libcudnn8=8.9.6.50-1+cuda12.2
libcudnn8-dev=8.9.6.50-1+cuda12.2

# This can be removed once NVIDIA publishes a cuda-12.3.2 Docker image.
# For now it ensures that we install at least version 12.3.107 of PTXAS,
# since 12.3.103 has a bug.

    --experimental_cc_shared_library \
    --@local_config_cuda//:enable_cuda \
    --define framework_shared_object=false \
    "somepath(//tensorflow/tools/pip_package:build_pip_package, " \
    "@local_config_cuda//cuda:cudart + "\
    "@local_config_cuda//cuda:cudart + "\
    "@local_config_cuda//cuda:cuda_driver + "\
    "@local_config_cuda//cuda:cudnn + "\
    "@local_config_cuda//cuda:curand + "\
    "@local_config_cuda//cuda:cusolver + "\

# Install devtoolset devel dependencies
COPY setup.sources.sh /setup.sources.sh
COPY setup.packages.sh /setup.packages.sh
COPY devel.packages.txt /devel.packages.txt
COPY cuda.packages.txt /cuda.packages.txt
RUN /setup.sources.sh && /setup.packages.sh /devel.packages.txt

# Install various tools.
# - bats: bash unit testing framework
#         NOTE: v1.6.0 seems to have a bug that made "git" in setup_file break

TFCI_BAZEL_TARGET_SELECTING_CONFIG_PREFIX=linux_arm64
# Note: this is not set to "--cpu", because that changes the package name
# to tensorflow_cpu. These ARM builds are supposed to have the name "tensorflow"
# despite lacking Nvidia CUDA support.
TFCI_BUILD_PIP_PACKAGE_ARGS="--repo_env=WHEEL_NAME=tensorflow"
TFCI_DOCKER_ENABLE=1
TFCI_DOCKER_IMAGE=gcr.io/tensorflow-sigs/build-arm64:tf-2-16-multi-python
TFCI_DOCKER_PULL_ENABLE=1

# Change the value of CACHEBUSTER when upgrading the toolchain, or when testing
# different compilation methods. E.g. for a PR to test a new CUDA version, set
# the CACHEBUSTER to the PR number.
build --action_env=CACHEBUSTER=20220325

# Use Python 3.X as installed in container image
build --action_env PYTHON_BIN_PATH="/usr/local/bin/python3"

[Docker container](https://www.tensorflow.org/install/docker), and
[build from source](https://www.tensorflow.org/install/source).

To install the current release, which includes support for
[CUDA-enabled GPU cards](https://www.tensorflow.org/install/gpu) *(Ubuntu and
Windows)*:

```
$ pip install tensorflow
```

Other devices (DirectX and MacOS-metal) are supported using

pkg debug/elf, const EM_CRX = 114
pkg debug/elf, const EM_CRX Machine
pkg debug/elf, const EM_CSR_KALIMBA = 219
pkg debug/elf, const EM_CSR_KALIMBA Machine
pkg debug/elf, const EM_CUDA = 190
pkg debug/elf, const EM_CUDA Machine
pkg debug/elf, const EM_CYPRESS_M8C = 161
pkg debug/elf, const EM_CYPRESS_M8C Machine
pkg debug/elf, const EM_D10V = 85
pkg debug/elf, const EM_D10V Machine
pkg debug/elf, const EM_D30V = 86

load(
    "//tensorflow:tensorflow.bzl",
    "check_deps",
    "if_google",
    "if_not_mobile",
    "tf_cc_test",
    "tf_copts",
    "tf_cuda_library",
    "tf_custom_op_library",
    "tf_kernel_library",
)
load("//tensorflow:tensorflow.default.bzl", "filegroup", "tf_cuda_cc_test")
load(
    "//tensorflow/core/tpu:build_defs.bzl",
    "if_libtpu_tf_status",
)

package(

#    build. This should also match the system you're using--you cannot build
#    the TF MacOS package from Linux.
#      Ex. linux_x86        -- x86_64 Linux platform
#      Ex. linux_x86_cuda   -- x86_64 Linux platform, with Nvidia CUDA support
#      Ex. macos_arm64      -- arm64 MacOS platform
# 3. Add modifiers. Some modifiers for local execution are:
#      Ex. disk_cache -- Use a local cache

    export flags="--config=opt -k"
    ```

    If the tests are to be run on the GPU, add CUDA paths to LD_LIBRARY_PATH and
    add the `cuda` option flag

    ```bash
    export LD_LIBRARY_PATH="${LD_LIBRARY_PATH}:/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64:$LD_LIBRARY_PATH"
    export flags="--config=opt --config=cuda -k"
    ```

    For example, to run all tests under tensorflow/python, do:

    ```bash