Crate tenferro_runtime 

Traced graph runtime and extension dispatch infrastructure for tenferro.

This crate owns graph construction, lowering to execution IR, graph execution, and backend-parametric extension runtime dispatch. Standard operations are lowered through the runtime’s internal operation vocabulary; tensor storage and backend kernels live in tenferro-tensor.

Use this crate directly when you want concrete tensor helpers or reusable traced graph execution without opting into autodiff. Start with TypedTensor when the scalar type is fixed in Rust, Tensor when dtype is selected at runtime, and TracedTensor plus GraphCompiler and GraphExecutor when the same expression should be compiled once and run repeatedly. Operation-family crates such as tenferro-einsum, tenferro-linalg, and tenferro-fft register extension runtimes with GraphExecutor when compiled execution reaches those operations.

User-facing guides live at https://tensor4all.org/tenferro-rs/guides/choosing-an-api.html and https://tensor4all.org/tenferro-rs/guides/execution-models.html.

Examples

use tenferro_runtime::{GraphCompiler, GraphExecutor, TracedTensor};
use tenferro_cpu::CpuBackend;

let x = TracedTensor::from_vec_col_major(vec![2], vec![1.0_f64, 2.0]).unwrap();
let y = (&x + &x).unwrap();
let mut compiler = GraphCompiler::new();
let program = compiler.compile(&y).unwrap();
let out = GraphExecutor::new(CpuBackend::default()).run(&program).unwrap();
assert_eq!(out.as_slice::<f64>().unwrap(), &[2.0, 4.0]);

Re-exports

pub use error::ContextId;

pub use error::Error;

pub use error::Result;

pub use extension_cache::ExtensionCacheKey;

pub use extension_cache::ExtensionCacheLimits;

pub use extension_cache::ExtensionCacheSelector;

pub use extension_cache::ExtensionCacheStore;

pub use extension_runtime::ExtensionExecutionContext;

pub use extension_runtime::ExtensionExecutor;

pub use extension_runtime::ExtensionRegistry;

pub use extension_runtime::ExtensionRuntime;

pub use extension_runtime::ExtensionRuntimeRegistryError;

pub use graph::GraphCompiler;

pub use graph::GraphCompilerCacheStats;

pub use graph::GraphExecutor;

pub use graph::GraphExecutorCacheStats;

pub use graph::GraphInstructionView;

pub use graph::GraphOpView;

pub use graph::GraphProgram;

pub use graph::GraphProgramInput;

pub use graph::GraphProgramLoweringShapeError;

pub use graph::GraphProgramLoweringView;

pub use traced::TracedTensor;

Modules

error

Error types for the tenferro runtime crate.

extension

Public surface for out-of-tree extension primitives.

extension_cache

Generic runtime caches for extension executors.

extension_runtime

Backend-parametric runtime dispatch for extension ops.

graph

Graph compilation APIs.

sym_dim

traced

Structs

CacheStats

Entry and retained-byte accounting for one cache.

CompilerOptions

DotGeneralConfig

DotGeneral dimension configuration.

GatherConfig

StableHLO gather dimension configuration.

OptimizerConfig

PadConfig

StableHLO pad configuration.

ScatterConfig

StableHLO scatter dimension configuration.

SliceConfig

Slice configuration.

SymDim

A symbolic tensor dimension expression used to build shape-agnostic graphs.

TypedTensor

Runtime typed tensor storage with compile-time scalar type and rank metadata.

TypedTensorView

Read-only borrowed view of typed tensor storage with arbitrary strides.

Enums

CompareDir

Comparison direction.

DType

Runtime scalar dtype tag.

Tensor

Dynamic tensor enum over the supported scalar types.

TensorRead

Read-only tensor input accepted by synchronous eager kernels.

TensorValue

Owned tensor value that can be compact or a lazy view.

TensorView

Dynamic read-only borrowed tensor view.

Traits

TensorBackend

Standard runtime backend over dynamic Tensor values.

TensorOpsExt

Backend-explicit concrete tensor operations.

TensorScalar

Sealed trait for scalar types that can be stored in a Tensor.

TypedTensorMaskOpsExt

Backend-explicit bool-mask operations for typed tensors.

TypedTensorOpsExt

Backend-explicit operations for dynamic-rank typed tensors.