Accelerating Eulerian Fluid Simulation With Convolutional Networks