For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo function

build_kernel_program

build_kernel_program(body: List[OpDesc], config: PipelineConfig, sched: ScheduleConfig = ScheduleConfig(scheduling=SchedulingStrategy.IDENTITY, sched_barrier_mask=85, auto_waits=True, drain_lgkm_mask=0, auto_drain=False, lds_contention_penalty=0, wait_lgkm_first=8, wait_vm_last=6, lgkm_per_load_a=0, lgkm_per_load_b=0, lgkm_after_last=False, minimal_barriers=False, omit_mma_set_prio=False, max_vgpr=999999, global_before_frag=False, barrier_before_pre_ops=False, inter_block_lgkm_drain=False, partial_prologue_drain=False, wrap_waits_with_sched_barrier=False), edges: List[DepEdge] = List()) -> PipelineProgram

Build the finalized PipelineProgram for a double-buffer kernel.

Runs the full pipeline: LDG construction → CSP/greedy scheduling → program building → redistribution → auto-waits → auto-drain → verification. Returns the PipelineProgram before expansion to entries.

Used by ScheduleCompiler.compile() for double-buffer kernel derivation and by derive_prologue_from_program() for prologue extraction.

Args:

body (List[OpDesc]): Loop body op list (typically PipelineSchedule.build_body()).
config (PipelineConfig): Pipeline configuration (depth, MMA grid, etc.).
sched (ScheduleConfig): Scheduling configuration (CSP knobs, wait overrides).
edges (List[DepEdge]): Optional dependency edges to drive scheduling. If empty (default), falls back to derive_edges_from_ops(body, config) (preserves existing behavior for callers that don't override PipelineSchedule.derive_edges). Schedules that need extra edges (e.g., cross-half FRAG→MMA for cross-stage rotation) should override derive_edges and pass them through here.

Returns:

PipelineProgram: Finalized PipelineProgram before expansion to entries.