D15514: [libbreezecommon] Use box blur instead of FFT approach

[Vlad Zagorodniy]

zzag created this revision.
zzag added reviewers: KWin, Plasma.
Herald added a project: Plasma.
Herald added a subscriber: plasma-devel.
zzag requested review of this revision.

REVISION SUMMARY
  Creation of shadows (especially for "Large" and "Very large" sizes) is
  a computation expensive task because we have to blur alpha channel of a
  shadow texture (which then will be tinted with desired shadow color).
  
  Currently, we use the following approach:
  
  - for blur radius less than 64, use naive 2-pass algorithm;
  - for blur radius greater than or equal to 64, use FFT.
  
  Even though the FFT approach is doing its the best, it still takes
  impresive amount of time to blur the alpha channel.
  
  What makes things even worse is that while we're blurring the alpha
  channel, we're blocking the main thread of KWin (decorations are not
  rendered in their own thread). This can result in frame drops (2 or 3
  frames, something like that).
  
  So, the only viable alternative is to use an approximation of the Gaussian
  blur. As such an approximation, I picked the box blur because it's quite
  simple, it's fast, and it takes small amount of iterations to have something
  similar to the true Gaussian blur.
  
  In general, there are no big differences between true gaussian shadows
  and approximated shadows:
  
  Before:
  F6264173: FFT shadows <https://phabricator.kde.org/F6264173>
  
  After:
  F6264175: after.png <https://phabricator.kde.org/F6264175>
  
  As a win, it takes much less time to generate decoration shadows.

REPOSITORY
  R31 Breeze

BRANCH
  zzag/box-shadow-helper-box-blur

REVISION DETAIL
  https://phabricator.kde.org/D15514

AFFECTED FILES
  CMakeLists.txt
  cmake/Modules/FindFFTW.cmake
  libbreezecommon/CMakeLists.txt
  libbreezecommon/breezeboxshadowhelper.cpp

To: zzag, #kwin, #plasma
Cc: plasma-devel, ragreen, Pitel, ZrenBot, lesliezhai, ali-mohamed, jensreuterberg, abetts, sebas, apol, mart
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/plasma-devel/attachments/20180914/2945ce5b/attachment.html>