[ERROR DPL-0036] Detailed placement failed -- error is not actionable

xobs commented 1 year ago

Description

When synthesizing a design, OpenLane produces the error [ERROR DPL-0036] Detailed placement failed.. It does not say why placement failed, nor does it explain how to address the error. It also doesn't report how to iterate to address the error, which is a problem when synthesis takes a lot of time.

Expected Behavior

When a step fails, it should report why the step failed. Ideally it should also give suggestions on how to fix the issue.

Environment report

user@Ondo:/opt/Si/work/caravel-gf180-pio/deps/openlane_src$ python3 ./env.py issue-survey
Kernel: Linux v5.15.74.2-microsoft-standard-WSL2
Distribution: ubuntu 22.04
Python: v3.10.6 (OK)
Container Engine: docker v20.10.21 (OK)
OpenLane Git Version: cb59d1f84deb5cedbb5b0a3e3f3b4129a967c988
pip: INSTALLED
python-venv: INSTALLED
---
PDK Version Verification Status: OK
---
Git Log (Last 3 Commits)

cb59d1f 2022-11-18T18:42:38+02:00 fix return values in `equally_spaced_sequence` (#1503) - Kareem Farid -  (grafted, HEAD, tag: 2022.11.19)
---
Git Remotes

origin  git@github.com:The-OpenROAD-Project/OpenLane (fetch)
origin  git@github.com:The-OpenROAD-Project/OpenLane (push)
user@Ondo:/opt/Si/work/caravel-gf180-pio/deps/openlane_src$

Reproduction material

issue_reproducible.tar.gz

The repository is available at https://github.com/xobs/caravel-gf180-pio

Relevant log output

[INFO]: Running Placement Resizer Design Optimizations (log: ../opt/Si/work/caravel-gf180-pio/openlane/wb_pio/runs/22_12_14_12_14/logs/placement/8-resizer.log)...
  [ERROR]: during executing openroad script /openlane/scripts/openroad/resizer.tcl
[ERROR]: Log: ../opt/Si/work/caravel-gf180-pio/openlane/wb_pio/runs/22_12_14_12_14/logs/placement/8-resizer.log
[ERROR]: Last 10 lines:
[INFO DPL-0035]  _075635_
[INFO DPL-0035]  _075635_
[INFO DPL-0035]  _088150_
[INFO DPL-0035]  _087790_
[INFO DPL-0035]  _087459_
[INFO DPL-0035]  _088111_
[INFO DPL-0035] message limit reached, this message will no longer print
[ERROR DPL-0036] Detailed placement failed.
Error: resizer.tcl, 62 DPL-0036
child process exited abnormally

[ERROR]: Creating issue reproducible...
[INFO]: Saving runtime environment...

xobs commented 1 year ago

Note that there is https://github.com/The-OpenROAD-Project/OpenLane/issues/1010 which seems to have run into a similar problem. However, that issue is filled mostly with people attempting to debug the symptoms of the error. I've opened this Issue because the error itself has no meaning -- it's just a bunch of numbers with no context and no indication of why placement failed.

Possible solutions are:

Print something about... DIE_AREA? PL_TARGET_DENSITY? FP_CORE_UTIL? I'm still not sure what these mean, or why there are two different knobs for die (what's a core and what's not-core?)
Show a visualisation of the placement process so that a user can intuit where the failure is
Suggest a different SYNTH_STRATEGY (does this affect things?)
Suggest that maybe the design is too complicated to be routed

maliberty commented 1 year ago

The simplest thing is to open the gui and look at where those instances are placed. There are many possible causes and it hard to say what the exact cause is.

xobs commented 1 year ago

Hmm... that would be a good thing to add to the output:

How to open the GUI
Which instances are causing the error
What to look for

maliberty commented 1 year ago

I don't think an error message should be a how-to book. The instances that are a problem are right there in the error message, as shown in the description above:

[INFO DPL-0035]  _075635_
[INFO DPL-0035]  _088150_
[INFO DPL-0035]  _087790_
[INFO DPL-0035]  _087459_

I would look at the area in which they are placed and how congested it is. Is there an excess of buffering (perhaps bad constraints)? Is it a channel between macros with limited space? Is the utilization very high overall? etc.

xobs commented 1 year ago

I do think an error message should explain what the error is. The error I'm seeing is:

