Imp: Multi-pass blur & D-Bus fading control

- Add multipass blur support. Note GLX Framebuffer support is required.
  My benchmark shows multipass blur brings 5% performance boost for X
  Render backend (3x3box). On GLX backend it brings 10% performance
  boost for 5x5box but negatively affects performance for 3x3box. Thanks
  to jrfonseca for advice. (#107)

- GLX backend: Cache blur texture for each window, for a 12% performance
  boost.

- Add D-Bus fading control. Thanks to yulan6248 for testing. (#112)

- Fix FAQ link in README.md. Thanks to lorenzos for report. (#111)

- Correctly deinitialize VSync on on-the-fly VSync method switch.

- X Render backend: Normalize blur kernel.

- Code clean-up.

- Known issue: Linear corruption on border of a window may appear with X
  Render multi-pass blur. Possible to fix but probably not worthwhile.

5 files changed, 765 insertions, 224 deletions

diff --git a/common.h b/common.h
index bf9f8ae01..cab29f76a 100644
--- a/common.h
+++ b/common.h
@@ -26,6 +26,10 @@
 // #define DEBUG_FRAME      1
 // #define DEBUG_LEADER     1
 // #define DEBUG_C2         1
+// #define DEBUG_GLX        1
+// #define DEBUG_GLX_GLSL   1
+// #define DEBUG_GLX_ERR    1
+// #define DEBUG_GLX_MARK   1
 // #define MONITOR_REPAINT  1
 
 // Whether to enable PCRE regular expression support in blacklists, enabled
@@ -40,8 +44,12 @@
 // #define CONFIG_LIBCONFIG_LEGACY 1
 // Whether to enable DRM VSync support
 // #define CONFIG_VSYNC_DRM 1
-// Whether to enable OpenGL VSync support
+// Whether to enable OpenGL support
 // #define CONFIG_VSYNC_OPENGL 1
+// Whether to enable GLX GLSL support
+// #define CONFIG_VSYNC_OPENGL_GLSL 1
+// Whether to enable GLX FBO support
+// #define CONFIG_VSYNC_OPENGL_FBO 1
 // Whether to enable DBus support with libdbus.
 // #define CONFIG_DBUS 1
 // Whether to enable condition support.
@@ -96,7 +104,7 @@
 
 // libGL
 #ifdef CONFIG_VSYNC_OPENGL
-#ifdef CONFIG_VSYNC_OPENGL_GLSL
+#if defined(CONFIG_VSYNC_OPENGL_GLSL) || defined(CONFIG_VSYNC_OPENGL_FBO)
 #define GL_GLEXT_PROTOTYPES
 #endif
 
@@ -183,6 +191,9 @@
 /// @brief Maximum OpenGL buffer age.
 #define CGLX_MAX_BUFFER_AGE 5
 
+/// @brief Maximum passes for blur.
+#define MAX_BLUR_PASS 5
+
 // Window flags
 
 // Window size is changed
@@ -343,6 +354,32 @@ struct _glx_texture {
 };
 #endif
 
+#ifdef CONFIG_VSYNC_OPENGL_GLSL
+typedef struct {
+  /// Fragment shader for blur.
+  GLuint frag_shader;
+  /// GLSL program for blur.
+  GLuint prog;
+  /// Location of uniform "offset_x" in blur GLSL program.
+  GLint unifm_offset_x;
+  /// Location of uniform "offset_y" in blur GLSL program.
+  GLint unifm_offset_y;
+  /// Location of uniform "factor_center" in blur GLSL program.
+  GLint unifm_factor_center;
+} glx_blur_pass_t;
+
+typedef struct {
+  /// Framebuffer used for blurring.
+  GLuint fbo;
+  /// Textures used for blurring.
+  GLuint textures[2];
+  /// Width of the textures.
+  int width;
+  /// Height of the textures.
+  int height;
+} glx_blur_cache_t;
+#endif
+
 typedef struct {
   Pixmap pixmap;
   Picture pict;
@@ -501,7 +538,7 @@ typedef struct {
   /// Background blur blacklist. A linked list of conditions.
   c2_lptr_t *blur_background_blacklist;
   /// Blur convolution kernel.
-  XFixed *blur_kern;
+  XFixed *blur_kerns[MAX_BLUR_PASS];
   /// How much to dim an inactive window. 0.0 - 1.0, 0 to disable.
   double inactive_dim;
   /// Whether to use fixed inactive dim opacity, instead of deciding
@@ -614,6 +651,8 @@ typedef struct {
   /// Current GLX Z value.
   int glx_z;
 #endif
+  // Cached blur convolution kernels.
+  XFixed *blur_kerns_cache[MAX_BLUR_PASS];
   /// Reset program after next paint.
   bool reset;
 
@@ -702,16 +741,7 @@ typedef struct {
   /// FBConfig-s for GLX pixmap of different depths.
   glx_fbconfig_t *glx_fbconfigs[OPENGL_MAX_DEPTH + 1];
 #ifdef CONFIG_VSYNC_OPENGL_GLSL
-  /// Fragment shader for blur.
-  GLuint glx_frag_shader_blur;
-  /// GLSL program for blur.
-  GLuint glx_prog_blur;
-  /// Location of uniform "offset_x" in blur GLSL program.
-  GLint glx_prog_blur_unifm_offset_x;
-  /// Location of uniform "offset_y" in blur GLSL program.
-  GLint glx_prog_blur_unifm_offset_y;
-  /// Location of uniform "factor_center" in blur GLSL program.
-  GLint glx_prog_blur_unifm_factor_center;
+  glx_blur_pass_t glx_blur_passes[MAX_BLUR_PASS];
 #endif
 #endif
 
@@ -908,6 +938,8 @@ typedef struct _win {
   /// Do not fade if it's false. Change on window type change.
   /// Used by fading blacklist in the future.
   bool fade;
+  /// Override value of window fade state. Set by D-Bus method calls.
+  switch_t fade_force;
   /// Callback to be called after fading completed.
   void (*fade_callback) (session_t *ps, struct _win *w);
 
@@ -950,6 +982,11 @@ typedef struct _win {
 
   /// Whether to blur window background.
   bool blur_background;
+
+#ifdef CONFIG_VSYNC_OPENGL_GLSL
+  /// Textures and FBO background blur use.
+  glx_blur_cache_t glx_blur_cache;
+#endif
 } win;
 
 /// Temporary structure used for communication between
@@ -1749,6 +1786,9 @@ force_repaint(session_t *ps);
 bool
 vsync_init(session_t *ps);
 
+void
+vsync_deinit(session_t *ps);
+
 #ifdef CONFIG_VSYNC_OPENGL
 /** @name GLX
  */
@@ -1788,9 +1828,13 @@ glx_tex_binded(const glx_texture_t *ptex, Pixmap pixmap) {
 void
 glx_set_clip(session_t *ps, XserverRegion reg, const reg_data_t *pcache_reg);
 
+#ifdef CONFIG_VSYNC_OPENGL_GLSL
 bool
 glx_blur_dst(session_t *ps, int dx, int dy, int width, int height, float z,
-    GLfloat factor_center, XserverRegion reg_tgt, const reg_data_t *pcache_reg);
+    GLfloat factor_center,
+    XserverRegion reg_tgt, const reg_data_t *pcache_reg,
+    glx_blur_cache_t *pbc);
+#endif
 
