Par4All Training

HPC Project offers theoretical and practical training sessions to help customers mastering Par4All. Training sessions can be organized in Paris or Montpellier (normal package) or onsite. Please contact us regarding specific demands in training.

Par4All training Program (5 days)

Parallel architectures

• SMP, clusters, MPP, GPU, hardware accelerators, embedded systems

Parallel programming models

• Parallelism & concurrency
• Parallel languages
• Parallel libraries
• Automatic parallelization

Par4All: automatic parallelization

• C parallelization
• Fortran parallelization
• Wild Cruncher: Scilab/Matlab parallelization
• Par4All options

Par4All programming rules

• Clean programming
• Express your programming style to maximize parallelization
• Profiling

Par4All stubs

• Using already optimized libraries
• Parallelizing a library instead of native implementation

Par4All Accel run-time

• Compilation system
• OpenMP GPU emulation
• CUDA for nVidia GPU and PGI x86
• OpenCL for GPU and manycores
• SCMP from CEA
• Adding your own target

Tweaking Par4All

• Improve generated code in compilation transit
• Understanding Par4All infrastructure:
  • Preprocessing, postprocessing, PyPS, PIPS, back-end compilation, run-time
    • Internal algorithms
    • Debugging Par4All
    • git infrastructure
    • Changing Par4All behaviour