Add 2
 BiasAdd 1
 Cast 1
 ConcatV2 1
 Const 7
 GreaterEqual 2
 MatMul 1
 Mul 5
 Select 2
 Sigmoid 3
 Snapshot 1
 Split 1
 Tanh 2
cluster 22 size 28
 Add 3
 BiasAdd 1
 Cast 1
 ConcatV2 1
 Const 5
 GreaterEqual 1
 MatMul 1
 Mul 5
 Select 3
 Sigmoid 3
 Snapshot 1
 Split 1
 Tanh 2
cluster 23 size 423
 Add 12
 AddN 28
 BiasAddGrad 6

//RUN: tf_tfl_translate --enable-stablehlo-conversion --input-mlir %s -o /tmp/temp.stablehlo; [ -f /tmp/temp.stablehlo ]


module {
func.func @main(%arg0: tensor<2xf32>) -> tensor<2xf32> {
  %0 = "tf.Sigmoid"(%arg0) : (tensor<2xf32>) -> tensor<2xf32>
  func.return %0 : tensor<2xf32>
}

}

TEST_F(CWiseUnaryGradTest, Sigmoid) {
  auto x_fn = [this](const int i) { return RV({0, -1, 1, -2, 2, -3, 3}); };
  TestCWiseGrad<float, float>(SIGMOID, x_fn);
}

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

TEST_F(CWiseUnaryGradTest, Sign) {

  // inconsistent with Conv 1x1 which always performs per channel quantization.
  bool disable_per_channel_for_dense_layers = false;

  // Whether to use fixed output ranges of the activation ops (tanh, sigmoid,
  // etc.) and not infer weight constants.
  // If this option is set, quantization emulation ops should be placed after
  // the ops in the input graph. This flag should be set to false for
  // post-training quantization.

            (CastValueToI64 $old, $shape),
            MHLO_RngDistributionValue<"NORMAL">),
          [(IsShapedTensor $shape)]>;

//===----------------------------------------------------------------------===//
// Sigmoid grad op.
//===----------------------------------------------------------------------===//

// TODO(hinsu): Handle unranked inputs by broadcasting constant one to the
// shape of $l instead of having it as a constant.

// =============================================================================
// Training OPs
// =============================================================================

// `grad = dy * y * (1 - y)`, where `y = sigmoid(x)`
def LegalizeSigmoidGrad : Pat<(TF_SigmoidGradOp $y, $dy),
  (TFL_MulOp $dy, (TFL_MulOp $y, (TFL_SubOp
        (Arith_ConstantOp ConstantAttr<RankedF32ElementsAttr<[]>, "1.0f">),
        $y, TFL_AF_None),

// be checked first if present.
// TODO: b/323478683: Consider deprecating this.
struct OpQuantScaleSpec {
  // Whether this op has a fixed range requirement (e.g. sigmoid)
  bool has_fixed_output_range = false;
  // Whether this op should have same operand and result scales (e.g. concat)
  bool has_same_scale_requirement = false;

  auto y = ConjugateHelper(grad_scope, op.output(0));
  grad_outputs->push_back(internal::SigmoidGrad(grad_scope, y, grad));
  return grad_scope.status();
}
REGISTER_GRADIENT_OP("Sigmoid", SigmoidGrad);

Status SignGrad(const Scope& scope, const Operation& op,
                const std::vector<Output>& grad_inputs,
                std::vector<Output>* grad_outputs) {

// CHECK-LABEL: argmin
// CHECK:  "tfl.arg_min"(%arg0, %arg1) : (tensor<3xi32>, tensor<i32>) -> tensor<i32>
}

func.func @sigmoid(%arg0: tensor<?x88xf32>) -> tensor<?x88xf32> {
  %0 = "tf.Sigmoid"(%arg0) : (tensor<?x88xf32>) -> tensor<?x88xf32>
  func.return %0 : tensor<?x88xf32>
// CHECK-LABEL: sigmoid
// CHECK:  "tfl.logistic"(%arg0) : (tensor<?x88xf32>) -> tensor<?x88xf32>
}

        $filter_width, $filter_height, ActFnAttr),
    [(HasOneUse $pool_out)]>;
}

// TODO(hinsu): Also fuse ops corresponding to SIGN_BIT fused
// activation functions.
// Currently we're not fusing tanh, sigmoid, hard_swish and other activations
// those cannot be simply translated into clamping.
foreach actFnPair = [[TFL_ReluOp, TFL_AF_Relu],
                     [TFL_Relu6Op, TFL_AF_Relu6],