No idea when it broke, but it isn't doing anything anymore, just giving a blank memory area. I was hoping to see if we can switch to using read-only memory for our pixel data, but since I can't test at all... this will have to wait.
I have updated to pyopencl version 2015.1, maybe this has something to do with it.
I am tempted to just remove the whole thing, we can resurrect it if we need to.
if the kernel was actually doing something, I could tell if this is worth applying or not
some tweaks to see if the queuing is part of the problem handle memoryviews, etc..
example code that does work
use the image interface instead of buffers, more like the working example
reinit before running to ensure we get a clean context, and avoid running selftest (different thread)
Some minor fixes in r10024, but despite all the attempts in the patches above, no clue as to what is broken.
Note: if / when we do fix it, r10058 uses read-only buffers so we may have to require a sufficiently new version of opencl to avoid errors.
Well, I have no idea what happened here, but it fixed itself after a reboot...
this ticket has been moved to: https://github.com/Xpra-org/xpra/issues/925