Merge branch 'master' into system-protobuf

Author: FloopCZ

Dockerfiles/BAZEL_VERSION

@@ -1 +1 @@
-3.1.0
+3.7.2

Dockerfiles/archlinux-cuda

@@ -10,4 +10,4 @@ RUN ./tensorflow_cc/Dockerfiles/install-archlinux.sh --cuda
 
 ENV NVIDIA_VISIBLE_DEVICES all
 ENV NVIDIA_DRIVER_CAPABILITIES compute,utility
-ENV NVIDIA_REQUIRE_CUDA "cuda>=10.1"
+ENV NVIDIA_REQUIRE_CUDA "cuda>=11.1"

Dockerfiles/install-archlinux.sh

@@ -59,5 +59,5 @@ fi
 # cleanup packages
 rm -rvf /usr/local/lib/bazel/
 rm -vf /usr/local/bin/bazel
-pacman --noconfirm -R jdk11-openjdk python python-numpy
+pacman --noconfirm -R jdk11-openjdk
 pacman --noconfirm -Rns $(pacman -Qtdq)

Dockerfiles/install-common.sh

@@ -19,11 +19,12 @@ chmod go+rX tensorflow_cc/tensorflow_cc/build
 cd tensorflow_cc/tensorflow_cc/build
 
 # build and install
-cmake ..
+cmake -DLOCAL_RAM_RESOURCES=2048 -DLOCAL_CPU_RESOURCES=2 ..
 make
 rm -rf /home/tensorflow_cc/.cache
 rm -rf /root/.cache
 make install
+ldconfig
 cd "$cwd"
 rm -rf tensorflow_cc/tensorflow_cc/build
 

Dockerfiles/ubuntu-cuda

@@ -1,4 +1,4 @@
-FROM nvidia/cuda:10.1-cudnn7-devel-ubuntu18.04
+FROM nvidia/cuda:11.1-cudnn8-devel-ubuntu18.04
 
 # copy the contents of this repository to the container
 COPY . tensorflow_cc

README.md

@@ -1,6 +1,6 @@
 # tensorflow_cc
-[![Build Status](http://elm.floop.cz:8080/buildStatus/icon?job=tensorflow_cc)](http://elm.floop.cz:8080/job/tensorflow_cc/)
-[![TF version](https://img.shields.io/badge/TF%20version-2.3.1-brightgreen.svg)]()
+[![Build Status](http://oak.floop.cz:8080/buildStatus/icon?job=tensorflow_cc)](http://oak.floop.cz:8080/job/tensorflow_cc/)
+[![TF version](https://img.shields.io/badge/TF%20version-2.5.0-brightgreen.svg)]()
 
 This repository makes possible the usage of the [TensorFlow C++](https://www.tensorflow.org/api_docs/cc/) API from the outside of the TensorFlow source code folders and without the use of the [Bazel](https://bazel.build/) build system.
 
@@ -46,12 +46,13 @@ sudo apt-get install cmake curl g++-7 git python3-dev python3-numpy sudo wget
 
 In order to build the TensorFlow itself, the build procedure also requires [Bazel](https://bazel.build/):
 ```
-curl https://bazel.build/bazel-release.pub.gpg | sudo apt-key add -
+curl -fsSL https://bazel.build/bazel-release.pub.gpg | gpg --dearmor > bazel.gpg
+sudo mv bazel.gpg /etc/apt/trusted.gpg.d/
 echo "deb [arch=amd64] https://storage.googleapis.com/bazel-apt stable jdk1.8" | sudo tee /etc/apt/sources.list.d/bazel.list
 sudo apt-get update && sudo apt-get install bazel
 ```
 
-If you require GPU support on Ubuntu, please also install NVIDIA CUDA Toolkit (>=10.1), NVIDIA drivers, cuDNN, and `cuda-command-line-tools` package.
+If you require GPU support on Ubuntu, please also install NVIDIA CUDA Toolkit (>=11.1), NVIDIA drivers, cuDNN, and `cuda-command-line-tools` package.
 The build procedure will automatically detect CUDA if it is installed in `/opt/cuda` or `/usr/local/cuda` directories.
 
