# copybara:uncomment_begin(google-only)
            # "-Wl,--undefined-version",
            # copybara:uncomment_end(google-only)
        ],
    }),
    linkstatic = 1,
    per_os_targets = True,
    roots = [
                "//tensorflow/c/experimental/filesystem:filesystem_interface",
                "//tensorflow/c/experimental/stream_executor:stream_executor",

  // or consider other ways of packing/unpacking parallel tensors.
  if (operation_name == std::string("TPUReplicatedInput")) {
    // Special-cased operation for packing per-device tensors into one parallel
    // tensor.
    if (inputs.size() != parallel_device.num_underlying_devices()) {
      std::string message(absl::StrCat(
          "The parallel device ", parallel_device_name, " expected ",

* <SIMILAR TO ABOVE SECTION, BUT FOR OTHER IMPORTANT CHANGES / BUG FIXES>
* <IF A CHANGE CLOSES A GITHUB ISSUE, IT SHOULD BE DOCUMENTED HERE>
* <NOTES SHOULD BE GROUPED PER AREA>

* `tf.lite`
    * Quantization for `FullyConnected` layer is switched from per-tensor to
      per-channel scales for dynamic range quantization use case (`float32`
      inputs / outputs and `int8` weights). The change enables new quantization

# Users can still include debug info for a specific kernel, e.g. with:
#     --config=dbg --per_file_copt=+tensorflow/core/kernels/identity_op.*@-g
# Since this .bazelrc file is synced between the tensorflow/tensorflow repo and
# the openxla/xla repo, also include debug info for files under xla/.
build:dbg --per_file_copt=+.*,-tensorflow.*,-xla.*@-g0
build:dbg --per_file_copt=+tensorflow/core/kernels.*@-g0

} TF_Filesystem_Option_Type;

typedef struct TF_Filesystem_Option {
  char* name;                         // null terminated, owned
  char* description;                  // null terminated, owned
  int per_file;                       // bool actually, but bool is not a C type
  TF_Filesystem_Option_Value* value;  // owned
} TF_Filesystem_Option;

/// SECTION 2. Function tables for functionality provided by plugins

  // once) and produces the gradient of the target tensors with respect to the
  // source tensors. The output gradients are used if not empty and not
  // null. The result is populated with one tensor per target element.
  // When running backward functions, builds zeros-like tensors for
  // incoming grads which are nullptrs, unless `build_default_zeros_grads`
  // is set to false.
  Status ComputeGradient(

  const char task1_name[] = "/job:localhost/replica:0/task:1/device:CPU:0";
  const char task2_name[] = "/job:localhost/replica:0/task:2/device:CPU:0";

  // Create one variable per task.
  TFE_TensorHandle* h0 = TestVariable(ctx, 1.0, task1_name);
  TFE_TensorHandle* h1 = TestVariable(ctx, 2.0, task2_name);
  TFE_TensorHandle* h2 = TestVariable(ctx, 3.0, task0_name);

TF_CAPI_EXPORT extern void TF_AddInputList(TF_OperationDescription* desc,
                                           const TF_Output* inputs,
                                           int num_inputs);

// Call once per control input to `desc`.
TF_CAPI_EXPORT extern void TF_AddControlInput(TF_OperationDescription* desc,
                                              TF_Operation* input);