We don't use it much because we prefer opengl, and even opengl is disabled we usually have swscale available.
The code looks fine (similar to the opencl version).
It could just be that because we return
RGBX instead of
RGB, we end up in a different code path:
paint_rgb32 instead of
trunk fix in r9983: I'm just not clever enough to figure out how to derive the byte order, so we hardcode it instead (big endian not tested)
This may need further work, as the same swapped colours occur with this fix applied (but with different library versions): #937.
this ticket has been moved to: https://github.com/Xpra-org/xpra/issues/922