Re: Using POPCNT and other advanced bit manipulation instructions - Mailing list pgsql-hackers

From Alvaro Herrera
Subject Re: Using POPCNT and other advanced bit manipulation instructions
Date
Msg-id 20190215160412.GA7354@alvherre.pgsql
Whole thread Raw
In response to Re: Using POPCNT and other advanced bit manipulation instructions  (Tom Lane <tgl@sss.pgh.pa.us>)
Responses Re: Using POPCNT and other advanced bit manipulation instructions
List pgsql-hackers
On 2019-Feb-15, Tom Lane wrote:

> Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> > Ah, I understand it now:
> > https://stackoverflow.com/questions/25683690/confusion-about-bsr-and-lzcnt/43443701#43443701
> > if you call LZCNT/TZCNT on a CPU that doesn't support it, it won't raise
> > SIGILL or anything ... it'll just silently compute the wrong result.
> > That's certainly not what I call a fallback!
> 
> Yeah, that's pretty nasty; it means there's no backstop for whether
> your choose function gets it right :-(

Hopefully other tests will fail in some visible way, though.  My fear is
whether we have such systems in buildfarm.

> Is POPCNT any better in this respect?

I couldn't find how is POPCNT encoded.  https://stackoverflow.com/a/28803917/242383

I did find these articles:
http://danluu.com/assembly-intrinsics/

https://stackoverflow.com/questions/25078285/replacing-a-32-bit-loop-counter-with-64-bit-introduces-crazy-performance-deviati

This suggests that this all a largely pointless exercise at least on
Intel and GCC/Clang.  It may be better on AMD ... but to get really
better performance we'd need to be coding the popcnt calls in assembly
rather than using the compiler intrinsics, even with -mpopcnt, because
the intrinsics suck.

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


pgsql-hackers by date:

Previous
From: John Naylor
Date:
Subject: Re: Reporting script runtimes in pg_regress
Next
From: Tomas Vondra
Date:
Subject: Re: Protect syscache from bloating with negative cache entries