Parallel Computer Architecture

892

CHAPTER

11 Latency Tolerance

1 2 4 8 16 32 64 1 2 4 8 16 32 64

Number of processors Number of processors

(a) Prefetching relative to original (b) Block transfer relative to prefetching

FIGURE 11.21 Performance benefits of prefetching remote data in a Fast Fourier Transform.

(a) shows the performance of the prefetched version relative to that of the original version. The graph

can be interpreted just as described in Figure 11.13: each curve shows the execution time of the

prefetched version relative to that of the nonprefetched version for the same cache block size

(B)

for

each number of processors, (b) shows the performance of the versio ...