[gmx-developers] bOptFFT

[David van der Spoel]

Erik Lindahl wrote:
> Hi,
> 
> The reason for this is that the optimization in FFTW2 uses slow 
> wallclock timers, so it frequently took several minutes at the start of 
> each run, which of course was a nuisance for short tests - that's why I 
> enabled it optionally a long time ago.
> 
> However, FFTW3 uses step counters with cycle accuracy and thus does the 
> same optimization in less than a second, which is thus always a good 
> idea to use.
> 
> We also support Intel MKL and the Built-in FFTPACK now, so it's a bit 
> unlogical to have an FFTW2-specific option in the mdp file - that's why 
> it was disabled. My primary feeling is to remove the option, but I'm 
> open to suggestions :-)
> 
> Cheers,
> 
> Erik
> 
> 
> 
> On Dec 22, 2006, at 2:59 PM, Ansgar Esztermann wrote:
> 
>> Hello everyone,
>>
>> the bOptFFT flag should tell fftw whether or not different algorithms
>> are to be timed at runtime. However, that option is not passed on to
>> the library. Instead, optimization is always on for fftw3 and always
>> off for fftw2. Back in August, there has been some discussion on this
>> via private email between Carsten Kutzner and David van der Spoel,
>> among others.
>>
>> Unless someone else is already working on the problem, I would like to
>> create a patch to fix this.
>>
>> Any objection?
>>

One reason to be able to turn it off is that you may want to test for 
reproducibility which is hard with this turned on.

-- 
David.
________________________________________________________________________
David van der Spoel, PhD, Assoc. Prof., Molecular Biophysics group,
Dept. of Cell and Molecular Biology, Uppsala University.
Husargatan 3, Box 596,  	75124 Uppsala, Sweden
phone:	46 18 471 4205		fax: 46 18 511 755
spoel at xray.bmc.uu.se	spoel at gromacs.org   http://folding.bmc.uu.se
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++