LAPACK v3.11.0

[h-vetinari]

Redo #114, now that MKL has miraculously been unblocked

[conda-forge-admin]

Hi! This is the friendly automated conda-forge-linting service.

I wanted to let you know that I linted all conda-recipes in your PR (recipe/meta.yaml) and found some lint.

Here's what I've got...

For recipe/meta.yaml:

• ❌ You're setting a constraint on the __osx virtual package directly; this should now be done by adding a build dependence on {{ stdlib("c") }}, and overriding c_stdlib_version in recipe/conda_build_config.yaml for the respective platform as necessary. For further details, please see META: {{ stdlib("c") }} migration conda-forge.github.io#2102.

For recipe/meta.yaml:

• ℹ️ The recipe is not parsable by parser conda-souschef (grayskull). This parser is not currently used by conda-forge, but may be in the future. We are collecting information to see which recipes are compatible with grayskull.
• ℹ️ The recipe is not parsable by parser conda-recipe-manager. The recipe can only be automatically migrated to the new v1 format if it is parseable by conda-recipe-manager.

_{This message was generated by GitHub Actions workflow run https://github.com/conda-forge/conda-forge-webservices/actions/runs/19385999972. Examine the logs at this URL for more detail.}

[isuruf]

We filter the symbols in this file whereas create-forwarder-dll doesn't.

[h-vetinari]

You mean this part, right?

blas-feedstock/recipe/create_forwarder_dll.py

Lines 35 to 38 in bb5fc01

	# A crude way to filter out unwanted symbols like fprintf which should
	# not have been exported in netlib libraries. probably a flang bug
	if symbol.startswith(("c", "s", "L", "d", "z", "i", "l", "x", "C", "R")):
	symbols_lib_pairs.append((symbol, libname))

Would you be open to add this as an option in https://github.com/isuruf/create-forwarder-dll?

[isuruf]

Yes, that part. I'm open to adding it. Maybe as a symbol filter regex?

[isuruf]

@carterbox can you run NVPL tests?

[h-vetinari]

Not very surprisingly (IMO), newaccelerate fails the 3.11 tests; AFAIU it only advertises compatibility with lapack 3.9. OTOH, the failures don't look catastrophically bad:

94% tests passed, 6 tests failed out of 93

Total Test time (real) =  41.32 sec

The following tests FAILED:
	  7 - LAPACK-xeigtsts_sec_in (Failed)
	 28 - LAPACK-xeigtstd_dec_in (Failed)
	 44 - LAPACK-xlintstc_ctest_in (Failed)
	 50 - LAPACK-xeigtstc_cec_in (Failed)
	 66 - LAPACK-xlintstz_ztest_in (Failed)
	 72 - LAPACK-xeigtstz_zec_in (Failed)

			-->   LAPACK TESTING SUMMARY  <--
		Processing LAPACK Testing output found in the TESTING directory
SUMMARY             	nb test run 	numerical error   	other error  
================   	===========	=================	================  
REAL             	385456		2	(0.001%)	3010	(0.781%)	
DOUBLE PRECISION	404304		0	(0.000%)	3032	(0.750%)	
COMPLEX          	345386		6032	(1.746%)	6745	(1.953%)	
COMPLEX16         	268973		7479	(2.781%)	6820	(2.536%)	

--> ALL PRECISIONS	1404119		13513	(0.962%)	19607	(1.396%)	

However, if we do not want to simply skip those xeigtst[sdcz] resp. xlintst[cz] tests, then I think we should probably allow the maximum lapack version to become independent per flavour; though we'd likely have to keep the compile-time pinning on 3.9 pretty much indefinitely then (for newaccelerate to remain something that can be switched in at runtime), which isn't very appealing TBH, but then I also don't see a viable alternative if we want to keep supporting this.

[carterbox]

I built 2569478 natively for both NVPL and openBLAS. The tests pass.

[h-vetinari]

@isuruf, I skipped the offending tests on newaccelerate (67bc0b7), which I think is the easier way forward for now, in terms of keeping a uniform baseline. PTAL

[h-vetinari]

Ping @isuruf

Planning to merge this in 72h unless there's further comments.

[rouault]

I'm sorry if I don't report in the right repo, but there's something wrong with the lapack/libblas 3.11 update.
It prevents from importing numpy on Windows:

C:\dev\miniconda3\envs\myenv\Lib\site-packages\numpy\__init__.py", line 454, in <module>
    from . import lib
  File "C:\dev\miniconda3\envs\myenv\Lib\site-packages\numpy\lib\__init__.py", line 18, in <module>
    from . import (
  File "C:\dev\miniconda3\envs\myenv\Lib\site-packages\numpy\lib\_arraypad_impl.py", line 8, in <module>
    from numpy.lib._index_tricks_impl import ndindex
  File "C:\dev\miniconda3\envs\myenv\Lib\site-packages\numpy\lib\_index_tricks_impl.py", line 8, in <module>
    import numpy.matrixlib as matrixlib
  File "C:\dev\miniconda3\envs\myenv\Lib\site-packages\numpy\matrixlib\__init__.py", line 4, in <module>
    from . import defmatrix
  File "C:\dev\miniconda3\envs\myenv\Lib\site-packages\numpy\matrixlib\defmatrix.py", line 13, in <module>
    from numpy.linalg import matrix_power
  File "C:\dev\miniconda3\envs\myenv\Lib\site-packages\numpy\linalg\__init__.py", line 87, in <module>
    from . import (
  File "C:\dev\miniconda3\envs\myenv\Lib\site-packages\numpy\linalg\_linalg.py", line 99, in <module>
    from numpy.linalg import _umath_linalg
ImportError: DLL load failed while importing _umath_linalg:

After downgrading with conda install "libblas<3.11", I can import numpy again

[h-vetinari]

@rouault, can you please open an issue and fill out the template (conda list is important for this one). I'm almost certain that I discovered the same issue in #154 and am working on a fix in conda-forge/lapack-feedstock#92

[rouault]

can you please open an issue

done in #155