return mkcall("countrunes", n.Type(), init, typecheck.Conv(n.X.(*ir.ConvExpr).X, types.Types[types.TSTRING]))
	}
	if isByteCount(n) {
		conv := n.X.(*ir.ConvExpr)
		walkStmtList(conv.Init())
		init.Append(ir.TakeInit(conv)...)
		_, len := backingArrayPtrLen(cheapExpr(conv.X, init))
		return len
	}
	if isChanLenCap(n) {
		name := "chanlen"
		if n.Op() == ir.OCAP {

  %0 = "tf.Conv2D"(%arg0, %arg1) {
    data_format = "NHWC", device = "", dilations = [1, 1, 1, 1], explicit_paddings = [],
    padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true
  } : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<*xf32>
  %1 = "tf.Relu6"(%0) {device = ""} : (tensor<*xf32>) -> tensor<*xf32>


  %3 = "tf.Conv2D"(%arg0, %arg1) {

// CHECK: %[[CONV:.+]] = stablehlo.convolution(%[[TRANSPOSE_0]], %[[CONST]]) dim_numbers = [b, 0, 1, f]x[0, 1, i, o]->[b, 0, 1, f], window = {pad = {{\[\[}}1, 1], [1, 1]]} {batch_group_count = 1 : i64, feature_group_count = 1 : i64} : (tensor<1x4x4x8xf32>, tensor<3x3x8x8xf32>) -> tensor<1x4x4x8xf32>
// CHECK: %[[TRANSPOSE_1:.+]] = stablehlo.transpose %[[CONV]], dims = [0, 3, 1, 2] : (tensor<1x4x4x8xf32>) -> tensor<1x8x4x4xf32>

                return true;
            }
            if (moduleConfiguration.equals("*")) {
                boolean include = true;
                for (String conf2 : dependencyConfigurations) {
                    if (conf2.startsWith("!") && conf2.substring(1).equals(configName)) {
                        include = false;
                        break;
                    }
                }

// CHECK: }

// CHECK-LABEL: private @composite_conv_with_bias_dynamic_fn_1
// CHECK: %[[CONV:.*]] = stablehlo.convolution(%arg0, %arg1)
// CHECK: %[[SHAPE_OF:.*]] = shape.shape_of %[[CONV]]
// CHECK: %[[DYNAMIC_BROADCAST_IN_DIM:.*]] = stablehlo.dynamic_broadcast_in_dim %arg2, %[[SHAPE_OF]]
// CHECK: %[[ADD:.*]] = stablehlo.add %[[CONV]], %[[DYNAMIC_BROADCAST_IN_DIM]]
// CHECK: return %[[ADD]] : tensor<?x28x28x16xf32>
// CHECK: }

                      weight_new_shape, &rewriter);

  // Replace the fc with conv.
  // The output would be [1, 1, width, output].
  auto conv_output_type = RankedTensorType::get({1, 1, width, output_size},
                                                output_type.getElementType());
  auto conv = rewriter.create<TFL::Conv2DOp>(
      fc_op.getLoc(), conv_output_type, reshaped_input, reshaped_weight,

  %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>

  // attribute which is shared.
  bool AreFuseCompatible(Conv2DOp conv, BiasAddOp bias_add,
                         PatternRewriter &rewriter) const override {
    // Verify that the data formats match and are valid for fusion.
    if (conv.getDataFormat() != bias_add.getDataFormat()) {
      (void)rewriter.notifyMatchFailure(conv, [&](Diagnostic &diag) {

    key: "narrow_range"
    value {
      b: true
    }
  }
  attr {
    key: "num_bits"
    value {
      i: 8
    }
  }
}
node {
  name: "BoxPredictor_4/ClassPredictor/Conv2D"
  op: "Conv2D"
  input: "input"
  input: "BoxPredictor_4/ClassPredictor/weights_quant/FakeQuantWithMinMaxVarsPerChannel"
  attr {
    key: "T"
    value {
      type: DT_FLOAT
    }
  }
  attr {

  %b = arith.constant dense<-1.23697901> : tensor<64xf32>
  %conv = "tfl.conv_2d"(%0, %w, %b) {dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "NONE", padding = "SAME", stride_h = 2 : i32, stride_w = 2 : i32} : (tensor<1x224x224x3xf32>, tensor<64x3x3x3xf32>, tensor<64xf32>) -> tensor<1x112x112x64xf32>