I am going to use C++ Armadillo library which handles matrices to generate matrix $B$ and $C$ from matrix $A$.
$$ A=[M_0,M_1,\ldots,M_{n-1}]^T $$
$$ B=[M_0,M_0,M_0,M_1,M_1,M_1,\ldots,M_{n-1},M_{n-1},M_{n-1}]^T $$
$$ C=[-M_0,0,+M_0,-M_1,0,+M_1,,\ldots,-M_{n-1},0,+M_{n-1}]^T $$
One possible way is to use a for loop doing that. But since for loop is slower than simple matrix operations involving graphic card processing, I was thinking if there is any simple matrix operation which can build matrix B and C.
Since $A$ is $(n-1)\times1$ while $B$ and $C$ are $3(n-1)\times1$, it is not possible to obtain B and C through a simple matrix multiplication like: $$B=D\times A, C=E\times A$$
I am not sure about tensors. Is there any GPU friendly operation for it?