Calculating PI Value In Different Parallel Framework.
We use Riemann Sum with Midpoint Rule to calculate the integration, which includes loops that may be optimized using parallel computing.
Use pthread as the parallel framework.
Use OpenMP as the parallel framework to calculate Pi.
Note that in macOS the default clang does not support OpenMP, thus it needs to be built with gcc or llvm.
gcc could be directly installed by
brew install gccand llvm could be installed via
brew install llvmUse MPI as the parallel framework.
Use CUDA to optimize the parallel computing process, which must be run under CUDA environment. i.e., you must have
at least a nVidia card and nvcc installed to compile and run the code.
Use mixed method of MPI and OpenMP to gain scalability and high parallel performance at the same time.
All experiments are carried out under Linux with nvcc and nVidia cards installed.
And I chose 2^30 as the STEP_NUM for all framework.
Parallel parameters are listed below:
| Framework | Parameters |
|---|---|
| OpenMP | 16 Threads |
| MPI | 16 Processes |
| PThread | 16 Threads |
| CUDA | 512 Threads / 64 Blocks |
| MPIOMP | 4 Processes / 4 Threads |
MIT.