Sarkars/upgrade r19 (#250)

ngraph_bridge/ngraph_builder.cc

@@ -29,6 +29,7 @@
 #include "ngraph/op/util/logical_reduction.hpp"
 
 #include "logging/ngraph_log.h"
+#include "ngraph_bridge/ngraph_api.h"
 #include "ngraph_bridge/ngraph_backend_manager.h"
 #include "ngraph_bridge/ngraph_builder.h"
 #include "ngraph_bridge/ngraph_conversions.h"
@@ -97,6 +98,9 @@ std::shared_ptr<TOpType> ConstructNgNode(const std::string& op_name,
   auto ng_node = std::make_shared<TOpType>(std::forward<TArg>(Args)...);
   ng_node->set_friendly_name(op_name);
   ng_node->add_provenance_tag(op_name);
+  if (config::IsLoggingPlacement()) {
+    cout << "TF_to_NG: " << op_name << " --> " << ng_node->get_name() << "\n";
+  }
   return ng_node;
 }
 

ngraph_bridge/ngraph_encapsulate_clusters.cc

@@ -763,10 +763,6 @@ Status EncapsulateClusters(
           TF_RETURN_IF_ERROR(
               ConvertGraphDefToGraph(opts, *gdef_for_current_encapsulate,
                                      &graph_for_current_encapsulate));
-          TF_RETURN_IF_ERROR(Builder::TranslateGraph(
-              input_shapes, static_input_map, &graph_for_current_encapsulate,
-              ng_function));
-          string serialized_ngfunc(ngraph::serialize(ng_function, 4));
 
           // get backend.
           // TODO: Note that this is code duplication of some stuff present
@@ -793,6 +789,14 @@ Status EncapsulateClusters(
           }
           TF_RETURN_IF_ERROR(BackendManager::CreateBackend(
               op_backend_name));  // Created a backend here. must free it
+          // TranslateGraph must be called AFTER CreateBackend because some TF
+          // ops like CNMS and gather use backend specific nodes
+          TF_RETURN_IF_ERROR(Builder::TranslateGraph(
+              input_shapes, static_input_map, &graph_for_current_encapsulate,
+              ng_function));
+          int json_indentation = 4;
+          string serialized_ngfunc(
+              ngraph::serialize(ng_function, json_indentation));
           std::unordered_map<std::string, std::string> additional_attribute_map;
           for (auto itr : node->attrs()) {
             // Find the optional attributes to be sent to the backend.
@@ -804,7 +808,7 @@ Status EncapsulateClusters(
             // For e.g. _ngraph_ice_cores --> ice_cores
             if (itr.first.find("_ngraph_") != std::string::npos) {
               // leave out _ngraph_aot_requested
-              if (itr.first.find("_ngraph_aot_requested") !=
+              if (itr.first.find("_ngraph_aot_requested") ==
                   std::string::npos) {
                 additional_attribute_map.insert(
                     {itr.first.substr(strlen("_ngraph_")), itr.second.s()});

ngraph_bridge/ngraph_encapsulate_impl.cc

@@ -134,32 +134,32 @@ Status NGraphEncapsulateImpl::GetNgExecutable(
     MemoryProfile(vm0, rss0);
 
     NGRAPH_VLOG(1) << "Compilation cache miss: " << m_name;
+    string serialized_ng_func;
     if (!m_do_aot) {
       TF_RETURN_IF_ERROR(Builder::TranslateGraph(input_shapes, static_input_map,
                                                  &m_graph, ng_function));
       ng_function->set_friendly_name(m_name);
+      int json_indentation = 4;
+      serialized_ng_func = ngraph::serialize(ng_function, json_indentation);
     } else {
       auto itr = m_aot_functions.find(signature);
       if (itr == m_aot_functions.end()) {
         return errors::Internal(
             "Expected to find AOT precompiled ng function of signature: ",
             signature);
       }
-      ng_function = ng::deserialize(itr->second);
+      serialized_ng_func = itr->second;
     }
 
