GH-39704: [C++][Parquet] Benchmark levels decoding #39705
Merged
Conversation
This file contains 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
|
|
github-actions
bot
added
Component: Parquet
Component: C++
awaiting review
|
@pitrou @emkornfield @wgtmac Would you mind take a look? Also cc @Hattonuri |
|
Benchmark on my MacOS with Release (-O2) |
|
This benchmark shows that, when not highly repeated, the RLE without bitpacking is slow 😅 After changing RLE to BIT_PACKED, the speed gets a bit faster when repeat is not high: We should mentioned that, our native unpack to int16 is slow =_=, this making decoding a bit slow. |
|
Is this use case relevant here? #34510 Reading a non nullable fixed size list is missing the fast path, it’d nice to see it in the benchmark (even if not improving yet). With all the AI nowadays I assume tensor storage will be more and more common. |
Yeah I think it's related, I think I can optimize unpack later, but maybe I need some help in optimizing RLE |
github-actions
bot
added
awaiting committer review
emkornfield
reviewed
Jan 22, 2024
github-actions
bot
added
awaiting changes
emkornfield
reviewed
Jan 22, 2024
|
@emkornfield @pitrou Updated, so sorry for the delaying |
github-actions
bot
added
awaiting change review
|
Result in my MacOS with Release(-O2): |
mapleFU
force-pushed
the
level-decoding-benchmark
branch
from
b402c93
to
fc0c1a5
Compare
wgtmac
approved these changes
Jan 28, 2024
|
Ping @pitrou @emkornfield for help |
pitrou
reviewed
Jan 31, 2024
pitrou
reviewed
Jan 31, 2024
pitrou
reviewed
Jan 31, 2024
|
Strange phenomenon: we get results like |
|
Oh, it seems Google benchmark has a weird behavior here. Unrelated to this PR though. |
|
Posted google/benchmark#1749 for the Google benchmark oddity. |
|
FTR, benchmark numbers here: |
|
@pitrou I think the benchmark result shows "batch_size" should be take into consideration, for example, when the batchsize grows, BITPACK doesn't get improved, however, RLE is well optimized |
|
Mybad, change vector for output to |
|
After merging your PR, Conbench analyzed the 5 benchmarking runs that have been run so far on merge-commit 0c88d13. There were no benchmark performance regressions. 🎉 The full Conbench report has more details. |
dgreiss
pushed a commit
to dgreiss/arrow
that referenced
this pull request
Feb 19, 2024
### Rationale for this change This patch add the level-decoding benchmark. It test: 1. Different max-level (for flat type, maximum level would be 1, for nested type, it would grows) 2. With different repeat ( repeated null / non-null is different from non-repeated data) 3. With different read-batch size. This part of logic is a bit tricky in original code ### What changes are included in this PR? Add Level decoding benchmark ### Are these changes tested? No need ### Are there any user-facing changes? no * Closes: apache#39704 Authored-by: mwish <maplewish117@gmail.com> Signed-off-by: Antoine Pitrou <antoine@python.org>
zanmato1984
pushed a commit
to zanmato1984/arrow
that referenced
this pull request
Feb 28, 2024
### Rationale for this change This patch add the level-decoding benchmark. It test: 1. Different max-level (for flat type, maximum level would be 1, for nested type, it would grows) 2. With different repeat ( repeated null / non-null is different from non-repeated data) 3. With different read-batch size. This part of logic is a bit tricky in original code ### What changes are included in this PR? Add Level decoding benchmark ### Are these changes tested? No need ### Are there any user-facing changes? no * Closes: apache#39704 Authored-by: mwish <maplewish117@gmail.com> Signed-off-by: Antoine Pitrou <antoine@python.org>
thisisnic
pushed a commit
to thisisnic/arrow
that referenced
this pull request
Mar 8, 2024
### Rationale for this change This patch add the level-decoding benchmark. It test: 1. Different max-level (for flat type, maximum level would be 1, for nested type, it would grows) 2. With different repeat ( repeated null / non-null is different from non-repeated data) 3. With different read-batch size. This part of logic is a bit tricky in original code ### What changes are included in this PR? Add Level decoding benchmark ### Are these changes tested? No need ### Are there any user-facing changes? no * Closes: apache#39704 Authored-by: mwish <maplewish117@gmail.com> Signed-off-by: Antoine Pitrou <antoine@python.org>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Rationale for this change
This patch add the level-decoding benchmark. It test:
What changes are included in this PR?
Add Level decoding benchmark
Are these changes tested?
No need
Are there any user-facing changes?
no