Direct3D 12 Memory Allocator
Loading...
Searching...
No Matches
Defragmentation

Interleaved allocations and deallocations of many objects of varying size can cause fragmentation over time, which can lead to a situation where the library is unable to find a continuous range of free memory for a new allocation despite there is enough free space, just scattered across many small free ranges between existing allocations.

To mitigate this problem, you can use defragmentation feature. It doesn't happen automatically though and needs your cooperation, because D3D12MA is a low level library that only allocates memory. It cannot recreate buffers and textures in a new place as it doesn't remember the contents of D3D12_RESOURCE_DESC structure. It cannot copy their contents as it doesn't record any commands to a command list.

Example:

allocator->BeginDefragmentation(&defragDesc, &defragCtx);
for(;;)
{
HRESULT hr = defragCtx->BeginPass(&pass);
if(hr == S_OK)
break;
else if(hr != S_FALSE)
// Handle error...
for(UINT i = 0; i < pass.MoveCount; ++i)
{
// Inspect pass.pMoves[i].pSrcAllocation, identify what buffer/texture it represents.
MyEngineResourceData* resData = (MyEngineResourceData*)pMoves[i].pSrcAllocation->GetPrivateData();
// Recreate this buffer/texture as placed at pass.pMoves[i].pDstTmpAllocation.
D3D12_RESOURCE_DESC resDesc = ...
ID3D12Resource* newRes;
hr = device->CreatePlacedResource(
pass.pMoves[i].pDstTmpAllocation->GetOffset(), &resDesc,
D3D12_RESOURCE_STATE_COPY_DEST, NULL, IID_PPV_ARGS(&newRes));
// Check hr...
// Store new resource in the pDstTmpAllocation.
// Copy its content to the new place.
cmdList->CopyResource(
}
// Make sure the copy commands finished executing.
cmdQueue->ExecuteCommandLists(...);
// ...
WaitForSingleObject(fenceEvent, INFINITE);
// Update appropriate descriptors to point to the new places...
hr = defragCtx->EndPass(&pass);
if(hr == S_OK)
break;
else if(hr != S_FALSE)
// Handle error...
}
defragCtx->Release();
void SetResource(ID3D12Resource *pResource)
Releases the resource currently pointed by the allocation (if any), sets it to new one,...
UINT64 GetOffset() const
Returns offset in bytes from the start of memory heap.
ID3D12Resource * GetResource() const
Returns D3D12 resource associated with this object.
Definition D3D12MemAlloc.h:512
ID3D12Heap * GetHeap() const
Returns memory heap that the resource is created in.
Represents defragmentation process in progress.
Definition D3D12MemAlloc.h:768
HRESULT BeginPass(DEFRAGMENTATION_PASS_MOVE_INFO *pPassInfo)
Starts single defragmentation pass.
HRESULT EndPass(DEFRAGMENTATION_PASS_MOVE_INFO *pPassInfo)
Ends single defragmentation pass.
@ DEFRAGMENTATION_FLAG_ALGORITHM_FAST
Definition D3D12MemAlloc.h:648
Parameters for defragmentation.
Definition D3D12MemAlloc.h:670
DEFRAGMENTATION_FLAGS Flags
Flags.
Definition D3D12MemAlloc.h:672
Allocation * pSrcAllocation
Allocation that should be moved.
Definition D3D12MemAlloc.h:706
Allocation * pDstTmpAllocation
Temporary allocation pointing to destination memory that will replace pSrcAllocation.
Definition D3D12MemAlloc.h:714
Parameters for incremental defragmentation steps.
Definition D3D12MemAlloc.h:722
DEFRAGMENTATION_MOVE * pMoves
Array of moves to be performed by the user in the current defragmentation pass.
Definition D3D12MemAlloc.h:746
UINT32 MoveCount
Number of elements in the pMoves array.
Definition D3D12MemAlloc.h:724

Although functions like D3D12MA::Allocator::CreateResource() create an allocation and a buffer/texture at once, these are just a shortcut for allocating memory and creating a placed resource. Defragmentation works on memory allocations only. You must handle the rest manually. Defragmentation is an iterative process that should repreat "passes" as long as related functions return S_FALSE not S_OK. In each pass:

