%fq = "tf.FakeQuantWithMinMaxVarsPerChannel"(%in, %mini, %maxi) {num_bits = 4, narrow_range = true} : (tensor<3x3x3x4xf32>, tensor<4xf32>, tensor<4xf32>) -> tensor<3x3x3x4xf32>
  %rst = "tf.Conv2D"(%arg, %fq) {T = "tfdtype$DT_FLOAT", data_format = "NHWC", dilations = [1, 2, 3, 1], padding = "SAME", strides = [1, 4, 5, 1]} : (tensor<256x32x32x3xf32>, tensor<3x3x3x4xf32>) -> tensor<256x8x7x4xf32>
  func.return %rst : tensor<256x8x7x4xf32>

  return %2 : tensor<?x2x2x1xi8>
}

// -----

// CHECK-LABEL: func @convolution_add_add_f32
func.func @convolution_add_add_f32(
    %lhs: tensor<?x3x2x1xf32>, %rhs: tensor<2x1x1x1xf32>,
    %zp_offset: tensor<?x2x2x1xf32>, %bias: tensor<1xf32>
  ) -> tensor<?x2x2x1xf32> {
  // CHECK-DAG: %[[conv:.*]] = mhlo.convolution

    %8 = "tf.Cast"(%5) {Truncate = false} : (tensor<1x3x2x2xi32>) -> tensor<1x3x2x2xf32>
    %9 = "tf.Mul"(%7, %8) : (tensor<f32>, tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xf32>
    %10 = "tf.Round"(%9) : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xf32>
    %11 = "tf.Cast"(%10) {Truncate = false} : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xi32>
    %12 = "tf.AddV2"(%11, %arg10) : (tensor<1x3x2x2xi32>, tensor<i32>) -> tensor<1x3x2x2xi32>

  // CHECK: return %[[RESULT]] : tensor<2x3x4x6xf32>
}

// -----

func.func @batchMatMulTwoDimAdjXY(%arg0: tensor<2x3x5x4xf32>, %arg1: tensor<2x3x6x5xf32>) -> tensor<2x3x4x6xf32> {
  %0 = "tf.BatchMatMul"(%arg0, %arg1) {adj_x = true, adj_y = true} : (tensor<2x3x5x4xf32>, tensor<2x3x6x5xf32>) -> tensor<2x3x4x6xf32>
  func.return %0 : tensor<2x3x4x6xf32>

  // CHECK-LABEL: batchMatMulTwoDimAdjXY

    indices_are_sorted = false
  } : (tensor<3x4x2xf32>, tensor<2x3x2xi64>) -> tensor<2x3x2x2xf32>
    %8 = "quantfork.qcast"(%7) {volatile} : (tensor<2x3x2x2xf32>) -> tensor<2x3x2x2x!quant.uniform<i8:f32, 0.13170163023705575:-1>>
    %9 = "quantfork.dcast"(%8) : (tensor<2x3x2x2x!quant.uniform<i8:f32, 0.13170163023705575:-1>>) -> tensor<2x3x2x2xf32>
    return %9 : tensor<2x3x2x2xf32>
  }

  // CHECK: quantized_dot_general_fn_1

^bb0(%arg0: tensor<256x32x32x3xf32>, %arg1: tensor<3x3x3x4xf32>) :
  %0 = "tf.DepthwiseConv2dNative"(%arg0, %arg1) {device = "", name = "MobilenetV2/expanded_conv/depthwise/depthwise", T = "tfdtype$DT_FLOAT", data_format = "NHWC", dilations = [1, 1, 1, 1], padding = "SAME", strides = [1, 1, 1, 1]} : (tensor<256x32x32x3xf32>, tensor<3x3x3x4xf32>) -> tensor<256x30x30x12xf32>

  %w = arith.constant dense<2.0> : tensor<3x3x3x3xf32>
  %q = "tfl.quantize"(%w) {qtype = tensor<3x3x3x3x!quant.uniform<i8<-127:127>:f32:0,{1.0,2.0,3.0}>>} : (tensor<3x3x3x3xf32>) -> tensor<3x3x3x3x!quant.uniform<i8<-127:127>:f32:0,{1.0,2.0,3.0}>>
  %dq = "tfl.dequantize"(%q) : (tensor<3x3x3x3x!quant.uniform<i8<-127:127>:f32:0,{1.0,2.0,3.0}>>) -> tensor<3x3x3x3xf32>

  // CHECK-LABEL: func @shape_from_const_input
  func.func @shape_from_const_input(%arg0: tensor<3x3x32x64xf32>, %arg1: tensor<200x24x24x64xf32>) -> tensor<?x?x?x?xf32> {
    %0 = "tf.Const"() {value = dense<[200, 26, 26, 32]> : tensor<4xi32>} : () -> tensor<4xi32>
    // CHECK: tf.Conv2DBackpropInput
    // CHECK-SAME: (tensor<4xi32>, tensor<3x3x32x64xf32>, tensor<200x24x24x64xf32>) -> tensor<200x26x26x32xf32>

// CHECK:           return %[[VAL_12]] : tensor<3x5x1x4xf32>
// CHECK:         }
func.func @convert_dot_general(%arg0: tensor<3x2x6x5x1xf32>, %arg1: tensor<3x2x4x6xf32>) -> tensor<3x5x1x4xf32> {
  %0 = "mhlo.dot_general"(%arg0, %arg1) {
    dot_dimension_numbers = #mhlo.dot<
      lhs_batching_dimensions = [0],