View page source

warp.tile_axpy#

warp.tile_axpy(
alpha: Any,
src: Tile[Any, tuple[int, ...]],
dest: Tile[Any, tuple[int, ...]],
) → None#

  • Kernel

  • Differentiable

Scale src by alpha and accumulate into dest.

Performs a fused multiply-add directly into the destination tile without creating an intermediate scaled tile.

Parameters:

  • alpha – Scalar multiplier (must match the tile’s underlying scalar type).

  • src – Source tile (must have same shape and dtype as dest).

  • dest – Destination tile, modified in place.

Example

@wp.kernel
def compute():

    dest = wp.tile_ones(dtype=float, shape=4) * 2.0
    src = wp.tile_ones(dtype=float, shape=4) * 3.0
    wp.tile_axpy(5.0, src, dest)

    print(dest)

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

[17 17 17 17] = tile(shape=(4), storage=register)