Uploading COLOR_RGB (format Vector3<u16>) data to a gpu point buffer will implicitly convert the data into Vector4<u32>. I guess this was done to allow accessing the data from compute shaders without relying on extensions for using 8-/16-bit types and non-vec4 alignment in shaders. But for vertex data, Vector3<u16> is usually supported without extensions and the conversion currently leads to much wasted space. I'd also argue that this implicit conversion is unexpected (at least when using the gpu point buffers in a context besides compute shaders).
Uploading COLOR_RGB (format
Vector3<u16>
) data to a gpu point buffer will implicitly convert the data intoVector4<u32>
. I guess this was done to allow accessing the data from compute shaders without relying on extensions for using 8-/16-bit types and non-vec4 alignment in shaders. But for vertex data,Vector3<u16>
is usually supported without extensions and the conversion currently leads to much wasted space. I'd also argue that this implicit conversion is unexpected (at least when using the gpu point buffers in a context besides compute shaders).