There is automatic protection for ordinary citizens (and non-citizens), but works of the US government and its employees are not protected by copyright.
2 Likes
Does the license on the repository not assert copyright protections?
Copyright (C) 2018 United States Government as represented by the
Administrator of the National Aeronautics and Space Administration.
All Other Rights Reserved.
Copyright protection is automatic in the US, you don’t need to assert a copyright to have it. On github, the terms of service allows public repositories to be forked by other github users, which somewhat bypasses the standard copyright laws, which still apply outside of what the github terms of service allow. If you make something public on github, other github users have permission to fork due to github terms of service.
I would take that at face value then, but it’s puzzling. As far as I understand, that is not a copyright statement that should legally exist.
Has anyone figured out why Numba was so incomprehensibly fast on the Laplace example in the tests from May 2020?
Also, has there been any advance on making Julia faster? We are losing almost an order of magnitude relative to C/Fortran.
2 Likes
one thing though, Julia is still on 1.2; Their python implementation is probably messed up.
1 Like
The Laplace/Jacobi benchmark is deeply flawed: the number of sweeps is adjusted according to the attained error. The C program and the Julia program in general execute different number of sweeps. If we force the number of sweeps to be constant, Julia is faster.
5 Likes
I filed an issue with the repository concerning the Jacobi iteration benchmark.https://github.com/JulesKouatchou/basic_language_comparison/issues/8
2 Likes
I revive this thread because I did found an other “benchmark paper” from NASA (2020): julia programming language benchmark using a flight simulation - NASA Technical Reports Server (NTRS)
I am however buffled by two “unusual” claims in that paper:
- Is it true that the JIT compiler “get better” as the number of runs of some code increase? I thought that, aside the first time where compilation was involved, the performances would have remained stable
- They use Julia 1.1 and they cite Distributed, but then they speak of multithreading advantages:
“The Julia Distributed package [1, Distributed Computing] adds functionality to extend the single execution thread used up until-now to multiple threads. The benchmark computer had 64 threads”
I thought Distributed deals with multiprocesses parallelism…
2 Likes
The report seems to be three years old (the latest citation is from 2018).
The problematic statements you list are indeed suspect.
2 Likes