How does the “-effort” option control the eco?
The same as the opt of ICE, the pin is divided into n parts according to the weight to scan, low, medium, high, extra high, extreme high respectively correspond to 1, 2, 3, 4, 5 copies. The ratio of the number of pins is 1:2:..:(2^(n-1)). High: (1:2:4)
Definition of weights
Hold path: the number of violated paths passing the current point
Hold GBA: the proportion of the number of valid endpoints covered by the current point
Setup path: the number of valid paths passing the current point or tns, with option control (delay_tns, number)
Setup GBA: the number of valid endpoints covered by the current point or the proportion of tns
Leakage: the number of endpoints covered by the current point (reverse order)
Why not do it directly from the point with the highest weight?
It can only be done in batches and can not achieve the optimal solution, which is determined by the characteristics of the graph. It can not be considered a lot can only follow the top from the back to thesweep, and the score can only be the main to calculate, otherwise the preanalysis time will be too long.
Reasons for not selecting the margin
The margin is dynamically changed, and the pre margin is large. When scanning to it, there may be basically no margin. The setup requires that the option first repair gain is the largest.
Can the tns effect be replaced?
In fact, you can support three types: by path number, tns, and incremental delay (not added, depending on the feedback). As far as the results of the current feedback are concerned, it is not expected to have a large efficiency improvement, but due to scanning. The improvement of the algorithm, the time overhead does not increase too much with the level increase. Therefore, it is OK to set it to high during normal operation. After verification, you can remove this option and run it directly by high.
Why not allow “size cell” and “insert buffer” in one round when fixing hold?
Because the size cell is changing the thrust, the position of the cell does not need to be changed, which is the lowest cost. The insert buffer takes into account the location, size and area, so it will first be used in the lowest cost way.
Why not allow “remove buffer” and other methods in one round when fixing hold?
For the fix hold, the buffer removal has no additional cost while getting the benefit.
This kind of scheme can be used as much as possible to achieve the maximum benefit.
How does “split net” handle multiple power domain design?
Principle:
only normal cells are allowed to be inserted, depending on the netlist, which domain should be inserted.
Rule determined by netlist:
1). When source is a normal cell, and its power domain can be inserted.
2). When the source is a special cell, the power domain in which it is locatedcannot be inserted. It needs to look at the load. If the load is also a specialcell, it cannot be inserted. It can only be inserted in the domain where theload is an ordinary cell.
3). In all candidate hier path, split net will choose the least hier path of create port.
How to keep route when “split_net”?
Currently,XTop doesn't output routing def because most of feedbacks about the original ICE "keep_route" function are negative. We suggest user to export macro eco commands such as "ecoAddRepeater" and feed it back to APR tools. This type of macro commands can be useful for preserve net routing. After splitting, the original routing will still be left on the layout. But if choosing to export atomic commands, all the eco touched routes will be deleted.
Does XTop support OCV/AOCV/POCV?
At present, XTop only support XX and incr_derate set on lib_cell by set_timing_derate command; compared to the traditional path timing update flow, just find the changed arc on the path. Because only one action will only change the limited arc, so just go forward Look at a few levels and then look at a few levels, and finally calculate the arc delay delta of these levels, and then directly modify the rest of the path data; and aocv and pocv must be calculated at least to the end of the path, and the amount of calculation will also increase a lot. They are not supported yet.
After fix transition, does it need to go back to APR and then fix other item further?
XTop supports to continue to fix other violation after fixing transition.
But fixing transition and going back to APR directly is our recommended flow. Because using split_net methods when fixing transition may cause many route changes.
======返回目录======
<<< 上一章：Parameter Settings
>>> 下一章：Interactive ECO