Improvement: --glx-use-copysubbuffermesa

- GLX backend: Add --glx-use-copysubbuffermesa, to use
  MESA_copy_sub_buffer to do partial screen update. Huge performance
  boost on mesa drivers for partial screen updates, but does not work
  for nvidia-drivers and may break VSync. Automagically overrides
  --glx-copy-from-front.

- Add rect_is_fullscreen() to reuse code. Misc changes.

4 files changed, 84 insertions, 12 deletions

diff --git a/common.h b/common.h
index 977643b60..2f8b6e20d 100644
--- a/common.h
+++ b/common.h
@@ -284,6 +284,8 @@ typedef int (*f_SwapIntervalSGI) (int interval);
 typedef void (*f_BindTexImageEXT) (Display *display, GLXDrawable drawable, int buffer, const int *attrib_list);
 typedef void (*f_ReleaseTexImageEXT) (Display *display, GLXDrawable drawable, int buffer);
 
+typedef void (*f_CopySubBuffer) (Display *dpy, GLXDrawable drawable, int x, int y, int width, int height);
+
 /// @brief Wrapper of a GLX FBConfig.
 typedef struct {
   GLXFBConfig cfg;
@@ -344,6 +346,8 @@ typedef struct {
   bool glx_no_stencil;
   /// Whether to copy unmodified regions from front buffer.
   bool glx_copy_from_front;
+  /// Whether to use glXCopySubBufferMESA() to update screen.
+  bool glx_use_copysubbuffermesa;
   /// Whether to try to detect WM windows and mark them as focused.
   bool mark_wmwin_focused;
   /// Whether to mark override-redirect windows as focused.
@@ -624,6 +628,8 @@ typedef struct {
   f_BindTexImageEXT glXBindTexImageProc;
   /// Pointer to glXReleaseTexImageEXT function.
   f_ReleaseTexImageEXT glXReleaseTexImageProc;
+  /// Pointer to glXCopySubBufferMESA function.
+  f_CopySubBuffer glXCopySubBufferProc;
   /// FBConfig-s for GLX pixmap of different depths.
   glx_fbconfig_t *glx_fbconfigs[OPENGL_MAX_DEPTH + 1];
 #ifdef CONFIG_VSYNC_OPENGL_GLSL
@@ -1491,6 +1497,15 @@ free_region(session_t *ps, XserverRegion *p) {
 }
 
 /**
+ * Check if a rectangle includes the whole screen.
+ */
+static inline bool
+rect_is_fullscreen(session_t *ps, int x, int y, unsigned wid, unsigned hei) {
+  return (x <= 0 && y <= 0
+      && (x + wid) >= ps->root_width && (y + hei) >= ps->root_height);
+}
+
+/**
  * Determine if a window has a specific property.
  *
  * @param ps current session
@@ -1621,6 +1636,9 @@ glx_render(session_t *ps, const glx_texture_t *ptex,
     int x, int y, int dx, int dy, int width, int height, int z,
     double opacity, bool neg, XserverRegion reg_tgt);
 
+void
+glx_swap_copysubbuffermesa(session_t *ps, XserverRegion reg);
+
 #ifdef CONFIG_VSYNC_OPENGL_GLSL
 GLuint
 glx_create_shader(GLenum shader_type, const char *shader_str);
diff --git a/compton.c b/compton.c
index 9e8623ea8..811bbf17e 100644
--- a/compton.c
+++ b/compton.c
@@ -1736,7 +1736,6 @@ paint_all(session_t *ps, XserverRegion region, win *t) {
   }
 
   // Free up all temporary regions
-  XFixesDestroyRegion(ps->dpy, region);
   XFixesDestroyRegion(ps->dpy, reg_tmp);
   XFixesDestroyRegion(ps->dpy, reg_tmp2);
 
@@ -1783,7 +1782,10 @@ paint_all(session_t *ps, XserverRegion region, win *t) {
       break;
 #ifdef CONFIG_VSYNC_OPENGL
     case BKEND_GLX:
-      glXSwapBuffers(ps->dpy, get_tgt_window(ps));
+      if (ps->o.glx_use_copysubbuffermesa)
+        glx_swap_copysubbuffermesa(ps, region);
+      else
+        glXSwapBuffers(ps->dpy, get_tgt_window(ps));
       break;
 #endif
     default:
@@ -1802,6 +1804,8 @@ paint_all(session_t *ps, XserverRegion region, win *t) {
   }
 #endif
 
