This work is devoted to the performance modeling of a vectorized multi-threaded implementation of a compressible hydrodynamics solver on current multicore CPUs. The underlying explicit time-advance scheme is based on a Lagrange+remap discretization. The algorithm consists in nine elementary kernels that exhibit, for most of them, a stencil update pattern. Using the Execution Cache Memory (ECM) model — a refinement of the Roofline model —, we devise a high fidelity performance model, allowing us to finely understand and overcome bottlenecks of the algorithm and accurately predict global code performance.