Could you add a description to the PR? Why is it needed? Why should the set_device thing be independent of the GHEX backend? Why are changes to benchmark.cpp needed? I don’t like the main benchmark file to be polluted by GHEX-specific stuff. Actually also the threading stuff should be moved into the communication-backend somehow. If that’s currently not possible, we have to adapt the interfaces to make sure it is possible, but if we proceed like now, the code becomes a mess (actually it already became with the threading).
Could you add a description to the PR? Why is it needed? Why should the set_device thing be independent of the GHEX backend? Why are changes to benchmark.cpp needed? I don’t like the main benchmark file to be polluted by GHEX-specific stuff. Actually also the threading stuff should be moved into the communication-backend somehow. If that’s currently not possible, we have to adapt the interfaces to make sure it is possible, but if we proceed like now, the code becomes a mess (actually it already became with the threading).