Merge branch 'master' into cxx_17

Author: FloopCZ

Dockerfiles/install-common.sh

@@ -24,6 +24,7 @@ make
 rm -rf /home/tensorflow_cc/.cache
 rm -rf /root/.cache
 make install
+ldconfig
 cd "$cwd"
 rm -rf tensorflow_cc/tensorflow_cc/build
 

README.md

@@ -83,13 +83,19 @@ mkdir build && cd build
 cmake ..
 make
 sudo make install
+sudo ldconfig
 ```
 
 **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

@@ -1,9 +1,10 @@
 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 ${version}
 )
-set(PROJECT_VERSION "${version}")
 
 option(ALLOW_CUDA "Try to find and use CUDA." ON)
 option(REQUIRE_CUDA "Make sure to find and use CUDA (implies ALLOW_CUDA)." OFF)
@@ -41,10 +42,10 @@ 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/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

@@ -61,9 +61,7 @@ if [ "$cuda_allowed" == true ] && [ "$cuda_available" == true ]; then
 
     # choose the right version of CUDA compiler
     if [ -z "$GCC_HOST_COMPILER_PATH" ]; then
-        if   hash gcc 2>/dev/null && version_gt 10.3 `gcc -dumpversion`; then
-            export GCC_HOST_COMPILER_PATH=${GCC_HOST_COMPILER_PATH:-"/usr/bin/gcc"}
-        elif hash gcc-10 2>/dev/null && version_gt 10.3 `gcc-10 -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"}

tensorflow_cc/patches/v2.5.0-Fix-NCCL-build-failure-with-CUDA-11.3.patch

@@ -0,0 +1,13 @@
+diff --git a/third_party/nccl/build_defs.bzl.tpl b/third_party/nccl/build_defs.bzl.tpl
+index ceb65e3b632f3..c875717403857 100644
+--- a/third_party/nccl/build_defs.bzl.tpl
++++ b/third_party/nccl/build_defs.bzl.tpl
+@@ -204,7 +204,7 @@ def _prune_relocatable_code_impl(ctx):
+             arguments = arguments,
+             mnemonic = "nvprune",
+         )
+-        output.append(outputs)
++        outputs.append(output)
+ 
+     return DefaultInfo(files = depset(outputs))
+