// Verify unused if with functions without side-effects is removed.
// CHECK-LABEL: main
func.func @main(%arg0: tensor<3x15x14x3xf32>) -> tensor<3x15x14x8xf32>
    attributes {tf.entry_function = {inputs = "input", outputs = "Conv2D"}} {
  %cst = arith.constant dense<[0, 1, 2, 3]> : tensor<4xi32>
  %cst_0 = arith.constant dense<1.000000e+00> : tensor<f32>
  %cst_1 = arith.constant dense<0.000000e+00> : tensor<8xf32>

    }
  }
  attr {
    key: "_class"
    value {
      list {
        s: "loc:@MobilenetV1/Conv2d_0/weights"
      }
    }
  }
}
node {
  name: "MobilenetV1/MobilenetV1/Conv2d_0/Conv2D"
  op: "Conv2D"
  input: "input"
  input: "MobilenetV1/Conv2d_0/weights/read"
  attr {
    key: "T"
    value {
      type: DT_FLOAT
    }
  }
  attr {
    key: "data_format"
    value {

    func.return %add : tensor<*xf32>
  }

  func.func @composite_conv2d_with_relu6_fn(%arg0: tensor<*xf32>, %arg1: tensor<*xf32>) -> tensor<*xf32> attributes {tf_quant.composite_function} {

   // Use identity op to avoid the filter being constant-folded.
    %identity = "tf.Identity"(%filter) : (tensor<*xi8>) -> tensor<*xi8>
    %2 = "tf.Cast"(%identity) {Truncate = false} : (tensor<*xi8>) -> tensor<*xf32>
    %3 = "tf.Conv2D"(%input, %2) {
      padding = "VALID", strides = [1, 1, 1, 1],
      attr_map = "strides:0,use_cudnn_on_gpu:1,padding:2,explicit_paddings:3,dilations:4"
    } : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>

func.func @not_lift_float_conv(%arg0: tensor<1x3x4x512xf32>) -> (tensor<*xf32>) {
  %cst = "tf.Const"() {value = dense<3.000000e+00> : tensor<2x3x512x512xf32>} : () -> tensor<2x3x512x512xf32>
  %0 = "tf.Conv2D"(%arg0, %cst) {
    data_format = "NHWC", device = "", dilations = [1, 1, 1, 1], explicit_paddings = [],
    padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true

// CHECK: %[[CONV:.+]] = stablehlo.convolution(%[[ARG1]], %[[ARG2]])
// CHECK-SAME: (tensor<1x3x4x3xf32>, tensor<2x3x3x2x!quant.uniform<i8:f32, 6.000000e-03:-128>>) -> tensor<1x3x4x2xf32>

    // Only rank size four input will be only available by the tf.Conv2D
    // operator verification.
    if (!input_type || input_type.isDynamicDim(3)) {
      return failure();
    }
    // Check if the given op is based on grouped convolution.
    // Dim size zero will be verified by the tf.Conv2D operator verification.
    if (input_type.getDimSize(3) % filter_type.getDimSize(2) != 0) {

func countrunes(string) int

// Convert non-interface type to the data word of a (empty or nonempty) interface.
func convT(typ *byte, elem *any) unsafe.Pointer

// Same as convT, for types with no pointers in them.
func convTnoptr(typ *byte, elem *any) unsafe.Pointer

// Specialized versions of convT for specific types.
// These functions take concrete types in the runtime. But they may

    }
  }
  attr {
    key: "_class"
    value {
      list {
        s: "loc:@conv_net_2d/conv_2d_0/w"
      }
    }
  }
}
node {
  name: "conv_net_2d_1/conv_2d_0/convolution"
  op: "Conv2D"
  input: "input"
  input: "conv_net_2d/conv_2d_0/w/read"
  attr {
    key: "T"
    value {
      type: DT_FLOAT
    }
  }
  attr {
    key: "data_format"
    value {
      s: "NCHW"
    }

    %5 = "tf.MatMul"(%1, %3) {
      attr_map = "transpose_a:0,transpose_b:1"
    } : (tensor<*xi32>, tensor<*xi32>) -> tensor<*xi32>
    func.return %5 : tensor<*xi32>
  }

  // Conv2D with int32 accumulation
  func.func private @internal_conv2d_fn(
                         %input : tensor<*xi8>, %filter : tensor<*xi8>,
                         %input_scale : tensor<*xf32>, %input_zp : tensor<*xi32>,