Slower run time...

[iklassen]

Hello,

We are currently upgrading Fortran code compiled with Digital Fortran 5 to Intel Fortran 9.0.028.

While some code executes faster, we are finding some programs run almost twice as slow. If the code and the machine are the same I would think there must be a compiler issue. Is there something specifically that I should be looking for? I've been playing with some compile options but haven't found anything yet that improves the speed.

Ian

[Steven_L_Intel1]

I would suggest you send an example to Intel Premier Support so that we can look at it. Be sure you are building a "release" configuration with optimization. You may also be enabling some run-time diagnostics that can slow things down.

[iklassen]

Steve, I don't think it will be possible to send out source code.

Something else I find unusual is that the new Intel compiled executable will run faster on some machines (compared to the previous Digital version) but on other machines run slower (again compared to the Digital version run on the same machine).

I would understand if the Intel versions always ran faster, but on some machines faster than others. But to have it actually run slower than the previous Digital version doesn't make sense. The conflicting information makes it difficult to isolate the problem.

Ian

[Steven_L_Intel1]

There's insufficient information here. I suggest you look at memory usage - is there more memory paging when it runs slower? It could be a lack of RAM. This would be mu initial guess based on what you've described.

[iklassen]

Sorry about the lack of information. I realize I don't have much to go on. Do you have any tests that I can run to help isolate the issue?

I thought it was a memory problem as well, but what doesn't jive is that the Digital version runs faster than the Intel version on the same machine. Does theIntel compilerhandle memory paging differently? The application definitely uses a lot of memory. Other than the /O options I didn't see any options for handling memory.

[Steven_L_Intel1]

There aren't any compiler options related to this. I would not normally expect a difference, but without seeing the program, I couldn't guess as the cause. The Windows task manager's Performance tab can be instructive - if you see the paging requests going way up, things are going to be slow.

It may just be that the code got a bit bigger and pushed you over the edge of your system configuration's boundary for virtual memory.

[lklawrie1]

I tried the upgrade (IA32 subset), recompiled my entire source and

a) the compile time seems a lot slower.

b) the program crashed -- which it was not crashing with 025.

So, I've gone back to 025 for now. Yes, I can upload everything to the premier site, but will take some time.

Linda