a) fixes the texture offset calculation
b) arranges he shadow pixmaps as border in the texture to avoid interpolation issues.
BUG: 280116
BUG: 282882
CCBUG: 291161
BUG: 293325
REVIEW: 103888
a) fixes the texture offset calculation
b) arranges he shadow pixmaps as border in the texture to avoid interpolation issues.
BUG: 280116
BUG: 282882
CCBUG: 291161
BUG: 293325
REVIEW: 103888
This patch reduces the number of QRegion and WindowQuadList operations
by drawing the opaque and translucent parts of the window within the
same bottom to top pass.
REVIEW: 103671
There seems to be a problem with nouveau GLES if you want to create an
EGLImageKHR more than once in a frame for the same pixmap. This patch
circumvents the problem in the way that it implements tfp the same way
as the mesa example in
mesa/demos/src/egl/opengles1/texture_from_pixmap.c does it. A nice
side effect of this is that it also avoids the overhead of recreating
the texture for every damaged window.
REVIEW: 103303
Additionally:
- hide the GLTexture implementation using dpointers
- drop the unused function SceneOpenGL::Texture::optimizeBindDamage()
- Texture::load now loads a new texture and does not update the existing one
REVIEW: 101999
Due to changes in build system we have always either OpenGL or OpenGL ES.
This allows to remove the KWIN_HAVE_OPENGL_COMPOSITING define. In the
effects the define is kept as KWIN_HAVE_OPENGL which can be used in
future to build also an XRender only effect system.
All the functionality of Overlay Window is moved to its own class
OverlayWindow. It is created and owned by class Scene, since almost
all function calls are called from this class.
REVIEW: 101866
Construct window quads which will end on the screen instead of
rendering the windows several times and using scissoring to
restrict to the area which will end on screen.
REVIEW: 101765
This commit merges the two signals clientClosed() and unmanagedClosed() to windowClosed() which
is now provided by Toplevel.
The approriate slots in effects.h and effects.cpp were merges as well, since they did the
same.
The direct method calls of the method windowClosed() in SceneOpenGL and SceneXRender were
removed and are now connected to the appropriate signal in windowAdded().
This commit just makes the declaration of windowClosed() in Class Scene be a Q_SLOT.
The inheriting classes SceneOpenGL and SceneXRender are updated as well.
The method windowGeometryShapeChanged() from the class Scene is now a slot. It is now connected to the signal geometryShapeChanged() which is sent from Toplevel instances Client and Unmanaged.
All direct method calls were deleted.
The method windowOpacityChanged is now a protected slot in class Scene. The implementations in the subclasses SceneOpenGL and SceneXRender are the same. The slots are connected to the singal opacityChanged() from Toplevel. The connection is done in the method windowAdded() in both SceneOpenGL and SceneXRender.
Copies the shadow parts into one image and creates a GLTexture
from the image, so that we can render the complete shadow with
just one texture and one painting pass.
Should remove most of the overhead involved when rendering the new Shadows.
As a side effect this should fix missing shadows with non-NPOT GPUs and
a rendering glitch reported with NVIDIA.
REVIEW: 101742
When using graphicssystem native the texture may be using
GL_TEXTURE_RECTANGLE_ARB instead of GL_TEXTURE_2D. This
needs to be honoured in order to make KWin work reliable
with e.g. R300 chipsets.
BUG: 274607
CCBUG: 276622
FIXED-IN: 4.7.0
The adjusted code for generating texture coordinates did not
take care of GL_TEXTURE_RECTANGLE_ARB textures causing the
generation of wrong texcoords.
BUG: 276622
FIXED-IN: 4.7.0
The uniforms textureWidth and textureHeight were only needed for
normal windows. For everything else it was just 1.0/1.0, that is
normalized.
The makeArrays method is changed to produce normalized texcoords
obsoleting the need for these uniforms. So two uniforms less, one
calculation in vertex shaders less and many many lines of code
removed.
At the same time makeArrays is also adjusted to take care of
yInverted of the texture, which is needed as we no longer can use
the enableUnnormalizedTexCoords which did the yInverted transformation.
REVIEW: 101646
by checking the graphicssystem at startup
also avoid pixmap and memory leaking on the xrender backend, validate some pointers
on deletion in SceneOpnGL and avoid attempts to render ::isNull pixmaps
With raster a QPixmap is no longer a XPixmap which fails all code
which assumes that an QPixmap is an XPixmap. Depending on were in
the codebase we either convert such pixmaps to images (OpenGL) or
create a XPixmap and use QPixmap::fromX11Pixmap to get a "real"
pixmap.
It is possible that there are more code pathes were we would need
a XPixmap. Currently tested is basic functionality of no-compositing,
XRender compositing, OpenGl/GLX and OpenGL ES/EGL compositing.
For OpenGL compositing raster might result in performance improvements,
for XRender it is possible that there are regressions when using raster.
By default KWin uses whatever is the default of the system, so we just
no longer enforce native.
Of course it is a bad idea to use graphicssystem OpenGL. As that
is broken anyways in Qt, we do not check for it.