 bool
 glx_dim_dst(session_t *ps, int dx, int dy, int width, int height, float z,
@@ -1812,7 +1856,78 @@ glx_create_shader(GLenum shader_type, const char *shader_str);
 GLuint
 glx_create_program(const GLuint * const shaders, int nshaders);
 #endif
+
+/**
+ * Free a GLX texture.
+ */
+static inline void
+free_texture_r(session_t *ps, GLuint *ptexture) {
+  if (*ptexture) {
+    assert(ps->glx_context);
+    glDeleteTextures(1, ptexture);
+    *ptexture = 0;
+  }
+}
+
+/**
+ * Free a GLX Framebuffer object.
+ */
+static inline void
+free_glx_fbo(session_t *ps, GLuint *pfbo) {
+#ifdef CONFIG_VSYNC_OPENGL_FBO
+  if (*pfbo) {
+    glDeleteFramebuffers(1, pfbo);
+    *pfbo = 0;
+  }
 #endif
+  assert(!*pfbo);
+}
+
+#ifdef CONFIG_VSYNC_OPENGL_GLSL
+/**
+ * Free data in glx_blur_cache_t on resize.
+ */
+static inline void
+free_glx_bc_resize(session_t *ps, glx_blur_cache_t *pbc) {
+  free_texture_r(ps, &pbc->textures[0]);
+  free_texture_r(ps, &pbc->textures[1]);
+  pbc->width = 0;
+  pbc->height = 0;
+}
+
+/**
+ * Free a glx_blur_cache_t
+ */
+static inline void
+free_glx_bc(session_t *ps, glx_blur_cache_t *pbc) {
+  free_glx_fbo(ps, &pbc->fbo);
+  free_glx_bc_resize(ps, pbc);
+}
+#endif
+#endif
+
+/**
+ * Free a glx_texture_t.
+ */
+static inline void
+free_texture(session_t *ps, glx_texture_t **pptex) {
+  glx_texture_t *ptex = *pptex;
+
+  // Quit if there's nothing
+  if (!ptex)
+    return;
+
+#ifdef CONFIG_VSYNC_OPENGL
+  glx_release_pixmap(ps, ptex);
+
+  free_texture_r(ps, &ptex->texture);
+
+  // Free structure itself
+  free(ptex);
+  *pptex = NULL;
+#endif
+  assert(!*pptex);
+}
 
 /**
  * Add a OpenGL debugging marker.
@@ -1846,29 +1961,6 @@ glx_mark_frame(session_t *ps) {
 #endif
 }
 
-static inline void
-free_texture(session_t *ps, glx_texture_t **pptex) {
-#ifdef CONFIG_VSYNC_OPENGL
-  glx_texture_t *ptex = *pptex;
-
-  // Quit if there's nothing
-  if (!ptex)
-    return;
-
-  glx_release_pixmap(ps, ptex);
-
-  // Free texture
-  if (ptex->texture) {
-    glDeleteTextures(1, &ptex->texture);
-    ptex->texture = 0;
-  }
-
-  // Free structure itself
-  free(ptex);
-  *pptex = NULL;
-#endif
-}
-
 ///@}
 
 /** @name DBus handling
@@ -1913,6 +2005,9 @@ void
 win_set_shadow_force(session_t *ps, win *w, switch_t val);
 
 void
+win_set_fade_force(session_t *ps, win *w, switch_t val);
+
+void
 win_set_focused_force(session_t *ps, win *w, switch_t val);
 
 void
@@ -1920,6 +2015,9 @@ win_set_invert_color_force(session_t *ps, win *w, switch_t val);
 
 void
 opts_init_track_focus(session_t *ps);
+
+void
+opts_set_no_fading_openclose(session_t *ps, bool newval);
 //!@}
 #endif
 
diff --git a/compton.c b/compton.c
index 7e005b638..f058bf927 100644
--- a/compton.c
+++ b/compton.c
@@ -70,6 +70,14 @@ static int (* const (VSYNC_FUNCS_WAIT[NUM_VSYNC]))(session_t *ps) = {
 #endif
 };
 
+/// Function pointers to deinitialize VSync.
+static void (* const (VSYNC_FUNCS_DEINIT[NUM_VSYNC]))(session_t *ps) = {
+#ifdef CONFIG_VSYNC_OPENGL
+  [VSYNC_OPENGL_SWC   ] = vsync_opengl_swc_deinit,
+  [VSYNC_OPENGL_MSWC  ] = vsync_opengl_mswc_deinit,
+#endif
+};
+
 /// Names of root window properties that could point to a pixmap of
 /// background.
 const static char *background_props_str[] = {
@@ -1276,13 +1284,11 @@ win_paint_shadow(session_t *ps, win *w,
 }
 
 /**
- * Create an alternative picture for a window.
+ * Create an picture.
  */
 static inline Picture
-win_build_picture(session_t *ps, win *w, XRenderPictFormat *pictfmt) {
-  const int wid = w->widthb;
-  const int hei = w->heightb;
-
+xr_build_picture(session_t *ps, int wid, int hei,
+    XRenderPictFormat *pictfmt) {
   if (!pictfmt)
     pictfmt = XRenderFindVisualFormat(ps->dpy, ps->vis);
 
@@ -1300,6 +1306,73 @@ win_build_picture(session_t *ps, win *w, XRenderPictFormat *pictfmt) {
 }
 
 /**
+ * @brief Blur an area on a buffer.
+ *
+ * @param ps current session
+ * @param tgt_buffer a buffer as both source and destination
+ * @param x x pos
+ * @param y y pos
+ * @param wid width
+ * @param hei height
+ * @param blur_kerns blur kernels, ending with a NULL, guaranteed to have at
+ *                    least one kernel
+ * @param reg_clip a clipping region to be applied on intermediate buffers
+ *
+ * @return true if successful, false otherwise
+ */
+static bool
+xr_blur_dst(session_t *ps, Picture tgt_buffer,
+    int x, int y, int wid, int hei, XFixed **blur_kerns,
+    XserverRegion reg_clip) {
+  assert(blur_kerns[0]);
+
+  // Directly copying from tgt_buffer to it does not work, so we create a
+  // Picture in the middle.
+  Picture tmp_picture = xr_build_picture(ps, wid, hei, NULL);
+
+  if (!tmp_picture) {
+    printf_errf("(): Failed to build intermediate Picture.");
+    return false;
+  }
+
+  if (reg_clip && tmp_picture)
+    XFixesSetPictureClipRegion(ps->dpy, tmp_picture, reg_clip, 0, 0);
+
+  Picture src_pict = tgt_buffer, dst_pict = tmp_picture;
+  for (int i = 0; blur_kerns[i]; ++i) {
+    assert(i < MAX_BLUR_PASS - 1);
+    XFixed *convolution_blur = blur_kerns[i];
+    int kwid = XFixedToDouble(convolution_blur[0]),
+        khei = XFixedToDouble(convolution_blur[1]);
+    bool rd_from_tgt = (tgt_buffer == src_pict);
+
+    // Copy from source picture to destination. The filter must
+    // be applied on source picture, to get the nearby pixels outside the
+    // window.
+    XRenderSetPictureFilter(ps->dpy, src_pict, XRFILTER_CONVOLUTION,
+        convolution_blur, kwid * khei + 2);
+    XRenderComposite(ps->dpy, PictOpSrc, src_pict, None, dst_pict,
+        (rd_from_tgt ? x: 0), (rd_from_tgt ? y: 0), 0, 0,
+        (rd_from_tgt ? 0: x), (rd_from_tgt ? 0: y), wid, hei);
+    xrfilter_reset(ps, src_pict);
+
+    {
+      XserverRegion tmp = src_pict;
+      src_pict = dst_pict;
+      dst_pict = tmp;
+    }
+  }
+
+  if (src_pict != tgt_buffer)
+    XRenderComposite(ps->dpy, PictOpSrc, src_pict, None, tgt_buffer,
+        0, 0, 0, 0, x, y, wid, hei);
+
+  free_picture(ps, &tmp_picture);
+
+  return true;
+}
+
+/**
  * Blur the background of a window.
  */
 static inline void
@@ -1321,46 +1394,63 @@ win_blur_background(session_t *ps, win *w, Picture tgt_buffer,
   switch (ps->o.backend) {
     case BKEND_XRENDER:
       {
-        // Directly copying from tgt_buffer does not work, so we create a
-        // Picture in the middle.
-        Picture tmp_picture = win_build_picture(ps, w, NULL);
+        // Normalize blur kernels
+        for (int i = 0; i < MAX_BLUR_PASS; ++i) {
+          XFixed *kern_src = ps->o.blur_kerns[i];
+          XFixed *kern_dst = ps->blur_kerns_cache[i];
+          assert(i < MAX_BLUR_PASS);
+          if (!kern_src) {
+            assert(!kern_dst);
+            break;
+          }
+
+          assert(!kern_dst
+              || (kern_src[0] == kern_dst[0] && kern_src[1] == kern_dst[1]));
+
+          // Skip for fixed factor_center if the cache exists already
+          if (ps->o.blur_background_fixed && kern_dst) continue;
+
+          int kwid = XFixedToDouble(kern_src[0]),
+              khei = XFixedToDouble(kern_src[1]);
 
