ARROW-11990: [C++][Compute] Handle errors consistently #10098
Closed
Conversation
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
Contributor
Author
|
Turns out a much bigger change than I thought. Some hints maybe helpful to reviewer:
|
Contributor
Author
|
@github-actions crossbow submit -g nightly |
|
Revision: b7229010b5413113a6274f7ad6d6d9ad900ff2e8 Submitted crossbow builds: ursacomputing/crossbow @ actions-336
|
Contributor
Author
|
Benchmark diff of all computer kernels, skylake, clang-9. Removed tests with less than 10% deviation. |
Contributor
Author
|
MacOS CI error is due to LLVM-12 update, https://issues.apache.org/jira/browse/ARROW-12467 |
Member
|
Here are the changes (only with |
Member
|
I'm in support of this. Performance improvements notwithstanding, the main benefits IMHO are:
|
Member
|
Out of curiosity, why does this improve performance? |
Member
|
@nealrichardson Not sure, probably it makes things easier for the compiler. |
Member
Member
|
@nealrichardson In the realm of wild guesses I would investigate before taking much stock in, a change like this... ...changes from setting a variable on a class (that variable could then potentially be read in a lot of different places) to an out parameter (which can only be seen by the caller). By reducing the visible scope of the variable you are writing too you increase the chance the compiler decides that no one else needs to see the change and it can keep it in a register instead of writing it out to RAM somewhere. |
Contributor
Author
|
Rebased and fixed merge conflicts. |
Contributor
Author
|
MinGW32 python test timeout looks a spurious issue happens occasionally. |
bkietz
approved these changes
Apr 26, 2021
Member
bkietz
left a comment
There was a problem hiding this comment.
Thanks a lot for this cleanup, looks great!
Two minor nits:
Arrow handles errors by returning Status/Result. But in compute kernels, errors are populated in KernelContext.status. This is not consistent, and updating KernelContext.status is not thread safe. This patch removes KernelContext.status and returns kernel errors as Status/Result.
Contributor
Author
|
RTools CI failure looks not related. @nealrichardson |
pitrou
approved these changes
Apr 27, 2021
Member
|
Really a nice improvement, thank you! |
Member
|
I do think the RTools failures are unrelated (I see them on other PRs), so will merge. |
This was referenced Apr 27, 2021
Member
|
FTR the R failure is This is because the 4.0.0 release just his CRAN, and apparently we have not bumped the version numbers post-release on master yet. @kszucs? |
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.
Arrow handles errors by returning Status/Result. But in compute kernels,
errors are populated in KernelContext.status. This is not consistent,
and updating KernelContext.status is not thread safe.
This patch removes KernelContext.status and returns kernel errors as
Status/Result.
See big performance improvement for arithmetic kernels, especially the
checked version (up to 4x).
Also see ~50% drops from some filter kernels. Will investigate deeper
as follow up task.