On Tue, Apr 30, 2013 at 1:55 PM, Simon Riggs <simon@2ndquadrant.com> wrote:
> On 30 April 2013 06:57, Simon Riggs <simon@2ndquadrant.com> wrote:
>
>> I'm about to light up the build farm with a trial commit of the
>> compiler instructions stuff.
>
> Amazingly that seemed to work.
Thanks for committing. Sorry about missing the .h file from the patch.
The two commits look good to me.
I can confirm that compiling with CFLAGS="-O2 -march=native" will
vectorize the committed code on GCC 4.7.
I also checked the situation on clang. clang-3.2 isn't able to
vectorize the loop even with vectorization options. I will check what
is stopping it. If any volunteer has a working build setup with ICC or
MSVC and is willing to run a couple of test compiles, I think we can
achieve vectorization there too.
> ISTM that we also need this patch to put memory barriers in place
> otherwise the code might be rearranged.
The compiler and CPU both have to preserve correctness when
rearranging code, so I don't think we care about it here. It might
matter if these routine could be called concurrently by multiple
backends for a single buffer, but in that case memory barriers won't
be enough, we'd need full exclusion.
Regards,
Ants Aasma
--
Cybertec Schönig & Schönig GmbH
Gröhrmühlgasse 26
A-2700 Wiener Neustadt
Web: http://www.postgresql-support.de
