Use pyperformance to test the performance of pyjion,SURPRISE!!!
zhuofeng6 opened this issue · 7 comments
If the numbers are 1.0x, then it's definitely not running Pyjion. (it would be slower or faster)
I've never tried running the Pyperformance benchmark suite in that way, I don't think it would work because of how it invokes Python interpreters.
In this case, what should be used to test the performance of pyjion?
There is a benchmark suite script in the source repo, it gets run on CI
https://github.com/tonybaloney/Pyjion/runs/5967372554?check_suite_focus=true
See https://github.com/tonybaloney/Pyjion/blob/develop/main/Tests/benchmarks/suite.py
python Tests/benchmarks/suite.py
If you want to add another benchmark you can to the folder in tests, e.g. bench_example
by adding a __benchmarks__
list to the bottom of the file.
python Tests/benchmarks/suite.py example
There are some existing benchmarks from the pyperformance suite already in the folder (e.g. delta blue, fannkuch, n-body).
This is only using the performance of pyjion.
How do compare performance with pyjion and without pyjion in python using
a benchmark suite script in the source repo
It runs each benchmark twice. Once without Pyjion and once with Pyjion. The number without is the numbers in purple.
The right 3 columns show the percentage difference between the two
WHAT! it is surprised me so much!
So the performance of using pyjion may be reduced?
WHAT! it is surprised me so much!
So the performance of using pyjion may be reduced?
Yes. Some benchmarks are faster and some are slower.