Start with LTO patch

Add LTO options to the Dockerfile. This includes a patch that is applied
to the openvino repo after it's downloaded.

Author: RH-steve-grubb

.bazelrc

@@ -17,6 +17,7 @@
 # Optimizations used for TF Serving release builds.
 build:release --copt=-mavx
 build:release --copt=-msse4.2
+build:linux --copt="-march=haswell"
 
 # Options used to build with CUDA.
 build:cuda --crosstool_top=@local_config_cuda//crosstool:toolchain

Dockerfile.redhat

@@ -26,6 +26,9 @@ SHELL ["/bin/bash", "-xo", "pipefail", "-c"]
 
 ARG JOBS=8
 ARG VERBOSE_LOGS=OFF
+ARG LTO_ENABLE=OFF
+ARG LTO_CXX_FLAGS=""
+ARG LTO_LD_FLAGS="-Wl,-z,relro"
 
 # hadolint ignore=DL3041
 RUN echo -e "max_parallel_downloads=8\nretries=50" >> /etc/dnf/dnf.conf && \
@@ -51,7 +54,10 @@ WORKDIR /boost
 RUN wget -nv https://sourceforge.net/projects/boost/files/boost/1.68.0/boost_1_68_0.tar.gz && \
 	if [ "$VERBOSE_LOGS" == "ON" ] ; then export BVERBOSE="-d+2" ; else export BVERBOSE="" ; fi && \
 	tar xf boost_1_68_0.tar.gz && cd boost_1_68_0 && ./bootstrap.sh && \
-	./b2 -j ${JOBS} cxxstd=17 link=static cxxflags='-fPIC' cflags='-fPIC' \
+	./b2 -j ${JOBS} cxxstd=17 link=static variant=release \
+		cxxflags="-fPIC $LTO_CXX_FLAGS" \
+		cflags="-fPIC $LTO_CXX_FLAGS" \
+		linkflags="$LTO_LD_FLAGS" \
 		--with-chrono --with-date_time --with-filesystem \
 		--with-program_options --with-system \
 		--with-random --with-thread --with-atomic --with-regex \
@@ -65,7 +71,9 @@ COPY third_party/pugixml /ovms/third_party/pugixml
 RUN git clone -b v1.13 https://github.com/zeux/pugixml && \
     cd pugixml && \
     patch -p1 < /ovms/third_party/pugixml/pugixml_v1.13_flags.patch && \
-    cmake -DBUILD_SHARED_LIBS=ON -DCMAKE_VERBOSE_MAKEFILE=${VERBOSE_LOGS} && \
+    cmake -DBUILD_SHARED_LIBS=ON -DCMAKE_VERBOSE_MAKEFILE=${VERBOSE_LOGS} \
+    -DCMAKE_CXX_FLAGS="${LTO_CXX_FLAGS}" \
+    -DCMAKE_SHARED_LINKER_FLAGS="${LTO_LD_FLAGS}" && \
     make all -j ${JOBS} && \
     cp -P libpugixml.so* /usr/lib64/
 
@@ -79,10 +87,10 @@ RUN git clone --recurse-submodules --depth 1 --branch v2.10.16 https://github.co
     patch -d /azure/azure-storage-cpp/ -p1 </ovms/third_party/azure/azure_sdk.patch
 
 WORKDIR /azure/cpprestsdk/Release/build.release
-RUN cmake .. -DCMAKE_VERBOSE_MAKEFILE=${VERBOSE_LOGS} -DCMAKE_CXX_FLAGS="-fPIC" -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON -DBoost_USE_STATIC_RUNTIME=ON -DBoost_USE_STATIC_LIBS=ON -DWERROR=OFF -DBUILD_SAMPLES=OFF -DBUILD_TESTS=OFF && make --jobs=$JOBS install
+RUN cmake .. -DCMAKE_VERBOSE_MAKEFILE=${VERBOSE_LOGS} -DCMAKE_CXX_FLAGS="-fPIC ${LTO_CXX_FLAGS}" -DCMAKE_SHARED_LINKER_FLAGS="${LTO_LD_FLAGS}" -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON -DBoost_USE_STATIC_RUNTIME=ON -DBoost_USE_STATIC_LIBS=ON -DWERROR=OFF -DBUILD_SAMPLES=OFF -DBUILD_TESTS=OFF && make --jobs=$JOBS install
 
