Dry-snow slab avalanche release can be separated in four distinct phases : (i) failure initiation in a weak snow layer, (ii) onset, (iii) dynamics of crack propagation in the weak layer and eventually (iv)slab release. While a lot has been done to study the first two phases, less is known about dynamic crack propagation, especially at slope scale.In this study, we used the Material Point Method and elastoplasticity to simulate the dynamics of20 m long Centered Propagation Saw Tests (PST). We improved the recent constitutive snow model of Gaume et al. (2018) by developing a new softening law based on the total plastic deformation (volumetric and deviatoric parts) to account for damages caused by shearing the weak layer. Interestingly, several regimes of propagations are observed depending on slope angle. For slope angles smaller than the friction angle (θ<φ), crack propagates faster in the downstream direction than in the upstream one. The propagation speed increases with slope angle and appear closely related to the bending mechanism which sustains the propagation. For slope angles higher than the friction angle (θ>φ), a sharp transition is observed once the crack reaches a critical length (lf). We interpret this transition as a change from slab bending to slab tension due to the increasing load in the downstream direction. An estimation of this critical length (lf) is proposed using a basic analytical shear model with residual friction similar to the one developped by McClung in 1979. The crack speed seems only bounded by the elastic wave speed in the slab. By this study, we explain the gap between propagation speeds based on 2 m PSTs and slope scale observations. Finally, our results bring back to light the pioneer work of McClung which appears sufficient to model slab avalanche release on steep terrain.