Merge lp:~vanvugt/ubuntu/oneiric/compiz/fix-763005-oneiric into lp:ubuntu/oneiric-proposed/compiz

[Daniel van Vugt]

Fix slow/stuttering display with fglrx (LP: #763005)

This branch was proposed, approved, and marked as merged back in November. But somewhere a mistake was made and the fix never made it into any branch.

[Didier Roche-Tolomelli (didrocks)]

Hey Daniel, thanks for you work there :)
There is a pending sru for compiz (which isn't approved yet as we need the fix to land in precise before it can go in oneiric). See http://people.canonical.com/~ubuntu-archive/pending-sru.html.

So, this one will go first (I hope we can get a first precise compiz release next week), and then, we can push this change for the next SRU (maybe with some more?)

Can you please just resubmit it when the new version lands in precise (and the SRU in oneiric-proposed) and subscribe me directly to it?

Oh, and btw, the versionning will be needed to be ubuntu6.2 now :)

review: Needs Resubmitting

[Sebastien Bacher (seb128)]

Daniel, I'm going to mark this one rejected, Omer is working on a SRU (see https://code.launchpad.net/~om26er/ubuntu/oneiric/compiz/sru-764330/+merge/92296) and will include your patch in his merge request so we can get both fixes in the same upload, technical it's superseded rather than rejected then ;-)

Unmerged revisions

254. By Daniel van Vugt on 2011-11-14

Fix slow/stuttering display with fglrx (LP: #763005)

=== modified file 'debian/changelog'
--- debian/changelog	2011-10-20 14:23:52 +0000
+++ debian/changelog	2012-01-26 10:38:29 +0000
@@ -1,3 +1,9 @@
+compiz (1:0.9.6+bzr20110929-0ubuntu7) oneiric-proposed; urgency=low
+
+  * Fix slow/stuttering display with fglrx (LP: #763005)
+
+ -- Daniel van Vugt <[e-mail]>  Mon, 14 Nov 2011 16:22:01 +0800
+
 compiz (1:0.9.6+bzr20110929-0ubuntu6) oneiric-proposed; urgency=low
 
   * debian/patches/rev_2821_fix_807487.patch:

