@@ -39,7 +39,7 @@ See the [issues](https://git.esa.informatik.tu-darmstadt.de/gems/gems/issues). B
## Internals
GeMS has the unique capability to construct dependence graphs that are guaranteed to be feasible, or infeasible, at the lower bound for the II search space (i.e. the usual MinII = max(RecMII, ResMII)). This is a two-part approach: First, a cycle in the dependence graph is constructed that defines the instance's desired MinII. Then, during the edge generation phase, GeMS has to ensure that no MinII-changing edge is added. We employ several quick checks to handle common situations, but have to invoke an actual modulo scheduler to check the (in-)feasibility of smaller subgraphs in some cases. GeMS internally uses two ILP-based modulo schedulers for this purpose: The formulation by Eichenberger and Davidson [3], and the Moovac formation [4].
GeMS has the unique capability to construct dependence graphs that are guaranteed to be feasible, or infeasible, at the lower bound for the II search space (i.e. the usual MinII = max(RecMII, ResMII)). Graph generation proceeds in two phases: First, a cycle in the dependence graph is constructed that defines the instance's desired MinII. Then, during the edge generation phase, GeMS has to ensure that no MinII-changing edge is added. We employ several quick checks to handle common situations, but have to invoke an actual modulo scheduler to check the (in-)feasibility of smaller subgraphs in some cases. GeMS internally uses two ILP-based modulo schedulers for this purpose: The formulation by Eichenberger and Davidson [3], and the Moovac formation [4].