-        if (!tmp_picture)
-          return;
+          // Allocate cache space if needed
+          if (!kern_dst) {
+            kern_dst = malloc((kwid * khei + 2) * sizeof(XFixed));
+            if (!kern_dst) {
+              printf_errf("(): Failed to allocate memory for blur kernel.");
+              return;
+            }
+            ps->blur_kerns_cache[i] = kern_dst;
+          }
 
-        XFixed *convolution_blur = ps->o.blur_kern;
-        int kwid = XFixedToDouble((ps->o.blur_kern[0])),
-            khei = XFixedToDouble((ps->o.blur_kern[1]));
+          // Modify the factor of the center pixel
+          kern_src[2 + (khei / 2) * kwid + kwid / 2] =
+            XDoubleToFixed(factor_center);
 
-        // Modify the factor of the center pixel
-        convolution_blur[2 + (khei / 2) * kwid + kwid / 2] = XDoubleToFixed(factor_center);
+          // Copy over
+          memcpy(kern_dst, kern_src, (kwid * khei + 2) * sizeof(XFixed));
+          normalize_conv_kern(kwid, khei, kern_dst + 2);
+        }
 
         // Minimize the region we try to blur, if the window itself is not
         // opaque, only the frame is.
-        if (WMODE_SOLID == w->mode && w->frame_opacity) {
+        XserverRegion reg_noframe = None;
+        if (WMODE_SOLID == w->mode) {
           XserverRegion reg_all = border_size(ps, w, false);
-          XserverRegion reg_noframe = win_get_region_noframe(ps, w, false);
+          reg_noframe = win_get_region_noframe(ps, w, false);
           XFixesSubtractRegion(ps->dpy, reg_noframe, reg_all, reg_noframe);
-          XFixesSetPictureClipRegion(ps->dpy, tmp_picture, reg_noframe, 0, 0);
           free_region(ps, &reg_all);
-          free_region(ps, &reg_noframe);
         }
-
-        // Copy the content to tmp_picture, then copy back. The filter must
-        // be applied on tgt_buffer, to get the nearby pixels outside the
-        // window.
-        XRenderSetPictureFilter(ps->dpy, tgt_buffer, XRFILTER_CONVOLUTION, convolution_blur, kwid * khei + 2);
-        XRenderComposite(ps->dpy, PictOpSrc, tgt_buffer, None, tmp_picture, x, y, 0, 0, 0, 0, wid, hei);
-        xrfilter_reset(ps, tgt_buffer);
-        XRenderComposite(ps->dpy, PictOpSrc, tmp_picture, None, tgt_buffer, 0, 0, 0, 0, x, y, wid, hei);
-
-        free_picture(ps, &tmp_picture);
+        xr_blur_dst(ps, tgt_buffer, x, y, wid, hei, ps->blur_kerns_cache,
+            reg_noframe);
+        free_region(ps, &reg_noframe);
       }
       break;
-#ifdef CONFIG_VSYNC_OPENGL
+#ifdef CONFIG_VSYNC_OPENGL_GLSL
     case BKEND_GLX:
+      // TODO: Handle frame opacity
       glx_blur_dst(ps, x, y, wid, hei, ps->glx_z - 0.5, factor_center,
-          reg_paint, pcache_reg);
+          reg_paint, pcache_reg, &w->glx_blur_cache);
       break;
 #endif
     default:
@@ -1447,7 +1537,7 @@ win_paint_win(session_t *ps, win *w, XserverRegion reg_paint,
 
   // Invert window color, if required
   if (BKEND_XRENDER == ps->o.backend && w->invert_color) {
-    Picture newpict = win_build_picture(ps, w, w->pictfmt);
+    Picture newpict = xr_build_picture(ps, wid, hei, w->pictfmt);
     if (newpict) {
       // Apply clipping region to save some CPU
       if (reg_paint) {
@@ -1993,14 +2083,8 @@ map_win(session_t *ps, Window id) {
   win_determine_shadow(ps, w);
 
   // Set fading state
-  w->in_openclose = false;
-  if (ps->o.no_fading_openclose) {
-    set_fade_callback(ps, w, finish_map_win, true);
-    w->in_openclose = true;
-  }
-  else {
-    set_fade_callback(ps, w, NULL, true);
-  }
+  w->in_openclose = true;
+  set_fade_callback(ps, w, finish_map_win, true);
   win_determine_fade(ps, w);
 
   win_determine_blur_background(ps, w);
@@ -2066,9 +2150,7 @@ unmap_win(session_t *ps, win *w) {
   // Fading out
   w->flags |= WFLAG_OPCT_CHANGE;
   set_fade_callback(ps, w, unmap_callback, false);
-  if (ps->o.no_fading_openclose) {
-    w->in_openclose = true;
-  }
+  w->in_openclose = true;
   win_determine_fade(ps, w);
 
   // Validate pixmap if we have to do fading
@@ -2195,7 +2277,9 @@ calc_dim(session_t *ps, win *w) {
  */
 static void
 win_determine_fade(session_t *ps, win *w) {
-  if ((ps->o.no_fading_openclose && w->in_openclose)
+  if (UNSET != w->fade_force)
+    w->fade = w->fade_force;
+  else if ((ps->o.no_fading_openclose && w->in_openclose)
       || win_match(ps, w, ps->o.fade_blacklist, &w->cache_fblst))
     w->fade = false;
   else
@@ -2593,6 +2677,7 @@ add_win(session_t *ps, Window id, Window prev) {
     .opacity_prop_client = OPAQUE,
 
     .fade = false,
+    .fade_force = UNSET,
     .fade_callback = NULL,
 
     .frame_opacity = 0.0,
@@ -3418,6 +3503,19 @@ win_set_shadow_force(session_t *ps, win *w, switch_t val) {
   if (val != w->shadow_force) {
     w->shadow_force = val;
     win_determine_shadow(ps, w);
+    ps->ev_received = true;
+  }
+}
+
+/**
+ * Set w->fade_force of a window.
+ */
+void
+win_set_fade_force(session_t *ps, win *w, switch_t val) {
+  if (val != w->fade_force) {
+    w->fade_force = val;
+    win_determine_fade(ps, w);
+    ps->ev_received = true;
   }
 }
 
@@ -3429,6 +3527,7 @@ win_set_focused_force(session_t *ps, win *w, switch_t val) {
   if (val != w->focused_force) {
     w->focused_force = val;
     win_update_focused(ps, w);
+    ps->ev_received = true;
   }
 }
 
@@ -3440,6 +3539,7 @@ win_set_invert_color_force(session_t *ps, win *w, switch_t val) {
   if (val != w->invert_color_force) {
     w->invert_color_force = val;
     win_determine_invert_color(ps, w);
+    ps->ev_received = true;
   }
 }
 
@@ -3465,6 +3565,20 @@ opts_init_track_focus(session_t *ps) {
   // Recheck focus
   recheck_focus(ps);
 }
+
+/**
+ * Set no_fading_openclose option.
+ */
+void
+opts_set_no_fading_openclose(session_t *ps, bool newval) {
+  if (newval != ps->o.no_fading_openclose) {
+    ps->o.no_fading_openclose = newval;
+    for (win *w = ps->list; w; w = w->next)
+      win_determine_fade(ps, w);
+    ps->ev_received = true;
+  }
+}
+
 //!@}
 #endif
 
