Commit Graph

8 Commits

Author SHA1 Message Date
commit-bot@chromium.org
c45ece5f7f NEON fast path for box blur
Calculate 8 channels in parallel by using 16-bits to store each channel. Due to the limitation of VQRDMULH, (int16 * int16 * 2 + 0x8000) >> 16, the fast path can only support kernelSize < 128.
8 significant bits are kept at least in each stage, the final error should less-equal than 1.

Pre-fetching memory for X-direction read. In fact pre-fetching memory doesn't help much for Y direction read, since it is a waste to load a cache line for only read 8 bytes.(I left it there to keep the symmetry. pre-fetch is cheap :) )

bench data on Nexus 10
before:
running bench [640 480]      blur_image_filter_large_10.00_10.00   8888:  cmsecs =  25081.48
running bench [640 480]      blur_image_filter_small_10.00_10.00   8888:  cmsecs =  25038.04
running bench [640 480]        blur_image_filter_large_1.00_1.00   8888:  cmsecs =  25209.04
running bench [640 480]        blur_image_filter_small_1.00_1.00   8888:  cmsecs =  24928.01
running bench [640 480]        blur_image_filter_large_0.00_1.00   8888:  cmsecs =  17160.98
running bench [640 480]       blur_image_filter_large_0.00_10.00   8888:  cmsecs =  17924.11
running bench [640 480]        blur_image_filter_large_1.00_0.00   8888:  cmsecs =  14609.19
running bench [640 480]       blur_image_filter_large_10.00_0.00   8888:  cmsecs =  14625.91

after:
running bench [640 480]      blur_image_filter_large_10.00_10.00   8888:  cmsecs =  14848.42
running bench [640 480]      blur_image_filter_small_10.00_10.00   8888:  cmsecs =  16037.29
running bench [640 480]        blur_image_filter_large_1.00_1.00   8888:  cmsecs =  14819.55
running bench [640 480]        blur_image_filter_small_1.00_1.00   8888:  cmsecs =  14563.69
running bench [640 480]        blur_image_filter_large_0.00_1.00   8888:  cmsecs =  11905.34
running bench [640 480]       blur_image_filter_large_0.00_10.00   8888:  cmsecs =  11883.85
running bench [640 480]        blur_image_filter_large_1.00_0.00   8888:  cmsecs =   9576.51
running bench [640 480]       blur_image_filter_large_10.00_0.00   8888:  cmsecs =   9793.84

BUG=
R=senorblanco@chromium.org, mtklein@google.com, reed@google.com, kevin.petit@arm.com, kevin.petit.arm@gmail.com

Author: zheng.xu@arm.com

Review URL: https://codereview.chromium.org/105893003

git-svn-id: http://skia.googlecode.com/svn/trunk@13036 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-01-13 08:16:45 +00:00
mtklein@google.com
26c6d58274 text size is an SkScalar, Windows reminds me.
BUG=
R=jvanverth@google.com

Review URL: https://codereview.chromium.org/23967014

git-svn-id: http://skia.googlecode.com/svn/trunk@11297 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-09-16 19:05:44 +00:00
mtklein@google.com
cfa7ba7753 srand() + rand() -> SkRandom
rand() makes these two GMs thread-unsafe.  When run concurrently they can
interfere with each other.  Use SkRandom instead to guarantee independence.

BUG=skia:1590
R=jvanverth@google.com

Review URL: https://codereview.chromium.org/24105003

git-svn-id: http://skia.googlecode.com/svn/trunk@11295 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-09-16 18:19:30 +00:00
rmistry@google.com
d6176b0dca Result of running tools/sanitize_source_files.py (which was added in https://codereview.appspot.com/6465078/)
This CL is part II of IV (I broke down the 1280 files into 4 CLs).
Review URL: https://codereview.appspot.com/6474054

git-svn-id: http://skia.googlecode.com/svn/trunk@5263 2bbb7eff-a529-9590-31e7-b0007b416f81
2012-08-23 18:14:13 +00:00
senorblanco@chromium.org
1e95d715d0 Clean up the use of AutoScratchTexture in the gaussian blur and morphology
filters.  Instead of passing in AutoScratchTextures for temporaries, we allocate
them inside the function and detach() after rendering.  Since the functions now
return a ref()'ed texture, we no longer ref() the result in filter_texture().

Also, the imageblur gm was passing a paint with an image filter both to
saveLayer()/restore(), and to every text draw call.  Back when only restore()
was applying filters, this was fine, but since we're now applying filters on all
draw calls, this means we're double-blurring in this GM.

I've reverted the Mac baselines for the imageblur GM to their previous versions;
hopefully this will be correct.



git-svn-id: http://skia.googlecode.com/svn/trunk@4659 2bbb7eff-a529-9590-31e7-b0007b416f81
2012-07-18 19:52:53 +00:00
tomhudson@google.com
75589257c6 Fix miscellaneous compiler warnings from Visual Studio 2010.
Changes serialization path for MorphologyImageFilter, handling of Windows
HRESULTS; otherwise just tweaks tests.



git-svn-id: http://skia.googlecode.com/svn/trunk@3642 2bbb7eff-a529-9590-31e7-b0007b416f81
2012-04-10 17:42:21 +00:00
senorblanco@chromium.org
54e01b2ab9 This change makes SkImageFilter and SkBlurImageFilter flattenable/unflattenable.
The imageblur test also needed to stop calling getDeviceSize(), since it was
 returning 99999x99999 for the  SkPicture case.



git-svn-id: http://skia.googlecode.com/svn/trunk@2699 2bbb7eff-a529-9590-31e7-b0007b416f81
2011-11-16 18:20:47 +00:00
senorblanco@chromium.org
60014ca387 Implement Gaussian blurs for images. The caller creates an an
SkBlurImageFilter, sets it on an SkPaint, passes that paint to saveLayer(),
draws the primitives which are to be blurred, then calls restore(), which
applies the blur.  The blurs have separate sizes in the horizontal and vertical
direction.  This feature is GPU-only for now.

NB:  Due to the clipping change, there are slight pixel differences on the
blurs_gpu and shadows_gpu tests, so those will require rebaselining on all
platforms, as will some of the WebKit layout tests (TBD).


Review URL:  http://codereview.appspot.com/5322068/



git-svn-id: http://skia.googlecode.com/svn/trunk@2643 2bbb7eff-a529-9590-31e7-b0007b416f81
2011-11-09 16:05:58 +00:00