Odd problem - AMD processor

[lklawrie]

We're seeing a difference in one file's execution on an AMD processor (sorry, don't know speed or details). I did not have any "require" extensions in the compile and have tried both "blended" and "pentium 4" "compile for" options.

Does anyone have any ideas of other compiler options that might be coming into play?

Linda

[Steven_L_Intel1]

The math library does its own CPU dispatching and you can't turn it off. You'll get one code path on AMD (or Pentium III or earlier) and different code paths for Pentium 4 and newer. I have seen this cause low-order result differences as different instruction sequences are used.

[lklawrie]

Result differences would be understandable. This is an "error condition" that we detect that is not detected on any pentium (or with several other compilers, debug modes, etc) but is showing up on the AMD machine. That's the odd part.

Same code, virtually, with all the other compilers.

Linda

[TimP]

Most 32-bit Athlons were P-III compatible. With recent Intel compilers, -QxK would work, but not -QxW. -QxW should be OK for AMD 64-bit desktop processors. Turion has been known to have problems with SSE2 instructions. Some of the issues are undocumented.

You also need to know whether a 32- or 64-bit OS is installed, and don't try to run incompatible combinations. 32-bit applications, with all necessary 32-bit libraries present, should run on 64-bit OS, but not the other way round.

There are quite a few issues to keep straight, and guessing won't cut it.

[Steven_L_Intel1]

At this point I would say that you'd have to figure out what computational differences are resulting in the "error" reports. You'll have to do that by adding code to display intermediate values, or if you're lucky, debug on the AMD system.

[lklawrie]

It may come down to that. Unfortunately, the team member with the AMD machine is in Chicago and I am in Colorado.

I did try the /Qxn compile -- crashed right out of the starting gate for him.

I do have some hints though as he is seeing some odd errors that I found when I ran the g95 compiler on a different problem (it does not see this problem).

Thanks for the help.

Linda