DRB050-functionparameter-orig-no.f95 has a data race about dataracebench HOT 4 CLOSED

Same for DRB052-indirectaccesssharebase-orig-no.f95 (private(idx1,idx2)) and DRB061-matrixvector1-orig-no.f95 (private(sum)).

In the C version, these variables are declared in the parallel region and therefore private. In Fortran they need explicit privatization.
I only checked the '-no.f95' codes, where TSan detected a "false positive". I guess the same privatization issue will be present in several of the '-yes.f95' codes.

DRB133-taskdep5-orig-omp45-no.f95 also has a race, because some depend clauses are missing (compared to the c code).

from dataracebench.

@jprotze, I agree with DRB050 ,52 and 61. And have fixed them too.

Regarding, DRB133, I'm not able to understand what is missing? Can you please help it here. Thanks.

from dataracebench.

All the tasks in the fortran version of the code should have the same depend clauses as in the C code. Especially the second task has no depend clauses.

from dataracebench.

Thanks. FIxed. #82

from dataracebench.