Rotated parallel mapping: A novel approach for mapping data parallel applications on CGRAs

Twenty Years of Automated Methods for Mapping Applications on CGRA

“…A solution adopted in many cases is to let the control flow managed by a host processor. But this reduces greatly the pos- Spatial mapping [23], [30], [31] GA [19] SA [32], [33] ILP [23], [34], [35] Temporal mapping [12], [16], [26], [36]- [40] SA [22] ILP [41] B&B [42] CP [43] SAT [17] SMT [44] Binding [14], [24], [28], [45]- [47] QEA [48] SA [30], [49], [50] ILP [15], [48] Scheduling [24], [28], [36], [46], [48], [50]- [52] ILP [15], [53] sibilities to use the CGRA and increases the communication overhead, loosing sometimes the benefit of the acceleration provided by the CGRA. Another approach is to provide the CGRA with extra hardware features to support the control flow.…”

“…The interaction between the CGRA and the memory is also of utmost importance as it defines the efficiency of the whole execution of the application. Various parameters of the memory can be considered for an efficient mapping: number of banks, communication bandwidth, and memory size [50], [65]- [68].…”