FIG. 2.

Flowchart of ARCHER_RT GPU code. The left panel describes the work for CPU which is the host and the right panel describes the work for GPU which is the device. Device cannot interact with outer world directly and can only send/receive data to/from host. Host and device communicate via PCI-E bus. Host first reads data and sends them to device; after simulation, device sends results back to host. Pseudorandom number (PRN) kernel is used to initiate the PRN streams. Transport kernel handles the MC simulation of particles. Batch simulation is employed. After all batches are done, the host synchronizes all thread on the device and processes and outputs results.