click on a benchmark to see an analysis of the checker results
| benchmark | number of test points |
average cpu time (s) |
minimum wirelength |
best (min W,max A) for full route |
|---|---|---|---|---|
| toy1 | 60 | 0.05 | 56 | (10, 1.00) |
| toy2 | 90 | 0.06 | 86 | (20, 1.19) |
| fract | 117 | 0.14 | 488 | (35, 1.00) |
| fract2 | 126 | 0.13 | 546 | (26, 1.24) |
| fract3 | 126 | 0.16 | 644 | (21, 1.33) |
| primary1 | 191 | 1.25 | 4017 | (76, 1.00) |
| primary2 | 100 | 36.38 | 34269 | (135, 1.00) |
| struct | 171 | 5.11 | 8919 | (50, 1.00) |
| biomed | 12 | 186.37 | 134404 | (175, 1.00) |
| industry1 | 25 | 29.27 | 40031 | (150, 1.00) |
| industry2 | 9 | 88.92 | * | * |
| (*the checker program is unable to determine statistics for the industry2 benchmark) | ||||
the table file format for checker results is
W A unrouted cellerr neterr utime stime
cpu times are based on a sun sparc ultra running sloaris (weh5201)
overall, these results were as expected. the graphs look like the handout, changing W and A have the expected result. as W decreases and A increases, the amount of wiring space in each cell decreases, causing more cells to fill over capacity. as W increases and A decreases, the amount of wiring space increases, making it easier for the router. the router does, however, hit a minimum wirelenth value, where further increasing W (and decreasing A) has no effect.
since the router seems to be fast and "reasonable" W values can be found to route every benchmark to 100% with no errors i think the initial design decisions were sufficient.
without other data to compare with it is difficult to make a much more detailed analysis.