Hey, I am implementing a naive partition vector of {1,0,…} by providing the -partvec argument for the amgx_mpi_capi binary in the examples folder of the latest AMGX release.
I am using CUDA Toolkit 7.5 and the latest NVIDIA Driver.
The partition vector is just a file with numberoflines=N of my NxN A matrix. First row is 1 and second is 0 and so on for all N lines. I get to this point:
Parsing configuration string: exception_handling=1 ;
Read partition vector, consisting of 1172788 rows
Read partition vector, consisting of 1172788 rows
Reading matrix dimensions in file: mymatrix.mtx
Reading data…
RHS vector was not found. Using RHS b=[1,…,1]^T
Solution vector was not found. Setting initial solution to x=[0,…,0]^T
Finished reading
Processing partition 1/2 size: 586394 offset 0 nnz 4301699 on device 0
and there the process hangs, running eternally. Nvidia-smi states both GPUs are occupied and there are 2 processes running on 100% on 2 of my CPU cores, exactly as supposed to. Seems like a deadlock to me.
Could anyone share any insight? :)