Suppose a data array d_a is residing inside the device(GPU) memory. If a programmer mistakenly calls cudaMemPrefetchAsync(d_a, size_t bytes, destination=gpu, 0), then will the data transfer happen ?
A unified memory prefetch operation is merely a hint, and a request to migrate the data to the specified target. If the data is already there, the prefetch request is a no-op.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.