Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[circle-mlir] Initial U2204 Dockerfile for CI #14616

Merged
merged 2 commits into from
Feb 10, 2025
Merged

[circle-mlir] Initial U2204 Dockerfile for CI #14616

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a63097f
[circle-mlir] Initial U2204 Dockerfile for CI
seanshpark Feb 6, 2025
846c343
update yapf
seanshpark Feb 7, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
213 changes: 213 additions & 0 deletions circle-mlir/infra/docker/u2204/Dockerfile
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,213 @@
FROM ubuntu:22.04 AS build

ENV DEBIAN_FRONTEND noninteractive

# Install 'add-apt-repository'
RUN apt-get update \
&& apt-get -qqy install apt-utils software-properties-common

# Build tool
RUN apt-get update \
&& apt-get -qqy install build-essential cmake git fakeroot autoconf automake libtool unzip wget \
devscripts debmake debhelper lcov

# additonal tools
RUN apt-get update \
&& apt-get -qqy install tree tmux

# Python
RUN apt-get update \
&& apt-get -qqy install python3 python3-pip python3-venv python3-dev python3-all dh-python
RUN python3 -m pip install --upgrade pip setuptools
RUN python3 -m pip install yapf==0.40.2 numpy h5py==3.8.0 einops

# TODO upgrade
ARG VER_TORCH=2.1.0
ARG VER_ONNX=1.16.0
ARG VER_ONNXRUNTIME=1.18.0

RUN python3 -m pip install torch==${VER_TORCH}
RUN python3 -m pip install onnx==${VER_ONNX}
RUN python3 -m pip install onnxruntime==${VER_ONNXRUNTIME}

# Clean archives (to reduce image size)
RUN apt-get clean -y

#
# For CIRCLE-MLIR
#
ARG NPROC=4
ARG WORK_DIR=/workdir
ARG FLATBUFFERS_HASH=a078130c
ARG ABSEIL_CPP_HASH=21510581
ARG LLVM_PROJECT_HASH=91088978d712cd7b33610c59f69d87d5a39e3113
ARG ONNX_MLIR_HASH=40615b362fe474d6cce4fd9c31a7ec51acd8f88a

