Announcement

Collapse
No announcement yet.

Fedora 31 Looking At Making It Easy To Use LLVM's LLD Linker

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Fedora 31 Looking At Making It Easy To Use LLVM's LLD Linker

    Phoronix: Fedora 31 Looking At Making It Easy To Use LLVM's LLD Linker

    Another late change proposal coming in for Fedora 31 is to allow update-alternatives to optionally point /usr/bin/ld as the default linker to /usr/bin/lld for the LLVM linker...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    I've struggled to get lld to work properly on my system, gold usually works fine

    There's been an ongoing discussion on the LKML about removing support for gold entirely

    Comment


    • #3
      Originally posted by FireBurn View Post
      I've struggled to get lld to work properly on my system, gold usually works fine

      There's been an ongoing discussion on the LKML about removing support for gold entirely
      I'd highly encourage you to file any bugs for these issues, it'll be worth it. lld nearly halved the LTO link time over gold for one of my projects (it requires Skia which can be quite large).

      Comment


      • #4
        The LLVM Linker is known for being a drop-in replacement to the GNU linkers while generally being much faster
        gnu linkers are not even drop-in replacements to themselves outside of trivial cases

        Comment


        • #5
          Originally posted by re:fi.64 View Post
          lld nearly halved the LTO link time over gold for one of my projects (it requires Skia which can be quite large).
          that's interesting. i thought most of lto time is spent by compiler. let me guess, your lto comparison included switching compilers? at least it doesn't mention supporting of non-clang lto
          Last edited by pal666; 26 July 2019, 02:51 PM.

          Comment


          • #6
            Originally posted by pal666 View Post
            that's interesting. i thought most of lto time is spent by compiler. let me guess, your lto comparison included switching compilers? at least it doesn't mention supporting of non-clang lto
            I'm not sure what the reason was, but both definitely took place with the same Clang version with the same flags; the only difference was -fuse-ld=lld.

            Comment


            • #7
              Originally posted by re:fi.64 View Post
              I'm not sure what the reason was, but both definitely took place with the same Clang version with the same flags; the only difference was -fuse-ld=lld.
              maybe clang's support of gold plugin was half-baked. i'd like to see comparison of lld+gcc vs gold+gcc but it doesn't look like first combination is possible

              Comment


              • #8
                Originally posted by pal666 View Post
                maybe clang's support of gold plugin was half-baked. i'd like to see comparison of lld+gcc vs gold+gcc but it doesn't look like first combination is possible
                GCC actually finally added it (maybe partly since some distros were shipping the patch anyway), though wouldn't there potentially be a similar issue with not-as-oft-used plugins?

                That being said, I believe Clang + gold was pretty extensively used at Google (they created gold after all) until lld came along.

                Comment

                Working...
                X