Hardware Reference
In-Depth Information
instructions on all multithreaded SIMD processors. Guz et al. [2009] have an interesting ana-
lytic model for these two architectures.
The researchers said that they selected the benchmark programs by analyzing the computa-
tional and memory characteristics of four recently proposed benchmark suites and then “for-
mulated the set of throughput computing kernels that capture these characteristics.” Figure 4.29
describes these 14 kernels, and Figure 4.30 shows the performance results, with larger num-
bers meaning faster.
FIGURE 4.29 Throughput computing kernel characteristics (from Table 1 in Lee et al.
[2010].) The name in parentheses identifies the benchmark name in this section. The authors
suggest that code for both machines had equal optimization effort.
 
Search WWH ::




Custom Search