[Intel-gfx] [PATCH 2/2] drm/i915: Do not force non-caching copies for pwrite along shmem path

[Chris Wilson]

On Fri, Mar 07, 2014 at 09:39:44AM +0100, Daniel Vetter wrote:
> On Fri, Mar 07, 2014 at 08:30:37AM +0000, Chris Wilson wrote:
> > We don't always want to write into main memory with pwrite. The shmem
> > fast path in particular is used for memory that is cacheable - under
> > such circumstances forcing the cache eviction is undesirable. As we will
> > always flush the cache when targeting incoherent buffers, we can rely on
> > that second pass to apply the cache coherency rules and so benefit from
> > in-cache copies otherwise.
> > 
> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> 
> Do you have some numbers on this? Looks good otherwise.

Comparative figures with 1333MHz DDR3 on crw:
0: Time to snooped copy 16384 bytes x 131072:       19.520µs, 839.4MiB/s
1: Time to snooped copy 16384 bytes x 131072:       19.444µs, 842.6MiB/s
2: Time to snooped copy 16384 bytes x 131072:       18.808µs, 871.1MiB/s

Oddly enough though, it was the removing the page flag accesses that
made the most impact at a higher level.

It will take a while longer to complete checks on pnv.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre