Thanks. What is your interest in using the window version of the
scatters? Some indication it might be better on some systems then the
standard message passing? Just for fun?
Barry
On Mar 30, 2010, at 7:33 PM, Jed Brown wrote:
> On Sun, 28 Mar 2010 21:56:58 -0500, Barry Smith <bsmith at mcs.anl.gov>
> wrote:
>>
>> Jed,
>>
>> My guess is that it is a fine patch.
>>
>> I put the MPI2 vecscatter together without much thought or
>> testing (it is probably no properly tested in make alltests).
>
> It certainly wasn't tested because it didn't work in 3.0.0. I
> thought I
> had fixed it when I eliminated some errors pre-3.1 and got correct
> results on 2 procs with bs=1, but there were still lingering bugs
> since
> I clearly forgot to test it on a real problem. I've fixed these and
> added -vecscatter_window to runex19_5 in 3.1 so that it is tested and
> not necessarily broken (indeed, I think it is correct now).
>
> Jed
