Re: What exactly is our CRC algorithm?

From: Abhijit Menon-Sen <ams(at)2ndQuadrant(dot)com>
To: Heikki Linnakangas <hlinnakangas(at)vmware(dot)com>
Cc: pgsql-hackers(at)postgresql(dot)org, Andres Freund <andres(at)2ndquadrant(dot)com>, Robert Haas <robertmhaas(at)gmail(dot)com>
Subject: Re: What exactly is our CRC algorithm?
Date: 2015-01-09 08:32:03
Message-ID: 20150109083203.GA13128@toroid.org
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Hi Heikki.

I've attached two regenerated CRC patches, split up as before.

1. The slicing-by-8 patch contains numerous changes:

a. A configure test for __builtin_bswap32
b. A comment referencing the slicing-by-8 paper (which is behind a
paywall, unfortunately, so I haven't even read it). Are more
comments needed? If so, where/what kind?
c. A byte-reversed big-endian version of the 8*256 table. In Linux,
there's only one table that uses __constant_swab32, but for us
it's simpler to have two tables.
d. Thanks to (c), we can avoid the bswap32 in the hot loop.
e. On big-endian systems, FIN_CRC32C now bswap32()s the CRC before
finalising it. (We don't need to do this in INIT_CRC32C only
because the initialiser is 0xFFFFFFFF.)

2. The sse4.2 patch has only some minor compile fixes.

I have built and tested both patches individually on little-endian
(amd64) and big-endian (ppc) systems. I verified that the _sse is
chosen at startup on the former, and _sb8 on the latter, and that
both implementations function correctly with respect to HEAD.

Please let me know if there's anything else I need to do.

-- Abhijit

Attachment Content-Type Size
0001-Implement-slicing-by-8-CRC-calculation.patch text/x-diff 63.2 KB
0002-Use-the-SSE4.2-CRC-instructions-where-available.patch text/x-diff 7.3 KB

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Dean Rasheed 2015-01-09 08:38:05 Re: Possible typo in create_policy.sgml
Previous Message Kyotaro HORIGUCHI 2015-01-09 08:30:03 Re: Async execution of postgres_fdw.