performance change when using PIXIE with typed.Dict

[dlee992]

This is a wonderful feature! Thanks for your effort!

I am curious if using PIXIE to "compile" the C implementation of typed.Dict to LLVM bitcode (?) and running some Dict benchmark, will the speed be much faster than current Numba way to call the C functions via ctypes under the hood?

Maybe this is not a correct question, since I don't know much about PIXIE for now.

[stuartarchibald]

I guess this question comes from things I've alluded to in Numba public meetings with regards to integrating PIXIE library support into Numba? The idea is that a PIXIE compiled variant of Numba's typed container libraries would be inlinable as bitcode by Numba's JIT compiler, this removing the current call overhead associated with calling an opaque function and permitting optimisation across the call boundary. Hopefully code using these containers would be faster as a result.