OpenMPTarget: Guard scratch memory usage in ParallelReduce

[masterleinad]

This should fix the occasional openmptarget.partitioning_by_args failures. AFAIK, we don't guard the usage of scratch memory in the OpenMPTarget backend so that we could run into race conditions when executing parallel_reduce from multiple simultaneously like partitioning_by_args does.
Names of the variables are up for bikeshedding, of course.

[rgayatri23]

Can we push this one post #6380 . That is where we actually start using scratch memory?
I am working on why it fails one of the algorithm unit tests.

[masterleinad]

Can we push this one post #6380 . That is where we actually start using scratch memory?

That pull request is really for the user-requested L0 scratch memory space in TeamPolicy and doesn't touch resize_scratch, right? The guard here really only is for the use of global/non-shared/L1 scratch space.

[masterleinad]

Retest this please.

[rgayatri23]

Can we push this one post #6380 . That is where we actually start using scratch memory?

That pull request is really for the user-requested L0 scratch memory space in TeamPolicy and doesn't touch resize_scratch, right? The guard here really only is for the use of global/non-shared/L1 scratch space.

Until the #6380 is merged, there is only global scratch in OpenMPTarget. Even if L0 is requested it comes from global memory.

[masterleinad]

Until the #6380 is merged, there is only global scratch in OpenMPTarget. Even if L0 is requested it comes from global memory.

Right, so the two pull requests are orthogonal.

[rgayatri23]

Its a good idea, since the scratch is on global memory, we want to avoid multiple instances accessing that location.
Where are you unlocking the mutex, or does it get destroyed when the instance is completed with its work?

[dalg24]

Looks good besides the static.

[masterleinad]

Where are you unlocking the mutex, or does it get destroyed when the instance is completed with its work?

All the scope_guards have a comment that says

// The constructor acquires the mutex which is released in the destructor.

[rgayatri23]

Oh, missed the comment. Thanks!