-    auto function_size = ng_function->get_graph_size() / 1024;  // kb unit
-
     // Serialize to nGraph if needed
     if (std::getenv("NGRAPH_ENABLE_SERIALIZE") != nullptr) {
       std::string file_name = "tf_function_" + m_name + ".json";
-      NgraphSerialize("tf_function_" + m_name + ".json", ng_function);
+      StringToFile("tf_function_" + m_name + ".json", serialized_ng_func);
 #if defined NGRAPH_DISTRIBUTED
       int rank_id;
       rank_id = ng::get_distributed_interface()->get_rank();
-      NgraphSerialize(
-          "tf_function_" + m_name + "_" + to_string(rank_id) + ".json",
-          ng_function);
+      StringToFile("tf_function_" + m_name + "_" + to_string(rank_id) + ".json",
+                   serialized_ng_func);
 #endif
     }
     // Evict the cache if the number of elements exceeds the limit
@@ -172,7 +172,7 @@ Status NGraphEncapsulateImpl::GetNgExecutable(
       int input_tensors_bytes_free = 0;
       evicted_ng_exec = m_ng_exec_map[m_lru.back()];
       m_ng_exec_map.erase(m_lru.back());
-      m_ng_function_map.erase(evicted_ng_exec);
+      m_serialized_ng_function_map.erase(evicted_ng_exec);
 
       // Call delete function here for the erased func
       op_backend->remove_compiled_function(evicted_ng_exec);
@@ -222,12 +222,12 @@ Status NGraphEncapsulateImpl::GetNgExecutable(
       }
     } catch (const std::exception& exp) {
       BackendManager::UnlockBackend(m_op_backend_name);
-      NgraphSerialize("tf_function_error_" + m_name + ".json", ng_function);
+      StringToFile("tf_function_error_" + m_name + ".json", serialized_ng_func);
       return errors::Internal("Caught exception while compiling op_backend: ",
                               exp.what(), "\n");
     } catch (...) {
       BackendManager::UnlockBackend(m_op_backend_name);
-      NgraphSerialize("tf_function_error_" + m_name + ".json", ng_function);
+      StringToFile("tf_function_error_" + m_name + ".json", serialized_ng_func);
       return errors::Internal("Error in compiling op_backend\n");
     }
     BackendManager::UnlockBackend(m_op_backend_name);
@@ -236,7 +236,7 @@ Status NGraphEncapsulateImpl::GetNgExecutable(
 
     SetNgExecMap(signature, ng_exec);
     // caching ng_function to serialize to ngraph if needed
-    SetNgFunctionMap(ng_exec, ng_function);
+    m_serialized_ng_function_map[ng_exec] = serialized_ng_func;
 
     m_lru.push_front(signature);
     // Memory after
@@ -245,9 +245,8 @@ Status NGraphEncapsulateImpl::GetNgExecutable(
     auto delta_res_mem = rss - rss0;
     NGRAPH_VLOG(1) << "NGRAPH_TF_CACHE_PROFILE: OP_ID: " << my_instance_id
                    << " Cache length: " << m_ng_exec_map.size()
-                   << "  Cluster: " << m_name << " Delta VM: " << delta_vm_mem
-                   << "  Delta RSS: " << delta_res_mem
-                   << "  Function size: " << function_size
+                   << " Cluster: " << m_name << " Delta VM: " << delta_vm_mem
+                   << " Delta RSS: " << delta_res_mem
                    << " KB Total RSS: " << rss / (1024 * 1024) << " GB "
                    << " VM: " << vm / (1024 * 1024) << " GB" << endl;
   }  // end of input signature not found in m_ng_exec_map
@@ -582,6 +581,12 @@ NGraphEncapsulateImpl::GetTensorsFromPipeline(
   return out_tpl;
 }
 
+void NGraphEncapsulateImpl::DumpNgFunction(
+    const string& file_name,
+    std::shared_ptr<ngraph::runtime::Executable> ng_exec) {
+  StringToFile(file_name, m_serialized_ng_function_map[ng_exec]);
+}
+
 }  // namespace ngraph_bridge
 
 }  // namespace tensorflow

ngraph_bridge/ngraph_encapsulate_impl.h

@@ -85,6 +85,9 @@ class NGraphEncapsulateImpl {
       const ng::element::Type& ng_element_type, const ng::Shape& ng_shape,
       std::shared_ptr<ng::runtime::Tensor> tensor_from_pipeline);
 
