anvl: Accelerated Array Computing and Automatic Differentiation

Accelerated array computing and code transformations for R. Numerical programs operating on arrays can be just-in-time compiled to optimized executables via 'XLA' – the same compiler that powers 'JAX' and 'TensorFlow' – and run on CPU or NVIDIA GPU from the same source. Also provides reverse-mode automatic differentiation, returning the gradient of a function as another R function.

Options

• anvl.backend (character(1)): the backend every operation runs on. Must be one of "pjrt" and "quickr". If not set, "pjrt" is used. Also see active_backend(), local_backend() and with_backend().

• anvl.default_dtypes (named character() | named list()): the data types an R double and an R integer settle on when they meet no typed array. See default_dtypes() for more details.

• anvl.default_device (character(1) | device object): the device a call that names none allocates on. Also see default_device(), local_default_device() and with_default_device().

Environment variables

• ANVL_DEFAULT_DTYPES: category=dtype pairs such as "float=f64,int=i64", used for every backend when the anvl.default_dtypes option is not set.

• ANVL_DEFAULT_DEVICE: a device identifier such as "cuda" or "cuda:0", used when the anvl.default_device option is not set.

Both are read once, when anvl is loaded, and kept as fallbacks for the options rather than set as them: getOption("anvl.default_device") stays NULL, and an option set later takes precedence. Changing a variable after anvl is loaded has no effect until the package is loaded again, e.g. in a new R session.

The remaining ones, all prefixed ANVL_TEST, affect only anvl's own test suite, not the package:

• ANVL_TEST: tests/testthat.R runs the tests only when this is "1", so R CMD check in a shell without it runs none of them.

• ANVL_TEST_SKIP_QUICKR: when set to anything non-empty, the tests that need the quickr backend are skipped – they are comparatively slow.

Third-Party Licenses

The anvl package itself is MIT-licensed. The CUDA backend dynamically loads NVIDIA software which is not bundled with anvl, but downloaded from NVIDIA's official redistributable channels by the CUDA toolkit R package (e.g. pjrt.cuda) at install time. Its use is governed by the NVIDIA CUDA Toolkit EULA, with the exception of cuDNN, which is covered by the NVIDIA cuDNN SLA, and NCCL, which is covered by its own license. By installing or using the CUDA backend you accept those terms.

See also

Useful links:

• https://r-xla.github.io/anvl/

• https://github.com/r-xla/anvl

• Report bugs at https://github.com/r-xla/anvl/issues

Author

Maintainer: Sebastian Fischer seb.fischer@tutamail.com (ORCID)

Authors:

• Sebastian Fischer seb.fischer@tutamail.com (ORCID)

• Daniel Falbel daniel@posit.co (ORCID)

• Tomasz Kalinowski tomasz@posit.co

• Nikolai German niko.german@gmail.com (ORCID)

Other contributors:

• Louis Aslett louis.aslett@durham.ac.uk (ORCID) [contributor]

• Hugh Graham hugh@belian.earth (ORCID) [contributor]