compile4gpu: Enable compilation of GPU-compatible user-defined operators...

In rkeops: Kernel Operations on GPU or CPU, with Autodiff, without Memory Overflows

Description

Set up rkeops compile options to compile user-defined operators that run on GPU. If CUDA is not available, user-defined operators will still be compiled without GPU support.

Usage

compile4gpu()

Details

Compiling GPU-compatible user-defined operators requires CUDA and nvcc (Nvidia compiler). If not available, user-defined operators will only be CPU-compatible.

Note: Default behavior is to compile GPU-compatible operators thus, if you do not modify rkeops options, it is optional to use the function compile4gpu.

When a GPU-compatible operator is compiled, you should call use_gpu() to ensure that computation will be run on GPU (difference between compilation and runtime options). GPU-compatible operators can run on CPU.

Value

None

Author(s)

Ghislain Durif

See Also

compile4cpu(), use_gpu(),

Examples

library(rkeops)
compile4gpu()

rkeops documentation built on Feb. 17, 2021, 5:08 p.m.