Adaptive Cpp CMake integration

[Helena2007]

I have completed the AdaptiveCpp CMake integration on the example project. When tested it runs on GPU. I have also completed the AdaptiveCpp CMake integration on the Geobox project of Geogram Library. The executable is generated and running. But when I tested it still ran on CPU instead of GPU. I have followed the same steps in both cases. When I analyzed the code I found that the bruteforce_nbody.cpp code in the example project had used sycl constructs in the cpp code. So, to do AdaptiveCpp CMake integration on any other project do we have to add sycl constructs in the cpp code? Is there any method to add sycl constructs to the cpp code other than manually?

[illuhad]

There is no compiler on the planet that can just take C++ code without any data parallel constructs and run it on GPU.

AdaptiveCpp can

• Run SYCL kernels on GPUs. This will require you to write data parallel kernels in SYCL for those components you wish to offload.
• Offload standard C++ parallel STL to GPUs, if you're using the std::execution::par_unseq policy. This will require you to formulate the parts of the program you wish to offload using C++ parallel STL algorithms, if the program is not already using them.