I have tried dividing the probability distribution function into codistributed 1D matrices inside the spmd block. However, the parallelised code runs slower than the serial code. I believe that I am doing it incorrectly. Please provide relevant study material and codes that might be of some help.