Facebook's BOLT Nearing Mainline LLVM For Optimizing Binaries

[phoronix]

Phoronix: Facebook's BOLT Nearing Mainline LLVM For Optimizing Binaries

Facebook's BOLT project for optimizing the performance out of compiled binaries is nearing the point of being added to LLVM's official source tree with its mono repository...

https://www.phoronix.com/scan.php?page=news_item&px=BOLT-Nearing-LLVM-Monorepo

[CochainComplex]

I allways wanted to check it out ...but at the moment I really don't have time. Such a pitty because the claims seem to be very impressive. LLVM or AOCC became very competitive regarding the performance of the compiled code.

[Alliancemd]

"Currently BOLT does not work on Windows" - it doesn't work on Linux as well, if you don't use an Intel processor.
It was supposed to have another technique of instrumenting the code but that one does not work as well.
It's either Linux on Intel CPU or nothing

[miskol]

From https://github.com/facebookincubator/BOLT
BOLT operates on X86-64 and AArch64 ELF binaries.
so AMD intel or arm cpus

[coder]

Does this work entirely by reducing instruction cache thrashing, or is there some other principle?

And presumably it can only optimize what's statically-linked into the executable? This would limit the benefit to code relying heavily on open source libraries for which dynamic linking is often mandated by license terms.