(Closes #2561) Add support for allocatables in the HoistLocalArraysTrans and check for kind/type dependent symbols

[sergisiso]

No description provided.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.90%. Comparing base (9a3e94b) to head (5159673).
⚠️ Report is 36 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2893   +/-   ##
=======================================
  Coverage   99.90%   99.90%           
=======================================
  Files         371      371           
  Lines       52105    52155   +50     
=======================================
+ Hits        52055    52105   +50     
  Misses         50       50           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[sergisiso]

Integration test passed but the change was not applied almost anywhere because the arrays of interest have:
real(kind=wp), ...
but the transformation validation does:

if isinstance(sym.datatype.intrinsic, DataTypeSymbol):
    continue
if isinstance(sym.datatype.precision, DataSymbol):
    continue

I think we could refine this to check that this symbol is not local to the subroutine and the subroutine doesn't have wilcard imports.

[arporter]

Nice work Sergi. I only have minor comments.
I see that examples/nemo/scripts/acc_loops_trans.py has hoisting of local arrays switched on and looking at the 'historical' plot, performance seems OK.

[sergisiso]

@arporter This is ready for another look

[arporter]

Almost there. Unfortunately last time I missed that it would be better to use reference_accesses() rather than hardwire the check on precision and type symbols. Once that's done, this can go on. Please can you also check why the docs build failed.

[sergisiso]

@arporter Trying to address your last comment exposed a passthrough bug that we need to solve, so for now I just added a TODO. The linkcheck was failing for password protected links - don't ask me why we didn't fail on these before. Ready for another look.

[arporter]

Thanks Sergi, all looks good now.
I'll fire off the integration tests to be on the safe side.

[arporter]

Integration tests were all green.
Will proceed to merge.