@@ -4222,6 +4336,8 @@ usage(int ret) {
     "  --blur-background-fixed.\n"
     "  A 7x7 Guassian blur kernel looks like:\n"
     "    --blur-kern '7,7,0.000003,0.000102,0.000849,0.001723,0.000849,0.000102,0.000003,0.000102,0.003494,0.029143,0.059106,0.029143,0.003494,0.000102,0.000849,0.029143,0.243117,0.493069,0.243117,0.029143,0.000849,0.001723,0.059106,0.493069,0.493069,0.059106,0.001723,0.000849,0.029143,0.243117,0.493069,0.243117,0.029143,0.000849,0.000102,0.003494,0.029143,0.059106,0.029143,0.003494,0.000102,0.000003,0.000102,0.000849,0.001723,0.000849,0.000102,0.000003'\n"
+    "  Up to 4 blur kernels may be specified, separated with semicolon, for\n"
+    "  multi-pass blur.\n"
     "  May also be one the predefined kernels: 3x3box (default), 5x5box,\n"
     "  7x7box, 3x3gaussian, 5x5gaussian, 7x7gaussian, 9x9gaussian,\n"
     "  11x11gaussian.\n"
@@ -4419,7 +4535,7 @@ parse_matrix_readnum(const char *src, double *dest) {
  * Parse a matrix.
  */
 static inline XFixed *
-parse_matrix(session_t *ps, const char *src) {
+parse_matrix(session_t *ps, const char *src, const char **endptr) {
   int wid = 0, hei = 0;
   const char *pc = NULL;
   XFixed *matrix = NULL;
@@ -4481,12 +4597,28 @@ parse_matrix(session_t *ps, const char *src) {
   }
 
   // Detect trailing characters
-  for ( ;*pc; ++pc)
+  for ( ;*pc && ';' != *pc; ++pc)
     if (!isspace(*pc) && ',' != *pc) {
       printf_errf("(): Trailing characters in matrix string.");
       goto parse_matrix_err;
     }
 
+  // Jump over spaces after ';'
+  if (';' == *pc) {
+    ++pc;
+    while (*pc && isspace(*pc))
+      ++pc;
+  }
+
+  // Require an end of string if endptr is not provided, otherwise
+  // copy end pointer to endptr
+  if (endptr)
+    *endptr = pc;
+  else if (*pc) {
+    printf_errf("(): Only one matrix expected.");
+    goto parse_matrix_err;
+  }
+
   // Fill in width and height
   matrix[0] = XDoubleToFixed(wid);
   matrix[1] = XDoubleToFixed(hei);
@@ -4502,7 +4634,12 @@ parse_matrix_err:
  * Parse a convolution kernel.
  */
 static inline XFixed *
-parse_conv_kern(session_t *ps, const char *src) {
+parse_conv_kern(session_t *ps, const char *src, const char **endptr) {
+  return parse_matrix(ps, src, endptr);
+}
+
+static bool
+parse_conv_kern_lst(session_t *ps, const char *src, XFixed **dest, int max) {
   static const struct {
     const char *name;
     const char *kern_str;
@@ -4519,8 +4656,30 @@ parse_conv_kern(session_t *ps, const char *src) {
   for (int i = 0;
       i < sizeof(CONV_KERN_PREDEF) / sizeof(CONV_KERN_PREDEF[0]); ++i)
     if (!strcmp(CONV_KERN_PREDEF[i].name, src))
-      return parse_matrix(ps, CONV_KERN_PREDEF[i].kern_str);
-  return parse_matrix(ps, src);
+      return parse_conv_kern_lst(ps, CONV_KERN_PREDEF[i].kern_str, dest, max);
+
+  int i = 0;
+  const char *pc = src;
+
+  // Free old kernels
+  for (i = 0; i < max; ++i) {
+    free(dest[i]);
+    dest[i] = NULL;
+  }
+
+  // Continue parsing until the end of source string
+  i = 0;
+  while (pc && *pc && i < max - 1) {
+    if (!(dest[i++] = parse_conv_kern(ps, pc, &pc)))
+      return false;
+  }
+
+  if (*pc) {
+    printf_errf("(): Too many blur kernels!");
+    return false;
+  }
+
+  return true;
 }
 
 #ifdef CONFIG_LIBCONFIG
@@ -5115,9 +5274,9 @@ get_cfg(session_t *ps, int argc, char *const *argv, bool first_pass) {
         break;
       case 301:
         // --blur-kern
-        free(ps->o.blur_kern);
-        if (!(ps->o.blur_kern = parse_conv_kern(ps, optarg)))
+        if (!parse_conv_kern_lst(ps, optarg, ps->o.blur_kerns, MAX_BLUR_PASS))
           exit(1);
+        break;
       case 302:
         // --resize-damage
         ps->o.resize_damage = atoi(optarg);
@@ -5187,7 +5346,7 @@ get_cfg(session_t *ps, int argc, char *const *argv, bool first_pass) {
   }
 
   // Fill default blur kernel
-  if (ps->o.blur_background && !ps->o.blur_kern) {
+  if (ps->o.blur_background && !ps->o.blur_kerns[0]) {
     // Convolution filter parameter (box blur)
     // gaussian or binomial filters are definitely superior, yet looks
     // like they aren't supported as of xorg-server-1.13.0
@@ -5200,12 +5359,12 @@ get_cfg(session_t *ps, int argc, char *const *argv, bool first_pass) {
       XDoubleToFixed(1), XDoubleToFixed(1), XDoubleToFixed(1),
       XDoubleToFixed(1), XDoubleToFixed(1), XDoubleToFixed(1),
     };
-    ps->o.blur_kern = malloc(sizeof(convolution_blur));
-    if (!ps->o.blur_kern) {
+    ps->o.blur_kerns[0] = malloc(sizeof(convolution_blur));
+    if (!ps->o.blur_kerns[0]) {
       printf_errf("(): Failed to allocate memory for convolution kernel.");
       exit(1);
     }
-    memcpy(ps->o.blur_kern, &convolution_blur, sizeof(convolution_blur));
+    memcpy(ps->o.blur_kerns[0], &convolution_blur, sizeof(convolution_blur));
   }
 }
 
@@ -5536,6 +5695,19 @@ vsync_opengl_oml_wait(session_t *ps) {
 
   return 0;
 }
+
+static void
+vsync_opengl_swc_deinit(session_t *ps) {
+  // The standard says it doesn't accept 0, but in fact it probably does
+  if (ps->glx_context && ps->glXSwapIntervalProc)
+    ps->glXSwapIntervalProc(0);
+}
+
+static void
+vsync_opengl_mswc_deinit(session_t *ps) {
+  if (ps->glx_context && ps->glXSwapIntervalMESAProc)
+    ps->glXSwapIntervalMESAProc(0);
+}
 #endif
 
 /**
@@ -5565,6 +5737,16 @@ vsync_wait(session_t *ps) {
 }
 
 /**
+ * Deinitialize current VSync method.
+ */
+void
+vsync_deinit(session_t *ps) {
+  if (ps->o.vsync && VSYNC_FUNCS_DEINIT[ps->o.vsync])
+    VSYNC_FUNCS_DEINIT[ps->o.vsync](ps);
+  ps->o.vsync = VSYNC_NONE;
+}
+
+/**
  * Pregenerate alpha pictures.
  */
 static void
@@ -6030,7 +6212,7 @@ session_init(session_t *ps_old, int argc, char **argv) {
       .blur_background_frame = false,
       .blur_background_fixed = false,
       .blur_background_blacklist = NULL,
-      .blur_kern = NULL,
+      .blur_kerns = { NULL },
       .inactive_dim = 0.0,
       .inactive_dim_fixed = false,
       .invert_color_list = NULL,
@@ -6096,12 +6278,6 @@ session_init(session_t *ps_old, int argc, char **argv) {
     .glXWaitVideoSyncSGI = NULL,
     .glXGetSyncValuesOML = NULL,
     .glXWaitForMscOML = NULL,
-
-#ifdef CONFIG_VSYNC_OPENGL_GLSL
-    .glx_prog_blur_unifm_offset_x = -1,
-    .glx_prog_blur_unifm_offset_y = -1,
-    .glx_prog_blur_unifm_factor_center = -1,
-#endif
 #endif
 
