This is an archive of the discontinued LLVM Phabricator instance.

Differential D26919

Update document. It is not an overstatement that LLD is 2x faster than gold.
Needs ReviewPublic

Authored by ruiu on Nov 21 2016, 9:47 AM.
This revision needs review, but all specified reviewers are disabled or inactive.

Details

Reviewers
• espindola
Summary

LLD is getting faster even though it was very fast from the beginning.
I think it is now fair to say that it is generally about 2x faster than gold.

Diff Detail

Event Timeline

ruiu updated this revision to Diff 78734.Nov 21 2016, 9:47 AM
ruiu retitled this revision from to Update document. It is not an overstatement that LLD is 2x faster than gold..
ruiu updated this object.
ruiu added a reviewer: • rafael.
ruiu added a subscriber: llvm-commits.
davide added a subscriber: davide.Nov 21 2016, 10:56 AM
davide added inline comments.
docs/NewLLD.rst
14

This is a strong claim. I'd rather leave the comment as is or rephrase as "it is generally faster than gold". If you really want to show the figures, well, maybe, we can add a section with some charts updated periodically (at each release), but I would refrain from making the statement so general.

emaste added a subscriber: emaste.Nov 21 2016, 11:22 AM
emaste added inline comments.
docs/NewLLD.rst
14

Perhaps we can get an updated version of @rafael's and @davide's BSDCan benchmark results, and reference specific results?

davide added inline comments.Nov 21 2016, 11:23 AM
docs/NewLLD.rst
14

Yes, that would do it. I think it's much better showing real numbers =)

davide added inline comments.Nov 21 2016, 11:24 AM
docs/NewLLD.rst
14

To clarify, I had in mind something like this http://www.dragonflybsd.org/performance/ but for link times.

ruiu added inline comments.Nov 21 2016, 11:41 AM
docs/NewLLD.rst
14

I agree that a graph would be better than the number. I don't think saying we are about 2x faster is a bad thing, we can change the tone to make it sound not too strong, but a graph would be more convincing.

• espindola edited reviewers, added: • espindola; removed: • rafael.Mar 15 2018, 10:08 AM

Revision Contents

PathSize
docs/
2 lines

Diff 78734

docs/NewLLD.rst

The ELF and COFF Linkers The ELF and COFF Linkers
======================== ========================
We started rewriting the ELF (Unix) and COFF (Windows) linkers in May 2015. We started rewriting the ELF (Unix) and COFF (Windows) linkers in May 2015.
Since then, we have been making a steady progress towards providing Since then, we have been making a steady progress towards providing
drop-in replacements for the system linkers. drop-in replacements for the system linkers.
Currently, the Windows support is mostly complete and is about 2x faster Currently, the Windows support is mostly complete and is about 2x faster
than the linker that comes as a part of Micrsoft Visual Studio toolchain. than the linker that comes as a part of Micrsoft Visual Studio toolchain.
The ELF support is in progress and is able to link large programs The ELF support is in progress and is able to link large programs
such as Clang or LLD itself. Unless your program depends on linker scripts, such as Clang or LLD itself. Unless your program depends on linker scripts,
you can expect it to be linkable with LLD. you can expect it to be linkable with LLD.
It is currently about 1.2x to 2x faster than GNU gold linker. It is about 2x faster than GNU gold linker.
davideUnsubmitted
Not Done
ReplyInline Actions

This is a strong claim. I'd rather leave the comment as is or rephrase as "it is generally faster than gold". If you really want to show the figures, well, maybe, we can add a section with some charts updated periodically (at each release), but I would refrain from making the statement so general.

davide: This is a strong claim. I'd rather leave the comment as is or rephrase as "it is generally…
emasteUnsubmitted
Not Done
ReplyInline Actions

Perhaps we can get an updated version of @rafael's and @davide's BSDCan benchmark results, and reference specific results?

emaste: Perhaps we can get an updated version of @rafael's and @davide's BSDCan benchmark results, and…
davideUnsubmitted
Not Done
ReplyInline Actions

Yes, that would do it. I think it's much better showing real numbers =)

davide: Yes, that would do it. I think it's much better showing real numbers =)
davideUnsubmitted
Not Done
ReplyInline Actions

To clarify, I had in mind something like this http://www.dragonflybsd.org/performance/ but for link times.

davide: To clarify, I had in mind something like this http://www.dragonflybsd.org/performance/ but for…
ruiuAuthorUnsubmitted
Not Done
ReplyInline Actions

I agree that a graph would be better than the number. I don't think saying we are about 2x faster is a bad thing, we can change the tone to make it sound not too strong, but a graph would be more convincing.

ruiu: I agree that a graph would be better than the number. I don't think saying we are about 2x…
We aim to make it a drop-in replacement for the GNU linker. We aim to make it a drop-in replacement for the GNU linker.
We expect that FreeBSD is going to be the first large system We expect that FreeBSD is going to be the first large system
to adopt LLD as the system linker. to adopt LLD as the system linker.
We are working on it in collaboration with the FreeBSD project. We are working on it in collaboration with the FreeBSD project.
The linkers are notably small; as of October 2016, The linkers are notably small; as of October 2016,
the COFF linker is about 7k lines and the ELF linker is about 18k lines, the COFF linker is about 7k lines and the ELF linker is about 18k lines,
▲ Show 20 Lines • Show All 328 Lines • Show Last 20 Lines