// CHECK-NEXT:   func.func @main(%arg0: tensor<8x8x1x207xf32>, %arg1: tensor<3x3x16x207xf32>) -> tensor<16x8x8x1xf32> {

  %0 = "tf.Conv2D"(%output, %cst) <{data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 2, 2, 1], use_cudnn_on_gpu = true}> {attr_map = "0:strides,1:use_cudnn_on_gpu,2:padding,3:explicit_paddings,4:dilations", device = ""} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x2x2x2xf32>

  %0 = "tf.Conv2D"(%arg0, %cst) {
    data_format = "NHWC", device = "", dilations = [1, 1, 1, 1], explicit_paddings = [],
    padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true
  } : (tensor<1x3x4x512xf32>, tensor<2x3x512x512xf32>) -> tensor<*xf32>
  func.return %0 : tensor<*xf32>

  %rst = "tf.Conv2D"(%arg, %fq) {T = "tfdtype$DT_FLOAT", data_format = "NHWC", dilations = [1, 2, 3, 1], padding = "SAME", strides = [1, 4, 5, 1]} : (tensor<256x32x32x3xf32>, tensor<3x3x3x16xf32>) -> tensor<256x8x7x16xf32>
  func.return %rst : tensor<256x8x7x16xf32>

// CHECK-DAG: %[[CONSTANT:.*]] = arith.constant dense<0.000000e+00> : tensor<16xf32>

    %padding = "tf.Const"() {value = dense<0> : tensor<3x2xi32>} : () -> tensor<3x2xi32>
    %strides = "tf.Const"() {value = dense<[3, 1, 1]> : tensor<3xi32>} : () -> tensor<3xi32>
    // expected-error @below {{'tf.XlaConvV2' op feature_group_count must be a positive number, got -1}}

    %0 = "tf.Conv2D"(%arg0, %arg1) {attr_map = "0:strides,1:use_cudnn_on_gpu,2:padding,3:explicit_paddings,4:dilations", data_format = "NHWC", device = "", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "VALID", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<1x2x2x3xf32>, tensor<2x2x3x2xf32>) -> tensor<*xf32>

  %rst = "tf.Conv2D"(%arg, %fq) {T = "tfdtype$DT_FLOAT", data_format = "NHWC", dilations = [1, 2, 3, 1], padding = "SAME", strides = [1, 4, 5, 1]} : (tensor<256x32x32x3xf32>, tensor<3x3x3x16xf32>) -> tensor<256x8x7x16xf32>
  func.return %rst : tensor<256x8x7x16xf32>

// CHECK-DAG: %[[CONSTANT:.*]] = arith.constant dense<0.000000e+00> : tensor<16xf32>

        offset = [0.5] * self.out_channel_size
        mean, variance = scale, offset
        out = nn_ops.conv2d(
            input_tensor,
            self.filters,
            strides=strides,
            dilations=dilations,
            padding=padding,
            data_format='NHWC',
            name='sample/conv',
        )
        if bias_fn is not None:

      }

      @Override
      public int compare(byte[] left, byte[] right) {
        int stride = 8;
        int minLength = Math.min(left.length, right.length);
        int strideLimit = minLength & ~(stride - 1);
        int i;

        /*
         * Compare 8 bytes at a time. Benchmarking on x86 shows a stride of 8 bytes is no slower

// For n < t.cutoff, the block is a simple lookup table. Otherwise, the block
// is a list of ranges with an accompanying value. Given a matching range r,
// the value for b is by r.value + (b - r.lo) * stride.
func (t *sparseBlocks) lookup(n uint32, b byte) uint16 {
	offset := t.offset[n]
	header := t.values[offset]
	lo := offset + 1
	hi := lo + uint16(header.lo)
	for lo < hi {
		m := lo + (hi-lo)/2
		r := t.values[m]