  1. D3D12MA::DefragmentationContext::BeginPass() function call:
    • Calculates and returns the list of allocations to be moved in this pass. Note this can be a time-consuming process.
    • Reserves destination memory for them by creating temporary destination allocations that you can query for their ID3D12Heap + offset using methods like D3D12MA::Allocation::GetHeap().
  2. Inside the pass, you should:
    • Inspect the returned list of allocations to be moved.
    • Create new buffers/textures as placed at the returned destination temporary allocations.
    • Copy data from source to destination resources if necessary.
    • Store the pointer to the new resource in the temporary destination allocation.
  3. D3D12MA::DefragmentationContext::EndPass() function call:
    • Frees the source memory reserved for the allocations that are moved.
    • Modifies source D3D12MA::Allocation objects that are moved to point to the destination reserved memory and destination resource, while source resource is released.
    • Frees ID3D12Heap blocks that became empty.

Defragmentation algorithm tries to move all suitable allocations. You can, however, refuse to move some of them inside a defragmentation pass, by setting pass.pMoves[i].Operation to D3D12MA::DEFRAGMENTATION_MOVE_OPERATION_IGNORE. This is not recommended and may result in suboptimal packing of the allocations after defragmentation. If you cannot ensure any allocation can be moved, it is better to keep movable allocations separate in a custom pool.

Inside a pass, for each allocation that should be moved:

  • You should copy its data from the source to the destination place by calling e.g. CopyResource().
  • If a resource doesn't contain any meaningful data, e.g. it is a transient render-target texture to be cleared, filled, and used temporarily in each rendering frame, you can just recreate this texture without copying its data.
  • If the resource is in D3D12_HEAP_TYPE_READBACK memory, you can copy its data on the CPU using memcpy().
  • If you cannot move the allocation, you can set pass.pMoves[i].Operation to D3D12MA::DEFRAGMENTATION_MOVE_OPERATION_IGNORE. This will cancel the move.
  • If you decide the allocation is unimportant and can be destroyed instead of moved (e.g. it wasn't used for long time), you can set pass.pMoves[i].Operation to D3D12MA::DEFRAGMENTATION_MOVE_OPERATION_DESTROY.

You can defragment a specific custom pool by calling D3D12MA::Pool::BeginDefragmentation or all the default pools by calling D3D12MA::Allocator::BeginDefragmentation (like in the example above).

Defragmentation is always performed in each pool separately. Allocations are never moved between different heap types. The size of the destination memory reserved for a moved allocation is the same as the original one. Alignment of an allocation as it was determined using GetResourceAllocationInfo() is also respected after defragmentation. Buffers/textures should be recreated with the same D3D12_RESOURCE_DESC parameters as the original ones.

You can perform the defragmentation incrementally to limit the number of allocations and bytes to be moved in each pass, e.g. to call it in sync with render frames and not to experience too big hitches. See members: D3D12MA::DEFRAGMENTATION_DESC::MaxBytesPerPass, D3D12MA::DEFRAGMENTATION_DESC::MaxAllocationsPerPass.

Thread safety: It is safe to perform the defragmentation asynchronously to render frames and other Direct3D 12 and D3D12MA usage, possibly from multiple threads, with the exception that allocations returned in D3D12MA::DEFRAGMENTATION_PASS_MOVE_INFO::pMoves shouldn't be released until the defragmentation pass is ended. During the call to D3D12MA::DefragmentationContext::BeginPass(), any operations on the memory pool affected by the defragmentation are blocked by a mutex.

What it means in practice is that you shouldn't free any allocations from the defragmented pool since the moment a call to BeginPass begins. Otherwise, a thread performing the allocation->Release() would block for the time BeginPass executes and then free the allocation when it finishes, while the allocation could have ended up on the list of allocations to move. A solution to freeing allocations during defragmentation is to find such allocation on the list pass.pMoves[i] and set its operation to D3D12MA::DEFRAGMENTATION_MOVE_OPERATION_DESTROY instead of calling allocation->Release(), or simply deferring the release to the time after defragmentation finished.

Mapping is out of scope of this library and so it is not preserved after an allocation is moved during defragmentation. You need to map the new resource yourself if needed.

Note
Defragmentation is not supported in custom pools created with D3D12MA::POOL_FLAG_ALGORITHM_LINEAR.