[LLVMdev] [announce] Chrome now uses clang for production builds on Linux
Sean Silva
chisophugis at gmail.com
Thu Nov 13 15:55:28 PST 2014
Congratulations!
On Thu, Nov 13, 2014 at 10:40 AM, Nico Weber <thakis at chromium.org> wrote:
> Hi,
>
> As of Chrome m38, which shipped to stable a bit over a month ago, we are
> using clang as our production compiler for Linux – meaning that the bits we
> ship to users on Linux are now created by clang. It's now also our default
> host compiler for Chrome/Android builds (the target compiler is still gcc).
> We used to use gcc 4.6 before. Maybe this is interesting to others, hence
> this (somewhat belated) announcement.
>
> We're using the same model we use on OS X: We use a pinned clang revision
> without any changes (these days, we sometimes cherrypick a few newer
> versions every now and then) that we update to ~trunk every few weeks.
>
> The switch went off without larger problems, except one: The linux
> binaries that clang produces by default don't run on older 32-bit debians,
> an issue that we didn't find until our stable release. This is due to clang
> making assumptions about alignment that aren't true on 32-bit in practice.
> We worked around this and got an updated stable out relatively quickly, but
> producing broken 32-bit binaries by default is a great opportunity for
> improvement :-) I filed http://llvm.org/PR21414 for this.
>
> With the switch to clang, binary size dropped about 8% while performance
> mostly stayed the same. Startup time improved a bit, some microbenchmarks
> got faster, others got slower (I debugged the biggest slowdown, that turned
> out to be a benchmark bug). If you're interested
> https://chromeperf.appspot.com/group_report?rev=282246 has details,
> ignore all rows that aren't for "linux-release" or "chromium-rel-linux".
> page_cycler.tough_layout_cases for example got 2.5% faster, which is nice.
>
> Thanks everyone for all your work building such a great compiler!
>
> Nico
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20141113/f58b6262/attachment.html>
More information about the llvm-dev
mailing list