=== added file 'debian/patches/fix_slow_vsync_lp763005.patch'
--- debian/patches/fix_slow_vsync_lp763005.patch	1970-01-01 00:00:00 +0000
+++ debian/patches/fix_slow_vsync_lp763005.patch	2012-01-26 10:38:29 +0000
@@ -0,0 +1,145 @@
+Description: Fix slow/stuttering display when sync to Vblank is enabled.
+ For fglrx the cause appears to be an ATI/AMD driver bug where
+ glXGetVideoSyncSGI and glXWaitVideoSyncSGI are particularly slow, so we
+ can't afford to call both on every redraw or we end up with 1/3-1/2 framerate.
+ For intel and really all drivers, another cause appears to be excessive
+ waiting in the compiz architecture not allowing enough CPU time to draw the
+ full frame before the next vertical refresh.
+ This patch addresses both possible problems, providing optimal framerates
+ with sync-to-vblank support, most of the time.
+Author: Daniel van Vugt <[e-mail]>
+Bug-Ubuntu: https://launchpad.net/bugs/763005
+
+===================================================================
+--- old/plugins/opengl/src/screen.cpp	2011-09-12 08:09:46 +0000
++++ new/plugins/opengl/src/screen.cpp	2011-09-30 06:19:53 +0000
+@@ -36,6 +36,9 @@
+ #include <math.h>
+ 
+ namespace GL {
++    typedef int (*GLXSwapIntervalProc) (int interval);
++
++    GLXSwapIntervalProc      swapInterval = NULL;
+     GLXBindTexImageProc      bindTexImage = NULL;
+     GLXReleaseTexImageProc   releaseTexImage = NULL;
+     GLXQueryDrawableProc     queryDrawable = NULL;
+@@ -234,6 +237,12 @@
+ 	    getProcAddress ("glXWaitVideoSyncSGI");
+     }
+ 
++    if (strstr (glxExtensions, "GLX_SGI_swap_control"))
++    {
++	GL::swapInterval = (GL::GLXSwapIntervalProc)
++	    getProcAddress ("glXSwapIntervalSGI");
++    }
++
+     glXMakeCurrent (dpy, CompositeScreen::get (s)->output (), priv->ctx);
+ 
+     glExtensions = (const char *) glGetString (GL_EXTENSIONS);
+@@ -999,21 +1008,41 @@
+ 		priv->lastViewport.height);
+ }
+ 
++namespace GL
++{
++
++void
++waitForVideoSync ()
++{
++    if (GL::waitVideoSync)
++    {
++	// Don't wait twice. Just in case.
++	if (GL::swapInterval)
++	    (*GL::swapInterval) (0);
++
++	// Docs: http://www.opengl.org/registry/specs/SGI/video_sync.txt
++	unsigned int frameno;
++	(*GL::waitVideoSync) (1, 0, &frameno);
++    }
++}
++
++void
++controlSwapVideoSync (bool sync)
++{
++    // Docs: http://www.opengl.org/registry/specs/SGI/swap_control.txt
++    if (GL::swapInterval)
++	(*GL::swapInterval) (sync ? 1 : 0);
++    else if (sync)
++	waitForVideoSync ();
++}
++
++} // namespace GL
++
+ void
+ PrivateGLScreen::waitForVideoSync ()
+ {
+-    unsigned int sync;
+-
+-    if (!optionGetSyncToVblank ())
+-	return;
+-
+-    if (GL::getVideoSync)
+-    {
+-	glFlush ();
+-
+-	(*GL::getVideoSync) (&sync);
+-	(*GL::waitVideoSync) (2, (sync + 1) % 2, &sync);
+-    }
++    if (optionGetSyncToVblank ())
++        GL::waitForVideoSync ();
+ }
+ 
+ void
+@@ -1085,10 +1114,23 @@
+ 
+     targetOutput = &screen->outputDevs ()[0];
+ 
+-    waitForVideoSync ();
++    glFlush ();
+ 
++    //
++    // FIXME: Actually fix the composite plugin to be more efficient;
++    // If GL::swapInterval == NULL && GL::waitVideoSync != NULL then the
++    // composite plugin should not be imposing any framerate restriction
++    // (ie. blocking the CPU) at all. Because the framerate will be controlled
++    // and optimized here:
++    //
+     if (mask & COMPOSITE_SCREEN_DAMAGE_ALL_MASK)
+     {
++	//
++	// controlSwapVideoSync is much faster than waitForVideoSync because
++	// it won't block the CPU. The waiting is offloaded to the GPU.
++	// Unfortunately it only works with glXSwapBuffers in most drivers.
++	//
++	GL::controlSwapVideoSync (optionGetSyncToVblank ());
+ 	glXSwapBuffers (screen->dpy (), cScreen->output ());
+     }
+     else
+@@ -1096,6 +1138,7 @@
+ 	BoxPtr pBox;
+ 	int    nBox, y;
+ 
++	waitForVideoSync ();
+ 	pBox = const_cast <Region> (tmpRegion.handle ())->rects;
+ 	nBox = const_cast <Region> (tmpRegion.handle ())->numRects;
+ 
+@@ -1151,7 +1194,7 @@
+ bool
+ PrivateGLScreen::hasVSync ()
+ {
+-   return (GL::getVideoSync && optionGetSyncToVblank ());
++   return (GL::waitVideoSync && optionGetSyncToVblank ());
+ }
+ 
+ void
+@@ -1159,7 +1202,8 @@
+ {
+     if (pendingCommands)
+     {
+-	glFinish ();
++	// glFlush! glFinish would block the CPU, which is bad.
++	glFlush ();
+ 	pendingCommands = false;
+     }
+ }
+

=== modified file 'debian/patches/series'
--- debian/patches/series	2011-10-20 14:23:52 +0000
+++ debian/patches/series	2012-01-26 10:38:29 +0000
@@ -18,3 +18,4 @@
 rev_2884_fix_874004.patch
 rev_2890_fix_879253.patch
 fix-886978.patch
+fix_slow_vsync_lp763005.patch