[Beowulf] cluster building advice?

Vincent Diepeveen diep at xs4all.nl
Wed Sep 26 14:59:13 PDT 2012


Yes easily.

Google for what linus posted there and what i posted there in code  
around 2007 already.

Where i showed how f'ed up GCC was, where it basically modified some  
simple code sample
to something ugly slow, instead of creating a CMOV instruction.

Basically it had the habit to do a comparision, if above zero then  
jump to the end of the function, execute 2 instructions there and  
jump bad.
Real real slow... ...getting the full stall of the pipeline when  
mispredicted.
Other compilers were on that code sample factor 30 faster of course.

Even my simple 32 bits C code it lost 35% there or so. And at C++  
codes it was factor 2+ slower for the bitboard chess engines
as some authors reported.

Linus posted something like: "there is no excuse to not generate CMOV  
instructions now that also intel core2 can execute them fast".

Another 100 of such examplers were there.

PGO speeded up visual studio / intel c++  20-25% and GCC profitted a  
meager 1.5% there.

All that has been changed in 4.6 to 4.7 gcc series. Still not  
perfect, but it makes up a lot of what it was behind.
For some bitboard chess engines it's not 2x slower now, it's for  
example for Stockfish the fastest compiler now.
In fact manages to vectorize.

Of course - only at intel processors. Core2 or even newer i7's.
Still duck slow at AMD.

Yet AMD is not of a big concern here...

If you are first slowest of the universe and then manage to get  
nearby the top compilers, then getting factor 2 faster is no big deal  
Mark.

That's why i wonder about all those distro's still just supporting  
gcc 4.4 and/or 4.5

On Sep 26, 2012, at 8:56 PM, Mark Hahn wrote:

>> If i want to slowdown factor 2 with some software i don't need to
>
> can you show an example of where using even a year-old release of gcc
> causes a factor of 2 slowdown?  especially for middleware or the  
> kernel.
>
> for actual applications, a factor of 2 is surprising but not  
> unbelievable,
> simply because any system will spend most of its time in the  
> application!
> kernel and middleware are simply not normally hotspots.
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin  
> Computing
> To change your subscription (digest mode or unsubscribe) visit  
> http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list