Mibo
InstanceWorldCache Type
Per-frame cache of staged per-instance world rows (VertexInstanceWorldPalette:
world matrix + chunk-local palette row) for skinned + instanced draws, SHARED by the
shadow pass and the forward PBR pass on the palette-texture backends (DX11/Vulkan).
Both passes stage the same rows every frame — the chunk plan is already shared via
PaletteChunkCache, so chunk-local
offsets match and one staging pass can serve both passes' vertex-buffer uploads
(each pass still uploads into its OWN DynamicVertexBuffer — the two passes never
share a buffer, preserving the no-race DX12 upload-ordering design).
Memo validity is per frame, same as PaletteChunkCache:
the buffer copies caller transforms into per-frame rented arrays at record time, so each
command's array is stable for the duration of the render flush, and
ReleaseAll — called once per frame alongside the palette cache's — drops every
memo. Keyed by array reference + count. The DX12 grouped path does NOT use this cache:
its forward/depth group budgets differ (PaletteGroup.MaxMatrices vs MaxMatricesDepth),
so chunk-local offsets differ per pass and staging stays per pass there.
Constructors
| Constructor |
Description
|
|
|
Instance members
| Instance member |
Description
|
Full Usage:
this.Obtain
Parameters:
Matrix[]
count : int
chunks : (int * int * 'T)[]
chunkTotal : int
Returns: VertexInstanceWorldPalette[]
Type parameters: 'T |
Returns the staged rows covering count instances under the given chunk plan ((chunkStart, chunkCount, _) triples), staging them on first request this frame. Row j of a chunk carries PaletteOffset j — chunk-local, matching the palette texture that holds only that chunk.
|
Full Usage:
this.ReleaseAll
|
Drops every memo (per-frame lifetime); pool arrays are kept (grow-only). |