Closed Kogarou-san closed 2 years ago
Oh so great, wasn't aware he finally ported. Will test and implement for the next week.
Done. Hopefully without any issues. Just keep in mind that nnedi3 and NNEDI3Cl don't output the same mathematical output.
Maybe adding a way to use CPU as an option or as a fallback might be useful, for those unwilling/unable to use oCL (unless nnedi3cl on the CPU is as fast as nnedi3, which I doubt) And just to be sure, did QTGMC also get updated to be able to use nnedi3cl as an deinterlacer?
Yes QTGMC+ is updated already with NNEDI3CL but bottleneck is elsewhere so not much of an improvement. Does NNEDI3CL have an option for CPU? I didn't see it... Integrating both options at the same time is a pain, unless I make a wrapper for both.
The ability to offload some of the compute to a GPU is still beneficial since you can spend those now free CPU cycles elsewhere, even when accounting for the overhead
Does NNEDI3CL have an option for CPU? I didn't see it...
It does not, unless you'd set it to run the oCL kernel on the CPU, but that's slower and defeats the purpose
Integrating both options at the same time is a pain, unless I make a wrapper for both.
A wrapper might come in handy, there isn't much difference in them besides no built-in rpow2 for CL, less prescreener options, and possibly different defaults, as far as I know
Check here: 47b055858eed471b5cdc7ae85939ddb0d70f907b Also in SMDegrain (not uploaded yet), still need to implement in QTGMCp()
I will take a look at it and play around tomorrow, thanks!
Self-explanatory. I'm still trying it out, since it's so recent, but Asd-g should be pretty quick at fixing bugs, once reported. Thank you.