Create a GPU-aware tensor

Create a GPU-aware tensor

Usage

cuda_tensor(x, device = c("auto", "cuda", "cpu"), dtype = NULL)

Arguments

x

Numeric vector, matrix, array, or another cudatensor.

device

One of "auto", "cuda", or "cpu". Auto selects CUDA only when cuda_available() is true.

dtype

One of "float64", "float32", or "integer".

Matrix and array dimnames, including names on a one-dimensional input, are retained as R metadata on both CPU and CUDA tensors. Floating dtypes accept IEEE Inf, -Inf, NaN, and R's floating NA; torch backends may normalize NA to NaN. Integer dtype rejects non-finite or fractional values because they have no exact integer representation. When x is already a tensor on the selected device, compatible floating dtype changes use its current backend without materializing the tensor on the host. Conversion to integer still validates exact representability on the host.

Value

A cudatensor object.

Examples

x <- cuda_tensor(matrix(1:6, nrow = 2), device = "cpu")
x
#> <cudatensor[2x3] device=cpu backend=base dtype=integer>
#>      [,1] [,2] [,3]
#> [1,]    1    3    5
#> [2,]    2    4    6