Modify and save database after simulation

[krabo0om]

I want to convert the .vpd files from a sim-debug (from VCS) to gziped .vcd files (for Voltus). Xcelium does not appear to be supported yet for gate-level simulation.

I have inserted an additional step after VCS's run_simulation to manipulate the waveforms that are given in sim.outputs.waveforms. As a result, their path/name changes. To reflect the changed path, I would like to update the sim.outputs.waveforms variable. First, for consistency and, second, to provide the correct paths for power analysis later.

However, when I use x.set_setting("sim.outputs.waveforms", new_waveform_paths) in my hook, it does not change the sim-debug-inputs.json file.

How can I make my changes permanent or is there a better way?

[harrisonliew]

If I'm not mistaken, I think you're using Hammer within Chipyard. A couple answers:

• Xcelium does support gate-level simulation, but it just hasn't been enabled in Chipyard yet.
• You could (and should) write to FSDB instead of VPD and converting to VCD. Just make sure the VCS version you are using isn't too new (see Investigate VPD, FSDB Support #769).
• The VCS plugin does not write the sim.outputs.waveforms key, as you may have found. Instead, it is set within Chipyard in sim.mk here. You will need to modify your sim.mk instead if you want to change the waveform file.

[krabo0om]

Thanks, yes, I'm using Hammer within Chipyard.

* Xcelium does support gate-level simulation, but it just hasn't been enabled in Chipyard yet.

Would that be easy to do for someone rather unfamiliar with Chipyard/Hammer like me? Would be the best solution.

* You could (and should) write to FSDB instead of VPD and converting to VCD. Just make sure the VCS version you are using isn't too new (see Investigate VPD, FSDB Support #769 ).

This exact issue was to motivation to convert it manually. VCS from 2019 appears to be too new ("empty" fdsb file according to voltus) and our version from 2017 is not a full 64bit installation. I can ask for 2018 to be installed. I am also limited to Voltus 2021 because 2022 crashes when reading ChipTop_power_constraints.tcl (although created by Innovus of the same version).

* The VCS plugin does not write the `sim.outputs.waveforms` key, as you may have found. Instead, it is set within Chipyard in [sim.mk here](https://github.com/ucb-bar/chipyard/blob/352cc773b55c509245aa4755450fa88130dba6a2/vlsi/sim.mk#L68-L70). You will need to modify your sim.mk instead if you want to change the waveform file.

That works. However, VCS still writes in the vpd format. Because my hook could be executed more than once, I have to check if the .vcd.gz file is actually a vpd or has already been converted to a vcd. Not a big problem, but also not "clean".
Currently, I'm using a hook before the power command to update the database and replace the power.input.waveforms paths with .vcd.gz.

[harrisonliew]

Would that be easy to do for someone rather unfamiliar with Chipyard/Hammer like me? Would be the best solution.

This will require that you dive into some Makefiles and understand how the Xcelium plugin works. If you want to try, you need to create a Xcelium-specific version of sim.mk using this Makefile as the reference for how to set the sim. keys. We just haven't gotten around to doing it yet, so if you are successful, a PR to Chipyard would be appreciated.

However, VCS still writes in the vpd format.

You probably have the USE_VPD Make variable set. Without it, it would generate FSDB by default. Anyways this doesn't matter for you since you can't use FSDB.

Because my hook could be executed more than once, I have to check if the .vcd.gz file is actually a vpd or has already been converted to a vcd.

Is there a reason why the hook could be executed more than once? Is this due to your own flow, or some Hammer issue?

Currently, I'm using a hook before the power command to update the database and replace the power.input.waveforms paths with .vcd.gz.

This is a good workaround for now.

[krabo0om]

You probably have the USE_VPD Make variable set. Without it, it would generate FSDB by default.

I tried all the combinations :)

Is there a reason why the hook could be executed more than once? Is this due to your own flow, or some Hammer issue?

I think it is due to my flow. Maybe it was also just an artifact of me testing the method repeatedly. In any case, the current workaround is good enough for now.

For completeness, I put the code here if anybody experiences the same problem:

import pathlib

def vpd2vcd_gz(x: HammerTool) -> bool:
    try:
      waveforms = x.get_setting("sim.outputs.waveforms")
    except KeyError:
        return True
    for waveform in waveforms:
        waveform = pathlib.Path(waveform)
        if waveform.suffix == '.vpd' and waveform.exists():
            vcd = waveform.with_suffix('.vcd')
            subprocess.run(['vcsplit', '-o', vcd, '-scope', x.get_setting('sim.inputs.tb_dut'), '-full64', waveform])
            subprocess.run(['gzip', vcd])
            waveform.unlink()
    return True

def vpd2vcd_path_fixes(x: HammerTool) -> bool:
    x.waveforms = [waveform.removesuffix('vpd') + ('vcd.gz') for waveform in x.waveforms]
    return True

class ExampleDriver(CLIDriver):
    def get_extra_sim_hooks(self) -> List[HammerToolHookAction]:
        extra_hooks = [
            HammerTool.make_post_insertion_hook("run_simulation", vpd2vcd_gz),
        ]   
        return extra_hooks

    def get_extra_power_hooks(self) -> List[HammerToolHookAction]:
        extra_hooks = [
            HammerTool.make_pre_insertion_hook("init_technology", vpd2vcd_path_fixes),
        ]   
        return extra_hooks

If you are using Chipyard, merge the code with the existing example-vlsi.