return absl::OkStatus();
    }

    std::vector<int64_t> dims(num_dims, kUnknownDim);
    TF_GraphGetTensorShape(graph_, output_,
                           reinterpret_cast<int64_t*>(dims.data()), num_dims,
                           &status);
    TF_RETURN_IF_ERROR(StatusFromTF_Status(&status));
    TF_RETURN_IF_ERROR(tensorflow::TensorShapeUtils::MakeShape(dims, shape));

    return absl::OkStatus();
  }

#include <vector>

#include "tensorflow/core/platform/types.h"

struct TFE_TensorDebugInfo {
  explicit TFE_TensorDebugInfo(const std::vector<int64_t>& dims)
      : dev_dims(dims) {}

  // Fully-padded, minor-to-major.
  std::vector<int64_t> dev_dims;
};

// CHECK: %[[BROADCAST_0:.*]] = stablehlo.broadcast_in_dim %[[CONST_0]], dims = [3] : (tensor<8xf32>) -> tensor<1x1x2x8xf32>
// CHECK: %[[MUL:.*]] = stablehlo.multiply %[[ARG]], %[[BROADCAST_0]] : tensor<1x1x2x8xf32>
// CHECK: %[[BROADCAST_1:.*]] = stablehlo.broadcast_in_dim %[[CONST_1]], dims = [3] : (tensor<8xf32>) -> tensor<1x1x2x8xf32>

                                                  float data[], int64_t dims[],
                                                  int num_dims);

// Get a Matrix TensorHandle with given float values and dimensions
TFE_TensorHandle* TestTensorHandleWithDimsFloat(TFE_Context* ctx, float data[],
                                                int64_t dims[], int num_dims);

// Get a Matrix TensorHandle with given int values and dimensions

                      absl::Span<const int64_t> dims, double abs_error) {
  TF_Tensor* analytical_tensor;
  auto s = GetValue(t, &analytical_tensor);
  ASSERT_EQ(errors::OK, s.code()) << s.message();

  int64_t num_elem_analytical = 1;
  auto num_dims_analytical = TF_NumDims(analytical_tensor);
  ASSERT_EQ(dims.size(), num_dims_analytical);
  for (int j = 0; j < num_dims_analytical; j++) {

TYPED_TEST_SUITE(ConstructScalarTensorHandleTest, SimpleTypes);

// This test constructs a scalar tensor for each of the types in "SimpleTypes",
// then wraps it in a TensorHandle. We then unwrap it back into a Tensor, and
// verify the expected dims, dtype, value, num bytes, and num elements.
TYPED_TEST(ConstructScalarTensorHandleTest,
           ValidTensorAttributesAfterConstruction) {
  Status status;
  RuntimeBuilder runtime_builder;

namespace tensorflow {
namespace {

std::string ConvertToMlirString(const std::vector<int64_t>& dims,
                                bool unknown_rank, DataType dtype) {
  TensorShapeProto shape;
  shape.set_unknown_rank(unknown_rank);
  for (int64_t dim : dims) {
    shape.add_dim()->set_size(dim);
  }
  mlir::MLIRContext context;
  mlir::Builder b(&context);

  auto dims_type =
      RankedTensorType::get(/*shape=*/{1}, rewriter.getIntegerType(64));
  auto dims = DenseIntElementsAttr::get(dims_type, {feature_dim});
  if (shape_value) {
    return rewriter.createOrFold<mhlo::DynamicBroadcastInDimOp>(
        loc, result_type, value1d, shape_value, dims);
  }
  assert(result_type.hasStaticShape());
  return rewriter.create<mhlo::BroadcastInDimOp>(loc, result_type, value1d,

constexpr char longTagParam[] = "LONGTAG____________________________";
constexpr float largeValueParam = 2352352.2623433;

#define BM_ScalarSummaryDev(device, dims, name, tag, value)                 \
  void BM_ScalarSummary##name##device(::testing::benchmark::State& state) { \
    TensorShape tensorshape(DIMARGS dims);                                  \
    auto g = BM_ScalarSummaryOp(tensorshape, #tag, value);                  \

    %11 = stablehlo.convert %3 : (tensor<1x1x3xi32>) -> tensor<1x1x3xf32>
    %12 = stablehlo.broadcast_in_dim %11, dims = [0, 1, 2] : (tensor<1x1x3xf32>) -> tensor<1x4x3xf32>  // Optional
    %13 = stablehlo.subtract %10, %12 : tensor<1x4x3xf32>  // Precalculated zp_neg.
    %14 = stablehlo.broadcast_in_dim %4, dims = [0, 1, 2] : (tensor<1x1x3xf32>) -> tensor<1x4x3xf32>  // Optional
    %15 = stablehlo.multiply %13, %14 : tensor<1x4x3xf32>  // s1 * s2