Add Simple Check to Prevent npasses>1 When Caliper Enabled

[michaelmckinsey1]

Summary

• This PR is a bugfix
• It does the following (modify list as needed):
  • Modifies src/common/RunParams.cpp
  • Fixes Incorrect attributes when npasses > 1 and using caliper #563

This PR addresses a known issue (#563) when using --npasses>1, where seemingly "incorrect" values will be propagated into the caliper files.

@MrBurmark and I discussed something like multiplying reps by npasses. Another option is we could rename reps to be reps/pass. However, @MrBurmark during this discussion, I forgot the issue of the other attributes being incorrectly set on the subsequent passes after the first pass. So the solution cannot be this simple.

To solve this:

1. We could add a guard so attributes are not set again. This would also require one of the solutions discussed above as well, as metrics would need to be divided by npasses and reps in post-processing.
2. flush Caliper files after each pass. So files would be named variant-tuning-passnum.cali like Base_Seq-1.cali. I think this would be a solution more befitting of Caliper's model than the first option.

This is just a temporary mechanism to prevent users from generating bad data by accident, until I implement a better solution, which I'm not sure will land before the 2025.12.0 release, so I am making this PR. This issue can also be really difficult for a user to catch, so it does need to be addressed in some form.

I tested some cases, which show what a user will see:

# If not building with caliper
### fine
raja-perf.exe --npasses 1 --dryrun
### fine
raja-perf.exe --npasses 2 --dryrun

# If building with Caliper
### fine
raja-perf.exe --npasses 1 --dryrun
### bad input
$ raja-perf.exe --npasses 2 --dryrun


Running with 1 MPI ranks...


Reading command line input...

Bad input: --npasses cannot be >1 when profiling with Caliper
Compiler: icpx-2025.2.0
...

[tdrwenski]

Seems good to me as a temporary fix to alert users!

I am okay with either solution as long as you can still get back the info about how many reps and passes were run. I don't think 10 reps and 2 passes is exactly equivalent to 20 reps and 1 pass so would be good to have that info.

[MrBurmark]

This seems like a reasonably simple stop-gap for the moment.
If you think that a file per pass is the solution that makes the most sense for caliper users then we can go to that.

[michaelmckinsey1]

Seems good to me as a temporary fix to alert users!

I am okay with either solution as long as you can still get back the info about how many reps and passes were run. I don't think 10 reps and 2 passes is exactly equivalent to 20 reps and 1 pass so would be good to have that info.

You are correct, they are not equivalent. There is different caching behavior for those two options.

[MrBurmark]

@rhornung67 Shall we merge?

[rhornung67]

@MrBurmark yes, go for it.