return errors::Unimplemented(
        "BroadcastMul is not supported in tracing mode yet.");
  }
  auto imm_ctx = dyn_cast<ImmediateExecutionContext>(ctx);
  AbstractTensorPtr minus_1(imm_ctx->CreateInt32Scalar(-1));
  ImmediateTensorHandlePtr dim(imm_ctx->CreateLocalHandle(minus_1.get()));
  AbstractTensorHandle* expand_dims_outputs;
  TF_RETURN_IF_ERROR(
      ops::ExpandDims(ctx, vec, dim.get(), &expand_dims_outputs, "ExpandDims"));

        } else {
          block->state = FetchState::ERROR;
        }
        block->cond_var.SignalAll();
        return;
      case FetchState::FETCHING:
        block->cond_var.WaitWithTimeout(&block->mu, absl::Minutes(1));
        if (block->state == FetchState::FINISHED) {
          return TF_SetStatus(status, TF_OK, "");
        }
        // Re-loop in case of errors.
        break;
      case FetchState::FINISHED:

    name = "Neg_Div_Grad_Y";
    TF_RETURN_IF_ERROR(Neg(ctx, upstream_grad, &temp_output,
                           name.c_str()));  // -U
    AbstractTensorHandlePtr MinusU(temp_output);

    name = "Mul_Div_Grad_Y";
    TF_RETURN_IF_ERROR(Mul(ctx, MinusU.get(), Z, &temp_output,
                           name.c_str()));  // -U*Z
    AbstractTensorHandlePtr UZ(temp_output);

    name = "Div_Grad_Y";