ENH: Comments about CPU compat / opt instructions

Dockerfile

@@ -47,7 +47,7 @@ RUN wget -qO /opt/otbtf/bin/bazelisk https://github.com/bazelbuild/bazelisk/rele
  && ln -s /opt/otbtf/bin/bazelisk /opt/otbtf/bin/bazel
 
 ARG BZL_TARGETS="//tensorflow:libtensorflow_cc.so //tensorflow/tools/pip_package:build_pip_package"
-# "--config=opt" will enable 'march=native' (otherwise edit CC_OPT_FLAGS in build-env-tf.sh)
+# "--config=opt" will enable 'march=native' (otherwise read comments about CPU compatibilty and edit CC_OPT_FLAGS in build-env-tf.sh)
 ARG BZL_CONFIGS="--config=nogcp --config=noaws --config=nohdfs --config=opt"
 # "--compilation_mode opt" is already enabled by default (see tf repo .bazelrc and configure.py)
 ARG BZL_OPTIONS="--verbose_failures --remote_cache=http://localhost:9090"

tools/docker/build-env-tf.sh

 ### TF - bazel build env variables
 
-# As in official TF wheels, you'll need to remove "-march=native" for old CPUs compatibity (no AVX2)
+# As in official TF wheels, you'll need to remove "-march=native" to ensure portability (avoid AVX2 / AVX512 compatibility issues)
+# You could also add CPUs instructions one by one, in this example to avoid only AVX512 but enable commons optimizations like FMA, SSE4.2 and AVX2
+#export CC_OPT_FLAGS="-Wno-sign-compare --copt=-mavx --copt=-mavx2 --copt=-mfma --copt=-mfpmath=both --copt=-msse4.2"
 export CC_OPT_FLAGS="-march=native -Wno-sign-compare"
 export GCC_HOST_COMPILER_PATH=$(which gcc)
 export PYTHON_BIN_PATH=$(which python)