[llvm-bugs] [Bug 34582] [OpenMP 4.5] Array segment mapping to device with default length (e.g. map(array[1:]) ) fails at runtime

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=34582

Jose Manuel Monsalve Diaz <josem at udel.edu> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |INVALID
             Status|NEW                         |RESOLVED

--- Comment #1 from Jose Manuel Monsalve Diaz <josem at udel.edu> ---
The problem boils down to the implicit mapping that occurs in the second
#pragma omp target. This implicit mapping is like having map(tofrom: a1d[0:N]).
Which goes agains the original mapping. 

#pragma omp target data map(from: a1d[1:])
{
#pragma omp target map(tofrom: isHost)
{
  // Some aid[] use.
}
}

Is equal to 

#pragma omp target data map(from: a1d[1:])
{
#pragma omp target map(tofrom: isHost) map(tofrom: a1d[0:N])
{
  // Some aid[] use.
}
}

This is generating a conflict. Even though the specs in the mapping section are
pretty clear what to do with respect to the reference counting, it doesn't say
anything regarding implicit mapping of a variable that is already in the table.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20170913/e2686a42/attachment-0001.html>