What do you mean by "core" code? As I mentioned in the other thread, there is some confusion when people say "Xinerama" because they're not clear on whether they mean the protocol extension or the implementation in the X.org server. There are three major implementations of the Xinerama protocol today:
- The Xinerama implementation in the X.org server, enabled by Option "Xinerama" in xorg.conf.
- The "TwinViewXineramaInfo" implementation in the NVIDIA X driver.
- The "fakerama" implementation in RandR 1.2.
№1 is incompatible with RandR (both 1.1 and 1.2). №2 is incompatible with Xinerama in that the TwinView screens will show up as one big screen according to the Xinerama protocol.
RandR 1.2 requires that any "CRTC" (basically a display device) be able to be placed anywhere on the desktop. This precludes being able to use multiple GPUs. A future revision of RandR will add GPU objects that can restrict the placement of CRTCs and allow a single RandR screen to span multiple GPUs.