Horton

Re: Improve CRC32C performance on SSE4.2

Nathan Bossart <nathandbossart@gmail.com>

View thread
From: Nathan Bossart <nathandbossart@gmail.com>
To: "Devulapalli, Raghuveer" <raghuveer.devulapalli@intel.com>
Cc: John Naylor <johncnaylorls@gmail.com>, "pgsql-hackers@lists.postgresql.org" <pgsql-hackers@lists.postgresql.org>, "Shankaran, Akash" <akash.shankaran@intel.com>
Date: 2025-02-12T22:01:58Z
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.

On Wed, Feb 12, 2025 at 09:48:57PM +0000, Devulapalli, Raghuveer wrote:
>> I think the idea behind USE_SSE42_CRC32C is to avoid the function pointer
>> overhead if possible.  I looked at switching to always using runtime checks for this
>> stuff, and we concluded that we'd better not [0].
> 
> Does that mean we want this feature for the new PCLMUL (and AVX-512) crc32c implementations too? The code for that will look a little ugly, I might need to think about a cleaner way to do this. 

Well, I suspect the AVX-512 version will pretty much always need the
runtime check given that its not available on a lot of newer hardware and
requires a bunch of extra runtime checks (see pg_popcount_avx512.c).  But
it might be worth doing for PCLMUL.  Otherwise, I think we'd have to leave
out the PCLMUL optimizations if built with -msse4.2 -mpclmul because we
don't want to regress existing -msse4.2 users with a runtime check.

-- 
nathan