svod_runtime/
lib.rs

1//! Runtime execution for svod kernels.
2//!
3//! Provides generic kernel execution interface with backend-specific implementations
4//! (LLVM JIT, native shared libraries, CUDA, etc.).
5//!
6//! # Execution Model
7//!
8//! `execution_plan` is the canonical runtime path and executes prepared
9//! operations in dependency order with hazard-aware host parallelism.
10//! The `executor` module remains available for explicit parallel scheduling
11//! scenarios.
12//!
13//! # Benchmarking
14//!
15//! The `benchmark` module provides timing utilities for measuring kernel
16//! execution performance, used by beam search auto-tuning.
17
18pub mod benchmark;
19pub mod clang;
20pub mod custom_function;
21pub mod device_registry;
22pub mod devices;
23pub(crate) mod dispatch;
24pub mod error;
25pub mod execution_plan;
26pub mod executor;
27pub mod jit_loader;
28pub mod kernel_cache;
29pub mod llvm;
30#[cfg(feature = "mlir")]
31pub mod mlir;
32pub mod profiler;
33
34#[cfg(test)]
35pub mod test;
36
37pub use benchmark::{BenchmarkConfig, BenchmarkResult, benchmark_kernel, warmup_thread_pool};
38pub use custom_function::run_custom_function;
39pub use device_registry::DEVICE_FACTORIES;
40pub use devices::cpu::{CpuBackend, create_cpu_device, create_cpu_device_with_backend};
41pub use devices::cpu_queue::CpuQueue;
42pub use error::*;
43pub use execution_plan::{
44    ExecutionPlan, ExecutionPlanBuilder, PreparedBufferView, PreparedCopy, PreparedCustomFunction, PreparedKernel,
45    PreparedOp,
46};
47pub use executor::{
48    DeviceContext, ExecutionGraph, ExecutionNode, KernelBufferAccess, SyncStrategy, UnifiedExecutor, global_executor,
49};
50pub use kernel_cache::*;
51pub use llvm::*;
52pub use profiler::KernelProfile;
svod_runtime/lib.rs

svod_runtime/
lib.rs