+  void DumpNgFunction(const string&,
+                      std::shared_ptr<ngraph::runtime::Executable>);
+
   // Accessors(getters and setters) for the private data members of
   // NgraphEncapsulateImpl class
   // needed by
@@ -148,19 +151,6 @@ class NGraphEncapsulateImpl {
 
   void ClearNgExecMap() { m_ng_exec_map.clear(); }
 
-  std::unordered_map<std::shared_ptr<ngraph::runtime::Executable>,
-                     std::shared_ptr<ngraph::Function>>
-  GetNgFunctionMap() {
-    return m_ng_function_map;
-  }
-
-  void SetNgFunctionMap(
-      const std::shared_ptr<ngraph::runtime::Executable>& exec,
-      const std::shared_ptr<ngraph::Function>& function) {
-    m_ng_function_map[exec] = function;
-  }
-
-  void ClearNgFunctionMap() { m_ng_function_map.clear(); }
   // TODO:sindhu have another get function for output_cache which is only
   // readable
   std::vector<std::pair<void*, shared_ptr<ng::runtime::Tensor>>>&
@@ -179,6 +169,10 @@ class NGraphEncapsulateImpl {
 
   void ClearNgExecOutputCache() { m_ng_exec_output_cache_map.clear(); }
 
+  void ClearNgExecSerializedFunctionCache() {
+    m_serialized_ng_function_map.clear();
+  }
+
   NGraphFreshnessTracker* GetNgraphFreshnessTracker() {
     return m_freshness_tracker;
   }
@@ -236,9 +230,8 @@ class NGraphEncapsulateImpl {
   // ng_function, ng_executable, Output and Input Cache maps
   std::unordered_map<std::string, std::shared_ptr<ngraph::runtime::Executable>>
       m_ng_exec_map;
-  std::unordered_map<std::shared_ptr<ngraph::runtime::Executable>,
-                     std::shared_ptr<ngraph::Function>>
-      m_ng_function_map;
+  std::unordered_map<std::shared_ptr<ngraph::runtime::Executable>, std::string>
+      m_serialized_ng_function_map;
 
   NgFunctionIOCache m_ng_exec_input_cache_map;
   NgFunctionIOCache m_ng_exec_output_cache_map;

ngraph_bridge/ngraph_encapsulate_op.cc

@@ -262,8 +262,8 @@ NGraphEncapsulateOp::~NGraphEncapsulateOp() {
   ng_encap_impl.ClearNgExecInputCache();
   ng_encap_impl.ClearNgExecOutputCache();
   ng_encap_impl.ClearNgExecMap();
-  ng_encap_impl.ClearNgFunctionMap();
   ng_encap_impl.ClearNgExecPipelinedTensorMap();
+  ng_encap_impl.ClearNgExecSerializedFunctionCache();
 
   // Release the backend
   NGRAPH_VLOG(2) << "~NGraphEncapsulateOp():: ReleaseBackend";
@@ -291,7 +291,6 @@ void NGraphEncapsulateOp::Compute(OpKernelContext* ctx) {
 
   std::vector<TensorShape> input_shapes;
   std::vector<const Tensor*> static_input_map;
-  std::shared_ptr<ngraph::Function> ng_function;
   std::shared_ptr<ngraph::runtime::Executable> ng_exec;
   ng::runtime::Backend* op_backend;
 
@@ -526,19 +525,17 @@ void NGraphEncapsulateOp::Compute(OpKernelContext* ctx) {
     try {
       ng_exec->call(ng_outputs, ng_inputs);
     } catch (const std::exception& exp) {
-      ng_function = ng_encap_impl.GetNgFunctionMap()[ng_exec];
       BackendManager::UnlockBackend(ng_encap_impl.GetOpBackend());
-      NgraphSerialize("tf_function_error_" + ctx->op_kernel().name() + ".json",
-                      ng_function);
+      ng_encap_impl.DumpNgFunction(
+          "tf_function_error_" + ctx->op_kernel().name() + ".json", ng_exec);
       OP_REQUIRES(ctx, false,
                   errors::Internal(
                       "Caught exception while executing nGraph computation: ",
                       exp.what(), "\n"));
     } catch (...) {
-      ng_function = ng_encap_impl.GetNgFunctionMap()[ng_exec];
       BackendManager::UnlockBackend(ng_encap_impl.GetOpBackend());
-      NgraphSerialize("tf_function_error_" + ctx->op_kernel().name() + ".json",
-                      ng_function);
+      ng_encap_impl.DumpNgFunction(
+          "tf_function_error_" + ctx->op_kernel().name() + ".json", ng_exec);
       OP_REQUIRES(
           ctx, false,
           errors::Internal("Error in executing the nGraph computation\n"));

