Host-level Operations

Tiled(x)

Wrapper for CUDA arrays to allow dispatching to cuTile kernels.

@. expr

Like Base.@. but wraps every value-position leaf in Tiled(), routing the broadcast through cuTile kernels.

using cuTile; const ct = cuTile
ct.@. C = A + sin(B)
# equivalent to: Tiled(C) .= Tiled(A) .+ sin.(Tiled(B))

For in-place assignment, the expression evaluates to the original destination array (C above), not its Tiled wrapper.