Struct AotCache 

pub struct AotCache { /* private fields */ }

On-disk AOT cache for optimized LIR modules.

Implementations

impl AotCache

pub fn new(root: impl Into<PathBuf>) -> Self

pub fn put_lir(&self, key: &str, lir: &LirModule) -> Result<LirFingerprint>

Persist an optimized LIR module. Returns its compile fingerprint.

pub fn get_lir(&self, key: &str) -> Result<LirModule>

Load a previously stored LIR module.

pub fn contains(&self, key: &str) -> bool

pub fn compile_graph_cached( &self, key: &str, device: Device, graph: Graph, options: &CompileOptions, ) -> Result<CompiledGraph, AotCacheError>

MIR graph → fusion pipeline → cached LIR → backend executable.

On a cache hit only the backend compile runs (fusion / vmap / autodiff already done). Keys must be unique per graph + options fingerprint the caller encodes in key.

pub fn compile_hir_cached( &self, key: &str, device: Device, hir: HirModule, options: &CompileOptions, ) -> Result<CompiledGraph, AotCacheError>

Compile HIR through the fusion pipeline, cache LIR, return executable.

pub fn specialize_cached( &self, base_key: &str, binding: &DimBinding, device: Device, template: &CompileResult, options: &CompileOptions, ) -> Result<CompiledGraph, AotCacheError>

Specialize a cached dynamic LIR template and persist the bound variant.