Many-Core Acceleration of Vertical Plane Radio Wave Propagation Prediction

Tianyi David Han, Cedomir Seguljac, Tarek S. Abdelrahman, Ivan Matosevic, Yves Lostanlen, and Grégory Gougeon


General-Purpose GPU Computing, Program Acceleration, Radio Wave Propagation Prediction


We accelerate a vertical plane radio wave propagation prediction application on GPUs. The application has abundant parallelism, but it exhibits uncoalesced memory accesses, divergent thread execution and contention for GPU register and memory resources – factors that make it challenging to accelerate on a GPU. We describe three acceleration strategies and optimizations to improve performance for each strategy. We evaluate the performance of the three strategies and their associated optimizations on GTX275 and GTX480 GPUs. We show that it is possible to obtain kernel speedups of up to 76× and overall application speedups of up to 48× over the sequential CPU implementation. However, the dynamic nature of data accesses in the algorithm makes the best strategy/optimizations dependent on input data and target GPU.

Important Links:

Go Back