Skip to content

Conversation

@ViralBShah
Copy link
Member

@ViralBShah ViralBShah commented Oct 15, 2025

It seems like Yggdrasil build of BLIS for mac aarch64 has some problems, which are not obvious from the logs. Given that things work fine on x64 on linux and windows, I am now thinking there's some problem with the default build on mac.

We also have some patches we carry from BLIS 1.0 and while they apply fine, I am not sure if they are required. Could those be causing the problems we are seeing?

cc @imciner2 @giordano

@codecov
Copy link

codecov bot commented Oct 15, 2025

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.66%. Comparing base (b6764e5) to head (1739aff).
⚠️ Report is 16 commits behind head on main.

Additional details and impacted files
@@           Coverage Diff           @@
##             main      #19   +/-   ##
=======================================
  Coverage   91.66%   91.66%           
=======================================
  Files           1        1           
  Lines          12       12           
=======================================
  Hits           11       11           
  Misses          1        1           

☔ 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.

@ViralBShah
Copy link
Member Author

I was able to fix the blis build by forcing the arch to be firestorm instead of arm64 for mac.

@ViralBShah ViralBShah merged commit e7e4008 into main Oct 15, 2025
12 of 14 checks passed
@ViralBShah ViralBShah deleted the vs/blis-2.0 branch October 15, 2025 05:26
@imciner2
Copy link
Contributor

Sounds like the SVE kernels in BLIS might have an issue on the M-series chips. I don't have one to try locally, unfortunately, so I guess firestorm is good enough for now.

@giordano
Copy link

Sounds like the SVE kernels in BLIS might have an issue on the M-series chips

Well, Apple Silicon doesn't have SVE, they also accurately avoided in M4 which is armv9, causing a mess in clang.

@imciner2
Copy link
Contributor

Well, Apple Silicon doesn't have SVE, they also accurately avoided in M4 which is armv9, causing a mess in clang.

Huh, I thought they had SVE in the M series. I wonder if this points to a problem with autodetection logic in BLIS then where it tries to use the wrong kernels and can't get the proper paramters.

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.

4 participants