Follow-up code review

openvinotoolkit · Dec 29, 2023 · aee3c36 · aee3c36
1 parent e631e6a
commit aee3c36
Show file tree

Hide file tree

Showing 3 changed files with 9 additions and 17 deletions.
diff --git a/src/plugins/intel_gpu/src/graph/gather.cpp b/src/plugins/intel_gpu/src/graph/gather.cpp
@@ -139,14 +139,12 @@ void gather_inst::on_execute() {
 void gather_inst::update_output_memory() {
     if (!can_be_optimized())
         return;
+    if (static_cast<bool>(_outputs[0]) && _network.get_engine().is_the_same_buffer(output_memory(), input_memory()))
+        return;
 
-    // need to run build_deps before checking same memory between input and output because deps are not setted yet in some case.
     if (_node != nullptr)
         build_deps();
 
-    if (static_cast<bool>(_outputs[0]) && _network.get_engine().is_the_same_buffer(output_memory(), input_memory()))
-        return;
-
     GPU_DEBUG_TRACE_DETAIL << id() << " : update_output_memory with mem of input " << get_node().get_dependency(0).id()
                            << " : " << input_memory_ptr()->buffer_ptr() << std::endl;
     _outputs[0] = input_memory_ptr();

diff --git a/src/plugins/intel_gpu/src/graph/network.cpp b/src/plugins/intel_gpu/src/graph/network.cpp
@@ -731,6 +731,10 @@ void network::allocate_primitives() {
     for (auto const& node : po) {
         if (node->can_be_optimized() && !node->is_dynamic()) {
             auto opt_inst = _primitives.at(node->id());
+            // build deps when prim_inst does not update dependencies yet.
+            if (!node->get_dependencies().empty() && opt_inst->dependencies().empty()) {
+                opt_inst->build_deps();
+            }
             opt_inst->update_output_memory();
         }
     }

diff --git a/src/plugins/intel_gpu/src/graph/primitive_inst.cpp b/src/plugins/intel_gpu/src/graph/primitive_inst.cpp
@@ -509,8 +509,7 @@ event::ptr primitive_inst::realloc_if_needed() {
 
     // Clear out memory if if was previously reused, but now primitive can't be optimized
     if (_node->is_type<gather>() && !can_be_optimized() && _outputs[0]
-        && (_impl_params->input_layouts[0].count() != 0 // check if layout is zero dimension before using dep_memory()
-            && _network.get_engine().is_the_same_buffer(dep_memory(0), output_memory(0)))) {
+            && dep_memory_ptr(0) && _network.get_engine().is_the_same_buffer(dep_memory(0), output_memory(0))) {
         _outputs[0] = nullptr;
         max_output_layout_size = 0;
     }
@@ -920,6 +919,8 @@ void primitive_inst::do_runtime_in_place_kv_cache() {
 void primitive_inst::do_runtime_skip_gather() {
     // Check pattern
     if (!get_node().is_type<gather>()
+        || !get_node().can_be_optimized()
+        || (_impl_params->get_output_layout().count() == 0)
         || _impl_params->has_fused_primitives()
         || _impl_params->get_input_layout(0).data_type != _impl_params->get_output_layout().data_type
         || get_node().get_dependency(1).is_constant() || get_node().get_dependency(1).is_type<data>())
@@ -932,17 +933,6 @@ void primitive_inst::do_runtime_skip_gather() {
     auto idx_shape = _impl_params->get_input_layout(1).get_shape();
     auto idx_rank = idx_shape.size();
 
-    // If all input layout count is zero, then it can be optimized because all input will be skipped for empty output.
-    if (_impl_params->get_input_layout(0).count() == 0 && _impl_params->get_input_layout(1).count() == 0) {
-        GPU_DEBUG_TRACE_DETAIL << "[do_runtime_skip_gather] " << id() << " : can_be_optimized" << std::endl;
-        GPU_DEBUG_TRACE_DETAIL << "            - Input layout : " << _impl_params->get_input_layout(0).to_short_string() << std::endl;
-        GPU_DEBUG_TRACE_DETAIL << "            - Indices layout : " << _impl_params->get_input_layout(1).to_short_string() << std::endl;
-        GPU_DEBUG_TRACE_DETAIL << "            - Gather axis : " << axis << std::endl;
-        GPU_DEBUG_TRACE_DETAIL << "            - Output layout : " << _impl_params->get_output_layout().to_short_string() << std::endl;
-        set_can_be_optimized(true);
-        return;
-    }
-
     if (idx_rank != 1) {
         GPU_DEBUG_TRACE_DETAIL << "-- Cannot optimize becuase of its indices rank " << idx_rank << std::endl;
         return;