/*padding*/(GetAvgPoolOpPadAttr $old_val),
                            /*stride_h*/(GetI32At<0> (GetAsVectorAttr<"stride"> $attrs)),
                            /*stride_w*/(GetI32At<1> (GetAsVectorAttr<"stride"> $attrs)),
                            /*fused_activation_function*/TFL_AF_None,
                            (returnType (GetNhwcReturnTypeFromNchw $old_val))), 

  // CHECK-DAG: %[[RESULT:.+]] = mhlo.convolution(%[[INPUT]], %[[NEW_FILTER]]) dim_numbers = [b, 0, 1, f]x[0, 1, i, o]->[b, 0, 1, f], window = {stride = [1, 1], pad = {{\[\[}}0, 0], [0, 0]], rhs_dilate = [1, 1]} {batch_group_count = 1 : i64, feature_group_count = 1 : i64} : (tensor<1x256x256x3xf32>, tensor<1x1x3x2xf32>) -> tensor<1x256x256x2xf32>

		}
	}
}

func TestDumper(t *testing.T) {
	var in [40]byte
	for i := range in {
		in[i] = byte(i + 30)
	}

	for stride := 1; stride < len(in); stride++ {
		var out bytes.Buffer
		dumper := Dumper(&out)
		done := 0
		for done < len(in) {
			todo := done + stride
			if todo > len(in) {
				todo = len(in)
			}
			dumper.Write(in[done:todo])
			done = todo
		}

		dumper.Close()

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

  int kh;  // Kernel height.
  int kw;  // Kernel width.

  int ph;  // Padding on "height" dimension (both sides).
  int pw;  // Padding on "width" dimension (both sides).

  int sh;  // Stride on "height" dimension.
  int sw;  // Stride on "width" dimension.

  bool ceil_mode;  // Rounding strategy (ceil or floor).
};

// Rounds the dimension based on the ceil mode.
int RoundDim(float dim, bool ceil_mode) {
  if (ceil_mode) {

// 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]

    %2 = stablehlo.convolution(%1, %0)
      dim_numbers = [b, 0, 1, f]x[0, 1, i, o]->[b, 0, 1, f],
      window = {
        stride = [1, 1], pad = [[0, 0], [1, 1]],
        lhs_dilate = [1, 1],
        rhs_dilate = [1, 1]
      }
      {
        batch_group_count = 1 : i64,
        feature_group_count = 1 : i64

  %5 = "tf.ReadVariableOp"(%arg3) : (tensor<!tf_type.resource<tensor<3920x10xf32>>>) -> tensor<3920x10xf32>
  %6 = mhlo.convolution(%arg0, %4) dim_numbers = [b, 0, 1, f]x[0, 1, i, o]->[b, 0, 1, f], window = {stride = [1, 1], pad = [[1, 1], [1, 1]], rhs_dilate = [1, 1]} {batch_group_count = 1 : i64, feature_group_count = 1 : i64} : (tensor<32x28x28x1xf32>, tensor<3x3x1x5xf32>) -> tensor<32x28x28x5xf32>