BRL-CAD has a regression test script called solids.sh that creates a bunch of primitives, renders an image of those primitives, and then compares that image to a reference image. On (most?) 64-bit platforms, the test is off by several RGB values for exactly 3 pixels.

This task involves figuring out why, exactly, this is occurring. It may be helpful to compare intermediate computation results from a 32-bit environment to see where the computations diverge, however slightly. Ultimately, the goal is to identify the cause and a recommended course of action to fix the divergence problem.

Code:

• regress/solids.sh
• src/librt
• src/liboptical