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
single_buffer_reorder
single_buffer_reorder(logical: List[OpDesc], config: PipelineConfig) -> List[OpDesc]
Single-buffer pipeline reorder: logical iteration → pipelined steady-state.
Reorders ops to overlap consecutive iterations, hiding memory latency behind compute. Adds one extra barrier (sync) for the pipelined steady-state, so the output has len(logical) + 1 ops.
Output order: frag[1..T-1], compute[0], sync, store_shared, load_global, compute[1..T-1], sync, frag[0]
Returns:
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!