[gmx-developers] cudaStreamSynchronize failed in cu_blockwait_nb

[Berk Hess]

On 10/22/2012 05:56 PM, Shirts, Michael (mrs5pt) wrote:
>>> But it is good practice that when things fail, even silly parameter choices,
>>> they fail gracefully, as it does help find OTHER bugs.
>> This was supposed to work, not fail and even hang the driver (and the
>> fix is submitted).
> Right, I was merely mentioning that finding bugs in less-common options is
> still a valuable exercise and wanted to commend the people who did it.
>
>> This was supposed to work, not fail and even hang the driver (and the
>> fix is submitted).
>> I was thinking again about changing cut-off electrostatics from a
>> warning to a note,
>> as some people still seem to be using it. But I guess there could be
>> valid uses of it.
> I'd keep it as a warning, not a note.  Any valid use will involve people who
> know what they are doing, and thus can think about it and decide they really
> do want to ignore the warning.
I mis-typed, I meant changing it from a note to a warning.
Currently it is a note, so this would actually require a change.

The check seems pretty fool-proof. Only if you have charges, it checks:
ir->coulombtype == eelCUT && ir->rcoulomb > 0 && !ir->implicit_solvent
So most valid cases (e.g. no cut-off) are checked.

Somewhat related, using reaction-field with an epsilon_rf not set to 
infinity
will produce the same artifacts as a plain cut-off, although to a lesser 
extent.
We might want to put the same note (not warning) there then.

Cheers,

Berk

>
> Best,
> ~~~~~~~~~~~~
> Michael Shirts
> Assistant Professor
> Department of Chemical Engineering
> University of Virginia
> michael.shirts at virginia.edu
> (434)-243-1821
>
>
>