On Tue, Mar 25, 2014 at 07:15:42PM -0700, Stéphane Marchesin wrote: > On Tue, Mar 25, 2014 at 12:27 AM, Chris Wilson <ch...@chris-wilson.co.uk> > wrote: > > On Mon, Mar 24, 2014 at 06:21:22PM -0700, Ben Widawsky wrote: > >> I am not clear why we've never enabled it by default for GEN7. Looking > >> at the git hostiry, it seems Rodrigo disabled it by default, and it's > >> never been turned on. Quite a few fixes have gone in over the past year, > >> and I think many of us are running this successfully. > >> > >> If there is some reason we know of why we don't enable this by default > >> on GEN7, then please ignore the patch, and forgive my laziness. > > > > Other than the major performance degredation due to our implementation, > > That sounds interesting, can you elaborate on the performance > degradation? I haven't noticed any, but of course I don't know how > it's supposed to behave...
The way we setup the FBC is that it causes it to be invalidated after every operation (not just batch, or upon flushing the framebuffer). The impact of this is that lightweight 3D rendering operations such as firefox are about 60% slower, game impact though is less than 10% (more often in the noise), and all but the most GPU bound of BLT operations are orders of magnitude slower. All of this is mitigable by disabling FBC invalidations except when we write to the scanout. https://bugs.freedesktop.org/show_bug.cgi?id=72023 -Chris -- Chris Wilson, Intel Open Source Technology Centre _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/intel-gfx
