Here's a patch series of the remaining prep work I have on hand for the ff_fragment_shader conversion to GLSL IR.
At this point, ff_fragment_shader is generating fewer Mesa IR instructions for glean texCombine than it was before, and the multitexturing openarena shader is generating 2/3 fewer instructions on 965, and there are no regressions on softpipe or 965 in piglit/ogles2conform (yeah, ogles2conform was relevant). However, there's a regression in openarena performance on my gen6 system of about 2%, apparently due to fragment shader performance (approximately constant relative loss with screen resolution), that we've been unable to track down. From debugging yesterday, it does appear to happen at the point of switching the ff_fragment_shader, not switching the backend on the 965 side. I haven't got around to testing other ff apps to see if this is generally the case, nor have I tested other drivers. If others are interested in this code, I'd encourage them to test the branch in my Mesa tree. _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev