From: | Michael Paquier <michael(at)paquier(dot)xyz> |
---|---|
To: | Postgres hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org> |
Subject: | Broken ./configure checks for __cpuid() and __cpuidex() |
Date: | 2025-07-29 02:21:41 |
Message-ID: | aIgwNYGVt5aRAqTJ@paquier.xyz |
Views: | Whole Thread | Raw Message | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
Hi all,
While looking at patch 0001 posted by Lukas at [1], the following bit
stood out as an independent fix:
+++ b/configure
unsigned int exx[4] = {0, 0, 0, 0};
- __get_cpuidex(exx[0], 7, 0);
+ __cpuidex(exx, 7, 0);
The patch posted is incorrect because it does not touch configure.ac
and an autoreconf -i or equivalent is missing, but it does not change
the fact that this looks like a valid issue to me.
While looking at the surroundings, I have noticed a second check
that's incorrect in ./configure.ac, with __cpuid(). I may be missing
something, of course, but these two typos have been introduced by
3dc2d62d0486, back in 2015. meson is doing these checks correctly. I
have not been able to test that on cirrus, it seems to run into
problems this morning. I don't know much about these instructions,
but it seems like __get_cpuid_count() and __get_cpuid() are much more
populate than the two others, so the impact may be lighter than I
suspect for builds using ./configure.
Equally, I think that one comment in meson.build is just a consequence
of these two typos.
The attached warrants a backpatch to me, thoughts?
Thanks,
[1]: https://www.postgresql.org/message-id/CAP53Pky-BN0Ui+A9no3TsU=GoMTFpxYSWYtp_LVaDH=y69BxNg@mail.gmail.com
--
Michael
From | Date | Subject | |
---|---|---|---|
Next Message | Michael Paquier | 2025-07-29 02:30:11 | Re: Reduce timing overhead of EXPLAIN ANALYZE using rdtsc? |
Previous Message | Andres Freund | 2025-07-29 01:56:28 | Re: Improve error reporting in 027_stream_regress test |