Struct ManagedMemoryHints 

pub struct ManagedMemoryHints { /* private fields */ }

Builder-style API for applying memory hints to a unified memory region.

Wraps a raw pointer + byte size and exposes methods that issue the appropriate mem_advise / mem_prefetch driver calls.

Construction

Use for_buffer for raw pointers or from_unified for UnifiedBuffer references.

Implementations

impl ManagedMemoryHints

pub fn for_buffer(ptr: u64, byte_size: usize) -> CudaResult<Self>

Creates a ManagedMemoryHints from a raw device pointer and byte size.

Errors

Returns CudaError::InvalidValue if byte_size is zero.

pub fn from_unified<T: Copy>(buf: &UnifiedBuffer<T>) -> CudaResult<Self>

Creates a ManagedMemoryHints from a UnifiedBuffer reference.

The pointer and byte size are extracted from the buffer.

Errors

Returns CudaError::InvalidValue if the buffer reports zero bytes (should not happen for a validly constructed buffer).

pub fn ptr(&self) -> u64

Returns the device pointer this hint set targets.

pub fn byte_size(&self) -> usize

Returns the byte size of the targeted region.

pub fn set_read_mostly(&self, device: &Device) -> CudaResult<()>

Marks the region as read-mostly on device, enabling read replicas.

pub fn unset_read_mostly(&self, device: &Device) -> CudaResult<()>

Removes the read-mostly hint for device.

pub fn set_preferred_location(&self, device: &Device) -> CudaResult<()>

Sets the preferred location to device for this region.

pub fn unset_preferred_location(&self, device: &Device) -> CudaResult<()>

Removes the preferred-location hint for device.

pub fn set_accessed_by(&self, device: &Device) -> CudaResult<()>

Indicates that device will access this memory region.

pub fn unset_accessed_by(&self, device: &Device) -> CudaResult<()>

Removes the accessed-by hint for device.

pub fn prefetch_to(&self, device: &Device, stream: &Stream) -> CudaResult<()>

Prefetches the entire region to device on stream.

pub fn prefetch_range( &self, offset_bytes: usize, count_bytes: usize, device: &Device, stream: &Stream, ) -> CudaResult<()>

Prefetches a sub-range of the region to device.

Parameters

• offset_bytes — byte offset from the start of the region.
• count_bytes — number of bytes to prefetch.

Errors

Returns CudaError::InvalidValue if the range [offset_bytes, offset_bytes + count_bytes) exceeds the buffer, or if count_bytes is zero.

pub fn apply_policy( &self, policy: &MigrationPolicy, device: &Device, ) -> CudaResult<()>

Applies a MigrationPolicy to this memory region.

For MigrationPolicy::Default this is a no-op. For other variants the corresponding advice hint(s) are issued.

Trait Implementations

impl Clone for ManagedMemoryHints

fn clone(&self) -> ManagedMemoryHints

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for ManagedMemoryHints

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.