-
Notifications
You must be signed in to change notification settings - Fork 434
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Out of tree build in conda environment / overdependency problem #3104
Comments
I have no idea what the mlir-python-bindings-feedstock in conda is. I wrote much of those Python bindings, and afaik, the MLIR project never converged on releasing them as a real package or anything ("MLIR" is not an entity that means much outside its role as a library to build certain kinds of compilers). Those bindings were written to be used by other projects wishing to create a Python API for their compilers: MLIR compilers are monolithic and generally don't interoperate across binary boundaries (not saying I care for this state, just reporting the news). So torch-mlir is building a standalone Python API under its namespace, using sources from upstream to do so. The CMake standalone example includes a minimal, standalone Python API setup: https://github.com/llvm/llvm-project/tree/main/mlir/examples/standalone iirc, there is CMake magic that gives the source sets names, but basically it is all of the files under It's been a long time since I've looked at any of that, but I think the |
(I should also say that all of this is open to be improved. The project contributors to these things have not invested here and largely use torch-mlir as a C++ library, as mentioned. But I am certainly not opposed to folks who care about other ways of using it investing to make it work) |
Thank you @stellaraccident ! I've resolve issue with missing sources - it was because feedstock build script was removing My question is it possible that there are missing dependencies set between |
Quite possible. I believe |
This should be fixed now, but I'm waiting for you to test it before I catch up the |
Hello there I'm trying to build out-of-tree
torch-mlir
in conda environment, but failing to do so because of numerous problems.I had to build locally
mlir-python-binding
to haveMLIRPythonExtension*
andMLIRPythonSources*
targets populated (conda-forge/mlir-python-bindings-feedstock#19), but then I'm getting this kind of error:It looks pretty wired for me, cause for some reason out-of-tree build reference to the source (not header) files of mlir project. Is it problem with overdependency on torch-mlir side, or it is "by-design". If it is by design, could you provide detailed explanation what source file is for in the downstream project (I'm relatively new to llvm)?
The text was updated successfully, but these errors were encountered: