Struct BufferPool 

pub struct BufferPool;

Facade over the GPU buffer pool.

Phase 1 status: documented no-op

The plan’s hand-rolled slab/coalescing pool was deliberately replaced by cubecl’s built-in allocator, which already does pooled slab allocation and reuse (configured through burn::backend::wgpu::MemoryConfiguration). burn 0.21 does not publicly expose cubecl 0.10’s ComputeClient::memory_cleanup, and memory_persistent_allocations does not exist in cubecl 0.10’s public API at all, so this facade currently does nothing. It exists so that the runtime can call pool.cleanup() / pool.pin_persistent() today and Phase 2 can back those calls with real handles (persistent KV/weight arenas) without changing call sites.

Implementations

impl BufferPool

pub fn new() -> Self

Creates a new pool facade.

pub fn pin_persistent(&self)

Pin long-lived allocations (weights, KV arena) so the pool never releases them. No-op in Phase 1 — cubecl’s pooled allocator keeps freed blocks for reuse anyway.

pub fn cleanup(&self)

Release cached free blocks back to the driver. No-op in Phase 1 — cubecl 0.10’s memory_cleanup is not reachable through burn’s public API.

Trait Implementations

impl Clone for BufferPool

fn clone(&self) -> BufferPool

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for BufferPool

impl Debug for BufferPool

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

Formats the value using the given formatter.

impl Default for BufferPool

fn default() -> BufferPool

Returns the “default value” for a type.