On 09/04/2015 10:37 AM, Chris Wilson wrote: > On Fri, Sep 04, 2015 at 09:59:03AM -0700, Jesse Barnes wrote: >> We just need to pass in an address to execute and some flags, since we >> don't have to worry about buffer relocation or any of the other usual >> stuff. Returns a fence to be used for synchronization. > > There is no need for a flush+fence as part of the execbuf except for > when the client requests one (having it optional in the flags as a > syscall convenience). Just allow the client to inject fences > when it wants and even then it may just elect to do all fencing itself > from userspace. Kernel fencing is only interesting for interclient > operation. Yeah that's easy enough to add; are you thinking userspace would just be polling periodically on some ranges? And only requesting fences for specific batches for blocking purposes? Jesse _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/intel-gfx
