[ https://issues.apache.org/jira/browse/IMPALA-9476?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17053743#comment-17053743 ]
Sahil Takiar edited comment on IMPALA-9476 at 3/6/20, 8:28 PM: --------------------------------------------------------------- Attaching output of running bloaty on an Impala release binary. Some notes from an offline discussion: * "5.5% 29.4Mi 43.0% 25.6Mi /home/systest/Impala/be/src/exprs/like-predicate.cc" looks suspicious, could be from the RE2 library * bit-packing.cc has a lot of templates and macros to improve Parquet scanner perf * Thrift generated code is problematic because it has a lot of "vector<>"s - so each different "vector<Foo*>" gets its own copy of a function * hive_metastore_types.cpp is 21K lines of code * "\-ffunction-sections \-Wl,--gc-sections" may help removed unused functions * We should look into LTO, though we'd probably need to futz with -fvisibility=hidden to get it to remove unused code, which might break our codegen was (Author: stakiar): Attaching output of running bloaty on an Impala release binary. Some notes from an offline discussion: * "5.5% 29.4Mi 43.0% 25.6Mi /home/systest/Impala/be/src/exprs/like-predicate.cc" looks suspicious, could be from the RE2 library * bit-packing.cc has a lot of templates and macros to improve Parquet scanner perf * Thrift generated code is problematic because it has a lot of "vector<>"s - so each different "vector<Foo*>" gets its own copy of a function * hive_metastore_types.cpp is 21K lines of code * "-ffunction-sections -Wl,--gc-sections" may help removed unused functions * We should look into LTO, though we'd probably need to futz with -fvisibility=hidden to get it to remove unused code, which might break our codegen > Run Bloaty McBloatface size profiler on Impala binaries > ------------------------------------------------------- > > Key: IMPALA-9476 > URL: https://issues.apache.org/jira/browse/IMPALA-9476 > Project: IMPALA > Issue Type: Task > Reporter: Sahil Takiar > Priority: Major > Attachments: Impala-Bloaty-McBloatface-Output.txt > > > Bloaty McBloatface (https://github.com/google/bloaty) is a size profiler for > binaries. We should run it against Impala binaries and see what improvements > we can make. -- This message was sent by Atlassian Jira (v8.3.4#803005) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-all-unsubscr...@impala.apache.org For additional commands, e-mail: issues-all-h...@impala.apache.org