[Openmp-dev] Libomptarget fatal error 1: failure of target construct while offloading is mandatory

[Alexey Bataev via Openmp-dev]

1. If the is trivial, it is guaranteed to be offloaded correctly.
2. No, 5.0 does not add anything like this.

Best regards,
Alexey Bataev

27 сент. 2019 г., в 21:59, Itaru Kitayama <itaru.kitayama at gmail.com> написал(а):


Do you mean, depending the type it might be offloaded correctly? Is this situation improved once 5.0 API is supported in trunk?

On Sat, Sep 28, 2019 at 10:22 Alexey Bataev <a.bataev at hotmail.com<mailto:a.bataev at hotmail.com>> wrote:
Non trivial types are not guaranteed to be offloaded correctly.

Best regards,
Alexey Bataev

> 27 сент. 2019 г., в 21:17, Itaru Kitayama via Openmp-dev <openmp-dev at lists.llvm.org<mailto:openmp-dev at lists.llvm.org>> написал(а):
>
> 
> Hi,
> This simple C++ program:
>
> #include <iostream>
>
> class Abstract {
> public:
>         virtual int f() = 0;
> };
> class Derived : public Abstract {
>         public:
>         int f() { return 1; };
> };
>
> int main() {
>         Abstract *p[10];
>         for (int i=0;i<10;i++) {
>             p[i] = new Derived;
>         }
>         int sum=0;
> #pragma omp target parallel for map(p[:10]) reduction(+: sum)
>         for (int i=0;i<10;i++) {
>                 sum += p[i]->f();
>         }
>         std::cout << "sum " << sum << std::endl;
> }
>
> can be built with the offloading flags of Clang and runs fine on host, but on
> the device it crashes at runtime leaving the error message:
>
> ==60659== NVPROF is profiling process 60659, command: ./a.out
> Libomptarget fatal error 1: failure of target construct while offloading is mandatory
> ==60659== Profiling application: ./a.out
> ==60659== Warning: 5 records have invalid timestamps due to insufficient device buffer space. You can configure the buffer space using the option --device-buffer-size.
> ==60659== Warning: 4 records have invalid timestamps due to insufficient semaphore pool size. You can configure the pool size using the option --profiling-semaphore-pool-size.
> ==60659== Profiling result:
> No kernels were profiled.
>
> Does this just indicate the limitation of the current trunk Clang implementation?
>
> _______________________________________________
> Openmp-dev mailing list
> Openmp-dev at lists.llvm.org<mailto:Openmp-dev at lists.llvm.org>
> https://lists.llvm.org/cgi-bin/mailman/listinfo/openmp-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/openmp-dev/attachments/20190928/ef2b621e/attachment.html>