On Tue, Nov 01, 2016 at 08:07:36AM +0000, Chris Wilson wrote: > On Mon, Oct 31, 2016 at 09:48:15PM -0700, Jason Ekstrand wrote: > > + * By doing relocations on the CPU, we can tell the kernel that it doesn't > > + * need to bother. We want to do this because the surface state buffer is > > + * used by every command buffer so, if the kernel does the relocations, it > > + * will always be busy and the kernel will always stall. > > Userspace relocation + NO_RELOC processing should also work out to be > fastest of the available approaches today. If you want to try an ignore > stalls flag, it shouldn't take too long to write (I'll get that done > today) and then you can compare the overhead of the kernel having to copy > the relocations from userspace. (You also want the fast execbuf relocation > processing patches from a few years ago...)
https://cgit.freedesktop.org/~ickle/linux-2.6/commit/?h=prescheduler&id=767804fe12608d905ba0b4e9590706537caa9956 [compile tested] -Chris -- Chris Wilson, Intel Open Source Technology Centre _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
