mbutrovich opened a new pull request, #3368:
URL: https://github.com/apache/datafusion-comet/pull/3368
This is a smaller scope alternative to #3366.
## Which issue does this PR close?
<!--
We generally require a GitHub issue to be filed for all bug fixes and
enhancements and this helps us generate change logs for our releases. You can
link an issue to this PR using the GitHub syntax. For example `Closes #123`
indicates that this PR will close issue #123.
-->
N/A.
## Rationale for this change
<!--
Why are you proposing this change? If this is already explained clearly in
the issue then this section is not needed.
Explaining clearly why changes are proposed helps reviewers understand your
changes and offer better suggestions for fixes.
-->
The release Docker builds (core-amd64-libs and core-arm64-libs) previously
used no explicit target-cpu, relying on rustc defaults. This produces binaries
targeting a minimal baseline that may miss important optimizations.
When running on a Graviton2 CPU, for example, we see significant time spent
in atomics that should have been optimized by LSE instructions.
For x86-64, x86-64-v3 is the appropriate baseline - it covers AVX2/FMA which
are available on all modern cloud instances (Haswell-era and newer, ~2013+).
For ARM64, neoverse-n1 is the appropriate baseline - it covers AWS
Graviton2, Ampere Altra, and newer cloud ARM processors. Key features include
LSE atomics which significantly improve concurrent workload performance. Code
compiled for neoverse-n1 also runs on Apple Silicon since M1 is a superset.
## What changes are included in this PR?
<!--
There is no need to duplicate the description in the issue here but it is
sometimes worth providing a summary of the individual changes in this PR.
-->
Updates Makefile to set explicit CPU targets for release builds:
- core-amd64-libs: adds -Ctarget-cpu=x86-64-v3
- core-arm64-libs: adds -Ctarget-cpu=neoverse-n1
- core-amd64: updates Linux target from haswell to x86-64-v3
- core-arm64: updates Linux target from native to neoverse-n1
Development targets (core, release, bench) continue to use native for
optimal local performance. x86-64-v3 was chosen over x86-64-v4 because v4
requires AVX-512, which is unavailable on AMD Zen 1-3 (EPYC Rome/Milan) and
older Intel chips. Most current cloud instances use these processors.
## How are these changes tested?
<!--
We typically require tests for all PRs in order to:
1. Prevent the code from being accidentally broken by subsequent changes
2. Serve as another way to document the expected behavior of the code
If tests are not included in your PR, please explain why (for example, are
they covered by existing tests)?
-->
These are build configuration changes. The release Docker build process
(dev/release/build-release-comet.sh) will use the new targets. The binaries
produced will be compatible with a wider range of cloud instances while still
enabling important optimizations like AVX2 (x86) and LSE atomics (ARM). We will
also try to test the Graviton2 instances and see if the atomics issue goes away.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
