typedef uint32_t VkFlags;
typedef VkFlags  VkDeviceQueueCreateFlags;
typedef uint32_t VkStructureType;

typedef struct VkDeviceQueueCreateInfo {
    VkStructureType             sType;
    const void*                 pNext;
    VkDeviceQueueCreateFlags    flags;
    uint32_t                    queueFamilyIndex;
    uint32_t                    queueCount;

  absl::flat_hash_map<std::pair<std::vector<int32_t>, std::vector<int32_t>>,
                      std::vector<uint8_t>>
      custom_op_data_map_;
  absl::flat_hash_map<std::pair<std::vector<int32_t>, std::vector<int32_t>>,
                      std::pair<uint64_t, uint64_t>>
      custom_op_idx_map_;

  // Points to TensorFlow and TFLite dialects, respectively. nullptr if the
  // dialect is not registered.

	case sys.MIPS, sys.MIPS64:
		eh.Machine = uint16(elf.EM_MIPS)
	case sys.Loong64:
		eh.Machine = uint16(elf.EM_LOONGARCH)
	case sys.ARM:
		eh.Machine = uint16(elf.EM_ARM)
	case sys.AMD64:
		eh.Machine = uint16(elf.EM_X86_64)
	case sys.ARM64:
		eh.Machine = uint16(elf.EM_AARCH64)
	case sys.I386:
		eh.Machine = uint16(elf.EM_386)
	case sys.PPC64:
		eh.Machine = uint16(elf.EM_PPC64)
	case sys.RISCV64:

  void add_adj_x(bool adj_x) {
    fbb_.AddElement<uint8_t>(BatchMatMulOptions::VT_ADJ_X, static_cast<uint8_t>(adj_x), 0);
  }
  void add_adj_y(bool adj_y) {
    fbb_.AddElement<uint8_t>(BatchMatMulOptions::VT_ADJ_Y, static_cast<uint8_t>(adj_y), 0);
  }
  void add_asymmetric_quantize_inputs(bool asymmetric_quantize_inputs) {

  if (element_ty.isF16()) {
    uint16_t raw_epsilon = Eigen::numext::bit_cast<uint16_t>(
        Eigen::NumTraits<Eigen::half>::epsilon());
    auto value = APFloat(APFloat::IEEEhalf(), APInt(16, raw_epsilon));
    return DenseElementsAttr::get(scalar_ty, value);
  } else if (element_ty.isBF16()) {
    uint16_t raw_epsilon = Eigen::numext::bit_cast<uint16_t>(
        Eigen::NumTraits<Eigen::bfloat16>::epsilon());

}

template <>
inline std::vector<int16_t> GetVector(DenseElementsAttr elements) {
  auto type = elements.getType();
  auto elemType = type.getElementType();
  if (elemType.isSignlessInteger(16)) {
    auto vec = llvm::to_vector(llvm::map_range(
        elements.getValues<APInt>(),
        [&](APInt value) -> int16_t { return value.getSExtValue(); }));
    return std::vector<int16_t>(vec.begin(), vec.end());
  }

             SP_DeviceMemoryBase* location, uint64_t size, TF_Status* status) {}

void Memset(const SP_Device* device, SP_Stream stream,
            SP_DeviceMemoryBase* location, uint8_t pattern, uint64_t size,
            TF_Status* status) {}

void Memset32(const SP_Device* device, SP_Stream stream,
              SP_DeviceMemoryBase* location, uint32_t pattern, uint64_t size,
              TF_Status* status) {}

          return true;
        }
      }
    }
    return false;
  }

  // Returns the producer op code of the specified tensor_idx.
  bool GetProducerOpCode(const Model* model, uint32_t subgraph_idx,
                         uint32_t tensor_idx, BuiltinOperator* op_code) {
    const auto subgraph = model->subgraphs()->Get(subgraph_idx);
    for (size_t op_idx = 0; op_idx < subgraph->operators()->size(); ++op_idx) {

#include <cpuid.h>
#include <stdint.h>
#include <x86intrin.h>

// Need to wrap __get_cpuid_count because it's declared as static.
int
gccgoGetCpuidCount(uint32_t leaf, uint32_t subleaf,
                   uint32_t *eax, uint32_t *ebx,
                   uint32_t *ecx, uint32_t *edx)
{
	return __get_cpuid_count(leaf, subleaf, eax, ebx, ecx, edx);
}

#pragma GCC diagnostic ignored "-Wunknown-pragmas"
#pragma GCC push_options

                                    clamp_max_double_);

    uint64_t signless_result;
    if (is_signed_) {
      int64_t clamped_int = static_cast<int8_t>(clamped);
      memcpy(&signless_result, &clamped_int, sizeof(clamped_int));
    } else {
      signless_result = static_cast<uint8_t>(clamped);
    }
    return APInt(storage_bit_width_, signless_result);
  }