Hi Francois
configure linux_intel_compilers
should do the trick. We distribute ccp4 built with the intel compilers for OS X. Part of this is that the macs cover a much smaller range of cpus than linux boxes, so optimisation is less of a problem. If you want speed make sure that you are compiling for your architecture.
My feeling, and I haven't checked this for some time, is that the intel compilers are a bit faster, 5-10% max, but that the gnu compilers have been closing the gap. A pointed out by the previous post the intel compilers basically have an assert that goes if CPU==AMD then stop. A number of years ago somebody published how to remove this from the code, and more recently intel lost a restrictive practice law case over it. So hopefully it will go away.
Charles Ballard
CCP4
ps- a lot of ccp4 is highly io dependent, so fast disks with decent cache can make a lot of difference
On 18 May 2010, at 01:58, Francois Berenger wrote:
> Hello,
>
> This is not what you asked for, but I think
> it is good to know: Intel compilers don't seem
> to like non-Intel CPUs:
>
> http://www.agner.org/optimize/blog/read.php?i=49
>
> Regards,
> F.
>
> Terry Lang wrote:
>> Hey Everyone,
>> I am considering switching from gcc to the Intel compiler in the hopes of making some of calculations run a bit faster. Has anyone ever tried compiling the ccp4 code base with the Intel compilers? Is there a difference in speed? What about in the reproducibility of the calculations? Any changes in statistics? Any information would be greatly appreciated!
>> Thanks,
>> Terry
|