Expand description
GPU execution and Loom compilation without a build-time native dependency.
Native libraries are verified and loaded on first use. See bundle for
offline setup.
Re-exports§
pub use execution::Access;pub use execution::Completion;pub use execution::Runtime;
Modules§
- artifacts
- Model artifact discovery and parsing.
- benchmark
- Reusable benchmark distributions and synchronized stage timing.
- bundle
- Reproducible native provisioning. No build script downloads or native link flags.
- execution
- Coordinated GPU/NPU execution with inferred memory dependencies. Coordinated execution for GPU and NPU kernels.
- fabric
- Native AMD device ownership through libamdf’s versioned C ABI. Discovery is passive; only opening a device activates hardware.
- gpu
- Low-level GPU execution. Kernel dispatch and external memory access require the caller to establish their safety and synchronization contracts.
- image
- Device-resident image transformations with bounded reusable plans.
- inference
- Shared model context, owned tensor transfers, and bounded inference slots.
- jobs
- Bounded host orchestration for asynchronous FFI callers and streaming pipelines. Jobs may block on coordinated GPU completions without occupying the caller’s scheduler. Cancellation never interrupts native work: running work drains.
- loom
- In-process Loom compilation with reusable indexed modules and verified disk artifacts.
- model
- Resident Loom workloads built from reusable buffers, kernels and graphs.
- npu
- Native XDNA compilation and execution through the shared AMD fabric.
Programs use canonical Loom
.xdnaimages; allocation is device-scoped. - plan_
cache - Bounded, single-flight preparation with idle-only LRU eviction.
- residency
- Opt-in, byte-budgeted model residency with transactional loader reservations.
- runner
hrx run— launch one Loom-compiled kernel on the GPU and dump its buffers.- storage
- Bounded GPU-authored file I/O over registered system memory.
- tensor
- Checked tensor metadata and owned views in one coordinated runtime.
Structs§
- Access
Graph - A raw GPU graph whose edges are inferred from declared buffer access.
- Access
View - One graph binding and the complete access performed through it.
- Buffer
- Device backing with allocation-budget ownership.
- Buffer
Pool - A bounded free list whose allocations are safe to reuse on one ordered stream.
- Constants
- Explicit scalar widths, packed in declaration order for HRX binding dispatch. Unlike direct kernargs this format has no implicit alignment or pointer slots. Export metadata gives only the total byte count. Mixed scalar widths must come from the kernel’s argument contract.
- Device
- An owned GPU address and execution domain.
- Event
- Immutable native completion point.
- Export
Info - Native entry’s checked argument and workgroup layout.
- Graph
- Owned native commands being recorded under checked dependency edges.
- Graph
Exec - Reusable prepared GPU graph and its last immutable completion point.
- Kernel
- Trusted native GPU entry and its declaration-order argument layout.
- Node
- A dependency node branded with its recording’s identity.
- Pooled
Buffer - Device storage returned to its
BufferPoolwhen the lease drops. - Readback
- Owned destination retained through native readback completion.
- Stream
- An ordered GPU stream with explicit prepared commands and bounded pools.
- Stream
Options - Native engines and instrumentation fixed before a stream allocates or submits.
- Submission
- Borrowed completion observer that also reclaims completed staging.
- Target
- An exact AMDGPU or XDNA deployment key.
- View
- Checked borrowed logical range in an owned buffer.
Enums§
- Error
- A runtime, validation, provisioning or compiler failure.
Constants§
- TARGET_
KEY - Default compiler target; devices report their architecture at runtime.
Traits§
- Scalar
- A supported fixed-width scalar for little-endian binding constants.
Type Aliases§
- Result
- The result of an HRX operation.