The code now does this. One can set IG < -1 in the GPU code and it will use the CPU RNG which shoudl address the issue fo tests failures with tests that use the random number generator. It's on the very very long todo list soemwhere before the next release but if someone want to volunteer to update the test cases great.
As for the other diffs - that needs a rework of dacdif most likely - or we build both SPFP and DPFP precision models and use the DPFP model for running the tests.
All the best
Ross
> On Mar 16, 2015, at 9:04 AM, Daniel Roe <daniel.r.roe.gmail.com> wrote:
>
> Hi All,
>
> I'd like to revisit the issue of the GPU test cases showing lots of
> diffs due to different underlying GPU hardware/RNGs etc. I think that
> 1) the option to use the CPU RNG so that stochastic tests will not
> rely on widely varying GPU RNGs (I think this was suggested by Jason a
> while back) should be implemented, and 2) the remaining tests should
> have their tolerances adjusted to what we believe is acceptable. My
> fear is that since there are so many differences now (on local
> machines here for example I get '39 file comparisons failed') it may
> be easier for bugs to creep in. Thoughts?
>
> -Dan
>
> --
> -------------------------
> Daniel R. Roe, PhD
> Department of Medicinal Chemistry
> University of Utah
> 30 South 2000 East, Room 307
> Salt Lake City, UT 84112-5820
> http://home.chpc.utah.edu/~cheatham/
> (801) 587-9652
> (801) 585-6208 (Fax)
>
> _______________________________________________
> 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 Mon Mar 16 2015 - 09:30:03 PDT
