DelayRepay: delayed execution for kernel fusion in Python
John Magnus Morton, Kuba Kaszyk, Lu Li, Jiawen Sun, Christophe Dubach, Michel Steuwer, Murray Cole, Michael O’Boyle · 2020
Python is a popular, dynamic language for data science and scientific computing. To ensure efficiency, significant numerical libraries are implemented in static native languages. However, performance suffers when switching between native and non-native code, especially if data has to be converted between native arrays and Python data structures. As GPU accelerators are increasingly used, this problem becomes particularly acute. Data and control has to be repeatedly transferred between the accelerator and the host.