Neura -- a dataflow dialect.

Build LLVM and Neura

Neura is pinned to LLVM commit 6146a88f60492b520a36f8f8f3231e15f3cc6082.

Prerequisites

Use Python 3.11.16 and install the native binding dependencies into the same environment that will configure LLVM and Neura:

python -m pip install pybind11==2.13.6 nanobind==2.15.0

The system build dependencies are CMake, Ninja, Clang, LLD, and optionally ccache.

Build LLVM/MLIR

From the llvm-project repository:

git checkout 6146a88f60492b520a36f8f8f3231e15f3cc6082

cmake -G Ninja -S llvm -B build \
  -DLLVM_ENABLE_PROJECTS="mlir;clang" \
  -DLLVM_BUILD_EXAMPLES=OFF \
  -DLLVM_TARGETS_TO_BUILD="Native" \
  -DCMAKE_BUILD_TYPE=Release \
  -DLLVM_ENABLE_ASSERTIONS=ON \
  -DCMAKE_C_COMPILER=clang \
  -DCMAKE_CXX_COMPILER=clang++ \
  -DCMAKE_CXX_STANDARD=17 \
  -DCMAKE_CXX_FLAGS="-std=c++17 -frtti" \
  -DLLVM_ENABLE_LLD=ON \
  -DMLIR_INSTALL_AGGREGATE_OBJECTS=ON \
  -DLLVM_ENABLE_RTTI=ON \
  -DMLIR_ENABLE_BINDINGS_PYTHON=ON \
  -DMLIR_BINDINGS_PYTHON_NB_DOMAIN=mlir \
  -DPython3_EXECUTABLE="$(which python)" \
  -DPython_EXECUTABLE="$(which python)"

cmake --build build

MLIR_ENABLE_BINDINGS_PYTHON=ON is required because Neura builds its Python package from the upstream MLIR Python bindings.

Build Neura

Neura is an out-of-tree MLIR project. Set LLVM_BUILD_DIR to the LLVM build directory; Neura derives the LLVM source and CMake package paths from it.

cd /path/to/dataflow
export LLVM_BUILD_DIR=/path/to/llvm-project/build

cmake -G Ninja -S . -B build \
  -DCMAKE_BUILD_TYPE=Release \
  -DPython3_EXECUTABLE="$(which python)" \
  -DPython_EXECUTABLE="$(which python)"

cmake --build build
cmake --build build --target NeuraPythonModules

The generated package is placed in build/python_packages/neura_core/neura_mlir.

Run tests

# Focused Python operation-binding test.
$LLVM_BUILD_DIR/bin/llvm-lit -v test/python/neura_binding.py

# Complete Neura test suite.
$LLVM_BUILD_DIR/bin/llvm-lit -v test

Some frontend tests additionally require PyTorch and a compatible torch-mlir wheel; the pinned CI setup is in .github/workflows/main.yml.

Sync test/e2e outputs into Zeonica_Testbench (submodule)

This repo vendors sarchlab/Zeonica_Testbench as a git submodule at test/benchmark/Zeonica_Testbench.

After running e2e compilation/tests, you can sync the generated artifacts into the testbench repo:

$ ./tools/sync_e2e_outputs_to_zeonica_testbench.sh

Mapping (per kernel K):

• test/e2e/K/tmp-generated-dfg.{dot,yaml} -> test/benchmark/Zeonica_Testbench/kernel/K/K-dfg.{dot,yaml}
• test/e2e/K/tmp-generated-instructions.{asm,yaml} -> test/benchmark/Zeonica_Testbench/kernel/K/K-instructions.{asm,yaml}

Contributing

Please refer to the Contributing Guide for code style, formatting, and contribution workflow.