As the compiler pipeline evolves, we might start to see implicit ordering constraints between passes (e.g. type_inference, legalization, backend-specific lowering). Currently, these dependencies are not explicitly encoded, which makes the pipeline more fragile and harder to maintain.
In established compiler frameworks such as LLVM and TVM, passes often declare dependencies on prior analyses or transformations. This allows the compiler to ensure that required preconditions are satisfied before executing a pass.
Proposal:
Introduce a lightweight pass dependency mechanism where each pass can declare required predecessors.
For example:
type_inference -> ETISS backend would declare a dependency on type_inference
As the compiler pipeline evolves, we might start to see implicit ordering constraints between passes (e.g. type_inference, legalization, backend-specific lowering). Currently, these dependencies are not explicitly encoded, which makes the pipeline more fragile and harder to maintain.
In established compiler frameworks such as LLVM and TVM, passes often declare dependencies on prior analyses or transformations. This allows the compiler to ensure that required preconditions are satisfied before executing a pass.
Proposal:
Introduce a lightweight pass dependency mechanism where each pass can declare required predecessors.
For example:
type_inference -> ETISS backend would declare a dependency on type_inference