Struct UnifiedAllocator 

pub struct UnifiedAllocator { /* private fields */ }

Unified allocator interface that can use different allocation strategies

Implementations

impl UnifiedAllocator

pub fn new( base_ptr: NonNull<u8>, total_size: usize, config: UnifiedConfig, ) -> Result<Self, AllocationError>

Create a new unified allocator

pub fn allocate( &mut self, size: usize, requested_allocator_type: AllocatorType, _alignment: Option<usize>, ) -> Result<NonNull<u8>, AllocationError>

Allocate memory using the unified interface

pub fn deallocate( &mut self, ptr: NonNull<u8>, size: usize, ) -> Result<(), AllocationError>

Deallocate memory

pub fn free( &mut self, ptr: *mut c_void, _allocator_type: AllocatorType, ) -> Result<(), AllocationError>

Free memory (alias for deallocate)

pub fn reallocate( &mut self, _ptr: *mut c_void, new_size: usize, allocator_type: AllocatorType, ) -> Result<*mut c_void, AllocationError>

Allocate a fresh block for a reallocation request.

Deliberately does not free _ptr (its parameter is unused by design, not by oversight): the sole caller, crate::memory::GpuMemorySystem::reallocate, already owns the complete, correct realloc contract one layer up – it tracks the old allocation’s real size (which this layer does not), copies the live data with it via copy_nonoverlapping, and only then frees the old pointer. Freeing _ptr here as well would race that caller: it still reads from _ptr after this call returns whenever the new address differs (the overwhelmingly common case, since this always allocates fresh rather than truly growing in place), which would make that read a use-after-free and the caller’s own free a double-free.

pub fn get_stats(&self) -> &UnifiedStats

Get unified statistics

pub fn get_config(&self) -> &UnifiedConfig

Get the active configuration (thresholds, enabled sub-allocators, default strategy).

allocate currently always uses the type its caller passes rather than deriving one from buddy_threshold/slab_threshold/ arena_threshold/enable_auto_routing – centralizing that decision here would change GpuMemorySystem::choose_allocator’s tested routing boundaries (its hardcoded 1 KiB / 1 MiB cutovers do not line up with this config’s threshold values), which is a deliberate behavior decision left to that caller rather than one this lint pass makes unilaterally. Exposed so callers can inspect (and eventually route against) the real configuration instead of duplicating their own hardcoded thresholds.

pub fn get_detailed_info(&self) -> DetailedAllocatorInfo

Get detailed allocator information

pub fn reset_allocator( &mut self, allocator_type: AllocatorType, ) -> Result<(), AllocationError>

Reset specific allocator

pub fn garbage_collect(&mut self) -> GarbageCollectionResult

Force garbage collection on applicable allocators

pub fn optimize_strategies(&mut self) -> Result<(), AllocationError>

Optimize allocation strategies based on performance data

Trait Implementations

impl Send for UnifiedAllocator

impl Sync for UnifiedAllocator