func.return %1, %2 : tensor<2x1xf32>, tensor<2x1xf32>

// CHECK:  %cst = arith.constant
// CHECK:  %[[FQ:.*]] = "tf.FakeQuantWithMinMaxVars"(%arg0, %arg1, %arg2)
// CHECK:  %[[T1:.*]] = "tf.Transpose"(%[[FQ]], %cst)
// CHECK:  %[[T2:.*]] = "tf.Transpose"(%[[FQ]], %cst)
}

// CHECK-LABEL: fakeQuantFollowedByReshape

// CHECK: "tfl.fully_connected"(%arg0, %arg1, [[NONE]])
// CHECK-SAME: (tensor<40x37xf32>, tensor<40x37xf32>, none) -> (tensor<40x40xf32>, tensor<40x40xf32>)
  %cst = "tfl.no_value"() {value = unit} : () -> none
  %0:2 = "tfl.fully_connected"(%arg0, %arg1, %cst) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<40x37xf32>, tensor<40x37xf32>, none) -> (tensor<40x40xf32>, tensor<40x40xf32>)

  func.return %1, %2 : tensor<2x1xf32>, tensor<2x1xf32>

// CHECK:  %cst = arith.constant
// CHECK:  %[[FQ:.*]] = "tf.FakeQuantWithMinMaxVars"(%arg0, %arg1, %arg2)
// CHECK:  %[[T1:.*]] = "tf.Transpose"(%[[FQ]], %cst)
// CHECK:  %[[T2:.*]] = "tf.Transpose"(%[[FQ]], %cst)
}

// CHECK-LABEL: fakeQuantFollowedByReshape

    return %0 : tensor<*xf32>
  }

// CHECK: %[[cst:.*]] = "arith.constant"() <{value = dense<0.000000e+00> : tensor<2x1024xf32>}> : () -> tensor<2x1024xf32>
// CHECK: %[[q_cst:.*]] = "quantfork.qcast"(%[[cst]]) : (tensor<2x1024xf32>) -> tensor<2x1024x!quant.uniform<i8<-127:127>:f32, 3.9370078740157481E-9>>

	ctx, cancel := context.WithCancel(context.Background())

	cst := newClientServerTest(t, mode, HandlerFunc(func(w ResponseWriter, r *Request) {
		w.(Flusher).Flush()
		<-testDone
	}))
	defer close(testDone)

	cst.c.Timeout = 1 * time.Hour
	req, _ := NewRequest("GET", cst.ts.URL, nil)
	req.Cancel = ctx.Done()
	res, err := cst.c.Do(req)
	if err != nil {
		t.Fatal(err)
	}
	cancel()

func.func @mean_4d_keepdim(%arg0: tensor<1x48x48x512xf32>) -> tensor<1x1x1x512xf32> {
  %cst = arith.constant dense<[1, 2]> : tensor<2xi32>
  %0 = "tfl.mean"(%arg0, %cst) {keep_dims = true} : (tensor<1x48x48x512xf32>, tensor<2xi32>) -> tensor<1x1x1x512xf32>
  func.return %0 : tensor<1x1x1x512xf32>
}

// CHECK-NEXT:  return 
//
module {
  func.func @serving_default() -> tensor<*xi32> {
    %cst = "tf.Const"() <{value = dense<"some_test.txt"> : tensor<!tf_type.string>}> : () -> tensor<!tf_type.string>
    %0 = "tf.VarHandleOp"() <{container = "", shared_name = "Variable"}> : () -> tensor<!tf_type.resource<tensor<!tf_type.string>>>

// If the tfl.dequantize op wasn't fused, we shouldn't quantize the floating
// point constant.
def : Pat<(TFL_DequantizeOp
             (TFL_QuantizeOp (Arith_ConstantOp F32ElementsAttr:$cst), $qt)),
          (TFL_ConstOp $cst)>;

// Transpose conv supports hybrid computation with quantized weights.
def FoldQuantWeightsIntoTposeConv : Pat<
  (TFL_TransposeConvOp
    $output_shape,
    (TFL_DequantizeOp $quant_weights),

// CHECK-LABEL: HandleReturnedDequantizeWithAnotherUse
func.func @HandleReturnedDequantizeWithAnotherUse(%arg0: tensor<128x16xf32>) -> (tensor<128x16xf32>, tensor<128xi32>) {
// CHECK-NEXT:  %[[cst:.*]] = arith.constant dense<1> : tensor<i32>
  %cst = arith.constant dense<1> : tensor<i32>
// CHECK-NEXT:  %[[softmax:.*]] = "tfl.softmax"(%arg0) <{beta = 1.000000e+00 : f32}> : (tensor<128x16xf32>) -> tensor<128x16xf32>

    "Is not a constant.">;

def : Pat<(StableHLO_DivOp $l,
            (StableHLO_ConstantOp:$divisor FloatElementsAttr<32>:$cst)),
          (StableHLO_MulOp $l,
            (StableHLO_DivOp
              (StableHLO_ConstantOp (CloneF32ElementsAttrWithOnes $cst)),
              $divisor)),