How to use bazel to build tensorflow as external dependencies to support gpu inference? - tensorflow

Question

I want to use tensorflow's c++ api to inference in my project, when I build it for cpu, it's ok. But when I build it for support gpu (bazel build --config=cuda ...), it has some error.
My WORKSPACE:
http_archive(
name = "org_tensorflow",
strip_prefix = "tensorflow-1.15.0",
sha256 = "a5d49c00a175a61da7431a9b289747d62339be9cf37600330ad63b611f7f5dc9",
urls = ["https://github.com/tensorflow/tensorflow/archive/v1.15.0.tar.gz"]
)
load("#org_tensorflow//tensorflow:workspace.bzl", "tf_workspace")
tf_workspace(tf_repo_name = "org_tensorflow")
# Uncomment the following two lines has the same error.
# load("#org_tensorflow//third_party/gpus:cuda_configure.bzl", "cuda_configure")
# cuda_configure(name="local_config_cuda")
My .bazelrc:
...
# config for using cuda
build:cuda --define=using_cuda=true --define=using_cuda_nvcc=true
build:cuda --action_env TF_NEED_CUDA=1
build:cuda --crosstool_top=#local_config_cuda//crosstool:toolchain
...
My BUILD:
cc_library(
name = "tf_utils",
srcs = ["tf_utils.cpp"],
hdrs = ["tf_utils.h"],
deps = ["#org_tensorflow//tensorflow/cc/saved_model:loader"
"#org_tensorflow//tensorflow/core:tensorflow"]
)
cc_binarary(
name = "test",
srcs = ["test.cpp"],
deps = ["tf_utils"]
)
When I run bazel build --config=cuda //:test
The error message is：
ERROR: external/com_google_protobuf/BUILD:405:1: C++ compilation of rule '#com_google_protobuf//:protoc' failed (Exit 1) crosstool_wrapper_driver_is_not_gcc failed: error executing command external/local_config_cuda/crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc -MD -MF bazel-out/host/bin/external/com_google_protobuf/_objs/protoc/main.d ... (remaining 46 argument(s) skipped)
Use --sandbox_debug to see verbose messages from the sandbox
src/main/tools/linux-sandbox-pid1.cc:413: "execvp(external/local_config_cuda/crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc, 0x6dfe10)": No such file or directory
Target //:test failed to build
Use --verbose_failures to see the command lines of failed build steps.
The error message shows no such file or directory, but it does exist.
# ls -la external/local_config_cuda/crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc
-rwxrwxr-x 1 root root 9129 Dec 1 14:14 external/local_config_cuda/crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc*
The tensorflow version is 1.15, bazel version is 3.1.0.
Does anyone know how to configure bazel correctly so that tensorflow can support gpu as an external dependency? Thanks very much.