value {
      list {
      }
    }
  }
  attr {
    key: "padding"
    value {
      s: "VALID"
    }
  }
  attr {
    key: "strides"
    value {
      list {
        i: 1
        i: 1
        i: 2
        i: 2
      }
    }
  }
  attr {
    key: "use_cudnn_on_gpu"
    value {

pkgpath unicode;
init unicode unicode..import;
var ASCII_Hex_Digit <type 1 *<type 2 "RangeTable" <type 3 struct { R16 <type 4 [] <type 5 "Range16" <type 6 struct { Lo <type -6>; Hi <type -6>; Stride <type -6>; }>>>; R32 <type 7 [] <type 8 "Range32" <type 9 struct { Lo <type -7>; Hi <type -7>; Stride <type -7>; }>>>; LatinOffset <type -11>; }>>>;
var Adlam <type 1>;
var Ahom <type 1>;
var Anatolian_Hieroglyphs <type 1>;
var Arabic <type 1>;
var Armenian <type 1>;

    %4 = "tf.Sub"(%3, %arg5) : (tensor<960x960x3x512xi32>, tensor<512xi32>) -> tensor<960x960x3x512xi32>
    %5 = "tf.Conv2D"(%1, %4) {dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<1x2240x2240x3xi32>, tensor<960x960x3x512xi32>) -> tensor<1x2240x1120x512xi32>
    %6 = "tf.Mul"(%arg2, %arg4) : (tensor<f32>, tensor<512xf32>) -> tensor<512xf32>

              mlir::cast<IntegerAttr>(op.getRhsDilationAttr()[i]).getInt());

      const int64_t output_size = (lhs_size_dilated + stride - 1) / stride;
      const int64_t total_padding = std::max(
          (output_size - 1) * stride + rhs_size_dilated - lhs_size_dilated,
          static_cast<int64_t>(0));
      const int64_t padding_begin = total_padding / 2;

  %cst = arith.constant dense<4> : tensor<i32>
  %stride = "tf.Add"(%arg2, %cst) : (tensor<i32>, tensor<i32>) -> tensor<i32>
  // CHECK: [[Result:%.*]]:3 = "tf.While"(%arg0, %arg1, %{{.+}} <{body = @tf.WhileRegion_body, cond = @tf.WhileRegion_cond
  %0:2 = "tf.WhileRegion"(%arg0, %arg1) (
    {
      ^bb0(%carg0: tensor<*xf32>, %carg1: tensor<i32>):
      %ne = "tf.NotEqual"(%carg1, %stride) : (tensor<i32>, tensor<i32>) -> tensor<i1>

const post = `
	default:
		return false
	}
	return true
}
`

const sratioCase = `
	case image.YCbCrSubsampleRatio%s:
		for y, sy := y0, sp.Y; y != y1; y, sy = y+1, sy+1 {
			dpix := dst.Pix[y*dst.Stride:]
			yi := (sy-src.Rect.Min.Y)*src.YStride + (sp.X - src.Rect.Min.X)
			%s

				// This is an inline version of image/color/ycbcr.go's func YCbCrToRGB.
				yy1 := int32(src.Y[yi]) * 0x10101

In detail, with the default NHWC format,

    output[b, i, j, k] =
        sum_{di, dj, q} input[b, strides[1] * i + di, strides[2] * j + dj, q] *
                        filter[di, dj, q, k]

Must have `strides[0] = strides[3] = 1`.  For the most common case of the same
horizontal and vertices strides, `strides = [1, stride, stride, 1]`.
  }];

  let arguments = (ins

    # Uniform Quantized op takes only the first and the second values for
    # strides.
    strides_to_check = (
        (strides[1], strides[2])
        if target_opset == quant_opts_pb2.UNIFORM_QUANTIZED
        else strides
    )
    strides_attr = attr_value_pb2.AttrValue(
        list=attr_value_pb2.AttrValue.ListValue(i=strides_to_check)
    )

    %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<*xf32>, tensor<*xf32>) -> tensor<*xf32>

					if !inclusive {
						if v == minSignedValue(limit.Type) {
							return false // < minint is never satisfiable.
						}
						v--
					}
					if init.isGenericIntConst() {
						// Use stride to compute a better lower limit.
						if init.AuxInt > v {
							return false
						}
						v = addU(init.AuxInt, diff(v, init.AuxInt)/uint64(step)*uint64(step))
					}
					if addWillOverflow(v, step) {