*   `tf.data`

    * `tf.data` now has an `autotune_options.initial_parallelism` option to
      control the initial parallelism setting used by autotune before the data
      pipeline has started running. The default is 16. A lower value reduces
      initial memory usage, while a higher value improves startup time.

## Keras

*  `keras.layers.experimental.DynamicEmbedding`

   char *s = s1;
 
   /* Find the end of S1.  */
-  do
-    c = *s1++;
-  while (c != '\0');
-
-  /* Make S1 point before next character, so we can increment
-     it while memory is read (wins on pipelined cpus).  */
-  s1 -= 2;
+  s1 += strlen (s1);
 
-  if (n >= 4)
-    {
-      size_t n4 = n >> 2;
-      do
-	{
-	  c = *s2++;
-	  *++s1 = c;
-	  if (c == '\0')
-	    return s;