[PATCH] D38266: Parallelize string merging.

[Rui Ueyama via Phabricator via llvm-commits]

ruiu created this revision.
Herald added a subscriber: emaste.

String merging is one of the most time-consuming functions in lld.
This patch parallelize it to speed it up. On my 2-socket 20-core
40-threads Xeon E5-2680 @ 2.8 GHz machine, this patch shorten the
clang debug build link time from 8.77s to 7.05s. It's a 20%
improvement and actually pretty noticeable. In this test condition,
lld is now 3x faster than gold.

When I limit the number of available cores to 2 using taskset command,
lld with or without this patch takes 11.74s or 11.71s, respectively,
to link the same executable. When I limit it to 1, they takes 16.66s
or 13.89s, respectively. So, as long as you have two available cores,
this patch seems a overall win.


https://reviews.llvm.org/D38266

Files:
  lld/ELF/InputSection.h
  lld/ELF/SyntheticSections.cpp
  lld/ELF/SyntheticSections.h
  lld/test/ELF/comment-gc.s
  lld/test/ELF/compressed-debug-input.s
  lld/test/ELF/debug-gc.s
  lld/test/ELF/gc-sections-merge.s
  lld/test/ELF/merge-string-align.s
  lld/test/ELF/merge.s

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D38266.116627.patch
Type: text/x-patch
Size: 12988 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170926/23ac8aaf/attachment.bin>