[Openmp-dev] Target CUDA RTL --> CUDA error is: an illegal memory access was encountered

[Alexey Bataev via Openmp-dev]

And it should fail since you're mapping non trivially copyable type.

Best regards,
Alexey Bataev

14 янв. 2020 г., в 19:14, Itaru Kitayama <itaru.kitayama at gmail.com> написал(а):


This piece of C++ program execution fails at run time.

#include <stdio.h>

class AbsBase {
public:
        virtual int f() = 0;
};

class Derived : public AbsBase {
private:
        int *arr = new int[100];
public:
        int f() { return arr[0]; }
        void fillarray() {
                arr[0] = 1234;
        }
};

int main() {
        AbsBase *absBase = new Derived();
        static_cast<Derived*>(absBase)->fillarray();
#pragma omp target parallel for map(to: absBase[0:1])
        for (int i=0;i<10;i++) {
                Derived d1(*static_cast<Derived*>(absBase));
                printf("arr[0] is %d\n", d1.f());
        }
}

On Sun, Jan 12, 2020 at 1:45 PM Doerfert, Johannes <jdoerfert at anl.gov<mailto:jdoerfert at anl.gov>> wrote:
On 01/12, Itaru Kitayama wrote:
> Do you guys have a timeframe for that feature
> Implemented?

I do not and I don't know anyone who will drive this right now.

As mentioned before, you should be able to "move/copy-create" the
elements on the device in order to use virtual functions.


> On Sun, Jan 12, 2020 at 12:51 Doerfert, Johannes <jdoerfert at anl.gov<mailto:jdoerfert at anl.gov>> wrote:
>
> > On 01/11, Alexey Bataev via Openmp-dev wrote:
> > > Virtual functions are not supported.
> >
> > Not yet ;).
> >
> > We'll get it with 5.1 so we might actually implement it soon. Till then,
> > you have to create the object on the device you call the virtual
> > function.
> >

--

Johannes Doerfert
Researcher

Argonne National Laboratory
Lemont, IL 60439, USA

jdoerfert at anl.gov<mailto:jdoerfert at anl.gov>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/openmp-dev/attachments/20200115/f0aff86c/attachment.html>