Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Chunk based iteration in accumulate_indices #13451

Merged
merged 11 commits into from
Nov 20, 2024
Merged

Chunk based iteration in accumulate_indices #13451

merged 11 commits into from
Nov 20, 2024

Conversation

jayzhan211
Copy link
Contributor

Which issue does this PR close?

Closes #.

Rationale for this change

What changes are included in this PR?

Are these changes tested?

Are there any user-facing changes?

@jayzhan211
filter chunk
90e8930 
Signed-off-by: Jay Zhan <[email protected]>
@jayzhan211
fmt
d0c8e05 
Signed-off-by: Jay Zhan <[email protected]>
@jayzhan211
acc
f3697b8 
Signed-off-by: Jay Zhan <[email protected]>
Dandandan
Dandandan reviewed Nov 16, 2024
View reviewed changes
datafusion/functions-aggregate-common/src/aggregate/groups_accumulator/accumulate.rs
let bit_chunks = filter.values().bit_chunks();
let group_indices_remainder = group_indices_chunks.remainder();

group_indices_chunks.zip(bit_chunks.iter()).for_each(
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm wondering if just using set_indices here is possible to iterate on valid indices?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm quite surprise that the fixed chunk size (64) is faster than iterating set_indices

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah it seems it is faster (generating faster code) when valid values are high enough.

@jayzhan211
BitIndexIterator
6d9f26c 
Signed-off-by: Jay Zhan <[email protected]>
@jayzhan211
cleanup
3721e74 
Signed-off-by: Jay Zhan <[email protected]>
@jayzhan211 jayzhan211 changed the title Chunk based filter in accumulate_indices Chunk based iteration in accumulate_indices Nov 17, 2024
@jayzhan211
Copy link
Contributor Author

jayzhan211 commented Nov 17, 2024 •
edited
Loading 

Comparing main and filter-chunk
--------------------
Benchmark clickbench_1.json
--------------------
┏━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query        ┃      main ┃ filter-chunk ┃        Change ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 0     │    0.26ms │       0.30ms │  1.16x slower │
│ QQuery 1     │   26.00ms │      25.44ms │     no change │
│ QQuery 2     │   55.35ms │      53.21ms │     no change │
│ QQuery 3     │   46.89ms │      45.90ms │     no change │
│ QQuery 4     │  307.76ms │     309.23ms │     no change │
│ QQuery 5     │  574.82ms │     570.49ms │     no change │
│ QQuery 6     │   26.30ms │      26.27ms │     no change │
│ QQuery 7     │   27.54ms │      29.15ms │  1.06x slower │
│ QQuery 8     │  356.26ms │     362.77ms │     no change │
│ QQuery 9     │  546.82ms │     557.03ms │     no change │
│ QQuery 10    │  131.44ms │     141.31ms │  1.08x slower │
│ QQuery 11    │  150.23ms │     152.91ms │     no change │
│ QQuery 12    │  614.05ms │     610.26ms │     no change │
│ QQuery 13    │  706.93ms │     688.08ms │     no change │
│ QQuery 14    │  566.77ms │     547.21ms │     no change │
│ QQuery 15    │  388.56ms │     383.52ms │     no change │
│ QQuery 16    │  932.13ms │     941.68ms │     no change │
│ QQuery 17    │  850.16ms │     838.91ms │     no change │
│ QQuery 18    │ 2798.37ms │    2790.18ms │     no change │
│ QQuery 19    │   39.54ms │      39.65ms │     no change │
│ QQuery 20    │  707.81ms │     702.99ms │     no change │
│ QQuery 21    │  971.53ms │     958.09ms │     no change │
│ QQuery 22    │ 2687.40ms │    2676.42ms │     no change │
│ QQuery 23    │ 5810.42ms │    5824.84ms │     no change │
│ QQuery 24    │  410.33ms │     385.74ms │ +1.06x faster │
│ QQuery 25    │  396.73ms │     405.53ms │     no change │
│ QQuery 26    │  448.47ms │     450.85ms │     no change │
│ QQuery 27    │ 1074.77ms │    1038.62ms │     no change │
│ QQuery 28    │ 8169.84ms │    7937.72ms │     no change │
│ QQuery 29    │  348.52ms │     355.17ms │     no change │
│ QQuery 30    │  520.21ms │     501.82ms │     no change │
│ QQuery 31    │  528.27ms │     525.37ms │     no change │
│ QQuery 32    │ 2780.65ms │    2652.40ms │     no change │
│ QQuery 33    │ 2678.26ms │    2671.76ms │     no change │
│ QQuery 34    │ 2572.40ms │    2586.40ms │     no change │
│ QQuery 35    │  526.14ms │     529.53ms │     no change │
│ QQuery 36    │  115.23ms │     112.60ms │     no change │
│ QQuery 37    │   77.91ms │      79.28ms │     no change │
│ QQuery 38    │   81.69ms │      82.17ms │     no change │
│ QQuery 39    │  203.75ms │     207.33ms │     no change │
│ QQuery 40    │   23.99ms │      24.14ms │     no change │
│ QQuery 41    │   22.93ms │      23.93ms │     no change │
│ QQuery 42    │   28.73ms │      28.82ms │     no change │
└──────────────┴───────────┴──────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary           ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (main)           │ 40332.13ms │
│ Total Time (filter-chunk)   │ 39875.01ms │
│ Average Time (main)         │   937.96ms │
│ Average Time (filter-chunk) │   927.33ms │
│ Queries Faster              │          1 │
│ Queries Slower              │          3 │
│ Queries with No Change      │         39 │
└─────────────────────────────┴────────────┘

@jayzhan211
count group
a889299 
Signed-off-by: Jay Zhan <[email protected]>
@jayzhan211 jayzhan211 marked this pull request as ready for review November 17, 2024 04:23
@jayzhan211 jayzhan211 requested a review from Dandandan November 17, 2024 04:24
@Dandandan
Copy link
Contributor

I wonder if any of the queries in that benchmark uses aggregate filters?

@jayzhan211
Copy link
Contributor Author

Neither Q7, Q10, nor Q24 uses an aggregate filter.

@Dandandan
Copy link
Contributor

Can we maybe find / create a query with filter? Maybe add one in clickbench_extended?

@jayzhan211 jayzhan211 marked this pull request as draft November 18, 2024 05:28
@jayzhan211
Copy link
Contributor Author

Can we maybe find / create a query with filter? Maybe add one in clickbench_extended?

aggregate filter required postgres dialect, it might be easier to run benchmark with criterion

@jayzhan211
clippy
568f749 
Signed-off-by: jayzhan211 <[email protected]>
@jayzhan211
taplo format
5c2b4bf 
Signed-off-by: jayzhan211 <[email protected]>
@jayzhan211 jayzhan211 marked this pull request as ready for review November 19, 2024 03:35
@jayzhan211
Copy link
Contributor Author

@Dandandan PTAL

@jayzhan211 jayzhan211 merged commit 963e8af into apache:main Nov 20, 2024
27 checks passed
@jayzhan211 jayzhan211 deleted the filter-chunk branch November 20, 2024 08:27
@jayzhan211
Copy link
Contributor Author

Thanks @Dandandan

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@Dandandan Dandandan Dandandan approved these changes

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@jayzhan211 @Dandandan