search

faster-cpython / bench_runner

Code for running pyperformance benchmarks on Github Action runners
BSD 3-Clause "New" or "Revised" License
3 stars 4 forks source link

Some pystats results are nonsensical #152

Closed mdboom closed 4 months ago

mdboom commented 4 months ago

Some of the pystats results for the Optimizer stats don't make sense -- they show more successful traces than trace attempts. For example, the hexiom ones:

Count Ratio
Optimization attemptsⓘ 1,540  
Traces createdⓘ 52,900 3,435.1%
Trace stack overflowⓘ 0 0.0%
Trace stack underflowⓘ 16,060 1,042.9%
Trace too longⓘ 0 0.0%
Trace too shortⓘ 1,197,960 77,789.6%
Inner loop foundⓘ 500 32.5%
Recursive callⓘ 0 0.0%
Low confidenceⓘ 100 6.5%
Executors invalidatedⓘ 0 0.0%
Traces executedⓘ 109,768,500  
Uops executedⓘ 2,948,922,980 2,686.5%

I can't reproduce this locally, so there must be something else going on.

mdboom commented 4 months ago

Moved to CPython: https://github.com/python/cpython/issues/116760