// CHECK: %[[v1:.*]] = "tf.Transpose"(%arg1, %[[cst_1]]) : (tensor<2x4x7x3xf32>, tensor<4xi32>) -> tensor<2x7x3x4xf32>
  // CHECK: "tf.BatchMatMulV2"(%[[v0]], %[[v1]]) <{adj_x = false, adj_y = false}> : (tensor<2x7x5x3xf32>, tensor<2x7x3x4xf32>) -> tensor<2x7x5x4xf32>
}

func.func @einsum_matrixdotprod(%arg0: tensor<2x5x7x3xf32>, %arg1: tensor<7x3x4xf32>) -> tensor<2x5x4xf32> {

  %cst = "tf.Const"() {value = dense<2.000000e+00> : tensor<2x3x3x1xf32>} : () -> tensor<2x3x3x1xf32>
  %cst_0 = "tf.Const"() {value = dense<0.400000e+00> : tensor<3xf32>} : () -> tensor<3xf32>
  %cst_1 = "tf.Const"() {value = dense<0.500000e+00> : tensor<3xf32>} : () -> tensor<3xf32>
  %identity = "tf.Identity"(%cst) : (tensor<2x3x3x1xf32>) -> tensor<2x3x3x1xf32>

    %cst_1 = "tf.Const"() {value = dense<3.000000e+00> : tensor<2x3x3x1xf32>} : () -> tensor<2x3x3x1xf32>
    %cst_2 = "tf.Const"() {value = dense<3.000000e+00> : tensor<2x3x3x2xf32>} : () -> tensor<2x3x3x2xf32>
    %0 = "tf.PartitionedCall"(%arg0, %cst_1) {_tfl_quant_trait = "fully_quantizable", config = "", config_proto = "", executor_type = "", f = @composite_depthwise_conv2d_fn} : (tensor<1x3x4x3xf32>, tensor<2x3x3x1xf32>) -> tensor<*xf32>

    %0 = "tf.DepthwiseConv2dNative"(%arg0, %cst_0) {data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 2, 2, 1]} : (tensor<1x3x4x3xf32>, tensor<2x3x3x1xf32>) -> tensor<1x2x2x3xf32>

    %cst_1 = "tf.Const"() {value = dense<3.000000e+00> : tensor<2x3x3x1xf32>} : () -> tensor<2x3x3x1xf32>
    %cst_2 = "tf.Const"() {value = dense<3.000000e+00> : tensor<2x3x3x2xf32>} : () -> tensor<2x3x3x2xf32>
    %0 = "tf.PartitionedCall"(%arg0, %cst_1) {_tfl_quant_trait = "fully_quantizable", config = "", config_proto = "", executor_type = "", f = @composite_depthwise_conv2d_fn} : (tensor<1x3x4x3xf32>, tensor<2x3x3x1xf32>) -> tensor<*xf32>

  %cst_1 = "tf.Const"() {value = dense<3.000000e+00> : tensor<2x3x3x2xf32>} : () -> tensor<2x3x3x2xf32>
  %0 = "tf.Conv2D"(%arg0, %cst_1) {
    data_format = "NHWC", device = "", dilations = [1, 1, 1, 1], explicit_paddings = [],
    padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true
  } : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<*xf32>

  %cst = "tf.Const"() {value = dense<0.000000e+00> : tensor<2xf32>} : () -> tensor<2xf32>
  %0 = "tf.Conv2D"(%arg0, %arg1) {data_format = "NHWC", device = "", dilations = [1, 1, 2, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<*xf32>

func.func @add_with_activation_transpose(%arg0: tensor<1x3x3x4xf32>) -> tensor<1x4x3x3xf32> {
  %0 = stablehlo.constant dense<2.000000e+00> : tensor<1x4x3x3xf32>
  %1 = stablehlo.transpose %arg0, dims = [0, 3, 1, 2] : (tensor<1x3x3x4xf32>) -> tensor<1x4x3x3xf32>
  %2 = stablehlo.add %1, %0 : tensor<1x4x3x3xf32>
  return %2 : tensor<1x4x3x3xf32>
}
// CHECK-SAME: (%[[ARG_0:.+]]: tensor<1x3x3x4xf32>) -> tensor<1x4x3x3xf32>

  %perm = arith.constant dense<[3, 0, 1, 2]> : tensor<4xi32>
  %0 = "tfl.transpose"(%arg0, %perm) : (tensor<2x3x4x1xf32>, tensor<4xi32>) -> tensor<1x2x3x4xf32>
  %cst = arith.constant dense<1.0> : tensor<5x2x3x4xf32>
  %1 = "tfl.add"(%0, %cst) { fused_activation_function = "NONE" } : (tensor<1x2x3x4xf32>, tensor<5x2x3x4xf32>) -> tensor<5x2x3x4xf32>
  func.return %1 : tensor<5x2x3x4xf32>
}

    %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}>>