and deploy TensorFlow. This folder is typically used by continuous integration
(CI) tools to build and test TensorFlow whenever there is a change to the
code. This folder is broken into subfolders that represent the level of support

      // Destroying pruning_thread_ will block until Prune() receives the above
      // notification and returns.
      pruning_thread_.reset();
    }
  }

  /// Read `n` bytes from `filename` starting at `offset` into `buffer`. It
  /// returns total bytes read ( -1 in case of errors ). This method will set
  /// `status` to:
  ///
  /// 1) The error from the remote filesystem, if the read from the remote
  ///    filesystem failed.

# Debug config
build:dbg -c dbg
# Only include debug info for files under tensorflow/, excluding kernels, to
# reduce the size of the debug info in the binary. This is because if the debug
# sections in the ELF binary are too large, errors can occur. See
# https://github.com/tensorflow/tensorflow/issues/48919.
# Users can still include debug info for a specific kernel, e.g. with:

"""Merge all JUnit test.xml files in one directory into one.

Usage: squash_testlogs.py START_DIRECTORY OUTPUT_FILE

Example: squash_testlogs.py /tf/pkg/testlogs /tf/pkg/merged.xml

Recursively find all the JUnit test.xml files in one directory, and merge any
of them that contain failures into one file. The TensorFlow DevInfra team

    const TF_Graph* fn_body, const char* fn_name,
    unsigned char append_hash_to_fn_name, int num_opers,
    const TF_Operation* const* opers, int ninputs, const TF_Output* inputs,
    int noutputs, const TF_Output* outputs, const char* const* output_names,
    int ncontrol_outputs, const TF_Operation* const* control_outputs,
    const char* const* control_output_names, const TF_FunctionOptions* opts,

    *   Added 8-bit/16-bit/32-bit int/uint support for the built-in op `right_shift`
    *   Added reference implementation for 16-bit int unquantized `add`.
    *   Added reference implementation for 16-bit int and 32-bit unsigned int unquantized `mul`.
    *   `add_op` supports broadcasting up to 6 dimensions.
    *   Added 16-bit support for `top_k`.

*   `tf.function`

// into its components.
//
// The filled `device` struct and the allocated `device_info` struct may be
// passed to TFE_RegisterCustomDevice. The `device_name` arguments must match.
void AllocateParallelDevice(const char* device_name,
                            const char* const* underlying_devices,
                            int num_underlying_devices,

    return absl::OkStatus();
  }
  Status SetAttrShapeList(const char* attr_name, const int64_t** dims,
                          const int* num_dims, int num_values) override {
    std::vector<PartialTensorShape> shapes;
    shapes.reserve(num_values);
    for (int i = 0; i < num_values; ++i) {
      if (num_dims[i] < 0) {
        shapes.emplace_back();
      } else {
        shapes.emplace_back(ArraySlice<int64_t>(

      required: true
  - type: textarea
    id: logs
    attributes:
      label: Relevant log output
      description: Please copy and paste any relevant log output. This will be automatically formatted into code, so no need for backticks.

set -exo pipefail

function is_continuous_or_release() {
  [[ "$KOKORO_JOB_TYPE" == "CONTINUOUS_INTEGRATION" ]] || [[ "$KOKORO_JOB_TYPE" == "RELEASE" ]]
}

# Move into the directory of the script
cd "$(dirname "$0")"

if is_continuous_or_release || [[ -z "$KOKORO_BUILD_ID" ]]; then
  # A continuous job is the only one to publish to latest
  TAG="latest-multi-python"
else