I have C++ code with several CUDA kernels. Some operations would be easier to do with the Thrust library. Can one mix thrust calls with CUDA kernels?
More specifically, assume I have a thrust device vector declared like this:
thrust::device_vector<int> d_myVector;
Then assume I use thrust to perform some operation on that device vector. Can I later access that device vector directly from a CUDA kernel? Or, do I have to copy the thrust device vector back to the host, and then convert it to something the CUDA kernel can use, and then copy it back to the device?