Load a double-precision (64-bit) floating-point element from memory into the lower of result, and zero the upper element. mem_addr does not need to be aligned on any particular boundary.