ABSTRACT
This chapter explores the buffering cost as the maximum number of live tokens present in it using a coarse-grained model of liveness. It explores an alternative cost for implementing buffers. Merged buffers versus lifetime analysis based sharing. Hence, when there are multiple iterations, the con-sumed-before-produced (CBP) can, at worst, offset the buffer by the number of tokens produced on one firing; the other firings can still share the buffer. Basically, the CBP captures token traffic for one firing, whereas the merging model captures the token traffic over several firings in a nested loop. Hence, when there are multiple iterations, the CBP can, at worst, offset the buffer by the number of tokens produced on one firing; the other firings can still share the buffer. The chapter shows that merging an input buffer with an output buffer will result in significantly less space required than their sums, or the maximum number of live tokens computed according to the coarse-grained model.
