Skip to content

[0.8] Implement optimized FP16 support for ARM architecture - [MOD-9078] #644

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.

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

Merged
merged 2 commits into from
Apr 7, 2025

Conversation

GuyAv46
Copy link
Collaborator

@GuyAv46 GuyAv46 commented Apr 7, 2025

Description

Backport of #620 to 0.8.

)

* implement L2 SVE with intermediate casting to f32

* implement IP SVE with f16 ops only

* implements L2 sve with no intermediate casting

* add SVE and SVE2 functions files

* add new files to cmake and use new implementations

* added benchmarks

* fix and switch implementation (due to sve2-only op)

* test with SVE2 intrinsics

* Revert "test with SVE2 intrinsics"

This reverts commit 06dd65c.

* remove redundant implementation

* move to 4 steps per iteration implementations

* add macro cleanup

* fix implementation

* refactor to use 4 accumulators

* added tests

* refactor accumulation

* add initial neon implementation

* fix build flags and file layout

* fix tests

* cleanup and L2 implementation with neon+fp16

* format

* fix test for any arch

* another attempt

* fix test

* rename step functions

* comment-in neon benchmarks

* fix benchmark

* review fixes

* more review fixes

* fixes and cleanup

* fix svwhilelt_b16 calls

* use vbslq_f16

* typo fix

* fix test for OSs that don't support fp16

* added back guards for a specific x86 test

(cherry picked from commit fcc8d78)
@GuyAv46 GuyAv46 requested a review from dor-forer April 7, 2025 11:25
@GuyAv46 GuyAv46 enabled auto-merge April 7, 2025 11:25
Copy link

codecov bot commented Apr 7, 2025

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 96.94%. Comparing base (38dbd58) to head (51d813a).
Report is 1 commits behind head on 0.8.

Additional details and impacted files
@@           Coverage Diff           @@
##              0.8     #644   +/-   ##
=======================================
  Coverage   96.94%   96.94%           
=======================================
  Files          91       91           
  Lines        5069     5069           
=======================================
  Hits         4914     4914           
  Misses        155      155           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

dor-forer
dor-forer previously approved these changes Apr 7, 2025
@GuyAv46 GuyAv46 added this pull request to the merge queue Apr 7, 2025
@github-merge-queue github-merge-queue bot removed this pull request from the merge queue due to failed status checks Apr 7, 2025
@GuyAv46 GuyAv46 enabled auto-merge April 7, 2025 12:26
@GuyAv46 GuyAv46 requested a review from dor-forer April 7, 2025 12:26
@GuyAv46 GuyAv46 added this pull request to the merge queue Apr 7, 2025
Merged via the queue into 0.8 with commit ba011f0 Apr 7, 2025
17 checks passed
@GuyAv46 GuyAv46 deleted the backport-620-to-0.8 branch April 7, 2025 13:45
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants