Horton

RE: Popcount optimization using AVX512

Amonson, Paul D <paul.d.amonson@intel.com>

View thread
From: "Amonson, Paul D" <paul.d.amonson@intel.com>
To: Nathan Bossart <nathandbossart@gmail.com>
Cc: Andres Freund <andres@anarazel.de>, Alvaro Herrera <alvherre@alvh.no-ip.org>, "Shankaran, Akash" <akash.shankaran@intel.com>, Noah Misch <noah@leadboat.com>, Tom Lane <tgl@sss.pgh.pa.us>, "Matthias van de Meent" <boekewurm+postgres@gmail.com>, "pgsql-hackers@lists.postgresql.org" <pgsql-hackers@lists.postgresql.org>
Date: 2024-03-14T19:50:46Z
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

> -----Original Message-----
> From: Nathan Bossart <nathandbossart@gmail.com>
> Sent: Monday, March 11, 2024 6:35 PM
> To: Amonson, Paul D <paul.d.amonson@intel.com>

> Thanks.  There's no need to wait to post the AVX portion.  I recommend using
> "git format-patch" to construct the patch set for the lists.

After exploring git format-patch command I think I understand what you need. Attached.
 
> > What exactly do you suggest here? I am happy to always call either
> > pg_popcount32() or pg_popcount64() with the understanding that it may
> > not be optimal, but I do need to know which to use.
> 
> I'm recommending that we don't change any of the code in the pg_popcount()
> function (which is renamed to pg_popcount_slow() in your v6 patch).  If
> pointers are 8 or more bytes, we'll try to process the buffer in 64-bit chunks.
> Else, we'll try to process it in 32-bit chunks.  Any remaining bytes will be
> processed one-by-one.

Ok, we are on the same page now. :)  It is already fixed that way in the refactor patch #1.

As for new performance numbers: I just ran a full suite like I did earlier in the process. My latest results an equivalent to a pgbench scale factor 10 DB with the target column having varying column widths and appropriate random data are 1.2% improvement with a 2.2% Margin of Error at a 98% confidence level. Still seeing improvement and no regressions.

As stated in the previous separate chain I updated the code removing the extra "extern" keywords.

Thanks,
Paul