repr(simd) for matrix types

[Jasper-Bekkers]

This is super low prio because I don't think anybody actually cares about OpTypeMatrix at the moment, but it might be nice to have.

[fu5ha]

related to this and also kinda #121 , what does repr(simd) actually do on these gpu types?

[khyperia]

Inside rustc, repr(simd) types are called "Vector", and, we just translate those directly to spir-v OpTypeVector.

rust-gpu/rustc_codegen_spirv/src/abi.rs

Lines 336 to 343 in 8681464

	Abi::Vector { ref element, count } => {
	let elem_spirv = trans_scalar(cx, ty, element, None, is_immediate);
	SpirvType::Vector {
	element: elem_spirv,
	count: count as u32,
	}
	.def(cx)
	}

So, this issue isn't actually about repr(simd) for matrix types, because we can't hijack the existing system for that, but rather probably creating a spirv builtin attribute that translates the struct to be OpTypeMatrix, but whatever, close enough. (The attribute would not publicly documented and only used in spirv-std, similar to the current storage_class)

[Boscop]

This is super low prio because I don't think anybody actually cares about OpTypeMatrix at the moment, but it might be nice to have.

I care about it :)
In fact for my use case I need Mat3 and Mat4 as uniforms that match the native mat3/mat4 types.

[khyperia]

I don't think this is on our roadmap any time soon, but is a relatively self-contained nugget of a task that a community member could feasibly do that wouldn't interfere much with anything else, so I'll leave this open for now even though it's technically kind of a wontfix.