Crate virtio_accel_coreml 

Core ML host backend for Apple Neural Engine capable Macs.

The production path accepts device-neutral TOSA 1.0 FlatBuffers, validates and analyzes them with virtio-accel-tosa, and lowers supported static floating-point graphs inside this host-native crate. Core ML models are configured with CPUAndNeuralEngine: supported operations may execute on the ANE, while Core ML remains free to place unsupported operations on the CPU. Program buffers are page-aligned allocations wrapped directly by MLMultiArray; output execution is accepted only when Core ML uses the same allocation as its output backing.

Structs

CoreMlAccelerator

Non-macOS placeholder that keeps workspace consumers portable.

CoreMlArtifact

Builder for the provider-owned Core ML path artifact.

Enums

ArtifactBuildError

Invalid Core ML path artifact construction.

FeatureRole

A Core ML model feature’s role in one virtio-accel binding.

InitError

Failure to initialize a Core ML backend instance.

LoweringError

Constants

ARTIFACT_FORMAT

Provider artifact format for CoreMlArtifact.

COREML_TOSA_TARGET

TOSA target currently lowered by the Core ML backend.

REQUIRED_RESIDENT_BYTES

The Core ML runtime does not publish a finite upper bound for model residency.

TARGET_IDENTITY

Core ML path-artifact ABI v1 targeting CPU plus Apple Neural Engine execution.

Functions

supports_tosa_dtype

Whether this lowering can expose dtype at a Core ML model boundary.

supports_tosa_operator

Whether the initial Core ML lowering tier can lower op for supported types and attributes.