Joe McDonnell has posted comments on this change. ( http://gerrit.cloudera.org:8080/21043 )
Change subject: IMPALA-12807: Add mold linker ...................................................................... Patch Set 1: (1 comment) http://gerrit.cloudera.org:8080/#/c/21043/1//COMMIT_MSG Commit Message: http://gerrit.cloudera.org:8080/#/c/21043/1//COMMIT_MSG@10 PS1, Line 10: linker that claims to be significantly faster than gold/lld. > Have any measurement on how much this impacts our builds? IMPALA-12807 has some numbers for an incremental build+link of impalad and it makes a big difference. >From what I see on a Jenkins job on Centos 7, it doesn't really change the >runtime of the automated builds very much (e.g. when they are building a lot >of backend tests), but it also doesn't regress them. I think link time is not >actually a big portion of our from-scratch build. C++ compilation takes up >more. I took a look at memory usage during that run, and it looks fine. My >guess is that the max memory usage for an individual link task is probably >similar, but if it can go more parallel on the CPU, then it can hold the >memory for a shorter duration. My plan is to make this an option a developer can set without changing the default yet. We currently have a "USE_GOLD_LINKER" option that defaults to true. I was going to change that to something like "IMPALA_LINKER" where it could be ld, gold, or mold. The default would stay gold for the initial commit. -- To view, visit http://gerrit.cloudera.org:8080/21043 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-Project: native-toolchain Gerrit-Branch: master Gerrit-MessageType: comment Gerrit-Change-Id: Ifc1ad3744ea9d0ff7bf111193652a21542d2b2cf Gerrit-Change-Number: 21043 Gerrit-PatchSet: 1 Gerrit-Owner: Joe McDonnell <[email protected]> Gerrit-Reviewer: Joe McDonnell <[email protected]> Gerrit-Reviewer: Michael Smith <[email protected]> Gerrit-Comment-Date: Tue, 20 Feb 2024 22:31:56 +0000 Gerrit-HasComments: Yes
