OpenGL performance regression?

Hi,

Since few months I have worked on a data structure that is used in the fragment shader. To use my structure, I need a stack of uints to keep in mind a list of nodes to visit. Before the last driver update (I update my drivers once a month), to use a stack of 64 units instead 32 units cost about 2 more times. Now, with 352.41 on Linux 64bit, it’s about 3 or 4 time slower.

Memory management modification?

Did you ever figure out what was going on here? What card are you using?