What a weird post.
Lea became very slow in recent Intel cpus? News to me. Hard to imagine them shipping such a cpu considering x86 compilers pattern match to Lea hella aggressively.
Weird to show the more complex arithmetic codegen as a downside of not allocating int boxes. Having to shift or add (or both) a lot is way cheaper than asking the GC for memory or dereferencing a pointer to a box.