The scheduling of parallel real-time tasks enables the efficient utilization of modern multiprocessor platforms for systems with real-time constrains. In this situation, the gang task model, in which each parallel sub-job has to be executed simultaneously, has shown significant performance benefits due to reduced context switches and more efficient intra-task synchronization. In this paper, we provide the first schedulability analysis for sporadic constrained-deadline gang task systems and propose a novel stationary gang scheduling algorithm. We show that the schedulability problem of gang task sets can be reduced to the uniprocessor self-suspension schedulability problem. Furthermore, we provide a class of partitioning algorithms to find a stationary gang assignment and show that it bounds the worst-case interference of each task. To demonstrate the effectiveness of our proposed approach, we evaluate it for implicit-deadline systems using randomized task sets under different settings, showing that our approach outperforms the state-of-the-art.