Consider providing a conda environment.yml which creates an environment including all needed open source EDA tools

[mithro]

Thanks to the work by @proppy tools like OpenROAD, the SkyWater PDK, Magic, Netgen, etc are all now packaged in conda @ https://github.com/hdl/conda-eda -- There are even packages for cross compilers for things like RISC-V architecture which can be found @ https://github.com/hdl/conda-compilers

This means that silicon compiler could provide a conda environment.yml file (https://docs.conda.io/projects/conda/en/latest/user-guide/tasks/manage-environments.html#create-env-file-manually) which installs all the requirements needed to run Silicon Compiler.

[proppy]

additionally maybe also consider packaging siliconcompiler w/ a recipe that depends on the external tools listed here https://docs.siliconcompiler.com/en/latest/user_guide/installation.html#external-tools, so that people could do conda install siliconcompiler to get a full blown installation.

[proppy]

potentially also depends on hdl/conda-eda#175 depending if siliconcompiler needs the full klayout functionality or can live with the subset distributed on pypi

[aolofsson]

Interesting! There are quite a lot of tricky dependancies between the tool versions that we are running into. We'll take a look to see how much work it would be. This does require peple to use conda, which is a barrier to some. (but perhaps not a big one?)

[mithro]

@aolofsson -- Conda is pretty common and well accepted in the ML and scientific worlds. Like all packaging systems it has some people who hate it.

[WRansohoff]

Thank you for the suggestion! We started looking into the LiteX conda-eda recipes, and they look very promising.

I do have a couple questions, which might stem from a misconfiguration or improper usage on my end. I used a fresh Ubuntu 20.04 VM to test these installation steps. After installing miniconda and activating a new environment, I can install surelog and yosys with the recommended commands, e.g.:

conda install --channel "LiteX-Hub" surelog

I encounter issues with magic and netgen, though - they produce errors referring to incompatible glibc versions, but the error messages appear to indicate that the system has a compatible version installed:

[...]
UnsatisfiableError: The following specifications were found to be incompatible with your system:

  - feature:/linux-64::__glibc==2.31=0
  - magic -> libgcc-ng[version='>=9.3.0'] -> __glibc[version='>=2.17']

Your installed version is: 2.31

The OpenROAD install command produces a similar 'UnsatisfiableErorr', but it doesn't provide more verbose information:

Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Solving environment: failed with repodata from current_repodata.json, will retry with next repodata source.
Collecting package metadata (repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Solving environment: \
Found conflicts! Looking for incompatible packages.
This can take several minutes.  Press CTRL-C to abort.
failed

UnsatisfiableError:

Are these common issues, possibly caused by missing dependencies? I can also open an issue on the conda-eda repository if they are atypical.

[mithro]

FYI - @proppy

[mithro]

Please do report the issue on the upstream repositories!

I have asked @proppy to look into what is going on here. I believe it is because these packages are still very new (less than a week old) and need a little bit more work. The only way things will get better is to have users finding and reporting these issues!

[proppy]

@WRansohoff some of those package do depend on conda-forge, can you confirm if/this also happen when using a conda-forge based environment? (either by using https://github.com/conda-forge/miniforge#miniforge3 or folllowing https://conda-forge.org/docs/user/introduction.html#how-can-i-install-packages-from-conda-forge)

[WRansohoff]

Ah, I must have missed that dependency. Using conda-forge to install the packages seems to work, e.g.:

conda install -c conda-forge --channel "LiteX-Hub" openroad

Thank you for the clarification, and for packaging open-source EDA tools like this!

We might want to look into packaging specific versions of the tools to ensure compatibility with our scripts - it looks like the conda scripts build yosys, openroad, etc from the tip of main. But it looks like the conda-eda README contains instructions for building the packages.

[proppy]

We might want to look into packaging specific versions of the tools to ensure compatibility with our scripts - it looks like the conda scripts build yosys, openroad, etc from the tip of main.

The intent behind this versioning schema is that https://github.com/hdl/conda-eda/actions build intermediate version daily and version them against the output of git-describe before uploading them https://anaconda.org/LiteX-Hub/repo.

That allows us to get most of the intermediate version packages, see for example https://anaconda.org/LiteX-Hub/magic/files.

Downstream tools (like siliconcompiler) are then free to depends on the (range of) version(s) they've been QA'ed against by specifying an explicit version constraint: https://docs.conda.io/projects/conda-build/en/latest/resources/package-spec.html#package-match-specifications

[WRansohoff]

Great, thank you!

We have a few things to work out before we can fully support using conda env create to setup a working dev environment, but your recipes made it easy to get most of the way there with a very small script!

[siddharth-joshi]

Hi all, really appreciate the work you've all been doing. I was curious if there has been movement here? I'm about to embark on a fresh install so was curious re: if SC had been included in conda?