Primitive coalescing for faster multi-sampling

trinibwoy

Meh
Legend
Supporter
Is Nvidia preparing for the age of tessellation? This patent outlines a method for processing adjacent triangles in parallel to reduce the overhead of edge pixel derivative calculations. This becomes a bigger problem as triangles get smaller due to higher quality models and/or tessellation.

http://patft.uspto.gov/netacgi/nph-...TXT&S1=nvidia.ASNM.&OS=an/nvidia&RS=AN/nvidia

A graphics processing apparatus, comprising: a raster stage configured to identify neighboring primitives for concurrent multi-sample processing as coalesced groups each coalesced group comprising adjacent non-overlapping triangles generated from a common surface; and a pixel shader configured to concurrently shade individual coalesced groups with the shading including texture operations to determine the color of pixels within coalesced groups; wherein said pixel shader is configured to evaluate pixel attributes at each multi-sample sample location within a coalesced group to concurrently perform multi-sampling of sample locations within the coalesced group for pixels straddling a common edge between neighboring primitives of said coalesced group and generate a single color while retaining z-data for a plurality of multi-sample locations for each pixel straddling a common edge between neighboring primitives of said coalesced group to perform z-testing.
 
In a way, that sounds a little like what we did in the PowerVR PCX1/2 drivers where we'd try to merge 2 triangles into a quad, since PCX* supported N-sided shapes.
 
Back
Top