IntelGPU backend: Allow more cases for clDNN gemm (#2187)

src/ngraph/runtime/intelgpu/intelgpu_backend.cpp

@@ -685,28 +685,30 @@ bool runtime::intelgpu::IntelGPUBackend::compile(shared_ptr<Function> func)
 
             if (get_input_type(op) == element::f32 && get_input_type(op, 1) == element::f32 &&
                 get_output_type(op) == element::f32 && input0_elem_count && input1_elem_count &&
-                (axes_count == 1) && (input0_shape.size() < 3) && (input1_shape.size() < 3) &&
-                !input0_shape.empty() && !input1_shape.empty())
+                (axes_count < 2) && (input0_shape.size() < 3) && (input1_shape.size() < 3))
             {
-                string input1_name = get_input_name(op, 1);
+                bool transpose0 = false;
+                bool transpose1 = false;
+
+                // If we have A[5] and B[] here, in cldnn we have A[1, 1, 1, 5] and B[1, 1, 1, 1]
+                // it needs to be reshaped into A[1, 1, 5, 1] and B[1, 1, 1, 1]
+                if ((input0_shape.size() == 1) && input1_shape.empty())
+                {
+                    transpose0 = true;
+                }
 
                 // If we have A[5] and B[5] here, in cldnn we have A[1, 1, 1, 5] and B[1, 1, 1, 5]
                 // it needs to be reshaped into A[1, 1, 1, 5] and B[1, 1, 5, 1]
                 if (!input0_shape.empty() && (input1_shape.size() == 1))
                 {
-                    const string new_name = input1_name + "_reshaped";
-                    Shape new_shape = input1_shape;
-                    new_shape.push_back(1);
-                    const cldnn::tensor reshaped_tensor =
-                        intelgpu_space::create_cldnn_tensor(new_shape);
-
-                    const cldnn::reshape reshape_op(new_name, input1_name, reshaped_tensor);
-                    topology.add(reshape_op);
-
-                    input1_name = new_name;
+                    transpose1 = true;
                 }
 
-                const cldnn::gemm dot_op(get_output_name(op), get_input_name(op, 0), input1_name);
+                const cldnn::gemm dot_op(get_output_name(op),
+                                         get_input_name(op, 0),
+                                         get_input_name(op, 1),
+                                         transpose0,
+                                         transpose1);
                 topology.add(dot_op);
             }
             else