Motivation: A large advantage of rust shaders in comparison to most extisting shaders are the support of modules and crates allowing to write reuseable code, which can be distributed fairly simple.

Providing libraries on-top of the std/core library would be a desirable target after the release of the compiler and the main library settles slowly. These libraries might contain code for example BRDF implementations (Lambertian, Phong, Oren-Nayar, ...) or common algorithms for noise creation.

Software rendering could be used as validation tool for testing the emitted SPIR-V code and also in later production stage to check for shader errors and regression. As the inspirv shaders are valid rust code this should be pretty straightfoward, but requiring some Rust backend only functions e.g. creation of interface structs.

Depends on #7

The movitation is to be able to compile a shader program also using the standard Rust compiler, allowing easy support for software renderer. This requires to implement all intrinsic functions with a Rust backend.

Currently only support C-like enums or consisting of unit types. SPIR-V (shader) doesn't support union types (afaik), so we probably will use a simple struct layout for emulating it:

enum Foo {
    A(u32),
    B(f32),
    C,
}

would be translated to

struct FooA(u32);
struct FooB(u32);
enum FooTy { A, B, C };

struct Foo {
   ty: FooTy,
   a: FooA,
   b: FooB,
}

This will need alot more memory than an corresponding enum, so complex enum types should be avoided.

Alternative

Keep the current state and disallow complex enum types, but add special handling for Option<T>.