Re: Using POPCNT and other advanced bit manipulation instructions

From: Alvaro Herrera <alvherre(at)2ndquadrant(dot)com>
To: Andrew Gierth <andrew(at)tao11(dot)riddles(dot)org(dot)uk>
Cc: David Rowley <david(dot)rowley(at)2ndquadrant(dot)com>, PostgreSQL Hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org>, Thomas Munro <thomas(dot)munro(at)enterprisedb(dot)com>
Subject: Re: Using POPCNT and other advanced bit manipulation instructions
Date: 2019-02-13 22:17:19
Message-ID: 20190213221719.GA15976@alvherre.pgsql
Views: Raw Message | Whole Thread | Download mbox
Thread:
Lists: pgsql-hackers

On 2019-Feb-13, Alvaro Herrera wrote:

> It definitely is ... plans have changed from using IndexOnly scans to
> Seqscans, which is likely fallout from the visibilitymap_count() change.

I think the problem here is that "unsigned long" is 32 bits in this
machine:
checking whether long int is 64 bits... no

and we have defined pg_popcount64() like this:

static int
pg_popcount64_sse42(uint64 word)
{
return __builtin_popcountl(word);
}

so it's counting bits in the lower half of the uint64.

If that's correct, then I think we need something like this patch. But
it makes me wonder whether we need a configure test for
__builtin_popcountll() and friends. I wonder if there's any compiler
that implements __builtin_popcountl() but not __builtin_popcountll() ...
and if not, then the test for __builtin_popcountl() should be removed,
and have everything rely on the one for __builtin_popcount().

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

Attachment Content-Type Size
uint64-is-longlong.patch text/x-diff 1.2 KB

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Tom Lane 2019-02-13 22:33:22 Re: Using POPCNT and other advanced bit manipulation instructions
Previous Message Andres Freund 2019-02-13 21:51:47 Re: subscriptionCheck failures on nightjar