Compose module crash

[LuckyEddie]

Hi, I have a problem where loading some files in the Compose module causes a full crash but the same files can be successfully loaded using the Open command. I thought that 'Open' used the Compose module in the background but something different is clearly happening here.

Here is an example of a file that causes this behaviour. The file was saved from Siril as 32-bit fit.
https://drive.google.com/file/d/1K3nkRI ... sp=sharing

ST 1.7.461 GPU on Linux 64-bit (also tried previous release with the same result).

Thanks,
Ed

[admin]

Thanks Ed,

I have no idea what the issue might be. I cannot replicate the problem (not on Linux, WIndows or macOS), and besides a sporadic "hang"/"freeze" issue, I have not had any other reports of this nature.

What are the specs of your machine? Might you be running out of storage space and/or memory? Have you got some sort of swap space configured?

Is there a specific sequence that always triggers the issue?

(as a completely unrelated sidenote; please avoid drizzling if not strictly necessary!)

[LuckyEddie]

The machine is a 2013 MacBook Pro, running Linux Mint 20.3, 16GB & 16GB swap file, Nvidia GT650M.
The issue isn't permanent. Sometimes I can reboot, run as normal and not have a problem, other times it hits. Once the problem has occurred, it is consistently there until I reboot the machine, closing and restarting ST makes no difference, and 'Open' never fails to load the file. I managed to get it to misbehave again a few minutes ago and having launched from a terminal (I usually run ST from a launcher) got this on stdout:

Error during sending the image data to the OpenCL buffer -4

I then ran clinfo, the output of which can be found at:
https://drive.google.com/file/d/1LonnGG ... sp=sharing

I bet it's another 'Nvidia event' but it's very odd that 'Open' never fails.

[admin]

Thank you for the update Ed - much appreciated.
It appears your GPU solution is running out of memory.

It's not so much the way the dataset is opened, it is the processing/analysis that is performed once it has been loaded (which is different and more arduous in the Compose module, and which is done on the GPU).

Thanks to your helpful clinfo output, we can see your GPU solution only has 1GB on board, which is shared with other applications and the window compositor. You may - possibly - find that processing a smaller dataset will not give you that error as quickly. Installing a package like PSensor may give you a convenient overview of GPU memory usage, so you can keep an eye on that.

All that said. The NVidia drivers on Linux are... poor. For example, there is also another quirk/bug that I run into daily with my NVidia cards on linux, which is that if I put my computer to sleep while StarTools is open, the driver stops functioning and I have to completely reboot before I can launch/use StarTools again. The driver stack completely forgets there is an OpenCL app running.

[LuckyEddie]

Hi Ivo,
That makes complete sense, thanks.
For anyone else who comes looking in similar circumstances, if you're running Nvidia under Linux then:
1. Using nvidia-smi at the terminal gives some useful usage info for some GPUs.
2. I feel your pain.