13.07.2015 Views

CPU-GPU Algorithms for Triangular Surface Mesh Simplification

CPU-GPU Algorithms for Triangular Surface Mesh Simplification

CPU-GPU Algorithms for Triangular Surface Mesh Simplification

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>CPU</strong>-<strong>GPU</strong> <strong>Algorithms</strong> <strong>for</strong> <strong>Triangular</strong> <strong>Surface</strong> <strong>Mesh</strong> <strong>Simplification</strong> 9all of the time is spent on the <strong>CPU</strong>. The <strong>GPU</strong> memory usage <strong>for</strong> the testedsplits is shown in Table 3.(a) Bunny(b) GargoyleFig. 4. The time taken to simplify the bunny and gargoyle meshes <strong>for</strong> each <strong>CPU</strong>-<strong>GPU</strong> split using the <strong>CPU</strong>-<strong>GPU</strong> naïve marking algorithm. As the <strong>CPU</strong>-<strong>GPU</strong> splitincreases, the <strong>CPU</strong> workload increases, and the <strong>GPU</strong> workload decreases.% of <strong>GPU</strong> timemesh 95-5 90-10 85-15 80-20 75-25 70-30 65-35 60-40 55-45 50-50armadillo 9.2 14.5 19.6 24.6 29.4 34.1 38.6 43.1 47.5 51.7bunny 5.4 8.8 12.1 15.2 18.2 21.1 23.9 26.5 29.0 31.4gargoyle 10.4 15.8 21.1 26.3 31.3 36.1 40.8 45.4 49.9 54.3hand 9.4 14.8 20.0 25.1 29.9 34.7 39.3 43.9 48.4 52.7horse 5.0 8.0 10.9 13.6 16.2 18.7 21.1 23.3 25.4 27.4kitten 8.2 13.4 18.4 23.2 27.9 32.5 36.8 41.2 45.5 49.6mesh 45-55 40-60 35-65 30-70 25-75 20-80 15-85 10-90 5-95 0-100armadillo 55.8 59.8 63.6 67.4 70.9 74.3 77.5 80.6 83.4 86.0bunny 33.6 35.7 37.7 39.5 41.3 42.9 44.4 45.8 47.2 48.5gargoyle 58.6 62.7 66.7 70.5 74.2 77.7 81.0 84.2 87.2 89.1hand 56.9 61.0 64.8 68.7 72.3 75.8 79.0 82.2 84.2 87.8horse 29.2 30.9 32.5 33.9 35.3 36.5 37.6 38.6 39.4 40.1kitten 53.6 57.4 61.1 64.8 68.2 71.4 74.5 77.4 80.1 82.5Table 2. The percentage of time spent on the <strong>GPU</strong> <strong>for</strong> various <strong>CPU</strong>-<strong>GPU</strong> splitsusing the <strong>CPU</strong>-<strong>GPU</strong> naïve marking algorithm.As seen in Figure 4, the gargoyle mesh shows an increase in running time,whereas the bunny mesh exhibits a decrease in running time, with respectto increasing the workload of the <strong>GPU</strong>. This indicates that if the mesh is

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!