### Table 1: Results on 26 small circuits with 22 or less PIs. pi gate exhaustive gate replace divide-and-conquer circuit

2005

"... In PAGE 3: ... Our results are compared with traditional input vector control methods in terms of leakage saving, run time, area and delay penalty. We conducted ex- periments on 69 benchmarks including 26 small circuits with 22 or fewer primary inputs ( Table1 ) and 43 large circuits (Figure 4). For each small circuit, we find its optimal MLV by exhaustive search.... In PAGE 3: ... To have a fair comparison with [1], we also collect the average leakage of 1,000 random input vectors for each large circuit. Table1 reports the results for the 26 small circuits. Column 4... ..."

Cited by 3

### Table 1. The divide-and-conquer coloring

"... In PAGE 2: ... Note that the colors used in supernodes 00 and 11 could be the same, but cannot be colors A and B. As an example, Table1 shows how this algorithm solves the coloring problem on dB(8, 2) recursively. A node v2v1 represented by the intersection of row v2 and column v1, and the symbol on the intersection is the color of that node.... ..."

### Table 2. The divide-and-conquer algorithms that atsl uses.

"... In PAGE 12: ....5. The Repertoire of Building Blocks. Tables 1 and 2 list the sorting al- gorithms that atsl currently uses as building blocks. Table 1 lists leaf algorithm, and Table2 lists divide-and-conquer compositions. Many of these building blocks are based on existing publicly-available codes.... ..."

### Table 1: Perfomance Results for divide-and-conquer example

1993

"... In PAGE 11: ... We sought a bound on the execution time of the program in a maximally parallel setting given durations for the fork and computation times. Table1 shows the performance of the toolset on several sizes of this example. The columns of the table give the size of the example, the number of tasks, the time required to generate the system of inequalities from the specification, the time to solve the system, the total time, and the size of the system of inequalities (all times are in seconds on a DECstation 5000/125).... ..."

Cited by 6

### Table 2: Average performance comparison with [1] algorithm in [1] gate replacement divide-and-conquer

2005

"... In PAGE 4: ... Be- cause their detailed results are not available and to make a fair com- parison, we can only compare the average performance and leakage reduction over the average leakage current of 1,000 random vec- tors. Table2 summerizes the performance improvement in the con- trol point insertion approach [1], our gate replacement algorithm, and the divide-and-conquer approach. Table 2: Average performance comparison with [1] algorithm in [1] gate replacement divide-and-conquer... ..."

Cited by 3

### Table 8: Statistics for 2D divide-and-conquer Delaunay triangulation of several point sets.

1997

"... In PAGE 48: ... (I have also tried perfect lattices with 53-bit integer coordinates, but ORIENT3D and INSPHERE never pass stage B; the perturbed lattices are preferred here because they occasionally force the predicates into stage C or D.) The results for 2D, which appear in Table8 , indicate that the four-stage predicates add about 8% to the total running time for randomly distributed input points, mainly because of the error bound tests. For the more difficult point sets, the penalty may be as great as 30%.... ..."

Cited by 86

### Table 8: Statistics for 2D divide-and-conquer Delaunay triangulation of several point sets. Timings are

1997

"... In PAGE 50: ... (I have also tried perfect lattices with 53-bit integer coordinates, but ORIENT3D and INSPHERE never pass stage B; the perturbed lattices are preferred here because they occasionally force the predicates into stage C or D.) The results for 2D, which appear in Table8 , indicate that the four-stage predicates add about 8% to the total running time for randomly distributed input points, mainly because of the error bound tests. For the more difficult point sets, the penalty may be as great as 30%.... ..."

Cited by 86

### Table 8: Statistics for 2D divide-and-conquer Delaunay triangulation of several point sets.

1997

"... In PAGE 48: ... (I have also tried perfect lattices with 53-bit integer coordinates, but ORIENT3D and INSPHERE never pass stage B; the perturbed lattices are preferred here because they occasionally force the predicates into stage C or D.) The results for 2D, which appear in Table8 , indicate that the four-stage predicates add about 8% to the total running time for randomly distributed input points, mainly because of the error bound tests. For the more difficult point sets, the penalty may be as great as 30%.... ..."

Cited by 86

### Table 8: Statistics for 2D divide-and-conquer Delaunay triangulation of several point sets. Timings are

1997

"... In PAGE 50: ... (I have also tried perfect lattices with 53-bit integer coordinates, but ORIENT3D and INSPHERE never pass stage B; the perturbed lattices are preferred here because they occasionally force the predicates into stage C or D.) The results for 2D, which appear in Table8 , indicate that the four-stage predicates add about 8% to the total running time for randomly distributed input points, mainly because of the error bound tests. For the more difficult point sets, the penalty may be as great as 30%.... ..."

Cited by 86

### Table 3: Statistics for 2D divide-and-conquer Delaunay triangulation of several point sets.

1996

"... In PAGE 10: ...lattices with 53-bit integer coordinates, but ORIENT3D and INSPHERE would never pass stage B; the perturbed lattices occasionally force the predicates into stage C or D.) The results for 2D, outlined in Table3 , indicate that the four-stage predicates add about 8% to the total running time for randomly distributed input points, mainly because of the errorboundtests. Forthe more difficult point sets, the penalty may be as great as 30%.... ..."

Cited by 47