and

Stream processors are becoming an affordable alternative to implement hardware assisted rendering techniques which were usually relegated to offline usage. We built a stream processing framework based on the Stream Programming Model concepts, selected the Photon Mapping algorithm and an NVIDIA GPU (Graphics Processing Unit) as a test case implementation of a Global Illumination technique. We defined a set of C++ classes to encapsulate the components (kernels and streams) of this new paradigm, using OpenGL and Cg language. Our application combines the Photon Splatting method and the BVH (Bounding Volumes Hierarchy) acceleration structure into a rendering pipeline relying almost entirely on the GPU. Finally, we evaluated its performance using a Cornell Box model