warp.tile_reshape#

warp.tile_reshape(
t: Tile[Any, tuple[int, ...]],
shape: tuple[int, ...],
) → Tile[Any, tuple[int, ...]]#
  • Kernel
  • Differentiable

Return a view of a tile with a new shape.

The view aliases the source tile, so writing through it modifies t. Taking the view requires shared storage, but the view itself is non-owning and allocates no additional storage.

Elements keep their order in memory, which for a row-major layout matches numpy.reshape().

Parameters:

  • t – Input tile to reshape. Must be contiguous in memory, not a strided or reversed view. Copy it into a new tile with tile_assign() first if needed.

  • shape – New shape, whose total number of elements must match t. Entries must be compile-time constants; at most one may be -1, which is inferred from the others.

Returns:

A non-owning tile that aliases t with the requested shape.

Example

@wp.kernel
def reshape_matrix(a: wp.array2d[float], out: wp.array2d[float]):
    t = wp.tile_load(a, shape=(2, 4))
    r = wp.tile_reshape(t, shape=(4, -1))

    wp.tile_store(out, r)

a = wp.array(np.arange(1, 9, dtype=np.float32).reshape(2, 4), dtype=float)
out = wp.zeros((4, 2), dtype=float)

wp.launch_tiled(reshape_matrix, dim=1, inputs=[a], outputs=[out], block_dim=4)

print(out.numpy())

[[1. 2.]
 [3. 4.]
 [5. 6.]
 [7. 8.]]