Reorder the separable blur passes into XXX/YYY, with an image transpose on the last...
authorsenorblanco@chromium.org <senorblanco@chromium.org@2bbb7eff-a529-9590-31e7-b0007b416f81>
Thu, 15 Nov 2012 20:27:35 +0000 (20:27 +0000)
committersenorblanco@chromium.org <senorblanco@chromium.org@2bbb7eff-a529-9590-31e7-b0007b416f81>
Thu, 15 Nov 2012 20:27:35 +0000 (20:27 +0000)
commit908276b3969cf8f8eec28026363897134c0e54e0
tree55fb3b7689972532535358a0f5cb53e5bd3809d7
parent0561a3ccd24e46259d825adf5507c7ee9531b9c5
Reorder the separable blur passes into XXX/YYY, with an image transpose on the last pass of each group.  This results in continuguous memory reads in all passes, giving a 22% speedup on theverge.skp over the previous separable implementation, and a 30%-50% improvement over the existing implementation (depending on platform).

Review URL: https://codereview.appspot.com/6851053

git-svn-id: http://skia.googlecode.com/svn/trunk@6445 2bbb7eff-a529-9590-31e7-b0007b416f81
src/effects/SkBlurMask.cpp