%q_weight = "quantfork.qcast"(%weight) : (tensor<2x3x3x2xf32>) -> tensor<2x3x3x2x!quant.uniform<i8:f32, 0.074855112561992565:-1>>
  %dq_weight = "quantfork.dcast"(%q_weight) : (tensor<2x3x3x2x!quant.uniform<i8:f32, 0.074855112561992565:-1>>) -> tensor<2x3x3x2xf32>
  %q_bias = "quantfork.qcast"(%bias) : (tensor<2xf32>) -> tensor<2x!quant.uniform<i32:f32, 0.044022349891595126>>

// CHECK-LABEL: func @transposeConv2D
func.func @transposeConv2D(%arg0: tensor<1x3x32x32xf32>, %arg1: tensor<1x1x3x8xf32>) -> tensor<1x8x32x32xf32> {

  // Convert input: NCHW -> NHWC
  %0 = "tf.Const"() {value = dense<[0, 2, 3, 1]> : tensor<4xi32>} : () -> tensor<4xi32>
  %1 = "tf.Transpose"(%arg0, %0) : (tensor<1x3x32x32xf32>, tensor<4xi32>) -> tensor<1x32x32x3xf32>

  // Compute in NHWC
  %2 = "tf.Conv2D"(%1, %arg1)
    {

// dilations, etc...). This test only verifies that changing convolution data
// layout will update all the attributes.

// CHECK-LABEL: func @transposeConv2D
func.func @transposeConv2D(%input: tensor<1x3x32x32xf32>, %filter: tensor<1x1x3x8xf32>) -> tensor<1x8x7x6xf32> {

  // CHECK: %[[ARG_PERM:.*]] = "tf.Const"() <{value = dense<[0, 2, 3, 1]> : tensor<4xi64>}>
  // CHECK: %[[ARG_TRANSPOSE:[0-9]*]] = "tf.Transpose"(%arg0, %[[ARG_PERM]])

  func.return %0 : tensor<?x32x32x16xf32>
}

func.func @testDepthwiseConv2D(tensor<1x112x112x3xf32>, tensor<1x3x3x32xf32>, tensor<32xf32>) -> tensor<1x112x112x32xf32> {
^bb0(%arg0: tensor<1x112x112x3xf32>, %arg1: tensor<1x3x3x32xf32>, %arg2: tensor<32xf32>):
  // CHECK: _arithmetic_count = 7626752 : i64

  func.return %0 : tensor<256x30x30x16xf32>
}

// -----

func.func @testDepthwiseConv(%arg0: tensor<1x112x112x32xf32>, %arg1: tensor<1x3x3x32xf32>, %arg2: tensor<32xf32>) -> tensor<1x112x112x32xf32> {
  // CHECK: tac.device = "GPU", tac.inference_type = "FLOAT"

    %8 = "tf.Cast"(%5) {Truncate = false} : (tensor<1x3x2x2xi32>) -> tensor<1x3x2x2xf32>
    %9 = "tf.Mul"(%7, %8) : (tensor<f32>, tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xf32>
    %10 = "tf.Round"(%9) : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xf32>
    %11 = "tf.Cast"(%10) {Truncate = false} : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xi32>
    %12 = "tf.AddV2"(%11, %arg10) : (tensor<1x3x2x2xi32>, tensor<i32>) -> tensor<1x3x2x2xi32>

func.func @DontFuseAddWithConvActivationFunc(%arg0: tensor<1x3x1x1xf32>) -> tensor<1x2x1x3xf32> {
  %cst = arith.constant dense<1.5> : tensor<1xf32>
  %cst_1 = arith.constant dense<0.0> : tensor<3xf32>
  %cst_2 = arith.constant dense<1.1> : tensor<3x2x1x1xf32>
  %0 = "tfl.add"(%arg0, %cst) {fused_activation_function = "RELU6"} : (tensor<1x3x1x1xf32>, tensor<1xf32>) -> tensor<1x3x1x1xf32>

    rhs_quantization_min_val = -128 : i64,
    rhs_quantization_max_val = 127 : i64
  } : (tensor<1x2x2x3xf32>, tensor<2x3x3x2x!tf_type.qint8>, tensor<f32>, tensor<i32>) -> tensor<1x3x3x2xf32>
  func.return %0 : tensor<1x3x3x2xf32>
}

//===----------------------------------------------------------------------===//
// tf.UniformQuantize and tf.UniformDequantize legalization

  %3 = "tfl.pseudo_const"() {value = dense<1.000000e+00> : tensor<16x3x3x16xf32>} : () -> tensor<16x3x3x16xf32>
  %4 = "tfl.pseudo_const"() {value = dense<1.000000e+00> : tensor<16xf32>} : () -> tensor<16xf32>

  // expected-error @+1 {{requires strides attribute length to be 4}}
  %0 = "tf.Conv2D"(%arg0, %arg1) {padding = "SAME", strides = [1, 1]} : (tensor<256x32x32x3xf32>, tensor<3x3x3x16xf32>) -> tensor<256x30x30x16xf32>
  func.return %0 : tensor<256x30x30x16xf32>
}

// -----

func.func @testConv2D(%arg0: tensor<256x32x32x3xf32>, %arg1: tensor<3x3x3x16xf32>) -> tensor<256x30x30x16xf32> {