 ##### Arch Linux:
@@ -66,7 +67,7 @@ sudo pacman -S cuda cudnn nvidia
 ```
 
 **Warning:** Newer versions of TensorFlow sometimes fail to build with the latest version of Bazel. You may wish
-to install an older version of Bazel (e.g., 3.1.0).
+to install an older version of Bazel (e.g., 3.7.2).
 
 #### 2) Clone this repository
 ```
@@ -82,13 +83,19 @@ mkdir build && cd build
 cmake ..
 make
 sudo make install
+sudo ldconfig
 ```
 
-**Warning:** Optimizations for Intel CPU generation `>=ivybridge` are enabled by default. If you have a
-processor that is older than `ivybridge` generation, you may wish to run `export CC_OPT_FLAGS="-march=native"`
+**Warning:** Optimizations for Intel CPU generation `>=haswell` are enabled by default. If you have a
+processor that is older than `haswell` generation, you may wish to run `export CC_OPT_FLAGS="-march=native"`
 before the build. This command provides the best possible optimizations for your current CPU generation, but
 it may cause the built library to be incompatible with older generations.
 
+**Warning:** In low-memory or many-cpu environments, the bazel scheduler can miss the resource consumption
+estimates and the build may be terminated by the out-of-memory killer.
+If that is your case, consider adding resource limit parameters to CMake, e.g.,
+`cmake -DLOCAL_RAM_RESOURCES=2048 -DLOCAL_CPU_RESOURCES=4 ..`
+
 #### 4) (Optional) Free disk space
 
 ```

tensorflow_cc/CMakeLists.txt

@@ -1,16 +1,18 @@
 cmake_minimum_required(VERSION 3.3 FATAL_ERROR)
+cmake_policy(SET CMP0048 NEW)  # Enable version parameter in project().
 file(STRINGS "${CMAKE_CURRENT_SOURCE_DIR}/PROJECT_VERSION" version)
 project(
   "tensorflow_cc"
-  VERSION "${version}"
+  VERSION ${version}
 )
 
-# If enabled, bazel has to be installed.
 option(ALLOW_CUDA "Try to find and use CUDA." ON)
 option(REQUIRE_CUDA "Make sure to find and use CUDA (implies ALLOW_CUDA)." OFF)
+set(LOCAL_RAM_RESOURCES 4096 CACHE STRING "The amount of local RAM resources passed to bazel scheduler (e.g., 4096).")
+set(LOCAL_CPU_RESOURCES HOST_CPUS CACHE STRING "The amount of local CPU cores passed to bazel scheduler (e.g., 2).")
 set(TENSORFLOW_TAG "v${version}" CACHE STRING "The tensorflow release tag to be checked out (default v${version}).")
 option(SYSTEM_PROTOBUF "Link to system protobuf." OFF)
-set(TARGET_CXX_STANDARD "cxx_std_11" CACHE STRING "C++ standard to be enforced when linking to TensorflowCC targets (e.g., cxx_std_11).")
+set(CMAKE_CXX_STANDARD 14 CACHE STRING "The C++ standard for building and linking the library (e.g., 14).")
 
 # -------------
 # CMake Options
@@ -22,9 +24,6 @@ set(CMAKE_LIBRARY_OUTPUT_DIRECTORY "${CMAKE_CURRENT_BINARY_DIR}/lib")
 include(CMakePackageConfigHelpers)
 set(CMAKECFG_INSTALL_DIR lib/cmake/TensorflowCC)
 
-set_property(CACHE TARGET_CXX_STANDARD PROPERTY STRINGS
-    "cxx_std_11" "cxx_std_14" "cxx_std_17" "cxx_std_20")
-
 # Configure the build_tensorflow script.
 configure_file("cmake/build_tensorflow.sh.in" "build_tensorflow.sh" @ONLY)
 
