Module builder 

IR program builders - construct the megakernel Program from vyre IR.

Two flavours:

• Interpreted (build_program_sharded) - If-tree opcode dispatch.
• JIT (build_program_jit) - payload processor fused directly.

Functions

build_program

Build the default megakernel IR (256 lanes × 1 workgroup, no custom opcodes).

build_program_jit

Build the JIT Megakernel IR where payload processor logic is fused into the body stream.

build_program_jit_slots

Build the JIT megakernel IR for an explicit number of ring slots.

build_program_priority

Build a priority-aware megakernel IR.

build_program_priority_slots

Build a priority-aware megakernel IR for an explicit ring slot count.

build_program_sharded

Build the megakernel IR with a custom workgroup size and optional custom opcodes.

build_program_sharded_no_io

Build the megakernel IR without the IO polling sidecar.

build_program_sharded_once_slots

Build a finite one-pass sharded megakernel IR for host-submitted batches.

build_program_sharded_once_slots_control_report_shared

Build a finite one-pass megakernel that reports completion through the control buffer only.

build_program_sharded_once_slots_shared

Shared-Arc variant of build_program_sharded_once_slots for hot runtime dispatchers that must not clone the megakernel template every launch.

build_program_sharded_slots

Build the megakernel IR for an explicit number of ring slots.

build_program_sharded_slots_shared

Build the sharded megakernel IR as a shared immutable template.

build_program_sharded_with_io_polling

Build the megakernel IR with the experimental IO polling sidecar.

build_program_sharded_with_workspace_adapter

Build the sharded megakernel IR with a consumer-owned resident workspace.

build_program_with_self_loading_miss_handler

Build the megakernel IR with a self-loading load-miss handler.

persistent_body

The body that runs once per iteration per lane. Exposed for tests and downstream crates that splice additional opcodes.

persistent_body_jit

The JIT body that runs once per iteration per lane.

persistent_body_priority

Priority-aware loop body. Replaces the per-lane 1:1 slot mapping with the scheduler’s priority scan.

persistent_body_priority_slots

Priority-aware loop body for an explicit ring slot count.

try_build_program_with_self_loading_miss_handler

Fallible variant of build_program_with_self_loading_miss_handler.

try_persistent_body

Fallible persistent body builder with explicit staging-allocation reporting.