Stream HLS build and example

INSTALL.md

Build

git clone https://github.com/UCLA-VAST/Stream-HLS
cd Stream-HLS
sudo add-apt-repository ppa:deadsnakes/ppa
sudo apt update
sudo apt install python3.11 python3.11-venv
python3.11 -m venv env
. env/bin/activate

sudo apt install lld

cd extern
git clone -b llvmorg-18.1.2 https://github.com/llvm/llvm-project.git
cd llvm-project/build
cd build
cmake -G Ninja ../llvm \
  -DLLVM_ENABLE_PROJECTS=mlir \
  -DCMAKE_BUILD_TYPE=Debug \
  -DLLVM_ENABLE_ASSERTIONS=ON \
  -DCMAKE_C_COMPILER=clang \
  -DLLVM_TARGETS_TO_BUILD=host \
  -DCMAKE_CXX_COMPILER=clang++ \
  -DLLVM_ENABLE_LLD=ON
cmake --build . --target check-mlir


#  obtain ampl licence https://portal.ampl.com/account/ampl/
python -m pip install amplpy --upgrade
python -m amplpy.modules install highs gurobi
python -m amplpy.modules activate 672f05a4-b6ba-430e-ad19-c3d73fd0242d

./build-streamhls.sh $(readlink -f extern/llvm-project)

cp env/lib/python3.11/site-packages/ampl_module_base/bin/* env/bin

python file

import torch_mlir
import torch 
import time
import torch
import torch.nn as nn

class Model(nn.Module):
  def forward(self, a):
    return a * a ** a

model = Model()
model.train(False)
model.eval()
inputs = torch.asarray([1.0, 2.0])

module = torch_mlir.compile(
        model,
        inputs,
        output_type="linalg-on-tensors",
        use_tracing=True)

path = 'test.mlir'
with open(path, 'w') as f:
    print(module.operation.get_asm(), file=f)

build file

set -ex

. ../env/bin/activate

python3 test.py

streamhls-opt ./test.mlir \
    -streamhls-host-pipeline \
    > host.mlir

streamhls-translate ./host.mlir \
    -emit-vivado-hls \
    -vitis-hls-weights-dir=data \
    -vitis-hls-is-host=true \
    -o host_tb.cpp

streamhls-opt test.mlir \
    -streamhls-kernel-pipeline="top-func=forward \
      graph-file=graph\
      report-file=report\
      optimize-schedule=1\
      parallelize-nodes=1\
      combined-optimization=0\
      board-dsps=1024 \
      tiling-limit=16 \
      time-limit-minutes=10 \
      bufferize-func-args=0 \
      optimize-conv-reuse=0 \
      minimize-on-chip-buffers=0 \
      debug-point=14" > kernel.mlir


streamhls-translate \
    kernel.mlir \
    -emit-vivado-hls \
    -o kernel.cpp