作者moonshade (翻滚吧!猫咪)
看板NTUGIEE_EDA
标题[研究] 大象
时间Sat May 28 20:01:51 2005
In this paper, the authors propose a placement technique for circuit
rows multi-voltage designs, in which rows of different voltage supplies
can be interwove and level converting filp-flops are placed across two
rows with different voltage supplies. The technique first runs Gordian
algorithm to get an initial solution. Based on the solution, it
assigns the voltage supply of each row by the majority voltage supply
of cells on the row. Then they propose an ILP algorithm to redistribute
every cell not compatible to the voltage
supply of the row it's currently on, to the nearest row with compatible
voltage supply. The experiments are based on ISCA98 benchmark, shows
the increment of wirelength is 20% in average.
1. Though the topic is interesting, the proposed technique is not
new. The ILP based legalizing method is a well-known technique,
and is considered time consuming especially for large scale designs.
2. Therefore, the reviewers think the authors should report the
runtimes of the experiments and try the algorithm on the larger benchmarks,
such as ISPD-02 IBM benchmark.
3. The wirelength increment is too large, which is 20% in average.
The reviewers think that this issue should be considered in global
placement stage, or the wirlength increment of this algorithm is not
considered efficient.
4. The “interleaves” at the second paragraph of section 2
should be “interweaves”.
--
1980以前的paper看起来都像用打字机打的,然後图是用糨糊黏上去的,
那个时代的人应该想不到今天电脑的能耐,
那麽,十年後的科技会是什麽样子呢?
--
※ 发信站: 批踢踢实业坊(ptt.cc)
◆ From: 140.112.48.60
※ 编辑: moonshade 来自: 140.112.48.60 (05/28 20:09)
※ 编辑: moonshade 来自: 140.112.48.60 (05/28 20:10)
※ 编辑: moonshade 来自: 140.112.48.60 (05/28 20:10)
1F:推 internal:line 10 show"s"没有s 140.112.48.60 05/28
2F:推 yellowfishie:第七行 to redistribute" " 也没有 s 218.164.38.150 05/28
3F:推 yellowfishie:第七行 Then they "propose" an 时态相同 218.164.38.150 05/28
4F:推 yellowfishie:1. is a well-know technique" "and is .. 218.164.38.150 05/28
5F:嘘 crazying:看来是不需要我了 @@a 140.112.25.195 05/28
6F:推 yellowfishie:is considered to "be" time"-"consuming 218.164.38.150 05/28
7F:推 crazying:话说回来 为什麽会按成嘘 ||| 140.112.25.195 05/29
8F:推 bluetai:因为你想嘘嘘? 219.86.34.17 05/29
9F:推 Morinfen:well-known 221.169.1.234 05/29
10F:→ Morinfen:1的最後 large scale design"s" 221.169.1.234 05/29
※ 编辑: moonshade 来自: 203.203.83.87 (05/29 00:55)
※ 编辑: moonshade 来自: 203.203.83.87 (05/29 00:57)
11F:推 moonshade:感谢~ 203.203.83.87 05/29