16 bit textures

[omri-polycam]

Hey guys,
I'm trying to use 16 bit per component textures and cannot seem to get it to work. I'm using this to encode XYZ coordinates as a std::uint16_t unsigned integer in each color component of each texel of an RGB texture and this texture is sampled in a custom vertex shader. The failure mode is that it seems that whenever using the texture(...).rgb function in the vertex shader it always evaluates to (0, 0, 0).
I've already verified that my shader works when encoding the coordinates as std::uint8_t color components in a common 8 bit per component RGB texture (Texture::Format::RGB, Texture::Type::UBYTE, Texture::InternalFormat::RGB8). I've also verified using a debugger that I'm passing in reasonable std::uint16_t encoded coordinates into the texture data.
I've tried (Texture::Format::RGB_INTEGER, Texture::Type::USHORT, Texture::InternalFormat::RGB16UI) with no success. I've also tried (Texture::Format::RGB, Texture::Type::HALF, Texture::InternalFormat::RGB16F) with the same texture sampling always evaluating to a 0 vector no matter what data I pass into the texture pixel buffer descriptor.
I'm currently running on MacOS with the metal backend but will also need this to work on iOS/Metal, Android/GL and WebAssembly/GL. Does anybody know if I'm passing the wrong texture format, type and internal format enumerations? Is there something special I need to do in my vertex shader? Is there something I need to set for format or precision in my sampler2d parameter in my material? Ideally texture(...).rgb evaluates to a 3D float vector with components from 0 to 1, but it's easy enough to work around if it evaluates from 0 to 65,535 as an integer. Any help is appreciated!

Thanks,
Omri

[romainguy]

Filament itself uses an RGB16F texture sampled for every pixel of every lit material:

    Texture* lut = Texture::Builder()
            .width(DFG_LUT_SIZE)
            .height(DFG_LUT_SIZE)
            .format(backend::TextureFormat::RGB16F)
            .build(engine);

    Texture::PixelBufferDescriptor buffer(DFG_LUT, byteCount,
            Texture::Format::RGB, Texture::Type::HALF);

    lut->setImage(engine, 0, std::move(buffer));

This works with every backend on every platform. The sampler for it uses Type::SAMPLER_2D, Format::FLOAT, Precision::MEDIUM.

[omri-polycam]

Thanks, are 16 bit bit integer components also available on every platform? This would be preferable as half precision floating point has a signifcand only marginally better than plain 8 bit and would not bias higher precision nearer to the origin/zero unlike floating point. If so, what are the type, format, internal format and precision to use it?

[romainguy]

It should work yes.

[omri-polycam]

Thanks, I got it working. One last question, for another application in the same shader, is TextureFormat::RGB32F available on all platforms or do I have to go with half precision?