Don't reemit hardware state *every* primitive.

src/mesa/pipe/i915simple/i915_context.h

    struct pipe_mipmap_tree *texture[PIPE_MAX_SAMPLERS];
    struct pipe_viewport_state viewport;
    GLuint dirty;
+   GLuint hw_dirty;
 
    GLuint *batch_start;
 

src/mesa/pipe/i915simple/i915_prim_emit.c

    unsigned *ptr;
    unsigned i;
 
-   i915_emit_hardware_state( i915 );
+   if (i915->dirty)
+      i915_update_derived( i915 );
+
+   if (i915->hw_dirty)
+      i915_emit_hardware_state( i915 );
 
    ptr = winsys->batch_start( winsys, nr * vertex_size, 0 );
    if (ptr == 0) {

src/mesa/pipe/i915simple/i915_state_derived.c

 		      I915_NEW_FRAMEBUFFER))
       compute_cliprect(i915);
 
+   if (i915->dirty)
+      i915->hw_dirty = 1;
+
    i915->dirty = 0;
 }

src/mesa/pipe/i915simple/i915_state_emit.c

       for (i = 0; i < dwords; i++)
 	 OUT_BATCH( prog[i] );
    }
+
+   i915->hw_dirty = 0;
 }