warp.tile_transpose#

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

  • Kernel

  • Differentiable

Transpose a tile.

For shared memory tiles, this operation will alias the input tile. Register tiles will first be transferred to shared memory before transposition.

Parameters:

a – Tile to transpose with shape=(M,N)

Returns:

Tile with shape=(N,M).