CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
    TFE_Execute(matmul, &retvals[0], &num_retvals, status);
    CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
    if (state.iterations() >= state.max_iterations && async) {
      TFE_Executor* executor = TFE_ContextGetExecutorForThread(ctx);
      TFE_ExecutorWaitForAllPendingNodes(executor, status);

     * read operations anyway:
     *
     * - All (unsynchronized) read operations must first read the "count" field, and should not
     * look at table entries if it is 0.
     *
     * - All (synchronized) write operations should write to the "count" field after structurally
     * changing any bin. The operations must not take any action that could even momentarily

   * Returns an iterator over the merged contents of all given {@code iterators}, traversing every
   * element of the input iterators. Equivalent entries will not be de-duplicated.
   *
   * <p>Callers must ensure that the source {@code iterators} are in non-descending order as this
   * method does not sort its input.
   *
   * <p>For any equivalent elements across all {@code iterators}, it is undefined which element is
   * returned first.

a single byte in the string.
</li>

<li>
The iteration order over maps is not specified
and is not guaranteed to be the same from one iteration to the next.
If a map entry that has not yet been reached is removed during iteration,
the corresponding iteration value will not be produced. If a map entry is
created during iteration, that entry may be produced during the iteration or

a single byte in the string.
</li>

<li>
The iteration order over maps is not specified
and is not guaranteed to be the same from one iteration to the next.
If a map entry that has not yet been reached is removed during iteration,
the corresponding iteration value will not be produced. If a map entry is
created during iteration, that entry may be produced during the iteration or

     * read operations anyway:
     *
     * - All (unsynchronized) read operations must first read the "count" field, and should not
     * look at table entries if it is 0.
     *
     * - All (synchronized) write operations should write to the "count" field after structurally
     * changing any bin. The operations must not take any action that could even momentarily

  /**
   * Returns the next element in the iteration, without advancing the iteration.
   *
   * <p>Calls to {@code peek()} should not change the state of the iteration, except that it
   * <i>may</i> prevent removal of the most recent element via {@link #remove()}.
   *
   * @throws NoSuchElementException if the iteration has no more elements according to {@link
   *     #hasNext()}
   */

  /**
   * Returns the next element in the iteration, without advancing the iteration.
   *
   * <p>Calls to {@code peek()} should not change the state of the iteration, except that it
   * <i>may</i> prevent removal of the most recent element via {@link #remove()}.
   *
   * @throws NoSuchElementException if the iteration has no more elements according to {@link
   *     #hasNext()}
   */

 * }</pre>
 *
 * ... the iteration order for {@link #keys()} is {@code [key1, key2, key1]}, and similarly for
 * {@link #entries()}. Unlike {@link LinkedHashMultimap}, the iteration order is kept consistent
 * between keys, entries and values. For example, calling:
 *
 * <pre>{@code
 * multimap.remove(key1, foo);
 * }</pre>
 *
 * <p>changes the entries iteration order to {@code [key2=bar, key1=baz]} and the key iteration

     * read operations anyway:
     *
     * - All (unsynchronized) read operations must first read the "count" field, and should not look
     * at table entries if it is 0.
     *
     * - All (synchronized) write operations should write to the "count" field after structurally
     * changing any bin. The operations must not take any action that could even momentarily cause a