[gmx-developers] Re: FFTW 3.1.2 and wisdom

[Erik Lindahl]

Hi Allen,

On Aug 10, 2007, at 5:23 AM, Allen Smith wrote:

>>
>> First, the actual 3D FFTs account for less than 10% of the runtime
>> with PME, so even a 30% improvement would barely be noticable for
>> your Gromacs performance.
>
> I'm curious as to why the plan calls are done with FFTW_MEASURE, then.

Oh, that's just my fault. Since the measuring is done in milliseconds  
when you have x86 hardware timers I didn't see any reasons not to use  
them when we first implemented it.

However, the drawback is that you are not guaranteed binary identical  
results from repeated runs, which has hit us a couple of times in  
debugging.

Cheers,

Erik