On Fri, Feb 16, 2024 at 04:15:05PM +0100, Tobias Burnus wrote:
> I have no idea whether that would - nor whether that would be
> the way forward. - Thoughts?
Don't have time to dive through this now in detail, just want to point out
why we ignore DECL_VALUE_EXPR on the magic var during gimplification for
the parallel case - gimplify_var_or_parm_decl has
/* When within an OMP context, notice uses of variables. */
if (gimplify_omp_ctxp && omp_notice_variable (gimplify_omp_ctxp, decl, true))
return GS_ALL_DONE;
/* If the decl is an alias for another expression, substitute it now. */
if (DECL_HAS_VALUE_EXPR_P (decl))
{
*expr_p = unshare_expr (DECL_VALUE_EXPR (decl));
return GS_OK;
}
return GS_ALL_DONE;
and the trick is to make sure omp_notice_variable returns true if it is
undesirable to expand the DECL_VALUE_EXPR of decl at that point.
And whether omp_notice_variable returns true or false depends on
lang_hooks.decls.omp_disregard_value_expr langhook.
And that one has
if (VAR_P (decl)
&& DECL_HAS_VALUE_EXPR_P (decl)
&& DECL_ARTIFICIAL (decl)
&& DECL_LANG_SPECIFIC (decl)
&& DECL_OMP_PRIVATIZED_MEMBER (decl))
return true;
to deal with this.
Jakub
