Michael Ho has posted comments on this change. Change subject: IMPALA-4397,IMPALA-3259: reduce codegen time and memory ......................................................................
Patch Set 7: (3 comments) http://gerrit.cloudera.org:8080/#/c/4956/7/be/src/codegen/llvm-codegen.cc File be/src/codegen/llvm-codegen.cc: Line 1073: module_ = NULL; What about destroying fns_to_jit_compile_ here too as it may still contain stale llvm::Function pointers to memory belonging to module_ ? Is it safer to also destroy other fields of similar nature too such as codegend_functions_, loaded_functions_, registered_exprs_ etc ? http://gerrit.cloudera.org:8080/#/c/4956/7/be/src/exec/partitioned-aggregation-node.cc File be/src/exec/partitioned-aggregation-node.cc: PS7, Line 1706: if (aggregate_evaluators_.size() > Expr::CODEGEN_INLINE_EXPRS_THRESHOLD) { > I experimented with a code change that disabled inlining always, and looked Do you think it's a reasonable middle ground to always inline MIN / MAX / SUM / COUNT only and leave the rest as not inlined ? The update function of NDV is probably too large to be worth inlining. http://gerrit.cloudera.org:8080/#/c/4956/8/common/thrift/PlanNodes.thrift File common/thrift/PlanNodes.thrift: PS8, Line 467: red is -- To view, visit http://gerrit.cloudera.org:8080/4956 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-MessageType: comment Gerrit-Change-Id: Id10015b49da182cb181a653ac8464b4a18b71091 Gerrit-PatchSet: 7 Gerrit-Project: Impala-ASF Gerrit-Branch: master Gerrit-Owner: Tim Armstrong <[email protected]> Gerrit-Reviewer: Marcel Kornacker <[email protected]> Gerrit-Reviewer: Michael Ho <[email protected]> Gerrit-Reviewer: Tim Armstrong <[email protected]> Gerrit-HasComments: Yes
