This is an awesome tip thank you so much. I have written an embarassingly parallel reader (since i am reading from several files) but i was worried that my main program won't be able to use more processors than the number of files, but this tip should save me so much trouble. Thank you.
2017-07-04 19:08 GMT+02:00 Barry Smith <[email protected]>: > > We recommend writing a utility program that runs on one process and > reads in the files putting the values into a matrix then using MatView() > with a binary viewer to save the matrix to a file. > > Then in your parallel program you simply call MatLoad() and it > efficiently loads up the matrix into your program in parallel. > > This means you don't need to write a parallel reader, you only need > to write a sequential one which is easier. > > Barry > > > On Jul 4, 2017, at 4:22 AM, errabii sohaib <[email protected]> > wrote: > > > > Hi, > > > > I am trying to use petsc4py and slepc4py to read a matrix that is > distributed on several files, In which i save the non zero elements by > specifiying three arrays (A, I, J) for Coeffs, rows and columns indices. > > I am currently using MUMPS and i simply read a file and pass the A,I,J > arrays to MUMPS with each process. I am trying to do the same, however i am > still confused how matrices are structured in PETSc and if its even > possible to read n files and pass the matrix to PETsc to solve with > slepc4py using atleast n processes. > > > > Thank you very much for your time, > > Sohaib. > > > > > > > >
