Struct MemoryPool 

pub struct MemoryPool { /* private fields */ }

A stream-ordered memory pool (CUDA 11.2+).

Implementations

impl MemoryPool

pub fn new(device_ordinal: i32) -> CudaResult<Self>

Creates a new memory pool on the given device.

This retains the device’s primary context for the lifetime of the pool (released when the pool is dropped), and binds every subsequent allocation and free issued through this pool to that context, so the pool is safe to share across threads without mixing pointers from different devices.

Errors

• CudaError::InvalidDevice if device_ordinal is negative.
• CudaError::NotInitialized if no CUDA driver is available.
• Other CudaError variants if cuDevicePrimaryCtxRetain fails (e.g. an out-of-range device ordinal).

pub fn raw_handle(&self) -> u64

Returns the raw pool handle.

Status

Returns 0 until the pool is properly initialised.

pub fn device_ordinal(&self) -> i32

Returns the device ordinal this pool targets.

pub fn stats(&self) -> PoolStats

Returns current pool statistics.

The statistics track allocation behaviour over the pool’s lifetime.

pub fn trim(&mut self, min_bytes: usize) -> CudaResult<()>

Trims the pool, releasing unused memory back to the OS.

Attempts to release memory such that the pool retains at most min_bytes of unused memory.

Errors

pub fn set_threshold(&mut self, bytes: usize) -> CudaResult<()>

Sets the threshold at which the pool will automatically release memory back to the OS.

When the pool’s unused memory exceeds bytes, subsequent frees will trigger automatic trimming.

Errors