TF_CAPI_EXPORT extern TF_Thread* TF_StartThread(const TF_ThreadOptions* options,
                                                const char* thread_name,
                                                void (*work_func)(void*),
                                                void* param);

// Waits for the given thread to finish execution, then deletes it.
TF_CAPI_EXPORT extern void TF_JoinThread(TF_Thread* thread);

  const std::vector<std::string> underlying_devices_;
  // A sequence of thread wrappers, one per device, for executing operations in
  // parallel.
  //
  // Conceptually this is a thread pool with one thread per device. It requires
  // less synchronization than a thread pool would for this task, since Execute
  // acquires each thread in order (and so only one Execute will schedule

// --------------------------------------------------------------------------
// The new graph construction API, still under development.

// Represents a computation graph.  Graphs may be shared between sessions.
// Graphs are thread-safe when used as directed below.
typedef struct TF_Graph TF_Graph;

// Return a new graph object.
TF_CAPI_EXPORT extern TF_Graph* TF_NewGraph(void);

// Destroy an options object. Graph will be deleted once no more

// Clears the internal caches in the TFE context. Useful when reseeding random
// ops.
TF_CAPI_EXPORT extern void TFE_ContextClearCaches(TFE_Context* ctx);

// Sets a thread-local device placement policy. After this call, other calls to
// TFE_Execute in the same thread will use the device policy specified here
// instead of the device policy used to construct the context. This has no
// effect on the device policy used by other program threads.

	for(i=0; i<nthread; i++)
		pthread_join(thread_id[i], 0);		
}

static void*
goDummyCallbackThread(void* p)
{
	int i, max;

	max = *(int*)p;
	for(i=0; i<max; i++)
		goDummy();
	return NULL;
}

int
callGoInCThread(int max)
{
	pthread_t thread;

	if (pthread_create(&thread, NULL, goDummyCallbackThread, (void*)(&max)) != 0)
		return -1;

	return 0;
}

void
doAdd(int max, int nthread)
{
	enum { MaxThread = 20 };
	int i;
	uintptr_t thread_id[MaxThread];
	
	if(nthread > MaxThread)
		nthread = MaxThread;
	for(i=0; i<nthread; i++)
		thread_id[i] = _beginthreadex(0, 0, addThread, &max, 0, 0);
	for(i=0; i<nthread; i++) {
		WaitForSingleObject((HANDLE)thread_id[i], INFINITE);
		CloseHandle((HANDLE)thread_id[i]);
	}
}

__stdcall

  // (now inaccessible) corresponding JVPs, but ForwardAccumulator's destructor
  // will release remaining references.
  //
  // This method is not thread-safe (and in general ForwardAccumulator is not
  // thread-safe).
  Status Accumulate(
      const string& op_type, const std::vector<TapeTensor>& input_tensors,
      const std::vector<TapeTensor>& output_tensors,

#include <memory>
#include <string>

#include "absl/base/thread_annotations.h"
#include "absl/synchronization/mutex.h"
#include "tensorflow/c/env.h"
#include "tensorflow/c/tf_status.h"

namespace tf_gcs_filesystem {

/// \brief An LRU cache of string keys and arbitrary values, with configurable
/// max item age (in seconds) and max entries.
///
/// This class is thread safe.
template <typename T>
class ExpiringLRUCache {

// Sets a custom Executor for the current thread. All nodes created by this
// thread will be added to this Executor. It will override the current executor.
TF_CAPI_EXPORT extern void TFE_ContextSetExecutorForThread(TFE_Context*,
                                                           TFE_Executor*);

// Returns the Executor for the current thread.
TF_CAPI_EXPORT extern TFE_Executor* TFE_ContextGetExecutorForThread(

        block_fetcher_(block_fetcher),
        timer_seconds_(timer_seconds),
        pruning_thread_(nullptr,
                        [](TF_Thread* thread) { TF_JoinThread(thread); }) {
    if (max_staleness_ > 0) {
      TF_ThreadOptions thread_options;
      TF_DefaultThreadOptions(&thread_options);
      pruning_thread_.reset(
          TF_StartThread(&thread_options, "TF_prune_FBC", PruneThread, this));
    }