Date: Wed, 13 Feb 2019 20:17:09 -0300
From: Alvaro Herrera <alvherre@2ndquadrant.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Andrew Gierth <andrew@tao11.riddles.org.uk>,
	David Rowley <david.rowley@2ndquadrant.com>,
	PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>,
	Thomas Munro <thomas.munro@enterprisedb.com>
Subject: Re: Using POPCNT and other advanced bit manipulation instructions
Message-ID: <20190213231709.GA12159@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <26998.1550097202@sss.pgh.pa.us>
User-Agent: Mutt/1.9.4 (2018-02-28)
Precedence: bulk

On 2019-Feb-13, Tom Lane wrote:

> Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> > and we have defined pg_popcount64() like this:
> 
> > static int
> > pg_popcount64_sse42(uint64 word)
> > {
> > 	return __builtin_popcountl(word);
> > }
> 
> That is clearly completely broken.

Pushed my proposed fix, which includes removing the configure tests for
builtins of varying widths.  I couldn't resist sorting entries
alphabetically in configure.in.  (I also used autoheader to produce the
new pg_config.h, which showed me that David had not used it to generate
his diffs there.)

For pg_config.h.win32 I used the compiler explorer tool I just learned
about, and came to the conclusion that MSVC's compiler does not
implement these builtins.

I didn't do anything about the const-cast-away in pg_popcount() yet.
I think that should use PointerIsAligned() instead of what it's doing
now.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services