Skip to content

Latest commit

 

History

History
85 lines (51 loc) · 4.56 KB

TestingHighLevel.md

File metadata and controls

85 lines (51 loc) · 4.56 KB

ONNX-MLIR: Build trouble-shooting and testing ONNX_MLIR

Trouble shooting the building of ONNX-MLIR

If you have issues during the first onnx-mlir build, you may need to check the cmake variables used by our build. See the last section of this page for help.

If you have used the source directory successfully for a while, you may experience difficulties to rebuild onnx-mlir after merging the latest changes from the main branch.

Below is a couple of steps you may perform. If any of them apply, it is recommended to remove the onnx-mlir/build subdirectory and rebuild from scratch using the cmake commands.

1) Checking the right commit of the llvm-project

If the latest onnx-mlir main branch has moved to a newer commit level of the llvm-project, the build process will typically experience multiple compiler failures related to LLVM and MLIR code.

Level required is found in the first code box of the Building ONNX-MLIR page next to the git checkout command.

Level used in the code is found by executing a git log in the llvm-project subdirectory.

If they don't match, please update the llvm project to the required level.

2) Checking the right third_party support

Typically, when we update the ONNX op level, it results in new software in the third_party/onnx subdirectory. Failing to update that code results typically in compiler failures related to ONNX dialect code.

It is easier to simply remove the third_party directory and then reinstalling the code using git submodule update --init --recursive.

3) Dialect update

Sometimes a dialect update requires the entire build directory to be rebuilt. Typical errors that you may see are missing declarations, for example to verifier methods. The recommendation is to simply remove the onnx-mlir/build subdirectory and rebuild from scratch using the cmake commands.

4) Protobuf related issues

If you run into protobuf related errors during the build, check the following potential causes:

  • protobuf version is too low or too new (relative to the prereq)
  • libprotobuf version and python binding version mismatch
  • llvm-project, onnx, and/or onnx-mlir are built against different versions of protobuf, because after updating protobuf you only rebuild one of them
  • llvm-project, onnx, and/or onnx-mlir may detect different versions of python3 (so watch their cmake output) if you have multiple python versions installed
  • cmake caches stuff and you should never use "make clean" when rebuilding. Instead remove everything under the build tree and start from scratch.

These and many other trickeries for setting up the build env are the reason why we recommend using the onnxmlir/onnx-mlir-dev docker image for development.

High level testing of ONNX-MLIR

To run the lit ONNX-MLIR tests, use the following command:

call cmake --build . --config Release --target check-onnx-lit

Or simply invoke the check-onnx-lit target for ninja or make in the build directory.

To run the numerical ONNX-MLIR tests, use the following command:

call cmake --build . --config Release --target check-onnx-numerical

Or simply invoke the check-onnx-numerical target for ninja or make in the build directory.

To run the doc ONNX-MLIR tests, use the following command after installing third_party ONNX shown below. Details to first install the third_party ONNX project are detailed here. Note that it is key to install the ONNX project's version listed in our third_party subdirectory, as ONNX-MLIR may be behind the latest version from the ONNX standard.

call cmake --build . --config Release --target check-docs

Or simply invoke the check-docs target for ninja or make in the build directory.

Summary of LLVM and ONNX-MLIR Cmake Variables

The following CMake variables from LLVM and ONNX-MLIR can be used when compiling ONNX-MLIR.

MLIR_DIR:PATH Path to to the mlir cmake module inside an llvm-project build or install directory (e.g., c:/repos/llvm-project/build/lib/cmake/mlir). This is required if MLIR_DIR is not already set from a previous cmake invocation.

LLVM_EXTERNAL_LIT:PATH Path to the lit tool. Defaults to an empty string and LLVM will find the tool based on MLIR_DIR if possible. This is required when MLIR_DIR points to an install directory.