Game Development Reference
In-Depth Information
Intel Gulftown limited to four cores. What I call a frame is the time DXUT gives
TGE in OnFrameMove () and OnD3D11RenderFrame (), which is highlighted in Figures
17.3 and 17.4 . ThetimewherenoGPAtasksappearistimespentinDXUT.This
time is not counted as frame time. DXUT is not a performance render engine and
is used here for simplicity. A real application would have a more optimal render
path for performance. GPA supports zooming into the timeline to better view the
data. Figures 17.3 a nd 17.4 are zoomed to show relevant events, so the horizontal
scale varies for each figure. Time is measured using the statistics and summary
windows on the left. The data in those windows is limited to the highlighted region
in GPA.
The first cut of the data looks promising ( Figures 17.3 and 17.4 ) . The animation
tasks are scheduling well. GPA tells us the wall time for the frame went from 5.23 ms
to 2.48 ms. Note that GPA times should be treated as approximate. The overhead
of logging data can affect absolute task execution time. Blank space on the thread
timeline indicates time not in a decorated task. This includes both driver time and
DXUT overhead. The total time spent in Animate Models increased from 4.5 ms
Figure 17.5. Task-based animation with memory allocations decorated.
 
Search WWH ::




Custom Search