Re: Help required in gcc 3.2.3 optimization

[Bernd Strieder]

Hello,

Pradeep  Kumar wrote:

> <P>
> &nbsp; <BR>
> Hi,<BR>
> <BR>
> &nbsp; &nbsp; &nbsp;We are finding the output differences between O2
> flagged , optimized version and a –g3 –gdwarf-2 flagged debugable
> version <BR>
> &nbsp; &nbsp; &nbsp;of a C++ application , which does some very heavy
> floating point xnumber crunching.&nbsp;  <BR> We are noticing


Please don't post HTML. You can see above, how most readers of this
newsgroup probably view your posting.

Due to optimization the app might do some calculations with higher
precision. E.g. on the i386 platform the floating point registers offer
higher precision than double. If intermediate results aren't stored to
memory, performance is gained, while the result has higher precision,
i.e. it is different. See documentation on -ffloat-store. Do you have a
worst-case estimate for the rounding errors in your calculations? Is it
larger than the difference you see? If yes, then you have no problem at
all.

Bernd Strieder