March 30, 2023
7:45 a.m.
On Wed Mar 29 09:40:55 2023 +0000, eric pouech wrote: > > Fwiw I don't think such low differences matter very much. I think it's > likely some diff in color sampling, and I think windows does a bad job > at it with a lot of color bleeding. There's already some differences > between windows versions in some tests. > I just want to be sure of that... > Digging just a bit more shows two interesting effets: > - it seems that gstreamer applies some smoothing effet in color > transition: the expected output has > clear transitions between bands while gstreamer is smoother > - (ascii dump of rgb colors, one char <= one different rgb color) > - expect: > ``` > line 58: $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ > line 59: $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ > line 60: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% > line 60: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% > ``` > - gstreamer: > ``` > line 58: ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;< > line 59: =================================================================================> > line 60: ?????????????????????????????????????????????????????????????????????????????????@ > line 61: AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAB > ``` > - note also the elements on end of line (also smoothing with vertical > band of test sample) > that's strange as yuv stores its information on a 2x2 rectangle, and all > the bands in source are expected to be aligned on this 2x2 thingie, so > either there's a misalignment of 1 somewhere (but would be on both > directions), or winegstreamer adds some interpolation in the process > looking at the diff of colors (picking same pixel in the middle of each > band) shows: > ``` > band 0: cdcdcd <> ff7bff > band 1: ff0000 <> fd0000 > band 2: 0000fe <> 0000fd > band 3: fe00ff <> fc00fd > band 4: 01ff00 <> 00fa00 > band 5: ffff00 <> fdfc00 > band 6: 00ffff <> 00fbfb > band 7: ffffff <> fdfdfd > ``` > diff:s for band 1 to 7 are very small, but band 0 is really more questionnable. > do these ring any bell? Yes, I think that's what I was seeing as well. There's also the same kind of differences in some other tests between windows version. Of course it'd be nice to be pixel perfect but I don't think it matters very much. It's also possibly related to the very small video frame size, or at least, it makes it more visible. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/2471#note_28404