|
马上注册,结交更多好友,享用更多功能,让你轻松玩转社区。
您需要 登录 才可以下载或查看,没有账号?注册
x
本帖最后由 yiyaowen 于 2024-10-17 17:14 编辑
大家好,我最近在跑一个数模混合电路后仿,xact寄生参数提取出的calibre网表有200M,使用APS+32个线程进行仿真,并打开了Post-Layout Optimization。因为开启了多线程和RC-Reduction,整个仿真过程并不慢,大概耗时35分钟就能跑完,但是从log可以看到,这其中有15分钟的时间花费在了网表的解析和其它准备工作上(也就是实际用来做spectre-trans的时间只有20分钟)。根据我的观察,准备仿真需要分配大概10个G的内存,这期间只有1个thread在工作:
Time for NDB Parsing: CPU = 669.241 s (11m 9.2s), elapsed = 670.763 s (11m 10.8s).
Time accumulated: CPU = 670.633 s (11m 10.6s), elapsed = 670.763 s (11m 10.8s).
Peak resident memory used = 6.77 Gbytes.
==========
Time for Elaboration: CPU = 135.585 s (2m 15.6s), elapsed = 135.917 s (2m 15.9s).
Time accumulated: CPU = 806.218 s (13m 26.2s), elapsed = 806.681 s (13m 26.7s).
Peak resident memory used = 8.41 Gbytes.
==========
Time for EDB Visiting: CPU = 25.6271 s, elapsed = 25.7052 s.
Time accumulated: CPU = 831.845 s (13m 51.8s), elapsed = 832.386 s (13m 52.4s).
Peak resident memory used = 9.43 Gbytes.
==========
Time for parsing: CPU = 112.375 s (1m 52.4s), elapsed = 112.727 s (1m 52.7s).
Time accumulated: CPU = 944.22 s (15m 44.2s), elapsed = 945.113 s (15m 45.1s).
Peak resident memory used = 9.78 Gbytes.
==========
然后是trans-analysis跑了大概20min
==========
Simulation started at: 4:18:19 PM, Thur Oct 17, 2024, ended at: 4:53:41 PM, Thur Oct 17, 2024, with elapsed time (wall clock): 2.12 ks (35m 22.1s).
==========
特别是这个NDB-Parsing,占用了很长的时间,而且它只使用1个thread。请问大家:1.这种情况是正常的吗?2.有没有什么办法减小用在解析上的时间,比如能否给这些工作也启用多线程?多谢了。
|
|