support bitand functions > 8 bit

[dave rice]

Summary of the enhancement request:

bitand (and bitor) operations function at 8 bit, I'd like to use bitand to work with the lower bits of some 10 and 16 bit video but it seems that bitand only finangles the first 8 bits.

Here is a command I would use to solo the 8th bit of the Y plane. I would like to do similarly with lower bits.

bit=8
ffplay 10bit_v210.mov -vf "extractplanes=y,lut=c0=bitand(val\,pow(2\,8-${bit}))*pow(2\,${bit})"

[Carl Eugen Hoyos]

Is it possible that bitand (and bitor) support > 8 bit but the lut filter does not?
(As you mentioned in ticket #2475.)

[Stefano Sabatini]

Replying to cehoyos:

Is it possible that bitand (and bitor) support > 8 bit but the lut filter does not?
(As you mentioned in ticket #2475.)

Yes, yuvrgb/yuv only supports 8bits pixel formats. What formats do you want to support?

[dave rice]

yuv422p10le

[Carl Eugen Hoyos]

So is this a duplicate of ticket #2475?

[dave rice]

No, ticket #2475 is for supporting >8 bit in lut, while ticket #3739 is for supporting values in bitand/bitor greater than 8 bit. Unfortunately in my example above this would only work if both lut and bitand supported >8 bit.

This uses geq

ffplay -f lavfi "mandelbrot,format=yuv422p10le,geq=lum=bitand(lum(X\,Y)\,128):cb=128:cr=128"

The 128 is 0b10000000, so the result zeros out bits 2-8 while bit 1 is left as it.

ffplay -f lavfi "mandelbrot,format=yuv422p10le,geq=lum=bitand(lum(X\,Y)\,1):cb=128:cr=128"

zeros out bits 1-7 and leaves 8. To normalize this you could multiple the bitand result by 255 to reveal the detail of bit position 8.

ffplay -f lavfi "mandelbrot,format=yuv422p10le,geq=lum=bitand(lum(X\,Y)\,1)*255:cb=128:cr=128"

From here I tried to increase y value of bitand to something >8 bit, such as 512, but realize that geq as well only outputs 8 bit.

[Carl Eugen Hoyos]

I tested the following and I believe it shows that bitand does support >8 bit

$ ffmpeg -f lavfi -i testsrc=r=1 -vf "select=bitand(n\, 262144)" -pix_fmt yuvj420p -vframes 1 out.jpg

[Carl Eugen Hoyos]

Replying to dericed:

geq as well only outputs 8 bit.

That wouldn't be a problem: At least in theory, a filter could exist that takes your 10 bit input, applies some calculations (in 10 bit and also with bitand) but outputs only 8 bit.
The problem is that both lut and geq only accept 8 bit input, so a scale filter is always inserted before these filters if the input file is 10 bit (or more).