i965: Cope with batch getting flushed in the middle of batchbuffer emits.

This isn't required for GEM (at least, yet), but the check_aperture code
for non-GEM results in batch getting flushed during emit.  brw_state_upload
restarts state emits, but a bunch of the state emit functions were assuming
that they would be called exactly once, after prepare and before new_batch.

Bug #17179.

src/mesa/drivers/dri/i965/brw_context.h

@@ -557,6 +557,11 @@ struct brw_context
 
       GLfloat *last_buf;
       GLuint last_bufsz;
+      /**
+       *  Whether we should create a new bo instead of reusing the old one
+       * (if we just dispatch the batch pointing at the old one.
+       */
+      GLboolean need_new_bo;
    } curbe;
 
    struct {

src/mesa/drivers/dri/i965/brw_curbe.c

@@ -282,7 +282,8 @@ static void prepare_constant_buffer(struct brw_context *brw)
       brw->curbe.last_bufsz = bufsz;
 
       if (brw->curbe.curbe_bo != NULL &&
-	  brw->curbe.curbe_next_offset + bufsz > brw->curbe.curbe_bo->size)
+	  (brw->curbe.need_new_bo ||
+	   brw->curbe.curbe_next_offset + bufsz > brw->curbe.curbe_bo->size))
       {
 	 dri_bo_unreference(brw->curbe.curbe_bo);
 	 brw->curbe.curbe_bo = NULL;

src/mesa/drivers/dri/i965/brw_draw.c

@@ -156,11 +156,13 @@ static void brw_emit_prim( struct brw_context *brw,
 static void brw_merge_inputs( struct brw_context *brw,
 		       const struct gl_client_array *arrays[])
 {
-   struct brw_vertex_element *inputs = brw->vb.inputs;
    struct brw_vertex_info old = brw->vb.info;
    GLuint i;
 
-   memset(inputs, 0, sizeof(*inputs));
+   for (i = 0; i < VERT_ATTRIB_MAX; i++)
+      dri_bo_unreference(brw->vb.inputs[i].bo);
+
+   memset(&brw->vb.inputs, 0, sizeof(brw->vb.inputs));
    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 
    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
@@ -171,7 +173,8 @@ static void brw_merge_inputs( struct brw_context *brw,
 	 if (arrays[i]->StrideB != 0)
 	    brw->vb.info.varying |= 1 << i;
 
-	 brw->vb.info.sizes[i/16] |= (inputs[i].glarray->Size - 1) << ((i%16) * 2);
+	 brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
+	    ((i%16) * 2);
       }
    }
 

src/mesa/drivers/dri/i965/brw_draw_upload.c

@@ -455,12 +455,6 @@ static void brw_emit_vertices(struct brw_context *brw)
 		input->offset);
       OUT_BATCH(brw->vb.max_index);
       OUT_BATCH(0); /* Instance data step rate */
-
-      /* Unreference the buffer so it can get freed, now that we won't
-       * touch it any more.
-       */
-      dri_bo_unreference(input->bo);
-      input->bo = NULL;
    }
    ADVANCE_BATCH();
 

src/mesa/drivers/dri/i965/brw_vtbl.c

@@ -97,8 +97,7 @@ static void brw_new_batch( struct intel_context *intel )
    /* Check that we didn't just wrap our batchbuffer at a bad time. */
    assert(!brw->no_batch_wrap);
 
-   dri_bo_unreference(brw->curbe.curbe_bo);
-   brw->curbe.curbe_bo = NULL;
+   brw->curbe.need_new_bo = GL_TRUE;
 
    /* Mark all context state as needing to be re-emitted.
     * This is probably not as severe as on 915, since almost all of our state