OpenMP, MPI and CUDA Parallelization of Recursive Functions

Final Project - High Performance Computing - A.Y. 2022/2023

- Edoardo Pastorino, 5169595

- Mahyar Sadeghi Garjan, 5283082

Introduction:

The aim of the project is to try to implement and explain a possible approach to parallelize recursive function, based on already existent experiments found online, with the usage of three of the most diffused interfaces for exploiting parallelism inside the code of a programming language. These API are OpenMP, MPI and CUDA. All these three parallelization methods are implemented, in our work, inside program in C language, due to the fact that are well supported by the language and by the relative compilers.

Project Structure:

.

├── fibonacci

│   └── fibonacci.c

├── README.md

├── Report_ProjectHPC_Pastorino_Sadeghi.pdf

├── fibonacci_OpenMP

│   ├── fibonacci_par1.c

│   ├── fibonacci_par2.c

│   └── fibonacci_par3.c

├── fibonacci_MPI

│   ├── fib.c

│   ├── MakeFile

│   └── mpi-fib-start.c

├── fibonacci_MPI_OMP

│   ├── omp_mpi.c

│   └── omp_mpi_v2.c

└── fibonacci_CUDA

    ├── fibonacciCUDA.cu

    └── fibonacciCUDA._NoRecursivecu

• fibonacci directory, contains the basic version of the fibonacci recursive method (fibonacci.c), without any kind of parallelization.

• fibonacci_OpenMP directory, contains the three versions of fibonacci with OpenMp multi-threading parallelism (fibonacci_par1.c, fibonacci_par2.c and fibonacci_par3.c).

• fibonacci_MPI directory, contains the files used for the multi-processes approach with MPI interface, the entry point of the configuration (mpi-fib-start.c) and the fibonacci program executed in the different processes (fib.c).

• fibonacci_MPI_OMP directory, contains the two versions of the integration between MPI and OMP (omp_mpi.c and omp_mpi_v2.c).

• Fibonacci_CUDA directory, contains the recursive version (fibonacciCUDA.cu) and the non recursive version (fibonacciCUDA_NoRecursive.cu) of fibonacci, using the GPU computation and parallelism.

General Instruction:

For compiling the Basic version of fibonacci:

gcc -std=c99 fibonacci.c -o heat.out

For compiling the three OpenMP fibonacci's versions:

icc -O2 -qopenmp -xHost fibonacci_par3.c -o fibonacci_par1.out

icc -O2 -qopenmp -xHost fibonacci_par3.c -o fibonacci_par2.out

icc -O2 -qopenmp -xHost fibonacci_par3.c -o fibonacci_par3.out

For compiling the files of MPI fibonacci's version:

mpiicc -o mpi-fib-start mpi2-fib-start.c

mpiicc -o fib fib.c

For executing the MPI fibonacci’s versions:

mpirun -np 1 mpi-fib-start 10

For compiling the two MPI + OMP fibonacci's versions:

mpiicc -qopenmp -xHost omp_mpi_v2.c -o omp_mpi.out

mpiicc -qopenmp -xHost omp_mpi_v2.c -o omp_mpi_v2.out

For executing the two MPI + OMP fibonacci's versions:

mpiexec -hostfile machinefile.txt -perhost 1 -np 8 ./omp_mpi.out 45

mpiexec -hostfile machinefile.txt -perhost 1 -np 8 ./omp_mpi_v2.out 45

For compiling the two CUDA fibonacci’s versions:

nvcc -o fibonacciCUDA.out fibonacciCUDA.cu 

nvcc -o fibonacciCUDA_NoRecursive.out fibonacciCUDA_NoRecursive.cu