For all TileDrawQuads within the same layer, they share a common
device_layer_quad and use_aa. So caching device_layer_quad and use_aa save
the effort to do the math and aa decision making for the coming TileDrawQuads
in the same layer.

Because TileDrawQuads from the same layer share the same SharedQuadState. We can
cache previous_shared_quad_state_ to decide if layer switches.
 

For all TileDrawQuads within the same layer, they share a common
device_layer_quad and use_aa. So caching device_layer_quad and use_aa save
the effort to do the math and aa decision making for the coming TileDrawQuads
in the same layer.

Because TileDrawQuads from the same layer share the same SharedQuadState. We can
cache previous_shared_quad_state_ to decide if layer switches.


BTW, one thing confused me is, when calculating ShouldAntialiasQuad, device_layer_quad is used. But why not use quad's rect or visible rect instead?