Request - Documentation Bug - Others validations: required: true - type:
markdown attributes: value: | Please make sure that this is a bug. As per our
[GitHub Policy](https://github.com/tensorflow/tensorflow/blob/master/ISSUES.md),we
only address code/doc bugs, performance issues, feature requests and
build/installation issues on GitHub.

-   type: markdown
    attributes:
      value: |

+++ b/sysdeps/aarch64/dl-machine.h
@@ -49,23 +49,11 @@ elf_machine_load_address (void)
   /* To figure out the load address we use the definition that for any symbol:
      dynamic_addr(symbol) = static_addr(symbol) + load_addr
 
-     The choice of symbol is arbitrary. The static address we obtain
-     by constructing a non GOT reference to the symbol, the dynamic
-     address of the symbol we compute using adrp/add to compute the

//
// Like a TF_Tensor, a TFE_TensorHandle refers to a tensor with a value, shape,
// type etc. Unlike a TF_Tensor, a TFE_TensorHandle may refer to such tensors
// placed in the memory of different devices or remote address spaces.
typedef struct TFE_TensorHandle TFE_TensorHandle;

TF_CAPI_EXPORT extern TFE_TensorHandle* TFE_NewTensorHandle(const TF_Tensor* t,
                                                            TF_Status* status);

#include "tensorflow/core/lib/llvm_rtti/llvm_rtti.h"
#include "tensorflow/core/platform/errors.h"

namespace tensorflow {
namespace gradients {
namespace {

// TODO(b/172558015): Using the pointer address as the identifier for the tensor
// may lead to collisions. Introduce another way to get a unique id for this
// tensor.
int64_t ToId(const AbstractTensorHandle* t) {
  return static_cast<int64_t>(reinterpret_cast<uintptr_t>(t));

    int num_dims, void* data, size_t len,
    void (*deallocator)(void* data, size_t len, void* arg),
    void* deallocator_arg, TF_Status* status);

// Retrieves the address space (i.e. job, replia, task) of the local host and
// saves it in the buffer.
TF_CAPI_EXPORT extern void TFE_HostAddressSpace(TFE_Context* ctx,
                                                TF_Buffer* buf);

}

void TFE_HostAddressSpace(TFE_Context* ctx, TF_Buffer* buf) {
  auto address_space = tensorflow::DeviceNameUtils::AddressSpace(
      tensorflow::unwrap(ctx)->HostCPUParsedName());
  auto str = tensorflow::DeviceNameUtils::ParsedNameToString(address_space);
  void* data = tensorflow::port::Malloc(str.length());
  str.copy(static_cast<char*>(data), str.length(), 0);
  buf->data = data;

namespace tf_read_only_memory_region {
typedef struct GCSMemoryRegion {
  const void* const address;
  const uint64_t length;
} GCSMemoryRegion;

void Cleanup(TF_ReadOnlyMemoryRegion* region) {
  auto r = static_cast<GCSMemoryRegion*>(region->plugin_memory_region);
  plugin_memory_free(const_cast<void*>(r->address));
  delete r;
}

const void* Data(const TF_ReadOnlyMemoryRegion* region) {

## Known Bugs

*   Using XLA:GPU with CUDA 9 and CUDA 9.1 results in garbage results and/or
    `CUDA_ILLEGAL_ADDRESS` failures.

    Google discovered in mid-December 2017 that the PTX-to-SASS compiler in CUDA
    9 and CUDA 9.1 sometimes does not properly compute the carry bit when
    decomposing 64-bit address calculations with large offsets (e.g. `load [x +
    large_constant]`) into 32-bit arithmetic in SASS.

  } else {
    const void* address =
        mmap(nullptr, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
    if (address == MAP_FAILED) {
      TF_SetStatusFromIOError(status, errno, path);
    } else {
      region->plugin_memory_region =
          new tf_read_only_memory_region::PosixMemoryRegion{
              address, static_cast<uint64_t>(st.st_size)};
      TF_SetStatus(status, TF_OK, "");

  std::vector<Tensor> all_input_tensors;
  // Update the vector with information from `input_tensors` if provided.
  if (input_tensors != nullptr) {
    // Note that we take the address of the elements in `all_input_tensors`
    // below. Allocate enough space so that no reallocation happens, which will
    // make the pointers invalid.
    all_input_tensors.reserve(num_inputs);