Many thanks to Philipp Knechtges for bringing up the issue, convincing
me that we need it and providing most of the patch.
REVIEW: 101132
CCMAIL: Philipp.Knechtges@rwth-aachen.de
On NVIDIA it is possible that the actual rendering gets delayed to
after the deletion of the pixmap during the end of fullscreen effects.
This was causing freezes. By using glFlush before deleting the pixmaps
we can ensure that the pixmap is not needed anymore after the pixmaps
are deleted.
BUG: 261323
FIXED-IN: 4.6.3
The Shadow is clearly an aspect of the compositor. Therefore the
Shadow has to be owned and controlled by the Scene::Window.
Nevertheless Toplevel needs to know about the Shadow cause of reading
the property.
For a complete documentation of new functionality refer to:
http://community.kde.org/KWin/Shadow
The current implementation includes a new Shadow class and Toplevel
holds a pointer to an instance of this class. The Shadow class reads
the data from the X11 Property. There is one extended class located
in SceneOpenGL to render the shadow.
Compositor is adjusted to include the shadow region into the painting
passes.
Implementation for XRender still missing and Shadow needs to respond
to size changes of the Toplevel to update cached shadow region and
WindowQuads.
Second part of cleaning up the lib directory: the effects library
now lives in libkwineffects/ directory.
For existing effects nothing changes as the install path is unchanged.
The change obsoletes the lib/ directory.
As glplatform.h has not yet been exported I dared to export it and
adjust the places where it is used.
CCMAIL: kwin@kde.org
Our primary target is Texture From Pixmap and it is supported
by all important drivers nowadays. If a driver is not able to
support TFP using OpenGL at all is probably no good idea and
XRender is more suited.
When rendering opaque (RGB-only) windows the alpha ends up to be 0
with blending disabled. This breaks subsequent rendering steps which
require blenden (e.g. Lanczos). Therefore a uniform is used to ensure
that the alpha channel is set to 1.
The vertex buffer implementation uses the shader manager to decide
whether core painting should be used or not. Shader manager is only
used by shaders using vertex attributes instead of gl_Vertex etc.
A color can be specified to render the geometry of the VBO.
For legacy painting glColor is used, for shader a uniform is set.
In order to allow rendering without texcoords, it is possible to pass
a null pointer as texcoords.
Shader added to scene which just renders a colored geometry without texturing.
Adds a shader to render an untransformed scene. renderGLGeometry is not used any more and replaced by generated triangles stored in a GLVertexBuffer.
The GLVertexBuffer has a new attribute to decide whether a core profile compatible rendering has to be performed.
Currently windows and EffectFrames can make use of the new shader.
The shader contains a debug mode which colours all rendered fragments in green. It is currently enabled in scene_opengl.
Rendering transformed geometries (without shader) is currently broken.
differecens to patch atteched to 258971:
- removed debug statements
- fixed indention...
- NON vsync strategy does not rely on the estimation, but on the time passed since the last repaint trigger, allowing a precise framerate
CCBUG: 258971
svn path=/trunk/KDE/kdebase/workspace/; revision=1210445
this should improve v'syncing, maybe v'synced "smoothness"
remaining and exposed issue are "dirty textures" w/o damage events (see the requst description)
can be diminished by increasing MaxFPS above the fastest update (or shadowed below the slowest one)
CCBUG: 258971
svn path=/trunk/KDE/kdebase/workspace/; revision=1207577
The GLX implementation in the X server appears to have a hardcoded limit
to how many pixmaps can be bound to textures simultaneously when using
indirect rendering, which we can end up exceeding with the changes
introduced in r1182198.
BUG: 256359
FIXED-IN: 4.5.5
svn path=/trunk/KDE/kdebase/workspace/; revision=1203578
This worksaround a problem with the nouveau driver causing
the text frames to be incorrectly rendered. We need to keep
the QPixmap around as long as we have a texture created from
that texture.
This applies for the text and the unstyled effect frame. For
the frames generated from Plasma's FrameSvgs it is not required.
Addresses freedesktop.org bug 30286
svn path=/trunk/KDE/kdebase/workspace/; revision=1184458
Prefer the GL_TEXTURE_2D target if the framebuffer configuration indicates
that it's supported.
This fixes a performance problem with the r600g driver.
freedesktop bug 30483.
svn path=/trunk/KDE/kdebase/workspace/; revision=1183978
We don't need to do this every time we bind the texture to a GL context,
even with strict binding.
svn path=/trunk/KDE/kdebase/workspace/; revision=1182198
For Software Rasterizer we never ever want to have OpenGL compositing.
No matter how many config options the user sets it has to be disabled.
Though falling back to XRender makes sense in the case of software
rasterization.
svn path=/trunk/KDE/kdebase/workspace/; revision=1176787
into KWin's global namespace. Morever None already kind of clashes with X's None.
CCMAIL: kde@martin-graesslin.com
svn path=/trunk/KDE/kdebase/workspace/; revision=1170588
This allows an effect to fade between old and new text/icon. As an example it's added to CoverSwitch.
