Load a double-precision (64-bit) floating-point element from memory into both elements of dst. mem_addr does not need to be aligned on any particular boundary.