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 struct

BarrierStrategy

struct BarrierStrategy

How the schedule emits barriers and barrier-adjacent fences.

Fields: minimal: Suppress per-block s_barriers and set_prio pairs; emit s_barrier only at top-of-half and the first cross-stage block. See ScheduleConfig.minimal_barriers. omit_set_prio: When minimal=True, drop the pre-MMA s_setprio[1] entirely. See ScheduleConfig.omit_mma_set_prio. sched_barrier_mask: Bitmask of which blocks get trailing schedule_barrier fences. Default: 0b01010101. wrap_waits_sched_barrier: Wrap each contiguous wait/barrier group with schedule_barrier on both sides. See ScheduleConfig.wrap_waits_with_sched_barrier. barrier_before_pre_ops: Move the pre_sync + barrier section ahead of the frag/global section in each block. See ScheduleConfig.barrier_before_pre_ops.

Fields

• ​minimal (Bool): Suppresses per-block s_barriers and set_prio pairs; emits s_barrier only at top-of-half and the first cross-stage block. See ScheduleConfig.minimal_barriers.
• ​omit_set_prio (Bool): When minimal=True, drops the pre-MMA s_setprio[1] entirely. See ScheduleConfig.omit_mma_set_prio.
• ​sched_barrier_mask (Int): Bitmask of which blocks get trailing schedule_barrier fences. Default: 0b01010101.
• ​wrap_waits_sched_barrier (Bool): Wraps each contiguous wait/barrier group with schedule_barrier on both sides. See ScheduleConfig.wrap_waits_with_sched_barrier.
• ​barrier_before_pre_ops (Bool): Moves the pre_sync + barrier section ahead of the frag/global section in each block. See ScheduleConfig.barrier_before_pre_ops.

Implemented traits

AnyType, Copyable, ImplicitlyDestructible, Movable

Methods

default

static default() -> Self

Returns the ping-pong default: full barriers, no minimal mode.

Returns:

Self: A BarrierStrategy matching the framework default.

minimal_no_set_prio

static minimal_no_set_prio(sched_barrier_mask: Int = 0, wrap_waits: Bool = False, barrier_before_pre_ops: Bool = False) -> Self

Returns the cross-stage-rotation default: minimal barriers, no set_prio.

Args:

• ​sched_barrier_mask (Int): Bitmask of which blocks get trailing schedule_barrier fences.
• ​wrap_waits (Bool): Wrap wait/barrier groups with schedule_barrier on both sides.
• ​barrier_before_pre_ops (Bool): Move pre_sync + barrier ahead of the frag/global section in each block.

Returns:

Self: A BarrierStrategy configured for the cross-stage rotation kernel pattern.

Was this page helpful?

Thank you! We'll create more content like this.

Thank you for helping us improve!

😔 What went wrong?

Some code doesn’t work

It includes inaccurate information

It's missing information I need

It was difficult to understand

Other

Submit