%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-NEXT:  %[[SELU_VAL:.*]] = "tf.Mul"(%[[ELU_VAL]], %[[SCALED_ALPHA]]) : (tensor<1x4x4x3xf32>, tensor<f32>) -> tensor<1x4x4x3xf32>
    // CHECK-NEXT:  %[[RES:.*]] = "tf.SelectV2"(%[[PRED]], %[[SCALED_FEATURES]], %[[SELU_VAL]]) : (tensor<1x4x4x3xi1>, tensor<1x4x4x3xf32>, tensor<1x4x4x3xf32>) -> tensor<1x4x4x3xf32>
    // CHECK-NEXT:  return %[[RES]] : tensor<1x4x4x3xf32>

func.func @NoPadStridedSliceNonNewAxisMask(%arg0: tensor<1x2x3x1xf32>) -> tensor<1x2x3x1xf32> {
  %cst = arith.constant dense<0> : tensor<4xi32>
  %cst_0 = arith.constant dense<1> : tensor<4xi32>
  %0 = "tf.StridedSlice"(%arg0, %cst, %cst, %cst_0) {begin_mask = 15 : i64, ellipsis_mask = 0 : i64, end_mask = 15 : i64, new_axis_mask = 0 : i64, shrink_axis_mask = 0 : i64} : (tensor<1x2x3x1xf32>, tensor<4xi32>, tensor<4xi32>, tensor<4xi32>) -> tensor<1x2x3x1xf32>

  %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>
// CHECK-DAG: %[[CONST_0:.+]] = stablehlo.constant

// CHECK:           %[[VAL_6:.*]] = "tfl.reshape"(%[[VAL_1]], %[[VAL_2]]) {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<1xf32>, tensor<4xi32>) -> tensor<1x1x1x1xf32>

  %result = "tf_device.launch"() ({
    %3 = "tf.Transpose"(%2, %1) : (tensor<1x8x4x4xf32>, tensor<4xi32>) -> tensor<1x4x4x8xf32>
    tf_device.return %3: tensor<1x4x4x8xf32>
  }) {device = "device"} : () -> tensor<1x4x4x8xf32>

  func.return %result : tensor<1x4x4x8xf32>

  // CHECK-DAG: %[[CONST1:.*]] = "tf.Const"() <{value = dense<[0, 3, 1, 2]> : tensor<4xi32>}>

    %3 = "tf.Identity"(%2) {device = ""} : (tensor<1x3x1x1xf32>) -> tensor<1x3x1x1xf32>
    return %3 : tensor<1x3x1x1xf32>
  }

// CHECK-LABEL: func @multiple_quantizable_ops_in_graph

// CustomOpNotWeightOnly-LABEL: QuantizeCustomOp
func.func @QuantizeCustomOp(%arg0: tensor<1x1x1x1xf32>) -> tensor<*xf32> attributes {tf.entry_function = {inputs = "input", outputs = "custom_op"}} {
  %0 = "quantfork.stats"(%arg0) {layerStats = dense<[0.000000e+00, 2.550000e+02]> : tensor<2xf32>} : (tensor<1x1x1x1xf32>) -> tensor<1x1x1x1xf32>
  %w = arith.constant dense<127.0> : tensor<1024x1x1x1xf32>

  %6 = "tf.Reshape"(%5, %2) {device = ""} : (tensor<2x16x1xf32>, tensor<4xi32>) -> tensor<2x4x4x1xf32>
  %7 = "tf.Identity"(%6) {device = ""} : (tensor<2x4x4x1xf32>) -> tensor<2x4x4x1xf32>
  func.return %7 : tensor<2x4x4x1xf32>
}

    rhs_quantization_min_val = -128 : i64,
    rhs_quantization_max_val = 127 : i64
  } : (tensor<1x6x6x3xf32>, tensor<2x3x3x2x!tf_type.qint8>, tensor<f32>, tensor<i32>) -> tensor<1x4x1x2xf32>
  func.return %0 : tensor<1x4x1x2xf32>
}

// -----

// CHECK-LABEL: func @uniform_quantized_convolution_hybrid_same
func.func @uniform_quantized_convolution_hybrid_same(%input: tensor<1x2x2x3xf32>) -> tensor<1x2x1x2xf32> {