Padding in Working with Lights chapter about learn-wgpu

[sotrh]

ModelVertex doesn't need padding because it's only used in a vertex buffer and you have to specify the VertexBufferLayout for those. This means that the vertex pipeline knows the exact offsets of the data you're trying to pass in.

Uniform layout is a special case and it does require padding. By default uniform blocks use the std140 alignment standard. You can read about alignment in the wgpu spec, and I have started a write up discussing in more detail.

From what I've read the reasoning this is performance. Rather than query the offsets for all the fields, the gpu simply expects fields to be at certain locations in memory. It takes whatever data is at that memory location and sends it to the shader. This means that if your struct is not padded correctly your shader will be pulling from the wrong parts of the struct. If ModelVertex were used as a uniform buffer position would overlap tex_coords in the shader meaning that tex_coords.x in the shader would map to tex_coords[1] in Rust.