Struct MappedBuffer 

pub struct MappedBuffer<T: Copy> { /* private fields */ }

A host-allocated, device-mapped (zero-copy) memory buffer.

The host memory is page-locked and accessible from both CPU code and GPU kernels. GPU accesses traverse the PCIe bus, making this suitable for small or infrequently-accessed data where the overhead of explicit transfers is not justified.

The buffer is freed automatically on drop via cuMemFreeHost.

Implementations

impl<T: Copy> MappedBuffer<T>

pub fn alloc(n: usize) -> CudaResult<Self>

Allocates a zero-copy host-mapped buffer of n elements.

The allocation uses cuMemAllocHost_v2 (page-locked pinned memory) and retrieves the corresponding device pointer via cuMemHostGetDevicePointer_v2. A CUDA context must be current on the calling thread.

Errors

Returns a CUDA driver error if allocation or mapping fails.

pub fn len(&self) -> usize

Returns the number of T elements in this buffer.

pub fn is_empty(&self) -> bool

Returns true if the buffer contains zero elements.

pub fn byte_size(&self) -> usize

Returns the byte size of this buffer.

pub fn as_device_ptr(&self) -> CUdeviceptr

Returns the raw device pointer for use in kernel parameters.

pub fn as_host_ptr(&self) -> *const T

Returns a raw const pointer to the host-side data.

pub fn as_host_ptr_mut(&mut self) -> *mut T

Returns a raw mutable pointer to the host-side data.

pub fn as_host_slice(&self) -> &[T]

Returns a shared slice over the host-side data.

Safety

The caller must ensure no concurrent GPU writes are in flight.

pub fn as_host_slice_mut(&mut self) -> &mut [T]

Returns a mutable slice over the host-side data.

Safety

The caller must ensure no concurrent GPU reads or writes are in flight.

Trait Implementations

impl<T: Copy> Drop for MappedBuffer<T>

fn drop(&mut self)

Executes the destructor for this type.

impl<T: Copy + Send> Send for MappedBuffer<T>

impl<T: Copy + Sync> Sync for MappedBuffer<T>