This is an out-of-tree MLIR RVTensor dialect.
The RISC-V Tensor dialect, abbreviated as RVTensor, introduces a subset of tensor operations frequently utilized in Deep Neural Networks (DNNs), primarily involving convolutions and matrix multiplications. The main goal of RVTensor is to facilitate the conversion of these high-level operations into vendor-specific RISC-V intrinsic operations, leveraging the capabilities of the vector and/or matrix Instruction Set Architecture (ISA). This ensures that these operators, from popular Machine Learning (ML) frameworks such as TensorFlow and PyTorch, can be efficiently represented within RVTensor. Additionally, it is anticipated that tools will be developed to continuously translate these operations from ML frameworks into RVTensor, enhancing interoperability and optimization for RISC-V based implementations.
This setup assumes that you have built LLVM and MLIR in $BUILD_DIR and installed them to $PREFIX. To build and launch the tests, run
mkdir build && cd build
cmake -G Ninja .. -DMLIR_DIR=$PREFIX/lib/cmake/mlir -DLLVM_EXTERNAL_LIT=$BUILD_DIR/bin/llvm-lit
cmake --build . --target check-rvtensorTo build the documentation from the TableGen description of the dialect operations, run
cmake --build . --target mlir-docNote: Make sure to pass -DLLVM_INSTALL_UTILS=ON when building LLVM with CMake in order to install FileCheck to the chosen
installation prefix.
This setup assumes that you build the project as part of a monolithic LLVM build via the LLVM_EXTERNAL_PROJECTS mechanism.
To build LLVM, MLIR, the example and launch the tests run
mkdir build && cd build
cmake -G Ninja `$LLVM_SRC_DIR/llvm` \
-DCMAKE_BUILD_TYPE=Release \
-DLLVM_TARGETS_TO_BUILD=host \
-DLLVM_ENABLE_PROJECTS=mlir \
-DLLVM_EXTERNAL_PROJECTS=rvtensor-dialect -DLLVM_EXTERNAL_RVTENSOR_DIALECT_SOURCE_DIR=../
cmake --build . --target check-rvtensorHere, $LLVM_SRC_DIR needs to point to the root of the monorepo.