KLAP Kernel Launch Aggregation and Promotion for Optimizing Dynamic Parallelis

KLAP: Kernel Launch Aggregation and Promotion for Optimizing Dynamic ParallelismIzzat El Hajj∗‡, Juan Gómez-Luna†, Cheng Li∗, Li-Wen Chang∗, Dejan Milojicic‡ and Wen-mei Hwu∗ ∗University of Illinois at Urbana-Champaign†Universidad de Córdoba ‡Hewlett-Packard Labselhajj2@illinois.edu, el1goluj@uco.es, {cli99, lchang20}@illinois.edu, dejan.milojicic@hpe.com, w-hwu@illinois.eduAbstract—Dynamic parallelism on GPUs simplifies the pro- gramming of many classes of applications that generate paral- le