Currently it tries to fetch data from three sources:

• The Intrinsics.td file of the latest commit on https://github.com/llvm/llvm-project
• https://github.com/GuillaumeGomez/llvmint
• https://github.com/antoyo/llvmint

The last two sources are because the last commit on llvm-project may have removed some intrinsics that stdarch uses. It should be possible to simplify this to just Intrinsics.td from llvm-project by picking the exact commit that the nightly version against which cg_gcc will build uses as LLVM version. If any intrinsics are removed there, stdarch has had to stop using them before, so the removal of the intrinsic wouldn't matter.

The manual parsing of llvm-tblgen output could also likely be replaced by usage of the --dump-json flag.