Currently only supported in OpenGL. XRender might be added, but I'm missing an idea for an effect to add it.
Most effects using EffectFrame require OpenGL anyway.
svn path=/trunk/KDE/kdebase/workspace/; revision=1160252
I missed this line in my previous change to disable the fallback which made the fallback happen nevertheless.
svn path=/trunk/KDE/kdebase/workspace/; revision=1157978
So it is more consistent (in KDE newspeak "elegant") with other selections and as a plus we get rid of all the custom rendering code in boxswitch.
svn path=/trunk/KDE/kdebase/workspace/; revision=1155051
Some effects (boxswitch and flipswitch) still need to be changed to not set the icon in each frame.
svn path=/trunk/KDE/kdebase/workspace/; revision=1152367
Rendering of the EffectFrame is moved into the scene as Scene::EffectFrame with a concrete implementation in SceneXrender and SceneOpenGL.
A factory method for an EffectFrame is added to the EffectsHandler, which is used by the effects.
Next step: pass the EffectFrame through all effects, so that effects can transform, blur, invert whatever it.
svn path=/trunk/KDE/kdebase/workspace/; revision=1151271
With Intel drivers currently the self check at KWin startup fails (no idea why) and KWin falls back to XRender. This is probably an explanation for all the complaints about slow kwin in 4.5 and the missing effects. So let's fail and make it possible to restart compositing via systemsettings and alt+shift+f12.
Nevertheless it would be nice to fix the the failing selfcheck...
svn path=/trunk/KDE/kdebase/workspace/; revision=1148315
to work correctly but as doing it there defeats the purpose of moving
the code to begin with there's no point in moving it.
BUG: 226049
svn path=/trunk/KDE/kdebase/workspace/; revision=1088054
functions and share the value with the KCM; Fallback to XRender
compositing if OpenGL fails to work correctly; Rearrange setting order
in options.h slightly and fix variable names
svn path=/trunk/KDE/kdebase/workspace/; revision=1079919
As this is a bigger commit I will wait with backporting to 4.3 for something about two or three weeks and will only backport if nobody yells.
BUG: 201780
svn path=/trunk/KDE/kdebase/workspace/; revision=1004096
This could be the solution to the performance regression when starting effects at activation change.
CCBUG: 191694
svn path=/trunk/KDE/kdebase/workspace/; revision=969215
at the same time (in other words, only when activating compositing using the kcm).
Currently selfcheck causes bad flicker (due to X mapping the overlay window
for too long?) which looks bad during KDE startup. With this patch, KDE startup
is without any flicker.
svn path=/trunk/KDE/kdebase/workspace/; revision=923842
(i.e. currently it allows to enable compositing even if self-check fails).
Not recommended of course, and it's be still nice to get self-check work
reliably.
CCBUG: 170085
svn path=/trunk/KDE/kdebase/workspace/; revision=860196
second test fails for some reason with BadAlloc in glXCreatePixmap().
Maybe nvidia bug, but could be also some hidden KWin bug (according
to GLX docs, glXCreatePixmap() can give BadAlloc only when it can't
allocate, which is nonsense here, but glXCreateWindow() says that
BadAlloc may mean calling it twice on the same window, so maybe
here it's twice with the same pixmap - I don't see where the bug
could be though).
svn path=/trunk/KDE/kdebase/workspace/; revision=857141
really works by simply trying to do it and test the result - create a small
testing window with known content, do the same with it like with normal
windows, grab the screen contents, compare with the original, doesn't match? -> fail.
It still would be nice to have something similar for performance.
svn path=/trunk/KDE/kdebase/workspace/; revision=854549
Therefore zScale, zTranslate and new RotationData are added to ScreenPaintData and WindowPaintData. So it is possible to define zTranslation and rotations without using OpenGL directly. The change only affects OpenGL compositing. XRender is not changed.
svn path=/trunk/KDE/kdebase/workspace/; revision=842018
actually needs to flush the output to the screen. Avoids windows
temporarily disappearing during KDE startup or similar visual glitches.
svn path=/trunk/KDE/kdebase/workspace/; revision=806387
geometry actually stays the same. Avoids large number of rebinds (with no
strict binding) with the launch feedback icon.
svn path=/trunk/KDE/kdebase/workspace/; revision=787948
Also slightly redo the #define's for effects, now it's:
- #ifdef KWIN_HAVE_COMPOSITING to check whether there's any compositing support at all
- #ifdef KWIN_HAVE_OPENGL_COMPOSITING to check for OpenGL-based compositing
- #ifdef KWIN_HAVE_XRENDER_COMPOSITING the same for XRender
CCMAIL: kwin@kde.org
svn path=/trunk/KDE/kdebase/workspace/; revision=749628
being v2+ (right now it says just GPL, which according to GPL itself
means any GPL). Decoration clients will come later.
CCMAIL: kwin@kde.org
svn path=/trunk/KDE/kdebase/workspace/; revision=742302
