Extending abstract GPU APIs to shared memory