Merge pull request #24309 from dkurt:gemm_ov_hotfix

Update OpenVINO init of new GEMM layer #24309

### Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

CI validation:

- [x] 2022.1.0: https://pullrequest.opencv.org/buildbot/builders/precommit_custom_linux/builds/100368
- [ ] 2021.4.2: https://pullrequest.opencv.org/buildbot/builders/precommit_custom_linux/builds/100373

Checklist:
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
      Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake

modules/dnn/perf/perf_gemm.cpp

@@ -160,8 +160,7 @@ PERF_TEST_P_(Gemm, gemm)
     }
 
     Net net;
-    int id = net.addLayerToPrev(lp.name, lp.type, lp);
-    net.connect(0, 0, id, 0);
+    net.addLayerToPrev(lp.name, lp.type, lp);
     net.setPreferableBackend(backend_id);
     net.setPreferableTarget(target_id);
 
@@ -221,14 +220,13 @@ PERF_TEST_P_(Gemm, innerproduct)
     }
 
     Net net;
-    int id = net.addLayerToPrev(lp.name, lp.type, lp);
-    net.connect(0, 0, id, 0);
+    net.addLayerToPrev(lp.name, lp.type, lp);
     net.setPreferableBackend(backend_id);
     net.setPreferableTarget(target_id);
 
     // warmup
     {
-        std::vector<std::string> input_names(2);
+        std::vector<std::string> input_names(1);
         input_names[0] = "A";
         net.setInputsNames(input_names);
         net.setInput(A, input_names[0]);

modules/dnn/src/layers/gemm_layer.cpp

@@ -287,33 +287,45 @@ public:
     virtual Ptr<BackendNode> initNgraph(const std::vector<Ptr<BackendWrapper> >& inputs,
                                         const std::vector<Ptr<BackendNode> >& nodes) CV_OVERRIDE
     {
-        auto& ieInpNode = nodes[0].dynamicCast<InfEngineNgraphNode>()->node;
+        auto ieInpNode = nodes[0].dynamicCast<InfEngineNgraphNode>()->node;
         std::shared_ptr<ngraph::Node> matmul;
-        int axis = -2;
 
         if (nodes.size() == 2)
         {
             auto& inp2 = nodes[1].dynamicCast<InfEngineNgraphNode>()->node;
-            matmul = std::make_shared<ngraph::op::MatMul>(ieInpNode, inp2, transA, transB);
+            matmul = std::make_shared<ngraph::op::MatMul>(ieInpNode, inp2, trans_a, trans_b);
         }
         else
         {
-            std::vector<int> shape(1 + normalize_axis(axis, ieInpNode->get_shape().size()), 0);
+            std::shared_ptr<ngraph::Node> ieWeights = std::make_shared<ngraph::op::Constant>(ngraph::element::f32, getShape(blobs[0]), blobs[0].data);
+
+            int flatten_axis = ieInpNode.get_shape().size() - ieWeights->get_shape().size();
+            if (flatten_axis > 0) {
+                std::vector<int> shape(1 + flatten_axis, 0);
                 shape[shape.size() - 1] = -1;
-            auto inp = std::make_shared<ngraph::op::v1::Reshape>(
+                ieInpNode = std::make_shared<ngraph::op::v1::Reshape>(
                     ieInpNode,
                     std::make_shared<ngraph::op::Constant>(ngraph::element::i32, ngraph::Shape{shape.size()}, shape.data()),
                     true
                 );
-
-            std::vector<size_t> weight_shape{(size_t)blobs[0].size[0], (size_t)blobs[0].size[1]};
-            auto ieWeights = std::make_shared<ngraph::op::Constant>(ngraph::element::f32, weight_shape, blobs[0].data);
-            matmul = std::make_shared<ngraph::op::MatMul>(inp, ieWeights, transA, transB);
+            }
+            matmul = std::make_shared<ngraph::op::MatMul>(ieInpNode, ieWeights, trans_a, trans_b);
+        }
+        if (alpha != 1.0f) {
+            matmul = std::make_shared<ngraph::op::v1::Multiply>(matmul,
+                std::make_shared<ngraph::op::Constant>(ngraph::element::f32, ngraph::Shape{1}, &alpha)
+            );
         }
 
         if (have_bias && const_C) {
-            auto bias_node = std::make_shared<ngraph::op::Constant>(ngraph::element::f32,
-                                              ngraph::Shape{(size_t)blobs.back().size[1]}, blobs.back().data);
+            Mat bias = blobs.back();
+            auto shape = bias.total() == bias.size[0] ? ngraph::Shape{bias.total()} : getShape(bias);
+            std::shared_ptr<ngraph::Node> bias_node = std::make_shared<ngraph::op::Constant>(ngraph::element::f32, shape, bias.data);
+            if (beta != 1.0f) {
+                bias_node = std::make_shared<ngraph::op::v1::Multiply>(bias_node,
+                    std::make_shared<ngraph::op::Constant>(ngraph::element::f32, ngraph::Shape{1}, &beta)
+                );
+            }
             matmul = std::make_shared<ngraph::op::v1::Add>(matmul, bias_node, ngraph::op::AutoBroadcastType::NUMPY);
         }
         return Ptr<BackendNode>(new InfEngineNgraphNode(matmul));