[mir_onnx] MatMul operation support (#6797)

* Added MatMulNodeConverter
* Fix registration and cmake

Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>

diff --git a/compiler/mir-onnx-importer/CMakeLists.txt b/compiler/mir-onnx-importer/CMakeLists.txt
index cce8e00..e319621 100644 (file)
--- a/compiler/mir-onnx-importer/CMakeLists.txt
+++ b/compiler/mir-onnx-importer/CMakeLists.txt
@@ -49,6 +49,8 @@ set(MIR_ONNX_IMPORTER_SOURCES
         Op/Gemm.h
         Op/Identity.cpp
         Op/Identity.h
+        Op/MatMul.cpp
+        Op/MatMul.h
         Op/GlobalAveragePool.cpp
         Op/GlobalAveragePool.h
         Op/Max.cpp

diff --git a/compiler/mir-onnx-importer/ONNXOpRegistration.h b/compiler/mir-onnx-importer/ONNXOpRegistration.h
index 04da1b2..2d458a0 100644 (file)
--- a/compiler/mir-onnx-importer/ONNXOpRegistration.h
+++ b/compiler/mir-onnx-importer/ONNXOpRegistration.h
@@ -31,6 +31,7 @@
 #include "Op/Gemm.h"
 #include "Op/GlobalAveragePool.h"
 #include "Op/Identity.h"
+#include "Op/MatMul.h"
 #include "Op/Max.h"
 #include "Op/MaxPool.h"
 #include "Op/Mul.h"
@@ -66,6 +67,7 @@ inline void registerSupportedOps()
   registry.registerConverter("GlobalAveragePool",
                              stdex::make_unique<GlobalAveragePoolNodeConverter>());
   registry.registerConverter("Identity", stdex::make_unique<IdentityNodeConverter>());
+  registry.registerConverter("MatMul", stdex::make_unique<MatMulNodeConverter>());
   registry.registerConverter("Max", stdex::make_unique<MaxNodeConverter>());
   registry.registerConverter("MaxPool", stdex::make_unique<MaxPoolNodeConverter>());
   registry.registerConverter("Mul", stdex::make_unique<MulNodeConverter>());
@@ -77,7 +79,6 @@ inline void registerSupportedOps()
   registry.registerConverter("Sigmoid", stdex::make_unique<SigmoidNodeConverter>());
   registry.registerConverter("Softmax", stdex::make_unique<SoftmaxNodeConverter>());
   registry.registerConverter("Sum", stdex::make_unique<SumNodeConverter>());
-  registry.registerConverter("Transpose", stdex::make_unique<TransposeNodeConverter>());
   registry.registerConverter("Unsqueeze", stdex::make_unique<UnsqueezeNodeConverter>());
   registry.registerConverter("Upsample", stdex::make_unique<UpsampleNodeConverter>());
 }

diff --git a/compiler/mir-onnx-importer/Op/MatMul.cpp b/compiler/mir-onnx-importer/Op/MatMul.cpp
new file mode 100644 (file)
index 0000000..b2107bc
--- /dev/null
+++ b/compiler/mir-onnx-importer/Op/MatMul.cpp
@@ -0,0 +1,63 @@
+/*
+ * Copyright (c) 2019 Samsung Electronics Co., Ltd. All Rights Reserved
+ *
+ * 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 "MatMul.h"
+
+#include "ONNXHelpers.h"
+
+#include "mir/ops/FullyConnectedOp.h"
+
+namespace mir_onnx
+{
+
+void MatMulNodeConverter::convert(const onnx::NodeProto &onnx_node, ConverterContext *context) const
+{
+  const auto opset_version = context->getOpsetVersion(onnx_node.domain());
+  if (opset_version >= 9)
+    convertV9(onnx_node, context);
+  else if (opset_version >= 1)
+    convertV1(onnx_node, context);
+  else
+    throw std::runtime_error("Not supported opset version on MatMul operation!");
+}
+
+void MatMulNodeConverter::convertV1(const onnx::NodeProto &onnx_node,
+                                    ConverterContext *context) const
+{
+  std::vector<mir::Operation::Output *> inputs = context->getNodeInputs(onnx_node);
+  mir::Graph *graph = context->getGraph();
+
+  assert(inputs.size() == 2);
+  auto A = inputs[0];
+  auto B = inputs[1];
+  // MatMul multiply N-dimentional matrix
+  // FullyConnected layer multiply only 2-dimentional matrix
+  if (A->getShape().rank() != 2 || B->getShape().rank() != 2)
+    throw std::runtime_error("Supported only 2D matrix multiplying!");
+  // Calculate A * B.
+  auto result = createOp<mir::ops::FullyConnectedOp>(graph, A, B)->getOutput(0);
+
+  context->setNodeOutputs(onnx_node, {result});
+}
+
+void MatMulNodeConverter::convertV9(const onnx::NodeProto &onnx_node,
+                                    ConverterContext *context) const
+{
+  // Other type constraints
+  convertV1(onnx_node, context);
+}
+
+} // namespace mir_onnx

diff --git a/compiler/mir-onnx-importer/Op/MatMul.h b/compiler/mir-onnx-importer/Op/MatMul.h
new file mode 100644 (file)
index 0000000..f418da8
--- /dev/null
+++ b/compiler/mir-onnx-importer/Op/MatMul.h
@@ -0,0 +1,37 @@
+/*
+ * Copyright (c) 2019 Samsung Electronics Co., Ltd. All Rights Reserved
+ *
+ * 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.
+ */
+
+#ifndef MIR_ONNX_OP_MATMUL_H
+#define MIR_ONNX_OP_MATMUL_H
+
+#include "ONNXNodeConverterRegistry.h"
+
+namespace mir_onnx
+{
+
+class MatMulNodeConverter : public NodeConverter
+{
+public:
+  void convert(const onnx::NodeProto &onnx_node, ConverterContext *context) const override;
+
+private:
+  void convertV1(const onnx::NodeProto &onnx_node, ConverterContext *context) const;
+  void convertV9(const onnx::NodeProto &onnx_node, ConverterContext *context) const;
+};
+
+} // namespace mir_onnx
+
+#endif // MIR_ONNX_OP_MATMUL_H