Make sure MI_WAIT packets are emitted for SWZ

Untested...

6 files changed, 75 insertions, 36 deletions

diff --git a/src/mesa/drivers/dri/i915tex/intel_batchbuffer.h b/src/mesa/drivers/dri/i915tex/intel_batchbuffer.h
index dfeb68364a..c2c1dee9bb 100644
--- a/src/mesa/drivers/dri/i915tex/intel_batchbuffer.h
+++ b/src/mesa/drivers/dri/i915tex/intel_batchbuffer.h
@@ -149,6 +149,31 @@ intel_batchbuffer_require_space(struct intel_batchbuffer *batch,
    batch->flags |= flags;
 }
 
+static INLINE GLubyte *
+intel_batchbuffer_get_space(struct intel_batchbuffer *batch, 
+			    GLuint segment,
+			    GLuint bytes, 
+			    GLuint flags )
+{
+   GLubyte *ptr;
+
+   intel_batchbuffer_require_space( batch, segment, bytes, flags );
+   ptr = batch->map + batch->segment_finish_offset[segment];
+
+   batch->segment_finish_offset[segment] += bytes;
+   return ptr;
+}
+
+
+static INLINE void
+intel_batchbuffer_put_back_space(struct intel_batchbuffer *batch, 
+			    GLuint segment,
+			    GLuint bytes )
+{
+   batch->segment_finish_offset[segment] -= bytes;
+}
+
+
 /* Here are the crusty old macros, to be removed:
  */
 #define BATCH_LOCALS
diff --git a/src/mesa/drivers/dri/i915tex/intel_classic.c b/src/mesa/drivers/dri/i915tex/intel_classic.c
index ceb17b84af..84ff341f79 100644
--- a/src/mesa/drivers/dri/i915tex/intel_classic.c
+++ b/src/mesa/drivers/dri/i915tex/intel_classic.c
@@ -239,15 +239,7 @@ static void classic_start_render( struct clip_render *render,
    /* Start a new batchbuffer, emit wait for pending flip.
     */
    if (start_of_frame)
-      intel_wait_flips(crc->intel, 0);
-
-#if 0
-   /* Flush any non-cliprect batch now:
-    */
-   intel_batchbuffer_require_space( crc->intel->batch, 0, 
-				    SEGMENT_IMMEDIATE,
-				    INTEL_BATCH_CLIPRECTS );
-#endif
+      intel_wait_flips_batch(crc->intel, GL_FALSE);
 
    intel_update_software_state( crc->intel );
 }
diff --git a/src/mesa/drivers/dri/i915tex/intel_swapbuffers.c b/src/mesa/drivers/dri/i915tex/intel_swapbuffers.c
index b60ebcc9f5..61ae7e2e2d 100644
--- a/src/mesa/drivers/dri/i915tex/intel_swapbuffers.c
+++ b/src/mesa/drivers/dri/i915tex/intel_swapbuffers.c
@@ -397,8 +397,9 @@ intelCopySubBuffer(__DRIdrawablePrivate * dPriv, int x, int y, int w, int h)
 
 
 /* Emit wait for pending flips */
-void
-intel_wait_flips(struct intel_context *intel, GLuint batch_flags)
+GLboolean
+intel_emit_wait_flips(struct intel_context *intel, 
+		      GLuint **ptr )
 {
    struct intel_framebuffer *intel_fb =
       (struct intel_framebuffer *) intel->ctx.DrawBuffer;
@@ -413,30 +414,49 @@ intel_wait_flips(struct intel_context *intel, GLuint batch_flags)
        intel_fb->pf_pipes) 
    {
       GLint pf_pipes = intel_fb->pf_pipes;
-      BATCH_LOCALS;
-
       /* Wait for pending flips to take effect */
-      BEGIN_BATCH(2, batch_flags);
-      OUT_BATCH(pf_pipes & 0x1 ? (MI_WAIT_FOR_EVENT | MI_WAIT_FOR_PLANE_A_FLIP)
-		: 0);
-      OUT_BATCH(pf_pipes & 0x2 ? (MI_WAIT_FOR_EVENT | MI_WAIT_FOR_PLANE_B_FLIP)
-		: 0);
-      ADVANCE_BATCH();
+      
+      (*ptr)[0] = ( (pf_pipes & 0x1) ? 
+		    (MI_WAIT_FOR_EVENT | MI_WAIT_FOR_PLANE_A_FLIP) :
+		    0 );
+
+      (*ptr)[1] = ( (pf_pipes & 0x2) ? 
+		    (MI_WAIT_FOR_EVENT | MI_WAIT_FOR_PLANE_B_FLIP) :
+		    0 );
+
+      *ptr += 2;
 
       intel_rb->pf_pending--;
+      return GL_TRUE;
    }
+
+   return GL_FALSE;
 }
 
+
+#define INTEL_WAIT_FLIP_BYTES (2*sizeof(int))
  
