New Linker "wild" "outperforms" mold

[theuserbl]

On Phoronix was written at
https://www.phoronix.com/news/Wild-Linker
"Wild: A Very Fast Linker Written In Rust, Aims To Outperform Mold Linker"

The linker can be found at
https://github.com/davidlattimore/wild
and is called "wild".

[rui314]

I haven't tried that yet, but it's interesting that they claimed wild is about 40% faster than mold. At some point, I think I need to take the time to see whether the claim is true and, if so, where the performance gains come from compared to mold. I'm busy with other projects these days, so I don't have time right now, though. If you've tried wild, please share your experience here.

[theuserbl]

If you've tried wild, please share your experience here.

I haven't tried it either. But I've read about it.

I think at
https://davidlattimore.github.io/posts/2025/09/02/rustforge-wild-performance-tricks.html
https://www.youtube.com/live/6Scgq9fBZQM?t=9246s
the author of wild describes, how he increases the speed of wild.

[fuhsnn]

Little test, linking release-build clang on Arch linux:

mold: mold 2.40.4 (083901f42dea2491be8bc7cf6e3b1e2b9a3c850d; compatible with GNU ld)
wild: Wild version 0.7.0 (compatible with GNU linkers)
lld: LLD 21.1.6 (compatible with GNU linkers)
ld: GNU ld (GNU Binutils) 2.45.0

linker/time/memory (max resident set size)
1 core
mold 0.753s 11000k
wild 0.510s 3872k
lld  2.163s 852020k
ld   2.904s 832716k
2 cores
mold 0.409s 10912k
wild 0.313s 3872k
lld  0.445s 845100k
4 cores
mold 0.242s 11008k
wild 0.175s 3872k
lld  0.365s 847960k
8 cores
mold 0.161s 11024k
wild 0.112s 3872k
lld  0.314s 852600k

[mati865]

Note that you need to add --no-fork for both Mold and Wild to get correct max RSS.

[fuhsnn]

Thanks, unfortunately I've removed the build folder so can't correct the result. Also should have said "threads" instead of "cores" because I used thread count options but no core pinning.