fixes

Signed-off-by: Vladimir Paramuzov <[email protected]>
vladimir-paramuzov · Dec 5, 2024 · fe2a5cf · fe2a5cf
1 parent 33af9c1
commit fe2a5cf
Show file tree

Hide file tree

Showing 3 changed files with 1 addition and 23 deletions.
diff --git a/src/plugins/intel_gpu/src/graph/graph_optimizer/prepare_buffer_fusing.cpp b/src/plugins/intel_gpu/src/graph/graph_optimizer/prepare_buffer_fusing.cpp
@@ -846,27 +846,6 @@ void prepare_buffer_fusing::run(program& p) {
                 if (user_info.first) {
                     node.get_users().front()->set_output_layout(user_info.second);
                 }
-
-                // In case that the rank of weight node of gemm is less than 4 and,
-                // it transforms to extend to 4 dims by adding 1 to begin().
-                // Therefore, the padding of crop_layout should be shifted properly.
-                const size_t TDIM = 4;
-                auto user = node.get_users().front();
-                bool allow_new_shape_infer = node.get_program().is_new_shape_infer();
-                if (!allow_new_shape_infer && user->is_type<gemm>() && user->get_dependency(1).id().compare(node.id()) == 0) {
-                    auto input_rank = user->get_kernel_impl_params()->typed_desc<gemm>()->weight_rank;
-                    if (input_rank < TDIM) {
-                        std::vector<int32_t> l_pad = {0, 0, 0, 0};
-                        std::vector<int32_t> u_pad = {0, 0, 0, 0};
-
-                        //shift right
-                        size_t shift_right = TDIM - input_rank;
-                        std::copy_n(crop_layout.data_padding._lower_size.begin(), l_pad.size() - shift_right, l_pad.begin() + shift_right);
-                        std::copy_n(crop_layout.data_padding._upper_size.begin(), u_pad.size() - shift_right, u_pad.begin() + shift_right);
-
-                        crop_layout.data_padding = padding(l_pad, u_pad);
-                    }
-                }
             }
             node.set_output_layout(crop_layout);
             node.can_be_optimized(true);

diff --git a/src/plugins/intel_gpu/src/graph/strided_slice.cpp b/src/plugins/intel_gpu/src/graph/strided_slice.cpp
@@ -177,7 +177,7 @@ void strided_slice_inst::update_output_memory() {
     if (!can_be_optimized())
         return;
 
-    if (node->get_program().is_new_shape_infer() && input_memory_ptr() == nullptr)
+    if (input_memory_ptr() == nullptr)
         return;
 
     if (static_cast<bool>(_outputs[0]) && _network.get_engine().is_the_same_buffer(output_memory(), input_memory()))

diff --git a/src/plugins/intel_gpu/tests/unit/dynamic_execution/skip_gather_at_runtime.cpp b/src/plugins/intel_gpu/tests/unit/dynamic_execution/skip_gather_at_runtime.cpp
@@ -62,7 +62,6 @@ TEST_P(skip_gather_at_runtime_test, runtime_skip) {
                         reorder("reorder", input_info("gather"), format::get_default_format(input1_rank), data_types::f32));
 
     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);