View page source

warp.tile_dot#

warp.tile_dot(
a: Tile[Any, tuple[int, ...]],
b: Tile[Any, tuple[int, ...]],
) → Tile[Any, tuple[Literal[1]]]#

  • Kernel

  • Differentiable

Compute the dot product of two tiles.

Computes a full contraction between corresponding elements and sums the results. For scalar tiles this is the standard dot product; for vector tiles each pair is contracted via wp.dot; for matrix tiles it is the Frobenius inner product (the sum of element-wise products over all axes).

Equivalent in Python to wp.tile_sum(a * b) for scalar tiles and to wp.tile_sum(wp.tile_map(wp.dot, a, b)) for vector tiles, but without the intermediate tile and shared-memory round trip the explicit forms would require. Matrix-typed tiles have no purely-elementwise Python equivalent because wp.dot is not defined for matrices.

Parameters:

  • a – First tile operand.

  • b – Second tile operand (must have same shape and dtype as a).

Returns:

A single-element tile holding the dot-product result. Index the tile at [0] to obtain the scalar value.

Example

@wp.kernel
def compute():

    a = wp.tile_ones(dtype=float, shape=64)
    b = wp.tile_ones(dtype=float, shape=64) * 2.0
    d = wp.tile_dot(a, b)

    print(d)

wp.launch_tiled(compute, dim=[1], inputs=[], block_dim=64)

[128] = tile(shape=(1), storage=register)