Eric Enderton and I wrote a chapter for the upcoming book, GPU Gems II on High Quality Antialiased Rasterization. It describes a way of using the GPU to render your scene with supersampled tiles and how to properly downsample the tiles with arbitrarily wide filters implemented with fragment programs. This technique allows you to render images larger than the GPU limits with arbitrarily high supersample rates and it forms the core of the Gelato hider.
Along with the article, we are also releasing our cross-platform GPU and utility libraries. This C++ code provides a commonly used set of graphics programming objects including basic vector and matrix math, argument parsing, and 1D filters. The higher level GPU library provides a GLUT-like API for creating windows and pbuffers that is object-oriented and stateful to make debugging complex GPGPU code much easier.
Here is the README file included with the code.