ABSTRACT Regulating the available gas mass inside galaxies proceeds through a delicate balance between inflows and outflows, but also through the internal depletion of gas due to star formation. At the same time, stellar feedback is the internal engine that powers the strong outflows. Since star formation and stellar feedback are both small-scale phenomena, we need a realistic and predictive subgrid model for both. We describe the implementation of supernova momentum feedback and star formation based on the turbulence of the gas in the ramses code. For star formation, we adopt the so-called multifreefall model. The resulting star formation efficiencies can be significantly smaller or bigger than the traditionally chosen value of $1\, {\rm per\, cent}$. We apply these new numerical models to a prototype cosmological simulation of a massive halo that features a major merger which results in the formation of an early-type galaxy without using AGN feedback. We find that the feedback model provides the first-order mechanism for regulating the stellar and baryonic content in our simulated galaxy. At high redshift, the merger event pushes gas to large densities and large turbulent velocity dispersions, such that efficiencies come close to $10\, {\rm per\, cent}$, resulting in large star formation rate (SFR). We find small molecular gas depletion time during the starburst, in perfect agreement with observations. Furthermore, at late times, the galaxy becomes quiescent with efficiencies significantly smaller than $1\, {\rm per\, cent}$, resulting in small SFR and long molecular gas depletion time.