     .xfixes_event = 0,
@@ -6151,6 +6327,14 @@ session_init(session_t *ps_old, int argc, char **argv) {
   // Allocate a session and copy default values into it
   session_t *ps = malloc(sizeof(session_t));
   memcpy(ps, &s_def, sizeof(session_t));
+#ifdef CONFIG_VSYNC_OPENGL_GLSL
+  for (int i = 0; i < MAX_BLUR_PASS; ++i) {
+    glx_blur_pass_t *ppass = &ps->glx_blur_passes[i];
+    ppass->unifm_factor_center = -1;
+    ppass->unifm_offset_x = -1;
+    ppass->unifm_offset_y = -1;
+  }
+#endif
   ps_g = ps;
   ps->ignore_tail = &ps->ignore_head;
   gettimeofday(&ps->time_start, NULL);
@@ -6513,7 +6697,10 @@ session_destroy(session_t *ps) {
   free(ps->o.display);
   free(ps->o.logpath);
   free(ps->o.config_file);
-  free(ps->o.blur_kern);
+  for (int i = 0; i < MAX_BLUR_PASS; ++i) {
+    free(ps->o.blur_kerns[i]);
+    free(ps->blur_kerns_cache[i]);
+  }
   free(ps->pfds_read);
   free(ps->pfds_write);
   free(ps->pfds_except);
diff --git a/compton.h b/compton.h
index 24325ab12..b3695af29 100644
--- a/compton.h
+++ b/compton.h
@@ -231,6 +231,9 @@ free_win_res(session_t *ps, win *w) {
   free(w->class_instance);
   free(w->class_general);
   free(w->role);
+#ifdef CONFIG_VSYNC_OPENGL_GLSL
+  free_glx_bc(ps, &w->glx_blur_cache);
+#endif
 }
 
 /**
@@ -594,6 +597,24 @@ set_tgt_clip(session_t *ps, XserverRegion reg, const reg_data_t *pcache_reg) {
   }
 }
 
+static bool
+xr_blur_dst(session_t *ps, Picture tgt_buffer,
+    int x, int y, int wid, int hei, XFixed **blur_kerns,
+    XserverRegion reg_clip);
+
+/**
+ * Normalize a convolution kernel.
+ */
+static inline void
+normalize_conv_kern(int wid, int hei, XFixed *kern) {
+  double sum = 0.0;
+  for (int i = 0; i < wid * hei; ++i)
+    sum += XFixedToDouble(kern[i]);
+  double factor = 1.0 / sum;
+  for (int i = 0; i < wid * hei; ++i)
+    kern[i] = XDoubleToFixed(XFixedToDouble(kern[i]) * factor);
+}
+
 static void
 paint_all(session_t *ps, XserverRegion region, XserverRegion region_real, win *t);
 
@@ -1095,6 +1116,12 @@ vsync_opengl_wait(session_t *ps);
 
 static int
 vsync_opengl_oml_wait(session_t *ps);
+
+static void
+vsync_opengl_swc_deinit(session_t *ps);
+
+static void
+vsync_opengl_mswc_deinit(session_t *ps);
 #endif
 
 static void
diff --git a/dbus.c b/dbus.c
index e48aa1a02..f2b019aa3 100644
--- a/dbus.c
+++ b/dbus.c
@@ -717,6 +717,7 @@ cdbus_process_win_get(session_t *ps, DBusMessage *msg) {
   cdbus_m_win_get_do(wmwin, cdbus_reply_bool);
   cdbus_m_win_get_do(leader, cdbus_reply_wid);
   cdbus_m_win_get_do(focused_real, cdbus_reply_bool);
+  cdbus_m_win_get_do(fade_force, cdbus_reply_enum);
   cdbus_m_win_get_do(shadow_force, cdbus_reply_enum);
   cdbus_m_win_get_do(focused_force, cdbus_reply_enum);
   cdbus_m_win_get_do(invert_color_force, cdbus_reply_enum);
@@ -770,8 +771,6 @@ cdbus_process_win_set(session_t *ps, DBusMessage *msg) {
     return true;
   }
 
-  ps->ev_received = true;
-
 #define cdbus_m_win_set_do(tgt, type, real_type) \
   if (!strcmp(MSTR(tgt), target)) { \
     real_type val; \
@@ -789,6 +788,14 @@ cdbus_process_win_set(session_t *ps, DBusMessage *msg) {
     goto cdbus_process_win_set_success;
   }
 
+  if (!strcmp("fade_force", target)) {
+    cdbus_enum_t val = UNSET;
+    if (!cdbus_msg_get_arg(msg, 2, CDBUS_TYPE_ENUM, &val))
+      return false;
+    win_set_fade_force(ps, w, val);
+    goto cdbus_process_win_set_success;
+  }
+
   if (!strcmp("focused_force", target)) {
     cdbus_enum_t val = UNSET;
     if (!cdbus_msg_get_arg(msg, 2, CDBUS_TYPE_ENUM, &val))
@@ -912,6 +919,11 @@ cdbus_process_opts_get(session_t *ps, DBusMessage *msg) {
   cdbus_m_opts_get_do(shadow_opacity, cdbus_reply_double);
   cdbus_m_opts_get_do(clear_shadow, cdbus_reply_bool);
 
+  cdbus_m_opts_get_do(fade_delta, cdbus_reply_int32);
+  cdbus_m_opts_get_do(fade_in_step, cdbus_reply_int32);
+  cdbus_m_opts_get_do(fade_out_step, cdbus_reply_int32);
+  cdbus_m_opts_get_do(no_fading_openclose, cdbus_reply_bool);
+
   cdbus_m_opts_get_do(blur_background, cdbus_reply_bool);
   cdbus_m_opts_get_do(blur_background_frame, cdbus_reply_bool);
   cdbus_m_opts_get_do(blur_background_fixed, cdbus_reply_bool);
@@ -961,6 +973,42 @@ cdbus_process_opts_set(session_t *ps, DBusMessage *msg) {
     goto cdbus_process_opts_set_success; \
   }
 
+  // fade_delta
+  if (!strcmp("fade_delta", target)) {
+    int32_t val = 0.0;
+    if (!cdbus_msg_get_arg(msg, 1, DBUS_TYPE_INT32, &val))
+      return false;
+    ps->o.fade_delta = max_i(val, 1);
+    goto cdbus_process_opts_set_success;
+  }
+
+  // fade_in_step
+  if (!strcmp("fade_in_step", target)) {
+    double val = 0.0;
+    if (!cdbus_msg_get_arg(msg, 1, DBUS_TYPE_DOUBLE, &val))
+      return false;
+    ps->o.fade_in_step = normalize_d(val) * OPAQUE;
+    goto cdbus_process_opts_set_success;
+  }
+
+  // fade_out_step
+  if (!strcmp("fade_out_step", target)) {
+    double val = 0.0;
+    if (!cdbus_msg_get_arg(msg, 1, DBUS_TYPE_DOUBLE, &val))
+      return false;
+    ps->o.fade_out_step = normalize_d(val) * OPAQUE;
+    goto cdbus_process_opts_set_success;
+  }
+
+  // no_fading_openclose
+  if (!strcmp("no_fading_openclose", target)) {
+    dbus_bool_t val = FALSE;
+    if (!cdbus_msg_get_arg(msg, 1, DBUS_TYPE_BOOLEAN, &val))
+      return false;
+    opts_set_no_fading_openclose(ps, val);
+    goto cdbus_process_opts_set_success;
+  }
+
   // unredir_if_possible
   if (!strcmp("unredir_if_possible", target)) {
     dbus_bool_t val = FALSE;
@@ -1002,6 +1050,7 @@ cdbus_process_opts_set(session_t *ps, DBusMessage *msg) {
     const char * val = NULL;
     if (!cdbus_msg_get_arg(msg, 1, DBUS_TYPE_STRING, &val))
       return false;
+    vsync_deinit(ps);
     if (!parse_vsync(ps, val)) {
       printf_errf("(): " CDBUS_ERROR_BADARG_S, 1, "Value invalid.");
       cdbus_reply_err(ps, msg, CDBUS_ERROR_BADARG, CDBUS_ERROR_BADARG_S, 1, "Value invalid.");
diff --git a/opengl.c b/opengl.c
index edd615132..635cd33ef 100644
--- a/opengl.c
+++ b/opengl.c
@@ -160,10 +160,13 @@ void
 glx_destroy(session_t *ps) {
 #ifdef CONFIG_VSYNC_OPENGL_GLSL
   // Free GLSL shaders/programs
-  if (ps->glx_frag_shader_blur)
-    glDeleteShader(ps->glx_frag_shader_blur);
-  if (ps->glx_prog_blur)
-    glDeleteProgram(ps->glx_prog_blur);
+  for (int i = 0; i < MAX_BLUR_PASS; ++i) {
+    glx_blur_pass_t *ppass = &ps->glx_blur_passes[i];
+    if (ppass->frag_shader)
+      glDeleteShader(ppass->frag_shader);
+    if (ppass->prog)
+      glDeleteProgram(ppass->prog);
+  }
 #endif
 
