DCTFilter (renewed)

[HolyWu]

https://github.com/HomeOfVapourSynthEvolution/VapourSynth-DCTFilter

Well, the existing DCTFilter is *a bit* slow, therefore here comes the new one. I simply use the DCT/IDCT transforms from FFTW and nothing fancy. The implementation of DCT/IDCT in FFTW is inefficient, hence it's still slower than AviSynth's equivalent, but at least 3~7 times faster than the old one.

[AzraelNewtype]

Just did a quick test with my fairly heavy processing script to see how it fares with just this switch (and changing the dct call in Deblock_QED since you kept the old filename but not the old namespace/function name)


Old
vspipe64.exe -s 2000 -e 9999 -p 42.vpy .
Output 8000 frames in 559.36 seconds (14.30 fps)

New
vspipe64.exe -s 2000 -e 9999 -p 42.vpy .
Output 8000 frames in 483.02 seconds (16.56 fps)


And this is in a script that's leaning into jivtc, occasionally ivtc_txt60mc, and smdegrain, so it's not like autodeblock (which calls Deblock_QED inside) was the only slow bit. Even out of isolation that's fairly impressive.