This patch-set implements a few vmx fast paths. These were the *most* low-hanging fruits I could find, which basically means I copied the sse2 implementation and modified the function calls for vmx :)
See individual commit messages for lowlevel-blt-bench numbers (upto 5x speedups). The patches also provide modest speedups in cairo benchmarks (using the trimmed traces): Speedups ======== t-firefox-canvas-swscroll 2182.58 -> 1507.09 : 1.45x speedup t-chromium-tabs 312.34 -> 223.55 : 1.40x speedup t-firefox-asteroids 507.44 -> 448.82 : 1.13x speedup t-firefox-paintball 871.85 -> 801.72 : 1.09x speedup Please review, Oded Oded Gabbay (3): vmx: implement fast path composite_over_8888_8888 vmx: implement fast path composite_add_8_8 vmx: implement fast path composite_add_8888_8888 pixman/pixman-vmx.c | 112 ++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 112 insertions(+) -- 2.4.3 _______________________________________________ Pixman mailing list Pixman@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/pixman