Parquet REL decoding w/IAA(Intel® In-Memory Analytics Accelerator)

["Zhao, Yaqi" <ya...@intel.com>]

Hi, All

We want to add new codes to enable a new RLE Decoding path by using Intel® In-Memory Analytics Accelerator (Intel® IAA).  The new solution provides in general higher performance than the current solution, and also consumes less CPU.

Intel® In-Memory Analytics Accelerator (Intel® IAA) is a hardware accelerator available in the upcoming generation of Intel® Xeon® Scalable processors ("Sapphire Rapids"). Its goal is to speed up common operations in analytics like data (de)compression and filtering.

We encountered 2 questions when submitting PR(14585<https://github.com/apache/arrow/pull/14585>) to Arrow repo:

  1.  Since IAA is a new accelerator in next-generation processors which will be published soon,  the current machine running CI cannot support this feature. I don’t know if it is acceptable that we run the CI offline, maybe on Intel’s lab environment.
  2.   Intel® Query Processing Library (QPL)<https://github.com/intel/qpl> is a library to access the hardware accelerator IAA, is it acceptable to added the dependency in the tool chain?


Hope to get the help from all the professionals on Arrow.

Thank you very much!
Best Regards
Yaqi