[gmx-users] Warning in compiling GMX4.0.2

[Mark Abraham]

Xavier Periole wrote:
> 
> Here a following up of the compilation on the IBM.
> 
> I got the following warning:
>      1500-036: (I) The NOSTRICT option (default at OPT(3)) has the 
> potential
> to alter the semantics of a program. Please refer to documentation on the
> STRICT/NOSTRICT option for more information.
> 
> Should that be a worry?

Depending on the architecture, different optimizations get added in as 
-Ox gets higher. You can selectively add "higher" things, e.g. -qhot, 
which is probably a default at -O4. You should start with the compiler 
man page to get some idea what this might affect. Look for -qstrict to 
see what optimizations might be involved here. For example, there are 
some floating-point optimizations, particularly involving reciprocals or 
reciprocal square-roots, that only produce approximate results. One 
might or might not want to use these, hence the warning. Other 
optimizations might enable the compiler to not check for underflows, 
division by zero or other such maths-fails. All these kind of things can 
mean that the optimized code produces a different result from non-optimized.

Roughly speaking, I'd say it's not a worry for GROMACS.

 >
 > Note that the warning disappeared with the option -qaltivec added to the
 > CFLAGS and CXXFLAGS.

That probably just means that the bit that was causing the warning to be 
issued is now being vectorized, e.g. a reciprocal square-root.

Mark