Ignoring 16 bit shader permutations?

[tomosh22]

Is there any reason why 16 bit shaders are ignored here? In my tests the 16 bit version of the FSR3 upscaler shading change pyramid shader is slightly faster with no noticeable visual difference.

https://github.com/GPUOpen-LibrariesAndSDKs/FidelityFX-SDK/blob/f4c1da8e92f3fe563b5c28c44e6267ce6b6b8eb2/Kits/FidelityFX/upscalers/fsr3/internal/ffx_fsr3upscaler_shaderblobs.cpp#L289