Performance improvement in Extractor
Created by: mantognini
When sampling stainless' performance, the deconstruct function is listed as one of the most -- if not the most -- time consuming part of the overall execution. In order to optimise it, we can re-order the pattern matching cases to put the most frequent at the top.
The order proposed here is based (for the significantly most frequent ones) on statistics acquired when verifying all stainless benchmark under verification/valid. On my workstation, I could see the following effect when executing sbt it:test:
| version | wall clock | user time |
|---|---|---|
| master | 39:30 min | 2771 s |
| patched v1 | 32:52 min | 2349 s |
| patched v2 | 30:49 min | 2261 s |
So the results are not negligible! :-)
You can also compare the results on laraquad: