// well with `MatMul` and remove `TestMatMul*` in
  // `mnist_gradients_test` when done.
  GTEST_SKIP();

  float A_vals[] = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f};
  int64_t A_dims[] = {3, 3};
  AbstractTensorHandlePtr A;
  {
    AbstractTensorHandle* A_raw;
    status_ = TestTensorHandleWithDims<float, TF_FLOAT>(

  ASSERT_EQ(errors::OK, status_.code()) << status_.message();

  auto ReluGradModel = BuildGradModel(ReluModel, registry_);

  float X_vals[] = {1.0f, 2.0f, 3.0f, -5.0f, -4.0f, -3.0f, 2.0f, 10.0f, -1.0f};
  int64_t X_dims[] = {3, 3};
  AbstractTensorHandlePtr X;
  {
    AbstractTensorHandle* X_raw;
    status_ = TestTensorHandleWithDims<float, TF_FLOAT>(

[#tf_type.shape<?>], dim_args_spec = [], function_list = [@_tf_func], module = "ML\EFR\07StableHLO_v0.12.0\00\01\19\05\01\05\01\03\05\03\09\07\09\0B\0D\03\8Fm\0F\01?\0B\07\0B\0B\0B\0B\0B\13\0B\0F\133\133\13\13S\0B\0B\0B\0B\0B\0B\0B\0B\0B\13\13\0B\13\13\03/\0B\0B\0B\13\1B\0B\0B\0B\0B\0B\0B\0F\13\0B\0B\0B\0B\0B\0B\0B\13\0B\0F\01\03\0F\03\0D3\07\0B\1B\17\07\02:\03\05\0F\1F\05\11\05\13\05\15\05\17\05\19\03\03\11\13\05\1B\11\01\05\17\01S\15\03\0B\05E\07S\09U\0B[\0DA\17\011\07\03\0B\05?\07]\09?\0BC\0D_\17\01'\07\17...

disabled_checks = [], function_list = [], has_token_input_output = false, module = "ML\EFR\01StableHLO_v0.9.0\00\01\17\05\01\03\01\03\05\03\07\07\09\0B\03]?\0B\01)\07\0F\0B+\0B\0F\0B\0B\0B3\0B\0B\0B\0B\0F\0B\0F\0B\13\0B\03\17\0F\13\0B\0B\0B\0F\13\0B\0B\0B\0B\01\05\0B\0F\03\07\17\17\07\02\D7\1F\11\03\05\05\0D\03\09\09\0B\0D\03\0F\03\05\11\05\0F\11\01\00\05\11\05\13\05\15\03\0B\15)\171\193\05;\1B=\05\17\05\19\05\1B\05\1D\1D\1F\01\05\1F\1D#%\05!\17'\A9\01\05#\03\03+\0D\03-/\1D%\1D'#\07\03\035\0D\0379\1D)...

  ASSERT_EQ(3, TF_OutputListNumOutputs(func_outputs));

  float expected_outputs[3][4] = {{1.0f, 1.0f, 1.0f, 1.0f},
                                  {2.0f, 0.0f, 0.0f, 2.0f},
                                  {2.0f, 2.0f, 2.0f, 2.0f}};

  float result_data[4];
  for (int idx = 0; idx < 3; ++idx) {
    TF_AbstractTensor* result = TF_OutputListGet(func_outputs, idx);

  collector.Collect(
      /*min=*/1.0f, /*max=*/10.f, /*histogram=*/{});
  collector.Collect(
      /*min=*/2.0f, /*max=*/5.f, /*histogram=*/{});

  statistics = collector.GetStatistics();

  EXPECT_TRUE(statistics.has_value());
  EXPECT_EQ(statistics.value().min_max_statistics().global_min(), 1.0f);
  EXPECT_EQ(statistics.value().min_max_statistics().global_max(), 10.0f);
}

  ASSERT_TRUE(v.ok()) << v.status().message();
  const TaggedValue& result = v.value();
  ExpectEquals(result.tensor().get(), 4.0f);
}

TEST_P(FunctionTest, IdentityN) {
  impl::TaggedValueTensor x = CreateScalarTensor<float, TF_FLOAT>(2.0f);
  impl::TaggedValueTensor y = CreateScalarTensor<float, TF_FLOAT>(4.0f);
  FunctionDef fdef = IdentityNFunc();
  AbstractFunctionPtr trace(new GraphFunction(fdef), /*add_ref=*/false);

  bool UseMlir() const { return strcmp(std::get<0>(GetParam()), "mlir") == 0; }
  bool UseFunction() const { return std::get<2>(GetParam()); }
};

TEST_P(GradientCheckerTest, TestMatMul) {
  float A_vals[] = {1.0f, 2.0f, 3.0f, 4.0f};
  int64_t A_dims[] = {2, 2};
  AbstractTensorHandlePtr A;
  {
    AbstractTensorHandle* A_raw;
    Status s = TestTensorHandleWithDims<float, TF_FLOAT>(ctx_.get(), A_vals,

def QuantActRangeReluN1To1Pattern :
  Pattern<
    (TFR_TFRQuantActRangeOp
     (TFR_ConstOp HasStringAttr<"RELU_N1_TO_1">:$act),
     (ConstantLikeMatcher F32Attr:$scale),
     (ConstantLikeMatcher I64Attr:$zp)),
    [(TFR_ConstantTensorOp (Arith_ConstantOp (Quantize<"-1.0f"> $scale, $zp))),

}

TEST(CCOpTest, CompositeOp) {
  Scope root = Scope::NewRootScope();
  auto l = Linear(root.WithOpName("layer0"), {{10.0f, -3.0f}},
                  {{.8f, .5f}, {.1f, .6f}}, {-8.0f, 31.0f});
  TF_EXPECT_OK(root.status());
  EXPECT_EQ(l.node()->name(), "layer0");
  Tensor out;
  test::GetTensor(root, l, &out);