Graphics Reference
In-Depth Information
Figure 1 3.1 6. The final result of our test scene rendering.
presents the time in milliseconds required to render a frame for a particular test. Thus, a
smaller number along the y-axis indicates better performance, while the scene complexity
increases to the right of the graph. Separate data plots are provided for each computer. The
test was run in both a single-threaded mode and a multi-threaded mode. 8
From this graph, we see a very clear performance improvement when using a mul-
tithreaded configuration, for both the dual-core and quad-core configurations. In addition,
there is a very linear change in frame time with the increasing number of objects in the
scene. This indicates that the application itself is CPU-bound—that is, rendering speed is
limited by how quickly the CPU can submit the needed API calls. By spreading the cost of
performing these API calls over multiple processor cores, we can effectively speed up our
overall frame rendering time. Figure 13.16 shows the final output of our test scene.
13.4 Conciusion
In this chapter we have developed an efficient dual-paraboloid-based technique to provide
dynamic environment mapping to objects within our scene. By using some of the new
8 The single-threaded mode renders directly with the immediate context, while the multi-threaded mode renders
one rendering pass into a command list, on a deferred context. The command lists are then submitted to the im-
mediate context for submission to the GPU.
Search WWH ::




Custom Search