@@ -39,15 +38,15 @@ include(TensorflowBase)
 # ------------------------------
 
 add_library(tensorflow_cc INTERFACE)
-target_compile_features(tensorflow_cc INTERFACE ${TARGET_CXX_STANDARD})
+target_compile_features(tensorflow_cc INTERFACE "cxx_std_${CMAKE_CXX_STANDARD}")
 
 # The include folders are sometimes contained under bazel-bin/bin/ and sometimes just bazel-bin.
 target_include_directories(
   tensorflow_cc INTERFACE
-  "${CMAKE_INSTALL_PREFIX}/include/tensorflow/bazel-bin/tensorflow/include"
-  "${CMAKE_INSTALL_PREFIX}/include/tensorflow/bazel-bin/tensorflow/include/src"
-  "${CMAKE_INSTALL_PREFIX}/include/tensorflow/bazel-bin/bin/tensorflow/include"
-  "${CMAKE_INSTALL_PREFIX}/include/tensorflow/bazel-bin/bin/tensorflow/include/src"
+  $<INSTALL_INTERFACE:include/tensorflow/bazel-bin/tensorflow/include>
+  $<INSTALL_INTERFACE:include/tensorflow/bazel-bin/tensorflow/include/src>
+  $<INSTALL_INTERFACE:include/tensorflow/bazel-bin/bin/tensorflow/include>
+  $<INSTALL_INTERFACE:include/tensorflow/bazel-bin/bin/tensorflow/include/src>
 )
 target_link_libraries(
   tensorflow_cc INTERFACE

tensorflow_cc/PROJECT_VERSION

@@ -1 +1 @@
-2.3.1
+2.5.0

tensorflow_cc/cmake/TensorflowBase.cmake

@@ -19,4 +19,7 @@ ExternalProject_Add(
           COMMAND mkdir -p "${CMAKE_CURRENT_BINARY_DIR}/tensorflow/bazel-bin/bin/tensorflow/include/src"
           COMMAND touch "${CMAKE_CURRENT_BINARY_DIR}/tensorflow/bazel-bin/tensorflow/include/src/__placeholder__.h"
           COMMAND touch "${CMAKE_CURRENT_BINARY_DIR}/tensorflow/bazel-bin/bin/tensorflow/include/src/__placeholder__.h"
+  # Fix compilation with CUDA 11.3.
+  PATCH_COMMAND cp "${CMAKE_CURRENT_SOURCE_DIR}/patches/v2.5.0-Fix-NCCL-build-failure-with-CUDA-11.3.patch" .
+        COMMAND patch -p1 < v2.5.0-Fix-NCCL-build-failure-with-CUDA-11.3.patch
 )

