Horton

Re: Improve CRC32C performance on SSE4.2

John Naylor <johncnaylorls@gmail.com>

View thread
From: John Naylor <johncnaylorls@gmail.com>
To: Nathan Bossart <nathandbossart@gmail.com>
Cc: "Devulapalli, Raghuveer" <raghuveer.devulapalli@intel.com>, "pgsql-hackers@lists.postgresql.org" <pgsql-hackers@lists.postgresql.org>, "Shankaran, Akash" <akash.shankaran@intel.com>
Date: 2025-03-06T11:45:40Z
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. Include _mm512_zextsi128_si512() in AVX-512 configure probes.

  2. Properly fix AVX-512 CRC calculation bug

  3. Workaround code generation bug in clang

  4. Compute CRC32C using AVX-512 instructions where available

  5. Inline CRC computation for small fixed-length input on x86

  6. Be more paranoid in configure's checks for CRC and POPCNT intrinsics.

Attachments

On Wed, Mar 5, 2025 at 10:52 PM Nathan Bossart <nathandbossart@gmail.com> wrote:
>
> On Wed, Mar 05, 2025 at 08:51:21AM +0700, John Naylor wrote:
> > That was my hunch too, but I wanted to be more sure, so I modified the
> > benchmark so it doesn't know the address of the next calculation until
> > it finishes the last calculation so we can hopefully see the latency
> > caused by indirection. It also does an additional calculation on
> > constant 20 bytes, like the WAL header. I also tweaked the length each
> > iteration so the branch predictor maybe has a harder time predicting
> > the constant 20 input. And to make it more challenging, I removed the
> > part that inlined all small inputs, so it inlines only constant
> > inputs:
>
> Would you mind sharing this test?

The test script is the same as here, except I only ran small lengths:

https://www.postgresql.org/message-id/CANWCAZahvhE-%2BhtZiUyzPiS5e45ukx5877mD-dHr-KSX6LcdjQ%40mail.gmail.com

...but I must have forgotten to attach the slightly tweaked patch set,
which I've done now. 0002 modifies the 0001 test module and 0006
reverts inlining non-constant input from 0005, just to see if I could
find a regression from indirection, which I didn't. If we don't need
it, it'd better to avoid inlining loops to keep from bloating the
binary.

> It sounds like you are running a
> workload with a mix of constant/inlined calls and function pointer calls to
> simulate typical usage for WAL, but I'm not 100% sure I'm understanding you
> correctly.

Exactly.

--
John Naylor
Amazon Web Services