Michael Ho has posted comments on this change. Change subject: IMPALA-4231: fix codegen time regression ......................................................................
Patch Set 2: (7 comments) http://gerrit.cloudera.org:8080/#/c/4623/2/be/src/exec/partitioned-aggregation-node.h File be/src/exec/partitioned-aggregation-node.h: PS2, Line 479: noexcept I wonder if the need for noexcept stems from the use of vector in these functions or their callees ? http://gerrit.cloudera.org:8080/#/c/4623/1/be/src/exec/partitioned-hash-join-builder-ir.cc File be/src/exec/partitioned-hash-join-builder-ir.cc: PS1, Line 32: inline bool PhjBuilder::AppendRow( > The codegen time was higher. Otherwise the runtime didn't seem to be notice I see. So, it appears that LLVM wasn't able to consolidate the multiple calls to FreeMessages() into a single one. Patch 2 also shows the reduction in IR size by not inlining some code in BufferedTupleStream so it's not 100% clear what the difference is. I believe you may have explored the following and concluded that they are not helpful: 1. keep the return type as Status but include the changes to BufferedTupleStream and use std::move() like you did in other places. 2. un-inline AppendRow() ? http://gerrit.cloudera.org:8080/#/c/4623/2/be/src/runtime/buffered-tuple-stream.h File be/src/runtime/buffered-tuple-stream.h: PS2, Line 244: appending succee the append succeeded Line 245: /// and sets 'status' to OK if appending failed but can be retried or returns; or to OK; If the append failed PS2, Line 245: or returns; ? PS2, Line 247: AddRow AddRow() http://gerrit.cloudera.org:8080/#/c/4623/2/be/src/util/bloom-filter.h File be/src/util/bloom-filter.h: PS2, Line 176: inline Do you still need inline here ? -- To view, visit http://gerrit.cloudera.org:8080/4623 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-MessageType: comment Gerrit-Change-Id: Idf0fdedabd488550b6db90167a30c582949d608d Gerrit-PatchSet: 2 Gerrit-Project: Impala-ASF Gerrit-Branch: master Gerrit-Owner: Tim Armstrong <[email protected]> Gerrit-Reviewer: Michael Ho Gerrit-Reviewer: Michael Ho <[email protected]> Gerrit-Reviewer: Tim Armstrong <[email protected]> Gerrit-HasComments: Yes
