Repository navigation
|
Hey @everyone, I have been looking into ABC's technology mapping code, especially
This seems reasonable for roughly symmetric CMOS libraries but what happens if the library is asymmetric cause its made on a different technology |
Replies: 1 comment
|
From what I understand, classic map actually tracks rise and fall separately and swaps them through inverting paths, and I think for liberty files this happens in abc's |
From what I understand, classic map actually tracks rise and fall separately and swaps them through inverting paths, and
max(rise_delay, fall_delay)is only the fallback for pins with an unknown phase.&nfdoes collapse to(rise_delay + fall_delay) / 2with a single arrival per node.I think for liberty files this happens in abc's
.lib-to-genlibconversion, which already averages rise/fall when fitting its delay model, so even map sees a symmetric library. That biases mapping decisions on asym libraries, but only mapping, and the post-mapping STA and sizing(stime, upsize, dnsize)use the full nldm tables with rise/fall separate, so most of it is recovered there. I think that if you want …