Most direct manipulation tasks rely on precise placements of the cursor on the object of interest. Commonly, this requires the knowledge of distinct application-dependent geometry attributes of this object computed on the CPU. In this work, we present a simple yet general GPU-based framework for computing such attributes without depending on application-specific algorithms. In particular, it provides, for each pixel of the rendered model, elements of discrete differential geometry that require only the vertex data and 1-ring connectivity of each vertex stored in video memory. We validate our framework by showing how, only by providing correct mouse placements even when the geometry is modified on GPU, it can support the implementation of many direct manipulation tasks presented on the literature. Implementation results are presented.