   // Free FBConfigs
@@ -199,8 +202,28 @@ glx_on_root_change(session_t *ps) {
  */
 bool
 glx_init_blur(session_t *ps) {
+  assert(ps->o.blur_kerns[0]);
+
+  // Allocate PBO if more than one blur kernel is present
+  if (ps->o.blur_kerns[1]) {
+#ifdef CONFIG_VSYNC_OPENGL_FBO
+    // Try to generate a framebuffer
+    GLuint fbo = 0;
+    glGenFramebuffers(1, &fbo);
+    if (!fbo) {
+      printf_errf("(): Failed to generate Framebuffer. Cannot do "
+          "multi-pass blur with GLX backend.");
+      return false;
+    }
+    glDeleteFramebuffers(1, &fbo);
+#else
+    printf_errf("(): FBO support not compiled in. Cannot do multi-pass blur "
+        "with GLX backend.");
+    return false;
+#endif
+  }
+
 #ifdef CONFIG_VSYNC_OPENGL_GLSL
-  // Build shader
   {
     static const char *FRAG_SHADER_BLUR_PREFIX =
       "#version 110\n"
@@ -235,73 +258,81 @@ glx_init_blur(session_t *ps) {
       shader_add = FRAG_SHADER_BLUR_ADD_GPUSHADER4;
     }
 
-    int wid = XFixedToDouble(ps->o.blur_kern[0]),
-        hei = XFixedToDouble(ps->o.blur_kern[1]);
-    int nele = wid * hei - 1;
-    int len = strlen(FRAG_SHADER_BLUR_PREFIX) + strlen(sampler_type) + strlen(extension) + (strlen(shader_add) + strlen(texture_func) + 42) * nele + strlen(FRAG_SHADER_BLUR_SUFFIX) + strlen(texture_func) + 12 + 1;
-    char *shader_str = calloc(len, sizeof(char));
-    if (!shader_str) {
-      printf_errf("(): Failed to allocate %d bytes for shader string.", len);
-      return false;
-    }
-    {
-      char *pc = shader_str;
-      sprintf(pc, FRAG_SHADER_BLUR_PREFIX, extension, sampler_type);
-      pc += strlen(pc);
-      assert(strlen(shader_str) < len);
-
-      double sum = 0.0;
-      for (int i = 0; i < hei; ++i) {
-        for (int j = 0; j < wid; ++j) {
-          if (hei / 2 == i && wid / 2 == j)
-            continue;
-          double val = XFixedToDouble(ps->o.blur_kern[2 + i * wid + j]);
-          if (0.0 == val)
-            continue;
-          sum += val;
-          sprintf(pc, shader_add, val, texture_func, j - wid / 2, i - hei / 2);
+    for (int i = 0; i < MAX_BLUR_PASS && ps->o.blur_kerns[i]; ++i) {
+      XFixed *kern = ps->o.blur_kerns[i];
+      if (!kern)
+        break;
+
+      glx_blur_pass_t *ppass = &ps->glx_blur_passes[i];
+
+      // Build shader
+      {
+        int wid = XFixedToDouble(kern[0]), hei = XFixedToDouble(kern[1]);
+        int nele = wid * hei - 1;
+        int len = strlen(FRAG_SHADER_BLUR_PREFIX) + strlen(sampler_type) + strlen(extension) + (strlen(shader_add) + strlen(texture_func) + 42) * nele + strlen(FRAG_SHADER_BLUR_SUFFIX) + strlen(texture_func) + 12 + 1;
+        char *shader_str = calloc(len, sizeof(char));
+        if (!shader_str) {
+          printf_errf("(): Failed to allocate %d bytes for shader string.", len);
+          return false;
+        }
+        {
+          char *pc = shader_str;
+          sprintf(pc, FRAG_SHADER_BLUR_PREFIX, extension, sampler_type);
           pc += strlen(pc);
           assert(strlen(shader_str) < len);
+
+          double sum = 0.0;
+          for (int j = 0; j < hei; ++j) {
+            for (int k = 0; k < wid; ++k) {
+              if (hei / 2 == j && wid / 2 == k)
+                continue;
+              double val = XFixedToDouble(kern[2 + j * wid + k]);
+              if (0.0 == val)
+                continue;
+              sum += val;
+              sprintf(pc, shader_add, val, texture_func, k - wid / 2, j - hei / 2);
+              pc += strlen(pc);
+              assert(strlen(shader_str) < len);
+            }
+          }
+
+          sprintf(pc, FRAG_SHADER_BLUR_SUFFIX, texture_func, sum);
+          assert(strlen(shader_str) < len);
         }
+        ppass->frag_shader = glx_create_shader(GL_FRAGMENT_SHADER, shader_str);
+        free(shader_str);
       }
 
-      sprintf(pc, FRAG_SHADER_BLUR_SUFFIX, texture_func, sum);
-      assert(strlen(shader_str) < len);
-#ifdef DEBUG_GLX_GLSL
-      fputs(shader_str, stdout);
-      fflush(stdout);
-#endif
-    }
-    ps->glx_frag_shader_blur = glx_create_shader(GL_FRAGMENT_SHADER, shader_str);
-    free(extension);
-    free(shader_str);
-  }
-
-  if (!ps->glx_frag_shader_blur) {
-    printf_errf("(): Failed to create fragment shader.");
-    return false;
-  }
+      if (!ppass->frag_shader) {
+        printf_errf("(): Failed to create fragment shader %d.", i);
+        return false;
+      }
 
-  ps->glx_prog_blur = glx_create_program(&ps->glx_frag_shader_blur, 1);
-  if (!ps->glx_prog_blur) {
-    printf_errf("(): Failed to create GLSL program.");
-    return false;
-  }
+      // Build program
+      ppass->prog = glx_create_program(&ppass->frag_shader, 1);
+      if (!ppass->prog) {
+        printf_errf("(): Failed to create GLSL program.");
+        return false;
+      }
 
+      // Get uniform addresses
 #define P_GET_UNIFM_LOC(name, target) { \
-  ps->target = glGetUniformLocation(ps->glx_prog_blur, name); \
-  if (ps->target < 0) { \
-    printf_errf("(): Failed to get location of uniform '" name "'. Might be troublesome."); \
-  } \
-}
+      ppass->target = glGetUniformLocation(ppass->prog, name); \
+      if (ppass->target < 0) { \
+        printf_errf("(): Failed to get location of %d-th uniform '" name "'. Might be troublesome.", i); \
+      } \
+    }
 
-  P_GET_UNIFM_LOC("factor_center", glx_prog_blur_unifm_factor_center);
-  if (!ps->o.glx_use_gpushader4) {
-    P_GET_UNIFM_LOC("offset_x", glx_prog_blur_unifm_offset_x);
-    P_GET_UNIFM_LOC("offset_y", glx_prog_blur_unifm_offset_y);
+      P_GET_UNIFM_LOC("factor_center", unifm_factor_center);
+      if (!ps->o.glx_use_gpushader4) {
+        P_GET_UNIFM_LOC("offset_x", unifm_offset_x);
+        P_GET_UNIFM_LOC("offset_y", unifm_offset_y);
+      }
+#undef P_GET_UNIFM_LOC
+    }
+    free(extension);
   }
 
