extern int sum1(int a, int b) asm("_sum1");
            extern int sum2(int a, int b) asm("_sum2");
            #endif

            int main () {
                printf("foo%dfoo%d", sum1(1, 0), sum2(1, 1));
                fflush(stdout);
                return 0;
            }

            """),
            sourceFile("asm", "foo1/sum.s", getAsmSource("sum1")),

	}
	if x := sum2(); x != 2*0 {
		println("sum 0", x)
		panic("fail")
	}
	if x := sum2(10); x != 2*10 {
		println("sum 10", x)
		panic("fail")
	}
	if x := sum2(1, 8); x != 2*9 {
		println("sum 9", x)
		panic("fail")
	}
	if x := sum3(1, 2, 3); x != 3*6 {
		println("sum 6", x)
		panic("fail")
	}
	if x := sum3(); x != 3*0 {
		println("sum 0", x)
		panic("fail")
	}
	if x := sum3(10); x != 3*10 {

	h1 := new(Hash)
	h1.WriteString("testing")
	sum1 := h1.Sum64()

	h1.Reset()
	h1.WriteString("testing")
	sum2 := h1.Sum64()

	if sum1 != sum2 {
		t.Errorf("different sum after resetting: %#x != %#x", sum1, sum2)
	}

	h2 := new(Hash)
	h2.SetSeed(h1.Seed())
	h2.WriteString("testing")
	sum3 := h2.Sum64()

	if sum1 != sum3 {
		t.Errorf("different sum on the same seed: %#x != %#x", sum1, sum3)
	}
}

	sum2 := 0
	cnt := 0
	for i := 0; i < 2; i++ {
		for {
			node := toMyNode(LFStackPop(stacks[i]))
			if node == nil {
				break
			}
			cnt++
			sum2 += node.data
			node.Next = 0
		}
	}
	if cnt != K {
		t.Fatalf("Wrong number of nodes %d/%d", cnt, K)
	}
	if sum2 != sum {
		t.Fatalf("Wrong sum %d/%d", sum2, sum)
	}

				t.Fatalf("UnmarshalBinary: %v", err)
			}
			h2.Write(buf[split:])
			h3.Write(buf[split:])
			sum2 := h2.Sum(nil)
			sum3 := h3.Sum(nil)
			if !bytes.Equal(sum2, sum) {
				t.Fatalf("Sum after MarshalBinary = %x, want %x", sum2, sum)
			}
			if !bytes.Equal(sum3, sum) {
				t.Fatalf("Sum after UnmarshalBinary = %x, want %x", sum3, sum)
			}
		})
	}

// calling convention by default, since this is redundant with the existing
// tests.
var cbFuncsRegABI = []cbFunc{
	{sum2},
	{sum3},
	{sum4},
	{sum5},
	{sum6},
	{sum7},
	{sum8},
	{sum9},
	{sum10},
	{sum9uint8},
	{sum9uint16},
	{sum9int8},
	{sum5mix},
	{sum5andPair},
	{sum9andGC},
}

func getCallbackTestFuncs() []cbFunc {

  // CHECK: %[[SUM1:.*]] = "tf.AddV2"(%arg2, %arg3)
  // CHECK: %[[SUM2:.*]] = "tf.AddV2"(%[[SUM0]], %[[SUM1]])
  // CHECK: %[[SUM3:.*]] = "tf.AddV2"(%[[SUM2]], %arg4)
  // return %[[SUM3]]
  %0 = "tf.AddN"(%arg0, %arg1, %arg2, %arg3, %arg4) : (tensor<*xf32>, tensor<*xf32>, tensor<*xf32>, tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>
  func.return %0 : tensor<*xf32>
}

// CHECK-LABEL: func @addN_variant

// CHECK-SAME:        window_strides = dense<[1, 3, 4, 1]>
// CHECK:           ^bb0(%[[ARG3:.*]]: tensor<f32>, %[[ARG4:.*]]: tensor<f32>):
// CHECK:             %[[SUM2:.*]] = mhlo.add %[[ARG3]], %[[ARG4]] : tensor<f32>
// CHECK:             mhlo.return %[[SUM2]] : tensor<f32>
// CHECK:           })
// CHECK-SAME:        -> tensor<2x4x6x7xf32>

//
// is lowered to:
//
//   %sum0 = "tf.AddV2"(%0, %1)
//   %result = "tf.AddV2"(%sum0, %2)
//
// While
//
//   %result = "tf.AddN"(%0, %1, %2, %3, %4)
//
// is lowered to:
//
//   %sum0 = "tf.AddV2"(%0, %1)
//   %sum1 = "tf.AddV2"(%2, %3)
//   %sum2 = "tf.AddV2"(%sum0, %sum1)
//   %result = "tf.AddV2"(%sum2, %4)
//
class LowerAddNOp : public RewritePattern {
 public:

  %0 = "tf.XlaVariadicReduce"(%arg0, %arg1) {dimensions_to_reduce = [], reducer = @sum2, device = "/job:localhost/replica:0/task:0/device:GPU:0"} : (tensor<3x4xf32>, tensor<f32>) -> tensor<?x?xf32>
  func.return %0 : tensor<?x?xf32>
}