/parallel-programming-for-hpc

Scientific codes in C/C++ with CUDA, OpenACC, FFTW, (cu)BLAS

Primary LanguageCMIT LicenseMIT

Parallel programming for HPC

Weekly exercises for the course in Parallel programming for HPC @ UniTS.

Topics

  • Distributed parallelism (MPI)
  • BLAS
  • (NVIDIA) GPU programming
    • CUDA
    • Theory & best practices
  • cuBLAS
  • OpenACC
  • FFTW

Some plots

The following time measurements were taken on standard nodes on Marconi-100.

Matrix multiplication (MPI, BLAS, cuBLAS)

2500x2500

5000x5000

Jacobi method --- heat diffusion (MPI, OpenACC)

FFT 3D (MPI, FFTW)