optimization flags

I have a program that uses alot of vectors. I have tried optimizing it at levels -O and -O2. The performance is much better at -O2. But the results the program gives are different. I have a test harness of about 150 tests that I have run berfore and after the optimization. Maybe the optimization is causing a problem with some of the floating point calculations. Does anyone know which flags in particular in -O2 may help improve the performance of the vectors?? I am using g++ version 3.4.2
Thanks
Mark

> But the results the program gives are different.
That usually means there is a bug (or an assumption) in your code.
If your program is correct (and your compiler code generator isn't broken, which seems unlikely), then optimisation should not affect the result.

> Maybe the optimization is causing a problem with some of the floating point calculations
Quite possibly.
http://www.daniweb.com/techtalkforum...ad.php?t=45388
Previously comparable floats may end up being non-comparable due to implicit extra precision.
Or put another way, if the compiler succeeds in keeping floating point values in floating point registers for longer, then expect some differences.

Post code?

Quote:

---

Originally Posted by Salem

Or put another way, if the compiler succeeds in keeping floating point values in floating point registers for longer, then expect some differences.

---

I really don't see why that would be if the code is written well... Copying from a register to memory would not cause a change... unless the type is coverted to a less precise vairable in the process (e.g. double is operated on and saved as a float which is then operated on with the result returned as a double). I think that you need to have a look through your code very carefully - easy to loose track of precision.

Another question - are the differences outside the level of precision you require or are they more significant?

> I really don't see why that would be if the code is written well
There's the rub - "if"

The code in question isn't well written otherwise it would not show up as having different behaviour depending on optimisation flags.

Maybe the OP should post code which shows the "problem", then perhaps we can give concrete help on how to fix it rather than just pages of waffle.