Skip to content

Experiment: Enable SLP vectorizer for faster codegen#73957

Merged
rschu1ze merged 2 commits intoClickHouse:masterfrom
rschu1ze:faster-codegen
Dec 30, 2024
Merged

Experiment: Enable SLP vectorizer for faster codegen#73957
rschu1ze merged 2 commits intoClickHouse:masterfrom
rschu1ze:faster-codegen

Conversation

@rschu1ze
Copy link
Copy Markdown
Member

After the recent upgrades to LLVM 16 and 18, folks started to complain about slower codegen: #66053 (comment)

Prior to the LLVM 16 upgrade, CHJIT force-enabled the SLP vectorizer (see https://github.com/ClickHouse/ClickHouse/pull/69654/files#diff-897102eb40b3e08e4997b80459a83b97ac7d555e4534942cd30581e4f92cefe2). This PR now does the same using the new LLVM pass manager. Let's see if it helps with performance.

Changelog category (leave one):

  • Not for changelog (changelog entry is not required)

@rschu1ze rschu1ze mentioned this pull request Dec 29, 2024
19 tasks
@robot-clickhouse robot-clickhouse added the pr-not-for-changelog This PR should not be mentioned in the changelog label Dec 29, 2024
@robot-ch-test-poll2
Copy link
Copy Markdown
Contributor

robot-ch-test-poll2 commented Dec 29, 2024

This is an automated comment for commit f6ba191 with description of existing statuses. It's updated for the latest CI running

❌ Click here to open a full report in a separate page

Check nameDescriptionStatus
Integration testsThe integration tests report. In parenthesis the package type is given, and in square brackets are the optional part/total tests❌ failure
Successful checks
Check nameDescriptionStatus
AST fuzzerRuns randomly generated queries to catch program errors. The build type is optionally given in parenthesis. If it fails, ask a maintainer for help✅ success
BuildsThere's no description for the check yet, please add it to tests/ci/ci_config.py:CHECK_DESCRIPTIONS✅ success
ClickBenchRuns [ClickBench](https://github.com/ClickHouse/ClickBench/) with instant-attach table✅ success
Compatibility checkChecks that clickhouse binary runs on distributions with old libc versions. If it fails, ask a maintainer for help✅ success
Docker keeper imageThe check to build and optionally push the mentioned image to docker hub✅ success
Docker server imageThe check to build and optionally push the mentioned image to docker hub✅ success
Docs checkBuilds and tests the documentation✅ success
Fast testNormally this is the first check that is ran for a PR. It builds ClickHouse and runs most of stateless functional tests, omitting some. If it fails, further checks are not started until it is fixed. Look at the report to see which tests fail, then reproduce the failure locally as described here✅ success
Flaky testsChecks if new added or modified tests are flaky by running them repeatedly, in parallel, with more randomization. Functional tests are run 100 times with address sanitizer, and additional randomization of thread scheduling. Integration tests are run up to 10 times. If at least once a new test has failed, or was too long, this check will be red. We don't allow flaky tests, read the doc✅ success
Install packagesChecks that the built packages are installable in a clear environment✅ success
Performance ComparisonMeasure changes in query performance. The performance test report is described in detail here. In square brackets are the optional part/total tests✅ success
Stateful testsRuns stateful functional tests for ClickHouse binaries built in various configurations -- release, debug, with sanitizers, etc✅ success
Stateless testsRuns stateless functional tests for ClickHouse binaries built in various configurations -- release, debug, with sanitizers, etc✅ success
Stress testRuns stateless functional tests concurrently from several clients to detect concurrency-related errors✅ success
Style checkRuns a set of checks to keep the code style clean. If some of tests failed, see the related log from the report✅ success
Unit testsRuns the unit tests for different release types✅ success
Upgrade checkRuns stress tests on server version from last release and then tries to upgrade it to the version from the PR. It checks if the new server can successfully startup without any errors, crashes or sanitizer asserts✅ success

@alexey-milovidov alexey-milovidov self-assigned this Dec 29, 2024
@rschu1ze
Copy link
Copy Markdown
Member Author

@rschu1ze rschu1ze added this pull request to the merge queue Dec 30, 2024
Merged via the queue into ClickHouse:master with commit e89364b Dec 30, 2024
@rschu1ze rschu1ze deleted the faster-codegen branch December 30, 2024 13:27
@robot-ch-test-poll robot-ch-test-poll added the pr-synced-to-cloud The PR is synced to the cloud repo label Dec 30, 2024
@rschu1ze
Copy link
Copy Markdown
Member Author

rschu1ze commented Jan 2, 2025

Yes that's still 20% slower than before LLVM 16 but nobody said that this PR was going to fix all performance issues 😄

The most reasonable explanation for the slowdown is the transition to the new pass manager: https://github.com/ClickHouse/ClickHouse/pull/69654/files#diff-897102eb40b3e08e4997b80459a83b97ac7d555e4534942cd30581e4f92cefe2

The missing SLP vectorizer enablement (this PR) was obvious but I couldn't find equivalents for

    pass_manager_builder.LoopVectorize = true;
    pass_manager_builder.RerollLoops = true;

in the new infrastructure. But I am not an expert in this, the code is actually copy-paste from https://llvm.org/docs/NewPassManager.html.

Further thoughts and inputs highly welcome ...

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

Labels

pr-not-for-changelog This PR should not be mentioned in the changelog pr-synced-to-cloud The PR is synced to the cloud repo

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants