Add TRT 7.1 Pipeline (#8073)

* Revert for testing TensorRT 7.1 * change to origianl googletest version * change machine * remove build arg * change back machine * revert back googletest version * Make it ready to merge to master * revert onnx-tensorrt to v7.1 * rename yml * use [[ ]] in bash command * add sudo * add chmod * add correct path * change another way to revert onnx-tensorrt * change docker image to manylinux build
2026-07-30 20:18:08 +00:00 · 2021-06-21 20:57:04 -07:00 · 2021-06-21 20:57:04 -07:00 · 27d1784d44
commit 27d1784d44
parent 3cd06cb38c
5 changed files with 298 additions and 5 deletions
--- a/tools/ci_build/github/azure-pipelines/linux-gpu-tensorrt-7-1-ci-pipeline.yml
+++ b/tools/ci_build/github/azure-pipelines/linux-gpu-tensorrt-7-1-ci-pipeline.yml
@ -0,0 +1,59 @@
+jobs:
+- job: Linux_CI_GPU_TENSORRT_Dev
+  pool: onnxruntime-tensorrt-linuxbuild
+  variables:
+    ALLOW_RELEASED_ONNX_OPSET_ONLY: '1'
+  timeoutInMinutes: 120
+  steps:
+    - checkout: self
+      clean: true
+      submodules: recursive
+
+    - template: templates/get-docker-image-steps.yml
+      parameters:
+        Dockerfile: tools/ci_build/github/linux/docker/Dockerfile.manylinux2014_cuda11_tensorrt7_1
+        Context: tools/ci_build/github/linux/docker
+        DockerBuildArgs: "--network=host --build-arg POLICY=manylinux2014 --build-arg PLATFORM=x86_64  --build-arg DEVTOOLSET_ROOTPATH=/opt/rh/devtoolset-9/root --build-arg PREPEND_PATH=/opt/rh/devtoolset-9/root/usr/bin: --build-arg LD_LIBRARY_PATH_ARG=/opt/rh/devtoolset-9/root/usr/lib64:/opt/rh/devtoolset-9/root/usr/lib:/opt/rh/devtoolset-9/root/usr/lib64/dyninst:/opt/rh/devtoolset-9/root/usr/lib/dyninst:/usr/local/lib64 --build-arg BUILD_UID=$( id -u )"
+        Repository: onnxruntimegpubuild
+    - task: CmdLine@2
+      inputs:
+        script: |
+          mkdir -p $HOME/.onnx
+          cd $(Build.SourcesDirectory)/cmake/external/onnx-tensorrt/
+          git checkout 7.1
+          cd $(Build.SourcesDirectory)
+          docker run --gpus all -e CC=/opt/rh/devtoolset-9/root/usr/bin/cc -e CXX=/opt/rh/devtoolset-9/root/usr/bin/c++ -e CFLAGS="-Wp,-D_FORTIFY_SOURCE=2 -Wp,-D_GLIBCXX_ASSERTIONS -fstack-protector-strong -fstack-clash-protection -fcf-protection -O3 -Wl,--strip-all" -e CXXFLAGS="-Wp,-D_FORTIFY_SOURCE=2 -Wp,-D_GLIBCXX_ASSERTIONS -fstack-protector-strong -fstack-clash-protection -fcf-protection -O3 -Wl,--strip-all" --rm \
+            --volume /data/onnx:/data/onnx:ro \
+            --volume $(Build.SourcesDirectory):/onnxruntime_src \
+            --volume $(Build.BinariesDirectory):/build \
+            --volume /data/models:/build/models:ro \
+            --volume $HOME/.onnx:/home/onnxruntimedev/.onnx \
+            -e ALLOW_RELEASED_ONNX_OPSET_ONLY=0 \
