Problem in Tensor Cores

wmma::fragment<wmma::matrix_a, M, N, K, double, wmma::row_major> a_frag;
...
wmma::load_matrix_sync(a_frag, A, K);
...
wmma::mma_sync(ab_frag, a_frag, b_frag, zero_frag)

It calculates the matrix A * matrix B.
Since I want to calculates many A * B and A won’t change, I’m thinking whether I can define a_frag and load A to a_frag only once, and then reuse a_frag. I tried but it failed. So I want to ask that will a_frag change after using mma_sync? Thanks!

It seems that the lifespan of fragment is only warp. And I can’t reuse fragment cross the warp thought it is in the same block.