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

reduce_sum

def reduce_sum[dtype: DType, InputFn: def[width: Int, alignment: Int, rank: Int](IndexList[rank]) -> SIMD[dtype, width] & RegisterPassable & ImplicitlyCopyable, OutputFn: def[width: SIMDLength, rank: Int](IndexList[rank], SIMD[dtype, width]) -> None & RegisterPassable & ImplicitlyCopyable, /, target: StringSpan[ImmStaticOrigin], *, reduce_dim: Int](input_fn: InputFn, output_fn: OutputFn, input_shape: Coord, context: Optional[DeviceContext] = None) where (eq InputFn.dtype, dtype) where (eq OutputFn.dtype, dtype)

Sums input values along reduce_dim via the Row free-form body layer: one ReduceSum phase plus a per-row emit.

input_fn / output_fn are value-based closures (a type parameter bound to a callable trait + a value argument), not comptime capturing[_] parameters.