Note that this code currently requires llvm-project@bd0cae6.
In order to build graphblas-opt, run python3 build.py from mlir_graphblas/src/. This will build graphblas-opt and run the tests for graphblas-opt. The built files will be stored in mlir_graphblas/src/build.
build.py does not rebuild from scratch by default. To perform a clean build, run python3 build.py -build-clean.
Ensure that clang-format is available (install the clang-tools conda package) and run:
./run-clang-format.py -r mlir_graphblas/src/
If changes required, can make changes in place with
./run-clang-format.py -i -r mlir_graphblas/src/
mlir-graphblas is transitioning away from lowering code targeting the SCF dialect and towards lowering code targeting
linalg.generic. This process is happening in tandem with changes to the sparse-tensor dialect's lowering of
linalg.generic with dynamically-shaped output. As a result, mlir-graphblas is temporarily pointing at the mlir-ac
conda package which is built from a branch off the LLVM project code. Once these changes are merged into the main branch
on LLVM, mlir-graphblas will be updated to target that. Until then, we will be out of sync with the LLVM project.