On Sun, Nov 21, 2010 at 9:23 AM, Christian König <deathsim...@vodafone.de> wrote: > Hi everybody, > > just wanted to note that I got the first implementation of the XvMC iDCT > code working. The luma/chroma scaling and clamping still need some work, > but beside from that the pure matrix multiplication seems to work fine. >
Nice work! > At the moment it uses a quite inefficient two stage approach and on > large resolutions mplayer is complaining that my system is to slow to > play this (with only around 8% CPU in use). So the bottleneck is no > longer the CPU and starts to be the GPU calculation. > Using tiled surfaces may help if you are doing a lot of texture lookups. There is support in r600g if you are using the latest drm and mesa code. > While working on the iDCT code I found a couple of bugs in the command > checker, and also implemented support for signed normalized colour > buffers in r600g, but I think I will spam the different mailing list > with separate patches about this. Make sure you are using the latest drm patches. Dave and I fixed a number of bugs in the CS checker in 2.6.37. Also, make sure you have the latest mesa bits. I fixed the buffer alignment handling in r600g. Alex _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev