Crate onnx_runtime_ir 

onnx-runtime-ir

The Graph intermediate representation (IR) for the ORT 2.0 runtime.

This crate is the stable contract that every downstream runtime crate (onnx-runtime-loader, onnx-runtime-ep-api, onnx-runtime-session, …) builds against. It is intentionally pure, safe Rust with no FFI and no device dependencies so it compiles standalone on any target.

It is a Rust port of the design captured in docs/ORT2.md §3 (Graph IR), §5 (Striding & Layout) and §11 (Dynamic Shape), itself inspired by the Python onnx-ir package.

What lives here

Concept	Type
Element type	DataType
Symbolic / static shapes	Shape, Dim, SymbolConstraints
Physical strided layout	TensorLayout, MemoryFormat
Device placement	DeviceType, DeviceId
Graph values (SSA edges)	Value, ValueId
Graph operations	Node, NodeId, Attribute
Constant / weight storage	TensorData, SparseTensorData, WeightRef
The graph itself	Graph
Errors	IrError, GraphError

Design guarantees

• SSA-like: every Value has at most one producer Node; node outputs are unique.
• First-class layout & device: every value carries a TensorLayout and an optional DeviceId, unlike upstream ONNX / onnx-ir.
• Mutable during optimization: the Graph mutation API keeps producer/consumer edges consistent so optimization passes can rewrite it, then it is shared immutably via Arc once frozen.

Deep algorithms whose full implementation belongs to a later task (e.g. per-op shape inference) are represented here only by their data model; the Graph operations that are cheap and foundational (topological ordering, validation, edge rewiring, broadcasting, stride arithmetic) are fully implemented and unit-tested so downstream crates compile against a stable, working surface.

Structs

Arena

A slot map: dense Vec storage with O(1) insert/remove/lookup by key.

Consumers

The unordered internal set of input slots that consume a value.

DeviceId

A specific device instance: a DeviceType plus an ordinal index.

Graph

A computation graph in SSA form.

Node

An operation in the graph.

NodeId

Unique identifier for a Node within a Graph.

SparseTensorData

A sparse constant tensor in COO form.

SymbolConstraints

Constraints on a symbolic dimension, used by shape bucketing, kernel specialization, and tiling (see docs/ORT2.md §11).

SymbolId

Unique identifier for a symbolic dimension.

TensorData

A concrete constant tensor held inline (e.g. an attribute value or a small initializer). Element bytes are stored little-endian and densely packed.

TensorLayout

First-class strided layout for a value.

Value

A value flowing through the graph — the output of one node and the input of zero or more others (SSA form: at most one producer).

ValueId

Unique identifier for a Value within a Graph.

Enums

Attribute

An ONNX operator attribute. Covers all attribute value kinds.

DataType

Supported tensor element types.

DeviceType

A class of compute device / execution backend.

Dim

A single dimension: either a concrete size or a symbol.

GraphError

A single structural defect in a Graph.

IrError

Crate-level error for fallible IR operations.

MemoryFormat

Memory-format hint used to pick vectorized kernels.

TypeProto

An ONNX TypeProto: the type of a value, which may be a tensor or a container of tensors (see docs/ORT2.md §3.2).

WeightRef

A reference to initializer (weight) data.

Constants

AI_ONNX_DOMAIN

The explicit spelling of the default ONNX operator domain.

Traits

ArenaKey

A type usable as an arena key: a newtype around a u32 index.

Functions

as_static_shape

Returns Some(dims) if every dimension of shape is static.

broadcast_shapes

Compute the output shape of a numpy-style broadcast of a and b.

compute_contiguous_strides

Compute row-major (C-order) contiguous strides, in elements, for a shape.

is_contiguous

Whether strides describe a row-major contiguous layout for shape.

is_default_domain

Whether a raw operator domain string denotes the default ONNX domain.

is_fully_static

Whether every dimension of shape is static.

normalize_domain

Canonicalize an operator domain to its post-load form.

static_shape

Convenience constructor for a fully-static shape.

Type Aliases

Result

A convenience Result alias for IR operations.

Shape

A tensor shape: an ordered list of Dims.

Usage

A node input slot that uses a value.