CVF to IVF, conditions for failure of Local variable storage = /Qsave

[salbruce]

In IVF 11, what kind of programming structure might fail with /Qsave or /Qauto but not the other two options for Local variable storage?

Some background: I am converting a number of dll projects from CVF to IVF 11. By default the Local variable storage is set to /Qsave. This has worked for all projects but one. In that case a subroutine fails. Trying all four options for Local variable storage I get:

Default local storage -- runs
All variables save (/Qsave) - subroutine fails
Scalar local variables automatic (/Qauto_scalar) - runs
Local variables automatic (/Qauto) -- fails.

Additional notes:
- No problems with CVF
- Failures happen with IVF /Qsave with both XP and Vista
- No variables are explicitly declared with Save or Automatic
- Array passed to the subroutine are allocatable and allocated before the subroutine call.
- Two arrays are allocated/deallocated within the subroutine
- Another project that calls the same subroutine does not fail with any of the settings for Local variable storage.
- Only Common variables are two pointers needed for calls to another dll. These are used in most of my projects with no problems.

Thanks in advance for your ideas.

[Steven_L_Intel1]

What does "fails" mean here?

Do you have any allocatable arrays declared in a subroutine or function (not as a dummy argument?) A known issue with /Qsave is that it makes those SAVE too, which is incorrect.

That /Qauto also "fails" suggests that you have code that depends on an array being SAVEd.

[Paul_Curtis]

The best way to ensure CVF-type behavior in IVF is to go through your code and add explicit SAVE attributes to all those quantities which need to be preserved. As noted, /Qsave does not exactly restore the CVF behavior, and in fact introduces lots of other complications and unwanted effects.

[salbruce]

Steve asked, "What does "fails" mean here?" I am calling an eigenanalysis subroutine. I generate an error code when the routine fails to converge on a solution after some number of iterations. This is what I mean by "fails." I know that this particular data set is no problem for eigenanalysis. The failure could mean any of a number of problems, such as a problem with the input array or a problem with some of the temporary storage arrays that I allocate within the routine. The routine is tried and true with numerous compilers. I was hoping to shortcut debugging this if I could figure out what the /Qsave was doing.

Steve, thanks for pointing out the problem with the use of /Qsave with the arrays with allocated arrays within the routine is a good reason not to use /Qsave. I am indeed allocating arrays within the routine.

I found that if I run the routine just once, no problem, but if I call it multiple times with /Qsave or /Qauto it fails.

Paul thanks for the suggestion of explicit SAVE. In this case I don't want the routine to save anything, but for future reference it sounds like using the SAVE attribute is much better than trying to use /Qsave.

Thanks again.