build:mkl --define=tensorflow_mkldnn_contraction_kernel=0
build:mkl --define=build_with_openmp=true
build:mkl -c opt

# config to build OneDNN backend with a user specified threadpool.
build:mkl_threadpool --define=build_with_mkl=true --define=enable_mkl=true
build:mkl_threadpool --define=tensorflow_mkldnn_contraction_kernel=0
build:mkl_threadpool --define=build_with_mkl_opensource=true

    * This allows adding extra code after a dependency is done. It can be used, for example, to close database connections.
    * Dependencies with `yield` can be normal or `async`, **FastAPI** will run normal dependencies in a threadpool.
    * They can be combined with normal dependencies.
    * It's possible to have arbitrary trees/levels of dependencies with `yield` and exit steps are handled in the correct order automatically.

    "@zlib//:__subpackages__",
    "@dlpack//:__subpackages__",
    "@arm_neon_2_x86_sse//:__subpackages__",
    "@cpuinfo//:__subpackages__",
    "@ruy//:__subpackages__",
    "@XNNPACK//:__subpackages__",
    "@pthreadpool//:__subpackages__",
    "@FXdiv//:__subpackages__",
    "@FP16//:__subpackages__",
    "@clog//:__subpackages__",
    "@flatbuffers//:__subpackages__",
    "@nccl_archive//:__subpackages__",

*   CTC loss and decoders in `contrib/ctc`
*   Basic support for `half` data type
*   Better support for loading user ops (see examples in `contrib/`)
*   Allow use of (non-blocking) Eigen threadpool with
    `TENSORFLOW_USE_EIGEN_THREADPOOL` define
*   Add an extension mechanism for adding network file system support
*   TensorBoard displays metadata stats (running time, memory usage and device