Upgrade BLISBLAS.jl

[amontoison]

@carstenbauer
I compiled an LP64 version of BLIS with Yggdrasil.
With Julia 1.9, we should be able to combine them with LAPACK_jll and LAPACK32_jll.

[codecov]

Codecov Report

❌ Patch coverage is 90.90909% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 95.00%. Comparing base (6938103) to head (a6ae8cb).
⚠️ Report is 6 commits behind head on main.

Files with missing lines	Patch %	Lines
src/BLISBLAS.jl	90.90%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main       #6      +/-   ##
==========================================
+ Coverage   91.66%   95.00%   +3.33%     
==========================================
  Files           1        1              
  Lines          12       20       +8     
==========================================
+ Hits           11       19       +8     
  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.

[amontoison]

@carstenbauer
I updated this PR because some users wanted to try it in Ipopt.jl. Here are a few key points:

• LBT requires at least Julia 1.9 to work on all platforms.
• BLIS serves only as an alternative to BLAS. Therefore, it must be combined with LAPACK to provide a full alternative to OpenBLAS.
• For many packages compiled with Yggdrasil, we need an LP64 BLAS/LAPACK compiled with 32-bit integers. This PR includes both LP64 and ILP64 versions, enabling functionality similar to MKL.jl or AppleAccelerate.jl.
• OpenBLAS is no longer necessary as we can now utilize LAPACK routines that rely on BLIS.

[carstenbauer]

Haven't reviewed this yet but note that almost all CI runs are failing. Can you address those failures first?

[amontoison]

Haven't reviewed this yet but note that almost all CI runs are failing. Can you address those failures first?

The issues seem related to blis. Since the OpenBLAS backend wasn't cleared, blis symbols were never used.
I opened PR #18 to verify it.

It turns out that using BLISBLAS.jl didn't change the default BLAS backend before this PR.

[imciner2]

We just merged BLIS 2.0 into Yggdrasil, so can this PR be revived and updated to allow BLIS 2.0?

[ViralBShah]

It should be straightforward to rebase this on main now. I am now requiring Julia 1.12 to keep things simple (and perhaps it provides a good starting point to make things work on older versions).

If we enable LAPACK now, we should be able to use the stock tests from LinearAlgebra (and delete the modified tests from LinearAlgebra I just checked in).

[ViralBShah]

We should be able to tag the 0.2 release after this PR is merged.

[amontoison]

@ViralBShah I rebased the PR.

[imciner2]

I think the test failures here are actually due to a bug in LBT - LAPACK_jll is not getting auto detected properly when it is linked against LBT. I think this PR to LBT should fix it: JuliaLinearAlgebra/libblastrampoline#168.

[ViralBShah]

@amontoison Should we register a new release or are there other things we should do before that?

[amontoison]

We can register a new release.

[ViralBShah]

Registered, along with a few more improvements. Verified this all works with Julia 1.9 onwards.