Structs§
- A2
- 2-byte alignment
- A4
- 4-byte alignment
- A16
- 16-byte alignment
- A32
- 32-byte alignment
- A64
- 64-byte alignment
- Aligned
- A newtype with alignment of at least
Abytes - Arena
Storage - Only stores the large arena value.
- Buffer
- Buffer
Mut - Buffer
Range - Model
Artifacts - Owned
Arena - Instance-owned model arena storage selected by the Oneliner
allocfeature. - Sequential
Executor - Executor that runs work items immediately in submission order.
- Shared
Arena - Tensor4D
- Four-dimensional tensor backed by an owned, aligned, fixed-size nested array.
- Work
Item - iree_
hal_ executable_ dispatch_ state_ v0_ t - iree_
hal_ executable_ environment_ v0_ t - iree_
hal_ executable_ library_ header_ t - iree_
hal_ executable_ workgroup_ state_ v0_ t - iree_
hal_ processor_ v0_ t
Enums§
Traits§
- Buffer
Source - Converts a generated storage item into the
BufferRefused by dispatch. - Executor
- Schedules work items and provides a completion barrier.
- Model
Inference - Typed tensor inference implemented directly by generated model instances.
- Model
Source - Metadata exposed by a model generated with
#[model].
Functions§
- Aligned
- Changes the alignment of
valueto be at leastAbytes - concurrent
- Runs a group of generated commands sequentially.
- dispatch⚠
- Dispatches an IREE workload and panics on failure.
- dispatch_
fn_ ⚠from_ library - Resolves an export ordinal from an IREE static library.
- fill⚠
- Fills a generated buffer range with one byte value.
- try_
dispatch ⚠ - Dispatches an IREE workload with the default executor.
- try_
dispatch_ ⚠with_ executor - Dispatches an IREE workload through a caller-provided executor.