tokenize only single groupNorm to subgraph

src/common/snippets/include/snippets/pass/group_norm_tokenization.hpp

@@ -0,0 +1,27 @@
+// Copyright (C) 2023 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#pragma once
+
+#include "openvino/pass/graph_rewrite.hpp"
+#include "openvino/pass/pattern/matcher.hpp"
+
+namespace ov {
+namespace snippets {
+namespace pass {
+
+/**
+ * @interface TokenizeGroupNormSnippets
+ * @brief Tokenize GroupNormalization to a subgraph
+ * @ingroup snippets
+ */
+class TokenizeGroupNormSnippets: public ov::pass::MatcherPass {
+public:
+    OPENVINO_RTTI("TokenizeGroupNormSnippets", "0");
+    TokenizeGroupNormSnippets();
+};
+
+}  // namespace pass
+}  // namespace snippets
+}  // namespace ov

src/common/snippets/src/op/subgraph.cpp

@@ -17,6 +17,7 @@
 #include "snippets/pass/set_softmax_ports.hpp"
 #include "snippets/pass/canonicalization.hpp"
 #include "snippets/pass/align_element_types.hpp"
+#include "snippets/pass/group_normalization_decomposition.hpp"
 #include "snippets/lowered/pass/validate_shapes.hpp"
 
 #include "snippets/utils.hpp"
@@ -75,7 +76,8 @@ auto Subgraph::is_domain_sensitive_op(const std::shared_ptr<ov::Node>& op) -> bo
            ov::is_type<ov::op::v8::Softmax>(op) ||
            ov::is_type<ov::op::v0::MatMul>(op) ||
            ov::is_type<ov::op::v1::Broadcast>(op) || // Broadcast is domain sensetive op because the output shape depends on
-           ov::is_type<ov::op::v3::Broadcast>(op);   // the both input and broadcast shapes (the both - are inputs of op). Note: is used only in MHA pattern
+           ov::is_type<ov::op::v3::Broadcast>(op) ||   // the both input and broadcast shapes (the both - are inputs of op). Note: is used only in MHA pattern
+           ov::is_type<ov::op::v12::GroupNormalization>(op);
 }
 
 void Subgraph::init_config() {
@@ -395,6 +397,7 @@ void Subgraph::data_flow_transformations(const BlockedShapeVector& blocked_input
         manager.register_pass<snippets::pass::FuseTransposeBrgemm>();
         manager.register_pass<snippets::pass::TransposeDecomposition>();
         manager.register_pass<snippets::pass::SetSoftmaxPorts>();
+        manager.register_pass<snippets::pass::GroupNormalizationDecomposition>();
     }
     manager.register_pass<snippets::pass::BroadcastToMoveBroadcast>();
     manager.register_pass<snippets::pass::ConvertConstantsToScalars>();

src/common/snippets/src/pass/group_norm_tokenization.cpp

@@ -0,0 +1,34 @@
+// Copyright (C) 2023 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#include "snippets/pass/group_norm_tokenization.hpp"
+
+#include "snippets/itt.hpp"
+#include "snippets/op/subgraph.hpp"
+#include "snippets/utils.hpp"
+
+#include "openvino/core/rt_info.hpp"
+#include "openvino/pass/pattern/op/wrap_type.hpp"
+
+ov::snippets::pass::TokenizeGroupNormSnippets::TokenizeGroupNormSnippets() {
+    MATCHER_SCOPE(TokenizeGroupNormSnippets);
+
+    auto group_norm_pattern = ov::pass::pattern::wrap_type<ov::op::v12::GroupNormalization>();
+
+    ov::matcher_pass_callback callback = [=](ov::pass::pattern::Matcher& m) {
+        OV_ITT_SCOPED_TASK(ov::pass::itt::domains::SnippetsTransform, "Snippets::pass::TokenizeGroupNormSnippets")
+        auto group_norm_node = ov::as_type_ptr<ov::op::v12::GroupNormalization>(m.get_match_root());
+
+        auto subgraph = op::Subgraph::wrap_node_as_subgraph(group_norm_node);
+        subgraph->get_rt_info()["originalLayersNames"] = group_norm_node->get_friendly_name();
+        ov::replace_node(group_norm_node, subgraph);
+        op::update_out_tensor_name(subgraph);
+
+        return true;
+    };
+    auto m = std::make_shared<ov::pass::pattern::Matcher>(group_norm_pattern, matcher_name);
+    register_matcher(m, callback);
+}
+
+

src/common/snippets/src/pass/group_normalization_decomposition.cpp

@@ -9,7 +9,6 @@
 #include "snippets/itt.hpp"
 #include "snippets/lowered/port_descriptor.hpp"
 #include "snippets/snippets_isa.hpp"
-#include "transformations/utils/utils.hpp"
 #include "openvino/core/rt_info.hpp"
 
 namespace ov {
@@ -32,7 +31,7 @@ GroupNormalizationDecomposition::GroupNormalizationDecomposition() {
         const auto bias = group_norm_node->input_value(2);
 
         const auto num_groups = static_cast<size_t>(group_norm_node->get_num_groups());
-        const auto eps = ov::op::util::cast_eps_to_float(group_norm_node->get_epsilon());
+        const float eps = static_cast<float>(group_norm_node->get_epsilon());
 
         // reshape [N, C, spatial] to [N, group, C / group, spatial]
         const auto orig_shape = group_norm_node->get_input_shape(0);

src/common/snippets/src/pass/tokenization.cpp

@@ -9,6 +9,7 @@
 #include "snippets/pass/common_optimizations.hpp"
 #include "snippets/pass/extract_reshapes_from_mha.hpp"
 #include "snippets/pass/mha_tokenization.hpp"
+#include "snippets/pass/group_norm_tokenization.hpp"
 #include "snippets/pass/collapse_subgraph.hpp"
 
 
@@ -82,6 +83,7 @@ bool SnippetsTokenization::run_on_model(const std::shared_ptr<ov::Model>& m) {
     manager.register_pass<ExtractReshapesFromMHA>();
     manager.register_pass<TokenizeMHASnippets>(m_config);
     manager.register_pass<TokenizeSnippets>();
+    manager.register_pass<TokenizeGroupNormSnippets>();
     manager.register_pass<CommonOptimizations>(m_config);
     manager.run_passes(m);