Re: amber-developers: Testing of PIMD / NEB? (Broken in AMBER 10)

From: Wei Zhang <zgjzweig.gmail.com>
Date: Thu, 11 Sep 2008 09:52:01 -0500

Hi,

     I totally agree. I would suggest to use "sanderrank.eq.0", since
it does not introduce a new
variable and "master_of_sander_group" is way to long.

     Sincerely,

     Wei



On Sep 10, 2008, at 8:06 PM, Scott Brozell wrote:

> Hi,
>
> On Tue, 9 Sep 2008, Wei Zhang wrote:
>
>> I think I found the bug which causes full PIMD fail at 8 cpus. In
>> force.f and runmd.f
>> there are the codes like the following:
>>
>> mpi_reduce( ..., commmaster, ... )
>>
>> which should be replaced by:
>>
>> if(master) mpi_reduce( ..., commmaster, ... )
>>
>> once we fix these, the problem will be gone.
>
> I haven't examined this code, but:
> mpi_reduce is a collective operation and is called by all processes
> in a group. If you are correct then the source of this bug is
> probably
> the confusing naming of masters. It seems that the name master
> should have
> been retired with the multisander approach.
>
> History will repeat itself. At the very least some clear comments
> with
> examples seem appropriate in multisander.f for using the various
> com's.
> Naturally, I would rename master with perhaps master_of_sander_group.
>
> Scott
>
>
>
>
>
Received on Thu Sep 11 2008 - 08:46:06 PDT
Custom Search