module {
  func.func @main() -> tensor<*xi32> {
    %0 = "tf.VarHandleOp"() {container = "c", shared_name = "v"} : () -> tensor<*x!tf_type.resource<tensor<*xi32>>>
    %cst = arith.constant dense<2> : tensor<4xi32>
    "tf.AssignAddVariableOp"(%0, %cst) {} : (tensor<*x!tf_type.resource<tensor<*xi32>>>, tensor<4xi32>) -> ()
    %1 = "tf.ReadVariableOp"(%0) {dtype = i32} : (tensor<*x!tf_type.resource<tensor<*xi32>>>) -> tensor<*xi32>

func.func @dot(%arg0: tensor<?x3xf32>) -> tensor<?x2xf32> {
  // CHECK: %[[cst:.*]] = stablehlo.constant
  // CHECK: %[[q1:.*]] = "quantfork.qcast"(%[[cst]])
  // CHECK-SAME: quant.uniform<i8:f32, 0.0040316890267764818:127>
  // CHECK: %[[dq1:.*]] = "quantfork.dcast"(%[[q1]])
  // CHECK-SAME: quant.uniform<i8:f32, 0.0040316890267764818:127>

  %cst = "tfl.pseudo_qconst"() {qtype = tensor<1x2x!quant.uniform<u8:f32, 1.0>>, value = dense<-76> : tensor<1x2xi8>} : () -> tensor<1x2x!quant.uniform<u8:f32, 1.0>>

    %cst = stablehlo.constant dense<0.000000e+00> : tensor<10x1x3xf32>
    %0 = stablehlo.dot_general %arg0, %arg1, batching_dims = [0] x [0], contracting_dims = [2] x [1], precision = [DEFAULT, DEFAULT] {mhlo.frontend_attributes = {grad_x = "false", grad_y = "false"}} : (tensor<10x1x1024xf32>, tensor<10x1024x3xf32>) -> tensor<10x1x3xf32>
    %1 = stablehlo.maximum %0, %cst : tensor<10x1x3xf32>
    return %1 : tensor<10x1x3xf32>

    %cst = arith.constant dense<0.000000e+00> : tensor<5xf32>
    %cst_0 = arith.constant dense<1.0> : tensor<5x384xf32>
    %cst_1 = arith.constant dense<1.0> : tensor<5x384xf32>

  %cst = "tf.Const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi32>} : () -> tensor<4xi32>
  %0 = "tf.AddV2"(%arg0, %arg1) {device = ""} : (tensor<1x1x2x3xf32>, tensor<2x3xf32>) -> tensor<1x1x2x3xf32>
  %1 = "tf.Transpose"(%0, %cst) {device = ""} : (tensor<1x1x2x3xf32>, tensor<4xi32>) -> tensor<1x2x1x3xf32>

  func.return %1 : tensor<1x2x1x3xf32>

// CHECK-LABEL: fuseMulIntoPerTensorConv2dWithQDQs
func.func @fuseMulIntoPerTensorConv2dWithQDQs(%arg0: tensor<256x32x32x3xf32>) -> tensor<256x8x7x3xf32> {
  %cst = arith.constant dense<1.5> : tensor<3xf32>
  %cst_0 = arith.constant dense<[1.0, 2.0, 3.0]> : tensor<3xf32>
  %w = arith.constant dense<2.0> : tensor<3x3x3x3xf32>

        self.assertIn('stablehlo.add %arg0, %cst : tensor<4xf32>', str(module))

  def test_tf_mlir_to_stablehlo(self):
    assembly = """
      module attributes {tf.versions = {bad_consumers = [], min_consumer = 0 : i32, producer = 268 : i32}} {
        func.func @main(%arg0 : tensor<?xf32>) -> tensor<?xf32> {
          %cst = "tf.Const"() {value = dense<1.0> : tensor<f32>} : () -> tensor<f32>

  %0 = "tf.BroadcastTo"(%arg0, %arg1) : (tensor<3xbf16>, tensor<2xi64>) -> tensor<3x3xbf16>
  func.return %0: tensor<3x3xbf16>

// CHECK-LABEL: broadcast_to_bf16
// CHECK:  [[CST:%.*]] = arith.constant dense<1.000000e+00> : tensor<3x3xbf16>
// CHECK:  [[MUL:%.*]] = tfl.mul(%arg0, [[CST]]) <{fused_activation_function = "NONE"}> : (tensor<3xbf16>, tensor<3x3xbf16>) -> tensor<3x3xbf16>
// CHECK:  return [[MUL]] : tensor<3x3xbf16>

      // Some loop invariant math
      %li0 = "tf.Div"(%a, %b) : (tensor<4xf32>, tensor<4xf32>) -> tensor<4xf32>
      %cst = arith.constant dense<22.0> : tensor<f32>
      %li1 = "tf.Mul"(%li0, %cst) : (tensor<4xf32>, tensor<f32>) -> tensor<4xf32>

      %final = "tf.Add"(%add, %li1) : (tensor<4xf32>, tensor<4xf32>) -> tensor<4xf32>