Availability of diagnostic tools for "numbarized" functions?

[Rafael Suchy]

Dear all, 

currently our research group is running a code (please see https://respy.readthedocs.io/en/latest/) that has a high ratio of Numba to Python code. Almost all functions are jitted. However, we have applications at hand that require to speeded up our code further. 

The problem: Once the code is converted to jitted functions it is not possible to internally profile those functions (to my best knowledge). Further optimization becomes very arduous since I need to factor out the code, guess the performance bottleneck, and re-write the code accordingly. That is, performance tracking boils down to monitor the time between function calls.

My question: Are there any diagnostic tools available for “numbarized" code parts? Any (web) resource that you have worked with and considered to be helpful is highly appreciated. 

Thank you in advance.

Best regards

Rafael 

[Siu Kwan Lam]

Hi Rafeal,

There are no diagnostic tool yet, but there are some ideas and experiments by the core devs. See this issue: https://github.com/numba/numba/issues/5028. 

Best,

Siu

[Philipp Eisenhauer]

Dear Siu,

Thanks a lot for your swift response ... and the pointer to the interesting discussion around #5028. Best, Philipp