Horton

Re: Detection of hadware feature => please do not use signal

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Thomas Munro <thomas.munro@gmail.com>
Cc: Heikki Linnakangas <hlinnaka@iki.fi>, Bastien Roucariès <rouca@debian.org>, pgsql-bugs@lists.postgresql.org
Date: 2024-11-27T19:57:03Z
Lists: pgsql-bugs 
I wrote:
> Early returns are that they're all picking the +simd option, and
> seem to be just as happy with that as they were without it.
> I wonder if we could just drop the "without" probe altogether.
> It'd save some configure cycles for other platforms (since they'll
> fruitlessly run through all of these tests).

Now that we have a nearly-complete census of the buildfarm,
the end result is that they are all happy with the +simd
spelling ... except for one.  "gull" rejects it:
  clang: error: unsupported argument 'armv8-a+crc+simd' to option '-march='
But without +simd it's okay, so removing the other spelling
would cause a performance regression on that machine.

I find this a bit surprising, because gull is using a reasonably
recent compiler (clang version 17.0.4).  We have much older
clang versions that accept +simd --- but they are all aarch64.
gull seems to be our only representative of clang on arm32.
So this seems like a gcc vs clang difference that we'll have
to put up with.

Anyway, I think this thread is now closed: nothing more to do
unless someone thinks of a platform we need more bespoke code
for.

			regards, tom lane

Commits

  1. Update configure probes for CFLAGS needed for ARM CRC instructions.

  2. Support runtime CRC feature probing on NetBSD/ARM using sysctl().

  3. Use auxv to check for CRC32 instructions on ARM.

  4. Improve our method for probing the availability of ARM CRC instructions.