Table 8.1 shows the results of applying GDPPO to the schedules generated by APGAN and RPMC on several practical SDF systems. The columns labeled “% Impr.” show the percentage of buffer memory reduction obtained by GDPPO. The QMF tree filter banks fall into a class of graphs for which APGAN is guaranteed to produce optimal results (see Chapter 7), and thus there is no room for GDPPO to produce improvement when APGAN is applied to these two examples. Overall, GDPPO produces an improvement in 11 out of the 14 heuristic/application combinations. A “significant” (greater than 5%) improvement is obtained in 9 of the 14 combinations; the mean improvement over all 14 combinations is 9.9%; and from the CD-DAT and DAT-CD examples, we see that it is possible to obtain very large reductions in the buffer memory requirement with GDPPO.
Unable to display preview. Download preview PDF.