[XLA:CPU] Make EmitTargetAddressForOp return void (well, technically Status).

This is a general cleanup -- less repeated code -- but it's also part of
an effort to use IrArray more and llvm::Value less.  In particular, many
callsites would take the llvm::Value returned by EmitTargetAddressForOp
and create an IrArray out of it, but then never attach AA info to that
array.  Having this function return void forces you to call
GetIrArrayForOp(), which attaches the AA metadata appropriately.

This change also gets rid of an unused arg to EmitTargetAddressForOp.

PiperOrigin-RevId: 171320201

1 files changed, 3 insertions, 1 deletions

diff --git a/tensorflow/contrib/all_reduce/python/all_reduce.py b/tensorflow/contrib/all_reduce/python/all_reduce.py
index 8e7f1791b8..22d7633ce2 100644
--- a/tensorflow/contrib/all_reduce/python/all_reduce.py
+++ b/tensorflow/contrib/all_reduce/python/all_reduce.py
@@ -762,6 +762,8 @@ def _reduce_non_singleton(input_tensors, red_f, un_op):
   if len(input_tensors) > 1:
     return red_f(input_tensors)
   else:
+    if not un_op:
+      return input_tensors
     output_tensors = []
     for t in input_tensors:
       with ops.colocate_with(t):
@@ -835,7 +837,7 @@ def _build_shuffle_hybrid(input_tensors, gather_devices, red_op, upper_level_f):
 
 
 def build_shuffle_then_ring(input_tensors, gather_devices, subdiv,
-                            red_n_op, red_op, un_op):
+                            red_n_op, red_op, un_op=None):
   """Construct hybrid of Shuffle within workers, Ring across workers."""
   def upper_builder(tensors):
     return build_ring_all_reduce(tensors, len(tensors), subdiv, [0],