Conversation
|
This looks reasonable to me in principle, though I wonder about one thing: iirc in the SMT-LIB benchmarking we noticed that a significant part of the overhead was due to loading olean files from the disk right? With this PR the number of file system accesses is going to increase quite a lot right? Apart from that I'd love to get the changes that you made to make Leanwuzla build on the current version again in though. I would ideally like to move the whole thing to the module system to reduce setup times a bit. |
|
A quick test on one of the SMT-LIB files I used to base benchmarks of in lean core shows that they are currently at 0.76s on Leanwuzla master, 0.87s with your changes and 0.67s with your changes - Fp. Would be kind of a shame to throw away those 10% of improvement 😅 |
|
Sure! I can break the PR into two! |
No description provided.