[Mesa-stable] [Mesa-dev] [PATCH 2/2] i965: Use BDW_MOCS_PTE for renderbuffers.

[Kenneth Graunke]

On Tuesday, September 30, 2014 10:33:42 AM Daniel Vetter wrote:
> On Tue, Sep 30, 2014 at 01:15:56AM -0700, Kenneth Graunke wrote:
> > Write-back caching cannot be used for buffers being scanned out by the
> > display engine; surfaces used for scan-out must be write-through or
> > uncached.  I originally chose WT for render targets because it works in
> > all cases.  However, we really want to use write-back caching where
> > possible, as it is more efficient.
> > 
> > Most renderbuffers are not used for scanout - off-screen FBOs certainly
> > are fine, and non-pageflipped backbuffers should be fine as well.  So
> > in most cases WB will work.  However, we don't know what will be used
> > for scan-out, so we instead simply use the PTE value specified by the
> > kernel, as it knows these things.
> > 
> > This matches our MOCS choice on Haswell.
> > 
> > Fixes performance regressions since commit ee4484be3dc827cf15bcf109f5
> > in a microbenchmark (spotted by Eero Tamminen).  Improves performance
> > in GLBenchmark 2.7/EgyptHD by 7.44362% +/- 0.496939% (n=55) on a
> > Broadwell GT2.
> > 
> > Signed-off-by: Kenneth Graunke <kenneth at whitecape.org>
> > Reported-by: Eero Tamminen <eero.t.tamminen at intel.com>
> > Cc: mesa-stable at lists.freedesktop.org
> > ---
> >  src/mesa/drivers/dri/i965/gen8_surface_state.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > Cc'd to stable because it's a pretty trivial change and provides a sizable
> > boost to performance on new hardware.
> 
> Both patches are Reviewed-by: Daniel Vetter <daniel.vetter at ffwll.ch>
> 
> Aside: Not using WT on display can lead to corruption (apparently bdw is
> fairly aggressive with writeback so hard to spot in reality), so imo
> definitely stable material.
> 
> With the hw display crc stuff we now support in the kernel/igt we could
> even write an automated testcase for these corruptions, but probably not
> worth the hassle.
> -Daniel

Well, we should have already been using WT when writing anything that hits the display.  The advantage of using the PTE entries' cache mode is that we should get WB for most (non-displayed) surfaces, but still get WT for anything displayed.

Thanks for the review!

--Ken
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.freedesktop.org/archives/mesa-stable/attachments/20140930/73d26e6d/attachment.sig>