Following on to Carl's update on our LCA talk, I've been trying to pull the 965 Render improvements out of the batchbuffer branch and onto master. Most of the changes in it are not actually dependent on batchbuffers or TTM. We can avoid reuploading programs without TTM. We can make bigger vertex buffers to reduce syncing without TTM. What the TTM branch (intel-batchbuffer) should be getting us is improved EXA pixmap migration performance, reduced memory fragmentation, more efficient buffer reuse, and possibly DRI2.
The current status of my work is at git://people.freedesktop.org/~anholt/xf86-video-intel on the 965-render-merge branch. It compiles, it runs, but the output is not quite what I hoped:
The code is mostly the same as what's on the branch, with just some changed buffer handling that should be all-or-nothing rendering correctness, not brokenness with colors. If I turn on xcompmgr, that desktop background goes thoroughly flourescent, and the windows partially translucent. Something very weird is going on, and I'm going to take a break and fix my 830/845 Render performance fix ("here test my patch, oh wait I didn't make it compile first") instead of bashing my head against this for now.
Welcome to the new LiveJournal
Some changes have been made to LiveJournal, and we hope you enjoy them! As we continue to improve the site on a daily basis to make your experience here better and faster, we would greatly appreciate your feedback about these changes. Please let us know what we can do for you!