What's the current status of ewa_lanczos in vo=gpu-next?

[Axiom415]

Back then you can't (cmiiw) use ewa_lanczossharp if you're using gpu-next. mpv wiki also stated this

ewa_lanczos defaults to using radius 3.2383... in gpu-next, rather than the radius 3.0 of gpu.

So ewa_lanczos is the new ewa_lanczossharp in gpu-next.
Is this still true now? When using mpv --no-config --vo=gpu-next --gpu-api=d3d11 --profile=high-quality, it uses ewa_lanczossharp instead of ewa_lanczos, does that mean ewa_lanczos in gpu-next is the same as in gpu?

[hooke007]

2ec79f8

[Raza-0007]

does that mean ewa_lanczos in gpu-next is the same as in gpu?

ewa_lanczos in vo=gpu-next defaults to the radius of 3.2. So it is the same as ewa_lanczossharp from vo=gpu

There is now a new ewa_lanczossharp in vo=gpu-next that also defaults to a radius of 3.2 but its blur has been reduced to 0.9 from 1, making it a little sharper than ewa_lanczos in vo=gpu-next.

If you were happy with ewa_lanczossharp in vo=gpu, then you will be happy with ewa_lanczos in vo=gpu-next as they are essentially the same. ewa_lanczossharp in vo=gpu-next is slightly sharper than the ewa_lanczossharp in vo=gpu

I hope this makes sense.

[hooke007]

mpv/video/out/filter_kernels.c

Lines 352 to 375 in 0035bb7

	#define JINC_R3 3.2383154841662362
	#define JINC_R4 4.2410628637960699
	const struct filter_kernel mp_filter_kernels[] = {
	// Spline filters
	{{SCALER_SPLINE16, 2, spline16}},
	{{SCALER_SPLINE36, 3, spline36}},
	{{SCALER_SPLINE64, 4, spline64}},
	// Sinc filters
	{{SCALER_SINC, 2, sinc, .resizable = true}},
	{{SCALER_LANCZOS, 3, sinc, .resizable = true}, .window = WINDOW_SINC},
	{{SCALER_GINSENG, 3, sinc, .resizable = true}, .window = WINDOW_JINC},
	// Jinc filters
	{{SCALER_JINC, JINC_R3, jinc, .resizable = true}, .polar = true},
	{{SCALER_EWA_LANCZOS, JINC_R3, jinc, .resizable = true}, .polar = true, .window = WINDOW_JINC},
	{{SCALER_EWA_HANNING, JINC_R3, jinc, .resizable = true}, .polar = true, .window = WINDOW_HANNING},
	// See <https://legacy.imagemagick.org/Usage/filter/nicolas/#upsampling>
	{{SCALER_EWA_GINSENG, JINC_R3, jinc, .resizable = true}, .polar = true, .window = WINDOW_SINC},
	// Slightly sharpened to minimize the 1D step response error (to better
	// preserve horizontal/vertical lines). Blur value determined by method
	// originally developed by Nicolas Robidoux for Image Magick, see:
	// <https://www.imagemagick.org/discourse-server/viewtopic.php?p=89068#p89068>
	{{SCALER_EWA_LANCZOSSHARP, JINC_R3, jinc, .blur = 0.9812505837223707, .resizable = true},
	.polar = true, .window = WINDOW_JINC},

https://github.com/haasn/libplacebo/blob/73ffd79bdab5de25a607cdf12e28205f3833a8ed/src/filters.c#L765-L801

Almostly the same.

[Raza-0007]

So according to the information you attached, what I wrote before is true... right?

[hooke007]

It doesn't make sense unless you didn't update mpv for over three years...
There is no difference in gpu/gpu-next after that commit.

[Raza-0007]

OK!
I thought that the OP was asking about ewa_lanczos and ewa_lanczossharp from the time before vo=gpu-next was introduced, and the difference between them from then and now.

[Axiom415]

I see, so now ewa_lanczos in gpu-next is ewa_lanczossharp in gpu, and ewa_lanczossharp in gpu-next is even sharper than before, albeit a little bit. Thanks!