Xpra: Ticket #839: fast restride and xor in cython

Follow up from #465, we should provide an optional image wrapper function to restride and xor in one pass. (and maybe even do the byte swapping to RGB?) This would be used by the rgb_encode to speed things up significantly, especially when we can / should xor the whole image buffer (ie: when shadowing a display and using a non-video encoding).

Bonus points if we can take advantage of realloc to do things in-place. Though I doubt this would be useful for the common case: when using an xshm image, we may not be allowed to write to it? or if we are, it would require extra work to ensure it does not get freed until after the packet has been sent... (or we could just force compression to ensure the buffer is consumed before we free the image)

Also, it's a shame that we cannot feed the data to lz4 one image row at a time, or we could then even combine restriding, xor, byteswapping and compression in one pass!

Another idea that just came up: the ability to provide the target buffer as an optional argument could be used with encoders like nvenc to copy directly into their input buffer as part of the restriding.



Mon, 04 May 2015 05:42:58 GMT - Antoine Martin: status, description changed


Mon, 04 May 2015 09:50:30 GMT - Antoine Martin:

memoryview was enabled by default for 0.16 in r9223.

The restride is now done as an optional method on the image wrapper class, only implemented for the ximage wrapper (also used by xshm image wrapper): done in r9235.

Looking at the other codepaths, they're all either fairly uncommon or cheap enough already. The only case that may warrant more work is the shadow servers: those don't use the x11 image wrapper and call get_rgb_rawdata.


Wed, 16 Sep 2015 12:10:25 GMT - Antoine Martin: status changed; resolution set

X11 shadow servers now use xshm to capture pixels (see #899 and r10246) so they will take advantage of this new code too.

This will do for now.


Sat, 23 Jan 2021 05:07:32 GMT - migration script:

this ticket has been moved to: https://github.com/Xpra-org/xpra/issues/839