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

block_prefix_sum

def block_prefix_sum[dtype: DType, //, num_elements: Int, thread_base: Int = Int(0), n_scoped_threads: Int = Int(0), barrier_id: Int = Int(-1)](_val: Scalar[dtype]) -> Scalar[dtype]

Performs a prefix sum (scan) operation across all threads in a block.

Parameters:

• ​dtype (DType): Element type of the values being scanned (inferred).
• ​num_elements (Int): Number of active threads in the block that contribute values to the scan.
• ​thread_base (Int): Absolute thread_idx.x of the first participating thread (0 in the standalone kernels; the FFN thread count when a comm thread-class runs the scan in a fused megakernel). All indices are relative to this base.
• ​n_scoped_threads (Int): Thread count the scan's block barriers scope to when barrier_id >= 0 (the comm thread-class size).
• ​barrier_id (Int): Named-barrier id for the scan's barriers, or negative (the default) for the full-block barrier().

Args:

• ​_val (Scalar[dtype]): The per-thread value to contribute to the inclusive prefix sum.

Returns:

Scalar[dtype]