Recent discoveries of dynamic ice VII and superionic ice highlight the importance of ionic diffusions in discriminating high-pressure (P) water phases. The rare event nature and the chemical bond breaking associated with these diffusions, however, make extensive simulations of these processes unpractical to ab initio and inappropriate for force field based methods. Using a first-principles neural network potential, we performed a theoretical study of water at 5–70 GPa and 300–3000 K. Long-time dynamics of protons and oxygens were found indispensable in discriminating several subtle states of water, characterized by proton’s and oxygen ion’s diffusion coefficients and the distribution of proton’s displacements. Within dynamic ice VII, two types of proton transfer mechanisms, i.e., translational and rotational transfers, were identified to discriminate this region further into dynamic ice VII T and dynamic ice VII R. The triple point between ice VII, superionic ice (SI), and liquid exists because the loosening of the bcc oxygen skeleton is prevented by the decrease of interatomic distances at high P’s. The melting of ice VII above ∼40 GPa can be understood as a process of two individual steps: the melting of protons and the retarded melting of oxygens, responsible for the forming of SI. The boundary of the dynamic ice VII and SI lies on the continuation line ice VII’s melting curve at low P’s. Based on these, a detailed phase diagram is given, which may shed light on studies of water under P’s in a wide range of interdisciplinary sciences.