[llvm] r281732 - Reapplying r278731 after fixing the problem that caused it to be reverted.

[Sanjoy Das via llvm-commits]

Hi Pankaj,

Chawla, Pankaj wrote:
 > />> A loop which stores to a global variable has no side effects?/
 >
 > This seems to be an oversight.

I talked to David on IRC.  We agreed that it is fine to ignore a store
to a global for the purposes of proving that a loop is finite, that
is:

   while (true) global++;

is undefined behavior.

He was mostly concerned about the naming, and wanted us to change the
name to reflect that we're not trying to rule out side effects in
general, but only the sort that would justify a well defined infinite
loop.

@David: is that a correct summary?

-- Sanjoy