Module planner 

Resident work-queue sizing, launch geometry, and work-item contracts.

Structs

BarrierElisionReport

Report returned by elide_value_flow_barriers.

ResidentGridLimits

Adapter limits that bound a megakernel worker-grid recommendation.

ResidentGridPlan

Resolved worker-grid plan shared by direct and policy-driven megakernel paths.

ResidentGridRequest

Logical work shape requested for a megakernel worker-grid recommendation.

ResidentLaunchGeometry

Host-side launch geometry for a finite megakernel dispatch.

ResidentQueueCapabilities

Capabilities surfaced by megakernel-aware backends.

ResidentQueueConfig

Configuration for one megakernel dispatch invocation.

ResidentQueueReport

Summary stats from one megakernel run.

ResidentQueueTelemetry

Production counters from one megakernel dispatch.

ResidentSizingPolicy

Shared worker-grid sizing policy for megakernel dispatch.

ResidentWorkItem

One work-queue item the megakernel worker consumes.

ResidentWorkloadHints

Optional scale signals that let the megakernel launch policy choose sparse, dense, hybrid, fused, or memory-constrained execution from real workload shape instead of queue length alone.

Functions

build_bellman_tn_order_program

Bellman tensor-network ordering Program builder. Wraps vyre_self_substrate::bellman_tn_order::bellman_tn_order_program.

build_kfac_autotune_step_program

KFAC autotune-step Program builder. Wraps vyre_self_substrate::kfac_autotune_step::kfac_autotune_step_program.

build_persistent_fixpoint_program

Build a persistent-fixpoint Program around a caller-supplied transfer body. Replaces a host-side loop { dispatch(); check } pattern with a single GPU-side dispatch that ping-pongs current ↔ next until changed[0] == 0 or max_iterations.

build_scallop_provenance_wide_program

Build a multi-word scallop-provenance Program. Wraps vyre_self_substrate::scallop_provenance_wide::scallop_provenance_wide_program for >32-rule lineage tracking (W=8 → 256 rules max).

build_sinkhorn_clustering_program

Build a sinkhorn dispatch-clustering Program. Wraps vyre_self_substrate::sinkhorn_dispatch_clustering::sinkhorn_clustering_program.

build_sinkhorn_full_clustering_program

for callers that need the full iterative-balance variant rather than the dispatch-clustering simplification.

default_worker_groups_from_limits

Compute a persistent-worker ceiling from adapter limits.

dispatch_grid_for

Compute the backend dispatch grid for a logical queue length.

elide_value_flow_barriers

Remove barriers between independent megakernel arms.

padded_slot_count

Round a logical slot count up to a whole workgroup.

worker_workgroup_size

Clamp the caller’s worker setting into the legal x dimension used by the current megakernel ABI.