Compressed Multisampling for Efficient Hardware Edge Antialiasing Philippe Beaudoin Pierre Poulin LIGUM D´ ep. I.R.O., Universit´ e de Montr´ eal Abstract Today’s hardware graphics accelerators incorporate tech- niques to antialias edges and minimize geometry-related sampling artifacts. Two such techniques, brute force su- persampling and multisampling, increase the sampling rate by rasterizing the triangles in a larger antialiasing buffer that is then filtered down to the size of the frame- buffer. The sampling rate is proportional to the num- ber of subsamples in the antialiasing buffer and, when no compression is used, to the memory it occupies. In turn, a larger antialiasing buffer implies an increase in bandwidth, one of the limiting resources for today’s ap- plications. In this paper we propose a mechanism to com- press the antialiasing buffer and limit the bandwidth re- quirements while maintaining higher sampling rates. The usual framebuffer-related functions of OpenGL are sup- ported: alpha blending, stenciling, color operations, and color masking. The technique is scalable, allowing for user-specified maximal and minimal sampling rates. The compression scheme includes a mechanism to nicely de- grade the quality when too much information would be required. A lower bound on the quality of the result- ing image is also available since the sampling rate will never be less than the user-specified minimal rate. The compression scheme is simple enough to be incorporated into standard hardware graphics accelerators. Software simulations show that, for a given bandwidth, our tech- nique offers improved visual results over multisampling schemes. Key words: graphics hardware, edge antialiasing, multi- sampling 1 Introduction Scan converting triangles is the core of today’s hardware graphics accelerators. This process, which is really the act of discretizing a continuous signal, is usually per- formed by sampling triangles at the center of each pixel of the screen. We know that sampling can give rise to ar- tifacts that are due to the presence of high frequencies in the continuous signal. In fact, the Nyquist formula tells us that such artifacts can occur as soon as the frequency of the input signal is greater than half the sampling rate. Signals with such high frequencies are often witnessed in textures and that is why various methods have been designed to try to minimize texture-related sampling ar- tifacts. The idea behind these techniques is to prefilter the input signal in order to remove all components above the Nyquist frequency. One such technique, trilinear mipmapping [17], is available on almost every hardware graphics accelerator. Even though textures are usually the most important source of high frequencies in the input signal, the geom- etry itself can cause artifacts that will not be handled by texture filtering. That is because the triangle edges create discontinuities in the input signal resulting in arbitrarily high frequencies. The presence of such frequencies can create various artifacts. One of these is the staircase pat- tern visible along polygon edges. This problem, often referred to as “jaggies”, is probably the most frequent ge- ometry sampling artifact. However, Moir´ e patterns can also occur and tend to become more important as the size of triangles decreases. Unfortunately, since the geometry-related input signal is not known in advance, prefiltering techniques such as mipmapping cannot be applied. In fact, the techniques available today in hardware do not filter out the compo- nents above the Nyquist frequency. Instead they increase the sampling rate, thus resulting in a higher Nyquist fre- quency leading to reduced artifacts. With the usual techniques, a higher sampling rate re- sults in an important increase in the internal bandwidth requirements of the graphics accelerator. Bandwith be- ing one of the most limiting resources in today’s appli- cations, high sampling rates can result in an important performance degradation. In this paper we present a technique to reduce the in- ternal bandwidth requirements for a given sampling rate through the use of a compressed antialiasing buffer. This technique supports the usual framebuffer-related func- tions of OpenGL such as alpha blending, stenciling, color operations, and color masking. The compression scheme includes a mechanism to nicely degrade the quality when