 WORKDIR /azure/azure-storage-cpp/Microsoft.WindowsAzure.Storage/build.release
-RUN CASABLANCA_DIR=/azure/cpprestsdk cmake .. -DCMAKE_CXX_FLAGS="-fPIC" -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON -DBoost_USE_STATIC_RUNTIME=ON -DBoost_USE_STATIC_LIBS=ON -DCMAKE_VERBOSE_MAKEFILE=${VERBOSE_LOGS} && make --jobs=$JOBS && make --jobs=$JOBS install
+RUN CASABLANCA_DIR=/azure/cpprestsdk cmake .. -DCMAKE_CXX_FLAGS="-fPIC ${LTO_CXX_FLAGS}" -DCMAKE_SHARED_LINKER_FLAGS="${LTO_LD_FLAGS}" -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON -DBoost_USE_STATIC_RUNTIME=ON -DBoost_USE_STATIC_LIBS=ON -DCMAKE_VERBOSE_MAKEFILE=${VERBOSE_LOGS} && make --jobs=$JOBS && make --jobs=$JOBS install
 ####### End of Azure SDK
 
 ####### Build OpenCV
@@ -99,6 +107,7 @@ SHELL ["/bin/bash", "-xo", "pipefail", "-c"]
 
 ARG JOBS=40
 ARG VERBOSE_LOGS=OFF
+ARG LTO_ENABLE=OFF
 
 # hadolint ignore=DL3041
 RUN dnf install -y https://rpmfind.net/linux/almalinux/8.10/PowerTools/x86_64/os/Packages/opencl-headers-2.2-1.20180306gite986688.el8.noarch.rpm && \
@@ -175,9 +184,12 @@ RUN dnf install -y https://dl.fedoraproject.org/pub/epel/epel-release-latest-9.n
 # hadolint ignore=DL3003
 RUN if [ "$ov_use_binary" == "0" ] ; then true ; else exit 0 ; fi ; git clone https://github.com/$ov_source_org/openvino.git /openvino && cd /openvino && git checkout $ov_source_branch && git submodule update --init --recursive
 RUN if [ "$ov_use_binary" == "0" ]; then true ; else exit 0 ; fi ; if ! [[ $debug_bazel_flags == *"py_off"* ]]; then true ; else exit 0 ; fi ; pip3 install --no-cache-dir -r /openvino/src/bindings/python/wheel/requirements-dev.txt
+WORKDIR /openvino
+COPY openvino-lto.patch .
+RUN if [ "$ov_use_binary" == "0" ]; then patch -p1 < openvino-lto.patch ; rm -f openvino-lto.patch ; fi
 WORKDIR /openvino/build
-RUN if [ "$ov_use_binary" == "0" ] && [[ $debug_bazel_flags == *"PYTHON_DISABLE=1"* ]]; then true ; else exit 0 ; fi ; if ! [[ $debug_bazel_flags == *"PYTHON_DISABLE=1"* ]]; then true ; else exit 0 ; fi ; cmake -DCMAKE_BUILD_TYPE=$CMAKE_BUILD_TYPE -DCMAKE_VERBOSE_MAKEFILE="${VERBOSE_LOGS}" -DENABLE_PYTHON=ON -DENABLE_INTEL_NPU=OFF -DENABLE_SAMPLES=0 -DCMAKE_CXX_FLAGS=" -Wno-error=parentheses "  ..
-RUN if [ "$ov_use_binary" == "0" ] ; then true ; else exit 0 ; fi ; cmake -DCMAKE_BUILD_TYPE="$CMAKE_BUILD_TYPE" -DCMAKE_VERBOSE_MAKEFILE="${VERBOSE_LOGS}" -DENABLE_SAMPLES=0 -DENABLE_INTEL_NPU=OFF -DCMAKE_CXX_FLAGS=" -Wno-error=parentheses " ..
+RUN if [ "$ov_use_binary" == "0" ] && [[ $debug_bazel_flags == *"PYTHON_DISABLE=1"* ]]; then true ; else exit 0 ; fi ; if ! [[ $debug_bazel_flags == *"PYTHON_DISABLE=1"* ]]; then true ; else exit 0 ; fi ; cmake -DCMAKE_BUILD_TYPE=$CMAKE_BUILD_TYPE -DCMAKE_VERBOSE_MAKEFILE="${VERBOSE_LOGS}" -DENABLE_LTO=${LTO_ENABLE} -DENABLE_PYTHON=ON -DENABLE_INTEL_NPU=OFF -DENABLE_SAMPLES=0 -DCMAKE_CXX_FLAGS=" -Wno-error=parentheses  ${LTO_CXX_FLAGS} " -DCMAKE_SHARED_LINKER_FLAGS="${LTO_LD_FLAGS}" ..
+RUN if [ "$ov_use_binary" == "0" ] ; then true ; else exit 0 ; fi ; cmake -DCMAKE_BUILD_TYPE="$CMAKE_BUILD_TYPE" -DCMAKE_VERBOSE_MAKEFILE="${VERBOSE_LOGS}" -DENABLE_LTO=${LTO_ENABLE} -DENABLE_SAMPLES=0 -DENABLE_INTEL_NPU=OFF -DCMAKE_CXX_FLAGS=" -Wno-error=parentheses ${LTO_CXX_FLAGS} " -DCMAKE_SHARED_LINKER_FLAGS="${LTO_LD_FLAGS}" ..
 RUN if [ "$ov_use_binary" == "0" ] ; then true ; else exit 0 ; fi ; make --jobs=$JOBS
 RUN if [ "$ov_use_binary" == "0" ] ; then true ; else exit 0 ; fi ; make install
 RUN if [ "$ov_use_binary" == "0" ] ; then true ; else exit 0 ; fi ; \
@@ -225,7 +237,7 @@ ARG ov_tokenizers_branch=85be884a69f10270703f81f970a5ee596a4c8df7
 # hadolint ignore=DL3003
 RUN git clone https://github.com/openvinotoolkit/openvino_tokenizers.git /openvino_tokenizers && cd /openvino_tokenizers && git checkout $ov_tokenizers_branch && git submodule update --init --recursive
 WORKDIR /openvino_tokenizers/build
-RUN cmake .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_VERBOSE_MAKEFILE="${VERBOSE_LOGS}"  && cmake --build . --parallel $JOBS ; cp /openvino_tokenizers/build/src/lib*.so /opt/intel/openvino/runtime/lib/intel64/
+RUN cmake .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_VERBOSE_MAKEFILE="${VERBOSE_LOGS}" -DCMAKE_CXX_FLAGS=" ${LTO_CXX_FLAGS} " -DCMAKE_SHARED_LINKER_FLAGS="${LTO_LD_FLAGS}"  && cmake --build . --parallel $JOBS ; cp /openvino_tokenizers/build/src/lib*.so /opt/intel/openvino/runtime/lib/intel64/
 
 # Build OpenVINO Model Server
 WORKDIR /ovms

openvino-lto.patch

@@ -0,0 +1,68 @@
+diff --git a/CMakeLists.txt b/CMakeLists.txt
+index 2187deb8e8..ad15a73395 100644
+--- a/CMakeLists.txt
++++ b/CMakeLists.txt
+@@ -65,6 +65,7 @@ if(ENABLE_COVERAGE)
+     include(cmake/coverage.cmake)
+ endif()
+ 
++set(CMAKE_INTERPROCEDURAL_OPTIMIZATION_RELEASE ${ENABLE_LTO})
+ if(APPLE AND CMAKE_OSX_DEPLOYMENT_TARGET AND
+     CMAKE_OSX_DEPLOYMENT_TARGET VERSION_LESS 10.15)
+     message(FATAL_ERROR "OpenVINO requires MACOSX_DEPLOYMENT_TARGET at least 10.15, specified ${CMAKE_OSX_DEPLOYMENT_TARGET}")
+diff --git a/src/frontends/common/CMakeLists.txt b/src/frontends/common/CMakeLists.txt
+index e4d2fe63e1..eace2a5593 100644
+--- a/src/frontends/common/CMakeLists.txt
++++ b/src/frontends/common/CMakeLists.txt
+@@ -52,7 +52,8 @@ target_link_libraries(${TARGET_NAME}_obj PRIVATE openvino::util openvino::core::
+ 
+ # TODO: fix lto
+ set_target_properties(${TARGET_NAME}_obj PROPERTIES
+-    INTERPROCEDURAL_OPTIMIZATION_RELEASE OFF)
++    INTERPROCEDURAL_OPTIMIZATION_RELEASE ${ENABLE_LTO})
++ov_add_compiler_flags(-march=haswell)
+ 
+ set(FRONTEND_LIB_SUFFIX "${FRONTEND_NAME_SUFFIX}${OV_BUILD_POSTFIX}")
+ if(APPLE)
+--- a/thirdparty/onnx/CMakeLists.txt
++++ b/thirdparty/onnx/CMakeLists.txt
+@@ -21,7 +21,10 @@ endif()
+ if(CMAKE_CXX_COMPILER_ID STREQUAL "MSVC")
+     # 4244 conversion from 'XXX' to 'YYY', possible loss of data
+     ov_add_compiler_flags(/wd4244)
++else()
++    ov_add_compiler_flags(-march=haswell)
+ endif()
++set(CMAKE_INTERPROCEDURAL_OPTIMIZATION_RELEASE ${ENABLE_LTO})
+ 
+ set(PYTHON_EXECUTABLE "${Python3_EXECUTABLE}")
+ set(ONNX_USE_PROTOBUF_SHARED_LIBS OFF CACHE BOOL "Use dynamic protobuf by ONNX library" FORCE)
+diff --git a/thirdparty/protobuf/CMakeLists.txt b/thirdparty/protobuf/CMakeLists.txt
+index a16bd56d11..2829926e68 100644
+--- a/thirdparty/protobuf/CMakeLists.txt
++++ b/thirdparty/protobuf/CMakeLists.txt
+@@ -10,6 +10,7 @@ set(BUILD_SHARED_LIBS OFF)
+ 
+ if(CMAKE_COMPILER_IS_GNUCXX OR OV_COMPILER_IS_CLANG OR (OV_COMPILER_IS_INTEL_LLVM AND UNIX))
+     ov_add_compiler_flags(-Wno-all)
++    ov_add_compiler_flags(-march=haswell)
+     if(CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 11 AND CMAKE_COMPILER_IS_GNUCXX)
+         ov_add_compiler_flags(-Wno-stringop-overflow)
+     endif()
+@@ -48,6 +49,7 @@ else()
+     set(protobuf_dir protobuf/cmake)
+ endif()
+ 
++set(CMAKE_INTERPROCEDURAL_OPTIMIZATION_RELEASE ${ENABLE_LTO})
+ add_subdirectory(${protobuf_dir} EXCLUDE_FROM_ALL)
+ get_directory_property(protobuf_VERSION DIRECTORY ${protobuf_dir} DEFINITION protobuf_VERSION)
+ 
+@@ -63,7 +65,7 @@ if(CMAKE_COMPILER_IS_GNUCXX OR OV_COMPILER_IS_CLANG OR OV_COMPILER_IS_INTEL_LLVM
+             CXX_VISIBILITY_PRESET default
+             C_VISIBILITY_PRESET default
+             VISIBILITY_INLINES_HIDDEN OFF
+-            INTERPROCEDURAL_OPTIMIZATION_RELEASE OFF)
++            INTERPROCEDURAL_OPTIMIZATION_RELEASE ${ENABLE_LTO})
+     endif()
+     ov_disable_all_warnings(${_protoc_libs} libprotobuf-lite)
+     set_target_properties(libprotobuf-lite PROPERTIES

third_party/opencv/opencv_cmake_flags.txt

@@ -1,6 +1,7 @@
 -D BUILD_LIST=core,improc,imgcodecs,calib3d,features2d,highgui,imgproc,video,videoio,optflow \
 -D OPENCV_EXTRA_MODULES_PATH=/opt/opencv_contrib_repo/modules \
 -D CMAKE_BUILD_TYPE=Release \
+-D ENABLE_LTO=${LTO_ENABLE} \
 -D CMAKE_INSTALL_PREFIX=/opt/opencv \
 -D OPENCV_LIB_INSTALL_PATH=lib \
 -D BUILD_TESTS=OFF \
@@ -39,4 +40,4 @@
 -D WITH_JPEG=ON \
 -D WITH_PNG=ON \
 -D WITH_TIFF=OFF \
--D WITH_OPENEXR=OFF
+-D WITH_OPENEXR=OFF