- Library for AXI based data transfers to be used with AXI IR
Download the most recent version of cmake installer and ninja binary and add to path:
https://askubuntu.com/questions/829310/how-to-upgrade-cmake-in-ubuntu#comment1262462_829311
https://cmake.org/download/
https://github.com/ninja-build/ninja/releases
Install clang-10 clang++-10 and lld-10.
Download and install latest version of SystemC and export its path:
# Follow instructions at the end of the page to use this path
export SYSTEMC_HOME=/opt/systemc/systemc-2.3.3
Execute the following commands where you want to have a copy of the LLVM project:
To setup the project, use:
$ git clone https://github.com/judeharis/axi_mlir.git axi4mlir
$ cd axi4mlir
$ git submodule init
$ git submodule updateThis should clone the axi4mlir branch of llvm-project.
A helper script is provided to build the llvm-project
./build_tools/build_llvm_dev_x86.sh \
llvm-project \
builds/llvm-project/build \
builds/llvm-project/install
# Which generate build files in builds/llvm-project/build-x86Add mlir binaries to path:
export PATH=$PATH:$(pwd)/builds/llvm-project/build-x86/bin
It is possible to compile the AXI API as a library to use in another project.
# In the LLVM_BUILD folder run
ninja mlir_axi_runner_utils This will generate 2 files:
<build_folder>/lib/libaxi_api_v1.so- to link with any C project<build_folder>/lib/libmlir_axi_runner_utils.so- to be used by mlir-cpu-runner
An example of using the API from a MLIR file can be found here.
# Prepare important folders
mkdir -p ~/Downloads/systemc/
sudo mkdir -p /opt/systemc/systemc-2.3.3
# Download
pushd ~/Downloads/systemc
wget https://accellera.org/images/downloads/standards/systemc/systemc-2.3.3.tar.gz
tar -xvf systemc-2.3.3.tar.gz
# Compile and install
mkdir ~/Downloads/systemc/systemc-2.3.3/build
pushd ~/Downloads/systemc/systemc-2.3.3/build
../configure --prefix=/opt/systemc/systemc-2.3.3/
make -j10
sudo make install
# Set the relevant environment variables
export SYSTEMC_HOME=/opt/systemc/systemc-2.3.3
popd
popdAvailable Options (end of the file)
After compiling llvm and installing SystemC, a systemc library can be compiled with:
cd $PROJ_ROOT/builds/llvm-project/build-x86
ninja mlir_syscaxi_runner_utils
# Or for all sysc accelerators:
cmake --build $PROJ_ROOT/builds/llvm-project/build-x86 --target \
mlir_syscaxi_runner_utils_accv1 \
mlir_syscaxi_runner_utils_accv2 \
mlir_syscaxi_runner_utils_accv3 \
mlir_syscaxi_runner_utils_accv4 \
mlir_syscaxi_runner_utils_conv_accv1 \
mlir_syscaxi_runner_utils_conv_accv3A 4 by 4 matmul example can be executed with mlir jitter, triggering a systemC simulation with:
$PROJ_ROOT/builds/llvm-project/build-x86/bin/mlir-opt \
-convert-linalg-to-loops -convert-scf-to-cf -convert-vector-to-llvm \
-convert-memref-to-llvm -convert-std-to-llvm -reconcile-unrealized-casts \
$PROJ_ROOT/llvm-project/mlir/test/axi4mlir-runner/run-axi-v1-data-copy.mlir | \
$PROJ_ROOT/builds/llvm-project/build-x86/bin/mlir-cpu-runner \
-O0 -e main -entry-point-result=void \
-shared-libs=$PROJ_ROOT/builds/llvm-project/build-x86/lib/libmlir_syscaxi_runner_utils.so \
-shared-libs=$PROJ_ROOT/builds/llvm-project/build-x86/lib/libmlir_runner_utils.so
The same mlir code is automatically executed with ninja check-mlir, but the
llvm tests only check for the existence of the dma mock library
(libmlir_mockaxi_runner_utils.so) without systemc simulation:
