Horton

Re: Popcount optimization using AVX512

Nathan Bossart <nathandbossart@gmail.com>

View thread
From: Nathan Bossart <nathandbossart@gmail.com>
To: "Amonson, Paul D" <paul.d.amonson@intel.com>
Cc: Tom Lane <tgl@sss.pgh.pa.us>, David Rowley <dgrowleyml@gmail.com>, Andres Freund <andres@anarazel.de>, Alvaro Herrera <alvherre@alvh.no-ip.org>, "Shankaran, Akash" <akash.shankaran@intel.com>, Noah Misch <noah@leadboat.com>, Matthias van de Meent <boekewurm+postgres@gmail.com>, "pgsql-hackers@lists.postgresql.org" <pgsql-hackers@lists.postgresql.org>
Date: 2024-03-27T22:00:10Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Fix __attribute__((target(...))) usage.

  2. Use __attribute__((target(...))) for AVX-512 support.

  3. Fix code for probing availability of AVX-512.

  4. Optimize visibilitymap_count() with AVX-512 instructions.

  5. Optimize pg_popcount() with AVX-512 instructions.

  6. Inline pg_popcount() for small buffers.

  7. Avoid function call overhead of pg_popcount() in syslogger.c.

  8. Refactor code for setting pg_popcount* function pointers.

  9. Inline pg_popcount{32,64} into pg_popcount().

  10. Remove MSVC scripts

  11. Use ARMv8 CRC instructions where available.

  12. Use Intel SSE 4.2 CRC instructions where available.

Attachments

On Mon, Mar 25, 2024 at 03:05:51PM -0500, Nathan Bossart wrote:
> On Mon, Mar 25, 2024 at 06:42:36PM +0000, Amonson, Paul D wrote:
>> Ok, CI turned green after my re-post of the patches.  Can this please get
>> merged?
> 
> Thanks for the new patches.  I intend to take another look soon.

Thanks for your patience.  I spent most of my afternoon looking into the
latest patch set, but I needed to do a CHECKPOINT and take a break.  I am
in the middle of doing some rather heavy editorialization, but the core of
your changes will remain the same (and so I still intend to give you
authorship credit).  I've attached what I have so far, which is still
missing the configuration checks and the changes to make sure the extra
compiler flags make it to the right places.

Unless something pops up while I work on the remainder of this patch, I
think we'll end up going with a simpler approach.  I originally set out to
make this look like the CRC32C stuff (e.g., a file per implementation), but
that seemed primarily useful if we can choose which files need to be
compiled at configure-time.  However, the TRY_POPCNT_FAST macro is defined
at compile-time (AFAICT for good reason [0]), so we end up having to
compile all the files in many cases anyway, and we continue to need to
surround lots of code with "#ifdef TRY_POPCNT_FAST" or similar.  So, my
current thinking is that we should only move the AVX512 stuff to its own
file for the purposes of compiling it with special flags when possible.  (I
realize that I'm essentially recanting much of my previous feedback, which
I apologize for.)

[0] https://postgr.es/m/CAApHDvrONNcYxGV6C0O3ZmaL0BvXBWY%2BrBOCBuYcQVUOURwhkA%40mail.gmail.com

-- 
Nathan Bossart
Amazon Web Services: https://aws.amazon.com