Graphics Reference
In-Depth Information
Figure 2.11. In-scattering only rendered on Intel HD graphics 5000 in high performance
profile.
high-resolution shadow maps. Even when taking this time into account, the
optimized algorithm is always faster.
Table 2.3 shows performance speed-ups obtained by using 1D min/max bi-
nary trees for different quality profiles for both test platforms. The RM column
displays the performance improvement to the ray marching algorithm alone; the
RM+MMT column shows the speed-up of the two steps: ray marching and con-
structing 1D min/max tees. The third column shows the performance improve-
ment of the whole algorithm.
As Table 2.3 shows, using 1D min/max trees significantly reduces the execu-
tion times of the ray marching step. The performance gain is higher for high-
resolution shadow maps (high quality profiles) as longer ray sections in shadow
map space can be e ciently processed. This results in a speed-up of up to 8 . 6
×
on NVidia and up to 5 . 3 × on the Intel platform. Taking the tree construction
NVidia GTX 680
Intel HD graphics 5000
Profile
RM
RM+MMT
Total
RM
RM+MMT
Total
High qual
8.6
5.3
3.8
5.3
2.6
1.7
Balanced
3.6
2.5
1.5
4.0
1.8
1.3
High perf
2.7
1.8
1.3
1.6
1.1
1.1
Tab l e 2. 3. Performance gains obtained by using 1D min/max trees.
Search WWH ::




Custom Search