Time Complexity of handler.copy()

Hello all,
I was looking into copying the contents of an accessor into the host memory and I came across two methods namely copy and the update_host method. I was wondering about the time complexity of the copy method, because if the implementation is similar to std::copy(which has a time complexity of O(n)) with device interfacing then it may nullify the gain due to parallelism.
So, is handler.copy() the right way to update the host memory from an accessor ?.

Hi,
the handler::copy function is intended to be a straightforward copy, in OpenCL terms it should generally translate to a single clEnqueueReadBuffer or something similar, basically a memcpy. The copy functions are the preferred way of interacting with host memory as well, at least for high performance code, because there is less synchronization involved compared to a host accessor. The copy function is instead executed asynchronously just like any other SYCL command group.