#include "tensorflow/c/experimental/filesystem/filesystem_interface.h"
#include "tensorflow/c/tf_status.h"

// Implementation of a filesystem for POSIX environments.
// This filesystem will support `file://` and empty (local) URI schemes.

static void* plugin_memory_allocate(size_t size) { return calloc(1, size); }
static void plugin_memory_free(void* ptr) { free(ptr); }

// SECTION 1. Implementation for `TF_RandomAccessFile`

    // Grad for Softmax Input
    TF_RETURN_IF_ERROR(BroadcastMul(
        ctx, grad_outputs[0], forward_outputs_[1],
        grad_inputs.subspan(0, 1)));  // upstream_grad * local softmax grad

    // Grad for labels is null
    grad_inputs[1] = nullptr;
    return absl::OkStatus();
  }
  ~SparseSoftmaxCrossEntropyWithLogitsGradientFunction() override {}

 private:

  const char remote_device_name[] =
      "/job:localhost/replica:0/task:1/device:CPU:0";
  const char local_device_name[] =
      "/job:localhost/replica:0/task:0/device:CPU:0";
  CheckRemoteMatMulExecutesOK(ctx, remote_device_name, local_device_name);

  TFE_Executor* executor = TFE_ContextGetExecutorForThread(ctx);
  TFE_ExecutorWaitForAllPendingNodes(executor, status);

#include "tensorflow/c/tf_file_statistics.h"
#include "tensorflow/c/tf_status.h"

// Implementation of a filesystem for POSIX environments.
// This filesystem will support `file://` and empty (local) URI schemes.

static void* plugin_memory_allocate(size_t size) { return calloc(1, size); }
static void plugin_memory_free(void* ptr) { free(ptr); }

// SECTION 1. Implementation for `TF_RandomAccessFile`

      return underlying_devices_;
    }
  }
  std::vector<std::string> local_names;
  local_names.reserve(underlying_devices_.size());
  for (const DeviceNameUtils::ParsedName& parsed_component :
       parsed_components) {
    local_names.push_back(
        absl::StrCat(parsed_component.type, ":", parsed_component.id));
  }
  return local_names;
}

std::unique_ptr<ParallelTensor> ParallelTensor::FromTensorHandles(

  // Pack 3 variable handles into one TFE_TensorHandle.
  // When remote is false, function device is placed on task0. Handle types are
  // REMOTE, REMOTE, LOCAL on task0. When remote is true, function device is
  // placed on task1, Handle types are LOCAL, REMOTE, LOCAL on task1.
  int num_replicas = 3;
  std::vector<TFE_TensorHandle*> handles = {h0, h1, h2};
  TFE_TensorHandle* packed_handle =

  const char remote_device[] = "/job:worker/replica:0/task:1/device:CPU:0";

  // Create a variable `var` on `worker2` using `local_ctx`.
  TFE_TensorHandle* handle_0 =
      CreateVariable(local_ctx, 1.2, remote_device, /*variable_name=*/"var");
  TF_Status* status = TF_NewStatus();
  TFE_ContextAsyncWait(local_ctx, status);
  EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TF_DeleteStatus(status);

// If the flag is not used, no shared objects are loaded.
//
// Every filesystem provides support for accessing URIs of form
// `[<scheme>://]<path>` where `<scheme>` is optional (if missing, we are
// accessing local paths). This test suite tests exactly one scheme for each
// invocation. By default, we are testing all schemes available but this can be
// restricted by using `--schemes` to specify a set of schemes to test.
//

    return nullptr;
  }
  tensorflow::TensorHandle* handle =
      tensorflow::TensorHandleFromInterface(tensorflow::unwrap(h));
  if (handle->Type() != TensorHandle::LOCAL) {
    status->status = tensorflow::errors::InvalidArgument(
        "DLPack doesn't support ", handle->TypeString(), " tensor");
    return nullptr;
  }
  const tensorflow::Tensor* tensor;

  } else {
    // Handles are on task0 (local), and task2, but op is on task1.
    matmul = MatMulOp(ctx, h0_task0, h1_task2);
  }
  if (remote) {
    TFE_OpSetDevice(matmul, task1_name, status);
    ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);
  } else if (!async) {
    // Set the local device to CPU to easily validate mirroring
    string cpu_device_name;