Horton

Re: define pg_structiszero(addr, s, r)

Bertrand Drouvot <bertranddrouvot.pg@gmail.com>

View thread View on postgr.es
From: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>
To: Ranier Vilela <ranier.vf@gmail.com>
Cc: Michael Paquier <michael@paquier.xyz>, David Rowley <dgrowleyml@gmail.com>, Peter Smith <smithpb2250@gmail.com>, Peter Eisentraut <peter@eisentraut.org>, Heikki Linnakangas <hlinnaka@iki.fi>, pgsql-hackers@lists.postgresql.org
Date: 2024-11-14T12:33:20Z
Lists: pgsql-hackers 
Hi,

On Thu, Nov 14, 2024 at 09:13:19AM -0300, Ranier Vilela wrote:
> Em qui., 14 de nov. de 2024 às 08:58, Bertrand Drouvot <
> Maybe I'm doing something wrong.
> But I'm testing in 32-bit, with the size set to 63, with v12 and I'm seeing
> the SIMD loop execute.

Yeah, that's expected and safe as each iteration reads 32 bytes on 32-bit.

> if (len < sizeof(size_t) * 8) // 8-63 bytes
> failed.
> 
> I expected that with size 63, it would be solved by case 2, or am I wrong?

Case 2 should be read as "in the 4-31" bytes range on 32-bit system as all 
comparisons are done in size_t.

What would be unsafe on 32-bit would be to read up to 32 bytes while len < 32
and that can not happen.

As mentioned up-thread the comments are wrong on 32-bit, indeed they must be read
as:

Case 1: len < 4 bytes
Case 2: len in the 4-31 bytes range
Case 3: len >= 32 bytes

Regards,

-- 
Bertrand Drouvot
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com

Commits

  1. Use pg_memory_is_all_zeros() in PageIsVerifiedExtended()

  2. Optimize pg_memory_is_all_zeros() in memutils.h

  3. Remove use of pg_memory_is_all_zeros() in bufpage.c

  4. Add pg_memory_is_all_zeros() in memutils.h