should MEM tracking be able to optimize this?

[Richard Kenner]

    "calc1" one has too many loads and stores...  Should MEM tracking be
    able to dissambiguate the non-dependend loads and stores in "calc1"?

No, only in calc2.  The point is that it can now disambiguate when you
have separate variables, like you do in calc2.  In calc1, what you
have are different *fields* of the same variable and that isn't what
is tracked.