650D looks a lot different, but doing a register comparison in QEMU with a model from the same generation, but different resolution (wait a minute, can't find any) should help. I'll compare them and post the results in the thread. Test code will be: call the lossless compression on some dummy "image" (can be zeroed out, but needs to have the correct size - can be faked). If silent.mo works on the tested model, even better.
(image height must be modulo 8 for lossless compression on 6D?)
Levas - MagicLantern - 6D
With current test build (4Feb2018Build from Dfort) (which uses the same modulo restrictions as other builds and cams) you have 3 outcomes with lossless in MLV_lite on the 6d:
Option 1: It works (after blacklevel fix)
Option 2: You get frames with bottom half corrupted, all frames in the MLV have this, not just a few, all frames.
Option 3: You get normal sized MLV's, but you can't extract any dng's out of it. Also while recording, emptying the buffer is painfully slow when selected a resolution which gives option 3 outcome.
Haven't tested all possible resolutions, but did a quick test with the current build from Dfort.
1824 x 768 = good frames
1808 x 1016 = good frames
2544 x 960(5xzoom) = good frames
2528 x 960(5xzoom) = good frames
2512 x 960(5xzoom) = good frames
1824 x 1026 = Bottom half corrupted frames
2592 x 960 (5xzoom) = good
2560 x 960 (5xzoom) = MLV where no dng's come out with mlv_dump.
If I remember correct, restricting the 6d to only modulo 8, didn't help, outcome was more resolutions had the bad MLV's, where no dng's came out.