// CHECK: %[[LOG1P:.*]] = "tf.Log1p"(%[[CONCAT]]) {device = "/job:localhost/replica:0/task:0/device:GPU:0"}
  // CHECK: return %[[LOG1P]]
  %0 = "tf.Log1p"(%arg0) : (tensor<?x1xf32>) -> tensor<?x1xf32>
  %1 = "tf.Log1p"(%arg1) : (tensor<?x1xf32>) -> tensor<?x1xf32>

  // y = log1p(x)
  // dy/dx = 1 / (1 + x)
  auto one = Cast(scope, Const(scope, 1.0), op.input(0).type());
  auto dydx = Reciprocal(scope, Add(scope, one, op.input(0)));
  // grad(x) = grad(y) * conj(dy/dx)
  grad_outputs->push_back(
      Mul(scope, grad_inputs[0], ConjugateHelper(scope, dydx)));
  return scope.status();
}
REGISTER_GRADIENT_OP("Log1p", Log1pGrad);

    }

    // To improve accuracy on platforms with less-precise log implementations,
    // compute log(lanczos_gamma_plus_one_half) at compile time and use log1p on
    // the device.
    // log(t) = log(kLanczosGamma + 0.5 + z)
    //        = log(kLanczosGamma + 0.5) + log1p(z / (kLanczosGamma + 0.5))
    Value t = rewriter.create<AddV2Op>(loc, lanczos_gamma_plus_one_half, z);
    Value z_div_lanczos_gamma_plus_one_half =

// Hoist coefficient-wise unary operation out of the Concat op:
//
//   %0 = "tf.Log1p"(%arg_0)
//   %1 = "tf.Log1p"(%arg_1)
//   ...
//   %n = "tf.Log1p"(%arg_n)
//   %m = "tf.ConcatV2"(%0, %1, ..., %n, %axis)
//
// Rewrite it to:
//
//   %0 = "tf.ConcatV2"(%arg_0, %arg_1, ..., %arg_n, %axis)
//   %1 = "tf.Log1p"(%0)
class HoistCwiseUnaryOutOfConcat : public OpRewritePattern<TF::ConcatV2Op> {
 public:

     * is close to 1 (why?), gives the following formula.
     */
    double fractionOfBitsSet = (double) bitCount / bitSize;
    return DoubleMath.roundToLong(
        -Math.log1p(-fractionOfBitsSet) * bitSize / numHashFunctions, RoundingMode.HALF_UP);
  }

  /** Returns the number of bits in the underlying bit array. */
  @VisibleForTesting
  long bitSize() {
    return bits.bitSize();

     * is close to 1 (why?), gives the following formula.
     */
    double fractionOfBitsSet = (double) bitCount / bitSize;
    return DoubleMath.roundToLong(
        -Math.log1p(-fractionOfBitsSet) * bitSize / numHashFunctions, RoundingMode.HALF_UP);
  }

  /** Returns the number of bits in the underlying bit array. */
  @VisibleForTesting
  long bitSize() {
    return bits.bitSize();

  func.return %0 : tensor<?xf32>
}

// CHECK-LABEL:   func @log1p(
// CHECK-SAME:                %[[VAL_0:.*]]: tensor<2xf32>) -> tensor<2xf32> {
// CHECK:           %[[VAL_1:.*]] = "tf.Log1p"(%[[VAL_0]]) : (tensor<2xf32>) -> tensor<2xf32>
// CHECK:           return %[[VAL_1]] : tensor<2xf32>
// CHECK:         }
func.func @log1p(%arg0: tensor<2xf32>) -> tensor<2xf32> {

  %0 = "tf.Log"(%arg0) : (tensor<?xf32>) -> tensor<?xf32>
  func.return %0 : tensor<?xf32>
}

// -----

// CHECK-LABEL: @log1p
func.func @log1p(%arg0: tensor<2xf32>) -> tensor<2xf32> {
  // CHECK:  mhlo.log_plus_one %arg0 : tensor<2xf32>
  %0 = "tf.Log1p"(%arg0) : (tensor<2xf32>) -> tensor<2xf32>
  func.return %0 : tensor<2xf32>
}

// -----

// CHECK-LABEL: func @log1p_dynamic

            "Atan", "Atanh", "Ceil", "Cos", "Cosh", "Sin", "Exp", "Expm1",
            "Floor", "IsFinite", "IsInf", "IsNan", "Inv", "Reciprocal", "Log",
            "Log1p", "Invert", "LogicalNot", "Ndtri", "Neg", "Rint", "Round",
            "Rsqrt", "Sigmoid", "Sign", "Sinh", "Softplus", "Softsign", "Sqrt",
            "Square", "Tan", "Tanh", "Real", "Imag", "Erf", "Erfc", "Erfinv",

  // CHECK:  %[[IS_ZERO:.*]] = "tf.Equal"(%[[X]], %[[ZERO]]) <{incompatible_shape_error = true}> : (tensor<*xf32>, tensor<f32>) -> tensor<*xi1>
  // CHECK:  %[[LOG:.*]] = "tf.Log1p"(%[[Y]]) : (tensor<*xf32>) -> tensor<*xf32>
  // CHECK:  %[[MUL:.*]] = "tf.Mul"(%[[X]], %[[LOG]]) : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>