How would you architect a Julia compiler-level optimization pipeline for domain-specific workloads?

You would extend Julia’s macro system and LLVM passes to specialize DSL transformations before code generation.

Domain-specific workloads benefit from custom compiler pipelines. Julia macros transform DSL syntax into optimized IR before type inference. Additional optimization hooks can guide LLVM passes for vectorization or fusion. This allows domain experts to embed performance guarantees directly into syntax.