[ODE] ODE slow on itanium?

[Bill Sellers]

Well gcc is certainly part of the problem. I've just recompiled  
everything using Intel's own icc and icpc using the -fast flag and it  
is quicker but it is still pretty slow. It was 3 times slower than  
the Opteron with gcc, now its only twice as slow which is still  
pretty disappointing (to me, since my University has just invested in  
a great big multiprocessor cluster based on itanium chips). I've also  
tried adding the -ftz flag which flushes denormals to zero on i64  
systems (according to the icc man page anyway) but it makes almost no  
difference at all so it's not that in my case.

Cheers
Bill

On 5 Dec 2006, at 21:57, Francois Beaune wrote:

> one thing to keep in mind is that itanium needs heavy
> assistance from the compiler. i don't know how good
> is gcc for itanium, but i suspect it is largely inferior to
> a native compiler for that processor.
>
> do you have access to the intel compiler for itanium?
>
> francois
>
> ----- Original Message -----
> From: "Andreas Fredriksson" <deplinenoise at gmail.com>
> To: "Bill Sellers" <wis at mac.com>
> Cc: "ode-list list" <ode at q12.org>
> Sent: Monday, December 04, 2006 20:38
> Subject: Re: [ODE] ODE slow on itanium?
>
>
>> On 12/3/06, Bill Sellers <wis at mac.com> wrote:
>>> Hi All,
>>>
>>> I've just been doing some benchmarking of my simulation on various
>>> processors and they all behave about as I expect except for a truly
>>> dreadful showing on a recent 1.6GHz Itanium system. It's over 3  
>>> times
>>> slower than an Opteron 252 which is hardly cutting edge anymore, and
>>> over twice as slow as the Xeon 5160 in my MacPro. They are all
>>> compiled with gcc using the standard "-O3 -ffastmath" options.  
>>> Anyone
>>> got any idea what's wrong since I don't think the Itanium should be
>>> this slow...
>>
>> Make sure you're not gettig denormals propagating through your
>> simulation; from what I've heard they're very slow on Itanium.  
>> They're
>> also very slow on AMD and Intel hardware, unless by chance the
>> precision of the simulation is different enough that you're only
>> seeing the problem on the Itanium box.
>>
>> Good luck,
>> Andreas
>>
>> -- 
>> My other car is a cons.
>> _______________________________________________
>> ODE mailing list
>> ODE at q12.org
>> http://q12.org/mailman/listinfo/ode
> _______________________________________________
> ODE mailing list
> ODE at q12.org
> http://q12.org/mailman/listinfo/ode