Crate daedalus_gpu

Expand description

GPU facade: backend selection, capability traits, opaque handles, and feature-gated backends. Feature matrix:

gpu-noop (default): deterministic noop backend, always available; zero-GPU environments still compile.
gpu-mock: deterministic mock backend for tests/CI.
gpu-wgpu: real wgpu backend (types remain internal), placeholder for now. Concurrency: backends are Send + Sync; clone the handle (cheap Arc) to share across tasks. Planner/runtime expectation: call select_backend once, inspect skipped reasons (for “why not GPU?”), then use the returned handle to allocate buffers/images without depending on any concrete GPU type.

Structs§

BackendSkip: Explanation for a backend that was skipped during selection.
ErasedPayload: Type-erased payload wrapper so runtimes can carry GPU-capable data without monomorphizing.
GpuAdapterInfo: Adapter information exposed to planner/runtime.
GpuBlockInfo: Block/stride info for formats (useful if compressed formats are added later).
GpuBufferHandle: Opaque buffer handle (no backend types).
GpuBufferId: Opaque buffer identifier.
GpuCapabilities: Capability query result.
GpuContextHandle: Shared handle wrapping a selected backend and diagnostics for why other backends were skipped.
GpuFormatFeatures: Per-format feature flags for planner/runtime decisions.
GpuImageHandle: Opaque image/texture handle (no backend types).
GpuImageId: Opaque image identifier.
GpuImageRequest: Request shape for image/texture creation.
GpuOptions: Adapter selection options.
GpuRequest: Request shape for resource creation.
GpuUsage: Usage flags for buffers/images; combinations are allowed.
NoopBackend: Noop backend: deterministic, does nothing.
SimpleBufferPool: Naive buffer pool that reuses freed handles by size and usage.
TransferStats: Upload telemetry for tracking bytes transferred.

BackendSkipReason: Reason a backend was skipped during selection.
GpuBackendKind: Backend kind identifiers.
GpuError: GPU error codes for diagnostics.
GpuFormat: Common GPU formats (minimal set for planner decisions).
GpuMemoryLocation: Memory location hint for GPU resources.
Payload: Generic payload that can carry either CPU data or a GPU representation.

BufferPool: Simple buffer pool trait; implementations provide recycling.
GpuBackend: GPU backend trait; no backend-specific types exposed.
GpuContext: Optional context trait if backends need per-thread context.
GpuSendable: Opt-in bridge to allow CPU types to participate in GPU segments. Users implement this for their own types to describe how to upload/download.

active_backend: Convenience for callers that just need to know the active backend kind.
format_bytes_per_pixel
select_backend: Select the best available backend given build-time features and runtime options. Order: preferred backend (if set), then wgpu, mock, noop.
upload_bytes: Simple upload helper: allocates an upload buffer and returns it with bytes staged length.
upload_r8_texture: Convenience helper for uploading an R8 (single-channel) texture with basic size/length validation.
upload_rgba8_texture: Convenience helper for uploading an RGBA8 texture with basic size/length validation.
validate_texture_bytes: Validate texture copy layout against capabilities.