sta: issues with set_max_fanout

[antonblanchard]

I'm trying to set a max fanout for all pins in a design. My understanding is this should do it:

set_max_fanout 5 [ current_design ]

But that doesn't appear to work since I end up with nets much higher fanouts (ie the resizer didn't repair them), and STA also doesn't report any max fanout violations.

Working with @maliberty and @tspyrou, we tried a number of other possible ways to add a fanout constraint:

set_max_fanout 5 [ get_nets -hier * ]
set_max_fanout 5 [ get_cells -of_object [get_nets -hier * ] ]
set_max_fanout 5 [ get_cells -hier * ]
set_max_fanout 5 [ get_cells ]
set_max_fanout 5 [ get_lib_cells * ]
set_max_fanout 5 [ get_lib_cells -of_objects [ get_cells -hier * ] ]

All seemed to fail.

Test case: set_max_fanout.tar.gz

This runs STA against a design with some large fan outs, and set_max_fanout set to 5:

Endpoint: _498_ (rising edge-triggered flip-flop clocked by clk)
Path Group: clk
Path Type: max

Fanout     Cap    Slew   Delay    Time   Description
-----------------------------------------------------------------------------
                          0.00    0.00   clock clk (rise edge)
                          0.00    0.00   clock source latency
                  0.98    0.72    0.72 ^ clk (in)
     1    0.22                           clk (net)
                  0.98    0.00    0.72 ^ clkbuf_0_clk/A (sky130_fd_sc_hd__clkbuf_16)
                  0.16    0.41    1.13 ^ clkbuf_0_clk/X (sky130_fd_sc_hd__clkbuf_16)
     2    0.14                           clknet_0_clk (net)
                  0.16    0.00    1.13 ^ clkbuf_1_0_0_clk/A (sky130_fd_sc_hd__clkbuf_8)
                  0.05    0.17    1.31 ^ clkbuf_1_0_0_clk/X (sky130_fd_sc_hd__clkbuf_8)
     1    0.02                           clknet_1_0_0_clk (net)
                  0.05    0.00    1.31 ^ clkbuf_1_0_1_clk/A (sky130_fd_sc_hd__clkbuf_8)
                  0.20    0.25    1.56 ^ clkbuf_1_0_1_clk/X (sky130_fd_sc_hd__clkbuf_8)
     2    0.11                           clknet_1_0_1_clk (net)
                  0.20    0.00    1.56 ^ clkbuf_2_0_0_clk/A (sky130_fd_sc_hd__clkbuf_8)
                  0.21    0.32    1.88 ^ clkbuf_2_0_0_clk/X (sky130_fd_sc_hd__clkbuf_8)
     2    0.12                           clknet_2_0_0_clk (net)
                  0.22    0.01    1.89 ^ clkbuf_3_1_0_clk/A (sky130_fd_sc_hd__clkbuf_8)
                  0.90    0.81    2.70 ^ clkbuf_3_1_0_clk/X (sky130_fd_sc_hd__clkbuf_8)
     8    0.54                           clknet_3_1_0_clk (net)
                  0.90    0.04    2.74 ^ clkbuf_leaf_93_clk/A (sky130_fd_sc_hd__clkbuf_16)
                  0.06    0.29    3.03 ^ clkbuf_leaf_93_clk/X (sky130_fd_sc_hd__clkbuf_16)
     2    0.01                           clknet_leaf_93_clk (net)
                  0.06    0.00    3.03 ^ _621_/CLK (sky130_fd_sc_hd__dfxtp_4)
                  1.56    1.48    4.51 ^ _621_/Q (sky130_fd_sc_hd__dfxtp_4)
    26    0.57                           a_registered[5] (net)
                  1.56    0.02    4.53 ^ U$$416/A2 (sky130_fd_sc_hd__a22o_4)
                  4.22    3.09    7.63 ^ U$$416/X (sky130_fd_sc_hd__a22o_4)
    65    1.56                           sel_1$4647 (net)
                  4.35    0.61    8.24 ^ U$$543/B2 (sky130_fd_sc_hd__a22o_1)
                  0.18    0.55    8.79 ^ U$$543/X (sky130_fd_sc_hd__a22o_1)
     1    0.01                           t$4711 (net)
                  0.18    0.00    8.79 ^ U$$544/A (sky130_fd_sc_hd__xor2_2)
                  0.22    0.15    8.94 v U$$544/X (sky130_fd_sc_hd__xor2_2)
     1    0.02                           booth_b6_m63 (net)

And yet:

max fanout violation count 0

[tspyrou]

Note the manual says list of ports or cells are the objects supported but set_max_fanout 5 [get_cells -hier *] gives this error. [ERROR STA-0467] unsupported object type Instance. Error: base.sdc, 13 STA-0467

[tspyrou]

@jjcherry56, I edited the line reading from an envi variable and hardcoded 5 since that is what @antonblanchard is trying to set.

The background is that there are some black boxes with no liberty causing some paths to be unconstrained. The resizer leaves behind fanout violations on the unconstrained paths as expected. He is trying to find a way to fix those. In the efabless flow the final timing analysis will be flat and netlists will fill in the black boxes but when openroad is run the black boxes are there. This flow has to be improved but in the meantime he is trying to make progress.

[jjcherry56]

The problem is the liberty file. "fanout" is not the number of loads, it is defined by the liberty fanout_load attribute on input pins and in this library the default value is zero:

default_fanout_load : 0.0000000000;

change it to 1 and you will see the expected behavior

default_fanout_load : 1.0;

[tspyrou]

@antonblanchard can you file an issue on openlane to fix this liberty? @donn fyi.

[antonblanchard]

Thanks @jjcherry56, I verified that fixing the liberty file fixed the issue.

I'm not sure if we have modify the liberty files in open_pdks (so we can fix this ourselves), but I submitted a PR against the sky130 repo at https://github.com/google/skywater-pdk-libs-sky130_fd_sc_hd/pull/7

[antonblanchard]

@tspyrou FYI the fix is now in open_pdks, and I also submitted a PR for ORFS. I see a commit in a PR from @jjcherry56 to take care of the OpenROAD tests.