search

arnetheduck / nlvm

LLVM-based compiler for the Nim language
Other
712 stars 42 forks source link

question about interpreter #44

Closed Nidrop closed 1 year ago

Nidrop commented 1 year ago

Firstly, if nlvm creates one object file, does that mean that the original nim will be faster when recompiling because it can reuse object files? If not, would an nlvm-based interpreter be efficient in theory? I think it is possible via LLVM lli

arnetheduck commented 1 year ago

Compiling nim comprises of 3 stages: generating C files, generating object files and finally linking the object files - nlvm skips the C stage and gains some performance, then loses it back because it indeed doesn't reuse object files then it gains it again because of other pipeline improvements such as a better linker by default.

It depends on the code and the changes between each compile, which approach is faster.

An interpreter is available in https://github.com/arnetheduck/nlvm/pull/31, based on ORC which is more or less what lli does.