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

Mojo function

once

def once[Emit: def() -> None & RegisterPassable & ImplicitlyCopyable, params: ContextParams, //](emit: Emit, ctx: Context[params])

Runs emit exactly once per (logical) output row.

• Warp tier: only lane 0 runs emit.
• Block tier: only thread 0 runs emit.
• Tiled / serial tier: each thread owns its own output(s), so every thread that runs the body also runs emit.
• Split-K tier: only thread 0 of the last-arriving block (per ctx._is_last_block, set by rowwise.reduce) runs emit.

emit is a value closure (its copy-captured state rides the value); this is an in-kernel thread predicate, not a launch boundary.

The body never observes lane_id() / thread_idx.x.

Parameters:

• ​Emit (def() -> None & RegisterPassable & ImplicitlyCopyable): The value-closure type of emit.
• ​params (ContextParams): The comptime dispatch parameters.

Args:

• ​emit (Emit): Callback to run on the canonical writer.
• ​ctx (Context[params]): The dispatch bundle passed to the body. The split-K branch reads ctx._is_last_block to gate emission.