Create a model for the GPU. This requires a working nvcc toolchain
and GPU device to work properly. Note that this will cache the
compilation within a session, so if you want to change GPU options
you will need to restart R. However, if you change the model
(e.g., by changing the result of options like rewrite_dims) the
model will recompile. Compilation speed is slow enough that the
cachine will be obvious.
compile_gpu(model = "lancelot", ..., real_type = "float", gpu = TRUE)The name of a sircovid model, either lancelot or
basic
Additional arguments passed to odin.dust::odin_dust_ and from there into either odin's options or the gpu options.
The type to use for floating point numbers. The default is "float" which differs from the CPU model but is typically what you want on the GPU.
The argument passed to odin.dust::odin_dust_ as
gpu. The default here is TRUE, but you may want to pass the
results of dust::dust_cuda_options in order to control
compilation.
A model generator, like lancelot that can run on the GPU available to this computer.