Merge branch 'tfl/send_recv_op' of github.com:NervanaSystems/ngraph into tfl/send_recv_op

src/ngraph/op/fused_op_tbl.hpp

@@ -24,8 +24,8 @@ NGRAPH_OP(ConvolutionBiasBackpropFiltersBias, ngraph::op)
 NGRAPH_OP(DepthToSpace, ngraph::op)
 NGRAPH_OP(Elu, ngraph::op)
 NGRAPH_OP(FakeQuantize, ngraph::op)
-NGRAPH_OP(GRN, ngraph::op)
 NGRAPH_OP(Gemm, ngraph::op)
+NGRAPH_OP(GRN, ngraph::op)
 NGRAPH_OP(GroupConvolution, ngraph::op)
 NGRAPH_OP(GroupConvolutionTranspose, ngraph::op)
 NGRAPH_OP(HardSigmoid, ngraph::op)
@@ -35,9 +35,9 @@ NGRAPH_OP(MVN, ngraph::op)
 NGRAPH_OP(Normalize, ngraph::op)
 NGRAPH_OP(PRelu, ngraph::op)
 NGRAPH_OP(ScaleShift, ngraph::op)
-NGRAPH_OP(SpaceToDepth, ngraph::op)
 NGRAPH_OP(ShuffleChannels, ngraph::op)
+NGRAPH_OP(SpaceToDepth, ngraph::op)
+NGRAPH_OP(Split, ngraph::op)
 NGRAPH_OP(SquaredDifference, ngraph::op)
 NGRAPH_OP(Squeeze, ngraph::op)
-NGRAPH_OP(Split, ngraph::op)
 NGRAPH_OP(Unsqueeze, ngraph::op)

src/ngraph/op/op_tbl.hpp

@@ -81,11 +81,12 @@ NGRAPH_OP(Cos, ngraph::op)
 NGRAPH_OP(Cosh, ngraph::op)
 NGRAPH_OP(Dequantize, ngraph::op)
 NGRAPH_OP(Divide, ngraph::op)
-NGRAPH_OP(DynBroadcast, ngraph::op)
 NGRAPH_OP(Dot, ngraph::op)
+NGRAPH_OP(DynBroadcast, ngraph::op)
 NGRAPH_OP(DynPad, ngraph::op)
 NGRAPH_OP(DynReshape, ngraph::op)
 NGRAPH_OP(DynSlice, ngraph::op)
+NGRAPH_OP(EmbeddingLookup, ngraph::op)
 NGRAPH_OP(Equal, ngraph::op)
 NGRAPH_OP(Erf, ngraph::op)
 NGRAPH_OP(Exp, ngraph::op)
@@ -119,13 +120,13 @@ NGRAPH_OP(Power, ngraph::op)
 NGRAPH_OP(Product, ngraph::op)
 NGRAPH_OP(Quantize, ngraph::op)
 NGRAPH_OP(QuantizedAvgPool, ngraph::op)
+NGRAPH_OP(QuantizedConvolution, ngraph::op)
 NGRAPH_OP(QuantizedConvolutionBias, ngraph::op)
 NGRAPH_OP(QuantizedConvolutionBiasAdd, ngraph::op)
 NGRAPH_OP(QuantizedConvolutionBiasSignedAdd, ngraph::op)
 NGRAPH_OP(QuantizedConvolutionRelu, ngraph::op)
-NGRAPH_OP(QuantizedConvolution, ngraph::op)
-NGRAPH_OP(QuantizedDotBias, ngraph::op)
 NGRAPH_OP(QuantizedDot, ngraph::op)
+NGRAPH_OP(QuantizedDotBias, ngraph::op)
 NGRAPH_OP(QuantizedMaxPool, ngraph::op)
 NGRAPH_OP(Recv, ngraph::op)
 NGRAPH_OP(Range, ngraph::op)
@@ -155,7 +156,6 @@ NGRAPH_OP(Subtract, ngraph::op)
 NGRAPH_OP(Sum, ngraph::op)
 NGRAPH_OP(Tan, ngraph::op)
 NGRAPH_OP(Tanh, ngraph::op)
-NGRAPH_OP(TopK, ngraph::op)
 NGRAPH_OP(Tile, ngraph::op)
+NGRAPH_OP(TopK, ngraph::op)
 NGRAPH_OP(Transpose, ngraph::op)
-NGRAPH_OP(EmbeddingLookup, ngraph::op)

src/ngraph/pass/fused_op_decomposition.cpp

@@ -13,36 +13,51 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //*****************************************************************************
-
 #include "ngraph/pass/fused_op_decomposition.hpp"
-
 #include "ngraph/graph_util.hpp"
 #include "ngraph/op/get_output_element.hpp"
-#include "ngraph/op/util/fused_op.hpp"
 
 using namespace std;
 using namespace ngraph;
 
