Trying Out LLVM 4.0's LLD Linker On Ubuntu 17.04 vs. GNU LD, GNU Gold

[phoronix]

Phoronix: Trying Out LLVM 4.0's LLD Linker On Ubuntu 17.04 vs. GNU LD, GNU Gold

With this week's LLVM 4.0 release making the LLD linker ready for production use on some platforms, namely ELF on x86_64 / AArch64, I decided to finally try it out on one of my test systems. I set LLD as the default linker on an Ubuntu 17.04 system and set off to run some benchmarks.

http://www.phoronix.com/vr.php?view=24372

[AllLinux]

Does it make any difference in kernel link times?
Does it even work linking the kernel?

[AsuMagic]

Is there any sort of LTO with LLD? How would different linkers compare binary performance-side?

[Michael]

Is there any sort of LTO with LLD? How would different linkers compare binary performance-side?

LLD supports LTO. With these tests, there weren't any difference in the generated binary performance, hence why I didn't bother including those numbers.

[davidbepo]

it will also be interesting to clang + lld
lld seems very good but it still has to improve

[chickenkiller]

What about memory consumption ?

[Marc Driftmeyer]

Try comparing Clang-LLD vs. GCC-LLD.

[carewolf]

You said in the beginning of the article that you would test gold both out of the box, and with --threads. Which one is in the graphs and where is the other?

[yurikoles]

Michael could you test impact of building kernel, xorg and mesa on recent CPUs with -march=native via gcc and llvm. It may be interesting, how many fps you will get by use of compiler optimization, since repo packages is built with -march=generic.

You may just use Gentoo, or recent ubuntu + some packages from source.

[pal666]

it can be faster, or slower, or fail. i'd call it alpha quality