In numba, I want to pass the config variable to a function as a compile-time constant. Specifically what I want to do is @njit def physics(config): flagA = config.flagA flagB = config.flagB aNumbaList = List() for i in range(100): if flagA: …

Hey @Sumit112192 , I’m not sure if it works for your purpose but you can access initial values of dictionaries as literals using “numba.overload”. from numba import njit, types, literally from numba.typed import Dict, List from numba.extending import overload @njit def get_config_1(): return …

Hi @Sumit112192 The problem is that numba.typed.List is a homogeneous container, i.e. its elements must all have the same type. However, in the following loop, i is an integer and i/10 is a float. What about using integer division instead? for i in range(100): if flagA: aNumbaList.appe…

@sschaer I agree. But if the config variables are treated as compile-time constants, there won’t be any else statement here since the flag is always True. So, it will never see that I am appending a float. flag1 = True @numba.njit def check1(): anotherNumbaList = List() for i in range(100)…

I don’t think there is a way to propagate literals through dictionaries. There is no such type as a literal dict. Numba knows about literal lists, so maybe that could work. But honestly, even if there was such a solution, relying on branch pruning to make typing work sounds like a very fragile solut…

I agree that it’s a bad idea. But for now, if this would work, it can enhance the speed of the code till the time we come up with a new efficient solution.

Could you use overload or generated_jit to get the effect you’re looking for based on the typing of the config variable(s)?

generated_jit has been deprecated, I believe.

[image] How to avoid multiple overload functions for the same config variable Support: How do I do ...? A somewhat close solution. So, there is this dead branch pruning in numba @numba.njit def aNumaFunc(flag): if flag is None: variable1 = None else …

Deprecated in the sense that it has become obsolete. The latest version doesn’t even have that.

Thanks for the heads-up, I wasn’t aware. I guess I missed the deletion alert in the release notes. Interestingly, dispatcher.py still seems to have the compiler EDIT: @DannyWeitekamp you might be interested in this if you haven’t picked up on it… I think at least some versions of CRE used generat…

Passing config variables to functions so they behave as compile time constants

Numba Support: How do I do ...?

Sumit112192 July 4, 2024, 12:56pm 9

Can you look at this question?

Topic		Replies	Views
How to avoid multiple overload functions for the same config variable Support: How do I do ...?	3	88	July 4, 2024
Coercing literal types for dispatch Numba	2	295	October 23, 2023
Unexpected behaviour with literally Development	2	648	August 12, 2020
Global constant dictionaries Support: How do I do ...?	1	93	October 14, 2025
Setting Class variable as function Polymorphism Community Support	3	273	September 18, 2023

Passing config variables to functions so they behave as compile time constants

Related topics