func.return %1 : tensor<1x1x1x16xf32>

// CHECK: %[[avgp:.*]] = "tfl.average_pool_2d"(%arg0)
// CHECK: %[[dq:.*]] = "tfl.dequantize"(%[[avgp]]) : (tensor<1x1x1x16x!quant.uniform<u8:f32, 7.812500e-03:128>>) -> tensor<1x1x1x16xf32>
// CHECK: return %[[dq]] : tensor<1x1x1x16xf32>
}

// CHECK-LABEL: QuantizeReshape2D

// 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>

    %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>

// 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>

  func.return %1 : tensor<1x1x1x1x1xf32>

  // CHECK-DAG: %[[CST1:.*]] = "tfl.no_value"() <{value}> : () -> none
  // CHECK-DAG: %[[CST2:.*]] = arith.constant dense<2.000000e+00> : tensor<1x1x1x1x1xf32>

^bb0(%arg0: tensor<1x7x7x16xf32>):
  // expected-error @+1 {{requires attribute 'padding'}}
  %0 = "tf.AvgPool"(%arg0) {T = "tfdtype$DT_FLOAT", ksize = [1, 7, 7, 1], strides = [1, 1, 1, 1]} : (tensor<1x7x7x16xf32>) -> tensor<1x1x1x16xf32>
  func.return %0 : tensor<1x1x1x16xf32>
}

// -----

func.func @testAvgPoolWrongPadding(tensor<1x7x7x16xf32>) -> tensor<1x1x1x16xf32> {

  %w = arith.constant dense<[[[[0.0]]], [[[127.0]]], [[[-127.0]]]]> : tensor<3x1x1x1xf32>
  %b = arith.constant dense<0.0> : tensor<3xf32>
  %conv = "tfl.conv_2d"(%arg0, %w, %b) {dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "RELU", padding = "SAME", stride_h = 1 : i32, stride_w = 1 : i32} : (tensor<1x5x5x1xf32>, tensor<3x1x1x1xf32>, tensor<3xf32>) -> tensor<1x5x5x3xf32>
  func.return %conv : tensor<1x5x5x3xf32>

// CHECK: %1 = "tfl.pad"(%0, %cst_0) : (tensor<1x1x8x1xf32>, tensor<4x2xi32>) -> tensor<1x1x11x1xf32>
// CHECK: %2 = "tfl.average_pool_2d"(%1) <{filter_height = 1 : i32, filter_width = 3 : i32, fused_activation_function = "NONE", padding = "VALID", stride_h = 1 : i32, stride_w = 2 : i32}> : (tensor<1x1x11x1xf32>) -> tensor<1x1x5x1xf32>

    } : (tensor<1x6x6x16xf32>) -> tensor<1x1x1x16xf32>
  func.return %1 : tensor<1x1x1x16xf32>

// CHECK: %0 = "tfl.dequantize"(%arg0)
// CHECK: %1 = "tfl.average_pool_2d"(%0)
// CHECK: %2 = "tfl.quantize"(%1)
// CHECK: %3 = "tfl.dequantize"(%2)
// CHECK: return %3 : tensor<1x1x1x16xf32>
}

// CHECK-LABEL: QuantizeMaximum

  func.func @simple_folding(%arg0: tensor<1x1x1x1xi32>, %arg1: tensor<1x1x1x1xf32>) -> tensor<?x?x?x?xf32> {
    // CHECK: %[[SHAPE:.*]] = "tf.Shape"
    // CHECK: %[[CONV:.*]] = "tf.Conv2DBackpropInput"(%[[SHAPE]]
    // CHECK-SAME: (tensor<4xi32>, tensor<1x1x1x1xf32>, tensor<1x1x1x1xf32>) -> tensor<1x1x1x1xf32>
    // CHECK: return %[[CONV]] : tensor<1x1x1x1xf32>
    %0 = "tf.Shape"(%arg0) : (tensor<1x1x1x1xi32>) -> tensor<4xi32>