-#undef P_GET_UNIFM_LOC
 
 #ifdef DEBUG_GLX_ERR
   glx_check_err(ps);
@@ -841,9 +872,9 @@ glx_set_clip(session_t *ps, XserverRegion reg, const reg_data_t *pcache_reg) {
   } \
   glBegin(GL_QUADS); \
  \
-  for (int i = 0; i < nrects; ++i) { \
+  for (int ri = 0; ri < nrects; ++ri) { \
     XRectangle crect; \
-    rect_crop(&crect, &rects[i], &rec_all); \
+    rect_crop(&crect, &rects[ri], &rec_all); \
  \
     if (!crect.width || !crect.height) \
       continue; \
@@ -856,21 +887,69 @@ glx_set_clip(session_t *ps, XserverRegion reg, const reg_data_t *pcache_reg) {
     cxfree(rects); \
   free_region(ps, &reg_new); \
 
+static inline GLuint
+glx_gen_texture(session_t *ps, GLenum tex_tgt, int width, int height) {
+  GLuint tex = 0;
+  glGenTextures(1, &tex);
+  if (!tex) return 0;
+  glEnable(tex_tgt);
+  glBindTexture(tex_tgt, tex);
+  glTexParameteri(tex_tgt, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+  glTexParameteri(tex_tgt, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
+  glTexParameteri(tex_tgt, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
+  glTexParameteri(tex_tgt, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
+  glTexImage2D(tex_tgt, 0, GL_RGB, width, height, 0, GL_RGB,
+      GL_UNSIGNED_BYTE, NULL);
+  glBindTexture(tex_tgt, 0);
+
+  return tex;
+}
+
+static inline void
+glx_copy_region_to_tex(session_t *ps, GLenum tex_tgt, int basex, int basey,
+    int dx, int dy, int width, int height) {
+  if (width > 0 && height > 0)
+    glCopyTexSubImage2D(tex_tgt, 0, dx - basex, dy - basey,
+        dx, ps->root_height - dy - height, width, height);
+}
+
+#ifdef CONFIG_VSYNC_OPENGL_GLSL
+/**
+ * Blur contents in a particular region.
+ */
 bool
 glx_blur_dst(session_t *ps, int dx, int dy, int width, int height, float z,
-    GLfloat factor_center, XserverRegion reg_tgt, const reg_data_t *pcache_reg) {
-  // Read destination pixels into a texture
-  GLuint tex_scr = 0;
-  glGenTextures(1, &tex_scr);
-  if (!tex_scr) {
-    printf_errf("(): Failed to allocate texture.");
-    return false;
-  }
+    GLfloat factor_center,
+    XserverRegion reg_tgt, const reg_data_t *pcache_reg,
+    glx_blur_cache_t *pbc) {
+  assert(ps->glx_blur_passes[0].prog);
+  const bool more_passes = ps->glx_blur_passes[1].prog;
+  const bool have_scissors = glIsEnabled(GL_SCISSOR_TEST);
+  const bool have_stencil = glIsEnabled(GL_STENCIL_TEST);
+  bool ret = false;
+
+  // Calculate copy region size
+  glx_blur_cache_t ibc = { .width = 0, .height = 0 };
+  if (!pbc)
+    pbc = &ibc;
 
   int mdx = dx, mdy = dy, mwidth = width, mheight = height;
+#ifdef DEBUG_GLX
+  printf_dbgf("(): %d, %d, %d, %d\n", mdx, mdy, mwidth, mheight);
+#endif
+
+  /*
   if (ps->o.resize_damage > 0) {
-    int inc_x = min_i(ps->o.resize_damage, XFixedToDouble(ps->o.blur_kern[0]) / 2),
-        inc_y = min_i(ps->o.resize_damage, XFixedToDouble(ps->o.blur_kern[1]) / 2);
+    int inc_x = 0, inc_y = 0;
+    for (int i = 0; i < MAX_BLUR_PASS; ++i) {
+      XFixed *kern = ps->o.blur_kerns[i];
+      if (!kern) break;
+      inc_x += XFixedToDouble(kern[0]) / 2;
+      inc_y += XFixedToDouble(kern[1]) / 2;
+    }
+    inc_x = min_i(ps->o.resize_damage, inc_x);
+    inc_y = min_i(ps->o.resize_damage, inc_y);
+
     mdx = max_i(dx - inc_x, 0);
     mdy = max_i(dy - inc_y, 0);
     int mdx2 = min_i(dx + width + inc_x, ps->root_width),
@@ -878,24 +957,57 @@ glx_blur_dst(session_t *ps, int dx, int dy, int width, int height, float z,
     mwidth = mdx2 - mdx;
     mheight = mdy2 - mdy;
   }
+  */
 
   GLenum tex_tgt = GL_TEXTURE_RECTANGLE;
   if (ps->glx_has_texture_non_power_of_two)
     tex_tgt = GL_TEXTURE_2D;
 
-  glEnable(tex_tgt);
-  glBindTexture(tex_tgt, tex_scr);
-  glTexParameteri(tex_tgt, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
-  glTexParameteri(tex_tgt, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
-  glTexParameteri(tex_tgt, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
-  glTexParameteri(tex_tgt, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
-  glTexImage2D(tex_tgt, 0, GL_RGB, mwidth, mheight, 0, GL_RGB, GL_UNSIGNED_BYTE, NULL);
-  glCopyTexSubImage2D(tex_tgt, 0, 0, 0, mdx, ps->root_height - mdy - mheight, mwidth, mheight);
+  // Free textures if size inconsistency discovered
+  if (mwidth != pbc->width || mheight != pbc->height)
+    free_glx_bc_resize(ps, pbc);
+
+  // Generate FBO and textures if needed
+  if (!pbc->textures[0])
+    pbc->textures[0] = glx_gen_texture(ps, tex_tgt, mwidth, mheight);
+  GLuint tex_scr = pbc->textures[0];
+  if (more_passes && !pbc->textures[1])
+    pbc->textures[1] = glx_gen_texture(ps, tex_tgt, mwidth, mheight);
+  pbc->width = mwidth;
+  pbc->height = mheight;
+  GLuint tex_scr2 = pbc->textures[1];
+#ifdef CONFIG_VSYNC_OPENGL_FBO
+  if (more_passes && !pbc->fbo)
+    glGenFramebuffers(1, &pbc->fbo);
+  const GLuint fbo = pbc->fbo;
+#endif
 
-#ifdef DEBUG_GLX
-  printf_dbgf("(): %d, %d, %d, %d\n", mdx, ps->root_height - mdy - mheight, mwidth, mheight);
+  if (!tex_scr || (more_passes && !tex_scr2)) {
+    printf_errf("(): Failed to allocate texture.");
+    goto glx_blur_dst_end;
+  }
+#ifdef CONFIG_VSYNC_OPENGL_FBO
+  if (more_passes && !fbo) {
+    printf_errf("(): Failed to allocate framebuffer.");
+    goto glx_blur_dst_end;
+  }
 #endif
 
+  // Read destination pixels into a texture
+  glEnable(tex_tgt);
+  glBindTexture(tex_tgt, tex_scr);
+  glx_copy_region_to_tex(ps, tex_tgt, mdx, mdy, mdx, mdy, mwidth, mheight);
+  /*
+  if (tex_scr2) {
+    glBindTexture(tex_tgt, tex_scr2);
+    glx_copy_region_to_tex(ps, tex_tgt, mdx, mdy, mdx, mdy, mwidth, dx - mdx);
+    glx_copy_region_to_tex(ps, tex_tgt, mdx, mdy, mdx, dy + height,
+        mwidth, mdy + mheight - dy - height);
+    glx_copy_region_to_tex(ps, tex_tgt, mdx, mdy, mdx, dy, dx - mdx, height);
+    glx_copy_region_to_tex(ps, tex_tgt, mdx, mdy, dx + width, dy,
+        mdx + mwidth - dx - width, height);
+  } */
+
   // Texture scaling factor
   GLfloat texfac_x = 1.0f, texfac_y = 1.0f;
   if (GL_TEXTURE_2D == tex_tgt) {
@@ -904,67 +1016,130 @@ glx_blur_dst(session_t *ps, int dx, int dy, int width, int height, float z,
   }
 
   // Paint it back
-  // Color negation for testing...
-  // glTexEnvf(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_COMBINE);
-  // glTexEnvf(GL_TEXTURE_ENV, GL_COMBINE_RGB, GL_REPLACE);
-  // glTexEnvf(GL_TEXTURE_ENV, GL_OPERAND0_RGB, GL_ONE_MINUS_SRC_COLOR);
+  if (more_passes) {
+    glDisable(GL_STENCIL_TEST);
+    glDisable(GL_SCISSOR_TEST);
+  }
 
