On Tue, Apr 22, 2025 at 10:47:31AM -0400, Jason Merrill wrote:
> On 4/21/25 6:46 AM, Nathaniel Shead wrote:
> > I don't really know how OpenMP works, hopefully this makes sense.
> > Bootstrapped and regtested on x86_64-pc-linux-gnu, OK for trunk?
> > And for 15 (I guess after release)?
>
> This is OK with a FIXME; presumably if we want to support running static
> constructors on the offload target we will eventually want to support that
> in modules as well.
The previous fixes were just to make sure omp_target mangles slightly
differently from !omp_target. Though those names were non-exported names,
while supposedly the mangle_module_global_init returned names are global.
In the end the omp_target ctors should be nohost and !omp_target ctors
host, but using the same mangled name for both could cause problems if the
host needs to refer to the target copy somewhere.
> > --- /dev/null
> > +++ b/gcc/testsuite/g++.dg/modules/openmp-1.C
> > @@ -0,0 +1,9 @@
> > +// PR c++/119864
> > +// { dg-do assemble }
> > +// { dg-additional-options "-fmodules -fopenmp" }
> > +// { dg-require-effective-target "fopenmp" }
> > +
> > +export module M;
> > +
> > +int foo();
> > +int x = foo();
I don't see why there should be omp_target case here though.
There is no #pragma omp declare target enter (x) or something similar.
Jakub
