Parallel Computing

MSc project experiences at UNIPD.

Requirements

Install MPI: sudo apt install libcr-dev mpich mpich-doc

Sequential version

Given two square matrices A and B of same size, C is compute as follows:

for (i = 0; i < SIZE; i++)
    for (j = 0; j < SIZE; j++)
        for (k = 0; k < SIZE; k++)
            C[i][j] += A[i][k] * B[k][j]

• Compile with gcc -o sequential -O3 sequential.c
• Run with ./sequential <SIZE> [unlock verbose mode]

Parallel version

Given two square matrices A and B of same size, C is compute exploiting parallelization.

There are three variants of optimization:

1. Just the parallelization of the algorithm, called "mono"
2. "mono" + transposition of matrix B in order to read both A and B matrices row-wise order during the inner loop. This leads to less cache misses.
3. "mono" + changing of the order of the loops: from i-j-k to k-i-j. In this way A[i][k] is constant in the inner loop, while B and C are scanned in row-wise order.

• Compile with mpicc -o parallel -O3 parallel.c
• Run with mpirun -np <NUM-OF-PROCESSORS> ./parallel <SIZE> <VARIANT: 0 | 1 | 2> [unlock verbose mode]

How to run on CAPRI server

Prepare a slurm file called test.slurm (just for example) like

#!/bin/bash

#SBATCH --job-name mmult
#SBATCH --output 2048_monokij_8.out
#SBATCH --error 2048_monokij_8.err
#SBATCH --mail-user leonardo.gemin@studenti.unipd.it
#SBATCH --mail-type ALL
#SBATCH --time 10:00
#SBATCH --ntasks 8
#SBATCH --partition allgroups
#SBATCH --mem 200M


mpicc -o parallel -O3 parallel.c

for ((i = 0; i < 10; i++)) do
    mpirun -np 8 ./parallel 2048 2
done

and submit the job through sbatch test.slurm.

Project report

Click here to see the report of the project.

Authors

• Massimiliano Viola
• Leonardo Gemin