/***************************************************************** KWin - the KDE window manager This file is part of the KDE project. Copyright (C) 2006 Lubos Lunak You can Freely distribute this program under the GNU General Public License. See the file "COPYING" for the exact licensing terms. Based on glcompmgr code by Felix Bellaby. Using code from Compiz and Beryl. ******************************************************************/ /* This is the OpenGL-based compositing code. It is the primary and most powerful compositing backend. Sources and other compositing managers: ======================================= - http://opengl.org - documentation - OpenGL Redbook (http://opengl.org/documentation/red_book/) - GLX docs (http://opengl.org/documentation/specs/glx/glx1.4.pdf) - extensions docs (http://www.opengl.org/registry/) - glcompmgr - http://lists.freedesktop.org/archives/xorg/2006-July/017006.html , - http://www.mail-archive.com/compiz%40lists.freedesktop.org/msg00023.html - simple and easy to understand - works even without texture_from_pixmap extension - claims to support several different gfx cards - compile with something like "gcc -Wall glcompmgr-0.5.c `pkg-config --cflags --libs glib-2.0` -lGL -lXcomposite -lXdamage -L/usr/X11R6/lib" - compiz - git clone git://anongit.freedesktop.org/git/xorg/app/compiz - the ultimate - glxcompmgr - git clone git://anongit.freedesktop.org/git/xorg/app/glxcompgr - a rather old version of compiz, but also simpler and as such simpler to understand - beryl - the community fork of Compiz - http://beryl-project.org - svn co http://svn.beryl-project.org/trunk/ - libcm (metacity) - cvs -d :pserver:anonymous@anoncvs.gnome.org:/cvs/gnome co libcm - not much idea about it, the model differs a lot from KWin/Compiz/Beryl - does not seem to be very powerful or with that much development going on */ #include "scene_opengl.h" #include #include "utils.h" #include "client.h" #include "deleted.h" #include "effects.h" #include "glutils.h" #include #include #include namespace KWinInternal { //**************************************** // SceneOpenGL //**************************************** // the config used for windows GLXFBConfig SceneOpenGL::fbcdrawable; // GLX content GLXContext SceneOpenGL::ctxbuffer; GLXContext SceneOpenGL::ctxdrawable; // the destination drawable where the compositing is done GLXDrawable SceneOpenGL::glxbuffer; GLXDrawable SceneOpenGL::last_pixmap; bool SceneOpenGL::tfp_mode; // using glXBindTexImageEXT (texture_from_pixmap) bool SceneOpenGL::strict_binding; // intended for AIGLX bool SceneOpenGL::db; // destination drawable is double-buffered bool SceneOpenGL::copy_buffer_hack; // workaround for nvidia < 1.0-9xxx drivers bool SceneOpenGL::supports_npot_textures; bool SceneOpenGL::supports_saturation; bool SceneOpenGL::shm_mode; XShmSegmentInfo SceneOpenGL::shm; // detect OpenGL error (add to various places in code to pinpoint the place) static void checkGLError( const char* txt ) { GLenum err = glGetError(); if( err != GL_NO_ERROR ) kWarning() << "GL error (" << txt << "): 0x" << QString::number( err, 16 ) << endl; } // attributes for finding a double-buffered destination window config static const int buffer_db_attrs[] = { GLX_CONFIG_CAVEAT, GLX_NONE, GLX_DOUBLEBUFFER, True, GLX_RED_SIZE, 1, GLX_GREEN_SIZE, 1, GLX_BLUE_SIZE, 1, GLX_RENDER_TYPE, GLX_RGBA_BIT, None }; // attributes for finding a non-double-buffered destination pixmap config static const int buffer_nondb_attrs[] = { GLX_CONFIG_CAVEAT, GLX_NONE, GLX_DOUBLEBUFFER, False, GLX_RED_SIZE, 1, GLX_GREEN_SIZE, 1, GLX_BLUE_SIZE, 1, GLX_RENDER_TYPE, GLX_RGBA_BIT, None }; // attributes for finding config for windows const int drawable_attrs[] = { GLX_CONFIG_CAVEAT, GLX_NONE, GLX_DOUBLEBUFFER, False, GLX_DEPTH_SIZE, 0, GLX_RED_SIZE, 1, GLX_GREEN_SIZE, 1, GLX_BLUE_SIZE, 1, GLX_ALPHA_SIZE, 1, GLX_RENDER_TYPE, GLX_RGBA_BIT, None }; // attributes for finding config for windows when using tfp const int drawable_tfp_attrs[] = { GLX_CONFIG_CAVEAT, GLX_NONE, GLX_DOUBLEBUFFER, False, GLX_DEPTH_SIZE, 0, GLX_RED_SIZE, 1, GLX_GREEN_SIZE, 1, GLX_BLUE_SIZE, 1, GLX_ALPHA_SIZE, 1, GLX_RENDER_TYPE, GLX_RGBA_BIT, GLX_BIND_TO_TEXTURE_RGBA_EXT, True, // additional for tfp None }; SceneOpenGL::SceneOpenGL( Workspace* ws ) : Scene( ws ) { // TODO add checks where needed int dummy; if( !glXQueryExtension( display(), &dummy, &dummy )) return; initGLX(); // check for FBConfig support if( !hasGLXVersion( 1, 3 ) && !hasGLExtension( "GLX_SGIX_fbconfig" )) return; strict_binding = false; // not needed now selectMode(); initBuffer(); // create destination buffer int