GPL: results of RePlAce is not legal

[mirkat1206]

Hello, I am testing the global placement of RePlAce with the testcases in standalone branch version, and I find the result is not quite what I suppose it should be.

I tried gcd_nontd_test.tcl, and I opened the result def file along with gcd.v in Cadence Innovus.
The resulting layout is like this:

Here are my questioins:
(1) Why are the standard cells not on the "row" the input lef/def file has defined?
In fact, none of the cells is on the correct y-coordinate of any of the predefined rows.

(2) Why is there overlapping?

(3) How can I get a result like the GIF in README.md?
Should I add any options when executing the binary file? If yes, what should I do exactly?

(4) I have noticed the limitation mentioned in README.md: Mixed-sized RePlAce with (LEF/DEF/Verilog) interface does not generate legalized placement.
Does it mean that RePlAce is not capable of generating legal placement for either single-size or mixed-sized standard cells placement?
Or, does it mean that RePlAce can generate legal placement for single-size standard cells placement but cannot for mixed-sized standard cells placement?

p.s. I only need the GPL results. DPL is not neccesary for me.

[maliberty]

Replace is a global placer; you need to run detailed placement to see non-overlapping, in-row placement. The gif will have overlaps if you could zoom in.

[mirkat1206]

Got it. I think I have some misunderstanding about GPL.
I'll look into how to run detailed placement. Thanks for your answer.