The 'ignore ARGB visuals' option from Kompmgr probably doesn't make
much sense, those (usually old) apps can be run with XLIB_SKIP_ARGB_VISUALS=1 set.
Also remove the alpha clear hack used for decorations - I think decorations
instead should be fixed not to "unintentionally" have alpha set.
svn path=/trunk/KDE/kdebase/workspace/; revision=689916
exposed a bug in our (ab)use of QX11Info::appDepth(). Make sure that
our buffers match the depth of the root window.
Thanks to Fredrik for the hint about DefaultDepth().
svn path=/trunk/KDE/kdebase/workspace/; revision=686998
for use in effects (and not only). Now a list of window quads (=window areas)
is created at the beginning of the paint pass, prepaint calls can modify
the split itself (i.e. divide it into more parts). The actual paint calls
can then modify these quads (i.e. transform their geometry). This will allow
better control of how the split is done and also allow painting e.g. only
the decoration differently. Still work in progress, but it works.
Also pass data to prepaint functions in a struct, as there is
already quite a number of them.
svn path=/trunk/KDE/kdebase/workspace/; revision=684893
is discarded. Windows that have previously been mapped and unmapped now
update properly when mapped again.
svn path=/trunk/KDE/kdebase/workspace/; revision=683977
only when the compositing pixmap changes (as in: a new one), not when its contents
change. This finally makes TFP faster than SHM (70% vs 45% of non-composited
performance) and should about match Compiz/Beryl. After this change also strict
binding may finally make sense.
svn path=/trunk/KDE/kdebase/workspace/; revision=676667
array when rendering.
- Rearrange some renderGLGeometry() parameters.
- Don't use opengl arrays and glDrawArrays() when rendering a small number of vertices since in this
case overhead of enabling/disabling the array is too big. Use immediate mode instead in such case.
svn path=/trunk/KDE/kdebase/workspace/; revision=675467
For 16bpp pixmaps, we use GL_RGB and GL_UNSIGNED_SHORT_5_6_5 with
glTex(Sub)Image2D, which means that SHM mode works with 16bpp X
servers.
Also, only create one XShmPixmap per damaged pixmap, instead of one
per (optimised) damage rectangle. Now we can use GL_RGB textures for
24bpp pixmaps, just like fallback mode does and tfp mode can.
svn path=/trunk/KDE/kdebase/workspace/; revision=659274
The render target is used to render the scene (or part of it) onto texture. This texture can then be used
e.g. to do some postprocessing.
Demo effect coming soon.
Move checkGLError() to kwineffects.*
Add GLTexture ctor which takes width and height and creates an empty texture (to be used with
GLRenderTarget to render onto it)
svn path=/branches/work/kwin_composite/; revision=655489
There's also a kwineffects library now, containing the effects API, which makes it possible to write
third-party effects.
API isn't complete yet and for now just two effects have been converted but I'm working on it :-)
svn path=/branches/work/kwin_composite/; revision=652226
Also add SceneOpenGL::Texture class, based on GLTexture. Optimised for SceneOpenGL::Window, this adds support for loading from an X Pixmap, as well as taking advantage of texture_from_pixmap/shm when available. Automatically detects what texture target should be used, so be sure to enableUnnormalizedTexCoords() before painting.
Make SceneOpenGL::Window, BoxSwitchEffect, and ExplosionEffect use the new classes.
svn path=/branches/work/kwin_composite/; revision=645125
It may fail (or "fail") if the window is not mapped or if the geometry
doesn't match, both of which may happen due to the asynchronous
nature of X.
svn path=/branches/work/kwin_composite/; revision=637741
Optionally, if SmoothScale is set to 2, trilinear filtering will be attempted instead of bilinear. This requires GL_ARB_texture_non_power_of_two, GL_EXT_framebuffer_object, and valid mipmaps.
svn path=/branches/work/kwin_composite/; revision=629453
Rivo Laks