Texture format for indirect light

[jwwalker]

I tried using TextureFormat::RGBA8UI for a cube map texture to be used in an IndirectLight, and got an assertion failure:

Fragment Function(main0): The pixel format (MTLPixelFormatRGBA8Uint) of the texture (name:<null>) bound at index 4 is incompatible with the data type (MTLDataTypeFloat) of the texture parameter (light_iblSpecular [[texture(0)]]). MTLPixelFormatRGBA8Uint is compatible with the data type(s) (
    uint,
    ushort
).

Does an IndirectLight always require a floating-point texture format, or does that depend on the platform?

[romainguy]

Metal has more constraints but you should use a float format anyway. IBLs encode HDR lighting after all. Just use the R11G11B10F format, it’s still 32 bit per pixel but contains float data. If your IBL is small you can use RGBA16F as it’s easier to encore (you can use our half.h API for that).

[prideout]

By the way are you using cmgen to generate the IBL? I'm surprised it can even generate textures in that format.

[romainguy]

It doesn’t, cmgen only generates R11G11B10F or RGBM.

[jwwalker]

That's a fair point; IndirectLight::Builder::reflections does say that the cube map should be generated by cmgen. I just thought that when I wanted to make an all-white cube map, it would be easier to make one from scratch rather than figure out how to use cmgen. Maybe that was a miscalculation on my part.

[romainguy]

You don't have to use cmgen, but on Metal you need to make sure the data is in a float format.