Module sequential 

Sequential CPU execution model for workgroup parity.

Replaces the “invocation scheduler” abstraction with an obvious sequential semantic: run invocation 0, then 1, then 2… inside a workgroup. At each barrier, the outer driver re-runs every invocation from the barrier checkpoint so shared-memory effects from earlier invocations are visible to later ones. This matches backend spec-compliant semantics under the parity contract.

Exposes a single entry point run_sequential_workgroup which the conform runner uses as its CPU oracle; crate::workgroup owns per-invocation state and memory types shared by the execution tree.

Structs

SequentialWorkgroup

Driver for sequential per-invocation execution inside a workgroup.

Constants

MAX_SHARED_BYTES

Maximum shared-memory allocation exported for test convenience so the sequential driver and workgroup memory model agree on bounds.