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
SingleBufferSchedule
struct SingleBufferSchedule[num_k_tiles: Int]
Declarative schedule for the default single-buffer matmul.
The algorithm declares logical ops (tag + buffer metadata only). The target cost model supplies resource, latency, and role. The framework derives the pipeline ordering and dependency edges.
Parameters
- num_k_tiles (
Int): Number of K-tiles along the contraction dimension, each producing one fragment load and one MMA compute op.
Fields
- hints (
AMDScheduleHints):
Implemented traits
AnyType,
ImplicitlyDeletable,
PipelineSchedule
Methods
__init__
def __init__(out self, hints: AMDScheduleHints, target: TargetProfile = mi355x_single_buffer_target())
config
def config(self) -> PipelineConfig
Returns:
PipelineConfig
build_body
def build_body(self) -> List[OpDesc]
Declare logical ops, annotate with cost model, reorder for pipeline.
Returns:
transform_kernel
def transform_kernel(self, ker: List[ScheduleEntry], body: List[OpDesc]) -> List[ScheduleEntry]
Append AMD schedule_group_barrier hints to kernel entries.
Args:
- ker (
List[ScheduleEntry]): The kernel schedule entries to append AMD schedule_group_barrier hints to. - body (
List[OpDesc]): The annotated body ops used to count fragment loads and split loop-carried vs non-loop-carried frags.
Returns:
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!