ngraph_bridge/ngraph_utils.cc

@@ -312,12 +312,16 @@ Status CheckAxisDimInRange(std::vector<int64> axes, size_t rank) {
 void NgraphSerialize(const std::string& file_name,
                      const std::shared_ptr<ngraph::Function>& ng_function) {
   NGRAPH_VLOG(0) << "Serializing graph to: " << file_name << std::endl;
-  std::string js = ngraph::serialize(ng_function, 4);
+  int json_indentation = 4;
+  StringToFile(file_name, ngraph::serialize(ng_function, json_indentation));
+}
+
+void StringToFile(const std::string& file_name, const std::string& contents) {
   std::ofstream f;
   f.exceptions(std::ofstream::failbit | std::ofstream::badbit);
   try {
     f.open(file_name);
-    f << js;
+    f << contents;
     f.close();
   } catch (std::ofstream::failure& e) {
     NGRAPH_VLOG(0) << "Exception opening/closing file " << file_name

ngraph_bridge/ngraph_utils.h

@@ -300,6 +300,8 @@ Status CheckAxisDimInRange(std::vector<int64> axes, size_t rank);
 void NgraphSerialize(const std::string&,
                      const std::shared_ptr<ngraph::Function>&);
 
+void StringToFile(const std::string&, const std::string&);
+
 // Collect the total memory usage through /proc/self/stat
 void MemoryProfile(long&, long&);
 

test/model_level_tests/models/MLP/getting_repo_ready.sh

@@ -1 +1 @@
-pip install -U keras
+pip install -U keras==2.2.5

test/model_level_tests/models/MLP/repo.txt

@@ -1 +1,2 @@
 https://github.com/keras-team/keras.git
+2.2.5

test/python/test_ngraph_serialize_flag.py

@@ -0,0 +1,62 @@
+# ==============================================================================
+#  Copyright 2019 Intel Corporation
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+# ==============================================================================
+"""Pytest for a simple run on model testing framework
+
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import pytest
+import platform
+import os
+
+import tensorflow as tf
+import numpy as np
+import re
+
+from common import NgraphTest
+import ngraph_bridge
+
+
+class TestNgraphSerialize(NgraphTest):
+
+    def test_ng_serialize_to_json(self):
+        initial_contents = set(os.listdir())
+        xshape = (3, 4, 5)
+        x = tf.placeholder(tf.float32, shape=xshape)
+        out = tf.nn.l2_loss(tf.abs(x))
+        values = np.random.rand(*xshape)
+
+        config = ngraph_bridge.update_config(tf.ConfigProto())
+        ngraph_enable_serialize = os.environ.pop('NGRAPH_ENABLE_SERIALIZE',
+                                                 None)
+        os.environ['NGRAPH_ENABLE_SERIALIZE'] = '1'
+        ngraph_bridge.enable()
+        with tf.Session(config=config) as sess:
+            out = sess.run((out), feed_dict={x: values})
+        os.environ.pop('NGRAPH_ENABLE_SERIALIZE', None)
+        if ngraph_enable_serialize is not None:
+            os.environ['NGRAPH_ENABLE_SERIALIZE'] = \
+                ngraph_enable_serialize
+
+        final_contents = set(os.listdir())
+        assert (len(final_contents) - len(initial_contents) == 1)
+        new_files = final_contents.difference(initial_contents)
+        flname = new_files.pop()
+        assert (flname.startswith('tf_function_') and flname.endswith('json'))
+        os.remove(flname)