OpenROAD 7f00621cb612fd94e15b35790afe744c89d433a7 
This program is licensed under the BSD-3 license. See the LICENSE file for details.
Components of this program may be licensed under more restrictive licenses which must be honored.
[INFO]: Setting RC values...
[INFO RSZ-0027] Inserted 76 input buffers.
[INFO RSZ-0028] Inserted 97 output buffers.
[INFO RSZ-0058] Using max wire length 22815um.
[INFO RSZ-0039] Resized 16830 instances.
[INFO RSZ-0042] Inserted 143 tie gf180mcu_fd_sc_mcu7t5v0__tiel instances.
[INFO DPL-0034] Detailed placement failed on:
[INFO DPL-0035]  _063554_
[INFO DPL-0035]  _091720_
[INFO DPL-0035]  _055760_
...
[INFO DPL-0035]  _065550_
[INFO DPL-0035]  _074226_
[INFO DPL-0035]  _073498_
[INFO DPL-0035]  _073498_
[INFO DPL-0035]  _064475_
[INFO DPL-0035]  _064475_
[INFO DPL-0035]  _064475_
[INFO DPL-0035] message limit reached, this message will no longer print
[ERROR DPL-0036] Detailed placement failed.
Error: resizer.tcl, 62 DPL-0036

What are those numbers? Are they wires? Are they cells? Are they logic elements? Are they indexes into some table? Are they distances? How do I cross-reference them with something useful?

The error message might not need to be a how-to book, but there should be something there that I can use with a search engine to be able to understand the problem better. Maybe a link to a page that describes how to open the GUI to investigate the error.

maliberty commented 1 year ago

The placer works on instances so they are instance names. The lousy names can be blamed on yosys; OR just uses whatever it gets as input.

xobs commented 1 year ago

Alright, so it might make sense to rewrite the error as:

Detailed placement failed on the following instances:

ANd the last message should read:

message limit reached, this message will no longer print -- there are %d remaining instances that were unable to be placed

Additionally, I should open an Issue with Yosys to try to get them to improve the names that are generated.

You mentioned that it's possible to open the GUI to inspect things. Can you show me where that's documented? The documentation at https://openroad.readthedocs.io/en/latest/main/src/gui/README.html#add-buttons-to-the-toolbarhttps://openroad.readthedocs.io/en/latest/main/src/gui/README.html#add-buttons-to-the-toolbar seems to just go over the TCL interface without mentioning any of the following:

What sorts of files does OpenROAD support viewing?
How can you open files in OpenROAD?
Which files should be inspected in order to troubleshoot the issue?

vijayank88 commented 1 year ago

@xobs

python3 ./env.py issue-survey should be ran where OPENLANE_ROOT set
For GF180 you need modify your config.json ex. using wrong CLOCK_NET and should be "PL_BASIC_PLACEMENT": 0, also "PL_TARGET_DENSITY": 0.85 its too high
"VDD_NETS": ["vccd1"], "GND_NETS": ["vssd1"], is not a valid pins for gf180

vijayank88 commented 1 year ago

@xobs which mpw tag you're cloned and running this project? This is reason behind we were closing the issue if Environment Report not updated means with invalid label.

vijayank88 commented 1 year ago

Due to high density cells placement too congested. Screenshot from 2022-12-14 11-28-46

xobs commented 1 year ago

I've updated the original issue with the output of the command -- there were some issues around https://github.com/efabless/caravel_user_project/issues/227 that caused the path to be wrong.

For PL_BASIC_PLACEMENT -- is there any documentation on this item? What does it mean to skip basic placement? It seems like a thing that would be good to do, and https://openlane-docs.readthedocs.io/en/rtd-w-markdown/configuration/README.html#placement doesn't say why it's a bad thing to run.

For PL_TARGET_DENSITY -- how would I know that 0.85 is too high? The tool itself recommended 0.75, and the documentation at https://openlane-docs.readthedocs.io/en/rtd-w-markdown/configuration/README.html#placement seems to indicate that values up to 1 are acceptable.

As for VDD_NETS and GND_NETS -- I have corrected those in my configuration. Should it have thrown an error with the invalid configuration?

I'm not sure what you're asking in the final question -- I cloned it from https://github.com/efabless/caravel_user_project, though it seems as though Github forgets the history when you clone a repository that way. I forked it yesterday.

vijayank88 commented 1 year ago

Refer latest document from here: https://openlane.readthedocs.io/en/latest/

xobs commented 1 year ago

Collecting the issues:

For OpenLane:

PL_BASIC_PLACEMENT should have an explanation of what it does and what are the implications of skipping it: https://github.com/The-OpenROAD-Project/OpenLane/pull/1576
PL_TARGET_DENSITY should have documentation on "vaguely" what acceptable parameters there are
A check should be added to ensure VDD_NETS and GND_NETS are valid: https://github.com/The-OpenROAD-Project/OpenLane/issues/1575

