Thanks to visit codestin.com
Credit goes to github.com

Skip to content

[wasm] Use packedsimd in hexconverter #115062

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
Merged
merged 46 commits into from
May 6, 2025
Merged

[wasm] Use packedsimd in hexconverter #115062

merged 46 commits into from
May 6, 2025
+493 −170

Conversation

lewing
Copy link
Member

@lewing lewing commented Apr 25, 2025 •
edited
Loading

Implement a PackedSimd vectorized version of Hexconverter.TryDecodeFromUtf16

Then fix several things that were broken in the AOT intrinsics to make that optimization work.

@dotnet-issue-labeler dotnet-issue-labeler bot added the needs-area-label An area label is needed to ensure this gets routed to the appropriate area owners label Apr 25, 2025
@lewing lewing requested review from kg and radekdoulik April 25, 2025 20:37
kg
kg reviewed Apr 25, 2025
View reviewed changes
Copy link
Member

@kg kg left a comment

Choose a reason for hiding this comment

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

Looks right, with the caveat that I didn't test the new implementation myself

@lewing lewing marked this pull request as ready for review April 26, 2025 01:41
@build-analysis build-analysis bot mentioned this pull request Apr 26, 2025
Open
@lewing
Copy link
Member Author

lewing commented Apr 26, 2025 •
edited
Loading

/ba-g coreclr failures are not related

@lewing
Copy link
Member Author

lewing commented Apr 26, 2025

Looks right, with the caveat that I didn't test the new implementation myself

the select mask was swapped but the tests caught it.

@lewing lewing removed the needs-area-label An area label is needed to ensure this gets routed to the appropriate area owners label Apr 26, 2025
@build-analysis build-analysis bot mentioned this pull request Apr 26, 2025
Open
@lewing
Copy link
Member Author

lewing commented Apr 27, 2025

Looks like there are some regressions with invalid values, taking a look.

@lewing
Copy link
Member Author

lewing commented Apr 27, 2025

It looks like the tests are only failing in windows without aggressive trimming, in CI, but not locally. ugh

@build-analysis build-analysis bot mentioned this pull request Apr 27, 2025
Open
@lewing
Copy link
Member Author

lewing commented Apr 28, 2025

@radekdoulik it looks like the failing paths are on the AOT builds

@build-analysis build-analysis bot mentioned this pull request Apr 29, 2025
Open
3 tasks
This was referenced Apr 29, 2025
Open
Open
@radekdoulik
Copy link
Member

@radekdoulik it looks like the failing paths are on the AOT builds

I have no luck replicating it locally. Tried both Release and Debug configurations.

I was running EMSDK_PATH=/Users/rodo/git/runtime/src/mono/browser/emsdk ./dotnet.sh build -c Debug /t:test /p:_WasmAllowAOTDebug=true /p:EnableAggressiveTrimming=true /p:RunAOTCompilation=true /p:TargetOS=browser src/libraries/System.Runtime/tests/System.Runtime.Tests and similar for Release.

@lewing
Copy link
Member Author

lewing commented Apr 29, 2025 •
edited
Loading

Passed locally with AOT on windows using the same version of chrome. I wonder if this is another msvc error in the cross compiler (see: #114786)

 info: Finished:    System.Runtime.Tests.dll
  info: Stored D:\dotnet\runtime\artifacts\bin\System.Runtime.Tests\Release\net10.0-browser\browser-wasm\AppBundle\xhar
  ness-output\testResults.xml results 17698363 bytes
  info: Finished uploading 17698363 bytes of RESULTXML
  info: Xml file was written to the provided writer.
  info:
  info: === TEST EXECUTION SUMMARY ===
  info: Tests run: 64996 Passed: 64882 Inconclusive: 0 Failed: 0 Ignored: 0 Skipped: 114
  info:
  info: test-main.js exiting WasmTestRunner.dll System.Runtime.Tests.dll with result 0 and linear memory 300154880 byte
  s
  info: MONO_WASM: forceDisposeProxies done: 14 imports, 0 exports, 0 GCHandles, 0 JSHandles.
  info: WASM EXIT 0
  info: Waiting to flush log messages with a timeout of 120 secs ..
  info: Closing 1 browser tabs before setting the main tab to config page and quitting.
  XHarness exit code: 0
  ----- end Tue 04/29/2025 17:03:46.84 ----- exit code 0 ----------------------------------------------------------
  XHarness artifacts: D:\dotnet\runtime\artifacts\bin\System.Runtime.Tests\Release\net10.0-browser\browser-wasm\AppBund
  le\xharness-output

Build succeeded.
    0 Warning(s)
    0 Error(s)

Time Elapsed 00:14:30.11

@lewing
Copy link
Member Author

lewing commented Apr 29, 2025

cc @akoeplinger

lewing added a commit that referenced this pull request May 5, 2025
@lewing
[wasm][aot] Reenable AOT System.Runtime.Intrinsics/tests and fix prob…
e8617b7 
…lems that exposed (#115204)

Enable System.Runtime.Intrinsics/tests in wasm smoke tests then fix the Shuffle intrinsics crash the tests exposed.

See #115062 for details.

The crash would happen when a non-constant index vector was passed to the llvm shuffle intrinsic which was exposed when the ShuffleNative path was added to the tests. The PR works around it by using the swizzle intrinsic which does not require a const index vector. If this does cause a regression it we should be able to restore the original implementation when we can verify a constant index vector is being passed.

This also fixes the PackedSimd.Min/Max intrinsic implementation which had accidentally reversed min an max.

Fixes #115203
* Fix the aot min/max intrinsics
This was referenced May 5, 2025
Open
Open
Open
Open
kg
kg reviewed May 5, 2025
View reviewed changes
This was referenced May 5, 2025
Draft
Open
@build-analysis build-analysis bot mentioned this pull request May 5, 2025
Open
@lewing
Copy link
Member Author

lewing commented May 6, 2025

I've extracted out some of the marginally related changes. What is left in the pr now is:

  • a fix for PackedSimd.BitwiseSelect argument ordering
  • removing OP_WASM_ONESCOMPLEMENT and using the common path for OP_ONES_COMPLEMENT now that llvm is fixed
  • additional tests
  • The original optimization (aka the first two commits)

I'm happy to split out the fixes completely from the optimization but I think it is reviewable as is.

@lewing lewing removed the NO-MERGE The PR is not ready for merge yet (see discussion for detailed reasons) label May 6, 2025
@lewing lewing requested a review from kg May 6, 2025 16:10
@lewing
Copy link
Member Author

lewing commented May 6, 2025

/ba-g failures are not related

@lewing lewing merged commit 2f8efe6 into dotnet:main May 6, 2025
152 of 166 checks passed
@lewing lewing deleted the hexconvert-packedsimd branch May 6, 2025 22:46
@build-analysis build-analysis bot mentioned this pull request May 6, 2025
Open
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@radekdoulik radekdoulik radekdoulik approved these changes

@maraf maraf Awaiting requested review from maraf

@akoeplinger akoeplinger Awaiting requested review from akoeplinger

@steveisok steveisok Awaiting requested review from steveisok steveisok is a code owner

@vitek-karas vitek-karas Awaiting requested review from vitek-karas vitek-karas is a code owner

@pavelsavara pavelsavara Awaiting requested review from pavelsavara

@BrzVlad BrzVlad Awaiting requested review from BrzVlad BrzVlad is a code owner

@kotlarmilos kotlarmilos Awaiting requested review from kotlarmilos kotlarmilos is a code owner

@kg kg Awaiting requested review from kg

Assignees

@lewing lewing

Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@lewing @radekdoulik @kg