gold linker is slowed down when LC_ALL != C |
||
Issue descriptionBuild libchrome.so, then try building it again with: export LC_ALL=C On my z620, in component mode: With LC_ALL=C: ~7 seconds Without LC_ALL=C: ~34 seconds While this is likely a bug in gold that should be fixed, we should just add set this variable in our linker wrapper script to benefit from it now. I've confirmed via "diff with.so without.so" that the .so's produced are identical.
,
Jul 29 2016
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/52f2a85089d3e3a85cee112a6ee6c7423394644b commit 52f2a85089d3e3a85cee112a6ee6c7423394644b Author: agrieve <agrieve@chromium.org> Date: Fri Jul 29 01:32:14 2016 Add LC_ALL=C to gcc linker wrappers to speed up links This speeds up an Android component build of libchrome.cr.so from 35s -> 10s on my z620. BUG= 632230 Review-Url: https://codereview.chromium.org/2190813002 Cr-Commit-Position: refs/heads/master@{#408549} [modify] https://crrev.com/52f2a85089d3e3a85cee112a6ee6c7423394644b/build/toolchain/gcc_link_wrapper.py [modify] https://crrev.com/52f2a85089d3e3a85cee112a6ee6c7423394644b/build/toolchain/gcc_solink_wrapper.py
,
Jul 29 2016
,
Aug 12 2016
Is there a gold bug for this?
,
Aug 13 2016
,
Sep 26 2016
For the record, which version of binutils/gold was this issue discovered on? I was checking to see if there's a similar issue with the binutils we have in Chromium OS and Android, but couldn't reproduce the linker slowdown described here.
,
Sep 27 2016
the public bug has the info about binutils version, if you want something more precise => the latest gold in src/third_party/android_tools/ndk exposes the issue for us |
||
►
Sign in to add a comment |
||
Comment 1 by dpranke@chromium.org
, Jul 28 2016