For OpenROAD:

~~Documentation should be added to show how to use the GUI: https://github.com/The-OpenROAD-Project/OpenROAD/pull/2619~~
~~The error message should indicate that the entity names that follow are the names of instances that were unable to be placed~~
The error message should indicate why they were unable to be placed (did it exceed the number of interations?)
~~The error message should indicate how many additional entities were unable to be placed~~

For Yosys:

The generated net names should be improved: https://github.com/The-OpenROAD-Project/OpenLane/pull/1577

vijayank88 commented 1 year ago

Replace with following config.json and try it

{
    "DESIGN_NAME": "wb_pio",
    "DESIGN_IS_CORE": 0,
    "VERILOG_FILES": ["dir::../../verilog/rtl/defines.v", "dir::../../verilog/rtl/wb_pio_top.v", "dir::../../verilog/rtl/fpga_pio/src/*.v"],
    "CLOCK_PERIOD": 10,
    "CLOCK_PORT": "wb_clk_i",
    "CLOCK_NET": "pio1.clk",
    "FP_SIZING": "absolute",
    "DIE_AREA": "0 0 1900 1600",
    "FP_PIN_ORDER_CFG": "dir::pin_order.cfg",
    "PL_BASIC_PLACEMENT": 0,
    "PL_TARGET_DENSITY": 0.40,
    "SYNTH_NO_FLAT": 0,
    "VDD_NETS": ["vdd"],
    "GND_NETS": ["vss"],
    "DIODE_INSERTION_STRATEGY": 4,
    "RUN_CVC": 0,
    "SYNTH_STRATEGY": "DELAY 4",
    "pdk::sky130*": {
        "FP_CORE_UTIL": 75,
        "RT_MAX_LAYER": "met4",
        "scl::sky130_fd_sc_hd": {
            "CLOCK_PERIOD": 10
        },
        "scl::sky130_fd_sc_hdll": {
            "CLOCK_PERIOD": 10
        },
        "scl::sky130_fd_sc_hs": {
            "CLOCK_PERIOD": 8
        },
        "scl::sky130_fd_sc_ls": {
            "CLOCK_PERIOD": 10,
            "SYNTH_MAX_FANOUT": 5
        },
        "scl::sky130_fd_sc_ms": {
            "CLOCK_PERIOD": 10
        }
    },
    "pdk::gf180mcuC": {
        "STD_CELL_LIBRARY": "gf180mcu_fd_sc_mcu7t5v0",
        "CLOCK_PERIOD": 24.0,
        "FP_CORE_UTIL": 20,
        "DPL_CELL_PADDING": 1,
        "RT_MIN_LAYER": "Metal1",
        "RT_MAX_LAYER": "Metal4",
        "SYNTH_MAX_FANOUT": 4
    }
}

Also https://github.com/xobs/caravel-gf180-pio/blob/main/verilog/rtl/wb_pio_top.v#L42 update power pins to vdd vss for gf180

xobs commented 1 year ago

That config.json does get to clock tree synthesis, but fails with the same sort of error (Detailed placement failed.).

Is there any documentation on how to get that nice graphic you showed? A search of https://openlane.readthedocs.io/en/latest/ does not yield any documentation on the -gui option.

vijayank88 commented 1 year ago

Use openroad -gui and load odb file with read_db <path-to>/name.odb

With GUI specific usage document available here

I've tried your repo locally and running detailed routing with updated config.json. Updated in the thread above.

xobs commented 1 year ago

I've been updating https://github.com/The-OpenROAD-Project/OpenLane/issues/1574#issuecomment-1350453510 with issues and pull requests that I've been following.

The specific thing that was missing was read_db, which is never called out as being the GUI's answer to the question "How do I open a file?" so I've added documentation to the README to explicitly call that out.

maliberty commented 1 year ago

The message limit is a property of the logger and is not specific to a particular message. We could add a total number to the message.

maliberty commented 1 year ago

Message improved in https://github.com/The-OpenROAD-Project/OpenROAD/pull/2626

maliberty commented 1 year ago

"The error message should indicate why they were unable to be placed" is impractical and can't be implemented

maliberty commented 1 year ago

You can run flow.tcl with -gui to open the gui.

vijayank88 commented 1 year ago

@xobs Are you able to complete flow?

xobs commented 1 year ago

I haven't, but that isn't related to the issues I've opened -- the routing failure is due to a design problem, so now I can work on simplifying the Verilog to get it to route.

kareefardi commented 1 year ago

I think this is completed now

The-OpenROAD-Project / OpenLane