Occlusion Culling
The HP fast-break program (page 8) enabled us to understand
representation of a more complex part is visible. Since a
customer requirements by analyzing what is important in
bounding box, or more generally a bounding volume, com-
OpenGL graphics today. As a result, we developed a technol-
pletely encloses the more complex part, it is possible to know
ogy called occlusion culling as an extension to OpenGL and
a priori that if the bounding volume is not visible then the
implemented it in the VISUALIZE fx graphics hardware.
complex part it encloses is not visible. Thus, the part that is
not visible does not need to be processed through the graphics
We found that the data sets many graphics workstation cus-
pipeline. The real benefit of occlusion culling comes when a
tomers are trying to visualize are very complex. These data
very complex part consisting of many vertices can be rejected,
sets have large numbers of small, complex components that
avoiding the expenditure of valuable time to process it.
are not always visible in the final images. For instance, when
rendering an airplane, all of the MCAD parts are present in the
For very complex data sets, such as the airplane mentioned
data set represented by potentially millions of polygons that
above or an automobile, a tremendous performance increase
must be processed. However, when this airplane is viewed
can be realized by using the HP occlusion culling technology.
from the outside only the outer surfaces are visible, not the fan
To date, several ISVs have begun using occlusion culling in
blades of the engine or the seats or bulkheads in the interior.
their applications and are seeing a 25 to 100 percent increase
in graphics performance. This magnitude of performance bene-
In a traditional 3D z-buffered graphics system, all polygons in
fit typically costs a customer several thousand dollars for the
a scene must be processed by the graphics pipeline because it
extra computational horsepower. HP includes this technology
is not known a priori which polygons will be visible and which
as standard in all VISUALIZE fx series graphics accelerators,
ones will be occluded (not visible). The notion of occlusion
giving even better price and performance results to our
culling, or removal of occluded objects, has been talked about
customers.
in the research community for several years. However, imple-
mentations tend to be in software where the performance is
The future of 3D graphics will continue toward visualizing ever
not at a satisfactory level.
more complex objects and environments. Occlusion culling
together with HP’s DirectModel technology (page 19) are
In the VISUALIZE fx series of graphics devices, HP developed
well positioned to be industry leaders in providing the technol-
a very efficient algorithm that tests objects for visibility.
ogy for 3D modeling applications.
An application program can very quickly use the occlusion
culling visibility test to determine if a simple bounding box
The primary responsibility of the interface chip is to sepa-
other pending operations. An example would be a texture
rate the streams of data that arrive from the host SPU into
cache download that is required to complete a primitive
three paths and arbitrate access among those paths.
that is currently being rasterized, a situation that would
lead to deadlock without the unbuffered path.
3D Path. Typically data from the host CPU looks very
much like the OpenGL API functions themselves. Data
2D Path. This path runs directly through the interface chip
following this first path is routed to the geometry chips.
to the texture and raster chips. The 2D path differs from
The geometry chips process the data and return the re-
the unbuffered path in the way its priority is handled. The
sults to the interface chip. These results are then sent on
interface chip manages priority among the three paths as
to the texture chips or directly to the raster chips if the
they all converge on the same set of wires between the
texture mapping subsystem is not installed. In either case
interface chip and the first texture chip. The unbuffered
the data is transmitted to and through all the texture and
path goes directly through the interface chip to those
raster chips in the system.
wires and has priority over the other two paths. Data
targeting the 2D path is held off until all preceding 3D
Unbuffered Path. This path passes data directly through
work in the geometry chip has been flushed through to
the interface chip to the texture and raster chips. This
the first texture chip.
provides a bypass method that allows traffic to get around
Article 4 • © 1998 Hewlett Packard Company
30
May 1998 • The Hewlett-Packard Journal

---

...

Este manual também é adequado para os modelos :
Computadores - HP Workstation x2000 (299.71 kb)
Computadores - HP Workstation xw8000 (299.71 kb)
Computadores - HP Workstation x4000 (299.71 kb)