Zig – SPIR-V Backend Progress

Zig's SPIR-V backend, which enables the creation of shaders and compute kernels, has seen significant advancements. Ali Cheraghi, a developer, has been working on revamping the backend, addressing issues such as the inability to express certain SPIR-V types in Zig's type system. The introduction of the `@SpirvType` builtin directly addresses this issue, allowing for more flexible and accurate representation of SPIR-V types. For instance, the `Sampler`, `Image`, `SampledImage`, and `RuntimeArray` types can now be expressed using this builtin. Additionally, execution mode information is now carried by the calling convention, eliminating the need for inline assembly `OpExecutionMode` instructions. Two new calling conventions, `spirv_task` and `spirv_mesh`, have been added to support mesh shading pipelines.

The improvements to Zig's SPIR-V backend and LLVM backend are part of a larger trend of refining and optimizing the language's capabilities, particularly in the areas of graphics and compute programming. The updates to the SPIR-V backend, in particular, bring Zig closer to being a viable option for developers working on graphics and compute-intensive projects. With the Vulkan API being a key target for SPIR-V, these advancements position Zig to better compete with other languages used in this domain, such as C++ and Rust. The fact that the SPIR-V backend can now utilize multi-threaded codegen, producing an Mir value like other self-hosted backends, and scheduling jobs on the compiler's thread pool, demonstrates the ongoing effort to enhance performance and efficiency.

The implications of these updates are significant for developers working on graphics and compute projects. With a more robust and efficient SPIR-V backend, developers can now explore the use of Zig for creating shaders and compute kernels, potentially benefiting from its safety features and performance capabilities. The redefinition of `@bitCast` semantics, driven by the need to align with the language proposal #19755, also ensures that the LLVM backend's implementation is consistent and well-defined. As the development of Zig continues, it will be interesting to see how these improvements translate into real-world adoption and the creation of new projects leveraging these capabilities. The increase in passing behavior tests, by nearly 10%, to 49% on the `spirv64-vulkan` target, indicates progress, but also highlights the need for continued refinement.