Mold Linker Performance Remains Very Compelling In 2024 Over GNU Gold/ld, LLVM lld

[phoronix]

Phoronix: Mold Linker Performance Remains Very Compelling In 2024 Over GNU Gold/ld, LLVM lld

The Mold high performance linker has long been known for offering excellent performance over GNU Gold/ld and LLVM lld while some fresh benchmark numbers reinforce the competitive advantage that persists today for this open-source project...

https://www.phoronix.com/news/Mold-Linker-2024-Performance

[Kjell]

Impressive

Thanks for the benchmark Michael

[anarki2]

So why isn't everyone using this already?

[satadru]

Mold has been the default linker used in building packages for the Chromebrew project for the last year or so, building for x86_64, armv7l, and i686 architectures. It has worked really well.

[dralley]

It probably won't be able to be the default linker without being able to support linking the Linux kernel.

https://fosdem.org/2024/schedule/event/fosdem-2024-2606-can-the-mold-linker-be-usr-bin-ld-/

But, Rui is working on it.

[ddriver]

So why isn't everyone using this already?

I guess many people can afford to wait an extra 30 seconds or so. Besides modl doesn't cover all use cases. Invite change and trouble in your build or wait 30 seconds? Not a big dilemma.

[miskol]

So why isn't everyone using this already?

I as developer when I debug something I want start app asap. But when I do release build I don't care about this anymore. So it is good for debug and dev. But you still need linker with best output

[brad0]

So why isn't everyone using this already?

Without it being able to cover all usage of linking an OS it's of no use. There is already an option that everyone is using that gains 95% of the performance.

[brad0]

I guess many people can afford to wait an extra 30 seconds or so. Besides modl doesn't cover all use cases. Invite change and trouble in your build or wait 30 seconds? Not a big dilemma.

There are already better options available, and yes that linking time really is annoying after doing some development on systems still using the BFD linker (Linux, Haiku, NetBSD/DragonFly). Also the memory usage is lower. I almost take it for granted the other OS's I use switched linkers numerous years ago.

[quaz0r]

I dont really know anything about linkers. Is there really much variance in the quality potential of their output? Is mold sacrificing output quality for performance? I feel like there are plenty of scenarios where slow antiquated crap can be replaced by stuff that is thoughtfully designed and programmed to be both performant and produce better output.

Anyway, I just installed a fresh gentoo system recently and enabled mold as the linker. There is like one or two packages you are instructed to disable mold on, I forget if I even use those packages, and I still use lld for the kernel and other than that everything seems to work just fine.