Crate alpha2_native_gpu 

Platform GPU interop for native decoded video surfaces.

FFmpeg ownership stays in alpha2-ffmpeg-bridge; this crate owns raw GPU handles, barriers, synchronization, and bounded surface accounting.

Structs

ExternalSyncPoint

Timeline synchronization point required before sampling an image.

ImportTicketStats

Allocation counters of a context’s import tickets.

ImportedPlaneDescriptor

One opaque native plane before it is wrapped by the active GPU backend.

ImportedVulkanFrame

An imported bi-planar 4:2:0 image and its two wgpu plane views.

NativeGpuIdentity

Stable identity of a native GPU device for cache and import validation.

NativeImageState

Native image state observed while the decoder surface is locked.

SharedVulkanChoice

The shared device the adapter choice made, its report, and why there is none if so.

SharedVulkanContext

One Vulkan logical device adopted by both the renderer and FFmpeg.

SharedVulkanError

Failure while creating a logical Vulkan device shared by wgpu and FFmpeg.

SurfaceBudget

Bounded decoder-surface allocator shared by decode sessions.

SurfaceBudgetError

A reservation exceeded either the count or byte limit.

SurfaceIdentity

Identity used to prevent imported texture reuse across decoder generations.

SurfaceReservation

A reservation released exactly once when its final owner is dropped.

SurfaceUsage

Current native surface allocation.

VideoEncodeExtent

Coded extents and coding limits one video-encode profile has on a device.

VulkanCandidate

One Vulkan adapter as the shared device’s choice sees it, read without opening a device.

VulkanEncodeCopier

Reusable NV12 handoff from the renderer’s plane textures to encoder frames, one per export.

VulkanImportLease

Owns the imported texture, FFmpeg frame, and native transition commands.

VulkanLayerVersion

A Vulkan layer’s specification version.

Enums

NativePlaneLayout

Supported decoded plane organizations.

Traits

NativeSurfaceLease

Native backing retained until every sampling submission has completed.

Functions

adapter_memory_budget

The device memory budget of adapter in bytes, or None where the API reports none.

adapter_pinned

Whether ALPHA2_GPU_ADAPTER pins one adapter, which probes opening their own devices then keep to.

choose_shared_vulkan

Makes the shared Vulkan device on the adapter policy takes, ranked as every host ranks adapters (alpha2_gpu::rank_adapters): under auto, the adapter encoding the most codecs on the GPU, then one that decodes when decode_allowed, then high-performance order. Without wanted nothing is made and the report says where the product device renders.

probe_vulkan_submission

Checks that a decoded frame can be imported: moves its planes to SHADER_READ_ONLY and back in one submission on its queue, through import-like views, and waits.

set_gpu_wait_timeout

Sets how long the platform interop helpers wait on GPU work before failing with a timeout.

shared_device_index

The loader physical-device index of the adapter the host’s choice makes the shared device on: the device index FFmpeg’s av_hwdevice_ctx_create takes. Probes that open their own device pass it, so they use the device an export runs on.

vulkan_candidates

Every Vulkan adapter, in enumeration order, with what it offers the shared device.