Mid-level intermediate representation (possibly using MLIR)?

[oroppas]

Out of curiosity, I wonder if nlvm has any plans to implement IR like other languages:

Julia https://docs.julialang.org/en/v1/devdocs/ssair/ Rust https://rustc-dev-guide.rust-lang.org/mir/index.html#the-mir-mid-level-ir Swift https://github.com/apple/swift/blob/main/docs/SIL.rst Flang https://github.com/llvm/llvm-project/tree/main/flang

Any comments will be greatly appreciated. Thanks!

[arnetheduck]

Yeah, I've thought about it, though it would actually be most powerful if done in the upstream compiler such that both the nlvm and C backends (which are quite similar at the end of the day) would use / benefit - one of the more difficult issues to deal with in nlvm is bug-for-bug compatibility.

Ideally it would also integrate with an llvm-based CTFE engine such that the nim VM is no longer used (currently there's a VM in nim which quite significantly differs from the runtime code, which incidentally also has its own IR) and instead, the same ir is used for both VM and final compile.

[Araq]

--gc:arc pretty much uses a Nim specific IR, it keeps changing though and some form of SSA would be useful to have.

[oroppas]

Perhaps,

CIL : Common MLIR abstraction for C / C++/ Fortran https://github.com/compiler-tree-technologies/cil https://llvm.org/devmtg/2020-09/slides/CIL_Common_MLIR_Abstraction.pdf https://www.youtube.com/watch?v=3gcw-8C9UbA

might help (if and only if this or similar technologies are implemented in MLIR one day)?