Folks, There were a lot of questions and concerns on the contest scoring function. I think it's a very positive thing to express your concerns and send me your feedbacks. In this way, we can make our contest more realistic and successful. Anyway, the biggest concern was that the previous scoring function encourages some overflows of edges because the computed WL could be cheaper in that way. Below, I attached an example sent by Mustafa Ozdal for better understanding. Also, Michael D. Moffitt made a lot of suggestions. Thanks Mustafa and Michael. So, here is the revised scoring function. TWL = 0; for each net if a net doesn't have an overflow edge WL = routed net WL else WL = ((max_overflow+capacity)/capacity)*sqrt(net degee)*HPWL /* here, max_overflow+capacity is a real usage */ TWL += WL Essentially, WL of a net with overflow edges is likely much bigger than that of routed WL without any overflow. Please note that WL is weighted by net degree. Also, when your router fails to route a net, its WL will be computed with the above function too. In other words, if your router successfully routes only 95% of nets, the rest 5% net's WLs will be calculated according to the above penalty function. Please remember that any solution in overflow group cannot beat the worst solution in non-overflow group. Thus, do your best to avoid unnecessary overflows. The above function will be used to determine your rank per each circuit. Your final rank score will be the sum of your ranks of all circuits, and the smallest rank score wins the contest. A few other issues: 1. I sent an email yesterday regarding pins on the tile boundary, but I realized that I made a mistake. Please follow the example shown in http://www.ispd.cc/planar.pdf In that example, assuming tile size is 10x10, (0,0)-(0,9) is one tile and (0,10)-(0,19) is the next tile. Thus, a pin at (0,10) belongs to the right (or upper) tile, not the left (lower) tile. Please use this definition. I apologize for the confusion. 2. This is a global routing contest. Thus, we won't consider detailed track assignment. As long as a tile-based path connects a source tile (a tile where a source pin belongs to) and a sink tile (similar definition), it is considered a valid global routing path. I'll post these clarifications at the global routing contest website tomorrow. Like always, let me know if you have any questions or concerns. By the way, an official global routing solution evaluation script will be posted soon. Good luck to everyone. Gi-Joon --

※ 發信站: 批踢踢實業坊(ptt.cc) ◆ From: 140.112.5.65