[jira] [Created] (ARROW-11905) SIMD info always returning none on MacOS

["Jonathan Keane (Jira)" <ji...@apache.org>]

Jonathan Keane created ARROW-11905:
--------------------------------------

             Summary: SIMD info always returning none on MacOS
                 Key: ARROW-11905
                 URL: https://issues.apache.org/jira/browse/ARROW-11905
             Project: Apache Arrow
          Issue Type: Bug
          Components: C++
            Reporter: Jonathan Keane


I'm helping with ARROW-11507 and in testing it I noticed something odd, I'm getting none and none for both simd_level and detected_simd_level which seems odd. I also tried it in pyarrow (below) and I'm getting the same thing (this was built off of HEAD of apache/arrow) so I suspect that this is something lower than either of them (and in both {{runtime_info}} is a think wrapper around the c++).

{code:python}
>>> import pyarrow as pa
>>> pa.runtime_info()
RuntimeInfo(simd_level='none', detected_simd_level='none')
{code}

I do see the following when building:

{code}
--   ARROW_SIMD_LEVEL=SSE4_2 [default=NONE|SSE4_2|AVX2|AVX512]
--       Compile-time SIMD optimization level
--   ARROW_RUNTIME_SIMD_LEVEL=MAX [default=NONE|SSE4_2|AVX2|AVX512|MAX]
--       Max runtime SIMD optimization level
and sse/avx etc are built:

-- Performing Test CXX_SUPPORTS_SSE4_2
-- Performing Test CXX_SUPPORTS_SSE4_2 - Success
-- Performing Test CXX_SUPPORTS_AVX2
-- Performing Test CXX_SUPPORTS_AVX2 - Success
-- Performing Test CXX_SUPPORTS_AVX512
-- Performing Test CXX_SUPPORTS_AVX512 - Success
{code}

I've also tried confirming that ARROW_USER_SIMD_LEVEL is unset, and tried setting it explicitly:

{code}
(pyarrow-source) jkeane@het python % ARROW_USER_SIMD_LEVEL=sse4_2 python
Python 3.9.2 (default, Feb 24 2021, 13:26:09)
[Clang 12.0.0 (clang-1200.0.32.29)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import pyarrow as pa
>>> pa.runtime_info()
RuntimeInfo(simd_level='none', detected_simd_level='none')
{code}


I've tested it on two macOS machines (both running on intel chips) and both had the same behavior above. This appears to work on linux machines just fine.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)