+            -e NIGHTLY_BUILD \
+            -e BUILD_BUILDNUMBER \
+            onnxruntimegpubuild \
+            /opt/python/cp37-cp37m/bin/python3 /onnxruntime_src/tools/ci_build/build.py \
+              --build_dir /build --cmake_generator Ninja \
+              --config Release \
+              --skip_submodule_sync \
+              --build_shared_lib \
+              --parallel \
+              --build_wheel \
+              --enable_onnx_tests --use_cuda --cuda_version=11.0 --cuda_home=/usr/local/cuda-11.0 --cudnn_home=/usr/local/cuda-11.0 \
+              --enable_pybind --build_java --build_nodejs \
+              --use_tensorrt --tensorrt_home /usr \
+              --cmake_extra_defines CMAKE_CUDA_HOST_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/cc  CMAKE_CUDA_ARCHITECTURES=52
+        workingDirectory: $(Build.SourcesDirectory)
+    - task: PublishTestResults@2
+      displayName: 'Publish unit test results'
+      inputs:
+        testResultsFiles: '**/*.results.xml'
+        searchFolder: '$(Build.BinariesDirectory)'
+        testRunTitle: 'Unit Test Run'
+      condition: succeededOrFailed()
+
+    - template: templates/component-governance-component-detection-steps.yml
+      parameters :
+        condition : 'ci_only'
+
+    - template: templates/clean-agent-build-directory-step.yml
--- a/tools/ci_build/github/linux/docker/Dockerfile.manylinux2014_cuda11_tensorrt7_1
+++ b/tools/ci_build/github/linux/docker/Dockerfile.manylinux2014_cuda11_tensorrt7_1
@ -0,0 +1,192 @@
+ARG BASEIMAGE=nvcr.io/nvidia/cuda:11.0-cudnn8-devel-centos7
+ARG POLICY=manylinux2014
+ARG PLATFORM=x86_64
+ARG DEVTOOLSET_ROOTPATH=
+ARG LD_LIBRARY_PATH_ARG=
+ARG PREPEND_PATH=
+
+#We need CUDA, TensorRT and manylinux. But the CUDA Toolkit End User License Agreement says NVIDIA CUDA Driver Libraries(libcuda.so, libnvidia-ptxjitcompiler.so) are only distributable in applications that meet this criteria:
+#1. The application was developed starting from a NVIDIA CUDA container obtained from Docker Hub or the NVIDIA GPU Cloud, and
+#2. The resulting application is packaged as a Docker container and distributed to users on Docker Hub or the NVIDIA GPU Cloud only.
+#So we use CUDA as the base image then add manylinux and TensorRT on top of it.
+
+#Build manylinux2014 docker image begin
+FROM $BASEIMAGE AS runtime_base
+ARG POLICY
+ARG PLATFORM
+ARG DEVTOOLSET_ROOTPATH
+ARG LD_LIBRARY_PATH_ARG
+ARG PREPEND_PATH
+LABEL maintainer="The ManyLinux project"
+
+ENV AUDITWHEEL_POLICY=${POLICY} AUDITWHEEL_ARCH=${PLATFORM} AUDITWHEEL_PLAT=${POLICY}_${PLATFORM}
+ENV LC_ALL=en_US.UTF-8 LANG=en_US.UTF-8 LANGUAGE=en_US.UTF-8
+ENV DEVTOOLSET_ROOTPATH=${DEVTOOLSET_ROOTPATH}
+ENV LD_LIBRARY_PATH=${LD_LIBRARY_PATH_ARG}
+ENV PATH=${PREPEND_PATH}${PATH}
+ENV PKG_CONFIG_PATH=/usr/local/lib/pkgconfig
+
+# first copy the fixup mirrors script, keep the script around
+COPY build_scripts/fixup-mirrors.sh /usr/bin/fixup-mirrors
+
+# setup entrypoint, this will wrap commands with `linux32` with i686 images
+COPY build_scripts/install-entrypoint.sh build_scripts/update-system-packages.sh /build_scripts/
+RUN bash /build_scripts/install-entrypoint.sh && rm -rf /build_scripts
+COPY manylinux-entrypoint /usr/local/bin/manylinux-entrypoint
+ENTRYPOINT ["manylinux-entrypoint"]
+
+COPY build_scripts/install-runtime-packages.sh build_scripts/update-system-packages.sh /build_scripts/
+RUN manylinux-entrypoint /build_scripts/install-runtime-packages.sh && rm -rf /build_scripts/
+
+COPY build_scripts/build_utils.sh /build_scripts/
+
+COPY build_scripts/install-autoconf.sh /build_scripts/
+RUN export AUTOCONF_ROOT=autoconf-2.71 && \
+    export AUTOCONF_HASH=431075ad0bf529ef13cb41e9042c542381103e80015686222b8a9d4abef42a1c && \
+    export AUTOCONF_DOWNLOAD_URL=http://ftp.gnu.org/gnu/autoconf && \
+    manylinux-entrypoint /build_scripts/install-autoconf.sh
+
+COPY build_scripts/install-automake.sh /build_scripts/
+RUN export AUTOMAKE_ROOT=automake-1.16.3 && \
+    export AUTOMAKE_HASH=ce010788b51f64511a1e9bb2a1ec626037c6d0e7ede32c1c103611b9d3cba65f && \
+    export AUTOMAKE_DOWNLOAD_URL=http://ftp.gnu.org/gnu/automake && \
+    manylinux-entrypoint /build_scripts/install-automake.sh
+
+COPY build_scripts/install-libtool.sh /build_scripts/
+RUN export LIBTOOL_ROOT=libtool-2.4.6 && \
+    export LIBTOOL_HASH=e3bd4d5d3d025a36c21dd6af7ea818a2afcd4dfc1ea5a17b39d7854bcd0c06e3 && \
+    export LIBTOOL_DOWNLOAD_URL=http://ftp.gnu.org/gnu/libtool && \
+    manylinux-entrypoint /build_scripts/install-libtool.sh
+
+COPY build_scripts/install-patchelf.sh /build_scripts/
+RUN export PATCHELF_VERSION=0.12 && \
+    export PATCHELF_HASH=3dca33fb862213b3541350e1da262249959595903f559eae0fbc68966e9c3f56 && \
+    export PATCHELF_DOWNLOAD_URL=https://github.com/NixOS/patchelf/archive && \
+    manylinux-entrypoint /build_scripts/install-patchelf.sh
+
+COPY build_scripts/install-libxcrypt.sh /build_scripts/
+RUN export LIBXCRYPT_VERSION=4.4.17 && \
+    export LIBXCRYPT_HASH=7665168d0409574a03f7b484682e68334764c29c21ca5df438955a381384ca07 && \
+    export LIBXCRYPT_DOWNLOAD_URL=https://github.com/besser82/libxcrypt/archive && \
+    manylinux-entrypoint /build_scripts/install-libxcrypt.sh
+
+
+FROM runtime_base AS build_base
+COPY build_scripts/install-build-packages.sh /build_scripts/
+RUN manylinux-entrypoint /build_scripts/install-build-packages.sh
+
+
+FROM build_base AS build_git
+COPY build_scripts/build-git.sh /build_scripts/
+RUN export GIT_ROOT=git-2.31.1 && \
+    export GIT_HASH=46d37c229e9d786510e0c53b60065704ce92d5aedc16f2c5111e3ed35093bfa7 && \
+    export GIT_DOWNLOAD_URL=https://www.kernel.org/pub/software/scm/git && \
+    manylinux-entrypoint /build_scripts/build-git.sh
+
+
+FROM build_base AS build_cmake
+COPY build_scripts/build-cmake.sh /build_scripts/
+RUN export CMAKE_VERSION=3.20.2 && \
+    export CMAKE_HASH=aecf6ecb975179eb3bb6a4a50cae192d41e92b9372b02300f9e8f1d5f559544e && \
+    export CMAKE_DOWNLOAD_URL=https://github.com/Kitware/CMake/releases/download && \
+    manylinux-entrypoint /build_scripts/build-cmake.sh
+
+
+FROM build_base AS build_swig
+COPY build_scripts/build-swig.sh /build_scripts/
+RUN export SWIG_ROOT=swig-4.0.2 && \
+    export SWIG_HASH=d53be9730d8d58a16bf0cbd1f8ac0c0c3e1090573168bfa151b01eb47fa906fc && \
+    export SWIG_DOWNLOAD_URL=https://sourceforge.net/projects/swig/files/swig/${SWIG_ROOT} && \
+    export PCRE_ROOT=pcre-8.44 && \
+    export PCRE_HASH=aecafd4af3bd0f3935721af77b889d9024b2e01d96b58471bd91a3063fb47728 && \
+    export PCRE_DOWNLOAD_URL=https://ftp.pcre.org/pub/pcre && \
+    manylinux-entrypoint /build_scripts/build-swig.sh
+
+
+FROM build_base AS build_cpython
+COPY build_scripts/build-sqlite3.sh /build_scripts/
+RUN export SQLITE_AUTOCONF_ROOT=sqlite-autoconf-3350500 && \
+    export SQLITE_AUTOCONF_HASH=f52b72a5c319c3e516ed7a92e123139a6e87af08a2dc43d7757724f6132e6db0 && \
+    export SQLITE_AUTOCONF_DOWNLOAD_URL=https://www.sqlite.org/2021 && \
+    manylinux-entrypoint /build_scripts/build-sqlite3.sh
+
+COPY build_scripts/build-openssl.sh /build_scripts/
+RUN export OPENSSL_ROOT=openssl-1.1.1k && \
+    export OPENSSL_HASH=892a0875b9872acd04a9fde79b1f943075d5ea162415de3047c327df33fbaee5 && \
+    export OPENSSL_DOWNLOAD_URL=https://www.openssl.org/source && \
+    manylinux-entrypoint /build_scripts/build-openssl.sh
+
+COPY build_scripts/build-cpython.sh /build_scripts/
+
+
+FROM build_cpython AS build_cpython36
+COPY build_scripts/cpython-pubkeys.txt /build_scripts/cpython-pubkeys.txt
+
+RUN manylinux-entrypoint /build_scripts/build-cpython.sh 3.6.13
+
+
+FROM build_cpython AS build_cpython37
+COPY build_scripts/cpython-pubkeys.txt /build_scripts/cpython-pubkeys.txt
+
+RUN manylinux-entrypoint /build_scripts/build-cpython.sh 3.7.10
+
+
+FROM build_cpython AS build_cpython38
+COPY build_scripts/ambv-pubkey.txt /build_scripts/cpython-pubkeys.txt
+
+RUN manylinux-entrypoint /build_scripts/build-cpython.sh 3.8.10
+
+
+FROM build_cpython AS build_cpython39
+COPY build_scripts/ambv-pubkey.txt /build_scripts/cpython-pubkeys.txt
+
+RUN manylinux-entrypoint /build_scripts/build-cpython.sh 3.9.5
+
+
+FROM build_cpython AS build_cpython310
+COPY build_scripts/cpython-pubkey-310-311.txt /build_scripts/cpython-pubkeys.txt
+
