my system.
I even disabled all the up-scaling with '0x0' target res.

waifu2x-ncnn-vulkan works on my system too

I have an AMD 6700XT and I'm running arch linux

[awused]

That's odd. I'll definitely need more information, but it is most likely an underlying gnome or AMD bug if it's breaking something so badly that you get booted back to GDM.

There should be logs somewhere, whether under dmesg or systemd or whatever you happen to be running.

One thing is to try building it without opencl (--default-features false), or removing this line https://github.com/awused/aw-man/blob/master/src/main.rs#L56 that fixes a gtk regression, but neither of those should cause your desktop environment to crash.

Does it reproduce under other desktop environments? Gnome tends to be unusual and unstable.

[arnxxau]

Hi, I disabled opencl and it works now.
I also tried to use the opencl version with kde but it crashes too
These are the dmesg and journalctl logs if you want to have a look

dmesg.txt
journalctl.txt

[awused]

From reading other bug reports for other issues, it seems that "Failed to initialize parser" is a generic error on GPU reset that doesn't point me to anything specific aw-man is doing. There might have been an error somewhere earlier in the logs, but without hardware on my end to test it'll be hard for me to do anything unless there's a clear "error at line X while parsing resample.cl" or "can't do Y" in the logs.

Even if my opencl code is invalid (which I don't believe it is, it's fairly simple), the worst that should happen is aw-man itself crashing. It looks to me like there's an underlying AMD or mesa bug, probably mesa, so you might consider opening a bug report against mesa and mentioning that this application causes it so they can try to debug it. I know the HD 7xxx series of AMD GPUs had some opencl teething issues, but I don't think there was anything for the 6xxx series.

Setting gpu_vram_limit_gb to 0 in the config should also work to disable opencl at run time in case you ever forget to disable it at build time.

Running aw-man --show-gpus when built with opencl might also be interesting, and if it shows more than one Device you could try forcing it to use the alternate ones with the prefix line in the config. But I wouldn't expect that to make a difference to a GPU reset.