-void intel_do_wait_flips( struct intel_context *intel ) 
+void intel_wait_flips_batch( struct intel_context *intel,
+			     GLboolean do_flush ) 
 {
-   intel_wait_flips( intel, 0 );
+   GLuint bytes = INTEL_WAIT_FLIP_BYTES;
+   GLuint *ptr = (GLuint *)intel_batchbuffer_get_space( intel->batch, 
+							0, 
+							bytes,
+							0 );
 
-   /* Flush the batchbuffer.  Later call to intelSpanRenderStart will
-    * ensure we wait for completion.
-    */
-   if (intel->batch->segment_finish_offset[0] != 0)
-      intel_batchbuffer_flush(intel->batch, GL_TRUE);
-
-   intel_batchbuffer_wait_last_fence( intel->batch );
+   if (intel_emit_wait_flips( intel, &ptr ))
+   {
+      /* Flush the batchbuffer.  Later call to intelSpanRenderStart will
+       * ensure we wait for completion.
+       */
+      if (do_flush) {
+	 intel_batchbuffer_flush(intel->batch, GL_TRUE);
+	 intel_batchbuffer_wait_last_fence( intel->batch );
+      }
+   }
+   else {
+      intel_batchbuffer_put_back_space( intel->batch, 0,
+					INTEL_WAIT_FLIP_BYTES );
+   }
 }
diff --git a/src/mesa/drivers/dri/i915tex/intel_swapbuffers.h b/src/mesa/drivers/dri/i915tex/intel_swapbuffers.h
index acd5ee262d..b938b691b2 100644
--- a/src/mesa/drivers/dri/i915tex/intel_swapbuffers.h
+++ b/src/mesa/drivers/dri/i915tex/intel_swapbuffers.h
@@ -28,10 +28,11 @@
 #ifndef INTEL_SWAPBUFFERS_H
 #define INTEL_SWAPBUFFERS_H
 
-extern void intel_wait_flips(struct intel_context *intel, GLuint batch_flags);
-
 extern void intelSwapBuffers(__DRIdrawablePrivate * dPriv);
 
-extern void intel_do_wait_flips( struct intel_context *intel );
+GLboolean intel_emit_wait_flips(struct intel_context *intel, GLuint **ptr );
+
+void intel_wait_flips_batch( struct intel_context *intel,
+			     GLboolean do_flush );
 
 #endif
diff --git a/src/mesa/drivers/dri/i915tex/intel_swrast.c b/src/mesa/drivers/dri/i915tex/intel_swrast.c
index 9e906cb3b8..78e48a7278 100644
--- a/src/mesa/drivers/dri/i915tex/intel_swrast.c
+++ b/src/mesa/drivers/dri/i915tex/intel_swrast.c
@@ -415,7 +415,7 @@ static void swrender_start_render( struct clip_render *render,
    /* Wait for pending flip.
     */
    if (start_of_frame)
-      intel_do_wait_flips(intel);
+      intel_wait_flips_batch(intel, GL_TRUE);
 
    /* Wait for last fence to clear:
     */
diff --git a/src/mesa/drivers/dri/i915tex/intel_swz.c b/src/mesa/drivers/dri/i915tex/intel_swz.c
index 32cd2c2605..dcca4c7cea 100644
--- a/src/mesa/drivers/dri/i915tex/intel_swz.c
+++ b/src/mesa/drivers/dri/i915tex/intel_swz.c
@@ -197,10 +197,6 @@ static void swz_start_render( struct clip_render *render,
 		       swz->zone_height);
 
 
-   /* Goes to the main batchbuffer: 
-    */
-//   intel_wait_flips(intel, INTEL_BATCH_NO_CLIPRECTS);
-
    memcpy( swz->last_driver_state,
 	   intel->state.current,
 	   intel->state.driver_state_size );
@@ -305,6 +301,11 @@ static void swz_flush( struct clip_render *render,
 		 swz->nr_vertices * swz->vertex_stride );
 
 
+      /* Add wait-for-flip code to the preamble:
+       */
+      intel_emit_wait_flips( intel, 
+			     (GLuint **)&swz->pre_post.ptr );
+
       zone_mi_flush( &swz->pre_post, 0 );
 
       /* Emit preamble - tweak cachemode0: