Add tf.BenchmarkConfig that returns a session config appropriate for benchmarking. At the moment, it returns a default config with only Grappler dependency optimizer disabled. Many benchmarks wrap the subgraph they want to time in control_flow_ops.group() to avoid including the overhead of copying the output back to the Python client in the measurement. In the graph, this only adds a control dependency between the subgraph output and the fetch node, which in turn (often) causes the dependency optimizer to turn all nodes in the graph into no-ops.

PiperOrigin-RevId: 216242463

1 files changed, 14 insertions, 0 deletions

diff --git a/tensorflow/python/platform/benchmark.py b/tensorflow/python/platform/benchmark.py
index fa17b17d10..4f7abb311a 100644
--- a/tensorflow/python/platform/benchmark.py
+++ b/tensorflow/python/platform/benchmark.py
@@ -27,6 +27,7 @@ import time
 import six
 
 from tensorflow.core.protobuf import config_pb2
+from tensorflow.core.protobuf import rewriter_config_pb2
 from tensorflow.core.util import test_log_pb2
 from tensorflow.python.client import timeline
 from tensorflow.python.platform import app
@@ -182,6 +183,19 @@ class Benchmark(six.with_metaclass(_BenchmarkRegistrar, object)):
         throughput=throughput, extras=extras)
 
 
+@tf_export("test.benchmark_config")
+def benchmark_config():
+  """Returns a tf.ConfigProto for disabling the dependency optimizer.
+
+    Returns:
+      A TensorFlow ConfigProto object.
+  """
+  config = config_pb2.ConfigProto()
+  config.graph_options.rewrite_options.dependency_optimization = (
+      rewriter_config_pb2.RewriterConfig.OFF)
+  return config
+
+
 @tf_export("test.Benchmark")
 class TensorFlowBenchmark(Benchmark):
   """Abstract class that provides helpers for TensorFlow benchmarks."""