warp.untile

warp.untile(a: Tile[Any, tuple[int, ...]]) → Any

Kernel

Differentiable

Convert a tile back to per-thread values.

This function converts a block-wide tile back to per-thread values.

• If the input tile is 1D, then the resulting value will be a per-thread scalar

• If the input tile is 2D, then the resulting value will be a per-thread vector of length M

param a:

A tile with dimensions shape=(M, block_dim)

returns:

A single value per-thread with the same data type as the tile

This example shows how to create a linear sequence from thread variables:

@wp.kernel
def compute():
    i = wp.tid()

    # create block-wide tile
    t = wp.tile(i)*2

    # convert back to per-thread values
    s = wp.untile(t)

    print(s)

wp.launch(compute, dim=16, inputs=[], block_dim=16)

Prints:

0
2
4
6
8
...