ViewCast Osprey PCI, Руководство пользователя

Страница: 85 / 126

Osprey PCI User Guide
ViewCast 79
Preview Pin to VMR9
Figure 57. Preview pin to VMR9
Video Mixing Renderer 9 is the newest video rendering method and the one on which Microsoft
supposedly will base its future development. The intent is to combine the functionality of the
Overlay Mixer plus Video Renderer in one module that takes advantage of the latest developments
in DirectShow. We are finding that at its present stage of development, with our hardware, VMR9
does not achieve the very high efficiency of YUY2-to-VMR7. Also, although VMR9 is supposed to
function as an overlay mixer for rendering captioning from the driver’s DShow CC or VBI pin, we
have never seen it function correctly.
Some Data Points
The following measurements are CPU percent on two machines – a fairly old P4, and a newer dual
Opteron 244. The video size is 640 x 480. The screen depth is 32 bits. The following abbreviations
are used:
YUY2
= the Osprey driver’s preview pin in YUY2 format
RGB15
= the Osprey driver’s preview pin in RGB15 format
RGB32
= the Osprey driver’s preview pin in RGB32 format
VR = old Video Renderer
VMR7 = Video Mixing Renderer 7
VMR9 = Video Mixing Renderer 9
AVI = AVI Decompressor
OVL = Overlay Mixer
The PostProc results are shown in two modes: with all post processing filters turned off, and with
the adaptive deinterlace filter turned on.
Generally these results show the following:

The great desirability of newer machines for video processing. This has to do with system
architecture more than raw CPU speed.

VMR7 is generally fastest.
If you don’t need the driver’s PostProcessing, then Direct Mode with
VMR7 is especially fast.

Results for specific pathways can be inconsistent across different machines. For example, on the
P4, YUV to VR is faster than RGB to VR; on the Opteron, RGB is faster.
In evaluating these benchmarks, bear in mind that all of them involve video rendering to the screen.
Depending on the exact pathway, video rendering can result in CPU utilization that is a lot higher
than for other capture scenarios. Specifically, writes to display adapter memory that are performed
by the CPU rather than with direct memory access (DMA) operations may be inordinately slow. If