Merge branch 'master' into louisfeng/NGMX-296-conv_bias

doc/sphinx/source/installation.rst

@@ -79,10 +79,11 @@ information about how to change or customize this location.
 
       $ cd build && cmake ../ [-DNGRAPH_USE_PREBUILT_LLVM=TRUE]
 
-#. (Optional) Run ``$ make [-jN]`` where ``-jN`` specifies the number of 
-   cores. The example here uses a configuration of ``j8``, which is 
-   good for a system install using an Intel® Xeon® (CPU processor). This step 
-   is **not recommended** with Docker / VM installs. 
+#. (Optional) Run ``$ make [-jN]`` where ``-jN`` specifies the number of physical
+   cores to use to build. The example here uses a configuration of ``j8``, 
+   which is good for a system install using an 8-core Intel® Xeon® CPU processor. 
+   This step is **not recommended** for machines with too little RAM available, 
+   such as those whose RAM is superceded by Docker or VM tasks.  
 
    .. code-block:: console
       

src/ngraph/CMakeLists.txt

@@ -191,6 +191,7 @@ if (NGRAPH_CPU_ENABLE AND LLVM_INCLUDE_DIR AND
         runtime/cpu/pass/cpu_assignment.cpp
         runtime/cpu/pass/cpu_fusion.cpp
         runtime/cpu/pass/cpu_layout.cpp
+        runtime/cpu/pass/cpu_nop_elimination.cpp
     )
     # LLVM binary builds are typically built without RTTI
     # The built-in headers are in a version-specific directory

src/ngraph/runtime/cpu/cpu_external_function.cpp

@@ -114,6 +114,7 @@
 #include "ngraph/runtime/cpu/pass/cpu_assignment.hpp"
 #include "ngraph/runtime/cpu/pass/cpu_fusion.hpp"
 #include "ngraph/runtime/cpu/pass/cpu_layout.hpp"
+#include "ngraph/runtime/cpu/pass/cpu_nop_elimination.hpp"
 
 #ifdef NGRAPH_DISTRIBUTED
 #include "ngraph/ops/allreduce.hpp"
@@ -275,6 +276,7 @@ void runtime::cpu::CPU_ExternalFunction::compile()
 
     ngraph::pass::Manager pass_manager;
 
+    pass_manager.register_pass<runtime::cpu::pass::CPUNopElimination>();
     pass_manager.register_pass<ngraph::pass::CoreFusion>();
     pass_manager.register_pass<runtime::cpu::pass::CPUFusion>();
     pass_manager.register_pass<runtime::cpu::pass::CPUAssignment>(this);

src/ngraph/runtime/cpu/pass/cpu_nop_elimination.cpp

+/*******************************************************************************
+* Copyright 2018 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.
+*******************************************************************************/
+
+#include <memory>
+#include <typeindex>
+#include <typeinfo>
+#include <unordered_map>
+
+#include "cpu_nop_elimination.hpp"
+#include "ngraph/ops/pad.hpp"
+
+#define TI(x) std::type_index(typeid(x))
+
+#define HANDLER_DECL(x)                                                                            \
+    static bool x(const std::shared_ptr<ngraph::Function>& function,                               \
+                  const std::shared_ptr<ngraph::Node>& node)
+
+HANDLER_DECL(eliminate_pad)
+{
+    auto pad = std::dynamic_pointer_cast<ngraph::op::Pad>(node);
+    if (pad->get_input_shape(0) == pad->get_output_shape(0))
+    {
+        function->replace_node(node, node->get_input_op(0));
+        return true;
+    }
+    return false;
+}
+
+static const std::unordered_map<std::type_index,
+                                std::function<bool(const std::shared_ptr<ngraph::Function>&,
+                                                   const std::shared_ptr<ngraph::Node>&)>>
+    dispatcher{{TI(ngraph::op::Pad), &eliminate_pad}};
+
+bool ngraph::runtime::cpu::pass::CPUNopElimination::run_on_function(
+    std::shared_ptr<ngraph::Function> function)
+{
+    bool clobbered = false;
+
+    for (const auto& n : function->get_ops())
+    {
+        // Work around a warning [-Wpotentially-evaluated-expression]
+        const Node& node = *n;
+        auto handler = dispatcher.find(TI(node));
+        if (handler != dispatcher.end())
+        {
+            clobbered = handler->second(function, n) || clobbered;
+        }
+    }
+
+    return clobbered;
+}

src/ngraph/runtime/cpu/pass/cpu_nop_elimination.hpp

+/*******************************************************************************
+* Copyright 2018 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.
+*******************************************************************************/
+
+#pragma once
+
+#include "ngraph/pass/pass.hpp"
+
+namespace ngraph
+{
+    namespace runtime
+    {
+        namespace cpu
+        {
+            namespace pass
+            {
+                class CPUNopElimination : public ngraph::pass::FunctionPass
+                {
+                public:
+                    bool run_on_function(std::shared_ptr<ngraph::Function> function) override;
+                };
+            }
+        }
+    }
+}