>>> Perri put in MKL's FFT for the CPU based hybrid precision model
>>> but I don't think that was added for double precision in the trunk - Perri?
>>
>>
>> Yes that is correct. No updates were made to the src files. Also, if
>> configure detects MKL_HOME in the environment and the intel compiler is
>> being used the FFTW_FFT execution path is followed as opposed to PUBFFT.
>
> Not sure what you mean by "that is correct" here. Is there a separate
> branch for pmemd/CPU/hybrid precision? Or is it in master?
Yes, there is a separate branch, the hybrid model ran into some issues and so wasn’t merged into master or added to the double precision trunk. The behavior you and Jason are seeing is expected and FFTW3 is definitely using double precision mode as the only changes introduced into the master branch by Ashraf and myself were in the configure2 script only.
Regards,
Perri
On Oct 14, 2014, at 8:48 AM, David A Case <case.biomaps.rutgers.edu> wrote:
> On Tue, Oct 14, 2014, Needham, Perri wrote:
>
>>> Perri put in MKL's FFT for the CPU based hybrid precision model
>>> but I don't think that was added for double precision in the trunk - Perri?
>>
>>
>> Yes that is correct. No updates were made to the src files. Also, if
>> configure detects MKL_HOME in the environment and the intel compiler is
>> being used the FFTW_FFT execution path is followed as opposed to PUBFFT.
>
> Not sure what you mean by "that is correct" here. Is there a separate
> branch for pmemd/CPU/hybrid precision? Or is it in master? Both
> Jason and I find that FFTW3 (with using the native code or the MKL
> alternative) seems to work fine in the master branch in what I assume
> is double-precision mode. I'm not sure if that is what you and Ross
> are expecting or not. Expanded explanation is welcome, since we may be
> missing something here.
>
> ...thx...dac
>
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Oct 14 2014 - 09:30:02 PDT
