Implicit Parallelization

Uses threads and shared memory (ideally via OpenMP) TODO: similar idea to SupR?

• the user specifies a maximal number of threads to use
• R decides internally when to use more than one thread, if allowed

• available in an experimental package pnmath^{pnmath_archive}

• requires no explicit user action once enabled (just call the compiled function in R?)

• Rcpp and inline

  • Rcpp is a C++ library facilitates the integration of R and C++
  • inline package provides functionality to dynamically define R functions and S4 methods with in-lined C, C++ and Fortran.

• OpenMP

  • It is a shared memory model.
  • It is a Lightweight approach.
  • Workload is distributed between threads.
  • Supported by many compilers: GNU, Intel, IBM, NAG and PGI.

  R has interfaces for calling compiled C/C++, Fortran codes.

  • Call compiled function written with Rcpp, OpenMP. Parallelism with OpenMP using threads. But thread safety is a big concern (TODO: how? this post explains why but I couldn't understand.)

    OpenMP in R Examples:

    • Parallelization in rcpp via OpenMP
    • OPENMP TUTORIAL, WITH R INTERFACE

• MPI