Use special memory for batch copy_to function only for CPU batches
Description
Using pinned memory for temporary batches leads to severe performance degradation in GPU runs. However, using aligned memory for CPU batches improves performance (because of better vectorization). Thus, use the special memory only for CPU batches if the calling batch also used special memory.
Checklist
-
I have checked that my code follows the Octopus coding standards
Edited by Sebastian Ohlmann