$PROJ_ROOT/builds/llvm-project/build-x86/bin/mlir-opt \
-convert-linalg-to-loops -convert-scf-to-cf -convert-vector-to-llvm \
-convert-memref-to-llvm -convert-std-to-llvm -reconcile-unrealized-casts \
$PROJ_ROOT/llvm-project/mlir/test/axi4mlir-runner/run-axi-v1-data-copy.mlir | \
$PROJ_ROOT/builds/llvm-project/build-x86/bin/mlir-cpu-runner \
-O0 -e main -entry-point-result=void \
-shared-libs=$PROJ_ROOT/builds/llvm-project/build-x86/lib/libmlir_mockaxi_runner_utils.so \
-shared-libs=$PROJ_ROOT/builds/llvm-project/build-x86/lib/libmlir_runner_utils.so A C(16x32) = A(16x8) x B(8x32) accelerator v1 example can be executed with
the mlir jitter, triggering a systemC simulation with:
$PROJ_ROOT/builds/llvm-project/build-x86/bin/mlir-opt \
-test-linalg-to-axi4mlir="flow-cpu-accumulation" \
-convert-linalg-to-loops -convert-scf-to-cf -convert-vector-to-llvm \
-convert-memref-to-llvm -convert-std-to-llvm -reconcile-unrealized-casts \
$PROJ_ROOT/llvm-project/mlir/test/axi4mlir-runner/run-matmul-v1accel.mlir | \
$PROJ_ROOT/builds/llvm-project/build-x86/bin/mlir-cpu-runner \
-O0 -e main -entry-point-result=void \
-shared-libs=$PROJ_ROOT/builds/llvm-project/build-x86/lib/libmlir_syscaxi_runner_utils.so \
-shared-libs=$PROJ_ROOT/builds/llvm-project/build-x86/lib/libmlir_runner_utils.so
export PROJ_ROOT=/working_dir
$PROJ_ROOT/builds/llvm-project/build-x86/bin/mlir-opt \
-test-linalg-to-axi4mlir="flow-cpu-accumulation" \
-convert-linalg-to-loops -convert-scf-to-cf \
-arith-expand \
-memref-expand \
-convert-vector-to-llvm \
-convert-memref-to-llvm \
-convert-arith-to-llvm \
-convert-std-to-llvm \
-reconcile-unrealized-casts \
$PROJ_ROOT/llvm-project/mlir/test/axi4mlir-runner/run-matmul-v1accel.mlir | \
$PROJ_ROOT/builds/llvm-project/build-x86/bin/mlir-translate --mlir-to-llvmir -o app.ll
$PROJ_ROOT/builds/llvm-project/build-x86/bin/clang -g \
--target=arm-linux-gnueabihf -march=armv7-a -marm -mfloat-abi=hard \
-c -o app.o app.ll
$PROJ_ROOT/builds/llvm-project/build-x86/bin/clang -g -o app app.o \
--target=arm-linux-gnueabihf \
-Wl,-rpath=$PROJ_ROOT/builds/llvm-project/build-runner-arm/lib \
-L$PROJ_ROOT/builds/llvm-project/build-runner-arm/lib \
-lmlir_runner_utils -lmlir_axi_runner_utils
# Running with qemu
qemu-arm -L /usr/arm-linux-gnueabihf ./app
To create a standalone image with all build dependencies installed follow execute:
./build-docker.sh
To enter the development container
./start-docker.sh
Steps:
- Enter development container
- With qemu-user, crossbuild-essential-armhf, libz-dev installed
- Compile bootstrap mlir-tblgen binaries
- required when cross compiling for arm
- Compile clang, llvm, mlir for x86 with the required targets
- Cross-compile mlir-runner libraries for ARM
- this requires arm libs and llvm-project/cmake targeting cross-compilation
- needs mlir-tblgen from step 2
- Compile and link desired application for arm, run with qemu
- Using compiler from step 3, libraries from step 4, and pointing qemu to look for the right libraries
# 1. Enter development container
./start-docker.sh # Further steps are executed inside docker container with automatic pwd mounted
# 2. Compile bootstrap mlir-tblgen binaries
./build_tools/build_bootstrap_tblgen.sh llvm-project builds/llvm-project/build builds/llvm-project/install
