We now emit movq %rdi,16(%r14,%rsi,8) instead of leaq 16(%r14),%rax movq %rdi,(%rax,%rsi,8) This helps e.g. byte array indexing.