WORKDIR ${WORK_DIR}

# Generate experimental patch file for onnx-mlir

RUN set -x \
&& { \
echo 'diff --git a/src/Dialect/ONNX/ONNXOps.td.inc b/src/Dialect/ONNX/ONNXOps.td.inc'; \
echo 'index d75ee6b6..f63a2910 100644'; \
echo '--- a/src/Dialect/ONNX/ONNXOps.td.inc'; \
echo '+++ b/src/Dialect/ONNX/ONNXOps.td.inc'; \
echo '@@ -1779,9 +1779,9 @@ def ONNXDequantizeLinearOp:ONNX_Op<"DequantizeLinear",'; \
echo ' `zero-point` is usually not used in the case of float8e4m3fn, float8e4m3fnuz, float8e5m2, float8e5m2fnuz quantization,'; \
echo " but the dequantization formula remains the same for consistency and 'x_scale' still determines the output type."; \
echo ' }];'; \
echo '- let arguments = (ins AnyTypeOf<[TensorOf<[I8]>, TensorOf<[UI8]>, TensorOf<[I32]>, TensorOf<[F8E4M3FN]>, TensorOf<[F8E4M3FNUZ]>, TensorOf<[F8E5M2]>, TensorOf<[F8E5M2FNUZ]>]>:$x,'; \
echo '+ let arguments = (ins AnyTypeOf<[TensorOf<[I8]>, TensorOf<[UI8]>, TensorOf<[I16]>, TensorOf<[I32]>, TensorOf<[F8E4M3FN]>, TensorOf<[F8E4M3FNUZ]>, TensorOf<[F8E5M2]>, TensorOf<[F8E5M2FNUZ]>]>:$x,'; \
echo ' AnyTypeOf<[TensorOf<[F32]>, TensorOf<[F16]>, TensorOf<[BF16]>]>:$x_scale,'; \
echo '- AnyTypeOf<[TensorOf<[I8]>, TensorOf<[UI8]>, TensorOf<[I32]>, TensorOf<[F8E4M3FN]>, TensorOf<[F8E4M3FNUZ]>, TensorOf<[F8E5M2]>, TensorOf<[F8E5M2FNUZ]>, NoneType]>:$x_zero_point,'; \
echo '+ AnyTypeOf<[TensorOf<[I8]>, TensorOf<[UI8]>, TensorOf<[I16]>, TensorOf<[I32]>, TensorOf<[F8E4M3FN]>, TensorOf<[F8E4M3FNUZ]>, TensorOf<[F8E5M2]>, TensorOf<[F8E5M2FNUZ]>, NoneType]>:$x_zero_point,'; \
echo ' DefaultValuedAttr<SI64Attr, "1">:$axis);'; \
echo ' let results = (outs AnyTypeOf<[TensorOf<[F32]>, TensorOf<[F16]>, TensorOf<[BF16]>]>:$y);'; \
echo ' let extraClassDeclaration = [{'; \
echo '@@ -5721,10 +5721,10 @@ def ONNXQuantizeLinearOp:ONNX_Op<"QuantizeLinear",'; \
echo ' }];'; \
echo ' let arguments = (ins AnyTypeOf<[TensorOf<[F32]>, TensorOf<[F16]>, TensorOf<[BF16]>, TensorOf<[I32]>]>:$x,'; \
echo ' AnyTypeOf<[TensorOf<[F32]>, TensorOf<[F16]>, TensorOf<[BF16]>, TensorOf<[I32]>]>:$y_scale,'; \
echo '- AnyTypeOf<[TensorOf<[I8]>, TensorOf<[UI8]>, TensorOf<[F8E4M3FN]>, TensorOf<[F8E4M3FNUZ]>, TensorOf<[F8E5M2]>, TensorOf<[F8E5M2FNUZ]>, NoneType]>:$y_zero_point,'; \
echo '+ AnyTypeOf<[TensorOf<[I8]>, TensorOf<[UI8]>, TensorOf<[I16]>, TensorOf<[F8E4M3FN]>, TensorOf<[F8E4M3FNUZ]>, TensorOf<[F8E5M2]>, TensorOf<[F8E5M2FNUZ]>, NoneType]>:$y_zero_point,'; \
echo ' DefaultValuedAttr<SI64Attr, "1">:$axis,'; \
echo ' DefaultValuedAttr<SI64Attr, "1">:$saturate);'; \
echo '- let results = (outs AnyTypeOf<[TensorOf<[I8]>, TensorOf<[UI8]>, TensorOf<[F8E4M3FN]>, TensorOf<[F8E4M3FNUZ]>, TensorOf<[F8E5M2]>, TensorOf<[F8E5M2FNUZ]>]>:$y);'; \
echo '+ let results = (outs AnyTypeOf<[TensorOf<[I8]>, TensorOf<[UI8]>, TensorOf<[I16]>, TensorOf<[F8E4M3FN]>, TensorOf<[F8E4M3FNUZ]>, TensorOf<[F8E5M2]>, TensorOf<[F8E5M2FNUZ]>]>:$y);'; \
echo ' let extraClassDeclaration = [{'; \
echo ' static int getNumberOfOperands() {'; \
echo ' return 3;'; \
echo ''; \
} > ${WORK_DIR}/onnx_quantdequant_i16.patch \
&& cat ${WORK_DIR}/onnx_quantdequant_i16.patch

# FLATBUFFERS build from source
RUN git clone --recursive https://github.com/google/flatbuffers.git \
&& cd flatbuffers && git checkout ${FLATBUFFERS_HASH}

RUN mkdir flatbuffers-build && cd flatbuffers-build \
&& cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=${WORK_DIR} ../flatbuffers \
&& make -j${NPROC} && make install

RUN rm -rf flatbuffers \
&& rm -rf flatbuffers-build

# ABSEIL-CPP build from source
RUN git clone --recursive https://github.com/abseil/abseil-cpp.git \
&& cd abseil-cpp && git checkout ${ABSEIL_CPP_HASH}

RUN mkdir abseil-cpp-build && cd abseil-cpp-build \
&& cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=${WORK_DIR} \
-DCMAKE_POSITION_INDEPENDENT_CODE=ON \
-DCMAKE_CXX_STANDARD=17 -DCMAKE_CXX_STANDARD_REQUIRED=ON \
../abseil-cpp \
&& make -j${NPROC} && make install

RUN rm -rf abseil-cpp \
&& rm -rf abseil-cpp-build

# NINJA
RUN git clone -b v1.10.2 https://github.com/ninja-build/ninja.git \
&& cd ninja && mkdir -p build && cd build \
&& cmake .. \
&& make -j${NPROC} install

