//...

      return %1 : tensor<1x3xf32>
    }

    func.func private @composite_dot_general_fn(%arg0: tensor<1x2xf32>, %arg1: tensor<2x3xf32>) -> tensor<1x3xf32> {
      %0 = stablehlo.dot_general %arg0, %arg1, contracting_dims = [1] x [0] : (tensor<1x2xf32>, tensor<2x3xf32>) -> tensor<1x3xf32>
      return %0 : tensor<1x3xf32>
    }
  )mlir";

    has_token_input_output = false, module = "", platforms = [],
    version = 5 : i64
  } : (tensor<1x2xf32>, tensor<2x3xf32>) -> tensor<1x3xf32>
  return %0 : tensor<1x3xf32>
}

// CHECK-LABEL: func.func @qdq_for_dot_general_weight_empty
// CHECK-SAME: (%[[ARG_0:.+]]: tensor<1x2xf32>) -> tensor<1x3xf32>
// CHECK: %[[CST:.+]] = "tf.Const"() <{value = dense<3.000000e-01> : tensor<2x3xf32>}> : () -> tensor<2x3xf32>

    %9 = stablehlo.convert %2 : (tensor<2x3xi8>) -> tensor<2x3xf32>
    %10 = stablehlo.dot_general %8, %9, contracting_dims = [1] x [0] : (tensor<1x2xf32>, tensor<2x3xf32>) -> tensor<1x3xf32>
    %11 = stablehlo.convert %3 : (tensor<1x3xi32>) -> tensor<1x3xf32>
    %12 = stablehlo.subtract %10, %11 : tensor<1x3xf32>  // q1 * q2 - z1 * q2
    %13 = stablehlo.multiply %12, %4 : tensor<1x3xf32>  // s1 * s2

    return %2 : tensor<1x2xf32>
  }

  func.func private @composite_add_fn(%arg0: tensor<1x2xf32>, %arg1: tensor<1x2xf32>) -> tensor<1x2xf32> attributes {_from_xla_call_module} {
    %0 = stablehlo.add %arg0, %arg1 : tensor<1x2xf32>
    %1 = stablehlo.add %0, %arg1 : tensor<1x2xf32>
    return %1 : tensor<1x2xf32>
  }
}

// -----

// =============================================================================

func.func @sigmoidGrad(%arg0: tensor<?x32xf32>, %arg1: tensor<?x32xf32>) -> tensor<?x32xf32> {
  %0 = "tf.SigmoidGrad"(%arg0, %arg1) : (tensor<?x32xf32>, tensor<?x32xf32>) -> tensor<?x32xf32>
  func.return %0 : tensor<?x32xf32>
// CHECK-LABEL: sigmoidGrad
// CHECK-NEXT: [[ONE:%.+]] = arith.constant dense<1.000000e+00> : tensor<f32>

    "vhlo.return_v1"(%lhs) : (tensor<f32>) -> ()
  }): (tensor<200x100x300xf32>, tensor<10x2xi32>, tensor<10x300xf32>) ->
      tensor<200x100x300xf32>
  func.return %0 : tensor<200x100x300xf32>
}

// CHECK-LABEL: func.func private @scatter(%arg0: tensor<200x100x300xf32>, %arg1: tensor<10x2xi32>, %arg2: tensor<10x300xf32>) -> tensor<200x100x300xf32> {

func.func @main(tensor<40x37xf32>, tensor<40x37xf32>) -> tensor<40x40xf32> {
^bb0(%arg0: tensor<40x37xf32>, %arg1: tensor<40x37xf32>):
  %cst = arith.constant dense<1.0> : tensor<40xf32>
  %0:2 = "tfl.fully_connected"(%arg0, %arg1, %cst) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<40x37xf32>, tensor<40x37xf32>, tensor<40xf32>) -> (tensor<40x40xf32>, tensor<40x40xf32>)

  %0 = arith.constant dense<-1> : tensor<1xi32>
  %1 = "tfl.reduce_max"(%arg0, %0) {keep_dims = true} : (tensor<10x10xf32>, tensor<1xi32>) -> tensor<10x1xf32>
  %2 = tfl.sub(%arg0, %1) {fused_activation_function = "NONE"} : (tensor<10x10xf32>, tensor<10x1xf32>) -> tensor<10x10xf32>
  %3 = "tfl.exp"(%2) : (tensor<10x10xf32>) -> tensor<10x10xf32>

// CHECK-LABEL: testFoldEnsureShapeOp
func.func @testFoldEnsureShapeOp(%arg0: tensor<10x20xf32>) -> (tensor<10x20xf32>, tensor<10x20xf32>, tensor<20x10xf32>) {
  %0 = "tf.EnsureShape"(%arg0) {shape = #tf_type.shape<10x20>} : (tensor<10x20xf32>) -> tensor<10x20xf32>
  %1 = "tf.EnsureShape"(%arg0) {shape = #tf_type.shape<?x20>} : (tensor<10x20xf32>) -> tensor<10x20xf32>
  // Failing case which should not be folded.