jit all the functions which are slow

[shreyasbapat]

🐞 Problem

Some methods are tooo slow! For which we brought einsteinpy.ijit.jit
🎯 Goal
Make einsteinpy fast!

💡 Possible solutions
Identify the slow methods, and

from einsteinpy.ijit import jit

@jit
def ..

📋 Steps to solve the problem

• Comment below about what you've started working on.
• Add, commit, push your changes
• Submit a pull request and add this in comments - Addresses #<put issue number here>
• Ask for a review in comments section of pull request
• Celebrate your contribution to this project 🎉

P.S : See #98 (comment)

[Tushar-Tyagi]

@shreyasbapat I would like to work on this issue.

[shreyasbapat]

Sure! Go ahead. Its a little tricky one though :)

[Tushar-Tyagi]

Any suggestions on how to identify the slow methords?

[ritzvik]

The only problem in jitting functions is njit does not support numpy array deceleration inside the function.
See this for more details.

If there is some solution to be found, try this on christoffels() in utils/kerr_utils.py.
Let's see what happens next.

[SuyashSalampuria]

I would like to take up this issue @ritzvik @shreyasbapat

[shreyasbapat]

@SuyashSalampuria One good first issue is enough. You can dive into more complicated issues now :)

[Starninja666]

Hey, I'm looking out for some slow methods right now, interested in solving this issue once I find them.

[shreyasbapat]

Sure. Will wait for a Pull Request

[Atheriz-46]

I would like to work on this.

[JeS24]

@Atheriz-46 Sure, but as mentioned by others, this issue is tricky due to the way numba and python work.

[ghost]

As this is still open, can i work on this?

[JeS24]

@sanju50201 It is open. Feel free.

[ghost]

using the JIT wrapper does increase the performance by few milli seconds, can i create a PR for the changes made?

[JeS24]

using the JIT wrapper does increase the performance by few milli seconds, can i create a PR for the changes made?

@sanju50201 Sure. Please also include a comparison table for the performance, something like With-jit | Without-jit.

[ghost]

this is with the jit wrapper:
437534 function calls (421186 primitive calls) in 1.356 seconds

works well when i just do tasks individually with jit and without 'jit', however i think the compiler limits itself from interacting with the instances

[JeS24]

however i think the compiler limits itself from interacting with the instances

@sanju50201 Could you expand on this? Are you facing any errors?

[ghost]

however i think the compiler limits itself from interacting with the instances

@sanju50201 Could you expand on this? Are you facing any errors?

Yes, this is the issue i'm facing:

numba.core.errors.TypingError: Failed in nopython mode pipeline (step: nopython frontend)
Untyped global name 'CartesianConversion': Cannot determine Numba type of <class 'type'>

[JeS24]

@sanju50201 Yeah, numba is quite limited when it comes to custom types (classes). See here & here.

However, in this particular case, you do not need to jit CartesianConversion (or other classes in the module), as the underlying functions are already using jit. See https://github.com/einsteinpy/einsteinpy/blob/main/src/einsteinpy/coordinates/utils.py.

You can perhaps experiment with the geodesics or the integrators modules, that aren't numba-accelerated. Note that the classes or their methods are usually incompatible with numba, even with type annotations. Try jit-ing the functions in the corresponding utils.py modules, as with the coordinates module above.

[ghost]

@sanju50201 Yeah, numba is quite limited when it comes to custom types (classes). See here & here.

However, in this particular case, you do not need to jit CartesianConversion (or other classes in the module), as the underlying functions are already using jit. See https://github.com/einsteinpy/einsteinpy/blob/main/src/einsteinpy/coordinates/utils.py.

You can perhaps experiment with the geodesics or the integrators modules, that aren't numba-accelerated. Note that the classes or their methods are usually incompatible with numba, even with type annotations. Try jit-ing the functions in the corresponding utils.py modules, as with the coordinates module above.

@JeS24 This gives me a clear picture, thank you
let me see on the other things you mentioned