Mojo function

reduce

reduce[val_type: DType, simd_width: Int, //, shuffle: fn[DType, Int](val: SIMD[$0, $1], offset: SIMD[uint32, 1]) -> SIMD[$0, $1], func: fn[DType, Int](SIMD[$0, $1], SIMD[$0, $1]) capturing -> SIMD[$0, $1]](val: SIMD[val_type, simd_width]) -> SIMD[val_type, simd_width]

Performs a generic warp-wide reduction operation using shuffle operations.

This is a convenience wrapper around lane_group_reduce that operates on the entire warp. It allows customizing both the shuffle operation and reduction function.

Example:

    from gpu.warp import reduce, shuffle_down

    # Compute warp-wide sum using shuffle down
    @parameter
    fn add[type: DType, width: Int](x: SIMD[type, width], y: SIMD[type, width]) capturing -> SIMD[type, width]:
        return x + y

    val = SIMD[DType.float32, 4](2.0, 4.0, 6.0, 8.0)
    result = reduce[shuffle_down, add](val)
    from gpu.warp import reduce, shuffle_down

    # Compute warp-wide sum using shuffle down
    @parameter
    fn add[type: DType, width: Int](x: SIMD[type, width], y: SIMD[type, width]) capturing -> SIMD[type, width]:
        return x + y

    val = SIMD[DType.float32, 4](2.0, 4.0, 6.0, 8.0)
    result = reduce[shuffle_down, add](val)

Parameters:

val_type (DType): The data type of the SIMD elements (e.g. float32, int32).
simd_width (Int): The number of elements in the SIMD vector.
shuffle (fn[DType, Int](val: SIMD[$0, $1], offset: SIMD[uint32, 1]) -> SIMD[$0, $1]): A function that performs the warp shuffle operation. Takes a SIMD value and offset and returns the shuffled result.
func (fn[DType, Int](SIMD[$0, $1], SIMD[$0, $1]) capturing -> SIMD[$0, $1]): A binary function that combines two SIMD values during reduction. This defines the reduction operation (e.g. add, max, min).

Args:

val (SIMD[val_type, simd_width]): The SIMD value to reduce. Each lane contributes its value.

Returns:

A SIMD value containing the reduction result broadcast to all lanes in the warp.