Support for the ICV nthreads-var in ompd_enumerate_icvs(), ompd_get_icv_from_scope() and ompd_get_icv_string_from_scope()

[jinisusan]

Review and pull request for providing support for the ICV nthreads-var. (#72).

Thanks!
Jini.

[jinisusan]

Gentle reminder !

Thanks,
Jini

[jprotze]

As I read the spec, ompd_get_icv_from_scope should always just provide a single integer in icv_value for which the tool provides the storage. If that is not possible, the function can return ompd_rc_incompatible.
In this case, the tool would need to use and parse ompd_get_icv_string_from_scope.

For ompd-nthreads-var this means, the function can return a value if the variable is just a single value, but ompd_rc_incompatible for nested execution with a list of values.

Does this make sense?

[jinisusan]

Thank you very much, Joachim, for pointing this out. You are right -- this makes sense. Will send an updated revision.

[jinisusan]

Hi again Joachim,

As I thought more about this, I feel that it might make more sense for the debugger to directly invoke ompd_get_icv_string_from_scope() for the ICVs like nthreads-var where the value of the ICV is a list (even if it constitutes of only a single element). This would avoid the (possibly) unnecessary calls from the debugger to ompd_get_icv_from_scope() for the cases where the value is a list of multiple elements. What do you think ? I am planning to ask about this in the omp-tools forum also.

Thanks,
Jini.

[jinisusan]

Have committed changes according to Joachim's and John's suggestions in omp-tools. Please take a look.

Thanks!
Jini

[jinisusan]

Gentle reminder ! Thanks !

[jinisusan]

Made commits to replace an assert call with a ompd_rc_error return in the case of the callback table not being available (as per the review comment in #75 )

[jprotze]

Note for OMPD specification: we might add a new return code: ompd_rc_incomplete.
For the currently available return codes, I would not expect meaningful values unless ompd_rc_ok.

[jinisusan]

Thanks, Joachim!