std::strncpy(buffer, "VALID", BUFFER_SIZE);
  TFE_OpSetAttrString(op, "padding", buffer, std::strlen(buffer));
  // Overwriting value in "buffer", should be fine since TFE_Op
  // shouldn't be holding on to it.
  std::strncpy(buffer, "NHWC", BUFFER_SIZE);
  TFE_OpSetAttrString(op, "data_format", buffer, std::strlen(buffer));

  TFE_OpSetAttrType(op, "T", TF_FLOAT);

// Returns the device of the operation that produced `h`. If `h` was produced by
// a copy, returns the destination device of the copy. Note that the returned
// device name is not always the device holding the tensor handle's memory. If
// you want the latter, use TFE_TensorHandleBackingDeviceName. This function
// will block till the operation that produces `h` has completed.

TF_CAPI_EXPORT void TF_SetXlaMinClusterSize(int size);

// Gets/Sets TF/XLA flag for whether(true) or not(false) to disable constant
// folding. This is for testing to ensure that XLA is being tested rather than
// Tensorflow's CPU implementation through constant folding.
TF_CAPI_EXPORT unsigned char TF_GetXlaConstantFoldingDisabled();
TF_CAPI_EXPORT void TF_SetXlaConstantFoldingDisabled(
    unsigned char should_enable);

unsigned char TF_GetXlaConstantFoldingDisabled() {
  return static_cast<unsigned char>(
      tensorflow::GetBuildXlaOpsPassFlags()->tf_xla_disable_constant_folding);
}

void TF_SetXlaConstantFoldingDisabled(unsigned char should_enable) {
  tensorflow::GetBuildXlaOpsPassFlags()->tf_xla_disable_constant_folding =
      static_cast<bool>(should_enable);
}

void TF_SetXlaMinClusterSize(int size) {
  tensorflow::MarkForCompilationPassFlags* flags =

  TF_TriState implementation_selector;
  TF_TriState function_optimization;
  TF_TriState common_subgraph_elimination;
  TF_TriState arithmetic_optimization;
  TF_TriState debug_stripper;
  TF_TriState constant_folding;
  TF_TriState shape_optimization;
  TF_TriState auto_mixed_precision;
  TF_TriState auto_mixed_precision_onednn_bfloat16;
  TF_TriState auto_mixed_precision_mkl;
  TF_TriState pin_to_host_optimization;

  CONFIG_TOGGLE(function_optimization);
  CONFIG_TOGGLE(common_subgraph_elimination);
  CONFIG_TOGGLE(arithmetic_optimization);
  CONFIG_TOGGLE(debug_stripper);
  CONFIG_TOGGLE(constant_folding);
  CONFIG_TOGGLE(shape_optimization);
  CONFIG_TOGGLE(auto_mixed_precision);
  CONFIG_TOGGLE(auto_mixed_precision_onednn_bfloat16);
  CONFIG_TOGGLE(auto_mixed_precision_mkl);
  CONFIG_TOGGLE(pin_to_host_optimization);

  /// was cached, a coordination lock, and state & condition variables.
  ///
  /// Thread safety:
  /// The iterator and timestamp fields should only be accessed while holding
  /// the block-cache-wide mu_ instance variable. The state variable should only
  /// be accessed while holding the Block's mu lock. The data vector should only
  /// be accessed after state == FINISHED, and it should never be modified.
  ///

        for all the RNG in Keras. We plan to switch on the new code path by
        default in tf 2.8, and the behavior change will likely to cause some
        breakage on user side (eg if the test is checking against some golden
        number). These 3 APIs will allow user to disable and switch back to
        legacy behavior if they prefer. In future (eg TF 2.10), we expect to