cuda.tile.scan

cuda.tile.scan(x, /, axis, func, identity, *, reverse=False)

Apply custom scan (inclusive prefix) function along axis.

Parameters:

• x – input tile or a tuple of tiles to be scanned. If a tuple is provided, shapes of the tiles in the tuple must be broadcastable to a common shape.

• axis (int) – an integer constant that specifies the axis to scan along.

• func – function for combining two values. If x is a single tile, then the function must take two 0d tile arguments and return the combined 0d tile. For example, lambda a, b: a + b or operator.add can be used to implement cumsum. If x is a tuple of N tiles, then the function takes 2N tiles and returns a tuple of N combined tiles. The first N arguments correspond to one of the groups of values being combined, while the rest correspond to the other.

• identity – a constant scalar or a tuple of constant scalars that specifies the identity element of the func.

• reverse (bool) – if True, the scan is performed in the reverse direction along the axis. Default: False.

Returns:

Scanned tile, or tuple of scanned tiles, depending on the type of x.