+RUN manylinux-entrypoint /build_scripts/build-cpython.sh 3.10.0b1
+
+
+FROM build_cpython AS all_cpython
+COPY --from=build_cpython36 /opt/_internal /opt/_internal/
+COPY --from=build_cpython37 /opt/_internal /opt/_internal/
+COPY --from=build_cpython38 /opt/_internal /opt/_internal/
+COPY --from=build_cpython39 /opt/_internal /opt/_internal/
+COPY --from=build_cpython310 /opt/_internal /opt/_internal/
+RUN hardlink -cv /opt/_internal
+
+
+FROM runtime_base
+COPY --from=build_git /manylinux-rootfs /
+COPY --from=build_cmake /manylinux-rootfs /
+COPY --from=build_swig /manylinux-rootfs /
+COPY --from=build_cpython /manylinux-rootfs /
+COPY --from=all_cpython /opt/_internal /opt/_internal/
+COPY build_scripts/finalize.sh build_scripts/update-system-packages.sh build_scripts/python-tag-abi-tag.py build_scripts/requirements.txt build_scripts/requirements-tools.txt /build_scripts/
+RUN manylinux-entrypoint /build_scripts/finalize.sh && rm -rf /build_scripts
+
+ENV SSL_CERT_FILE=/opt/_internal/certs.pem
+
+CMD ["/bin/bash"]
+
+#Build manylinux2014 docker image end
+
+#Install TensorRT 7.1.3
+RUN yum install -y wget
+RUN cd /tmp &&\
+    wget --no-check-certificate https://developer.download.nvidia.com/compute/machine-learning/repos/rhel7/x86_64/nvidia-machine-learning-repo-rhel7-1.0.0-1.x86_64.rpm &&\
+    rpm -Uvh nvidia-machine-learning-repo-*.rpm
+RUN yum install -y libnvinfer7-7.1.3-1.cuda11.0 libnvparsers7-7.1.3-1.cuda11.0 libnvinfer-plugin7-7.1.3-1.cuda11.0 libnvonnxparsers7-7.1.3-1.cuda11.0 libnvinfer-devel-7.1.3-1.cuda11.0 libnvparsers-devel-7.1.3-1.cuda11.0 libnvinfer-plugin-devel-7.1.3-1.cuda11.0
+
+#Add our own dependencies
+ADD scripts /tmp/scripts
+RUN cd /tmp/scripts && /tmp/scripts/manylinux/install_centos.sh && /tmp/scripts/manylinux/install_deps.sh && rm -rf /tmp/scripts
+
+ARG BUILD_UID=1001
+ARG BUILD_USER=onnxruntimedev
+RUN adduser --uid $BUILD_UID $BUILD_USER
+WORKDIR /home/$BUILD_USER
+USER $BUILD_USER
+ENV PATH /usr/local/gradle/bin:/usr/local/dotnet:$PATH
--- a/tools/ci_build/github/linux/docker/Dockerfile.ubuntu_tensorrt7_1
+++ b/tools/ci_build/github/linux/docker/Dockerfile.ubuntu_tensorrt7_1
@ -0,0 +1,28 @@
+# Tag: nvcr.io/nvidia/tensorrt:20.07.1-py3
+# Label: com.nvidia.cuda.version: 11.0.194
+# Label: com.nvidia.cudnn.version: 8.0.1
+# Ubuntu 18.04
+FROM nvcr.io/nvidia/tensorrt:20.07.1-py3
+
+ARG PYTHON_VERSION=3.6
+
+ADD scripts /tmp/scripts
+RUN /tmp/scripts/install_ubuntu.sh -p $PYTHON_VERSION && /tmp/scripts/install_os_deps.sh && /tmp/scripts/install_python_deps.sh -p $PYTHON_VERSION && rm -rf /tmp/scripts \
+    && rm /usr/local/bin/cmake && rm /usr/local/bin/ctest && rm -r /usr/local/share/cmake-3.14
+
+WORKDIR /root
+
+# Allow configure to pick up GDK and CuDNN where it expects it.
