[GitHub] [arrow] jianxind commented on pull request #7314: ARROW-8996: [C++] AVX2/AVX512 runtime support for aggregate sum kernel

2020-06-28 Thread GitBox


jianxind commented on pull request #7314:
URL: https://github.com/apache/arrow/pull/7314#issuecomment-650865605


   Close this one as reference of intrinsic. I will try to work out a new 
approach without intrinsic.



This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org




[GitHub] [arrow] jianxind commented on pull request #7314: ARROW-8996: [C++] AVX2/AVX512 runtime support for aggregate sum kernel

2020-06-22 Thread GitBox


jianxind commented on pull request #7314:
URL: https://github.com/apache/arrow/pull/7314#issuecomment-647334741


   > I think we also need a way of setting max runtime instruction set for 
runtime dispatch (apologies if there is one and I missed it)
   
   Thanks, currently no. But it can easily supported by hooking into 
CpuInfo::Init(),  tweak hardware_flags_ while checking the user ENV settings, 
something like "ARROW_RUNTIME_SIMD_LEVEL=none make unittest".
   
   For split PR without intrinsics, I will try to work this. Now I used some 
compiler pragma(clang attribute push or GCC push_options) to support the 
intrinsic API, it's defined to NULL for MSVC as MSVC supports directly using 
intrinsic API without compiler flag. Need find a new method for new ways 
without intrinsic, maybe something like set_source_files_properties can help 
here.



This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org