The goal of this crate is to translate compute shaders in SPIR-V format into Rust code that can be executed on CPU. The main use case is to be able to run the piet-gpu shader pipeline even on hardware with inadequate or obsolete GPU infrastructure.
There are a number of projects with similar scope. The most ambitious are SwiftShader and Mesa Lavapipe, which run a significant fraction of all of Vulkan. However, they are based on doing JIT translation at runtime, which is a significant dependency and impacts startup time.
Probably the closest existing work is glsl-to-cxx, which translates GLSL to C++ for use in WebRender. There are a few differences; we start with SPIR-V, to leverage existing shader compilation infrastructure (and to reduce friction for source languages other than GLSL). We target Rust mostly because it has standard SIMD support, and also to minimize non-Rust tooling dependencies. And we target compute shaders rather than vertex+fragment, because that's what piet-gpu uses.
Another related project is the spirv to ispc translator. This is based on an earlier version of spirv-cross, which had C++ output (now removed).
A significant amount of work in OpenCL is also geared to being able to run workloads on CPU, as is Android Renderscript.
The first milestone for this crate would be to generate fairly simplistic scalar code, to get the shaders running at all. A further direction would be to map it to SIMD code, using many of the techniques of ispc. Major techniques would include detection of uniform, detection of linear memory access patterns (so SIMD load/store intrinsics could be used), and organization of access to workgroup-shared memory around barriers.
The license is MIT or Apache 2.0, at your choice.