Vectorize Hashing in FlatGroupByHash#19302
Merged
pettyjamesm merged 3 commits intotrinodb:masterfrom Oct 19, 2023
Merged
Conversation
Contributor
|
test-other-modules failure is unrelated. I just fixed that in master. Sorry :) |
pettyjamesm
force-pushed
the
vectorize-flatgroupbyhash-hashing
branch
from
09ac658 to
68f7df6
Compare
Member
Author
No worries, thanks for the heads up- just rebased to pick up the fix. |
|
Started benchmark workflow for this PR.
|
pettyjamesm
force-pushed
the
vectorize-flatgroupbyhash-hashing
branch
from
3bbddf6 to
9fd7533
Compare
|
Started benchmark workflow for this PR.
Benchmark Summary:Top 5 duration differences
|
pettyjamesm
force-pushed
the
vectorize-flatgroupbyhash-hashing
branch
from
9fd7533 to
8df5f3a
Compare
martint
approved these changes
Oct 18, 2023
Member
martint
left a comment
martint
left a comment
There was a problem hiding this comment.
Squash the first three commits, as they are all essentially incremental parts of the same change.
core/trino-main/src/main/java/io/trino/operator/FlatHashStrategyCompiler.java
Outdated
Show resolved
Hide resolved
Implements column-wise hash calculations in FlatHashCompiler and changes FlatGroupByHash to use it to implement a batched approach to first computing a range of position hashes and then attempting to insert those positions using the hashes that were precomputed. By calculating position hashes in a columnar tarversal path, we can avoid repeated expensive bounds checking and allow the JIT to unroll loops into much more efficient forms. Additionally, when attempting to insert the positions into the FlatGroupByHash after precomputing the hash code we can start loading the relevant portition of the hash table into memory sooner since we aren't intermixing computing the hash with the memory accesses.
pettyjamesm
force-pushed
the
vectorize-flatgroupbyhash-hashing
branch
from
8df5f3a to
1a75e64
Compare
Member
|
Please note that we added a release notes entry for the performance improvement @pettyjamesm .. for future reference and as an example. |
This was referenced Dec 10, 2025
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
Implements column-wise hash calculations in FlatHashCompiler and changes FlatGroupByHash to use it to implement a batched approach to first computing a range of position hashes and then attempting to insert those positions using the hashes that were precomputed.
By calculating position hashes in a columnar traversal, we can avoid repeated expensive bounds checking per access, allow the JIT to emit more efficient unrolled loops, and access the memory in a way that friendlier and more predictable for CPU caches.
Additionally, when attempting to insert the positions into the FlatGroupByHash after precomputing the hash code we can start loading the relevant portition of the hash table into memory sooner since we aren't intermixing computing the hash with the memory accesses.
Since re-hashing is still performed row-at-a-time, the performance improvement is much more significant when the number of groups is small and when the number of columns in the hash calculation are larger.
BenchmarkGroupByHash.addPagesresults generally improve between 5-50% depending on the scenario:Release notes
(x) This is not user-visible or is docs only, and no release notes are required.
( ) Release notes are required. Please propose a release note for me.
( ) Release notes are required, with the following suggested text: