[GPU] Fix static_impl != nullptr assertion for reorder node (#27786)

### Details: - Fix ”Assertion `static_impl != nullptr' failed“ issue for reorder (Convolution_0_reorder_1) node by skipping mark node in shape flow. ### Tickets: - [CVS-154336](https://jira.devtools.intel.com/browse/CVS-154336) --------- Signed-off-by: yuan.xiong <[email protected]>
openvinotoolkit · Dec 23, 2024 · d52f1da · d52f1da
1 parent 6e02445
commit d52f1da
Show file tree

Hide file tree

Showing 2 changed files with 113 additions and 0 deletions.
diff --git a/src/plugins/intel_gpu/src/graph/graph_optimizer/mark_shape_of_subgraphs.cpp b/src/plugins/intel_gpu/src/graph/graph_optimizer/mark_shape_of_subgraphs.cpp
@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: Apache-2.0
 //
 
+#include "broadcast_inst.h"
 #include "shape_of_inst.h"
 #include "read_value_inst.h"
 #include "reshape_inst.h"
@@ -86,6 +87,13 @@ bool mark_shape_of_subgraphs::can_mark_node(const program_node& node) {
         return false;
     }
 
+    // skip mark_node for broadcast node if dependency nodes are data and shape_of
+    auto& dependencies = node.get_dependencies();
+    if (node.is_type<broadcast>() && dependencies.size() == 2) {
+        if (dependencies[0].first->is_type<data>() && dependencies[1].first->is_type<shape_of>())
+            return false;
+    }
+
     return true;
 }
 

diff --git a/src/plugins/intel_gpu/tests/unit/passes/mark_shape_of_subgraphs_test.cpp b/src/plugins/intel_gpu/tests/unit/passes/mark_shape_of_subgraphs_test.cpp
@@ -318,3 +318,108 @@ TEST(mark_shape_of_subgraphs, gather_compressed_no_mark) {
     ASSERT_FALSE(check_subgraph(prog->get_node("shape_of"), prog->get_node("gather_compressed")));
     ASSERT_FALSE(check_subgraph(prog->get_node("shape_of"), prog->get_node("concat")));
 }
+
+TEST(mark_shape_of_subgraphs, broadcast_not_existed_after_shapeof) {
+    auto& engine = get_test_engine();
+    auto input_layout_dynamic = layout{ov::PartialShape{ov::Dimension::dynamic(), 4, ov::Dimension::dynamic(), ov::Dimension::dynamic()},
+                                       data_types::f32, format::bfyx};
+    auto data_0 = engine.allocate_memory({ ov::PartialShape{4}, data_types::i32, format::bfyx });
+    set_values(data_0, {1, 4, 1, 1});
+    auto weights = engine.allocate_memory({ data_types::f16, format::bfyx, {1152, 4, 1, 1} });
+
+    topology topology;
+    topology.add(input_layout("input", input_layout_dynamic));
+    topology.add(data("data_0", data_0));
+    topology.add(data("weights", weights));
+    topology.add(shape_of("shape_of", input_info("input"), data_types::i32));
+    topology.add(reshape("reshape", input_info("shape_of"), input_info("data_0"), false, {}));
+    topology.add(convolution("convolution", input_info("reshape"), "weights", "", 1, {1, 1}, {1, 1}, {0, 0}, {0, 0}, false));
+
+    ExecutionConfig config = get_test_default_config(engine);
+    config.set_property(ov::intel_gpu::allow_new_shape_infer(true));
+    config.set_property(ov::intel_gpu::optimize_data(true));
+    network network(engine, topology, config);
+
+    auto prog = network.get_program();
+    ASSERT_NE(prog, nullptr);
+
+    ASSERT_TRUE(check_subgraph(prog->get_node("shape_of"), prog->get_node("convolution")));
+}
+
+TEST(mark_shape_of_subgraphs, broadcast_w_data_and_direct_shapeof_no_mark) {
+    auto& engine = get_test_engine();
+    auto input_layout_dynamic = layout{ov::PartialShape{ov::Dimension::dynamic(), 4, ov::Dimension::dynamic(), ov::Dimension::dynamic()},
+                                       data_types::f32, format::bfyx};
+    auto data_0 = engine.allocate_memory({ ov::PartialShape{1}, data_types::i32, format::bfyx });
+    set_values(data_0, {0});
+    auto weights = engine.allocate_memory({ data_types::f16, format::bfyx, {1152, 4, 2, 2} });
+
+    topology topology;
+    topology.add(input_layout("input", input_layout_dynamic));
+    topology.add(data("data_0", data_0));
+    topology.add(shape_of("shape_of", input_info("input"), data_types::i32));
+    topology.add(broadcast("broadcast", input_info("data_0"), input_info("shape_of"), {}, ov::op::BroadcastType::BIDIRECTIONAL));
+    topology.add(data("weights", weights));
+    topology.add(convolution("convolution", input_info("broadcast"), "weights", "", 1, {1, 1}, {1, 1}, {0, 0}, {0, 0}, false));
+
+    ExecutionConfig config = get_test_default_config(engine);
+    config.set_property(ov::intel_gpu::allow_new_shape_infer(true));
+    config.set_property(ov::intel_gpu::optimize_data(true));
+    network network(engine, topology, config);
+
+    auto prog = network.get_program();
+    ASSERT_NE(prog, nullptr);
+
+    ASSERT_FALSE(check_subgraph(prog->get_node("shape_of"), prog->get_node("convolution")));
+    ASSERT_FALSE(check_subgraph(prog->get_node("shape_of"), prog->get_node("broadcast")));
+}
+
+TEST(mark_shape_of_subgraphs, broadcast_w_data_and_indirect_shapeof) {
+    auto& engine = get_test_engine();
+    auto input_layout_dynamic = layout{ov::PartialShape{ov::Dimension::dynamic(), 4, ov::Dimension::dynamic(), ov::Dimension::dynamic()},
+                                       data_types::f32, format::bfyx};
+    auto data_0 = engine.allocate_memory({ ov::PartialShape{1}, data_types::i32, format::bfyx });
+    set_values(data_0, {0});
+
+    topology topology;
+    topology.add(input_layout("input", input_layout_dynamic));
+    topology.add(data("data_0", data_0));
+    topology.add(shape_of("shape_of", input_info("input"), data_types::i32));
+    topology.add(gather("gather", input_info("shape_of"), input_info("data_0"), 0, 0, {}));
+    topology.add(broadcast("broadcast", input_info("data_0"), input_info("gather"), {}, ov::op::BroadcastType::BIDIRECTIONAL));
+
+    ExecutionConfig config = get_test_default_config(engine);
+    config.set_property(ov::intel_gpu::allow_new_shape_infer(true));
+    config.set_property(ov::intel_gpu::optimize_data(true));
+    network network(engine, topology, config);
+
+    auto prog = network.get_program();
+    ASSERT_NE(prog, nullptr);
+
+    ASSERT_TRUE(check_subgraph(prog->get_node("shape_of"), prog->get_node("broadcast")));
+}
+
+TEST(mark_shape_of_subgraphs, broadcast_w_direct_shapeof_and_data) {
+    auto& engine = get_test_engine();
+    auto input_layout_dynamic = layout{ov::PartialShape{ov::Dimension::dynamic(), 4, ov::Dimension::dynamic(), ov::Dimension::dynamic()},
+                                       data_types::f32, format::bfyx};
+    auto target_shape = engine.allocate_memory({ ov::PartialShape{4}, data_types::i32, format::bfyx });
+    set_values(target_shape, {4, 4, 1, 1});
+
+    topology topology;
+    topology.add(input_layout("input", input_layout_dynamic));
+    topology.add(data("target_shape", target_shape));
+    topology.add(shape_of("shape_of", input_info("input"), data_types::i32));
+    topology.add(broadcast("broadcast", input_info("shape_of"), input_info("target_shape"), {}, ov::op::BroadcastType::BIDIRECTIONAL));
+    topology.add(reshape("reshape", input_info("input"), input_info("broadcast"), false, ov::PartialShape{4, 4, 1, 1}));
+
+    ExecutionConfig config = get_test_default_config(engine);
+    config.set_property(ov::intel_gpu::allow_new_shape_infer(true));
+    config.set_property(ov::intel_gpu::optimize_data(true));
+    network network(engine, topology, config);
+
+    auto prog = network.get_program();
+    ASSERT_NE(prog, nullptr);
+
+    ASSERT_TRUE(check_subgraph(prog->get_node("shape_of"), prog->get_node("broadcast")));
+}