- Version: 1.8.2
- Status:
- License: Apache 2.0
- Author: Drew Schmidt (package creator)
Thrust is a high-level C++ library for developing parallel code to be run on either a GPU or a multicore processor. It achieves this by selecting a "backend" at compilation time. Supported backends are multicore processors (OpenMP or Intel Thread Building Blocks) and NVIDIA GPU's (CUDA). So you only have to write your code once, but you can target different hardware architectures.
This is merely a convenient packaging of Thrust for use with R packages (I had nothing to do with creating the Thrust C++ library). To use the package, simply set your LinkingTo field to include thrust (this package). This is conceptually similar to the BH package for Boost.
Some file/folder names were changed in order to satisfy R CMD check. Most users should not be affected by these changes, because the files are buried pretty deep in the source tree and would likely only be directly included by a Thrust developer. For a full list of changes, see inst/Thrust_CHANGES of the package source, or run thrust::thrust_changes().
The package is provided under the same license as Thrust, Apache 2.0, and its version number mirrors it as well.
The development version is maintained on GitHub, and can easily be installed by any of the packages that offer installations from GitHub:
### Pick your preference
devtools::install_github("wrathematics/thrust")
ghit::install_github("wrathematics/thrust")
remotes::install_github("wrathematics/thrust")Basically you just link with the package and write your kernels using the Thrust API. You can find the Thrust documentation here.
Some other useful packages to this end:
- For an example of how to accommodate multiple backends, see the fastrand package (particularly its configure.ac).
- If you plan to support GPU's, you will likely want to compute on 32-bit floats. See the float package.
- For developing your kernels, you may wish to use the Rcpp package, although I'd be surprised if you haven't heard of it already.