//go:noescape
func Or8(ptr *uint8, val uint8)

//go:noescape
func And(ptr *uint32, val uint32)

//go:noescape
func Or(ptr *uint32, val uint32)

//go:noescape
func And32(ptr *uint32, val uint32) uint32

//go:noescape
func Or32(ptr *uint32, val uint32) uint32

//go:noescape
func Anduintptr(ptr *uintptr, val uintptr) uintptr

//go:noescape

    %cast = "tf.Cast"(%accumulation) {Truncate = false} : (tensor<*xi32>) -> tensor<*xf32>
    %mul = "tf.Mul"(%cast, %rescale_factor) : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>
    %add = "tf.AddV2"(%mul, %float_out_zp) : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>
    func.return %add : tensor<*xf32>
  }

func Or8(ptr *uint8, val uint8)

// NOTE: Do not add atomicxor8 (XOR is not idempotent).

//go:noescape
func And(ptr *uint32, val uint32)

//go:noescape
func Or(ptr *uint32, val uint32)

//go:noescape
func And32(ptr *uint32, val uint32) uint32

//go:noescape
func Or32(ptr *uint32, val uint32) uint32

//go:noescape
func And64(ptr *uint64, val uint64) uint64

//go:noescape
func Or64(ptr *uint64, val uint64) uint64

			}
		}
	}
}

func TestSameLocalAddrs(t *testing.T) {
	testCases := []struct {
		addr1       string
		addr2       string
		sameAddr    bool
		expectedErr error
	}{
		{"", "", false, errors.New("unable to process empty address")},
		{":9000", ":9000", true, nil},
		{"localhost:9000", ":9000", true, nil},

  TF_Operation* one = ScalarConst(1, params_->body_graph, s_);
  ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);
  TF_Operation* add2 = Add(add1, one, params_->body_graph, s_, "add2");
  ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);
  params_->body_outputs[0] = {add2, 0};
  params_->body_outputs[1] = params_->body_inputs[1];

  // Finalize while loop
  ExpectOK();

func Add(x, y, carry uint) (sum, carryOut uint) {
	if UintSize == 32 {
		s32, c32 := Add32(uint32(x), uint32(y), uint32(carry))
		return uint(s32), uint(c32)
	}
	s64, c64 := Add64(uint64(x), uint64(y), uint64(carry))
	return uint(s64), uint(c64)
}

// Add32 returns the sum with carry of x, y and carry: sum = x + y + carry.
// The carry input must be 0 or 1; otherwise the behavior is undefined.

	MOVL	val+8(FP), CX
casloop:
	MOVL 	CX, DX
	MOVL	(BX), AX
	ORL	AX, DX
	LOCK
	CMPXCHGL	DX, (BX)
	JNZ casloop
	MOVL 	AX, ret+16(FP)
	RET

// func And32(addr *uint32, v uint32) old uint32
TEXT ·And32(SB), NOSPLIT, $0-20
	MOVQ	ptr+0(FP), BX
	MOVL	val+8(FP), CX
casloop:
	MOVL 	CX, DX
	MOVL	(BX), AX
	ANDL	AX, DX
	LOCK
	CMPXCHGL	DX, (BX)
	JNZ casloop
	MOVL 	AX, ret+16(FP)

    ASSERT_EQ(errors::OK, status_.code()) << status_.message();
    y.reset(y_raw);
  }

  // TODO(srbs): Rename ops::Add to ops::AddV2 and AddRegister to
  // AddV2Registerer.
  status_ = registry_.Register("AddV2", AddRegisterer);
  ASSERT_EQ(errors::OK, status_.code()) << status_.message();

  ASSERT_NO_FATAL_FAILURE(CompareNumericalAndAutodiffGradients(

	MOVW	val+8(FP), R5
	MOVW	(R4), R3
repeat:
	OR	R5, R3, R6
	CS	R3, R6, (R4) // if R3==(R4) then (R4)=R6 else R3=(R4)
	BNE	repeat
	MOVW	R3, ret+16(FP)
	RET

// func And32(addr *uint32, v uint32) old uint32
TEXT ·And32(SB), NOSPLIT, $0-20
	MOVD	ptr+0(FP), R4
	MOVW	val+8(FP), R5
	MOVW	(R4), R3
repeat:
	AND	R5, R3, R6
	CS	R3, R6, (R4) // if R3==(R4) then (R4)=R6 else R3=(R4)
	BNE	repeat

// NOTE: Do not add atomicxor8 (XOR is not idempotent).

//go:noescape
func And(ptr *uint32, val uint32)

//go:noescape
func Or(ptr *uint32, val uint32)

//go:noescape
func And32(ptr *uint32, val uint32) uint32

//go:noescape
func Or32(ptr *uint32, val uint32) uint32

//go:noescape
func And64(ptr *uint64, val uint64) uint64

//go:noescape