// CHECK-NEXT: [[loop_counter:%.*]] = "tf.AddV2"(%arg2, [[cst_1]])
// CHECK-NEXT: [[weight:%.*]] = "tf.ReadVariableOp"(%arg5)
// CHECK-NEXT: [[mpy:%.*]] = "tf.MatMul"(%arg6, [[weight]])
// CHECK-NEXT: [[element_index:%.*]] = "tf.AddV2"(%arg3, [[cst_1]])
// CHECK-NEXT: [[bias:%.*]] = "tf.GatherV2"(%arg7, %arg3, [[cst_0]])
// CHECK-NEXT: [[res:%.*]] = "tf.AddV2"([[mpy]], [[bias]])

	var x [128]uint32 // give x its own cache line
	sink = &x
	for i := 0; i < b.N; i++ {
		atomic.And32(&x[63], uint32(i))
	}
}

func BenchmarkAnd32Parallel(b *testing.B) {
	var x [128]uint32 // give x its own cache line
	sink = &x
	b.RunParallel(func(pb *testing.PB) {
		i := uint32(0)
		for pb.Next() {
			atomic.And32(&x[63], i)
			i++
		}
	})
}

func BenchmarkAnd64(b *testing.B) {

  %0 = "tf.Identity"(%cst) {device = ""} : (tensor<1xf32>) -> tensor<1xf32>
  %1 = "tf.AddV2"(%0, %cst_1) {device = ""} : (tensor<1xf32>, tensor<f32>) -> tensor<1xf32>
  %2 = "tf.Conv2DBackpropInput"(%arg0, %arg1, %arg2) {strides = [1, 2, 2, 1], padding="SAME", dilations=[1, 1, 1, 1]}: (tensor<4xi32>, tensor<3x3x1x32xf32>, tensor<15x14x14x32xf32>) -> tensor<15x28x28x1xf32>
  %3 = "tf.AddV2"(%2, %1): (tensor<15x28x28x1xf32>, tensor<1xf32>) -> tensor<15x28x28x1xf32>

      *literal2, xla::LiteralUtil::CreateR2<int32_t>({{3, 4}})));
}

TEST_F(PjRtExecutionUtilTest, PopulateCtxOutputs) {
  XlaOpRegistry::RegisterCompilationKernels();
  TF_EXPECT_OK(NodeDefBuilder("AddV2", "AddV2")
                   .Input(FakeInput(DT_INT32))
                   .Input(FakeInput(DT_INT32))
                   .Attr("T", DT_INT32)
                   .Device("/job:localhost/replica:0/task:0/device:XLA_CPU:0")

// Define test modules that are deserialized to module ops.
static const char *const module_with_add =
    R"(module {
func.func @main(%arg0: tensor<3x4x5xf32>, %arg1: tensor<3x4x5xf32>) -> tensor<3x4x5xf32> {
  %0 = "tf.AddV2"(%arg0, %arg1) : (tensor<3x4x5xf32>, tensor<3x4x5xf32>) -> tensor<3x4x5xf32>
  func.return %0 : tensor<3x4x5xf32>
}
}
)";

static const char *const module_with_sub =
    R"(module {

func rewriteValuedec64_OpMul64(v *Value) bool {
	v_1 := v.Args[1]
	v_0 := v.Args[0]
	b := v.Block
	typ := &b.Func.Config.Types
	// match: (Mul64 x y)
	// result: (Int64Make (Add32 <typ.UInt32> (Mul32 <typ.UInt32> (Int64Lo x) (Int64Hi y)) (Add32 <typ.UInt32> (Mul32 <typ.UInt32> (Int64Hi x) (Int64Lo y)) (Select0 <typ.UInt32> (Mul32uhilo (Int64Lo x) (Int64Lo y))))) (Select1 <typ.UInt32> (Mul32uhilo (Int64Lo x) (Int64Lo y))))
	for {
		x := v_0

    %3 = "tf.LookupTableSizeV2"(%1) {device = ""} : (tensor<!tf_type.resource>) -> tensor<i64>
    %4 = "tf.AddV2"(%2, %3) {device = ""} : (tensor<i64>, tensor<i64>) -> tensor<i64>
    %5 = "tf.LookupTableFindV2"(%0, %arg0, %cst) {device = ""} : (tensor<!tf_type.resource>, tensor<?x!tf_type.string>, tensor<i64>) -> tensor<*xi64>
    %6 = "tf.AddV2"(%5, %4) {device = ""} : (tensor<*xi64>, tensor<i64>) -> tensor<*xi64>
    return %6 : tensor<*xi64>
  }

					continue
				}
				v.reset(OpAdd32)
				v.AddArg2(y, z)
				return true
			}
		}
		break
	}
	// match: (Add32 (Add32 i:(Const32 <t>) z) x)
	// cond: (z.Op != OpConst32 && x.Op != OpConst32)
	// result: (Add32 i (Add32 <t> z x))
	for {
		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
			if v_0.Op != OpAdd32 {
				continue
			}
			_ = v_0.Args[1]

attributes  {tf.entry_function = {inputs = "input", outputs = "output_node"}} {
  %0 = arith.constant dense<2.000000e+00> : tensor<f32>
  %1 = arith.constant dense<1.000000e+00> : tensor<f32>
  %2 = "tf.AddV2"(%arg0, %1) {T = "tfdtype$DT_FLOAT", device = "", name = "StatefulPartitionedCall/add"} : (tensor<1xf32>, tensor<f32>) -> tensor<1xf32>

			Valu("addr1", OpAddr, ptrType, 0, nil, "sb"),
			Valu("addr2", OpAddr, ptrType, 0, nil, "sb"),
			Valu("addr3", OpAddr, ptrType, 0, nil, "sb"),
			Valu("zero1", OpZero, types.TypeMem, 1, c.config.Types.Bool, "addr3", "start"),
			Valu("store1", OpStore, types.TypeMem, 0, c.config.Types.Bool, "addr1", "v", "zero1"),
			Valu("store2", OpStore, types.TypeMem, 0, c.config.Types.Bool, "addr2", "v", "store1"),