[GPU] Fix dynamic layout calling of get_dims() in prepare_buffer_fusing

[yuanxion]

Details

• Changed to get single dim from partial_shape when get_dims is not available for dynamic layout

Description of the issue

Symptom

E2E test for Pytorch_YoloV9_api_2_True will be failed at "get_dims()" for "__module.model.22/aten::cat/Concat_4" node during prepare_buffer_fusing pass.

Root cause

"__module.model.22/aten::cat/Concat_4" node has 2 dependent nodes as its inputs. The layout of 1 of the dependent node is dynamic and used to call "get_dims()", which caused the issue.

How to fix it

The concat_axis dim in dims is not dynamic, and can be used to get_length().

The code and line that caused this issue

• get_dims

openvino/src/plugins/intel_gpu/src/graph/graph_optimizer/prepare_buffer_fusing.cpp

Line 323 in 4d25dd6

auto input_length = pred_layout.get_dims()[concat_axis];

Reproduction step and snapshot

python -m pytest test_ovc_mo.py --tb=native --env_conf=.automation/env_config.yml --test_conf=test_config_local.yml -m "not launch_only_if_manually_specified " --pregen_irs=python_api_serialized/irs_mapping.csv --tf_models_version=1.15.2 --modules pipelines/production/pytorch/heavy/yolov9.py -k "Pytorch_YoloV9_api_2_True" --dynamism_type=None

Problematic graph

• get_dims with dynamic layout (dep 1)

Checklist

• Is it a proper fix? (not a workaround)
• Did you include test case for this fix, if necessary?
• Did you review existing test that can be extended to cover this scenario? Which test did you review?

Tickets:

• CVS-101242