Memory access to GlobalAddress + Immediate Offset can be optimised (by using Immediate expressions to wrap the GlobalAddress + Immediate Offset, which will be handled by the linker).
See this example where we could remove the move instruction and have this load instruction: lw r21, zero, my_table + 4