-bool ngraph::pass::FusedOpDecomposition::run_on_node(std::shared_ptr<ngraph::Node> node)
+pass::FusedOpDecomposition::FusedOpDecomposition(op_query_t callback)
+    : m_has_direct_support{callback}
+{
+}
+
+bool pass::FusedOpDecomposition::run_on_node(shared_ptr<Node> node)
 {
     bool modified = false;
 
-    if (auto fused_op = std::dynamic_pointer_cast<ngraph::op::util::FusedOp>(node))
+    if (auto fused_op = dynamic_pointer_cast<op::util::FusedOp>(node))
     {
-        if (m_callback && m_callback(*node))
+        if (m_has_direct_support && m_has_direct_support(*node))
         {
             // Op supported by backend. Do not decompose
             return modified;
         }
+
         auto subgraph_outputs = fused_op->decompose_op();
+        // Run recursively untill no more fused ops
+        auto subgraph = extract_subgraph(subgraph_outputs, fused_op->get_arguments());
+        for (auto subgraph_node : subgraph)
+        {
+            if (auto nested_fused_op = dynamic_pointer_cast<op::util::FusedOp>(subgraph_node))
+            {
+                if (!(m_has_direct_support && m_has_direct_support(*nested_fused_op)))
+                {
+                    run_on_node(nested_fused_op);
+                }
+            }
+        }
+
         size_t i = 0;
         for (auto output_node : subgraph_outputs)
         {
             for (size_t j = 0; j < output_node->get_outputs().size(); j++, i++)
             {
                 // TODO: Provenance
-                std::set<ngraph::descriptor::Input*> fop_users{
-                    begin(fused_op->get_outputs().at(i).get_inputs()),
+                set<descriptor::Input*> fop_users{begin(fused_op->get_outputs().at(i).get_inputs()),
                                                   end(fused_op->get_outputs().at(i).get_inputs())};
                 for (auto fop_user : fop_users)
                 {
@@ -52,7 +67,7 @@ bool ngraph::pass::FusedOpDecomposition::run_on_node(std::shared_ptr<ngraph::Nod
                         if (goe->get_n() == i && !goe->get_output_inputs(0).empty())
                         {
                             // Replace GOE users
-                            std::set<ngraph::descriptor::Input*> goe_users{
+                            set<descriptor::Input*> goe_users{
                                 begin(goe->get_outputs().at(0).get_inputs()),
                                 end(goe->get_outputs().at(0).get_inputs())};
                             for (auto goe_user : goe_users)
@@ -80,8 +95,3 @@ bool ngraph::pass::FusedOpDecomposition::run_on_node(std::shared_ptr<ngraph::Nod
 
     return modified;
 }
-
-pass::FusedOpDecomposition::FusedOpDecomposition(op_query_t callback)
-    : m_callback{callback}
-{
-}

src/ngraph/pass/fused_op_decomposition.hpp

@@ -16,6 +16,9 @@
 
 #pragma once
 
+#include <memory>
+
+#include "ngraph/op/util/fused_op.hpp"
 #include "ngraph/pass/pass.hpp"
 
 namespace ngraph
@@ -25,13 +28,24 @@ namespace ngraph
         class FusedOpDecomposition : public NodePass
         {
         public:
+            /// \brief  Function signature type for callback used to check whether provided node
+            ///         is supported by backend.
             using op_query_t = std::function<bool(const Node& node)>;
 
+            ///
+            /// \brief      Constructor for the Fused operation decomposition pass.
+            ///
+            /// \param[in]  callback  The function object used to determine whether current backend
+            ///                       provide direct support for passed node. Should have signature:
+            ///                       bool fn(const Node&)
+            ///
             FusedOpDecomposition(op_query_t callback = nullptr);
             bool run_on_node(std::shared_ptr<ngraph::Node> node) override;
 
         private:
-            op_query_t m_callback = nullptr;
+            /// \brief A function returning whether provided Node is supported by current backend.
+            ///        The returned bool value is used to control whether decompose operator or not.
+            op_query_t m_has_direct_support = nullptr;
         };
     }
 }

src/ngraph/runtime/cpu/cpu_external_function.cpp

@@ -1180,7 +1180,6 @@ void runtime::cpu::CPU_ExternalFunction::register_common_passes(
     REGISTER_KNOBBED_PASS(RecurrentReshapeElimination, false, ngraph::pass);
     REGISTER_KNOBBED_PASS_WITH_ARGS(
         CoreFusion, true, ngraph::pass, ngraph::pass::FusionType::ALL_FUSIONS);
-    REGISTER_KNOBBED_PASS_WITH_ARGS(FusedOpDecomposition, true, ngraph::pass, is_supported);
     REGISTER_KNOBBED_PASS(CPUFusion, true, runtime::cpu::pass);
     REGISTER_KNOBBED_PASS(CPUQuantFusion, true, runtime::cpu::pass);
     REGISTER_KNOBBED_PASS(CPUHorizontalFusion, true, runtime::cpu::pass);