id: Python
    attributes:
      label: Python version
      placeholder: e.g., 3.9
  - type: input
    id: Bazel
    attributes:
      label: Bazel version
      description: If compiling from source
  - type: input
    id: Compiler
    attributes:
      label: GCC/compiler version
      description: If compiling from source
  - type: input
    id: Cuda
    attributes:
      label: CUDA/cuDNN version
  - type: input

 # endif
 #endif
 
+/* Undefine (also defined in libc-symbols.h).  */
+#undef __attribute_copy__
+#if __GNUC_PREREQ (9, 0)
+/* Copies attributes from the declaration or type referenced by
+   the argument.  */
+# define __attribute_copy__(arg) __attribute__ ((__copy__ (arg)))
+#else
+# define __attribute_copy__(arg)
+#endif
+
 #endif	 /* sys/cdefs.h */
diff --git a/stdlib/setenv.c b/stdlib/setenv.c

      // starting on the next operation.
      finished_join_.wait(l);
    }
    context_ = context;
    operation_name_ = operation_name;
    step_id_ = step_id;
    op_inputs_ = inputs;
    attributes_ = attributes;
    expected_max_outputs_ = expected_max_outputs;
    cancellation_manager_ = &cancellation_manager;
    execution_state_ = ExecutionState::kReadyToExecute;
  }
  start_execute_.notify_one();
}

  // its corresponding inputs from the input ParallelTensors. Wraps the
  // resulting per-device and per-output TFE_TensorHandles into one
  // ParallelTensor per output of the original operation.
  //
  // Attributes are forwarded to executed operations unmodified.
  //
  // The returned optional has a value if and only if `status` evaluates to
  // TF_OK. Bad statuses are forwarded from underlying `TFE_Execute` calls, or

    const ParallelDevice& parallel_device,
    const std::string& parallel_device_name, TFE_Context* context,
    std::vector<MaybeParallelTensorUnowned> inputs, const char* operation_name,
    const TFE_OpAttrs* attributes, int expected_max_outputs,
    TF_Status* status) {
  absl::optional<std::vector<MaybeParallelTensorOwned>> result;
  // TODO(allenl): We should remove "TPU" from these op names at the very least,

  const char* operation_name = TFE_OpGetName(original_op, s);
  if (TF_GetCode(s) != TF_OK) return;
  const TFE_OpAttrs* attributes = TFE_OpGetAttrs(original_op);

  TFE_Op* op(TFE_NewOp(context, operation_name, s));
  if (TF_GetCode(s) != TF_OK) return;
  TFE_OpAddAttrs(op, attributes);
  TFE_OpSetDevice(op, dev->underlying_device.c_str(), s);
  if (TF_GetCode(s) != TF_OK) return;

// Return an allreduce op adding up input tensor `in` from `group_size` workers.
TFE_Op* AllReduceOp(TFE_Context* ctx, TFE_TensorHandle* in, int group_size);

// Return a SendOp op `op_name` with send input tensor `in` and attributes
// `send_device`, `recv_device`, and `send_device_incarnation` set.
TFE_Op* SendOp(TFE_Context* ctx, TFE_TensorHandle* in,
               const std::string& op_name, const std::string& send_device,

    compiling from source placeholder: validations: required: false

-   type: input id: Compiler attributes: label: GCC/Compiler version
    description: if compiling from source placeholder: validations: required:
    false

-   type: input id: Cuda attributes: label: CUDA/cuDNN version description:
    placeholder: validations: required: false

  // destroy an instance of this class.
  virtual void Release() = 0;

  // Creates an operation builder and ties it to this context.
  // The returned object can be used for setting operation's attributes,
  // adding inputs and finally executing (immediately or lazily as in tracing)
  // it in this context.
  virtual AbstractOperation* CreateOperation() = 0;

  op->node.AddAttr(attr_name, attr_val);
  RecordMutation(graph, *op, "setting attribute");
}

void TF_ClearAttr(TF_Graph* graph, TF_Operation* op, const char* attr_name,
                  TF_Status* status) {
  using tensorflow::RecordMutation;
  mutex_lock l(graph->mu);
  op->node.ClearAttr(attr_name);
  RecordMutation(graph, *op, "clearing attribute");
}

void TF_SetFullType(TF_Graph* graph, TF_Operation* op,