Skip to main content

Crate sim_lib_compute_model

Crate sim_lib_compute_model 

Source
Expand description

Modeled resident tensor compute provider.

The modeled provider is a deterministic TensorExecutor for tests, recipes, and future hardware-provider conformance. It executes through the published sim-lib-numbers-tensor CPU semantics, then wraps successful results in resident storage owned by site/compute/model.

Structs§

ComputeModelLib
Loadable library that registers the modeled compute site.
ModeledComputeProfile
Configuration for a modeled compute site.
ModeledComputeSnapshot
Snapshot of modeled executor counters.
ModeledResidentSegment
One resident segment in the modeled storage arena.
ModeledResidentStorage
Resident storage that models readback into host tensor cells.
ModeledTensorExecutor
Deterministic tensor executor that models resident compute placement.
ResidentHandle
Opaque resident allocation handle owned by the modeled site.
ResidentOdeExecution
Result plus provider counters for a resident ODE execution.
ResidentOdeExecutor
Executes checked tensor ODE plans through an active tensor executor.
ResidentOdePlan
Checked resident ODE plan.

Enums§

ModeledComputeFault
Fault injected into the modeled executor or resident storage.
ResidentOdeKind
ODE execution family requested by the resident adapter.
ResidentRhsLowering
Provider-side lowering classification for a tensor RHS.

Statics§

RECIPES
Cookbook recipes for this lib, embedded at build time.

Functions§

compute_model_lib_symbol
Runtime library symbol for the modeled compute provider.
compute_model_site_symbol
Site symbol exported by the modeled compute provider.
modeled_executor_symbol
Stable symbol for the modeled tensor executor.