You are correct that we haven't keep compatible with latest mesa on the cl-gl sharing extension. The current implementation is such a hacky way that it highly depends on the mesa/dri driver's internal implementation. Just as you mentioned, the mesa/dri driver have changed a lot recently, and the cl-gl extension in beignet is broken with the upstream git master version.
You can see that the major focus of our team is not cl-gl sharing extension currently. Please feel free to do it at your own pace. And don't need to keep backward compatible with the old mesa repository. IMO, to keep it with latest upstream mesa is always a good thing so far. On Sat, May 3, 2014 at 2:16 AM, Steven Newbury <[email protected]>wrote: > Beignet has got a little left behind upstream Mesa changes so I've been > patching to compile against mesa/master for my gentoo-opencl overlay. > Eric Anholt has just landed a significant patch series dropping support > for regions in the i965 driver. Unfortunately, the Beignet DRI resource > sharing code depended on that functionality, so I've been working on > converting it to use buffer objects like the other previous users of the > region code. > > First an old issue that I've just worked around for now, I can't figure > out how to fix, but needs addressing: > > i965/intel_context.h has been merged with i965/brw_context.h which in > turn needs to be compiled with -std=c99/gnu99, which then breaks > beignet/src/cl_extensions.c, where this happens: > > > > cl_extensions.c:19:12: error: initializer element is not constant > > DECL_ALL_EXTENSIONS > > ^ > > cl_utils.h seriously clashes with headers pulled in by brw_context.h. > > There are a small number of easy fixes, for which I'll see if I can make > conditional on Mesa version and send to the list. > > As for the big one, the intel_region struct is gone, I've had a shot at > converting the code, I'll keep plugging away when I have time in a > couple of days, but if somebody else wants to give it a go don't let me > hold you back! :) A question though, should the code need to also > compile against older release versions of Mesa, or is it sufficient to > just target current upstream and have it maybe land in a branch? > > > _______________________________________________ > Beignet mailing list > [email protected] > http://lists.freedesktop.org/mailman/listinfo/beignet >
_______________________________________________ Beignet mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/beignet
