%w = arith.constant dense<[[[[-1.0, 1.0]]], [[[1.0, 2.0]]], [[[-2.0, 1.0]]]]> : tensor<3x1x1x2xf32>
  %b = arith.constant dense<[1.0e-2, 2.1473647e1, -2.1473647e2]> : tensor<3xf32>
  %transpose_conv = "tfl.transpose_conv"(%arg1, %w, %0, %b) {
    padding = "SAME", stride_h = 1 : i32, stride_w = 1 : i32, fused_activation_function = "NONE"
  } : (tensor<4xi32>, tensor<3x1x1x2xf32>, tensor<1x5x5x2xf32>, tensor<3xf32>) -> tensor<1x5x5x3xf32>

  return %0 : tensor<1x1x1x2xf32>
}
func.func private @XlaCallModule_aten.avg_pool2d.default.impl_5(%arg0: tensor<1x1x1x7xf32>) -> tensor<1x1x1x2xf32>

// CHECK-LABEL: avg_pool2d_6
// CHECK: %cst = arith.constant dense<[0, 2, 3, 1]> : tensor<4xi32>
// CHECK: %0 = "tfl.transpose"(%arg0, %cst) : (tensor<1x1x1x7xf32>, tensor<4xi32>) -> tensor<1x1x7x1xf32>

  %cst_0 = "tf.Const"() {value = dense<-1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %cst_1 = "tf.Const"() {value = dense<1.000000e+00> : tensor<f32>} : () -> tensor<f32>
  %0 = "tf.Conv2D"(%arg0, %cst) {data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 1, 1]} : (tensor<1x3x4x3xf32>, tensor<1x1x3x2xf32>) -> tensor<1x3x4x2xf32>

// CHECK:           %[[VAL_6:.*]] = "tfl.reshape"(%[[VAL_1]], %[[VAL_2]]) : (tensor<1xf32>, tensor<4xi32>) -> tensor<1x1x1x1xf32>
// CHECK:           %[[VAL_7:.*]] = "tfl.concatenation"(%[[VAL_5]], %[[VAL_6]]) <{axis = 3 : i32, fused_activation_function = "NONE"}> : (tensor<1x1x1x1xf32>, tensor<1x1x1x1xf32>) -> tensor<1x1x1x2xf32>

// CHECK:  [[VAL_1:%.*]] = "tfl.reshape"(%2, %[[CST]]) {tac.device = "GPU",  tac.inference_type = "FLOAT"} : (tensor<1xf32>, tensor<4xi32>) -> tensor<1x1x1x1xf32>
// CHECK:  [[VAL_2:%.*]] = "tfl.concatenation"([[VAL_0]], [[VAL_1]]) <{axis = 3 : i32, fused_activation_function = "NONE"}> {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<1x1x1x1xf32>, tensor<1x1x1x1xf32>) -> tensor<1x1x1x2xf32>

// CHECK-LABEL: conv2d_backprop_input_with_add
func.func @conv2d_backprop_input_with_add(%arg0: tensor<4xi32>, %arg1: tensor<3x3x1x32xf32>, %arg2: tensor<15x14x14x32xf32>) -> tensor<15x28x28x1xf32> {
  %0 = "tf.Conv2DBackpropInput"(%arg0, %arg1, %arg2) {strides = [1, 2, 2, 1], padding="SAME", dilations=[1, 1, 1, 1]}: (tensor<4xi32>, tensor<3x3x1x32xf32>, tensor<15x14x14x32xf32>) -> tensor<15x28x28x1xf32>

    %1 = stablehlo.constant dense<1.000000e+03> : tensor<1x1x1x1xf32>  // Input inverse scale.
    %2 = stablehlo.constant dense<-128> : tensor<1x1x1x1xi8>  // Input zero point.
    %3 = stablehlo.constant dense<1> : tensor<3x3x4x4xi8>  // Quantized filter tensor.
    %4 = stablehlo.constant dense<3.000000e+03> : tensor<1x1x1x4xf32>
    %5 = stablehlo.constant dense<4.000000e+03> : tensor<1x1x1x1xf32>  // Output inverse scale.

    %1 = "tfl.reshape"(%arg1, %cst) {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<1xf32>, tensor<4xi32>) -> tensor<1x1x1x1xf32>
    %2 = "tfl.concatenation"(%0, %1) {axis = 3 : i32, fused_activation_function = "NONE", tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<1x1x1x1xf32>, tensor<1x1x1x1xf32>) -> tensor<1x1x1x2xf32>

// CHECK: return %[[v0]]
}

func.func @batch_multilhs_einsum(%arg0: tensor<2x1x1x11xf32>, %arg1: tensor<2x11x2xf32>) -> tensor<2x1x1x2xf32> {
  %0 = "tf.Einsum"(%arg0, %arg1) {T = "tfdtype$DT_FLOAT", equation = "BiNj,BjS->BiNS"} : (tensor<2x1x1x11xf32>, tensor<2x11x2xf32>) -> tensor<2x1x1x2xf32>
  func.return %0 : tensor<2x1x1x2xf32>
// CHECK-LABEL: batch_multilhs_einsum
// CHECK-DAG: %[[cst:.*]] = arith.constant dense<[2, 1, 11]> : tensor<3xi64>

// CHECK:           %[[VAL_8:.*]] = "tfl.reshape"(%[[VAL_7]], %[[VAL_3]]) {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<1x1x1x2xf32>, tensor<1xi32>) -> tensor<2xf32>
// CHECK:           %[[VAL_9:.*]] = "tfl.reshape"(%[[VAL_8]], %[[VAL_4]]) {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<2xf32>, tensor<2xi32>) -> tensor<2x1xf32>