Make LLVM fast again
I’m not sure whether this has been true in the past, but it certainly isn’t true now. Each LLVM release is a few percent slower than the last. LLVM 10 put some extra effort in this area, and somehow managed to make Rust compilation a whole 10% slower, for as yet unknown reasons.
This is a really interesting read on compilers and performance benchmarking.
Also kind of insane that this investment hasn’t been done on a project where even a 10% win probably saves lots of CPU cycles worldwide.
“The server this runs on only has 2 cores, so a full LLVM build can take more than two hours. For smaller changes, building LLVM from ccache and compiling the benchmarks takes about 20 minutes. This is too slow to test every single commit, but we don’t really need to do that, as long as we automatically bisect any ranges with significant changes.”
2 cores is what - $10/month on a lot of VPS providers? (though that might not have enough memory or disk to compile llvm)Posted on 2020-05-11T06:03:16+0000