MIR is not suitable IR as we basically have to reconstruct the control flow, which we already had in HIR!
MIR changes quite a lot atm (early stage), therefore HIR might be more suitable?
Use a rust fork instead of relaying on rustc_private as it could be easier to upstream changes and modify pieces of the compiler (often non-public functions..)
DXIL: Mircosofts' directx compiler makes quite a lot progress and gets a SPIR-V backend (implemented by some google employees). Moving towards HIR/HAIR -> HLSL AST based on clang might be a nicer approach. Optionally first start with a macro based approach ala glassfull
Some general thoughts regarding the project:
rustc_private
as it could be easier to upstream changes and modify pieces of the compiler (often non-public functions..)