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

  // CHECK: %[[CONV2D:[0-9]*]] = "tf.Conv2D"(%[[ARG_TRANSPOSE]], %arg1)
  // CHECK-SAME: data_format = "NHWC"
  // CHECK-SAME: dilations = [1, 3, 4, 2]
  // CHECK-SAME: explicit_paddings = [1, 2, 5, 6, 7, 8, 3, 4]
  // CHECK-SAME: padding = "EXPLICIT"

// CHECK: %[[cast:.*]] = "tf.Cast"(%[[cst]]) <{Truncate = false}> : (tensor<2x3x3x2xbf16>) -> tensor<2x3x3x2xf32>
// CHECK: %[[conv:.*]] = "tf.Conv2D"(%arg0, %[[cast]])
// CHECK: %[[identity:.*]] = "tf.IdentityN"(%[[conv]]) {device = ""} : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xf32>
// CHECK: return %[[identity]] : tensor<1x3x2x2xf32>

// -----

func.func @conv_with_non_constant_filter(%arg0: tensor<1x3x4x3xf32>, %arg1: tensor<2x3x3x2xf32>) -> tensor<*xf32> {
  %cst = "tf.Const"() {value = dense<0.000000e+00> : tensor<2xf32>} : () -> tensor<2xf32>
  %0 = "tf.Conv2D"(%arg0, %arg1) {data_format = "NHWC", 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>

// CHECK: return %0 : tensor<*xf32>
}

// -----

module {
  func.func @conv2d(%arg0: tensor<1x3x4x512xf32>) -> (tensor<*xf32>) {
    %cst_0 = "tf.Const"() {value = dense<0.000000e+00> : tensor<2xf32>} : () -> tensor<2xf32>

      %outputs_2, %control_3 = tf_executor.island(%control_1) wraps "tf.Const"() {value = dense<0.000000e+00> : tensor<5x5x32x16xf32>} : () -> tensor<5x5x32x16xf32>
      %outputs_4, %control_5 = tf_executor.island wraps "tf.Conv2D"(%outputs, %outputs_2) {data_format = "NHWC", device = "", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "VALID", strides = [1, 1, 1, 1], use_cudnn_on_gpu = true} : (tensor<*xf32>, tensor<5x5x32x16xf32>) -> tensor<*xf32>

// CHECK: return %0 : tensor<*xf32>
}

// -----

module {
  func.func @conv2d(%arg0: tensor<1x3x4x3xf32>) -> (tensor<*xf32>) {
    %cst_0 = "tf.Const"() {value = dense<0.000000e+00> : tensor<2xf32>} : () -> tensor<2xf32>

  // CHECK-SAME: [1.300000e+01, 2.800000e+01], [1.500000e+01, 3.200000e+01], [1.700000e+01, 3.600000e+01]
  // CHECK: %[[CONV:.*]] = "tf.Conv2D"(%arg0, %[[CST]]) <{data_format = "NHWC", dilations = [1, 2, 3, 1], explicit_paddings = [], padding = "SAME", strides = [1, 4, 5, 1], use_cudnn_on_gpu = true}>
  // CHECK: return %[[CONV]] : tensor<1x28x23x2xf32>
}

// CHECK-LABEL: @notfuseMulIntoConv2d
// filter and multiply are not broadcastable

  using OpRewritePattern<TFL::SplitVOp>::OpRewritePattern;

  LogicalResult matchAndRewrite(TFL::SplitVOp splitv_op,
                                PatternRewriter& rewriter) const override;
};

// Ensure bias for conv2d op.
struct EnsureBiasForConv2d : public OpRewritePattern<TFL::Conv2DOp> {
  using OpRewritePattern<TFL::Conv2DOp>::OpRewritePattern;

  LogicalResult matchAndRewrite(TFL::Conv2DOp conv_op,

// CHECK: %[[DEQUANTIZE:.*]] = "quantfork.dcast"(%[[QUANTIZE]])
// CHECK: %[[CONV:.*]] = "tf.Conv2D"(%arg0, %[[DEQUANTIZE]])
// CHECK: return %[[CONV]]
}

// CHECK-LABEL: perChannelFakeQuantWithConv2D
func.func @perChannelFakeQuantWithConv2D(tensor<256x32x32x3xf32>) -> (tensor<256x8x7x16xf32>) {
^bb0(%arg: tensor<256x32x32x3xf32>) :

// CHECK: %[[cast:.*]] = "tf.Cast"(%[[cst]]) <{Truncate = false}> : (tensor<2x3x3x2xbf16>) -> tensor<2x3x3x2xf32>
// CHECK: %[[conv:.*]] = "tf.Conv2D"(%[[ARG0]], %[[cast]])
// CHECK: %[[identity:.*]] = "tf.IdentityN"(%[[conv]]) {device = ""} : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xf32>
// CHECK: return %[[identity]] : tensor<1x3x2x2xf32>

// -----

// Tests that `tf.BatchFunction` is inlined.