g77: Different results when compiling with/without -fbounds-check ?!

[Toon Moene]

Richard Ems wrote:

> After 2 days searching for errors and comparing my test case result's
> with different option combinations, I found
> that compiling with -fbounds-check and without it returns (for my test
> case) different results.

I can reproduce your results using g77-2.95.2 on i686-linux.

As far as I can see, the differences are in the 7th digit, not
unreasonable for a 32-bit floating point number.

The reason they are different at all is caused by the fact that the
extra code that -fbounds-check introduces leads to more intermediate
stores of register values into memory.  On the Intel architecture that
means that more rounding from 80-bit floating point values to 32-bit
ones take place in case of -fbounds-check than when not compiling that
way.

You could work around this by adding the -ffloat-store option in all
cases - however, that would slow down your program considerably.

Hope this helps,

-- 
Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
GNU Fortran 95: http://g95.sourceforge.net/ (under construction)