[uwu]

similar thing from google: http://lists.llvm.org/pipermail/llvm-dev/2019-September/1353...

> Inspired by the performance gains and to address the scalability issue of BOLT, we went about designing a scalable infrastructure that can perform BOLT-like post-link optimizations.

> Our experiments on large real-world applications and SPEC with code layout show that Propeller can optimize as effectively as BOLT, with just 20% of its memory footprint and time overhead.

[thedance]

One thing I really wanted to be able to do with Propeller was a post-link optimization of a Go program. Go's compiler is both fast and terrible. If you glance at the text, opportunities for simple optimizations abound. Unfortunately the layout of the program, references between the debug information and the function entry points, and Propeller having been designed for C++ programs kept me from getting very far. I did get a little ways, but not far enough that I could keep a long-running Go program from crashing.

[BubRoss]

Did you mean bolt instead of propeller? Bolt optimizes binaries, propeller is an llvm fork with extra information and a special linker.

Neither is made to do the optimizations you are talking about, that is what regular llvm does.

[dang]

Currently discussed at https://news.ycombinator.com/item?id=22655316.