Rect:           r,
	}
}

func (p *YCbCr) Opaque() bool {
	return true
}

func yCbCrSize(r Rectangle, subsampleRatio YCbCrSubsampleRatio) (w, h, cw, ch int) {
	w, h = r.Dx(), r.Dy()
	switch subsampleRatio {
	case YCbCrSubsampleRatio422:
		cw = (r.Max.X+1)/2 - r.Min.X/2
		ch = h
	case YCbCrSubsampleRatio420:
		cw = (r.Max.X+1)/2 - r.Min.X/2
		ch = (r.Max.Y+1)/2 - r.Min.Y/2

000001c0  b1 f3 66 30 f8 aa 90 f7  90 49 19 48 4c 25 4a 1f  |..f0.....I.HL%J.|
000001d0  12 9d 67 32 79 bb 53 d8  c5 d1 b4 6e 89 75 49 c0  |..g2y.S....n.uI.|
000001e0  65 86 ac 72 23 2f 97 d3  ae e2 64 79 5e e2 10 4e  |e..r#/....dy^..N|
000001f0  55 0c c6 70 d3 2e 4a 6c  b0 73 0a 11 eb ae f7 a1  |U..p..Jl.s......|
00000200  a1 f0 5f 67 45 46 d3 8c  11 ff 21 62 7d ed f9 0e  |.._gEF....!b}...|

// `dx` must be nullptr or have size `ny`.
// If `dx` is nullptr, the implementation will use dx of `OnesLike` for all
// shapes in `y`.
// The partial derivatives are returned in `dy`. `dy` should be allocated to
// size `nx`.
//
// Gradient nodes are automatically named under the "gradients/" prefix. To
// guarantee name uniqueness, subsequent calls to the same graph will

  // clang-format off
  return GradForUnaryCwise(g, {
      {{"y"}, "Tanh", {"x"}},
      {{"y2"}, "Square", {"y"}, {}, {"dy"}},
      FunctionDefHelper::Const("one", 1.0f),
      {{"a"}, "Sub", {"one", "y2"}},
      {{"dx"}, "Mul", {"dy", "a"}},
  });
  // clang-format on
}
REGISTER_OP_GRADIENT("Supported", SupportedGrad);

// A gradient containing an unsupported operator.

<mxfile host="65bd71144e">
    <diagram id="BkDNbdtn8_9fWQybnc8v" name="Page-1">
        <mxGraphModel dx="741" dy="1167" grid="1" gridSize="10" guides="1" tooltips="1" connect="1" arrows="1" fold="1" page="1" pageScale="1" pageWidth="1920" pageHeight="1200" math="0" shadow="0">
            <root>
                <mxCell id="0"/>
                <mxCell id="1" parent="0"/>

//===--------------------------------------------------------------------===//

func.func @stridedSliceGrad(%dy: tensor<4x8xf32>, %begin: tensor<2xi64>, %end: tensor<2xi64>, %strides: tensor<2xi64>, %shape: tensor<2xi64>) -> tensor<?x?xf32> {
  // CHECK: tf.StridedSliceGrad
  %0 = "tf.StridedSliceGrad"(%shape, %begin, %end, %strides, %dy) : (tensor<2xi64>, tensor<2xi64>, tensor<2xi64>, tensor<2xi64>, tensor<4x8xf32>) -> tensor<?x?xf32>

  // CHECK: %[[RESULT:.*]] = "tf.Mul"(%[[DY]], %[[SUB]]) : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>
  %0 = "tf.TanhGrad"(%y, %dy) : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>

  // CHECK: return %[[RESULT]]
  func.return %0 : tensor<*xf32>
}

// CHECK-LABEL: func @tanhgrad_complex
// CHECK-SAME: (%[[Y:.*]]: tensor<*xcomplex<f32>>, %[[DY:.*]]: tensor<*xcomplex<f32>>)

000003f0  d6 e8 f3 57 17 03 03 00  35 9e 6f 39 92 2e 32 10  |...W....5.o9..2.|
00000400  03 cd f0 28 2c 1a 32 77  19 c8 39 38 60 0c 28 b7  |...(,.2w..98`.(.|
00000410  8c 3a d8 50 a1 44 e4 d6  c5 64 59 88 2d a4 23 c9  |.:.P.D...dY.-.#.|
00000420  26 d1 96 0c c9 5d da 84  3e 8a 7d fe 80 77        |&....]..>.}..w|
>>> Flow 3 (client to server)
00000000  14 03 03 00 01 01 17 03  03 00 19 87 49 99 a6 d9  |............I...|

000001e0  6b df 54 88 a8 f3 af 0d  4a 1b 29 91 3a bb 90 27  |k.T.....J.).:..'|
000001f0  29 3a 97 17 03 03 00 17  09 78 1d f3 95 2e 92 af  |):.......x......|
00000200  d8 af 64 59 fb 77 e3 85  de f0 37 be 11 91 2f 17  |..dY.w....7.../.|
00000210  03 03 00 13 4e a2 06 c3  90 6f d3 b0 04 a8 25 32  |....N....o....%2|

  op_tape_.erase(op_it);
}

// Terminology:
//
//  - op: a possibly composite operation, which has an entry in the tape
//  - target: dy in dx/dy
//  - source: dx in dx/dy
//  - tensor: one of the many inputs or outputs of an operation
//
// Below here we do the gradient algorithm. It works as follows:
//