+  XFixesDestroyRegion(ps->dpy, region);
+
 #ifdef DEBUG_REPAINT
   print_timestamp(ps);
   struct timespec now = get_time_timespec();
@@ -4151,9 +4155,14 @@ usage(void) {
     "  negative effect on performance. (My test shows a 10% slowdown.)\n"
     "--glx-copy-from-front\n"
     "  GLX backend: Copy unmodified regions from front buffer instead of\n"
-    "  redrawing them all. My tests show a 10% decrease in performance\n"
-    "  when the whole screen is modified, but a 20% increase when only 1/4\n"
-    "  is, so this optimization is not enabled by default.\n"
+    "  redrawing them all. My tests with nvidia-drivers show a 10% decrease\n"
+    "  in performance when the whole screen is modified, but a 20% increase\n"
+    "  when only 1/4 is. My tests on nouveau show terrible slowdown.\n"
+    "--glx-use-copysubbuffermesa\n"
+    "  GLX backend: Use MESA_copy_sub_buffer to do partial screen update.\n"
+    "  My tests on nouveau shows a 200% performance boost when only 1/4 of\n"
+    "  the screen is updated. May break VSync and is not available on some\n"
+    "  drivers. Overrides --glx-copy-from-front.\n"
 #undef WARNING
 #ifndef CONFIG_DBUS
 #define WARNING WARNING_DISABLED
@@ -4620,6 +4629,7 @@ get_cfg(session_t *ps, int argc, char *const *argv, bool first_pass) {
     { "glx-copy-from-front", no_argument, NULL, 292 },
     { "benchmark", required_argument, NULL, 293 },
     { "benchmark-wid", required_argument, NULL, 294 },
+    { "glx-use-copysubbuffermesa", no_argument, NULL, 295 },
     // Must terminate with a NULL entry
     { NULL, 0, NULL, 0 },
   };
@@ -4898,6 +4908,10 @@ get_cfg(session_t *ps, int argc, char *const *argv, bool first_pass) {
         // --benchmark-wid
         ps->o.benchmark_wid = strtol(optarg, NULL, 0);
         break;
+      case 295:
+        // --glx-use-copysubbuffermesa
+        ps->o.glx_use_copysubbuffermesa = true;
+        break;
       default:
         usage();
         break;
diff --git a/compton.h b/compton.h
index c16c59612..b590b0ead 100644
--- a/compton.h
+++ b/compton.h
@@ -427,10 +427,8 @@ dump_drawable(session_t *ps, Drawable drawable) {
  */
 static inline bool
 win_is_fullscreen(session_t *ps, const win *w) {
-  return (w->a.x <= 0 && w->a.y <= 0
-      && (w->a.x + w->widthb) >= ps->root_width
-      && (w->a.y + w->heightb) >= ps->root_height
-      && !w->bounding_shaped);
+  return rect_is_fullscreen(ps, w->a.x, w->a.y, w->widthb, w->heightb)
+      && !w->bounding_shaped;
 }
 
 static void
diff --git a/opengl.c b/opengl.c
index a0bed63bb..3f1f419b7 100644
--- a/opengl.c
+++ b/opengl.c
@@ -96,6 +96,15 @@ glx_init(session_t *ps, bool need_render) {
       printf_errf("(): Failed to acquire glXBindTexImageEXT() / glXReleaseTexImageEXT().");
       goto glx_init_end;
     }
+
+    if (ps->o.glx_use_copysubbuffermesa) {
+      ps->glXCopySubBufferProc = (f_CopySubBuffer)
+        glXGetProcAddress((const GLubyte *) "glXCopySubBufferMESA");
+      if (!ps->glXCopySubBufferProc) {
+        printf_errf("(): Failed to acquire glXCopySubBufferMESA().");
+        goto glx_init_end;
+      }
+    }
   }
 
   // Acquire FBConfigs
@@ -543,10 +552,13 @@ void
 glx_paint_pre(session_t *ps, XserverRegion *preg) {
   ps->glx_z = 0.0;
   // glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
+
   // OpenGL doesn't support partial repaint without GLX_MESA_copy_sub_buffer,
-  // we currently redraw the whole screen or copy unmodified pixels from
+  // we could redraw the whole screen or copy unmodified pixels from
   // front buffer with --glx-copy-from-front.
-  if (!ps->o.glx_copy_from_front || !*preg) {
+  if (ps->o.glx_use_copysubbuffermesa || !*preg) {
+  }
+  else if (!ps->o.glx_copy_from_front) {
     free_region(ps, preg);
   }
   else {
@@ -612,7 +624,6 @@ glx_set_clip(session_t *ps, XserverRegion reg) {
     glDepthMask(GL_FALSE);
     glStencilOp(GL_REPLACE, GL_KEEP, GL_KEEP);
 
-
     glBegin(GL_QUADS);
 
     for (int i = 0; i < nrects; ++i) {
@@ -853,6 +864,37 @@ glx_render(session_t *ps, const glx_texture_t *ptex,
   return true;
 }
 
+/**
+ * Swap buffer with glXCopySubBufferMESA().
+ */
+void
+glx_swap_copysubbuffermesa(session_t *ps, XserverRegion reg) {
+  int nrects = 0;
+  XRectangle *rects = XFixesFetchRegion(ps->dpy, reg, &nrects);
+
+  if (1 == nrects && rect_is_fullscreen(ps, rects[0].x, rects[0].y,
+        rects[0].width, rects[0].height)) {
+    glXSwapBuffers(ps->dpy, get_tgt_window(ps));
+  }
+  else {
+    glx_set_clip(ps, None);
+    for (int i = 0; i < nrects; ++i) {
+      const int x = rects[i].x;
+      const int y = ps->root_height - rects[i].y - rects[i].height;
+      const int wid = rects[i].width;
+      const int hei = rects[i].height;
+
+#ifdef DEBUG_GLX
+      printf_dbgf("(): %d, %d, %d, %d\n", x, y, wid, hei);
+#endif
+      ps->glXCopySubBufferProc(ps->dpy, get_tgt_window(ps), x, y, wid, hei);
+    }
+  }
+
+  if (rects)
+    XFree(rects);
+}
+
 #ifdef CONFIG_VSYNC_OPENGL_GLSL
 GLuint
 glx_create_shader(GLenum shader_type, const char *shader_str) {