-  glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
-#ifdef CONFIG_VSYNC_OPENGL_GLSL
-  glUseProgram(ps->glx_prog_blur);
-  if (ps->glx_prog_blur_unifm_offset_x >= 0)
-    glUniform1f(ps->glx_prog_blur_unifm_offset_x, texfac_x);
-  if (ps->glx_prog_blur_unifm_offset_y >= 0)
-    glUniform1f(ps->glx_prog_blur_unifm_offset_y, texfac_y);
-  if (ps->glx_prog_blur_unifm_factor_center >= 0)
-    glUniform1f(ps->glx_prog_blur_unifm_factor_center, factor_center);
+  bool last_pass = false;
+  for (int i = 0; !last_pass; ++i) {
+    last_pass = !ps->glx_blur_passes[i + 1].prog;
+    assert(i < MAX_BLUR_PASS - 1);
+    const glx_blur_pass_t *ppass = &ps->glx_blur_passes[i];
+    assert(ppass->prog);
+
+    assert(tex_scr);
+    glBindTexture(tex_tgt, tex_scr);
+
+#ifdef CONFIG_VSYNC_OPENGL_FBO
+    if (!last_pass) {
+      static const GLenum DRAWBUFS[2] = { GL_COLOR_ATTACHMENT0 };
+      glBindFramebuffer(GL_FRAMEBUFFER, fbo);
+      glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, tex_scr2, 0);
+      glDrawBuffers(1, DRAWBUFS);
+      if (glCheckFramebufferStatus(GL_FRAMEBUFFER)
+          != GL_FRAMEBUFFER_COMPLETE) {
+        printf_errf("(): Framebuffer attachment failed.");
+        goto glx_blur_dst_end;
+      }
+    }
+    else {
+      static const GLenum DRAWBUFS[2] = { GL_BACK };
+      glBindFramebuffer(GL_FRAMEBUFFER, 0);
+      glDrawBuffers(1, DRAWBUFS);
+      if (have_scissors)
+        glEnable(GL_SCISSOR_TEST);
+      if (have_stencil)
+        glEnable(GL_STENCIL_TEST);
+    }
 #endif
 
-  {
-    P_PAINTREG_START();
+    // Color negation for testing...
+    // glTexEnvf(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_COMBINE);
+    // glTexEnvf(GL_TEXTURE_ENV, GL_COMBINE_RGB, GL_REPLACE);
+    // glTexEnvf(GL_TEXTURE_ENV, GL_OPERAND0_RGB, GL_ONE_MINUS_SRC_COLOR);
+
+    glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
+    glUseProgram(ppass->prog);
+    if (ppass->unifm_offset_x >= 0)
+      glUniform1f(ppass->unifm_offset_x, texfac_x);
+    if (ppass->unifm_offset_y >= 0)
+      glUniform1f(ppass->unifm_offset_y, texfac_y);
+    if (ppass->unifm_factor_center >= 0)
+      glUniform1f(ppass->unifm_factor_center, factor_center);
+
     {
-      const GLfloat rx = (crect.x - mdx) * texfac_x;
-      const GLfloat ry = (mheight - (crect.y - mdy)) * texfac_y;
-      const GLfloat rxe = rx + crect.width * texfac_x;
-      const GLfloat rye = ry - crect.height * texfac_y;
-      const GLfloat rdx = crect.x;
-      const GLfloat rdy = ps->root_height - crect.y;
-      const GLfloat rdxe = rdx + crect.width;
-      const GLfloat rdye = rdy - crect.height;
+      P_PAINTREG_START();
+      {
+        const GLfloat rx = (crect.x - mdx) * texfac_x;
+        const GLfloat ry = (mheight - (crect.y - mdy)) * texfac_y;
+        const GLfloat rxe = rx + crect.width * texfac_x;
+        const GLfloat rye = ry - crect.height * texfac_y;
+        GLfloat rdx = crect.x - mdx;
+        GLfloat rdy = mheight - crect.y + mdy;
+        GLfloat rdxe = rdx + crect.width;
+        GLfloat rdye = rdy - crect.height;
+
+        if (last_pass) {
+          rdx = crect.x;
+          rdy = ps->root_height - crect.y;
+          rdxe = rdx + crect.width;
+          rdye = rdy - crect.height;
+        }
 
 #ifdef DEBUG_GLX
-      printf_dbgf("(): %f, %f, %f, %f -> %d, %d, %d, %d\n", rx, ry, rxe, rye, rdx, rdy, rdxe, rdye);
+        printf_dbgf("(): %f, %f, %f, %f -> %f, %f, %f, %f\n", rx, ry, rxe, rye, rdx, rdy, rdxe, rdye);
 #endif
 
-      glTexCoord2f(rx, ry);
-      glVertex3f(rdx, rdy, z);
+        glTexCoord2f(rx, ry);
+        glVertex3f(rdx, rdy, z);
 
-      glTexCoord2f(rxe, ry);
-      glVertex3f(rdxe, rdy, z);
+        glTexCoord2f(rxe, ry);
+        glVertex3f(rdxe, rdy, z);
 
-      glTexCoord2f(rxe, rye);
-      glVertex3f(rdxe, rdye, z);
+        glTexCoord2f(rxe, rye);
+        glVertex3f(rdxe, rdye, z);
 
-      glTexCoord2f(rx, rye);
-      glVertex3f(rdx, rdye, z);
+        glTexCoord2f(rx, rye);
+        glVertex3f(rdx, rdye, z);
+      }
+      P_PAINTREG_END();
+    }
+
+    glUseProgram(0);
+
+    // Swap tex_scr and tex_scr2
+    {
+      GLuint tmp = tex_scr2;
+      tex_scr2 = tex_scr;
+      tex_scr = tmp;
     }
-    P_PAINTREG_END();
   }
 
-#ifdef CONFIG_VSYNC_OPENGL_GLSL
-  glUseProgram(0);
+  ret = true;
+
+glx_blur_dst_end:
+#ifdef CONFIG_VSYNC_OPENGL_FBO
+  glBindFramebuffer(GL_FRAMEBUFFER, 0);
 #endif
-  glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
   glBindTexture(tex_tgt, 0);
-  glDeleteTextures(1, &tex_scr);
   glDisable(tex_tgt);
+  if (have_scissors)
+    glEnable(GL_SCISSOR_TEST);
+  if (have_stencil)
+    glEnable(GL_STENCIL_TEST);
+
+  if (&ibc == pbc) {
+    free_glx_bc(ps, pbc);
+  }
 
 #ifdef DEBUG_GLX_ERR
   glx_check_err(ps);
 #endif
 
-  return true;
+  return ret;
 }
+#endif
 
 bool
 glx_dim_dst(session_t *ps, int dx, int dy, int width, int height, float z,
@@ -1156,7 +1331,7 @@ glx_render(session_t *ps, const glx_texture_t *ptex,
       }
 
 #ifdef DEBUG_GLX
-      printf_dbgf("(): Rect %d: %f, %f, %f, %f -> %d, %d, %d, %d\n", i, rx, ry, rxe, rye, rdx, rdy, rdxe, rdye);
+      printf_dbgf("(): Rect %d: %f, %f, %f, %f -> %d, %d, %d, %d\n", ri, rx, ry, rxe, rye, rdx, rdy, rdxe, rdye);
 #endif
 
 #define P_TEXCOORD(cx, cy) { \
@@ -1318,6 +1493,11 @@ glx_swap_copysubbuffermesa(session_t *ps, XserverRegion reg) {
 #ifdef CONFIG_VSYNC_OPENGL_GLSL
 GLuint
 glx_create_shader(GLenum shader_type, const char *shader_str) {
+#ifdef DEBUG_GLX_GLSL
+  printf("glx_create_shader(): ===\n%s\n===\n", shader_str);
+  fflush(stdout);
+#endif
+
   bool success = false;
   GLuint shader = glCreateShader(shader_type);
   if (!shader) {