Do we need to recompile entire code base ?

Question

Do we need to recompile the entire code base to utilize the efficiency of a Sun SPARC Enterprise T5220 Server.
We will not be able to perform any of the optimization on the code, as it affects the working nature of the product.
So I'm not able to utilize http://www.spec.org/cpu2006/flags/Sun-Solaris-Studio12-and-gccfss4.2.html#Section_compiler
any of the above flags.
I'm able to run the software built on solaris 10 using gcc-3.4.2. can you please suggest is it required to build the entire codebase or not ?
are there any special flags with gcc or gfss compilers for getting better performance on multi core architectures?

Answer 1

We will need a little bit more information about the nature of the application etc, to make any reasonable recommendations.
You mention
"We will not be able to perform any of the optimization on the code, as it affects the working nature of the product. "
Compiler optimizations are always safe, as long as your code is correct. Some subset of the compiler flags enable more aggressive optimizations and can alter the precision of the floating point operations like -fsimple=2. These are however, documented and you can avoid such flags if you feel it may effect the stability of your application. -fast is a good starting point.
There are a wealth of tools offered by Gccfss and Sun Studio. I am listing them here in no specific order.
1. Please take a look at Performance Analyzer, works with Gccfss.
http://developers.sun.com/sunstudio/documentation/ss12/mr/READMEs/analyzer.html
You can download it from developers.sun.com/sunstudio
2. ATS is a tool that is offered part of the distribution, whose whole purpose it to help you tune your application. It is already bundled in your download. You can read more about it here
http://cooltools.sunsource.net/ats/
Let it run on your binary, and it will automatically suggest the best options for your application.
3. You can try -xautopar which enable automatic parallelization. The benefit largely depends on the parallelization opportunities that can be discovered by the compiler. This will heavily depend on your application though.
4. You should definitely try "-fast -xipo", the former which is a macro flag which enables aggressive optimizations in the compiler, the latter which enables inter procedural optimizations.
Thanks
Prashanth