Struct CudaArray 

pub struct CudaArray { /* private fields */ }

RAII wrapper for a CUDA array (1-D or 2-D).

Created by CudaArray::create_1d or CudaArray::create_2d; freed by Drop.

A CudaArray can be bound to a CudaTextureObject or CudaSurfaceObject for hardware-accelerated sampling.

Implementations

impl CudaArray

pub fn create_1d( width: usize, format: ArrayFormat, num_channels: u32, ) -> CudaRtResult<Self>

Allocate a 1-D CUDA array with width elements of the given format and channel count (num_channels must be 1, 2, or 4).

Mirrors cudaMallocArray (1-D form).

Errors

Returns CudaRtError::NotSupported if the driver does not expose cuArrayCreate_v2, or propagates driver errors.

pub fn create_2d( width: usize, height: usize, format: ArrayFormat, num_channels: u32, ) -> CudaRtResult<Self>

Allocate a 2-D CUDA array with width × height elements.

Mirrors cudaMallocArray (2-D form).

Errors

Returns CudaRtError::NotSupported if cuArrayCreate_v2 is absent.

pub unsafe fn copy_from_host_raw( &self, src: *const c_void, byte_count: usize, ) -> CudaRtResult<()>

Copy a contiguous host buffer into the entire array (synchronous).

data must contain exactly width * height.max(1) * num_channels elements of the appropriate type.

Mirrors cudaMemcpyToArray (host-to-array).

Errors

Returns an error if the driver does not support cuMemcpyHtoA_v2 or if the copy fails.

Safety

src must be valid for reading byte_count bytes.

pub fn copy_from_host<T: Copy>(&self, src: &[T]) -> CudaRtResult<()>

Copy a typed host slice into the array (synchronous, type-safe helper).

Errors

Forwards errors from Self::copy_from_host_raw.

pub unsafe fn copy_to_host_raw( &self, dst: *mut c_void, byte_count: usize, ) -> CudaRtResult<()>

Copy the entire array into a host buffer (synchronous, raw pointer).

Mirrors cudaMemcpyFromArray (array-to-host).

Errors

Returns an error if cuMemcpyAtoH_v2 is absent or the copy fails.

Safety

dst must be valid for writing byte_count bytes.

pub fn copy_to_host<T: Copy>(&self, dst: &mut [T]) -> CudaRtResult<()>

Copy the entire array into a typed host slice (synchronous, type-safe).

Errors

Forwards errors from Self::copy_to_host_raw.

pub unsafe fn copy_from_host_async_raw( &self, src: *const c_void, byte_count: usize, stream: CudaStream, ) -> CudaRtResult<()>

Asynchronously copy a host buffer into the array on stream.

Mirrors cudaMemcpyToArrayAsync.

Errors

Returns an error if cuMemcpyHtoAAsync_v2 is absent.

Safety

The caller must ensure src remains valid until the stream operation completes (i.e., until the stream is synchronized).

pub fn raw(&self) -> CUarray

Returns the raw CUarray handle (for use in resource descriptors).

pub const fn width(&self) -> usize

Width of the array in elements.

pub const fn height(&self) -> usize

Height of the array in elements (0 for 1-D arrays).

pub const fn format(&self) -> ArrayFormat

Element format of this array.

pub const fn num_channels(&self) -> u32

Number of channels (1, 2, or 4).

Trait Implementations

impl Drop for CudaArray

fn drop(&mut self)

Executes the destructor for this type.