Simple Models of the Impact of Overlap in Bucket Rendering
Milton Chen,
Gordon Stoll,
Homan Igehy,
Kekoa Proudfoot, and
Pat Hanrahan,
Stanford University
Appears in the Proceedings of the 1998 Eurographics/SIGGRAPH Workshop on Graphics Hardware
Abstract:
Bucket rendering is a technique in which the framebuffer is subdivided
into coherent regions that are rendered independently. The primary
benefits of this technique are the decrease in the size of the working set
of framebuffer memory required during rendering and the possibility of
processing multiple regions in parallel. The drawbacks of this technique
are the cost of computing the regions overlapped by each triangle and the
redundant work required in processing triangles multiple times when they
overlap multiple regions. Tile size is a critical parameter in bucket
rendering systems: smaller tile sizes allow smaller memory footprints and
better parallel load balancing but exacerbate the problem of redundant
computation.
In this paper, we use mathematical models, instrumentation, and
trace-driven simulation to evaluate the impact of overlap and conclude
that the problem of overlap is limited in scope. If triangles are small,
the overlap factor itself is also small. If triangles are large, overlap
is high but pixel work dominates the rendering time. In pipelined
rendering systems, the worst-case impact of overlap occurs when the area
of an input triangle is equal to the area for which the pipeline is
balanced-that is, the triangle-related computation time is equal to the
pixel-related computation time. Thus, as the current trends of
exponentially increasing triangle rate, slowly increasing screen
resolution, and increasing per-pixel computation continue to push this
balance point toward triangles with smaller area, bucket rendering systems
will be able to utilize smaller tiles efficiently.
Paper without images:
Paper with images:
Copyright Notice (from ACM):
Permission to make digital or hard copies of part or all of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. Copyrights for
components of this work owned by others than ACM must be honored.
Abstracting with credit is permitted. To copy otherwise, to republish, to
post on servers or to redistribute to lists, requires prior specific
permission and/or a fee.
[email protected]