While we generally give the line number where a handled error occurs (such as an IO error) , we don’t have an equivalent to “-traceback”. The best way to replicate this feature is to use PGI debugger, pgdbg, which comes standard with the compilers.
As for benchmarking, the thing to keep in mind is that the only true benchmark is your code. You can look at benchmark results others have done, but unless your code is similar, they may not be applicable. The ideal situation would be for you to find a representive data set for your application, run each compiler’s generated executable, and compare the times. Of course, make your best effort to find the optimal flag set for each.
A good comparison of compilers can be found at Polyhedron’s website http://www.polyhedron.com. You’ll find that Intel is faster with some codes, while PGI is faster on others. (Note: we found and fixed a missed opportunity with fatigue so you’ll see a vast performance improvement with the PGI 6.0 release)
The caveat to using this and other published benchmarks is that the compiler companies will tune their performance for these codes. While we look at these codes for performance opportunities, we generally focus our efforts on real world application performance. PGI is the compiler to use for applications, not just benchmarking.
If you need help in finding the best optimization for your code, let us know.