Skip to main content
. Author manuscript; available in PMC: 2014 May 1.
Published in final edited form as: J Parallel Distrib Comput. 2013 Jan 16;73(5):10.1016/j.jpdc.2013.01.001. doi: 10.1016/j.jpdc.2013.01.001

Figure 4.

Figure 4

Percentage of computation time spent in each step. Left: The baseline Toeplitz strategy with the direct evaluations of Ql and FHy; Right: The fast Toeplitz strategy with Ql and FHy computed via the GPU-accelerated gridding. The total execution times are normalized between 0 and 1 to emphasize the relative time of each step. For comparison, Step 3 is identical for both methods. The performance benefits from gridding are clearly demonstrated by the reduction of the fraction of the time spent in Steps 1 and 2.