RUN rm -rf ninja

# bazel to build protobuf from source
ARG BAZEL_ARCH="x86_64"
ARG BAZEL_VERSION=5.2.0
RUN wget -q https://github.com/bazelbuild/bazel/releases/download/${BAZEL_VERSION}/bazel-${BAZEL_VERSION}-linux-${BAZEL_ARCH} \
-O /usr/bin/bazel \
&& chmod a+x /usr/bin/bazel

# protobuf build from source
RUN git clone -b v21.12 https://github.com/google/protobuf.git \
&& cd protobuf \
&& git submodule update --init --recursive

RUN cd protobuf \
&& ./autogen.sh \
&& ./configure --with-zlib \
&& make -j${NPROC} && make install && ldconfig

RUN rm -rf protobuf \
&& rm -f /usr/bin/bazel

# Download LLVM and ONNX-MLIR
# NOTE download two huge repos at once to fast fail for network problems
RUN git clone --recursive https://github.com/llvm/llvm-project.git \
&& cd llvm-project && git checkout ${LLVM_PROJECT_HASH}

RUN git clone https://github.com/onnx/onnx-mlir.git \
&& cd onnx-mlir \
&& git checkout ${ONNX_MLIR_HASH} \
&& git submodule update --init --recursive

# LLVM/MLIR build from source
RUN mkdir llvm-project-build && cd llvm-project-build \
&& cmake -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=${WORK_DIR} \
-DLLVM_ENABLE_PROJECTS=mlir -DLLVM_TARGETS_TO_BUILD=host \
-DLLVM_ENABLE_ASSERTIONS=ON -DLLVM_ENABLE_RTTI=ON \
-DCMAKE_JOB_POOL_COMPILE:STRING=compile -DCMAKE_JOB_POOL_LINK:STRING=link \
-DCMAKE_JOB_POOLS:STRING='compile=4;link=1' \
../llvm-project/llvm

RUN cd llvm-project-build && cmake --build . && cmake --install .

# ONNX-MLIR build from source
# Patch to accept INT16 for QuantizeLinear/DequantizeLinear
RUN cd onnx-mlir \
&& patch -p1 < ${WORK_DIR}/onnx_quantdequant_i16.patch

RUN mkdir onnx-mlir-build && cd onnx-mlir-build \
&& cmake -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=${WORK_DIR} \
-DMLIR_DIR=${WORK_DIR}/llvm-project-build/lib/cmake/mlir \
-DPython3_ROOT_DIR=/usr/bin \
-DCMAKE_JOB_POOL_COMPILE:STRING=compile -DCMAKE_JOB_POOL_LINK:STRING=link \
-DCMAKE_JOB_POOLS:STRING='compile=4;link=1' \
-DONNX_MLIR_BUILD_TESTS=OFF -DONNX_MLIR_ENABLE_MHLO=OFF -DLLVM_ENABLE_ASSERTIONS=ON \
-DONNX_MLIR_DECOMP_ONNX_CONVTRANSPOSE=OFF \
../onnx-mlir

RUN cd onnx-mlir-build && cmake --build . && cmake --install .

# LLVM/MLIR clean up
RUN rm -rf llvm-project
RUN rm -rf llvm-project-build

# ONNX-MLIR clean up
RUN rm -rf onnx-mlir/.git \
&& rm -rf onnx-mlir/third_party/* \
&& rm -rf onnx-mlir-build/test \
&& rm -rf onnx-mlir-build/Release
RUN find ./onnx-mlir-build -name *.o -exec rm {} \;
RUN find ./onnx-mlir-build -name *.so -exec rm {} \;

# bin cleanup and strip: 2.7G -> 1.2G
RUN rm -f ./bin/onnx-mlir-opt \
&& rm -f ./bin/onnx-mlir-reduce
RUN cd ./bin && strip -s *
RUN cd ./lib && strip -s *.so.*git

# more cleanup
RUN rm -f /usr/local/lib/libprotoc.* \
&& rm -rf ./onnx-mlir-build/docs/* \
&& rm -rf ./onnx-mlir-build/third_party/* \
&& rm -f ./onnx-mlir-build/.ninja_* \
&& rm -f ./onnx-mlir-build/build.ninja

# copy from build
FROM ubuntu:22.04

COPY --from=build / /

# Add user account
ARG USER_ACCOUNT=circlemlir

RUN groupadd -g 1000 ${USER_ACCOUNT} \
&& useradd -r -u 1000 -g ${USER_ACCOUNT} -ms /bin/bash ${USER_ACCOUNT}
USER ${USER_ACCOUNT}
WORKDIR /home/${USER_ACCOUNT}
CMD ["bash"]