All Pairwise Computation problem using MPI

Efficient distributed algorithm for all pairwise computations using MPI with a logical 1D ring topology and loop unrolling with sequential implementation for benchmark and correctness verification.

Usage

Single process (sequential only)

mpirun -np 1 ./pairwise <rows> <cols> <datatype>

Multiple processes (parallel with sequential verification)

mpirun -np <num_processes> ./pairwise <rows> <cols> <datatype>

Input args

rows: Positive integer N cols: Positive integer M datatype: float or double only - case sensitive num_processes: Number of MPI processes - strictly positive

Examples

mpirun -np 1 ./pairwise 512 256 double
mpirun -np 4 ./pairwise 1024 512 float
mpirun -np 8 ./pairwise 2048 1024 double

Output

Single Process Mode

Performing sequential computation on A: 1024 rows, 512 cols
Sequential computation time: 0.125000 seconds
Peak memory usage: 45.250000 MB

Multi Process Mode

Performing sequential computation on A: 1024 rows, 512 cols
Sequential computation time: 0.125000 seconds
Performing parallel computation on A: 1024 rows, 512 cols
Parallel computation time: 0.045000 seconds
Peak memory usage: 52.125000 MB
Verification: PASSED

Benchmark Results

The benchmark script generates a CSV file with metrics:

DataType,MatrixSize,Processes,SequentialTime,ParallelTime,Speedup,Efficiency,PeakMemoryMB,Verified
float,1024,4,0.125000,0.045000,2.7778,0.6944,52.125000,PASSED
double,2048,8,1.250000,0.180000,6.9444,0.8681,198.750000,PASSED

CSV Columns

• DataType: float or double
• MatrixSize: Matrix dimension (M×M result from M×N input)
• Processes: Number of MPI processes
• SequentialTime: Sequential execution time in seconds
• ParallelTime: Parallel execution time in seconds
• Speedup: SequentialTime / ParallelTime
• Efficiency: Speedup / Processes
• PeakMemoryMB: Peak RSS memory usage in megabytes
• Verified: PASSED/FAILED correctness check