The docs say Vec3 is 12 bytes but I can't get that behavior on spirv.

[jre0]

I'm compiling as spirv using rust-gpu and I have this struct:

#[repr(C)]
pub struct Cp {
    vector: Vec3,
    index: u32,
}

The Vec3 appears to have 4 bytes of padding, taking up 16 bytes (making Cp take 20 in total). But the documentation says Vec3 has zero padding with size of 12 bytes. The glam code has this attrib #[cfg_attr(target_arch = "spirv", repr(simd))] which must make Vec3 like Vec3A when compiled to spirv. I tried to turn this off with scalar-math feature flag but it had no effect. Is there a way to make Vec3 always act like a fixed array of 3 f32 elements? Based on the documentation, I could use Vec3A on CPU and GPU sides to solve my problem but I would like a way to use less storage (not too interested in squeezing out that last drop of speed with SIMD). Thanks!

[bitshifter]

I'm not an expert on rust-gpu as generally those changes have been contributed. All of the vector types appear to use repr(simd) on spirv, not just Vec3, so I think the docs just don't cover rust-gpu behaviour at all. I assume that you are talking about the table under https://docs.rs/glam/latest/glam/#vec3a-and-mat3a? The docs could be updated to say this doesn't apply to spirv here, it would be better if someone who is more familiar with rust-gpu could make those docs clarifications though.

[jre0]

Thanks for the answer! Yes, that's the exact bit of documentation I was referring to. I would be interested in writing something for the docs about spirv behavior.