GH-20339: [C++] Add residual filter support to swiss join #39487
Conversation
|
|
zanmato1984
changed the title
| @@ -1123,29 +1143,6 @@ uint32_t SwissTableForJoin::payload_id_to_key_id(uint32_t payload_id) const { | |||
| return static_cast<uint32_t>(first_greater - entries) - 1; | |||
| } | |||
|
|
|||
| void SwissTableForJoin::payload_ids_to_key_ids(int num_rows, const uint32_t* payload_ids, | |||
There was a problem hiding this comment.
Deleted because used nowhere in existing code base.
github-actions
bot
added
awaiting committer review
|
@zanmato1984 would it be possible to add some test cases? |
Yes, I am working on that. |
|
@zanmato1984 I will completely review once the test cases are there. |
|
I tried to rebase and get things messy. Let me fix my branch first. Sorry for the trouble, guys :( |
zanmato1984
force-pushed
the
swiss-join-filter
branch
from
118c93e to
629cdd7
Compare
|
Hi @westonpace , I've addressed the comments you previously left except one thing I'm uncertain about in my last post. Would you please take a look again? Thanks! |
|
After merging your PR, Conbench analyzed the 7 benchmarking runs that have been run so far on merge-commit 0ce7267. There were no benchmark performance regressions. 🎉 The full Conbench report has more details. It also includes information about 11 possible false positives for unstable benchmarks that are known to sometimes produce them. |
Rationale for this change
Add residual filter support to swiss join.
What changes are included in this PR?
JoinResidualFilteras a centralized structure to evaluate residual filter in swiss join. It has various flavors of filtering for various join types. Zero-overhead is guaranteed for trivial filters (literal true and sometimes literal false/null). More detailed explanation in code comments.JoinProbeProcessor::OnNextBatch) to better cope withJoinResidualFiltercalls.Are these changes tested?
Legacy UTs (
HashJoin.Random,HashJoin.ResidualFilterandHashJoin.TrivialResidualFilter) cover part of this change. New fine-grained residual filter cases added as well.Are there any user-facing changes?
No.