%cst_0 = "tf.Const"() {value = dense<[-2.000000e+00, 3.000000e+00]> : tensor<2xf32>} : () -> tensor<2xf32>
    %0 = "quantfork.qcast"(%cst) : (tensor<2x2x3x2xf32>) -> tensor<2x2x3x2x!quant.uniform<i8<-127:127>:f32:3, {4.000000e-03,5.000000e-03}>>

func.func @fake_quant_conv(%arg0: tensor<1x3x4x3xf32>, %arg1: tensor<2x3x3x2xf32>) -> tensor<*xf32> {
  %cst = "tf.Const"() {value = dense<0.000000e+00> : tensor<2xf32>} : () -> tensor<2xf32>
  %0 = "tf.FakeQuantWithMinMaxArgs"(%arg1) {device = "", max = 2.000000e+00 : f32, min = -1.000000e+00 : f32, narrow_range = false, num_bits = 8 : i64} : (tensor<2x3x3x2xf32>) -> tensor<*xf32>

  %2 = "tfl.add"(%0, %1) { fused_activation_function = "NONE" } : (tensor<1x2x3x4xf32>, tensor<5x2x3x4xf32>) -> tensor<5x2x3x4xf32>
  func.return %2 : tensor<5x2x3x4xf32>
}

// CHECK: %cst = arith.constant dense<[3, 0, 1, 2]> : tensor<4xi32>
// CHECK: %0 = tfl.add(%arg0, %arg1) <{fused_activation_function = "NONE"}> : (tensor<2x3x4x1xf32>, tensor<2x3x4x5xf32>) -> tensor<2x3x4x5xf32>

  %output, %min, %max, %histogram = "tf.CustomAggregator"(%arg0) <{calibration_method = 5 : i32, id = "skipping_id", num_bins = 32 : i32, max_percentile = 0.000000e+00 : f32, min_percentile = 0.000000e+00 : f32}> : (tensor<1x3x4x3xf32>)...

  %3 = "tfl.depthwise_conv_2d"(%arg0, %2, %0) {depth_multiplier = 4 : i32, dilation_h_factor = 2 : i32, dilation_w_factor = 2 : i32, fused_activation_function = "NONE", padding = "VALID", stride_h = 4 : i32, stride_w = 5 : i32} : (tensor<1x224x224x3xf32>, tensor<32x3x3x3xf32>, tensor<32xf32>) -> tensor<1x112x112x32xf32>

  func.return %3 : tensor<1x112x112x32xf32>

  %output, %min, %max, %histogram = "tf.CustomAggregator"(%arg0) <{calibration_method = 5 : i32, id = "0", num_bins = 32 : i32, max_percentile = 0.000000e+00 : f32, min_percentile = 0.000000e+00 : f32}> : (tensor<1x3x4x3xf32>)...

  %cst2 = arith.constant dense<[1.0, 2.0]> : tensor<2xf32>
  %0 = "tf.Conv2D"(%arg0, %cst0) {T = "tfdtype$DT_FLOAT", data_format = "NHWC", dilations = [1, 2, 3, 1], padding = "SAME", strides = [1, 4, 5, 1]} : (tensor<1x112x112x3xf32>, tensor<1x3x3x2xf32>) -> tensor<1x28x23x2xf32>

  %w = arith.constant dense<-1.0> : tensor<32x3x3x3xf32>
  %3 = "tfl.quantize"(%w) {qtype = tensor<32x3x3x3x!quant.uniform<u8<1:255>:f32, 0.1>>} : (tensor<32x3x3x3xf32>) -> tensor<32x3x3x3x!quant.uniform<u8<1:255>:f32, 0.1>>
  %4 = "tfl.dequantize"(%3) : (tensor<32x3x3x3x!quant.uniform<u8<1:255>:f32, 0.1>>) -> tensor<32x3x3x3xf32>

            index_vector_dim = 2>,
          slice_sizes = array<i64: 1, 2, 2>,
          indices_are_sorted = false
        } : (tensor<3x4x2xf32>, tensor<2x3x2xi64>) -> tensor<2x3x2x2xf32>
        func.return %2 : tensor<2x3x2x2xf32>
      }
    }
  )mlir";
  OwningOpRef<ModuleOp> module_op = ParseModuleOpString(kModuleGather);
  ASSERT_TRUE(module_op);

  func::FuncOp main_fn = FindMainFuncOp(*module_op);

// CHECK: %[[cast:.*]] = "tf.Cast"(%[[cst]]) <{Truncate = false}> : (tensor<2x3x3x2xbf16>) -> tensor<2x3x3x2xf32>
// CHECK: %[[conv:.*]] = "tf.Conv2D"(%[[ARG0]], %[[cast]])
// CHECK: %[[identity:.*]] = "tf.IdentityN"(%[[conv]]) {device = ""} : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xf32>
// CHECK: return %[[identity]] : tensor<1x3x2x2xf32>