Another important aspect not brought up is the semantics of the programming language itself. FORTRAN doesn't have memory aliasing (at least in arrays -- it's b

Another important aspect not brought up is the semantics of the programming language itself. FORTRAN doesn't have memory aliasing (at least in arrays -- it's been a while since I wrote any FORTRAN code) so the compiler can make assumptions (and from them optimizations) that it cannot in, say, C or C++. The author points out the "myopic" nature of compilation, implying that a more global view of the code might allow the compiler to prove that no aliasing happens in a particular code path. I think that problem is Turing complete.