OpenGL Insights

298 III Bending the Pipeline

using fully programmable vertex pulling makes me believe that AMD GPUs are less

dependent on efficient post-transform cache usage.

Based on the results, my verdict is that programmable vertex pulling has no over-

head compared to fixed-function vertex pulling in the following cases:

• When rendering nonindexed primitives (e.g., triangle strips).

• When rendering indexed primitives using fixed-function index handling.

While programmable index handling is also an option, when the meshes rely

heavily on the usage of the post-transform cache, the per formance of fully program-

mable vertex pulling can be prohibitive.

21.4 Application ...