[Mesa-dev] [PATCH 2/2] glx: Make glXWaitGL actually flush the GL command stream.

[Eric Anholt]

We were only flushing it in the fake-front case, which won't be enough if
you're doing something like drawing into a GLXPixmap and using X rendering
to composite from it.
---
 src/glx/dri2_glx.c | 28 ++++++++++++++++++++++++----
 1 file changed, 24 insertions(+), 4 deletions(-)

diff --git a/src/glx/dri2_glx.c b/src/glx/dri2_glx.c
index 3cdd249..d813788 100644
--- a/src/glx/dri2_glx.c
+++ b/src/glx/dri2_glx.c
@@ -664,10 +664,30 @@ dri2_wait_gl(struct glx_context *gc)
    struct dri2_drawable *priv = (struct dri2_drawable *)
       GetGLXDRIDrawable(gc->currentDpy, gc->currentDrawable);
 
-   if (priv == NULL || !priv->have_fake_front)
-      return;
-
-   dri2_copy_drawable(priv, DRI2BufferFrontLeft, DRI2BufferFakeFrontLeft);
+   if (priv != NULL) {
+      if (priv->have_fake_front) {
+         dri2_copy_drawable(priv, DRI2BufferFrontLeft, DRI2BufferFakeFrontLeft);
+      } else {
+         struct dri2_screen *psc = (struct dri2_screen *) priv->base.psc;
+
+         /* From the GLX 1.4 spec, page 33:
+          *
+          *     "OpenGL calls made prior to glXWaitGL are guaranteed to be
+          *      executed before X rendering calls made after glXWaitGL. While
+          *      the same result can be achieved using glFinish, glXWaitGL
+          *      does not require a round trip to the server, and is therefore
+          *      more efficient in cases where the client and server are on
+          *      separate machines."
+          *
+          * In DRI2, we assume that the kernel serializes operations on BOs
+          * appropriately, so a glFlush here that gets the GL commands down to
+          * the kernel is enough to ensure that later X rendering comes after
+          * it.
+          */
+         if (psc->f)
+            (*psc->f->flush) (priv->driDrawable);
+      }
+   }
 }
 
 /**
-- 
1.8.3.rc0