Skip to content

Conversation

@larsoner
Copy link
Member

@larsoner larsoner commented Jul 11, 2023

Plan:

  1. Show the problem with a timeout by reverting to not ultraslowtest rather than not slowtest:

    • Timed out @ 75% complete in 1h 5m 43s on 43256dd / 2.0.0.dev0+466.gc143f7b93 (OpenBLAS 0.3.23 with 1 thread)
    • np.show_config (in the _infos.sh step) has no Supported SIMD extensions section.
  2. Try the 1.25.1 wheel from PyPI:

    • Completed in 1h 1m 30s on ac99726 / 1.25.1 (OpenBLAS 0.3.23 with 1 thread)

    • 77% complete in ~43m (manual inspection)

    • np.show_config in the _infos.sh step says:

      Supported SIMD extensions in this NumPy install:
          baseline = SSE,SSE2,SSE3
          found = SSSE3,SSE41,POPCNT,SSE42,AVX,F16C,FMA3,AVX2
          not found = AVX512F,AVX512CD,AVX512_KNL,AVX512_KNM,AVX512_SKX,AVX512_CLX,AVX512_CNL,AVX512_ICL
      
  3. Try the 1.25.0rc1 wheel from scientific-python-nightly-wheels

    • Completed in 52m 40s on 3e2a5fc / 1.25.0rc1+218.g0e5a362fd (OpenBLAS 0.3.23 with 1 thread)

    • 76% complete in ~36m (manual inspection)

    • np.show_config (in the _infos.sh step) says:

      Supported SIMD extensions in this NumPy install:
          baseline = SSE,SSE2,SSE3
          found = SSSE3,SSE41,POPCNT,SSE42,AVX,F16C,FMA3,AVX2
          not found = AVX512F,AVX512CD,AVX512_KNL,AVX512_KNM,AVX512_SKX,AVX512_CLX,AVX512_CNL,AVX512_ICL
      

(2) and (3) being faster than (1) will help isolate the time frame of changes and ensure it's not an issue with all SPNW wheels. Looks like the 1.25 RC wheel up here was ~23 days ago, so not an awesome bisect but at least a start.

@larsoner
Copy link
Member Author

It sounds like this is a known issue at the NumPy end because they removed / are refactoring SIMD instruction support!

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.

1 participant