### TABLE V COMPARISON WITH SSCH: DISJOINT FLOWS

### Table 1: Classification of XY-disjoint cases

2007

"... In PAGE 19: ... Let n = |N| and h = |G/N|, and then let k = rn and m = sn, so that r and s are the orders of the images of X and Y in the quotient G/N. When G/N is of type (A2), we may suppose without loss of generality that v = s/2 is odd, and so can take v to be odd in all four of the first cases from Table1 in Section 4. With the above notation assumed, we will use the following: Lemma 8.... In PAGE 19: ... square Corollary 8.3 If G/N is of type (A1), (A2), (B1) or (B2), as in Table1 (with v odd in case (A2)), then gcd(v, n) = 1; moreover, if G/N is of type (A2), then gcd(u, n) = 1 whenever u or n is odd. Proof.... ..."

### Table 2. Two examples of disjoint paths

### Table 1: Classification of XY-disjoint cases

2007

"... In PAGE 19: ... Let n = |N| and h = |G/N|, and then let k = rn and m = sn, so that r and s are the orders of the images of X and Y in the quotient G/N. When G/N is of type (A2), we may suppose without loss of generality that v = s/2 is odd, and so can take v to be odd in all four of the first cases from Table1 in Section 4. With the above notation assumed, we will use the following: Lemma 8.... In PAGE 19: ... square Corollary 8.3 If G/N is of type (A1), (A2), (B1) or (B2), as in Table1 (with v odd in case (A2)), then gcd(v, n) = 1; moreover, if G/N is of type (A2), then gcd(u, n) = 1 whenever u or n is odd. Proof.... ..."

### Table 3: Solution times for disjoint problem instances

"... In PAGE 16: ... Table 11 in Appendix A lists these averages and the standard deviation of the ten test cases in each group. Table3 presents the solution speed and the ability to solve test cases completely for these methods. For each group, the average solution time (over all ten test cases) is shown in CPU seconds.... ..."

### Table 10: Breakdown of correct predictions. R = Memory Renaming, D = Store Set Depen- dence Prediction, A = Hybrid Address Prediction, V = Hybrid Value Prediction, NP = loads that were not predicted by any of the predictors, Miss = all pre- dictors mispredicted these loads, Other = the remaining contributions of loads for the columns not shown. Each column represents the disjoint percentage of loads that were correctly predicted by the combination of predictors in the col- umn header. For example, the column labelled VD corresponds to the percent of loads that were correctly predicted by both value and dependence prediction, but not by either rename or address prediction. The \other quot; column contains predictor combinations which were responsible for close to 0% of the load cov- erage. For example, since almost all loads could be predicted with dependence prediction, the quot;V quot;, quot;A quot;, and quot;D quot; columns were close to 0% and were combined into the Other column, along with other combinations close to 0%.

2000

"... In PAGE 32: ...6 Table 11: Percent of correct and incorrect Rename, Value, Dependence, and Address pre- dictions for Data Cache hits and misses. Table10 shows the contribution of di erent predictor combinations. The numbers rep- resent the percentage of executed loads that were correctly predicted by the combination of predictors in the column header.... In PAGE 32: ... In Figure 16, after combining value prediction with dependence and address prediction, little performance improvement is seen by adding in memory renaming. One reason for this can be seen in Table10 , which shows that value prediction correctly predicts 27.7% of the loads, which memory renaming either mispredicts or chooses not to predict.... ..."

Cited by 25

### Table 1: How often combinations beat non- combinations for highest R-precision and F1 in 20 disjoint test/train sets, considering all feature set sizes. Parentheses contain sign test p-values.

2006

"... In PAGE 2: ... While combination methods achieved higher R-precision over the majority of individual set sizes, recall that we se- lected methods from our preliminary study to attain a high- est peak R-precision. Table1 shows how often each method in each pair obtained the highest peak over all set sizes. Because many complete tasks require thresholding to pro- duce label sets, we also report here microaveraged F1.... ..."

Cited by 1