is it possible to have a square unary operator op(x) = x*x, eg. CUTENSOR_OP_POW2, in a future version of cutensor library ?
or are the unary operators available in cutensor library based on hardware instruction set of tensor cores ??
is it possible to have a square unary operator op(x) = x*x, eg. CUTENSOR_OP_POW2, in a future version of cutensor library ?
or are the unary operators available in cutensor library based on hardware instruction set of tensor cores ??