}

// Op: Log1p()
// Summary: Computes natural logarithm of (1 + x) element-wise.
//
// Description:
//   I.e., \\(y = \log_e (1 + x)\\).
//
//   Example:
//
//   ```python
//   x = tf.constant([0, 0.5, 1, 5])
//   tf.math.log1p(x) ==> [0., 0.4054651, 0.6931472, 1.7917595]
//   ```
Status Log1p(AbstractContext* ctx, AbstractTensorHandle* const x,

}

TEST_F(CWiseUnaryGradTest, Log1p) {
  auto x_fn = [this](const int i) { return RV({0, 1e-6, 1, 2, 3, 4, 100}); };
  TestCWiseGrad<float, float>(LOG1P, x_fn);
}

TEST_F(CWiseUnaryGradTest, Log1p_Complex) {
  auto x_fn = [this](const int i) {
    return CRV({{0, 0}, {1e-6, 0}, {2, -1}, {1, 2}, {3, 4}});
  };
  TestCWiseGrad<complex64, complex64>(LOG1P, x_fn);
}

TEST_F(CWiseUnaryGradTest, Sinh) {

}

Status Log1pModel(AbstractContext* ctx,
                  absl::Span<AbstractTensorHandle* const> inputs,
                  absl::Span<AbstractTensorHandle*> outputs) {
  return ops::Log1p(ctx, inputs[0], &outputs[0], "Log1p");
}

Status DivNoNanModel(AbstractContext* ctx,
                     absl::Span<AbstractTensorHandle* const> inputs,
                     absl::Span<AbstractTensorHandle*> outputs) {

def LogOfSoftmax : Pat<
  (TF_LogOp:$src (TF_SoftmaxOp $arg)), (TF_LogSoftmaxOp:$dest $arg), [],
  [(CopyAttrs $src, $dest)]>;

// Canonicalize: Log(1.0 + x) to Log1p(x)
//
// We currently do this rewrite only if the constant `1` is a scalar, because
// it is safely broadcastable to any shape. To be able to canonicalize when

                 absl::Span<AbstractTensorHandle* const> grad_outputs,
                 absl::Span<AbstractTensorHandle*> grad_inputs) override {
    // TODO(vnvo2409): Add control dependency
    /* Given upstream grad U and a Log1p op: Y = log(1 + X), the gradients are:
     *
     *    dX = U / (1 + X)
     *
     */

    AbstractTensorHandle* upstream_grad = grad_outputs[0];
    AbstractTensorHandle* X = forward_inputs_[0];

     * 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();

        assertEquals(BigIntegerMath.log10(valueOf(x), mode), IntMath.log10(x, mode));
      }
    }
  }

  // Relies on the correctness of log10(int, FLOOR) and of pow(int, int).
  @GwtIncompatible // pow()
  public void testLog10Exact() {
    for (int x : POSITIVE_INTEGER_CANDIDATES) {
      int floor = IntMath.log10(x, FLOOR);
      boolean expectSuccess = IntMath.pow(10, floor) == x;

        assertEquals(BigIntegerMath.log10(valueOf(x), mode), LongMath.log10(x, mode));
      }
    }
  }

  // Relies on the correctness of log10(long, FLOOR) and of pow(long, int).
  @GwtIncompatible // TODO
  public void testLog10Exact() {
    for (long x : POSITIVE_LONG_CANDIDATES) {
      int floor = LongMath.log10(x, FLOOR);
      boolean expectedSuccess = LongMath.pow(10, floor) == x;

  // Relies on the correctness of log10(BigInteger, {HALF_UP,HALF_DOWN}).
  @GwtIncompatible // TODO
  public void testLog10HalfEven() {
    for (BigInteger x : POSITIVE_BIGINTEGER_CANDIDATES) {
      int halfEven = BigIntegerMath.log10(x, HALF_EVEN);
      // Now figure out what rounding mode we should behave like (it depends if FLOOR was
      // odd/even).
      boolean floorWasEven = (BigIntegerMath.log10(x, FLOOR) & 1) == 0;