Page 25 - 中国仿真学会通讯2020第1期
P. 25
The computational number of GCS tiles is 1 computational time and speedup for Line1 and
when the occlusion point locates in the first GCS Line2 are shown in Table 4 and Table 5
tile while is 2 when the occlusion point locates in respectively.
the second GCS tile or the computational result is
intervisible. The computing in the 2nd GCS tile is Table 3. The computer environment.
always proceeded after finishing the computing in Items Details
the 1st GCS tile no matter on CPU or GPU in this CPU Intel Core i7-7700K @ 4.20GHz
paper. RAM 24.0GB
OS Windows 10, 64bits
The computer environment is listed in Table GPU NVIDIA GeForce GTX 1080 Ti
3. Based on the parallel implementation CUDA v10.0
mentioned in previous section and different Compiler Visual Studio 2015
heights of viewpoints and targets, intervisibility
computing is executed to compare computational
time and speedup between host and device. The
Table 4. The computational time of host and device for Line1.
Computing on Host( CPU) Computing on Device( GPU)
heights( m) Speedup
RSGs Time( ms) RSGs time( ms) 2.06
4.83
500 3315 1.32 16595 0.64 5.42
8.18
1000 7750 3.09 16595 0.64 10.33
10.33
1500 8845 3.52 16595 0.65
2000 13406 5.32 16595 0.65
2500 16595×2 13.22 16595×2 1.28
3000 16595×2 13.22 16595×2 1.28
Note: The number of GCS tiles is 2 when the heights are 2500m and 3000m for Line1.
Table 5. The computational time of host and device for Line2.
Computing on Host( CPU) Computing on Device( GPU)
heights( m) Speedup
RSGs Time( ms) RSGs time( ms) 0.02
0.97
500 1 0.01 16595 0.64 3.06
7.77
1000 1557 0.62 16595 0.64 10.33
10.33
1500 4898 1.96 16595 0.64
2000 12617 5.05 16595 0.65
2500 16595×2 13.22 16595×2 1.28
3000 16595×2 13.22 16595×2 1.28
Note: The number of GCS tiles is 2 when the heights are 2500m and 3000m for Line2.
According to 500m and 1000m heights of viewpoints and targets for Line2, the
22