+# (Note: $CUDNN_VERSION is defined by NVidia's base image)
+RUN _CUDNN_VERSION=$(echo $CUDNN_VERSION | cut -d. -f1-2) && \
+    mkdir -p /usr/local/cudnn-$_CUDNN_VERSION/cuda/include && \
+    ln -s /usr/include/cudnn.h /usr/local/cudnn-$_CUDNN_VERSION/cuda/include/cudnn.h && \
+    mkdir -p /usr/local/cudnn-$_CUDNN_VERSION/cuda/lib64 && \
+    ln -s /etc/alternatives/libcudnn_so /usr/local/cudnn-$_CUDNN_VERSION/cuda/lib64/libcudnn.so && \
+    ln -s /usr/local/cudnn{-$_CUDNN_VERSION,}
+
+ARG BUILD_USER=onnxruntimedev
+ARG BUILD_UID=1000
+WORKDIR /home/$BUILD_USER
+RUN adduser --gecos 'onnxruntime Build User' --disabled-password $BUILD_USER --uid $BUILD_UID
+USER $BUILD_USER
--- a/tools/ci_build/github/linux/run_build.sh
+++ b/tools/ci_build/github/linux/run_build.sh
@ -53,7 +53,15 @@ else
            --use_cuda \
            --cuda_home /usr/local/cuda \
            --cudnn_home /usr/local/cudnn-$_CUDNN_VERSION/cuda $BUILD_EXTR_PAR
-    elif [ $BUILD_DEVICE = "tensorrt" ]; then
+    elif [[ $BUILD_DEVICE = "tensorrt"* ]]; then
+        if [ $BUILD_DEVICE = "tensorrt-v7.1" ]; then
+            CUR_PWD=$(pwd)
+            cd $SCRIPT_DIR/../../../../cmake/external/onnx-tensorrt/
+            git remote update
+            git checkout 7.1
+            cd $CUR_PWD
+            COMMON_BUILD_ARGS=${COMMON_BUILD_ARGS/"--skip_submodule_sync"}
+        fi
        _CUDNN_VERSION=$(echo $CUDNN_VERSION | cut -d. -f1-2)
        python3 $SCRIPT_DIR/../../build.py --build_dir /build \
            --config Release $COMMON_BUILD_ARGS \
--- a/tools/ci_build/github/linux/run_dockerbuild.sh
+++ b/tools/ci_build/github/linux/run_dockerbuild.sh
@ -83,10 +83,16 @@ elif [ $BUILD_DEVICE = "gpu" ]; then
        $GET_DOCKER_IMAGE_CMD --repository "onnxruntime-$IMAGE" \
            --docker-build-args="--build-arg BASEIMAGE=nvcr.io/nvidia/cuda:11.1.1-cudnn8-devel-${BUILD_OS} --build-arg BUILD_USER=onnxruntimedev --build-arg BUILD_UID=$(id -u) --build-arg PYTHON_VERSION=${PYTHON_VER} --build-arg INSTALL_DEPS_EXTRA_ARGS=\"${INSTALL_DEPS_EXTRA_ARGS}\" --build-arg USE_CONDA=${USE_CONDA} --network=host" \
            --dockerfile Dockerfile.ubuntu_gpu_training --context .
-elif [ $BUILD_DEVICE = "tensorrt" ]; then
-        # TensorRT container release 20.12
-        IMAGE="$BUILD_OS-cuda11.1-cudnn8.0-tensorrt7.2"
-        DOCKER_FILE=Dockerfile.ubuntu_tensorrt
+elif [[ $BUILD_DEVICE = "tensorrt"* ]]; then
+        if [ $BUILD_DEVICE = "tensorrt-v7.1" ]; then
+            # TensorRT container release 20.07
+            IMAGE="$BUILD_OS-cuda11.0-cudnn8.0-tensorrt7.1"
+            DOCKER_FILE=Dockerfile.ubuntu_tensorrt7_1
+        else
+            # TensorRT container release 20.12
+            IMAGE="$BUILD_OS-cuda11.1-cudnn8.0-tensorrt7.2"
+            DOCKER_FILE=Dockerfile.ubuntu_tensorrt
+        fi
        $GET_DOCKER_IMAGE_CMD --repository "onnxruntime-$IMAGE" \
            --docker-build-args="--build-arg BUILD_USER=onnxruntimedev --build-arg BUILD_UID=$(id -u) --build-arg PYTHON_VERSION=${PYTHON_VER} --network=host --build-arg POLICY=manylinux2014 --build-arg PLATFORM=x86_64  --build-arg DEVTOOLSET_ROOTPATH=/opt/rh/devtoolset-9/root --build-arg PREPEND_PATH=/opt/rh/devtoolset-9/root/usr/bin: --build-arg LD_LIBRARY_PATH_ARG=/opt/rh/devtoolset-9/root/usr/lib64:/opt/rh/devtoolset-9/root/usr/lib:/opt/rh/devtoolset-9/root/usr/lib64/dyninst:/opt/rh/devtoolset-9/root/usr/lib/dyninst:/usr/local/lib64" \
            --dockerfile $DOCKER_FILE --context .