Value filter_size, int64_t dilation_rate,
                        int64_t stride, tensorflow::Padding padding_type,
                        Type shape_scalar_type, Value *padding_low,
                        Value *padding_high) const {
    // Stride must be > 0
    if (stride <= 0) return false;
    // Dilation rate must be >= 1
    if (dilation_rate < 1) return false;

  // Begin:        1,   4,   -3
  // End:          8,  65,   42
  // Stride:       1,   4,   -1
  // Begin mask:   0,   0,    1  (= 1)
  // End mask:     1,   0,    0  (= 4)

  // So result shape:
  // Dim #0: begin mask (1) -> begin = 0; end 8 canonicalized to 4: so 4
  // Dim #1: 4 to 65 stride 4: so 16
  // Dim #2: begin -3 + 1024 = 1021; end mask (1) -> end = -1: so 1022

	//     rlen = j - i
	//     rptr = ptr + (mask(rcap) & (i * stride))
	//
	// Where mask(x) is 0 if x==0 and -1 if x>0 and stride is the width
	// of the element type.
	stride := s.constInt(types.Types[types.TINT], ptr.Type.Elem().Size())

	// The delta is the number of bytes to offset ptr by.
	delta := s.newValue2(mulOp, types.Types[types.TINT], i, stride)

	// If we're slicing to the point where the capacity is zero,

// CHECK:           %[[VAL_2:.*]] = mhlo.convolution(%[[VAL_0]], %[[VAL_1]]) dim_numbers = [b, 0, f]x[0, i, o]->[b, 0, f], window = {stride = [1], pad = {{\[\[}}0, 0]], lhs_dilate = [1], rhs_dilate = [1]} {batch_group_count = 1 : i64, feature_group_count = 1 : i64, precision_config = [#mhlo<precision DEFAULT>, #mhlo<precision DEFAULT>]} : (tensor<16x?x256xbf16>, tensor<1x256x256xbf16>)...