Interpreting Profiler output

[Intel_C_Intel]

I got the profiler running, now I need to understand the output. Function timing gives me the following statistics (for the top few functions):

        Func           Hit 
        Time   %      Count  Function 
------------------------------------- 
    2048.406  11.5  767622 _for_check_env_name (for_io_util.obj) 
    1875.643  10.6    2959 _SOLVEFLOWP2@0 (solveflowp2.obj) 
    1286.843   7.2  681089 _PROPMIX@72 (propmix.obj) 
    1105.056   6.2 3241186 __OtsFill (sfill.obj) 
     856.845   4.8 9406814 __OtsStringCompareEqlPadded (scmpeqlp.obj) 
     603.523   3.4  248465 _LENST@8 (cha.obj) 
     577.943   3.3  137287 _FMASS@168 (tfmass.obj) 
     420.372   2.4  596982 _PROPTHRMS@52 (propthrms.obj) 
     417.869   2.4 1071130 __FIIfexp_ (intrini.obj) 

The functions that are not in all caps are not in my code and come from the Fortran libraries, I presume. How do I find out where _for_check_env_name is being called? For this particular case the hit count for this routine increases with the square of the number of timesteps I take. For other cases it is barely used.

I tried collecting attribution data with the /AT option to PREP, but I can't make heads or tail out of the extra output.

[Steven_L_Intel1]

Interesting. I would have thought the _for_check_env_name would be called only for an OPEN. I'll have to ask the RTL folks about this tomorrow. Do you do a lot of OPENs?

Steve

[Steven_L_Intel1]

My RTL buddy says that the likely candidate is the exception handler, which could be triggered if you had compiled /fpe:0 and were getting lots of underflows. Sound likely?

Steve

[Intel_C_Intel]

It's probably the exception handler because I do have lots of underflows for this case and it is compiled with fpe:0. I'll try changing the compiler setting and use a model that doesn't generate as many underflows.

I get e-mail replies to these posts that include the text:

View/reply at Interpreting Profiler output

but this URL doesn't seem to work.