tensorflow_cc/cmake/build_tensorflow.sh.in

@@ -7,19 +7,17 @@ function version_gt {
 }
 
 # configure environmental variables
-export CC_OPT_FLAGS=${CC_OPT_FLAGS:-"-march=ivybridge"}
+export CC_OPT_FLAGS=${CC_OPT_FLAGS:-"-march=haswell"}
 export TF_NEED_GCP=${TF_NEED_GCP:-0}
 export TF_NEED_HDFS=${TF_NEED_HDFS:-0}
 export TF_NEED_OPENCL=${TF_NEED_OPENCL:-0}
-export TF_NEED_OPENCL_SYCL=${TF_NEED_OPENCL_SYCL:-0}
 export TF_NEED_TENSORRT=${TF_NEED_TENSORRT:-0}
 export TF_NEED_NGRAPH=${TF_NEED_NGRAPH:-0}
 export TF_NEED_JEMALLOC=${TF_NEED_JEMALLOC:-1}
 export TF_NEED_VERBS=${TF_NEED_VERBS:-0}
 export TF_NEED_MKL=${TF_NEED_MKL:-1}
 export TF_DOWNLOAD_MKL=${TF_DOWNLOAD_MKL:-1}
 export TF_NEED_MPI=${TF_NEED_MPI:-0}
-export TF_ENABLE_XLA=${TF_ENABLE_XLA:-1}
 export TF_NEED_AWS=${TF_NEED_AWS:-0}
 export TF_NEED_GDR=${TF_NEED_GDR:-0}
 export TF_CUDA_CLANG=${TF_CUDA_CLANG:-0}
@@ -58,12 +56,14 @@ if [ "$cuda_allowed" == true ] && [ "$cuda_available" == true ]; then
     export TF_CUDA_COMPUTE_CAPABILITIES=${TF_CUDA_COMPUTE_CAPABILITIES:-"3.5,7.0"}  # default from configure.py
     export TF_CUDA_PATHS=${TF_CUDA_PATHS:-"/opt/cuda,/usr/local/cuda,/usr/local,/usr/cuda,/usr"}
     export TF_CUDA_VERSION="$(nvcc --version | sed -n 's/^.*release \(.*\),.*/\1/p')"
-    export TF_NCCL_VERSION="$(find /opt /usr -name 'libnccl.so.*' | tail -n1 | sed -r 's/^.*\.so\.//')"
-    export TF_CUDNN_VERSION="$(find /opt /usr -name 'libcudnn.so.*' | tail -n1 | sed -r 's/^.*\.so\.//')"
+    export TF_NCCL_VERSION="$(find /opt /usr -name 'libnccl.so.*' -path '*/cuda*' | tail -n1 | sed -r 's/^.*\.so\.//')"
+    export TF_CUDNN_VERSION="$(find /opt /usr -name 'libcudnn.so.*' -path '*/cuda*' | tail -n1 | sed -r 's/^.*\.so\.//')"
 
     # choose the right version of CUDA compiler
     if [ -z "$GCC_HOST_COMPILER_PATH" ]; then
-        if   hash gcc-9 2>/dev/null && version_gt 9.4 `gcc-9 -dumpversion`; then
+        if hash gcc-10 2>/dev/null && version_gt 10.3 `gcc-10 -dumpversion`; then
+            export GCC_HOST_COMPILER_PATH=${GCC_HOST_COMPILER_PATH:-"/usr/bin/gcc-10"}
+        elif hash gcc-9 2>/dev/null && version_gt 9.4 `gcc-9 -dumpversion`; then
             export GCC_HOST_COMPILER_PATH=${GCC_HOST_COMPILER_PATH:-"/usr/bin/gcc-9"}
         elif hash gcc-8 2>/dev/null && version_gt 8.5 `gcc-8 -dumpversion`; then
             export GCC_HOST_COMPILER_PATH=${GCC_HOST_COMPILER_PATH:-"/usr/bin/gcc-8"}
@@ -93,10 +93,24 @@ else
     export TF_NEED_CUDA=0
 fi
 
+# set bazel options for c++17 standard if necessary
+cxx_std_opts=""
+if [ "@CMAKE_CXX_STANDARD@" == 17 ]; then
+    echo "Using C++17 standard"
+    cxx_std_opts="--config=c++17_gcc"
+fi
+
 # configure and build
 ./configure
-bazel build --config=opt \
+JVM_RAM_RESOURCES=$((@LOCAL_RAM_RESOURCES@ / 4))
+JVM_RAM_RESOURCES=$((${JVM_RAM_RESOURCES} < 1024 ? 1024 : ${JVM_RAM_RESOURCES}))
+bazel --host_jvm_args="-Xmx${JVM_RAM_RESOURCES}m" \
+      build --config=opt \
             --config=monolithic \
+            --local_ram_resources=@LOCAL_RAM_RESOURCES@ \
+            --local_cpu_resources=@LOCAL_CPU_RESOURCES@ \
+            --discard_analysis_cache \
+            $cxx_std_opts \
             $cuda_config_opts \
             tensorflow:libtensorflow_cc.so \
             tensorflow:install_headers