[In testing, I was able to reproduce Federico's results for most, but
not all, of the test images. There might be some additional
platform-specific discrepancies that need ironed out, but this is a
solid step forward in any case.
Results for a quick run against just the image backend on my system:
--bryce]
Signed-off-by: Bryce Harrington <bryce@bryceharrington.org>
Bryce Harrington <bryce@bryceharrington.org>
Signed-off-by: Bryce Harrington <b.harrington@samsung.com>
(And include a git add missed from commit
ccd48b3464
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Tue Sep 30 14:06:21 2014 +0100
test: Remove more duplicated reference images
but were mostly invalidated by the rasteriser changes anyway).
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
When updating the quorem between cells, we would lose the overflow
increment as it was only applied locally and not preserved by updating
the quorem.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
If we have both a argb32 and rgb24 reference image that are identical,
we can replace them with a plain reference image. I also prefer to have
argb32/rgb24 versions of the reference images if rgb24 differs from the
plain reference.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reducing the number of passes has the usual change in the antialiasing
side-effects, as well as the boon of being faster (and theorectically more
accurate through reduced loss of dynamic range.)
On an i5-2520m:
swfdec-giant-steps-full 3240.43 -> 2651.36: 1.22x speedup
grads-heat-map 166.84 -> 136.79: 1.22x speedup
swfdec-giant-steps 940.19 -> 796.24: 1.18x speedup
ocitysmap 953.51 -> 831.96: 1.15x speedup
webkit-canvas-alpha 13924.01 -> 13115.70: 1.06x speedup
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Having discovered that I had recreated the references images with a
buggy rasteriser, we now need to recreate them with the correct reference
settings.
