{[]int{0}, false},
			{[]int{0, -1}, false},
			{[]int{0, 0}, false},
			{[]int{0, 0, -1}, false},
			{[]int{0, -1, 0}, false},
			{[]int{0, -1, 0, -1}, false},
			{[]int{0, 1}, true},
			{[]int{0, 1, -1}, true},
			{[]int{0, -1, 1}, true},
			{[]int{0, -1, 1, -1}, true},
			{[]int{1}, false},
			{[]int{1, -1}, false},
			{[]int{2}, true},
			{[]int{2, -1}, true},
		},
	}, {

7578,1,5.56,19.5,0,0.5,0,0,,,20,2,1,166,48,green,0.00,3.9,0.0,31,16,8.50,1312,209.01,1,Manhattan,020901,1020901,E,MN09,Morningside Heights,3802,1163,121,1,Manhattan,012100,1012100,E,MN15,Clinton,3807^3389508,2,2014-03-31 14:28:58,2014-03-31 14:44:48,N,1,-73.956329345703125,40.813201904296875,-73.981666564941406,40.779743194580078,1,3.03,13.5,0,0.5,2.7,0,,,16.7,1,1,166,143,green,0.16,0.0,0.0,56,36,8.28,1312,209.01,1,Manhattan,020901,1020901,E,MN09,Morningside Heights,3802,1946,159,1,Manhattan,015...

// CHECK: %[[QCONST_1:.+]] = "tfl.pseudo_qconst"() <{qtype = tensor<2x!quant.uniform<i32:f32:0, {1.500000e+01,1.500000e+01}>>, value = dense<0> : tensor<2xi32>}> : () -> tensor<2x!quant.uniform<i32:f32:0, {1.500000e+01,1.500000e+01}>>

		return S64 | 0<<30 | 0<<29 | 0x11<<24

	case ACMN, AADDS:
		return S64 | 0<<30 | 1<<29 | 0x11<<24

	case AMOVW, AADDW:
		return S32 | 0<<30 | 0<<29 | 0x11<<24

	case ACMNW, AADDSW:
		return S32 | 0<<30 | 1<<29 | 0x11<<24

	case ASUB:
		return S64 | 1<<30 | 0<<29 | 0x11<<24

	case ACMP, ASUBS:
		return S64 | 1<<30 | 1<<29 | 0x11<<24

	case ASUBW:

  func.return %1#0, %1#1: tensor<3x4x2xf32>, tensor<3x4x2xi32>
}

// -----

// CHECK-LABEL: topk_d
func.func @topk_d(%arg0: tensor<?x8xf32>) -> (tensor<?x2xf32>, tensor<?x2xi32>) {
  %0 = arith.constant dense<2> : tensor<i32>
  %1:2 = "tfl.topk_v2"(%arg0, %0) : (tensor<?x8xf32>, tensor<i32>) -> (tensor<?x2xf32>, tensor<?x2xi32>)
  func.return %1#0, %1#1: tensor<?x2xf32>, tensor<?x2xi32>
}

  auto shape = elements.getType().getShape();
  if (!shape.empty()) {
    // Checks that elements are essentially 1d.
    assert(elements.getNumElements() == shape.back());
  }
  std::vector<int64_t> shape_data = {1, 1, 1, 1};
  const int vector_length = elements.getNumElements();
  if (is_depthwise)
    shape_data[3] = vector_length;
  else
    shape_data[0] = vector_length;
  auto new_shape =

        builder_, opcode_index, builder_.CreateVector(operands),
        builder_.CreateVector(results), tflite::BuiltinOptions_NONE,
        /*builtin_options=*/0, 0, tflite::CustomOptionsFormat_FLEXBUFFERS, 0, 0,
        1, 1);
  }
  if (IsModelBiggerThan2GB(custom_option_vector.size())) {
    require_use_buffer_offset_ = true;
    return tflite::CreateOperator(
        builder_, opcode_index, builder_.CreateVector(operands),

// and "x <= y". Because if either or both of the operands are
// NaNs, all three of (x < y), (x == y) and (x > y) are false,
// and ARM Manual says FCMP instruction sets PSTATE.<N,Z,C,V>
// of this case to (0, 0, 1, 1).
(Less32F x y) => (LessThanF (FCMPS x y))
(Less64F x y) => (LessThanF (FCMPD x y))

// For an unsigned integer x, the following rules are useful when combining branch
// 0 <  x  =>  x != 0
// x <= 0  =>  x == 0

  func.return %0 : tensor<1x1x1x16xf32>
}

// -----

// CHECK-LABEL: func @testValidConv2D
func.func @testValidConv2D(%arg0: tensor<256x32x32x3xf32>, %arg1: tensor<3x3x3x16xf32>) -> tensor<256x32x32x16xf32> {

</p>

<pre>
[...]Point{{1.5, -3.5}, {0, 0}}     // same as [...]Point{Point{1.5, -3.5}, Point{0, 0}}
[][]int{{1, 2, 3}, {4, 5}}          // same as [][]int{[]int{1, 2, 3}, []int{4, 5}}
[][]Point{{{0, 1}, {1, 2}}}         // same as [][]Point{[]Point{Point{0, 1}, Point{1, 2}}}
map[string]Point{"orig": {0, 0}}    // same as map[string]Point{"orig": Point{0, 0}}