loadModule: Load a GPU module and access its Routines
In duncantl/RCUDA: R Bindings for the CUDA Library for GPU Computing

Description Usage Arguments Value Author(s) References See Also Examples

These functions load a module and allow us to get a reference to any of the named functions/routines or global variables it contains.

loadModule(filename, ctx = cuGetContext(TRUE),
           isCode = is(filename, "raw") ||
                     is(filename, "AsIs") ||
                       grepl(" __cudaparm", filename),  
           ...)
getFunction(module, name) 
## S4 method for signature 'CUmodule'
x$name
## S4 method for signature 'CUmodule,character,missing'
x[[i,j, globalVar = FALSE, ...]]

`filename`	the name of the file containing the GPU module. This can be in PTX, cubin or fatbin format. It is important that the code be compiled correctly for the appropriate architecture/GPU device. If not, an error is raised.
`ctx`	the context into which to load the module. If none has been created, this creates default contet that is used for this and subsequent computations.
`isCode`	a logical value that indicates whether `filename` is actually the PTX code itself or the name of a file containing the code. This allows us to obtain the code in R as a string and load it directly. One can use this argument or alternatively wrap the PTX text within a call to `I()` to indicate that it is the PTX code and not the name of a file. We also use some heuristic to determine if the text is PTX code.
`...`	additional parameters for methods
`module,x`	the module object returned by `loadModule` in which we will find the named function
`name,i`	the name of the function to find in the module. It is imperative that the kernel we want be compiled with the declaration `extern "C"` so that its name is not mangled.
`j`	missing/not supplied
`globalVar`	a logical that controls whether we are looking for a variable or a routine/function in the module.

loadModule returns an reference to the C object and is of class CUmodule in R.

getFunction, if it succeeds, returns an object of class CUfunction and is a reference to the C data type representing this routine.

Duncan Temple Lang

http://docs.nvidia.com/cuda/cuda-driver-api/group__CUDA__MODULE.html#group__CUDA__MODULE_1g366093bd269dafd0af21f1c7d18115d3

.cuda

if(getNumDevices() > 0) {
  ctx = cuGetContext(TRUE)
  ptx = system.file("sampleKernels", "dnormOutput.ptx", package = "RCUDA")
  m = loadModule(ptx)
  kernel = m$dnorm_kernel
}