copyFromDevice: Copy data betwen R and GPU
In duncantl/RCUDA: R Bindings for the CUDA Library for GPU Computing

Description Usage Arguments Value Author(s) See Also Examples

These functions perform the (potentially) necessary copying of data between the host and the GPU device. The functions codeToDevice and copyFromDevice are the explicit functions. We can use R's subsetting to copy the elements with ref[] and ref[] = x.

copyFromDevice(obj, nels, type)
copyToDevice(obj, to = cudaMalloc(length(obj), elType = elType, strict = strict),
              elType = typeof(obj), strict = !missing(elType) )
cudaMalloc(numEls, sizeof = 4L, elType = NA, 
            strict = !missing(elType) || inherits(elType, "AsIs"))

`obj`	the R object to be copied to the GPU or the reference to the GPU memory that is to be copied back to R
`nels,numEls`	the number of elements to be copied or to be allocated
`type`	the type of the elements being copied
`elType`	the name of the R/GPU type to be allocated
`sizeof`	the size of the element in the array/vector on the GPU to be allocated
`to`	a reference to space on the GPU to store the data being copied from R to the GPU
`strict`	a logical value that controls whether we treat the value `elType` as verbatim and use it exactly or whether we can map it to another type. If this is `FALSE`, `"double"` is mapped to `"float"`.

cudaMalloc returns a generic pointer to memory on the GPU.

copyToDevice returns an object that knows both the type and the number of elements of the array and so can retrieve this from the GPU.

copyFromDevice returns the R vector corresponding to the data on the GPU.

Duncan Temple Lang

.cuda

if(getNumDevices() > 0) {

 cuGetContext()

 N = 1e5
 x = rnorm(N)

 cx = copyToDevice(x)
 length(cx[])
   # or explicitly with
 vals = copyFromDevice(cx)
 length(vals)

 i = sample(1:N, 100)
   # subsetting on the GPU
 cx[i]
}