# 3. Compile clang, llvm, mlir for x86 with the required targets
# This will take a while (20min using 16c/32t)
./build_tools/build_llvm_dev_x86.sh llvm-project builds/llvm-project/build builds/llvm-project/install
# 4. Cross-compile mlir-runner libraries for ARM
./build_tools/build_runner_arm.sh llvm-project builds/llvm-project/build builds/llvm-project/install
# 5. Compile and link desired application for arm, run with qemu
cd cross-comp/hello-world-mlir
make
make run-arm
# Or for individual steps on AOT compiling app.mlir into an arm binary
cp $PROJ_ROOT/llvm-project/mlir/test/Integration/Dialect/Linalg/CPU/test-conv-1d-nwc-wcf-call.mlir app.mlir
$PROJ_ROOT/builds/llvm-project/build-x86/bin/mlir-opt -convert-linalg-to-loops -convert-scf-to-cf -convert-linalg-to-llvm -lower-affine -convert-scf-to-cf --convert-memref-to-llvm -convert-std-to-llvm -reconcile-unrealized-casts -o app-llvm.mlir app.mlir
$PROJ_ROOT/builds/llvm-project/build-x86/bin/mlir-translate -mlir-to-llvmir -o app.ll app-llvm.mlir
$PROJ_ROOT/builds/llvm-project/build-x86/bin/clang --target=arm-linux-gnueabihf -march=armv7-a -marm -mfloat-abi=hard -c -o app.o app.ll
$PROJ_ROOT/builds/llvm-project/build-x86/bin/clang -o app app.o --target=arm-linux-gnueabihf -Wl,-rpath=$PROJ_ROOT/builds/llvm-project/build-runner-arm/lib -L$PROJ_ROOT/builds/llvm-project/build-runner-arm/lib -lmlir_runner_utils
# Running with qemu
qemu-arm -L /usr/arm-linux-gnueabihf ./appOnce these steps finished, build folders will be available in:
├── builds/build-runner-arm # (ARM) targets: mlir_runner_utils mlir_axi_runner_utils mlir_runner_utils
├── builds/build-tblgen-x86 # (x86) targets: mlir-tblgen clang-tblgen llvm-tblgen
└── builds/build-x86 # (x86) targets: mlir-opt mlir-translate mlir_runner_utils mlir_axi_runner_utils
export PROJ_ROOT=/working_dir
cmake --build $PROJ_ROOT/builds/llvm-project/build-runner-arm \
--target mlir_c_runner_utils mlir_runner_utils mlir_axi_runner_utils mlir_mockaxi_runner_utils
cmake --build $PROJ_ROOT/builds/llvm-project/build-x86 \
--target mlir_c_runner_utils mlir_runner_utils \
mlir_axi_runner_utils mlir_mockaxi_runner_utils \
mlir_syscaxi_runner_utils \
mlir_syscaxi_runner_utils_accv1 \
mlir_syscaxi_runner_utils_accv2 \
mlir_syscaxi_runner_utils_accv3 \
mlir_syscaxi_runner_utils_accv4 \
mlir_syscaxi_runner_utils_conv_accv1 \
mlir_syscaxi_runner_utils_conv_accv3
To recompile inside docker, with PROJ_ROOT=/working_dir:
cmake --build /working_dir/builds/llvm-project/build-x86 --target clang opt mlir-opt mlir-translate mlir-cpu-runner mlir_c_runner_utils mlir_runner_utils mlir_axi_runner_utils mlir_mockaxi_runner_utils mlir-lsp-server FileCheck && cmake --build /working_dir/builds/llvm-project/build-runner-arm --target mlir_c_runner_utils mlir_runner_utils mlir_axi_runner_utils mlir_mockaxi_runner_utils
cmake --build /working_dir/builds/llvm-project/build-x86 --target clang opt mlir-opt mlir-translate mlir-lsp-server FileCheck \
mlir-cpu-runner mlir_c_runner_utils mlir_runner_utils mlir_axi_runner_utils mlir_mockaxi_runner_utils \
mlir_syscaxi_runner_utils \
mlir_syscaxi_runner_utils_accv1 \
mlir_syscaxi_runner_utils_accv2 \
mlir_syscaxi_runner_utils_accv3 \
mlir_syscaxi_runner_utils_accv4 \
mlir_syscaxi_runner_utils_conv_accv1 \
mlir_syscaxi_runner_utils_conv_accv3 \
&& \
cmake --build /working_dir/builds/llvm-project/build-runner-arm --target mlir_c_runner_utils mlir_runner_utils mlir_axi_runner_utils mlir_mockaxi_runner_utils