Jakub,

The attached patch and ChangeLog reflect the updates from your
review: https://gcc.gnu.org/ml/gcc-patches/2015-11/msg00703.html.

The most significant change was the one that you suggestion:

On 11/06/2015 02:28 PM, Jakub Jelinek wrote:
 > Perhaps what would work is stick the exit clauses you need for automatic
 > vars in the function inside of some pointer map / hash table / whatever,
 > and then in gimplify_bind_expr in the
 >    /* Add clobbers for all variables that go out of scope.  */
 > if if flag_openacc && the pointer_map /hash table has any entries look
 > up each variable in there and collect the clauses from those vars that go
 > out of scope, after the loop if any were collected construct the statement
 > you need prepend it to cleanup (so that it works before restoring VLA memory
 > and before the clobber stmts).

This particular change allowed for the removal of all the
'stuff looks broken' code. Thanks for the suggestion.

The following change I did not address:

On 11/06/2015 01:03 PM, Jakub Jelinek wrote
 >> @@ -5841,6 +5863,8 @@ omp_default_clause (struct gimplify_omp_ctx *ctx, tree 
decl,
 >>         flags |= GOVD_FIRSTPRIVATE;
 >>         break;
 >>       case OMP_CLAUSE_DEFAULT_UNSPECIFIED:
 >> +      if (is_global_var (decl) && device_resident_p (decl))
 >> +flags |= GOVD_MAP_TO_ONLY | GOVD_MAP;
 >
 > I don't think you want to do this except for (selected or all?)
 > OpenACC contexts.  Say, I don't see why one couldn't e.g. try to mix
 > OpenMP host parallelization or tasking with OpenACC offloading,
 > and that affecting in weird way OpenMP semantics.

A colleague is adding code to allow for the detection of OpenACC contexts.
This change has yet make it to trunk. I need some guidance from you whether
I can leave the code as is and resolve the issue at stage3 time,
or remove the code and the associated function device_resident_()
and address the issue at stage 3.

OK to commit to trunk?

Thanks!
Jim