vis_buffer, vis_drawable; glXGetFBConfigAttrib( display(), fbcbuffer, GLX_VISUAL_ID, &vis_buffer ); glXGetFBConfigAttrib( display(), fbcdrawable, GLX_VISUAL_ID, &vis_drawable ); kDebug( 1212 ) << "Buffer visual: 0x" << QString::number( vis_buffer, 16 ) << ", drawable visual: 0x" << QString::number( vis_drawable, 16 ) << endl; initRenderingContext(); // Initialize OpenGL initGL(); if( db ) glDrawBuffer( GL_BACK ); has_waitSync = glXGetVideoSync ? true : false; // Check whether certain features are supported supports_npot_textures = hasGLExtension( "GL_ARB_texture_non_power_of_two" ) || hasGLVersion(2, 0); supports_saturation = ((hasGLExtension("GL_ARB_texture_env_crossbar") && hasGLExtension("GL_ARB_texture_env_dot3")) || hasGLVersion(1, 4)) && (glTextureUnitsCount >= 4) && glActiveTexture != NULL; // OpenGL scene setup glMatrixMode( GL_PROJECTION ); glLoadIdentity(); // swap top and bottom to have OpenGL coordinate system match X system glOrtho( 0, displayWidth(), displayHeight(), 0, 0, 65535 ); glMatrixMode( GL_MODELVIEW ); glLoadIdentity(); checkGLError( "Init" ); kDebug( 1212 ) << "DB:" << db << ", TFP:" << tfp_mode << ", SHM:" << shm_mode << ", Direct:" << bool( glXIsDirect( display(), ctxbuffer )) << endl; } SceneOpenGL::~SceneOpenGL() { for( QMap< Toplevel*, Window >::Iterator it = windows.begin(); it != windows.end(); ++it ) (*it).free(); // do cleanup after initBuffer() if( wspace->overlayWindow()) { if( hasGLXVersion( 1, 3 )) glXDestroyWindow( display(), glxbuffer ); XDestroyWindow( display(), buffer ); wspace->destroyOverlay(); } else { glXDestroyPixmap( display(), glxbuffer ); XFreeGC( display(), gcroot ); XFreePixmap( display(), buffer ); } if( shm_mode ) cleanupShm(); if( !tfp_mode && !shm_mode ) { if( last_pixmap != None ) glXDestroyPixmap( display(), last_pixmap ); glXDestroyContext( display(), ctxdrawable ); } glXDestroyContext( display(), ctxbuffer ); checkGLError( "Cleanup" ); } void SceneOpenGL::selectMode() { // select mode - try TFP first, then SHM, otherwise fallback mode shm_mode = false; tfp_mode = false; if( options->glMode == Options::GLTFP ) { if( initTfp()) tfp_mode = true; else if( initShm()) shm_mode = true; } else if( options->glMode == Options::GLSHM ) { if( initShm()) shm_mode = true; else if( initTfp()) tfp_mode = true; } if( !tfp_mode && !findConfig( drawable_attrs, &fbcdrawable )) assert( false ); // use copy buffer hack from glcompmgr (called COPY_BUFFER there) - nvidia drivers older than // 1.0-9xxx don't update pixmaps properly, so do a copy first copy_buffer_hack = !tfp_mode && !shm_mode; // TODO detect that it's nvidia < 1.0-9xxx driver } bool SceneOpenGL::initTfp() { if( glXBindTexImageEXT == NULL || glXReleaseTexImageEXT == NULL ) return false; if( !findConfig( drawable_tfp_attrs, &fbcdrawable )) return false; return true; } bool SceneOpenGL::initShm() { int major, minor; Bool pixmaps; if( !XShmQueryVersion( display(), &major, &minor, &pixmaps ) || !pixmaps ) return false; if( XShmPixmapFormat( display()) != ZPixmap ) return false; const int MAXSIZE = 4096 * 2048 * 4; // TODO check there are not larger windows // TODO check that bytes_per_line doesn't involve padding? shm.readOnly = False; shm.shmid = shmget( IPC_PRIVATE, MAXSIZE, IPC_CREAT | 0600 ); if( shm.shmid < 0 ) return false; shm.shmaddr = ( char* ) shmat( shm.shmid, NULL, 0 ); if( shm.shmaddr == ( void * ) -1 ) { shmctl( shm.shmid, IPC_RMID, 0 ); return false; } #ifdef __linux__ // mark as deleted to automatically free the memory in case // of a crash (but this doesn't work e.g. on Solaris ... oh well) shmctl( shm.shmid, IPC_RMID, 0 ); #endif KXErrorHandler errs; XShmAttach( display(), &shm ); if( errs.error( true )) { #ifndef __linux__ shmctl( shm.shmid, IPC_RMID, 0 ); #endif shmdt( shm.shmaddr ); return false; } return true; } void SceneOpenGL::cleanupShm() { shmdt( shm.shmaddr ); #ifndef __linux__ shmctl( shm.shmid, IPC_RMID, 0 ); #endif } void SceneOpenGL::initRenderingContext() { bool direct_rendering = options->glDirect; if( !tfp_mode && !shm_mode ) direct_rendering = false; // fallback doesn't seem to work with direct rendering KXErrorHandler errs; ctxbuffer = glXCreateNewContext( display(), fbcbuffer, GLX_RGBA_TYPE, NULL, direct_rendering ? GL_TRUE : GL_FALSE ); if( ctxbuffer == NULL || !glXMakeContextCurrent( display(), glxbuffer, glxbuffer, ctxbuffer ) || errs.error( true )) { // failed if( !direct_rendering ) assert( false ); glXDestroyContext( display(), ctxbuffer ); direct_rendering = false; // try again ctxbuffer = glXCreateNewContext( display(), fbcbuffer, GLX_RGBA_TYPE, NULL, GL_FALSE ); if( ctxbuffer == NULL || !glXMakeContextCurrent( display(), glxbuffer, glxbuffer, ctxbuffer )) assert( false ); } if( !tfp_mode && !shm_mode ) { ctxdrawable = glXCreateNewContext( display(), fbcdrawable, GLX_RGBA_TYPE, ctxbuffer, direct_rendering ? GL_TRUE : GL_FALSE ); } } // create destination buffer void SceneOpenGL::initBuffer() { if( findConfig( buffer_db_attrs, &fbcbuffer ) && wspace->createOverlay()) { // we have overlay, try to create double-buffered window in it XVisualInfo* visual = glXGetVisualFromFBConfig( display(), fbcbuffer ); XSetWindowAttributes attrs; attrs.colormap = XCreateColormap( display(), rootWindow(), visual->visual, AllocNone ); buffer = XCreateWindow( display(), wspace->overlayWindow(), 0, 0, displayWidth(), displayHeight(), 0, QX11Info::appDepth(), InputOutput, visual->visual, CWColormap, &attrs ); if( hasGLXVersion( 1, 3 )) glxbuffer = glXCreateWindow( display(), fbcbuffer, buffer, NULL ); else glxbuffer = buffer; wspace->setupOverlay( buffer ); db = true; XFree( visual ); } else if( findConfig( buffer_nondb_attrs, &fbcbuffer )) { // cannot get any double-buffered drawable, will double-buffer using a pixmap db = false; XGCValues gcattr; gcattr.subwindow_mode = IncludeInferiors; gcroot = XCreateGC( display(), rootWindow(), GCSubwindowMode, &gcattr ); buffer = XCreatePixmap( display(), rootWindow(), displayWidth(), displayHeight(), QX11Info::appDepth()); glxbuffer = glXCreatePixmap( display(), fbcbuffer, buffer, NULL ); } else assert( false ); } // print info about found configs static void debugFBConfig( GLXFBConfig* fbconfigs, int i, const int* attrs ) { int pos = 0; while( attrs[ pos ] != (int)None ) { int value; if( glXGetFBConfigAttrib( display(), fbconfigs[ i ], attrs[ pos ], &value ) == Success ) kDebug( 1212 ) << "ATTR: 0x" << QString::number( attrs[ pos ], 16 ) << ": 0x" << QString::number( attrs[ pos + 1 ], 16 ) << ": 0x" << QString::number( value, 16 ) << endl; else kDebug( 1212 ) << "ATTR FAIL: 0x" << QString::number( attrs[ pos ], 16 ) << endl; pos += 2; } } // find config matching the given attributes and possibly the given X visual bool SceneOpenGL::findConfig( const int* attrs, GLXFBConfig* config, VisualID visual ) { int cnt; GLXFBConfig* fbconfigs = glXChooseFBConfig( display(), DefaultScreen( display()), attrs, &cnt ); if( fbconfigs != NULL ) { if( visual == None ) { *config = fbconfigs[ 0 ]; kDebug( 1212 ) << "Found FBConfig" << endl; debugFBConfig( fbconfigs, 0, attrs ); XFree( fbconfigs ); return true; } else { for( int i = 0; i < cnt; ++i ) { int value; glXGetFBConfigAttrib( display(), fbconfigs[ i ], GLX_VISUAL_ID, &value ); if( value == (int)visual ) { kDebug( 1212 ) << "Found FBConfig" << endl; *config = fbconfigs[ i ]; debugFBConfig( fbconfigs, i, attrs ); XFree( fbconfigs ); return true; } } } } #if 0 // for debug fbconfigs = glXGetFBConfigs( display(), DefaultScreen( display()), &cnt ); for( int i = 0; i < cnt; ++i ) { kDebug( 1212 ) << "Listing FBConfig:" << i << endl; debugFBConfig( fbconfigs, i, attrs ); } if( fbconfigs != NULL ) XFree( fbconfigs ); #endif return false; } // the entry function for painting void SceneOpenGL::paint( QRegion damage, ToplevelList toplevels ) { foreach( Toplevel* c, toplevels ) { assert( windows.contains( c )); stacking_order.append( &windows[ c ] ); } grabXServer(); glXWaitX(); glPushMatrix(); int mask = 0; paintScreen( &mask, &damage ); // call generic implementation glPopMatrix(); ungrabXServer(); // ungrab before flushBuffer(), it may wait for vsync flushBuffer( mask, damage ); // do cleanup stacking_order.clear(); checkGLError( "PostPaint" ); } // wait for vblank signal before painting void SceneOpenGL::waitSync() { // NOTE that vsync has no effect with indirect rendering if( waitSyncAvailable() && options->glVSync ) { unsigned int sync; glFlush(); glXGetVideoSync( &sync ); glXWaitVideoSync( 2, ( sync + 1 ) % 2, &sync ); } } // actually paint to the screen (double-buffer swap or copy from pixmap buffer) void SceneOpenGL::flushBuffer( int mask, QRegion damage ) { if( mask & PAINT_SCREEN_REGION )// make sure not to go outside visible screen damage &= QRegion( 0, 0, displayWidth(), displayHeight()); if( db ) { if( mask & PAINT_SCREEN_REGION ) { waitSync(); if( glXCopySubBuffer ) { foreach( QRect r, damage.rects()) { // convert to OpenGL coordinates int y = displayHeight() - r.y() - r.height(); glXCopySubBuffer( display(), glxbuffer, r.x(), y, r.width(), r.height()); } } else { // no idea why glScissor() is used, but Compiz has it and it doesn't seem to hurt glEnable( GL_SCISSOR_TEST ); glDrawBuffer( GL_FRONT ); int xpos = 0; int ypos = 0; foreach( QRect r, damage.rects()) { // convert to OpenGL coordinates int y = displayHeight() - r.y() - r.height(); // Move raster position relatively using glBitmap() rather // than using glRasterPos2f() - the latter causes drawing // artefacts at the bottom screen edge with some gfx cards // glRasterPos2f( r.x(), r.y() + r.height()); glBitmap( 0, 0, 0, 0, r.x() - xpos, y - ypos, NULL ); xpos = r.x(); ypos = y; glScissor( r.x(), y, r.width(), r.height()); glCopyPixels( r.x(), y, r.width(), r.height(), GL_COLOR ); } glBitmap( 0, 0, 0, 0, -xpos, -ypos, NULL ); // move position back to 0,0 glDrawBuffer( GL_BACK ); glDisable( GL_SCISSOR_TEST ); } } else { waitSync(); glXSwapBuffers( display(), glxbuffer ); } glXWaitGL(); XFlush( display()); } else { glFlush(); glXWaitGL(); waitSync(); if( mask & PAINT_SCREEN_REGION ) foreach( QRect r, damage.rects()) XCopyArea( display(), buffer, rootWindow(), gcroot, r.x(), r.y(), r.width(), r.height(), r.x(), r.y()); else XCopyArea( display(), buffer, rootWindow(), gcroot, 0, 0, displayWidth(), displayHeight(), 0, 0 ); XFlush( display()); } } void SceneOpenGL::paintGenericScreen( int mask, ScreenPaintData data ) { if( mask & PAINT_SCREEN_TRANSFORMED ) { // apply screen transformations glPushMatrix(); glTranslatef( data.xTranslate, data.yTranslate, 0 ); glScalef( data.xScale, data.yScale, 1 ); } Scene::paintGenericScreen( mask, data ); if( mask & PAINT_SCREEN_TRANSFORMED ) glPopMatrix(); } void SceneOpenGL::paintBackground( QRegion region ) { if( region == infiniteRegion()) { glClearColor( 1, 1, 1, 1 ); // white glClear( GL_COLOR_BUFFER_BIT ); } else { glColor4f( 1, 1, 1, 1 ); // white glBegin( GL_QUADS ); foreach( QRect r, region.rects()) { glVertex2i( r.x(), r.y()); glVertex2i( r.x() + r.width(), r.y()); glVertex2i( r.x() + r.width(), r.y() + r.height()); glVertex2i( r.x(), r.y() + r.height()); } glEnd(); } } void SceneOpenGL::windowAdded( Toplevel* c ) { assert( !windows.contains( c )); windows[ c ] = Window( c ); } void SceneOpenGL::windowClosed( Toplevel* c, Deleted* deleted ) { assert( windows.contains( c )); if( deleted != NULL ) { // replace c with deleted Window& w = windows[ c ]; w.updateToplevel( deleted ); windows[ deleted ] = w; windows.remove( c ); } else { windows[ c ].free(); windows.remove( c ); } } void SceneOpenGL::windowDeleted( Deleted* c ) { assert( windows.contains( c )); windows[ c ].free(); windows.remove( c ); } void SceneOpenGL::windowGeometryShapeChanged( Toplevel* c ) { if( !windows.contains( c )) // this is ok, shape is not valid return; // by default Window& w = windows[ c ]; w.discardShape(); w.discardTexture(); } void SceneOpenGL::windowOpacityChanged( Toplevel* ) { #if 0 // not really needed, windows are painted on every repaint // and opacity is used when applying texture, not when // creating it if( !windows.contains( c )) // this is ok, texture is created return; // on demand Window& w = windows[ c ]; w.discardTexture(); #endif } //**************************************** // SceneOpenGL::Window //**************************************** SceneOpenGL::Window::Window( Toplevel* c ) : Scene::Window( c ) , texture( 0 ) , texture_target( 0 ) , texture_y_inverted( false ) , bound_glxpixmap( None ) , currentXResolution( -1 ) , currentYResolution( -1 ) , requestedXResolution( 0 ) , requestedYResolution( 0 ) , verticesDirty( false ) { } void SceneOpenGL::Window::free() { discardTexture(); } void SceneOpenGL::Window::requestVertexGrid(int maxquadsize) { requestedXResolution = (requestedXResolution <= 0) ? maxquadsize : qMin(maxquadsize, requestedXResolution); requestedYResolution = (requestedYResolution <= 0) ? maxquadsize : qMin(maxquadsize, requestedYResolution); } void SceneOpenGL::Window::createVertexGrid(int xres, int yres) { int oldcount = verticeslist.count(); verticeslist.clear(); foreach( QRect r, shape().rects()) { // First calculate number of columns/rows that this rect will be // divided into int cols = (xres <= 0) ? 1 : (int)ceilf( r.width() / (float)xres ); int rows = (yres <= 0) ? 1 : (int)ceilf( r.height() / (float)yres ); // Now calculate actual size of each cell int cellw = r.width() / cols; int cellh = r.height() / rows; int maxx = r.x() + r.width(); int maxy = r.y() + r.height(); for( int x1 = r.x(); x1 < maxx; x1 += cellw ) { int x2 = qMin(x1 + cellw, maxx); for( int y1 = r.y(); y1 < maxy; y1 += cellh ) { int y2 = qMin(y1 + cellh, maxy); // Add this quad to vertices' list verticeslist.append( Vertex( x1, y1 )); verticeslist.append( Vertex( x1, y2 )); verticeslist.append( Vertex( x2, y2 )); verticeslist.append( Vertex( x2, y1 )); } } } Client* c = qobject_cast(window()); kDebug( 1212 ) << k_funcinfo << "'" << (c ? c->caption() : "") << "': Resized vertex grid from " << oldcount/4 << " quads (minreso: " << currentXResolution << "x" << currentYResolution << ") to " << verticeslist.count()/4 << " quads (minreso: " << xres << "x" << yres << ")" << endl; currentXResolution = xres; currentYResolution = yres; verticesDirty = false; } void SceneOpenGL::Window::resetVertices() { // This assumes that texcoords of the vertices are unchanged. If they are, // we need to do this in some other way (or maybe the effects should then // clean things up themselves) for(int i = 0; i < verticeslist.count(); i++) { verticeslist[i].pos[0] = verticeslist[i].texcoord[0]; verticeslist[i].pos[1] = verticeslist[i].texcoord[1]; } verticesDirty = false; } void SceneOpenGL::Window::prepareVertices() { if( requestedXResolution != currentXResolution || requestedYResolution != currentYResolution ) createVertexGrid( requestedXResolution, requestedYResolution ); else if( verticesDirty ) resetVertices(); // Reset requests for the next painting requestedXResolution = 0; requestedYResolution = 0; } void SceneOpenGL::Window::prepareForPainting() { prepareVertices(); // We should also bind texture here so that effects could access it in the // paint pass } void SceneOpenGL::Window::findTextureTarget() { unsigned int target = 0; if( tfp_mode && glXQueryDrawable && bound_glxpixmap != None ) glXQueryDrawable( display(), bound_glxpixmap, GLX_TEXTURE_TARGET_EXT, &target ); else { if( supports_npot_textures || ( isPowerOfTwo( toplevel->width() ) && isPowerOfTwo( toplevel->height() ))) target = GLX_TEXTURE_2D_EXT; else target = GLX_TEXTURE_RECTANGLE_EXT; } switch( target ) { case GLX_TEXTURE_2D_EXT: texture_target = GL_TEXTURE_2D; texture_scale_x = 1.0f / toplevel->width(); texture_scale_y = 1.0f / toplevel->height(); break; case GLX_TEXTURE_RECTANGLE_EXT: texture_target = GL_TEXTURE_RECTANGLE_ARB; texture_scale_x = 1.0f; texture_scale_y = 1.0f; break; default: assert( false ); } } // Bind the window pixmap to an OpenGL texture. void SceneOpenGL::Window::bindTexture() { if( texture != 0 && toplevel->damage().isEmpty() && !options->glAlwaysRebind ) // interestingly with some gfx cards always rebinding is faster { // texture doesn't need updating, just bind it glBindTexture( texture_target, texture ); return; } // Get the pixmap with the window contents Pixmap pix = toplevel->windowPixmap(); // HACK // When a window uses ARGB visual and has a decoration, the decoration // does use ARGB visual. When converting such window to a texture // the alpha for the decoration part is broken for some reason (undefined?). // I wasn't lucky converting KWin to use ARGB visuals for decorations, // so instead simply set alpha in those parts to opaque. // Without alpha_clear_copy the setting is done directly in the window // pixmap, which seems to be ok, but let's not risk trouble right now. // TODO check if this isn't a performance problem and how it can be done better Client* c = dynamic_cast< Client* >( toplevel ); bool alpha_clear = c != NULL && c->hasAlpha() && !c->noBorder(); bool alpha_clear_copy = true; bool copy_buffer = (( alpha_clear && alpha_clear_copy ) || copy_buffer_hack ); if( copy_buffer ) { Pixmap p2 = XCreatePixmap( display(), pix, toplevel->width(), toplevel->height(), toplevel->depth()); GC gc = XCreateGC( display(), pix, 0, NULL ); XCopyArea( display(), pix, p2, gc, 0, 0, toplevel->width(), toplevel->height(), 0, 0 ); pix = p2; XFreeGC( display(), gc ); } if( alpha_clear ) { XGCValues gcv; gcv.foreground = 0xff000000; gcv.plane_mask = 0xff000000; GC gc = XCreateGC( display(), pix, GCPlaneMask | GCForeground, &gcv ); XFillRectangle( display(), pix, gc, 0, 0, c->width(), c->clientPos().y()); XFillRectangle( display(), pix, gc, 0, 0, c->clientPos().x(), c->height()); int tw = c->clientPos().x() + c->clientSize().width(); int th = c->clientPos().y() + c->clientSize().height(); XFillRectangle( display(), pix, gc, 0, th, c->width(), c->height() - th ); XFillRectangle( display(), pix, gc, tw, 0, c->width() - tw, c->height()); XFreeGC( display(), gc ); } if( copy_buffer || alpha_clear ) glXWaitX(); if( shm_mode ) { // non-tfp case, copy pixmap contents to a texture findTextureTarget(); if( texture == None ) { glGenTextures( 1, &texture ); glBindTexture( texture_target, texture ); texture_y_inverted = false; glTexImage2D( texture_target, 0, GL_RGBA, toplevel->width(), toplevel->height(), 0, GL_BGRA, GL_UNSIGNED_BYTE, NULL ); // TODO hasAlpha() ? // glCopyTexImage2D( texture_target, 0, // toplevel->hasAlpha() ? GL_RGBA : GL_RGB, // 0, 0, toplevel->width(), toplevel->height(), 0 ); } else glBindTexture( texture_target, texture ); if( !toplevel->damage().isEmpty()) { XGCValues xgcv; xgcv.graphics_exposures = False; xgcv.subwindow_mode = IncludeInferiors; GC gc = XCreateGC( display(), pix, GCGraphicsExposures | GCSubwindowMode, &xgcv ); QRegion damage = optimizeBindDamage( toplevel->damage(), 100 * 100 ); foreach( QRect r, damage.rects()) { // TODO for small areas it might be faster to not use SHM to avoid the XSync() Pixmap p = XShmCreatePixmap( display(), rootWindow(), shm.shmaddr, &shm, r.width(), r.height(), toplevel->depth()); XCopyArea( display(), pix, p, gc, r.x(), r.y(), r.width(), r.height(), 0, 0 ); XSync( display(), False ); glXWaitX(); glTexSubImage2D( texture_target, 0, r.x(), r.y(), r.width(), r.height(), GL_BGRA, GL_UNSIGNED_BYTE, shm.shmaddr ); glXWaitGL(); XFreePixmap( display(), p ); } XFreeGC( display(), gc ); } texture_y_inverted = true; toplevel->resetDamage( toplevel->rect()); } else if( tfp_mode ) { // tfp mode, simply bind the pixmap to texture if( texture == None ) glGenTextures( 1, &texture ); if( bound_glxpixmap != None && !strict_binding ) // release old if needed { glXReleaseTexImageEXT( display(), bound_glxpixmap, GLX_FRONT_LEFT_EXT ); glXDestroyGLXPixmap( display(), bound_glxpixmap ); } static const int attrs[] = { GLX_TEXTURE_FORMAT_EXT, GLX_TEXTURE_FORMAT_RGBA_EXT, None }; // the GLXPixmap will reference the X pixmap, so it will be freed automatically // when no longer needed bound_glxpixmap = glXCreatePixmap( display(), fbcdrawable, pix, attrs ); findTextureTarget(); int value; glXGetFBConfigAttrib( display(), fbcdrawable, GLX_Y_INVERTED_EXT, &value ); texture_y_inverted = value ? true : false; glBindTexture( texture_target, texture ); if( !strict_binding ) glXBindTexImageEXT( display(), bound_glxpixmap, GLX_FRONT_LEFT_EXT, NULL ); toplevel->resetDamage( toplevel->rect()); } else { // non-tfp case, copy pixmap contents to a texture findTextureTarget(); GLXDrawable pixmap = glXCreatePixmap( display(), fbcdrawable, pix, NULL ); glXMakeContextCurrent( display(), pixmap, pixmap, ctxdrawable ); if( last_pixmap != None ) glXDestroyPixmap( display(), last_pixmap ); // workaround for ATI - it leaks/crashes when the pixmap is destroyed immediately // here (http://lists.kde.org/?l=kwin&m=116353772208535&w=2) last_pixmap = pixmap; glReadBuffer( GL_FRONT ); glDrawBuffer( GL_FRONT ); if( texture == None ) { glGenTextures( 1, &texture ); glBindTexture( texture_target, texture ); texture_y_inverted = false; glCopyTexImage2D( texture_target, 0, toplevel->hasAlpha() ? GL_RGBA : GL_RGB, 0, 0, toplevel->width(), toplevel->height(), 0 ); } else { glBindTexture( texture_target, texture ); QRegion damage = optimizeBindDamage( toplevel->damage(), 30 * 30 ); foreach( QRect r, damage.rects()) { // convert to OpenGL coordinates (this is mapping // the pixmap to a texture, this is not affected // by using glOrtho() for the OpenGL scene) int gly = toplevel->height() - r.y() - r.height(); glCopyTexSubImage2D( texture_target, 0, r.x(), gly, r.x(), gly, r.width(), r.height()); } } glXWaitGL(); if( db ) glDrawBuffer( GL_BACK ); glXMakeContextCurrent( display(), glxbuffer, glxbuffer, ctxbuffer ); glBindTexture( texture_target, texture ); texture_y_inverted = false; toplevel->resetDamage( toplevel->rect()); } // if using copy_buffer, the pixmap is no longer needed (either referenced // by GLXPixmap in the tfp case or not needed at all in non-tfp cases) if( copy_buffer ) XFreePixmap( display(), pix ); } QRegion SceneOpenGL::Window::optimizeBindDamage( const QRegion& reg, int limit ) { if( reg.rects().count() <= 1 ) return reg; // try to reduce the number of rects, as especially with SHM mode every rect // causes X roundtrip, even for very small areas - so, when the size difference // between all the areas and the bounding rectangle is small, simply use // only the bounding rectangle int size = 0; foreach( QRect r, reg.rects()) size += r.width() * r.height(); if( reg.boundingRect().width() * reg.boundingRect().height() - size < limit ) return reg.boundingRect(); return reg; } void SceneOpenGL::Window::enableTexture() { glEnable( texture_target ); glBindTexture( texture_target, texture ); if( tfp_mode && strict_binding ) { assert( bound_glxpixmap != None ); glXBindTexImageEXT( display(), bound_glxpixmap, GLX_FRONT_LEFT_EXT, NULL ); } if( options->smoothScale != 0 ) // default to yes { glTexParameteri( texture_target, GL_TEXTURE_MIN_FILTER, GL_LINEAR ); glTexParameteri( texture_target, GL_TEXTURE_MAG_FILTER, GL_LINEAR ); } else { glTexParameteri( texture_target, GL_TEXTURE_MIN_FILTER, GL_NEAREST ); glTexParameteri( texture_target, GL_TEXTURE_MAG_FILTER, GL_NEAREST ); } } void SceneOpenGL::Window::disableTexture() { if( tfp_mode && strict_binding ) { assert( bound_glxpixmap != None ); glBindTexture( texture_target, texture ); glXReleaseTexImageEXT( display(), bound_glxpixmap, GLX_FRONT_LEFT_EXT ); } glBindTexture( texture_target, 0 ); glDisable( texture_target ); } void SceneOpenGL::Window::discardTexture() { if( texture != 0 ) { if( tfp_mode ) { if( !strict_binding ) glXReleaseTexImageEXT( display(), bound_glxpixmap, GLX_FRONT_LEFT_EXT ); glXDestroyGLXPixmap( display(), bound_glxpixmap ); bound_glxpixmap = None; } glDeleteTextures( 1, &texture ); } texture = 0; } // paint the window void SceneOpenGL::Window::performPaint( int mask, QRegion region, WindowPaintData data ) { // check if there is something to paint (e.g. don't paint if the window // is only opaque and only PAINT_WINDOW_TRANSLUCENT is requested) bool opaque = isOpaque() && data.opacity == 1.0; if( mask & ( PAINT_WINDOW_OPAQUE | PAINT_WINDOW_TRANSLUCENT )) {} else if( mask & PAINT_WINDOW_OPAQUE ) { if( !opaque ) return; } else if( mask & PAINT_WINDOW_TRANSLUCENT ) { if( opaque ) return; } // paint only requested areas if( region != infiniteRegion()) // avoid integer overflow region.translate( -x(), -y()); region &= shape(); if( region.isEmpty()) return; bindTexture(); glPushMatrix(); // do required transformations int x = toplevel->x(); int y = toplevel->y(); if( mask & PAINT_WINDOW_TRANSFORMED ) { x += data.xTranslate; y += data.yTranslate; } glTranslatef( x, y, 0 ); if(( mask & PAINT_WINDOW_TRANSFORMED ) && ( data.xScale != 1 || data.yScale != 1 )) glScalef( data.xScale, data.yScale, 1 ); // setup blending of transparent windows glPushAttrib( GL_ENABLE_BIT ); if( !opaque ) { glEnable( GL_BLEND ); glBlendFunc( GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA ); } if( data.saturation != 1.0 && supports_saturation ) { // First we need to get the color from [0; 1] range to [0.5; 1] range glActiveTexture( GL_TEXTURE0 ); glTexEnvi( GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_COMBINE ); glTexEnvi( GL_TEXTURE_ENV, GL_COMBINE_RGB, GL_INTERPOLATE ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE0_RGB, GL_TEXTURE ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND0_RGB, GL_SRC_COLOR ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE1_RGB, GL_CONSTANT ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND1_RGB, GL_SRC_COLOR ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE2_RGB, GL_CONSTANT ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND2_RGB, GL_SRC_ALPHA ); const float scale_constant[] = { 1.0, 1.0, 1.0, 0.5}; glTexEnvfv( GL_TEXTURE_ENV, GL_TEXTURE_ENV_COLOR, scale_constant ); enableTexture(); // Then we take dot product of the result of previous pass and // saturation_constant. This gives us completely unsaturated // (greyscale) image // Note that both operands have to be in range [0.5; 1] since opengl // automatically substracts 0.5 from them glActiveTexture( GL_TEXTURE1 ); float saturation_constant[] = { 0.5 + 0.5*0.30, 0.5 + 0.5*0.59, 0.5 + 0.5*0.11, data.saturation }; glTexEnvi( GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_COMBINE ); glTexEnvi( GL_TEXTURE_ENV, GL_COMBINE_RGB, GL_DOT3_RGB ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE0_RGB, GL_PREVIOUS ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND0_RGB, GL_SRC_COLOR ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE1_RGB, GL_CONSTANT ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND1_RGB, GL_SRC_COLOR ); glTexEnvfv( GL_TEXTURE_ENV, GL_TEXTURE_ENV_COLOR, saturation_constant ); enableTexture(); // Finally we need to interpolate between the original image and the // greyscale image to get wanted level of saturation glActiveTexture( GL_TEXTURE2 ); glTexEnvi( GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_COMBINE ); glTexEnvi( GL_TEXTURE_ENV, GL_COMBINE_RGB, GL_INTERPOLATE ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE0_RGB, GL_TEXTURE0 ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND0_RGB, GL_SRC_COLOR ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE1_RGB, GL_PREVIOUS ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND1_RGB, GL_SRC_COLOR ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE2_RGB, GL_CONSTANT ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND2_RGB, GL_SRC_ALPHA ); glTexEnvfv( GL_TEXTURE_ENV, GL_TEXTURE_ENV_COLOR, saturation_constant ); // Also replace alpha by primary color's alpha here glTexEnvi( GL_TEXTURE_ENV, GL_COMBINE_ALPHA, GL_REPLACE ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE0_ALPHA, GL_PRIMARY_COLOR ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND0_ALPHA, GL_SRC_ALPHA ); // And make primary color contain the wanted opacity glColor4f( data.opacity, data.opacity, data.opacity, data.opacity ); enableTexture(); if( toplevel->hasAlpha() || data.brightness != 1.0f ) { glActiveTexture( GL_TEXTURE3 ); glTexEnvi( GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_COMBINE ); glTexEnvi( GL_TEXTURE_ENV, GL_COMBINE_RGB, GL_MODULATE ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE0_RGB, GL_PREVIOUS ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND0_RGB, GL_SRC_COLOR ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE1_RGB, GL_PRIMARY_COLOR ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND1_RGB, GL_SRC_COLOR ); if( toplevel->hasAlpha() ) { // The color has to be multiplied by both opacity and brightness float opacityByBrightness = data.opacity * data.brightness; glColor4f( opacityByBrightness, opacityByBrightness, opacityByBrightness, data.opacity ); // Also multiply original texture's alpha by our opacity glTexEnvi( GL_TEXTURE_ENV, GL_COMBINE_ALPHA, GL_MODULATE ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE0_ALPHA, GL_TEXTURE0 ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND0_ALPHA, GL_SRC_ALPHA ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE1_ALPHA, GL_PRIMARY_COLOR ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND1_ALPHA, GL_SRC_ALPHA ); } else { // Color has to be multiplied only by brightness glColor4f( data.brightness, data.brightness, data.brightness, data.opacity ); // Alpha will be taken from previous stage glTexEnvi( GL_TEXTURE_ENV, GL_COMBINE_ALPHA, GL_REPLACE ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE0_ALPHA, GL_PREVIOUS ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND0_ALPHA, GL_SRC_ALPHA ); } enableTexture(); } glActiveTexture(GL_TEXTURE0 ); } else if( data.opacity != 1.0 || data.brightness != 1.0 ) { // the window is additionally configured to have its opacity adjusted, // do it if( toplevel->hasAlpha()) { float opacityByBrightness = data.opacity * data.brightness; glTexEnvi( GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_MODULATE ); glColor4f( opacityByBrightness, opacityByBrightness, opacityByBrightness, data.opacity); } else { // Multiply color by brightness and replace alpha by opacity float constant[] = { data.brightness, data.brightness, data.brightness, data.opacity }; glTexEnvi( GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_COMBINE ); glTexEnvi( GL_TEXTURE_ENV, GL_COMBINE_RGB, GL_MODULATE ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE0_RGB, GL_TEXTURE ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND0_RGB, GL_SRC_COLOR ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE1_RGB, GL_CONSTANT ); glTexEnvi( GL_TEXTURE_ENV, GL_OPERAND1_RGB, GL_SRC_COLOR ); glTexEnvi( GL_TEXTURE_ENV, GL_COMBINE_ALPHA, GL_REPLACE ); glTexEnvi( GL_TEXTURE_ENV, GL_SOURCE0_ALPHA, GL_CONSTANT ); glTexEnvfv( GL_TEXTURE_ENV, GL_TEXTURE_ENV_COLOR, constant ); } } enableTexture(); // update texture matrix to handle GL_TEXTURE_2D and GL_TEXTURE_RECTANGLE glMatrixMode( GL_TEXTURE ); glLoadIdentity(); glScalef( texture_scale_x, texture_scale_y, 1 ); if( !texture_y_inverted ) { // Modify texture matrix so that we could always use non-opengl // coordinates for textures glScalef(1, -1, 1); glTranslatef(0, -height(), 0); } glMatrixMode( GL_MODELVIEW ); if(verticeslist.isEmpty()) createVertexGrid(0, 0); // Enable arrays glEnableClientState( GL_VERTEX_ARRAY ); glVertexPointer(3, GL_FLOAT, sizeof(Vertex), verticeslist[0].pos); glEnableClientState( GL_TEXTURE_COORD_ARRAY ); glTexCoordPointer(2, GL_FLOAT, sizeof(Vertex), verticeslist[0].texcoord); // Render if( mask & PAINT_WINDOW_TRANSFORMED ) // Just draw the entire window, no clipping glDrawArrays( GL_QUADS, 0, verticeslist.count() ); else { // Make sure there's only a single quad (no transformed vertices) // Clip using scissoring glEnable( GL_SCISSOR_TEST ); region.translate( x, y); // Back to screen coords int dh = displayHeight(); foreach( QRect r, region.rects()) { // Scissor rect has to be given in OpenGL coords glScissor(r.x(), dh - r.y() - r.height(), r.width(), r.height()); glDrawArrays( GL_QUADS, 0, verticeslist.count() ); } glDisable( GL_SCISSOR_TEST ); } // Restore texture matrix glMatrixMode( GL_TEXTURE ); glLoadIdentity(); glMatrixMode( GL_MODELVIEW ); // Disable arrays glDisableClientState( GL_VERTEX_ARRAY ); glDisableClientState( GL_TEXTURE_COORD_ARRAY ); glPopMatrix(); if( data.opacity != 1.0 || data.saturation != 1.0 || data.brightness != 1.0f ) { if( data.saturation != 1.0 && supports_saturation ) { glActiveTexture(GL_TEXTURE3); glDisable( texture_target ); glActiveTexture(GL_TEXTURE2); glDisable( texture_target ); glActiveTexture(GL_TEXTURE1); glDisable( texture_target ); glActiveTexture(GL_TEXTURE0); } glTexEnvi( GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE ); glColor4f( 0, 0, 0, 0 ); } glPopAttrib(); disableTexture(); } } // namespace