Open to contributions. Submit a pull request and I'm happy to take a look.
Calculating the run time of various ways of generating a histogram using different parallel and distributed computing algorithms. Using the serial version as a base for comparision.
Note : I haven't uploaded the serial version yet. And there are a few seg faults in MPI for now. I'll redo the whole MPI thing.
- MPI : Using a distributed memory system to distribute data over nodes and simultaneously reduce and generate a histogram.
- OMP : Using a shared memory system to parallelize blocks of code to optimize performance.
- OMP + MPI : Using parallelized
forloops over distributed memory to generate a histogram.
Clone the repo and generate a data file for the program to read.
gcc -g -Wall -o create_file create_file.c -std=c99 (I'm using for loops so I need c++ implementation.)
./create_file 'name of data file' 'number of elements'
To compile the exec files :
gcc -g -Wall -mpi -fopenmp -o execfile mpi_file.c -std=c99 (Pick and choose the optional depending on the program you're running.)
Note: You want to pay close attention to the order of tags used in compilation. Example gcc -g -Wall -o -fopenmp execfile mpi_file.c -std=c99 will not compile.
All the programs take similar command line arguments.
- Serial version:
./nameofexec 'data file to read from' '# of intervals of histogram' - MPI version:
./nameofexec '# of nodes of MPI' 'data_file_to_read_from' '# of intervals' - OMP version:
./nameofexec '# of threads' 'data_file_to_read_from' '# of intervals'
I've added error checks in the programs to ensure that the minimum # of command line arguments are provided by the user.
Also painstakingly made sure that there are no segfaults.
Please ensure you generate a data file BEFORE you run the exec file.
I've tried to add verbose comments about functions wherever I can, if you feel I'm lacking submit a pull request and I'll take a look.