W(i, j, x, y) denotes the weight of the edge between Instruction i and Instruction j, which is defined as the span of that edge. It is used to estimate the influence of the second factor on GF value. t(j) is the execution time of Instruction j. Nc(j) denotes the number of clusters that Instruction j can be scheduled in. ��(j, x, y) is the number of active inter-cluster data communications concerning from Instruction j, which is a member of the neighborhood of Instruction i, to Cluster x at Cycle y. It is mainly used to estimate the influence of the third factor on GF value.4.1.2. Calculation of RF Value Repulsion force value RF(i, x, y) represents the resource availability when Instruction i is to be prescheduled to Schedule-Point (x, y). There are two factors that will influence the RF value.

The available resources in each cluster. For the purpose of minimizing the number of execution cycles, we need to distribute instructions evenly in each cluster, which means we would like to pre-schedule instructions to cluster which has more available resources.The existed inter-cluster data communications in each cluster. As we know, for the purpose of balance the distribution of inter-cluster data communications, it is beneficial to pre-schedule instructions to cluster which has smaller number of existed inter-cluster data communications. In step 6 of Algorithm 1, M(j) is the mobility of Instruction j, which indicates the possibility of Instruction i to move between different cycles. ��(j, x, y) denotes the possibility that Instruction j is in Schedule-Point (x, y).

��j��(j, x, y) represents current resource occupation at Schedule-Point (x, y). It is used to calculate the influence of the first factor on RF value. ��(x, y) is the number of existed active inter-cluster data communications from other clusters to Cluster x at Cycle y. It is used to calculate the influence of the second factor on RF value.4.1.3. Calculation of BF Value As discussed before, instruction scheduling process for RFCC VLIW architecture has three tasks: (1) minimizing the number of inter-cluster data communications; (2) balancing the distribution of inter-cluster data communications to minimize the situation where the number of concurrent inter-cluster data communications exceeds the number of registers in the global register file or the number of read or write ports to the global register file from one cluster at a single clock cycle; (3) minimizing the number of execution cycle.

In order to fulfill the first task, the instruction should be prescheduled to the schedule point that has the largest GF value. For the third task, the instruction should be prescheduled to the schedule Anacetrapib point with the least RF value. And for the second task, we would like to schedule instruction to the schedule point with the largest GF value and the least RF value.