| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
Arnold Neumaier <arnold.neumaier@xxxxxxxxxxxx> wrote on 07/20/2011 10:54:00 AM:
> On Wed, July 20, 2011 16:43, Nate Hayes wrote:
> > Arnold Neumaier wrote:
. . .
> One cannot (and should not) require an optimizing compiler to yield the
> _same_ answer; already optimizing x/3+x/3+x/3 to x will change the
> interval output (in this case to something better).
>
> But one must insist on that an optimization never changes a correct answer
> for the unoptimized version into a false answer in the optimized version.
>
> Arnold Neumaier
Most compilers have an option to preserve exact semantics even when optimizing (IBM's is "strict"). Some users need exact semantics even when optimizing, and when the option is used it would be incorrect to replace "x/3+x/3+x/3" with "x". It is better to some but wrong to others.
- Ian McIntosh IBM Canada Lab Compiler Back End Support and Development