![]() Registered Member ![]()
|
Dear all, I have now spent several days baffled by this issue so please, any help and comment is appreciated.
The code below generates NaN values when ran on two processors, that is, one processors sends garbage to MPI while the other resolves perfectly. I dont know what is going on but apparently it could be down to pointers sent to MPI? Can anyone help?
I am no expert at all, so please be patient with me. I should also make clear that the version below of the code works absolutely fine, but sending eigen type vectors (or maybe any vectors?) to MPI fails spectacularly. This code is fine
Last edited by leokay on Fri Jul 11, 2014 2:00 pm, edited 1 time in total.
|
![]() Registered Member ![]()
|
I know there is a 'sig.data()' instead of the reference '&sig' somewhere there, so feel free to change it to that when playing with the code, but you still get the NaN values.
Apparently this is the issue, does anyone know how to fix it https://lists.sdsc.edu/pipermail/npaci- ... 38886.html Please, also note that the start vector [4,3] should descend faster to the true answer (minimum), which is why I know for sure there is a problem. If I run the code by sending integers to MPI, the start vector [4,3] always causes the abort. |
Registered users: Bing [Bot], claydoh, Google [Bot], rblackwell, Yahoo [Bot]