Trapping Floating-Point Exceptions

[jeff_ryman]

My understanding is that Intel Fortran presently has no options to trap floating-point exceptions, as is possible with CVF. Please correct me if I am wrong. Will a future version have the option to trap these exceptions? I am trying to make a decision on a future compiler purchase, and this is an important issue for me. It caused me to dump Ryan-McFarland and switch to MS Fortran in MS-DOS days.

I also note that neither CVF nor Intel Fortran score particularly well on the Polyhedron Diagnostic benchmarks. Have the kinds of errors they don't diagnose been a real issue for anyone, or are these particular benchmarks not very relevant? I do note the excellent scores on execution speed, but diagnostics are more important to me.

[jeff_ryman]

I should have included integer overflow in my question about trapping exceptions.

[Steven_L_Intel1]

It is our intent that the future Intel Visual Fortran will have most if not all of CVF's capabilities in the area of exception handling.

The Polyhedron diagnostic tests are artificial in that some of the diagnostics are much more important than others, but a simple count ratio is included. We do expect our diagnostic capability to continue to improve over the coming releases, but our focus is not on scoring 100% on the Polyhedron suite. (Lahey did, and then two weeks later Polyhedron added more tests that Lahey didn't pass...)

Many of the tests not passed by CVF or IF are irrelevant with normal good Fortran coding practices, including explicit interfaces for all routines.

Steve

[jeff_ryman]

Thanks for the comments. I will have to wait and see what exception handling is included in IVF. I just purchased CVF v5.0 on eBay and will soon upgrade to CVF 6.x, so I will be ready for a potential upgrade to IVF later. If you are looking for input on future products, I will say that I consider not trapping exceptions to be a show-stopper. I understand that many benchmarks are artificial, and that you can't play the "chase the benchmark" game.

[Steven_L_Intel1]

Oh, we'll trap the exceptions. (You'll be happier with 6.6 than with 5.0 in this regard, BTW.) It's things such as intent mismatches across different source files that we're not likely to catch.

Steve